# Случайные графы Эрдеша-Реньи и задача изоморфизма графов А.М. Райгородский, М. Тихомиров, А. Голованов ## 1 Блок 1 ## 1.1 Случайный граф Эрдеша-Реньи Определение. Граф $G=(V, E)$ состоит из множества вершин $V$ и множества ребер $E$, где под ребром мы понимаем неупорядоченную пару различных вершин. Легко видеть, что в графе может быть максимум $\binom{n}{2}=\frac{n(n-1)}{2}$ ребер. Определение. Зафиксируем $p \in(0,1)$ и назовем вероятностью графа (в модели, или в вероятностном пространстве, Эрдеша-Реньи) с $n$ вершинами $\{1,2, \ldots, n\}$ и $е$ рёбрами число $\mathrm{P}(G)=$ $\mathrm{P}_{p}(G):=p^{e}(1-p)^{\frac{n(n-1)}{2}-e}$. Вероятностъю семейства (или, что то же самое, свойства) графов с вершинами $1,2, \ldots, n$ называется сумма вероятностей входящих в него графов. Если $\Omega$ - множество всех графов с $n$ вершинами, то $\mathrm{P}(\Omega)=1$. При вышеприведенном формальном определении это теорема. Вот ее доказательство: $$ \left.\mathrm{P}(\Omega)=\sum_{k=0}^{\binom{n}{2}}\binom{n}{2}\right) p^{k}(1-p)^{\binom{n}{2}-k}=(p+(1-p))^{\binom{n}{2}}=1 $$ Определение. Случайной величиной называется функция, определённая на множестве графов с вершинами $1,2, \ldots, n$. Например, количество рёбер графа - случайная величина. Пусть случайная величина $Y$ принимает $k$ различных значений $y_{1}, \ldots, y_{k}$. Тогда Определение. математическим ожиданием (мат. ожиданием) случайной величины $Y$ называется её «взвешенное среднее» $$ \mathrm{E} Y:=\sum_{s=1}^{k} y_{s} \mathrm{P}\left(Y^{-1}\left(y_{s}\right)\right) $$ где $Y^{-1}\left(y_{s}\right)$ - множество всех графов $G$, для которых $Y(G)=y_{s}$. Последнюю вероятность обозначают $\mathrm{P}\left(Y=y_{s}\right)$. Определение. Два случайных события $A$ и $B$ называются независимыми, если $\mathrm{P}(A \cap B)$ (вероятность одновременного наступления $A$ и $B)=\mathrm{P}(A) \cdot \mathrm{P}(B)$ (произведение вероятностей $A$ и $B$ ). Случайные величины $f$ и $g$ независимы, если события $f=x$ и $g=y$ независимы для всех $x$ и $y$. Комментарий: это определение справледливо только для дискретных случайных величин, то есть принимающих не более чем счетное число значений. Задача 1. Пусть граф $H$ состоит из $n$ вершин и $m$ ребер. Какова вероятность, что $G(n, p)=H$, т.е. при генерации случайного графа согласно определению выше получится в точности граф $H$ ? Задача 2. Какова вероятность, что в графе $G(n, p)$ будет ровно $m$ ребер? Задача 3. Какова вероятность, что в графе $G(n, p)$ не будет изолированных вершин (т.е. вершин, не соединенных ребром ни с какой другой вершиной)? Пусть случайная величина $f$ принимает численные значения, и вероятность того, что $f=x$ равна $p_{x}$. Математическим ожиданием (также матожиданием или средним значением) $f$ называется число $\sum_{x} x \cdot p_{x}$. Будем обозначать матожидание величины $f$ как $\mathrm{E} f$. Задача 4. Докажите формулы: a) $\mathrm{E}[c \cdot f]=c \cdot \mathrm{E} f$, если $c$ - константа; b) $\mathrm{E}[f+g]=\mathrm{E} f+\mathrm{E} g$ (в том числе и для зависимых $f$ и $g$ !); c) если $f$ и $g$ независимы, то $\mathrm{E}[f \cdot g]=\mathrm{E} f \cdot \mathrm{E} g$; Задача 5. Найдите матожидание следующих величин: а) число ребер графа $G(n, p)$; b) степень вершины $G(n, p)$ (поскольку определение $G(n, p)$ симметрично, то выбор конкретной вершины неважен); c) количество треугольников в $G(n, p)$; d) количество гамильтоновых путей и циклов в $G(n, p)$ (гамильтонов путь/цикл - путь/цикл, проходящий через каждую вершину ровно один раз); е) клик (полных подграфов) на $k$ вершинах. Определение. Дисперсией $\mathrm{D} f$ случайной величины $f$ называют величину $\mathrm{E}\left[(f-\mathrm{E} f)^{2}\right]$. Задача 6. Докажите формулы: a) $\mathrm{D}[c \cdot f]=c^{2} \cdot \mathrm{D} f$, если $c$ - константа; b) $\mathrm{D} f=\mathrm{E}\left[f^{2}\right]-(\mathrm{E} f)^{2}$; c) если $f$ и $g$ независимы, то $D[f+g]=\mathrm{D} f+\mathrm{D} g$. Задача 7. Найдите дисперсию степени вершины $G(n, p)$. Задача 8. Найдите дисперсию числа треугольников $G(n, p)$ (будьте аккуратны с независимостью случайных величин!). ## 1.2 Алгоритмы на графах и их сложность Пусть у нас есть описание некоторого графа и мы хотим проверить какое-то его свойство, например, связность. Воспользуемся следующим алгоритмом: Обход в глубину (DFS, depth-first search). процедура dfs (вершина $v$ ): used $_{v}:=1$ для каждой вершины $u$ - соседа $v$ : если $\operatorname{used}_{u}=0$, запустить $\operatorname{dfs}(u)$ функция is_connected (граф $G$ ) : used $_{v}=0$ для всех вершин $v$ $\mathrm{dfs}$ (произвольная вершина $G$ ) если $\operatorname{used}_{v}=1$ для всех вершин $v \in G$, вернуть дА иначе вернуть НЕТ Описание алгоритма приведено в формате псевдокода. Для лучшего понимания того, как работает алгоритм, рекомендуется проэмулировать работу этого кода мысленно или на бумаге на некоторых несложных входных данных. Обратите внимание, что процедура $\mathrm{dfs}$ является рекурсивной, т.е. ее запуски могут вкладываться друг в друга, при этом завершение одного запуска возвращает работу программы в предыдущий. Задача 9. Докажите, что данный алгоритм верно проверяет граф на связность. Задача 10. Докажите, что данный алгоритм совершает $O(n+m)$ действий (т.е. не более $c(n+m)$ для некоторой константы $c$, не зависящей от входного графа) на графе с $n$ вершинами и $m$ ребрами. Задача 11. Модифицируйте данный алгоритм для поиска количества компонент связности. Определение. Мостом в графе $G$ будем называть ребро, удаление которого приводит к увеличению количества компонент связности. Задача 12. Постройте алгоритм, по описанию графа определяющий все его мосты, количества действий которого составляет: a) $O\left((n+m)^{2}\right)$; b) $O(n(n+m))$; c) $\left(^{*}\right) O(n+m)$. Рассмотрим другой метод для той же задачи: Поиск в ширину (bfs, breadth-first search). Псевдокод: процедура bfs (вершина $v$ ): queue $=\{\mathrm{v}\} / /$ очередь вершин used $_{v}:=1$ пока queue $\neq \varnothing$ : вершина $w$ = начало queue удалить $w$ из начала queue для каждой вершины $u$ - соседа $w$ : если used $_{u}=0$ : used $_{u}=1$ добавить $u$ в конец queue функция is_connected (граф $G$ ): $\operatorname{used}_{v}=0$ для всех вершин $v / /$ used имеет размер $\mathrm{n}$ $\mathrm{bfs}$ (произвольная вершина $G$ ) если $\operatorname{used}_{v}=1$ для всех вершин $v \in G$, вернуть дА иначе вернуть НET Задача 13. Докажите, что алгоритм корректен, и найдите асимптотику времени его работы. Задача 14. Модифицируйте алгоритм для поиска длин кратчайших путей от выбранной вершины $v$ до всех остальных. Задача 15. Пусть для каждого ребра $е$ графа $G$ определена его длина - неотрицательное число $w(e)$. Постройте алгоритм, находяющий длины кратчайших путей от выбранной вершины $v$ до всех остальных со временем работы: a) $O\left(n^{2}+m\right)$; b) (*) $O(n+m \log n)$. ## 1.3 Задача об изоморфизме графов Определение. Графы $G=\left(V_{G}, E_{G}\right)$ и $H=\left(V_{H}, E_{H}\right)$ называются изоморфными, если существует биекция $\pi: V_{G} \rightarrow V_{H}$, такая что неупорядоченная пара вершин $v u$ присутствует в $E_{G}$ тогда и только тогда, когда пара $\pi(v) \pi(u)$ присутствует в $E_{H} \cdot \pi$ называется изоморфизмом графов $G$ и $H$. Задача 16. Нарисуйте все попарно неизоморфные графы с 3 и 4 вершинами. Определение. Будем называть инвариантом функцию $f$, сопоставляющую каждому графу $G$ некоторое значение $f(G)$, такую что $f(G)=f(H)$ для любой пары изоморфных графов $G$ и $H$. Очевидно, что из $f(G) \neq f(H)$ следует, что $G$ и $H$ неизоморфны, однако обратное может быть неверно (т.е. $f(G)=f(H)$ может выполняться и для неизоморфных графов). Определение. Определим лексикографическое или алфавитное сравнение двух последовательностей следующим образом: - пустая последовательность меньше любой непустой последовательности; - если первый элемент последовательности $A$ меньше первого элемент последовательности $B$, то $A1$. Однако, существуют быстрые алгоритмы для проверки изоморфизма графов в ряде частных случаев. Задача 19. Предъявите алгоритм, проверяющий изоморфизм графов за время $O(n+m)$ в следующих случаях: a) степени вершин графов не превосходят 2; b) графы не содержат циклов; c) (*) каждый из графов содержит ровно один простой цикл. Если предполагать, что входные графы для задачи изоморфизма сгенерированы случайно, можно пытаться пользоваться простыми алгоритмами для проверки их изоморфизма и оценивать вероятность того, что они могут сработать неправильно. Наша цель - проанализировать вероятность отказа следующего простого алгоритма: функция invariant (граф $G$ ) : для каждой вершины $v$ графа $G$ : построить $l_{v}$ - упорядоченный список степеней соседей $v$ если среди $l_{v}$ есть одинаковые списки, вернуть ОТКАЗ иначе вернуть лексикографически отсортированный набор $l_{v}$ функция are_isomorphic (графы $G, H$ ): если invariant $(G)$ или invariant $(H)$ вернул ОТКАЗ, вернуть ОТКАЗ иначе вернуть invariant $(G)=$ invariant $(H)$ Задача 20. Опишите способ реализации данного алгоритма за время $O(n+m)$. Задача 21. Рассмотрим пару вершин $a$ и $b$. Докажите, что $l_{a}=l_{b}$ тогда и только тогда, когда после удаления $a$ и $b$ из графа набор степеней бывших соседей $a$ совпадает с набором степеней бывших соседей $b$. Задача 22. Докажите, что вероятность отказа не превосходит $n^{2} \cdot \mathrm{P}\left(l_{a}=l_{b}\right)$. Таким образом, если мы покажем, что $\mathrm{P}\left(l_{a}=l_{b}\right)$ мала, то мала будет и вероятность отказа. Из задачи 21 следует, что необходимо оценить вероятность такого события: при независимом выборе двух случайных наборов вершин $N_{a}$ и $N_{b}$ (каждая вершина выбирается из $\{1, \ldots, n-2\}$ независимо с вероятностью $p$ ) наборы степеней вершин из $N_{a}$ и $N_{b}$ в случайном графе $G=G(n-2, p)$ совпадут ## 2 Блок 2 ## 2.1 Комбинаторные величины и их асимптотика Определения. Будем говорить, что: - $f(n)=o(g(n))$, если $\lim _{n \rightarrow \infty} \frac{f(n)}{g(n)}=0$. - $f(n)=O(g(n))$, если $\lim _{n \rightarrow \infty} \frac{f(n)}{g(n)}<\infty$ (т.е. отношение ограничено сверху некоторым конечным числом). - $f(n)=\omega(g(n))$, если $\lim _{n \rightarrow \infty} \frac{f(n)}{g(n)}=\infty$. - $f(n)=\Omega(g(n))$, если $\lim _{n \rightarrow \infty} \frac{f(n)}{g(n)}>0$. - $f(n)=\Theta(g(n))$, если $0<\lim _{n \rightarrow \infty} \frac{f(n)}{g(n)}<\infty$. - $f(n) \sim g(n)$, если $\lim _{n \rightarrow \infty} \frac{f(n)}{g(n)}=1$. В некоторых из следующих задач надо будет «найти асимптотику» для некоторых функций. Это значит, что для каждой из предложенных функций $f(n)$ надо найти функцию в «явном» виде $g(n)$ (грубо говоря, в записи этой функции присутствуют только сложение, вычитание, умножение, деление, возведение в степень, а также функция в замкнутой форме, то есть, не содержит знаков “...", " $\Sigma$ ", "П"), что $f(n) \sim g(n)$. Задача 23. Докажите, что $\frac{2^{n}}{n+1} \leqslant\binom{ n}{\left\lfloor\frac{n}{2}\right\rfloor} \leqslant 2^{n}$. Задача 24. Найдите асимптотику для $\sqrt[n]{\binom{n}{\left\lfloor\frac{n}{2}\right\rfloor}}$. Указание: к чему стремится величина $\sqrt[n]{\frac{1}{n+1}}$ при $n \rightarrow \infty$ ? Задача 25. Докажите, что $\frac{3^{3 n}}{3 n+1} \leqslant 2^{2 n}\binom{3 n}{n} \leqslant 3^{3 n}$. Указание: найдите комбинаторную интерпретацию для чисел $3^{3 n}$ и $2^{2 n}\binom{3 n}{n}$. Задача 26. Найдите асимптотику для $\sqrt[n]{\binom{3 n}{n}}$. Задача 27. Пусть $0<\alpha<1$. Докажите, что $\binom{n}{\alpha n}=\left(\alpha^{-\alpha}(1-\alpha)^{\alpha-1}+o(1)\right)^{n}$. Без доказательства приведем следующую важную формулу: Формула Стирлинга. $n!\sim \sqrt{2 \pi n}\left(\frac{n}{e}\right)^{n}$ (здесь $e=\lim _{n \rightarrow \infty}\left(1+\frac{1}{n}\right)^{n}$ - второй замечательный предел). Задача 28. Найдите асимптотику для $\binom{2 n}{n}$. Задача 29. Найдите асимптотику для $\binom{n^{2}}{n}$. Задача 30. Найдите асимптотику для $(2 n-1)!!=1 \cdot 3 \cdot \ldots \cdot(2 n-1)$. Задача 31. Докажите, что количество различных неизоморфных графов на $n$ вершинах составляет $\Omega\left(\frac{1}{\sqrt{2 \pi n}}\left(\frac{e 2^{(n-1) / 2}}{n}\right)^{n}\right)$. Определение. Бернуллиева величина $c$ параметром $p$ - величина, равная 1 с вероятностью $p$, и равная 0 с вероятностью $1-p$. Обозначим $b(k, n, p)$ вероятность того, что среди $n$ независимых в совокупности Бернуллиевых величин с параметром $p$ ровно $k$ будут равны 1 . Задача 32. Найдите формулу для $b(k, n, p)$. Задача 33. Найдите асимптотику величины $\max _{k} b(k, n, p)$ при условии $p=\omega(1 / n)$. Указание: при каком $k$ достигается максимум $b(k, n, p)$ ? Задача 34. Пусть $p=\omega(1 / n)$. Докажите, что вероятность совпадения степеней двух выбранных заранее вершин $G(n, p)$ составляет $O(1 / \sqrt{p n})$. Оценка Чернова. Если $X$ - сумма $n$ независимых Бернуллиевых величин с параметром $p$, то $\mathrm{P}(|X-p n|>p n \delta) \leq 2 e^{-(\delta p n)^{2} / 3 p n}=2 e^{-\delta^{2} p n / 3}$. (Замечание: стоит обратить внимание на то, что $\mathrm{E} X=$ pn.) Задача 35. При условии $p=\omega(1 / n)$ докажите, что $|b(k, n, p)-b(k-1, n-1, p)|=O(\sqrt{\ln n} / p n)$. a) Запишите $Y=b(k, n, p)-b(k-1, n-1, p)$ в виде $\delta b(k, n, p)$, где $\delta$ - некоторая величина. b) Оцените $|Y|$ непосредственно для маленьких значений $|\delta|$ и при помощи оценки Чернова для больших значений $|\delta|$. Найдите оптимальную границу между «маленькими» и «большими» значениями $|\delta|$. Задача 36. При условии $p=\omega(1 / n)$ докажите, что: a) $|b(k, n, p)-b(k, n-1, p)|=O(\sqrt{\ln n} / p n)$. b) $|b(k-1, n-1, p)-b(k, n-1, p)|=O(\sqrt{\ln n} / p n)$. Задача 37. Пусть $p=\omega(\ln n / n),|k-p n|=o(\sqrt{p n / \ln n})$. Докажите, что $b(k, n, p)=\Theta(1 / \sqrt{p n})$. ## 2.2 Реберный мартингал и интервалы, содержащие степени случайного графа Определение. Условной вероятностъю события $A$ относительно события $B$ называется величина $\mathrm{P}(A \mid B)=\mathrm{P}(A \cap B) / \mathrm{P}(B)$. Условным матожиданием случайной величины $f$ относительно события $B$ называется величина $\mathrm{E}[F \mid B]=\sum_{x} x \cdot \mathrm{P}(A=x \mid B)$. Неформально, характеристики, условные относительно события $B$, являются усреднениями по исходам, принадлежащим событию $B$. Пусть $f$ и $g$ - случайные величины. Тогда условным матожиданием $f$ относительно $g$ называется случайная величина $\mathrm{E}[f \mid g]$, принимающая значение $x$ с вероятностью $\sum_{\mathrm{E}[f \mid g=z]=x} \mathrm{P}[g=z]$. Другими словами, условное матожидание $f$ относительно $g$ - это величина $h$, которую можно «сгенерировать» так: сгенерировать $z$ - значение $g$, после чего взять условное матожидание $f$ относительно события $g=z$ и вернуть в качестве значения $h$. Задача 38. Докажите, что если $c$ - константа, то $\mathrm{E}[f \mid c]=\mathrm{E} f$. Задача 39. Что такое $\mathrm{E}[f \mid f]$ ? Задача 40. Докажите, что $\mathrm{E}[f+g \mid h]=\mathrm{E}[f \mid h]+\mathrm{E}[g \mid h]$. Указание: обратите внимание, что $c$ обеих сторон от равенства стоят случайные величины, поэтому доказать нужно их равенство при каждом исходе. Задача 41. Докажите, что $\mathrm{E}[\mathrm{E}[f \mid g]]=\mathrm{E} f$. Ранее мы хотели оценить вероятность того, что $l_{a}=l_{b}$. Напомним: $N_{a}$ и $N_{b}$-случайные наборы вершин графа $G=G(n-2, p)$. Если наборы степеней вершин $N_{a}$ и $N_{b}$ совпадают, то внутри любого интервала степеней количество представителей этих наборов совпадает. Рассмотрим интервал $I=\left[D_{0}, D_{0}+R-1\right]$. Обозначим $f$ количество вершин $G(n, p)$, степени которых попадают в интервал $I$. Пронумеруем всевозможные ребра графа на $n$ вершинах от 1 до $\binom{n}{2}$ в порядке $\{1,2\},\{1,3\}, \ldots$, $\{2,3\}, \ldots$ Пусть $f$ - некоторая характеристика графа $G(n, p)$ (например, та, которая определена выше). Обозначим $C_{i}$ случайную величину, равную 1 , если ребро $i$ присутствует в $G(n, p)$, и 0 в противном случае. Определение. Реберным мартингалом, отвечающим $f$, будем называть последовательность случайных величин $f_{0}, \ldots, f_{\binom{n}{2}}$, где $f_{i}=\mathrm{E}\left[f \mid C_{1}, \ldots, C_{i}\right]$. Неформально, значение $f_{i}$ пытается «угадать» значение $f$, имея лишь информацию о наличии или отсутствии первых $i$ ребер графа. На самом деле мартингалы можно определить в гораздо более общем случае. Определение. Мартингал - это любая последовательность случайных величин $f_{0}, \ldots, f_{m}$, такая что для каждого $i$ выполнено: $$ f_{i-1}=\mathrm{E}\left[f_{i} \mid f_{i-1}, \ldots, f_{0}\right] $$ Задача 42. Пусть $Y_{1}, \ldots, Y_{m}$ - независимые случайные величины, каждая из которых принимает значения $+1,-1$ с равными вероятностями. Положим $X_{i}=Y_{1}+\ldots+Y_{i}$. Докажите, что $X_{1}, \ldots, X_{m}$ образуют мартингал. Задача 43. Докажите, что математические ожидания всех случайных вЙеличин в любом мартингале совпадают. Задача 44. (*) Докажите, что реберный мартингал является мартингалом. Неравенство Азумы. Пусть $X_{0}, \ldots, X_{n}$ - мартингал. Пусть также $\left|X_{i}-X_{i-1}\right|$ не превосходит $c_{i}$ при любом исходе. Тогда $$ \mathrm{P}\left[X_{n}-X_{0} \geq t\right] \leq e^{-t^{2} / 2 \sum_{i=1}^{n} c_{i}^{2}} $$ Рассмотрим ребро $i=\{a, b\}$, и пусть $a0$. Задача 51. (**) Приведите алгоритм проверки изоморфизма графов Эрдеша-Реньи с полиномиальным временем работы и экспоненциально маленькой вероятностью отказа для нетривиального диапазона вероятности ребра $p$. Задача 1. Все ребра, которые есть в $H$, должны быть и в графе $G=G(n, p)$, причем на тех же местах, а ребер, которых нет в графе $H$, не должно быть и в $G$. Вероятность того, что $m$ ребер есть в графе $G$ на нужных местах $-p^{m}$, а вероятность отсутствия $\frac{n(n-1)}{2}-m$ ребер на остальных местах есть $(1-p)^{\frac{n(n-1)}{2}-m}$. Так как события независимы, то вероятность того, что $G=H$, равна $p^{m}(1-p)^{\frac{n(n-1)}{2}-m}$. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-009.jpg?height=84&width=1824&top_left_y=457&top_left_x=172) способов). Вероятность того, что они все будут присутствовать есть $p^{m}$, а того, что остальные отсутсвовать $-(1-p)^{\binom{n}{2}-m}$, значит общая вероятность равна $\left(\begin{array}{c}\left(\begin{array}{c}n \\ 2 \\ m\end{array}\right)\end{array} p^{m}(1-p)^{\binom{n}{2}-m}\right.$. Задача 3. Coming soon... Задача 4. а) Пусть $f$ принимает $k$ различных значений $f_{1}, \ldots, f_{k}$. Тогда $\mathrm{E}[c \cdot f]=\sum_{s=1}^{k} c f_{s} \mathrm{P}\left(f^{-1}\left(f_{s}\right)\right)=$ $c \sum_{s=1}^{k} f_{s} \mathrm{P}\left(f^{-1}\left(f_{s}\right)\right)=c \cdot \mathrm{E} f$ b) Пусть $f$ и $g$ на графах $G_{1}, \ldots, G_{k}$ принимают значения $f_{1}, \ldots, f_{k}$ и $g_{1}, \ldots, g_{k}$ соответственно. Тогда $\mathrm{E}[f+g]=\sum_{s=1}^{k}\left(f_{s}+g_{s}\right) \mathrm{P}\left(G_{s}\right)=\sum_{s=1}^{k} f_{s} \mathrm{P}\left(G_{s}\right)+\sum_{s=1}^{k} g_{s} \mathrm{P}\left(G_{s}\right)=\mathrm{E} f+\mathrm{E} g$; c) $\mathrm{E}[f \cdot g]=\sum_{k, m}\left(f_{k} g_{m}\right) \mathrm{P}\left(f=f_{k}, g=g_{m}\right)=\sum_{k} x_{k} \mathrm{P}\left(f=f_{k}\right) \cdot \sum_{m} x_{m} \mathrm{P}\left(g=g_{m}\right)=\mathrm{E} f \cdot \mathrm{E} g$. ## Задача 5. a) Пусть число ребер $-a$ (случайная величина). Можно заметить, что $a=a_{1}+\ldots+a_{\binom{n}{2}}$, где $a_{i}-$ индикатор наличия $i$-го ребра (случайная величина). Тогда $\mathrm{E}[a]=\mathrm{E}\left[\sum_{i=1}^{\binom{n}{2}} a_{i}\right]=\sum_{i=1}^{\binom{n}{2}} \mathrm{E} a_{i}=\binom{n}{2} \cdot p$; b) Аналогично пункту $a$ ), только $a=a_{1}+\ldots+a_{n-1}$. И значит $\mathrm{E}[a]=(n-1) \cdot p$; c) Пусть число треугольников $-a$ (случайная величина). Можно заметить, что $a=a_{1}+\ldots+a_{\binom{n}{3}}$, где $a_{i}$ - индикатор наличия $i$-го треугольника (случайная величина). Тогда $\mathrm{E}[a]=\mathrm{E}\left[\sum_{i=1}^{\binom{n}{3}} a_{i}\right]=$ $\sum_{i=1}^{\binom{n}{2}} \mathrm{E} a_{i}=\binom{n}{3} \cdot p^{3}$ d) Так как количество путей есть $\frac{n!}{2}$. То математическое ожидания их кол-ва $-\frac{n!}{2} \cdot p^{n-1}$; е) Так как количество циклов есть $\frac{(n-1)!}{2}$. То математическое ожидания их кол-ва $-\frac{(n-1)!}{2} \cdot p^{n}$. Задача 6. a) $\mathrm{D}[c \cdot f]=\mathrm{E}\left[(c f-\mathrm{E}[c f])^{2}\right]=\mathrm{E}\left[(c f-c \cdot \mathrm{E} f)^{2}\right]=c^{2} \cdot \mathrm{E}\left[(f-\mathrm{E} f)^{2}\right]=c^{2} \cdot \mathrm{E} f$; b) $\mathrm{D} f=\mathrm{E}\left[\left(f-(\mathrm{E} f)^{2}\right]=\mathrm{E}\left[f^{2}-2 f \cdot \mathrm{E} f+(\mathrm{E} f)^{2}\right]=\mathrm{E}\left[f^{2}\right]-2(\mathrm{E} f)^{2}+(\mathrm{E} f)^{2}=\mathrm{E}\left[f^{2}\right]-(\mathrm{E} f)^{2}\right.$; c) $\mathrm{D}[f+g]=\mathrm{E}\left[(f+g)^{2}\right]-(\mathrm{E}[f+g])^{2}=\mathrm{E}\left[f^{2}+2 f \cdot g+g^{2}\right]-(\mathrm{E} f)^{2}-2 \mathrm{E} f \cdot \mathrm{E} g-(\mathrm{E} g)^{2}=\mathrm{D} f-\mathrm{D} g$. Задача 7. Пусть степень вершины - $a$ (случайная величина). Можно представить $a$ как сумму $a_{1}+\ldots+a_{n-1}$, где $a_{i}-$ индикатор наличия $i$-го исходящго ребра. Тогда $\mathrm{D} a=\mathrm{D}\left[\sum_{i=1}^{n-1} a_{i}\right]=\sum_{i=1}^{n-1} \mathrm{D} a_{i}=$ $(n-1)\left(p-p^{2}\right)$. Задача 8. Пусть число треугольников $G(n, p)-f$ (случайная величина). Представим $f$ в виде ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-009.jpg?height=85&width=1892&top_left_y=2321&top_left_x=108) как мы знаем $\mathrm{E} f=\binom{n}{3} \cdot p^{3}$, найдем $\mathrm{E}\left[f^{2}\right]=\mathrm{E}\left[\left(\sum_{i=1}^{\binom{n}{3}} f_{i}\right)^{2}\right]=\mathrm{E}\left[\sum_{i=1}^{\binom{n}{3}} f_{i}^{2}+2 \cdot \sum_{i=1, j=1, i!=j}^{\binom{n}{3}} f_{i} f_{j}\right]=\mathrm{E}\left[\sum_{i=1}^{\binom{n}{3}} f_{i}+2 \cdot\right.$ $\left.\sum_{i=1, j=1, i!=j}^{\binom{n}{3}} f_{i} f_{j}\right]=\mathrm{E} f+2 \cdot \sum_{i=1, j=1, i!=j}^{\binom{n}{3}} \mathrm{E}\left[f_{i} f_{j}\right]$. Осталось подсчитать $\sum_{i=1, j=1, i!=j}^{\binom{n}{3}} \mathrm{E}\left[f_{i} f_{j}\right]:$ 1. Если так оказалось, что $f_{i}$ и $f_{j}$ независимы (им соответствуют реберно не пересекающиеся треугольники), то $\mathrm{E}\left[f_{i} f_{j}\right]=\mathrm{E} f_{i} \cdot \mathrm{E} f_{j}=p^{6}$. Всего таких пар треугольников $\frac{1}{2} \cdot\binom{n}{3} \cdot\left(\binom{n-3}{3}+\binom{n-3}{2}\right)$; 2. Иначе треугольники событий $f_{i}, f_{j}$ пересекаются по одному ребру, значит $\mathrm{E}\left[f_{i} f_{j}\right]=p^{5}$. И их количество равно $\frac{1}{2} \cdot\binom{n}{5} \cdot\binom{n}{3}$ Значит $\mathrm{E} f=\binom{n}{3} \cdot p^{3}+\binom{n}{3} \cdot\left(\binom{n-3}{3}+\binom{n-3}{2}\right) \cdot p^{6}+\binom{n}{5} \cdot\binom{n}{3} \cdot p^{5}-\left(\binom{n}{3} \cdot p^{3}\right)^{2}$. Задача 9. Dfs всегда найдет путь до вершины, если он существует. Доказательство: Для произвольного графа G: База. Пути из не более, чем 1 вершины алгоритм находит верно (путь от начальной вершины до нее же). Предположение: Алгоритм посещает все вершины, находящиеся на расстоянии не более, чем $k-1$, от начальной. Шаг: Рассмотрим любую вершину $v$, находящуюся на расстоянии $k$ от начальной. Она, очевидно, соединена ребром с какой-то вершиной, находящейся на расстоянии $k$ - 1 от начальной - назовем еe $w$. Значит, мы перейдем в вершину $v$ при просмотре вершины $w$. Задача 10. Алгоритм просматривает каждое ребро один раз, и выполняет для каждой вершины константное число действий. Обозначая число вершин как $n$, а ребер - как $m$, получаем, что время работы $-O(n+m)$. Задача 11. Достаточно пройтись по всем вершинам и запуститься от них, если $u s e d_{v}=0$, и увеличить ответ. Этот алгоритм корректен, так как при обходе посещается вся компонента. Задача 12. Запустим обход в глубину из произвольной вершины графа; обозначим её через root. Заметим следующий факт: Пусть мы находимся в обходе в глубину, просматривая сейчас все рёбра из вершины $v$. Тогда, если текущее ребро $(v, t o)$ таково, что из вершины to и из любого её потомка в дереве обхода в глубину нет обратного ребра в вершину $v$ или какого-либо её предка, то это ребро является мостом. В противном случае оно мостом не является. Теперь осталось научиться проверять этот факт для каждой вершины эффективно. Для этого воспользуемся "временами входа в вершину вычисляемыми алгоритмом поиска в глубину. Итак, пусть $\operatorname{tin}[v]$ - это время захода поиска в глубину в вершину $v$. Теперь введём массив $f u p[v]$, который и позволит нам отвечать на вышеописанные запросы. Время $f u p[v]$ равно минимуму из времени захода в саму вершину $\operatorname{tin}[v]$, времён захода в каждую вершину $p$, являющуюся концом некоторого обратного ребра $(v, p)$, а также из всех значений $f u p[t o]$ для каждой вершины $t o$, являющейся непосредственным сыном $v$ в дереве поиска: $$ f u p[v]=\min \left\{\begin{array}{l} \operatorname{tin}[v] \\ \operatorname{tin}[p], \text { для всех }(v, p)-\text { обратное ребро, } \\ f u p[t o], \text { для всех }(v, t o)-\text { ребро дерева, где tо не предок } \end{array}\right. $$ Тогда, из вершины $v$ или её потомка есть обратное ребро в её предка тогда и только тогда, когда найдётся такой сын $t o$, что $f u p[t o] \leqslant t i n[v]$. Таким образом, если для текущего ребра ( $v, t o)$ (принадлежащего дереву поиска) выполняется $f u p[t o]>\operatorname{tin}[v]$, то это ребро является мостом; в противном случае оно мостом не является. Задача 13. Доказательство корректности проводится как и для $D F S$ при помощи индукции. Докажем время работы данного алгоритма: В очередь добавляются только непосещенные вершины, поэтому каждая вершина посещается не более одного раза. Операции внесения в очередь и удаления из нее требуют $O(1)$ времени, так что общее время работы с очередью составляет $O(|V|)$ операций. Для каждой вершины $v$ рассматривается не более $\operatorname{deg}(v)$ ребер, инцидентных ей. Так как $\sum_{v \in V} \operatorname{deg}(v)=2|E|$, то время, используемое на работу с ребрами, составляет $O(|E|)$. Поэтому общее время работы алгоритма поиска в ширину - $O(|V|+|E|)$. Задача 14. Достаточно подсчитать вспомогательную величину $d[v]=$ кратчайшее растояние до стартовой вершины. Положим $d[s]=0$ ( $s$ - стартовая вершина). И если мы хотим добавить непосещенную вершину $v$ в очередь, то положим $d[v]=d[u]+1$, где вершина $u$ - предок вершины $v$. Задача 15. Пусть $s$ - вершина от которой мы хотим посчитать расстояния. Заведём массив $d[]$, в котором для каждой вершины $v$ будем хранить текущую длину $d[v]$ кратчайшего пути из $s$ в $v$. Изначально $d[s]=0$, а для всех остальных вершин эта длина равна бесконечности: $$ d[v]=\infty, v \neq s $$ Кроме того, для каждой вершины $v$ будем хранить, помечена она ещё или нет, т.е. заведём булевский массив $u[]$. Изначально все вершины не помечены, т.е. $$ u[v]=\text { false } $$ Сам алгоритм Дейкстры состоит из $\mathrm{n}$ итераций. На очередной итерации выбирается вершина $v \mathrm{c}$ наименьшей величиной $d[v]$ среди ещё не помеченных, т.е.: $$ d[v]=\min _{p: u[p]=\text { false }} d[p] $$ Выбранная таким образом вершина $v$ отмечается помеченной. Далее, на текущей итерации, из вершины $v$ производятся релаксации: просматриваются все рёбра $(v, t o)$, исходящие из вершины $v$, и для каждой такой вершины to алгоритм пытается улучшить значение $d[t o]$. Пусть длина текущего ребра равна len, тогда в виде кода релаксация выглядит как: $$ d[t o]=\min (d[t o], d[v]+\text { len }) $$ На этом текущая итерация заканчивается, алгоритм переходит к следующей итерации. При этом в конце концов, после $n$ итераций, все вершины графа станут помеченными, и алгоритм свою работу завершает. Утверждается, что найденные значения $d[v]$ и есть искомые длины кратчайших путей из $s$ в $v$. Время работы алгоритма зависит от двух операций: извлечение минимума и изменение значения $d[v]$. При простейшей реализации эти операции потребуют соответственно $O(n)$ и $\mathrm{O}(1)$ времени. Учитывая, что первая операция всего выполняется $O(n)$ раз, а вторая - $O(m)$, получаем асимптотику простейшей реализации алгоритма Дейкстры: $O\left(n^{2}+m\right)$. Если же использовать структуру куча, то время работы будет $O(n+n \log n+m \log n)=O(n+m \log n)$. Задача 16. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-012.jpg?height=933&width=1087&top_left_y=241&top_left_x=213) ## Задача 17. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-012.jpg?height=320&width=570&top_left_y=1350&top_left_x=238) ## Задача 18. Задача 19. Для начала научимся различать два дерева $G$ и $H$. Найдем центры (вершина, которая имеет наименьшее максимальное расстояние до всех остальных) этих деревьев и сделаем их корнями. Если так оказалось, что центра два, то подвесим их за фиктивную (новую) вершину. Далее нам нужно сравнить два подвешенных дерева на изоморфизм (так, чтобы корень одного перешел в корень другого). Теперь научимся кодировать подвешенное дерево. Для этого будем нумеровать вершины. Сопоставим всем висячим вершинам число - 0 . Далее мы будем кодировать вершины через упорядоченный список кодировок потомков. Если такой список уже закодирован, то пронумируем вершину соответствующим числом, иначе сопоставим вершине и списку следующее число. И, значит, для того, чтобы сравнить $G$ и $H$ достаточно сравнить кодировки их корней. Если же в графе есть цикл, то мы можем закодировать все вершины на нем, если считать, что текущая вершина корень поддерева. далее просто нужно быстро сравнить на равенство два пронумерованных цикла. Для достижения необходимой асимптотики $O(n+m)$ нужно использовать структуру, которая за длину списка добавляет его и за длину проверяет его наличие, а также нужно уметь сравнивать две занумерованные окружности на равенство за количество чисел на них. Это предлагается сделать самостоятельно. # Erdös-Renyi random graphs and Graph Isomorphism problem A. M. Raigorodsky, M. Tikhomirov, A. Golovanov ## 1 Block \#1 ### 1.1 Erdös-Renyi random graph A graph $G=(V, E)$ consists of a set of vertices $V$ together with a set of edges $E$, where each edge represents an unordered pair of vertices. One can easily see that any graph can contain no more than $\binom{n}{2}=\frac{n(n-1)}{2}$ edges. Fix an arbitrary number $p \in(0,1)$ and denote by a probability of a graph (in the Erdös-Renyi probability space) with $n$ vertices $\{1,2, \ldots, n\}$ and $e$ edges the value of the following expression $$ \mathrm{P}(G)=\mathrm{P}_{p}(G):=p^{e}(1-p)^{\frac{n(n-1)}{2}-e} $$ A probability of a family (or of a property, which is the same) of graphs with vertices numbers $1,2, \ldots, n$ is a sum of probabilities of all graphs of this family. If $\Omega$ stands for the set of all graphs with $n$ vertices then $P(\Omega)=1$. According to the definition above, this is a theorem which is to be proved, and here is the proof: $$ \left.P(\Omega)=\sum_{k=0}^{\binom{n}{2}}\binom{n}{2}\right) p^{k}(1-p)^{\binom{n}{2}-k}=(p+(1-p))^{\binom{n}{2}}=1 $$ We say that a random variable is any function defined on the set of all graphs with vertices numbered $1,2, \ldots, n$. For example, the number of edges in a graph is a random variable. Let a random variable $Y$ take $k$ distinct values $y_{1}, \ldots, y_{k}$. Define a mathematical expectation of $Y$ as its "weighted average" $$ \mathrm{E} Y:=\sum_{s=1}^{k} y_{s} \mathrm{P}\left(Y^{-1}\left(y_{s}\right)\right) $$ where $Y^{-1}\left(y_{s}\right)$ stands for the set of all graphs $G$ with $Y(G)=y_{s}$ hold. The last probability in the expression above is denoted by $\mathrm{P}\left(Y=y_{s}\right)$. We call tow events $A$ and $B$ independent (of each other) if $\mathrm{P}(A \cap B)$ (the probability that both $A$ and $B$ happen) $=P(A) \cdot P(B)$. We call that two random variables $f$ and $g$ are independent (of each other) if events $f=x$ and $g=y$ are independent for every pair of $x$ and $y$. NB: this definition is correct only for discrete random variables, that is, for random variables which take no more than countable number of values. Problem 1. Let a graph $H$ consist of $n$ vertices and $m$ edges. What is the probability that $G(n, p)=H$, i.e. after generating a random graph according to the definition above the exact copy of $H$ is obtained? Problem 2. What is the probability of $G(n, p)$ having exactly $m$ edges? Problem 3. What is the probability of the event that $G(n, p)$ contains no isolated vertices (that is, vertices which are not connected with any other vertices by any edge)? Let a random variable $f$ take reals (maybe a countable number of them), and the probability of the event $f=x$ equals $p_{x}$. We say that the mathematical expectation of $f$ is the value of the expression $\sum_{x} x \cdot p_{x}$. We denote the expectation of $f$ by $\mathrm{E} f$. Problem 4. Prove the following: a) $\mathrm{E}[c \cdot f]=c \cdot \mathrm{E} f$ if $c$ is constant; b) $E[f+g]=\mathrm{E} f+\mathrm{E} g$ (even if $f$ and $g$ are not independent!); c) if variables $f$ and $g$ are independent then $\mathrm{E}[f \cdot g]=E f \cdot E g$; Problem 5. Find the expectation of the following variables: a) the number of edges of $G(n, p)$; b) the degree of any vertex of $G(n, p)$ (since the definition of $G(n, p)$ is symmetric, one can choose any arbitrary vertex); c) the number of triangles of $G(n, p)$; d) the number of Hamilton paths and cycles of $G(n, p)$ (a Hamilton path/cycle is any path/cycle which visits each vertex exactly once); e) the number of cliques (that is, complete subgraphs) of $k$ vertices. A variance $\mathrm{D} f$ of a random variable $f$ is $\mathrm{E}\left[(f-\mathrm{E} f)^{2}\right]$. Problem 6. Prove the following: a) $\mathrm{D}[c \cdot f]=c^{2} \cdot \mathrm{D} f$ if $c$ is constant; b) $\mathrm{D} f=\mathrm{E}\left[f^{2}\right]-(\mathrm{E} f)^{2}$; c) if $f$ and $g$ are independent then $D[f+g]=\mathrm{D} f+\mathrm{D} g$. Problem 7. Find the variance of the degree of any vertex of $G(n, p)$. Problem 8. Find the variance of the number of triangles $G(n, p)$ (be careful with the dependence of random variables!). ### 1.2 Algorithms on graphs and their complexities Assume that we have a description of any graph and we want to check if one of its properties (for example, connectivity) is hold. Consider the following algorithm: Depth-first search (DFS). procedure dfs (vertex $v$ ): $\operatorname{used}_{v}:=1$ for each vertex $u$ - a neighbor of $v$ : if used $_{u}=0$ then call $\operatorname{dfs}(u)$ function is_connected (graph $G$ ): used $_{v}=0$ for all vertices $v$ dfs(arbitrary vertex of $G$ ) if used $_{v}=1$ holds for all $v \in G$, return YES otherwise return NO The description above is represented by a pseudocode. For the better explanation of the algorithm we recommend you to run it in your mind or on paper for some small input. Please notice that the procedure $\mathrm{df} \mathbf{s}$ is recursive, that is, it can be called from itself, and once the last call is finished, the workflow returns to the previous call. Problem 9. Prove that DFS checks connectivity correctly. Problem 10. Prove that the algorithm performs $O(n+m)$ operations (no more than $c(n+m)$ for some constant $c$ which doesn't depend on the input graph) for a graph with $n$ vertices and $m$ edges. Problem 11. Modify this algorithm for obtaining the number of connected components of a graph. Problem 12. An edge of a graph is called a bridge if its removal increases the number of connected components. Describe an algorithm of obtaining all bridges in a graph with the following number of actions performed: a) $O\left((n+m)^{2}\right)$; b) $O(n(n+m))$; c) $\left(^{*}\right)(n+m)$. Consider another solution of the initial problem: Breadth-first search (BFS). Pseudocode: procedure bfs (vertex $v$ ): queue $=\{\mathrm{v}\} / /$ a queue of vertices $\operatorname{used}_{v}:=1$ while queue $\neq \varnothing$ : vertex $w=$ the first element of queue remove $w$ from the beginning of queue for each vertex $u$ which is a neighbor of $w$ : if used $_{u}=0$ : $\operatorname{used}_{u}:=1$ add $u$ to the end of queue function is_connected (graph $G$ ): $\operatorname{used}_{v}:=0$ for all vertices $v / /$ used is of size $\mathrm{n}$ bfs(arbitrary vertex of $G$ ) if used $_{v}=1$ for all vertices $v \in G$ then return YES otherwise return NO Problem 13. Prove that the algorithm is correct and find its complexity. Problem 14. Modify the algorithm for obtaining the lengths of the shortest paths from a fixed vertex $v$ to each other. Problem 15. Assume that each edge $e$ of a graph $G$ has its own length which is a nonnegative real and is denoted by $w(e)$. Build an algoritm of obtaining the lengths of all shortest paths from $v$ to each other vertex with the following number of actions performed: a) $O\left(n^{2}+m\right)$; b) (*) $O(n+m \log n)$. ### 1.3 Graph Isomorphism problem Two graphs $G=\left(V_{G}, E_{G}\right)$ and $H=\left(V_{H}, E_{H}\right)$ are called isomorphic (to each other) once there is such bijection $\pi: V_{G} \rightarrow V_{H}$ between the sets of their vertices that every unordered pair of vertices $v u$ occurs in $E_{G}$ iff $\pi(v) \pi(u)$ occurs in $E_{H}$. Such bijection $\pi$ is called an isomorphism of graphs $G$ and $H$. Problem 16. Draw all possible graphs (modulo isomorphism) with 3 or 4 vertices. Define an invariant as any function $f$ which for every graph $G$ returns some value $f(G)$ in such a way that $f(G)=f(H)$ for any isomorphic graphs $G$ and $H$. It's obvious that $f(G) \neq f(H)$ implies that $G$ and $H$ are not isomorphic, while the inverse doesn't have to hold (that is, $f(G)=f(H)$ can be true for non-isomorphic graphs). Define a lexicographic or alphabetical comparison of two sequences as follows: - the empty sequence is the least; - if the first element of $A$ is less than the first element of $B$ then $A1$. However, there are fast algorithms to the GI problem for a variety of cases. Problem 19. Provide algorithms with $O(n+m)$ time complexity to the problem for the following cases: a) all degrees are no more than 2 ; b) graphs don't contain cycles; c) (*) each of the graphs contains exactly one simple cycle. In assumption that graphs are generated randomly it makes sense to use some simple algorithms and estimate the probability of error. Our current goal is to analyze such probability for the following algorithm: ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-016.jpg?height=269&width=1295&top_left_y=1578&top_left_x=108) Problem 20. Describe how to implement the algorithm above for $O(n+m)$. Problem 21. Consider a pair of vertices $a$ and $b$. Prove that $l_{a}=l_{b}$ holds iff after $a$ 's and $b$ 's removal the list of $a$ 's ex-neighbors' degrees and the list of $b$ 's ex-neighbors' degrees are the same (modulo permutation). Problem 22. Prove that the decline probability doesn't exceed $n^{2} \cdot \mathrm{P}\left(l_{a}=l_{b}\right)$. Thus, once we show that $\mathrm{P}\left(l_{a}=l_{b}\right)$ is small, the decline probability will be small as well. It follows from the 21-st problem that the following probability is to be estimated: generate $G=G(n-2, p)$ and pick two sets $N_{a}$ and $N_{b}$, including each vertex of $G$ independently with probability $p$, then the set $N_{a}$ of degrees coincides with $N_{b}$. ## 2 Block \#2 ### 2.1 Combinatorial values and their asymptotics Definitions. We say that - $f(n)=o(g(n))$, if $\lim _{n \rightarrow \infty} \frac{f(n)}{g(n)}=0$. - $f(n)=O(g(n))$, if $\lim _{n \rightarrow \infty} \frac{f(n)}{g(n)}<\infty$ (that is, the ratio is upper-bounded by some real number). - $f(n)=\omega(g(n))$, if $\lim _{n \rightarrow \infty} \frac{f(n)}{g(n)}=\infty$. - $f(n)=\Omega(g(n))$, if $\lim _{n \rightarrow \infty} \frac{f(n)}{g(n)}>0$. - $f(n)=\Theta(g(n))$, if $0<\lim _{n \rightarrow \infty} \frac{f(n)}{g(n)}<\infty$. - $f(n) \sim g(n)$, if $\lim _{n \rightarrow \infty} \frac{f(n)}{g(n)}=1$. In some of following problems you'll be offered to "find an asymptotic" for several functions. That means that for each of the functions $f(n)$ you are asked to find any function $g(n)$ is its closed form (informally, there are only additions, subtractions, multiplications, divisions, powers which occur in the definition of this function, and it also doesn't contain skipped parts behind "...", " $\Sigma$ ", " $\Pi$ " etc.) such that $f(n) \sim g(n)$. Problem 23. Prove that $\frac{2^{n}}{n+1} \leqslant\binom{ n}{\left\lfloor\frac{n}{2}\right\rfloor} \leqslant 2^{n}$. Problem 24. Fund an asymptotic for $\sqrt[n]{\binom{n}{\left\lfloor\frac{n}{2}\right\rfloor}}$. Hint what does $\sqrt[n]{\frac{1}{n+1}}$ tend to for $n \rightarrow \infty$ ? Problem 25. Show that $\frac{3^{3 n}}{3 n+1} \leqslant 2^{2 n}\binom{3 n}{n} \leqslant 3^{3 n}$. Hint: find a combinatorial meaning for $3^{3 n}$ and $2^{2 n}\binom{3 n}{n}$. Problem 26. Find an asymptotic for $\sqrt[n]{\binom{3 n}{n}}$. Problem 27. Let $0<\alpha<1$. Show that $\binom{n}{\alpha n}=\left(\alpha^{-\alpha}(1-\alpha)^{\alpha-1}+o(1)\right)^{n}$. The following important formula is given without a proof: Stirling's formula. $n!\sim \sqrt{2 \pi n}\left(\frac{n}{e}\right)^{n}$ (here $e=\lim _{n \rightarrow \infty}\left(1+\frac{1}{n}\right)^{n}-$ the second great limit). Problem 28. Find an asymptotic for $\binom{2 n}{n}$. Problem 29. Find an asymptotic for $\binom{n^{2}}{n}$. Problem 30. Find an asymptotic for $(2 n-1)!!=1 \cdot 3 \cdot \ldots \cdot(2 n-1)$. Problem 31. Show that the number of non-isomorphic graphs with $n$ vertices equals $\Omega\left(\frac{1}{\sqrt{2 \pi n}}\left(\frac{e 2^{(n-1) / 2}}{n}\right)\right.$ Definition. A Bernoulli random variable with parameter $p$ is a variable which equals 1 with probability $p$ and 0 with the remaining probability $1-p$. Denote by $b(k, n, p)$ the probability that among $n$ independent Bernoulli random variables of $p$ exactly $k$ equal 1. Problem 32. Find the formula for $b(k, n, p)$. Problem 33. Find an asymptotic for $\max _{k} b(k, n, p)$ under the condition that $p=\omega(1 / n)$. Hint: what is the optimal value of $k$ ? Problem 34. Let $p=\omega(1 / n)$. Show that the probability that the degrees of two fixed vertices of $G(n, p)$ coincide is $O(1 / \sqrt{p n})$. Chernoff's bound. If $X$ is the sum of $n$ independent Bernoulli random variables of $p$ then $\mathrm{P}(|X-p n|>$ $p n \delta) \leq 2 e^{-\delta^{2} p n / 3}$. Problem 35. Under the condition that $p=\omega(1 / n)$ show that $|b(k, n, p)-b(k-1, n-1, p)|=$ $O(\sqrt{\ln n} / p n)$. a) Represent $Y=b(k, n, p)-b(k-1, n-1, p)$ as $\delta b(k, n, p)$ where $\delta$ is finite. b) Estimate $|Y|$ straightforwardly for small values of $|\delta|$ and using Chernoff's bound for a bigger values of $|\delta|$. Find an optimal value between "small" and "big" values of $|\delta|$. Problem 36. Under the condition of $p=\omega(1 / n)$ prove that: a) $|b(k, n, p)-b(k, n-1, p)|=O(\sqrt{\ln n} / p n)$. b) $|b(k-1, n-1, p)-b(k, n-1, p)|=O(\sqrt{\ln n} / p n)$. Problem 37. Let $p=\omega(\ln n / n),|k-p n|=o(\sqrt{p n / \ln n})$. Show that $b(k, n, p)=\Theta(1 / \sqrt{p n})$. ### 2.2 Edge martingale and intervals containing the degrees of a random graph Call a conditional probability of an event $A$ with respect to the event $B$ the value of $P(A \cap B) / P(B)$. A conditional expectation of $f$ wrt the event $B$ is the value of $\mathrm{E}[F \mid B]=\sum_{x} x \cdot \mathrm{P}(A=x \mid B)$. Informally, characteristics which are conditional under $B$ are just averages over the outcomes of $B$. Let $f$ and $g$ be random variables. Then call a conditional expectation of $f$ wrt $g$ (and denote it by $\mathrm{E}[f \mid g])$ the random variable which equals $x$ with the probability of $\sum_{E[f \mid g=z]=x} \mathrm{P}[g=z]$. In other words, the conditional expectation of $f$ wrt $g$ is such variable $h$ that can be generated in the following way: generate the value $z$ of $g$ then getting the expectation of $f$ wrt the event $g=z$ and return as $h$ 's outcome. Problem 38. Prove that is $c$ is constant then $\mathrm{E}[f \mid c]=\mathrm{E} f$. Problem 39. Find $\mathrm{E}[f \mid f]$ ? Problem 40. Show that $\mathrm{E}[f+g \mid h]=\mathrm{E}[f \mid h]+\mathrm{E}[g \mid h]$. Hint: note that both left hand side and right hand side contain random variables, so they have to be proved to be equal to each other for each outcome. Problem 41. Show that $\mathrm{E}[\mathrm{E}[f \mid g]]=\mathrm{E} f$. Return to the event we aim to estimate (the end of the block \#1). If degrees of $N_{a}$ coincide with degrees of $N_{b}$ then inside every interval the numbers of $N_{a}$ 's and $N_{b}$ 's representers coincide as well. Consider an arbitrary interval $I=\left[D_{0}, D_{0}+R-1\right]$. Denote by $f$ the number of vertices of $G(n, p)$ with degree from $I$. Enumerate all possible edges of a graph with $n$ vertices from 1 to $\binom{n}{2}$ in the following order: $\{1,2\}$, $\{1,3\}, \ldots,\{2,3\}, \ldots$ Let $f$ be any characteristic of a graph $G(n, p)$ (for example, the above defined $\mathrm{f}$ ). Denote by $C_{i}$ the random variable being equal to 1 if the $i$-th edge occurs in $G(n, p)$ and 0 otherwise. Call an edge martingale $f$ the sequence of random variables $f_{0}, \ldots, f_{\binom{n}{2}}$, where $f_{i}=E\left[f \mid C_{1}, \ldots, C_{i}\right]$. Informally saying, the value of $f_{i}$ tries to "guess" the value of $f$ using the information of presences/absences of the first $i$ edges. Actually we can define martingales in common case. Martingale is any sequence of random variables $f_{0}, \ldots, f_{m}$ such that $f_{i-1}=E\left[f \mid f_{i-1}, \ldots, f_{0}\right]$ for each $i$. Problem 42. Let $Y_{1}, \ldots, Y_{m}$ be independent random variables and each of them takes values $+1,-1$ with the equal probabilities. Let $X_{i}=Y_{1}+\cdots+Y_{i}$. Show that the sequence of $X_{1}, \ldots, X_{m}$ is a martingale. Problem 43. Show that in any martingale expectations of all random variables are the same. Problem 44. (*) Show that the edge martingale is a martingale. Azuma's inequality. Let $X_{0}, \ldots, X_{n}$ be a martingale. Let $\left|X_{i}-X_{i-1}\right|$ be no more than $c_{i}$ for each outcome. Then $$ \mathrm{P}\left[X_{n}-X_{0} \geq t\right] \leq e^{-t^{2} / 2 \sum_{i=1}^{n} c_{i}^{2}} $$ Consider an edge $i=\{a, b\}$ (and $a0$. Problem 51. (**) Provide an algorithm to the GI problem for Erdös-Renyi graphs with the polynomial time complexity (over the number of vertices and edges) and an exponentially small decline probability for some non-trivial interval of $p$-s. ## Точки Ферма, прямые Эйлера и кое-что еще Н.Белухов, А.Заславский О.Заславский, П.Кожевников, Д.Креков ${ }^{1}$ На судъбу полагаясь доверчиво, Абсолютно не сущую суть Я ленивой рукою вычерчивал Без претензий на что-нибудъ. А.Великий ## Обозначения и вводные факты Для данного неравностороннего треугольника $A B C$ будем обозначать через $M, H, O, I, I_{a}, I_{b}, I_{c}$ центр тяжести, ортоцентр, центр описанной окружности, центр вписанной окружности, центры вневписанных окружностей, касающихся сторон $B C, C A, A B$ соответственно. Точки $O, M, H$ лежат на одной прямой, которая называется прямой Эйлера, причем $M$ делит отрезок $O H$ в отношении $1: 2$. Середины сторон треугольника, основания его высот и середины отрезков $A H, B H, C H$ лежат на одной окружности, которая называется окружноствю Эйлера или окружностью девяти точек. Центром этой окружности является середина $E$ отрезка $O H$. Для любой точки $X$ прямые, симметричные прямым $A X, B X, C X$ относительно биссектрис углов $A, B, C$ соответственно, пересекаются в одной точке $X^{*}$ (или параллельны), которая называется изогонально сопряженной $X$. Точка $L=M^{*}$ называется точкой Лемуана. Любой треугольник можно аффинным преобразованием перевести в правильный. При обратном преобразовании описанная и вписанная окружности правильного треугольника переходят в эллипсы, которые называются описанным и вписанным эллипсами Штейнера. ## 1 Точки Торричелли, Наполеона и Аполлония Пусть $A B T_{c}, B C T_{a}, C A T_{b}$ - правильные треугольники, построенные во внешнюю от $\triangle A B C$ сторону, $N_{a}, N_{b}, N_{c}$ - центры этих треугольников. Аналогично, пусть $A B T_{c}^{\prime}, B C T_{a}^{\prime}, C A T_{b}^{\prime}$ - правильные треугольники, построенные во внутреннюю от $\triangle A B C$ сторону, $N_{a}^{\prime}, N_{b}^{\prime}, N_{c}^{\prime}$ - центры этих треугольников. 1.1. Докажите, что ${ }^{2}$ прямые $A T_{a}, B T_{b}, C T_{c}$ пересекаются в одной точке и прямые $A T_{a}^{\prime}, B T_{b}^{\prime}, C T_{c}^{\prime}$ пересекаются в одной точке.[^0] Полученные точки называются точками Торричелли или точками Ферма и обозначаются соответственно $T_{1}$ и $T_{2}$. 1.2. Прямые $A N_{a}, B N_{b}, C N_{c}$ пересекаются в одной точке и прямые $A N_{a}^{\prime}$, $B N_{b}^{\prime}, C N_{c}^{\prime}$ пересекаются в одной точке. Полученные точки называются точками Наполеона и обозначаются соответственно $N_{1}$ и $N_{2}$. 1.3 . a) Треугольники $N_{a} N_{b} N_{c}$ и $N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}$ - правильные. b) Центры этих треугольников совпадают с $M$. c) Разность их площадей равна площади $\triangle A B C$. 1.4. Пусть углы $\triangle A B C$ меньше $120^{\circ}$. a) Стороны $\triangle A B C$ видны из точки $T_{1}$ под равными углами. b) Сумма расстояний от вершин $\triangle A B C$ до точки $T_{1}$ меньше, чем до любой другой точки плоскости. Для $\triangle A B C$ с углами, не превосходящими $120^{\circ}$, сумма расстояний от вершин до точки $T_{1}$ называется расстоянием Ферма. 1.5. Если $\triangle A B C$ и $\triangle A^{\prime} B^{\prime} C^{\prime}$ удовлетворяют условиям $\angle A+\angle A^{\prime}=\angle B+\angle B^{\prime}=$ $\angle C+\angle C^{\prime}=120^{\circ}$, то отношения их расстояний Ферма к радиусам описанных окружностей совпадают. 1.6 . a) Точки $T_{1}, N_{a}^{\prime}, N_{b}^{\prime}, N_{c}^{\prime}$ лежат на одной окружности. b) Точки пересечения этой окружности с отрезками $A T_{a}, B T_{b}, C T_{c}$, отличные от $T$, делят эти отрезки в отношении $1: 2$. c) Сформулируйте и докажите аналогичное утверждение для точек $T_{2}, N_{a}$, $N_{b}, N_{c}$. 1.7. a) $C T_{1} \perp N_{a} N_{b}$. b) $C N_{1} \perp T_{a} T_{b}$. c) Сформулируйте и докажите аналогичные утверждения для точек $T_{2}, N_{2}$. 1.8 . a) Проекции точек $T_{1}^{*}, T_{2}^{*}$ на стороны $\triangle A B C$ образуют правильные треугольники. b) $T_{i}^{*} A \cdot B C=T_{i}^{*} B \cdot C A=T_{i}^{*} C \cdot A B, i=1,2$. Точки, обладающие указанными свойствами, называются точками Аполлонил. Таким образом точки Торричелли и Аполлония изогонально сопряжены. 1.9. Пусть точка $T_{1}$ лежит внутри $\triangle A B C$. Тогда три бильярдных шара, выпущенных из нее одновременно с равными скоростями в направлениях, противоположных вершинам, после отражения от соответствующих сторон столкнутся в одной точке. 1.10. Прямая $T_{1}^{*} T_{2}^{*}$ проходит через $O$. 1.11. Прямые $T_{1}^{*} T_{2}^{*}, T_{1} T_{2}, N_{1} N_{2}$ пересекаются в точке $L$. 1.12. Прямые $T_{1} N_{1}, T_{2} N_{2}$ пересекаются в точке $O$. 1.13. Прямые $T_{1} N_{2}, T_{2} N_{1}$ пересекаются в точке $E$. 1.14. Прямые $T_{1} T_{2}^{*}, T_{2} T_{1}^{*}$ пересекаются в точке $M$. 1.15. Середина $M_{T}$ отрезка $T_{1} T_{2}$ лежит на окружности Эйлера. ## 2 Прямые Эйлера и эллипсы Штейнера ${ }^{3}$ 2.1. Прямые Эйлера $\triangle A B T_{1}, \triangle B C T_{1}, \triangle C A T_{1}, \triangle A B T_{2}, \triangle B C T_{2}, \triangle C A T_{2}$ пересекаются в точке $M$. Пусть $M_{A}, M_{B}$ и $M_{C}$ - середины $B C, C A$ и $A B$ соответственно, а $l_{A}, l_{B}$ и $l_{C}$ - прямые Эйлера $\triangle A T_{1} T_{2}, \triangle B T_{1} T_{2}$ и $\triangle C T_{1} T_{2}$. 2.2. Прямые $l_{A}, l_{B}, l_{C}$ параллельны соответственно $M_{A} M_{T}, M_{B} M_{T}, M_{C} M_{T}$. 2.3. Прямые Эйлера всех десяти треугольников, вершины которых принадлежат множеству $\left\{A, B, C, T_{1}, T_{2}\right\}$, пересекаются в $M$. 2.4. Пусть $T_{3}$ - такая точка, что $\triangle T_{1} T_{2} T_{3}$ - правильный. Тогда прямые, проходящие через $A, B, C$ и параллельные соответственно $B T_{3}, C T_{3}, A T_{3}$, пересекаются в одной точке. Пусть $P$ - такая точка, что $\triangle T_{1} T_{2} P$ - правильный, ориентированный противоположно $\triangle A B C$. Пусть $Q$ - такая точка, что $\triangle T_{1} T_{2} Q$ - правильный, ориентированный так же, как $\triangle A B C$. 2.5. Прямая $l_{A}$ параллельна $C P$ и $B Q$, прямая $l_{B}$ параллельна $A P$ и $C Q$, прямая $l_{C}$ параллельна $B P$ и $A Q$. 2.6. Найдите геометрическое место таких точек $X$, что прямые, проходящие через $A, B, C$ и параллельные $B X, C X, A X$ соответственно, пересекаются в одной точке. 2.7. Точки $P$ и $Q$ лежат на описанном эллипсе Штейнера. 2.8 . a) Прямые, проходящие через $A, B, C$ и параллельные соответственно $l_{A}$, $l_{B}, l_{C}$, пересекаются в некоторой точке $R$. b) $R$ лежит на описанном эллипсе Штейнера. 2.9. У $\triangle A B C$ и $\triangle P Q R$ площади равны, а центры тяжести совпадают. 2.10. $M_{T}$ лежит на вписанном эллипсе Штейнера. 2.11. Прямая $T_{1} T_{2}$ является нормалью к вписанному эллипсу Штейнера.[^1] ## 3 Гипербола Кипера Пусть $A B C^{\prime}, B C A^{\prime}, C A B^{\prime}$ - подобные, одинаково ориентированные равнобедренные треугольники с основаниями $A B, B C, C A$. 3.1. a) Прямые $A A^{\prime}, B B^{\prime}, C C^{\prime}$ пересекаются в одной точке. b) Геометрическим местом точек пересечения является равносторонняя гипербола, проходящая через $A, B, C, M, H$. Она называется гиперболой Кипера. Обозначим через $X(\varphi)$ точку гиперболы Кипера, соответствующую углу $\varphi$ при основании равнобедренных треугольников. Причем угол $\varphi$ будем считать положительным, если треугольники построены во внешнюю от $\triangle A B C$ сторону, и отрицательным в противном случае. 3.2. Точка $X^{*}(\varphi)$ лежит на прямой $O L$. 3.3. a) Все прямые $X(\varphi) X(-\varphi)$ проходят через $L$. b) Все прямые $X(\varphi) X(\pi / 2-\varphi)$ проходят через $O$. c) Все прямые $X(\varphi) X(\pi / 2+\varphi)$ проходят через $E$. 3.4. Точки $X\left(\varphi_{1}\right), X\left(\varphi_{2}\right), X^{*}\left(\varphi_{3}\right)$ лежат на одной прямой тогда и только тогда, когда $\varphi_{1}+\varphi_{2}+\varphi_{3} \equiv 0(\bmod \pi)$. 3.5. a) Перпендикуляры из точек $A, B, C$ на прямые $B^{\prime} C^{\prime}, C^{\prime} A^{\prime}, A^{\prime} B^{\prime}$ соответственно пересекаются в одной точке. b) Эта точка лежит на гиперболе Кипера. ## 4 Кубика Нейберга 4.1. Пусть $\triangle U V W$ - правильный, $X$ - произвольная точка плоскости, $X_{u}$, $X_{v}, X_{w}$ - точки, симметричные $X$ относительно прямых $V W, W U, U V$ соответственно. a) Прямые $U X_{u}, V X_{v}, W X_{w}$ пересекаются в одной точке. b) Прямые Эйлера $\triangle X U V, \triangle X V W$ и $\triangle X W U$ пересекаются в одной точке. 4.2. Для точки $X$, не лежащей на описанной окружности $\triangle A B C$, утверждения предыдущей задачи выполняются тогда и только тогда, когда прямая $X X^{*}$ параллельна прямой Эйлера. Геометрическим местом таких точек $X$, что прямая $X X^{*}$ параллельна прямой Эйлера, является кривая третьего порядка, которая называется кубикой Нейберга. 4.3. Кубика Нейберга проходит через $A, B, C, O, H, I, I_{a}, I_{b}, I_{c}, T_{1}, T_{2}, T_{1}^{*}$, $T_{2}^{*}$. 4.4. a) Если кубика Нейберга содержит некоторую прямую, то она также содержит некоторую окружность. b) Если $A C=B C$, то кубика Нейберга состоит из биссектрисы угла $C$ и окружности с центром $C$ и радиусом $A B$. c) Если $\angle C=60^{\circ}$, то кубика Нейберга состоит из внешней биссектрисы угла $C$ и окружности, симметричной описанной окружности $\triangle A B C$ относительно прямой $A B$. d) Если $\angle C=120^{\circ}$, то кубика Нейберга состоит из внутренней биссектрисы угла $C$ и окружности, симметричной описанной окружности $\triangle A B C$ относительно прямой $A B$. е) Свойство из п.а) выполняется тогда и только тогда, когда прямая Эйлера параллельна одной из биссектрис треугольника (внутренней или внешней). В этом случае соответствующая биссектриса целиком принадлежит кубике Нейберга. f) Свойства из пп. а) и е) выполняются только в случаях, перечисленных в ппг. b), c), d). 4.5. Если точка $D$ лежит на кубике Нейберга $\triangle A B C$, то $A$ лежит на кубике Нейберга $\triangle B C D$ и аналогичные свойства выполнены для вершин $B, C$. ## Список литературы [1] Акопян А.В., Заславский А.А. Геометрические свойства кривых второго порядка. М.: МЦНМО, 2011. [2] N.Beluhov. Ten Concurrent Euler Lines. Unpublished, 2009. (Частично опубликовано как Ten Concurrent Euler Lines, http://forumgeom.fau.edu/ FG2009volume9/FG200924.pdf .) [3] A.Zaslavsky. Geometry of Kiepert and Grinberg-Myakishev Hyperbolas. http: //www.jcgeometry .org/Articles/Volume1/JCG2012V1pp65-71.pdf [4] D.Banerjee, S.Mukherjee. Neuberg Locus And Its Properties. http://www. jcgeometry.org/Articles/Volume2/JCG2013V2pp26-38.pdf ## Точки Ферма, прямые Эйлера и кое-что еще Н.Белухов, А.Заславский, О.Заславский, П.Кожевников, Д.Креков ${ }^{1}$ Решения ## 1 Точки Торричелли, Наполеона и Аполлония 1.1. Первое решение. Применяя теорему синусов к треугольникам $A T_{a} B$ и $A T_{a} C$, получаем $$ \sin \angle B A T_{a}: \sin \angle A B T_{a}=B T_{a}: A T_{a}=C T_{a}: A T_{a}=\sin \angle C A T_{a}: \sin \angle A C T_{a} $$ Следовательно, $\sin \angle B A T_{a}: \sin \angle C A T_{a}=\sin (\angle B+\pi / 3): \sin (\angle C+\pi / 3)$. Аналогично получаем, что $\sin \angle A C T_{c}: \sin \angle B C T_{c}=\sin (\angle A+\pi / 3): \sin (\angle B+\pi / 3)$ и $\sin \angle C B T_{b}: \sin \angle A B T_{b}=\sin (\angle C+\pi / 3): \sin (\angle A+\pi / 3)$. Значит, по теореме Чевы $A T_{a}, B T_{b}$ и $C T_{c}$ пересекаются в одной точке. Второе утверждение доказывается аналогично. Второе решение. Так как поворот на $60^{\circ}$ вокруг точки $A$ переводит $\triangle A B T_{b}$ в $\triangle A T_{c} C$, прямые $B T_{b}$ и $C T_{c}$ проходят через общую точку окружностей, описанных около $\triangle A B T_{c}$ и $\triangle C A T_{b}$. Использовав аналогичное рассуждение для точек $B$ и $C$, получаем, что прямые $A T_{a}, B T_{b}, C T_{c}$ и описанные окружности $\triangle A B T_{c}$, $\triangle B C T_{a}, \triangle C A T_{b}$ проходят через одну точку (рис.1). Аналогично доказывается вторая часть задачи. 1.2. Доказательство аналогично первому решению предыдущей задачи. 1.3. Первое решение. a) Применяя теорему косинусов к треугольнику $C N_{a} N_{b}$, получаем $N_{a} N_{b}^{2}=\frac{C A^{2}}{3}+\frac{C B^{2}}{3}-2 \frac{C A \cdot C B}{3} \cos (\angle C+\pi / 3)=\frac{C A^{2}+C B^{2}+A B^{2}}{6}+\frac{2 S_{A B C}}{\sqrt{3}}$. Очевидно, для других сторон получаем такие же выражения. Для второго треугольника доказательство аналогично. b) Следует из легко проверяемых векторных равенств $A \vec{N}_{b}+B \vec{N}_{c}+C \vec{N}_{a}=\overrightarrow{0}$ и $A \vec{N}_{b}^{\prime}+B \vec{N}_{c}^{\prime}+C \vec{N}_{a}^{\prime}=\overrightarrow{0}$. c) Следует из полученных в п.а) выражений для сторон правильных треугольников $N_{a} N_{b} N_{c}$ и $N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}$. Второе решение. a) Так как $A T_{1}$ - общая хорда окружностей $A T_{c} B T_{1}$ и $A T_{b} C T_{1}$, она перпендикулярна линии центров $N_{b} N_{c}$ этих окружностей. Аналогично $B T_{b} \perp N_{c} N_{a}$ и $C T_{c} \perp N_{a} N_{b}$. Значит $\triangle N_{a} N_{b} N_{c}$ - правильный, для $\triangle N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}$ доказательство аналогично.[^2] ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-027.jpg?height=1230&width=1179&top_left_y=263&top_left_x=468) Рис. 1 c) Заметим, что $\triangle A N_{b} N_{c} \simeq \triangle T_{1} N_{b} N_{c}, \triangle B N_{c} N_{a} \simeq \triangle T_{1} N_{c} N_{a}$ и $\triangle C N_{a} N_{b} \simeq$ $\triangle T_{1} N_{a} N_{b}$. Следовательно, $$ S_{N_{a} N_{b} N_{c}}=\frac{1}{2}\left(S_{A B C}+S_{A B N_{c}}+S_{B C N_{a}}+S_{C A N_{b}}\right) $$ Аналогично $$ S_{N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}}=\frac{1}{2}\left(-S_{A B C}+S_{A B N_{c}^{\prime}}+S_{B C N_{a}^{\prime}}+S_{C A N_{b}^{\prime}}\right) $$ 1.4. a) Следует из второго решения задачи 1.1. b) Пусть $X$ - произвольная точка плоскости, а поворот на $60^{\circ}$ вокруг $A$, переводящий $\triangle A B T_{b}$ в $\triangle A T_{c} C$, переводит $T_{1}$ и $X$ в $T^{\prime}$ и $Y$ соответственно. Тогда (рис.2) $A X+B X+C X=C X+X Y+Y T_{c} \geq C T_{c}=C T_{1}+T_{1} T^{\prime}+T^{\prime} T_{c}=A T_{1}+B T_{1}+C T_{1}$. Равенство достигается только при $X \equiv T_{1}$. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-028.jpg?height=1027&width=681&top_left_y=267&top_left_x=711) Рис. 2 (Неравенство $X A+X B \geq X T_{c}$ можно также получить из теоремы Птолемея для четырехугольника $A B T_{c} X$.) 1.5. Пусть $A T_{1}, B T_{1}$ и $C T_{1}$ повторно пересекают описанную около $\triangle A B C$ окружность $k$ в точках $A^{\prime \prime}, B^{\prime \prime}$ и $C^{\prime \prime}$ соответственно. Тогда $\triangle A^{\prime \prime} B^{\prime \prime} C^{\prime \prime}$ подобен $\triangle A^{\prime} B^{\prime} C^{\prime}$ и остается доказать, что расстояния Ферма для $\triangle A B C$ и $\triangle A^{\prime \prime} B^{\prime \prime} C^{\prime \prime}$ равны. Первое решение. Пусть $\triangle B^{\prime \prime} C^{\prime \prime} T_{a}^{\prime \prime}$ - правильный и построенный во внешнюю сторону от $\triangle A^{\prime \prime} B^{\prime \prime} C^{\prime \prime}$. Из решения предыдущей задачи следует, что расстояния Ферма равны $A T_{a}$ и $A^{\prime \prime} T_{a}^{\prime \prime}$ (рис.3). Пусть $B T_{a}, C T_{a}, B^{\prime \prime} T_{a}^{\prime \prime}$ и $C^{\prime \prime} T_{a}^{\prime \prime}$ повторно пересекают $k$ в точках $B_{1}, C_{1}, B_{1}^{\prime \prime}$ и $C_{1}^{\prime \prime}$. Счетом углов получаем, что $B C_{1}^{\prime \prime}\left\|C B_{1}^{\prime \prime}\right\| B^{\prime \prime} C_{1} \| C^{\prime \prime} B_{1}$. Следовательно, чертеж симметричен относительно серединного перпендикуляра $s_{A A^{\prime \prime}}$ к отрезку $A A^{\prime \prime}$. Значит, отрезки $A T_{a}$ и $A^{\prime \prime} T_{a}^{\prime \prime}$ симметричны относительно $s_{A A^{\prime \prime}}$, а их длины равны. Второе решение. Пусть окружность $k^{\prime}$, проходящая через $T_{1}$ и концентричная с $k$, повторно пересекает прямые $A A^{\prime \prime}, B B^{\prime \prime}$ и $C C^{\prime \prime}$ в точках $K_{a}, K_{b}$ и $K_{c}$ соответственно. Рассмотрим случай, когда точки $A, T_{1}, K_{a}, A^{\prime \prime} ; B, T_{1}, K_{b}$, $B^{\prime \prime}$ и $C, K_{c}, T_{1}, C^{\prime \prime}$ лежат на соответствующих прямых именно в таком порядке, остальные случаи разбираются аналогично. В силу симметрии $A T_{1}=K_{a} A^{\prime \prime}, B T_{1}=K_{b} B^{\prime \prime}$ и $C K_{c}=T_{1} C^{\prime \prime}$. Осталось доказать, что $T_{1} K_{a}+T_{1} K_{b}=T_{1} K_{c}$, но это следует из теоремы Птолемея для четырехугольника $T_{1} K_{a} K_{b} K_{c}$. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-029.jpg?height=759&width=1252&top_left_y=263&top_left_x=426) Рис. 3 1.6 . a) Очевидно, $N_{c}^{\prime}$ лежит на описанной окружности $k_{c}$ треугольника $A B T_{c}$, причем $N_{c}^{\prime} T_{c}$ - диаметр этой окружности. Так как $T_{1}$ тоже лежит на $k_{c}$, то $\angle N_{c}^{\prime} T_{1} C=\pi / 2$. С другой стороны, $M_{C} M: M_{C} C=M_{C} N_{c}: M_{C} T_{c}=1: 3$, т.е. $N_{c} M \| C T_{c}$. Следовательно, $N_{c} M$ - серединный перпендикуляр к $N_{c}^{\prime} T_{1}$ и $M N_{c}^{\prime}=M T_{1}$ (рис.4). b) Так как центром окружности $N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}$ является точка $M$ и $\angle N_{c}^{\prime} T_{1} C=\pi / 2$, вторая точка пересечения окружности $N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}$ с прямой $C T_{c}$ симметрична $N_{c}^{\prime}$ относительно $M$ (рис.4). Поскольку прямая $C M$ делит отрезок $N_{c}^{\prime} T_{c}$ в отношении $1: 3$, утверждение задачи следует из теоремы Менелая. 1.7 . a) См. второе решение задачи 1.3a). b) Применив теорему косинусов к треугольникам $A T_{b} N_{c}$ и $B T_{a} N_{c}$, получаем, $N_{c} T_{a}^{2}-N_{c} T_{b}^{2}=C T_{a}^{2}-C T_{b}^{2}$, что, очевидно, равносильно утверждению задачи. 1.8. a) Точки, симметричные $T_{1}^{*}$ относительно $A B, B C, C A$, образуют треугольник, для которого прямые $T_{1} A, T_{1} B, T_{1} C$ являются серединными перпендикулярами к сторонам. Следовательно, углы этого треугольника равны $60^{\circ}$ (рис.5). b) Пусть $X, Y$ - проекции $T_{1}^{*}$ на $A C, B C$ соответственно. Тогда четырехугольник $C X T_{1}^{*} Y$ вписан в окружность с диаметром $C T_{1}^{*}$, т.е. $X Y=C T_{1}^{*} \cdot \sin \angle C$ и искомое утверждение следует из предыдущего пункта. 1.9. Прямые $A T_{1}, B T_{1}, C T_{1}$ проходят через точки, симметричные $T_{1}^{*}$ относительно $B C, C A, A B$ соответственно. Поэтому после отражения от сторон все три шара попадут в точку $T_{1}^{*}$, причем за одинаковое время (рис.5). (Следствие из этой задачи предлагалось на 28 Турнире городов, весна, 10-11.7. Прямые, симметричные $A T_{1}, B T_{1}, C T_{1}$ относительно противоположных сторон треугольника, пересекаются в одной точке.) ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-030.jpg?height=1014&width=724&top_left_y=268&top_left_x=690) Рис. 4 ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-030.jpg?height=629&width=723&top_left_y=1404&top_left_x=693) Рис. 5 1.10. Геометрическим местом точек $X$, удовлетворяющих условию $A X$ : $B X=A C: B C$, является окружность $\Omega_{a}$, перпендикулярная описанной окружности треугольника $A B C$. Так как точки $T_{1}^{*}, T_{2}^{*}$ являются общими точками этой и двух аналогичных окружностей $\Omega_{b}, \Omega_{c}$, они инверсны относительно описанной окружности треугольника. 1.11. Точки $T_{1}^{*}, T_{2}^{*}$ лежат на окружностях $\Omega_{a}, \Omega_{b}, \Omega_{c}$, центрами которых являются точки $S_{a}, S_{b}, S_{c}$ пересечения сторон треугольника с касательными к его описанной окружности в противоположных вершинах. Прямая $S_{a} S_{b} S_{c}$ является полярой точки $L$ относительно описанной окружности, следовательно, она перпендикулярна прямой $O L$. Но проходящая через $O$ прямая $T_{1}^{*} T_{2}^{*}$ также перпендикулярна линии центров. Два других утверждения задачи являются частными случаями задачи 3.3 . 1.12. Частный случай задачи 3.3 . 1.13. Частный случай задачи 3.3 . 1.14. Частный случай теоремы о трех парах изогональных точек (см. [1]). 1.15. Следует из задач 3.1, 3.2. ## 2 Прямые Эйлера и эллипсы Штейнера Во всех решениях этого раздела мы рассматриваем случай, когда $T_{1}$ лежит внутри $\triangle A B C$, а четырехугольник $A B C T_{2}$ - выпуклый. Другие случаи разбираются аналогично. 2.1. Покажем, что прямая Эйлера $\triangle B C T_{1}$ проходит через $M$. Для остальных треугольников доказательство аналогично. Пусть $M_{a}$ - центр тяжести $\triangle B C T_{1}$. Так как $N_{a}$ - центр описанной окружности $\triangle B C T_{1}$, прямая Эйлера этого треугольника совпадает с $M_{a} N_{a}$ (рис.6). ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-032.jpg?height=585&width=1136&top_left_y=759&top_left_x=489) Рис. 6 Так как точки $A, T_{1}$ и $T_{a}$ лежат на одной прямой, а $M, M_{a}$ и $N_{a}$ делят отрезки $M_{A} A, M_{A} T_{1}$ и $M_{A} T_{a}$ соответственно в отношении $1: 2$, эти точки также лежат на одной прямой. 2.2. Пусть $M_{a}$ и $O_{a}$ - центр тяжести и центр описанной окружности соответственно $\triangle A T_{1} T_{2}$. Поскольку $N_{b} N_{c}$ - серединный перпендикуляр к отрезку $A T_{1}$, а $N_{b}^{\prime} N_{c}^{\prime}$ - серединный перпендикуляр к $A T_{2}$, прямые $N_{b} N_{c}$ и $N_{b}^{\prime} N_{c}^{\prime}$ пересекаются в точке $O_{a}$ (рис.7). Обозначим через $x$ острый угол между прямыми $M O_{a}$ и $N_{b}^{\prime} N_{c}^{\prime}$. Так как $M$ - общий центр $\triangle N_{a} N_{b} N_{c}$ и $\triangle N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}$, то $\angle M N_{c} N_{b}=\angle M N_{b}^{\prime} N_{c}^{\prime}=$ $30^{\circ}$. Значит, четырехугольник $M O_{a} N_{c} N_{b}^{\prime}$ - вписанный и $$ \angle N_{b} N_{c} N_{b}^{\prime}=\angle O_{a} N_{c} N_{b}^{\prime}=180^{\circ}-\angle N_{b}^{\prime} M O_{a}=\angle M O_{a} N_{b}^{\prime}+\angle O_{a} N_{b}^{\prime} M=x+30^{\circ} $$ Заметим, что $\triangle A N_{b} T_{b}, \triangle A N_{c} B$ и $\triangle A N_{b}^{\prime} C$ - подобны и одинаково ориентированы. Поэтому $A$ - центр подобия, переводящего $\triangle N_{b} N_{c} N_{b}^{\prime}$ в $\triangle T_{b} B C$. Следовательно, $$ \angle T_{1} B C=\angle T_{b} B C=\angle N_{b} N_{c} N_{b}^{\prime}=x+30^{\circ} $$ Пусть точка $T^{\prime}$ симметрична $T_{1}$ относительно $M_{A}$. Тогда $\angle B T^{\prime} C=\angle B T_{1} C$ $=120^{\circ}$. Поэтому четырехугольник $B T^{\prime} C T_{2}$ вписанный и $$ \angle B T_{2} T^{\prime}=\angle B C T^{\prime}=\angle T_{1} B C=x+30^{\circ} $$ ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-033.jpg?height=1041&width=962&top_left_y=268&top_left_x=576) Рис. 7 Так как прямые $B T_{2}$ и $N_{a}^{\prime} N_{c}^{\prime}$ перпендикулярны, угол между прямыми $B T_{2}$ и $N_{b}^{\prime} N_{c}^{\prime}$ равен $30^{\circ}$. Отсюда, с учетом равенства $\angle B T_{2} T^{\prime}=x+30^{\circ}$ получаем, что $M O_{a} \| T^{\prime} T_{2}$. Так как $M_{A}$ и $M_{T}$ - середины $T_{1} T^{\prime}$ и $T_{1} T_{2}$ соответственно, то $M_{A} M_{T} \| T^{\prime} T_{2}$. Так как $M$ и $M_{a}$ делят отрезки $A M_{A}$ и $A M_{T}$ соответственно в отношении $2: 1$, то $M M_{a} \| M_{A} M_{T}$. Таким образом, $M O_{a}\left\|T^{\prime} T_{2}\right\| M_{A} M_{T} \| M M_{a}$. Следовательно, $M, M_{a}$ и $O_{a}$ лежат на одной прямой, $M$ лежит на $l_{A}$ и $l_{A} \| M_{A} M_{T}$. 2.3. Первое решение. Следует из задачи 2.1 и решения задачи 2.2 . Второе решение. Применив утверждение задачи $4.4 \mathrm{~d})$ к $\triangle A B T_{1}$ и точке $T_{2}$, получим, что $l_{A}, l_{B}$ и прямая Эйлера $\triangle A B T_{2}$ пересекаются в одной точке. Аналогично $l_{A}, l_{B}$ и прямая Эйлера $\triangle A B T_{1}$ пересекаются в одной точке. Отсюда по задаче 2.1 получаем, что $l_{A}$ и $l_{B}$ пересекаются в точке $M$. 2.4. В зависимости от ориентации $\triangle T_{1} T_{2} T_{3}$ существуют две точки $T_{3}$, удовлетворяющих условию: $T_{3} \equiv P$ или $T_{3} \equiv Q$. Обычно в геометрических задачах ориентация несущественна. Однако, в данной задаче доказательства для двух положений точки $T_{3}$ принципиально различны. Случай 1. $T_{3} \equiv P$. Пусть точка $A^{\prime}$ изогонально сопряжена $A$ относительно $\triangle T_{1} T_{2} P$, а $A^{\prime \prime}$ симметрична $A^{\prime}$ относительно $T_{1} T_{2}$ (рис.8). Тогда $\angle A^{\prime \prime} T_{1} T_{2}=\angle T_{2} T_{1} A^{\prime}=180^{\circ}-\angle P T_{1} A=\left(\right.$ так как $\angle A T_{1} C=120^{\circ}$ и ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-034.jpg?height=596&width=919&top_left_y=271&top_left_x=598) Рис. 8 $\left.\angle P T_{1} T_{2}=60^{\circ}\right)=\angle C T_{1} T_{2}$. Следовательно $T_{1}, A^{\prime \prime}$ и $C$ лежат на одной прямой. Аналогично $T_{2}, A^{\prime \prime}$ и $C$ лежат на одной прямой. Значит, $A^{\prime \prime} \equiv C$. Композиция симметрий относительно внутренней биссектрисы $\angle T_{1} P T_{2}$ и прямой $T_{1} T_{2}$ переводит прямую $A P$ в $C Q$, значит, $A P \| C Q$. Аналогично $B P \| A Q$ и $C P \| B Q$. Таким образом, если $T_{3} \equiv P$, то три прямые пересекаются в точке $Q$. Случай 2. $T_{3} \equiv Q$. Пусть прямые $A P$ и $B Q$ пересекаются в точке $X$, а $\triangle R Y B$ и $\triangle Z C P$ являются образами $\triangle A Q X$ при соответствующих параллельных переносах (рис.9). ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-034.jpg?height=794&width=897&top_left_y=1544&top_left_x=603) Рис. 9 Тогда точки $C, Y, Z$ и $R$ лежат на одной прямой, $A R\|B Q\| C P, A P\|B R\| C Q$ и $A Q\|B P\| C R$. Следовательно, если $T_{3} \equiv Q$, то три прямых пересекаются в точке $R$. 2.5. Первое решение. По задаче 2.2 прямые $l_{A}$ и $M_{A} M_{T}$ параллельны. По задаче 2.4 прямые $C P$ и $B Q$ параллельны. Так как $M_{T}$ - середина $P Q$, получаем, что $M_{A} M_{T}$ средняя линия полосы между $C P$ и $B Q$, т.е. эти три прямые параллельны . Второе решение. По задаче 4.1b) $P$ лежит на кубике Нейберга $\triangle A T_{1} T_{2}$. Так как $P$ и $C$ изогонально сопряжены, то по задаче $4.2 l_{A} \| C P$. 2.6. Пусть $X$ - такая точка, что прямые, проходящие через $A, B, C$ и параллельные соответственно $B X, C X, A X$, пересекаются в некоторой точке $Y$. Рассмотрим аффинное преобразование $\alpha$, переводящее $\triangle A B C$ в правильный $\triangle A^{\prime} B^{\prime} C^{\prime}$. Пусть $\alpha(X)=X^{\prime}$ и $\alpha(Y)=Y^{\prime}$. Тогда прямые, проходящие через $A^{\prime}$, $B^{\prime}, C^{\prime}$ и параллельные соответственно $B^{\prime} X^{\prime}, C^{\prime} X^{\prime}, A^{\prime} X^{\prime}$ пересекаются в $Y^{\prime}$. Рассмотрим поворот $\rho$, переводящий $A^{\prime}$ в $B^{\prime}, B^{\prime}$ в $C^{\prime}$ и $C^{\prime}$ в $A^{\prime}$. Пусть $\rho\left(Y^{\prime}\right)=Z^{\prime}$. Тогда каждый из углов $\angle X^{\prime} A^{\prime} Z^{\prime}, \angle X^{\prime} B^{\prime} Z^{\prime}$ и $\angle X^{\prime} C^{\prime} Z^{\prime}$ равен либо $60^{\circ}$, либо $120^{\circ}$. Следовательно, точки $X^{\prime}, Z^{\prime}, A^{\prime}, B^{\prime}$ и $C^{\prime}$ лежат на одной окружности. Таким образом, $X^{\prime}$ лежит на окружности, описанной около $\triangle A^{\prime} B^{\prime} C^{\prime}$. Соответственно, $X$ лежит на описанном эллипсе Штейнера $\triangle A B C$. 2.7. Следует из задач 2.4 и 2.6 . 2.8. Следует из задачи 2.5 и случая 2 задачи 2.4 . 2.9-2.11. Определим аффинное преобразование $\alpha$ и $\triangle A^{\prime} B^{\prime} C^{\prime}$ аналогично решению задачи 2.6. Пусть $\alpha(\triangle P Q R)=\triangle P^{\prime} Q^{\prime} R^{\prime}$. Аналогично решению задачи 2.6 получаем, что $\triangle A^{\prime} B^{\prime} C^{\prime}$ и $\triangle P^{\prime} Q^{\prime} R^{\prime}$ - равные правильные треугольники с общим центром. Значит, их площади равны, а центры тяжести совпадают. Сделав обратное преобразование, получим, что оба этих свойства сохраняются для $\triangle A B C$ и $\triangle P Q R$. Так как середина $P^{\prime} Q^{\prime}$ лежит на окружности $e^{\prime}$, вписанной в $\triangle A^{\prime} B^{\prime} C^{\prime}$ и $\triangle P^{\prime} Q^{\prime} R^{\prime}$, середина $P Q$ лежит на общем вписанном эллипсе Штейнера е $\triangle A B C$ и $\triangle P Q R$. Так как $P^{\prime} Q^{\prime}$ касается $e^{\prime}$ и касание сохраняется при аффинном преобразовании, то $P Q$ касается $e$ в точке $M_{T}$. Поскольку прямая $T_{1} T_{2}$ - перпендикуляр к $P Q$ в точке $M_{T}$, она является нормалью к $e$. ## 3 Гипербола Кипера 3.1 . a) Аналогично задачам 1.1 и 1.2 . b) Воспользуемся следующим, легко проверяемым утверждением (подробнее см. [1], [3]). Лемма. Пусть даны две точки $A, B$ и отображение $f$, которое переводит каждую прямую $l$, проходящую через $A$, в прямую $f(l)$, проходящую через $B$, сохраняя двойные отношения четверок прямых. Тогда геометрическим местом точек пересечения $l$ и $f(l)$ является коника, проходящая через $A$ и $B$ (если $f(A B)=A B$, коника распадается на $A B$ и еще одну прямую). Легко видеть, что отображение, переводящее прямую $A A^{\prime}$ в $B B^{\prime}$, удовлетворяет условию леммы. Следовательно, геометрическим местом точек пересечения этих прямых будет проходящая через $A$ и $B$ коника. Если треугольники $A B^{\prime} C$ и $C A^{\prime} B$ с углами при основании, равными $\angle C$, построены во внутреннюю сторону, то прямые $A A^{\prime}$ и $B B^{\prime}$ совпадают с $A C$ и $B C$ соответственно, поэтому точка $C$ также лежит на этой конике. Кроме того, коника, очевидно, проходит через точки $M$ и $H$ и, следовательно, является равносторонней гиперболой. Примечание. Поскольку коника однозначно определяется пятью точками, геометрическим местом точек пересечения прямых $A A^{\prime}$ и $C C^{\prime}$ будет та же гипербола. Отсюда можно получить еще одно доказательство п.а). 3.2. При изогональном сопряжении описанная около треугольника коника переходит в прямую. Поскольку гипербола Кипера проходит через $M$ и $H$, сопряженная прямая совпадает с прямой $O L$. Заметим, что инверсные относительно описанной окружности точки $T_{1}^{*}, T_{2}^{*}$ при изогональном сопряжении переходят в противоположные точки гиперболы. Следовательно, точка $M_{T}$ является центром гиперболы Кипера и, значит, лежит на окружности Эйлера. Отсюда и из задачи 2.9 получаем, что точка $R$ лежит на описанной окружности $\triangle A B C$. Обозначим через $A^{\prime}(\varphi), B^{\prime}(\varphi), C^{\prime}(\varphi)$ вершины равнобедренных треугольников с углом при основании $\varphi$ так, что прямые $A A^{\prime}(\varphi), B B^{\prime}(\varphi)$ и $C C^{\prime}(\varphi)$ пересекаются в точке $X(\varphi)$. 3.3. Первое решение. a) Если проецировать гиперболу Кипера из точки $A$ на серединный перпендикуляр к $B C$, то точки $X(\varphi)$ и $X(-\varphi)$ перейдут в точки $A^{\prime}(\varphi)$ и $A^{\prime}(-\varphi)$, симметричные относительно $B C$. Поэтому отображение гиперболы на себя, переводящее $X(\varphi)$ в $X(-\varphi)$, сохраняет двойные отношения. Также сохраняет двойные отношения отображение, переводящее точку $X$ во вторую точку пересечения гиперболы с прямой $X L$. Поэтому для доказательства утверждения достаточно найти три значения $\varphi$, при которых эти отображения совпадают. При $\varphi=\angle A$ точка $X(-\varphi)$ совпадает с $A$, а $X(\varphi)$ лежит на симедиане из $A$. Аналогично получаем совпадение отображений при $\varphi=\angle B$ и $\varphi=\angle C$. Два других пункта доказываются аналогично. Второе решение. а), b) Следует из задачи 3.4. с) Аналогично первому решению задачи 3.4. Воспользуемся тем, что $A E$ проходит через $A^{\prime}\left(90^{\circ}-\angle A\right)$ и аналогичными утверждениями для $B E$ и $C E$. Частными случаями этой задачи при $\varphi=\pi / 3$ и $\varphi=\pi / 6$ являются утверждения задач 1.11-1.13. 3.4. Первое решение. (Н. Белухов) Используем следующую лемму. Лемма. Пусть $\varphi_{1}+\varphi_{2}=\angle A$. Тогда прямые $A A^{\prime}\left(\varphi_{1}\right)$ и $A A^{\prime}\left(\varphi_{2}\right)$ симметричны относительно биссектрис угла $A$. Доказательство. Пусть $L_{A}$ и $L_{A}^{\prime}$ - середины соответственно дуги $B C$, не содержащей $A$, и дуги $B A C$ окружности $k$, описанной вокруг $\triangle A B C$ (рис.10). Тогда $B L_{A}$ и $B L_{A}^{\prime}$ биссектрисы $\angle A^{\prime}\left(\varphi_{1}\right) B A^{\prime}\left(\varphi_{2}\right)$ в $\triangle B A^{\prime}\left(\varphi_{1}\right) A^{\prime}\left(\varphi_{2}\right)$. Следовательно, окружность $k$, описанная вокруг $\triangle B L_{A} L_{A}^{\prime}$ является окружностью Аполло- ния, соответствующей вершине $B$, для $\triangle B A^{\prime}\left(\varphi_{1}\right) A^{\prime}\left(\varphi_{2}\right)$. Поэтому $A L_{A}$ и $A L_{A}^{\prime}-$ биссектрисы $\angle A^{\prime}\left(\varphi_{1}\right) A A^{\prime}\left(\varphi_{2}\right)$. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-037.jpg?height=986&width=697&top_left_y=426&top_left_x=711) Рис. 10 Пусть теперь $X\left(\varphi_{1}\right) X\left(\varphi_{2}\right)$ пересекает $A B$ в точке $Y$. Необходимым и достаточным условием того, что точка $X^{*}\left(\varphi_{3}\right)$ лежит на прямой $X\left(\varphi_{1}\right) X\left(\varphi_{2}\right)$ будет совпадение точек пересечения $A X^{*}\left(\varphi_{3}\right)$ и $B X^{*}\left(\varphi_{3}\right)$ с $X\left(\varphi_{1}\right) X\left(\varphi_{2}\right)$. Это равносильно тому, что двойное отношение прямых $A Y$, $A X\left(\varphi_{1}\right), A X\left(\varphi_{2}\right)$ и $A X^{*}\left(\varphi_{3}\right)$ равно двойному отношению $B Y, B X\left(\varphi_{1}\right), B X\left(\varphi_{2}\right)$ и $B X^{*}\left(\varphi_{3}\right)$. Прямые $A Y, A X\left(\varphi_{1}\right), A X\left(\varphi_{2}\right)$ и $A X^{*}\left(\varphi_{3}\right)$ пересекают серединный перпендикуляр к $A C$ в точках $A^{\prime}(-\angle A), A^{\prime}\left(\varphi_{1}\right), A^{\prime}\left(\varphi_{2}\right)$ и (по лемме) $A^{\prime}\left(\angle B-\varphi_{3}\right) \equiv$ $A^{\prime}\left(\angle B+\varphi_{1}+\varphi_{2}\right)$. Следовательно, углы между этими прямыми равны $\angle A+\varphi_{1}$, $\varphi_{2}-\varphi_{1}$ и $\angle B+\varphi_{1}$. Аналогично получаем, что прямые $B Y, B X\left(\varphi_{1}\right), B X\left(\varphi_{2}\right)$ и $B X^{*}\left(\varphi_{3}\right)$ образуют между собой те же углы, но в противоположном порядке. Второе решение. Обозначим сумму $\varphi_{1}+\varphi_{2}$ через $2 \varphi_{0}$. Рассуждая также, как в предыдущей задаче, получаем, что четырехугольник $X\left(\varphi_{1}\right) X\left(\varphi_{0}\right) X\left(\varphi_{2}\right) X\left(\pi / 2+\varphi_{0}\right)$ - гармонический, т.е прямая $X\left(\varphi_{1}\right) X\left(\varphi_{2}\right)$ проходит через полюс прямой $X\left(\varphi_{0}\right) X\left(\pi / 2+\varphi_{0}\right)$. Но последняя прямая при любом $\varphi_{0}$ проходит через точку $E$. Следовательно все прямые $X\left(\varphi_{1}\right) X\left(\varphi_{2}\right)$ проходят через некоторую точку $X$ поляры $E$, которая по предыдущей задаче совпадает с прямой $O L$. Поскольку гипербола Кипера является изогональным образом прямой $O L$, полученная выше точка $X$ будет точкой, изогонально сопряженной некоторой точке $X\left(\varphi_{3}\right)$ гиперболы. Чтобы установить связь между $\varphi_{3}$ и $\varphi_{1}, \varphi_{2}$, воспользуемся теоремой о двух парах изогонально сопряженных точек. Пусть $X\left(\varphi_{1}\right), X\left(\varphi_{2}\right)$ - две точки гиперболы Кипера, $X^{*}\left(\varphi_{1}\right), X^{*}\left(\varphi_{2}\right)$ - изогонально сопряженные им точки прямой $O L$. По теореме о двух парах изогонально сопряженных точек прямые $X\left(\varphi_{1}\right) X^{*}\left(\varphi_{2}\right)$ и $X^{*}\left(\varphi_{1}\right) X\left(\varphi_{2}\right)$ пересекаются на гиперболе Кипера. Параметр, соответствующий этой точке равен с одной стороны $f\left(\varphi_{1}\right)-\varphi_{2}$, а с другой $-f\left(\varphi_{2}\right)-\varphi_{1}$, где $f$ - некоторая, пока неизвестная функция. Отсюда получаем $f\left(\varphi_{1}\right)+\varphi_{1}=f\left(\varphi_{2}\right)+\varphi_{2}=$ const, и подставив, например, $\varphi_{2}=0$, получаем $f(\varphi)=-\varphi$, что равносильно утверждению задачи. 3.5. a) Поскольку перпендикуляры из вершин треугольника $A^{\prime} B^{\prime} C^{\prime}$ на стороны треугольника $A B C$ пересекаются в точке $O$, перпендикуляры из вершин треугольника $A B C$ на стороны треугольника $A^{\prime} B^{\prime} C^{\prime}$ также пересекаются в одной точке. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-038.jpg?height=1049&width=1203&top_left_y=1153&top_left_x=453) Рис. 11 b) Воспользуемся следующей леммой. Лемма. Прямые $A^{\prime}(\varphi) B^{\prime}(\varphi)$ и $C C^{\prime}(\pi / 2-\varphi)$ перпендикулярны. Действительно, из леммы следует, что центром ортологичности треугольников $A B C$ и $A^{\prime} B^{\prime} C^{\prime}$ является точка $X(\pi / 2-\varphi)$. Доказательство 1. Пусть точки $A^{\prime \prime}$ и $B^{\prime \prime}$ симметричны $C$ относительно $A^{\prime}(\varphi)$ и $B^{\prime}(\varphi)$ соответственно. Тогда $\angle C A B^{\prime \prime}=\angle C B A^{\prime \prime}=\pi / 2$ и $A^{\prime \prime} B^{\prime \prime} \|$ $A^{\prime}(\varphi) B^{\prime}(\varphi)$ (рис.11). Обозначим через $C^{\prime \prime}$ проекцию $C$ на прямую $A^{\prime \prime} B^{\prime \prime}$. Так как четырехугольник $C C^{\prime \prime} B A^{\prime \prime}$ вписан в окружность с диаметром $C A^{\prime \prime}$, то $\angle B C^{\prime \prime} A^{\prime \prime}=\angle B C A^{\prime \prime}=\varphi$. Аналогично $\angle A C^{\prime \prime} B^{\prime \prime}=\varphi$. Таким образом, четырехугольник $A C^{\prime}(\pi / 2-\varphi) B C^{\prime \prime}$ вписанный. Следовательно, $\angle A C^{\prime \prime} C^{\prime}(\pi / 2-\varphi)=\angle A B C^{\prime}(\pi / 2-\varphi)=\pi / 2-\varphi$ и точки $C, C^{\prime \prime}, C^{\prime}(\pi / 2-\varphi)$ лежат на одной прямой, что и требуется. Доказательство 2. Пусть точка $A^{\prime \prime}$ симметрична $A$ относительно $C B^{\prime}(\varphi)$. Тогда $\triangle C A^{\prime}(\varphi) B$ и $\triangle C B^{\prime}(\varphi) A^{\prime \prime}$ подобны и одинаково ориентированы. Значит, $\triangle C A^{\prime}(\varphi) B^{\prime}(\varphi)$ и $\triangle C B A^{\prime \prime}$ также подобны и одинаково ориентированы, т.е угол между прямыми $A^{\prime}(\varphi) B^{\prime}(\varphi)$ и $B A^{\prime \prime}$ равен $\varphi$. (рис.12) ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-039.jpg?height=1190&width=943&top_left_y=972&top_left_x=580) Рис. 12 Аналогично $\triangle A C^{\prime}(\pi / 2-\varphi) B$ и $\triangle A C A^{\prime \prime}$ подобны и одинаково ориентированы. Следовательно, $\triangle A C C^{\prime}(\pi / 2-\varphi)$ и $\triangle A A^{\prime \prime} B$ подобны и одинаково ориентированы, а угол между прямыми $C C^{\prime}(\pi / 2-\varphi)$ и $A^{\prime \prime} B$ равен $\pi / 2-\varphi$. ## 4 Кубика Нейберга 4.1. a) Все три прямые проходят через одну из точек Наполеона треугольника $X_{u} X_{v} X_{w}$. b) (Из [2]) Первое решение. Два треугольника, образованных центрами тяжести и центрами описанных окружностей $\triangle X U V, \triangle X V W$ и $\triangle X W U$ opтологичны, причем оба центр ортологичности совпадают с центром $\triangle U V W$. Следовательно, эти треугольники перспективны. Второе решение. Указание. Все три прямые проходят через одну из точек Наполеона треугольника, образованного центрами окружностей, описанных около $\triangle X U V, \triangle X V W$ и $\triangle X W U$. 4.2. См. [4]. 4.3. (Из [2]) Приведем элементарное доказательство того, что прямые Эйлера $\triangle A B X, \triangle B C X$ и $\triangle C A X$ пересекаются в одной точке, если $X \equiv I, I_{a}, I_{b}, I_{c}$, $T_{1}^{*}$, или $T_{2}^{*}$. Воспользуемся следующим утверждением. Лемма. Пусть точка $X$ не лежит на окружности $k$, описанной около $\triangle A B C$, а прямые $A X, B X$ и $C X$ повторно пересекают $k$ в точках $A^{\prime}, B^{\prime}$ и $C^{\prime}$ соответственно. Тогда прямые Эйлера $\triangle A B X, \triangle B C X$ и $\triangle C A X$ пересекаются в одной точке тогда и только тогда, когда прямые Эйлера $\triangle A^{\prime} B^{\prime} X, \triangle B^{\prime} C^{\prime} X$ и $\triangle C^{\prime} A^{\prime} X$ пересекаются в одной точке. Доказательство. Пусть $O_{a}$ - центр окружности, описанной около $\triangle B C X$, аналогично определим $O_{b}, O_{c}, O_{a}^{\prime}, O_{b}^{\prime}$ и $O_{c}^{\prime}$. Предположим, что прямые Эйлера $\triangle A^{\prime} B^{\prime} X, \triangle B^{\prime} C^{\prime} X$ и $\triangle C^{\prime} A^{\prime} X$ пересекаются в точке $Y^{\prime}$. Пусть точка $Y^{\prime \prime}$ изогонально сопряжена $Y^{\prime}$ относительно $\triangle O_{a}^{\prime} O_{b}^{\prime} O_{c}^{\prime}$, а $Y$ - такая точка, что четырехугольники $O_{a} O_{b} O_{c} Y$ и $O_{a}^{\prime} O_{b}^{\prime} O_{c}^{\prime} Y^{\prime \prime}$ подобны. Тогда прямые Эйлера $\triangle A B X, \triangle B C X$ и $\triangle C A X$ пересекаются в точке $Y$. Вернемся к задаче. Если $X$ совпадает с $I, I_{a}, I_{b}$ или $I_{c}$, то $X$ - ортоцентр соответствующего $\triangle A^{\prime} B^{\prime} C^{\prime}$ и утверждение задачи следует из леммы. Если $X$ совпадает с $T_{1}^{*}$ или $T_{2}^{*}$, то $\triangle A^{\prime} B^{\prime} C^{\prime}$ - правильный и утверждение следует из леммы и задачи $4.1 \mathrm{~b})$. 4.4 . a) Так как кубика Нейберга переходит в себя при изогональном сопряжении, она содержит точки (мнимые) пересечения бесконечной прямой с описанной окружностью треугольника. Любая коника, проходящая через эти точки, является окружностью. b) Кубика Нейберга, очевидно, содержит, биссектрису угла $C$. Кроме того, она содержит $A, B$ и точки, симметричные им относительно противоположных сторон. c) По задаче 4.1b) кубика Нейберга проходит через точки $T_{a}, T_{b}$ и $T_{c}^{\prime}$. Поскольку она проходит также через $C$, получаем, что внешняя биссектриса $\angle C$ пересекает кубику в четырех различных точках и, значит, целиком содержится в ней. Поскольку $A, B, O$ и $H$ лежат на окружности, симметричной описанной окружности треугольника относительно прямой $A B$, эта окружность также целиком содержится в кубике. d) Доказательство аналогично предыдущему пункту. e) Предположим, что кубика Нейберга распадается на прямую $l$ и окружность $s$. Поскольку $s$ не может совпадать с описанной окружностью $\triangle A B C$, можно без ограничения общности предположить, что $l$ содержит $C$. Так как кубика Нейберга изогонально сопряжена самой себе, прямая $l^{\prime}$, изогонально сопряженная $l$, также целиком содержится в кубике. Значит $l \equiv l^{\prime}$. По задаче 4.2, $l$ параллельна прямой Эйлера. f) (Из [2]) Воспользуемся утверждением п.е). Пусть $L_{C}$ - середина дуги $A C B$ окружности $k$, описанной около $\triangle A B C$. Предположим, что $A C \neq B C$ и прямая Эйлера параллельна $C L_{C}$. Тогда $C L_{C} O H$ - параллелограмм, следовательно $C H$ равно радиусу описанной окружности $\triangle A B C$ и $\angle C=60^{\circ}$. 4.5. См. [4]. ## Список литературы [1] Акопян А.В., Заславский А.А. Геометрические свойства кривых второго порядка. М.: МЦНМО, 2011. [2] N.Beluhov. Ten Concurrent Euler Lines. Unpublished, 2009. (A section of this paper was published as Ten Concurrent Euler Lines, http://forumgeom.fau. edu/FG2009volume9/FG200924.pdf .) [3] A.Zaslavsky. Geometry of Kiepert and Grinberg-Myakishev Hyperbolas. http: //www.jcgeometry.org/Articles/Volume1/JCG2012V1pp65-71.pdf [4] D.Banerjee, S.Mukherjee. Neuberg Locus And Its Properties. http://www. jcgeometry.org/Articles/Volume2/JCG2013V2pp26-38.pdf ## Fermat Points and Euler Lines N. Beluhov, A. Zaslavsky, P. Kozhevnikov, D. Krekov, O. Zaslavsky ${ }^{1}$ ## Preliminaries Given a nonequilateral $\triangle A B C$, we write $M, H, O, I, I_{a}, I_{b}$, and $I_{c}$ for its medicenter, orthocenter, circumcenter, incenter, and excenters opposite $A, B$, and C. The points $O, M$, and $H$ lie on a line known as the Euler line. The point $M$ divides the segment $O H$ in ratio $1: 2$. The midpoints of the sides of $\triangle A B C$, the feet of its altitudes, and the midpoints of the segments $A H, B H$, and $C H$ lie on a circle known as the Euler circle (or the nine-point circle). Its center is the midpoint $E$ of the segment $O H$. Let $X$ be an arbitrary point in the plane of $\triangle A B C$. Then the reflections of the lines $A X, B X$, and $C X$ in the bisectors of $\angle A, \angle B$, and $\angle C$ meet at a point $X^{*}$ (or are parallel, in which case $X^{*}$ is a point at infinity) known as the isogonal conjugate of $X$. The point $L=M^{*}$ is known as Lemoine's point. For every triangle there exists an affine transformation mapping it onto an equilateral triangle. The inverse transformation maps the circumcircle and the incircle of that equilateral triangle onto two ellipses known as the circumscribed and inscribed Steiner ellipses. ## 1 The Fermat, Napoleon, and Apollonius points Let $\triangle A B T_{c}, \triangle B C T_{a}$, and $\triangle C A T_{b}$ be equilateral triangles constructed externally on the sides of $\triangle A B C$ and let $N_{a}, N_{b}$, and $N_{c}$ be their centers. Analogously, let $\triangle A B T_{c}^{\prime}, \triangle B C T_{a}^{\prime}$, and $\triangle C A T_{b}^{\prime}$ be equilateral triangles constructed internally on the sides of $\triangle A B C$ and let $N_{a}^{\prime}, N_{b}^{\prime}$, and $N_{c}^{\prime}$ be their centers. 1.1. Show that ${ }^{2}$ the lines $A T_{a}, B T_{b}$, and $C T_{c}$ are concurrent, and that so are the lines $A T_{a}^{\prime}, B T_{b}^{\prime}$, and $C T_{c}^{\prime}$. The concurrency points thus obtained are known as the two Fermat points (or as the two Torricelli points). We denote them by $T_{1}$ and $T_{2}$. 1.2. The lines $A N_{a}, B N_{b}$, and $C N_{c}$ are concurrent, as are the lines $A N_{a}^{\prime}, B N_{b}^{\prime}$, and $C N_{c}^{\prime}$. The concurrency points thus obtained are known as the two Napoleon points. We denote them by $N_{1}$ and $N_{2}$. 1.3 . a) Both of $\triangle N_{a} N_{b} N_{c}$ and $\triangle N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}$ are equilateral.[^3]b) The centers of both of those triangles coincide with $M$. c) The difference of the areas of those triangles equals the area of $\triangle A B C$. 1.4. Let all angles of $\triangle A B C$ be less than $120^{\circ}$. a) The point $T_{1}$ sees all sides of $\triangle A B C$ at equal angles. b) The sum of the distances from $T_{1}$ to the vertices of $\triangle A B C$ does not exceed the analogous sum for any other point in the plane. When all angles of $\triangle A B C$ are less than $120^{\circ}$, the sum of the distances from $T_{1}$ to its vertices is known as the Fermat length. 1.5. Let $\triangle A B C$ and $\triangle A^{\prime} B^{\prime} C^{\prime}$ satisfy $\angle A+\angle A^{\prime}=\angle B+\angle B^{\prime}=\angle C+\angle C^{\prime}=120^{\circ}$. Then the ratio of the Fermat length to the circumradius is the same in both triangles. 1.6 . a) The points $T_{1}, N_{a}^{\prime}, N_{b}^{\prime}$, and $N_{c}^{\prime}$ lie on a circle. b) The second intersection points of that circle with the segments $A T_{a}, B T_{b}$, and $C T_{c}$ divide those segments in ratio $1: 2$. c) Formulate and prove analogous statements for the points $T_{2}, N_{a}, N_{b}$, and $N_{c}$. 1.7. a) $C T_{1} \perp N_{a} N_{b}$. b) $C N_{1} \perp T_{a} T_{b}$. c) Formulate and prove analogous statements for the points $T_{2}$ and $N_{2}$. 1.8. a) The projections of $T_{1}^{*}$ and $T_{2}^{*}$ onto the sides of $\triangle A B C$ are the vertices of two equilateral triangles. b) $T_{i}^{*} A \cdot B C=T_{i}^{*} B \cdot C A=T_{i}^{*} C \cdot A B$ for $i=1$ and 2 . The two points possessing properties a) and b) are known as the Apollonius points. Thus the points of Fermat and Apollonius are isogonal conjugates. 1.9. Let $T_{1}$ lie inside $\triangle A B C$. Then three billiard balls starting from $T_{1}$ simultaneously and traveling with equal speed in directions opposite the three vertices meet again after reflecting in the three sides. 1.10. $O$ lies on $T_{1}^{*} T_{2}^{*}$. 1.11. The lines $T_{1}^{*} T_{2}^{*}, T_{1} T_{2}$, and $N_{1} N_{2}$ meet at $L$. 1.12. The lines $T_{1} N_{1}$ and $T_{2} N_{2}$ meet at $O$. 1.13. The lines $T_{1} N_{2}$ and $T_{2} N_{1}$ meet at $E$. 1.14. The lines $T_{1} T_{2}^{*}$ and $T_{2} T_{1}^{*}$ meet at $M$. 1.15. The midpoint $M_{T}$ of $T_{1} T_{2}$ lies on the Euler circle. ## 2 Euler lines and Steiner ellipses ${ }^{3}$ 2.1. The Euler lines of $\triangle A B T_{1}, \triangle B C T_{1}, \triangle C A T_{1}, \triangle A B T_{2}, \triangle B C T_{2}$, and $\triangle C A T_{2}$ meet at $M$. Let $M_{A}, M_{B}$, and $M_{C}$ be the midpoints of $B C, C A$, and $A B$ and let $l_{A}, l_{B}$, and $l_{C}$ be the Euler lines of $\triangle A T_{1} T_{2}, \triangle B T_{1} T_{2}$, and $\triangle C T_{1} T_{2}$.[^4] 2.2. The line $l_{A}$ is parallel to $M_{A} M_{T}$, the line $l_{B}$ is parallel to $M_{B} M_{T}$, and the line $l_{C}$ is parallel to $M_{C} M_{T}$. 2.3. The Euler lines of all ten triangles with vertices in the set $\left\{A, B, C, T_{1}, T_{2}\right\}$ meet at $M$. 2.4. Let $T_{3}$ be such a point that $\triangle T_{1} T_{2} T_{3}$ is equilateral. Then the line through $A$ parallel to $B T_{3}$, the line through $B$ parallel to $C T_{3}$, and the line through $C$ parallel to $A T_{3}$ are concurrent. Let $P$ be such a point that $\triangle T_{1} T_{2} P$ is equilateral and oriented oppositely to $\triangle A B C$. Let $Q$ be such a point that $\triangle T_{1} T_{2} Q$ is equilateral and oriented identically to $\triangle A B C$. 2.5. The line $l_{A}$ is parallel to $C P$ and $B Q$, the line $l_{B}$ is parallel to $A P$ and $C Q$, and the line $l_{C}$ is parallel to $B P$ and $A Q$. 2.6. Find the locus of points $X$ such that the line through $A$ parallel to $B X$, the line through $B$ parallel to $C X$, and the line through $C$ parallel to $A X$ are concurrent. 2.7. The points $P$ and $Q$ lie on the circumscribed Steiner ellipse. 2.8 . a) The line through $A$ parallel to $l_{A}$, the line through $B$ parallel to $l_{B}$, and the line through $C$ parallel to $l_{C}$ meet at a point $R$. b) $R$ lies on the circumscribed Steiner ellipse. 2.9. $\triangle A B C$ and $\triangle P Q R$ have equal area and a common medicenter. 2.10. $M_{T}$ lies on the inscribed Steiner ellipse. 2.11. The line $T_{1} T_{2}$ is a normal to the inscribed Steiner ellipse. ## 3 The Kiepert hyperbola Let $\triangle A B C^{\prime}, \triangle B C A^{\prime}$, and $\triangle C A B^{\prime}$ be similar, identically oriented isosceles triangles of bases $A B, B C$, and $C A$. 3.1 . a) The lines $A A^{\prime}, B B^{\prime}$, and $C C^{\prime}$ are concurrent. b) The locus of the intersection of $A A^{\prime}, B B^{\prime}$, and $C C^{\prime}$ is a rectangular hyperbola containing $A, B, C, M$, and $H$. This hyperbola is known as the Kiepert hyperbola. Let $X(\varphi)$ be the point on the Kiepert hyperbola corresponding to angle $\varphi$ at the bases of the three isosceles triangles. We consider $\varphi$ to be positive when all isosceles triangles are constructed externally to $\triangle A B C$, and negative when they are constructed internally. 3.2. The point $X^{*}(\varphi)$ lies on $O L$. 3.3. a) All lines $X(\varphi) X(-\varphi)$ meet at $L$. b) All lines $X(\varphi) X(\pi / 2-\varphi)$ meet at $O$. c) All lines $X(\varphi) X(\pi / 2+\varphi)$ meet at $E$. 3.4. The points $X\left(\varphi_{1}\right), X\left(\varphi_{2}\right)$, and $X^{*}\left(\varphi_{3}\right)$ are collinear if and only if $\varphi_{1}+\varphi_{2}+$ $\varphi_{3} \equiv 0(\bmod \pi)$. 3.5. a) The perpendiculars from $A, B$, and $C$ onto the lines $B^{\prime} C^{\prime}, C^{\prime} A^{\prime}$, and $A^{\prime} B^{\prime}$ are concurrent. b) Their point of concurrence lies on the Kiepert hyperbola. ## 4 The Neuberg cubic 4.1. Let $\triangle U V W$ be equilateral, let $X$ be an arbitrary point in the plane, and let $X_{u}, X_{v}$, and $X_{w}$ be the reflections of $X$ in the lines $V W, W U$, and $U V$. a) The lines $U X_{u}, V X_{v}$, and $W X_{w}$ are concurrent. b) The Euler lines of $\triangle X U V, \triangle X V W$, and $\triangle X W U$ are concurrent. 4.2. Let $X$ be any point in the plane not on the circumcircle of $\triangle A B C$. Then $X$ possesses properties a) and b) above if and only if $X X^{*}$ is parallel to the Euler line. The locus of the points $X$ such that $X X^{*}$ is parallel to the Euler line is a cubic known as the Neuberg cubic. 4.3. The Neuberg cubic contains $A, B, C, O, H, I, I_{a}, I_{b}, I_{c}, T_{1}, T_{2}, T_{1}^{*}$, and $T_{2}^{*}$. 4.4 . a) If the Neuberg cubic contains a straight line, then it also contains a circle. b) If $A C=B C$, then the Neuberg cubic consists of the internal bisector of $\angle C$ and the circle of center $C$ and radius $A B$. c) If $\angle C=60^{\circ}$, then the Neuberg cubic consists of the external bisector of $\angle C$ and the reflection of the circumcircle of $\triangle A B C$ in the line $A B$. d) If $\angle C=120^{\circ}$, then the Neuberg cubic consists of the internal bisector of $\angle C$ and the reflection of the circumcircle of $\triangle A B C$ in the line $A B$. e) Part a) holds if and only if an angle bisector (internal or external) is parallel to the Euler line. In that case, that angle bisector is the straight-line part of the Neuberg cubic. f) Parts a) and e) hold only in the cases listed in parts b), c), and d). 4.5. If $D$ lies on the Neuberg cubic for $\triangle A B C$, then $A$ lies on the Neuberg cubic for $\triangle B C D$, and analogously for $B$ and $C$. ## References [1] Akopyan A. V., Zaslavsky A. A. Geometricheskie Svoistva Krivih Vtorogo Poryadka. MCCME, 2011. [2] N. Beluhov. Ten Concurrent Euler Lines. Unpublished, 2009. (A section of this paper was published as Ten Concurrent Euler Lines, http://forumgeom.fau . edu/FG2009volume9/FG200924.pdf.) [3] A. Zaslavsky. Geometry of Kiepert and Grinberg-Myakishev Hyperbolas. http: //www.jcgeometry.org/Articles/Volume1/JCG2012V1pp65-71.pdf. [4] D. Banerjee, S. Mukherjee. Neuberg Locus And Its Properties. http://www. jcgeometry.org/Articles/Volume2/JCG2013V2pp26-38.pdf. ## Fermat Points and Euler Lines N. Beluhov, A. Zaslavsky,
P. Kozhevnikov, D. Krekov, O. Zaslavsky ${ }^{1}$
Solutions ## 1 The Fermat, Napoleon, and Apollonius points 1.1. Solution 1. By the sine theorem for $\triangle A T_{a} B$ and $\triangle A T_{a} C$, $\sin \angle B A T_{a}: \sin \angle A B T_{a}=B T_{a}: A T_{a}=C T_{a}: A T_{a}=\sin \angle C A T_{a}: \sin \angle A C T_{a}$. Therefore, $\sin \angle B A T_{a}: \sin \angle C A T_{a}=\sin (\angle B+\pi / 3): \sin (\angle C+\pi / 3)$. Analogously, $\sin \angle A C T_{c}: \sin \angle B C T_{c}=\sin (\angle A+\pi / 3): \sin (\angle B+\pi / 3)$ and $\sin \angle C B T_{b}: \sin \angle A B T_{b}=\sin (\angle C+\pi / 3): \sin (\angle A+\pi / 3)$. By the sine Ceva theorem, the statement of the problem follows. The second part of the problem is resolved analogously. Solution 2. Since a $60^{\circ}$ rotation about $A$ maps $\triangle A B T_{b}$ onto $\triangle A T_{c} C$, the lines $B T_{b}$ and $C T_{c}$ and the circumcircles of $\triangle A B T_{c}$ and $\triangle C A T_{b}$ meet at a point. By reasoning analogously about $B$ and $C$, we obtain that $A T_{a}, B T_{b}, C T_{c}$, and the circumcircles of $\triangle A B T_{c}, \triangle B C T_{a}$, and $\triangle C A T_{b}$ all meet at a point. (Figure 1) The second part of the problem is resolved analogously. 1.2. The solution is analogous to the first solution of the previous problem. 1.3. Solution 1. a) By the cosine theorem for $\triangle C N_{a} N_{b}$, we obtain $N_{a} N_{b}^{2}=\frac{C A^{2}}{3}+\frac{C B^{2}}{3}-2 \frac{C A \cdot C B}{3} \cos (\angle C+\pi / 3)=\frac{A B^{2}+B C^{2}+C A^{2}}{6}+\frac{2 S_{A B C}}{\sqrt{3}}$. Since the right-hand side is symmetric, we obtain identical expressions for the two remaining sides of $\triangle N_{a} N_{b} N_{c}$. The proof for $\triangle N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}$ is analogous. $\overrightarrow{0}$. b) By the easily verified identities $A \vec{N}_{b}+B \vec{N}_{c}+C \vec{N}_{a}=\overrightarrow{0}$ and $A \vec{N}_{b}^{\prime}+B \vec{N}_{c}^{\prime}+C \vec{N}_{a}^{\prime}=$ c) By the expressions for the sides of $\triangle N_{a} N_{b} N_{c}$ and $\triangle N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}$ obtained in a). Solution 2. a) Since $A T_{1}$ is a common chord of the circumcircles of $A T_{c} B T_{1}$ and $A T_{b} C T_{1}$, it is perpendicular to the line $N_{b} N_{c}$ connecting the centers of those circles. Analogously, $B T_{b} \perp N_{c} N_{a}$ and $C T_{c} \perp N_{a} N_{b}$. Thus $\triangle N_{a} N_{b} N_{c}$ is equilateral, and $\triangle N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}$ is handled analogously. c) We have that $\triangle A N_{b} N_{c} \simeq \triangle T_{1} N_{b} N_{c}, \triangle B N_{c} N_{a} \simeq \triangle T_{1} N_{c} N_{a}$, and $\triangle C N_{a} N_{b} \simeq$ $\triangle T_{1} N_{a} N_{b}$. Therefore, $$ S_{N_{a} N_{b} N_{c}}=\frac{1}{2}\left(S_{A B C}+S_{A B N_{c}}+S_{B C N_{a}}+S_{C A N_{b}}\right) $$[^5] ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-048.jpg?height=1230&width=1165&top_left_y=263&top_left_x=469) Figure 1 Analogously, $$ S_{N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}}=\frac{1}{2}\left(-S_{A B C}+S_{A B N_{c}^{\prime}}+S_{B C N_{a}^{\prime}}+S_{C A N_{b}^{\prime}}\right) $$ The claim follows. 1.4 . a) By the second solution of Problem 1.1. b) Let $X$ be any point in the plane. Let the $60^{\circ}$ rotation about $A$ mapping $\triangle A B T_{b}$ onto $\triangle A T_{c} C$ also map $T_{1}$ and $X$ onto $T^{\prime}$ and $Y$. Then (Figure 2) $A X+B X+C X=C X+X Y+Y T_{c} \geq C T_{c}=C T_{1}+T_{1} T^{\prime}+T^{\prime} T_{c}=A T_{1}+B T_{1}+C T_{1}$. Equality is attained if and only if $X \equiv T_{1}$. (That $X A+X B \geq X T_{c}$ also follows by Ptolemy's theorem for the quadrilateral $A B T_{c} X$.) 1.5. Let $A T_{1}, B T_{1}$, and $C T_{1}$ meet the circumcircle $k$ of $\triangle A B C$ for the second time at $A^{\prime \prime}, B^{\prime \prime}$, and $C^{\prime \prime}$. Then $\triangle A^{\prime \prime} B^{\prime \prime} C^{\prime \prime}$ is similar to $\triangle A^{\prime} B^{\prime} C^{\prime}$ and we are left to show that the Fermat lengths of $\triangle A B C$ and $\triangle A^{\prime \prime} B^{\prime \prime} C^{\prime \prime}$ are equal. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-049.jpg?height=1025&width=675&top_left_y=271&top_left_x=714) Figure 2 Solution 1. Let $\triangle B^{\prime \prime} C^{\prime \prime} T_{a}^{\prime \prime}$ be equilateral and external to $\triangle A^{\prime \prime} B^{\prime \prime} C^{\prime \prime}$. By the solution of the previous problem, the Fermat lengths of the two triangles equal $A T_{a}$ and $A^{\prime \prime} T_{a}^{\prime \prime}$. (Figure 3 ) ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-049.jpg?height=762&width=1266&top_left_y=1625&top_left_x=424) Figure 3 Let $B T_{a}, C T_{a}, B^{\prime \prime} T_{a}^{\prime \prime}$, and $C^{\prime \prime} T_{a}^{\prime \prime}$ meet $k$ for the second time at $B_{1}, C_{1}, B_{1}^{\prime \prime}$, and $C_{1}^{\prime \prime}$. By an angle chase, $B C_{1}^{\prime \prime}\left\|C B_{1}^{\prime \prime}\right\| B^{\prime \prime} C_{1} \| C^{\prime \prime} B_{1}$. Therefore, the figure is symmetric with respect to the perpendicular bisector $s_{A A^{\prime \prime}}$ of $A A^{\prime \prime}$. It follows that the segments $A T_{a}$ and $A^{\prime \prime} T_{a}^{\prime \prime}$ are reflections of each other in $s_{A A^{\prime \prime}}$ and that their lengths are equal, as needed. Solution 2. Let $k^{\prime}$ be the circle through $T_{1}$ concentric with $k$, and let $k^{\prime}$ meet the lines $A A^{\prime \prime}, B B^{\prime \prime}$, and $C C^{\prime \prime}$ for the second time at $K_{a}, K_{b}$, and $K_{c}$. We consider the configuration when the points $A, T_{1}, K_{a}, A^{\prime \prime} ; B, T_{1}, K_{b}, B^{\prime \prime}$; and $C, K_{c}, T_{1}, C^{\prime \prime}$ lie in this order on the corresponding lines, and all other configurations are analogous. By symmetry, $A T_{1}=K_{a} A^{\prime \prime}, B T_{1}=K_{b} B^{\prime \prime}$, and $C K_{c}=T_{1} C^{\prime \prime}$. We are left to prove that $T_{1} K_{a}+T_{1} K_{b}=T_{1} K_{c}$, and this holds by Ptolemy's theorem for $T_{1} K_{a} K_{b} K_{c}$. 1.6 . a) The point $N_{c}^{\prime}$ lies on the circumcircle $k_{c}$ of $\triangle A B T_{c}$, with $N_{c}^{\prime} T_{c}$ being a diameter of $k_{c}$. Since $T_{1}$ also lies on $k_{c}, \angle N_{c}^{\prime} T_{1} C=\pi / 2$. On the other hand, $M_{C} M: M_{C} C=$ $M_{C} N_{c}: M_{C} T_{c}=1: 3$, so $N_{c} M \| C T_{c}$. Therefore, $N_{c} M$ is the perpendicular bisector of $N_{c}^{\prime} T_{1}$ and $M N_{c}^{\prime}=M T_{1}$. (Figure 4) ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-050.jpg?height=1011&width=721&top_left_y=1077&top_left_x=691) Figure 4 b) Since $M$ is the circumcenter of $\triangle N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}$ and $\angle N_{c}^{\prime} T_{1} C=\pi / 2$, the second intersection point of the circumcircle of $\triangle N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}$ and the line $C T_{c}$ is the reflection of $N_{c}^{\prime}$ in $M$. (Figure 4) Since the line $C M$ divides the segment $N_{c}^{\prime} T_{c}$ in ratio $1: 3$, the claim follows by Menelaus' theorem. 1.7 . a) See the second solution to problem 1.3 a). b) By the cosine theorem for $\triangle A T_{b} N_{c}$ and $\triangle B T_{a} N_{c}$, we have that $N_{c} T_{a}^{2}-N_{c} T_{b}^{2}=$ $C T_{a}^{2}-C T_{b}^{2}$ and the claim follows. 1.8 . a) The reflections of $T_{1}^{*}$ in the sides of $\triangle A B C$ form a triangle $\delta$ such that the lines $T_{1} A, T_{1} B$, and $T_{1} C$ are the perpendicular bisectors of its sides. (Figure 5) Therefore, all of $\delta$ 's angles equal $60^{\circ}$. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-051.jpg?height=656&width=729&top_left_y=607&top_left_x=687) Figure 5 b) Let $X$ and $Y$ be the projections of $T_{1}^{*}$ onto $A C$ and $B C$. Then the quadrilateral $C X T_{1}^{*} Y$ is inscribed in a circle on diameter $C T_{1}^{*}$. Therefore, $X Y=C T_{1}^{*} \cdot \sin \angle C$ and the claim follows by a). 1.9. The lines $A T_{1}, B T_{1}$, and $C T_{1}$ contain the reflections of $T_{1}^{*}$ in the sides of $\triangle A B C$. (Figure 5) Therefore, all three billiard balls eventually arrive at $T_{1}^{*}$, and at the same moment. (A corollary of this problem was on the 28th Tournament of Towns, Spring, 10-11.7: The reflections of $A T_{1}, B T_{1}$, and $C T_{1}$ in the opposite sides of $\triangle A B C$ are concurrent.) 1.10. The locus of the points $X$ such that $A X: B X=A C: B C$ is a circle $\Omega_{c}$ perpendicular to the circumcircle $k$ of $\triangle A B C$. Since $T_{1}^{*}$ and $T_{2}^{*}$ are the intersection points of $\Omega_{a}, \Omega_{b}$, and $\Omega_{c}$, they are mutually inverse with respect to $k$. 1.11. The centers $S_{a}, S_{b}$, and $S_{c}$ of $\Omega_{a}, \Omega_{b}$, and $\Omega_{c}$ are the intersections of the sides of $\triangle A B C$ with the tangents to $k$ at the opposite vertices. Since the line $S_{a} S_{b} S_{c}$ is the polar of $L$ with respect to $k$, it is perpendicular to $O L$. Since the line $T_{1}^{*} T_{2}^{*}$ contains $O$ and is perpendicular to $S_{a} S_{b} S_{c}$, it coincides with $O L$. The two remaining parts of the problem are special cases of Problem 3.3. 1.12. A special case of Problem 3.3. 1.13. A special case of Problem 3.3. 1.14. A special case of the three pairs of isogonal conjugates theorem (see [1]). 1.15. By Problems 3.1 and 3.2 . ## 2 Euler lines and Steiner ellipses In all solutions for this section, we only consider the configuration when $T_{1}$ lies inside $\triangle A B C$ and the quadrilateral $A B C T_{2}$ is convex. All other configurations are handled analogously. 2.1. We show that the Euler line of $\triangle B C T_{1}$ passes through $M$. All other lines in the problem are handled analogously. Let $M_{a}$ be the medicenter of $\triangle B C T_{1}$. Since $N_{a}$ is the circumcenter of $\triangle B C T_{1}$, the Euler line of $\triangle B C T_{1}$ is $M_{a} N_{a}$. (Figure 6) ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-052.jpg?height=583&width=1136&top_left_y=755&top_left_x=489) Figure 6 Since the points $A, T_{1}$, and $T_{a}$ are collinear and the points $M, M_{a}$, and $N_{a}$ divide the segments $M_{A} A, M_{A} T_{1}$, and $M_{A} T_{a}$ in the same ratio $1: 2$, they are collinear as well. 2.2. Let $M_{a}$ and $O_{a}$ be the medicenter and circumcenter of $\triangle A T_{1} T_{2}$. Since $N_{b} N_{c}$ is the perpendicular bisector of $A T_{1}$ and $N_{b}^{\prime} N_{c}^{\prime}$ is the perpendicular bisector of $A T_{2}$, the lines $N_{b} N_{c}$ and $N_{b}^{\prime} N_{c}^{\prime}$ meet at $O_{a}$. (Figure 7) Let $x$ be the acute angle between the lines $M O_{a}$ and $N_{b}^{\prime} N_{c}^{\prime}$. Since $M$ is the common center of $\triangle N_{a} N_{b} N_{c}$ and $\triangle N_{a}^{\prime} N_{b}^{\prime} N_{c}^{\prime}, \angle M N_{c} N_{b}=\angle M N_{b}^{\prime} N_{c}^{\prime}$ $=30^{\circ}$. Consequently, the quadrilateral $M O_{a} N_{c} N_{b}^{\prime}$ is cyclic and $$ \angle N_{b} N_{c} N_{b}^{\prime}=\angle O_{a} N_{c} N_{b}^{\prime}=180^{\circ}-\angle N_{b}^{\prime} M O_{a}=\angle M O_{a} N_{b}^{\prime}+\angle O_{a} N_{b}^{\prime} M=x+30^{\circ} $$ Notice that $\triangle A N_{b} T_{b}, \triangle A N_{c} B$, and $\triangle A N_{b}^{\prime} C$ are similar and identically oriented. It follows that similitude of center $A$ maps $\triangle N_{b} N_{c} N_{b}^{\prime}$ onto $\triangle T_{b} B C$. Thus $$ \angle T_{1} B C=\angle T_{b} B C=\angle N_{b} N_{c} N_{b}^{\prime}=x+30^{\circ} $$ Let $T^{\prime}$ be the reflection of $T_{1}$ in $M_{A}$. Then $\angle B T^{\prime} C=\angle B T_{1} C=120^{\circ}$. Consequently, the quadrilateral $B T^{\prime} C T_{2}$ is cyclic and $$ \angle B T_{2} T^{\prime}=\angle B C T^{\prime}=\angle T_{1} B C=x+30^{\circ} $$ Since the lines $B T_{2}$ and $N_{a}^{\prime} N_{c}^{\prime}$ are perpendicular, the lines $B T_{2}$ and $N_{b}^{\prime} N_{c}^{\prime}$ make a $30^{\circ}$ angle. Thus from $\angle B T_{2} T^{\prime}=x+30^{\circ}$ it follows that $M O_{a} \| T^{\prime} T_{2}$. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-053.jpg?height=1041&width=965&top_left_y=266&top_left_x=577) Figure 7 Since $M_{A}$ and $M_{T}$ are midpoints of $T_{1} T^{\prime}$ and $T_{1} T_{2}, M_{A} M_{T} \| T^{\prime} T_{2}$. Since $M$ and $M_{a}$ divide the segments $A M_{A}$ and $A M_{T}$ in the same ratio 2:1, $M M_{a} \| M_{A} M_{T}$. Consequently, $M O_{a}\left\|T^{\prime} T_{2}\right\| M_{A} M_{T} \| M M_{a}$. It follows that the points $M, M_{a}$, and $O_{a}$ are collinear, $M$ lies on $l_{A}$, and $l_{A} \| M_{A} M_{T}$. 2.3. Solution 1. By Problem 2.1 and the solution to Problem 2.2. Solution 2. By Problem $4.4 \mathrm{~d})$ applied to $\triangle A B T_{1}$ and the point $T_{2}, l_{A}, l_{B}$, and the Euler line of $\triangle A B T_{2}$ are concurrent. Analogously, $l_{A}, l_{B}$, and the Euler line of $\triangle A B T_{1}$ are concurrent. By Problem 2.1, it follows that $l_{A}$ and $l_{B}$ meet at $M$. 2.4. There exist two distinct points $T_{3}$ satisfying the conditions of the problem, corresponding to the two possible orientations of $\triangle T_{1} T_{2} T_{3}$, namely $T_{3} \equiv P$ and $T_{3} \equiv$ $Q$. Ordinarily, orientation does not matter much in geometry problems. However, in this problem the two cases for $T_{3}$ need to be handled very differently. Case 1. $T_{3} \equiv P$. Let $A^{\prime}$ be the isogonal conjugate of $A$ in $\triangle T_{1} T_{2} P$ and let $A^{\prime \prime}$ be the reflection of $A^{\prime}$ in the line $T_{1} T_{2}$. (Figure 8) Then $\angle A^{\prime \prime} T_{1} T_{2}=\angle T_{2} T_{1} A^{\prime}=180^{\circ}-\angle P T_{1} A=\left(\right.$ since $\angle A T_{1} C=120^{\circ}$ and $\left.\angle P T_{1} T_{2}=60^{\circ}\right)=\angle C T_{1} T_{2}$. It follows that the points $T_{1}, A^{\prime \prime}$, and $C$ are collinear. Analogously, the points $T_{2}, A^{\prime \prime}$, and $C$ are collinear. Thus $A^{\prime \prime} \equiv C$. Reflection in the internal bisector of $\angle T_{1} P T_{2}$ followed by reflection in $T_{1} T_{2}$ maps ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-054.jpg?height=594&width=916&top_left_y=272&top_left_x=599) Figure 8 the line $A P$ onto the line $C Q$, so $A P \| C Q$. Analogously, $B P \| A Q$ and $C P \| B Q$. Therefore, when $T_{3} \equiv P$, the three lines in the problem meet at $Q$. Case 2. $T_{3} \equiv Q$. Let $X$ be the intersection point of $A P$ and $B Q$, and let $\triangle R Y B$ and $\triangle Z C P$ be translation copies of $\triangle A Q X$. (Figure 9) ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-054.jpg?height=783&width=895&top_left_y=1300&top_left_x=604) Figure 9 Then the points $C, Y, Z$, and $R$ are collinear, $A R\|B Q\| C P, A P\|B R\| C Q$, and $A Q\|B P\| C R$. Therefore, when $T_{3} \equiv Q$, the three lines in the problem meet at $R$. 2.5. Solution 1. By Problem 2.2, $l_{A}$ is parallel to $M_{A} M_{T}$. By Problem 2.4, CP and $B Q$ are parallel. Since $M_{T}$ is the midpoint of $P Q$, this means that $M_{A} M_{T}$ is the midline of the two parallel lines $C P$ and $B Q$, and is thus parallel to both of them. Solution 2. By Problem 4.1 b), $P$ lies on the Neuberg cubic for $\triangle A T_{1} T_{2}$. Since the isogonal conjugate of $P$ with respect to that triangle is $C$, by Problem $4.2 l_{A}$ is parallel to $C P$. 2.6. Let $X$ be such that the line through $A$ parallel to $B X$, the line through $B$ parallel to $C X$, and the line through $C$ parallel to $A X$ are concurrent at $Y$. Let $\alpha$ be an affine transformation mapping $\triangle A B C$ to an equilateral $\triangle A^{\prime} B^{\prime} C^{\prime}$. Let $\alpha(X)=X^{\prime}$ and $\alpha(Y)=Y^{\prime}$. Then the line through $A^{\prime}$ parallel to $B^{\prime} X^{\prime}$, the line through $B^{\prime}$ parallel to $C^{\prime} X^{\prime}$, and the line through $C^{\prime}$ parallel to $A^{\prime} X^{\prime}$ are concurrent at $Y^{\prime}$. Consider a rotation $\rho$ concentric with $\triangle A^{\prime} B^{\prime} C^{\prime}$ such that $\rho$ maps $A^{\prime}$ onto $B^{\prime}, B^{\prime}$ onto $C^{\prime}$, and $C^{\prime}$ onto $A^{\prime}$. Let $\rho\left(Y^{\prime}\right)=Z^{\prime}$. Then each of $\angle X^{\prime} A^{\prime} Z^{\prime}, \angle X^{\prime} B^{\prime} Z^{\prime}$, and $\angle X^{\prime} C^{\prime} Z^{\prime}$ equals either $60^{\circ}$ or $120^{\circ}$. Consequently, the points $X^{\prime}, Z^{\prime}, A^{\prime}, B^{\prime}$, and $C^{\prime}$ are concyclic. It follows that $X^{\prime}$ lies on the circumcircle of $\triangle A^{\prime} B^{\prime} C^{\prime}$. Therefore, $X$ lies on the circumscribed Steiner ellipse of $\triangle A B C$. 2.7. By Problems 2.4 and 2.6 . 2.8. By Problem 2.5 and Case 2 of the solution to Problem 2.4. 2.9-2.11. Define the affine transformation $\alpha$ and $\triangle A^{\prime} B^{\prime} C^{\prime}$ as in the solution to Problem 2.6. Let $\alpha(\triangle P Q R)=\triangle P^{\prime} Q^{\prime} R^{\prime}$. As in the solution to Problem 2.6, $\triangle A^{\prime} B^{\prime} C^{\prime}$ and $\triangle P^{\prime} Q^{\prime} R^{\prime}$ are concentric and equal equilateral triangles. Thus they have the same area and a common medicenter. Since both of those properties are preserved by affine transformations, so do $\triangle A B C$ and $\triangle P Q R$. Since the midpoint of $P^{\prime} Q^{\prime}$ lies on the common incircle $e^{\prime}$ of $\triangle A^{\prime} B^{\prime} C^{\prime}$ and $\triangle P^{\prime} Q^{\prime} R^{\prime}$, the midpoint of $P Q$ lies on the common inscribed Steiner ellipse $e$ of $\triangle A B C$ and $\triangle P Q R$. Since $P^{\prime} Q^{\prime}$ is tangent to $e^{\prime}$ at its midpoint and tangency is preserved by affine transformations, $P Q$ is tangent to $e$ at $M_{T}$. Since $T_{1} T_{2}$ is perpendicular to $P Q$ at $M_{T}$, it is a normal to $e$. ## 3 The Kiepert hyperbola 3.1 . a) Similarly to Problems 1.1 and 1.2. b) We are going to need the following easily verified claim (see [1] and [3] for details). Lemma. Let $A$ and $B$ be two points in the plane, and let $f$ be a transformation mapping every line through $A$ onto a line $f(l)$ through $B$ in such a way that crossratios are preserved. Then the locus of the intersection points of $l$ and $f(l)$ as $l$ varies is a conic through $A$ and $B$. (When $f(A B)=A B$, that conic degenerates into the union of the line $A B$ and one more straight line.) The transformation $f$ defined by $f\left(A A^{\prime}\right)=B B^{\prime}$ satisfies the conditions of the Lemma. Therefore, the required locus is a conic through $A$ and $B$. When $\triangle A B^{\prime} C$ and $\triangle B A^{\prime} C$ are constructed internally and their base angles equal $\angle C$, the lines $A A^{\prime}$ and $B B^{\prime}$ coincide with $A C$ and $B C$, and we obtain that $C$ also lies on this conic. Since the required locus also contains $M$ and $H$ (corresponding to setting the base angles of the three isosceles triangles to $0^{\circ}$ and to letting them tend to $90^{\circ}$ ), it is a rectangular hyperbola. Note. Since five points uniquely determine a conic, the locus of the intersection of $A A^{\prime}$ and $C C^{\prime}$ is the same hyperbola. This gives us a second proof of part a). 3.2. Isogonal conjugation with respect to a triangle maps a circumscribed conic onto a straight line. Since the Kiepert hyperbola contains $M$ and $H$, its isogonal conjugate is the line $O L$. Since the points $T_{1}^{*}$ and $T_{2}^{*}$ are mutually inverse with respect to the circumcircle, isogonal conjugation maps them onto opposite points on the Kiepert hyperbola. Therefore, $M_{T}$ is the center of the Kiepert hyperbola and thus lies on the Euler circle. By Problem 2.9, we obtain also that the point $R$ lies on the circumcircle of $\triangle A B C$. Let $A^{\prime}(\varphi), B^{\prime}(\varphi)$, and $C^{\prime}(\varphi)$ be the third vertices of the isosceles triangles corresponding to base angle $\varphi$, so that $A A^{\prime}(\varphi), B B^{\prime}(\varphi)$, and $C C^{\prime}(\varphi)$ meet at $X(\varphi)$. 3.3. Solution 1. a) When the Kiepert hyperbola is projected from $A$ onto the perpendicular bisector of $B C$, the points $X(\varphi)$ and $X(-\varphi)$ are mapped onto the points $A^{\prime}(\varphi)$ and $A^{\prime}(-\varphi)$ symmetric with respect to $B C$. Therefore, the transformation from the Kiepert hyperbola onto itself mapping $X(\varphi)$ onto $X(-\varphi)$ preserves cross-ratios. The transformation mapping $X$ onto the second intersection point of the Kiepert hyperbola with the line $X L$ also preserves cross-ratios. Therefore, to establish the claim it suffices to find three values of $\varphi$ such that the two transformations coincide. When $\varphi=\angle A, X(-\varphi)$ coincides with $A$ and $X(\varphi)$ lies on the symmedian through $A$. Analogously, the two transformations coincide when $\varphi=\angle B$ and $\varphi=\angle C$. b), c) Similarly to part a). Solution 2. a), b) By Problem 3.4. c) Similarly to the first solution of Problem 3.4. Use the fact that $A E$ contains $A^{\prime}\left(90^{\circ}-\angle A\right)$ and analogously for $B E$ and $C E$. The special cases $\varphi=\pi / 3$ and $\varphi=\pi / 6$ of this problem yield Problems 1.11-1.13. 3.4. Solution 1. (N. Beluhov) We use the following lemma. Lemma. Let $\varphi_{1}+\varphi_{2}=\angle A$. Then the lines $A A^{\prime}\left(\varphi_{1}\right)$ and $A A^{\prime}\left(\varphi_{2}\right)$ are symmetric with respect to the bisectors of $\angle A$. Proof. Let $L_{A}$ and $L_{A}^{\prime}$ be the midpoints of the $\operatorname{arcs} B C$ not containing $A$ and $B A C$ of the circumcircle $k$ of $\triangle A B C$. (Figure 10) Then $B L_{A}$ and $B L_{A}^{\prime}$ are the bisectors of $\angle A^{\prime}\left(\varphi_{1}\right) B A^{\prime}\left(\varphi_{2}\right)$ in $\triangle B A^{\prime}\left(\varphi_{1}\right) A^{\prime}\left(\varphi_{2}\right)$. It follows that the circumcircle $k$ of $\triangle B L_{A} L_{A}^{\prime}$ is the Apollonius circle for $B$ in $\triangle B A^{\prime}\left(\varphi_{1}\right) A^{\prime}\left(\varphi_{2}\right)$. Therefore, $A L_{A}$ and $A L_{A}^{\prime}$ are the bisectors of $\angle A^{\prime}\left(\varphi_{1}\right) A A^{\prime}\left(\varphi_{2}\right)$. Let the line $X\left(\varphi_{1}\right) X\left(\varphi_{2}\right)$ meet $A B$ at $Y$. For $X^{*}\left(\varphi_{3}\right)$ to lie on $X\left(\varphi_{1}\right) X\left(\varphi_{2}\right)$, it is necessary and sufficient that $A X^{*}\left(\varphi_{3}\right)$ and $B X^{*}\left(\varphi_{3}\right)$ meet the line $X\left(\varphi_{1}\right) X\left(\varphi_{2}\right)$ at the same point. This holds if and only if the cross-ratio of the four lines $A Y, A X\left(\varphi_{1}\right), A X\left(\varphi_{2}\right)$, and $A X^{*}\left(\varphi_{3}\right)$ equals the cross-ratio of the four lines $B Y, B X\left(\varphi_{1}\right), B X\left(\varphi_{2}\right)$, and $B X^{*}\left(\varphi_{3}\right)$. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-057.jpg?height=997&width=702&top_left_y=271&top_left_x=709) Figure 10 The lines $A Y, A X\left(\varphi_{1}\right), A X\left(\varphi_{2}\right)$, and $A X^{*}\left(\varphi_{3}\right)$ meet the perpendicular bisector of $A C$ at $A^{\prime}(-\angle A), A^{\prime}\left(\varphi_{1}\right), A^{\prime}\left(\varphi_{2}\right)$, and, by the Lemma, $A^{\prime}\left(\angle B-\varphi_{3}\right) \equiv A^{\prime}\left(\angle B+\varphi_{1}+\varphi_{2}\right)$. It follows that the angles between successive lines equal $\angle A+\varphi_{1}, \varphi_{2}-\varphi_{1}$, and $\angle B+\varphi_{1}$. Analogously, the angles between successive lines in the sequence $B Y, B X\left(\varphi_{1}\right)$, $B X\left(\varphi_{2}\right)$, and $B X^{*}\left(\varphi_{3}\right)$ are the same, only ordered in reverse. Solution 2. Let $\varphi_{1}+\varphi_{2}=2 \varphi_{0}$. Reasoning as in the previous problem, we obtain that the quadrilateral $X\left(\varphi_{1}\right) X\left(\varphi_{0}\right) X\left(\varphi_{2}\right) X\left(\pi / 2+\varphi_{0}\right)$ is harmonic, i.e., the line $X\left(\varphi_{1}\right) X\left(\varphi_{2}\right)$ contains the pole of the line $X\left(\varphi_{0}\right) X\left(\pi / 2+\varphi_{0}\right)$. However, for all $\varphi_{0}$ the line $X\left(\varphi_{0}\right) X\left(\pi / 2+\varphi_{0}\right)$ contains $E$. It follows that when $\varphi_{1}$ and $\varphi_{2}$ vary so that $\varphi_{1}+\varphi_{2}$ remains equal to $2 \varphi_{0}$, all lines $X\left(\varphi_{1}\right) X\left(\varphi_{2}\right)$ meet at a point $X$ on the polar of $E$, which by the previous problem coincides with the line $O L$. Since the Kiepert hyperbola and the line $O L$ are isogonal conjugates, $X$ is the isogonal conjugate of some point $X\left(\varphi_{3}\right)$ on the Kiepert hyperbola. To work out the relation between $\varphi_{1}$ and $\varphi_{2}$ on the one hand, and $\varphi_{3}$ on the other hand, we are going to use the two pairs of isogonal conjugates theorem. Let $X\left(\varphi_{1}\right)$ and $X\left(\varphi_{2}\right)$ be two points on the Kiepert hyperbola and let $X^{*}\left(\varphi_{1}\right)$ and $X^{*}\left(\varphi_{2}\right)$ be their isogonal conjugates on the line $O L$. By the two pairs of isogonal conjugates theorem, the lines $X\left(\varphi_{1}\right) X^{*}\left(\varphi_{2}\right)$ and $X^{*}\left(\varphi_{1}\right) X\left(\varphi_{2}\right)$ meet in a point on the Kiepert hyperbola. The value of $\varphi$ corresponding to that point equals, on the one hand, $f\left(\varphi_{1}\right)-\varphi_{2}$, and, on the other hand, $f\left(\varphi_{2}\right)-\varphi_{1}$, where $f$ is some (as of yet unknown) function. It follows that $f\left(\varphi_{1}\right)+\varphi_{1}=f\left(\varphi_{2}\right)+\varphi_{2}=$ const. Substituting, say, $\varphi_{2}=0$, we obtain $f(\varphi)=-\varphi$, completing the solution. 3.5. a) Since the perpendiculars from the vertices of $\triangle A^{\prime} B^{\prime} C^{\prime}$ onto the sides of $\triangle A B C$ are concurrent (being the perpendicular bisectors of the sides of $\triangle A B C$ ), the perpendiculars from the vertices of $\triangle A B C$ onto the sides of $\triangle A^{\prime} B^{\prime} C^{\prime}$ are concurrent as well. b) We use the following lemma. Lemma. The lines $A^{\prime}(\varphi) B^{\prime}(\varphi)$ and $C C^{\prime}(\pi / 2-\varphi)$ are perpendicular. Indeed, it follows from the Lemma that the orthologic center of $\triangle A B C$ and $\triangle A^{\prime} B^{\prime} C^{\prime}$ is the point $X(\pi / 2-\varphi)$. Proof 1. Let $A^{\prime \prime}$ and $B^{\prime \prime}$ be the reflections of $C$ in $A^{\prime}(\varphi)$ and $B^{\prime}(\varphi)$. Then $\angle C A B^{\prime \prime}=\angle C B A^{\prime \prime}=\pi / 2$ and $A^{\prime \prime} B^{\prime \prime} \| A^{\prime}(\varphi) B^{\prime}(\varphi)$. (Figure 11) ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-058.jpg?height=1035&width=1203&top_left_y=973&top_left_x=453) Figure 11 Let $C^{\prime \prime}$ be the projection of $C$ onto the line $A^{\prime \prime} B^{\prime \prime}$. Since the quadrilateral $C C^{\prime \prime} B A^{\prime \prime}$ is inscribed in a circle of diameter $C A^{\prime \prime}$, we obtain that $\angle B C^{\prime \prime} A^{\prime \prime}=$ $\angle B C A^{\prime \prime}=\varphi$. Analogously, $\angle A C^{\prime \prime} B^{\prime \prime}=\varphi$. It follows that the quadrilateral $A C^{\prime}(\pi / 2-\varphi) B C^{\prime \prime}$ is cyclic. Thus $\angle A C^{\prime \prime} C^{\prime}(\pi / 2-$ $\varphi)=\angle A B C^{\prime}(\pi / 2-\varphi)=\pi / 2-\varphi$ and $C, C^{\prime \prime}$, and $C^{\prime}(\pi / 2-\varphi)$ are collinear, as needed. Proof 2. Let $A^{\prime \prime}$ be the reflection of $A$ in the line $C B^{\prime}(\varphi)$. Then $\triangle C A^{\prime}(\varphi) B$ and $\triangle C B^{\prime}(\varphi) A^{\prime \prime}$ are similar and identically oriented. Therefore, so are $\triangle C A^{\prime}(\varphi) B^{\prime}(\varphi)$ and $\triangle C B A^{\prime \prime}$ and the lines $A^{\prime}(\varphi) B^{\prime}(\varphi)$ and $B A^{\prime \prime}$ make an angle of $\varphi$. (Figure 12) ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-059.jpg?height=1188&width=956&top_left_y=268&top_left_x=579) Figure 12 Analogously, $\triangle A C^{\prime}(\pi / 2-\varphi) B$ and $\triangle A C A^{\prime \prime}$ are similar and identically oriented. Therefore, so are $\triangle A C C^{\prime}(\pi / 2-\varphi)$ and $\triangle A A^{\prime \prime} B$ and the lines $C C^{\prime}(\pi / 2-\varphi)$ and $A^{\prime \prime} B$ make an angle of $\pi / 2-\varphi$. ## 4 The Neuberg cubic 4.1. a) All three lines meet at one of the two Napoleon points of $\triangle X_{u} X_{v} X_{w}$. b) (From [2]) Solution 1. The two triangles formed by the medicenters and circumcenters of $\triangle X U V, \triangle X V W$, and $\triangle X W U$ are orthologic, and both orthologic centers coincide with the center of $\triangle U V W$. However, two orthologic triangles with coinciding orthologic centers are always perspective. Solution 2. Hint. All three lines meet at one of the two Napoleon points of the triangle formed by the circumcenters of $\triangle X U V, \triangle X V W$, and $\triangle X W U$. 4.2. See [4]. 4.3. (From [2]) We give an elementary proof that the Euler lines of $\triangle A B X$, $\triangle B C X$, and $\triangle C A X$ are concurrent when $X \equiv I, I_{a}, I_{b}, I_{c}, T_{1}^{*}$, and $T_{2}^{*}$. We use the following lemma. Lemma. Let $X$ be a point not on the circumcircle $k$ of $\triangle A B C$, and let $A X$, $B X$, and $C X$ meet $k$ for the second time at $A^{\prime}, B^{\prime}$, and $C^{\prime}$. Then the Euler lines of $\triangle A B X, \triangle B C X$, and $\triangle C A X$ are concurrent if and only if the Euler lines of $\triangle A^{\prime} B^{\prime} X, \triangle B^{\prime} C^{\prime} X$, and $\triangle C^{\prime} A^{\prime} X$ are concurrent. Proof. Let $O_{a}$ be the circumcenter of $\triangle B C X$, and define $O_{b}, O_{c}, O_{a}^{\prime}, O_{b}^{\prime}$, and $O_{c}^{\prime}$ analogously. Suppose that the Euler lines of $\triangle A^{\prime} B^{\prime} X, \triangle B^{\prime} C^{\prime} X$, and $\triangle C^{\prime} A^{\prime} X$ are concurrent at $Y^{\prime}$. Let $Y^{\prime \prime}$ be the isogonal conjugate of $Y^{\prime}$ in $\triangle O_{a}^{\prime} O_{b}^{\prime} O_{c}^{\prime}$ and let $Y$ be such that the quadrilaterals $O_{a} O_{b} O_{c} Y$ and $O_{a}^{\prime} O_{b}^{\prime} O_{c}^{\prime} Y^{\prime \prime}$ are similar. Then the Euler lines of $\triangle A B X, \triangle B C X$, and $\triangle C A X$ are concurrent at $Y$. When $X$ coincides with $I, I_{a}, I_{b}$, or $I_{c}, X$ is the orthocenter of $\triangle A^{\prime} B^{\prime} C^{\prime}$ and the claim follows by the Lemma. When $X$ coincides with $T_{1}^{*}$ or $T_{2}^{*}, \triangle A^{\prime} B^{\prime} C^{\prime}$ is equilateral and the claim follows by the Lemma and Problem $4.1 \mathrm{~b}$ ). 4.4. a) Since the Neuberg cubic is its own isogonal conjugate, it contains the (imaginary) intersection points of the line at infinity and the circumcircle. Every conic through this pair of points is a circle. b) The Neuberg cubic contains the internal bisector of $\angle C$. Furthermore, it contains $A, B$ and their reflections in the opposite sides. c) By Problem $4.1 \mathrm{~b}$ ), the Neuberg cubic contains $T_{a}, T_{b}$, and $T_{c}^{\prime}$. Since it also contains $C$, it meets the external bisector of $\angle C$ at four distinct points and therefore contains it. Since $A, B, O$, and $H$ all lie on the reflection of the circumcircle in the line $A B$, the circle part of the Neuberg cubic coincides with this circle. d) Similarly to part c). e) Suppose that the Neuberg cubic consists of a straight line $l$ and a circle $s$. Since $s$ cannot contain all vertices of $\triangle A B C$, without loss of generality $l$ contains $C$. Since the Neuberg cubic is its own isogonal conjugate, the isogonal conjugate $l^{\prime}$ of $l$ also belongs to the Neuberg cubic. Thus $l \equiv l^{\prime}$. By Problem 4.2, $l$ is then parallel to the Euler line. f) (From [2]) We are going to use e). Let $L_{C}$ be the midpoint of the arc $A C B$ of the circumcircle $k$ of $\triangle A B C$. Suppose that $A C \neq B C$ and the Euler line is parallel to $C L_{C}$. Then $C L_{C} O H$ is a parallelogram, so $C H$ equals the circumradius of $\triangle A B C$ and $\angle C=60^{\circ}$. 4.5. See $[4]$. ## References [1] Akopyan A. V., Zaslavsky A. A. Geometricheskie Svoistva Krivih Vtorogo Poryadka. MCCME, 2011. [2] N. Beluhov. Ten Concurrent Euler Lines. Unpublished, 2009. (A section of this paper was published as Ten Concurrent Euler Lines, http://forumgeom.fau. edu/FG2009volume9/FG200924.pdf.) [3] A. Zaslavsky. Geometry of Kiepert and Grinberg-Myakishev Hyperbolas. http: //www.jcgeometry.org/Articles/Volume1/JCG2012V1pp65-71.pdf. [4] D. Banerjee, S. Mukherjee. Neuberg Locus And Its Properties. http://www. jcgeometry.org/Articles/Volume2/JCG2013V2pp26-38.pdf. ## Игра «Избегание линий»- 1 Задачу представляют Бурсиан О., Кохась Д., Кохась К. ## 1 Предварительные задачи Мы будем рассматривать игры, в которых два игрока - Петя и Вася (они же Алиса и Боб, они же Первый и Второй, ну, или попросту Р1 и Р2) - по-очереди совершают ходы. В наших играх они будут что-нибудь красить, причем каждый игрок будет использовать свой «личный» цвет: для определенности пусть Петя все красит в пурпурный цвет (а лучше, не в пурпурный, а в первый), а Вася пусть красит, эээ..., в васильковый (а лучше, второй)! Во всех играх Петя ходит первым. «А Вася?» - спросите вы. А Вася ходит вторым! Обычно за один ход разрешается покрасить один еще неокрашенный объект (если в условии явно не указано противоположное). Чего они хотят добиться своими раскрашиваниями, указано в условиях задач, но стоит иметь в виду, что, вообще говоря, им, быть может, и не удастся добиться желаемого, тогда игра окончиться ничьей. Поэтому на вопрос «Кто выиграет?» вполне возможен ответ «Никто!». 1.1. Петя и Вася по-очереди красят вершины додекаэдра. Игрок, после хода которого две вершины какого-то ребра окажутся окрашенными в один цвет, проигрывает. Кто выиграет при правильной игре? 1.2. Петя и Вася по-очереди красят вершины какого-то многогранника. Игрок, после хода которого все вершины какой-то грани окажутся окрашенными в один цвет, выигрывает. Кто выиграет при правильной игре? 1.3. На окружности поставлено 99 точек. Петя и Вася по-очереди красят эти точки. Игрок, после хода которого две соседние точки окажутся окрашенными одинаково, проигрывает. Кто выиграет при правильной игре? 1.4. На окружности стоят $n$ точек. Петя и Вася по-очереди, каждый - своим фломастером, проводят хорды, в том числе, можно соединять соседние точки. За один ход можно провести сразу несколько хорд. Игрок, после хода которого образуется треугольник с вершинами на исходной окружности, все стороны которого окрашены одинаково, проигрывает. Кто выиграет при правильной игре? Достаточно выяснить вопрос для достаточно больших значений $n$. ## 2 Доски с пересекающимися линиями Пусть на плоскости или еще где-нибудь отмечено произвольное конечное множество точек - будем называть эти точки узлами - и проведено несколько линий: каждая линия проходит через несколько узлов. Линию можно представлять себе, как «кривую произвольной формы», проходящую через узлы. Пусть для любых двух линий множества узлов, принадлежащих этим линиям, не совпадают. Фактически линии просто задают подмножества в множестве узлов. Будем говорить, что две линии пересекаются, если они имеют хотя бы один общий узел. Геометрические подробности расположения линий, в том числе, их пересечения не в узлах, нам несущественны. Позже мы будем использовать конфигурацию из узлов и линии как игровую доску и введем дополнительное требования, которому должна удовлетворять доска. А пока несколько предварительных утверждений. 2.1. На доске $n$ узлов. Все линии попарно пересекаются. a) Докажите, что количество линий не больше $2^{n-1}$. b) Пусть количество линий равно $2^{n-1}$. Докажите, что если к линии добавить узел, то снова получится линия. 2.2. На доске $n$ узлов. Все линии попарно пересекаются. Докажите, что можно добавить несколько линий так, чтобы их количество стало равно $2^{n-1}$ и чтобы по-прежнему все линии пересекались. 2.3. На доске $n$ узлов и несколько линий. Пусть наименьшая линия содержит $k$ узлов. Докажите, что если любые $k+1$ линий имеют хотя бы один общий узел, то и все линии вообще тоже пересекаются в каком-то узле. 2.4. Дан кубик $3 \times 3 \times 3$. Пусть его единичные кубики будут узлами доски, можно считать, что это тройки целых чисел $\left(x_{1}, x_{2}, x_{3}\right), 0 \leqslant x_{1}, x_{2}, x_{3} \leqslant 2$. Пусть все линии на доске содержат ровно по 3 узла, а именно, будем считать, что линия, проходящая через узлы $\left(x_{1}, x_{2}, x_{3}\right.$ ) и $\left(y_{1}, y_{2}, y_{3}\right)$, проходит также через узел ( $\left.-x_{1}-y_{1},-x_{2}-y_{2},-x_{3}-y_{3}\right)$ (вычисления по модулю 3 ). Кто выиграет на этой доске? ## Игра «Избегание линий»- 2 Задачу представляют Бурсиан О., Кохась Д., Кохась К. ## 3 Общий вариант игры «Избегание линий» Пусть дана игровая доска с узлами и линиями, как в предыдущем разделе. Петя и Вася поочереди красят узлы. Игрок, после хода которого все узлы какой-то линии окажутся окрашенными одинаково, проигрывает. Если этого не произошло к тому времени, когда все узлы на доске оказались покрашенными, объявляется ничья. Напомним, что перестановка - это взаимно однозначное отображение конечного множества на себя. Мы будем рассматривать перестановки множества узлов доски. Будем говорить, что перестановка узлов доски сохраняет линии, если для любой линии образ набора узлов, из которых состоит эта линия, - тоже линия. Мы будем рассматривать только доски, обладающие свойством транзитивности: для любых двух узлов $u$ и $v$ существует перестановка $\sigma$ узлов доски, сохраняющая линии, такая что $\sigma(u)=v$. Пример. Игра Сим. На окружности стоят $n$ точек. Петя и Вася, по-очереди проводят хорды, каждый - своим фломастером, в том числе, можно соединять соседние точки. В отличие от задачи 1.4, за один ход можно провести одну хорду. Игрок, после хода которого образуется треугольник с вершинами на исходной окружности, все стороны которого окрашены одинаково, проигрывает. Чтобы превратить Сим в транзитивную игру с избеганием линий, обозначим исходные точки $A_{1}$, $A_{2}, \ldots, A_{n}$ и рассмотрим доску, на которой заданы узлы $N_{i j}, 1 \leqslant i1$, существует доска, на которой Петя выигрывает. Вы можете разбирать отдельные частные случаи этого вопроса. 3.8. Докажите, что Петя не сможет выиграть ни на какой доске, содержащей ровно $2^{n}$ узлов. 3.9. Открытый вопрос. Для каких простых чисел $n$ существует доска с $n$ узлами, на которой Петя выигрывает? Например, что вы можете сказать о небольших значениях $n: n=3,5,7,11,13$ ? ## Игра «Избегание линий»- 3 Задачу представляют Бурсиан О., Кохась Д., Кохась К. ## 4 Специальные доски Все доски, рассматриваемые в этой серии задач, предполагаются транзитивными. Обозначим через $A$ множество таких $n$, для которых существует транзитивная доска с $n$ узлами, на которой любые две линии пересекаются и количество линий равно $2^{n-1}$. Вопрос о том, какие натуральные числа принадлежат множеству $A$, а какие нет, до конца не выяснен. Перестановка $\tau$ точек множества $X$ называется произведением перестановок $\rho$ и $\sigma$, если $\tau(k)=$ $\rho(\sigma(k))$ при всех $k \in X$, т. е. если $\tau$ - это композиция отображений $\rho$ и $\sigma$. Умножение перестановок, вообще говоря, некоммутативно. Набор перестановок $G$ называют группой, если он замкнут относительно операции произведения перестановок. В этом случае $G$ обязательно содержит тождественную перестановку $e$, а также для каждого $\sigma \in G$ в наборе $G$ содержится перестановка, обратная к $\sigma$, т. е. такая перестановка $\bar{\sigma}$, что $\bar{\sigma} \sigma=\sigma \bar{\sigma}=e$. Назовем перестановку флипом, если она разбивает все точки множества на пары и меняет местами точки в каждой паре. Назовем перестановку флопом, если длина каждого ее цикла равна степени 2. Пусть $B$ - множество тех $n$, для которых можно построить группу перестановок множества $[n]=\{1,2, \ldots, n\}$, которая действует на множестве $[n]$ транзитивно и не содержит флопов. 4.1. а) Пусть для данной доски существует флип, сохраняющий линии. Докажите, что Петя не сможет выиграть на такой доске. b) В магазине игрушек продается доска, на которой $n$ узлов, где $n$ четное, все линии попарно пересекаются и их количество равно $2^{n-1}$. Докажите, что существует доска, на которой $n$ узлов, выигрышная для первого. 4.2. Пусть транзитивная группа $G$ перестановок множества $[2 k]$ содержит перестановку $\tau$, которая отображает некоторое $k$-элементное множество $X$ на его дополнение $[2 k] \backslash X$. Докажите, что група $G$ содержит флоп. 4.3. На доске $n$ узлов. Все линии попарно пересекаются. При этом никакой флоп не является перестановкой, сохраняющей линии. Докажите, что можно добавить несколько линий так, чтобы их количество стало равно $2^{n-1}$, доска осталась транзитивной, а флопы по-прежнему не сохраняли бы линии. 4.4. Пусть $n=2^{a} \cdot b$, где $b>1$ нечетно. Докажите, что существует доска с $n$ узлами, на которой нет флипов. 4.5. Докажите, что $A=B$. 4.6. Докажите, что множество $A$ замкнуто относительно умножения: если $k, n \in A$, то $k n \in A$. 4.7. Докажите, что если $b>1$ - нечетное число, то $2 b \in A$. 4.8. Докажите, что если $b>3$ - нечетное число, то $4 b \in A$. ## 5 Некоторые факты из теории групп Мы приводим несколько теорем из теории групп, они требуются в решении задачи 3.8 Относительно элементарная постановка вопроса и его сведение с помощью задачи 4.1 к факту Любая транзитивная группа перестановок множества из $2^{n}$ элементов содержит флип оставляет шанс на то, что существует решение, не требующее всей этой техники. 5.1. Теорема Лагранжа. a) Пусть $G$ - конечная группа, $H$ - ее подгруппа. Тогда $|G|$ : $|H|$. b) Пусть $x \in G$, обозначим через $k$ наименьшее натуральное число, для которого $x^{k}=e$. Это число называется порядком элемента $x$. Тогда $|G|: k$. 5.2. Пусть $G$ - конечная группа. Тогда для любого элемента $h \in G$ число элементов в его классе сопряженных элементов, т. е. в множестве $G_{h}=\left\{a h a^{-1}, a \in G\right\}$ является делителем числа $|G|$. Более точно: пусть $N_{h}=\{x \in G: h x=x h\}$. Тогда множество $N_{h}$ представляет собой группу и $\left|G_{h}\right| \cdot\left|N_{h}\right|=|G|$. 5.3. Пусть $G$ - конечная группа перестановок множества $M$. Для каждых $x \in M, g \in G$ обозначим через $g x$ образ точки $x$ под действием перестановки $g$. Множество $\mathcal{O}_{x}=\{g x, g \in$ $G\}$ называется орбитой элемента $x$, а множество $S_{x}=\{g \in G: g x=x\}$ - стабилизатором точки $x$. Тогда для любой точки $x \in M \quad\left|\mathcal{O}_{x}\right| \cdot\left|S_{x}\right|=|G|$. 5.4. Пусть $G$ - конечная группа. Множество $C=\{c \in G: \forall g \in G g c=c g\}$ называется центром группы $G$. Оно непусто, поскольку всегда $e \in C$. Тогда если $|G|=p^{n}$, где $p-$ простое число, то $C \neq\{e\}$. 5.5. Теорема Силова. Пусть $G$ - группа из $n$ элементов, $p^{a}$ - максимальная степень простого числа, на которую делится $n$. Тогда группа $G$ содержит подгруппу $H$, состоящую из $p^{a}$ элементов. ## Решения ## 1 Предварительные задачи 1.1. Ответ: выигрывает Вася. Он может пользоваться центрально-симметричной стратегией и тогда не проиграет. При этом игра не может окончиться ничьей, поскольку если вершины пятиугольной грани додекаэдра покрашены в два цвета, обязательно найдутся две соседние вершины, окрашенные одинаково. 1.2. Ответ: возможно два исхода: выигрывает Петя, либо ничья. Первый вариант возможен, например, на октаэдре, второй - на тетраэдре. Ни на каком многограннике Вася не может выиграть, потому что если у Васи есть выигрышная стратегия, то Петя может ее украсть. 1.3. Ответ: выигрывает Вася. Своим первым ходом он отметит точку через один от Петиной и дальше будет пользоваться симметричной стратегией. Другая возможная стратегия для Васи - каждым ходом красить точку, соседнюю по часовой стрелке с точкой, которую только что закрасил Петя. 1.4. Ответ: при $n \geqslant 6$ выигрывает Вася. Это следствие 5 из [2]. По задаче 3.2 Петя не может выиграть. При $n \geqslant 6$ игра не может закончиться ничьей, так как рамсеевское число $R(3,3)=6$. ## 2 Доски с пересекающимися линиями 2.1. а) Множество и его дополнение не могут одновременно оказаться линиями. Кстати, если количество линий равно $2^{n-1}$, для любой пары взаимно дополнительных множеств одно является линией, а другое нет. b) Если $F$ - линия, $F \subset G$, то $F \cap(X \backslash G)=\varnothing$. Следовательно, $X \backslash G$ - не линия. Тогда по решению п. а) $G$ - линия. 2.2. Это [1, лемма 2.2]. Пусть $X$ - множество узлов доски, $\mathcal{L}$ - множество линий. по условию любые две линии пересекаются. Множество линий, которое мы хотим построить, должно удовлетворять утверждению 2.1 b), возьмем на заметку все множества узлов, которые содержат линию: положим $$ \mathcal{L}_{1}=\{B \subset X: A \subset B \text { для некоторой линии } A \in \mathcal{L}\} $$ В частности, $\mathcal{L} \subset \mathcal{L}_{1}$. Если найдется множество $B$, которое пересекается с каждой линией, но ни одну линию не содержит (это значит, что его дополнение $\bar{B}$ тоже пересекается с каждой линией и ни одну из них не содержит), то одно из множеств $B, \bar{B}$ тоже стоит назначить линией. Выберем то из них, в котором больше узлов (в случае равенства числа узлов выберем любое). Пусть $\mathcal{L}_{2}$ - совокупность множеств, выбранных таким образом. Проверим, что $\mathcal{L}_{1} \cup \mathcal{L}_{2}$ - семейство попарно пересекающихся линий, в котором ровно $2^{n-1}$ элементов. То, что любые две линии в нем пересекаются, ясно по построению. Пусть $Y-$ произвольное подмножество $X, \bar{Y}$ - его дополнение. Если $Y$ содержит какую-то линию, то $Y \in \mathcal{L}_{1} ;$ если $Y$ не пересекается какой-нибудь линией, то $\bar{Y} \in \mathcal{L}_{1}$; если же и то, и другое не выполнено, то $Y$ или $\bar{Y}$ лежит в $\mathcal{L}_{2}$. 2.3. Это утверждение Ph. Zumstein. Если $L=\left\{x_{1}, x_{2}, \ldots, x_{k}\right\}$ - минимальная линия, и ни одна из ее точек не является общей точкой пересечения всех множеств, то для каждого $i$, $1 \leqslant i \leqslant k$ существует линия $L_{i}$, для которой $x_{i} \notin L_{i}$. Но тогда пересечение $L \cap L_{1} \cap \ldots \cap L_{k}$ пусто. 2.4. Ответ: выигрывает Петя. Сделаем несколько замечаний об устройстве игровой доски. Прежде всего, любые два различных кубика действительно определяют линию, состоящую из трех кубиков (в том смысле, что третий кубик, найденный по рецепту из условия задачи, не совпадает с исходными двумя). Далее, правило, описывающее линии, весьма симметрично: если по кубикам $A$ и $B$ конструкция из условия задачи задает кубик $C$, то по кубикам $A$ и $C$ эта же конструкция дает кубик $B$, а по кубикам $B$ и $C$ эта конструкция дает кубик $A$. Также нетрудно проверить, что через каждый кубик проходит 13 различных линий. Заметим, что все кубики на этой доске, кроме кубика ( $0,0,0)$, можно разбить на пары с помощью отображения: $x \mapsto-x$. Пусть Петя первым ходом закрасит кубик $(0,0,0)$, после чего каждым ходом он будет закрашивать кубик, парный к предыдущему ходу соперника. Пользуясь этой стратегией, Петя не проиграет. Осталось доказать, что позиция, которая получится в конце игры, не может быть ничейной. Допустим, что ничейная позиция все-таки получилась. В финальной позиции 14 кубиков закрашено Петей. Имеется $C_{14}^{2}=7 \cdot 13$ всевозможных пар Петиных кубиков, и каждая пара $A, B$ задает линию, содержащую некоторый третий кубик $C$. Поскольку позиция ничейная, этот кубик закрашен Васей. Перебирая все $7 \cdot 13$ пар, мы обнаружим $7 \cdot 13$ кубиков, закрашенных Васей. Но Вася закрасил всего 13 кубиков, значит, какой-то из этих кубиков мы «обнаружили» не менее 7 раз. Таким образом, мы обнаружили 7 линий $\left(A_{1}, B_{1}, C\right)$, $\left(A_{2}, B_{2}, C\right), \ldots,\left(A_{7}, B_{7}, C\right)$, проходящих через Васин кубик $C$. У этих линий кубики $A_{i}, B_{i}$ закрашены Петей, и никакие два кубика среди этих 14 кубиков не могут совпадать. Значит, эти линии содержат все 14 кубиков, закрашенных Петей, а остальные 6 линий, проходящие через точку $C$, целиком состоят из Васиных кубиков. Противоречие. ## 3 Общий вариант игры с линиями 3.1. Это теорема 3 из [2]. Вася может украсть стратегию. Действительно, пусть Петя имеет выигрышную стратегию $S_{x}$, согласно которой первым ходом следует покрасить узел $x$. Пусть $\{x, y\}$ - какая-нибудь линия. В силу транзитивности существует выигрышная (для первого игрока) стратегия $S_{y}$, начинающаяся с хода в узел $y$. Тогда Вася может игнорировать тот факт что узел $x$ закрашен и играть как бы за первого по стратегии $S_{y}$. Эта стратегия никогда не потребует от него, чтобы он походил в узел $x$, так как это проигрышный ход, поэтому Вася всегда будет иметь возможность ходить по стратегии. 3.2. Это теорема 4 из [2]. Вася может украсть стратегию. Пусть существует стратегия первого игрока, которая рекомендует первым ходом закрасить множество узлов $S$. Докажем, что у Васи имеется способ не проиграть, каким бы ни был первый ход Пети. Пусть первым ходом Петя закрасил множество $U$. Если $U$ - это всё поле, то игра уже закончилась и Вася не проиграл. В противном случае на доске имеется неокрашенный узел $x$. В силу транзитивности существует выигрышная стратегия первого игрока, начинающаяся с закрашивания множества узлов $S^{\prime}$, содержащего $x$. Пусть тогда Вася закрасит узел $x$ и остальные неокрашенные узлы из $S^{\prime}$, если таковые имеются, после чего будет далее играть в соответствии с этой выигрышной стратегией. Наличие Петиных узлов в $S^{\prime}$ может только облегчить выигрыш, в то время как узлы, закрашенные Петей за пределами $S^{\prime}$, можно мысленно причислить к его второму ходу. 3.3. Ответ: да. Пример из [2]. Расположим 6 узлов в виде двух рядов по три узла, линии это трехточечные множества одного из двух видов: либо это множество содержит по одному узлу из каждой пары, причем в верхнем ряду четное число узлов, либо содержит две точки из одной пары и одну точку из следующей по циклу пары. То, что эта доска транзитивная и первый выигрывает, проверяется непосредственно. 3.4. Это теорема 6 из [2]. Пусть $n=p q$. Будем считать, что на доске нарисовано $q$ непересекающихся кругов, в каждом круге находится $p$ узлов. Положим $p^{\prime}=(p+1) / 2, q^{\prime}=(q+1) / 2$. Рассмотрим семейство $W$, содержащее все множества из $p^{\prime} q^{\prime}$ узлов, каждое из которых содержит по $p^{\prime}$ узлов из каких-то $q^{\prime}$ кругов. В качестве множества линий $L$ возьмем все множества из $p^{\prime} q^{\prime}$ узлов, не входящие в $W$. Транзитивность такой доски тривиальна, если мы ограничим себя перестановками, которые переставляют круги целиком, а также точки в них. Так как каждая пара множеств из $W$ пересекается, для доказательства того, что Первый выигрывает, достаточно проверить, что за первые $p^{\prime} q^{\prime}$ своих ходов он сможет построить множество из $W$ (тогда Второй следующим ходом построит линию). Назовем круг активным, если Первый закрасил в нем не менее одного узла, но менее, чем $p^{\prime}$. Назовем круг заполненным, если Первый закрасил в нем $p^{\prime}$ узлов. Стратегия Первого состоит в том, что он применяет первое из следующих правил, которое удается применить. 1) Если Второй только что ходил в активный круг, то Первый ходит туда же. 2) Если меньше $q^{\prime}$ кругов являются активными или заполненными, Первый ходит в пустой круг. 3) Первый ходит в любой активный круг. Правило 1 означает, что после хода Первый закрасил в этом активном круге больше узлов, чем Второй. Правило 2 означает, что после хода Первого более половины непустых кругов активны или полны. В результате после $p^{\prime} q^{\prime}$ ходов Первый закрасит ровно по $p^{\prime}$ узлов в точности в $q^{\prime}$ кругах. 3.5. Это теорема 24 из [2]. Нужно взять $2 k+1$ копий доски из решения задачи 3.3 (а линиями будем считать множества из трех узлов в каждой из копий, которые ранее там считались линиями). Транзитивность очевидна. Один экземпляр Петя оставит для содержательной игры (в ней он выигрывает по задаче 3.3), а остальные разобьет на пары и будет совершать в них симметричные ходы (и тут он не проиграет). 3.6. Это следствие 25 из [2]. Достаточно модифицировать пример из задачи 3.5. Пусть $k>2 r$ и при этом настолько крупное, насколько нам хочется. Возьмем $n=2 k+1$ экземпляров доски из решения задачи 3.3. Пусть $L-$ множество линий длины 3 на этой доске. Зададим линии длины $r$ : пусть $$ L^{\prime}=\left\{\ell^{\prime} \in[n]^{(r)}: \ell^{\prime} \supset \ell \text { для некоторого } \ell \in L\right\} $$ Петина стратегия та же, что в задаче 3.5: на первой доске Петя поддерживает игру из задачи 3.3 , на остальных парах досок Петя ходит симметрично. В какой-то момент Вася построит линию $\ell$ длины 3 из множества $L$ (а Петя к этому моменту еще не построил такой линии). Если к этому времени сделано $r$ пар ходов, то Вася проиграл: линия $\ell$ плюс остальные его ходы определяют линию из множества $L^{\prime}$. Если же $r$ ходов еще не сделано, то Петя начинает просто тянуть время, делая каждый свой ход на новую доску. 3.7. а) Это предложение 10 из [2]. Пусть $n=2 b, b^{\prime}=(b-1) / 2$. Будем считать, что доска это клетчатый цилиндр $b \times 2$, для которого фиксировано направление обхода и клетки которого - это узлы. Будем обозначать клетки парами $(x, \epsilon)$, где $x$ - остаток по модулю $b$, $\epsilon=0$ или 1. Множество из $b$ клеток будем называть выигрышным, если оно - или содержит по одной клетке в каждом столбце и при этом нечетное число клеток в верхней строке - или содержит ровно один «полный» столбец (в котором обе клетки принадлежат множеству) и при этом «пустой» столбец (не содержащий клеток множества) отстоит от полного не далее чем на $b^{\prime}$ клеток в направлении обхода. Обозначим через $W$ семейство всех выигрышных множеств. Очевидно, выигрышное множество содержит половину всех клеток доски и дополнение выигрышного множества не является выигрышным, так что в семействе $W$ любые два множества пересекаются. Дополнение любого выигрышного множества будем считать линией. Полученная доска транзитивна, поскольку повороты цилиндра, а также одновременные перестановки узлов в двух столбцах сохраняют линии. Проверим, что первый выигрывает на такой доске. Если $A$ - это множество клеток, содержащее не более одной клетки в каждом столбце, обозначим через $\bar{A}$ множество, содержащее вторые клетки из тех столбцов, где имеются клетки $A$. Позиции в игре будем обозначать как пары $(A, B)$, где $A$ - множество клеток, закрашенных к данному моменту первым игроком, а $B$ - множество клеток, закрашенных вторым. Заметим, что в позициях вида $(A \cup(x, \epsilon), \bar{A} \cup(z, \delta))$, где $1 \leqslant z-x \leqslant b^{\prime}$, у первого игрока есть простой форсированный выигрыш: он должен закрасить клетку ( $x, 1-\epsilon$ ) и после этого играть так, чтобы закрасить по одной клетке в каждом столбце, за исключением столбца $z$. Благодаря условию на $x$ и $z$, в конце игры первый закрасит выигрышное множество второго вида (а второй игрок закрасит линию). Опишем теперь стратегию первого игрока. Она состоит в том, что если нет простого форсированного выигрыша, то первый должен оставлять после своего хода позицию вида $(A \cup(x, \epsilon), \bar{A})$ для какого-нибудь множества $A$, содержащего не более одной клетки в каждом столбце и еще одной клетки $(x, \epsilon)$. Допустим, что в второй игрок закрашивает в ответ клетку $(z, \delta)$. Возможны следующие случаи. 1) Если $1 \leqslant z-x \leqslant b^{\prime}$, то первый выиграет далее с помощью форсированного выигрыша. 2) Если $(z, \delta) \neq(x, 1-\epsilon)$, то первый ответит на это закрашиванием клетки $(z, 1-\delta)$, чем обеспечит после своего хода требуемый вид позиции. 3) Если $(z, \delta)=(x, 1-\epsilon)$, то после хода второго получилась позиция $(B, \bar{B})$, в которой все столбцы либо полные (обе клетки раскрашены), либо пустые. Выберем самый первый пустой столбец $x^{\prime}$ и если $x^{\prime}1$, there exists a board such that Alice wins. You may suggest solutions for partial cases of this question. 3.8. Prove that Alice can not win on any board that contains exactly $2^{n}$ nodes. 3.9. Open question. For which prime numbers $n$ there exists a board with $n$ nodes for which Alice wins? For example, what can you say about small prime numbers: $n=3,5,7,11,13$ ? ## "Line avoidance" game - 3 Bursian O., Kokhas D., Kokhas K. ## 4 Special boards We assume that all the boards under consideration are transitive. Denote by $A$ a set of numbers $n$ for which a transitive board with $n$ nodes exists such that any its two lines are intersecting and the number of lines equals $2^{n-1}$. What integers $n$ belong to the set $A$ is an open question. Remind that a permutation is a one-to-one mapping of the set $[n]=\{1,2, \ldots, n\}$ onto itself. A permutation $\tau$ is called a product of permutations $\rho$ and $\sigma$ if $\tau(k)=\rho(\sigma(k))$ for all $k \in[n]$ i.e. if $\tau$ is a composition of the mappings $\rho$ and $\sigma$. Multiplication of permutations is non commutative in the general case. A set $G$ of permutations is called a group if it closed under taking the operation of permutation product. In this case $G$ necessarily contains the identical permutation $e$ and for any $\sigma \in G$ the set $G$ contains permutation $\bar{\sigma}$, which is inverse to $\sigma$, i.e. $\bar{\sigma} \sigma=\sigma \bar{\sigma}=e$. We say that a permutation is a flip if it partitions the points of the set $[n]$ onto pairs and maps each point to the point of its pair. We call a permutation a flop if the length of any of its cycles equals a power of 2 . Let $B$ be a set of numbers $n$ for which there exists a group of permutations of the set $[n]$ that does not contain a flop and such that this group acts transitively on the set $[n]$. 4.1. a) If a flip preserving lines exists for the given board then Alice can not win on that board. b) In the toys shop they sale a board with $n$ nodes where $n$ is even, all the lines are pairwise intersecting and their number is $2^{n-1}$. Prove that there exists a board with $n$ nodes such that Alice wins on it. 4.2. Let a transitive group $G$ of permutations of the set $[2 k]$ contain a permutation $\tau$ that maps some $k$-element set $X$ to its complement $[2 k] \backslash X$. Prove that group $G$ contains a flop. 4.3. Let a board contain $n$ nodes and all the lines be pairwise intersecting. Let there is no a flop that preserves lines on this board. Prove that we can add several lines so that the number of lines is $2^{n-1}$, the board remains to be transitive and there is no a flop preserving lines. 4.4. Let $n=2^{a} \cdot b$ where $b>1$ is odd. Prove that there exists a board with $n$ nodes such that any flip does not preserve the lines. 4.5. Prove that $A=B$. 4.6. Prove that the set $A$ is closed under multiplication i.e. $k, n \in A$ implies that $k n \in A$. 4.7. Prove that if $b>1$ is an odd number then $2 b \in A$. 4.8. Prove that if $b>3$ is an odd number then $4 b \in A$. ## Solutions ## 1 Preliminary problems 1.1. Answer: the first player wins. If all the vertices of dodecahedron are colored in two colors then each face of the dodecahedron contains 2 neighbor vertices colored in the same color. So a draw is impossible in this game. The first player does not lose when uses central symmetry strategy. 1.2. Answer: two possible results of the game are either a draw or the first player wins For a tetrahedron the result is a draw. For octahedron it is easy to see that the first player wins. For any polyhedron the second player can not win because the first player can steal his winning strategy. 1.3. Answer: the second player wins. Enumerate points along the circle. Let the first player color point 1 at his first move. Then the second player colors point 98 in his first move and after that uses symmetrical (with respect to a symmetry $k \leftrightarrow 99-k$ ) strategy. 1.4. Answer: the second player wins for $n \geqslant 6$. The first player can not win due to problem 3.2. A draw is impossible for $n \geqslant 6$ because Ramsey number $R(3,3)=6$. ## 2 Boards with intersecting lines 2.1. a) A set and its complement can not both be lines. And just in case, if the number of lines is $2^{n-1}$, then in any pair of the complementary sets exactly one is a line. b) Let $X$ be a set of the board nodes, $G$ be an arbitrary subset of $X$. If $F$ is a line, $F \subset G$, then $F \cap(X \backslash G)=\varnothing$. Therefore $X \backslash G$ is not a line. Then $G$ is a line by p. a). 2.2. [1, lemma 2.2]. Let $X$ be a set of the board nodes and $\mathcal{L}$ be a set of the lines. We know that any two lines intersect. The set of lines that we want to construct should satisfy $2.1 \mathrm{~b}$ ). Consider all the sets of nodes that contain line, let $$ \mathcal{L}_{1}=\{B \subset X: A \subset B \text { for some line } A \in \mathcal{L}\} $$ Particularly, $\mathcal{L} \subset \mathcal{L}_{1}$. If there exists a set $B$ that intersects any line and in the same moment does not contain any line (it means that its complement $\bar{B}$ intersects any line and does not contain any line, too) then it's worth to assign one of the sets $B, \bar{B}$ to be a "new" line. Choose one of the two sets that contains more nodes (in the case of equality of the numbers of nodes we can choose any of them). Let $\mathcal{L}_{2}$ be a family of sets that we choose by the described way. Check that $\mathcal{L}_{1} \cup \mathcal{L}_{2}$ is a family of pairwise intersecting lines that contains exactly $2^{n-1}$ elements. It's clear by the construction that any two lines in this family are intersecting. Let $Y$ be an arbitrary subset of $X$ and $\bar{Y}$ be its complement. If $Y$ contains a line then $Y \in \mathcal{L}_{1}$; else if $Y$ does not intersect any line then $\bar{Y} \in \mathcal{L}_{1}$; in all other cases $Y$ or $\bar{Y}$ is contained in $\mathcal{L}_{2}$. 2.3. It's the proposition of $\mathrm{Ph}$. Zumstein. If $L=\left\{x_{1}, x_{2}, \ldots, x_{k}\right\}$ is a minimal line and no one of its points is a common point of the intersection of all sets then for any $i, 1 \leqslant i \leqslant k$, there exists a line $L_{i}$ for which $x_{i} \notin L_{i}$. But then the intersection $L \cap L_{1} \cap \ldots \cap L_{k}$ is empty. ### 2.4. Answer: Alice wins. We make some notes about construction of the game board. First of all any two distinct cubes indeed define a line consisting of 3 cubes (in the sense that the third cube found by the receipt of problem statement does not coincide with the two initial ones). Further, the rule describing lines is very symmetrical: if construction from the problem statement gives the cube $C$ by the cubes $A$ and $B$ then the same construction gives the cube $B$ by the cubes $A$ and $C$ and the cube $A$ by the cubes $B$ and $C$. It is not hard to check also that each cube belongs to 13 different lines. Note that all cubes on this board except the cube $(0,0,0)$ can be partitioned into pairs by the mapping $x \mapsto-x$. Let Alice color the cube $(0,0,0)$ at her first move. After that let she color at each move the cube corresponding to the cube colored by her opponent at the previous move. Alice can not lose using this strategy. It remains to prove that the final position of the game can not be a draw. Assume that we have a draw in the end of the game (no matter what strategy the players used). 14 cubes in this position are colored by Alice. Each of $\binom{14}{2}=7 \cdot 13$ pairs of these cubes defines a line that contains a third cube $C$. Since we have a draw here this cube is colored by Bob. So for each 7 - 13 pairs we detect a cube colored by Bob. But Bob has colored 13 cubes totally, therefore we detect one of his cubes at least 7 times. Thus, we find 7 lines $\left(A_{1}, B_{1}, C\right),\left(A_{2}, B_{2}, C\right), \ldots$, $\left(A_{7}, B_{7}, C\right)$, that contain Bob's cube $C$. All the 14 cubes $A_{i}, B_{i}$ here are colored by Alice and no two of them coincide. Hence these lines contain all Alice's cubes and then other 6 lines that pass trough $C$ consist of Bob's cubes. A contradiction. ## 3 The general case of "Line avoidance" game 3.1. $[2$, theorem 3]. If a winning strategy for the first player exists, it can start by coloring of an arbitrary node. Denote by $S_{z}$ the strategy that starts from the coloring of node $z$. Let Alice use a winning strategy $S_{x}$. We will prove that Bob can steal Alice's strategy. Let $\{x, y\}$ be an arbitrary line that contains node $x$. Then Bob can ignore the fact that the node $x$ has been already colored and plays as the first player according to the strategy $S_{y}$. Since Alice has colored node $x$ she will never color it again. As to Bob his strategy $S_{y}$ never requires to color node $x$ because he loses in that case. Therefore Bob can always move according to the strategy $S_{y}$. So he will not lose. 3.2. $[2$, theorem 4]. Let a winning strategy for Alice exist, assume that the first move of Alice according to this strategy is a coloring of a set of nodes $S$. We will prove that for each Alice's first move Bob has an answer that allows him do not lose. Let Alice color a set $U$ at the first move. If $U$ is the whole board then the game is already finished and Bob has not lost. In the opposite case there exists node $x$ in the board that has not been colored yet. The game is transitive, so there exists a winning strategy $F$ for the first player that begins from coloring of a set $S^{\prime}$ containing $x$. Let Bob color node $x$ and those nodes of $S^{\prime}$ that are not colored yet. Let he play after that according to the strategy $F$. There are 2 obstacles in this way. 1) In fact some nodes in $S^{\prime}$ were possibly colored by Alice. It is not a disadvantage for Bob, and does not prevent him to use strategy $F$ due to avoidance nature of the game. 2) Several nodes outside the set $S^{\prime}$ (namely, the nodes in $U \backslash S^{\prime}$ ) have been colored by Alice before Bob's first move. Then consider Alice's second move, let she color set $V$. Now Bob can think that the game begins from coloring of $S^{\prime}$ by him at the first move and continues with coloring of $U \backslash S^{\prime}$ and $V$ by Alice at the second move. 3.3. Answer: yes. Example from [2]. Arrange 6 nodes in 2 rows, 3 nodes in a row, let lines be the sets containing 3 points of one of two kinds: either this set contains exactly one of two nodes from each column and even number of nodes in the upper row or it contains two points from one column and one point from the next column in cyclic order. It is now verified directly that this board is transitive and the first player wins. 3.4. $[2$, theorem 6]. Let $n=p q$. We assume that $q$ non intersecting circles are drawn on the board, each circle contains $p$ nodes. Let $p^{\prime}=(p+1) / 2, q^{\prime}=(q+1) / 2$. Choose an arbitrary $q^{\prime}$ circles and choose $p^{\prime}$ nodes in each of these circles. Consider a family of all sets $W$ consisting of $p^{\prime} q^{\prime}$ nodes that can be obtained by this way for all choices. Let lines be sets of $p^{\prime} q^{\prime}$ nodes that do not belong to $W$. Transitivity of this board is provided by permutations that permute all nodes in each circle only or permute the whole circles only and by compositions of these permutations. It is clear that any two sets of $W$ are intersecting. So if we want to prove that Alice wins it is sufficient to check that she can construct a set from $W$ after her $p^{\prime} q^{\prime}$ moves (then Bob on his next move will construct a line). We call a circle active if Alice has colored in it at least one node but less than $p^{\prime}$ nodes. We call a circle full if Alice has colored $p^{\prime}$ nodes in it. The winning strategy for Alice is to apply the first rule from the list below that could be applied. 1) If the last move of Bob was in active circle then Alice chooses any node in this active circle. 2) If less than $q^{\prime}$ circles are active or full then Alice chooses any node in empty circle. 3) Alice chooses any node in any active circle. The first rule guarantees that Alice always has colored more nodes than Bob in any active circle. The second rule guarantees that more than one half of non empty circles are active or full after Alice's move. According to this strategy after $p^{\prime} q^{\prime}$ Alice will color exactly $p^{\prime}$ nodes in exactly $q^{\prime}$ circles. 3.5. [2, theorem 24]. Let a board consist of $2 k+1$ copies of the 6 -node board from solution 3.3 (and lines are sets of 3 nodes, belonging to the same copy and forming a line in it in a sense of solution 3.3). The transitivity of this board is clear. Now let Alice play on the first board as in the solution 3.3. All other boards she splits onto pairs and plays symmetrically on them. 3.6. [2, corollary 25]. We modify the example of the solution 3.5 . Let $k>2 r$ and $k$ be as big as we will need. Consider the 6-node board from solution 3.3, let $L$ be the set of lines on this board as in solution 3.3, each line contains 3 nodes. Consider a big board consisting of $n=2 k+1$ copies of 6 -node board. Define a set of lines in it (each lines contains $r$ nodes) as $$ L^{\prime}=\left\{\ell^{\prime} \text { is a set of } r \text { nodes : } \ell^{\prime} \supset \ell \text { for some } \ell \in L\right\} $$ Alise's strategy looks like that in the problem 3.5: Alice plays the game from the problem 3.3 on the first board, and plays symmetrically on others. Then at some moment of time Bob constructs a 3-node line $\ell \in L$ (and Alice has constructed no lines from $L$ yet). If $r$ pairs of moves have been made till this moment, Bob loses because line $\ell$ plus all other his moves form a line from the set $L^{\prime}$. If less than $r$ pairs of moves have been made, then Alice begins just to wait making each move on a new board. 3.7. a) $[2$, proposition 10$]$. Let $n=2 b, b^{\prime}=(b-1) / 2$. We may think that the board is a square cylinder $b \times 2$, for which we fix the (cyclic) direction of its long side and the nodes are the cells of the cylinder. Denote nodes by pairs $(x, \epsilon)$, where $x$ is a residue modulo $b, \epsilon=0$ or 1 . We call a set of $b$ cells winning if - either it contains one cell in each column and the total number of cells in the upper row is odd; - or it contains the unique "full" column (i.e. both cells of this column belong to the set) and the "empty" column (without cells of the set) is situated at most $b$ ' cells later in the chosen direction. Denote by $W$ the set of all winning sets. It is clear that a winning set contains one half of all cells of the board and that the complement of the winning set is not a winning set. Therefore any 2 sets in $W$ intersect. We call a set a line if it is the complement of a winning set. The obtained board is transitive because the rotations of the cylinder along the long side and along the short side preserve lines. Check that Alice wins on this board. For a set $A$ that contains at most one cell in each column denote by $\bar{A}$ a set that contains the second cells of those columns which contain cell of $A$. In each moment denote positions of a game by pair $(A, B)$ where $A$ is the set that has been colored by Alice and $B$ is the set that has been colored by Bob. Observe that for positions of the form $(A \cup(x, \epsilon), \bar{A} \cup(z, \delta))$, where $1 \leqslant z-x \leqslant b^{\prime}$, Alice has a simple forced victory: she just should color cell $(x, 1-\epsilon)$ and then continue a game by coloring one cell in each column except column $z$. Due to the restrictions on $x$ and $z$ Alice will color a winning set of the second type at the end of the game (and hence Bob will color a line). Now describe the strategy of Alice. If there are no simple forced victory, Alice should leave to Bob after her move a position of the form $(A \cup(x, \epsilon), \bar{A})$, where set $A$ contains at most one cell in each column, and $(x, \epsilon)$ is an arbitrary additional cell. Look at this strategy in details. Assume that Bob colors cell $(z, \delta)$ in his next move. Then the following cases are possible. 1) $1 \leqslant z-x \leqslant b^{\prime}$. Then Alice has a forced victory. 2) $(z, \delta) \neq(x, 1-\epsilon)$. Then Alice colors cell $(z, 1-\delta)$ and obtains a position in agreement with her strategy. 3) $(z, \delta)=(x, 1-\epsilon)$. Here we have after Bob's move a position of the form $(B, \bar{B})$, in which all columns are either full (both cells are colored) or empty. Then choose the first empty column $x^{\prime}$. If $x^{\prime}1$, then $2 b \in A$. (iv) if $b$ is odd and $b>3$, then $4 b \in A$. (v) $A$ has density 1. The pioneering work on this question was done by Isbell in the 1960s. He conjectured [10] the existence of a function $m$ such that, if $b$ is odd and $a \geqslant m(b)$, then $2^{a} . b \notin A$. He claimed to have proved this statement in [11], but the proof is incorrect, and the question is still open. In this direction, Theorem 1(i) is straightforward, and (ii) is established in [4]. However, the obvious conjecture, that $2^{a} . b \notin A$ if $b$ is odd and $2^{a}>b$, is false: Peter Neumann (personal communication) showed that $56 \in A$, and his construction easily extends to give two infinite families of examples. We give a variety of constructions in section 4. We also give lower bounds for the size of a set in such a family. THEOREM 3. Let $\mathscr{F}$ be an intersecting family of size $2^{n-1}$ and $F \in \mathscr{F}$. (i) If $\mathscr{F}$ is transitive, then $|F| \geqslant n^{1 / 2}$. (ii) If $\mathscr{F}$ is regular, then $|F| \geqslant \frac{1}{2} \log _{2}(\pi n / 2)$. The first inequality is essentially best possible; but we have no examples remotely near the second. We also investigate the function $f(n)$ defined by $2^{n-1}-f(n)=\max \{|\mathscr{F}|: \mathscr{F}$ is a transitive intersecting family of subsets of an $n$-set $\}$. (Another interpretation of $f(n)$ will be given later.) Clearly $f(n)=0$ iff $n \in A$; we evaluate $f(n)$ in some other cases: THEOREM 4. (i) $f\left(2^{a}\right)=\Sigma_{i=0}^{a-1} 2^{2^{i}}$ for $a \geqslant 1$. (ii) $f(12)=48$. We also consider $t$-wise intersecting families with $t>2$, that is, families in which any $t$ sets have non-empty intersection. The family of all sets containing $x$ is $t$-wise intersecting for all $t$ and has cardinality $2^{n-1}$. But with our global assumptions, 4 -wise intersecting families are smaller: THEOREM 5. (i) If $\mathscr{F}$ is 4 -wise intersecting and transitive, then $|\mathscr{F}| \leqslant 2^{n} . \exp \left(-(n \ln 2 / 2)^{1 / 3}\right)$. (ii) If $\mathscr{F}$ is 4-wise intersecting and regular and closed under taking supersets, then $|\mathscr{F}| \leqslant 2^{n} \cdot n^{-0.347}$. Problem. Is it true that a 3-wise intersecting transitive (or regular) family $\mathscr{F}$ must have $|\mathscr{F}|=o\left(2^{n}\right) ?$ The method of proof of Theorem 5 gives a result which is of independent interest. THEOREM 6. Let $G$ be a transitive permutation group on an $n$-set $X$, and $T$ a $t$-subset of $X$. let $$ \mathscr{F}=\{F: F \subseteq X, F \cap g(T) \neq \varnothing \text { for all } g \in G\} $$ Then $|\mathscr{F}| \leqslant\left(2^{t}-1\right)^{n / t}$. Equality holds, for example, if $T$ is a block of imprimitivity for $G$. REMARK. By analogy with the alternative characterisation of the set $A$ given before Theorem 1, we could define, for any prime $p$, the set $$ \begin{aligned} A_{p}= & \{n \in \mathbb{N}: \text { there is a transitive permutation group of } \\ & \text { degree } n \text { containing no fixed-point-free } p \text {-element }\} \end{aligned} $$ and conjecture the existence of a function $m_{p}$ suc that, if $n=p^{a} . b$ with $p \nmid b$ and $a \geqslant m_{p}(b)$, then $n \notin A_{p}$. It is trivial that if $n=p^{a} . b$ with $a \geqslant 1$ and $b1$ contains a fixedpoint-free $p$-element for some prime $p$. We always use log to denote logarithms to base 2, and in for natural logarithms. We are grateful to Roy Meshulam and Peter Neumann for helpful discussions. ## 2. Transitive Intersecting Families For the remainder of the paper, $X$ denotes a set of cardinality $n$, and $\mathscr{F}$ a family of subsets of $X$. Lemma 2.1. Let $\mathscr{F}$ be an intersecting family. Then $|\mathscr{F}| \leqslant 2^{n-1}$; equality holds if and only if $\mathscr{F}$ contains one of each complementary pair of subsets of $X$. If $|\mathscr{F}|=2^{n-1}$, then $\mathscr{F}$ is closed under taking supersets. Proof. All but the last sentence follows from the observation that $\mathscr{F}$ contains at most one of each complementary pair of subsets. If $|\mathscr{F}|=2^{n-1}, F \in \mathscr{F}$, and $G \supseteq F$, then $F \cap(X \backslash G)=\varnothing$, so $X \backslash G \notin \mathscr{F}$, and $G \in \mathscr{F}$. A set $B$ is called a blocking set for the family $\mathscr{F}$ if it meets every member of $\mathscr{F}$ and contains none; that is, $F \cap B \neq \varnothing$ and $F \cap(X \backslash B) \neq \varnothing$ for all $F \in \mathscr{F}$. Obviously, the complement of a blocking set is a blocking set. Lemma 2.2. Let $\mathscr{F}$ be an intersecting family. Let $$ \mathscr{F}_{1}=\{G \subseteq X: F \subseteq G \text { for some } F \in \mathscr{F}\} $$ and let $\mathscr{F}_{2}$ consist of one of each complementary pair of blocking sets of $\mathscr{F}$, chosen according to the rule that if $|B| \neq|X \backslash B|$ then the larger set is chosen. Then $\mathscr{F}_{1} \cup \mathscr{F}_{2}$ is an interesting family of cardinality $2^{\text {n-1 }}$ containing $\mathscr{F}$. ProOF. It is clear that $\mathscr{F}_{1} \cup \mathscr{F}_{2}$ is intersecting. Let $Y$ be any subset of $X$. If $Y$ contains a member of $\mathscr{F}$, then $Y \in \mathscr{F}_{1}$; if $Y$ is disjoint from a member of $\mathscr{F}$, then $X \backslash Y \in \mathscr{F}_{1}$; and if neither holds, then $Y$ is an $\mathscr{F}$-blocking set, so either $Y$ or $X \backslash Y$ is in $\mathscr{F}_{2}$. Proposition 2.3. Let $\mathscr{F}$ be an intersecting family, and $G$ a subgroup of $\operatorname{Aut}(\mathscr{F})$ which contains no fixed-point-free 2-element. Then there is an intersecting family $\mathscr{F}^{\prime} \supseteq \mathscr{F}$ with $\left|\mathscr{F}^{\prime}\right|=2^{n-1}$ and $G \leqslant \operatorname{Aut}\left(\mathscr{F}^{\prime}\right)$. Proof. A permutation of $X$ interchanges some complementary pair of subsets iff all its cycles have even length, that is, iff some odd power of it is a fixed-point-free 2-element (interchanging the same pair of sets). So, if $G$ contains no fixed-point-free 2-element, then the orbits of $G$ on $\frac{1}{2} n$-sets fall into 'dual' pairs, the dual of an orbit consisting of all complements of sets in that orbit. Thus the choices in Lemma 2.2 can be made in a $G$-invariant way. REMARK. If $n$ is odd, there are no fixed-point-free 2-elements; so any intersecting family can be enlarged to one of size $2^{n-1}$ without destroying any automorphisms. From this result, we obtain Isbell's characterisation of the set $A$ : COROLLARY $2.4 n \in A$ iff there is a transitive permutation group of degree $n$ containing no fixed-point-free 2-element. PrOOF. The reverse implication is immediate from Proposition 2.3, taking $\mathscr{F}=\varnothing$. For the forward implication, the first sentence of the proof of the proposition shows that, if $\mathscr{F}$ is intersecting and $|\mathscr{F}|=2^{n-1}$, then Aut $(\mathscr{F})$ contains no fixed-point-free 2-elements. Let $m_{r}(n)$ and $m_{t}(n)$ be the maximum sizes of regular (resp. transitive) intersecting families of sets; and, for $n$ even, let $m_{r}^{\prime}(n)$ and $m_{r}^{\prime}(n)$ be the maximum sizes of regular (resp. transitive) intersecting families of $\frac{1}{2} n$-sets. For $0 \leqslant k \leqslant n$, and any family $\mathscr{F}$, let $\mathscr{F}_{k}$ denote the family $$ \{F \subseteq X: F \in \mathscr{F},|F|=k\} $$ Lemma 2.5 (i) if $\mathscr{F}$ is transitive, then $\mathscr{F}_{k}$ is transitive for all $k$. (ii) If $F_{k}$ is regular for all $k$, then $\mathscr{F}$ is regular. The proof is obvious. The converse of (i) is clearly false, since we may take transitive families $\mathscr{F}_{k}$ and $\mathscr{F}_{l}$ for which $\operatorname{Aut}\left(\mathscr{F}_{k} \cup \mathscr{F}_{l}\right)=\operatorname{Aut}\left(\mathscr{F}_{k}\right) \cap \operatorname{Aut}\left(\mathscr{F}_{l}\right)$ is intransitive. It is true, however, that if $\operatorname{Aut}\left(\mathscr{F}_{k}\right)$ is transitive and $\operatorname{Aut}\left(\mathscr{F}_{l}\right)=S_{n}$ for all $l \neq k$, then $\operatorname{Aut}(\mathscr{F})$ is transitive. The converse of (ii) is false, even if $\mathscr{F}$ is intersecting and $|\mathscr{F}|=2^{n-1}$; as the following example shows. EXAmPLE 2.6. Let $X$ be a large finite set whose cardinality is coprime to 10 . Choose a random partition of $X$ into 10 parts, labelled with the 2 -subsets of $\{1, \ldots, 5\}$, say $X_{12}, \ldots$, $X_{45}$. let $A_{i}$ be the union of the four parts whose labels contain $i$; then $A_{i} \cap A_{j}=X_{i j}$ for $i \neq j$. Let $P$ be a cyclic group of permutations of $X$ (any transitive group of polynomially bounded size would do), and set $$ \mathscr{A}=\left\{\pi\left(A_{i}\right): \pi \in P, 1 \leqslant i \leqslant 5\right\} $$ It can be shown that, with probability exponentially close to $1, \mathscr{A}$ is an intersecting antichain and $\left|A_{i}\right|<\frac{1}{2} n$ for all $i$. Now let $\mathscr{A}^{\prime}$ be obtained as in Proposition 2.3; that is, $\mathscr{A}^{\prime}$ consists of all sets containing a member of $\mathscr{A}$ together with all blocking sets of size greater than $n$. Finally, set $$ \mathscr{F}=\mathscr{A}^{\prime} \backslash\left\{A_{1}, \ldots, A_{5}\right\} \cup\left\{X \backslash A_{1}, \ldots, X \backslash A_{5}\right\} $$ Then $\mathscr{F}$ is intersecting and $|\mathscr{F}|=\left|\mathscr{A}^{\prime}\right|=2^{n-1}$. Also, $\mathscr{A}^{\prime}$ is transitive, hence regular; and, since each point lies in two of the five sets $A_{i}$, the passage from $\mathscr{A}^{\prime}$ to $\mathscr{F}$ increases the degree by one. Moreover, since $5 \nmid n$, not all $A_{i}$ have the same size, so $\mathscr{F}_{k}$ is not regular for some $k$. PROPOSITION 2.7 (i) $2^{n-1}-m_{t}(n)=\frac{1}{2}\binom{n}{n}-m_{i}^{\prime}(n)$. (ii) $2^{n-1}-m_{r}(n) \leqslant \frac{1}{2}\binom{n}{2_{n}^{n}}-m_{r}^{\prime}(n)$. Proof. The inequality $\leqslant$ is obtained in each case by taking a transitive or regular intersecting family of $n$-sets of maximum size and adjoining all sets of size greater than $\frac{1}{2} n$ (using Lemma 2.4(11) and the remark following it). The reverse inequality is obtained by taking a transitive intersecting family $\mathscr{F}$ of maximum size and considering $\mathscr{F}_{\frac{1}{n} n}$. We do not know whether equality necessarily holds in (ii). However, it is known for which $n$ the right-hand side is zero, in view of the following result of Brace and Daykin [2]: THEOREM $2.8 m_{r}^{\prime}(n)=\frac{1}{2}\left(\frac{n}{2} n\right)$ iff $n$ is not a power of 2. It follows that $m_{r}(n)=2^{n-1}$ if $n$ is not a power of 2 . Problem 2.9. For $n=2^{a}, a \geqslant 1$, is there a regular intersecting family of cardinality $2^{n-1}$ ? If $\mathscr{F}$ were such a family, then $\mathscr{F}_{k n}$ would not be regular; this easily implies that $\mathscr{F}$ would contain sets of cardinality less than $\frac{1}{2} n$. It can be shown that no such family exists for $n=2,4,8$. ## 3. On Numbers Not in A Let $G$ be a permutation group on $X$, where $|X|=n$. We say that a subset $Y$ of $X$ is flipped by $G$ if some element of $G$ interchanges $Y$ with its complement. (Such a permutation must have all its cycles of even length; and, if $Y$ is flipped by $G$, then some fixed-point-free 2-element flips it.) Let $f(n)$ be the minimum, over all transitive permutation groups $G$ of degree $n$, of the number of sets flipped by $G$. Clearly we can take the minimum over all minimal transitive groups $G$, those such that every proper subgroup of $G$ is intransitive. LEMMA 3.1. (i) $m_{t}^{\prime}(n)=\frac{1}{2}\binom{n}{n}-\frac{1}{2} f(n)$. (ii) $m_{t}(n)=2^{n-1}-\frac{1}{2} f(n)$. Proof. (i) The $\frac{1}{2} n$-sets not flipped by a group $G$ fall into dual pairs of orbits, as in Proposition 2.3. A maximal $G$-invariant intersecting family of $\frac{1}{2} n$-sets consists of the union of one of each dual pair of orbits. (ii) Clear from (i) and Proposition 2.7(i). Hence $f(n)=0$ iff $n \in A$. The first family of numbers not in $A$ are the powers of 2 , about which the next result gives complete information. THEOREM 3.2 (i) For $n>1, n$ is a power of 2 iff every transitive permutation group of degree $n$ contains a fixed-point-free involution. (ii) If $n=2^{a}(a \geqslant 1)$, then $n \notin A$ and $f(n)=\Sigma_{i=0}^{a-1} 2^{2 i}$; moreover, the configuration of $f(n)$ sets of size $\frac{1}{2} n$ is unique. Proof. Let $G$ be a transitive group of degree $2^{a}(a \geqslant 1)$. A Sylow 2 -subgroup of $G$ is transitive; without loss of generality, we may assume that $G$ is a 2-group. Let $z$ be an involution in the centre of $G$. Then $z$ is fixed-point-free and flips $2^{2 u-1}$ sets, namely, all those containing one point from ech cycle of $z$. Now let $B$ be the set of cycles of $z$, and $H$ the group induced on $B$ by $G$. Then $H$ is a transitive 2-group of degree $2^{a-1}$. By induction, $H$ flips at least $\sum_{i=0}^{a-2} 2^{2 i}$ sets. Each of these yields a set (the union of the corresponding cycles of $z$ ) flipped by $G$; and none of these coincides with a set already described, since none of the earlier sets is a union of cycles of $z$. Thus $$ f(n) \geqslant \sum_{i=0}^{a-1} 2^{2^{i}} $$ The cyclic group of order $n$ realises the value $\Sigma_{i=0}^{a-1} 2^{2^{i}}$, since for $j=0, \ldots, a-1$, the elements of order $2^{j+1}$ are fixed-point-free and all flip the same $2^{2 a j-1}$ sets, all of which are fixed by elements of smaller order. The uniqueness of the minimal configuration also follows from the above inductive proof: the configuration on $2 n$ points is found from that on $n$ points by replacing each point $x$ by two points $x_{1}, x_{2}$, and then adjoining each set which contains exactly one point from each pair. Finally, we prove the converse of (i). Suppose that $n=2^{a} \cdot b$, with $b$ odd and $b>1$. Let $X=\left\{x_{i j}: 1 \leqslant i \leqslant 2^{a}, 1 \leqslant j \leqslant b\right\}$. For $1 \leqslant j \leqslant b$, let $s_{j}$ be the permutation $\left(x_{1 j} x_{2 j} \cdots x_{2 j}\right)$, and let $t$ be the permutation $$ \prod_{i=1}^{2^{a}}\left(x_{i 1} x_{i 2} \cdots x_{i b}\right) $$ Let $$ N=\left\{s_{1}^{k_{1}} \cdots s_{b}^{k_{b}}: k_{1}+\cdots+k_{b} \equiv 0\left(\bmod 2^{a}\right)\right\} $$ and let $C$ be the group generated by $t$. The product $G=N C$ is transitive of degree $n$. Any involution $z$ in this group lies in $N$ and has the form $s_{1}^{e_{1}} \cdots s_{b}^{e_{b}}$, where $e_{j}=0$ or $2^{a-i}$ for $j=1, \ldots, b$; since $b$ is odd, $e_{j}=0$ for some $j$, and $z$ fixes the point $x_{1 j}$. Remark 3.3. (i) Note that $m_{t}^{\prime}(2)=m_{t}^{\prime}(4)=0$; these are the only even numbers $n$ for which $m_{t}^{\prime}(n)=0$. (ii) The extremal configuration of $m_{t}^{\prime}\left(2^{a}\right)$ intersecting sets is not unique, because of the choices of orbits involved. (ii) One remaining problem is that of determining the minimal transitive 2-groups, other than cyclic groups, which flip exactly $f(n)$ sets. (Both the Klein group of order 4 and the quaternion group of order 8 have this property.) By Theorem 1 , we have $3.2^{a} \notin A$ for $a \geqslant 2$. For the first of these values, we can compute the value of $f$ (Theorem 4(ii)). Proof OF TheOrem 4(ii). It is straightforward to show that the group defined in the last part of the proof of Theorem 3.2 flips exactly 48 sets. We must establish that any minimal transitive group of degree 12 flips at least 48 sets. Lemma 3.4. (i) The only minimal transitive groups of degree 6 are the cyclic and dihedral groups (acting regularly) and the alternating group $A_{4}$ (acting on the cosets of a subgroup of order 2). (ii) Any minimal transitive group of degree 12 is a $\{2,3\}$-group. ProOF. We give the argument for (ii); (i) is similar but easier. First, using Sims' list of primitive groups [14] and the subgroup structure of these groups, we see that a minimal transitive group of degree 12 is necessarily imprimitive. If it has a block of imprimitivity of size 3 or 4 , it is contained in $S_{3}$ wr $S_{4}$ or $S_{4}$ w $S_{3}$, and so is necessarily $a\{2,3\}$ group. If it has a block of size 2 , then either the group $\bar{G}$ induced on the set of blocks is minimal transitive of degree 6 (in which case the result follows from (i)), or else a proper transitive subgroup of $\bar{G}$ acts with two orbits of length 6 (each of which is also a block for $G$ ). So we may assume that $G$ has two blocks of size 6 , and that the group induced by a block on its setwise stabiliser is one of PSL(2, 5), PGL $(2,5), A_{6}$ or $S_{6}$. Let $H$ be the subgroup fixing the two blocks. Then $H$ has a unique minimal characteristic subgroup, which is normal in $G$; by minimality, we may assume that $H$ is $\operatorname{PSL}(2,5), \operatorname{PSL}(2,5) \times \operatorname{PSL}(2,5), A_{6}$, or $A_{6} \times A_{6}$ if $H=S \times S$, then $G=S$ wr $C_{2}$, and $G$ contains a proper transitive subgroup $S_{0} \times C_{2}$, where $S_{0}$ is a diagonal subgroup of $S \times S$. If $H$ is simple, $g \in G \backslash H$, and $K$ is a transitive subgroup of $H$ normalised by $g$, then $K\langle g\rangle$ is a transitive subgroup of $G$. The proof is finished by showing that any coset of $\operatorname{Inn}(H)$ in $\operatorname{Aut}(H)$ contains an automorphism which normalises a transitive subgroup $K$ of $H$. ( $K$ can be chosen to be a Sylow 2- or 3-normaliser of $H$.) We return now to the proof that $f(12)=48$. Let $G$ be a transitive group of degree 12 . We may assume that $G$ contains no fixed-point-free involution, since such an involution flips 64 sets. Also, we may assume that $G$ is minimal transitive. From this it follows, both that $G$ is a $\{2,3\}$-group (Lemma 3.4(ii)), and that any odd permutation in $G$ must be fixed-point-free (for otherwise the even permutations in $G$ would form a proper transitive subgroup). By Burnside's theorem, $G$ is solvable, and so a minimal normal subgroup $N$ of $G$ is elementary abelian, and has orbits of length 2,3 or 4 . We treat these cases in turn. $N$-orbits of length 2 . We can identify $N$ with a binary linear code of length 6 , in which all words have even weight and no word has weight 6 , and which has a transitive automorphism group. Up to isomorphism, the only such code is spanned by ( $\left.\begin{array}{lllll}1 & 1 & 1 & 1 & 0\end{array}\right)$ and (1 100011 ). So $N \simeq V_{4}$. Moreover, $G / N$ is a minimal transitive group of degree 6 , necessarily regular or isomorphic to $A_{4}$, by Lemma 3.4(i). If $G / N$ is regular, then a Sylow 2-subgroup $P$ of $G$ has order 8 , and all its involutions lie in $N$ (since elements outside $N$ are fixed-point-free); so $P$ is abelian, and $P \leqslant C_{G}(N)$. If $G / N \simeq S_{3}$, then $G$ is generated by its Sylow 2-subgroups, and $N \leqslant Z(G)$; if $G / N \simeq C_{6}$, then $G$ has a normal Sylow 2-subgroup $P \simeq C_{4} \times C_{2}$ admitting no automorphism of order 3 , so we have $P \leqslant Z(G)$. This is impossible, since $Z(G)$ is semiregular. So we have $G / N \simeq A_{4}$. Then $G$ has a normal Sylow 2 -subgroup $P$ of order 16 with an automorphism of order 3 acting non-trivially on both $N$ and $P / N$. This implies that $P$ is abelian, whence regular on each of its orbits of length 4 . Thus there are only three distinct stabilisers in $P$, containing between them 10 elements. The remaining 6 elements of $P$ fall into 3 inverse pairs of elements of order 4, each a product of two 2-cycles and two 4-cycles and so flipping 16 sets; and there is no overlap, since a set flipped by such an element contains two $N$-orbits lying in distinct $P$-orbits and the inverse pair is determined by these $P$-orbits. So $G$ flips 48 sets. $N$-orbits of length 3 . Now $N$ is an elementary abelian 3-group; and, if $K$ is the kernel of the action of $G$ on the set of $N$-orbits, then $|G / K|=4$. If $|N|=3$, then any 2-element normalises $N$, so $N$ and a Sylow 2-subgroup generate a regular subgroup, containing a fixed-point-free involution. So we may assume that $|N| \geqslant 9$. From the fact that each coset of $K$ in $G$ except $K$ itself contains a fixed-point-free 2-element, we see that, for each pair of $N$-orbits, there is a set flipped by $G$ containing two points from each of these orbits and one point from each of the other two. No such set can be fixed by an element of $N$, so there are at least $6 \cdot 9=54$ sets flipped by $G$. $N$-orbits of length 4 . Since $N$ acts regularly on each orbit, the number of fixed points of any element of $N$ is a multiple of 4 . Since the average number of fixed points is the number of orbits, namely 3 , some element of $N$ is a fixed-point-free involution. REMARK 3.5. The proof of Theorem 1(ii) given by [4] is a straightforward induction starting with the case $n=12$. Can similar methods yield lower bounds for $f\left(2^{a} .3\right)$ for $a \geqslant 3$ ? ## 4. NUmbers IN A In this section, we prove Theorem 2, and give a variety of constructions exhibiting further members of $A$. PROOF OF THEOREM 2. (i) If $G_{i}$ is a transitive permutation group on $X_{i}$ containing no fixed-point-free 2-element for $i=1,2$, then $G_{1} \times G_{2}$ acting on $X_{1} \times X_{2}$ has the same property. (ii) A set of odd size admits no fixed-point-free 2-element. (iii) The group constructed in Theorem 3.2, with degree $n=2^{a} \cdot b$, with $b>1$ odd, contains no fixed-point-free involution; but, if $a=1$, it contains no 2-element of order greater than 2. (v) It follows from (i)-(iii) that, if $B$ is the set of natural numbers having at least as many odd as even prime factors (counted with multiplicities), then $B \subseteq A$; it suffices to show that $B$ has density 1 . Take $\varepsilon>0$ arbitrary. Choose $k$ so large that $2^{k}>1 / \varepsilon$. let $\mathbb{N} \backslash B=C_{1} \cup C_{2}$, where $$ C_{1}=\left\{n \in \mathbb{N} \backslash B: 2^{k+1} \mid b\right\}, \quad C_{2}=\mathbb{N} \backslash B \backslash C_{1} $$ Then $C_{1}$ has density at most $1 / 2^{k+1}<\varepsilon / 2$. Also, every number in $C_{2}$ has at most $2 k$ prime factors; it suffices to show that the set $D_{m}$ of numbers with exactly $m$ prime factors has density 0 . This is true because $$ \sum_{\substack{n \leqslant X \\ n \in D_{m}}} 1 / n \leqslant\left(\sum_{\substack{p \leqslant X \\ p \in D_{1}}} 1 / p\right)^{m} \sim(\ln \ln x)^{m} $$ while $\Sigma_{n \leqslant x} 1 / n \sim \ln x$. (iv) First, we re-cast slightly the group used in (iii). Let $H$ be the additive group of all binary $b$-tuples of even weight, and $C$ the cyclic group of order $b$ of co-ordinate permutations. Then $C$ acts on $H$, and we take $G$ to be the semi-direct product. Since $b$ is odd, any element of $H$ has an entry 0 , and so some conjugate of it lies in the subgroup $H_{1}$ of $H$ defined by the equation $x_{1}=0$. Thus, representing $G$ on the $2 b$ cosets of $H_{1}$, every 2-element (that is, every element of $H$ ) has a fixed point. To establish (iv), we use the same group, but replace $H_{1}$ by the subgroup $H_{2}$ defined by the equations $x_{2}=0, x_{1}=x_{3}$. Every element $h$ of $H$ has an odd number of zeros, hence a run of consecutive zeros of odd length (regarding it as cyclically ordered). Thus $h$ has either three consecutive zeros, or a run . . 101 ...; in either case, some cyclic shift is in $H_{2}$. Now, if $b>3$, the equations $x_{1}+\cdots+x_{b}=0, x_{2}=0, x_{1}=x_{3}$ are independent; so $\left|H: H_{2}\right|=4,\left|G: H_{2}\right|=4 b$, establishing the result. This construction suggests an obvious generalisation. By (i), it is reasonable to concentrate on the case when $b=p$ is an odd prime. Define a function $w(p)$ as follows. Let $C$ be the cyclic group of order $p$, and $V$ the unique (up to similarity) non-trivial irreducible $C$-module over GF(2). The dimension of $V$ is the order of $2 \bmod p$, say $d$, and $V$ can be realised as the additive group of $\mathrm{GF}\left(2^{d}\right)$, with $C$ the unique subgroup of order $p$ in the multiplicative group. Now let $w(p)$ be the maximum codimension of a subspace $W$ of $V$ with the property that the images of $W$ under $C$ cover $V$. Proposition 4.1. $2^{a} \cdot p \in A$ for all $a \leqslant w(p)$. Proof. Let $W$ be as in the definition, with codimension $w(p)$. Let $U$ be a subspace containing $W$ and having codimension $a$. Now let $G$ be the semi-direct product of $V$ by $C$, acting on the cosets of $U$. Proposition 4.2. Let $p$ be a prime greater than 3 , and let $d$ be the order of $2 \bmod p$. Then $$ \max (2,\lfloor 2 \log p\rfloor-d) \leqslant w(p) \leqslant\lfloor\log p\rfloor $$ The upper bound is attained if $p=\left(2^{d}-1\right) /\left(2^{e}-1\right)$ for some $e$ dividing $d$, and so, in particular, if $p$ is a Fermat or Mersenne prime. ProOF. The lower bound 2 is established by the same argument as Theorem 2(iv); for the other lower bound see (3). The upper bound is trivial: $|V|=2^{d},|W|=2^{d-w(p)}$, and $p$ images of $W$ cover $V$. Suppose that $e$ divides $d$. Then we can regard $V$ as a $\operatorname{GF}\left(2^{e}\right)$-space of dimension $d / e$. If $p=\left(2^{d}-1\right) /\left(2^{e}-1\right)$, then $C$ permutes transitively the one-dimensional $\mathrm{GF}\left(2^{e}\right)$-spaces, so the images of one of these (say $W$ ) cover $V$; and the GF(2)-codimension of $W$ is $d-e$. This result yields a number of primes $p$ for which $w(p)>2$ (although as yet, with present knowledge, only finitely many): in addition to Fermat and Mersenne primes greater than 7, there are others given by the last assertion (for example, $w(73)=6$ ), or by the lower bound involving $d$ (for example, $w(178481) \geqslant 11)$. We have established by computation the values $w(11)=w(13)=2, w(23)=w(43)=3, w(89)=6$. It is an open problem to establish that $w(p)>2$ for infinitely many $p$; but perhaps it is even the case that $w(p) \rightarrow \infty$ as $p \rightarrow \infty$. The upper bound in Proposition 4.2 means that we can never construct a member of $A$ whose 2-part exceeds its odd part by this method. Moreover, the multiplicative closure of $A$ does not lift this limitation either. To do better with these methods, we need to ensure that the number of conjugates of the subgroup $U$ of $V$ exceeds the index of $V$ in $G$. For this, we need to use a non-abelian group $V$. This argument is due to Peter Neumann. The next result extends his observation that $56 \in A$. Proposition 4.3. (i) If $n$ is even, then $2^{a}\left(2^{n}-1\right) \in$ for all $a \leqslant 3 n / 2-2$. (ii) If $n$ is odd, then $2^{a}\left(2^{n}-1\right) \in A$ for all $a \leqslant 2 n-2$. Proof. (i) Let $q=2^{2^{n}}$. Let $G$ be the stabiliser of a point in the group $\operatorname{PGU}(3, q)$ (in its usual 2-transitive permutation representation). Then $G$ has a normal subgroup $V$ of order $q^{3}$, and $G / V$ is cyclic of order $q^{2}-1$. Moreover, the following are proved by straightforward calculation: (a) The centre of $V$ is elementary abelian of order $q$, and $V / Z(V)$ is elementary abelian of order $q^{2}$; (b) $\left|C_{G}(V)\right|=q^{2}$ for all $v \in V \backslash Z(V)$; (c) $G / V$ acts transitively on $Z(V)$ and on $V / Z(V)$ by conjugation. It follows that $V$ is the union of three $G$-conjugacy classes, namely $\{1\}, Z(V) \backslash\{1\}$, and $V \backslash Z(V)$. So, if $H$ is a subgroup of $G$ containing an element of $V \backslash Z(V)$ (all such elements have order 4, and their squares lie in $Z(V))$, then $G$, acting on the cosets of $H$, has no fixed-point-free 2-element. Moreover, $|G: H|=\left(2^{n}-1\right) 2^{3 n / 2-f}$, where $|H|=2^{f}$; here $f$ can be any number in the range $2 \leqslant f \leqslant 3 n / 2$. (ii) If $n$ is odd, let $q=2^{n}$, and let $G$ be the stabiliser of a point in $S z(q)$. Then $|G|=$ $q^{2}(q-1)$. If $H$ contains an element of order 4 then, again, $H$ meets every conjugacy class in $V=0_{2}(G)$, and the conclusion follows as in (i). Remark 4.4. Many more constructions can be devised. For example, if $G$ is a group with elementary abelian Sylow 2-subgroups and only one conjugacy class of involutions (such as $P S L\left(2,2^{n}\right)$, or the Ree group ${ }^{2} G_{2}\left(3^{2 n+1}\right)$ ), and $H$ any subgroup of $G$ of even order, then $|G: H| \in A$. But for every number which we have been able to show to lie in $A$, we can in fact demonstrate this using Theorem 2 and Propositions 4.1-4.3. ## 5. On 4-Wise Intersecting Transitive FamiLies In this section, we prove Theorem 5(i) and Theorem 6. The proofs depend on a result from [5] which we now state. Let $\bar{x}=\left(x_{1}, \ldots, x_{n}\right)$ be a random vector-variable. For $A \subseteq X=\{1, \ldots, n\}$, let $\bar{x}_{A}$ denote the restriction of $\bar{x}$ to the co-ordinates in $A$, that is, $\bar{x}_{A}=\left(x_{a}: a \in A\right)$. For example, $\bar{x}_{\{i\}}$ is simply the random variable $x_{i}$. Let $H_{A}=H\left(\bar{x}_{A}\right)$ be the binary entropy of $\bar{x}_{A}$. Proposition 5.1 [5]. Suppose that $A_{1}, \ldots, A_{s}$ are subsets of $X$ such that each element of $X$ lies in at least $d$ of them. Then we have $$ d H(\bar{x}) \leqslant \sum_{i 1}^{s} H\left(\bar{x}_{A_{i}}\right) $$ We identify a subset $F$ of $X$ with its characteristic function $\bar{x}(F)$, a zero-one vector of length $n$. Given a family $\mathscr{F}$ of subsets of $X$, we use the uniform distribution on $\mathscr{F}$; that is, $$ p(\bar{x}=\bar{x}(F))=1 /|\mathscr{F}| \text { for } F $$ Thus we have $H(\bar{x})=\Sigma_{i=1}^{m}-(1 / m) \log (1 / m)=\log m$, where $m=|\tilde{F}|$. Proof OF TheORem 6. Let $G$ be a transitive permutation group on $X, T$ a $t$-subset of $X$, and $$ \mathscr{F}=\{F \subseteq X: F \cap g(T) \neq \varnothing \quad \text { for all } g \in G\} $$ and consider the random variable $\bar{x}$ defined above. For every $g \in G, \bar{x}_{g(T)}$ takes at most $2^{t}-1$ values, since $(0, \ldots, 0)$ is excluded. Thus we have $$ H\left(\bar{x}_{g(T)}\right) \leqslant \log \left(2^{t}-1\right) $$ On the other hand, the transitivity of $G$ implies that every element of $X$ lies in exactly $|G| t / n$ sets $g(T)$. Thus, by Proposition 5.1, $$ |G| t / n \log |\mathscr{F}| \leqslant|G| \log \left(2^{t}-1\right) $$ or, equivalently, $$ |\tilde{F}| \leqslant\left(2^{t}-1\right)^{n / t} $$ In [7] it was shown that a transitive 4-wise intersecting family $\mathscr{F}$ satisfies $|\mathscr{F}| \leqslant 2^{n} / n^{1 / 2}$. Theorem 5(i), which we now prove, improves this bound considerably. We need the following easy consequence of a theorem of Katona [13]. Proposition 5.2. Let $r$ be a positive integer, and $\mathscr{F}$ a family of sets satisfying $\left|F \cap F^{\prime}\right| \geqslant r$ for all $F, F^{\prime} \in \mathscr{F}$. Then $$ |\mathscr{F}| \leqslant 2^{n} \mathrm{e}^{-r^{2} / 2 n} $$ Proof OF TheOREm 5(i). Let $\mathscr{F}$ be transitive and 4-wise intersecting, and $G=\operatorname{Aut}(\mathscr{F})$. Choose $F_{1}, F_{2} \in \mathscr{F}$ so that $\left|F_{1} \cap F_{2}\right|$ is minimal. Set $T=F_{1} \cap F_{2}$ and $t=|T|$. Since $\mathscr{F}$ is 4-wise intersecting, the family $$ \mathscr{F}(g(T))=\{F \cap g(T): F \in \mathscr{F}\} $$ is intersecting, and so has cardinality at most $2^{t-1}$ (by Lemma 2.1). Thus $$ H\left(\bar{x}_{g(T)}\right) \leqslant \log 2^{t-1}=t-1 $$ By Proposition 5.1, $$ |G| t / n \log |\mathscr{F}| \leqslant|G|(t-1) $$ so $|\mathscr{F}| \leqslant 2^{n} / 2^{n / t}$. Thus the result is true if $t \leqslant\left(2 n^{2} \ln 2\right)^{1 / 3}$. However, if $t>\left(2 n^{2} \ln 2\right)^{1 / 3}$, the conclusion is immediate from Katona's result (Proposition 5.2). ## 6. Regular Filters A family $\mathscr{F}$ is called a filter if $F \in \mathscr{F}$ and $G \supseteq F$ imply $G \in \mathscr{F}$. (This is not standard; we do not reuire $\mathscr{F}$ to be closed under intersection.) For $x \in X$, the degree $d(x)$ of $x$ in $\mathscr{F}$ is the number of members of $\mathscr{F}$ containing $x$. We require a simple lemma. Lemma 6.1. Suppose that $\mathscr{F}$ is a filter on $X$ and $A \subseteq X$ is such that $F \cap A \neq \varnothing$ for all $F \in \mathscr{F}$. Then the average degree of points of $A$ in $\mathscr{F}$ is at least $\frac{1}{2}|\mathscr{F}|\left(1+1 /\left(2^{a}-1\right)\right)$, where $a=|A|$. Proof. By [12] or more easily by Hall's theorem, the incidence matrix of $b$-sets and $(a-b)$-sets of an $a$-set $A$ (with incidence $=$ inclusion) is non-singular. The existence of a non-zero term in the expansion of the determinant shows that, if $b<\frac{1}{2} a$, there is a matching $B \mapsto B^{*}$ from $b$-sets to $(a-b)$-sets so that $B \subseteq B^{*}$. Defining $$ f(B)=|\{F \in \mathscr{F}: F \cap A=B\}| $$ the fact that $\mathscr{F}$ is a filter shows that $f(B) \leqslant f\left(B^{*}\right)$ for all sets $B$ with $|B|<\frac{1}{2} a$. The average degree of points in $A$ is $\alpha$, say, where $$ \begin{gathered} a \alpha=\sum_{X \in A} d_{\mathscr{F}}(x)=\sum_{\varnothing \neq B \subseteq A}|B| f(B) \\ \sum_{\substack{\varnothing \neq B \subseteq A \\ |B|<\frac{1}{2} a}}\left(|B| f(B)+(a-|B|) f\left(B^{*}\right)\right)+\frac{1}{2} a \sum_{B \subseteq A} f(B)+a f(A) \geqslant \frac{1}{2} a \sum_{\varnothing \neq B \subseteq A} f(B)+\frac{1}{2} a f(A) \\ =\frac{1}{2} a|\mathscr{F}|+\frac{1}{2} a f(A) \end{gathered} $$ Again, since $\mathscr{F}$ is a filter and $F \cap A \neq \varnothing, f(A) \geqslant|\mathscr{F}| /\left(2^{a}-1\right)$, from which the result follows. Remark. The result can be adjusted if the condition $$ F \in \mathscr{F} \Rightarrow|F \cap A| \geqslant t $$ holds, for fixed $t$. Proof OF TheOrem 5(ii). Let $\mathscr{F}$ be a 4-wise intersecting regular filter. Choose $F_{1}, F_{2}$, $F_{3} \in \mathscr{F}$ so that $A=F_{1} \cap F_{2} \cap F_{3}$ has minimal cardinality, say $a$. By [8], we have $|\mathscr{F}| \leqslant$ $2^{n}\left(\frac{1}{2}(\sqrt{5}-1)\right)^{a}$. Hence we are finished if $a>\frac{1}{2} \log n-\log \log n$; so suppose that $$ a \leqslant \frac{1}{2} \log n-\log \log n $$ If $d$ is the degree of $\mathscr{F}$, the lemma yields $$ d /|\mathscr{F}| \geqslant \frac{1}{2}\left(1+1 /\left(2^{a}-1\right)\right) $$ Then for the uniform random variable $\bar{x}$ of the last section, we have $$ H\left(x_{i}\right) \leqslant H\left(\frac{1}{2}\left(1+1 /\left(2^{a}-1\right)\right)\right. $$ and Proposition 5.1 (with $s=n, A_{i}=\{i\}$ ) yields $$ \log |\mathscr{F}| \leqslant n H\left(\frac{1}{2}\left(1+1 /\left(2^{a}-1\right)\right)\right) $$ or $$ |\mathscr{F}| \leqslant 2^{n} \cdot 2^{n H\left(\frac{2}{2}\left(1+1 /\left(2^{a}-1\right)\right)\right)}<2^{n} \cdot \mathrm{e}^{-n / 2^{2 a+1}}<2^{n} / n $$ Proof of Theorem 3. The first part is straightforward. By [1], if $G$ is transitive on $X$, and $Y \subseteq X$ with $|Y|<|X|^{1 / 2}$, then $Y \cap g(Y)=\varnothing$ for some $g \in G$, and $Y$ cannot be contained in any $G$-invariant intersecting family. To see that the result is close to best possible, take a projective plane with a transitive automorphism group $G$ (for example, a Desarguesian plane), and enlarge it to an intersecting family of size $2^{n-1}$ admitting $G$, using the remark following Proposition 2.3. For the second part, suppose that $\mathscr{F}$ is a regular intersecting family of cardinality $2^{n-1}$. Then $\mathscr{F}$ is a filter, by Lemma 2.1. Let $F$ be a member of $\mathscr{F}$ of cardinality $a$. By Lemma 6.1, the degree of any point is at least $\frac{1}{2} \cdot 2^{n-1}\left(1+1 /\left(2^{a}-1\right)\right)$. On the other hand, the sum of the degrees of all points is equal to the sum of the sizes of all sets in $\mathscr{F}$. This does not exceed the sum of the sizes of the $2^{n-1}$ largest subsets of $X$, which is easily seen to be $$ \frac{1}{2} n\left(2^{n-1}+\binom{n}{\left\lfloor\frac{1}{2}(n-1)\right\rfloor}\right. $$ Thus $$ \begin{gathered} 2^{n-1}\left(1+1 /\left(2^{a}-1\right)\right) \leqslant 2^{n-1}+\binom{n}{\left\lfloor\frac{1}{2}(n-1)\right\rfloor} \\ 2^{a}-1 \geqslant 2^{n-1} /\binom{n}{\left\lfloor\frac{1}{2}(n-1)\right\rfloor} \end{gathered} $$ from which the result follows. ## REFERENCES 1. B. J. Birch, R. G. Burns, S. O. Macdonald and P. M. Neumann. On the orbit-sizes of permutation groups containing elements separating finite subsets, Bull. Austral. Math. Soc., 14 (1976), 7-10. 2. A. Brace and D. E. Daykin, Sperner-type theorems for finite sets, in: Combinatorics, D. R. Woodall and D. J. A. Welsh, eds, 18-37, Inst. Maths Applics, Southend-on-Sea, 1972. 3. P. J. Cameron, Four lectures on projective geometry, in: Finite geometries, C. A. Baker and L. M. Batten, eds, 27-63, Marcel Dekker, New York, 1985. 4. P. J. Cameron, L. G. Kovacs, M. F. Newman and C. E. Praeger, Fixed-point-free permutations in transitive permutation groups of prime power order, Q. J. Math., Oxford (2), 36 (1985), 273-278. 5. F. R. K. Chung, P. Frankl, R. L. Graham and J. B. Shearer, Some intersection theorems for ordered sets and graphs, J. Combin. Theory, Ser. A, 43 (1986), 23-37. 6. B. Fein, W. M. Kantor and M. Schacher, Relative Brauer groups II, J. Reine Angew. Math., 328 (1981), 39-57. 7. P. Frankl, Regularity conditions and intersecting hypergraphs, Proc. Am. Math. Soc., 82 (1981), 309-311. 8. P. Frankl, An intersection theorem for finite sets, Bull. Austral. Math. Soc., 15 (1976), 73-79. 9. J. R. Isbell, Homogeneous games, I, Math. Student, 25 (1957), 123-128. 10. J. R. Isbell, Homogeneous games, II, Proc. Am. Math. Soc., 11 (1960), 159-161. 11. J. R. Isbell, Homogeneous games, III, in: Advances in game theory, M. Dresher, L. S. Shapley and A. W. Tucker, eds 255-265, Princeton Univ. Press, Princeton, N.J., 1984. 12. W. M. Kantor, On incidence matrices of finite projective and affine space, Math. Z., 124 (1972), 315-318. 13. Gy. Katona, An intersection theorem for systems of finite sets, Acta Math. Hungar., 15 (1964), 329-337. 14. C. C. Sims, Computational methods in the study of permutation groups, in: Computational problems in abstract algebra, J. Leech ed., 169-183, Pergamon, Oxford, 1970. Received 5 March 1987 and in revised form 25 July 1988 P. J. CAMERon School of Mathematical Sciences, Queen Mary College, London E1 4NS, U.K. P. FrankL C.N.R.S., Université Paris VII, 75005 Paris, France and W. M. KAntOR Department of Mathematics, University of Oregon, Eugene, Oregon 97403, U.S.A. # Transitive Avoidance Games J. Robert Johnson*† Imre Leader ${ }^{\ddagger \S} \quad$ Mark Walters ${ }^{* \uparrow}$ July 12, 2016 #### Abstract Positional games are a well-studied class of combinatorial game. In their usual form, two players take turns to play moves in a set ('the board'), and certain subsets are designated as 'winning': the first person to occupy such a set wins the game. For these games, it is well known that (with correct play) the game cannot be a second-player win. In the avoidance (or misère) form, the first person to occupy such a set loses the game. Here it would be natural to expect that the game cannot be a first-player win, at least if the game is transitive, meaning that all points of the board look the same. Our main result is that, contrary to this expectation, there are transitive games that are first-player wins, for all board sizes which are not prime or a power of 2. Further, we show that such games can have additional properties such as stronger transitivity conditions, fast winning times, and 'small' winning sets. ## 1 Introduction Many natural combinatorial games may be viewed as 'achievement games' as follows. We have a finite set (the board) and some of its subsets are designated as special and called lines. Two players take it in turn to claim a (previously unclaimed) point of the board; the first player to complete a line[^6]is declared the winner. If all points have been claimed but neither player has completed a line then the game is considered a draw. In this paper we consider the avoidance (or misère) variant - the game is played exactly as above except the first player to complete a line loses. Simmons [13] described the first example of such a game that we are aware of (the game of Sim), and Harary [7] introduced the more general mathematical framework. Since then such games have been considered by many authors - see, e.g., Beck [3, 2] and Slany [14]. For more general misère games see, e.g., Conway [4], and Albert and Nowakowski [1]. There has also been a substantial amount of work on Avoider/Enforcer games, which are the 'Maker/Breaker' analogue for avoidance games - see, e.g., Lu [12], Beck [3], and Hefetz, Krivelevich and Szabó [9]. For achievement games, a simple strategy stealing argument shows that the game is either a draw or a first player win (with perfect play). However, for avoidance games the situation is not clear. Indeed Beck [2] states "The general open problem is to find the avoidance version of the strategy stealing argument." At first glance it looks as though the first player has a disadvantage since he has 'more' points than the second player. Of course in general this is not true: indeed take any game that is a second player win and add a single point not in any line. The first player picks this new point on his first turn and thus reduces the new game to the old game with Player II (the second player) playing first. However, this is a rather trivial example: we have artificially given Player I an advantage by giving him a special 'safe' point he can pick. Thus, it is natural to insist that the game is transitive: that is, that the automorphism group of the game acts transitively on the board's points. (The automorphism group is the group of all permutations of the board $X$ that preserve the family $\mathcal{L}$ of lines.) Informally, 'all points look the same.' One might guess that a transitive avoidance game must be a second player win: what advantage can there be to going first when all points are the same? Indeed, this intuition is correct when all the lines have size 2. In general, however, it turns out that there may be advantages to playing first. A priori, there are two reasons why it seems to be harder to find a game that is a Player I win when $n$ (the board size) is odd: first, Player I has an extra move and, secondly, Player II has the final choice of move (as Player I's last move is forced). However, whilst we shall give examples of first player wins for both $n$ even and $n$ odd, it turns out that the simplest examples of such a game occur in the case $n$ odd (see Section 3 ). Having found examples of Player I wins we turn to the central question we address in this paper: namely for which $n$ does there exist a transitive avoidance game that is a Player I win. Our main result shows that, for most $n$, such games do exist. Theorem 1. Suppose that $n$ is neither a power of 2 nor a prime. Then there is a transitive avoidance game on $n$ points that is a Player I win. The case when $n$ is divisible by a large power of 2 is much harder than the case when $n$ is odd, or $n$ is equal to $2 \bmod 4$. It relies on a careful study of subsets of $\mathbb{Z}_{2^{m}}$ (the integers modulo $2^{m}$ ) under rotation. This may be of independent interest. There are two cases not covered by Theorem 1: powers of 2, and primes. For powers of 2 we have the following result. Theorem 2. Suppose that $n$ is a power of 2. Then no transitive avoidance game on $n$ points is a Player I win. It turns out that this is a fairly simple consequence of a well known group theoretic result. For the remaining case of $n$ prime there are examples for which there are Player I wins (e.g., 11 and 13), but we do not know what happens for any prime larger than 13. In many of our constructions Player II does not lose until his last move. It is easy to modify these examples so that Player I can force Player II to lose before time $(1-\varepsilon) n$. However, we also give examples of games where Player II loses in time $o(n)$. Interestingly these even include examples where the losing lines are all of bounded size (size 3 in fact). We also consider avoidance games where on each turn each player is allowed to pick more than one point (this is usually called the plus version of the game; see Slany [14]). One might think that picking more than one point would never help a player, but it turns out that in any transitive plus game Player II can use this extra freedom to guarantee that he does not lose. A related phenomenom was proved by Hefetz, Krivelevich, Stojaković and Szabó [8] for a biased version of the Avoider/Enforcer game where moving to the plus version of the game simplified the behaviour substantially. Our argument applies to some previously studied games. For example, it gives a simple strategy stealing proof of the previously unknown fact that the plus version of the Ramsey Avoidance Game (see Section 2) is never a Player I win. In Section 2 we give two situations where strategy stealing arguments do work: when there is a line of size 2 , and for the plus version of the games. Then in Section 3 we give a simple construction of Player I wins for all composite odd $n$. Section 4 is the heart of the paper. In it we show that there are Player I wins for all even sizes, except powers of 2. In Section 5 we discuss some natural variants of the game such as lines with bounded size, games which are 'more' transitive, and games where Player I can win quickly. We show that our constructions so far (or simple modifications of them) are able to provide first player wins in these cases too. Then, in Section 6 we give a natural game which satisfies all of these stronger properties simultaneously - it is a fast Player I win, the lines have bounded size, and it is more transitive (more precisely it is edge/line transitive - see the section for the formal definition). We conclude with a discussion of some open problems. We will use the following notation for games throughout the paper. We will denote the board by $X$, the board size $|X|$ by $n$, and the set of losing lines by $\mathcal{L}$. We abbreviate Player I and Player II to PI and PII respectively. ## 2 Games for which strategy stealing works ### 2.1 A line of size 2 If the family of losing lines $\mathcal{L}$ has any set of size 2 then there is a strategy stealing argument. Note that this includes the case when all lines have size 2: i.e., where $\mathcal{L}$ corresponds to the edge set of some vertex transitive graph. Theorem 3. A transitive avoidance game with any line of size 2 is not a PI win. Proof. Suppose (for contradiction) that the game is a PI win. Suppose that PI's first move is $x$. Since the game is transitive, every point is in a line of size 2 so we can choose $y$ such that $\{x, y\} \in \mathcal{L}$. Using the transitivity of the game again we see that there is a winning strategy, $\Phi$ say, for PI with first move $y$. PII now plays strategy $\Phi$ ignoring the fact that PI has already claimed $x$. This could only go wrong if $\Phi$ tells PII to play $x$. However, since $\{x, y\}$ is a line, and thus PII would lose anyway if he played $x$, this cannot happen. Note that we made crucial use of the existence of a line of size 2 in the 'PII would lose anyway' statement at the end of the proof. We remark that a similar argument shows that in any transitive avoidance game any strategy of PI that promises never to play a particular point is not winning. Thus, winning strategies for PI must be global; i.e., they must examine the entire board. ### 2.2 The plus version As described in the introduction, the plus variant of an avoidance game is the same as the avoidance game except that, on each move, a player may choose to pick as many points as he likes rather than just one; in other words, on each move, each player chooses a (non-empty) set of points. Theorem 4. The plus version of a transitive avoidance game is not a PI win. Proof. Since the players can pick an arbitrary number of points we cannot tell which player's move it is just by looking at the current position, and we will make use of this fact in our proof. Thus, we introduce some notation for the position. A position is a pair of disjoint sets $(S, T)$ where $S$ denotes the set of all points picked so far by the player whose turn it is, and $T$ the set of all points picked by the other player. From this position we will name the player whose move it is the current player. Let $G$ be the automorphism group of the game. Suppose for the sake of a contradiction that PI has a winning strategy and let $S$ be PI's first (set) move in this strategy. After this move the position is $(\emptyset, S)$. Since PI is playing a winning strategy we know that position $(S, T)$ is a current player win for any non-empty set $T$. Thus, for any $g \in G$ and any non-empty set $T$ disjoint from $g(S)$, the position $(g(S), T)$ is also a current player win. To obtain a contradiction we now show that the game is not in fact is a PI win. Suppose that PI plays some set $U$. If $U$ is the whole board then the game is over and PII had definitely not lost. Thus, we may assume that PI does not pick all the points. We have to give a strategy for PII. Pick $x \notin U$ and $g \in G$ which maps some $s \in S$ to $x$. Let $S^{\prime}=g(S)$. PII plays $S^{\prime} \backslash U$, which contains $x$ so is non-empty. Now, suppose that PI plays $W$. Then, it is PII's turn and the position is $\left(S^{\prime} \backslash U, U \cup W\right)$. Since $S^{\prime} \backslash U \subset S^{\prime}$ and $$ \left(S^{\prime} \backslash U\right) \cup(U \cup W)=S^{\prime} \cup U \cup W=S^{\prime} \cup\left(\left(U \backslash S^{\prime}\right) \cup W\right) $$ we see that $\left(S^{\prime} \backslash U, U \cup W\right)$ is no worse for the current player than $\left(S^{\prime},(U \backslash\right.$ $\left.\left.S^{\prime}\right) \cup W\right)$. As noted above, this latter position is a current player win so $\left(S^{\prime} \backslash U, U \cup W\right)$ is also a current player win; i.e., PII wins as claimed. Some plus version avoidance games have been studied previously: in particular, the Ramsey Avoidance Game $R A G(N, s)$ (see, e.g., [14, 2]). This game is played with board the edge set $E=E\left(K_{N}\right)$ and lines are the $\binom{s}{2}$ subsets of $E$ corresponding to $K_{s}$ subgraphs for some fixed $s$. Corollary 5. The plus version of Ramsey Avoidance Game $\operatorname{RAG}(N, s)$ is not a PI win. In particular, if $N \geq R(s, s)$ (where $R(s, s)$ denotes the Ramsey number) then it is a PII win. Proof. The game is obviously transitive so Theorem 4 implies that it is not a PI win. In the case $N \geq R(s, s)$ the game cannot be a draw - when the board is full (at least) one player must have a $K_{s}$ - and so must be a PII win. ## 3 First player wins for odd board sizes In this section we show that there are first player wins for all odd composite board sizes. This is in contrast to the special cases discussed in the previous section. Theorem 6. Suppose that $n=p q$ is odd. Then there is a transitive avoidance game on $[n]$ that is a PI win. Proof. First we define the game. View the board, $[n]$, as $q$ sets $A_{1}, A_{2}, \ldots, A_{q}$ of $p$ points. Let $p^{\prime}=(p+1) / 2$ and $q^{\prime}=(q+1) / 2$. We call each $A_{i}$ a bucket. Let $\mathcal{W}$ be all subsets of $[n]$ of size $p^{\prime} q^{\prime}$ consisting of $p^{\prime}$ from each of $q^{\prime}$ buckets. Define the lines $\mathcal{L}=[n]^{\left(p^{\prime} q^{\prime}\right)} \backslash \mathcal{W}$. We claim that the avoidance game on $(n, \mathcal{L})$ is transitive and a PI win. The transitivity is trivial since we can permute the bins, and permute the points in any bin. Thus we just need to prove that it is a PI win. Since, every pair of sets in $\mathcal{W}$ meet, we see that if PI can form a set in $\mathcal{W}$ with his first $p^{\prime} q^{\prime}$ points then PII's first $p^{\prime} q^{\prime}$ points must form a set not in $\mathcal{W}$, i.e., these points must form a line in $\mathcal{L}$. Thus, it suffices to show that PI can guarantee to form a set in $\mathcal{W}$ with his first $p^{\prime} q^{\prime}$ points. We call a bucket active if PI has played at least one point in it but not yet $p^{\prime}$ points in it. We say it is full if he has played $p^{\prime}$ points in it. PI's strategy is to play according to first of the following rules that applies. 1. If PII has just played in an active bucket then PI plays in the same bucket. 2. If less than $q^{\prime}$ buckets are either full or active then PI plays in an empty bucket. 3. PI plays in any active bucket. Rule 1 implies that, after his turn, PI always has strictly more points than PII in any active bucket. Rule 2 implies that, after PI's turn, strictly more than half the non-empty buckets are active or full. Thus after $p^{\prime} q^{\prime}$ moves PI has exactly $p^{\prime}$ points in exactly $q^{\prime}$ buckets. ## 4 First player wins for even board sizes ### 4.1 Isbell Families We saw in the introduction that one advantage for the first player is that he goes first and can pick a 'special' point, but obviously this is not possible for a transitive game. A second possible advantage for the first player occurs if the board has an even number of points: the second player has no choice on his last turn, whereas the first player does always have a choice. We use this 'forced move' to construct examples of games with even size boards that are first player wins. It turns out that our avoidance game is closely related to an existing game idea, namely that of a fair game, from a very different context (see Isbell [10] and [11]). To avoid confusion with other notions of fair game we will call the families involved Isbell families. Definition. An Isbell family on a set $[n]$ is a family $\mathcal{F}$ of subsets of $[n]$ such that $\mathcal{F}$ is an up-set containing exactly one of $X$ and $[n] \backslash X$ for each set $X$, and having a transitive automorphism group. We remark that an Isbell family must be intersecting, that is any two sets in the family meet. Proposition 7. Suppose that $n$ is even and that there exists an Isbell family on $[n]$. Then there is a transitive avoidance game with board $[n]$ that is a first player win. Proof. Let $\mathcal{F}$ be the sets in the Isbell family. We define the lines $\mathcal{L}=$ $\mathcal{F} \cap[n]^{(n / 2)}$. We show that this avoidance game is a PI win. Obviously, this game is transitive. Consider the achievement game on board $[n]$ with winning lines $\mathcal{W}=$ $[n]^{(n / 2)} \backslash \mathcal{L}$. By the definition of an Isbell family every $n / 2$ sized set is either in $\mathcal{W}$ or its complement is in $\mathcal{W}$. Hence, a draw is impossible in this achievement game. Thus, by the standard strategy stealing result, this achievement game must be a PI win. PI follows exactly the same strategy in the avoidance game. At the end his $n / 2$ points form a set in $\mathcal{W}$ so not in $\mathcal{L}$ (so he has not lost) and PII's $n / 2$ points are the complement of PI's set, so must form a set in $\mathcal{L}$ and PII has lost. It is not known for exactly which $n$ Isbell families exist but most cases are known. In particular, Isbell [10] showed that they do exist for $n=2 b$ with $b>1$ odd; and Cameron, Frankl and Kantor [5] showed they do exist for $n=4 b$ with $b>3$ and odd. Thus, for all these cases we have transitive avoidance games which are first player wins. However, Cameron, Frankl and Kantor [5] also showed that Isbell families do not exist for $n=2^{a}$, or for $n=3 \times 2^{a}$ for $a \geq 2$, so we cannot use them to prove all the remaining cases of Theorem 1. For concreteness let us describe the Isbell family, and thus the PI win avoidance game, on 6 points. We think of the six points as being arranged in a grid of two rows and three columns. The Isbell family $\mathcal{F}$ is the up-set generated by the family of all 3 -sets that either contain one point from each pair and an even number of points in the top row, or contain both points in one pair and one point in the next pair cyclicly. It is easy to check that this family is transitive and that every set is either in $\mathcal{F}$ or its complement is in $\mathcal{F}$. Thus, $\mathcal{F}$ is indeed an Isbell family, and the avoidance game with lines $[6]^{(3)} \cap \mathcal{F}$ (i.e., the lines are exactly the generating sets described above) is a PI win. Indeed, this is easy to verify by hand. In contrast, there do not exist first player wins when $n$ is a power of 2 . In order to prove this we start with a simple lemma. Lemma 8. Let $(n, \mathcal{L})$ be a transitive avoidance game and suppose that its automorphism group contains a fixed-point-free involution. Then the game is not a PI win. Proof. Let $G$ be the game's automorphism group and let $g \in G$ be a fixedpoint-free involution. Then $g$ partitions the board into pairs. PII's strategy is to play the point paired with PI's previous move; i.e., if PI plays $x$, then PII plays $g(x)$. Suppose that, after PII's move, PII has played all the points in $T$. Then PI must have played all the points in $g(T)$. Thus, if PII's move was losing - that is, if $T$ contains a losing set - then PI must have already lost. The following theorem is an immediate consequence. Theorem 9. A transitive avoidance game on a set of size $2^{a}$ for some $a$ is not a PI win. Proof. Let $G$ be the automorphism group of the game. Then $G$ acts transitively on the board $X$ which has size $2^{a}$. By a standard result (see e.g., [5]) from group theory, it follows that $g$ has a fixed-point-free involution. Thus, by Lemma 8 the game is not a PI win. We would like to extend Proposition 7 to show that there is a transitive avoidance game that is a PI win for all even sizes except powers of 2 . In the proof of Proposition 7 we relied on the existence of Isbell families. These were useful because they gave a transitive intersecting family consisting of half of the $\frac{n}{2}$-subsets. Since there do not exist Isbell families for sizes $3 \times 2^{a}$, we cannot use the same technique. Indeed, since the union of any such intersecting family and the family of all subsets of size strictly greater than $n / 2$ is an Isbell family, such intersecting families only exist when Isbell families exist. Instead, we look for a smaller intersecting family $\mathcal{F}$ for which PI can win the achievement game. Since $\mathcal{F}$ is not a maximal intersecting family it is possible that neither player forms a set from $\mathcal{F}$ (i.e., a draw is possible in the achievement game), so we cannot use a strategy stealing argument to show that PI has a winning strategy for the achievement game. Thus, we need to define a winning strategy explicitly. ### 4.2 Special case: $n=2 b$ with $b$ odd As our proof in the general case is rather involved, we start by illustrating it in the simple case of $n=2 b$ with $b$ odd; this case is covered by Isbell's results on the existence of Isbell Families, but we will give an example with a much smaller intersecting family for all $n \geq 10$. (For $n=6$ the family is exactly the Isbell family on six points described earlier.) We remark that we give a proof that generalises to the full case, rather than the simplest proof for this special case. Proposition 10. For any odd $b \geq 3$ there is a transitive avoidance game of size $2 b$. Proof. Let $n=2 b$ and $b^{\prime}=(b-1) / 2$. We think of $[n]$ as $\mathbb{Z}_{b} \times \mathbb{Z}_{2}$ : i.e., as $b$ pairs. We start by defining some winning sets $\mathcal{W}$. They are all of size $b$, and are 1. all sets which contain exactly one point from each pair and an odd number of which have a 1 in the $\mathbb{Z}_{2}$-coordinate, 2. all sets with both elements of exactly one pair, that have the (unique) pair with neither element at most $b^{\prime}$ later cyclicly. Obviously the complement of any set in this family is not in the family and thus we see that this family is intersecting. We define the board of our avoidance game to be $[n]$ and the lines $\mathcal{L}$ to be the complements of the sets in $\mathcal{W}$. We claim that this avoidance game is transitive and a PI win. To see that the game is transitive, it is enough to observe that its automorphism group contains the following elements - cycle the $b$ pairs, - in two pairs swap the elements (i.e., swap $(x, 0)$ and $(x, 1)$, and swap $(y, 0)$ and $(y, 1)$ for some $x$ and $y$ ). Thus, to complete the proof, we just need to show that the game is a PI win. Indeed, if we can show that PI can guarantee to make a set in $\mathcal{W}$, then we know that PII will finish with a set in $\mathcal{L}$, and so will lose. We will denote positions in the game as ordered pairs $(A, B)$ of subsets of the board where this means PI has played the points in $A$ and PII has played the points in $B$. For a set $A$ containing at most one of each pair, we write $\bar{A}$ for the 'opposite' points: that is the points in the other half of each pair that meets $A$. First, define a position to be a direct win if it is of the form $(A \cup$ $\{(x, y)\}, \bar{A} \cup\{(z, w)\})$ for some $A$ and $1 \leq z-x \leq b^{\prime}$, and observe that PI has a simple winning strategy from any such position. Indeed, he plays $(x, y+1)$ and for the rest of the game makes sure he gets one of every other pair set except he never plays the point $(z, w+1)$ opposite to $(z, w)$. This means that after PI's last turn he has one of every pair, except he has both of the $x^{\text {th }}$ pair and neither of the $z^{\text {th }}$ pair. The constraint on $z$ means that this is a winning set under Condition 2 above. Our strategy for PI is as follows. Unless the position is a direct win, he will ensure that the position after his turn is of the form $(A \cup\{(x, y)\}, \bar{A})$ for some set $A$ and point $(x, y)$. Now from the position $(A \cup\{(x, y)\}, \bar{A})$ suppose that PII plays a point $(z, w)$. We split into several cases. 1. If $1 \leq z-x \leq b^{\prime}$ then the position is a direct win and he wins by following the simple strategy given above. 2. If $(z, w)$ is any point other than $(x, y+1)$ (i.e., unless PII plays the opposite point to $(x, y))$ then PI plays the point $(z, w+1)$ (i.e., opposite where PII played). Obviously this is possible and keeps the position being of the form above, so PI follows the same strategy from the new position. 3. The last case is if PII plays the point $(x, y+1)$, i.e., the opposite point to $(x, y)$. Thus, before PI plays, the position is $(B, \bar{B})$ where $B=A \cup\{(x, y)\}$ : i.e., every pair is either full or empty. Let $x^{\prime}$ be the first empty pair. If $x^{\prime}1$, an $r$-maximal point of a cyclic pair set $A$ is a point $x$ where the intersection of $A$ with the interval $\{x, x+1, x+2, \ldots, x+r-1\}$ is maximal in lexicographic order over all the sets formed by intersecting $A$ with an interval of length $r$ of $[m]$. We say $x$ is maximal if it is an $m$-maximal point. We define $r$-minimal and minimal similarly. Lemma 11. Any partial cyclic pair set in $[m]$ has a unique maximal point. Proof. Since the set contains exactly one of some opposite pair the set is not fixed by cycling by $m / 2$, the order of the stabiliser is not divisible by 2. Since the set has size $m=2^{a}$ this means the stabiliser has order 1 ; in particular, no two cyclic shifts of $A$ are the same so there must be a unique maximal point. In the case when $a=1$ discussed above, i.e., for board sizes $2 b$, it was important that when PI came to decide what happened in the final empty pair he already knew what would happen in all the remaining pairs. In that case that was trivial: they were already completely filled. In the general case, when PI decides what happens in the final empty bin the remaining bins are non-empty but this does not mean they are full. However, the following key lemma shows that PI has some control over what happens in these later bins. Lemma 12. Suppose that $m$ and $m^{\prime}$ are as above, and that $A$ is a partial pair set in $\mathbb{Z}_{m}$. Then there exist values $s, t$ and $z_{1}, z_{2}$ such that - if all free points in $\left[z_{1}, z_{1}+m^{\prime}\right)$ are placed in $A$ then regardless of which of the remaining free points are placed in $A$ the maximum lies in $[t-s, t]$ - if all free points in $\left[z_{2}, z_{2}+m^{\prime}\right)$ are placed in $A$ then regardless of which of the remaining free points are placed in $A$ the maximum lies in $\left[t, t+2 m^{\prime}-s\right)$. We postpone the proof of this technical lemma to later. First, we show that we can deduce the existence of transitive avoidance games of all even sizes (except powers of 2) that are PI wins from it. The deduction is similar to the proof of Proposition 10. Theorem 13. Suppose that $n=2^{a} b$ for some odd $b>1$. Then there is a transitive avoidance game on $n$ points which is a Player 1 win. Proof. Let $m=2^{a}, m^{\prime}=m / 4$ and $b^{\prime}=(b-1) / 2$. First, we define the winning sets $\mathcal{W}$. These will all have size $n / 2$ and the lines in our avoidance game will be the family of their complements. We view $n$ as being $\mathbb{Z}_{b} \times Z_{m}$. For any point $(x, y)$ we define its opposite point to be the opposite point in the same bin: i.e., $\left(x, y+2 m^{\prime}\right)$. The winning sets $\mathcal{W}$ are all the sets 1. that contain exactly one of each opposite pair and the sum over all bins of the maximal points $\bmod m$ lies in the interval $[0, m / 2)$. 2. that contain both of exactly one opposite pair, say the pair $\left\{y, y+2 m^{\prime}\right\}$ in bin $j$, and the unique empty opposite pair is either (a) in one of the bins between $j+1$ and $j+b^{\prime}$, (b) in bin $j$ and of the form $\left\{z, z+2 m^{\prime}\right\}$ for some $y+1 \leq z \leq y+m^{\prime}-1$ (i.e., the empty pair is between 1 and $m^{\prime}-1$ after the full pair in the same bin) This is an intersecting family; indeed, since all the sets have size $n / 2$, we just need to check that the complement of any set in the family is not in the family. In the first case, this follows since the maximal point of the complement of a set containing exactly one of each pair is the maximal point of the set plus $m / 2$, so the sum of the maximal points changes by $\mathrm{bm} / 2 \equiv m / 2$ modulo $m$. In the second case it is trivial. Also this family is transitive: indeed, the automorphism group contains the elements - cycle the $b$ bins - rotate each bin $i$ by an amount $r_{i}$ with $\sum_{i} r_{i}=0$. As in Proposition 10 we will denote a position in the game by an ordered pair $(A, B)$ of subsets of the board, where $A$ denotes the points played by PI, and $B$ the points played by PII. For any set $A$ containing at most one of each opposite pair we write $\bar{A}$ for the set of points opposite to $A$. Also as in Proposition 10, some positions have simple direct wins. There are two types. A position of the form $(A \cup\{(x, y)\}, \bar{A} \cup\{(z, w)\})$ is a direct win of type-1 if $1 \leq z-x \leq b^{\prime}$. It is a direct win of type-2 if $z=x$ and $0r$, bins $[0, j)$ are full but bin $j$ is not. In such a position, the maximal points $u_{i}$ for bins $[0, j)$ have all been determined and, moreover, all the remaining bins are non-empty. Thus, for $j \leq ir$, this stays in the region $\left[0,2 m^{\prime}\right)$ as the bins fill up. We show that PI can achieve this by induction. First, we show that he can play in bin $r$ to ensure that $G(r+1) \in\left[0,2 m^{\prime}\right)$. This is trivial: PI picks a point $u$ and, as above, guarantees to pick all the points in $\left[u, u+m^{\prime}\right)$. It is easy to see that, however PII plays, the maximum lies in $\left[u-m^{\prime}, u\right]$. Thus, regardless of the game so far, by choosing $u$ correctly, PI can ensure that $G(r+1) \in\left[0,2 m^{\prime}\right)$. Now suppose that $j>r$ and that $x=G(j)$. Inductively, we know that $x \in\left[0,2 m^{\prime}\right)$, so one of $\left[x-s_{j}, x\right]$ and $\left[x, x+2 m^{\prime}-s_{j}\right)$ is a subset of $\left[0,2 m^{\prime}\right)$. Therefore, by the above observation, PI can play all the free points in $\left[z_{1}, z_{1}+m^{\prime}\right)$ or $\left[z_{2}, z_{2}+m^{\prime}\right)$ respectively, where $z_{1}, z_{2}$ are as in Lemma 12 . Then, however PII plays, $u_{j} \in\left[t_{j}-s_{j}, t_{j}\right]$ or $u_{j} \in\left[t_{j}, t_{j}+2 m^{\prime}-s_{j}\right)$ and $$ G(j+1)=G(j)+u_{j}-t_{j} \in\left[0,2 m^{\prime}\right) $$ Thus, $G(j+1) \in\left[0,2 m^{\prime}\right)$ and the induction is complete. When the process finishes the 'guess' $G(b)$ equals the actual sum of the maximum points and so this sum is in $\left[0,2 m^{\prime}\right)$ and $\mathrm{PI}$ wins. ### 4.4 Proof of Lemma 12 In this section fix $m=2^{a}$ and let $m^{\prime}=m / 4$. We start with some some notation. Suppose that $A$ is a partial cyclic pair set in $\mathbb{Z}_{m}$. Then $\left.A\right|_{[x, x+r)}$ denotes the restriction of $A$ to the interval $[x, x+r)$, rotated to be a subset of $[0, r)$. We write $A+\left.\mathbb{1}\right|_{x, x+r}$ for the union of $A$ and the set of all free points in $[x, x+r)$. Note that this is not just $A \cup[x, x+r)$ as there may be points of $[x, x+r)$ that are opposite to points in $A$ and these are not added; in particular, $A+\left.\mathbb{1}\right|_{x, x+r}$ is a partial cyclic pair set. Finally we write $A_{\max }$ for the set $A$ with all the free points added. Note that $A_{\max }$ is not a pair set (unless $A$ was already a full pair set). Next we need some simple lemmas about maximal points. We start with a trivial observation that we use repeatedly. Lemma 14. Suppose that $A$ is a full pair set and that $x$ is $r$-maximal. Then $x+2 m^{\prime}$ is r-minimal. Proof. Immediate from the definition of a cyclic pair set. Lemma 15. Let $A$ be any subset of $[m]$. Then any $r$-maximal point is $r^{\prime}$-maximal for all $r^{\prime} \leq r$. Proof. This is trivial from properties of the lexicographic order. Lemma 16. Let $A$ be any subset of $[m]$ and suppose that $x$ is an $r$-maximal point and $x+r$ is an $r^{\prime}$-maximal point. Then $x$ is an $\left(r+r^{\prime}\right)$-maximal point. Proof. This is trivial from properties of the lexicographic order. Corollary 17. Suppose that $A$ is a partial cyclic pair set and $x$ is a maximal point. Then the point $x-r$ is not an $r$-maximal point. Proof. By the previous lemma $x-r$ would be an $(r+m)$-maximal point so an $m$-maximal point (i.e., a maximal point) contradicting the uniqueness of the maximum point (Lemma 11). The same holds for $A_{\max }$ even though it is not a cyclic pair set. Corollary 18. Suppose that $A$ is a partial cyclic pair set and $x$ is a maximal point of $A_{\max }$. Then the point $x-r$ is not an $r$-maximal point of $A_{\max }$. Proof. Again, by Lemma 16, $x-r$ would be an $(r+m)$-maximal point of $A_{\max }$ so an $m$-maximal point of $A_{\max }$. However, by a similar argument to Lemma 11, $A_{\max }$ has a unique maximal point (the proof of that lemma just uses the fact that the set has exactly one point of some pair). Lemma 19. Suppose that $A$ is a (full) pair set, that 0 is maximal, that $x$ is $r$-minimal for some $x$ and $r$, and that $y$ is any $x$-maximal point. Then $y+x$ is r-minimal. Proof. Since 0 and $y$ are both $x$-maximal we have that $\left.A\right|_{[0, x)}=\left.A\right|_{[y, y+x)}$ and, since 0 is maximal, we have $\left.A\right|_{[0, x+r)} \geq\left. A\right|_{[y, y+x+r)}$. Combining these we get $\left.A\right|_{[x, x+r)} \geq\left. A\right|_{[y+x, y+x+r)}$. But since $x$ is $r$-minimal, we have $\left.A\right|_{[x, x+r)} \leq$ $\left.A\right|_{[y+x, y+x+r)}$ and hence $\left.A\right|_{[x, x+r)}=\left.A\right|_{[y+x, y+x+r)}$. Thus, $y+x$ is $r$-minimal. Lemma 20. Suppose that $A$ is a (full) pair set and that 0 is maximal. Then, for all $r<2 m^{\prime}$, no point in $[0, r]$ is $r$-minimal. In particular, this holds for $r=m^{\prime}$. Proof. Suppose that $x \in[0, r]$ is $r$-minimal. Since $r \geq x$ Lemma 19 implies that, for any $y$ that is $x$-maximal, $y+x$ is $x$-minimal, and this is the form we shall use. Now $x$ is $r$-minimal, so $x+2 m^{\prime}$ is $r$-maximal and thus, since $x \leq r$, also $x$-maximal. By Lemma 19, $2 x+2 m^{\prime}$ is $x$-minimal, and hence $2 x$ is $x$ maximal. By Lemma 19 again, we have that $3 x$ is $x$-minimal, so $3 x+2 m^{\prime}$ is $x$-maximal. Repeating we see that $4 x+2 m^{\prime}$ is $x$-minimal so $4 x$ is $x$-maximal etc. In particular $k x$ is $x$-maximal for even $k$ and $x$-minimal for odd $k$. Thus the pair set $A$ is periodic with period $2 x \leq 2 r2 m^{\prime}$ then, by Lemma $23(\mathrm{~d})$, any set extending $A^{3}$ has maximum point in $\left[x_{2}, x_{3}\right]$ and any set extending $A^{4}$ has maximum point in $\left[x_{3}, 0\right]$ (recall $x_{4}=0$ ). Thus, in this case we are done with $t=x_{3}$ and $s=x_{3}-x_{2}$. Thus we may assume $x_{2} \leq 2 m^{\prime}$ and thus, by Lemma 23(c), that that there are no free points in $\left[x_{2}, 2 m^{\prime}\right)$. We apply Lemma $23(\mathrm{~b})$ again, this time with $x=x_{2}$ and $y=2 m^{\prime}$. This shows that $x_{1}$ is $m^{\prime}$-maximal in $A_{\max }$. If $x_{3}-x_{1}<2 m^{\prime}$ then, as above, we are done: any set extending $A^{2}$ has maximum point in $\left[x_{1}, x_{2}\right]$ and any set extending $A^{3}$ has maximum point in $\left[x_{2}, x_{3}\right]$. Thus we may assume $x_{3}-x_{1} \geq 2 m^{\prime}$ and, in particular, that $x_{1} \leq m^{\prime}$. Hence, Lemma 23(c) implies that $A$ has no free points in $\left[x_{1}, m^{\prime}\right)$. To summarise, we have that, for each $k, x_{k}$ is $m^{\prime}$-maximal in $A_{\max }$, $x_{k} \leq k m^{\prime}$, and $A$ does not have any free points in the interval $\left[x_{k}, k m^{\prime}\right)$. If, for any $k$ we have $x_{k+2}-x_{k}<2 m^{\prime}$ then as above we are done: any set extending $A^{k+1}$ has maximum point in $\left[x_{k}, x_{k+1}\right]$ and any set extending $A^{k+2}$ has maximum point in $\left[x_{k+1}, x_{k+2}\right]$. Hence the only remaining case is that both $x_{2}=2 m^{\prime}$ (i.e $x_{2}=x_{0}+2 m^{\prime}$ ) and $x_{3}=x_{1}+2 m^{\prime}$. Then both 0 and $2 m^{\prime}$ are $m^{\prime}$-maximal in $A_{\max }$ and, in particular, $\left.A\right|_{\left[0, m^{\prime}\right)}=\left.A\right|_{\left[2 m^{\prime}, 3 m^{\prime}\right)}$. This means that there can be no fixed point (i.e., non-free point) in these intervals as such a point would be fixed differently in the two sets. Hence all points in these intervals are free. But, if this is the case, then 0 is the maximal point in any set $B$ extending $A^{0}$. Indeed, since 0 is maximal in $A_{\max }$ we must have that the preceding element (element ' $-1^{\prime}$ ) is fixed not in $A$. Thus, in the interval $\left[2 m^{\prime}, 4 m^{\prime}\right)$, the set $B$ consists of $m^{\prime}$ zeros, $m^{\prime}-1$ points we don't know about followed by another zero. Thus it is trivial to see that, whatever $B$ is, no point in $\left[2 m^{\prime}, 4 m^{\prime}\right)$ is $m^{\prime}$-maximal (it would have to have $m^{\prime}$ ones following it) and the result follows. ## 5 Some Simple Observations Before proceeding further we collect some simple observations. Note that, although some of the proofs are a little long to write out, all the results in this section are essentially trivial consequences of our work so far. ### 5.1 Bounded size lines In Theorem 3 we showed that all transitive games with a line of size 2 are a PII win. However, this is best possible since our first example of an even sized PI win was a game on 6 points with all lines of size 3. It is easy to extend this to find arbitrarily large examples of PI win games with all lines of size 3 . Theorem 24. For all $n=12 k+6$ there is a transitive avoidance game on $n$ points, with all lines of size 3 , that is a first player win. Proof. Let $\mathcal{H}_{0}$ be the transitive avoidance game on 6 points that is a PI win given by Proposition 10 and let $\Phi$ be a winning strategy for $\mathcal{H}_{0}$. Let $\mathcal{H}$ be the disjoint union of $2 k+1$ copies of $\mathcal{H}_{0}$ which we write as $[2 k+1] \times \mathcal{H}_{0}$. Obviously $\mathcal{H}$ has $12 k+6$ points, all lines have size 3 , and it is transitive. Thus we just need to show that this game is a PI win. Let $f$ be an involution from $[2 k+1] \rightarrow[2 k+1]$ fixing 1 and having no other fixed point. PI starts by playing according to the winning strategy in the first copy of $\mathcal{H}_{0}$. For all subsequent moves he plays as follows. Suppose that PII has just played a point $(x, y) \in[2 k+1] \times \mathcal{H}_{0}$. If $x \neq 1$ then PI plays $(f(x), y)$; if $x=1$ then PI plays according to $\Phi$ in the first copy of $\mathcal{H}_{0}$. It is easy to see that PI can follow this strategy (i.e., he never has to play a point that has already been played), and that this strategy is a PI win. One might wonder whether size 3 is special in the above theorem, but it is easy to show that there are arbitrarily large games that are PI wins and have all lines of any fixed size greater than 2. Corollary 25. For any $r \geq 3$ and $n_{0}$ there is a transitive game on $n>n_{0}$ points with lines all of size $r$ that is a first player win. Proof. Take the game $\mathcal{H}$ with lines $\mathcal{L}$ given by Theorem 24 with $k=$ $\max \left(n_{0}, 2 r\right)$. Define the new game $\mathcal{H}^{\prime}$ to have the same board and to have lines $$ \mathcal{L}^{\prime}=\left\{L^{\prime} \in[n]^{(r)}: L^{\prime} \supset L \text { for some } L \in \mathcal{L}\right\} $$ Let PI play as above. Then at some point PII forms a set $L \in \mathcal{L}$. If PII has played at least $r$ points then he has formed a set $L^{\prime} \in \mathcal{L}^{\prime}$ (any superset of a set $L$ which has size $r$ is in $\mathcal{L}^{\prime}$ ). If PII has not played this many points then PI just has to continue playing until PII has played $r$ points without forming a set in $\mathcal{L}^{\prime}$ himself. This is trivial: on each turn PI plays in a copy of $\mathcal{H}_{0}$ that has not been played in previously (where $\mathcal{H}_{0}$ is as in the construction for $\mathcal{H})$. This is possible as $k \geq 2 r$. ### 5.2 Fast Player I wins In the examples of PI wins we have given so far PII can avoid losing for a long time: indeed, in the even case he need not lose until the last turn of the game, and in the odd case he can play for at least time $n / 4$ (in the construction in Theorem 6 for board size $p q$ all lines have size $\left.p^{\prime} q^{\prime} \geq p q / 4\right)$. However, an iterated variant of the odd size 'majority of majorities' construction in Theorem 6 shows that there are games where PII must lose in time $o(n)$. Theorem 26. For any $\varepsilon>0$ there is a game $\mathcal{H}$ on $n$ points such that PII must lose before time $\varepsilon$ n. The proof of this Theorem, whilst simple, is a little tedious to write out. Since we prove a substantially stronger result in the next section we omit the proof. ### 5.3 Transitivity So far we have required that the game be transitive on the points. However, many natural games are more transitive than this. For example the Ramsey Avoidance Game is also line transitive (recall that edges in that game correspond to our points, and complete graphs there are our lines). In fact, the Ramsey Avoidance Game is point/line transitive in that any line, and any point in that line can be mapped, by an element of the automorphism group, to any other line and point in that line. All our PI wining games so far are not transitive on the lines so it is natural to ask whether this extra transitivity is enough to rule out the possibility of a PI win. We answer this in the next section. We remark that our 'majority of majorities' game may appear transitive on the lines, and it is true that the it is transitive on the set $\mathcal{W}$ of 'allowed subsets'. However, it is not transitive on the set $\mathcal{L}$ of lines. ## 6 The Torus Game In this section we introduce a very natural game that simultaneously satisfies all of the properties we discussed in the previous section: PI can force a win in time $o(n)$, all lines have size 3, and it is point/line transitive (indeed it is even more: it is also possible to map any point and line not containing it to any other point and line not containing it). Definition. The torus game $\mathcal{T}_{q}(d)$ is the game on board $\mathbb{Z}_{q}^{d}$, with lines $\mathcal{L}$ defined to be all subsets of the form $\{x, x+y, x+2 y, \ldots x+(q-1) y\}$ for $x, y \in \mathbb{Z}_{q}^{d}$ and $y \neq 0$. We show that the game $\mathcal{T}_{3}(d)$ is an avoidance game satisfying all the conditions mentioned at the start of this section. Obviously, all lines have size 3, and it is transitive (translation by any element of $\mathbb{Z}_{3}^{d}$ is in its automorphism group). The other stronger transitive properties mentioned above are also simple to verify. Theorem 27. The game $\mathcal{T}_{3}(d)$ is not a PII win. Proof. The map $g: \mathbb{Z}_{3}^{d} \rightarrow \mathbb{Z}_{3}^{d}$ defined by $g(x)=-x$ is an involution of the board $X$ with a single fixed point 0 . PI plays 0 on his first go. Then on each subsequent turn he plays $g(y)$ where $y$ is the point PII just played. Observe that this is a valid strategy: PI never plays a point that has already been played. Now suppose that at some point PI forms a set $L \in \mathcal{L}$. This set cannot contain 0 since all lines containing 0 are of the form $\{-y, 0, y\}$, and if $\mathrm{PI}$ has played $y$ then PII has played $-y$. Thus, if PI has all the points of $L$ then PII has already played all points of $-L$ which is also a line in $\mathcal{L}$. Thus PII has already lost. Corollary 28. For all sufficiently large $d$ the game $\mathcal{T}_{3}(d)$ is a PI win. Moreover, if $t_{d}$ is the longest PII can avoid losing then $t_{d} / 3^{d} \rightarrow 0$ as $d \rightarrow \infty$. Proof. View the torus $\mathbb{Z}_{3}^{d}$ as a Hales-Jewett cube $[3]^{d}$. Observe that any combinatorial line in the Hales-Jewett cube is a line in the game sense (i.e. is in $\mathcal{L})$. Now for any $d$ greater than the Hales-Jewett number $H J(q, 2)$ (i.e., the smallest $d$ such that any two colouring of the cube $[3]^{d}$ contains a monochromatic combinatorial line) the game cannot be a draw, and thus must be a PI win. To prove the bound on the time observe that, by the density version of the Hales-Jewett Theorem [6] (or, indeed, standard cap-set results), there exists a sequence $\varepsilon_{d}$ tending to zero such that any set of size $\varepsilon_{d} 3^{d}$ in $[3]^{d}$ contains a combinatorial line. Thus, by time $\varepsilon_{d} 3^{d}$ one player must have lost, so PII must have lost. Hence $t_{d} / 3^{d}<\varepsilon_{d}$ so $t_{d} / 3^{d} \rightarrow 0$ as claimed. We conclude with an example showing that there are also even-sized boards where PI can win quickly. Moreover, in this example all lines have size 3. Theorem 29. There are games $\mathcal{H}_{d}$ with board size $n_{d}=6 \cdot 3^{d}$ which are PI wins and, moreover, PII loses game $\mathcal{H}_{d}$ in time o $\left(n_{d}\right)$. Proof. The construction is an extension of the above. Let $\mathcal{H}_{1}$ be the game on six points that is a PI win given by Proposition 10. We define $\mathcal{H}_{d}$ to be $\mathcal{T}_{3}(d) \times \mathcal{H}_{1}$ where the lines in this product are a line in one of the component directions and constant in the other. More precisely, the board is the set $\mathbb{Z}_{q}^{d} \times \mathcal{H}_{1}$ and the lines are the set $$ \left(\bigcup_{\substack{x \in \mathcal{T}_{3}(d) \\ L \in \mathcal{L}\left(\mathcal{H}_{1}\right)}}\{x\} \times L\right) \cup\left(\bigcup_{\substack{L \in \mathcal{L}\left(\mathcal{T}_{3}(d)\right) \\ y \in \mathcal{H}_{1}}} L \times\{y\}\right) $$ We have to show that PI wins this game and that the game ends quickly. First, we give a winning PI strategy. Let $\Phi$ denote the winning strategy for PI in $\mathcal{H}_{1}$. As usual we view the board of $\mathcal{T}_{3}(d)$ as $\{-1,0,1\}^{d}$. PI starts by playing in the $(0,0, \ldots, 0)$ copy of $\mathcal{H}_{1}$ and plays according to $\Phi$. We call this copy of $\mathcal{H}_{1}$ the zero copy. Now for all subsequent moves PI does the following. If PII just played in the zero copy the PI also plays in the zero copy and follows the strategy $\Phi$. If PII played in any other copy $\left(x_{1}, x_{2}, \ldots, x_{d}\right)$ of $\mathcal{H}_{1}$ then PI plays the same point in the antipodal copy $\left(-x_{1},-x_{2}, \ldots,-x_{d}\right)$ of $\mathcal{H}_{1}$. Observe that PI cannot lose by forming any triple not including a point in the zero copy of $\mathcal{H}_{1}$ as PII would have formed the antipodal set and would have already lost. Furthermore PI cannot lose by forming a triple not wholly contained in the zero copy as PII would have the antipodal point. Finally, PI does not form a losing triple in $\mathcal{H}_{1}$ as he is following the winning strategy $\Phi$ there. To show that the game ends quickly consider the points in $\mathcal{T}_{3}(d) \times\{y\}$ for some $y \in \mathcal{H}_{1}$. If a player has more than $\varepsilon\left\|\mathcal{T}^{d}\right\|$ points in this set then, by the density Hales-Jewett theorem, provided $d$ is sufficiently large, he must have a combinatorial line which is a losing set in our game. Thus, if PII has not lost he has played a total of at most $6 \varepsilon\left|\mathcal{T}^{d}\right|=\varepsilon\left|\mathcal{H}_{d}\right|$ points. ## 7 Open Problems Our first open question concerns the case when $n$ is prime. Question 1. For which primes $n$ does there exist a transitive avoidance game on $n$ points that is a first player win? We know almost nothing in this case. When $n=3,5$ or 7 there is no transitive avoidance game that is a PI win. Indeed, 3 is trivial; 5 follows immediately from Theorem 3 . The case $|V|=7$ is slightly trickier but, by Theorem 3, we only need to consider the case where all lines have size three and, since 7 is prime, we may assume that the cyclic group $C_{7}$ acts on the board. This reduces the problem to a manageable number of cases. However, for 11 and 13 there are transitive avoidance games that are first player wins. These were found by computer search. The games we find are of the following form: there is a transitive intersecting family $\mathcal{W}$ of $\frac{n-1}{2}$-sets that PI can guarantee to make one of in his first $\frac{n-1}{2}$ moves. Thus setting $\mathcal{L}$ to be all other $\frac{n-1}{2}$-subsets we get the required game. For $n=11, \mathcal{W}$ is the set of all affine copies of the set $\{0,1,2,4,5\}$. For $n=13, \mathcal{W}$ is the set of all affine copies of the any of the sets $\{0,1,2,4,5,6\}$, $\{0,1,2,4,5,7\}$ or $\{0,1,3,4,5,7\}$. However, we do not have a 'nice' strategy for either game. We remark that we searched for games transitive under the affine group as this reduced the number of orbits to a level manageable by computer search. There are other PI wins which are less symmetric, and there are some which are not of this 'maker' form. We do not know the anything about the answer to this question for any primes greater than 13. Indeed, we do not even know if there are infinitely many such primes. Moreover, the number of possible transitive games for 17 is extremely large (even if we restrict to affinely transitive games) which makes a computer search, even for the next open case, impractical. Theorem 24 and Corollary 25 answer the question of for which $n$ and $r$ there exist transitive avoidance games on $n$ points with lines all of size $r$ that are PI wins, for infinitely many values of $n$ and $r$. However, the full characterisation remains open. Question 2. For which $n$ and $r$ does there exist a transitive avoidance game on $n$ points with all lines of size $r$ that is a PI win. In particular we do not know the full characterisation even for $r=3$. A family of games of particular interest is the class of sim-like games, which is defined as follows. These games have board the edge set of the complete graph $K_{n}$. The lines are sets of edges that are isomorphic to some forbidden graph (or family of graphs). For example, the Ramsey Avoidance Game $\operatorname{RAG}(n, k)$ is of this form: the board is the edge set of $K_{n}$ and the lines are the subgraphs isomorphic to $K_{k}$. (They have been called sim-like as the first non-trivial Ramsey Avoidance Game, $\operatorname{RAG}(6,3)$, is commonly called Sim.) We do not know if there are any sim-like games that are PI wins. Question 3. Does there exist a sim-like game that is a PI win? One particular property of sim-like game is that they have a large automorphism group. Indeed, the automorphism group of a sim-like game played on $K_{n}$ trivially contains $S_{n}$. We do not know whether this itself is enough to force a PII win. More generally it would be interesting to characterise for which automorphism groups there exist PI wins. Question 4. For which groups $G$ does there exist a transitive avoidance game with automorphism group $G$ that is a PI win? Finally, there is a natural definition of an infinite avoidance game. In this case the board has infinite size (with all lines finite) and a player loses if he forms a line. If the play continues forever with neither player losing then the game is deemed a draw. Question 5. Is there an infinite transitive avoidance game that is a PI win? We have no intuition as to the correct answer to this question; indeed, we do not even know the answer when all losing lines have size three. We remark that any such game must have 'many' losing lines, to avoid PII having an easy draw - for example, by picking a faraway point that does not complete a losing line. ## References [1] M. H. Albert and R. J. Nowakowski, editors. Games of no chance 3, volume 56 of Mathematical Sciences Research Institute Publications. Cambridge University Press, Cambridge, 2009. Papers from the Combinatorial Game Theory Workshop held in Banff, AB, June 2005. [2] J. Beck. Ramsey games. Discrete Math., 249(1-3):3-30, 2002. Combinatorics, graph theory and computing (Louisville, KY, 1999). [3] J. Beck. Combinatorial games, volume 114 of Encyclopedia of Mathematics and its Applications. Cambridge University Press, Cambridge, 2008 . [4] E. R. Berlekamp, J. H. Conway, and R. K. Guy. Winning ways for your mathematical plays. A K Peters, Ltd., Natick, MA, second edition, 2001. [5] P. J. Cameron, P. Frankl, and W. M. Kantor. Intersecting families of finite sets and fixed-point-free 2-elements. European J. Combin., $10(2): 149-160,1989$. [6] H. Furstenberg and Y. Katznelson. A density version of the HalesJewett theorem. J. Anal. Math., 57:64-119, 1991. [7] F. Harary. Achievement and avoidance games for graphs. In Graph theory (Cambridge, 1981), volume 13 of Ann. Discrete Math., pages 111-119. North-Holland, Amsterdam-New York, 1982. [8] D. Hefetz, M. Krivelevich, M. Stojaković, and T. Szabó. AvoiderEnforcer: the rules of the game. J. Combin. Theory Ser. A, 117(2):152163, 2010. [9] D. Hefetz, M. Krivelevich, and T. Szabó. Avoider-enforcer games. J. Combin. Theory Ser. A, 114(5):840-853, 2007. [10] J. R. Isbell. Homogeneous games. Math. Student, 25:123-128, 1957. [11] J. R. Isbell. Homogeneous games. II. Proc. Amer. Math. Soc., 11:159161, 1960. [12] X. Lu. A matching game. Discrete Math., 94(3):199-207, 1991. [13] G. Simmons. The game of SIM. J. Rec. Math., 2(2):66, 1969. [14] W. Slany. Graph ramsey games. Electronic Colloquium on Computational Complexity (ECCC), (47), 1999. # Игра в кварки А. Балакин, Ю. Петрова, М. Скопенков 1. Плоскость разбита на клетки в виде правильных шестиугольников. За ход король перемещается на любую соседнюю клетку, см. рисунок. Ходы вверх, вправо-вниз и влево-вниз назовем положительными, остальные - отрицательными. Король сделал несколько ходов и вернулся на исходную клетку. Чему может быть равна разность между числом положительных и отрицательных ходов? Эта незатейливая математическая задача имеет фундаментальное физическое следствие: заряд любой свободной частицы кратен заряду электрона. Хотите увидеть, как школьная математика помогает понять физику частиц? Читайте дальше! Знаний физики не потребуется, мы дадим всю необходимую информацию. Начнем с неформального введения; математические определения появятся после задачи 9. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-121.jpg?height=322&width=306&top_left_y=690&top_left_x=1663) ## Элементарные частицы и места их обитания ${ }^{1}$ Вещество состоит из атомов, атомы - из ядра и электронов, ядра - из протонов и нейтронов. Все это частицы вещества. Частицы бывают элементарные и составные, т.е., состоящие из других частиц. Так, протон состоит из двух $u$-кварков и одного $d$-кварка, а сами кварки уже элементарны; см. рисунок. Для наших целей неважно, что такое кварки; можно представлять себе просто клейкие шарики, которые слипаются, образуя другие частицы. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-121.jpg?height=211&width=208&top_left_y=1154&top_left_x=1735) -протон- Если в эксперименте частица отделена от всех остальных (т.е. движется независимо от них), то она называется свободной. Так, обычно нейтрон содержится в ядре, но может стать свободным в ходе ядерной реакции. Эта подборка задач о пленении кварков (основные задачи - 11, 27, 33.b): Кварки не бывают свободными. Это значит нечто парадоксальное: кварки существуют, но их в принципе невозможно наблюдать в эксперименте. Мы обсудим как причины, так и следствия этого (пока не доказанного) утверждения ${ }^{2}$. Каждая частица имеет заряд. Так, заряд электрона равен -1 (таков выбор единиц измерения), заряд $u$-кварка равен $+2 / 3$. Заряд составной частицы равен сумме зарядов составляющих. Так, заряд протона равен +1 . Каждая частица имеет античастииу с противоположным зарядом. Так, заряд $u$-антикварка равен $-2 / 3$. Список всего нескольких из известных кварков и свободных составных частиц приведен в таблице. | частица | | | | античастица | | | | | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | | название | обозначение | состав | заряд | название | обозначение | состав | заряд | | Элеметнарные частицы: кварки (слева) и антикварки (справа) | | | | | | | | | | $\bar{u}$ | - | $+2 / 3$ | u-антикварк | $\overline{\bar{v}}$ | - | $-2 / 3$ | | $d$-кварк | $d$ | - | $-1 / 3$ | $d$-антикварк | $\bar{d}$ | - | $+1 / 3$ | Составные частицы из кварка и антикварка | пион $\pi^{+}$ $u \bar{d}$ +1 пион $\pi^{-}$ $\bar{u} d$ -1
протон
нейтрон $p$ иu +1 антипротон $\bar{p}$ $\bar{u} \bar{u} \bar{d}$ $?$ | | :--- | 2. Заполните клетки таблицы, в которых стоят знаки вопроса.[^7] Каждая частица еще имеет ияет (со светом он никак не связан). Это уже не число, а вектор на плоскости. В дальнейшем будем считать, что $R B^{\prime} G R^{\prime} B G^{\prime}$ - правильный шестиугольник с центром в начале координат $O$. Кварк может иметь один из цветов $\overrightarrow{O R}, \overrightarrow{O G}, \overrightarrow{O B}$ ("красный”, “зеленый”, "синий”). Т.е. на самом деле, существует 3 разных $u$-кварка и 3 разных $d$-кварка. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-122.jpg?height=237&width=280&top_left_y=70&top_left_x=1690) Цвет античастицы противоположен цвету частицы. Так, античастица красного $u$-кварка имеет цвет $\overrightarrow{O R}$ (“бирюзовый”). Цвет составной частицы - это векторная сумма цветов составляющих. Например, протон, состоящий из красного $u$-кварка, зеленого $u$-кварка и синего $d$-кварка, имеет нулевой цвет. Настоящая причина пленения кварков - это пленение цвета: Частицы ненулевого цвета не бывают свободными. Для разминки обсудим простейшие следствия пленения цвета. 3. а) Для каких $q$ и $p$ с условием $p+q \leqslant 3$ есть частица нулевого цвета из $q$ кварков и $p$ антикварков ${ }^{3}$ ? b) Докажите, что каждая частица нулевого цвета, состоящая из кварков и антикварков, указанных в таблице выше, обязательно имеет целочисленный заряд. Начнем наш путь к пониманию причин пленения цвета. Нам потребуется теория, описывающая взаимодействие кварков. Все известные взаимодействия, кроме гравитации, описываются калибровочной теорией. Это касается не только кварков, но и каждодневных явлений, вроде примагничивания проводников с током. Как мы сейчас увидим, идея калибровочной теории очень проста. ## Игрушечная модель калибровочной теории Несколько городов соединены дорогами в форме решетки $M \times N$; см. рисунок. В каждом городе свои товары (в неограниченном количестве). Например, в городе $A$ - апельсины, в городе $B$ бананы. Для пары соседних городов $A$ и $B$ фиксирован курс обмена $U(A B)>0$, например, 2 банана за апельсин. Курс симметричен, т.е. $U(B A)=U(A B)^{-1}$ : за 2 банана получаем назад свой апельсин. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-122.jpg?height=588&width=1592&top_left_y=1351&top_left_x=230) Хитрый горожанин может проехать вокруг квадрата $A B C D$ размера $1 \times 1$, меняясь по пути, в результате чего его начальный запас товаров умножится на $U(A B) U(B C) U(C D) U(D A)$. Так, на рисунке слева он получит 8-кратную прибыль. Обозначим $U(A B C D):=U(A B) U(B C) U(C D) U(D A)$. Как мы увидим, удобнее рассматривать логарифмы этих множителей. Все, что понадобится знать о логарифмах, это следующее определение. Если $y=2^{x}$ для действительных $x$ и $y$, то $x$ называется логарифмом числа $y$ и обозначается $x=\log _{2} y$. Так, $\log _{2} 2=1, \log _{2} \sqrt{2}=\frac{1}{2}, \log _{2} 1=0, \log _{2} \frac{1}{2}=-1$. $\mathrm{B}$ частности, путешествие вокруг квадрата $A B C D$ в том или ином направлении даст прибыль, если $\log _{2} U(A B C D) \neq 0$. Общий доход от спекуляиий измеряется суммой $S[U]$ величин $\log _{2}^{2} U(A B C D)$ по всем квадратам $A B C D$ размера $1 \times 1$ (обходятся они против часовой). Так, на рисунке слева $$ S[U]=\log _{2}^{2} U(A B C D)+\log _{2}^{2} U(D C F E)=\left(\log _{2} 8\right)^{2}+\left(\log _{2} \frac{1}{2}\right)^{2}=10 $$ Вы - король и можете устанавливать курсы обмена везде, кроме границы решетки. Вы устанавливаете их, чтобы минимизировать величину $S[U]$. Полученные курсы назовем оптилальными. 4. Наведите порядок в королевстве на рисунке слева, т.е. подберите значение $x$, для которого минимален общий доход от спекуляций на рисунке в центре.[^8] Оптимальные курсы обмена можно найти приближенно на компьютере, перебирая с маленьким шагом все возможные курсы из некорого интервала и сравнивая соответствующие значения $S[U]$. 5. Сделайте это для королевств на рисунках в центре и справа. А есть ли более быстрый алгоритм? Заметим, что замена переменных $x(A B):=\log _{2} U(A B)$ сильно упрощает выражение для дохода: $$ S[x]:=\sum_{\text {все квадраты } A B C D \text { размера } 1 \times 1}(x(A B)+x(B C)+x(C D)+x(D A))^{2} $$ Для квадрата $A B C D$ размера $1 \times 1$ обозначим $x(A B C D):=x(A B)+x(B C)+x(C D)+x(D A)$. Так, на рисунке слева $x(A B C D)=3$. Обозначим через $W$ множитель, на который умножится запас товаров при путешествии по границе всей решетки против часовой стрелки. Так, на рисунке слева $W=4$. 6. Предположим, что фиксированные курсы обмена на границе такие, как на рисунке, т.е. $$ U(A B)= \begin{cases}2, & \text { если дорога } A B \text { принадлежит северной или южной границе решетки } \\ 1, & \text { и направлена против часовой стрелки вдоль границы; }\end{cases} $$ Наведите порядок в королевстве для следующих размеров решетки, т.е. заполните таблицу": | Решетка | $1 \times 2$ | $1 \times 3$ | $1 \times N$ | $2 \times 2$ | | :--- | :---: | :---: | :---: | :---: | | Величина $W$ | 4 | | | | | Минимальное значение $S[U]$ | | | | | | Оптимальные курсы для всех дорог | | | | | | Величины $x(A B)$ для всех дорог $A B$ | | | | | | Величины $x(A B C D)$ для квадратов | | | | | 7. а) Для каких значений курса обмена на границе решетки $M \times N$ можно достичь равенства $S[U]=0$ ? b)* Пусть $S[U]=0$. Может ли горожанин получить прибыль, двигаясь по замкнутому пути? с) Для каких значений $M$ и $N$ оптимальные курсы однозначно определяются курсами на границе? d) Как связаны $M, N, W$ и минимальный доход от спекуляций? е) Какая из решеток $-8 \times 8$ или $7 \times 9$ - дает меньший доход от спекуляций при одинаковом $W$ ? 8. Пусть теперь разрешено менять курсы обмена на границе. Какое минимальное значение величины $$ \frac{1}{2} S[U]-j \log _{2} W $$ можно получить для данного вещественного числа $j$ на решетке $M \times N$ ? ## Физическая интерпретация Наша игрушечная модель позволяет описать магнитное взаимодействие проводников с током. Пусть дано вещественное число $j$. Граница решетки - это рамка с током силы $j$. Ток в рамке создает магнитное поле. Величина $x(A B C D)$ для квадрата $A B C D$ размера $1 \times 1$ - это магнитный поток через квадрат, сумма $\frac{1}{2} S[x]$ - это энергия магнитного поля, а $\frac{1}{2} S[x]-j \log _{2} W$ - общая энергия системы. Всякая система стремится минимизировать энергию. Поэтому величины $x(A B)$ выбираются так, чтобы минимизировать величину $\frac{1}{2} S[x]-j \log _{2} W$. (Осторожно: это грубая модель!) Если расположение проводников не фиксировано, то проводники стремятся сдвинуться так, чтобы общая энергия уменьшилась. Это означает притяжение или отталкивание проводников. При этом движение проводников понимается как изменение количества клеток в решетке, а не размера самих клеток. 9. а) Стремится ли рамка с током уменьшить или увеличить свою площадь? b) Притягиваются или отталкиваются два проводника с противоположно направленными токами? c)* А если направления токов совпадают? Указание. Замкните проводники вдалеке, чтобы получилась рамка (для b)) или 2 рамки (для с)).[^9] Взаимодействие кварков описывается квантовой калибровочной теорией. Ее главное отличие в том, что курсы обмена становятся случайными и не обязательно положительными. Для простоты будем считать, что курсы обмена принимают всего два значения: +1 и -1 , а коротко: "+" и "-". Обход по границе квадрата меняет запас товаров, если на его границе нечетное число знаков "-". Пришло время для точных математических формулировок. Сначала мы сформулируем нашу новую модель, затем пару задач, и только после этого постепенно дадим все необходимые определения. ## Игрушечная модель квантовой калибровочной теории Рассмотрим клетчатую полосу $1 \times N$; см. рисунок. Каждой из сторон квадратов $1 \times 1$ случайным образом сопоставим знак "+" или "-" (так, что все расстановки знаков имеют одинаковую вероятность; определение вероятности см. перед задачей 12). Для расстановки $U$ обозначим через $S^{\prime}[U]$ количество квадратов $1 \times 1$, на сторонах которых нечетное число ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-124.jpg?height=275&width=146&top_left_y=385&top_left_x=1820) знаков "-". Например, на рисунке $S^{\prime}[U]=2$. Математическое ожидание случайной величины $S^{\prime}[U]$ назовем энергией $E(N)$ электромагнитного взаимодействия кварка и антикварка на расстоянии $N$. (Определения случайной величины и математического ожидания см. перед и после задачи 17.) Замечание. Последний термин - неделимый; слова 'кварк', 'антикварк', 'электромагнитное взаимодейстие' не имеют формального смысла по-отдельности. Острожно: модель крайне грубая! 10. Найдите энергию $E(N)$ для полосок размера: $1 \times 1 ; 1 \times 2 ; 1 \times 3 ; 1 \times N$; a) приближенно с помощью компьютерной симуляции; b) точно. 11. (Пленение кварков в 1-мерном пространстве.) Существует ли такое число $E_{0}$, что $E(N) \leqslant E_{0}$ для каждого $N$ ? (То же неформально: конечное или бесконечное количество энергии требуется, чтобы развести кварк и антикварк очень далеко друг от друга?) ## Классическая вероятность ${ }^{5}$ Знакомство с теорией вероятностей полезно начинать на «физическом» уровне строгости, как в книгах [Shen], [KZhP]. Здесь же мы сразу даём «математические» определения. Однако мы приводим многие задачи на «практическом» языке и показываем на примерах, как их формализовать. Формализацию остальных задач оставляем читателю. Такая формализация является первым шагом решения, от которого может зависеть ответ. Рассмотрим эксперимент, имеющий $m$ равновозможных исходов, например бросание игральной кости, вытаскивание карты из колоды и т.д. Если интересующее нас событие (например, выпадение шестёрки, вытаскивание туза и т.д.) происходит в $a$ из этих исходов, то вероятность события считают равной $p=a / m$. Это пояснение полезно для начинающего, но не является математическим определением. Вот математическое определение. Вероятностъю подмножества $A$ конечного множества $M$ называется число $$ P(A)=P_{M}(A):=|A| /|M| $$ Далее, если не оговорено противное, множество $M$ фиксировано и пропускается из обозначений. Тогда вероятность определена для всех его подмножеств. Их часто называют событиями. 12. Из колоды в 52 карты вытаскивается одна карта. Найдите вероятность того, что она окажется (a) чёрной масти; (b) тузом; (с) картинкой; (d) дамой пик; (е) королём или бубной. Например, в задаче 12 (с) множество $M$ («всех возможных исходов») совпадает с множеством карт в колоде, а множество $A$ («исходов, в которых происходит рассматриваемое событие») - с множеством картинок. Так эта и многие другие вероятностные задачи могут быть строго сформулированы на комбинаторном языке. 13. Монета бросается 3 раза. Найдите вероятность выпадения (a) трёх орлов; (b) двух орлов и решки.[^10] Для решения некоторых из вышеприведённых задач полезны следующие. 14. (а) Правило сложения. Пусть $A \cap B=\emptyset$. Выразите $P(A \cup B)$ через $P(A)$ и $P(B)$. (b) Выразите вероятность $P(A \cup B)$ через $P(A), P(B)$ и $P(A \cap B)$. (с) Правило умножения. Выразите вероятность $P_{M \times N}(A \times B)$ через $P_{M}(A)$ и $P_{N}(B)$. 15. а) Сторонам квадрата случайно сопоставлены знаки "+" и "-". Какова вероятность, что общее количество знаков "-" нечетно? b) Какова вероятность, что для полоски $1 \times 2$ выполнено $S^{\prime}[U]=0$ ? А $S^{\prime}[U]=1$ ? А $S^{\prime}[U]=2$ ? Следующая задача подводит к понятиям случайной величины и математического ожидания. 16. а) Вам предлагается такая игра. Вы платите 2 конфеты, затем бросается игральная кость, и вы получаете столько конфет, сколько очков выпадает. Выгодна ли вам эта игра? b) Правила те же, только в случае выпадения 1 очка вы платите 100 конфет. (У вас достаточно конфет, чтобы заплатить.) Выгодня ли вам эта игра? c) Банк предлагает вам стабильный доход совершенно бесплатно. Вы кладете в банк 8 конфет, после чего бросается игральная кость. Если выпадает 2,3 или 4 очка, то вы получаете назад свой вклад плюс еще 1 конфету вдобавок. Если выпадает 5 или 6 очков ("рост рынка"), то вы получите даже плюс 2 конфеты вдобавок ${ }^{6}$. Выгодна ли вам эта игра? Числовая функция $X$, заданная на $M$, называется случайной величиной. Множество пар $\left(x_{i}, p_{i}\right), i=$ $1,2, \ldots$, где $\left\{x_{1}, x_{2}, \ldots\right\}$ - множество возможных значений случайной величины $X$, а $p_{i}=P(\{m \in$ $\left.\left.M: X(m)=x_{i}\right\}\right), i=1,2, \ldots,-$ соответствующие им вероятности, называется распределением случайной величины $X$. Событие $\left\{m \in M: X(m)=x_{i}\right\}$ в дальнейшем сокращённо обозначается $X=x_{i}$. 17. Найдите распределение случайной величины $S^{\prime}[U]$ для полоски $1 \times 2$. Математическим ожиданием или средним значением случайной величины $X$ называется сумма $$ \mathrm{E}(X):=\sum x_{i} p_{i}=x_{1} P\left(X=x_{1}\right)+x_{2} P\left(X=x_{2}\right)+\ldots $$ 18. Найдите математическое ожидание случайной величины $S^{\prime}[U]$ для полоски $1 \times 2$. Договоримся вместо знака "+" писать +1 , а вместо знака " - " писать -1 . Пусть случайная величина $X_{k}$ равна произведению чисел на сторонах $k$-го сверху квадрата $1 \times 1$ в полоске $1 \times N$, а $W^{\prime}-$ произведению всех чисел на границе полоски. (Аналогично определяется $W^{\prime}$ для решетки $M \times N$.) 19. Выразите $S^{\prime}[U]$ и $W^{\prime}$ через $X_{1}$ и $X_{2}$ для полоски $1 \times 2$. Найдите $\mathrm{E}\left(X_{1}\right), \mathrm{E}\left(X_{2}\right)$ и $\mathrm{E}\left(W^{\prime}\right)$. 20. а) Докажите, что математическое ожидание случайной величины $X$ на $M$ равно $\sum_{m \in M} X(m) /|M|$. b) Докажите, что если $\mathrm{E}(X) \leqslant x$, то существует $m \in M: X(m) \leqslant x$. c) Случайная величина $X$ при всех $m \in M$ принимает одно и то же значение $\mu$. Найдите $\mathrm{E}(X)$. d) Пусть $a, b$ - вещественные числа, а $X, Y$ - случайные величины. Всегда ли верно равенство $\mathrm{E}(a X+b Y)=a \mathrm{E}(X)+b \mathrm{E}(Y)$ ? е) А равенство $\mathrm{E}(X Y)=\mathrm{E}(X) \mathrm{E}(Y)$ ? Перейдем теперь к более точной модели взаимодействия кварков (дальше мы работам уже с ней). ## Солидная модель квантовой калибровочной теории Зафиксируем число $c>1$, называемое константой взаимодействия. Каждой дороге решетки $M \times N$ случайным образом сопоставим знак "+" или "-" так, что вероятность расстановки $U$ равна $$ P[U]:=\frac{c^{-S^{\prime}[U]}}{\sum_{\text {все расстановки } V} c^{-S^{\prime}[V]}} $$ Энергией электромагнитного взаимодействия кварка и антикварка на расстоянии $N$ назовем число $$ E_{M, c}(N):=-\frac{1}{M} \log _{2} \mathrm{E}\left(W^{\prime}\right) $$ Наша цель - найти ее. (Здесь $\mathrm{E}\left(W^{\prime}\right)$ - это математическое ожидание случайной величины $W^{\prime}$, определенной перед задачей 19. Мы не знаем простого объяснения такой формулы для энергии.)[^11] ## Статистическая вероятность В этой модели используется такое более общее определение вероятности. Пусть задано множество $M$ и каждому $m \in M$ поставлено в соответствие неотрицательное число $P(m)$, причём сумма всех этих чисел равна 1. Тогда вероятностью события $A$ называется сумма чисел $P(m)$ по всем $m \in A$. 21. а) Для решетки $1 \times 1$ найдите вероятность каждой расстановки знаков, величины $\mathrm{E}\left(W^{\prime}\right)$ и $E_{1,2}(1)$. b) То же, если разрешаются только расстановки, где на всех дорогах, кроме верхней, стоит "+". 22. Как изменится вероятность расстановки, если поменять знаки на всех дорогах из одного города? Следующее определение обобщает ситуацию правила умножения 14 (с). Пусть сначала вероятности $P(m)$ всех элементов $m$ множества $M$ одинаковы. Подмножества (т. е. события) $A$ и $B \neq \emptyset$ множества $M$ независимы, если доля (т. е. вероятность) множества $A \cap B$ в $B$ равна доле (т.е. вероятности) множества $A$ в $M$. Приведём симметричную переформулировку, которая работает и для $B=\emptyset$, и для более общего определения вероятности, когда не все числа $P(m)$ равны. Подмножества $A$ и $B$ конечного множества $M$ называются независимыми, если $$ P(A \cap B)=P(A) \cdot P(B) $$ Основной пример независимых подмножеств - в множестве всех клеток шахматной доски подмножество клеток в первых трёх её строках и подмножество клеток в последних четырёх её столбцах. 23. Независимы ли следующие подмножества, если вероятности элементов множества $M$ одинаковы? (a) Подмножества $\{1,2\} \subset\{1,2,3,4\}$ и $\{1,3\} \subset\{1,2,3,4\}=M$. (b) Подмножества $\{1,2\} \subset\{1,2,3,4,5,6\}$ и $\{1,3\} \subset\{1,2,3,4,5,6\}=M$. Случайные величины $X$ и $Y$ называются независимыми, если события $X=x_{i}$ и $Y=y_{j}$ независимы при любых $x_{i}, y_{j}$, т. е. $$ P\left(\left\{m \in M: X(m)=x_{i} \text { и } Y(m)=y_{j}\right\}\right)=P\left(X=x_{i}\right) P\left(Y=y_{i}\right) $$ 24. Независимы ли случайные величины $X_{1}$ и $X_{2}$ на полоске $1 \times 2$ (см. определение перед задачей 19 )? 25. Докажите, что если случайные величины $X$ и $Y$ независимы, то математическое ожидание их произведения равно произведению их математических ожиданий: $\mathrm{E}(X Y)=\mathrm{E}(X) \mathrm{E}(Y)$. 26. (Закон площади Вильсона) Пусть $c=2, M$ и $N$ произвольны. Найдите $\mathrm{E}\left(W^{\prime}\right)$ и энергию $E_{M, 2}(N)$. При каких $M_{1}, N_{1}, M_{2}, N_{2}$ решетка $M_{1} \times N_{1}$ дает меньшее значение $\mathrm{E}\left(W^{\prime}\right)$, чем решетка $M_{2} \times N_{2}$ ? 27. (Пленение кварков в 2 -мерном пространстве.) Конечное или бесконечное количество энергии требуется, чтобы развести кварк и антикварк очень далеко друг от друга в 2-мерном пространстве? 28. Исследуйте 3-мерную решетку $N \times N \times N$ и 4-мерную решетку $N \times N \times N \times N$ с помощью компьютерной симуляции для различных $c \in[2 ; 3]$. Для них $S^{\prime}[U]$ по-прежнему определяется как число квадратов $1 \times 1$ с нечетным числом знаков " - ", а $W^{\prime}$ - как произведение знаков по границе одной из 2 -мерных граней. Конечное или бесконечное количество энергии требуется, чтобы развести кварк и антикварк очень далеко друг от друга в 3 - и 4 -мерном пространстве? 29. Докажите следующие утверждения. а) Отношение энергий $E_{M, c}(N) / E(N)$ не зависит $N$. b) При $c \rightarrow \infty$ энергия $E_{M, c}(N)$ стремится к минимальному значению величины $S^{\prime}[U]$, т.е. к 0 . ## Геометрический взгляд на калибровочную теорию* Посмотрим на нашу модель с новой стороны (этот текст неформален и далее не используется). Заряды частиц бывают положительные и отрицательные. Какой из них считать положительным - условность: это просто выбор направления на числовой оси. А что если выбор "направления оси" в каждой точке пространства делать независимо? Скажем, в каждом городе решетки $M \times N$ выбирать направление по-своему. Тогда для каждой дороги нужно указать, как меняется направление оси при переходе от города к соседнему. Можно представить себе ситуацию, когда при обходе квадрата $1 \times 1$ направление оси заменяется на противоположное (как направление перпендикуляра к ленте Мебиуса при ее обходе). Геометрический взгляд на калибровочную теорию состоит в том, что такие "дефектные" квадраты - это и есть магнитное поле, и они несут энергию. Поэтому в моделях мы считали энергию через количество "дефектных" квадратов. Конечно, буквально эту конструкцию понимать нельзя: при обходе по замкнутому контуру знак заряда, конечно, не изменится. "Направление оси" - это нечто другое, чем знак заряда. 30. (Теорема Гаусса-Бонне) Рассмотрим а) куб; b) правильный тетраэдр; с)* выпуклый многогранник. Два вектора, лежащих в соседних гранях, назовем параллельными, если они образуют равные "вертикальные" углы с общей стороной этих граней (т.е. становятся параллельными и одинаково направленными, если "развернуть" две грани вокруг общей стороны, см. рисунок). Пусть $f_{1}, f_{2}, \ldots, f_{k}$, $f_{1}$ - все грани вокруг вершины $v$ в естественном порядке. Начнем с произвольного вектора $\vec{e}_{1} \subset f_{1}$ и возьмем вектор $\vec{e}_{2} \subset f_{2}$ параллельный $\vec{e}_{1}$, потом вектор $\vec{e}_{3} \subset f_{3}$ параллельный $\vec{e}_{2}, \ldots$, наконец, вектор $\vec{e}_{k+1} \subset f_{1}$ параллельный $\vec{e}_{k}$. Обозначим через $\phi_{v}$ ориентированный угол между $\vec{e}_{k+1}$ и $\vec{e}_{1}$. Найдите сумму углов $\phi_{v}$ по всем вершинам $v$. Цвет кварка - уже не число, а вектор, для которого возможны три направления. Поэтому в теории сильного (т.е. цветового) взаимодействия на дорогах ставятся перестановки трех цветов. ## Перестановки Перестановка множества - запись элементов этого множества в некотором порядке. Если говорить более строго, перестановкой множества называется взаимно однозначное отображение этого множества на себя (т. е. биекция). Перестановка множества $\left\{a_{1}, a_{2}, \ldots, a_{n}\right\}$, переводящая $a_{k}$ в $f\left(a_{k}\right)$, записывается в виде $$ \left(\begin{array}{cccc} a_{1} & a_{2} & \ldots & a_{n} \\ f\left(a_{1}\right) & f\left(a_{2}\right) & \ldots & f\left(a_{n}\right) \end{array}\right) $$ обычно $a_{k}=k$ для всех $k=1, \ldots, n$. Композицией перестановок $f$ и $g$ называется перестановка $f \circ g$, определённая формулой $(f \circ g)(x):=f(g(x))$. 31. Найдите композиции (a) $\left(\begin{array}{lll}1 & 2 & 3 \\ 2 & 1 & 3\end{array}\right) \circ\left(\begin{array}{lll}1 & 2 & 3 \\ 3 & 1 & 2\end{array}\right)$; (b) $\left(\begin{array}{lll}1 & 2 & 3 \\ 2 & 3 & 1\end{array}\right) \circ\left(\begin{array}{lll}1 & 2 & 3 \\ 3 & 1 & 2\end{array}\right)$. 32. Для любых ли двух перестановок $x$ и $y$ выполнено $x \circ y=y \circ x$ ? ## Игрушечная модель неабелевой калибровочной теории Зафиксируем число $c>1$. Обозначим через $S_{3}$ множество всех перестановок множества $\{1,2,3\}$. Неподвижная точка перестановки $f$ - это такое $x \in\{1,2,3\}$, что $f(x)=x$. След $\operatorname{Tr}(f)$ перестановки $f$ - это число ее неподвижных точек минус 1 . Например, $\operatorname{Tr}\left(\begin{array}{lll}1 & 2 & 3 \\ 3 & 2 & 1\end{array}\right)=0$. Каждой дороге $A B$ решетки $M \times N$ случайным образом сопоставим перестановку $U(A B) \in S_{3}$ так, что $U(A B) \circ U(B A)$ - тождественная перестановка, и вероятность расстановки $U$ равна $$ P[U]:=\frac{c^{-S^{\prime \prime}[U]}}{\sum_{\text {все расстановки } V} c^{-S^{\prime \prime}[V]}} $$ где $$ S^{\prime \prime}[U]:=\sum_{\substack{\text { все квадраты } \\ \text { (вершины обозначены против часовой стрелки })}}(2-\operatorname{Tr}(U(A B) \circ U(B C) \circ U(C D) \circ U(D A))) \text {. } $$ (Т.е. чем меньше элементов остаются неподвижными при обходе квадрата, тем больший вклад в $S^{\prime \prime}[U]$ он дает.) Обозначим через $W^{\prime \prime}$ след произведения перестановок на всех граничных дорогах, если обходить границу против часовой стрелки, начиная с левого нижнего угла. Величина $-\frac{1}{M} \log _{2} \mathrm{E}\left(W^{\prime \prime}\right)$ называется энергией сильного взаимодействия кварка и антикварка на расстоянии $N$. 33.* а) Придумайте понятие математического ожидания случайной перестановки из $S_{3}$, обладающее как можно большим числом свойств математического ожидания случайной величины. Подсказка. Нам не хватает операции сложения перестановок. Она появится сама собой, если вспомнить, что $1,2,3$ - это цвета кварков, цвета - векторы, а перестановка - отображение. b) Найдите математическое ожидание случайной величины $W^{\prime \prime}$ при $c=2$. Докажите, что найдутся такие числа $a, b, B \in \mathbb{R}$, что $b a^{M N} \leqslant \mathrm{E}\left(W^{\prime \prime}\right) \leqslant B a^{M N}$ для всех $M, N$. 34. ("Сущность" Проблемы Тысячелетия) То же неравенство для 4-мерной решетки $N \times N \times N \times N$ и прямоугольника $M \times N$ на одной из ее 2 -мерных граней. ## Дополнительные задачи Эти сложные задачи выдаются тем, кто решил большинство из предложенных выше задач. 35. Решите аналог задачи 33.b, если определение следа перестановки заменить на следующее: $$ \operatorname{Tr}^{\prime}(x)= \begin{cases}+3, & \text { если } x-\text { тождественная, т.е. } x=\left(\begin{array}{lll} 1 & 2 & 3 \\ 1 & 2 & 3 \end{array}\right) \\ 0, & \text { если } x-\text { цикл длины } 3, \text { т.е. } x=\left(\begin{array}{lll} 1 & 2 & 3 \\ 2 & 3 & 1 \end{array}\right) \text { или } x=\left(\begin{array}{lll} 1 & 2 & 3 \\ 3 & 1 & 2 \end{array}\right) \\ -1, & \text { если } x-\text { транспозиция, т.е. } x=\left(\begin{array}{lll} 1 & 2 & 3 \\ 2 & 2 & 3 \end{array}\right), x=\left(\begin{array}{lll} 1 & 3 \\ 3 & 2 & 1 \end{array}\right) \text { или } x=\left(\begin{array}{lll} 1 & 2 & 3 \\ 1 & 3 & 2 \end{array}\right)\end{cases} $$ а в формуле для $S^{\prime \prime}[U]$ выражение $2-\operatorname{Tr}(\ldots)$ заменить на $3-\operatorname{Tr}^{\prime}(\ldots)$. 36. (Разложение сильной связи) Рассмотрим солидную модель на 3 -мерной решетке $N \times N \times N$ из задачи 28. Найдите асимптотику выражения $\mathrm{E}\left(W^{\prime}\right)$ при $c \rightarrow \infty$, т.е. такую функцию $f_{N}(c)$, что $\lim _{c \rightarrow \infty} \frac{\mathrm{E}\left(W^{\prime}\right)}{f_{N}(c)}=1$ для каждого $N$. ## Вместо эпилога (Подводные камни) Мы надеемся, что некоторые из наших читателей заинтересовались элементарными частицами и хотят узнать о них побольше. В качестве эпилога мы дадим несколько предупреждений таким читателям. В научно-популярной литературе теория элементарных частиц обычно слишком упрощена. Эта подборка задач не исключение. Рассмотренные нами игрушечные модели довольно грубые, и относиться к ним стоит с некоторой долей критики. Их единственное преимущество - простота; если принимать их слишком серьезно, то они могут привести к развитию неправильной физической интуиции. Только “Солидная модель квантовой калибровочной теории" признана и действительно рассматривается в физической литературе. А для настоящего понимания теории элементарных частиц необходимо хорошее знание как физики, так и математики. Задача, которую мы назвали "сущностью проблемы тысячелетия", ни в коем случае не равносильна проблеме тысячелетия в теории Янга-Миллса. И даже не является частным случаем последней. Это скорее наиболее содержательная - по очень субъективному мнению авторов - часть проблемы, освобожденная от технических деталей. Хотя мы и пришли к ней, обсуждая пленение кварков, по сути она ближе к другому явлению: близкодействию ядерных сил. Также хотим заметить, что на сегодняшний день нет почти никаких математических результатов по калибровочной теории на решетках; обычно имеется только численное моделирование. Наконец, есть "теории Новой Физики", которые разрабатываются без каких-либо объективных критериев: такие теории не имеют ни экспериментального, ни математического подтверждения (а некоторые имеют экспериментальные опровержения). ## Указания, решения, ответы 1. Ответ: любому целому числу, кратному 3. Указание. Расставим остатки 0,1 и 2 при делении на три в клетки как показано на рисунке. Заметим, что положительный ход из любой клетки прибавляет 1 к остатку, на котором стоит король, а отрицательный - вычитает 1. 2. Ответ: | протон | $p$ | $u u d$ | +1 | антипротон | $\bar{p}$ | $\bar{u} \bar{u}$ | -1 | | :--- | :--- | :--- | :---: | :--- | :---: | :--- | :---: | | нейтрон | $n$ | $u d d$ | 0 | антинейтрон | $\bar{n}$ | $\bar{u} \bar{d} \bar{d}$ | 0 | ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-128.jpg?height=397&width=420&top_left_y=1837&top_left_x=1549) 3a. Ответ: $(q, p)=(1,1),(3,0)$ или $(0,3)$. Эта задача решается небольшим перебором, приведём пример рассуждения для одного случая. Почему $q$ и $p$ не могут равняться соответственно 2 и 0 : если два вектора в сумме дают 0 , то они противоположны, а два кварка не могут иметь противоположные цвета. 3 b. Указание. Пусть $p \leqslant q$. Заметим, что сумма зарядов произвольных кварка и антикварка - целое число. Значит, можно рассматривать только сумму зарядов $q-p$ кварков. Также заметим, что сумма зарядов любой тройки кварков - целое число. Из задачи 1 следует, что для частицы с нулевым цветом $q-p$ кратно 3 , то есть $q-p$ кварков можно разбить на тройки. 4. Ответ: $x=0$. Указание: $$ S[U]=\log _{2}^{2}\left(2 \cdot 1 \cdot \frac{1}{2^{x}} \cdot 1\right)+\log _{2}^{2}\left(2^{x} \cdot 1 \cdot 2 \cdot 1\right)=(1-x)^{2}+(1+x)^{2}=2+2 x^{2} $$ Минимум достигается при $x=0$. 6. Ответ: В таблице ниже в третьей и четвертой строках в первых трех столбцах перечислены величины для всех неграничных дорог, отсчет которых идет сверху вниз. При этом считается, что все эти дороги направлены слева направо. В этих же строках в последнем столбце считается, что все неграничные дороги направлены из центра к границам, а $x$ обозначает произвольное действительное число (то есть для каждого $x$ есть свое решение). Если для всех дорог (или квадратов) значение величины одно и то же, то в соответствующую клетку ставится только одно число. | Решетка | $1 \times 2$ | $1 \times 3$ | $1 \times N$ | $2 \times 2$ | | :--- | :---: | :---: | :---: | :---: | | Величина $W$ | 4 | 4 | 4 | 16 | | Минимум $S[U]$ | 2 | $\frac{4}{3}$ | $\frac{4}{N}$ | 4 | | Оптимальные курсы | 1 | $2^{-1 / 3}, 2^{1 / 3}$ | $2^{-1+\frac{2}{N}}, 2^{-1+\frac{4}{N}}, \ldots 2^{-1+\frac{2 N-2}{N}}$ | $2^{x}$ | | Величины $x(A B)$ | 0 | $-\frac{1}{3}, \frac{1}{3}$ | $-1+\frac{2}{N},-1+\frac{4}{N}, \ldots-1+\frac{2 N-2}{N}$ | $x$ | | Величины $x(A B C D)$ | 1 | $\frac{2}{3}$ | $\frac{2}{N}$ | 1 | Указание. В решении этой задачи, как и некоторых последующих, очень полезно неравенство между средним квадратическим и средним арифметическим, которое формулируется следующим образом: для любых действительных чисел $x_{1}, x_{2}, \ldots x_{n}$ верно неравенство: $$ \sqrt{\frac{x_{1}^{2}+x_{2}^{2}+\ldots+x_{n}^{2}}{n}} \geqslant \frac{x_{1}+x_{2}+\ldots+x_{n}}{n} $$ причем равенство достигается тогда и только тогда, когда $x_{1}=x_{2}=\ldots=x_{n} \geqslant 0$. Для решетки $1 \times N$ за $x_{1}, x_{2}, \ldots x_{n}$ стоит принять величины $x(A B C D)$ для всех квадратов $A B C D$ (занумерованных сверху вниз). Для решетки $2 \times 2$ обозначим курсы обмена в направлении из центра к границе за $2^{x}, 2^{y}, 2^{z}$ и $2^{t}$ против часовой стрелки. Тогда $$ \begin{aligned} S[U]= & (x+1-y)^{2}+(-z+y+1)^{2}+(1-t+z)^{2}+(1-x+t)^{2}= \\ = & 2 x^{2}+2 y^{2}+2 z^{2}+2 t^{2}+4-2 x y-2 y z-2 z t-2 t x= \\ & =(x-y)^{2}+(y-z)^{2}+(z-t)^{2}+(t-x)^{2}+4 \geqslant 4 \end{aligned} $$ Замечание. Есть и другие решения, некоторые из которых опираются на существование оптимальных курсов. В таком случае последнее нужно доказывать, а это довольно сложно, см. [SSU, Принцип достижения наименьшего значения на стр. 30]. 7a. (E. Павлов) Oтвет: $S[U]=0$ может достигаться тогда и только тогда, когда $W=1$. В самом деле, $S[U]=0$ тогда и только тогда, когда для каждого $1 \times 1$ квадрата $A B C D$ выполняется $U(A B C D)=1$. Рассмотрим произведение ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-129.jpg?height=161&width=572&top_left_y=1827&top_left_x=742) В этом произведении каждая неграничная дорога появляется дважды, один раз как $U(A B)$ и один раз как $U(B A)$, которые уничтожают друг друга. Граничные дороги появляются один раз с ориентацией против часовой стрелки. Это означает, что $\prod=W$. Так как каждый член в $\prod$ это просто 1 , из этого следует, что $W=\prod=1$ это необходимое условие. Теперь предположим, что $W=1$. Покажем, что существуют курсы $U(A B)$, такие что $S[U]=0$. Обозначим через $a_{0}, a_{1}, \cdots, a_{2 M+2 N-1}$ вершины на границе, начиная с нижней левой вершины. Далее введем функцию $P(A)$ на вершинах решетки следующим образом: $$ P(A)= \begin{cases}1, & \text { если } A \text { не на границе или нижняя левая вершина. } \\ \prod_{k=0}^{n-1} U\left(a_{k} a_{k-1}\right), & \text { если } A \text { это вершина } a_{\mathrm{n}} \text { на границе. }\end{cases} $$ Наконец, установим $U(A B)$ на неграничных дорогах по формуле $U(A B)=\frac{P(B)}{P(A)}$. Для каждого квадрата $A B C D$, мы имеем $$ \log _{2}^{2}(U(A B) U(B C) U(C D) U(D A))=\log _{2}^{2}\left(\frac{P(B)}{P(A)} \frac{P(C)}{P(B)} \frac{P(D)}{P(C)} \frac{P(A)}{P(D)}\right)=\log _{2}^{2}(1)=0 $$ Это означает, что $S[U]=0$, ч.т.д. 7b*. Достаточно построить функцию $P(A)$ на множестве городов, такую что $U(A B)=\frac{P(B)}{P(A)}$ для каждой дороги $A B$. Для каждого города $A$ возьмем за $P(A)$ произведение всех курсов обмена для $Г$-образного пути, начинающегося в нижнем левом городе и заканчивающегося в $A$. Тогда условие $U(A B C D)=1$ для каждого $1 \times 1$ квадрата $A B C D$ дает условие $U(A B)=\frac{P(B)}{P(A)}$. 7с. Для $M, N>1$ курсы не определены однозначно. Для внутреннего города мы можем изменить единицы измерения, например, обменивать дюжину яблок вместо одного. Такая калибровочная трансформация умножает курсы обмена для всех дорог, начинающихся в городе, в то же время сохраняя $S[U]$. 7d. Oтвет: $2^{\sqrt{M \cdot N \cdot S_{\min }}}=W$, где $S_{\text {min }}$ - минимальное значение $S[U]$. Указание. По неравенству между средним квадратическим и средним арифметическим: $$ \begin{gathered} S[U]=\sum_{\text {все квадраты } A B C D \text { размера } 1 \times 1} x(A B C D)^{2} \geqslant \\ \geqslant \frac{1}{M N}\left(\sum_{\text {все квадраты } A B C D \text { размера } 1 \times 1} x(A B C D)\right)^{2}=\frac{\log _{2}^{2} W}{M N} \end{gathered} $$ Осталось доказать, что всегда можно установить такие курсы обмена внутри королевства, чтобы равенство достигалось, то есть чтобы значения $x(A B C D)$ были равны $\frac{\log _{2} W}{M N}$ для всех квадратов $1 \times 1$. Для прямоугольника $1 \times N$ можно последовательно определять курсы на еще не определенных сторонах квадратов, обходя их сверху вниз. Эти курсы определяются однозначно, а для последнего квадрата величина $x(A B C D)$ будет равна $\frac{\log _{2} W}{M N}$, потому что сумма этих величин для всех квадратов фиксирована. Для прямоугольника $M \times N$ предлагается установить на дорогах, нарисованных жирными линиями (см. рис.), единичный курс обмена, а затем доопределить курсы на остальных дорогах. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-130.jpg?height=440&width=534&top_left_y=1248&top_left_x=767) 7e. Ответ: $8 \times 8$. Это следует из формулы в задаче $7 \mathrm{~d}$. 8. Ответ: $-M N j^{2} / 2$. Указание. Обозначим $w=\log _{2} W$. Для фиксированного $w$, минимум $S[U]$ равняется $\frac{w^{2}}{M N}$ по Задаче 7.d. Остается найти минимум квадратного многочлена $\frac{w^{2}}{2 M N}-j w$. 9a. Ответ: рамка стремится увеличить свою площадь. Первый шаг в решении этой задачи (и многих других) - математическая формализация. Пункт а) формализуется так: Рассмотрим минимальное значение $\frac{1}{2} S[x]-j \log _{2} W$ на решетке $M \times N$ как функцию от $M N$ при фиксированном ј. Возрастает или убывает эта функция? После формализации решение следует из Задачи 8.е. На этом факте основана работа любого элекрического двигателя или генератора. 9b. Ответ: проводники с противоположно направленными токами отталкиваются. Математическая формализация: рассмотрим минимальное значение $\frac{1}{2} S[x]-j \log _{2} W$ на решетке $M \times N$ как функцию от $N$ при фиксированных $M$ и ј. Возрастает или убывает эта функция? 9c*. Oтвет: проводники с одинаково направленными токами притягиваются. Математическая формализация: рассмотрим минимальное значение $\frac{1}{2} S[x]-j \log _{2} W$ на решетке $M \times$ $(L-N)$ как функцию от $N$ при фиксированных $L, M$ и j. Возрастает или убывает эта функция? 10b. Ответ: $\frac{1}{2}, 1, \frac{3}{2}, \frac{N}{2}$. Возьмем произвольный квадрат полосы $1 \times N$ и рассмотрим случайную величину, равную 0 , если на его сторонах четное количество знаков "-", и 1 , если нечетное. Путем недлинного перебора легко убедиться, что значения 0 и 1 равновероятны, то есть математическое ожидание этой величины равно $\frac{1}{2}$. Заметим, что величина $S^{\prime}[U]$ равна сумме рассматриваемых величин по всем квадратам $1 \times 1$, и, так как математическое ожидание суммы равно сумме математических ожиданий, выполняется равенство $E(N)=\frac{N}{2}$. 11. Ответ: нет. Указание. В этой модели энергия $E(N)=\frac{N}{2}$ стремится к бесконечности при стремлении $N$ к бесконечности. 12. Ответи: a) $\frac{1}{2}$; b) $\frac{1}{13}$; c) $\frac{3}{13}$; d) $\frac{1}{52}$; e) $\frac{15}{52}$. 13. Ответы: а) $\frac{1}{8}$; b) $\frac{3}{8}$. 14a. Omвет: $P(A \cup B)=P(A)+P(B)$. Пусть $A$ и $B$ - подмножества множества $M$. Так как $A$ и $B$ не пересекаются, $|A \cup B|=|A|+|B|$. Тогда по определению вероятность $$ P(A \cup B)=\frac{|A \cup B|}{|M|}=\frac{|A|+|B|}{|M|}=\frac{|A|}{|M|}+\frac{|B|}{|M|}=P(A)+P(B) $$ 14b. Oтвет: $P(A \cup B)=P(A)+P(B)-P(A \cap B)$. 14c. Oтвет: $P_{M \times N}(A \times B)=P_{M}(A) P_{N}(B)$. 15. Ответы: a) $\frac{1}{2}$; b) $\frac{1}{4}, \frac{1}{2}, \frac{1}{4}$. Приведем пример рассуждения для пункта b). Разобьем все расстановки знаков на пары следующим образом: две расстановки состоят в паре тогда и только тогда, когда они отличаются только знаком на верхней стороне. Заметим, что замена знака на верхней стороне меняет четность количества знаков "-" на сторонах верхнего квадрата. Значит, таких расстановок, что $S^{\prime}[U]=1$, ровно столько, сколько расстановок, дающих $S^{\prime}[U]=0$ и $S^{\prime}[U]=2$, вместе взятых. То есть половина от общего числа расстановок. Аналогично объединим в пары расстановки, дающие $S^{\prime}[U]=0$ и $S^{\prime}[U]=2$, отличающиеся только знаком на центральной стороне, и поймем, что расстановок, дающих $S^{\prime}[U]=0$, столько же, сколько и дающих $S^{\prime}[U]=2$. 16. Ответы: а) да; b) нет; с) нет. 17. Ответы: $\{(0,1 / 4),(1,1 / 2),(2,1 / 4)\}$. На самом деле, это задача 15. 18. Ответ: 1. 19. Omвeт: $S^{\prime}[U]=\left(2-X_{1}-X_{2}\right) / 2, W^{\prime}=X_{1} X_{2}, \mathrm{E}\left(X_{1}\right)=\mathrm{E}\left(X_{2}\right)=\mathrm{E}\left(W^{\prime}\right)=0$. 20. Ответы: с) $\mu$; d) да; е) нет. Контрпример к е $): \mathrm{E}\left(X_{1}^{2}\right)=\mathrm{E}(1)=1 \neq 1 / 4=\mathrm{E}\left(X_{1}\right) \mathrm{E}\left(X_{1}\right)$. 21. а) Oтвет: Вероятность расстановки с четным количеством знаков "-" равна $\frac{c}{8(c+1)}$, с нечетным $-\frac{1}{8(c+1)}$; $\mathrm{E}\left(W^{\prime}\right)=\frac{c-1}{c+1}, E_{1,2}(1)=\log _{2} 3$. b) Oтвет: Вероятность расстановки без знаков "-" равна $\frac{c}{c+1}$, с одним знаком "-" - $\frac{1}{c+1}$; $\mathrm{E}\left(W^{\prime}\right)$ и $E_{1,2}(1)$ такие же, как и в пункте а). 22. Ответ: вероятность не изменится. 23. Ответы: a) да; b) нет. 24. Ответ: да. Указание. Следующее упрощение солидной модели очень удобно для этой и многих других задач. Изменение знаков всех дорог, исходящих из определенного города, называется калибровочным преобразованием. По Задаче 22 это действие не меняет вероятность расстановки. Оно не меняет ни случайные величины $X_{k}$, ни $W^{\prime}$. Для произвольной расстановки, с помощью последовательности калибровочных трансформаций можно сделать все знаки на жирных отрезках на рисунке плюсами. Следовательно, можно полагать, что только такие расстановки, что все жирные отрезки имеют знак " + ", разрешены в модели. Далее, знаки всех остальных отрезков однозначно определены величинами $X_{k}$ на всех квадратах. Следовательно, наша модель эквивалентна следующему процессу Бернулли: Пусть вероятность набора чисел $X_{1}, \ldots, X_{M N} \in\{+1,-1\}$ равна $$ P\left(X_{1}, \ldots, X_{M N}\right)=\frac{c^{\left(M N-X_{1}-\cdots-X_{M N}\right) / 2}}{\sum_{\text {все возможсные наборы }\left(Y_{1}, \ldots, Y_{M N}\right)} c^{\left(M N-Y_{1}-\cdots-Y_{M N}\right) / 2}} $$ Тогда энергия $E_{M, c}(N)=-\frac{1}{M} \log _{2} \mathrm{E}\left(X_{1} \cdots X_{M N}\right)$. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-131.jpg?height=431&width=534&top_left_y=2389&top_left_x=767) 26. Oтвет: $\mathrm{E}(W)=\frac{1}{3^{M N}}, E_{M, 2}(N)=N \log _{2} 3$. Указание. Используйте упрощенную модель из решения Задачи 24 . Покажите, что $X_{1}, \ldots, X_{M N}$ независимы, следовательно $\mathrm{E}\left(X_{1} \cdots \cdots X_{M N}\right)=\mathrm{E}\left(X_{1}\right) \cdots \cdot \mathrm{E}\left(X_{M N}\right)$. (Не забудьте определить независимость для более чем 2 случайных величин.) Покажите, что каждое $\mathrm{E}\left(X_{k}\right)$ не зависит от размера решетки, следовательно $\mathrm{E}\left(X_{k}\right)=1 / 3$ по Задаче 21 . 27. Ответ: бесконечное. Это следует из Задачи 26. 28. Для 4 -мерной решетки, при $c=1+\sqrt{2}$ должен быть виден сильный фазовый переход $[C,(16.37)$ и карт. 9.1]. 29. а)-с) Ответ: 4л. Указание: вычислите сумму углов граней двумя способами. 30. Ответ: нет. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-132.jpg?height=83&width=1884&top_left_y=518&top_left_x=86) действительные числа $a, b, c, d$ даны из равенств $$ \begin{aligned} & f(\overrightarrow{O R})=a \overrightarrow{O R}+c \overrightarrow{O G} \\ & f(\overrightarrow{O G})=b \overrightarrow{O R}+d \overrightarrow{O G} \end{aligned} $$ Тогда математическое ожидание случайной перестановки определено буквально той же формулой, что и математическое ожидание случайной величины, но сложение и умножение понимаются как операции над матрицами. Обратите внимание, что такое ожидание есть $2 \times 2$ матрица, а не перестановка. b) Omвет: $\mathrm{E}\left(W^{\prime \prime}\right)=2\left(\frac{7}{16}\right)^{M N}$. Указание. Рассуждайте аналогично решению Задачи 26 . Для начала вычислите $\mathrm{E}(U(A B C D))=\left(\begin{array}{cc}7 / 16 & 0 \\ 0 & 7 / 16\end{array}\right)$. Затем покажите, что случайные перестановки $U(A B C D)$ независимы для различных $1 \times 1$ квадратов $A B C D$, следовательно, $\mathrm{E}\left(X_{1} \circ \cdots \circ X_{M N}\right)=\mathrm{E}\left(X_{1}\right) \ldots \mathrm{E}\left(X_{M N}\right)=$ ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-132.jpg?height=80&width=1899&top_left_y=1136&top_left_x=84) $\operatorname{Tr}\left(\mathrm{E}\left(W^{\prime \prime}\right)\right)$. 35. Oтвет: $\mathrm{E}\left(W^{\prime \prime}\right)=\left(\frac{17}{23}\right)^{M N}+2 \cdot\left(\frac{14}{23}\right)^{M N}$. Указание. Рассуждайте аналогично решению задачи 33.b. 36. Предответ: $\mathrm{E}\left(W^{\prime}\right) \simeq$ const $/ c^{N^{2}}$. ## Благодарности Авторы благодарны А. Штерну и А. Смирнову за помощь в проверке работ школьников и С. Галкину, А. Канелю, А. Семенову, А. Скопенкову за ценные замечания. ## Список литературы [KZhP] Колмогоров А. Н., Журбенко И. Г., Прохоров А.В. Введение в теорию вероятностей. (Библиотечка «Квант»; вып. 135). М.: МЦНМО, 2015. [SSU] Скопенков М., Смыкалов В., Устинов А., Случайные блуждания и электрические цепи // Математические просвещение, 3-я серия, выпуск 16. М.: МЦНМО, 2012. С. 25-47. Исправлено в: А. Юрьвв, Случайные блуждания возвращаются // Математические просвещение, 3-я серия, выпуск 20. М.: МЦНМО, 2016. С. 243-246. [Shen] Шень А. Вероятность: примеры и задачи. М.: МЦНМО, 2008. [EMZ] Элементы математики в задачах: от кружков и олимпиад - к профессии. / Под общ. ред.: М. Б. Скопенков, А. Б. Скопенков, А. А. Заславский. М. : МЦНМО, 2017 (в печати). [C] M. Creutz, Quarks, Gluons and Lattices, Cambridge Univ. Press, 1983 - Science - 169 pp. [R] H. Rothe, Lattice Gauge Theories: An Introduction, World Scientific, $2005-350$ p. # Quarks game A. Balakin, Yu. Petrova, M. Skopenkov 1. The plane is divided into cells in the form of regular hexagons. A King can move to any neighbor cell, see the picture. Moves up, right-down and left-down will be called positive and the rest - negative. The King made several moves and returned to the original cell. What are possible values for the difference between the number of positive and negative moves? This simple mathematical problem has fundamental physical consequence: the charge of any free particle is a multiple of the electron charge. Want to see how school mathematics helps to understand particle physics? Read on! Knowledge of physics is not required, we are going to give all the necessary information. Let us start with an informal introduction; mathematical definitions appear after Problem 9. ## Elementary particles and where they live ${ }^{1}$ ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-133.jpg?height=320&width=302&top_left_y=722&top_left_x=1668) Substance consists of atoms, atoms consist of nuclea and electrons, nuclea consist of protons and neutrons. All that are particles of substance. Particles can be either elementary or composed, i.e., consisting of other particles. E.g., a proton is composed from two up-quarks and one downquark, while the quarks are elementary; see the figure. For our purposes it is not important what are quarks; just imagine sticky balls glued together to form other particles. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-133.jpg?height=200&width=202&top_left_y=1168&top_left_x=1738) If a particle is separated from any other particles (i.e., moves independently on them) in -protonan experiment, then the particle is called free. E.g., normally a neutron is contained in a nucleus, but it can be made free in nuclear reactions. This collection of problems is about confinement of quarks (main problems are 11, 27, 33.b): ## Quarks cannot be free. What this actually means is absolutely paradoxical: quarks exist but cannot be observed in an experiment in principle. We are going to discuss both reasons and consequences of this (so far unproved) assertion ${ }^{2}$. Each particle has a charge. E.g., an electron has charge -1 (this is how the measurement units are chosen), and an up-quark has charge $+2 / 3$. The charge of a composed particle equals to the sum of charges of the contents. E.g., a proton has charge +1 . Each particle has an antiparticle having opposite charge. E.g., an up-antiquark has charge $-2 / 3$. The list of just a few known quarks and free composed particles is given in the table. | Particle | | | | Antiparticle | | | | | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | | Name | Notation | Content | Charge | Name | Notation | Content | Charge | | Elementary particles: quarks (to the left) and antiquarks (to the right) | | | | | | | | | up-quark | $u$ | - | $+2 / 3$ | up-antiquark | $\bar{u}$ | 5 | $-2 / 3$ | | down-quark | $d$ | - | $-1 / 3$ | down-antiquark | $\bar{d}$ | - | $+1 / 3$ | | Composite particles from a quark and an antiquark | | | | | | | | | pion | $\pi^{+}$ | $u \bar{d}$ | +1 | pion | $\pi^{-}$ | $\bar{u} d$ | -1 | | Composite particles from 3 quarks (to the left) or 3 antiquarks (to the right) | | | | | | | | | proton
neutron | $p$
$n$ | $u u d$
$?$ | +1
0 | antiproton
antineutron | $\bar{p}$
$\bar{n}$ | $\bar{u} \bar{u} \bar{d}$
$?$ | $?$
$?$ | 2. Complete the entries with a questionmark in the table.[^12] Each particle has also a color (it has nothing to do with light). It is not a number, but a vector in the plane. Hereafter we assume that $R B^{\prime} G R^{\prime} B G^{\prime}$ is a regular hexagon with the center at the origin $O$. A quark can have one of the colors $\overrightarrow{O R}, \overrightarrow{O G}, \overrightarrow{O B}$ ("red", "green", "blue"). Thus there actually 3 distinct up-quarks and 3 down-quarks. The color of an antiparticle is opposite to the color of a particle. E.g., the antiparticle ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-134.jpg?height=245&width=297&top_left_y=69&top_left_x=1688) of a red up-quark has color $\overrightarrow{O R}$ ("cyan"). The color of a composed particle is the vector sum of colors of the components. E.g., a proton consisting of a red up-quark, a green up-quark, and a blue down-quark, has vanishing color. Actually the reason for confinement of quarks is confinement of color: Particles with nonvanishing color cannot be free. As a warm-up, let us discuss basic consequences of confinement of color. 3. a) For which $q$ and $p$ satisfying $p+q \leqslant 3$, there is a particle with vanishing color consisting of $q$ quarks and $p$ antiquarks ${ }^{3}$ ? b) Prove that each particle with vanishing color consisting of quarks and antiquarks from the above table has integer charge. Now start the way to understanding the reasons of confinement of color. We shall need a theory describing interaction of quarks. All known interactions except gravitation are described by gauge theory. This concerns not only interaction of quarks but also everyday-observed effects like magnetic interaction of conductors with currents. As we shall see now, the idea of gauge theory is very simple. ## Toy model of gauge theory Several cities are connected by roads in the shape of an $M \times N$ grid; see the figure. Each city has its own type of goods (in unlimited quantity). E.g., city $A$ has apples and city $B$ has bananas. For two neighboring cities $A$ and $B$ an exchange rate $U(A B)>0$ is fixed, e.g., 2 bananas for an apple. The rate is symmetric, i.e., $U(B A)=U(A B)^{-1}$ : one gets back an apple for 2 bananas. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-134.jpg?height=588&width=1600&top_left_y=1351&top_left_x=228) A cunning citizen can travel and exchange along a $1 \times 1$ square $A B C D$ to multiply his initial amount of goods by a factor of $U(A B) U(B C) U(C D) U(D A)$. E.g., in the figure to the left the factor is $2 \cdot 1 \cdot 4 \cdot 1=8$. Denote $U(A B C D):=U(A B) U(B C) U(C D) U(D A)$. We shall see that the logarithms of these factors are more convenient quantities. All we need to know about logarithms is the following definition. If $y=2^{x}$ for some real numbers $x$ and $y$, then $x$ is called the logarithm of $y$ and is denoted by $x=\log _{2} y$. E.g., $\log _{2} 2=1, \log _{2} \sqrt{2}=\frac{1}{2}, \log _{2} 1=0, \log _{2} \frac{1}{2}=-1$. In particular, a trip along the square $A B C D$ gives profit in one of the directions, if $\log _{2} U(A B C D) \neq 0$. The total speculation profit is measured by the sum $S[U]$ of the values $\log _{2}^{2} U(A B C D)$ over all $1 \times 1$ squares $A B C D$ (say, bypassed counterclockwise). E.g., in the figure to the left $$ S[U]=\log _{2}^{2} U(A B C D)+\log _{2}^{2} U(D C F E)=\left(\log _{2} 8\right)^{2}+\left(\log _{2} \frac{1}{2}\right)^{2}=10 $$ You are the king, who can set exchange rates except those on the boundary of the grid. You set them to minimize the quantity $S[U]$. The resulting rates are called optimal. 4. Clean up the kingdom in the left figure, i.e., find the number $x$ for which the total speculation profit in the middle figure is minimal.[^13] Optimal exchange rates can be found approximately on the computer, scrolling with a small step all possible rates from a certain interval and comparing the corresponding values of $S[U]$. 5. Do it for the kingdoms in the figures in the middle and to the right. Is there any faster algorithm? Notice that the change of variables $x(A B):=\log _{2} U(A B)$ simplifies the expression for the profit a lot: $$ S[x]:=\sum_{\substack{\text { all } 1 \times 1 \text { squares } A B C D \\ \text { bypassed counterclockwise }}}(x(A B)+x(B C)+x(C D)+x(D A))^{2} $$ For each $1 \times 1$ square $A B C D$ denote $x(A B C D):=x(A B)+x(B C)+x(C D)+x(D A)$. E.g., in the figure to the left, $x(A B C D)=3$. Denote by $W$ the factor multiplying the initial amount of goods for a counterclockwise travel around the whole boundary. E.g., in the figure to the left, $W=4$. 6. Assume that the fixed rates at the boundary are as in the figure, i.e., $$ U(A B)= \begin{cases}2, & \text { if } A B \text { is on the southern or northern border of the grid } \\ \text { and is directed counterclockwise along the boundary; } \\ 1, & \text { if } A B \text { is on the eastern or western border of the grid. }\end{cases} $$ Clean up the kingdom for the following particular grid sizes, i.e., complete the table ${ }^{4}$ : | Grid | $1 \times 2$ | $1 \times 3$ | $1 \times N$ | $2 \times 2$ | | :--- | :---: | :---: | :---: | :---: | | The value $W$ | 4 | | | | | The minimal value for $S[U]$ | | | | | | Optimal rates $U(A B)$ for all roads $A B$ | | | | | | Values $x(A B)$ for all roads $A B$ | | | | | | Values $x(A B C D)$ for all $1 \times 1$ squares | | | | | 7. a) For which values of the boundary rates in an $M \times N$ grid one can achieve $S[U]=0$ ? b)* Assume that $S[U]=0$. Can the citizen get a profit by moving along a closed path? c) For which values of $M$ and $N$ the optimal rates are uniquely determined by the boundary rates? d) How $M, N, W$ and the minimal speculation profit are related? e) Which of the grids $-8 \times 8$ or $7 \times 9$ - has smaller speculation profit, if $W$ is the same? 8. Now we are allowed to change boundary rates. What is the minimal possible value of the expression $$ \frac{1}{2} S[U]-j \log _{2} W $$ that we can obtain for a fixed real number $j$ on the $M \times N$ grid? ## Physical interpretation Our toy model allows to describe roughly magnetic interaction of conductors with currents. Let $j$ be a real number. The grid boundary is a frame with a current strength $j$. Current in the frame creates a magnetic field. The value $x(A B C D)$ at a $1 \times 1$ square $A B C D$ represents the magnetic flux through the square, the sum $S[x]$ represents the magnetic field energy, and $\frac{1}{2} S[U]-j \log _{2} W$ represents the energy of the whole system. Each system tries to minimize its energy. Thus the values $x(A B)$ are set to minimize the expression $\frac{1}{2} S[U]-j \log _{2} W$. Beware: the model is very rough! If the positions of the conductors are not fixed, i.e., the grid size is not fixed, then the system tries to move the conductors to decrease the energy. This means attraction or repulsion of conductors. Movement of conductors is understood as change of the number of cells, but not the sizes of the cells themselves. 9. a) Does a frame with current tend to reduce or increase its area? b) Do parallel conductors with opposite currents magnetically attract or repulse? c)* And if the current directions are the same? Hint. Close the conductors far away to obtain a frame (for b)) or two frames (for c)).[^14] The quark interaction is described by quantum gauge theory. The main difference is that the exchange rates are random and not necessarily positive. For simplicity let us assume that the exchange rates take only two values +1 and -1 , shortly: "+" and "-". Bypassing the boundary of the square changes the amount of goods, if there is an odd number of signs " - " on its boundary. Now it is time for precise mathematical definitions and theorems. First we state the modified model, then a pair of problems, and only after that introduce all the required definitions little by little. ## Toy model for quantum gauge theory To each road $A B$ of the $1 \times N$ grid in the figure, randomly assign a sign " + " or "-" (so that all possible assignments of signs have equal probability; see the definition of probability before Problem 12). For each assignment $U$ let us denote by $S^{\prime}[U]$ the number of squares $1 \times 1$ having an odd number of signs "-" on the sides. E.g., $S^{\prime}[U]=2$ in the figure. The expectation of the ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-136.jpg?height=285&width=148&top_left_y=420&top_left_x=1822) random variable $S^{\prime}[U]$ is called the energy $E(N)$ of electromagnetic interaction between a quark and an antiquark at distance $N$. (See the definitions of a random variable and its expectation before and after Problem 17.) Remark. The latter term is unsplittable; the words 'quark', 'antiquark', 'electromagnetic interaction' have no separate formal meaning. Beware: the model is very rough! 10. Find the energy $E(N)$ for the grids: $1 \times 1 ; 1 \times 2 ; 1 \times 3 ; 1 \times N$; a) approximately using a computer simulation; b) exactly. 11. (Quarks confinement in 1-dimensional space.) Does there exist a number $E_{0}$ such that $E(N) \leqslant E_{0}$ for all $N$ ? (The same informally: is the amount of energy required to move a quark and an antiquark far away from each other finite or infinite?) ## Classical probability ${ }^{5}$ It is useful to start learning probability theory on the "physical" level of rigor as in books [Shen], [KZhP]. Here we give "mathematical" definitions from the beginning. However many problems are stated using "practical" language and we show how to formalize some of them. The formalization of the remaining problems is left to the reader. Such formalization is the first step of the solution, on which the answer can depend. Consider an experiment that has $m$ equally probable outcomes, such as rolling a dice, drawing a card out of a deck, etc. If the event in question (for example, the fall of the six, drawing an Ace, etc.) occurs in $a$ outcomes, then the probability of the event is said to be $p=a / \mathrm{m}$. This explanation is useful for beginners, but it is not a mathematically rigorous definition. Here is the mathematical definition. The probability of a subset $A$ of a finite set $M$ is the following number $$ P(A)=P_{M}(A):=|A| /|M| $$ Hereafter, unless otherwise stated, the set $M$ is fixed and omitted from the notation. Then the probability is defined for all its subsets. They are often called events. 12. From the deck of 52 cards one card is drawn. Find the probability that it is (a) of black suit; (b) an Ace; (c) with a picture; (d) Queen of spades; (e) a King or a diamond. E.g. in the problem 12 (c) the set $M$ ("the set of all possible events") coincides with the set of all cards in the deck, the set $A$ ("the set of events that we consider") is the set of all cards with pictures. This is the way to formalize this problem (and many other probabilistic problems). 13. A coin is tossed 3 times. Find the probability of getting (a) three Heads; (b) two Heads and one Tail.[^15] To solve some of the above-stated problems, the following rules are useful. 14. (a) Addition rule. Let $A \cap B=\emptyset$. Express $P(A \cup B)$ in terms of $P(A)$ and $P(B)$. (b) Express the probability $P(A \cup B)$ in terms of $P(A), P(B)$ and $P(A \cap B)$. (c) Multiplication Rule. Express the probability $P_{M \times N}(A \times B)$ in terms of $P_{M}(A)$ and $P_{N}(B)$. 15. a) Signs " + " and " - " are randomly assigned to the 4 sides of a square. Find the probability that the number of signs " - " is odd. b) What is the probability that for the $1 \times 2$ grid we have $S^{\prime}[U]=0$ ? $S^{\prime}[U]=1$ ? $S^{\prime}[U]=2$ ? The following problem leads to the notions of a random variable and the expectation. 16. a) I suggest you the following game. You pay 2 candies, then a die is rolled, and you get as many candies as there occur points on the die. Is the game profitable for you? b) Now the rules are the same except that you pay 100 candies, when 1 point occurs. (You have enough candies to pay.) Is the game profitable for you? c) A bank promises you a reliable profit absolutely for free. You put your 8 candies into the bank, then a die is rolled. If 2,3 , or 4 points occur, then you get back your contibution of 8 candies plus 1 candy in addition. If 5 or 6 points occur ("market growth"), then you get even plus 2 candies in addition to your contribution. But if 1 point occurs ("crysis"), then you loose the whole your contribution ("We are sorry but we cannot do anything for you - that's crysis.") Is the game profitable for you? A real-valued function $X$ defined on the set $M$ is called a random variable. The set of pairs $\left(x_{i}, p_{i}\right), i=$ $1,2, \ldots$, where $\left\{x_{1}, x_{2}, \ldots\right\}$ is the set of all possible values of the random variable $X$, and $p_{i}=P(\{m \in$ $\left.\left.M: X(m)=x_{i}\right\}\right), i=1,2, \ldots$, are the corresponding probabilities, is called the distribution of the random variable $X$. Further, we denote by $X=x_{i}$ the event $\left\{m \in M: X(m)=x_{i}\right\}$. 17. Find the distribution of the random variable $S^{\prime}[U]$ for the grid $1 \times 2$. The mathematical expectation (or mean-value) of the random variable $X$ is the sum $$ \mathrm{E}(X):=\sum x_{i} p_{i}=x_{1} P\left(X=x_{1}\right)+x_{2} P\left(X=x_{2}\right)+\ldots $$ 18. Find the expectation of the random variable $S^{\prime}[U]$ for the grid $1 \times 2$. Write +1 instead of " + " and -1 instead of " - ". Let the random variable $X_{k}$ be the product of all numbers on the sides of the $k$-th square from the top in the $1 \times N$ grid and the random variable $W^{\prime}$ be the product of all numbers on the boundary of the grid. (Similarly we define $W^{\prime}$ for the $M \times N$ grid.) 19. Express $S^{\prime}[U]$ and $W^{\prime}$ in terms of $X_{1}$ and $X_{2}$ for the grid $1 \times 2$. Find $\mathrm{E}\left(X_{1}\right), \mathrm{E}\left(X_{2}\right)$ and $\mathrm{E}\left(W^{\prime}\right)$. 20. (a) Prove that the expectation of the random variable $X$ defined on $M$ is equal to $\sum_{m \in M} X(m) /|M|$. (b) Prove that if $\mathrm{E}(X) \leqslant x$, then there exists $m \in M: X(m) \leqslant x$. (c) A random variable $X$ takes the same value $\mu$ for all $m \in M$. Find $\mathrm{E}(X)$. d) Let $a, b$ be real numbers and $X, Y$ be random variables. Is the following equality always true $\mathrm{E}(a X+b Y)=a \mathrm{E}(X)+b \mathrm{E}(Y) ?$ (e) And what about $\mathrm{E}(X Y)=\mathrm{E}(X) \mathrm{E}(Y)$ ? Now let us proceed to a more accurate model of quark interaction (further we work in this model). ## Solid model for quantum gauge theory Fix a number $c>1$ called the interaction constant. To each road $A B$ of the $M \times N$ grid, randomly assign a sign " + " or " - " so that an assignment $U$ has the probability $$ P[U]=\frac{c^{-S^{\prime}[U]}}{\sum_{\text {all the assignments } V} c^{-S^{\prime}[V]}} $$ The energy of the electromagnetic interaction between a quark and an antiquark at distance $N$ is $$ E_{M, c}(N)=-\frac{1}{M} \log _{2} \mathrm{E}\left(W^{\prime}\right) $$ Our aim is to find it. (Here $\mathrm{E}\left(W^{\prime}\right)$ is the expectation of the random variable $W^{\prime}$, defined before the problem 18. We do not know any easy explanation of the formula.) ## Statistical probability The following more general definition of the probability is used in this model. Consider a set $M$ and for each $m \in M$ assign a non-negative number $P(m)$, so that the sum of all such numbers is equal to 1 . Then the probability of the event $A$ is the sum of all $P(m)$ over all $m \in A$. 21. a) Find the probability of each assignment of signs for the $1 \times 1$ grid. Compute $\mathrm{E}\left(W^{\prime}\right)$ and $E_{1,2}(1)$. b) The same, if one considers only the assignments such that all roads except the top one have sign "+". 22. How the probability of an assignment changes, if the signs of all roads coming from a city are changed? The following definition generalizes the multiplication rule from 14 (c). First, assume that the probabilities $P(m)$ are the same for all elements $m \in M$. Two subsets (or events) $A$ and $B \neq \emptyset$ of the finite set $M$ are called independent, if the probability of the set $A \cap B$ in $B$ is equal to the probability of the set $A$ in $M$. Let us give a symmetric reformulation, which works for $B=\emptyset$ and for more general definitions of the probability, when not all probabilites $P(m)$ are equal. The subsets $A$ and $B$ of the finite set $M$ are called independent, if $$ P(A \cap B)=P(A) \cdot P(B) $$ The main example of the independent subsets is the following. In the set of all cells of the chessboard the subset of cells in first three rows and the subset of cells in last four columns are independent. 23. Are the following subsets independent, if all elements of $M$ have equal probabilities? (a) The subsets $\{1,2\} \subset\{1,2,3,4\}$ and $\{1,3\} \subset\{1,2,3,4\}=M$. (b) The subsets $\{1,2\} \subset\{1,2,3,4,5,6\}$ and $\{1,3\} \subset\{1,2,3,4,5,6\}=M$. Random variables $X$ and $Y$ are called independent, if the events $X=x_{i}$ and $Y=y_{j}$ are independent for any $x_{i}, y_{j}$, i. e. $$ P\left(\left\{m \in M: X(m)=x_{i} \text { and } Y(m)=y_{j}\right\}\right)=P\left(X=x_{i}\right) P\left(Y=y_{i}\right) $$ 24. Are the random variables $X_{1}$ and $X_{2}$ independent in the $1 \times 2$ grid (see the definition before Pr. 19)? 25. Prove that if the random variables $X$ and $Y$ are independent, then the mathematical expectation of their product is the product of their mathematical expectations: $\mathrm{E}(X Y)=\mathrm{E}(X) \mathrm{E}(Y)$. 26. (Wilson's area law) Let $c=2, M$ and $N$ are arbitrary. Compute $\mathrm{E}\left(W^{\prime}\right)$ and the energy $E_{M, 2}(N)$. For which $M_{1}, N_{1}, M_{2}, N_{2}$ the $M_{1} \times N_{1}$ grid has smaller $\mathrm{E}\left(W^{\prime}\right)$ than the $M_{2} \times N_{2}$ grid? 27. (Quarks confinement in 2-dimensional space.) Is the amount of energy required to move a quark and an antiquark far away from each other finite or infinite in 2-dimentional space? 28. Investigate 3-dimensional grid $N \times N \times N$ and 4-dimensional grid $N \times N \times N \times N$ experimentally by a numeric simulation for various $c \in[2 ; 3]$. For them, $S^{\prime}[U]$ is still defined as the number of $1 \times 1$ squares with an odd number of signs " - ", and $W^{\prime}$ is defined as the product of signs along the boundary of one of the 2-dimensional faces. Is the amount of energy required to move a quark and an antiquark far away from each other in 3- and 4-dimentional space finite or infinite? 29. Prove that: a) the ratio of the energies $E_{M, c}(N) / E(N)$ does not depend on $N$ b) for $c \rightarrow \infty$ the energy $E_{M, c}(N)$ tends to the minimal possible value of $S^{\prime}[U]$, i.e., to 0 . ## Geometrical view on gauge theory Let us look at our model from the new side (this text is informal and is not used further.) The charges of the particles can be positive or negative. It is a convention which one to consider positive: is just a choice of direction on the real axis. And what happens if the choice of the "axis direction" at each point of the space is done independently? For example, in each city of the $M \times N$ grid, choose the direction in its own way. Then for each road you need to specify how the direction of the axis changes as you move from the city to its neighbor. One can imagine a situation, when bypassing a $1 \times 1$ square, the direction of the axis is changed to the opposite one (as the direction of the perpendicular to the Möbius strip when it is bypassed.) Geometrical view on gauge theory is that such "defective" squares are the magnetic field, and they carry energy. Therefore we computed energy through the number of "defective" squares in our models. Of course, this construction can not be understood literally: when bypassing, a charge sign, of course, is not changed. The "axis direction" is something different from the charge sign. 30. (Gauss-Bonnet theorem) Consider: a) a cube; b) a regular tetrahedron; c)* a convex polytope. Two vectors lying in adjacent faces are called parallel, if they form equal "vertical" angles with the common side of the faces (i.e., they become parallel and looking in the same direction, if one "unfolds" the two faces around the common side; see the figure). Let $f_{1}, f_{2}, \ldots, f_{k}$ be all faces around a vertex $v$ in the natural order. Start with an arbitrary vector $\vec{e}_{1} \subset f_{1}$ and take a vector $\vec{e}_{2} \subset f_{2}$ parallel to $\vec{e}_{1}$, then a vector $\vec{e}_{3} \subset f_{3}$ parallel to $\vec{e}_{2}, \ldots$, and, finally, a vector $\vec{e}_{k+1} \subset f_{1}$ parallel to $\vec{e}_{k}$. Denote by $\phi_{v}$ the oriented angle between $\vec{e}_{k+1}$ and $\vec{e}_{1}$. Find the sum of the angles $\phi_{v}$ over all vertices $v$. Quark color is not a number, but a vector for which three directions are possible. This is why permutations of three colors are put on the roads in the theory of strong (i.e., color) interaction. ## Permutations A permutation of a set is an arrangement of the elements of the set in some order. More strictly, a permutation of the set is a one-to-one correspondence of this set onto itself (i.e. a bijection). The permutation of the set $\left\{a_{1}, a_{2}, \ldots, a_{n}\right\}$, that to an element $a_{k}$ assigns the element $f\left(a_{k}\right)$, is written as follows $$ \left(\begin{array}{cccc} a_{1} & a_{2} & \ldots & a_{n} \\ f\left(a_{1}\right) & f\left(a_{2}\right) & \ldots & f\left(a_{n}\right) \end{array}\right) $$ usually $a_{k}=k$ for all $k=1, \ldots, n$. The composition of permutations $f$ and $g$ is the permutation $f \circ g$ defined by the formula $(f \circ g)(x):=f(g(x))$. 31. Find the compositions: (a) $\left(\begin{array}{lll}1 & 2 & 3 \\ 2 & 1 & 3\end{array}\right) \circ\left(\begin{array}{lll}1 & 2 & 3 \\ 3 & 1 & 2\end{array}\right)$; (b) $\left(\begin{array}{lll}1 & 2 & 3 \\ 2 & 3 & 1\end{array}\right) \circ\left(\begin{array}{lll}1 & 2 & 3 \\ 3 & 1 & 2\end{array}\right)$. 32. Is it true that $x \circ y=y \circ x$ for any pair of permutations $x$ and $y$ ? ## Toy model for non-Abelian gauge theory Fix a number $c>1$. Denote by $S_{3}$ a set of all permutations of the set $\{1,2,3\}$. A fixed point of the permutation $f$ is such $x \in\{1,2,3\}$ that $f(x)=x$. The trace $\operatorname{Tr}(f)$ of a permutation $f$ is the number of its fixed points minus 1 . For example, $\operatorname{Tr}\left(\begin{array}{lll}1 & 2 & 3 \\ 3 & 2 & 1\end{array}\right)=0$. To each road $A B$ of the $M \times N$ grid randomly assign a permutation $U(A B) \in S_{3}$ so that $U(A B) \circ U(B A)$ is the identical permutation, and the probability of an assignment $U$ is equal to $$ P[U]:=\frac{c^{-S^{\prime \prime}[U]}}{\sum_{\text {all the assignments } V} c^{-S^{\prime \prime}[V]}} $$ where $$ S^{\prime \prime}[U]:=\sum_{\substack{\text { all } 1 \times 1 \times 1 \text { squares } A B C D \\ \text { bypassed counterclockwise }}}(2-\operatorname{Tr}(U(A B) \circ U(B C) \circ U(C D) \circ U(D A))) $$ (I.e. the fewer elements remain fixed when bypassing the square, the greater contribution it makes to $S^{\prime \prime}[U]$.) Let $W^{\prime \prime}$ be the trace of the product of all permutations on the boundary sides bypassed counterclockwise and starting from the lower-left corner. The value $-\frac{1}{M} \log _{2} \mathrm{E}\left(W^{\prime \prime}\right)$ is called the energy of strong interaction of a quark and an antiquark at distance $N$. 33. * a) Invent the notion of an expectation of a random permutation from $S_{3}$ having as many properties of the expectation of a random variable as possible. Hint. We lack the operation of adding permutations. It appears by itself, if we recall that 1,2,3 are quark colors, colors are vectors, and permutation is a mapping. b) Compute the expectation of $W^{\prime \prime}$ for $c=2$. Prove that there are $a, b, B \in \mathbb{R}$ such that for all $M, N$ we have $b a^{M N} \leqslant \mathrm{E}\left(W^{\prime \prime}\right) \leqslant B a^{M N}$. 34. (The "essence" of the Millenium problem) Prove the same inequality for the 4-dimensional $N \times N \times$ $N \times N$ grid and a rectangle $M \times N$ on one of its 2-dimensional faces. ## Additional problems These hard problems are given to those who have solved most of the problems above. 35. Solve the analogue of Problem 33.b, if the definition of permutation trace is changed as follows: $$ \operatorname{Tr}^{\prime}(x)= \begin{cases}+3, & \text { if } x \text { is identical, i.e. } x=\left(\begin{array}{lll} 1 & 2 & 3 \\ 1 & 2 & 3 \end{array}\right) \\ 0, & \text { if } x \text { is a cycle of length } 3, \text { i.e. } x=\left(\begin{array}{lll} 1 & 2 & 3 \\ 2 & 3 & 1 \end{array}\right) \text { or } x=\left(\begin{array}{lll} 1 & 2 & 3 \\ 3 & 1 & 2 \end{array}\right) \\ -1, & \text { if } x \text { is a transposition, i.e. } x=\left(\begin{array}{llll} 1 & 2 & 3 \\ 2 & 1 & 3 \end{array}\right), x=\left(\begin{array}{llll} 1 & 2 & 3 \\ 3 & 2 & 1 \end{array}\right), \text { or } x=\left(\begin{array}{lll} 1 & 2 & 3 \\ 1 & 3 & 2 \end{array}\right)\end{cases} $$ and the expression $2-\operatorname{Tr}(\ldots)$ is replaced by $3-\operatorname{Tr}^{\prime}(\ldots)$ in the formula for $S^{\prime \prime}[U]$. 36. (Strong coupling expansion) Consider the solid model on a 3-dimensional grid $N \times N \times N$; see Problem 28. Find an asymptotic form for $\mathrm{E}\left(W^{\prime}\right)$ as $c \rightarrow \infty$, i.e., a function $f_{N}(c)$ such that $\lim _{c \rightarrow \infty} \frac{\mathrm{E}\left(W^{\prime}\right)}{f_{N}(c)}=1$ for fixed $N$. ## Underwater riffs We hope that at least some of our readers have become interested in elementary particles and want to learn more about them. As an epilogue, let us give a few warnings to such readers. In popular science, theory of elementary particles is usually oversimplified. This sequence of problems is not an exception. The toy models introduced here are very rough and should be considered with a grain of salt. Simplicity is their only advantage; if taken too seriously, the models could even give a wrong physical intuition. Only the model called "solid model for quantum gauge theory" is a respectful one truly considered in physical literature. Real understanding of particles theory requires excellent knowledge of both physics and mathematics. The problem called "the essence of the Millenium Problem" is not equivalent to the Millenium Problem in Yang-Mills theory at all. It is not even a particular case. It is rather the most essential - in our very subjective opinion - part of the problem freed from technical details. Although we came to the problem during the discussion of quarks confinement, it is closer to another phenomenon: short range of nuclear forces. We should also remark that nowadays there are almost no mathematical results in lattice gauge theory; what we have is usually just a numeric simulation. Finally, there are "theories of New Physics" which are developed without any objective truth criterion: such theories are supported by neither experimental nor mathematical proofs (and some of them have experimental disproofs). ## Hints, solutions, answers 1. Answer: any integer multiple of 3 . Hint. Let us write the remainders 0,1 and 2 (when dividing by three) into cells as in the figure below. Note that a positive move from any cell adds 1 to the remainder on which the king stands, and a negative move subtracts 1. 2. Answer: | proton
neutron | $p$ | $u u d$ | +1 | antiproton | $\bar{p}$ | $\bar{u} \bar{u} \bar{d}$ | -1 | | :--- | :---: | :---: | :---: | :--- | :---: | :---: | :---: | | antineutron | $\bar{n}$ | $\bar{u} \bar{d} \bar{d}$ | 0 | | | | | ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-140.jpg?height=329&width=417&top_left_y=1863&top_left_x=1568) 3. a) Answer: $(q, p)=(1,1),(3,0)$ or $(0,3)$. Hint. This problem is solved by a brute-force exhaustion, let us give an example of reasoning for one case. Why $q$ and $p$ can not be equal to 2 and 0 respectively: if the sum of two vectors is 0 , then they are opposite, but two quarks can not have opposite colors. b) Let $p \leqslant q$. Note that the sum of the charges of an arbitrary quark and an antiquark is an integer. Hence, we can consider only the sum of the charges of $q-p$ quarks. Also note that the sum of the charges of any three quarks is an integer. It follows from Problem 1 that for a particle with zero color $q-p$ is a multiple of 3 , that is, $q-p$ quarks can be divided into triples. 4. Answer: $x=0$. Hint: $$ S[U]=\log _{2}^{2}\left(2 \cdot 1 \cdot \frac{1}{2^{x}} \cdot 1\right)+\log _{2}^{2}\left(2^{x} \cdot 1 \cdot 2 \cdot 1\right)=(1-x)^{2}+(1+x)^{2}=2+2 x^{2} $$ The minimum is reached at $x=0$. 6. Answer. In the table below, in the third and fourth lines in the first three columns there are values for all non-boundary roads (from top down). It is assumed that all these roads are directed from left to right. In the same lines in the last column it is considered that all non-boundary roads are directed from the center to the borders, and by $x$ we denote an arbitrary real number (that is, there is a solution for each $x$ ). If for all roads (or squares) the value is the same, then only one number is placed in the corresponding cell. | Grid | $1 \times 2$ | $1 \times 3$ | $1 \times N$ | $2 \times 2$ | | :--- | :---: | :---: | :---: | :---: | | The value $W$ | 4 | 4 | 4 | 16 | | Minimal $S[U]$ | 2 | $\frac{4}{3}$ | $\frac{4}{N}$ | 4 | | Optimal rates | 1 | $2^{-1 / 3}, 2^{1 / 3}$ | $2^{-1+\frac{2}{N}}, 2^{-1+\frac{4}{N}}, \ldots 2^{-1+\frac{2 N-2}{N}}$ | $2^{x}$ | | Values $x(A B)$ | 0 | $-\frac{1}{3}, \frac{1}{3}$ | $-1+\frac{2}{N},-1+\frac{4}{N}, \ldots-1+\frac{2 N-2}{N}$ | $x$ | | Values $x(A B C D)$ | 1 | $\frac{2}{3}$ | $\frac{2}{N}$ | 1 | Hint. In the solution of this problem, as well as some subsequent ones, it is very useful the inequality of arithmetic and quadratic means ( $A M-Q M$ inequality), which is formulated as follows: for all real numbers $x_{1}, x_{2}, \ldots x_{n}$ the inequality holds: $$ \sqrt{\frac{x_{1}^{2}+x_{2}^{2}+\ldots+x_{n}^{2}}{n}} \geqslant \frac{x_{1}+x_{2}+\ldots+x_{n}}{n} $$ and equality is achieved if and only if $x_{1}=x_{2}=\ldots=x_{n} \geqslant 0$. For the $1 \times N$ grid one should take the values of $x_{1}, x_{2}, \ldots x_{n}$ equal to $x(A B C D)$ for all squares $A B C D$ (enumerated from top down). For the $2 \times 2$ grid denote by $2^{x}, 2^{y}, 2^{z}$ and $2^{t}$ the exchange rates for non-boundary roads directed from the center to the boundary and going counterclockwise. Then $$ \begin{aligned} S[U]= & (x+1-y)^{2}+(-z+y+1)^{2}+(1-t+z)^{2}+(1-x+t)^{2}= \\ = & 2 x^{2}+2 y^{2}+2 z^{2}+2 t^{2}+4-2 x y-2 y z-2 z t-2 t x= \\ & =(x-y)^{2}+(y-z)^{2}+(z-t)^{2}+(t-x)^{2}+4 \geqslant 4 \end{aligned} $$ Remark. There are other solutions, some of them rely on the existence of optimal rates. In this case, the latter must be proved, and this is rather difficult, see [SSU, The principle of achieving the smallest value on p.30]. 7a. (E. Pavlov) Answer: $S[U]=0$ can be achieved if and only if $W=1$. Indeed, $S[U]=0$ if and only if for each $1 \times 1$ square $A B C D$ we have $U(A B C D)=1$. Consider the product $$ \prod=\prod_{\substack{\text { all } 1 \times 1 \\ \text { squares } A B C D}} U(A B C D) $$ In this product each non-boundary road appears twice, once as $U(A B)$ and once as $U(B A)$, cancelling each other. The boundary roads appear once with a counterclockwise orientation. This means that $\Pi=W$. Since every term in $\prod$ is just 1 it follows that $W=\prod=1$ is a necessary condition. Now assume that $W=1$. Let us show that there exist rates $U(A B)$ such that $S[U]=0$. First, we denote by $a_{0}, a_{1}, \cdots, a_{2 M+2 N-1}$ the vertices on the boundary, starting from the bottom left vertex. Next, we introduce the function $P(A)$ on the vertices of the grid by: $$ P(A)= \begin{cases}1, & \text { if } A \text { is not on the boundary or the bottom left vertex. } \\ \prod_{k=0}^{n-1} U\left(a_{k} a_{k-1}\right), & \text { if } A \text { is the vertex } a_{\mathrm{n}} \text { on the boundary. }\end{cases} $$ Finally, we extend $U(A B)$ to the non-boundary roads by the formula $U(A B)=\frac{P(B)}{P(A)}$. For each square $A B C D$, we have $$ \log _{2}^{2}(U(A B) U(B C) U(C D) U(D A))=\log _{2}^{2}\left(\frac{P(B)}{P(A)} \frac{P(C)}{P(B)} \frac{P(D)}{P(C)} \frac{P(A)}{P(D)}\right)=\log _{2}^{2}(1)=0 $$ This means that $S[U]=0$, q.e.d. 7b**. It suffices to construct a function $P(A)$ on the set of cities such that $U(A B)=\frac{P(B)}{P(A)}$ for each road $A B$. For a city $A$, set $P(A)$ to be the product of the rates for the $\Gamma$-shaped route starting from the bottom-left city and ending at $A$. Then the condition $U(A B C D)=1$ for each $1 \times 1$ square $A B C D$ implies the condition $U(A B)=\frac{P(B)}{P(A)}$. 7c. For $M, N>1$ the rates are not unique. For an interior city, one can change the units of measurements, e.g., exchange dozens of apples instead of single ones. Such gauge transformation multiplies the rates for all the roads starting from the city by the same value but preserves $S[U]$. 7d. Answer: $2^{\sqrt{M \cdot N \cdot S_{\min }}}=W$, where $S_{\min }$ is the minimal value of $S[U]$. Hint. By AM-QM inequality we have: $$ S[U]=\sum_{\substack{\text { all } 1 \times 1 \\ \text { squares } A B C D}} x(A B C D)^{2} \geqslant \frac{1}{M N}\left(\sum_{\substack{\text { all } 1 \times 1 \\ \text { squares } A B C D}} x(A B C D)\right)^{2}=\frac{\log _{2}^{2} W}{M N} $$ It remains to prove that it is always possible to establish such exchange rates in the kingdom so that equality is achieved, that is, the values of $x(A B C D)$ are equal to $\frac{\log _{2} W}{M N}$ for all $1 \times 1$ squares. For a rectangle $1 \times N$ one can successively determine the rates on the undefined roads, bypassing them from the top down. These rates are uniquely determined, and for the last square $x(A B C D)$ is equal to $\frac{\log _{2} W}{M N}$, because the sum of these quantities is fixed for all squares. For the rectangle $M \times N$, one can establish an exchange rate equal to 1 on the roads drawn in bold lines (see the figure). And then establish the rates on the remaining roads as written above. 7e. Answer: $8 \times 8$. This follows from the formula in the problem $7 \mathrm{~d}$. 8. Answer: $-M N j^{2} / 2$. Hint. Denote $w=\log _{2} W$. For fixed $w$, the minimum of $S[U]$ equals $\frac{w^{2}}{M N}$ by Problem 7.d. It remains to find the minimum of the quadratic polynomial $\frac{w^{2}}{2 M N}-j w$. 9a. Answer: it tends to increase the area. The first step in solving this problem (and many others) is mathematical formalization. Problem a) is formalized as follows: Consider the minimal value of $\frac{1}{2} S[x]-j \log _{2} W$ on the $M \times N$ grid as a function of $M N$ with fixed $j$. Is this function increasing or decreasing? After such formalization, the solution follows from Problem 8. The work of any electric motor and generator is based on this fact. 9b. Answer: opposite currents repulse. Mathematical formalization: consider the minimal value of $\frac{1}{2} S[x]-j \log _{2} W$ on the $M \times N$ grid as a function of $N$ with fixed $M$ and $j$. Is this function increasing or decreasing? 9c*. Answer: currents with the same directions attract. Mathematical formalization: consider the minimal value of $\frac{1}{2} S[x]-j \log _{2} W$ on the $M \times(L-N)$ grid as a function of $N$ with fixed $L, M$, and $j$. Is this function increasing or decreasing? 10b. Answer: $\frac{1}{2}, 1, \frac{3}{2}, \frac{N}{2}$. Take an arbitrary square of the $1 \times N$ grid and consider a random variable equal to 0 if there is an even number of signs " - " on its sides, and 1 , if odd. By a not very long brute-force search it is easy to check that 0 and 1 are equally probable, hence, the mathematical expectation of this variable is $\frac{1}{2}$. Note that the value $S^{\prime}[U]$ is equal to the sum of such random variables over all $1 \times 1$ squares. Since mathematical expectation of sum is equal to sum of mathematical expectations, we obtain $E(N)=\frac{N}{2}$. 11. Answer: no. Hint. In this model the energy $E(N)=\frac{N}{2}$ tends to infinity as $N \rightarrow \infty$. 12. Answers: a) $\frac{1}{2}$; b) $\frac{1}{13}$; c) $\frac{3}{13}$; d) $\frac{1}{52}$; e) $\frac{15}{52}$. 13. Answers: a) $\frac{1}{8} ;$ b) $\frac{3}{8}$. 14a. Answer: $P(A \cup B)=P(A)+P(B)$. Let $A$ and $B$ be subsets of the set $M$. Since $A$ and $B$ do not intersect, $|A \cup B|=|A|+|B|$. Hence, by definition of probability we get $$ P(A \cup B)=\frac{|A \cup B|}{|M|}=\frac{|A|+|B|}{|M|}=\frac{|A|}{|M|}+\frac{|B|}{|M|}=P(A)+P(B) $$ 14b. Answer: $P(A \cup B)=P(A)+P(B)-P(A \cap B)$. 14c. Answer: $P_{M \times N}(A \times B)=P_{M}(A) P_{N}(B)$. 15. Answers: a) $\frac{1}{2}$; b) $\frac{1}{4}, \frac{1}{2}, \frac{1}{4}$. We give an example of the reasoning for b). We divide into pairs all the arrangements of signs in the following way: two arrangements are in the pair if and only if they differ only by the sign on the upper side. Note that replacing the sign on the upper side changes the parity of the number of "-" signs on the sides of the upper square. Hence, the number of arrangements with $S^{\prime}[U]=1$ equals to the number of all arrangements with values $S^{\prime}[U]=0$ and $S^{\prime}[U]=2$ taken together. That is a half of the total number of arrangements. Similarly, we divide into pairs all the arrangements that give $S^{\prime}[U]=0$ and $S^{\prime}[U]=2$, and differ only in the sign on the central side. The same as above argument gives us that the number of assignments with $S^{\prime}[U]=0$ equals to the number of assignments with $S^{\prime}[U]=2$. 16. Answers: a) yes; b) no; c) no. 17. Answers: $\{(0,1 / 4),(1,1 / 2),(2,1 / 4)\}$. This is actually Problem 15. 18. Answer: 1 . 19. Answer: $S^{\prime}[U]=\left(2-X_{1}-X_{2}\right) / 2, W^{\prime}=X_{1} X_{2}, \mathrm{E}\left(X_{1}\right)=\mathrm{E}\left(X_{2}\right)=\mathrm{E}\left(W^{\prime}\right)=0$. 20. Answers: c) $\mu$; d) yes; e) no. Counterexample to e): $\mathrm{E}\left(X_{1}^{2}\right)=\mathrm{E}(1)=1 \neq 1 / 4=\mathrm{E}\left(X_{1}\right) \mathrm{E}\left(X_{1}\right)$. 21. a) Answer: The probability of an assignment with an even number of signs "-" is $\frac{c}{8(c+1)}$, with an odd $\frac{1}{8(c+1)} ; \mathrm{E}\left(W^{\prime}\right)=\frac{c-1}{c+1}, E_{1,2}(1)=\log _{2} 3$. b) Answer: The probability of the assignment without signs "_" is $\frac{c}{c+1}$, with a sign "-" - $\frac{1}{c+1} ; \mathrm{E}\left(W^{\prime}\right)$ and $E_{1,2}(1)$ are the same as in a). 22. Answer: the probability does not change. 23. Answers: a) yes; b) no. 24. Answer: yes. Hint. The following simplification of the solid model is very useful for this and many other problems. Changing the signs of all the roads coming from one particular city is called a gauge transformation. By Problem 22 this does not change the probability of an assignment. It changes neither the random variables $X_{k}$ nor $W^{\prime}$. For an arbitrary assignment, by a sequence of gauge transformations one can turn all the signs at the solid segments in the figure into pluses. Thus one may assume that only the assignments such that all the solid segments have signs "+" are allowed in the model. Next, the signs of all the other segments are uniquely determined by the values of $X_{k}$ at all the squares. Thus our model is equivalent to the following Bernoulli process: Let the probability of a collection of numbers $X_{1}, \ldots, X_{M N} \in\{+1,-1\}$ be $$ P\left(X_{1}, \ldots, X_{M N}\right)=\frac{c^{\left(M N-X_{1}-\cdots-X_{M N}\right) / 2}}{\sum_{\text {all possible collections }\left(Y_{1}, \ldots, Y_{M N}\right)} c^{\left(M N-Y_{1}-\cdots-Y_{M N}\right) / 2}} $$ Then the energy $E_{M, c}(N)=-\frac{1}{M} \log _{2} \mathrm{E}\left(X_{1} \cdots X_{M N}\right)$. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-143.jpg?height=437&width=548&top_left_y=1529&top_left_x=754) 26. Answer: $\mathrm{E}(W)=\frac{1}{3^{M N}}, E_{M, 2}(N)=N \log _{2} 3$. Hint. Use the simplified model from the solution of Problem 24. Show that $X_{1}, \ldots, X_{M N}$ are independent, hence $\mathrm{E}\left(X_{1} \cdots X_{M N}\right)=\mathrm{E}\left(X_{1}\right) \cdots \cdot \mathrm{E}\left(X_{M N}\right)$. (Do not forget to define independence for more than 2 random variables.) Show that each $\mathrm{E}\left(X_{k}\right)$ does not depend on the size of the grid, hence $\mathrm{E}\left(X_{k}\right)=1 / 3$ by Problem 21 . 27. Answer: infinite. This follows from Problem 26. 28. For the $4 D$-grid, a strong phase transition at $c=1+\sqrt{2}$ should be visible [1, (16.37) and Figure 9.1]. 29. a)-c) Answer: $4 \pi$. Hint: compute the sum of the angles of all the faces in two ways. 30. Answer: no. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-143.jpg?height=77&width=1884&top_left_y=2366&top_left_x=86) numbers $a, b, c, d$ are given by the equations $$ \begin{aligned} & f(\overrightarrow{O R})=a \overrightarrow{O R}+c \overrightarrow{O G} \\ & f(\overrightarrow{O G})=b \overrightarrow{O R}+d \overrightarrow{O G} \end{aligned} $$ Then the expectation of a random permutation is defined literally by the same formula as the expectation of a random variable, but addition and multiplication are understood as operations over matrices. Beware that such expectation is a $2 \times 2$ matrix rather than a permutation. b) Answer: $\mathrm{E}\left(W^{\prime \prime}\right)=2\left(\frac{7}{16}\right)^{M N}$. Hint. Argue analogously to the solution of Problem 26. First compute $\mathrm{E}(U(A B C D))=\left(\begin{array}{cc}7 / 16 & 0 \\ 0 & 7 / 16\end{array}\right)$. Then show that random permutations $U(A B C D)$ are independent for distinct $1 \times 1$ squares $A B C D$, hence $\mathrm{E}\left(X_{1} \circ \cdots \circ X_{M N}\right)=\mathrm{E}\left(X_{1}\right) \ldots \mathrm{E}\left(X_{M N}\right)=\left(\begin{array}{cc}(7 / 16)^{M N} & 0 \\ 0 & (7 / 16)^{M N}\end{array}\right)$. Finally, define the trace of a $2 \times 2$ matrix by $\operatorname{Tr}\left(\begin{array}{ll}a & b \\ c & d\end{array}\right):=a+d$ and show that $\mathrm{E}\left(\operatorname{Tr}\left(W^{\prime \prime}\right)\right)=\operatorname{Tr}\left(\mathrm{E}\left(W^{\prime \prime}\right)\right)$. 35. Answer: $\mathrm{E}\left(W^{\prime \prime}\right)=\left(\frac{17}{23}\right)^{M N}+2 \cdot\left(\frac{14}{23}\right)^{M N}$. Hint. Argue analogously to the solution of Problem 33.b. 36. Semi-answer: $\mathrm{E}\left(W^{\prime}\right) \simeq \mathrm{const} / c^{N^{2}}$. ## Acknowledgements The authors are grateful to A. Stern and A. Smirnov for help with checking the works of the students, and to S. Galkin, A. Kanel, A. Semenov, A. Skopenkov for useful remarks. ## References [1] M. Creutz, Quarks, Gluons and Lattices, Cambridge Univ. Press, 1983 - Science - 169 pp. [2] H. Rothe, Lattice Gauge Theories: An Introduction, World Scientific, 2005 - 350 p. [EMZ] Elements of mathematics in problems. Through circles and olympiads to profession. Ed. by A. Skopenkov, M. Skopenkov, and A. Zaslavskiy. Moscow Center for Continuous Mathematical Education, 2017, in print (in Russian) [KZhP] Kolmogorov A.N., Zhurbenko I.G., Prohorov A.V. Introduction to the Theory of Probability. (Bibliotechka "Kvant"; issue 135). M.: MCCME, 2015. [Shen] Shen A. Probability: examples and problems. M.: MCCME, 2008. [SSU] M. Skopenkov, V. Smykalov, A. Ustinov, Random walks and electric networks, Mat. Prosv. 3rd ser. 16 (2012), 25-47 (in Russian). Minor correction in: A. Yuryev, Random walks return, Mat. Prosv. 3rd ser. 20 (2016), 243-246. # Серии парных теннисных матчей и упаковки продуктов Ю. В. Таранников, И. И. Богданов, К. Г. Куюмжиян,
М. С. Лобанов, А. С. Москалев, А. В. Саускан, С. И. Токарев. ## ЧАСТЬ I
УСЛОВИЯ ЗАДАЧ ## ЗАДАЧИ ДО ПРОМЕЖУТОЧНОГО ФИНИША. ## Задача о серии парных теннисных матчей и ее вариации Задача 1 В теннисном клубе $n$ теннисистов хотят провести серию парных (двое на двое) матчей так, чтобы каждый из теннисистов имел своим противником каждого из остальных ровно в одном матче. На партнеров по паре никаких ограничений нет. При каких $n$ возможна такая серия матчей? Задача 1 допускает также формулировку на теоретико-множественном языке. Пусть $A$ - множество теннисистов. Матчем назовем неупорядоченную совокупность $\left\{\left\{a_{1}, a_{2}\right\}\left\{a_{3}, a_{4}\right\}\right\}$ двух неупорядоченных непересекающихся пар элементов множества $A$. Здесь элементы $a_{1}, a_{2}, a_{3}, a_{4}$ принадлежат $A$ и попарно различны. Расписанием назовем множество матчей. Тогда задача 1 изучает расписания, в которых для любых двух различных элементов $a$ и $b$ из $A$ существует в точности один матч, в котором элементы $a$ и $b$ встречаются в разных парах. Задача 2 В теннисном клубе $n$ теннисистов хотят провести серию парных (двое на двое) матчей так, чтобъ каждый из теннисистов имел своим противником каждого из остальных не более чем в одном матче. На партнеров по паре никаких ограничений нет. Какое наибольшее число матчей можно провести? Задача 2 изучает расписания, в которых для любых двух различных элементов $a$ и $b$ из $A$ существует не более одного матча, в котором элементы $a$ и $b$ встречаются в разных парах. Для выписывания расписания серии матчей можно использовать алгебраическую запись: $$ \left(a_{1}+a_{2}\right)\left(a_{3}+a_{4}\right) $$ Это означает, что пара теннисистов $a_{1}$ и $a_{2}$ играет против пары теннисистов $a_{3}$ и $a_{4}$. Дадим определение ииклического расписания. Пронумеруем игроков числами от 0 до $n-1$. Расписание будем называть «циклическим», если для любого целого $i$ вместе с каждым матчем $\{\{a, b\}\{c, d\}\}$ в расписании будет присутствовать также матч $\left\{\left\{a_{i}, b_{i}\right\}\left\{c_{i}, d_{i}\right\}\right\}$, где $a_{i}=a+i(\bmod n)$. Числа $b_{i}, c_{i}, d_{i}$ определяются аналогично. Для удобства формулировок в последующих задачах этого раздела назовем совершенными все расписания, о которых идет речь в задаче 1 (т. е. в которых для любых двух различных элементов $a$ и $b$ из $A$ существует в точности один матч, в котором элементы $a$ и $b$ встречаются в разных парах, или, говоря о теннисистах, все расписания, в которых каждый из теннисистов имел своим противником каждого из остальных ровно в одном матче). Назовем совершенным каждое такое число $n$, для которого существует хотя бы одно совершенное расписание. Задача 3 Придумать «циклическое» совершенное расписание для любого совершенного $n$. Задача 4* Наложим дополнительное условие: у каждого из теннисистов каждый партнер по паре должен повторяться не более одного раза (т. е. на теоретико-множественном языке любые два элемента находятся вместе в одной паре не более чем в одном матче). Придумайте совершенные расписания с этим дополнительным условием для как можно большего числа различных $n$ (решение задачи для произвольного $n$ является открытой проблемой). Задача 5 Пусть число $n$ - совершенное. В теннисном клубе $n$ теннисистов хотят провести серию парных (двое на двое) матчей. Спрашивается, можно ли составить совершенное расписание так, чтобы партнеров по паре у каждого было больше, чем $\frac{n}{4}$. Задача 6 Наложим такое условие: у каждого из теннисистов каждый партнер по паре должен повторяться ровно один раз, а каждый соперник - ровно два раза. Придумайте расписание с этим условием для $n=5,8,9$. Задача $7^{* *}$ (открытая проблема)При каких $n$ существует расписание с условием из задачи 6? Назовем $m$-матчем неупорядоченную совокупность $\left\{\left\{a_{1}, \ldots, a_{m}\right\}\left\{b_{1}, \ldots, b_{m}\right\}\right\}$ двух неупорядоченных подмножеств (или команд) множества $A$, где все элементы $a_{1}, \ldots, a_{m}, b_{1}, \ldots, b_{m}$ принадлежат $A$ и попарно различны. Тогда 2 -матч является просто матчем. Назовем $m$-расписанием множество $m$-матчей. Тогда 2 -расписание является просто расписанием. Назовем совершенным $m$ расписание, в котором для любых двух различных элементов $a$ и $b$ из $A$ существует в точности один $m$-матч, в котором элементы $a$ и $b$ встречаются в разных командах. Задача 8 Докажите, что если совершенное т-расписание для $n$ теннисистов существует, то $n-1$ делится на $\mathrm{m}^{2}$. Задача 9 При каких $n$ существует совершенное $m$-расписание для $n$ теннисистов, если $m$ четно? Задача 10 Существует ли совершенное 3-расписание для 10 теннисистов? Задача 11**(открытая проблема) При каких $n$ существует совершенное $m$-расписание для $n$ теннисистов, если т нечетно? Задача 12 Внесем (только в рамках этой задачи) следующие изменения в определения. В матче разрешим элементу содержаться одновременно в обеих парах (т. е. разрешим теннисисту одновременно играть за обе пары по обе стороны сетки). От совершенного расписания будем требовать, чтобы для любых двух не обязательно различных элементов а и $b$ из $A$ существовал в точности один матч, в котором элементы а и в встречаются в разных парах (т. е. дополнительно потребуем, чтобы каждый теннисист сыграл сам против себя). Найдите все $n$, при которых возможно такое совершенное расписание. ## Упаковки продуктов Загадка 13 Чем интересно выражение $$ \begin{array}{r} \quad\left(x_{1}+x_{2}\right)\left(x_{3}+x_{4}\right)\left(x_{5}+x_{6}\right)+\left(x_{1}+x_{2}\right)\left(x_{4}+x_{6}\right)\left(x_{8}+x_{9}\right)+\left(x_{1}+x_{2}\right)\left(x_{7}+x_{9}\right)\left(x_{8}+x_{10}\right)+ \\ +\left(x_{1}+x_{3}\right)\left(x_{2}+x_{5}\right)\left(x_{7}+x_{8}\right)+\left(x_{1}+x_{4}\right)\left(x_{5}+x_{7}\right)\left(x_{6}+x_{9}\right)+\left(x_{1}+x_{5}\right)\left(x_{2}+x_{3}\right)\left(x_{9}+x_{10}\right)+ \\ +\left(x_{1}+x_{6}\right)\left(x_{3}+x_{10}\right)\left(x_{4}+x_{8}\right)+\left(x_{1}+x_{7}\right)\left(x_{2}+x_{10}\right)\left(x_{5}+x_{6}\right)+\left(x_{1}+x_{10}\right)\left(x_{2}+x_{7}\right)\left(x_{3}+x_{4}\right)+ \\ +\left(x_{2}+x_{8}\right)\left(x_{3}+x_{7}\right)\left(x_{4}+x_{9}\right)+\left(x_{2}+x_{9}\right)\left(x_{5}+x_{10}\right)\left(x_{6}+x_{8}\right)+\left(x_{3}+x_{5}\right)\left(x_{4}+x_{9}\right)\left(x_{7}+x_{10}\right)+ \\ +\left(x_{3}+x_{5}\right)\left(x_{6}+x_{8}\right)\left(x_{7}+x_{10}\right)+\left(x_{3}+x_{8}\right)\left(x_{4}+x_{6}\right)\left(x_{5}+x_{9}\right)+\left(x_{4}+x_{7}\right)\left(x_{6}+x_{10}\right)\left(x_{8}+x_{9}\right) ? \end{array} $$ Мономом будем называть произведение переменных. Во всех последующих задачах рассматриваются только мономы без повторяющихся переменных. Два монома, отличающиеся лишь порядком переменных, считаются одинаковыми. Задачу о серии парных теннисных матчей можно переформулировать в терминах алгебраических выражений, а именно произведений сумм пар переменных. Сопоставим каждому из $n$ теннисистов одну из переменных $x_{1}, x_{2}, \ldots, x_{n}$. Выражение $\left(x_{i}+x_{j}\right)\left(x_{k}+x_{l}\right)$ будет сопоставлено матчу, в котором пара теннисистов, соответствующих переменным $x_{i}$ и $x_{j}$ встречается с парой теннисистов, соответствующих переменным $x_{k}$ и $x_{l}$. Такие выражения будем называть $(n, 2)$-продуктами ( 2 - потому что в матче участвуют две команды, в дальнейшем этот параметр будет обобщен). Требование, чтобы никакие два теннисиста не встретились в качестве соперников более одного раза означает, что после раскрытия по правилу дистрибутивности скобок в $(n, 2)$-продуктах, соответствующих всем матчам, никакой моном длины 2 не встречается более одного раза. Требование, чтобы любые два теннисиста встретились в качестве соперников в точности один раз означает, что после раскрытия скобок в $(n, 2)$-продуктах, соответствующих всем матчам, каждый моном длины 2 (состоящий из разных переменных, порядок которых в мономе не важен) встречается в точности один раз. Теперь дадим более общее определение, сделав количество скобок в продукте равным произвольному фиксированному числу $k$. Пусть $n, k \in \mathbf{Z}, 0 \leq 2 k \leq n$. Будем называть ( $n, k$ )-продуктом (или просто продуктом) следующее выражение со скобками, являющееся произведением двучленов: $$ P=\prod_{i=1}^{k}\left(x_{i, 1}+x_{i, 2}\right) $$ где $x_{i, 1}, x_{i, 2}, i=1, \ldots, k$, - неповторяющиеся переменные из множества $\left\{x_{1}, \ldots, x_{n}\right\}$. Фигурирующие в записи продукта знаки произведения и сложения считаем обычными алгебраическими операциями. ${ }^{1} 2$ Разложением $(n, k)$-продукта $P$ назовем совокупность $2^{k}$ мономов длины $k$, получающихся после раскрытия скобок в продукте $P$ по правилу дистрибутивности. Считаем, что разложением $(n, 0)$ продукта является моном длины 0 . Будем говорить, что сумма продуктов $\sum_{i=1}^{s} P_{i}$ является упаковкой,[^16]если разложения никаких двух продуктов $P_{i}$ и $P_{j}, i \neq j$, не содержат общих мономов. Число $s$ продуктов в сумме назовем длиной упаковки продуктов. Введем величины $A_{n, k}$. Через $A_{n, k}$ обозначим максимально возможное значение длины упаковки $(n, k)$-продуктов. Будем называть упаковку $(n, k)$-продуктов совершенной, если каждый моном длины $k$ встречается в разложении в точности одного продукта. Задача 14 Докажите, что $A_{n, 0}=1$. Задача 15 Докажите, что $A_{n, 1}=\left\lfloor\frac{n}{2}\right\rfloor$. Задача 16 Докажите, что $A_{n, 2}=\binom{\frac{n}{2}}{2}$ при четном $n$. Задача 17 Докажите, что $A_{n,\left\lfloor\frac{n}{2}\right\rfloor}=1$. Задача 18 Докажите, что $A_{n, \frac{n}{2}-1}=\frac{n}{2}$ при четном $n$. Задача 19 Докажите, что $A_{n, k} \leq \frac{\binom{n}{k}}{2^{k}}$. Задача 20 (проверка на внимательность) Дожажите, что $A_{10,3}=15$. Задача 21 Докажите, что $A_{n, k} \leq\binom{ n}{2 k}$. Задача 22 Докажите, что при $00$, что наибольший корень многочлена $x^{n}-\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x^{k}$ при любом $n$ строго менъше $2-c$. Задача 39** (открытая проблема) Найдите верхний предел наибольшего корня многочлена $x^{n}-$ $\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x^{k}$ nрu $n \rightarrow \infty$. ## ЧАСТЬ II ## УСЛОВИЯ ЗАДАЧ ## ЗАДАЧИ ПОСЛЕ ПРОМЕЖУТОЧНОГО ФИНИША. ## Упаковки «больших» продуктов Пусть $m, t \in \mathbf{Z}, 0 \leq t \leq m$. Будем называть большим ( $m, t$ )-продуктом произведение, в которое в качестве множителя входят выражения трех типов: а) $x_{i}$; б) $\left(x_{i}+x_{j}\right)$, в) $\left(x_{i}+1\right)$; при этом каждая из переменных $x_{1}, \ldots, x_{m}$ встречается в выражении в точности один раз; суммарное число множителей типов а) и б) равно в точности $t$. Разложением большого $(m, t)$-продукта $P$ назовем совокупность мономов, получающихся после раскрытия скобок в продукте $P$, порядок следования переменных в мономе не имеет значения. Будем говорить, что сумма больших продуктов $\sum_{i=1}^{s} P_{i}$ является упаковкой, если разложения никаких двух продуктов $P_{i}$ и $P_{j}, i \neq j$, не содержат общих мономов. Число $s$ продуктов в сумме назовем длиной упаковки больших продуктов. Упаковку длины $s$ больших $(m, t)$-продуктов будем называть $s$-длинной. Если один или несколько параметров упаковки больших $(m, t)$-продуктов нам или совсем не важны, либо определяются в зависимости от процедуры построения, то мы будем заменять эти параметры на прочерки. Например, если мы задаем только параметр $t$, то будем говорить об упаковке больших $(-, t)$-продуктов. Пусть $m, r, t \in \mathbf{Z}, 0 \leq r \leq t \leq m$. Большой $(m, t)$-продукт будем называть большим $(m, t, r)$ продуктом, (обратите внимание, появляется новый параметр $r$ ), если число множителей типа б) не превосходит $r$. Особый интерес для приложений представляют $s$-длинные упаковки больших $(m, t, r)$-продуктов, когда $s=2^{r}$. Если для некоторой $s$-длинной упаковки больших $(m, t, r)$-продуктов выполнено $s>2^{r}$, то можно просто удалить «лишние» большие продукты. Упаковку больших ( $m, t, r)$-продуктов будем называть жесткой, если все переменные намертво разбиты на пары, т. е. в каждом большом продукте упаковки в выражениях типа б) каждая переменная может участвовать только вместе со своей «партнершей», а если некоторая переменная участвует в некотором большом продукте в выражении типа а), то и ее «партнерша» тоже участвует в этом большом продукте в выражении типа а). На языке теннисистов свойство жесткости можно сформулировать так: теннисисты, как и в обычных теннисных турнирах, жестко разбиты на пары, и партнеров по паре менять нельзя. Задача 40 Придумайте 2-длинную упаковку больших (3,2,1)-продуктов. Задача 41 Придумайте 8-длинную жесткую упаковку больших (8,5,3)-продуктов. Задача 42 Придумайте 16-длинную жесткую упаковку больших (12,6,4)-продуктов. Задача 43 Докажите, что если существуют $s_{1}$-длинная упаковка больших ( $\left.m_{1}, t_{1}, r_{1}\right)$-продуктов и $s_{2}$-длинная упаковка больших ( $\left.m_{2}, t_{2}, r_{2}\right)$-продуктов, то существует $s_{1} s_{2}$-длинная упаковка больших ( $\left.m_{1}+m_{2}, t_{1}+t_{2}, r_{1}+r_{2}\right)$-продуктов. Задача 44 Покажите, как из упаковки $(n, k)$-продуктов получить упаковку больших $(n, n-k)$ продуктов той же длины. Задача построения $2^{r}$-длинной упаковки больших ( $\left.m, t, r\right)$-продуктов с как можно меньшей величиной отношения $\frac{t}{t+r}$ важна для проблемы построения корреляционно-иммунных булевых функций с максимально возможной нелинейностью. Взаимосвязь рассматриваемой задачи и указанной проблемы слишком сложна, чтобы обсуждать ее здесь. В рамках настоящего проекта булевы функции в дальнейшем не упоминаются и знания даже их определений для решения задач не требуется. Задача 45 Доказать, что при $r>0$ для любой $2^{r}$-длинной упаковки больших $(m, t)$-продуктов выполнено $\frac{t}{t+r}>1 / 2$. Открытая проблема: можно ли $2^{r}$-длинными упаковками больших $(m, t, r)$-продуктов (выбирая параметры $m, t$ и $r$ по своему усмотрению) величину $\frac{t}{t+r}$ сколь угодно близко приблизить к $1 / 2$ ? Заметим, что задача 43 не дает продвижений в проблеме построения $2^{r}$-длинной упаковки больших $(m, t, r)$-продуктов с как можно меньшей величиной отношения $\frac{t}{t+r}$, так как $\frac{t_{1}+t_{2}}{t_{1}+t_{2}+r_{1}+r_{2}} \geq$ $\min \left(\frac{t_{1}}{t_{1}+r_{1}}, \frac{t_{2}}{t_{2}+r_{2}}\right)$. Задача б/н. Придумайте конструкцию, позволяющую строить упаковки больших $(-, t)$-продуктов. Главная цель: сделать величину $\frac{t}{t+r}$ настолько близкой к $1 / 2$, насколько это получится. Как вариант действий: придумать рекурсивную конструкцию, исходя из упаковок $(n, k)$ продуктов и некоторых начальных упаковок больших ( $n, u$ )-продуктов. Обозначим через $S_{m, t}$ максимально возможную длину упаковки больших $(m, t)$-продуктов. Задача 46 Докажите, что $n р и$ $m \geq t \geq n$ величины $S_{m, t}$ удовлетворяют рекуррентному неравенству $$ S_{m, t} \geq \sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} S_{m-n, t-n+k} $$ Если нам нужно построить упаковку больших $(-, t, r)$-продуктов и при этом важен параметр $r$, то можно несколько раз воспользоваться задачей 46 , а после этого удалить все большие продукты, в которых число множителей типа б) превышает $r$. Задача 47 Пусть $n$ - натуральное число. Пусть последовательность $\left\{u_{\ell}\right\}$ состоит из положительных чисел и удовлетворяет соотношению $$ u_{\ell+n}=\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} u_{\ell+k} $$ при всех натуральных $\ell$. Обозначим через $X_{\max }$ наибольший корень многочлена $P(x)=x^{n}-$ $\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x^{k}$. Докажите, что существуют положительные константы $C_{1}$ и $C_{2}$ такие, что $C_{1} X_{\max }^{\ell} \leq u_{\ell} \leq C_{2} X_{\max }^{\ell}$ при всех достаточно больших $\ell$. Задача 48 Докажите, что для любого $\varepsilon>0$ можно, пользуясь задачей 46 при $n=2$, построить такую $2^{r}$-длинную упаковку больших $(-, t)$-продуктов, что $\frac{t}{t+r}<\frac{1}{1+\log _{2}\left(\frac{\sqrt{5}+1}{2}\right)}+\varepsilon=0.5902 \ldots+\varepsilon$. Указание. Примите во внимание то, что наибольший корень многочлена $x^{n}-\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x^{k}$ при $n=2$ равен $\frac{1+\sqrt{5}}{2}$. Задача 49 Докажите, что для любого $\varepsilon>0$ можно построить такую $2^{r}$-длинную упаковку больших (-,t,r)-продуктов, что $\frac{t}{t+r}<\frac{1}{1+\log _{2}\left(\frac{\sqrt{5}+1}{2}\right)}+\varepsilon=0.5902 \ldots+\varepsilon$. Задача 50* Докажите, что для любой $2^{r}$-длинной жесткой упаковки больших $(-, t)$-продуктов имеет место неравенство $\frac{t}{t+r}>\frac{1}{1+\log _{2}\left(\frac{\sqrt{5}+1}{2}\right)}=0.5902 \ldots$ Задача 51* Постройте 2r-длинную упаковку больших (-,t)-продуктов, для которой $\frac{t}{t+r}<$ $\frac{1}{1+\log _{2}\left(\frac{\sqrt{5}+1}{2}\right)}=0.5902 \ldots$ Задача 52* Постройте $2^{r}$-длинную упаковку больших $(-, t, r)$-продуктов, для которой $\frac{t}{t+r}<$ $\frac{1}{1+\log _{2}\left(\frac{\sqrt{5}+1}{2}\right)}=0.5902 \ldots$ ## РЕШЕНИЯ И КОММЕНТАРИИ ## ЧАСТЬ I ## ЗАДАЧИ ДО ПРОМЕЖУТОЧНОГО ФИНИША. ## Задача о серии парных теннисных матчей и ее вариации Задача 1 предлагалась на заключительном этапе всероссийской олимпиады школьников 1993 года [3]. Автор задачи - С. Токарев. Задача 1. Пусть описываемая в задаче серия матчей проведена. Тогда все противники одного теннисиста разбиваются на пары, игравшие против него в одном матче, поэтому $n$ нечетно. Все $\frac{n(n-1)}{2}$ пар теннисистов разбиваются на четверки пар противников в одном матче. Следовательно, число этих пар $\frac{n(n-1)}{2}$ кратно четырем, откуда $n-1=8 l$. Докажем, что при любом натуральном $l$ описанная в условии задачи серия матчей для $n=8 l+1$ возможна. Для $l=1$ нумеруем теннисистов числами от 1 до 9 и предъявляем расписание в явном виде: 1 и 7 против 2 и $3 ; 1$ и 9 против 4 и $5 ; 1$ и 5 против 6 и $7 ; 1$ и 3 против 8 и $9 ; 2$ и 3 против 4 и $5 ;$ 2 и 4 против 6 и $8 ; 7$ и 8 против 6 и $9 ; 5$ и 7 против 4 и $8 ; 3$ и 9 против 2 и 6. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-154.jpg?height=661&width=672&top_left_y=1193&top_left_x=683) Рассмотрим теперь произвольное $n$, сравнимое с 1 по модулю 8. Выделим одного особого теннисиста, а остальных разобьем на восьмерки. Объединяя особого теннисиста с каждой из восьмерок, организуем расписание матчей девяток, как указано выше. Таким образом, особый теннисист уже встретился в качестве соперника с любым другим; любые два теннисиста из одной восьмерки также встретились между собой в качестве соперников. Осталось добиться, чтобы любые два теннисиста из разных восьмерок встретились в качестве соперников. Для этого внутри восьмерок жестко разбиваем теннисистов на пары, и для любых двух разных восьмерок проводим матч каждой пары из одной восьмерки против каждой пары из другой восьмерки. Таким образом, требуемое расписание матчей составлено. Ответ. При $n \equiv 1(\bmod 8)$. Задача 2. Пусть сначала $n$ четно. Тогда у каждого теннисиста число потенциальных соперников равно $n-1$ и нечетно. В каждом матче, в котором теннисист участвует, у него два соперника. Поэтому каждый теннисист может сыграть не более чем в $\frac{n-2}{2}$ матчах. Отсюда, суммируя по всем теннисистам и деля на 4 (поскольку в матче участвуют четыре теннисиста), получаем, что общее число матчей не превосходит $\frac{n(n-2)}{8}$. В то же время такое число матчей провести можно: жестко разобьем теннисистов на пары и пусть каждая пара сыграет с каждой. Таким образом, при четном $n$ максимальное число матчей равно $\frac{n(n-2)}{8}$. Пусть теперь $n$ нечетно. Рассуждениями, аналогичными вышеприведенным, получаем, что число матчей не превосходит $\frac{n(n-1)}{8}$; для того, чтобы эта оценка достигалась, необходимо, чтобы $n \equiv 1$ $(\bmod 8)$, при этом любые два теннисиста встретятся в качестве противников в точности один раз. В задаче 1 показано, что при $n \equiv 1(\bmod 8)$ такое расписание составить можно. Таким образом, при $n \equiv 1(\bmod 8)$ максимальное число матчей равно $\frac{n(n-1)}{8}$. Рассмотрим теперь остальные случаи нечетного $n$. Построим конструкции для этих случаев. Выделяем одного особого теннисиста, а остальных делим на восьмерки и одну неполную группу, в которой 2,4 или 6 теннисистов. Объединяя особого теннисиста с каждой из восьмерок, организуем расписание матчей девяток, как указано выше. Внутри восьмерок и неполной группы жестко разбиваем теннисистов на пары, и для любых двух разных восьмерок (или неполной группы) проводим матч каждой пары из одной восьмерки против каждой пары из другой восьмерки (или неполной группы). Таким образом, любые два теннисиста уже встретились между собой в качестве соперников, за исключением следующего случая: оба эти теннисиста из неполной группы, объединенной с особым теннисистом. Назовем это объединение остаточной группой, в нее входит $m$ теннисистов, $m=3,5,7$, где $m \equiv n(\bmod 8)$. Проведем дополнительно некоторые матчи с участием только теннисистов из остаточной группы. Ясно, что нельзя добиться, чтобы каждый теннисист в этом случае встретился с каждым другим в качестве соперника. Определим, какое минимальное число пар теннисистов может не встретиться друг с другом к качестве соперника. Поскольку $n$ нечетно, каждый теннисист может не встретиться в качестве соперника с четным числом разных теннисистов, поэтому общее число пар теннисистов, которые не встретятся друг с другом в качестве соперников, не может равняться в точности 1 или 2. Рассмотрим теперь оставшиеся серии $n$ по отдельности. Пусть $m \equiv n \equiv 3(\bmod 8)$. В остаточной группе три пары не встретившихся между собой в качестве соперников теннисистов, а меньше и быть не может, как мы показали выше. Максимальное число матчей равно $\frac{\frac{n(n-1)}{2}-3}{4}=\frac{(n-3)(n+2)}{8}$. Пусть $m \equiv n \equiv 5(\bmod 8)$. В остаточной группе проведем матч 1 и 2 против 3 и 4. После этого в остаточной группе остается шесть пар не встретившихся между собой в качестве соперников теннисистов. Если бы прошел еще один матч, то осталось бы только две пары не встретившихся между собой в качестве соперников теннисистов, чего быть не может, как мы показали выше. Максимальное число матчей равно $\frac{\frac{n(n-1)}{2}-6}{4}=\frac{(n-4)(n+3)}{8}$. Пусть $m \equiv n \equiv 7(\bmod 8)$. В остаточной группе проведем матчи 1 и 2 против 3 и $4 ; 2$ и 4 против 6 и $7 ; 1$ и 3 против 5 и $6 ; 1$ и 5 против 2 и 7. После этого в остаточной группе остается пять пар не встретившихся между собой в качестве соперников теннисистов. Если бы прошел еще один матч, то осталась бы только одна пара не встретившихся между собой в качестве соперников теннисистов, чего быть не может, как мы показали выше. Максимальное число матчей равно $\frac{\frac{n(n-1)}{2}-5}{4}=\frac{(n-2)(n+1)}{8}-1$. Ответ. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-156.jpg?height=262&width=666&top_left_y=308&top_left_x=683) Задача 3. Мы уже знаем из решения задачи 1 , что хотя бы одно совершенное расписание существует лишь при $n \equiv 1(\bmod 8)$. Пусть $n=8 l+1$. Пронумеруем теннисистов числами от 0 до $8 l$ и включим в расписание матчи 0 и 1 против $4 i+2$ и $4 i+4, i=0,1, \ldots, l-1$, а также все их циклические сдвиги. Тогда разности между номерами теннисистов в выписанных матчах принимают по разу все значения между 1 и $4 l$, а с учетом циклических сдвигов матчей любые два теннисиста окажутся соперниками ровно в одном матче. Задача 4. Расположим игроков по кругу и занумеруем их последовательно вычетами по модулю $n=8 l+1$. В случае $l=1$ возьмем матч $\{\{0,1\}\{2,4\}\}$ и все его циклические сдвиги. В дальнейших примерах, когда мы включаем в расписание матч, мы добавляем в него и все его циклические сдвиги. $$ \begin{aligned} & l=2 \quad l=3 \quad l=4 \quad l=5 \quad l=6 \quad l=7 \\ & \{\{0,1\}\{2,8\}\} \quad\{\{0,1\}\{2,12\}\} \quad\{\{0,1\}\{2,18\}\} \quad\{\{0,1\}\{2,20\}\} \quad\{\{0,1\}\{2,26\}\} \quad\{\{0,1\}\{2,27\}\} \\ & \{\{0,2\}\{5,13\}\} \quad\{\{0,2\}\{5,10\}\} \quad\{\{0,2\}\{5,14\}\} \quad\{\{0,2\}\{5,17\}\} \quad\{\{0,2\}\{5,22\}\} \quad\{\{0,2\}\{5,36\}\} \\ & \{\{0,3\}\{7,19\}\} \quad\{\{0,3\}\{7,25\}\} \quad\{\{0,3\}\{7,32\}\} \quad\{\{0,3\}\{7,21\}\} \quad\{\{0,3\}\{7,17\}\} \\ & \{\{0,4\}\{10,24\}\} \quad\{\{0,4\}\{10,18\}\} \quad\{\{0,4\}\{10,37\}\} \quad\{\{0,4\}\{10,28\}\} \\ & \{\{0,5\}\{13,30\}\} \quad\{\{0,5\}\{13,35\}\} \quad\{\{0,5\}\{13,25\}\} \\ & \{\{0,6\}\{15,38\}\} \quad\{\{0,6\}\{15,41\}\} \\ & \{\{0,7\}\{18,45\}\} \end{aligned} $$ Расстояние между парой игроков $a$ и $b$ определим как минимальное расстояние между этими игроками по окружности (расстояние между соседними равно единице). Рассматривая расстояния между парами противников, заметим что они по разу принимают значения от 1 и $4 l$. Значит любая пара противников встречается в расписании ровно один раз. Рассматривая расстояния между парами союзников, заметим, что они все различны. Значит у каждого теннисиста партнер по паре повторяется не более одного раза. Замечание 2 В ходе конференции в задаче 4 В. Ретинским и Д. Захаровым была найдена конструкция, позволяющая строить много примеров для новых значений парамера $n$. Представляется, что эта конструкция дает бесконечную серию значений параметра, но доказательства этого пока нет. Задача 5. Занумеруем всех игроков вычетами по модулю $n=8 l+1$. Рассмотрим циклическое расписание, порожденное следующими матчами: $$ \begin{aligned} & \{\{0,1\}\{l+1,3 l+1\}\} \\ & \{\{0,3\}\{l+2,3 l+2\}\} \\ & \{\{0,5\}\{l+3,3 l+3\}\} \\ & \vdots \\ & \{\{0,2 l-5\}\{2 l-2,4 l-2\}\} \\ & \{\{0,2 l-3\}\{2 l-1,4 l-1\}\} \\ & \{\{0,2 l-1\}\{2 l, 4 l\}\} \end{aligned} $$ Рассматривая расстояния между парами противников, заметим что они по разу принимают значения от 1 и $4 l$. Значит любая пара противников встречается в расписании ровно один раз. Выпишем расстояния между парами союзников: $\{1,3,5, \ldots, 2 l-5,2 l-3,2 l-1,2 l\}$. Число этих расстояний $l+1$, а значит число различных союзников равно $2 l+2$, что больше, чем $\frac{n}{4}$. Задача 6. Пусть $n=5$. Возьмем расписание шахматного турнира, тур - номер матча, только соперники в туре теперь партнеры против другой пары. Пусть $n=8$. Занумеровав семерых теннисистов элементами кольца вычетов по модулю 7 , а восьмого обозначив буквой $A$, для $k=0,1, \ldots, 6$ проведем матч $\{k, k+1\}$ против $\{k+3, A\}$ и матч $\{k+2, k+5\}$ против $\{k+4, k+6\}$. Пусть $n=9$. Сопоставим теннисистам элементы матрицы $3 \times 3$ и пусть для каждых $(i, j), i=$ $1,2,3, j=1,2,3$, происходит следующее. 1) Теннисист $i j$ отдыхает. 2) пара из строки, содержащей $i j$, играет против пары из столбца, содержащего $i j$. 3) Удалим из матрицы $3 \times 3$ строку $i$ и столбец $j$. Получится матрица $2 \times 2$. Для этой матрицы $2 \times 2$ пара теннисистов, соответствующих одной диагонали, играет против пары теннисистов, соответствующих другой диагонали. Например: $i=2$, $j=3$. Тогда теннисист 23 отдыхает, проводятся матч $\{21,22\}$ против $\{13,33\}$ и матч $\{31,12\}$ против $\{11,32\}$. Замечание 3 В ходе конференции в задаче 7 В. Ретинским и Д. Захаровым была найдена конструкция, позволяющая строить расписание для бесконечной серии значений параметра $n$. Задача 8. Пусть сначала $m$ нечетно, тогда $n-1$ должно делиться на $m^{2}$. Общее число пар $n(n-1) / 2$ должно делиться на число пар $m^{2}$, встречающихся в матче, т. е. $m^{2} \mid n(n-1) / 2$. С другой стороны, произвольный игрок в каждом матче играет с кратным $m$ числом людей, т. е. $m \mid n-1$. Отсюда $m^{2} \mid n-1$. Пусть теперь $m$ четно, тогда $n-1$ должно делиться на $2 m^{2}$. Повторяем те же рассуждения, что и для нечетного $m$, но теперь из $m^{2} \mid n(n-1) / 2$ и $m \mid n-1$ следует $2 m^{2} \mid n-1$. Тем самым мы доказали утверждение задачи 8 , а для четного $m$ получили даже более сильный результат, который мы используем в дальнейшем при решении задачи 9. Задача 9. В решении задачи 8 мы уже получили, что если $m$ четно, то $n-1$ делится на $2 m^{2}$. В любом случае, вне зависимости от четности $m$, если $n-1$ делится на $2 m^{2}$, то искомое расписание существует. Покажем это. Сначала предъявим циклическое расписание для $n=2 m^{2}+1$. Это циклические сдвиги матча $\{0,1,2, \ldots, m-1\}$ против $\left\{m, 2 m, 3 m, \ldots, m^{2}\right\}$. Между любыми двумя игроками расстояние по кругу не больше $m^{2}$, т. е. равно $a m-b$ при $1 \leq a \leq m$ и $0 \leq b \leq m-1$. Тогда при некотором циклическом сдвиге нашего матча они будут образами игроков $b$ и $m a$, т. е. сыграют. Осталось показать, что из существования совершенного $m$-расписания для $n_{1}$ и $n_{2}$ следует существование его для $n_{1}+n_{2}-1$ игроков. Напомним, что $m \mid n_{1}-1$ и $m \mid n_{2}-1$. Осуществим оба расписания на двух множествах игроков с одним общим. После этого останется осуществить все игры между игроком из первых $n_{1}-1$ и игроком из последних $n_{2}-1$. Это можно сделать, разбив множество из первых $n_{1}-1$ игроков и множество из последних $n_{2}-1$ игроков на группы по $m$ игроков, и проведя матчи каждой группы из первого множества против каждой группы из второго множества. Замечание 4 Из приведенного решения задачи 9 помимо утверждения этой задачи следует, что если при некотором нечетном $m$ требуемое возможно при некотором четном $n$ (таком, что $n-1$ делится на $m^{2}$ ), то и при бо́льших четных $n$ (с тем же условием) требуемое возможно. Задача 10. Если совершенное 3 -расписание есть, то в нем $\binom{10}{2} / 3^{2}=5$ матчей, и каждый игрок участвует в трех из них. Рассмотрим игрока 1 и три матча, в которых он участвовал, назовем плохими. Максимум с 6 другими игроками он присутствовал в одной тройке во время матча. Значит, есть хотя бы три игрока $2,3,4$, не побывавшие в одной тройке с 1 . Каждый из них участвовал ровно в одном плохом матче (в противоположной тройке, нежели 1), т. е. каждый из них участвовал в обоих неплохих матчах. Кроме того, все попарные встречи игроков $2,3,4$ происходили в неплохих матчах. Но в каждом из них было четное число таких встреч, а всего их должно быть 3 . Противоречие. Задача 12. Пусть описываемая в задаче серия матчей проведена. Тогда все противники одного теннисиста разбиваются на пары, за исключением двоих, которые будут в паре с этим теннисистом, когда он играет против самого себя, поэтому $n$ нечетно. Все возможные пары противников разбиваются на четверки пар, игравших в одном матче. Следовательно, число этих пар $\frac{n(n-1)}{2}+n=\frac{n(n+1)}{2}$ кратно четырем, откуда $n+1=8 l$. Для $l=1$ нумеруем теннисистов числами от 0 до 6 . Включим в расписание матч 0 и 1 против 1 и 3 . В этом матче разности между номерами теннисистов принимают по разу значения $0,1,2,3$. Поэтому если мы включим в расписание также все циклические сдвиги этого матча, то любые два теннисиста (включая совпадающих) окажутся соперниками ровно в одном матче. Для произвольного $l$ пронумеруем теннисистов числами от 0 до $8 l-2$ и включим в расписание матчи 0 и 1 против $4 i+1$ и $4 i+3, i=0,1, \ldots, l-1$, а также все их циклические сдвиги. Тогда разности между номерами теннисистов в выписанных матчах принимают по разу все значения между 0 и $4 l-1$, а с учетом циклических сдвигов матчей любые два теннисиста (включая совпадающих) окажутся соперниками ровно в одном матче. Ответ. При $n \equiv 7(\bmod 8)$. ## Упаковки продуктов Загадка 13. Если раскрыть скобки по распределительному закону, то получится $15 \cdot 2^{3}=120$ мономов длины 3, при этом все эти мономы будут попарно различными. В то же время всего существует в точности $\binom{10}{3}=120$ различных мономов длины 3 от 10 переменных. Поэтому при раскрытии скобок в загадке 13 каждый возможный моном длины 3 от 10 переменных возникнет в точности один раз. Поэтому загадка 13 дает пример совершенной упаковки $(10,3)$-продуктов. Задача 14. Моном длины 0 существует только один, и получить его можно из единственного продукта, содержащего нуль скобок. Задача 15. Существует в точности $n$ мономов длины 1 от $n$ переменных. Каждый $(n, 1)$-продукт при разложении дает 2 монома, отсюда следует верхняя оценка $A_{n, 1} \leq\left\lfloor\frac{n}{2}\right\rfloor$. В то же время мы получим упаковку длины $\left\lfloor\frac{n}{2}\right\rfloor$, если возьмем в каждый ( $n, 1$ )-продукт разные переменные. Задача 16. При четном $n$ для каждой переменной $x_{i}$ есть $n-1$ отличная от нее переменная. Пусть есть упаковка $(n, 2)$-продуктов. Рассмотрим совокупность всех таких $(n, 2)$-продуктов этой упаковки, в одной из скобок которых содержится переменная $x_{i}$. Тогда в другой скобке этих продуктов должны содержаться попарно разные переменные, иначе какой-то моном длины 2 при разложении возникнет больше одного раза. Поэтому каждая из $n$ переменных содержится не более чем в $\left\lfloor\frac{n-1}{2}\right\rfloor=\frac{n}{2}-1$ продуктах. В каждый ( $n, 2$ )-продукт входит 4 переменные, поэтому $$ A_{n, 2} \leq \frac{1}{4} n\left(\frac{n}{2}-1\right)=\binom{\frac{n}{2}}{2} $$ В то же время можно составить упаковку из такого числа продуктов. Жестко объединим $n$ переменных в $\frac{n}{2}$ пар и из любых двух различных пар составим продукт. Получится упаковка из в точности $\binom{\frac{n}{2}}{2}$ продуктов. Замечание 5 Фактически задача 16 эквивалентна случаю четного $n$ в задаче 2 . Задача 17. Пусть $k=\left\lfloor\frac{n}{2}\right\rfloor$. Пусть есть два ( $n, k$ )-продукта. Сопоставим каждой переменной вершину графа. Для каждой скобки каждого из двух продуктов соединим ребром вершины, соответствующие переменным из этой скобки. Тогда полученный граф разобьется на циклы четной длины и (при нечетном $n$ ) цепочку с нечетным числом вершин. Выделим в каждом цикле вершины через одну (любым из двух способов). В цепочке тоже выделим вершины через одну, беря только неконцевые вершины. Тогда множество выделенных вершин задает моном, который, как легко видеть, получается при разложении каждого из двух продуктов. Поэтому больше одного продукта в упаковку взять нельзя. Задача 18. Пусть $k=\frac{n}{2}-1$. Тогда в каждый $(n, k)$-продукт не входят в точности две из имеющихся $n$ переменных. Мы утверждаем, что для любых двух продуктов упаковки множества не входящих в них переменных не пересекаются. Если это не так, то мы рассуждениями из решения задачи 17 получаем моном, получающийся при разложении обоих продуктов. Отсюда $A_{n, \frac{n}{2}-1} \leq \frac{n}{2}$. В то же время можно составить упаковку из такого числа продуктов. Жестко объединим $n$ переменных в $\frac{n}{2}$ пар и составим все возможные продукты, содержащие все сформированные пары переменных, кроме одной. Легко видеть, что любой моном из разложения одного такого продукта не содержится в разложении никакого другого. Поэтому получится упаковка из в точности $\frac{n}{2}$ продуктов. Задача 19. Всего различных мономов длины $k$ от $n$ переменных в точности $\binom{n}{k}$, разложение каждого $(n, k)$-продукта дает $2^{k}$ таких мономов. Отсюда $A_{n, k} \leq \frac{\binom{n}{k}}{2^{k}}$. Задача 20. Верхняя оценка следует из задачи 19. Пример упаковки длины 15 приведен в загадке 13. Задача 21. Если множества переменных входящих в два $(n, k)$-продукта совпадают, то мы рассуждениями из решения задачи 17 получаем моном, получающийся при разложении обоих продуктов. Отсюда $A_{n, k} \leq\binom{ n}{2 k}$. Задача 22. Возьмем упаковку ( $n, k$ )-продуктов максимальной длины, т. е. длины $A_{n, k}$. Ко множеству переменных каждого из продуктов упаковки добавим не принадлежащую ему переменную всеми возможными способами (которых в точности $n-2 k$ ). Все получившиеся таким способом множества из $2 k+1$ переменных должны быть попарно различными в силу задачи 17 . Отсюда $A_{n . k}(n-2 k) \leq\binom{ n}{2 k+1}$ или $$ A_{n . k} \leq \frac{\binom{n}{2 k+1}}{n-2 k}=\frac{n!}{(2 k+1)!(n-2 k-1)!(n-2 k)}=\frac{\binom{n}{2 k}}{2 k+1} $$ Теперь из множества переменных каждого из продуктов максимальной упаковки удалим принадлежащую ему переменную всеми возможными способами (которых в точности $2 k$ ). Все получившиеся таким способом множества из $2 k-1$ переменных должны быть попарно различными в силу той же задачи 17. Отсюда $A_{n . k} \cdot 2 k \leq\binom{ n}{2 k-1}$ или $$ A_{n . k} \leq \frac{\binom{n}{2 k-1}}{2 k}=\frac{n!}{(2 k-1)!(n-2 k+1)!\cdot 2 k}=\frac{\binom{n}{2 k}}{n-2 k+1} $$ Задача 23. Жестко объединим переменные в $\left\lfloor\frac{n}{2}\right\rfloor$ пар и возьмем в продукт произвольные сочетания по $k$ из них. Легко видеть, что любой моном из разложения одного такого продукта не содержится в разложении никакого другого. Поэтому получится упаковка из в точности $\binom{\left\lfloor\frac{n}{2}\right\rfloor}{ k}$ продуктов. Задача 24. Выделим отдельно пару переменных $x_{n-1}$ и $x_{n}$. Образуем следующую совокупность продуктов. Включим в нее все продукты, входящие в упаковку $(n-2, k)$-продуктов максимальной длины, а также все продукты, входящие в упаковку ( $n-2, k-1$ )-продуктов максимальной длины, с дописанной к ним скобкой $\left(x_{n-1}+x_{n}\right.$ ). Легко видеть, что любой моном из разложения одного продукта из образованной совокупности не содержится в разложении никакого другого. Отсюда вытекает требуемое рекуррентное неравенство. Задача 25. Разобьем исходный набор переменных на два, состоящих из $n_{1}$ и $n_{2}$ различных переменных. Составим следующую упаковку продуктов: для каждого $i, 0 \leq i \leq k$, перемножим упаковки, соответствующие числам $A_{n_{1}, i}$ и $A_{n_{2}, k-i}$, а потом сложим результаты. Отсюда и из того, что длина упаковки, полученной из произведения упаковок, равна произведению длин этих упаковок получаем требуемое неравенство. Задача 26. Получить совершенную упаковку ( $n-1, k-1$ )-продуктов из совершенной упаковки $(n, k)$-продуктов можно следующим способом. Возьмем какую-то переменную, скажем, $x_{n}$. Найдем все продукты в упаковке, содержащие переменную $x_{n}$. Откинем в них скобку с этой переменной. Получим ( $n-1, k-1$ )-продукты. Несложно увидеть, что их сумма и будет искомой упаковкой. Задача 27. См. решение задачи 29. Задача 28. Покажем, что остаток $n$ по модулю 8 может быть равен только 2. Допустим, что для $n=8 k+i$ существует совершенная упаковка ( $8 k+i, 3)$-продуктов. Тогда в силу задачи 26 существует совершенная упаковка $(8 k+i-1,2)$-продуктов. По задаче 1 совершенная упаковка $(m, 2)$-продуктов может существовать только при $m=8 l+1$. Отсюда следует, что $n$ может быть равно только $n=8 k+2$. В то же время, применяя задачу 29 к совершенной упаковке ( 10,3 )-продуктов, получим, что для всех чисел вида $n=8 k+2$ существует совершенная упаковка. Ответ. При $n=8 k+2$, где $k$ - натуральное число. Задача 29. По задаче 26 , для всех $i=0,1, \ldots, k$ существуют совершенная упаковка $\left(n_{1}-i, k-i\right)$ продуктов и совершенная упаковка ( $\left.n_{2}-i, k-i\right)$-продуктов. Пусть $X, Y, Z$ - непересекающиеся множества переменных, причем $|X|=n_{1}-k,|Y|=n_{2}-k, Z=\left\{z_{0}, z_{1}, \ldots, z_{k}\right\}$. Образуем следующее объединение упаковок продуктов. При каждом $i=0,1, \ldots, k$ возьмем совершенную упаковку ( $n_{1}-k+i, i$ )-продуктов на множестве переменных $X \cup\left\{z_{0}, \ldots, z_{i-1}\right\}$ (назовем ее $M_{i}$ ) и совершенную упаковку ( $\left.n_{2}-i, k-i\right)$-продуктов на множестве переменных $Y \cup\left\{z_{i+1}, \ldots, z_{k}\right\}$ (назовем ее $N_{i}$ ). Произведения произвольного продукта из $M_{i}$ и произвольного продукта из $N_{i}$ дадут, очевидно, упаковку ( $n_{1}+n_{2}-k+1, k$ )-продуктов на множестве переменных $X \cup Y \cup Z$, которую мы обозначим через $U_{i}$. Мы утверждаем, что объединение $U$ всех этих упаковок является совершенной упаковкой $\left(n_{1}+n_{2}-k+1, k\right)$-продуктов. Формально, $$ \bigsqcup_{i=0}^{k} U_{i}, \quad U_{i}=P\left(X \sqcup\left\{\bigsqcup_{j=0}^{i-1} z_{j}\right\}, i\right) \times P\left(Y \sqcup\left\{\bigsqcup_{j=i+1}^{k} z_{j}\right\}, k-i\right) $$ где $P(.,$.$) - соответствующие совершенные упаковки.$ Покажем сначала, что любой моном $m$ степени $k$ встречается при раскрытии скобок не более одного раза. Пусть он встретился в разложениях некоторых продуктов из $U_{i}$ и $U_{j}$ при $i0$. Следовательно у многочлена $\widetilde{P}$ есть корень, больший $\alpha$. Задача 34. При $x \geq 2$ верно следующее неравенство: $$ x^{n} \geq(1+x / 2)^{n}=\sum_{k=0}^{n}\binom{n}{k}(x / 2)^{k}>\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor}\binom{n}{k}(x / 2)^{k} \geq \sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x^{k} $$ где последнее неравенство следует из утверждения задачи 19. Следовательно $x^{n}-\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x^{k}>0$ при любом $x \geq 2$. Задача 35. Ясно, что $P_{n}(0) \leq 0$, поэтому наибольший действительный корень многочлена $P_{n}(x)-$ неотрицательный. Обозначим, $P_{n}^{+}(x)=x^{n}+\sum_{k=0}^{n-1} a_{n, k} x^{k}$. Все $a_{n, k}$ неотрицательны, поэтому многочлен $P_{n}^{+}(x)$ не имеет положительных корней. Отсюда наибольший действительный корень многочлена $P_{n}^{\prime}(x)=P_{n}(x) P_{n}^{+}(x)$ такой же, как и у многочлена $P_{n}(x)$. Имеем $P_{n}^{\prime}(x)=P_{n}(x) P_{n}^{+}(x)=x^{2 n}-\left(\sum_{k=0}^{n-1} a_{n, k} x^{k}\right)^{2}=x^{2 n}-\sum_{k=0}^{2 n-1} u_{2 n, k} x^{k}$, где $u_{2 n, k}=\sum_{i=0}^{k} a_{n . i} a_{n, k-i}$. В то же время из условия, что если к любой фразе справа дописать любую фразу, то снова получится фраза, имеем $a_{2 n, k} \geq \sum_{i=0}^{k} a_{n . i} a_{n, k-i}$. Следовательно, $a_{2 n, k} \geq u_{2 n, k}$ для любого $k$. Пусть $x_{0}$ - наибольший (действительный) корень многочлена $P_{n}(x)$. Мы уже получили, что $x_{0}$ также является наибольшим (действительным) корнем многочлена $P_{n}^{\prime}(x)$, т. е. $P^{\prime}\left(x_{0}\right)=0$. Поскольку $a_{2 n, k} \geq u_{2 n, k} \geq 0$, из задачи 33 следует, что наибольший корень $P_{2 n}(x)$ не меньше $x_{0}$. Загадка 36. Ответ. Неравенство из задачи 25. Задача 37. Пусть $x_{0}^{n}=\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x_{0}^{k}$, тогда верно $$ x_{0}^{2 n}=\left(\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x_{0}^{k}\right)^{2} \leq \sum_{k=0}^{n}\left(\sum_{i=0}^{k} A_{n, i} \cdot A_{n, k-i}\right) x_{0}^{k} \leq \sum_{k=0}^{n} A_{2 n, k} x_{0}^{k} $$ где последнее неравенство следует из утверждения задачи 25. Следовательно, справедливо неравенство $x^{2 n}-\sum_{k=0}^{n} A_{2 n, k} x^{k} \leq 0$ при $x=x_{0}$. В то же время, так как коэффициент при старшей степени больше нуля, то при достаточно больших $x$ многочлен принимает положительные значения, а, значит, он имеет корень на луче $\left[x_{0}, \infty\right)$. Задача 38. Нам потребуется следующее неравенство. Лемма 1 Справедливо неравенство $$ \binom{n}{k} \leq \frac{n^{n}}{k^{k}(n-k)^{n-k}} $$ Доказательство. Раскладывая $n^{n}=(k+(n-k))^{n}$ по биному Ньютона и выделяя в нем $k$-й член, получаем $$ n^{n}=\sum_{i=0}^{n}\binom{n}{k} k^{i}(n-k)^{n-i} \geq\binom{ n}{k} k^{k}(n-k)^{n-k} $$ что и требовалось доказать. С помощью леммы 1 можно привести результаты задач 19 и 21 к более удобной форме (несколько ослабив их). Именно, обозначив $\lambda=k / n$, из задачи 19 получаем $$ A_{n, k} \leq \frac{\binom{n}{k}}{2^{k}} \leq \frac{n^{n}}{(\lambda n)^{\lambda n}((1-\lambda) n)^{(1-\lambda) n} \cdot 2^{\lambda n}}=\left((2 \lambda)^{-\lambda}(1-\lambda)^{-(1-\lambda)}\right)^{n} $$ а из задачи 21 , аналогичным образом, $$ A_{n, k} \leq\binom{ n}{2 k} \leq\left((2 \lambda)^{-2 \lambda}(1-2 \lambda)^{-(1-2 \lambda)}\right)^{n} $$ Обозначим через $P_{n}(x)$ наш многочлен, а через $\alpha$ (или $\alpha_{n}$ ) - его наибольший корень. Пусть $\ell$ - номер наибольшего слагаемого в сумме $\sum_{k=0}^{\lfloor n / 2\rfloor} A_{n, k} \alpha^{k}$. Тогда из оценок выше имеем $$ \alpha^{n}=\sum_{k=0}^{\lfloor n / 2\rfloor} A_{n, k} \alpha^{k} \leq n A_{n, \ell} \alpha^{\ell} \leq n \alpha^{\ell} \cdot\left(\min \left\{(2 \lambda)^{-\lambda}(1-\lambda)^{-(1-\lambda)}, \quad(2 \lambda)^{-2 \lambda}(1-2 \lambda)^{-(1-2 \lambda)}\right\}\right)^{n} $$ где $\lambda=\ell / n$. Отсюда, извлекая корень степени $n-\ell=(1-\lambda) n$, получаем $$ \alpha \leq \sqrt[n-\ell]{n} \cdot\left(\min \left\{(2 \lambda)^{-\lambda}(1-\lambda)^{-(1-\lambda)}, \quad(2 \lambda)^{-2 \lambda}(1-2 \lambda)^{-(1-2 \lambda)}\right\}\right)^{1 /(1-\lambda)}=: \sqrt[n-\ell]{n} \cdot f(\lambda) $$ Заметим сразу, что $\sqrt[n-\ell]{n} \leq \sqrt[n]{n^{2}}$, поскольку $\ell \leq n / 2$. Мы покажем, что максимальное значение $f_{0}$ функции $f$ при $\lambda \in(0,1 / 2]$ меньше 2 , и что его и можно принять за требуемую величину $2-c$. Начнем с последнего утверждения. По задаче 37 у каждого многочлена $P_{s}(x)$ при $s=n 2^{q}$ есть корень, не меньший $\alpha$; значит. по доказанному выше $\alpha \leq \sqrt[s]{s^{2}} \cdot f_{0}$. Если устремить показатель $q$ к бесконечности, то $\sqrt[s]{s^{2}} \rightarrow 1$; значит, $\alpha \leq f_{0}$. Осталось показать, что $f_{0}<2$. Заметим, что $f(\lambda)=\min \{g(\lambda), h(\lambda)\}$, где $$ \left.g(\lambda)=\left((2 \lambda)^{-\lambda}(1-\lambda)^{-(1-\lambda)}\right)^{1 /(1-\lambda)} \quad \text { и } \quad h(\lambda)=\left((2 \lambda)^{-2 \lambda}(1-2 \lambda)^{-(1-2 \lambda)}\right)\right)^{1 /(1-\lambda)} $$ Исследуем каждую из этих функций по отдельности. Производная логарифма функции $g(x)$ равна $$ (\ln g(\lambda))^{\prime}=\left(-\frac{\lambda}{1-\lambda} \ln 2 \lambda-\ln (1-\lambda)\right)^{\prime}=-\left(\frac{\lambda}{1-\lambda}\right)^{\prime} \cdot \ln 2 \lambda=-\frac{\ln 2 \lambda}{1-\lambda}-\frac{\lambda \ln 2 \lambda}{(1-\lambda)^{2}}>0 $$ значит, $g$ монотонно возрастает на нашем интервале. При этом $g(1 / 2)=2$. Итак, для любого $\lambda_{0}<$ $1 / 2$ имеем $g\left(\lambda_{0}\right)<2$. С другой стороны, при $\lambda \rightarrow 1 / 2-0$ имеем $h(\lambda) \rightarrow(1 \cdot 1)^{2}=1$, поскольку $x^{x} \rightarrow 1$ при $x \rightarrow+0$. Значит, существует такое $\lambda_{0}<1 / 2$, что $h(\lambda)<3 / 2$ при всех $\lambda \in\left[\lambda_{0}, 1 / 2\right]$. Итак, при всех $\lambda \in(0,1 / 2]$ имеем $f(\lambda) \leq \max \left\{g\left(\lambda_{0}\right), 3 / 2\right\}<2$, то есть и $f_{0}<2$. Замечание 6 Можно показать, что функция $h$ убывает в достаточно большой левой полуокрестности точки $1 / 2$. В этой полуокрестности есть точка $\lambda_{0}$, в которой $g\left(\lambda_{0}\right)=h\left(\lambda_{0}\right)$; это значит, что $f_{0}=g\left(\lambda_{0}\right)=h\left(\lambda_{0}\right)$. Компьютерные вычисления показывают, что $\lambda_{0}=0,435954877 \ldots$ и $f_{0}=1,971043304 \ldots$ ## ЧАСТЬ II ## РЕШЕНИЯ ЗАДАЧ ## ЗАДАЧИ ПОСЛЕ ПРОМЕЖУТОЧНОГО ФИНИША. ## Упаковки «больших» продуктов Задача 40. Например, $$ \begin{aligned} & \left\{x_{1} x_{2}\left(x_{3}+1\right)\right. \\ & \left.\left(x_{1}+x_{2}\right) x_{3}\right\} \end{aligned} $$ Задача 41. Например, $$ \begin{aligned} & \left\{\left(x_{1}+1\right)\left(x_{2}+1\right)\left(x_{3}+x_{4}\right) x_{5} x_{6} x_{7} x_{8}\right. \\ & x_{1} x_{2}\left(x_{3}+1\right)\left(x_{4}+1\right)\left(x_{5}+x_{6}\right) x_{7} x_{8} \\ & x_{1} x_{2} x_{3} x_{4}\left(x_{5}+1\right)\left(x_{6}+1\right)\left(x_{7}+x_{8}\right) \\ & \left(x_{1}+x_{2}\right) x_{3} x_{4} x_{5} x_{6}\left(x_{7}+1\right)\left(x_{8}+1\right) \\ & x_{1} x_{2}\left(x_{3}+x_{4}\right)\left(x_{5}+x_{6}\right)\left(x_{7}+x_{8}\right) \\ & \left(x_{1}+x_{2}\right) x_{3} x_{4}\left(x_{5}+x_{6}\right)\left(x_{7}+x_{8}\right) \\ & \left(x_{1}+x_{2}\right)\left(x_{3}+x_{4}\right) x_{5} x_{6}\left(x_{7}+x_{8}\right) \\ & \left.\left(x_{1}+x_{2}\right)\left(x_{3}+x_{4}\right)\left(x_{5}+x_{6}\right) x_{7} x_{8}\right\} \end{aligned} $$ Задача 42. Например, ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-164.jpg?height=673&width=1029&top_left_y=1296&top_left_x=510) Задача 43. Пусть существуют $s_{1}$-длинная упаковка больших $\left(m_{1}, t_{1}, r_{1}\right)$-продуктов и $s_{2}$-длинная упаковка больших $\left(m_{2}, t_{2}, r_{2}\right)$-продуктов. Занумеруем переменные, которые входят в большие продукты этих упаковок, так, чтобы множество переменных первой упаковки не пересекалось со множеством переменных второй упаковки. После этого устроим «прямое произведение» упаковок, т. е. каждый продукт первой упаковки умножим на каждый продукт второй упаковки. Несложно проверить, что получится упаковка с требуемыми параметрами, т. е. $s_{1} s_{2}$-длинная упаковка больших $\left(m_{1}+m_{2}, t_{1}+t_{2}, r_{1}+r_{2}\right)$-продуктов. Задача 44. Пусть есть $P$ - упаковка ( $n, k$ )-продуктов. К каждому продукту упаковки $P$ допишем в качестве множителей типа а) недостающие переменные, которых для каждого продукта в точности $n-2 k$. Суммарное число множителей в каждом продукте станет равным $k+(n-2 k)=n-k$, все они типов а) или б). Поэтому мы получили совокупность $P^{\prime}$ больших ( $n, n-k$ )-продуктов в том же количестве, что и в упаковке $P$. Покажем, что совокупность $P^{\prime}$ является упаковкой. Рассмотрим два произвольных разных больших продукта $p_{1}$ и $p_{2}$ из $P^{\prime}$. Сопоставим множителям, входящим в эти большие продукты, вершины двудольного графа $G$ : множителям большого продукта $p_{1}$ - вершины левой доли $G$, множителям большого продукта $p_{2}$ - вершины правой доли $G$. Две вершины графа $G$ соединим ребром, если соответствующие вершинам множители содержат общую переменную. Тогда вершины $G$, соответствующие множителям типа а) имеют степень 1 , а вершины, соответствующие множителям типа б), имеют степень 2 . Докажем, что разложения больших продуктов $p_{1}$ не содержат общих мономов. Предположим противное: такой моном есть. Это равносильно тому, что в графе $G$ существует совершенное паросочетание (мощности $n-k$ ). Удалим из $G$ все ребра этого паросочетания. После этой процедуры степени всех вершин уменьшатся на 1 и, таким образом, из каждой вершины, соответствующей множителю типа б), будет выходить ровно по одному ребру, из других вершин ребер выходить не будет. Несложно понять, что множество оставшихся ребер будет задавать общий моном двух исходных продуктов из $P$, чего быть не может, потому что $P$ - упаковка. Полученное противоречие доказывает, что и $P^{\prime}$ является упаковкой. Задача 45. Любой большой продукт из упаковки содержит $m$ переменных. Поэтому для него сумма числа множителей типа а), удвоенного числа множителей типа б) и числа множителей типа в) равна $m$. По определению для любого продукта из упаковки сумма числа множителей типа а) и числа множителей типа б) равна $t$. Вычитая из первого равенства второе, получаем, что для любого продукта из упаковки сумма числа множителей типа б) и числа множителей типа в) равна $m-t$. Заметим, что множители типов б) и в) содержат по два слагаемых, в отличие от множителей типа а), поэтому число мономов в декомпозиции любого большого продукта упаковки равно в точности $2^{m-t}$, а суммарное число мономов в декомпозициях всех больших продуктов упаковки равно в точности $2^{m-t+r}$. Всего мономов от $m$ переменных в точности $2^{m}$. Поэтому, если декомпозиции всех больших продуктов упаковки содержат не все возможные мономы от $m$ переменных, то $2^{m-t+r}<2^{m}$, отсюда $t>r$, что и требовалось доказать. Если декомпозиции всех больших продуктов упаковки содержат в совокупности все возможные мономы от $m$ переменных, то они содержат и моном 1 , который мог появиться только если в каком-то большом продукте упаковки все множители относятся к типу в). Но тогда все возможные мономы появляются при декомпозиции одного этого большого продукта. Отсюда длина упаковки равна $s=1=2^{0}$, что невозможно по условию задачи. Задача 46. Пусть $\mathcal{B}_{m-n, i}$ - произвольные $S_{m-n, i}$-длинные упаковки больших ( $\left.m-n, i\right)$-продуктов, $i=t-n, \ldots, t-\left\lceil\frac{n}{2}\right\rceil$, заданные на множестве переменных $x_{1}, \ldots, x_{m-n}$, а $\mathcal{P}_{n, k}$ - упаковки $(n, k)$ продуктов мощности соответственно $A_{n, k}, k=0, \ldots,\left\lfloor\frac{n}{2}\right\rfloor$, заданные на множестве переменных $x_{m-n+1}, \ldots, x_{n}$. Для каждого $k=0, \ldots,\left\lfloor\frac{n}{2}\right\rfloor$, следуя решению задачи 44 , превратим упаковку $\mathcal{P}_{n, k}$ в упаковку $\mathcal{P}_{n, n-k}^{\prime}$ той же мощности $A_{n, k}$ больших ( $\left.n, n-k\right)$-продуктов, после чего устроим «прямое произведение» упаковки $\mathcal{B}_{m-n, t-n+k}$ на упаковку $\mathcal{P}_{n, n-k}^{\prime}$, т. е. каждый продукт первой упаковки умножим на каждый продукт второй упаковки. После этого объединим полученные для всех $k=0, \ldots,\left\lfloor\frac{n}{2}\right\rfloor$ большие ( $\left.m, t\right)$-продукты в одну совокупность. Эта совокупность является упаковкой, потому что разложения полученных для разных $k$ больших продуктов содержат разное количество переменных из множества $\left\{x_{m-n+1}, \ldots, x_{n}\right\}$ и поэтому не могут содержать общие мономы, а при одном $k$ в разложениях больших продуктов все мономы разные, потому что исходные $\mathcal{B}_{m-n, i}$ и $\mathcal{P}_{n, n-k}^{\prime}$ являлись упаковками. Задача 47. Заметим, что последовательность $w_{\ell}=X_{\max }^{\ell}$ удовлетворяет тому же рекуррентному уравнению $$ w_{\ell+n}=\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} w_{\ell+k} $$ (на самом деле, это верно для любого корня многочлена $P(x)$ вместо $X_{\max }$ ). Поскольку числа $u_{1}, u_{2}, \ldots, u_{n}$ положительны, существуют такие положительные константы $C_{1}$ и $C_{2}$, что неравенства $C_{1} w_{\ell} \leq u_{\ell} \leq C_{2} w_{\ell}$ выполнены при всех $\ell=1,2, \ldots, n$. Покажем теперь индукцией по $\ell$, что эти неравенства выполнены и при всех натуральных $\ell$. База при $\ell \leq n$ уже проверена. Для шага индукции при $\ell \geq n+1$ достаточно заметить, что $$ u_{\ell}=\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} u_{\ell-n+k} \geq \sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k 1} w_{\ell-n+k}=C_{1} w_{\ell} $$ неравенство $u_{\ell} \leq C_{2} w_{\ell}$ проверяется аналогично. Замечание 7 То же рассуждение работает, если известно лишь, что в последовательности $\left\{u_{\ell}\right\}$ есть $n$ положительных членов подряд. Задача 48. Устроим рекурсивную конструкцию на основе неравенства из задачи 46 при $n=2$. Начнём с некоторых натуральных $m_{0}, t_{0}$ и некоторых непустых семейств $s\left(m_{0}, t_{0}+i\right)$-длинных упаковок $\mathcal{B}_{m_{0}, t_{0}+i}, i=0,1$. Мы последовательно построим упаковки $B_{m_{0}+2 d, t_{0}+d+i}, i=0,1$, следующим образом. На $d$-м шаге упаковка $B_{m_{0}+2 d, t_{0}+d+1}$ получается из (уже построенных) упаковок $B_{m_{0}+2(d-1), t_{0}+(d-1)+i}, i=0,1$, при помощи конструкции из решения задачи 46. Упаковка же $B_{m_{0}+2 d, t_{0}+d}$ получается из $B_{m_{0}+2(d-1), t_{0}+d}$ дописыванием множителей типа в) для двух добавленных переменных. Длины полученных упаковок удовлетворяют рекуррентному соотношению $$ s(-, t)=\sum_{k=0}^{1} A_{2, k} \cdot s(-, t-2+k)=s(-, t-2)+s(-, t-1) $$ с характеристическим многочленом $$ x^{2}-\sum_{k=0}^{1} A_{2, k} x^{k}=x^{2}-x-1 $$ В силу задачи 47 поведение величины $s(m, t)$ для некоторых $m$ (зависящих от $r$ ) и некоторой положительной константы $C$ удовлетворяет соотношению $$ s(m, t) \geq C X_{\max }^{t} $$ где $X_{\max }=1.6180 \ldots-$ наибольший корень многочлена (2). Отсюда $\log _{2} s(m, t) \geq t \log _{2} X_{\max }-\log _{2} C$. Отбрасывая «лишние» продукты до ближайшей снизу степени двойки и полагая $r=\left\lfloor\log _{2} s(m, t)\right\rfloor$, получаем при $t \rightarrow \infty$ последовательность $2^{r}$-длинных упаковок больших $(-, t)$-продуктов, для которых $\frac{t}{t+r} \rightarrow \frac{1}{1+\log _{2}\left(X_{\max }\right)}=0.5902 \ldots$ Задача 49. Пусть $\mathcal{B}_{t}$ - упаковка больших $(-, t)$-продуктов, построенная в решении задачи 48. Было доказано, что число больших продуктов в $\mathcal{B}_{t}$ не меньше чем $C \varphi^{t}$, где $\varphi=\frac{\sqrt{5}+1}{2}=1.6180 \ldots$ Достаточно доказать, что в упаковке $\mathcal{B}_{t}$ не более $C^{\prime} c^{t}$ больших продуктов (при некоторой $c<\varphi$ ), у которых число множителей типа б) больше чем $r$. Обозначим через $\rho(t, r)$ количество больших продуктов с больше чем $r$ множителями в упаковке $\mathcal{B}_{t}$. Тогда из конструкции легко видеть, что имеет место рекуррентное соотношение $\rho(t, r)=\rho(t-2, r)+\rho(t-1, r-1)$. Определим следующие константы: $d=\log _{2} \varphi, A=\frac{3-d}{2}, \alpha=\frac{\ln \frac{3-d}{1-d}}{2 \ln A}, \beta=1-\alpha$. Докажем оценку $$ \rho(t, r) \leq \gamma A^{\alpha t+\beta r} $$ для некоторого положительного $\gamma$ и всех $0 \leq r \leq t$. Доказательство индукцией по $t$. Выберем $\gamma$ так, чтобы при $t=t_{0}, t_{0}+1$ оценка (3) была верна - это база индукции. Шаг индукции. Пусть оценка (3) верна для $t^{\prime}0$. Покажем, что для упаковки $\mathcal{B}_{m, t}$ при $j \geq(2 / 3+\varepsilon) t(1+o(1))$, начиная с некоторого $t$ выполнено $$ \frac{l_{j-2}(t+2)}{l_{j}(t)}>15 $$ Для произвольного большого продукта $\alpha$ упаковки $\mathcal{B}_{m, t}$ обозначим через $n_{i}(\alpha), i=0,1,2,3,4,5$, число кусков переменных длины 10 в продукте $\alpha$ (не считая префиксного куска, с которого начинает действовать конструкция, и кусков, состоящих только из множителей типа в)), содержащих в точности $i$ множителей типа б). Пусть $j_{0}(\alpha)$ - число множителей типа б) в префиксе $\alpha$ (с которого начала действовать конструкция), а $t_{0}(\alpha)$ - суммарное число множителей типов а) и б) в префиксе $\alpha$. Для соотношения числа $j(\alpha)$ числа множителей типа б) к суммарному числу $t(\alpha)$ множителей типов а) и б) в строке $\alpha$ имеем $$ \frac{j(\alpha)}{t(\alpha)}=\frac{5 n_{5}(\alpha)+4 n_{4}(\alpha)+3 n_{3}(\alpha)+2 n_{2}(\alpha)+n_{1}(\alpha)+j_{0}(\alpha)}{5 n_{5}(\alpha)+6 n_{4}(\alpha)+7 n_{3}(\alpha)+8 n_{2}(\alpha)+9 n_{1}(\alpha)+10 n_{0}(\alpha)+t_{0}(\alpha)} $$ Нас интересуют только такие большие продукты $\alpha$ из упаковки $\mathcal{B}_{m, t}$, для которых, начиная с некоторого $t$, выполнено $\frac{j(\alpha)}{t(\alpha)}>\frac{2}{3}+\varepsilon^{\prime}, 0<\varepsilon^{\prime}<\varepsilon$. Поэтому, исходя из (6), можно считать, что $\min _{\alpha \in \mathcal{B}_{m, t}} n_{5}(\alpha) \rightarrow \infty$ при $t \rightarrow \infty$ и, начиная с некоторого $t$, для любого большого продукта $\alpha$ из упаковки $\mathcal{B}_{m, t}$ выполнено $n_{5}(\alpha)>n_{3}(\alpha)+1$. Обозначим через $S\left(t, j, n_{5}\right)$ множество больших продуктов из упаковки $\mathcal{B}_{m, t}$, содержащих в точности $j$ множителей типа б) и в точности $n_{5}$ кусков длины 10 , состоящих из 5 множителей типа б). Для заданных $j$ и достаточно большого $t$ для всех значений $n_{5}$, для которых множество $S\left(t, j, n_{5}\right)$ непусто, заменим в каждом большом продукте $\alpha$ из $S\left(t, j, n_{5}\right)$ один из кусков длины 10 с 5 множителями типа б) на допустимый кусок длины 10 с 3 множителями типа б). Это можно сделать $15 n_{5}$ способами. Мы получим большой продукт из упаковки $\mathcal{B}_{m, t+2}$, содержащий в точности $j-2$ множителей типа б), который мог быть получен таким способом из $n_{3}(\alpha)+12 / 3+\varepsilon$. Используя неравенство (5), имеем $$ \frac{\sum_{j=k_{0}}^{t} l_{j}(t)}{s(t)}<\frac{\sum_{j=k_{0}-2 d}^{t-2 d} l_{j}(t+2 d)}{15^{d} s(t)}<\frac{s(t+2 d)}{15^{d} s(t)} $$ С учётом задачи 47, для бесконечно многих $t$ выполнено неравенство $\frac{s(t+2 d)}{s(t)} \leq 2 X_{\max }^{2 d}$; при этих $t$ имеем $$ \frac{\sum_{j=k_{0}}^{t} l_{j}(t)}{s(t)}<2\left(\frac{X_{\max }^{2}}{15}\right)^{d} \rightarrow 0, \quad d \rightarrow \infty $$ Таким образом, мы показали, что число больших продуктов с числом множителей типа б), асимптотически превосходящим $t \log _{2} X_{\max }=0.7274 \ldots$, действительно часто бывает мало по сравнению с $s(t)$. (На самом деле, можно показать, что оно будет мало всегда, начиная с некоторого $t$.) Следовательно, удаляя из упаковки $\mathcal{B}_{m, t}$, построенной в решении задачи 51 , все большие продукты с числом множителей типа б), превосходящим указанную границу, мы все равно получим последовательность $2^{r}$-длинных упаковок больших $(m, t, r)$-продуктов, для которой $\frac{t}{t+r} \rightarrow \frac{1}{1+\log _{2}\left(X_{\max }\right)}=0.5789 \ldots$ ## Список литературы [1] Tarannikov Y. V. Generalized proper matrices and constructing of $m$-resilient Boolean functions with maximal nonlinearity for expanded range of parameters. - Siberian Electronic Mathematical Reports, 2014. - V. 11, pp. 229-245 (http://semr.math.nsc.ru/v11/p229-245.pdf). [2] Sauskan A. V., Tarannikov Yu. V. On packings of (n,k)-products. - Siberian Electronic Mathematical Reports, 2016. - V. 13, pp. 888-896 (http://semr.math.nsc.ru/v13/p888-896.pdf). [3] Агаханов Н. Х., Богданов И. И., Кожевников П. А., Подлипский О. К., Терешин Д. А. Всероссийские олимпиады школьников по математике 1993-2006: Окружной и финальный этапы. Под ред. Н. Х. Агаханова. - М.: МЦНМО, 2007, Задача 447, с. 58, 276-277. (http://www.math.ru/lib/files/pdf/olimp/Vseross.pdf) [4] Таранников Ю. В. Несократимые разложения однородных произведений двучленов для построения $m$-устойчивых функций с максимально возможной нелинейностью // Проблемы теоретической кибернетики. Материалы XII международной конференции (Казань, 16-20 июня 2014 г.). - Казань: Отечество, 2014. - pp. 271-272. (http://agora.guru.ru/ptk2014/files/ptk2014.pdf) [5] Таранников Ю. В. О возможности построения $m$-устойчивых функций с оптимальной нелинейностью в рамках одного метода // Материалы XII Международного семинара «Дискретная математика и ее приложения» имени академика О. Б. Лупанова (Москва, МГУ, 20-25 июня 2016 г.), М.: Изд-во механико-математического факультета МГУ, 2016. - pp. 394-397. (https://istina.msu.ru/media/icollections/collectioneditorship/218/904/26389465/sbornik.pdf) # Series of tennis doubles matches and packings of products Yu. Tarannikov, I. Bogdanov, K. Kuyumzhiyan,
M. Lobanov, A. Moskalev, A. Sauskan, S. Tokarev.
PART I ## THE PROBLEMS TO SOLVE BEFORE THE INTERMEDIATE FINAL. ## Series of tennis doubles matches problem and its variations Problem 1 In a tennis club $n$ tennis players want to organize a series of doubles matches (a doubles match is a match in which a team of two players play against a team of the other two players) in a way that each player plays against each of the other in one match exactly. There are no restrictions on the ways of forming teams each time. For what values of $n$ is the series possible? Problem 1 can be formulated in set-theoretical language. Let $A$ be a set of tennis players. We will define match as an irregular set $\left\{\left\{a_{1}, a_{2}\right\}\left\{a_{3}, a_{4}\right\}\right\}$ two irregular disjoint pairs of elements of the set $A$. Here the elements $a_{1}, a_{2}, a_{3}, a_{4}$ belong to $A$ and are pairwise different. We will call a set of matches timetable. Then problem 1 explores the timetables in which for any two different elements $a$ and $b$ from $A$ there is exactly one match in which the elements $a$ and $b$ are in different pairs. Problem 2 In a tennis club $n$ tennis players want to organize a series of doubles matches in a way that each player doesn't play against each player in more than one match. There are no restrictions on the ways of forming teams each time. What is the largest number of matches that can be organized? Problem 2 explores the timetables in which for any two different elements $a$ and $b$ from $A$ there is no more than one match in which the elements $a$ and $b$ are in different pairs. For making a timetable of the series of matches an algebraic notation can be used: $$ \left(a_{1}+a_{2}\right)\left(a_{3}+a_{4}\right) $$ This means that a pair of players $a_{1}$ and $a_{2}$ plays against a pair of players $a_{3}$ and $a_{4}$. Let's define a cyclic timetable. Let's number the players with numbers from 0 to $n-1$. The timetable is called a "cyclic" one, if for any integer $i$ with any match $\{\{a, b\}\{c, d\}\}$ in the timetable there is also a match $\left\{\left\{a_{i}, b_{i}\right\}\left\{c_{i}, d_{i}\right\}\right\}, a_{i}=a+i(\bmod n)$. The numbers $b_{i}, c_{i}, d_{i}$ are defined similarly. For the convenience of formulating of the next problems we will call all the timetables that were described in problem 1 perfect timetables (which means that in perfect timetables for any different elements $a$ and $b$ from $A$ there is exactly one match in which the elements $a$ and $b$ are in different pairs, or, if we talk about tennis players, all the timetables in which each player plays against each of the other in exactly one match). Let's call perfect any number $n$ for which there is at least one perfect timetable. Problem 3 Construct a "cyclic" perfect timetable for any perfect $n$. Problem 4* Let's impose an additional condition: each tennis player can not play with each of the other tennis players in one team in more than one match (which in set-theoretical language means that any two elements are in the same pair in no more than one match). Construct perfect timetables with this additional condition for as many integer $n$ as possible (for arbitrary $n$ this problem is open). Problem 5 Suppose number $n$ is perfect. In a tennis club $n$ tennis players want to organize a series of doubles matches. The question is can one make a perfect timetable so that each player has a number of partners larger than $\frac{n}{4}$. Problem 6 Let's impose a condition: each player has the same partner exactly once and the same rival exactly twice. Make a timetable that meets this condition for $n=5,8,9$. Problem 7 (an open problem) For what $n$ is there a timetable described in problem 6? Let's call am m-match unordered set $\left\{\left\{a_{1}, \ldots, a_{m}\right\}\left\{b_{1}, \ldots, b_{m}\right\}\right\}$ of two unordered subsets (or teams) of a set $A$, where all elements $a_{1}, \ldots, a_{m}, b_{1}, \ldots, b_{m}$ belong $A$ and are pairwise different. Then 2-match is just a match. Let's call m-timetable a set of $m$-matches. Then 2 -timetable is just a timetable. Let's call perfect an $m$-timetable in which for any two different elements $a$ and $b$ from $A$ there is exactly one $m$-match in which elements $a$ and $b$ are in different teams. Problem 8 Prove that if there exists a perfect m-timetable for $n$ tennis players then $n-1$ is divided by $m^{2}$. Problem 9 For what values of $n$ is there a perfect $m$-timetable for $n$ tennis players if $m$ is even? Problem 10 Is there a perfect 3-timetable for 10 tennis players? Problem 11**( ) For what values of $n$ there exists a perfect $m$-timetable for $n$ tennis players if $m$ is odd? Problem 12 Let's make (exclusively for this problem) the following modifications in the definitions. In a match let's allow an element to be simultaneously in both pairs (which means that a player is allowed to play simultaneously for both pairs astride the net). A perfect timetable in this particular case must meet the condition: for any two not necessarily different elements $a$ and $b$ from $A$ there is exactly one match in which elements $a$ and $b$ are in different pairs (this means that in addition we demand that each player also plays against himself). Find all $n$ for which there is such perfect timetable. ## Packings of products Puzzle 13 What makes the mathematical expression interesting $$ \begin{array}{r} \quad\left(x_{1}+x_{2}\right)\left(x_{3}+x_{4}\right)\left(x_{5}+x_{6}\right)+\left(x_{1}+x_{2}\right)\left(x_{4}+x_{6}\right)\left(x_{8}+x_{9}\right)+\left(x_{1}+x_{2}\right)\left(x_{7}+x_{9}\right)\left(x_{8}+x_{10}\right)+ \\ +\left(x_{1}+x_{3}\right)\left(x_{2}+x_{5}\right)\left(x_{7}+x_{8}\right)+\left(x_{1}+x_{4}\right)\left(x_{5}+x_{7}\right)\left(x_{6}+x_{9}\right)+\left(x_{1}+x_{5}\right)\left(x_{2}+x_{3}\right)\left(x_{9}+x_{10}\right)+ \\ +\left(x_{1}+x_{6}\right)\left(x_{3}+x_{10}\right)\left(x_{4}+x_{8}\right)+\left(x_{1}+x_{7}\right)\left(x_{2}+x_{10}\right)\left(x_{5}+x_{6}\right)+\left(x_{1}+x_{10}\right)\left(x_{2}+x_{7}\right)\left(x_{3}+x_{4}\right)+ \\ +\left(x_{2}+x_{8}\right)\left(x_{3}+x_{7}\right)\left(x_{4}+x_{9}\right)+\left(x_{2}+x_{9}\right)\left(x_{5}+x_{10}\right)\left(x_{6}+x_{8}\right)+\left(x_{3}+x_{5}\right)\left(x_{4}+x_{9}\right)\left(x_{7}+x_{10}\right)+ \\ +\left(x_{3}+x_{5}\right)\left(x_{6}+x_{8}\right)\left(x_{7}+x_{10}\right)+\left(x_{3}+x_{8}\right)\left(x_{4}+x_{6}\right)\left(x_{5}+x_{9}\right)+\left(x_{4}+x_{7}\right)\left(x_{6}+x_{10}\right)\left(x_{8}+x_{9}\right) ? \end{array} $$ Let's call a product of variables a monomial. In all following problems we consider only monomials without repeating variables. Two monomials that are only different in the order of variables are considered the same. The series of tennis doubles matches problem can be reformulated in terms of algebraic expressions, notably in products of sums of pairs of variables. Let's put in correspondence to each of $n$ tennis players one of variables $x_{1}, x_{2}, \ldots, x_{n}$. The expression $\left(x_{i}+x_{j}\right)\left(x_{k}+x_{l}\right)$ will be put in correspondence to the tennis match in which a pair of players corresponding to the variables $x_{i}$ and $x_{j}$ play against a pair of players corresponding to the variables $x_{k}$ and $x_{l}$. Let's call such expressions ( $n, 2$ )-products $(2$ - because there are two teams in a tennis match, this parameter will be generalized in what follows). The requirement that no two players play against each other more than once means the following: after expanding (according to the distribution rule) brackets in ( $n, 2$ )-products that correspond to all matches, no monomial of length 2 appears more than once. The requirement that any two tennis players play against each other exactly once means that after expanding brackets in ( $n, 2$ )-products, corresponding to all matches, each monomial of length 2 (that consists of different variables, of which the order in a monomial is not important) appears exactly once. Now let's give a more general definition by making the quantity of brackets in product equal to an arbitrary fixed number $k$. Let $n, k \in \mathbf{Z}, 0 \leq 2 k \leq n$. We will call the following expression with brackets, that is a product of the binominals, $(n, k)$-product (or simply product) : $$ P=\prod_{i=1}^{k}\left(x_{i, 1}+x_{i, 2}\right) $$ where $x_{i, 1}, x_{i, 2}, i=1, \ldots, k$ are the nonrecurring variables from the set $\left\{x_{1}, \ldots, x_{n}\right\}$. The product and summation signs (that appear in the note of product) correspond to regular algebraic operations. ${ }^{12}$ The decomposition $(n, k)$-product $P$ is the set of $2^{k}$ monomials of length $k$, appearing after expanding brackets in product $P$ according to the rule of distribution. It is assumed that decomposition of $(n, 0)$ product is the monomial of length 0 . The sum of products $\sum_{i=1}^{s} P_{i}$ is called a packing, if the decompositions of any two products $P_{i}$ and $P_{j}, i \neq j$, do not contain common monomials. We will call number s of products in sum the length of packing of products. Let's define values $A_{n, k}$. $A_{n, k}$ is the maximum value of the length of the packing of $(n, k)$-products. We will call a packing of $(n, k)$-products perfect if each monomial of length $k$ appears in the decomposition of exactly one product. Problem 14 Prove that $A_{n, 0}=1$. Problem 15 Prove that $A_{n, 1}=\left\lfloor\frac{n}{2}\right\rfloor$. Problem 16 Prove that $A_{n, 2}=\binom{\frac{n}{2}}{2}$ for even $n$. Problem 17 Prove that $A_{n,\left\lfloor\frac{n}{2}\right\rfloor}=1$. Problem 18 Prove that $A_{n, \frac{n}{2}-1}=\frac{n}{2}$ for even $n$. Problem 19 Prove that $A_{n, k} \leq \frac{\binom{n}{k}}{2^{k}}$.[^18] Problem 20 (attention test) Prove that $A_{10,3}=15$. Problem 21 Prove that $A_{n, k} \leq\binom{ n}{2 k}$. Problem 22 Prove that for $00$, that the largest root of the polynomial $x^{n}-\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x^{k}$ for any $n$ is strictly less than $2-c$. Problem 39* (an open problem) Find the upper limit of the largest root of the polynomial $x^{n}-$ $\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x^{k}$ for $n \rightarrow \infty$. ## PART II ## FORMULATIONS OF PROBLEMS ## THE PROBLEMS TO SOLVE AFTER THE INTERMEDIATE FINAL. ## Packings of big products Let $m, t \in \mathbf{Z}, 0 \leq t \leq m$. We define a $\operatorname{big}(m, t)$-product as the product in which the factors are of the following three types: a) $x_{i}$; b) $\left(x_{i}+x_{j}\right)$, c) $\left(x_{i}+1\right)$; each of the variables $x_{1}, \ldots, x_{m}$ should appear in the expression exactly once; the total number of factors of types a) and b) should equal $t$. The decomposition of a big $(m, t)$-product $P$ is a set of monomials appearing after expanding the brackets in the product $P$ (the order of variables in the monomial does not matter). We say that a sum of big products $\sum_{i=1}^{s} P_{i}$ is a packing, if the decompositions of any two products $P_{i}$ and $P_{j}, i \neq j$, contain no common monomials. The length of the packing is the number $s$ of products in the sum. We say that a packing is $s$-long if its length equals $s$. If one or several parameters of a packing of big $(m, t)$-products are either unimportant to us or are determined in the construction procedure, then we will replace these parameters with dashes. For example, if we set up only the parameter $t$, then we will talk about a packing of big $(-, t)$-products. Let $m, r, t \in \mathbf{Z}, 0 \leq r \leq t \leq m$. We say that a big $(m, t)$-product is a big $(m, t, r)$-product (a new parameter appears!) if the number of factors of type b) does not exceed $r$. When $s=2^{r}$, the $s$-long packings of big $(m, t, r)$-products are especially important for applications. If for some $s$-long packing of big $(m, t, r)$-products the inequality $s>2^{r}$ holds, then we can simply delete "superfluous" big products. We say that a packing of big $(m, r, t)$-products is rigid if the variables are rigidly combined into pairs $\left(x_{i}, x_{j}\right)$ so that in every big products of the packing either the variables $x_{i}$ and $x_{j}$ form one factor of type b), or both are factors of type a), or both lie in factors of type c). In "tennis players" language, the property of rigidness can be formulated as follows: the tennis players, as it happens in usual tennis tournaments, are rigidly divided into pairs and one cannot change his or her partner. Problem 40 Construct a 2-long packing of big (3,2,1)-products. Problem 41 Construct a 2-long rigid packing of big (8,5,3)-products. Problem 42 Construct a 16-long rigid packing of big (12,6,4)-products. Problem 43 Prove that if there exist an $s_{1}$-long packing of big $\left(m_{1}, t_{1}, r_{1}\right)$-products and an $s_{2}$-long packings of big $\left(m_{2}, t_{2}, r_{2}\right)$-products, then there exists an $s_{1} s_{2}$-long packing of big $\left(m_{1}+m_{2}, t_{1}+t_{2}, r_{1}+r_{2}\right)$ products. Problem 44 Find a way to transform a packing of ( $n, k)$-products into a packing of big ( $n, n-k)$-products of the same length. The problem of constructing appropriate packings of big $(m, t, r)$-products with the least possible value of the ratio $\frac{t}{t+r}$ is important to the problem of construction of correlation-immune Boolean functions with the maximum possible nonlinearity. The interrelation of these two problems is too complicated to be discussed here. In the present project Boolean functions are not mentioned in the following and knowledge of their definitions is not needed. Problem 45 Prove that if $r>0$ then for any $2^{r}$-long packing of big $(m, t)$-products we have $\frac{t}{t+r}>1 / 2$. An open problem: Is it possible to construct a $2^{r}$-long packing of big $(m, t, r)$-products (with any choice of parameters $m, t$, and $r$ ) in order to achieve the value $\frac{t}{t+r}$ arbitrarily close to $1 / 2$ ? Note that the Problem 43 provides no advantage in the task of constructing $2^{r}$-long packings of big $(m, t, r)$-products minimizing the value of $\frac{t}{t+r}$. Indeed, $\frac{t_{1}+t_{2}}{t_{1}+t_{2}+r_{1}+r_{2}} \geq \min \left(\frac{t_{1}}{t_{1}+r_{1}}, \frac{t_{2}}{t_{2}+r_{2}}\right)$. Problem without number. Find a construction of packings of big $(-, t)$-products. The main aim is: to make the value $\frac{t}{t+r}$ as close to $1 / 2$ as possible. One possible approach: to find a recursive construction dealing with packings of $(n, k)$-products and some initial packings of big $(n, u)$-products. Denote by $S_{m, t}$ the maximal possible length of the packing of big $(m, t)$-products. Problem 46 Prove that for $m \geq t \geq n$ the values $S_{m, t}$ satisfy the following recurrent inequality $$ S_{m, t} \geq \sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} S_{m-n, t-n+k} $$ If one needs to construct a packing of big $(-, r, t)$-products and the parameter $r$ is important, then one can still use problem 46 , and after that delete all the big products in which the number of multipliers of type b) exceeds $r$. Problem 47 Let $n$ be a positive integer. Let the sequence $\left\{u_{\ell}\right\}$ consists of positive numbers and satisfies the condition $$ u_{\ell+n}=\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} u_{\ell+k} $$ for all positive integers $\ell$. Denote by $X_{\max }$ the largest root of the polynomial $P(x)=x^{n}-\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x^{k}$. Prove that there exist positive constants $C_{1}$ and $C_{2}$ such that $C_{1} X_{\max }^{\ell} \leq u_{\ell} \leq C_{2} X_{\max }^{\ell}$ for all sufficiently big $\ell$. Problem 48 Prove that for any $\varepsilon>0$ one can implement problem $46($ for $n=2)$ in order to construct $a 2^{r}$-long packing of big $(-, t)$-products such that $\frac{t}{t+r}<\frac{1}{1+\log _{2}\left(\frac{\sqrt{5}+1}{2}\right)}+\varepsilon=0.5902 \ldots+\varepsilon$. Hint. Notice that for $n=2$ the largest root of the polynomial $x^{n}-\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x^{k}$ equals $\frac{1+\sqrt{5}}{2}$. Problem 49 Prove that for any $\varepsilon>0$ one can construct a $2^{r}$-long packing of big $(-, t, r)$-products such that $\frac{t}{t+r}<\frac{1}{1+\log _{2}\left(\frac{\sqrt{5}+1}{2}\right)}+\varepsilon=0.5902 \ldots+\varepsilon$. Problem 50* Prove that for any $2^{r}$-long rigid packing of big $(-, t)$-products we have $\frac{t}{t+r}>\frac{1}{1+\log _{2}\left(\frac{\sqrt{5}+1}{2}\right)}=$ $0.5902 \ldots$ Problem 51* Construct a $2^{r}$-long packing of big $\left(-\right.$, t)-products such that $\frac{t}{t+r}<\frac{1}{1+\log _{2}\left(\frac{\sqrt{5}+1}{2}\right)}=0.5902 \ldots$ Problem 52* Construct a $2^{r}$-long packing of big $(-, t, r)$-products such that $\frac{t}{t+r}<\frac{1}{1+\log _{2}\left(\frac{\sqrt{5}+1}{2}\right)}=$ $0.5902 \ldots$ ## SOLUTIONS AND COMMENTS ## PART I ## THE PROBLEMS TO SOLVE BEFORE THE INTERMEDIATE FINAL. ## The problem on a series of tennis doubles matches and its variations Problem 1 was used at the Final round of the All-Russian mathematical olympiad in 1993. The author of the problem is S. Tokarev. Problem 1. Suppose that the described series of matches was organized. Then all the opponents of one tennis player can be divided into pairs, thus $n$ is odd. All $\frac{n(n-1)}{8}$ possible pairs of opponents can be divided into groups of four which played in one match. Hence, the number of these pairs is a multiple of 4 , which gives us $n-1=8 l$. Let's prove that for any natural $l$ the series of matches described in the problem is possible for $n=8 l+1$. For $l=1$ let's give to tennis players umbers from 1 to 9 and we explicitly produce the timetable: 1 and 7 against 2 and $3 ; 1$ and 9 against 4 and $5 ; 1$ and 5 against 6 and $7 ; 1$ and 3 against 8 and 9;2 and 3 against 4 and $5 ; 2$ and 4 against 6 and $8 ; 7$ and 8 against 6 and $9 ; 5$ and 7 against 4 and $8 ; 3$ and 9 against 2 and 6 . ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-179.jpg?height=667&width=662&top_left_y=1217&top_left_x=688) Let's consider now the arbitrary $n \equiv 1(\bmod 8)$. Let's single out one special tennis player and divide all the rest players into groups of eight. Combining the special tennis player with each of these groups of eight, we organize a timetable of matches in the groups of nine as shown above. Thus, the special tennis player has already played against every other player; any two players from one group of eight have also played against each other. What is left is to make happen that any two players from different groups of eight play against each other. For this purpose we rigidly subdivide players within each group of eight into pairs. For any two different groups of eight we will organize a match of every pair from one group against every pair from the other group. The required timetable of matches is constructed. Answer. $n \equiv 1(\bmod 8)$. Problem 2. Assume first that $n$ is even. Then each tennis player has $n-1$ potential opponents, this number being odd. In each match a player plays, he has two opponents. That's why each player can play no more than in $\frac{n-2}{2}$ matches. Summing over all players and dividing by 4 (since there are 4 players in a match), we get that the total number of matches is no more than $\frac{n(n-2)}{8}$. This number of matches can indeed be organized: let's divide rigidly the players into pairs and let each pair play with each of the others. Thus, for an even $n$ the maximum number of matches is $\frac{n(n-2)}{8}$. Assume now that $n$ is odd. By a similar reasoning, we get that the number of matches is no more than $\frac{n(n-1)}{8}$. This bound can be met only when $n \equiv 1(\bmod 8)$ and any two players play against each other exactly once. As we saw in problem 1 , such a timetable can be constructed for $n \equiv 1(\bmod 8)$. Thus, for such values of $n$ the maximum number of matches is $\frac{n(n-1)}{8}$. Now we deal with the remaining cases of odd $n$. First, we construct the schedules for these values of $n$. Let's single out one special tennis player and divide the rest into groups of eight and one incomplete group containing 2, 4, or 6 tennis players. Combining a special tennis player with each of these groups of eight, we organize a timetable of matches in the groups of nine, as shown in problem 1. Within the groups of eight and the incomplete group, we divide rigidly the players into pairs, and we organize a match between any two pairs from different groups. . Thus, any two players play against each other, unless they both lie in the incomplete group augmented by the special player. What is left is to organize some matches between the players from the incomplete group augmented by the special player; we call this group a remaining group. It contains $m$ players, where $m=3,5$, or 7 , and $m \equiv n(\bmod 8)$. Clearly, it is impossible to make each player of this group play against each of the others. Let's determine the minimum number of pairs of players which ill not play. Since $n$ is odd, each player will not play against an even number of other players; therefore, the total number of pairs of players which will not play cannot equal exactly 1 or 2 . Now we consider the possible values of $m$ separately. Let $m \equiv n \equiv 3(\bmod 8)$. In the remaining group, there are three pairs who didn't play against each other, and we cannot have fewer, as was shown above. So the maximum number of matches equals $\frac{\frac{n(n-1)}{2}-3}{4}=\frac{(n-3)(n+2)}{8}$. Let $m \equiv n(\bmod 8) \equiv 5(\bmod 8)$. In the remaining group we organize a match of 1 and 2 against 3 and 4. After this, in the remaining group there are 6 pairs of players who didn't play against each other. If there were one more match, then there would be only two such pairs, which is impossible. Thus the maximum number of matches equals $\frac{\frac{n(n-1)}{2}-6}{4}=\frac{(n-4)(n+3)}{8}$. Let $m \equiv n(\bmod 8) \equiv 7(\bmod 8)$. In the remaining group we organize the following matches; 2 and 4 against 6 and $7 ; 1$ and 3 against 5 and $6 ; 1$ and 5 against 2 and 7 . After that, the remaining group contains five pairs who didn't play against each other. If there were one more match, than there would be only one such pair, which is impossible. Thus the maximum number of matches equals $\frac{\frac{n(n-1)}{2}-5}{4}=\frac{(n-2)(n+1)}{8}-1$. Answer. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-180.jpg?height=256&width=617&top_left_y=1829&top_left_x=708) Problem 3. We know from problem 1 that a perfect timetable exists only for $n \equiv 1(\bmod 8)$. Let $n=8 l+1$. Number the players from 0 to $8 l$ and include into the timetable the matches 0 and 1 against $4 i+2$ and $4 i+4$ for all $i=0,1, \ldots, l-1$, along with their cyclic shifts. Then the differences between the opponents' numbers in the matches possess all the values between 1 and $4 l$ (each value is possessed once). Taking into account the cyclic shifts, we conclude that any two players are opponents in exactly one match. Problem 4. Arrange the players in a circle and number them consecutively by the residue classes modulo $n=8 l+1$. In case $l=1$ let's take a match $\{\{0,1\}\{2,4\}\}$ and all its cyclic shifts. In the following examples, when we include a match in the timetable we also include all its cyclic shifts. | $l=2$ | $l=3$ | $l=4$ | $l=5$ | $l=6$ | $l=7$ | | :--- | :--- | :--- | :--- | :--- | :--- | | $\{\{0,1\}\{2,8\}\}$ | $\{\{0,1\}\{2,12\}\}$ | $\{\{0,1\}\{2,18\}\}$ | $\{\{0,1\}\{2,20\}\}$ | $\{\{0,1\}\{2,26\}\}$ | $\{\{0,1\}\{2,27\}\}$ | | $\{\{0,2\}\{5,13\}\}$ | $\{\{0,2\}\{5,10\}\}$ | $\{\{0,2\}\{5,14\}\}$ | $\{\{0,2\}\{5,17\}\}$ | $\{\{0,2\}\{5,22\}\}$ | $\{\{0,2\}\{5,36\}\}$ | | | $\{\{0,3\}\{7,19\}\}$ | $\{\{0,3\}\{7,25\}\}$ | $\{\{0,3\}\{7,32\}\}$ | $\{\{0,3\}\{7,21\}\}$ | $\{\{0,3\}\{7,17\}\}$ | | | | $\{\{0,4\}\{10,24\}\}$ | $\{\{0,4\}\{10,18\}\}$ | $\{\{0,4\}\{10,37\}\}$ | $\{\{0,4\}\{10,28\}\}$ | | | | | $\{\{0,5\}\{13,30\}\}$ | $\{\{0,5\}\{13,35\}\}$ | $\{\{0,5\}\{13,25\}\}$ | | | | | | $\{\{0,6\}\{15,38\}\}$ | $\{\{0,6\}\{15,41\}\}$ | | | | | | $\{\{0,7\}\{18,45\}\}$ | | We define the distance between a pair of players $a$ and $b$ as the minimal distance between them along the circle; thus, the distance between the neighboring players is 1 . Notice that the distances between pairs of opponents in the table above take on each value from 1 and $4 l$ exactly once. This means that any pair of opponents appears in a timetable exactly once. On the other hand, the distances between the pairs of partners (i.e. a pair of team-mates) are all different. This means that each tennis player has the same partner no more than once. Remark 2 During the conference in problem 4 V. Retinsky and D. Zakharov invented a construction that allows to generate many examples for new values of parameter $n$. It seems that this construction provides examples for infinitely many values of $n$ but up to now we are aware of no proof for this fact. Problem 5. Number all the players by the residue classes modulo $n=8 l+1$. Take the cyclic schedule generated by the following matches. $$ \begin{aligned} & \{\{0,1\}\{l+1,3 l+1\}\} \\ & \{\{0,3\}\{l+2,3 l+2\}\} \\ & \{\{0,5\}\{l+3,3 l+3\}\} \\ & \vdots \\ & \{\{0,2 l-5\}\{2 l-2,4 l-2\}\} \\ & \{\{0,2 l-3\}\{2 l-1,4 l-1\}\} \\ & \{\{0,2 l-1\}\{2 l, 4 l\}\} \end{aligned} $$ Again, the distances between pairs of opponents in the table above take on each value from 1 and $4 l$ exactly once. The distances between the pairs of team-mates take on the values from the set $\{1,3,5, \ldots, 2 l-5,2 l-3,2 l-1,2 l\}$. The number of these distances is $l+1$, and this means that the number of different team-mates of one player equals $2 l+2$, which is larger than $\frac{n}{4}$. Problem 6. Let $n=5$. Consider a timetable of a chess tournament. Now, each tour consists of two pairs, and we say that in our schedule these pairs play against each other. Let $n=8$. Number the seven players by the residues modulo 7 and denote the eighth player by $A$. For $k=0,1, \ldots, 6$, arrange a match $\{k, k+1\}$ against $\{k+3, A\}$ and a match $\{k+2, k+5\}$ against $\{k+4, k+6\}$. Let $n=9$. To each player we put into correspondence an entry of a $3 \times 3$ matrix. For each entry $(i, j)$, $i, j=1,2,3$, we perform the following. 1) Let the player $i j$ relax. 2) Let the pair from the row containing $i j$ play against the pair from the column containing $i j$. 3) Remove the row and column containing $i j$; in the remaining $2 \times 2$ matrix, let the pair of players in one diagonal play against the pair in the other diagonal. For example, if $i=2, j=3$, then the player 23 is relaxing, and we arrange the matches $\{21,22\}$ against $\{13,33\}$ and $\{31,12\}$ against $\{11,32\}$. Remark 3 During the conference in problem 7 V. Retinsky and D. Zakharov invented a construction that allows to generate the schedules for infinitely many values of $n$. Problem 8. Assume first that $m$ is odd. Then $n-1$ should to divisible by $m^{2}$. Indeed, the total number of pairs $n(n-1) / 2$ should be divisible by the number of pairs $m^{2}$ that appear in a match, i.e. $m^{2} \mid n(n-1) / 2$. On the other hand, any player in each match plays with a number of people divisible by $m$, i.e. $m \mid n-1$. Hence, $m^{2} \mid n-1$. If $m$ is even, then $n-1$ has to be a multiple of $2 m^{2}$. Indeed, the same reasoning shows that $m^{2}$ $n(n-1) / 2$ and $m \mid n-1$, but in our case these relations yield $2 m^{2} \mid n-1$. So, we have solved the problem and, moreover, we have got a stronger statement for even $m$ which we will use in the solution of problem 9 . Problem 9. In the solution of problem 8 we saw that $n-1$ is divisible by $2 m^{2}$ if $m$ is even. Now we show that a required schedule exists for all $n$ divisible by $2 m^{2}$, even regardless of the parity of $m$. Firstly, we present a cyclic timetable for $n=2 m^{2}+1$. It consists of the cyclic shifts of match $\{0,1,2, \ldots, m-1\}$ against $\left\{m, 2 m, 3 m, \ldots, m^{2}\right\}$. The distance between any two players is at most $m^{2}$, i.e. it equals $a m-b$ for some $1 \leq a \leq m$ and $0 \leq b \leq m-1$. Then, under a certain cyclic shift of our match, our players are the images of the players $b$ and $m a$, i.e. they will play. Now it suffices to show that the existence of perfect $m$-schedules for $n_{1}$ players and for $n_{2}$ players implies the existence of such schedule for $n_{1}+n_{2}-1$ players. Recall that $m \mid n_{1}-1$ and $m \mid n_{2}-1$. Implement both schedules on two sets of players with one common player. After that, it remains to make all the games between a player from the first $n_{1}-1$ ones and a player from the last $n_{2}-1$ ones. It is now easy: divide the set of $n_{1}-1$ players, as well as the set of $n_{2}-1$ players, into groups by $m$, and perform a match between each group from the first set against each group from the second set. Remark 4 The solution of problem 9 shows also the following fact. Assume that $m$ is odd. If for some even $n$ (such that $n-1$ is divisible by $m^{2}$ ) a required schedule exists, then such schedule also exists for all larger even $n$ (satisfying the same condition). Problem 10. If a perfect 3-timetable exists, then it contains $\binom{10}{2} / 3^{2}=5$ matches and each player participates in three of them. Consider the player 1; call three matches in which he participates bad. In the three bad matches, this player had at most 6 teammates. This means that there are three players 2 , 3, 4 which were not his teammates. Each of them participated in exactly one bad match (in the opposite group to 1), i.e. each of them participated in both non-bad matches. Moreover, all the pairwise meetings between the players $2,3,4$ were in non-bad matches. But in each of them there was an even number of such meetings, while their total number should be 3. A contradiction. Problem 12. Suppose the series of matches described in the problem was organized. Then all the opponents of one player can be divided into pairs, except two of them, who will be in a pair with this player when he plays against himself. Therefore, $n$ is odd. All the possible pairs of opponents are divided into groups of four pairs who played in one match. Hence, the number of these pairs $\frac{n(n-1)}{2}+n=\frac{n(n+1)}{2}$ is a multiple of four, which gives us $n+1=8 l$. For $l=1$ we number the from 0 to 6 . Let's include in the timetable the match 0 and 1 against 1 and 3. In this match the differences between opponents' numbers possess the values $0,1,2,3$ (each value is possessed once). So, if we also include in the timetable all the cyclic shifts of this match, than any two players (including those who coincide), are opponents in exactly one match. For arbitrary $l$ we number the players from 0 to $8 l-2$ and include in the timetable the matches 0 and 1 against $4 i+1$ and $4 i+3, i=0,1, \ldots, l-1$, along with their cyclic shifts. Then the differences between the opponents' numbers in the matches possess all the values between 0 and $4 l-1$ (each value is possessed once). Therefore, any two players (including those who coincide) are opponents in exactly one match. Answer. $n \equiv 7(\bmod 8)$. ## Packing of products Puzzle 13. Expanding the brackets, we get $15 \cdot 2^{3}=120$ monomials of length 3 , and all the monomials will be pairwise different. At the same time, there are exactly $\binom{10}{3}=120$ possible monomials of length 3 in 10 variables. Therefore, after expanding the brackets in puzzle 13 each possible monomial of length 3 in 10 variables appears exactly once. Hence puzzle 13 gives us an example of a perfect packing of $(10,3)$-products. Problem 14. There exists exactly one monomial of length 0 and we can get if from the only product that contains zero brackets. Problem 15. There exist exactly $n$ monomials of length 1 in $n$ variables. Each ( $n, 1$ )-product contains two such monomials, so $A_{n, 1} \leq\left\lfloor\frac{n}{2}\right\rfloor$. This bound is achieved if we take $\left\lfloor\frac{n}{2}\right\rfloor$ sums of different variables. Problem 16. Since $n$ is even, for each variable $x_{i}$ there is $n-1$ variable different from it. Consider any packing of $(n, 2)$-products. Consider the set of all $(n, 2)$-products in this packing which contain $x_{i}$ in one of the brackets. Consider the variables contained in the other bracket of such products; all these vaiables should be pairwise distinct, otherwise we will get the same monomial twice. That's why each of $n$ variables is contained in no more than $\left\lfloor\frac{n-1}{2}\right\rfloor=\frac{n}{2}-1$ products. Each (n,2)-product contains 4 variables, that's why $$ A_{n, 2} \leq \frac{1}{4} n\left(\frac{n}{2}-1\right)=\binom{\frac{n}{2}}{2} $$ At the same time it is possible to make a packing with this number of products. Let's combine rigidly $n$ variables into $\frac{n}{2}$ pairs and take a product of any two different pairs. We will get a packing made out of exactly $\binom{\frac{n}{2}}{2}$ products. Remark 5 In fact, problem 16 is equivalent to the case of even $n$ in problem 2. Problem 17. Let $k=\left\lfloor\frac{n}{2}\right\rfloor$. Consider any two $(n, k)$-products. Let's match each variable with a vertex of a graph. For each bracket of each of two products let's join the corresponding vertices by an edge. The obtained graph is partitioned into cycles of even length and (for odd $n$ ) one chain with an odd number of vertices. Let's choose every second vertex in each cycle. In the chain (if it exists), we also choose every second vertex (so that the border vertices are not chosen). Then the set of the chosen vertices gives a monomial appearing $\mathrm{n}$ the decompositions of both products. Thus, a packing cannot contain more than one product. Problem 18. Let $k=\frac{n}{2}-1$. Then each $(n, k)$-product doesn't contain exactly two from the given $n$ variables. We assert that for any two products of the packing the sets of variables that are not contained in these products are disjoint. Indeed, otherwise we would have two products on the $2 k+1$ variables, which by problem 17 cannot appear in a packing. Hence, $A_{n, \frac{n}{2}-1} \leq \frac{n}{2}$. On the other hand, there exists a packing with $k+1$ products. Let's partition rigidly $n$ variables into $\frac{n}{2}$ pairs and make all the possible products, that contain the sums of all pairs except one. It is easy to see that any monomial from the decomposition of one such product is not in the decomposition of any other. Hence, we get a packing made out of exactly $\frac{n}{2}$ products. Problem 19. In total, there are exactly $\binom{n}{k}$ different monomials of length $k$ of $n$ variables, the decomposition of each $(n, k)$-product gives $2^{k}$ such monomials. Hence, $A_{n, k} \leq \frac{\binom{n}{k}}{2^{k}}$. Problem 20. The upper bound follows from problem 19. The example of packing of length 15 is demonstrated in puzzle 13 . Problem 21. If the sets of variables in two $(n, k)$-products coincide, then by problem 17 their decompositions contain a common monomial. Hence, $A_{n, k} \leq\binom{ n}{2 k}$. Problem 22. Consider a packing of $(n, k)$-products of the maximal length. i. e. of length $A_{n, k}$. Take any product, consider the set of variables in it, and add to this set an arbitrary variable from the remaining ones (this can be performed in $n-2 k$ ways for every product). By problem 17 , all the sets of $2 k+1$ variables obtained in this manner are distinct. Hence, $A_{n . k}(n-2 k) \leq\binom{ n}{2 k+1}$, or $$ A_{n . k} \leq \frac{\binom{n}{2 k+1}}{n-2 k}=\frac{n!}{(2 k+1)!(n-2 k-1)!(n-2 k)}=\frac{\binom{n}{2 k}}{2 k+1} $$ Now, take any product, consider the set of variables in it, and remove one of them (this can be made in $2 k$ ways for every product). By the same problem, all such sets are also distinct. Hence $A_{n . k} \cdot 2 k \leq\binom{ n}{2 k-1}$, or $$ A_{n . k} \leq \frac{\binom{n}{2 k-1}}{2 k}=\frac{n!}{(2 k-1)!(n-2 k+1)!\cdot 2 k}=\frac{\binom{n}{2 k}}{n-2 k+1} $$ Problem 23. Let's combine rigidly the variables into $\left\lfloor\frac{n}{2}\right\rfloor$ pairs and let's take into a product the arbitrary combinations of $k$ of them. It is easy to see that no two such products have a common monomial in their decompositions. Hence, we get a packing made out of exactly $\binom{\left\lfloor\frac{n}{2}\right\rfloor}{ k}$ products. Problem 24. Single out a pair of variables $x_{n-1}$ and $x_{n}$. Now we construct the following set of products. Include in it all the products from a packing of $(n-2, k)$-products of maximum length. Also, take a packing of $(n-2, k-1)$-products of maximal length, add a new bracket $\left(x_{n-1}+x_{n}\right)$ to each of them, and add all obtained products to our new packing. One can easily see that the decompositions of the obtained products do not share any monomial. Hence, the required recurrent inequality follows. Problem 25. Divide the initial set into two subsets of $n_{1}$ and $n_{2}$ different variables, respectively. Now we construct the following packing of products: for each $i, 0 \leq i \leq k$, we multiply the packings corresponding to numbers $A_{n_{1}, i} \quad A_{n_{2}, k-i}$ and then sum up all obtained packings. Clearly, we obtain a new packing. Since the length of the packing obtained by multiplying two packings equals to their lengths' product, we get the required inequality. Problem 26. One can get the perfect packing of $(n-1, k-1)$-products from a perfect packing of $(n, k)$-products in the following way. Take any variable, say, $x_{n}$. Find all the products in the packing that contain $x_{n}$, and remove from each of them the bracket containing $x_{n}$. We get a set of $(n-1, k-1)$-products. It is not hard to see that their sum is the required packing. Problem 27. See the solution to problem 29. Problem 28. We will show that $n$ should be congruent to 2 modulo 8. Assume that there exists a perfect packing of $(8 k+i, 3)$-products. Then, by problem 26 , there exists a perfect packing of $(8 k+i-1,2)$ products, which is known to exist only for $8 l+1$ variables. Thus, $n=8 k+2$. On the other hand, by applying problem 27 to the perfect packing of (10,3)-products, we get that for all the numbers of the form $n=8 k+2$ there exists a perfect packing. Answer: $n=8 k+2$ for all positive integers $k$. Problem 29. Due to problem 26, for every $i=0,1, \ldots, k$ there exist perfect packings of $\left(n_{1}-i, k-i\right)$ and of $\left(n_{2}-i, k-i\right)$-products. Let $X, Y, Z$ be pairwise disjoint sets of variables such that $|X|=n_{1}-k$, $|Y|=n_{2}-k, Z=\left\{z_{0}, z_{1}, \ldots, z_{k}\right\}$. Construct the following collection of packings. For every $i=0,1, \ldots, k$ we take a perfect packing $M_{i}$ of $\left(n_{1}-k+i, i\right)$-products on the set of variables $X \cup\left\{z_{0}, \ldots, z_{i-1}\right\}$, along with a perfect packing $N_{i}$ of $\left(n_{2}-i, k-i\right)$-products on the set of variables $Y \cup\left\{z_{i+1}, \ldots, z_{k}\right\}$. Multiply each products in $M_{i}$ by each product in $N_{i}$; clearly, the obtained expressions form a packing of $\left(n_{1}+n_{2}-k+1, k\right)$-products on the set of variables $X \cup Y \cup Z$; denote this packing by $U_{i}$. We claim that the union $U$ of all these packings is a perfect packing of $\left(n_{1}+n_{2}-k+1, k\right)$-products. Formally,, $$ \bigsqcup_{i=0}^{k} U_{i}, \quad U_{i}=P\left(X \sqcup\left\{\bigsqcup_{j=0}^{i-1} z_{j}\right\}, i\right) \times P\left(Y \sqcup\left\{\bigsqcup_{j=i+1}^{k} z_{j}\right\}, k-i\right) $$ where $P(\ldots)$ are corresponding perfect packings. We start with showing that every monomial $m$ of degree $k$ appears in the decompositions of all products in $U$ at most once. Arguing indirectly, assume that $m$ appears in decompositions of two products lying in $U_{i}$ and $U_{j}, i0$. Hence, the polynomial $\widetilde{P}$ has the root larger than $\alpha$. Problem 34. For every $x \geq 2$, we have $$ x^{n} \geq(1+x / 2)^{n}=\sum_{k=0}^{n}\binom{n}{k}(x / 2)^{k}>\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor}\binom{n}{k}(x / 2)^{k} \geq \sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x^{k} $$ where the last inequality follows from problem 19. Hence, $x^{n}-\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x^{k}>0$ for any $x \geq 2$. Problem 35. It is clear that $P_{n}(0) \leq 0$, therefore the largest real root of the polynomial $P_{n}(x)$ is nonnegative. Denote $P_{n}^{+}(x)=x^{n}+\sum_{k=0}^{n-1} a_{n, k} x^{k}$. All $a_{n, k}$ are nonnegative, therefore the polynomial $P_{n}^{+}(x)$ has no positive roots. Hence, the largest real root of the polynomial $P_{n}^{\prime}(x)=P_{n}(x) P_{n}^{+}(x)$ is the same as that of the polynomial $P_{n}(x)$. We have $P_{n}^{\prime}(x)=P_{n}(x) P_{n}^{+}(x)=x^{2 n}-\left(\sum_{k=0}^{n-1} a_{n, k} x^{k}\right)^{2}=x^{2 n}-\sum_{k=0}^{2 n-1} u_{2 n, k} x^{k}$, where $u_{2 n, k}=$ $\sum_{i=0}^{k} a_{n . i} a_{n, k-i}$. By the problem condition, the concatenation of two phrases is also a phrase, so $a_{2 n, k} \geq$ $\sum_{i=0}^{k} a_{n . i} a_{n, k-i}$. Hence, $a_{2 n, k} \geq u_{2 n, k}$ for any $k$. Let $x_{0}$ be the largest (real) root of the polynomial $P_{n}(x)$; then $x_{0}$ is also the largest (real) root of the polynomial $P_{n}^{\prime}(x)$, i.e. $P^{\prime}\left(x_{0}\right)=0$. Since $a_{2 n, k} \geq u_{2 n, k} \geq 0$, by problem 33 the largest positive root of $P_{2 n}(x)$ is at least $x_{0}$. Puzzle 36. Answer. The inequality from problem 25. Problem 37. Assume that $x_{0}^{n}=\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x_{0}^{k}$. Then $$ x_{0}^{2 n}=\left(\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} x_{0}^{k}\right)^{2} \leq \sum_{k=0}^{n}\left(\sum_{i=0}^{k} A_{n, i} \cdot A_{n, k-i}\right) x_{0}^{k} \leq \sum_{k=0}^{n} A_{2 n, k} x_{0}^{k} $$ where the last inequality follows from problem 25. Hence, the inequality $x^{2 n}-\sum_{k=0}^{n} A_{2 n, k} x^{k} \leq 0$ also holds for $x=x_{0}$. Since the coefficient of the highest power is positive, for large enough $x$ the polynomial takes on positive values, hence it has a root on a ray $\left[x_{0}, \infty\right)$. Problem 38. We will use the following inequality. Lemma 1 The following inequality holds: $$ \binom{n}{k} \leq \frac{n^{n}}{k^{k}(n-k)^{n-k}} $$ Proof. Expand $n^{n}=(k+(n-k))^{n}$ and single out the $k$-term to obtain $$ n^{n}=\sum_{i=0}^{n}\binom{n}{k} k^{i}(n-k)^{n-i} \geq\binom{ n}{k} k^{k}(n-k)^{n-k} $$ as required. Now we implement Lemma 1 to bring the results of problems 19 and 21 to a more handy form (by weakening them a little). Denote $\lambda=k / n$. Problem 19 yields $$ A_{n, k} \leq \frac{\binom{n}{k}}{2^{k}} \leq \frac{n^{n}}{(\lambda n)^{\lambda n}((1-\lambda) n)^{(1-\lambda) n} \cdot 2^{\lambda n}}=\left((2 \lambda)^{-\lambda}(1-\lambda)^{-(1-\lambda)}\right)^{n} $$ Analogously, problem 21 implies $$ A_{n, k} \leq\binom{ n}{2 k} \leq\left((2 \lambda)^{-2 \lambda}(1-2 \lambda)^{-(1-2 \lambda)}\right)^{n} $$ Let $P_{n}(x)$ be our polynomial, and let $\alpha$ (or $\alpha_{n}$ ) be its largest root. Let $\ell$ be the index of the largest term in a sum $\sum_{k=0}^{\lfloor n / 2\rfloor} A_{n, k} \alpha^{k}$. Then the bounds above show that $$ \alpha^{n}=\sum_{k=0}^{\lfloor n / 2\rfloor} A_{n, k} \alpha^{k} \leq n A_{n, \ell} \alpha^{\ell} \leq n \alpha^{\ell} \cdot\left(\min \left\{(2 \lambda)^{-\lambda}(1-\lambda)^{-(1-\lambda)}, \quad(2 \lambda)^{-2 \lambda}(1-2 \lambda)^{-(1-2 \lambda)}\right\}\right)^{n} $$ where $\lambda=\ell / n$. Hence, extracting the root of the degree $n-\ell=(1-\lambda) n$, we get $$ \alpha \leq \sqrt[n-\ell]{n} \cdot\left(\min \left\{(2 \lambda)^{-\lambda}(1-\lambda)^{-(1-\lambda)}, \quad(2 \lambda)^{-2 \lambda}(1-2 \lambda)^{-(1-2 \lambda)}\right\}\right)^{1 /(1-\lambda)}=: \sqrt[n-\ell]{n} \cdot f(\lambda) $$ Notice here that $\sqrt[n-\ell]{n} \leq \sqrt[n]{n^{2}}$, since $\ell \leq n / 2$. We will show that the maximum value $f_{0}$ of function $f$ for $\lambda \in(0,1 / 2]$ is smaller than 2 , and that it can be regarded as the required value $2-c$. We start with the latter assertion. According to problem 37, each polynomial $P_{s}(x)$ with $s=n 2^{q}$ has a root not smaller than $\alpha$; hence the obtained bounds show $\alpha \leq \sqrt[s]{s^{2}} \cdot f_{0}$. As the parameter $q$ tends to infinity, we have $\sqrt[s]{s^{2}} \rightarrow 1$; hence, $\alpha \leq f_{0}$. It remains to prove that $f_{0}<2$. Notice that $f(\lambda)=\min \{g(\lambda), h(\lambda)\}$, where $$ \left.g(\lambda)=\left((2 \lambda)^{-\lambda}(1-\lambda)^{-(1-\lambda)}\right)^{1 /(1-\lambda)} \quad \text { and } \quad h(\lambda)=\left((2 \lambda)^{-2 \lambda}(1-2 \lambda)^{-(1-2 \lambda)}\right)\right)^{1 /(1-\lambda)} $$ We investigate each of these functions separately. The derivative of logarithm of function $g(x)$ equals $$ (\ln g(\lambda))^{\prime}=\left(-\frac{\lambda}{1-\lambda} \ln 2 \lambda-\ln (1-\lambda)\right)^{\prime}=-\left(\frac{\lambda}{1-\lambda}\right)^{\prime} \cdot \ln 2 \lambda=-\frac{\ln 2 \lambda}{1-\lambda}-\frac{\lambda \ln 2 \lambda}{(1-\lambda)^{2}}>0 $$ hence, $g$ is monotonically increasing on our interval. Moreover, $g(1 / 2)=2$. So, for any $\lambda_{0}<1 / 2$ we have $g\left(\lambda_{0}\right)<2$. On the other hand, as $\lambda \rightarrow 1 / 2-0$ we have $h(\lambda) \rightarrow(1 \cdot 1)^{2}=1$, since $x^{x} \rightarrow 1$ as $x \rightarrow+0$. Hence, there exists $\lambda_{0}<1 / 2$ such that $h(\lambda)<3 / 2$ for any $\lambda \in\left[\lambda_{0}, 1 / 2\right]$. So, for any $\lambda \in(0,1 / 2]$ we have $f(\lambda) \leq \max \left\{g\left(\lambda_{0}\right), 3 / 2\right\}<2$, i. e. we also have $f_{0}<2$. Remark 6 It can be shown that the function $h$ is decreasing in a large enough left semi-neighborhood of the point $1 / 2$. In this semi-neighborhood there is a point $\lambda_{0}$ such that $g\left(\lambda_{0}\right)=h\left(\lambda_{0}\right)$; this means that $f_{0}=g\left(\lambda_{0}\right)=h\left(\lambda_{0}\right)$. Computer calculations show that $\lambda_{0}=0,435954877 \ldots f_{0}=1,971043304 \ldots$ ## SOLUTIONS AND COMMENTS PART II ## THE PROBLEMS TO SOLVE AFTER THE INTERMEDIATE FINAL. ## Packings of big products Problem 40. For example, $$ \begin{aligned} & \left\{x_{1} x_{2}\left(x_{3}+1\right)\right. \\ & \left.\left(x_{1}+x_{2}\right) x_{3}\right\} \end{aligned} $$ Problem 41. For example, $$ \begin{aligned} & \left\{\left(x_{1}+1\right)\left(x_{2}+1\right)\left(x_{3}+x_{4}\right) x_{5} x_{6} x_{7} x_{8}\right. \\ & x_{1} x_{2}\left(x_{3}+1\right)\left(x_{4}+1\right)\left(x_{5}+x_{6}\right) x_{7} x_{8} \\ & x_{1} x_{2} x_{3} x_{4}\left(x_{5}+1\right)\left(x_{6}+1\right)\left(x_{7}+x_{8}\right) \\ & \left(x_{1}+x_{2}\right) x_{3} x_{4} x_{5} x_{6}\left(x_{7}+1\right)\left(x_{8}+1\right) \\ & x_{1} x_{2}\left(x_{3}+x_{4}\right)\left(x_{5}+x_{6}\right)\left(x_{7}+x_{8}\right) \\ & \left(x_{1}+x_{2}\right) x_{3} x_{4}\left(x_{5}+x_{6}\right)\left(x_{7}+x_{8}\right) \\ & \left(x_{1}+x_{2}\right)\left(x_{3}+x_{4}\right) x_{5} x_{6}\left(x_{7}+x_{8}\right) \\ & \left.\left(x_{1}+x_{2}\right)\left(x_{3}+x_{4}\right)\left(x_{5}+x_{6}\right) x_{7} x_{8}\right\} \end{aligned} $$ Problem 42. For example, $$ \begin{aligned} & \left\{x_{1} x_{2} x_{3} x_{4} x_{5} x_{6}\left(x_{7}+1\right)\left(x_{8}+1\right)\left(x_{9}+1\right)\left(x_{10}+1\right)\left(x_{11}+1\right)\left(x_{12}+1\right)\right. \\ & \left(x_{1}+x_{2}\right) x_{3} x_{4}\left(x_{5}+1\right)\left(x_{6}+1\right)\left(x_{7}+x_{8}\right) x_{9} x_{10}\left(x_{11}+1\right)\left(x_{12}+1\right) \\ & \left(x_{1}+x_{2}\right) x_{3} x_{4}\left(x_{5}+1\right)\left(x_{6}+1\right)\left(x_{7}+1\right)\left(x_{8}+1\right)\left(x_{9}+x_{10}\right) x_{11} x_{12} \\ & \left(x_{1}+x_{2}\right) x_{3} x_{4}\left(x_{5}+1\right)\left(x_{6}+1\right) x_{7} x_{8}\left(x_{9}+1\right)\left(x_{10}+1\right)\left(x_{11}+x_{12}\right) \\ & \left(x_{1}+1\right)\left(x_{2}+1\right)\left(x_{3}+x_{4}\right) x_{5} x_{6}\left(x_{7}+x_{8}\right) x_{9} x_{10}\left(x_{11}+1\right)\left(x_{12}+1\right) \\ & \left(x_{1}+1\right)\left(x_{2}+1\right)\left(x_{3}+x_{4}\right) x_{5} x_{6}\left(x_{7}+1\right)\left(x_{8}+1\right)\left(x_{9}+x_{10}\right) x_{11} x_{12} \\ & \left(x_{1}+1\right)\left(x_{2}+1\right)\left(x_{3}+x_{4}\right) x_{5} x_{6} x_{7} x_{8}\left(x_{9}+1\right)\left(x_{10}+1\right)\left(x_{11}+x_{12}\right) \\ & x_{1} x_{2}\left(x_{3}+1\right)\left(x_{4}+1\right)\left(x_{5}+x_{6}\right)\left(x_{7}+x_{8}\right) x_{9} x_{10}\left(x_{11}+1\right)\left(x_{12}+1\right) \\ & x_{1} x_{2}\left(x_{3}+1\right)\left(x_{4}+1\right)\left(x_{5}+x_{6}\right)\left(x_{7}+1\right)\left(x_{8}+1\right)\left(x_{9}+x_{10}\right) x_{11} x_{12} \\ & x_{1} x_{2}\left(x_{3}+1\right)\left(x_{4}+1\right)\left(x_{5}+x_{6}\right) x_{7} x_{8}\left(x_{9}+1\right)\left(x_{10}+1\right)\left(x_{11}+x_{12}\right) \\ & x_{1} x_{2}\left(x_{3}+1\right)\left(x_{4}+1\right)\left(x_{5}+x_{6}\right)\left(x_{7}+x_{8}\right)\left(x_{9}+x_{10}\right)\left(x_{11}+x_{12}\right) \\ & \left(x_{1}+x_{2}\right) x_{3} x_{4}\left(x_{5}+1\right)\left(x_{6}+1\right)\left(x_{7}+x_{8}\right)\left(x_{9}+x_{10}\right)\left(x_{11}+x_{12}\right) \\ & \left(x_{1}+1\right)\left(x_{2}+1\right)\left(x_{3}+x_{4}\right) x_{5} x_{6}\left(x_{7}+x_{8}\right)\left(x_{9}+x_{10}\right)\left(x_{11}+x_{12}\right) \\ & \left(x_{1}+x_{2}\right)\left(x_{3}+x_{4}\right)\left(x_{5}+x_{6}\right) x_{7} x_{8}\left(x_{9}+1\right)\left(x_{10}+1\right)\left(x_{11}+x_{12}\right) \\ & \left(x_{1}+x_{2}\right)\left(x_{3}+x_{4}\right)\left(x_{5}+x_{6}\right)\left(x_{7}+x_{8}\right) x_{9} x_{10}\left(x_{11}+1\right)\left(x_{12}+1\right) \\ & \left.\left(x_{1}+x_{2}\right)\left(x_{3}+x_{4}\right)\left(x_{5}+x_{6}\right)\left(x_{7}+1\right)\left(x_{8}+1\right)\left(x_{9}+x_{10}\right) x_{11} x_{12}\right\} \end{aligned} $$ Problem 43. Take an $s_{1}$-long packing of big $\left(m_{1}, t_{1}, r_{1}\right)$-products and an $s_{2}$-long packing of big $\left(m_{2}, t_{2}, r_{2}\right)$-products. Enumerate the variables of these packings so that no variable participates in both packings. After that we take a "direct product" of the two packings, multiplying every big product from the first packing by every big product from the second one. One can easily check that we obtan a packing with the required parameters, i.e., an $s_{1} s_{2}$-packing of big $\left(m_{1}+m_{2}, t_{1}+t_{2}, r_{1}+r_{2}\right)$-products. Problem 44. Let $P$ be a packing of $(n, k)$-products. For every product in $P$, take all $n-2 k$ variables that are absent in it and multiply the product by all those variables as factors of type a); we will get some big product in $n$ variables with $k+(n-2 k)=n-k$ factors of types a) and b) - i.e., a big ( $n, n-k)$-product. Performing this operation to all products in $P$, we get a collection $P^{\prime}$ of big $(n, n-k)$-products of the same length as $P$. Let us show that $P^{\prime}$ is a packing. Consider any two different big products $p_{1}$ and $p_{2}$ in $P^{\prime}$. Construct the following bipartite graph $G$ : the vertices of the left (resp., right) part correspond to the factors of $p_{1}$ (resp., $p_{2}$ ), and two vertices are connected with an edge if and only if the corresponding factor have a common variable. Thus, the vertices corresponding to factors of type a) and b) have degrees 1 and 2 , respectively. We need to show that the decompositions of $p_{1}$ and $p_{2}$ contain no common monomials. Assume the contrary. In the graph language, the existence of such monomial is equivalent to the existence of a perfect matching in $G$. Remove all edges of this matching from the graph; all degrees of vertices decrease by 1 , so after the removal each vertex of type b) has degree 1 , and each vertex of type a) has no edges adjacent to it. But then the set of remaining edges corresponds to a monomial that appears in both initial products (which were transformed to $p_{1}$ and $p_{2}$ ). This is impossible, since $P$ is a packing. The obtained contradiction shows that $P^{\prime}$ is a packing. Problem 45. Consider any big product $p$ in the packing; let $a, b$, and $c$ be the numbers of its factors of types a), b), and c), respectively. Since $p$ is a ( $m, t)$-product, we have $a+2 b+c=m$ and $a+b=t$; therefore, $b+c=(a+2 b+c)-(a+b)=m-t$. Notice now that each factor of types b) and c) contains two summands, while that of type a) contains only one summand. So, the decomposition of $p$ contains exactly $2^{b+c}=2^{m-t}$ monomials. Since the packing contains $2^{r}$ big products, the total number of monomials in their decompositions is $2^{m-t+r}$. Now, the total number of monomials in $m$ variables is $2^{m}$. So, if the decompositions of the big products from the packing contain not all possible monomials, then we obtain $2^{m-t+r}<2^{m}$, whence $t>r$, as required. Otherwise, one of these decompositions contains monomial 1 which could happen only if all factors in the big product are of type c). But then the decomposition of this big product already contains all possible monomials, so the packing contains ony one big product. THis is prohibited by the problem condition. Problem 46. Let $\mathcal{B}_{m-n, i}, i=t-n, \ldots, t-\left\lceil\frac{n}{2}\right\rceil$, be arbitrary $S_{m-n, i}$-long packings of big $(m-n, i)-$ products; the set of variables for each of them is $\left\{x_{1}, \ldots, x_{m-n}\right\}$. Let $\mathcal{P}_{n, k}, k=0, \ldots,\left\lfloor\frac{n}{2}\right\rfloor$, be arbitrary packings of $(n, k)$-products of length $A_{n, k}$, the set of variables for each of them is $\left\{x_{m-n+1}, \ldots, x_{n}\right\}$. For every $k=0, \ldots,\left\lfloor\frac{n}{2}\right\rfloor$, according to problem 44, we transform the packing $\mathcal{P}_{n, k}$ into an $A_{n, k}$-long packing $\mathcal{P}_{n, n-k}^{\prime}$ of big $(n, n-k)$-products. After that we again form a "direct product" of the packings $\mathcal{B}_{m-n, t-n+k}$ and $\mathcal{P}_{n, n-k}^{\prime}$, i.e., multiply each product of the former one by each product of the latter; we obtain a collection of big $(m, t)$-products. Finally, we take the union of all obtained collections over all $k=0, \ldots,\left\lfloor\frac{n}{2}\right\rfloor$. We claim that the obtained union is a packing of $(m, t)$-products. Indeed, consider any two different big products in this collection. If they lie in different "direct products", then the monomials in their decompositions contain different number of variables from $\left\{x_{m-n+1}, \ldots, x_{n}\right\}$, so these decompositions share no monomial. Otherwise, these decompositions have no common monomial, since $\mathcal{B}_{m-n, i}$ and $\mathcal{P}_{n, n-k}^{\prime}$ are packings. Problem 47. Notice that the sequence $w_{\ell}=X_{\max }^{\ell}$ satisfies the same recurrence relation $$ w_{\ell+n}=\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} w_{\ell+k} $$ (in fact, this is true for every root of the polynomial $P(x)$ in place of $X_{\max }$ ). Since the numbers $u_{1}, u_{2}, \ldots, u_{n}$ are positive, there exist some positive constants $C_{1}$ and $C_{2}$ such that the inequalities $C_{1} w_{\ell} \leq u_{\ell} \leq C_{2} w_{\ell}$ hold for every $\ell=1,2, \ldots, n$. Now we use the induction to show that these inequalities hold for all positive integer $\ell$; the base case $\ell \leq n$ has been already established. To prove the induction step, for $\ell \geq n+1$ we have $$ u_{\ell}=\sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k} u_{\ell-n+k} \geq \sum_{k=0}^{\left\lfloor\frac{n}{2}\right\rfloor} A_{n, k 1} w_{\ell-n+k}=C_{1} w_{\ell} $$ the inequality $u_{\ell} \leq C_{2} w_{\ell}$ can be checked similarly. Remark 7 The same arguments work if we know only that the sequence $\left\{u_{\ell}\right\}$ contains $n$ positive elements in a row. Problem 48. We arrange a recursive construction based on the inequality from problem 46 for $n=2$. Fix any positive integers $m_{0}, t_{0}$ and take any $s\left(m_{0}, t_{0}+i\right)$-long packings $\mathcal{B}_{m_{0}, t_{0}+i}$ of big $\left(m_{0}, t_{0}+i\right)$-products for $i=0,1$. We will successively construct packings $B_{m_{0}+2 d, t_{0}+d+i}, i=0,1$, as follows. On the $d$ th step, the packing $B_{m_{0}+2 d, t_{0}+d+1}$ is obtained from (already constructed) packings $B_{m_{0}+2(d-1), t_{0}+(d-1)+i}$, $i=0,1$, using the construction from the solution of problem 46. The packing $B_{m_{0}+2 d, t_{0}+d}$ is obtained from $B_{m_{0}+2(d-1), t_{0}+d}$ by augmenting its products with factors of type c) containing two new variables. The lengths of the obtained packings satisfy the recurrence relation $$ s(-, t)=\sum_{k=0}^{1} A_{2, k} \cdot s(-, t-2+k)=s(-, t-2)+s(-, t-1) $$ with a characteristic polynomial $$ x^{2}-\sum_{k=0}^{1} A_{2, k} x^{k}=x^{2}-x-1 $$ Due to problem 47 , the values of $s(m, t)$ for some $m$ (depending on $t$ ) satisfy the relation $$ s(m, t) \geq C X_{\max }^{t} $$ where $C$ is some positive constant, and $X_{\max }=1.6180 \ldots$ is the largest root of the polynomial (2). Hence, $\log _{2} s(m, t) \geq t \log _{2} X_{\max }-\log _{2} C$. Removing "superfluous" big products to make their number to be a power of 2 , and setting $r=\left\lfloor\log _{2} s(m, t)\right\rfloor$, we obtain for $t \rightarrow \infty$ a sequence of $2^{r}$-long packings of big $(-, t)$-products such that $\frac{t}{t+r} \rightarrow \frac{1}{1+\log _{2}\left(X_{\max }\right)}=0.5902 \ldots$ Problem 49. Let $\mathcal{B}_{t}$ be the packing of big $(-, t)$-products constructed in the solution of problem 48 . We have already shown that the length of $\mathcal{B}_{t}$ is at least $C \varphi^{t}$, where $\varphi=\frac{\sqrt{5}+1}{2}=1.6180 \ldots$ Now, it suffices to show that the packing $\mathcal{B}_{t}$ contains at most $C^{\prime} c^{t}$ big products containing more than $r$ factors of type b) ( (for some $c<\varphi)$. Denote by $\rho(t, r)$ the number of big products in $\mathcal{B}_{t}$ with more than $r$ factors of type $\mathrm{b}$ ). Then the construction yields the recurrence relation $\rho(t, r)=\rho(t-2, r)+\rho(t-1, r-1)$. Now we define the following constants: $d=\log _{2} \varphi, A=\frac{3-d}{2}, \alpha=\frac{\ln \frac{3-d}{1-d}}{2 \ln A}, \beta=1-\alpha$. We prove that the inequality $$ \rho(t, r) \leq \gamma A^{\alpha t+\beta r} $$ holds for all $0 \leq r \leq t$, where $\gamma$ is some fixed constant. The proof goes by indution on $\ell$, For the base case, we choose $\gamma>0$ so that the estimate (3) holds for $t=t_{0}, t_{0}+1$; we will show that this value of $\gamma$ fits. To perform an induction step, assume that (3) holds for $t^{\prime}0$. We claim that, if $t$ is sufficiently large and $j \geq(2 / 3+\varepsilon) t(1+o(1))$, then we have $$ \frac{l_{j-2}(t+2)}{l_{j}(t)}>15 $$ For an arbitrary big product $\alpha$ in $\mathcal{B}_{m, t}$, denote by $n_{i}(\alpha), i=0,1,2,3,4,5$, the number of "pieces" of variables of length 10 in the product $\alpha$ containing exactly $i$ factors of type b) (we do not take into account the prefix piece obtained from the induction base, neither the pieces containing factors of type c)). Let $j_{0}(\alpha)$ be the number of factors of type b) in the prefix of $\alpha$, and let $t_{0}(\alpha)$ be the total number of factors of types a) and b) in the prefix of $\alpha$. Now, the ratio of the number $j(\alpha)$ of factors of type b) to the total numbert $(\alpha)$ of factors of types a) and b) is expressed as $$ \frac{j(\alpha)}{t(\alpha)}=\frac{5 n_{5}(\alpha)+4 n_{4}(\alpha)+3 n_{3}(\alpha)+2 n_{2}(\alpha)+n_{1}(\alpha)+j_{0}(\alpha)}{5 n_{5}(\alpha)+6 n_{4}(\alpha)+7 n_{3}(\alpha)+8 n_{2}(\alpha)+9 n_{1}(\alpha)+10 n_{0}(\alpha)+t_{0}(\alpha)} $$ For our claim, we are interested only in big products $\alpha$ in the packing $\mathcal{B}_{m, t}$ (for sufficiently large $t$ ) which satisfy $\frac{j(\alpha)}{t(\alpha)}>\frac{2}{3}+\varepsilon^{\prime}, 0<\varepsilon^{\prime}<\varepsilon$. Therefore, due to (6), we may assume that $\min _{\alpha \in \mathcal{B}_{m, t}} n_{5}(\alpha) \rightarrow \infty$ as $t \rightarrow \infty$, and that for sufficiently large $t$ ever big product $\alpha \in \mathcal{B}_{m, t}$ we are interested in satisfies $n_{5}(\alpha)>n_{3}(\alpha)+1$. Denote by $S\left(t, j, n_{5}\right)$ the set of big products in $\mathcal{B}_{m, t}$ which contain exactly $j$ factors of type b), as well as exactly $n_{5}$ pieces of length 10 with exactly 5 factors of type b). Take some values of $j$ and $t$ (where $t$ is sufficiently large), and consider any value of $n_{5}$ such that the set $S\left(t, j, n_{5}\right)$ is nonempty. In every big product $\alpha$ in $S\left(t, j, n_{5}\right)$ we replace one length 10 piece with 5 factors of type b) by an appropriate length 10 piece with 3 such factors. This can be made in $15 n_{5}$ ways, each of which results in a big product from $\mathcal{B}_{m, t+2}$ containing exactly $j-2$ factors of type b); every such product is obtained from $n_{3}(\alpha)+12 / 3+\varepsilon$. Using the inequality (5) we get $$ \frac{\sum_{j=k_{0}}^{t} l_{j}(t)}{s(t)}<\frac{\sum_{j=k_{0}-2 d}^{t-2 d} l_{j}(t+2 d)}{15^{d} s(t)}<\frac{s(t+2 d)}{15^{d} s(t)} $$ From problem 47 we get that for infinitely many values of $t$ the inequality $\frac{s(t+2 d)}{s(t)} \leq 2 X_{\max }^{2 d}$ holds; for such values of $t$ we have $$ \frac{\sum_{j=k_{0}}^{t} l_{j}(t)}{s(t)}<2\left(\frac{X_{\max }^{2}}{15}\right)^{d} \rightarrow 0, \quad d \rightarrow \infty $$ Thus, the number of big products with at least $t \log _{2} X_{\max }=0.7274 \ldots$ factors of type b) is asymptotically smaller than $s(t)$, at least for infinitely many values of $t$. (In fact, one can show that this holds for all sufficiently large values of $t$.) Consequently, for such values of $t$, removing such big products from the packing $\mathcal{B}_{m, t}$ constructed in the solution of problem 51, we still get a sequence of $2^{r}$-long packings of big $(m, t, r)$-products such that $\frac{t}{t+r} \rightarrow \frac{1}{1+\log _{2}\left(X_{\max }\right)}=0.5789 \ldots$ ## References [1] Tarannikov Y. V. Generalized proper matrices and constructing of $m$-resilient Boolean functions with maximal nonlinearity for expanded range of parameters. - Siberian Electronic Mathematical Reports, 2014. - V. 11, pp. 229-245 (http://semr.math.nsc.ru/v11/p229-245.pdf). [2] Sauskan A. V., Tarannikov Yu. V. On packings of (n,k)-products. — Siberian Electronic Mathematical Reports, 2016. — V. 13, pp. 888-896 (http://semr.math.nsc.ru/v13/p888-896.pdf). [3] Agahanov N. H., Bogdanov I. I., Kozhevnikov P. A., Podlipskij O. K., Tereshin D. A., Vserossijskie olimpiady shkol'nikov po matematike 1993-2006: Okruzhnoj i final'nyj jetapy, Pod red. N. H. Agahanova, M.: MCNMO, 2007, Problem 447, pp. 58, 276-277 (in Russian). [4] Tarannikov Yu. V., Nesokratimye razlozhenija odnorodnyh proizvedenij dvuchlenov dlja postroenija m-ustojchivyh funkcij s maksimal'no vozmozhnoj nelinejnost'ju, Problemy teoreticheskoj kibernetiki. Materialy XII mezhdunarodnoj konferencii (Kazan, 16-20 June 2014), Kazan': Otechestvo, (2014), 271-272 (in Russian). [5] Tarannikov Yu. V., O vozmozhnosti postroenija m-ustojchivyh funkcij s optimal'noj nelinejnost'ju $v$ ramkah odnogo metoda, Materialy XII Mezhdunarodnogo seminara Diskretnaja matematika i ee prilozhenija imeni akademika O. B. Lupanova (Moscow, MSU, 20-25 June 2016.), M.: Izd-vo mehaniko-matematicheskogo fakul'teta MSU, (2016), 394-397 (in Russian). # Инварианты изображений графов на плоскости представляют А. Еннэ, Т. Зайцев, А. Рябичев и А. Скопенков * ## Содержание 1 Задачи до промежуточного финиша ..... 3 1.1 Теоремы Радона и Тверберга для плоскости ..... 3 1.2 Линейные реализации графов ..... 4 1.3 Основные результаты о планарности графов ..... 6 1.4 Число пересечения для ломаных на плоскости ..... 8 1.5 Инвариант самопересечения изображения графа ..... 10 Указания и решения к задачам до промежуточного финиша ..... 11 2 Задачи после промежуточного финиша ..... 18 2.1 Полиномиальный алгоритм распознавания планарности ..... 18 2.2 Топологическая теорема Радона для плоскости ..... 20 $2.3 \mathrm{~K}$ доказательству теоремы Тверберга для плоскости ..... 20 Указания и решения к задачам после промежуточного финиша ..... 23 3 Дополнительные задачи для продвинутых команд ..... 27 3.1 Пересечения со знаком и для отображений в прямую ..... 27 3.2 Топологическая гипотеза Тверберга для плоскости ..... 28 Введение Основные результаты данного текста следуюшие: - полиномиальный алгоритм распознавания планарности графов (утверждение 1.3.6) и объяснение, как его придумать (§2.1); - теоремы Радона, Тверберга и Езайдына для плоскости (линейные и топологические) 1.1.1.c, 1.1.5.c, 2.2.1, 3.2.2, 3.1.4.d, 3.2.5 и их элементарные доказательства (§§1.1, 1.1, 3.2; для теорем Тверберга и Езайдына в частных случаях). Элементарные доказательства, которые мы приводим, не упоминают конфигурационных пространств и когомологических препятствий. Однако основное содержание этого текста - введение в алгебрачческую топологию (точнее, в теорию конфигурационных пространств и когомологических препятствий) мотивированное алгоритмическими, комбинаторными и геометрическими задачами. Мы представим некоторые идеи *Мы благодарны И. Богданову, Р. Карасеву, Ю. Макарычеву, Г. Челнокову и М. Танцеру за плодотворные обсуждения, а также Е. Морозову, Е. Стрельцовой и А. Щепиновой за перевод частей текста. А. Еннэ: Петрозаводский Государственный Университет. Т. Зайцев: Высшая Школа Экономики. А. Рябичев: Высшая Школа Экономики, Независимый Московский Университет. А. Скопенков: Московский Физико-Технический Институт, Независимый Московский Университет. Частично поддержан Российским Фондом Фундаментальных Исследований, гранты номер 07-0100648a и 15-01-06302, грантами фонда Саймонса и грантом фонда Д. Зимина «Династия». Email: skopenko@mccme.ru. http://www.mccme.ru/ skopenko решения топологической гипотезы Тверберга, поставленной в 1966 г. и окончательно решенной в 2015 г., см. обзоры [BZ, Sk16]. Два вышеуказанных направления связаны между собой понятием числа ван Кампе$н а$, обобщения которого требуются для обоих пунктов. Идея более общего препятствия (=инварианта) ван Кампена нужна и для придумывания полиномиального алгоритма распознавания планарности, и для формулировки теоремы Езайдына. Для решения приводимых задач не нужно специальных знаний, все новые определения будут даны. При этом потребуется (и будет далее развиваться) опыт сообразительности, т.е. математическая культура. ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-197.jpg?height=193&width=699&top_left_y=527&top_left_x=732) Вот схема (наиболее существенной) зависимости пунктов. Таким образом, начинать можно с п. 1.1, 1.2 или 1.4. Важные утверждения называются 'теоремами', 'леммами' и т.д. Они также являются задачами для решения, если явно не указано противоположное. Как правило, мы приводим формулировку утверждения перед его доказательством. ${ }^{1} \mathrm{~B}$ таких случаях для доказательства утверждения могут потребоваться следующие задачи. Это всегда явно оговаривается в подсказках, а иногда и прямо в тексте. ## Соглашения Если текст задачи является утверждением, то в задаче требуется его доказать. Если номер задачи помечен звездочкой (например, $\left.(b)^{*}\right)$, эта задача посложнее соседних. Такие задачи можно отложить до тех пор, пока не будут решены остальные. Участник (или группа участников) конференции, решающий задачи проекта, получает «боб» за каждое записанное решение, оцененное в «十» или «+.». Дополнительные бобы могут выдаваться за красивые решения, решения сложных проблем, или оформление некоторых решений в системе $\mathrm{T}_{\mathrm{E}} \mathrm{X}$. У жюри бесконечно много бобов. Решения можно сдавать и устно, отдавая один боб за каждые пять попыток (неважно, удачных или нет). Если вы застряли на какой-нибудь задаче, советуем перейти к следующим, они могут помочь. Приглашаем участников, работающим над проектом, обсуждать с жюри все возникающие вопросы. Особо успешным решателям вы выдаем дополнительные задачи для исследования. Пожалуйста, сообщите нам, если Вам знакомы какие-то из основных задач. Если Вы подтвердите свои знания, сообщив нам строгие доказательства некоторых из них, Вам будет разрешено не получать плюсы по всем этим задачам.[^20] ## 1 Задачи до промежуточного финиша ## 1.1 Теоремы Радона и Тверберга для плоскости 1.1.1. (а) Существуют такие 4 точки на плоскости, что для любого их разбиения на две пары отрезок, соединяющий точки в первой паре, не пересекает отрезок, соединяющий точки во второй паре. (b) Существуют такие 4 точки на плоскости, ни одна из которых не лежит внутри треугольника, образованного оставшимися точками. (c) Теорема Радона для плоскости. Для любых 4 точек на плоскости либо одна из них лежит внутри треугольника, образованного оставшимися точками, либо их можно разбить на две пары так, что отрезок, соединяющий точки в первой паре, пересекает отрезок, соединяющий точки во второй паре. Выпуклой оболочкой конечного набора точек плоскости называется наименьший (по включению, или по площади) выпуклый многоугольник, их содержащий. Теорему Радона для плоскости можно переформулировать так: любые 4 точки на плоскости можно разбить на два множества, выпуклые оболочки которых пересекаются. 1.1.2. (Ср. с задачей 2.2.2) Если никакие 3 из 4 точек на плоскости не лежат на прямой, то существует ровно одно их (радоновское) разбиение на два множества, выпуклые оболочки которых пересекаются. 1.1.3. (а) Существуют 6 точек плоскости, при любом разбиении которых на три множества выпуклые оболочки этих трех множеств не имеют общей точки. (b) Существуют 7 точек на плоскости, для любой нумерации $1, \ldots, 7$ которых точка 1 не лежит в одном из (двумерных) треугольников 234 и 567. (c) Существуют 7 точек на плоскости, для любой нумерации $1, \ldots, 7$ которых точка пересечения отрезков 12 и 34 не лежит в треугольнике 567. 1.1.4. (а) Вершины любого выпуклого 8-угольника на плоскости можно разбить на 3 множества, выпуклые оболочки которых имеют общую точку. (b)* То же для 7 -угольника. (c) Любые 15 точек плоскости можно разбить на 3 множества, выпуклые оболочки которых имеют общую точку. (d)* То же для 7 точек. Подсказка к доказательству утверждений 1.1.4.d и 1.1.5.c приведена в задаче 2.3.8. 1.1.5. (а) Для любого $r$ существуют $3 r-3$ точки плоскости, при любом разбиении которых на $r$ множеств выпуклые оболочки этих множеств не имеют общей точки. (b) Для любого $r$ существует такое $N$, что любые $N$ точек плоскости можно разбить на $r$ множеств, выпуклые оболочки которых имеют общую точку. (c)* Теорема Тверберга для плоскости. Для любого $r$ любые $3 r-2$ точки плоскости можно разбить на $r$ множеств, выпуклые оболочки которых имеют общую точку. ## 1.2 Линейные реализации графов В этом и следующем пунктах мы представляем две формализации понятия реализуемости графа на плоскости. Важны оба способа; второй способ использует первый. (Они оказываются эквивалентными по теореме Фари 1.3 .4 ; их многомерные версии не эквивалентны.) Под $k$ точками на плоскости подразумевается $k$-элементное подмножество плоскости, т.е., считается, что эти $k$ точек различны. Утверждение 1.2.1. ${ }^{2}$ (а) Из любых 5 точек на плоскости можсно выбрать две такие непересекающиеся пары точек, что отрезок, соединяющий точки первой пары, пересекает отрезок, соединяющий точки второй пары. Более того, если никакие 3 из них не лежат на прямой, то количество точек пересечения внутренностей отрезков, соединяющих данные точки, нечетно. (b) Даны две тройки точек на плоскости. Тогда существуют два пересекающихся отрезка, не имеющие общих вершин, каждый из которых соединяет точки из разных троек. Утверждение 1.2.1 легко доказывается с помощью рассмотрения выпуклой оболочки точек. Теорема 1.2.2 (Общего положения). Для любого $n$ существуют такие $n$ точек в трехмерном пространстве, что отрезки, их соединяющие, не имеют общих внутренних точек. 1.2.3. Для любых пяти точек $1,2,3,4,5$ на плоскости если отрезки (a) $j k, 1 \leq j 3 Дополнительные задачи для продвинутых команд ## 3.1 Пересечения со знаком и для отображений в прямую Здесь определяется препятствие Ван Кампена к $\mathbb{Z}$-планарности графа (задачи 3.1.13.1.4) и к $\mathbb{Z}_{2}$-вложимости графа в пря.мую (задача 3.1.5). Эти обобщения конструкций из п. 2.1 формально не используются далее. Однако эти простые обобщения полезно проделать перед более сложными обобщениями в задачах 3.2.4, 3.2.5. Граф называется $\mathbb{Z}$-планарным, если существует его отображение общего положения в плоскость, при котором сумма знаков точек пересечения $f$-образов любых двух несмежных ребер равна нулю, для некоторых (или, эквивалентно, для любых) ориентаций на этих ребрах [Sk16, Fig. 4]. 3.1.1. (а) Нарисуйте две (незамкнутые) ориентированные ломаные общего положения, пересекающиеся в четном числе точек, сумма знаков которых не равна нулю. (b) Граф планарен тогда и только тогда, когда он $\mathbb{Z}$-планарен. Ориентируем ребра графа $K$. Пусть $f: K \rightarrow \mathbb{R}^{2}$ - отображение общего положения. Поставим в соответствие упорядоченной паре $(\sigma, \tau)$ несмежных ребер сумму $f \sigma \cdot f \tau$ знаков точек пересечения их $f$-образов. Обозначим через $\widetilde{K}$ множество упорядоченных пар несмежных ребер графа $K$. Тогда получено отображение $\widetilde{K} \rightarrow \mathbb{Z}$. Назовем его целочисленной расстановкой пересечений. В этом пункте слово «целочисленная» пропускается. 3.1.2. (a) Ориентируйте ребра графа $K_{5}$ и найдите расстановку пересечений для линейного отображения общего положения $K_{5} \rightarrow \mathbb{R}^{2}$, образы вершин при котором образуют выпуклый 5 -угольник. (b) $f \sigma \cdot f \tau=-f \tau \cdot f \sigma$. (c) Как меняется расстановка пересечений при изменении ориентации ребра? (d) Как меняется расстановка пересечений при изменении ориентации плоскости, т.е. при композиции $f$ с осевой симметрией? (е)* В условиях задачи 2.1.4.b $f \sigma \cdot f \tau$ равно числу оборотов вектора при обходе по $\partial(\sigma \times \tau)$. 3.1.3. (I-V) Как меняется расстановка пересечений при преобразованиях Райдемайстера на рис. 9.I-V? (На рис. 9.V вершина $A$ не является концом ребра $\tau$.) Элементарной (целочисленной) кограницей пары $(A, \sigma)$, в которой $A$ - вершина, не являющаяся концом ребра $\sigma$, называется отображение $\delta(A, \sigma): \widetilde{K} \rightarrow \mathbb{Z}$, при котором - в единицу отображается каждая пара $(\sigma, \tau)$, для которой $\tau$ выходит из $A$, и каждая пара $(\tau, \sigma)$, для которой $\tau$ входит в $A$, - в минус единицу отображается каждая пара $(\sigma, \tau)$, для которой $\tau$ входит в $A$, и каждая пара $(\tau, \sigma)$, для которой $\tau$ выходит из $A$, - в ноль отображаются все остальные пары. Ответ к задаче 3.1.3.V - к расстановке пересечений добавляется $\delta(A, \sigma)$. Назовем расстановки $N_{1}, N_{2}: \widetilde{K} \rightarrow \mathbb{Z}$ (целочисленно) когомологичными, если $$ N_{1}-N_{2}=m_{1} \delta\left(A_{1}, \sigma_{1}\right)+\ldots+m_{k} \delta\left(A_{k}, \sigma_{k}\right) $$ для некоторых вершин $A_{1}, \ldots, A_{k}$, ребер $\sigma_{1}, \ldots, \sigma_{k}$ и целых чисел $m_{1}, \ldots, m_{k}$ (не обязательно различных). 3.1.4. (a), (b) Сформулируйте и докажите целочисленные аналоги задач 2.1.7.ab. (c) Удвоенная расстановка пересечений любого отображения общего положения графа в плоскость когомологична нулю. 3.1.5. (а) Для любых отображения $f: K \rightarrow \mathbb{R}$ общего положения графа $K$ (определите, что это такое!) и пары $\{A B, C D\}$ несмежных ребер графа $K$ $$ |f(A) \cap f(C D)|+|f(B) \cap f(C D)|+|f(A B) \cap f(C)|+|f(A B) \cap f(D)| \equiv 0 $$ (b)* Для любых отображения $f: K \rightarrow \mathbb{R}$ общего положения графа $K$ назовем расстановкой пересечений сопоставление числа $|f(A) \cap f(\sigma)|$ каждой паре $A, \sigma$ из вершины и не содержащего ее ребра. Придумайте хотя бы два аналога преобразований Райдемайстера на рис. 9 для отображений графов в прямую. Выясните, как меняется расстановка пересечений при таких аналогах. (c)* Определим граф $K^{*(1)}$. Его вершины - неупорядоченные пары $\{A, B\}$ различных вершин графа $K$. Для каждой пары $A, B C$ из вершины и не содержащего ее ребра в графе $K$ соединим ребром в графе $K^{*(1)}$ вершины $\{A, B\}$ и $\{A, C\}$. Это ребро обозначается $A \times B C=B C \times A$. Найдите $K^{*(1)}$, если $K$ - цикл с тремя вершинами, триод, $K_{4}$. (d) Для вершины $B$ графа $G$ назовем элементарной кограницей $\delta_{G} B$ вершины $B$ расстановку единиц на ребрах графа $G$, содержащих эту вершину, и нулей на остальных ребрах. (Иными словами, $\delta_{G} B$ соответствует множеству ребер с концом $B$.) Назовем расстановки $\omega_{1}$ и $\omega_{2}$ когомологичными, если $\omega_{1}-\omega_{2}$ является суммой некоторых элементарных кограниц $\delta_{K^{*(1)}}\{A, B\}$. Верно ли, что существует отображение $f: K \rightarrow \mathbb{R}$ общего положения графа $K$, для которого $f(A) \notin f(\sigma)$ для любой вершины $A$ и ребра $\sigma \not \supset A$, тогда и только тогда, когда расстановка пересечений некоторого отображения общего положения $f^{\prime}: K \rightarrow \mathbb{R}$ когомологична нулю? (е)* Коциклом называется такая расстановка нулей и единиц на ребрах графа $K^{*(1)}$, что сумма четырех чисел ребрах $A \times C D, B \times C D, C \times A B, D \times A B$ четна для любых непересекающихся ребер $A B, C D$ графа $K$ (ср. с п. (а)). Для каждого коцикла $\nu$ сопоставим неупорядоченной паре $\{A B, C D\}$ непересекающихся ребер графа $K$ сумму двух чисел на «противоположных» ребрах $A \times C D$ и $B \times C D$ «прямоугольника» $A B \times C D$. Т.е. определим отображение $\mathrm{Sq}^{1} \nu: K^{*} \rightarrow \mathbb{Z}_{2}$ формулой $$ \mathrm{Sq}^{1} \nu\{A B, C D\}:=\nu(A \times C D)+\nu(B \times C D)=\nu(A B \times C)+\nu(A B \times D) $$ Докажите, что $\mathrm{Sq}^{1}(\mu+\nu)=\mathrm{Sq}^{1} \mu+\mathrm{Sq}^{1} \nu$. (f) ${ }^{*}$ Выразите $\mathrm{Sq}^{1}(\mu \nu)$ через $\mu, \nu, \mathrm{Sq}^{1} \mu, \mathrm{Sq}^{1} \nu$. (Здесь сумма и произведение коциклов - пореберные.) $(\mathrm{g}) * \delta_{K^{*(1)}}\{A, B\}-$ коцикл. (h)* $\mathrm{Sq}^{1} \delta_{K^{*(1)}}\{A, B\}=\sum_{\sigma \ni B} \delta(A, \sigma)=: \delta\left(A \times \delta_{K} B\right)$. ## 3.2 Топологическая гипотеза Тверберга для плоскости Числом оборотов замкнутой ориентированной ломаной $A_{1} \ldots A_{n}$ вокруг не лежащей на ней точки $O$ называется следующая сумма ориентированных углов, деленная на $2 \pi$ : $$ \operatorname{deg}_{A_{1} \ldots A_{n}} O:=\left(\angle A_{1} O A_{2}+\angle A_{2} O A_{3}+\ldots+\angle A_{n-1} O A_{n}+\angle A_{n} O A_{1}\right) / 2 \pi $$ 3.2.1. (а) Найдите число оборотов (произвольно ориентированного) правильного пятиугольника вокруг его центра и вокруг точки пересечения прямых, содержащих несмежные стороны. (b) Для каждой ломаной (с произвольной ориентацией) на рис. 4 и точки на ваш выбор (в любой из ограниченной областей) найдите число оборотов ломаной вокруг точки. 3.2.2. * (3) Для любого отображения $f: K_{7} \rightarrow \mathbb{R}^{2}$ можно так занумеровать вершины числами $1, \ldots, 7$, что либо - число оборотов образа цикла 567 вокруг некоторой точки пересечения образов ребер 12 и 34 не равно нулю, либо - число оборотов образа каждого из циклов 567 и 234 вокруг образа точки 1 не равно нулю. (4) Топологическая теорема Тверберга для плоскости. Если $r$ - степень простого, то для любого отображения $f: K_{3 r-2} \rightarrow \mathbb{R}^{2}$ можно так занумеровать вершины числами $1, \ldots, 3 r-2$, что либо - число оборотов образа каждого из циклов $3 t-1,3 t, 3 t+1, t=2,3, \ldots, r-1$, вокруг некоторой точки пересечения образов ребер 12 и 34 не равно нулю, либо - число оборотов образа каждого из циклов $3 t-1,3 t, 3 t+1, t=1,2,3, \ldots, r-1$, вокруг образа точки 1 не равно нулю. (Случаи $r=5, r$ простого, $r=4, r=2^{k}$ засчитываются как отдельные пункты.) (6) Топологическая проблема Тверберга для плоскости. Верен ли аналог утверждения (4), если $r$ - не степень простого? Подсказки к задаче 3.2.2 приведены в следующей задаче, cf. problem 2.3.8. Вy [Sc05, SZ] Теорема 3.2.2 эквивалентна следующему утверждению: Если $r$ - степень простого, то для любого отображения ( $3 r-1)$-мерного симплекса в плоскость существует $r$ попарно непересекающихся граней, образы которых имеют общую точку. Аналогичным образом можно переформулировать задачу 3.2.2. 3.2.3. * (а) Для любого отображения общего положения $f: K_{7} \rightarrow \mathbb{R}^{2}$ $$ \sum_{\substack{\left|R_{1}\right|=1 \\\left|R_{2}\right|=\left|R_{3}\right|=3}} \operatorname{csgn} R \operatorname{deg}_{f R_{2}}\left(f R_{1}\right) \operatorname{deg}_{f R_{3}}\left(f R_{1}\right)+\sum_{\substack{\left|R_{1}\right|=\left|R_{2}\right|=2,\left|R_{3}\right|=3, X \in f R_{1} \cap f R_{2}}} \operatorname{csgn} R \operatorname{sgn} X \operatorname{deg}_{f R_{3}} X \equiv 2 $$ Здесь суммирование происходит по всем радужным разбиениям $R=\left(R_{1}, R_{2}, R_{3}\right)$ с указанными ниже знака суммы свойствами, а $f R_{i}$ есть $f$-образ вершины $R_{i}$, ребра $R_{i}$ или ориентированного цикла $R_{i}=a b c, aN(k, l)$ и любого разбиения множества $1, \ldots, N$ на $k$ подмножеств, одно из подмножеств содержит арифметическую прогрессию длины $l$. Верхниие оценки на числа $N(k, l)$ просто чудовишные. Например, $N(k, 3)$ имеет порядок экспонента от экспоненты и т.д. $k$ раз, для $N(k, 4)$ вместо экспоненты итерируется функция $N(k, 3)$ и т.д. (лучшие оценки были получены совсем недавно и неэлементарными методами). Нижние оценки - только экспоненциальные. Этой тореме, в частности, был посвящен проект на Летней конференции Турнира Городов в 1995 году, так что ее доказательство не является нашей целью. Смежные вопросы над которыми предлагается подумать участникам таковы: 1. Можно ли раскрасить множество натуральных чисел так, чтобы бесконечной одноцветной арифметической прогрессии не нашлось? 2. Докажите экспоненциальную нижнюю оценку: $N(s, l)>s^{l / 2}$. Попробуйте построить явную конструкцию. 3. Можно ли раскрасить более половины вершин какого-нибудь правильного $n$ угольника в чёрный цвет так, что объединение исходной раскраски и любых десяти её поворотов не покроют чёрным цветом все клетки? Постарайтесь получить по-возможности лучшие оценки на $n$, а также явные конструкции. Теорема Семереди (она же гипотеза Эрдеша-Турана, поставленная в 1936 году) утверждает, что если $A$ - произвольное подмножество натурального ряда и последовательность $a_{n}=$ ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-256.jpg?height=57&width=1589&top_left_y=1544&top_left_x=146) ность $a_{k_{1}}, \ldots, a_{k_{i}}, \ldots$ стремящаяся к точке отличной от нуля), тогда для любого натурального $k>3$ множество $A$ содержит арифметическую прогрессию длины $k$. Это один из основных результатов Семереди (полученный в 1975 году), за которые он получил премию Абеля. Для прогрессий длинны 3 этот результат получен Клаусом Ротом в 1953 году. https://ru.wikipedia.org/wiki/Теорема_Семереди. Один из основных результатов Теренса Тао явилось доказательство того, что среди простых чисел найдется сколь угодно длинная арифметическая прогрессия. Результат И.Д.Шкредова относится к нахождению чёрного треугольника в квадрате $N \times N$ при раскраске некоторой доли клеток в черный цвет (см. http://www.mathnet.ru/php/archive.phtml?wshow=paper\&jrnid=im\&paperid=912\&option_lang=rus). Наша цель - посмотреть на эту тематику с игровой точки зрения. Возможно, этот альтернативный взгляд позволит лучше понять вопросы, связанные с теоремами Ван-Дер-Вардена и Семереди и получить новые результаты, в частности, оценки в теореме Ван-Дер-Вардена. ## Серия А 1. Двое играют в следующую игру. На бесконечной в обе стороны клетчатой полоске первый ставит в свободную клетку один крестик, второй ставит в 1000 свободных клеток нули. Цель первого игрока - создать арифметическую прогрессию из крестиков длины 3 , цель второго ему помешать. a) Докажите, что у первого игрока существует выигрышная стратегия. б) Докажите, что найдется число $n$, такое что первый сможет гарантированно завершить игру за $n$ ходов. в) Найдите минимальное $n$ в условиях пункта б). г) Пусть теперь первый и второй играют на бесконечной в одну сторону полосе. Найдите минимальное число ходов достаточное для выигрыша первого игрока. 2. Двое играют на бесконечной клетчатой плоскости в следующую игру. Первый ходит крестиком в свободную клетку, второй отвечает ему 1000 нулями. Цель первого - создать равнобедренный прямоугольный треугольник из крестиков, катеты которого параллельны осям координат. Цель второго - помешать ему. a) Докажите, что у первого игрока существует выигрышная стратегия б) Найдите минимально число ходов необходимое для гарантированного выигрыша первого. 3. Сможет ли первый создать арифметическую прогрессию длины 4 в условиях задачи 1 ? 4. В условиях задачи 2 может ли первый построить квадрат? 5. Пусть в игре на полоске первый ставит один крестик, а второй отвечает ему $n$ ноликами. Может ли первый создать арифметическую прогрессию длины $k$ ? 6. Оцените минимальное число ходов необходимое для выигрыша первого в предыдущей задаче. 7. Теперь два игрока играют в игру на плоскости. Первый ставит одну красную точку, второй отвечает $k$ синими. Докажите, что для любого набора из $n$ точек $F$, первый сможет создать фигуру $F^{\prime}$ подобную $F$ (то есть получающуюся из $F$ композицией гомотетии и параллельного переноса), все вершины которой красные. ## Серия Б Обобщенная теорема ван-дер-Вардена: Пусть рациональные точки плоскости раскрашены в конечное число цветов. Тогда для любой фигуры $M$ состоящей из конечного числа рациональных точек, можно найти одноцветную фигуру $M^{\prime}$ подобную $M$ (то есть получающуюся из $M$ композицией гомотетии и параллельного переноса) состоящую из рациональных точек. 1. Докажите, что в формулировке обобщенной теоремы Ван-Дер-Вардена понятие подобного множества нельзя заменить на понятие конгруэнтного (или равного согласно терминологии современной школьной программы). Иными словами, одноцветного множества, конгруэнтного множеству $M$ может и не найтись. (*) А что если рассмотреть группу преобразований, сохраняющих площадь или гиперболических поворотов (а именно, преобразований, сжимающих одну ось координат $O X$ и растягивающих другую $O Y$ в одно и то же число раз)? 2. Рациональные точки плоскости раскрашены в несколько цветов. Доказать, что найдется прямоугольный треугольник, с катетами, параллельными осям координат с одноцветными вершинами. (При решении этой задачи разрешено пользоваться теоремой Ван-Дер-Вардена). 3. (** решение этой задачи неизвестно) В условиях предыдущей задачи докажите, что найдется одноцветный прямоугольник площади 1 с одноцветными вершинами. Назовем симплекс в $n$-мерном пространстве стандартным если у него есть $n$ ребер, параллельным осям координат. 4. Рациональные точки $n$-мерного пространства раскрашены в несколько цветов. Доказать, что найдется стандартный симплекс с одноцветными вершинами. (При решении этой задачи разрешено пользоваться теоремой Ван-Дер-Вардена). 5. Сформулируйте и докажите игровую форму предыдущих пунктов. Вторая задача в этой серии в игровой форме становится уже подъемной. В этой связи возникает вопрос: 6. (*) Сформулируйте и решите игровую форму задачи об образе данного конечного множества под действием гиперболических поворотов. ## Серия В 1. Имеется неограниченное количество игральных досок $8 \times 8$. Первый своим ходом ставит 2 фишки на любые из них, второй отвечает постановкой одной черной фишки на любую из досок. a) Может ли первый гарантированно замостить одну из досок? б) За какое наименьшее число ходов первый сможет это сделать? 2. Двое играют в крестики - нолики на бесконечной в одну сторону клетчатой полоске. Первый ходит двумя крестиками, второй отвечает ноликом. Может ли первый получить 100 крестиков идущих подряд? 3. Может ли первый в условиях первой задачи добиться этого за $2^{45}$ ходов? 4. А за $2^{90}$ хода? ## Серия $\Gamma$ Двое художников играют в следующую игру. Первый ставит на плоскости точку и соединяет с несколькими поставленными ранее. При этом пересекать ранее проведенные дуги нельзя. Второй красит поставленную первым точку так, чтобы никакие две одноцветные точки не были соединены дугой. 1. Может ли первый заставить второго использовать более $n$ цветов? 2. Может ли первый достичь этого за полиномиальное число ходов (от $n$ )? 3. Нарисовано дерево. Двое по очереди красят вершины так, чтобы соседние вершины были раскрашены в разные цвета. Может ли первый заставить второго использовать более четырех цветов? 4. Может ли первый, в условиях предыдущей задачи, заставить второго использовать более трех цветов? 5. Имеется двудольный граф $G$. Двое по очереди красят его вершины так, чтобы соседние вершины были раскрашены в разные цвета. При этом второй игрок старается использовать как можно меньше цветов, а первый хочет заставить его использовать как можно больше цветов. Верно ли, что при любом натуральном $N$ двудольный граф $G$ можно подобрать так, что второму не хватит $N$ цветов? ## Серия Д 1. Дан полный граф на $n$ вершинах. Двое по очереди красят его рёбра. Первый красит одно ребро красным, второй красит 100 ребер в синий цвет, и так повторяется, пока все рёбра не будут покрашены. Может ли первый при достаточно больших $n$ добиться появления полного подграфа со 100 вершинами и всеми красными рёбрами? 2. В красный цвет раскрашены $99 \%$ рёбер полного графа из $n$ вершин. Верно ли, что при достаточно большом $n$ найдётся полный подграф из 1000 вершин с ребрами красного цвета? 3. Каждое $k$-элементное подмножество множества $\{1, \ldots, n\}$ раскрашено в один из $s$ цветов (например, при $k=2$ получаем раскраску рёбер полного графа на $n$ вершинах). Докажите, что при достаточно большом $n$ найдётся такое подмножество $U \subset\{1, \ldots, n\}$, что все его $k$ элементные подмножества - одного и того же цвета, причём если $x$-минимальный элемент из $U$, то число элементов в $U$ не меньше $\exp (x)+s$. ## Теорема Ван Дер Вардена и игры 2 ## Серия $\mathrm{E}$ Проблемы Ван Дер Варденского типа можно рассматривать в классическом, плотностном (типа теоремы Семиреди) и в игровом аспектах. ## Классическая версия 1. Целочисленные точки плоскости раскрашены в $k$ цветов. Докажите, что найдется одноцветный прямоугольник с вершинами в целых точках. 2. Целочисленные точки пространства раскрашены в $k$ цветов. Докажите, что найдется одноцветный параллелепипед с вершинами в целых точках. 3. Сформулируйте и докажите $n$-мерный аналог предыдущих задач. 4. Докажите, что аналог задачи $\mathrm{E} 1$ верен для раскраски целочисленных точек прямоугольника: а) Размера $(k+1) \times\left(k^{(k+1)}+1\right)$ б) Размера $(k+1) \times\left(k C_{k}^{2}+1\right)$ 5. Сформулируйте и докажите аналоги предыдущей задачи для: a) Трехмерной сетки б) $n$-мерной сетки 6. Целочисленные точки квадрата $S \times S$ красятся в $k$ цветов. Получите оценку на $S$ при которой найдется одноцветный прямоугольник. 7. Назовем $\mathrm{l}$-решеткой множество из $l^{2}$ точек, получаемых как пересечение $l$ вертикальных прямых с $l$ горизонтальными. Докажите, что при раскраске плоскости в $k$ цветов найдется $l$-решетка. 8. Целочисленные точки квадрата $S \times S$ красятся в $k$ цветов. Получите оценку на $S$ при которой найдется $l$-решетка. ## Плотностная версия Определим понятие плотности для целочисленных точек на плоскости: скажем, что какой-либо цвет имеет плотность $k$, если последовательность $$ a_{n}=\frac{\text { Количество черных точек внутри квадрата } n \times n \text { с центром в точке }(0 ; 0)}{n^{2}} $$ стремится к $k$. 9. Докажите, что при раскраске некоторых точек плоскости в черный цвет с плотностью $1 \%$ найдется черный прямоугольник. 10. Сформулируйте и докажите аналогичное условие для: a) Пространства б) $n$-мерного пространства 11. Докажите, что при раскраске некоторых точек плоскости в черный цвет с плотностью $1 \%$ найдется одноцветная $l$-решетка. ## Игровая версия 12. Двое играют на бесконечной клетчатой плоскости в следующую игру. Первый ходит крестиком в свободную клетку, второй отвечает ему $k$ нулями. Цель первого - прямоугольник, стороны которого параллельны осям координат. Цель второго - помешать ему. Докажите, что первый сможет победить. 13. В условиях предыдущей задачи найдите минимальное число ходов необходимых первому для победы. 14. Решите 12 задачу для параллелепипеда в пространстве. Постарайтесь как можно лучше оценить число ходов. 15. Решите задачу 12 для $l$-решеток. Постарайтесь как можно лучше оценить число ходов. ## Дополнительные задачи к серии А 8. Докажите в условиях задачи A1, что первый сможет построить прогрессию длины 5 . ## Дополнительные задачи к серии Б 7. Пусть квадрат $S \times S$ разбит на многоугольники. Некоторые из них покрашены в черный цвет (с границей) так, что суммарная площадь закрашенных многоугольников равна $1 \%$ от площади всего квадрата. а) Докажите, что существует такое $S$, что внутри квадрата всегда можно найти прямоугольный треугольник со сторонами, параллельными сторонам квадрата, единичной площади с черными вершинами. б)* Докажите, что существует такое $S$, что внутри квадрата всегда можно найти прямоугольник единичной площади с черными вершинами. ## Теорема Ван Дер Вардена и игры. Решения ## Серия А 1. Ответ: 336. Первый ставит 334 крестика куда угодно, 335 ставит далеко справа от уже поставленых крестиков и ноликов так, чтобы 335 крестик образовывал со всеми предыдущими крестиками 3 потенциальных места для выигрыша и 336 ходом заканчивает партию. Второй в свою очередь может не проиграть за 335 ходов, так как каждый новый крестик образует с каждым предыдущим крестиком как максимум 3 угрозы завершения матча на следующем ходу, которые второй и будет закрывать ноликами. В пункте г) ответ такой же, стратегия второго остается неизменной, 2. Ответ: 253. Стратегия и первого, и второго аналогичны предыдущей задаче, с поправкой на то, что каждым первый может создавать 4 угрозы ставя крестики в ряд. 3. Да, сможет. Приведем решение Гаркавенко Андрея. Первый ставит $R$ крестиков в начале и объединяет их в фигуру $X$. Пусть фигура $X$ занимает $k$ клеток. После построения первой фигуры $X$ первый выбирает $k$ клеток справа без нулей и начинает в них воссоздавать фигуру $X$. Полностью она у него конечно же не получится, но получится точно хотя бы $\frac{1}{1000}$ часть. После этого первый повторяет данный процесс, только строит уже подфигуру последней подфигуры, пока не получит $l$ вложенных подфигур типа $X$ (выбрав $R$ достаточно большим, можно гарантировать для каждого фиксированного $l$ сколь угодно большой размер $l$-той подфигуры $X$ ). После этого первый строит подфигуру последней фигуры, но повернутую на 180 градусов, обозначим ее за $X^{\prime}$. Заметим, что между $X^{\prime}$ и любой до этого построенной подфигурой $X$ можно поставить крестик так, чтобы элементы образовывали $\left|X^{\prime}\right|$ последовательностей длины 3. Такой ход будет существовать, так как можно ставить все крестики только в четные клетки. Теперь если выбрать $R$ такое что $X^{\prime}$ больше 1001, чтобы $l=1001\left|X^{\prime}\right|$ тогда во время построения $X^{\prime}$ второй не успеет закрыть все клетки средние клетки между $X^{\prime}$ и каждой из $l$ подфигур $X$ и первый ставя соедующим ходом крест в один из таких центров создает как минимум 1001 угрозу последовательности длины 4 и выигрывает. ## Серия 5 Все известные результаты задач данной серии можно найти в дипломной работе В. 3. Шарича https://istina.msu.ru/diplomas/60736779/ ## Серия В 1. Первый строит $2^{62}$ досок с одной фишкой за $2^{63}$ хода, из них делает $2^{61}$ досок с двумя фишками и так далее, пока до хода второго не получит две доски с 62 крестиками. Второй уничтожит как максимум одну из них и первый поставит сдвоенные фишки на оставшуюся, затратив в сумме $2^{64}-1$ ход. Требуется минимум $2^{64}-1$ ход, так как если назвать весом доски с $k$ белыми фишками величину $2^{k}-1$, то после хода первого и второго суммарный вес изменяется на 1 , а требуется получить $2^{64}-1$. 2. Первому для победы достаточно повторить алгоритм из предыдущей задачи, только досками считать отрезки вида ( $100 k+1 ; 101 k)$. 3. Заметим, что второй может объединить фишки с номерами вида $(50 k+1 ; 51 k)$ в доски из первой задачи и помешать первому за $2^{45}$ ходов собрать полную последовательность такого вида, осталось заметить, что в любой последовательности из 100 чисел идущих подряд есть отрезок из 50 чисел вида ( $50 k+1 ; 51 k)$. ## Серия $\Gamma$ 1. Да, сможет. Докажем утверждение по индукции. Пусть первый умеет строить фигуру с такими свойствами: а) На раскраску второму всегда потребуется хотя бы $k$ цветов. б) Каждая новая точка во время построения данной фигуры лежала во внешней грани. Тогда построить фигуру с такими же свойствами, только для $k+1$ цвета можно так: первый строит $k$ фигуру $k$ раз и каждый раз останавливается когда второй поставит новый еще не использовавшийся до этого цвет, потом начинает строить следующую $k$ фигуру. Каждый раз новый цвет оказывается на границе фигуры по условию б). И заканчивает построение постановкой точки вне всех предыдущих фигур, соединяя с $k$ цветами. 3. Покажем, что второ й может поддерживать следующую ситуацию: закрашенные ранее вершины разбивают дерево на части, в каждой не более 2 окрашенных вершины. В самом деле. Пусть первый в одной из таких частей красит вершину. Получается дерево с 3 окрашенными вершинами.В этом случае либо оно распадается на деревья с 2 окрашенными вершинами, либо можно указать вершину, при выбрасывании которой все 3 покрашенные вершины окажутся в разных компонентах. Второй туда и ходит четвертым цветом. 4. Конструкция в данной задаче была предложена участниками Георгиевым Филипом, Широковских Светой и Пушкаревой Марией и является крайне сложной. 5. Достаточно ставить каждый раз цвет который ставил второй в другую долю, это будет вынуждать его не использовать этот цвет еще раз в случае полного двудольного графа. ## Серия Д 1. Докажем по индукции, что для любого $k$ существует $n(k)$ такое, что при игре на графе из $n(k)$ вершин у первого получится построить полный граф на $k$ вершинах с красными рёбрами. Для $k=1$ это очевидно. Теперь из существования такого $n(k)$ выведем существование $n(k+1)$. Опишем стратегию первого для $k+1$, считая стратегию для $k$ известной. Он выбирает произвольную вершину $A$ и старается получить $n(k)$ вершин, соединённых с $A$ красными рёбрами и попарно не соединённых между собой. Если ему это удастся, то, по предположению идукции, он может построить полный граф на $k$ вершинах среди этих $n(k)$ соединённых с $A$ вершин, и этот граф вместе с $A$ будет искомым полным графом. Осталось показать, как первый может содинить $A$ и сколь угодно много не попарно не соединённых рёбрами вершин. Сначала первый соединяет $A$ и $1000 n(k)$ вершин, причём каждая следующая вершина до этого не должна быть соединена ни с одной из предыдущих. Средняя «синия степень» этих $1000 n(k)$ вершин не более 200 , значит существует множество вершин $S$, такое что $|S|=1000 n(k) / 2$ и из каждой вершины из $S$ выходит не более 400 синих рёбер. Теперь докажем, что $S$ уже содержит подмножество $P$, состоящее из $n(k)$ попарно не соединённых вершин. Будем добавлять в $P$ вершины по одной. Начнём с произвольной вершины из $S$, зачеркнём все соединённые с ней синими рёбрами, их не более 400 , затем добавим любую ещё не зачёркнутую вершину из $S$, зачеркнём все соединённые с ней. Повторим эту процедуру, пока в $P$ не окажется $n(k)$ вершин. Ясно, что вершин в $S$ хватит и $P$ - искомое множество. Как было сказано, применим к $P$ предположение индукции, полученный граф вместе с $A$ - полный граф с красными рёбрами. 2. Это неверно: для сколь угодно большого $n$ можно построить граф на $n$ вершинах, $99 \%$ рёбер которого покрашены в красный цвет, в котором не полного подграфа на 1000 вершинах. Изложенное ниже решение на летней конференции было придумано Петром Кучерявым. Разделим $n$ вершин на 999 долей размерами $\left[\frac{n}{999}\right]$ и $\left[\frac{n}{999}\right]+1$. Соединим вершины красным рёбром тогда и только тогда, когда они лежат в разных долях, если получилось более $99 \%$ красных рёбер, некоторые из них можно перестать считать красными. Подграфа на 1000 вершинах с красными рёбрами нет, так как среди 1000 вершин найдутся две, лежащие в одной доле, то есть не соединённые красным ребром. Несложно понять, что для достаточно больших $n$ красных рёбер будет не менее $99 \%$. 3. Сформулируем две стандартные леммы, из которых будет следовать решение задачи. Лемма 0.1 (Бесконечная теорема Рамсея для гиперграфов.) Каждое $k$-элементное подмножество множества натуральных чисел раскрашено в 1 из $s$ цветов. Тогда найдётся такое бесконечное подмножество множества натуральных чисел, что все его $k$-элементные подмножества одного цвета. Эта лемма - обобщение теоремы Рамсея для обычных графов, частным случаем которой является, например, утверждение о том, что среди любых шести людей найдётся трое попарно знакомых или трое попарно незнакомых. Лемма 0.2 (Компактность пространства гиперграфов.) Дано бесконечное множество раскрасок $k$-элементных подмножеств множеств $\left\{1 \ldots n_{i}\right\}$ в в цветов, причём среди $n_{i}$ есть сколь угодно большие. Тогда можно раскрасить все $k$-элементные подмножества натурального ряда так, что для любого $N$ в этой раскраске цвета $k$-элементных подмножеств $\{1 \ldots N\}$ совпадают с их цветами из какой-то из исходных раскрасок. Эту лемму можно считать следствием леммы Кёнига о деревьях, даже компактности канторова множества (или множества $p$-адических чисел). Её не очень сложно доказать самому. Подобные рассуждение часто используются для того, чтобы свести бесконечный случай к конечному. Доказательство. [Вывод задачи из лемм.] Допустим, условие задачи не выполняется. Построим бесконечную последовательность из раскрасок подмножеств возрастающих нача натурального ряда, для которого не выполняется условие задачи. С помощью второй леммы, получим из неё раскраску $k$-элементных подмножеств всех натуральных чисел. По первой лемме в ней есть бесконечное подмножество, все $k$-элементные подмножества которого одного цвета. Из него можно выбрать конечное подмножество, удовлетворяющее условию задачи. Но оно целиком лежит в одной из выбранных вначале раскрасок, которые для которых условие задачи не должно выполняться. Противоречие, значит утверждение задачи верно. ## Серия E Решения задач 1-8 можно найти в статье В. Знака https://www.mccme.ru/free-books/matprosk.html С более полной версией решений можно будет ознакомиться на сайте Турнира Городов ## Van der Waerden's theorem and games The well-known Van der Waerden's theorem states: If the natural numbers are colored in finite number of colors then one can find a finite arbitrarily large arithmetic progression, all of the same color. The theorem can be reformulated as follows: Let $k$ and $l$ be natural numbers. Then there exists a number $N(k, l)$ such that for any natural $N>N(k, l)$ and any partitioning of the set $1, \ldots, N$ into $k$ subsets, one of the subsets contains an arithmetic progression of length $l$. Upper bounds for the numbers $N(k, l)$ are huge. For example, complexity of the upper bound for $N(k, 3)$ is an exponent of the exponent etc. $k$ times taken. For $N(k, 4)$, instead of the exponent, the function $N(k, 3)$ is iterated, and so on. (The best estimates were obtained quite recently and by non-elementary methods). Lower estimates are only exponential. In particular, the project at the Summer Conference of the Tournament of Towns in 1995 was devoted to this theorem. Its proof is not our goal. Related questions on which is suggested to think for participants are: 1. Is it possible to color the set of natural numbers so that there would be no infinite monochromatic arithmetic progression? 2. Prove an exponential lower bound: $N(s, l)>s^{l / 2}$. Implement an explicit construction. 3. Is it possible to color black more than half of the vertices of some regular $n$-gon so that the union of the original coloring and any ten its rotations will not make all vertices black? Try to get the best estimates for $n$ as well as explicit constructions. Szemeredi's theorem (also known as Erdös-Turan conjecture, 1936) states that if $A$ is an arbitrary set of natural numbers and the sequence $a_{n}=\frac{A \cap\{1, \ldots, N\}}{N}$ has a subsequence with a nonzero limit point then for any natural $k>3$ the set $A$ contains an arithmetic progression of length $k$. This is one of the main results of Szemeredi (obtained in 1975), for which he received the Abel Prize. For progressions of length 3 this result was obtained by Klaus Roth in 1953. https://en.wikipedia.org/wiki//Szemer\�\�di\'s theorem. One of the main results of Terence Tao was the proof that among prime numbers there is an arbitrarily long arithmetic progression. The result of I.D.Shkhedov refers to finding of a black corner in the square $N \times N$ when a certain fraction of cells is colored black. (http://www.mathnet.ru/php/archive.phtml?wshow=paper\&jrnid=im\&paperid=912\&option_lang=rus). Our main goal is to look at this topic from the game point of view. Perhaps this alternative view will enable us to better understand the issues related to the theorems of Van der Waerden and Szemerédi and to obtain new results, in particular, the estimates in the Van der Waerden's theorem. ## Series A 1. Two players play the following game. On a both-ways endless checkered strip, the first player puts one cross into a free cell, and the second player puts 1000 zeroes in free cells. The goal of the first player is to create an arithmetic progression of crosses of length 3, and the goal of the second player is to prevent him. A) Prove that the first player has a winning strategy. B) Prove that there is a number $n$ such that the first player safely can finish the game in $n$ moves. C)Try to estimate the minimum number of moves necessary for the first to surely win under the assumptions above. D)Solve previous task for one-way strip. 2. Two players are playing the foloowing game on an infinite checkered plane. The first one puts a cross in a free cell, the second one answers by putting 1000 zeros. The goal of the first player is to create an isosceles rectangular triangle of crosses whose sides are parallel to the coordinate axes. A) Prove that the first player has a winning strategy B) Find the minimum number of moves necessary for the guaranteed win of the first. 3. Can the first player create an arithmetic progression of length 4 under the assumptions of the problem 1 ? 4. Under the assumptions of problem 2, can the first player construct a square? 5. Suppose that in the game on the strip the first player puts one cross, and the second one answers by putting $n$ zeros. Can the first one create an arithmetic progression of length $k$ ? 6. Estimate the minimum number of moves necessary for the first player to win in the previous problem. 7. Now two players play a game on the plane. The first one puts one red dot, the second one answers by putting $k$ blue dots. Prove that for any set $F$ that consists of $n$ points, the first player can create a figure $F^{\prime}$ similar to $F$ (that is, obtained from $F$ by a composition of homothety and parallel translation), all of whose vertices are red. ## Series B Generalized van der Waerden theorem: Let rational points of the plane be painted in a finite number of colors. Then for any figure $M$ consisting of a finite number of rational points, one can find a one-color figure $M^{\prime}$ similar to $M$ (that is, obtained from $M$ by a composition of homothety and parallel transport) consisting of rational points. 1. Prove that in the formulation of the generalized Van der Waerden's theorem, the concept of similar set can not be replaced by the concept of congruent (or equal set, according to the terminology of the modern school program). In other words, a one-color set congruent to the set $M$ not becessarily exists. And what if we consider the group of transformations preserving the area, or hyperbolic rotations (namely, transformations that shrink the $X$-axis and dilate $Y$-axis with the same factor? 2. The rational points of the plane are colored in several colors. Prove that there is a right triangle with sides parallel to the coordinate axes and monochrome vertices. (For solving this problem, Van der Waerden's theorem may be used.) 3. (** The solution of this problem is unknown.) Under the assumptions of the previous problem, prove that there is a monochrome rectangle of area 1 with monochrome vertices. We call a simplex in the $n$-dimensional space standard if it has $n$ edges parallel to the coordinate axes. 4. Rational points of the $n$-dimensional space are colored in several colors. Prove that there is a standard simplex with monochrome vertices. (For solving this problem, Van der Waerden's theorem may be used.) 5. Formulate and prove the game form of the previous items. The second problem of this series reduced to the game form is solvable. In this connection, the question arises: 6. Formulate and solve the game form of the problem about the image of a given finite set under the action of hyperbolic rotations. ## Series C 1. There is an unlimited number of boards $8 \times 8$. The first player puts 2 chips on any of them, the second one answers setting one black chip on any of the boards. a) Can the first player pave one of the boards for sure? b) What is the smallest number of moves sufficient for the first player to do this? 2. Two play tic-tac-toe on a one-way endless checkered strip. The first one puts two crosses at each move, the second one replies with a zero. Can the first player obtain 100 consecutive crosses in a row? 3. Under the assumptions of the first, can the first player obtain the above result in $2^{45}$ moves? 4. And in $2^{90}$ turnes? ## Series D Two painters play the following game. The first one marks a point on a plane and connects it with some points that were marked before. Previously drawn arcs cannot be crossed. The second painter colors the point just marked by the first painter in such a manner that no two points of the same color are connected by an arc. 1. Can the first painter make the second one use more than $n$ colors? 2. Can the first painter achieve that in polinomial (in $n$ ) number of moves? 3. A tree is drawn. Two players paint its vertices in turn so that no two neighbouring vertices are of the same color. Can the first player make the second one use more than four colors? 4. Under assumptions of the previous problem, can the first player make the second one use more than three colors? 5. Let $G$ be a bipartite graph. Two players in turn color its vertices so that any two neighbouring vertices are of different color. The second player tries to use the minimum possible number of colors, and the first one wants to make him use the maximum possible number of colors as possible. Is it true that for any natural $N$ this bipartite graph $G$ can be chosen so that $N$ is not a sufficient number of colors for the second player? ## Series E 1. A complete graph with $n$ vertices is given. Two players paint its edges in turn. Theirst player colors one edge red, the other one colors 100 edges blue, and this is repeated until all edges are painted. Can the first player obtain for sure for sufficiently large $n$ a complete subgraph with 100 vertices all whose edges are red? 2. $99 \%$ of a complete graph's edges are colored red. Given sufficiently large $n$, does there exist a complete subgraph on 1000 vertices with red edges? 3. Every subset with $k$ elements in the set $\{1, \ldots, n\}$ is colored into one of $s$ colors (for example, for $k=2$ we get a coloring of edges of a complete graph with $n$ vertices). Prove that for sufficiently large $n$ there exists a subset $U \subset\{1, \ldots, n\}$ such that all its subsets having $k$ elements are of the same color, and if $x$ is the minimal element of $U$ then the number of elements in $U$ is at least $\exp (x)+s$. [^0]: ${ }^{1}$ Проект подготовлен Н.Белуховым и А.Заславским. Представлен на Конференции А.Заславским, О.Заславским, П.Кожевниковым и Д.Крековым. ${ }^{2} \mathrm{~B}$ дальнейшем в формулировках задач слова "Докажите, что" будут опускаться. [^1]: ${ }^{3}$ Задачи 2.2-2.11 этого раздела и их решения взяты из неопубликованной работы Н.Белухова [2]. [^2]: ${ }^{1}$ Проект подготовлен Н.Белуховым и А.Заславским. Представлен на Конференции А.Заславским, О.Заславским, П.Кожевниковым и Д.Крековым. [^3]: ${ }^{1}$ Problems and solutions by N. Beluhov and A. Zaslavsky. Conference presentation by A. Zaslavsky, P. Kozhevnikov, D. Krekov, and O. Zaslavsky. ${ }^{2}$ In all problems of the form "Show that so-and-so holds" to follow, the words "Show that" are omitted. [^4]: $[2]$. ${ }^{3}$ Problems 2.2-2.11 in this section and their solutions from N. Beluhov's unpublished manuscript [^5]: ${ }^{1}$ Problems and solutions by N. Beluhov and A. Zaslavsky. Conference presentation by A. Zaslavsky, P. Kozhevnikov, D. Krekov, and O. Zaslavsky. [^6]: *School of Mathematical Sciences, Queen Mary, University of London, London E1 4NS, England ${ }^{\dagger}$ r.johnson@qmul.ac.uk ${ }^{\ddagger}$ Department of Pure Mathematics and Mathematical Statistics, Centre for Mathematical Sciences, Wilberforce Road, Cambridge CB3 0WB, England. ${ }^{\S}$ I. Leader@dpmms . cam . ac . uk ${ }^{m}$ m. walters@qmul.ac.uk [^7]: ${ }^{1}$ Это только изложение общепринятой теории и не претендует на истину в последней инстанции. ${ }^{2}$ Доказательство одного очень близкого утверждения составляет одну из Проблем Тысячелетия, сущность которой мы тоже постараемся объяснить. [^8]: ${ }^{3}$ Кварков может быть и больше: частицы из 4 кварков и антикварков были открыты в 2014 , из 5 - в 2015 . [^9]: ${ }^{4}$ Команда с максимальным числом заполненных клеток к промежуточному финишу получит такое же число апельсинов. [^10]: ${ }_{5}^{5}$ Данный раздел с небольшими изменениями заимствован из книги [EMZ]. [^11]: ${ }^{6}$ А если выпадет 1 очко, то это "кризис", и вы теряете весь свой вклад. Кризис есть кризис. [^12]: ${ }^{1}$ This is only an exposition of a commonly accepted theory and does not pretend to be the final truth about particles. ${ }^{2}$ Actually the proof of a related assertion is one of the Millenium problems, the essence of which we shall also explain. [^13]: ${ }^{3}$ There can be more quarks: particles from 4 quarks or antiquarks were discovered in 2014, from 5 - in 2015. [^14]: ${ }^{4}$ The team with the maximal number of completed entries at the semifinal will receive the same number of apples. [^15]: ${ }^{5}$ This section is taken from [EMZ] with small changes. [^16]: ${ }^{1}$ Формально говоря, если в $(n, k)$-продукте раскрыть скобки по распределитедьному закону, то получится многочлен от переменных $x_{1}, \ldots, x_{n}$. ${ }^{2}$ На теоретико-множественном языке определение ( $n, k$ )-продукта можно сформулировать так. Пусть $A$ - множество из $n$ элементов. Тогда $(n, k)$-продукт - это неупорядоченная совокупность $k$ неупорядоченных непересекающихся пар элементов множества $A$. [^17]: ${ }^{3}$ На самом деле, сформулированное условие является критерием того, что продукты, соответствующие строкам множества $W$, образуют упаковку. Читатель приглашается самостоятельно доказать это утверждение. [^18]: ${ }^{1} 1$. Formally, when expanding the brackets in $(n, k)$-product according to the distribution rule the result will be the polynomial in variables $x_{1}, \ldots, x_{n}$. ${ }^{2}$ In set-theoretical language the definition of $(n, k)$-product can be formulated as following. Let $A$ be the set of $n$ elements. Then $(n, k)$-product is an unordered set of $k$ unordered disjoint pairs of elements of the set $A$. [^19]: ${ }^{3}$ In fact, the formulated condition is a criterion for the big products corresponding to the words of the set $W$ to form a rigid packing. A curious reader is invited to prove this criterion. [^20]: ${ }^{1}$ Часто происходит обратное: формулировки красивых результатов и важных проблем, ради которых была придумана теория, приводятся только после продолжительного изучения этой теории (или не приводятся совсем). Это способствует появлению представления о математике как науке, изучающей немотивированные понятия и теории. Такое представление принижает ценность математики. [^21]: ${ }^{2}$ Это «линейные» версии непланарности графов $K_{5}$ и $K_{3,3}$. Но доказываются они проще, ибо не требуют леммы о четности 1.4.4. ${ }^{3}$ Общепринятый термин для понятия графа, данного здесь - граф без петель и кратных ребер или простой граф. Графы $G_{1}$ и $G_{2}$ называются изоморфными, если существует взаимно однозначное отображение $f$ множества $V_{1}$ вершин графа $G_{1}$ на множество $V_{2}$ вершин графа $G_{2}$, удовлетворяющее следующему условию: вершины $A, B \in V_{1}$ соединены ребром тогда и только тогда, когда вершины $f(A), f(B) \in V_{2}$ соединены ребром. [^22]: ${ }^{4}$ Поскольку для плоского графа с $n$ вершинами и $e$ ребрами выполнено $e \leq 3 n-6$, и поскольку существуют планарные графы с $e=3 n-6$, «сложность» по количеству ребер «такая же», как по количеству вершин. [^23]: ${ }^{5}$ Благопристойность - свойство именно неупорядоченной пары кавалеров и жены. Например, если $A, B$ могут танцевать с женой $C$ и даже $B, C$ могут танцевать с женой $A$, то не обязательно $C, A$ могут танцевать с женой $B$. А вот переформулировка на математическом языке. Не существует пять семейств $A_{1}, \ldots, A_{5}$ (неупорядоченных) пар (различных элементов) из $\{1,2,3,4,5\}$ таких, что - никакое $j \in\{1,2,3,4,5\}$ не входит ни в одну пару из $A_{j}$, и - для любых четырех различных чисел $i, j, k, l \in\{1,2,3,4,5\}$ количество элементов в $\{i, j\}$, лежащих между $k$ и $l$, имеет ту же четность, что и [^24]: ${ }^{6}$ Если Вы выводите теорему Жордана из формулы Эйлера, то подумайте, как доказывается формула Эйлера. Будьте осторожны: в доказательстве теоремы 1.2 из [Pr04] лемма о четности 1.4.4 используется неявно, когда в первом предложении на стр. 20 написано, что четность меняется непрерывно. 7 Это нетривиально, поскольку ломаные могут иметь самопересечения, и поскольку теорема Жордана 1.4.3.b нетривиальна. Выводить лемму о четности из теоремы Жордана или формулы Эйлера неразумно, ибо их доказательства используют лемму о четности. Будьте осторожны: утверждение $[\operatorname{Pr} 04$, задача 1.2] неверно для «восьмерки» и треугольника, проходящего через точку самопересечения «восьмерки». [^25]: ${ }^{8}$ Компьютерные версии рисунков 3,4 и 10 были подготовлены И. Деркачем, В. Ковыршиной и Ю. Тихоновым. [^26]: ${ }^{9}$ При доказательстве леммы нельзя использовать без обоснования то, что любые два отображения общего положения одного графа в плоскость можно перевести друг в друга преобразованиями Райдемайстера на рис. 9. Впрочем, доказать лемму проще, чем этот факт. [^27]: ${ }^{10}$ Также это можно вывести из 2.1.6.d и 2.1.7.а ![](https://cdn.mathpix.com/cropped/2024_04_28_76011221cbab4e24c027g-218.jpg?height=46&width=1594&top_left_y=2362&top_left_x=298) [Sk16, §3.4]. [^28]: ${ }^{*}$ We are grateful to I. Bogdanov, G. Chelnokov, R. Karasev, Yu. Makarychev and M. Tancer for useful discussions, and to E. Morozov, A. Shchepinova and E. Streltsova for translation of certain parts of this text. A. Enne: Petrozavodsk State University. T. Zaitsev: Higher School of Economics. A. Ryabichev: Higher School of Economics, Independent University of Moscow. A. Skopenkov: Moscow Institute of Physics and Technology, Independent University of Moscow. Supported in part by RFBR, Grants No. 15-01-06302, the D. Zimin Dynasty foundation, and the Simons-IUM fellowship. Email: skopenko@mccme.ru. http://www.mccme.ru/ skopenko. [^29]: ${ }^{1}$ Often the opposite happens, when the formulations of beautiful results or important problems, for which the theory is developed, are given after long time of learning this theory (or not given at all). This approach promotes the wrong view that mathematics studies unmotivated notions and theories. Such a promotion belittles the value of mathematics. [^30]: ${ }^{2}$ These are 'linear' versions of the nonplanarity of the complete graph $K_{5}$ on 5 vertices and the bipartite graph $K_{3,3}$, see fig. 1, left. But they can be proved easier because the Parity Lemma 1.4.4 is not required for the proof. ${ }^{3}$ The first sentence indeed follows by the 'moreover' part: put the points in general position by a small shift so that no intersection points of segments with disjoint vertices are added. (Or else, for non-generalposition points Proposition 1.2.1 is obvious: if points $A, B, C$ among given 5 points belong to one line, $B$ between $A$ and $C$, and $D$ is any other given point, then segments $A C$ and $B D$ intersect). ${ }^{4}$ The commonly term is a graph without multiple edges or loops or a simple graph. The graphs $G_{1}$ and $G_{2}$ are called isomorphic if there is a 1-1 correspondence $f: V_{1} \rightarrow V_{2}$ between the set $V_{1}$ of vertices of $G_{1}$ and the set $V_{2}$ of the vertices of $G_{2}$ such that vertices $A, B \in V_{1}$ are adjacent in $G_{1}$ if and only if $f(A), f(B) \in V_{2}$ are adjacent in $G_{2}$. [^31]: ${ }^{5}$ The 'complexity' in the number of edges is 'the same' as the complexity in the number of vertices, because for a planar graph with $n$ vertices and $e$ edges we have $e \leq 3 n-6$ and there are planar graphs with $n$ vertices and $e$ edges such that $e=3 n-6$. [^32]: ${ }^{6}$ If you deduce the Jordan Theorem from the Euler Formula, think how to prove the Euler Formula. ${ }^{7}$ This is not trivial because the polygonal lines may have self-intersections and because the Jordan Theorem 1.4.3.b is not obvious. It is not reasonable to deduce the Parity Lemma from the Jordan Theorem or the Euler Theorem because this will make a vicious circle. [^33]: ${ }^{8}$ The computer versions of Figures 3, 4 and 10 was prepared by I. Derkach, V. Kovirshina and Yu. Tikhonov. [^34]: ${ }^{9}$ This also may be deduced from Problems 2.1.6.d and 2.1.7.a ${ }^{10}$ Statements 2.1.4.b and 3.1.2.e can also be proved analogously to the proof that $v(f)=o(f)$ for $r=2$ in [Sk16, §3.4].