Q
stringlengths
4
3.96k
A
stringlengths
1
3k
Result
stringclasses
4 values
Proposition 6.3.1 Let \( \mathcal{A} \) and \( \mathcal{B} \) be two monotone increasing families of subsets of \( N = \{ 1,2,\ldots, n\} \) and let \( \mathcal{C} \) and \( \mathcal{D} \) be two monotone decreasing families of subsets of \( N \) . Then\n\n\[ \Pr \left( {\mathcal{A} \cap \mathcal{B}}\right) \geq \Pr \l...
Proof. Let \( f : P\left( N\right) \rightarrow {\mathbb{R}}^{ + } \) be the characteristic function of \( \mathcal{A} \), i.e., \( f\left( A\right) = 0 \) if \( A \notin \mathcal{A} \) and \( f\left( A\right) = 1 \) if \( A \in \mathcal{A} \) . Similarly, let \( g \) be the characteristic function of \( B \) . By the a...
Yes
\[ \alpha \left( G\right) \geq \mathop{\sum }\limits_{{v \in V}}\frac{1}{{d}_{v} + 1} \]
Proof. Let \( < \) be a uniformly chosen total ordering of \( V \) . Define\n\n\[ I = \{ v \in V : \{ v, w\} \in E \Rightarrow v < w\} \]\n\nLet \( {X}_{v} \) be the indicator random variable for \( v \in I \) and \( X = \mathop{\sum }\limits_{{v \in V}}{X}_{v} = \left| I\right| \) . For each \( v \)\n\n\[ E\left\lbrac...
Yes
Theorem 2 [Turán (1941) ]Turán's Theorem Let \( H \) have \( n \) vertices and e edges. Then \( \alpha \left( H\right) \geq m \) and \( \alpha \left( H\right) = m \Leftrightarrow H \cong {G}_{n, e} \) .
Proof. \( {G}_{n, e} \) has \( \mathop{\sum }\limits_{{v \in V}}{\left( {d}_{v} + 1\right) }^{-1} = m \) since each clique contributes 1 to the sum. Fixing \( e = \mathop{\sum }\limits_{{v \in V}}{d}_{v}/2,\mathop{\sum }\limits_{{v \in V}}{\left( {d}_{v} + 1\right) }^{-1} \) is minimized with the \( {d}_{v} \) as close...
Yes
Theorem 7.2.1 [Azuma’s Ineqality] Let \( 0 = {X}_{0},\ldots ,{X}_{m} \) be a martingale with\n\n\[ \left| {{X}_{i + 1} - {X}_{i}}\right| \leq 1 \]\n\nfor all \( 0 \leq i < m \) . Let \( \lambda > 0 \) be arbitrary. Then\n\n\[ \Pr \left\lbrack {{X}_{m} > \lambda \sqrt{m}}\right\rbrack < {e}^{-{\lambda }^{2}/2}. \]
Proof. Set, with foresight, \( \alpha = \lambda /\sqrt{m} \) . Set \( {Y}_{i} = {X}_{i} - {X}_{i - 1} \) so that \( \left| {Y}_{i}\right| \leq 1 \) and \( E\left\lbrack {{Y}_{i} \mid {X}_{i - 1},{X}_{i - 2},\ldots ,{X}_{0}}\right\rbrack = 0 \) . Then, as in A.1.16,\n\n\[ E\left\lbrack {{e}^{\alpha {Y}_{i}} \mid {X}_{i ...
Yes
Theorem 7.2.3 When \( f \) satisfies the edge Lipschitz condition, the corresponding edge exposure martingale satisfies \( \left| {{X}_{i + 1} - {X}_{i}}\right| \leq 1 \) . When \( f \) satisfies the vertex Lipschitz condition the corresponding vertex exposure martingale satisfies \( \mid {X}_{i + 1} - \) \( {X}_{i} \m...
We prove these results in a more general context later. They have the intuitive sense that if knowledge of a particular vertex or edge cannot change \( f \) by more than one then exposing a vertex or edge should not change the expectation of \( f \) by more than one.
No
Theorem 7.2.4 [Shamir and Spencer (1987) ] Let \( n, p \) be arbitrary and let \( c = \) \( E\left\lbrack {\chi \left( G\right) }\right\rbrack \) where \( G \sim G\left( {n, p}\right) \) . Then \[ \Pr \left\lbrack {\left| {\chi \left( G\right) - c}\right| > \lambda \sqrt{n - 1}}\right\rbrack < 2{e}^{-{\lambda }^{2}/2}....
Proof. Consider the vertex exposure martingale \( {X}_{1},\ldots ,{X}_{n} \) on \( G\left( {n, p}\right) \) with \( f\left( G\right) = \) \( \chi \left( G\right) \) . A single vertex can always be given a new color so the vertex Lipschitz condition applies. Now apply Azuma’s Inequality in the form of Corollary 7.2.2. \...
Yes
Lemma 7.3.1 \( E\left\lbrack Y\right\rbrack \geq \frac{{n}^{2}}{2{k}^{4}}\left( {1 + o\left( 1\right) }\right) \) .
Proof. Let \( \mathcal{K} \) denote the family of \( k \) -cliques of \( G \) so that \( f\left( k\right) = \mu = E\left\lbrack \left| \mathcal{K}\right| \right\rbrack \) . Let \( W \) denote the number of unordered pairs \( \{ A, B\} \) of \( k \) -cliques of \( G \) with \( 2 \leq \left| {A \cap B}\right| < k \) . Th...
Yes
Theorem 7.3.2\n\n\[ \Pr \left\lbrack {\omega \left( G\right) < k}\right\rbrack < {e}^{-\left( {c + o\left( 1\right) }\right) \frac{{n}^{2}}{{\ln }^{8}n}} \]
Proof. Let \( {Y}_{0},\ldots ,{Y}_{m}, m = \left( \begin{array}{l} n \\ 2 \end{array}\right) \), be the edge exposure martingale on \( G\left( {n,1/2}\right) \) with the function \( Y \) just defined. The function \( Y \) satisfies the edge Lipschitz condition\nas adding a single edge can only add at most one clique to...
Yes
Lemma 7.3.4 Let \( \alpha, c \) be fixed \( \alpha > \frac{5}{6} \) . Let \( p = {n}^{-\alpha } \) . Then almost always every \( c\sqrt{n} \) vertices of \( G = G\left( {n, p}\right) \) may be \( 3 \) -colored.
Proof. If not, let \( T \) be a minimal set which is not 3-colorable. As \( T - \{ x\} \) is 3-colorable, \( x \) must have internal degree at least 3 in \( T \) for all \( x \in T \) . Thus if \( T \) has \( t \) vertices it must have at least \( \frac{3t}{2} \) edges. The probability of this occuring for some \( T \)...
Yes
Theorem 7.4.1 Let \( L \) satisfy the Lipschitz condition. Then the corresponding martingale satisfies\n\n\[ \left| {{X}_{i + 1}\left( h\right) - {X}_{i}\left( h\right) }\right| \leq 1 \]\n\nfor all \( 0 \leq i < m, h \in {A}^{B} \) .
Proof. Let \( H \) be the family of \( {h}^{\prime } \) which agree with \( h \) on \( {B}_{i + 1} \) . Then\n\n\[ {X}_{i + 1}\left( h\right) = \mathop{\sum }\limits_{{{h}^{\prime } \in H}}L\left( {h}^{\prime }\right) {w}_{{h}^{\prime }} \]\n\nwhere \( {w}_{{h}^{\prime }} \) is the conditional probability that \( g = {...
Yes
For all \( \epsilon > 0 \) there exists \( \delta > 0 \) so that the following holds. Suppose Paul has a strategy for finding \( Y \) such that every line of questioning has total variance at most \( {\sigma }^{2} \) . Then \[ \Pr \left\lbrack {\left| {Y - E\left\lbrack Y\right\rbrack }\right| > {\alpha \sigma }}\right...
For simplicity we replace \( Y \) by \( Y - E\left\lbrack Y\right\rbrack \) so that we shall henceforth assume \( E\left\lbrack Y\right\rbrack = 0 \) . By symmetry we shall bound only the upper tail of \( Y \) . We set, with foresight, \( \lambda = \alpha /\left\lbrack {\sigma \left( {1 + \epsilon }\right) }\right\rbra...
Yes
Theorem 7.5.2\n\n\[ \Pr \left\lbrack {X - E\left\lbrack X\right\rbrack > \lambda \sqrt{n}}\right\rbrack < {e}^{-{\lambda }^{2}/2}, \]\n\n\[ \Pr \left\lbrack {X - E\left\lbrack X\right\rbrack < - \lambda \sqrt{n}}\right\rbrack < {e}^{-{\lambda }^{2}/2}. \]
Proof. Consider \( \{ - 1, + 1{\} }^{n} \) as the underlying probability space with all \( \left( {{\epsilon }_{1},\ldots ,{\epsilon }_{n}}\right) \) equally likely. Then \( X \) is a random variable and we define a martingale \( {X}_{0},\ldots ,{X}_{n} = \) \( X \) by exposing one \( {\epsilon }_{i} \) at a time. The ...
Yes
Theorem 7.5.3 Let \( \epsilon ,\lambda > 0 \) satisfy \( {e}^{-{\lambda }^{2}/2} = \epsilon \) . Then \[ \left| A\right| \geq \epsilon {2}^{n} \Rightarrow \left| {B\left( {A,{2\lambda }\sqrt{n}}\right) }\right| \geq \left( {1 - \epsilon }\right) {2}^{n}. \]
Proof. Consider \( \{ 0,1{\} }^{n} \) as the underlying probability space, all points equally likely. For \( y \in \{ 0,1{\} }^{n} \) set \[ X\left( y\right) = \mathop{\min }\limits_{{x \in A}}\rho \left( {x, y}\right) \] Let \( {X}_{0},{X}_{1},\ldots ,{X}_{n} = X \) be the martingale given by exposing one coordinate o...
Yes
\[ \rho \left( {A,\overrightarrow{x}}\right) = \mathop{\min }\limits_{{\overrightarrow{v} \in V\left( {A,\overrightarrow{x}}\right) }}\left| \overrightarrow{v}\right| \]
Proof. Let \( \overrightarrow{v} \in V\left( {A,\overrightarrow{x}}\right) \) achieve this minimum. The hyperplane through \( \overrightarrow{v} \) perpendicular to the line from the origin to \( \overrightarrow{v} \) then separates \( V\left( {A,\overrightarrow{x}}\right) \) from the origin so that all \( \overrightar...
Yes
Theorem 7.6.2\n\n\[ \int_{\Omega} \exp \left[ \frac{1}{4} \rho^2(A, \overrightarrow{x}) \right] d\overrightarrow{x} \leq \frac{1}{\Pr[A]}. \]
Proof.[Theorem 7.6.2] We use induction on the dimension \( n \). For \( n = 1, \rho(A, \overrightarrow{x}) = 1 \) if \( \overrightarrow{x} \notin A \), zero otherwise so that\n\n\[ \int \exp \left[ \frac{1}{4} \rho^2(A, \overrightarrow{x}) \right] = \Pr[A] + (1 - \Pr[A]) e^{1/4} \leq \frac{1}{\Pr[A]} \]\n\nas the inequ...
Yes
Theorem 7.7.1 Under the above assumptions and for all \( b, t \)\n\n\[ \Pr \left\lbrack {X \leq b - t\sqrt{f\left( b\right) }}\right\rbrack \Pr \left\lbrack {X \geq b}\right\rbrack \leq {e}^{-{t}^{2}/4}. \]
Proof. Set \( A = \{ x : h\left( x\right) < b - t\sqrt{f\left( b\right) }\} \) . Now suppose \( h\left( y\right) \geq b \) . We claim \( y \notin {A}_{t} \) . Let \( I \) be a set of indices of size at most \( f\left( b\right) \) that certifies \( h\left( y\right) \geq b \) as given above. Define \( {\alpha }_{i} = 0 \...
Yes
Theorem 1 [Weierstrass Approximation Theorem] For every continuous real function \( f : \left\lbrack {0,1}\right\rbrack \mapsto R \) and every \( \epsilon > 0 \), there is a polynomial \( p\left( x\right) \) such that \( \left| {p\left( x\right) - f\left( x\right) }\right| \leq \epsilon \) for all \( x \in \left\lbrack...
Proof. Since a continuous \( f : \left\lbrack {0,1}\right\rbrack \mapsto R \) is uniformly continuous there is a \( \delta > 0 \) such that if \( x,{x}^{\prime } \in \left\lbrack {0,1}\right\rbrack \) and \( \left| {x - {x}^{\prime }}\right| \leq \delta \) then \( \left| {f\left( x\right) - f\left( {x}^{\prime }\right)...
Yes
Theorem 8.1.1 [The Janson Inequality] Let \( {B}_{i}, i \in I,\Delta, M,\mu \) be as above and assume all \( \Pr \left\lbrack {B}_{i}\right\rbrack \leq \epsilon \) . Then\n\n\[ M \leq \Pr \left\lbrack {{ \land }_{i \in I}\overline{{B}_{i}}}\right\rbrack \leq M{e}^{\frac{1}{1 - \epsilon }\frac{\Delta }{2}} \]\n\nand, fu...
For each \( i \in I \)\n\n\[ \Pr \left\lbrack \overline{{B}_{i}}\right\rbrack = 1 - \Pr \left\lbrack {B}_{i}\right\rbrack \leq {e}^{-\Pr \left\lbrack {B}_{i}\right\rbrack } \]\n\nso, multiplying over \( i \in I \) ,\n\n\[ M \leq {e}^{-\mu }. \]
No
Theorem 8.3.1 Suppose there is a constant \( \mu \) so that\n\n\[ E\left\lbrack X\right\rbrack = {S}^{\left( 1\right) } \rightarrow \mu \]\n\nand such that for every fixed \( r \)\n\n\[ E\left\lbrack {{X}^{\left( r\right) }/r!}\right\rbrack = {S}^{\left( r\right) } \rightarrow {\mu }^{r}/r! \]\n\nThen\n\n\[ \Pr \left\l...
Proof. We do only the case \( t = 0 \) . Fix \( \epsilon > 0 \) . Choose \( s \) so that\n\n\[ \left| {\mathop{\sum }\limits_{{r = 0}}^{{2s}}{\left( -1\right) }^{r}\frac{{\mu }^{r}}{r!} - {e}^{-\mu }}\right| \leq \frac{\epsilon }{2}. \]\n\nThe Bonferroni Inequalities state that, in general, the inclusion-exclusion form...
Yes
Lemma 8.4.1 With the above notation and for any integer \( s \)\n\n\[ \Pr \left\lbrack {\text{ there exists a disfam }J,\left| J\right| = s}\right\rbrack \leq \frac{{\mu }^{s}}{s!}. \]
Proof. Let \( \mathop{\sum }\limits^{ * } \) denote the sum over all \( s \) -sets \( J \subseteq I \) with no \( j \sim {j}^{\prime } \) . Let \( \mathop{\sum }\limits^{o} \) denote the sum over ordered \( s \) -tuples \( \left( {{j}_{1},\ldots ,{j}_{s}}\right) \) with \( \left\{ {{j}_{1},\ldots ,{j}_{s}}\right\} \) f...
Yes
Lemma 8.4.2 With the above notation and for any integer \( s \)\n\n\[\n\begin{aligned} \Pr \left\lbrack {\text{ there exists a maxdisfam }J,\left| J\right| = s}\right\rbrack & \leq \frac{{\mu }^{s}}{s!}{e}^{-{\mu }_{s}}{e}^{\frac{\Delta }{2}} \\ & \leq \frac{{\mu }^{s}}{s!}{e}^{-\mu }{e}^{s\nu }{e}^{\frac{\Delta }{2}}....
Proof. As in Lemma 8.4.1 we bound this probability by \( \mathop{\sum }\limits^{ * } \) of \( J = \left\{ {{j}_{1},\ldots ,{j}_{s}}\right\} \) being a maxdisfam. For this to occur \( J \) must first be a disfam and then \( { \land }^{ * }\overline{{B}_{i}} \), where \( { \land }^{ * } \) is the conjunction over all \( ...
Yes
Theorem 8.5.1 Set \( p = \frac{\ln n}{n}\omega \left( n\right) \) where \( \omega \left( n\right) \rightarrow \infty \) arbitrarily slowly. Then in \( G\left( {n, p}\right) \) almost always
\[ \deg \left( x\right) \sim \left( {n - 1}\right) p \] for all vertices \( \mathbf{x} \). This is actually a large deviation result. It suffices to show the following.
Yes
Theorem 8.5.2 Set \( p = \frac{\ln n}{n}\omega \left( n\right) \) where \( \omega \left( n\right) \rightarrow \infty \) arbitrarily slowly. Let \( x \in G \) be fixed. Fix \( \epsilon > 0 \) . Then\n\n\[ \Pr \left\lbrack {\left| {\deg \left( x\right) - \left( {n - 1}\right) p}\right| > \epsilon \left( {n - 1}\right) p}...
Proof. As \( \deg \left( x\right) \sim B\left( {n - 1, p}\right) \), i.e., it is a Binomial random variable with the above parameters, we have from A.1.14 that\n\n\[ \Pr \left\lbrack {\left| {\deg \left( x\right) - \left( {n - 1}\right) p}\right| > \epsilon \left( {n - 1}\right) p}\right\rbrack < 2{e}^{-{c}_{\epsilon }...
Yes
Theorem 8.5.4 Let \( p \) be such that \( \mu > > \ln n \) . Let \( x \in G \) be fixed. Fix \( \epsilon > 0 \) Then\n\n\[ \Pr \left\lbrack {\left| {N\left( x\right) - \mu }\right| > {\epsilon \mu }}\right\rbrack = o\left( {n}^{-1}\right) . \]
Proof. We shall prove this under the further assumption \( p = {n}^{-2/3 + o\left( 1\right) } \) (or, equivalently, \( \mu = {n}^{o\left( 1\right) } \) ) which could be removed by technical methods. We now have, in the notation of Lemmas 8.4.1,8.4.2 \( {\nu \mu },\Delta = o\left( 1\right) \) . Let \( P \) denote the Po...
Yes
Theorem 8.6.2 [Erdős (1956)] There is a set \( S \) for which \( f\left( n\right) = \Theta \left( {\ln n}\right) \) . That is, there is a set \( S \) and constants \( {c}_{1},{c}_{2} \) so that for all sufficiently large \( n \)\n\n\[ \n{c}_{1}\ln n \leq f\left( n\right) \leq {c}_{2}\ln n \n\]
Proof. Define \( S \) randomly by\n\n\[ \n\Pr \left\lbrack {x \in S}\right\rbrack = {p}_{x} = \min \left\lbrack {{10}\sqrt{\frac{\ln x}{x}},1}\right\rbrack . \n\]\n\nFix \( n \) . Now \( f\left( n\right) \) is a random variable with mean\n\n\[ \n\mu = E\left\lbrack {f\left( n\right) }\right\rbrack = \frac{1}{2}\mathop{...
Yes
Theorem 8.6.3 [Erdős and Tetali (1990) ] There is a set \( S \) for which \( g\left( n\right) = \) \( \Theta \left( {\ln n}\right) \) . That is, there is a set \( S \) and constants \( {c}_{1},{c}_{2} \) so that for all sufficiently large \( n \)\n\n\[{c}_{1}\ln n \leq g\left( n\right) \leq {c}_{2}\ln n\]
Proof. Define \( S \) randomly by\n\n\[ \Pr \left\lbrack {x \in S}\right\rbrack = {p}_{x} = \min \left\lbrack {{10}{\left( \frac{\ln x}{{x}^{2}}\right) }^{1/3},\frac{1}{2}}\right\rbrack . \]\n\nFix \( n \) . Now \( g\left( n\right) \) is a random variable and\n\n\[ \mu = E\left\lbrack {g\left( n\right) }\right\rbrack =...
Yes
For all \( k \) there exists \( \epsilon > 0 \) so that for all sufficiently large \( n \) there exist graphs \( G \) on \( n \) vertices with \( \chi \left( G\right) > k \) and yet \( \chi \left( {\left. G\right| }_{S}\right) \leq 3 \) for every set \( S \) of vertices of size at most \( {\epsilon n} \) .
Proof. For a given \( k \) let \( c,\epsilon > 0 \) satisfy (with foresight)\n\n\[ c > 2{k}^{2}H\left( {1/k}\right) \ln 2 \]\n\n\[ \epsilon < {e}^{-5}{3}^{3}{c}^{-3} \]\n\nwhere \( H\left( x\right) = - x{\log }_{2}x - \left( {1 - x}\right) {\log }_{2}\left( {1 - x}\right) \) is the entropy function. Set \( p = c/n \) a...
Yes
Theorem 9.1.1 For the tournaments \( {T}_{p} \) described above,\n\n\[ c\left( {T}_{p}\right) \leq \frac{1}{2}\left( \begin{array}{l} p \\ 2 \end{array}\right) + O\left( {{p}^{3/2}\log p}\right) \]
In order to prove this theorem we need some preparations. Let \( \chi \) be the quadratic residue character defined on the elements of the finite field \( {GF}\left( p\right) \) by \( \chi \left( y\right) = \) \( {y}^{\left( {p - 1}\right) /2} \) . Equivalently, \( \chi \left( y\right) \) is 1 if \( y \) is a nonzero s...
No
Lemma 9.1.2 For any two subsets \( A \) and \( B \) of \( {GF}\left( p\right) \); \( \left| {\mathop{\sum }\limits_{{i \in A}}\mathop{\sum }\limits_{{j \in B}}{d}_{ij}}\right| \leq {\left| A\right| }^{1/2}{\left| B\right| }^{1/2}{p}^{1/2}. \)
Proof. By the Cauchy-Schwarz Inequality and by the fact above: \[ {\left( \mathop{\sum }\limits_{{i \in A}}\mathop{\sum }\limits_{{j \in B}}{d}_{ij}\right) }^{2} \leq \left| A\right| \left( {\mathop{\sum }\limits_{{i \in A}}{\left( \mathop{\sum }\limits_{{j \in B}}{d}_{ij}\right) }^{2}}\right) \] \[ \leq \left| A\right...
Yes
Theorem 9.2.1 For every partition of the set of vertices \( V \) into two disjoint subsets \( B \) and \( C \) :\n\n\[ e\left( {B, C}\right) \geq \frac{\left( {d - \lambda }\right) \left| B\right| \left| C\right| }{n}. \]
Proof. Put \( \left| V\right| = n, b = \left| B\right|, c = \left| C\right| = n - b \) . Let \( D = {dI} \) be the \( n \) by \( n \) scalar matrix with the degree of regularity of \( G \) on its diagonal. Observe that for any real vector \( x \) of length \( n \) (considered as a function \( x : V \mapsto R \) ) we ha...
Yes
Corollary 9.2.2 If \( \lambda \) is the second largest eigenvalue of a d-regular graph \( G \) with \( n \) vertices, then \( G \) is an \( \left( {n, d, c}\right) \) -expander for \( c = \frac{d - \lambda }{2d} \) .
Proof. Let \( W \) be a set of \( w \leq n/2 \) vertices of \( G \) . By Theorem 9.2.1 there are at least \( \frac{\left( {d - \lambda }\right) w\left( {n - w}\right) }{n} \geq \frac{\left( {d - \bar{\lambda }}\right) w}{2} \) edges from \( W \) to its complement. Since no vertex in the complement is adjacent to more t...
Yes
Theorem 9.2.4 Let \( G = \\left( {V, E}\\right) \) be a d-regular graph on \( n \) vertices, and suppose the absolute value of each of its eigenvalues but the first one is at most \( \\lambda \) . For a vertex \( v \\in V \) and a subset \( B \) of \( V \) denote by \( N\\left( v\\right) \) the set of all neighbours of...
Proof. Let \( A \) be the adjacency matrix of \( G \) and define a vector \( f : V \\mapsto R \) by \( f\\left( v\\right) = 1 - b \) for \( v \\in B \) and \( f\\left( v\\right) = - b \) for \( v \\notin B \) . Clearly \( \\mathop{\\sum }\\limits_{{v \\in V}}f\\left( v\\right) = 0 \), i.e., \( f \) is orthogonal to the...
Yes
Corollary 9.2.5 Let \( G = \left( {V, E}\right), d, n \) and \( \lambda \) be as in Theorem 9.2.4. Then for every two sets of vertices \( B \) and \( C \) of \( G \), where \( \left| B\right| = {bn} \) and \( \left| C\right| = {cn} \) we have:\n\n\[ \left| {e\left( {B, C}\right) - {cbdn}}\right| \leq \lambda \sqrt{bc}n...
Proof. By Theorem 9.2.4\n\n\[ \mathop{\sum }\limits_{{v \in C}}{\left( \left| {N}_{B}\left( v\right) \right| - bd\right) }^{2} \leq \mathop{\sum }\limits_{{v \in V}}{\left( \left| {N}_{B}\left( v\right) \right| - bd\right) }^{2} \leq {\lambda }^{2}b\left( {1 - b}\right) n. \]\n\nThus, by the Cauchy Schwarz inequality;\...
Yes
Theorem 9.2.7 Let \( G = \left( {V, E}\right) \) be a d-regular graph on \( n \) vertices, and suppose that each of its eigenvalues but the first one is at most \( \lambda \) . Let \( C \) be a set of \( {cn} \) vertices of \( G \) . Then, for every \( l \), the number of walks of length \( l \) in \( G \) that avoid \...
Proof. Let \( A \) be the adjacency matrix of \( G \) and let \( {A}^{\prime } \) be the adjacency matrix of its induced subgraph on the complement of \( C \) . We claim that the maximum eigenvalue of \( {A}^{\prime } \) is at most \( \left( {1 - c}\right) d + {c\lambda } \) . To prove this claim we must show that for ...
Yes
Corollary 9.2.8 Let \( G = \left( {V, E}\right), d, n,\lambda, C \) and \( c \) be as in Theorem 9.2.7 and suppose\n\n\[ \left( {1 - c}\right) d + {c\lambda } \leq \frac{d}{\sqrt{2}} \]\n\nThen, for every \( l \), the probability that a randomly chosen walk of length \( l \) in \( G \) avoids \( C \) is at most \( {2}^...
Proof. The number of walks of length \( l \) in \( G \) that avoid \( C \) is at most \( \left( {1 - c}\right) n((1 - \) \( c)d + {c\lambda }{)}^{l} \leq n{d}^{l}{2}^{-l/2} \), by Theorem 9.2.7. Since the total number of walks is \( n{d}^{l} \) , the desired result follows. -
Yes
Theorem 1 Let \( G = \left( {V, E}\right) \) be a vertex-transitive graph. For an integer \( k \) and for two (not neccessarily distinct) vertices \( u, v \) of \( G \), let \( {P}^{k}\left( {u, v}\right) \) denote the probability that a random walk of length \( k \) starting at \( u \) ends at \( v \) . Then, for ever...
Proof. We need the following simple inequality, sometimes attributed to Chebyschev.\n\nClaim 2 For every sequence \( \left( {{a}_{1},\ldots ,{a}_{n}}\right) \) of \( n \) reals and for any permutation \( \pi \) of \( \{ 1,\ldots, n\} \) :\n\n\[ \n\mathop{\sum }\limits_{{i = 1}}^{n}{a}_{i}{a}_{\pi \left( i\right) } \leq...
Yes
Theorem 10.1.1 Let \( H \) be a strictly balanced graph with \( v \) vertices, e edges and a automorphisms. Let \( c > 0 \) be arbitrary. Let \( A \) be the property that \( G \) contains no copy of \( H \) . Then with \( p = c{n}^{-v/e} \) ,
Proof. Let \( {A}_{\alpha },1 \leq \alpha \leq \left( \begin{array}{l} n \\ v \end{array}\right) v!/a \), range over the edge sets of possible copies of \( H \) and let \( {B}_{\alpha } \) be the event \( G\left( {n, p}\right) \supseteq {A}_{\alpha } \) . We apply Janson’s Inequality. As\n\n\[ \mathop{\lim }\limits_{{n...
Yes
Theorem 10.3.1 [Bollobás (1988) ] Almost always\n\n\\[ \chi \\left( G\\right) ) \\sim \\frac{n}{2{\\log }_{2}n}. \\]
Proof. Let \\( \\alpha \\left( G\\right) = \\omega \\left( \\bar{G}\\right) \\) denote, as usual, the independence number of \\( G \\) . The complement of \\( G \\) has the same distribution \\( G\\left( {n,1/2}\\right) \\) . Hence \\( \\alpha \\left( G\\right) \\leq \\left( {2 + o\\left( 1\\right) }\\right) {\\log }_{...
Yes
Theorem 10.7.2 For any irrational \( \alpha ,0 < \alpha < 1 \), setting \( p = p\left( n\right) = {n}^{-\alpha } \), and for any First Order A,
Both proofs are only outlined.
No
Lemma 10.7.3 For every First Order \( A \) there is a \( t = t\left( A\right) \) so that if \( G, H \) are any graphs with \( G \vDash A \) and \( H \vDash \neg A \) then Spoiler wins \( {EHR}\left\lbrack {G, H, t}\right\rbrack \) .
A detailed proof would require a formal analysis of the First Order language so we give only an example. Let \( A \) be the property \( {\forall }_{x}{\exists }_{y}\left\lbrack {x \sim y}\right\rbrack \) of not containing an isolated point and set \( t = 2 \) . Spoiler begins by selecting an isolated point \( {y}_{1} \...
No
Theorem 10.7.4 A function \( p = p\left( n\right) \) satisfies the Zero-One Law if and only if for every \( t \), letting \( G\left( {n, p\left( n\right) }\right), H\left( {m, p\left( m\right) }\right) \) be independently chosen random graphs on disjoint vertex sets\n\n\[ \mathop{\lim }\limits_{{m, n \rightarrow \infty...
Proof. We prove only the \
No
Theorem 10.7.5 For any fixed \( p,0 < p < 1 \), and any \( s, G\left( {n, p}\right) \) almost always has the full level \( s \) extension property.
Proof. For every distinct \( {u}_{1},\ldots ,{u}_{a},{v}_{1},\ldots ,{v}_{b}, x \in G \) with \( a + b \leq s \) we define \( {E}_{{u}_{1},\ldots ,{u}_{a},{v}_{1},\ldots ,{v}_{b}, x} \) to be the event that \( \left\{ {x,{u}_{i}}\right\} \in E\left( G\right) ,1 \leq i \leq a \) and \( \left\{ {x,{v}_{j}}\right\} \notin...
Yes
Lemma 10.7.6 [Generic Extension] Let \( \left( {R, H}\right) \), as given above, be safe. Let \( t \geq 0 \) be an arbitrary, but fixed, integer. Then in \( G \sim G\left( {n,{n}^{-\alpha }}\right) \) almost surely for all \( {x}_{1},\ldots ,{x}_{r} \) there exist \( {y}_{1},\ldots ,{y}_{v} \) such that\n\n1. \( {y}_{1...
Proof. From Exercise 5 almost surely every \( {x}_{1},\ldots ,{x}_{r} \) has \( \Theta \left( {{n}^{v}{p}^{e}}\right) \left( {R, H}\right) \) extensions \( {y}_{1},\ldots ,{y}_{v} \) . Our rough notion will be that the number of these \( {y}_{1},\ldots ,{y}_{v} \) that fail to be generic, in any of the bounded number o...
No
Lemma 10.7.7 [Finite Closure] Let \( \alpha, r > 0 \) be fixed. Set \( \varepsilon \) equal to the minimal value of \( \frac{{e\alpha } - v}{v} \) over all integers \( v, e \) with \( 1 \leq v \leq t \) and \( {e\alpha } - v > 0 \) . Let \( K \) be such that \( r - {K\varepsilon } < 0 \) . Then in \( G\left( {n,{n}^{-\...
Proof. If not there would be a rigid \( t \) -chain \( X = {X}_{0} \subset {X}_{1} \subset \ldots \subset {X}_{L} = Y \) with \( K + r < \left| Y\right| < K + r + t \) . Letting \( \left( {{X}_{i - 1},{X}_{i}}\right) \) have type \( \left( {{v}_{i},{e}_{i}}\right) \) the restriction of \( G \) to \( Y \) would have \( ...
Yes
Theorem 11.2.2 Let \( f = f\left( {{x}_{1},\ldots ,{x}_{n}}\right) \) be a function and let \( C \) be a \( C\left( {\infty, s, d, t}\right) \) - circuit computing \( f \), where \( s \cdot {\left( \frac{1}{2}\right) }^{t} \leq {0.5} \) . Then either \( f \) or its complement \( \bar{f} \) has a minterm of size at most...
Proof. Let us apply to \( C \), repeatedly, \( d - 2 \) times a random \( 1/\left( {10t}\right) \) - restriction. Each of these random restrictions, when applied to any bottom subcircuit of depth 2, transforms it by Lemma 11.2.1 with probability at least \( 1 - {\left( \frac{1}{2}\right) }^{t} \) from a \( t \) -Or-And...
Yes
Corollary 11.2.3 For any \( d \geq 2 \), there is no \( C\left( {\infty ,\frac{1}{2} \cdot {2}^{\frac{1}{10}{n}^{1/\left( {d - 1}\right) }}, d,\frac{1}{10}{n}^{1/\left( {d - 1}\right) }}\right) \) - circuit that computes the parity function \( f\left( {{x}_{1},\ldots ,{x}_{n}}\right) = {x}_{1} \oplus \cdots \oplus {x}_...
Proof. Assuming there is such a circuit we obtain, by Theorem 11.2.2, that the value of \( f \) can be fixed by assigning values to at most \( n - \frac{1}{2}{n}^{1/\left( {d - 1}\right) } + \frac{1}{10}{n}^{1/\left( {d - 1}\right) } < n \) variables. This is false, and hence there is no such circuit.
Yes
Lemma 11.3.2 There is no polynomial \( P\left( {{x}_{1},\ldots ,{x}_{n}}\right) \) over \( {GF}\left( 3\right) \) of degree at most \( \sqrt{n} \) which is equal to the parity of \( {x}_{1},\ldots ,{x}_{n} \) for a set \( S \) of at least \( {0.9} \cdot {2}^{n} \) distinct binary vectors \( \left( {{x}_{1},\ldots ,{x}_...
Proof. Suppose this is false, and suppose \( S \subset \{ 0,1{\} }^{n},\;\left| S\right| \geq {0.9} \cdot {2}^{n} \) and \( P\left( {{x}_{1},\ldots ,{x}_{n}}\right) = {x}_{1} \oplus \cdots \oplus {x}_{n} \) for all \( \left( {{x}_{1},\ldots ,{x}_{n}}\right) \in S \) . Define a polynomial \( Q = Q\left( {{y}_{1},\ldots ...
Yes
Corollary 11.3.3 There is no circuit of depth \( d \) and size \( s \leq \frac{1}{10}{2}^{\frac{1}{2}{n}^{1/{2d}}} \) computing the parity of \( {x}_{1},{x}_{2},\ldots ,{x}_{n} \) using Not, And, Or and \( {\operatorname{Mod}}_{3} \) gates.
Proof. Suppose this is false and let \( C \) be such a circuit. Put \( \ell = \frac{1}{2} \cdot {n}^{1/{2d}} \) . By Lemma 11.3.1 there is a polynomial \( P = P\left( {{x}_{1},\ldots ,{x}_{n}}\right) \) over \( {GF}\left( 3\right) \), whose degree is at most \( {\left( 2\ell \right) }^{d} = \sqrt{n} \), which is equal ...
Yes
Lemma 11.4.2 For all members \( {M}_{i} \) of \( M\left( C\right) \)\n\n\[ \n{A}_{i} - \left( {\mathop{\bigcup }\limits_{{j \leq i}}{\delta }_{ \sqcap }^{j}}\right) \subseteq {M}_{i} \subseteq {A}_{i} \cup \underset{j \leq i}{ \cup }{\delta }_{ \sqcup }^{j}.\n\]\n\n(11.10)
Proof. We apply induction on \( i \) . For \( i < 0\;{M}_{i} = {A}_{i} \) and thus (11.10) holds. Assuming (11.10) holds for all \( {M}_{j} \) with \( j < i \) we prove it for \( i \) . If \( {A}_{i} = {A}_{\ell } \cup {A}_{k} \) , then, by the induction hypothesis\n\n\[ \n{M}_{i} = {M}_{\ell } \cup {M}_{k} \cup {\delt...
Yes
Lemma 11.5.1 Let \( f = f\left( {{x}_{1},\ldots ,{x}_{n}}\right) \) be a non-atom Boolean function of \( n \) variables. Then there is an \( i,1 \leq i \leq n \) and an \( \varepsilon \in \{ 0,1\} \) such that for the function \( g = f\left( {{x}_{1},\ldots ,{x}_{i - 1},\varepsilon ,{x}_{i + 1},\ldots ,{x}_{n}}\right) ...
Proof. Fix a formula \( F \) computing \( f \) with \( l = L\left( f\right) \) And and \( {Or} \) gates. \( F \) can be represented by a binary tree each of whose \( l + 1 \) leaves is labeled by an atom \( {x}_{i} \) or \( {\bar{x}}_{i} \) . Let us choose, randomly, a variable \( {x}_{i},1 \leq i \leq n \) according t...
Yes
Corollary 11.5.2 If \( f = f\left( {{x}_{1},\ldots ,{x}_{n}}\right) \) and \( L\left( f\right) \leq {\left( \frac{n}{k}\right) }^{3/2} - 1 \), then one can assign values to \( n - k \) variables so that the resulting function \( g \) is an atom.
Proof. Repeated application of Lemma 11.5.1 \( n - k \) times yields a \( g \) with\n\n\[ \left( {L\left( g\right) + 1}\right) \leq \mathop{\prod }\limits_{{i = k + 1}}^{n}{\left( 1 - \frac{1}{i}\right) }^{3/2}\left( {L\left( f\right) + 1}\right) = {\left( k/n\right) }^{3/2}\left( {L\left( f\right) + 1}\right) \leq 1.\...
Yes
Theorem 1 Let \( \mathcal{F} \) be an antichain. Then\n\n\[ \mathop{\sum }\limits_{{A \in \mathcal{F}}}\frac{1}{\left( \begin{matrix} n \\ \left| A\right| \end{matrix}\right) } \leq 1 \]
Proof. Let \( \sigma \) be a uniformly chosen permutation of \( \{ 1,\ldots, n\} \) and set\n\n\[ {\mathcal{C}}_{\sigma } = \{ \{ \sigma \left( j\right) : 1 \leq j \leq i\} : 0 \leq i \leq n\} \]\n\n(The cases \( i = 0, n \) give \( \varnothing ,\{ 1,\ldots, n\} \in \mathcal{C} \) respectively.) Define a random variabl...
Yes
Corollary 2 [Sperner’s Theorem] Let \( \mathcal{F} \) be an antichain. Then\n\n\[ \left| \mathcal{F}\right| \leq \left( \begin{matrix} n \\ \lfloor n/2\rfloor \end{matrix}\right) \]
Proof. The function \( \left( \begin{array}{l} n \\ x \end{array}\right) \) is maximized at \( x = \lfloor n/2\rfloor \) so that\n\n\[ 1 \geq \mathop{\sum }\limits_{{A \in \mathcal{F}}}\frac{1}{\left( \begin{matrix} n \\ \left| A\right| \end{matrix}\right) } \geq \frac{\left| \mathcal{F}\right| }{\left( \begin{matrix} ...
Yes
Theorem 12.1.1 Let \( \mathcal{A} \) be a family of \( n \) subsets of an \( m \) -set \( \Omega \) . Then\n\n\[ \operatorname{disc}\left( \mathcal{A}\right) \leq \sqrt{{2m}\ln \left( {2n}\right) } \]
Proof. Let \( \chi : \Omega \rightarrow \{ - 1, + 1\} \) be random. For \( A \subset \Omega \) let \( {X}_{A} \) be the indicator random variable for \( \left| {\chi \left( A\right) }\right| > \alpha \) where we set \( \alpha = \sqrt{{2m}\ln \left( {2n}\right) } \) . If \( \left| A\right| = a \) then \( \chi \left( A\r...
Yes
Theorem 12.2.1 [Spencer (1985a) ] Let \( \mathcal{A} \) be a family of \( n \) subsets of an \( n \) -element set \( \Omega \) . Then \[ \operatorname{disc}\left( \mathcal{A}\right) < 6\sqrt{n} \]
With \( \chi : \Omega \rightarrow \{ - 1, + 1\} \) random, \( A \in \mathcal{A},\chi \left( A\right) \) has zero mean and variance at most \( \sqrt{n} \) . If \( \left| {\chi \left( A\right) }\right| > 6\sqrt{n} \) then \( \chi \left( A\right) \) is at least six standard deviations off the mean. The probability of this...
No
Lemma 12.2.3 Let \( \mathcal{A} \) be a family of \( n \) subsets of an \( r \) -set \( \Omega \) with \( r \leq {10}^{-9}n \) . Then there is a partial coloring \( \chi \) of \( \Omega \) with at most \( {10}^{-{40}}r \) points uncolored so that
We outline the argument which leaves room to spare. Let \( {A}_{1},\ldots ,{A}_{n} \) denote the sets of \( \mathcal{A} \) . Let \( \chi : \Omega \rightarrow \{ - 1, + 1\} \) be random. For \( 1 \leq i \leq n \) define\n\n\[ \n{b}_{i} = \text{ nearest integer to }\frac{\chi \left( {A}_{i}\right) }{{20}\sqrt{r}\sqrt{\ln...
No
Theorem 12.3.1 Let \( \mathcal{A} \) be a family of \( n \) sets on \( m \) points with \( m \geq n \) . Suppose that lindisc \( \left( {\left. \mathcal{A}\right| }_{X}\right) \leq K \) for every subset \( X \) of at most \( n \) points. Then \( \overline{\operatorname{lindisc}}\left( \mathcal{A}\right) \leq K \) .
Proof. Let \( {p}_{1},\ldots ,{p}_{m} \in \left\lbrack {0,1}\right\rbrack \) be given. We define a reduction process. Call index \( j \) fixed if \( {p}_{j} \in \{ 0,1\} \), otherwise call it floating, and let \( F \) denote the set of floating indices. If \( \left| F\right| \leq n \) then halt. Otherwise, let \( {y}_{...
Yes
Corollary 12.3.3 Let \( \mathcal{A} \) be a family of \( n \) sets on \( m \) points. Suppose disc \( \left( {\left. \mathcal{A}\right| }_{X}\right) \leq K \) for every subset \( X \) with at most \( n \) points. Then \( \operatorname{disc}\left( \mathcal{A}\right) \leq {2K} \) .
Proof. For every \( X \subseteq \Omega \) with \( \left| X\right| \leq n \), herdisc \( \left( {\left. \mathcal{A}\right| }_{X}\right) \leq K \) so by Theorem 12.3.2 lindisc \( \left( {\left. \mathcal{A}\right| }_{X}\right) \leq K \) . By Theorem 12.3.1 lindisc \( \left( \mathcal{A}\right) \leq \bar{K} \) . But\n\n\[ \...
Yes
For any family \( \mathcal{A} \) of \( n \) sets of arbitrary size
Apply Theorem 12.2.1 and Corollary 12.3.3.
No
Theorem 12.5.1 Let \( \mathcal{A} \) be a finite family of finite sets, no restriction on either the number of sets nor on the cardinality of the sets, with \( \deg \left( \mathcal{A}\right) \leq t \) . Then\n\n\[ \operatorname{disc}\left( \mathcal{A}\right) \leq {2t} - 1 \]
Proof. For convenience write \( \mathcal{A} = \left\{ {{A}_{1},\ldots ,{A}_{m}}\right\} \) with all \( {A}_{i} \subseteq \Omega = \{ 1,\ldots, n\} \) . To each \( j \in \Omega \) there is assigned a value \( {\mathbf{x}}_{\mathbf{j}} \) which will change as the proof progresses. Initially all \( {x}_{j} = 0 \) . At the...
No
Theorem 13.1.1 For every \( d \geq 1 \) there is a set of at least \( \left\lfloor {\frac{1}{2}{\left( \frac{2}{\sqrt{3}}\right) }^{d}}\right\rfloor \) points in the \( d \) -dimensional Euclidean space \( {R}^{d} \), such that all angles determined by three points from the set are strictly less than \( \pi /2 \) .
Proof.[Theorem 13.1.1] We select the points of a set \( X \) in \( {R}^{d} \) from the vertices of the \( d \) -dimensional cube. As usual, we view the vertices of the cube, which are \( 0,1 \) -vectors of length \( d \), as the characteristic vectors of subsets of a \( d \) -element set; i.e., each 0,1-vector \( a \) ...
Yes
Theorem 13.2.1 Let \( {I}_{1},{I}_{2},\ldots ,{I}_{n} \) be parallel unit intervals in the plane, where\n\n\[ \n{I}_{i} = \{ \left( {x, y}\right) : x = i,0 \leq y \leq 1\} .\n\]\n\nFor each \( i \) let us choose a point \( {p}_{i} \) randomly and independently from \( {I}_{i} \) according to a uniform distribution. Let...
Proof. We first estimate the probability that the triangle determined by the points \( {p}_{i},{p}_{i + a} \) and \( {p}_{i + k} \) is empty, for some fixed \( i, a \) and \( k = a + b \geq 3 \) . Let \( A = \left( {i, x}\right) \) , \( B = \left( {i + a, y}\right) \) and \( C = \left( {i + k, z}\right) \) be the point...
Yes
Theorem 13.3.1 For all \( m \geq n \) ,\n\n\[ d\left( {m, n}\right) \leq \left( {n + 1}\right) /2 + \sqrt{\frac{n - 1}{2}\log m} \]
For the proof, we need a definition and two lemmas. For a vector \( \mathbf{a} = \left( {{a}_{1},\ldots ,{a}_{n}}\right) \) of +1,-1-entries, the number of sign-changes in \( \mathbf{a} \) is the number of indices \( i \) , \( 1 \leq i \leq n - 1 \) such that \( {a}_{i} = - {a}_{i + 1} \) . For a matrix \( A \) of \( +...
No
Lemma 13.3.2 For any matrix \( A \) of \( + 1, - 1 \) -entries, \( d\left( A\right) \leq s\left( A\right) + 1 \) .
Proof. Let \( A = \left( {a}_{i, j}\right) \) be an \( m \) by \( n \) matrix of \( + 1, - 1 \) entries and suppose \( s = s\left( A\right) \) . Let \( {t}_{1} < {t}_{2} < \ldots < {t}_{n} \) be arbitrary reals, and define \( n \) points \( {P}_{1},{P}_{2},\ldots ,{P}_{n} \) in \( {R}^{s + 1} \) by: \( {P}_{j} = \left(...
Yes
Lemma 13.3.3 For every \( m \) by \( n \) matrix \( A \) of \( + 1, - 1 \) -entries there is a matrix \( B \) obtained from \( A \) by multiplying some of the columns of \( A \) by -1, such that \( s\left( B\right) \leq \left( {n - 1}\right) /2 + \sqrt{\frac{n - 1}{2}\log m} \) .
Proof. For each column of \( A \), randomly and independently, choose a number \( \epsilon \in \{ + 1, - 1\} \), where each of the two choices is equally likely, and multiply this column by \( \epsilon \) . Let \( B \) be the random sign-matrix obtained in this way. Consider an arbitrary fixed row of \( B \) . One can ...
Yes
Lemma 13.4.1 If \( \\left( {X, R}\\right) \) is a range space of VC-dimension \( d \) with \( \\left| X\\right| = n \) points then \( \\left| R\\right| \\leq g\\left( {d, n}\\right) \) .
Proof. We apply induction on \( n + d \) . The assertion is trivially true for \( d = 0 \) and \( n = 0 \) . Assuming it holds for \( n \) and \( d - 1 \) and for \( n - 1 \) and \( d - 1 \) we prove it for \( n \) and \( d \) . Let \( S = \\left( {X, R}\\right) \) be a range space of VC-dimension \( d \) on \( n \) po...
Yes
Corollary 13.4.3 Let \( \left( {X, R}\right) \) be a range space of VC-dimension \( d \geq 2 \), and let \( \left( {X,{R}_{h}}\right) \) be the range space on \( X \) in which \( {R}_{h} = \left\{ {\left( {{r}_{1} \cap \ldots \cap {r}_{h}}\right) : {r}_{1},\ldots {r}_{h} \in R}\right\} \) . Then \( {VC}\left( {X,{R}_{h...
Proof. Let \( A \) be an arbitrary subset of cardinality \( n \) of \( X \) . By Corollary 13.4.2 \( \left| {{P}_{R}\left( A\right) }\right| \leq g\left( {d, n}\right) \leq {n}^{d} \) . Since each member of \( {P}_{{R}_{h}}\left( A\right) \) is an intersection of \( h \) members of \( {P}_{R}\left( A\right) \) it follo...
Yes
Theorem 13.4.4 There is a positive constant \( c \) such that if \( \left( {X, R}\right) \) is any range-space of VC-dimension at most \( d, A \subset X \) is a finite subset and \( \epsilon ,\delta > 0 \), then a random subset \( B \) of cardinality \( s \) of \( A \) where \( s \) is at least the minimum between \( \...
\[ \frac{c}{{\epsilon }^{2}}\left( {\operatorname{dlog}\frac{d}{\epsilon } + \log \frac{1}{\delta }}\right) \] is an \( \epsilon \) -sample for \( A \) with probability at least \( 1 - \delta \) .
Yes
Theorem 13.4.5 Let \( \\left( {X, R}\\right) \) be a range space of VC-dimension \( d \), let \( A \) be a finite subset of \( X \) and supppose \( 0 < \\epsilon ,\\delta < 1 \) . Let \( N \) be a set obtained by \( m \) random independent draws from \( A \), where\n\n\[ m \\geq {max}\\left( {\\frac{4}{\\epsilon }{log}...
Proof.[Theorem 13.4.5] Let \( \\left( {X, R}\\right) \) be a range space with VC-dimension \( d \), and let \( A \) be a subset of \( X \) of cardinality \( \\left| A\\right| = n \) . Suppose \( m \) satisfies (13.2), and let \( N = \\left( {{x}_{1},\\ldots ,{x}_{m}}\\right) \) be obtained by \( m \) independent random...
Yes
Lemma 13.5.2 Let \( \\left( {A,\\mathcal{F}}\\right), n, h, t \) and \( c \) be as above, let \( B \) be a finite subset of \( p > 1 \) points of \( A \), and let \( \\mathcal{G} \) be a collection of \( m \) (not necessarily distinct) members of \( \\mathcal{F} \). Then there are two distinct points \( \\mathbf{x}, y ...
Proof. We may and will assume that \( p \) is larger than \( c + 1 \). Let \( g \) be the largest integer such that \( c{g}^{t} \\leq p - 1 \), that is, \( g = \\left\\lfloor {\\left( \\frac{p - 1}{c}\\right) }^{1/t}\\right\\rfloor \). Let \( L \) be a random collection of \( g \) members of \( \\mathcal{G} \), each pi...
Yes
Theorem 13.5.3 Let \( \left( {A,\mathcal{F}}\right) \) be a finite range space, where \( \left| A\right| = n \), and suppose that its dual shatter function \( h \) satisfies \( h\left( g\right) \leq c{g}^{t} \) for some fixed \( c, t > 0 \) . Then, there is a \( {C}^{\prime } = {C}^{\prime }\left( {c, t}\right) \) such...
Proof. Without loss of generality, assume that the number of points of \( A \) is even (otherwise, simply omit a point). By Theorem 13.5.1 there is a Hamilton path \( {x}_{1}{x}_{2}\ldots {x}_{n} \) on these points such that each member of \( \mathcal{F} \) is stabbed by at most \( C{n}^{1 - 1/t}\log n \) edges of the ...
Yes
Theorem 1 Let \( C \) be bounded, convex and centrally symmetric about the origin. Then\n\n\[ \delta \left( C\right) \geq {2}^{-\left( {n - 1}\right) } \]
Proof. Fix \( \epsilon > 0 \) . Normalize so \( \mu = \mu \left( C\right) = 2 - \epsilon \) . For any real \( z \) let \( {C}_{z} \) denote the \
No
Theorem 14.1.1 [Shannon’s Theorem] Let \( p \in \left( {0,{.5}}\right) \) be fixed. For \( \epsilon > 0 \) arbitrarily small there exists a Coding Scheme with rate of transmission greater than \( 1 - H\left( p\right) - \epsilon \) and probability of incorrect transmission less than \( \epsilon \) .
Proof. Let \( \delta > 0 \) be such that \( p + \delta < {.5} \) and \( H\left( {p + \delta }\right) < H\left( p\right) + \epsilon /2 \) . For \( n \) large set \( m = n\left( {1 - H\left( p\right) - \epsilon }\right) \), guaranteeing the rate of transmission. Let \( f \) : \( \{ 0,1{\} }^{m} \rightarrow \{ 0,1{\} }^{n...
Yes
Theorem 14.1.2 Let \( p \in \left( {0,{.5}}\right) \) be fixed. For \( \epsilon > 0 \) arbitrarily small there exists a Group Code with rate of transmission greater than \( 1 - H\left( p\right) - \epsilon \) and probability of incorrect transmission less than \( \epsilon \) .
Proof. For \( 1 \leq i \leq m \) let \( {u}_{i} \in \{ 0,1{\} }^{m} \) be that vector with a one in position \( i \) , all other entries zero. Let \( f\left( {u}_{1}\right) ,\ldots, f\left( {u}_{m}\right) \) be chosen randomly and independently and then extend \( f \) by setting\n\n\[ f\left( {{\epsilon }_{1}{u}_{1} + ...
Yes
Corollary 14.2.2 If\n\n\[ n > \frac{{2}^{q}}{\mathop{\sum }\limits_{{i = 0}}^{k}\left( \begin{matrix} q \\ i \end{matrix}\right) } \]\n\nthen Carole wins the \( \left( {n, q, k}\right) \) -Liar Game.
Proof.[Theorem 14.2.1] Fix a strategy for Paul. Now Carole plays randomly! That is, at each round, after Paul has selected a set \( S \) of chips Carole flips a coin - if it comes up heads she moves every chip not in \( S \) one position to the left and if it comes up tails she moves every chip in \( S \) one position ...
Yes
Theorem 14.3.1 If \( \mathop{\sum }\limits_{i}{x}_{i}{2}^{-i} < 1 \) then Carole wins the \( \left( {{x}_{1},\ldots ,{x}_{k}}\right) \) -Tenure Game.
Proof. Fix a strategy for Paul. Now Carole plays randomly! That is, at each round, after Paul has selected a set \( S \) of chips Carole flips a coin - if it comes up heads she moves every chip not in \( S \) one position to the left and if it comes up tails she moves every chip in \( S \) one position to the left. For...
Yes
Lemma 14.3.2 If a set of chips has weight at least one it may be split into two parts each of weight at least one half.
Proof. There must be two chips at some position \( i \), otherwise the weight is less than one. If there are two chips at position 1 simply split them. If there are two chips at position \( i > 1 \) glue them together, and consider them as one superchip at position \( i - 1 \) . Then the proof follows by induction on t...
No
Theorem 14.3.3 If \( \sum {x}_{i}{2}^{-i} \geq 1 \) then Paul wins the \( \left( {{x}_{1},\ldots ,{x}_{k}}\right) \) -Tenure Game.
Proof. The initial weight is at least one. Applying the Lemma Paul splits the chips into two parts each of weight at least one half and sets \( S \) equal one of the parts. Carole moves all chips in one part one position to the left, doubling their weight, leaving a new position of weight at least one. Thus the weight ...
Yes
Theorem 14.4.1 If \( \Pr \left\lbrack {\left| {S}_{n}\right| > \alpha }\right\rbrack < {n}^{-1} \) then \( \operatorname{VAL}\left( n\right) \leq \alpha \) .
Proof. Consider the game a win for Pusher if the final \( {\left| P\right| }_{\infty } > \alpha \) . Suppose Chooser announces that she will flip a fair coin each round to determine whether to reset \( P \) as \( P + v \) or \( P - v \) . Let \( {x}_{i} \) be the \( i \) -th coordinate for the final value of the positi...
Yes
Lemma 14.6.1 Let \( X, Y \) and \( Z \) be three random variables taking values in \( S, T \) and \( U \), respectively. Then\n\n1. \( H\left( X\right) \leq {\log }_{2}\left| S\right| \) .\n\n2. \( H\left( {X, Y}\right) \geq H\left( X\right) \) .
## Proof.\n\n1. Since the function \( \log z \) is concave it follows, by Jensen’s Inequality, that\n\n\[ H\left( X\right) = \mathop{\sum }\limits_{{i \in S}}P\left( {X = i}\right) \log \left( \frac{1}{P\left( {X = i}\right) }\right) \]\n\n\[ \leq \log \left( {\mathop{\sum }\limits_{{i \in S}}P\left( {X = i}\right) \fr...
Yes
Proposition 14.6.2 Let \( X = \left( {{X}_{1},\ldots ,{X}_{n}}\right) \) be a random variable taking values in the set \( S = {S}_{1} \times {S}_{2} \times \ldots \times {S}_{n} \), where each of the coordinates \( {X}_{i} \) of \( X \) is a random variable taking values in \( {S}_{i} \). Then\n\n\[ H\left( X\right) \l...
Proof. This follows by induction from Lemma 14.6.1, part 3. ∎
No
Corollary 14.6.3 Let \( \mathcal{F} \) be a family of subsets of \( \{ 1,2,\ldots, n\} \) and let \( {p}_{i} \) denote the fraction of sets in \( \mathcal{F} \) that contain \( i \) . Then\n\n\[ \left| \mathcal{F}\right| \leq {2}^{\mathop{\sum }\limits_{{i = 1}}^{n}H\left( {p}_{i}\right) } \]\n\nwhere \( H\left( y\righ...
Proof. Associate each set \( F \in \mathcal{F} \) with its characteristic vector \( v\left( F\right) \), which is a binary vector of length \( n \) . Let \( X = \left( {{X}_{1},\ldots ,{X}_{n}}\right) \) be the random variable taking values in \( \{ 0,1{\} }^{n} \), where \( P\left( {X = v\left( F\right) }\right) = 1/\...
Yes
Proposition 14.6.4 Let \( X = \left( {{X}_{1},\ldots ,{X}_{n}}\right) \) and \( S \) be as above. If \( \mathcal{G} \) is a family of subsets of \( \{ 1,\ldots, n\} \) and each \( i \in \{ 1,\ldots, n\} \) belongs to at least \( k \) members of \( \mathcal{G} \) then\n\n\[ \n{kH}\left( X\right) \leq \mathop{\sum }\limi...
Proof. We apply induction on \( k \) . For \( k = 1 \), replace each set \( G \in \mathcal{G} \) by a subset of it to obtain a family \( {\mathcal{G}}^{\prime } \) whose members form a partition of \( \{ 1,\ldots, n\} \) . By Lemma 14.6.1, part 2, \( \mathop{\sum }\limits_{{G \in \mathcal{G}}}H\left( {X\left( G\right) ...
Yes
Corollary 14.6.5 Let \( \mathcal{F} \) be a family of vectors in \( {S}_{1} \times {S}_{2}\ldots \times {S}_{n} \) . Let \( \mathcal{G} = \) \( \left\{ {{G}_{1},{G}_{2},\ldots {G}_{m}}\right\} \) be a collection of subsets of \( N = \{ 1,2,\ldots, n\} \), and suppose that each element \( i \in N \) belongs to at least ...
Proof. Let \( X = \left( {{X}_{1},\ldots ,{X}_{n}}\right) \) be the random variable taking values in \( \mathcal{F} \), where \( P\left( {X = F}\right) = \frac{1}{\left| \mathcal{F}\right| } \) for all \( F \in \mathcal{F} \) . By Proposition 14.6.4\n\n\[ \n{kH}\left( X\right) \leq \mathop{\sum }\limits_{{i = 1}}^{m}H\...
Yes
Corollary 14.6.7 [Chung et al. (1986) ] Let \( N \) be a finite set, and let \( \mathcal{F} \) be a family of subsets of \( N \) . Let \( \mathcal{G} = \left\{ {{G}_{1},\ldots {G}_{m}}\right\} \) be a collection of subsets of \( N \), and suppose that each element of \( S \) belongs to at least \( k \) members of \( \m...
\[ {\left| \mathcal{F}\right| }^{k} \leq \mathop{\prod }\limits_{{i = 1}}^{m}\left| {\mathcal{F}}_{i}\right| \]
Yes
Corollary 14.6.8 Let \( \\mathcal{F} \) be a family of graphs on the labeled set of vertices \( \\{ 1,2,\\ldots t\\} \), and suppose that for any two members of \( \\mathcal{F} \) there is a triangle contained in both of them. Then\n\n\[ \n\\left| \\mathcal{F}\\right| < \\frac{1}{4}{2}^{\\left( \\begin{array}{l} t \\\\...
Proof. Let \( N \) be the set of all \( \\left( \\begin{array}{l} t \\\\ 2 \\end{array}\\right) \) unordered pairs of vertices in \( T = \\{ 1,2\\ldots, t\\} \) , and consider \( \\mathcal{F} \) as a family of subsets of \( N \) . Let \( \\mathcal{G} \) be the family of all subsets of \( N \) consisting of the edge-set...
Yes
Theorem 15.2.1 Suppose \( n = {2}^{k} - 1 \) and \( d = {2t} + 1 \) . Then there exists a symmetric probability space \( \Omega \) of size \( 2{\left( n + 1\right) }^{t} \) and \( d \) -wise independent random variables \( {y}_{1},\ldots ,{y}_{n} \) over \( \Omega \) each of which takes the values 0 and 1 with probabil...
The space and the variables are explicitly constructed, given a representation of the field \( F = {GF}\left( {2}^{k}\right) \) as a \( k \) -dimensional algebra over \( {GF}\left( 2\right) \) . Proof. Let \( {x}_{1},\ldots ,{x}_{n} \) be the \( n \) nonzero elements of \( F \), represented as column-vectors of length ...
Yes
Lemma 15.2.2 Any set of \( d = {2t} + 1 \) columns of \( H \) is linearly independent over \( {GF}\left( 2\right) \) .
Proof. Let \( J \subset \{ 1,2,\ldots, n\} \) be a subset of cardinality \( \left| J\right| = {2t} + 1 \) of the set of indices of the columns of \( H \) . Suppose that \( \mathop{\sum }\limits_{{j \in J}}{z}_{j}{H}_{j} = 0 \), where \( {H}_{j} \) denotes the \( j \) -th column of \( H \) and \( {z}_{j} \in {GF}\left( ...
Yes
Proposition 15.2.3 If the random variables \( {y}_{1},\ldots ,{y}_{n} \) over the sample space \( \Omega \) are \( d \) -wise independent and none of them is almost constant then \( \left| \Omega \right| \geq m\left( {n, d}\right) \) .
Proof. Clearly we may assume that the expected value of each \( {y}_{j} \) is 0 (since otherwise we can replace \( {y}_{j} \) by \( {y}_{j} - E\left( {y}_{j}\right) \) ). For each subset \( S \) of \( \{ 1,\ldots, n\} \), define \( {\alpha }_{S} = \mathop{\prod }\limits_{{j \in S}}{y}_{j} \) . Observe that since no \( ...
Yes
Theorem 1 The crossing number of any simple graph \( G = \left( {V, E}\right) \) with \( \left| E\right| \geq 4\left| V\right| \) is at least \( \frac{{\left| E\right| }^{3}}{{64}{\left| V\right| }^{2}} \) .
Proof. By Euler’s formula any simple planar graph with \( n \) vertices has at most \( {3n} - 6 \) edges, implying that the crossing number of any simple graph with \( n \) vertices and \( m \) edges is at least \( m - \left( {{3n} - 6}\right) > m - {3n} \) . Let \( G = \left( {V, E}\right) \) be a graph with \( \left|...
Yes
Theorem 2 Let \( P \) be a set of \( n \) distinct points in the plane, and let \( L \) be a set of \( m \) distinct lines. Then, the number of incidences between the members of \( P \) and those of \( L \) (that is, the number of pairs \( \left( {p, l}\right) \) with \( p \in P, l \in L \) and \( p \in l \) ) is at mo...
Proof. Denote the number of incidences by \( I \) . Let \( G = \left( {V, E}\right) \) be the graph whose vertices are all members of \( P \), where two are adjacent if and only if they are consecutive points of \( P \) on some line in \( L \) . Clearly, \( \left| V\right| = n \) and \( \left| E\right| = I - m \) . Not...
Yes
Theorem 3 For any three sets \( A, B \) and \( C \) of \( s \) real numbers each,\n\n\[ \left| {A \cdot B + C}\right| = \left| {\{ {ab} + c : a \in A, b \in B, c \in C\} }\right| \geq \Omega \left( {s}^{3/2}\right) . \]\n
Proof. Put \( R = A \cdot B + C,\left| R\right| = r \) and define\n\n\[ P = \{ \left( {a, t}\right) : a \in A, t \in R\} ,\;L = \{ y = {bx} + c : b \in B, c \in C\} . \]\n\nThus \( P \) is a set of \( n = {sr} \) points in the plane, \( L \) is a set of \( m = {s}^{2} \) lines in the plane, and each line \( y = {bx} + ...
Yes
Proposition 1 Let \( G = \left( {V, E}\right) \) be a triangle-free graph on \( n \) vertices with maximum degree at most \( d \geq 1 \) . Then\n\n\[ \alpha \left( G\right) \geq \frac{n\log d}{8d} \]\n\nwhere the logarithm here and in what follows is in base 2.
Proof. If, say, \( d < {16} \) the result follows from the trivial bound \( \alpha \left( G\right) \geq n/\left( {d + 1}\right) \) and hence we may and will assume that \( d \geq {16} \) . Let \( W \) be a random independent set of vertices in \( G \), chosen uniformly among all independent sets in \( G \) . For each v...
Yes
Theorem 2 [Ajtai et al. (1980) ] There exists an absolute constant \( b \) such that \( r\left( {3, k}\right) \leq b{k}^{2}/\log k \) for every \( k > 1 \) .
Proof. Let \( G = \left( {V, E}\right) \) be a triangle-free graph on \( 8{k}^{2}/\log k \) vertices. If \( G \) has a vertex of degree at least \( k \) then its neighborhood contains an independent set of size \( k \) . Otherwise, by proposition 1 above, \( G \) contains an independent set of size at least \( \frac{8{...
Yes
Proposition 1.1 Let \( {\left( {X}_{n}\right) }_{n \geq 1} \) be a sequence of real random variables such that, for every \( n \geq 1,{X}_{n} \) follows the \( \mathcal{N}\left( {{m}_{n},{\sigma }_{n}^{2}}\right) \) -distribution. Suppose that \( {X}_{n} \) converges in \( {L}^{2} \) to X. Then:\n\n(i) The random varia...
(i) The convergence in \( {L}^{2} \) implies that \( {m}_{n} = E\left\lbrack {X}_{n}\right\rbrack \) converges to \( E\left\lbrack X\right\rbrack \) and \( {\sigma}_{n}^{2} = \) \( \operatorname{var}\left( {X}_{n}\right) \) converges to \( \operatorname{var}\left( X\right) \) as \( n \rightarrow \infty \) . Then, setti...
Yes
Proposition 1.2 Under the preceding assumptions, the random variables \( {X}_{1},\ldots ,{X}_{d} \) are independent if and only if the covariance matrix \( {\left( \operatorname{cov}\left( {X}_{j},{X}_{k}\right) \right) }_{1 \leq j, k \leq d} \) is diagonal or equivalently if and only if \( {q}_{X} \) is of diagonal fo...
Proof If the random variables \( {X}_{1},\ldots ,{X}_{d} \) are independent, the covariance matrix \( {\left( \operatorname{cov}\left( {X}_{j},{X}_{k}\right) \right) }_{j, k = 1,\ldots d} \) is diagonal. Conversely, if this matrix is diagonal, we have for every \( u = \mathop{\sum }\limits_{{j = 1}}^{d}{u}_{j}{e}_{j} \...
Yes
Proposition 1.7 If \( {\left( {X}_{t}\right) }_{t \in T} \) is a Gaussian process, the closed linear subspace of \( {L}^{2} \) spanned by the variables \( {X}_{t}, t \in T \), is a Gaussian space, which is called the Gaussian space generated by the process \( X \) .
Proof It suffices to observe that an \( {L}^{2} \) -limit of centered Gaussian variables is still centered Gaussian, by Proposition 1.1.
No
Theorem 1.9 Let \( H \) be a centered Gaussian space and let \( {\left( {H}_{i}\right) }_{i \in I} \) be a collection of linear subspaces of \( H \) . Then the subspaces \( {H}_{i}, i \in I \), are (pairwise) orthogonal in \( {L}^{2} \) if and only the \( \sigma \) -fields \( \sigma \left( {H}_{i}\right), i \in I \), a...
Proof Suppose that the \( \sigma \) -fields \( \sigma \left( {H}_{i}\right) \) are independent. Then, if \( i \neq j \), if \( X \in {H}_{i} \) and \( Y \in {H}_{j} \) ,\n\n\[ E\left\lbrack {XY}\right\rbrack = E\left\lbrack X\right\rbrack E\left\lbrack Y\right\rbrack = 0, \]\n\nso that the linear spaces \( {H}_{i} \) a...
Yes
Corollary 1.10 Let \( H \) be a (centered) Gaussian space and let \( K \) be a closed linear subspace of \( H \) . Let \( {p}_{K} \) denote the orthogonal projection onto \( K \) in the Hilbert space \( {L}^{2} \), and let \( X \in H \) . (i) We have \[ E\left\lbrack {X \mid \sigma \left( K\right) }\right\rbrack = {p}_...
## Proof (i) Let \( Y = X - {p}_{K}\left( X\right) \) . Then \( Y \) is orthogonal to \( K \) and, by Theorem \( {1.9}, Y \) is independent of \( \sigma \left( K\right) \) . Then, \[ E\left\lbrack {X \mid \sigma \left( K\right) }\right\rbrack = E\left\lbrack {{p}_{K}\left( X\right) \mid \sigma \left( K\right) }\right\r...
Yes
Theorem 1.11 Let \( \Gamma \) be a symmetric function of positive type on \( T \times T \) . There exists, on an appropriate probability space \( \left( {\Omega ,\mathcal{F}, P}\right) \), a centered Gaussian process whose covariance function is \( \Gamma \) .
Example Consider the case \( T = \mathbb{R} \) and let \( \mu \) be a finite measure on \( \mathbb{R} \), which is also symmetric (i.e. \( \mu \left( {-A}\right) = \mu \left( A\right) \) ). Then set, for every \( s, t \in \mathbb{R} \) ,\n\n\[ \Gamma \left( {s, t}\right) = \int {\mathrm{e}}^{\mathrm{i}\xi \left( {t - s...
Yes