# Об изогональном сопряжении, точках Микеля, прямых Гаусса и
др.
Н.Белухов, А.Заславский, П.Кожевников
## 1 Вводные задачи
1. Изогональное сопряжение. Дан треугольник $A B C$ и точка $P$.
а)Докажите, что прямые, симметричные $A P, B P, C P$ относительно биссектрис соответствующих углов, пересекаются в одной точке или параллельны. Полученная точка $P^{\prime}$ называется изогонально сопряженной $P$ относительно треугольника $A B C$.
b)Докажите, что $P^{\prime}$ - бесконечно удаленная точка (т.е. три соответствующие прямые параллельны) тогда и только тогда, когда $P$ лежит на описанной окружности треугольника $A B C$.
с)Найдите образ при изогональном сопряжении окружности, проходящей через две вершины треугольника.
d) ${ }^{1}$ Докажите, что проекции $P$ и $P^{\prime}$ на стороны $A B C$ лежат на одной окружности. Как звучит это утверждение, если $P^{\prime}$ бесконечно удалена?
е)Если $X, X^{\prime}$ и $Y, Y^{\prime}$ - две пары изогонально сопряженных точек, то $X Y \cap X^{\prime} Y^{\prime}$ и $X Y^{\prime} \cap X^{\prime} Y$ изогонально сопряжены.
Дан четырехугольник $A B C D$ и точка $P$.
f)Докажите, что, если три из четырех прямых, симметричных $A P, B P, C P, D P$ относительно биссектрис соответствующих углов, пересекаются в одной точке, то четвертая также проходит через эту точку.
g)Докажите, что точка, изогонально сопряженная $P$ относительно четырехугольника, существует тогда и только тогда, когда проекции $P$ на стороны лежат на одной окружности (причем на этой окружности лежат и проекции изогонально сопряженной точки).
Коникой, вписанной в многоугольник, называется коника, касающаяся всех прямых, содержащих стороны многоугольника.
h)Докажите, что фокусы вписанной в треугольник коники изогонально сопряжены.
i)Докажите, что фокус любой параболы, касающейся прямых $A B, B C, C A$, лежит на описанной окружности треугольника $A B C$.
2. Точка Микеля. Дан четырехугольник $A B C D$. Прямые $A B$ и $C D$ пересекаются в точке $E, A D$ и $B C$ - в точке $F$.
а)Докажите, что в обозначениях предыдущей задачи окружности, описанные около треугольников $A B F, C D F, A D E$ и $C D E$, пересекаются в одной точке $M$ (точке Микеля четверки прямых $A B, B C, C D, D A)$.
b)Докажите, что $M$ - центр поворотной гомотетии, переводящей отрезок $B E$ в $F D$ (или $D E$ в $F B$, и т.д.)
с)Два таракана $B$ и $C$ ползут с постоянными скоростями по двум прямым, пересекающимся в точке $A$. Докажите, что окружности $A B C$ проходят через фиксированную точку, а прямые $B C$ касаются фиксированной параболы.
d)(IMO2005) Дан выпуклый четырехугольник $A B C D$, стороны $B C$ и $A D$ которого равны, но не параллельны. Пусть $E$ и $F$ - внутренние точки отрезков $B C$ и $A D$ соответственно такие, что $B E=D F$. Прямые $A C$ и $B D$ пересекаются в точке $P$, прямые $B D$ и $E F$ пересекаются в точке $Q$, прямые $E F$ и $A C$ пересекаются в точке $R$. Рассмотрим треугольники $P Q R$, получаемые для всех таких точек $E$ и $F$. Докажите, что окружности, описанные около всех этих треугольников, имеют общую точку, отличную от $P$.
е)Выясните связь точки Микеля со вписанными кониками.[^0]
f)Докажите, что проекции точки Микеля на стороны четырехугольника лежат на одной прямой, перпендикулярной прямой Гаусса. Как связана эта прямая со вписанной в четырехугольник параболой?
3. Прямая Гаусса. Дан четырехугольник $A B C D$. Прямые $A B$ и $C D$ пересекаются в точке $E, A D$ и $B C$ - в точке $F$.
а)Докажите, что середины отрезков $A C, B D$ и $E F$ лежат на одной прямой (прямой Гаусса четырехугольника $A B C D$, или четверки прямых $A B, B C, C D, D A$ ).
b)Докажите, что центры окружностей, проходящих через проекции пары изогональных относительно четырехугольника точек, лежат на прямой Гаусса четырехугольника.
с)Докажите, что точка Микеля четырехугольника изогонально сопряжена бесконечно удаленной точке его прямой Гаусса.
d)Докажите, что центры коник, вписанных в четырехугольник, лежат на его прямой Гаусса.
е)(Всероссийская олимпиада 2009) На сторонах $A B$ и $B C$ параллелограмма $A B C D$ выбраны точки $A_{1}$ и $C_{1}$ соответственно. Отрезки $A C_{1}$ и $C A_{1}$ пересекаются в точке $P$. Описанные окружности треугольников $A A_{1} P$ и $C C_{1} P$ вторично пересекаются в точке $Q$, лежащей внутри треугольника $A C D$. Докажите, что $\angle P D A=$ $\angle Q B A$.
## 2 Три Микеля для квартетов.
В задачах этого раздела рассматривается следующая конструкция и используются следующие обозначения. $A, B, C, D$ - четыре точки общего положения. $X$ - точка Микеля прямых $A B, A C, B D, C D, Y$ - точка Микеля прямых $A B, A D, B C, C D, Z$ - точка Микеля прямых $B C, A C, B D, A D . P_{X}=A D \cap B C, P_{Y}=A C \cap B D, P_{Z}=A B \cap C D$. $K_{X}, L_{X}$ - середины $B C$ и $A D, K_{Y}, L_{Y}$ - середины $A C$ и $B D, K_{Z}, L_{Z}$ - середины $A B$ и $C D . \Gamma_{X}, \Gamma_{Y}, \Gamma_{Z}$ - прямые (Гаусса для соответствующих четверок прямых) $K_{X} L_{X}, K_{Y} L_{Y}$, $K_{Z} L_{Z}$.
4. Докажите, что прямые $A X, B Y, C Z$ пересекаются в одной точке $D^{\prime}$ или параллельны. Аналогично определяются точки $A^{\prime}, B^{\prime}, C^{\prime}$ как центры перспективы треугольника $X Y Z$ с треугольниками $D C B, C D A, B A D$.
5. Докажите, что $A^{\prime}, B^{\prime}, C^{\prime}, D^{\prime}$ - точки, изогонально сопряженные $A, B, C, D$ относительно треугольника $X Y Z$.
6. Докажите, что $X, Y, Z$ - точки Микеля для четверки точек $A^{\prime}, B^{\prime}, C^{\prime}, D^{\prime}$.
7. Докажите, что прямые $A A^{\prime}, B B^{\prime}, C C^{\prime}$ и $D D^{\prime}$ параллельны.
8. Докажите, что прямые $A D, A^{\prime} D^{\prime}$ и $Y Z$ пересекаются в одной точке (и аналогичные пересечения).
9.
а)Докажите, что точки $X, Z, P_{Y}, K_{Y}, L_{Y}$ лежат на одной окружности - $\omega_{Y}$. Окружности $\omega_{X}, \omega_{Z}$ определяются аналогично.
b)Докажите, что окружности $\omega_{X}, \omega_{Y}, \omega_{Z}$ проходят через одну точку $T$ (или совпадают).
c)Докажите, что прямые $X P_{X}, Y P_{Y}, Z P_{Z}$ проходят через $T$.
## 3 Квартеты для трех Микелей.
Дан треугольник $X Y Z$. Определим преобразование $\psi_{X}$, как композицию симметрии относительно биссектрисы угла $X$ и инверсии с центром $X$ и таким радиусом $R$, что $R^{2}=X Y \cdot X Z$. Аналогично определим $\psi_{Y}, \psi_{Z}$.
10. Докажите, что
a) $\psi_{X}(Y)=Z, \psi_{X}(Z)=Y$
b) $\psi_{X}^{2}$ - тождественное преобразование;
с) Композиция $\psi_{X}, \psi_{Y}$ и $\psi_{Z}$ - тождественное преобразование.
Пусть $D$ - произвольная точка, $A=\psi_{X}(D), B=\psi_{Y}(D), C=\psi_{Z}(D)$.
11. Докажите, что $\triangle X D Z \sim \triangle X Y A$ и $\triangle X D Y \sim \triangle X Z A$.
12. Докажите, что каждое из преобразований $\psi_{X}, \psi_{Y}, \psi_{Z}$ переводит набор из четырех точек $A, B, C, D$ в себя.
Будем называть набор из четырех (не обязательно различных) точек $A, B, C, D$ квартетом. Из последней задачи следует, что вся плоскость может быть разбита на квартеты.
13. Докажите, что четверка точек, изогонально сопряженных квартету - квартет.
14. Найдите квартеты, содержащие
а)центр $I$ вписанной окружности треугольника $X, Y, Z$;
b)центр $O$ его описанной окружности.
c)Найдите неподвижные точки преобразования $\psi_{Z}$ и соответствующие квартеты.
15.
a)Докажите, что $X$ - точка Микеля прямых $A B, A C, B D, C D$.
b)Докажите обратное утверждение: если $X, Y, Z$ - точки Микеля, определяемые точками $A, B, C, D$, то $A, B, C, D$ образуют квартет.
16. Докажите, что каждое из преобразований $\psi_{X}, \psi_{Y}, \psi_{Z}$ коммутирует с изогональным сопряжением относительно треугольника $X Y Z$.
17. Пусть точки $A, B, C, D$ образуют квартет относительно треугольника $X Y Z, A^{\prime}, B^{\prime}, C^{\prime}, D^{\prime}$ изогонально сопряжены им. Тогда существуют вписанные в треугольник коники с фокусами $A$ и $A^{\prime}, B$ и $B^{\prime}, C$ и $C^{\prime}$, $D$ и $D^{\prime}$.
а)Докажите, что эти коники гомотетичны друг другу.
b)Докажите, что середины шести отрезков, соединяющих центры этих коник, лежат на гомотетичной им конике, описанной около треугольника $X Y Z$.
18. Внутри треугольника $A B C$ лежат две изогонально сопряженные точки $M$ и $N$. Известно, что $A M$. $A N \cdot B C=B M \cdot B N \cdot A C=C M \cdot C N \cdot A B=k$.
а)Докажите, что середина $M N$ совпадает с центром тяжести треугольника.
b)Выразите $k$ через стороны треугольника.
## 4 Дополнительные задачи.
19.
а)Пусть $A, B, C, D$ - квартет, $A^{\prime}, B^{\prime}, C^{\prime}, D^{\prime}$ - изогонально сопряженный квартет; $P_{X}-$ точка пересечения $A D$ и $B C, P_{Y}-A C$ и $B D, P_{Z}-A B$ и $C D$. Точки $Q_{X}, Q_{Y}, Q_{Z}$ определяются аналогично по точкам $A^{\prime}, B^{\prime}, C^{\prime}, D^{\prime}$. Докажите, что прямые $P_{X} Q_{X}, P_{Y} Q_{Y}, P_{Z} Q_{Z}$ пересекаются в одной точке, лежащей на описанной окружности треугольника $X Y Z$ (из предыдущих обозначений).
b)В обозначениях предыдущего пункта докажите, что прямые $P_{X} Q_{Y}, P_{Y} Q_{X}$ и $X Y$ пересекаются в одной точке.
с)Обозначим точку, полученную в предыдущем пункте через $Z^{\prime}$. Докажите, что $Z Z^{\prime}$ параллельна $A A^{\prime}, B B^{\prime}, C C^{\prime}, D D^{\prime}$.
d)Пусть $D_{1}, D_{1}^{\prime}$ и $D_{2}, D_{2}^{\prime}$ - две пары изогонально сопряженных точек такие, что $D_{1} D_{1}^{\prime} \|$ $D_{2} D_{2}^{\prime}$. Докажите, что прямые $A_{1} A_{2}, B_{1} B_{2}, C_{1} C_{2}, D_{1} D_{2}$ пересекаются в одной точке ( $A_{1}, B_{1}, C_{1}, D_{1}$ и $A_{2}, B_{2}, C_{2}, D_{2}$ - квартеты).
20. Даны точки $A, B, C, D$. Известно, что треугольник $X Y Z$ перспективен каждому из треугольников $A B C, B C D, C D A, D A B$ (именно в таком порядке вершин). Точки $D^{\prime}, A^{\prime}, B^{\prime}, C^{\prime}-$ соответствующие центры перспективы. Докажите, что прямые $A A^{\prime}, B B^{\prime}, C C^{\prime}, D D^{\prime}$ пересекаются в одной точке.
## Об изогональном сопряжении, точках Микеля, прямых Гаусса и др. Решения
Н.Белухов, А.Заславский, П.Кожевников
## 1 Вводные задачи
1.
а)Непосредственно следует из теоремы Чевы в форме синусов.
b)Указание. Доказывается счетом углов.
c) Пусть $P$ лежит на этой окружности. Тогда угол $A P B$ имеет фиксированную величину. Но так как сумма углов четырехугольника $C A P B$ равна $360^{\circ}$, то фиксирована и сумма углов $C A P$ и $C B P$. А значит будет фиксирован угол $A P^{\prime} B$, то есть $P^{\prime}$ лежит на фиксированной окружности, проходящей через точки $A$ и $B$.
d)Обозначим через $P_{A}, P_{B}, P_{C}$ точки, симметричные $P$ относительно $B C, C A, A B$. Так как $P_{A} C=P C=P_{B} C$, серединный перпендикуляр к отрезку $P_{A} P_{B}$ проходит через $C$ и совпадает с биссектрисой угла $P_{A} C P_{B}$. Но легко видеть, что этой биссектрисой является луч $C P^{\prime}$. Следовательно, $P^{\prime}$ - центр описанной окружности треугольника $P_{A} P_{B} P_{C}$. Применив гомотетию с центром $P$ и коэффициентом $1 / 2$, получим, что середина отрезка $P P^{\prime}$ является центром окружности, проходящей через проекции $P$ на стороны. Аналогично получаем, что эта же точка является центром окружности, проходящей через проекции $P^{\prime}$, а поскольку она равноудалена от проекций $P$ и $P^{\prime}$ на любую прямую, обе окружности совпадают.
Если $P^{\prime}$ бесконечно удалена, получаем теорему Симсона: основания перпендикуляров, опущенных из точки на описанной окружности треугольника на стороны треугольника, лежат на одной прямой.
е) Обозначим точку пересечения $X Y^{\prime}$ с $Y X^{\prime}$ через $P$, а $X Y$ с $X^{\prime} Y^{\prime}$ через $Q$. Пусть $X A \cap$ $Y X^{\prime}=X_{A}$, а $Y^{\prime} A \cap Y X^{\prime}=Y_{A}$. Также обозначим через $Q_{A}$ точку пересечения $Q A$ с $Y X^{\prime}$. Покажем, что прямые $A Q$ и $A P$ симметричны относительно биссектрисы угла $A$. Тогда, так как аналогичное утверждение верно и для остальных углов треугольника, точки $P$ и $Q$ будут изогонально сопряжены, что и требуется.
Рассмотрим двойное отношение точек ( $Y, Q_{A}, X_{A}, X^{\prime}$ ). Спроецируем его из точки $Q$ на прямую $A X^{\prime}$. Тогда $Y$ перейдет в $X^{\prime}, Q_{A}$ перейдет в $A, X_{A}$ перейдет сама в себя, а $X^{\prime}$ перейдет в пересечение прямых $X^{\prime} Y^{\prime}$ и $X A$. Спроецируем получившееся двойное отношение из точки $Y^{\prime}$ на прямую $Y X^{\prime}$. Тогда $A$ перейдет в $Y_{A}, X$ перейдет в $P, X_{A}$ опять останется на месте, а образ точки $X^{\prime}$ вернется в $X^{\prime}$. То есть $\left(Y, Q_{A}, X_{A}, X^{\prime}\right)=\left(P, Y_{A}, X_{A}, X^{\prime}\right)=\left(Y_{A}, P, X^{\prime}, X_{A}\right)$. А значит при симметрии относительно биссектрисы угла $A$ прямая $A Q$ перейдет в прямую $A P$, ч.т.д.
Другое доказательство так же можно прочитать в статье А.Акопяна и А.Заславского "Разные взгляды на изогональное сопряжение" "Математическое просвещение" №11, 2007.
f) Обозначим эту точку пересечения $P^{\prime}$. Пусть прямые $A B$ и $C D$ пересекаются в точке $K$. Тогда для одного из треугольников $K B C$ и $K D A$ точка $P^{\prime}$ будет точкой пересечения хотя бы двух прямых симметричных прямым, соединяющим $P$ с вершинами треугольников. А следовательно она будет являться точкой, изогонально сопряженной точке $P$. А следовательно, она лежит на прямой, симметричной $K P$ относительно биссектрисы угла $K$. Но тогда и для оставшегося треугольника она является изогонально сопряженной точкой точке $P$, а следовательно, лежит на всех четырех прямых, симметричных $A P, B P, C P$ и $D P$ относительно биссектрис соответствующих углов
g)Пусть проекции точки $P$ лежат на одной окружности. Тогда, рассуждая, как в п.d),
получаем, что точка $P^{\prime}$, симметричная $P$ относительно центра окружности, изогонально сопряжена $P$. Обратное утверждение доказывается аналогично.
h)См. указанную выше статью.
i)Из обратной теоремы Симсона получаем, что нам надо доказать, что основания перпендикуляров из фокуса параболы на стороны треугольника лежат на одной прямой. А это верно, поскольку, если симметрично отразить фокус параболы относительно любой касательной к этой параболе, то он попадет на директрису этой параболы.
2.
а)Пусть описанный окружности треугольников $A B F$ и $C D F$ вторично пересекаются в точке $M$. Тогда из вписанности получаем $\angle(A M, M D)=\angle(A M, M F)+\angle(M F, M D)=$ $\angle(B A, B F)+\angle(C F, C D)=\angle(B A, C D)=\angle(A E, E D)$. То есть $M$ так же лежит и на описанной окружности треугольника $A D E$. Аналогично для оставшегося треугольника. b)Для этого нам достаточно доказать, что треугольники $M B E$ и $M F D$ подобны. Для этого покажем, что угол $M B E$ равен углу $M F D$. Тогда углы $M E B$ и $M D F$ равны по аналогичным соображениям, а следовательно, треугольники подобны по двух углам. $\angle(E B, B M)=\angle(C E, C M)=\angle(C D, C M)=\angle(F D, F M)$, ч.т.д.
с)Пусть вектор скорости таракана $B$ равен $\vec{b}$, а таракана $C-\vec{c}$. Тогда пусть $\overrightarrow{C C^{\prime}}=\vec{c}$, а $\overrightarrow{B B^{\prime}}=\vec{b}$. Тогда точка пересечения описанных окружностей треугольников $A B C$ и $A B^{\prime} C^{\prime}$ (точка $P$ ) будет центр поворотной гомотетии, переводящий отрезок $B B^{\prime}$ в $C C^{\prime}$. А следовательно и переводящий всю прямую $A B$ в прямую $A C$ с коэффициентом отношения скоростей тараканов. Следовательно, так как в какой-то момент эта гомотетия переводит одного таракана в другого, то она всегда будет делать это. А тогда для любого другого положения тараканов $B_{0}, C_{0}$ для четверки прямых $B_{0} B, B C, C C_{0}, C_{0} B_{0}$ точка $P$ всегда будет являться центром такой поворотной гомотетии, а следовательно и точкой Микеля. Заметив еще, что если отразить точку $P$ симметрично относительно прямой $B C$ то получится треугольник $P B P^{\prime}$, который так же при движении $B$ будет оставаться все время подобным самому себе (так как треугольник $P B C$ таковым остается). Следовательно всевозможные точки $P^{\prime}$ можно получить поворотной гомотетией из прямой $A B$ с центром в точке $P$. То есть это тоже будет прямая. Обозначим ее через $l$. Тогда прямая $C B$ все время будет касаться параболы с фокусом в точке $P$ и директрисой $l$. Точкой касания будет являться точка пересечения прямой $C B$ с перпендикуляром, восстановленным в точке $P^{\prime}$ к прямой $l$.
d)Пусть точка $E$ ползет с постоянной скоростью от точки $B$ к точке $C$, а точка $F$ от $D$ к $A$ с такой же скоростью. Тогда условие сохраняется в любой момент времени. С другой стороны посмотрим на то как будут двигаться точки $R$ и $Q$. Покажем, что точка $Q$ двигается с постоянной скоростью, тогда и точка $R$ по аналогичным причинам будет двигаться с постоянной скоростью, и задача сведется к предыдущей. Заметим, что углы $E Q B$ и $F Q D$ равны. Так же равны стороны $E B$ и $F D$ треугольников $E B Q$ и $F D Q$. А следовательно, по теореме синусов равны и их описанные окружности. А следовательно, так как углы $E B Q$ и $F D Q$ фиксированы, то постоянно и отношение $E Q$ к $Q F$. А следовательно, точка $Q$ так же двигается с постоянной скоростью, ч.т.д.
е)Точка Микеля - фокус вписанной в четырехугольник параболы.
f)Заметим, что если взять любые три из четырех проекций, то они лежат на одной прямой по теореме Симсона для соответственного треугольника. Следовательно, все они лежат на одной прямой. Несложно показать, что если сделать гомотетию с центром в точке, для которой применяется теорема Симсона, с коэффициентом 2 , то образ прямой Симсона для этой точки пройдет через ортоцентр треугольника. Поэтому если сделать гомотетию с центром в точке Микеля и коэффициентом 2 то полученная прямая пройдет через ортоцентры
всех треугольников. А если взять три окружности с диаметрами на диагоналях (всех трех) четырехугольника, то все ортоцентры будут иметь одинаковые степени относительно этих трех окружностей. А следовательно, это их общая радикальная ось и эта прямая перпендикулярна линии центров, то есть прямой Гаусса. Для вписанной параболы ее фокус будет лежать на описанных окружностях всех треугольников. Следовательно, точка Микеля и будет фокусом этой параболы. А основания перпендикуляров будут лежать на прямой, являющейся касательной к этой параболе в ее вершине. Прямая, проходящая через ортоцентры будет директрисой этой параболы. А значит прямая Гаусса будет параллельна главной оси параболы.
3.
а)Обозначим рассматриваемые середины через $M$ (середина $A C$ ), $N$ (середина $B D$ ) и $T$ (середина $E F$ ). Пусть середины треугольника $A B F$ - точки $F^{\prime}, A^{\prime}$ и $B^{\prime}$. Заметим, что $M$ лежит на $F^{\prime} B^{\prime}, N$ на $F^{\prime} A^{\prime}, T$ на $A^{\prime} B^{\prime}$. Если сделать гомотетии с центрами в вершинах треугольника $A B F$ и коэффициентом два получаем $\frac{\overrightarrow{F^{\prime} M}}{\overrightarrow{M B^{\prime}}}=\frac{\overrightarrow{B C}}{\overrightarrow{C F}}, \frac{\overrightarrow{B^{\prime} T}}{T A^{\prime}}=\frac{\overrightarrow{A E}}{\overrightarrow{E C}}, \frac{\overrightarrow{A^{\prime} N}}{N F^{\prime}}=\frac{\overrightarrow{F D}}{\overrightarrow{D A}}$. Перемножив эти три равенства получим справа по теореме Менелая -1 , а следовательно, по этой же теореме примененной в обратную сторону получаем требуемое.
b) Указание. Прямая Гаусса четырехугольника $A B C D$ является геометрическим местом точек, для которых $X S_{X A B}+S_{X C D}=S_{X B C}+S_{X D A}$ (площади ориентированные).
c) Следует из задачи $2 \mathrm{f}$.
d)Это переформулировка п.b)
е)Заметим, что $Q$ - точка Микеля для четверки прямых $A B, B C, C A_{1}, A C_{1}$. Ввиду $3 с$ достаточно понять, что $D P$ параллельна прямой Гаусса этой четверки прямых. Но это верно, так как при гомотетии с центром $B$ и коэффициентом $1 / 2 D P$ переходит в прямую Гаусса.
## 2 Три Микеля для квартетов.
4. Из задач 13,15 следует, что прямые $A X, B Y, C Z$ проходят через точку $D^{\prime}$, изогонально сопряженную $D$ относительно $X Y Z$.
5. Непосредственно следует из задачи 13.
6. Непосредственно следует из задач 13,15 .
7. По задаче 11 треугольники $X D Z$ и $X Y A, X D^{\prime} Z$ и $X Y A^{\prime}$ подобны. Следовательно, $X A$ : $X D^{\prime}=(X A: X Z)\left(X Z: X D^{\prime}\right)=(X Y: X D)\left(X A^{\prime}: X Y\right)=X A^{\prime}: X D$, что равносильно утверждению задачи.
8. Следует из предыдущей задачи и теоремы о трех центрах гомотетии, примененной к отрезкам $A A^{\prime}, D D^{\prime}$ и $B^{\prime} B$. Действительно, $Z$ является центром гомотетии, переводящей $A$ в $B^{\prime}$, а $A^{\prime}-$ в $B$ и т.д.
Другое решение можно сразу получить из утверждения задачи 1е).
9.
а)Точка $X$ является центром поворотной гомотетии, переводящей $C$ в $D$, а $A$ - в $B$. Поскольку $K_{Y}$ при этой гомотетии переходит в $L_{Y}$, угол $K_{Y} X L_{Y}$ равен углу между прямыми $A C$ и $B D$. Следовательно, $X$ лежит на окружности $P_{Y} K_{Y} L_{Y}$. Аналогично получаем, что $Z$ тоже лежит на этой окружности.
b)Так как $X$ лежит на окружности $A P_{Y} B, \angle X P_{Y} B=\angle X A B$. Аналогично, $\angle B P_{Y} Z=$ $\angle B C Z$. Из этих и четырех аналогичных равенств получаем, что $\angle X P_{Y} Z+\angle Z P_{X} Y+$ $\angle Y P_{Z} X=\pi$, откуда, очевидно, следует утверждение задачи.
c)Из решения задачи 15 видно, что $\psi_{X}\left(P_{Y}\right)=P_{Z}$ и т.п. Это означает, что точки $P_{X}$, $P_{Y}, P_{Z}$ входят в один квартет, т.е. $\psi_{X}\left(P_{X}\right)=\psi_{Y}\left(P_{Y}\right)=\psi_{Z}\left(P_{Z}\right)$. Утверждение задачи означает, что эта точка изогонально сопряжена $T$ или $\psi_{Z}(T)$ изогонально сопряжена $P_{Z}$. Заметим, что $\psi_{Z}$ переводит проходящие через $T$ окружности $Z X P_{Y}$ и $Z Y P_{X}$ в прямые $Y P_{X}$ и $X P_{Y}$, так что $\psi_{Z}(T)$ - точка пересечения этих прямых. Из равенств $\angle P_{X} Y X=\angle Z Y P_{Z}$, $\angle P_{Y} X Y=\angle Z X P_{Z}$ получаем искомое утверждение.
## 3 Квартеты для трех Микелей.
10. a)-b) Непосредственно следует из определения.
c) Каждое из преобразований $\psi_{X}, \psi_{Y}, \psi_{Z}$ является круговым (т.е. переводит любую окружность в окружность или прямую) и сохраняющим ориентацию. Значит, их композиция также обладает этими свойствами. Кроме того, из пп. а)-b) следует, что она оставляет неподвижными точки $X, Y, Z$. Но круговое преобразование, сохраняющее ориентацию, однозначно определяется образами трех точек. Заметим, что утверждение задачи верно независимо от порядка применения преобразований $\psi_{X}, \psi_{Y}, \psi_{Z}$. Следовательно, эти преобразования коммутируют друг с другом.
Второе решение. Положим $\psi_{X}(D)=A, \psi_{Y}(A)=C$. Достаточно доказать, что треугольники $Y D Z$ и $C X Z$ совмещаются поворотной гомотетией.
Имеем (многократно пользуемся подобиями $Y X D \sim A X Z$ и аналогичными): $\angle(Y D, D Z)=$ $\angle(Y D, D X)+\angle(D X, D Z)=\angle(A Z, Z X)+\angle(Y X, A Y)=\angle(A Z, Z Y)+\angle(Z Y, Z X)+\angle(Y X, A Y)=$ $\angle(X C, C Y)+\angle(Z Y, Z X)+\angle(C Y, Z Y)=\angle(C X, Z X) ; \frac{Y D}{C X}=\frac{Y D}{A Z} \cdot \frac{A Z}{C X}=\frac{D X}{X Z} \cdot \frac{Y A}{Y X}=\frac{D X}{X Z} \cdot \frac{D Z}{D X}=\frac{D Z}{X Z}$, что и требовалось.
11. По определению $\psi_{X} \angle Z X D=\angle A X Y$ и $X D \cdot X A=X Y \cdot X Z$, откуда сразу следует первое подобие. Второе доказывается аналогично
12. Из задачи 10 следует, что, например, $\psi_{X} \circ \psi_{Y}=\psi_{Z}^{-1}=\psi_{Z}$. Следовательно, $\psi_{Y}(A)=$ $\psi_{Y}\left(\psi_{X}(D)\right)=\psi_{Z}(D)=C$, т.е. $\psi_{Y}$ меняет местами точки $A$ и $C, B$ и $D$. Аналогично получаем, что $\psi_{X}$ меняет местами $A$ и $D, B$ и $C$, а $\psi_{Z}-A$ и $B, C$ и $D$.
13. Пусть $D^{\prime}, A^{\prime}$ - точки, изогонально сопряженные $D, A$. Тогда $A^{\prime}$ лежит на прямой $X D$, а $D^{\prime}$ - на прямой $X A$. Кроме того, $\angle X D^{\prime} Z=\pi-\angle Z X D^{\prime}-\angle D^{\prime} Z X=\pi-\angle D X Y-\angle Y Z D=$ $\angle Z D X+\angle X Y Z-\pi$. Но по задаче $11 \angle Z D X=\angle A Y X$, т.е. $\angle X D^{\prime} Z=\angle X Y A^{\prime}$. Значит, треугольники $X D^{\prime} Z$ и $X A^{\prime} Y$ подобны и $A^{\prime}=\psi_{X}\left(D^{\prime}\right)$.
14.
а)Центры вписанной и трех вневписанных окружностей треугольника $X Y Z$.
b)Точка $O$ и три точки, симметричные вершинам треугольника $X Y Z$ относительно противоположных сторон.
c)Из определения $\psi_{Z}$ следует, что его неподвижные точки должны лежать на биссектрисе угла $Z$ и окружности с центром $Z$ и радиусом $\sqrt{Z X \cdot Z Y}$. Таких точек две, обозначим их $U, V$. Из задачи 10 получаем, что $\psi_{X}(U)=\psi_{Y}\left(\psi_{Z}(U)\right)=\psi_{Y}(U)=\psi_{Z}\left(\psi_{X}(U)\right)$, т.е. $\psi_{X}(U)$ - тоже неподвижная точка $\psi_{Z}$. Очевидно, $\psi_{X}(U) \neq U$, следовательно, $\psi_{X}(U)=$ $\psi_{Y}(U)=V$, и искомый квартет (как для $U$, так и для $V$ ) состоит из дважды повторенных точек $U, V$. Более того, ясно, что точки, изогонально сопряженные $U, V$, также являются неподвижными точками $\psi_{Z}$. Значит, $U$ и $V$ изогонально сопряжены, а полученный квартет совпадает со своим сопряженным.
15.
a)Из определения $\psi_{X}$ и результата задачи 12 следует, что $X$ - центр поворотной гомотетии, переводящей $A$ в $B$, а $C$ в $D$. По задаче $2 \mathrm{~b}$ ) этот центр совпадает с точкой Микеля b)Так как $X$ - точка Микеля, то , например, треугольники $X B D$ и $X A C$ подобны. Пусть $P_{X}, P_{Y}, P_{Z}$ - точки пересечения $A D$ и $B C, A C$ и $B D, A B$ и $C D$. Тогда треугольники $X B P_{Y}$ и $X P_{Z} C$ подобны, следовательно, $X A \cdot X D=X B \cdot X C=X P_{Y} \cdot X P_{Z}=R_{X}^{2}$ и углы $A X D, B X C, P_{Y} X P_{Z}$ имеют общую биссектрису $l$. Композиция инверсии с центром $X$ и радиусом $R_{X}$ и симметрии относительно $l$ переводит треугольники $A D P_{Y}$ и $B C P_{Y}$ соответственно в $D A P_{Z}$ и $C B P_{Z}$. Следовательно, общая точка $Z$ описанных окружностей двух первых треугольников переходит в $Y$, т.е. эта композиция совпадает с $\psi_{X}$.
16. Из задачи 13 следует, что композиция $\psi_{X}$ и изогонального сопряжения, примененных в любом порядке, переводит точку $D$ в $A^{\prime}$.
## On isogonal conjugacy, Miquel points, (Newton-)Gauss lines, etc. N.Beluhov, A.Zaslavsky, P.Kozhevnikov
## 1 Introductory problems
1. Isogonal conjugacy. Given a triangle $A B C$ and a point $P$.
a)Prove that lines symmetric to $A P, B P, C P$ in the bisectors of corresponding angles are concurrent or parallel. The common point $P^{\prime}$ of these lines is called isogonal conjugate to $P$ with respect to $A B C$.
b)Prove that $P^{\prime}$ is a point at infinity (i.e. three corresponding lines are parallel) iff $P$ lies on the circumcircle of $A B C$.
c) Determine the image isogonal conjugacy of a circle passing through two of three points $A, B$, C.
d) ${ }^{1}$ Prove that all projections of $P$ and $P^{\prime}$ to the sidelines of $A B C$ are concyclic. Reformulate the statement above for the case when $P^{\prime}$ is a point at infinity.
e)For two pairs $X, X^{\prime}$ and $Y, Y^{\prime}$ of isogonal conjugate points, prove that $X Y \cap X^{\prime} Y^{\prime}$ and $X Y^{\prime} \cap X^{\prime} Y$ are isogonal conjugates.
Given a quadrilateral $A B C D$ and a point $P$.
f)Suppose that three of four lines symmetric to $A P, B P, C P, D P$ in the bisectors of corresponding angles are concurrent. Prove that all four lines are concurrent.
g)Prove that for a point $P$ there exists an isogonal conjugate $P^{\prime}$ iff projections of $P$ to the sidelines of $A B C D$ are concyclic (if $P^{\prime}$ exists, then all projections of $P$ and $P^{\prime}$ to the sidelines of $A B C D$ are concyclic).
A conic is said to be inscribed to a polygon if it touches all the sidelines of this polygon.
h)Prove that foci of a conic inscribed to a triangle are isogonal conjugates.
i)Prove that focus of a parabola inscribed to a triangle lies on its circumcircle.
2. Miquel point. Given a quadrilateral $A B C D$. Let $E=A B \cap C D, F=A D \cap B C$.
a)Prove that (in notation of the previous problem) circumcircles of triangles $A B F, C D F, A D E$, $C D E$ have a common point $M$ (Miquel point for a quadruple of lines $A B, B C, C D, D A$ ).
b)Prove that $M$ is a center of spiral similarity that takes segment $B E$ to $F D$ (or $D E$ to $F B$, etc.)
c) Two bugs $B$ and $C$ move, each at a constant speed, along two lines intersecting at $A$. Prove that all the circles $A B C$ have a common point, and * all the lines $B C$ touch fixed parabola.
d)(IMO2005) Let $A B C D$ be a convex quadrilateral with sides $B C$ and $A D$ equal in length and not parallel. Let $E$ and $F$ be interior points of the sides $B C$ and $A D$ such that $B E=D F$. The lines $A C$ and $B D$ meet at $P$, the lines $B D$ and $E F$ meet at $Q$, the lines $E F$ and $A C$ meet at $R$. Consider all triangles $P Q R$ as $E$ and $F$ vary. Prove that the circumcircles of these triangles have a common point other than $P$.
e)Establish a connection between Miquel point and inscribed conics.
f)Prove that the projections of Miquel point to the sidelines of a quadrilateral lie on a line perpendicular to Gauss line. Establish a connection between this line and a parabola inscribed to the quadrilateral.
3. Gauss line. Given a quadrilateral $A B C D$. Let $E=A B \cap C D, F=A D \cap B C$.
a)Prove that the midpoints of the segments $A C, B D, E F$ lie on a line (that is called Gauss line of $A B C D$, or Gauss line of quadruple of lines $A B, B C, C D, D A$ ).
b)Prove that the center of the circle passing through the projections of a pair of isogonal conjugates lies on Gauss line.
c)Prove that Miquel point is isogonal conjugate to the infinite point of Gauss line.
d)Prove that centers of conics inscribed to a quadrilateral lie on Gauss line.[^1]
e)(All-Russian Olympiad 2009) Let $A_{1}$ and $C_{1}$ be points on the sides $A B$ and $B C$ of parallelogram $A B C D$. Let $P=A C_{1} \cap C A_{1}$. Circumcircles of triangles $A A_{1} P$ and $C C_{1} P$ meet for the second time at point $Q$ lying inside triangle $A C D$. Prove that $\angle P D A=\angle Q B A$.
## 2 Three Miquels for a Quartet.
In this section we use the following notation. Let $A, B, C, D$ be four points such that no three of them are collinear. Let $X$ be Miquel point for the quadruple of lines $A B, A C, B D, C D$, let $Y$ be Miquel point for the quadruple of lines $A B, A D, B C, C D$, let $Z$ be Miquel point for the quadruple of lines $B C, A C, B D, A D$. We set $P_{X}=A D \cap B C, P_{Y}=A C \cap B D, P_{Z}=A B \cap C D$. Let $K_{X}$ and $L_{X}$ be midpoints of the segments $B C$ and $A D$ respectively, similarly, let $K_{Y}, L_{Y}$ be midpoints of $A C, B D$, let $K_{Z}, L_{Z}$ be midpoints of $A B, C D$. Let $\Gamma_{X}=K_{X} L_{X}, \Gamma_{Y}=K_{Y} L_{Y}$, $\Gamma_{Z}=K_{Z} L_{Z}$ be Gauss lines for the corresponding quadruples of lines.
4. Prove that $A X, B Y, C Z$ have a common point $D^{\prime}$, or parallel. Similarly define $A^{\prime}, B^{\prime}, C^{\prime}$.
5. Prove that $A^{\prime}, B^{\prime}, C^{\prime}, D^{\prime}$ are isogonal conjugates to $A, B, C, D$ with respect to triangle $X Y Z$.
6. Prove that $X, Y, Z$ are Miquel points for quadruples of lines joining $A^{\prime}, B^{\prime}, C^{\prime}, D^{\prime}$.
7. Probe that lines $A A^{\prime}, B B^{\prime}, C C^{\prime}, D D^{\prime}$ are parallel.
8. Prove that $A D, A^{\prime} D^{\prime}, Y Z$ are concurrent (find other analogous intersections).
9.
a)Prove that points $X, Z, P_{Y}, K_{Y}, K_{Y}$ lie on a certain circle $\omega_{Y}$. Similarly define circles $\omega_{X}$, $\omega_{Z}$.
b)Prove that $\omega_{X}, \omega_{Y}, \omega_{Z}$ have a common point $T$.
c) Prove that $X P_{X}, Y P_{Y}, Z P_{Z}$ meet at $T$.
## 3 Quartets for three Miquels.
Let $X Y Z$ be a triangle. Define a transformation $\psi_{X}$ as the symmetry in the bisector of angle $X$ followed by the inversion with center $X$ and radius $R=\sqrt{X Y \cdot X Z}$. Similarly define transformations $\psi_{Y}, \psi_{Z}$.
10. Prove that
a) $\psi_{X}(Y)=Z, \psi_{X}(Z)=Y$
b) $\psi_{X}^{2}$ is the identity transformation;
c)Product $\psi_{Z} \psi_{Y} \psi_{X}$ is the identity transformation.
Let $D$ be an arbitrary point, let $A=\psi_{X}(D), B=\psi_{Y}(D), C=\psi_{Z}(D)$.
11. Prove that $\triangle X D Z \sim \triangle X Y A$ and $\triangle X D Y \sim \triangle X Z A$.
12. Prove that each of the transformations $\psi_{X}, \psi_{Y}, \psi_{Z}$ takes the 4 -element set $\{A, B, C, D\}$ to itself. A 4 -element set of points $\{A, B, C, D\}$ defined as above is said to be a quartet. From the previous problem it follows that all the plane except $X, Y, Z$ is partitioned into quartets.
13. Prove that four isogonal conjugates to points of a quartet is a quartet.
14. Find all the quartets containing
a)the incenter $I$ of triangle $X Y Z$;
b)the circumcenter $O$ of triangle $X Y Z$.
c)Find the invariant points for $\psi_{Z}$, and corresponding quartets.
15.
a)Prove that $X$ is Miquel point for the quadruple of lines $A B, A C, B D, C D$.
b) Formulate similar statements for $Y, Z$.
c) Prove the converse: if $X, Y, Z$ are Miquel points defined by $A, B, C, D$, then $A, B, C, D$ us a quartet (for $X, Y, Z$ ).
16. Prove that each of transformations $\psi_{X}, \psi_{Y}, \psi_{Z}$ commutes with the isogonal conjugacy with respect to $X Y Z$.
17. Suppose $A, B, C, D$ be a quartet with respect to $X Y Z$, let $A^{\prime}, B^{\prime}, C^{\prime}, D^{\prime}$ be isogonal conjugates to $A, B$, $C, D$ respectively. Consider four conics having pairs of foci $A$ and $A^{\prime}, B$ and $B^{\prime}, C$ and $C^{\prime}, D$ and $D^{\prime}$.
a)Prove that these conics are homothetic to each other.
b)Prove that midpoints of six segments joining centers of these conics lie on a certain conic that is homothetic to them and passing through $X, Y, Z$.
18. Let $M, N$ be a pair of isogonal conjugates with respect to triangle $A B C$ lying inside $A B C$. It appears that $A M \cdot A N \cdot B C=B M \cdot B N \cdot A C=C M \cdot C N \cdot A B=k$.
a)Prove that the midpoint of $M N$ is the gravity center of $A, B, C$.
b)Find $k$ in terms of side lengths of $A B C$.
## 4 Additional problems.
19.
a)Let $A, B, C, D$ be a quartet, $A^{\prime}, B^{\prime}, C^{\prime}, D^{\prime}$ be conjugated quartet; let $P_{X}$ be intersection point of $A D$ and $B C, P_{Y}$ - of $A C$ and $B D, P_{Z}-$ of $A B$ and $C D$. Points $Q_{X}, Q_{Y}, Q_{Z}$ are defined similarly by points $A^{\prime}, B^{\prime}, C^{\prime}, D^{\prime}$. Prove that lines $P_{X} Q_{X}, P_{Y} Q_{Y}, P_{Z} Q_{Z}$ are concurent in the point, which lie on the circumcircle of triangle $X Y Z$ (notations as above).
b)In previous notations prove that lines $P_{X} Q_{Y}, P_{Y} Q_{X}$ and $X Y$ concur.
c) Let $Z^{\prime}$ be the point obtained in b). Prove that line $Z Z^{\prime}$ is parallel to $A A^{\prime}, B B^{\prime}, C C^{\prime}, D D^{\prime}$.
d) Let $D_{1}, D_{1}^{\prime}$ and $D_{2}, D_{2}^{\prime}$ be two pairs of isogonally conjugated points such that $D_{1} D_{1}^{\prime} \|$ $D_{2} D_{2}^{\prime}$. Prove that lines $A_{1} A_{2}, B_{1} B_{2}, C_{1} C_{2}, D_{1} D_{2}$ concur $\left(A_{1}, B_{1}, C_{1}, D_{1}\right.$ and $A_{2}, B_{2}, C_{2}, D_{2}$ are quartets).
20. Given points $A, B, C, D$. It is known that triangle $X Y Z$ is perspective to each of triangles $A B C, B C D, C D A, D A B$ (with indicated order of vertices). Points $D^{\prime}, A^{\prime}, B^{\prime}, C^{\prime}$ are respective centers of perspective. Prove that lines $A A^{\prime}, B B^{\prime}, C C^{\prime}, D D^{\prime}$ concur.
## On isogonal conjugacy, Miquel points, (Newton-)Gauss lines, etc. Solutions.
N.Beluhov, A.Zaslavsky, P.Kozhevnikov
## 1 Introductory problems
1.
a)Follows from sine Ceva theorem.
b)Proof by counting angles.
c)Let $P$ be a point of a given circle. Then measure of angle $A P B$ is fixed. Hence the sum of measures of angles $C A P$ and $C B P$ is fixed. Therefore, measure of angle $A P^{\prime} B$ is fixed, hence $P^{\prime}$ lies on a fixed circle passing through $A$ and $B$.
d) By $P_{A}, P_{B}, P_{C}$ denote points symmetric to $P$ in $B C, C A, A B$, respectively. Since $P_{A} C=$ $P C=P_{B} C$, the perpendicular bisector of the segment $P_{A} P_{B}$ passes through $C$, and hence it is the bisector $C P^{\prime}$ of angle $P_{A} C P_{B}$. Hence $P^{\prime}$ is the circumcenter of triangle $P_{A} P_{B} P_{C}$. By homothety with center $P$ and ratio $1 / 2$, the midpoint $T$ of $P P^{\prime}$ is the center of the circle passing through projections of $P$ to the sidelines. Similarly, $T$ is the center of the circle passing through projections of $P^{\prime}$. These two circles coincide since $T$ is equidistant from projections of $P$ and $P^{\prime}$ to a centain line.
In the case when $P^{\prime}$ is a point at infinity we obtain the Theorem on Simson line. for a point lying on the circumcircle, its projections of to the sidelines are collinear.
e) Let the common point of $X Y^{\prime}$ and $Y X^{\prime}$ be $P$ and the common point of $X Y$ and $X^{\prime} Y^{\prime}$ be $Q$. Let $X A \cap Y X^{\prime}=X_{A}$, and $Y^{\prime} A \cap Y X^{\prime}=Y_{A}$. Also call as $Q_{A}$ intersection point of $Q A$ and $Y X^{\prime}$. Prove that lines $A Q$ and $A P$ are symmetric with respect to the bisector of angle $A$. Since this is true for all angles points $P$ and $Q$ are isogonaly conjugated.
Consider cross-ratio $\left(Y, Q_{A}, X_{A}, X^{\prime}\right)$. Project these points from $Q$ to line $A X^{\prime}$. The map of $Y$ is $X^{\prime}$, the map of $Q_{A}$ is $A, X_{A}$ transforms to itself and $X^{\prime}$ transforms to the common point of lines $X^{\prime} Y^{\prime}$ and $X A$. Now project obtained ratio from $Y^{\prime}$ to line $Y X^{\prime}$. The map of $A$ is $Y_{A}$, the map of $X$ is $P, X_{A}$ transforms to itself and the map of $X^{\prime}$ transforms back to $X^{\prime}$. Thus $\left(Y, Q_{A}, X_{A}, X^{\prime}\right)=\left(P, Y_{A}, X_{A}, X^{\prime}\right)=\left(Y_{A}, P, X^{\prime}, X_{A}\right)$. It means that $A Q$ is the reflection of $A P$ in the bisector of angle $A$.
Another proof see also in the book of A.Akopyan and A.Zaslavsky "Geometry of conics" AMS, 2007.
f) By $P^{\prime}$ denote the point of intersection. Let $A B$ and $C D$ meet at $K$. For one of triangles $K B C$ and $K D A, P^{\prime}$ is isogonal conjugate to $P$. Hence, $P$ lies on the line symmetric to $K P$ in the bisector of angle $K$. For the other of two triangles $K B C$ and $K D A, P^{\prime}$ is also isogonal conjugate to $P$. We obtain that $P$ lies on all four lines symmetric to $A P, B P, C P$, and $D P$, in the bisectors of corresponding angles.
g)Suppose that projections of $P$ are concyclic. Then, similarly to p.d), obtain that point $P^{\prime}$ symmetric to $P$ in the center of the circle is isogonal conjugate to $P$. The converse statement is proved anagolously.
h)See the book mentioned above.
i)By the statement converse to the Theorem on Simson line, it is sufficient to prove that the projections of parabola focus to the sidelines are collinear. Since a point symmetric to the focus in any tangent lies on a directrix, the required statement follows.
2.
a) Let the circumcircles of $A B F$ and $C D F$ meet for the second time at $M$. We have $\angle(A M, M D)=\angle(A M, M F)+\angle(M F, M D)=\angle(B A, B F)+\angle(C F, C D)=\angle(B A, C D)=$
$\angle(A E, E D)$, hence $M$ lies on the circumcircle of $A D E$. Similarly for the other triangle.
b)It is sufficient to prove that triangles $M B E$ and $M F D$ are similar. We have $\angle(E B, B M)=$ $\angle(C E, C M)=\angle(C D, C M)=\angle(F D, F M)$. Angles $M E B$ and $M D F$ are equal by the same reasoning.
c) Let $B^{\prime}$ and $C^{\prime}$ are the positions of bugs at some moment. The intersection point $P$ of the circumcircles of triangles $A B C$ and $A B^{\prime} C^{\prime}$ is the center of spiral similarity taking line $B B^{\prime}$ to $C C^{\prime}$ (ratio of this similarity is equal to the ratio of the speeds of the bugs). Hence $P$ is Miquel point for quadruple of lines $B^{\prime} B, B C, C C^{\prime}, C^{\prime} B^{\prime}$.
d)Suppose $E$ moves from $B$ to $C$ at some constant speed, while $F$ moves from $D$ to $A$ at the same speed. The condition holds at any moment. We show that each of points $Q$ and $R$ moves at a constant speed, so statement of the problem follows from the previous problem. Note that $\angle E Q B=\angle F Q D, E B=F D$. Hence the circumcircles of triangles $E B Q F D Q$ are equal, and ratio $E Q / Q F$ is constant. Therefore $Q$ moves at a constant speed. Similarly, for $R$.
e)Miquel point is the focus of a parabola inscribed ro the quadrilateral.
f)By Therem on Simson line, eah three of four projections are collinear. Hence all four projections are collinear. It is known that homothety with center at some point of the circumcircle and ratio 2 takes Simson line to the line passing through the orthocenter. Therefore, by homothety with center at Miquel point and ratio 2 we obtain the line $l$ passing through orthocenters of four triangles. The orthocenters have equal powers with respect to three circles constructed on the diagonals of quadrilateral, hence $l$ is the radical axis for these three circles that is perpendicular to Gauss line (passing through the centers of the circles). For the inscribed parabola, the focus lies on the circumcircles of the triangles. Hence, Miquel point is the focus. The projections of Miquel point lie on a tangent to the parabola at its vertex. The line passing through the orthocenters is the directrix of the parabola. Hence, Gauss line is parallel to the axis od parabola.
3.
a)By $M$ denote the midpoint of $A C, N-$ midpoint of $B D, T$ - midpoint of $E F$. Let $F^{\prime}$, $A^{\prime}$ and $B^{\prime}$ are the midpoints of the sides of $A B F$. Note that $M$ lies on $F^{\prime} B^{\prime}, N$ lies on $F^{\prime} A^{\prime}$, $T$ lies on $A^{\prime} B^{\prime}$. By homotheties with centers at vertices of $A B F$ and ratio $2, \frac{\overrightarrow{F^{\prime} M}}{\overrightarrow{M B^{\prime}}}=\frac{\overrightarrow{B C}}{\overrightarrow{C F}}$, $\frac{\overrightarrow{B^{\prime} T}}{\overrightarrow{T A^{\prime}}}=\frac{\overrightarrow{A E}}{\overrightarrow{E C}}, \frac{\overrightarrow{A^{\prime} N}}{\overrightarrow{N F^{\prime}}}=\frac{\overrightarrow{F D}}{\overrightarrow{D A}}$. Multiplying these three equalities, and applying Menelaus theorem, we obtain the required.
b)Hint. Gauss line of quadrilateral $A B C D$ is the locus of points $X$ such that $S_{X A B}+S_{X C D}=$ $S_{X B C}+S_{X D A}$ (here the areas are oriented).
c) Follows from $2 \mathrm{f}$.
d) This is reformulation of b).
## 2 Three Miquels for a Quartet.
4. From Problems 13, 15 it follows that $A X, B Y, C Z$ pass through $D^{\prime}$ (isogonal conjugate to $D$ with respect to triangle $X Y Z$ ).
5. Follows directly from Problem 13.
6. Follows directly from Problems 13, 15 .
7. By problem 11, $\triangle X D Z \sim \triangle X Y A, \triangle X D^{\prime} Z \sim \triangle X Y A^{\prime}$. Hence $X A: X D^{\prime}=(X A$ : $X Z)\left(X Z: X D^{\prime}\right)=(X Y: X D)\left(X A^{\prime}: X Y\right)=X A^{\prime}: X D$, that equivalent to the statement of the problem.
8. Follows from the previous problem and Theorem on three homotheties applied to the segments $A A^{\prime}, D D^{\prime}$, and $B^{\prime} B$. Indeed, $Z$ is the center of homothety that takes $A$ to $B^{\prime}, A^{\prime}$ to $B$, etc. Alternative solution could be easily derived from Problem 1e).
9.
a) $X$ is a center of a spiral similarity that takes $C$ to $D, A$ to $B$. Since $K_{Y}$ is the image of $L_{Y}$ under this spiral similarity, angle $K_{Y} X L_{Y}$ equals to $\angle(A C, B D)$. Hence $X$ lies on the circle $P_{Y} K_{Y} L_{Y}$. Similarly obtain that $Z$ lies on the same circle.
b) Since $X$ lies on the circle $A P_{Y} B$, we have $\angle X P_{Y} B=\angle X A B$. Similarly, $\angle B P_{Y} Z=\angle B C Z$. From these and four analogous equalities it follows that $\angle X P_{Y} Z+\angle Z P_{X} Y+\angle Y P_{Z} X=\pi$, that implies the statement of the problem.
c)From problem 15 it is clear that $\psi_{X}\left(P_{Y}\right)=P_{Z}$, etc. This means that $P_{X}, P_{Y}, P_{Z}$ belong to a certain quartet, so $\psi_{X}\left(P_{X}\right)=\psi_{Y}\left(P_{Y}\right)=\psi_{Z}\left(P_{Z}\right)$. We need to show that $\psi_{Z}\left(P_{Z}\right)$ is isogonal conjugate to $T$, or equivalently, $\psi_{Z}(T)$ is isogonal conjugate to $P_{Z}$. Note that $\psi_{Z}$ takes circles $Z X P_{Y}$ and $Z Y P_{X}$ (passing through $T$ ) to lines $Y P_{X}$ and $X P_{Y}$, hence $\psi_{Z}(T)=Y P_{X} \cap X P_{Y}$. From equalities $\angle P_{X} Y X=\angle Z Y P_{Z}, \angle P_{Y} X Y=\angle Z X P_{Z}$ obtain the required statement.
## 3 Quartets for three Miquels.
10. a)-b) Follows directly from the definition.
c) Each of the transformations $\psi_{X}, \psi_{Y}, \psi_{Z}$ is circular (i.e. takes a circle either to a circle or to a line) and preserves the orientation. Hence any product of $\psi_{Z}, \psi_{Y}, \psi_{X}$ is a circular transformation preserving the orientation. Moreover, from a)-b) follows $X, Y, Z$ are invariant points for $\psi_{Z} \psi_{Y} \psi_{X}$. Note that a circular transformation preserving the orientation is uniquely defined by the images of three points. The statement of the problems is independent of the order of $\psi_{X}, \psi_{Y}, \psi_{Z}$ in its product. Hence $\psi_{X}, \psi_{Y}, \psi_{Z}$ commute to each other.
11. By definition of $\psi_{X}, \angle Z X D=\angle A X Y$ and $X D \cdot X A=X Y \cdot X Z$, that implies the first similarity. The second similarity is proved analogously.
12. From Problem 10 it follows, in particular, that $\psi_{X} \circ \psi_{Y}=\psi_{Z}^{-1}=\psi_{Z}$. Therefore, $\psi_{Y}(A)=$ $\psi_{Y}\left(\psi_{X}(D)\right)=\psi_{Z}(D)=C$, so $\psi_{Y}$ interchanges the points in the pairs $(A, C),(B, D)$. Similarly we obtain that $\psi_{X}$ interchanges the points in the pairs $(A, D),(B, C)$, while $\psi_{Z}$ interchanges the points in the pairs $(A, B),(C, D)$.
13. Let $D^{\prime}, A^{\prime}$ be isogonal conjugates to $D, A$ respectively. Then $A^{\prime}$ lies on $X D, D^{\prime}$ lies on $X A$. Moreover, $\angle X D^{\prime} Z=\pi-\angle Z X D^{\prime}-\angle D^{\prime} Z X=\pi-\angle D X Y-\angle Y Z D=\angle Z D X+\angle X Y Z-\pi$. By Problem 11 we have $\angle Z D X=\angle A Y X$, i.e. $\angle X D^{\prime} Z=\angle X Y A^{\prime}$. Hence $\triangle X D^{\prime} Z \sim \triangle X A^{\prime} Y$, and $A^{\prime}=\psi_{X}\left(D^{\prime}\right)$.
14.
a) The incenter and the excenters of $X Y Z$.
b)Point $O$ and three points symmetric to $X, Y, Z$ in the opposite sidelines of $X Y Z$.
c) From the definition of $\psi_{Z}$ it follows that its invariant point lie on bisector of angle $X Z Y$, and on the circle with center $Z$ and radius $\sqrt{Z X \cdot Z Y}$. There are two such points $U$ and $V$. From Problem 10 we obtain that $\psi_{X}(U)=\psi_{Y}\left(\psi_{Z}(U)\right)=\psi_{Y}(U)=\psi_{Z}\left(\psi_{X}(U)\right)$, i.e. $\psi_{X}(U)$ is also an invariant point for $\psi_{Z}$. It is clear that $\psi_{X}(U) \neq U$, therefore, $\psi_{X}(U)=\psi_{Y}(U)=V$, and a required quartet (both for $U$ and for $V$ ) is $U, U, V, V$. Further, the isogonal conjugates to $U, V$ are also invariant under $\psi_{Z}$. Hence $U$ and $V$ are isogonal conjugates to each other, and quartet $U, U, V, V$ coincides to conjugate quartet.
15.
a)From the definition of $\psi_{X}$ and the Problem 12 it follows that $X$ is the center of spiral similarity taking $A$ to $B$, and $C$ to $D$. By Problem 2b) this center is Miquel point.
b) $Y$ is Miquel point for the quadruple of lines $A B, B C, A D, C D ; Z$ is Miquel point for the quadruple $A D, A C, B D, B C$.
c) $X$ is Miquel point, hence (in particular) $\triangle X B D \sim \triangle X A C$. Let $P_{X}=A D \cap B C, P_{Y}=$ $A C \cap B D, P_{Z}=A B \cap C D$. We have $X A \cdot X D=X B \cdot X C=X P_{Y} \cdot X P_{Z}=R_{a}^{2}$, and the angles $A X D, B X C, P_{Y} X P_{Z}$ have a common bisector $l$. The inversion with center $X$ and radius $R_{a}$ followed by the symmetry in $l$ takes triangles $A D P_{Y}$ and $B C P_{Y}$ to $D A P_{Z}$ and $C B P_{Z}$, respectively. Therefore it tales intersection point $Y$ of the circles $A D P_{Y}$ and $B C P_{Y}$ to $Z$. Hence the product of an inversion and a symmetry defined above is $\psi_{X}$.
16. From Problem 13 it follows that the product of $\psi_{X}$ and the isogonal conjugacy (in any order) takes $D$ to $A^{\prime}$.
# Раскраски и кластеры
А. Белов-Канель,
И. Иванов-Погодаев, А. Малистов, М. Харитонов
1. Плоскость раскрашена а) в два цвета b) в три цвета. Докажите, что найдутся две точки одного цвета, расстояние между которыми 1.
2. Тот же вопрос для пространства, раскрашенного в 4 цвета.
3. Тот же вопрос для $n$-мерного пространства, раскрашенного в $n+1$ цвет.
4. Плоскость раскрашена в два цвета. Докажите, что в одном из цветов укладываются все расстояния.
5. Тот же вопрос для $n$-мерного пространства, раскрашенного в $n$ цветов.
6. Тот же вопрос для плоскости, раскрашенной в 3 цвета.
7. Тот же вопрос для $n$-мерного пространства раскрашенного в $\mathrm{n}+1$ цвет.
8. Раскрасьте плоскость в возможно меньшее число цветов, чтобы не было единичного отрезка с одноцветными концами.
$K$ настоящему времени неизвестно минимальное число $x$ цветов, такое что при некоторой раскраске плоскости в $x$ цветов нет отрезка единичной длины с одноцветными вершинами. Известно только, что $4 \leqslant x \leqslant 7$.
Задача упрощается, если искать «почти единичные» отрезки.
9. Плоскость раскрашена a) в четыре цвета b) в пять цветов. Докажите, что найдутся две точки одного цвета, расстояние между которыми отличается от единицы не более чем на 0,001 .
Решение пункта $b$ предыдущей задачи основывается на следующем факте:
10. Клетки плоскости раскрашены в два цвета (общая граница клеток разных цветов считается пестрой, каждая клетка раскрашена полностью в один цвет). Докажите, что найдется одноцветная ломаная (все точки покрашены в один цвет), концы которой находятся на расстоянии больше 1000.
11. Пространственное обобщение предыдущей задачи на пространственную трехмерную решетку, клетки которой покрашены в три цвета
12. То же для $n$-мерной решетки, покрашенной в $n$ цветов.
13. Куб $k \times k \times k$ разбит на $k^{3}$ единичных кубиков, каждый из которых покрашен в красный, синий или зеленый цвет. Докажите, что найдется одноцветная ломаная, соединяющая противоположные грани. Сформулируйте и докажите $n$-мерное обобщение. Докажите также, что при увеличении количества цветов на единицу утверждение становится неверным.
14. Трехмерное пространство покрашено в 9 цветов. Докажите, что найдутся две одноцветные точки, расстояние между которыми отличается от единицы меньше, чем на 0,001 . Обобщите задачу на $n$-мерный случай.
Определение. Назовем кластером множество связанных клеток. Две клетки, имеющие общую точку, считаются связанными. Результаты задачи 12 поддаются дальнейшему обобщению.
15. Пусть все клетки единичной $n$-мерной решетки покрашены в $k$ цветов $(kИ. Иванов-Погодаев, А. Малистов, М. Харитонов
18. Слой между двумя параллельными прямыми раскрашен в 2 цвета. Докажите, что в нём найдутся 2 одноцветные точки на единичном расстоянии.
19. Слой между двумя параллельными плоскостями раскрашен в 4 цвета. Докажите, что в нём найдутся 2 одноцветные точки на единичном расстоянии.
20. а) Лемма Шпернера. Треугольник, вершины $A B C$ которого раскрашены в цвета $1,2,3$ соответственно, разбит на треугольники. Вершины этих треугольников раскрашены в цвета $1,2,3$ так, чтобы точки, лежащие на $[A B]$, были раскрашены в цвета 1,$2 ;$ на $[B C]-$ 2,$3 ;[C A]-3,1$. Тогда найдётся треугольник, вершины которого раскрашены в разные цвета.
b) Сформулируйте и докажите это утверждение для $n$-мерного пространства.
21. а) Докажите, что не существует непрерывного отображения диска на свою границу, тождественного на этой границе (т. е. любая точка границы переходит сама в себя). Такие отображения называются ретрактами.
b) Докажите теорему Брауэра о неподвижной точке: Непрерывное отображение диска в себя имеет неподвижную точку.
c) Сформулируйте и докажите $n$-мерные аналоги этих фактов.
22. Индуктивное определение размерности. 0-мерное множество: если все его точки лежат в разных компонентах связности. 1-мерное: не 0 -мерное, любые 2 точки которого разделяются 0 -мерным множеством, и т.д. Докажите, что $\mathbb{R}^{n}$ согласно этому определению $n$-мерно.
## Раскраски и кластеры
## М. Матдинов
Вариации на задачи про кластеры
23. Пусть $k$-мерный куб со ребром $n$ разбит на $n^{k}$ маленьких $k$-мерных кубиков со стороной 1 , раскрашенных в $\ell$ цветов. Рассмотрим все тройки цветов $a, b, c$. Рассмотрим множество точек, каждая из которых покрашена во все эти цвета. Окружим все эти точки окрестностью радиуса 2 и рассмотрим объединения этих окрестностей и связные компоненты такого объединения. Пусть каждая такая связная компонента для любого цвета имеет диаметр не больше $d$.
Тогда найдётся константа $C(k, d, \ell)>0$ такая, что существует кластер объёма $C(k, d, \ell) \cdot n^{k-1}$.
a) Докажите это для $k=3$,
b) для произвольного $k$.
24** Сформулируйте условие, аналогичное условию предыдущей задачи, для наборов из $m$ цветов. Общая гипотеза: существует константа $C(k, m, d, \ell)>0$ такая, что при раскраске $k$-мерного куба с ребром $n$ в $l$ цветов найдётся кластер объёма $C(k, m, d, \ell) \cdot n^{k+2-m}$.
Данная задача является своего рода обобщением задачи 15 . Её решение нам не известно.
## Colorings and clusters
## A. Belov-Kanel,
I. Ivanov-Pogodaev, A. Malistov, M. Kharitonov
1. Each point of the plane is colored by one of a) two colors; b) three colors. Prove that there exist two points separated by the distance 1 which have the same color.
2. Prove the same for the space colored by 4 colors.
3. Prove the same for the $n$-dimensional space colored by $n+1$ colors.
4. The plane is colored by two colors. Prove that one can choose the color such that for any distance $d$ there exist the points separated by the distance $d$ and colored by this color.
5. Prove the same for the $n$-dimensional space colored by $n$ colors.
6. Prove the same for the plane colored by 3 colors.
7. Prove the same for the $n$-dimensional space colored by $n+1$ colors.
8. Color the plane such that there are no unit segment with the edges colored by the same color. Try to use the minimal number of colors. So far, it remains an open question to find the minimal number of colors $x$ such for some coloring there are no unit segment with the single-colored edges. It is known that $4 \leq x \leq 7$.
If we look for "almost unit" segments then the problem became more simple.
9. The plane is colored by a) four colors; b) five colors. Prove that there exist two points of the same color such that the distance between them is different of 1 by less then 0.001 .
The solution of the b-part of the previous problem is based on the following fact:
10. Consider the cell-like plane. Suppose that any cell is fully colored by one of two colors. Let the common edge of the two cells be colored by both cell-colors. Prove that there exists a single-colored polygonal path such that the edges of the path are separated for the distance more then 1000 .
11. Prove the generalization of the previous problem for 3-dimensional cube lattice colored by 3 colors.
12. Prove the generalization of the previous problem for $n$-dimensional cube lattice colored by $n$ colors.
13. Suppose that a cube $k \times k \times k$ consists of $k^{3}$ unit cubes which are colored by red, blue and green colors. Prove that there exists a single-colored polygonal path which connects opposite faces of the big cube. Formulate and prove the $n$-dimensional generalization. Prove that if we use $n+1$ colors then the statement is not true.
14. 3-dimensional space is colored by 9 colors. Prove that there exist two single-colored points which are separated by the distance $d$ such that $|d-1|<0.001$. Generalize this problem for $n$-dimensional case.
Definition. Connected set of cells called claster (Two cells with common point are connected).
The achievements of the problem 12 can be further generalized.
15. Suppose that the cells of the $n$-dimensional lattice are colored by $k$ colors and $kI. Ivanov-Pogodaev, A. Malistov, M. Kharitonov
There is a mistake in the first version of the problem 16. Here is the right version.
16. Show that the statement of the problem 12 can be obtained from the following fact, which is base for topological definition of dimension: if $n$-dimensional space is covered by some open sets of bounded diameter then there exists a point covered $n+1$ times.
17. Suppose the layer between two parallel lines is colored by two colors. Prove that there exist 2 points of the same color inside the layer such that distance $(A, B)=1$.
18. The same problem for the layer between two parallel planes, colored by 4 colors.
19. a) Sperner Lemma. Triangle $A B C$ is divided by small triangles; vertices of these triangles are colored by 3 colors such that $A$ is colored by color $1, B$ by color $2, C$ by color 3 . Vertices on $[A B]$ are colored by colors 1 or 2 ; vertices on $[B C]$ - by colors 2 or 3 , vertices on $[C A]$ - by colors 3 or 1 . Prove that there exists a small triangle, which vertices are colored by different colors.
b) Generalize this lemma for $n$-dimensional space (consider also $n=1$ ).
21. a) Prove that there is no continuous mapping of disc onto its boundary $R$ such that it is identical on the border. This mapping is called Retract. b) Prove the Brauer theorem: Continuous mapping $F$ of the disc into itself has a stable point, i.e. there exists a point $x_{0}$ such that $F\left(x_{0}\right)=x_{0}$. c) Generalize this for $n$-dimensional space.
22. Inductive definition of dimension. 0-dimensional space is a space such that any 2 points are situated in different connected components, 1-dimensional space is a space such that any 2 points can be separated by 0 -dimensional space (and it is not 0 -dimensional). $n$-dimensional space is a space, such that any two points are separated by $n-1$-dimensional space (and it is not ( $n-1$ )-dimensional). Prove that $\mathbb{R}^{n}$ is $n$-dimensional space according this definition.
# Colorings and clusters
M. Matdinov
Additional problems
23. ${ }^{\star} k$-dimensional cube $n \times n \times \cdots \times n$ is divided by $n^{k} k$-dimensional subcubes $1 \times 1 \times \cdots \times 1$ colored by $\ell$ colors. Consider all triples of colors $(a, b, c)$. Consider the set of points colored by these 3 colors simultaneously. Surround each of them by circle of radius 2 . Consider connected components of the union of these circles. Suppose that all such components for all triples $(a, b, c)$ have diameter less than $d$.
Then there exists a positive constant $C(k, d, \ell)>0$ such that for any coloring there exists a cluster of volume $C(k, d, \ell) \cdot n^{k-1}$.
a) Prove that for $k=3$.
b) Prove that for all $k$.
24.* Generalize condition of the previous problem for $m$-tuples of colours. General hypothesis: there exists a constant $C(k, m, d, \ell)>0$ such that for each coloring of $k$-dimensional cube $n \times n \times \cdots \times n$ by $l$ colors there exists a cluster of volume $C(k, m, d, \ell) \cdot n^{k+2-m}$.
This hypothesis can be considered as generalization of the problem 15 . We don't know how to solve it.
## Colorings and clusters Solutions
1a. В равностороннем треугольнике со стороной 1 по принципу Дирихле найдутся 2 вершины одного цвета.

1a. Consider the regular triangle with the side 1. Using Dirichlet principle we can find 2 vertices colored by the same color.

1b. Предположим обратное. Тогда рассмотрим точку $A$ цвета 1 . Если такой не найдётся, задача сводится к 1а. Докажем, что на расстоянии $\sqrt{3}$ от $A$ все точки цвета 1. Рассмотрим произвольный равносторонний треугольник $A B C$ со стороной 1 . Точки $B$ и $C$ раскрашены в цвета 2 и 3. Рассмотрим равносторонний треугольник $\mathrm{BCD}$. Точка $\mathrm{D}$ покрашена в первый цвет. Расстояние между $A$ и $D$ равно $\sqrt{3}$. Повторяя такую опе-
рацию для произвольных точек получим, что все точки на расстоянии $\sqrt{3}$ от $A$ покрашены в первый цвет, а т.к. среди них найдутся 2 точки на расстоянии 1 , имеем требуемое утверждение.
2. Аналогично $1 \mathrm{~b}$, только берём не равносторонние треугольники, а равносторонние тетраэдры.
3. Аналогично 2 , только берём не равносторониие тетраэдры, а $n$ мерные симплексы.
4. Предположим обратное. Тогда у цветов 1 и 2 найдутся такие расстояния $x$ и $y$ соответственно, без ограничения общности $x \geqslant y$, что $x$ не укладывается в цвете 1 , а $y$ - в 2 . Тогда рассмотрим точку $A$ цвета 1 . Вокруг неё опишем окружность радиуса $x$. Все точки этой окружности покрашены в цвет 2. Тогда на ней найдутся две точки цвета 2 , расстояние между которыми - $y$.
5. Предположим противное. Аналогично задаче 4 рассмотрим расстояния $x_{1} \geqslant x_{2} \geqslant \ldots \geqslant x_{n}$ такие, что $x_{1}$ не укладывается в цвете $1, x_{2}-$ в цвете $2, \ldots, x_{n}$ - в цвете $n$. Далее доказательство проводим по индукции. Если точки цвета 1 нет, то применяем индукционное предположение. Рассматриваем точку 1 -го цвета и описываем вокруг неё $n$-мерную сферу радиуса $r_{1}=x_{1}$. На ней только точки 2 -го, $\ldots, n$-го цветов. Теперь рассмотрим на ней точку 2 -го цвета. Если её нет, то задача опять сводится к случаю меньшей размерности. Описываем вокруг этой точки сферу радиуса $x_{2}$. В пересечении этих двух сфер получаем сферу меньшей размерности и радиуса $r_{2}$, раскрашенной в цвета $3,4, \ldots, n$. Продолжая этот процесс далее получаем точку, которая не может быть покрашена ни в какой из цветов. Несложно доказать, что $d_{i}>r_{i} \geqslant \frac{\sqrt{2}}{2} d_{i}$, откуда следует требуемое утверждение.
6. We shall act similarly to the problem 4. Assume the contrary. Suppose that there are no two points of color 1 on the distance $x_{1}$ from each other, there are no two points of color 2 on the distance $x_{2}$ from each other, etc. We can also suppose that $x_{1} \geqslant x_{2} \geqslant \ldots \geqslant x_{n}$.
Let us consider a point $A_{1}$ of the color 1 and sphere $S_{1}$ with center $A_{1}$ with the radius $r_{1}=x_{1}$. Next we consider a point $A_{2} \in S_{1}$ of color 2 and sphere $S_{2} \subset S_{1}$ obtained by intersection of $S_{1}$ with sphere of radius $x_{2}$ centered in the point of $A_{2}$. Similarly, we construct point $A_{3}$ and $(n-4)$-dimensional sphere $S_{3}$ and so on.
If we can not find point on the sphere $S_{k}$ with color $k+1$, we proceed with next color and distance. Finally we get a point, which can not be colored in any color and get a contradiction.
The only thing we have to take care that process can be continued on the each step, i.e. all spheres will be not empty. This can be guaranteed by proving that $2 \cdot r_{i}>d_{i+1}$.
8. Раскраска в 7 цветов. Границы шестиугольников покрашены в любой цвет, их диаметр - 0,99 : (рисунок)

The coloring of the plane in 7 colors is based on the hexagonal lattice. diameter of each hexagon is 0.99 (see picture).
9b. Предположим противное. Разобьём всю плоскость на квадратики со стороной $\varepsilon / 1000$. Если в квадратике найдутся точки 3 цветов, то все точки, расположенные от него на расстоянии от $1-\varepsilon / 2$ до $1+\varepsilon / 2$ раскрашены в 2 цвета, значит среди них найдутся 2 точки одного цвета, расстояние между которыми отличается от единицы не больше чем на $\varepsilon$.
Если каждый квадратик раскрашен не более чем в 2 цвета, то будем считать его квадратиком одного из этих цветов. Без ограничения общности найдётся квадратик 1-го цвета. Рассмотрим кластер максимальной площади из квадратиков 1го цвета. Если с ним по внешней границе граничат кластеры двух цветов, то найдётся квадрат со стороной $\varepsilon / 10$, содержащий точки 3 цветов. Далее доказываем аналогично первой части доказательства.
Если этот кластер 1-го цвета граничит только с кластером одного цвета, без ограничения общности 2 го, то рассматриваем этот кластер 2 -го цвета. Он кроме кластера первоначального цвета граничит только с одним кластером. Продолжая эту операцию, получаем кластер диаметра не меньше 2 , а в таком кластере найдутся 2 точки на расстоянии, отличающемся от единицы не больше чем на $\varepsilon$. Противоречие.
9b. Suppose contrary. Let us divide the plane onto squares with the side $\varepsilon / 1000$.
Suppose we can find 3 points of 3 different colors $a, b c$ in one of these squares. Let $A$ be center of such square. Consider a circle $C$ of radii 1 with the center $A$. It is clear that all its point are colored with remaining two colors. If both of them are present (otherwise we can find 2 points of the same color of distance 1), then there are two points in $C$ of different colors arbitrary close to each other. By considering intersection point of $C$ and circle $C^{\prime}$ of radii 1 centered by one of these point, we get a contradiction.
Suppose there are no such square. Let us color each square in color of an arbitrary point in it. Then no three squares of pairwise different colors meet and we have cluster with arbitrary big diameter, greater than 1 . Then we have pair of points which we need.
14. The proof is similar to the proof of problem $9 \mathrm{~b}$. We divide a space on small cubes and color each of them in the color of its arbitrary point. If there is no pairwise contacted cubes of 4 different colors, then there exist a cluster diameter grater than 1 and we are done. (This fact can be obtained from dimension theorem quite similar as Lebeg theorem). Next we consider a sphere $S$ of radii 1 centered by center of one of these cubes and continue.
In order to proceed next step we divide equator of $S$ onto 4 equal parts $P_{1}, P_{2}, P_{3}, P_{4}$ and consider north hemisphere of $S$ as a "square" with edges $P_{i}$. Next we proceed similarly to the problem $9 \mathrm{~b}$.
14. Разбиваем пространство на кубики с ребром $\varepsilon / 1000$. Раскрашиваем их в один из цветов, которые в них содержатся. Из задачи 16 находим кубик, в котором найдутся точки 4-ёх цветов. Тогда точки на расстоянии от $1-\varepsilon / 2$ до $1+\varepsilon / 2$ раскрашены в 5 цветов. Далее действуем аналогично задаче $9 b$.
15. Если $n$-мерное пространство покрыто открытыми множествами ограниченного диаметра, то есть точка, покрытая $n+1$ раз.
Указание. Пусть все открытые множества ограничены диаметром $d$. Рассмотрим $n$-мерный правильный симплекс со стороной $1000 d$.
Присвоим всем вершинам симплекса цвета от 1 до $n+1$. Также присвоим всем открытым множествам цвета от 1 до $n+1$ так, чтобы множества покрывающие вершины имели цвета этих вершин, множества, покрывающие ребра имели цвета одной из двух вершин этих рёбер, множества, покрывающие $k$-мерные грани имели цвет одной из $k+1$ вершин, ограничивающих грань.
Рассмотрим непрерывное отображение внутренних точек симплекса на его границу. Каждая точка, покрытая один раз переходит в вершину
соответствующего цвета. Точка, покрытая $k$ раз переходит в точку на $k$ мерной грани, вершины которой раскрашены в соответствующие цвета, при этом точка выбирается как средневзвешенный центр масс с весами, равными расстояниям до границы соответствующего множества.
Такое отображение является ретрактом, а непрерывного ретракта не существует.
17. Note. Suppose that all open sets are bounded by diameter $d$. Consider $n$-dimensional simplex with edges equal to $1000 d$.
Let us assign the colors from 1 up to $n+1$ for the simplex vertices. Also we assign these colors to the open sets such that the following conditions hold: the sets covering vertices colored by its colors; the sets covering edge colored by one of its vertices colors; the sets covering $k$-dimensional face colored by one of this face vertices colors.
Consider the continuous mapping from simplex to its boundary. Each point covered by one set is transformed to the vertex colored by the color of this set. Each point covered by $k$ sets is transformed to a point on the $k$ dimensional face which vertices are colored by the corresponding colors. This point is situated at weighted mass center of these vertices according to the distances to the boundaries of the corresponding sets.
This mapping is a retract. But continuous retract does not exist.
10,11,12, 16. Указание. Все кластеры ограничены, иначе мы можем найти длинный путь в неограниченном кластере. Каждому кластеру, состоящему из кубов, сопоставим открытое множество, состоящее из точек кластера и некоторой $\varepsilon$-окрестности $\left(\varepsilon=10^{-3}\right.$ ). Из топологического факта задачи 17 следует, что существует точка, покрытая $n+1$ множеством. Из принципа Дирихле следует, что существуют две пересекающиеся окрестности кластеров одного цвета. Тогда это должен быть один кластер.
10,11,12, 16. Note. All the clusters are bounded. If not, there exists a long path in some unbounded cluster. For each cluster consider an open set formed by the cubes of the cluster and its $\varepsilon$-neighborhood. Using the fact of problem 17 we obtain that there exists a point covered by $n+1$ sets. Hence this point is covered by two clusters of the same color. But it is impossible.
13. Указание. Пусть существует раскрашенный куб $k \times k \times k$ без сквозного пути. Отразим куб относительно каждой его грани в соответствующие этим граням стороны. Будем отражать получившиеся кубы относительно других граней, заполняя отражениями исходного куба все новые и новые области. Получим заполнение пространства отражениями
нашего куба. Для любого кластера в исходном кубе существуют три соприкасающиеся по вершине грани, которых этот кластер не касается. В виду построенных отражений этот кластер не касается всех граней некоторого куба $2 k \times 2 k \times 2 k$, внутри которого он находится. Таким образом все кластеры ограничены, что противоречит факту задачи 12 .
14. Note. Suppose that there exists a colored cube $k \times k \times k$ having no path from some face to the opposite one. Let us reflect the cube using each its face. Then we reflect these reflected cubes again using other faces. So we can fill the space by reflections of our initial cube. For any cluster in the cube there exist three faces having some common vertex such that the cluster does not intersect them. It is clear that our cluster is bounded by some $2 k \times 2 k \times 2 k$ cube. Hence all cluster are bounded. Using the fact from problem 12 we obtain a contradiction.
## Задачи о покрытиях и функции роста
А.Толпыго, Б.Френкин, М.Прасолов, И.Богданов
## Предисловие.
Тема данного цикла - покрытия фигур однотипными фигурами (как правило - кругами или шарами). Требуется оценить их количество, их суммарную площадь и т. п.
Наиболее трудными здесь, естественно, являются задачи, где требуется дать точную оценку. K счастью, оказывается, что основные приложения таких задач как раз не требуют точного ответа, достаточно знать порядок соответствующей величины. Точно соответствующие определения будут сформулированы в цикле В.
## Цикл А.
## A1
Нетрудно покрыть единичный квадрат кругом площади $\pi / 2$. А можно ли покрыть квадрат несколькими кругами, суммарная площадь которых меньше $\pi / 2$ ? Круги могут пересекаться и выходить за пределы квадрата.
## A2
Требуется покрыть единичный квадрат несколькими кругами, радиус каждого из которых равен $r$. Пусть $N(r)$ - минимальное число кругов, которыми это можно сделать.
Очевидно, если $r \rightarrow 0$, то $N(r) \rightarrow \infty$.
Найти характер стремления к бесконечности этой функции (как быстро она растет)?
## A3
Та же задача, если требуется покрыть единичный куб несколькими (пересекающимися) шарами радиуса $r$.
## Цикл В.
В предыдущих задачах подразумевалось, что всякому интуитивно понятно, что такое «скорость роста». Но дальше нам потребуется точное определения: что такое «рост функции»?
Дать такое определение не очень легко, и оно будет дано чуть ниже. Вначале же мы, еще не давая точного определения «роста функции», сформулируем свойства этого «роста». А именно:
Далее рассматриваются только функции $f(x)$ такие, что
$(*) f(x)$ определена для всех $x$, больше некоторого $a$ (как правило, $a=0$, но удобнее разрешить более общий случай - напр. функции типа $\ln (x-1)$ ). Кроме того, предполагается, что $f(x)$ всюду положительна (во всяком случае, при $x>a$ ), нестрого возрастает и стремится к бесконечности.
Обозначим рост функции $f(x)$ через $[f]$. В частности, для простоты будем далее обозначать рост функции $x^{n}$ просто $n$. Таким образом, $n=\left[x^{n}\right]$.
Мы хотим, чтобы рост функции обладал следующими свойствами:
(1) Если для всех $x$, больших некоторого $b$ ( $b$ не обязательно совпадает с $a$ ) выполняется $f(x)>g(x)$, то $[f] \geqslant[g]$.
(2) Пусть $A, B, C$ - произвольные положительные числа, и $g(x)=C f(A x+B)$. Тогда $[g]=[f]$.
(3) Отсюда понятно, что вполне возможна ситуация, когда одновременно $[f] \geqslant[g]$ и $[g] \geqslant[f]$. В этом случае мы также полагаем, что $[f]=[g]$.
Если же $[f] \geqslant[g]$, но неверно, что $[g] \geqslant[f]$, то мы пишем: $[f]>[g]$.
Однако до сих пор мы не дали определения: что же такое рост? Правильный ответ состоит в том, что рост как раз и есть нечто, удовлетворяющее всем перечисленным свойствам.
Если говорить более формально, то следует рассмотреть класс функций, удовлетворяющих (*), и разбить его на классы эквивалентности: $f \sim g$, если $[f]=[g]$ (в соответствии с пп. $(2),(3))$.
Каждый из этих классов и называется ростом всех входящих в него функций. При этом некоторые классы больше других, так что мы можем говорить о том, что такая-то функция растет быстрее (ее рост больше), чем другая.
Докажите, что $1<2$.
(Напоминание: 1 и 2 - не числа, а рост функций).
Однако если для обычных чисел всегда верно одно из трех: либо $a>b$, либо $a=b$, либо $a0$ выполняется неравенство $[f]\alpha$ квадрат можно покрыть кругами площади $\beta$. (Постарайтесь также выяснить, можно ли его покрыть кругами площади ровно $\alpha$ ).
Эта задача подразделяется на три. Для формулировки первой укажем, что одно из решений задачи A1 состоит в следующем: квадрат покрывается сеткой из правильных шестиугольников, и каждый шестиугольник покрывается кругом. Отношение площади круга к площади шестиугольника равно $\gamma=\frac{2 \pi}{3 \sqrt{3}}$. Поскольку площадь сетки немного больше площади квадрата, то суммарная площадь кругов (обозначим её $\beta$ ) будет больше $\gamma$, но разность $\beta-\gamma$ можно сделать сколь угодно малой, взяв достаточно малые шестиугольники.
## D1
Докажите, что если все радиусы кругов должны быть одинаковыми, то приведённая выше конструкция оптимальна, то есть $\alpha=\gamma$.
## D2
Найдите $\alpha$, если разрешается брать круги двух произвольных радиусов.
## D3
Найдите $\alpha$, если разрешается брать круги произвольных радиусов без всяких ограничений.
(Предупреждение. Не думайте, что задачи упорядочены по возрастанию сложности!)
## D4-D6
Та же задача, но требуется покрыть куб со стороной 1 шарами (шары, разумеется, могут пересекаться). (Здесь достаточно дать хорошие оценки; точный ответ неизвестен).
## D7
Требуется покрыть квадрат несколькими кругами равного радиуса так, чтобы каждая точка была покрыта не менее, чем $N$ кругами. Докажите, что при некотором $N$ суммарную площадь кругов можно сделать меньше, чем $N \gamma$.
## D8
Верно ли утверждение задачи $\mathrm{D} 7$ при $N=2$ ?
## Цикл Е.
## E1
На столе лежит листок бумаги в клеточку. Поверх него положен еще один лист бумаги в клеточку; клетки на обоих листах квадратные и одного размера, но второй лист положен наискось, так что его линии не параллельны линиям первого. Верхний листок прозрачный, и видно, как его линии делят один из квадратов нижнего листка.
На какое максимальное число частей может быть разделен нижний квадрат?
А на какое минимальное?
## E2
На столе лежит листок бумаги в клеточку размером $10000 \times 10000$. Поверх него положен еще один лист бумаги в клеточку размером $1000 \times 2000$; клетки на обоих листах квадратные и одного размера. Верхний листок прозрачный, и видно, как линии нижнего листка делят квадраты верхнего листка на части; таким образом, мы видим, что число частей верхнего листка больше двух миллионов.
Докажите, что это число меньше десяти миллионов.
## Цикл F.
Возвращаемся к задачам того же типа, что в цикле С. Однако теперь рассмотрим несколько иную конструкцию. Число $\varepsilon$ мы будем считать фиксированным (напр. $\varepsilon=1$ ), но будем рассматривать неограниченную фигуру $\Phi$ - например, всю плоскость, полуплоскость, и т.п.
Рассмотрим произвольную точку $O$ данной фигуры и фигуру $\Phi(R)$, состоящую из всех точек, принадлежащих данной фигуре и отстоящих от $O$ на расстояние не более $R$ (пересечение данной фигуры с кругом или, может быть, шаром с центром в данной точке).
Обозначим через $N(R)$ число точек минимальной $\varepsilon$-сети в фигуре $\Phi(R)$. Пусть $\chi$ - скорость роста функции $N(R)$. Если эта функция не стремится к бесконечности при $R \rightarrow \infty$, положим условно $\chi=0$.
$\chi$ называется объёмной характеристикой данной фигуры.
## F1
У каких фигур объемная характеристика равна нулю?
## F2
Найти объемную характеристику следующих фигур: (а) плоскости, (б) полуплоскости, (в) полосы, заключенной между двумя параллельными прямыми.
## F3
Докажите, что объемная характеристика фигуры не зависит от того, как выбрана точка $O$.
## F4
Даны два положительных числа $\varepsilon$ и $\delta$. Верно ли, что для любой фигуры $\Phi$ её объёмная характеристика, вычисленная с помощью кругов радиуса $\varepsilon$ и радиуса $\delta$, одна и та же?
## F5
Найти объемную характеристику внутренности параболы и внешней части параболы.
## F6
Найти объемную характеристику каждой из трех частей, на которые гипербола $x y=1$ делит плоскость.
## F7
Какие числовые значения может принимать объемная характеристика фигуры (фигуры на плоскости, фигуры в пространстве)?
## F8
Может ли объемная характеристика фигуры не быть числом (т.е. быть функцией, которая не сравнима с числами); например, может ли она, подобно функции из задачи В4, быть больше 1 , меньше 2 и притом несравнима с числом $3 / 2$ ?
## Решения
## A1
Ответ. Можно.
Один из способов состоит в том, чтобы покрыть квадрат одним кругом радиуса $\frac{\sqrt{2}}{2}$, затем немного уменьшить этот радиус. Останется 4 не покрытых уголка, которые можно покрыть четырьмя маленькими кругами.
Другой способ приведен в комментариях к циклу задач D.
## $\mathrm{A} 2, \mathrm{~A} 3$
В первом случае функция $N(r)$ растет как квадрат (пользуясь терминологией цикла В, $[N(r)]=2)$, во втором случае - как куб.
## B1
Каковы бы ни были $A, B, C$, при больших $x$ всегда $x^{2}>C(A x+B)$. Таким образом, $\left[x^{2}\right] \geqslant[x]$, но неверно, что $[x] \geqslant\left[x^{2}\right]$. Это и значит, что $[2]>[1]$.
## B2, B3
Искомую функцию можно построить, например, так. Построим последовательность быстро увеличивающихся интервалов (нам подойдут, например, интервалы: $[2,4],[4,16]$, $\left.[16,256], \ldots,\left[2^{2^{n}}, 2^{2 n+1}\right], \ldots\right)$, и рассмотрим вначале функцию $g(x)$, которая равна $x^{\frac{3}{2}}$ на нечетных по счету отрезках, и равна $x^{\frac{5}{2}}$ на четных.
Достаточно очевидно, что такая функция растет быстрее, чем $x$, но медленнее, чем $x^{3}$. Главный ее недостаток в том, что она разрывна (в точках $4,16, \ldots$ ) и беда не в том, что разрывна, а в том, что из-за этого она не монотонна.
Однако это легко поправимо. Положим функцию $f(x)$ на $r$-м по счету выделенном отрезке равной $g(x)+A_{r}$, где числа $A_{r}$ подбираются так, чтобы функция стала непрерывной. Конкретно положим $A_{1}=0, A_{2}=4^{\frac{3}{2}}-4^{\frac{5}{2}}=-56$, и т. д.
Получающаяся функция удовлетворяет требованию задачи. Действительно, её график по-прежнему лежит между графиками функций $x^{3 / 2}$ и $x^{5 / 2}$. Покажем, что она несравнима с $x^{2}$. Рассмотрим отрезок $\left[d, d^{2}\right]$, где $d=2^{2^{2 n}}$. тогда $f(d) \leqslant d^{5 / 2}$, откуда $f\left(d^{2}\right) \leqslant$ $d^{5 / 2}+\left(\left(d^{2}\right)^{5 / 2}-d^{3 / 2}\right) \leqslant d^{3}+d^{5 / 2} \leqslant 2\left(d^{2}\right)^{3 / 2}$. Значит, если $[f] \geqslant a$, то $a \geqslant 3 / 2$. Аналогично, рассматривая отрезок $\left[d, d^{2}\right]$, где $d=2^{2^{2 n+1}}$, получаем $f\left(d^{2}\right) \geqslant \frac{1}{2}\left(d^{2}\right)^{5 / 2}$, откуда $[f] \leqslant a$ лишь при $a \leqslant 3 / 2$.
Соответственно, построенная функция и функция $x^{2}$ дают решение задачи В2.
## B4
Годится, например, функция $f(x)=\ln x$.
## B5
Годится, например, функция $f(x)=2^{x}$. В самом деле, тогда $f^{2}=4^{x}=f(2 x)$, и согласно определению, это значит, что рост функций одинаков.
## B6
Ответ. Да, существует.
Достаточно, например, чтобы выполнялось соотношение $\ln f(x)=f(x / 2)$, или, что то же самое, $f(2 x)=\exp (f(x))$. Для этого рассмотрим произвольную возрастающую функцию $f(x)$ на отрезке $[1,2]$ так, чтобы $f(1)=1\left(\frac{1}{\varepsilon}\right)^{2}$ (соответственно, $M(\varepsilon)>\left(\frac{1}{\varepsilon}\right)^{3}$ для куба).
С другой стороны, используя конструкцию из правильных 6 -угольников, приведённую перед условием D1, мы легко убеждаемся в том, что можно покрыть весь круг (с небольшим «заходом наружу») правильными 6 -угольниками. Покрыв эти 6 -угольники кругами, мы получаем (при достаточно малых $\varepsilon$ ) оценку $M(\varepsilon)\left(\frac{1}{\delta}\right)^{2}$. Остальные оценки получаются аналогично.
## C3
Очевидно, что если $\delta$-решетка максимальна, то круги радиуса $\delta$ с центрами в этих точках полностью покрывают фигуру, так что такая решетка является также и $\delta$-сетью, хотя не обязательно максимальной. Во всяком случае, это значит, что $N(\delta) \geqslant M(\delta)$.
С другой стороны, в круге радиуса $\frac{\delta}{2}$ может находиться не более одной точки $\delta$-решетки. Отсюда сразу следует, что если $\varepsilon<\frac{\delta}{2}$, то $M(\varepsilon) \geqslant N(\delta)$.
Таким образом, $M\left(\frac{\varepsilon}{3}\right) \geqslant N(\delta) \geqslant M(\delta)$, откуда и следует утверждение задачи.
## C4
Искомую фигуру можно построить, в частности, как пересечение бесконечного числа фигур. А именно, пусть $\Phi_{1}$ - круг радиуса 2 с центром в начале координат.
Далее, $\Phi_{2}$ есть фигура, вписанная в $\Phi_{1}$. А именно, $\Phi_{2}$ есть объединение восьми кругов радиуса $\frac{1}{2}$ (т.е. в 4 раза меньше) с центрами в точках $\left(0,-\frac{3}{4}\right) ;\left(0,-\frac{1}{4}\right) ;\left(0, \frac{1}{4}\right)\left(0, \frac{3}{4}\right)$, и аналогично еще 4 круга с центрами на оси ординат. (Эти круги частично пересекаются.)
Теперь в каждый из этих кругов мы вписываем 8 кругов радиуса $1 / 8$, расположенных точно так же. Это значит, что их центры лежат на прямых, проходящих через центр очередного круга параллельно одной из осей. Эти 64 круга образуют фигуру $\Phi_{3}$ и т. д. Обозначим через $\Phi$ пересечение всех фигур $\Phi_{i}$.
Очевидно, получившаяся фигура как раз и может быть покрыта либо одним кругом радиуса 2 , либо 8 кругами радиуса $\frac{1}{2}$, и т. д. С другой стороны, все центры кругов радиуса
$2^{1-2 n}$ (их количество равно $2^{3 n}$ ) образуют $2^{-2 n}$-решётку и лежат в нашей фигуре. Отсюда и находим её размерность.
## C5
Ответ. Реализуются все размерности между 1 и 2 . Также размерность может не быть числом. Размерность может быть меньше 1 только для несвязной фигуры.
## C6
Ответ. Размерность спирали может быть различной.
В самом деле, допустим сначала, что длина спирали конечна. Так будет, например, если $f(\varphi)=e^{-\varphi}$ : в этом случае длина каждого следующего витка спирали пропорциональна длине первого витка, так что длина всей спирали равна сумме убывающей геометрической прогрессии.
Пусть $L$ - длина всей спирали.
Поскольку кусок спирали длины $2 r$ заведомо можно покрыть кругом радиуса $r$, очевидно, что всю спираль можно будет покрыть кругами в количестве $\frac{L}{2 r}$, а это означает, что размерность равна 1 .
С другой стороны, если спираль плотно покрывает круг, то ее размерность будет больше 1. Но что значит «плотно»?
Для примера проведем внутри круга, в котором лежит спираль, концентрические окружности радиусов $\frac{1}{2}, \frac{1}{3}, \ldots \frac{1}{n}, \ldots$ Они разделяют круг на концентрические кольца.
Допустим, что в $k$-м кольце лежит $10^{k}$ витков спирали, расположенных равномерно (в этом кольце). Это будет «достаточно плотное» расположение витков, и легко убедиться, что размерность такой спирали больше 1 .
Для доказательства сформулируем общее утверждение:
Утверждение Т. Пусть $\Phi$ - некоторая фигура, размер которой значительно больше ширины витка спирали. Тогда площадь $\Phi$ приблизительно равна произведению ширины витка на длину той части спирали, которая содержится внутри $\Phi$.

Строго говоря, это утверждение весьма неточно, и легко привести к нему контрпримеры. Однако для тех двух фигур, которые нам только и понадобятся, оно верно и достаточно очевидно. Поэтому пока оставим вопрос открытым, и перейдем собственно к доказательству.
Пусть дано некоторое, достаточно малое $\varepsilon$. Выберем $k$ таким образом, чтобы, одновременно выполнялись два неравенства: с одной стороны, $1 / k^{2} \gg \varepsilon \gg 10^{-k}$, с другой - чтобы ширина витка в $k$-том по счету концентрическом кольце (обозначим ее $\delta$ ) была много меньше $\varepsilon$. Поскольку, по предположению, ширина витка меньше чем $1 / 10^{-k}$, то очевидно, что эти условия вполне совместимы.
Пусть $\Phi$ - $k$-тое кольцо (оно задается неравенством $1 / k A.Tolpygo, B.Frenkin, M.Prasolov, I.Bogdanov
## Preface
This cycle is devoted to coverings of arbitrary figures by figures of a prescribed type (usually by disks or balls). The task is to estimate their number, their total area etc.
Naturally, the most difficult are the problems where the precise estimate is required. Fortunately it turns out that basic applications of these problems just don't require an exact answer, and it suffices to know the order of the respective value. The related definitions will be formulated accurately in Part B.
## Part A.
A1.
It is not difficult to cover the unit square by a disk of area $\pi / 2$. However, is it possible to cover the square by several disks of total area less than $\pi / 2$ ? The disks may intersect and get out of the square.
## A2.
Cover the unit square by several disks of the same radius $r$. Let $N(r)$ be the minimal number of disks sufficient for this.
Clearly $r \rightarrow 0$ implies $N(r) \rightarrow \infty$.
Determine the type of tending to infinity for this function. In other words: how fast does it grow?
A3.
The same task but it is required to cover the unit cube by several (possibly intersecting) balls of radius $r$.
## Part B.
In the above problems we assumed that the notion of the "growth rate" is intuitively clear. But below we shall need the precise definition for the "growth of a function".
This definition is not quite obvious, and we will give it a bit later. Before that, having no precise definition for the "growth of a function", we will formulate certain properties of this "growth". Specifically:
In the sequel, we consider only functions $f(x)$ such that
(*) $f(x)$ is defined for all $x$ exceeding some $a$ (usually $a=0$, but for convenience we allow a more general case, for example, functions like $\ln (x-1))$. Moreover we assume that $f(x)$ is positive everywhere (at any rate, for $x>a$ ), non-strictly increases and tends to infinity.
Denote the growth of $f(x)$ by $[f]$. In the sequel, for simplicity we denote the growth of $x^{n}$ by $n$. Thus, $n=\left[x^{n}\right]$.
We require that the growth of a function has the following features:
(1) If for all $x$ exceeding some $b$ ( $b$ not necessarily equals $a$ ) we have $f(x)>g(x)$ then $[f]$ $\geq[g]$.
(2) Let $A, B, C$ be arbitrary positive numbers, and $g(x)=C f(A x+B)$. Then $[g]=[f]$.
(3) This easily implies that a situation is possible when both relations $[f] \geq[g]$ and $[g] \geq$ $[f]$ hold. In this case we set $[f]=[g]$ as well.
If $[f] \geq[g]$ but not $[g] \geq[f]$ then we write $[f]>[g]$.
However we still have not given a definition for growth. The correct answer is that growth is just something subject to the above conditions.
Speaking more formally, we have to consider the class of functions subject to $\left({ }^{*}\right)$ and split it into equivalence classes: $f \sim g$ iff $[f]=[g]$ (according to items (2), (3)).
Each of these classes is called the growth of all functions contained in it. Some of these classes occur to be greater than others, so that we may say that a certain function grows faster (its growth is greater) than another one.
Prove that $1<2$.
## B1.
(A reminder: 1 and 2 are not numbers but the growth of some functions.)
For ordinary numbers there are three alternatives: either $a>\mathrm{b}$, or $a=b$, or $a0$ we have $[f]\alpha$, this square can be covered by disks of total area $\beta$. (Try also to determine whether the square can be covered by disks of total area equal to $\alpha$.)
We split this problem into three different ones. To formulate the first problem, we recall one of the solutions for problem A1: tile the square by the regular hexagons, and cover each hexagon by a disk. The ratio of areas of a hexagon and a disk is $\gamma=2 \pi / 3 \sqrt{3}$. Since the total area of hexagons is larger than the area of the square, the total area of disks $\beta$ will be greater than $\gamma$, but the difference $\beta-\gamma$ can be made arbitrarily small by taking sufficiently small hexagons.
## D1.
Suppose that all the radii of the disks should be equal. Prove that the above construction is optimal, that is, $\alpha=\gamma$.
## D2.
Find $\alpha$ if it is allowed to use the disks of two distinct radii.
D3.
Find $\alpha$ if it is allowed to use the disks of arbitrary radii without any restrictions.
(Warning. The problems are NOT necessarily ordered by the ascending difficulty!)
## D4 - D6.
The same problem but the task is to cover the unit cube by balls which of course may intersect. (Here it suffices to give some good estimates; the precise answer is not known.)
D7.
We need to cover the square with some disks of equal radii so that each point is covered by at least $N$ disks. Prove that there exists some $N$ such that it is possible to make this with total area of the disks smaller than $N \gamma$.
## D8.
Does the statement from $\mathrm{D} 7$ hold for $N=2$ ?
## Part E.
## E1.
A checked sheet of paper lies on a table. Another checked sheet of paper is put on it; the squares on both sheets have the same size but the lines on the second list aren't parallel to those on the first sheet. The upper sheet is transparent, so that we see how its lines dissect some fixed square of the lower sheet.
Determine the maximum and the minimal possible number of parts in a dissection of a lower square.
## E2.
A checked sheet of paper of size $10000 * 10000$ lies on a table. Above it, another checked sheet of paper of size $1000 * 2000$ is put; the squares on both sheets have the same size. The upper sheet is transparent, so that we see how the lines of the lower sheet dissect the squares of the upper one. So, the number of parts of the upper list occurs to be greater than two millions.
Prove that this number is less than ten millions.
## Part F.
Now we return to the problems of the same type as in cycle B. However, we consider a somewhat different construction. We assume the number $\varepsilon$ to be fixed (for instance $\varepsilon=1$ ) but consider an unbounded figure $\Phi$, for instance, the whole plane, a half-plane etc.
Consider an arbitrary point $O$ of the given figure, and construct the figure $\Phi(R)$ consisting of all points of the given figure whose distance from $O$ does not exceed $R$ (in other words, $\Phi(R)$ is the intersection of the given figure with a disk or a ball with center $O$ and radius $R$ ).
Denote by $N(R)$ the number of points in a minimal $\varepsilon$-net in the figure $\Phi(R)$. Let $\chi$ be the growth rate of the function $N(R)$; if this function does not tend to infinity as $R \rightarrow \infty$, we conventionally say that $\chi=0$.
The number $\chi$ is called the volume characteristics of the given figure.
## F1.
Which figures have a zero volume characteristics?
## F2.
Determine the volume characteristic of the following figures: (a) the plane, (b) a halfplane, (c) a strip bounded by parallel lines.
F3.
Prove that the volume characteristic of a figure does not depend on the choice of point $O$.
F4.
Given two positive numbers $\varepsilon$ and $\delta$. Is it true, that the volume characteristics of a figure $\Phi$ calculated using the disks of radius $\varepsilon$ is the same as that calculated using the disks of radius $\varepsilon$ ?
## F5.
Determine the volume characteristic of the interior and of the exterior of a parabola.
F6.
Determine the volume characteristic for each of three parts of the plane obtained by its dissection by the hyperbola $x y=1$.
## F7.
What are possible numerical values of the volume characteristic of a figure (in the plane or in the space)?
F8.
Can the volume characteristic of a figure be not a number (in other words, can $N(R)$ be a function incomparable with some numbers)? For instance, can it be greater than 1, less than 2 and incomparable with $3 / 2$, like the function from Problem B4?
# Problems on coverings and growth functions
A. Tolpygo, B. Frenkin, M. Prasolov, I. Bogdanov
## Solutions
## A1
Answer. Yes, it is possible.
One of the possible ways is to cover a square by a disk of radius $\frac{\sqrt{2}}{2}$, then decrease this radius a bit, and cover the four regions which are uncovered by four small disks.
A different way is presented in the problem statements before problem D1.
## A2, A3
In the first case the growth of $N(r)$ is quadratic (in terms of part $\mathrm{B},[N(r)]=2$ ), while in the second case it is cubic.
## B1
For any $A, B$ and a sufficiently large $x$ we have $x^{2}>A x+B$. Thus $\left[x^{2}\right] \geq[x]$ but it is false that $\left[x^{2}\right] \leq[x]$. This just means that $2>1$.
## B2, B3
For instance, the required function can be constructed as follows. Construct a sequence of rapidly increasing intervals (for example, the intervals $[2,4],[4,16],[16,256], \ldots,\left[2^{2^{n}}, 2^{2^{n+1}}\right] \ldots$ would fit), and consider first the function $g(x)$ equal to $x^{\frac{3}{2}}$ on odd intervals, and equal to $x^{\frac{5}{2}}$ on even intervals.
Obviously, this function grows faster than $x$ but slower than $x^{3}$. The main its defect is its discontinuity (at points $4,16, \ldots$ ) which implies a non-monotonicity. But this can be easily corrected. Define $f(x)$ at the $r$ th interval as $g(x)+A_{r}$, where the constants $A_{r}$ are chosen so that the function becomes continuous. Namely, let $A_{1}=0, A_{2}=4^{\frac{3}{2}}-4^{\frac{5}{2}}=-56$, and so on.
The resulting function satisfies the conditions of the problem. Actually, its graph lies between the graphs of $x$ and $x^{3}$ as before. Now we show that it is incomparable with $x^{2}$. Consider an interval $\left[d, d^{2}\right]$, where $d=2^{2^{2 n}}$. Then we have $f(d) \leq d^{5 / 2}$, and hence $f\left(d^{2}\right) \leq d^{5 / 2}+$ $\left(\left(d^{2}\right)^{5 / 2}-d^{3 / 2}\right) \leq d^{3}+d^{5 / 2} \leq 2\left(d^{2}\right)^{3 / 2}$. Therefore, if $[f] \geq a$, then $a \geq 5 / 2$. Analogously, considering the interval $\left[d, d^{2}\right]$ for $d=2^{2^{2 n+1}}$, we get $f\left(d^{2}\right) \geq \frac{1}{2}\left(d^{2}\right)^{5 / 2}$, wherefore the relation $[f] \leq a$ implies $a \leq 3 / 2$.
Correspondingly, the constructed function and $x^{2}$ provide the solution for Problem B2.
## B4
For instance, the function $f(x)=\ln x$ fits.
## B5
For instance, the function $f(x)=2^{x}$ fits. Indeed, we have $f^{2}=4^{x}=f(2 x)$, and the growth of the functions is the same by the definition.
Answer. Yes, it exists.
It suffices to provide for instance the relation $\ln f(x)=f(x / 2)$, or equivalently $f(2 x)=$ $\exp (f(x))$. To get this, take any increasing function on the interval $[1,2]$ such that $f(1)=1$ i $f(2)=e=\exp (f(1))$; then our relation determines uniquely the function $f \mathrm{n}$ the intervals $[2,4]$, $[4,8]$, and so on. Clearly, we get a desired example.
Answer. $M(\varepsilon)=\left\lceil\frac{a}{2 \varepsilon}\right\rceil, N(\delta)=\left\lfloor\frac{a}{\delta}+1\right\rfloor$.
$\mathrm{C} 2$
The best possible estimate is not required in this problem, so the answer is not unique.
We present one of the possible estimates.
(a) For $M(\varepsilon)$.
Since the disks of radius $\varepsilon$ must cover the whole unit disk, their total area must exceed the area of the unit disk. Hence $M(\varepsilon)>\left(\frac{1}{\varepsilon}\right)^{2}$ (respectively, $M(\varepsilon)>\left(\frac{1}{\varepsilon}\right)^{3}$ for the cube).
On the other hand, using the construction shown in the Problems section above D1, we can easily see that the whole disk (with a minor "overcoming") can be covered by regular hexagons. Covering these hexagons by disks, we arrive (for a sufficiently small $\varepsilon$ ) to the estimate $M(\varepsilon)\left(\frac{1}{\delta}\right)^{2}$. The other estimates are obtained similarly to the ones above.
## C3
Clearly, a maximal $\delta$-lattice is also a $\delta$-net (the respective disks cover the whole figure). On the other hand, a disk of radius $\frac{\delta}{2}$ can contain not more than one point of a $\delta$-lattice. This immediately implies that if $\varepsilon<\frac{\delta}{2}$ then $M(\varepsilon) \geq N(\delta)$.
Thus $M\left(\frac{\varepsilon}{3}\right) \geq N(\delta) \geq M(\delta)$, and this implies the statement of the problem.
## $\mathrm{C} 4$
We construct the required figure as an intersection of an infinite number of figures. Actually, let $\Phi_{1}$ be the disk of radius 2 with the center at the origin. Furthermore, inscribe a figure $\Phi_{2}$ into $\Phi_{1}$ as follows. Let $\Phi_{2}$ be the union of 8 disks of radius $\frac{1}{2}$ (that is, 4 times less) with centers at $\left(0,-\frac{3}{4}\right) ;\left(0,-\frac{1}{4}\right) ;\left(0, \frac{1}{4}\right) ;\left(0, \frac{3}{4}\right)$, and of 4 similar disks with centers on the $y$-axis. (Some of these disks do intersect.)
Next, into each of these disks we inscribe 8 disks of radius $\frac{1}{8}$ arranged similarly. Namely, their centers lie on lines passing through the center of the disk in question and parallel to one of the coordinate axes. These 64 disks form the figure $\Phi_{3}$. The figures $\Phi_{4}, \Phi_{5}, \ldots$ are constructed similarly; let $\Phi$ be the intersection of all these figures.
Obviously the resulting figure can be covered either by a single disk of radius 2 , or by 8 disks of radius $\frac{1}{2}$, or so on. On the other hand, all the centers of constructed disks of radius $2^{1-2 n}$ form a $2^{-2 n}$-lattice in $\Phi$, and there are $2^{3 n}$ such centers. This determines the dimension of $\Phi$.
## C5
Answer. The dimension of a (connected) figure can attain each value from $[1,2]$ (on the plane). The dimension can be incomparable with some numbers as well. It can be less than 1 only for a non-connected figure.
## C6
Answer. The dimensions of different helices may be different.
Let us assume first that a helix has finite length. (This is true, for example, when $f(\varphi)=e^{-\varphi}$, since in this case the length of each turn is proportional to the length of the first one, thus the length of the whole helix is the sum of a descending geometrical progression.)
Let $L$ be the length of the whole helix. Since a piece of the helix having length $2 r$ obviously can be covered by a disk of radius $r$, the whole helix can be covered by $\frac{L}{2 r}$ disks, which means that the dimension is equal to 1 .
On the other hand, if a helix covers a disk densely, then its dimension is greater than 1. We are left to define what is "densely".
To give an example, let us draw concentrical circles of radii $\frac{1}{2}, \frac{1}{3}, \ldots \frac{1}{n}, \ldots$ in the unit disk. They dissect the disk into concentrical rings. Suppose that the $k$ th ring contains $10^{k}$ turns of the helix arranged uniformly (in this ring). We claim that this is a "sufficiently dense" arrangement of turns, that is, the dimension of this helix is greater than 1 .
For the proof we formulate a general
Proposition T. Let $\Phi$ be some figure in a ring, the size of $\Phi$ being much larger than the width of a helix turn. Then the area of $\Phi$ is approximately equal to the product of the turn width and the total length of the parts of the helix lying inside $\Phi$.

Strictly speaking, this proposition is incorrect; one can easily find some counterexamples. But we will apply it only to the disks and the rings, for which it holds, and can be easily proved.
So, we turn to the dimension of our helix. Consider some small $\varepsilon$. Choose integer $k$ such that $1 / k^{2} \gg \varepsilon \gg 10^{-k}$, that is, $\varepsilon$ much smaller that the width of $k$ th ring, but the turn width $\delta$ in this ring is much smaller than $\varepsilon$. Note that such $k$ can be found if $\varepsilon$ is small enough. Denote by $\Phi$ the $k$ the ring.
Now, apply the Proposition T to $\Phi$ and to each disk of radius $\varepsilon$ in the covering. Let $L$ be the length of the piece of the helix lying inside $\Phi$; then the area of $\Phi$ is $\pi\left(\frac{1}{k^{2}}-\frac{1}{(k+1)^{2}}\right) \approx L \delta$. On the other hand, each disk will contain the pieces of helix with total length $\approx \pi \varepsilon^{2} / \delta$. This means that the number of disks covering the part of a helix inside $\Phi$ is (almost) at least the ratio of these two values. In other words, these disks have almost the area sufficient to cover the whole $\Phi$.
From this, we cannot make a conclusion that the dimension of a helix is 2 , though $\Phi$ has a dimension 2: actually, we choose different figures $\Phi$ for different $\varepsilon$. But, since $k$ grows much slower than $\varepsilon$, one can easily see that the dimension of $\Phi$ is definitely greater than 1 , and this was exactly our goal.
## D1
Lemma. Given $N$ polygons lying inside unit disks such that each polygon contains the center of its disk, and the total amount of vertices of polygons does not exceed $6 N$. Then the total area of polygons is at most $N \cdot \frac{3 \sqrt{3}}{2}$ (so, this estimate is sharp when the polygons are the regular hexagons).
Proof. Dissect each polygon into triangles by the radii of its disk. The doubled area of each triangle is not greater than $\sin \alpha$, where $\alpha$ is the angle at the central vertex. So, the doubled total area of given polygons equals to $\sin \alpha_{1}+\sin \alpha_{2}+\ldots+\sin \alpha_{n}$, where $\alpha_{i}$ are the corresponding angles; we have $\alpha_{1}+\cdots+\alpha_{n}=\pi N$, and $n \leq 6 N$. Moreover, adding some zero angles one can assume that $n=6 N$. Finally, one can notice that the graph of the function $\sin x$ on $[0, \pi]$ is concave, so the function attains its maximal value when $\alpha_{i}=\frac{\pi}{3}$. So we get the maximal area if all polygons are regular hexagons.
Solution of problem D1. We will prove the statement for an arbitrary polygon $T$ with angles not exceeding $2 \pi / 3$. Suppose that $T$ is covered by $N$ equal disks. Divide this unit square into convex polygons by the following rule: for each disk center, its polygon contains all the points such that this center is the closest center to them (see Figure). The discs are equal, therefore the sides of polygons are the parts of common chords of our disks. Since all points are covered, each polygon lies in a corresponding disk. Moreover, each polygon obviously contains the center of its disk.

Now we estimate the average value of the angle of our polygon. The average angle in each vertex of our dissection is not greater than $2 \pi / 3$ (it can be easily seen separately for the vertices inside $T$, on the sides of $T$, and for the vertices of $T$ - exactly here we use the estimates for the angles of $T$ ). It follows easily that the total number of vertices of the polygons does not exceed $6 N$. Hence by the Lemma the total area of the polygons (which is 1 ) is at most $N \frac{3 \sqrt{3}}{2}$, and the total area of the disks is $N \cdot \pi \varepsilon^{2} \geq \frac{2}{3 \sqrt{3} \varepsilon^{2}}=\gamma$, as desired.
Remark. If $T$ is a regular hexagon, then the best way to cover it is to use exactly one disk; all other ways are strictly worse.
With some minimal changes, the proof above is valid for any polygon with at most six sides.
## D2
The optimal construction is the following one.
We cover a part of the square by the disks of a larger radius, and the remaining part will be covered by the smaller disks using the method from D1 (that is, using a covering by the small hexagons). Naturally, to reach an (almost) optimal configuration by this method, we should take the radii $r_{1}$ and $r_{2}$ such that $1 \gg r_{1} \gg r_{2}$.
Cover a unit square by regular hexagon lattice. Denote a side of a hexagon by $a$ and put on each hexagon a disk of radius $r_{1}$. We wish these disks to intersect but not to cover the whole square; these conditions rewrite as $\frac{\sqrt{3}}{4} a2$, we get the desired covering.
## D4
Naturally, as in D1, the radii should be small enough, and the main queation is about the arrangement of the ball centers. By analogy with D1, it is presumably better to take some "dense" packing of balls and try to expand it. We present a dense packing of non-intersecting balls; then it remains to increase their radii to obtain the covering of the whole cube.
Let the centers of the balls in a "first layer" lie in the vertices of a triangular lattice in one plane. The next layer will contain the balls also forming the triangular lattice; moreover, each center in the second layer will form a regular tetrahedron with three centers from the first layer.
The third layer is constructed in a same manner from the second one, and so on. Notice here that, having made two first layers, one can construct the third one in two (essentially different) ways; but the density of both coverings will be the same. To visualize his, we remark that if one starts with the first layer in a form of the regular triangle with side $n$, then we can obtain a pyramid with $n$ layers.
It remains to calculate the ratio of the radii of the original and the expanded balls. We mention (without a proof) that this ratio is equal to the ratio of the circumdiameter and the side of a regular octahedron, or $\sqrt{2}$.

D5
It seems that the optimal configuration has a form similar to that in D2. That is, we take the configuration from the previous problem, decrease the radii a bit, and cover the remainng spaceby the balls of a smaller radius.
## D6
The answer for D6 is the same as for D3. The solution is similar due to the fact that the volume of the ball is greater than half of the volume of the corresponding cube (in 4-dimensional case this is not true, hence one should upgrade the proof a bit).
Remark. Notice that in D4-D5 we provide only some plausible reasonings on an optimal example; conversely, in D6 we show an outline of the full solution.
## D7
Cover the square as in D1. Part of the square which is covered twice consists of equal figures; we call such a figure $a$ lens. Cover a hexagon by lenses as in the figure: a hexagon is covered by rhombs, and each rhomb can be covered by a lens.

Denote by $N$ a number of rhombs in this covering. Each rhomb can be obtain from the initial one by a shift. So, let us shift the covering $N$ times to obtain a square covered $N+1$ times. The correcting of the boundary effect is left to the reader.
## D8
Jury does not know the solution.
## E1
We start with two elementary lemmas.
Lemma 1. The number of dissection parts equals to $1+a+b+c$ where $a, b$ respectively are numbers of horizontal and vertical lines which intersect the lower square and $c$ is a number of vertices of the upper sheet which lies in the lower square.
Proof by induction is easy: delete all the horizontal and vertical lines, and then draw them one by one.
Lemma 2. If a base and the corresponding altitude of triangle are at least 1 and this triangle lies in the square then its base coincide with the side of the square.
Proof follows from the fact that if a unit square contains a triangle with area $\geq 1 / 2$, then this area is $1 / 2$; moreover, in this case the triangle and the square have a common side.
Answer. The number of parts equals to 4,5 or 6 . Examples are shown in the figure.

Solution. First, notice that each projection of a unit square onto some line has a length between 1 and $\sqrt{2}$; this means that $a$ and $b$ can be only 1 or 2 . Moreover, it is easy to see than $0 \leq c \leq 2$. This means, by Lemma 1 , that the desired number lies between 3 and 7 . We are left to show that the border cases are impossible.
Suppose that the number of parts equales to 3 , thus $a=b=1$ and $c=0$. This means that the lower square is covered by three upper squares (see the left figure below); this contradicts Lemma 2.
The case of 7 parts is similar in the sense that it follows from the same Lemma 2 (see the right figure below).

E2
We will find an approximate estimate for the number of pieces; the further details are left to the reader. We will assume that the upper rectangle is drawn on the plane; so we erase it and then we reconstruct it in several steps. Now we assume that the sides of top (small) rectangle are oriented vertically and horizontally, while the lower rectangle is sloped.
First, we draw the boundary of the upper rectangle; then it will be split into a bit more than two millions parts (almost all of them - except those on the border - are the unit squares). Next, we draw the vertical and horizontal lines (of the upper rectangle) one by one. Each line increases the number of parts by the number of its intersections with other lined (drawn up to this moment). Hence, the total increment will equal to the total number of the points of intersection (where the 3- and 4-fold points are considered in an appropriate manner). So, we are to estimate this number. There are not more than two millions points of intersection of horizontal and vertical lines with each other. The remaining points are the points of intersection of lines from different sheets.
Let $\alpha$ be the angle between a horizontal and (some) sloped line. Then a horizontal line (of length 1000) intersects approximately $1000 \sin \alpha$ lower lines of this type, and approximately $1000 \cos \alpha$ sloped lines of another type, so all horizontal lines add approximately $2000 \cdot 1000(\sin \alpha+$ $\cos \alpha$ ) points. Similarly, the vertical lines add approximately $1000 \cdot 2000(\sin \alpha+\cos \alpha)$ points. Note that the expression $\sin \alpha+\cos \alpha$ reaches its maximum value $\sqrt{2}$ when $\alpha=\pi / 4$, so the
obtained bound for the number of parts is approximately $2 \cdot 10^{6}+2 \cdot 10^{6}+2 \cdot 2 \cdot 10^{6} \sqrt{2} \approx 2 \cdot 4.83 \cdot 10^{6}$ parts. It is left to see that the errors in our calculations sum up at less than one million.
## F1
Answer. Figure has a zero volume characteristics if and only if it is bounded.
## F2
Answer. A volume characteristics of the plane and the halfplane equals to 2 . A volume characteristics of the strip equals to 1 .
## F3
Denote by $N_{1}(R)$ and $N_{2}(R)$ functions corresponding to points $O_{1}$ and $O_{2}$. Then $N_{1}(R+$ $\left.O_{1} O_{2}\right) \geq N_{2}(R)$ and $N_{1}(R) \leq N_{2}\left(R+O_{1} O_{2}\right)$. Thus $\left[N_{1}\right]=\left[N_{2}\right]$.
## F4
Answer. Yes.
Assume that $\varepsilon>\delta$. Then each disk of radius $\delta$ can be covered by a disk of radius $\varepsilon$. So $N_{\delta}(R) \geq N_{\varepsilon}(R)$. (Here, $N_{\varepsilon}$ and $N_{\delta}$ denote the two functions defined by the disks of the corresponding radii.)
Conversely, each disk of radius $\epsilon$ can be covered by $A$ disks of radius $\delta$ (for some constant $A$ ). Then $N_{\delta}(R) \leq A \cdot N_{\varepsilon}(R)$, quod erat demonstrandum.
## F5
Answer. $\frac{3}{2}, 2$.
## F6
Answer. A volume characteristics of all of them equals to 2 .
## F7, F8
If a figure is unbounded and connected then its volume characteristics is at least 1. Volume characteristics of a figure is at most volume characteristics of the whole plane/space, i.e. 2 or 3 . These are the only restrictions.
## Случайные блуждания и электрические цепи
Дмитрий Баранов, Михаил Скопенков, Алексей Устинов
Цель нашего проекта - доказать следующую теорему и изучить смежные вопросы.
Теорема Пойа. (а) Если человек случайным образом перемещается по 2-мерной решетке, то он вернется в начальную точку с вероятностью 1.
(b) Если же он перемещается по 3-мерной решетке, то вероятность его возврата в начальную точку строго меньше 1.
Точные формулировки даны ниже. Предлагаемый подход к доказательству основан на физической интерпретации. Никаких специальных знаний физики не требуется.

2-мерная решетка
3-мерная решетка
## 1. Одномерные блуждания
Мы сначала сформулируем задачу и лишь потом дадим необходимые определения.
1.1. Человек ходит по отрезку улицы, состоящему из 5 кварталов. Начав свой путь на границе кварталов в точке $x$, он с вероятностью $1 / 2$ перемещается на один квартал влево и с вероятностью $1 / 2$ - на один квартал вправо. Подойдя к границе кварталов, он опять выбирает направление движения случайным образом. Если он оказывается в точке 5 (его дом) или в точке 0 (бар), то он прекращает движение: см Рис 1.

Рис. 1: Случайное движение по улице; см. задачу 1.1.
(A)* Напишите компьютерную программу, моделирующую движение этого человека. Запустите ее много раз и определите процент числа случаев, в которых он приходит домой. Вы можете использовать этот способ для угадывания ответов в последующих задачах.
E-mail address: dimbaranov@mail.ru, skopenkov@rambler.ru, ustinov.alexey@gmail.com
(B) Пусть $P_{T}(x)$ - вероятность того, что человек, начавший свое движение в точке $x$ и сделавший не более $T$ ходов, оказался дома. Заполните следующую таблицу десятичными дробями с точностью до сотых.
Таблица 1: Вероятности $P_{T}(x)$ для малых $T$
| $T^{x}$ | 0 | 1 | 2 | 3 | 4 | 5 |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| 1 | 0.00 | 0.00 | 0.00 | 0.00 | 0.50 | 1.00 |
| 2 | | | | | | |
| 3 | | | | | | |
| 4 | | | | | | |
(C) Найдите вероятность $P(x)$ того, что человек дойдет до дома через какое-то количество ходов.
Определение. (А) Предположим, что у некоторого эксперимента имеется $n$ равновероятных исходов, и событие $X$ происходит ровно в $m$ из них. Тогда вероятностью события $X$ называется число $P_{1}(X):=m / n$.
Например, вероятность выпадения орла при бросании монеты $-1 / 2$; вероятность выпадения 6 очков на кубике $-1 / 6$; вероятность пойти направо по нашей улице $1 / 2$.
(B) Теперь предположим, что событие $X$ зависит от последовательности таких экспериментов. Последовательность из $T$ экспериментов имеет $n^{T}$ возможных исходов. Предположим, что событие $X$ происходит ровно при $m_{T}$ исходов из них. Тогда вероятностью события $X$ называется число $P_{T}(X):=m_{T} / n^{T}$.
Например, есть ровно 4 возможных исхода при бросании монеты 2 раза:
| 1 -ый бросок
2 -ой бросок | орел
орел | орел
решка | решка
орел | решка
решка |
| :---: | :---: | :---: | :---: | :---: |
Пусть событие $X$ состоит в появлении решки хотя бы один раз. Событие $X$ происходит в 3 случаях из 4 возможных. Поэтому вероятность события $X$ есть $P_{2}(X)=3 / 4$.
Вероятность получения более 10 очков при бросании двух кубиков $-1 / 12$, так как это событие происходит в ровно 3 случаях $(5+6,6+5$ или $6+6$ ) из 36 возможных. Вероятность того, что человек, начавший с точки 3 , сдвинется вправо два раза подряд составляет $1 / 4$.
(C) Пусть теперь событие $X$ зависит от бесконечной последовательности таких экспериментов. Мы будем называть число $P(X)$ вероятностъю события $X$, если вероятности $P_{T}(X)$ стремятся к числу $P(X)$ при стремлении $T$ к бесконечности ${ }^{1}$.
Например, вероятность выпадения решки хотя бы один раз в бесконечной серии бросков составляет $P(X)=1$, так как $P_{T}(X)=1-1 / 2^{T}$ стремится к 1 при стремлении $T$ к бесконечности.
То, что вероятности $P(X)$ существуют для всех событий $X$, рассматриваемых в проекте, можно использовать без доказательства.[^2]
1.2. Петя и Паша играют на монетки; всего у них есть 5 монеток; в каждом раунде Петя выигрывает у Паши одну монетку с вероятностью $1 / 2$ и проигрывает с вероятностью $1 / 2$; они играют до тех пор, пока у Пети не станет 0 монеток (он проиграл) или 5 (он выиграл все монеты Паши). Найдите вероятность $P(x)$ того, что Петя выиграет, начав игру с $x$ монетками.
1.3. Предположим, нашего "путешественника" сносит в одну сторону; точнее, пусть он каждый раз перемещается вправо с вероятностью $p$ и влево с вероятностью $q=1-p$. Найдите вероятности $P(x)$ в этом случае.
1.4. Предположим, что вы играете на деньги; сначала у вас 20 монет, а у вашего соперника - 50 монет; в каждой игре вы выигрываете одну монету с веротностью 0.45 и проигрываете с вероятностью 0.55 ; игра продолжается до тех пор, пока у кого-либо не закончатся деньги. Найдите вероятность своего разорения.
Определение. Электрическая цепь - это связный конечный граф, у которого каждому ребру $x y$ приписано положительное вещественное число, называемое его проводимостъю ${ }^{2} C(x y)$, и задано два непересекающихся выделенных множества вершин $(P$ и $N$ ). Вершины из множества $N$ соединены с отрицательным полюсом батарейки и землей, а вершины из множества $P$ - с положительным; см рисунок 2.
Потенииалы вершин $v(x)$ определяются следующими аксиомами:
1. Граничные условия. Если $x \in N$, то $v(x)=0$. Если $x \in P$, то $v(x)=1$.
2. Закон Кирхгофа. Если $x \notin P \cup N$, то $\sum_{x y} C(x y)(v(x)-v(y))=0$, где суммирование ведется по всем ребрам $x y$, содержащим вершину $x$.
Число $i(x y):=C(x y)(v(x)-v(y))$ называется током, идущим по ребру $x y ; i(x):=$ $\sum_{x y} i(x y)$ - током, втекающим в цепь через вершину $x$ (так, $i(x)=0$ для каждого $x \notin P \cup N$ по аксиоме 2$) ; C:=\sum_{x \in P} i(x)$ называется эффективной проводимостью цепи между множествами $P$ и $N ; Q:=\sum_{x y} C(x y)(v(x)-v(y))^{2}$, где суммирование ведется по всем ребрам цепи, называется тепловой мощностью цепи.
1.5. Одинаковые резисторы соединены последовательно и подключены к батарейке в 1 вольт как показано на рисунке 2 . Найдите потенциалы $v(x)$ в точках $x=0,1,2,3,4,5$. Вы можете использовать программы, эмулирующие электрические цепи, для угадывания ответа.

Рис. 2: Электрическая цепь; см. задачу 1.3 .[^3]
1.6. Рассмотрим цепь с вершинами $0,1, \ldots, n$, ребрами $01,12, \ldots,(n-1) n$ единичной проводимости, и выделенными множествами $N=\{0\}, P=\{n\}$.
(A) Принцип максимума. Функция $v(x)$, удовлетворяющая аксиоме 2 достигает своего максимума и минимума в вершинах из множества $P \cup N$.
(B) Единственность. Если $v(x)$ и $u(x)$ - две функции, удовлетворяющие аксиомам $1-2$, то $v(x)=u(x)$ для всех $x$.
(C) Найдите потенциалы $v(x)$ и эффективную проводимость данной цепи. $\mathrm{K}$ чему они стремятся при стремлении числа $n$ к бесконечности?
1.7. Сформулируйте и докажите 1 -мерную теорему Пойа.
## 2. Двумерные блуждания
2.1. Рассмотрим город, схема которого приведена на рисунке 3 слева. Отрезки обозначают улицы. Пути отхода помечены буквой $E$, а буквой $P$ помечены точки, занятые полицией. Найдите с точностью до сотых вероятность $P(x)$ того, что начав свой путь в точке $x$, человек убежит, а не попадет в руки полиции. Из точки $x=(a, b)$ он перемещается в каждую из точек $(a+1, b),(a-1, b),(a, b+1),(a, b-1)$ с вероятностью $1 / 4$. Если он достигает одной из точек $E$ или $P$, то его передвижения заканчиваются. 2.2. Найдите потенциалы $v(x)$ в цепи из единичных резисторов на рисунке 3 справа.

Рис. 3: Случайное движение по городу и электрическая цепь; см. задачи 2.1 и 2.2 .
2.3. Паук перемещается случайным образом по ребрам
(А) куба;
(B) октаэдра;
(C) додекаэдра;
(D) икосаэдра;
если он начинает движение в точке $a$, то какова вероятность того, что он достигнет противоположной вершины $h$ быстрее, чем вернется в начальную вершину $a$; см. рисунок 4 слева?

Рис. 4: Случайное блуждание по кубу и электрическая цепь; см. задачи 2.3(А) и $2.5(\mathrm{~A})$.
2.4. Следующие преобразования сохраняют эффективную проводимость цепи:
(A) замена двух резисторов, соединенных последовательно, на один резистор проводимости $1 /\left(\frac{1}{C_{1}}+\frac{1}{C_{2}}\right)$; см. рисунок 5 слева;
(B) замена двух параллельно соединенных резисторов на один резистор с проводимостью $C_{1}+C_{2}$; см. рисунок 5 справа;
(C) объединение двух вершин с одинаковыми потенциалами в одну новую вершину.

Рис. 5: Последовательное и параллельное соединение; см. задачу 2.4.
2.5. Найдите эффективную проводимость между
(1) противоположными вершинами; (2)* смежными вершинами;
(А) куба;
(B) октаэдра;
(C) додекаэдра;
(D) икосаэдра;
с ребрами единичной проводимости; см рисунок 4 справа.
2.6. Пьяный турист выходит из отеля и перемещается случайным образом по улицам Парижа, схема центра которого приведена на рисунке 6 слева. Найдите вероятность того, что он дойдет до Триумфальной арки до того, как доберется до окраины города.

Рис. 6: Туристическая карта Парижа и решеточная цепь; см. задачи 2.6 и 2.7 .
2.7. Проводимость между вершинами (A)* $a$ и $b ;(\mathbf{B})^{* *} a$ и $c$; двумерной решетки из единичных резисторов равна 2 и $\pi / 2$, соответственно; см. Рис. 6 справа.
Для 2-мерной решетки в определении потенциала мы добавляем еще одну аксиому:
3. $v(x)$ стремится к $1 / 2$ при стремлении расстояния между $x$ и некоторой фиксированной вершиной к бесконечности.
Разрешается пользоваться без доказательства существованием функции $v(x)$, удовлетворяющей аксиомам $1-3$.
2.8. Закон монотонности. Разрезание каких-либо ребер цепи может только уменьшить эффективную проводимость между данными вершинами; см. рисунок 7 слева. Объединение каких-либо вершин в одну вершину может только увеличить эффективную проводимость между данными вершинами; см. рисунок 7 в центре.
2.9. (А) Докажите, что эффективная проводимость между центром и границей квадратной решетки $4 \times 4$ из единичных резисторов меньше 3 ; см рисунок 7 справа.
(B) $\mathrm{K}$ чему стремится проводимость между центром и границей квадратной решетки $2 n \times 2 n$ из единичных резисторов при стремлении $n$ к бесконечности?

Разрезание

Соединение

Рис. 7: Разрезание, соединение, и квадратная решетка $4 \times 4$; см. задачи 2.8 и 2.9 .
2.10. Докажите теорему Пойа для 2-мерной решетки.
## 3. Сопротивление правильных графов
Задачи данного раздела могут помочь при решении задач предыдущего.
Функция $v(x)$ на вершинах электрической цепи называется гармонической, если она удовлетворяет аксиоме 2 из определения электрической цепи. Вершины из множества $P \cup N$ назовем граничными, остальные - внутренними.
3.1. (A) Принцип суперпозиции. Если функции $u(x)$ и $v(x)$ - гармонические и $a, b \in$ $\mathbb{R}$, то и функция $a u(x)+b v(x)$ - гармоническая.
(B) Принцип максимума. Гармоническая на конечной электрической цепи функция принимает свое наибольшее и наименьшее значения на граничных вершинах.
(C) Единственность. Если $u(x)$ и $v(x)$ - две гармонические функции, совпадающие во всех граничных вершинах конечной электрической цепи, то $u(x)=v(x)$ во всех вершинах $x$ графа.
(D) Человек, случайно блуждающий по конечному городу, с вероятностью 1 побывает на всех перекрестках его улиц.
3.2. (А)* Альтернатива Фредголъма. Пусть имеется система линейных уравнений
$$
\left\{\begin{array}{c}
a_{1,1} x_{1}+\ldots+a_{1, n} x_{n}=b_{1} \\
\ldots \ldots \ldots \ldots \ldots \ldots \\
a_{n, 1} x_{1}+\ldots+a_{n, n} x_{n}=b_{n}
\end{array}\right.
$$
в которой число уравнений равно числу неизвестных. Докажите, что всегда выполняется одна из следующих возможностей:
1. для любых $b_{1}, \ldots, b_{n}$ система имеет ровно одно решение (в частности, при $b_{1}=$ $\ldots=b_{n}=0$ существует только нулевое решение);
2. для некоторых $b_{1}, \ldots, b_{n}$ система неразрешима, а для некоторых (в том числе нулевых) имеет бесконечно много решений.
(B)* Задача Дирихле. Докажите, что для любой конечной электрической цепи существует функция $v(x)$, удовлетворяющая аксиомам $1-2$.
3.3. (A) Вариационный принцип. Пусть $v(x)$ - произвольная функция на вершинах конечной электрической цепи, удовлетворяющая аксиоме 1, но не обязательно аксиоме 2. Занумеруем вершины графа числами $1, \ldots, n$ и пусть $1, \ldots, k$ - внутренние вершины. Обозначим $v_{1}:=v(1), v_{2}:=v(2), \ldots, v_{n}=v(n)$. Будем рассматривать $v_{1}, \ldots, v_{k}$ как переменные. Рассмотрим тепловую мощность $Q\left(v_{1}, \ldots, v_{k}\right):=$ $\sum_{x y} C(x y)\left(v_{x}-v_{y}\right)^{2}$ как функцию переменных $v_{1}, \ldots, v_{k}$. Докажите, что функция $Q\left(v_{1}, \ldots, v_{k}\right)$ принимает наименьшее значение, когда функция $v(x)$ - гармоническая. (B)* Докажите, что для функции $Q\left(v_{1}, \ldots, v_{k}\right)$ существует ровно один набор значений $v_{1}, \ldots, v_{k}$, при котором она достигает своего минимального значения. Используя это, дайте второе доказательство того факта, что для конечной электрической цепи функция $v(x)$, заданная аксиомами $1-2$, существует и единственна.
(C) Закон сохранения энергии. Докажите, что минимальное значение величины $Q\left(v_{1}, \ldots, v\right.$ численно равно эффективной проводимости $C$.
(D) Принцип монотонности. Докажите, что если в цепи одну проводимость увеличить, то эффективная проводимость не уменьшится.
(E) Из 2-мерной решетки выбросили произвольное множество ребер. Докажите, что по-прежнему случайное блуждание с вероятностью 1 вернется в исходную вершину.
Предположим, что имеется граф $\Gamma$, у которого сопротивление каждого ребра равно 1. Возьмем в графе $\Gamma$ два смежных ребра $A B$ и $A C$. Эти ребра назовем эквивалентными, если существует перестановка вершин графа, переводящая соединенные ребром вершины в соединенные ребром вершины, при которой $A$ переходит в $A$ и $B-$ в $C$. Вершину графа назовем центром симметрии графа $\Gamma$, если все ребра, содержащие ее, эквивалентны. Граф Г называется правильным, если все его вершины - центры симметрии графа.
Примеры правильных графов: правильные многогранники любой размерности; правильные решетки на евклидовой плоскости, плоскости Лобачевского и их многомерных аналогах; симметричные решетки на торе и т. п. Нетривиальный пример: граф ромбододекаэдра. Это многогранник, который получается, если к каждой грани куба приставить по четырехугольной пирамиде так, что все треугольники, граничащие по ребрам куба, сольются в ромбы. Поверхность ромбододекаэдра состоит из 12 ромбов. Он нетривиален тем, что его вершины имеют разную степень (3 и 4).
3.4. (A) Пусть правильный граф содержит $n$ вершин, $A_{1}$ и $A_{2}$ - соседние вершины степеней $k_{1}$ и $k_{2}$, соответственно. Докажите, что сопротивление между ними равно
$$
\left(\frac{1}{k_{1}}+\frac{1}{k_{2}}\right)\left(1-\frac{1}{n}\right)
$$
(B) Если же взять 2 -мерную решетку, то $1 / n$ в последней формуле нужно заменить нулем.
3.5. $\mathrm{K}$ двум соседним вершинам проволочного (А) икосаэдра; (В) додекаэдра; подвели напряжение так, что по соединяющему их ребру потек ток $I$. Какой при этом будет течь ток по диаметрально противоположному ребру?
3.6. * Докажите, что суммы $D_{n}=\sum_{k=0}^{n} \frac{1}{C_{n}^{k}}$ и $F_{n}=\frac{n+1}{2^{n+1}} \sum_{k=1}^{n+1} \frac{2^{k}}{k}$ равны друг другу. Как эти суммы связаны с сопротивлением многомерного куба? В качестве следствия получите, что порядок вхождения двойки в число $\sum_{k=1}^{n} \frac{2^{k}}{k}$ стремится к бесконечности с ростом $n$.
## 4. Трехмерные блуждания

Рис. 8: (Слева) бинарное дерево глубины 3; (в центре) модифицированное бинарное дерево глубины 3; (справа) разрешенные пересечения ребер в этом дереве; см. задачи 4.1, 4.2 и 4.4.
4.1. Найдите сопротивление бинарного дерева глубины (А) 3; (B) 2010, составленного из единичных резисторов (см. рис. 8 слева).
4.2. Найдите сопротивление модифицированного бинарного и троичного деревьев глубины 2010 , в которых каждый резистор на $k$-ом уровне заменяется на $2^{k}$ последовательно соединенных единичных резисторов (см. рис. 8 в центре).
4.3. Какие из деревьев, упомянутых (А) в задаче 4.1; (B) в задаче 4.2; можно вырезать из трехмерной решетки?
4.4. А если разрешаются пересечения (см. рис. 8 справа) ребер на равном расстоянии от корня?
4.5. Докажите теорему Пойа для 3 -мерной решетки.
## 5. Сопротивление кольца*
Во всех задачах этого раздела будет фигурировать квадратная металлическая сетка. Будет предполагаться, что все ее узлы - это точки двумерной целочисленной решетки. Соединены друг с другом только соседние узлы (расстояние между которыми равно единице). Сопротивление ребра между любыми соседними узлами также считается равным единице.
5.1. Источник тока подключается к узлам сетки с координатами $(0,0)$ и $(1,0)$. Докажите, что в узлах $(2,2)$ и $(3,2)$ будут одинаковые потенциалы (см. замечание в задаче 3.4(В)).
5.2. Из сетки вырезан квадрат размерами $n \times n$, граничные точки которого соединены шиной с нулевым сопротивлением. Докажите, что сопротивление между любым узлом квадрата и его границей не превосходит $\sqrt{n}$.
5.3. В условиях задачи 5.2 источник тока подключен к внутреннему узлу квадрата и к границе. Потенциал на границе равен нулю. Докажите, что если источник подает ток $\varepsilon$, то в каждой из точек квадрата потенциал не превосходит $\varepsilon \sqrt{n}$.
5.4. Для функций, заданных в узлах целочисленной решетки определим
$$
\Delta f(x, y):=f(x-1, y)+f(x+1, y)+f(x, y-1)+f(x, y+1)-4 f(x, y)
$$
Пусть $r(x, y)=\sqrt{x^{2}+y^{2}}$. Докажите, что для функции $f(x, y):=\ln r(x, y)$ (при $r(x, y) \geq 2$ ) выполнено $\Delta f(x, y)=O\left(\frac{1}{r^{4}(x, y)}\right)$.
Здесь и далее для двух функций $A$ и $B$ запись $A=O(B)$ означает, что для некоторой положительной константы $c$ всегда выполняется неравенство $|A| \leq c B$.
5.5. Из металлической сетки вырезано кольцо с внутренним радиусом $r_{1} n$ и внешним $-r_{2} n$ (центры обоих кругов - в начале координат). Если некоторое ребро разрезано, то сопротивление оставшегося куска пропорционально его длине. На внутренний контур кольца подается напряжение $\ln r_{1}$, а на внешний $-\ln r_{2}$. Докажите, что в каждой точке кольца $(x, y)$ потенциал имеет вид $U_{n}(x, y)=\ln r(x, y)+O\left(\frac{1}{n^{3 / 2}}\right)$.
5.6. С помощью равенства $\operatorname{arctg} x=x+O\left(x^{3}\right)$ докажите, что при $0 \leq y3 r_{1} / 2$ (достаточное для того, чтобы квадрат, описанный около внутреннего круга, целиком содержался бы во внешнем). Докажите, что между внутренним и внешним контурами течет ток $2 \pi+O\left(\frac{1}{\sqrt{n}}\right)$. Выведите отсюда формулу для сопротивления кольца
$$
R\left(r_{1} n, r_{2} n\right)=\frac{1}{2 \pi} \ln \frac{r_{2}}{r_{1}}+O\left(\frac{1}{\sqrt{n}}\right)
$$
5.9. Докажите, что формула (1) выполняется и без дополнительного ограничения $r_{2}>$ $3 r_{1} / 2$.
5.10. С помощью равенства (1) уточните оценки в задачах 5.2, 5.3 и докажите формулы с более точными остаточными членами в задачах 5.5 и 5.8 :
$$
U(x, y)=\ln r(x, y)+O\left(\frac{\ln n}{n^{2}}\right), \quad R\left(r_{1} n, r_{2} n\right)=\frac{1}{2 \pi} \ln \frac{r_{2}}{r_{1}}+O\left(\frac{\ln n}{n}\right)
$$
## 6. Сложные задачи*
6.1. Теорема Лиувилля.Пусть заданная на $\mathbb{Z}^{2}$ функция $f(m, n)$ удовлетворяет неравенству $0 \leq f(m, n) \leq 1$ и условию
$$
f(m, n)=\frac{1}{4}(f(m-1, n)+f(m+1, n)+f(m, n-1)+f(m, n+1))
$$
для всех $m, n \in \mathbb{Z}$. Докажите, что $f(m, n)$ является константой.
6.2. Существование потенциалов. Докажите, что существует такая функция $f(m, n)$ на $\mathbb{Z}^{2}$, что $f(0,0)=0, f(0,1)=1$, для каждого $(m, n) \neq(0,0),(0,1)$ условие $(2)$ выполняется, и $f(m, n)$ стремится к $1 / 2$ при стремлении $r(m, n):=\sqrt{m^{2}+n^{2}}$ к бесконечности.
6.3. Функция Грина. Пусть $f(m, n)$ - сопротивление 2-мерной решетки между началом координат и точкой $(m, n)$.
(A) Докажите, что для каждого $(m, n) \neq(0,0)$ условие (2) выполнено.
(B) Докажите, что $f(m, n)=g(r(m, n))+O(1)$ для некоторой функции $g(x)$.
(C) Докажите, что сопротивление между и центром и границей диска радиуса $r$, вырезанного из 2 -мерной решетки равно $\frac{1}{2 \pi} \ln r+O(1)$.
(D) Докажите, что $f(m, n)=\frac{1}{2 \pi} \ln r(m, n)+O(1)$.
6.4. Найдите с точностью до сотых вероятность того, что при случайном блуждании по 3 -мерной решетке мы вернемся в начальную точку.
6.5. Робот ходит по вершинам 3 -мерной решетки, переходя каждый раз в одну из соседних вершин. В одной из вершин находится клад; робот находит его, когда оказывается в вершине с кладом. Существует ли программа, управляющая движением робота и использующая конечный объем памяти и генератор случайных чисел, такая, что робот найдет клад с вероятностью 1?
## 7. Указания и решения
1.1 (А) Правильность работы программы проверяется следующим образом: разность между "настоящей" и посчитанной вероятностями должна быть пропорциональна числу $\frac{1}{\sqrt{n}}$, где $n$ - число экспериментов.
(B) Ответ смотрите в таблице.
Таблица 2: Вероятности $P_{T}(x)$ и $P(x)$
| | $x$ | 0 | 1 | 2 | 3 | 4 | 5 |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| $T$ | | | | | | | |
| 1 | | 0.00 | 0.00 | 0.00 | 0.00 | 0.50 | 1.00 |
| 2 | | 0.00 | 0.00 | 0.00 | 0.25 | 0.50 | 1.00 |
| 3 | | 0.00 | 0.00 | 0.13 | 0.25 | 0.63 | 1.00 |
| 4 | | 0.00 | 0.06 | 0.13 | 0.38 | 0.63 | 1.00 |
| | $P(x)$ | 0.00 | 0.20 | 0.40 | 0.60 | 0.80 | 1.00 |
(C) Omвет: $P(x)=x / 5$; см последнюю строку в таблице выше.
Доказательство. Рассмотрим случайное блуждание по точкам $0,1,2, \ldots, n$. Обозначим за $P(x)$ вероятность дойти из $x$ до $N$ раньше, чем до 0 . Рассмотрим получившуюся функцию $P(x)$, определенную в точках $x=0,1,2, \ldots, n$. Она обладает следующими свойствами:
1. $P(0)=0$ и $P(n)=1$.
2. $P(x)=\frac{1}{2} P(x-1)+\frac{1}{2} P(x+1)$ для каждого $x=1,2, \ldots, n-1$.
Свойство 1 следует из того, что при достижении точек 0 и $n$ перемещения заканчиваются; для игры на монетки это означает конец игры. Свойство 2 заключается в том, что вероятность попасть домой из внутренней точки $x$ равна среднему арифметическому вероятностей попадания домой из соседних точек. Свойство 2 выводится из следующего утверждения:
Базовое Утверждение. Пусть $E$ - некоторое событие, $F$ и $G$ - два события, из которых всегда случается ровно одно. Тогда
$$
P(E)=P(F) \cdot P(E \text { последует за } F)+P(G) \cdot P(E \text { последует за } G)
$$
В нашем случае $E=$ "человек дойдет до бара", $F=$ "первый раз он пойдет налево" и $G=$ "первый раз он пойдет направо". Тогда получаем $P(E)=P(x), P(F)=P(G)=1 / 2, P(E$ последует за $F)=$ $P(x-1), P(E$ последует за $G)=P(x+1)$ и свойство 2 доказано.
Из этих двух свойств вытекает, что $P(x)$ представляет собой арифметическую прогрессию $P(x)=$ $x / n$.
1.2 Omвет: $P(x)=x / 5$; эта задача эквивалентна задаче 1.1(С).
1.3 Ответ: $P(x)=\frac{(q / p)^{x}-1}{(q / p)^{5}-1}$.
Указание: Рассуждайте так же, как в решении задачи 1.1(C). Покажите, что свойства 1-2 надо заменить на такие:
1. $P(0)=0$ и $P(n)=1$.
2. $P(x)=q P(x-1)+p P(x+1)$ для каждого $x=1,2, \ldots, n-1$.
Выберите $A$ и $B$ такими, чтобы функция $f(x)=A(q / p)^{x}+B$ удовлетворяла новым свойствам 1-2. 1.4 Oтвет: $\approx 99.995 \%$. Точное значение: $1-\frac{(0.55 / 0.45)^{20}-1}{(0.55 / 0.45)^{70}-1}$; смотрите решение задачи 1.3 .
1.5 Oтвет: $v(x)=x / 5$. Указание. Из аксиом 1-2 следует, что функция $v(x)$ будет линейной для этой цепи.
1.6 (А) Пусть $M$ - максимум функции $v(x)$. Тогда если $v(x)=M$ для $x \notin P \cup N$, то это же равенство должно быть выполнено для $v(x-1)$ и $v(x+1)$ так как $v(x)$ - среднее арифметическое $v(x-1)$ и $v(x+1)$. Если $x-1$ оказалась внутренней точкой, применяем то же самое рассуждение и получаем $f(x-2)=M$; продолжая рассуждение, получаем $f(0)=M$. Для минимального значения аналогично.
(В) Положим $h(x)=v(x)-u(x)$. Тогда для любой внутренней точки $x$ имеем:
$$
\frac{h(x-1)+h(x+1)}{2}=\frac{v(x-1)+v(x+1)}{2}-\frac{u(x-1)+u(x+1)}{2}
$$
и поэтому функция $h(x)$ также удовлетворяет аксиоме 2 . Но $h(x)=0$ при $x$ из $P \cup N$; из принципа максимума получаем, что максимальное и минимальное значения $h$ равны 0 . Значит, $h(x)=0$ и $v(x)=u(x)$.
(C) Oтвет: $v(x)=x / n, C=1 / n ; C \rightarrow 0$ и $v(x) \rightarrow 0$ для каждого фиксированного $x$ при $n \rightarrow \infty$.
Указание: Легко проверить, что функция $f(x):=x / n$ удовлетворяет аксиомам $1-2$. Из единственности (см 1.6(В)) следует, что $v(x)=x / n$.
1.7 Теорема. При случайном блуждании по 1-мерной решетке вероятность вернуться когда-либо в начальную точку равна 1.
Доказательство. Пусть $P$ - вероятность вернуться когда-либо в начальную точку. Обозначим $P_{n}$ вероятность вернуться в начальную точку до попадания в $n$ или $-n$. Предположим, что все эти вероятности существуют. Тогда $P_{n} \leq P \leq 1$ для любого $n$.
Сейчас мы докажем, что $P_{n}=1-1 / n$. После первого "хода" человек попадает в одну из точек 1 и -1 с вероятностью $1 / 2$. Если он оказался в точке 1 , то из задачи 1.1 (С) получаем, что вероятность вернуться в начало до попадания в точку $n$ равна $1-1 / n$. Если он оказался в точке -1 , рассуждаем аналогично. Применяя Базовое Утверждение из решения задачи 1.1(С), получаем $P_{n}=\frac{1}{2}\left(1-\frac{1}{n}\right)+$ $\frac{1}{2}\left(1-\frac{1}{n}\right)=1-\frac{1}{n}$. (Еще можно было заметить, что $P_{n}=1-C$, где $C=1 / n-$ проводимость цепи из задачи 1.6.)
Так как $1-1 / n \leq P \leq 1$ для каждого $n$, то $P$ равно 1 .
2.1 Ответ: см. рисунок 9 слева.
| | | 1 | 1 | |
| :---: | :---: | :---: | :---: | :---: |
| | 1 | .823 | .787 | 1 |
| 1 | .876 | .506 | .323 | 0 |
| | 1 | 0 | 0 | |

Рис. 9: Вероятности $P(x)$ или потенциалы $v(x)$; см. задачи 2.1 и 2.2 .
Указание. Схема города представлена на рисунке 9 справа. Вероятности $P(x)$ обозначены за $a, b, c$, $d$, и $e$. Как и в 1-мерном случае, функция $P(x)$ удовлетворяет аксиомам 1-2 из опредления электрической цепи. Отсюда мы получаем систему линейных уравнений:
$$
\begin{aligned}
a & =(b+d+2) / 4 \\
b & =(a+c+2) / 4 \\
c & =(d+3) / 4 \\
d & =(a+c+e) / 4 \\
e & =(b+d) / 4
\end{aligned}
$$
Ответ получаем, решая эту систему.
Замечание. Нахождение точного решения для двухмерной "задачи Дирихле" - дело сложное; поэтому мы рассмотрим два метода нахождения приближенных решений.
Первый метод использует случайные блуждания. Он называется методом Монте-Карло, так как случайные блуждания связаны с вероятностями, а в Монте-Карло находится известные игорные дома, азартные игры в которых тоже связаны с вероятностями. Мы моделируем много случайных блужданий из точки $x$ и находим долю путей, закончившихся в точках $E$. Из закона больших чисел следует, что полученная оценка будет приближением для "настоящей" вероятности $P(x)$. Этот яркий и простой метод позволяет находить решения, но он не очень эффективен.
Теперь опишем более эффективный метод релаксации. Напомним, что мы ищем функцию с заданными значениями на границе у которое значение в любой внутренней точке равно среднему
арифметическому значений ее соседей. Возьмем какую-нибудь функцию с подходящими граничными значениями и возьмем некоторую внутреннюю точку. В общем случае значение функции не будет равно среднему арифметическому значений в соседних точках. Тогда попробуем "подогнать": положим новое значение функции в этой точке равным среднему арифметическому значений в соседних точках. Теперь будем по очереди брать остальные внутренние точки и делать с ними ту же операцию. Когда мы пройдем по всем внутренним точкам, функция не будет удовлетворять аксиоме 2, так как после изменения значения функции в одной точке мы могли изменить значения в соседних с ней точках, нарушив равенство. Тем не менее, полученная функция будет "лучше" удовлетворять аксиоме 2, чем та функция, с которой мы начали; повторяя этот процесс (проходя каждый раз по всем внутренним точкам) мы будем получать приближения к решению лучше и лучше.
2.2 Oтвет: см. рисунок 9 слева; эта задача эквивалентна задаче 2.1.
2.3 Ответ: (A) $2 / 5$; (B) $1 / 2$; (C) $2 / 7$; (D) $2 / 5$.
Указание. Сведем задачу к задаче 2.5 при помощи следующего утверждения:
Физическая интерпретация вероятности. Вероятность того, что случайное блуждание по графу $G$ из вершины а достигнет вершины $h$ до возврата в а, равна
$$
P=C / \operatorname{deg} a
$$
где $C$ - проводимость графа $G$ (все резисторы единичные) между $a$ и $h$, а $\operatorname{deg} a$ - число ребер, выходящих из вершины $a$.
2.4 (C) Указание. Функция $v(x)$ однозначно определена на вершинах получившейся цепи. Проверьте, что она удовлетворяет аксиомам $1-2$.
2.5 Ответ: (1A) $6 / 5$; (1B) 2 ; (1C) 6/7; (1D) 2 .
(2A) $12 / 7$; (2B) $12 / 5$; (2C) 30/19; (2D) 30/11.
Короткое решение смотрите в разделе 3 .
(2A) Указание. Соединим точки $a$ и $b$ с батарейкой; см. рисунок 4 справа. Потенциалы в точках $c$ и $d$ равны из симметрии; аналогично в точках $e$ и $f$. Таким образом, наша схема эквивалентна схеме, изображенной на рисунке 10 слева.
Используя формулы для параллельного и последовательного соединения резисторов, эта цепь сводится в к цепи из одного резистора сопротивлением $7 / 12$ ом (см рисунок 10 справа). Таким образом, сопротивление равно $7 / 12$.
2.6 Ответ: $1 / 7$. Решение аналогично решению задачи 2.3 .
2.7 (А). Короткое решение смотрите в разделе 3.
(B) Авторам неизвестно элементарное решение задачи. Красивое решение, использующее дискретное преобразование Фурье, вы можете найти в книге [7].
2.8 Смотрите раздел 3 .
2.9 (В) Ответ: $C \rightarrow 0$ при $n \rightarrow \infty$.
Указание. Применим закон монотонности: объединим вместе точки, расположенные на квадратах, как показано на рисунке 11 сверху. Полученная цепь эквивалентна цепи на рисунке 11 в центре. Так как можно заменить $n$ параллельных резисторов в 1 ом на один резистор в $1 / n$ ом, цепь эквивалентна цепи на рисунке 11 снизу. Проводимость этой цепи равна
$$
\frac{1}{\sum_{k=1}^{n} \frac{1}{8 k-4}}
$$
Это число стремится к нулю при стремлении $n$ к бесконечности. Так как проводимость старой цепи не больше, она тоже стремится к нулю.
2.10 Указание. Пусть $P$ - вероятность того, что при случайном блуждании по 2 -мерной решетке мы вернемся в начальную точку. Обозначим за $P_{n}$ вероятность того, что случайное блуждание вернется в начальную точку до достижения граничных точек квадрата $2 n \times 2 n$ с центром в начальной точке. Предположим, что все эти вероятности существуют. Ясно, что $P_{n} \leq P \leq 1$ для каждого $n$. Из физической интерпретации вероятности получаем, что $P_{n}=1-C / 4$, где $C$ - эффективное сопротивление

$1 \int^{\frac{7}{5}}$
b) $\left\lvert\, \frac{1}{1+\frac{5}{7}}-\frac{7}{12}\right.$
Рис. 10: Упрощение цепи; см решение задачи 2.5.

Рис. 11: Объединение в квадратной цепи и эквивалентная цепь; см решение задачи 2.9.
между центром и границей квадрата $2 n \times 2 n$. Из решения задачи $2.9(\mathrm{~B})$ следует, что $C$ стремится к нулю при стремлении $n$ к бесконечности. Поэтому $P_{n} \rightarrow 1$ при $n \rightarrow \infty$ и $P$ равно 1 .
3.1 См., например, статью [6].
3.2 (В) Рассмотрим случайное блуждание по электрической цепи. Пусть $P_{T}(x)$ - вероятность того, что, стартуя из вершины $x$ и делая $T$ шагов, мы достигнем положительного полюса батареи раньше, чем отрицательного. Ясно, что при фиксированном $x$ последовательность $P_{T}(x)$ возрастает, значит, имеет предел $P(x)$. Функция $P(x)$ удовлетворяет аксиомам $1-2$.
Замечание. Существование и единственность решения системы Кирхгофа - факты действительно фундаментальные. Например, из единственности решения следует теорема Дена о том, что прямоугольник с иррациональным отношением сторон нельзя разрезать на квадраты. Из существования решения (в непрерывном случае) следует теорема Римана о конформном отображении [2].
3.3 См., например, статью [6].
3.4 (А) Пусть правильный граф содержит $n$ вершин и $A_{1}, A_{2}$ - соседние вершины степеней $k_{1}$ и $k_{2}$ соответственно. Рассмотрим сначала ситуацию, когда в вершину $A_{1}$ подается ток $\frac{n-1}{n}$, а из всех остальных вершин вытекает ток $\frac{1}{n}$. В силу правильности графа по ребру $A_{1} A_{2}$ будет течь ток $\frac{1}{k_{1}}\left(1-\frac{1}{n}\right)$. Если же ток $\frac{n-1}{n}$ подается в вершину $A_{2}$, а из остальных вершин вытекает ток $\frac{1}{n}$, то по ребру $A_{1} A_{2}$ будет течь ток $\frac{1}{k_{2}}\left(1-\frac{1}{n}\right)$. Объединим обе ситуации, заменив во втором случае все токи на противоположные. Тогда получится, что в вершину $A_{1}$ подается единичный ток, который вытекает из вершины $A_{2}$. При этом по ребру $A_{1} A_{2}$ течет ток $\left(\frac{1}{k_{1}}+\frac{1}{k_{2}}\right)\left(1-\frac{1}{n}\right)$. Значит, сопротивление между вершинами $A_{1} A_{2}$ равно $\left(\frac{1}{k_{1}}+\frac{1}{k_{2}}\right)\left(1-\frac{1}{n}\right)$.
Если граф бесконечный, то $1 / n$ нужно заменить нулем. Корректность обосновывается с помощью предельного перехода.
Формула для сопротивления между соседними вершинами правильного графа принадлежит А.Б. Ходул Вместе с определением правильного графа она взята из статьи [4].
(B) Решим сначала задачу на физическом уровне строгости. Наложим шину с нулевым сопротилением на периметр прямоугольника $[-N, N+1] \times[-N, N]$. Поскольку при подключении батарейки к узлам $(0,0)$ и $(1,0)$ потенциалы на бесконечности стремятся к нулю, то наложение шины мало повлияет на значение искомого сопротивления. (Дальнейшие рассуждения также проводятся с точностью до погрешности, которая стремится к нулю с ростом $N$.) Если источник единичного тока подключен к точке $(0,0)$ и шине, то из точки $(0,0)$ в каждый из четырех соседних узлов течет ток равный $1 / 4$. Если же источник единичного тока подключен к шине и точке $(1,0)$, то из четырех соседних узлов в точку $(1,0)$ втекает ток равный $1 / 4$. Значит, при подключении обоих источников тока по ребру $(0,0)-(1,0)$ будет течь ток $1 / 2$, и разность потенциалов в этих точках тоже будет равна $1 / 2$. Но между ними течет общий ток равный единице, поэтому эквивалентное сопротивление решетки между соседними узлами равно $1 / 2$.
Придадим этим рассуждениям математическую строгость. Снова предположим, что источник единичного тока подключен к точке $(0,0)$ и шине с нулевым сопротилением, наложенной на периметр прямоугольника $[-N, N+1] \times[-N, N]$. Согласно задаче 5.2 сопротивление $R$ такого графа не превосходит $N^{1 / 2}$, поэтому при нулевом потенциале на шине потенциал $V$ в точке $(0,0)$ не превосходит $N^{1 / 2}$ :
$$
V=I R=R \leq N^{1 / 2}
$$
Покажем, что потенциал в точках, близких к наложенной шине, мало отличается от нуля. Обозначим через $u_{j}$ наибольший потенциал на периметре прямоугольника $[-j, j+1] \times[-j, j]$. Тогда из гармоничности распределения потенциалов следует, что $u_{j-1} \geq 2 u_{j}-u_{j+1}(1 \leq j \leq N-1)$. Поэтому, если $u_{N-1}=\varepsilon$ (по предположению $u_{N}=0$ ), то для всех $j$ в пределах $0 \leq j \leq N$ будет выполняться неравенство $u_{j} \geq(N-j) \varepsilon$. В частности, $u_{0}=V \geq N \varepsilon$. Учитывая неравенство (3), получаем, что $\varepsilon \leq N^{-1 / 2}$.
Если теперь наложить шину по периметру квадрата $[-N, N] \times[-N, N]$, то получится новое (симметричное) распределение потенциалов, которое, согласно принципу максимума, отличается от исходного не более чем на $N^{-1 / 2}$. Значит, до переноса правого края шины четыре тока, выходящие из точки $(0,0)$ отличались от $1 / 4$ не более чем на $N^{-1 / 2}$. Аналогично, если источник тока подключен к
шине и к точке $(1,0)$, то из четырех соседних узлов в точку $(1,0)$ втекают четыре тока, отличающиеся от $1 / 4$ не более чем на $N^{-1 / 2}$.
Объединяя обе ситуации, получаем, что в прямоугольнике $[-N, N+1] \times[-N, N]$ с закороченным периметром источник единичного тока подключен к узлам $(0,0)$ и $(1,0)$, а по ребру $(0,0)-(1,0)$ течет ток $1 / 2+O\left(N^{-1 / 2}\right)$. Значит, разность потенциалов, а соответственно и сопротивление, равны $1 / 2+O\left(N^{-1 / 2}\right)$.
Для завершения доказательства применим аксиому 3. Рассмотрим батарейку, подключенную к точкам $(0,0)$ и $(1,0)$ так, что потенциалы в этих точках равны $1 / 4$ и $-1 / 4$ соответственно. Ток, протекающий между $(0,0)$ и $(1,0)$ обозначим через $I$. Выберем прямоугольник $[-N, N+1] \times[-N, N]$ так, чтобы потенциалы на его периметре по модулю не превосходили некоторого $\varepsilon>0$. При замене всех потенциалов на периметре прямоугольника нулями, согласно принципу максимума, все потенциалы внутри так же изменятся не более чем на $\varepsilon>0$. Тогда получится, что ток $I$ течет между узлами с разностью потенциалов $1 / 2+O(\varepsilon)$, а сопротивление между которыми равно $1 / 2+O\left(N^{-1 / 2}\right)$. Следовательно $I=1+O(\varepsilon)+O\left(N^{-1 / 2}\right.$ ). Так как $\varepsilon$ может быть выбрано сколь угодно малым, а $N$ растет при уменьшении $\varepsilon$, то ток $I$ в точности равен единице. Он течет между узлами с разностью потенциалов $1 / 2$, значит, эквивалентное сопротивление решетки между соседнми узлами в точности равно $1 / 2$.
3.5 Пусть теперь $B_{1}$ и $B_{2}$ - вершины графа, диаметрально противоположные $A_{1}$ и $A_{2}$ соответственно. Мы доказали, что если в $A_{1}$ подается единичный ток, который вытекает из вершины $A_{2}$, то по ребру $A_{1} A_{2}$ течет ток $I=\left(\frac{1}{k_{1}}+\frac{1}{k_{2}}\right)\left(1-\frac{1}{n}\right)$. Ток текущий при этом по ребру $B_{2} B_{1}$ обозначим через $x$. Дополнительно подключим источник единичного тока к вершинам $B_{1}$ и $B_{2}$ (в $B_{2}$ подается, из $B_{1}$ вытекает). Тогда по каждому из ребер $A_{1} A_{2}$ и $B_{2} B_{1}$ будет течь ток $I+x$. Но то же распределение токов получится, если единичный ток подается в вершину $A_{1}$ и вытекает из $B_{1}$, а дополнительный ток подается в $B_{2}$ и вытекает из $A_{2}$. В такой ситуации в силу правильности графа первый источник по ребру $A_{1} A_{2}$ дает ток $\frac{1}{k_{1}}$, а второй $-\frac{1}{k_{2}}$. Отсюда
$$
\begin{aligned}
I+x & =\left(\frac{1}{k_{1}}+\frac{1}{k_{2}}\right)\left(1-\frac{1}{n}\right)+x=\frac{1}{k_{1}}+\frac{1}{k_{2}} \\
x & =\left(\frac{1}{k_{1}}+\frac{1}{k_{2}}\right) \frac{1}{n}, \quad \frac{x}{I}=\frac{1}{n-1}
\end{aligned}
$$
Поэтому для икосаэдра получается ток $\frac{I}{11}$, додекаэдра $-\frac{I}{19}$, ромбододекаэдра $-\frac{I}{13}$, куба $-\frac{I}{7}$. 3.6 Указание. Суммы $D_{n}$ и $F_{n}$ удовлетворяют одному и тому же рекуррентному соотношению. Например, $D_{n}=1+\frac{n+1}{2 n} D_{n-1}$. Кроме того, $D_{0}=F_{0}=1$. Следовательно, они равны друг другу. Значит, при $n \geq 1$
$$
\sum_{k=1}^{n} \frac{2^{k}}{k}=\frac{2^{n}}{n} F_{n-1}=\frac{2^{n}}{n} D_{n-1}=\frac{2^{n}}{n} \sum_{k=0}^{n-1} \frac{1}{C_{n-1}^{k}}
$$
Для завершения доказательства нужно оценить степень вхождения числа 2 в общий знаменатель дробей из полученной суммы с помощью формулы Лежандра для показателя, с которым простое число входит в разложение факториала.
Сопротивление $n$-мерного проволочного куба (у которого каждое ребро имеет единичное сопротивление) между противоположными вершинами $R_{n}$ связанно с данными суммами равенствами (более подробно см. в статье [5]).
$$
D_{n}=F_{n}=(n+1) R_{n+1}
$$
4.1 Указание. Докажите по индукции, что сопротивление бинарного дерева глубины $n$ из единичных резисторов равно $1-\frac{1}{2^{n}}$.
4.2 Указание. Потенциалы в точках, расположенных на одинаковом расстоянии от корня дерева, равны из симметрии. Объединив такие точки в бинарном дереве, получим цепь, изображенную на рисунке 12. Ее сопротивление равно $\frac{1}{2} \cdot n=\frac{n}{2}$. Для троичного дерева аналогично получаем $R=$ $\frac{1}{3}+\frac{2}{9}+\cdots+\frac{2^{n-1}}{3^{n}}$. Отсюда $R=1-\frac{2^{n}}{3^{n}}$.
4.3 Указание. Двоичное дерево глубины 3 вырезать не сложно. Покажем, что двоичное дерево глубины 2010 вырезать нельзя. Если его удалось вырезать, то все его вершины расположены на расстоянии

Рис. 12: Подсчет сопротивления дерева; см решение задачи 4.2.

Рис. 13: Вырезание двоичного дерева с пересечениями из плоскости; см решение задачи 4.4.
не более 2010 от корня; отсюда получаем, что дерево находится в кубе со стороной $2 \cdot 2010+1$. Поэтому число его вершин не превосходит $4021^{3} \leq 2^{36}$. С другой стороны, число его вершин равно $2^{2011}-1$. Полученное противоречие завершает доказательство. Задача о вырезании модифицированного дерева решается не просто.
4.4 Указание. Двоичное дерево вырезать нельзя; рассуждайте аналогично решению задачи 4.3, пользуясь тем, что более двух вершин склеиться не могут. Модифицированное двоичное дерево можно вырезать из плоскости (см рисунок 13), а троичное из пространства аналогичным образом (см рисунок 14). Доказательство проводится индукцией по глубине дерева.
4.5 Указание. Для любого $n=2^{i}$ - 1 рассмотрим множество вершин $(x, y, z)$, где $|x|+|y|+|z| \leq n$. Пусть $R_{i}$ - сопротивление между началом координат и границей такой фигуры. Как известно из задачи 4.4, из такой части решетки можно вырезать модифицированное троичное дерево глубины $i$ с пересечениями ребер на равном расстоянии от корня. Легко заметить, что сопротивление дерева с такими пересечениями равно сопротивлению такого же дерева без пересечений. Как известно из задачи 4.2, сопротивления модифицированных троичных деревьев не превосходят 1. Поэтому не превосходят 1 и сопротивления вырезаемых деревьев с пересечениями. Из закона монотонности получаем, что $R_{i} \leq 1$. Значит, при подключении батарейки в 1 вольт ток будет не меньше 1 . Следовательно, потенциалы в вершинах, соседних с началом координат будут не больше $1-\frac{1}{6}=\frac{5}{6}$. Они равны вероятности возврата в начало координат до попадания на границу. Переходя к пределу, получаем требуемое.
5.1 Рассмотрим точки $A(2,3), B(3,3), C(1,2), D(2,2), E(3,2), F(4,2), G(2,1), H(3,1)$. Сначала подсоединим источник тока, который подает единичный ток в начало координат, а вторым концом подсоединен к контуру с нулевым сопротивлением, наложенному на периметр квадрата $[-R, R]^{2}$. Из-за сим-

Рис. 14: Вырезание троичного дерева с пересечениями из пространства; см решение задачи 4.4.
метрии для некоторого числа $I$ будут выполняться равенства $i(C D)=i(G D)=i(D A)=i(D E)=I$. Рассмотрим теперь другую ситуацию, когда единичный ток подается на периметр квадрата $[-R, R]^{2}$ и выходит из точки $(1,0)$. Тогда $i(D E) \approx i(H E) \approx i(E B) \approx i(E F) \approx-I$. Здесь равенство понимается с точностью до малого $\varepsilon$, которое стремится к нулю с ростом $R$ (как и в задаче 3.4 , это следует из аксиомы 3). Комбинируя обе ситуации получаем, что когда источник тока подключен к узлам $(0,0)$, $(1,0)$, а на периметр квадрата $[-R, R]^{2}$ наложен контур с нулевым сопротивлением, по ребру $D E$ течет ток меньший, чем $\varepsilon$. Устремляя $R$ в бесконечность (и снова применяя аксиому 3 ), приходим к утверждению задачи.
Отметим, что при решении мы без доказательства пользовались сложным утверждением о существовании и единственности.
5.2 Рассмотрите какое-нибудь дерево, соединяющее данную точку с периметром квадрата.
5.3 Примените принцип максимума.
5.4 Запишем оператор Лапласа в виде
$$
\Delta f(x, y)=f(x-1, y)+f(x+1, y)-2 f(x, y)+f(x, y-1)+f(x, y+1)-2 f(x, y)
$$
Тогда для функции $f(x, y)=\ln r(x, y)$
$$
\begin{gathered}
f(x-1, y)+f(x+1, y)-2 f(x, y)=\frac{1}{2} \ln \frac{\left((x+1)^{2}+y^{2}\right)\left((x-1)^{2}+y^{2}\right)}{\left(x^{2}+y^{2}\right)^{2}}= \\
=\frac{1}{2} \ln \left(1+\frac{2 x+1}{r^{2}}\right)\left(1+\frac{-2 x+1}{r^{2}}\right)=\frac{1}{2} \ln \left(\left(1+\frac{1}{r^{2}}\right)^{2}-\frac{4 x^{2}}{r^{4}}\right)
\end{gathered}
$$
Аналогично
$$
f(x, y-1)+f(x, y+1)-2 f(x, y)=\frac{1}{2} \ln \left(\left(1+\frac{1}{r^{2}}\right)^{2}-\frac{4 y^{2}}{r^{4}}\right)
$$
Поэтому
$$
\begin{aligned}
& \Delta f(x, y)=\frac{1}{2} \ln \left(\left(1+\frac{1}{r^{2}}\right)^{2}-\frac{4 x^{2}}{r^{4}}\right)\left(\left(1+\frac{1}{r^{2}}\right)^{2}-\frac{4 y^{2}}{r^{4}}\right)= \\
& =\frac{1}{2} \ln \left(1-\frac{1}{r^{4}}+\frac{16 x^{2} y^{2}}{r^{8}}\right)=\frac{1}{2} \ln \left(1+O\left(\frac{1}{r^{4}}\right)\right)=O\left(\frac{1}{r^{4}}\right)
\end{aligned}
$$
5.5 Если точка $(x, y)$ лежит вблизи границы кольца, то в ней имеет смысл изменить определение оператора Лапласа, чтобы оно согласовывалось с правилами Кирхгофа. Например, если для некоторых $a, b \in[0,1)$ точки $(x-a, y)$ и $(x, y-b)$ попадают на границу, то будем считать, что
$$
\Delta f(x, y)=\frac{f(x-a, y)-f(x, y)}{a}+\frac{f(x, y-b)-f(x, y)}{b}+f(x+1, y)+f(x, y+1)-2 f(x, y)
$$
Тогда для функции $f(x, y)=\ln r(x, y)$ в такой точке
$$
\frac{f(x-a, y)-f(x, y)}{a}+f(x+1, y)-f(x, y)=\frac{1}{2 a} \ln \left(1+\frac{-2 a x+a^{2}}{r^{2}}\right)+\frac{1}{2} \ln \left(1+\frac{2 x+1}{r^{2}}\right)=O\left(\frac{1}{r^{2}}\right)
$$
Аналогично
$$
\frac{f(x, y-b)-f(x, y)}{b}+f(x, y+1)-f(x, y)=O\left(\frac{1}{r^{2}}\right)
$$
Таким образом $\Delta f(x, y)=O\left(r^{-2}\right)$, причем эта оценка остается справедливой, если из четырех соседних с $(x, y)$ точек лишь одна лежит за пределами кольца.
Рассмотрим теперь функцию $f(x, y)=U_{n}(x, y)-\ln r(x, y)$. Она равна нулю на границе кольца, а во всех внутренних точках удовлетворяет уравнению $\Delta f(x, y)=\varphi(x, y)$, где $\varphi(x, y)=O\left(n^{-2}\right)$ в точках вблизи границы, и $\varphi(x, y)=O\left(n^{-4}\right)$ в остальных точках кольца.
Значения функции $\varphi(x, y)$ - это токи, которые подаются в соотвествующие узлы кольца. Потенциалы, индуцированные токами в точках вблизи границы оцениваются как $O\left(n^{-2}\right)$. Действительно, если, считать, что во все внутренние точки области имеют равный потенциал $U$, то в точки границы текут токи не меньше чем $U$. Поэтому, при замене всех токов на $U$ потенциалы внутри области не увеличатся.
Оценим теперь потенциал порождаемый токами в остальных точках (отделенных от границы). Число таких точек есть $O\left(n^{2}\right.$ ), и ток в каждой из них (согласно задаче 5.3) приводит к потенциалам не превосходящим $O\left(n^{-7 / 2}\right)$. Поэтому общий потенциал, порожденный внутренними токами есть $O\left(n^{-3 / 2}\right)$. То есть $f(x, y)=O\left(n^{-3 / 2}\right)$.
5.6 Воспользуемся равенством
$$
\operatorname{arctg} x-\operatorname{arctg} y=\operatorname{arctg} \frac{x-y}{1+x y}
$$
которое справедливо при $|x y|<1$. Тогда
$$
\begin{gathered}
\operatorname{arctg} \frac{y+1}{R}-\operatorname{arctg} \frac{y}{R}=\operatorname{arctg} \frac{1 / R}{1+y(y+1) / R^{2}}= \\
=\operatorname{arctg}\left(\frac{R}{y^{2}+R^{2}}+O\left(\frac{1}{R^{2}}\right)\right)=\frac{R}{R^{2}+y^{2}}+O\left(\frac{1}{R^{2}}\right)
\end{gathered}
$$
5.7 Просуммируйте формулу из задачи 5.6.
5.8 Для нахождения сопротивления кольца найдем ток, который будет через него протекать при условии, что на внутренний контур кольца подается напряжение $\ln n r_{1}$, а на внешний $-\ln n r_{2}$. Будем искать ток протекающий через периметр квадрата $[-R-1 / 2, R+1 / 2]^{2}$, где $R=\left[r_{1} n\right]+1$. Посчитаем его приближенно, заменяя потенциалы в узлах на значения функции $\ln r(x, y)$. Всего будет просуммировано $O(n)$ токов, каждый с погрешностью $O\left(n^{-3 / 2}\right)$. Поэтому итоговая погрешность будет равна $O\left(n^{-1 / 2}\right)$.
В силу симеетрии квадрата, протекающий через его периметр ток может быть записан в виде
$$
I=8 \sum_{y=0}^{R}(\ln r(R+1, y)-\ln r(R, y))+O\left(\frac{1}{n^{1 / 2}}\right)
$$
Так как
$$
\ln r(R+1, y)-\ln r(R, y)=\frac{R}{R^{2}+y^{2}}+O\left(\frac{1}{R^{2}}\right)
$$
то, применяя формулу из задачи 5.7 , получаем нужное равенство $I=2 \pi+O\left(n^{-1 / 2}\right)$.
5.9 Как и в предыдущей задаче, для вычисления сопротивления нужно найти ток, протекающий через замкнутую ломаную, опоясывающую внутреннюю окружность. Снова для приближенного нахождения тока потенциалы в узлах решетки можно заменить на значения функции $\ln r(x, y)$. Если теперь ломаную заменить на описанный около нее квадрат, то внутри полученного контура появится $O\left(n^{2}\right)$ новых источников тока, в каждом из которых втекает или вытекает ток равный $O\left(n^{-4}\right)$. Значит, искомое значение для суммарного тока отличается от уже найденного тока через периметр квадрата $I=2 \pi+O\left(n^{-1 / 2}\right)$ не более чем на $O\left(n^{-2}\right)$.
5.10 Докажем, что в задаче 5.2 сопротивление между любым узлом квадрата и его границей есть $O(\ln n)$. Тогда и во всех следующих задачах при буквальном повторении доказательств все остаточные члены умножаться на $\frac{\ln n}{\sqrt{n}}$.
Вместо квадрата рассмотрим треугольник, вырезанный из квадратной сетки прямыми $x=0$, $y=0, x+y=n$, и оценим его сопротивление между началом координат и гипотенузой. Будем предполагать, что в целых точках на отрезка $x+y=k, x, y \geq 0$ потенциалы равны $V_{k}=\sum_{j=2}^{k+1} \frac{1}{j}$ $(0 \leq k \leq n)$. В частности, в начале координат потенциал нулевой. Будем также предполагать, что в каждую точку на прямой $x+y=k$ втекает ток $1 / k$, то есть через каждый уровень протекает единичный ток. Чтобы ситуация не противоречила закону Ома, сопротивления внутри треугольника
нужно будет увеличить. (Сопротивления, соединяющие точки вида $(0, k),(k, 0)$ с точками $(0, k+1)$, $(k+1,0)$ остаются единичными.) Чтобы выполнялся закон Кирхгофа, из точки $(j, k-j)$ должны течь токи $\frac{k-j}{k(k+1)}$ и $\frac{j+1}{k(k+1)}$ в точки $(j, k-j+1)$ и $(j+1, k-j)$ соответственно. Посколку разность потенциалов равна $1 /(n+1)$, то единичные сопротивления нужно будет заменить на сопротивления $\frac{k}{k-j}$ и $\frac{k}{j+1}$ соответственно. Полученная схема имеет сопротивление $V_{n} \leq \ln n$. Значит, сопротивление исходной схемы также не превосходит $\ln n$.
## 8. Благодарности
Большинство задач частей 1, 2 и 4 данного проекта заимствованы из статьи П. Дойля и Дж. Снелл [6]. Авторы благодарны И. Богданову, В. Бугаенко и М. Прасолову за помощь при переводе данногог проекта.
[1] I. Benjamini and O. Schramm, Random walks and harmonic functions on infinite planar graphs using square tilings, Ann. Prob. 24:3 (1996), 1219-1238.
[2] J. Cannon, W. Floyd, W. Parry, Squaring rectangles: the finite Riemann mapping theorem, Contemp. Math. 169 (1994), 133-211.
[3] P. G. Doyle and J. L. Snell, Random walks and electric networks, Mathematical Association of America, 1984, http://arxiv.org/abs/math.PR/0001057.
[4] Г.А.Гальперин "Мой друг Андрей Ходулчв" (Математическое просвещение, сер. 3, вып. 4 (2000), $8-32$,
http://www.mccme.ru/free-books/matpros/i5008032.pdf.zip.
[5] Ф. Недемейер и Я. А. Смородинский, Сопротивление ребер многомерного куба, Квант, № 6, 1986.
[6] M. Prasolov and M. Skopenkov, Tiling by rerctangles and alternating current, submitted (2010). http: //arxiv.org/abs/1002.1356.
[7] F. Spitzer, Principles of random walks, Springer-Verlag, 1976.
## Random walks through electrical networks
Dmitry Baranov, Mikhail Skopenkov, Alexey Ustinov
The aim of the project is to prove the following result and investigate related problems. The Polya Theorem. (a) A man which is randomly walking in a 2-dimensional lattice will return to the initial point with probability 1.
(b) A man which is randomly walking in a 3-dimensional lattice will return to the initial point with probability strictly less than 1.
Accurate statements are given in the project. The suggested approach to the result is based on a physical interpretation. However, no physical background is assumed.

2-dimensional lattice
1-dimsensional latlice
## 1. Walking in one dimension
Let us first state a problem and then give all the necessary definitions.
1.1. A man walks along a 5-block stretch of Madison Avenue. He starts at corner $x$ and, with probability $1 / 2$, walks one block to the right and, with probability $1 / 2$, walks one block to the left; when he comes to the next corner he again randomly chooses his direction along Madison Avenue. He continues until he reaches corner 5, which is home, or corner 0 , which is a bar. If he reaches either home or the bar, he stays there; see Figure 1.
Bar

Hıне
Figure 1: Random walk along Madison Avenue; see Problem 1.1.
(A)* Write a computer program which models the motion of the man. Run the program a large number of times, and find the percentage of cases in which the man returns home. You may use this to guess the answers in further problems.
E-mail address: dimbaranov@mail.ru, skopenkov@rambler.ru, ustinov.alexey@gmail.com
(B) Let $P_{T}(x)$ be the probability that the man, starting at corner $x$ and making at most $T$ "moves", will reach home. Complete the following table by 2-digit decimals.
Table 1: The probabilities $P_{T}(x)$ for small $T$
| $T^{x}$ | 0 | 1 | 2 | 3 | 4 | 5 |
| :--- | :---: | :---: | :---: | :---: | :---: | :---: |
| 1 | 0.00 | 0.00 | 0.00 | 0.00 | 0.50 | 1.00 |
| 2 | | | | | | |
| 3 | | | | | | |
| 4 | | | | | | |
(C) Find the probability $P(x)$ that the man will reach home eventually.
Definition. (A) Suppose that an experiment has $n$ equally possible outcomes, and an event $X$ occurs in exactly $m$ of the outcomes. Then the probability of the event $X$ is by definition the number $P_{1}(X):=m / n$.
For instance, the probability of getting tails in a coin throw is $1 / 2$; the probability of getting 6 points in a die roll is $1 / 6$; the probability that our walker moves one block to the right is $1 / 2$.
(B) Now suppose that the event $X$ depends on a sequence of such experiments. A sequence of $T$ experiments has $n^{T}$ possible outcomes. Assume that $X$ occurs for exactly $m_{T}$ outcomes among them. Then the probability of $X$ is the number $P_{T}(X):=m_{T} / n^{T}$.
For instance, there are 4 possible outcomes of throwing a coin 2 times:
| 1st throw
2 nd throw | heads
heads | heads
tails | tails
heads |
| :---: | :---: | :---: | :---: |
Let the event $X$ be getting tails in at least one throw. The event $X$ occurs for 3 cases among the 4 possible ones. Thus the probability of the event $X$ is $P_{2}(X)=3 / 4$.
The probability of getting more than 10 points in two die rolls is $1 / 12$ because this event occurs for 3 cases $(5+6,6+5$ or $6+6)$ among the 36 possible ones. The probability that our walker, starting at corner 3 , consequently moves right twice is $1 / 4$.
(C) Finally, suppose that the event $X$ depends on an infinite sequence of such experiments. We say that the probability of the event $X$ is $P(X)$, if the probabilities $P_{T}(X)$ tend to a number $P(X)$ as $T$ tends to infinity ${ }^{1}$.
For instance, the probability of getting tails at least once in an infinite sequence of coin throws is $P(X)=1$, because $P_{T}(X)=1-1 / 2^{T}$ tends to 1 as $T$ tends to infinity.
You may use without proof that the probability $P(X)$ exists for all the events $X$ considered in the project.
1.2. Peter and Paul match pennies; they have a total of 5 pennies; on each match, Peter wins one penny from Paul with probability $1 / 2$ and loses one with probability $1 / 2$; they play until Peter's fortune reaches 0 (he is ruined) or reaches 5 (he wins all Paul's money). Find the probability $P(x)$ that Peter wins if he starts with $x$ pennies.[^4]
1.3. Assume that our walker has a tendency to drift in one direction: more specifically, assume that each step is to the right with probability $p$ or to the left with probability $q=1-p$. Find the probability $P(x)$ in this case.
1.4. You are gambling against a professional gambler; you start with 20 dollars and the gambler with 50 dollars; you play a game in which you win one dollar with probability 0.45 and lose one dollar with probability 0.55 ; play continues until you or the gambler runs out of money. Find the probability of being ruined.
Definition. An electrical network is a connected finite graph with a positive real number (conductance $\left.{ }^{2} C(x y)\right)$ assigned to each edge $x y$, and two disjoint marked sets of vertices $(P$ and $N)$. The vertices of the set $N$ are joined with the ground and the negative pole of a battery, and the vertices of $P$ are joined with the positive pole; see Figure 2.
The voltages $v(x)$ of the vertices in the network are defined by the following axioms:
1. Boundary condition. If $x \in N$ then $v(x)=0$; if $x \in P$ then $v(x)=1$.
2. Kirchhoff current law. If $x \notin P \cup N$ then $\sum_{x y} C(x y)(v(x)-v(y))=0$, where the summation is over all the edges $x y$ containing the vertex $x$.
The number $i(x y):=C(x y)(v(x)-v(y))$ is the current through edge $x y ; i(x):=$ $\sum_{x y} i(x y)$ is the current flowing inside the network through vertex $x$ (thus $i(x)=0$ for each $x \notin P \cup N$ by axiom 2); $C:=\sum_{x \in P} i(x)$ is the effective conductance of the network between the subsets $P$ and $N ; Q:=\sum_{x y} C(x y)(v(x)-v(y))^{2}$, where the summation is over all the edges of the network, is the heat power of the network.
1.5. We connect equal resistors in series and put a unit voltage across the ends as in Figure 2. Find the voltages $v(x)$ established at the points $x=0,1,2,3,4,5$. Hereafter you may use network-simulation software to guess the answer.

Figure 2: An electrical network; see Problem 1.5.
1.6. Consider the network with the vertices $0,1, \ldots, n$, the edges $01,12, \ldots,(n-1) n$ of unit conductance, and marked sets $N=\{0\}, P=\{n\}$.
(A) Maximum Principle. A function $v(x)$ satisfying the above axiom 2 takes on its maximum value and its minimum value on the set $P \cup N$.
(B) Uniqueness Principle. If $v(x)$ and $u(x)$ are two functions satisfying the above axioms 1-2 then $v(x)=u(x)$ for all $x$.
(C) Find the voltages $v(x)$ and the effective conductance of the network. To which numbers tend these values as $n$ tends to infinity?
1.7. State and prove an analogue of the Polya theorem for the 1-dimensional lattice.[^5]
## 2. Walking in two dimensions
2.1. Consider the town in Figure 3 to the left. Segments represent streets. Large dots marked $E$ indicate escape routes and those marked $P$ are police. Find with 2-digit precision the probability $P(x)$ that our walker, starting at an interior point $x$, will reach an escape route before he reaches a policeman. The walker moves from $x=(a, b)$ to each of the four neighboring points $(a+1, b),(a-1, b),(a, b+1),(a, b-1)$ with probability $1 / 4$. If he reaches one of the points $E$ or $P$, he remains at this point.
2.2. Find the voltages $v(x)$ in the network (of unit resistors) in Figure 3 to the right.

Figure 3: Random walk in a town and an electrical network; see Problems 2.1 and 2.2.
2.3. A bug walks randomly on the edges of
(A) a cube;
(B) an octahedron;
(C) a dodecahedron;
(D) an icosahedron;
If the bug starts at a vertex $a$, what is the probability that it reaches food at the opposite vertex $h$ before returning to $a$; see Figure 4 to the left?

Figure 4: Random walk in a cube and an electrical network; see Problems 2.3(A) and 2.5(A).
2.4. The following transformations preserve the effective conductance of a network:
(A) replacing two resistors connected in series by a single resistor whose conductance is 1/ $\left(\frac{1}{C_{1}}+\frac{1}{C_{2}}\right)$; see Figure 5 to the left;
(B) replacing two resistors connected in parallel by a single resistor whose conductance is $C_{1}+C_{2}$; see Figure 5 to the right;
(C) shortening together two vertices having the same voltage.

Figure 5: Series and parallel connections; see Problem 2.4.
2.5. Find the effective conductance between (1) opposite; (2)* adjacent; vertices of
(A) a cube;
(B) an octahedron;
(C) a dodecahedron;
(D) an icosahedron;
with edges of unit conductance; see Figure 4 to the right.
2.6. A drunken tourist starts at her hotel and walks at random through the streets of the idealized Paris shown in Figure 6 to the left. Find the probability that she reaches the Arc de Triomphe before she reaches the outskirts of town.
2.7. The conductance between vertices (A)* $a$ and $b ;(\mathbf{B})^{* *} a$ and $c$; of the 2 -dimensional lattice (of unit resistors) equals to 2 and $\pi / 2$, respectively; see Figure 6 to the right.
Notice that for the 2-dimensional lattice we need to modify the definition of the voltages $v(x)$ by adding one more axiom:
3. Condition at infinity. $v(x)$ tends to $1 / 2$ as the distance between $x$ and a fixed vertex tends to infinity.
It is allowed to use without proof that there exists a function $v(x)$ satisfying axioms 1-3.

Figure 6: Paris tourist map and a lattice network; see Problems 2.6 and 2.7.
2.8. Rayleigh's Monotonicity Law. Cutting certain edges can only decrease the effective conductance between two given nodes; see Figure 7 to the left. Shorting certain sets of nodes together can only increase the effective conductance of the network between two given nodes; see Figure 7 in the middle.
2.9. (A) Prove that the conductance between the center and the boundary of a square $4 \times 4$ lattice of unit resistors is less than 3 ; see Figure 7 to the right.
(B) To which number tends the conductance between the center and the boundary of a square $2 n \times 2 n$ lattice of unit resistors as $n$ tends to infinity?
2.10. Prove the Polya theorem for the 2-dimensional lattice.

Short

Figure 7: Cutting, shortening, and a square $4 \times 4$ lattice; see Problems 2.8 and 2.9.
## 3. Conductances of symmetric graphs
The problems of this sections may help to solve the problems from the previous one.
A function $v(x)$ on vertices of an electrical network is harmonic, if it satisfies axiom 2 from the definition of an electrical network. The vertices from the set $P \cup N$ are called boundary vertices, the remaining vertices are called interior ones.
3.1. (A) Superposition principle. If functions $u(x)$ and $v(x)$ are harmonic and $a, b \in \mathbb{R}$ then the function $a u(x)+b v(x)$ is harmonic.
(B) Maximum principle. Prove that a harmonic function $u(x)$ defined on a finite electrical network takes its maximum and minimum values on boundary vertices.
(C) Uniqueness principle. If harmonic functions $u(x)$ and $v(x)$ coincide at each boundary vertex of a finite network then $u(x)=v(x)$ for each vertex $x$ of the network.
(D) A man randomly walking in a finite town visits all street corners with probability 1.
3.2. (A)* Fredholm's alternative. Take a system of linear equations
$$
\left\{\begin{array}{c}
a_{1,1} x_{1}+\ldots+a_{1, n} x_{n}=b_{1} \\
\ldots \ldots \ldots \ldots \ldots \ldots \\
a_{n, 1} x_{1}+\ldots+a_{n, n} x_{n}=b_{n}
\end{array}\right.
$$
in which the number of equations equals to the number of unknowns. Then exactly one of the following alternatives holds:
1. for any $b_{1}, \ldots, b_{n}$ the system has a unique solution (in particular, for $b_{1}=\ldots=$ $b_{n}=0$ there is only zero solution);
2. for some $b_{1}, \ldots, b_{n}$ the system has no solutions, and for some other (in particular, for $\left.b_{1}=\ldots=b_{n}=0\right)$ it has infinitely many solutions.
(B)* Dirichlet's problem. Prove that for any finite electrical network there exist a function $v(x)$ satisfying axioms 1-2.
3.3. (A) Variational principle. Let $v(x)$ be an arbitrary function on vertices of a finite electrical network satisfying axiom 1 , but not necessarily axiom 2 . Enumerate the vertices by $1, \ldots, n$ and let $1, \ldots, k$ be the interior vertices. Denote by $v_{1}:=v(1), v_{2}:=v(2)$, $\ldots, v_{n}=v(n)$. Consider $v_{1}, \ldots, v_{k}$ as independent variables. Consider the heat power $Q\left(v_{1}, \ldots, v_{k}\right):=\sum_{x y} C(x y)\left(v_{x}-v_{y}\right)^{2}$ as a function in variables $v_{1}, \ldots, v_{k}$. Prove that the function $Q\left(v_{1}, \ldots, v_{k}\right)$ takes its minimum when the function $v(x)$ is harmonic.
(B) Prove that for the function $Q\left(v_{1}, \ldots, v_{k}\right)$ there exist a unique sequence $v_{1}, \ldots, v_{k}$, for which the function takes its minimum. Apply this to obtain the second proof that for each finite electrical network there is a unique function $v(x)$ satisfying axioms 1-2.
(C) Energy conservation law. Prove that the minimal value of the heat power $Q\left(v_{1}, \ldots, v_{k}\right)$ numerically equals to the effective conductance $C$.
(D) Rayleigh's Monotonicity Law. If the conductances of the edges of a network are increased, the effective conductance can only increase.
(E) One removed an arbitrary set of edges from the 2-dimensional lattice. Prove that a random walk still returns to the initial vertex with probability 1.
Let $\Gamma$ be an electrical network of unit resistors. Take a pair of adjacent edges $A B$ and $A C$ of the network. These edges are called equivalent, if there is a permutation of vertices of the network, taking adjacent vertices to adjacent ones, and taking $A$ to $A$ and $B$ to $C$. A vertex is a symmetry center of the network $\Gamma$, if all the edges containing the vertex are equivalent. The network $\Gamma$ is regular, if all its vertices are symmetry centers.
Examples of regular graphs: regular polyhedra of arbitrary dimensions, regular lattices of arbitrary dimensions, regular lattices in a torus etc. A nontrivial example: graph of a rombododecahedron. The vertices of this polyhedron have different degrees (3 and 4).
3.4. (A) A regular network contains $n$ vertices. Let $A_{1}$ and $A_{2}$ be two adjacent vertices of degrees $k_{1}$ and $k_{2}$, respectively. Prove that the resistance between them is
$$
\left(\frac{1}{k_{1}}+\frac{1}{k_{2}}\right)\left(1-\frac{1}{n}\right)
$$
(B) If the network is 2-dimensional lattice then $1 / n$ should be replaced by 0 .
3.5. The battery is joined with two adjacent vertices of (A) an icosahedron; (B) a dodecahedron; so that the current through the edge joining the two vertices is $I$. Find the current through the opposite edge.
3.6. * Prove that the sums $D_{n}=\sum_{k=0}^{n} \frac{1}{C_{n}^{k}}$ and $F_{n}=\frac{n+1}{2^{n+1}} \sum_{k=1}^{n+1} \frac{2^{k}}{k}$ are equal. How these sums are connected with the conductance of an $n$-dimensional cube? Apply this to prove that the order of 2 in the number $\sum_{k=1}^{n} \frac{2^{k}}{k}$ tends to infinity as $n$ tends to infinity.
## 4. Walking in three dimensions
4.1. Find the conductance of a binary tree of depth (A) 3; (B) 2010; made of unit resistors; see Figure 8 to the left.
4.2. Find the conductance of a modified binary and a trinary trees of depth 2010, in which each resistor at $k$-th level is replaced by $2^{k}$ unit resistors in series; see Figure 8 in the middle.
4.3. Which trees from (A) Problem 4.1; (B) Problem 4.2; can be cut out from the 3-dimensional lattice?

Figure 8: (Left) A binary tree of depth 3; (middle) a modified binary tree of depth 3; (right) allowed intersections of edges of the trees; see Problems 4.1, 4.2, and 4.4.
4.4. - And if one allows intersections of edges at equal distance from the root of the tree; see Figure 8 to the right?
4.5. Prove the Polya theorem for the 3-dimensional lattice.
## 5. Conductance of a ring*
Consider a metal 2-dimensional lattice with unit resistance of each edge.
5.1. The battery is joined with nodes $(0,0)$ and $(1,0)$. Prove that the voltages at the nodes $(2,2)$ and $(3,2)$ are the same; see the remark in Problem 3.4(B).
5.2. Prove that the resistance between any node of a square $n \times n$ lattice and the boundary is less than $\sqrt{n}$.
5.3. The battery is joined with an interior node $A$ of the square $n \times n$ lattice and with the boundary. The voltage at the boundary is zero. Prove that if the current through node $A$ is $\varepsilon$, then the voltage at each other node is less than $\varepsilon \sqrt{n}$.
5.4. Denote by $\Delta f(x, y)=f(x-1, y)+f(x+1, y)+f(x, y-1)+f(x, y+1)-4 f(x, y)$ and $r(x, y)=\sqrt{x^{2}+y^{2}}$. If $f(x, y)=\ln r(x, y)$ for $r(x, y) \geq 2$ then $\Delta f(x, y)=O\left(\frac{1}{r^{4}(x, y)}\right)$.
Hereafter for two functions $A$ and $B$ we write $A=O(B)$, if there exist a positive constant $c$ we have $|A| \leq c B$.
5.5. A ring with inner radius $r_{1} n$, outer radius $r_{2} n$, and center at the origin is cut out from the 2-dimensional lattice. If an edge is cut then the resistance of the part is proportional to its length. Assign the voltage $\ln n r_{1}$ to the inner boundary circle and $\ln n r_{2}$ - to the outer one. Let $U_{n}(x, y)$ be the voltage at node $(x, y)$. Prove that for each $(x, y)$ inside the ring $U_{n}(x, y)=\ln r(x, y)+O\left(\frac{1}{n^{3 / 2}}\right)$.
5.6. Using $\arctan x=x+O\left(x^{3}\right)$ prove that for each $0 \leq y3 r_{1} / 2$ (so that our ring contains a square). Prove that the current incoming through the inner boundary is $2 \pi+O\left(\frac{1}{\sqrt{n}}\right)$. Apply this to
get the following formula for the conductance of the ring:
$$
R\left(r_{1} n, r_{2} n\right)=\frac{1}{2 \pi} \ln \frac{r_{2}}{r_{1}}+O\left(\frac{1}{\sqrt{n}}\right)
$$
5.9. Prove formula (1) without additional assumption $r_{2}>3 r_{1} / 2$.
5.10. Using (1) make estimations in Problems 5.2, 5.3 more precise and prove more precise formulas in Problems 5.5, 5.8:
$$
U_{n}(x, y)=\ln r(x, y)+O\left(\frac{\ln n}{n^{2}}\right), \quad R\left(r_{1} n, r_{2} n\right)=\frac{1}{2 \pi} \ln \frac{r_{2}}{r_{1}}+O\left(\frac{\ln n}{n}\right)
$$
## 6. Challenge*
6.1. Liouville's theorem. Suppose that a function $f(m, n)$ on $\mathbb{Z}^{2}$ satisfies the inequality $0 \leq f(m, n) \leq 1$ and the equality
$$
f(m, n)=\frac{1}{4}(f(m-1, n)+f(m+1, n)+f(m, n-1)+f(m, n+1))
$$
for each $m, n \in \mathbb{Z}$. Prove that the function $f(m, n)$ is constant.
6.2. Existence of a voltage. Prove that there exists a function $f(m, n)$ on $\mathbb{Z}^{2}$ such that $f(0,0)=0, f(0,1)=1$, for each $(m, n) \neq(0,0),(0,1)$ equality (2) holds, and $f(m, n)$ tends to $1 / 2$ as $r(m, n):=\sqrt{m^{2}+n^{2}}$ tends to infinity.
6.3. Green's function. Let $f(m, n)$ be the resistance of the 2 -dimensional lattice between the origin and the point $(m, n)$.
(A) Prove that for each $(m, n) \neq(0,0)$ equality (2) holds.
(B) Prove that $f(m, n)=g(r(m, n))+O(1)$ for some function $g(x)$.
(C) Prove that the resistance between the center and the boundary of a disc of radius $r$ cut from the 2-dimensional lattice equals to $\frac{1}{2 \pi} \ln r+O(1)$.
(D) Prove that $f(m, n)=\frac{1}{2 \pi} \ln r(m, n)+O(1)$.
6.4. Find with 2 digit precision the probability that a random walk on a 3-dimensional lattice eventually returns to the initial point.
6.5. Robot walks on the vertices of the 3-dimensional lattice, each time moving from a vertex to one of the neighbors. One of the vertices contains a treasure, which is found when the robot reaches the vertex. Is there a program for a robot using a finite memory and a random number generator such that the robot finds the treasure with probability 1 ?
## 7. Hints to the solutions
1.1 (A) To check the correctness of the program we use the following criterion: the difference between the percentage and the probability should be approximately inverse proportional to the square root of the number of times the program is run.
(B) See the answer in the table. The proof is straightforward.
Table 2: The probabilities $P_{T}(x)$ and $P(x)$
| | $x$ | 0 | 1 | 2 | 3 | 4 | 5 |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| $T$ | | | | | | | |
| 1 | | 0.00 | 0.00 | 0.00 | 0.00 | 0.50 | 1.00 |
| 2 | | 0.00 | 0.00 | 0.00 | 0.25 | 0.50 | 1.00 |
| 3 | | 0.00 | 0.00 | 0.13 | 0.25 | 0.63 | 1.00 |
| 4 | | 0.00 | 0.06 | 0.13 | 0.38 | 0.63 | 1.00 |
| | $P(x)$ | 0.00 | 0.20 | 0.40 | 0.60 | 0.80 | 1.00 |
(C) Answer: $P(x)=x / 5$; see the last row of the above table.
Proof. Consider a random walk on the integers $0,1,2, \ldots, n$. Let $P(x)$ be the probability, starting at $x$, of reaching $N$ before 0 . We regard $P(x)$ as a function defined on the points $x=0,1,2, \ldots, n$. The function $P(x)$ has the following properties:
1. $P(0)=0$ and $P(n)=1$.
2. $P(x)=\frac{1}{2} P(x-1)+\frac{1}{2} P(x+1)$ for each $x=1,2, \ldots, n-1$.
Property 1 follows from our convention that 0 and $n$ are traps; if the walker reaches one of these positions, he stops there; in the game interpretation, the game ends when one player has all of the pennies. Property 2 states that, for an interior point, the probability $P(x)$ of reaching home from $x$ is the average of the probabilities $P(x-1)$ and $P(x+1)$ of reaching home from the points that the walker may go to from $x$. We can derive property 2 from the following basic fact about probability:
Basic Fact. Let $E$ be any event, and $F$ and $G$ be events such that one and only one of the events $F$ or $G$ will occur. Then
$$
P(E)=P(F) \cdot P(E \text { given } F)+P(G) \cdot P(E \text { given } G)
$$
In this case, let $E$ be the event "the walker ends at the bar", $F$ the event "the first step is to the left", and $G$ the event "the first step is to the right". Then, if the walker starts at $x, P(E)=P(x)$, $P(F)=P(G)=1 / 2, P(E$ given $F)=P(x-1), P(E$ given $G)=P(x+1)$, and property 2 follows.
Properties 1-2 imply together that $P(x)$ is the arithmetic progression $P(x)=x / n$.
1.2 Answer: $P(x)=x / 5$; in fact this problem is equivalent to 1.1(C).
1.3 Answer: $P(x)=\frac{(q / p)^{x}-1}{(q / p)^{5}-1}$.
Hint: Argue as in the solution of Problem 1.1(C). Show that properties 1-2 in the solution should be replaced by
1. $P(0)=0$ and $P(n)=1$.
2. $P(x)=q P(x-1)+p P(x+1)$ for each $x=1,2, \ldots, n-1$.
Show that you can choose $A$ and $B$ so that the function $f(x)=A(q / p)^{x}+B$ satisfies these modified properties.
1.4 Answer: $\approx 99.995 \%$. The exact answer is $1-\frac{(0.55 / 0.45)^{20}-1}{(0.55 / 0.45)^{70}-1}$; the problem is equivalent to 1.3 .
1.5 Answer: $v(x)=x / 5$. Hint. Axioms 1-2 imply that $v(x)$ is linear for this network.
1.6 (A) Let $M$ be the largest value of $v(x)$. Then if $v(x)=M$ for $x \notin P \cup N$, the same must be true for $v(x-1)$ and $v(x+1)$ since $v(x)$ is the average of these two values. If $x-1$ is still an interior point, the same argument implies that $f(x-2)=M$; continuing in this way, we eventually conclude that $f(0)=M$. That same argument works for the minimum value $m$.
(B) Let $h(x)=v(x)-u(x)$. Then if $x$ is any interior point,
$$
\frac{h(x-1)+h(x+1)}{2}=\frac{v(x-1)+v(x+1)}{2}-\frac{u(x-1)+u(x+1)}{2}
$$
and $h(x)$ also satisfies axiom 2. But $h(x)=0$ for $x$ in $P \cup N$, and hence, by the Maximum Principle, the maximum and minimum values of $h$ are 0 . Thus $h(x)=0$ for all $x$, and $v(x)=u(x)$ for all $x$.
(C) Answer: $v(x)=x / n, C=1 / n ; C \rightarrow 0$ and $v(x) \rightarrow 0$ for each fixed $x$ as $n \rightarrow \infty$.
Hint: It is easy to check that the function $f(x):=x / n$ satisfies axioms $1-2$. By uniqueness principle it follows that $v(x)=x / n$.
1.7 Theorem. A random walk, starting at the origin of the 1-dimensional lattice, eventually returns to the origin with probability 1.
Proof. Let $P$ be the probability that a random walk, starting at the origin, eventually returns to the origin. Let $P_{n}$ be the probability that a random walk, starting at the origin, returns to the origin before reaching the points $n$ and $-n$. Assume that all these probabilities exist. Clearly, then $P_{n} \leq P \leq 1$ for each $n$.
Let us prove that $P_{n}=1-1 / n$. After first "move" our walker comes to either point 1 or -1 with probability 1/2. Given that he comes to 1 by Problem 1.1(C) the probability that he returns to the origin before reaching the point $n$ equals to $1-1 / n$. Analogously, given that he comes to -1 the probability that he returns to the origin before reaching the point $-n$ equals to $1-1 / n$. Applying Basic Fact from the solution of Problem 1.1(C) one gets $P_{n}=\frac{1}{2}\left(1-\frac{1}{n}\right)+\frac{1}{2}\left(1-\frac{1}{n}\right)=1-\frac{1}{n}$. (Alternatively, one can observe that $P_{n}=1-C$, where $C=1 / n$ is the conductance of the network from Problem 1.6.)
So $1-1 / n \leq P \leq 1$ for each $n$, hence $P$ must be 1 .
2.1 Answer: see Figure 9 to the left.

Figure 9: The probabilities $P(x)$ or, equivalently, the voltages $v(x)$; see Problems 2.1 and 2.2 .
Hint. The town is shown again in Figure 9 to the right. The probabilities $P(x)$ are denoted by $a, b, c, d$, and $e$. Similarly to 1-dimensional case, the function $P(x)$ satisfies axioms 1-2 from the definition of an electrical network. Thus we get a system os linear equations:
$$
\begin{aligned}
a & =(b+d+2) / 4 \\
b & =(a+c+2) / 4 \\
c & =(d+3) / 4 \\
d & =(a+c+e) / 4 \\
e & =(b+d) / 4
\end{aligned}
$$
Solving the system, we get the answer.
Remark. Finding the exact solution to a "Dirichlet problem" in two dimensions is not always a simple matter, so we will consider two methods for generating approximate solutions.
First let us present a method using random walks. This method is known as a Monte Carlo method, since random walks are random, and gambling involves randomness, and there is a famous gambling casino in Monte Carlo. We start many random walks at $x$ and count the percentage of walks reaching the points marked by $E$. By the law of averages (the law of large numbers in probability theory), the estimate that we obtain this way will approach the true expected probability $P(x)$. This method is a colorful way to solve the problem, but quite inefficient.
Now let us present the more efficient method of relaxations. Recall that we are looking for a function that has specified boundary values, for which the value at any interior point is the average of the values at its neighbors. Begin with any function having the specified boundary values, pick an interior point, and
see what is happening there. In general, the value of the function at the point we are looking at will not be equal to the average of the values at its neighbors. So adjust the value of the function to be equal to the average of the values at its neighbors. Now run through the rest of the interior points, repeating this process. When you have adjusted the values at all of the interior points, the function that results will not satisfy axiom 2 , because most of the time after adjusting the value at a point to be the average value at its neighbors, we afterwards came along and adjusted the values at one or more of those neighbors, thus destroying the harmony. However, the function that results after running through all the interior points is more nearly to satisfy axiom 2 than the function we started with; if we keep repeating this averaging process, running through all of the interior points again and again, the function will approximate more and more closely the solution to our problem.
2.2 Answer: see Figure 9 to the left; this problem is equivalent to 2.1.
2.3 Answer: (A) $2 / 5$;
(B) $1 / 2$; (C
(C) $2 / 7$; (D) $2 / 5$.
Hint. Reduce to Problem 2.5 using the following simple result:
Physical interpretation of probability. The probability that a random walk in a graph $G$, starting at a vertex $a$, reaches a vertex $h$ before returning to the initial point a, equals to
$$
P=C / \operatorname{deg} a
$$
where $C$ is the conductance of the graph $G$ (of unit resistors) between $a$ and $h$, and deg $a$ is the number of edges containing the vertex $a$.
2.4 (C) Hint. The function $v(x)$ is well-defined on the vertices of the network obtained by the shortening. Check that $v(x)$ still satisfies axioms $1-2$.
2.5 Answer: (1A) $6 / 5$; (1B) 2 ; (1C) $6 / 7$; (1D) 2 .
(2A) $12 / 7$; (2B) $12 / 5$; (2C) 30/19; (2D) 30/11.
For a short solution refer to section 3 .
(2A) Hint. We put a unit battery between $a$ and $b$; see Figure 4 to the right. Then, by symmetry, the voltages at $c$ and $d$ will be the same as will those at $e$ and $f$. Thus our circuit is equivalent to the circuit shown in Figure 10 to the left.
Using the laws for the effective resistance of resistors in series and parallel, this network can be successively reduced to a single resistor of resistance 7/12 ohms, as shown in Figure 10 to the right. Thus the effective resistance is $7 / 12$.
2.6 Answer: 1/7. Argue analogously to the solution of Problem 2.3 .
2.7 (A). For a short solution refer to section 3 .
(B) The authors do not know elementary solution of the problem. A nice solution based on discrete Fourier transformation can be found in the book [7].
### 2.8 Refer to section 3.
2.9 (B) Answer: $C \rightarrow 0$ as $n \rightarrow \infty$.
Hint. We apply Monotonicity Law as follows: short together nodes on squares about the origin, as shown in Figure 11 in the top. The network we obtain is equivalent to the network shown in Figure 11 in the middle. Now as $n 1$-ohm resistors in parallel are equivalent to a single resistor of resistance $1 / n$ ohms, the modified network is equivalent to the network shown in Figure 11 in the bottom. The conductance of this network is
$$
\frac{1}{\sum_{k=1}^{n} \frac{1}{8 k-4}}
$$
This number tends to zero as $n$ tends to infinity. As the conductance of the old network can only be smaller, we conclude that it too must tend to zero.
2.10 Hint. Let $P$ be the probability that a random walk on the 2-dimensional lattice, starting at the origin, eventually returns to the origin. Let $P_{n}$ be the probability that a random walk, starting at the origin, returns to the origin before reaching the boundary points of the $2 n \times 2 n$ square centered at the origin. Assume that all these probabilities exist. Clearly, then $P_{n} \leq P \leq 1$ for each $n$. By the physical interpretation of the probability one gets $P_{n}=1-C / 4$, where $C$ is the effective conductance between the

$1 \int^{\frac{7}{5}}$

Figure 10: Simplification of a network; see the solution of Problem 2.5.

Figure 11: Shortening a square network and an equivalent network; see the solution of Problem 2.9.
center and the boundary of the $2 n \times 2 n$ square. By Problem 2.9(B) $C$ tends to zero as $n$ tends to infinity. Thus $P_{n} \rightarrow 1$ as $n \rightarrow \infty$, hence $P$ must be 1 .
3.1-3.3 E. g., see [6].
3.4 (A) First consider the following network: a current flowing through the vertex $A_{1}$ equals to $\frac{n-1}{n}$ and to $-\frac{1}{n}$ for all other vertices. Since graph network is regular the current flowing through the edge $A_{1} A_{2}$ equals to $\frac{1}{k_{1}}\left(1-\frac{1}{n}\right)$. Now consider a similar network: a current flowing through the vertex $A_{2}$ equals to $-\frac{n-1}{n}$ and to $\frac{1}{n}$ for all other vertices. Sum two networks and obtain by superposition principle the network with unit current source connected to $A_{1}$ and $A_{2}$. The current flowing through the edge $A_{1} A_{2}$ in obtained network equals to $\left(\frac{1}{k_{1}}+\frac{1}{k_{2}}\right)\left(1-\frac{1}{n}\right)$. Therefore, the resistance between $A_{1}$ and $A_{2}$ equals to $\left(\frac{1}{k_{1}}+\frac{1}{k_{2}}\right)\left(1-\frac{1}{n}\right)$.
If the graph is infinite substitute $\frac{1}{n}$ by zero. An explanation of this answer bases on limiting process.
The formula of the resistance between two adjacent vertices of the regular graph is due to A.B.Hodulev. The definition of a regular graph and this formula are taken from [4].
(B) First we introduce some physical explanations. Apply a bus of zero resistance to the boundary of rectangle $[-N, N+1] \times[-N, N]$. Since the potential tends to zero at infinity, applying a bus changes the resistance a little bit. (Next we give our explanations with error which tends to zero as $N$ grows to infinity). If a current source of unit current is connected with point $(0,0)$ and the bus then the current flowing through the edges outgoing from $(0,0)$ equals to $\frac{1}{4}$. If current source of unit current is connected to the bus and the point $(1,0)$ then the current flowing through the edges ingoing to $(1,0)$ equals to $\frac{1}{4}$. Therefore if both current sources are connected then the current flowing through the edge $(0,0)-(1,0)$ equals to $\frac{1}{2}$. So a potential difference between these points equals to $\frac{1}{2}$ also. But the current flowing through source connecting these points equals to 1 . Therefore the resistance between these points equals to $\frac{1}{2}$.
Now we make our arguments more strict. Again suppose that a current source is connected to the point $(0,0)$ and a bus of zero resistance applied to the boundary of rectangle $[-N, N+1] \times[-N, N]$. According to Problem 5.2 the resistance of such graph is at most $N^{1 / 2}$. Hence if the bus is joined with the ground, the potential at the point $(0,0)$ is at most $N^{\frac{1}{2}}$ :
$$
V=I R=R \leq N^{1 / 2}
$$
We will show that the potential at the points which are close to the bus differs from zero a little. Denote the maximal potential at the points of the boundary of the rectangle $[-j, j+1] \times[-j, j]$ by $u_{j}$. From harmonicity of potential distribution it follows that $u_{j-1} \geq 2 u_{j}-u_{j+1}(1 \leq j \leq N-1)$. So if $u_{N-1}=\varepsilon$ (by assumption $\left.u_{n}=0\right)$ then for all $j$ such that $0 \leq j \leq N$ an inequality $u_{j} \geq(N-j) \varepsilon$ holds. In particular, $u_{0}=V \geq N \varepsilon$. Using inequality (3), we obtain that $\varepsilon \leq N^{-1 / 2}$.
This time apply a bus to the boundary of the square $[-N, N] \times[-N, N]$. The obtained potential distribution is symmetric. Due to the maximum principle it differs from the initial one at most by $N^{-1 / 2}$. Therefore, before the moving of the right side of the bus four currents outcoming from the point $(0,0)$ differed from $\frac{1}{4}$ at most by $N^{-\frac{1}{2}}$. Similarly, if the source is connected to a bus and the point $(1,0)$ then ingoing currents to the point $(1,0)$ differ from $\frac{1}{4}$ at most by $N^{-\frac{1}{2}}$.
Unite these two situations. So by superposition principle we obtain that in the network obtained from the rectangle by short-circuiting its boundary and connecting the unit current source to the points $(0,0)$ and $(1,0)$ the current flowing through the edge $(0,0)-(1,0)$ equals to $1 / 2+O\left(N^{-1 / 2}\right)$. Hence the potential difference and, therefore, resistance equal to $1 / 2+O\left(N^{-1 / 2}\right)$.
Consider the initial network. To finish the proof apply the axiom 3. Assume that potentials at the points $(0,0)$ and $(1,0)$ equal to $\frac{1}{4}$ and $-\frac{1}{4}$ respectively. Denote the current flowing through the source by $I$. Choose the rectangle $[-N, N+1] \times[-N, N]$ such that a potential on its boundary is smaller than some $\varepsilon>0$. By maximum principle if one substitute all potentials at the boundary of the rectangle by zero (fixing the current through the battery) then potentials at all points will change at most by $\varepsilon$. In particular the potential difference between $(0,0)$ and $(1,0)$ will equal to $1 / 2+O(\varepsilon)$. But resistance between them equals to $1 / 2+O\left(N^{-1 / 2}\right)$. Therefore $I=1+O(\varepsilon)+O\left(N^{-1 / 2}\right)$. As $\varepsilon$ tends to zero, $N$ grows to infinity and we obtain that $I=1$. Therefore the resistance of the lattice between adjacent vertices equals to $\frac{1}{2}$.

Figure 12: Calculation of the resistance of a tree; see solution of Problem 4.2.
3.5 Let $B_{1}$ and $B_{2}$ be the vertices of the graph which are opposite to $A_{1}$ and $A_{2}$ respectively. By Problem 3.4(A) the current flowing through the edge $A_{1} A_{2}$ equals to $I=\left(\frac{1}{k_{1}}+\frac{1}{k_{2}}\right)\left(1-\frac{1}{n}\right)$ if the source of given network is of unit current. Denote the current flowing through the edge $B_{2} B_{1}$ by $x$. Additionally connect the unit current source to vertices $B_{1}$ and $B_{2}$ such that the current flows in the vertex $B_{2}$. Then the current flowing through edges $A_{1} A_{2}$ and $B_{2} B_{1}$ equals to $I+x$. But if one connect two unit current sources to $A_{1}, B_{1}$ and $B_{2}, A_{2}$ (current flows in $A_{1}$ and in $B_{2}$ ) then the current distribution will be the same. By Problem 3.4 the first source gives a current $\frac{1}{k_{1}}$ through $A_{1} A_{2}$ and the second $-\frac{1}{k_{2}}$. Therefore
$$
\begin{gathered}
I+x=\left(\frac{1}{k_{1}}+\frac{1}{k_{2}}\right)\left(1-\frac{1}{n}\right)+x=\frac{1}{k_{1}}+\frac{1}{k_{2}} \\
x=\left(\frac{1}{k_{1}}+\frac{1}{k_{2}}\right) \frac{1}{n}, \quad \frac{x}{I}=\frac{1}{n-1}
\end{gathered}
$$
So for icosahedron the answer is $\frac{I}{11}$, for a dodecahedron $-\frac{I}{19}$, for a rhombdodecahedron $-\frac{I}{13}$, for a cube $-\frac{I}{7}$.
3.6 Hint. Sums $D_{n}$ and $F_{n}$ fit the same recurrence relation. For instance, $D_{n}=1+\frac{n+1}{2 n} D_{n-1}$. Also $D_{0}=F_{0}=1$. Therefore $D_{n}=F_{n}$. So if $n \geq 1$ then
$$
\sum_{k=1}^{n} \frac{2^{k}}{k}=\frac{2^{n}}{n} F_{n-1}=\frac{2^{n}}{n} D_{n-1}=\frac{2^{n}}{n} \sum_{k=0}^{n-1} \frac{1}{C_{n-1}^{k}}
$$
To finish the proof bound the power of the number 2 in the factorization of the common denominator of the fractions from the obtained sum using Legendre formula for the power of the prime number in the factorization of a factorial.
The resistance $R_{n}$ between two opposite vertices of an $n$-dimensional cube (with unit resistance on each its edge) is related with the sums $D_{n}$ and $F_{n}$ by the relations
$$
D_{n}=F_{n}=(n+1) R_{n+1}
$$
(see details in the paper [5]).
4.1 Hint. Prove by induction that the resistance of a binary tree of depth $n$ constructed of unit resistors equals to $1-\frac{1}{2^{n}}$.
4.2 Hint. Voltages at points situated at the same distance from the tree root are equal in virtue of symmetry. Shortening such points in a binary tree we receive a series from Figure 12. It's resistance equals to $\frac{1}{2} \cdot n=\frac{n}{2}$. Similarly for a trinary tree we receive $R=\frac{1}{3}+\frac{2}{9}+\cdots+\frac{2^{n-1}}{3^{n}}$. Hence $R=1-\frac{2^{n}}{3^{n}}$.
4.3 Hint. It is not difficult to cut a tree of depth 3 . Let us show that it is impossible to cut a tree of depth 2010. Suppose we cut it, thus all it's vertices are situated at the distance not more than 2010 from the root; hence the tree is contained in a cube with the side $2 \cdot 2010+1$. So it has no more than $4021^{3} \leq 2^{36}$ vertices. On the other hand the number of it's vertices equals to $2^{2011}-1$. The contradiction completes the proof. The problem of cutting of a modified tree couldn't be solved easily.
4.4 Hint. A binary tree could not be cut; the arguments are similar to the solution 4.3, taking in account that more than two vertices could not be glued. A modified binary tree could be cut from the plane(see


Figure 13: Cutting of a binary tree with intersections from the plain; see solution 4.4.

Figure 14: Cutting of a trinary tree with intersections from the space; see solution 4.4.
Figure 13), and similarly a trinary tree could be cut from the space (see Figure 14). Proofs could be done using induction by the tree depth.
4.5 Hint. For any $n=2^{i}-1$ let us consider the set of vertices $(x, y, z)$, where $|x|+|y|+|z| \leq n$. Let $R_{i}$ be the resistance between the origin and the border of the figure. As we already know from Problem 4.4, it is possible to cut from such a part of the lattice a modified trinary tree of depth $i$ with edges intersections at the same distance from the root. As we know from Problem 4.2, the resistances of modified trinary trees not more than 1 . So the resistances of modified trinary trees also not more than 1 . From the monotonicity law we receive that $R_{i} \leq 1$. Hence switching on a battery of 1 Volt the current will be not more than 1. Hence the voltages at the vertices joint to the origin will be not more than $1-\frac{1}{6}=\frac{5}{6}$. They equal to the probability or returning to the origin before reaching of the border. Taking the limit we obtain the statement we need.
5.1 Consider the points $A(2,3), B(3,3), C(1,2), D(2,2), E(3,2), F(4,2), G(2,1), H(3,1)$. First let the power supply deliver a unit current to the origin, the second clip being connected to the perimeter of the square $[-R, R]^{2}$ (we set the zero resistance to this perimeter). From the symmetry reasons we get the equalities $i(C D)=i(G D)=i(D A)=i(D E)=I$ for some number $I$.
Now, consider the second situation, when the same current is delivered to the perimeter of the same square (with zero resistance as well), the second clip being connected to the point $(1,0)$. Then we get $i(D E) \approx i(H E) \approx i(E B) \approx i(E F) \approx-I$, where the equalties are accurate within some small $\varepsilon$ which tends to zero as $R \rightarrow \infty$ (similarly to Problem 3.4, it follows from axiom 3). Combining both situations, we get the following. Suppose that the power supply is connected to $(0,0)$ and $(1,0)$, while the perimeter of square $[-R, R]^{2}$ is replaced by a zero resistance loop; then the current flowing through edge $D E$ is less than $\varepsilon$. Taking the limit as $R \rightarrow \infty$ (and applying axiom 3 again), we obtain the desired statement.
Note that we used without proof a difficult fact on the existence and uniqueness.
5.2 Consider an arbitrary tree connecting the given point with the perimeter of the square.
### 5.3 Apply the maximum principle.
5.4 Write the Laplace operator in the form
$$
\Delta f(x, y)=f(x-1, y)+f(x+1, y)-2 f(x, y)+f(x, y-1)+f(x, y+1)-2 f(x, y)
$$
Then for the function $f(x, y)=\ln r(x, y)$ we get
$$
\begin{gathered}
f(x-1, y)+f(x+1, y)-2 f(x, y)=\frac{1}{2} \ln \frac{\left((x+1)^{2}+y^{2}\right)\left((x-1)^{2}+y^{2}\right)}{\left(x^{2}+y^{2}\right)^{2}}= \\
=\frac{1}{2} \ln \left(1+\frac{2 x+1}{r^{2}}\right)\left(1+\frac{-2 x+1}{r^{2}}\right)=\frac{1}{2} \ln \left(\left(1+\frac{1}{r^{2}}\right)^{2}-\frac{4 x^{2}}{r^{4}}\right)
\end{gathered}
$$
Similarly,
$$
f(x, y-1)+f(x, y+1)-2 f(x, y)=\frac{1}{2} \ln \left(\left(1+\frac{1}{r^{2}}\right)^{2}-\frac{4 y^{2}}{r^{4}}\right)
$$
Hence
$$
\begin{aligned}
& \Delta f(x, y)=\frac{1}{2} \ln \left(\left(1+\frac{1}{r^{2}}\right)^{2}-\frac{4 x^{2}}{r^{4}}\right)\left(\left(1+\frac{1}{r^{2}}\right)^{2}-\frac{4 y^{2}}{r^{4}}\right)= \\
& =\frac{1}{2} \ln \left(1-\frac{1}{r^{4}}+\frac{16 x^{2} y^{2}}{r^{8}}\right)=\frac{1}{2} \ln \left(1+O\left(\frac{1}{r^{4}}\right)\right)=O\left(\frac{1}{r^{4}}\right)
\end{aligned}
$$
5.5 For a point $(x, y)$ nearby the boundary of the ring, it makes sense to change the definition of the Laplace operator, in order to agree with the Kirchhoff rules. For instance, if the points $(x-a, y)$ and $(x, y-b)$ for some $a, b \in[0,1)$ lie on the boundary, then we set
$$
\Delta f(x, y)=\frac{f(x-a, y)-f(x, y)}{a}+\frac{f(x, y-b)-f(x, y)}{b}+f(x+1, y)+f(x, y+1)-2 f(x, y)
$$
Thus, considering the function the function $f(x, y)=\ln r(x, y)$ in such a point we have
$$
\frac{f(x-a, y)-f(x, y)}{a}+f(x+1, y)-f(x, y)=\frac{1}{2 a} \ln \left(1+\frac{-2 a x+a^{2}}{r^{2}}\right)+\frac{1}{2} \ln \left(1+\frac{2 x+1}{r^{2}}\right)=O\left(\frac{1}{r^{2}}\right)
$$
Analogously,
$$
\frac{f(x, y-b)-f(x, y)}{b}+f(x, y+1)-f(x, y)=O\left(\frac{1}{r^{2}}\right)
$$
So, we get $\Delta f(x, y)=O\left(r^{-2}\right)$; moreover, this estimate remains valid if only one of the points neighboring to $(x, y)$ lies outside the ring.
Consider now the function $f(x, y)=U_{n}(x, y)-\ln r(x, y)$. It vanishes on the boundary of the ring, while in each interior point it satisfies the equation $\Delta f(x, y)=\varphi(x, y)$, where $\varphi(x, y)=O\left(n^{-2}\right)$ at the points nearby the boundary, and $\varphi(x, y)=O\left(n^{-4}\right)$ at all other points of the ring.
Interpret the values $\varphi(x, y)$ as the currents delivered to the corresponding points of the ring. The potentials at the points nearby the boundary can be estimated by the maximum principle (cf. Problem 3.1). Actually, the potential on the boundary is zero, the current between our point $(x, y)$ and the closest point of the boundary is $O\left(n^{-2}\right)$, therefore the potential at $(x, y)$ is $O\left(n^{-2}\right)$ as well. By the maximum principle, all such currents induce the potentials not exceeding $O\left(n^{-2}\right)$ in all points of our region.
Now, we are left to estimate the potentials generated by the currents at other points of the ring (that are those far from the boundary). The number of those points is $O\left(n^{2}\right)$, and the current in each of them induces the potentials not exceeding $O\left(n^{-7 / 2}\right.$ ) (according to Problem 5.3). Thus, the total potential generated by our points is $O\left(n^{-3 / 2}\right)$. Hence, we finally get $f(x, y)=O\left(n^{-3 / 2}\right)$.
5.6 We involve the relation
$$
\arctan x-\arctan y=\arctan \frac{x-y}{1+x y}
$$
which holds when $|x y|<1$. So, we get
$$
\begin{aligned}
& \arctan \frac{y+1}{R}-\arctan \frac{y}{R}=\arctan \frac{1 / R}{1+y(y+1) / R^{2}}= \\
= & \arctan \left(\frac{R}{y^{2}+R^{2}}+O\left(\frac{1}{R^{2}}\right)\right)=\frac{R}{R^{2}+y^{2}}+O\left(\frac{1}{R^{2}}\right)
\end{aligned}
$$
5.7 Sum up the formula from problem 5.6.
5.8 To find the resistance of the ring, we find the current flowing through it when the inner and the outer loops of the ring are under the voltages $\ln n r_{1}$ and $\ln n r_{2}$ respectively. Let us find the current through the perimeter of a square $[-R-1 / 2, R+1 / 2]^{2}$, where $R=\left[r_{1} n\right]+1$. We will calcuat it approximately, changing the potentials at all the points by the corresponding values of the function $\ln r(x, y)$. Thus, we sum up $O(n)$ of currents, each with the error of $O\left(n^{-3 / 2}\right)$. Hence the total error is $O\left(n^{-1 / 2}\right)$.
By the symmetry of the square, the current flowing trough its perimeter can be found as
$$
I=8 \sum_{y=0}^{R}(\ln r(R+1, y)-\ln r(R, y))+O\left(\frac{1}{n^{1 / 2}}\right)
$$
Since
$$
\ln r(R+1, y)-\ln r(R, y)=\frac{R}{R^{2}+y^{2}}+O\left(\frac{1}{R^{2}}\right)
$$
we can apply the formula from problem 5.7 obtaining the desired relation $I=2 \pi+O\left(n^{-1 / 2}\right)$.
5.9 Similarly to the previous problem, to calculate the resistance we will find the current flowing through some closed broken line. Again, for the approximate calculation we replace the potentials by the values of the function $\ln r(x, y)$. Now let us replace our broken line by the square circumscribed around it; there will be $O\left(n^{2}\right)$ new current sources inside this contour, in each of them there appears an (incoming or outcoming) current of order $O\left(n^{-4}\right)$. Therefore, the desired value of the total current differs from the
current $I=2 \pi+O\left(n^{-1 / 2}\right)$ through the perimeter of the square (which was found above) by at most $O\left(n^{-2}\right)$.
5.10 Let us prove that in Problem 5.2 the resistance between any vertex of the square and its boundary equals to $O(\ln n)$. This allows us to multiply residual terms by $\frac{\ln n}{\sqrt{n}}$ in solutions of all next problems.
Instead of the square consider a triangle cut from the square lattice by lines $x=0, y=0, x+y=n$, and bound its resistance between its origin and its hypotenuse. Assume that a potential at integer points of the segment $x+y=k, x, y \geq 0$ equals to $V_{k}=\sum_{j=2}^{k+1} \frac{1}{j}(0 \leq k \leq n)$. In particular the potential at the origin equals zero. Also assume that the current flowing into each vertex at the line $x+y=k$ equals to $\frac{1}{k}$, i.e. through each level flows unit current. Increase resistances inside the triangle to fit Ohm law. (Resistances connecting points of the form $(0, k),(k, 0)$ with points $(0, k+1),(k+1,0)$ stay unit.) To fit Kirchhoff law currents flowing from the point $(j, k-j)$ should equal to $\frac{k-j}{k(k+1)}$ and $\frac{j+1}{k(k+1)}$ and flow to the points $(j, k-j+1)$ and $(j+1, k-j)$ respectively. Since potential difference equals to $\frac{1}{n+1}$ then unit resistances must be substituted by $\frac{k}{k-j}$ and $\frac{k}{j+1}$ respectively. The resistance of the obtained network equals to $V_{n} \leq \ln n$. So the resistance of the initial network is smaller than $\ln n$ also.
## 8. Acknowledgements
Most of the problems from sections 1, 2 and 4 are taken from the paper of P. Doyle and J. Snell [3]. The authors are grateful to I. Bogdanov, V. Bugaenko and M. Prasolov for help in translation.
[1] I. Benjamini and O. Schramm, Random walks and harmonic functions on infinite planar graphs using square tilings, Ann. Prob. 24:3 (1996), 1219-1238.
[2] J. Cannon, W. Floyd, W. Parry, Squaring rectangles: the finite Riemann mapping theorem, Contemp. Math. 169 (1994), 133-211.
[3] P. G. Doyle and J. L. Snell, Random walks and electric networks, Mathematical Association of America, 1984, http://arxiv.org/abs/math.PR/0001057.
[4] G. A. Galperin, My friend Andrey Khodulev, Mat. Prosv. 3rd series, 4 (2000), 8-32, http://www. mccme.ru/free-books/matpros/i5008032.pdf.zip.
[5] F. Nedemeyer, A. Smorodinskiy, Resistance of edges of higher dimensional cube, Kvant 6 (1986), in Russian.
[6] M. Prasolov and M. Skopenkov, Tiling by rerctangles and alternating current, submitted (2010). http: //arxiv.org/abs/1002.1356.
[7] F. Spitzer, Principles of random walks, Springer-Verlag, 1976.
# Tiling by rectangles and alternating current
M. Prasolov ${ }^{\mathrm{a}}$, M. Skopenkov ${ }^{\mathrm{b}, \mathrm{c}, *}$
${ }^{a}$ Moscow State University, Faculty of Mechanics and mathematics,
Leninskie Gory, 1, GSP-1, Moscow, 119991, Russian Federation
${ }^{b}$ Institute for information transmission problems of the Russian Academy of Sciences,
Bolshoy Karetny per. 19, bld. 1, Moscow, 127994, Russian Federation
${ }^{c}$ King Abdullah University of Science and Technology,
P.O. Box 2187, 4700 KAUST, 23955-6900 Thuwal, Kingdom of Saudi Arabia
\begin{abstract}
This paper is on tilings of polygons by rectangles. A celebrated physical interpretation of such tilings due to R.L. Brooks, C.A.B. Smith, A.H. Stone and W.T. Tutte uses directcurrent circuits. The new approach of the paper is an application of alternating-current circuits. The following results are obtained:
- a necessary condition for a rectangle to be tilable by rectangles of given shapes;
- a criterion for a rectangle to be tilable by rectangles similar to it but not all homothetic to it;
- a criterion for a generic polygon to be tilable by squares.
\end{abstract}
These results generalize the ones of C. Freiling, R. Kenyon, M. Laczkovich, D. Rinne and G. Szekeres.
Keywords:
Tiling, rectangle, orthogonal polygon, alternating current
2010 MSC: $52 \mathrm{C} 20,94 \mathrm{C} 05,31 \mathrm{C} 20,30 \mathrm{C} 15,60 \mathrm{~J} 10$
## 1. Introduction
A rectangle $a \times b$, where $a$ and $b$ are integers, can be tiled by $a \cdot b$ squares. Thus a rectangle with rational side ratio can be tiled by squares. In 1903 M. Dehn proved the converse assertion:
Theorem 1.1. [10] A rectangle can be tiled by squares (not necessarily equal) if and only if the ratio of two orthogonal sides of the rectangle is rational.
Although this assertion is expectable, the proof is complicated. After original proof, many improvements have been made $[2,3,18,25,32]$.
The most interesting for us is the approach of R.L. Brooks, C.A.B. Smith, A.H. Stone and W.T. Tutte [3]. To a tiling of a rectangle they assign a direct-current circuit, and then deduce Theorem 1.1 from certain properties of the circuit. They also apply the technique to find a tiling of a square by squares of distinct sizes, see the figure in the front cover of the journal [13].[^6]
We study finite tilings by arbitrary nondegenerate rectangles. The sides of rectangles are assumed to be parallel to coordinate axes, i.e., either vertical or horizontal. By the ratio of a rectangle we mean the length of the horizontal side divided by the length of the vertical one. We study the following problem posed in $[16$, p. 218] and $[19$, p. 3]:
Problem 1.2. Which rectangles can be tiled by rectangles of given ratios $c_{1}, \ldots, c_{n}$ ?
A related problem of signed tilings is solved in [19].
For $n=1$ and $c_{1}=1$ the question of Problem 1.2 is answered by Theorem 1.1. A necessary condition for arbitrary $n$ was actually proved by M. Dehn: if a rectangle of ratio $c$ can be tiled by rectangles of ratios $c_{1}, \ldots, c_{n}$ then $c$ is (the value of) a rational function in $c_{1}, \ldots, c_{n}$ with rational coefficients.
This function depends only on "combinatorial structure" of the tiling. For instance, if a rectangle of ratio $c$ is dissected into 2 rectangles of ratios $c_{1}$ and $c_{2}$ by a vertical (respectively, horizontal) cut then $c\left(c_{1}, c_{2}\right)=c_{1}+c_{2}$ (respectively, $\left.c\left(c_{1}, c_{2}\right)=\frac{c_{1} c_{2}}{c_{1}+c_{2}}\right)$. The problem reduces to description of possible functions $c\left(c_{1}, \ldots, c_{n}\right)$. By the mentioned physical interpretation this is equivalent to a natural problem: describe possible formulas $c\left(c_{1}, \ldots, c_{n}\right)$ expressing the conductance of a planar direct-current circuit through the conductances $c_{1}$, $\ldots, c_{n}$ of individual resistors.
The main idea of the paper is to apply alternating-current circuits (equivalently, circuits with complex-valued conductances) to the above problems. Our first result is
Theorem 1.3. Suppose that a rectangle of ratio $c$ can be tiled by rectangles of ratios $c_{1}$, $\ldots, c_{n}$. Then $c=C\left(c_{1}, \ldots, c_{n}\right)$ for some rational function $C\left(z_{1}, \ldots, z_{n}\right)$ such that
(1) $C\left(z_{1}, \ldots, z_{n}\right)$ has rational coefficients, i.e., $C\left(z_{1}, \ldots, z_{n}\right) \in \mathbb{Q}\left(z_{1}, \ldots, z_{n}\right)$;
(2) $C\left(z_{1}, \ldots, z_{n}\right)$ is degree 1 homogeneous, i.e., $C\left(t z_{1}, \ldots, t z_{n}\right)=t C\left(z_{1}, \ldots, z_{n}\right)$;
(3) if $\operatorname{Re} z_{1}, \ldots, \operatorname{Re} z_{n}>0$ then $\operatorname{Re} C\left(z_{1}, \ldots, z_{n}\right)>0$.
Problem 1.4. Is the converse theorem true for $n \geq 3$ ?
Parts (1) and (2) of Theorem 1.3 were actually proved by Dehn, see also [17, Lemma 4]. Case $n=1$ (respectively, $n=2$ ) of both Theorem 1.3 and its converse is equivalent to Theorem 1.1 (respectively, to [16, Theorem 5], see also Theorem 3.1 below). For $n \geq 3$ the converse theorem cannot be proved by our method, see Example 3.2.
Theorem 1.3 has a clear physical meaning, see $\S 2.4$. But this theorem (even together with its converse) is not algorithmic, i.e., it does not give an algorithm to decide if there exists a required tiling. Thus it is interesting to get less general but algorithmic results.
A result of this kind was obtained independently by C. Freiling, D. Rinne in 1994 and M. Laczkovich, G. Szekeres in 1995. It uses the following notion. An algebraic conjugate of an algebraic number $c$ is a complex root of the minimal integral polynomial of $c$.
Theorem 1.5. [17,22] For $c>0$ the following 3 conditions are equivalent:
(1) a square can be tiled by rectangles of ratios $c$ and $1 / c$;
(2) the number $c$ is algebraic and all its algebraic conjugates have positive real parts;
(3) for certain positive rational numbers $d_{1}, \ldots, d_{m}$ we have
$$
d_{1} c+\frac{1}{d_{2} c+\cdots+\frac{1}{d_{m} c}}=1
$$
We present a new short self-contained proof of this result. This new proof is an example of a natural application of alternating-current circuits. We also get a new algorithmic result:
Theorem 1.6. For a number $c>0$ the following 3 conditions are equivalent:
(1) a rectangle of ratio $c$ can be tiled by rectangles of ratios $c$ and $1 / c$ (in such a way that there is at least one rectangle of ratio $1 / c$ in the tiling);
(2) the number $c^{2}$ is algebraic and all its algebraic conjugates distinct from $c^{2}$ are negative real numbers.
(3) for certain positive rational numbers $d_{1}, \ldots, d_{m}$ we have
$$
\frac{1}{d_{1} c+\frac{1}{d_{2} c+\cdots+\frac{1}{d_{m} c}}}=c
$$
More algorithmic results can be found in [16, p. 224]. For similar results on tiling by triangles see [29]. For higher dimensional generalizations see [25].
We also consider tilings of arbitrary (not necessarily convex) polygons by rectangles. This generalization reveals new connections between tilings and electrical circuits.
We apply direct-current circuits with several terminals to get a criterion for a generic polygon to be tilable by squares (Theorem 4.2 below, again not algorithmic). This result generalizes Theorem 1.1 and [21, Theorems 9 and 12]. An easier related problem of signed tiling by squares is solved in $[15,20]$.
We apply alternating-current circuits with several terminals to get a short proof of a generalization of Theorem 1.5 to polygons with rational vertices [28] (Theorem 4.3 below). We also give basic results on electrical impedance tomography for alternatingcurrent circuits, cf. [9, 6, 7, 23].
There is a close relationship among electrical circuits, discrete harmonic functions and random walks on graphs $[11,24,1]$. Our results have equivalent statements in the language of each of the theories, e.g., see Corollary 4.9 below.
The paper splits naturally into two formally independent parts: $\S \S 1-3$ and $\S \S 4-6$.
The first part contains the proof of Theorems 1.3, 1.5 and 1.6. In $\S 2$ the basics of electrical circuits and their connection with tilings are recalled. In $\S 3$ the results of $\S 1$ are proved.
The second part concerns some variations. In $\S 4$ the results on tilings of polygons, electrical impedance tomography and random walks are stated. In $\S 5$ the results of $\S 2$ are generalized to electrical circuits with several terminals. In $\S 6$ the results of $\S 4$ are proved.
## 2. Main ideas
### 2.1. Electrical circuits
Our approach is based on electrical circuits theory [26]. However, the reader is not assumed to be familiar with physics. In this section we recall all the required physical concepts (although the presentation is formal and physical meaning is explained very briefly). This section does not contain new results. For short proofs see $\S 5$.
An electrical network is a connected graph with a nonnegative real number (conductance) assigned to each edge, and two marked (boundary) vertices.
For simplicity assume that the graph does not have neither multiple edges nor loops. Although all the concepts below can be adopted easily for the graphs with multiple edges. We say that electrical network is planar if the graph is drawn in the unit disc in such a way that the boundary vertices are in the boundary of the disc and the edges do not intersect each other.
Fix an enumeration of the vertices $1,2, \ldots, n$ of the graph such that 1 and 2 are the boundary ones. It is convenient to denote the number of boundary vertices by $b=2$. Let $m$ the number of edges. Denote by $c_{k l}$ the conductance of the edge between the vertices $k$ and $l$. Set $c_{k l}=0$ if there is no edge between $k$ and $l$ in the graph.
An electrical circuit is an electrical network along with two real numbers $U_{1}$ and $U_{2}$ (incoming voltages) assigned to the boundary vertices.
Each electrical circuit gives rise to certain numbers $U_{k}$, where $1 \leq k \leq n$ (voltages at the vertices), and $I_{k l}$, where $1 \leq k, l \leq n$ (currents through the edges). These numbers are defined by the following axioms:
(C) The Ohm law. For each pair of vertices $k, l$ we have $I_{k l}=c_{k l}\left(U_{k}-U_{l}\right)$.
(I) The Kirchhoff current law. For each vertex $k>b$ we have $\sum_{l=1}^{n} I_{k l}=0$.
Informal meaning of law (I) is that electrical charge is not aggregated at the nonboundary vertices. In other words, these laws assert that $U_{k}$ is a discrete harmonic function. The numbers $U_{k}$ and $I_{k l}$ are well-defined by these axioms by the following classical result.
Theorem 2.1. [31] For any electrical circuit the system of linear equations (C),(I) in variables $U_{k}, b0$ then $\frac{\partial}{\partial c_{j}} C\left(c_{1}, \ldots, c_{m}\right) \geq 0$; if the edge $j$ is essential then the latter inequality is strict;
(5) [5] if $\operatorname{Re} c_{1}, \ldots, \operatorname{Re} c_{m}>0$ then $\operatorname{Re} C\left(c_{1}, \ldots, c_{m}\right)>0$.
Remark 2.4. (A. Akopyan, private communication) Property (4) follows from (1), (2) and (5). Property (5) does not follow from (1), (2) and (4), e.g., the function $C\left(c_{1}, c_{2}\right)=$ $\left(c_{1}+c_{2}\right) \frac{c_{1}^{2}+c_{2}^{2}}{c_{1}^{2}+2 c_{2}^{2}}$ satisfies (1), (2), (4) but not (5).
Property (5) concerns the extension of the function $C\left(c_{1}, \ldots, c_{m}\right)$ to the complex plane. This fundamental property does not seem to be payed attention for direct-current circuits. Certainly it is well-known for alternating-current circuits. Short proof of the lemma is given in $\S 5.1$.
### 2.4. Alternating-current circuits
Let us explain informal physical meaning of fundamental Lemma 2.3(5) and condition (3) of Theorem 1.3. This is not used elsewhere in the paper and the reader may easily skip this subsection.
Informally, an alternating-current circuit is a collection of conductors, condensers, inductors and a single alternating-voltage source connected with each other.
Formally, an alternating-current circuit is a graph with the following structure:
- two marked (boundary) vertices;
- two functions (voltages) $\tilde{U}_{1}(t)=U \cos \omega t$ and $\tilde{U}_{2}(t)=0$ assigned to them;
- division the edges into three types (conductors, condensers and inductors);
- a positive number $\tilde{c}_{k l}$ assigned to each edge (called conductance, capacitance or inductance, depending on the type of the edge).
The voltages $\tilde{U}_{k}(t)$ and the currents $\tilde{I}_{k l}(t)$ are defined by the following axioms:
(官) The generalized Ohm law. For each edge $k l$ we have
$$
\tilde{I}_{k l}(t)= \begin{cases}\tilde{c}_{k l}\left(\tilde{U}_{k}(t)-\tilde{U}_{l}(t)\right) & \text { if } k l \text { is a conductor } \\ \tilde{c}_{k l} \frac{d}{d t}\left(\tilde{U}_{k}(t)-\tilde{U}_{l}(t)\right) & \text { if } k l \text { is a condenser } \\ \tilde{c}_{k l} \int_{\pi / 2 \omega}^{t}\left(\tilde{U}_{k}(t)-\tilde{U}_{l}(t)\right) d t & \text { if } k l \text { is an inductor }\end{cases}
$$
(I) The Kirchhoff current law. For each vertex $k \neq 1,2$ we have $\sum_{l=1}^{n} \tilde{I}_{k l}(t)=0$.
The voltages and the currents can be found using the following well-known algorithm. Denote by $i=\sqrt{-1}$. Put $U_{1}=U, U_{2}=0$ and
$$
c_{k l}=\left\{\begin{aligned}
\tilde{c}_{k l}, & \text { if } k l \text { is a conductor } \\
i \omega \tilde{c}_{k l}, & \text { if } k l \text { is a condenser } \\
\frac{1}{i \omega} \tilde{c}_{k l}, & \text { if } k l \text { is an inductor. }
\end{aligned}\right.
$$
Define the complex numbers $U_{k}, 3 \leq k \leq n$, and $I_{k l}, 1 \leq k, l \leq n$, by direct-current laws (C), (I). Then $\tilde{U}_{k}(t)=\operatorname{Re}\left(U_{k} e^{i \omega t}\right), \tilde{\tilde{I}}_{k l}(t)=\operatorname{Re}\left(I_{k l} e^{i \omega t}\right)$. In this sense alternating-current circuits are "equivalent" to direct-current circuits with complex-valued conductances (also called admittances).
Notice that always $\operatorname{Re} c_{k l} \geq 0$. Physically this means nonnegative energy dissipation at the edge $k l$ (which is $\operatorname{Re} c_{k l}\left|U_{k}-U_{l}\right|^{2}$ ). Thus a physical meaning of Lemma 2.3(5) is: "a network consisting of elements dissipating energy also dissipates energy".
### 2.5. Positive real functions
This subsection is used in the proof of only assertions $(2) \Longrightarrow(3)$ in Theorems 1.5 and 1.6 .
Consider electrical circuits, in which all the edges have conductances $z$ and $1 / z, \operatorname{Re} z>$ 0 . (They have a natural physical meaning: circuits consisting of condensers and inductors with incoming voltage of complex frequency $z / i$.) Let us describe possible conductances $C(z)$ of such electrical circuits. By Lemma 2.3(1), (2) and (5) the functions $C(z)$ are positive real, i.e., satisfy condition (1) of the following lemma. Denote by $\operatorname{Re} \infty=0$, $C(\infty)=\lim _{z \rightarrow 0} C(1 / z)$ and $C^{\prime}(\infty)=\lim _{z \rightarrow 0}(C(1 / z))^{\prime}$.
Lemma 2.5. [5, 14, 16, Lemma 4] For an odd function $C(z) \in \mathbb{R}(z)$ the following 5 conditions are equivalent:
(1) if $\operatorname{Re} z>0$ then $\operatorname{Re} C(z)>0$;
(2) if $C(z)=1$ then $\operatorname{Re} z>0$;
(3) if $C(z)=0$ then $\operatorname{Re} z=0$ and $C^{\prime}(z)>0$ (here $z \in \mathbb{C}$ or $z=\infty$ );
(4) either $C(z)$ or $1 / C(z)$ equals
$$
d_{1} z \prod_{k=1}^{n} \frac{z^{2}+a_{k}^{2}}{z^{2}+b_{k}^{2}}
$$
for some integer number $n \geq 0$ and real numbers $d_{1}>0, a_{1}>b_{1}>a_{2}>\cdots>b_{n} \geq 0$;
(5) either $C(z)$ or $1 / C(z)$ equals
$$
d_{1} z+\frac{1}{d_{2} z+\cdots+\frac{1}{d_{m} z}}
$$
for some integer number $m \geq 1$ and real numbers $d_{1}, \ldots, d_{m}>0$.
Parts of the lemma are proved in [5, 14] and in [16] using the results of [30]. A short proof is given in $\S 5.3$.
## 3. Proof of main results
### 3.1. Proof of Theorem 1.3
Hereafter in an electrical circuit or network we allow the conductances to be arbitrary complex numbers with positive real part. This generalization of the above notion is motivated by $\S 2.4$ (and describes both direct- and alternating-current circuits). Theorem 1.3 is an easy consequence of the results of $\S 2$ :
Proof of Theorem 1.3. Suppose that a rectangle of ratio $c$ can be tiled by rectangles of ratios $c_{1}, \ldots, c_{n}$. By Lemma 2.2 there is an electrical network of conductance $c$ consisting of edges of conductances $c_{1}, \ldots, c_{n}$. For each $k=1, \ldots, n$ replace each edge of conductance $c_{k}$ in the network by an edge of complex conductance $z_{k}$, Re $z_{k}>0$. Let $C\left(z_{1}, \ldots, z_{n}\right)$ be the conductance of the obtained network. The function $C\left(z_{1}, \ldots, z_{n}\right)$ has the properties (1)-(3) of Theorem 1.3 by Lemma 2.3(1),(2) and (5).
### 3.2. Proof of Theorem 1.5
Proof of Theorem 1.5. (3) $\Longrightarrow$ (1) [16] Suppose that condition (3) of Theorem 1.5 holds and, say, $m$ is odd. Take a unit square. Cut off a rectangle of ratio $d_{1} c$ from the square by a vertical cut. The remaining part is a rectangle of ratio
$$
1-d_{1} c=\frac{1}{d_{2} c+\cdots+\frac{1}{d_{m} c}}
$$
Now cut off a rectangle of ratio $1 / d_{2} c$ from the remaining part by a horizontal cut. We get a rectangle of ratio
$$
d_{3} c+\frac{1}{d_{4} c+\cdots+\frac{1}{d_{m} c}}
$$
Continue this process alternating vertical and horizontal cuts. Condition (3) guaranties that after step $(m-1)$ we get a rectangle of ratio $d_{m} c$. We obtain a tiling of the square by rectangles of ratios $d_{1} c, 1 / d_{2} c, d_{3} c, 1 / d_{4} c, \ldots, d_{m} c$. Since all $d_{k} \in \mathbb{Q}$ one can chop the tiling into rectangles of ratios $c$ and $1 / c$.
$(1) \Longrightarrow(2)$. Suppose that a square is tiled by rectangles of ratios $c$ and $1 / c$. By Lemma 2.2 there exists an electrical network of conductance 1 with edge conductances $c$ and $1 / c$. Replace each edge of conductance $c$ (respectively, $1 / c$ ) in this network by an edge of conductance $z \in \mathbb{C}$ (respectively, $1 / z$ ). Let $C(z)$ the conductance of the obtained network. Then $C(z) \in \mathbb{Q}(z)$ by Lemma 2.3(1).
Since $C(c)=1$ it follows that $c$ is algebraic $(C(z)$ is nonconstant because $C(-c)=$ $-C(c)=-1$ by Lemma 2.3(2)). Let $z$ be an algebraic conjugate of $c$. Then still $C(z)=1$.
Let us prove that $\operatorname{Re} z>0$. Indeed, first assume that $\operatorname{Re} z<0$. Then $\operatorname{Re}(-z)>0$ and $\operatorname{Re}(-1 / z)>0$. Thus by Lemma 2.3(5) we have $0<\operatorname{Re} C(-z)=-\operatorname{Re} C(z)=-1$, a contradiction. Now assume that $\operatorname{Re} z=0$. Let $z_{k} \rightarrow z$, where each $\operatorname{Re} z_{k}<0$. Still $0<\operatorname{Re} C\left(-z_{k}\right)=-\operatorname{Re} C\left(z_{k}\right) \rightarrow-1$, a contradiction. Thus $\operatorname{Re} z>0$.
$(2) \Longrightarrow(3)$ [16] Let $p(z)$ be a minimal polynomial of $c$. Put $C(z)=\frac{p(-z)-p(z)}{p(-z)+p(z)}$. Then $C(c)=1, C(z) \in \mathbb{Q}(z), C(z)$ is odd and all the roots of the equation $C(z)=1$ have positive real part. By Lemma $2.5(2) \Longrightarrow(5)$ the function $C(z)$ satisfies condition (5) of Lemma 2.5. Since $C(z) \in \mathbb{Q}(z)$ it follows by Euclidean algorithm that all $d_{k} \in \mathbb{Q}$. Substituting $z=c$ we get the required condition.
### 3.3. Proof of Theorem 1.6
The proof follows the ideas of $\S 3.2$ and $\S 5.3$.
Proof of Theorem 1.6. (3) $\Longrightarrow$ (1) Analogously to the proof of Theorem 1.5(3) $\Longrightarrow$ (1).
$(1) \Longrightarrow(2)$. Suppose that a rectangle of ratio $c$ is tiled by rectangles of ratios $c$ and $1 / c$. Rotating through $\pi / 2$ and stretching the figure we get a square tiled by squares and rectangles of ratio $c^{2}$. By Lemma 2.2 there exists an electrical circuit of conductance 1 with edge conductances 1 and $c^{2}$, in which all the edges are essential. Since there is at least one rectangle of ratio $1 / c$ in the initial tiling, it follows that the network contains at least one edge of conductance $c^{2}$. Replace each edge of conductance $c^{2}$ (respectively, 1) in the network by an edge of conductance $z \in \mathbb{C}$ (respectively, $w \in \mathbb{C}$ ). Let $C(z, w)$ the conductance of the obtained network. Denote by $C(z)=C(z, 1)$.
Let us prove that $c^{2}$ is algebraic. Indeed, by Lemma 2.3(4) we have $C^{\prime}\left(c^{2}\right)>0$ because there is at least one essential edge of conductance $c^{2}$ in the network. Thus $C(z)$ is nonconstant. By Lemma 2.3(1) it follows that $C(z) \in \mathbb{Q}(z)$. Since $C\left(c^{2}\right)=1$ it follows that $c^{2}$ is algebraic.
Let $z$ be an algebraic conjugate of $c^{2}$ distinct from $c^{2}$ itself. Then $C(z, 1)=C\left(c^{2}\right)=1$.
Let us prove that $z$ is a negative real number. First assume $\operatorname{Im} z<0$. Then $\operatorname{Re} i z>0$. By Lemma 2.3(2) it follows that $\operatorname{Re} C(i z, i)=\operatorname{Re}(i C(z, 1))=\operatorname{Re} i=0$. Since $C(i z, i)$ is a rational function it follows that any neighborhood of $i z$ contains a point $z^{\prime}$ such that $\operatorname{Re} C\left(z^{\prime}, i\right)<0$. Taking sufficiently small neighborhood we get $\operatorname{Re} z^{\prime}>0$ because Re $i z>0$. By continuity a neighborhood of $i$ contains a point $w^{\prime}$ such that $\operatorname{Re} w^{\prime}>0$ and still $\operatorname{Re} C\left(z^{\prime}, w^{\prime}\right)<0$. The obtained inequalities contradict to Lemma 2.3(5). Case $\operatorname{Im} z>0$ is violated similarly. Assume now $z>c^{2}$. Then by Lemma 2.3(4) we have $1=C(z)>C\left(c^{2}\right)=1$, a contradiction. Case $0 \leq z\cdots>$ $b_{n}>0$. Take a polynomial $q(z)$ with rational coefficients such that $q(z)=z \prod_{k=1}^{n}\left(z^{2}+a_{k}^{2}\right)$, where $a_{1}>b_{1}>a_{2} \cdots>b_{n}>0$. Consider the odd rational function $C(z)=q(z) /(z q(z)-$ $\left.p\left(z^{2}\right)\right)$. We have $C(c)=1 / c$.
Let us check that the function $C(z)$ satisfies condition (3) of Lemma 2.5. The roots of $C(z)$ are the numbers $0, \pm i a_{1}, \ldots, \pm i a_{n}$. A direct evaluation shows that for each $l=$ $1, \ldots, n$
$$
C^{\prime}\left( \pm i a_{l}\right)=-\frac{q^{\prime}\left( \pm i a_{l}\right)}{p\left(-a_{l}^{2}\right)}=\frac{2 a_{l}^{2}}{\left(c^{2}+a_{l}^{2}\right)\left(a_{l}^{2}-b_{l}^{2}\right)} \prod_{k \neq l} \frac{a_{k}^{2}-a_{l}^{2}}{b_{k}^{2}-a_{l}^{2}}>0
$$
by the assumption $a_{1}>b_{1}>a_{2}>\cdots>b_{n}>0$. Analogously $C^{\prime}(0)=-q^{\prime}(0) / p(0)>0$.
Then by Lemma 2.5(3) $\Longrightarrow$ (5) the function $C(z)$ satisfies condition (5) of Lemma 2.5. Since $C(z) \in \mathbb{Q}(z)$ it follows by Euclidean algorithm that all $d_{k} \in \mathbb{Q}$. Substituting $z=c$ we get the required condition.
### 3.4. Remarks to main results
Let us define inductively a series-parallel electrical network. By definition, a network consisting of a single edge is series-parallel. If $a$ and $b$ are two series-parallel networks then both their series and parallel "unions" (see figure 1) are series-parallel.
Theorem 3.1. [5] If a function $C\left(c_{1}, c_{2}\right)$ satisfies conditions (1)-(3) of Theorem 1.3 then $C\left(c_{1}, c_{2}\right)$ is the conductance of a series-parallel electrical network with edge conductances $c_{1}$ and $c_{2}$.
Proof. By conditions (1)-(3) of Theorem 1.3 we have $C\left(c_{1}, c_{2}\right)=\sqrt{c_{1} c_{2}} C(z, 1 / z)$, where $z=\sqrt{c_{1} / c_{2}}$, and the function $C(z)=C(z, 1 / z)$ satisfies condition (1) of Lemma 2.5. By Lemma $2.5(1) \Longrightarrow(5)$ it satisfies condition (5). Therefore, say, for $m$ even and $C(0)=0$,
$$
C\left(c_{1}, c_{2}\right)=d_{1} c_{1}+\frac{1}{\frac{d_{2}}{c_{2}}+\frac{1}{d_{3} c_{1}+\cdots+\frac{d_{m}}{c_{2}}}}
$$
All the numbers $d_{k} \in \mathbb{Q}$ by the Euclidean algorithm. Now the required series-parallel network is constructed analogously to the proof of Theorem 1.5(3) $\Longrightarrow$ (1).
Example 3.2. A generalization of Theorem 3.1 to the case of 3 variables $c_{1}, c_{2}, c_{3}$ is not true. E.g., consider the network with 4 vertices and edge conductances $c_{13}=c_{1}$, $c_{23}=c_{2}, c_{24}=c_{1}, c_{14}=c_{2}, c_{34}=c_{3}$. By Lemma 2.3(4) and a symmetry argument it follows that $\partial C\left(c_{1}, c_{2}, c_{3}\right) / \partial c_{3}=0$ if $c_{1}=c_{2}$. So $C\left(c_{1}, c_{2}, c_{3}\right)$ cannot be the conductance of a series-parallel network, because all the edges of such networks are essential.
## 4. Variations
### 4.1. Tilings of polygons by rectangles
In this subsection we study the following problem.
Problem 4.1. Which polygons can be tiled by rectangles of given ratios $c_{1}, \ldots, c_{n}$ ?
Case $n=1, c_{1}=1$ of the problem is a description of polygons which can be tiled by squares, a problem posed in [15]. In case of hexagons such a description was obtained by R. Kenyon [21]. We give such description for a wide class of polygons.
Hereafter $P$ is an orthogonal polygon, i.e., a polygon with sides parallel to coordinate axes. Assume that $P$ is simple, i.e., the boundary $\partial P$ has one connected component. Enumerate the sides parallel to the $x$-axis counterclockwise in $\partial P$. Let $I_{u}$ be the signed length of the side $u$, where the sign of $I_{u}$ is "+" ("-") if the $P$ locally lies below (above) the side $u$. Let $U_{u}$ be the $y$-coordinate of the side $u$. Assume that $P$ is generic, i.e., the numbers $U_{1}, \ldots, U_{b}$ are pairwise distinct.
We need the following notion [9]. A sequence of boundary vertices $\left(p_{1}, \ldots, p_{k}, q_{1}, \ldots, q_{k}\right)$ of a planar network is circular, if the sequence $\left(p_{1}, \ldots, p_{k}, q_{k}, \ldots, q_{1}\right)$ is in counterclockwise order in the boundary of the unit disc. Denote by $\Omega_{b}$ the set of real $b \times b$ matrices $C_{u v}$ satisfying the following properties:
- $C_{u v}$ is symmetric;
- the sum of the entries of $C_{u v}$ in each row is zero;
- if $\left(p_{1}, \ldots, p_{k}, q_{1}, \ldots, q_{k}\right)$ is a circular sequence then $(-1)^{k} \operatorname{det}\left\{C_{p_{i} q_{j}}\right\}_{i, j=1}^{k} \geq 0$.
Theorem 4.2. Let $P$ be a generic orthogonal polygon with $b$ horizontal sides having signed lengths $I_{1}, \ldots, I_{b}$ and $y$-coordinates $U_{1}, \ldots, U_{b}$. Then the following two conditions are equivalent:
(1) the polygon $P$ can be tiled by squares;
(2) there is a matrix $C_{u v} \in \Omega_{b}$ with rational entries such that $I_{v}=\sum_{u=1}^{b} C_{u v} U_{u}$ for each $v=1, \ldots, b$.
Cases $b=2$ and $b=3$ of this theorem are equivalent to Theorem 1.1 and [21, Theorem 9], respectively. Theorem 4.2 is algorithmic in the particular case when $U_{1}, \ldots, U_{b}$ are linearly independent over $\mathbb{Q}$. Proof of the theorem is constructive, i.e., gives an algorithm to construct the required tiling if the latter exists. Theorem 4.2 does not necessarily hold for nongeneric polygons, e.g., for an orthogonal polygon with
$$
U_{1}=U_{3}=0, \quad U_{2}=2, \quad U_{4}=-4, \quad I_{1}=\sqrt{2}, \quad I_{2}=2, \quad I_{3}=2-\sqrt{2}, \quad I_{4}=-4
$$
We also give a short proof of the following result:
Theorem 4.3. [28] A generic orthogonal polygon with rational vertices can be tiled by rectangles of ratios $c$ and $1 / c$ if and only if a square can be tiled by rectangles of ratios $c$ and $1 / c$.
### 4.2. Electrical impedance tomography
Our approach to Problem 4.1 follows the idea of $[21,7]$ and uses electrical networks with several terminals.
Hereafter we allow electrical circuits to have several boundary vertices $1, \ldots, b$ with prescribed voltages $U_{1}, \ldots, U_{b}$. If an electrical circuit is planar, we assume that the boundary vertices are enumerated counterclockwise along the boundary of the unit disc. We do not assume that an electrical circuit is connected but require that each connected component contains a boundary vertex. The voltages and currents in such circuits are defined by the Ohm and the Kirchhoff current laws (C) and (I) from $\S 2$.
Consider the linear map $\mathbb{C}^{b} \rightarrow \mathbb{C}^{b}$ which takes the vector of voltages $\left(U_{1}, \ldots, U_{b}\right)$ to the vector of incoming currents $\left(I_{1}, \ldots, I_{b}\right)=\left(\sum_{k=1}^{n} I_{1 k}, \ldots, \sum_{k=1}^{n} I_{b k}\right)$ flowing inside the network through the vertices $1, \ldots, b$, respectively. The matrix $C_{w v}$ of this linear map is called the response of the network. This matrix is symmetric [9].
We reduce the results of $\S 4.1$ to the following problems even more interesting in themselves:
- Direct problem. Describe possible responses of electrical networks.
- Inverse problem. Describe possible networks having a given response.
These problems are solved for planar direct-current networks [9, 6, 7, 23]. Let us state certain deep results of Y. Colin de Verdière, E.B. Curtis and J.A. Morrow.
Theorem 4.4. [9, 8, 7, Theorem 5] The set of all possible responses of planar electrical networks with $b$ boundary vertices and positive edge conductances is the set $\Omega_{b}$.
An electrical network is minimal (or critical) if it has minimal number of edges among all planar electrical networks with positive edge conductances and with the same response. The minimality of a network depends only on its graph [7]. In $[9,8, \S 9]$ an algorithm for finding edge conductances in a minimal network with given response is presented. This algorithm implies the following result.
Theorem 4.5. $[9, \S 6.4]$ Conductances of the edges in a minimal electrical network are uniquely determined by the response of the network. Each edge conductance is a rational function with rational coefficients in the entries of the response.
For alternating-current circuits the direct problem is probably open. Let us state some basic results. The rest of $\S 4$ is not used in the proof of the above results.
Theorem 4.6. For $b=2$ or $b=3$ the following 2 conditions are equivalent:
(1) $C_{u v}$ is the response of a connected electrical network with $b$ boundary vertices and with edge conductances having positive real parts;
(2) $C_{u v}$ is a complex $b \times b$ matrix has the following 4 properties:
- $C_{u v}$ is symmetric;
- the sum of the entries of $C_{u v}$ in each row is zero;
- $\operatorname{Re} C_{u v}$ is non-negatively definite;
- if $\sum_{1 \leq u, v \leq b} \operatorname{Re} C_{u v} U_{u} U_{v}=0$ then $U_{1}=\cdots=U_{b}$.
Problem 4.7. Does this result remain true for arbitrary $b \geq 4$ ?
Unlike direct-current networks nonboundary vertices in alternating-current networks can be detected by the response. For instance, by Theorem 4.6 there are electrical networks with response $\left(\begin{array}{ccc}2 & 1 & -3 \\ 1 & 2 & -3 \\ -3 & -3 & 6\end{array}\right)$; any such network necessarily has nonboundary vertices.
### 4.3. Random walks
A random work on an electrical network (or on a weighted graph) is the Markov chain with the transition matrix $P_{k l}=c_{k l} / \sum_{j=1}^{n} c_{j k}$. Such Markov chain is ergodic and reversible. Denote by $k_{1} l_{1}, \ldots, k_{m} l_{m}$ all the edges of the Markov chain. The following theorem allows to translate the results of $\S 1-\S 2$ to the language of random walks.
Theorem 4.8. [11, page 42] Let $P\left(c_{k_{1} l_{1}}, \ldots, c_{k_{m} l_{m}}\right)$ be the probability that a random walk starting at vertex 1 reaches vertex 2 before returning to 1. Let $C\left(c_{k_{1} l_{1}}, \ldots, c_{k_{m} l_{m}}\right)$ be the conductance of the network (with boundary vertices 1 and 2 ). Then $P\left(c_{k_{1} l_{1}}, \ldots, c_{k_{m} l_{m}}\right)=$ $C\left(c_{k_{1} l_{1}}, \ldots, c_{k_{m} l_{m}}\right) /\left(c_{12}+\cdots+c_{1 n}\right)$.
For instance, a translation of Lemmas 2.3(1) and (5) is:
Corollary 4.9. The probability $P\left(c_{k_{1} l_{1}}, \ldots, c_{k_{m} l_{m}}\right)$ is a rational function in $c_{k_{1} l_{1}}, \ldots, c_{k_{m} l_{m}}$. If $\operatorname{Re} c_{k_{1} l_{1}}, \ldots, \operatorname{Re} c_{k_{m} l_{m}}>0$ then $\operatorname{Re}\left(\left(c_{12}+\cdots+c_{1 n}\right) P\left(c_{k_{1} l_{1}}, \ldots, c_{k_{m} l_{m}}\right)\right)>0$.
The latter result does not necessarily hold for nonreversible Markov chains, e.g., for a Markov chain with vertices 1, 2, 3, 4 and oriented edges 14, 42, 43 .
Nonreversible planar Markov chains have a geometric interpretation as tilings of trapezoids by trapezoids [21]. Here a trapezoid is a 4 -gon with two sides parallel to the $x$-axis. The ratio of the trapezoid is the length of the horizontal middle edge divided by the hight. Natural problems are: generalize the results of the paper to tilings by trapezoids; infinite tilings; signed tilings.
## 5. Generalization of main ideas
### 5.1. Electrical circuits
Our approach is based on a generalization of the results of $\S 2$ to electrical circuits with $b$ terminals. Short proofs of the results of $\S 2$ are obtained in this section as particular case $b=2$. Our proof of Lemma 5.2(3), generalizing Lemma 2.3(3), is probably new. All the proofs are based on the following fundamental energy conservation law.
Claim 5.1. Let $E(U, I)$ be a bilinear function. Consider an electrical network with the vertices $1, \ldots, n$ such that $1, \ldots, b$ are the boundary ones. Suppose that the numbers $U_{k}$, $1 \leq k \leq n$, and $I_{k l}, 1 \leq k, l \leq n$, satisfy laws (C),(I) from §2. Set $I_{u}=\sum_{k=1}^{n} I_{u k}$. Then
$$
\sum_{1 \leq k0$ or $c_{k l}=0$. Thus each $\operatorname{Re} \bar{c}_{k l}\left|U_{k}-U_{l}\right|^{2}=0$. Since all the connected components of the circuit contain boundary vertices it follows that all $U_{k}$ are equal. Hence each $U_{k}=0, I_{k l}=0$ and thus each $I_{k l}^{I}=I_{k l}^{I I}, U_{k}^{I}=U_{k}^{I I}$.
Existence. The number of linear equations in the system (C),(I) equals the number of variables. By the previous paragraph the system has a unique solution for $U_{1}=\cdots=$ $U_{b}=0$. Thus by the finite-dimensional Fredholm alternative it has a solution for any $U_{1}, \ldots, U_{b}$.
The following result generalizes Lemma 2.3.
Lemma 5.2. Suppose that an electrical network has $b$ boundary vertices and $m$ edges of conductances $c_{1}, \ldots, c_{m}$. Then the response of the network $C_{u v}\left(c_{1}, \ldots, c_{m}\right)$ has the following properties:
(1) $C_{u v}\left(c_{1}, \ldots, c_{m}\right) \in \mathbb{Q}\left(c_{1}, \ldots, c_{m}\right)^{b \times b}$;
(2) $C_{u v}\left(c_{1}, \ldots, c_{m}\right)$ is degree 1 homogeneous;
(3) $\frac{\partial}{\partial c_{j}} C_{u v}\left(c_{1}, \ldots, c_{m}\right)=\left(V_{k u}-V_{l u}\right)\left(V_{k v}-V_{l v}\right)$, where $k$ and $l$ are the endpoints of the edge $j$ and $V_{p q}$ is the matrix of the linear map $\left(U_{1}, \ldots, U_{b}\right) \mapsto\left(U_{1}, \ldots, U_{n}\right)$;
(4) if $c_{1}, \ldots, c_{m}>0$ then $\frac{\partial}{\partial c_{j}} C_{u v}\left(c_{1}, \ldots, c_{m}\right)$ is non-negatively definite;
(5) if $\operatorname{Re} c_{1}, \ldots, \operatorname{Re} c_{m}>0$ then $\operatorname{Re} C_{u v}\left(c_{1}, \ldots, c_{m}\right)$ is non-negatively definite.
Proof of Lemma 5.2. (1) By Theorem 2.1 and the Crammer rule the solution $\left\{I_{k l}\left(U_{1}, \ldots, U_{b}\right)\right\}$ of the system of linear equations (C), (I) consists of linear functions in $U_{1}, \ldots, U_{b}$ with coefficients being rational functions in $c_{1}, \ldots, c_{m}$. So the entries of the matrix of the linear map $\left(U_{1}, \ldots, U_{b}\right) \mapsto \sum_{k=1}^{n} I_{u k}\left(U_{1}, \ldots, U_{b}\right)$ are rational functions in $c_{1}, \ldots, c_{m}$.
(2) Consider the system of linear equations obtained from laws (C), (I) by substituting $t c_{1}, \ldots, t c_{m}$ for $c_{1}, \ldots, c_{m}$. It defines the same voltages as the initial one and the currents are scaled by $t$. So $C\left(t c_{1}, \ldots, t c_{m}\right)=t C\left(c_{1}, \ldots, c_{m}\right)$.
(3) Set $E(U, I)=\frac{\partial U}{\partial c_{k} I} I-U \frac{\partial I}{\partial c_{k l}}$. Then $E\left(U_{k}-U_{l}, I_{k l}\right)=\left(U_{k}-U_{l}\right)^{2}$ and $E\left(U_{p}-U_{q}, I_{p q}\right)=$ 0 for $p q \neq k l$. Thus by Claim 5.1 we have
$$
\begin{aligned}
\sum_{1 \leq u, v \leq b} \frac{\partial C_{u v}}{\partial c_{k l}} U_{u} U_{v}=\sum_{1 \leq u \leq b} E\left(U_{u}, I_{u}\right) & =\sum_{1 \leq p0$ or $c_{k l}=0$. Take $U_{1}, \ldots, U_{b} \in \mathbb{R}$. By Claim 5.1 we have
$$
\begin{aligned}
& \sum_{1 \leq u, v \leq b} \operatorname{Re} C_{u v} U_{u} U_{v}=\sum_{1 \leq u \leq b} \operatorname{Re}\left(U_{u} \bar{I}_{u}\right)= \\
&=\sum_{1 \leq k0$, incoming voltages $U_{1}, \ldots, U_{b}$ and incoming currents $I_{1}, \ldots, I_{b}$.
Remark 5.5. Condition (2) itself does not guarantee the existence of a rectangular polygon with horizontal sides of signed lengths $I_{1}, \ldots, I_{b}$ and $y$-coordinates $U_{1}, \ldots, U_{b}$. Lemma $5.4(1) \Longrightarrow(2)$ is not necessarily true for nongeneric polygons.
Proof of Lemma 5.4. (1) $\Longrightarrow(2)$. Take a generic polygon $P$ tiled by rectangles.
Let us construct the graph of the required network, see figure 3. Consider the union of the horizontal sides of all rectangles of the tiling. This union splits into several disjoint segments called horizontal cuts. Paint red (bold) all horizontal cuts except small neighborhoods of their endpoints. Paint blue (dashed) the vertical centerline of each rectangle in the tiling.
Contract all red segments. Then the blue set "becomes" a graph $G$ and the polygon $P$ "becomes" a topological disc $D$ (since the $y$-coordinates of the horizontal sides of $P$ are distinct it follows that each read segment has not more than one common point with $\partial P)$. Denote by $1, \ldots, b$ the vertices of the graph $G$ obtained from the red segments in the horizontal cuts containing the sides of $P$ and by $b+1, \ldots, n$ - the other vertices.

Figure 3: Construction of an electrical network
Clearly, $G \subset D, G \cap \partial D=\{1, \ldots, b\}$ and each connected component of $G$ contains a boundary vertex. Thus $G$ is a graph of a planar network.
Let us define the voltages, currents and conductances in the network. For each vertex $k=1, \ldots, n$ of the graph $G$ set $U_{k}$ to be the $y$-coordinate of the horizontal red segment contracted to the vertex. For each edge $k l$ of the graph $G$, obtained from the vertical centerline of a rectangle in the tiling, set $I_{k l}$ and $c_{k l}$ to be the horizontal side (with an appropriate sign) and the ratio of the rectangle, respectively. The laws (C), (I) are now checked directly. The constructed network is the required.

Figure 4: Construction of a tiling
$(2) \Longrightarrow(1)$. Take an electrical network as in (2). Construct a tiling of $P$ as follows.
Let $e$ be an edge of the network. Denote by $e \uparrow(e \downarrow)$ the endpoint of $e$ with higher (lower) voltage (it is well-defined by the assumption that all the edges are essential). By a face we mean a connected component of the complement to the network in the unit disc $D$. Denote by $e \leftarrow(e \rightarrow)$ the face that borders the edge $e$ from the left-hand (right-hand) side while one moves along the edge $e$ from $e \uparrow$ to $e \downarrow$.
By law (I) it follows that to each face $f$ one can assign a number $I_{f}$ in such a way that $I_{k l \leftarrow}-I_{k l \rightarrow}=I_{k l}$. Without loss of generality assume $\min _{f} I_{f}=\min _{(x, y) \in P} x$, where the minimum in the left-hand side is over all the faces $f$ meeting $\partial D$.
Let $P_{e}$ be the rectangle with the vertices $\left(I_{e \rightarrow}, U_{e \uparrow}\right),\left(I_{e \rightarrow}, U_{e \downarrow}\right),\left(I_{e \leftarrow}, U_{e \uparrow}\right),\left(I_{e \leftarrow}, U_{e \downarrow}\right)$. The rectangles $P_{e}$, where $e$ runs through all the edges of the network, tile the polygon $P$ by the following two claims ( $P_{e}$-s cover $P$ by Claim 5.6 and do not overlap by Claim 5.7).
Claim 5.6. $\bigcup_{e} P_{e}=P$.
Proof. It suffices to prove that $\partial \bigcup_{e} P_{e} \subset \partial P$. Since $\partial P$ is a simple closed curve in the plane and $\bigcup_{e} P_{e}$ is bounded, the claim will follow.
We need the following description of the boundary $\partial P$, see figure 4 . Boundary vertices split $\partial D$ into $b$ arcs. Start from vertex $b$ and move along the circle $\partial D$ counterclockwise. Enumerate the arcs in the order they appear in the motion. Denote by $f(v)$ the face containing the arc $v$. Denote by $H_{v}$ the segment joining the points $\left(I_{f(v)}, U_{v}\right)$ and $\left(I_{f(v+1)}, U_{v}\right)$. Denote by $V_{v}$ the segment joining the points $\left(I_{f(v)}, U_{v-1}\right)$ and $\left(I_{f(v)}, U_{v}\right)$, where we set $U_{0}=U_{b}$. Clearly, $\partial P=\bigcup_{v=1}^{b}\left(H_{v} \cup V_{v}\right)$.
Take a "generic" point $p \in \partial \bigcup_{e} P_{e}$, say, in a horizontal side of the "polygon" $\bigcup_{e} P_{e}$. The point $p$ necessarily belongs to a horizontal side of a rectangle in the tiling, say, to the top side of a rectangle $P_{e}$. Denote by $v=e \uparrow$ the vertex of $e$ of higher voltage.
Draw a horizontal line $H$ through the top side of the rectangle $P_{e}$. We say that a rectangle $P_{d}$ is adjacent if the vertex $v$ is an endpoint of the edge $d$. Adjacent rectangles border upon the line $H$ either from above or from below.
First assume that $v$ is nonboundary. A simple induction shows that each point of $H$ (except a finite set) is bordered by the same number of adjacent rectangles $P_{d}$ from above and from below. Since the rectangle $P_{e}$ borders upon the point $p$ from below and $p$ is "generic" it follows that some adjacent rectangle $P_{d}$ borders upon it from above. Thus $p$ belongs to Int $P_{e} \cup P_{f} \subset \operatorname{Int} \bigcup_{e} P_{e}$, a contradiction.
So $v$ is a boundary vertex. Analogously to the above each point of $H-H_{v}$ (except a finite set) is bordered by the same number of adjacent rectangles $P_{d}$ from above and from below. Hence $p \in H_{v}$ and thus $p \in \partial P$.
Claim 5.7. $\sum_{e} \operatorname{Area}\left(P_{e}\right)=\operatorname{Area}(P)$.
Proof. This follows immediately from Claim 5.1 because $\operatorname{Area}\left(P_{k l}\right)=\left(U_{k}-U_{l}\right) I_{k l}$ and Area $(P)=\sum_{1 \leq u \leq b} U_{u} I_{u}$.
### 5.3. Positive real functions
Let us prove Lemma 2.5. For a generalization to the case $b>2$ see [12].
Proof of Lemma 2.5. (1) $\Longrightarrow$ (2). Indeed, if $\operatorname{Re} z \leq 0$ then $\operatorname{Re} C(z)=-\operatorname{Re} C(-z) \leq 0$ and thus $C(z) \neq 1$.
$(2) \Longrightarrow(1)$. Consider the equation $C(z)=w$. Move $w$ continuously in the half-plane $\operatorname{Re} w>0$. The roots cannot cross the line $\operatorname{Re} z=0$ (because $\operatorname{Re} z=0$ implies $\operatorname{Re} C(z)=0$ for an odd function $C(z) \in \mathbb{R}(z))$. Thus for each $w$ in the half-plane $R e w>0$ all roots of $C(z)=w$ are in the half-plane $\operatorname{Re} z>0$. Since $C(z)$ is odd it follows that the same is true for the half-planes $\operatorname{Re} w<0, \operatorname{Re} z<0$. So (1) holds.
$(1) \Longrightarrow$ (3). Suppose that $C(z)=0$, where $z \in \mathbb{C}$. Then $\operatorname{Re} z=0$ because $\operatorname{Re} z>$ $0 \Longrightarrow \operatorname{Re} C(z)>0$ and $\operatorname{Re} z<0 \Longrightarrow \operatorname{Re} C(z)=-\operatorname{Re} C(-z)<0$. Since condition (1) and its converse hold in a neighborhood of the point $z$ it follows that $C^{\prime}(z)>0$. A simple limiting argument proves the same for $z=\infty$.
$(3) \Longrightarrow(4)$ Assume for simplicity that $C(\infty) \neq 0$. Let $z_{1}, \ldots, z_{m}$ be the roots of $C(z)$. Since $C^{\prime}\left(z_{k}\right)>0$ it follows that the roots are simple. Thus $C(z)$ has not more than $m$ poles. The roots split the projective line Re $z=0$ into $m$ "segments". Since $C^{\prime}\left(z_{k}\right)>0$ it follows that for sufficiently small $\epsilon>0$ we have $C\left(z_{k}-i \epsilon\right)<0$ and $C\left(z_{k}+i \epsilon\right)>0$. By intermediate value theorem it follows that each of the segments contains a pole of $C(z)$. Thus all the $m$ poles of $C(z)$ belong to the line $\operatorname{Re} z=0$ and alternate with the roots. So (4) holds.
$(4) \Longrightarrow(5)$. Denote by $h t C(z)$ the sum of the degrees of the nominator and the denominator of $C(z)$. The proof is by induction over $h t C(z)$. If $h t C(z)=1$ then there is nothing to prove. Assume that, say, $C(z)$ equals the expression from condition (4), where $n \geq 1$ and $b_{n} \neq 0$.
Denote by $r(z)=1 /\left(C(z)-d_{1} z\right)$ and $q(z)=1 / C(z)$. Let us prove that $r(z)$ satisfies condition (3). Indeed, the roots of $r(z)$ are the numbers $\pm i b_{1}, \ldots, \pm i b_{n}$. For each $l=$ $1, \ldots, n$
$$
r^{\prime}\left( \pm i b_{l}\right)=q^{\prime}\left( \pm i b_{l}\right)=\frac{2}{d_{1}\left(a_{l}^{2}-b_{l}^{2}\right)} \prod_{k \neq l} \frac{b_{k}^{2}-b_{l}^{2}}{a_{k}^{2}-b_{l}^{2}}>0
$$
by the condition $a_{1}>b_{1}>a_{2}>\cdots>b_{n} \geq 0$.
Hence by Lemma $2.5(3) \Longrightarrow(4)$ it follows that $r(z)$ satisfies condition (4) as well. On the other hand $h t r(z)0$.
Since $E(z) \in \mathbb{Q}(z)$ and $E(c) \in \overline{\mathbb{Q}}$ it follows that $c$ is algebraic $(E(z)$ is nonconstant because $E(-c)=-E(c)<0$ by Lemma 5.2(2)). Let $z$ be an algebraic conjugate of $c$. Then $E(z)=E(c)>0$.
Let us prove that $\operatorname{Re} z>0$. Indeed, first assume that $\operatorname{Re} z<0$. Then by Lemma 5.2(5) we have $0 \leq \operatorname{Re} E(-z)=-\operatorname{Re} E(z)<0$, a contradiction. A simple limiting argument shows that assumption $\operatorname{Re} z=0$ also leads to a contradiction. Thus $\operatorname{Re} z>0$.
### 6.3. Proof of Theorem 4.6
Proof of Theorem 4.6. (1) $\Longrightarrow$ (2). This follows from Lemma 5.2(5) and Remark 5.3.
$(2) \Longrightarrow(1)$. For $b=2$ there is nothing to prove. Assume that $b=3$. Let $\delta>0$ be a small number, $r_{u v}=-\operatorname{Re} C_{u v}-\delta, m_{u v}=-\operatorname{Im} C_{u v}$. By the assumption of the theorem it follows that $\left(\begin{array}{cc}C_{11} & C_{12} \\ C_{21} & C_{22}\end{array}\right)$ is positively definite. Thus $\left(\begin{array}{cc}r_{31}+r_{12} & -r_{12} \\ -r_{12} & r_{12}+r_{23}\end{array}\right)$ is positively definite for sufficiently small $\delta$. Hence $r_{12}+r_{23}, r_{31}+r_{12}, r_{12} r_{23}+r_{23} r_{31}+r_{31} r_{12}>0$. Analogously $r_{23}+r_{31}>0$. Thus at least two of the numbers $r_{12}, r_{23}, r_{31}$ are positive.
If $r_{12}, r_{23}, r_{31}>0$ then the required network is a triangle 123 with edge conductances $c_{k l}=r_{k l}+i m_{k l}+\delta$.
Now assume that exactly one of the numbers $r_{12}, r_{23}, r_{31}$, say, $r_{31}$ is nonpositive. Take a large number $M$ and denote by $\Delta_{M}=r_{12} r_{23}+r_{23} r_{31}+r_{31} r_{12}+i M\left(r_{23}+r_{12}\right)$. The required network is a complete graph on the vertices $1,2,3,4$ with edge conductances
$$
\begin{array}{llc}
c_{12}=\quad i m_{12}+\delta, & c_{14}= & \Delta_{M} / r_{23} \\
c_{23}=\quad i m_{23}+\delta, & c_{34}= & \Delta_{M} / r_{12} \\
c_{31}=i m_{31}+\delta-i M, & c_{24}= & \Delta_{M} /\left(r_{31}+i M\right)
\end{array}
$$
Clearly, for $M^{2}>\left(r_{12} r_{23}+r_{23} r_{31}+r_{31} r_{12}\right)\left|r_{31}\right| /\left(r_{23}+r_{12}\right)$ we have all $\operatorname{Re} c_{k l}>0$.
Let us show by electrical transformations that the network has response $C_{u v}$. Indeed, replace the "letter $Y$ " formed by the edges 14, 24 and 34 by a "triangle $\Delta$ " formed by 3 new edges of conductances $c_{12}^{\prime}=r_{12}, c_{23}^{\prime}=r_{23}$ and $c_{31}^{\prime}=r_{31}+i M$. This $Y \Delta$-transformation does not change the response [21, page 12]. The obtained network has 3 pairs of multiple edges. Thus it has the same response as a triangle with edge conductances $r_{12}+i m_{12}+\delta$, $r_{23}+i m_{23}+\delta, r_{31}+i m_{31}+\delta$. So the network has the response $C_{u v}$.
## Acknowledgements
The authors are grateful to A. Akopyan for many useful discussions and suggestions over the years. M. Skopenkov was supported in part by INTAS grant 06-1000014-6277, Russian Foundation of Basic Research grants 06-01-72551-NCNIL-a, Moebius Contest Foundation for Young Scientists and Euler Foundation.
## References
[1] I. Benjamini and O. Schramm, Random walks and harmonic functions on infinite planar graphs using square tilings, Ann. Prob. 24:3 (1996), 1219-1238.
[2] V.G. Boltianskii, Hilbert's Third Problem (trans. by R. Silverman), V. H. Winston \& Sons, Washington D.C., 1978.
[3] R.L. Brooks, C.A.B. Smith, A.H. Stone and W.T. Tutte, The Dissection of Rectangles into Squares, Duke Math. J. 7 (1940), 312-340.
[4] J. Cannon, W. Floyd, W. Parry, Squaring rectangles: the finite Riemann mapping theorem, Contemp. Math. 169 (1994), 133-211.
[5] W. Cauer, Die Verwirklichung der Wechselstromwiderst ände vorgeschriebener Frequenzabh ängigkeit, Archiv für Elektrotechnik 17 (1926), 355-388 (in German).
[6] Y. Colin de Verdière, Réseaux électriques planaires I, Comm. Math. Helv. 69:1 (1994), 351-374 (in French).
[7] Y. Colin de Verdière, I. Gitler and D. Vertigan, Réseaux électriques planaires II, Comm. Math. Helv. 71:1 (1996), 144-167 (in French).
[8] E.B. Curtis, D. Ingerman and J.A. Morrow, Circular planar graphs and resistor networks, Lin. Alg. Appl. 283 (1998), 115-150.
[9] E.B. Curtis and J.A. Morrow, Inverse problems for electrical networks, Series on Applied Mathematics 13, World Scientific, Singapore, 2000.
[10] M. Dehn, Über die Zerlegung von Rechtecken in Rechtecke, Math. Ann. 57 (1903), 314-332 (in German).
[11] P.G. Doyle and J.L. Snell, Random walks and electric networks, Mathematical Association of America, 1984, http://arxiv.org/abs/math.PR/0001057.
[12] R.J. Duffin, Elementary operations which generate network matrices, Proc. AMS 6:3 (1955), 335-339.
[13] A.J.W. Duijvestijn, Simple perfect squared square of lowest order, J. Comb. Theory B 25 (1978), 240-243.
[14] R.M. Foster, A reactance theorem, Bell System Techn. J. 3 (1924), 259-267.
[15] C. Freiling, R. Hunter, C. Turner and R. Wheeler, Tiling with Squares and AntiSquares, Amer. Math. Monthly 107:3 (2000), 195-204.
[16] C. Freiling, M. Laczkovich, D. Rinne, Rectangling a rectangle, Discr. Comp. Geometry 17 (1997), 217-225.
[17] C. Freiling, D. Rinne, Tiling a square with similar rectangles, Math. Res. Lett. 1 (1994), 547-558.
[18] H. Hadwiger, Vorlesungen Uber Inhalt, Oberflähe und Isoperimetrie, Springer-Verlag, 1957 (in German).
[19] K. Keating and J.L. King, Shape tiling, Elect. J. Comb. 4:2 (1997), R12.
[20] K. Keating and J.L. King, Signed tilings with squares, J. Comb. Theory A 85:1 (1999), 83-91.
[21] R. Kenyon, Tilings and discrete Dirichlet problems, Israel J. Math. 105:1 (1998), $61-84$.
[22] M. Laczkovich, G. Szekeres, Tiling of the square with similar rectangles, Discr. Comp. Geometry 13 (1995), 569-572.
[23] G.F. Lawler and J. Sylvester, Determining resistances from boundary measurements in finite networks, SIAM J. Discr. Math., 2 (1989), 211-239.
[24] L. Lovasz, Random walks on graphs: a survey, In: Combinatorics, Paul Erdos is Eighty, D. Milos, V.T. Sos, and T. Szony, Eds., Budapest, Hungary: Janos Bolyai Math. Soc., 1996, 353-398.
[25] V.G. Pokrovskii, Slicings of $n$-dimensional parallelepipeds, Math. Notes 33:2 (1983), $137-140$.
[26] M. Prasolov, M. Skopenkov, Dissections of a metal rectangle, Kvant (2008) (in Russian), submitted.
[27] J.W.S. Rayleigh, On the theory of resonance, In: Collected scientific papers 1 (1899), $33-75$.
[28] Z. Su and R. Ding, Tilings of orthogonal polygons with similar rectangles or triangles, J. Appl. Math. Comp. 17:1 (2005), 343-350.
[29] B. Szegedy, Tilings of the square with similar right triangles, Combinatorica 21:1 (2001), 139-144.
[30] H.S. Wall, Analytic theory of continued fractions, Chelsea Pub. Co., Bronx, N.Y., 1967, 433 p.
[31] H. Weyl, Repartición de corriente en uno red conductora, Rev. Mat. Hisp. Amer. 5 (1923), 153-164 (in Spanish).
[32] I.M. Yaglom, How to dissect a square? Mathematicheskaya bibliotechka, Nauka, Moscow, 1968, 112 p. (in Russian), http://ilib.mirror1.mccme.ru/djvu/yaglom/ square.htm.
# Random walks and electric networks
Peter G. Doyle J. Laurie Snell
Version 3.02, 5 January 2000
Copyright (C) 1999, 2000 Peter G. Doyle and J. Laurie Snell
Derived from work(s)
Copyright (C) 1984 The Mathematical Association of America
This work is freely redistributable under the terms of
the GNU General Public License
as published by the Free Software Foundation.
This work comes with ABSOLUTELY NO WARRANTY.
## Preface
Probability theory, like much of mathematics, is indebted to physics as a source of problems and intuition for solving these problems. Unfortunately, the level of abstraction of current mathematics often makes it difficult for anyone but an expert to appreciate this fact. In this work we will look at the interplay of physics and mathematics in terms of an example where the mathematics involved is at the college level. The example is the relation between elementary electric network theory and random walks.
Central to the work will be Polya's beautiful theorem that a random walker on an infinite street network in $d$-dimensional space is bound to return to the starting point when $d=2$, but has a positive probability of escaping to infinity without returning to the starting point when $d \geq 3$. Our goal will be to interpret this theorem as a statement about electric networks, and then to prove the theorem using techniques from classical electrical theory. The techniques referred to go back to Lord Rayleigh, who introduced them in connection with an investigation of musical instruments. The analog of Polya's theorem in this connection is that wind instruments are possible in our three-dimensional world, but are not possible in Flatland (Abbott [四).
The connection between random walks and electric networks has been recognized for some time (see Kakutani [12], Kemeny, Snell, and
Knapp [14], and Kelly [13]). As for Rayleigh's method, the authors first learned it from Peter's father Bill Doyle, who used it to explain a mysterious comment in Feller ([5], p. 425, Problem 14). This comment suggested that a random walk in two dimensions remains recurrent when some of the streets are blocked, and while this is ticklish to prove probabilistically, it is an easy consequence of Rayleigh's method. The first person to apply Rayleigh's method to random walks seems to have been Nash-Williams [24]. Earlier, Royden [30] had applied Rayleigh's method to an equivalent problem. However, the true importance of Rayleigh's method for probability theory is only now becoming appreciated. See, for example, Griffeath and Liggett [9], Lyons [20], and Kesten [16].
Here's the plan of the work: In Section 1] we will restrict ourselves to the study of random walks on finite networks. Here we will establish the connection between the electrical concepts of current and voltage and corresponding descriptive quantities of random walks regarded as finite state Markov chains. In Section a we will consider random walks on infinite networks. Polya's theorem will be proved using Rayleigh's method, and the proof will be compared with the classical proof using probabilistic methods. We will then discuss walks on more general infinite graphs, and use Rayleigh's method to derive certain extensions of Polya's theorem. Certain of the results in Section 2 were obtained by Peter Doyle in work on his Ph.D. thesis.
To read this work, you should have a knowledge of the basic concepts of probability theory as well as a little electric network theory and linear algebra. An elementary introduction to finite Markov chains as presented by Kemeny, Snell, and Thompson [15] would be helpful.
The work of Snell was carried out while enjoying the hospitality of Churchill College and the Cambridge Statistical Laboratory supported by an NSF Faculty Development Fellowship. He thanks Professors Kendall and Whittle for making this such an enjoyable and rewarding visit. Peter Doyle thanks his father for teaching him how to think like a physicist. We both thank Peter Ney for assigning the problem in Feller that started all this, David Griffeath for suggesting the example to be used in our first proof that 3-dimensional random walk is recurrent (Section 2.2.9), and Reese Prosser for keeping us going by his friendly and helpful hectoring. Special thanks are due Marie Slack, our secretary extraordinaire, for typing the original and the excessive number of revisions one is led to by computer formatting.
## 1 Random walks on finite networks
### 1.1 Random walks in one dimension
### 1.1.1 A random walk along Madison Avenue
A random walk, or drunkard's walk, was one of the first chance processes studied in probability; this chance process continues to play an important role in probability theory and its applications. An example of a random walk may be described as follows:
A man walks along a 5-block stretch of Madison Avenue. He starts at corner $x$ and, with probability $1 / 2$, walks one block to the right and, with probability $1 / 2$, walks one block to the left; when he comes to the next corner he again randomly chooses his direction along Madison Avenue. He continues until he reaches corner 5, which is home, or corner 0 , which is a bar. If he reaches either home or the bar, he stays there. (See Figure 1.)

Figure 1:
The problem we pose is to find the probability $p(x)$ that the man, starting at corner $x$, will reach home before reaching the bar. In looking at this problem, we will not be so much concerned with the particular form of the solution, which turns out to be $p(x)=x / 5$, as with its general properties, which we will eventually describe by saying " $p(x)$ is the unique solution to a certain Dirichlet problem."
### 1.1.2 The same problem as a penny matching game
In another form, the problem is posed in terms of the following game: Peter and Paul match pennies; they have a total of 5 pennies; on each match, Peter wins one penny from Paul with probability $1 / 2$ and loses one with probability $1 / 2$; they play until Peter's fortune reaches 0 (he
is ruined) or reaches 5 (he wins all Paul's money). Now $p(x)$ is the probability that Peter wins if he starts with $x$ pennies.
### 1.1.3 The probability of winning: basic properties
Consider a random walk on the integers $0,1,2, \ldots, N$. Let $p(x)$ be the probability, starting at $x$, of reaching $N$ before 0 . We regard $p(x)$ as a function defined on the points $x=0,1,2, \ldots, N$. The function $p(x)$ has the following properties:
(a) $p(0)=0$.
(b) $p(N)=1$.
(c) $p(x)=\frac{1}{2} p(x-1)+\frac{1}{2} p(x+1)$ for $x=1,2, \ldots, N-1$.
Properties (a) and (b) follow from our convention that 0 and $\mathrm{N}$ are traps; if the walker reaches one of these positions, he stops there; in the game interpretation, the game ends when one player has all of the pennies. Property (c) states that, for an interior point, the probability $p(x)$ of reaching home from $x$ is the average of the probabilities $p(x-1)$ and $p(x+1)$ of reaching home from the points that the walker may go to from $x$. We can derive (c) from the following basic fact about probability:
Basic Fact. Let $E$ be any event, and $F$ and $G$ be events such that one and only one of the events $F$ or $G$ will occur. Then
$$
\mathbf{P}(E)=\mathbf{P}(F) \cdot \mathbf{P}(E \text { given } F)+\mathbf{P}(G) \cdot \mathbf{P}(E \text { given } G)
$$
In this case, let $E$ be the event "the walker ends at the bar", $F$ the event "the first step is to the left", and $G$ the event "the first step is to the right". Then, if the walker starts at $x, \mathbf{P}(E)=p(x)$, $\mathbf{P}(F)=\mathbf{P}(G)=\frac{1}{2}, \mathbf{P}(E$ given $F)=p(x-1), \mathbf{P}(E$ given $G)=p(x+1)$, and (c) follows.
### 1.1.4 An electric network problem: the same problem?
Let's consider a second apparently very different problem. We connect equal resistors in series and put a unit voltage across the ends as in Figure 2 .
Voltages $v(x)$ will be established at the points $x=0,1,2,3,4,5$. We have grounded the point $x=0$ so that $v(0)=0$. We ask for the voltage $v(x)$ at the points $x$ between the resistors. If we have $N$ resistors, we

Figure 2:
make $v(0)=0$ and $v(N)=1$, so $v(x)$ satisfies properties (a) and (b) of Section 1.1.3. We now show that $v(x)$ also satisfies (c).
By Kirchhoff's Laws, the current flowing into $x$ must be equal to the current flowing out. By Ohm's Law, if points $x$ and $y$ are connected by a resistance of magnitude $R$, then the current $i_{x y}$ that flows from $x$ to $y$ is equal to
$$
i_{x y}=\frac{v(x)-v(y)}{R}
$$
Thus for $x=1,2, \ldots, N-1$,
$$
\frac{v(x-1)-v(x)}{R}+\frac{v(x+1)-v(x)}{R}=0
$$
Multiplying through by $R$ and solving for $v(x)$ gives
$$
v(x)=\frac{v(x+1)+v(x-1)}{2}
$$
for $x=1,2, \ldots, N-1$. Therefore, $v(x)$ also satisfies property (c).
We have seen that $p(x)$ and $v(x)$ both satisfy properties (a), (b), and (c) of Section 1.1.3. This raises the question: are $p(x)$ and $v(x)$ equal? For this simple example, we can easily find $v(x)$ using Ohm's Law, find $p(x)$ using elementary probability, and see that they are the same. However, we want to illustrate a principle that will work for very general circuits. So instead we shall prove that these two functions are the same by showing that there is only one function that satisfies these properties, and we shall prove this by a method that will apply to more general situations than points connected together in a straight line.
Exercise 1.1.1 Referring to the random walk along Madison Avenue, let $X=p(1), Y=p(2), Z=p(3)$, and $W=p(4)$. Show that properties (a), (b), and (c) of Section 1.1.3 determine a set of four linear equations with variables $X, Y, Z$ and $W$. Show that these equations have a unique solution. What does this say about $p(x)$ and $v(x)$ for this special case?
Exercise 1.1.2 Assume that our walker has a tendency to drift in one direction: more specifically, assume that each step is to the right with probability $p$ or to the left with probability $q=1-p$. Show that properties (a), (b), and (c) of Section 1.1.3 should be replaced by
(a) $p(0)=0$.
(b) $p(N)=1$.
(c) $p(x)=q \cdot p(x-1)+p \cdot p(x+1)$.
Exercise 1.1.3 In our electric network problem, assume that the resistors are not necessarily equal. Let $R_{x}$ be the resistance between $x$ and $x+1$. Show that
$$
v(x)=\frac{\frac{1}{R_{x-1}}}{\frac{1}{R_{x-1}}+\frac{1}{R_{x}}} v(x-1)+\frac{\frac{1}{R_{x}}}{\frac{1}{R_{x-1}}+\frac{1}{R_{x}}} v(x+1)
$$
How should the resistors be chosen to correspond to the random walk of Exercise 1.1.2?
### 1.1.5 Harmonic functions in one dimension; the Uniqueness Principle
Let $S$ be the set of points $S=\{0,1,2, \ldots, N\}$. We call the points of the set $D=\{1,2, \ldots, N-1\}$ the interior points of $S$ and those of $B=\{0, N\}$ the boundary points of $S$. A function $f(x)$ defined on $S$ is harmonic if, at points of $D$, it satisfies the averaging property
$$
f(x)=\frac{f(x-1)+f(x+1)}{2}
$$
As we have seen, $p(x)$ and $v(x)$ are harmonic functions on $S$ having the same values on the boundary: $p(0)=v(0)=0 ; p(N)=v(N)=$ 1. Thus both $p(x)$ and $v(x)$ solve the problem of finding a harmonic
function having these boundary values. Now the problem of finding a harmonic function given its boundary values is called the Dirichlet problem, and the Uniqueness Principle for the Dirichlet problem asserts that there cannot be two different harmonic functions having the same boundary values. In particular, it follows that $p(x)$ and $v(x)$ are really the same function, and this is what we have been hoping to show. Thus the fact that $p(x)=v(x)$ is an aspect of a general fact about harmonic functions.
We will approach the Uniqueness Principle by way of the Maximum Principle for harmonic functions, which bears the same relation to the Uniqueness Principle as Rolle's Theorem does to the Mean Value Theorem of Calculus.
Maximum Principle . A harmonic function $f(x)$ defined on $S$ takes on its maximum value $M$ and its minimum value $m$ on the boundary.
Proof. Let $M$ be the largest value of $f$. Then if $f(x)=M$ for $x$ in $D$, the same must be true for $f(x-1)$ and $f(x+1)$ since $f(x)$ is the average of these two values. If $x-1$ is still an interior point, the same argument implies that $f(x-2)=M$; continuing in this way, we eventually conclude that $f(0)=M$. That same argument works for the minimum value $m$. $\diamond$
Uniqueness Principle. If $f(x)$ and $g(x)$ are harmonic functions on $S$ such that $f(x)=g(x)$ on $B$, then $f(x)=g(x)$ for all $x$.
Proof. Let $h(x)=f(x)-g(x)$. Then if $x$ is any interior point,
$$
\frac{h(x-1)+h(x+1)}{2}=\frac{f(x-1)+f(x+1)}{2}-\frac{g(x-1)+g(x+1)}{2}
$$
and $h$ is harmonic. But $h(x)=0$ for $x$ in $B$, and hence, by the Maximum Principle, the maximum and mininium values of $h$ are 0 . Thus $h(x)=0$ for all $x$, and $f(x)=g(x)$ for all $x$. $\diamond$
Thus we finally prove that $p(x)=v(x)$; but what does $v(x)$ equal? The Uniqueness Principle shows us a way to find a concrete answer: just guess. For if we can find any harmonic function $f(x)$ having the right boundary values, the Uniqueness Principle guarantees that
$$
p(x)=v(x)=f(x)
$$
The simplest function to try for $f(x)$ would be a linear function; this leads to the solution $f(x)=x / N$. Note that $f(0)=0$ and $f(N)=1$ and
$$
\frac{f(x-1)+f(x+1)}{2}=\frac{x-1+x+1}{2 N}=\frac{x}{N}=f(x)
$$
Therefore $f(x)=p(x)=v(x)=x / N$.
As another application of the Uniqueness Principle, we prove that our walker will eventually reach 0 or $N$. Choose a starting point $x$ with $00$
$$
\mathbf{P}\left(-k<\frac{S_{n}-n p}{\sqrt{n p q}}
1 \& 0 \& 0 \& 0 \& 0
0 \& 1 \& 0 \& 0 \& 0
1 / 2 \& 0 \& 0 \& 1 / 2 \& 0
0 \& 0 \& 1 / 2 \& 0 \& 1 / 2
0 \& 1 / 2 \& 0 \& 1 / 2 \& 0\end{array}\right)\).
The matrix $\mathbf{N}=(\mathbf{I}-\mathbf{Q})^{-1}$ is called the fundamental matrix for the absorbing chain $\mathbf{P}$. The entries $N_{i j}$ of this matrix have the following probabilistic interpretation: $N_{i j}$ is the expected number of times that
the chain will be in state $s_{j}$ before absorption when it is started in $s_{i}$. (To see why this is true, think of how $(\mathbf{I}-\mathbf{Q})^{-1}$ would look if it were written as a geometric series.) Let $\mathbf{1}$ be a column vector of all 1's. Then the vector $\mathbf{t}=\mathbf{N I}$ gives the expected number of steps before absorption for each starting state.
The absorption probabilities $\mathbf{B}$ are obtained from $\mathbf{N}$ by the matrix formula
$$
\mathbf{B}=(\mathbf{I}-\mathbf{Q})^{-1} \mathbf{R}
$$
This simply says that to get the probability of ending up at a given absorbing state, we add up the probabilities of going there from all the non-absorbing states, weighted by the number of times we expect to be in those (non-absorbing) states.
For our random walk example
$$
\begin{gathered}
\mathbf{Q}=\left(\begin{array}{ccc}
0 & \frac{1}{2} & 0 \\
\frac{1}{2} & 0 & \frac{1}{2} \\
0 & \frac{1}{2} & 0
\end{array}\right) \\
\mathbf{I}-\mathbf{Q}=\left(\begin{array}{ccc}
1 & -\frac{1}{2} & 0 \\
-\frac{1}{2} & 1 & -\frac{1}{2} \\
0 & -\frac{1}{2} & 1
\end{array}\right) \\
\left.\begin{array}{ccc}
1 & 2 & 3 \\
1 \\
\mathbf{3} & 1 & \frac{1}{2} \\
1 & 2 & 1 \\
\frac{1}{2} & 1 & \frac{3}{2}
\end{array}\right) \\
\mathbf{t}=\mathbf{N} \mathbf{1}=\left(\begin{array}{ccc}
\frac{3}{2} & 1 & \frac{1}{2} \\
1 & 2 & 1 \\
\frac{1}{2} & 1 & \frac{3}{2}
\end{array}\right)\left(\begin{array}{c}
1 \\
1 \\
1
\end{array}\right)=\left(\begin{array}{c}
3 \\
4 \\
3
\end{array}\right) \\
\mathbf{B}=\mathbf{N R}=\left(\begin{array}{ccc}
\frac{3}{2} & 1 & \frac{1}{2} \\
1 & 2 & 1 \\
\frac{1}{2} & 1 & \frac{3}{2}
\end{array}\right)\left(\begin{array}{cc}
\frac{1}{2} & 0 \\
0 & 0 \\
0 & \frac{1}{2}
\end{array}\right)=\begin{array}{c}
4 \\
1 \\
2 \\
3
\end{array}\left(\begin{array}{cc}
\frac{3}{4} & \frac{1}{4} \\
\frac{1}{2} & \frac{1}{2} \\
\frac{1}{4} & \frac{3}{4}
\end{array}\right)
\end{gathered}
$$
Thus, starting in state 3 , the probability is $3 / 4$ of reaching 4 before 0 ; this is in agreement with our previous results. From $\mathbf{t}$ we see that the expected duration of the game, when we start in state 2 , is 4 .
For an absorbing chain $\mathbf{P}$, the $n$th power $\mathbf{P}^{n}$ of the transition probabilities will approach a matrix $\mathbf{P}^{\infty}$ of the form
$$
\mathbf{P}^{\infty}=\left(\begin{array}{ll}
\mathbf{I} & \mathbf{0} \\
\mathbf{B} & \mathbf{Q}
\end{array}\right)
$$
We now give our Markov chain version of the Dirichlet problem. We interpret the absorbing states as boundary states and the non-absorbing states as interior states. Let $B$ be the set of boundary states and $D$ the set of interior states. Let $f$ be a function with domain the state space of a Markov chain $\mathbf{P}$ such that for $i$ in $D$
$$
f(i)=\sum_{j} P_{i j} f(j)
$$
Then $f$ is a harmonic function for $\mathbf{P}$. Now $f$ again has an averaging property and extends our previous definition. If we represent $f$ as a column vector $\mathbf{f}, f$ is harmonic if and only if
$$
\mathbf{P f}=\mathbf{f}
$$
This implies that
$$
\mathbf{P}^{2} \mathbf{f}=\mathbf{P} \cdot \mathbf{P f}=\mathbf{P f}=\mathbf{f}
$$
and in general
$$
\mathbf{P}^{n} \mathbf{f}=\mathbf{f}
$$
Let us write the vector $\mathbf{f}$ as
$$
\mathbf{f}=\binom{\mathbf{f}_{B}}{\mathbf{f}_{D}}
$$
where $\mathbf{f}_{B}$ represents the values of $f$ on the boundary and $\mathbf{f}_{D}$ values on the interior. Then we have
$$
\binom{\mathbf{f}_{B}}{\mathbf{f}_{D}}=\left(\begin{array}{cc}
\mathbf{I} & \mathbf{0} \\
\mathbf{B} & \mathbf{Q}
\end{array}\right)\binom{\mathbf{f}_{B}}{\mathbf{f}_{D}}
$$
and
$$
\mathbf{f}_{D}=\mathbf{B} \mathbf{f}_{B}
$$
We again see that the values of a harmonic function are determined by the values of the function at the boundary points.
Since the entries $B_{i j}$ of $\mathbf{B}$ represent the probability, starting in $i$, that the process ends at $j$, our last equation states that if you play a game in which your fortune is $f_{j}$ when you are in state $j$, then your expected final fortune is equal to your initial fortune; that is, fairness is preserved. As remarked above, from Markov chain theory $\mathbf{B}=\mathbf{N R}$ where $\mathbf{N}=(\mathbf{I}-\mathbf{Q})^{-1}$. Thus
$$
\mathbf{f}_{D}=(\mathbf{I}-\mathbf{Q})^{-1} \mathbf{R} \mathbf{f}_{B}
$$
(To make the correspondence between this solution and the solution of Section 1.2.5, put $\mathbf{A}=\mathbf{I}-\mathbf{Q}$ and $\mathbf{u}=\mathbf{R f}_{B}$.)
A general discussion of absorbing Markov chains may be found in Kemeny, Snell, and Thompson [15].
Exercise 1.2.9 Consider the game played on the grid in Figure 11. You start at an interior point and move randomly until a boundary

Figure 11:
point is reached and obtain the payment indicated at this point. Using Markov chain methods find, for each starting state, the expected value of the game. Find also the expected duration of the game.
### 1.3 Random walks on more general networks
### 1.3.1 General resistor networks and reversible Markov chains
Our networks so far have been very special networks with unit resistors. We will now introduce general resistor networks, and consider what it means to carry out a random walk on such a network.
A graph is a finite collection of points (also called vertices or nodes) with certain pairs of points connected by edges (also called branches). The graph is connected if it is possible to go between any two points by moving along the edges. (See Figure 12.)
We assume that $G$ is a connected graph and assign to each edge $x y$ a resistance $R_{x y}$; an example is shown in Figure 13. The conductance of an edge $x y$ is $C_{x y}=1 / R_{x y}$; conductances for our example are shown in Figure 14.

Connected graph

Disconnected graph
Figure 12:

Resistances
Figure 13:

Figure 14:
We define a random walk on $G$ to be a Markov chain with transition matrix $\mathbf{P}$ given by
$$
P_{x y}=\frac{C_{x y}}{C_{x}}
$$
with $C_{x}=\sum_{y} C_{x y}$. For our example, $C_{a}=2, C_{b}=3, C_{c}=4$, and $C_{d}=5$, and the transition matrix $\mathbf{P}$ for the associated random walk is

Its graphical representation is shown in Figure 15.
Since the graph is connected, it is possible for the walker to go between any two states. A Markov chain with this property is called an ergodic Markov chain. Regular chains, which were introduced in Section 1.2.6, are always ergodic, but ergodic chains are not always regular (see Exercise 1.3.1).
For an ergodic chain, there is a unique probability vector $\mathbf{w}$ that is a fixed vector for $\mathbf{P}$, i.e., $\mathbf{w P}=\mathbf{w}$. The component $w_{j}$ of $\mathbf{w}$ represents the proportion of times, in the long run, that the walker will be in state $j$. For random walks determined by electric networks, the fixed vector is given by $w_{j}=C_{j} / C$, where $C=\sum_{x} C_{x}$. (You are asked to prove this

Figure 15:
in Exercise 1.3.2.) For our example $C_{a}=2, C_{b}=3, C_{c}=4, C_{d}=5$, and $C=14$. Thus $\mathbf{w}=(2 / 14,3 / 14,4 / 14,5 / 14)$. We can check that $\mathbf{w}$ is a fixed vector by noting that
$$
\left(\begin{array}{llll}
\frac{2}{14} & \frac{3}{14} & \frac{4}{14} & \frac{5}{14}
\end{array}\right)\left(\begin{array}{cccc}
0 & 0 & \frac{1}{2} & \frac{1}{2} \\
0 & 0 & \frac{1}{3} & \frac{2}{3} \\
\frac{1}{4} & \frac{1}{4} & 0 & \frac{1}{2} \\
\frac{1}{5} & \frac{2}{5} & \frac{2}{5} & 0
\end{array}\right)=\left(\begin{array}{llll}
\frac{2}{14} & \frac{3}{14} & \frac{4}{14} & \frac{5}{14}
\end{array}\right)
$$
In addition to being ergodic, Markov chains associated with networks have another property called reversibility. An ergodic chain is said to be reversible if $w_{x} P_{x y}=w_{y} P_{y x}$ for all $x, y$. That this is true for our network chains follows from the fact that
$$
C_{x} P_{x y}=C_{x} \frac{C_{x y}}{C_{x}}=C_{x y}=C_{y x}=C_{y} \frac{C_{y x}}{C_{y}}=C_{y} P_{y x}
$$
Thus, dividing the first and last term by $C$, we have $w_{x} P_{x y}=w_{y} P_{y x}$.
To see the meaning of reversibility, we start our Markov chain with initial probabilities $\mathbf{w}$ (in equilibrium) and observe a few states, for example
$$
\begin{array}{llll}
a & c & b & d
\end{array}
$$
The probability that this sequence occurs is
$$
w_{a} P_{a c} P_{c b} P_{b d}=\frac{2}{14} \cdot \frac{1}{2} \cdot \frac{1}{4} \cdot \frac{2}{3}=\frac{1}{84}
$$
The probability that the reversed sequence
$$
d b c a
$$
occurs is
$$
w_{d} P_{d b} P_{b c} P_{c a}=\frac{5}{14} \cdot \frac{2}{5} \cdot \frac{1}{3} \cdot \frac{1}{4}=\frac{1}{84}
$$
Thus the two sequences have the same probability of occurring.
In general, when a reversible Markov chain is started in equilibrium, probabilities for sequences in the correct order of time are the same as those with time reversed. Thus, from data, we would never be able to tell the direction of time.
If $\mathbf{P}$ is any reversible ergodic chain, then $\mathbf{P}$ is the transition matrix for a random walk on an electric network; we have only to define $C_{x y}=w_{x} P_{x y}$. Note, however, if $P_{x x} \neq 0$ the resulting network will need a conductance from $x$ to $x$ (see Exercise (1.3.4). Thus reversibility characterizes those ergodic chains that arise from electrical networks. This has to do with the fact that the physical laws that govern the behavior of steady electric currents are invariant under time-reversal (see Onsager [25]).
When all the conductances of a network are equal, the associated random walk on the graph $G$ of the network has the property that, from each point, there is an equal probability of moving to each of the points connected to this point by an edge. We shall refer to this random walk as simple random walk on $G$. Most of the examples we have considered so far are simple random walks. Our first example of a random walk on Madison Avenue corresponds to simple random walk on the graph with points $0,1,2, \ldots, N$ and edges the streets connecting these points. Our walks on two dimensional graphs were also simple random walks.
Exercise 1.3.1 Give an example of an ergodic Markov chain that is not regular. (Hint: a chain with two states will do.)
Exercise 1.3.2 Show that, if $\mathbf{P}$ is the transition matrix for a random walk determined by an electric network, then the fixed vector $\mathbf{w}$ is given by $w_{x}=\frac{C_{x}}{C}$ where $C_{x}=\sum_{y} C_{x y}$ and $C=\sum_{x} C_{x}$.
Exercise 1.3.3 Show that, if $\mathbf{P}$ is a reversible Markov chain and $a, b, c$ are any three states, then the probability, starting at $a$, of the cycle $a b c a$ is the same as the probability of the reversed cycle $a c b a$. That is $P_{a b} P_{b c} P_{c a}=P_{a c} P_{c b} P_{b a}$. Show, more generally, that the probability of going around any cycle in the two different directions is the same. (Conversely, if this cyclic condition is satisfied, the process is reversible. For a proof, see Kelly [13].)
Exercise 1.3.4 Assume that $\mathbf{P}$ is a reversible Markov chain with $P_{x x}=$ 0 for all $x$. Define an electric network by $C_{x y}=w_{x} P_{x y}$. Show that the Markov chain associated with this circuit is $\mathbf{P}$. Show that we can allow $P_{x x}>0$ by allowing a conductance from $x$ to $x$.
Exercise 1.3.5 For the Ehrenfest urn model, there are two urns that together contain $N$ balls. Each second, one of the $N$ balls is chosen at random and moved to the other urn. We form a Markov chain with states the number of balls in one of the urns. For $N=4$, the resulting transition matrix is
$$
\mathbf{P}=\begin{gathered}
\\
0 \\
1 \\
2 \\
3 \\
4
\end{gathered}\left(\begin{array}{ccccc}
0 & 1 & 2 & 3 & 4 \\
0 & 1 & 0 & 0 & 0 \\
\frac{1}{4} & 0 & \frac{3}{4} & 0 & 0 \\
0 & \frac{1}{2} & 0 & \frac{1}{2} & 0 \\
0 & 0 & \frac{3}{4} & 0 & \frac{1}{4} \\
0 & 0 & 0 & 1 & 0
\end{array}\right)
$$
Show that the fixed vector $\mathbf{w}$ is the binomial distribution $\mathbf{w}=\left(\frac{1}{16}, \frac{4}{16}, \frac{6}{16}, \frac{4}{16}, \frac{1}{16}\right)$. Determine the electric network associated with this chain.
### 1.3.2 Voltages for general networks; probabilistic interpretation
We assume that we have a network of resistors assigned to the edges of a connected graph. We choose two points $a$ and $b$ and put a one-volt battery across these points establishing a voltage $v_{a}=1$ and $v_{b}=0$, as illustrated in Figure 16. We are interested in finding the voltages $v_{x}$ and the currents $i_{x y}$ in the circuit and in giving a probabilistic interpretation to these quantities.
We begin with the probabilistic interpretation of voltage. It will come as no surprise that we will interpret the voltage as a hitting probability, observing that both functions are harmonic and that they have the same boundary values.

1 volt
Figure 16:
By Ohm's Law, the currents through the resistors are determined by the voltages by
$$
i_{x y}=\frac{v_{x}-v_{y}}{R_{x y}}=\left(V_{x}-v_{y}\right) C_{x y}
$$
Note that $i_{x y}=-i_{y x}$. Kirchhoff's Current Law requires that the total current flowing into any point other than $a$ or $b$ is 0 . That is, for $x \neq a, b$
$$
\sum_{y} i_{x y}=0
$$
This will be true if
$$
\sum_{y}\left(v_{x}-v_{y}\right) C_{x y}=0
$$
or
$$
v_{x} \sum_{y} C_{x y}=\sum_{y} C_{x y} v_{y}
$$
Thus Kirchhoff's Current Law requires that our voltages have the property that
$$
v_{x}=\sum_{y} \frac{C_{x y}}{C_{x}} v_{y}=\sum_{y} P_{x y} v_{y}
$$
for $x \neq a, b$. This means that the voltage $v_{x}$ is harmonic at all points $x \neq a, b$.
Let $h_{x}$ be the probability, starting at $x$, that state $a$ is reached before $b$. Then $h_{x}$ is also harmonic at all points $x \neq a, b$. Furthermore
$$
v_{a}=h_{a}=1
$$
and
$$
v_{b}=h_{b}=0
$$
Thus if we modify $\mathbf{P}$ by making $a$ and $b$ absorbing states, we obtain an absorbing Markov chain $\overline{\mathbf{P}}$ and $v$ and $h$ are both solutions to the Dirichlet problem for the Markov chain with the same boundary values. Hence $v=h$.
For our example, the transition probabilities $\bar{P}_{x y}$ are shown in Figure 17. The function $v_{x}$ is harmonic for $\bar{P}$ with boundary values $v_{a}=$

Figure 17:
$1, v_{b}=0$.
To sum up, we have the following:
Intrepretation of Voltage. When a unit voltage is applied between $a$ and $b$, making $v_{a}=1$ and $v_{b}=0$, the voltage $v_{x}$ at any point $x$
represents the probability that a walker starting from $x$ will return to $a$ before reaching $b$.
In this probabilistic interpretation of voltage, we have assumed a unit voltage, but we could have assumed an arbitrary voltage $v_{a}$ between $a$ and $b$. Then the hitting probability $h_{x}$ would be replaced by an expected value in a game where the player starts at $x$ and is paid $v_{a}$ if $a$ is reached before $b$ and 0 otherwise.
Let's use this interpretation of voltage to find the voltages for our example. Referring back to Figure 17, we see that
$$
\begin{gathered}
v_{a}=1 \\
v_{b}=0 \\
v_{c}=\frac{1}{4}+\frac{1}{2} v_{d} \\
v_{d}=\frac{1}{5}+\frac{2}{5} v_{c}
\end{gathered}
$$
Solving these equations yields $v_{c}=\frac{7}{16}$ and $v_{d}=\frac{3}{8}$. From these voltages we obtain the current $i_{x y}$. For example $i_{c d}=\left(\frac{7}{16}-\frac{3}{8}\right) \cdot 2=\frac{1}{8}$. The resulting voltages and currents are shown in Figure 18. The
voltage at $c$ is $\frac{7}{16}$ and so this is also the probability, starting at $c$, of reaching $a$ before $b$.
### 1.3.3 Probabilistic interpretation of current
We turn now to the probabilistic interpretation of current. This interpretation is found by taking a naive view of the process of electrical conduction: We imagine that positively charged particles enter the network at point $a$ and wander around from point to point until they finally arrive at point $b$, where they leave the network. (It would be more realistic to imagine negatively charged particles entering at $b$ and leaving at $a$, but realism is not what we're after.) To determine the current $i_{x y}$ along the branch from $x$ to $y$, we consider that in the course of its peregrinations the point may pass once or several times along the branch from $x$ to $y$, and in the opposite direction from $y$ to $x$. We may now hypothesize that the current $i_{x y}$ is proportional to the expected net number of movements along the edge from $x$ to $y$, where movements from $y$ back to $x$ are counted as negative. This hypothesis is correct, as we will now show.
The walker begins at $a$ and walks until he reaches $b$; note that if he returns to $a$ before reaching $b$, he keeps on going. Let $u_{x}$ be the

Figure 18:
expected number of visits to state $x$ before reaching $b$. Then $u_{b}=0$ and, for $x \neq a, b$,
$$
u_{x}=\sum_{y} u_{y} P_{y x}
$$
This last equation is true because, for $x \neq a, b$, every entrance to $x$ must come from some $y$.
We have seen that $C_{x} P_{x y}=C_{y} P_{y x}$; thus
$$
u_{x}=\sum_{y} u_{y} \frac{P_{x y} C_{x}}{C_{y}}
$$
or
$$
\frac{u_{x}}{C_{x}}=\sum_{y} P_{x y} \frac{u_{y}}{C_{y}}
$$
This means that $v_{x}=u_{x} / C_{x}$ is harmonic for $x \neq a, b$. We have also $v_{b}=0$ and $v_{a}=u_{a} / C_{a}$. This implies that $v_{x}$ is the voltage at $x$ when we put a battery from $a$ to $b$ that establishes a voltage $u_{a} / C_{a}$ at $a$ and voltage 0 at $b$. (We remark that the expression $v_{x}=u_{x} / C_{x}$ may be understood physically by viewing $u_{x}$ as charge and $C_{x}$ as capacitance; see Kelly [13] for more about this.)
We are interested in the current that flows from $x$ to $y$. This is
$i_{x y}=\left(v_{x}-v_{y}\right) C_{x y}=\left(\frac{u_{x}}{C_{x}}-\frac{u_{y}}{C_{y}}\right) C_{x y}=\frac{u_{x} C_{x y}}{C_{x}}-\frac{u_{y} C_{y x}}{C_{y}}=u_{x} P_{x y}-u_{y} P_{y x}$.
Now $u_{x} P_{x y}$ is the expected number of times our walker will go from $x$ to $y$ and $u_{y} P_{y x}$ is the expected number of times he will go from $y$ to $x$. Thus the current $i_{x y}$ is the expected value for the net number of times the walker passes along the edge from $x$ to $y$. Note that for any particular walk this net value will be an integer, but the expected value will not.
As we have already noted, the currents $i_{x y}$ here are not those of our original electrical problem, where we apply a 1 -volt battery, but they are proportional to those original currents. To determine the constant of proportionality, we note the following characteristic property of the new currents $i_{x y}$ : The total current flowing into the network at $a$ (and out at $b$ ) is 1 . In symbols,
$$
\sum_{y} i_{a y}=1
$$
Indeed, from our probabilistic interpretation of $i_{x y}$ this sum represents the expected value of the difference between the number of times our
walker leaves $a$ and enters $a$. This number is necessarily one and so the current flowing into $a$ is 1 .
This unit current flow from $a$ to $b$ can be obtained from the currents in the original circuit, corresponding to a 1-volt battery, by dividing through by the total amount of current $\sum_{y} i_{a y}$ flowing into $a$; doing this to the currents in our example yields the unit current flow shown in Figure 19.

The unit curtent flow
Figure 19:
This shows that the constant of proportionality we were seeking to determine is the reciprocal of the amount of current that flows through the circuit when a I-volt battery is applied between $a$ and $b$. This quantity, called the effective resistance between $a$ and $b$, is discussed in detail in Section 1.3.4.
To sum up, we have the following:
Interpretation of Current. When a unit current flows into $a$ and out of $b$, the current $i_{x y}$ flowing through the branch connecting $x$ to $y$ is equal to the expected net number of times that a walker, starting at $a$ and walking until he reaches $b$, will move along the branch from $x$ to $y$. These currents are proportional to the currents that arise when a unit voltage is applied between $a$ and $b$, the constant of proportionality being the effective resistance of the network.
We have seen that we can estimate the voltages by simulation. We can now do the same for the currents. We have to estimate the expected value for the net number of crossings of $x y$. To do this, we start a large number of walks at $a$ and, for each one, record the net number of crossings of each edge and then average these as an estimate for the expected value. Carrying out 10,000 such walks yielded the results shown in Figure 20.

Estimates of currents by simulation

Calculated curtemts
Figure 20:
The results of simulation are in good agreement with the theoretical values of current. As was the case for estimates of the voltages by simulation, we have statistical errors. Our estimates have the property that the total current flowing into $a$ is 1 , out of $b$ is 1 , and into any other point it is 0 . This is no accident; the explanation is that the history of each walk would have these properties, and these properties are not destroyed by averaging.
Exercise 1.3.6 Kingman [17] introduced a different model for current flow. Kelly [13] gave a new interpretation of this model. Both authors use continuous time. A discrete time version of Kelly's interpretation would be the following: At each point of the graph there is a black or a white button. Each second an edge is chosen; edge $x y$ is chosen with probability $C_{x y} / C$ where $C$ is the sum of the conductances. The
buttons on the edge chosen are then interchanged. When a button reaches $a$ it is painted black, and when a button reaches $b$ it is painted white. Show that there is a limiting probability $p_{x}$ that site $x$ has a black button and that $p_{x}$ is the voltage $v_{x}$ at $x$ when a unit voltage is imposed between $a$ and $b$. Show that the current $i_{x y}$ is proportional to the net flow of black buttons along the edge $x y$. Does this suggest a hypothesis about the behavior of conduction electrons in metals?
### 1.3.4 Effective resistance and the escape probability
When we impose a voltage $v$ between points $a$ and $b$, a voltage $v_{a}=v$ is established at $a$ and $v_{b}=0$, and a current $i_{a}=\sum_{x} i_{a x}$ will flow into the circuit from the outside source. The amount of current that flows depends upon the overall resistance in the circuit. We define the effective resistance $R_{\mathrm{eff}}$ between $a$ and $b$ by $R_{\mathrm{eff}}=v_{a} / i_{a}$. The reciprocal quantity $C_{\mathrm{eff}}=1 / R_{\mathrm{eff}}=i_{a} / v_{a}$ is the effective conductance. If the voltage between $a$ and $b$ is multiplied by a constant, then the currents are multiplied by the same constant, so $R_{\text {eff }}$ depends only on the ratio of $v_{a}$ to $i_{a}$.
Let us calculate $R_{\text {eff }}$ for our example. When a unit voltage was imposed, we obtained the currents shown in Figure 18. The total current flowing into the circuit is $i_{a}=9 / 16+10 / 16=19 / 16$. Thus the effective resistance is
$$
R_{\mathrm{eff}}=\frac{v_{a}}{i_{a}}=\frac{1}{\frac{19}{16}}=\frac{16}{19}
$$
We can interpret the effective conductance probabilistically as an escape probability. When $v_{a}=1$, the effective conductance equals the total current $i_{a}$ flowing into $a$. This current is
$i_{a}=\sum_{y}\left(v_{a}-v_{y}\right) C_{a y}=\sum_{y}\left(v_{a}-v_{y}\right) \frac{C_{a y}}{C_{a}} C_{a}=C_{a}\left(1-\sum_{y} P_{a y} v_{y}\right)=C_{a} p_{\mathrm{esc}}$
where $p_{\text {esc }}$ is the probability, starting at $a$, that the walk reaches $b$ before returning to $a$. Thus
$$
C_{\mathrm{eff}}=C_{a} p_{\mathrm{esc}}
$$
and
$$
p_{\mathrm{esc}}=\frac{C_{\mathrm{eff}}}{C_{a}}
$$
In our example $C_{a}=2$ and we found that $i_{a}=19 / 16$. Thus
$$
p_{\mathrm{esc}}=\frac{19}{32}
$$
In calculating effective resistances, we shall use two important facts about electric networks. First, if two resistors are connected in series, they may be replaced by a single resistor whose resistance is the sum of the two resistances. (See Figure 21.) Secondly, two resistors in parallel

$\xrightarrow{R_{1}+R_{2}}$
Figure 21:
may be replaced by a single resistor with resistance $R$ such that
$$
\frac{1}{R}=\frac{1}{R_{1}}+\frac{1}{R_{2}}=\frac{R_{1} R_{2}}{R_{1}+R_{2}}
$$
(See Figure 22.)

Figure 22:
The second rule can be stated more simply in terms of conductances: If two resistors are connected in parallel, they may be replaced by a single resistor whose conductance is the sum of the two conductances.
We illustrate the use of these ideas to compute the effective resistance between two adjacent points of a unit cube of unit resistors, as shown in Figure 23. We put a unit battery between $a$ and $b$. Then,

Figure 23:
by symmetry, the voltages at $c$ and $d$ will be the same as will those at $e$ and $f$. Thus our circuit is equivalent to the circuit shown in Figure 24.

Figure 24:
Using the laws for the effective resistance of resistors in series and parallel, this network can be successively reduced to a single resistor of resistance 7/12 ohms, as shown in Figure 25. Thus the effective resistance is $7 / 12$. The current flowing into $a$ from the battery will be

a

Figure 25:
$i_{a}=\frac{1}{R_{\text {eff }}}=12 / 7$. The probability that a walk starting at $a$ will reach $b$ before returning to $a$ is
$$
p_{\mathrm{esc}}=\frac{i_{a}}{C_{a}}=\frac{\frac{12}{7}}{3}=\frac{4}{7}
$$
This example and many other interesting connections between electric networks and graph theory may be found in Bollobas [2].
Exercise 1.3.7 A bug walks randomly on the unit cube (see Figure 26). If the bug starts at $a$, what is the probability that it reaches food

Figure 26:
at $b$ before returning to $a$ ?
Exercise 1.3.8 Consider the Ehrenfest urn model with $N=4$ (see Exercise 1.3.5). Find the probability, starting at 0 , that state 4 is reached before returning to 0 .
Exercise 1.3.9 Consider the ladder network shown in Figure 27. Show that if $R_{n}$ is the effective resistance of a ladder with $n$ rungs then $R_{1}=2$ and
$$
R_{n+1}=\frac{2+2 R_{n}}{2+R_{n}}
$$
Use this to show that $\lim _{n \rightarrow \infty} R_{n}=\sqrt{2}$.
Exercise 1.3.10 A drunken tourist starts at her hotel and walks at random through the streets of the idealized Paris shown in Figure 28. Find the probability that she reaches the Arc de Triomphe before she reaches the outskirts of town.

Figure 27:

Figure 28:
### 1.3.5 Currents minimize energy dissipation
We have seen that when we impose a voltage between $a$ and $b$ voltages $v_{x}$ are established at the points and currents $i_{x y}$ flow through the resistors. In this section we shall give a characterization of the currents in terms of a quantity called energy dissipation. When a current $i_{x y}$ flows through a resistor, the energy dissipated is
$$
i_{x y}^{2} R_{x y}
$$
this is the product of the current $i_{x y}$ and the voltage $v_{x y}=i_{x y} R_{x y}$. The total energy dissipation in the circuit is
$$
E=\frac{1}{2} \sum_{x, y} i_{x y}^{2} R_{x y}
$$
Since $i_{x y} R_{x y}=v_{x}-v_{y}$, we can also write the energy dissipation as
$$
E=\frac{1}{2} \sum_{x, y} i_{x y}\left(v_{x}-v_{y}\right)
$$
The factor $1 / 2$ is necessary in this formulation since each edge is counted twice in this sum. For our example, we see from Figure 18 that
$$
E=\left(\frac{9}{16}\right)^{2} \cdot 1+\left(\frac{10}{16}\right)^{2} \cdot 1+\left(\frac{7}{16}\right)^{2} \cdot 1+\left(\frac{2}{16}\right)^{2} \cdot \frac{1}{2}+\left(\frac{12}{16}\right)^{2} \cdot \frac{1}{2}=\frac{19}{16}
$$
If a source (battery) establishes voltages $v_{a}$ and $v_{b}$ at $a$ and $b$, then the energy supplied is $\left(v_{a}-v_{b}\right) i_{a}$ where $i_{a}=\sum_{x} i_{a x}$. By conservation of energy, we would expect this to be equal to the energy dissipated. In our example $v_{a}-v_{b}=1$ and $i_{a}=\frac{19}{16}$, so this is the case. We shall show that this is true in a somewhat more general setting.
Define a flow $\mathbf{j}$ from $a$ to $b$ to be an assignment of numbers $j_{x y}$ to pairs $x y$ such that
(a) $j_{x y}=-j_{y x}$
(b) $\sum_{y} j_{x y}=0$ if $x \neq a, b$
(c) $j_{x y}=0$ if $x$ and $y$ are not adjacent.
We denote by $j_{x}=\sum_{y} j_{x y}$ the flow into $x$ from the outside. By (b) $j_{x}=0$ for $x \neq a, b$. Of course $j_{b}=-j_{a}$. To verify this, note that
$$
j_{a}+j_{b}=\sum_{x} j_{x}=\sum_{x} \sum_{y} j_{x y}=\frac{1}{2} \sum_{x, y}\left(j_{x y}+j_{y x}\right)=0
$$
since $j_{x y}=-j_{y x}$.
With this terminology, we can now formulate the following version of the principle of conservation of energy:
Conservation of Energy. Let $w$ be any function defined on the points of the graph and $\mathbf{j}$ a flow from $a$ to $b$. Then
$$
\left(w_{a}-w_{b}\right) j_{a}=\frac{1}{2} \sum_{x, y}\left(w_{x}-w_{y}\right) j_{x y}
$$
Proof.
$$
\begin{aligned}
\sum_{x, y}\left(w_{x}-w_{y}\right) j_{x y} & =\sum_{x}\left(w_{x} \sum_{y} j_{x y}\right)-\sum_{y}\left(w_{y} \sum_{x} j_{x y}\right) \\
& =w_{a} \sum_{y} j_{a y}+w_{b} \sum_{y} j_{b y}-w_{a} \sum_{x} j_{x a}-w_{b} \sum_{x} j_{x b} \\
& =w_{a} j_{a}+w_{b} j_{b}-w_{a}\left(-j_{a}\right)-w_{b}\left(-j_{b}\right) \\
& =2\left(w_{a}-w_{b}\right) j_{a}
\end{aligned}
$$
Thus
$$
\left(w_{a}-w_{b}\right) j_{a}=\frac{1}{2} \sum_{x, y}\left(w_{x}-w_{y}\right) j_{x y}
$$
as was to be proven. $\diamond$
If we now impose a voltage $v_{a}$ between $a$ and $b$ with $v_{b}=0$, we obtain voltages $v_{x}$ and currents $i_{x y}$. The currents $\mathbf{i}$ give a flow from $a$ to $b$ and so by the previous result, we conclude that
$$
v_{a} i_{a}=\frac{1}{2} \sum_{x, y}\left(v_{x}-v_{y}\right) i_{x y}=\frac{1}{2} \sum_{x, y} i_{x y}^{2} R_{x y}
$$
Recall that $R_{\mathrm{eff}}=v_{a} / i_{a}$. Thus in terms of resistances we can write this as
$$
i_{x y}^{2} R_{\mathrm{eff}}=\frac{1}{2} \sum_{x, y} i_{x y}^{2} R_{x y}
$$
If we adjust $v_{a}$ so that $i_{a}=1$, we call the resulting flow the unit current flow from $a$ to $b$. The unit current flow from $a$ to $b$ is a particular example of a unit flow from $a$ to $b$, which we define to be any flow $i_{x y}$ from $a$ to $b$ for which $i_{a}=-i_{b}=1$. The formula above shows that the energy dissipated by the unit current flow is just $R_{\text {eff. }}$. According to a basic result called Thomson's Principle, this value is smaller than the energy dissipated by any other unit flow from $a$ to $b$. Before proving this principle, let us watch it in action in the example worked out above.
Recall that, for this example, we found the true values and some approximate values for the unit current flow; these were shown in Figure 20. The energy dissipation for the true currents is
$$
E=R_{\mathrm{eff}}=\frac{16}{19}=.8421053
$$
Our approximate currents also form a unit flow and, for these, the energy dissipation is
$$
\bar{E}=(.4754)^{2} \cdot 1+(.5246)^{2} \cdot 1+(.3672)^{2} \cdot 1+(.1082)^{2} \cdot \frac{1}{2}+(.6328)^{2} \cdot \frac{1}{2}=.8421177
$$
We note that $\bar{E}$ is greater than $E$, though just barely.
Thomson's Principle. (Thomson [33]). If $\mathbf{i}$ is the unit flow from $a$ to $b$ determined by Kirchhoff's Laws, then the energy dissipation $\frac{1}{2} \sum_{x, y} i_{x y}^{2} R_{x y}$ minimizes the energy dissipation $\frac{1}{2} \sum_{x, y} j_{x y}^{2} R_{x y}$ among all unit flows $\mathbf{j}$ from $a$ to $b$.
Proof. Let $\mathbf{j}$ be any unit flow from $a$ to $b$ and let $d_{x y}=j_{x y}-i_{x y}$. Then $\mathbf{d}$ is a flow from $a$ to $b$ with $d_{a}=\sum_{x} d_{a x}=1-1=0$.
$$
\begin{aligned}
\sum_{x, y} j_{x y}^{2} R_{x y} & =\sum_{x, y}\left(i_{x y}+d_{x y}\right)^{2} R_{x y} \\
& =\sum_{x, y} i_{x y}^{2} R_{x y}+2 \sum_{x, y} i_{x y} R_{x y} d_{x y}+\sum_{x, y} d_{x y}^{2} R_{x y} \\
& =\sum_{x, y} i_{x y}^{2} R_{x y}+2 \sum_{x, y}\left(v_{x}-v_{y}\right) d_{x y}+\sum_{x, y} d_{x y}^{2} R_{x y}
\end{aligned}
$$
Setting $\mathbf{w}=\mathbf{v}$ and $\mathbf{j}=\mathbf{d}$ in the conservation of energy result above shows that the middle term is $4\left(v_{a}-v_{b}\right) d_{a}=0$. Thus
$$
\sum_{x, y} j_{x y}^{2} R_{x y}=\sum_{x, y} i_{x y}^{2} R_{x y}+\sum_{x, y} d_{x y}^{2} R_{x y} \geq \sum_{x, y} i_{x y}^{2} R_{x y}
$$
This completes the proof. $\diamond$
Exercise 1.3.11 The following is the so-called "dual form" of Thomson's Principle. Let $u$ be any function on the points of the graph $G$ of a circuit such that $u_{a}=1$ and $u_{b}=0$. Then the energy dissipation
$$
\frac{1}{2} \sum_{x, y}\left(u_{x}-u_{y}\right)^{2} C_{x y}
$$
is minimized by the voltages $v_{x}$ that result when a unit voltage is established between $a$ and $b$, i.e., $v_{a}=1, v_{b}=0$, and the other voltages are determined by Kirchhoff's Laws. Prove this dual principle. This second principle is known nowadays as the Dirichlet Principle, though it too was discovered by Thomson.
Exercise 1.3.12 In Section 1.2.4 we stated that, to solve the Dirichlet problem by the method of relaxations, we could start with an arbitrary initial guess. Show that when we replace the value at a point by the average of the neighboring points the energy dissipation, as expressed in Exercise 1.3.11, can only decrease. Use this to prove that the relaxation method converges to a solution of the Dirichlet problem for an arbitrary initial guess.
### 1.4 Rayleigh's Monotonicity Law
### 1.4.1 Rayleigh's Monotonicity Law
Next we will study Rayleigh's Monotonicity Law. This law from electric network theory will be an important tool in our future study of random walks. In this section we will give an example of the use of this law.
Consider a random walk on streets of a city as in Figure 29. Let

Figure 29:
$p_{\text {esc }}$ be the probability that a walker starting from $a$ reaches $b$ before returning to $a$. Assign to each edge a unit resistance and maintain a voltage of one volt between $a$ and $b$; then a current $i_{a}$ will flow into the circuit and we showed in Section 1.3 .4 that
$$
p_{\mathrm{esc}}=\frac{i_{a}}{C_{a}}=\frac{i_{a}}{4}
$$
Now suppose that one of the streets (not connected to $a$ ) becomes blocked. Our walker must choose from the remaining streets if he reaches a corner of this street. The modified graph will be as in Figure 30. We want to show that the probability of escaping to $b$ from $a$ is

Figure 30:
decreased.
Consider this problem in terms of our network. Blocking a street corresponds to replacing a unit resistor by an infinite resistor. This should have the effect of increasing the effective resistance $R_{\text {eff }}$ of the circuit between $a$ and $b$. If so, when we put a unit voltage between $a$ and $b$ less current will flow into the circuit and
$$
p_{\mathrm{esc}}=\frac{i_{a}}{4}=\frac{1}{4 R_{\mathrm{eff}}}
$$
will decrease.
Thus we need only show that when we increase the resistance in one part of a circuit, the effective resistance increases. This fact, known as Rayleigh's Monotonicity Law, is almost self-evident. Indeed, the father of electromagnetic theory, James Clerk Maxwell, regarded this to be the case. In his Treatise on Electricity and Magnetism ([21], p. 427), he wrote
If the specific resistance of any portion of the conductor be changed, that of the remainder being unchanged, the
resistance of the whole conductor will be increased if that of the portion is increased, and diminished if that of the portion is diminished. This principle may be regarded as self-evident ....
Rayleigh's Monotonicity Law. If the resistances of a circuit are increased, the effective resistance $R_{\text {eff }}$ between any two points can only increase. If they are decreased, it can only decrease.
Proof. Let $\mathbf{i}$ be the unit current flow from $a$ to $b$ with the resistors $R_{x y}$. Let $\mathbf{j}$ be the unit current flow from $a$ to $b$ with the resistors $\bar{R}_{x y}$ with $\bar{R}_{x y} \geq R_{x y}$. Then
$$
\bar{R}_{\mathrm{eff}}=\frac{1}{2} \sum_{x, y} j_{x y}^{2} \bar{R}_{x y} \geq \frac{1}{2} \sum_{x, y} j_{x y}^{2} R_{x y}
$$
But since $\mathbf{j}$ is a unit flow from $a$ to $b$, Thomson's Principle tells us that the energy dissipation, calculated with resistors $R_{x y}$, is bigger than that for the true currents determined by these resistors: that is
$$
\frac{1}{2} \sum_{x, y} j_{x y}^{2} R_{x y} \geq \frac{1}{2} \sum_{x, y} i_{x y}^{2} R_{x y}=R_{\mathrm{eff}}
$$
Thus $\bar{R}_{\text {eff }} \geq R_{\text {eff. }}$. The proof for the case of decreasing resistances is the same.
Exercise 1.4.1 Consider a graph $G$ and let $R_{x y}$ and $\bar{R}_{x y}$ be two different assignments of resistances to the edges of $G$. Let $\hat{R}_{x y}=\bar{R}_{x y}+R_{x y}$. Let $R_{\text {eff }}, \bar{R}_{\text {eff }}$, and $\hat{R}_{\text {eff }}$ be the effective resistances when $R, \bar{R}$, and $\hat{R}$, respectively, are used. Prove that
$$
\hat{R}_{\mathrm{eff}} \geq \bar{R}_{\mathrm{eff}}+R_{\mathrm{eff}}
$$
Conclude that the effective resistance of a network is a concave function of any of its component resistances (Shannon and Hagelbarger [31.)
Exercise 1.4.2 Show that the effective resistance of the circuit in Figure 31 is greater than or equal to the effective resistance of the circuit in Figure 32 . Use this to show the following inequality for $R_{i j} \geq 0$ :
$$
\frac{1}{\frac{1}{R_{11}+R_{12}}+\frac{1}{R_{21}+R_{22}}} \geq \frac{1}{\frac{1}{R_{11}}+\frac{1}{R_{21}}}+\frac{1}{\frac{1}{R_{12}}+\frac{1}{R_{22}}}
$$
See the note by Lehman [18] for a proof of the general Minkowski inequality by this method.

Figure 31:

Figure 32:
Exercise 1.4.3 Let $\mathbf{P}$ be the transition matrix associated with an electric network and let $a, b, r, s$ be four points on the network. Let $\overline{\mathbf{P}}$ be a transition matrix defined on the state-space $S=\{a, b, r, s\}$. Let $\bar{P}_{i i}=0$ and for $i \neq j$ let $\bar{P}_{i j}$ be the probability that, if the chain $\mathbf{P}$ is started in state $i$, then the next time it is in the set $S-\{i\}$ it is in the state $j$. Show that $\overline{\mathbf{P}}$ is a reversible Markov chain and hence corresponds to an electric network of the form of a Wheatstone Bridge, shown in Figure 33. Explain how this proves that, in order to prove the Monotonicity Law, it is sufficient to prove that $R_{\text {eff }}$ is a monotone function of the component resistances for a Wheatstone Bridge. Give a direct proof of the Monotonicity Law for this special case.
### 1.4.2 A probabilistic explanation of the Monotonicity Law
We have quoted Maxwell's assertion that Rayleigh's Monotonicity Law may be regarded as self-evident, but one might feel that any argument in terms of electricity is only self-evident if we know what electricity is. In Cambridge, they tell the following story about Maxwell: Maxwell was lecturing and, seeing a student dozing off, awakened him, asking, "Young man, what is electricity?" "I'm terribly sorry, sir," the student

Figure 33:
replied, 'I knew the answer but I have forgotten it." Maxwell's response to the class was, "Gentlemen, you have just witnessed the greatest tragedy in the history of science. The one person who knew what electricity is has forgotten it."
To say that our intuition about the Monotonicity Law is only as solid as our understanding of electricity is not really a valid argument, of course, because in saying that this law is self-evident we are secretly depending on the analogy between electricity and the flow of water (see Feynman [6], Vol. 2, Chapter 12). We just can't believe that if a water main gets clogged the total rate of flow out of the local reservoir is going to increase. But as soon as we admit this, some pedant will ask if we're talking about flows with low Reynolds number, or what, and we'll have to admit that we don't understand water any better than we understand electricity.
Whatever our feelings about electricity or the flow of water, it seems desirable to have an explanation of the Monotonicity Law in terms of our random walker. We now give such an explanation.
We begin by collecting some notation and results from previous sections. As usual, we have a network of conductances (streets) and a walker who moves from point $x$ to point $y$ with probability
$$
P_{x y}=\frac{C_{x y}}{C_{x}}
$$
where $C_{x y}$ is the conductance from $x$ to $y$ and $C_{x}=\sum_{y} C_{x y}$. We choose two preferred points $a$ and $b$. The walker starts at $a$ and walks until he reaches $b$ or returns to $a$. We denote by $v_{x}$ the probability that the walker, starting at $a$, reaches $a$ before $b$. Then $v_{a}=1, v_{b}=0$, and the
function $v_{x}$ is harmonic at all points $x \neq a, b$. We denote by $p_{\text {esc }}$ the probability that the walker, starting at $a$, reaches $b$ before returning to a. Then
$$
p_{\mathrm{esc}}=1-\sum_{x} p_{a x} v_{x}
$$
Now we have seen that the effective conductance between $a$ and $b$ is
$$
C_{a} p_{\text {esc }}
$$
We wish to show that this increases whenever one of the conductances $C_{r s}$ is increased. If $a$ is different from $r$ or $s$, we need only show that $p_{\text {esc }}$ increases. The case where $r$ or $s$ coincides with $a$ is easily disposed of (see Exercise 1.4.4). The case where $r$ or $s$ coincides with $b$ is also easy (see Exercise 1.4.5). Hence from now on we will assume that $r, s \neq a$ and $r, s \neq b$.
Instead of increasing $C_{r s}$, we can think of adding a new edge of conductance $\epsilon$ between $r$ and $s$. (See Figure 34.) We will call this

Figure 34:
new edge a "bridge" to distinguish it from the other edges. Note that the graph with the bridge added will have more than one edge between
$r$ and $s$ (unless there was no edge between $r$ and $s$ in the original graph), and this will complicate any expression that involves summing over edges. Everything we have said or will say holds for graphs with "multiple edges" as well as for graphs without them. So far, we have chosen to keep our expressions simple by assuming that an edge is determined by its endpoints. The trade-off is that in the manipulations below, whenever we write a sum over edges we will have to add an extra term to account for the bridge.
Why should adding the bridge increase the escape probability? The first thing you think is, "Of course, it opens up new possibilities of escaping!" The next instant you think, "Wait a minute, it also opens up new possibilities of returning to the starting point. What ensures that the first effect will outweigh the second?" As we shall see, the proper reply is, "Because the walker will cross the bridge more often in the good direction than in the bad direction." To turn this blithe reply into a real explanation will require a little work, however.
To begin to make sense of the notion that the bridge gets used more often in the good direction than the bad, we will make a preliminary argument that applies to any edge of any graph. Let $G$ be any graph, and let $r s$ be any edge with endpoints not $a$ or $b . v_{r}>v_{s}$. Since the walker has a better chance to escape from $s$ than from $r$, this means that to cross this edge in the good direction is to go from $r$ to $s$ We shall show that the walker will cross the edge from $r$ to $s$ more often on the average than from $s$ to $r$.
Let $u_{x}$ be the expected number of times the walker is at $x$ and $u_{x y}$ the expected number of times he crosses the edge $x y$ from $x$ to $y$ before he reaches $b$ or returns to $a$. The calculation carried out in Section 1.3.3 shows that $u_{x} / C_{x}$ is harmonic for $x \neq a, b$ with $u_{a} / C_{a}=1 / C_{a}$ and $b_{b} / C_{b}=0$. But the function $v_{x} / C_{a}$ also has these properties, so by the Uniqueness Principle
$$
\frac{u_{x}}{C_{x}}=\frac{v_{x}}{C_{a}}
$$
Now
$$
u_{r s}=u_{r} P_{r s}=u_{r} \frac{C_{r s}}{C_{r}}=v_{r} \frac{C_{r s}}{C_{a}}
$$
and
$$
u_{s r}=u_{s} P_{s r}=u_{s} \frac{C_{s r}}{C_{s}}=v_{s} \frac{C_{s r}}{C_{a}}
$$
Since $C_{r s}=C_{s r}$, and since by assumption $v_{r} \geq v_{s}$, this means that $u_{r s} \geq u_{s r}$. Therefore, we see that any edge leads the walker more often to the more favorable of the points of the edge.
Now let's go back and think about the graph with the bridge. The above discussion shows that the bridge helps in the sense that, on the average, the bridge is crossed more often in the direction that improves the chance of escaping. While this fact is suggestive, it doesn't tell us that we are more likely to escape than if the bridge weren't there; it only tells us what goes on once the bridge is in place. What we need is to make a "before and after" comparison.
Recall that we are denoting the conductance of the bridge by $\epsilon$. To distinguish the quantities pertaining to the walks with and without the bridge, we will put $(\epsilon)$ superscripts on quantities that refer to the walk with the bridge, so that, e.g., $p_{\text {esc }}^{(\epsilon)}$ denotes the escape probability with the bridge.
Now let $d^{(\epsilon)}$ denote the expected net number of times the walker crosses from $r$ to $s$. As above, we have
$$
d^{(\epsilon)}=u_{r}^{(\epsilon)} \frac{\epsilon}{C_{r}+\epsilon}-u_{s}^{(\epsilon)} \frac{\epsilon}{C_{s}+\epsilon}=\left(\frac{u_{r}^{(\epsilon)}}{C_{r}+\epsilon}-\frac{u_{s}^{(\epsilon)}}{C_{s}+\epsilon}\right) \epsilon
$$
Claim.
$$
p_{\mathrm{eSC}}^{(\epsilon)}=p_{\mathrm{esc}}+\left(v_{r}-v_{s}\right) d^{(\epsilon)}
$$
Why. Every time you use the bridge to go from $r$ to $s$, you improve your chances of escaping by
$$
\left(1-v_{s}\right)-\left(1-v_{r}\right)=v_{r}-v_{s}
$$
assuming that you would continue your walk without using the bridge. To get the probability of escaping with the bridge, you take the probability of escaping without the bridge, and correct it by adding in the change attributable to the bridge, which is the difference in the original escape probabilities at the ends of the bridge, multiplied by the net number of times you expect to cross the bridge. $\odot$
Proof. Suppose you're playing a game where you walk around the graph with the bridge, and your fortune when you're at $x$ is $v_{x}$, which is the probability that you would return to $a$ before reaching $b$ if the bridge weren't there. You start at $a$, and walk until you reach $b$ or return to $a$.
This is not a fair game. Your initial fortune is 1 since you start at $a$ and $v_{a}=1$. Your expected final fortune is
$$
1 \cdot\left(1-p_{\mathrm{eSc}}^{(\epsilon)}\right)+0 \cdot p_{\mathrm{eSc}}^{(\epsilon)}=1-p_{\mathrm{eSc}}^{(\epsilon)}
$$
The amount you expect to lose by participating in the game is
$$
p_{\mathrm{eSc}}^{(\epsilon)}
$$
(Note that escaping has suddenly become a bad thing!)
Let's analyze where it is that you expect to lose money. First of all, you lose money when you take the first step away from $a$. The amount you expect to lose is
$$
1-\sum_{x} P_{a x}^{(\epsilon)} v_{x}=p_{\mathrm{esc}}
$$
Now if your fortune were given by $v_{x}^{(\epsilon)}$ instead of $v_{x}$, the game would be fair after this first step. However, the function $v_{x}$ is not harmonic for the walk with the bridge; it fails to be harmonic at $r$ and $s$. Every time you step away from $r$, you expect to lose an amount
$$
v_{r}-\left(\sum_{x} \frac{C_{r s}}{C_{r}+\epsilon} v_{x}+\frac{\epsilon}{C_{r}+\epsilon} v_{s}\right)=\left(v_{r}-v_{s}\right) \frac{\epsilon}{C_{r}+\epsilon}
$$
Similarly, every time you step away from $s$ you expect to lose an amount
$$
\left(v_{s}-v_{r}\right) \frac{\epsilon}{C_{s}+\epsilon}
$$
The total amount you expect to lose by participating in the game is:
$$
\begin{aligned}
& \text { expected loss at first step }+ \\
& (\text { expected loss at } r) \cdot(\text { expected number of times at } r)+ \\
& (\text { expected loss at } s) \cdot(\text { expected number of times at } s) \\
= & p_{\text {esc }}+ \\
& \left(v_{r}-v_{s}\right) \frac{\epsilon}{C_{r}+\epsilon} u_{r}^{(\epsilon)}+ \\
& \left(v_{s}-v_{r}\right) \frac{\epsilon}{C_{s}+\epsilon} u_{s}^{(\epsilon)} \\
= & p_{\text {esc }}+\left(v_{r}-v_{s}\right) d^{(\epsilon)}
\end{aligned}
$$
Equating this with our first expression for the expected cost of playing the game yields the formula we were trying to prove.
According to the formula just established,
$$
p_{\mathrm{eSc}}^{(\epsilon)}-p_{\mathrm{esc}}=\left(v_{r}-v_{s}\right)\left(\frac{u_{r}^{(\epsilon)}}{C_{r}+\epsilon}-\frac{u_{s}^{(\epsilon)}}{C_{s}+\epsilon}\right) \epsilon
$$
For small $\epsilon$, we will have
$$
\frac{u_{r}^{(\epsilon)}}{C_{r}+\epsilon}-\frac{u_{s}^{(\epsilon)}}{C_{s}+\epsilon} \approx \frac{u_{r}}{C_{r}}-\frac{u_{s}}{C_{s}}=\frac{v_{r}}{C_{a}}-\frac{v_{s}}{C_{a}}
$$
so for small $\epsilon$
$$
p_{\mathrm{eSC}}^{(\epsilon)}-p_{\mathrm{esc}} \approx\left(v_{r}-v_{s}\right)^{2} \frac{\epsilon}{C_{a}}
$$
This approximation allows us to conclude that
$$
p_{\mathrm{eSc}}^{(\epsilon)} \geq p_{\mathrm{eSc}} \geq 0
$$
for small $\epsilon$. But this is enough to establish the monotonicity law, since any finite change in $\epsilon$ can be realized by making an infinite chain of graphs each of which is obtained from the one before by adding a bridge of infinitesimal conductance.
To recapitulate, the difference in the escape probabilities with and without the bridge is obtained by taking the difference between the original escape probabilities at the ends of the bridge, and multiplying by the expected net number of crossings of the bridge. This quantity is positive because the walker tends to cross the bridge more often in the good direction than in the bad direction.
Exercise 1.4.4 Give a probabilistic argument to show that $C_{a} p_{\text {esc }}$ increases with $C_{a r}$ for any $r$. Give an example to show that $p_{\text {esc }}$ by itself may actually decrease.
Exercise 1.4.5 Give a probabilistic argument to show that $C_{a} p_{\text {esc }}$ increases with $C_{r b}$ for any $r$.
Exercise 1.4.6 Show that when $v_{r}=v_{s}$, changing the value of $C_{r s}$ does not change $p_{\text {esc }}$.
Exercise 1.4.7 Show that
$$
\frac{\partial}{\partial R_{r s}} R_{\mathrm{eff}}=i_{r s}^{2}
$$
Exercise 1.4.8 In this exercise we ask you to derive an exact formula for the change in escape probability
$$
p_{\mathrm{eSc}}^{(\epsilon)}-p_{\mathrm{eSc}}
$$
in terms of quantities that refer only to the walk without the bridge.
(a) Let $N_{x y}$ denote the expected number of times in state $y$ for a walker who starts at $x$ and walks around the graph without the bridge until he reaches $a$ or $b$. It is a fact that
$$
u_{r}^{(\epsilon)}=u_{r}+u_{r}^{(\epsilon)} \frac{\epsilon}{C_{r}+\epsilon}\left(N_{s r}+1-N_{r r}\right)+u_{s}^{(\epsilon)} \frac{\epsilon}{C_{s}+\epsilon}\left(N_{r r}-N_{s r}\right)
$$
Explain in words why this formula is true.
(b) This equation for $u_{r}^{(\epsilon)}$ can be rewritten as follows:
$$
\frac{C_{r}}{C_{r}+\epsilon} u_{r}^{(\epsilon)}=u_{r}+d^{(\epsilon)}\left(N_{s r}-N_{r r}\right)
$$
Prove this formula. (Hint: Consider a game where your fortune at $x$ is $N_{x r}$, and where you start from $a$ and walk on the graph with the bridge until you reach $b$ or return to $a$.)
(c) Write down the corresponding formula for $u_{s}^{(\epsilon)}$, and use this formula to get an expression for $d^{(\epsilon)}$ in terms of quantities that refer to the walk without the bridge.
(d) Use the expression for $d^{(\epsilon)}$ to express $p_{\mathrm{eSc}}^{(\epsilon)}-p_{\mathrm{esc}}$ in terms of quantities that refer to the walk without the bridge, and verify that the value of your expression is $\geq 0$ for $\epsilon \geq 0$.
Exercise 1.4.9 Give a probabilistic interpretation of the energy dissipation rate.
### 1.4.3 A Markov chain proof of the Monotonicity Law
Let $\mathbf{P}$ be the ergodic Markov chain associated with an electric network. When we add an $\epsilon$ bridge from $r$ to $s$, we obtain a new transition matrix $\mathbf{P}^{(\epsilon)}$ that differs from $\mathbf{P}$ only for transitions from $r$ and $s$. We can minimize the differences between $\mathbf{P}$ and $\mathbf{P}^{(\epsilon)}$ by replacing $\mathbf{P}$ by the matrix $\hat{\mathbf{P}}$ corresponding to the circuit without the bridge but with an $\epsilon$ conductance added from $r$ to $r$ and from $s$ to $s$. This allows the chain to stay in states $r$ and $s$ but does not change the escape probability
from $a$ to $b$. Thus, we can compare the escape probabilities for the two matrices $\hat{\mathbf{P}}$ and $\mathbf{P}^{(\epsilon)}$, which differ only by
$$
\begin{array}{ll}
\hat{P}_{r r}=\frac{\epsilon}{C_{r}+\epsilon} & P_{r r}^{(\epsilon)}=0 \\
\hat{P}_{r s}=\frac{C_{r s}}{C_{r}+\epsilon} & P_{r s}^{(\epsilon)}=\frac{C_{r s}+\epsilon}{C_{r}+\epsilon} \\
\hat{P}_{s s}=\frac{\epsilon}{C_{s}+\epsilon} & P_{s s}^{(\epsilon)}=0 \\
\hat{P}_{s r}=\frac{C_{s r}}{C_{S}+\epsilon} & P_{s r}^{(\epsilon)}=\frac{C_{s r}+\epsilon}{C_{s}+\epsilon}
\end{array}
$$
We make states $a$ and $b$ into absorbing states. Let $\hat{\mathbf{N}}$ and $\mathbf{N}^{(\epsilon)}$ be the fundamental matrices for the absorbing chains obtained from $\hat{\mathbf{P}}$ and $\mathbf{P}^{(\epsilon)}$ respectively. Then $\hat{\mathbf{N}}=(\mathbf{I}-\hat{\mathbf{Q}})^{-1}$ and $\mathbf{N}^{(\epsilon)}=\left(\mathbf{I}-\mathbf{Q}^{(\epsilon)}\right)^{-1}$ where $\hat{\mathbf{Q}}$ and $\mathbf{Q}^{(\epsilon)}$ differ only for the four components involving only $r$ and $s$. That is,

where $\mathbf{h}$ is the column vector with only components $r$ and $s$ non-zero
$$
\mathbf{h}=\begin{array}{r}
r \\
s
\end{array}\binom{\frac{\epsilon}{C_{r}+\epsilon}}{-\frac{\epsilon}{C_{s}+\epsilon}}
$$
and $\mathbf{k}$ is a row vector with only components $r$ and $s$ non-zero
$$
\mathbf{k}=\left(\begin{array}{cc}
r & s \\
-1 & 1
\end{array}\right)
$$
J. G. Kemeny has pointed out to us that if $\mathbf{A}$ is any matrix with inverse $\mathbf{N}$ and we add to $\mathbf{A}$ a matrix of the form $-\mathbf{h k}$, then $\mathbf{A}-\mathbf{h k}$ has an inverse if and only if $\mathbf{k N h} \neq 1$ and, if so, $\overline{\mathbf{N}}=(\mathbf{A}-\mathbf{h k})^{-1}$ is given by
$$
\overline{\mathbf{N}}=\mathbf{N}+c(\mathbf{N h})(\mathbf{k N})
$$
where $c=1 /(1-\mathbf{k N h})$. You are asked to prove this in Exercise 1.4.10. Adding $-\mathbf{h k}$ to $\mathbf{A}=\mathbf{I}-\hat{\mathbf{Q}}$ and using this result, we obtain
$$
\mathbf{N}^{(\epsilon)}=\hat{\mathbf{N}}+c(\hat{\mathbf{N}} \mathbf{h})(\mathbf{k} \hat{\mathbf{N}})
$$
Using the simple nature of $\mathbf{h}$ and $\mathbf{k}$ we obtain
$$
N_{i j}^{(\epsilon)}=\hat{N}_{i j}+\left(\frac{\hat{N}_{i r} \epsilon}{C_{r}+\epsilon}-\frac{\hat{N}_{i s} \epsilon}{C_{s}+\epsilon}\right)\left(\hat{N}_{s j}-\hat{N}_{r j}\right)
$$
and
$$
c=\frac{1}{1+\frac{\hat{N}_{r r} \epsilon}{C_{r}+\epsilon}-\frac{\hat{N}_{s r} \epsilon}{C_{r}+\epsilon}+\frac{\hat{N}_{s s} \epsilon}{C_{s}+\epsilon}-\frac{\hat{N}_{r s} \epsilon}{C_{s}+\epsilon}}
$$
Since $\hat{N}_{r r}$ is the expected number of times in $r$ starting in $r$ and $\hat{N}_{s r}$ is the expected number of times in $r$ starting in $s, \hat{N}_{r r} \geq \hat{N}_{s r}$. Similarly $\hat{N}_{s s} \geq \hat{N}_{r s}$ and so the denominator of $c$ is $\geq 1$. In particular, it is positive.
Recall that the absorption probabilities for state $b$ are given by
$$
B_{x b}=\sum_{y} N_{x y} P_{y b}
$$
Since $P_{x b}^{(\epsilon)}=\hat{P}_{x b}$,
$$
B_{x b}^{(\epsilon)}=\hat{B}_{x b}+c\left(\frac{\hat{N}_{x r} \epsilon}{C_{r}+\epsilon}-\frac{\hat{N}_{x s} \epsilon}{C_{s}+\epsilon}\right)\left(\hat{B}_{s b}-\hat{B}_{r b}\right)
$$
Since $P_{a x}^{(\epsilon)}=\hat{P}_{a x}$,
$$
p_{\mathrm{eSc}}^{(\epsilon)}=\hat{p}_{\mathrm{esc}}+c\left(\frac{\hat{u}_{r} \epsilon}{C_{r}+\epsilon}-\frac{\hat{u}_{s} \epsilon}{C_{s}+\epsilon}\right)\left(\hat{B}_{s b}-\hat{B}_{r b}\right)
$$
where $\hat{u}_{x}$ is the expected number of times that the ergodic chain $\hat{P}$, started at state $a$, is in state $x$ before returning to $a$ reaching $b$ for the first time. The absorption probability $B_{x a}$ is the quantity $v_{x}$ introduced in the previous section. As shown there, reversibility allows us to conclude that
$$
\frac{\hat{u}_{x}}{\hat{C}_{x}}=\frac{\hat{B}_{x a}}{\hat{C}_{a}}=\frac{\hat{B}_{x a}}{C_{a}}
$$
so that
$$
p_{\mathrm{eSc}}^{(\epsilon)}=p_{\mathrm{esc}}+\frac{\epsilon c}{C_{a}}\left(\hat{B}_{s b}-\hat{B}_{r b}\right)^{2}
$$
and this shows that the Monotonicity Law is true.
The change from $\mathbf{P}$ to $\hat{\mathbf{P}}$ was merely to make the calculations easier. As we have remarked, the escape probabilities are the same for the two chains as are the absorption probabilities $B_{i b}$. Thus we can remove the hats and write the same formula.
$$
p_{\mathrm{eSc}}^{(\epsilon)}=p_{\mathrm{esc}}+\frac{\epsilon c}{C_{a}}\left(B_{s b}-B_{r b}\right)^{2}
$$
The only quantity in this final expression that seems to depend upon quantities from $\hat{\mathbf{P}}$ is $c$. In Exercise 1.4.11 you are asked to show that $c$ can also be expressed in terms of the fundamental matrix $\mathbf{N}$ obtained from the original $\mathbf{P}$.
Exercise 1.4.10 Let $\mathbf{A}$ be a matrix with inverse $\mathbf{N}=\mathbf{A}^{-1}$ Let $\mathbf{h}$ be a column vector and $\mathbf{k}$ a row vector. Show that
$$
\overline{\mathbf{N}}=(\mathbf{A}-\mathbf{h k})^{-1}
$$
exists if and only if $\mathbf{k N h} \neq 1$ and, if so,
$$
\overline{\mathbf{N}}=\mathbf{N}+\frac{(\mathrm{Nh})(\mathrm{kN})}{1-\mathrm{kNh}}
$$
Exercise 1.4.11 Show that $c$ can be expressed in terms of the fundamental matrix $\mathbf{N}$ of the original Markov chain $\mathbf{P}$ by
$$
c=\frac{1}{1+\frac{N_{r r} \epsilon}{C_{r}}-\frac{N_{s r} \epsilon}{C_{r}}+\frac{N_{s s} \epsilon}{C_{s}}-\frac{N_{r s} \epsilon}{C_{s}}}
$$
## 2 Random walks on infinite networks
### 2.1 Polya's recurrence problem
### 2.1.1 Random walks on lattices
In 1921 George Polya [26] investigated random walks on certain infinite graphs, or as he called them, "street networks". The graphs he considered, which we will refer to as lattices, are illustrated in Figure 35.
To construct a $d$-dimensional lattice, we take as vertices those points $\left(x_{1}, \ldots, x_{d}\right)$ of $\mathbf{R}^{d}$ all of whose coordinates are integers, and we join each vertex by an undirected line segment to each of its $2 d$ nearest neighbors. These connecting segments, which represent the edges of our graph, each have unit length and run parallel to one of the coordinate axes of $\mathbf{R}^{d}$. We will denote this $d$-dimensional lattice by $\mathbf{Z}^{d}$. We will denote the origin $(0,0, \ldots, 0)$ by $\mathbf{0}$.
Now let a point walk around at random on this lattice. As usual, by walking at random we mean that, upon reaching any vertex of the graph, the probability of choosing any one of the $2 d$ edges leading out of that vertex is $\frac{1}{2 d}$. We will call this random walk simple random walk in $d$ dimensions.
When $d=1$, our lattice is just an infinite line divided into segments of length one. We may think of the vertices of this graph as representing the fortune of a gambler betting on heads or tails in a fair coin

1-dimensional lattice

2.dimensional battice
Figure 35:
tossing game. Simple random walk in one dimension then represents the vicissitudes of his or her fortune, either increasing or decreasing by one unit after each round of the game.
When $d=2$, our lattice looks like an infinite network of streets and avenues, which is why we describe the random motion of the wandering point as a "walk".
When $d=3$, the lattice looks like an infinite "jungle gym", so perhaps in this case we ought to talk about a "random climb", but we will not do so. It is worth noting that when $d=3$, the wanderings of our point could be regarded as an approximate representation of the random path of a molecule diffusing in an infinite cubical crystal. Figure 36 shows a simulation of a simple random walk in three dimensions.
### 2.1.2 The question of recurrence
The question that Polya posed amounts to this: "Is the wandering point certain to return to its starting point during the course of its wanderings?" If so, we say that the walk is recurrent. If not, that is, if there is a positive probability that the point will never return to its starting point, then we say that the walk is transient.
If we denote the probability that the point never returns to its starting point by $p_{\text {esc }}$, then the chain is recurrent if $p_{\text {esc }}=0$, and transient if $p_{\text {esc }}>0$.

Figure 36: $\boldsymbol{\Omega}$
We will call the problem of determining recurrence or transience of a random walk the type problem.
### 2.1.3 Polya's original question
The definition of recurrence that we have given differs from Polya's original definition. Polya defined a walk to be recurrent if, with probability one, it will pass through every single point of the lattice in the course of its wanderings. In our definition, we require only that the point return to its starting point. So we have to ask ourselves, "Can the random walk be recurrent in our sense and fail to be recurrent in Polya's sense?"
The answer to this question is, "No, the two definitions of recurrence are equivalent." Why? Because if the point must return once to its starting point, then it must return there again and again, and each time it starts away from the origin, it has a certain non-zero probability of hitting a specified target vertex before returning to the origin. And anyone can get a bull's-eye if he or she is allowed an infinite number of darts, so eventually the point will hit the target vertex.
Exercise 2.1.1 Write out a rigorous version of the argument just given.
### 2.1.4 Polya's Theorem: recurrence in the plane, transience in space
In [26], Polya proved the following theorem:
Polya's Theorem. Simple random walk on a $d$-dimensional lattice is recurrent for $d=1,2$ and transient for $d>2$.
The rest of this work will be devoted to trying to understand this theorem. Our approach will be to exploit once more the connections between questions about a random walk on a graph and questions about electric currents in a corresponding network of resistors. We hope that this approach, by calling on methods that appeal to our physical intuition, will leave us feeling that we know "why" the theorem is true.
Exercise 2.1.2 Show that Polya's theorem implies that if two random walkers start at $\mathbf{0}$ and wander independently, then in one and two dimensions they will eventually meet again, but in three dimensions there is positive probability that they won't.
Exercise 2.1.3 Show that Polya's theorem implies that a random walker in three dimensions will eventually hit the line defined by $x=2, z=0$.
### 2.1.5 The escape probability as a limit of escape probabilities for finite graphs
We can determine the type of an infinite lattice from properties of bigger and bigger finite graphs that sit inside it. The simplest way to go about this is to look at the lattice analog of balls (solid spheres) in space. These are defined as follows: Let $r$ be an integer - this will be the radius of the ball. Let $G^{(r)}$ be the graph gotten from $\mathbf{Z}^{d}$ by throwing out vertices whose distance from the origin is $>r$. By "distance from the origin" we mean here not the usual Euclidean distance, but the distance "in the lattice"; that is, the length of the shortest path along the edges of the lattice between the two points. Let $\partial G^{(r)}$ be the "sphere" of radius $r$ about the origin, i.e., those points that are exactly $r$ units from the origin. In two dimensions, $\partial G^{(r)}$ looks like a square. (See Figure 37.) In three dimensions, it looks like an octahedron.

Figure 37:
We define a random walk on $G^{(r)}$ as follows: The walk starts at $\mathbf{0}$ and continues as it would on $\mathbf{Z}^{d}$ until it reaches a point on $\partial G^{(r)}$
and then it stays at this point. Thus the walk on $G^{(r)}$ is an absorbing Markov chain with every point of $\partial G^{(r)}$ an absorbing state.
Let $p_{\text {esc }}^{(r)}$ be the probability that a random walk on $G^{(r)}$ starting at $\mathbf{0}$, reaches $\partial G^{(r)}$ before returning to 0 . Then $p_{\text {esc }}^{(r)}$ decreases as $r$ increases and $p_{\text {esc }}=\lim _{r \rightarrow \infty} p_{\text {eSC }}^{(r)}$ is the escape probability for the infinite graph. If this limit is 0 , the infinite walk is recurrent. If it is greater than 0 , the walk is transient.
### 2.1.6 Electrical formulation of the type problem
Now that we have expressed things in terms of finite graphs, we can make use of electrical methods. To determine $p_{\text {esc }}$ electrically, we simply ground all the points of $\partial G^{(r)}$, maintain $\mathbf{0}$ at one volt, and measure the current $i^{(r)}$ flowing into the circuit. (See Figure 38.)

Figure 38:
From Section 1.3.4, we have
$$
p_{\mathrm{esc}}^{(r)}=\frac{i^{(r)}}{2 d}
$$
where $d$ is the dimension of the lattice. (Remember that we have to divide by the number of branches coming out of the starting point.) Since the voltage being applied is $1, i^{(r)}$ is just the effective conductance between $\mathbf{0}$ and $\partial G^{(r)}$, i.e.,
$$
i^{(r)}=\frac{1}{R_{\mathrm{eff}}^{(r)}}
$$
where $R_{\mathrm{eff}}^{(r)}$ is the effective resistance from $\mathbf{0}$ to $\partial G^{(r)}$. Thus
$$
p_{\mathrm{eSc}}^{(r)}=\frac{1}{2 d R_{\mathrm{eff}}^{(r)}}
$$
Define $R_{\text {eff }}$, the effective resistance from the origin to infinity, to be
$$
R_{\mathrm{eff}}=\lim _{r \rightarrow \infty} R_{\mathrm{eff}}^{(r)}
$$
This limit exists since $R_{\mathrm{eff}}^{(r)}$ is an increasing function of $r$. Then
$$
p_{\mathrm{esc}}=\frac{1}{2 d R_{\mathrm{eff}}}
$$
Of course $R_{\text {eff }}$ may be infinite; in fact, this will be the case if and only if $p_{\mathrm{esc}}=0$. Thus the walk is recurrent if and only if the resistance to infinity is infinite, which makes sense.
The success of this electrical formulation of the type problem will come from the fact that the resistance to infinity can be estimated using classical methods of electrical theory.
### 2.1.7 One Dimension is easy, but what about higher dimensions?
We now know that simple random walk on a graph is recurrent if and only if a corresponding network of 1-ohm resistors has infinite resistance "out to infinity". Since an infinite line of resistors obviously has infinite resistance, it follows that simple random walk on the 1-dimensional lattice is recurrent, as stated by Polya's theorem.
What happens in higher dimensions? We are asked to decide whether a $d$-dimensional lattice has infinite resistance to infinity. The difficulty is that the $d$-dimensional lattice $\mathbf{Z}^{d}$ lacks the rotational symmetry of the Euclidean space $\mathbf{R}^{d}$ in which it sits.

Random walk recurrence?

Resistance to infinity infinite?
Figure 39:
Figure 40:
To see how this lack of symmetry complicates electrical problems, we determine, by solving the appropriate discrete Dirichlet problem, the voltages for a one-volt battery attached between $\mathbf{0}$ and the points of $\partial G^{(3)}$ in $\mathbf{Z}^{2}$. The resulting voltages are:
| | | | 0 | | | |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| | 0 | .091 | 0 | | | |
| 0 | .091 | .364 | .364 | .182 | 0 | |
| | 0 | .182 | .364 | .182 | .091 | 0. |
| | | 0 | .091 | 0 | | |
| | | | 0 | | | |
The voltages at points of $\partial G^{(1)}$ are equal, but the voltages at points of $\partial G^{(2)}$ are not. This means that the resistance from $\mathbf{0}$ to $\partial G^{(3)}$ cannot be written simply as the sum of the resistances from $\mathbf{0}$ to $\partial G^{(1)}, \partial G^{(1)}$ to $\partial G^{(2)}$, and $\partial G^{(2)}$ to $\partial G^{(3)}$. This is in marked contrast to the case of a continuous resistive medium to be discussed in Section 2.1.8.
Exercise 2.1.4 Using the voltages given for $G^{(3)}$, find $R_{\mathrm{eff}}^{(3)}$ and $p_{\mathrm{eSc}}^{(3)}$.
Exercise 2.1.5 Consider a one-dimensional infinite network with resistors $R_{n}=1 / 2^{n}$ from $n$ to $n+1$ for $n=\ldots,-2,-1,0,1,2, \ldots$ Describe the associated random walk and determine whether the walk is recurrent or transient.
Exercise 2.1.6 A random walk moves on the non-negative integers; when it is in state $n, n>0$, it moves with probability $p_{n}$ to $n+1$ and with probability $1-p_{n}$, to $n-1$. When at $\mathbf{0}$, it moves to 1 . Determine a network that gives this random walk and give a criterion in terms of the $p_{n}$ for recurrence of the random walk.
### 2.1.8 Getting around the lack of rotational symmetry of the lattice
Suppose we replace our $d$-dimensional resistor lattice by a (homogeneous, isotropic) resistive medium filling all of $\mathbf{R}^{d}$ and ask for the effective resistance to infinity. Naturally we expect that the rotational symmetry will make this continuous problem easier to solve than the original discrete problem. If we took this problem to a physicist, he or she would probably produce something like the scribblings illustrated
in Figure 41, and conclude that the effective resistance is infinite for $d=1,2$ and finite for $d>2$. The analogy to Polya's theorem is obvious, but is it possible to translate these calculations for continuous media into information about what happens in the lattice?

Figure 41:
This can indeed be done, and this would certainly be the most natural approach to take. We will come back to this approach at the end of the work. For now, we will take a different approach to getting around the asymmetry of the lattice. Our method will be to modify the lattice in such a way as to obtain a graph that is symmetrical enough so that we can calculate its resistance out to infinity. Of course, we will have to think carefully about what happens to that resistance when we make these modifications.
### 2.1.9 Rayleigh: shorting shows recurrence in the plane, cutting shows transience in space
Here is a sketch of the method we will use to prove Polya's theorem.
To take care of the case $d=2$, we will modify the 2 -dimensional resistor network by shorting certain sets of nodes together so as to get a new network whose resistance is readily seen to be infinite. As shorting
can only decrease the effective resistance of the network, the resistance of the original network must also be infinite. Thus the walk is recurrent when $d=2$.
To take care of the case $d=3$, we will modify the 3 -dimensional network by cutting out certain of the resistors so as to get a new network whose resistance is readily seen to be finite. As cutting can only increase the resistance of the network, the resistance of the original network must also be finite. Thus the walk is transient when $d=3$.
The method of applying shorting and cutting to get lower and upper bounds for the resistance of a resistive medium was introduced by Lord Rayleigh. (Rayleigh [29]; see also Maxwell 21], Jeans [11, Polya and Szego [28]). We will refer to Rayleigh's techniques collectively as Rayleigh's short-cut method. This does not do Rayleigh justice, for Rayleigh's method is a whole bag of tricks that goes beyond mere shorting and cutting - but who can resist a pun?
Rayleigh's method was apparently first applied to random walks by C. St. J. A. Nash-Williams [24], who used the shorting method to establish recurrence for random walks on the 2-dimensional lattice.
### 2.2 Rayleigh's short-cut method
### 2.2.1 Shorting and cutting
In its simplest form, Rayleigh's method involves modifying the network whose resistance we are interested in so as to get a simpler network. We consider two kinds of modifications, shorting and cutting. Cutting involves nothing more than clipping some of the branches of the network, or what is the same, simply deleting them from the network. Shorting involves connecting a given set of nodes together with perfectly conducting wires, so that current can pass freely between them. In the resulting network, the nodes that were shorted together behave as if they were a single node.
### 2.2.2 The Shorting Law and the Cutting Law; Rayleigh's idea
The usefulness of these two procedures (shorting and cutting) stems from the following observations:
Shorting Law. Shorting certain sets of nodes together can only decrease the effective resistance of the network between two given nodes.

Cut

Short
Figure 42:
Cutting Law. Cutting certain branches can only increase the effective resistance between two given nodes.
These laws are both equivalent to Rayleigh's Monotonicity Law, which was introduced in Section 1.4.1 (see Exercise 2.2.1):
Monotonicity Law. The effective resistance between two given nodes is monotonic in the branch resistances.
Rayleigh's idea was to use the Shorting Law and the Cutting Law above to get lower and upper bounds for the resistance of a network. In Section 2.2 .3 we apply this method to solve the recurrence problem for simple random walk in dimensions 2 and 3.
Exercise 2.2.1 Show that the Shorting Law and the Cutting Law are both equivalent to the Monotonicity Law.
### 2.2.3 The plane is easy
When $d=2$, we apply the Shorting Law as follows: Short together nodes on squares about the origin, as shown in Figure 43. The network we obtain is equivalent to the network shown in Figure 44.
Now as $n$ 1-ohm resistors in parallel are equivalent to a single resistor of resistance $\frac{1}{n}$ ohms, the modified network is equivalent to the network shown in Figure 45. The resistance of this network out to infinity is

Figure 43: $\boldsymbol{\AA}$

Figure 44:

Figure 45:
$$
\sum_{n=0}^{\infty} \frac{1}{8 n+4}=\infty
$$
As the resistance of the old network can only be bigger, we conclude that it too must be infinite, so that the walk is recurrent when $d=2$.
Exercise 2.2.2 Using the shorting technique, give an upper bound for $p_{\mathrm{esc}}^{(3)}$, and compare this with the exact value obtained in Exercise 2.1.4.
### 2.2.4 Space: searching for a residual network
When $d=3$, what we want to do is delete certain of the branches of the network so as to leave behind a residual network having manifestly finite resistance. The problem is to reconcile the "manifestly" with the "finite". We want to cut out enough edges so that the effective resistance of what is left is easy to calculate, while leaving behind enough edges so that the result of the calculation is finite.
### 2.2.5 Trees are easy to analyze
Trees - that is, graphs without circuits - are undoubtedly the easiest to work with. For instance, consider the full binary tree, shown in Figure 46. Notice that sitting inside this tree just above the root are two copies of the tree itself. This self-similarity property can be used to compute the effective resistance $R_{\infty}$ from the root out to infinity. (See Exercise 2.2.3.) It turns out that $R_{\infty}=1$. We will demonstrate this below by a more direct method.
To begin with, let us determine the effective resistance $R_{n}$ between the root and the set of $n$th generation branch points. To do this, we should ground the set of branch points, hook the root up to a 1-volt

I lee lull binary dre
Figure 46:

Root
Figure 47:
battery, and compute
$$
R_{n}=\frac{1}{\text { current through battery }}
$$
For $n=3$, the circuit that we would obtain is shown in Figure 48 .

1 volt
Figure 48:
In the resulting circuit, all branch points of the same generation are at the same voltage (by symmetry). Nothing happens when you short together nodes that are at the same potential. Thus shorting together branch points of the same generation will not affect the distribution of currents in the branches. In particular, this modification will not affect the current through the battery, and we conclude that
$$
R_{n}=\frac{1}{\text { current in original circuit }}=\frac{1}{\text { current in modified circuit }}
$$
For $n=3$, the modified circuit is shown in Figure 49. This picture shows that
$$
R_{3}=\frac{1}{2}+\frac{1}{4}+\frac{1}{8}=1-\frac{1}{2^{3}}
$$
More generally,
$$
R_{n}=\frac{1}{2}+\frac{1}{4}+\ldots+\frac{1}{2^{n}}=1-\frac{1}{2^{n}}
$$

Figure 49:
Letting $n \rightarrow \infty$, we get
$$
R_{\infty}=\lim _{n \rightarrow \infty} R_{n}=\lim _{n \rightarrow \infty} 1-\frac{1}{2^{n}}=1
$$
Figure 50 shows another closely related tree, the tree homogeneous of degree three: Note that all nodes of this tree are similar - there is no intrinsic way to distinguish one from another. This tree is obviously a close relative of the full binary tree. Its resistance to infinity is $2 / 3$.
Exercise 2.2.3 (a) Show, using the self-similarity of the full binary tree, that the resistance $R_{\infty}$ to infinity satisfies the equation
$$
R_{\infty}=\frac{R_{\infty}+1}{2}
$$
and conclude that either $R_{\infty}=1$ or $R_{\infty}=\infty$.
(b) Again using the self-similarity of the tree, show that
$$
R_{n+1}=\frac{R_{n}+1}{2}
$$
where $R_{n}$ denotes the resistance out to the set of the $n$ th-generation branch points. Conclude that
$$
R_{\infty}=\lim _{n \rightarrow \infty} R_{n}=1
$$

Figure 50:
### 2.2.6 The full binary tree is too big
Nothing could be nicer than the two trees we have just described. They are the prototypes of networks having manifestly finite resistance to infinity. Unfortunately, we can't even come close to finding either of these trees as a subgraph of the three-dimensional lattice. For in these trees, the number of nodes in a "ball" of radius $r$ grows exponentially with $r$, whereas in a $d$-dimensional lattice, it grows like $r$, i.e., much slower. (See Figure 51.) There is simply no room for these trees in any finite-dimensional lattice.
### 2.2.7 $\mathrm{NT}_{3}$ : a "three-dimensional" tree
These observations suggest that we would do well to look for a nice tree $\mathrm{NT}_{3}$ where the number of nodes within a radius $r$ of the root is on the order of $r^{3}$. For we might hope to find something resembling $\mathrm{NT}_{3}$ in the 3-dimensional lattice, and if there is any justice in the world, this tree would have finite resistance to infinity, and we would be done.
Before introducing $\mathrm{NT}_{3}$, let's have a look at $\mathrm{NT}_{2}$, our choice for the tree most likely to succeed in the 2-dimensional lattice (see Figure 52). The idea behind $\mathrm{NT}_{2}$ is that, since a ball of radius $r$ in the graph ought to contain something like $r^{2}$ points, a sphere of radius $r$ ought
Full binary tree
1

3

7

$2^{r+1}-1$
Tree homogeneous of degree 3

22
$3 \cdot 2^{r}-2$
Figure 51:

etc.
Figure 52: $\boldsymbol{2}$
to contain something like $r$ points, so the number of points in a sphere should roughly double when the radius of the sphere is doubled. For this reason, we make the branches of our tree split in two every time the distance from the origin is (roughly) doubled.
Similarly, in a 3-dimensional tree, when we double the radius, the size of a sphere should roughly quadruple. Thus in $\mathrm{NT}_{3}$, we make the branches of our tree split in four where the branches of $\mathrm{NT}_{2}$ would have split in two. $\mathrm{NT}_{3}$ is shown in Figure 53. Obviously, $\mathrm{NT}_{3}$ is none too

Figure 53:
happy about being drawn in the plane.
### 2.2.8 $\quad \mathrm{NT}_{3}$ has finite resistance
To see if we're on the right track, let's work out the resistance of our new trees. These calculations are shown in Figures 54 and 55 .
As we would hope, the resistance of $\mathrm{NT}_{2}$ is infinite, but the resistance of $\mathrm{NT}_{3}$ is not.
Exercise 2.2.4 Use self-similarity arguments to compute the resistance of $\mathrm{NT}_{2}$ and $\mathrm{NT}_{3}$.
### 2.2.9 But does $\mathrm{NT}_{3}$ fit in the three-dimensional lattice?
We would like to embed $\mathrm{NT}_{3}$ in $\mathbf{Z}^{3}$. We start by trying to embed $\mathrm{NT}_{2}$ in $\mathbf{Z}^{2}$. The result is shown in Figure 56 .
To construct this picture, we start from the origin and draw 2 rays, one going north, one going east. Whenever a ray intersects the line $x+y=2^{n}-1$ for some $n$, it splits into 2 rays, one going north, and one
going east. The sequence of pictures in Figure 57 shows successively larger portions of the graph, along with the corresponding portions of $\mathrm{NT}_{2}$.
Of course this isn't really an embedding, since certain pairs of points that were distinct in $\mathrm{NT}_{2}$ get identified, that is, they are made to correspond to a single point in the picture. In terms of our description, sometimes a ray going north and a ray going east pass through each other. This could have been avoided by allowing the rays to "bounce" instead of passing through each other, at the expense of embedding not $\mathrm{NT}_{2}$ but a close relative - see Exercise 2.2.7. However, because the points of each identified pair are at the same distance from the root of $\mathrm{NT}_{2}$, when we put a battery between the root and the $n$th level they will be at the same potential. Hence, the current flow is not affected by these identifications, so the identifications have no effect on $R_{\text {eff }}$. For our purposes, then, we have done just as well as if we had actually embedded $\mathrm{NT}_{2}$.
To construct the analogous picture in three dimensions, we start three rays off from the origin going north, east, and up. Whenever a ray intersects the plane $x+y+z=2^{n}-1$ for some $n$, it splits into three rays, going north, east, and up. This process is illustrated in Figure 58. Surprisingly, the subgraph of the 3-dimensional lattice obtained in this way is not $\mathrm{NT}_{3}$ ! Rather, it represents an attempt to embed the tree shown in Figure 59. We call this tree $\mathrm{NT}_{2.5849 \ldots}$.. because it is $2.5849 \ldots$. .dimensional in the sense that when you double the radius

Figure 54:

Figure 55:

Figure 56:


Figure 57: $\boldsymbol{\Omega}$


Figure 59: \&
of a ball, the number of points in the ball gets multiplied roughly by 6 and
$$
6=2^{\log _{2} 6}=2^{2.5849 \ldots}
$$
Again, certain pairs of points of $\mathrm{NT}_{2.5849 . .}$ have been allowed to correspond to the same point in the lattice, but once again the intersections have no effect on $R_{\text {eff. }}$.
So we haven't come up with our embedded $\mathrm{NT}_{3}$ yet. But why bother? The resistance of $\mathrm{NT}_{2.5849 . . .}$ out to infinity is
$$
\frac{1}{3}+\frac{2}{9}+\frac{4}{27}+\ldots=\frac{1}{3}\left(1+\frac{2}{3}+\left(\frac{2}{3}\right)^{2}+\ldots\right)=\frac{1}{3} \frac{1}{1-\frac{2}{3}}=1
$$
Thus we have found an infinite subgraph of the 3-dimensional lattice having finite resistance out to infinity, and we are done.
Exercise 2.2.5 This exercise deals with the escape probability $p_{\text {esc }}$ for simple random walk in 3 dimensions. The idea is to turn upper and lower bounds for the resistance of the lattice into bounds for $p$ esc. Bounds are the best we can ask for using our method. The determination of the exact value will be discussed in Section 2.3.5. It is roughly 66 .
(a) Use a shorting argument to get an upper bound for $p_{\text {esc }}$.
(b) We have seen that the resistance of the 3-dimensional lattice is at most one ohm. Show that the corresponding lower bound for $p_{\text {esc }}$ is $1 / 6$ Show that this lower bound can be increased to $1 / 3$ with no extra effort.
Exercise 2.2.6 Prove that simple random walk in any dimension $d>$ 3 is transient.
Exercise 2.2.7 Show how the not-quite embeddings of $\mathrm{NT}_{2}$ and $\mathrm{NT}_{2.5849 \ldots}$ can be altered to yield honest-to-goodness embeddings of "stretchedout" versions of these trees, obtained by replacing each edge of the tree by three edges in series. (Hint: "bounce".)
### 2.2.10 What we have done; what we will do
We have finally finished our electrical proof of Polya's theorem. The proof of recurrence for $d=1,2$ was straight-forward, but this could hardly be said of the proof for $d=3$. After all, we started out trying
to embed $\mathrm{NT}_{3}$ and ended up by not quite embedding something that was not quite $\mathrm{NT}_{3}$ !
This is not bad in itself, for one frequently sets out to do something and in the process of trying to do it gets a better idea. The real problem is that this explicit construction is just too clever, too artificial. We seem to be saying that a simple random walk in 3 dimensions is transient because it happens to contain a beautifully symmetrical subgraph that is in some sense $2.5849 \ldots$-dimensional! Fine, but what if we hadn't stumbled upon this subgraph? Isn't there some other, more natural way?
We will see that indeed there are more natural approaches to showing transience for $d=3$. One such approach uses the same idea of embedding trees, but depends on the observation that one doesn't need to be too careful about sending edges to edges. Another approach, based on relating the lattice not to a tree but to Euclidean space, was already hinted at in Section 2.1.8. The main goal for the rest of this work will be to explore these more natural electrical approaches to Polya's theorem.
Before jumping into this, however, we are going to go back and take a look at a classical-i.e., probabilistic - approach to Polya's theorem. This will give us something to compare our electrical proofs with.
### 2.3 The classical proofs of Polya's Theorem
### 2.3.1 Recurrence is equivalent to an infinite expected number of returns
For the time being, all of our random walks will be simple. Let $u$ be the probability that a random walker, starting at the origin, will return to the origin. The probability that the walker will be there exactly $k$ times (counting the initial time) is $u^{k}(1-u)$. Thus, if $m$ is the expected number of times at the origin,
$$
m=\sum_{k=1} \infty k u^{k-1}(1-u)=\frac{1}{1-u}
$$
If $m=\infty$ then $u=1$, and hence the walk is recurrent. If $m<\infty$ then $u<1$, so the walk is transient. Thus $m$ determines the type of the walk.
We shall use an alternate expression for $m$. Let $u_{n}$ be the probability that the walk, starting at $\mathbf{0}$, is at $\mathbf{0}$ on the nth step. Since the walker
starts at $\mathbf{0}, u_{0}=1$. Let $e_{n}$ be a random variable that takes on the value 1 if, at time $n$ the walker is at 0 and 0 otherwise. Then
$$
T=\sum_{n=0}^{\infty} e_{n}
$$
is the total number of times at $\mathbf{0}$ and
$$
m=\mathbf{E}(T)=\sum_{n=0}^{\infty} \mathbf{E}\left(e_{n}\right)
$$
But $\mathbf{E}\left(e_{n}\right)=1 \cdot u_{n}+0 \cdot\left(1-u_{n}\right)=u_{n}$. Thus
$$
m=\sum_{n=0}^{\infty} u_{n}
$$
Therefore, the walk will be recurrent if the series $\sum_{n=0}^{\infty}$ diverges and transient if it converges.
Exercise 2.3.1 Let $N_{\mathbf{x y}}$ be the expected number of visits to $\mathbf{y}$ for a random walker starting in $\mathbf{x}$. Show that $N_{\mathrm{xy}}$ is finite if and only if the walk is transient.
### 2.3.2 Simple random walk in one dimension
Consider a random walker in one dimension, started at $\mathbf{0}$. To return to $\mathbf{0}$, the walker must take the same number of steps to the right as to the left; hence, only even times are possible. Let us compute $u_{2 n}$. Any path that returns in $2 n$ steps has probability $1 / 2^{n}$. The number of possible paths equals the number of ways that we can choose the $n$ times to go right from the $2 n$ possible times. Thus
$$
u_{2 n}=\binom{2 n}{n} \frac{1}{2^{2 n}}
$$
We shall show that $\sum_{n} u_{2 n}=\infty$ by using Stirling's approximation:
$$
n!\sim \sqrt{2 \pi n} e^{-n} n^{n}
$$
Thus
$$
u_{2 n}=\frac{(2 n!)}{n!n!} \frac{1}{2^{2 n}} \sim \frac{\sqrt{2 \pi \cdot 2 n} e^{-2 n}(2 n)^{2 n}}{\left(\sqrt{2 \pi n} e^{-n} n^{n}\right)^{2} 2^{2 n}}=\frac{1}{\sqrt{\pi n}}
$$
Therefore,
$$
\sum_{n} u_{2 n} \sim \sum_{n} \frac{1}{\sqrt{\pi n}}=\infty
$$
and a simple random walk in one dimension is recurrent.
Recall that this case was trivial by the resistor argument.
Exercise 2.3.2 We showed in Section 1.1.5 that a random walker starting at $x$ with $01 / 2$. Let $h_{x}$ be the probability, starting at $x$, that the walker ever reaches 0 . Use Exercise [.1.9 to show that $h_{x}=(q / p)^{x}$ for $x \geq 0$ and $h_{x}=1$ for $x<0$. Show that this walk is transient.
Exercise 2.3.4 For a simple random walk in one dimension, it follows from Exercise 1.1 .7 that the expected time, for a walker starting at $x$ with $00$ with $R_{x y}=\frac{1}{C_{x y}}$. The graph $G$ together with the conductances $\mathbf{C}=\left(C_{x y}\right)$ is called a network and denoted by $(G, \mathbf{C})$. Given a network $(G, \mathbf{C})$, we define a random walk by
$$
P_{x y}=\frac{C_{x y}}{C_{x}}
$$
where $C_{x}=\sum_{y} C_{x y}$. When all the conductances are equal, we obtain a random walk that moves along each edge with the same probability: In agreement with our previous terminology, we call this walk simple random walk on $G$.
We have now a quite general class of infinite-state Markov chains. As in the case of finite networks, the chains are reversible Markov chains: That is, there is a positive vector $\mathbf{w}$ such that $w_{x} P_{x y}=w_{y} P_{y x}$. As in the finite case, we can take $w_{x}=C_{x}$, since $C_{x} P_{x y}=C_{x y}=C_{y x}=$ $C_{y} P_{y x}$.
### 2.4.2 The type problem
Let $(G, \mathbf{C})$ be an infinite network with random walk $\mathbf{P}$. Let $\mathbf{0}$ be a reference point. Let $p_{\text {esc }}$ be the probability that a walk starting at $\mathbf{0}$ will never return to 0 . If $p_{\text {esc }}=0$ we say that $\mathbf{P}$ is recurrent, and if $p_{\text {esc }}>0$ we say that it is transient. You are asked to show in Exercise 2.4.1 that the question of recurrence or transience of $\mathbf{P}$ does not depend upon the choice of the reference point. The type problem is the problem of determining if a random walk (network) is recurrent or transient.
In Section 2.1.5 we showed how to rephrase the type problem for a lattice in terms of finite graphs sitting inside it. In Section 2.1.6 we showed that the type problem is equivalent to an electrical network problem by showing that simple random walk on a lattice is recurrent if and only if the lattice has infinite resistance to infinity. The same arguments apply with only minor modifications to the more general infinite networks as well. This means that we can use Rayleigh's shortcut method to determine the type of these more general networks.
Exercise 2.4.1 Show that the question of recurrence or transience of $\mathbf{P}$ does not depend upon the choice of the reference point.
### 2.4.3 Comparing two networks
Given two sets of conductances $\mathbf{C}$ and $\overline{\mathbf{C}}$ on $G$, we say that $(G, \overline{\mathbf{C}})<$ $(G, \mathbf{C})$ if $\bar{C}_{x y}R_{x y}$ for all $x y$. Assume that $(G, \overline{\mathbf{C}})<(G, \mathbf{C})$. Then by the Monotonicity Law,
$\bar{R}_{\text {eff }} \geq R_{\text {eff. }}$. Thus if random walk on $(G, \overline{\mathbf{C}})$ is transient, i.e., if $\bar{R}_{\text {eff }}<$ $\infty$, then random walk on $(G, \mathbf{C})$ is also transient. If random walk on $(G, \mathbf{C})$ is recurrent, i.e., if $R_{\text {eff }}=\infty$, then random walk on $(G, \overline{\mathbf{C}})$ is also recurrent.
Theorem. If $(G, \mathbf{C})$ and $(G, \overline{\mathbf{C}})$ are networks, and if there exist constants $u, v$ with $0E$. We know that $\mathbf{P}$ is reversible with $\mathbf{w} \mathbf{P}=\mathbf{w}$, where $w_{x}$ is the number of edges coming out of $x$. From its construction, $\overline{\mathbf{P}}$ is also reversible and $\mathbf{w} \overline{\mathbf{P}}=\mathbf{w} . \overline{\mathbf{P}}$ is the random walk on a network $\left(G_{k}, \overline{\mathbf{C}}\right)$ with $\bar{C}_{x y}=$ $w_{x} \bar{P}_{x y}$. If $\bar{P}_{x y}>0$, there is a path $x, x_{1}, x_{2}, \ldots, x_{m-1}, y$ in $G$ from $x$ to $y$ of length $m \leq k$. Then
$$
\bar{P}_{x y} \geq \frac{1}{k}\left(\frac{1}{E}\right)^{m} \geq \frac{1}{k}\left(\frac{1}{E}\right)^{k}
$$
Thus
$$
0<\frac{1}{k}\left(\frac{1}{E}\right)^{k} \leq \bar{P}_{x y} \leq 1
$$
and
$$
0<\frac{1}{k}\left(\frac{1}{E}\right)^{k} \leq \bar{C}_{x y} \leq E
$$
Therefore, by the theorem on the irrelevance of bounded twiddling proven in Section 2.4.3, $\overline{\mathbf{P}}$ and simple random walk on $G_{k}$ are of the same type. So $G$ and $G_{k}$ are of the same type.
NOTE: This is the only place where we use probabilistic methods of proof. For the purist who wishes to avoid probabilistic methods, Exercise 2.4.9 indicates an alternative electrical proof.
We show how this theorem can be used. We say that a graph $G$ can be embedded in a graph $\bar{G}$ if the points $x$ of $G$ can be made to correspond in a one-to-one fashion to points $\bar{x}$ of $\bar{G}$ in such a way that if $x y$ is an edge in $G$, then $\bar{x} \bar{y}$ is an edge in $\bar{G}$.
Theorem. If simple random walk on $G$ is transient, and if $\bar{G}$ can be embedded in a $k$-fuzz $\bar{G}_{k}$ of $\bar{G}$ then simple random walk on $\bar{G}$ is also transient. Simple random walk on $G$ and $\bar{G}$ are of the same type if each graph can be embedded in a $k$-fuzz of the other graph.
Proof. Assume that simple random walk on $G$ is transient and that $G$ can be embedded in a $k$-fuzz $\bar{G}_{k}$ of $\bar{G}$. Since $R_{\text {eff }}$ for $G$ is finite and $G$ can be embedded in $\bar{G}_{k}, R_{\text {eff }}$ for $\bar{G}_{k}$ is finite. By our previous theorem, the same is true for $\bar{G}$ and simple random walk on $\bar{G}$ is transient.
If we can embed $G$ in $\bar{G}_{k}$ and $\bar{G}$ in $G_{k}$, then the random walk on $G$ is transient if and only if the random walk on $\bar{G}$ is. $\diamond$
Exercise 2.4.4 We have assumed that there is a bound $E$ for the number of edges coming out of any point. Show that if we do not assume this, it is not necessarily true that $G$ and $G_{k}$ are of the same type. (Hint: Consider a network something like that shown in Figure 62.)

Figure 62:
### 2.4.5 Comparing general graphs with lattice graphs
We know the type of simple random walk on a lattice $\mathbf{Z}^{d}$. Thus to determine the type of simple random walk on an arbitrary graph $G$, it is natural to try to compare $\mathrm{G}$ with $\mathbf{Z}^{d}$. This is feasible for graphs that can be drawn in some Euclidean space $\mathbf{R}^{d}$ in a civilized manner.
Definition. A graph $G$ can be drawn in a Euclidean space $\mathbf{R}^{d}$ in a civilized manner if its vertices can be embedded in $\mathbf{R}^{d}$ so that for some $r<\infty, s>0$
(a) The length of each edge is $\leq r$.
(b) The distance between any two points is $>s$.
Note that we make no requirement about being able to draw the edges of $G$ so they don't intersect.
Theorem. If a graph can be drawn in $\mathbf{R}^{d}$ in a civilized manner, then it can be embedded in a $k$-fuzz of the lattice $\mathbf{Z}^{d}$.
Proof. We carry out the proof for the case $d=2$. Assume that $G$ can be drawn in a civilized manner in $\mathbf{R}^{2}$. We want to show that $G$ can be embedded in a $k$-fuzz of $\mathbf{Z}^{2}$. We have been thinking of $\mathbf{Z}^{2}$ as
being drawn in $\mathbf{R}^{2}$ with perpendicular lines and adjacent points a unit distance apart on these lines, but this embedding is only one particular way of representing $\mathbf{Z}^{2}$. To emphasize this, let's talk about $L^{2}$ instead of $\mathbf{Z}^{2}$. Figure 63 shows another way of drawing $L^{2}$ in $\mathbf{R}^{2}$. From a

Figure 63:
graph-theoretical point of view, this is the same as $\mathbf{Z}^{2}$. In trying to compare $G$ to $L$, we take advantage of this flexibility by drawing $L^{2}$ so small that points of $G$ can be moved onto points of $L^{2}$ without bumping into each other.
Specifically, let $L^{2}$ be a two-dimensional rectangular lattice with lines a distance $s / 2$ apart. In any square of $L^{2}$, there is at most one point of $G$. Move each point $x$ of $G$ to the southwest corner $\bar{x}$ of the square that it is in, as illustrated in Figure 64.
Now since any two adjacent points $x, y$ in $G$ were within $r$ of each other in $\mathbf{R}^{2}$, the corresponding points $\bar{x}, \bar{y}$ in $L^{2}$ will have Euclidean distance $Задачу представляли И.Богданов, В.Бугаенко, К.Куюмжиян, К.Кохась, А.Скопенков,
Г.Челноков
## 1 Неравенство Шапиро
В октябре 1954 г. в журнале "American Mathematical Monthly" появилась задача американского математика Гарольда Шапиро:
Для положительных чисел $x_{1}, x_{2}, \ldots, x_{n}$ докажите неравенство
$$
\frac{x_{1}}{x_{2}+x_{3}}+\frac{x_{2}}{x_{3}+x_{4}}+\ldots+\frac{x_{n-1}}{x_{n}+x_{1}}+\frac{x_{n}}{x_{1}+x_{2}} \geqslant \frac{n}{2}
$$
причем равенство может достигаться, только если все знаменатели равны между собой.
B "Monthly" в отличие, например, от журнала "Квант" допускалась публикация задач, которые никто не умел решать, причем читателей об этом не предупреждали. Так было и на этот раз. У автора было решение только для $n=3$ и 4
В предлагаемых ниже задачах вместо положительности всех чисел $x_{k}$ можно требовать, чтобы все числа $x_{k}$ были неотрицательными, а все знаменатели - ненулевыми. Если неравенство для положительных чисел уже доказано, то из него несложно вывести неравенство для неотрицательных чисел, для которых знаменатели не обращаются в нуль. Обозначим
$$
f\left(x_{1}, x_{2}, \ldots, x_{n}\right)=\frac{x_{1}}{x_{2}+x_{3}}+\frac{x_{2}}{x_{3}+x_{4}}+\ldots+\frac{x_{n-1}}{x_{n}+x_{1}}+\frac{x_{n}}{x_{1}+x_{2}}
$$
1.1. Докажите неравенство (1) при $n=3,4,5,6$.
1.2. Докажите, что неравенство (1) неверно:
а) при $n=20$;
b) при $n=14$;
c) при $n=25$.
1.3. Докажите неравенство (1) для монотонных последовательностей.
1.4. Докажите, что если неравенство (1) неверно при $n=m$, то оно неверно и при $n=m+2$.
1.5. Докажите, что если неравенство (1) неверно при $n=m$, где $m$ нечетно, то оно неверно и при всех $n$, бо́льших $m$.
1.6. Докажите, что неравенство (1) верно при $n=8,10,12$ и $n=7,9,11,13,15,17,19,21,23$. Как следует из утверждения задачи 1.4, достаточно доказать неравенство лишь при $n=12$ и $n=23$.
1.7. Докажите, что $f\left(x_{1}, x_{2}, \ldots, x_{n}\right)+f\left(x_{n}, x_{n-1}, \ldots, x_{1}\right) \geqslant n$.
1.8. Предположим, что в точке $a_{1}, a_{2}, \ldots, a_{n}>0$ функция $f\left(x_{1}, x_{2}, \ldots, x_{n}\right)$ имеет локальный минимум.
a) Если $n$ четно, докажите, что $f\left(a_{1}, a_{2}, \ldots, a_{n}\right)=n / 2$.
b*) Докажите аналогичное утверждение для нечетных $n$.
c) Докажите с помощью пунктов а) и б) неравенство для $n=7$ и $n=8$.
1.9. Докажите неравенство $f\left(x_{1}, x_{2}, \ldots, x_{n}\right) \geqslant c n$ для следующих значений $c$
a) $c=1 / 4$
b) $c=(\sqrt{2}-1)$;
c) $c=5 / 12$.
## 2 Полезные и родственные неравенства
Докажите следующие неравенства в предположении, что все числа $x_{k}$ положительны. Проверьте, что выделенные жирным шрифтом константы нельзя заменить на бо́льшие (при каждом $n$ ).
2.1. Неравенство Морделла.
а) Для любых неотрицательных чисел $x_{1}, x_{2}, \ldots, x_{n}$ имеет место неравенство
$$
\left(\sum_{k=1}^{n} x_{k}\right)^{2} \geqslant \min \left\{\frac{\mathbf{n}}{\mathbf{2}}, \mathbf{3}\right\} \cdot \sum_{k=1}^{n} x_{k}\left(x_{k+1}+x_{k+2}\right)
$$
b) Установите, для каких неотрицательных чисел $x_{1}, x_{2}, \ldots, x_{n}$ неравенство Морделла обращается в равенство.
2.2. Для всех неотрицательных чисел $x_{1}, x_{2}, \ldots, x_{n}$ докажите неравенство
$$
\left(\sum_{k=1}^{n} x_{k}\right)^{2} \geqslant \min \left\{\frac{\mathbf{n}}{\mathbf{3}}, \frac{\mathbf{8}}{\mathbf{3}}\right\} \cdot \sum_{k=1}^{n} x_{k}\left(x_{k+1}+x_{k+2}+x_{k+3}\right)
$$
2.3. а) При $n \leqslant 8$ докажите неравенство
$$
\frac{x_{1}}{x_{2}+x_{3}+x_{4}}+\frac{x_{2}}{x_{3}+x_{4}+x_{5}}+\ldots+\frac{x_{n-1}}{x_{n}+x_{1}+x_{2}}+\frac{x_{n}}{x_{1}+x_{2}+x_{3}} \geqslant \frac{n}{3}
$$
$\left.\mathrm{b}^{*}\right)$ Верно ли это неравенство еще при каких-нибудь натуральных $n$ ?
2.4. $\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2} \geqslant 4\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n-1} x_{n}+x_{n} x_{1}\right) ; \quad n \geqslant 4$.
2.5. $\sum_{k=1}^{n} \frac{x_{k}}{x_{k+1}+x_{k+2}} \geqslant \sum_{k=1}^{n} \frac{x_{k+1}}{x_{k}+x_{k+1}}$.
2.6. $\frac{x_{1}}{x_{n}+x_{2}}+\frac{x_{2}}{x_{1}+x_{3}}+\ldots+\frac{x_{n-1}}{x_{n-2}+x_{n}}+\frac{x_{n}}{x_{n-1}+x_{1}} \geqslant \mathbf{2} ; \quad n \geqslant 4$.
2.7. $\frac{x_{1}+x_{2}}{x_{1}+x_{3}}+\frac{x_{2}+x_{3}}{x_{2}+x_{4}}+\ldots+\frac{x_{n-1}+x_{n}}{x_{n-1}+x_{1}}+\frac{x_{n}+x_{1}}{x_{n}+x_{2}} \geqslant \mathbf{4} ; \quad n \geqslant 4$.
2.8. $\frac{x_{1}}{x_{n}+x_{3}}+\frac{x_{2}}{x_{1}+x_{4}}+\ldots+\frac{x_{n-1}}{x_{n-2}+x_{1}}+\frac{x_{n}}{x_{n-1}+x_{2}} \geqslant \mathbf{3} ; \quad n \geqslant 6$.
2.9. $\frac{x_{2}+x_{3}}{x_{1}+x_{4}}+\frac{x_{3}+x_{4}}{x_{2}+x_{5}}+\ldots+\frac{x_{n}+x_{1}}{x_{n-1}+x_{2}}+\frac{x_{1}+x_{2}}{x_{n}+x_{3}} \geqslant \mathbf{6} ; \quad n \geqslant 6$.
2.10. $\frac{x_{1}+x_{2}}{x_{1}+x_{4}}+\frac{x_{2}+x_{3}}{x_{2}+x_{5}}+\ldots+\frac{x_{2004}+x_{1}}{x_{2004}+x_{3}} \geqslant \mathbf{6}$.
2.11. $\frac{x_{1}}{x_{n}+x_{4}}+\frac{x_{2}}{x_{1}+x_{5}}+\ldots+\frac{x_{n-1}}{x_{n-2}+x_{2}}+\frac{x_{n}}{x_{n-1}+x_{3}} \geqslant 4$, где $n-$ четное число, большее 7 .
2.12. $\sum_{k=1}^{n} \frac{x_{k}^{2}}{x_{k+1}^{2}-x_{k+1} x_{k+2}+x_{k+2}^{2}} \geqslant\left[\frac{\mathbf{n}+\mathbf{1}}{\mathbf{2}}\right]$.
## Неравенство Шапиро
## 3 Добавление после промежуточного финиша
1.10. а) Для любого натурального $n$ существует такое число $q_{n}>1$, что при всех вещественных числах $x_{1}$, $x_{2}, \ldots, x_{n} \in\left[\frac{1}{q_{n}} ; q_{n}\right]$ имеет место неравенство (1).
$\mathrm{b}^{*}$ ) Существует ли такое $q>1$, что при всех натуральных $n$ и при всех $x_{i} \in\left[\frac{1}{q} ; q\right]$ выполнено неравенство (1)?
1.11. Пусть $S=f\left(x_{1}, x_{2}, \ldots, x_{n}\right)$ - левая часть неравенства Шапиро. Обозначим через $a_{1}, a_{2}, \ldots, a_{n}$ числа $x_{2} / x_{1}, x_{3} / x_{2}, \ldots, x_{n} / x_{n-1}, x_{1} / x_{n}$, расположенные в порядке возрастания.
a) Докажите, что $S \geqslant \frac{1}{a_{1}\left(1+a_{n}\right)}+\frac{1}{a_{2}\left(1+a_{n-1}\right)}+\ldots+\frac{1}{a_{n}\left(1+a_{1}\right)}$;
б) Пусть $b_{k}= \begin{cases}\frac{1}{a_{k} a_{n+1-k}}, & a_{k} a_{n+1-k} \geqslant 1 \\ \frac{2}{a_{k} a_{n+1-k}+\sqrt{a_{k} a_{n+1-k}}}, & a_{k} a_{n+1-k}<1 .\end{cases}$
Докажите, что $2 S \geqslant b_{1}+b_{2}+\ldots+b_{n} ;$
в) Пусть $g$ - наибольшая выпуклая функция, не превосходящая функций $e^{-x}$ и $2\left(e^{x}+e^{x / 2}\right)^{-1}$. Докажите, что $2 S \geqslant g\left(\ln \left(a_{1} a_{n}\right)\right)+g\left(\ln \left(a_{2} a_{n-1}\right)\right)+\ldots+g\left(\ln \left(a_{n} a_{1}\right)\right) \geqslant n g(0)$.
г) Докажите, что для любого $\lambda>g(0) / 2$ существует такое натуральное число $n$ и такие положительные числа $x_{1}, x_{2}, \ldots, x_{n}$, что $S \leqslant \lambda n$.
## Решения
1.1. $\underline{n=3}$. Пусть $S=x_{1}+x_{2}+x_{3}$. Как нетрудно видеть, функция $f(t)=\frac{t}{S-t}$ выпукла на промежутке $[0 ; S)$. Запишем для нее неравенство Йенсена
$$
\frac{f\left(x_{1}\right)+f\left(x_{2}\right)+f\left(x_{3}\right)}{3} \geqslant f\left(\frac{x_{1}+x_{2}+x_{3}}{3}\right)=f\left(\frac{S}{3}\right)=\frac{1}{2}
$$
Это и есть требуемое неравенство.
$\underline{n=4}$. Неравенство циклическое. Напишем данные числа последовательно в вершинах квадрата. По диагонали проведем стрелочки от меньшего числа к большему. Тогда у одной из сторон квадрата обе вершины концы стрелочек. Перенумеруем числа так, чтобы это была сторона $x_{4} x_{1}$. Таким образом, можно считать, что $x_{1} \geqslant x_{3}, x_{4} \geqslant x_{2}$. Для переменных, упорядоченных этим способом, имеет место очевидное неравенство
$$
\frac{x_{1}}{x_{2}+x_{3}}+\frac{x_{3}}{x_{4}+x_{1}} \geqslant \frac{x_{1}}{x_{4}+x_{3}}+\frac{x_{3}}{x_{2}+x_{1}}
$$
Воспользуемся им для доказательства неравенства Шапиро
$\frac{x_{1}}{x_{2}+x_{3}}+\frac{x_{2}}{x_{3}+x_{4}}+\frac{x_{3}}{x_{4}+x_{1}}+\frac{x_{4}}{x_{1}+x_{2}} \geqslant \frac{x_{1}}{x_{4}+x_{3}}+\frac{x_{2}}{x_{3}+x_{4}}+\frac{x_{3}}{x_{2}+x_{1}}+\frac{x_{4}}{x_{1}+x_{2}}=\frac{x_{1}+x_{2}}{x_{3}+x_{4}}+\frac{x_{3}+x_{4}}{x_{1}+x_{2}}=a+a^{-1} \geqslant 2$.
$\underline{n=5}$. Заметим, что функция $f(t)=1 /(S-t)$ выпукла на интервале $[0 ; S)$. Воспользуемся тогда неравенством Йенсена для пяти чисел
$$
a_{1} f\left(t_{1}\right)+a_{2} f\left(t_{2}\right)+a_{3} f\left(t_{3}\right)+a_{4} f\left(t_{4}\right)+a_{5} f\left(t_{5}\right) \geqslant f\left(a_{1} t_{1}+a_{2} t_{2}+a_{3} t_{3}+a_{4} t_{4}+a_{5} t_{5}\right)
$$
где $a_{i} \geqslant 0, \sum a_{i}=1$. В качестве $a_{i}$ возьмем $a_{i}=\frac{x_{i}}{S}$, и пусть $t_{i}=x_{i}+x_{i-1}+x_{i-2}, i=1, \ldots, 5$ (считаем, что нумерация переменных циклическая: $x_{0}=x_{5}, x_{-1}=x_{4}$ ). Тогда $f\left(t_{i}\right)=\frac{1}{S-t_{i}}=\frac{1}{x_{i+1}+x_{i+2}}$, и значит, левая часть неравенства (2) есть в точности левая часть неравенства Шапиро. Посмотрим, что представляет собой правая часть.
$$
\frac{1}{S-\sum_{i=1}^{5} a_{i} t_{i}}=\frac{1}{S-\sum_{i=1}^{5} \frac{x_{i}}{S}\left(x_{i}+x_{i-1}+x_{i-2}\right)}=\frac{S}{S^{2}-\sum_{i=1}^{5} x_{i}\left(x_{i}+x_{i-1}+x_{i-2}\right)}
$$
Как нетрудно убедиться, раскрыв скобки, знаменатель представляет собой в точности сумму попарных произведений набора переменных $x_{i}$. Заметим еще, что силу однородности доказываемого неравенства, можно считать, что $S=x_{1}+x_{2}+x_{3}+x_{4}+x_{5}=1$. Итак, правая часть неравенства (2), есть величина, обратная сумме попарных произведений переменных $x_{i}$, подчиненных условию $x_{1}+x_{2}+x_{3}+x_{4}+x_{5}=1$. Очевидно, минимум правой части достигается в том случае, когда сумма попарных произведений максимальна. Хорошо известно, что это будет в случае, когда все переменные равны. Нетрудно видеть, что правая часть в этом случае равна $5 / 2$.
Кстати, при $n=4$ аналогичное доказательство тоже работает.
$\underline{n=6}$. Аналогично предыдущему решению. Функция $f(t)=1 /(S-t)$ выпукла на интервале $[0 ; S)$ Рассмотрим неравенство Йенсена для шести чисел
$$
\sum_{i=1}^{6} a_{i} f\left(t_{i}\right) \geqslant f\left(\sum_{i=1}^{6} a_{i} t_{i}\right)
$$
Пусть $a_{i}=\frac{x_{i}}{S}, t_{i}=x_{i}+x_{i-1}+x_{i-2}+x_{i-3}, i=1, \ldots, 6$ (считаем, что нумерация переменных циклическая: $x_{0}=x_{6}, x_{-1}=x_{5}, x_{-2}=x_{4}$ ). Тогда $f\left(t_{i}\right)=\frac{1}{S-t_{i}}=\frac{1}{x_{i+1}+x_{i+2}}$, и значит, левая часть неравенства (1.1) есть в точности левая часть неравенства Шапиро. Посмотрим, что представляет собой правая часть.
$$
\frac{1}{S-\sum_{i=1}^{6} a_{i} t_{i}}=\frac{1}{S-\sum_{i=1}^{6} \frac{x_{i}}{S}\left(x_{i}+x_{i-1}+x_{i-2}+x_{i-3}\right)}=\frac{S}{S^{2}-\sum_{i=1}^{6} x_{i}\left(x_{i}+x_{i-1}+x_{i-2}+x_{i-3}\right)}
$$
Как нетрудно убедиться, раскрыв скобки, знаменатель представляет собой в точности сумму попарных произведений набора переменных $x_{i}$, кроме произведений $x_{1} x_{4}, x_{2} x_{5}, x_{3} x_{6}$. Эту сумму можно записать в виде $\left(x_{1}+x_{4}\right)\left(x_{2}+x_{5}\right)+\left(x_{1}+x_{4}\right)\left(x_{3}+x_{6}\right)+\left(x_{2}+x_{5}\right)\left(x_{3}+x_{6}\right)$. Обозначив $A=x_{1}+x_{4}, B=x_{2}+x_{5}, C=x_{3}+x_{6}$, мы можем записать правую часть нашего неравенства Йенсена в виде
$$
\frac{A+B+C}{A B+B C+A C}
$$
В силу однородности доказываемого неравенства, можно считать, что $S=x_{1}+x_{2}+x_{3}+x_{4}+x_{5}=A+B+C=1$. Тогда очевидно, что выражение (3) не меньше 3 , в силу неравенства $(A+B+C)^{2} \geqslant 3(A B+B C+A C)$. чтд
Замечание. K сожалению, дальше этот способ не работает. Если мы применяем метод множителей Лагранжа, чтобы найти максимум знаменателей в правой части (при фиксированной сумме переменных), то система уравнений получается линейной, следовательно, подозрительная точка единственна. При $n=6$ квадратичная форма второго дифференциала в этой точке является положительно определенной и вырожденной. При $n=7$ она уже не является знакоопределенной (Я проверил это в Марle. KK). Таким образом, при $n=7$ неравенство Йенсена дает слишком грубую оценку снизу.
Другое решение этой задачи получится, если применить неравенство Коши-Буняковского для наборов чисел
$$
\begin{array}{rlll}
\sqrt{\frac{x_{1}}{x_{2}+x_{3}}}, & \sqrt{\frac{x_{2}}{x_{3}+x_{4}}}, \ldots, & \sqrt{\frac{x_{n}}{x_{1}+x_{2}}} \\
\sqrt{x_{1}\left(x_{2}+x_{3}\right)}, & \sqrt{x_{2}\left(x_{3}+x_{4}\right)}, & \ldots, & \sqrt{x_{n}\left(x_{1}+x_{2}\right)}
\end{array}
$$
Получится
$$
\frac{x_{1}}{x_{2}+x_{3}}+\frac{x_{2}}{x_{3}+x_{4}}+\ldots+\frac{x_{n}}{x_{1}+x_{2}} \geqslant \frac{\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2}}{x_{1}\left(x_{2}+x_{3}\right)+x_{2}\left(x_{3}+x_{4}\right)+\ldots+x_{n}\left(x_{1}+x_{2}\right)}
$$
По неравенству Морделла (задача 2.1) при $n \leqslant 6$ правая часть этого неравенства не меньше, чем $n / 2$.
1.2. а) Если в качестве $x_{1}, x_{2}, \ldots, x_{20}$ взять
$$
\begin{aligned}
& 1+5 \varepsilon, \quad 6 \varepsilon, \quad 1+4 \varepsilon, \quad 5 \varepsilon, \quad 1+3 \varepsilon, \quad 4 \varepsilon, \quad 1+2 \varepsilon, \quad 3 \varepsilon, \quad 1+\varepsilon, \quad 2 \varepsilon \\
& 1+2 \varepsilon, \quad \varepsilon, \quad 1+3 \varepsilon, \quad 2 \varepsilon, \quad 1+4 \varepsilon, \quad 3 \varepsilon, \quad 1+5 \varepsilon, \quad 4 \varepsilon, \quad 1+6 \varepsilon, \quad 5 \varepsilon
\end{aligned}
$$
то левая часть неравенства будет меньше чем $10-\varepsilon^{2}+c \varepsilon^{3}$ для некоторого $c$. Следовательно, при достаточно малом $\varepsilon$ она будет меньше 10. Этот пример принадлежит Лайтхиллу; опубликован в "Monthly" [22].
b) Если в качестве $x_{1}, x_{2}, \ldots, x_{14}$ взять
$$
1+7 \varepsilon, 7 \varepsilon, 1+4 \varepsilon, 6 \varepsilon, 1+\varepsilon, 5 \varepsilon, 1,2 \varepsilon, 1+\varepsilon, 0,1+4 \varepsilon, \varepsilon, 1+6 \varepsilon, 4 \varepsilon
$$
то правая часть неравенства будет меньше чем $7-2 \varepsilon^{2}+c \varepsilon^{3}$ и, следовательно, при достаточно малом $\varepsilon$ она будет меньше 7. Это пример Цулауфа [27].
А вот еще пример [24], целочисленный, в нем некоторые переменные равны нулю.
$$
0,42,2,42,4,41,5,39,4,38,2,38,0,40
$$
c) Примеры, подтверждающие что при $n=25$ неравенство неверно, были построены на компьютере в 1970 г. Дейкиным [10] и Малкольмом [18]. Ниже приведен пример Дейкина (он в отличие от примера Малкольма целочисленный):
$0,85,0,101,0,120,14,129,41,116,59,93,64,71,63,52,60,36,58,23,58,12,62,3,71$.
А вот еще пример Р. Алексеева и Е. Фошкина (приведен в [3]).
$32,0,37,0,43,0,50,0,59,8,62,21,55,29,44,32,33,31,24,30,16,29,10,29,4$.
1.3. Утверждение задачи опубликовано в [13]. Мы приводим короткое изящное решение.
Пусть $x_{1} \geqslant x_{2} \geqslant \cdots \geqslant x_{n}>0$. Заметим, что произведение $n$ дробей $\frac{x_{k}+x_{k+1}}{x_{k+1}+x_{k+2}}$ равно 1 . Тогда из неравенства о средних заключаем, что
$$
\sum_{k=1}^{n} \frac{x_{k}+x_{k+1}}{x_{k+1}+x_{k+2}} \geqslant n=\sum_{k=1}^{n} \frac{x_{k+1}+x_{k+2}}{x_{k+1}+x_{k+2}}
$$
Следовательно,
$$
\sum_{k=1}^{n} \frac{x_{k}}{x_{k+1}+x_{k+2}} \geqslant \sum_{k=1}^{n} \frac{x_{k+2}}{x_{k+1}+x_{k+2}}=\sum_{k=1}^{n} \frac{x_{k+1}}{x_{k}+x_{k+1}}
$$
Теперь мы воспользуемся следующим известным неравенством, которое будем называть транснеравенством: пусть имеются два набора чисел $a_{1} \geqslant \ldots \geqslant a_{n}$ и $b_{1} \geqslant \ldots \geqslant b_{n}$ Тогда для любой перестановки $k_{1}, \ldots, k_{n}$ чисел $1, \ldots, n$ имеет место неравенство
$$
a_{1} b_{1}+a_{2} b_{2}+\ldots a_{n} b_{n} \geqslant a_{1} b_{k_{1}}+a_{2} b_{k_{2}}+\ldots a_{n} b_{k_{n}} \geqslant a_{1} b_{n}+a_{2} b_{n-1}+\ldots a_{n} b_{1}
$$
Два раза воспользуемся транснеравеством
$$
\begin{aligned}
\sum_{k=1}^{n} \frac{x_{k}}{x_{k+1}+x_{k+2}} & =\sum_{k=1}^{n-2} \frac{x_{k}}{x_{k+1}+x_{k+2}}+\frac{x_{n-1}}{x_{n}+x_{1}}+\frac{x_{n}}{x_{1}+x_{2}} \geqslant \\
& \geqslant \sum_{k=1}^{n-2} \frac{x_{k}}{x_{k+1}+x_{k+2}}+\frac{x_{n-1}}{x_{1}+x_{2}}+\frac{x_{n}}{x_{n}+x_{1}} \underset{(* *)}{\geqslant} \\
& \geqslant \sum_{k=1}^{n} \frac{x_{k}}{x_{k}+x_{k+1}}
\end{aligned}
$$
Здесь неравенство (*) - это транснеравенство для наборов из двух чисел: $x_{n-1} \geqslant x_{n}$ и $\frac{1}{x_{n}+x_{1}} \geqslant \frac{1}{x_{1}+x_{2}}$; а неравенство (**) - это транснеравенство для обратно упорядоченных наборов чисел $x_{1}, x_{2}, \ldots, x_{n-1}$ и $\frac{1}{x_{1}+x_{2}}, \frac{1}{x_{2}+x_{3}}, \ldots, \frac{1}{x_{n-1}+x_{n}}$.
Таким образом,
$$
2 \sum_{k=1}^{n} \frac{x_{k}}{x_{k+1}+x_{k+2}} \geqslant \sum_{k=1}^{n} \frac{x_{k}}{x_{k}+x_{k+1}}+\sum_{k=1}^{n} \frac{x_{k+1}}{x_{k}+x_{k+1}}=n
$$
Для убывающего набора чисел $x_{i}$ решение аналогично, поскольку применяя неравенство о средних мы вообще не пользовались упорядочением, а применяя транснеравенство, использовали то, что наборы $x_{i}$ и $\frac{1}{x_{i}+x_{i+1}}$ упорядочены по-разному.
А вот более прямолинейное решение из [3].
Разберем сначала случай $x_{1} \geqslant x_{2} \geqslant \cdots \geqslant x_{n}>0$. Для краткости обозначим левую часть неравенства через $f_{n}\left(x_{1}, x_{2}, \ldots, x_{n}\right)$. Заметим, что $f_{2}\left(x_{1}, x_{2}\right)=1$. Поэтому если мы покажем, что
$$
f_{n+1}\left(x_{1}, x_{2}, \ldots, x_{n}, x_{n+1}\right)-f_{n}\left(x_{1}, x_{2}, \ldots, x_{n}\right) \geqslant \frac{1}{2}
$$
то, воспользовавшись методом математической индукции, получим
$$
f_{n}\left(x_{1}, x_{2}, \ldots, x_{n}\right) \geqslant f_{n-1}\left(x_{1}, x_{2}, \ldots, x_{n-1}\right)+\frac{1}{2} \geqslant f_{n-2}\left(x_{1}, x_{2}, \ldots, x_{n-2}\right)+1 \geqslant \cdots \geqslant f_{2}\left(x_{1}, x_{2}\right)+\frac{n-2}{2}=\frac{n}{2}
$$
Рассмотрим разность
$$
f_{n+1}\left(x_{1}, x_{2}, \ldots, x_{n}, x_{n+1}\right)-f_{n}\left(x_{1}, x_{2}, \ldots, x_{n}\right)=\frac{x_{n-1}}{x_{n}+x_{n+1}}+\frac{x_{n}}{x_{n+1}+x_{1}}+\frac{x_{n+1}}{x_{1}+x_{2}}-\frac{x_{n-1}}{x_{n}+x_{1}}-\frac{x_{n}}{x_{1}+x_{2}}
$$
Она не зависит от чисел $x_{3}, x_{4}, \ldots, x_{n-2}$, поэтому мы можем считать их любыми числами из интервала $\left[x_{2} ; x_{n-1}\right]$. Слагаемые, содержащие $x_{n-1}$, имеют вид
$$
\frac{x_{n-1}}{x_{n}+x_{n+1}}-\frac{x_{n-1}}{x_{n}+x_{1}}=x_{n-1} \cdot \frac{x_{1}-x_{n+1}}{\left(x_{n}+x_{n+1}\right)\left(x_{n}+x_{1}\right)}
$$
т.е. равны $x_{n-1}$, умноженному на неотрицательное число. Следовательно, при уменьшении $x_{n-1}$ разность может только уменьшиться. Поэтому можно считать, что $x_{n-1}=x_{n}$. Слагаемые, содержащие $x_{2}$, дают
$$
\frac{x_{n+1}-x_{n}}{x_{1}+x_{2}}
$$
что уменьшается при уменьшении $x_{2}$, поэтому можно считать, что $x_{2}=x_{n-1}=x_{n}$. Следовательно, достаточно доказать неравенство
$$
\begin{aligned}
0 & \leqslant f_{n+1}\left(x_{1}, x_{n}, \ldots, x_{n}, x_{n}, x_{n+1}\right)-f_{n}\left(x_{1}, x_{n}, \ldots, x_{n}, x_{n}\right)-\frac{1}{2}= \\
& =\frac{x_{n+1}-2 x_{n}}{x_{1}+x_{n}}+\frac{x_{n}}{x_{n}+x_{n+1}}+\frac{x_{n}}{x_{1}+x_{n+1}}-\frac{1}{2}=\frac{\left(x_{n}-x_{n+1}\right)\left(x_{1}^{2}+2 x_{n}^{2}+x_{n} x_{n+1}-x_{1} x_{n}-x_{1} x_{n+1}-2 x_{n+1}^{2}\right)}{\left(x_{1}+x_{n}\right)\left(x_{1}+x_{n+1}\right)\left(x_{n}+x_{n+1}\right)}
\end{aligned}
$$
Таким образом, достаточно доказать, что
$$
x_{1}^{2}+2 x_{n}^{2}+x_{n} x_{n+1}-x_{1} x_{n}-x_{1} x_{n+1}-2 x_{n+1}^{2} \geqslant 0
$$
Последнее выражение уменьшается при увеличении $x_{n+1}$, поэтому можно считать, что $x_{n+1}=x_{n}$. Но в этом случае неравенство превращается в очевидное: $x_{1}^{2}+x_{n}^{2}-2 x_{1} x_{n} \geqslant 0$.
Теперь разберем случай $0\sqrt{2} \cdot \sqrt{\frac{x_{k}+x_{k+1}}{x_{k+1}+x_{k+2}}}
\end{aligned}
$$
Так как произведение $n$ чисел $\sqrt{\frac{x_{1}+x_{2}}{x_{2}+x_{3}}}, \sqrt{\frac{x_{2}+x_{3}}{x_{3}+x_{4}}}, \ldots, \sqrt{\frac{x_{n}+x_{1}}{x_{1}+x_{2}}}$ равно 1 , то из неравенства Коши следует, что их сумма не меньше $n$. Поэтому исходная сумма в левой части неравенства Шапиро не меньше, чем $\sqrt{2} n-n=$ $(\sqrt{2}-1) n$.
c) Аналогично предыдущему пункту каждую из дробей $\frac{x_{k}}{x_{k+1}+x_{k+2}}, k=1,2, \ldots, n$, запишем в виде
$$
\frac{x_{k}}{x_{k+1}+x_{k+2}}=\frac{x_{k}+\beta x_{k+1}}{x_{k+1}+x_{k+2}}+\alpha \cdot \frac{\beta x_{k+1}+x_{k+2}}{x_{k+1}+x_{k+2}}-\alpha
$$
а параметры $\alpha$ и $\beta$ подберем так, чтобы это равенство оказалось верным. Как нетрудно видеть, для этого нужно, чтобы $\beta+\alpha \beta=\alpha$, т. е. $\beta=\alpha /(\alpha+1)$. Тогда
$$
\begin{aligned}
\frac{x_{k}+\beta x_{k+1}}{x_{k+1}+x_{k+2}}+\alpha \cdot \frac{\beta x_{k}+x_{k+1}}{x_{k}+x_{k+1}} \geqslant 2 \sqrt{\alpha \frac{\left(x_{k}+\beta x_{k+1}\right)\left(\beta x_{k}+x_{k+1}\right)}{\left(x_{k}+x_{k+1}\right)\left(x_{k+1}+x_{k+2}\right)}}= \\
=2 \sqrt{\alpha \frac{\beta\left(x_{k}+x_{k+1}\right)^{2}+(\beta-1)^{2} x_{k} x_{k+1}}{\left(x_{k}+x_{k+1}\right)\left(x_{k+1}+x_{k+2}\right)}}>2 \sqrt{\alpha \beta \frac{x_{k}+x_{k+1}}{x_{k+1}+x_{k+2}}}=\frac{2 \alpha}{\sqrt{\alpha+1} \cdot \sqrt{\frac{x_{k}+x_{k+1}}{x_{k+1}+x_{k+2}}}}
\end{aligned}
$$
Значит,
$$
\begin{array}{r}
\frac{x_{1}}{x_{2}+x_{3}}+\frac{x_{2}}{x_{3}+x_{4}}+\ldots+\frac{x_{n-1}}{x_{n}+x_{1}}+\frac{x_{n}}{x_{1}+x_{2}} \geqslant \frac{2 \alpha}{\sqrt{\alpha+1}}\left(\sqrt{\frac{x_{1}+x_{2}}{x_{2}+x_{3}}}+\sqrt{\frac{x_{2}+x_{3}}{x_{3}+x_{4}}}+\ldots+\sqrt{\frac{x_{n}+x_{1}}{x_{1}+x_{2}}}\right)-\alpha n> \\
>\frac{2 \alpha}{\sqrt{\alpha+1}} n-\alpha n=\left(\frac{2 \alpha}{\sqrt{\alpha+1}}-\alpha\right) n
\end{array}
$$
Максимум выражения $g(\alpha)=\frac{2 \alpha}{\sqrt{\alpha+1}}-\alpha$ достигается при $\alpha=\alpha_{0} \approx 1.1479$ (корень кубического уравнения $\left.g^{\prime}(\alpha)=0\right)$, при этом $g\left(\alpha_{0}\right) \approx 0.4186$. При $\alpha=\frac{5}{4} \quad g(\alpha)=\frac{5}{12} \approx 0.416$ - неплохое приближение.
1.10. а) Утверждение принадлежит В. Кыртоаже [9].
Положим для краткости $y_{k}=x_{k}+x_{k+1}$. В новых обозначениях неравенство (1) примет вид
$$
\frac{x_{1}}{y_{2}}+\frac{x_{2}}{y_{3}}+\ldots+\frac{x_{n}}{y_{1}} \geqslant \frac{n}{2}
$$
Еще немного его преобразуем:
$$
\sum_{k=1}^{n} \frac{2 q_{n}^{2} x_{k}-y_{k+1}}{y_{k+1}} \geqslant n\left(q_{n}^{2}-1\right)
$$
здесь $q_{n}$ - это параметр, значение которого мы подберем позже так, чтобы все рассматриваемые неравенства оказались верными. Поскольку
$$
2 q_{n}^{2} x_{k}-y_{k+1}=\left(q_{n}^{2} x_{k}-x_{k+1}\right)+\left(q_{n}^{2} x_{k}-x_{k+2}\right) \geqslant 0
$$
по неравенству Коши-Буняковского для наборов чисел
$$
\left\{\sqrt{\frac{2 q_{n}^{2} x_{k}-y_{k+1}}{y_{k+1}}}\right\} \text { и }\left\{\sqrt{\left(2 q_{n}^{2} x_{k}-y_{k+1}\right) y_{k+1}}\right\}
$$
имеем
$$
\sum_{k=1}^{n} \frac{2 q_{n}^{2} x_{k}-y_{k+1}}{y_{k+1}} \geqslant \frac{\left(\sum_{k=1}^{n}\left(2 q_{n}^{2} x_{k}-y_{k+1}\right)\right)^{2}}{\sum_{k=1}^{n}\left(2 q_{n}^{2} x_{k}-y_{k+1}\right) y_{k+1}}
$$
Таким образом, достаточно показать, что
$$
A^{2}=\left(\sum_{k=1}^{n}\left(2 q_{n}^{2} x_{k}-y_{k+1}\right)\right)^{2} \geqslant n\left(q_{n}^{2}-1\right) \sum_{k=1}^{n}\left(2 q_{n}^{2} x_{k}-y_{k+1}\right) y_{k+1}=n\left(q_{n}^{2}-1\right) B
$$
Поскольку $\sum_{k=1}^{n} y_{k}=2 \sum_{k=1}^{n} x_{k}$, имеем равенства
$$
\begin{aligned}
& A=\left(q_{n}^{2}-1\right) \sum_{k=1}^{n} y_{k} \\
& B=2 q_{n}^{2} \sum_{k=1}^{n} x_{k} y_{k+1}-\sum_{k=1}^{n} y_{k}^{2}=2 q_{n}^{2} \sum_{k=1}^{n} y_{k} y_{k+1}-\left(q_{n}^{2}+1\right) \sum_{k=1}^{n} y_{k}^{2}
\end{aligned}
$$
Следовательно, осталось доказать, что
$$
\left(q_{n}^{2}-1\right)\left(\sum_{k=1}^{n} y_{k}\right)^{2} \geqslant n\left(2 q_{n}^{2} \sum_{k=1}^{n} y_{k} y_{k+1}-\left(q_{n}^{2}+1\right) \sum_{k=1}^{n} y_{k}^{2}\right)
$$
Преобразуем левую часть с помощью соотношения
$$
\left(\sum_{k=1}^{n} y_{k}\right)^{2}=n \sum_{k=1}^{n} y_{k}^{2}-\sum_{i1$.
Замечание. С ростом $n$ найденные $q_{n}$ стремятся к единице.
b)
1.11.
1.11. (а) Обозначим $k_{i}:=x_{i+1} / x_{i}$. Тогда
$$
S=\frac{1}{k_{1}\left(k_{2}+1\right)}+\frac{1}{k_{2}\left(k_{3}+1\right)}+\cdots+\frac{1}{k_{n}\left(k_{1}+1\right)} \geqslant \frac{1}{a_{1}\left(a_{n}+1\right)}+\frac{1}{a_{2}\left(a_{n-1}+1\right)}+\cdots+\frac{1}{a_{n}\left(a_{1}+1\right)}
$$
(b) Неравенство справедливо, поскольку
$$
\frac{1}{a_{i}\left(a_{n+1-i}+1\right)}+\frac{1}{a_{n+1-i}\left(a_{i}+1\right)}=\frac{1+\frac{a_{i} a_{n+1-i}-1}{\left(1+a_{i}\right)\left(1+a_{n+1-i}\right)}}{a_{i} a_{n+1-i}} \geqslant b_{i}
$$
где последнее неравенство справедливо, поскольку $\left(1+a_{i}\right)\left(1+a_{n+1-i}\right) \geqslant\left(1+\sqrt{a_{i} a_{n+1-i}}\right)^{2}$.
(с) Первое неравенство $2 S \geqslant g\left(\ln \left(a_{1} a_{n}\right)\right)+g\left(\ln \left(a_{2} a_{n-1}\right)\right)+\cdots+g\left(\ln \left(a_{n} a_{1}\right)\right)$ справедливо, поскольку $g(x)$ меньше и $e^{-x}$, и $2\left(e^{x}+e^{x / 2}\right)^{-1}$. Второе неравенство справедливо, по неравенству Йенсена, поскольку $g$ выпукла.
(d) $[2]$
2.1. Это неравенство доказано в статье [20].
a) При $n=3$ и $n=5$ после раскрытия скобок получим неравенство
$$
(n-1)\left(a_{1}+a_{2}+\ldots+a_{n}\right)^{2} \geqslant 2 n \sum_{i8$. Передвинув, если нужно, числа по циклу, можно добиться того, что $x_{4} \geqslant x_{1}, x_{4} \geqslant x_{2}$ и $x_{4} \geqslant x_{3}$, например, сделав $x_{4}$ наибольшим. При $r=1,2,3$ или 4 обозначим через $a_{r}$ сумму всех чисел $x_{k}$, для которых $k \equiv r(\bmod 4), k \leqslant n$. Тогда $x_{1}+x_{2}+\ldots+x_{n}=a_{1}+a_{2}+a_{3}+a_{4}$ и, следовательно, по неравенству (7)
Положим для краткости
$$
3\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2}=3\left(a_{1}+a_{2}+a_{3}+a_{4}\right)^{2} \geqslant 8\left(a_{1} a_{2}+a_{2} a_{3}+a_{3} a_{4}+a_{4} a_{1}\right) \geqslant 8 \cdot \sum_{(i-k) \div 4} x_{i} x_{k}
$$
$$
A=\sum_{(i-k) \nvdash 4} x_{i} x_{k} \quad \text { и } \quad B=\sum_{k=1}^{n} x_{k}\left(x_{k+1}+x_{k+2}+x_{k+3}\right)
$$
и проверим, что $A \geqslant B$. Действительно, при $n \equiv 0(\bmod 4)$ все слагаемые из суммы $B$ содержатся в сумме $A$; при $n \equiv 1(\bmod 4)$ в сумме $A$ по сравнению с суммой $B$ недостает лишь слагаемого $x_{n} x_{1}$, которое не превосходит содержащегося в ней слагаемого $x_{n} x_{4} ;$ при $n \equiv 2(\bmod 4)$ в сумме $A$ по сравнению с суммой $B$ недостает слагаемых $x_{n-1} x_{1}$ и $x_{n} x_{2}$, которые не превосходят соответственно слагаемых $x_{n-1} x_{4}$ и $x_{n} x_{4}$; и наконец, при $n \equiv 3(\bmod 4)$ в сумме $A$ по сравнению с суммой $B$ недостает слагаемых $x_{n-2} x_{1}, x_{n-1} x_{2}$ и $x_{n} x_{3}$, которые не превосходят соответственно слагаемых $x_{n-2} x_{4}, x_{n-1} x_{4}$ и $x_{n} x_{4}$.
Итак, во всех случаях $A \geqslant B$. Таким образом,
$$
3\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2} \geqslant 8 A \geqslant 8 B=8 \sum_{k=1}^{n} x_{k}\left(x_{k+1}+x_{k+2}+x_{k+3}\right)
$$
2.3. а) Это неравенство доказано в статье [11] с использованием преобразования Фурье. Мы приводим элементарное рассуждение. По неравенству Коши-Буняковского имеем
$$
\frac{x_{1}}{x_{2}+x_{3}+x_{4}}+\frac{x_{2}}{x_{3}+x_{4}+x_{5}}+\ldots+\frac{x_{n-1}}{x_{n}+x_{1}+x_{2}}+\frac{x_{n}}{x_{1}+x_{2}+x_{3}} \geqslant \frac{\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2}}{\sum x_{k}\left(x_{k+1}+x_{k+2}+x_{k+3}\right)} \geqslant \frac{n}{3}
$$
последнее по задаче 2.2 .
b)
2.4. Задача Б. Гинзбурга [1, задача 187], предлагалась на Всесоюзной олимпиаде по математике 1972 г. С помощью циклической перестановки чисел можно добиться того, что $x_{1} \leqslant x_{2}$. Пусть $S=x_{1}+x_{2}+\ldots+x_{n}$, $S_{1}=x_{1}+x_{3}+\ldots, S_{2}=x_{2}+x_{4}+\ldots$ Тогда $S_{1}^{2}+S_{2}^{2} \geqslant\left(S_{1}+S_{2}\right)^{2} / 2=S^{2} / 2$, откуда
$$
\frac{S^{2}}{2} \geqslant S^{2}-S_{1}^{2}-S_{2}^{2}=2 \sum_{(i-k) ; 2} x_{i} x_{k}
$$
Если $n$ - четно, то в последней сумме содержатся все слагаемые вида $x_{k} x_{k+1}$, а если нечетно, то отсутствует слагаемое $x_{n} x_{1}$, зато вместо него имеется большее слагаемое $x_{n} x_{2}$. Таким образом,
$$
\frac{S^{2}}{2} \geqslant 2\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n} x_{1}\right)
$$
2.5. См. решение задачи 1.3 до неравенства (4) (к этому моменту в упомянутом решении упорядочение переменных еще не использовалось).
2.6. Это задача А. Прокопьева, Турнир городов, 1981-82, [4], также опубликована в журнале "Квант", 1982, №6, задача M749.
Обозначим левую часть неравенства через $L_{n}$. При $n=4$
$$
L_{4}=\frac{x_{1}+x_{3}}{x_{2}+x_{4}}+\frac{x_{2}+x_{4}}{x_{1}+x_{3}}=a+a^{-1} \geqslant 2
$$
При $n>4$ рассуждаем по индукции. Так как неравенство циклическое, можно считать, что $x_{n+1}$ - наименьшее из всех чисел. Тогда отбросим в сумме $L_{n+1}$ последнее слагаемое, а потом уменьшим два других, получим
$$
L_{n+1} \geqslant \frac{x_{1}}{x_{n+1}+x_{2}}+\ldots+\frac{x_{n}}{x_{n-1}+x_{n+1}} \geqslant \frac{x_{1}}{x_{n}+x_{2}}+\ldots+\frac{x_{n}}{x_{n-1}+x_{n}}=L_{n}
$$
Чтобы показать, что константа в правой части точная, возьмем
$$
x_{1}=x_{2}=1, \quad x_{3}=t, \quad x_{4}=t^{2}, \quad \ldots, \quad x_{n}=t^{n-2}
$$
При $t \rightarrow+0$ первые два слагаемых стремятся к 1 , остальные - к 0 .
Используя неравенство Коши-Буняковского подобно тому как это делается в решении следующей задачи, читатель без труда придумает другое доказательство, сводящее данное неравенство к неравенству из задачи 2.4 .
2.7. Мы почерпнули условие этой задачи в статье [10]. Положим для краткости $S=x_{1}+x_{2}+\ldots+x_{n}$. Воспользуемся неравенством Коши-Буняковского для наборов чисел $\left\{\frac{x_{k}+x_{k+1}}{x_{k}+x_{k+2}}\right\}$ и $\left\{\left(x_{k}+x_{k+1}\right)\left(x_{k}+x_{k+2}\right)\right\}$,
получим
$$
\frac{x_{1}+x_{2}}{x_{1}+x_{3}}+\frac{x_{2}+x_{3}}{x_{2}+x_{4}}+\ldots+\frac{x_{n-1}+x_{n}}{x_{n-1}+x_{1}}+\frac{x_{n}+x_{1}}{x_{n}+x_{2}} \geqslant \frac{4\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2}}{\sum_{k=1}^{n}\left(x_{k}+x_{k+1}\right)\left(x_{k}+x_{k+2}\right)}
$$
Таким образом, достаточно установить неравенство
$$
S^{2} \geqslant \sum_{k=1}^{n}\left(x_{k}+x_{k+1}\right)\left(x_{k}+x_{k+2}\right)=\sum_{k=1}^{n} x_{k}^{2}+2 \sum_{k=1}^{n} x_{k} x_{k+1}+\sum_{k=1}^{n} x_{k} x_{k+2}
$$
которое получается с помощью раскрытия скобок в левой части, ибо при $n \geqslant 4$ слагаемые $x_{k} x_{k+1}$ и $x_{k} x_{k+2}$ при $k=1,2, \ldots, n$ различны.
Покажем, что константу 4 нельзя увеличить. Положим $x_{k}=a^{k-1}$ при $k=1,2, \ldots, n-1$ и $x_{n}=a^{n-2}$. При $a \rightarrow \infty$ первые $n-3$ слагаемых стремятся к нулю, а оставшиеся к 1,2 и 1.
2.8. Мы почерпнули условие этой задачи в статье [6]. Воспользуемся неравенством Коши-Буняковского для наборов чисел $\left\{\frac{x_{k}}{x_{k-1}+x_{k+2}}\right\}$ и $\left\{x_{k}\left(x_{k-1}+x_{k+2}\right)\right\}$, получим
$$
\frac{x_{1}}{x_{n}+x_{3}}+\frac{x_{2}}{x_{1}+x_{4}}+\ldots+\frac{x_{n-1}}{x_{n-2}+x_{1}}+\frac{x_{n}}{x_{n-1}+x_{2}} \geqslant \frac{\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2}}{\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n} x_{1}\right)+\left(x_{1} x_{3}+x_{2} x_{4}+\ldots+x_{n} x_{2}\right)}
$$
Правая часть полученного неравенства не меньше 3 в силу неравенства Морделла (задача 2.1).
Покажем, что константу 3 нельзя увеличить. Возьмем набор чисел $x_{k}=a^{k-1}$ при $k=1,2, \ldots, n-2$ и $x_{n-1}=x_{n}=1$. При $a \rightarrow 0$ первое и два последних слагаемых стремятся к единице, а остальные к нулю.
2.9. Мы почерпнули условие этой задачи в статье [5]. Сложим два неравенства из 2.8 (для прямого и обратного порядков чисел).
Покажем, что константу 6 нельзя увеличить. Возьмем набор чисел $x_{k}=a^{k-1}$ при $k=1,2, \ldots, n-2$ и $x_{n-1}=x_{n}=1$. При $a \rightarrow 0$ последние четыре слагаемых стремятся соответственно к $1,2,2,1$; остальные слагаемые стремятся к нулю.
2.10. В [19] это утверждение для произвольного $n$ высказано в качестве гипотезы.
Авторы следующего доказательства - участники конференции P.Milošević и M. Bukić.
Доказываемое неравенство есть сумма двух неравенств при $n=2004$ - неравенства из задачи 2.8 и неравенства
$$
\frac{x_{1}}{x_{1}+x_{4}}+\frac{x_{2}}{x_{2}+x_{5}}+\ldots+\frac{x_{n}}{x_{n}+x_{3}} \geqslant 3
$$
Докажем последнее неравенство. При $n=3 m$ оно является суммой трех неравенств:
$$
\begin{aligned}
& \frac{x_{1}}{x_{1}+x_{4}}+\frac{x_{4}}{x_{4}+x_{7}}+\ldots+\frac{x_{n-2}}{x_{n-2}+x_{1}} \geqslant 1 \\
& \frac{x_{2}}{x_{2}+x_{5}}+\frac{x_{5}}{x_{5}+x_{8}}+\ldots+\frac{x_{n-1}}{x_{n-1}+x_{2}} \geqslant 1 \\
& \frac{x_{3}}{x_{3}+x_{6}}+\frac{x_{6}}{x_{6}+x_{9}}+\ldots+\frac{x_{n}}{x_{n}+x_{3}} \geqslant 1
\end{aligned}
$$
Каждое из этих неравенств записывается в виде
$$
\frac{1}{1+a_{1}}+\frac{1}{1+a_{3}}+\ldots+\frac{1}{1+a_{m}} \geqslant 1 \quad \text { где } a_{1} a_{2} \ldots a_{m}=1
$$
Это неравенство доказывается по индукции. База $m=2$
$$
\frac{1}{1+a_{1}}+\frac{1}{1+\frac{1}{a_{1}}}=1 \geqslant 1
$$
Для обоснования перехода проверяем что
$$
\frac{1}{1+b}+\frac{1}{1+c} \geqslant \frac{1}{1+b c}
$$
Это делается непосредственно с помощью домножения на знаменатели и раскрытия скобок.
Приводим доказательство А. Храброва. Будем доказывать неравенство
$$
Z=\frac{x_{1}+x_{2}}{x_{1}+x_{4}}+\frac{x_{2}+x_{3}}{x_{2}+x_{5}}+\ldots+\frac{x_{3 n}+x_{1}}{x_{3 n}+x_{3}} \geqslant 6
$$
Для удобства будем считать, что нумерация переменных циклическая: $x_{3 n+k}=x_{k}$. Положим для краткости $(r=0,1$ и 2$)$
$$
S_{r}=\sum_{k=1}^{n} x_{3 k+r}, \quad X_{r}=\sum_{k=1}^{n} \frac{x_{3 k+r}}{x_{3 k+r}+x_{3 k+3+r}} \quad \text { и } \quad Y_{r}=\sum_{k=1}^{n} \frac{x_{3 k+r+1}}{x_{3 k+r}+x_{3 k+3+r}}
$$
Докажем сначала, что $X_{r} \geqslant 1$. Чтобы не усложнять формулы, ограничимся случаем $r=0$. Заметим, что
$$
X_{0} S_{0}^{2} \geqslant X_{0}\left(\sum_{k=1}^{n} x_{3 k}^{2}+\sum_{k=1}^{n} x_{3 k} x_{3 k+3}\right)=X_{0}\left(\sum_{k=1}^{n} x_{3 k}\left(x_{3 k}+x_{3 k+3}\right)\right) \geqslant S_{0}^{2}
$$
последнее - по неравенству Коши-Буняковского. Поэтому, $X_{0} \geqslant 1$.
Далее проверим неравенство $Y_{r} \geqslant S_{r+1} / S_{r}$ (мы полагаем $S_{3}=S_{0}$ ). Опять рассмотрим лишь случай $r=0$.
$$
Y_{0} S_{0} S_{1} \geqslant Y_{0}\left(\sum_{k=1}^{n} x_{3 k} x_{3 k+1}+\sum_{k=1}^{n} x_{3 k+1} x_{3 k+3}\right)=Y_{0}\left(\sum_{k=1}^{n} x_{3 k+1}\left(x_{3 k}+x_{3 k+3}\right)\right) \geqslant S_{1}^{2}
$$
в последнем неравенстве мы снова применили неравенство Коши-Буняковского. Стало быть, $Y_{0} \geqslant S_{1} / S_{0}$.
Сложим все доказанные неравенства и воспользуемся неравенством о средних, получим
$$
Z=X_{0}+X_{1}+X_{2}+Y_{0}+Y_{1}+Y_{2} \geqslant 3+\frac{S_{1}}{S_{0}}+\frac{S_{2}}{S_{1}}+\frac{S_{0}}{S_{2}} \geqslant 6
$$
Покажем, что константу 6 нельзя увеличить. Возьмем набор чисел $x_{1}=x_{2}=x_{3}=1, x_{k}=a^{n-k+1}$ при $k=3,4, \ldots, n$. При $a \rightarrow 0$ первое и второе слагаемое стремятся к 2 , третье, а также последнее - к 1 , остальные слагаемые стремятся к нулю.
2.11. Доказательство А. Храброва. Положим для краткости $S=x_{1}+x_{2}+\ldots+x_{n}$ и $T=\sum_{(i-k) ; 2} x_{i} x_{k}$. По неравенству Коши-Буняковского для наборов чисел $\left\{\frac{x_{k}}{x_{k-1}+x_{k+3}}\right\}$ и $\left\{x_{k}\left(x_{k-1}+x_{k+3}\right)\right\}$ имеем
$$
\frac{x_{1}}{x_{n}+x_{4}}+\frac{x_{2}}{x_{1}+x_{5}}+\ldots+\frac{x_{n-1}}{x_{n-2}+x_{2}}+\frac{x_{n}}{x_{n-1}+x_{3}} \geqslant \frac{\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2}}{\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n} x_{1}\right)+\left(x_{1} x_{4}+x_{2} x_{5}+\ldots+x_{n} x_{3}\right)}
$$
Таким образом, достаточно показать, что
$$
S^{2} \geqslant 4\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n} x_{1}\right)+4\left(x_{1} x_{4}+x_{2} x_{5}+\ldots+x_{n} x_{3}\right)
$$
В решении задачи 2.4 мы установили неравенство $S^{2} \geqslant 4 T$, см. (8). Поэтому достаточно показать, что
$$
T \geqslant\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n} x_{1}\right)+\left(x_{1} x_{4}+x_{2} x_{5}+\ldots+x_{n} x_{3}\right)
$$
Поскольку $n$ четно, то все слагаемые из правой суммы содержатся и в левой сумме.
Покажем, что константу 4 нельзя увеличить. Возьмем набор чисел $x_{k}=a^{k-1}$ при $k=1,2, \ldots, n-3$ и $x_{n-2}=x_{n-1}=x_{n}=1$. При $a \rightarrow+0$ первое слагаемое и три последних стремятся к единице, а остальные к нулю.
2.12. Мы взяли эту задачу в статье [14].
Заметим, что $a^{2}-a b+b^{2} \leqslant \max \{a, b\}^{2}$.
Пусть $x_{i_{1}}$ - наибольшее из чисел $x_{1}, x_{2}, \ldots, x_{n} ; x_{i_{2}}$ - наибольшее из двух следующих за $x_{i_{1}}$ чисел; $x_{i_{3}}-$ наибольшее из двух следующих за $x_{i_{2}}$ чисел и т. д. Будем строить эту последовательность чисел до тех пор пока не дойдем до такого $k$, что наибольшее из двух следующих за $x_{i_{k}}$ чисел - это $x_{i_{1}}$.
Ясно, что $k \geqslant n / 2$ и поэтому $k \geqslant\left[\frac{n+1}{2}\right]$.
$$
\sum_{k=1}^{n} \frac{x_{k}^{2}}{x_{k+1}^{2}-x_{k+1} x_{k+2}+x_{k+2}^{2}} \geqslant \sum_{j=1}^{k} \frac{x_{i_{j}}^{2}}{x_{i_{j+1}}^{2}} \geqslant k
$$
последнее по неравенству о среднем арифметическом и среднем геометрическом.
Выражение $\left[\frac{n+1}{2}\right]$ в правой части нельзя увеличить, поскольку если положить $x_{k}=1$ при нечетных $k$ и $x_{k}=0$ при четных $k$, то левая часть будет в точности равна $\left[\frac{n+1}{2}\right]$.
2.13. а) Мы взяли эту задачу в статье [10].
Можно считать, что $x_{3}$ - наибольшее. Тогда первое слагаемое не меньше единицы. Кроме того заметим, что сумма двух соседних слагаемых тоже не меньше единицы:
$$
\frac{x_{k}+x_{k+2}}{x_{k}+x_{k+1}}+\frac{x_{k+1}+x_{k+3}}{x_{k+1}+x_{k+2}}=1+\frac{x_{k} x_{k+1}+x_{k+2}^{2}+x_{k} x_{k+3}+x_{k+1} x_{k+3}}{\left(x_{k}+x_{k+1}\right)\left(x_{k+1}+x_{k+2}\right)} \geqslant 1
$$
b) Мы взяли эту задачу в статье [17].
c) Случай $n \leqslant 4$ разобран в [10], случай $n=5-$ в [25].
d) Вот контрпримеры из [25]. $n=6: 381,0,334,29,340,49$;
$n=13: 41,0,28,0,19,4,17,10,18,18,20,29,18$.
2.14. a) ([11]) и b) ([12]) следуют из неравенства Коши-Бушяковского и задач 2.2 a) и b).
c) $[11]$
d, e, f)
2.15. Мы взяли эту задачу в статье [15].
Пусть $x_{i_{1}}$ - наибольшее из чисел $x_{1}, x_{2}, \ldots, x_{n} ; x_{i_{2}}$ - наибольшее из $m$ следующих за $x_{i_{1}}$ чисел; $x_{i_{3}}$ наибольшее из $m$ следующих за $x_{i_{2}}$ чисел и т. д. Будем строить эту последовательность чисел до тех пор пока не дойдем до такого $k$, что наибольшее из $m$ следующих за $x_{i_{k}}$ чисел - это $x_{i_{1}}$.
Ясно, что $k \geqslant n / m$ и поэтому $k \geqslant\left[\frac{n+m-1}{m}\right]$. Таким образом,
$$
\sum_{k=1}^{n} \frac{x_{k}}{x_{k+1}+x_{k+2}+\ldots+x_{k+m}} \geqslant \sum_{j=1}^{k} \frac{x_{i_{j}}}{m x_{i_{j+1}}} \geqslant \frac{k}{m}
$$
последнее - по неравенству о среднем арифметическом и среднем геометрическом.
2.16. а) Мы взяли эту задачу в статье [20].
При $n=m$ имеем очевидное равенство. При $n=m+1$ и $n=2 m+1$ это неравенство - частный случай неравенства (6). При $n=2 m$ неравенство можно переписать в виде
$$
\left(x_{1}-x_{m+1}\right)^{2}+\left(x_{2}-x_{m+2}\right)^{2}+\ldots+\left(x_{m}-x_{2 m}\right)^{2} \geqslant 0
$$
в котором оно очевидно.
Пусть теперь $n=m+2$. Положим для краткости $s=x_{1}+x_{2}+\ldots+x_{n}$. Нам нужно проверить неравенство
$$
\frac{n-2}{n} s^{2} \geqslant \sum_{k=1}^{n} x_{k}\left(x_{k+1}+x_{k+2}+\ldots+x_{k+m}\right)=\sum_{k=1}^{n} x_{k}\left(s-x_{k}-x_{k+n-1}\right)=s^{2}-\sum_{k=1}^{n} x_{k}\left(x_{k}+x_{k+n-1}\right)
$$
Или, что тоже самое,
$$
\frac{2 s^{2}}{n} \leqslant \sum_{k=1}^{n} x_{k}\left(x_{k}+x_{k+n-1}\right)=\frac{1}{2} \sum_{k=1}^{n}\left(x_{k}+x_{k+n-1}\right)^{2}
$$
Последнее неравенство является очевидным следствием неравенства Коши-Буняковского:
$$
n \sum_{k=1}^{n}\left(x_{k}+x_{k+n-1}\right)^{2} \geqslant\left(\sum_{k=1}^{n}\left(x_{k}+x_{k+n-1}\right)\right)^{2}
$$
Перейдем теперь к случаю $n=2 m+2$. При $1 \leqslant r \leqslant m+1$ обозначим через $a_{r}$ сумму всех чисел $x_{k}$, для которых $k \equiv r(\bmod m+1), k \leqslant n$. Тогда $x_{1}+x_{2}+\ldots+x_{n}=a_{1}+a_{2}+\ldots+a_{m}$. Заметим, что
$$
\sum_{im$.
1.6. Prove the inequality (1) for $n=8,10,12$ and for $n=7,9,11,13,15,17,19,21,23$. Due to the statement of the previous problem it is sufficient to prove the inequality only for $n=12$ and $n=23$.
1.7. Prove that $f\left(x_{1}, x_{2}, \ldots, x_{n}\right)+f\left(x_{n}, x_{n-1}, \ldots, x_{1}\right) \geqslant n$.
1.8. Assume that the function $f\left(x_{1}, x_{2}, \ldots, x_{n}\right)$ has a local minimum in the point $\left(a_{1}, a_{2}, \ldots, a_{n}\right), a_{1}, a_{2}, \ldots, a_{n}>0$.
a) Prove that $f\left(a_{1}, a_{2}, \ldots, a_{n}\right)=n / 2$ if $n$ is even.
b*) Prove the same statement for odd $n$.
c) Use the statements niiSa) and b) to prove the inequality for $n=7$ and $n=8$.
1.9. Prove the inequality $f\left(x_{1}, x_{2}, \ldots, x_{n}\right) \geqslant c n$ for the following values of the constant $c$ :
a) $c=1 / 4$
b) $c=(\sqrt{2}-1)$;
c) $c=5 / 12$.
## 2 Useful and related inequalities
Prove the following inequalities assuming that all the $x_{k}$ 's are positive. Prove that the constants printed in bold can not be decreased (for each $n$ ).
2.1. Mordell's inequality.
a) $\left(\sum_{k=1}^{n} x_{k}\right)^{2} \geqslant \min \left\{\frac{\mathbf{n}}{\mathbf{2}}, \mathbf{3}\right\} \cdot \sum_{k=1}^{n} x_{k}\left(x_{k+1}+x_{k+2}\right)$.
b) Find all $n$-tuples $x_{1}, x_{2}, \ldots, x_{n}$ such that the equality is achieved.
2.2. $\left(\sum_{k=1}^{n} x_{k}\right)^{2} \geqslant \min \left\{\frac{\mathbf{n}}{\mathbf{3}}, \frac{8}{\mathbf{3}}\right\} \cdot \sum_{k=1}^{n} x_{k}\left(x_{k+1}+x_{k+2}+x_{k+3}\right)$.
2.3. пïSa) Prove that for $n \leqslant 8$
$$
\frac{x_{1}}{x_{2}+x_{3}+x_{4}}+\frac{x_{2}}{x_{3}+x_{4}+x_{5}}+\ldots+\frac{x_{n-1}}{x_{n}+x_{1}+x_{2}}+\frac{x_{n}}{x_{1}+x_{2}+x_{3}} \geqslant \frac{n}{3}
$$
b) For which $n>8$ this inequality is also true?
2.4. $\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2} \geqslant 4\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n-1} x_{n}+x_{n} x_{1}\right) ; \quad n \geqslant 4$.
2.5. $\sum_{k=1}^{n} \frac{x_{k}}{x_{k+1}+x_{k+2}} \geqslant \sum_{k=1}^{n} \frac{x_{k+1}}{x_{k}+x_{k+1}}$.
2.6. $\frac{x_{1}}{x_{n}+x_{2}}+\frac{x_{2}}{x_{1}+x_{3}}+\ldots+\frac{x_{n-1}}{x_{n-2}+x_{n}}+\frac{x_{n}}{x_{n-1}+x_{1}} \geqslant \mathbf{2} ; \quad n \geqslant 4$.
2.7. $\frac{x_{1}+x_{2}}{x_{1}+x_{3}}+\frac{x_{2}+x_{3}}{x_{2}+x_{4}}+\ldots+\frac{x_{n-1}+x_{n}}{x_{n-1}+x_{1}}+\frac{x_{n}+x_{1}}{x_{n}+x_{2}} \geqslant \mathbf{4} ; \quad n \geqslant 4$.
2.8. $\frac{x_{1}}{x_{n}+x_{3}}+\frac{x_{2}}{x_{1}+x_{4}}+\ldots+\frac{x_{n-1}}{x_{n-2}+x_{1}}+\frac{x_{n}}{x_{n-1}+x_{2}} \geqslant \mathbf{3} ; \quad n \geqslant 4$.
2.9. $\frac{x_{2}+x_{3}}{x_{1}+x_{4}}+\frac{x_{3}+x_{4}}{x_{2}+x_{5}}+\ldots+\frac{x_{n}+x_{1}}{x_{n-1}+x_{2}}+\frac{x_{1}+x_{2}}{x_{n}+x_{3}} \geqslant \mathbf{6} ; \quad n \geqslant 6$.
2.10. $\frac{x_{1}+x_{2}}{x_{1}+x_{4}}+\frac{x_{2}+x_{3}}{x_{2}+x_{5}}+\ldots+\frac{x_{2004}+x_{1}}{x_{2004}+x_{3}} \geqslant \mathbf{6}$.
2.11. $\frac{x_{1}}{x_{n}+x_{4}}+\frac{x_{2}}{x_{1}+x_{5}}+\ldots+\frac{x_{n-1}}{x_{n-2}+x_{2}}+\frac{x_{n}}{x_{n-1}+x_{3}} \geqslant 4$, where $n>5$ is even.
2.12. $\sum_{k=1}^{n} \frac{x_{k}^{2}}{x_{k+1}^{2}-x_{k+1} x_{k+2}+x_{k+2}^{2}} \geqslant\left[\frac{\mathbf{n}+\mathbf{1}}{\mathbf{2}}\right]$.
## 3 After the intermediate finish
1.10. a) Prove that for each $n$ there exists $q_{n}>1$, such that for all real $x_{1}, x_{2}, \ldots, x_{n} \in\left[\frac{1}{q_{n}} ; q_{n}\right]$ the inequality (1) holds.
$\left.\mathrm{b}^{*}\right)$ Is it possible to choose $q>1$, such that for all integers $n>0$ and for all $x_{i} \in\left[\frac{1}{q} ; q\right]$ the inequality (1) holds?
1.11. Let $S=f\left(x_{1}, x_{2}, \ldots, x_{n}\right)$ be the left hand side of Shapiro's inequality. Denote by $a_{1}, a_{2}, \ldots, a_{n}$ the numbers $x_{2} / x_{1}, x_{3} / x_{2}, \ldots, x_{n} / x_{n-1}, x_{1} / x_{n}$, arranged in increasing order.
a) Prove that $S \geqslant \frac{1}{a_{1}\left(1+a_{n}\right)}+\frac{1}{a_{2}\left(1+a_{n-1}\right)}+\ldots+\frac{1}{a_{n}\left(1+a_{1}\right)}$;
b) Let $b_{k}=\left\{\begin{array}{ll}\frac{1}{a_{k} a_{n+1-k}}, & a_{k} a_{n+1-k} \geqslant 1 \\ \frac{a_{k} a_{n+1-k}+\sqrt{a_{k} a_{n+1-k}}}{}, & a_{k} a_{n+1-k}<1 .\end{array} \quad\right.$ Prove that $2 S \geqslant b_{1}+b_{2}+\ldots+b_{n}$;
c) Let $g$ be the maximal convex function that does not exceed both functions $e^{-x}$ nïS $2\left(e^{x}+e^{x / 2}\right)^{-1}$. Prove that $2 S \geqslant g\left(\ln \left(a_{1} a_{n}\right)\right)+g\left(\ln \left(a_{2} a_{n-1}\right)\right)+\ldots+g\left(\ln \left(a_{n} a_{1}\right)\right) \geqslant n g(0)$.
d) Prove that for each $\lambda>g(0)$ there exist a nonnegative integer $n$ and positive numbers $x_{1}, x_{2}, \ldots, x_{n}$, such that $S \leqslant \lambda n$.
## Solutions
1.1. $\underline{n=3}$. Let $S=x_{1}+x_{2}+x_{3}$. It is easy to see that the function $f(t)=\frac{t}{S-t}$ is convex on the interval $[0 ; S)$. Apply the Jensen inequality to it:
$$
\frac{f\left(x_{1}\right)+f\left(x_{2}\right)+f\left(x_{3}\right)}{3} \geqslant f\left(\frac{x_{1}+x_{2}+x_{3}}{3}\right)=f\left(\frac{S}{3}\right)=\frac{1}{2}
$$
We are done.
$\underline{n=4}$. This inequality is cyclic. Write down the values of $x_{i}$ 's successively at the vertices of a square. Then on each diagonal put an arrow leading from the smaller value to the greater one. Notice that there is a side of the square with two tails on it. Re-number the $x_{i}$ 's in such a manner that this side becomes $x_{4} x_{1}$. Now we may assume that $x_{1} \geqslant x_{3}, x_{4} \geqslant x_{2}$. For the variables with these restrictions the following inequality is true:
$$
\frac{x_{1}}{x_{2}+x_{3}}+\frac{x_{3}}{x_{4}+x_{1}} \geqslant \frac{x_{1}}{x_{4}+x_{3}}+\frac{x_{3}}{x_{2}+x_{1}}
$$
Indeed, re-write it in the following way:
$$
x_{1}\left(\frac{1}{x_{2}+x_{3}}-\frac{1}{x_{4}+x_{3}}\right) \geqslant x_{3}\left(\frac{1}{x_{2}+x_{1}}-\frac{1}{x_{4}+x_{1}}\right)
$$
Reduce both hands to a common denominator, cancel $x_{4}-x_{2}$ in both hands (if $x_{4}-x_{2}=0$, we already have the equality), and multiply both hands to the product of denominators. We obtain the evident (since $x_{1} \geqslant x_{3}$ ) inequality
$$
x_{1}\left(x_{2}+x_{1}\right)\left(x_{4}+x_{1}\right) \geqslant x_{3}\left(x_{2}+x_{3}\right)\left(x_{4}+x_{3}\right)
$$
Use it to prove Shapiro's inequality:
$\frac{x_{1}}{x_{2}+x_{3}}+\frac{x_{2}}{x_{3}+x_{4}}+\frac{x_{3}}{x_{4}+x_{1}}+\frac{x_{4}}{x_{1}+x_{2}} \geqslant \frac{x_{1}}{x_{4}+x_{3}}+\frac{x_{2}}{x_{3}+x_{4}}+\frac{x_{3}}{x_{2}+x_{1}}+\frac{x_{4}}{x_{1}+x_{2}}=\frac{x_{1}+x_{2}}{x_{3}+x_{4}}+\frac{x_{3}+x_{4}}{x_{1}+x_{2}}=a+a^{-1} \geqslant 2$.
$n=5$. Notice that the function $f(t)=1 /(S-t)$ is convex on the interval $[0 ; S)$. So we can apply the Jensen inequality with $n=5$ :
$$
a_{1} f\left(t_{1}\right)+a_{2} f\left(t_{2}\right)+a_{3} f\left(t_{3}\right)+a_{4} f\left(t_{4}\right)+a_{5} f\left(t_{5}\right) \geqslant f\left(a_{1} t_{1}+a_{2} t_{2}+a_{3} t_{3}+a_{4} t_{4}+a_{5} t_{5}\right)
$$
where $a_{i} \geqslant 0, \sum a_{i}=1$. Take $a_{i}=\frac{x_{i}}{S}$, and let $t_{i}=x_{i}+x_{i-1}+x_{i-2}, i=1, \ldots, 5$ (we assume that the variables are enumerated cyclically: $\left.x_{0}=x_{5}, x_{-1}=x_{4}\right)$. Then $f\left(t_{i}\right)=\frac{1}{S-t_{i}}=\frac{1}{x_{i+1}+x_{i+2}}$, and it means that the left-hand side of inequality (2) coincides with the left-hand side of Shapiro's inequality. Now consider the right-hand side of 2 :
$$
\frac{1}{S-\sum_{i=1}^{5} a_{i} t_{i}}=\frac{1}{S-\sum_{i=1}^{5} \frac{x_{i}}{S}\left(x_{i}+x_{i-1}+x_{i-2}\right)}=\frac{S}{S^{2}-\sum_{i=1}^{5} x_{i}\left(x_{i}+x_{i-1}+x_{i-2}\right)}
$$
Open the brackets. It is easy to see that the denominator is the sum of pairwise products of the set of variables $x_{i}$. Since the initial inequality is homogeneous, we may assume that $S=x_{1}+x_{2}+x_{3}+x_{4}+x_{5}=1$. Now the right-hand side of inequality (2) is the inverse number to the sum of pairwise products of the variables $x_{i}$, satisfying one condition $x_{1}+x_{2}+x_{3}+x_{4}+x_{5}=1$. The right-hand side reaches its minimum when the sum of pairwise products reaches its maximum. It is well-known that for it all the variables should be equal. But the right-hand side equals $5 / 2$ in this point.
The analogous proof also works for $n=4$.
$\underline{n=6}$. Proceed as above. The function $f(t)=1 /(S-t)$ is convex on the interval $[0 ; S)$. So we can apply the Jensen inequality with $n=6$ :
$$
\sum_{i=1}^{6} a_{i} f\left(t_{i}\right) \geqslant f\left(\sum_{i=1}^{6} a_{i} t_{i}\right)
$$
Let $a_{i}=\frac{x_{i}}{S}, t_{i}=x_{i}+x_{i-1}+x_{i-2}+x_{i-3}, i=1, \ldots, 6$ (we assume that the variables are enumerated cyclically: $\left.x_{0}=x_{6}, x_{-1}=x_{5}, x_{-2}=x_{4}\right)$. Then $f\left(t_{i}\right)=\frac{1}{S-t_{i}}=\frac{1}{x_{i+1}+x_{i+2}}$, and this means that the left-hand side of the inequality (1.1) coincides with the left-hand side of Shapiro's inequality. Now consider the right-hand side of (1.1):
$$
\frac{1}{S-\sum_{i=1}^{6} a_{i} t_{i}}=\frac{1}{S-\sum_{i=1}^{6} \frac{x_{i}}{S}\left(x_{i}+x_{i-1}+x_{i-2}+x_{i-3}\right)}=\frac{S}{S^{2}-\sum_{i=1}^{6} x_{i}\left(x_{i}+x_{i-1}+x_{i-2}+x_{i-3}\right)}
$$
Open the brackets. It is easy to see that the denominator is the sum of pairwise products of the variables $x_{i}$ 's but the products $x_{1} x_{4}, x_{2} x_{5}$, and $x_{3} x_{6}$. This sum can be re-written as $\left(x_{1}+x_{4}\right)\left(x_{2}+x_{5}\right)+\left(x_{1}+x_{4}\right)\left(x_{3}+x_{6}\right)+\left(x_{2}+x_{5}\right)\left(x_{3}+x_{6}\right)$. Denote $A=x_{1}+x_{4}, B=x_{2}+x_{5}, C=x_{3}+x_{6}$. The right-hand side of (1.1) can be re-written as
$$
\frac{A+B+C}{A B+B C+A C}
$$
Since the initial inequality is homogeneous, we may assume that $S=x_{1}+x_{2}+x_{3}+x_{4}+x_{5}=A+B+C=1$. Now it is clear that the expression (3) is greater than or equal to 3 , since $(A+B+C)^{2} \geqslant 3(A B+B C+A C)$. пїSпї̈
Remark. Unfortunately, this method does not work for $n>6$.
Second solution. Apply the Cauchy-Bunyakovsky inequality to the sets of numbers
$$
\begin{array}{cccc}
\sqrt{\frac{x_{1}}{x_{2}+x_{3}}}, & \sqrt{\frac{x_{2}}{x_{3}+x_{4}}}, \ldots, & \sqrt{\frac{x_{n}}{x_{1}+x_{2}}} \\
\sqrt{x_{1}\left(x_{2}+x_{3}\right)}, & \sqrt{x_{2}\left(x_{3}+x_{4}\right)}, & \ldots, \quad \sqrt{x_{n}\left(x_{1}+x_{2}\right)}
\end{array}
$$
We obtain
$$
\frac{x_{1}}{x_{2}+x_{3}}+\frac{x_{2}}{x_{3}+x_{4}}+\ldots+\frac{x_{n}}{x_{1}+x_{2}} \geqslant \frac{\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2}}{x_{1}\left(x_{2}+x_{3}\right)+x_{2}\left(x_{3}+x_{4}\right)+\ldots+x_{n}\left(x_{1}+x_{2}\right)}
$$
Use Mordell's inequality (problem 2.1). When $n \leqslant 6$, it gives us that the right-hand side of this inequality is greater than or equal to $n / 2$.
1.2. a) [22] Take as $x_{1}, x_{2}, \ldots, x_{20}$ numbers
$$
\begin{array}{llllllllll}
1+5 \varepsilon, & 6 \varepsilon, & 1+4 \varepsilon, & 5 \varepsilon, & 1+3 \varepsilon, & 4 \varepsilon, & 1+2 \varepsilon, & 3 \varepsilon, & 1+\varepsilon, & 2 \varepsilon, \\
1+2 \varepsilon, & \varepsilon, & 1+3 \varepsilon, & 2 \varepsilon, & 1+4 \varepsilon, & 3 \varepsilon, & 1+5 \varepsilon, & 4 \varepsilon, & 1+6 \varepsilon, & 5 \varepsilon .
\end{array}
$$
Then $f\left(x_{1}, \ldots, x_{20}\right)<10-\varepsilon^{2}+c \varepsilon^{3}<10$ for some $c$ and small enough $\varepsilon$.
b) [27] Take as $x_{1}, x_{2}, \ldots, x_{14}$ numbers
$$
1+7 \varepsilon, 7 \varepsilon, 1+4 \varepsilon, 6 \varepsilon, 1+\varepsilon, 5 \varepsilon, 1,2 \varepsilon, 1+\varepsilon, 0,1+4 \varepsilon, \varepsilon, 1+6 \varepsilon, 4 \varepsilon
$$
Then $f\left(x_{1}, \ldots, x_{20}\right)<7-2 \varepsilon^{2}+c \varepsilon^{3}<7$ for some $c$ and small enough $\varepsilon$.
An alternative example [24]:
$$
0,42,2,42,4,41,5,39,4,38,2,38,0,40
$$
c) $[10],[18]$. Take
$0,85,0,101,0,120,14,129,41,116,59,93,64,71,63,52,60,36,58,23,58,12,62,3,71$.
Alternatively, in [3] the following example is given:
$32,0,37,0,43,0,50,0,59,8,62,21,55,29,44,32,33,31,24,30,16,29,10,29,4$.
1.3. The statement of the problem is published in [13]. We present here a short nice solution.
Let $x_{1} \geqslant x_{2} \geqslant \cdots \geqslant x_{n}>0$. Observe that the product of $n$ fractions $\frac{x_{k}+x_{k+1}}{x_{k+1}+x_{k+2}}$ is equal to 1 . Then by Cauchy inequality we conclude that
$$
\sum_{k=1}^{n} \frac{x_{k}+x_{k+1}}{x_{k+1}+x_{k+2}} \geqslant n=\sum_{k=1}^{n} \frac{x_{k+1}+x_{k+2}}{x_{k+1}+x_{k+2}}
$$
Hence
$$
\sum_{k=1}^{n} \frac{x_{k}}{x_{k+1}+x_{k+2}} \geqslant \sum_{k=1}^{n} \frac{x_{k+2}}{x_{k+1}+x_{k+2}}=\sum_{k=1}^{n} \frac{x_{k+1}}{x_{k}+x_{k+1}}
$$
Now we will apply the rearranging inequality: Let $a_{1} \geqslant \ldots \geqslant a_{n}$ and $b_{1} \geqslant \ldots \geqslant b_{n}$ be two sets of numbers. Then for each permutation $k_{1}, \ldots, k_{n}$ of numbers $1, \ldots, n$ the following inequality holds
$$
a_{1} b_{1}+a_{2} b_{2}+\ldots a_{n} b_{n} \geqslant a_{1} b_{k_{1}}+a_{2} b_{k_{2}}+\ldots a_{n} b_{k_{n}} \geqslant a_{1} b_{n}+a_{2} b_{n-1}+\ldots a_{n} b_{1}
$$
Use the rearranging inequality twice
$$
\begin{aligned}
\sum_{k=1}^{n} \frac{x_{k}}{x_{k+1}+x_{k+2}} & =\sum_{k=1}^{n-2} \frac{x_{k}}{x_{k+1}+x_{k+2}}+\frac{x_{n-1}}{x_{n}+x_{1}}+\frac{x_{n}}{x_{1}+x_{2}} \geqslant \underset{(*)}{\geqslant} \\
& \geqslant \sum_{k=1}^{n-2} \frac{x_{k}}{x_{k+1}+x_{k+2}}+\frac{x_{n-1}}{x_{1}+x_{2}}+\frac{x_{n}}{x_{n}+x_{1}} \underset{(* *)}{\geqslant} \\
& \geqslant \sum_{k=1}^{n} \frac{x_{k}}{x_{k}+x_{k+1}}
\end{aligned}
$$
The inequality $(*)$ here is the rearranging inequality for two pairs of numbers: $x_{n-1} \geqslant x_{n}$ and $\frac{1}{x_{n}+x_{1}} \geqslant \frac{1}{x_{1}+x_{2}}$; and the inequality $(* *)$ is the rearranging inequality for the sets $x_{1}, x_{2}, \ldots, x_{n-1}$ and $\frac{1}{x_{1}+x_{2}}, \frac{1}{x_{2}+x_{3}}, \ldots, \frac{1}{x_{n-1}+x_{n}}$ that have opposite ordering.
Thus
$$
2 \sum_{k=1}^{n} \frac{x_{k}}{x_{k+1}+x_{k+2}} \geqslant \sum_{k=1}^{n} \frac{x_{k}}{x_{k}+x_{k+1}}+\sum_{k=1}^{n} \frac{x_{k+1}}{x_{k}+x_{k+1}}=n
$$
For the decreasing set $x_{i}$ the solution is similar because we do not use the order of the variables when we apply the Cauchy inequality, and for the rearranging inequalities we need the fact that the sets $x_{i}$ and $\frac{1}{x_{i}+x_{i+1}}$ have different orderings.
1.4. [3] It is easy to see that $f_{n+2}\left(x_{1}, x_{2}, \ldots, x_{n}, x_{1}, x_{2}\right)=f_{n}\left(x_{1}, x_{2}, \ldots, x_{n}\right)+1$. Therefore if $f_{n}\left(x_{1}, x_{2}, \ldots, x_{n}\right)<$ $n / 2$, then $f_{n+2}\left(x_{1}, x_{2}, \ldots, x_{n}, x_{1}, x_{2}\right)<(n+2) / 2$.
1.5. [3] Assume that $f_{m}\left(x_{1}, x_{2}, \ldots, x_{m}\right)<\frac{m}{2}$. For each $k$ let us calculate the difference
$$
\begin{aligned}
& f_{m+1}\left(x_{1}, \ldots, x_{k}, x_{k}, x_{k+1}, \ldots, x_{m}\right)-f_{m}\left(x_{1}, x_{2}, \ldots, x_{m}\right)-\frac{1}{2}= \\
& =\frac{x_{k-1}}{2 x_{k}}+\frac{x_{k}}{x_{k}+x_{k+1}}-\frac{x_{k-1}}{x_{k}+x_{k+1}}-\frac{1}{2}=\frac{\left(x_{k}-x_{k-1}\right)\left(x_{k}-x_{k+1}\right)}{2 x_{k}\left(x_{k}+x_{k+1}\right)} .
\end{aligned}
$$
If $\left(x_{k}-x_{k-1}\right)\left(x_{k}-x_{k+1}\right) \leqslant 0$, then
$$
f_{n+1}\left(x_{1}, x_{2}, \ldots, x_{k}, x_{k}, x_{k+1}, \ldots, x_{m}\right)<\frac{m+1}{2}
$$
and we are done. If $n$ is odd, we can always choose $k$ such that $\left(x_{k}-x_{k-1}\right)\left(x_{k}-x_{k+1}\right) \leqslant 0$ because otherwise the product of the (odd number of) inequalities $\left(x_{k}-x_{k-1}\right)\left(x_{k+1}-x_{k}\right)<0$ for all $k$ is
$$
\left(x_{2}-x_{1}\right)^{2}\left(x_{3}-x_{2}\right)^{2} \ldots\left(x_{m}-x_{m-1}\right)^{2}\left(x_{1}-x_{m}\right)^{2}<0
$$
Thus if for odd $n$ the Shapiro inequality is wrong then for $n+1$ it is wrong, too. It remains to apply the statement of the previous problem.
1.6. $[7,8]$
1.7. [28] Let $y_{k}=x_{k}+x_{k+1}$. Then
$$
\frac{x_{1}+x_{4}}{x_{2}+x_{3}}+\frac{x_{2}+x_{5}}{x_{3}+x_{4}}+\ldots+\frac{x_{n}+x_{3}}{x_{1}+x_{2}}=\sum_{k=1}^{n} \frac{y_{k}-y_{k+1}+y_{k+2}}{y_{k+1}}=\sum_{k=1}^{n} \frac{y_{k}}{y_{k+1}}+\sum_{k=1}^{n} \frac{y_{k+2}}{y_{k+1}}-n \geqslant n
$$
because by Cauchy inequality each sum is at least $n$.
1.8. The statements a), b) were published in [21].
a) !!! This short proof is taken from [8].
Denote for brevity $a=\left(a_{1}, a_{2}, \ldots, a_{n}\right), x=\left(x_{1}, x_{2}, \ldots, x_{n}\right)$, and $u=(-1,1,-1,1, \ldots,-1,1)$.
Observe that
$$
\frac{\partial f}{\partial x_{k}}(x)=\frac{1}{x_{k+1}+x_{k+2}}-\frac{x_{k-2}}{\left(x_{k-1}+x_{k}\right)^{2}}-\frac{x_{k-1}}{\left(x_{k}+x_{k+1}\right)^{2}}
$$
It is easy to see that we have an identity
$$
f(x+t u)=f(x)+t \sum_{k=1}^{n}(-1)^{k} \frac{\partial f}{\partial x_{k}}(x)
$$
Since $a$ is the minimum point, we have
$$
\frac{\partial f}{\partial x_{k}}(a)=0
$$
Therefore $f(a+t u)=f(a)$ if all the coordinates of the point $a+t u$ are positive. Hence $a+t u$ is the minimum point of the function $f$ as well. Hence,
$$
\frac{\partial f}{\partial x_{k}}(a+t u)=0
$$
So
and
$$
\frac{1}{a_{k+1}+a_{k+2}}-\frac{a_{k-2}}{\left(a_{k-1}+a_{k}\right)^{2}}-\frac{a_{k-1}}{\left(a_{k}+a_{k+1}\right)^{2}}=0
$$
$$
\frac{1}{a_{k+1}+a_{k+2}}-\frac{a_{k-2}+t(-1)^{k-2}}{\left(a_{k-1}+a_{k}\right)^{2}}-\frac{a_{k-1}+t(-1)^{k-1}}{\left(a_{k}+a_{k+1}\right)^{2}}=0
$$
Subtract the first equality from the second:
$$
\frac{t}{\left(a_{k-1}+a_{k}\right)^{2}}-\frac{t}{\left(a_{k}+a_{k+1}\right)^{2}}=0
$$
Therefore,
$$
a_{k-1}+a_{k}=a_{k}+a_{k+1}
$$
and hence
$$
a_{1}=a_{3}=a_{5}=\cdots=a_{n-1} \quad \text { and } \quad a_{2}=a_{4}=a_{6}=\cdots=a_{n}
$$
Thus, $f(a)=n / 2$.
b) This short proof is taken from [7]. Denote for brevity $a=\left(a_{1}, a_{2}, \ldots, a_{n}\right), x=\left(x_{1}, x_{2}, \ldots, x_{n}\right), y=$ $\left(y_{1}, y_{2}, \ldots, y_{n}\right), z=\left(z_{1}, z_{2}, \ldots, z_{n}\right)$, where $y_{k}=x_{k}+x_{k+1}$ and $z_{k}=1 / y_{n+1-k}$.
Set
$$
S(x)=\frac{x_{1}}{x_{2}+x_{3}}+\frac{x_{2}}{x_{3}+x_{4}}+\ldots+\frac{x_{n-1}}{x_{n}+x_{1}}+\frac{x_{n}}{x_{1}+x_{2}}=\sum_{k=0}^{n-1} \frac{x_{k}}{y_{k+1}}
$$
Observe that
$$
\frac{\partial f}{\partial x_{k}}(x)=\frac{1}{x_{k+1}+x_{k+2}}-\frac{x_{k-2}}{\left(x_{k-1}+x_{k}\right)^{2}}-\frac{x_{k-1}}{\left(x_{k}+x_{k+1}\right)^{2}}
$$
It is easy to check the following identities:
$$
\frac{a}{b}+\frac{c}{d}=\frac{a+c}{b+d}+\frac{\frac{a}{b^{2}}+\frac{c}{d^{2}}}{\frac{1}{b}+\frac{1}{d}}
$$
Hence,
$$
\begin{aligned}
\frac{x_{k-2}}{x_{k-1}+x_{k}}+\frac{x_{k-1}}{x_{k}+x_{k+1}} & =\frac{x_{k-2}+x_{k-1}}{\left(x_{k-1}+x_{k}\right)+\left(x_{k}+x_{k+1}\right)}+\frac{\frac{x_{k-2}}{\left(x_{k-1}+x_{k}\right)^{2}}+\frac{x_{k-1}}{\left(x_{k}+x_{k+1}\right)^{2}}}{\frac{1}{x_{k-1}+x_{k}}+\frac{1}{x_{k}+x_{k+1}}}= \\
& =\frac{y_{k-2}}{y_{k-1}+y_{k}}+\frac{z_{n-k}-\frac{\partial f}{\partial x_{k}}(x)}{z_{n-k+1}+z_{n-k+2}}
\end{aligned}
$$
Therefore,
$$
2 S(x)=S(y)+S(z)-\sum_{k=1}^{n} \frac{\frac{\partial f}{\partial x_{k}}(x)}{z_{n-k+1}+z_{n-k+2}}
$$
If $x$ is a minimum point then we have $2 S(x)=S(y)+S(z)$. Hence $S(x)=S(y)=S(z)$.
Let $u:=\left(x_{1}+x_{2}+\cdots+x_{n}\right) / n$. Consider the transformation $M: \mathbb{R}^{n} \rightarrow \mathbb{R}^{n}$ defined by
$$
M(x)=\left(\frac{x_{1}+x_{2}}{2}, \frac{x_{2}+x_{3}}{2}, \ldots, \frac{x_{n}+x_{1}}{2}\right) .
$$
Let $M_{k}(x)$ be its $k$-th iteration. Observe that $S(x)=S(y)=S(M(x))=\cdots=S\left(M_{k}(x)\right)$. It is clear that $\lim _{k \rightarrow \infty} M_{k}(x)=(u, u, \ldots, u)$. Then
$$
S(x)=\lim _{k \rightarrow \infty} S\left(M_{k}(x)\right)=S((u, u, \ldots, u))=\frac{n}{2}
$$
c) $[16],[7,8]$
1.9. These solutions are taken from [3].
niiS) The problem was presented at the Third USSR mathematical olympiad, 1969. Probably it was originally published in [14].
Let $x_{i_{1}}$ be the maximal number among $x_{1}, x_{2}, \ldots, x_{n} ; x_{i_{2}}$ be the maximum of the two next numbers after $x_{i_{1}}$ (i.e. of $x_{i_{1}+1}$ and $x_{i_{1}+2}$ ); $x_{i_{3}}$ be the maximum of the two next numbers after $x_{i_{2}}$, and so on. We will continue this sequence till the step number $k$ when the maximum of the two next after $x_{i_{k}}$ numbers is $x_{i_{1}}$.
It is clear that $k \geqslant n / 2$. We have
$$
\frac{x_{1}}{x_{2}+x_{3}}+\frac{x_{2}}{x_{3}+x_{4}}+\ldots+\frac{x_{n}}{x_{1}+x_{2}} \geqslant \frac{x_{i_{1}}}{2 x_{i_{2}}}+\frac{x_{i_{2}}}{2 x_{i_{3}}}+\ldots+\frac{x_{i_{k}}}{2 x_{i_{1}}}
$$
The last expression is at least $k / 2$ by the Cauchy inequality therefore it is at least $n / 4$.
b) Rewrite each of the fractions $\frac{x_{k}}{x_{k+1}+x_{k+2}}, k=1,2, \ldots, n$, in the form
$$
\frac{x_{k}}{x_{k+1}+x_{k+2}}=\frac{x_{k}+\frac{1}{2} x_{k+1}}{x_{k+1}+x_{k+2}}+\frac{\frac{1}{2} x_{k+1}+x_{k+2}}{x_{k+1}+x_{k+2}}-1
$$
We obtain $2 n$ fractions. Combine them by pairs: the first and the last, the second and the third, the fourth and the fifth and so on. Now estimate the sum of each pair from below
$$
\begin{aligned}
\frac{\frac{1}{2} x_{k}+x_{k+1}}{x_{k}+x_{k+1}}+\frac{x_{k}+\frac{1}{2} x_{k+1}}{x_{k+1}+x_{k+2}} \geqslant 2 \sqrt{\frac{\left(\frac{1}{2} x_{k}+x_{k+1}\right)\left(x_{k}+\frac{1}{2} x_{k+1}\right)}{\left(x_{k}+x_{k+1}\right)\left(x_{k+1}+x_{k+2}\right)}}= \\
\quad=2 \sqrt{\left(\frac{1}{2}+\frac{x_{k} x_{k+1}}{4\left(x_{k}+x_{k+1}\right)^{2}}\right) \frac{x_{k}+x_{k+1}}{x_{k+1}+x_{k+2}}}>\sqrt{2} \cdot \sqrt{\frac{x_{k}+x_{k+1}}{x_{k+1}+x_{k+2}}}
\end{aligned}
$$
Since the product of $n$ numbers $\sqrt{\frac{x_{1}+x_{2}}{x_{2}+x_{3}}}, \sqrt{\frac{x_{2}+x_{3}}{x_{3}+x_{4}}}, \ldots, \sqrt{\frac{x_{n}+x_{1}}{x_{1}+x_{2}}}$ equals 1 , then by the Cauchy inequality their sum is at least $n$. Therefore $f\left(x_{1}, \ldots, x_{n}\right) \geqslant \sqrt{2} n-n=(\sqrt{2}-1) n$.
c) As in the previous solution rewrite each of the fractions $\frac{x_{k}}{x_{k+1}+x_{k+2}}, k=1,2, \ldots, n$, in the form
$$
\frac{x_{k}}{x_{k+1}+x_{k+2}}=\frac{x_{k}+\beta x_{k+1}}{x_{k+1}+x_{k+2}}+\alpha \cdot \frac{\beta x_{k+1}+x_{k+2}}{x_{k+1}+x_{k+2}}-\alpha
$$
where $\alpha$ and $\beta$ are parameters chosen to make the equality true. For such a choice of $\alpha$ and $\beta$ we need $\beta+\alpha \beta=\alpha$, i.e. $\beta=\alpha /(\alpha+1)$. Then
$$
\begin{aligned}
\frac{x_{k}+\beta x_{k+1}}{x_{k+1}+x_{k+2}}+\alpha \cdot \frac{\beta x_{k}+x_{k+1}}{x_{k}+x_{k+1}} \geqslant 2 \sqrt{\alpha \frac{\left(x_{k}+\beta x_{k+1}\right)\left(\beta x_{k}+x_{k+1}\right)}{\left(x_{k}+x_{k+1}\right)\left(x_{k+1}+x_{k+2}\right)}}= \\
\quad=2 \sqrt{\alpha \frac{\beta\left(x_{k}+x_{k+1}\right)^{2}+(\beta-1)^{2} x_{k} x_{k+1}}{\left(x_{k}+x_{k+1}\right)\left(x_{k+1}+x_{k+2}\right)}}>2 \sqrt{\alpha \beta \frac{x_{k}+x_{k+1}}{x_{k+1}+x_{k+2}}}=\frac{2 \alpha}{\sqrt{\alpha+1}} \cdot \sqrt{\frac{x_{k}+x_{k+1}}{x_{k+1}+x_{k+2}}}
\end{aligned}
$$
Therefore
$$
\begin{array}{r}
\frac{x_{1}}{x_{2}+x_{3}}+\frac{x_{2}}{x_{3}+x_{4}}+\ldots+\frac{x_{n-1}}{x_{n}+x_{1}}+\frac{x_{n}}{x_{1}+x_{2}} \geqslant \frac{2 \alpha}{\sqrt{\alpha+1}}\left(\sqrt{\frac{x_{1}+x_{2}}{x_{2}+x_{3}}}+\sqrt{\frac{x_{2}+x_{3}}{x_{3}+x_{4}}}+\ldots+\sqrt{\frac{x_{n}+x_{1}}{x_{1}+x_{2}}}\right)-\alpha n> \\
>\frac{2 \alpha}{\sqrt{\alpha+1}} n-\alpha n=\left(\frac{2 \alpha}{\sqrt{\alpha+1}}-\alpha\right) n
\end{array}
$$
For $\alpha=\frac{5}{4}$ we have $c=5 / 12$.
Remark. This is a good approximation.The expression $g(\alpha)=\frac{2 \alpha}{\sqrt{\alpha+1}}-\alpha$ reaches its maximal value at $\alpha=$ $\alpha_{0} \approx 1.1479$ (this is a root of the cubic equation $g^{\prime}(\alpha)=0$ ), and the minimum value is $g\left(\alpha_{0}\right) \approx 0.4186$. For $\alpha=\frac{5}{4}$ we have $g(\alpha)=\frac{5}{12} \approx 0.416$.
1.10. [9]. Set $y_{k}=x_{k}+x_{k+1}$. We need to prove that
$$
\frac{x_{1}}{y_{2}}+\frac{x_{2}}{y_{3}}+\ldots+\frac{x_{n}}{y_{1}} \geqslant \frac{n}{2}
$$
or
$$
\sum_{k=1}^{n} \frac{2 q_{n}^{2} x_{k}-y_{k+1}}{y_{k+1}} \geqslant n\left(q_{n}^{2}-1\right)
$$
We suppose that the parameter $q_{n}$ will be chosen later. Since
$$
2 q_{n}^{2} x_{k}-y_{k+1}=\left(q_{n}^{2} x_{k}-x_{k+1}\right)+\left(q_{n}^{2} x_{k}-x_{k+2}\right) \geqslant 0
$$
by the Cauchy-Bunyakovsky inequality for sets
$$
\left\{\sqrt{\frac{2 q_{n}^{2} x_{k}-y_{k+1}}{y_{k+1}}}\right\} \text { and }\left\{\sqrt{\left(2 q_{n}^{2} x_{k}-y_{k+1}\right) y_{k+1}}\right\}
$$
we have
$$
\sum_{k=1}^{n} \frac{2 q_{n}^{2} x_{k}-y_{k+1}}{y_{k+1}} \geqslant \frac{\left(\sum_{k=1}^{n}\left(2 q_{n}^{2} x_{k}-y_{k+1}\right)\right)^{2}}{\sum_{k=1}^{n}\left(2 q_{n}^{2} x_{k}-y_{k+1}\right) y_{k+1}}
$$
So it suffices to prove that
$$
A^{2}:=\left(\sum_{k=1}^{n}\left(2 q_{n}^{2} x_{k}-y_{k+1}\right)\right)^{2} \geqslant n\left(q_{n}^{2}-1\right) \sum_{k=1}^{n}\left(2 q_{n}^{2} x_{k}-y_{k+1}\right) y_{k+1}=: n\left(q_{n}^{2}-1\right) B
$$
Since $\sum_{k=1}^{n} y_{k}=2 \sum_{k=1}^{n} x_{k}$, we have
$$
\begin{aligned}
& A=\left(q_{n}^{2}-1\right) \sum_{k=1}^{n} y_{k} \\
& B=2 q_{n}^{2} \sum_{k=1}^{n} x_{k} y_{k+1}-\sum_{k=1}^{n} y_{k}^{2}=2 q_{n}^{2} \sum_{k=1}^{n} y_{k} y_{k+1}-\left(q_{n}^{2}+1\right) \sum_{k=1}^{n} y_{k}^{2}
\end{aligned}
$$
So it remains to prove that
$$
\left(q_{n}^{2}-1\right)\left(\sum_{k=1}^{n} y_{k}\right)^{2} \geqslant n\left(2 q_{n}^{2} \sum_{k=1}^{n} y_{k} y_{k+1}-\left(q_{n}^{2}+1\right) \sum_{k=1}^{n} y_{k}^{2}\right)
$$
Transform the left-hand side using the relation
$$
\left(\sum_{k=1}^{n} y_{k}\right)^{2}=n \sum_{k=1}^{n} y_{k}^{2}-\sum_{i1$.
Remark. When $n$ tends to infinity, the values $q_{n}$ which are found above tend to 1 .
b)
1.11. (a) Denote $k_{i}:=x_{i+1} / x_{i}$. Then
$$
S=\frac{1}{k_{1}\left(k_{2}+1\right)}+\frac{1}{k_{2}\left(k_{3}+1\right)}+\cdots+\frac{1}{k_{n}\left(k_{1}+1\right)} \geqslant \frac{1}{a_{1}\left(a_{n}+1\right)}+\frac{1}{a_{2}\left(a_{n-1}+1\right)}+\cdots+\frac{1}{a_{n}\left(a_{1}+1\right)}
$$
(b) The inequality holds because
$$
\frac{1}{a_{i}\left(a_{n+1-i}+1\right)}+\frac{1}{a_{n+1-i}\left(a_{i}+1\right)}=\frac{1+\frac{a_{i} a_{n+1-i}-1}{\left(1+a_{i}\right)\left(1+a_{n+1-i}\right)}}{a_{i} a_{n+1-i}} \geqslant b_{i}
$$
where the latter inequality holds because $\left(1+a_{i}\right)\left(1+a_{n+1-i}\right) \geqslant\left(1+\sqrt{a_{i} a_{n+1-i}}\right)^{2}$.
(c) The first inequality $2 S \geqslant g\left(\ln \left(a_{1} a_{n}\right)\right)+g\left(\ln \left(a_{2} a_{n-1}\right)\right)+\cdots+g\left(\ln \left(a_{n} a_{1}\right)\right)$ holds because $g(x)$ is less than both $e^{-x}$ and $2\left(e^{x}+e^{x / 2}\right)^{-1}$. The second inequality holds by the Jensen inequality because $g$ is convex.
(d) $[\mathrm{Dr}]$
2.1. a) $[20]$
For $n=4$ we need to prove that
$$
\left(x_{1}+x_{2}+x_{3}+x_{4}\right)^{2} \geqslant 2 x_{1} x_{2}+2 x_{2} x_{3}+2 x_{3} x_{4}+2 x_{4} x_{1}+4 x_{1} x_{3}+4 x_{2} x_{4}
$$
This follows from the inequality
$$
x_{1}^{2}+x_{2}^{2}+x_{3}^{2}+x_{4}^{2} \geqslant 2 x_{1} x_{3}+2 x_{2} x_{4}
$$
For $n=3$ and $n=5$ re-write the inequality. We need to prove that
$$
(n-1)\left(a_{1}+a_{2}+\ldots+a_{n}\right)^{2} \geqslant 2 n \sum_{i8$. We may suppose that $x_{4} \geqslant x_{1}, x_{4} \geqslant x_{2}$, and $x_{4} \geqslant x_{3}$. For $r=1,2,3$, and 4 denote by $a_{r}$ the sum of all $x_{k}$ such that $k \equiv r(\bmod 4)$ and $k \leqslant n$. Then $x_{1}+x_{2}+\ldots+x_{n}=a_{1}+a_{2}+a_{3}+a_{4}$. Hence by $(7)$
Set
$$
3\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2}=3\left(a_{1}+a_{2}+a_{3}+a_{4}\right)^{2} \geqslant 8\left(a_{1} a_{2}+a_{2} a_{3}+a_{3} a_{4}+a_{4} a_{1}\right) \geqslant 8 \cdot \sum_{(i-k) ; 4} x_{i} x_{k}
$$
$$
A:=\sum_{(i-k) ; 4} x_{i} x_{k} \quad \text { and } \quad B:=\sum_{k=1}^{n} x_{k}\left(x_{k+1}+x_{k+2}+x_{k+3}\right)
$$
We have $A \geqslant B$ because
- for $n \equiv 0(\bmod 4)$ all the summands of $B$ are contained in $A$;
- for $n \equiv 1(\bmod 4)$ the sum $A$ contains all the summands of $B$ except $x_{n} x_{1}$, but $x_{n} x_{1}$ does not exceed $x_{n} x_{4}$;
- for $n \equiv 2(\bmod 4)$ the sum $A$ contains all the summands of $B$ except $x_{n-1} x_{1}$ and $x_{n} x_{2}$, but these summands do not exceed $x_{n-1} x_{4}$ and $x_{n} x_{4}$;
- for $n \equiv 3(\bmod 4)$ the sum $A$ contains all the summands of $B$ except $x_{n-2} x_{1}, x_{n-1} x_{2}$ and $x_{n} x_{3}$, but these summands do not exceed $x_{n-2} x_{4}, x_{n-1} x_{4}$, and $x_{n} x_{4}$.
Hence
$$
3\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2} \geqslant 8 A \geqslant 8 B=8 \sum_{k=1}^{n} x_{k}\left(x_{k+1}+x_{k+2}+x_{k+3}\right)
$$
2.3. a) Cf. [11]. By the Cauchy-Bunyakovsky inequality and Problem 2.2 we have
$$
\frac{x_{1}}{x_{2}+x_{3}+x_{4}}+\frac{x_{2}}{x_{3}+x_{4}+x_{5}}+\ldots+\frac{x_{n-1}}{x_{n}+x_{1}+x_{2}}+\frac{x_{n}}{x_{1}+x_{2}+x_{3}} \geqslant \frac{\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2}}{\sum x_{k}\left(x_{k+1}+x_{k+2}+x_{k+3}\right)} \geqslant \frac{n}{3}
$$
b) ???
2.4. [1, Problem 187]. We may assume that $x_{1} \leqslant x_{2}$. Set
$$
S:=x_{1}+x_{2}+\ldots+x_{n}, \quad S_{1}:=x_{1}+x_{3}+\ldots, \quad S_{2}:=x_{2}+x_{4}+\ldots
$$
Then $S_{1}^{2}+S_{2}^{2} \geqslant\left(S_{1}+S_{2}\right)^{2} / 2=S^{2} / 2$. Hence
$$
\frac{S^{2}}{2} \geqslant S^{2}-S_{1}^{2}-S_{2}^{2}=2 \sum_{(i-k) ; 2} x_{i} x_{k}
$$
If $n$ is even, then the last sum contains all the summands of form $x_{k} x_{k+1}$. If $n$ is odd, then the summand $x_{n} x_{1}$ is missing, however the sum contains a greater summand $x_{n} x_{2}$. So
$$
\frac{S^{2}}{2} \geqslant 2\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n} x_{1}\right)
$$
2.5. See the solution of 1.3 up to the inequality (4).
2.6. Induction on $n \geqslant n=4$. Denote the left-hand side by $L_{n}$. We have
$$
L_{4}=\frac{x_{1}+x_{3}}{x_{2}+x_{4}}+\frac{x_{2}+x_{4}}{x_{1}+x_{3}}=a+a^{-1} \geqslant 2
$$
Let us prove the inductive step. We may assume that $x_{n+1}$ is the minimal of all $x_{i}$ 's. Now remove the last summand from $L_{n+1}$, and then decrease two others. We obtain
$$
L_{n+1} \geqslant \frac{x_{1}}{x_{n+1}+x_{2}}+\ldots+\frac{x_{n}}{x_{n-1}+x_{n+1}} \geqslant \frac{x_{1}}{x_{n}+x_{2}}+\ldots+\frac{x_{n}}{x_{n-1}+x_{n}}=L_{n}
$$
In order to show that the constant 2 is sharp, take
$$
x_{1}=x_{2}=1, \quad x_{3}=t, \quad x_{4}=t^{2}, \quad \ldots, \quad x_{n}=t^{n-2}
$$
When $t \rightarrow+0$, the first two summands tend to 1 and the remaining tends to 0 .
2.7. [10]. Set $S:=x_{1}+x_{2}+\ldots+x_{n}$. Use the Cauchy-Bunyakovsky inequality for sets $\left\{\frac{x_{k}+x_{k+1}}{x_{k}+x_{k+2}}\right\}$ and $\left\{\left(x_{k}+x_{k+1}\right)\left(x_{k}+x_{k+2}\right)\right\}$. We obtain
$$
\frac{x_{1}+x_{2}}{x_{1}+x_{3}}+\frac{x_{2}+x_{3}}{x_{2}+x_{4}}+\ldots+\frac{x_{n-1}+x_{n}}{x_{n-1}+x_{1}}+\frac{x_{n}+x_{1}}{x_{n}+x_{2}} \geqslant \frac{4\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2}}{\sum_{k=1}^{n}\left(x_{k}+x_{k+1}\right)\left(x_{k}+x_{k+2}\right)}
$$
So it suffices to prove that
$$
S^{2} \geqslant \sum_{k=1}^{n}\left(x_{k}+x_{k+1}\right)\left(x_{k}+x_{k+2}\right)=\sum_{k=1}^{n} x_{k}^{2}+2 \sum_{k=1}^{n} x_{k} x_{k+1}+\sum_{k=1}^{n} x_{k} x_{k+2}
$$
This can be shown by opening brackets in the left-hand side, because for $n \geqslant 4$ all the summands $x_{k} x_{k+1}$ and $x_{k} x_{k+2}$, where $k=1,2, \ldots, n$, are different.
In order to show that the constant 4 is sharp, take $x_{k}=a^{k-1}$ for $k=1,2, \ldots, n-1$ and $x_{n}=a^{n-2}$. When $a \rightarrow \infty$, the first $n-3$ summands tend to 0 and the remaining summands tend to 1,2 and 1 .
Using the Cauchy-Bunyakovsky inequality as it is done in the solution of the next problem, the reader will easily find another solution of this problem reducing it to the inequality from Problem 2.4.
2.8. [6]. Use the Cauchy-Bunyakovsky inequality for sets $\left\{\frac{x_{k}}{x_{k-1}+x_{k+2}}\right\}$ nïS $\left\{x_{k}\left(x_{k-1}+x_{k+2}\right)\right\}$. We obtain
$$
\frac{x_{1}}{x_{n}+x_{3}}+\frac{x_{2}}{x_{1}+x_{4}}+\ldots+\frac{x_{n-1}}{x_{n-2}+x_{1}}+\frac{x_{n}}{x_{n-1}+x_{2}} \geqslant \frac{\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2}}{\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n} x_{1}\right)+\left(x_{1} x_{3}+x_{2} x_{4}+\ldots+x_{n} x_{2}\right)}
$$
So it suffices to prove that
$$
S^{2} \geqslant 3\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n} x_{1}\right)+3\left(x_{1} x_{3}+x_{2} x_{4}+\ldots+x_{n} x_{2}\right)=: 3 Y
$$
where $S:=x_{1}+x_{2}+\ldots+x_{n}$. Set
$$
S_{1}:=x_{1}+x_{4}+\ldots, \quad S_{2}=x_{2}+x_{5}+\ldots \quad \text { and } \quad S_{3}=x_{3}+x_{6}+\ldots
$$
Then $S=S_{1}+S_{2}+S_{3}$ and $S_{1}^{2}+S_{2}^{2}+S_{3}^{2} \geqslant S^{2} / 3$. We may assume that $x_{3} \geqslant x_{1}$ and $x_{3} \geqslant x_{2}$. Notice that
$$
S^{2} \geqslant \frac{3}{2}\left(S^{2}-S_{1}^{2}-S_{2}^{2}-S_{3}^{2}\right)=3 \sum_{(i-k) ; 3} x_{i} x_{k}=: 3 Z
$$
- If $n \equiv 0(\bmod 3)$, then all the summands of $Y$ are contained in $Z$.
- If $n \equiv 1(\bmod 3)$, then $Z$ contains all the summands of $Y$ except $x_{n} x_{1}$, but this summand does not exceed $x_{n} x_{3}$.
- If $n \equiv 2(\bmod 3)$, then $Z$ contains all the summands of $Y$ except $x_{n-1} x_{1}$ and $x_{n} x_{2}$, but these summands do not exceed $x_{n-1} x_{3}$ and $x_{n} x_{3}$.
Hence $S^{2} \geqslant 3 Y \geqslant 3 Z$, which proves the initial inequality.
In order to show that the constant 3 is sharp, take $x_{k}=a^{k-1}$ for $k=1,2, \ldots, n-2$ and $x_{n-1}=x_{n}=1$. When $a \rightarrow 0$, the first and the last two summands tend to 1 , while the remaining summands tend to 0 .
2.9. [5]. The inequality is obtained by summing two inequalities of 2.8 (for the direct and the opposite order of variables).
In order to show that the constant 6 is sharp, take $x_{k}=a^{k-1}$ for $k=1,2, \ldots, n-2$ and $x_{n-1}=x_{n}=1$. When $a \rightarrow 0$, the last four summands tend to $1,2,2,1$, respectively; the remaining tend to 0 .
2.10. This is conjectured in [19].
The following proof is due to P.Milošević nïS M. Bukić, participants of the Conference.
This inequality can be represented as sum of two inequalities for $n=2004-$ the inequality from Problem 2.8 and the inequality
$$
\frac{x_{1}}{x_{1}+x_{4}}+\frac{x_{2}}{x_{2}+x_{5}}+\ldots+\frac{x_{n}}{x_{n}+x_{3}} \geqslant 3
$$
Prove the last inequality. For $n=3 m$ it is the sum of three inequalities:
$$
\begin{aligned}
& \frac{x_{1}}{x_{1}+x_{4}}+\frac{x_{4}}{x_{4}+x_{7}}+\ldots+\frac{x_{n-2}}{x_{n-2}+x_{1}} \geqslant 1 \\
& \frac{x_{2}}{x_{2}+x_{5}}+\frac{x_{5}}{x_{5}+x_{8}}+\ldots+\frac{x_{n-1}}{x_{n-1}+x_{2}} \geqslant 1 \\
& \frac{x_{3}}{x_{3}+x_{6}}+\frac{x_{6}}{x_{6}+x_{9}}+\ldots+\frac{x_{n}}{x_{n}+x_{3}} \geqslant 1
\end{aligned}
$$
Each of these inequalities can be re-written as
$$
\frac{1}{1+a_{1}}+\frac{1}{1+a_{3}}+\ldots+\frac{1}{1+a_{m}} \geqslant 1 \quad \text { where } a_{1} a_{2} \ldots a_{m}=1
$$
This can be shown by induction. The base $m=2$ is the following inequality:
$$
\frac{1}{1+a_{1}}+\frac{1}{1+\frac{1}{a_{1}}}=1 \geqslant 1
$$
To prove the induction step, let us check that
$$
\frac{1}{1+b}+\frac{1}{1+c} \geqslant \frac{1}{1+b c}
$$
This can be done directly by reducing to a common denominator and opening brackets.
Here is the proof of A. Khrabrov. Let us prove that
$$
Z:=\frac{x_{1}+x_{2}}{x_{1}+x_{4}}+\frac{x_{2}+x_{3}}{x_{2}+x_{5}}+\ldots+\frac{x_{3 n}+x_{1}}{x_{3 n}+x_{3}} \geqslant 6
$$
Set $x_{3 n+k}:=x_{k}$ and, for $r=0,1,2$,
$$
S_{r}:=\sum_{k=1}^{n} x_{3 k+r}, \quad X_{r}:=\sum_{k=1}^{n} \frac{x_{3 k+r}}{x_{3 k+r}+x_{3 k+3+r}}, \quad \text { and } \quad Y_{r}:=\sum_{k=1}^{n} \frac{x_{3 k+r+1}}{x_{3 k+r}+x_{3 k+3+r}}
$$
First we prove that $X_{r} \geqslant 1$. Consider only the case $r=0$. Then
$$
X_{0} S_{0}^{2} \geqslant X_{0}\left(\sum_{k=1}^{n} x_{3 k}^{2}+\sum_{k=1}^{n} x_{3 k} x_{3 k+3}\right)=X_{0}\left(\sum_{k=1}^{n} x_{3 k}\left(x_{3 k}+x_{3 k+3}\right)\right) \geqslant S_{0}^{2}
$$
where the last inequality holds by the Cauchy-Bunyakovsky inequality. So $X_{0} \geqslant 1$.
Now prove that $Y_{r} \geqslant S_{r+1} / S_{r}$ (we set $S_{3}:=S_{0}$ ). Consider only the case $r=0$.
$$
Y_{0} S_{0} S_{1} \geqslant Y_{0}\left(\sum_{k=1}^{n} x_{3 k} x_{3 k+1}+\sum_{k=1}^{n} x_{3 k+1} x_{3 k+3}\right)=Y_{0}\left(\sum_{k=1}^{n} x_{3 k+1}\left(x_{3 k}+x_{3 k+3}\right)\right) \geqslant S_{1}^{2}
$$
where the last inequality holds by the Cauchy-Bunyakovsky inequality. So $Y_{0} \geqslant S_{1} / S_{0}$.
Summing up all the proved inequalities we obtain
$$
Z=X_{0}+X_{1}+X_{2}+Y_{0}+Y_{1}+Y_{2} \geqslant 3+\frac{S_{1}}{S_{0}}+\frac{S_{2}}{S_{1}}+\frac{S_{0}}{S_{2}} \geqslant 6
$$
In order to show that the constant 6 is sharp, take $x_{1}=x_{2}=x_{3}=1, x_{k}=a^{n-k+1}$ for $k=3,4, \ldots, n$. When $a \rightarrow 0$, the first and the second summands tend to 2 , the third and the last tend to 1 , and the remaining summands tend to 0 .
2.11. This proof is due to A. Khrabrov. Set $S=x_{1}+x_{2}+\ldots+x_{n}$ and $T=\sum_{(i-k) ; 2} x_{i} x_{k}$. By the Cauchy-Bunyakovsky inequality for sets $\left\{\frac{x_{k}}{x_{k-1}+x_{k+3}}\right\}$ and $\left\{x_{k}\left(x_{k-1}+x_{k+3}\right)\right\}$ we have
$$
\frac{x_{1}}{x_{n}+x_{4}}+\frac{x_{2}}{x_{1}+x_{5}}+\ldots+\frac{x_{n-1}}{x_{n-2}+x_{2}}+\frac{x_{n}}{x_{n-1}+x_{3}} \geqslant \frac{\left(x_{1}+x_{2}+\ldots+x_{n}\right)^{2}}{\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n} x_{1}\right)+\left(x_{1} x_{4}+x_{2} x_{5}+\ldots+x_{n} x_{3}\right)}
$$
So it suffices to prove that
$$
S^{2} \geqslant 4\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n} x_{1}\right)+4\left(x_{1} x_{4}+x_{2} x_{5}+\ldots+x_{n} x_{3}\right)
$$
In the solution of problem 2.4 we proved that $S^{2} \geqslant 4 T$, see (8). So it suffices to prove that
$$
T \geqslant\left(x_{1} x_{2}+x_{2} x_{3}+\ldots+x_{n} x_{1}\right)+\left(x_{1} x_{4}+x_{2} x_{5}+\ldots+x_{n} x_{3}\right)
$$
Since $n$ is even, all the summands of the right-hand sum are contained in the left-hand sum.
In order to show that the constant 6 is sharp, take $x_{k}=a^{k-1}$ and $k=1,2, \ldots, n-3$ and $x_{n-2}=x_{n-1}=x_{n}=1$. When $a \rightarrow+0$ the first summand and the three last summands tend to 1 , and the remaining summands tend to 0 .
2.12. [14]. Note that $a^{2}-a b+b^{2} \leqslant \max \{a, b\}^{2}$.
Let $x_{i_{1}}$ be the maximal number of $x_{1}, x_{2}, \ldots, x_{n}$. Let $x_{i_{2}}$ be the maximal number of $x_{i_{1}+1}$ and $x_{i_{1}+2}$. Let $x_{i_{3}}$ be the maximal number of $x_{i_{2}+1}$ and $x_{i_{2}+2}$, and so on. There exists a number $k$ such that $x_{i_{k+1}}=x_{i_{1}}$. Hence
$$
\sum_{k=1}^{n} \frac{x_{k}^{2}}{x_{k+1}^{2}-x_{k+1} x_{k+2}+x_{k+2}^{2}} \geqslant \sum_{j=1}^{k} \frac{x_{i_{j}}^{2}}{x_{i_{j+1}}^{2}} \geqslant k \geqslant\left[\frac{n+1}{2}\right]
$$
where the latter inequality holds because $k \geqslant n / 2$.
In order to show that the constant $\left[\frac{n+1}{2}\right]$ is sharp, take $x_{k}=1$ for odd $k$ and $x_{k}=0$ for even $k$. Then the left-hand side is $\left[\frac{n+1}{2}\right]$.
## References
[1] Василъев Н. Б., Егоров А. А. Задачи Всесоюзных математических олимпиад. М.: Наука, 1988.
[2] Дринфелъд В. Г. Об одном циклическом неравенстве // Мат. заметки. 1971. Т. 9. №2. С. 113-119.
[3] Курляндчик Л. Д., Файбусович А. История одного неравенства // Квант. 1991. №4. С. 14-18.
[4] Толпыго A. К. Тысяча задач Международного математического Турнира городов. М.: МЦНМО, 2009.
[5] Чимэдиэрэн С. Нэгэн орчилт нийлбэр // Математикийн олимпиадын цуврал. 1999. Т. 22. (На монгольск. яз.).
[6] Чимэдиэрэн C., Адбласурен B., Батболд C. Оценка в одной циклической сумме // Монгол улсын их сургууль, Эрдэм шинжилгээний бичиг. 2000. Т. 7 (168). С. 79-84.
[7] Bushell P. J. Shapiro's Cyclic Sum // Bull. London Math. Soc. 1994. Vol. 26. No 6. P. 564-574
[8] Bushell P. J., McLeod J. B. Shapiro's cyclic inequality for even $n$ // J. Inequal. \& Appl., 2002. Vol. 7(3). P. 331-348
[9] Cîrtoaje V. Crux Mathematicorum. 2006. Vol. 32. No. 8. Problem 3195.
[10] Daykin D. E. Inequalities for certain cyclic sums // Proc. Edinburgh Math. Soc. (2) 1970/71. Vol. 17. P. 257-262.
[11] Diananda P. H. Extensions of an inequality of H. S. Shapiro // Amer. Math. Monthly 1959. Vol. 66. P. 489-491.
[12] Diananda P. H. On a conjecture of L. J. Mordell regarding an inequality involoving quadratic forms // J. London Math. Soc. 1961. Vol. 36. P. 185-192.
[13] Diananda P. H. Inequalities for a class of cyclic and other sums // J. London Math. Soc. 1962. Vol. 37. P. 424-431.
[14] Diananda P. H. Some cyclic and other inequalities // Proc. Cambridge Philos. Soc. 1962. Vol. 58. P. 425-427.
[15] Diananda P. H. Some cyclic and other inequalities, II // Proc. Cambridge Philos. Soc. 1962. Vol. 58. P. 703-705.
[16] Diananda P. H. On a cyclic sum // Proc. Glasgow Math. Assoc. 1963. Vol. 6. P. 11-13.
[17] Elbert A. On a cyclic inequality // Period. Math. Hungarica. 1973. Vol. 4. № 2-3. P. 163-168.
[18] Malcolm M. A. A note on a conjecture of L. J. Mordell // Math. Comp. 1971. Vol. 25. P. 375-377.
[19] Mitrinović D. S., Pečarić J. E., Fink A. M. Classical and new inequalities in analysis. Kluwer Academic Publishers Group, Dordrecht, 1993. (Mathematics and its Applications (East European Series), Vol. 61).
[20] Mordell L. J. On the inequality $\sum_{r=1}^{n} \frac{x_{r}}{x_{r+1}+x_{r+2}} \geqslant \frac{n}{2}$ and some others // Abh. Math. Sem. Univ. Hamburg. 1958. Vol. 22. P. 229-240.
[21] Nowosad P. Isoperimetric eigenvalue problems in algebras // Comm. Pure Appl. Math. 1968. Vol. 21. P. 401-465.
[22] Shapiro H. S., Northover F. H. Amer. Math. Monthly. 1956. Vol. 63. № 3. P. 191-192.
[23] Tanahashi K., Tomiyama J. Indecomposable positive maps in matrix algebras // Canad. Math. Bull. 1988. Vol. 31. № 3. P. 308-317.
[24] Troesch B. A. Full solution of Shapiro's cyclic inequality // Notices Amer. Math. Soc. 1985. Vol. 39. № 4. P. 318.
[25] Vukmirović J. A note on an inequality for the cyclic sums introduced by D. E. Daykin // Math. Balk. 1978. Vol. 8. P. $293-297$.
[26] Yamagami S. Cyclic inequalities // Proc. Amer. Math. Soc. 1993. Vol. 118. № 2. P. 521-527.
[27] Zulauf A. Note on a conjecture of L. J. Mordell // Abh. Math. Sem. Univ. Hamburg. 1958. Vol. 22. P. 240-241.
[28] Zulauf A. Note on an Inequality // Math. Gazette. 1962. Vol. 46. №355. P. 41-42.
## ТРОПИЧЕСКАЯ ГЕОМЕТРИЯ
## А. Заславский, Ф. Нилов, А. Скопенков, М. Скопенков
## Краткий обзор. ${ }^{1}$
16-я проблема Гильберта спрашивает, каким может быть количество и взаимное расположение кривых, образующих подмножество плоскости, заданное уравнением $\sum_{i+j \leq d} a_{i j} x^{i} y^{j}=0$. Более аккуратная формулировка и примеры приводятся в части А. ${ }^{2}$ Цель данного цикла задач - частичное решение 16 -й проблемы Гильберта для $d=6$, а именно, построение требуемых подмножеств в наиболее содержательном случае (см. Основную Теорему ниже).
Для подмножества плоскости, заданного многочленом от двух переменных с некоторыми конкретными коэффициентами, определить число и расположение кривых не так-то просто (даже вооружившись современным компьютером). При решении задач части В Вы нащупаете формулировку основной леммы, которая позволит легко сделать это для многочленов некоторого специального вида. Вы увидите, как при рисовании подмножеств, заданных уравнениями вида $\sum_{i+j \leq d}\left(a_{i j} x^{i} y^{j}\right)^{N}=0$, где $N$ - большое нечетное число, естественно возникает тропическая геометрия. Используя ее, Вы сможете строить такие подмножества с различным расположением овалов.
Исходные соображения тропической геометрии элементарны. Заменим умножение на сложение, а сложение - на операцию, связанную со сложением таким же дистрибутивным законом, каким сложение связано с умножением. В качестве такой операции можно взять максимум $\max \{a, b\}$ пары чисел $a$ и $b$. При такой замене функция $\sum_{i+j \leq d} b_{i j} x^{i} y^{j}$ перейдет в функцию вида (проверьте!): $f(x, y)=\max _{i+j \leq d}\left(i x+j y+b_{i j}\right)$. Множество точек излома такой функции называется тропической кривой. ${ }^{3}$
Например, прямая на плоскости задается уравнением $A x+B y+C=0$.

Рис. 1: Левая часть данного уравнения при нашей замене переходит в функцию вида $f(x, y)=\max \{x+a, y+b, c\}$. Множество точек излома полученной функции $f(x, y)$ имеет вид, показанный на рисунке 1 (проверьте!). Так определяется тропическая прямая. Для тропических прямых сохраняются многие свойства обычных прямых. "Экспериментальной" проверке этих свойств посвящена часть С.
## А. Примеры алгебраических кривых.
Многочленом (от двух переменных) называется функция $F: \mathbb{R}^{2} \rightarrow \mathbb{R}$, для которой существуют числа $d$ и $a_{i j}, 0 \leq i, j \leq d$, такие что $F(x, y)=\sum_{i+j \leq d} a_{i j} x^{i} y^{j}$. Вы можете пользоваться без доказательства следующим фактом: для данной функции $F$ такие числа единственны с точностью до увеличения $d$ и выбора новых $a_{i j}$ равными нулю.
Множеством нулей многочлена $F$ называется множество $F^{-1}(0):=\left\{(x, y) \in \mathbb{R}^{2} \mid F(x, y)=0\right\}$.
A1. Однозначно ли множество $F^{-1}(0)$ определяет многочлен $F$ ?
A2. Какие из следующих множеств являются множествами нулей многочленов?
(a) прямая;
(b) окружность;
(с) точка;
(d) отрезок;
(е) объединение двух прямых;
(f) "хрюшка" (объединение 6 окружностей) на рисунке 2.[^7]
Степень многочлена - это наименьшее возможное число $d$, для которого найдутся требуемые $a_{i j}$. (Комментарий. Степень - это наибольшее возможное $d$, для которого $a_{i, d-i} \neq 0$ для некоторого представления многочлена и некоторого $i$.)
A3. (а) Сколько точек может быть в пересечении прямой с множеством нулей многочлена степени $d$ ?
(b) Множество нулей многочлена нечетной степени неограничено (то есть, не содержится ни в каком в диске).
Многочлен $F$ приводим, если $F=G \cdot H$ для некоторых многочленов $G$ и $H$.
Кривые $^{4}$. Функция $\gamma:[a, b] \rightarrow \mathbb{R}$ называется дифференцируемой в точке $t_{0}$, если для некоторого числа $A$ и любого $\varepsilon>0$ существует $\delta$, такое, что для любого
$$
t \in\left(t_{0}-\delta, t_{0}+\delta\right) \quad \text { выполнено } \quad\left|\gamma(t)-\gamma\left(t_{0}\right)-A\left(t-t_{0}\right)\right|<\varepsilon\left|t-t_{0}\right|
$$
Отображение $\gamma:[a, b] \rightarrow \mathbb{R}^{2}$ можно рассматривать как упорядоченную пару функций $\gamma_{1}, \gamma_{2}:[a, b] \rightarrow \mathbb{R}$. Дифференцируемость отображения $\gamma=\left(\gamma_{1}, \gamma_{2}\right)$ равносильна дифференцируемости функций $\gamma_{1}$ и $\gamma_{2}$.
(Гладкой) кривой на плоскости называется дифференцируемое отображение $\gamma:[a, b] \rightarrow \mathbb{R}^{2}$ (или $\gamma$ : $\left.\mathbb{R} \rightarrow \mathbb{R}^{2}\right)$. Кривая $\gamma:[a, b] \rightarrow \mathbb{R}^{2}$ называется замкнутой, если $\gamma(a)=\gamma(b)$.
В задачах A4.cfg и А7 достаточно привести пример многочлена; доказательство его свойств не требуется.
A4. (а) Существует неприводимый многочлен степени 3 , множество нулей которого содержит замкнутую кривую.
(b) То же для степени 4.
(c) Существует неприводимый многочлен степени 4, множество нулей которого содержит две замкнутых кривые, одну внутри другой.
(d) Если множество нулей многочлена степени 4 содержит две замкнутых кривые, одну внутри другой, то это множество не содержит никаких других точек.
(e) Верен ли аналог утверждения (d) для неприводимого многочлена степени 5 ?
(f) Существует многочлен степени 4, множество нулей которого содержит 4 замкнутые кривые.

Рис. 2:
(g) Существует многочлен степени 4 , множество нулей которого содержит 3 замкнутые кривые.
Овалы . Компоненты связности, на которые распадается множество нулей многочлена, называются ветвями. (Наличие неограниченных ветвей усложняет изучение множеств нулей многочленов.) Для неограниченной ветви $B$ прямые, соединяющие начало координат $O$ с точками на ветви $B$, стремящимися к "бесконечности", стремятся к некоторой "предельной" прямой. Две неограниченные ветви элементарно эквивалентны, если их "предельные" прямые совпадают.
A5. Неограниченные ветви гиперболы $x y=1$ элементарно эквивалентны.
Две неограниченные ветви эквивалентны, если существует соединяющая их цепочка ветвей, в которой каждые две соседние ветви элементарно эквивалентны. Множество нулей многочлена невырождено, если все его ветви - гладкие кривые. Овал невырожденного множества нулей многочлена - это либо замкнутая кривая (содержащаяся в этом множестве), либо класс эквивалентности неограниченных ветвей. (Заметим, что это определение отлично от "правильного", приводящегося в учебниках.)
A6. Найдите все $h$, при которых множество нулей невырождено, и посчитайте количество овалов для многочлена
(a) $x y(x+y-1)-h$.
(b) $x^{3}-x+h-y^{2}$. (ответ может зависеть от $h$ ).
A7. Существует многочлен степени 5 , множество нулей которого невырождено и состоит из 7 овалов.
16-я проблема Гильберта. Каким может быть количество и взаимное расположение овалов у невырожденного множества нулей многочлена степени d?
Мы не приводим формального определения "расположения" овалов. Такое определение потребовало бы понятие проективизации многочлена.[^8]
Основная теорема. (а) Существует многочлен степени 6, множество нулей которого невырождено и состоит из 11 овалов.
(b) Существуют три многочлена степени 6, множество нулей каждого из которых невырождено и состоит из 11 овалов, а расположение овалов для всех трех многочленов различно.
## В. Тропическая кривая как предел алгебраических.
B1. Нарисуйте множество нулей многочлена
(a) $x-y-1$;
(a') $x^{1001}-y^{1001}-1$
(b) $x+y-1$;
(b') $x^{1001}+y^{1001}-1$;
$\begin{array}{ll}\text { (c) } x y=x+y ; & \text { (c') } x^{1001} y^{1001}=x^{1001}+y^{1001} \text {; }\end{array}$
$\begin{array}{ll}\text { (d) } x^{2}+y^{2}-4 x-4 y-2=0 ; & \text { (d') } x^{2002}+y^{2002}-4^{1001} x^{1001}-4^{1001} y^{1001}-2^{1001} \text {; }\end{array}$
(e') $x^{3003}+2^{1001} x^{1001} y^{2002}-3^{1001} x^{1001} y^{1001}+y^{2002}-x^{1001}-2^{1001}$.
Обозначим через
$$
F_{N}(x, y)=\sum_{i+j \leq d}\left(a_{i j} x^{i} y^{j}\right)^{N}
$$
семейство многочленов, зависящее от нечетного числа $N \geq 1$.
При замене переменных $u=x^{N}, v=y^{N}$ каждый многочлен $F_{N}$ переходит в многочлен $\sum_{i+j \leq d} a_{i j}^{N} u^{i} v^{j}$ степени $d$. Поэтому для решения 16 -й проблемы Гильберта полезно научиться определять число и расположения овалов для множества $F_{N}^{-1}(0)$.
B2. Количество овалов для $F_{N}^{-1}(0)$ может отличаться от количества овалов у $F_{1}^{-1}(0)$.
Обозначим через $B_{R}$ круг радиуса $R$ с центром в начале координат 0 .
B3. Для любых $\varepsilon, R>0$ найдется $N_{0}>0$, такое что для всех нечетных $N>N_{0}$ пересечение множества нулей многочлена
(a) $x^{2 N}-x^{N}-y^{N}$ с кругом $B_{R}$ содержится в $\varepsilon$-окрестности объединения прямых $x=0, x= \pm 1$, $x= \pm y$ и парабол $y= \pm x^{2}$.
(b) $F$ с кругом $B_{R}$ содержится в $\varepsilon$-окрестности объединения множеств нулей всевозможных многочленов $a_{i j} x^{i} y^{j} \pm a_{k l} x^{k} y^{l}$, в которых $(i, j) \neq(k, l), i+j \leq d, k+l \leq d$.
Обозначим $\mathbb{R}_{+}:=(0,+\infty)$ множество положительных чисел и через $\mathbb{R}_{+}^{2}:=(0,+\infty)^{2}$ угол, заданный неравенствами $x>0, y>0$. Определим отображение $L O G: \mathbb{R}_{+}^{2} \rightarrow \mathbb{R}^{2}$ bу $\operatorname{LOG}(x, y)=$ $\left(\log _{2} x, \log _{2} y\right)$.
B4. (abcde) Нарисуйте $L O G$-образ (то есть, образ при отображении $L O G$ ) пересечения множеств нулей многочленов ( $\mathrm{a}^{\prime} \mathrm{b}^{\prime} \mathrm{c}^{\prime} \mathrm{d}^{\prime} \mathrm{e}$ ') из задачи $\mathrm{B1}$ с углом $\mathbb{R}_{+}^{2}$.
Тропическим многочленом называется функция
$$
f(x, y):=\max _{i+j \leq d}\left(i x+j y+b_{i j}\right)
$$
Обозначим множества $f^{p q}=\left\{(x, y) \in \mathbb{R}^{2} \mid f(x, y)=p x+q y+b_{p q}\right\}$. Некоторые пары таких множеств пересекаются друг с другом (по границе). Объединение всех таких попарных пересечений называется тропической кривой (степени $d$ ). (Это в точности множество точек, в которых график функции $f(x, y)$ имеет "излом".)
Далее будем считать, что $a_{i j} \neq 0$ при всех $i, j$, удовлетворяющих неравенству $i+j \leq d$. Тропическая кривая соответствует семейству многочленов $F_{N}$, если $b_{i j}=\log _{2}\left|a_{i j}\right|$. Это определение мотивировано важной задачей $\mathrm{B} 6 \mathrm{~b}$ ниже.
B5. Нарисуйте тропическую кривую, соответствующую семейству многочленов
(a) $(a x)^{N}+(b y)^{N}+c^{N}$ ?
(b) $\left(a x^{2}\right)^{N}+(2 b x y)^{N}+\left(c y^{2}\right)^{N}$ ?
(Ответ может зависеть от $a, b, c$.)
Обозначим через $\Delta_{R}$ треугольник, заданный неравенствами $x \geq-R, y \geq-R, x+y \leq R$.
B6. (а) Для любых $\varepsilon, R>0$ найдется $N_{0}$, такое что для всех нечетных $N>N_{0}$ пересечение $L O G$-образа множества нулей многочлена $x^{2 N}-x^{N}-y^{N}$ с треугольником $\Delta_{R}$ содержится в $\varepsilon$ окрестности объединения луча $y=2 x, x \geq 0$ и луча $x=0, y \geq 0$.
(b) Для любых чисел $\left\{a_{i j}\right\}_{i+j \leq d}$ и $\varepsilon, R>0$ найдется $N_{0}>0$, такое что для всех нечетных $N>N_{0}$ множество $\operatorname{LOG}\left(F_{N}^{-1}(0)\right) \cap \Delta_{R}$ содержится в $\varepsilon$-окрестности тропической кривой, соответствующей семейству $F_{N}$.
## С. Тропические прямые и окружности.
Эта часть цикла задач представляет собой художественный конкурс: предлагается экспериментально проверить некоторые теоремы тропической геометрии путем рисования аккуратных чертежей. Избранные рисунки будут выставлены для всеобщего обозрения. "Задачи" С1-С10 не оцениваются (хотя точные формулировки и доказательства каких-нибудь из этих утверждений будут награждаться "плюсиками"). Часть "задачи" можно пропустить, если упоминаемая теорема евклидовой геометрии Вам неизвестна. Вся часть $С$ не нужна для решения 16-й проблемь Гильберта и может быть опущена.
Рассмотрим плоскость с фиксированной системой координат. Тропической прямой ("лапой") называется объединение трех лучей на плоскости, выходящих из одной точки (называемой вершиной), один из которых направлен строго влево, другой строго вниз, а третий (называемый диагональным) - вправовверх под углом ровно $45^{\circ}$.
C1. Существуют две различные тропические прямые, пересекающиеся в двух различных точках.
Будем говорить, что две точки находятся в общем положении, если евклидова прямая, проходящая через эти точки, не параллельна координатным осям и прямой $x=y$.
C2. (а) Через две точки общего положения проходит единственная тропическая прямая.
(b) Если вершины двух тропических прямых находятся в общем положении, то эти тропические прямые имеют единственную общую точку.
Будем говорить, что две тропические прямые параллельны, если вершина одной из них лежит на диагональном луче другой.
C3. Пусть точка $A$ и вершина тропической прямой $b$ находятся в общем положении. Тогда через точку $A$ можно провести единственную тропическую прямую, параллельную $b$.
Будем говорить, что две тропические прямые перпендикулярны, если прямые, которые содержат их диагональные лучи, симметричны относительно прямой $x=y$.
C4. Пусть точка $A$ и вершина тропической прямой $b$ находятся в общем положении. Тогда через точку $A$ можно провести единственную тропическую прямую, перпендикулярную $b$.
Тропический треугольник ("паук") - это объединение трех тропических прямых, вершины которых находятся (попарно) в общем положении.
C5. Нарисуйте чертежи к тропическим аналогам следующих теорем.
(a) Высоты треугольника пересекаются в одной точке.
(b) Теорема Паппа.
(c) Теорема Дезарга.
Пусть даны две точки $A$ и $B$. Тропической окружноствю ("цаплей") назовем ГМТ $X$ таких, что существуют две перпендикулярные тропических прямые, одна из которых проходит через $A$ и $X$, а другая - через $B$ и $X$. (Не забывайте, что через две точки $A$ и $X$, вообще говоря, можно провести несколько тропических прямых!)
C6. (а) Нарисуйте тропическую окружность. Как зависит это множество от расположения точек $A$ и $B$ ?
(b) У любого ли треугольника существует описанная окружность?
(c) Теорема Паскаля.
C7*. Придумайте понятие середины отрезка в тропической геометрии, так чтобы выполнялась теорема о медианах треугольника.
## ТРОПИЧЕСКАЯ ГЕОМЕТРИЯ
## А. Заславский, Ф. Нилов, А. Скопенков, М. Скопенков
Основная серия задач состоит из двух частей - окончания части $\mathrm{B}$ и новой части $\mathrm{D}$. Задачи части $\mathrm{D}$ не используют (за исключением явно оговоренных случаев) понятий и результатов других частей проекта. Поэтому их можно решать, не принимая участия в других частях проекта.
## В. Теорема Виро о склейке.
B3. Для любых $\varepsilon, R>0$ найдется $N_{0}>0$, такое что для всех нечетных $N>N_{0}$ пересечение множества нулей многочлена
(c) $x^{2 N}-x^{N}-y^{N}$ с кругом $B_{R}$ и с первой четвертью $(x>0, y>0)$ содержится в $\varepsilon$-окрестности объединения множеств
$$
\{(1, y) \mid 0 \leq y \leq 1\}, \quad 0 \leq x=y \leq 1 \quad \text { и } \quad y=x^{2} \geq 1
$$
(d) $x^{2 N}-x^{N}-y^{N}$ с кругом $B_{R}$ и со второй четвертью $(x<0, y>0)$ содержится в $\varepsilon$-окрестности множества, симметричного объединению из (с) относительно оси $O y$.
B7. Сформулируйте и докажите аналог пункта В3d для третьей и четвертой четвертей.
B8. Пересечение множества нулей многочлена $x^{2 N}-x^{N}-y^{N}$ с третьей четвертью пусто.
B9. Для любых $\varepsilon, R>0$ найдется $N_{0}>0$, такое что для всех нечетных $N>N_{0}$ пересечение множества нулей многочлена $x^{2 N}-x^{N}-y^{N}$ с кругом $B_{R}$ и
(a) первой четвертью содержится в $\varepsilon$-окрестности объединения множеств $\{(1, y) \mid 0 \leq y \leq 1\}$ и $y=$ $x^{2} \geq 1$.
(b) второй четвертью содержится в $\varepsilon$-окрестности объединения множеств $0 \leq-x=y \leq 1$ и $y=x^{2} \geq 1$.
B10. Сформулируйте и докажите аналог задачи В9 для четвертой четверти.
Сформулируем теорему Виро о склейке, которая позволяет найти число и взаимное расположение овалов для некоторых алгебраических кривых.
B11. Каждая тропическая кривая является конечным объединением отрезков и лучей.
Определение кривой Виро и ее овалов. Возьмем тропическую кривую, соответствующую набору числел $a_{i j} \neq 0$. Тропическая кривая является конечным объединением ребер (отрезков и лучей), которые пересекаются в вершинах (т.е. в общих точках ребер). Гранъю тропической кривой называется компонента связности ее дополнения в плоскости. Каждой грани соответствует пара $(p, q)$ таких чисел, что $p x+q y+\log _{2}\left|a_{p q}\right|=\max _{i+j \leq d, a_{i j} \neq 0}\left(i x+j y+\log _{2}\left|a_{i j}\right|\right)$ для точек $(x, y)$ этой грани, а также соответствует знак коэффициента $a_{p q}$. B этом определении мы используем не $\left\{a_{i j}\right\}$, а тропическую кривую, на гранях которой расставлены пары чисел и знаки ${ }^{5}$.
Перенесем тропическую кривую параллельно, чтобы ее вершины оказались в угле $x>0, y>0$. Обозначим через $U_{p, q .00}$ образ при этом переносе грани тропической кривой, на которой стоит пара $(p, q)$. Обозначим $U_{p, q, 01}, U_{p, q, 10}$ и $U_{p, q, 11}$ - образы множества $U_{p, q, 00}$ при симметриях относительно оси $x$, оси $y$ и $(0,0)$, соответственно. Продолжим расстановку знаков с первой координатной четверти на всю плоскость по следующему правилу: при отражении области $U_{p, q}$ относительно оси $O x$ знак области умножается на $(-1)^{q}$, а при отражении относительно оси $O y-$ на $(-1)^{p}$. (То есть, $\operatorname{sgn} U_{p, q, s t}=(-1)^{s p+t q} \operatorname{sgn} U_{p, q, 00}$.) Определим кривую Виро как объединение $\cup\left\{U_{\alpha} \cap U_{\beta} \mid \operatorname{sgn} U_{\alpha} \neq \operatorname{sgn} U_{\beta}\right\}$ тех ребер тропической кривой, которые разделяют грани разных знаков. Две неограниченные компоненты связности кривой Виро называются
- элементарно эквивалентны, если они содержат лучи, симметричные относительно точки $(0,0)$.
- эквивалентны если существует последовательность компонент от одной к другой, в которой любые две последовательные компоненты элементарно эквивалентны.
Овалом кривой Виро называется либо замкнутая ломаная, содержащаяся в кривой Виро, или класс эквивалентности компонент связности.
Следующей теоремой разрешается пользоваться без доказательства:
Теорема Харнака. Невырожденное множество нулей многочлена степени $d$ не может содержать больше $\frac{(d-1)(d-2)}{2}+1$ овалов.
В12.* Теорема Виро о склейке. Пусть кривая Виро, построенная по набору чисел $a_{i j} \neq 0$, содержит ровно $\frac{(d-1)(d-2)}{2}+1$ овалов. Тогда существует такое $N$, что множество нулей многочлена[^9]$\sum_{i+j \leq d} a_{i j}^{N} u^{i} v^{j}$ невырождено, а число и взаимное расположение его овалов такое же, как у соответствующей кривой Bиро.
## D. Построение примеров в 16 -й проблеме Гильберта.
Цель задач серии $\mathrm{D}$ - научиться описывать тропические кривые на чисто комбинаторном языке, и тем самым получить комбинаторный метод построения примеров кривых в $16-$ й проблеме Гильберта.
Напомним, что тропическая кривая степени $d$ - это множество точек излома графика функции $\max _{i+j \leq d}\left\{i x+j y+b_{i j}\right\}$ (подробнее см. выше, абзац после задачи В4).
D1. (а) Проверьте, что тропическая кривая степени 1 выглядит так, как показано на рисунке 1. (Сравните с определением тропической прямой в части C).
(b) Из каждой вершины тропической кривой выходит как минимум 3 ребра.
Припишем каждому ребру тропической кривой кратность по следующему правилу. Предположим, что в одной из областей, граничащих с этим ребром, максимальной является величина $i x+j y+b_{i j}$, а другой - величина $i^{\prime} x+j^{\prime} y+b_{i^{\prime} j^{\prime}}$. Тогда прямая, содержащая данный отрезок, задается уравнением

чисел $i-i^{\prime}$ и $j-j^{\prime}$.
Будем обозначать на рисунках кратные ребра тропической кривой двойными (тройными и т.д.) линиями.
D2. Тропические кривые степени $d$ обладают следующими свойствами:
(a) Наклон каждого ребра рационален.
(b) В каждой вершине выполняется следующее условие сбалансированности. Обозначим через $v_{i}$ вектор с началом в данной вершине, имеющий направление $i$-го ребра, выходящего из вершины, и равный кратчайшему целочисленному вектору с данным направлением, умноженному на кратность ребра. Тогда $\sum v_{i}=0$ для каждой вершины.
(c) Имеется $3 d$ бесконечных рёбер, взятых с учетом кратностей, $d$ из которых направлены строго влево, $d$ направлены строго вниз, и $d$ направлены вправо-вверх с углом наклона $45^{\circ}$.
D3. (а) Тропический многочлен $\max _{i+j \leq d}\left\{i x+j y+b_{i j}\right\}$ восстанавливается по своей тропической кривой однозначно с точностью до добавления постоянной.
(b) Всякий граф на плоскости с прямыми ребрами и предписанными кратностями, удовлетворяющий свойствам (a), (b) и (с) задачи D2, является тропической кривой степени $d$.
Мы говорим, что две тропические кривые имеют одинаковую конфигурацию, если у них совпадает комбинаторный тип графа и наклон его рёбер (но не обязательно их длины и положение).
D4. Нарисуйте 5 различных конфигураций тропических кривых второй степени.
Для решения следующих задач достаточно прочитать в предыдущем пункте абзац "Определение кривой Виро и ее овалов".
D5. Из какого максимального числа овалов может состоять кривая Виро при $d=$
(a) 2 ; (b) 3 ; (c) 4 ; (d) 5? (Доказательства максимальности мы не требуем. Сравните ответ с задачами A4f и А7.)
D6*. Напишите программу на компьютере, которая:
(a) рисует все конфигурации тропических кривых данной степени $d$;
(b) по данной конфигурации тропической кривой и набору знаков, приписанных областям $U_{i j}$ дополнения к ней, определяет количество овалов у кривой Виро.
D7*. (ab) Докажите Основную теорему (разрешается пользоваться теоремой Виро о склейке без доказательства).
## РЕШЕНИЯ ЗАДАЧ
A1. Ответ: нет. Например, прямая $x=0$ является множеством нулей разных многочленов $F(x, y)=x$ и $G(x, y)=x^{2}$.
A2. Ответ: a, b, c, e, f.
Примеры. (а) Любая прямая на плоскости задается уравнением $A x+B y+C=0$ с некоторыми числами $A, B, C$.
(b) Уравнение окружности: $\left(x-x_{0}\right)^{2}+\left(y-y_{0}\right)^{2}-R^{2}=0$, где $\left(x_{0}, y_{0}\right)$ - координаты центра, $R$ - paдиус.
(c) Уравнение точки $\left(x_{0}, y_{0}\right):\left(x-x_{0}\right)^{2}+\left(y-y_{0}\right)^{2}=0$.
(е) Уравнение объединения двух прямых: $(A x+B y+C)(a x+b y+c)=0$, где $A x+B y+C=0-$ уравнение первой, $a x+b y+c=0$ - уравнение второй прямой.
(f) Уравнение объединения 6 окружностей: $\prod_{k=0}^{6}\left(\left(x-x_{k}\right)^{2}+\left(y-y_{k}\right)^{2}-R_{k}^{2}\right)=0$, где $\left(x-x_{k}\right)^{2}+(y-$ $\left.y_{k}\right)^{2}-R_{k}^{2}=0$ - уравнение $k$-й окружности.
Невозможность в пункте (d) напрямую следует из задачи АЗа.
A3. (а) Ответ: либо от 0 до $d$, либо прямая содержится в нашем множестве нулей.
Параметризуем прямую $l: x=x_{0}+\alpha \cdot t, y=y_{0}+\beta \cdot t$. Подставляя эти выражения в многочлен, получим многочлен $P(t)$ степени не более, чем $d$. Многочлен $P(t)$ либо имеет не более, чем $d$ вещественных корней, либо тождественно равен 0 .
Покажем, что для любого $d^{\prime}0, f(1)<0, f(2)>0$, то два корня $c_{1}$ и $c_{2}$ уравнения $f(c)=0$ лежат на интервале $(0,1)$, а третий корень лежит на интервале $(1,2)$. Поэтому $D(c)=0$ ровно в двух точках $c_{1}$ и $c_{2}$ интервала $(0,1)$, причём $D(c)>0$ для любого $c \in\left(c_{1}, c_{2}\right)$ и $D(c)<0$ для остальных точек этого интервала. (Для определённости считаем, что $c_{1}1 / 27$ - один овал. При $h=0$ и $h=1 / 27$ алгебраическая кривая вырождена. Приведем решение.
Обозначим $f(x, y):=x y(x+y-1)+h$. Введём обозначения для точек пересечения прямых $O x, O y$ и $x+y-1=0$ и областей, на которые эти прямые разбивают плоскость:
$$
A:=(1,0), \quad B:=(0,1)
$$
$C:=(0,0), X:=\{(x, y) \mid x>0, y>0, x+y<1\}, X_{A}:=y<0, x+y>1, X_{B}:=x<0, x+y>1$, $X_{C}:=x<0, y<0, Y_{A}:=x<0, y>0, x+y<1, Y_{B}:=x>0, y<0, x+y<1, Y_{C}:=x>0, y>0, x+y>1$.
Ясно, что $f(x, y)=h$ в точках прямых $O x, O y$ и $x+y-1=0, f(x, y)h$ в точках областей $Y_{A}, Y_{B}$ и $Y_{C}$. Поэтому при $h>0$ нули многочлена $f(x, y)$ могут лежать только в $X_{A}, X_{B}, X_{C}$ и $X$, при $h<0$ они могут лежать только в $Y_{A}, Y_{B}$ и $Y_{C}$.
Пусть $h<0$. Обозначим $y_{A}:=Y_{A} \cap f^{-1}(0)$. Аналогично определим $y_{B}$ и $y_{C}$.
Докажем, что $y_{A}$ является связной компонентой множества $f^{-1}(0)$ нулей $f$. Координаты $x$ точек пересечения прямых $y=c$ с $f^{-1}(0)$ удовлетворяют уравнению $x^{2}+(c-1) x+\frac{h}{c}=0$. Дискриминант $D=D(c)$ этого уравнения равен $(c-1)^{2}-\frac{4 h}{c}$. Поскольку $h<0$, то для любого $c \in R_{+}, D(c)>0$. Значит, каждая из прямых $y=c$, где $c \in R_{+}$, пересекает $F$ ровно в двух точках $\left(x_{1,2}(c), c\right)$, таких, что $x_{1,2}(c)=\frac{ \pm \sqrt{D}-(c-1)}{2}$. Определим
$$
\gamma: R_{+} \rightarrow \mathbb{R}^{2} \quad \text { формулой } \quad\left\{\left(x_{2}(t), t\right) \quad t \in R_{+}\right.
$$
Так как функция $x_{2}(c)$ гладкая, то отображение $\gamma(t)$ гладкое. Поскольку $\gamma\left(R_{+}\right)=y_{A}$, то $y_{A}$ является связной компонентой множества $f^{-1}(0)$ нулей $f$. Аналогично $y_{B}$ и $y_{C}$ являются связными компонентами множества $f^{-1}(0)$ нулей $f$.
Нетрудно проверить, что направление прямой $O x$ является предельным для ветви $y_{C}$.
Аналогично, это направление является предельным для ветви $y_{A}$. Поэтому ветви $y_{A}$ и $y_{C}$ элементарно эквивалентны. Аналогично, ветви $y_{A}$ и $y_{B}$ элементарно эквивалентны, поскольку направление прямой $x+y-1=0$ является для этих ветвей предельным. Значит, ветви $y_{A}, y_{B}$ и $y_{C}$ эквивалентны и поэтому образуют один овал.
При $h=0$ алгебраическая кривая $f$ вырождена.
Пусть $h>0$. Введём обозначения
$$
x:=X \cap f^{-1}(0), \quad x_{A}:=X_{A} \cap f^{-1}(0), \quad x_{B}:=X_{B} \cap f^{-1}(0) \quad \text { и } \quad x_{C}:=X_{C} \cap f^{-1}(0)
$$
Связность и эквивалентность $x_{A}, x_{B}$ и $x_{C}$ устанавливается аналогично случаю $h<0$. Поэтому они образуют один овал при любом $h>0$. Если множество $x$ не пусто и не представляет собой одну точку, то оно является овалом (доказательство аналогично решению задачи А4а).
Покажем, что множество точек $x$ не пусто только при $h \in\left(0, \frac{1}{27}\right]$. Ясно, что $x$ пусто тогда и только тогда, когда $D(c)<0$ при любом $c \in(0,1)$. Производная $D^{\prime}(c)>0$ при $c \in(0,1 / 3), D^{\prime}(c)=0$ при $c=1 / 3$, $D^{\prime}(c)<0$ при $c \in(1 / 3,1)$. Поэтому в точке $c=1 / 3$ достигается максимум функции $D(c)$ на интервале $(0,1)$. Значит, $D(c)<0$ при любом $c \in(0,1)$ тогда и только тогда, когда $D(1 / 3)=4 / 9-\frac{4 h}{c}<0$, т.е. $h>1 / 27$. При $h=1 / 27$ множество $x$ состоит из одной точки. Таким образом, при $h \in(0,1 / 27)$ множество нулей $f^{-1}(0)$ состоит из двух овалов, при $h=1 / 27$ алгебраическая кривая вырождена, при $h>1 / 27$ множество $f^{-1}(0)$ состоит из одного овала.
(b) Ответ: Один овал при $h \in\left(-\frac{2}{3 \sqrt{3}}, \frac{2}{3 \sqrt{3}}\right)$, два овала при $h \in\left(-\infty,-\frac{2}{3 \sqrt{3}}\right)$ и $h \in\left(\frac{2}{3 \sqrt{3}}, \infty\right)$, алгебраическая кривая $x^{3}-x+h-y^{2}$ вырождена при $h= \pm \frac{2}{3 \sqrt{3}}$. Указание. Аналогично (а).
A7. Указание. Рассмотрите многочлен $x\left((x-1)^{2}+y^{2}-2\right)\left((x+1)^{2}+y^{2}-2\right)+\frac{1}{100}$.
B1. (a') Указание. См. рисунок 3.а'. Обоснуем рисунок. Ясно, что множество нулей многочлена $x^{2001}-$ $y^{2001}-1$ лежит "ниже" прямой $y=x$ и симметрично относительно прямой $x+y=0$. Поэтому мы можем рассматривать только случай $y>-x$. Точки $(0,-1)$ и $(1,0)$ являются точками пересечения нашего множества с осями координат. Если $x=1+\epsilon$, где $\epsilon>0$, то для $y$ выполнено неравенство (1001є) $\frac{1}{1001}1 / 1001$ значения $x$ и $y$ приблизительно равны. Если $1-\epsilonx$. При $x \in[0,1000 / 1001]$ значение $y$ приблизительно равно $x$. При $x \in[1000 / 1001,1]$ значение $y$ изменяется от -1 до $-\infty$. При $x \in(1,1002 / 1001)$ выполнено $y>x$. При $x \geqslant 1002 / 1001$ значение $y$ приблизительно равно 1.
(d') Указание. См. рисунок 3.d'. Ясно, что множество нулей многочлена $x^{2002}+y^{2002}-4^{1001} x^{1001}-$ $4^{1001} y^{1001}-2^{1001}$ симметрично относительно прямой $y=x$. Поэтому можно рассматривать только случай $x \geqslant y$. Точек с координатой $x \geqslant 4$ среди множества нулей нашего многочлена нет. При $x \in(-1 / 2,1 / 2]$ значение $y$ приблизительно равно $-1 / 2$. При $x \in(1 / 2,4003 / 1001]$ значение $y$ приблизительно равно $-x$. При $x \in(4003 / 1001,4)$ значение $y$ изменяется от -4 до 4 .


a

$a^{\prime}$

b

b'

c

c'

d

d'
Рис. 3:
B2. Пусть $F(x, y)=x^{3}-p x+q-y^{2}$, где $p, q>0$. Тогда множество нулей многочлена $F$ состоит из двух овалов, если многочлен $f(x)=x^{3}-p x+q$ имеет три действительных корня, и из одного овала, если $f(x)$ имеет один действительный корень. Приравняв производную многочлена $f$ к нулю, убеждаемся, что он имеет локальный максимум в точке $x_{1}=-\sqrt{p / 3}$ и локальный минимум в точке $x_{2}=\sqrt{p / 3}$. Соответственно, $f(x)$ имеет три корня тогда и только тогда, когда $f\left(x_{2}\right)27 q^{2}$. Аналогично получаем, что множество нулей многочлена $F_{N}(x, y)$ состоит из двух овалов при $4 p^{3 N}>27 q^{2 N}$ и из одного овала в противном случае. Очевидно, что при $1<\frac{p^{3}}{q^{2}}<\frac{27}{4}$ первое неравенство не выполняется, а второе выполняется при достаточно больших $N$.
B3. (a) Пункт (a) является частным случаем пункта (b).
(b) Указание. Предположим, что в некоторой точке $(x, y)$ значения всех одночленов $a_{i j} x^{i} y^{j}$ различны по абсолютной величине, причем $\left|a_{k l} x^{k} y^{l}\right|>\left|a_{i j} x^{i} y^{j}\right|$ для всех пар $(i, j) \neq(k, l)$. Тогда при $N \rightarrow \infty$ $\left|\frac{a_{i j} x^{i} y^{j}}{a_{k l} x^{k} y^{N}}\right|^{N} \rightarrow 0$. Следовательно, при всех достаточно больших $N\left|a_{k l} x^{k} y^{l}\right|^{N}$ превосходит сумму абсолютных
величин остальных одночленов и равенство $F_{N}(x, y)=0$ невозможно. Поэтому множество $F_{N}^{-1}(0)$ при больших $N$ стремится к некоторому подмножеству объединения множеств, задаваемых равенствами вида $\left|a_{i j} x^{i} y^{j}\right|=\left|a_{k l} x^{k} y^{l}\right|$.
Решение. Пусть дано $\epsilon, R>0$. Обозначим через $\Gamma$ объединение всех кривых, задаваемых уравнениями $a_{i j} x^{i} y^{j} \pm a_{k l} x^{k} y^{l}=0$. Рассмотрим произвольную точку $\left(x_{0}, y_{0}\right) \in B_{R}$ на расстоянии больше $\epsilon$ от Г. Будем считать, что эта точка лежит в той части $\mathbb{R}^{2}-\Gamma$, в которой максимальное по модулю значение принимает одночлен $a_{k l} x^{k} y^{l}$. Пусть $\left(x_{1}, y_{0}\right) \in \Gamma$ - ближайшая к $\left(x_{0}, y_{0}\right)$ точка на прямой $y=y_{0}$. Тогда $\left|y_{1}-y_{0}\right|>\epsilon$ и $\left|a_{i j} x_{1}^{i} y_{0}^{j}\right|=\left|a_{k l} x_{1}^{k} y_{0}^{l}\right|$ для некоторых $i, j$.
Оценим отношение
$$
\frac{a_{i j} x_{0}^{i} y_{0}^{j}}{a_{k l} x_{0}^{k} y_{0}^{l}}=\frac{a_{i j}}{a_{k l}} x_{0}^{i-k} y_{0}^{j-l}=\left(\frac{x_{0}}{x_{1}}\right)^{i-k}
$$
Так как $\left|a_{k l} x_{0}^{k} y_{0}^{l}\right|>\left|a_{i j} x_{0}^{i} y_{0}^{j}\right|$, то $\left|x_{0} / x_{1}\right|<1$. Поскольку $\left|x_{0}\right|,\left|x_{1}\right|0$ значение $x$ приблизительно равно $y$.
(b) Указание. См. рис. 4.b. Обоснование аналогично (a).
(c) При $x \in(0,1 / 1001)$ значения $y$ изменяются от 0 до $+\infty$, при $x \geqslant 1 / 1001$ значение $y$ приблизительно равно 0 .
(d) Ясно, что множество нулей функции $f(x, y):=2^{2002 x}+2^{2002 y}-4^{1001} 2^{1001 x}-4^{1001} 2^{1001 y}-2^{1001}$ симметрично относительно прямой $y=x$. Поэтому можно рассматривать только случай $x \geqslant y$. При $x \geqslant 2003 / 1001$ значение $y$ приблизительно равно $x / 2+1$, при $x \in(2,2003 / 1001)$ значение $y$ изменяется от $-\infty$ до 2 .

a

b

c d
Рис. 4:
B6. (а) Рассуждая, как в задаче В3b, получаем, что пересечение множества нулей с первой координатной четвертью лежит вблизи объединения множеств $x^{2}=y \geq x, x^{2}=x \geq y, x=y \geq x^{2}$. Однако, коэффициенты при одночленах $x^{N}$ и $y^{N}$ у данного многочлена имеют одинаковые знаки, так что в окрестности последнего множества $F_{N}$ не может обращаться в нуль. При логарифмическом отображении первое из указанных множеств переходит в луч $y=2 x, x \geq 0$, а второе - в луч $x=0, y \leq 0$.
(b) Аналогично предыдущему пункту получаем, что множество нулей многочлена $F_{N}$ при больших $N$ лежит в окрестности объединения множеств, задаваемых соотношениями $a_{i j} x_{i} y^{j}=a_{k l} x^{k} y^{l} \geq a_{p q} x^{p} y^{q}$.
При логарифмическом отображении эти множества переходят в множества вида $i x+j y=k x+l y \geq$ $p x+q y$, каждое из которых является отрезком или лучом. Объединение этих множеств задает тропическую кривую.
B7. Рассуждая аналогично задаче B3d, получаем, что множества, содержащие пересечения множества нулей с первой и четвертой четвертью, симметричны относительно оси абсцисс, а с первой и третьей относительно начала координат.
B8. Так как в третьей четверти $x<0, y<0$, то все одночлены, входящие в многочлен $F_{N}$ положительны, т.е. равенство $F_{N}(x, y)=0$ невозможно.
B9. (а) Решение аналогично решению задачи В6а.
(b) Из задачи B3d следует, что пересечение множества нулей с второй четвертью лежит в окрестности множеств $x^{2}=y \geq-x, x^{2}=-x \geq y,-x=y \geq x^{2}$. Поскольку знаки одночленов $-x^{N}$ и $x^{2 N}$ во второй четверти совпадают, то пересечение множества нулей со второй четвертью содержится в окрестности только первого и третьего из указанных множеств.
B10. Рассуждая, как в задаче B9b, получаем, что искомое пересечение содержится в окрестности объединения множеств $(1, y), 0 \geq y \geq-1$ и $0 \leq x=-y \leq 1$.
B11. Каждое ребро тропической кривой задается системой из одного уравнения и нескольких неравенств вида $i x+j y+b_{i j}=k x+l y+b_{k l} \geq p x+q y+b_{p q}$. Если эта система совместна, то уравнение определяет прямую, а неравенства высекают на этой прямой отрезок или луч.
B12. Указание. Действительно, рассмотрим отдельно поведение множества нулей многочлена $F_{N}(x, y)$ в каждом из четырёх квадрантов. Отображение $L O G:(\mathbb{R}-\{0\})^{2} \rightarrow \mathbb{R}^{2},(x, y) \mapsto\left(\log _{2}|x|, \log _{2}|y|\right)$ переводит каждый из квадрантов на плоскость взаимно однозначно. Выберем один из квадрантов (например, $x, y>0$ ), и отождествим его с плоскостью указанным отображением. Тропическая кривая, задаваемая тропическим многочленом $\max _{i+j \leq d}\left\{i x+j y+b_{i j}\right\}, b_{i j}=\log _{2}\left|a_{i j}\right|$ разбивает тропическую плоскость на области. Внутри каждой области поведение многочлена $F_{N}(x, y)$ определяется поведением одного из мономов $\left(a_{i j} x_{i} y_{j}\right)^{N}$, и в зависимости от знака коэффициента $a_{i j}$ (а также выбранного квадранта) многочлен $F_{N}$ в данной области либо положителен, либо отрицателен. Закрасим каждую из областей дополнения к тропической кривой в один из двух цветов, в соответствии со знаком многочлена $F_{N}$ в этой области. Если две соседние области, граничащие вдоль некоторого ребра, окрашены в разные цвета, то по тереме о промежуточном значении вдоль этого ребра проходит ветвь множества нулей многочлена $F_{N}$. Если же обе соседние области окрашены в одинаковые цвета, то вблизи этого ребра нет вещественных точек кривой. Таким образом, для больших нечетных значений параметра $N$ множество нулей многочлена $F_{N}$ (в выбранном квадранте) приближённо изображается набором некоторого количества явно перечисляемых рёбер тропической кривой, а множество нулей многочлена $F_{N}$ во всей плоскости приближенно задается кривой Виро.
В принципе, множество нулей многочлена $F_{N}$ могло бы иметь больше ветвей, чем у кривой Виро например, могли бы существовать "маленькие" овалы вблизи вершин тропической кривой. Отсутствие "лишних" ветвей (и овалов) гарантирует нам предположение, что число овалов у кривой Виро равно $(d-1)(d-2) / 2+1$ и теорема Харнака.
Замечание. Авторам задачи неизвестно, остается ли верной теорема о склейке Виро без предположения, что число овалов у кривой Виро равно $(d-1)(d-2) / 2+1$.
C5. (a) См. рис. 5.a
(b) См. рис. 5.b.
(c) См. рис. 6 .
D1. (а) Указание. Функция $\max \{x+a, y+b, c\}$ имеет следующее поведение. При $x$ и $y$ отрицательных и больших по абсолютной величине максимальным из трёх величин является постоянное значение $c$. При увеличении $x$ значение функции не меняется до тех пор, пока точка $(x, y)$ не пересечет вертикальную прямую $x+a=c$. Правее этой прямой максимальной является величина $x+a$. Аналогично, при движении точки $(x, y)$ вверх переход к величине $y+b$ осуществляется на горизонтальной прямой $y+b=c$, вдоль которой максимум достигается на двух конкурирующих выражениях $y+q$ и $c$. Наконец, области, в которых значение функции $f(x, y)$ совпадает с выражениями $x+a$ и $y+b$, разделяются лучом прямой $x+a=y+b$, имеющей наклон 1 . Все три полученных луча сходятся в точке $(c-a ; c-b)$.
D2. (а) Очевидно.
(b) Указание. Рассмотрим вершину кривой и предположим, что к этой вершине подходит $r$ областей (дополнения к тропической кривой), в которых максимум достигается на функциях $i_{1} x+j_{1} y+b_{i_{1} j_{1}}, \ldots, i_{r} x+$

a

b
Рис. 5:

Рис. 6:
$j_{r} y+b_{i_{r} j_{r}}$, соответственно (мы считаем области занумерованными против часовой стрелки). Тогда, очевидно, выполняется векторное равенство
$$
\binom{i_{2}-i_{1}}{j_{2}-j_{1}}+\cdots+\binom{i_{r}-i_{r-1}}{j_{r}-j_{r-1}}+\binom{i_{1}-i_{r}}{j_{1}-j_{r}}=0
$$
Осталось заметить только, что вектор $\binom{i_{s+1}-i_{s}}{j_{s+1}-j_{s}}$ отличается от вектора $v_{s}$, участвующего в условии сбалансированности, только лишь поворотом на $90^{\circ}$.
(c) Указание. Докажем, например, что тропическая кривая степени $d$ имеет (с учетом кратности) ровно $d$ горизонтальных лучей, направленных влево. Действительно, будем рассматривать только ту часть плоскости, в которой координата $x$ отрицательна и очень велика по абсолютной величине. Ясно, что в этой части максимальной может быть только одна из величин $j y+a_{0 j}, j=0,1, \ldots, d$. Ясно также, что в этой части при больших по абсолютной величине отрицательных $y$ максимальна величина $a_{00}$, а при больших по абсолютной величине положительных $y$ максимальна величина $d y+a_{0 d}$. Пусть при увеличении $y$ максимальными последовательно становятся величины $a_{00}, j_{1} y+a_{0 j_{1}}, j_{2} y+a_{0 j_{2}}, \ldots, j_{k} y+a_{0 j_{k}}, d y+a_{0 d}$. Легко видеть, что $00$ there exist $\delta$ such that for any
$$
t \in\left(t_{0}-\delta, t_{0}+\delta\right) \quad \text { we have } \quad\left|\gamma(t)-\gamma\left(t_{0}\right)-A\left(t-t_{0}\right)\right|<\varepsilon\left|t-t_{0}\right|
$$
A map $\gamma:[a, b] \rightarrow \mathbb{R}^{2}$ can be considered as an ordered pair of functions $\gamma_{1}, \gamma_{2}:[a, b] \rightarrow \mathbb{R}$. A map $\gamma:[a, b] \rightarrow \mathbb{R}^{2}$ is differentiable if both functions $\gamma_{1}, \gamma_{2}$ are differentiable.
A (smooth) curve in the plane is a differentiable map $\gamma: \mathbb{R} \rightarrow \mathbb{R}^{2}$ (or $\gamma:[a, b] \rightarrow \mathbb{R}^{2}$ ).
In problems A4.cfg and A7 you only need to give an example of a polynomial; proving its properties is not required.
A4. (a) There is an irreducible polynomial of degree 3 whose zero set contains a closed curve.
(b) The same for degree 4 .
(c) There is an irreducible polynomial of degree 4 whose zero set consists of two closed curves one inside the other.
(d) If the zero set of a polynomial of degree 4 contains two closed curves one inside the other, then the zero set contains no other points.
(e) Is the analogue of (d) correct for an irreducible polynomial of degree 5?
(f) There is a polynomial of degree 4 whose zero set contains 4 closed curves.
(g) There is a polynomial of degree 4 whose zero set contains 3 closed curves.
Ovals . Connected components of the zero set of a polynomial are called branches. (Existence of unbounded branches makes the investigation of zero sets harder.) For an unbounded branch $B$ the lines joining the origin $O$ with the points of $B$ have a "limit" line. Two unbounded branches are elementary equivalent if their "limit" lines coincide.
A5. The infinite branches of hyperbola $x y=1$ are elementary equivalent.
Two infinite branches are equivalent if there is a sequence of branches joining them, in which sequence each two consecutive branches are elementary equivalent. A zero set is nondegenerate if it is a disjoint union of smooth curves. An oval of a non-degenerate zero set of a polynomial is either a closed curve (contained in the zero set) or an equivalence class of unbounded branches. (Note that this definition is different from the "correct" one given in textbooks.)
A6. Find all $h$ such that the zero set is non-degenerate and find the number of ovals for the polynomial
(a) $x y(x+y-1)-h$.
(b) $x^{3}-x+h-y^{2}$.
(the answer could depend on $h$ ).
A7. There is a polynomial of degree 5 whose zero set is non-degenerate and consists of 7 ovals.
Hilbert's 16th problem. What could be the number and mutual arrangement of ovals of a non-degenerate zero set of a polynomial of degree d?
We do not assign any formal meaning to the words 'mutual arrangement'. Such a meaning can be assigned, but requires projectivization of a polynomial.
Main Theorem. (a) There is a polynomial of degree 6 whose zero set is non-degenerate and consists of 11 ovals.
(b) There are three polynomials of degree 6 each whose zero sets are non-degenerate and consist of 11 ovals each, with different mutual arrangement of ovals.
## B. Tropical curve as a limit of algebraic curves.
B1. Draw the zero sets of
(a) $x-y-1$;
(a') $x^{1001}-y^{1001}-1$;
(b) $x+y-1$;
(b) $x^{1001}+y^{1001}-1$;
(c) $x y=x+y$;
(c') $x^{1001} y^{1001}=x^{1001}+y^{1001}$;

(e') $x^{3003}+2^{1001} x^{1001} y^{2002}-3^{1001} x^{1001} y^{1001}+y^{2002}-x^{1001}-2^{1001}$.
Denote by
$$
F_{N}(x, y)=\sum_{i+j \leq d}\left(a_{i j} x^{i} y^{j}\right)^{N}
$$[^11]a family of polynomials depending on an odd number $N \geq 1$. Under the transformation of variables $u=x^{N}, v=y^{N}$ each polynomial $F_{N}$ goes to the polynomial $\sum_{i+j \leq d} a_{i j}^{N} u^{i} v^{j}$ of degree $d$. So for solution of the Hilbert 16th problem it is worth to determine the number and mutual arrangement of ovals of $F_{N}^{-1}(0)$.
B2. The number of ovals of $F_{N}^{-1}(0)$ can be different from that of $F_{1}^{-1}(0)$.
Denote by $B_{R}$ the ball of radius $R$ centered at 0 .
B3. (a) For each $\varepsilon, R>0$ there is $N_{0}>0$ such that for each odd $N>N_{0}$ the intersection of the zero set of $x^{2 N}-x^{N}-y^{N}$ with $B_{R}$ is contained in $\varepsilon$-neighborhood of the union of the lines $x=0, x=1$, $x=y$ and the parabola $y=x^{2}$.
(b) For each $\varepsilon, R>0$ there is $N_{0}>0$ such that for each odd $N>N_{0}$ the set $F_{N}^{-1}(0) \cap B_{R}$ is contained in $\varepsilon$-neighborhood of the union of the zero sets of all the polynomials $a_{i j} x^{i} y^{j}-a_{k l} x^{k} y^{l}$, in which $(i, j) \neq(k, l), i+j \leq d, k+l \leq d$.
Denote by $\mathbb{R}_{+}:=[0,+\infty)$ the set of positive numbers and by $\mathbb{R}_{+}^{2}:=[0,+\infty)^{2}$ the angle defined by the inequalities $x>0, y>0$. Define a map $L O G: \mathbb{R}_{+}^{2} \rightarrow \mathbb{R}^{2}$ by $L O G(x, y)=\left(\log _{2} x, \log _{2} y\right)$.
B4. (abcde) Draw the $L O G$-image of the intersection with $\mathbb{R}_{+}^{2}$ of the zero sets of polynomials (a'b'c'd'e') of B1.
A tropical polynomial is a function
$$
\begin{gathered}
f(x, y):=\max _{i+j \leq d}\left(i x+j y+b_{i j}\right) \\
\text { Let } \quad f^{p q}=\left\{(x, y) \in \mathbb{R}^{2} \mid f(x, y)=p x+q y+b_{p q}\right\}
\end{gathered}
$$
The union of intersections of different $f^{p q}$ is a tropical curve. (This is the set of "break points" of $f$.)
Assume further that all $a_{i j} \neq 0$ for $i+j \leq d$. The tropical curve corresponds to the family of polynomials $F_{N}$, if $b_{i j}=\log _{2}\left|a_{i j}\right|$. This definition is motivated by important problem B6b below.
B5. Draw the tropical curve corresponding to the family of polynomials
(a) $(a x)^{N}+(b y)^{N}+c^{N}$ ?
(b) $\left(a x^{2}\right)^{N}+(2 b x y)^{N}+\left(c y^{2}\right)^{N}$ ?
(the answer could depend on $a, b, c$.)
Denote by $\Delta_{R}$ the triangle given by the inequalities $x \geq-R, y \geq-R, x+y \leq R$.
B6. (a) For each $\varepsilon, R>0$ there is $N_{0}$ such that for $N>N_{0}$ the intersection of the $L O G$-image of the zero set of the polynomial $x^{2 N}-x^{N}-y^{N}$ with the triangle $\Delta_{R}$ is contained in $\varepsilon$-neighborhood of the union of the ray $y=2 x, x \geq 0$ and the ray $x=0, y \geq 0$.
(b) For each numbers $\left\{a_{i j}\right\}_{i+j \leq d}$ and $\varepsilon, R>0$ there is $N_{0}>0$ such that for $N>N_{0}$ the set $\operatorname{LOG}\left(F_{N}^{-1}(0) \cap \mathbb{R}_{+}^{2}\right) \cap \Delta_{R}$ is contained in $\varepsilon$-neighborhood of the intersection of the tropical curve corresponding to $F_{N}$.
## C. Tropical lines and circles.
This part of the project is an contest in art: it is suggested to check the theorems of tropical geometry experimentally by drawing accurate figures. Selected figures will be exposed for public viewing. "Problems" C1C10 are not graded (although for accurate statement and proving some of these assertions additional points would be awarded). Ignore part of a "problem" if you do not know the corresponding theorem of Euclidean geometry. The whole part $C$ of the project is not required for the solution of the Hilbert 16th problem and can be skipped.
Consider the plane with fixed Cartesian coordinate system. A tropical line ('leg") is a union of three rays with common origin (called the vertex), one of them going "west", the other going "south" and the third going "north-east".
C1. There are different tropical lines intersecting at two different points.
Two points are in general position if the Euclidean line passing through these points is not parallel either to coordinate axes or to the line $x=y$.
C2. (a) For each two points in general position there is a unique tropical line passing through these points. (b) If the vertices of two tropical lines are in general position, then the lines have the only common point.
Two tropical lines are parallel if the vertex of one lies on the "north-eastern" ray of the other.
C3. If a point $A$ is in general position with the vertex of a tropical line $b$, then there is a unique tropical line passing through $A$ and parallel to $b$.
Two tropical lines are perpendicular if the Euclidean lines containing their "north-eastern" rays are symmetric with respect to the line $x=y$.
C4. If a point $A$ is in general position with the vertex of a tropical line $b$, then there is a unique tropical line passing through $A$ and perpendicular to $b$.
A tropical triangle ("spider") is the union of three tropical lines whose vertices are (pairwise) in general position.
C5. Draw figures to tropical analogues of the following theorems.
(a) The heights of a triangle intersect at a common point.
(b) the Pappus theorem.
(c) The Desargue theorem.
(d) The Sondat theorem.
For given points $A$ and $B$ a tropical circle ("heron") is the set of points $X$ for which there are orthogonal tropical lines, one of them passing through $A$ and $X$ and the other passing through $B$ and $X$. (Recall that there could be different tropical lines passing through $A$ and $X$.)
C6. (a) Draw a tropical circle. How does this set depend on $A, B$ ?
(b) Is it true that each tropical triangle has a circumscribed tropical circle?
(c) The Pascal theorem.
C7.* Define the tropical middle point of a tropical segment so that the tropical medians of a tropical triangle would intersect in a common point.
## TROPICAL GEOMETRY
## F. Nilov, A. Skopenkov, M. Skopenkov, A. Zaslavsky
The main problem's complect consists of two parts: the end of Part B and a new part D. Problems of part D use (except some explicitly indicated cases) neither notions nor results of previous parts of the project. So one may solve these problems without taking part in previous parts.
## B. The Viro patchworking theorem.
B3. For each $\varepsilon, R>0$ there is $N_{0}>0$ such that for each odd $N>N_{0}$ the intersection of the zero set of the polynomial
(c) $x^{2 N}-x^{N}-y^{N}$ with the disk $B_{R}$ and with the first coordinate quarter $(x>0, y>0)$ is contained in the $\varepsilon$-neighborhood of the union of the sets
$$
\{(1, y) \mid 0 \leq y \leq 1\}, \quad 0 \leq x=y \leq 1 \quad \text { and } \quad y=x^{2} \geq 1
$$
(d) $x^{2 N}-x^{N}-y^{N}$ with the disk $B_{R}$ and with the second coordinate quarter $(x<0, y>0)$ is contained in the $\varepsilon$-neighborhood of the union of the set that is symmetric to the union from (c) w.r.t. $O y$.
B7. State and prove the analogues of B3d for the third and the fourth coordinate quarters.
B8. The intersection of the zero set of the polynomial $x^{2 N}-x^{N}-y^{N}$ with the third coordinate quarter is empty.
B9. For each $\varepsilon, R>0$ there is $N_{0}>0$ such that for each odd $N>N_{0}$ the intersection of the zero set of the polynomial $x^{2 N}-x^{N}-y^{N}$ with the disk $B_{R}$ and with
(a) the first coordinate quarter is contained in the $\varepsilon$-neighborhood of the union of the sets $\{(1, y) \mid 0 \leq$ $y \leq 1\}$ and $y=x^{2} \geq 1$.
(b) the second coordinate quarter is contained in the $\varepsilon$-neighborhood of the union of the sets $0 \leq$ $-x=y \leq 1$ and $y=x^{2} \geq 1$.
B10. State and prove the analogue of B9 for the fourth coordinate quarter.
Let us state the Viro patchworking theorem that allows to find the number and mutual arrangement of ovals for certain special algebraic curves.
B11. Each tropical curve is a finite union of segments and rays.
Definition of the Viro curve and its ovals. Take the tropical curve corresponding to $\left\{a_{i j}\right\}$. The tropical curve is a finite union of edges (segments and rays) that intersect at vertices (i.e. at common points of edges). A face of the tropical curve is a connected component if its complement in the plane. To each face there corresponds a pair $(p, q)$ of integers such that $p x+q y+\log _{2}\left|a_{p q}\right|=$ $\max _{i+j \leq d, a_{i j} \neq 0}\left(i x+j y+\log _{2}\left|a_{i j}\right|\right)$ for points $(x, y)$ of this face, and the sign of $a_{p, q}$. In this definition we use not $\left\{a_{i j}\right\}$ but the tropical curve whose faces are marked with pairs of integers and signs.
Make a parallel transfer so that the vertices of the tropical curve would move into the angle $x>$ $0, y>0$. Define $U_{p, q, 00}$ to be the image of the face of the tropical curve marked by $(p, q)$ under this parallel transfer. Let $U_{p, q, 01}, U_{p, q, 10}$ and $U_{p, q, 11}$ be the symmetric images of $U_{p, q, 00}$ under the symmetries with respect to the $x$-axis, $y$-axis and $(0,0)$, respectively. Extend the given disposition of signs from the first coordinate quarter to the whole plane as follows: under the symmetry of $U_{p q}$ w.r.t. the $x$ axis the sign is multiplied by $(-1)^{q}$, while under the symmetry of $U_{i j}$ w.r.t. the $y$-axis the sign is multiplied by $(-1)^{p}$. (Thus $\operatorname{sgn} U_{p q, s t}=(-1)^{p s+q t} \operatorname{sgn} U_{p q, 00}$.) Define the Viro curve to be the union $\cup\left\{U_{\alpha} \cap U_{\beta} \mid \operatorname{sgn} U_{\alpha} \neq \operatorname{sgn} U_{\beta}\right\}$ of those edges of the tropical curve that split faces of different signs (see Figure). Two unbounded connected components of the Viro curve are
- elementary equivalent if they contain rays symmetric w.r.t the origin $(0,0)$.
- equivalent if there is a sequence of components joining them, in which sequence each two consecutive components are elementary equivalent.
An oval of the Viro curve is either a closed broken line contained in the Viro curve or an equivalence class of unbounded connected components.
You can use the following theorem without proof:
The Harnak Theorem. A non-degenerate zero set of a polynomial of degree $d$ cannot have more than $\frac{(d-1)(d-2)}{2}+1$ ovals.
B12.* The Viro patchworking theorem. Let the Viro curve assigned to the family of polynomials $F_{M}=\sum_{i+j \leq d}\left(a_{i j} x^{i} y^{j}\right)^{M}$ with all $a_{i j} \neq 0$ contain exactly $\frac{(d-1)(d-2)}{2}+1$ ovals. Then there exist $N$ such that the zero set of the polynomial $\sum_{i+j \leq d} a_{i j}^{N} u^{i} v^{j}$ is non-degenerate, and the number and mutual arrangement of the ovals are the same as those of the corresponding Viro curve.
## D. Construction of examples in the Hilbert 16th problem.
The aim of part D is to describe tropical curves using purely combinatorial method and to obtain a purely combinatorial construction of examples in 16th Hilbert problem.
Let us recall that tropical curve of degree $d$ is the set of "break points" of graph of the function $\max _{i+j \leq d}\left\{i x+j y+b_{i j}\right\}$ (see the details above, after problem B4).
D1. (a) Check that a tropical curve of degree 1 looks like picture 1. (Compare with the definition of a tropical line in part C).
(b) Each vertex of a tropical curve is contained in at least 3 edges.
To any edge of a tropical curve assign its multiplicity as follows. Suppose that value $i x+j y+b_{i j}$ is maximal in one of faces bounded by this edge, and value $i^{\prime} x+j^{\prime} y+b_{i^{\prime} j^{\prime}}$ is maximal in the other one. So the line, which contains the given segment, has the equation $\left(i-i^{\prime}\right) x+\left(j-j^{\prime}\right) y+\left(b_{i j}-b_{i^{\prime} j^{\prime}}\right)=0$. We define multiplicity of the given edge as the greatest common divisor of numbers $i-i^{\prime}$ and $j-j^{\prime}$.
In pictures we shall denote the multiplicate edges of a tropical curve with double (triple, and so on) lines.
D2. Tropical curves of degree $d$ have the following properties:
(a) The slope of any edge is a rational number.
(b) For any vertex the following balance condition holds. Denote by $v_{i}$ a vector beginning at the given vertex parallel to $i$-th edge starting from the vertex, and equal to the shortest vector with integer coordinates and given direction, multiplied by edge's multiplicity. Then $\sum v_{i}=0$.
(c) There are $3 d$ infinite edges (counted with multiplicity), $d$ of them are directed (strictly) to the "west", $d$ - to the "south" and $d$ - to the "north-east" with slope angle $45^{\circ}$.
D3. (a) One may uniquely restorate a tropical polynomial $\max _{i+j \leq d}\left\{i x+j y+b_{i j}\right\}$ (up to adding a constant) by its tropical curve.
(b) If the edges of a graph in the plane are segments and rays with given multiplicities, and the conditions (a), (b), (c) of problem D2 are satisfied, then the graph is a tropical curve of degree $d$.
If two tropical curves have the same combinatorial type of their graphs and the same slopes of their edges (but not nesessary their lengths and positions), we shall say that these curves have it the same configuration.
D4. Draw 5 different configurations of tropical curves of degree two.
All information required for solving the following problems you can find in the paragraph "Definition of Viro curve and its ovals" contained in the previous part.
D5. What maximal number of ovals may have Viro curve if $d=$ do not require the proof of maximality. Compare your answer with problems A4f and A7).
D6*. Write down a computer program which:
(a) draws all configurations of tropical curves of given degree $d$;
(b) given a tropical curve configuration and given the set of signs "plus, minus" assigned to all the faces $U_{i j}$ of its complement - the program checks the number of Viro curve ovals.
D7*. (ab) Prove the Main Theorem (you may use Viro patchworking theorem without proof).
## SOLUTIONS
A1. Answer:no. For example, the line $x=0$ is a set of zeros for different polynomials $F(x, y)=x$ and $G(x, y)=x^{2}$.
A2. Answer: a, b, c, e, f.
Examples. (a) Any line on the plane has an equation $A x+B y+C=0$ for some numbers $A, B, C$.
(b) Equation of a circle: $\left(x-x_{0}\right)^{2}+\left(y-y_{0}\right)^{2}-R^{2}=0$, where $\left(x_{0}, y_{0}\right)$ are coordinates of centre, $R$ is radius.
(c) Equation of a point $\left(x_{0}, y_{0}\right):\left(x-x_{0}\right)^{2}+\left(y-y_{0}\right)^{2}=0$.
(e) Equation of a unite of two lines: $(A x+B y+C)(a x+b y+c)=0$, where $A x+B y+C=0$ is an equation of the first line, $a x+b y+c=0$ - of a second one.
(f) Equation of a unite of 6 circles: $\prod_{k=0}^{6}\left(\left(x-x_{k}\right)^{2}+\left(y-y_{k}\right)^{2}-R^{k}\right)=0$, where $\left(x-x_{k}\right)^{2}+\left(y-y_{k}\right)^{2}-R_{k}^{2}=0$ is an equation of $k$-th circle.
Impossibility in point (d) is consequence of Problem A3a.
A3. (a) Let us parametrize the line $l: x=x_{0}+\alpha \cdot t, y=y_{0}+\beta \cdot t$. Substituting these formulas in the polynomial, we'll get a new polynomial $P(t)$, its degree no more than $d$. So polynomial $P(t)$ has no more than $d$ real roots, or equals to zero everywhere. Now let us prove that for any $d^{\prime}0, f(1)<0, f(2)>0$, it follows that two roots $c_{1}$ and $c_{2}$ of the equation $f(c)=0$ belong to the interval $(0,1)$, and the third root belongs to the interval $(1,2)$. Therefore $D(c)=0$ precisely in two points $c_{1}$ and $c_{2}$ of the interval $(0,1)$, and $D(c)>0$ for any $c \in\left(c_{1}, c_{2}\right)$ and $D(c)<0$ for remaining points of the interval $(0,1)$. (We assume w. l. g. that $c_{1}1 / 27$. If $h=0$ or $h=1 / 27$, the algebraic curve is degenerate. Here is the proof.
Denote $f(x, y):=x y(x+y-1)+h$. Further, denote the points of intersection of lines $O x, O y$ and $x+y-1=0$ and regions, into which the plane is divided by the lines as follows:
$$
A:=(1,0), \quad B:=(0,1)
$$
$C:=(0,0), X:=\{(x, y) \mid x>0, y>0, x+y<1\}, X_{A}:=y<0, x+y>1, X_{B}:=x<0, x+y>1$, $X_{C}:=x<0, y<0, Y_{A}:=x<0, y>0, x+y<1, Y_{B}:=x>0, y<0, x+y<1, Y_{C}:=x>0, y>0, x+y>1$.
Obviously $f(x, y)=h$ if (x,y) belongs to one of lines $O x, O y$ or $x+y-1=0$, and $f(x, y)h$ when (x,y) belongs to one of regions $Y_{A}, Y_{B}$ и $Y_{C}$. So if $h>0$ then zeros of polynomial $f(x, y)$ may lie only in $X_{A}, X_{B}, X_{C}$ and $X$, and if $h<0$ they may lie only in $Y_{A}, Y_{B}$ and $Y_{C}$.
Suppose $h<0$. Denote $y_{A}:=Y_{A} \cap f^{-1}(0)$. Definitions of $y_{B}$ and of $y_{C}$ are similar.
Let's prove that $y_{A}$ is a connected componenta of the set $f^{-1}(0)$ of zeros of $f$. Coordinates $x$ of points of intersection lines $y=c$ and $f^{-1}(0)$ are roots of the equation $x^{2}+(c-1) x+\frac{h}{c}=0$. The discriminant $D=D(c)$ of this equation equals to $(c-1)^{2}-\frac{4 h}{c}$. As $h<0$ so for any $c \in R_{+}, D(c)>0$. It follows, that any line $y=c$, where $c \in R_{+}$, intersects $F$ in two points (no more no less) namely $\left(x_{1,2}(c), c\right)$ such that $x_{1,2}(c)=\frac{ \pm \sqrt{D}-(c-1)}{2}$. Denote
$$
\gamma: R_{+} \rightarrow \mathbb{R}^{2} \quad \text { by formula } \quad\left\{\left(x_{2}(t), t\right) \quad t \in R_{+}\right.
$$
The function $x_{2}(c)$ is smooth, so the curve $\gamma(t)$ is smooth also. $\gamma\left(R_{+}\right)=y_{A}$ implies that $y_{A}$ is a connected componenta of the set $f^{-1}(0)$ of zeros of $f$. Similarly $y_{B}$ и $y_{C}$ are connected componentas of the same set. It is easy to prove that the direction of line $O x$ is a "limit direction" for the branch $y_{C}$. Similarly, this direction is a "limit direction" for branch $y_{A}$. So branches $y_{A}$ and $y_{C}$ are elementary equivalent. Similarly, branches $y_{A}$ and $y_{B}$ are elementary equivalent, as for these branches the direction of line $x+y-1=0$ is a "limit one". So, branches $y_{A}, y_{B}$ and $y_{C}$ are equivalent and form one oval.
If $h=0$, then an algebraic curve $f$ is a degenerate one. Suppose $h>0$. Denote
$$
x:=X \cap f^{-1}(0), \quad x_{A}:=X_{A} \cap f^{-1}(0), \quad x_{B}:=X_{B} \cap f^{-1}(0) \quad \text { and } \quad x_{C}:=X_{C} \cap f^{-1}(0)
$$
One proves that $x_{A}, x_{B}$ and $x_{C}$ are connected and equivalent in the same way as in the case $h<0$. So they form one oval for any $h>0$. If the set $x$ is non-empty and has more than one point, it is an oval (the proof is similar to solution of Problem A4(a)).
Let us prove that the set of points $x$ isn't empty only if $h \in\left(0, \frac{1}{27}\right]$. It is clear that $x$ is empty if and only if $D(c)<0$ for any $c \in(0,1)$. Derivative $D^{\prime}(c)>0$ if $c \in(0,1 / 3), D^{\prime}(c)=0$ if $c=1 / 3, D^{\prime}(c)<0$ if $c \in(1 / 3,1)$. So the function $D(c)$ has its maximum on intervale $(0,1)$ in the point $c=1 / 3$. So $D(c)<0$ for any $c \in(0,1)$ if and only if $D(1 / 3)=4 / 9-\frac{4 h}{c}<0$, i.e. $h>1 / 27$. If $h=1 / 27$, then the set $x$ consists from only one point. So, if $h \in(0,1 / 27)$, then the set of zeros $f^{-1}(0)$ consists from two ovals, if $h=1 / 27$, then the algebraic curve is degenerate, if $h>1 / 27$ then the set $f^{-1}(0)$ has only one oval.
(b) Answer: one oval if $h \in\left(-\frac{2}{3 \sqrt{3}}, \frac{2}{3 \sqrt{3}}\right)$, two ovals if $h \in\left(-\infty,-\frac{2}{3 \sqrt{3}}\right)$ and $h \in\left(\frac{2}{3 \sqrt{3}}, \infty\right)$, an algebraic curve $x^{3}-x+h-y^{2}$ is degenerate if $h= \pm \frac{2}{3 \sqrt{3}}$. Hint. The situation is similar to (a).
A7. Hint. Consider the polynomial $x\left((x-1)^{2}+y^{2}-2\right)\left((x+1)^{2}+y^{2}-2\right)+\frac{1}{100}$.
B1. (a') Hint. See figure 3.a'. Why the picture is right one? It is clear that the line $x+y=0$ is an axis of symmetry for the set of zeros of our polynomial. So we may study only the case $y>-x$. Moreover. it lies under
the line $y=x$. The set of zeros intersects with coordinate axes in points $(0,-1)$ and $(1,0)$. If $x=1+\epsilon(\epsilon>0)$ holds (1001 $)^{\frac{1}{1001}}1 / 1001$, then $x$ is approximately equal to $y$. If $1-\epsilon0$. Then the set of zeros of the polynomial $F$ consists of two ovals, if a polynomial $f(x)=x^{3}-p x+q$ has three real roots, and it consists of one oval, if $f(x)$ has one real root. After solving the equation $f^{\prime}(x)=0$, we'll see, that $f(x)$ has a local maximum in the point $x_{1}=-\sqrt{p / 3}$ and a local minimum in point $x_{2}=\sqrt{p / 3}$. It follows, that $f(x)$ has three roots if and only if $f\left(x_{2}\right)27 q^{2}$. Analogously, we can prove that a set of zeros of a polynomial $F_{N}(x, y)$ consists of two ovals if $4 p^{3 N}>27 q^{2 N}$ and of one oval otherwise. It is obvious that if $1<\frac{p^{3}}{q^{2}}<\frac{27}{4}$ the first inequality doesn't hold, and the second holds for sufficiently big $N$.
B3. (a) Point (a) is a specific case of (b).
(b) Hint. Suppose that there is a point $(x, y)$ such that in it values of all monomials $a_{i j} x^{i} y^{j}$ differ by their modules, and $\left|a_{k l} x^{k} y^{l}\right|>\left|a_{i j} x^{i} y^{j}\right|$ for all pairs $(i, j) \neq(k, l)$. Then, when $N \rightarrow \infty$, we have $\left|\frac{a_{i j} x^{i} y^{j}}{a_{k l} x^{k} y^{l}}\right|^{N} \rightarrow 0$. So for any sufficiently big $N\left|a_{i j} x^{i} y^{j}\right|^{N}$ is more than the sum of modules of all the rest monomials, so the equality $F_{N}(x, y)=0$ is impossible. So, for sufficiently big $N$ the set $F_{N}^{-1}(0)$ tends to some subset of the union of sets, which are defined by equalities of the type $\left|a_{i j} x^{i} y^{j}\right|=\left|a_{k l} x^{k} y^{l}\right|^{N}$.
(c) It follows from the statement of previous problem. One must consider it for a polynomial $F_{N}=x^{2 N}-$ $x^{N}-y^{N}$.
(d) A set of zeros of a polynomial $x^{2 N}-x^{N}-y^{N}$, which lie in a second quadrant, is symmetric with respect to ordinate axis to the set of zeros of a polynomial $x^{2 N}+x^{N}-y^{N}$, which lie in a first quadrant. So our statement is a consequence of the statement of point (b).
B4. (a) Hint. See figure 4.a. Let us explain why the picture is correct. It is clear that the set of zeros of the function $f(x, y):=2^{1001 x}-2^{1001 y}-1$ lies on the right of the axis $O y$. If $y<0$, then $x$ is approximately equals to zero, and If $y>0$, then $x$ is approximately equals to $y$.
(b) Hint. See figure 4.b. Everything is analogous to (a).
(c) Hint. See figure 4.c.

a

b

c
Figure 4.
d
B6. (a) We may follow the method used in problem B3b, and we'll see that the intersection of zero set with the 1st coordinate quadrant lies near the union of sets $x^{2}=y \geq x, x^{2}=x \geq y, x=y \geq x^{2}$. But given polynomial has the same signs of coefficients of monomials $x^{N}$ and $y^{N}$, so $F_{N}$ cannot equal zero near the last set. Logarithmical map brings the first named set to the ray $y=2 x, x \geq 0$, and the second one - to the ray $x=0, y \leq 0$.
B7. Arguing, as in Problem B3d, we see that the intersection of zero set with the first quadrant, is symmetric with respect to abscissa axis to the intersection of zero set with the fourth quadrant, and is symmetric with respect to coordinate beginning to the intersection with third quadrant.
B8. In the third quadrant $x<0, y<0$. It means that all monomials in $F_{N}$ are positive, so the equality $F_{N}(x, y)=0$ is impossible.
B9. (a) The solution is analogous to the solution of Problem B6a.
(b) It follows from the problem B3d, that the intersection of a set of zeros with the second quadrante lies near sets $x^{2}=y \geq-x, x^{2}=-x \geq y,-x=y \geq x^{2}$. Signs of monomials $-x^{N}$ and $x^{2 N}$ coincide in the second quadrant. So the intersection of a set of zeros with the second quadrant lie only near the first and the third of the named sets.
B10. Arguing, as in Problem B9b, we see that the intersection we study lies near the unite of sets $(1, y), 0 \geq$ $y \geq-1$ and $0 \leq x=-y \leq 1$.
B11. Any edge of a tropical curve may be defined by the system which consists of one equation and some inequalities of type $i x+j y+b_{i j}=k x+l y+b_{k l} \geq p x+q y+b_{p q}$. If this system is compatible, then the equation defines some line, and inequalities show that one must take a ray or segment instead of all line.
B12. Hint. Really, let us study zeros of a polynomial $F_{N}(x, y)$ in each quadrant separately. The map $L O G$ : $(\mathbb{R}-\{0\})^{2} \rightarrow \mathbb{R}^{2},(x, y) \mapsto\left(\log _{2}|x|, \log _{2}|y|\right)$ is a bijection of each quadrant to the plane. Let us take any quadrant (for example $x, y>0$ ), and let us identify it with the plane with this map. A tropical curve, corresponding to a tropical polynomial $\max _{i+j \leq d}\left\{i x+j y+b_{i j}\right\}, b_{i j}=\log _{2}\left|a_{i j}\right|$, divides a tropical plane to some areas. In each of this areas one of the monomials $\left(a_{i j} x_{i} y_{j}\right)^{N}$ defines the behavior of the polynomial $F_{N}(x, y)$, and it is positive or negative, correspondingly to the sign of a coefficients $a_{i j}$ (of course, it depends also from the quadrant). Let us paint each area of the complement to tropical curve in one color, if $F_{N}$ is positive in this area, and in other color, if $F_{N}$ is negative in it. If two areas $i$-th common edge are painted in different colors, then, by the Theorem on intermediate value near this edge lies some branch of the set of zeros of $F_{N}$. Now, if such two areas are painted in the same color, then no real point of the curve lies near this edge. So, for big odd values of $N$ the set of zeros of $F_{N}$ (in chosen quadrant) may be approximately shown as a set of some edges of a tropical curve (which may be implicitly named), and the set of zeros of $F_{N}$ in all the plane is, approximately, a Viro curve.
Principally, the set of zeros of $F_{N}$ could have more branches, than Viro curve - for example, there could be some "small" ovals near vertices of tropical curve. But the supposition about the number of Viro curve ovals (we suppose it has $(d-1)(d-2) / 2+1$ ovals), combined with Harnak theorem guarantees us from superfluous branches and ovals.
Remark. Authors of the problem don't know, is Viro patchworking theorem is true without the supposition that Viro curve has $(d-1)(d-2) / 2+1$ ovals.
C5. (a) See figure 5.a.
(b) See figure 5.b.
(c) See figure 6 .

a

b
Figure 5 .

Figure 6.
D1. (a) Hint. The behavior of a function $\max \{x+a, y+b, c\}$ is such one. If $x$ and $y$ are negative and big (by module) then the constanta $c$ is the biggest of our three values. When $x$ increases, nothing changes till the point $(x, y)$ will intersect the vertical line $x+a=c$. After such intersection the value $x+a$ is maximal. Similarly, when the point $(x, y)$ moves up, maximal value is still $c$ till it would reach the horizontal line $y+b=c$. On it both values $y+b$ and $c$ are maximal, later - only $y+b$. At last, the areas, in which maximal value is $x+a$ or $y+b$, are divided by the ray of the line $x+a=y+b$. All three rays have the common point $(c-a, c-b)$
D2. (a) It is obvious.
(b) Hint. Let us take any vertice of the curve. Suppose that there are $r$ areas (supplements to tropical curve) which are near this vertice, and that in these areas maximal are functions $i_{1} x+j_{1} y+b_{i_{1} j_{1}}, \ldots, i_{r} x+j_{r} y+b_{i_{r} j_{r}}$, respectively (we suppose that the areas are numerated in positive direction, against the clock needle). Then the equality is obvious:
$$
\binom{i_{2}-i_{1}}{j_{2}-j_{1}}+\cdots+\binom{i_{r}-i_{r-1}}{j_{r}-j_{r-1}}+\binom{i_{1}-i_{r}}{j_{1}-j_{r}}=0
$$
Now one has to notice only, that the vector $\binom{i_{s+1}-i_{s}}{j_{s+1}-j_{s}}$ differs from the vector $v_{s}$ in the "balance condition" only by turn on $90^{\circ}$.
(c) Hint. Let us prove, for example, that the tropical curve of degree $d$ has exactly $d$ horizontal rays (counting with multiplicity, of course). Let us study only the part of the plane, where coordinate $x$ is negative and very big by module. Obviously in this part only one of the values $j y+a_{0 j}, j=0,1, \ldots, d$ may be the maximal one. It is
obvious also, that in this part, when $y$ is negative and big by module, then $a_{0} 0$ is maximal, and when $y$ is positive and big by module, then $d y+a_{0 d}$ is maximal. Let $y$ grow, and suppose, that maximal value will be (successively) $a_{00}, j_{1} y+a_{0 j_{1}}, j_{2} y+a_{0 j_{2}}, \ldots, j_{k} y+a_{0 j_{k}}, d y+a_{0 d}$. One easily sees, that $00$ существует число $T_{0}$ такое, что для каждого $T>T_{0}$ выполнено $\left|P(X)-P_{T}(X)\right|<\varepsilon$.
[^3]: ${ }^{2}$ Величина, обратная проводимости, называется сопротивлением.
[^4]: ${ }^{1}$ Formally, this means that for each $\varepsilon>0$ there is a number $T_{0}$ such that for each $T>T_{0}$ we have $\left|P(X)-P_{T}(X)\right|<\varepsilon$.
[^5]: ${ }^{2}$ The reciprocal of conductance is called resistance.
[^6]: ${ }^{*}$ Corresponding author
Email address: skopenkov@rambler.ru, Tel.: +966 531557 960, Fax: +966 28020064 (M. Skopenkov)
[^7]: ${ }^{1}$ Не переживайте, если Вам что-нибудь непонятно в этом кратком обзоре. Вы можете пропустить его и начать решать задачи с любой из частей А или С.
${ }^{2} \Pi$ Пи $d \leq 5$ ответ был известен еще в 19-м веке. Гильберт формулировал свою проблему для $d=6$, в этом случае решение было получено Гудковым. Для $d=7$ проблема была решена О.Я. Виро с использованием методов тропической геометрии. Для $d \geq 8$ данный вопрос до сих пор открыт.
${ }^{3}$ Не следует думать, что многочлену от двух переменных с некоторыми конкретными коэффициентами соответствует какая-то определенная тропическая кривая. Связь между многочленами и тропическими кривыми сложнее. В некотором смысле, тропическая кривая является "пределом" целого семейства подмножеств, заданных многочленами от двух переменных - см. часть В.
[^8]: ${ }^{4}$ Следующие определения нужны только для аккуратного обоснования примеров в задаче А4.
[^9]: ${ }^{5}$ Пару чисел $(p, q)$, соответствующую грани, легко восстановить по самой тропической кривой (подумайте, как!)
[^10]: ${ }^{6}$ Do not worry if you do not understand something in this brief overview. You can omit it and start solving problems from either part A or part C.
${ }^{7}$ For $d \leq 5$ the answer was known as early as in 19th century. Hilbert stated his problem for $d=6$. The solution for this case was obtained by Gudkov. For $d=7$ the problem was solved by Viro using methods of tropical geometry. For $d \geq 8$ the problem is open.
[^11]: ${ }^{8}$ These definitions are required only for the accurate proofs of problem A4.