# Покрасьте мою шляпу в 3,5 цвета!
K. Кохась, А. Латышев
Задачу представляют: О. Бурсиан, Д. Кохась Д, К. Кохась, В. Ретинский
## 1 Знакомьтесь: игра в шляпы
Пусть дан неориентированный граф $G$, в каждой вершине которого находится один мудрец и один сундук со шляпами разного цвета. Мудрецы знакомы друг с другом, граф $G$, расположение мудрецов по вершинам графа и все цвета шляп в сундуках зафиксированы и известны всем. В частности, каждый мудрец понимает, в какой вершине находится каждый из остальных мудрецов. Судья проводит с мудрецами следующий тест. Он надевает каждому мудрецу шляпу из его сундука. Каждый мудрец видит только шляпы мудрецов, находящихся в соседних вершинах графа, но не видит своей шляпы и не знает ее цвета. Мудрецы не общаются во время теста. По команде судьи мудрецы одновременно записывают каждый на своей бумажке названия нескольких цветов (сколько именно цветов должен указывать мудрец, определяется дополнительным правилом). Считается, что мудрецы успешно прошли тест $=$ «выиграли», если хотя бы один из них угадал хотя бы в одной из своих попыток.
Перед тестом мудрецам сообщили правила теста и дали возможность устроить совещание, на котором они должны определить публичную стратегию. Публичность означает, что все, включая судью, знают ее. Стратегия мудрецов должна быть детерминированной - каждый мудрец должен записывать на бумажку цвета, исходя только из того, какие цвета он видит у соседей. Будем говорить, что стратегия выигрышная, если при любой раскладке шляп хотя бы один мудрец угадает цвет надетой на него шляпы, т. е. упомянет этот цвет в своем списке. Также будем говорить, что мудрецы выигрывают, если они имеют выигрышную стратегию, и проигрывают, если не имеют.
Таким образом, игра в шляпы не есть игра в привычном понимании этого слова. Она длится всего один ход. Вот несколько задач на этот сюжет.
1.1. Каждому из двух мудрецов надевают шляпу белого, синего, красного или зеленого цвета. Каждый из них видит шляпу другого, но не видит свою. Они должны одновременно попытаться угадать цвет своей шляпы, написав на своей бумажке два цвета. Докажите, что мудрецы могут заранее договориться действовать так, чтобы хотя бы один из них угадал.
1.2. Каждому из двух мудрецов надевают шляпу одного из пяти возможных цветов. Каждый из них видит шляпу другого, но не видит свою. Они должны одновременно попытаться угадать цвет своей шляпы, написав на бумажке: первый - какие-то два цвета, второй какие-то три цвета. Докажите, что мудрецы могут заранее договориться действовать так, чтобы хотя бы один из них угадал.
1.3. Пять мудрецов стоят по кругу возле непрозрачного баобаба. Шах надел на голову каждого из мудрецов красную, синюю, желтую или зеленую шляпу. Мудрец не знает цвет своей шляпы и видит только двух соседних по кругу мудрецов. Не общаясь, мудрецы одновременно должны высказать предположение о цвете своей шляпы. При этом они боятся оказаться слишком удачливыми. Как им следует действовать, чтобы при любом раскладе шляп цвет своей шляпы угадало не более двух мудрецов?
1.4. Султан устроил экзамен шести придворным мудрецам. По правилу экзамена султан размещает 5 мудрецов в 5 ям, расположенных по кругу, а шестого мудреца сажает на вышку в центре круга. На лбу у каждого из первых пяти мудрецов султан пишет число 1,2 или 3 ; на лбу у центрального мудреца султан пишет число от 1 до 243. Мудрец на вышке видит числа на всех остальных мудрецах, а те видят его число (но не видят друг друга). Все мудрецы должны одновременно попытаться угадать свои числа: для этого мудрецы в яме должны назвать по два числа, а мудрец на вышке - одно. Султан заранее объяснил мудрецам правила экзамена и дал им время посовещаться до начала экзамена. Могут ли мудрецы действовать так, чтобы хотя бы один из них заведомо угадал свое число?
Мы будем отождествлять вершину графа $G$ и мудреца, который в ней находится. Будем считать, что цвета пронумерованы числами $0,1,2,3, \ldots$ и что в сундуке мудреца $v$ лежат шляпы с цветами от 0 до некоторого числа $h(v)-1$.
Игрой в шляпи назовем тройку $\langle G, h, g\rangle$, где $G=\langle V, E\rangle-$ граф, $h: V \rightarrow \mathbb{N}$ - функция, сопоставляющая вершине количество (цветов) шляп, лежащих в сундуке в этой вершине, $g: V \rightarrow \mathbb{N}$ - функция, показывающая, сколько попыток угадывания имеет каждый мудрец. Функцию $h$ мы будем называть «шляпностью», а $g$ - функцией угадывания или числом попыток. Если $h$ - натуральное число, символом $\star h$ будем обозначать функцию на $V$, принимающую постоянное значение $h$. Вместо обозначения $\langle G, h, \star 1\rangle$ мы будем использовать более короткое обозначение $\langle G, h\rangle$.
1.5. Докажите, что если игра $\langle G, h, g\rangle$ выигрышная, то для всех натуральных $k$ игра $\langle G, k \cdot h, k \cdot g\rangle$ тоже выигрышная.
1.6. Дана игра $\langle G, h, g\rangle$. Пусть $K \subset G$ - антиклика (множество вершин, между которыми не проведено ни одного ребра) и для всех $v \in K h(v)>g(v)$. Докажите, что существует расклад шляп, для которого ни один мудрец из $K$ не угадывает.
1.7. Пусть $h$ и $g$ натуральные числа, $G=\langle G, \star h, \star g\rangle$ - выигрышная игра, $r^{\prime} \leqslant \frac{h}{g}$ - рациональное число. Докажите, что существуют натуральные числа $h^{\prime}$ и $g^{\prime}$, для которых $\frac{h^{\prime}}{g^{\prime}}=r^{\prime}$ и игра $\left\langle G, \star h^{\prime}, \star g^{\prime}\right\rangle$ - выигрышная.
1.8. Сформулируйте и докажите аналог предыдущего утверждения для непостоянных функций шляпности и угадывания.
1.9. Обозначим через $K_{n}$ полный граф на $n$ вершинах. Докажите, что в игре $\left\langle K_{n}, h, g\right\rangle$ мудрецы выигрывают тогда и только тогда, когда $\sum_{v \in K_{n}} \frac{g(v)}{h(v)} \geqslant 1$.
## 2 Пути и деревья
Теория игры в шляпы на полном графе из трех вершин исчерпывается утверждением задачи 1.9. Рассмотрим игру на более простом графе - пути $P_{3}$.
2.1. Докажите, что в игре $\left\langle P_{3}, \star 3, \star 1\right\rangle$ мудрецы проигрывают.
2.2. а) Докажите, что игра ${ }^{\frac{3}{10}} \frac{3}{10} \frac{3}{5}$ - выигрышная (числитель дроби возле вершины обозначает количество попыток, а знаменатель - шляпность).
б) Докажите, что игра $\stackrel{\frac{3}{11}}{\bullet}{ }^{\frac{3}{10}} \frac{3}{5}-$ проигрышная.
в) Докажите, что игра $\overbrace{}^{\frac{s}{t(s)}} \cdot \frac{s}{t(s)} \cdot \frac{s}{t(s)}$ где $t(s)=s^{2}+s+1,-$ проигрышная.
Пусть дан граф $G$ и натуральное число $s$. Наибольшее число шляп $h$, для которого игра $\langle G, \star h, \star s\rangle$ выигрышная, называется $s$-шляпным числом графа и обозначается $\mathrm{HG}_{s}(G)$. При $s=1$ мы называем это число просто шляпным числом и обозначаем $\operatorname{HG}(G)$.
2.3. Докажите, что для любых натуральных $n$ и $s$ игра $\stackrel{\frac{s}{2 s}}{A} \cdot \cdots$ на пути $P_{n}$ проигрышная. Все вершины, кроме крайней вершины $A$, имеют шляпность $4 s-1$ и $s$ попыток.
2.4. Докажите, что существуют пути $P_{n}$, для которых $\mathrm{HG}_{2}\left(P_{n}\right)=6, \mathrm{HG}_{3}\left(P_{n}\right)=10, \mathrm{HG}_{4}\left(P_{n}\right)=14$.
2.5. Докажите, что для любого натурального $s$ игра $\left\langle P_{n}, \star(4 s-2), \star s\right\rangle$ выигрышная при $n \geqslant 2 s$.
2.6. а) Пусть $t(s)=s^{2}+s+1$. Докажите, что для любого дерева $T$ игра $\langle T, \star t(s), \star s\rangle$ проигрышная.
b) Пусть $K_{1, n}$ - граф «звезда» (дерево, состоящее из корня и $n$ листьев). Докажите, что при больших $n$ игра $\left\langle K_{1, n}, \star\left(s^{2}+s\right), \star s\right\rangle$ - выигрышная.
c) Докажите, что для любого натурального $h$ существует такое натуральное $n$, что игра на графе $K_{1, n}$ - выигрышная, если шляпности всех периферийных мудрецов равны 3 и все они имеют по одной попытке угадывания, а шляпность центрального мудреца равна $h$ и у него две попытки.
## 3 Конструкторы
3.1. Пусть $\langle G, h, g\rangle$ - выигрышная игра, $A_{1}$ и $A_{2}$ - вершины графа $G$, не соединенные ребром, причем $h\left(A_{1}\right)=h\left(A_{2}\right), \widetilde{G}$ - граф, получающийся из $G$ склеиванием вершин $A_{1}$ и $A_{2}$ в одну вершину $A$. Пусть функции $\widetilde{h}, \widetilde{g}$ на вершинах графа $\widetilde{G}$ совпадают с $h$ и $g$ во всех вершинах, кроме $A_{1}$ и $A_{2}$, и $\widetilde{h}(A)=h\left(A_{1}\right), \widetilde{g}(A)=g\left(A_{1}\right)+g\left(A_{2}\right)$. Тогда игра $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle$ тоже выигрышная. (И тогда, если игра $\langle\widetilde{G}, h, \widetilde{g}\rangle$ проигрышная, то и $\langle G, h, g\rangle$ проигрышная.)
Пусть $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}, g_{1}\right\rangle, \mathcal{G}_{2}=\left\langle G_{2}, h_{2}, g_{2}\right\rangle$ - две игры, такие что $V_{1} \cap V_{2}=\{v\}$. Пусть $G=G_{1}+_{v} G_{2}$ - объединение графов $G_{1}$ и $G_{2}$, в котором обе вершины $v$ объединены в одну вершину. Зададим функции $h, g: V_{1} \cup V_{2} \rightarrow \mathbb{N}$ :
$$
h(u)=\left\{\begin{array}{ll}
h_{i}(u), & u \in V_{i} \backslash\{v\},(i=1,2), \\
h_{1}(v) h_{2}(v), & u=v,
\end{array} \quad g(u)= \begin{cases}g_{i}(u), & u \in V_{i} \backslash\{v\},(i=1,2) \\
g_{1}(v) g_{2}(v), & u=v\end{cases}\right.
$$
Игру $\mathcal{G}=\langle G, h, g\rangle$ будем обозначать $\mathcal{G}_{1} \times{ }_{v} \mathcal{G}_{2}$ (рис. 1).

Рис. 1. Игра $G_{1} \times{ }_{v} G_{2}$
3.2. Теорема о произведении. Если мудрецы выигрывают в играх $\mathcal{G}_{1}$ и $\mathcal{G}_{2}$, то они выигрывают и в игре $\mathcal{G}_{1} \times{ }_{v} \mathcal{G}_{2}$.
3.3. Пусть $G=G_{1}+_{A} G_{2}$, где $G_{1}$ и $G_{2}$ - графы, у которых $V\left(G_{1}\right) \cap V\left(G_{2}\right)=\{A\}$, а игры $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}, g_{1}\right\rangle$ и $\mathcal{G}_{2}=\left\langle G_{2}, h_{2}, g_{2}\right\rangle$ - проигрышные, и при этом
$$
g_{1}(A)=g_{2}(A)=s, \quad h_{1}(A) \geqslant h_{2}(A)=s+1
$$
Тогда игра $\mathcal{G}=\left\langle G_{1}+{ }_{A} G_{2}, h, g\right\rangle$ - проигрышная, где
$$
h(x)= \begin{cases}h_{i}(x), & x \in V_{i} \backslash\{A\}, \quad(i=1,2), \\ h_{1}(A), & x=A\end{cases}
$$
3.4. Удаление полуребра. Пусть $\langle G, h, g\rangle$ - выигрышная игра, $A B$ - ребро графа $G, \widetilde{G}$ - граф, получающийся из $G$ заменой ребра $A B$ на ориентированное ребро $B \rightarrow A$ (те. мудрец $A$ не видит мудреца $B$, но $B$ видит $A$ ). Пусть функция $\widetilde{g}$ на вершинах графа $G$ совпадает с $g$ во всех вершинах, кроме $A$, и $\widetilde{g}(A)=h(B) g(A)$. Тогда игра $\langle\widetilde{G}, h, \widetilde{g}\rangle$ тоже выигрышная. (И тогда, если игра $\langle\widetilde{G}, h, \widetilde{g}\rangle$ проигрышная, то и $\langle G, h, g\rangle$ проигрышная.)
3.5. Теорема о подстановке. Пусть $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}, g_{1}\right\rangle$ и $\mathcal{G}_{2}=\left\langle G_{2}, h_{2}, g_{2}\right\rangle$ - выигрышные игры. Пусть $A$ - произвольная вершина графа $G_{2}$. Рассмотрим новый граф $G$, который получается из $G_{2}$ заменой вершины $A$ на граф $G_{1}$ (каждая вершина $G_{1}$ соединяется новыми ребрами с бывшими соседями вершины $A$, см. рис. 2). Тогда игра $\langle G, h, g\rangle$ - выигрышная, где
$$
h(u)=\left\{\begin{array}{ll}
h_{2}(u), & u \in V\left(G_{2}\right) \backslash\{A\}, \\
h_{1}(u) h_{2}(A), & u \in V\left(G_{1}\right),
\end{array} \quad g(u)= \begin{cases}g_{2}(u), & u \in V\left(G_{2}\right) \backslash\{A\} \\
g_{1}(u) g_{2}(A), & u \in V\left(G_{1}\right)\end{cases}\right.
$$
3.6. Подстановка с сокращением. Пусть $\mathcal{G}=\langle G, h, \star s\rangle, \mathcal{G}^{\prime}=\left\langle G^{\prime}, h^{\prime}, g^{\prime}\right\rangle$ - выигрышные игры. Пусть $A$ - вершина графа $G^{\prime}$, причем $h^{\prime}(A)=s$. Пусть $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle-$ выигрышная игра, получающаяся

Рис. 2. Подстановка графа на место вершины $A$.
подстановкой игры $\mathcal{G}$ на место вершины $A$ в игру $\mathcal{G}^{\prime}$ (как в задаче 3.5). По правилу подстановки для всех подставленных вершин $v$
$$
\widetilde{h}(v)=h(v) h^{\prime}(A)=s \cdot h(v), \quad \widetilde{g}(v)=g(v) g^{\prime}(A)=s \cdot g^{\prime}(A)
$$
Рассмотрим новые функции $h^{*}, g^{*}$ на графе $\widetilde{G}$, которые отличаются от $\widetilde{h}, \widetilde{g}$ только значениями в подставленных вершинах $v$, и это отличие - сокращение на $s$ :
$$
h^{*}(v)=h(v), \quad g^{*}(v)=g^{\prime}(A)
$$
Тогда игра $\left\langle\widetilde{G}, h^{*}, g^{*}\right\rangle$ - тоже выигрышная.
3.7. Раздувание вершины. Пусть $\mathcal{G}=\langle G, h, g\rangle$ - выигрышная игра, $A \in V(G), \widetilde{G}$ - граф, получающийся из $G$ подстановкой на место вершины $A$ клики $B$, состоящей из $g(A)$ вершин. Тогда игра $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle$ тоже выигрышная, где
$$
\widetilde{h}(v)=\left\{\begin{array}{ll}
h(v), & v \in V(G) \backslash\{A\} \\
h(A), & v \in B
\end{array} \quad \widetilde{g}(v)= \begin{cases}g(v), & v \in V(G) \backslash\{A\} \\
1, & v \in B\end{cases}\right.
$$
## 4 «Лепестки» и «етунии»
Лепестком будем называть граф, изображенный на рис. 3 , вершину лепестка степени $n-1$ (т. е. верхнюю вершину на рис. 3) будем называть черенком лепестка.
Будем строить из лепестков более сложные графы. Возьмем два лепестка $L_{1}$ и $L_{2}$, обозначим в каждом из них одну из вершин буквой $v_{1}$ и построим граф $M_{2}=L_{1}+_{v_{1}} L_{2}$. Возьмем граф $M_{2}$ и лепесток $L_{3}$, обозначим в каждом из них одну из вершин буквой $v_{2}$ и рассмотрим граф $M_{3}=M_{2} t_{v_{2}} L_{3}$ и т. д. Графы, которые можно получить с помощью конечного числа таких операций, будем называть петуниями.
Петунию будем называть королевской петунией (рис. 4), если на каждом шаге ее построения вершина $v_{i}$ была черенком лепестка $L_{i+1}$.
4.1. Пусть $G$ - лепесток из $n$ вершин, см. рис. 3 , а функция шляпности $h$ задается так: шляпность верхней вершины равна 2, шляпности остальных вершин равны 7. Докажите, что в игре $\langle G, h\rangle$ мудрецы проигрывают.

Рис. 3. Лепесток из $n$ вершин

Рис. 4. Королевская петуния
4.2. Пусть $G$ - лепесток из $n$ вершин. Положим $f(s)=s^{2}+s$. Докажите, что $\mathrm{HG}_{s}(G) \leqslant f(f(s))$.
4.3. Пусть $M$ - петуния, $h_{s}$ - наибольшее натуральное число, для которого игра $\left\langle M, \star h_{s}, \star s\right\rangle$ выигрышная. Докажите, что $h_{s} \leqslant f(f(f(s)))$.
4.4. а) Докажите, что $\mathrm{HG}_{s}(G)=4 s(s+1)-2$, где $G$ - лепесток из $n$ вершин, где $n$ достаточно велико (рис. 3).
b) Докажите это же равенство, если $G$ - королевская петуния с достаточно крупными лепестками.
## 5 Дополнительные задачи
## В серию 1
5.1. Пусть $K_{1, n}$ - граф «звезда», а $h=\left(h_{0}, \ldots, h_{n}\right), g=\left(g_{0}, \ldots, g_{n}\right)$ - произвольные функции количеств шляп и числа попыток, где $1 \leqslant g_{i} \leqslant h_{i}$ при всех $i$, нулевой индекс соответствует центральной вершине графа. Докажите, что существование $k$, для которого $\left\langle K_{1, n}, k \cdot h, k \cdot g\right\rangle$ является выигрышной, равносильно неравенству
$$
\frac{g_{0}}{h_{0}} \geqslant \prod_{i=1}^{n}\left(1-\frac{g_{i}}{h_{i}}\right)
$$
## В серию 2
5.2. Пусть $h$ и $g$ натуральные числа, причем $g^{2}-3 g h+h^{2}<0$. Докажите, что игра $\left\langle P_{3}, \star h, \star g\right\rangle-$ выигрышная.
Дробным шляпным числом графа $G$ назовем величину $\hat{\mu}(G)=\sup \left\{\frac{h}{g}\right.$ : игра $\langle G, \star h, \star g\rangle$ - выигрышная $\}$.
Как следует из задачи 1.9, $\mathrm{HG}\left(K_{n}\right)=\hat{\mu}\left(K_{n}\right)=n, \mathrm{HG}_{s}\left(K_{n}\right)=s n$. В общем случае, $\hat{\mu}\left(K_{n}\right) \geqslant \frac{1}{s} \mathrm{HG}_{s}(G) \geqslant \operatorname{HG}(G)$.
5.3. Докажите, что $\hat{\mu}\left(K_{3}\right)=\frac{3+\sqrt{5}}{2}$.
## В серию 3
5.4. Пусть $\left\langle G_{2}, h_{2}\right\rangle$ - проигрышная игра, $H_{2} \subset G_{2}$ - клика в $G_{2}$. Пусть $G_{1}$ полный граф. Зададим на нем функцию шляпности $h_{1}$ так, чтобы выполнялось соотношение
$$
\left(\sum_{u \in G_{1}} \frac{1}{h_{1}(u)}\right)\left(\prod_{v \in H_{2}} h_{2}(v)\right)<1
$$
Пусть $G$ - это граф, получающийся объединением графов $G_{1}$ и $G_{2}$ и добавлением всех ребер между вершинами $G_{1}$ и $H_{2}$ (рис. 5). Докажите, что игра $\langle G, h\rangle$ - проигрышная, если
$$
h(v)= \begin{cases}h_{1}(v), & v \in G_{1} \\ h_{2}(v), & v \in G_{2}\end{cases}
$$

Рис. 5. Пример к задаче 5.4. Число вершин в $G_{1}$ и $H_{2}$ не обязано совпадать
5.5. Пусть $\mathcal{G}=\langle G, h\rangle$ - проигрышная игра, $A$ - любая вершина графа $G$. Рассмотрим граф $G^{\prime}=\left\langle V^{\prime}, E^{\prime}\right\rangle$, получающийся добавлением к графу $G$ новой висячей вершины $B: V^{\prime}=V \cup\{B\}$, $E^{\prime}=E \cup\{A B\}$. Тогда мудрецы проигрывают в игре $\left\langle G^{\prime}, h^{\prime}\right\rangle$, где $h^{\prime}(B)=2, h^{\prime}(A)=2 h(A)-1$ и $h^{\prime}(u)=h(u)$ для остальных вершин $u \in V$.
5.6. Пусть $\mathcal{G}=\langle G, h, g\rangle$ - некоторая игра, $A \in V(G)$ соединена ребрами со всеми остальными вершинами графа $G, h(A)=s+1, g(A)=s$, и $\mathcal{G}^{\prime}=\left\langle G \backslash\{A\}, h^{\prime},(s+1) \cdot g^{\prime}\right\rangle$, где $h^{\prime}=\left.h\right|_{V(G) \backslash\{A\}}$, $g^{\prime}=\left.g\right|_{V(G) \backslash\{A\}}$. Тогда игры $\mathcal{G}$ и $\mathcal{G}^{\prime}$ эквивалентны (мудрецы выигрывают в $\mathcal{G}$ тогда и только тогда, когда они выигрывают в $\mathcal{G}^{\prime}$ ).
## Серия 5
5.7. а) Выигрышный граф $G$ содержит «длинный мост»- двузвенный путь $A B C$, при удалении которого граф распадается на две компоненты: $G_{1}$ (содержащую вершину $A$ ) и $G_{2}$ (содержащую вершину $C$ ). Пусть шляпность вершины $B$ равна 5. Докажите, что хотя бы одна из игр $\left\langle G_{1},\left.h\right|_{G_{1}}\right\rangle$, $\left\langle G_{2},\left.h\right|_{G_{2}}\right\rangle$ выигрышная.
b) Пусть граф $\widetilde{G}$ получен подразбиением произвольного графа $G$ (т. е. каждое ребро графа $G$ заменили на двузвенный путь). Докажите, что игра $\langle\widetilde{G}, \star 5\rangle$ - проигрышная.
5.8. Пусть $\mathcal{G}=\langle G, h\rangle$ - выигрышная игра, причем максимальная в следующем смысле: при увеличении функции шляпности в любой вершине мудрецы проигрывают и, кроме того, с функцией шляпности $h$ мудрецы не могут выиграть ни на каком подграфе графа $G$. Допустим, что граф $G$ содержит ребро-мост $A B$. Докажите, что игра $\mathcal{G}$ представима в виде произведения игр.
5.9. Мудрецы $A$ и $B$ имеют по одной попытке угадывания, видят друг друга и всех остальных мудрецов в графе (а те видят их), $h(A)=2, h(B)=3$. Докажите, что если заменить этих двух мудрецов на одного мудреца $C$, который видит остальных, а остальные его, и при этом $h(C)=6$, $g(C)=5$, то результат игры не изменится.
5.10. Даны натуральные числа $s$ и $d$. Пусть $G$ - произвольный граф, вершины которого разбиты на два множества $V(G)=A \cup B$, причем каждая вершина из $A$ имеет не более $d$ соседей из $B$. Докажите, что $\mathrm{HG}_{s}(G) \leqslant \mathrm{HG}_{s^{\prime}}(G[A])$, где $s^{\prime}=s\left(\mathrm{HG}_{s}(G[B])+1\right)^{d}$, а $G[A]$ и $G[B]$ - индуцированные подграфы на множествах $A$ и $B$.
# Покрасьте мою шляпу в 3,5 цвета!
K. Кохась, А. Латышев
Задачу представляют: О. Бурсиан, Д. Кохась Д, К. Кохась, В. Ретинский
## 1 Знакомьтесь: игра в шляпы
Пусть дан неориентированный граф $G$, в каждой вершине которого находится один мудрец и один сундук со шляпами разного цвета. Мудрецы знакомы друг с другом, граф $G$, расположение мудрецов по вершинам графа и все цвета шляп в сундуках зафиксированы и известны всем. В частности, каждый мудрец понимает, в какой вершине находится каждый из остальных мудрецов. Судья проводит с мудрецами следующий тест. Он надевает каждому мудрецу шляпу из его сундука. Каждый мудрец видит только шляпы мудрецов, находящихся в соседних вершинах графа, но не видит своей шляпы и не знает ее цвета. Мудрецы не общаются во время теста. По команде судьи мудрецы одновременно записывают каждый на своей бумажке названия нескольких цветов (сколько именно цветов должен указывать мудрец, определяется дополнительным правилом). Считается, что мудрецы успешно прошли тест $=$ «выиграли», если хотя бы один из них угадал хотя бы в одной из своих попыток.
Перед тестом мудрецам сообщили правила теста и дали возможность устроить совещание, на котором они должны определить публичную стратегию. Публичность означает, что все, включая судью, знают ее. Стратегия мудрецов должна быть детерминированной - каждый мудрец должен записывать на бумажку цвета, исходя только из того, какие цвета он видит у соседей. Будем говорить, что стратегия выигрышная, если при любой раскладке шляп хотя бы один мудрец угадает цвет надетой на него шляпы, т. е. упомянет этот цвет в своем списке. Также будем говорить, что мудрецы выигрывают, если они имеют выигрышную стратегию, и проигрывают, если не имеют.
Таким образом, игра в шляпы не есть игра в привычном понимании этого слова. Она длится всего один ход. Вот несколько задач на этот сюжет.
1.1. Каждому из двух мудрецов надевают шляпу белого, синего, красного или зеленого цвета. Каждый из них видит шляпу другого, но не видит свою. Они должны одновременно попытаться угадать цвет своей шляпы, написав на своей бумажке два цвета. Докажите, что мудрецы могут заранее договориться действовать так, чтобы хотя бы один из них угадал.
1.2. Каждому из двух мудрецов надевают шляпу одного из пяти возможных цветов. Каждый из них видит шляпу другого, но не видит свою. Они должны одновременно попытаться угадать цвет своей шляпы, написав на бумажке: первый - какие-то два цвета, второй какие-то три цвета. Докажите, что мудрецы могут заранее договориться действовать так, чтобы хотя бы один из них угадал.
1.3. Пять мудрецов стоят по кругу возле непрозрачного баобаба. Шах надел на голову каждого из мудрецов красную, синюю, желтую или зеленую шляпу. Мудрец не знает цвет своей шляпы и видит только двух соседних по кругу мудрецов. Не общаясь, мудрецы одновременно должны высказать предположение о цвете своей шляпы. При этом они боятся оказаться слишком удачливыми. Как им следует действовать, чтобы при любом раскладе шляп цвет своей шляпы угадало не более двух мудрецов?
1.4. Султан устроил экзамен шести придворным мудрецам. По правилу экзамена султан размещает 5 мудрецов в 5 ям, расположенных по кругу, а шестого мудреца сажает на вышку в центре круга. На лбу у каждого из первых пяти мудрецов султан пишет число 1,2 или 3 ; на лбу у центрального мудреца султан пишет число от 1 до 243. Мудрец на вышке видит числа на всех остальных мудрецах, а те видят его число (но не видят друг друга). Все мудрецы должны одновременно попытаться угадать свои числа: для этого мудрецы в яме должны назвать по два числа, а мудрец на вышке - одно. Султан заранее объяснил мудрецам правила экзамена и дал им время посовещаться до начала экзамена. Могут ли мудрецы действовать так, чтобы хотя бы один из них заведомо угадал свое число?
Мы будем отождествлять вершину графа $G$ и мудреца, который в ней находится. Будем считать, что цвета пронумерованы числами $0,1,2,3, \ldots$ и что в сундуке мудреца $v$ лежат шляпы с цветами от 0 до некоторого числа $h(v)-1$.
Игрой в шляпи назовем тройку $\langle G, h, g\rangle$, где $G=\langle V, E\rangle-$ граф, $h: V \rightarrow \mathbb{N}$ - функция, сопоставляющая вершине количество (цветов) шляп, лежащих в сундуке в этой вершине, $g: V \rightarrow \mathbb{N}$ - функция, показывающая, сколько попыток угадывания имеет каждый мудрец. Функцию $h$ мы будем называть «шляпностью», а $g$ - функцией угадывания или числом попыток. Если $h$ - натуральное число, символом $\star h$ будем обозначать функцию на $V$, принимающую постоянное значение $h$. Вместо обозначения $\langle G, h, \star 1\rangle$ мы будем использовать более короткое обозначение $\langle G, h\rangle$.
1.5. Докажите, что если игра $\langle G, h, g\rangle$ выигрышная, то для всех натуральных $k$ игра $\langle G, k \cdot h, k \cdot g\rangle$ тоже выигрышная.
1.6. Дана игра $\langle G, h, g\rangle$. Пусть $K \subset G$ - антиклика (множество вершин, между которыми не проведено ни одного ребра) и для всех $v \in K h(v)>g(v)$. Докажите, что существует расклад шляп, для которого ни один мудрец из $K$ не угадывает.
1.7. Пусть $h$ и $g$ натуральные числа, $G=\langle G, \star h, \star g\rangle$ - выигрышная игра, $r^{\prime} \leqslant \frac{h}{g}$ - рациональное число. Докажите, что существуют натуральные числа $h^{\prime}$ и $g^{\prime}$, для которых $\frac{h^{\prime}}{g^{\prime}}=r^{\prime}$ и игра $\left\langle G, \star h^{\prime}, \star g^{\prime}\right\rangle$ - выигрышная.
1.8. Сформулируйте и докажите аналог предыдущего утверждения для непостоянных функций шляпности и угадывания.
1.9. Обозначим через $K_{n}$ полный граф на $n$ вершинах. Докажите, что в игре $\left\langle K_{n}, h, g\right\rangle$ мудрецы выигрывают тогда и только тогда, когда $\sum_{v \in K_{n}} \frac{g(v)}{h(v)} \geqslant 1$.
## 2 Пути и деревья
Теория игры в шляпы на полном графе из трех вершин исчерпывается утверждением задачи 1.9. Рассмотрим игру на более простом графе - пути $P_{3}$.
2.1. Докажите, что в игре $\left\langle P_{3}, \star 3, \star 1\right\rangle$ мудрецы проигрывают.
2.2. а) Докажите, что игра ${ }^{\frac{3}{10}} \frac{3}{10} \frac{3}{5}$ - выигрышная (числитель дроби возле вершины обозначает количество попыток, а знаменатель - шляпность).
б) Докажите, что игра $\stackrel{\frac{3}{11}}{\bullet}{ }^{\frac{3}{10}} \frac{3}{5}-$ проигрышная.
в) Докажите, что игра $\overbrace{}^{\frac{s}{t(s)}} \cdot \frac{s}{t(s)} \cdot \frac{s}{t(s)}$ где $t(s)=s^{2}+s+1,-$ проигрышная.
Пусть дан граф $G$ и натуральное число $s$. Наибольшее число шляп $h$, для которого игра $\langle G, \star h, \star s\rangle$ выигрышная, называется $s$-шляпным числом графа и обозначается $\mathrm{HG}_{s}(G)$. При $s=1$ мы называем это число просто шляпным числом и обозначаем $\operatorname{HG}(G)$.
2.3. Докажите, что для любых натуральных $n$ и $s$ игра $\stackrel{\frac{s}{2 s}}{A} \cdot \cdots$ на пути $P_{n}$ проигрышная. Все вершины, кроме крайней вершины $A$, имеют шляпность $4 s-1$ и $s$ попыток.
2.4. Докажите, что существуют пути $P_{n}$, для которых $\mathrm{HG}_{2}\left(P_{n}\right)=6, \mathrm{HG}_{3}\left(P_{n}\right)=10, \mathrm{HG}_{4}\left(P_{n}\right)=14$.
2.5. Докажите, что для любого натурального $s$ игра $\left\langle P_{n}, \star(4 s-2), \star s\right\rangle$ выигрышная при $n \geqslant 2 s$.
2.6. а) Пусть $t(s)=s^{2}+s+1$. Докажите, что для любого дерева $T$ игра $\langle T, \star t(s), \star s\rangle$ проигрышная.
b) Пусть $K_{1, n}$ - граф «звезда» (дерево, состоящее из корня и $n$ листьев). Докажите, что при больших $n$ игра $\left\langle K_{1, n}, \star\left(s^{2}+s\right), \star s\right\rangle$ - выигрышная.
c) Докажите, что для любого натурального $h$ существует такое натуральное $n$, что игра на графе $K_{1, n}$ - выигрышная, если шляпности всех периферийных мудрецов равны 3 и все они имеют по одной попытке угадывания, а шляпность центрального мудреца равна $h$ и у него две попытки.
## 3 Конструкторы
3.1. Пусть $\langle G, h, g\rangle$ - выигрышная игра, $A_{1}$ и $A_{2}$ - вершины графа $G$, не соединенные ребром, причем $h\left(A_{1}\right)=h\left(A_{2}\right), \widetilde{G}$ - граф, получающийся из $G$ склеиванием вершин $A_{1}$ и $A_{2}$ в одну вершину $A$. Пусть функции $\widetilde{h}, \widetilde{g}$ на вершинах графа $\widetilde{G}$ совпадают с $h$ и $g$ во всех вершинах, кроме $A_{1}$ и $A_{2}$, и $\widetilde{h}(A)=h\left(A_{1}\right), \widetilde{g}(A)=g\left(A_{1}\right)+g\left(A_{2}\right)$. Тогда игра $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle$ тоже выигрышная. (И тогда, если игра $\langle\widetilde{G}, h, \widetilde{g}\rangle$ проигрышная, то и $\langle G, h, g\rangle$ проигрышная.)
Пусть $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}, g_{1}\right\rangle, \mathcal{G}_{2}=\left\langle G_{2}, h_{2}, g_{2}\right\rangle$ - две игры, такие что $V_{1} \cap V_{2}=\{v\}$. Пусть $G=G_{1}+_{v} G_{2}$ - объединение графов $G_{1}$ и $G_{2}$, в котором обе вершины $v$ объединены в одну вершину. Зададим функции $h, g: V_{1} \cup V_{2} \rightarrow \mathbb{N}$ :
$$
h(u)=\left\{\begin{array}{ll}
h_{i}(u), & u \in V_{i} \backslash\{v\},(i=1,2), \\
h_{1}(v) h_{2}(v), & u=v,
\end{array} \quad g(u)= \begin{cases}g_{i}(u), & u \in V_{i} \backslash\{v\},(i=1,2) \\
g_{1}(v) g_{2}(v), & u=v\end{cases}\right.
$$
Игру $\mathcal{G}=\langle G, h, g\rangle$ будем обозначать $\mathcal{G}_{1} \times{ }_{v} \mathcal{G}_{2}$ (рис. 1).

Рис. 1. Игра $G_{1} \times{ }_{v} G_{2}$
3.2. Теорема о произведении. Если мудрецы выигрывают в играх $\mathcal{G}_{1}$ и $\mathcal{G}_{2}$, то они выигрывают и в игре $\mathcal{G}_{1} \times{ }_{v} \mathcal{G}_{2}$.
3.3. Пусть $G=G_{1}+_{A} G_{2}$, где $G_{1}$ и $G_{2}$ - графы, у которых $V\left(G_{1}\right) \cap V\left(G_{2}\right)=\{A\}$, а игры $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}, g_{1}\right\rangle$ и $\mathcal{G}_{2}=\left\langle G_{2}, h_{2}, g_{2}\right\rangle$ - проигрышные, и при этом
$$
g_{1}(A)=g_{2}(A)=s, \quad h_{1}(A) \geqslant h_{2}(A)=s+1
$$
Тогда игра $\mathcal{G}=\left\langle G_{1}+{ }_{A} G_{2}, h, g\right\rangle$ - проигрышная, где
$$
h(x)= \begin{cases}h_{i}(x), & x \in V_{i} \backslash\{A\}, \quad(i=1,2), \\ h_{1}(A), & x=A\end{cases}
$$
3.4. Удаление полуребра. Пусть $\langle G, h, g\rangle$ - выигрышная игра, $A B$ - ребро графа $G, \widetilde{G}$ - граф, получающийся из $G$ заменой ребра $A B$ на ориентированное ребро $B \rightarrow A$ (те. мудрец $A$ не видит мудреца $B$, но $B$ видит $A$ ). Пусть функция $\widetilde{g}$ на вершинах графа $G$ совпадает с $g$ во всех вершинах, кроме $A$, и $\widetilde{g}(A)=h(B) g(A)$. Тогда игра $\langle\widetilde{G}, h, \widetilde{g}\rangle$ тоже выигрышная. (И тогда, если игра $\langle\widetilde{G}, h, \widetilde{g}\rangle$ проигрышная, то и $\langle G, h, g\rangle$ проигрышная.)
3.5. Теорема о подстановке. Пусть $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}, g_{1}\right\rangle$ и $\mathcal{G}_{2}=\left\langle G_{2}, h_{2}, g_{2}\right\rangle$ - выигрышные игры. Пусть $A$ - произвольная вершина графа $G_{2}$. Рассмотрим новый граф $G$, который получается из $G_{2}$ заменой вершины $A$ на граф $G_{1}$ (каждая вершина $G_{1}$ соединяется новыми ребрами с бывшими соседями вершины $A$, см. рис. 2). Тогда игра $\langle G, h, g\rangle$ - выигрышная, где
$$
h(u)=\left\{\begin{array}{ll}
h_{2}(u), & u \in V\left(G_{2}\right) \backslash\{A\}, \\
h_{1}(u) h_{2}(A), & u \in V\left(G_{1}\right),
\end{array} \quad g(u)= \begin{cases}g_{2}(u), & u \in V\left(G_{2}\right) \backslash\{A\} \\
g_{1}(u) g_{2}(A), & u \in V\left(G_{1}\right)\end{cases}\right.
$$
3.6. Подстановка с сокращением. Пусть $\mathcal{G}=\langle G, h, \star s\rangle, \mathcal{G}^{\prime}=\left\langle G^{\prime}, h^{\prime}, g^{\prime}\right\rangle$ - выигрышные игры. Пусть $A$ - вершина графа $G^{\prime}$, причем $h^{\prime}(A)=s$. Пусть $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle-$ выигрышная игра, получающаяся

Рис. 2. Подстановка графа на место вершины $A$.
подстановкой игры $\mathcal{G}$ на место вершины $A$ в игру $\mathcal{G}^{\prime}$ (как в задаче 3.5). По правилу подстановки для всех подставленных вершин $v$
$$
\widetilde{h}(v)=h(v) h^{\prime}(A)=s \cdot h(v), \quad \widetilde{g}(v)=g(v) g^{\prime}(A)=s \cdot g^{\prime}(A)
$$
Рассмотрим новые функции $h^{*}, g^{*}$ на графе $\widetilde{G}$, которые отличаются от $\widetilde{h}, \widetilde{g}$ только значениями в подставленных вершинах $v$, и это отличие - сокращение на $s$ :
$$
h^{*}(v)=h(v), \quad g^{*}(v)=g^{\prime}(A)
$$
Тогда игра $\left\langle\widetilde{G}, h^{*}, g^{*}\right\rangle$ - тоже выигрышная.
3.7. Раздувание вершины. Пусть $\mathcal{G}=\langle G, h, g\rangle$ - выигрышная игра, $A \in V(G), \widetilde{G}$ - граф, получающийся из $G$ подстановкой на место вершины $A$ клики $B$, состоящей из $g(A)$ вершин. Тогда игра $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle$ тоже выигрышная, где
$$
\widetilde{h}(v)=\left\{\begin{array}{ll}
h(v), & v \in V(G) \backslash\{A\} \\
h(A), & v \in B
\end{array} \quad \widetilde{g}(v)= \begin{cases}g(v), & v \in V(G) \backslash\{A\} \\
1, & v \in B\end{cases}\right.
$$
## 4 «Лепестки» и «етунии»
Лепестком будем называть граф, изображенный на рис. 3 , вершину лепестка степени $n-1$ (т. е. верхнюю вершину на рис. 3) будем называть черенком лепестка.
Будем строить из лепестков более сложные графы. Возьмем два лепестка $L_{1}$ и $L_{2}$, обозначим в каждом из них одну из вершин буквой $v_{1}$ и построим граф $M_{2}=L_{1}+_{v_{1}} L_{2}$. Возьмем граф $M_{2}$ и лепесток $L_{3}$, обозначим в каждом из них одну из вершин буквой $v_{2}$ и рассмотрим граф $M_{3}=M_{2} t_{v_{2}} L_{3}$ и т. д. Графы, которые можно получить с помощью конечного числа таких операций, будем называть петуниями.
Петунию будем называть королевской петунией (рис. 4), если на каждом шаге ее построения вершина $v_{i}$ была черенком лепестка $L_{i+1}$.
4.1. Пусть $G$ - лепесток из $n$ вершин, см. рис. 3 , а функция шляпности $h$ задается так: шляпность верхней вершины равна 2, шляпности остальных вершин равны 7. Докажите, что в игре $\langle G, h\rangle$ мудрецы проигрывают.

Рис. 3. Лепесток из $n$ вершин

Рис. 4. Королевская петуния
4.2. Пусть $G$ - лепесток из $n$ вершин. Положим $f(s)=s^{2}+s$. Докажите, что $\mathrm{HG}_{s}(G) \leqslant f(f(s))$.
4.3. Пусть $M$ - петуния, $h_{s}$ - наибольшее натуральное число, для которого игра $\left\langle M, \star h_{s}, \star s\right\rangle$ выигрышная. Докажите, что $h_{s} \leqslant f(f(f(s)))$.
4.4. а) Докажите, что $\mathrm{HG}_{s}(G)=4 s(s+1)-2$, где $G$ - лепесток из $n$ вершин, где $n$ достаточно велико (рис. 3).
b) Докажите это же равенство, если $G$ - королевская петуния с достаточно крупными лепестками.
## 5 Дополнительные задачи
## В серию 1
5.1. Пусть $K_{1, n}$ - граф «звезда», а $h=\left(h_{0}, \ldots, h_{n}\right), g=\left(g_{0}, \ldots, g_{n}\right)$ - произвольные функции количеств шляп и числа попыток, где $1 \leqslant g_{i} \leqslant h_{i}$ при всех $i$, нулевой индекс соответствует центральной вершине графа. Докажите, что существование $k$, для которого $\left\langle K_{1, n}, k \cdot h, k \cdot g\right\rangle$ является выигрышной, равносильно неравенству
$$
\frac{g_{0}}{h_{0}} \geqslant \prod_{i=1}^{n}\left(1-\frac{g_{i}}{h_{i}}\right)
$$
## В серию 2
5.2. Пусть $h$ и $g$ натуральные числа, причем $g^{2}-3 g h+h^{2}<0$. Докажите, что игра $\left\langle P_{3}, \star h, \star g\right\rangle-$ выигрышная.
Дробным шляпным числом графа $G$ назовем величину $\hat{\mu}(G)=\sup \left\{\frac{h}{g}\right.$ : игра $\langle G, \star h, \star g\rangle$ - выигрышная $\}$.
Как следует из задачи 1.9, $\mathrm{HG}\left(K_{n}\right)=\hat{\mu}\left(K_{n}\right)=n, \mathrm{HG}_{s}\left(K_{n}\right)=s n$. В общем случае, $\hat{\mu}\left(K_{n}\right) \geqslant \frac{1}{s} \mathrm{HG}_{s}(G) \geqslant \operatorname{HG}(G)$.
5.3. Докажите, что $\hat{\mu}\left(K_{3}\right)=\frac{3+\sqrt{5}}{2}$.
## В серию 3
5.4. Пусть $\left\langle G_{2}, h_{2}\right\rangle$ - проигрышная игра, $H_{2} \subset G_{2}$ - клика в $G_{2}$. Пусть $G_{1}$ полный граф. Зададим на нем функцию шляпности $h_{1}$ так, чтобы выполнялось соотношение
$$
\left(\sum_{u \in G_{1}} \frac{1}{h_{1}(u)}\right)\left(\prod_{v \in H_{2}} h_{2}(v)\right)<1
$$
Пусть $G$ - это граф, получающийся объединением графов $G_{1}$ и $G_{2}$ и добавлением всех ребер между вершинами $G_{1}$ и $H_{2}$ (рис. 5). Докажите, что игра $\langle G, h\rangle$ - проигрышная, если
$$
h(v)= \begin{cases}h_{1}(v), & v \in G_{1} \\ h_{2}(v), & v \in G_{2}\end{cases}
$$

Рис. 5. Пример к задаче 5.4. Число вершин в $G_{1}$ и $H_{2}$ не обязано совпадать
5.5. Пусть $\mathcal{G}=\langle G, h\rangle$ - проигрышная игра, $A$ - любая вершина графа $G$. Рассмотрим граф $G^{\prime}=\left\langle V^{\prime}, E^{\prime}\right\rangle$, получающийся добавлением к графу $G$ новой висячей вершины $B: V^{\prime}=V \cup\{B\}$, $E^{\prime}=E \cup\{A B\}$. Тогда мудрецы проигрывают в игре $\left\langle G^{\prime}, h^{\prime}\right\rangle$, где $h^{\prime}(B)=2, h^{\prime}(A)=2 h(A)-1$ и $h^{\prime}(u)=h(u)$ для остальных вершин $u \in V$.
5.6. Пусть $\mathcal{G}=\langle G, h, g\rangle$ - некоторая игра, $A \in V(G)$ соединена ребрами со всеми остальными вершинами графа $G, h(A)=s+1, g(A)=s$, и $\mathcal{G}^{\prime}=\left\langle G \backslash\{A\}, h^{\prime},(s+1) \cdot g^{\prime}\right\rangle$, где $h^{\prime}=\left.h\right|_{V(G) \backslash\{A\}}$, $g^{\prime}=\left.g\right|_{V(G) \backslash\{A\}}$. Тогда игры $\mathcal{G}$ и $\mathcal{G}^{\prime}$ эквивалентны (мудрецы выигрывают в $\mathcal{G}$ тогда и только тогда, когда они выигрывают в $\mathcal{G}^{\prime}$ ).
## Серия 5
5.7. а) Выигрышный граф $G$ содержит «длинный мост»- двузвенный путь $A B C$, при удалении которого граф распадается на две компоненты: $G_{1}$ (содержащую вершину $A$ ) и $G_{2}$ (содержащую вершину $C$ ). Пусть шляпность вершины $B$ равна 5. Докажите, что хотя бы одна из игр $\left\langle G_{1},\left.h\right|_{G_{1}}\right\rangle$, $\left\langle G_{2},\left.h\right|_{G_{2}}\right\rangle$ выигрышная.
b) Пусть граф $\widetilde{G}$ получен подразбиением произвольного графа $G$ (т. е. каждое ребро графа $G$ заменили на двузвенный путь). Докажите, что игра $\langle\widetilde{G}, \star 5\rangle$ - проигрышная.
5.8. Пусть $\mathcal{G}=\langle G, h\rangle$ - выигрышная игра, причем максимальная в следующем смысле: при увеличении функции шляпности в любой вершине мудрецы проигрывают и, кроме того, с функцией шляпности $h$ мудрецы не могут выиграть ни на каком подграфе графа $G$. Допустим, что граф $G$ содержит ребро-мост $A B$. Докажите, что игра $\mathcal{G}$ представима в виде произведения игр.
5.9. Мудрецы $A$ и $B$ имеют по одной попытке угадывания, видят друг друга и всех остальных мудрецов в графе (а те видят их), $h(A)=2, h(B)=3$. Докажите, что если заменить этих двух мудрецов на одного мудреца $C$, который видит остальных, а остальные его, и при этом $h(C)=6$, $g(C)=5$, то результат игры не изменится.
5.10. Даны натуральные числа $s$ и $d$. Пусть $G$ - произвольный граф, вершины которого разбиты на два множества $V(G)=A \cup B$, причем каждая вершина из $A$ имеет не более $d$ соседей из $B$. Докажите, что $\mathrm{HG}_{s}(G) \leqslant \mathrm{HG}_{s^{\prime}}(G[A])$, где $s^{\prime}=s\left(\mathrm{HG}_{s}(G[B])+1\right)^{d}$, а $G[A]$ и $G[B]$ - индуцированные подграфы на множествах $A$ и $B$.
## Решения
1.1. Назовём белый и красный цвета светлыми, а синий и зелёный - тёмными. Тогда пусть первый назовёт оба тёмных цвета, если на втором надет светлый, и наоборот; а второй - назовёт оба светлых цвета, если на первом надет светлый, и наоборот. Нетрудно понять, что в данной стратегии хотя бы один из мудрецов угадывает свой цвет.
1.2. Перенумерум цвета шляп от 1 до 5 . Тогда пусть первый игрок назовёт цвета в предположении, что сумма номеров на шляпах первого и второго даёт остаток 0 или 1 при делении на 5 , а второй - в предположении, что эта же сумма даёт остаток 2,3 или 4 при делении на 5 . Так как реальная сумма существует и даёт какой-то остаток, то мудрец с верным предположением угадает свой цвет.
1.3. Сначала напомним, как играть в игру, если мудрецов всего два, шляпы могут быть лишь двух цветов (обозначим их 0 и 1) и требуется, чтобы обязательно кто-то угадал. Стратегия здесь такая - один мудрец (назовем его уравнитель) проверяет гипотезу «цвета шляп одинаковы», другой (различитель) - «цвета шляп различны». Отметим, что по этой стратегии для любого расклада шляп один из мудрецов угадает, а другой нет.
Введем граф видимости - мудрецы это вершины, пары соседних мудрецов - ребра, по условию этот граф - цикл из пяти вершин. Обозначим ребра $a, b, c, d, e$. Будем считать, что цвет шляпы мудреца - это двузначное бинарное число: 00, 01, 10 или 11 , причем записывая это число, будем помечать его разряды метками ребер, инцидентных вершине, где сидит мудрец. Например, если из вершины выходят ребра $a$ и $b$, то мы подписываем снизу под одним из разрядов «a», а под другим «b» (под каким разрядом какая метка - неважно, главное, чтобы все мудрецы пользовались этими пометками единообразно).
В начале для каждого ребра мудрецы договариваются, кто на этом ребре уравнитель, а кто различитель. При угадывании на каждом ребре $x$ происходит следующее: каждый мудрец на этом ребре смотрит лишь на разряд $x$ в цвете своего соседа и вычисляет цвет своего разряда $x$ в соответствии со своей ролью на этом ребре. Таким образом, каждый мудрец, посмотрев налево и направо, вычисляет оба разряда для цвета своей шляпы и называет полученный цвет в качестве ответа.
Очевидно, мудрец угадает цвет своей шляпы, только если он правильно угадал оба бита. Поскольку граф содержит лишь 5 ребер, лишь 5 битов были угаданы правильно, и следовательно, не более двух мудрецов верно указали цвет своей шляпы.
1.4. Сопоставим каждому цвету центрального мудреца последовательность из 5 цифр, каждая из которых $-1,2$ или 3 . Стратегия $i$-го мудреца в яме: посмотреть на $i$-ю цифру центрального мудреца и назвать 2 остальные. Стратегия центрального: назвать цвет, $i$-я цифра которого равна цифре $i$-го мудреца из ямы. Тогда если никакой мудрец из ямы не угадал, то центральный верно назовёт свой цвет.
1.5. Для каждого мудреца $v$ рассмотрим $k \cdot h(v)$ цветов шляп и разобьём их на $h(v)$ групп по $k$ цветов, которые мы будем называть мегацветами. Тогда в игре $\langle G, k \cdot h, k \cdot g\rangle$ каждый мудрец $v$ может понять мегацвета всех своих соседей, и, согласно стратегии для игры $\langle G, h, g\rangle$, назвать $g(v)$ своих мегацветов. Но эти $g(v)$ мегацветов соответствуют $k \cdot g(v)$ обычным цветам. Нетрудно поверить, что если стратегия для игры $\langle G, h, g\rangle$ была выигрышной, то полученная стратегия также будет выигрышной.
1.6. Дадим произвольные шляпы мудрецам не из $K$. Тогда для мудрецов из $K$ станет ясно, какой ответ дает каждый из них по своей стратегии. Осталось каждому дать такую шляпу, чтобы он не угадал.
1.7. Пусть $r^{\prime}=\frac{p}{q}$. Если $G=\langle G, \star h, \star g\rangle$ - выигрышная, то игра $\langle G, \star p h, \star p g\rangle$ - тоже выигрышная. А так как $\frac{p}{q} \leqslant \frac{h}{g}$, то $p g \leqslant q h$, откуда игра $\langle G, \star p h, \star q h\rangle$ - также выигрышная, так как увеличение количества попыток работающую стратегию не портит. Тогда $h^{\prime}=p h, q^{\prime}=q h$ являются искомыми.
1.8. Пусть $G=\langle G, h, g\rangle$ - выигрышная игра, $r^{\prime}: V \rightarrow \mathbb{Q}$ - функция, удовлетворяющая условию $01$.
2.3. Для каждого натурального $s$ докажем утверждение индукцией по $n$. База $n=1$, т. е. проигрышность игры $\stackrel{\frac{s}{2 s}}{\stackrel{\leftrightarrow}{A}} \stackrel{\frac{s}{4 s-1}}{\stackrel{\circ}{s}}$ следует из утверждения задачи 1.9 .
Переход. Рассмотрим три крайние вершины $A, B, C$. Рассмотрим всевозможные назначения цвета шляпы мудрецу $B$. Мудрец $A$ называет в совокупности $s(4 s-1)$ цветов из множества $\{0,1, \ldots, 2 s-1\}$. Значит, какой-то цвет $c_{A}$ встречается в его ответах не более $\left[\frac{s(4 s-1)}{2 s}\right]=2 s-1$ раз. Выдадим мудрецу $A$ шляпу этого цвета. Мудрец $B$ видит шляпу $c_{A}$ и знает, для каких $2 s-1$ цветов его шляпы мудрец $A$ назовет цвет $c_{A}$. Поэтому мудрец $B$ может считать, что цвет его шляпы берется из множества $C_{B}$, состоящего из $4 s-1-(2 s-1)=2 s$ цветов. В этот момент чертик объявляет, что построит расклад шляп на оставшемся графе, выдав мудрецу $B$ какой-то цвет из $C_{B}$, и сообщает остальным мудрецам, что это за множество цветов. Тогда на оставшемся графе происходит игра из индукционного предположения.
2.4. Неравенство $\mathrm{HG}_{2}\left(P_{4}\right) \geqslant 6$ выполняется в силу того, что игра

выигрышная.
Неравенство $\mathrm{HG}_{3}\left(P_{6}\right) \geqslant 10$ выполняется в силу того, что игра

выигрышная (множители по краям выигрышные о утверждению задачи 2.2).
Наконец, неравенство $\operatorname{HG}_{4}\left(P_{10}\right) \geqslant 14$ выполняется в силу того, что игра $G(u) \times \times_{u} \stackrel{\frac{1}{2}}{\stackrel{\bullet}{u}} \stackrel{\frac{1}{2}}{\omega} \times{ }_{w} G(w)$


с задачей 2.3 здесь увеличена шляпность вершины $A$ ). При $s=2,3,4$ это дает, кстати, при всех $n$ неравенства $\mathrm{HG}_{2}\left(P_{n}\right) \leqslant 6, \mathrm{HG}_{3}\left(P_{n}\right) \leqslant 10, \mathrm{HG}_{4}\left(P_{n}\right) \leqslant 14$.
2.5. Рассмотрим на графе $P_{s}$ функцию шляпности $h$ :
$$
h\left(v_{i}\right)= \begin{cases}4 s-2 & \text { при } 1 \leqslant is+1$ мудрец $A$ по этой стратегии может называть больше $s+1$ цвета, т. е. больше, чем его шляпность в игре $G_{2}$. Однако в этом
случае упоминание «посторонних» цветов не способствует выигрышу. Значит, на $G_{2}$ существует опровергающий расклад $\psi$. Если $\psi(A)=\varphi_{j}(A)$, то $\left.\psi \cup \varphi_{j}\right|_{V\left(G_{1}\right) \backslash A}$ - расклад, опровергающий стратегию $f$ в игре $\mathcal{G}$.
3.4. Это очевидно: на графе $\widetilde{G}$ мудрец $A$ должен сначала назвать $g(A)$ цветов, которые он называет по стратегии для графа $G$, если у мудреца $B$ первый цвет шляпы, потом еще $g(A)$ цветов, которые он называет, если у мудреца $B$ второй цвет шляпы, и т. д.
3.5. Для каждого натурального $N$ множество $\{0,1, \ldots, N-1\}$ обозначим для краткости через $[N]$.
Для каждой вершины $u$ подставляемого графа $G_{1}$ определим её цвет в игре $\langle G, h, g\rangle$, как композитный цвет из множества $\left[h_{1}(u)\right] \times\left[h_{2}(A)\right]$. Первую компоненту $u$ будет искать по своей изначальной стратегии из игры $\mathcal{G}_{1}$, а вторую - по стратегии $A$ из игры $\mathcal{G}_{2}$. Соседи вершины $A$ из графа $G_{2}$ в новом графе $G$ видят весь подграф $G_{1}$, и потому могут определить, кто угадал свою первую компоненту цвета. Пусть $B$ - первый из таких мудрецов (в лексикографическом порядке), вершины графа $G_{2} \backslash\{A\}$ могут играть по стратегии игры $\mathcal{G}_{2}$, используя вторую компоненту цвета $B$ в качестве цвета $A$. Поскольку $\mathcal{G}_{2}$ - выигрышная игра, какая-то из вершин побеждает. Если это вершина из $G_{2} \backslash\{A\}$, то она угадывает свой цвет и в графе $G$. Если же в $\mathcal{G}_{2}$ побеждала вершина $A$, то $B$ правильно нашла и вторую компоненту своего цвета.
3.6. Для доказательства мы модифицируем стратегию конструктора подстановки из предыдущей задачи. В силу этой конструкции вершина $v$ после подстановки получает композитный цвет из множества $[h(v)] \times\left[h^{\prime}(A)\right]$, а стратегия мудреца $v$ состоит в том, что он вычисляет $G$-компоненту своего ответа, т. е. выбирает $s=g(v)$ цветов $c_{1}, \ldots, c_{s} \in[h(v)]$, вычисляет $G^{\prime}$-компоненту своего ответа, т. е. выбирает $g^{\prime}(A)$ цветов $e_{1}, \ldots, e_{g^{\prime}(A)} \in\left[h^{\prime}(A)\right]$, после чего называет все пары цветов вида $\left(c_{i}, e_{j}\right)$.
Введем изменение в конструкцию подстановки и опишем, как играют мудрецы в изменившейся ситуации. Изменение затрагивает только подставляемых мудрецов $v$ и состоит в том, что им назначается $h^{*}(v)=h(v)$ и $g^{*}(v)=g^{\prime}(A)$. Таким образом, шляпа мудреца $v$ вместо композитного цвета будет иметь цвет «всего лишь» из множества $[h(v)]$, каковое по-прежнему воспринимается его соседями из $N_{G}(v)$ и $N_{G^{\prime}}(A)$ как $G$-компонента его цвета.
Стратегия каждого подставляемого мудреца $v$ состоит из двух фаз. Первая фаза: поглядев на соседей в $G$, мудрец $v$ вычисляет « $G$-компоненту» своего ответа, т. е. некоторый набор из $s$ цветов $c_{1}, \ldots, c_{s} \in[h(v)]$. Далее мудрец $v$ отождествляет построенный набор с множеством $\left[h^{\prime}(A)\right]$ (по правилу $c_{i} \mapsto i$; напомним, что $h^{\prime}(A)=s$ ). После этого происходит вторая фаза: он смотрит на своих соседей в графе $G^{\prime}$ и применяет стратегию мудреца $A$, называя $g^{\prime}(A)$ цветов из своего новообретенного множества $\left[h^{\prime}(A)\right]$.
Осталось описать стратегию мудрецов из $N_{G^{\prime}}(A)$. Все они видят целиком граф $G$, поэтому знают, какой набор цветов каждый мудрец $v$ отождествил с множеством $\left[h^{\prime}(A)\right]$. Кроме того, все они знают, кто из $V(G)$ правильно угадал $G$-компоненту своего цвета, допустим, первым из таких мудрецов (в лексикографическом порядке) является мудрец $w$. То, что мудрец $w$ угадал $G$-компоненту своего цвета, означает, что цвет его шляпы лежит в построенном им множестве $\left[h^{\prime}(A)\right]$. Тогда на второй фазе мудрецы графа $G^{\prime} \backslash\{A\}$ просто разыгрывают выигрышную стратегию игры $\mathcal{G}^{\prime}$, подставляя $w$ с его построенным множеством $\left[h^{\prime}(A)\right]$ на место $A$, а мудрец $w$, как определено выше, тоже фактически играет по этой стратегии. В результате кто-то из них угадает.
3.7. Каждый мудрец из $V(G) \backslash\{A\}$, который видит мудрецов $B$, рассчитывает «виртуальный цвет мудреца $A$ »
$$
c_{A}=\sum_{v \in B} c_{v} \quad(\bmod h(A))
$$
и играет по стратегии из игры $\mathcal{G}$. Что касается мудрецов из $B$, они берут себе по одному ответу $a_{i}$ из стратегии мудреца $A$, и мудрец $v_{i}$ называет цвет
$$
a_{i}-\sum_{v \in B, v \neq v_{i}} c_{v} \quad(\bmod h(A))
$$
(таким образом $i$-й мудрец проверяет гипотезу $c_{A}=a_{i}$ ).
4.1. Пусть $B$ - вершина шляпности 2. Применим для вершины $B$ утверждение конструктора «удаление половины ребра» (задача 3.4), сделав эту вершину невидимой для остальных вершин. Тогда остальные мудрецы не видят $B$, имеют по две попытки, и проигрывают по утверждению задачи 2.6 б). Выдав им опровергающий расклад шляп, судья сделает так, что и мудрец $B$ не угадает.
4.2. Зафиксируем число $s$. Рассмотрим случай, когда верхняя вершина лепестка имеет шляпность $s+1$, а остальные вершины - шляпность $h$. Действуя, как в предыдущей задаче, мы делаем верхнюю вершину невидимой для остальных, от этого число попыток угадывания у остальных вершин, расположенных на пути $P_{n-1}$, становится равно $s(s+1)=f(s)$. Отсюда по утверждению задачи 2.6б) мы получаем, что при $h>f(f(s))$ мудрецы проигрывают. Следовательно, $t_{s} \leqslant$ $f(f(s))$.
4.3. Будем доказывать индукцией по числу лепестков, что при $h>f(f(f(s)))$ игра проигрышная. База индукции - один лепесток $G$ - с запасом доказана в п. а): игра $\langle G, \star h, \star s\rangle$ проигрышная уже при $h>f(f(s))$.
Но нам понадобится одно близкое утверждение - «модифицированная база индукции»: игра $\langle G, \star \bar{h}, \star s\rangle$ проигрышная при $h>f(f(f(s)))$, где через $\star \bar{h}$ обозначена функция шляпности, равная $h$ во всех вершинах лепестка $G$, кроме одной вершины $B$, шляпность которой равна $s+1$.
Докажем это утверждение. Если $B$ - верхняя вершина лепестка $G$, это опять утверждение базы. Пусть $B$ - произвольная вершина лепестка в нижнем ряду (рис. 7 , слева). Пользуясь конструктором «удаление половины ребра» задачи 3.4 , сделаем вершину $B$ невидимой для остальных вершин, в результате число попыток у ее соседей станет равно $f(s)$. Теперь вершину $B$ можно вообще удалить из $G$ - ее никто не видит, у нее шляпность $s+1$ и всего $s$ попыток угадывания, ей не суждено угадать. Как нетрудно видеть, оставшийся граф представляет собой объединение двух лепестков с общей верхней вершиной (или один лепесток, но этот случай тривиален) и с увеличенным до $f(s)$ количеством попыток угадывания у некоторых вершин (рис. 7 , в центре). Добавим к графу горизонтальное ребро между бывшими соседями $B$ (рис. 7 , справа), оно могло бы помочь мудрецам выиграть. В результате получился лепесток, вершины которого имеют не более $f(s)$ попыток угадывания, а их шляпности больше $f(f(f(s))$ ). По утверждению базы мудрецы все-таки проигрывают.

Рис. 7. Удаляем вершину $B$ из лепестка $G$
Теперь докажем индукционный переход. Рассмотрим петунию $M_{n}=M_{n-1} t_{v_{n}} L_{n}$. В силу модифицированного утверждения базы игра $\left\langle L_{n}, \star \bar{h}, \star s\right\rangle$ проигрышная при $h>f(f(f(s)))$, где через $\star \bar{h}$ обозначена функция шляпности, равная $h$ во всех вершинах лепестка $G$, кроме $v_{n}$, а шляпность $v_{n}$ равна $s+1$. Игра $\left\langle M_{n-1}, \star h, \star s\right\rangle$ проигрышная по предположению индукции. Осталось заметить, что игра $\left\langle M_{n}, \star h, \star s\right\rangle$ может быть получена с помощью конструктора задачи 3.3 из проигрышных игр $\left\langle M_{n-1}, \star h, \star s\right\rangle$ и $\left\langle L_{n}, \star \bar{h}, \star s\right\rangle$ и поэтому тоже является проигрышной.
4.4. а) Оценка $\mathrm{HG}_{s}(G)<4 s(s+1)-1$. Пусть $A$ - черенок лепестка, причем $h(A)=s+1$, а остальные вершины $v$ имеют шляпность $h(v)=4 s(s+1)-1$. Достаточно проверить проигрышность игры $\langle G, h, \star s\rangle$. По утверждению задачи 5.6, эта игра эквивалентна игре на пути $P_{n}$, где у всех вершин шляпность равна $4 s(s+1)-1$, а число попыток $-s(s+1)$. Но как следует из задачи 2.3 , эта игра проигрышная.
Покажем, что игра $\langle G, \star 4 s(s+1)-2, \star s\rangle$ - выигрышная при больших $n$.
Пусть $\mathcal{G}_{0}=\left\langle P_{k}, \star 4 s(s+1)-2, \star s(s+1)\right\rangle$. Будем считать, что число $k$ достаточно велико, тогда по результату задачи 2.5 , игра $\mathcal{G}_{0}$ выигрышная.
Аналогично задаче 2.6 с) можно доказать, что для любого натурального $h$ существует такое натуральное $n$, что игра на графе $K_{1, n}$ - выигрышная, если шляпности всех периферийных мудрецов равны $s+1$ и все они имеют по одной попытке угадывания, а шляпность центрального мудреца равна $h$ и у него $s$ попыток. Положим $h=4 s(s+1)-2$ и подберем подходящее $n$. Подставим с сокращением на место каждого периферийного мудреца игру $\mathcal{G}_{0}$. Мы получим выигрышную игру, где у всех вершин шляпности равны $4 s(s+1)-2$, количества попыток равны $s$, а граф представляет собой подграф большого лепестка.
b) Как мы проверили в п. а), игра $\langle G, h, \star s\rangle$ - проигрышная, где $G$ - лепесток с большим числом вершин, а $h$ - функция, задающая черенку шляпность $s+1$, а остальным вершинам шляпность $4 s(s+1)-1$. По утверждению задачи 3.3 приклеивание такого лепестка по черенку к другой проигрышной игре (где число попыток в вершине, куда мы его приклеиваем, равно $s$ ) снова дает проигрышную игру. Но королевская петуния по определению задается последовательными черенковыми приклеиваниями лепестков! Таким образом, игра на королевской петунии с большими лепестками, где все вершины имеют шляпность $4 s(s+1)$ - 1 и $s$ попыток угадывания, кроме (первого, корневого) черенка, имеющего шляпность $s+1$ и $s$ попыток угадывания, проигрышная. Это дает оценку $\mathrm{HG}_{s}(G)<4 s(s+1)-1$.
Тогда $\mathrm{HG}_{s}(G)=4 s(s+1)-2$, поскольку такая шляпность реализуется уже на лепестках, которые являются королевскими петуниями, хотя и не слишком развесистыми.
5.1. Необходимость. Рассмотрим один из $h_{0}$ цветов шляпы центрального мудреца. Тогда стратегии остальных вершин уже определены, и в $\prod_{i=1}^{n}\left(h_{i}-g_{i}\right)$ случаев ни одна из висячих вершин не угадывает свой цвет. Следовательно, это должен делать центральный. Но центральный мудрец может это сделать суммарно только в $g_{0} \prod_{i=1}^{n} h_{i}$ случаях. Отсюда и следует неравенство, равносильное неравенству из условия.
Достаточность. Покажем, что игра $\left\langle K_{1, n},\left(N \cdot h_{0} ; h_{1}, \ldots, h_{n}\right),\left(N \cdot g_{0} ; g_{1}, \ldots, g_{n}\right)\right\rangle$ при $N=\prod_{i=1}^{n} h_{i}$ - выигрышная.
Закодируем $h_{0} \cdot h_{1} \cdot \ldots \cdot h_{n}$ цветов центрального мудреца наборами $\left(c_{0} ; c_{1}, \ldots, c_{n}\right)$, где $0 \leqslant c_{i}3 g-h$. Тогда найдётся такое $k$, что оба числа $c_{C}-\left[k \cdot \frac{h}{g}\right]$ и $c_{C}-\left[(k+3 g-h) \cdot \frac{h}{g}\right]$ лежат в $I_{A} \cap I_{C}$ (элементы пересечения $I_{A} \cap I_{C}$ могут быть записаны двумя способами: $c_{A}-i=c_{C}-\left[\ell \cdot \frac{h}{g}\right]$, в качестве $k$ можно число $\ell$, соответствующее минимально возможному $i$ ). Поскольку оба элемента лежат в множестве $I_{A}$, состоящем из подряд идущих остатков, расстояние между ними не превосходит $g-1$ :
$$
\left(c_{C}-\left[k \cdot \frac{h}{g}\right]\right)-\left(c_{C}-\left[(k+3 g-h) \cdot \frac{h}{g}\right]\right) \leqslant g-1
$$
что равносильно
$$
\left[(k+3 g-h) \cdot \frac{h}{g}\right]-\left[k \cdot \frac{h}{g}\right] \leqslant g-1
$$
Избавляясь от округления, получаем следствие:
$$
(k+3 g-h) \cdot \frac{h}{g}-0.5-k \cdot \frac{h}{g}-0.5 \leqslant g-1
$$
Последнее равносильно неравенству $(3 g-h) \cdot \frac{h}{g} \leqslant g$, т. е. $0 \leqslant g^{2}-3 g h+h^{2}$, что противоречит условию.
5.3. По утверждению задачи 5.1 существование $k$, при котором игра $\left\langle P_{3}, \star k h, \star k g\right\rangle=\left\langle K_{1,2}, \star k h, \star k g\right\rangle$ выигрышная, эквивалентна условию
$$
\left(1-\frac{g}{h}\right)\left(1-\frac{g}{h}\right) \leqslant \frac{g}{h}
$$
а это для неотрицательных $h$ равносильно неравенству $g^{2}-3 g h+h^{2} \leqslant 0$, что при $1 \leqslant g \leqslant h$ эквивалентно неравенству
$$
\frac{h}{g} \leqslant \frac{3+\sqrt{5}}{2}
$$
Теперь утверждение задачи очевидно.
5.4. Рассмотрим произвольную стратегию мудрецов в игре $\langle G, h\rangle$. Произведение $\prod_{v \in H_{2}} h_{2}(v)$ перечисляет расклады шляп на $H_{2}$, задание каждого из них фиксирует стратегию мудрецов на $G_{1}$. При фиксированном раскладе шляп на $H_{2}$ будем перебирать расклады шляп на $G_{1}$. Тогда сумма $\sum_{u \in G_{1}} \frac{1}{h_{1}(u)}$ подсчитывает наибольшую возможную долю тех раскладов на $G_{1}$, где хотя бы один мудрец из $G_{1}$ угадывает. Тогда произведение $\left(\sum_{u \in G_{1}} \frac{1}{h_{1}(u)}\right) \prod_{v \in H_{2}} h_{2}(v)$ подсчитывает наибольшую возможную долю тех раскладов на $G_{1}$, где хотя бы один мудрец из $G_{1}$ угадывает, если мы каждому мудрецу даем $\prod_{v \in H_{2}} h_{2}(v)$ попыток для угадывания. Таким образом, выполнение неравенства из условия задачи означает, что существует расклад шляп $\alpha$ на $G_{1}$, для которого ни один мудрец из $G_{1}$ не угадывает, какие бы шляпы при этом ни выдавались бы мудрецам из $H_{2}$. Если выдать мудрецам из $G_{1}$ расклад $\alpha$, ни один из них не угадает и при этом будет полностью задана стратегия мудрецов на графе $G_{2}$, пригодная для игры $\left\langle G_{2}, h_{2}\right\rangle$. Поскольку эта игра проигрышная, расклад шляп $\alpha$ можно дополнить раскладом шляп на $G_{2}$, для которого никто из $G_{2}$ также не угадает.
5.5. Пусть мудрецы зафиксировали стратегию на графе $G^{\prime}$. Построим проигрышный расклад шляп для этой стратегии. Стратегия мудреца $B$ для каждого из $2 h(A)-1$ возможных цветов шляпы $A$ предписывает назвать один из двух цветов. Какой-то из этих двух цветов называется не более чем $h(A)-1$ раз. Дадим мудрецу $B$ шляпу этого цвета, это фиксирует стратегию мудреца $A$ на оставшемся графе $G$. Теперь, чтобы $B$ не угадал, мы попытаемся дать $A$ шляпу одного из оставшихся $h(A)$ цветов (если их осталось больше, оставим ровно $h(A)$ ). С этим ограничением нам удастся построить расклад шляп на графе $G$ так, что никто на $G$ не угадает, поскольку игра $\mathcal{G}$ проигрышная.
5.6. If $\mathcal{G}$ is a winning game, we remove by problem 3.4 all half-edges $\overrightarrow{v A}$ and obtain the winning game. In this game $A$ has $s$ guesses, $s+1$ colors and no infotmation, therefore we can assign $A$ 's color such that $A$ does not guess. But now the remaining sages play the game $\mathcal{G}^{\prime}$. Hence $\mathcal{G}^{\prime}$ is winning.
If the game $\mathcal{G}^{\prime}$ is winning, substitute $\mathcal{G}^{\prime}$ in the winning game $\stackrel{\frac{s}{s+1}}{\stackrel{\Delta}{s}} \quad \frac{1}{s+1}$ in place of vertex $B$ by problem 3.6. We obtain a winning game $\mathcal{G}$.
5.7. а) Пусть $\mathcal{G}$ - проигрышная игра на пути $A B C$ со шляпностями вершин $h(A)=h(C)=2$, $h(B)=5$. Если бы обе игры из условия оказались проигрышными, то исходная игра получалась бы из этих игр и игры $\mathcal{G}$ с помощью конструктора задачи 3.3 (где $s=1, g_{1}=g_{2}=\star 1$ ) и тоже оказалась бы проигрышной.
b) Множество $V(\widetilde{G})$ состоит из вершин графа $G$ и множества новых вершин $V_{1}$ (расположенных в середине двузвенных путей). Зададим функцию на $V(\widetilde{G})$ :
$$
h(v)= \begin{cases}2, & v \in V(G) \\ 5, & v \in V_{1}\end{cases}
$$
Достаточно проверить, что игра $\langle\widetilde{G}, h\rangle$ проигрышная. Это очевидно. Действительно, каждый мудрец из $V_{1}$ имеет двух соседей в $\widetilde{G}$ со шляпностью 2 , поэтому он называет по своей стратегии не более четырех цветов - сразу же дадим ему шляпу того цвета, который он не называет, и он не угадает. Теперь все ответы мудрецов из $V(G)$ детерминированы, и мы тоже дадим каждому мудрецу шляпу того цвета, который он не называет.
5.8. Пусть при удалении моста граф $G$ распадается на компоненты $G_{1}$ (содержащую вершину $B$ ) и $G_{2}$ (содержащую вершину $A$ ). Введем функции шляпности $h_{1}$ и $h_{2}$ на этих графах по правилу
$$
h_{1}(x)=\left\{\begin{array}{ll}
h(x), & x \in V\left(G_{1}\right) \backslash\{B\}, \\
\left\lceil\frac{h(B)}{2}\right\rceil, & x=B
\end{array} \quad h_{2}(x)= \begin{cases}h(x), & x \in V\left(G_{2}\right) \backslash\{A\} \\
\left\lceil\frac{h(A)}{2}\right\rceil, & x=A\end{cases}\right.
$$
Обозначим через $\mathcal{G}_{B A}$ игру на графе $K_{2}$ с вершинами $B$ и $A$, где шляпности обеих вершин равны 2 . Эта игра выигрышная. Наконец, пусть $\mathcal{G}_{1}^{\prime}=\left\langle G_{1},\left.h\right|_{G_{1}}\right\rangle$ - проигрышная игра в силу свойств функции $h$.
Если игры $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}\right\rangle$ и $\mathcal{G}_{2}=\left\langle G_{2}, h_{2}\right\rangle$ обе выигрышные, то игра $\mathcal{G}_{1} \times{ }_{B} \mathcal{G}_{B A} \times_{A} \mathcal{G}_{2}$ тоже выигрышная. Если при этом значения $h(A)$ и $h(B)$ четные, мы получаем требуемое разложение игры $\mathcal{G}$ в произведение игр. Если же хотя бы одно из чисел $h(A), h(B)$ нечетное, то функция шляпности полученной игры мажорирует $h$ и по свойствам функции $h$ такая игра не может быть выигрышной.
Осталось разобрать случай, когда хотя бы одна игра $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}\right\rangle$ или $\mathcal{G}_{2}=\left\langle G_{2}, h_{2}\right\rangle$ проигрышная, пусть это будет игра $\mathcal{G}_{2}$. Рассмотрим игру $\mathcal{G}_{2}^{\prime}$, полученную применением конструктора задачи 5.5 к игре $\mathcal{G}_{2}$ : в ней к проигрышному графу $G_{2}$ добавлена новая вершина $B$, соединенная с $A$, при этом шляпность вершины $A$ стала равна
$$
2\left\lceil\frac{h(A)}{2}\right\rceil-1 \leqslant h(A)
$$
шляпность вершины $B$ равна 2 , а шляпности остальных вершин задаются функцией $h$. Применим теперь конструктор задачи 3.3 к играм $\mathcal{G}_{1}^{\prime}$ и $\mathcal{G}_{2}^{\prime}$ (считаем, что $s=1, g_{1}=g_{2}=\star 1$ ). Мы получим проигрышную игру на графе $G$, в которой функция шляпности не превосходит $h$. Это невозможно.
5.9. Обозначим исходную игру через $\mathcal{G}_{1}=\left(G_{1}, h_{1}, g_{1}\right)$, где $g_{1} \equiv 1$. Подграф графа $G_{1}$, получающийся из $G_{1}$ удалением вершины $A$, обозначим через $G_{2}$, подграф графа $G_{2}$, получающийся из $G_{2}$ удалением вершины $B$, обозначим через $G_{3}$, а через $G_{4}$ обозначим граф, полученный из $G_{3}$
добавлением вершины $C$, которая всех видит и которую все видят, т. е. фактически $G_{4}$ получается из $G_{2}$ переименованием вершины $B$ в $C$. Рассмотрим игры $\mathcal{G}_{2}=\left(G_{2}, h_{2}, g_{2}\right), \mathcal{G}_{3}=\left(G_{3}, h_{3}, g_{3}\right)$, $\mathcal{G}_{4}=\left(G_{4}, h_{4}, g_{4}\right)$, где
$$
\begin{aligned}
& h_{2}(v)=\left\{\begin{array}{ll}
h_{1}(v) & v \in G_{3}, \\
3 & v=B
\end{array}, \quad h_{3}(v)=\left.h_{1}\right|_{G_{3}}, \quad h_{4}(v)= \begin{cases}h_{1}(v) & v \in G_{3} \\
6 & v=C\end{cases} \right. \\
& g_{2} \equiv 2 \\
& g_{3} \equiv 6 \\
& g_{4}(v)= \begin{cases}1 & v \in G_{3} \\
5 & v=C\end{cases}
\end{aligned}
$$
Допустим, что игра $\mathcal{G}_{1}$ выигрышная. Удалим по задаче 3.4 все полуребра, ведущие в вершины $A$ и $B$, мы получим выигрышную игру, в которой у остальных участников (мудрецов из $G_{3}$ ) число попыток стало равно 6 , а стратегии не зависят от цветов шляп $A$ и $B$. Если бы можно было выдать на $G_{3}$ расклад шляп, на котором никто из $G_{3}$ не угадывает, это фиксировало бы стратегии мудрецов $A$ и $B$, играющих на ребре $A B$ в игру $\stackrel{\frac{1}{2}}{\stackrel{\bullet}{A} \quad \stackrel{1}{3}}$, в в результате все мудрецы проиграли бы, что невозможно. Следовательно, сужение игры на граф $G_{3}$, т. е. игра $\mathcal{G}_{3}$, является выигрышной. Сделаем по задаче 3.6 подстановку этой игры с сокращением в игру $\underset{C}{\frac{5}{6}} \stackrel{\frac{1}{6}}{\stackrel{\circ}{*}}$ на место вершины $v$, получаем выигрышную игру $\mathcal{G}_{4}$.
Пусть теперь наоборот игра $\mathcal{G}_{4}$ выигрышная. Мы должны заменить игрока $C$ на двух игроков $A$ (шляпности 2) и $B$ (шляпности 3 ). Предъявим выигрышную стратегию игроков в полученной игре. Все игроки из $G_{3}$ будут пользоваться стратегией игры $\mathcal{G}_{4}$, интерпретируя пару (цвет $A$, цвет $B$ ) как композитный цвет игрока $C$. Покажем, как можно «конвертировать» стратегию игрока $C$ в пару стратегий $A$ и $B$.
Цвет игрока $C$ - это элемент множества $\mathcal{C}=\{0,1\} \times\{0,1,2\}$. Пусть при состоявшемся раскладе шляп соседей $C$ должен назвать все цвета из множества $\mathcal{C}$, кроме $(1,2)$. Тогда действия игроков $A$ и $B$ состоят в том, что $A$ называет цвет, совпадающий с четность цвета $B$, а $B$ называет цвет 0 или 1 , противоположной четности к цвету $A$. Игрок $C$ угадывал по своей стратегии, если у него был один из цветов $(0,0),(0,2),(1,1)$ или $(0,1),(1,0)$. По нашему правилу в первых трех случаях угадает $A$, в остальных двух $B$. Аналогично распределяются роли $A$ и $B$ для других пятерок ответов $C$.
5.10. Покажем, что игра $\left\langle G, \star \mathrm{HG}_{s^{\prime}}(G[A])+1, \star s\right\rangle$, где $s^{\prime}=s\left(\mathrm{HG}_{s}(G[B])+1\right)^{d}$, - проигрышная. Для этого построим расклад шляп, для которого мудрецы проиграют. Поскольку $\mathrm{HG}_{s^{\prime}}(G[A]) \geqslant s^{\prime} \geqslant$ $\mathrm{HG}_{s}(G[B])+1$, т. е. шляпность в рассматриваемой игре больше, чем $\mathrm{HG}_{s}(G[B])+1$, достаточно рассмотреть случай, когда шляпность мудрецов из $B$ равна $\mathrm{HG}_{s}(G[B])+1$. Используя задачу 3.4 , удалим всё полурёбра из $A$ в $B$, сделав множество $B$ невидимым для множества $A$. Так как в результате этого действия мы у каждого мудреца из $A$ стираем не более $d$ полуребер, число попыток у мудреца вырастет, но не будет превосходить $s^{\prime}$. При этом стратегии у мудрецов из $A$ теперь не зависят от расклада шляп на $B$. Значит, мы можем считать, что на графе $G[A]$ происходит игра $\left\langle G[A], \star \mathrm{HG}_{s^{\prime}}(G[A])+1, \star s^{\prime}\right\rangle$. По определению $s$-шляпного числа эта игра проигрышная. Значит, существует расклад шляп на $A$, такой что никто из $A$ не угадывает свой цвет. Выдадим мудрецам из $A$ этот раклад, тогда у мудрецов из $B$ будет детерминирована стратегия игры на $G[B]$. Поскольку шляпности мудрецов из $B$ больше $\mathrm{HG}_{s}(G[B])$, мы можем каждому выдать цвет, который он не назвал. Мудрецы проиграли.
# Paint my hat in 3.5 colors!
K. Kokhas, A. Latyshev
Project team: O. Bursian, D. Kokhas, K. Kokhas, V. Retinskiy
## 1 Let us introduce: the HATS game!
Let an undirected graph $G$ be given, one sage and one chest with hats of different colors are located in each of its vertices. All the sages are acquainted with each other. Graph $G$, the location of the sages in the vertices of the graph and the colors of hats in all chests are fixed and known to everybody. In particular, each sage understands, in which vertex each of the others sages is located. The referee performs the following test with the sages. He puts a hat on the head of each sage, the hat is taken from the sage's chest. Each sage sees only the hats of the sages located in the neighbouring vertices of the graph, he does not see his own hat and does not know its color. The sages cannot communicate during the test. At the command of the referee each of the sages writes names of several colors on his paper simultaneously (how many colors the sages has to mention, is determined by the additional rule). We say that the sages have passed the test successfully $=$ «have won», if at least one of them wrote the color of his hat in his paper.
The sages have been informed of the rules of the test before the testing and they have the possibility to hold a meeting, in which they must to define their public strategy. The publicity means that all the participants, including the referee, know this strategy. The strategy of the sages has to be deterministic, that is each sage has to write colors on his paper looking only the colors that he sees on his neighbours. We call the strategy winning if for any hats placement at least one sage will guess correctly the color of the hat on his head, i. e. mention this color in the his list of guesses. We say that the sages win, if they have a winning strategy, and that they lose, if they have not.
Therefore, the HATS game is not a game in a sense as it is ordinarily understood. This game lasts only one move.
1.1. The referee puts a hat of white, blue, red or green color on the head of each of two sages. Each of them sees the hat of the other, but does not see his own hat. Each of them writes on his own paper two colors simultaneously. They try to guess correctly the colors of their own hats. Prove that the sages can come to an agreement in the meeting before the test in such a way that at least one of them will guess correctly.
1.2. The referee puts a hat of five possible colors on the head of each of two sages. Each of them sees the hat of the other, but does not see his own hat. Each of them try to guess correctly the color of his own hat. The first sage writes on his own paper two colors and the second - three colors simultaneously. Prove that the sages can come to an agreement in the meeting before test in such a way that at least one of them will guess correctly.
1.3. Five sages stand around the non-transparent baobab. Shah has put red, blue, yellow or green hat the head of each of the sages. Sage does not know the color of his own hat and sees only the two neighbouring sages. As usual, without any communication each sages must makes one assumptions about the colors of his hat. But they fear be too lucky. How they should act to guarantee that for any placement of hats no more than two sages guess correctly the colors of their hats?
1.4. Sultan examines six court sages. By the rule of the examination the sultan locates 5 sages in 5 pits positioned around a circle, and locates the sixth sage in the tower in the center of the circle. The sultan writes one of the numbers 1,2 or 3 on the forehead of each of the first five sages and writes a number from 1 to 243 on the forehead of the central sage. The sage in the tower sees the numbers of all the other sages, and these sages see his number (but do not see each other). All the sages must simultaneously try to guess correctly their numbers: the sages in the pits must say two numbers and the sage in the tower - one number. The sultan has explained to the sages the rules of the examination beforehand and has given time to communicate before the beginning of the examination. Can the sages act so that at least one of them certainly guess correctly his number?
We identify a vertex of graph $G$ and the sage located in it. We assume that the colors are numbered by $0,1,2,3, \ldots$ and that the chest of sage $v$ contains hats of colors from 0 to some number $h(v)-1$.
The HaTs game is the triple $\langle G, h, g\rangle$, where $G=\langle V, E\rangle-$ a graph, $h: V \rightarrow \mathbb{N}$ - a function that for each vertex $v$ equals the number of colors of hats keeping in the chest in this vertex, $g: V \rightarrow \mathbb{N}-$ a function equal to the number of guesses of each sage. We call function $h$ a «hatness», and $g$ - a function of guesses or the number of attempts. For each non negative integer $h$ we denote by $\star h$ the function on $V$ possessing the constant value $h$. Instead of the notation $\langle G, h, \star 1\rangle$ we will use shorter notation $\langle G, h\rangle$.
1.5. Prove that if the game $\langle G, h, g\rangle$ is winning, then for each non negative integer $k$ the game $\langle G, k \cdot h, k \cdot g\rangle$ is winning, too.
1.6. Game $\langle G, h, g\rangle$ is given. Let $K \subset G$ is anticlique (a set of vertices such that there is no edge connected any pair of them) and for each $v \in K h(v)>g(v)$. Prove that there exists a hats placement, for which none of the sages in $K$ guesses correctly.
1.7. Let $h$ and $g$ be natural numbers, $G=\langle G, \star h, \star g\rangle$ be a winning game, $r^{\prime} \leqslant \frac{h}{g}$ be a rational number. Prove that there exist natural numbers $h^{\prime}$ and $g^{\prime}$ such that $\frac{h^{\prime}}{g^{\prime}}=r^{\prime}$ and game $\left\langle G, \star h^{\prime}, \star g^{\prime}\right\rangle$ is winning. 1.8. Formulate and prove the analogue of the previous statement for non-constant functions of hatness and guessing.
1.9. Denote by $K_{n}$ a complete graph on $n$ vertices. Prove that the game $\left\langle K_{n}, h, g\right\rangle$ is winning if and only if
$$
\sum_{v \in K_{n}} \frac{g(v)}{h(v)} \geqslant 1
$$
## 2 Paths and trees
The theory of HATS game on the complete graph $K_{3}$ is given by the problem statement 1.9. Now consider a path $P_{3}$ which is less complicated graph.
2.1. Prove that the sages lose in the game $\left\langle P_{3}, \star 3, \star 1\right\rangle$.
2.2. a) Prove that the game $\frac{\frac{3}{10}}{0^{10}} \frac{3}{5}$ is winning (the numerator is the number of guesses, and the denominator is the hatness).
b) Prove that the game $0^{\frac{3}{11}} \frac{\frac{3}{10}}{0^{\circ}} \frac{3}{5}$ is losing.
c) Prove that the game $\stackrel{\frac{s}{t(s)}}{\frac{s}{t(s)}} \cdot \frac{s}{t(s)}$ is losing, where $t(s)=s^{2}+s+1$.
Let $G$ be a graph and $s$ be a non negative integer. Denote by $\mathrm{HG}_{s}(G)$ the $s$-hat number of $G$, i.e. the maximum number of hats $h$ for which the game $\langle G, \star h, \star s\rangle$ is winning. For $s=1$ this number is called hat number of $G$ and is denoted by $\mathrm{HG}(G)$.
2.3. Prove that for any non negative integers $n$ and $s$ the game $\stackrel{\frac{s}{2 s}}{\stackrel{\Delta}{A}} \longrightarrow \cdots$ on path $P_{n}$ is losing. Here all vertices except the leftmost vertex $A$ have hatness $4 s-1$ and $s$ guesses.
2.4. Prove that one can find $n$ such that $\mathrm{HG}_{2}\left(P_{n}\right)=6, \mathrm{HG}_{3}\left(P_{n}\right)=10, \mathrm{HG}_{4}\left(P_{n}\right)=14$.
2.5. Prove that for any non negative integer $s$ the game $\left\langle P_{n}, \star(4 s-2), \star s\right\rangle$ is winning for $n \geqslant 2 s$.
2.6. a) Let $t(s)=s^{2}+s+1$. Prove that for each tree $T$ the game $\langle T, \star t(s), \star s\rangle$ is losing.
b) Len $K_{1, n}$ be "a star" graph (i.e. a tree consisting of a root and $n$ leaves). Prove that for sufficiently large $n$ the game $\left\langle K_{1, n}, \star\left(s^{2}+s\right), \star s\right\rangle$ is winning.
c) Prove that for any non negative integer $h$ there exists integer $n$ such that the game on graph $K_{1, n}$ is winning if all the sages in pendant vertices have one guess and hatness 3 , and the central sage has 2 guesses and hatness $h$.
## 3 Constructors
3.1. Let $\langle G, h, g\rangle$ be a winning game, $A_{1}$ and $A_{2}$ be vertices of $G$, not connected with an edge and such that $h\left(A_{1}\right)=h\left(A_{2}\right)$. We glue vertices $A_{1}$ and $A_{2}$ of $G$ into one new vertex $A$, denote by $\widetilde{G}$ the obtained graph. Let functions $\widetilde{h}$ and $\widetilde{g}$ defined on the set of vertices of $\widetilde{G}$ coincide with $h$ and $g$ in all vertices except $A_{1}$ and $A_{2}$, and $\widetilde{h}(A)=h\left(A_{1}\right), \widetilde{g}(A)=g\left(A_{1}\right)+g\left(A_{2}\right)$. Then the game $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle$ is also winning. (And then if the game $\langle\widetilde{G}, h, \widetilde{g}\rangle$ is losing, then the game $\langle G, h, g\rangle$ is losing too.)
Let $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}, g_{1}\right\rangle, \mathcal{G}_{2}=\left\langle G_{2}, h_{2}, g_{2}\right\rangle$ be two games such that $V_{1} \cap V_{2}=\{v\}$. Let $G=G_{1}+{ }_{v} G_{2}$ be the union of graphs $G_{1}$ and $G_{2}$, in which both vertices $v$ are glued into one new vertex. Define functions $h, g: V_{1} \cup V_{2} \rightarrow \mathbb{N}$ :
$$
h(u)=\left\{\begin{array}{ll}
h_{i}(u), & u \in V_{i} \backslash\{v\},(i=1,2), \\
h_{1}(v) h_{2}(v), & u=v,
\end{array} \quad g(u)= \begin{cases}g_{i}(u), & u \in V_{i} \backslash\{v\},(i=1,2) \\
g_{1}(v) g_{2}(v), & u=v\end{cases}\right.
$$
We say that the game $\mathcal{G}=\langle G, h, g\rangle$ is a product of games $\mathcal{G}_{1}$ and $\mathcal{G}_{2}$ and denote it by $\mathcal{G}_{1} \times{ }_{v} \mathcal{G}_{2}$ (fig. 1).

Рис. 1. The product $G_{1} \times{ }_{v} G_{2}$
3.2. The theorem about the product. If the sages win in games $\mathcal{G}_{1}$ and $\mathcal{G}_{2}$, then they also win in game $\mathcal{G}_{1} \times{ }_{v} \mathcal{G}_{2}$.
3.3. Let $G=G_{1}+{ }_{A} G_{2}$, where $G_{1}$ and $G_{2}$ are graphs, for which $V\left(G_{1}\right) \cap V\left(G_{2}\right)=\{A\}$. Let games $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}, g_{1}\right\rangle$ and $\mathcal{G}_{2}=\left\langle G_{2}, h_{2}, g_{2}\right\rangle$ be losing, and the following conditions hold:
$$
g_{1}(A)=g_{2}(A)=s, \quad h_{1}(A) \geqslant h_{2}(A)=s+1
$$
Then game $\mathcal{G}=\left\langle G_{1}+{ }_{A} G_{2}, h\right\rangle$ is losing, where
$$
h(x)=\left\{\begin{array}{ll}
h_{i}(x), & x \in V_{i} \backslash\{A\}(i=1,2), \\
h_{1}(A), & x=A
\end{array} \quad g(x)= \begin{cases}g_{i}(x), & x \in V_{i} \backslash\{A\}(i=1,2) \\
s, & x=A\end{cases}\right.
$$
3.4. A half-edge removal. Let $\langle G, h, g\rangle$ be a winning game, $A B$ be an edge of graph $G, \widetilde{G}$ be the graph obtained from $G$ by replacing edge $A B$ by directed edge $B \rightarrow A$ (ie. sage $A$ does not see sage $B$, but $B$ sees $A$ ). Let function $\widetilde{g}$ on the vertices of graph $G$ coincide with $g$ in all vertices except $A$, and $\widetilde{g}(A)=h(B) g(A)$. Then game $\langle\widetilde{G}, h, \widetilde{g}\rangle$ is winning too. (And therefore, if game $\langle\widetilde{G}, h, \widetilde{g}\rangle$ is losing, then $\langle G, h, g\rangle$ is also losing.)
3.5. Substitution theorem. Let $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}, g_{1}\right\rangle$ be $\mathcal{G}_{2}=\left\langle G_{2}, h_{2}, g_{2}\right\rangle$ be winning games. Let $A$ be an arbitrary vertex of graph $G_{2}$. Consider the new graph $G$ obtained from $G_{2}$ by substitution of graph $G_{1}$ on the place of vertex $v$ (each vertex $G_{1}$ is connected with former neighbours of vertex $A$ by new edges, see fig. 2). Then game $\langle G, h, g\rangle$ is winning, where
$$
h(u)=\left\{\begin{array}{ll}
h_{2}(u), & u \in V\left(G_{2}\right) \backslash\{A\}, \\
h_{1}(u) h_{2}(A), & u \in V\left(G_{1}\right),
\end{array} \quad g(u)= \begin{cases}g_{2}(u), & u \in V\left(G_{2}\right) \backslash\{A\} \\
g_{1}(u) g_{2}(A), & u \in V\left(G_{1}\right)\end{cases}\right.
$$
3.6. Substitution with reducing. Let $\mathcal{G}=\langle G, h, \star s\rangle, \mathcal{G}^{\prime}=\left\langle G^{\prime}, h^{\prime}, g^{\prime}\right\rangle$ be winning games. Let $A$ be a vertex of graph $G^{\prime}$, and $h^{\prime}(A)=s$. Let $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle$ be the winning game obtained by the substitution of

Рис. 2. Substitution of a graph on the place of vertex $A$.
game $\mathcal{G}$ on the place of vertex $A$ to game $\mathcal{G}^{\prime}$ (as in problem 3.5). By the rule of the substitution for all substituting vertices $v$
$$
\widetilde{h}(v)=h(v) h^{\prime}(A)=s \cdot h(v), \quad \widetilde{g}(v)=g(v) g^{\prime}(A)=s \cdot g^{\prime}(A)
$$
Consider new functions $h^{*}, g^{*}$ on graph $\widetilde{G}$, which differ from $\widetilde{h}, \widetilde{g}$ only by the values in substituting vertices $v$, and this difference is the cancellation by $s$ :
$$
h^{*}(v)=h(v), \quad g^{*}(v)=g^{\prime}(A)
$$
Then game $\left\langle\widetilde{G}, h^{*}, g^{*}\right\rangle$ is also winning.
3.7. Blowing up of a vertex. Let $\mathcal{G}=\langle G, h, g\rangle$ be winning game, $A \in V(G), \widetilde{G}$ be the graph obtained from $G$ by the substitution of clique $B$ consisting of $g(A)$ vertices on the place of vertex $A$. Then game $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle$ is also winning, where
$$
\widetilde{h}(v)=\left\{\begin{array}{ll}
h(v), & v \in V(G) \backslash\{A\}, \\
h(A), & v \in B
\end{array} \quad \widetilde{g}(v)= \begin{cases}g(v), & v \in V(G) \backslash\{A\} \\
1, & v \in B\end{cases}\right.
$$
## 4 "Petals" and "petunias"
We define a petal graph to be a graph $G$ obtained from a path by adding a vertex $v$ adjacent to every vertex of this path, see fig. 3, we say that $v$ is the stem of $G$.
Then, we define a petunia to be a graph constructed in the following way. Take two petals $L_{1}$ and $L_{2}$, denote one vertex in each of them by $v_{1}$, and construct a graph $M_{2}=L_{1}+_{v_{1}} L_{2}$. After that consider graph $M_{2}$ and a new petal $L_{3}$ denote one vertex in each of them by $v_{2}$, and construct a graph $M_{3}=M_{2}+_{v_{2}} L_{3}$ and so on.
A royal petunia is a petunia (puc. 4), for which the vertex $v_{i}$ in each step of its construction were chosen as the stem of petal $L_{i+1}$.
4.1. Let $G$ be a petal of $n$ vertices, see fig. 3, let the stem has hatness 2 and the other vertices have hatness 7 . Prove that the sages lose in the game $\langle G, h\rangle$.

Рис. 3. A petal of $n$ vertices

Рис. 4. A royal petunia
4.2. Let $G$ be a petal of $n$ vertices, $f(s)=s^{2}+s$. Prove that $\mathrm{HG}_{s}(G) \leqslant f(f(s))$.
4.3. Let $M$ be a petunia, $h_{s}$ be maximum integer such that the game $\left\langle M, \star h_{s}, \star s\right\rangle$ is winning. Prove that $h_{s} \leqslant f(f(f(s)))$.
4.4. a) Prove that $\mathrm{HG}_{s}(G)=4 s(s+1)-2$, where $G$ is a petal of $n \geqslant 2 s+1$ vertices (fig. 3$)$.
b) Prove the same equality if $G$ is a royal petunia.
# Paint my hat in 3.5 colors!
K. Kokhas, A. Latyshev
Project team: O. Bursian, D. Kokhas, K. Kokhas, V. Retinskiy
## 1 Let us introduce: the HATS game!
Let an undirected graph $G$ be given, one sage and one chest with hats of different colors are located in each of its vertices. All the sages are acquainted with each other. Graph $G$, the location of the sages in the vertices of the graph and the colors of hats in all chests are fixed and known to everybody. In particular, each sage understands, in which vertex each of the others sages is located. The referee performs the following test with the sages. He puts a hat on the head of each sage, the hat is taken from the sage's chest. Each sage sees only the hats of the sages located in the neighbouring vertices of the graph, he does not see his own hat and does not know its color. The sages cannot communicate during the test. At the command of the referee each of the sages writes names of several colors on his paper simultaneously (how many colors the sages has to mention, is determined by the additional rule). We say that the sages have passed the test successfully = "have won", if at least one of them wrote the color of his hat in his paper.
The sages have been informed of the rules of the test before the testing and they have the possibility to hold a meeting, in which they must to define their public strategy. The publicity means that all the participants, including the referee, know this strategy. The strategy of the sages has to be deterministic, that is each sage has to write colors on his paper looking only the colors that he sees on his neighbours. We call the strategy winning if for any hats placement at least one sage will guess correctly the color of the hat on his head, i. e. mention this color in the his list of guesses. We say that the sages win, if they have a winning strategy, and that they lose, if they have not.
Therefore, the HATS game is not a game in a sense as it is ordinarily understood. This game lasts only one move.
1.1. The referee puts a hat of white, blue, red or green color on the head of each of two sages. Each of them sees the hat of the other, but does not see his own hat. Each of them writes on his own paper two colors simultaneously. They try to guess correctly the colors of their own hats. Prove that the sages can come to an agreement in the meeting before the test in such a way that at least one of them will guess correctly.
1.2. The referee puts a hat of five possible colors on the head of each of two sages. Each of them sees the hat of the other, but does not see his own hat. Each of them try to guess correctly the color of his own hat. The first sage writes on his own paper two colors and the second - three colors simultaneously. Prove that the sages can come to an agreement in the meeting before test in such a way that at least one of them will guess correctly.
1.3. Five sages stand around the non-transparent baobab. Shah has put red, blue, yellow or green hat the head of each of the sages. Sage does not know the color of his own hat and sees only the two neighbouring sages. As usual, without any communication each sages must makes one assumptions about the colors of his hat. But they fear be too lucky. How they should act to guarantee that for any placement of hats no more than two sages guess correctly the colors of their hats?
1.4. Sultan examines six court sages. By the rule of the examination the sultan locates 5 sages in 5 pits positioned around a circle, and locates the sixth sage in the tower in the center of the circle. The sultan writes one of the numbers 1, 2 or 3 on the forehead of each of the first five sages and writes a number from 1 to 243 on the forehead of the central sage. The sage in the tower sees the numbers of all the other sages, and these sages see his number (but do not see each other). All the sages must simultaneously try to guess correctly their numbers: the sages in the pits must say two numbers and the sage in the tower - one number. The sultan has explained to the sages the rules of the examination beforehand and has given time to communicate before the beginning of the examination. Can the sages act so that at least one of them certainly guess correctly his number?
We identify a vertex of graph $G$ and the sage located in it. We assume that the colors are numbered by $0,1,2$, $3, \ldots$ and that the chest of sage $v$ contains hats of colors from 0 to some number $h(v)-1$.
The HATS game is the triple $\langle G, h, g\rangle$, where $G=\langle V, E\rangle-$ a graph, $h: V \rightarrow \mathbb{N}$ - a function that for each vertex $v$ equals the number of colors of hats keeping in the chest in this vertex, $g: V \rightarrow \mathbb{N}$ - a function equal to the number of guesses of each sage. We call function $h$ a "hatness", and $g-$ a function of guesses or the number of attempts. For each non negative integer $h$ we denote by $\star h$ the function on $V$ possessing the constant value $h$. Instead of the notation $\langle G, h, \star 1\rangle$ we will use shorter notation $\langle G, h\rangle$.
1.5. Prove that if the game $\langle G, h, g\rangle$ is winning, then for each non negative integer $k$ the game $\langle G, k \cdot h, k \cdot g\rangle$ is winning, too.
1.6. Game $\langle G, h, g\rangle$ is given. Let $K \subset G$ is anticlique (a set of vertices such that there is no edge connected any pair of them) and for each $v \in K h(v)>g(v)$. Prove that there exists a hats placement, for which none of the sages in $K$ guesses correctly.
1.7. Let $h$ and $g$ be natural numbers, $G=\langle G, \star h, \star g\rangle$ be a winning game, $r^{\prime} \leqslant \frac{h}{g}$ be a rational number. Prove that there exist natural numbers $h^{\prime}$ and $g^{\prime}$ such that $\frac{h^{\prime}}{g^{\prime}}=r^{\prime}$ and game $\left\langle G, \star h^{\prime}, \star g^{\prime}\right\rangle$ is winning. 1.8. Formulate and prove the analogue of the previous statement for non-constant functions of hatness and guessing.
1.9. Denote by $K_{n}$ a complete graph on $n$ vertices. Prove that the game $\left\langle K_{n}, h, g\right\rangle$ is winning if and only if
$$
\sum_{v \in K_{n}} \frac{g(v)}{h(v)} \geqslant 1
$$
## 2 Paths and trees
The theory of HATS game on the complete graph $K_{3}$ is given by the problem statement 1.9. Now consider a path $P_{3}$ which is less complicated graph.
2.1. Prove that the sages lose in the game $\left\langle P_{3}, \star 3, \star 1\right\rangle$.
2.2. a) Prove that the game $\frac{\frac{3}{10}}{\frac{3}{10}} \frac{3}{5}$ is winning (the numerator is the number of guesses, and the denominator is the hatness).
b) Prove that the game $0^{\frac{3}{11}} \frac{\frac{3}{10}}{0^{\circ}} \frac{3}{5}$ is losing.
c) Prove that the game $\stackrel{\frac{s}{t(s)}}{\frac{s}{t(s)}} \cdot \frac{s}{t(s)}$ is losing, where $t(s)=s^{2}+s+1$.
Let $G$ be a graph and $s$ be a non negative integer. Denote by $\mathrm{HG}_{s}(G)$ the $s$-hat number of $G$, i.e. the maximum number of hats $h$ for which the game $\langle G, \star h, \star s\rangle$ is winning. For $s=1$ this number is called hat number of $G$ and is denoted by $\operatorname{HG}(G)$.
2.3. Prove that for any non negative integers $n$ and $s$ the game $\stackrel{\frac{s}{2 s}}{\stackrel{A}{A}} \cdot \cdots$ on path $P_{n}$ is losing. Here all vertices except the leftmost vertex $A$ have hatness $4 s-1$ and $s$ guesses.
2.4. Prove that one can find $n$ such that $\mathrm{HG}_{2}\left(P_{n}\right)=6, \mathrm{HG}_{3}\left(P_{n}\right)=10, \mathrm{HG}_{4}\left(P_{n}\right)=14$.
2.5. Prove that for any non negative integer $s$ the game $\left\langle P_{n}, \star(4 s-2), \star s\right\rangle$ is winning for $n \geqslant 2 s$.
2.6. a) Let $t(s)=s^{2}+s+1$. Prove that for each tree $T$ the game $\langle T, \star t(s), \star s\rangle$ is losing.
b) Len $K_{1, n}$ be "a star" graph (i.e. a tree consisting of a root and $n$ leaves). Prove that for sufficiently large $n$ the game $\left\langle K_{1, n}, \star\left(s^{2}+s\right), \star s\right\rangle$ is winning.
c) Prove that for any non negative integer $h$ there exists integer $n$ such that the game on graph $K_{1, n}$ is winning if all the sages in pendant vertices have one guess and hatness 3 , and the central sage has 2 guesses and hatness $h$.
## 3 Constructors
3.1. Let $\langle G, h, g\rangle$ be a winning game, $A_{1}$ and $A_{2}$ be vertices of $G$, not connected with an edge and such that $h\left(A_{1}\right)=h\left(A_{2}\right)$. We glue vertices $A_{1}$ and $A_{2}$ of $G$ into one new vertex $A$, denote by $\widetilde{G}$ the obtained graph. Let functions $\widetilde{h}$ and $\widetilde{g}$ defined on the set of vertices of $\widetilde{G}$ coincide with $h$ and $g$ in all vertices except $A_{1}$ and $A_{2}$, and $\widetilde{h}(A)=h\left(A_{1}\right), \widetilde{g}(A)=g\left(A_{1}\right)+g\left(A_{2}\right)$. Then the game $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle$ is also winning. (And then if the game $\langle\widetilde{G}, h, \widetilde{g}\rangle$ is losing, then the game $\langle G, h, g\rangle$ is losing too.)
Let $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}, g_{1}\right\rangle, \mathcal{G}_{2}=\left\langle G_{2}, h_{2}, g_{2}\right\rangle$ be two games such that $V_{1} \cap V_{2}=\{v\}$. Let $G=G_{1}+_{v} G_{2}$ be the union of graphs $G_{1}$ and $G_{2}$, in which both vertices $v$ are glued into one new vertex. Define functions $h, g: V_{1} \cup V_{2} \rightarrow \mathbb{N}$ :

We say that the game $\mathcal{G}=\langle G, h, g\rangle$ is a product of games $\mathcal{G}_{1}$ and $\mathcal{G}_{2}$ and denote it by $\mathcal{G}_{1} \times_{v} \mathcal{G}_{2}$ (fig. 1).

Figure 1. The product $G_{1} \times{ }_{v} G_{2}$
3.2. The theorem about the product. If the sages win in games $\mathcal{G}_{1}$ and $\mathcal{G}_{2}$, then they also win in game $\mathcal{G}_{1} \times{ }_{v} \mathcal{G}_{2}$.
3.3. Let $G=G_{1}+{ }_{A} G_{2}$, where $G_{1}$ and $G_{2}$ are graphs, for which $V\left(G_{1}\right) \cap V\left(G_{2}\right)=\{A\}$. Let games $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}, g_{1}\right\rangle$ and $\mathcal{G}_{2}=\left\langle G_{2}, h_{2}, g_{2}\right\rangle$ be losing, and the following conditions hold:
$$
g_{1}(A)=g_{2}(A)=s, \quad h_{1}(A) \geqslant h_{2}(A)=s+1
$$
Then game $\mathcal{G}=\left\langle G_{1}+{ }_{A} G_{2}, h, g\right\rangle$ is losing, where
$$
h(x)=\left\{\begin{array}{ll}
h_{i}(x), & x \in V_{i} \backslash\{A\}(i=1,2), \\
h_{1}(A), & x=A
\end{array} \quad g(x)= \begin{cases}g_{i}(x), & x \in V_{i} \backslash\{A\}(i=1,2) \\
s, & x=A\end{cases}\right.
$$
3.4. A half-edge removal. Let $\langle G, h, g\rangle$ be a winning game, $A B$ be an edge of graph $G, \widetilde{G}$ be the graph obtained from $G$ by replacing edge $A B$ by directed edge $B \rightarrow A$ (ie. sage $A$ does not see sage $B$, but $B$ sees $A$ ). Let function $\widetilde{g}$ on the vertices of graph $G$ coincide with $g$ in all vertices except $A$, and $\widetilde{g}(A)=h(B) g(A)$. Then game $\langle\widetilde{G}, h, \widetilde{g}\rangle$ is winning too. (And therefore, if game $\langle\widetilde{G}, h, \widetilde{g}\rangle$ is losing, then $\langle G, h, g\rangle$ is also losing.)
3.5. Substitution theorem. Let $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}, g_{1}\right\rangle$ be $\mathcal{G}_{2}=\left\langle G_{2}, h_{2}, g_{2}\right\rangle$ be winning games. Let $A$ be an arbitrary vertex of graph $G_{2}$. Consider the new graph $G$ obtained from $G_{2}$ by substitution of graph $G_{1}$ on the place of vertex $v$ (each vertex $G_{1}$ is connected with former neighbours of vertex $A$ by new edges, see fig. 2). Then game $\langle G, h, g\rangle$ is winning, where
$$
h(u)=\left\{\begin{array}{ll}
h_{2}(u), & u \in V\left(G_{2}\right) \backslash\{A\}, \\
h_{1}(u) h_{2}(A), & u \in V\left(G_{1}\right),
\end{array} \quad g(u)= \begin{cases}g_{2}(u), & u \in V\left(G_{2}\right) \backslash\{A\} \\
g_{1}(u) g_{2}(A), & u \in V\left(G_{1}\right)\end{cases}\right.
$$
3.6. Substitution with reducing. Let $\mathcal{G}=\langle G, h, \star s\rangle, \mathcal{G}^{\prime}=\left\langle G^{\prime}, h^{\prime}, g^{\prime}\right\rangle$ be winning games. Let $A$ be a vertex of graph $G^{\prime}$, and $h^{\prime}(A)=s$. Let $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle$ be the winning game obtained by the substitution of

Figure 2. Substitution of a graph on the place of vertex $A$.
game $\mathcal{G}$ on the place of vertex $A$ to game $\mathcal{G}^{\prime}$ (as in problem 3.5). By the rule of the substitution for all substituting vertices $v$
$$
\widetilde{h}(v)=h(v) h^{\prime}(A)=s \cdot h(v), \quad \widetilde{g}(v)=g(v) g^{\prime}(A)=s \cdot g^{\prime}(A)
$$
Consider new functions $h^{*}, g^{*}$ on graph $\widetilde{G}$, which differ from $\widetilde{h}, \widetilde{g}$ only by the values in substituting vertices $v$, and this difference is the cancellation by $s$ :
$$
h^{*}(v)=h(v), \quad g^{*}(v)=g^{\prime}(A)
$$
Then game $\left\langle\widetilde{G}, h^{*}, g^{*}\right\rangle$ is also winning.
3.7. Blowing up of a vertex. Let $\mathcal{G}=\langle G, h, g\rangle$ be winning game, $A \in V(G), \widetilde{G}$ be the graph obtained from $G$ by the substitution of clique $B$ consisting of $g(A)$ vertices on the place of vertex $A$. Then game $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle$ is also winning, where

## 4 "Petals" and "petunias"
We define a petal graph to be a graph $G$ obtained from a path by adding a vertex $v$ adjacent to every vertex of this path, see fig. 3, we say that $v$ is the stem of $G$.
Then, we define a petunia to be a graph constructed in the following way. Take two petals $L_{1}$ and $L_{2}$, denote one vertex in each of them by $v_{1}$, and construct a graph $M_{2}=L_{1}+_{v_{1}} L_{2}$. After that consider graph $M_{2}$ and a new petal $L_{3}$ denote one vertex in each of them by $v_{2}$, and construct a graph $M_{3}=M_{2}+_{v_{2}} L_{3}$ and so on.
A royal petunia is a petunia (puc. 4), for which the vertex $v_{i}$ in each step of its construction were chosen as the stem of petal $L_{i+1}$.
4.1. Let $G$ be a petal of $n$ vertices, see fig. 3 , let the stem has hatness 2 and the other vertices have hatness 7. Prove that the sages lose in the game $\langle G, h\rangle$.

Figure 3. A petal of $n$ vertices

Figure 4. A royal petunia
4.2. Let $G$ be a petal of $n$ vertices, $f(s)=s^{2}+s$. Prove that $\mathrm{HG}_{s}(G) \leqslant f(f(s))$.
4.3. Let $M$ be a petunia, $h_{s}$ be maximum integer such that the game $\left\langle M, \star h_{s}, \star s\right\rangle$ is winning. Prove that $h_{s} \leqslant f(f(f(s)))$.
4.4. a) Prove that $\mathrm{HG}_{s}(G)=4 s(s+1)-2$, where $G$ is a petal of $n$ vertices for sufficiently large $n$ (fig. 3).
b) Prove the same equality if $G$ is a royal petunia that hat sufficiently big petals.
## 5 Problems after intermediate finish
## To section 1
5.1. Let $K_{1, n}$ be a "star" graph, and $h=\left(h_{0}, \ldots, h_{n}\right), g=\left(g_{0}, \ldots, g_{n}\right)$ be arbitrary functions of hatness and number of attempts, where $1 \leqslant g_{i} \leqslant h_{i}$ for all $i$, the zero index corresponds to the central vertex of the graph. Prove that the existence of $k$, for which $\left\langle K_{1, n}, k \cdot h, k \cdot g\right\rangle$ is winning, is equivalent to the inequality
$$
\frac{g_{0}}{h_{0}} \geqslant \prod_{i=1}^{n}\left(1-\frac{g_{i}}{h_{i}}\right)
$$
## To section 2
5.2. Let $h$ and $g$ be positive integers, and $g^{2}-3 g h+h^{2}<0$. Prove that game $\left\langle P_{3}, \star h, \star g\right\rangle$ is winning.
Fractional hat guessing number of graph $G$ we call the value $\hat{\mu}(G)=\sup \left\{\frac{h}{g}:\langle G, \star h, \star g\rangle\right.$ is winning $\}$. As it follows from problem 1.9, $\mathrm{HG}\left(K_{n}\right)=\hat{\mu}\left(K_{n}\right)=n, \mathrm{HG}_{s}\left(K_{n}\right)=s n$. In the general case, $\hat{\mu}\left(K_{n}\right) \geqslant \frac{1}{s} \mathrm{HG}_{s}(G) \geqslant \mathrm{HG}(G)$.
5.3. Prove that $\hat{\mu}\left(K_{3}\right)=\frac{3+\sqrt{5}}{2}$.
## To section 3
5.4. Let $\left\langle G_{2}, h_{2}\right\rangle$ be a losing game, $H_{2} \subset G_{2}$ be a clique in $G_{2}$. Let $G_{1}$ be a complete graph. Define a hatness function $h_{1}$ on it in such a way that the relation
$$
\left(\sum_{u \in G_{1}} \frac{1}{h_{1}(u)}\right)\left(\prod_{v \in H_{2}} h_{2}(v)\right)<1
$$
holds. Let $G$ be the graph obtained by union of graphs $G_{1}$ and $G_{2}$ with adding all edges between vertices $G_{1}$ and $H_{2}$ (fig. 5). Prove that game $\langle G, h\rangle$ is losing if
$$
h(v)= \begin{cases}h_{1}(v), & v \in G_{1} \\ h_{2}(v), & v \in G_{2}\end{cases}
$$

Figure 5. Example to problem 5.4. The number of vertices in $G_{1}$ and $H_{2}$ should not necessarily coincide
5.5. Let $\mathcal{G}=\langle G, h\rangle$ be a losing game, $A$ be an arbitrary vertex of graph $G$. Consider graph $G^{\prime}=$ $\left(V^{\prime}, E^{\prime}\right)$ obtained by adding to graph $G$ new pendant vertex $B: V^{\prime}=V \cup\{B\}, E^{\prime}=E \cup\{A B\}$. Then the sages lose in game $\left\langle G^{\prime}, h^{\prime}\right\rangle$, where $h^{\prime}(B)=2, h^{\prime}(A)=2 h(A)-1$ and $h^{\prime}(u)=h(u)$ for other vertices $u \in V$.
5.6. Let $\mathcal{G}=\langle G, h, g\rangle$ be a game, a vertex $A \in V(G)$ be connected with all other vertices of $G$, $h(A)=s+1, g(A)=s$, and $\mathcal{G}^{\prime}=\left\langle G \backslash\{A\}, h^{\prime},(s+1) \cdot g^{\prime}\right\rangle$, where $h^{\prime}=\left.h\right|_{V(G) \backslash\{A\}} g^{\prime}=\left.g\right|_{V(G) \backslash\{A\}}$. Then the games $\mathcal{G}$ and $\mathcal{G}^{\prime}$ are equivalent.
## To section 5
5.7. a) A winning graph $G$ contains a "long bridge" - two-link path $A B C$, such that after removing this path the graph falls into two components of connectivity: $G_{1}$ (containing vertex $A$ ) and $G_{2}$ (containing vertex $C$ ). Let hatness of vertex $B$ equals 5. Prove that at least one of the games $\left\langle G_{1},\left.h\right|_{G_{1}}\right\rangle,\left\langle G_{2},\left.h\right|_{G_{2}}\right\rangle$ is winning.
b) Let graph $\widetilde{G}$ is obtained by subpartition of an arbitrary graph $G$ (i. e. each edge of graph $G$ has been replaced by two-link path). Prove that game $\langle\widetilde{G}, \star 5\rangle$ is losing.
5.8. Let $\mathcal{G}=\langle G, h\rangle$ be a winning game, and it is maximal in the following sense: when hatness function increase in any vertex, the sages lose and, besides that, there does not exist a subgraph of $G$, on which the sages can win with hatness function $h$. Suppose that graph $G$ contains edge-bridge $A B$. Prove that game $\mathcal{G}$ can be represented as the product of games.
5.9. Sages $A$ and $B$ have 1 guess, see each other and all other sages in the graph (and the others see them), $h(A)=2, h(B)=3$. Prove that if to replace these two sages by one sage $C$ that sees the others, and the others see him, and $h(C)=6, g(C)=5$, then the result of the game does not change.
5.10. Given natural numbers $s$ and $d$. Let $G$ be an arbitrary graph, the vertices of which are partitioned into two sets $V(G)=A \cup B$, and each vertex of $A$ has no more than $d$ neighbours from $B$. Prove that $\mathrm{HG}_{s}(G) \leqslant \mathrm{HG}_{s^{\prime}}(G[A])$, where $s^{\prime}=s\left(\mathrm{HG}_{s}(G[B])+1\right)^{d}, G[A]$ and $G[B]$ are induced subgraphs on sets $A$ and $B$.
## Solutions
1.1. Let us call white and red colors light, and blue and green colors dark. Then let the first sage write both dark colors, if he sees on the second a light color, and vice versa; and the second sage write both light colors, if he sees on the first a light color, and vice versa. It is not difficult to understand that at least one of the sages guesses correctly.
1.2. Renumber the hat colors from 1 to 5 . Then let the first player write colors under the assumption that the sum of the numbers on the hats of the first and the second sages gives remainder 0 or $1 \mathrm{mod}$ 5, and the second - under the assumption that the same sum gives remainder 2, 3 or 4 mod 5 . Since the sum really gives some remainder, the sage with the correct assumption guesses correctly his color.
1.3. First remind how to play the game, when there are only two sages: the hats can be colored in only two colors (denote the colors by 0 and 1 ) and it is required that somebody necessarily guessed correctly. Here is a strategy: one sage (call him equalizer) checks hypothesis "the hat colors are identical", the other (distinguisher) - "the hat colors are different". Note that by this strategy for any hats placement one of the sages guesses correctly and the other does not.
Consider a visibility graph: the sages are vertices, the pairs of neighbouring sages are edges, then this graph is a cycle on five vertices. Denote its edges by $a, b, c, d, e$. We assume that the color of sage's hat is a two-digit binary number: $00,01,10$ or 11 ; writing this number we will mark its bits by the labels of the edges incident to the vertex. For example, if there are two outgoing edges $a$ and $b$ from the vertex, then we subscribe one of the bits " $a$ ", and the other - "b" (the order of labels is not important, all the neighbours of the sage see this labelling).
Let the sages at the endpoints of each edge come to agreement, who on this edge is equalizer, and who is distinguisher. The guessing on each edge $x$ happens as follows: each sage on this edge looks only at bit $x$ of his neighbour's color and calculates the color of his own bit $x$ in accordance with his role on this edge. Therefore, each sage casting a glance to the left and to the right, calculates both bits and names the obtained color as his answer.
It is evident that a sage guesses correctly the color of his hat only if he has guessed correctly both bits. Since the graph contains only 5 edges, only 5 bits have been guessed correctly, and hence, at most two sages have guessed correctly the colors of their hats.
1.4. Put into correspondence to each color of the central sage the sequence of 5 digits, each of which is 1,2 or 3 . The strategy of $i$-th sage in pit: look at $i$-th digit of the central sage and name the other two digits. The strategy of the central sage: name the color, $i$-th digit of which is equal to the digit of $i$-th sage from pit. If no sage from pit has guessed correctly, then the central guesses his color correctly.
1.5. For each sage $v$ consider $k \cdot h(v)$ hat colors and split them into $h(v)$ groups containing $k$ colors each, we call them megacolors. Then in game $\langle G, k \cdot h, k \cdot g\rangle$ each sage $v$ can understand megacolors of all his neighbours, and, according to the strategy for game $\langle G, h, g\rangle$, name $g(v)$ of his own megacolors. But these $g(v)$ megacolors correspond to $k \cdot g(v)$ usual colors. It is not difficult to see that if the strategy for game $\langle G, h, g\rangle$ is winning, then the obtained strategy is winning too.
1.6. Give arbitrary hats to the sages that are not from $K$. Then all answers of the sages from $K$ according the strategy are determined. It remains to give a hat to each of them such that he will not guess correctly.
1.7. Let $r^{\prime}=\frac{p}{q}$. If $G=\langle G, \star h, \star g\rangle$ is winning, then game $\langle G, \star p h, \star p g\rangle$ is also winning. And since $\frac{p}{q} \leqslant \frac{h}{g}$, then $p g \leqslant q h$, whence game $\langle G, \star p h, \star q h\rangle$ is also winning, because the increasing of the number of attempts cannot destroy the working strategy. Then $h^{\prime}=p h, q^{\prime}=q h$ are the desired.
1.8. Let $G=\langle G, h, g\rangle$ be a winning game, $r^{\prime}: V \rightarrow \mathbb{Q}$ be a function such that $01$.
2.3. For each natural we $s$ prove the statement by induction on $n$. Base case $n=1$, i. e. the loss in

Induction step. Consider three leftmost vertices $A, B, C$. Consider all possible hat color assignments to sage $B$. Sage $A$ names $s(4 s-1)$ colors from set $\{0,1,2, \ldots, 2 s-1\}$ in total. Therefore some color $c_{A}$ occurs in his answers at most $\left[\frac{s(4 s-1)}{2 s}\right]=2 s-1$ times. Give to sage $A$ the hat of this color. Then sage $B$ sees color $c_{A}$ and knows, for which $2 s-1$ colors of his hat sage $A$ names color $c_{A}$. So sage $B$ may assume that the color of his own hat is taken from set $C_{B}$ consisting of $4 s-1-(2 s-1)=2 s$ colors. At that moment the devilkin (the other name of the referee) declares that in current hats placement the color of $B$ 's hat belongs to $C_{B}$ and inform the other sages what is the set $C_{B}$. Then the game from induction step takes place on the remained graph and it is losing.

winning by theorem of product.

(the leftmost and rightmost multipliers are winning by problem 2.2).


In view of the statement of problem 2.3 any game in the form $\stackrel{\frac{s}{4 s-1}}{A} \cdot \frac{s}{4 s-1} \cdot \frac{s}{4 s-1}$ is losing (as compared to problem 2.3 here the hatness of vertex $A$ has been increased). For $s=2,3,4$ this gives, by the way, for all $n$ the inequalities $\mathrm{HG}_{2}\left(P_{n}\right) \leqslant 6, \mathrm{HG}_{3}\left(P_{n}\right) \leqslant 10, \mathrm{HG}_{4}\left(P_{n}\right) \leqslant 14$.
2.5. Consider a hatness function $h$ on graph $P_{s}$ :
$$
h\left(v_{i}\right)= \begin{cases}4 s-2 & \text { for } 1 \leqslant is+1$ sage $A$ by this strategy can name more than $s+1$ colors, i. e. more than his hatness in game $G_{2}$. But in this case mention of "outsider" colors does help to win. Therefore there exists disproving placement $\psi$ on $G_{2}$. If $\psi(A)=\varphi_{j}(A)$, then $\left.\psi \cup \varphi_{j}\right|_{V\left(G_{1}\right) \backslash A}$ is a disproving hats placement for strategy $f$ in game $\mathcal{G}$.
3.4. It is evident. At first, sage $A$ on graph $\widetilde{G}$ has to name the $g(A)$ colors, which he names by the strategy for graph $G$, when $B$ 's hat is painted in the first color. After that sage $A$ names the $g(A)$ colors, which he names when $B$ 's hat is painted in the second color and so on.
3.5. For each natural $N$ denote the set $\{0,1, \ldots, N-1\}$ by $[N]$ for short.
For each vertex $u$ of substituted graph $G_{1}$ define its color in game $\langle G, h, g\rangle$ as a pair from set $\left[h_{1}(u)\right] \times\left[h_{2}(A)\right]$. Let sage $u$ look for the first component of his color by the strategy of game $\mathcal{G}_{1}$, and the second component by the strategy of $A$ in game $\mathcal{G}_{2}$. The neighbours of vertex $A$ from graph $G_{2}$ in new graph $G$ see the whole subgraph $G_{1}$, and therefore can determine, who has guessed correctly the first component. Let $B$ be the first of these sages (in lexicographical order). Then the vertices of graph $G_{2} \backslash\{A\}$ can play by the strategy of game $\mathcal{G}_{2}$, using the second component of color $B$ as a color of $A$. Since $\mathcal{G}_{2}$ is a winning game, some of the vertices win. If it is vertex from $G_{2} \backslash\{A\}$, then it guesses correctly its color in graph $G$ too. And if the winner of $\mathcal{G}_{2}$ was vertex $A$, then $B$ correctly found both components of its color.
3.6. For proof we modify the strategy from the previous solution. In view of this construction vertex $v$ after substitution gets a composite color from set $[h(v)] \times\left[h^{\prime}(A)\right]$, and the strategy of sage $v$ consists in calculating both components of his color, i. e. he chooses $s=g(v)$ colors $c_{1}, \ldots, c_{s} \in[h(v)]$, calculates $G^{\prime}$-component of his color, i. e. chooses $g^{\prime}(A)$ colors $e_{1}, \ldots, e_{g^{\prime}(A)} \in\left[h^{\prime}(A)\right]$, and after that he names all the pairs of colors $\left(c_{i}, e_{j}\right)$.
We will change the construction of substitution and describe how the sages play in changed situation. The change affects only the sages $v \in G$, we assign for these sages new hatness and number of guesses: $h^{*}(v)=h(v)$ and $g^{*}(v)=g^{\prime}(A)$. Therefore now $v$ 's hat has a color from set $[h(v)]$ (instead of a composite color), that is interpreted by his neighbours from $N_{G}(v)$ and $N_{G^{\prime}}(A)$ as $G$-component of his color as before.
The strategy of each sage $v \in G$ consists of two phases. The first phase: casting glances at the neighbours in $G$, sage $v$ calculates " $G$-component" of his color, i. e. a set consisting of $s$ colors $c_{1}, \ldots$, $c_{s} \in[h(v)]$. Further, sage $v$ identifies the obtained set and $\left[h^{\prime}(A)\right]$ (by the rule $c_{i} \mapsto i$; remind that
$\left.h^{\prime}(A)=s\right)$. After that the second phase begins: he looks at his neighbours in graph $G^{\prime}$ and apply the strategy of sage $A$ naming $g^{\prime}(A)$ colors from his newfound set $\left[h^{\prime}(A)\right]$.
It remains to describe strategy of the sages from $N_{G^{\prime}}(A)$. They all see the whole graph $G$, so they know, what set of colors each sage $v$ has identified with set $\left[h^{\prime}(A)\right]$. Besides that, they all know, who from $V(G)$ has guessed correctly $G$-component of his color. Let $w$ be the first of these sages in lexicographical order. Since sage $w$ has guessed correctly $G$-component of his color, the color of his hat belongs to the set $\left[h^{\prime}(A)\right]$, that he has constructed in the first phase. Then during the second phase the sages of graph $G^{\prime} \backslash\{A\}$ simply play the winning strategy of game $\mathcal{G}^{\prime}$, substituting $w$ with its constructed set $\left[h^{\prime}(A)\right]$ in the place of $A$, and sage $w$ actually plays by this strategy too, as explained above. As a result, somebody of them will guess correctly.
3.7. Each neighbour of $A$ in $V(G) \backslash\{A\}$ now sees the whole set $B$, computes a "virtual color of sage $A$ "
$$
c_{A}=\sum_{v \in B} c_{v} \quad(\bmod h(A))
$$
and plays by the strategy from game $\mathcal{G}$. As for the sages from $B$, they take for themselves one answer $a_{i}$ each from the strategy of sage $A$, and sage $v_{i}$ names color
$$
a_{i}-\sum_{v \in B, v \neq v_{i}} c_{v} \quad(\bmod h(A))
$$
(therefore, $i$-th sage verifies hypothesis $c_{A}=a_{i}$ ).
4.1. Let $B$ be the vertex of hatness 2. Apply for vertex $B$ the statement of constructor "removing half-edge" (problem 3.4) making this vertex invisible for the other vertices. Then the other sages do not see $B$, have two attempts, and lose by the statement of problem $2.6 \mathrm{~b}$ ). Giving them a disproving hats placement, the referee will make so that sage $B$ will not guess correctly too.
4.2. Fix number $s$. Consider the case, when the upper vertex has hatness $s+1$, and the other vertices have hatness $h$. Acting as in the previous problem, we make the upper vertex invisible for the others, as a result, the number of guesses of the vertices on path $P_{n-1}$ becomes equal to $s(s+1)=f(s)$. By the statement of problem $2.6 \mathrm{~b}$ ) the sages lose for $h>f(f(s))$. Thus $t_{s} \leqslant f(f(s))$.
4.3. We will prove by induction on the number of petals that for $h>f(f(f(s)))$ the game is losing. The base case, one petal $G$, follows from the previous problem: game $\langle G, \star h, \star s\rangle$ is already losing for $h>f(f(s))$.
But we need one more relative statement - a "modified base of the induction": game $\langle G, \star \bar{h}, \star s\rangle$ is losing for $h>f(f(f(s)))$, where $\star \bar{h}$ denotes the hatness function that is equal to $h$ in all the vertices of petal $G$, except one vertex $B$ for which $\bar{h}(B)=s+1$.
Proof of the statement. If $B$ is the upper vertex of petal $G$, then this is again the statement of the base. Let $B$ be an arbitrary vertex of petal on path $P_{n-1}$ (fig. 7, left). Using constructor "removal of half-edge" (problem 3.4), make vertex $B$ invisible for the other vertices. As a result the number of guesses of its neighbours becomes equal to $f(s)$. Now vertex $B$ can be deleted from $G$, because nobody sees it, it has hatness $s+1$ and $s$ guesses, so it is fated to fail in guessing. It can be easily seen that the remained graph is a union of two petals with common upper vertex (or one petal, but this case is trivial) and its vertices have $s$ or $f(s)$ guesses (fig. 7, center). Add to graph horizontal edge between former neighbours of $B$ (fig. 7, right), this edge can help to the sages to win. As a result, we have obtained a petal, which vertices have hatnesses at least $f(f(f(s)))$ and at most $f(s)$ guesses. By the statement of the base the sages nevertheless lose.
Now prove the inductive step. Consider petunia $M_{n}=M_{n-1}+_{v_{n}} L_{n}$. In view of modified statement of the base game $\left\langle L_{n}, \star \bar{h}, \star s\right\rangle$ is losing for $h>f(f(f(s)))$, where we denote by $\star \bar{h}$ the hatness function that is equal to $h$ in all the vertices of petal $G$, except $v_{n}$, and hatness of $v_{n}$ is equal to $s+1$. Game $\left\langle M_{n-1}, \star h, \star s\right\rangle$ is losing by the induction hypothesis. It remains to note that game $\left\langle M_{n}, \star h, \star s\right\rangle$ can be obtained by constructor of problem 3.3 from losing games $\left\langle M_{n-1}, \star h, \star s\right\rangle$ and $\left\langle L_{n}, \star \bar{h}, \star s\right\rangle$ and therefore is losing too.

Figure 7. Removing of vertex $B$ from petal $G$
4.4. a) Estimation $\operatorname{HG}_{s}(G)<4 s(s+1)$ - 1. Let $A$ be the stem of the petal, $h(A)=s+1$, and the other vertices $v$ have hatness $h(v)=4 s(s+1)-1$. It is sufficient to verify that game $\langle G, h, \star s\rangle$ is losing. By the statement of problem 5.6, this game is equivalent to the game on path $P_{n}$, where all vertices have hatness equal to $4 s(s+1)-1$ and $s(s+1)$ guesses. But this game is losing by problem 2.3.
Now we will prove that game $\langle G, \star 4 s(s+1)-2, \star s\rangle$ is winning for sufficiently large $n$.
Let $\mathcal{G}_{0}=\left\langle P_{k}, \star 4 s(s+1)-2, \star s(s+1)\right\rangle$. For sufficiently large $k$ this game is winning by problem 2.5.
Similarly to the problem $2.6 \mathrm{c}$ ), one can prove that for any natural $h$ there exists such natural $n$, that game on graph $K_{1, n}$ is winning, if the hatnesses of all peripheral sages are equal to $s+1$ and they all have one attempt for guessing, and the hatness of the central sage is equal to $h$ and he has $s$ attempts. Set $h=4 s(s+1)-2$ and choose suitable $n$. Substitute with reducing game $\mathcal{G}_{0}$ in the place of each peripheral sage ${ }^{1}$. We obtain a winning game, where the hatnesses of all vertices are equal to $4 s(s+1)-2$, the numbers of attempts are equal to $s$, and the graph is a subraph of a large petal.
b) As we have checked in p. a), game $\langle G, h, \star s\rangle$ is losing, where $G$ is a petal with large number of vertices, and $h$ is the function defining hatness of the stem by $s+1$, and the hatnesses of the other vertices by $4 s(s+1)-1$. By the statement of problem 3.3 gluing of stem of such petal to a vertex of another losing game with $s$ guesses gives again a losing game. But royal petunia by definition is constructed by consecutive stem's glueings of petals! Therefore a game on a royal petunia with large petals, where all the vertices have hatness $4 s(s+1)-1$ and $s$ guesses, except the first (rooted) stem with hatness $s+1$ and $s$ guesses, is losing. It gives the estimation $\mathrm{HG}_{s}(G)<4 s(s+1)-1$.
Then $\mathrm{HG}_{s}(G)=4 s(s+1)-2$, since this hatness is realized already on petals that are royal petunias too though not very branchy.
5.1. "If" case. Fix one of $h_{0}$ colors of the central sage's hat. Then the strategies of the other vertices are determined, and in $\prod_{i=1}^{n}\left(h_{i}-g_{i}\right)$ cases none of pendant vertices guesses correctly. Therefore the central sage must guess. But the central sage can do this only in $g_{0} \prod_{i=1}^{n} h_{i}$ cases in total. We obtain an inequality that is equivalent to the inequality from the condition.
"Only if" case. We will show that for $N=\prod_{i=1}^{n} h_{i}$ game $\left\langle K_{1, n},\left(N \cdot h_{0} ; h_{1}, \ldots, h_{n}\right),\left(N \cdot g_{0} ; g_{1}, \ldots, g_{n}\right)\right\rangle$ is winning.
Encode $h_{0} \cdot h_{1} \cdot \ldots \cdot h_{n}$ colors of the central sage by sets $\left(c_{0} ; c_{1}, \ldots, c_{n}\right)$, where $0 \leqslant c_{i}3 g-h$. Then there exists $k$ such that both numbers $c_{C}-\left[k \cdot \frac{h}{g}\right]$ and $c_{C}-\left[(k+3 g-h) \cdot \frac{h}{g}\right]$ belong to $I_{A} \cap I_{C}$ (the elements of $I_{A} \cap I_{C}$ can be written in two forms: $c_{A}-i=c_{C}-\left[\ell \cdot \frac{h}{g}\right]$; the number $\ell$ that corresponds to the minimal possible $i$, can be taken as $k$ ). Since both numbers belong to set $I_{A}$, consisting of consecutive remainders, the distance between them does not exceed $g-1$ :
$$
\left(c_{C}-\left[k \cdot \frac{h}{g}\right]\right)-\left(c_{C}-\left[(k+3 g-h) \cdot \frac{h}{g}\right]\right) \leqslant g-1
$$
that is equivalent
$$
\left[(k+3 g-h) \cdot \frac{h}{g}\right]-\left[k \cdot \frac{h}{g}\right] \leqslant g-1
$$
Getting rid of rounding, we obtain the corollary:
$$
(k+3 g-h) \cdot \frac{h}{g}-0.5-k \cdot \frac{h}{g}-0.5 \leqslant g-1
$$
The last is equivalent to the inequality $(3 g-h) \cdot \frac{h}{g} \leqslant g$, i. e. $0 \leqslant g^{2}-3 g h+h^{2}$, that contradicts the condition.
5.3. By the statement of problem 5.1 the existence of $k$, for which game $\left\langle P_{3}, \star k h, \star k g\right\rangle=\left\langle K_{1,2}, \star k h, \star k g\right\rangle$ is winning, is equivalent to the condition
$$
\left(1-\frac{g}{h}\right)\left(1-\frac{g}{h}\right) \leqslant \frac{g}{h}
$$
For non-negative $h$ it is equivalent to the inequality $g^{2}-3 g h+h^{2} \leqslant 0$, that for $1 \leqslant g \leqslant h$ is equivalent to the inequality
$$
\frac{h}{g} \leqslant \frac{3+\sqrt{5}}{2}
$$
Now the problem statement is evident.
5.4. Consider an arbitrary strategy of the sages in game $\langle G, h\rangle$. The product $\prod_{v \in H_{2}} h_{2}(v)$ enumerates hats placements on $H_{2}$. When we choose each of these hats placements, we fix strategy of the sages on $G_{1}$. For the fixed hats placement on $H_{2}$ consider hats placements on $G_{1}$. The sum $\sum_{u \in G_{1}} \frac{1}{h_{1}(u)}$ estimates from above the fraction of those placements on $G_{1}$, where at least one sage from $G_{1}$ guesses
correctly. Then the product $\left(\sum_{u \in G_{1}} \frac{1}{h_{1}(u)}\right) \prod_{v \in H_{2}} h_{2}(v)$ estimates from above the maximum fraction of those placements on $G_{1}$, where at least one sage from $G_{1}$ guesses correctly provided that each sage makes $\prod_{v \in H_{2}} h_{2}(v)$ guesses. Therefore the inequality from the problem condition means that there exists hats placement $\alpha$ on $G_{1}$, for which none of sages from $G_{1}$ guesses correctly, whatever hats were given to the sages from $H_{2}$. Therefore after assigning hats placement $\alpha$ to sages from $G_{1}$, none of them guesses correctly and the strategies of the other sages on graph $G_{2}$ are completely determined and now are suitable for game $\left\langle G_{2}, h_{2}\right\rangle$. Since this game is losing, hats placement $\alpha$ can be enlarged to hats placement on $G_{2}$, for which nobody from $G_{2}$ guesses correctly too.
5.5. Let the sages have fixed a strategy on graph $G^{\prime}$. We will construct a disproving hats placement for this strategy. The strategy of sage $A$ for each of $2 h(B)-1$ possible colors of hat of $B$ prescribes to name one of two colors. Some of these two colors is named at most $h(B)-1$ times. Give to sage $A$ the hat of this color, this will fix the strategy of sage $B$ on the remained graph $G$. Now, to prevent correct guessing of $A$, we give to $B$ a hat of one of at least $h(B)$ remained colors. Since game $\mathcal{G}$ is losing, we can construct hats placement on graph $G$ so that nobody on $G$ will guess correctly.
5.6. If $\mathcal{G}$ is a winning game, we remove by problem 3.4 all half-edges $\overrightarrow{v A}$ and obtain the winning game. In this game $A$ has $s$ guesses, $s+1$ colors and no information, therefore we can assign $A$ 's color such that $A$ does not guess. But now the remaining sages play the game $\mathcal{G}^{\prime}$. Hence $\mathcal{G}^{\prime}$ is winning.

3.6. We obtain a winning game $\mathcal{G}$.
5.7. a) Let $\mathcal{G}$ be a losing game on path $A B C$ where $h(A)=h(C)=2, h(B)=5$. If both games were losing, then the initial game would be obtained from these games and game $\mathcal{G}$ by constructor of problem 3.3 (where $s=1, g_{1}=g_{2}=\star 1$ ) and were losing too.
b) Set $V(\widetilde{G})$ consists of the vertices of graph $G$ and the set of new vertices $V_{1}$ (that are situated in the middles of two-link paths). Define a function on $V(\widetilde{G})$ :
$$
h(v)= \begin{cases}2, & v \in V(G) \\ 5, & v \in V_{1}\end{cases}
$$
It is sufficient to verify that game $\langle\widetilde{G}, h\rangle$ is losing. It is evident. Indeed, each sage from $V_{1}$ has two neighbours in $\widetilde{G}$ with hatness 2 , so he names by his strategy at most four colors. Then we give him a hat of the color, that he does not name, and he will not guess correctly. Now all the answers of the sages from $V(G)$ are determined, and we give to each sage a hat of the color that he does not name, too.
5.8. Let after deletion of the bridge graph $G$ fall into components $G_{1}$ (containing vertex $B$ ) and $G_{2}$ (containing vertex $A$ ). Define hatness functions $h_{1}$ and $h_{2}$ on these graphs by the rule
$$
h_{1}(x)=\left\{\begin{array}{ll}
h(x), & x \in V\left(G_{1}\right) \backslash\{B\}, \\
\left\lceil\frac{h(B)}{2}\right\rceil, & x=B
\end{array} \quad h_{2}(x)= \begin{cases}h(x), & x \in V\left(G_{2}\right) \backslash\{A\} \\
\left\lceil\frac{h(A)}{2}\right\rceil, & x=A\end{cases}\right.
$$
Let $\mathcal{G}_{1}^{\prime}=\left\langle G_{1},\left.h\right|_{G_{1}}\right\rangle$, it is a losing game due to properties of function $h$.
If games $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}\right\rangle$ and $\mathcal{G}_{2}=\left\langle G_{2}, h_{2}\right\rangle$ are both winning, then game $\mathcal{G}_{1} \times_{B} \stackrel{2}{\bullet_{B}} \quad \stackrel{2}{A} \times{ }_{A} \mathcal{G}_{2}$ is also winning. If the values $h(A)$ and $h(B)$ are even, we obtain the desired decomposition in the product of games. But if at least one of the numbers $h(A), h(B)$ is odd, then the hatness function of the obtained game majorizes $h$ and by properties of function $h$ such game cannot be winning, a contradiction.
It remains to consider the case, when at least one of games game $\mathcal{G}_{1}=\left\langle G_{1}, h_{1}\right\rangle$ or $\mathcal{G}_{2}=\left\langle G_{2}, h_{2}\right\rangle$ is losing, let it be game $\mathcal{G}_{2}$. Let us apply constructor of problem 5.5 to loosing game $\mathcal{G}_{2}$ : take a new
vertex $B$ of hatness 2 connected with $A$, let the hatness of vertex $A$ become equal to
$$
2\left\lceil\frac{h(A)}{2}\right\rceil-1 \leqslant h(A)
$$
and the hatnesses of other vertices be defined by function $h$. Denote the obtained game by $\mathcal{G}_{2}^{\prime}$. Apply now constructor of problem 3.3 to games $\mathcal{G}_{1}^{\prime}$ and $\mathcal{G}_{2}^{\prime}$ (we assume that $s=1, g_{1}=g_{2}=\star 1$ ). We will obtain a losing game on graph $G$, in which hatness function does not exceed $h$. It is impossible.
5.9. Denote the initial game by $\mathcal{G}_{1}=\left(G_{1}, h_{1}, g_{1}\right)$, where $g_{1} \equiv 1$. Denote by $G_{2}$ the subgraph of graph $G_{1}$ obtained from $G_{1}$ by removing vertex $A$, denote by $G_{3}$ the subgraph of $G_{2}$ obtained from $G_{2}$ by removing vertex $B$, and denote by $G_{4}$ the graph obtained from $G_{3}$ by adding vertex $C$, which is connected to all other vertices, i. e. in fact $G_{4}$ is obtained from $G_{2}$ by renaming vertex $B$ to $C$. Consider games $\mathcal{G}_{2}=\left(G_{2}, h_{2}, g_{2}\right), \mathcal{G}_{3}=\left(G_{3}, h_{3}, g_{3}\right), \mathcal{G}_{4}=\left(G_{4}, h_{4}, g_{4}\right)$, where
$$
\begin{aligned}
& h_{2}(v)=\left\{\begin{array}{ll}
h_{1}(v) & v \in G_{3}, \\
3 & v=B
\end{array}, \quad h_{3}(v)=\left.h_{1}\right|_{G_{3}}, \quad h_{4}(v)= \begin{cases}h_{1}(v) & v \in G_{3} \\
6 & v=C\end{cases} \right. \\
& g_{2} \equiv 2 \\
& g_{3} \equiv 6 \\
& g_{4}(v)= \begin{cases}1 & v \in G_{3} \\
5 & v=C\end{cases}
\end{aligned}
$$
Suppose that game $\mathcal{G}_{1}$ is winning. When we remove by problem 3.4 all the half-edges leading from vertices $A$ and $B$, we obtain a winning game, in which the number of guesses of all other sages (i.e. the sages from $G_{3}$ ) become equal to 6 , and the strategies do not depend of colors of $A$ 's and $B$ 's hats. Assume that there exists a hats placement on $G_{3}$, for which nobody from $G_{3}$ guesses correctly. This hats placement determines the strategies of sages $A$ and $B$, playing on edge $A B$ in game $\stackrel{\frac{1}{2}}{\stackrel{\bullet}{*}} \quad \stackrel{\frac{1}{3}}{B}$, and as a result all the sages lose. That is impossible. Therefore the restriction of the game to graph $G_{3}$, i.e. game $\mathcal{G}_{3}$, is winning. Making by problem 3.6 substitution of this game with reducing in game $\underset{C}{\frac{5}{6}} \underset{v}{\frac{1}{6}}$ in the place of vertex $v$, we obtain winning game $\mathcal{G}_{4}$.
Conversely, let game $\mathcal{G}_{4}$ be winning. We wish to replace player $C$ by two players $A$ (with hatness 2) and $B$ (with hatness 3 ). Let us demonstrate the winning strategy of the players in the obtained game. All players from $G_{3}$ will use the strategy of game $\mathcal{G}_{4}$, interpreting pair (color $A$, color $B$ ) as composite color of player $C$. Show how one can "convert" the strategy of player $C$ to a pair of strategies of $A$ and $B$.
The color of player $C$ is an element of the set $\mathcal{C}=\{0,1\} \times\{0,1,2\}$. Let for the current hats placement of his neighbours $C$ must name all the colors from set $\mathcal{C}$, except $(1,2)$. Then the actions of players $A$ and $B$ consist in that $A$ names the color with the same parity as $B$ 's color and $B$ names color 0 or 1 of opposite parity to $A$ 's color. Player $C$ has guessed correctly, if one of colors $(0,0),(0,2)$, $(1,1)$ or $(0,1),(1,0)$ was on his head. By our rule in the first three cases $A$ guesses correctly, in the other two $B$ guesses correctly.
The roles of $A$ and $B$ are assigned similarly for the other sets of five guesses of $C$.
5.10. Show that game $\left\langle G, \star \mathrm{HG}_{s^{\prime}}(G[A])+1, \star s\right\rangle$, where $s^{\prime}=s\left(\mathrm{HG}_{s}(G[B])+1\right)^{d}$, is losing. For this we construct hats placement such that the sages will lose. Since $\mathrm{HG}_{s^{\prime}}(G[A]) \geqslant s^{\prime} \geqslant \mathrm{HG}_{s}(G[B])+1$, i.e. the hatness in the game under consideration is larger than $\mathrm{HG}_{s}(G[B])+1$, it is sufficient to consider the case, when the hatness of the sages from $B$ is equal to $\operatorname{HG}_{s}(G[B])+1$. Applying problem 3.4, remove all half-edges from $A$ in $B$, making set $B$ invisible for set $A$. Since during this action we erase at most $d$ half-edges for each sage from $A$, the number of sages' guesses will increase, but will not exceed $s^{\prime}$. And the strategy of the sages from $A$ now does not depend on the hats placement on $B$. Therefore we can assume that they play a game $\left\langle G[A], \star \mathrm{HG}_{s^{\prime}}(G[A])+1, \star s^{\prime}\right\rangle$ on graph $G[A]$. By the definition of $s$-hat number this game is losing. Thus, there exists a hats placement on $A$, such that nobody from $A$ guesses correctly. Give to the sages from $A$ this placement, then the strategy of the sages from $B$ of game on $G[B]$ is determined. Since the hatnesses of the sages from $B$ are greater than $\mathrm{HG}_{s}(G[B])$, we can assign colors on $B$ so that the sages will lose.
# Многочлены в теории чисел
Проект подготовили и представляют:
Павел Козлов, Илья Богданов, Павел Кожевников, Андрей Рябичев, Борис Френкин, Navid Safaei
## Аннотация
В этом проекте мы собираемся познакомить участников с несколькими замечательными задачами и идеями, возникающими на стыке теории чисел и теории многочленов. Этот проект будет состоять из нескольких разделов, каждый из них будет по-своему интересен.
В разделе 0 собраны вспомогательные факты, которые будет полезно знать для решения задач.
Во разделе 1 речь пойдет о делителях (и простых делителей) значений различных многочленов с целыми коэффициентами. Здесь удается поработать с некоторыми конкретными многочленами, а также установить некоторые общие, причем весьма красивые и удивительные, результаты. Например, оказывается, для любого непостоянного многочлена с целыми коэффициентами существует бесконечно много простых чисел вида $4 k+1$, на которые делятся его значения в целых точках. А если потребовать от многочлена дополнительно наличие вещественного корня, то это же можно сказать о простых числах вида $4 k+3$.
Раздел 2 посвящён в основном одной сложной теоретико-числовой задаче, в которой описываются все ситуации, когда произведения $n$ последовательных чисел, увеличенные на константу, являются точными степенями. По ходу решения этой задачи участники ознакомятся с аналитическими и числовыми методами, полезными и для решения других задач.
В разделе 3 собраны разные задачи, представляющие самостоятельный интерес и в некотором смысле родственные другим задачам проекта.
## 0 Вспомогательные факты
## 0.1 Основы теории чисел
Перечислим некоторые теоретико-числовые сведения, которые могут потребоваться при решении задач.
- Деление с остатком и алгоритм Евклида.
- Линейное представление НОД.
- Основная теорема арифметики (о существовании и единственности разложения натурального числа на простые сомножители).
Следствие основной теоремы арифметики - наличие для любого натурального числа $n>1$ канонического разложения $n=p_{1}^{\alpha_{1}} \ldots p_{k}^{\alpha_{k}}$, где $p_{1}, \ldots, p_{k}$ - различные простые числа, $\alpha_{1}, \ldots, \alpha_{k}$ - степени их вхождения в разложение.
Степень вхождения простого числа $p$ в разложение целого $n$ ниже будем обозначать $v_{p}(n)$.
- Китайская теорема об остатках.
Одна из ее переформулировок такова: Пусть даны $n$ бесконечных в обе стороны арифметических прогрессий, состоящих из целых чисел, причем разности прогрессий $d_{1}, \ldots, d_{n}$ попарно взаимно просты. Тогда пересечение этих прогрессий - прогрессия с разностью $d_{1} \ldots d_{n}$.
- Сравнения по модулю и вычеты.
Запись $a \equiv b(\bmod n)(a$ и $b$ сравнимы по модулю $n$ ) для целых $a, b$ и натурального $n$ означает, что $a$ и $b$ имеют равные остатки при делении на $n$, или что $n \mid(a-b)$.
Вычетом по модулю $n$ называют класс эквивалентности относительно отношения равенства по модулю $n$ (каждый такой класс представляет собой арифметическую прогрессию с разностью $n$ ); впрочем, вычетом иногда называют и любого представителя класса. Все $n$ вычетов образуют полную систему вычетов, которую обозначаем $\mathbb{Z}_{n}$. Можно считать, что $\mathbb{Z}_{n}=\{0,1, \ldots, n-1\}$, если для каждого вычета выбрать соответствующего представителя. На множестве вычетов естественно вводятся операции сложения и умножения. Обратным к вычету $a$ называется вычет $a^{-1}$ такой, что $a a^{-1}$ равно 1 в $\mathbb{Z}_{n}$, т.е. $a a^{-1} \equiv 1(\bmod n)$. Вычет $a$ обратим (т.е. имеет обратный) тогда и только тогда, когда НОД $(a, n)=1$.
- Теоремы Эйлера и Ферма.
Теорема Эйлера гласит, что для взаимно простых $a \in \mathbb{Z}$ и $n \in \mathbb{N}$ выполнено $a^{\varphi(n)} \equiv 1$ $(\bmod n)$, где $\varphi(n)$ - функция Эйлера числа $n$, т.е. количество чисел из множества $\{1, \ldots, n\}$, взаимно простых с $n$.
Теорема Ферма - частный случай теоремы Эйлера для простого $n=p$.
- Показателем целого числа $a$ по модулю натурального $n$ называется наименьшее натуральное $k$ такое, что $a^{k} \equiv 1(\bmod n)$. (Ясно, что показатель определен только в случае НОД $(a, n)=1$.) Показатель обозначаем $\operatorname{ord}_{n}(a)$.
Отметим следующее свойство показателя: $a^{m} \equiv 1(\bmod n)$ тогда и только тогда, когда $\operatorname{ord}_{n}(a) \mid m$. В силу теоремы Эйлера отсюда, в частности, следует, что $\operatorname{ord}_{n}(a) \mid \varphi(n)$.
Если $\operatorname{ord}_{n}(a)=\varphi(n)$, то степени $1, a, a^{2}, \ldots, a^{\varphi(n)-1}$ пробегают все вычеты, взаимнопростые с $n$. В таком случае $a$ называют первообразным корнем по модулю $n$. Первообразный корень существует тогда и только тогда, когда $n$ равно $2,4, p^{\alpha}$ или $2 p^{\alpha}$ для простого $p$ и натурального $\alpha$.
- Напомним LTE-лемму (Lifting the Exponent Lemma).
Пусть $p$ - нечётное простое число, $a, b$ - различные целые. Тогда если $p \mid a-b$, то
$$
v_{p}\left(a^{n}-b^{n}\right)=v_{p}(a-b)+v_{p}(n)
$$
Если же $p=2$, то это верно при условии $4 \mid a-b$.
- Ненулевой вычет $a$ по простому нечетному модулю $p$ называется квадратичным, если сравнение $x^{2} \equiv a(\bmod p)$ имеет непустое множество решений. Количество (ненулевых) квадратичных вычетов $\frac{p-1}{2}$ - столько же, сколько невычетов.
Справедлив критерий Эйлера: ненулевой вычет $a$ является квадратичным тогда и только тогда, когда $a^{\frac{p-1}{2}} \equiv 1(\bmod p)$.
## 0.2 Теория делимости в $\mathbb{Q}[x]$
Множество многочленов с рациональными коэффициентами обозначим $\mathbb{Q}[x]$, а множество многочленов с целыми коэффициентами - $\mathbb{Z}[x]$. Многочлены степени не меньше 1 назовём непостоянными, так что постоянные многочлены - это многочлены, тождественно равные некоторому числу. Непостоянный многочлен $f \in \mathbb{Q}[x]$ степени $d$ называется приводимьм над $\mathbb{Q}$ (или приводимым в $\mathbb{Q}[x]$ ), если он может быть представлен в виде произведения двух многочленов степени меньшей чем $d$ с рациональными же коэффициентами. Каждый непостоянный многочлен из $\mathbb{Q}[x]$ является либо приводимым, либо неприводимым над $\mathbb{Q}$. Так же, как и целые числа, многочлены из $\mathbb{Q}[x]$ можно делить с остатком (например, "уголком"). В $\mathbb{Q}[x]$ определяется НОД многочленов $f_{1}, f_{2}, \ldots, f_{k}$ как многочлен наибольшей степени, на который делится каждый из многочленов $f_{1}, f_{2}, \ldots, f_{k}$. НОД многочленов единственен с точностью до умножения на ненулевое число, НОД двух многочленов можно находить с помощью алгоритма Евклида. Многочлены, НОД которых равен 1, называются взаимно простыми. Аналогами теорем о линейном представлении НОД и основной теоремы арифметики для целых чисел являются следующие теоремы:
Факт 0.1 (Теорема о линейном представлении НОД). Для любых многочленов $f_{1}, f_{2}, \ldots, f_{k} \in \mathbb{Q}[x]$ найдутся такие многочлены $u_{1}, u_{2}, \ldots, u_{k} \in \mathbb{Q}[x]$, что
$$
\text { НОД }\left(f_{1}, f_{2}, \ldots, f_{k}\right)=u_{1} f_{1}+u_{2} f_{2}+\ldots u_{k} f_{k}
$$
Факт 0.2 (Основная теорема арифметики). Непостоянный многочлен $f \in \mathbb{Q}[x]$ представим в виде $f=h_{1} \cdot h_{2} \cdot \ldots \cdot h_{r}$, где $h_{i}$ - неприводимые над $\mathbb{Q}$ многочлены, причем если $f=h_{1}^{\prime} \cdot h_{2}^{\prime} \cdot \ldots \cdot h_{s}^{\prime}$ - другое такое разложение, то $s=r$ и для некоторой перенумерации $h_{1}^{\prime}, h_{2}^{\prime} \ldots h_{r}^{\prime}$ будет выполнено
$$
h_{1}^{\prime}=u_{1} h_{1}, \quad h_{2}^{\prime}=u_{2} h_{2}, \quad \ldots, \quad h_{r}^{\prime}=u_{r} h_{r}
$$
где $u_{1}, u_{2} \ldots u_{r}$ - ненулевые рациональные числа.
Группируя в разложении на неприводимые многочлены одинаковые сомножители, можно получить каноническое разложение, подобно каноническому разложению натурального числа на простые сомножители.
## 0.3 Лемма Гаусса
Назовем содержанием многочлена с целыми коэффициентами НОД всех его коэффициентов. Далее содержание многочлена $f$ обозначим $d(f)$. Если $d(f)=1$, то многочлен $f$ называют примитивным.
Факт 0.3. Пусть $f, g \in \mathbb{Z}[x]$. Тогда $d(f g)=d(f) \cdot d(g)$. В частности, произведение двух примитивных многочленов снова будет примитивным многочленом.
Доказательство. Начнем с последнего утверждения. Пусть $f(x)=a_{0}+a_{1} x+\ldots a_{n} x^{n}$ и $g(x)=b_{0}+b_{1} x+\ldots b_{m} x^{m}$ - примитивные многочлены из $\mathbb{Z}[x]$, произведение которых не является примитивным, то есть для некоторого простого числа $p$ верна делимость $p \mid d(f g)$.
Выберем наименьшие индексы $s, t$, для которых $p \nmid a_{s}, p \nmid b_{t}$. Такие индексы найдутся в силу примитивности $f$ и $g$. Коэффициент при $x^{s+t}$ в многочлене $f g$ будет равен
$$
c_{s+t}=a_{s} b_{t}+\left(a_{s+1} b_{t-1}+a_{s+2} b_{t-2}+\ldots\right)+\left(a_{s-1} b_{t+1}+a_{s-2} b_{t+2}+\ldots\right)
$$
Так как $a_{s-i}$ и $b_{t-i}$ при $i>0$ делятся на $p$ по условию и $p \mid c_{s+t}$ по предположению, то $p \mid a_{s} b_{t}$, а это противоречит выбору индексов $s$ и $t$.
Переходя к общему случаю, запишем произвольные многочлены $f, g \in \mathbb{Z}[x]$ в виде
$$
f=d(f) f_{0}, \quad g=d(g) g_{0}
$$
где $f_{0}, g_{0}$ - примитивные многочлены. Так как $f g=d(f) d(g) \cdot f_{0} \cdot g_{0}$ и по доказанному $d\left(f_{0} \cdot g_{0}\right)=1$, то $d(f g)=d(f) d(g)$, что и требовалось доказать.
Факт 0.4 (Лемма Гаусса). Многочлен из $\mathbb{Z}$, неприводимый над $\mathbb{Z}$, также неприводим над $\mathbb{Q}$.
Доказательство. Предположим, что для некоторых многочленов $h \in \mathbb{Z}[x]$ и $f, g \in \mathbb{Q}[x]$ выполняется равенство $h=f g$. После умножения обеих частей этого равенства на HOK знаменателей и вынесения HOK числителей всех коэффициентов у $f$ и $g$, оно переписывается в виде:
$$
a h=b f_{0} g_{0}
$$
где $a, b \in \mathbb{Z}$, а $f_{0}$ и $g_{0}$ - примитивные многочлены над $\mathbb{Z}$. По предыдущему пункту, $a \cdot d(h)=b$, так что после сокращения получается разложение
$$
h=d(h) f_{0} g_{0}
$$
поэтому многочлен $h$ приводим также и в $\mathbb{Z}[x]$.
С помощью леммы Гаусса и основной теоремы арифметики для $\mathbb{Q}[x]$ можно получить следующий
Факт 0.5 (Основная теорема арифметики в $\mathbb{Z}[x]$ ). Непостоянный многочлен $f \in \mathbb{Z}[x]$ представим в виде $f=h_{1} \cdot h_{2} \cdot \ldots \cdot h_{r}$, где $h_{i} \in \mathbb{Z}[x]$ - неприводимые над $\mathbb{Q}$ (непостоянные) многочлены.
Если в последнем разложении $f=h_{1} h_{2} \ldots h_{r}$ сгруппировать пропорциональные сомножители и вынести из каждого сомножителя НОД его коэффициентов, то мы придем к каноническому разложению многочлена с целыми коэффициентами
$$
f=u \cdot g_{1}^{\alpha_{1}} g_{2}^{\alpha_{2}} \ldots g_{k}^{\alpha_{k}}
$$
где $g_{i} \in \mathbb{Z}[x]$ - попарно взаимно простые примитивные неприводимые над $\mathbb{Q}$ (непостоянные) многочлены, $u \in \mathbb{Z}, u \neq 0, k \geqslant 1, \alpha_{i} \geqslant 1$.
Каноническое разложение единственно с точностью до домножения сомножителей на $\pm 1$.
## 0.4 Лемма Гензеля
В этой параграфе опишем очень важный результат, который не один раз пригодится в дальнейшем. Для начала заметим, что для любого постоянного многочлена $f$ верно следующее тождество:
$$
f(x+y)=f(x)
$$
Дальше, для любого линейного многочлена $f$ верно соотношение:
$$
f(x+y)=f(x)+c y
$$
где константа $c$ равна коэффициенту при $x$ или, другими словами, первой производной многочлена $f$.
Вышенаписанное несложно обобщить на случай произвольного многочлена (например, с помощью индукции по степени многочлена). Верен следующий факт: если $f-$ многочлен степени $d$ с целыми коэффициентами, то справедливо равенство:
$$
f(x+y)=f(x)+f_{1}(x) y+f_{2}(x) y^{2}+\ldots+f_{d}(x) y^{d}
$$
где $f_{i}, i=1,2, \ldots, d$ - многочлены с целыми коэффициентами степени не выше $d-1$; $f_{i}$ явно выражаются как $f_{i}=\frac{f^{(i)}}{i!}$.
Непосредственным следствием этого утверждения является следующая очень полезная лемма:
Факт 0.6 (Лемма Гензеля). Пусть $f \in \mathbb{Z}$ - многочлен с целыми коэффициентами. Тогда для любого простого $q$, натурального $s$ и целых $r, n$ верно следующее сравнение
$$
f\left(n+r q^{s}\right) \equiv f(n)+r q^{s} f^{\prime}(n) \quad\left(\bmod q^{2 s}\right)
$$
## 1 Делители значений многочлена и приложения
Введем следующие обозначения. Для многочлена $f \in \mathbb{Z}[x]$ через $D(f)$ обозначим множество натуральных чисел $m$, для которых $f(n)$ делится на $m$ при некотором целом $n$. Иначе говоря, $m \in D(f)$ тогда и только тогда, когда сравнение $f(x) \equiv 0(\bmod m)$ имеет целочисленное решение.
Более общо, для многочленов $f_{1}, f_{2}, \ldots, f_{k} \in \mathbb{Z}[x]$ через $D\left(f_{1}, f_{2}, \ldots, f_{k}\right)$ обозначим множество натуральных чисел $m$ таких, что система сравнений
$$
f_{i}(x) \equiv 0 \quad(\bmod m), \quad i=1,2, \ldots, k
$$
имеет целочисленное решение. Через $P\left(f_{1}, f_{2}, \ldots, f_{k}\right)$ будем обозначать множество всех простых чисел, принадлежащих $D\left(f_{1}, f_{2}, \ldots, f_{k}\right)$.
Ясно, что если $f(n)$ делится на $m$, то и $f(n+m t)$ делится на $m$ для любого целого $t$. Таких образом, множество целых чисел $x$, удовлетворяющих системе (1), либо пустое, либо является объединением бесконечных в обе стороны арифметических прогрессий с разностью $m$.
Очевидно, вместе с каждым $m \in D(f)$ в множество $D(f)$ входят и все натуральные делители $m$. Для постоянного многочлена $f=u \in \mathbb{Z}$ множество $D(u)$ совпадает с множеством (натуральных) делителей числа $u$. Легко видеть, что для непостоянного многочлена $f \in \mathbb{Z}[x]$ множество $D(f)$ бесконечно.
Задача 1.1. а) Докажите, что для любых заданных натуральных $m_{1}, \ldots, m_{n}$ существует многочлен $f \in \mathbb{Z}[x]$, для которого $m_{1}, \ldots, m_{n}$ не принадлежат $D(f)$.
б) Докажите, что $D(f g) \supset D(f)$ для любых двух многочленов $f, g \in \mathbb{Z}[x]$.
в) Докажите, что $D(f(g)) \subset D(f)$ для любых двух многочленов $f, g \in \mathbb{Z}[x]$.
Задача 1.2. Для любого непостоянного многочлена $f \in \mathbb{Z}[x]$ множество $P(f)$ бесконечно.
Задача 1.3. Если $f_{1}, f_{2}, \ldots, f_{k} \in \mathbb{Z}[x], k \geqslant 2$ - взаимно простые в совокупности многочлены, то множество $D\left(f_{1}, f_{2}, \ldots, f_{k}\right)$ конечно. (И, следовательно, множество $P\left(f_{1}, f_{2}, \ldots, f_{k}\right)$ конечно.)
Задача 1.4. Пусть $f \in \mathbb{Z}[x]$. Докажите, что если $m_{1}, m_{2} \in D(f)$, и $\operatorname{HOД~}\left(m_{1}, m_{2}\right)=1$, то $m_{1} m_{2} \in D(f)$.
Задача 1.5. Найдите $P(f)$, если
а) $f(x)=x^{2}+1$
б) $f(x)=x^{2}+x+1$
в) $f(x)=x^{p-1}+x^{p-2}+\ldots+x+1$, где $p>2-$ простое число.
Задача 1.6. а) Ясно, что если многочлен $f \in \mathbb{Z}[x]$ имеет целый корень, то $P(f)$ совпадает с множеством всех простых чисел. Докажите, что обратное утверждение неверно.
б) Пусть $P(f)$ совпадает с множеством всех простых чисел. Следует ли отсюда, что у $f$ есть рациональный корень?
Для многочлена $f \in \mathbb{Z}[x]$ через $P_{\alpha}(f)$ обозначим множество простых чисел $p$ таких, что $v_{p}(f(n))=\alpha$ для некоторого целого $n$.
Ясно, что если $v_{p}(f(n))=\alpha$, то и $v_{p}\left(f\left(n+p^{\alpha+1} k\right)\right)=\alpha$ при любом целом $k$. Таких образом, множество целых чисел $x$ таких, что $v_{p}(f(x))=\alpha$, либо пустое, либо является объединением бесконечных в обе стороны арифметических прогрессий с разностью $p^{\alpha+1}$.
Легко видеть, что $P_{\alpha}(f) \subset P(f)$, причем для $f \in \mathbb{Z}[x]$, не равного тождественно нулю, $P(f)=\bigcup_{\alpha=1}^{\infty} P_{\alpha}(f)$. Отметим также, что для $f \in \mathbb{Z}[x]$ и $k \in \mathbb{N}$ выполнено $P_{\alpha}(f)=P_{k \alpha}\left(f^{k}\right)$.
Задача 1.7. Пусть $f \in \mathbb{Z}[x]$.
a) Если $p \in P(f) \backslash P\left(f, f^{\prime}\right)$, то $p \in P_{\alpha}(f)$ для сколь угодно больших $\alpha$.
б) Докажите, что множества $P_{\alpha}(f) \backslash P\left(f, f^{\prime}\right)$ совпадают с $P(f) \backslash P\left(f, f^{\prime}\right)$ при $\alpha=1,2,3, \ldots$
Задача 1.8. Ясно, что если многочлен $f \in \mathbb{Z}[x]$ имеет целый корень, то $D(f)=\mathbb{N}$. Верно ли обратное утверждение?
Задача 1.9. Пусть $f \in \mathbb{Z}[x]$ таков, что при каждом натуральном $n$ значение многочлена $f(n)$ является нетривиальной точной степенью некоторого натурального числа, т.е. $f(n)=m(n)^{s(n)}$, где $m(n)$ и $s(n)>1$ - натуральные числа.
a) Докажите, что $f$ приводим над $\mathbb{Q}$.
б) Докажите, что существуют натуральное $s>1$ и многочлен $q \in \mathbb{Z}[x]$ такие, что $f=q^{s}$.
Задача 1.10. Докажите, что Для любого непостоянного многочлена $f \in \mathbb{Z}[x]$ множество $P(f)$ содержит бесконечно много простых чисел вида $4 k+1$.
## 2 Об одной задаче Сендерова
Задача 2.1. Многочлен $Q(x)$ с вещественными коэффициентами таков, что его старший коэффициент рациональный, и для некоторого натурального $k>1$ выполняется $Q(x)^{k} \in \mathbb{Z}[x]$. Докажите, что $Q(x) \in \mathbb{Z}[x]$.
Задача 2.2. Про многочлен $P(x)$ чётной степени с целыми коэффициентами известно, что его значения при любых достаточно больших натуральных $n$ и старший коэффициент являются квадратами целых чисел. Докажите, что существует многочлен $Q(x)$ с целыми коэффициентами такой, что $P(x)=Q(x)^{2}$.
Задача 2.3. Про многочлен $P(x)$ с целыми коэффициентами известно только то, что его значения при любых достаточно больших натуральных $n$ являются квадратами целых чисел. Докажите, что существует многочлен $Q(x)$ с целыми коэффициентами такой, что $P(x)=Q(x)^{2}$.
Здесь и далее используется следующее обозначение:
$$
P_{k}(x)=x(x+1) \ldots(x+k-1)
$$
где $k$ - некоторое натуральное число.
Задача 2.4. Пусть приведённый многочлен $Q(x)$ степени $l>2$ с действительными коэффициентами таков, что для некоторого положительного $b$ каждый из многочленов $R(x)=Q(x)-b$ и $S(x)=Q(x)+b$ имеет ровно $l$ различных действительных корней. Пусть $r_{1}4$ и $c$ не является квадратом другого многочлена с целыми коэффициентами.
Задача 2.6. Докажите, что для каждого комплексного $a$ степень НОД многочленов $P_{k}(x)-a$ и $P_{k}^{\prime}(x)$ не превосходит 2 , более того, если $k$ нечётно, то она не превосходит 1 .
Задача 2.7. Предположим, что существуют многочлены $Q(x), R(x)$ степени, не меньшей 2 , с рациональными коэффициентами такие, что $P_{k}(x)=R(Q(x))$. Докажите, что тогда $k$ чётно, и степень многочлена $Q(x)$ равна ровно 2.
Задача 2.8 (В. А. Сендеров). Найдите все натуральные числа $k, c$ такие, что выражение $P_{k}(n)+c$ является точной степенью для всех натуральных чисел $n$.
## 3 Разное
Задача 3.1. Даны многочлены $f, g \in \mathbb{Z}[x]$. Известно, что для бесконечно многих натуральных $n$ число $f(n)$ делится на $g(n)$.
а) Докажите, если если старший коэффициент $g$ равен 1 , то $g \mid f$ (в $\mathbb{Z}[x]$ ).
б) Докажите, что всегда найдется бесконечная арифметическая прогрессия $A$ натуральных чисел такая, что $g(n) \mid f(n)$ для любого $n \in A$.
Задача 3.2. Дан многочлен $f \in \mathbb{Z}[x]$ и натуральные числа $a_{1}, a_{2}, \ldots, a_{m}$. Известно, что для любого целого $n$ число $f(n)$ делится хотя бы на одно из чисел $a_{1}, a_{2}, \ldots, a_{m}$. Докажите, что существует такое натуральное $k$, что $k \leq m$ и $f(n)$ делится на $a_{k}$ для любого целого $n$.
Каждый многочлен $f \in \mathbb{Z}[x]$ определяет функцию $\mathbb{Z}_{n} \rightarrow \mathbb{Z}_{n}$, поскольку вычет $f(m)$ по модулю $n$ зависит только от вычета $m$ по модулю $n$.
Задача 3.3. а) Докажите, что если функция $\mathbb{Z}_{n} \rightarrow \mathbb{Z}_{n}$ реализуется многочленом из $\mathbb{Z}[x]$, то она реализуется и многочленом из $\mathbb{Z}[x]$, степень которого не превышает $n-1$.
б) Для каких $n$ любая функция $\mathbb{Z}_{n} \rightarrow \mathbb{Z}_{n}$ реализуется многочленом из $\mathbb{Z}[x]$ ?
в) Найдите все натуральные $n>1$, для которых верно следующее утверждение: если $f \in \mathbb{Z}, \operatorname{deg} f \leqslant n-1$ таков, что $f(m)$ делится на $n$ при всех целых $m$, то все коэффициенты $f$ делятся на $m$.
г) Для каких $n$ любая функция $\mathbb{Z}_{n} \rightarrow \mathbb{Z}_{n}$, удовлетворяющая условию
$$
\text { НОД }(a-b, n) \mid \text { НОД }(f(a)-f(b), n), \quad a, b \in \mathbb{Z}_{n}
$$
реализуется многочленом из $\mathbb{Z}[x]$ ?
Задача 3.4. Пусть $f \in \mathbb{Z}[x]$ задает биекцию по любому простому модулю. Докажите, что $f$ линейный.
Задача 3.5. Про многочлены $f$ и $g$ с целыми коэффициентами известно, что при любом целом $x$ число $f(g(x))-x$ делится на данное целое $n$. Докажите, что число $g(f(x))-x$ тоже делится на $n$ при любом целом $x$.
# Многочлены в теории чисел
(Продолжение)
## 0 Вспомогательные факты
## 0.5 Алгебраические числа и минимальные многочлены
Комплексное число $\alpha$ называется алгебраическим, если существует ненулевой многочлен $f \in \mathbb{Q}[x]$ такой, что $f(\alpha)=0$. Например, $\sqrt{2}+\sqrt{3}$ - алгебраическое число, поскольку $f(\sqrt{2}+\sqrt{3})=0$, где $f(x)=x^{4}-10 x^{2}+1$. Очевидно, множество алгебраических чисел содержит $\mathbb{Q}$.
Ненулевой многочлен $f \in \mathbb{Q}[x]$ минимальной степени, для которого $f(\alpha)=0$, называется минимальным многочленом алгебраического числа $\alpha$. Степень этого многочлена называется степенъю числа $\alpha$.
Несложно понять, что минимальный многочлен любого алгебраического числа всегда неприводим в $\mathbb{Q}[x]$. Кроме того, любой многочлен $h \in \mathbb{Q}[x]$, имеющий $\alpha$ своим корнем, делится на минимальный многочлен. Действительно, разделим многочлен $h$ на минимальный многочлен $f$ с остатком:
$$
h=f q+r
$$
где $q, r \in \mathbb{Q}[x]$ и $\operatorname{deg} r<\operatorname{deg} f$. Подставив $x=\alpha$ в уравнение выше, найдём, что $r(\alpha)=0$. Следовательно, из условия минимальности $f$ получаем, что $r$ тождественно равен 0 , то есть $h$ делится на $f$.
Из доказанного немедленно следует
Факт 0.7. Каждый неприводимый многочлен $f \in \mathbb{Q}[x]$ является минимальным многочленом для каждого из своих корней.
Задача 0.8. а) Пусть $f$ - неприводимый многочлен с рациональными коэффициентами. Докажите, что $f$ и $f^{\prime}$ взаимно просты.
б) Пусть $f$ - многочлен в $\mathbb{Q}[x]$. Докажите, что многочлены $f$ и $f^{\prime}$ не взаимно просты тогда и только тогда, когда существует непостоянный многочлен $g$ такой, что $g^{2} \mid f$.
в) Докажите, что у неприводимого (в $\mathbb{Q}[x]$ ) многочлена нет кратных комплексных корней (и, следовательно, количество его комплексных корней совпадает с его степенью).
Непустое множество комплексных чисел $\mathbb{F}$ называется числовым полем, если оно состоит не только из нуля, и вместе с любыми числами $a, b \in \mathbb{F}$ оно содержит также числа $a+b, a b,-a$ и $a^{-1}$ (последнее - в случае $a \neq 0$ ). Иначе говоря, в поле можно совершать все обычные арифметические операции (сложение, умножение, вычитание и деление на ненулевое число), получая числа, также принадлежащие этому полю. (В этом проекте под полями понимаются именно числовые поля.)
Поскольку большинство алгебраических выкладок использует ровно эти операции, многие факты, доказательства которых использует лишь эти действия, продолжают оставаться верными над любым полем. Так, обозначая через $\mathbb{F}[x]$ множество многочленов с коэффициентами из $\mathbb{F}$, нетрудно показать, что в $\mathbb{F}[x]$ верны теоремы о линейном представлении НОД и основная теорема арифметики (факты 0.1 и 0.2).
Здесь полезно заметить, что поиск НОД двух многочленов (алгоритмом Евклида) не зависит от того, над каким полем это происходит. Значит, и полученный НОД не зависит от поля (лишь бы все коэффициенты многочленов лежали в этом поле).
Аналогично определению выше, комплексное число $\alpha$ называется алгебраическим над полем $\mathbb{F}$, если $\alpha$ является корнем ненулевого многочлена из $\mathbb{F}[x]$. Такой многочлен минимальной степени называется минимальным многочленом числа $\alpha$ над полем $\mathbb{F}$. Сформулированные выше факты об алгебраических числах также переносятся на случай произвольного поля.
Важные базовые примеры полей получаются из следующего факта.
Факт 0.9. а) Пусть $\alpha$ - алгебраическое число. Тогда множество
$$
\mathbb{Q}[\alpha]=\{f(\alpha): f \in \mathbb{Q}[x]\}
$$
является полем.
б) Аналогично, если $\alpha$ - алгебраическое число над полем $\mathbb{F}$, то множество
$$
\mathbb{F}[\alpha]=\{f(\alpha): f \in \mathbb{F}[x]\}
$$
является полем.
Поле $\mathbb{F}[\alpha]$ называется расширением поля $\mathbb{F}$ алгебраическим числом $\alpha$.
Доказательство. Мы докажем сразу пункт б).
Для элементов $g(\alpha), h(\alpha) \in \mathbb{F}[\alpha]$ очевидно, что $g(\alpha)+h(\alpha),-g(\alpha)$ и $g(\alpha) h(\alpha)$ лежат в $\mathbb{F}[\alpha]$. Осталось доказать, что $g(\alpha)^{-1} \in \mathbb{F}[\alpha]$, если, конечно, $g(\alpha) \neq 0$,
Пусть $f \in \mathbb{F}[x]$ - минимальный многочлен числа $\alpha$ над $\mathbb{F}$; тогда $g$ не делится на $f$. поскольку $g(\alpha) \neq 0$. Из неприводимости $f$ теперь вытекает, что $Н О Д(f, g)=1$, и по теореме о линейном представлении НОД существуют такие многочлены $a, b \in \mathbb{F}[x]$, что $1=a f+b g$. Подставляя в это равенство $\alpha$, получаем
$$
1=a(\alpha) f(\alpha)+b(\alpha) g(\alpha)=b(\alpha) g(\alpha)
$$
то есть число $g(\alpha)^{-1}=b(\alpha)$ лежит в $\mathbb{F}[x]$.
Замечание. Применяя деление на $f$ с остатком, несложно получить, что каждый элемент в $\mathbb{F}[\alpha]$ представляется как $g(\alpha)$, где $g \in \mathbb{F}[x]$ и $\operatorname{deg} g<\operatorname{deg} f$ (проведите эту выкладку!).
Следующую задачу можно решать по-разному; один из возможных путей решения опирается на такую известную лемму.
Факт 0.10. Пусть задана система однородных линейных уравнений, то есть система вида
$$
\begin{gathered}
a_{11} x_{1}+a_{12} x_{2}+\ldots+a_{1 n} x_{n}=0 \\
a_{21} x_{1}+a_{22} x_{2}+\ldots+a_{2 n} x_{n}=0 \\
\vdots \\
a_{k 1} x_{1}+a_{k 2} x_{2}+\ldots+a_{k n} x_{n}=0
\end{gathered}
$$
где $a_{i j}$ - фиксированные (скажем, рациональные) коэффициенты, а $x_{i}$ - неизвестные. Тогда, если количество уравнений $k$ меньше количества неизвестных $n$, то у системы есть (рациональное) решение, в котором не все значения переменных нулевые.
Эту лемму можно доказать, например, последовательным исключением неизвестных. Лемма верна и над любым полем: если коэффициенты принадлежат полю $\mathbb{F}$, то можно найти и решение, в котором все значения неизвестных лежат в $\mathbb{F}$.
Следующая задача также допускает обобщение над любым полем, но для простоты восприятия мы формулируем её над полем $\mathbb{Q}$.
Задача 0.11. а) Пусть $\alpha$ - алгебраическое число, и $\mathbb{F}=\mathbb{Q}[\alpha]$. Тогда любое число в $\mathbb{F}$ алгебраично.
б) Пусть число $\alpha$ алгебраично, а число $\beta$ алгебраично над $\mathbb{F}=\mathbb{Q}[\alpha]$. Тогда $\beta$ алгебраично.
в) Множество $\overline{\mathbb{Q}}$ всех алгебраических чисел является полем.
г) Более того, любое число, алгебраическое над $\overline{\mathbb{Q}}$, лежит в $\overline{\mathbb{Q}}$. (Такие поля называются алгебраччески замкнутыми.
Пусть $\alpha$ и $\beta$ - алгебраические числа; положим $\mathbb{F}=\mathbb{Q}[\alpha]$. Тогда поле $\mathbb{F}[\beta]$ обозначается также как $\mathbb{Q}[\alpha, \beta]$. Нетрудно проверить, что $\mathbb{Q}[\alpha, \beta]=\mathbb{Q}[\beta, \alpha]$.
Аналогично можно определить поле $\mathbb{Q}\left[\alpha_{1}, \ldots, \alpha_{n}\right]$ при $\alpha_{i} \in \overline{\mathbb{Q}}$.
Задача 0.12 (Теорема о примитивном элементе). а) Пусть $\alpha$ и $\beta$ - два алгебраических числа. Докажите, что существует алгебраическое число $\gamma$ такое, что $\mathbb{Q}[\alpha, \beta]=\mathbb{Q}[\gamma]$. Более того, это число можно искать в виде $\gamma=\alpha+t \beta$ при $t \in \mathbb{Q}$.
б) Пусть $\alpha_{1}, \alpha_{2}, \ldots, \alpha_{n}$ - алгебраические числа. Докажите, что существует алгебраическое число $\gamma$ такое, что $\mathbb{Q}\left[\alpha_{1}, \alpha_{2}, \ldots, \alpha_{n}\right]=\mathbb{Q}[\gamma]$. В каком виде можно искать $\gamma$ в этом случае?
## 0.6 Круговые многочлены
Комплексное число $z$ называется примитивным корнем степени $n$ из 1 , если $z^{n}=1$, но $z^{k} \neq 1$ при $1 \leqslant k1, a_{i}$ - целье, $p-$ простое. Докажите, что $f$ задает функцию $\mathbb{Z}_{p} \rightarrow \mathbb{Z}_{p}$, не являющую биекцией, если
a) $d=p-1$;
б) $d$ - делитель $p-1$.
Задача 3.8. Пусть $f \in \mathbb{Z}[x]$ задает биекцию по модулю $n$. Докажите, что существует такой $g \in \mathbb{Z}[x]$, что $f(g(x))-x$ делится на $n$ при любом целом $x$.
## Многочлены вешения теории чисел
## 0.5 Алгебраические числа и минимальные многочлены
0.8. а) Предположим противное, существует непостоянный многочлен $d$ с рациональными коэффициентами, который делит как $f$, так и $f^{\prime}$. Поскольку $f$ неприводим, то $d=c \cdot f$, следовательно, $f$ должен делить $f^{\prime}$, что невозможно, поскольку $\operatorname{deg} f^{\prime}<\operatorname{deg} f$.
б) Пусть описанный многочлен $g$ существует, $f=g^{2} h$. Тогда
$$
f^{\prime}=g^{2} h^{\prime}+2 g g^{\prime} h=g \cdot\left(g h^{\prime}+2 g^{\prime} h\right)
$$
и потому $g \mid \operatorname{НОД}\left(f, f^{\prime}\right)$. Значит, $f$ и $f^{\prime}$ не взаимно просты.
Наоборот, если многочлена $g$ не существует, то в каноническом разложении
$$
f=h_{1} h_{2} \ldots h_{k}
$$
на неприводимые сомножители все многочлены $h_{1}, \ldots, h_{k}$ различны (более точно, среди них нет пропорциональных). Если бы $f$ и $f^{\prime}$ не были взаимно простыми, то их НОД был бы произведением некоторых из $h_{i}$; пусть для определённости он делится на $h_{1}$, то есть $h_{1} \mid f^{\prime}$. Но
$$
f^{\prime}=h_{1}^{\prime} h_{2} \ldots h_{k}+h_{1} h_{2}^{\prime} h_{3} \ldots h_{k}+h_{1} h_{2} h_{3}^{\prime} \ldots h_{k}+\ldots+h_{1} h_{2} \ldots h_{k}^{\prime}
$$
где все слагаемые, кроме, возможно, первого, делятся на $h_{1}$. Значит, и первое слагаемое тоже должно на этот многочлен делиться. Поскольку $h_{1}$ неприводим, делиться на него должен один из сомножителей. Это невозможно, ибо $h_{1} \nmid h_{i}$ при $i \geqslant 2$ по нашему предположению, и $h_{1} \nmid h_{1}^{\prime}$ по пункту а)
в) Пусть многочлен $f$ неприводим. По пункту а), многочлены $f$ и $f^{\prime}$ взаимно просты. По пункту б), это значит, что квадрат непостоянного многочлена не может делить $f$; в частности, $f$ не может делиться на многочлен вида $(x-\alpha)^{2}$ при $\alpha \in \mathbb{C}$. Это и значит, что у $f$ нет кранных корней.
0.11. а) Пусть $\gamma \in \mathbb{F}[\alpha]$, и пусть $n$ - степень числа $\alpha$. Тогда при каждом $i=0,1, \ldots, n$ имеем $\gamma^{i}=g_{i}(\alpha)$, где $g_{i} \in \mathbb{F}[x]$ и $\operatorname{deg} g_{i}2$, то среди корней $\Phi_{n}(x)$ нет действительных, и эти корни разбиваются на пары комплексно сопряженных чисел модуля 1. Следовательно $\Phi_{n}(x)$ представляется в виде произведения многочленов вида $x^{2}+a x+1$, поэтому $\Phi_{n}(0)=1$.
б) Заметим, что $\Phi_{1}(1)=0$, а также 1 не является корнем никакого другого кругового многочлена. Если же $n$ делится на некоторое простое $p$, то по задаче 0.14 а $\Phi_{n}(1)=\Phi_{n p}(1)$. Таким образом, достаточно вычислить $\Phi_{n}(1)$ для бесквадратных $n$.
Если $n-$ простое, то $\Phi_{n}(1)=n$. Если же $n$ делится по крайней мере на два простых числа, одно из которых равно $p$, то $n / p$ - это натуральное число, большее 1 и по задаче 0.146
$$
\Phi_{n}(1)=\frac{\Phi_{n / p}(1)}{\Phi_{n / p}(1)}=1
$$
Суммируя вышенаписанное, получаем, что при $n>1$
$$
\Phi_{n}(1)= \begin{cases}p & , \text { если } n=p^{k} \\ 1 & , \text { если } n \neq p^{k}\end{cases}
$$
Замечание. Оба пункта также можно решить по индукции, используя разложение $x^{n}-1$ в произведение круговых многочленов.
0.17. а) Решение задачи 0.86 проходит дословно, за одним исключением - именно, надо по-другому обосновать, что $h_{1} \nmid h_{1}^{\prime}$. Дело в том, что степень $h_{1}^{\prime}$, конечно, меньше степени $h_{1}$, однако производная многочлена из $\mathbb{Z}_{p}[x]$ может оказаться нулевой, даже если его степень больше нуля (пример: $\left(x^{p}-1\right)^{\prime}=p x^{p-1}=0$ ). Так что нам надо показать, что $h_{1}^{\prime} \neq 0$, если $h_{1} \in \mathbb{Z}_{p}[x]$ - неприводимый многочлен.
Это делается так. Из бинома Ньютона следует, что для многочленов $a, b \in \mathbb{Z}_{p}[x]$ выполнено равенство $(a+b)^{p}=a^{p}+b^{p}$ (то же равенство, естественно, верно для элементов самого $\mathbb{Z}_{p}$ ). Применяя это свойство несколько раз, получаем, что
$$
\left(a_{0}+a_{1} x+a_{2} x^{2}+\ldots+a_{k} x^{k}\right)^{p}=a_{0}+a_{1} x^{p}+a_{2} x^{2 p}+\ldots+a_{k} x^{p k}
$$
Таким образом, если $h_{1}^{\prime}=0$ (что означает, что степени всех мономов в $h_{1}$ кратны $p$ ), то $h_{1}$ является $p$-й степенью другого многочлена и потому не может быть неприводимым. Это завершает решение.
б) Предположим, что многочлены $\Phi_{m}(x)$ и $\Phi_{n}(x)$ делятся на некоторый непостоянный многочлен $g(x)$ в $\mathbb{Z}_{p}[x]$. Тогда из разложения $x^{m n}-1$ в произведение круговых многочленов следует, что $g(x)^{2} \mid x^{m n}-1$ (в $\mathbb{Z}_{p}[x]$, естественно). Следовательно, по пункту а) многочлен $x^{m n}-1$ и его производная $m n x^{m n-1}$ не взаимно просты в $\mathbb{Z}_{p}[x]$, что, очевидно, неверно, ибо $p \nmid m n$ и $x \nmid x^{m n}-1$.
## 1 Делители значений многочлена и приложения
1.1. а) Достаточно, чтобы все коэффициенты, кроме свободного члена, делились на $m_{1} m_{2} \ldots m_{n}$.
б) Следует из того, что $f(n) \mid f(n) g(n)$.
в) Следует из того, что множество значений $f(g)$ в целых точках является подмножеством множества значений $f$ в целых точках.
1.2. Предположим противное: пусть $P(f)$ состоит из $k$ простых чисел, то есть $P(f)=\left\{p_{1}, p_{2}, \ldots, p_{k}\right\}$. Возьмем $a \in \mathbb{Z}$, для которого $f(a)=b \neq 0$. Зададим многочлен $g$ равенством $g(x)=\frac{f(a+b p x)}{b}$, где $p=p_{1} p_{2} \ldots p_{k}$. Как нетрудно видеть, $g-$ непостоянный многочлен с целыми коэффициентами, причем $P(g) \subset P(f)$. Свободный член $g(0)=1$, а все остальные коэффициенты многочлена $g$ делятся на $p$, следовательно в множестве $P(g)$ не может лежать ни одно из чисел $p_{1}, p_{2}, \ldots, p_{k}$. Но тогда множество $P(g)$ пусто - противоречие. (Заметим, что проведенное рассуждение обобщает доказательство Евклида бесконечности множества простых чисел.)
1.3. По теореме о линейном представлении НОД найдутся такие многочлены $u_{1}, u_{2}, \ldots, u_{k} \in \mathbb{Q}[x]$, что
$$
u_{1} f_{1}+u_{2} f_{2}+\ldots+u_{k} f_{k}=1
$$
Если домножить многочлены $u_{1}, \ldots, u_{k}$ на натуральное число $m$, равное HOK знаменателей всех ненулевых коэффициентов $u_{1}, \ldots, u_{k}$, получим многочлены
$v_{1}, v_{2}, \ldots, v_{k} \in \mathbb{Z}[x]$. Домножая равенство $u_{1} f_{1}+u_{2} f_{2}+\ldots+u_{k} f_{k}=1$ на $m$, получим $v_{1} f_{1}+v_{2} f_{2}+\ldots+v_{k} f_{k}=m$. Видим, что если при некотором $n$ числа $f_{1}(n), f_{2}(n)$, $\ldots, f_{k}(n)$ делятся на $s$, то $m$ также делится на $s$. Отсюда следует, что множество $D\left(f_{1}, f_{2}, \ldots, f_{k}\right)$ может содержать лишь делители числа $m$, и значит оно конечно.
1.4. $f(n)$ делится на $m_{i}$ для всех $n$ из некоторой арифметической прогрессии с разностью $m_{i}$. Так как НОД $\left(m_{1}, m_{2}\right)=1$, согласно $\mathrm{KTO}$, прогрессии с разностями $m_{1}$ и $m_{2}$ пересекаются. Тогда $n$ из их пересечения таково, что $m_{1} m_{2} \mid f(n)$.
1.5. а) -1 - квадратичный вычет по модулю $p$ вида $4 k+1$ и невычет по модулю $4 k+3$ (это следует, например, из критерия Эйлера). Это означает, что простые вида $4 k+1$ входят в $P(f)$, а простые вида $4 k+3$ - нет. Кроме того, очевидно, $2 \in P(f)$.
в) $f(1)=p$, поэтому $p \in P(f)$.
Пусть $f(n)$ делится на простое $q \neq p$. Так как многочлен $x^{p}-1$ делится на $f$ (в $\mathbb{Z}[x])$, то $n^{p} \equiv 1(\bmod q)$, откуда $\operatorname{ord}_{p}(n)$ равен 1 или $p$, поскольку он делит $p$.
В первом случае $n \equiv 1(\bmod q)$, откуда $f(n)=n^{p-1}+n^{p-2}+\ldots+n+1 \equiv p(\bmod q)$. Имеем $p \equiv 0(\bmod q)$, откуда $p=q$ - противоречие.
Во втором случае $q-1$ должно делиться на $\operatorname{ord}_{p}(n)=p$, т.е. $q \equiv 1(\bmod p)$.
Наоборот, пусть простое $q \equiv 1(\bmod p)$. Пусть $a$ - первообразный корень по модулю $q$, и пусть $n=a^{\frac{q-1}{p}}$. Тогда $n \not \equiv 1(\bmod q)$, но $n^{p} \equiv 1(\bmod q)$, следовательно $f(n)=\frac{n^{p}-1}{n-1}$ делится на $q$.
Таким образом, в $P(f)$ входят все простые вида $p k+1$ и $p$.
1.6. а) Если $f_{k}(x)=k x-1$ для некоторого натурального $k$, то $P\left(f_{k}\right)$ совпадает с множеством всех простых чисел, за исключением делителей $k$. Тогда ясно, что для $f(x)=(2 x-1)(3 x-1)$ множество $P(f)$ совпадает с множеством всех простых чисел.
б) Не следует. Для каждого простого $p$ и целых $t, s$ хотя бы одно из чисел $t, s$, $t s$ является квадратичным вычетом или нулевым вычетом по модулю $p$. Значит, для многочлена
$$
f_{t, s}(x)=\left(x^{2}-t\right)\left(x^{2}-s\right)\left(x^{2}-t s\right)
$$
выполнено $p \in P\left(f_{t, s}\right)$. Достаточно взять $t, s$ такие, что $t, s, t s$ отличны от точных квадратов.
1.7. а) Покажем индукцией по $\alpha$, что найдется целое $n_{\alpha}$ такое, что $p^{\alpha} \mid f\left(n_{\alpha}\right)$.
Для $\alpha=1$ утверждение следует из включения $p \in P(f)$. Пусть утверждение доказано для некоторого $\alpha \in \mathbb{N}$, и $f\left(n_{\alpha}\right)=p^{\alpha} m$ для некоторого $m \in \mathbb{Z}$. По лемме Гензеля
$$
f\left(n_{\alpha}+p^{\alpha} z\right) \equiv p^{\alpha}\left(m+z f^{\prime}\left(n_{\alpha}\right)\right) \quad\left(\bmod p^{\alpha+1}\right)
$$
Но так как $p \notin P\left(f, f^{\prime}\right)$, то $f^{\prime}\left(n_{\alpha}\right)$ не делится на $p$, поэтому достаточно выбрать $z=z_{0}$ так, чтобы $m+z_{0} f^{\prime}\left(n_{\alpha}\right)$ делилось на $p$, и положить $n_{\alpha+1}=n_{\alpha}+p^{\alpha} z_{0}$.
б) Возьмём $n_{\alpha}$ из пункта а). Тогда $f\left(n_{\alpha}+p^{\alpha}\right)$ делится на $p^{\alpha}$, а $f^{\prime}\left(n_{\alpha}\right)$ не делится на $p$, и из леммы Гензеля следует, что $v_{p}\left(f\left(n_{\alpha}+p^{\alpha}\right)-f\left(n_{\alpha}\right)\right)=\alpha$. Значит, либо $v_{p}\left(f\left(n_{\alpha}\right)\right)$, либо $v_{p}\left(f\left(n_{\alpha}+p^{\alpha}\right)\right)$ равен в точности $\alpha$.
1.8. Обратное неверно. Покажем, что существует такой многочлен даже без рациональных корней. Возьмем $f_{t, s}$ из решения задачи 1.6, где $t, s$ и $t s$ - не точные квадраты. Пусть нечетное простое $p$ не является делителем $t$ и $s$. Тогда $p \notin P\left(h, h^{\prime}\right)$ для каждого из многочленов $h(x)=x^{2}-a$, где $a$ равно $t, s$ или $t s$. Тогда, согласно задаче 1.7, $p^{\alpha} \in D\left(f_{t, s}\right)$ для сколь угодно больших $\alpha$.
Для конечного множества простых делителей $q$ чисел $t$ и $s$ возьмем $t^{\prime}, s^{\prime}$, не делящихся ни на одно из таких простых $q$ и такие, что $t^{\prime}, s^{\prime}, t^{\prime} s^{\prime}$ - не точные квадраты. Тогда $q^{\alpha} \in D\left(f_{t^{\prime}, s^{\prime}}\right)$ для сколь угодно больших $\alpha$.
Наконец, для $g(x)=x^{2}+x+2$, согласно задаче $1.7,2^{\alpha} \in D(g)$ для сколь угодно больших $\alpha$.
Согласно задаче 1.4, $f=f_{t, s} f_{t^{\prime}, s^{\prime}} g$ - искомый многочлен - такой, что $D(f)=\mathbb{N}$.
Замечание. На самом деле искомый многочлен можно найти просто в виде $f_{t, s}$ для подходящих $t$ и $s$.
1.9. б) Пусть $f=a h_{1}^{\alpha_{1}} \ldots h_{t}^{\alpha_{t}}$, где $a \in \mathbb{Z}$, а $h_{1}, \ldots h_{t}$ - попарно не пропорциональные примитивные неприводимые многочлены с целыми коэффициентами.
Так как НОД $\left(h_{1}, h_{1}^{\prime}\right)=1$, согласно задачам $1.2,1.3$ множество
$$
P_{1}\left(h_{1}\right) \backslash\left(P\left(h_{1}, h_{1}^{\prime}\right) \cup P\left(h_{1}, a\right) \cup P\left(h_{1}, h_{2}\right) \cup \ldots \cup P\left(h_{1}, h_{t}\right)\right)
$$
бесконечно; выберем в нем $p_{1}$ и, в согласии с задачей 1.7 , найдем $n_{1}$ такое, что $v_{p_{1}}\left(h_{1}\left(n_{1}\right)\right)=1$. Согласно определению $p_{1}$, для всех $m$ из прогрессии $n_{1}+k p_{1}^{2}$ выполнено $v_{p_{1}}\left(h_{1}(m)\right)=1$, в то время как $a, h_{2}(m), \ldots, h_{t}(m)$ не делятся на $p_{1}$. Значит, для всех $m$ из прогрессии $n_{1}+k p_{1}^{2}$ выполнено $v_{p_{1}}(f(m))=\alpha_{1}$.
Аналогично определим $p_{i}, i=2, \ldots, t$ и соответствующие прогрессии. По $\mathrm{KTO}$, все $t$ прогрессий пересекаются. Пусть $m$ принадлежит всем прогрессиям. Тогда с одной стороны, по условию, $f(m)=u^{s}$, для некоторых целых $u$ и $s>1$. С другой стороны, $v_{p_{i}}(f(m))=\alpha_{i}, i=1, \ldots, t$, поэтому все $\alpha_{i}$ кратны $s$, тем самым, $f=a r^{s}$, где $r \in \mathbb{Z}[x]$. Так как $f(m)=u^{s}=\operatorname{ar}(m)^{s}$, константа $a$ является точной $s$-й степенью. Окончательно, $f$ представим в искомом виде $f=q^{s}$.
Замечание. Условие задачи можно ослабить, потребовав, чтобы точными степенями были значения $f$ лишь на некоторой бесконечной арифметической прогрессии $A$. В решении при выборе $p_{i}$ тогда требуем еще, чтобы $p_{i}$ не было делителем разности этой прогрессии. Тогда наши прогрессии с разностями $p_{i}^{2}$ пересекутся с прогрессией $A$.
1.10. Задача непосредственно следует из следующей (при помощи 1.5а). Но мы приведём и непосредственное решение.
Мы докажем существование одного простого $p=4 k+1 \in P(f)$; доказательство того, что таких простых бесконечно много, можно теперь получить так же, как и в задаче 1.2 .
Пусть $\operatorname{deg} f=n$. Рассмотрим на минутку многочлен $f(x+i)$, где $i-$ мнимая единица. Он представляется в виде $f(x+i)=P_{1}(x)+i P_{2}(x)$, где $P_{1}(x), P_{2}(x) \in \mathbb{Z}[x]$, причём $\operatorname{deg} P_{1}=n, \operatorname{deg} P_{2}=n-1 \geqslant 0$. Значит, существует такое $k \in \mathbb{Z}$, что числа $a=P_{1}(k)$ и $b=P_{2}(k)$ ненулевые, причём $\left|P_{1}(k)\right|>\left|P_{2}(k)\right|$. Заменив $P(x)$ на $P(x+k)$, мы можем считать, что $f(i)=a+b i$, где $|a|>|b|>0$; тогда $f(-i)=a-b i$.
Пусть теперь $d=\operatorname{НОД~}(a, b), a=a^{\prime} d, b=b^{\prime} d$; тогда $\left|a^{\prime}\right| \geqslant|a| /|b|>1$. Число $a^{\prime 2}+b^{\prime 2}$ больше 2 и не может делиться на 4; значит, оно имеет нечётный простой делитель $p$. Поскольку числа $a^{\prime}$ и $b^{\prime}$ взаимно просты, они не могут делиться на $p$, и оно имеет вид $p=4 k+1$.
Положим $Q(x)=f(x)-a-b x$. Значения $Q(x)$ в точках $\pm i$ равны нулю, поэтому он делится на $(x-i)(x+i)=x^{2}+1$. Поскольку $x^{2}+1$ приведённый, получаем, что $f(x)=a+b x+\left(x^{2}+1\right) R(x)$ для некоторого $R(x) \in \mathbb{Z}[x]$.
Пусть теперь $n-$ такое число, что $n b^{\prime} \equiv-a^{\prime}(\bmod p)$. Тогда $a+b n=d\left(a^{\prime}+b^{\prime} n\right)$ делится на $p$ и $b^{\prime 2}\left(n^{2}+1\right)=\left(b^{\prime} n\right)^{2}+b^{\prime 2} \equiv a^{\prime 2}+b^{\prime 2} \equiv 0(\bmod p)$; поскольку $b^{\prime}$ не делится на $p$, получаем, что $p \mid n^{2}+1$. Значит, и число $f(n)=(a+b n)+\left(n^{2}+1\right) R(n)$ делится на $p$.
### 1.11. Мы докажем сразу пункт б).
Ясно, что достаточно ограничиться случаем, когда все многочлены $f_{i}$ неприводимы над $\mathbb{Z}$ (если они приводимы, достаточно применить факт для некоторых их неприводимых делителей).
Пусть $\alpha_{i}$ - корень многочлена $f_{i}$, при $i=1,2, \ldots, k$. По теореме о примитивном элементе, существует алгебраическое число $\gamma$ такое, что $\mathbb{Q}\left[\alpha_{1}, \alpha_{2}, \ldots, \alpha_{k}\right]=\mathbb{Q}[\gamma]$. В частности, это означает, что существуют такие многочлены $A_{i} \in \mathbb{Q}[x]$, что $\alpha_{i}=A_{i}(\gamma)$.
Пусть $g$ - минимальный многочлен числа $\gamma$. Заметим, что $f_{i}\left(A_{i}(\gamma)\right)=0$, так что многочлены $f_{i}\left(A_{i}(x)\right)$ делятся на $g$ при всех $i=1,2, \ldots, k$, то есть
$$
f_{i}\left(A_{i}(x)\right)=g(x) h_{i}(x)
$$
Коэффициенты многочленов из равенств выше могут быть нецелыми (но они рациональны); пусть $N$ - наибольшее простое число, на которое делится какой-нибудь из знаменателей этих коэффициентов.
По задаче 1.2, существует бесконечно много простых чисел, делящих значения $g$ в целых точках. Пусть $p>N$ - такое простое число, и пусть $g(n)$ - соответствующее значение. Тогда числители рациональных чисел $f_{i}\left(A_{i}(n)\right)=g(n) h_{i}(n)$ делятся на $p$, поскольку знаменатели чисел $P_{i}(n)$ и $Q_{i}(n)$ на него не делятся.
Единственная оставшаяся проблема заключается в том, что числа $A_{i}(n)$ - рациональные, а не целые. Однако их знаменатели также не делятся на $p$. Значит, если $A_{i}(n)=k_{i} / \ell_{i}$ для взаимно простых целых $k_{i}$ и $\ell_{i}$, то можно выбрать такое целое $m_{i}$, что $k_{i} \equiv \ell_{i} m_{i}(\bmod p)$. Тогда $f_{i}\left(m_{i}\right) \equiv f_{i}\left(k_{i} / \ell_{i}\right) \equiv 0(\bmod p)$. Значит, $p \in P\left(f_{i}\right)$ при всех $i=1,2, \ldots, k$.
1.12. Если $x \mid f$, то $D(f)=\mathbb{N}$, и утверждение очевидно. В противном случае пусть $b \neq 0$ - свободный член; заменяя $f$ на $f(b x) / b \in \mathbb{Z}[x]$, можем считать, что $b=1$. Без ограничения общности $f$ считаем неприводимым, а его старший коэффициент положительным. Как и в задаче 1.10, достаточно показать, что найдется хотя бы одно простое $p \equiv 3(\bmod 4)$, такое что $p \in P(f)$.
a) Предположим, что $\operatorname{deg} f$ нечетна. Рассмотрим большое отрицательно целое число $-n$, делящееся на 4 . Тогда $f(-n)<0$ и $f(-n) \equiv f(0) \equiv 1(\bmod 4)$, поэтому $f(-n)=-(4 k+3)$ для некоторого натурального $k$. Тогда в разложении $4 k+3$ на простые множители должно содержаться искомое простое $p \equiv 3(\bmod 4)$.
б) Если $D(f)$ содержит натуральное число, сравнимое с 3 по модулю 4, можно применить те же рассуждения, что и выше. В противном случае, поскольку $f$ - неприводимый, у него нет кратных корней. Так как он имеет вещественный корень, найдется интервал $(\alpha, \beta)$, на котором $f$ принимает только отрицательные значения.
Теперь выберем рациональное число $\frac{a}{b} \in(\alpha, \beta)$ (где $a$ и $b$ взаимно простые), такое что $b$ не имеет простых делителей вида $4 k+3$, а $4 \mid$ a (можно положить, например, $b=5^{N}$ для подходящего $\left.N\right)$. Вспомним, что $f(a) \equiv 1(\bmod 4)$. Более того, $f(a)>0$, иначе $-f(a)$ - натуральное число вида $4 k+3$, и работает рассуждение из пункта а). Тогда $f\left(\frac{a}{b}\right)=-\frac{c}{d}$, где $d=b^{n} \equiv 1(\bmod 4)$, и натуральное $c$ удовлетворяет условию
$$
c=-b^{n} f\left(\frac{a}{b}\right) \equiv-f(a) \equiv 3 \quad(\bmod 4)
$$
Поэтому найдется простое $p \mid c$, такое что $p \equiv 3(\bmod 4)$. По нашему предположению $p \nmid b$; значит, найдется $a^{\prime} \in \mathbb{Z}$, такое что $a \equiv b a^{\prime}(\bmod p)$. Тогда имеем $b^{n} f\left(a^{\prime}\right) \equiv b^{n} f\left(\frac{a}{b}\right)=-c \equiv 0(\bmod p)$, и искомое $p$ найдено.
1.13. Решение 1. Предположим, противное, тогда для некоторого целого $a$ многочлены $\Phi_{m}(x)$ и $\Phi_{n}(x)$ делятся на $x-a$ в $\mathbb{Z}_{p}[x]$, что противоречит задаче 0.176.
1.13. Решение 2. Предположим, что $\Phi_{m}(k)$ и $\Phi_{n}(k)$ делятся на $p$. Тогда $k^{m}-1$ и $k^{n}-1$ тоже делятся на $p$. Пусть $mv_{p}\left(k^{m}-1\right)$. Но так как $n$ не делится на $p$, это противоречит LTE-лемме для $a=k^{m}$ и $b=1$.
1.14. Следует напрямую из задачи 0.14 , поскольку $2 \in P(f)$ если и только если хотя бы одно из чисел $f(0), f(1)$ четно.
1.15. а) Пусть $\Phi_{n}(m)$ делится на простое число $p$. Тогда $m^{n} \equiv 1(\bmod p)$. Пусть $d=\operatorname{ord}_{p}(m)$, тогда $d \mid n$ и $d \mid p-1$.
Если $d=n$, то $n \mid p-1$ и $p \equiv 1(\bmod n)$.
Иначе $dv_{p}\left(m^{d}-1\right)
$$
С другой стороны, применив LTE-лемму для $a=m^{d}$ и $b=1$, имеем
$$
v_{p}\left(m^{n}-1\right)-v_{p}\left(m^{d}-1\right)=v_{p}(n / d)
$$
Значит, $p \mid(n / d)$, откуда $p \mid n$.
б) Следует из а) и задачи 1.2 .
1.16. а) Пусть $k=v_{p}\left(m^{d}-1\right)$. Тогда для любого целого $a$ имеем:
- если $d \nmid a$, то $p \nmid m^{a}-1$ (вытекает из свойств порядка по модулю $p$ );
- если $d \mid a$, то $v_{p}\left(m^{a}-1\right)=k+v_{p}(a / d)$ (из LTE-леммы).
Значит, $n_{i}=d p^{i}$ - это наименьшее число $n$, для которого $p^{k+i} \mid m^{n}-1$.
Выведем из разложения $x^{n}-1$ на круговые многочлены формулу
$$
\Phi_{n}=\frac{x^{n}-1}{\operatorname{HOK}\left\{x^{d}-1: d \mid n, d0
$$
Перепишем его в виде
$$
P(x)=Q(x)^{2}+r(x)=\left(b_{k} x^{k}+b_{k-1} x^{k-1}+\ldots+b_{0}\right)^{2}+r(x)
$$
где $\operatorname{deg} r(x) \leqslant k-1$. Чтобы понять, что это возможно, сравним коэффициенты при $x^{t+k}$ для $t=0,1, \ldots, k-1$ у двух разных форм записи: $a_{k+t}=2 b_{k} b_{t}+\sum_{i=1}^{k-1-t} b_{t+i} b_{k-1}$. Легко видеть, что по этим соотношениям последовательно и однозначно восстанавливаются $b_{k-1}, b_{k-2}, \ldots, b_{1}, b_{0}$. Также видно, что все $b_{i}$ являются рациональными числами; обозначим через $M$ HOK их знаменателей.
По условию, для всех достаточно больших натуральных $n$ мы имеем $P(n)=y_{n}^{2}$, где $y_{n} \in \mathbb{Z}$. После домножения на $M^{2}$ мы получим соотношение $\left(M y_{n}\right)^{2}=Q_{1}(n)^{2}+M^{2} r(n)$, где $Q_{1}(x)=M Q(x) \in \mathbb{Z}[x]$. Если у $r(x)$ положительный старший коэффициент, то при достаточно больших натуральных $n$ верно неравенство $\left(M y_{n}\right)^{2}>Q_{1}(n)^{2}$ и, следовательно, $\left(M y_{n}\right)^{2} \geqslant\left(1+Q_{1}(n)\right)^{2}$. Аналогично, если старший коэффициент многочлена $r(x)$ отрицательный, то $\left(M y_{n}\right)^{2} \leqslant\left(-1+Q_{1}(n)\right)^{2}$. В обоих случаях выполняется соотношение $2\left|Q_{1}(n)\right| \leqslant\left|M^{2} r(n)-1\right|$ при всех достаточно больших натуральных $n$, что, очевидно, неверно, так как $\operatorname{deg} Q_{1}>\operatorname{deg} r$.
Следовательно, многочлен $r(x)$ тождественно равен нулю, и $P(x)=Q(x)^{2}$. Осталось доказать, что многочлен $Q(x)$ имеет целые коэффициенты, а это следует из задачи 2.1.
2.3. В условии этой задачи у нас нет дополнительных ограничений, как в задаче 2.2 , но мы применим следующий красивый трюк. Очевидно, что существует целое $\ell$ такое, что многочлены $P(x)$ и $P(x+\ell)$ не имеют общих комплексных корней (и потому взаимно просты). Многочлен $H(x)=P(x) \cdot P(x+\ell)$ гарантированно имеет чётную степень и старший коэффициент, являющийся точным квадратом, а также точными квадратами являются и его значения во всех достаточно больших целых точках.
По задаче 2.2 существует многочлен $P_{1}(x) \in \mathbb{Z}[x]$ такой, что $P(x) \cdot P(x+\ell)=P_{1}(x)^{2}$. Рассмотрим его каноническое разложение в $\mathbb{Z}[x]$ :
$$
P_{1}=u h_{1} h_{2} \ldots h_{k}
$$
где $u \in \mathbb{Z}$, а $h_{i}$ - примитивные неприводимые многочлены. Из взаимной простоты многочленов $P(x)$ и $P(x+\ell)$ с учётом леммы Гаусса вытекает, что, скажем, $P(x)$ есть произведение некоторого целого $u_{1}$ и квадратов некоторых из многочленов $h_{i}$. При этом, поскольку у $P(x)$ есть ненулевые значения в целых точках, являющиеся квадратами, число $u_{1}$ - тоже квадрат. Следовательно, $P(x)$ является квадратом многочлена с целыми коэффициентами, что и требовалось доказать.
Отметим, что эту задачу можно решить и по-другому, повторяя решение задачи 1.9 .
2.4. Заметим, что каждый интервал $\left(r_{i}, r_{i+1}\right)$ содержит хотя бы один корень производной $R^{\prime}(x)=Q^{\prime}(x)$; мы получили $l-1$ различных корней, так что мы перечислили все корни этой производной, и на каждом интервале такой корень единственен. Аналогичное утверждение верно для многочлена $S(x)$. Тогда для любого $i4$, то степень многочленов $Q(x) \pm b$ больше 2 , и по теореме Виета у
этих многочленов одинакова сумма корней и одинакова сумма попарных произведений корней. Значит, совпадают и суммы их квадратов, а именно:
$$
0^{2}+3^{2}+\ldots+(k-4)^{2}+(k-1)^{2}=1^{2}+2^{2}+\ldots+(k-3)^{2}+(k-2)^{2}
$$
но это равенство не может выполняться, так как для любого $t$
$$
(t-4)^{2}+(t-1)^{2}=2 t^{2}-10 t+17>2 t^{2}-10 t+13=(t-3)^{2}+(t-2)^{2}
$$
Следовательно, $k=4$, при этом $Q(x)-b=x(x+3), Q(x)+b=(x+1)(x+2)$. Значит, $b=1$ и $c=b^{2}=1$. Пара чисел $(k, c)=(4,1)$ является единственным решением задачи.
2.6. Структура многочлена $P_{k}(x)$ такова, что его производная $P_{k}^{\prime}(x)$ имеет ровно $k-1$ корень $r_{1}, \ldots, r_{k-1}$, причём эти корни можно упорядочить так, чтобы выполнялось неравенство
$$
-(k-1)\left|P_{k}\left(r_{i}\right)\right|$. По симметрии для $i \geqslant \frac{k+2}{2}$ выполняется обратное неравенство $\left|P_{k}\left(r_{i-1}\right)\right|<\left|P_{k}\left(r_{i}\right)\right|$.
Таким образом, для каждого комплексного $a$ уравнение $P_{k}(x)=a$ имеет не более двух корней из множества $\left\{r_{1}, \ldots, r_{m-1}\right\}$ корней $P_{k}^{\prime}(x)$. Дальше, если $k$ нечётно, то $P_{k}\left(r_{i}\right)=-P_{k}\left(r_{k-i}\right)$, следовательно количество этих корней в этом случае не больше одного. Из этого и следует требуемое, поскольку все корни $P_{k}^{\prime}$ простые (т.е. некратные).
2.7. Заметим, что $P_{k}^{\prime}(x)=R^{\prime}(Q(x)) \cdot Q^{\prime}(x)$.
Пусть $r$ - некий комплексный корень многочлена $R^{\prime}$. Тогда $r$ - общий корень многочленов $R^{\prime}(x)$ и $R(x)-R(r)$, то есть $x-r \mid \operatorname{НОД}\left(R^{\prime}(x), R(x)-R(r)\right)$. Подставляя $Q(x)$ вместо $x$, получаем $Q(x)-r \mid$ НОД $\left(R^{\prime}(Q(x)), R(Q(x))-R(r)\right)$. Наконец, поскольку $R^{\prime}(Q(x)) \mid P_{k}^{\prime}(x)$, отсюда вытекает, что $Q(x)-r \mid \operatorname{HOД}\left(P_{k}^{\prime}(x), P_{k}(x)-R(r)\right)$.
Теперь из задачи 2.6 следует, что степень многочлена $Q(x)-r$, как и многочлена $Q(x)$ должна быть ровно 2 , причём $k$ должно быть чётно, что и требовалось доказать.
2.8. Решение 1. По задаче 1.9
$$
P_{k}(x)+c=x(x+1) \ldots(x+k-1)+c=Q(x)^{s}
$$
для некоторого натурального числа $s \geqslant 2$ и некоторого многочлена $Q(x)$ с комплексными коэффициентами. Пусть старший коэффициент $Q(x)$ равен $a$, тогда, поскольку
$a^{s}=1$, то можно заменить $Q(x)$ на $\frac{1}{a} Q(x)$, поэтому можно считать, что многочлен $Q(x)$ является приведённым, а тогда по задаче 2.1 многочлен $Q$ имеет целые коэффициенты. Не умаляя общности, можно считать, что $Q(x)$ не является степенью другого многочлена, в противном случае мы можем заменить $s$ на другое подходящее натуральное число.
Заметим, что
$$
P_{k}^{\prime}(x)=\left(P_{k}(x)+c\right)^{\prime}=\left(Q(x)^{s}\right)^{\prime}=Q^{\prime}(x) \cdot Q(x)^{s-1}
$$
Значит, $Q^{s-1}$ делит НОД $\left(P_{k}+c, P_{k}^{\prime}\right)$. Но по задаче 2.7, степень этого НОД не превосходит 2. Поэтому возможны лишь следующие случаи: (1) $s=3, \operatorname{deg} Q=1 ;(2) s=2$, $\operatorname{deg} Q=1 ;$ и (3) $s=2, \operatorname{deg} Q=2$.
В случае (1) получаем, что $k=3$ нечётно, но тогда степень НОД даже не больше 1 , так что этот случай невозможен.
В случае (2) имеем $x(x+1)+c=(x+a)^{2}$, откуда $a=1 / 2$, то есть $Q \notin \mathbb{Z}[x]$.
В случае (3) имеем $x(x+1)(x+2)(x+3)+c=\left(x^{2}+a x+b\right)^{2}$; сравнивая коэффициенты при $x^{3}$ и $x^{2}$ в левой и правой частях, получаем $a=3, b=1$, а тогда $c=1$, и мы приходим к единственному ответу $(k, c)=(4,1)$.
2.8. Решение 2. Как мы уже отметили в первом решении, существуют натуральное $s>1$ и многочлен $Q(x) \in \mathbb{Z}[x]$ такие, что
$$
x(x+1) \ldots(x+k-1)+c=Q(x)^{s}
$$
Полагая $x=0$ мы находим, что $c=Q(0)^{s}=b^{s}$. Таким образом, верно равенство
$$
x(x+1) \ldots(x+k-1)=Q(x)^{s}-b^{s}
$$
Следовательно, $Q(-j)^{s}=b^{s}$ для всех $j=0,1, \ldots, k-1$. Если $s$ нечётно, то многочлен $Q(x)-b$ имеет минимум $k$ различных корней $0,-1, \ldots,-k+1$ и имеет степень не меньше $k$. Но тогда степень многочлена $Q(x)^{s}$ не меньше $s k$, что больше степени многочлена $P_{k}(x)=x(x+1) \ldots(x+k-1)$, а это невозможно.
Предположим, что $s$ чётно, в таком случае достаточно рассмотреть случай $s=2$, а он разобран в задаче 2.5 .
2.9. Заметим сразу, что
$$
f(x)=\frac{x^{n+1}-1}{x-1}
$$
Пусть $\alpha$ - произвольный комплексный корень многочлена $g^{\prime}(x)$. Как и в решении

$$
\operatorname{deg} \text { НОД }\left(f^{\prime}(x), f(x)-g(\alpha)\right) \geqslant 2
$$
Следовательно, существует как минимум два (возможно, совпадающие) корня $r, s$ многочлена
$$
f^{\prime}(x)=\frac{n x^{n+1}-(n+1) x^{n}+1}{(x-1)^{2}}
$$
такие, что $f(r)=f(s)$. Легко заметить, что 1 не является корнем $f^{\prime}$, и что у $f^{\prime}$ нет кратных корней, так что $r \neq s$.
Из формулы для $f^{\prime}$ следует, что
$$
n\left(s^{n+1}-1\right)=(n+1)\left(s^{n}-1\right) \quad \text { и } \quad n\left(r^{n+1}-1\right)=(n+1)\left(r^{n}-1\right)
$$
Далее, из равенства $f(r)=f(s)$ следует, что
$$
\frac{s^{n+1}-1}{s-1}=\frac{r^{n+1}-1}{r-1}
$$
Из соотношений выше следует тогда, что и
$$
\frac{s^{n}-1}{s-1}=\frac{r^{n}-1}{r-1}
$$
Вычитая это равенство из предыдущего, получаем $s^{n}=r^{n}$, а тогда и $s-1=r-1$, то есть $s=r$, что невозможно. Поэтому натуральных $n$, удовлетворяющих условию задачи, не существует.
## 3 Разное
3.1. а) Разделим $f$ на $g$ с остатком: $f=q g+r$, где $q, r \in \mathbb{Z}[x]$, $\operatorname{deg} r<\operatorname{deg} g$. Тогда найдется $N$ такое, что для всех $n>N$ выполнено $|r(n)|<|g(n)|$, и если $n>N$ таково, что $g(n) \mid f(n)$, то $r(n)=0$. Получаем, что $r(n)=0$ для бесконечно многих значений $n$, поэтому $r$ - нулевой многочлен.
б) После деления $f$ на $g$ с остатком в $\mathbb{Q}[x]$ представим неполное частное в виде $q / m$, где $m \in \mathbb{N}$ и $g \in \mathbb{Z}[x]$, то есть $f=\frac{q g}{m}+r$, где $r \in \mathbb{Q}[x], \operatorname{deg} r<\operatorname{deg} g$. Тогда найдется $N$ такое, что для всех $n>N$ выполнено $|r(n)|<|g(n)| / m$, и если $n>N$ таково, что $g(n) \mid f(n)$, то $r(n)=0$. Получаем, что $r(n)=0$ для бесконечно многих значений $n$, поэтому $r$ - нулевой многочлен.
Далее, пусть для некоторого $n \in \mathbb{N} g(n) \mid f(n)$, т.е. $m \mid q(n)$. Но тогда $m \mid q(n+m t)$ для любого целого $t$, и значит, $g(n+m t) \mid f(n+m t)$.
3.2. Предположим противное, тогда для каждого $i=1,2, \ldots, m$ найдется целое $k_{i}$ такое, что $f\left(k_{i}\right)$ не делится на $p_{i}^{\alpha_{i}}$, где $p_{i}$ - простое, а $p_{i}^{\alpha_{i}}$ - делитель $a_{i}$. Если $p_{i}=p_{j}$ и $\alpha_{i} \leqslant \alpha_{j}$, то для индекса $j$ можно переопределить $\alpha_{j}=\alpha_{i}$ и $k_{j}=k_{i}$. После таких переопределений считаем, что равным $p_{i}$ соответствуют одни и те же $\alpha_{i}$ и $k_{i}$. Согласно $\mathrm{KTO}$, найдется $k$ в пересечении прогрессий вида $k_{i}+p_{i}^{\alpha_{i}} t$. Тогда $f(k)$ не делится ни на одно из $p_{i}^{\alpha_{i}}$, и следовательно, ни на одно из $a_{i}$. Противоречие.
Каждый многочлен $f \in \mathbb{Z}[x]$ определяет функцию $\mathbb{Z}_{n} \rightarrow \mathbb{Z}_{n}$, поскольку вычет $f(m)$ по модулю $n$ зависит только от вычета $m$ по модулю $n$.
3.3. а) Разделим данный многочлен $f$ на $x(x-1) \ldots(x-(n-1))$ с остатком: $f=x(x-1) \ldots(x-(n-1)) g+r$, где где $q, r \in \mathbb{Z}[x], \operatorname{deg} r3$ не удовлетворяет (*).
Если $p \equiv 1(\bmod 4)$, возьмем $a=0$ и $b$ такое, что $b^{2} \equiv-1(\bmod p)$. Тогда $a^{2}+a b+b^{2}+1$ делится на $p$, а $a-b$ не делится.
Теперь пусть $p \equiv 3(\bmod 4)$, так что -1 - не квадратичный вычет по модулю $n$. Достаточно найти такой вычет $x$, что $x^{2}+x+1$ - квадратичный невычет. Действительно, для такого $x$, вычет $-\left(x^{2}+x+1\right)^{-1} \bmod p$ будет квадратичным, поэтому найдется ненулевой вычет $a$, такой что
$$
-1 \equiv a^{2}\left(x^{2}+x+1\right)=(a x)^{2}+a \cdot a x+a^{2} \quad(\bmod p)
$$
тогда $(a x)^{2}+a \cdot a x+a^{2}+1$ делится на $p$, в то время как $a x-a$ не делится, за исключением случая $x \equiv 1(\bmod p)$. Но если $x \equiv 1(\bmod p)$, то заменим его на $x \equiv-2(\bmod p)$, не изменив вычет $x^{2}+x+1$.
Предположим, что напротив, все вычеты вида $x^{2}+x+1$ - квадратичные или нулевые, тогда то же верно для $4\left(x^{2}+x+1\right)=4(2 x+1)^{2}+3$. Так как $2(2 x+1)$ пробегает все вычеты, получаем следующее условие: если вычет $y$ квадратичный, то $y+3$ - тоже. Но, итерируя $y \mapsto y+3$, мы можем получить любой вычет, в том числе и не квадратичный. Противоречие.
3.7. Ясно, что $p>2$. Для решения нам понадобится формула
$$
0^{k}+1^{k}+\ldots+(p-1)^{k} \equiv\left\{\begin{array}{ll}
0, & 1 \leqslant k \leqslant p-2 ; \\
-1, & k=p-1
\end{array} \quad(\bmod p)\right.
$$
При $k=p-1$ это следует из малой теоремы Ферма. При меньших $k$ выберем первообразный корень $t$ по модулю $p$; тогда сумма в левой части (по модулю $p$ ) останется той же самой при домножении на $t^{k}$. Поскольку $t^{k} \not \equiv 1(\bmod p)$, отсюда следует, что сумма кратна $p$.
а) Предположим противное, тогда $f(0)+f(1)+\ldots+f(p-1) \equiv 0+1+\ldots+(p-1) \equiv 0$ $(\bmod p)$. С другой стороны, из формул выше получаем $f(0)+f(1)+\ldots+f(p-1) \equiv-1$ $(\bmod p)$. Противоречие.
б) Пусть $p-1=d k$. Применим соображения из пункта а) для многочлена $g=f^{k}$. Для него
$$
\begin{aligned}
-1 \equiv g(0)+g(1)+\ldots+g(p-1) \equiv(f(0))^{k}+(f(1))^{k}+\ldots+(f(p-1))^{k} & \equiv \\
& \equiv 0^{k}+1^{k}+\ldots+(p-1)^{k} \equiv 0 \quad(\bmod p)
\end{aligned}
$$
3.8. Так как $f$ задает перестановку $\sigma$ на множестве $\mathbb{Z}_{n}$, то
$$
g=f(f(f(\ldots)))(k \text { итераций })
$$
задает перестановку $\sigma^{k}$. Поэтому достаточно взять $k$ такое, что $\sigma^{k}$ является тождественной перестановкой (т.е. чтобы $k$ делилось на все длины независимых циклов, на которые распадается $\sigma$, подойдет, например, $k=n!$ ) и положить
$$
g=f(f(f(\ldots)))(k-1 \text { итераций }) .
$$
## Список литературы.
1. В. Прасолов. Многочлены. М: МЦНМО, 2003.
2. П.Кожевников, В.Сендеров. Делители значений многочлена. - Сборник «Задачи Санкт-Петербургской олимпиады по математике 2010 года». - СПб.: Невский Диалект, 2010, с. 121-129.
3. Д. Клюев, И. Богданов, А. Канель-Белов. Решение задачи 18.7. Математическое просвещение, серия 3,2015 , выпуск 19, с. 264-266
4. Н. Сафаеи, Об одной задаче В.А. Сендерова, Квант, 2021, № 10, с. 2-11
5. Polynomial Problems from the Awesomemath Summer Program (Xyz) by Titu Andreescu, Navid Safaei, Alessandro Ventullo.
# Polynomials in number theory
Project is prepared and presented by:
Ilya Bogdanov, Boris Frenkin, Pavel Kozhevnikov, Pavel Kozlov, Andrey Ryabichev, Navid Safaei
#### Abstract
In this project you can get acquainted with some remarkable problems and ideas arising at the intersection of number theory and theory of polynomials. This project consists of several sections each of which is interesting by itself.
Section 0 contains some technical facts which will be quite helpful in other sections.
In Section 1, we investigate the sets of divisors of values of a polynomial with integer coefficients. We start with investigating some concrete polynomials; after that, we suggest to prove some general facts which appear to be quite curious and beautiful. In particular, it appears that for each non-constant polynomial with integer coefficients there are infinitely many primes of the form $p=4 k+1$ dividing some values of the polynomial. On the other hand, under the extra restriction that the polynomial has at least one real root, the same holds for primes of the form $p=4 k+3$.
Section 2 is mainly devoted to one difficult number-theoretical problem describing all situations when a product of $n$ consecutive integers increased by a constant is a perfect power. While solving this problem, you will learn both analytical and number-theoretical methods which are quite useful for many other problems.
In Section 3 we collect miscellaneous problems which are interesting by themselves, but they are also related to the problems in other Sections, by means of methods needed for their solutions.
## 0 Auxiliary facts
### 0.1 Basic concepts in number theory
Let us list some number-theoretical facts which can be helpful in solving problems.
- Division with remainder and Euclid's algorithm.
- Linear representation of the g.c.d.
- The fundamental theorem of arithmetics (i.e., existence and uniqueness of prime factorization).
Corollary: existence of the canonical representation of integer $n>1$; namely $n=p_{1}^{\alpha_{1}} \cdots p_{k}^{\alpha_{k}}$, where $p_{1}, \ldots, p_{k}$ are distinct primes, and $\alpha_{1}, \ldots, \alpha_{k}$ are exponents.
The exponent of $p$ in the decomposition of $n$ further denoted by $v_{p}(n)$.
- Chinese Remainers Theorem.
One of its reformulations reads as follows: consider $n$ (both-sides) infinite arithmetic progressions consisting of integers, such that the steps $d_{1}, \ldots, d_{n}$ are pairwise coprime. Then the intersection of those progressions is exactly an arithmetic progression with step $d_{1} \cdots d_{n}$.
- Congruences and residues modulo $n$.
Congruence $a \equiv b(\bmod n)$ (which reads as ' $a$ and $b$ are congruent modulo $n$ ') for integers $a, b$ and a positive integer $n$ means that $a$ and $b$ have equal remainders while diveded by $n$, or, equivalently, that $n \mid(a-b)$.
A residue modulo $n$ is an equivalence class with respect to congruence modulo $n$ (each residue is an arithmetic progression with step $n$ ); but, a representative of this class is also called a residue. All $n$ residues form a complete residue system, which is denoted by $\mathbb{Z}_{n}$. If we fix a representative for eash residue, we may put $\mathbb{Z}_{n}=\{0,1, \ldots, n-1\}$. On $\mathbb{Z}_{n}$, the operations of addition and multiplication are naturally defined. The inverse for a residue $a$ is a residue $a^{-1}$ such that $a a^{-1}$ is equal to 1 in $\mathbb{Z}_{n}$, i.e., $a a^{-1} \equiv 1(\bmod n)$. The residue $a$ is invertible (i.e. having an inverse) iff $\operatorname{gcd}(a, n)=1$.
- Euler's and Fermat's theorems.
Euler's theorem states that for coprime $a \in \mathbb{Z}$ and $n \in \mathbb{N}$ we have $a^{\varphi(n)} \equiv 1(\bmod n)$, where $\varphi(n)$ is Euler's totient function, i.e., the number of numbers from $\{1, \ldots, n\}$ coprime to $n$.
Fermat's theorem is a particular case of Euler's theorem for a prime $n=p$ (in which case. $\varphi(p)=p-1$ ).
- The order of an integer $a$ modulo $n$ is the least positive integer $k$ such that $a^{k} \equiv 1$ $(\bmod n)$. (It is clear that the order is defined only if $\operatorname{gcd}(a, n)=1$.) Let us denote the order of $a$ modulo $n$ by $\operatorname{ord}_{n}(a)$.
We note the following property of the order: $a^{m} \equiv 1(\bmod n)$ iff $\operatorname{ord}_{n}(a) \mid m$. In particular, by Euler's theorem, it follows that $\operatorname{ord}_{n}(a) \mid \varphi(n)$.
If $\operatorname{ord}_{n}(a)=\varphi(n)$, then $1, a, a^{2}, \ldots, a^{\varphi(n)-1}$ are all the residues which are coprime to $n$. In this case $a$ is called a primitive root modulo $n$. A primitive root exists iff $n$ equals 2,4 , $p^{\alpha}$, or $2 p^{\alpha}$ for a prime $p>2$ and a positive integer $\alpha$.
- Let us recall the LTE-lemma (Lifting the Exponent Lemma).
Let $p$ be an odd prime, and $a, b$ be distinct integers not divisible by $p$. If $p \mid a-b$, then
$$
v_{p}\left(a^{n}-b^{n}\right)=v_{p}(a-b)+v_{p}(n)
$$
For $p=2$ the same equality holds whenever $4 \mid a-b$.
- A non-zero residue $a$ modulo an odd prime $p$ is called quadratic, if the congruence $x^{2} \equiv a(\bmod p)$ has a non-empty solution. The number of (non-zero) quadratic residues is $\frac{p-1}{2}$ - the same as the number of non-quadratic non-zero residues.
Euler's criterion states that a non-zero residue $a$ is quadratic iff $a^{\frac{p-1}{2}} \equiv 1(\bmod p)$.
### 0.2 Theory of divisibility in $\mathbb{Q}[x]$
Let us denote by $\mathbb{Q}[x]$ the set of polynomials with rational coefficients, and by $\mathbb{Z}[x]$ the set of polynomials with integer coefficients. We say that polynomials of degree not less than 1 are non-constant. A non-constant polynomial $f \in \mathbb{Q}[x]$ of degree $d$ is called reducible over $\mathbb{Q}$ (or, reducible in $\mathbb{Q}[x]$ ) if it can be represented as a product of two polynomial of degree less
than $d$ with retional coefficients. Each non-constant polynomial in $\mathbb{Q}[x]$ is eqither reducible, or irredusible over $\mathbb{Q}$. Similarly to the case of integers, one can divide a polynomial in $\mathbb{Q}[x]$ by another nonzero polynomial in $\mathbb{Q}[x]$ with remainder (e.g., via long division). In $\mathbb{Q}[x]$, the g.c.d. of polynomials $f_{1}, f_{2}, \ldots, f_{k}$ is defined as a polynomials of the greatest degree which divides each of $f_{1}, f_{2}, \ldots, f_{k}$. A g.c.d. is unique up to multiplication by a non-zero constant; the g.c.d. of two polynomials can be found by Euclid's algorithm. Polynomials whose g.c.d. equals 1 are called coprime. Analogues of the theorem on a linear representation of g.c.d. and the fundamental theorem of arithmetics are the following theorems:
Fact 0.1 (Theorem on linear representation of g.c.d.). For any $f_{1}, f_{2}, \ldots, f_{k} \in \mathbb{Q}[x]$ there exist $u_{1}, u_{2}, \ldots, u_{k} \in \mathbb{Q}[x]$ such that
$$
\operatorname{gcd}\left(f_{1}, f_{2}, \ldots, f_{k}\right)=u_{1} f_{1}+u_{2} f_{2}+\cdots u_{k} f_{k}
$$
Fact 0.2 (The fundamental theorem of arithmetic). Each non-constant polynomial $f \in \mathbb{Q}[x]$ can be represented as $f=h_{1} \cdot h_{2} \cdots h_{r}$, where $h_{i}$ are irreducible over $\mathbb{Q}$ polynomials. If $f=h_{1}^{\prime} \cdot h_{2}^{\prime} \cdots h_{s}^{\prime}$ is some other such decomposition, then $s=r$, and for some permutation of $h_{1}^{\prime}, h_{2}^{\prime}, \ldots, h_{r}^{\prime}$ we have
$$
h_{1}^{\prime}=u_{1} h_{1}, \quad h_{2}^{\prime}=u_{2} h_{2}, \quad \ldots, \quad h_{r}^{\prime}=u_{r} h_{r}
$$
where $u_{1}, u_{2}, \ldots, u_{r}$ are non-zero rational numbers.
Collecting powers of equal polynomials, one may obtain a canonical representation, similar to canonacal representation of a potitive integer as a product of powers of primes.
### 0.3 Gauss' Lemma
For a polynomial $f \in \mathbb{Z}[x]$, we define the content $d(f)$ of $f$ as the g.c.d. of all its coefficients. A polynomial $f$ is primitive if $d(f)=1$.
Fact 0.3. Let $f, g \in \mathbb{Z}[x]$. Then $d(f g)=d(f) \cdot d(g)$. In particular, the product of two primitive polynomials is primitive.
Proof. We start with the second statement. Assume, for the sake of contradiction, that there exist primitive polynomials $f(x)=a_{0}+a_{1} x+\cdots+a_{n} x^{n}$ and $g(x)=b_{0}+b_{1} x+\cdots+b_{m} x^{m}$ in $\mathbb{Z}[x]$ such that their product is not primitive, so that $p \mid d(f g)$ for some prime $p$.
Let $s, t$ be the least indices such that $p \nmid a_{s}, p \nmid b_{t}$. Such indices exist since $f$ and $g$ are primitive. The coefficiet of $x^{s+t}$ in $f g$ is equal to
$$
c_{s+t}=a_{s} b_{t}+\left(a_{s+1} b_{t-1}+a_{s+2} b_{t-2}+\cdots\right)+\left(a_{s-1} b_{t+1}+a_{s-2} b_{t+2}+\cdots\right)
$$
Since $a_{s-i}$ and $b_{t-i}$ are divisible by $p$ for $i>0$, and $p \mid c_{s+t}$, we obtain $p \mid a_{s} b_{t}$ which contradicts the choice of $s$ and $t$.
To prove the general statement, let us represent arbitrary polynomials $f, g \in \mathbb{Z}[x]$ as
$$
f=d(f) f_{0}, \quad g=d(g) g_{0}
$$
where $f_{0}, g_{0}$ are primitive. Since $f g=d(f) d(g) \cdot f_{0} \cdot g_{0}$, and $d\left(f_{0} \cdot g_{0}\right)=1$, we get $d(f g)=d(f) d(g)$, as required.
Fact 0.4 (Gauss' Lemma). A polynomial $\mathbb{Z}$ irreducible over $\mathbb{Z}$ is also irreducible over $\mathbb{Q}$.
Proof. Assume that for some polynomials $h \in \mathbb{Z}[x]$ and $f, g \in \mathbb{Q}[x]$ we have $h=f g$. After multiplying both sides by the l.c.m. of the denominators of all coefficients in $f$ and $g$, and dividing by the g.c.d. of all numeratorsu, we have:
$$
a h=b f_{0} g_{0}
$$
where $a, b \in \mathbb{Z}$, and $f_{0}, g_{0}$ are primitive polynomials over $\mathbb{Z}$. By the previous, $a \cdot d(h)=b$, so that after cancellation we have
$$
h=d(h) f_{0} g_{0}
$$
hence $h$ is reducible over $\mathbb{Z}[x]$.
Using Gauss' lemma and the fundamental theorem of arithmetic for $\mathbb{Q}[x]$, one may prove the following
Fact 0.5 (The fundamental theorem of arithmetic in $\mathbb{Z}[x]$ ). A non-constant polynomial $f \in \mathbb{Z}[x]$ can be presented as $f=h_{1} \cdot h_{2} \cdots h_{r}$, where $h_{i} \in \mathbb{Z}[x]$ are irreducible over $\mathbb{Q}$ (non-constant) polynomials.
Dividing by constants, one may transform $f=h_{1} h_{2} \cdots h_{r}$ to a canonical representation
$$
f=u \cdot g_{1}^{\alpha_{1}} g_{2}^{\alpha_{2}} \cdots g_{k}^{\alpha_{k}}
$$
where $g_{i} \in \mathbb{Z}[x]$ are pairwise coprime primitive irreducible over $\mathbb{Q}$ (non-constant) polynomials, $u \in \mathbb{Z}, u \neq 0, k \geqslant 1, \alpha_{i} \geqslant 1$.
The canonical representation is unique up to multyplying factors by $\pm 1$.
### 0.4 Hensel's Lemma
Here we present an important result, which will be helpful further. First, let us note that for a constant polynomila we have:
$$
f(x+y)=f(x)
$$
Next, for ant linear $f$ we have
$$
f(x+y)=f(x)+c y
$$
where $c$ is the coefficient of $x$, or, in other word, the derivative of $f$.
The above notes can be generalized for the case of an arbitrary polynomial (e.g., by induction over the degree) in the following way: if $f$ is a polynomials of degree $d$ with integer coefficients, then
$$
f(x+y)=f(x)+f_{1}(x) y+f_{2}(x) y^{2}+\cdots+f_{d}(x) y^{d}
$$
where $f_{i}, i=1,2, \ldots, d$ are polynomials with integer coefficients of degree not greater than $d-i$; in fact, $f_{i}=\frac{f^{(i)}}{i!}$.
The corollary of this statement is the following useful
Fact 0.6 (Hensel's Lemma). Let $f \in \mathbb{Z}$. Then for any prime $q$, positive integer $s$, and integers $r, n$, we have
$$
f\left(n+r q^{s}\right) \equiv f(n)+r q^{s} f^{\prime}(n) \quad\left(\bmod q^{2 s}\right)
$$
## 1 Divisors of polynomial values, with applications
Let us fix the following notation. For $f \in \mathbb{Z}[x]$ denote by $D(f)$ the set of all positive integers $m$ such that $f(n)$ is divisible by $m$ for some integer $n$. In other words, $m \in D(f)$ iff $f(x) \equiv 0(\bmod m)$ has an integer root.
More generally, for $f_{1}, f_{2}, \ldots, f_{k} \in \mathbb{Z}[x]$ denote by $D\left(f_{1}, f_{2}, \ldots, f_{k}\right)$ the set of all positive integers $m$ such that the system
$$
f_{i}(x) \equiv 0 \quad(\bmod m), \quad i=1,2, \ldots, k
$$
has an integer root. Denote by $P\left(f_{1}, f_{2}, \ldots, f_{k}\right)$ the set of all primes in $D\left(f_{1}, f_{2}, \ldots, f_{k}\right)$.
It is clear that if $f(n)$ is divisible by $m$, then $f(n+m t)$ is divisible by $m$ для for any integer $t$. Thus, the set of integers $x$ satisfying (1) is either empty, or is a union of (both-sides) infinite arithmetic progressions with step $m$.
Obviously, if $m \in D(f)$, then all positive divisors of $m$ also belong to $D(f)$. For a constant polynomial $f=u \in \mathbb{Z}$ the set $D(u)$ coincides with the set of all positive divisors of $u$. It is easy to show that for a non-constant $f \in \mathbb{Z}$ the set $D(f)$ is infinite.
Problem 1.1. a) Prove that for any positive integers $m_{1}, \ldots, m_{n}$ there exists a polynomial $f \in \mathbb{Z}[x]$ such that neither of $m_{1}, \ldots, m_{n}$ lies in $D(f)$.
b) Prove that $D(f g) \supset D(f)$ for any two polynomials $f, g \in \mathbb{Z}[x]$.
c) Prove that $D(f(g)) \subset D(f)$ for any two polynomials $f, g \in \mathbb{Z}[x]$.
Problem 1.2. Prove that for any non-constant polynomial $f \in \mathbb{Z}[x]$ the set $P(f)$ is infinite.
Problem 1.3. Prove that if $f_{1}, f_{2}, \ldots, f_{k} \in \mathbb{Z}[x], k \geqslant 2$, are coprime polynomials, then the set $D\left(f_{1}, f_{2}, \ldots, f_{k}\right)$ is finite. (And, therefore, $P\left(f_{1}, f_{2}, \ldots, f_{k}\right)$ is finite.)
Problem 1.4. Let $f \in \mathbb{Z}[x]$. Prove that if $m_{1}, m_{2} \in D(f)$ and $\operatorname{gcd}\left(m_{1}, m_{2}\right)=1$, then $m_{1} m_{2} \in D(f)$.
Problem 1.5. Find $P(f)$ if
a) $f(x)=x^{2}+1$;
b) $f(x)=x^{2}+x+1$;
c) $f(x)=x^{p-1}+x^{p-2}+\cdots+x+1$, where $p>2$ is a prime.
Problem 1.6. a) It is clear that if $f \in \mathbb{Z}[x]$ has an integer root, then $P(f)$ is the set of all primes. Show that the converse is false.
b) Assume that $P(f)$ is the set of all primes. Determine if it follows that $f$ has a rational root.
For a polynomial $f \in \mathbb{Z}[x]$ denote by $P_{\alpha}(f)$ the set of all primes $p$ such that $v_{p}(f(n))=\alpha$ for some integer $n$.
It is clear that if $v_{p}(f(n))=\alpha$, then $v_{p}\left(f\left(n+p^{\alpha+1} k\right)\right)=\alpha$ for any integer $k$. Thus the set of integers $x$ such that $v_{p}(f(x))=\alpha$, is either empty, or is a union of (both-sides) infinite arithmetic progressions with step $p^{\alpha+1}$.
It is easy to see thet $P_{\alpha}(f) \subset P(f)$, moreover, for a non-zero $f \in \mathbb{Z}[x]$ we have $P(f)=\bigcup_{\alpha=1}^{\infty} P_{\alpha}(f)$. Note that for $f \in \mathbb{Z}[x]$ and $k \in \mathbb{N}$ we have $P_{\alpha}(f)=P_{k \alpha}\left(f^{k}\right)$.
Problem 1.7. Let $f \in \mathbb{Z}[x]$.
a) If $p \in P(f) \backslash P\left(f, f^{\prime}\right)$, then $p \in P_{\alpha}(f)$ for an unbounded set of VALUES OF $\alpha$.
b) Prove that, in fact, EACH OF the sets $P_{\alpha}(f) \backslash P\left(f, f^{\prime}\right)$ coincides with $P(f) \backslash P\left(f, f^{\prime}\right)$ for ALL $\alpha=1,2,3, \ldots$.
Problem 1.8. It is clear that if a polynomial $f \in \mathbb{Z}[x]$ has an integer root, then $D(f)=\mathbb{N}$. Determine if the converse is true.
Problem 1.9. Let $f \in \mathbb{Z}[x]$ be a polynomial such that for any positive integer $n$ the value $f(n)$ is a perfect power, i.e., $f(n)=m(n)^{s(n)}$, for some positive integers $m(n)$ и $s(n)>1$.
a) Prove that $f$ is reducible (over $\mathbb{Q}$ ).
b) Prove that there exist a positive integer $s>1$ and a polynomial $q \in \mathbb{Z}[x]$ such that $f=q^{s}$.
Problem 1.10. Prove that for a non-constant $f \in \mathbb{Z}[x]$ the set $P(f)$ contains infinitely many primes of the form $4 k+1$.
## 2 On a problem of V.A. Senderov
Problem 2.1. Let $Q(x)$ be a polynomial with real coefficients such that its leading coefficient is rational. It is known that $Q(x)^{k} \in \mathbb{Z}[x]$ for some positive integer $k>1$. Prove that $Q(x) \in \mathbb{Z}[x]$.
Problem 2.2. A polynomial $P(x)$ of even degree with integer coefficients is given. The leading coefficient of $P(x)$ is a perfect square, and $P(n)$ is a perfect square for any large enough positive integer $n$. Prove that there is a polynomial $Q(x)$ with integer coefficients such that $P(x)=Q(x)^{2}$.
Problem 2.3. A polynomial $P(x)$ with integer coefficients is given such that $P(n)$ is a perfect square for any large enough positive integer $n$. Prove that there is a polynomial $Q(x)$ with integer coefficients such that $P(x)=Q(x)^{2}$.
For every positive integer $k$, we introduce the polynomial:
$$
P_{k}(x)=x(x+1) \cdots(x+k-1)
$$
Problem 2.4. Let $Q(x)$ be a monic polynomial of degree $l>2$ with real coefficients such that $R(x)=Q(x)-b$ has exactly $l$ different real roots $r_{1}, r_{2}, \ldots, r_{l}$, while $S(x)=Q(x)+b$ has exactly $l$ different real roots $s_{1}, s_{2}, \ldots, s_{l}$, for some positive real $b$. Sort the elements of the set $\left\{r_{1}, r_{2}, \ldots, r_{l}, s_{1}, s_{2}, \ldots, s_{l}\right\}$ under the assumptions $r_{1}4$ and $c$.
Problem 2.6. Prove that for any complex number $a$ the degree of the g.c.d. of polynomials $P_{k}(x)-a$ and $P_{k}^{\prime}(x)$ does not exceed 2. Moreover, if $k$ is odd then that degree is at most 1 .
Problem 2.7. Suppose that there exist polynomials $Q(x)$ and $R(x)$ of degree at least 2 with rational coefficients such that $P_{k}(x)=R(Q(x))$. Prove that $k$ is even, and $\operatorname{deg} Q(x)=2$.
Problem 2.8 (V. A. Senderov). Find all positive integers $k$ and $c$ such that $P_{k}(n)+c$ is a perfect power for any positive integer $n$.
## 3 Miscellaneous
Problem 3.1. Let $f, g \in \mathbb{Z}[x]$. Assume that the number $f(n)$ is divisible by $g(n)$ for infinitely many positive integers $n$.
a) Prove that if the leading coefficient of $g$ equals 1 , then $g \mid f$ (in $\mathbb{Z}[x]$ ).
b) Prove that there exists an infinite arithmetic progression $A$ of positive integers such that $g(n) \mid f(n)$ for any $n \in A$.
Problem 3.2. Let $f \in \mathbb{Z}[x]$, and let $a_{1}, a_{2}, \ldots, a_{m}$ be positive integers. Given that for any integer $n$ the number $f(n)$ is divisible by at lease one of $a_{1}, a_{2}, \ldots, a_{m}$, prove that there exists a positive integer $k \leq m$ such that $f(n)$ is divisible by $a_{k}$ for any integer $n$.
Each polynomial $f \in \mathbb{Z}[x]$ determines a function $\mathbb{Z}_{n} \rightarrow \mathbb{Z}_{n}$, since $f(m)$ modulo $n$ is uniquely defined by the residue of $m$ modulo $n$.
Problem 3.3. a) Prove that if the function $\mathbb{Z}_{n} \rightarrow \mathbb{Z}_{n}$ is determined by a polynomial in $\mathbb{Z}[x]$, then it is also determined by a polynomial in $\mathbb{Z}[x]$ whise degree is at most $n-1$.
b) Find all $n$ such that any fuction $\mathbb{Z}_{n} \rightarrow \mathbb{Z}_{n}$ can be determined by a polynomial in $\mathbb{Z}[x]$.
c) Find all positive integers $n>1$ satisfying the following condition: if $f \in \mathbb{Z}$, $\operatorname{deg} f \leqslant n-1$, is such that $f(m)$ is divisible by $n$ for all integers $m$, then all coefficients of $f$ are divisible by $n$.
d) Find all $n$ such that any function $\mathbb{Z}_{n} \rightarrow \mathbb{Z}_{n}$ satisfying
$$
\operatorname{gcd}(a-b, n) \mid \operatorname{gcd}(f(a)-f(b), n), \quad a, b \in \mathbb{Z}_{n}
$$
is determined by a polynomial from $\mathbb{Z}[x]$.
Problem 3.4. Let $f \in \mathbb{Z}[x]$ defines a bijection modulo any prime number. Prove that $f$ has degree 1 .
Problem 3.5. Let $f, g \in \mathbb{Z}[x]$ be polynomials such that for any integer $x$ the number $f(g(x))-x$ is divisible by a fixed integer $n$. Prove that $g(f(x))-x$ is also divisible by $n$ for any integer $x$.
## Polynomials in number theory
(Continuation)
### 0.5 Algebraic numbers and minimal polynomials
A complex number $\alpha$ is called algebraic if there exists a non-zero polynomial $f \in \mathbb{Q}[x]$ such that $f(\alpha)=0$. E.g., $\sqrt{2}+\sqrt{3}$ is algebraic, since $f(\sqrt{2}+\sqrt{3})=0$, где $f(x)=x^{4}-10 x^{2}+1$. Obviously, the set of algebraic numbers contains $\mathbb{Q}$.
A non-zero polynomial $f \in \mathbb{Q}[x]$ of minimal degree such that $f(\alpha)=0$ is called a minimal polynomial of algebraic number $\alpha$. The degree of this polunomial is called the degree of $\alpha$.
It is easy to see that the minimal polynomial of any algebraic number is irreducible in $\mathbb{Q}[x]$. Moreover, any polynomial $h \in \mathbb{Q}[x]$ having $\alpha$ as a root, is divisible by the minimal polynomial. Indeed, let us divide $h$ by the minimal polynomial $f$ with remainder:
$$
h=f q+r
$$
where $q, r \in \mathbb{Q}[x]$ and $\operatorname{deg} r<\operatorname{deg} f$. Plug in $x=\alpha$ to obtain $r(\alpha)=0$. By minimality of $f$, we see that $r$ vanishes, i.e., $h$ is divisible by $f$.
The next fact follows immediately from the above.
Fact 0.7. Any irreducible $f \in \mathbb{Q}[x]$ is the minimal polynomial for each of its roots.
Problem 0.8. a) Let $f \in \mathbb{Q}[x]$ be an irreducible polynomial. Prove that $f$ and $f^{\prime}$ are coprime.
b) Let $f \in \mathbb{Q}[x]$. Prove that $f$ and $f^{\prime}$ are not coprime iff there exists a non-constant $g \in \mathbb{Q}[x]$ such that $g^{2} \mid f$.
c) Prove that an irreducible (in $\mathbb{Q}[x]$ ) polynomial has no multiple complex roots (and hence, the number of its complex roots equals its degree).
A non-empty set of complex numbers $\mathbb{F}$ is called a number field if it contains some nonzero number and, along with any numbers $a, b \in \mathbb{F}$, it also contains the numbers $a+b$, $a b,-a$ and $a^{-1}$ (the latter - only if $a \neq 0$ ). In other words, in the field, one can perform all usual arithmetic operations (addition, multiplication, subtraction, and division by a nonzero number), obtaining numbers that also belong to the field. (In this project, number fields are referred to as fields.)
Since most algebraic transformations use exactly the four mentioned operations, many facts whose proofs boil down to only these operations hold over any field. E.g., denoting by $\mathbb{F}[x]$ the set of polynomials with coefficients from $\mathbb{F}$, it is easy to show that the theorem on the linear representation of g.c.d. and the fundamental theorem of arithmetic are true in $\mathbb{F}[x]$ (see facts $0.1,0.2$ ).
It is useful to note here that the procedure of determining the g.c.d. of two polynomials (by Euclid's algorithm) does not depend on the field over which we work. This means that the resulting g.c.d. does not depend on the field (the only restriction is that all the coefficients of the polynomials must lie in this field).
Similarly to the definition above, a complex number $\alpha$ is said to be algebraic over a field $\mathbb{F}$ if $\alpha$ is a root of a nonzero polynomial in $\mathbb{F}[x]$. Such a polynomial of minimal degree is called minimal polynomial of $\alpha$ over the field $\mathbb{F}$. The above facts about algebraic numbers also carry over to the case of an arbitrary field.
Important (yet basic) examples of number fields are provided by the following fact.
Fact 0.9. a) Let $\alpha$ be the algebraic niumber. Then the set
$$
\mathbb{Q}[\alpha]=\{f(\alpha): f \in \mathbb{Q}[x]\}
$$
is a field.
b) Similarly, if $\alpha$ is algebraic over $\mathbb{F}$, then the set
$$
\mathbb{F}[\alpha]=\{f(\alpha): f \in \mathbb{F}[x]\}
$$
is a field.
The field $\mathbb{F}[\alpha]$ is called the extension of the field $\mathbb{F}$ by the algebraic number $\alpha$.
Proof. We prove the more general statement b).
For arbitrary $g(\alpha), h(\alpha) \in \mathbb{F}[\alpha]$ it is clear thet $g(\alpha)+h(\alpha),-g(\alpha)$, and $g(\alpha) h(\alpha)$ lie in $\mathbb{F}[\alpha]$. It remains to show that $g(\alpha)^{-1} \in \mathbb{F}[\alpha]$, if $g(\alpha) \neq 0$,
Let $f \in \mathbb{F}[x]$ be the minimal polynomial of $\alpha$ over $\mathbb{F}$; then $g$ is not divisible by $f$. since $g(\alpha) \neq 0$. Irreducibility of $f$ now implies that $\operatorname{gcd}(f, g)=1$, and by the theorem on the linear representation of the g.c.d. there are polynomials $a, b \in \mathbb{F}[x]$ such that $1=a f+b g$. Substituting $\alpha$ into this equality, we get
$$
1=a(\alpha) f(\alpha)+b(\alpha) g(\alpha)=b(\alpha) g(\alpha)
$$
i.e., $g(\alpha)^{-1}=b(\alpha)$ is in $\mathbb{F}[x]$.
Remark. Using division by $f$ with a remainder, it is easy to get that each element in $\mathbb{F}[\alpha]$ is represented as $g(\alpha)$, where $g \in \mathbb{F}[x]$ and $\operatorname{deg} g<\operatorname{deg} f$ (check this out!).
The next problem can be solved in different ways; one possible solution is based on the following well-known lemma.
Fact 0.10. Consider a system of homogeneous linear equations, that is - a system of the form
$$
\begin{gathered}
a_{11} x_{1}+a_{12} x_{2}+\ldots+a_{1 n} x_{n}=0 \\
a_{21} x_{1}+a_{22} x_{2}+\ldots+a_{2 n} x_{n}=0 \\
\vdots \\
a_{k 1} x_{1}+a_{k 2} x_{2}+\ldots+a_{k n} x_{n}=0
\end{gathered}
$$
where $a_{i j}$ are fixed (say, rational) coefficients, and $x_{i}$ are variables. Then, if the number $k$ of equations is less than the number $n$ of variables, then the system has a (rational) solution in which not all values of the variables are zero.
This lemma can be proved, for example, by successive elimination of variables. The lemma is also true over any field: if the coefficients belong to the field $\mathbb{F}$, then one can also find a non-trivial solution in which all the values of the variables lie in $\mathbb{F}$.
The next problem can also be generalized over any field, but for ease of perception we formulate it over the field $\mathbb{Q}$.
Problem 0.11. a) Let $\alpha$ be an algebraic number, and put $\mathbb{F}=\mathbb{Q}[\alpha]$. Then any number in $\mathbb{F}$ is algebraic.
b) Let $\alpha$ be an algebraic number, and let number $\beta$ be algebraic over $\mathbb{F}=\mathbb{Q}[\alpha]$. Then $\beta$ is algebraic.
c) The set $\overline{\mathbb{Q}}$ of all algebraic numbers is a field.
d) Moreover, any number algebraic over $\overline{\mathbb{Q}}$ lies in $\overline{\mathbb{Q}}$. (Such fields are called algebraically closed.)
Let $\alpha$ and $\beta$ be two algebraic numbers; put $\mathbb{F}=\mathbb{Q}[\alpha]$. Then the field $\mathbb{F}[\beta]$ is also denoted by $\mathbb{Q}[\alpha, \beta]$. It is easy to check that $\mathbb{Q}[\alpha, \beta]=\mathbb{Q}[\beta, \alpha]$.
Similarly, one can define the field $\mathbb{Q}\left[\alpha_{1}, \ldots, \alpha_{n}\right]$ for $\alpha_{i} \in \overline{\mathbb{Q}}$.
Problem 0.12 (Primitive element theorem). a) Let $\alpha$ and $\beta$ be two algebraic numbers. Prove that there exists an algebraic number $\gamma$ such that $\mathbb{Q}[\alpha, \beta]=\mathbb{Q}[\gamma]$. Moreover, this number can be chosen in the form $\gamma=\alpha+t \beta$ for $t \in \mathbb{Q}$.
b) Let $\alpha_{1}, \alpha_{2}, \ldots, \alpha_{n}$ be algebraic numbers. Prove that there exists an algebraic number $\gamma$ such that $\mathbb{Q}\left[\alpha_{1}, \alpha_{2}, \ldots, \alpha_{n}\right]=\mathbb{Q}[\gamma]$. In what form can $\gamma$ be sought for in this case?
### 0.6 Cyclotomic polynomials
A complex number $z$ is called a primitive $n$-th degree root of unity if $z^{n}=1$, but $z^{k} \neq 1$ при $1 \leqslant k1$, and let $p$ be a prime number. Prove that the function $\mathbb{Z}_{p} \rightarrow \mathbb{Z}_{p}$ determined by $f$ is not a bijection, if
a) $d=p-1$;
b) $d$ is a divisor of $p-1$,
Problem 3.8. Let $f \in \mathbb{Z}[x]$ deftermine a bijection modulo $n$. Prove that there exists $g \in \mathbb{Z}[x]$ such that $f(g(x))-x$ is divisible by $n$ for any integer $x$.
## Polynomials in number theory
## 0 Auxiliary facts
Collecting powers of equal polynomials, one may obtain a canonical representation, similar to canonacal representation of a potitive integer as a product of powers of primes.
### 0.5 Algebraic numbers and minimal polynomials
0.8. a) Assume the contrary, that there exists a non-constant $d \in \mathbb{Q}[x]$ such that $d$ divides both $f$ and $f^{\prime}$. Since $f$ is irreducible, we have $d=c \cdot f$, hence $f$ divides $f^{\prime}$. It is impossible, since $\operatorname{deg} f^{\prime}<\operatorname{deg} f$.
b) First, auume that such $g$ exists. We have
$$
f^{\prime}=g^{2} h^{\prime}+2 g g^{\prime} h=g \cdot\left(g h^{\prime}+2 g^{\prime} h\right)
$$
hence $g \mid \operatorname{gcd}\left(f, f^{\prime}\right)$. Therefore, $f$ and $f^{\prime}$ are not coprime.
Conversely, if the polynomial $g$ does not exist, then in the canonical decomposition
$$
f=h_{1} h_{2} \ldots h_{k}
$$
into irreducible factors, all the polynomials $h_{1}, \ldots, h_{k}$ are different (more precisely, there are no proportional ones among them). If $f$ and $f^{\prime}$ are not coprime, then their g.c.d. would be the product of some of $h_{i}$; let, for definiteness, it is divisible by $h_{1}$, i.e. $h_{1} \mid f^{\prime}$. But
$$
f^{\prime}=h_{1}^{\prime} h_{2} \ldots h_{k}+h_{1} h_{2}^{\prime} h_{3} \ldots h_{k}+h_{1} h_{2} h_{3}^{\prime} \ldots h_{k}+\ldots+h_{1} h_{2} \ldots h_{k}^{\prime}
$$
where all terms, except possibly the first one, are divisible by $h_{1}$. This means that the first term must also be divisible by this polynomial. Since $h_{1}$ is irreducible, one of the factors must be divisible by it. This is impossible, because $h_{1} \nmid h_{i}$ for $i \geqslant 2$ by our assumption, and $h_{1} \nmid h_{1}$ by item a).
c) Let the polynomial $f$ be irreducible. By item a), the polynomials $f$ and $f^{\prime}$ are coprime. By item $\mathrm{b}$ ), this means that the square of a non-constant polynomial does not divide $f$; in particular, $f$ is not divisible by a polynomial of the form $(x-\alpha)^{2}$ for $\alpha \in \mathbb{C}$. This means that $f$ has no multiple roots.
0.11. a) Let $\gamma \in \mathbb{F}[\alpha]$, and let $n$ be the degree of $\alpha$. Then for each $i=0,1, \ldots, n$, we have $\gamma^{i}=g_{i}(\alpha)$, where $g_{i} \in \mathbb{F}[x]$ and $\operatorname{deg} g_{i}2$, then $\Phi_{n}(x)$ has no real roots, all the roots' absolute values equal to 1 , and the roots fall into pairs of conjugates. Hence $\Phi_{n}(x)$ is a product of polynomials of the form $x^{2}+a x+1$, therefore, $\Phi_{n}(0)=1$.
b) Note that $\Phi_{1}(1)=0$, and 1 is not a root of any other cyclotomical polynomial. If $n$ is divisible ny a prime $p$, then by problem $0.14 \mathrm{a}$, we have $\Phi_{n}(1)=\Phi_{n p}(1)$. Thus it suffices to calculate $\Phi_{n}(1)$ for square-free positive integers $n$.
If $n$ is a prime, then $\Phi_{n}(1)=n$. If $n$ is divisible by two distinct primes $p$ and $q$, then $n / p$ is an integer greater than 1 , and by problem $0.14 \mathrm{~b}$, we have
$$
\Phi_{n}(1)=\frac{\Phi_{n / p}(1)}{\Phi_{n / p}(1)}=1
$$
Finally, we conclude that for $n>1$
$$
\Phi_{n}(1)= \begin{cases}p & , \text { если } n=p^{k} \\ 1 & , \text { если } n \neq p^{k}\end{cases}
$$
Remark. The problem can also be solved by induction, using decomposition of $x^{n}-1$ as a product of cyclotomic polynomials.
0.17. a) The solution of problem $0.8 \mathrm{~b}$ works literally, with one exception. Namely, we need a different argument showing that $h_{1} \nmid h_{1}^{\prime}$. The issue is that, although the degree of $h_{1}^{\prime}$ is still smaller than that of $h_{1}$, it does not by itself yield the result, as the derivative of a non-constant polynomial may vanish in $\mathbb{Z}_{p}[x]$ (example: $\left(x^{p}-1\right)^{\prime}=p x^{p-1}=0$ ). Thus, we need to show additionally that $h_{1}^{\prime} \neq 0$, if $h_{1} \in \mathbb{Z}_{p}[x]$ is irreducible.
THis can be done as follows. From Newton's binomial formula, it is easy to see that $(a+b)^{p}=a^{p}+b^{p}$ for all $a, b \in \mathbb{Z}_{p}[x]$ (in particular, this holds for elements in $\mathbb{Z}_{p}$ ). Applying this property several times, we obtain
$$
\left(a_{0}+a_{1} x+a_{2} x^{2}+\ldots+a_{k} x^{k}\right)^{p}=a_{0}+a_{1} x^{p}+a_{2} x^{2 p}+\ldots+a_{k} x^{p k}
$$
Thus, if we had $h_{1}^{\prime}=0$ (which means that the degrees of monomials in $h_{1}$ are all divisible by $p$ ), then $h_{1}$ would be the $p$-th power of another polynomial; hence it was not irreducible. This finishes the solution.
b) Arguing indirectly, assume that both $\Phi_{m}(x)$ and $\Phi_{n}(x)$ are divisible by some nonconstant polynomial $g(x)$ in $\mathbb{Z}_{p}[x]$. From the factorization of $x^{m n}-1$ into a product of cyclotomic polynomials, we obtain $g(x)^{2} \mid x^{m n}-1$ (surely, in $\mathbb{Z}_{p}[x]$ ). Part a) now yields that the polynomial $x^{m n}-1$ is not coprime with its derivative $m n x^{m n-1}$ in $\mathbb{Z}_{p}[x]$, which is obviously wrong, as $p \nmid m n$ and $x \nmid x^{m n}-1$.
## 1 Divisors of polynomial values, with applications
1.1. a) It suffices to put all coefficients, except for the last one, to be divisible by the product $m_{1} m_{2} \cdots m_{n}$, and set the last coefficient to be 1 .
b) Follows from $f(n) \mid f(n) g(n)$.
c) It follows from the fact that the set of values of $f(g)$ at integer points is a subset of the set of values of $f$ at integer points.
1.2. Assume the contrary: let $P(f)$ consist of $k$ primes, i.e. $P(f)=\left\{p_{1}, p_{2}, \ldots, p_{k}\right\}$. Take $a \in \mathbb{Z}$ for which $f(a)=b \neq 0$. We define the polynomial $g$ by the equality $g(x)=\frac{f(a+b p x)}{b}$, where $p=p_{1} p_{2} \cdots p_{k}$. It is easy to see that $g$ is a non-constant polynomial with integer coefficients, and $P(g) \subset P(f)$. The constant term $g(0)=1$, and all other coefficients of the polynomial $g$ are divisible by $p$, therefore, none of the numbers $p_{1}, p_{2}, \ldots, p_{k}$ can lie in the set $P(g)$. But then the set $P(g)$ is empty, which is a contradiction. (Note that the above argument generalizes Euclid's proof of the infinity of the set of primes.)
1.3. By the theorem on a linear representation of the g.c.d., there exist polynomials $u_{1}, u_{2}, \ldots, u_{k} \in \mathbb{Q}[x]$ such that
$$
u_{1} f_{1}+u_{2} f_{2}+\cdots+u_{k} f_{k}=1
$$
If we multiply the polynomials $u_{1}, \ldots, u_{k}$ by $m$ that is the l.c.m. of the denominators of all nonzero coefficients $u_{1}, \ldots, u_{k}$, we get the polynomials $v_{1}, v_{2}, \ldots, v_{k} \in \mathbb{Z}[x]$. Multiplying
the equality $u_{1} f_{1}+u_{2} f_{2}+\cdots+u_{k} f_{k}=1$ by $m$, we get $v_{1} f_{1}+v_{2} f_{2}+\cdots+v_{k} f_{k}=m$. We see that if for some $n$ the numbers $f_{1}(n), f_{2}(n), \ldots, f_{k}(n)$ are divisible by $s$, then $m$ is also divisible by $s$. This implies that the set $D\left(f_{1}, f_{2}, \ldots, f_{k}\right)$ can contain only divisors of the number $m$, and hence it is finite.
1.4. $f(n)$ is divisible by $m_{i}$ for all $n$ from some arithmetic progression with step $m_{i}$. Since $\operatorname{gcd}\left(m_{1}, m_{2}\right)=1$, by the Chinese remainders theorem, progressions with steps $m_{1}$ and $m_{2}$ intersect. Then $n$ from this intersection satisfies $m_{1} m_{2} \mid f(n)$.
1.5. a) -1 is a quadratic residue modulo $p$ of the form $4 k+1$ and non-quadratic residue modulo $p$ of the form $4 k+3$ (e.g., it follows from Euler's criteria). This means that all primes of the form $4 k+1$ are in $P(f)$, while all primes of the form $4 k+3$ are outside $P(f)$. Moreover, obviously, $2 \in P(f)$.
c) $f(1)=p$, hence $p \in P(f)$.
Let $f(n)$ be divisible by a prime $q \neq p$. Since $x^{p}-1$ is divisible by $f$ (in $\mathbb{Z}[x]$ ), then $n^{p} \equiv 1(\bmod q)$, hence $\operatorname{ord}_{p}(n)$ equals 1 or $p$, since it divides $p$.
In the first case $n \equiv 1(\bmod q)$, it follows $f(n)=n^{p-1}+n^{p-2}+\cdots+n+1 \equiv p(\bmod q)$. We have $p \equiv 0(\bmod q)$, откуда $p=q-$ a contradiction.
In the second case $q-1$ divides $\operatorname{ord}_{p}(n)=p$, i.e., $q \equiv 1(\bmod p)$.
Conversely, let prime $q \equiv 1(\bmod p)$. Let $a$ be a primitive root modulo $q$, and let $n=a^{\frac{q-1}{p}}$. Therefore, $n \not \equiv 1(\bmod q)$, but $n^{p} \equiv 1(\bmod q)$, hence $f(n)=\frac{n^{p}-1}{n-1}$ is divisible by $q$.
Thus $P(f)$ consists of all primes of the form $p k+1$, and $p$.
1.6. a) If $f_{k}(x)=k x-1$ for some positive integer $k$, then $P\left(f_{k}\right)$ coincides with the set of all primes, except divisors of $k$. Threrfore, it is clear that for $f(x)=(2 x-1)(3 x-1)$ the set $P(f)$ coincides with the set of all primes.
b) It does not follow. For each prime $p$ and integers $t, s$ at least one of numbers $t, s, t s$ is either zero residue, or a quadratic residue modulo $p$. Hence, for the polynomial
$$
f_{t, s}(x)=\left(x^{2}-t\right)\left(x^{2}-s\right)\left(x^{2}-t s\right)
$$
we have $p \in P\left(f_{t, s}\right)$. It suffises to take $t, s$ such that $t, s, t s$ are not perfect squares.
1.7. a) Let us show by induction on $\alpha$, that there exists an integer $n_{\alpha}$ such that $p^{\alpha} \mid f\left(n_{\alpha}\right)$.
For $\alpha=1$ the statement follows from $p \in P(f)$. Assume that the statement is proved for some $\alpha \in \mathbb{N}$, and $f\left(n_{\alpha}\right)=p^{\alpha} m$ for some $m \in \mathbb{Z}$. By Hensel's lemma,
$$
f\left(n_{\alpha}+p^{\alpha} z\right) \equiv p^{\alpha}\left(m+z f^{\prime}\left(n_{\alpha}\right)\right) \quad\left(\bmod p^{\alpha+1}\right)
$$
Since $p \notin P\left(f, f^{\prime}\right), f^{\prime}\left(n_{\alpha}\right)$ is not divisible by $p$, hence it suffices to choose $z=z_{0}$ so that $m+z_{0} f^{\prime}\left(n_{\alpha}\right)$ is divisible by $p$, and set $n_{\alpha+1}=n_{\alpha}+p^{\alpha} z_{0}$.
b) Take $n_{\alpha}$ from a). Then $f\left(n_{\alpha}+p^{\alpha}\right)$ is divisible by $p^{\alpha}$, and $f^{\prime}\left(n_{\alpha}\right)$ is not divisible by $p$. Thus from Hensel's lemma it follows that $v_{p}\left(f\left(n_{\alpha}+p^{\alpha}\right)-f\left(n_{\alpha}\right)\right)=\alpha$. Hence either $v_{p}\left(f\left(n_{\alpha}\right)\right)$, or $v_{p}\left(f\left(n_{\alpha}+p^{\alpha}\right)\right)$ is equal to $\alpha$.
1.8. The converse is not true. Let us prove the existence of such polynomial without rational roots. Take $f_{t, s}$ from the solution of problems 1.6 , where $t, s$ and $t s$ are not perfect squares. Assume that an odd prime $p$ is a divisor of neither of $s$ and $t$. Then $p \notin P\left(h, h^{\prime}\right)$ for each of the polynomials $h(x)=x^{2}-a$ where $a$ is equal to $t, s$ or $t s$. Then, according to the problem 1.7, $p^{\alpha} \in D\left(f_{t, s}\right)$ for arbitrarily large $\alpha$.
For a finite set of prime divisors $q$ of the numbers $t$ and $s$ take $t^{\prime}, s^{\prime}$ not divisible by any such primes $q$ and such that $t^{\prime}, s^{\prime}, t^{\prime} s^{\prime}$ are not perfect squares. Then $q^{\alpha} \in D\left(f_{t^{\prime}, s^{\prime}}\right)$ for arbitrarily large $\alpha$.
Finally, for $g(x)=x^{2}+x+2$, according to the problem $1.7,2^{\alpha} \in D(g)$ for arbitrarily large $\alpha$.
By problem 1.4, $f=f_{t, s} f_{t^{\prime}, s^{\prime}} g$ is a required polynomial such that $D(f)=\mathbb{N}$.
Remark. In fact, a required polynomial can be found just in the form $f_{t, s}$ for an appropriate choice of $t$ and $s$.
1.9. b) Let $f=a h_{1}^{\alpha_{1}} \cdots h_{t}^{\alpha_{t}}$, where $a \in \mathbb{Z}$, and $h_{1}, \ldots h_{t}$ are pairwise non-proportional primitive irreducible polynomials with integer coefficients.
Since $\operatorname{gcd}\left(h_{1}, h_{1}^{\prime}\right)=1$, by problems $1.2,1.3$, the set
$$
P_{1}\left(h_{1}\right) \backslash\left(P\left(h_{1}, h_{1}^{\prime}\right) \cup P\left(h_{1}, a\right) \cup P\left(h_{1}, h_{2}\right) \cup \cdots \cup P\left(h_{1}, h_{t}\right)\right)
$$
is infinite; take some $p_{1}$ from this set, and by problem 1.7, there exists $n_{1}$ such that $v_{p_{1}}\left(h_{1}\left(n_{1}\right)\right)=1$. By the choice of $p_{1}$, for all $m$ from progression $n_{1}+k p_{1}^{2}$ we have $v_{p_{1}}\left(h_{1}(m)\right)=1$, while $a, h_{2}(m), \ldots, h_{t}(m)$ are not divisible by $p_{1}$. Therefore, for all $m$ from progression $n_{1}+k p_{1}^{2}$ we have $v_{p_{1}}(f(m))=\alpha_{1}$.
Similarly, we define $p_{i}, i=2, \ldots, t$ and the corresponding progressions. By the Chinese remainders theorem, all $t$ progressions intersect. Let $m$ belong to all progressions. Then, on one hand, by assumption, $f(m)=u^{s}$, for some integers $u$ and $s>1$. On the other hand, $v_{p_{i}}(f(m))=\alpha_{i}, i=1, \ldots, t$, therefore all $\alpha_{i}$ are multiples of $s$, thus $f=a r^{s}$, where $r \in \mathbb{Z}[x]$. Since $f(m)=u^{s}=\operatorname{ar}(m)^{s}$, the constant $a$ is a perfect $s$-th power. Finally, we represent $f$ in the required form $f=q^{s}$.
Remark. The condition of the problem can be weakened by requiring that the values of $f$ be perfect powers only on some infinite arithmetic progression $A$. In the solution when choosing $p_{i}$ then we also require that $p_{i}$ is not a divisor of the step of $A$. Then our progressions with steps $p_{i}^{2}$ intersect with the progression $A$.
1.10. The problem follows directly from the next one (using problem 1.5a). But we will also provide a direct solution.
We will prove the existence of one prime $p=4 k+1 \in P(f)$; the proof that there are infinitely many such primes can now be obtained in the same way as in problem 1.2.
Let $\operatorname{deg} f=n$. Consider the polynomial $f(x+i)$, where $i$ is the imaginary unit. It is represented as $f(x+i)=P_{1}(x)+i P_{2}(x)$, where $P_{1}(x), P_{2}(x) \in \mathbb{Z}[x]$, and $\operatorname{deg} P_{1}=n$, $\operatorname{deg} P_{2}=n-1 \geqslant 0$. Hence, there exists $k \in \mathbb{Z}$ such that the numbers $a=P_{1}(k)$ and
$b=P_{2}(k)$ are nonzero, and $\left|P_{1}(k)\right|>\left|P_{2}(k)\right|$. Replacing $P(x)$ with $P(x+k)$, we can assume that $f(i)=a+b i$, where $|a|>|b|>0$; then $f(-i)=a-b i$.
Let $d=\operatorname{gcd}(a, b), a=a^{\prime} d, b=b^{\prime} d$; hence $\left|a^{\prime}\right| \geqslant|a| /|b|>1$. The number $a^{\prime 2}+b^{\prime 2}$ is greater than 2 and is not divisible by 4 ; hence it has an odd prime divisor $p$. Since $a^{\prime}$ and $b^{\prime}$ are coprime, $p=4 k+1$.
Let us set $Q(x)=f(x)-a-b x$. The values of $Q$ at $\pm i$ both equal zero, hence $Q$ is divisible by $(x-i)(x+i)=x^{2}+1$. Since $x^{2}+1$ have the leading coefficient 1 , we have $f(x)=a+b x+\left(x^{2}+1\right) R(x)$ for some $R(x) \in \mathbb{Z}[x]$.
Now let $n$ be such that $n b^{\prime} \equiv-a^{\prime}(\bmod p)$. Hence $a+b n=d\left(a^{\prime}+b^{\prime} n\right)$ is divisible by $p$ and $b^{\prime 2}\left(n^{2}+1\right)=\left(b^{\prime} n\right)^{2}+b^{\prime 2} \equiv a^{\prime 2}+b^{\prime 2} \equiv 0(\bmod p)$; since $b^{\prime}$ is not divisible by $p$, we have $p \mid n^{2}+1$. Therefore, $f(n)=(a+b n)+\left(n^{2}+1\right) R(n)$ is also divisible by $p$.
### 1.11. We will prove the more general statement b).
If is clear that it suffices to consider the case of irreducible over $\mathbb{Z}$ polynomials $f_{i}$ (for reducible polynomials it suffices to apply the statement for some of their irreducible divisors).
Let $\alpha_{i}$ be a root of $f_{i}$, for $i=1,2, \ldots, k$. By the Primitive element theorem, there exists an algebraic number $\gamma$ such that $\mathbb{Q}\left[\alpha_{1}, \alpha_{2}, \ldots, \alpha_{k}\right]=\mathbb{Q}[\gamma]$. In particular, there exist polynomials $A_{i} \in \mathbb{Q}[x]$ such that $\alpha_{i}=A_{i}(\gamma)$.
Let $g$ be the minimal polynomial of $\gamma$. Note that $f_{i}\left(A_{i}(\gamma)\right)=0$, so that $f_{i}\left(A_{i}(x)\right)$ is divisible by $g$ for all $i=1,2, \ldots, k$, i.e.,
$$
f_{i}\left(A_{i}(x)\right)=g(x) h_{i}(x)
$$
Coefficients of polynomials in the above equalities can be non=integers (but they are rationals); let $N$ be the maximal prime which divides at least one denominator of a coefficient.
By Problem 1.2, there exist infinitely many primes in $P(g)$. Let $p \in P(g)$ be a prime such that $p>N$, and let $p \mid g(n)$. Then the numerators of rationals $f_{i}\left(A_{i}(n)\right)=g(n) h_{i}(n)$ are divisible by $p$, since the denominators of $P_{i}(n)$ and $Q_{i}(n)$ are not divisible by $p$.
The only remaining trouble is that the numbers $A_{i}(n)$ are rational numbers, but not necessarily integers. However, their denominators are also not divisible by $p$. Hence if $A_{i}(n)=k_{i} / \ell_{i}$ for coprime integers $k_{i}$ and $\ell_{i}$, then one can take $m_{i}$, such that $k_{i} \equiv \ell_{i} m_{i}$ $(\bmod p)$. Therefore, $f_{i}\left(m_{i}\right) \equiv f_{i}\left(k_{i} / \ell_{i}\right) \equiv 0(\bmod p)$. Hence $p \in P\left(f_{i}\right)$ for all $i=1,2, \ldots, k$.
1.12. If $x \mid f$, then $D(f)=\mathbb{N}$, and the claim is obvious. Otherwise, let $b \neq 0$ be the constant term; replacing $f$ with $f(b x) / b \in \mathbb{Z}[x]$ we may assume that $b=1$. Without loss of generality, $f$ is irreducible, and its leading coefficient is positive. As in problem 1.10, it suffices to show that there exists at least one prime $p \equiv 3(\bmod 4), p \in P(f)$.
a) Assume that $\operatorname{deg} f$ is odd. Consider a large negative integer $-n$ divisible by 4 . Then $f(-n)<0$ and $f(-n) \equiv f(0) \equiv 1(\bmod 4)$, so $f(-n)=-(4 k+3)$ for some positive integer $k$. Then the prime decomposition of $4 k+3$ must contain a desired prime $p \equiv 3$ $(\bmod 4)$.
b) If $D(f)$ contains a positive integer congruent to 3 modulo 4 , one may apply the same reasoning as above. Otherwise, since $f$ is irreducible, all its roots are simple. As it has a real root, there exists an interval $(\alpha, \beta)$ on which $f$ attains only negative values.
Choose now a rational number $\frac{a}{b} \in(\alpha, \beta)$ (here $a$ and $b$ are coprime) such that $b$ has no prime divisors of the form $4 k+3$, while $4 \mid a$ (one may even choose $b=5^{N}$ for an appropriate $N)$. Recall that $f(a) \equiv 1(\bmod 4)$. Moreover, $f(a)>0$, otherwise $-f(a)$ is positive integer of the form $4 k+3$, and the same reasoning as in the part a) works. Then $f\left(\frac{a}{b}\right)=-\frac{c}{d}$, where $d=b^{n} \equiv 1(\bmod 4)$, and the positive integer $c$ satisfies
$$
c=-b^{n} f\left(\frac{a}{b}\right) \equiv-f(a) \equiv 3 \quad(\bmod 4)
$$
Therefore, there exists a prime $p \mid c$ such that $p \equiv 3(\bmod 4)$. By our assumption, $p \nmid b$; so there exist $a^{\prime} \in \mathbb{Z}$ such that $a \equiv b a^{\prime}(\bmod p)$. Then we have $b^{n} f\left(a^{\prime}\right) \equiv b^{n} f\left(\frac{a}{b}\right)=-c \equiv 0$ $(\bmod p)$, and the desired $p$ has been found.
1.13. Solution 1. Assume the contrary, so that $\Phi_{m}(x)$ and $\Phi_{n}(x)$ are divisible by $x-a$ in $\mathbb{Z}_{p}[x]$, which contradicts the problem $0.17 \mathrm{~b}$.
1.13. Solution 2. Assume that $\Phi_{m}(k)$ and $\Phi_{n}(k)$ are divisible by $p$. Then $k^{m}-1$ and $k^{n}-1$ are divisible by $p$. Let $mv_{p}\left(k^{m}-1\right)$. But since $n$ is not divisible by $p$, this contradicts LTE-lemma for $a=k^{m}$ and $b=1$.
1.14. Immediately follows from the problem 0.14 , since $2 \in P(f)$ iff at least one of numbers $f(0), f(1)$ is even.
1.15. a) Assume that $\Phi_{n}(m)$ is divisible by a prime $p$. Then $m^{n} \equiv 1(\bmod p)$. Let $d=\operatorname{ord}_{p}(m)$, then $d \mid n$ and $d \mid p-1$.
If $d=n$, then $n \mid p-1$ and $p \equiv 1(\bmod n)$.
Otherwise $dv_{p}\left(m^{d}-1\right)
$$
On the other hand, applying LTE-lemma for $a=m^{d}$ and $b=1$, we have
$$
v_{p}\left(m^{n}-1\right)-v_{p}\left(m^{d}-1\right)=v_{p}(n / d)
$$
It follows $p \mid(n / d)$, hence $p \mid n$.
b) Follows from a) and problem 1.2.
1.16. a) Let $k=v_{p}\left(m^{d}-1\right)$. Then for any integer $a$ we have:
- if $d \nmid a$, then $p \nmid m^{a}-1$ (follows from properties of order modulo $p$ );
- if $d \mid a$, then $v_{p}\left(m^{a}-1\right)=k+v_{p}(a / d)$ follows from LTE-lemma).
Hence $n_{i}=d p^{i}$ is the minimal number $n$ satisfying $p^{k+i} \mid m^{n}-1$.
Let us derive from decomposition of $x^{n}-1$ into cyclotomic polynomials the following formula:
$$
\Phi_{n}=\frac{x^{n}-1}{\operatorname{lcm}\left\{x^{d}-1: d \mid n, d0\right)$. We will rewrite it in the form
$$
P(x)=Q(x)^{2}+r(x)=\left(b_{k} x^{k}+b_{k-1} x^{k-1}+\ldots+b_{0}\right)^{2}+r(x)
$$
where $\operatorname{deg} r(x) \leqslant k-1$. To show this is possible, compare the coefficients of $x^{t+k}$ for $t=0,1, \ldots, k-1$ on both sides: $a_{k+t}=2 b_{k} b_{t}+\sum_{i=1}^{k-1-t} b_{t+i} b_{k-1}$. It is easy to see that the
coefficients $b_{k-1}, b_{k-2}, \ldots, b_{1}, b_{0}$ can be uniquely determined via this system of equations. Moreover, all the $b_{i}$ will be rational. Denote by $M$ the LCM of the denominators of those coeggicients.
By the problem conditions, for every large enough positive integer $n$ we have $P(n)=y_{n}^{2}$, where $y_{n} \in \mathbb{Z}$. Multiplying by $M^{2}$ we get the relation $\left(M y_{n}\right)^{2}=Q_{1}(n)^{2}+M^{2} r(n)$, where $Q_{1}(x)=M Q(x) \in \mathbb{Z}[x]$. If the leading coefficient in $r$ is positive, then for all large enough $n$ we have $\left(M y_{n}\right)^{2}>Q_{1}(n)^{2}$, whence $\left(M y_{n}\right)^{2} \geqslant\left(1+Q_{1}(n)\right)^{2}$. Similarly, if the leading coefficient in $r$ is negative, then $\left(M y_{n}\right)^{2} \leqslant\left(-1+Q_{1}(n)\right)^{2}$. In both cases, the inequality $2\left|Q_{1}(n)\right| \leqslant\left|M^{2} r(n)-1\right|$ holds for all large enoung positive integers $n$, which cannot hold since $\operatorname{deg} Q_{1}>\operatorname{deg} r$.
It follows that $r(x)$ vanishes identically, so that $P(x)=Q(x)^{2}$. It remains to show that the coefficients of $Q$ are all integers; this dollows from problem 2.1.
2.3. In this problem we do not have additional restrictions as in 2.2 ; to work this out, we apply the following interesting trick. It is clear that there exists an integer $\ell$ such that the polynomials $P(x)$ and $P(x+\ell)$ share no common roots (and hence are coprime). Now the polynomial $H(x)=P(x) \cdot P(x+\ell)$ already has an even degree, and its leading coefficient is a perfect square. Also, the values of $R$ at all large enoung positive integers are perfect squares.
By problem 2.2, there exists a polynomial $P_{1}(x) \in \mathbb{Z}[x]$ such that $P(x) \cdot P(x+\ell)=P_{1}(x)^{2}$. Consider its canonical factorization in $\mathbb{Z}[x]$ :
$$
P_{1}=u h_{1} h_{2} \ldots h_{k}
$$
where $u \in \mathbb{Z}$, and the $h_{i}$ are primitive irreducible polynomials. From coprimality of $P(x)$ and $P(x+\ell)$, taking into account Gauss' lemma, it follows that, say, $P(x)$ is a product of some integer $u_{1}$ and squares of some of the $h_{i}$. Since the value of $P$ at some integer point is a nonzero square, the number $u_{1}$ is also a perfect square. Therefore, $P(x)$ is a square of a polynomial with integer coefficients, as desired.
Note that one can solve this problem in another way, repeating the solution of the problem 1.9.
2.4. Notice that each interval of the form $\left(r_{i}, r_{i+1}\right)$ contains a root of the derivative $R^{\prime}(x)=Q^{\prime}(x)$; this way, we have found $l-1$ distinct roots, so these are all the roots of $P^{\prime}$. A similar argument applies to the roots of $S$. Hence, for each $i4$, then $\operatorname{deg}(Q(x) \pm b)>2$, and by the Vieta theorem the sums of the roots of both polynomials must coincide, as well as the sums of pairwise products of the roots. Hence the sums of squares of the roots also must coincide, i.e.,
$$
0^{2}+3^{2}+\ldots+(k-4)^{2}+(k-1)^{2}=1^{2}+2^{2}+\ldots+(k-3)^{2}+(k-2)^{2}
$$
But this equality cannot hold, as
$$
(t-4)^{2}+(t-1)^{2}=2 t^{2}-10 t+17>2 t^{2}-10 t+13=(t-3)^{2}+(t-2)^{2}
$$
for every $t$.
Consequently, $k=4$, and $Q(x)-b=x(x+3), Q(x)+b=(x+1)(x+2)$. Therefore, $b=1$ and $c=b^{2}=1$. So the pair $(k, c)=(4,1)$ is the unique solution.
2.6. The structure of polynomial $P_{k}(x)$ yields that the derivative $P_{k}^{\prime}(x)$ has exactly $k-1$ roots, and those roots can be enumerated as $r_{1}, \ldots, r_{k-1}$ so that
$$
-(k-1)\left|P_{k}\left(r_{i}\right)\right|$ holds. By symmetry, for $i \geqslant \frac{k+2}{2}$ the converse inequality $\left|P_{k}\left(r_{i-1}\right)\right|<\left|P_{k}\left(r_{i}\right)\right|$ holds.
THus, for every complex $a$ at most two roots of equation $P_{k}(x)=a$ belong to the set $\left\{r_{1}, \ldots, r_{m-1}\right\}$ of roots of $P_{k}^{\prime}(x)$. Moreover, if $k$ is odd, then $P_{k}\left(r_{i}\right)=-P_{k}\left(r_{k-i}\right)$, so even at most one of the roots is a root of $P_{k}^{\prime}$. This yields the desired statement, as all the roots of $P_{k}^{\prime}$ are simple.
2.7. Notice that $P_{k}^{\prime}(x)=R^{\prime}(Q(x)) \cdot Q^{\prime}(x)$.
Let $r$ be a complex root of $R^{\prime}(x)$. Then $r$ is a common root of $R^{\prime}(x)$ and $R(x)-R(r)$, hence $x-r \mid \operatorname{gcd}\left(R^{\prime}(x), R(x)-R(r)\right)$. Substituting $Q(x)$ in place of $x$, we obtain $Q(x)-r \mid \operatorname{gcd}\left(R^{\prime}(Q(x)), R(Q(x))-R(r)\right)$. Finally, since $R^{\prime}(Q(x)) \mid P_{k}^{\prime}(x)$, it follows that $Q(x)-r \mid \operatorname{gcd}\left(P_{k}^{\prime}(x), P_{k}(x)-R(r)\right)$.
Problem 2.6 yields now that the degree of $Q(x)-r$ (coinciding with that of $Q(x)$ ) equals 2 , and in this case $k$ must be even. This is exactly what we aimed to prove.
2.8. Solution 1. Due to problem 1.9, we have
$$
P_{k}(x)+c=x(x+1) \ldots(x+k-1)+c=Q(x)^{s}
$$
for some integer $s \geqslant 2$ and some polynomial $Q$ with complex coefficients. Let $a$ be the leading coefficient of $Q$; then, since $a^{s}=1$, one may replace $Q$ by $\frac{1}{a} Q$ not affecting equality (*). Thus we assume that $Q$ is monic, and by problem 2.1, we in fact have $Q \in \mathbb{Z}[x]$. Without loss of generality, we assume that $Q$ is not a non-trivial power of a different polynomial, otherwise we would replace $s$ with a larger number,
Notice that
$$
P_{k}^{\prime}(x)=\left(P_{k}(x)+c\right)^{\prime}=\left(Q(x)^{s}\right)^{\prime}=Q^{\prime}(x) \cdot Q(x)^{s-1}
$$
Hence, $Q^{s-1}$ divides $\operatorname{gcd}\left(P_{k}+c, P_{k}^{\prime}\right)$. But, by problem 2.7, the degree of that g.c.d. does not exceed 2 . Therefore, the only possible cases are (1) $s=3, \operatorname{deg} Q=1 ;(2) s=2, \operatorname{deg} Q=1$; and (3) $s=2, \operatorname{deg} Q=2$.
In case (1) we get that $k=3$ is odd; but in this case, the degree of the g.c.d. should not exceed 1 , thus this case is impossible.
In case (2) we have $x(x+1)+c=(x+a)^{2}$, whence $a=1 / 2$, so that $Q \notin \mathbb{Z}[x]$.
In case (3) we have $x(x+1)(x+2)(x+3)+c=\left(x^{2}+a x+b\right)^{2}$; comparing the coefficients of $x^{3}$ and $x^{2}$, we obtain $a=3, b=1$; therefore, $c=1$, and we arrive at the unique answer $(k, c)=(4,1)$.
2.8. Solution 2. Again, we start with equation $(*)$. where $Q \in \mathbb{Z}[x]$. Plugging $x=0$, we get $c=Q(0)^{s}=b^{s}$. Therefore, we have
$$
x(x+1) \ldots(x+k-1)=Q(x)^{s}-b^{s}
$$
Consequently, $Q(-j)^{s}=b^{s}$ for all $j=0,1, \ldots, k-1$. If $s$ is odd, the polynomial $Q(x)-b$ has at least $k$ distinct roots, namely $0,-1, \ldots,-k+1$, and hence $\operatorname{deg} Q \geqslant k$. But then $\operatorname{deg} Q^{s} \geqslant s k>k=\operatorname{deg}\left(P_{k}+c\right)$, which is impossible.
Assume now that $s$ is even. Then $P_{k}+c$ is a square of a different polynomial; this situation is completely investigated in problem 2.5. So we arrive at the same answer.
2.9. Denote
$$
f(x)=x^{n}+x^{n-1}+\ldots+1=\frac{x^{n+1}-1}{x-1}
$$
Let $\alpha$ be any complex root of $g^{\prime}(x)$. As in the solution of 2.7 , we obtain
$$
h(x)-\alpha \mid \operatorname{gcd}\left(f^{\prime}(x), f(x)-g(\alpha)\right)
$$
Therefore, $\operatorname{deg} \operatorname{gcd}\left(f^{\prime}(x), f(x)-g(\alpha)\right) \geqslant 2$.
It follows that there exist at least two (possibly equal) roots $r, s$ of polynomial
$$
f^{\prime}(x)=\frac{n x^{n+1}-(n+1) x^{n}+1}{(x-1)^{2}}
$$
such that $f(r)=f(s)$. It is easy to notice that 1 is not a root of $f^{\prime}$, and that $f^{\prime}$ has no multiple roots, so that $r \neq s$.
Now the formula for $f^{\prime}$ yields
$$
n\left(s^{n+1}-1\right)=(n+1)\left(s^{n}-1\right) \quad \text { and } \quad n\left(r^{n+1}-1\right)=(n+1)\left(r^{n}-1\right)
$$
Further, from $f(r)=f(s)$ we get
$$
\frac{s^{n+1}-1}{s-1}=\frac{r^{n+1}-1}{r-1}
$$
The previous relations now give us
$$
\frac{s^{n}-1}{s-1}=\frac{r^{n}-1}{r-1}
$$
Subtracting this equality from the previous one, we obtain $s^{n}=r^{n}$, whence $s-1=r-1$, i.e., $s=r$, which is impossible. Thus, no value of $n$ satisfies the problem requirements.
## 3 Miscellaneous
3.1. a) Let us divide $f$ by $g$ with remainder: $f=q g+r$, where $q, r \in \mathbb{Z}[x]$, $\operatorname{deg} r<\operatorname{deg} g$. There exists an integer $N$ such that for all $n>N$ we have $|r(n)|<|g(n)|$. Thus, if $n>N$ satisfies $g(n) \mid f(n)$, then $r(n)=0$. We get that $r(n)=0$ for infinitely many values of $n$, so $r$ is the zero polynomial.
b) After dividing $f$ by $g$ with a remainder in $\mathbb{Q}[x]$, we have $f=\frac{q g}{m}+r$, where $r \in \mathbb{Q}[x]$, $\operatorname{deg} r<\operatorname{deg} g, q \in \mathbb{Z}[x], m \in \mathbb{N}$. Then there exists an integer $N$ such that $|r(n)|<|g(n)| / m$ for all $n>N$, and if $n>N$ is such that $g(n) \mid f(n)$, then $r(n)=0$. We get that $r(n)=0$ for infinitely many values of $n$, so $r$ is the zero polynomial.
Further, assume that $g(n) \mid f(n)$ for some $n \in \mathbb{N}$, i.e., $m \mid q(n)$. Then $m \mid q(n+m t)$ for any integer $t$, and hence $g(n+m t) \mid f(n+m t)$.
3.2. Assume the contrary; then for each $i=1,2, \ldots, m$ there exists an integer $k_{i}$ such that $f\left(k_{i}\right)$ is not divisible by $p_{i}^{\alpha_{i}}$, where $p_{i}$ is a prime, and $p_{i}^{\alpha_{i}}$ is a divisor of $a_{i}$. If $p_{i}=p_{j}$ and $\alpha_{i} \leqslant \alpha_{j}$, then one may redefine $\alpha_{j}$ and $k_{j}$ as $\alpha_{j}=\alpha_{i}$ and $k_{j}=k_{i}$. All such changes having been made, we assume that equal primes $p_{i}$ correspond to equal numbers $\alpha_{i}$ and $k_{i}$. By the Chinese remainders theorem, there exists an integer $k$ in the intersection of progressions of the form $k_{i}+p_{i}^{\alpha_{i}} t$. Then $f(k)$ is not divisible by any of $p_{i}^{\alpha_{i}}$, hence is not divisible by any of $a_{i}$. A contradiction.
3.3. a) Let $f$ be a polynomial determining the given function. Divide it with remainder by the polynomial $x(x-1) \cdots(x-(n-1))$ :
$$
f=x(x-1) \cdots(x-(n-1)) g+r
$$
where $q, r \in \mathbb{Z}[x]$, $\operatorname{deg} r3$ does not satisfy $(*)$.
If $p \equiv 1(\bmod 4)$, take $a=0$, and choose $b$ satisfying $b^{2} \equiv 1(\bmod p)$. Then $a^{2}+a b+b^{2}+1$ is divisible by $p$, while $a-b$ is not.
Suppose now that $p \equiv 3(\bmod 4)$, thus -1 is a non-quadratic residue modulo $n$. It suffices to find an integer $x$ such that $x^{2}+x+1$ is a non-quadratic residue. Indeed, for such $x$, the residue $-\left(x^{2}+x+1\right)^{-1}$ is quadratic, hence the exists a non-zero residue $a$ such that
$$
-1 \equiv a^{2}\left(x^{2}+x+1\right) \equiv(a x)^{2}+a \cdot a x+a^{2}+1 \quad(\bmod p)
$$
then $(a x)^{2}+a \cdot a x+a^{2}+1$ is divisible by $p$, while $a x-a$ is not divisible, except the case $x \equiv 1(\bmod p)$. Finally, if $x \equiv 1(\bmod p)$, then replace it by $x \equiv-2(\bmod p)$ with the same resudue for $x^{2}+x+1$.
Assume, to the contrary, that each residue of the form $x^{2}+x+1$ is either quadratic or zero residue; then the same is true for $4\left(x^{2}+x+1\right)=4(2 x+1)^{2}+3$. Since $2(2 x+1)$ runs over all residues, we obtain the following: if the residue $y$ is either quadratic or zero modulo $p$, then the same is true for $y+3$. But iterating $y \mapsto y+3$ it is possible to obtain all the residues including a non-quadratic residue, which is a contradiction.
3.7. It is clear that $p>2$. We need the following statement:
$$
0^{k}+1^{k}+\ldots+(p-1)^{k} \equiv\left\{\begin{array}{ll}
0, & 1 \leqslant k \leqslant p-2 ; \\
-1, & k=p-1
\end{array} \quad(\bmod p)\right.
$$
For $k=p-1$, it follows from Fermat's theorem. For $kО. Никитенко, А. Петухов, А. Скопенков *
## Содержание
1 Мотивировки и некоторые основные результаты ..... 1
2 Вырожденные матрицы ..... 5
3 Ранг матрицы ..... 6
4 Вложения по модулю 2 графов в поверхности ..... 7
5 Ранг матриц с соотношениями ..... 10
6 Classification of symmetric bilinear forms ..... 11
7 Rank of matrix with relations: generalization ..... 13
## 1 Мотивировки и некоторые основные результаты
Замечание (мотивировки; формально не используется далее)
«Восполнение матриц - задача восстановления недостающих элементов в матрице, известной только частично. Одним из примеров такой задачи является восполнение матрицы с рейтингами фильмов в задаче Нетфликса: дана матрица рейтингов, в которой $i, j$-ый элемент равен оценке, поставленной пользователем $i$ фильму $j$, если такая оценка есть, и пропущен в противном случае; мы хотим восполнить недостающие элементы этой матрицы, чтобы предсказать, насколько пользователям понравятся эти фильмы, и датъ хорошие рекомендации...» [МС] Пропущенные элементы матрицы заполняются так, чтобы минимизировать ранг восполненной матрицы. Все нужные определения (например, ранга) даны далее.
Для простоты мы рассматриваем матрицы над множеством $\mathbb{Z}_{2}=\{0,1\}$ всех остатков по модулю 2 (с операциями сложения и умножения). Мы приведем интересные
*A. Воропаев (Москва). Т. Гараев: Московский Государственный Университет. С. Дженжер, A. Скопенков: Московский Физико-Технический Институт. О. Никитенко: Алтайский Технический Университет (Барнаул). А. Петухов: Институт Проблем Передачи Информации им. А.А. Харкевича (Москва). А. Скопенков: Независимый Московский Университет, https://users .mccme.ru/skopenko/. Благодарим Е. Когана за разрешение использовать его текст [Ko21], В. Ретинского и Я. Абрамова за полезные обсуждения, Д. Деомидова и Ф. Нилова за переводы фрагментов текста, А. Рябичева и издательство МЦНМО за разрешение использовать подготовленные ими рисунки.
элементарные результаты из линейной алгебры. Эти результаты позволяют построить алгоритмы, оценивающие ранг частично заполненных матриц для частного случая восполнения диагонали (предложение 1.1 и теоремы 1.3, 1.4, см. также предложение 1.2). Далее мы рассмотрим более сложную задачу. Вместо соотношений вида $M_{i j}=a_{i j}$ для некоторых элементов $M_{i j}$ матрицы $M$ (где $a_{i j}$ это известные элементы) мы рассмотрим более сложные соотношения на элементы матрицы. Мы оценим минимальный ранг матриц с такими соотношениями (теоремы 5.1, 7.1).
Краткий обзор истории этой и близких задач см. в [MC, NKS] и [Ko21, Remark 4]. Эти результаты имеют приложения к вложениям графов в неориентируемые поверхности (включая вложение по модулю 2 , см. начало §5) и вложениям $k$-мерных «гиперграфов» в $2 k$-мерные поверхности, см. [KS21, KS21e, DS22]. В частности, теоремы 1.3, 1.4 влекут существование полиномиальных (по количеству рёбер) алгоритмов, распознающих «слабую реализуемость» «графов с вращениями» на неориентируемых поверхностях (см. замечание ниже).
Обозначим через $\mathbb{Z}_{2}^{s \times n}=\left(\mathbb{Z}_{2}^{s}\right)^{n}$ множество всех $s \times n$ матриц с элементами из $\mathbb{Z}_{2}$.
Предложение 1.1. (а) Пусть имеется симметричная матрица с элементами из $\mathbb{Z}_{2}$. Тогда следующие условия равносильны:
- можно поменять некоторые элементы на главной диагонали так, чтобы все ненулевые строки матрицы оказались равны;
- нельзя сделать такую одинаковую перестановку строк и столбцов ${ }^{1}$, что в верхнем левом углу полученной матрицы будет стоять подматрица вида:
$$
\left(\begin{array}{ccc}
* & 1 & 1 \\
1 & * & 0 \\
1 & 0 & *
\end{array}\right) \quad \text { или }\left(\begin{array}{cccc}
* & 1 & 0 & 0 \\
1 & * & 0 & 0 \\
0 & 0 & * & 1 \\
0 & 0 & 1 & *
\end{array}\right)
$$
где через * обозначены произвольные (возможно, различные) элементы.
(b) Существует алгоритм сложности $O\left(n^{2}\right)$, который для любой матрицы $M \in$ $\mathbb{Z}_{2}^{n \times n}$ определяет, можно ли поменять некоторые элементы на главной диагонали так, чтобы в получившейся матрице все ненулевые строки были равны.
Часть «только тогда» пункта (а) можно сдавать отдельно.
Алгоритмические результаты в этом тексте могут быть пропущены участниками, ориентированными на теорию, поскольку эти результаты являются простыми следствиями математических результатов. Сложность алгоритма - количество «элементарных» шагов в этом алгоритме. Алгоритм имеет сложность $O(f(n))$, если сложность не превосходит $C f(n)$ для некоторой константы $C>0$ для всех $n$.
Квадратная матрица $M \in \mathbb{Z}_{2}^{n \times n}$ называется вырожденной, если сумма её нескольких различных столбцов (ненулевого количества столбцов) равна нулевому столбцу (т. е. столбцу, состоящему только из нулей). Иначе матрица называется невырожденной. Вводные задачи о вырожденных матрицах, полезные для следующего результата, приведены в $\S 2$.[^1]
Предложение 1.2. (а) Для любой матрицы $M \in \mathbb{Z}_{2}^{n \times n}$ можно поменять несколько элементов на главной диагонали так, чтобы получившаяся матрица была вырожденной.
(b) То се саменой «вырожденной» на «невырожденной».
Рассмотрим матрицу $M \in \mathbb{Z}_{2}^{s \times n}$. Ранг $\operatorname{rk} M$ - максимальное количество столбцов матрицы $M$ таких, что сумма никаких из них не равна нулю. (Это «размерность» «векторного пространства», образованного столбцами матрицы). Вводные задачи на ранг, полезные для следующих результатов, приведены в $\S 3$.
Для $M \in \mathbb{Z}_{2}^{n \times n}$ обозначим через $R(M)$ минимальный из рангов всех матриц, полученных путем изменения некоторых чисел на главной диагонали матрицы $M$.
Матрица называется диагональной, если все её значения вне главной диагонали равны 0.
Теорема 1.3. (a') Чтобы сделать квадратную матрииу ранга $k$ из квадратной матрицъ ранга $n$ изменением некоторых чисел на главной диагонали, необходимо изменить не менее $|n-k|$ чисел.
(а) Для любой невырожденной матрицы $M \in \mathbb{Z}_{2}^{n \times n}$ неравенство $R(M) \leqslant k$ равносильно существованию диагональной матрицы $D$ с не более чем $k$ нулями на главной диагонали такой, что $\operatorname{rk}(M+D) \leqslant k$.
(b) Для любого фиксированного $k$ существует алгоритм сложности $O\left(n^{k+3}\right)$, определяющий для любой матриць $M \in \mathbb{Z}_{2}^{n \times n}$, верно ли, что $R(M) \leqslant k$.
Единичной матрицей $E$ называется диагональная матрица, у которой все значения на диагонали равны 1.
Теорема 1.4. (а) Для любой невырожденной матрицъ $M \in \mathbb{Z}_{2}^{n \times n}$ и диагональной матриц,ы $D \in \mathbb{Z}_{2}^{n \times n}$ верно, что $2 \operatorname{rk}(M+D) \geqslant \operatorname{rk}(M+E)$.
(b) Существует алгоритм сложности $O\left(n^{4}\right)$, который для матрицы $M \in \mathbb{Z}_{2}^{n \times n}$ находит число $k$ такое, что $k / 2 \leqslant R(M) \leqslant k$.
Замечание (определение слабой реализуемости; формально не используется далее) Иероглифом называется (неориентированное циклическое) слово длины $2 n$ из $n$ букв, в котором каждая буква встречается дважды.

Рис. 1: Диск с ленточками, соответствующий иероглифу $a a b b c c$ (слева) и аabcbc (справа)
Возьмем границу выпуклого многоугольника. Отметим на ней непересекающиеся отрезки, отвечающие буквам данного слова, в том порядке, в котором буквы идут в слове. Для каждой буквы соединим (необязательно в плоскости) соответствующие ей два
отрезка ленточкой (т. е. «растянутым» и «помятым» прямоугольником) так, чтобы разные ленточки не пересекались. Ленточки могут быть перекрученными или неперекрученными. Диском с ленточками, отвечающим данному слову, называется объединение построенных (двумерного) выпуклого многоугольника и ленточек.
Назовем иероглиф слабо реализуемым на ленте Мёбиуса, если из неё можно вырезать некоторый диск с ленточками, соответствующий данному иероглифу. Можно аналогично определить слабую реализуемость на бутылке Клейна и других неориентируемых поверхностях.
Две буквы $a, b$ в иероглифе $H$ перекрещиваются в $H$, если они чередуются в циклической последовательности данного иероглифа (т. е. если они идут в циклическом порядке $a b a b$, а не $a a b b)$. Определим матрицу перекрещиваний $M(H) \in \mathbb{Z}_{2}^{n \times n}$ иероглифа $H$ следующим образом. На главной диагонали поставим нули. Поставим единицу в клетку $(i, j)$ для $i \neq j$, если буквы $i, j$ перекрещиваются в $H$, в противном случае поставим ноль.
Верен следующий результат: Иероглиф Н слабо реализуем на ленте Мёбиуса тогда и только тогда, когда $R(M(H)) \leqslant 1$.
См. подробнее [Bi20], [Ko21, Appendix], [Sk20, §2].
## Рекомендации участникам
Если условие задачи является формулировкой утверждения, то в задаче требуется это утверждение доказать. Если задача выделена словом «теорема» («лемма», «следствие» и т.д.), то её утверждение более важное. Как правило, мы приводим (в виде задачи) формулировку красивого или важного утверждения перед его доказательством. В таких случаях для доказательства утверждения могут потребоваться последующие задачи. Мы не лишаем Вас удовольствия самостоятельно найти момент, когда Вы наконец-то сможете доказать такое утверждение. Вообще, если Вы застряли на какой-то задаче, попробуйте перейти к следующим, они могут оказаться полезными. Замечания и задачи, помеченные звёздочками, формально не используются в дальнейшем. В тексте определения важных понятий помечены жирным шрифтом, чтобы затем было проще их найти. Приглашаем Вас обсуждать с жюри возникающие вопросы. Те, кто успешно работают над проектом, завоюют право получить интересные дополнительные задачи для исследования.
Участник (или команда), решающий задачи проекта, получает «боб» за каждое письменное решение для пользователя (не являющееся просто ответом), оцененное в «+»или «+.».
См. рекомендации https://www.mccme.ru/circles/oim/home/pism.pdf
Дополнительные бобы могут выдаваться за красивые решения, решения сложных задач или оформление некоторых решений в системе $\mathrm{T}_{\mathrm{E}} \mathrm{X}$. У жюри бесконечно много бобов. У каждого участника (или каждой команды) в начале один боб. Решения можно сдавать и устно, и письменно для соавтора, отдавая один боб за каждые пять попыток (неважно, удачных или нет).
Пожалуйста, сообщите нам, если Вы знаете решения каких-то из предложенных задач. Это не противоречит Вашему участию в проекте, но это и не обязывает Вас решать этот проект. После проверки у Вас некоторых из задач, названных Вами решёнными заранее, Вы сможете пользоваться результатами всех этих задач. При этом решения этих задач не будут считаться Вашим достижением на ЛКТГ. Зато у Вас появится возможность дойти до более сложных задач. Мы будем рады их выдать, они уже готовы!
(В проекте задачи были разбиты следующим образом: сначала задачи до теоремы 4.3 включительно, затем некоторые решения к простым задачам из $\S \S 1-2$, затем остальные задачи и оставшиеся решения.)
## 2 Вырожденные матрицы
2.1. (а1-а4) Какие из следуюших матрии являются вырожденными?
$$
A_{1}=\left(\begin{array}{ccc}
0 & 0 & 0 \\
0 & 0 & 0 \\
0 & 0 & 0
\end{array}\right), \quad A_{2}=\left(\begin{array}{ccc}
1 & 1 & 1 \\
1 & 1 & 1 \\
1 & 1 & 1
\end{array}\right), \quad A_{3}=\left(\begin{array}{ccc}
1 & 1 & 1 \\
1 & 0 & 1 \\
1 & 1 & 1
\end{array}\right), \quad A_{4}=\left(\begin{array}{ccc}
0 & 1 & 1 \\
0 & 0 & 1 \\
1 & 0 & 0
\end{array}\right)
$$
2.2. (а) Вырожденность матрицы сохраняется при перестановке столбцов (строк).
(b) Вырожденность матрицы сохраняется при добавлении одного столбца (одной строки) к другому (другой).
(с) Любая матрица может быть превращена в диагональную преобразованиями из пунктов ( $a, b)$.
(d) Матрица вырождена тогда и только тогда, когда она не может быть превращена в единичную преобразованиями из ( $a, b$ ).
(f) Квадратная матрица вырождена тогда и только тогда, когда сумма её некоторых строк (ненулевого количества строк) является нулевой строкой.
(g) Существует алгоритм сложности $O\left(n^{3}\right)$, определяющий, является ли матрица $n \times n$ вырожденной.
Для матрицы $M \in \mathbb{Z}_{2}^{n \times n}$ положим $\operatorname{det} M:=0$, если она вырождена, и $\operatorname{det} M:=1$ иначе. Это число называется определителем матрицы $M$. Другое обозначение:
$$
\operatorname{det}\left(\begin{array}{ll}
M_{1,1} & M_{1,2} \\
M_{2,1} & M_{2,2}
\end{array}\right)=\left|\begin{array}{ll}
M_{1,1} & M_{1,2} \\
M_{2,1} & M_{2,2}
\end{array}\right|, \quad \operatorname{det} M=\left|\begin{array}{ccc}
M_{1,1} & \ldots & M_{1, n} \\
\vdots & \ddots & \vdots \\
M_{n, 1} & \ldots & M_{n, n}
\end{array}\right|
$$
2.3. (a) $\left|\begin{array}{ll}a & b \\ c & d\end{array}\right|=a d+b c$.
(b) $\operatorname{det}\left(a_{1}+b_{1}, a_{2}, \ldots, a_{n}\right)=\operatorname{det}\left(a_{1}, a_{2}, \ldots, a_{n}\right)+\operatorname{det}\left(b_{1}, a_{2}, \ldots, a_{n}\right)$. Здесь и ниже $a_{j}, b_{1} \in$ $\mathbb{Z}_{2}^{n}$ это столбиы длины $n$.
(c) $\operatorname{det}\left(a_{1}, \ldots, a_{n}\right)=\sum_{i=1}^{n} a_{i, n} \operatorname{det}\left(a_{1}^{-}, \ldots, a_{i-1}^{-}, a_{i+1}^{-}, \ldots, a_{n}^{-}\right)$, где каждый столбеи, $a_{i}^{-} \in$ $\mathbb{Z}_{2}^{n-1}$ получается из столбца $a_{i}$ удалением последней координаты.
$(d)^{*} \operatorname{det} M=\sum_{\sigma \in S_{n}} \prod_{i=1}^{n} M_{i, \sigma(i)}$, где $S_{n}-$ множество перестановок (т.е. взаимно однозначных соответствий) $\sigma:[n] \rightarrow[n]$.
2.4. (а1-а4) Для каждой матрицы из задачи 2.1 выясните, можно ли изменить некоторые элементы на главной диагонали так, чтобы полученная матрица была вырожденной.
(b1-b4) То же с заменой «вырожденной» на «невырожденной».
Лемма 2.5. (а) Пусть $M \in \mathbb{Z}_{2}^{n \times n}$ - матрица с нулями на главной диагонали. Определим последовательность $M^{(i)}, i=0,1,2, \ldots, n$ рекурсивно.
- $M^{(0)}:=M$;
- $M^{(i)}$ есть результат заменъ в $M^{(i-1)}$ элемента $M_{i, i}^{(i-1)}=0$ на $1+\delta_{i}$, где $\delta_{i}:=$ $\operatorname{det} M_{[i] \times[i]}^{(i-1)}$ есть определитель верхней левой $i \times i$-подматрицъ матриць $M^{(i-1)}$.
Тогда матрица $M^{(n)}$ невырождена.
(b) Существует алгоритм сложности $O\left(n^{4}\right)$, который для матрицы $M \in \mathbb{Z}_{2}^{n \times n}$ находит набор таких значений из $\mathbb{Z}_{2}$, что после подстановки их на главную диагональ матрицы $M$ получится невырожденная матрица.
## 3 Ранг матрицы
3.1. (а1-а4) Найдите $\operatorname{rk} M$ для матриц,
$$
M=\left(\begin{array}{llll}
0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0
\end{array}\right), \quad\left(\begin{array}{llll}
1 & 1 & 1 & 1 \\
1 & 1 & 1 & 1 \\
1 & 1 & 1 & 1
\end{array}\right), \quad\left(\begin{array}{llll}
1 & 1 & 1 & 0 \\
1 & 0 & 1 & 0 \\
1 & 1 & 1 & 1
\end{array}\right), \quad\left(\begin{array}{llll}
0 & 1 & 1 & 1 \\
0 & 0 & 1 & 1 \\
1 & 0 & 0 & 1
\end{array}\right)
$$
(b1-b4) Найдите $R(M)$ для матрии из задачи 2.1.
3.2. Выберем матрицу $M \in \mathbb{Z}_{2}^{s \times n}$.
(а) Из матрицы Мможно выбрать $\mathrm{rk} M$ столбцов так, чтобы любой столбец, был суммой нескольких выбранных столбцов.
(b) Пусть имеется $k$ столбцов (не обязательно матрицы $M$ ) таких, что любой столбец, матриць $M$ есть сумма нескольких из них. Тогда $\operatorname{rk} M \leqslant k$.
(с) Ранг подматрицы не превосходит ранга матрицъ.
3.3. (а) Перестановка столбцов (или строк) не меняет ранга матрицы.
(b) Добавление одного столбиа к другому столбиу (или одной строки к другой) не меняет ранга матрицъ.
(c) Ранг матрицы равен максимальному количеству её строк таких, что сумма никаких из них не равна нулевой строке.
(d) Ранг матриць равен максимальному размеру её невырожденной квадратной подматрицы.
Квадратная матрица, в которую выставлены значения из $\mathbb{Z}_{2}$, называется чётной, если все её значения на главной диагонали равны нулю.
3.4. (а) Для любой $M \in \mathbb{Z}_{2}^{s \times n}$ все ненулевые строки равны тогда и только тогда, когда rk $M \leqslant 1$.
(b) Для симметричной матриць $M \in \mathbb{Z}_{2}^{n \times n}$ все ненулевые строки равны тогда и только тогда, когда, используя одинаковые перестановки строк и столбцов, из неё можно получить матрииу, левый верхний угол которой состоит из единии, а элементы вне этого квадрата равны нулю.
(с) Ранг любой ненулевой чётной симметричной матрицы больше единицы.
3.5. Величина $R(M)$ необязательно сохраняется при
(a) перестановках столбцов;
(b) добавлении одного столбца к другому.
3.6. (а) Существует алгоритм сложности $O\left(n^{3}\right)$, вычисляющий ранг матрицы из $\mathbb{Z}_{2}^{s \times n}, s \leqslant n$.
(b)* Для всякого числа $k$ существует алгоритм сложности $O\left(n^{2}\right)$, проверяющий свойство «rk $M \leqslant k » д л я ~ M \in \mathbb{Z}_{2}^{s \times n}, s \leqslant n$.
Лемма 3.7. Пусть $M, D$ - это матрицы одного размера, в клетках которых стоят элементы $\mathbb{Z}_{2}$. Тогда
(a) $\operatorname{rk}(M+D) \leqslant \operatorname{rk} M+\operatorname{rk} D$;
(b) $\operatorname{rk}(M+D) \geqslant \operatorname{rk} M-\operatorname{rk} D$.
3.8. Существует алгоритм сложности $O\left(n^{k+3}\right)$, находящий по матрище $M \in \mathbb{Z}_{2}^{n \times n}$ диагональную матрицу $D$ такую, что
(a) $\operatorname{rk}(M+D) \leqslant k$,
(b) $\operatorname{rk}(M+D)=k$,
при условии, что такая матрица $D$ существует.
3.9. Для любых ли $m, k \leqslant n$ и матрицы $M \in \mathbb{Z}_{2}^{n \times n}$ ранга $m$ верно следующее: если матрица ранга $k$ может быть получена изменением некоторого количества диагональных элементов матрицы $M$, то это может быть сделано изменением ровно $|m-k|$ элементов?
3.10. (a,b,c) Найдите число матрич, ранга $k$ в $\mathbb{Z}_{2}^{n \times n}$ для $k=0,1,2$.
## 4 Вложения по модулю 2 графов в поверхности
Данный раздел не нужен для понимания дальнейшего текста, однако он дает мотивацию к $\S 5$.
Мы будем обозначать через $S$ либо тор, либо сферу с ручками, либо ленту Мёбиуса, либо бутылку Клейна, либо какую-то другую 2-мерную поверхность. Их простое определение можно найти, например, в §2.1 в
[Sk20]=https://www.mccme.ru/circles/oim/obstruct.pdf
Далее изображения графов на $S$ могут иметь самопересечения. Вложением называется изображение графа без самопересечений.
4.1. Существуют вложения графов (a1,a2,a3) $K_{5}, K_{6}, K_{7}$ в тор;
(b1,b2) $K_{5}, K_{6}$ в ленту Мёбиуса;

(d) $K_{m}$ в сферу с каким-то числом (зависящим от $m$ ) ручек.
Замечание. В данной задаче требуются не строгие доказательства, а большие, понятные и желательно красивые рисунки.
4.2. Граф $K_{5}$ может быть изображён на плоскости так, что изображения (то есть образы) любых двух несмежных рёбер пересекаются в чётном числе точек.
Точкой самопересечения изображения называется точка на изображении, которая отвечает более чем одной точке графа.
Говорят, что изображение графа находится в общем положении, если
- любой точке самопересечения отвечают ровно две точки графа;
- изображение любой вершины не является точкой самопересечения;

Рис. 2: Трансверсальное пересечение и нетрансверсальное пересечение
- изображение графа имеет конечное число точек самопересечения, и
- в любых таких точках самопересечение трансверсальное (рис. 2)2.
Изображение графа в общем положении называется $\mathbb{Z}_{2}$-вложением, если изображения любых двух несмежных рёбер пересекаются в чётном числе точек.
Замечание. Пусть $S$ - либо плоскость, либо тор, либо лента Мёбиуса. Если граф допускает $\mathbb{Z}_{2}$-вложение в $S$, то этот граф допускает вложение в $S$. Тем не менее, существует граф, имеющий $\mathbb{Z}_{2}$-вложение в сферу с 4 ручками, но не вложимый в неё. См. ссылки в [Bi21, Remark 1.3.b,c].
Теорема 4.3. Если граф $К$ допускает $\mathbb{Z}_{2}$-вложение в сферу с $g$ ручками, то
(a) $g \geqslant(m-4) / 3$ для $K=K_{m}$.
(b) $g \geqslant(m-5)^{2} / 16$ для $K=K_{m}$.
(c) $g \geqslant(n-2)^{2} / 4$ для $K=K_{n, n}$.
Идея доказательства теоремы 4.3 заключается в том, что на поверхности, в которую большой граф допускает $\mathbb{Z}_{2}$-вложение, пересечения кривых устроены достаточно сложно (в смысле ранга некоторой матрицы; см. утверждение 4.5). Точнее, теорема 4.3.а [РТ19] следует из теорем 4.4 и 5.1.b вместе с утверждением 4.5 (все ниже). Теорема 4.3.b следует из пункта теоремы 4.3.c (докажите!). Теорема 4.3.c доказана в [FK19], см. структурированное изложение в [DS22]. Аналогично, утверждения 4.5 и 5.2 (вместе с теоремой 4.4 и её неориентированным аналогом) влекут несуществование $\mathbb{Z}_{2}$ вложения $K_{8}$ в тор, и $K_{7}$ в ленту Мёбиуса (или даже $K_{7}$ в бутылку Клейна, чей аналог для невложимости не следует из неравенства Эйлера). Аналогичным образом результат о невложимости в более высоких размерностях следует из теоремы 7.1.
Обозначим через $|X|_{2} \in \mathbb{Z}_{2}$ чётность числа элементов в конечном множестве $X$.
Говорят, что замкнутые кривые $\gamma_{1}, \ldots, \gamma_{p}$ на $S$ находятся в общем положении, если изображение графа (несвязного объединения $p$ циклов), полученное из этих кривых, находится в общем положении. Их $p \times p$-матрица пересечений $G$ определена как
$$
G_{i, j}:= \begin{cases}\left|\gamma_{i} \cap \gamma_{j}\right|_{2}, & \text { если } i \neq j \\ \left|\gamma_{i} \cap \gamma_{i}^{\prime}\right|_{2}, & \text { если } i=j\end{cases}
$$
где $\gamma_{i}^{\prime}$-это кривая, близкая к кривой $\gamma_{i}$ и в общем положении с ней.
Пользуйтесь следующей «гомологической теоремой Бетти» без доказательства.[^2]
Теорема 4.4. Для любых кривых $\gamma_{1}, \ldots, \gamma_{p}$ в общем положении
(a) на сфере с $g$ ручками ранг их матрицы пересечений не превосходит $2 g$.
(b) на диске с $m$ лентами Мёбиуса ранг их матрицв пересечений не превосходит $m$.
Здесь диск с $k$ лентами Мёбиуса - это фигура, изображённая на рис. 1, слева. Более точно, диск с $m$ лентами Мёбиуса - это объединение диска и $m$ попарно непересекающихся ленточек, концы которых приклеены к $2 m$ попарно непересекающимся дугам на граничной окружности диска (ленточки могут не лежать в плоскости диска) так, что
- ориентации концов каждой из ленточек, заданные ориентацией граничной окружности диска, «сохраняют направление вдоль ленточки», и
- ленточки «разделены», т. е. существуют $m$ попарно непересекающихся дуг $A_{i}$ граничной окружности диска такие, что концы $i$-ой ленточки приклеены к двум непересекающимся дугам, содержащимся в $A_{i}, i=1,2, \ldots, m$.
Вы можете делать аппроксимации общего положения на интуитивном уровне.

Рис. 3: Слева: $K_{3}$ и $K_{4}$ на торе. Справа: $K_{5}$ на торе
4.5. Рассмотрим произвольное вложение (или $\mathbb{Z}_{2}$-вложение) $f: K_{n} \rightarrow S$. Рассмотрим произвольное отображение $f^{\prime}: K_{n} \rightarrow S$ в общем положении с $f$ и близкое $\kappa f$. Для любых попарно различных $i, j, k \in[n]$ обозначим через $\langle i j k\rangle$ цикл длины три в $K_{n}$. Положим
$$
i j k \wedge p q r:=\left|f\langle i j k\rangle \cap f^{\prime}\langle p q r\rangle\right|_{2}
$$
Тогда
(4.5.1) $123 \wedge 456=0$.
(4.5.2) $123 \wedge 456+123 \wedge 567+123 \wedge 467+123 \wedge 457=0$.
$123 \wedge 345+123 \wedge 346+123 \wedge 356+123 \wedge 456=0$.
$123 \wedge 234+123 \wedge 235+123 \wedge 245+123 \wedge 345=0$. $123 \wedge 123+123 \wedge 124+123 \wedge 134+123 \wedge 234=0$.
См. рис. 3, слева. Для одной формулы, охватывающей эти четыре, см. свойство линейной зависимости в §5.
(4.5.3) $125 \wedge 345+135 \wedge 245+145 \wedge 235=1$.
См. рис. 3, справа. Подсказка: выведите из (В) ниже.
Замечание. (А) Для любых попарно различных точек $A_{1}, A_{2}, A_{3}, A_{4}$ на прямой существует только одна «чередующаяся» раскраска в два цвета.
(B) Для любых попарно различных точек $A_{1}, A_{2}, A_{3}, A_{4}$ на окружности
$$
\left|A_{1} A_{2} \cap A_{3} A_{4}\right|+\left|A_{1} A_{3} \cap A_{2} A_{4}\right|+\left|A_{1} A_{4} \cap A_{2} A_{3}\right|=1
$$
(B') Для любого отображения $f: K_{5} \rightarrow \mathbb{R}^{2}$ «общего положения» число точек пересечений, образованных образами несмежных рёбер в $\mathbb{R}^{2}$, нечётно.
Простая импликация $(A) \Rightarrow\left(B^{\prime}\right)$ приводится в [Sk14] (в линейном случае; в PL случае импликация аналогична).
В отличие от невложимости, отсутствие $\mathbb{Z}_{2}$-вложимости не следует из неравенства Эйлера [Sk20, §2.4] (как и (B') не следует из формулы Эйлера для планарных графов).
## 5 Ранг матриц с соотношениями
Мы сокращаем $\{i\}$ до $i$. Назовём $\binom{[m]}{3}$-матрицей такую симметричную квадратную матрицу с элементами из $\mathbb{Z}_{2}$, строки и столбцы которой соответствуют всем 3 -элементным подмножествам множества $[m]$, и для которой выполнены следующие свойства:
(тривиальность) $A_{P, Q}=0$, если $P \cap Q=\varnothing$;
(линейная зависимость) для любых 4 -элементного и 3 -элементного подмножеств $F, P \subset[m]$
$$
\sum_{i \in F} A_{F-i, P}=0
$$
(нетривиальность) для любых $i \in[m]$ и 4 -элементного подмножества $F \subset[m]-i$ выполнено $A_{F, i}=1$, где
$$
A_{F, i}:=\sum_{\{X, Y\}: F \cup i=X \cup Y,|X|=|Y|=3, X \cap Y=i} A_{X, Y}=\sum_{\{\sigma, \tau\}: F=\sigma \sqcup \tau,|\sigma|=|\tau|=2} A_{i \sqcup \sigma, i \sqcup \tau}
$$
По утверждению 4.5 , некоторая $\binom{[m]}{3}$-матрица строится по $\mathbb{Z}_{2}$-вложению $f: K_{m} \rightarrow S$ в поверхность. Действительно, положим $A_{\{i, j, k\},\{p, q, r\}}:=i j k \wedge p q r$. Если поверхность $S$ ориентируема, то такая матрица $A$ чётна (т. е., $A_{P, P}=0$ для любого 3 -элементного подмножества $P \subset[m])$.
Теорема 5.1. (а) Если $A$ является $\binom{[m]}{3}$-матрицей, то $\mathrm{rk} A \geqslant \frac{m-4}{3}$.
(b) Более того, если $A$ чётна, то $\operatorname{rk} A \geqslant \frac{2(m-4)}{5}$.
Вы можете отдельно представлять решения следующих частных случаев теоремы 5.1. Более сильные оценки доказаны в $\S 7$.
5.2. (а)* Не существует $\binom{[7]}{3}$-матриць ранга 1.
(b)* Не существует чётной $\binom{[8]}{3}$-матрицъ ранга менее 3.
Теорему 5.1 можно вывести из предложений 5.7.a,b.
Следующее утверждение в дальнейшем не используется. В его доказательстве не обязательно явно предоставлять матрицу, достаточно описать её построение. Мы знаем только решение, использующее утверждения 4.1, 4.5 и теорему 4.4.
5.3. (а) Существует ненулевая ( $\binom{[4]}{3}$-матрица.
(b) Существует $\binom{[5]}{3}$-матрица.
(с) Для любого $m \geqslant 5$ существует $\binom{[m]}{3}$-матрица.
$\left(a^{\prime}, b^{\prime}, c^{\prime}\right)$ То же для чётных матрии.
(d) Существует $\binom{[5]}{3}$-матрица ранга 1.
(е) Существует чётная $\binom{[5]}{3}$-матрица ранга 2.
(f) Существует $\binom{[6]}{3}$-матрица ранга 1.
$(g) *$ Существует $\binom{[8]}{3}$-матриц, ранга более 2.
5.4. Пусть $B$ - квадратная матрица размера $\binom{m-1}{3}$, полученная из $\binom{[m]}{3}$-матрицы удалением строк и столбцов, соответствующих всем подмножествам, содержащим элемент $m$. Тогда $B$ является $\binom{[m-1]}{3}$-матрицей.
5.5. (а) Пусть $B$ - квадратная матрица размера $\binom{m-3}{3}$, полученная из $\binom{[m]}{3}$-матрицы $A$ удалением строк и столбиов, соответствующих подмножествам, содержащим хотя бы один элемент множества $X:=\{m, m-1, m-2\}$. Если $A_{X, X}=1$, mо $\operatorname{rk} A>\operatorname{rk} B$.
(b) Пусть $B$-квадратная матрица, полученная из $\binom{[m]}{3}$-матрицы $A$ удалением строк и столбцов, соответствующих подмножествам, содержащим хотя бы один элемент 3 -элементных подмножеств $X, Y \subset[m]$. Если $A_{X, X}=A_{Y, Y}=0$ и $A_{X, Y}=1$, mo rk $A \geqslant \operatorname{rk} B+2$.
Обозначим через $r_{m}$ минимальный ранг $\binom{[m]}{3}$-матрицы. Обозначим через $\widetilde{r_{m}}$ минимальный ранг чётной $\binom{[m]}{3}$-матрицы. Очевидно, $r_{m}=\widetilde{r_{m}}=0$ для $m \leqslant 4$, и $r_{m} \leqslant \widetilde{r_{m}}$. Нетривиальность означает, что $r_{5}, \widetilde{r_{5}} \geqslant 1$. Теорема 5.1 утверждает, что $r_{m} \geqslant \frac{m-4}{3}$ и $\widetilde{r_{m}} \geqslant \frac{2(m-4)}{5}$.
5.6. (a,b) Haüdume $r_{5}, r_{6}$ и $\widetilde{r_{5}}, \widetilde{r_{6}}, \widetilde{r_{7}}$.
(с) Обе последовательности $r_{m}, \widetilde{r_{m}}$ не убывают.
Предложение 5.7. (a) $r_{m} \geqslant \min \left\{r_{m-3}+1, \widetilde{r_{m}}\right\}$ (точнее, либо $r_{m}=\widetilde{r_{m}}$, либо $r_{m} \geqslant$ $\left.r_{m-3}+1\right)$
(b) $\widetilde{r_{m}} \geqslant \widetilde{r_{m-5}}+2$.
## 6 Classification of symmetric bilinear forms
Fix a symmetric matrix $A \in \mathbb{Z}_{2}^{n \times n}$. For $U, V \in \mathbb{Z}_{2}^{n}$ let
$$
A(U, V)=U \cdot{ }_{A} V:=\sum_{i, j=1}^{n} A_{i, j} U_{i} V_{j} \quad\left(=U^{T} A V\right)
$$
A basis of $\mathbb{Z}_{2}^{n}$ is an inclusion-minimal ordered set of vectors such that every vector from $\mathbb{Z}_{2}^{n}$ is the sum of some vectors from this set.
Теорема 6.1. For $n=2$ there is a basis $X_{1}, X_{2}$ of $\mathbb{Z}_{2}^{2}$ and numbers $\gamma_{1}, \gamma_{2} \in \mathbb{Z}_{2}$ such that either
(i) for any $a_{1}, a_{2}, b_{1}, b_{2} \in \mathbb{Z}_{2}$ we have
$$
\left(a_{1} X_{1}+a_{2} X_{2}\right) \cdot{ }_{A}\left(b_{1} X_{1}+b_{2} X_{2}\right)=\gamma_{1} a_{1} b_{1}+\gamma_{2} a_{2} b_{2}, \quad \text { or }
$$
(ii) for any $a_{1}, a_{2}, b_{1}, b_{2} \in \mathbb{Z}_{2}$ we have
$$
\left(a_{1} X_{1}+a_{2} X_{2}\right) \cdot{ }_{A}\left(b_{1} X_{1}+b_{2} X_{2}\right)=a_{1} b_{2}+a_{2} b_{1}
$$
Recall that problems stated after theorems are hints to proofs of the theorems.
6.2. Assume that $n=2, X \in \mathbb{Z}_{2}^{2}$ and $X \cdot{ }_{A} X=1$.
(a) For any $P \in \mathbb{Z}_{2}^{2}$ there is $\lambda_{X, P} \in \mathbb{Z}_{2}$ such that for $P_{X}:=P+\lambda_{X, P} X$ we have $P_{X} \cdot{ }_{A} X=0$.
(b) There is a basis $X_{1}=X$, $X_{2}$ of $\mathbb{Z}_{2}^{2}$ and numbers $\gamma_{1}=1, \gamma_{2} \in \mathbb{Z}_{2}$ such that the property (i) of Theorem 6.1 holds.
6.3. Assume that $n=2, X, Y \in \mathbb{Z}_{2}^{2}$ and $X \cdot{ }_{A} Y=1, X \cdot{ }_{A} X=Y \cdot{ }_{A} Y=0$. Then $X_{1}:=X$, $Y_{1}:=Y$ is a basis of $\mathbb{Z}_{2}^{2}$ such that the property (ii) of Theorem 6.1 holds.
Теорема 6.4. For $n=3$ there is a basis $X_{1}, X_{2}, X_{3}$ of $\mathbb{Z}_{2}^{3}$ and numbers $\gamma_{1}, \gamma_{2}, \gamma_{3} \in \mathbb{Z}_{2}$ such that either
(i) for any $a_{1}, a_{2}, a_{3}, b_{1}, b_{2}, b_{3} \in \mathbb{Z}_{2}$ we have
$$
\left(a_{1} X_{1}+a_{2} X_{2}+a_{3} X_{3}\right) \cdot{ }_{A}\left(b_{1} X_{1}+b_{2} X_{2}+b_{3} X_{3}\right)=\gamma_{1} a_{1} b_{1}+\gamma_{2} a_{2} b_{2}+\gamma_{3} a_{3} b_{3}, \quad \text { or }
$$
(ii) for any $a_{1}, a_{2}, a_{3}, b_{1}, b_{2}, b_{3} \in \mathbb{Z}_{2}$ we have
$$
\left(a_{1} X_{1}+a_{2} X_{2}+a_{3} X_{3}\right) \cdot{ }_{A}\left(b_{1} X_{1}+b_{2} X_{2}+b_{3} X_{3}\right)=a_{1} b_{2}+a_{2} b_{1}+\gamma_{3} a_{3} b_{3}
$$
6.5. Assume that $X, Y \in \mathbb{Z}_{2}^{3}$ and $X \cdot{ }_{A} Y=1, X \cdot{ }_{A} X=Y \cdot{ }_{A} Y=0$.
(a) For any $P \in \mathbb{Z}_{2}^{3}$ there are $\lambda_{X, Y, P}, \lambda_{Y, X, P} \in \mathbb{Z}_{2}$ such that for $P_{X, Y}:=P+\lambda_{X, Y, P} Y+$ $\lambda_{Y, X, P} X$ we have $P_{X, Y} \cdot{ }_{A} X=P_{X, Y} \cdot{ }_{A} Y=0$.
(b) There is a basis $X_{1}=X, X_{2}=Y, X_{3}$ of $\mathbb{Z}_{2}^{3}$ and a number $\gamma_{3} \in \mathbb{Z}_{2}$ such that the property (ii) of Theorem 6.4 holds.
Теорема 6.6. There are $k, l$ and a basis $X_{1}, Y_{1}, \ldots, X_{k}, Y_{k}, Z_{1}, \ldots, Z_{n-2 k}$ of $\mathbb{Z}_{2}^{n}$ such that $2 k+l \leqslant n$ and for any $a, a^{\prime}, b, b^{\prime} \in \mathbb{Z}_{2}^{k}$ and $c, c^{\prime} \in \mathbb{Z}_{2}^{n-2 k}$ we have
$$
\begin{gathered}
\left(a_{1} X_{1}+b_{1} Y_{1}+\ldots+a_{k} X_{k}+b_{k} Y_{k}+c_{1} Z_{1}+\ldots+c_{n-2 k} Z_{n-2 k}\right) \cdot A \\
\cdot_{A}\left(a_{1}^{\prime} X_{1}+b_{1}^{\prime} Y_{1}+\ldots+a_{k}^{\prime} X_{k}+b_{k}^{\prime} Y_{k}+c_{1}^{\prime} Z_{1}+\ldots+c_{n-2 k}^{\prime} Z_{n-2 k}\right)= \\
\quad=a_{1} b_{1}^{\prime}+a_{1}^{\prime} b_{1}+\ldots+a_{k} b_{k}^{\prime}+a_{k}^{\prime} b_{k}+c_{1} c_{1}^{\prime}+\ldots+c_{l} c_{l}^{\prime}
\end{gathered}
$$
If $A$ is even, then $l=0$.
6.7. Assume that $X \in \mathbb{Z}_{2}^{n}$ and $X \cdot{ }_{A} X=1$.
(a) State and prove the n-dimensional analogue of Assertion 6.2.a.
(b) There is a basis $X, E_{1}, \ldots, E_{n-1}$ of $\mathbb{Z}_{2}^{n}$ and a symmetric matrix $B \in \mathbb{Z}_{2}^{(n-1) \times(n-1)}$ such that for any $a, b \in \mathbb{Z}_{2}$ and $\lambda, \mu \in \mathbb{Z}_{2}^{n-1}$ we have
$$
\left(a X+\lambda_{1} E_{1}+\ldots+\lambda_{n-1} E_{n-1}\right) \cdot{ }_{A}\left(b X+\mu_{1} E_{1}+\ldots+\mu_{n-1} E_{n-1}\right)=a b+\lambda \cdot_{B} \mu
$$
6.8. Assume that $X, Y \in \mathbb{Z}_{2}^{n}$ and $X \cdot{ }_{A} Y=1, X \cdot{ }_{A} X=Y \cdot{ }_{A} Y=0$.
(a) State and prove the n-dimensional analogue of Assertion 6.5.a.
(b) There is a basis $X, Y, E_{1}, \ldots, E_{n-2}$ of $\mathbb{Z}_{2}^{n}$ and a symmetric matrix $B \in \mathbb{Z}_{2}^{(n-2) \times(n-2)}$ such that for any $a_{X}, a_{Y}, b_{X}, b_{Y} \in \mathbb{Z}_{2}$ and $\lambda, \mu \in \mathbb{Z}_{2}^{n-2}$ we have $\left(a_{X} X+a_{Y} Y+\lambda_{1} E_{1}+\ldots+\lambda_{n-2} E_{n-2}\right) \cdot{ }_{A}\left(b_{X} X+b_{Y} Y+\mu_{1} E_{1}+\ldots+\mu_{n-2} E_{n-2}\right)=a_{X} b_{Y}+a_{Y} b_{X}+\lambda \cdot{ }_{B} \mu$.
## 7 Rank of matrix with relations: generalization
The following results are 'higher-dimensional' (and more strong) generalizations of Theorem 5.1, Assertions 5.4 and 5.5, and Proposition 5.7. They give a simplified well-structured exposition of [PT19, Theorem 1].
An $\binom{[m]}{l}$-matrix is a symmetric square matrix with $\mathbb{Z}_{2}$-entries whose rows and whose columns correspond to all $l$-element subsets of $[m]$, and for which (triviality) and the following properties hold:
(linear dependence) for each $(l+1)$-element and $l$-element subsets $F, P \subset[m]$
$$
\sum_{i \in F} A_{F-i, P}=0
$$
(non-triviality) for each $i \in[m]$ and $(2 l-2)$-element subset $F \subset[m]-i$ we have $A_{F, i}=1$, where
$$
A_{F, i}:=\sum_{\{X, Y\}: F \cup i=X \cup Y, X \cap Y=i,|X|=|Y|=l} A_{X, Y}=\sum_{\{\sigma, \tau\}: F=\sigma \sqcup \tau,|\sigma|=l-1} A_{i \sqcup \sigma, i \sqcup \tau}
$$
Analogously to Assertion 4.5, an $\binom{[m]}{l}$-matrix is constructed by a $\mathbb{Z}_{2}$-embedding of the $(l-1)$-dimensional skeleton of the $(m-1)$-dimensional simplex to a $2(l-1)$-dimensional manifold.
Теорема 7.1. Suppose $l \geqslant 3$ and $A$ is an $\binom{[m]}{l}$-matrix.
(a) Then $\operatorname{rk} A \geqslant \frac{m-2 l+2}{l-1}$. (b) If, moreover, $A$ is even, then $\operatorname{rk} A \geqslant \frac{2(m-2 l+2)}{l}$.
You can deduce Theorem 7.1 from Propositions 7.4.a,b.
7.2. Let $A^{\prime}$ be the square matrix of size $\binom{m-1}{l}$ obtained from an $\binom{[m]}{l}$-matrix by deleting rows and columns corresponding to all subsets containing $m$. Then $A^{\prime}$ is an $\binom{[m-1]}{l}$-matrix.
7.3. Let $A$ be an $\binom{[m]}{l}$-matrix and $X:=\{m-l+1, m-l+2, \ldots, m\}$.
$\left(a, b\right.$ ') Let $B$ be the square matrix of size $\binom{m-l}{l}$ obtained from $A$ by deleting rows and columns corresponding to subsets containing at least one of the elements of $X$.
If $A_{X, X}=1$, then $\operatorname{rk} A>\operatorname{rk} B$.
If $A_{X, X}=A_{Y, Y}=0$ and $A_{X, Y}=1$ for some $Y \subset[m]$, then $\operatorname{rk} A \geqslant \operatorname{rk} B+2$.
(b) Let $C$ be the square matrix obtained from $A$ by deleting rows and columns corresponding to subsets containing at least one element of $X$ or of certain l-element subset $Y \subset[m]$. If $A_{X, X}=A_{Y, Y}=0$ and $A_{X, Y}=1$, then $\operatorname{rk} A \geqslant \operatorname{rk} C+2$.
(a') For l-element subsets $P, Q \subset[m-l+1]$ define
$$
D_{P, Q}:=A_{P, Q}+A_{P, X} A_{Q, X}
$$
If $A_{X, X}=1$, then $\operatorname{rk} D<\operatorname{rk} A$ and $D$ is an $\binom{[m-l+1]}{l}$-matrix.
Assertions 7.3.a,b are only required to illustrate the idea of Assertions 7.3.a',b' by proving much easier results giving estimates $\operatorname{rk} A \geqslant \frac{m-2 l+2}{l}$ and, for $A$ even, $\operatorname{rk} A \geqslant \frac{2(m-2 l+2)}{2 l-1}$.
Denote by $r_{m}$ the minimal rank of an $\binom{[m]}{l}$-matrix. Denote by $\widetilde{r_{m}}$ the minimal rank of an even $\binom{[m]}{l}$-matrix. Clearly, $r_{m}=\widetilde{r_{m}}=0$ for $m \leqslant 2 l-2$, both sequences $r_{m}, \widetilde{r_{m}}$ are non-decreasing, and $r_{m} \leqslant \widetilde{r_{m}}$. The non-triviality implies that $r_{2 l-1}, \widetilde{r_{2 l-1}} \geqslant 1$. Theorem 7.1 asserts that $r_{m} \geqslant \frac{m-2 l+2}{l-1}$ and $\widetilde{r_{m}} \geqslant \frac{2(m-2 l+2)}{l}$.
Предложение 7.4. (a) $r_{m} \geqslant \min \left\{r_{m-l+1}+1, \widetilde{r_{m}}\right\}$ (more precisely, either $r_{m}=\widetilde{r_{m}}$, or $\left.r_{m} \geqslant r_{m-l+1}+1\right)$
(b) $\widetilde{r_{m}} \geqslant \widetilde{r_{m-l}}+2$.
Proof of Proposition 7.4.a also uses an algebraic version (b) of the higher-dimensional analogue of the following result (a).
Предложение 7.5. (a) Denote by $X=\left(\begin{array}{c}\binom{[5]}{2}\end{array}\right)$ the set of unordered pairs of 2-element subsets of [5]. For any $i \in[5]$ and a partition [5] $-i=\sigma \sqcup \tau$ into disjoint 2-element sets denote
$$
T_{i,\{\sigma, \tau\}}:=\{\{\alpha, \beta\} \in X: \alpha \subset \sigma \sqcup i, \beta \subset \tau \sqcup i\}
$$
Denote by $A_{i}$ the sum modulo 2 (i. e., the symmetric difference) of sets $T_{i,\{\sigma, \tau\}}$ over all non-ordered partitions [5] $-i=\sigma \sqcup \tau$ as above. Then
$$
A_{i}=\{\{\alpha, \beta\} \in X: \alpha \cap \beta=\varnothing\}
$$
and so is independent of $i$.
(b) Let $A$ be a symmetric square matrix with $\mathbb{Z}_{2}$-entries whose rows and whose columns correspond to all l-element subsets of $[m]$. If A satisfies the linear dependence property (from the definition of an $\binom{[m]}{l}$-matrix), then $A_{F, i}$ depends only on $F \sqcup i$ not on $(F, i)$.
## Указания и решения к $\S \S 1-7$
Доказательство предложения 1.1 см. в [Ві20].
1.2. (а) Изменим числа на главной диагонали матрицы $M$ таким образом, чтобы сумма элементов в каждой строке стала чётной. Тогда получившаяся матрица будет вырожденной.
(b) Используйте индукцию. См. лемму 2.5.a.
1.3. (а) Любая матрица с коэффициентами из $\mathbb{Z}_{2}$, полученная из матрицы $M$ заменой элементов на главной диагонали, может быть представлена единственным образом как $M+D$, где $D$ - диагональная матрица. По лемме 3.7.b, для всякой диагональной матрицы $D$ с более чем $k$ нулями на главной диагонали, мы имеем $\operatorname{rk}(M+D) \geqslant$ $\operatorname{rk} M-\operatorname{rk} D>n-(n-k)=k$.
(b) Алгоритм (b) строится с помощью (a) и леммы 2.5.b. Заметим, что алгоритм леммы 2.5.b имеет сложность $O\left(n^{4}\right)$. Легко видеть, что перебор всех диагональных матриц $n \times n$ с не более $k$ нулями на главной диагонали использует
$$
O\left(n\binom{n}{0}+n\binom{n}{1}+\ldots+n\binom{n}{k}\right) \stackrel{(*)}{=} O\left((k+1) n\binom{n}{k}\right)=O\left(n \cdot n^{k}\right)=O\left(n^{k+1}\right)
$$
операций. Здесь равенство $(*)$ выполняется, так как можно считать, что $n \geqslant 2 k$. Таким образом, ввиду утверждения 3.6.b, сложность всего алгоритма равна $O\left(n^{4}\right)+O\left(n^{k+1} n^{2}\right)=$ $O\left(n^{k+3}\right)$ (т. к. $k \geqslant 1$ ).
1.4. (a) Обозначим через $n$ число столбцов в $M$ и в $D$. По лемме 3.7.b мы имеем
$$
\begin{aligned}
& 2 \operatorname{rk}(M+D)=\operatorname{rk}(M+D)+\operatorname{rk}((M+E)+(E+D)) \geqslant \\
& \geqslant(\operatorname{rk} M-\operatorname{rk} D)+(\operatorname{rk}(M+E)-\operatorname{rk}(E+D))= \\
& \quad=n-\operatorname{rk} D+\operatorname{rk}(M+E)-(n-\operatorname{rk} D)=\operatorname{rk}(M+E)
\end{aligned}
$$
(b) Обозначим через $M_{n}$ матрицу, полученную применением алгоритма леммы 2.5.b к матрице $M$. Положим $k:=\operatorname{rk}\left(M_{n}+E\right)$. Тогда $R(M)=\operatorname{rk}(M+D)$ для некоторой диагональной матрицы $D$. Вместе с пунктом (а) это влечёт $k / 2 \leqslant R(M) \leqslant k$, что и требовалось.
Число $k$ может быть вычислено за $O\left(n^{3}\right)$ операций. Таким образом, общая сложность алгоритма равна $O\left(n^{4}\right)+O\left(n^{3}\right)=O\left(n^{4}\right)$.
2.1. Oтвет: Матрицы $A_{1}, A_{2}, A_{3}$ вырождены. Матрица $A_{4}$ невырождена.
2.2. Указание: в (a)-(b) выделите максимальную невырожденую подматрицу; в (c) воспользуйтесь индукцией.
Пункт (а) понятен.
Зафиксируем матрицу $M$.
(b) Для матрицы $M^{\prime}$ обозначим через row $_{i \rightarrow i+j} M^{\prime}$ модификацию матрицы $M^{\prime}$, при котором $i$-ая строка переходит в сумму $i$-ой и $j$-ой строки; $\operatorname{col}_{i \rightarrow i+j} M^{\prime}$ определяются аналогично.
Для доказательства пункта (b) достаточно показать, что $M$ вырождена тогда и только тогда, когда $\operatorname{row}_{i \rightarrow i+j} M$ вырождена и тогда и только тогда, когда $\operatorname{row}_{i \rightarrow i+j} M$ вырождена.
Далее, заметим, что $\operatorname{row}_{i \rightarrow i+j} \operatorname{row}_{i \rightarrow i+j} M=M=\operatorname{col}_{i \rightarrow i+j} \operatorname{col}_{i \rightarrow i+j} M$. Таким образом, достаточно показать, что если $M$ вырождена, то и $\operatorname{row}_{i \rightarrow i+j} M$, и $\operatorname{col}_{i \rightarrow i+j} M$ вырождены.
Предположим, что сумма столбцов с номерами $c_{1}, c_{2}, \ldots, c_{s}$ в $M$ равна 0 . Тогда сумма столбцов с теми же номерами в $\operatorname{row}_{i \rightarrow i+j} M$ равна 0 . Если $i \notin\left\{c_{1}, \ldots, c_{s}\right\}$, то сумма столбцов с теми же номерами в матрице $\operatorname{col}_{i \rightarrow i+j} M$ равна 0 . Если $i, j \in\left\{c_{1}, \ldots, c_{s}\right\}$, то сумма столбцов с номерами $\left\{c_{1}, \ldots, c_{s}\right\} \backslash j$ в матрице $\operatorname{col}_{i \rightarrow i+j} M$ равна 0 . Если $i \in$ $\left\{c_{1}, \ldots, c_{s}\right\}, j \notin\left\{c_{1}, \ldots, c_{s}\right\}$, тогда сумма столбцов с номерами $\left\{c_{1}, \ldots, c_{s}\right\} \cup\{j\}$ в матрице $\mathrm{col}_{i \rightarrow i+j} M$ равна 0 . Это доказывает (b).
(c) Мы явно покажем, как матрицу $M$ привести к диагональному виду.
Если все элементы в матрице $M$ равны 0 , то $M$ диагональна. Пусть в матрице $M$ есть ненулевой элемент. Переставим строку с этим элементом с верхней строкой, и столбец с этим элементом с левым столбцом. Прибавим верхнюю строку полученной матрицы к остальным строкам с ненулевым левым элементом. Аналогично прибавим левый столбец полученной матрицы к остальным столбцам с ненулевым верхним элементом. В полученной матрице в левом столбце и верхней строке все элементы нулевые, кроме левого верхнего. Удалим из полученной матрицы верхнюю строку и левый столбец.
Повторим описанную процедуру индуктивно к полученной подматрице. В итоге $M$ приведется к диагональной матрице.
(d) С помощью п. (с) мы можем перевести матрицу $M$ в диагональную матрицу, пользуясь преобразованиями из п. (a,b); также матрица $M$ вырождена тогда и только тогда, когда полученная матрица диагональна. Остается заметить, что диагональная матрица невырождена тогда и только тогда, когда она равна единичной.
(f) Следует из прямых аналогов пунктов (a)-(d) для строк.
(g) Алгоритм строится по пункту (c). У алгоритма есть $n$ больших шагов, один из которых описан во втором абзаце решения пункта (с). Каждый большой шаг содержит не более одной перестановки строк, не более одной одной перестановки столбцов и до $2 n$ прибавлений строк и столбцов. Таким образом, сложность всего алгоритма равна $O(n)+n \cdot O\left(n^{2}\right)=O\left(n^{3}\right)$.
2.3. (а) Формула справедлива, так как матрица из $\mathbb{Z}_{2}^{2 \times 2}$ вырождена тогда и только
тогда, когда или в ней есть нулевая строка, или нулевой столбец, или в когда в ней совпадают строки и совпадают столбцы (в последнем случае все элементы матрицы единицы).
Другое решение. Ниже приведины все матрицы из $\mathbb{Z}_{2}^{2 \times 2}$ с точностью до перестановки строк и столбцов:
$$
\left(\begin{array}{ll}
1 & 0 \\
0 & 1
\end{array}\right),\left(\begin{array}{ll}
1 & 1 \\
0 & 1
\end{array}\right),\left(\begin{array}{ll}
0 & 0 \\
0 & 0
\end{array}\right),\left(\begin{array}{ll}
1 & 1 \\
1 & 1
\end{array}\right),\left(\begin{array}{ll}
1 & 1 \\
0 & 0
\end{array}\right),\left(\begin{array}{ll}
1 & 0 \\
1 & 0
\end{array}\right),\left(\begin{array}{ll}
1 & 0 \\
0 & 0
\end{array}\right)
$$
Первые две матрицы невырождены, в то время как оставшиеся вырождены. Легко проверить формулу на данных матрицах.
(d) Следует из пунктов (b,c).
Далее приводим альтернативное прямое решение.
Рассмотрим шахматную доску размера $n \times n$. Правилъной расстановкой ладей для данной шахматной доски называется такая расстановка $n$ ладей на доске, что они не бьют друг друга. Правильной $M$-расстановкой ладей для данной шахматной доски называется такая правильная расстановка ладей, что все ладьи стоят на клетках, отвечающих единичным элементам матрицы $M$.
Обозначим через $\operatorname{det}^{*} M$ чётность числа правильных $M$-расстановок ладей. Тогда (d) можно переформулировать так: $\operatorname{det} M=\operatorname{det}^{*} M$. Равенство следует из того, что
- преобразования из 2.2.a, 2.2.b coxраняют $\operatorname{det}^{*} M$ и
- $\operatorname{det} M^{\prime}=\operatorname{det}^{*} M^{\prime}$ для диагональной матрицы $M^{\prime}$.
2.4. Для любой невырожденной матрицы из задачи 2.1 мы показали, как поменять элементы на главной диагонали, что бы сделать её невырожденной; покажем обратное для $A_{4}$ :
$$
A_{1} \rightarrow\left(\begin{array}{ccc}
1 & 0 & 0 \\
0 & 1 & 0 \\
0 & 0 & 1
\end{array}\right), \quad A_{2}, A_{3} \rightarrow\left(\begin{array}{ccc}
0 & 1 & 1 \\
1 & 0 & 1 \\
1 & 1 & 1
\end{array}\right), \quad A_{4} \rightarrow\left(\begin{array}{ccc}
0 & 1 & 1 \\
0 & 1 & 1 \\
1 & 0 & 0
\end{array}\right)
$$
2.5. (а) В следующем абзаце мы докажем индукцией по $i \geqslant 1$, что определитель $\Delta_{i}:=\operatorname{det} M_{[i] \times[i]}^{(i)}$ левой верхней $i \times i$-угловой подматрицы матрицы $M^{(i)}$ равен 1 . Таким образом, $\operatorname{det} M^{(n)}=\Delta_{n}=1$.
База $i=1$ выполняется, поскольку $\Delta_{1}=1+\delta_{0}=1$. Докажем шаг индукции $i-1 \rightarrow i$. Применим формулу разложения определителя $\Delta_{i}$ по последней строке соответствующей подматрицы матрицы $M^{(i)}$ (утверждение 2.3.c). Поскольку $M_{i, i}^{(i-1)}=M_{i, i}=0$ и $\Delta_{i-1}=1$, мы получаем $\Delta_{i}=\delta_{i}+\left(1+\delta_{i}\right) \Delta_{i-1}=1$.
(b) Алгоритм строится по пункту (a). Алгоритм по сути является вычислением определителей $n$ квадратных подматриц размеров $1,2, \ldots, n$. Значит, по утверждению $2.2 . \mathrm{g}$ его сложность равна $O\left(1^{3}+2^{3}+\ldots+n^{3}\right)=O\left(n \cdot n^{3}\right)=O\left(n^{4}\right)$.
3.1. Ответы: (a1) 0; (a2) 1; (a3), (a4) 3; (b1) 0; (b2), (b3) 1 ; (b4) 2 .
3.2 Указание к пункту ( $\mathrm{b}$ ): найдите число различных сумм из $k$ столбцов.
Пункт (а) следует из определения ранга матрицы.
(b) По определению ранга, число различных сумм из столбцов матрицы $M$ равно $2^{\mathrm{rk} M}$. С другой стороны, число таких сумм не превосходит $2^{k}$. Следовательно, $2^{k} \geqslant 2^{\text {rk } M}$ и $k \geqslant \operatorname{rk} M$.
3.3. Доказательства пунктов (a), (b) аналогичны доказательствам пунктов 2.2.a,b.
3.4. Пункт (а) ясен.
(b) Если для ненулевой симметричной матрицы $M$ существует такая перестановка столбцов и строк, то $\mathrm{rk} M=1$ по утверждению 3.2.b.
Возьмём симметричную матрицу $M$ ранга 1 . В качестве требуемой перестановки можно взять любую перестановку, отображающую ненулевые строки матрицы $M$ в первую строку. Действительно, возьмём любые ненулевые строки с номерами $i, j$. Если $M_{i, j}=0$, то существует такая ненулевая строка с номером $k$, для которой $M_{i, k}=1$. Тогда строки с номерами $j$ и $k$ являются различными ненулевыми строками матрицы $M$ ранга 1. Противоречие. Следовательно, $M_{i, j}=1$.
(c) Возьмите ненулевые строки и примените аргументы из предыдущего пункта.
3.5. (a) $R\left(\begin{array}{lll}1 & 0 & 1 \\ 0 & 0 & 1 \\ 0 & 1 & 0\end{array}\right)=2, \quad R\left(\begin{array}{lll}1 & 0 & 1 \\ 0 & 1 & 0 \\ 0 & 0 & 1\end{array}\right)=1$.
(b) $R\left(\begin{array}{lll}1 & 0 & 1 \\ 0 & 1 & 1 \\ 0 & 1 & 0\end{array}\right)=2, \quad R\left(\begin{array}{lll}1 & 0 & 1 \\ 0 & 1 & 0 \\ 0 & 0 & 1\end{array}\right)=1$.
3.6. Подсказка к пункту (а): см. утверждение 2.2 .
(a) Алгоритм из доказательства утверждения 2.2.c обеспечивает диагональную матрицу того же ранга и имеет требуемую сложность. Ранг диагональной матрицы равен количеству ненулевых элементов в ней.
(b) Нужно построить такое множество $S_{k}$ из столбцов, что
- эти столбцы составляют невырожденную подматрицу;
- первые $k$ столбцов матрицы $M$ являются суммами некоторых столбцов из множества $S_{k}$.
Если $\left|S_{k}\right|>r$ для некоторого $k=1, \ldots, n$, то ответ «нет». Если для любого $k=$ $1, \ldots, n$ справедливо неравенство $\left|S_{k}\right| \leqslant r$, то ответ «да».
Ответ корректен, так как $\left|S_{1}\right| \leqslant\left|S_{2}\right| \leqslant \ldots \leqslant\left|S_{n}\right|$, и $\left|S_{n}\right|>r$ равносильно неравенству rk $M>r$.
Положим $S_{1}:=\varnothing$, если первый столбец матрицы $M$ равен 0 , и $S_{1}:=\{1\}$ иначе.
Определим $S_{k+1}$ по $S_{k}$. Составим множество всех сумм столбцов матрицы $M$ с индексами из $S_{k}$ (требуется $O(n)$ операций, так как $\left|S_{k}\right| \leqslant r$ ). Затем сравним $(k+1)$-ый столбец матрицы $M$ со всеми суммами из этого множества (на это потребуется не более $2^{r} O(n)=O(n)$ операций). Если $(k+1)$-ый столбец матрицы $M$ равен хотя бы одной сумме, то $S_{k+1}:=S_{k}$. Иначе $S_{k+1}:=S_{k} \cup\{k+1\}$.
Легко проверить, что общая сложность алгоритма равна $O\left(n^{2}\right)$.
3.7. Пункт (b) следует из пункта (a), поскольку
$$
\operatorname{rk} M=\operatorname{rk}(M+D+D) \leqslant \operatorname{rk}(M+D)+\operatorname{rk} D \Longrightarrow \operatorname{rk}(M+D) \geqslant \operatorname{rk} M-\operatorname{rk} D
$$
Теперь мы докажем пункт (а). Выберем столбцы из утверждения 3.2.а для матрицы $M$ и матрицы $D$. Тогда каждый столбец матрицы $M+D$ является суммой нескольких из выбранных $\operatorname{rk} M+\operatorname{rk} D$ столбцов. По утверждению 3.2.b $\operatorname{rk}(M+D) \leqslant \operatorname{rk} M+\operatorname{rk} D$.
3.8. Утверждение и доказательство утверждения 3.8 аналогичны утверждению и доказательству теоремы 1.3.b.
3.9. Ответ: нет. Для матрицы $M$ ниже и $k=1$ утверждение неверно.
$$
\left(\begin{array}{lll}
0 & 1 & 1 \\
1 & 0 & 1 \\
1 & 1 & 0
\end{array}\right)
$$
3.10. Ответы: (a) 1 ; (b) $\left(2^{n}-1\right)^{2}$; (c) $\left(2^{n}-1\right)^{2}\left(2^{n}-2\right)^{2} / 6$.
(a) Существует ровно одна матрица ранга 0 : матрица, у которой все элементы нулевые.
(b) Для матрица ранга 1 все столбцы, содержащие ненулевой элемент, совпадают. Следовательно, такие матрицы находятся во взаимно однозначном соответствии с упорядоченными парами, образованными
- непустым подмножеством множества столбцов («ненулевые столбцы»), и
- ненулевым вектором $v \in \mathbb{Z}_{2}^{n}$ («вектор-столбец»).
Поэтому существуют $\left(2^{n}-1\right)^{2}$ таких матриц.
(c) Зафиксируем матрицу $M$ ранга 2 . Тогда существует пара $(v, w)$ столбцов матрицы $M$, образующих невырожденную матрицу. Любой другой столбец-это либо 0 , либо $v$, либо $w$, либо $v+w$ (см. утверждение 3.2). Это множество $S=S_{M}$ из четырёх векторов не зависит от выбора двух столбцов $v, w$; мы назовем его оболочкой столбцов матрицы $M$. (Оно является двумерным векторным подпространством в $\mathbb{Z}_{2}^{n}$.)
Каждая оболочка столбцов определяется любой упорядоченной парой векторов в ней. Любая оболочка столбцов содержит ровно 6 таких упорядоченных пар. Значит, всего существует $\left(2^{n}-1\right)\left(2^{n}-2\right) / 6$ оболочек столбцов. Мы докажем, что существуют ровно $\left(2^{n}-1\right)\left(2^{n}-2\right)$ матриц ранга 2 для данной оболочки столбцов. Таким образом, существует $\left(2^{n}-1\right)^{2}\left(2^{n}-2\right)^{2} / 6$ матриц ранга 2 .
Первое доказательство. Сопоставим матрице $M$ два множества-множество столбцов $X$ матрицы $M$, которые равны $v$ или $v+w$, и множество столбцов $Y$ матрицы $M$, которые равны $w$ или $v+w$. Из условия $\operatorname{rk} M=2$ следует, что оба множества должны быть непустыми и $X \neq Y$. Более того, саму матрицу $M$ по паре ( $X, Y$ множеств можно восстановить. Пар $(X, Y)$ различных непустых множеств ровно $\left(2^{n}-1\right)\left(2^{n}-2\right)$.
Bторое доказательство. Для данного 4 -элементного множества $S=\{0, v, w, v+w\}$ рассмотрим матрицу $M$, оболочкой столбцов которой является $S$. Рассмотрим матрицу $M$ в качестве отображения $\phi_{M}$ из множества $[n]$ столбцов в $S$. Условие $\operatorname{rk} M=2$ равносильно следующему условию:
$(*)$ образ отображения $\phi_{M}$ содержит по крайней мере два вектора из $v, w, v+w$.
Существует всего $4^{n}$ отображений $[n] \rightarrow S$. Существует всего $2^{n}$ отображений $[n] \rightarrow$ $\{0, v\}$. То же самое верно для пары $\{0, v\}$, заменённой на $\{0, w\}$ или $\{0, v+w\}$. Есть лишь одно отображение $[n] \rightarrow\{0\}$. Значит, существуют ровно $4^{n}-3 \cdot 2^{n}+2=\left(2^{n}-1\right)\left(2^{n}-2\right)$ отображений, удовлетворяющих условию (*).
Замечание. Вообще, количесто матриц ранга $k$ из $\mathbb{Z}_{2}^{m \times n}$ равно
$$
\frac{2^{k(k-1) / 2} \prod_{i=0}^{k-1}\left(2^{m-i}-1\right) \Pi_{i=0}^{k-1}\left(2^{n-i}-1\right)}{\prod_{i=0}^{k-1}\left(2^{k-i}-1\right)}
$$
См. теорему 7.1.5 в [АСМ29, p. 299] (эта теорема ещё более общая; для нашего случая матриц над $\mathbb{Z}_{2}$ возьмите $q=2, \operatorname{GF}(2)=\mathbb{Z}_{2}$ ).
4.1. (а1-а3) Изящная реализация графа $K_{5}$ на торе показана на рис. 4 , слева. Реализации графов $K_{6}$ и $K_{7}$ аналогичны, см. рис. 4, в центре.
Другое решение (эта идея также работает для ( $\mathrm{c}, \mathrm{d}$ )): нарисуйте граф $K_{5}$ на плоскости с одним самопересечением, в малой окрестности точки самопересечения прикрепите ручку, а потом поднимите одно из рёбер как мост над другим ребром, проведя его по ручке, см. рис. 5 , слева.
(b2) См. рис. 4, справа.

Рис. 4: Реализация непланарных графов

Рис. 5: Слева: разрешение самопересечения добавлением ручки. Справа: «чётная не общего положения реализация» графа $K_{5}$ на плоскости
4.2. См. рис. 5, справа.
5.1. (b) Индукция по $m$. База $m \leqslant 4$ очевидна. Из предложения 5.7.b и предположения индукции получаем
$$
\widetilde{r_{m}} \geqslant \widetilde{r_{m-5}}+2 \geqslant \frac{2(m-5-4)}{5}+2=\frac{2(m-4)}{5}
$$
(a) Индукция по $m$. База $m \leqslant 4$ очевидна. Из предложения 5.7.a, теоремы 5.1.b и предположения индукции получаем
$$
r_{m} \geqslant \min \left\{r_{m-3}+1, \widetilde{r_{m}}\right\} \geqslant \min \left\{\frac{m-3-4}{3}+1, \frac{2(m-4)}{5}\right\}=\frac{m-4}{3}
$$
5.3. Через $A^{f}$ обозначим $\binom{[m]}{3}$-матрицу, построенную по вложению $f: K_{m} \rightarrow S$, как описано в начале $\S 5$.
(a) Рассмотрим подграф $K_{4}$ на вершинах $1,2,3,4$ графа $K_{5}$. Рис. 3 справа задает вложение $f$ графа $K_{4}$ в тор. Матрица $A^{f}$ - искомая.
(b) Рис. 3 справа задает вложение $f$ графа $K_{5}$ в тор. Матрица $A^{f}$ - искомая.
(c) По утверждению 4.1.d, существует вложение $f$ графа $K_{m}$ в сферу с ручками. Матрица $A^{f}$ - искомая.
(a', b', c') Матрицы из пунктов (a), (b) и (c) - искомые.
(d, e, f) Следуют из задач 5.6.a,b.
5.4. Следует из определений ( $\binom{[m]}{3}$-матрицы и $\binom{[m-1]}{3}$-матрицы.
5.5. (a, b) См. доказательство предложений 7.4.a,b ниже.
5.6. (а) Так как $\binom{[m]}{3}$-матрица для $m \geqslant 5$ невырождена, то $r_{5}, r_{6} \geqslant 1$. Значения $r_{5}=r_{6}=1$ достигаются на матрицах $A^{f}$ для $f$ из утверждения 4.1.b1, $\mathrm{b} 2$.
(b) Из утверждения 3.4.c следует, что $\widetilde{r_{5}}, \widetilde{r_{6}}, \widetilde{r_{7}} \geqslant 2$. Значения $\widetilde{r_{5}}=\widetilde{r_{6}}=\widetilde{r_{7}}=2$ достигаются на матрицах $A^{f}$ для $f$ из утверждений 4.1.a1, a2, a3.
(c) Для $\binom{[m]}{3}$-матрицы $A$ её $\binom{[m-1]}{3}$-подматрица $B$, введённая в утверждении 5.4, имеет такой же ранг, или меньший. Следовательно, $r_{m} \geqslant r_{m-1}$. Если матрица $A$ чётная, то и матрица $B$ чётная. Следовательно, $\widetilde{r_{m}} \geqslant \widetilde{r_{m-1}}$.
5.7. (a) (Take $l=3$.) Take an $\binom{[m]}{l}$-matrix $A$ such that $\operatorname{rk} A=r_{m}$. If $A$ is even, then $r_{m}=\widetilde{r_{m}}$, so we are done. Otherwise there is an $l$-element subset $X \subset[m]$ such that $A_{X, X}=1$. Let $B$ be the 'restriction' of $A$ to $l$-element subsets of $[m]-X$.
Then
$$
r_{m}=\operatorname{rk} A \geqslant \operatorname{rk} B+1 \geqslant r_{m-l}+1, \quad \text { where }
$$
- the first inequality follows by Assertion 5.5.a;
- the second inequality holds because $B$ is a $\binom{[m]-X}{l}$-matrix by Assertion 5.4.
(b) (Take $l=3$.) Take an even $\binom{[m]}{l}$-matrix $A$ such that $\operatorname{rk} A=\widetilde{r_{m}}$. By the non-triviality $A \neq 0$. Hence there are $l$-element subsets $X, Y \subset[m]$ such that $A_{X, Y}=1$. Let $C$ be the 'restriction' of $A$ to $l$-element subsets of $[m]-X-Y$.
Then
$$
\widetilde{r_{m}}=\operatorname{rk} A \geqslant \operatorname{rk} C+2 \geqslant \widetilde{r_{m-2 l+1}}+2, \quad \text { where }
$$
- the first inequality follows by Assertion 5.5.b;
- the second inequality holds because $C$ is a $\binom{[m]-X-Y}{l}$-matrix by Assertion 5.4, and because $A_{X, Y}=1$, so by the triviality $X \cap Y \neq \varnothing$, hence $|[m]-X-Y| \geqslant m-2 l+1$.
6.7. (a) $\lambda_{X, P}=X \cdot{ }_{A} P$.
6.8. (a) $\lambda_{X, Y, P}=X \cdot{ }_{A} P, \lambda_{Y, X, P}=Y \cdot{ }_{A} P$.
7.2. (For 5.4 take $l=3$.) It is obvious that all the conditions for the mentioned submatrix are satisfied.
7.3. (a) (For 5.5.a take $l=3$.) Let $B^{\prime}$ be the 'restriction' of $A$ to $X$ and to $l$-element subsets of $[m]-X$. Then
$$
\operatorname{rk} A \geqslant \operatorname{rk} B^{\prime}=\operatorname{rk} B+1
$$
where equality holds because by the triviality $B_{X, Z}^{\prime}=0$ for any $Z \subset[m]-X$.
(b) (For 5.5.b take $l=3$.) Let $C^{\prime}$ be the 'restriction' of $A$ to $X, Y$ and $l$-element subsets of $[m]-X-Y$. Then
$$
\operatorname{rk} A \geqslant \operatorname{rk} C^{\prime}=\operatorname{rk} C+2
$$
where equality holds because by the triviality $C_{X, Z}^{\prime}=C_{Y, Z}^{\prime}=0$ for any $Z \subset[m]-X-Y$.
(b') Take a basis of $\mathbb{Z}_{2}^{\binom{m}{l}}$ corresponding to $l$-element subsets of $[m]$. Define a bilinear form $A$ on $\mathbb{Z}_{2}^{\binom{m}{l}}$ by setting $A(P, Q):=A_{P, Q}$ for basic vectors $P, Q$. Take any $l$-element set $P \subset[m]$. Let
$$
\bar{P}=\bar{P}(X, Y):=P+A_{X, P} Y+A_{Y, P} X
$$
Recall that
$$
A_{X, Y}=A_{Y, X}=1 \quad \text { and } \quad A_{X, X}=A_{Y, Y}=0
$$
Hence
$$
A(\bar{P}, X)=A(\bar{P}, Y)=0
$$
(i. e., $\bar{P}$ is the orthogonal projection of $P$ to the orthogonal complement of $\langle X, Y\rangle$ with respect to $A$ ). By the triviality, for $P \subset[m]-X$ we have $\bar{P}=P+A_{Y, P} X$. Hence for every $l$-element sets $P, Q \subset[m]-X$ we have
$$
A(\bar{P}, \bar{Q})=A_{P, Q}+0+0+0=B_{P, Q}
$$
(I. e., $B$ is the Gramian matrix with respect to $A$ of the 'projections' $\bar{P}$ of $l$-element sets $P \subset[m]-X$.) Let $B^{\prime}$ be the Gramian matrix with respect to $A$ of $X, Y$ and the 'projections' $\bar{R}$ of $l$-element sets $R \subset[m]-X$. I. e., $B_{P, Q}^{\prime}=A(\widehat{P}, \widehat{Q})$, where $\widehat{P}=P$ if $P \in\{X, Y\}$, and $\widehat{P}=\bar{P}$ otherwise ( $\widehat{Q}$ is defined analogously). Then
- $B_{X, Y}^{\prime}=B_{Y, X}^{\prime}=1, B_{X, X}^{\prime}=B_{Y, Y}^{\prime}=0$ (by $(*)$ ),
- $B_{X, P}^{\prime}=B_{P, X}^{\prime}=B_{Y, P}^{\prime}=B_{P, Y}^{\prime}=0$ for $P \neq X, Y$ (by $(* *)$ ), and
- $B_{P, Q}^{\prime}=B_{P, Q}$ for $P, Q \subset[m]-X$ (by $(* * *)$ ).
Hence $\operatorname{rk} B+2=\operatorname{rk} B^{\prime} \leqslant \operatorname{rk} A$.
(a') In this paragraph we prove that $\operatorname{rk} D<\operatorname{rk} A$. Take a basis of $\mathbb{Z}_{2}^{\binom{m}{l}}$ corresponding to l-element subsets of $[m]$. Define a bilinear form $A$ on $\mathbb{Z}_{2}^{\binom{m}{l}}$ by setting $A(P, Q):=A_{P, Q}$ for basic vectors $P, Q$. Let $P_{X}$ be the orthogonal projection of $P$ to the orthogonal complement of $X$ (with respect to $A$ ), i. e., $P_{X}:=P+A_{P, X} X$. We have
$$
\begin{aligned}
& A\left(P_{X}, Q_{X}\right)=A(P, Q)+A\left(A_{P, X} X, Q\right)+A\left(P, A_{Q, X} X\right)+A\left(A_{P, X} X, A_{Q, X} X\right)= \\
& \quad=A_{P, Q}+A_{P, X} A_{X, Q}+A_{P, X} A_{Q, X}+A_{P, X} A_{Q, X} A_{X, X}=A_{P, Q}+A_{P, X} A_{Q, X}=D_{P, Q}
\end{aligned}
$$
Then $D$ is the Gramian matrix (with respect to $A$ ) of the projections of subsets of $[m-l+1]$. Let $D^{\prime}$ be the Gramian matrix (with respect to $A$ ) of $X$ and the projections of subsets of $[m-l+1]$. We have $D_{P, Q}=D_{P, Q}^{\prime}$ for all subsets $P, Q \subset[m-l+1]$. Furthermore, $D_{X, P}^{\prime}=$ $D_{P, X}^{\prime}=0$ for any basic vector $P \neq X$ and $D_{X, X}^{\prime}=A_{X, X}=1$. Thus $\operatorname{rk} D=\operatorname{rk} D^{\prime}-1<\operatorname{rk} A$.
In this paragraph we prove that $D$ satisfies the triviality property. If $P \cap Q=\varnothing$, then either $P \cap X=\varnothing$, or $Q \cap X=\varnothing$. Hence $D_{P, Q}=A_{P, Q}+A_{P, X} A_{Q, X}=0+0=0$.
In this paragraph we prove that $D$ satisfies the linear dependence property. For each $(l+1)$-element and $l$-element subsets $F, P \subset[m-l+1]$ we have
$$
\sum_{i \in F} D_{F-i, P}=\sum_{i \in F} A_{F-i, P}+A_{P, X} \sum_{i \in F} A_{F-i, X}=0
$$
In this paragraph we prove that $D$ satisfies the non-triviality property. By Proposition 7.5.b for $D$, we may assume that $i \neq m-l+1$. Then for each summand $D_{i \sqcup \sigma, i \sqcup \tau}$ of $D_{F, i}$ at least one of the sets $i \sqcup \sigma, i \sqcup \tau$ does not contain $m-l+1$ and hence does not intersect $X$. Hence $D_{i \sqcup \sigma, i \sqcup \tau}=A_{i \sqcup \sigma, i \sqcup \tau}+A_{i \sqcup \sigma, X} A_{i \sqcup \tau, X}=A_{i \sqcup \sigma, i \sqcup \tau}$. Thus $D_{F, i}=A_{F, i}=1$.
7.4. (a) Take an $\binom{[m]}{l}$-matrix $A$ such that $\operatorname{rk} A=r_{m}$. If $A$ is even, then $r_{m}=\widetilde{r_{m}}$, so we are done. Otherwise there is an $l$-element subset $X \subset[m]$ such that $A_{X, X}=1$. Without loss of generality $X=\{m-l+1, m-l+2, \ldots, m\}$. Then by Assertion 7.3.a'
$$
r_{m}=\operatorname{rk} A \geqslant \operatorname{rk} D+1 \geqslant r_{m-l+1}+1, \quad \text { where }
$$
- $D$ is the matrix defined in Assertion 7.3.a';
- the first inequality follows from Assertion 7.3.a';
- the second inequality holds because $D$ is an $\binom{[m-l+1]}{l}$-matrix by Assertion 7.3.a'.
(b) Take an even $\binom{[m]}{l}$-matrix $A$ such that $\mathrm{rk} A=\widetilde{r_{m}}$. By the non-triviality $A \neq 0$. Let $X, Y, B$ be defined as in Assertion 7.3.b'. Then
$$
\widetilde{r_{m}}=\operatorname{rk} A \geqslant \operatorname{rk} B+2 \geqslant r_{m-l}+2, \quad \text { where }
$$
- the first inequality follows from Assertion 7.3.b';
- the second inequality holds because $B$ is an even $\binom{[m]-X}{l}$-matrix by Assertion 7.2.
7.5. (a) It suffices to check that for each pair $\{\alpha, \beta\}$ the number of sets $T_{i,\{\sigma, \tau\}}$ containing $\{\alpha, \beta\}$ is odd if and only if $\alpha \cap \beta=\varnothing$ (hence this parity not depend on $i$ ). Clearly, $|\alpha \cap \beta| \leqslant 2$.
Assume $|\alpha \cap \beta|=2$. Then $\{\alpha, \beta\} \notin T_{i,\{\sigma, \tau\}}$ for all $i, \sigma, \tau$ and hence $\{\alpha, \beta\} \notin A_{i}$ for all $i \in[5]$.
Assume $|\alpha \cap \beta|=1$. It suffices to consider the case $\alpha=\{1,2\}, \beta=\{1,3\}$. Then $\{\alpha, \beta\} \in T_{i,\{\sigma, \tau\}}$ iff $i=1$ and $\{\sigma, \tau\}$ is either $\{\{2,4\},\{3,5\}\}$ or $\{\{2,5\},\{3,4\}\}$ Therefore $\{\alpha, \beta\} \notin A_{i}$ for all $i \in[5]$.
Assume $|\alpha \cap \beta|=0$. It suffices to consider the case $\alpha=\{1,2\}, \beta=\{3,4\}$. Then $\{\alpha, \beta\} \in T_{i,\{\sigma, \tau\}}$ iff either
- $i=1$ and $\{\sigma, \tau\}=\{\{1,2,5\},\{1,3,4\}\}$, or
- $i=2$ and $\{\sigma, \tau\}=\{\{1,2,5\},\{2,3,4\}\}$, or
- $i=3$ and $\{\sigma, \tau\}=\{\{1,2,3\},\{3,4,5\}\}$, or
- $i=4$ and $\{\sigma, \tau\}=\{\{1,2,4\},\{3,4,5\}\}$, or
- $i=5$ and $\{\sigma, \tau\}=\{\{1,2,5\},\{3,4,5\}\}$.
Therefore $\{\alpha, \beta\} \in A_{i}$ for every $i \in[5]$.
(b) It suffices to prove that $A_{G \sqcup i, j}=A_{G \sqcup j, i}$ for each $i, j \in[m]$ and $(2 l-3)$-element subset $G \subset[m]-i-j$. Denote $\bar{\sigma}:=\{i, j\} \sqcup \sigma$. Then
$$
\begin{aligned}
A_{G \sqcup j, i}+A_{G \sqcup i, j} & \stackrel{(1)}{=} \sum_{\{(\sigma, \tau): G=\sigma \sqcup \tau,|\sigma|=l-2\}}\left(A_{\bar{\sigma}, i \sqcup \tau}+A_{\bar{\sigma}, j \sqcup \tau}\right) \stackrel{(2)}{=} \\
& =\sum_{\{(\sigma, \tau):} \sum_{G=\sigma \sqcup \tau,|\sigma|=l-2\}} A_{\bar{\sigma}, \overline{\tau-t}} \stackrel{(3)}{=} \sum_{t \in G} \sum_{\{(\sigma, \nu): G-t=\sigma \sqcup \nu,|\sigma|=l-2\}} A_{\bar{\sigma}, \bar{\nu}} \stackrel{(4)}{=} 0, \quad \text { where }
\end{aligned}
$$
- equality (1) holds because $A_{G \sqcup j, i}$ is equal to the sum of the first summands $A_{\bar{\sigma}, i \sqcup \tau}$, and $A_{G \sqcup i, j}$ is equal to the sum of the second summands $A_{\bar{\sigma}, j \sqcup \tau}$;
- equality (2) holds by the linear dependence for $F=\bar{\tau}, P=\bar{\sigma}$;
- equality (3) is obtained by changes of the order of summation and of variable $\nu=\tau-t$;
- equality (4) holds because ordered decompositions $(\sigma, \nu)$ of $G-t$ into $(l-2)$-element subsets $\sigma, \nu$ split into pairs $\{(\sigma, \nu),(\nu, \sigma)\}$ and $A_{\bar{\sigma}, \bar{\nu}}+A_{\bar{\nu}, \bar{\sigma}}=0$.
## Список литературы
[ACM29] D. Hachenberger, D. Jungnickel. Topics in Galois Fields (2020).
[Bi20] * A. Бикеев. Реализуемость дисков с ленточками на ленте Мёбиуса. Математическое просвещение. Сер. 3. Вып.28. 2021. С 150-158. Исправления в печати. arXiv:2010.15833.
[Bi21] A. I. Bikeev, Criteria for integer and modulo 2 embeddability of graphs to surfaces, arXiv:2012.12070v2.
[DS22] S. Dzhenzher and A. Skopenkov, To the Kühnel conjecture on embeddability of $k$-complexes into $2 k$-manifolds, arXiv:2208.04188.
[FK19] R. Fulek, J. Kynčl, $\mathbb{Z}_{2}$-genus of graphs and minimum rank of partial symmetric matrices, 35th Intern. Symp. on Comp. Geom. (SoCG 2019), Article
No. 39; pp. 39:1-39:16, https://drops.dagstuhl.de/opus/volltexte/2019/ 10443/pdf/LIPIcs-SoCG-2019-39.pdf. We refer to numbering in arXiv version: arXiv:1903.08637.
[IF] *http://www.map.mpim-bonn.mpg.de/Intersection_form
[Ko21] E. Kogan. On the rank of $\mathbb{Z}_{2}$-matrices with free entries on the diagonal, arXiv:2104.10668.
[KS21] * E. Kogan and A. Skopenkov. A short exposition of the Patak-Tancer theorem on non-embeddability of $k$-complexes in $2 k$-manifolds, arXiv:2106.14010.
[KS21e] E. Kogan and A. Skopenkov. Embeddings of $k$-complexes in $2 k$-manifolds and minimum rank of partial symmetric matrices, arXiv:2112.06636.
[MC] https://en.wikipedia.org/wiki/Matrix_completion\#Low_rank_matrix_ completion
[NKS] L. T. Nguyen, J. Kim, B. Shim. Low-rank matrix completion: a contemporary survey. arXiv:1907.11705
[PT19] P. Paták and M. Tancer. Embeddings of $k$-complexes into $2 k$-manifolds. arXiv:1904.02404.
[Sk14] * A. Skopenkov, Realizability of hypergraphs and Ramsey link theory, arXiv:1402.0658.
[Sk20] * A. Скопенков, Алгебраическая топология с геометрической точки зрения, МЦНМО, Москва, 2020 (2ое издание).
Фрагмент книги: http://www.mccme.ru/circles/oim/obstruct.pdf. Фрагмент английской версии:
https://www.mccme.ru/circles/oim/obstructeng.pdf.
## Low rank matrix completion
presented by S. Dzhenzher, T. Garaev, O. Nikitenko, A. Petukhov, A. Skopenkov, A. Voropaev*
## Contents
1 Motivation and some main results ..... 1
2 Degenerate matrices ..... 4
3 The rank of a matrix ..... 5
4 Modulo 2 embeddings of graphs to surfaces ..... 6
5 Rank of matrix with relations ..... 9
6 Classification of symmetric bilinear forms ..... 10
7 Rank of matrix with relations: generalization ..... 12
## 1 Motivation and some main results
Remark (motivation; formally is not used later)
'Matrix completion is the task of filling in the missing entries of a partially observed matrix... One example is the movie-ratings matrix, as appears in the Netflix problem: Given a ratings matrix in which each entry $(i, j)$ represents the rating of movie $j$ by customer $i$, if customer $i$ has watched movie $j$ and is otherwise missing, we would like to predict the remaining entries in order to make good recommendations to customers on what to watch next...' [MC] The remaining entries are predicted so as to minimize the rank of the completed matrix. All the required definitions (of rank etc.) are given below. For a brief overview of the history of this and related problems, see [MC, NKS], [Ko21, Remark 4].
Here for simplicity we consider matrices with entries in the set $\mathbb{Z}_{2}=\{0,1\}$ of all residues modulo 2 (with the sum and product operations). We present interesting elementary results in linear algebra. These results allow us to construct algorithms estimating minimal rank of partial matrices for the particular case of filling the diagonal (Proposition 1.1 and Theorems
${ }^{*}$ S. Dzhenzher, A. Skopenkov: Moscow Institute of Physics and Technology. T. Garaev: Moscow State University. O. Nikitenko: Altay Technical University (Barnaul). A. Petukhov: Institute for Information Transmission Problems (Moscow). A. Skopenkov: Independent University of Moscow, https://users . mccme.ru/skopenko/. A. Voropaev: (Moscow).
We are grateful to E. Kogan for allowing us to use his text [Ko21], to V. Retinskiy and Ya. Abramov for useful discussions, to D. Deomidov and F. Nilov for translating parts of the text, to A. Ryabichev and MCCME publishing house for allowing us to use figures they prepared.
1.3, 1.4, see also Proposition 1.2). Then we consider a more complicated problem. Instead of relations $M_{i j}=a_{i j}$ for some elements $M_{i j}$ of matrix $M$ (where $a_{i j}$ are given numbers) we consider more complicated relations on matrix elements. We estimate minimal rank of matrices with such relations (Theorems 5.1, 7.1.)
These results have applications to embeddings of graphs in surfaces (including embedding modulo 2 , see §4), and of $k$-dimensional 'hypergraphs' to $2 k$-dimensional surfaces, see [KS21, KS21e, DS22]. In particular, Theorems 1.3 and 1.4 give polynomial (in the number of edges) algorithms recognizing 'weak realizability' of 'graphs with rotations' on non-orientable surfaces (see the remark on weak realizability below).
Denote by $\mathbb{Z}_{2}^{s \times n}=\left(\mathbb{Z}_{2}^{s}\right)^{n}$ the set of all $s \times n$ matrices with entries in $\mathbb{Z}_{2}$.
Proposition 1.1. (a) For a symmetric matrix with $\mathbb{Z}_{2}$-entries the following conditions are equivalent:
- some entries on the main diagonal can be changed so that in the resulting matrix all non-zero rows are equal;
- it is impossible to make the same permutation of rows and of columns ${ }^{1}$ so that the upper left square will be one of the submatrices
$$
\left(\begin{array}{ccc}
* & 1 & 1 \\
1 & * & 0 \\
1 & 0 & *
\end{array}\right) \quad \text { or } \quad\left(\begin{array}{cccc}
* & 1 & 0 & 0 \\
1 & * & 0 & 0 \\
0 & 0 & * & 1 \\
0 & 0 & 1 & *
\end{array}\right)
$$
where by * are denoted arbitrary (possibly different) elements.
(b) There is an algorithm with the complexity of $O\left(n^{2}\right)$ deciding for a matrix $M \in \mathbb{Z}_{2}^{n \times n}$ whether some entries on the main diagonal can be changed so that in the resulting matrix all non-zero rows are equal.
You can submit separately your solution of the 'only if' implication of part (a).
The algorithmic results in this text could be omitted by theoretically-minded students because they are easy corollaries of mathematical results. The complexity of an algorithm is the number 'elementary' steps in this algorithm. An algorithm has complexity $O(f(n))$ if there is $C>0$ such that the complexity does not exceed $C f(n)$ for any $n$.
A square matrix $M \in \mathbb{Z}_{2}^{n \times n}$ is called degenerate if the sum of its several columns (a non-zero number of columns) is the zero column (i. e., the column consisting of zeros only). A matrix is called non-degenerate otherwise. Introductory problems on degenerate matrices useful for the following result are presented in $\S 2$.
Proposition 1.2. (a) For any matrix $M \in \mathbb{Z}_{2}^{n \times n}$ some entries on the main diagonal can be changed so that the resulting matrix would be degenerate.
(b) The same with 'degenerate' replaced by 'non-degenerate'.
The rank $\operatorname{rk} M$ of a matrix $M \in \mathbb{Z}_{2}^{s \times n}$ is the maximal number of columns of $M$ none of whose sums is zero. (This is the 'dimension' of the 'vector space' formed by the columns of the matrix.) Introductory problems on rank useful for the following results are presented in $\S 3$.
For a matrix $M \in \mathbb{Z}_{2}^{n \times n}$ let $R(M)$ be the minimal rank of all the matrices obtained by changing some entries on the main diagonal of $M$.
A matrix is said to be diagonal if all its entries outside of the main diagonal are zeros.[^3]
Theorem 1.3. (a') To make a square matrix of rank $k$ out of a square matrix of rank $n$ by changing some diagonal entries, one needs to change at least $|n-k|$ entries.
(a) For any non-degenerate matrix $M \in \mathbb{Z}_{2}^{n \times n}$ the inequality $R(M) \leq k$ is equivalent to the existence of a diagonal matrix $D$ with at most $k$ zeroes on the main diagonal such that $\operatorname{rk}(M+D) \leq k$.
(b) For any fixed $k$ there is an algorithm with the complexity of $O\left(n^{k+3}\right)$ deciding for a matrix $M \in \mathbb{Z}_{2}^{n \times n}$ whether $R(M) \leq k$.
The identity matrix $E$ is the diagonal matrix whose diagonal elements are units.
Theorem 1.4. (a) For any non-degenerate matrix $M \in \mathbb{Z}_{2}^{n \times n}$ and diagonal matrix $D \in \mathbb{Z}_{2}^{n \times n}$ we have $2 \operatorname{rk}(M+D) \geq \operatorname{rk}(M+E)$.
(b) There is an algorithm with the complexity of $O\left(n^{4}\right)$ calculating for a matrix $M \in \mathbb{Z}_{2}^{n \times n}$ a number $k$ such that $k / 2 \leq R(M) \leq k$.
Remark (weak realizability; formally is not used later)
A hieroglyph on $n$ letters is an unoriented cyclic letter sequence of length $2 n$ such that each letter from the sequence appears in the sequence twice.
Take a hieroglyph on $n$ letters. Take a convex polygon with $2 n$ sides. Put the letters in the hieroglyph on the sides of the convex polygon in the nonoriented cyclic order. For each letter glue the ends of a ribbon to the pair of sides corresponding to the letter so that the glued ribbons are pairwise disjoint. The ribbons can be either twisted or not twisted. Call the resulting surface a disk with ribbons corresponding to the hieroglyph (see Figure 1).

Figure 1: Disk with ribbons corresponding to the hieroglyph $a a b b c c$ (left) and $a a b c b c$ (right)
A hieroglyph $H$ is called weakly realizable on the Möbius band if some disk with ribbons corresponding to $H$ can be cut out of the Möbius band. Analogously one defines weak realizability on the Klein bottle and other non-orientable surfaces.
Two letters $a, b$ in a hieroglyph $H$ overlap in $H$ if they interlace in the cyclic sequence of the hieroglyph (i. e., if they appear in the sequence in the order $a b a b$ but not $a a b b$ ). Define the overlap matrix $M(H) \in \mathbb{Z}_{2}^{n \times n}$ of a hieroglyph $H$ as follows. Put zeros on the main diagonal. Put 1 in the cell $(i, j)$ for $i \neq j$ if the letters $i, j$ overlap in $H$, and put 0 otherwise.
Hieroglyph $H$ is weakly realizable on the Möbius band if and only if $R(M(H)) \leq 1$.
See more in [Bi20], [Ko21, Appendix], [Sk20, §2].
## Recommendations for participants
If a mathematical statement is formulated as a problem, then the objective is to prove this statement. (Open-ended questions are called challenges or riddles; here one must come up with a clear wording, and a proof.) If a problem is named 'theorem' ('lemma', 'corollary', etc.), then this statement is considered to be more important. Usually we formulate beautiful or important statement before giving a sequence of results (lemmas, assertions, etc.) which
constitute its proof. We give hints on that after the statements but we do not want to deprive you of the pleasure of finding the right moment when you finally are ready to prove the statement. In general, if you are stuck on a certain problem, try looking at the next ones. They may turn out to be helpful. Remarks and problems marked by star are not used in the sequel. Important definitions are highlighted in bold for easy navigation. You are welcomed to consult the jury on any questions on the project. If you successfully work on the project, you can get interesting extra problems.
For every solution written for a user marked with either ' + ' or ' + .' a student (or a group of students) gets a 'bean' (see recommendations in p. 3, 'How to write a proof for a user' of https://www.mccme.ru/circles/oim/multicomb.pdf). The jury may also award extra beans for beautiful solutions, solutions of hard problems, or solutions typeset in $\mathrm{T}_{\mathrm{E}} \mathrm{X}$. The jury has infinitely many beans. Every participant (or team) initially has 1 bean. You may submit a solution in oral form or as written for a developer; you lose a bean with every 5 attempts (successful or not).
Please notify us if you already know solutions of several problems. If you confirm your knowledge by presenting some solutions, you will be allowed not to receive plus-marks for the problems, but to use them in solutions of other problems.
(In the project the tasks were separated as follows: the tasks before and including Theorem 4.3, then some solutions to simple tasks from $\S \S 1-2$, then other tasks and solutions.)
## 2 Degenerate matrices
2.1. (a1-a4) Which of the following matrices are degenerate?
$$
A_{1}=\left(\begin{array}{lll}
0 & 0 & 0 \\
0 & 0 & 0 \\
0 & 0 & 0
\end{array}\right), \quad A_{2}=\left(\begin{array}{ccc}
1 & 1 & 1 \\
1 & 1 & 1 \\
1 & 1 & 1
\end{array}\right), \quad A_{3}=\left(\begin{array}{ccc}
1 & 1 & 1 \\
1 & 0 & 1 \\
1 & 1 & 1
\end{array}\right), \quad A_{4}=\left(\begin{array}{ccc}
0 & 1 & 1 \\
0 & 0 & 1 \\
1 & 0 & 0
\end{array}\right)
$$
2.2. (a) Degeneracy is not changed under permutation of columns (rows).
(b) Degeneracy is not changed under adding one column (row) to another.
(c) Any matrix can be changed to a diagonal matrix by transformations from $(a, b)$.
(d) A matrix is degenerate if and only if it cannot be changed by transformations from $(a, b)$ to the identity matrix.
(f) A square matrix is degenerate if and only if the sum of its several rows (a non-zero number of rows) is the zero row.
(g) There is an algorithm with the complexity of $O\left(n^{3}\right)$ checking the degeneracy of an $n \times n$ matrix.
For a matrix $M \in \mathbb{Z}_{2}^{n \times n}$ define $\operatorname{det} M:=0$ if $M$ is degenerate, and $\operatorname{det} M:=1$ otherwise. This is called the determinant of $M$. Another notation is
$$
\operatorname{det}\left(\begin{array}{ll}
M_{1,1} & M_{1,2} \\
M_{2,1} & M_{2,2}
\end{array}\right)=\left|\begin{array}{ll}
M_{1,1} & M_{1,2} \\
M_{2,1} & M_{2,2}
\end{array}\right|, \quad \operatorname{det} M=\left|\begin{array}{ccc}
M_{1,1} & \ldots & M_{1, n} \\
\vdots & \ddots & \vdots \\
M_{n, 1} & \ldots & M_{n, n}
\end{array}\right|
$$
2.3. (a) $\left|\begin{array}{ll}a & b \\ c & d\end{array}\right|=a d+b c$.
(b) $\operatorname{det}\left(a_{1}+b_{1}, a_{2}, \ldots, a_{n}\right)=\operatorname{det}\left(a_{1}, a_{2}, \ldots, a_{n}\right)+\operatorname{det}\left(b_{1}, a_{2}, \ldots, a_{n}\right)$. Here and below $a_{j}, b_{1} \in \mathbb{Z}_{2}^{n}$ are columns of length $n$.
(c) $\operatorname{det}\left(a_{1}, \ldots, a_{n}\right)=\sum_{i=1}^{n} a_{i, n} \operatorname{det}\left(a_{1}^{-}, \ldots, a_{i-1}^{-}, a_{i+1}^{-}, \ldots, a_{n}^{-}\right)$, where every column $a_{i}^{-} \in$ $\mathbb{Z}_{2}^{n-1}$ is obtained from the column $a_{i}$ by deleting the last coordinate.
$(d)^{*} \operatorname{det} M=\sum_{\sigma \in S_{n}} \prod_{i=1}^{n} M_{i, \sigma(i)}$, where $S_{n}$ is the set of all permutations (i. e., 1-1 correspondences) $\sigma:[n] \rightarrow[n]$.
2.4. (a1-a4) For every matrix of Problem 2.1 find if some entries on the main diagonal can be changed so that the resulting matrix would be degenerate.
(b1-b4) The same with 'degenerate' replaced by 'non-degenerate'.
Lemma 2.5. (a) Let $M \in \mathbb{Z}_{2}^{n \times n}$ be a matrix with zeros on the main diagonal. Define the sequence $M^{(i)}, i=0,1,2, \ldots, n$ recursively as follows:
- $M^{(0)}:=M$, and
- $M^{(i)}$ is the result of replacing in $M^{(i-1)}$ the element $M_{i, i}^{(i-1)}=0$ by $1+\delta_{i}$, where $\delta_{0}=0$ and $\delta_{i}:=\operatorname{det} M_{[i] \times[i]}^{(i-1)}$ is the determinant of the left upper $i \times i$-corner submatrix of $M^{(i-1)}$.
Then the matrix $M_{n}$ is non-degenerate.
(b) There is an algorithm with the complexity of $O\left(n^{4}\right)$ which for a matrix $M \in \mathbb{Z}_{2}^{n \times n}$ finds some numbers from $\mathbb{Z}_{2}$ to replace the entries on the main diagonal of $M$ so that the resulting matrix is non-degenerate.
## 3 The rank of a matrix
3.1. (a1-a4) Find rk $M$ for
$$
M=\left(\begin{array}{llll}
0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0
\end{array}\right), \quad\left(\begin{array}{cccc}
1 & 1 & 1 & 1 \\
1 & 1 & 1 & 1 \\
1 & 1 & 1 & 1
\end{array}\right), \quad\left(\begin{array}{cccc}
1 & 1 & 1 & 0 \\
1 & 0 & 1 & 0 \\
1 & 1 & 1 & 1
\end{array}\right), \quad\left(\begin{array}{llll}
0 & 1 & 1 & 1 \\
0 & 0 & 1 & 1 \\
1 & 0 & 0 & 1
\end{array}\right)
$$
(b1-b4) Find $R(M)$ for the matrices from Problem 2.1.
3.2. Take a matrix $M \in \mathbb{Z}_{2}^{s \times n}$.
(a) One can choose $\mathrm{rk} M$ columns of $M$ such that every column is the sum of some chosen columns.
(b) Assume that there are $k$ columns (not necessarily of $M$ ) such that every column of $M$ is the sum of some of them. Then $\operatorname{rk} M \leq k$.
(c) The rank of a submatrix does not exceed the rank of a matrix.
3.3. (a) A permutation of columns (or of rows) does not change the rank of a matrix.
(b) Adding one column to another one (or one row to another one) does not change the rank of a matrix.
(c) The rank of a matrix equals to the maximal number of its rows none of whose sums is zero.
(d) The rank of a matrix equals to the maximal size of its non-degenerate square submatrix.
A square matrix with $\mathbb{Z}_{2}$-entries is called even if all the entries on the main diagonal are zeros.
3.4. (a) For $M \in \mathbb{Z}_{2}^{s \times n}$ all non-zero rows are equal if and only if $\operatorname{rk} M \leq 1$.
(b) For a symmetric matrix $M \in \mathbb{Z}_{2}^{n \times n}$ all non-zero rows are equal if and only if by same permutation of rows and of columns it is possible to obtain a matrix whose upper left square is filled by ones, and all other elements are zeros.
(c) The rank of any non-zero symmetric even matrix is greater than one.
3.5. The number $R(M)$ is not necessarily preserved by
(a) permutation of columns;
(b) adding one column to another one.
3.6. (a) There is an algorithm with the complexity of $O\left(n^{3}\right)$ which calculates the rank of a matrix from $\mathbb{Z}_{2}^{s \times n}, s \leq n$.
(b)* For a fixed integer $k$ there is an algorithm with the complexity of $O\left(n^{2}\right)$ deciding for $M \in \mathbb{Z}_{2}^{s \times n}, s \leq n$, whether $\operatorname{rk} M \leq k$.
Lemma 3.7. Let $M, D$ be matrices of the same size with entries in $\mathbb{Z}_{2}$. Then
(a) $\operatorname{rk}(M+D) \leq \operatorname{rk} M+\operatorname{rk} D$;
(b) $\operatorname{rk}(M+D) \geq \operatorname{rk} M-\operatorname{rk} D$.
3.8. There is an algorithm with the complexity of $O\left(n^{k+3}\right)$ finding for $M \in \mathbb{Z}_{2}^{n \times n}$ a diagonal matrix $D$ such that
(a) $\operatorname{rk}(M+D) \leq k ; \quad$ (b) $\operatorname{rk}(M+D)=k$
under the assumption that such a matrix $D$ exists.
3.9. Is it correct that for any $m, k \leq n$ and a matrix $M \in \mathbb{Z}_{2}^{n \times n}$ of rank $m$, if a matrix of rank $k$ can be obtained by changing some entries on the diagonal of $M$, then this can be done by changing exactly $|m-k|$ entries?
3.10. $(a, b, c)$ Find the number of matrices of rank $k$ in $\mathbb{Z}_{2}^{n \times n}$ for $k=0,1,2$.
## 4 Modulo 2 embeddings of graphs to surfaces
This section is formally not used later, but serves as an additional motivation for $\S 5$.
Denote by $S$ the torus, or sphere with handles, or the Möbius band, or the Klein bottle, or a 2-dimensional surface. Their simple definitions can be found e. g. in $\$ 2.1$ of
[Sk20]=https://www.mccme.ru/circles/oim/obstructeng.pdf
Below graph drawings on $S$ may have self-intersections. An embedding is a graph drawing without self-intersections.
4.1. There are embeddings $(a 1, a 2, a 3)$ of $K_{5}, K_{6}, K_{7}$ in the torus;
$(b 1, b 2)$ of $K_{5}, K_{6}$ in the Möbius band;
(c) of $K_{8}$ in the sphere with two handles;
(d) of $K_{m}$ in the sphere with some number of handles (depending on $m$ ).
Remark. In this problem you need to give not rigorous proofs, but large, comprehensible, and preferably beautiful pictures.
4.2. The graph $K_{5}$ can be drawn in the plane so that the drawings (i. e., the images of) every two non-adjacent edges intersect at an even number of points.
A self-intersection point of a drawing is a point on the drawing to which corresponds more than one point of the graph itself.
A graph drawing is said to be general position if
- to every self-intersection point there corresponds exactly two points of the graph;
- every drawing of a vertex is not a self-intersection point,
- the drawing has finitely many self-intersection points, and

Figure 2: A transverse intersection and a non-transverse intersection
- at every such point the self-intersection is transverse (Figure 2)2.
A general position graph drawing is a $\mathbb{Z}_{2}$-embedding if the drawings of every two nonadjacent edges intersect at an even number of points.
Remark. Let $S$ be either the plane or the torus or the Möbius band. If a graph has a $\mathbb{Z}_{2}$-embedding to $S$, then the graph has an embedding in $S$. However, there is a graph having a $\mathbb{Z}_{2}$-embedding to the sphere with 4 handles but not an embedding in the sphere with 4 handles. See references in [Bi21, Remark 1.3.b,c].
Theorem 4.3. If graph $K$ has a $\mathbb{Z}_{2}$-embedding to the sphere with $g$ handles, then
(a) $g \geq(m-4) / 3$ for $K=K_{m}$.
(b) $g \geq(m-5)^{2} / 16$ for $K=K_{m}$.
(c) $g \geq(n-2)^{2} / 4$ for $K=K_{n, n}$.
Theorem 4.3 is proved by showing that on a surface to which a large graph has a $\mathbb{Z}_{2^{-}}$ embedding, the intersections of closed curves are sufficiently complicated (in the sense of rank of certain matrix; cf. Assertion 4.5). More precisely, Theorem 4.3.a [PT19] follows by Theorems 4.4 and 5.1.b together with Assertion 4.5 (all below). Theorem 4.3.b follows by Theorem 4.3.c (prove!). Theorem 4.3.c is proved in [FK19], see a well-structured exposition in [DS22]. Analogously, Assertions 4.5 and 5.2 (together with Theorem 4.4 and its nonorientable analogue) imply the non- $\mathbb{Z}_{2}$-embeddability of $K_{8}$ to the torus, and of $K_{7}$ to the Möbius band (or even of $K_{7}$ to the Klein bottle; the non-embeddability of $K_{7}$ to the Klein bottle does not follow from the Euler inequality). Analogous non-embeddabiliy result in higher dimensions follows by Theorem 7.1.
Denote by $|X|_{2} \in \mathbb{Z}_{2}$ the parity of the number of elements in a finite set $X$.
Closed curves $\gamma_{1}, \ldots, \gamma_{p}$ on $S$ are said to be in general position if the graph drawing (of disjoint union of $p$ cycles) formed by this curves is in general position. Their intersection $p \times p$ - matrix $G$ is defined as
$$
G_{i, j}:= \begin{cases}\left|\gamma_{i} \cap \gamma_{j}\right|_{2}, & i \neq j \\ \left|\gamma_{i} \cap \gamma_{i}^{\prime}\right|_{2}, & i=j\end{cases}
$$[^4]where $\gamma_{i}^{\prime}$ is a curve close to $\gamma_{i}$ in general position to $\gamma_{i}$. Please use the following 'Homology Betti Theorem' without proof.
Theorem 4.4. For any closed general position curves $\gamma_{1}, \ldots, \gamma_{p}$ on
(a) the sphere with $g$ handles the rank of their intersection matrix does not exceed $2 g$.
(b) the disk with $m$ Möbius bands the rank of their intersection matrix does not exceed $m$.
Here the disk with $k$ Möbius bands is the figure shown on the left of fig. 1. More precisely, the disk with $m$ Möbius bands is the union of a disk and $m$ pairwise disjoint ribbons having their ends glued to $2 m$ pairwise disjoint arcs on the boundary circle of the disk (the ribbons do not have to lie in the plane of the disk) so that
- the orientations of the ends of each ribbon given by an orientation of the boundary circle of the disk have 'the same direction along the ribbon', and
- the ribbons are 'separated', i. e. there are $m$ pairwise disjoint $\operatorname{arcs} A_{i}$ on the boundary circle of the disk such that the ends of the $i$-th ribbon are glued to two disjoint arcs contained in $A_{i}, i=1,2, \ldots, m$.
You can make approximations by general position drawings at an intuitive level.

Figure 3: Left: $K_{3}$ and $K_{4}$ on the torus. Right: $K_{5}$ on the torus
4.5. Take any embedding (or $\mathbb{Z}_{2}$-embedding) $f: K_{n} \rightarrow S$. Take any map $f^{\prime}: K_{n} \rightarrow S$ in general position to $f$, and close to $f$. For any pairwise different numbers $i, j, k \in[n]$ denote by $\langle i j k\rangle$ the cycle of length 3 in $K_{n}$ passing through $i, j, k$. Let
$$
i j k \wedge p q r:=\left|f\langle i j k\rangle \cap f^{\prime}\langle p q r\rangle\right|_{2}
$$
Then
(4.5.1) $123 \wedge 456=0$.
(4.5.2) $123 \wedge 456+123 \wedge 567+123 \wedge 467+123 \wedge 457=0$.
$123 \wedge 345+123 \wedge 346+123 \wedge 356+123 \wedge 456=0$.
$123 \wedge 234+123 \wedge 235+123 \wedge 245+123 \wedge 345=0$.
$123 \wedge 123+123 \wedge 124+123 \wedge 134+123 \wedge 234=0$.
See Figure 3, left. For one formula covering these four formulas see the linear dependence property in $\$ 5$.
(4.5.3) $125 \wedge 345+135 \wedge 245+145 \wedge 235=1$.
See Figure 3, right. Hint: deduce from (B) below.
Remark. (A) For any pairwise distinct points $A_{1}, A_{2}, A_{3}, A_{4}$ in the line there is exactly one 'intertwined' coloring into two colors.
(B) For any pairwise distinct points $A_{1}, A_{2}, A_{3}, A_{4}$ on the circle
$$
\left|A_{1} A_{2} \cap A_{3} A_{4}\right|+\left|A_{1} A_{3} \cap A_{2} A_{4}\right|+\left|A_{1} A_{4} \cap A_{2} A_{3}\right|=1
$$
(B') For any 'general position' map $f: K_{5} \rightarrow \mathbb{R}^{2}$ the number of intersection points in $\mathbb{R}^{2}$ formed by images of disjoint edges is odd.
A simple deduction of $(A) \Rightarrow\left(B^{\prime}\right)$ is presented in [Sk14] (for the linear case; for the PL case the deduction is analogous). Observe that ( $\left.\mathrm{B}^{\prime}\right)$ does not follow from Euler formula for planar graphs. Analogously, the non- $\mathbb{Z}_{2}$-embeddability to surfaces (unlike the nonembeddability) does not follow from the Euler inequality for surfaces [Sk20, §2.4].
## 5 Rank of matrix with relations
We shorten $\{i\}$ to $i$. An $\binom{[m]}{3}$-matrix is a symmetric square matrix with $\mathbb{Z}_{2}$-entries whose rows and whose columns correspond to all 3 -element subsets of $[m]$, and for which the following properties hold:
(triviality) $A_{P, Q}=0$ if $P \cap Q=\emptyset$;
(linear dependence) for each 4-element and 3-element subsets $F, P \subset[m]$
$$
\sum_{i \in F} A_{F-i, P}=0
$$
(non-triviality) for each $i \in[m]$ and 4-element subset $F \subset[m]-i$ we have $A_{F, i}=1$, where
$$
A_{F, i}:=\sum_{\{X, Y\}: F \cup i=X \cup Y, X \cap Y=i,|X|=|Y|=3} A_{X, Y}=\sum_{\{\sigma, \tau\}: F=\sigma \sqcup \tau,|\sigma|=|\tau|=2} A_{i \sqcup \sigma, i \sqcup \tau}
$$
By Assertion 4.5, an $\binom{[m]}{3}$-matrix is constructed from a $\mathbb{Z}_{2}$-embedding $f: K_{m} \rightarrow S$ to a surface. Indeed, set $A_{\{i, j, k\},\{p, q, r\}}:=i j k \wedge p q r$. If the surface $S$ is orientable, then the constructed matrix $A$ is even (i. e., $A_{P, P}=0$ for each 3 -element subset $P \subset[m]$ ).
Theorem 5.1. (a) If $A$ is an $\binom{[m]}{3}$-matrix, then $\operatorname{rk} A \geq \frac{m-4}{3}$.
(b) If, moreover, $A$ is even, then $\operatorname{rk} A \geq \frac{2(m-4)}{5}$.
You can submit separately proofs of the following particular cases of Theorem 5.1. For more strong estimates see $\S 7$.
5.2. (a)* There are no $\binom{[7]}{3}$-matrices of rank 1 .
(b)* There are no even $\binom{[8]}{3}$-matrices of rank smaller than 3 .
You can deduce Theorem 5.1 from Proposition 5.7.a,b.
The following assertion is not used in the sequel. In its proof you do not need to explicitly give the matrix, just describe the construction. We only know a proof using Assertions 4.1, 4.5, and Theorem 4.4.
5.3. (a) There is a non-zero $\binom{[4]}{3}$-matrix.
(b) There is a $\binom{[5]}{3}$-matrix.
(c) For any $m \geq 5$ there is an $\binom{[m]}{3}$-matrix.
$\left(a^{\prime}, b^{\prime}, c^{\prime}\right)$ The same for even matrices.
(d) There is a $\binom{[5]}{3}$-matrix of rank 1 .
(e) There is an even $\binom{[5]}{3}$-matrix of rank 2.
(f) There is a $\binom{[6]}{3}$-matrix of rank 1 .
(g)* There is an $\binom{[8]}{3}$-matrix of rank greater than 2.
5.4. Let $A^{\prime}$ be the square matrix of size $\binom{m-1}{3}$ obtained from an $\binom{[m]}{3}$-matrix by deleting rows and columns corresponding to all subsets containing $m$. Then $A^{\prime}$ is an $\binom{[m-1]}{3}$-matrix.
5.5. (a) Let $B$ be the square matrix of size $\binom{m-3}{3}$ obtained from an $\binom{[m]}{3}$-matrix $A$ by deleting rows and columns corresponding to subsets containing at least one element of $X:=\{m, m-$ $1, m-2\}$. If $A_{X, X}=1$, then $\operatorname{rk} A>\operatorname{rk} B$.
(b) Let $C$ be the square matrix obtained from an $\binom{[m]}{3}$-matrix $A$ by deleting rows and columns corresponding to subsets containing at least one element of certain 3-element subsets $X, Y \subset[m]$. If $A_{X, X}=A_{Y, Y}=0$ and $A_{X, Y}=1$, then $\operatorname{rk} A \geq \operatorname{rk} C+2$.
Denote by $r_{m}$ the minimal rank of an $\binom{[m]}{3}$-matrix. Denote by $\widetilde{r_{m}}$ the minimal rank of an even $\binom{[m]}{3}$-matrix. Clearly, $r_{m}=\widetilde{r_{m}}=0$ for $m \leq 4$, and $r_{m} \leq \widetilde{r_{m}}$. The non-triviality implies that $r_{5}, \widetilde{r_{5}} \geq 1$. Theorem 5.1 asserts that $r_{m} \geq \frac{m-4}{3}$ and $\widetilde{r_{m}} \geq \frac{2(m-4)}{5}$.
5.6. (a,b) Find $r_{5}, r_{6}$ and $\widetilde{r_{5}}, \widetilde{r_{6}}, \widetilde{r_{7}}$.
(c) Both sequences $r_{m}, \widetilde{r_{m}}$ are non-decreasing.
Proposition 5.7. (a) $r_{m} \geq \min \left\{r_{m-3}+1, \widetilde{r_{m}}\right\}$ (more precisely, either $r_{m}=\widetilde{r_{m}}$ or $r_{m} \geq$ $\left.r_{m-3}+1\right)$;
(b) $\widetilde{r_{m}} \geq \widetilde{r_{m-5}}+2$.
## 6 Classification of symmetric bilinear forms
Fix a symmetric matrix $A \in \mathbb{Z}_{2}^{n \times n}$. For $U, V \in \mathbb{Z}_{2}^{n}$ let
$$
A(U, V)=U \cdot{ }_{A} V:=\sum_{i, j=1}^{n} A_{i, j} U_{i} V_{j} \quad\left(=U^{T} A V\right)
$$
A basis of $\mathbb{Z}_{2}^{n}$ is an inclusion-minimal ordered set of vectors such that every vector from $\mathbb{Z}_{2}^{n}$ is the sum of some vectors from this set.
Theorem 6.1. For $n=2$ there is a basis $X_{1}, X_{2}$ of $\mathbb{Z}_{2}^{2}$ and numbers $\gamma_{1}, \gamma_{2} \in \mathbb{Z}_{2}$ such that either
(i) for any $a_{1}, a_{2}, b_{1}, b_{2} \in \mathbb{Z}_{2}$ we have
$$
\left(a_{1} X_{1}+a_{2} X_{2}\right) \cdot{ }_{A}\left(b_{1} X_{1}+b_{2} X_{2}\right)=\gamma_{1} a_{1} b_{1}+\gamma_{2} a_{2} b_{2}, \quad \text { or }
$$
(ii) for any $a_{1}, a_{2}, b_{1}, b_{2} \in \mathbb{Z}_{2}$ we have
$$
\left(a_{1} X_{1}+a_{2} X_{2}\right) \cdot{ }_{A}\left(b_{1} X_{1}+b_{2} X_{2}\right)=a_{1} b_{2}+a_{2} b_{1}
$$
Recall that problems stated after theorems are hints to proofs of the theorems.
6.2. Assume that $n=2, X \in \mathbb{Z}_{2}^{2}$ and $X \cdot{ }_{A} X=1$.
(a) For any $P \in \mathbb{Z}_{2}^{2}$ there is $\lambda_{X, P} \in \mathbb{Z}_{2}$ such that for $P_{X}:=P+\lambda_{X, P} X$ we have $P_{X} \cdot{ }_{A} X=0$.
(b) There is a basis $X_{1}=X$, $X_{2}$ of $\mathbb{Z}_{2}^{2}$ and numbers $\gamma_{1}=1, \gamma_{2} \in \mathbb{Z}_{2}$ such that the property (i) of Theorem 6.1 holds.
6.3. Assume that $n=2, X, Y \in \mathbb{Z}_{2}^{2}$ and $X \cdot{ }_{A} Y=1, X \cdot{ }_{A} X=Y \cdot{ }_{A} Y=0$. Then $X_{1}:=X$, $Y_{1}:=Y$ is a basis of $\mathbb{Z}_{2}^{2}$ such that the property (ii) of Theorem 6.1 holds.
Theorem 6.4. For $n=3$ there is a basis $X_{1}, X_{2}, X_{3}$ of $\mathbb{Z}_{2}^{3}$ and numbers $\gamma_{1}, \gamma_{2}, \gamma_{3} \in \mathbb{Z}_{2}$ such that either
(i) for any $a_{1}, a_{2}, a_{3}, b_{1}, b_{2}, b_{3} \in \mathbb{Z}_{2}$ we have
$$
\left(a_{1} X_{1}+a_{2} X_{2}+a_{3} X_{3}\right) \cdot{ }_{A}\left(b_{1} X_{1}+b_{2} X_{2}+b_{3} X_{3}\right)=\gamma_{1} a_{1} b_{1}+\gamma_{2} a_{2} b_{2}+\gamma_{3} a_{3} b_{3}, \quad \text { or }
$$
(ii) for any $a_{1}, a_{2}, a_{3}, b_{1}, b_{2}, b_{3} \in \mathbb{Z}_{2}$ we have
$$
\left(a_{1} X_{1}+a_{2} X_{2}+a_{3} X_{3}\right) \cdot{ }_{A}\left(b_{1} X_{1}+b_{2} X_{2}+b_{3} X_{3}\right)=a_{1} b_{2}+a_{2} b_{1}+\gamma_{3} a_{3} b_{3}
$$
6.5. Assume that $X, Y \in \mathbb{Z}_{2}^{3}$ and $X \cdot{ }_{A} Y=1, X \cdot{ }_{A} X=Y \cdot{ }_{A} Y=0$.
(a) For any $P \in \mathbb{Z}_{2}^{3}$ there are $\lambda_{X, Y, P}, \lambda_{Y, X, P} \in \mathbb{Z}_{2}$ such that for $P_{X, Y}:=P+\lambda_{X, Y, P} Y+$ $\lambda_{Y, X, P} X$ we have $P_{X, Y} \cdot{ }_{A} X=P_{X, Y} \cdot{ }_{A} Y=0$.
(b) There is a basis $X_{1}=X, X_{2}=Y, X_{3}$ of $\mathbb{Z}_{2}^{3}$ and a number $\gamma_{3} \in \mathbb{Z}_{2}$ such that the property (ii) of Theorem 6.4 holds.
Theorem 6.6. There are $k, l$ and a basis $X_{1}, Y_{1}, \ldots, X_{k}, Y_{k}, Z_{1}, \ldots, Z_{n-2 k}$ of $\mathbb{Z}_{2}^{n}$ such that $2 k+l \leq n$ and for any $a, a^{\prime}, b, b^{\prime} \in \mathbb{Z}_{2}^{k}$ and $c, c^{\prime} \in \mathbb{Z}_{2}^{n-2 k}$ we have
$$
\begin{gathered}
\left(a_{1} X_{1}+b_{1} Y_{1}+\ldots+a_{k} X_{k}+b_{k} Y_{k}+c_{1} Z_{1}+\ldots+c_{n-2 k} Z_{n-2 k}\right) \cdot A \\
\cdot_{A}\left(a_{1}^{\prime} X_{1}+b_{1}^{\prime} Y_{1}+\ldots+a_{k}^{\prime} X_{k}+b_{k}^{\prime} Y_{k}+c_{1}^{\prime} Z_{1}+\ldots+c_{n-2 k}^{\prime} Z_{n-2 k}\right)= \\
=a_{1} b_{1}^{\prime}+a_{1}^{\prime} b_{1}+\ldots+a_{k} b_{k}^{\prime}+a_{k}^{\prime} b_{k}+c_{1} c_{1}^{\prime}+\ldots+c_{l} c_{l}^{\prime}
\end{gathered}
$$
If $A$ is even, then $l=0$.
6.7. Assume that $X \in \mathbb{Z}_{2}^{n}$ and $X \cdot{ }_{A} X=1$.
(a) State and prove the n-dimensional analogue of Assertion 6.2.a.
(b) There is a basis $X, E_{1}, \ldots, E_{n-1}$ of $\mathbb{Z}_{2}^{n}$ and a symmetric matrix $B \in \mathbb{Z}_{2}^{(n-1) \times(n-1)}$ such that for any $a, b \in \mathbb{Z}_{2}$ and $\lambda, \mu \in \mathbb{Z}_{2}^{n-1}$ we have
$$
\left(a X+\lambda_{1} E_{1}+\ldots+\lambda_{n-1} E_{n-1}\right) \cdot{ }_{A}\left(b X+\mu_{1} E_{1}+\ldots+\mu_{n-1} E_{n-1}\right)=a b+\lambda \cdot_{B} \mu
$$
6.8. Assume that $X, Y \in \mathbb{Z}_{2}^{n}$ and $X \cdot{ }_{A} Y=1, X \cdot{ }_{A} X=Y \cdot{ }_{A} Y=0$.
(a) State and prove the n-dimensional analogue of Assertion 6.5.a.
(b) There is a basis $X, Y, E_{1}, \ldots, E_{n-2}$ of $\mathbb{Z}_{2}^{n}$ and a symmetric matrix $B \in \mathbb{Z}_{2}^{(n-2) \times(n-2)}$ such that for any $a_{X}, a_{Y}, b_{X}, b_{Y} \in \mathbb{Z}_{2}$ and $\lambda, \mu \in \mathbb{Z}_{2}^{n-2}$ we have
$$
\left(a_{X} X+a_{Y} Y+\lambda_{1} E_{1}+\ldots+\lambda_{n-2} E_{n-2}\right) \cdot{ }_{A}\left(b_{X} X+b_{Y} Y+\mu_{1} E_{1}+\ldots+\mu_{n-2} E_{n-2}\right)=a_{X} b_{Y}+a_{Y} b_{X}+\lambda \cdot_{B} \mu
$$
## 7 Rank of matrix with relations: generalization
The following results are 'higher-dimensional' (and more strong) generalizations of Theorem 5.1, Assertions 5.4 and 5.5, and Proposition 5.7. They give a simplified well-structured exposition of [PT19, Theorem 1].
An $\binom{[m]}{l}$-matrix is a symmetric square matrix with $\mathbb{Z}_{2}$-entries whose rows and whose columns correspond to all $l$-element subsets of $[m]$, and for which (triviality) and the following properties hold:
(linear dependence) for each $(l+1)$-element and $l$-element subsets $F, P \subset[m]$
$$
\sum_{i \in F} A_{F-i, P}=0
$$
(non-triviality) for each $i \in[m]$ and $(2 l-2)$-element subset $F \subset[m]-i$ we have $A_{F, i}=1$, where
$$
A_{F, i}:=\sum_{\{X, Y\}: F \cup i=X \cup Y, X \cap Y=i,|X|=|Y|=l} A_{X, Y}=\sum_{\{\sigma, \tau\}: F=\sigma \sqcup \tau,|\sigma|=l-1} A_{i \sqcup \sigma, i \sqcup \tau}
$$
Analogously to Assertion 4.5, an $\binom{[m]}{l}$-matrix is constructed by a $\mathbb{Z}_{2}$-embedding of the $(l-1)$-dimensional skeleton of the $(m-1)$-dimensional simplex to a $2(l-1)$-dimensional manifold.
Theorem 7.1. Suppose $l \geq 3$ and $A$ is an $\binom{[m]}{l}$-matrix.
(a) Then $\operatorname{rk} A \geq \frac{m-2 l+2}{l-1}$.
(b) If, moreover, $A$ is even, then $\operatorname{rk} A \geq \frac{2(m-2 l+2)}{l}$.
You can deduce Theorem 7.1 from Propositions 7.4.a,b.
7.2. Let $A^{\prime}$ be the square matrix of size $\binom{m-1}{l}$ obtained from an $\binom{[m]}{l}$-matrix by deleting rows and columns corresponding to all subsets containing $m$. Then $A^{\prime}$ is an $\binom{[m-1]}{l}$-matrix.
7.3. Let $A$ be an $\binom{[m]}{l}$-matrix and $X:=\{m-l+1, m-l+2, \ldots, m\}$.
$\left(a, b\right.$ ') Let $B$ be the square matrix of size $\binom{m-l}{l}$ obtained from $A$ by deleting rows and columns corresponding to subsets containing at least one of the elements of $X$.
If $A_{X, X}=1$, then $\operatorname{rk} A>\operatorname{rk} B$.
If $A_{X, X}=A_{Y, Y}=0$ and $A_{X, Y}=1$ for some $Y \subset[m]$, then $\operatorname{rk} A \geq \operatorname{rk} B+2$.
(b) Let $C$ be the square matrix obtained from $A$ by deleting rows and columns corresponding to subsets containing at least one element of $X$ or of certain l-element subset $Y \subset[m]$. If $A_{X, X}=A_{Y, Y}=0$ and $A_{X, Y}=1$, then $\operatorname{rk} A \geq \operatorname{rk} C+2$.
(a') For l-element subsets $P, Q \subset[m-l+1]$ define
$$
D_{P, Q}:=A_{P, Q}+A_{P, X} A_{Q, X}
$$
If $A_{X, X}=1$, then $\operatorname{rk} D<\operatorname{rk} A$ and $D$ is an $\binom{[m-l+1]}{l}$-matrix.
Assertions 7.3.a,b are only required to illustrate the idea of Assertions 7.3.a',b' by proving much easier results giving estimates $\operatorname{rk} A \geq \frac{m-2 l+2}{l}$ and, for $A$ even, $\operatorname{rk} A \geq \frac{2(m-2 l+2)}{2 l-1}$.
Denote by $r_{m}$ the minimal rank of an $\left(\begin{array}{c}{\left[\begin{array}{c}m \\ l\end{array}\right)}\end{array}\right)$-matrix. Denote by $\widetilde{r_{m}}$ the minimal rank of an even $\binom{[m]}{l}$-matrix. Clearly, $r_{m}=\widetilde{r_{m}}=0$ for $m \leq 2 l-2$, both sequences $r_{m}, \widetilde{r_{m}}$ are non-decreasing, and $r_{m} \leq \widetilde{r_{m}}$. The non-triviality implies that $r_{2 l-1}, \widetilde{r_{2 l-1}} \geq 1$. Theorem 7.1 asserts that $r_{m} \geq \frac{m-2 l+2}{l-1}$ and $\widetilde{r_{m}} \geq \frac{2(m-2 l+2)}{l}$.
Proposition 7.4. (a) $r_{m} \geq \min \left\{r_{m-l+1}+1, \widetilde{r_{m}}\right\}$ (more precisely, either $r_{m}=\widetilde{r_{m}}$, or $r_{m} \geq$ $\left.r_{m-l+1}+1\right)$;
(b) $\widetilde{r_{m}} \geq \widetilde{r_{m-l}}+2$.
Proof of Proposition 7.4.a also uses an algebraic version (b) of the higher-dimensional analogue of the following result (a).
Proposition 7.5. (a) Denote by $X=\binom{(55)}{2}$ the set of unordered pairs of 2-element subsets of [5]. For any $i \in[5]$ and a partition [5] $-i=\sigma \sqcup \tau$ into disjoint 2-element sets denote
$$
T_{i,\{\sigma, \tau\}}:=\{\{\alpha, \beta\} \in X: \alpha \subset \sigma \sqcup i, \beta \subset \tau \sqcup i\}
$$
Denote by $A_{i}$ the sum modulo 2 (i. e., the symmetric difference) of sets $T_{i,\{\sigma, \tau\}}$ over all non-ordered partitions [5] $-i=\sigma \sqcup \tau$ as above. Then
$$
A_{i}=\{\{\alpha, \beta\} \in X: \alpha \cap \beta=\emptyset\}
$$
and so is independent of $i$.
(b) Let $A$ be a symmetric square matrix with $\mathbb{Z}_{2}$-entries whose rows and whose columns correspond to all l-element subsets of $[m]$. If $A$ satisfies the linear dependence property (from the definition of an $\binom{[m]}{l}$-matrix), then $A_{F, i}$ depends only on $F \sqcup i$ not on $(F, i)$.
## Hints and solutions to some problems
See proof of Proposition 1.1 in [Bi20].
1.2. (a) Change the numbers on the main diagonal of $M$ so that the sum of the entries in each row is even. The resulting matrix is degenerate.
(b) Use induction. See Lemma 2.5.b.
1.3. (a) Any matrix formed as a result of putting numbers from $\mathbb{Z}_{2}$ in the elements on the main diagonal of $M$ can be uniquely represented as the sum $M+D$ where $D$ is a diagonal matrix. By Lemma 3.7.b for every diagonal matrix $D$ with more than $k$ zeros on the main diagonal we have $\operatorname{rk}(M+D) \geq \operatorname{rk} M-\operatorname{rk} D>n-(n-k)=k$.
(b) The algorithm of (b) is constructed using (a) and Lemma 2.5.b. The algorithm given by Lemma 2.5.b has complexity $O\left(n^{4}\right)$. There is an algorithm searching through all diagonal $n \times n$ matrices with $\leq k$ zeroes on the main diagonal with the complexity of
$$
O\left(n\binom{n}{0}+n\binom{n}{1}+\ldots+n\binom{n}{k}\right) \stackrel{(*)}{=} O\left((k+1) n\binom{n}{k}\right)=O\left(n \cdot n^{k}\right)=O\left(n^{k+1}\right)
$$
Here ${ }^{*}$ ) holds because we may assume that $n \geq 2 k$. Thus, by Assertion 3.6.b the complexity of the whole algorithm is $O\left(n^{4}\right)+O\left(n^{k+1} n^{2}\right)=O\left(n^{k+3}\right)$ (since $\left.k \geq 1\right)$.
1.4. (a) Denote by $n$ the number of columns of $M$ and of $D$. By Lemma 3.7.b we have
$$
\begin{aligned}
& 2 \operatorname{rk}(M+D)=\operatorname{rk}(M+D)+\operatorname{rk}((M+E)+(E+D)) \\
& \geq(\operatorname{rk} M-\operatorname{rk} D)+(\operatorname{rk}(M+E)-\operatorname{rk}(E+D)) \\
& \quad=n-\operatorname{rk} D+\operatorname{rk}(M+E)-(n-\operatorname{rk} D)=\operatorname{rk}(M+E)
\end{aligned}
$$
(b) Let $M_{n}$ be the matrix obtained by applying the algorithm of Lemma 2.5.b to the matrix $M$. Let $k:=\operatorname{rk}\left(M_{n}+E\right)$. We have $R(M)=\operatorname{rk}(M+D)$ for some diagonal matrix $D$. Hence by (a) $k / 2 \leq R(M) \leq k$ as required.
The number $k$ can be computed in time $O\left(n^{3}\right)$. Hence the total complexity of the algorithm is $O\left(n^{4}\right)+O\left(n^{3}\right)=O\left(n^{4}\right)$.
2.1. Answers: $A_{1}, A_{2}$ and $A_{3}$ are degenerate, while $A_{4}$ is non-degenerate.
2.2. Hints: (a)-(b) track the maximal non-degenerate submatrix; (c) use induction.
Part (a) is clear.
(b) For a matrix $M$ denote by $\operatorname{row}_{i \rightarrow i+j} M$ the matrix obtained from $M$ by replacing the $i$ th row by the sum of the $i$ th row and the $j$ th row. The matrix $\operatorname{col}_{i \rightarrow i+j} M$ is defined similarly.
It is clear that to prove part (b) we have to show that the matrices $M, \operatorname{row}_{i \rightarrow i+j} M$ and $\mathrm{col}_{i \rightarrow i+j} M$ are degenerate or not simultaneously.
Next, observe that $\operatorname{row}_{i \rightarrow i+j} \operatorname{row}_{i \rightarrow i+j} M=M=\operatorname{col}_{i \rightarrow i+j} \operatorname{col}_{i \rightarrow i+j} M$. Thus it suffices to prove that if $M$ is degenerate then both $\operatorname{row}_{i \rightarrow i+j} M$ and $\operatorname{col}_{i \rightarrow i+j} M$ are degenerate.
Assume that the sum of columns $c_{1}, c_{2}, \ldots, c_{s}$ of $M$ equals zero.
Then the sum of the 'same' columns of $\operatorname{row}_{i \rightarrow i+j} M$ equals zero.
If $i \notin\left\{c_{1}, \ldots, c_{s}\right\}$ then the sum of 'the same' columns of $\operatorname{col}_{i \rightarrow i+j} M$ equals zero. If $i, j \in\left\{c_{1}, \ldots, c_{s}\right\}$ then the sum of columns indexed by $\left\{c_{1}, \ldots, c_{s}\right\}-\{j\}$ of $\operatorname{col}_{i \rightarrow i+j} M$ equals zero. If $i \in\left\{c_{1}, \ldots, c_{s}\right\}$ and $j \notin\left\{c_{1}, \ldots, c_{s}\right\}$ then the sum of columns indexed by $\left\{c_{1}, \ldots, c_{s}, j\right\}$ of $\operatorname{col}_{i \rightarrow i+j} M$ equals zero.
This completes the proof of (b).
(c) We will show explicitly how to produce a diagonal matrix out of $M$.
If all entries of $M$ are 0 then $M$ is already diagonal. If $M$ has a non-zero entry then we place this entry in the top-left corner by permuting the row of this entry with the top row, and the column of this entry with the left column. For the obtained matrix add the top row to other rows and the left column to other columns. All entries in the left column and the top row except the the top-left entry become zeros. Delete the top row and the left column of the obtained matrix.
Repeat the procedure inductively for the obtained submatrix. In the end this will produce a diagonal matrix.
(d) By part (c) we can change $M$ into a diagonal matrix using transformations from parts (a, b); also $M$ is degenerate if and only if the new matrix is degenerate. It remains to mention that a diagonal matrix is non-degenerate iff it is the identity matrix.
(f) This follows from (a)-(d).
(g) The algorithm is constructed in the solution of part (c). The algorithm has $n$ major steps, a single major step is described in the second paragraph of (c). Each major step requires at most one permutation of rows, at most one permutation of columns and up to $2 n$ additions of rows and columns. Thus the complexity of the whole algorithm is $O(n)+n$. $O\left(n^{2}\right)=O\left(n^{3}\right)$.
2.3. (a) The formula follows because a matrix from $\mathbb{Z}_{2}^{2 \times 2}$ is degenerate if and only if either it has a zero row, or it has a zero column, or rows are the same and columns are the same (in the latter case all entries are ones).
Alternatively, here are all matrices from $\mathbb{Z}_{2}^{2 \times 2}$ up to permutations of rows and columns:
$$
\left(\begin{array}{ll}
1 & 0 \\
0 & 1
\end{array}\right),\left(\begin{array}{ll}
1 & 1 \\
0 & 1
\end{array}\right),\left(\begin{array}{ll}
0 & 0 \\
0 & 0
\end{array}\right),\left(\begin{array}{ll}
1 & 1 \\
1 & 1
\end{array}\right),\left(\begin{array}{ll}
1 & 1 \\
0 & 0
\end{array}\right),\left(\begin{array}{ll}
1 & 0 \\
1 & 0
\end{array}\right),\left(\begin{array}{ll}
1 & 0 \\
0 & 0
\end{array}\right)
$$
The first two are non-degenerate, and the other are degenerate. It is easy to verify the formula for each of them.
(d) This follows by $(\mathrm{b}, \mathrm{c})$.
Here is an alternative direct proof. Consider $n \times n$ chessboard. A rook placement for such a chessboard is a placement of $n$ rooks on that board with the condition that they do not beat each other. A rook $M$-placement for such a chessboard is a rook placement such that all rooks are staying on cells corresponding to unit entries of $M$. Denote by det* $M$ the parity of the amount of rook $M$-placements. Then (d) can be restated as follows: $\operatorname{det} M=\operatorname{det}^{*} M$. This follows because
- transformations of 2.2.a, 2.2.b preserve $\operatorname{det}^{*} M$, and
- $\operatorname{det} M^{\prime}=\operatorname{det}^{*} M^{\prime}$ for a diagonal matrix $M^{\prime}$.
2.4. For any degenerate matrix of Problem 2.1 we show how to change entries on the main diagonal to make it non-degenerate; we show the opposite for $A_{4}$ :
$$
A_{1} \rightarrow\left(\begin{array}{ccc}
1 & 0 & 0 \\
0 & 1 & 0 \\
0 & 0 & 1
\end{array}\right), \quad A_{2}, A_{3} \rightarrow\left(\begin{array}{ccc}
0 & 1 & 1 \\
1 & 0 & 1 \\
1 & 1 & 1
\end{array}\right), \quad A_{4} \rightarrow\left(\begin{array}{lll}
0 & 1 & 1 \\
0 & 1 & 1 \\
1 & 0 & 0
\end{array}\right)
$$
2.5. (a) In the following paragraph we prove by induction on $i \geq 1$ that the determinant $\Delta_{i}:=\operatorname{det} M_{[i] \times[i]}^{(i)}$ of the left upper $i \times i$-corner submatrix of $M^{(i)}$ is equal to 1 . Then $\operatorname{det} M^{(n)}=\Delta_{n}=1$.
Base $i=1$ follows because $\Delta_{1}=1+\delta_{0}=1$. Let us prove the inductive step $i-1 \rightarrow i$. Apply the decomposition formula for the determinant $\Delta_{i}$ by the last row of the corresponding
submatrix of $M^{(i)}$ (Assertion 2.3.c). Since $M_{i, i}^{(i-1)}=M_{i, i}=0$ and $\Delta_{i-1}=1$, we have $\Delta_{i}=\delta_{i}+\left(1+\delta_{i}\right) \Delta_{i-1}=1$.
(b) The algorithm is constructed by (a). The algorithm is essentially a computation of the determinants of $n$ square submatrices of sizes $1,2, \ldots, n$. Hence by Assertion 2.2.g its complexity is $O\left(1^{3}+2^{3}+\ldots+n^{3}\right)=O\left(n \cdot n^{3}\right)=O\left(n^{4}\right)$.
3.1. Answers: (a1) 0; (a2) 1 ; (a3), (a4) 3 ; (b1) 0 ; (b2), (b3) 1 ; (b4) 2 .
3.2. Hint to (b): find the number of sums of $k$ columns.
Part (a) follows from the definition of $\mathrm{rk} M$.
(b) By definition of rk the number of different sums of columns of $M$ is $2^{\mathrm{rk} M}$. On the other hand the number of such sums does not exceed $2^{k}$. Therefore $2^{k} \geq 2^{\mathrm{rk} M}$, hence $k \geq \operatorname{rk} M$.
3.3. The proofs of are similar to the proofs of Assertion 2.2.
3.4. Part (a) is clear.
(b) If for a non-zero symmetric matrix $M$ there exists such a permutation of rows and columns, then $\mathrm{rk} M=1$ by Assertion 3.2.b.
We now take a symmetric matrix $M$ of rank 1 . As the required permutation we can take any permutation mapping non-zero rows of $M$ to the first rows. Indeed, take any non-zero rows $i, j$. If $M_{i, j}=0$ then there exists a non-zero row $k$ such that $M_{i, k}=1$. Hence the $j$ th and the $k$ th rows are distinct non-zero rows of the matrix $M$ of rank one. A contradiction. Hence $M_{i, j}=1$.
(c) Pick a nonzero row and apply the above argument.
3.5. (a) $R\left(\begin{array}{lll}1 & 0 & 1 \\ 0 & 0 & 1 \\ 0 & 1 & 0\end{array}\right)=2, \quad R\left(\begin{array}{lll}1 & 0 & 1 \\ 0 & 1 & 0 \\ 0 & 0 & 1\end{array}\right)=1$.
(b) $R\left(\begin{array}{lll}1 & 0 & 1 \\ 0 & 1 & 1 \\ 0 & 1 & 0\end{array}\right)=2, \quad R\left(\begin{array}{lll}1 & 0 & 1 \\ 0 & 1 & 0 \\ 0 & 0 & 1\end{array}\right)=1$.
3.6. Hint for (a): cf. Assertion 2.2.
(a) The algorithm from the proof of 2.2.c provides a diagonal matrix of the same rank, and has the required complexity. The rank of a diagonal matrix is equal to the number of non-zero entries in it.
(b) We shall construct a set $S_{k}$ of columns such that
- these columns constitute a non-degenerate submatrix;
- the first $k$ columns of the matrix $M$ are sums of several columns from the set $S_{k}$.
If $\left|S_{k}\right|>r$ for some $k=1, \ldots, n$ then the answer is ' $\mathrm{NO}$ '. If for all $k=1, \ldots, n$ we have $\left|S_{k}\right| \leq r$ then the answer is 'Yes'. The answer is correct because $\left|S_{1}\right| \leq\left|S_{2}\right| \leq \ldots \leq\left|S_{n}\right|$, and because $\left|S_{n}\right|>r$ is equivalent to $\mathrm{rk} M>r$.
Set $S_{1}:=\emptyset$ if the first column of $M$ is 0 then, and $S_{1}:=\{1\}$ otherwise.
Let us define $S_{k+1}$ from $S_{k}$. We form the set of all sums of columns of $M$ with indices from $S_{k}$ (it takes $O(n)$ operations because $\left|S_{k}\right| \leq r$ ). Then we compare the $(k+1)$ st column of $M$ with all sums from this set (this will take at most $2^{r} O(n)=O(n)$ operations). If the $(k+1)$ st column of $M$ equals to at least one of the sums then $S_{k+1}:=S_{k}$. Otherwise we set $S_{k+1}:=S_{k} \cup\{k+1\}$.
It is easy to verify that the total complexity of the algorithm is $O\left(n^{2}\right)$.
3.7. Part (b) follows from (a) because
$$
\operatorname{rk} M=\operatorname{rk}(M+D+D) \leq \operatorname{rk}(M+D)+\operatorname{rk} D \quad \Rightarrow \quad \operatorname{rk}(M+D) \geq \operatorname{rk} M-\operatorname{rk} D
$$
We now prove (a). Choose columns from Assertion 3.2.a for $M$ and for $D$. Then every column of $M+D$ is the sum of some of chosen $\operatorname{rk} M+\operatorname{rk} D$ columns. By Assertion 3.2.b $\operatorname{rk}(M+D) \leq \operatorname{rk} M+\operatorname{rk} D$.
3.8. The statement and the proof of Assertion 3.8 are similar to the statement and the proof of Theorem 1.3.b.
3.9. Answer: no. If $M$ is a $3 \times 3$-matrix given next and $k=1$ then the statement is false.
$$
\left(\begin{array}{lll}
0 & 1 & 1 \\
1 & 0 & 1 \\
1 & 1 & 0
\end{array}\right)
$$
3.10. Answers: (a) 1 ; (b) $\left(2^{n}-1\right)^{2}$; (c) $\left(2^{n}-1\right)^{2}\left(2^{n}-2\right)^{2} / 6$.
(a) There exists only one matrix of rank 0 , the matrix all whose entries are zeroes.
(b) For the matrix of rank 1 all columns containing a non-zero entry are the same. Hence such matrices are in 1-1 correspondence with ordered pairs formed by
- a non-empty subset of the set of columns ('non-zero columns'), and
- a nonzero vector in $v \in \mathbb{Z}_{2}^{n}$ ('column vector').
Therefore there are $\left(2^{n}-1\right)^{2}$ such matrices.
(c) Fix a matrix $M$ of rank 2. Then there exists a pair $(v, w)$ of columns of $M$ forming a non-degenerate matrix. Any other column is either 0 or $v$, or $w$, or $v+w$ (see Assertion 3.2). This set $S=S_{M}$ of four vectors does not depend on a choice of the two columns $v, w$; we call it the column span of $M$. (This is a 2-dimensional vector subspace of $\mathbb{Z}_{2}^{n}$.)
Each column span is defined by any ordered pair of non-zero vectors in it. Each column span contains exactly 6 such ordered pairs. Hence there are $\left(2^{n}-1\right)\left(2^{n}-2\right) / 6$ column spans. In the following paragraph we prove that there are exactly $\left(2^{n}-1\right)\left(2^{n}-2\right)$ matrices of rank 2 for a given column span. Hence there are $\left(2^{n}-1\right)^{2}\left(2^{n}-2\right)^{2} / 6$ matrices of rank 2 .
First proof. To a matrix $M$ there correspond the set $X$ of columns of $M$ equal to $v$ or to $v+w$, and the set $Y$ of columns of $M$ equal $w$ or $v+w$. Since $\operatorname{rk} M=2$, both sets are non-empty and $X \neq Y$. Moreover, matrix $M$ can be reconstructed from $X, Y$. There are $\left(2^{n}-1\right)\left(2^{n}-2\right)$ pairs $(X, Y)$ of distinct non-empty subsets.
Second proof. For a given 4-element set $S=\{0, v, w, v+w\}$, regard a matrix $M$ with the column span $S$ as a map $\phi_{M}$ from the set $[n]$ of columns to $S$. We have $\operatorname{rk} M=2$ if and only if
$\left(^{*}\right)$ the image of $\phi_{M}$ contain at least two of vectors $v, w, v+w$.
There are $4^{n}$ maps $[n] \rightarrow S$. There are $2^{n}$ maps $[n] \rightarrow\{0, v\}$. The same holds for $\{0, v\}$ replaced either by $\{0, w\}$ or by $\{0, v+w\}$. There is only one map $[n] \rightarrow\{0\}$. Hence there are exactly $4^{n}-3 \cdot 2^{n}+2=\left(2^{n}-1\right)\left(2^{n}-2\right)$ maps satisfying the condition $(*)$.
Remark. More generally, the number of matrices of $\operatorname{rank} k$ in $\mathbb{Z}_{2}^{m \times n}$ is equal to
$$
\frac{2^{k(k-1) / 2} \prod_{i=0}^{k-1}\left(2^{m-i}-1\right) \prod_{i=0}^{k-1}\left(2^{n-i}-1\right)}{\prod_{i=0}^{k-1}\left(2^{k-i}-1\right)}
$$
See theorem 7.1.5 in [ACM29, p. 299] (this theorem is even more general; for our case of matrices over $\mathbb{Z}_{2}$ take $\left.q=2, \operatorname{GF}(2)=\mathbb{Z}_{2}\right)$.
4.1. (a1-a3) A beautiful realization of the graph $K_{5}$ on the torus is shown in Figure 4, left. Realizations of $K_{6}$ and $K_{7}$ are analogous, see Figure 4, middle.
Another solutions (whose idea works for (c,d)). Draw the graph $K_{5}$ in the plane with one self-intersection point. In a small neighborhood of the point, attach a handle and lift one of the edges 'bridgelike' over the other edge to the handle, see Figure 5, left.

Figure 4: Realization of nonplanar graphs

Figure 5: Left: resolving intersection by adding a handle.
Right: a 'non-general position even drawing' of $K_{5}$ in the plane
(b2) See Figure. 4, right.
4.2. See Fig. 5, right.
5.1. (b) Induction on $m$. The base $m \leq 4$ is clear. By Assertion 5.7.b and induction hypothesis we have
$$
\widetilde{r_{m}} \geq \widetilde{r_{m-5}}+2 \geq \frac{2(m-5-4)}{5}+2=\frac{2(m-4)}{5}
$$
(a) Induction on $m$. The base $m \leq 4$ is clear. By Assertion 5.7.a, Theorem 5.1.b and induction hypothesis we have
$$
r_{m} \geq \min \left\{r_{m-3}+1, \widetilde{r_{m}}\right\} \geq \min \left\{\frac{m-3-4}{3}+1, \frac{2(m-4)}{5}\right\}=\frac{m-4}{3}
$$
5.3. Denote by $A^{f}$ the $\binom{[m]}{3}$-matrix constructed via an embedding $f: K_{m} \rightarrow S$, see the beginning of Section 5 .
(a) Consider the $K_{4}$-subgraph of $K_{5}$ with vertices 1, 2, 3, 4 together with its embedding to the torus defined by Figure 3 , right side. Then $A^{f}$ is a non-trivial $\binom{[m]}{3}$-matrix.
(b) Let $f$ be the embedding of $K_{5}$ into the torus defined by the right side of Figure 3. Then $A^{f}$ has the required property.
(c) By Assertion 4.1.d there exists an embedding $f: K_{m} \rightarrow S$ where $S$ is a sphere with several handles. Then $A^{f}$ has the required property.
(a', b', c') The matrices $A^{f}$ from (a, b, c) respectively satisfy the needed conditions.
Parts (d, e, f) follow from Assertion 5.6.a, b.
5.4. See proof of Assertion 7.2 below.
5.5. (a, b) See proof of Assertions 7.3.a,b below.
5.6. (a) Recall that if $m \geq 5$ then every $\binom{[m]}{3}$-matrix is not a zero matrix. This implies that $r_{5}, r_{6} \geq 1$. Let $f$ be an embedding defined by Assertion 4.1.b1, b2. It follows from Theorem 4.4 that $\mathrm{rk} A^{f} \leq 1$. Hence $r_{5}=r_{6}=1$.
(b) Assertion 3.4.c implies that $\widetilde{r_{5}}, \widetilde{r_{6}}, \widetilde{r_{7}} \geq 2$. We have $\operatorname{rk} A^{f}=2$ for $f$ defined by Assertion 4.1.a1, a2, a3 and hence $\widetilde{r_{5}}=\widetilde{r_{6}}=\widetilde{r_{7}}=2$.
(c) Let $A$ be an $\binom{[m]}{3}$-matrix and let $B$ be an $\binom{[m-1]}{3}$-submatrix of $A$ introduced in Assertion 5.4. We have $\operatorname{rk} B \leq \operatorname{rk} A$ and therefore $r_{m} \geq r_{m-1}$.
If $A$ is even then $B$ is even and therefore $\widetilde{r_{m}} \geq \widetilde{r_{m-1}}$.
5.7. (a) (Take $l=3$.) Take an $\binom{[m]}{l}$-matrix $A$ such that $\operatorname{rk} A=r_{m}$. If $A$ is even, then $r_{m}=\widetilde{r_{m}}$, so we are done. Otherwise there is an l-element subset $X \subset[m]$ such that $A_{X, X}=1$. Let $B$ be the 'restriction' of $A$ to $l$-element subsets of $[m]-X$.
Then
$$
r_{m}=\operatorname{rk} A \geq \operatorname{rk} B+1 \geq r_{m-l}+1, \quad \text { where }
$$
- the first inequality follows by Assertion 5.5.a;
- the second inequality holds because $B$ is a $\binom{[m]-X}{l}$-matrix by Assertion 5.4.
(b) (Take $l=3$.) Take an even $\binom{[m]}{l}$-matrix $A$ such that $\mathrm{rk} A=\widetilde{r_{m}}$. By the non-triviality $A \neq 0$. Hence there are $l$-element subsets $X, Y \subset[m]$ such that $A_{X, Y}=1$. Let $C$ be the 'restriction' of $A$ to $l$-element subsets of $[m]-X-Y$.
Then
$$
\widetilde{r_{m}}=\operatorname{rk} A \geq \operatorname{rk} C+2 \geq \widetilde{r_{m-2 l+1}}+2, \quad \text { where }
$$
- the first inequality follows by Assertion 5.5.b;
- the second inequality holds because $C$ is a $\left({ }^{[m]-X-Y}\right)$-matrix by Assertion 5.4, and because $A_{X, Y}=1$, so by the triviality $X \cap Y \neq \emptyset$, hence $|[m]-X-Y| \geq m-2 l+1$.
6.7. (a) $\lambda_{X, P}=X \cdot{ }_{A} P$.
6.8. (a) $\lambda_{X, Y, P}=X \cdot{ }_{A} P, \lambda_{Y, X, P}=Y \cdot{ }_{A} P$.
7.2. (For 5.4 take $l=3$.) It is obvious that all the conditions for the mentioned submatrix are satisfied.
7.3. (a) (For 5.5.a take $l=3$.) Let $B^{\prime}$ be the 'restriction' of $A$ to $X$ and to $l$-element subsets of $[m]-X$. Then
$$
\operatorname{rk} A \geq \operatorname{rk} B^{\prime}=\operatorname{rk} B+1
$$
where equality holds because by the triviality $B_{X, Z}^{\prime}=0$ for any $Z \subset[m]-X$.
(b) (For 5.5.b take $l=3$.) Let $C^{\prime}$ be the 'restriction' of $A$ to $X, Y$ and $l$-element subsets of $[m]-X-Y$. Then
$$
\operatorname{rk} A \geq \operatorname{rk} C^{\prime}=\operatorname{rk} C+2
$$
where equality holds because by the triviality $C_{X, Z}^{\prime}=C_{Y, Z}^{\prime}=0$ for any $Z \subset[m]-X-Y$.
(b') Take a basis of $\mathbb{Z}_{2}^{\binom{m}{l}}$ corresponding to $l$-element subsets of $[m]$. Define a bilinear form $A$ on $\mathbb{Z}_{2}^{\binom{m}{l}}$ by setting $A(P, Q):=A_{P, Q}$ for basic vectors $P, Q$. Take any $l$-element set $P \subset[m]$. Let
$$
\bar{P}=\bar{P}(X, Y):=P+A_{X, P} Y+A_{Y, P} X
$$
Recall that
$$
A_{X, Y}=A_{Y, X}=1 \quad \text { and } \quad A_{X, X}=A_{Y, Y}=0
$$
Hence
$$
A(\bar{P}, X)=A(\bar{P}, Y)=0
$$
(i. e., $\bar{P}$ is the orthogonal projection of $P$ to the orthogonal complement of $\langle X, Y\rangle$ with respect to $A$ ). By the triviality, for $P \subset[m]-X$ we have $\bar{P}=P+A_{Y, P} X$. Hence for every $l$-element sets $P, Q \subset[m]-X$ we have
$$
A(\bar{P}, \bar{Q})=A_{P, Q}+0+0+0=B_{P, Q}
$$
(I. e., $B$ is the Gramian matrix with respect to $A$ of the 'projections' $\bar{P}$ of $l$-element sets $P \subset[m]-X$.) Let $B^{\prime}$ be the Gramian matrix with respect to $A$ of $X, Y$ and the 'projections' $\bar{R}$ of $l$-element sets $R \subset[m]-X$. I. e., $B_{P . Q}^{\prime}=A(\widehat{P}, \widehat{Q})$, where $\widehat{P}=P$ if $P \in\{X, Y\}$, and $\widehat{P}=\bar{P}$ otherwise ( $\widehat{Q}$ is defined analogously). Then
- $B_{X, Y}^{\prime}=B_{Y, X}^{\prime}=1, B_{X, X}^{\prime}=B_{Y, Y}^{\prime}=0$ (by (*)),
- $B_{X, P}^{\prime}=B_{P, X}^{\prime}=B_{Y, P}^{\prime}=B_{P, Y}^{\prime}=0$ for $P \neq X, Y$ (by $(* *)$ ), and
- $B_{P, Q}^{\prime}=B_{P, Q}$ for $P, Q \subset[m]-X$ (by (***)).
Hence $\operatorname{rk} B+2=\operatorname{rk} B^{\prime} \leq \operatorname{rk} A$.
(a') In this paragraph we prove that $\operatorname{rk} D<\operatorname{rk} A$. Take a basis of $\mathbb{Z}_{2}^{\binom{m}{l}}$ corresponding to $l$-element subsets of $[m]$. Define a bilinear form $A$ on $\mathbb{Z}_{2}^{\binom{m}{l}}$ by setting $A(P, Q):=A_{P, Q}$ for basic vectors $P, Q$. Let $P_{X}$ be the orthogonal projection of $P$ to the orthogonal complement of $X$ (with respect to $A$ ), i. e., $P_{X}:=P+A_{P, X} X$. We have
$$
\begin{aligned}
& A\left(P_{X}, Q_{X}\right)=A(P, Q)+A\left(A_{P, X} X, Q\right)+A\left(P, A_{Q, X} X\right)+A\left(A_{P, X} X, A_{Q, X} X\right)= \\
& \quad=A_{P, Q}+A_{P, X} A_{X, Q}+A_{P, X} A_{Q, X}+A_{P, X} A_{Q, X} A_{X, X}=A_{P, Q}+A_{P, X} A_{Q, X}=D_{P, Q}
\end{aligned}
$$
Then $D$ is the Gramian matrix (with respect to $A$ ) of the projections of subsets of $[m-l+1]$. Let $D^{\prime}$ be the Gramian matrix (with respect to $A$ ) of $X$ and the projections of subsets of $[m-l+1]$. We have $D_{P, Q}=D_{P, Q}^{\prime}$ for all subsets $P, Q \subset[m-l+1]$. Furthermore, $D_{X, P}^{\prime}=$ $D_{P, X}^{\prime}=0$ for any basic vector $P \neq X$ and $D_{X, X}^{\prime}=A_{X, X}=1$. Thus $\operatorname{rk} D=\operatorname{rk} D^{\prime}-1<\operatorname{rk} A$.
In this paragraph we prove that $D$ satisfies the triviality property. If $P \cap Q=\emptyset$, then either $P \cap X=\emptyset$, or $Q \cap X=\emptyset$. Hence $D_{P, Q}=A_{P, Q}+A_{P, X} A_{Q, X}=0+0=0$.
In this paragraph we prove that $D$ satisfies the linear dependence property. For each $(l+1)$-element and $l$-element subsets $F, P \subset[m-l+1]$ we have
$$
\sum_{i \in F} D_{F-i, P}=\sum_{i \in F} A_{F-i, P}+A_{P, X} \sum_{i \in F} A_{F-i, X}=0
$$
In this paragraph we prove that $D$ satisfies the non-triviality property. By Proposition 7.5.b for $D$, we may assume that $i \neq m-l+1$. Then for each summand $D_{i \sqcup \sigma, i \sqcup \tau}$ of $D_{F, i}$ at least one of the sets $i \sqcup \sigma, i \sqcup \tau$ does not contain $m-l+1$ and hence does not intersect $X$. Hence $D_{i \sqcup \sigma, i \sqcup \tau}=A_{i \sqcup \sigma, i \sqcup \tau}+A_{i \sqcup \sigma, X} A_{i \sqcup \tau, X}=A_{i \sqcup \sigma, i \sqcup \tau}$. Thus $D_{F, i}=A_{F, i}=1$.
7.4. (a) Take an $\binom{[m]}{l}$-matrix $A$ such that $\operatorname{rk} A=r_{m}$. If $A$ is even, then $r_{m}=\widetilde{r_{m}}$, so we are done. Otherwise there is an $l$-element subset $X \subset[m]$ such that $A_{X, X}=1$. Without loss of generality $X=\{m-l+1, m-l+2, \ldots, m\}$. Then by Assertion 7.3.a'
$$
r_{m}=\operatorname{rk} A \geq \operatorname{rk} D+1 \geq r_{m-l+1}+1, \quad \text { where }
$$
- $D$ is the matrix defined in Assertion 7.3.a';
- the first inequality follows from Assertion 7.3.a';
- the second inequality holds because $D$ is an $\binom{[m-l+1]}{l}$-matrix by Assertion 7.3.a'.
(b) Take an even $\binom{[m]}{l}$-matrix $A$ such that $\mathrm{rk} A=\widetilde{r_{m}}$. By the non-triviality $A \neq 0$. Let $X, Y, B$ be defined as in Assertion 7.3.b'. Then
$$
\widetilde{r_{m}}=\operatorname{rk} A \geq \operatorname{rk} B+2 \geq r_{m-l}+2, \quad \text { where }
$$
- the first inequality follows from Assertion 7.3.b';
- the second inequality holds because $B$ is an even $\left({ }_{[}^{[m]-X}\right)$-matrix by Assertion 7.2.
7.5. (a) It suffices to check that for each pair $\{\alpha, \beta\}$ the number of sets $T_{i,\{\sigma, \tau\}}$ containing $\{\alpha, \beta\}$ is odd if and only if $\alpha \cap \beta=\emptyset$ (hence this parity not depend on $i$ ). Clearly, $|\alpha \cap \beta| \leq 2$.
Assume $|\alpha \cap \beta|=2$. Then $\{\alpha, \beta\} \notin T_{i,\{\sigma, \tau\}}$ for all $i, \sigma, \tau$ and hence $\{\alpha, \beta\} \notin A_{i}$ for all $i \in[5]$.
Assume $|\alpha \cap \beta|=1$. It suffices to consider the case $\alpha=\{1,2\}, \beta=\{1,3\}$. Then $\{\alpha, \beta\} \in T_{i,\{\sigma, \tau\}}$ iff $i=1$ and $\{\sigma, \tau\}$ is either $\{\{2,4\},\{3,5\}\}$ or $\{\{2,5\},\{3,4\}\}$ Therefore $\{\alpha, \beta\} \notin A_{i}$ for all $i \in[5]$.
Assume $|\alpha \cap \beta|=0$. It suffices to consider the case $\alpha=\{1,2\}, \beta=\{3,4\}$. Then $\{\alpha, \beta\} \in T_{i,\{\sigma, \tau\}}$ iff either
- $i=1$ and $\{\sigma, \tau\}=\{\{1,2,5\},\{1,3,4\}\}$, or
- $i=2$ and $\{\sigma, \tau\}=\{\{1,2,5\},\{2,3,4\}\}$, or
- $i=3$ and $\{\sigma, \tau\}=\{\{1,2,3\},\{3,4,5\}\}$, or
$\bullet i=4$ and $\{\sigma, \tau\}=\{\{1,2,4\},\{3,4,5\}\}$, or
$\bullet i=5$ and $\{\sigma, \tau\}=\{\{1,2,5\},\{3,4,5\}\}$.
Therefore $\{\alpha, \beta\} \in A_{i}$ for every $i \in[5]$.
(b) It suffices to prove that $A_{G \sqcup i, j}=A_{G \sqcup j, i}$ for each $i, j \in[m]$ and $(2 l-3)$-element subset $G \subset[m]-i-j$. Denote $\bar{\sigma}:=\{i, j\} \sqcup \sigma$. Then
$$
\begin{aligned}
A_{G \sqcup j, i}+A_{G \sqcup i, j} & \stackrel{(1)}{=} \sum_{\{(\sigma, \tau): G=\sigma \sqcup \tau,|\sigma|=l-2\}}\left(A_{\bar{\sigma}, i \sqcup \tau}+A_{\bar{\sigma}, j \sqcup \tau}\right) \stackrel{(2)}{=} \\
& =\sum_{\{(\sigma, \tau): G=\sigma \sqcup \tau,|\sigma|=l-2\}} \sum_{t \in \tau} A_{\bar{\sigma}, \overline{\tau-t}} \stackrel{(3)}{=} \sum_{t \in G} \sum_{\{(\sigma, \nu): G-t=\sigma \sqcup \nu,|\sigma|=l-2\}} A_{\bar{\sigma}, \bar{\nu}} \stackrel{(4)}{=} 0, \quad \text { where }
\end{aligned}
$$
- equality (1) holds because $A_{G \sqcup j, i}$ is equal to the sum of the first summands $A_{\bar{\sigma}, i \sqcup \tau}$, and $A_{G \sqcup i, j}$ is equal to the sum of the second summands $A_{\bar{\sigma}, j \sqcup \tau}$;
- equality (2) holds by the linear dependence for $F=\bar{\tau}, P=\bar{\sigma}$;
- equality (3) is obtained by changes of the order of summation and of variable $\nu=\tau-t$;
- equality (4) holds because ordered decompositions $(\sigma, \nu)$ of $G-t$ into $(l-2)$-element subsets $\sigma, \nu$ split into pairs $\{(\sigma, \nu),(\nu, \sigma)\}$ and $A_{\bar{\sigma}, \bar{\nu}}+A_{\bar{\nu}, \bar{\sigma}}=0$.
## References
[ACM29] D. Hachenberger, D. Jungnickel. Topics in Galois Fields (2020).
[Bi20] * A. Bikeev. Realizability of discs with ribbons on the Möbius strip. Mat. Prosveschenie, 28 (2021); erratum to appear. arXiv:2010.15833.
[Bi21] A. I. Bikeev, Criteria for integer and modulo 2 embeddability of graphs to surfaces, arXiv:2012.12070v2.
[DS22] S. Dzhenzher and A. Skopenkov, To the Kühnel conjecture on embeddability of $k$-complexes into $2 k$-manifolds, arXiv:2208.04188.
[FK19] R. Fulek, J. Kynčl, $\mathbb{Z}_{2}$-genus of graphs and minimum rank of partial symmetric matrices, 35th Intern. Symp. on Comp. Geom. (SoCG 2019), Article No. 39; pp. 39:1-39:16, https://drops.dagstuhl.de/opus/volltexte/2019/ 10443/pdf/LIPIcs-SoCG-2019-39.pdf. We refer to numbering in arXiv version: arXiv:1903.08637.
[Ko21] E. Kogan. On the rank of $\mathbb{Z}_{2}$-matrices with free entries on the diagonal, arXiv:2104.10668.
[KS21] * E. Kogan and A. Skopenkov. A short exposition of the Patak-Tancer theorem on non-embeddability of $k$-complexes in $2 k$-manifolds, arXiv:2106.14010.
[KS21e] E. Kogan and A. Skopenkov. Embeddings of $k$-complexes in $2 k$-manifolds and minimum rank of partial symmetric matrices, arXiv:2112.06636.
[MC] * https://en.wikipedia.org/wiki/Matrix_completion\#Low_rank_matrix_ completion
[NKS] * L. T. Nguyen, J. Kim, B. Shim. Low-rank matrix completion: a contemporary survey. arXiv:1907.11705.
[PT19] P. Paták and M. Tancer. Embeddings of $k$-complexes into $2 k$-manifolds. arXiv:1904.02404.
[Sk14] * A. Skopenkov, Realizability of hypergraphs and Ramsey link theory, arXiv:1402.0658.
[Sk20] * A. Skopenkov, Algebraic Topology From Geometric Viewpoint (in Russian), MCCME, Moscow, 2020 (2nd edition). Part of the book: http://www.mccme. ru/circles/oim/obstruct.pdf . Part of the English translation: https://www. mccme.ru/circles/oim/obstructeng.pdf.
Books, surveys and expository papers in this list are marked by the stars.
# Движение точек
Давид Бродский
Проект представляют: Давид Бродский, Алексей Заславский
Олег Заславский, Иван Фролов и Фёдор Нилов
Движение - это жизнь, а жизнь - это движение!
## 1 Проективное движение
Аристотель
Все утверждения, сформулированные в этом параграфе в качестве лемм и теорем можно использовать без доказательства. В основном это стандартные утверждения проективной геометрии желающие могут самостоятельно осознать их истинность.
Определение 1.1. Проективная плоскость. Проективной плоскостью $\mathbb{R}^{2}$ мы будем называть обычную плоскость, дополненную бесконечно удаленными точками, в каждой из которых пересекается свой класс параллельных прямых. Таким образом, на каждой прямой появляется новая бесконечно удаленная точка, общая для всех параллельных между собой прямых. Все бесконечно удаленные точки образуют бесконечно удаленную прямую. Прямая, дополненная бесконечно удаленной точкой, называется проективной.
Определение 1.2. Двойное отношение четверки точек на проективной прямой. Пусть точки $A, B, C, D$ лежат на проективной прямой $\ell$. Двойным отношением точек $(A, B ; C, D)$ на прямой $\ell$ будем называть величину
$$
\frac{\overrightarrow{C A}}{\overrightarrow{C B}}: \frac{\overrightarrow{D A}}{\overrightarrow{D B}}
$$
Будем считать, что бесконечно удаленная точка делит любой отрезок в отношении $1: 1$.
Определение 1.3. Пучок прямых. Пучком прямых $\mathcal{L}_{A}$ точки $A$ будем называть множество всех прямих, проходящих через точку $A$.
Отметим, что множество всех прямых одного направления образуют пучок бесконечно удаленной точки.
Определение 1.4. Двойное отношение четверки прямых одного пучка. Пусть прямые $a, b, c, d$ проходят через точку $O$ (множество прямих, проходящ,их через точку $O$, называется пучком). Выберем произвольным образом на этих прямых направляющие векторы $\vec{v}_{a}, \vec{v}_{b}, \vec{v}_{c}, \vec{v}_{d}$. Двойным отношением прямых $(a, b ; c, d)$ будем называть величину
$$
\frac{\sin \angle\left(\vec{v}_{c}, \vec{v}_{a}\right)}{\sin \angle\left(\vec{v}_{c}, \vec{v}_{b}\right)}: \frac{\sin \angle\left(\vec{v}_{d}, \vec{v}_{a}\right)}{\sin \angle\left(\vec{v}_{d}, \vec{v}_{b}\right)}
$$
Замечание: Вообще говоря, знак $\sin \angle\left(\vec{v}_{x}, \vec{v}_{y}\right)$ не определяется прямыми $x$ и $у$ однозначно, так как направляющий вектор можно выбрать двумя способами. Однако, если поменять направление одного из векторов, то знак инвертируется сразу у двух синусов, и значение двойного отношения не изменится.
Определение 1.5. Двойное отношение точек на окружности. Пусть точки $A, B, C, D$ лежат на окружности $\Omega$. Отметим еще одну точку $O$ на окружности. Двойным отношением на окружности $\Omega$ точек $(A, B ; C, D)$ будем называть двойное отношение прямых $(O A, O B ; O C, O D)$. Корректность определения следует из того, что величина вписанного угла постоянна.
Лемма 1.1. Пусть точки $A, B, C, D$ лежат на прямой $\ell$, точка $O$ вне этой прямой. Тогда
$$
(A, B ; C, D)=(O A, O B ; O C, O D)
$$
1. Убедитесь, что если завести на плоскости комплексные координаты, то двойное отношение четырех точек на прямой или окружности с координатами $a, b, c, d$ считается как $\frac{c-a}{c-b}: \frac{d-a}{d-b}$.
В дальнейшем мы определим двойное отношение четырех точек на произвольной конике. Для них результат последней задачи НЕ ВЕРЕН.
Определение 1.6. Гомография. Будем называть отображение $\mathcal{F}$ проективным (или гомографией), если
$$
(\mathcal{F}(A), \mathcal{F}(B) ; \mathcal{F}(C), \mathcal{F}(D))=(A, B ; C, D)
$$
Отображение может действовать между прямыми, пучками и окружностями (кониками).
Лемма 1.2. Пусть $A, B, C \in \ell, k \in \mathbb{R}$, тогда существует единственная точка $D \in \ell$ такая, что $(A, B, C, D)=k$.
Теорема 1.1. Любую гомографию прямых можно разложить в композицию нескольких центральных проекций и параллельных переносов.
Определение 1.7. Проективное движение. Зафиксируем некоторую проективную прямую $\mathcal{T}$. Будем говорить, что переменная точка $X$ движется по проективной прямой или окружности $\ell$ проективно, если существует проективное отображение $\mathcal{F}: \mathcal{T} \rightarrow \ell$ такое, что $X=\mathcal{F}(t \in \mathcal{T})$. Прямая $\mathcal{T}$ играет роль параметра времени.
Давайте поясним простыми словами, что в этом определении имеется в виду. Что означает, что точка $X$ движется с течением времени? Формально это значит, что для каждого момента $t$ существует положение точки $X$ на плоскости, соответствующее этому моменту. Такое соответствие может быть выражено функцией $\mathcal{F}(t)$, которая по сути действует из координатной прямой в плоскость. Однако нам иногда будет удобно подставлять «бесконечное» время, и, для того чтобы формализовать эту идею, вместо обычной прямой времени мы будем брать проективную. Так вот проективное движение с течением времени это любое движение точки по прямой или окружности, при котором выполняется странное условие на сохранение двойных отношений. То есть, если взять любые четыре момента времени (как точки на проективной прямой), то их двойное отношение должно быть равно двойному отношению четырех положений точки $X$, соответствующих этим моментам времени.
Аналогично определим прямую, проективно вращающуюся в пучке:
Определение 1.8. Проективно вращающаяся прямая. Будем говорить, что переменная прямал $\ell$ вращается в пучке $\mathcal{L}$ проективно, если существует проективное отображение $\mathcal{F}: \mathcal{T} \rightarrow \mathcal{L}$ такое, что $\ell=\mathcal{F}(t \in \mathcal{T})$.
Договоримся о некоторых обозначениях специальных гомографий, которые в дальнейшем будем использовать.
- $\ell \xrightarrow{S} \mathcal{L}_{S}$ проекция с центром в точке $S$, отображающая прямую $\ell$ в пучок прямых $\mathcal{L}_{S}$.
- $\mathcal{R}_{S}^{\psi}$ - поворот с центром в точке $S$ против часовой стрелки на угол $\psi$.
- $\mathcal{H}_{S}^{k}$ - гомотетия с центром в точке $S$ и коэффициентом $k$.
Покажем, как проективное движение помогает решать задачи.
Пример 1.1. Пусть $A_{1}$ - точка касания вписанной окружности треугольника $A B C$ со стороной $B C, D$ - произвольная точка на $B C$. Обозначим за $I_{B}$ и $I_{C}$ центры вписанных окружностей $\triangle A B D, \triangle A C D$ соответственно. Докажите, что $\angle I_{B} A_{1} I_{C}=90^{\circ}$
Доказательство. Разобьем решение на несколько типовых шагов.
1. Зафиксируем $\triangle A B C$ и будем проективно двигать точку $I_{B}$ по биссектрисе $\angle A B C$.
2. Построим гомографию $\mathcal{F}: B I \rightarrow B I$, тождественность которой равносильна задаче.
Оформим «цепочку» гомографий в виде таблички, за $\ell_{b}, \ell_{c}$ обозначены биссектрисы $\angle B$ и $\angle C$ соответственно:

В верхней строке указаны прямые и пучки, переводящиеся друг в друга гомографиями, в нижней - объекты, которые по ним движутся. Очевидно, что задача равносильна проверке того, что $I_{B}=I_{B}^{\prime}$.
3. Проверим, что задача верна в каких-нибудь трех положениях точки $I_{B}$. Положения обычно удобно выбирать вырожденные. В этой задаче мы рассмотрим $I_{B}=B, I_{B}=I, I_{B}=P$, где $P$ центр вписанной окружности треугольника $A B A_{1}$. В каждом из них задача очевидна.
Поскольку гомография однозначно восстанавливается по образам трех точек, построенное отображение из прямой $\ell_{1}$ в себя тождественно, поэтому задача верна и при любом другом положении точки $I_{B}$.
Давайте потренируемся решать задачи, используя проективное движение. Во многих задачах полезно помнить, что направления суть точки бесконечно удаленной прямой.
2. Прямая $\ell$ проективно вращается вокруг фиксированной точки $P$. Точка $S \neq P$ - фиксирована. Докажите, что основание перпендикуляра из точки $S$ на $\ell$ движется проективно. Какова будет траектория движения этой точки?
3. Внешние биссектрисы $B B_{1}$ и $C C_{1}$ треугольника $A B C$ пересекаются в точке $I_{A}$. Прямая $\ell$, проходящая через $I_{A}$, пересекает прямые $A B$ и $A C$ в точках $X$ и $Y$ соответственно. Докажите, что прямые, симметричные прямым $B Y$ и $C X$ относительно $B B_{1}$ и $C C_{1}$, пересекаются на $B_{1} C_{1}$.
4. Дан треугольник $A B C$ и точки $B_{1}, C_{1}$ на сторонах $A C, A B$ такие, что прямые $B B_{1}, C C_{1}$ пересекаются на высоте треугольника $A A_{1}$. Докажите, что прямые $A_{1} B_{1}$ и $A_{1} C_{1}$ симметричны относительно $A A_{1}$.
5. На чевианах $A A_{1}, B B_{1}, C C_{1}$ (то есть, пересекающихся в одной точке) треугольника $A B C$ выбраны точки $A_{2}, B_{2}$ и $C_{2}$ соответственно. Положим $A_{3}=B C_{2} \cap B_{2} C$. Точки $B_{3}$ и $C_{3}$ определяются аналогично. Докажите, что прямые $A A_{3}, B B_{3}$ и $C C_{3}$ пересекаются в одной точке.
6. Дан ромб $A B C D$ с острым углом $B$. Точка $O$ - центр описанной окружности треугольника $A B C$. На продолжении луча $O C$ за точку $C$ выбрана точка $P$. Прямая $P D$ пересекается с прямой, проходящей через точку $O$ параллельно стороне $A B$, в точке $Q$. Докажите, что $\angle A Q O=\angle P B C$.
7. Остроугольный треугольник $A B C$ вписан в окружность $\Omega$ и описан около окружности $\omega$. Точка $P$ выбирается на отрезке, соединяющим центры $\Omega$ и $\omega$. Обозначим за $A^{\prime}, B^{\prime}$ и $C^{\prime}$ вторые пересечения прямых $A P, B P$ и $C P$ с $\Omega$. Докажите, что внутренние биссектрисы $\angle B A^{\prime} C, \angle C B^{\prime} A$ и $\angle A C^{\prime} B$ пересекаются на линии центров $\Omega$ и $\omega$.
8. Четырехугольник $A B C D$ описан около окружности с центром $I$, на прямых $A I$ и $C I$ выбраны точки $P$ и $Q$ соответственно так, что угол $\angle P B Q=\frac{1}{2} \angle A B C$. Докажите, что угол $\angle P D Q=\frac{1}{2} \angle A D C$.
9. Обозначим за $S$ проекцию ортоцентра треугольника $A B C$ на его медиану $A M$. Окружность $\omega$ проходит через точки $A$ и $S$ и пересекает отрезки $A B$ и $A C$ в точках $Q$ и $P$ соответственно. Докажите, что отрезки $B P$ и $C Q$ пересекаются на $\omega$.
10. (a) Пусть $f$ - гомография проективной прямой $\ell$ в себя. Параметризуем конечные точки прямой переменной $x$. Докажите, что отображение $f$ дробно-линейно, то есть $f(x)=\frac{a x+b}{c x+d}$, где числа $a, b, c$ и $d$ фиксированы. (b) Пусть $f$ - гомография прямых $\ell_{1}$ и $\ell_{2}$. Заведем на конечной части плоскости декартовы координаты. Докажите, что $f(x, y)=\left(\frac{a_{1} x+b_{1} y+c_{1}}{d x+e y+f}, \frac{a_{2} x+b_{2} y+c_{2}}{d x+e y+f}\right)$, где числа $a_{1}, a_{2}, b_{1}, b_{2}, c_{1}, c_{2}, d, e, f$ - фиксированы. (c) Докажите, что любое отображение плоскости $\mathbb{R}^{2}$ такого вида, который описан в пункте (b) может быть однозначно продолжено до проективного преобразования проективной плоскости.
11. Отображение $f$ из комплексной прямой $\mathbb{C}$ в себя задается в декартовых координатах как $f(z)=\frac{P(z)}{Q(z)}$, где $P$ и $Q$ - многочлены. Пусть дополнительно известно, что $f$ биективно. Докажите, что $f$ сохраняет двойные отношения любой четверки комплексных чисел.
## 2 Проективное движение +
Удивительным образом оказывается, что инверсия, суженная на окружность или прямую, также сохраняет двойные отношения!
Теорема 2.1. Пусть при инверсии $\mathcal{I}$ окружность или прямая $\Omega$ переходит в окружность или прямую $\widetilde{\Omega}$. Тогда для любых четырех точек $A, B, C, D \in \Omega$ верно $(A, B ; C, D)=(\mathcal{I}(A), \mathcal{I}(B) ; \mathcal{I}(C), \mathcal{I}(D))$.
Доказательство. Ранее было доказано, что двойное отношение четвёрки точек на прямой или окружности равно двойному отношению комплексных координат этих точек. Воспользуемся этим.
Введём комплексные координаты так, что окружность, относительно которой строится инверсия $\mathcal{I}$, задаётся уравнением $z \bar{z}=1$. Образ $\mathcal{I}(z)$ произвольной точки $z$ плоскости будет $\mathcal{I}(z)=\frac{1}{\bar{z}}$.
Обозначим за $a, b, c$ и $d$ комплексные координаты точек и напишем:

$$
\begin{aligned}
& =\frac{\bar{c}-\bar{a}}{\bar{c}-\bar{b}}: \frac{\bar{d}-\bar{a}}{\bar{d}-\bar{b}}=\frac{\overline{c-a}}{c-b}: \frac{d-a}{d-b}=\frac{c-a}{c-b}: \frac{d-a}{d-b}=(a, b ; c, d)
\end{aligned}
$$
Теорема позволяет сделать следующие крайне полезное наблюдение:
12. Проекция окружности на себя. а) Пусть $\Omega$ - окружность, $S$ - произвольная точка плоскости, не лежащая на окружности. Каждой точке $X \in \Omega$ сопоставим вторую точку $\mathcal{F}(X)$ пересечения прямой $S X$ с $\Omega$. Тогда отображение $\mathcal{F}: \Omega \rightarrow \Omega$ проективно. b) Докажите, что отображение, сопоставляющие точке $X$ прямую $S X \in \mathcal{L} \mathrm{HE}$ проективно.
13. Перебрасывание окружности на касательную. Пусть точка $X$ движется по прямой $\ell$, которая касается окружности $\Omega$. Обозначим основание второй касательной из $X$ к $\Omega$ за $Y=\mathcal{F}(X)$. Тогда отображение $\mathcal{F}: \ell \rightarrow \Omega$ проективно.
14. Перебрасывание окружности на себя через прямую. а) Точка $X$ лежит на окружности $\Omega, \ell$ - фиксированная прямая, не касающаяся $\Omega$. Пусть касательная к $\Omega$, восстановленная в $X$, пересекает $\ell$ в точке $Z, Y=\mathcal{F}(X)$ - основание второй касательной из $Z$ к $\Omega$. Тогда отображение $\mathcal{F}: \Omega \rightarrow \Omega$ проективно. b) Докажите, что отображение, сопоставляющие точке $X$ точку $Z \in \ell \mathrm{HE}$ проективно.
Доказанные выше леммы часто помогают решать задачи проективным движением.
15. Пусть $\gamma_{A}, \gamma_{B}, \gamma_{C}$ - вневписанные окружности треугольника $A B C$, касающиеся сторон $B C, C A, A B$ соответственно. Обозначим через $\ell_{A}$ общую внешнюю касательную окружностей $\gamma_{B}$ и $\gamma_{C}$, отличную от $B C$. Аналогично определим прямые $\ell_{B}, \ell_{C}$. Из точки $P$, лежащей на $\ell_{A}$, проведем отличную от $\ell_{A}$ касательную к $\gamma_{B}$ и найдем точку $X$ ее пересечения с $\ell_{C}$. Аналогично найдем точку $Y$ пересечения касательной из $P$ к $\gamma_{C}$ с $\ell_{B}$. Докажите, что прямая $X Y$ касается $\gamma_{A}$.
16. Остроугольный треугольник $A B C$ с ортоцентром $H$ вписан в окружность $\omega$ с центром $O$. Прямая $l$ проходит через $H$ и пересекает меньшие дуги $A B$ и $A C$ в точках $P$ и $Q$ соответственно. Пусть $A A^{\prime}$ - диаметр окружности $\omega$. Прямые $A^{\prime} P$ и $A^{\prime} Q$ пересекают $B C$ в точках $K$ и $L$ соответственно. Докажите, что точки $O, K, L$ и $A^{\prime}$ лежат на одной окружности.
17. Пусть $S$ - проекция центра $I$ окружности на диагональ $A C$ описанного около нее четырехугольника $A B C D$. Докажите, что $\angle B S A=\angle D S A$.
18. Пятиугольник $A B C D E$ описан около окружности $\omega$. Пары лучей $E A$ и $C B, A E$ и $C D$, $A B$ и $D C, B C$ и $E D$ пересекаются в точах $P, Q, X, Y$ соответственно. Окружность $\omega$ касается $A E$ в точке $R$. Оказалось, что $X Y \| A E$. Пусть окружности $(A X Q),(P Y E)$ пересекаются в точках $S, T$. Докажите, что точки $S, T, R$ лежат на одной прямой.
19. Прямая $\ell$ проходит через центр $O$ описанной около треугольника $A B C$ окружности, пересекая стороны $A B, A C$ треугольника в точках $P, Q$. Докажите, что одна из точек пересечения окружностей, построенных на отрезках $B Q, C P$ как на диаметрах, лежит на окружности девяти точек треугольника $A B C$, а вторая - на описанной около него окружности.
20. 2023 прямые пересекаются в одной точке. В каждый из 4026-и углов вписано по окружности; окружности касаются друг друга по циклу. На сторонах углов отмечено по точке. Известно, что для всех углов, кроме одного, отрезок, соединяющий отмеченные точки на сторонах, касается вписанной в угол окружности. Докажите, что для оставшегося угла это также верно.
21. Зафиксируем стороны $A B$ и $A C$ треугольника $A B C$, а также вписанную в него окружность $\omega$. Докажите, что точка касания $\omega$ с отрезком $B C$ проективно зависит от точки Фейербаха этого треугольника.
Теорема 2.2. Частный случай теоремы Понселе, можно использовать без доказательства. Треугольник $A B C$ вписан в окружность $\Omega$ и описан около окружности $\omega$ (возможно, внеиним образом). Пусть точка $A^{\prime}$ также лежит на $\Omega$, касательные из нее к $\omega$ вторично пересекают $\Omega$ в точках $B^{\prime}$ и $C^{\prime}$. Тогда прямая $B^{\prime} C^{\prime}$ касается $\omega$.
Давайте заметим, что в данной конфигурации имеется пара биективно зависящих друг от друга точек: каждой точке $A$ описанной окружности соответствует точка касания стороны $B C$ со вписанной окружностью. Возникает подозрение, что подобное отображение с описанной окружности на вписанную также проективно.
Лемма 2.1. Гомографии конфигурации Понселе. Пусть треугольник АВС вращается по Понселе с сохранением вписанной окружности. Тогда точка $A_{1}$ - касание отрезка BC со вписанной окружностъю, точки $S_{A}$ и $T_{A}$ - середины дуг, стягивающих хорду $B C$, точка $I_{A}$ - центр вневписанной окружности, касающейся отрезка $B C$; все перечисленные точки проективно зависят от точки $A$.
Отображение, переводящую точку $A$ в точку $A_{1}$, будем называть гомографией Понселе и обозначать ее $\mathcal{P}$.
Доказательство. Обозначим через $I$ и $O$ центры вписанной и описанной окружностей треугольника $A B C$ соответственно. Заметим, что
- Середина $S_{A}$ дуги $B C$ есть образ точки $A$ при проекции описанной окружности на себя с центром в точке $I$.
- Точка $T_{A}$ диаметрально противоположна точке $S_{A}$.
- Точка $A_{1}$ - образ точки $S_{A}$ при положительной гомотетии, переводящей описанную окружность во вписанную.
- В силу леммы о трезубце точка $I_{A}$ - образ середины $S_{A}$ дуги $B C$ при гомотетии с центром в точке $I$ и коэффициентом 2.
Явно предъявлены гомографии, по точке $A$ строящие точки $S_{A}, T_{A}, A_{1}, I_{A}$, а значит они действительно проективно зависят от $A$.
Обычно гомографии Понселе очень хорошо работают вместе на пару с методом полиномиального движения точек, который обобщает проективное движение. Но некоторые трудные задачи удается решить лишь уже известным нам аппаратом.
22. Выпуклый шестиугольник $A Q C P B R$ вписан в окружность $\Omega$, и при этом треугольники $A B C$ и $P Q R$ описаны около одной и той же окружности $\gamma$. Прямая $\ell$, параллельная прямой $B C$ и не совпадающая с ней, касается $\gamma$. Пусть $P_{1}$ - точка пересечения прямой $\ell$ и отрезка $Q R$. Докажите, что $\angle P A B=\angle P_{1} A C$.
23. Треугольники $A B C$ и $D E F$ описаны около окружности $\omega$ и вписаны в окружность $\gamma$. Обозначим за $K$ и $L$ точки касания отрезков $B C$ и $E F$ с окружностью $\omega$. Положим $N$ и $M$ - вторые пересечения $A L$ и $D K$ с $\gamma$ соответственно. Докажите, что прямые $A M, E F, B C, N D$ пересекаются в одной точке.
24. Пусть оказалось, что в конфигурации Понселе вершины $A$ и $B$ вращающегося треугольника зависят друг от друга проективно. Докажите, что вращающийся треугольник - правильный.
Возможно, последняя задача пока покажется вам сложной. Предлагаем вернутся к ней после того, как Вы разберетесь с принципом относительности.
Хорошо известно, что теорема Понселе верна не только для треугольника, но и для любого $\mathrm{n}$-угольника. В связи с этим возникает такой естественный вопрос:
25. Пусть многоугольник $A_{1} A_{2} \ldots A_{2 k+1}$ вращается по Понселе. Верно ли, что точка $A_{k+1}$ проективно зависит от точки касания звена $A_{1} A_{2 k+1}$ со вписанной окружностью?
Нам не известно элементарное доказательство этого утверждения.
## 3 Коники
Определение 3.1. Коника. Невырожденной коникой мы будем называть образ окружности после проективного преобразования. Вырожденной коникой называется пара прямых или прямая.
Можно без доказательства пользоваться тем, что любая невырожденная коника это окружность, эллипс, парабола или гипербола; а также тем, что любая коника является множеством точек $(x, y)$, удовлетворяющих уравнению $a x^{2}+b y^{2}+c x y+d x+e y+f=0$, где $a, b, c, d, e, f-$ фиксированные вещественные числа.
Без доказательства разрешается использовать следующее утверждение:
Теорема 3.1. Через любые пять точек, никакие три из которых не лежат на одной прямой, можно провести невырожденную конику, причем единственным образом.
Определение 3.2. Двойное отношение четырех точек на конике. Пусть $A, B, C, D-$ точки на конике $\Omega$, а $S$ - еще одна точка на ней же. Будем называть двойным отношением четверки точек $(A, B ; C, D)$ на конике двойное отношение четверки прямых $(S A, S B ; S C, S D)$.
26. Докажите, что это определение двойного отношения корректно.
27. На конике $\mathcal{C}$ нашлись три точки $A, B$ и $C$, лежащие на одной прямой. Докажите, что $\mathcal{C}-$ вырожденная.
28. а) Прямые $a_{t}$ и $b_{t}$ проективно вращаются вокруг точек $A$ и $B$ соответственно. Докажите, что точка пересечения прямых $a_{t}$ и $b_{t}$ проективно движется по некоторой конике (возможно, вырожденной) b) Точки $A_{t}$ и $B_{t}$ проективно движутся по прямым $a$ и $b$ соответственно. Докажите, что прямая $A_{t} B_{t}$ огибает некоторую конику (или все время проходит через фиксированную точку).
При каких условиях в предыдущей задаче коника получается вырожденной?
29. Точки $A_{t}$ и $B_{t}$ движутся по прямой с постоянными скоростями. Какую конику огибает соединяющая их прямая?
Коники зачастую оказываются геометрическим местом точек различных объектов.
30. Даны окружность и прямая, пересекающиеся в точках $A$ и $B$. Найдите ГМТ точек, для которых касательная к окружности равна расстоянию до прямой.
31. На сторонах остроугольного треугольника $A B C$ как на основаниях строятся подобные равнобедренные треугольники $B A_{1} C, C B_{1} A, A C_{1} B$ (либо все внутрь, либо все наружу). Докажите, что прямые $A A_{1}, B B_{1}, C C_{1}$ пересекаются в одной точке, и найдите их ГМТ.
Иногда применение коник оказывается полезным в задачах, напрямую с ними не связанными.
32. В остроугольном неравнобедренном треугольнике $A B C$ отмечены изогонально сопряжённые точки $P$ и $Q$. Точка $W$ - середина дуги $B A C$ окружности $(A B C)$. Прямые $W P$ и $W Q$ второй
раз пересекают окружность $(A B C)$ в точках $X$ и $Y$ соответственно. Через точки $P$ и $Q$ проведены прямые, параллельные прямой $A W$; эти прямые пересекают стороны $A B, A C$ в точках $P_{B}, P_{C}, Q_{B}$, $Q_{C}$. Докажите, что точки $X, Y, P_{B}, P_{C}, Q_{B}, Q_{C}$ лежат на одной окружности.
33. Окружность, вписанная в неравнобедренный треугольник $A B C$, касается его сторон $B C$, $C A$ и $A B$ в точках $A_{1}, B_{1}$ и $C_{1}$ соответственно. Три мухи ползли по прямым $A A_{1}, B B_{1}$ и $C C_{1}$ с постоянными скоростями так, что в какой-то момент они находились в точках $A, B$ и $C$, а в другой момент были в точках $A_{1}, B_{1}$ и $C_{1}$. В некоторый момент времени все три мухи находились на прямой $p_{1}$, а в некоторый другой момент - на прямой $p_{2}$. Докажите, что $p_{1} \perp p_{2}$.
34. На сторонах $A D, C D$ четырехугольника $A B C D$ выбраны точки $P, Q$ так, что $\angle A B P=$ $\angle C B Q$. Обозначим за $S$ точку пересечения $C P$ с $A Q$. Докажите, что $\angle P B S=\angle Q B D$.
35. Рассмотрим произвольный треугольник $A B C$ с центром $I$ вписанной окружности. Прямая $\ell$ пересекает прямые $A I, B I$ и $C I$ соответственно в точках $D, E, F$, отличных от $A, B, C$ и $I$. Серединные перпендикуляры к отрезкам $A D, B E$ и $C F$ образуют треугольник $\Delta$ с описанной окружностью $\omega$. Докажите, что окружности $\omega$ и $(A B C)$ касаются.
## 4 Полиномиальное движение
Определение 4.1. Проективная плоскость. Проективной плоскостью называется множество всех прямых в пространстве, проходящих через фиксированную точку $O$, а эти прямые называются точками проективной плоскости.
На каждой прямой в $\mathbb{R}^{3}$, проходящей через начало координат, можно выбрать любую точку $(x, y, z)$, отличную от начала координат. Таким образом, все ненулевые тройки $[x: y: z]$ с точностью до пропорциональности кодируют точки проективной плоскости и называются однородными координатами на плоскости. Это определение естественно согласуется с данным ранее интуитивным: если зафиксировать в пространстве обычную плоскость $\alpha \mid z=1$, не содержащую точку $O$, то каждой ее точке $A$ сопоставляется прямая $O A$, а каждой бесконечно удаленной точке некоторого направления сопоставляется прямая, проходящая через точку $O$ параллельно $\alpha$ соответствующего направления. Тройка $[0: 0: 0]$ не задает никакую точку проективной плоскости.
Заметим, что оси $O X$ и $O Y$ трехмерных координат параллельны плоскости $\alpha$, а ось $O Z$ перпендикулярна ей. Спроецировав $O X$ и $O Y$ на $\alpha$, получим стандартную систему координат в этой плоскости. Точка с координатами $(x, y)$ в этих координатах имеет однородные координаты $[x: y: 1]$. Бесконечно удаленные точки проективной плоскости имеют однородные координаты $[x: y: 0]$
Пусть $p$ и $q$ - две различные точки проективной плоскости $\alpha$. Посмотрим на плоскость, проходящую через точки $p, q$ и $O,-$ она задается уравнением $a x+b y+c z=0$, где $a, b$ и $c$ - фиксированные числа, выбранные с точностью до пропорциональности. Однородными координатами прямой $p q$ мы будем называть тройку чисел $[a: b: c]$ с точностью до пропорциональности. Тройка $[0: 0: 0]$ не задает никакую прямую.
Легко видеть, что точка $\left[x_{0}: y_{0}: z_{0}\right]$ лежит на прямой $[a: b: c]$ тогда и только тогда, когда $a x_{0}+b x_{0}+c z_{0}=0$, откуда прямой подстановкой следует следующее утверждение:
Лемма 4.1. Коэффициент прямой, проходящей через точки с координатами $\left[x_{1}: y_{1}: z_{1}\right]$ и $\left[x_{2}\right.$ : $\left.y_{2}: z_{2}\right]$, можно записать как
$$
\left[y_{1} z_{2}-z_{1} y_{2}: z_{1} x_{2}-x_{1} z_{2}: x_{1} y_{2}-y_{1} x_{2}\right]
$$
а координату точки пересечения двух прямых $\left[a_{1}: b_{1}: c_{1}\right] u\left[a_{2}: b_{2}: c_{2}\right]$
$$
\left[b_{1} c_{2}-c_{1} b_{2}: c_{1} a_{2}-a_{1} c_{2}: a_{1} b_{2}-b_{1} a_{2}\right]
$$
Аналогичным образом можно определить однородные координаты на проективной прямой, кодируя ее точки парами чисел $[x: y]$ с точностью до пропорциональности. Поскольку время - это точка проективной прямой, ее можно рассматривать как пару чисел $\left[t_{1}: t_{2}\right]$.
Раннее мы рассматривали отображения из проективной прямой в проективную плоскость (а конкретно, в прямые и коники на ней), сохраняющие двойные отношения. Сейчас мы расширим наш арсенал полиномиальными отображениями, то есть такими функциями $\mathcal{F}: \mathbb{R} \mathbb{P}^{1} \rightarrow \mathbb{R P}^{2}$, которые на вход берут пару чисел $t_{1}, t_{2}$, а на выходе выдают тройку многочленов $P\left(t_{1}, t_{2}\right), Q\left(t_{1}, t_{2}\right)$ и $R\left(t_{1}, t_{2}\right)$, отвечающие тройке однородных координат точки на проективной плоскости.
Очевидно, что на многочлены $P, Q$ и $R$ нужно наложить некоторые условия, чтобы указанное отображение корректно отображало точки проективной прямой-времени в проективную плоскость.
Многочлены должны быть однородными ( $t_{1}^{2}+2 t_{1} t_{2}$ подходит, а $t^{3}+2 t_{1} t_{2}$ - нет) и одинаковой степени, чтобы можно было пропорционально заменять координаты. Также мы не хотим, чтобы для каких-то $t_{1}, t_{2}$ все три многочлена разом обращались в ноль, для этого потребуем их взаимную простоту.
Определение 4.2. Степенная зависимость. Будем говорить, что степень зависимости точки $X$ от времени равна $k$, если однородные координаты точки $X$ можно записать как $\left[P_{1}\left(t_{1}, t_{2}\right)\right.$ : $\left.P_{2}\left(t_{1}, t_{2}\right): P_{3}\left(t_{1}, t_{2}\right)\right]$, где $P_{i}$ - однородные полиномы степени $k$ от времени, которые взаимно просты в совокупности. Аналогично определим степени зависимости прямых.
Легко понять, что такое определение корректно.
Лемма 4.2. О сложении степеней. Пусть точки $X$ и $Y$ движутся со степенями а и $b$ соответственно. Тогда степень прямой $X Y$ не выше, чем $a+b$.
Доказательство. Обозначим однородные координаты точек за $\left[x_{1}: x_{2}: x_{3}\right]$ и $\left[y_{1}: y_{2}: y_{3}\right]$ соответственно, тогда координаты прямой $X Y$ задаются как $\left[x_{2} y_{3}-x_{3} y_{2}: x_{3} y_{1}-x_{1} y_{3}: x_{1} y_{2}-y_{1} x_{2}\right]$. Если $x_{i}$ - это многочлены степени не выше, чем $a$, а $y_{i}$ - не выше чем $b$, то указанные выражения полиномы степени не выше, чем $a+b$. Отметим, что оценка может быть не точна только в том случае, когда получившаяся тройка многочленов не взаимно проста в совокупности.
Лемма 4.3. Пусть точка $X$ степени 1 движется по прямой, не проходящей через точку $S$. Тогда степень прямой $S X$ также равна 1.
Доказательство. Из леммы о сложении степеней следует, что степень прямой $S X$ не выше, чем $0+1$. Так как прямая не неподвижна, то оценка точна.
В следующей главе мы докажем более общую теорему: если точка $X$ степени $a$ движется по произвольной траектории, не проходящей через точку $S$, то степень прямой $S X$ равна степени точки $X$.
Теорема 4.1. Степенъ зависимости точки $X$, проективно едущей по прямой, равна 1. Степень проективно вращающейся в пучке прямой также равна 1.
Доказательство. Для начала покажем, что любое проективное отображение из прямой в прямую можно разложить в композицию параллельных переносов и центральных проекций.
Пусть точки $A, B$ и $C$ прямой $\ell_{1}$ переходят в точки $A_{1}, B_{1}$ и $C_{1}$ прямой $\ell_{2}$. Если $\ell_{1} \| \ell_{2}$, то спроецируем $\ell_{1}$ на любую не параллельную им прямую и будем решать задачу для пары не параллельных прямых. Сделаем параллельный перенос прямой $\ell_{1}$, переводящий точку $A$ в точку $A_{1}$, точки $B$ и $C$ перешли при нем в $B^{\prime}$ и $C^{\prime}$. Обозначим за $S$ пересечение $B_{1} B^{\prime}$ и $C_{1} C^{\prime}$ (возможно, бесконечно удаленное). Проекция с центром в точке $S$, переводящая $\ell_{1}^{\prime}$ в $\ell_{2}$, реализует искомую гомографию.
Очевидно, что параллельный перенос не меняет степень зависимости, поэтому нужно проверить, что центральная проекция проективно движущейся по прямой точки не меняет степень зависимости - это следует из предыдущей леммы.
Теорема 4.2. Проективные преобразования плоскости линейно меняют однородные координаты.
Лемма 4.4. Об удвоении степени на конике. Точка $X$, проективно бегающая по окружности (конике), имеет степень зависимости 2.
Доказательство. Отметим на окружности две фиксированные точки $A, B$, тогда прямые $a=A X$ и $b=B X$ вращаются проективно, следовательно точка $X$ определяется как пересечение двух прямых со степенями зависимости 1 , тогда по лемме о сложении степеней получаем требуемое.
Замечание: Легко понять, верна и двойственная лемма: Пусть прямая $\ell$ вращается вокруг окружности (коники) так, что точка касания имеет степень движется проективно. Тогда степень $\ell$ оценивается как 2.
Теорема 4.3. Чтобы проверить, что точка степени $k$ всегда совпадает с точкой степени $l$, достаточно проверить $k+l+1$ положение.
Доказательство. Заметим, что совпадение точек в момент времени $\left[t_{1}: t_{2}\right]$ равносильно тому, что отношения многочленов от $t_{1}, t_{2}$, задающих координаты точек, равны:
$$
\left\{\begin{array}{l}
\frac{P_{x}\left(t_{1}, t_{2}\right)}{P_{y}\left(t_{1}, t_{2}\right)}=\frac{Q_{x}\left(t_{1}, t_{2}\right)}{Q_{y}\left(t_{1}, t_{2}\right)} \\
\frac{P_{y}\left(t_{1}, t_{2}\right)}{P_{z}\left(t_{1}, t_{2}\right)}=\frac{Q_{y}\left(t_{1}, t_{2}\right)}{Q_{z}\left(t_{1}, t_{2}\right)} \\
\frac{P_{z}\left(t_{1}, t_{2}\right)}{P_{x}\left(t_{1}, t_{2}\right)}=\frac{Q_{z}\left(t_{1}, t_{2}\right)}{Q_{x}\left(t_{1}, t_{2}\right)}
\end{array}\right.
$$
Что равносильно (если соответствующая пара знаменателей не обращается в ноль):
$$
\left\{\begin{array}{l}
P_{x}\left(t_{1}, t_{2}\right) Q_{y}\left(t_{1}, t_{2}\right)=P_{y}\left(t_{1}, t_{2}\right) Q_{x}\left(t_{1}, t_{2}\right) \\
P_{x}\left(t_{1}, t_{2}\right) Q_{z}\left(t_{1}, t_{2}\right)=P_{z}\left(t_{1}, t_{2}\right) Q_{x}\left(t_{1}, t_{2}\right)
\end{array}\right.
$$
А для того, чтобы проверить тождественное совпадение однородных многочленов от двух переменных степени не выше чем $k+l$, достаточно проверить $k+l+1$ не пропорциональное положение. Действительно: если многочлен обнуляется в точках $\left(x_{i}, y_{i}\right)$ при $y_{i} \neq 0$, то разделив $P\left(t_{1}, t_{2}\right)$ на $t_{2}^{d}$ где $d$ - степень многочлена, получим полином от одной переменной $\frac{t_{1}}{t_{2}}$, количество корней которого больше его степени, поэтому он тождественно равен нулю. Если же один из поданных нам моментов времени $[t: 0]$, то можно вынести $t_{2}$ как общий множитель и применить к оставшемуся предыдущее рассуждение.
Аналогично, для проверки инцидентности прямой степени $k$ и точки степени $l$ также достаточно проверять $k+l+1$ положение.
Сформулированный набор лемм является «базовым» и уже помогает решать множество не простых задач. Убедимся в этом на примере:
Пример 4.1. На прямой $B C$ неравнобедренного треугольника $A B C$ выбраны точки $P$ и $Q$ так, что $B P=C Q$. Пусть $\omega-$ вписанная окружность треугольника, $\omega_{A}-$ вневписанная, касающаяся отрезка BC.S и T на окружностях $\omega$ и $\omega_{A}$ соответственно таковы, что $P S$ касается $\omega$, a $Q T$ касается $\omega_{A}$. $A S$ и $A T$ пересекают $B C$ в $X$ и $Y$ соответственно. Доказать, что $B X=C Y$.
1. Введем на плоскости однородную систему координат и время. Будем двигать точку $P$ проективно по прямой $B C$. Будем обозначать степень зависимости точки за $d$.
2. Точка $S$ проективно зависит от $P$ и движется по окружности, поэтому, согласно лемме об удвоении степени точки на конике, $d(S) \leq 2$.
3. По лемме о сложении степеней $d(A S) \leq d(S)+d(A) \leq 2+0=2$
4. В силу той же леммы $d(X) \leq d(B C)+d(A S) \leq 0+2=2$. Аналогично $d(Y) \leq 2$, так как $P$ и $Q$ симметричны относительно $M$ середины $B C$, то есть, проективно зависимы.
5. Обозначим $Y^{\prime}=\mathcal{S}_{M}(X)$. Очевидно, $d\left(Y^{\prime}\right)=d(X) \leq 2$, значит, мы хотим доказать, что две точки со степенями зависимости не более чем 2 совпадают, для этого достаточно проверить $2+2+1=5$ положений.
6. Положения $P=C, P=B, P=A_{1}$ - точка касания вписанной окружности, $P=M$ - середина $B C, P=\infty$ очевидны.
При решении задач полезно понимать, как корректно обрабатывать вырожденные случаи. Пусть в какой-то момент мы хотим провести прямую через две совпадающие точки: из формул следует, что координаты этой прямой будут $[0: 0: 0]$ (то есть, у трех многочленов, задающих координаты прямой, будет общий множитель), что автоматически гарантирует обнуление всех последующих однородных многочленов; поэтому можно считать, что во всех вырожденных случаях условие задачи автоматически выполнено. Однако при таком подходе нужно действовать аккуратно. Разберем такой пример: точка $X$ проективно движется по конике, точка $S$ фиксирована на ней. Пусть требуется доказать, что некоторая точка $Y$ степени 1 всегда лежит на прямой $S X$. Мы хотим правильно интерпретировать прямую $S X$, когда $S=X$. Если мы оцениваем степень прямой $S X$ по лемме о сложении степеней, то ее степень не выше 2 , и, при $X=S$, прямая $S X$ «нулевая», что дает нам бесплатное положение (эффект того, что мы изначально не точно оценили степень). После этого, для решения задачи нам нужно найти еще три положения. Если же мы думаем про $S X$ как про прямую, проективно вращающуюся в пучке точки $S$, то нам изначально нужно проверять три положения, но при $X=S$ прямую $S X$, конечно, надо интерпретировать как касательную, а не как «нулевую».
Попробуйте попрактиковаться с теорией, решив несколько упражнений.
36. Точки $A$ и $B$ движутся с постоянными скоростями по двум прямым. Докажите, что направление прямой $A B$ проективно.
37. Три точки движутся проективно. Сколько положений достаточно проверить, чтобы убедиться, что они всегда лежат на одной прямой? Тот же вопрос для трех проективно вращающихся прямых.
38. Точки $X$ и $Y$ движутся со степенями $a$ и $b$. Докажите, что середина отрезка, их соединяющего, движется со степенью не выше, чем $a+b$.
39. На окружности $\omega$ фиксирована точка $P$ и движется точка $A$ степени $a$. Точка $B$ выбирается так, что дуга $P B$ в два раза больше дуги $P A$ (если считать против ч.с.). Докажите, что степень $B$ не выше $2 a$.
40. Полярное преобразование не меняет степень зависимости.
Теперь перейдем к решению задач.
41. Выпуклый шестиугольник $A B C D E F$ вписан в окружность. Треугольники $A C E$ и $B D F$ в пересечении образуют шестиугольник. Докажите, что главные диагонали этого шестиугольника пересекаются в одной точке.
42. Через ортоцентр остроугольного треугольника провели две перпендикулярные прямые. Докажите, что середины отрезков, которые эти прямые высекают на сторонах или продолжениях сторон треугольника, лежат на одной прямой.
43. Дан треугольник $A B C$ и три точки $P, Q, R$, лежащие на одной прямой. Прямые $A P, B P$ и $C P$ пересекают описанную окружность треугольника $A B C$ в точках $A^{\prime}, B^{\prime}$ и $C^{\prime}$. Прямые $A^{\prime} Q, B^{\prime} Q$ и $C^{\prime} Q$ пересекают ту же окружность в точках $A^{\prime \prime}, B^{\prime \prime}$ и $C^{\prime \prime}$. Прямые $A^{\prime \prime} R, B^{\prime \prime} R, C^{\prime \prime} R$ пересекают ту же окружность в точках $A^{\prime \prime \prime}, B^{\prime \prime \prime}$ и $C^{\prime \prime \prime}$. Доказать, что прямые $A A^{\prime \prime \prime}, B B^{\prime \prime \prime}, C C^{\prime \prime \prime}$ пересекаются в одной точке, лежащей на прямой, проходящей через точки $P, Q$ и $R$.
44. На описанной окружности треугольника $A B C$ отмечена точка $X$. Прямые $B X$ и $C X$ пересекают высоты $C C_{1}$ и $B B_{1}$ в точках $P, Q$ соответственно. Докажите, что середина отрезка $P Q$ лежит на прямой $B_{1} C_{1}$.
45. В остроугольном треугольнике $A B C$, вписанном в окружность $\Omega$ с центром $O$, проведена высота $A H_{A}$. Прямые $\ell_{A}, \ell_{B}, \ell_{C}$ касаются окружности $\Omega$ в точках $A, B, C$ соответственно. Точка $S$ - ортоцентр треугольника, образованного прямыми $\ell_{A}, \ell_{B}, \ell_{C}$. Докажите, что прямые $O H_{A}$ и $S H_{A}$ симметричны относительно прямой $B C$.
46. Вневписанная окружность треугольника $A B C$ имеет центр $I_{A}$, касается отрезка $B C$ в точке $A_{1}$ и касается прямых $A B, A C$ в точках $C_{1}, B_{1}$ соответственно. На прямой $I_{A} C_{1}$ выбрана точка $P$ так, что $A P \perp B I_{A}$. На прямой $I_{A} B_{1}$ выбрана точка $Q$ так, что $A Q \perp C I_{A}$. Докажите, что точки $P$, $Q, A_{1}$ лежат на одной прямой.
47. На прямой Эйлера неравнобедренного треугольника $A B C$ отмечена точка $X$, лежащая внутри треугольника; точка $O$ - центр окружности $(A B C)$. Прямые $A X, B X, C X$ пересекают соответственные стороны треугольника $A B C$, в точках $A_{1}, B_{1}, C_{1}$. Докажите, что окружности $\left(A O A_{1}\right)$, $\left(B O B_{1}\right),\left(C O C_{1}\right)$ имеют две общие точки.
48. Дан треугольник $A B C$ с ортоцентром $H$. На описанной около треугольника окружности выбираются точки $A_{1}, B_{1}$ и $C_{1}$ так, что прямые $A A_{1}, B B_{1}$ и $C C_{1}$ конкуррентны. Обозначим за $A_{2}, B_{2}$ и $C_{2}$ точки, симметричные $A_{1}, B_{1}$ и $C_{1}$ относительно середин соответствующих сторон треугольника. Докажите, что точки $A_{2}, B_{2}, C_{2}$ и $H$ лежат на одной окружности.
49. Дан треугольник $A B C$ с ортоцентром $H$. На описанной около треугольника окружности выбираются точки $A_{1}, B_{1}$ и $C_{1}$ так, что прямые $A A_{1}, B B_{1}$ и $C C_{1}$ конкуррентны. Обозначим за $A_{2}, B_{2}$ и $C_{2}$ точки, симметричные $A_{1}, B_{1}$ и $C_{1}$ относительно соответствующих сторон треугольника. Докажите, что точки $A_{2}, B_{2}, C_{2}$ и $H$ лежат на одной окружности.
50. Теорема Торнера. Пусть точки $P, Q$ инверсны относительно окружности $A B C, P_{C}$ симметрична $P$ относительно $A B, P_{C} Q$ пересекает $A B$ в точке $C^{\prime}$. Аналогично определяются точки $A^{\prime}$, $B^{\prime}$. Тогда $A^{\prime}, B^{\prime}, C^{\prime}$ лежат на одной прямой.
51. Выпуклый четырехугольник $A B C D$ таков, что $\angle B=\angle D$. Докажите, что середина диагонали $B D$ лежит на общей внутренней касательной к окружностям, вписанным в треугольники $A B C$
и $A C D$.
52. Дан треугольник $A B C$. Обозначим за $A_{1}$ пересечение средней линии, параллельной $B C$, с прямой, соединяющей основания высот на стороны $A B, A C$. Аналогично определим точки $B_{1}, C_{1}$. Докажите, что ортоцентр треугольника $A_{1} B_{1} C_{1}$ лежит на прямой Эйлера треугольника $A B C$.
53. Диагонали вписанного в окружность в $\omega$ четырехугольника $A B C D$ пересекаются в точке $P$. Обозначим за $I_{A}, I_{B}, I_{C}$ и $I_{D}$ центры окружностей, вписанных в треугольники $A P B, B P C, C P D$ и $D P A$ соответственно. Пусть $S_{A}, S_{B}, S_{C}$ и $S_{D}$ середины «меньших» дуг $A B, B C, C D$ и $D A$ окружности $\omega$. Докажите, что прямые $I_{A} S_{A}, I_{B} S_{B}, I_{C} S_{C}$ и $I_{D} S_{D}$ пересекаются в одной точке.
## 5 Полиномиальное движение +
В этой главе собрано три сюжета, которые можно изучать почти независимо друг от друга: обобщения ранее доказанных теорем, принцип относительности, и комбинирование теоремы Понселе с движением.
Для того, чтобы доказывать более продвинутые теоремы, связанные с полиномиальным движением, нам понадобится более мощный арсенал, чем вещественные числа. Все определения и теоремы предыдущей главы работают и для комплексных чисел. Вместо проективной прямой $\mathbb{R} \mathbb{P}^{1}$ мы будем рассматривать $\mathbb{C P}^{1}$ - множество прямых, проходящих через $(0,0)$ в $\mathbb{C}^{2}$. Однородные координаты на проективной комплексной прямой это пары чисел $\left[z_{1}: z_{2}\right]$, выбирающиеся с точностью до пропорциональности. То же и с комплексной проективной плоскостью $\mathbb{C P}^{2}$ и ее однородными координатами, пучками и т.д. Отметим, что многие результаты предыдущих глав также верны и для комплексных чисел. Комплексную единичную окружность на $\mathbb{C}^{2}$ нужно мыслить как множество решений уравнения $x^{2}+y^{2}=1$ в комплексных числах, а на $\mathbb{C P}^{2}$ как множество точек $[x: y: z]$, для которых $x^{2}+y^{2}=z^{2}$. Любая невырожденная коника получается из окружности проективным преобразованием (переводящем прямые в прямые) и может быть задана множеством нулей однородного неприводимого полинома $P(x, y, z)=0$.
54. Докажите, что комплексная прямая и невырожденная комплексная коника имеют не более двух общих точек. Могут ли они не пересекаться?
Комплексные числа лучше вещественных тем, что любой непостоянный полином $f(x)$ имеет корень (этим можно пользоваться без доказательств). Более того, любая кривая степени $n$ на $\mathbb{C P}^{2}$ (то есть, множество нулей однородного полинома степени $n$ ) пересекается с кривой степени $m$ по $m n$ точкам с учетом кратности. Это утверждение называется обобщенной теоремой Безу. В этой главе теорема Безу нам не понадобится - мы будем использовать лишь такой его частный случай:
55. а) Докажите, что однородный непостоянный многочлен $f\left(t_{1}, t_{2}\right)$ делится на некоторый линейный однородный многочлен $a t_{1}+b t_{2} \mathbf{b}$ ). Докажите, что любой однородный многочлен $f\left(t_{1}, t_{2}\right)$ единственным образом (с точностью до перестановок и умножения на константы) раскладывается в произведение $p_{1}^{\alpha_{1}} p_{2}^{\alpha_{2}} \ldots p_{n}^{\alpha_{n}}$, где $p_{i}$ - однородные линейные полиномы от двух переменных.
Не вооруженным взглядом видно, что последняя задача - аналог основной теоремы арифметики.
Лемма 5.1. Комплексная лемма о сложении степеней. Пусть точки $X$ и $Y$ движутся со степенями а и $b$ соответственно. Тогда степень прямой $X Y$ не выше чем $a+b$, причем если точки $X$ и $Y$ в любой момент времени различны, то эта оценка точная.
Доказательство. Обозначим однородные координаты точек за $\left[x_{1}: x_{2}: x_{3}\right]$ и $\left[y_{1}: y_{2}: y_{3}\right]$ соответственно, тогда координаты прямой $X Y$ задаются как $\left[x_{2} y_{3}-x_{3} y_{2}: x_{3} y_{1}-x_{1} y_{3}: x_{1} y_{2}-y_{1} x_{2}\right]$. Если $x_{i}$ - это многочлены степени не выше, чем $a$, а $y_{i}$ - не выше чем $b$, то указанные выражения полиномы степени не выше, чем $a+b$. Если оценка не точна, то у трех получившихся многочленов есть общий непостоянный множитель $d\left(t_{1}, t_{2}\right)$. Выберем у $d$ любой линейный делитель $p=a t_{1}+b t_{2}$. Тогда в момент времени $[b:-a]$ прямая $X Y$ была нулевая, то есть точки $X$ и $Y$ совпадали, противоречие.
Лемма 5.2. Пусть точка $X$ степени а движется по прямой, не проходящей через точку $S$. Тогда степень прямой $S X$ также равна а.
Доказательство. Следует из предыдущего утверждения.
Суммируя все вышеперечисленное получаем:
Теорема 5.1. Пусть $f$ - гомография прямых или пучков. Тогда $f$ не меняет степень зависимости.
Осталось разобраться с гомографиями, задействующими коники. Мы уже доказали, что при перебрасывании точки на конику ее степень удваивается. Верно и обратное.
Теорема 5.2. (О сбрасывании точки с коники). Точка $X$ степени $k$ движется по конике, точка $S$ фиксирована на ней. Тогда степень прямой $S X$ равна $\frac{k}{2}$, в частности, $k$ - четно.
Эту теорему, разбитую на задачи, Вы докажете (или нет) самостоятельно (все объекты в них предполагаются комплексными).
56. Дана коника $\omega$. Выберем произвольную точку $S$ BHE коники и прямую $\ell$, не содержащую $S$. Пусть $f$ - проекция коники из точки $S$ на прямую $\ell$. Докажите, что у всех точек прямой по два прообраза, кроме, быть может, конечного числа.
57. Точка $X$ полиномиально движется по конике $\omega$. Выберем НА $\omega$ произвольную точку $T$ и рассмотрим проекцию $\omega$ из $T$ на прямую $z=0$. Пусть образ точки $X$ при этой проекции - это $X^{\prime}$. Таким образом, мы построили отображение $g$ из комплексной прямой-времени в прямую $z=0$. a) Докажите, что существует такое число $k$, что у всех точек прямой $z=0$ кроме, быть может, конечного числа ровно $k$ прообразов при отображении $g$. b) Докажите, что точка $X$ посетила почти все точки $\omega$ ровно $k$ раз.
58. Обозначим за $X^{\prime \prime}$ образ точки $X$ при отображении $f$ из пред-предыдущей задачи. Посчитайте двумя способами, сколько раз точка $X^{\prime \prime}$ посетила общую точку прямой $\ell$ и выведите отсюда утверждение теоремы.
Начнем следующий сюжет с некоторого мотивирующего примера. Предположим, что вершины некоторого треугольника движутся со степенями $a, b$ и $c$ соответственно. Как можно оценить степень центра описанной около него окружности? Несложные вычисления показывают, что степень середины отрезка, соединяющего точки со степенями $a$ и $b$, движется со степенью не выше, чем $a+b$. Направление прямой, соединяющей вершины степеней $a$ и $b$, также имеет степень не выше, $a+b$, поэтому серединный перпендикуляр к соответствующей стороне треугольника имеет степень не выше, чем $(a+b)+(a+b)=2(a+b)$ (согласно лемме о сложении степеней). Аналогично, серединный перпендикуляр к стороне треугольника с вершинами степени $a$ и $c$ имеет степень не выше, чем $2(a+c)$. Пересекая эти серединные перпендикуляры и вновь применяя принцип сложения степеней, получим, что центр окружности движется со степенью не большей, чем $4 a+2 b+2 c$.
Это оценка заставляет задуматься: получившееся выражение не симметрично по степеням вершин треугольника, что странно. Вероятно, это должно означать, что предъявленная нами оценка не точна. Более того, появляется разумное подозрение, что реально можно оценить степень просто как $2(a+b+c)$. Давайте заведем инструмент, который позволит нам доказать нашу гипотезу.
Определение 5.1. Полиномиальная подстановка. Предположим, что на проективной плоскости зафиксированы произвольные подмножества $P_{1}, P_{2}, \ldots P_{n}$ и для любого набора точек $p_{1} \in$
$P_{1}, p_{2} \in P_{2} \ldots, p_{n} \in P_{n}$ определена точка $\left[p_{0_{x}}: p_{0_{y}}: p_{0_{z}}\right]=\mathcal{R}\left(p_{1}, p_{2}, \ldots, p_{n}\right)$, причем существуют однородные многочлены $R_{x}, R_{y}, R_{z}$ от 3 п переменных такие, что $p_{0_{x}}=R_{x}\left(p_{1_{x}}, p_{1_{y}} \ldots p_{n_{z}}\right)$ и аналогично для $p_{0_{y}}$ и $p_{0_{z}}$. Более того потребуем, чтобы многочлены $\mathcal{R}_{x}, \mathcal{R}_{y}$ и $\mathcal{R}_{z}$ были однородными многочленами одинаковой степени отдельно по координатам каждой из точек $p_{i}$. Такое $\mathcal{R}$ будем называть полиномиальной подстановкой от $n$ переменных.
Определение 5.2. Относительная степень зависимости. Пусть $\mathcal{R}$ - некоторая полиномиальная подстановка от двух переменных, точки $q_{1}$ и $q_{2}$ движутся с некоторыми степенями зависимости, также $q=\mathcal{R}\left(q_{1}, q_{2}\right)$. Пусть $r_{1}$ наименьшее такое натуральное число, что для любого фиксированного момента времени $\left[\tau_{1}: \tau_{2}\right]$ степень зависимости точки $\mathcal{R}\left(q_{1}, q_{2}\left(\tau_{1}, \tau_{2}\right)\right.$ ) не выше, чем $r_{1}$ (при разных фиксированных $\left[\tau_{1}: \tau_{2}\right]$ степень может быть разной). Такое $r_{1}$ мы будем называть относительно степенью точки $q$ от точки $q_{1}$.
Аналогично определяются относительные степени точки, зависящей более, чем от двух. Нужно фиксировать все точки $q_{i}$ и смотреть на степень зависимости точки $q$, когда она зависит только от одной переменной точки.
59. Принцип относительности. Пусть точки $q_{1}, q_{2} \ldots, q_{n}$ движутся полиномиально, и относительные степени точки $q=\mathcal{R}\left(q_{1}, q_{2_{t}}, \ldots q_{n_{t}}\right)$ равны $r_{1}, r_{2}, \ldots r_{n}$ для некоторой полиномиальной подстановки $\mathcal{R}$. Тогда степень зависимости точки $q$ не выше, чем $r_{1}+r_{2}+\ldots+r_{n}$.
60. Пусть вершины треугольника двигаются со степенями $a, b$ и $c$ соответственно, тогда центр описанной около этого треугольника окружности и его ортоцентр оба имеют степень зависимости не больше, чем $2(a+b+c)$.
61. Точки $A$ и $B$ движутся по конике $\omega$ проективно. а) Докажите, что прямая $A B$ огибает некоторую конику $\gamma$. b) Докажите, что коники $\omega$ и $\gamma$ касаются в двух точках на $\mathbb{C P}^{2}$ (то есть, точки касания могут иметь комплексные координаты). с) Докажите, что в конфигурации Понселе соседние вершины треугольника зависят друг от друга проективно тогда и только тогда, когда вращающийся треугольник - правильный.
62. Точки $A$ и $B$ едут по одной и той же конике проективно. Докажите, что пересечение касательных к конике в точках $A$ и $B$ проективно движется по некоторой конике.
63. Пусть точка $A$ движется по конике со степенью $2 a$, точка $B$ движется по произвольной траектории со степенью $b$. Обозначим за $C$ второе пересечение прямой $A B$ с коникой. Тогда степень зависимости точки $C$ не выше чем $2 a+2 b$.
64. Точки движутся со степенями $a, b, c$ и $d$. Докажите, что для того, чтобы проверить то, что они лежат на одной окружности, достаточно $2(a+b+c+d)+1$ положение.
Часто оказывается эффективным применение поризма Понселе не только ко вписанной и описанной окружности треугольника, но и ко вневписанной и описанной. При такой конфигурации есть шесть вырожденных положений: параметризуем динамику касательной $\ell$, вращающейся вокруг вневписанной окружности; можно выбрать два положения, когда $\ell$ касается описанной и вневписанной окружности, два положения, когда $\ell$ проходит через общие точки окружностей, а также два положения, когда получающиеся треугольник равнобедренный.
65. Обозначим за $\Omega$ и $\omega$ описанную и вневписанную окружности треугольника $A B C$. Пусть они
пересекаются в точках $X$ и $Y$. Общие внешние касательные $\Omega$ и $\omega$ касаются $\Omega$ в $U$ и $V$. Докажите, что касательные к $\omega$ в $X$ и $Y$ проходят через $U$ и $V$.
66. Обозначим за $\omega$ вневписанную окружность треугольника $A B C$, касающуюся отрезка $B C$. Общие внешние касательные к $\omega$ и $(A B C)$ касаются $(A B C)$ в точках $X$ и $Y$. Докажите, что прямая $X Y$ проходит через основания биссектрис, проведенных из вершин $B$ и $C$.
67. Вневписанная окружность $\omega$ треугольника $A B C$ касается отрезка $B C$ в точке $A_{1}$ и пересекает описанную окружность ( $A B C$ ) в точках $X$ и $Y$. Касательные к $\omega$ в точках $X$ и $Y$ пересекаются в точке $Z$. Обозначим за $S$ середину дуги $B A C$. Докажите, что прямые $S A_{1}$ и $A Z$ пересекаются на $(A B C)$.
68. Вневписанная окружность $\omega$ треугольника $A B C$ касается отрезка $B C$ в точке $D$ и пересекает описанную окружность $(A B C)$ в точках $K$ и $L$. Обозначим за $E$ основание высоты треугольника из вершины $A$. Докажите, что на прямых $K D$ и $L D$ найдутся такие точки $V$ и $N$, что четырехугольник $E V A N$ - ромб.
69. Прямая $\ell$ степени $d$ вращается вокруг точки $O$. Зафиксируем два различных вектора $\overrightarrow{O A}$ и $\overrightarrow{O B}$, с началом в точке $O$. Докажите, что $\frac{\sin \angle(\ell, \overrightarrow{O A})}{\sin \angle(\ell, \overrightarrow{O B})}=\frac{P\left(t_{1}, t_{2}\right)}{Q\left(t_{1}, t_{2}\right)}$, где $P\left(t_{1}, t_{2}\right)$ и $Q\left(t_{1}, t_{2}\right)$ - однородные полиномы степени $d$ от двух переменных.
70. Пусть $A B C$ треугольник с описанной окружностью $\omega$ и вневписанной окружностью $\Omega_{A}$, касающейся отрезка $B C$. Обозначим пересечения этих окружностей за $X$ и $Y$ соответственно. Точки $P$ и $Q$ - проекции точки $A$ на касательные в точках $X$ и $Y$ к $\Omega_{A}$. Точка $R$ - пересечение касательной к окружности $(A P X)$ в точке $P$ с касательной к окружности $(A Q Y)$ в точке $Q$. Докажите, что $A R \perp B C$.
## Решения некотрых задач
6. Дан ромб $A B C D$ с острым углом $B$. Точка $O$ - центр описанной окружности треугольника $A B C$. На продолжении луча $O C$ за точку $C$ выбрана точка $P$. Прямая $P D$ пересекается с прямой, проходящей через точку $O$ параллельно стороне $A B$, в точке $Q$. Докажите, что $\angle A Q O=\angle P B C$.
Доказательство. Зафиксируем ромб и будем двигать точку $P$ по прямой $O C$ проективно.
$$
\begin{aligned}
& \mathcal{L}_{A} \quad \xrightarrow{A} O C \quad \xrightarrow{D} \quad \mathcal{L}_{D} \quad \xrightarrow{D} \quad \ell \quad \xrightarrow{A} \mathcal{L}_{A} \xrightarrow{f} \mathcal{L}_{B} \\
& \begin{array}{cccccc}
A P & P & D P & Q & A Q & A P^{\prime}
\end{array}
\end{aligned}
$$
Мы хотим проверить, что $A P=A P^{\prime}$, где $f$ сдвиг на вектор $\vec{A} B$ и поворот на угол $B 2$ против чс. Положения $P=O, P=B, P=\inf$ очевидны.
11. Отображение $f$ из комплексной прямой $\mathbb{C}$ в себя задается в декартовых координатах как $f(z)=\frac{P(z)}{Q(z)}$, где $P$ и $Q-$ многочлены. Пусть дополнительно известно, что $f$ биективно. Докажите, что $f$ сохраняет двойные отношения любой четверки комплексных чисел.
Доказательство. НУО многочены взаимно просты. Известно, что для любого $c \in \mathbb{C}$ уравнение $P(z)-c Q(z)=0$ имеет ровно один корень. Выберем $c$ так, чтобы член при старшей чтепени не сокрощался. Предположим, что $P$ или $Q$ степени хотя бы 2 , тогда многочлен $P(z)-c Q(z)$ должен иметь кратные корни, то есть для единственного его корня $z_{0}$ выполненно $P\left(z_{0}\right)-c Q\left(z_{0}\right)=0$ и $P^{\prime}\left(z_{0}\right)-c Q^{\prime}\left(z_{0}\right)=0$. Домножим первое равенство на $Q^{\prime}\left(z_{0}\right)$, второе на $Q\left(z_{0}\right)$ и вычтем. Получим $P\left(z_{0}\right) Q^{\prime}\left(z_{0}\right)-P^{\prime}\left(z_{0}\right) Q\left(z_{0}\right)=0$, то есть, при $Q\left(z_{0}\right) \neq=0$ (а это верно почти при всех $c$ ) $\left(\frac{P}{Q}\right)^{\prime}\left(z_{0}\right)=0$. Таким образом, для всех $z$ кроме, быть может, конечного числа верно $\left(\frac{P}{Q}\right)^{\prime}\left(z_{0}\right)=0$, откуда $\frac{P(z)}{Q(z)}=$ const, что противоречит взаимной простоте многочленов.
14. Проекция окружности на себя. а) Пусть $\Omega$ - окружность, $S$ - произвольная точка плоскости, не лежащая на окружности. Каждой точке $X \in \Omega$ сопоставим вторую точку $\mathcal{F}(X)$ пересечения прямой $S X$ с $\Omega$. Тогда отображение $\mathcal{F}: \Omega \rightarrow \Omega$ проективно. b) Докажите, что отображение, сопоставляющие точке $X$ прямую $S X \in \mathcal{L} \mathrm{HE}$ проективно.
## Доказательство.
Точка $S$ вне $\Omega$ :
Заметим, что $S X \cdot S Y=$ const. Тогда проекцию окружности на себя можно представить как $Y=\mathcal{I}_{S}^{\sqrt{S X \cdot S Y}}$
Точка $S$ внутри $\Omega$ : Заметим, что $Y=\mathcal{S}_{S} \circ \mathcal{I}_{S}^{\sqrt{S X \cdot S Y}}$
Пункт b) следует из того, что люое проективное отображение должно быть проективно.
13. Перебрасывание окружности на касательную. Пусть точка $X$ движется по прямой $\ell$, которая касается окружности $\Omega$. Обозначим основание второй касательной из $X$ к $\Omega$ за $Y=\mathcal{F}(X)$. Тогда отображение $\mathcal{F}: \ell \rightarrow \Omega$ проективно.
Доказательство. $Y=\mathcal{H}_{S}^{2} \circ \mathcal{I}_{\Omega}$, где $S$ точка касания прямой $\ell$ с окружностью.
14. Перебрасывание окружности на себя через прямую. а) Точка $X$ лежит на окружности $\Omega, \ell$ - фиксированная прямая, не касающаяся $\Omega$. Пусть касательная к $\Omega$, восстановленная в $X$, пересекает $\ell$ в точке $Z, Y=\mathcal{F}(X)$ - основание второй касательной из $Z$ к $\Omega$. Тогда отображение $\mathcal{F}: \Omega \rightarrow \Omega$ проективно. b) Докажите, что отображение, сопоставляющие точке $X$ точку $Z \in \ell \mathrm{HE}$ проективно.
Доказательство. Пусть $L$ полюс прямой $\ell$ относительно $\Omega$. Заметим, что $X Z$ поляра точки $Y \in \ell$, значит, $L \in X Z$, тогда $\mathcal{F}$ просто проекция окружности на себя с центром в фиксированной точке $L$.
Пункт b) следует из того, что люое проективное отображение должно быть проективно.
22. Выпуклый шестиугольник $A Q C P B R$ вписан в окружность $\Omega$, и при этом треугольники $A B C$ и $P Q R$ описаны около одной и той же окружности $\gamma$. Прямая $\ell$, параллельная прямой $B C$ и не совпадающая с ней, касается $\gamma$. Пусть $P_{1}$ - точка пересечения прямой $\ell$ и отрезка $Q R$. Докажите, что $\angle P A B=\angle P_{1} A C$.
Доказательство. 1. Будем двигать $P$ по $\Omega$ проективно, зафиксировав треугольник $A B C$, тогда треугольник $P Q R$ будет вращаться по Понселе. По доказанной лемме точка $P_{2}$ касания $\gamma \mathrm{c}$ отрезком $Q R$ это образ точки $P$ при гомографии Понселе. Согласно лемме о перебрасывании на касательную, примененной к окружности $\gamma$ и касательной $\ell$ получаем, что точка $P_{1}$ проективно зависит от точки $P_{2}$, обозначим соответствующую гомографию за $\mathcal{F}$, за $g$ - внутреннюю биссектрису $\angle B A C$.

3. Утверждение задачи равносильно совпадению точек $P$ и $P^{\prime}$, поэтому достаточно доказать его для каких-нибудь трех положений точки $P$. Положения $P=A, B, C$ очевидны.
4. а) Прямые $a_{t}$ и $b_{t}$ проективно вращаются вокруг точек $A$ и $B$ соответственно. Докажите, что точка пересечения прямых $a_{t}$ и $b_{t}$ проективно движется по некоторой конике (возможно, вырожденной) b) Точки $A_{t}$ и $B_{t}$ проективно движутся по прямым $a$ и $b$ соответственно. Докажите, что прямая $A_{t} B_{t}$ огибает некоторую конику (или все время проходит через фиксированную точку).
Доказательство. Рассмотрим три точки $X, Y, Z$, принадлежащие ГМТ. Проведем конику $\Omega$ через точки $A, B, X, Y, Z$. Пусть прямая $l_{T}$, проходящая через $A$ пересекает конику в точке $T$, а соответствующая ей прямая, проходящая через $B$ пересекает $\Omega$ в точке $T^{\prime}$. Обозначим за $l_{i}$ прямые, проходящие через $A, k_{i}$ - через $B$. Тогда: $(X, Y ; Z, T)=\left(\ell_{X}, \ell_{Y} ; \ell_{Z} ; \ell_{T}\right)=\left(k_{X}, k_{Y} ; k_{Z} ; k_{T^{\prime}}\right)=\left(X, Y ; Z, T^{\prime}\right)$
Откуда следует, что $T=T^{\prime}$. Более того видно, что точка $T$ движется по конике проективно.
Двойственное утверждение получектся полярным преобразованием.
32. В остроугольном неравнобедренном треугольнике $A B C$ отмечены изогонально сопряжённые точки $P$ и $Q$. Точка $W$ - середина дуги $B A C$ окружности $(A B C)$. Прямые $W P$ и $W Q$ второй раз пересекают окружность ( $A B C$ ) в точках $X$ и $Y$ соответственно. Через точки $P$ и $Q$ проведены прямые, параллельные прямой $A W$; эти прямые пересекают стороны $A B, A C$ в точках $P_{B}, P_{C}, Q_{B}$, $Q_{C}$. Докажите, что точки $X, Y, P_{B}, P_{C}, Q_{B}, Q_{C}$ лежат на одной окружности.
Доказательство. Точки $P_{B}, P_{C}, Q_{B}, Q_{C}$ образуют равнобедренную трапецию, поэтому лежат на одной окружности. Мы докажем, что точки $P_{B}, P_{C}, X, Y$ лежат на одной окружности; аналогично доказывается про точки $Q_{B}, Q_{C}, X, Y$. Если эти три окружности не совпадают, то их радикальные оси $P_{B} P_{C}, Q_{B} Q_{C}$ и $X Y$ пересекаются в одной точке. Из соображений непрерывности можно понять, что окружности совпадают.
Фиксируем треугольник $A B C$ и точки $P_{B}, P_{C}$ и двигаем точку $P$ по прямой $P_{B} P_{C}$. Тогда точка $Q$ движется по конике, проходящей через вершины треугольника $A B C$, а также точку $W$, изогонально сопряженную бесконечно удаленной точке прямой $P_{B} P_{C}$. Поэтому прямые $W P, W Q$, а значит, и точки $X, Y$ двигаются проективно. Пусть $X P_{B}$ и $Y P_{C}$ вторично пересекают окружность $(A B C)$ в точках $S$ и $T$. Достаточно доказать, что $S T \| P_{B} P_{C}$. Поскольку точки $S$ и $T$ движутся проективно, то достаточно проверить 3 положения. Положения, когда $P=P_{B}, P=P_{C}$, и когда $P$ на окружности $(A B C)$, очевидны.
36. Точки $A$ и $B$ движутся с постоянными скоростями по двум прямым. Докажите, что направление прямой $A B$ проективно.
Доказательство. Напомним, что направление прямой $A B$ - бесконечно удаленная точка прямой $A B$. Выберем неподвижную точку $O$ и отложим от нее вектор $\overrightarrow{O C}=\overrightarrow{A B}$. Поскольку точка $C$ движется линейно, то направление прямой $O C$ зависит от времени проективно, что и требовалось.
37. Три точки движутся проективно. Сколько положений достаточно проверить, чтобы убедиться, что они всегда лежат на одной прямой? Тот же вопрос для трех проективно вращающихся прямых.
Доказательство. Проективно движущиеся по прямым точки имеют степень зависимости 1. Проведем прямую через две из данных точек, она имеет степень зависимости не выше 2. Чтобы проверить, что на ней лежит оставшаяся точка степени 1 , нужно проверить 4 положения. Рассуждение для проективно вращающихся прямых двойственно рассуждению для точек.
38. Точки $X$ и $Y$ движутся со степенями $a$ и $b$. Докажите, что середина отрезка, их соединяющего, движется со степенью не выше, чем $a+b$.
Доказательство. Пусть точки имеют однородные координаты $\left[x_{1}: y_{1}: z_{1}\right]$ и $\left[x_{2}: y_{2}: z_{2}\right]$. Тогда их декартовы координаты $\left(\frac{x_{1}}{z_{1}}, \frac{y_{1}}{z_{1}}\right)$ и $\left(\frac{x_{2}}{z_{2}}, \frac{y_{2}}{z_{2}}\right)$, а декартовы координаты середины имеют вид $\left(\frac{x_{1} z_{2}+x_{2} z_{1}}{2 z_{1} z_{2}}, \frac{y_{1} z_{2}+y_{2} z_{1}}{2 z_{1} z_{2}}\right)$. Тогда однородные координаты середины $\left[x_{1} z_{2}+x_{2} z_{1}: y_{1} z_{2}+y_{2} z_{1}: 2 z_{1} z_{2}\right]-$ многочлены степени $a+b$ от времени.
40. Полярное преобразование не меняет степень зависимости.
Доказательство. Проективные преобразования плоскости линейны в однородных координатах, поэтому не меняют степень зависимости. Сделаем проективное преобразование, переводящее конику в единичную окружность $x^{2}+y^{2}=z^{2}$. Несложно видеть, что поляра точки $[x: y: z]$ относительно этой окружности - прямая с координатами $[x: y:-z]$, откуда следует, что степень зависимости сохраняется.
47. На прямой Эйлера неравнобедренного треугольника $A B C$ отмечена точка $X$, лежащая внутри треугольника; точка $O$ - центр окружности $(A B C)$. Прямые $A X, B X, C X$ пересекают соответственные стороны треугольника $A B C$, в точках $A_{1}, B_{1}, C_{1}$. Докажите, что окружности $\left(A O A_{1}\right)$, $\left(B O B_{1}\right),\left(C O C_{1}\right)$ имеют две общие точки.
Доказательство. Считаем, что треугольник неравнобедренный. Пусть $A_{2}, B_{2}, C_{2}$ - образы $A_{1}, B_{1}$, $C_{1}$ при инверсии относительно описанной окружности. Будем доказывать, что прямые $A A_{2}, B B_{2}$, $C C_{2}$ пересекаются в одной точке. Фиксируем треугольник $A B C$ и двигаем $X$ по прямой Эйлера. Заметим, что $A_{2}, B_{2}, C_{2}$ двигаются проективно по окружностям, поэтому их степени зависимости равны 2. Нужно проверить 7 положений. Если $X$ лежит на одной из сторон треугольника, утверждение очевидно. Положения, когда $X$ - ортоцентр, точка пересечения медиан и центр описанной окружности, легко проверяются. Найдем 7 -ое положение: пусть окружность ( $B O C$ ) пересекает $A B$ и $A C$ в точках $X$ и $Y$; и пусть $M$ и $N$ - середины $A B$ и $A C$. Тогда $X N$ и $Y M$ пересекаются в точке $O$, и по теореме Паппа точка $X$ пересечения $C X$ и $B Y$ лежит на прямой Эйлера. Она и будет седьмым положением.
# Point moving
David Brodsky
Presented by: David Brodsky, Alexey Zaslavskiy, Oleg Zaslavskiy, Ivan Frolov and Fedor Nilov
## 1 Projective movement
Lemmas and theorems formulated in this section may be used without proof. As usually they are standard assertions of projective geometry - if you want it is not hard to verify their correctness.
Definition 1.1 Projective plane. The projective plane $\mathbb{R P}^{2}$ is the Euclidean plane completed by infinite points, each such point is the common point of some class of parallel lines. So each line is completed by one infinite point, and this point is the same for all parallel lines. All infinite points form the infinite line. A line completed by an infinite point is called a projective line.
Definition 1.2 Cross-ratio of four points on a projective line Let points $A, B, C, D$ lie on a projective line $\ell$. The cross-ratio $(A, B ; C, D)$ of these points on $\ell$ is the value
$$
\frac{\overrightarrow{C A}}{\overrightarrow{C B}}: \frac{\overrightarrow{D A}}{\overrightarrow{D B}}
$$
We suppose that an infinite point divide any segment in ratio $1: 1$.
Definition 1.3 Pencil of lines. The pencil of lines $\mathcal{L}_{A}$ of point $A$ is the set of all lines passing through A.
Not that the set of all parallel lines is the pencil of the corresponding infinite point.
Definition 1.4 Cross-ratio of four lines on a pencil. Let lines $a, b, c, d$ pass through a point $O$ (the set of lines passing through $O$ is a pencil). Choose arbitrary directing vectors $\vec{v}_{a}, \vec{v}_{b}, \vec{v}_{c}, \vec{v}_{d}$ on these lines. The cross-ratio $(a, b ; c, d)$ of the given lines is the value
$$
\frac{\sin \angle\left(\vec{v}_{c}, \vec{v}_{a}\right)}{\sin \angle\left(\vec{v}_{c}, \vec{v}_{b}\right)}: \frac{\sin \angle\left(\vec{v}_{d}, \vec{v}_{a}\right)}{\sin \angle\left(\vec{v}_{d}, \vec{v}_{b}\right)}
$$
Remark: Generally the sign of $\sin \angle\left(\vec{v}_{x}, \vec{v}_{y}\right)$ is not defined by the lines $x$ and $y$, because we may choose the directing vector by two ways. But if we change the direction of one vector, the signs of two sines invert, and the value of the cross-ratio does not change.
Definition 1.5 Cross-ratio of points on a circle Let points $A, B, C, D$ lie on a circle $\Omega$. Mark an arbitrary point $O$ on this circle. The cross-ratio $(A, B ; C, D)$ of points on $\Omega$ is he cross-ratio of lines $(O A, O B ; O C, O D)$. This definition is correct because the value of an inscribed angle is constant.
Lemma 1.1 Let points $A, B, C, D$ lie on a line $\ell$, and a point $O$ lie not on this line. Then
$$
(A, B ; C, D)=(O A, O B ; O C, O D)
$$
1. Let we consider points of the plane as complex numbers. Make sure that the cross-ratio of four points of a line or a circle corresponding to $a, b, c, d$ equals $\frac{c-a}{c-b}: \frac{d-a}{d-b}$.
Later we will define the cross-ratio for four points of an arbitrary conic. In this case the result of the last problem IS NOT CORRECT.
Definition 1.6 Homography. A map $\mathcal{F}$ is called projective, or homography if
$$
(\mathcal{F}(A), \mathcal{F}(B) ; \mathcal{F}(C), \mathcal{F}(D))=(A, B ; C, D)
$$
. The map may transform a line, a pencil or a circle (a conic) to a line, a pencil or a circle (a conic).
Lemma 1.2 Let $A, B, C \in \ell, k \in \mathbb{R}$, then there exists a unique point $D \in \ell$ such that $(A, B, C, D)=k$.
Theorem 1.1 Each homography of lines may be presented as a composition of several central projections and parallel translations.
Definition 1.7 Projective movement. Fixe some projective line $\mathcal{T}$. Say that a point $X$ moves on a projective line or a circle $\ell$ projectively, if there exists a projective map $\mathcal{F}: \mathcal{T} \rightarrow \ell$ such that $X=\mathcal{F}(t \in \mathcal{T})$. The line $\mathcal{T}$ may be considered as the axis of time.
A projective rotation of a line on a pencil is defined similarly.
Definition 1.8 The line rotating projectively The line $\ell$ rotates projectively in the pencil $\mathcal{L}$ if there exists a projective map $\mathcal{F}: \mathcal{T} \rightarrow \mathcal{L}$
such that $\ell=\mathcal{F}(t \in \mathcal{T})$.
Later we will use the following notations for several special homographies.
- $\ell \xrightarrow{S} \mathcal{L}_{S}$ - projection centered at point $S$, mapping a line $\ell$ to a pencil $\mathcal{L}_{S}$.
- $\mathcal{R}_{S}^{\psi}$ - rotation centered at $S$ to angle $\psi$ counterclockwise.
- $\mathcal{H}_{S}^{k}$ - homothety centered at $S$ with coefficient $k$
Demonstrate, how the projective movement helps to solve problems.
Example 1.1 Let $A_{1}$ be the touching point of the incircle of triangle $A B C$ with the side $B C$, and $D$ be an arbitrary point on $B C$. Denote as $I_{B}$ and $I_{C}$ the incenters of triangles $\triangle A B D, \triangle A C D$ respectively. Prove that $\angle I_{B} A_{1} I_{C}=90^{\circ}$
Divide the solution into several typical steps.
1. Fix $\triangle A B C$ and move projectively $I_{B}$ on the bisector of $\angle A B C$.
2. Construct a homography $\mathcal{F}: B I \rightarrow B I$, the identity of this homography is equivalent to the required assumption.
Present «a chain» of homographies as a table, denote as $\ell_{b}, \ell_{c}$ the bisectors of $\angle B$ and $\angle C$ perspectively:

We indicate the lines and the pencils transformed by the homographies in the top row, and the corresponding moving objects in the bottom one. The assumption of the problem is equivalent to the identity $I=I^{\prime}$
3. Verify that the assumption is correct for some three positions of $I_{B}$. As usually it is useful to choose degenerated positions. For the given problem consider $I_{B}=B, I_{B}=I, I_{B}=P$, where $P$ is the incenter of triangle $A B A_{1}$. For each of these cases the problem is clear. Since a homography is uniquely defined by the images of three points, the constructed homography of $\ell_{1}$ is the identity, hence the assumption is correct for any position of $I_{B}$.
Now train to solve problems using the projective movement. In several problems it is useful to remember that directions are points of the infinite line.
2. A line $\ell$ rotates projectively around a fixed point $P$. A point $S \neq P$ is fixed. Prove that the projection of $S$ to $\ell$ moves projectively. Find the trajectory of this point
3. The external bisectors $B B_{1}$ and $C C_{1}$ of triangle $A B C$ meet at point $I_{A}$. A line $\ell$ passing through $I_{A}$ meets $A B$ and $A C$ at points $X$ and $Y$ respectively. Prove that the reflections of $B Y$ and $C X$ about $B B_{1}$ and $C C_{1}$ respectively meet on $B_{1} C_{1}$.
4. Let points $B_{1}, C_{1}$ lie on the sides $A C, A B$ of triangle $A B C$ in such a way that lines $B B_{1}, C C_{1}$ meet on the altitude $A A_{1}$. Prove that the lines $A_{1} B_{1}$ and $A_{1} C_{1}$ are symmetric with respect to $A A_{1}$.
5. Points $A_{2}, B_{2}$ and $C_{2}$ lie on cevians $A A_{1}, B B_{1}, C C_{1}$ (i.e. concurrent lines) respectively of triangle $A B C$. Let $A_{3}=B C_{2} \cap B_{2} C$. points $B_{3}$ and $C_{3}$ are defined similarly. Prove that $A A_{3}, B B_{3}$ and $C C_{3}$ concur.
6. Let $A B C D$ be a rhombus with acute angle $B$. Let $O$ the circumcenter of triangle $A B C$. A point $P$ lie on the extension of $O C$ beyond $C$. The line $P D$ meets the line passing through $O$ and parallel to $A B$ at point $Q$. prove that $\angle A Q O=\angle P B C$.
7. Let $A B C$ be an acute-angled triangle with circumcircle $\Omega$ and incircle $\omega$. A point $P$ lies on a segment joining the centers of $\Omega$ and $\omega$. Denote as $A^{\prime}, B^{\prime}$ and $C^{\prime}$ the second common points of $A P, B P$ and $C P$ with $\Omega$. Prove that the internal bisectors of $\angle B A^{\prime} C, \angle C B^{\prime} A$ and $\angle A C^{\prime} B$ concur at a point lying on the center line of $\Omega$ and $\omega$.
8. Let $A B C D$ be a circumscribed quadrilateral with incenter $I$. Points $P$ and $Q$ lie on $A I$ and $C I$ respectively in such a way that $\angle P B Q=\frac{1}{2} \angle A B C$. Prove that $\angle P C Q=\frac{1}{2} \angle A D C$
9. Let $S$ be the projection of the orthocenter of triangle $A B C$ to the median $A M$. A circle $\omega$ passes through $A$ and $S$, and meets $A B$ and $A C$ at $Q$ and $P$ respectively. Prove that $B P$ and $C Q$ meet on $\omega$.
10. a) Let $f$ be a homography of a projective line $\ell$ to itself. Parametrize the finite points of the line by $x$. Prove that $f$ is fractionally linear, i.e. $f(x)=\frac{a x+b}{c x+d}$, where $a, b, c$ and $d$ are fixed numbers. b) Let $f$ be a homography from $\ell_{1}$ to $\ell_{2}$. Set Cartesian coordinates on a plane. Prove that $f(x, y)=\left(\frac{a_{1} x+b_{1} y+c_{1}}{d x+e y+f}, \frac{a_{2} x+b_{2} y+c_{2}}{d x+e y+f}\right)$, where $a_{1}, a_{2}, b_{1}, b_{2}, c_{1}, c_{2}, d, e$ are fixed numbers. c) Prove that any map of $\mathbb{R}^{\not}$ defined by the formulas of p. b) may be uniquely extended to a projective map of the projective plane.
11. A map $f$ from a complex line $\ell_{1}$ to itself is defined as $f(z)=\frac{P(z)}{Q(z)}$, where $P, Q$ are polynomials. Let $f$ be a bijection. Prove that $f$ conserve the cross-ratio of any four points.
## 2 Projective movement +
It is a wonder that an inversion limited to a line or a circle also conserve cross-ratios!
Theorem 2.1 Let an inversion $\mathcal{I}$ maps a circle or a line $\Omega$ to a circle or a line $\widetilde{\Omega}$. Then for any four points $A, B, C, D \in \Omega$ we have $(A, B ; C, D)=(\mathcal{I}(A), \mathcal{I}(B) ; \mathcal{I}(C), \mathcal{I}(D))$
The simplest proof of this theorem use the following properties of complex numbers: a number equals to its conjugated if and only if it is real; the cross-ratio of four complex numbers is real if and only if the corresponding points are collinear or concyclic; the inversion centered at the origin with radius 1 maps $z$ to $\frac{1}{\bar{z}}$.
Denoting the complex coordinates of points as $a, b, c$ and $d$ we obtain:
$$
\begin{gathered}
\left(a^{*}, b^{*} ; c^{*}, d^{*}\right)=\left(\frac{1}{\bar{a}}, \frac{1}{\bar{b}} ; \frac{1}{\bar{c}}, \frac{1}{\bar{d}}\right)=\frac{\frac{1}{\bar{c}}-\frac{1}{\bar{a}}}{\frac{1}{\bar{c}}-\frac{1}{\overline{\bar{d}}}-\frac{1}{\overline{\bar{a}}}}= \\
=\frac{\overline{\bar{b}}-\frac{1}{\bar{b}}}{\bar{c}-\bar{a}}: \frac{\bar{d}-\bar{a}}{\bar{d}-\bar{b}}=\frac{\overline{c-a}}{c-b}: \frac{d-a}{d-b}=\frac{c-a}{c-b}: \frac{d-a}{d-b}=(a, b ; c, d)
\end{gathered}
$$
This allows to do the following useful remark:
12. Projection of a circle to itself. a) Let $\Omega$ be a circle, and $S$ be an arbitrary point not lying on it. Map each point $X \in \Omega$ to the second common point $\mathcal{F}(X)$ of $S X$ with $\Omega$. Then the map $\mathcal{F}: \Omega \rightarrow \Omega$ is projective. b) Prove that the map transforming a point $X$ to the line $S X$ is NOT projective
13. Transfering of a circle to a tangent Let $X$ move along a line $\ell$ touching a circle $\Omega$. Let the second tangent from $X$ to $\Omega$ touche it at $Y=\mathcal{F}(X)$. Then $\mathcal{F}: \ell \rightarrow \Omega$ is projective.
14. Transfering of a circle to itself via a line. Let $X$ lie on a circle $\Omega$, and $\ell$ be a fixed line. Let the tangent to $\Omega$ at $X$ meet $\ell$ at point $Z$, and $Y=\mathcal{F}(X)$ be the base of the second tangent from $Z$ to $\Omega$. Then $\mathcal{F}: \Omega \rightarrow \Omega$ is projective.
The lemmas proved above help to solve problems using the projective movement.
15. Let $\gamma_{A}, \gamma_{B}, \gamma_{C}$ - be the excircles of triangle $A B C$ touching the sides $B C, C A, A B$ respectively. Denote as $\ell_{A}$ the common external tangent of $\gamma_{B}$ and $\gamma_{C}$ distinct from $B C$. Define $\ell_{B}, \ell_{C}$ similarly. Draw from a point lying on $\ell_{A}$ the tangent to $\gamma_{B}$ distinct from $\ell_{A}$ and find its common point $X$ with $\ell_{C}$. Similarly find the common point $Y$ of the tangent from $P$ to $\gamma_{C}$ and $\ell_{B}$. Prove that the line $X Y$ touches $\gamma_{A}$.
16. An acute-angled triangle $A B C$ with orthocenter $H$ is inscribed into a circle $\omega$ centerd at $O$. A line $l$ passes through $H$ and meets the minor $\operatorname{arcs} A B$ and $A C$ at points $P$ and $Q$ respectively. Let $A A^{\prime}$ be the diameter of $\omega$. The lines $A^{\prime} P$ and $A^{\prime} Q$ meet $B C$ at points $K$ and $L$ respectively. Prove that $O, K$, $L$ and $A^{\prime}$ are concyclic.
17. Let $S$ be the projection of the incenter $I$ of an circumscribed quadrilateral $A B C D$ to the diagonal $A C$. Prove that $\angle B S A=\angle D S A$
18. A pentagon $A B C D E$ is circumscribed around a circle $\omega$. The pairs of rays $E A$ and $C B, A E$ and $C D, A B$ and $D C, B C$ and $E D$ meet at points $P, Q, X, Y$ respectively. The circle $\omega$ touches $A E$ at point $R$. It is known that $X Y \| A E$. Let the circles $(A X Q),(P Y E)$ meet at points $S, T$. Prove that $S, T, R$ are collinear.
19. A line $\ell$ passes through the circumcenter $O$ of triangle $A B C$ and meets the sides $A B, A C$ at points $P, Q$ respectively. Prove that one common point of the circles with diameters $B Q, C P$ lies on the nine-points-circle of triangle $A B C$, and the second one lies on its circumcircle.
20. 2023 lines concur. A circle is inscribed into each of 4046 formed angles in such a way that the circle inscribed into adjacent angles are tangent. A point is marked on each side of the angles. For each angle except one the segment joining the points marked on the sides of this angle touches the inscribed circle. Prove that this is correct for the remaining angle.
21. F ix the sidelines $A B, A C$ and the incircle $\omega$ of triangle $A B C$. Prove that the touching point of $\omega$ with the side $B C$ projectively depends on the Feuerbach point of this triangle
Theorem 2.2 (May be used without proof) Let $A B C$ be a triangle with circumcircle $\Omega$ and incircle (or excircle) $\omega$. Let $A^{\prime}$ lie on $\Omega$, and the tangents from it to $\omega$ meet $\Omega$ for the second time at $B^{\prime}$ and $C^{\prime}$. Then the line $B^{\prime} C^{\prime}$ touches $\omega$.
Note that we have in this configuration a bijection: each point $A$ of the circumcircle corresponds to the touching point of $B C$ with the incircle. A hypothesis emerges that the corresponding map between to circles is projective.
Lemma 2.1 Homographies of Poncelet configuration. Consider a Poncelet rotation of triangle $A B C$ conserving the circumcircle and the incircle. Let $A_{1}$ be the touching point of side $B C$ with the incircle, $S_{A}$ and $T_{A}$ be the midpoints of two arcs $B C$, and $I_{A}$ be the center of the excircle touching the side $B C$, then all these points projectively depend on $A$.
Call the map transforming $A$ to $A_{1}$ as Poncelet homography and denote it as $\mathcal{P}$.
Mark the incenter $I$ and the circumcenter $O$ of triangle $A B C$. Note that
- The midpoint $S_{A}$ of arc $B C$ not containing $A$ is the image of $A$ in the projection of the circumcirle to itself centered at $I$.
- The point $T_{A}$ is the image of $S_{A}$ in the projection centered at $O$.
- Point $A_{1}$ is obtained from $S_{A}$ by the homothety mapping the circumcircle to the incircle.
- By the trident theorem $I_{A}$ is the image of the midpoint $S_{A}$ of $\operatorname{arc} B C$ in the homothety centered at $I$ with coefficient 2.
The homographies mapping $A$ to $S_{A}, T_{A}, A_{1}, I_{A}$ are presented, thus these points projectively depend on $A$
Usually the Poncelet homographies work effectively with the method of polynomial movement generalizing the projective movement. But it is possible to solve several difficult problems using the methods described above.
22. A convex hexagon $A Q C P B R$ is inscribed into a circle $\Omega$ in such a way that the triangles $A B C$ and $P Q R$ have a common incircle $\gamma$. A line $\ell$ parallel to $B C$ and distinct from it touches $\gamma$. Let $P_{1}$ be the common point of $\ell$ and $Q R$. Prove $\angle P A B=\angle P_{1} A C$.
23. Triangles $A B C$ and $D E F$ have a common incircle $\omega$ and circumcircle $\gamma$. Let $L$ and $K$ be the touching points of $B C$ and $E F$ respectively with $\omega$, and $M, N$ be the second common points of $A L$ and $D K$ respectively with $\gamma$, Prove that $A M, E F, B C, N D$ concur.
24. Let the vertices $A$ and $B$ of a triangle projevtively move in the Poncelet rotation. Prove that this triangle is regular.
The last problem may seem too hard. We propose to return to it after the learning of the relativity principle.
It is known that the Poncelet theorem is correct not only for a triangle but for ane $n$-gon. So the following question is natural:
25. Consider a Poncelet rotation of polygon $A_{1} A_{2} \ldots A_{2 k+1}$. Does $A_{k+1}$ projectively depend on the touching point of side $A_{1} A_{2 k+1}$ with the incircle?
Now we do not know an elementary proof of this assertion.
## 3 Conics
Definition 3.1 Conic. An image of a circle in a projective map of the plane is called a non-degenerated conic. A degenerated conic is a pair of lines (possibly coinciding).
You may use without proof that any non-degenerated conic is a circle, an ellipse, a parabola, or a hyperbola; and that any conic is the set of points $(x, y)$ satisfying to an equation $a x^{2}+b y^{2}+c x y+d x+$ $e y+f=0$, where $a, b, c, d, e, f$ are real constants.
The following assertion may be used without proof:
Theorem 3.1 There exists a unique conic passing through five point such that any three of them are not collinear.
Definition 3.2 Cross-ratio of four points on the conic Let $A, B, C, D$ lie on a conic $\Omega$, and $S$ be an arbitrary point of this conic. The cross-ratio $(A, B ;, C, D)$ on the conic equals to the cross-ratio $(S A, S B ; S C, S D$.
26. Prove that this definition is correct.
27. Three points $A, B$ and $C$ of a conic $\mathcal{C}$ are collinear. Prove that $\mathcal{C}$ is degenerated.
28. a) Lines $a_{t}$ and $b_{t}$ rotate projectively around points $A$ and $B$ respectively. Prove that the common point of $a_{t}$ and $b_{t}$ move projectively along some conic (probably degenerated) b) Points $A_{t}$ and $B_{t}$ move projectively along lines $a$ and $b$ respectively. Prove that the line $A_{t} B_{t}$ envelops some conic (or passes through a fixed point).
When the conic in the previous problem is degenerated?
29. Points $A_{t}$ and $B_{t}$ move along two lines with fixed velocities. Which is the conic envelopped by the line joining these points?
Conics often appear as locus of points.
30. A circle and a line meeting at points $A$ and $B$ are give. Find the locus of points such that the tangent to the circle equals the distant to the line.
31. Similar isosceles triangles $B A_{1} C, C B_{1} A, A C_{1} B$ have the sides of triangle $A B C$ as the bases (All triangles are constructed inside $A B C$, or all triangles are constructed outside it). Prove that the lines $A A_{1}, B B_{1}, C C_{1}$ concur and find the locus of their common points.
The using of conics may be useful in problems not coherent directly wit them.
32. Let points $P$ и $Q$ be isogonally conjugated with respect to an acute-angled scalene triangle $A B C$. A point $W$ is the midpoint of arc $B A C$ of the circumcircle of $A B C$. The lines $W P$ and $W Q$ meet the
circumcircle of $A B C$ for the second time at points $X$ and $Y$ respectively. The lines passing through $P$ and $Q$ and parallel to $A W$ meet $A B, A C$ at points $P_{B}, P_{C}, Q_{B}, Q_{C}$. Prove that $X, Y, P_{B}, P_{C}, Q_{B}, Q_{C}$ are concyclic.
33. The incircle of scalene triangle $A B C$ touches $B C, C A$, and $A B$ at points $A_{1}, B_{1}$, and $C_{1}$ respectively. Three flies creep along the lines $A A_{1}, B B_{1}$, and $C C_{1}$ with fixed velocities in such a way that in some moment they were at points $A, B$, and $C$, and in some other time they were at $A_{1}, B_{1}$, and $C_{1}$ respectively. In some time all three flies were collinear on a line $p_{1}$, and in some other time they were collinear on a line $p_{2}$. Prove that $p_{1} \perp p_{2}$.
34. Points $P, Q$ lie on the sides $A D, C D$ of quadrilateral $A B C D$ in such a way that $\angle A B P=\angle C B Q$. Denote as $S$ the common point of $C P$ and $A Q$. Prove that $\angle P B S=\angle Q B D$.
35. Let $A B C$ be a triangle with incenter $I$. A line $\ell$ meets $A I, B I$, and $C I$ respectively at points $D, E, F$ distinct from $A, B, C$, and $I$. The perpendicular bisectors to segments $A D, B E$, and $C F$ form a triangle $\Delta$ with incircle $\omega$. Prove that $\omega$ and the circumcircle of $A B C$ are tangent.
## 4 Polynomial dependences
Definition 4.1 Projective plane. Projective plane is the set of all lines passing through a fixed point $O$ of the space, these lines are called points of the projective plane.
We can choose on any line in $\mathbb{R}^{3}$ passing through the origin an arbitrary point $(x, y, z)$ distinct from the origin. So all non-zero triplets $[x: y: z]$ code points of the projective plane and are called homogenious coordinates. This definition naturally corresponds to the previous one: if a plane $\alpha \mid z=1$ not containing $O$ is fixed in the space, then each point $A$ of this plane corresponds to the line $O A$, and each infinite point corresponds to the line passing through $O$, parallel to $\alpha$, and having the same direction. The triple $[0: 0: 0]$ does not define any point of the projectibe plane.
Note that the axes $O X$ and $O Y$ in the space are parallel to $\alpha$, and the axis $O Z$ is perpendicular this plane. Projecting $O X$ and $O Y$ to $\alpha$ we obtain the standart coordinates in this plane. The point $(x, y)$ of $\alpha$ has homogenious coordinates $[x: y: 1]$. The infinite points of the projective plane have homogenious coordinates $[x: y: 0]$.
Let $p, q$ be two different points of the projective plane. Consider the plane passing through $p, q$, and $O$ - it is defined by an equation $a x+b y+c z=0$, where a triple $[a: b: c]$ is defined up to a factor. We call $[a: b: c]$ the homogenious coordinates of a line. The triple $[0: 0: 0$ does not define any line.
It is clear that the point $\left[x_{0}: y_{0}: z_{0}\right.$ lies on the line $\left[a: b: c\right.$ if and only if $a x_{0}+b y_{0}+c z_{0}=0$. From this we obtain by substitution the following
Lemma 4.1 The coordinates of the line passing through the points $\left[x_{1}: y_{1}: z_{1}\right]$ and $\left[x_{2}: y_{2}: z_{2}\right]$ may be defined as
$$
\left[y_{1} z_{2}-z_{1} y_{2}: z_{1} x_{2}-x_{1} z_{2}: x_{1} y_{2}-y_{1} x_{2}\right]
$$
and the common point of the lines $\left[a_{1}: b_{1}: c_{1}\right]$ and $\left[a_{2}: b_{2}: c_{2}\right]$ has the coordinates
$$
\left[b_{1} c_{2}-c_{1} b_{2}: c_{1} a_{2}-a_{1} c_{2}: a_{1} b_{2}-b_{1} a_{2}\right]
$$
Similarly we may define the homogenious coordinates on the projective line coding its points by the pairs $[x: y]$ defined up to factor. Since the time is a point of the projective line we can consider it as a pair $\left[t_{1}: t_{2}\right]$.
Earlier we considered the maps from the projective line to the projective plane (and concretely to lines or conics on it) conserving the cross-ratios. Now we add the polynomial maps i.e. the functions $\mathcal{F}: \mathcal{R} \mathcal{P}^{1} \rightarrow$ $\mathcal{R} \mathcal{P}^{2}$ mapping a pair $\left[t_{1}: t_{2}\right]$ to a triple of polynomials $P\left(t_{1}, t_{2}\right), Q\left(t_{1}, t_{2}\right), R\left(t_{1}, t_{2}\right)$ corresponding to he homogenious coordinates of a point of the projective plane.
Clearly the polynomials $P, Q, R$ have to satisfy to several conditions. If this map cotrrectly transforms the points of projective line to the points of the projective plane the polynomials have to be homogenious $\left(t_{1}^{2}+2 t_{1} t_{2}\right.$ satisfies and $t_{1}^{3}+2 t_{1} t_{2}$ does not satisfy) and they degrees have to be equal. Also all three polynomials may not equal zero, thus we suppose that they are relatively prime.
Definition 4.2 Power law Say that the degree of dependence of point $X$ on the time equals $k$, if the homogenious coordinates of $X$ may be defined as $\left[P_{1}\left(t_{1}, t_{2}\right): P_{2}\left(t_{1}, t_{2}\right): P_{3}\left(t_{1}, t_{2}\right)\right]$, where $P_{i}$ are relatively prime polynomials of degree $k$. The degrees of dependence of lines are defined similarly.
It is easy to see that this definition is correct.
Lemma 4.2 Addition of degrees. Let points $X$ and $Y$ move with degrees $a$ and $b$ respectively. Then the degree of the line $X Y$ is not greater than $a+b$.
Denote the coordinates of $X$ and $Y$ as $\left[x_{1}: x_{2}: x_{3}\right]$ and $\left[y_{1}: y_{2}: y_{3}\right]$ respectively. Then the coordinates of $X Y$ are $y_{1} z_{2}-y_{2} z_{1}: z_{1} x_{2}-z_{2} x_{1}: x_{1} y_{2}-x_{2} y_{1}$. If the degrees of polynomials $x_{i}$ are not greater than $a$, and the degrees of $y_{i}$ are not greater than $b$, then the degrees of these polynomials are not greater than $a+b$. This estimation is not precise only if the obtained polynomials are not relatively prime.
Lemma 4.3 Let a point $X$ of degree 1 move one the line not passing trough a point $S$. Then the degree of line $S X$ also equals 1.
By the addition degree lemma the degree of $S X$ is not greater than $0+1=1$. Since the line is not constant this estimation is precise.
In the following chapter we prove the general theorem: if the point $X$ moves on an arbitrary trajectory, then the degree of $S X$ equals to the degree of $X$.
Theorem 4.1 The degree of point $X$ projectively moving on a line equals 1. The degree of the line rotating projectively is also 1.
Firstly demonstrate that any projective map between two lines may be presented as a composition of central projections and translations.
Let the map transform the points $A_{1}, B_{1}, C_{1}$ of line $\ell_{1}$ to the points $A_{2}, B_{2}, C_{2}$ of $\ell_{2}$. If $\ell_{1} \| \ell_{2}$ project $\ell_{1}$ to any line not parallel to it. Let a translation of $\ell_{1}$ mapping $A_{1}$ to $A_{2}$ map $B_{1}$ and $C_{1}$ to $B^{\prime}$ and $C^{\prime}$ respectively. Denote as $S$ the common point of $B^{\prime} B_{2}$ and $C^{\prime} C_{2}$ (may be infinite). The projection centered at $S$ realize the required homography.
Clearly the translation does not change the degree of dependence. Hence we have to prove that that the projection does not change the degree. his follows from the previous lemma.
This theorem yields that if $Y$ projectively depend on $X$ and both points move on the lines then their degrees are equal. This is also correct for two rotating lines. Even more is true.
Theorem 4.2 A homography of lines transforms the homogenious coordinates of their points linearly.
This theorem may be used without proof.
Lemma 4.4 Redoubling of degree on a conic. $A$ point $X$ moving projectively along a circle (a conic) has degree 2.
Fix two points $A, B$ on a circle, then the lines $a=A X$ and $b=B X$ rotate projectively, therefore $X$ is the common point of two lines with degrees 1 , and my the addition of degrees lemma we obtain the required assertion. point.
Similarly the homography of a line (a pencil) to a circle may at most redouble the degree of any
Remark: Clearly the dual lemma is also correct: let a line $\ell$ rotates around a circle (a conic) in such a way that the degree of touching point equals $k$. Then the degree of $\ell$ is not grater than $2 k$
Theorem 4.3 To prove the coincidence of two points with degrees $k$ and $l$ it is sufficient to verify $k+l+1$ dispositions.
Not that the coincidence of points at time $\left[t_{1}: t_{2}\right]$ is equivalent to the equality of the ratios of polynomials on $t_{1}, t_{2}$ defining the coordinates of points:
$$
\left\{\begin{array}{l}
\frac{P_{x}\left(t_{1}, t_{2}\right)}{P_{y}\left(t_{1}, t_{2}\right)}=\frac{Q_{x}\left(t_{1}, t_{2}\right)}{Q_{y}\left(t_{1}, t_{2}\right)} \\
\frac{P_{x}\left(t_{1}, t_{2}\right)}{P_{z}\left(t_{1}, t_{2}\right)}=\frac{Q_{x}\left(t_{1}, t_{2}\right)}{Q_{z}\left(t_{1}, t_{2}\right)}
\end{array}\right.
$$
This is equivalent to:
$$
\left\{\begin{array}{l}
P_{x}\left(t_{1}, t_{2}\right) Q_{y}\left(t_{1}, t_{2}\right)=P_{y}\left(t_{1}, t_{2}\right) Q_{x}\left(t_{1}, t_{2}\right) \\
P_{x}\left(t_{1}, t_{2}\right) Q_{z}\left(t_{1}, t_{2}\right)=P_{z}\left(t_{1}, t_{2}\right) Q_{x}\left(t_{1}, t_{2}\right)
\end{array}\right.
$$
And to prove the identity of two homogemious polynomials with degrees not grater than $k+l$ it is sufficient to verify $k+l+1$ Not proportional dispositions. In fact if the polynomial equals zero at points $\left(x_{i}, y_{i}\right)$ where $y_{i} \neq 0$, then dividing $P\left(t_{1}, t_{2}\right)$ onto $t_{2}^{d}$ where $d$ is the degree of polynomial, we obtain a polynomial on $\frac{t_{1}}{t_{2}}$ with the number of roots greater than its degree, hence it is equal to zero. If several $y_{i}$ equal to zero, consider another time parameter $\left[\tau_{1}: \tau_{2}\right]$ such that the infinite time $\tau$ does not correspond to any $\left(x_{i}, y_{i}\right)$.
Similarly to prove that a line with degree $k$ passes through a point with degree $l$ it is sufficient to verify $k+l+1$ dispositions.
The lemmas formulated above are «basic» and allow to solve many hard problems. Consider an example:
Example 4.1 Points $P$ and $Q$ lie on the sideline $B C$ of scalene triangle $A B C$ in such a way that $B P=C Q$. Let $\omega$ be the incircle of the triangle, and $\omega_{A}$ be the excircle touching the side $B C$. Points $S$ and $T$ lie on $\omega$ and $\omega_{A}$ respectively in such a way that $P S$ touches $\omega$, and $Q T$ touches $\omega_{A}$. Let $A S$ and $A T$ meet $B C$ at points $X$ and $Y$ respectively. Probe that $B X=C Y$.
1. Choose a time and homogenious coordinates on the plane. Move $P$ projectively along $B C$. Denote the degree of a point as $d$.
2. Point $S$ projectively depend on $P$ and moves along the circle, hence ny the redoubling of degree lemma $d(S) \leq 2$.
3. By the addition of degrees lemma $d(A S) \leq d(S)+d(A) \leq 2+0=2$
4. By the same lemma $d(X) \leq d(B C)+d(A S) \leq 0+2=2$. Similarly $d(Y) \leq 2$, because $P$ and $Q$ are symmetric with respect to the midpoint $M$ of $B C$, i.e $Q$ projectively depend on $P$.
5. Let $Y^{\prime}=\mathcal{S}_{M}(X)$. Clearly $d\left(Y^{\prime}\right)=d(X) \leq 2$, thus we have to prove two points with degrees 2 coincide, for this it is sufficient to verify $2+2+1=5$ dispositions.
6. The dispositions $P=C, P=B, P=A_{1}$ - the touching point of the incircle, $P=M, P=\infty$ are clear.
Solving problems it is useful to understand how degenerated cases habe to be considered. Let in some moment we have to draw the line through two coinciding points: By the formula the coordinates of this line are $[0: 0: 0]$, which yields that all following polynomials equal zero. Hence the required assertion is always correct in degenereted dispositions. But we have to be accurate, consider the following example: a point $X$ moves projectively on a conic, and $S$ is a fixed point of this conic. Let we want to prove that several point $Y$ of degree 1 lies always on $S X$. We have to interpretate correctly the line $S X$, when $S=X$. If we estimate the degree of $S X$ by the degrees addition lemma its degree do not exceed 2 , and if $X=S$ $S X$ is the «zero line», this give us one disposition (because we did not estimate the degree precisely) and we have to find three other disposition. But if we consider $S X$ as a line rotating projectively in the pencil of $S$, then we initially have to verify three dispositions but when $X=S$ the line has to be considered as the tangent and not the «zero line».
Now try to solve several exercises.
36. Points $A$ and $B$ move along two lines with constant velocities. Prove that the direction of line $A B$ is projective.
37. Three points move projectively. How many positions have to be verified for prove that they are always collinear? The same question for three lines rotating projectively.
38. P oints $X$ and $Y$ move with degrees $a$ and $b$ respectively. Prove that the degree of the midpoint of segment is not greater than $a+b$.
39. Let $P$ be a fixed point of a circle $\omega$, and $A$ move with degree $a$. Let $B$ be such point that the arc $P B$ is twice greater than the $\operatorname{arc} P A$ (we count the length of an arc counterclockwise). Prove that the degree of $B$ is at most $2 b$.
40. P olar transformations do not change the degree of dependence.
Now solve several problems.
41. Let $A B C D E F$ be a convex cyclic hexagon. The intersection of triangles $A C E$ and $B D F$ is a hexagon. Prove that its main diagonals concur.
42. Two perpendicular lines meet at the orthocenter of an acute-angled triangle. Prove that the midpoints of segments carving by these lines on the sidelines are collinear.
43. A triangle $A B C$ and three collinear points $P, Q, R$ are given. The lines $A P, B P$, and $C P$ meet the circumcircle of $A B C$ at points $A^{\prime}, B^{\prime}$, and $C^{\prime}$ respectively. The lines $A^{\prime} Q, B^{\prime} Q, C^{\prime} Q$ meet the circumcircle at points $A^{\prime \prime}, B^{\prime \prime}$, and $C^{\prime \prime}$. The lines $A^{\prime \prime} R, B^{\prime \prime} R, C^{\prime \prime} R$ meet this circle at points $A^{\prime \prime \prime}, B^{\prime \prime \prime}$, and $C^{\prime \prime \prime}$. Prove that the lines $A A^{\prime \prime \prime}, B B^{\prime \prime \prime}, C C^{\prime \prime \prime}$ concur at a point lying on the line passing through $P, Q$, and $R$.
44. A point $X$ is marked on the circumcircle of triangle $A B C$. The lines $B X$ and $C X$ meet the altitudes $C C_{1}$ and $B B_{1}$ at points $P$ and $Q$ respectively. Prove that the midpoint of segment $P Q$ lies on $B_{1} C_{1}$.
45. Let $A H_{A}$ be the altitude of an acute-angled triangle $A B C$, and $O$ be the center of its circumcircle $\Omega$. The lines $\ell_{A}, \ell_{B}, \ell_{C}$ touch $\Omega$ at $A, B, C$ respectively. Let $S$ be the orthocenter of triangle formed by $\ell_{A}, \ell_{B}, \ell_{C}$. Prove that the lines $O H_{A}$ and $S H_{A}$ are symmetric with respect to $B C$.
46. The excircle of triangle $A B C$ centered at $I_{A}$ touches the side $B C$ at $A_{1}$ and touches the sidelines $A B, A C$ at $C_{1}, B_{1}$ respectively. A point $P$ of the line $I_{A} C_{1}$ is such that $A P \perp B I_{A}$. A point $Q$ of the line $I_{A} B_{1}$ is such that $A Q \perp C I_{A}$. Prove that $P, Q, A_{1}$ are collinear.
47. Let $X$ lie inside a scalene triangle $A B C$ on its Euler line; and $O$ be the circumcenter of $A B C$. The lines $A X, B X, C X$ meet the opposite sides of $A B C$ at $A_{1}, B_{1}, C_{1}$ respectively. Prove that the circle $\left(A O A_{1}\right),\left(B O B_{1}\right),\left(C O C_{1}\right)$ are coaxial.
48. A triangle $A B C$ with orthocenter $H$ is given. Points $A_{1}, B_{1}$, and $C_{1}$ kie on the circumcircle of the triangle in such a way that the lines $A A_{1}, B B_{1}$, and $C C_{1}$ concur. Denote as $A_{2}, B_{2}$, and $C_{2}$ the reflections of $A_{1}, B_{1}$, and $C_{1}$ about the midpoints of the corresponding sidelines. Prove that $A_{2}, B_{2}, C_{2}$, and $H$ are concyclic.
49. A triangle $A B C$ with orthocenter $H$ is given. Points $A_{1}, B_{1}$, and $C_{1}$ kie on the circumcircle of the triangle in such a way that the lines $A A_{1}, B B_{1}$, and $C C_{1}$ concur. Denote as $A_{2}, B_{2}$, and $C_{2}$ the reflections of $A_{1}, B_{1}$, and $C_{1}$ about the corresponding sidelines. Prove that $A_{2}, B_{2}, C_{2}$, and $H$ are concyclic.
50. The Turner theorem. Let points $P, Q$ be inverse with respect to the circumcircle of triangle $A B C, P_{C}$ be the reflection of $P$ about $A B$, and $P_{C} Q$ meet $A B$ at point $C^{\prime}$. Points $A^{\prime}, B^{\prime}$ are defined similarly. Prove that $A^{\prime}, B^{\prime}, C^{\prime}$ are collinear.
51. Let $A B C D$ be a convex quadrilateral with $\angle B=\angle D$. Prove that the midpoint of $B D$ lies on a common internal tangent to the incircles of triangles $A B C$ and $A C D$.
52. A triangle $A B C$ is given. Denote as $A_{1}$ the common point of the medial line parallel to $B C$ and the line joining the feet of altitudes to $A B, A C$. Points $B_{1}, C_{1}$ are defined similarly. Prove that the orthocenter of triangle $A_{1} B_{1} C_{1}$ lies on the Euler line of triangle $A B C$.
53. Let $A B C D$ be a cyclic quadrilateral, $\omega$ be its circumcircle, and $P$ be the common point of its diagonals. Denote as $I_{A}, I_{B}, I_{C}$, and $I_{D}$ the incenters of triangles $A P B, B P C, C P D$, and $D P A$ respectively. Let $S_{A}, S_{B}, S_{C}$, and $S_{D}$ be the midpoints of «minor» $\operatorname{arcs} A B, B C, C A$, and $D A$ of $\omega$. Prove that the lines $I_{A} S_{A}, I_{B} S_{B}, I_{C} S_{C}$, and $I_{D} S_{D}$ concur.
## 5 Polynomial movement +
This chapter contains three subjects which be learned independently: generalizations of the theorems proved above, the relativity principle, and the combination of the movement with the Poncelet theorem.
To prove the advanced theorems concerning the polynomial movement we need to use the instruments most powerful than the real numbers. All definitions and theorems of the previous chapter are working for the complex numbers. Instead the real projective line $\mathbb{R P}^{1}$ we will consider $\mathbb{C P}^{1}-$ the set of lines in $\mathbb{C}^{2}$ passing through $(0,0)$. Homogenious coordinates on the complex projective line are pairs $\left[z_{1}: z_{2}\right]$ defined up to factor. The complex projective plane $\mathbb{C P}^{2}$, the homogenius coordinates on it, the pencils, etc. are defined similarly. Not that many results of previous chapters are correct for the complex numbers too. The unit comples circle is the set of solutions in $\mathbb{C}^{2}$ of the equation $x^{2}+y^{2}=1$, or the set of points $[x: y: z]$ in $\mathbb{C P}^{2}$ such that $x^{2}+y^{2}=z^{2}$. Any non-degenerated conic may be obtained from a circle by a projective map (transforming the lines to the lines) and may be defined as the set of roots of an irreducible homogenious polynomial $P(x, y, z)=0$.
54. Prove that a complex line and a non-degenerated complex conic have at most two common points. Do they intersect obviously?
The complex numbers are better than the real ones because any polynomial $f(x)$ distinct from a constant has a root (this may be used without proof). Furthermore any line of degree $n$ on $\mathbb{C P}^{2}$ (i.e the set of roots of a homogenious polynomial of degree $n$ ) meets a line of degree $m$ at $m n$ points counting their multiplicity. This assertion is the general Bezou theorem. In this chapter we will use only the following patial case of the Besou theorem:
55. a) Prove that a homogenious non-constant polynomial $f\left(t_{1}, t_{2}\right)$ may be divided to some linear polynomial $a t_{1}+b t_{2} \mathbf{b}$ ). Prove that any homogenious polynomial $f\left(t_{1}, t_{2}\right)$ may be uniquely (up to permutations and common factors) presented as a product $p_{1}^{\alpha_{1}} p_{2}^{\alpha_{2}} \ldots p_{n}^{\alpha_{n}}$, where $p_{i}$ are homogenious linear polynomials.
Clearly the last problem is similar to the main theorem of the arithmetic.
Lemma 5.1 (The complex degrees addition lemma). Let points $X$ and $Y$ move with degrees $a$ and $b$ respectively. Then the degree of the line $X Y$ is at most $a+b$, and if $X$ and $Y$ are distinct at any time, this estimation is precise.
Denote the homogenious coordinates of points as $\left[x_{1}: x_{2}: x_{3}\right]$ and $\left[y_{1}: y_{2}: y_{3}\right]$ respectively, Then the line $X Y$ is defined by th coordinates $\left[y_{1} z_{2}-z_{1} y_{2}: z_{1} x_{2}-x_{1} z_{2}: x_{1} y_{2}-y_{1} x_{2}\right]$. If the degree of $x_{i}$ is at most $a$, and the degree of $y_{i}$ is at most $b$ the degrees of these polynomials do not exceed $a+b$. If this estimation is not precise then three obtained polynomials have a common divisor $d\left(t_{1}, t_{2}\right)$. Choose any linear divisor $p=a t_{1}+b t_{2}$ of $d$. Then at time $[b:-a] X Y$ is «zero line»m i.e $X$ and $Y$ coinside, contradiction.
Lemma 5.2 Let a point $X$ of degree a move on a line not passing through a point $S$. Then the degree of line $S X$ also equals a.
Follows from the previous assertion.
Summing we obtain:
Theorem 5.1 Let $f$ be a homography of lines or pencils. Then $f$ does not change the degree of dependence.
Now we have to learn homographies using conics. We proved that the transfering of a point to a conic redoubles its degree. The inverse assertion is also true.
Theorem 5.2 (Throwing off conic). Let a point $X$ of degree $k$ move on a conic, and $S$ be fixed on it. Then the degree of line $S X$ equals $\frac{k}{2}$, in partial $k$ is even.
You will prove this theorem solving the problems (we suppose that all object are complex).
56. A conic $\omega$ is given. Choose an arbitrary point $S$ NOT on this conic and a line $\ell$ not passing through $S$. Let $f$ be the projection of the conic from $S$ to $\ell$. Prove that any point of the line has two prototypes (excepting may be a finite set of points).
57. A point $X$ moves polynomially on a conic $\omega$. Choose an arbitrary point $T \mathrm{ON} \omega$ and consider the projection of $\omega$ from $S$ to the line $z=0$. Let this projection map $X$ to $X^{\prime}$. So we defined the map $g$ from the line of tome to the line $z=0$. a) Prove that there exists a number $k$ such that each point of line $z=0$ (excepting may be a finite set) has exactly $k$ prototypes in the map $g$. b) Prove that $X$ coincide with almost any point of $\omega$ exactly $k$ times.
58. Denote as $X^{\prime \prime}$ the image of $X$ in the map $f$ from the problem preceding the previous one. Count by two methods how many times $X^{\prime \prime}$ coincides with a general point of $\ell$ and obtain from this the theorem.
To start the next subject consider an example. Suppose that the vertices of any triangle move with degrees $a, b$, and $c$ respectively. How can we estimate the degree of its circumcenter? It is not hard to see that the degree of the mid point of the segment joining two points with degrees $a$ and $b$ is at most $a+b$. The direction of the line passing through the points with degrees aandbalsohasthedegreenotexceedinga + $\mathrm{b}$, hencethedegreeoftheperpendicularbisectorisnotgreaterthan $(\mathrm{a}+\mathrm{b})+(\mathrm{a}+\mathrm{b})=2(\mathrm{a}+\mathrm{b})$ (bytheadditiondegre + c).Meetingthesebisector sandusingtheadditionprincipleweobtainthatthedegreeofthecircumcenterisnotgreat $+2 \mathrm{~b}+2 \mathrm{c}$.
This result sets thinking: the obtained estimation is not symmetric. Probably this means that the estimation is not precise. Furthemore we can suppose that the real estimation equals $2(a+b+c)$. The following instrument allows to prove this hypothesis.
Definition 5.1 Polynomial substitution Suppose that subsets $P_{1}, P_{2}, \ldots P_{n}$ of projective plane are fixed, and for any points $p_{1} \in P_{1}, p_{2} \in P_{2} \ldots, p_{n} \in P_{n}$ a point $\left[p_{0_{x}}: p_{0_{y}}: p_{0_{z}}\right]=\mathcal{R}\left(p_{1}, p_{2}, \ldots, p_{n}\right)$ is defined such that there exist homogenious polynomials $R_{x}, R_{y}, R_{z}$ on $3 n$ variables, $p_{0_{x}}=R_{x}\left(p_{1_{x}}, p_{1_{y}} \ldots p_{n_{z}}\right)$, and $p_{0_{y}}, p_{0_{z}}$ are defined similarly. Call such $\mathcal{R}$ a polynomial substitution on $n$ points.
Definition 5.2 Relative degree of dependence. Let $\mathcal{R}$ be some polynomial substitution on two points, points $q_{1}, q_{2}$ move with several degrees, and $q=\mathcal{R}\left(q_{1}, q_{2}\right)$. Let $r_{1}$ be the minimal natural number such that for any time $\left[\tau_{1}: \tau_{2}\right]$ the degree of $\mathcal{R}\left(q_{1}, q_{2}\left(\tau_{1}, \tau_{2}\right)\right)$ is not greater than $r_{1}$ (the degree may be different for different $\left[\tau_{1}: \tau_{2}\right]$ ). Call such $r_{1}$ the relative degree of dependence of $q$ on $q_{1}$.
The relative degrees of point depending on more than two points are defined similarly. Fix all $q_{i}$ except one and see the degree of dependence of $q$ on the remaining point.
59. Relativity principle. Let points $q_{1}, q_{2} \ldots, q_{n}$ move polynomially, and the relative degrees of $q=\mathcal{R}\left(q_{1 t}, q_{2_{t}}, \ldots q_{n_{t}}\right)$ equal $r_{1}, r_{2}, \ldots r_{n}$ for some polynomial substitution $\mathcal{R}$. Then the degree of $q$ is not grater than $r_{1}+r_{2}+\ldots+r_{n}$.
60. Let the verices of a triangle move with degrees $a, b$, and $c$ respectively. Then the degrees of the circumcenter and the orthocenter of this triangle are njt greater than $2(a+b+c)$.
61. Let points $A$ and $B$ projectively move along a conic $\omega$. a) Prove that the line $A B$ envelops some conic $\gamma$ (or passes through a fixed point). b) Prove that $\omega$ and $\gamma$ touch at two points of $\mathbb{C P}^{\not \_\not}$ (i.e. the coordinates of touching points may be complex).
62. Let points $A$ and $B$ projectively move along a conic $\omega$. Prove that the common point of tangents to the conic at $A$ and $B$ moves projectively along some conic or a line.
63. Let $A$ move along a conic with degree $2 a$, and $B$ move along an arbitrary trajectory with degree $b$. Denote as $C$ the second common point of line $A B$ with the conic. Then the degree of $C$ is not greater than $2 a+2 b$.
64. Four points move with degrees $a, b, c$, and $d$. Then to prove that they are concyclic $2(a+b+c+d)+1$ disposition are sufficient.
It may be useful to apply the Poncelet theorem not only to the circumcircle and the incircle of a triangle, but also to its circumcircle and excircle. There are six degenerated disposition in this configuration: parametrizing a tangent $\ell$ to the excircle we can consider two dispositions when $\ell$ touches the circumcircle and the excircle, two disposition when it passes through the common points of these circles, and two disposition when the triangle is isosceles.
65. Denote the circumcircle and the excircle of triangle $A B C$ as $\Omega$ and $\omega$ respectively. Let they meet at points $X$ and $Y$. The common external tangents to $\Omega$ and $\omega$ touch $\Omega$ at points $U$ and $V$. Prove that the tangents to $\omega$ at $X$ and $Y$ pass through $U$ and $V$.
66. Let $\omega$ de the excircle of triangle $A B C$ touching the side $B C$. The common external tangents to $\omega$ and $(A B C)$ touch $(A B C)$ at point $X$ and $Y$. Prove that the line $X Y$ passes through the feet of bisectors from $B$ and $C$.
67. The excircle $\omega$ of triangle $A B C$ touches the side $B C$ at point $A_{1}$ and meets the circumcircle of $A B C$ at points $X$ and $Y$. The tangents to $\omega$ at $X$ and $Y$ meet at point $Z$. Denote as $S$ the midpoint of arc $B A C$. Prove that the lines $S A_{1}$ and $A Z$ meet on the circumcircle of $A B C$.
68. The excircle $\omega$ of triangle $A B C$ touches the side $B C$ at point $A_{1}$ and meets the circumcircle of $A B C$ at points $X$ and $Y$. The lines $A_{1} X$ and $A_{1} Y$ meet the perpendicular bisector to the altitude from $A$ at points $U, V$. Prove that $A U A_{1} V$ is a parallelogram.
69. Let a line $\ell$ of degree $d$ rotate around a point $O$. Fix two different vectors $\overrightarrow{O A}$ and $\overrightarrow{O B}$ with origin $O$. Prove that $\frac{\sin \angle(\ell, \overrightarrow{O A})}{\sin \angle(\ell, \overrightarrow{O B})}=P\left(t_{1}, t_{2}\right)$, where $P\left(t_{1}, t_{2}\right)$ is a homogenious polynomials with degree
$d$.
70. Let $\omega$ be the circumcircle of triangle, and $\Omega_{A}$ be the excircle touching the side $B C$. Denote the common points of these circles as $X$ and $Y$. Let $P$ and $Q$ be the projections of $A$ to the tangents at $X$ and $Y$ to $\Omega_{A}$. The tangent to the circle $(A P X)$ at $P$ meets the tangent to the circle $(A Q Y)$ at $Q$ at point $R$. Prove that $A R \perp B C$.
## Дискретная задача Дидоны и проблема Бобылева
Ивлев Ф.А., Исмаилов А., Канель-Белов А.Я., Иванов-Погодаев И.А., Golafshan М.
## А Предварительные задачи
Вводная серия относится к классическим фактам о задаче Дидоны о минимальной поверхности. Пусть $\Phi$ - фигура минимального периметра с заданной площадью $S$.
A1 Докажите, что любая хорда фигуры $\Phi$, делящая пополам ее периметр, делит пополам ее площадь и наоборот.
A2 Докажите, что хорда из предыдущей задач перпендикулярна границе фигуры $\Phi$.
A3 (Задача Дидоны) Докажите, что фигура $\Phi$ максимальной площади заданного периметра есть круг.
A4 Укажите кривую минимальной длины, которая делит равносторонний треугольник на две равные по площади части.
A5 Докажите, что среди всех $n$-угольников заданного периметра максимальную площадь имеет правильный.
A6 Докажите, что среди всех $n$-угольников с заданными длинами сторон максимальную площадь имеет описаный.
A7 Докажите, что площадь $n$-угольника из предыдущей задачи не зависит от порядка его сторон.
A8 (Многомерная задача Дидоны) Докажите, что тело заданной площади поверхности с максимальным объемом есть шар.
A9 Докажите, что тетраэдр заданной площади поверхности с максимальным объемом правильный.
A10 Докажите что параллелепипед заданной площади поверхности с максимальным объемом есть куб.
A11 (Задача для исследования) Что можно сказать про многомерные обобщения?
## В Дискретная задача Дидоны
В1 Клеточный многоугольник с клетками двух цветов назовем хорошим, если в нем ровно четверть клеток - черная. Верно ли, что любой хороший квадрат $12 \times 12$ можно разрезать на 9 хороших многоугольников?
Рассмотрим бесконечную клетчатую плоскость (квадратную, треугольную, шестиугольную). Пусть отмечено несколько клеток. Отмеченная клетка называется граничной, если она граничит хотя бы с одной неотмеченной клеткой. Пусть имеется $n$ отмеченных граничных клеток. Дискретной задачей Дидоны будем называть вопрос максимизации числа отмеченных клеток.
B2 Решите дискретную задачу Дидоны для квадратной решетки (граничные - это по стороне).
B3 Решите дискретную задачу Дидоны для квадратной решетки (граничные - это по стороне или углу).
B4 Решите дискретную задачу Дидоны для правильной шестиугольной решетки (граничние - это по стороне).
B5 Решите дискретную задачу Дидоны для правильной треугольной решетки (граничние - это по стороне).
B6 Решите дискретную задачу Дидоны для кубической решетки (граничные - это по грани).
B7 * Придумайте многомерное обобщение для кубической решетки.
Задачу также можно поставить немного по-другому: например, если множество граничных клеток расположено внутри некоторого квадрата, и мы не учитываем клетки на границе этого квадрата. Множество граничных клеток, не попадающих на границу квадрата (или, аналогично, куба) будем называть свободной поверхностью.
B8 Решите дискретную задачу Дидоны для свободной поверхности участка квадратной решетки в квадрате $k \times k$ (то есть, в свободной поверхности $n$ клеток, граничные клетки - по стороне, мы минимизируем число отмеченных клеток в квадрате $k \times k$ ).
В9 * Решите дискретную задачу Дидоны для свободной поверхности участка кубической решетки в кубе $k \times k \times k$ (граничные - это по грани).
B10 * Решите дискретную задачу Дидоны для свободной поверхности участка многомерной кубической решетки в кубе $k \times k \times \cdots \times k$ (граничные - это по грани).
В11 (Открытый вопрос) Выведите из предыдущего пункта проблему куба (см. цикл С).
## С Свойства многомерья
Начнем с известной задачи В.И.Арнольда
C1 Какой процент объёма занимает мякоть в стомерном арбузе диаметра 1 метр, если толщина корки - 1 см?
Продолжим тему.
$\mathrm{C} 2 \mathrm{~K}$ чему стремится объем $n$-мерного шара радиуса 2022 при $n \rightarrow \infty$ ?
С3 Докажите, что в $n$-мерный куб при $n \gg 1$ можно поместить главное здание МГУ.
Расстоянием между множествами $A$ и $B$ будем называть такое максимальное число $d$, что любое расстояние между точками $x$ и $y$, где $x \in A, y \in B$, будет не меньше $d$.
C4 Докажите, что сечением многомерного куба плоскостью может быть многоугольник, сколь угодно близкий к окружности (то есть, расстояние между многоугольником и окружностью может быть сделано меньше любого заранее заданного числа $\delta>0$ ).
C5 В многомерном единичном кубе дано множество $M$ объема 0,99 и точка $A$. Докажите, что расстояние от $M$ до $A$ может быть сколь угодно большим.
Тем не менее, есть убежденность в положительном решении следующей проблемы.
Проблема многомерного куба. В $n$-мерном кубе единичного объема расположены два множества $M_{1}$ и $M_{2}$ объема $\varepsilon$ каждое. Тогда расстояние между ними не превосходит некоторой константы $F(\varepsilon)$.
Замечание. Константа $F$ зависит только от $\varepsilon$, но не от размерности.
Решение проблемы многомерного куба нам не известно, атака на нее есть одна из целей этого проекта.
C6 **. Проблема многомерного шара. В $n$-мерном шаре единичного объема расположено два множества $M_{1}$ и $M_{2}$ объема $\varepsilon$ каждое. Тогда расстояние между ними не превосходит некоторой константы $G(\varepsilon)$.
C7 Решите проблему многомерного шара для выпуклых тел.
C8 * (Проблема Бобылева.) Решите проблему многомерного куба для выпуклых тел.
Замечание. Можно поставить аналогичные вопросы для симплексов (многомерных тетраэдров) и многомерных октаэдров. Однако такие вопросы в данный момент представляются нам преждевременными, по крайней мере до решения проблемы куба.
Функциональный анализ изучает многомерные и бесконечномерные пространства. Проблемы куба и шара несомненно прольют дополнительный свет на соответствующую проблематику, в особенности относящуюся к теории меры и к пониманию устройства бесконечномерных пространств.
При чем тут задача Дидоны?
Открытый вопрос. Проблема минимальной свободной поверхности в кубе. Опишем следующий класс подмножеств точек единичного $k$-мерного куба. Выбираем любое натуральное число $n \leqslant k$, выбираем $n$ координат и включаем в подмножество все точки, у которых выбранные координаты неотрицательны, причём сумма их квадратов не превосходит некоторого $C$, а остальные координаты - числа от 0 до 1 . Включаем в класс все множества, получающиеся всевозможными такими выборами. Тогда подмножество куба, имеющее фиксированный объём и минимальную свободную поверхность (то есть, площадь той части поверхности, которая не выходит на границу куба, минимальна), достигается на одном из множеств этого класса.
С9 Проблема минимальной свободной поверхности в шаре. Множество объема $V$ в шаре единичного объема $B$ с минимальной площадью той части поверхности, которая не выходит на границу шара, устроено как $B \cap B^{\prime}$, где $B^{\prime}$ - шар, чья поверхность перпендикулярна поверхности $B$.
C10 Выведите проблему куба из проблемы минимальной свободной поверхности в кубе.
Указание. Если $M$ - некоторое множество точек внутри куба, то при $\delta \rightarrow 0$ объем $\delta$-окрестности $M$ асимптотически равен $\operatorname{Vol}(M)+\delta S(M)$.
Замечание. Асимптотика $\operatorname{Vol}(M)+\delta S(M)$ лежит в основе определения площади поверхности по Минковскому.
C11 Выведите проблему шара из проблемы минимальной свободной поверхности в шаре.
О дискретной задаче Дидоны. Хотя проблема минимальной свободной поверхности в кубе представляется очень трудной, дискретный аналог этой задачи (достаточный для решения задачи куба), как нам кажется, поддается решению.
## D Нумерация клеток куба
D1 В квадрате $8 \times 8$ расставлены числа от 1 до 64 . Докажите, что найдутся две соседние по стороне клетки, числа в которых отличаются не менее чем на 5 .
D2 Клетки доски $N \times N$ занумерованы числами от 1 до $N^{2}$. Докажите, что найдутся две соседние по стороне клетки, разность номеров которых не меньше $N$.
D3 Тот же вопрос для куба со стороной $N$.
D4 Тот же вопрос для многомерного куба.
## Е Кривая Пеано
E1 Докажите, что существует непрерывное отображение отрезка на квадрат.
E2 Докажите, что бесконечно много пар точек обязаны склеиться.
E3 Верно ли, что бесконечно много троек точек обязаны склеиться?
E4 Верно ли, что бесконечно много четверок точек обязаны склеиться?
E5 Исследуйте многомерные обобщения.
E6 Исследуйте системы склеенных точек и расстояния между их прообразами.
## Свойства многомерья (ещё).
C12.
a) Найдите минимально возможное $n$ такое, что равносторонний треугольник со стороной 100 содержится в $n$-мерном единичном кубе $[0,1]^{n}$. Каков максимальный периметр треугольника, лежащего в 11 -мерном единичном кубе?
б) Тот же вопрос для квадрата со стороной 10.
в) (Задача на исследование) Аналогичный вопрос для вложения $k$-мерного куба в единичный $n$-мерный куб.
г) Каков максимальный радиус трехмерного шара, лежащего внутри четырехмерного единичного куба?
## Дискретная изопериметрическая задача
Будем рассматривать решётку $\mathcal{L}$, заданную набором попарно не параллельных векторов $\vec{v}_{1}, \ldots, \vec{v}_{n}$ :
$$
\mathcal{L}=\left\{\sum_{i=1}^{n} a_{i} \vec{v}_{i} \mid a_{i} \in \mathbb{Z}\right\}
$$
Пусть $\mathcal{L}$ обладает таким свойством: в любом конечном куске плоскости (например, квадрате) содержится конечное количество точек $\mathcal{L}$.
Для фигуры $F$ определим понятия: $S(F)$ - площадь $F, P(F)$ - периметр $F, c(F)$ количество точек $\mathcal{L}$ в $F$.
F1. Докажите, что существуют такие константы $\alpha, \beta>0$, что для любой фигуры $F$
$$
c(F) \in[\alpha S(F)-\beta P(F) ; \alpha S(F)+\beta P(F)]
$$
Графом нашей решётки назовём $G=(\mathcal{L}, E)$, где ребром соединена любая пара точек из $\mathcal{L}$, отличающихся на $\vec{v}_{i}$ для некоторого $i$. Для любого $S \subset \mathbb{R}^{2}$ (хотя можно считать $S \subset \mathcal{L}$ ) определим $\partial S$, как подмножество рёбер из $E$, один конец которых входит в $S$, а другой - нет. Нас будет интересовать минимизация
$$
\frac{|\partial S|}{\sqrt{|S \cap \mathcal{L}|}}
$$
при фиксированном $|S \cap \mathcal{L}|$.
Пусть нам дан простой многоугольник $M$. Для $t>0$ за $t M$ обозначим результат применения гомотетии с коэффициентом $t$ к нашему многоугольнику, т. е. раздуваем его в $t$ раз. Введём величину
F2. Находим $\gamma(M)$ :
$$
\gamma(M)=\lim _{t \rightarrow \infty} \frac{|\partial(t M)|}{\sqrt{c(t M)}}
$$
F2.1 Докажите, что $c(t M) \sim \alpha t^{2} S(M)$ при $t \rightarrow \infty$.
F2.2 Для работы с $\partial(t M)$ рассмотрим одну из сторон $M$, заданную вектором $\vec{a}$. Докажите, что количество ребёр из $\partial(t M)$, проходящих через эту сторону, входит в $[\alpha f(\vec{a}) t-C ; \alpha f(\vec{a}) t+C]$, где $C$ - константа, не зависящая от $t$, и
$$
f(\vec{a})=\sum_{i=1}^{n} \mid\left[\vec{a}, \vec{v}_{i}\right]
$$
F2.3 Совмещая предыдущие два пункта, выведите формулу для $\gamma(M)$.
В общем случае задача минимизации
$$
\frac{|\partial S|}{\sqrt{|S \cap \mathcal{L}|}}
$$
кажется трудной. В частности, неясно, как показать, что множеству $S$ стоит иметь какойто хороший вид, например быть выпуклым. Однако за счёт полученной нами формулы для $\gamma(M)$ мы можем что-то понять про «оптимальный» $M$.
F3. Докажите, что «оптимальный» $M$ является выпуклым, иначе говоря, для минимизации $\gamma(M)$ будем лишь рассматривать выпуклые многоугольники.
Отметим, что здесь могут быть актуальны теорема Жордана, т. е. что понятие «внутренности» простого многоугольника вообще имеет смысл, и критерий выпуклости многоугольника: все внутренние углы не превосходят $180^{\circ}$.
F4. Докажите, что «оптимальный» $M$ имеет стороны, параллельные векторам $\vec{v}_{i}$, задающим решётку, т. е. будем лишь рассматривать многоугольники такого типа.
F5. Докажите, что для простого многоугольника $M$, образованного последовательностью векторов $\vec{a}_{i}$ (его стороны)
$$
S(M)=\left|\frac{1}{2} \sum_{i0$ ).
C5 In a multidimensional unit cube, a set $M$ of volume 0,99 and a point $A$ are given. Prove that the distance from $M$ to $A$ can be arbitrarily large.
Nevertheless, there is a conviction in a positive solution to the following problem.
The problem of a multidimensional cube. In the $n$-dimensional cube of a unit volume there are two sets $M_{1}$ and $M_{2}$ of volume $\varepsilon$ each. Then the distance between them does not exceed some constant $F(\varepsilon)$.
Remark. The constant $F$ depends only on $\varepsilon$, but not on the dimension.
We do not know the solution to the multidimensional cube problem, an attack on it is one of the goals of this project.
C6 **. The multidimensional ball problem. In an $n$-dimensional sphere of a unit volume there are two sets $M_{1}$ and $M_{2}$ of volume $\varepsilon$ each. Then the distance between them does not exceed some constant $G(\varepsilon)$.
C7 Solve the multidimensional ball problem for convex bodies.
C8 * (Bobylev's problem.) Solve the multidimensional cube problem for convex bodies.
Remark. Similar questions can be posed for simplices (multidimensional tetrahedra) and multidimensional octahedra. However, such questions seem premature to us at the moment, at least until the cube problem is solved.
Functional analysis studies multidimensional and infinite-dimensional spaces. The problems of the cube and the sphere will undoubtedly shed additional
light on the relevant problems, especially those related to the theory of measure and to the understanding of the structure of infinite-dimensional spaces.
What does Dido's task have to do with it?
An open question. The problem of the minimum free surface in a cube.
We describe the following class of subsets of points of a unit $k$-dimensional cube. We choose any natural number $n \leqslant k$, choose $n$ coordinates and include in the subset all points whose selected coordinates are non - negative, and the sum of their squares does not exceed some $C$, and the remaining coordinates are numbers from 0 to 1 . We include in the class all the sets obtained by all possible such choices. Then a subset of the cube having a fixed volume and a minimal free surface (that is, the area of the part of the surface that does not extend to the boundary of the cube is minimal) is achieved on one of the sets of this class.
C9 The problem of the minimum free surface in the ball. The set of volume $V$ in a ball of unit volume $B$ with the minimum area of the part of the surface that does not extend to the boundary of the ball is arranged as $B \cap B^{\prime}$, where $B^{\prime}$ is a ball whose surface is perpendicular to the surface of $B$.
C10 Derive the cube problem from the problem of the minimum free surface in the cube.
Hint. If $M$ is some set of points inside the cube, then for $\delta \rightarrow 0$ the volume The $\delta$-neighborhood of $M$ is asymptotically equal to $\operatorname{Vol}(M)+$ $\delta S(M)$.
Remark. The asymptotics of $\operatorname{Vol}(M)+\delta S(M)$ is the basis for determining the surface area by Minkowski.
C11 Derive the ball problem from the problem of the minimum free surface in the ball.
On the discrete Dido problem. Although the problem of the minimal free surface in a cube seems to be very difficult, a discrete analogue of this problem (sufficient to solve the cube problem), it seems to us, can be solved.
## D) Numbering of cube cells
D1 The numbers from 1 to 64 are placed in an $8 \times 8$ square. Prove that there are two adjacent cells on the side, the numbers in which differ by at least 5.
D2 In the cell, the boards $N \times N$ are numbered with numbers from 1 to $N^{2}$. Prove that there are two adjacent cells on the side whose number difference is not less than $N$.
D3 The same question for a cube with side $N$.
D4 The same question for a multidimensional cube.
## E) The Peano curve
E1 Prove that there is a continuous mapping of a segment to a square.
E2 Prove that infinitely many pairs of points are bound to stick together.
E3 Is it true that infinitely many triples of points are bound to stick together?
E4 Is it true that infinitely many fours of points are bound to stick together?
E5 Explore multidimensional generalizations.
E6 Explore the systems of glued points and the distances between their prototypes.
[^0]: ${ }^{1}$ We need here the following more general version of the constructor than that in problem 3.6.
Let $\mathcal{G}=\left\langle G, h, \star s g_{0}\right\rangle, \mathcal{G}^{\prime}=\left\langle G^{\prime}, h^{\prime}, g^{\prime}\right\rangle$ be winning games. Let $A$ be a vertex of graph $G^{\prime}$, and $h^{\prime}(A)=s$. Let $\langle\widetilde{G}, \widetilde{h}, \widetilde{g}\rangle$ be the winning game obtained by the substitution of game $\mathcal{G}$ on the place of vertex $A$ to game $\mathcal{G}^{\prime}$ (as in problem 3.5). By the rule of the substitution for all substituting vertices $v$
$$
\widetilde{h}(v)=h(v) h^{\prime}(A)=s \cdot h(v), \quad \widetilde{g}(v)=g(v) g^{\prime}(A)=s g_{0} \cdot g^{\prime}(A)
$$
Consider new functions $h^{*}, g^{*}$ on graph $\widetilde{G}$, which differ from $\widetilde{h}, \widetilde{g}$ only by the values in substituting vertices $v$, and this difference is the cancellation by $s$ :
[^1]: ${ }^{1}$ Это означает, что строки и столбы последовательно занумерованы числами от 1 до $n$ и выбрана перестановка $f$ этих чисел; строки и столбцы переставляются так, что $i, j$-ый элемент становится $f(i), f(j)$-ым.
[^2]: ${ }^{2}$ Строго говоря, трансверсальность легко определить только для PL (кусочно-линейного) изображения графа. PL кривые на торе легко определить, если рассматривать тор как склейку многоугольника. Тогда $P L$ кривая на торе - это семейство ломаных на многоугольнике, удовлетворяющее некоторым условиям (выпишите эти условия!). Похожим образом любая поверхность $S$ может быть получена склейкой многоугольника. (Для ленты Мёбиуса и бутылки Клейна см. [Sk20, §2.1]; для сферы с ручками см. визуализацию в https://www.youtube.com/watch?v=G1yyfPShgqw и в https://www.youtube.com/watch?v=U5N5mg3MePM.) Это позволяет определить PL кривые на $S$. Тогда изображение графа на $S$ называется $P L$ (кусочно-линейным), если изображение любого ребра этого графа PL. Другая формализация приведена в [Sk20, §4, §5].
[^3]: ${ }^{1}$ This means that the rows and columns are numbered by $1, \ldots, n$ (where $n=3,4$ ) and the permutation of the set $[n]$ is applied both to the rows and to the columns.
[^4]: ${ }^{2}$ Strictly speaking, the transversality is only easy to define for PL (piecewise-linear) graph drawings. PL curves on the torus can be easily defined by regarding the torus as obtained from a rectangle by gluing. A $P L$ curve on the torus is then a family of polygonal lines in the rectangle satisfying certain conditions (work out these conditions!). In a similar way, other surfaces $S$ can be obtained from plane polygons by gluing. (For Möbius band and Klein bottle see [Sk20, §2.1]; for spheres with handles see visualization in https: //www.youtube.com/watch?v=G1yyfPShgqw and in https://www.youtube.com/watch?v=U5N5mg3MePM.) This allows one to define PL curves on $S$. A graph drawing on $S$ is called $P L$ if the drawing of every edge is PL. Another formalizations are given in $[\mathrm{Sk} 20, \S 4, \S 5]$.