Past exam of the mathematics course of the University of Cambridge / 2016 / iii / Paper 109 (Combinatorics)

For a family $\mathcal{F}\subseteq \left(\genfrac{}{}{0ex}{}{[n]}{r}\right)$, let its lower shadow $\partial \mathcal{F}$ consist of all $(r-1)$-sets contained in members of $\mathcal{F}$. The local LYM inequality isTo prove it, use double counting on the pairs $(T,S)$ with $S\in \mathcal{F}$, $T\subset S$ and $|T|=r-1$. Each $S$ contributes $r$ pairs; each $T\in \partial \mathcal{F}$ contributes at most $n-r+1$. Hence $r|\mathcal{F}|\le (n-r+1)|\partial \mathcal{F}|$, exactly the displayed Local LYM inequality after using $r\left(\genfrac{}{}{0ex}{}{n}{r}\right)=(n-r+1)\left(\genfrac{}{}{0ex}{}{n}{r-1}\right)$.

To deduce the LYM inequality, start with an antichain $\mathcal{A}$ in the Boolean lattice and assign it weight ${\sum }_{S\in \mathcal{A}}{\left(\genfrac{}{}{0ex}{}{n}{|S|}\right)}^{-1}$. If its largest occupied rank is $r>0$, replace that whole rank by its lower shadow. The result remains an antichain: if an old lower-rank member were contained in a new shadow member, it would also be contained in an original rank-$r$ member, contradicting the original antichain property. Nor can a new member be contained in an old one, since every old member has size at most $r-1$ and equality would give the same contradiction. The Local LYM inequality shows that this replacement does not decrease the weight. Repeating eventually leaves an antichain on rank zero, whose weight is at most one. ThereforeFor the empty family the conclusion is immediate; the same argument includes the singleton family containing the empty set.

The antichains of maximum size are exactly the full middle levels. More precisely, for even $n$ only $\left(\genfrac{}{}{0ex}{}{[n]}{n/2}\right)$ attains the bound in Sperner theorem; for odd $n$ either $\left(\genfrac{}{}{0ex}{}{[n]}{(n-1)/2}\right)$ or $\left(\genfrac{}{}{0ex}{}{[n]}{(n+1)/2}\right)$ does. To prove that there are no others, put $M=\left(\genfrac{}{}{0ex}{}{n}{\lfloor n/2\rfloor }\right)$. If $|\mathcal{A}|=M$, the LYM inequality and $\left(\genfrac{}{}{0ex}{}{n}{r}\right)\le M$ imply equality in every term's bound $1/\left(\genfrac{}{}{0ex}{}{n}{r}\right)\ge 1/M$. Thus every member lies on a middle level. For even $n$, there is only one such level, so it must be full.

For odd $n=2r+1$, let $X$ be the rank-$r$ members and $Y$ the rank-$(r+1)$ members. Consider the bipartite graph of inclusions between those levels. Both parts have size $M$, and the degree of a vertex is $r+1$. If $N(X)$ is the neighbourhood of $X$, double counting gives $|N(X)|\ge |X|$. Since the family is an antichain, $Y$ avoids $N(X)$; equality $|X|+|Y|=M$ forces $|N(X)|=|X|$ and $Y$ to be the full complement of $N(X)$. Equality in the degree count means that every edge into $N(X)$ comes from $X$, so there are no edges between $X\cup N(X)$ and its complement.

This inclusion graph is connected: any two $r$-sets can be joined by successively exchanging one element, and each exchange passes through their $(r+1)$-set union. Every upper vertex has a lower neighbour. Thus $X$ must be empty or the whole lower level, proving the classification. For $n=1$ the graph is just one edge, and for $n=0$ the unique maximum family is $\{\varnothing \}$. This establishes equality in Sperner theorem without assuming that equality in the LYM inequality alone always forces a middle level.

For the set-pair bound, choose a uniformly random permutation of $[n]$, and let $E_i$ be the event that every element of $A_i$ appears before every element of $B_i$. Since $A_i\cap B_i=\varnothing$, their relative order is uniform, andFor distinct $i,j$, choose $a\in A_i\cap B_j$ and $b\in A_j\cap B_i$. These are distinct because $A_i\cap B_i$ is empty. On $E_i$, $a$ precedes $b$; on $E_j$, $b$ precedes $a$. Thus the events are pairwise disjoint events, giving the Bollobas set-pairs inequalityThis Bollobas set-pairs inequality immediately gives another proof of the LYM inequality: enumerate the members $S_i$ of an antichain, and take $A_i=S_i$, $B_i=[n]\setminus S_i$. For $i\ne j$, incomparability means $S_i\setminus S_j\ne \varnothing$, while $A_i\cap B_i=\varnothing$. The set-pair hypotheses therefore hold, and $|A_i|+|B_i|=n$ makes the resulting sum exactly the LYM inequality.

Identify the Boolean hypercube $Q_n$ with the subsets of $[n]$. For a family $\mathcal{H}$, write $N(\mathcal{H})$ for its closed vertex neighbourhood, consisting of the family and all vertices at Hamming distance one from it. Order subsets first by increasing size and, within a fixed size, use lexicographic order: at the smallest coordinate on which they differ, the set containing that coordinate comes first. This is the simplicial order on the discrete cube. Harper's theorem states that if $I$ is its initial segment of size $|\mathcal{H}|$, thenThus simplicial initial segments minimize the external vertex boundary for each prescribed family size. The closed-neighbourhood and external-boundary formulations are equivalent because $|N(\mathcal{H})\setminus \mathcal{H}|=|N(\mathcal{H})|-|\mathcal{H}|$. No edge boundary is involved in this assertion.

We now give the deduction Harper theorem implies the Kruskal-Katona theorem, keeping track of the two different orders. For a family $\mathcal{F}$ of $r$-sets with $r\ge 1$, adjoin every smaller set:Its closed vertex neighbourhood is exactlywhere $\nabla \mathcal{F}$ is the upper shadow. Applying Harper theorem to $\mathcal{H}$ shows that, among $r$-uniform families of fixed size, the lexicographic initial segment minimizes the upper shadow: all the lower-level contributions to the two neighbourhood sizes are identical.

To convert this into a statement about the lower shadow of a $k$-uniform family $\mathcal{G}$, take $r=n-k$ and map each member $S$ to $R([n]\setminus S)$, where $R(i)=n+1-i$ reverses the ground-set coordinates. Complementation changes lower shadows to upper shadows, and the coordinate reversal preserves their sizes. Moreover, the resulting lexicographic order corresponds exactly to colexicographic order on the original $k$-sets: $S$ precedes $T$ in colexicographic order precisely when the largest element of $S△T$ lies in $T$. After complementation and reversal, the smallest differing coordinate lies in the image of $S$, as required for the stated lexicographic order. Hence a colexicographic initial segment minimizes the lower shadow, which is the Kruskal-Katona theorem. The cases $k=n$ and $k=0$ are immediate and need no adjoining construction.

For its usual numerical form, write the unique greedy binomial representationThen the Kruskal-Katona theorem saysHere is why the shadow of the colexicographic initial segment has exactly that size. Its first block consists of all $k$-sets in $[a_k]$. If any members remain, they have the form $\{a_k+1\}\cup T$, where $T$ runs through a colexicographic initial segment of $(k-1)$-sets. The lower shadow consists of all $(k-1)$-sets in $[a_k]$, together with $\{a_k+1\}$ adjoined to the lower shadow of that remainder. These two blocks are disjoint. Recursing gives precisely the displayed sum. For $m=0$, both the initial segment and its shadow are empty.

The Erdős-Ko-Rado theorem states that an intersecting family $\mathcal{A}\subseteq \left(\genfrac{}{}{0ex}{}{[n]}{r}\right)$, with $1\le r\le n/2$, satisfiesThe bound is attained by all $r$-sets containing one fixed element. For the proof Erdős-Ko-Rado theorem from shadows, take complements to obtain an $(n-r)$-uniform family $\mathcal{B}$ of the same size, and let $D$ be its rank-$r$ lower shadow. Every member of $D$ is disjoint from some member of $\mathcal{A}$, so $D\cap \mathcal{A}=\varnothing$. Iterating the Kruskal-Katona theorem bounds $|D|$ below by the rank-$r$ shadow of the colexicographic initial segment of size $|\mathcal{B}|$. This iteration is valid because lower shadows of colexicographic initial segments are again colexicographic initial segments, as the preceding block description shows.

If $|\mathcal{A}|>\left(\genfrac{}{}{0ex}{}{n-1}{r-1}\right)=\left(\genfrac{}{}{0ex}{}{n-1}{n-r}\right)$, that initial segment contains all $(n-r)$-sets in $[n-1]$ and at least one further set containing $n$. Since $r\le n-r\le n-1$, its rank-$r$ shadow contains every $r$-set in $[n-1]$ and at least one $r$-set containing $n$. Thus $|D|>\left(\genfrac{}{}{0ex}{}{n-1}{r}\right)$, implyingcontrary to their disjointness. This proves the bound. If $r>n/2$, all $r$-sets intersect and the trivial maximum is instead $\left(\genfrac{}{}{0ex}{}{n}{r}\right)$.

For an intersecting family in a product alphabet, label the alphabet by $\mathbb{Z}/k\mathbb{Z}$. Partition the words into the classesEach class has $k$ words, and any two distinct words in it disagree in every coordinate. An intersecting family therefore contains at most one word per class. There are $k^{n-1}$ classes, and fixing the first coordinate gives an intersecting family of that size. Consequently the exact maximum isFor $k=1$ this just says that the single word may be included.

Finally, for a weakly intersecting family in a product alphabet, associate to each word its support $S(x)=\{i:x_i>1\}$. The occurring supports form an intersecting family of nonempty subsets, and the fibre over support $S$ has $(k-1{)}^{|S|}$ words. Put $w=k-1$. When $k\ge 2$, $w\ge 1$, and a support family can contain at most one of each complementary pair $S,[n]\setminus S$. For odd $n$, the larger support in each pair has size greater than $n/2$, soAll supports of size greater than $n/2$ are mutually intersecting, so taking every word over those supports achieves this bound. Thus the required extremal family is the strict-majority support family, with sizeThis is the weighted intersecting family bound on an odd Boolean lattice. For $k>2$, strict inequality between the complementary fibre weights forces this extremal family to be unique. For $k=2$, it is a largest family but need not be the only one; for example, all supports containing one fixed coordinate also attain the same size. For $k=1$, no word has a nonempty support, so the maximum is zero, again agreeing with the formula.

Give a vertex $S\subseteq [n]$ the binary value $\nu (S)={\sum }_{i\in S}{2}^{i-1}$. The binary order on the discrete cube orders vertices by increasing $\nu$; its initial segment $I_m$ consists of the values $0,\dots ,m-1$. Write $e(\mathcal{A})$ for the number of cube edges whose two endpoints lie in $\mathcal{A}$, and ${\partial }_e\mathcal{A}$ for its external edge boundary. The degree sum identity isThe exact edge-isoperimetric inequality iswhere $\mathrm{wt}(j)$ is the number of ones in the binary expansion of $j$. Equality is attained by $I_m$. Indeed, $I_m$ is a down-set, so every downward neighbour of each vertex is also present. Counting each internal edge at its upper endpoint gives $e(I_m)={\sum }_{j<m}\mathrm{wt}(j)$. We prove the equivalent edge-isoperimetric theorem for binary initial segments, $e(\mathcal{A})\le e(I_m)$, by mathematical induction using section compression in binary order.

Fix a coordinate $i$ and write ${\mathcal{A}}_0,{\mathcal{A}}_1\subseteq Q_{n-1}$ for the two sections with coordinate $i$ absent or present, deleting that coordinate. Their inherited binary order is the binary order on the remaining coordinates. Replace each section by the binary initial segment of its own size. This preserves $|\mathcal{A}|$. The edges split into internal edges in the two sections and crossing edges, givingBy the mathematical induction hypothesis, the two section edge counts do not decrease. After compression, the sections are nested initial segments, so their intersection has the largest possible size, $\min \{|{\mathcal{A}}_0|,|{\mathcal{A}}_1|\}$. Thus each section compression does not decrease the internal edge count.

Apply these compressions repeatedly whenever one changes the family. The integer potential ${\sum }_{S\in \mathcal{A}}\nu (S)$ decreases strictly at every nontrivial step: within either fixed-coordinate section, the initial segment uniquely minimizes the sum of the vertex values among sets of the prescribed size. Therefore the process terminates with a family compressed in every section.

We need a small structural fact about families compressed in every binary section. If such a family is not a binary initial segment, choose a missing vertex $x$ and a present vertex $y$ with $\nu (x)<\nu (y)$. They cannot agree in any coordinate: in that coordinate's section, the compressed initial segment would contain $x$ whenever it contains $y$. Thus $y=[n]\setminus x$. The same argument applies to every inversion. There cannot be any vertex strictly between $x$ and $y$ in binary order: a present one would form an inversion with $x$, and a missing one would form an inversion with $y$, forcing it to be $y$ or $x$, respectively. Hence $\nu (y)=\nu (x)+1$. Together with $\nu (x)+\nu (y)=2^n-1$, this givesEvery vertex below $x$ must be present, and every vertex above $y$ absent, by the same inversion argument. Therefore the only possible exception to a binary initial segment isFor $n\ge 2$, this replaces the top vertex of an $(n-1)$-dimensional subcube by the singleton $\{n\}$. Removing that vertex loses $n-1$ internal edges, and adding the singleton creates exactly one edge, to the empty set. ThusThe dimensions $n=0,1$ are immediate. This finishes the mathematical induction: compression leads either to the desired initial segment or to an exception with no larger edge count. Hence the exact edge-isoperimetric inequality in the discrete cube holds, and binary initial segments maximize the number of contained edges at every size.

For the square-face assertion, let $f(\mathcal{A})$ count the two-dimensional cube faces all of whose vertices are in $\mathcal{A}$. We use mathematical induction again, now also using the edge theorem just proved. Faces lying wholly in one coordinate section contribute $f({\mathcal{A}}_0)+f({\mathcal{A}}_1)$. Every face crossing coordinate $i$ is determined by an edge lying in both sections, soCompress the two sections as before. Their internal face counts do not decrease by mathematical induction. Put $c=|{\mathcal{A}}_0\cap {\mathcal{A}}_1|$ and $b=\min \{|{\mathcal{A}}_0|,|{\mathcal{A}}_1|\}$. The edge theorem and nesting of binary initial segments giveAfter compression, the intersection is precisely $I_b$. Thus section compression does not decrease the face count either. The same decreasing potential gives termination, and the same classification leaves only a binary initial segment or ${\mathcal{E}}_n$.

In the exceptional family, removing the top vertex loses $\left(\genfrac{}{}{0ex}{}{n-1}{2}\right)$ square faces, one for each pair of downward coordinate directions. The added singleton lies in no square face, because it is the only vertex in its upper section. ConsequentlyDimensions below two have no square faces and start the mathematical induction. We have proved that binary initial segments maximize contained square faces:For an explicit value, the down-set property counts every face at its unique top vertex, givingThis establishes binary initial segments maximize contained square faces for arbitrary sizes, rather than only sizes that are powers of two.

We prove both the general modular intersection bound for a set family and its sharper uniform version, the Frankl-Wilson theorem. Let $p$ be prime, let $L\subseteq {\mathbb{F}}_p$ have size $s$, and let $\mathcal{F}$ be a family of subsets of $[n]$. Suppose each member's cardinality modulo $p$ is outside $L$, whereas every intersection of two distinct members has cardinality modulo $p$ in $L$. Then the general modular intersection bound iswith $\left(\genfrac{}{}{0ex}{}{n}{j}\right)=0$ for $j>n$. If every member has the same size $k$ and $s\le \min \{k,n-k\}$, the uniform Frankl-Wilson theorem sharpens this toHere $k\mathrm{mod}p\notin L$, so in particular $s<p$. Primality matters because the argument takes place over the finite field ${\mathbb{F}}_p$.

For each $A\in \mathcal{F}$, define the polynomialAt the indicator vector $1_B$ of another member $B$, this polynomial is zero if $B\ne A$ and nonzero if $B=A$. Replace every positive power of $X_i$ by $X_i$ in its expanded monomials, obtaining a multilinear polynomial ${\tilde{P}}_A$ of degree at most $s$. This multilinear reduction on the Boolean cube preserves values at every zero-one vector.

The resulting polynomials are linearly independent. Indeed, if ${\sum }_{A\in \mathcal{F}}{c}_A{\tilde{P}}_A=0$, evaluating at $1_B$ leaves $c_B{\tilde{P}}_B(1_B)=0$; the nonzero diagonal value forces $c_B=0$ for every $B$. But multilinear polynomials of degree at most $s$ lie in the vector space with basis $\{{\prod }_{i\in S}{X}_i:|S|\le s\}$, whose dimension is ${\sum }_{j=0}^s\left(\genfrac{}{}{0ex}{}{n}{j}\right)$. The bound on the dimension of a vector space gives the general modular intersection bound for a set family. This is the polynomial method in combinatorics: the intersection conditions produce an evaluation matrix with nonzero diagonal and zero off-diagonal entries.

To prove the sharper uniform bound, we first establish injectivity of inclusion between adjacent set layers. Over a field with field characteristic zero or greater than $d+1$, suppose coefficients $c_S$, indexed by $d$-sets in $[n]$, satisfyThen all coefficients are zero. For $d=0$ this is immediate. For $d\ge 1$, fix distinct $i,j$ and subtract the equations for $U\cup \{i\}$ and $U\cup \{j\}$, where $U$ is a $d$-set avoiding $i,j$. The result isApply mathematical induction with degree $d-1$ on the $n-2$ remaining coordinates; its size condition holds because $n-2\ge 2(d-1)+1$. This forces $c_{S\cup \{i\}}=c_{S\cup \{j\}}$ for every such $S$. Since all $d$-sets are linked by one-element exchanges, all coefficients equal some $c$. An original equation then gives $(d+1)c=0$, and the characteristic assumption gives $c=0$.

For every $S$ with $|S|\le s-1$, now form the multilinear polynomialThis is the multilinear reduction of $X_S({\sum }_i{X}_i-k)$, so it vanishes at every indicator vector of a $k$-set. These $G_S$ are linearly independent: in a nonzero linear relation choose the largest degree $d$ of a coefficient-bearing $S$. The degree-$(d+1)$ terms have coefficients ${\sum }_{S\subset T,|S|=d}{c}_S$. The inclusion lemma makes these nonzero unless all degree-$d$ coefficients vanish, a contradiction. It applies because $d\le s-1$, $p>s$, and $n\ge 2s$.

The $G_S$ are also independent jointly with the ${\tilde{P}}_A$. Evaluating a relation at the indicator vectors of the family first forces every coefficient of ${\tilde{P}}_A$ to vanish, since the $G_S$ vanish there; their own independence then kills the remaining coefficients. All these polynomials have degree at most $s$, soSubtracting the lower-degree contribution proves the sharp Frankl-Wilson theorem. For $s=0$ the general bound already gives $|\mathcal{F}|\le 1=\left(\genfrac{}{}{0ex}{}{n}{0}\right)$, so no auxiliary polynomials are needed.

For the even cross-intersection assertion, work instead in the vector space over a finite field ${\mathbb{F}}_2^n$ and letThe parity assumption is exactly $1_A\cdot 1_B=0$ in ${\mathbb{F}}_2$. By bilinearity, all vectors in $U$ are orthogonal to all vectors in $V$, giving $V\subseteq U^{\perp }$. The coordinate dot product is nondegenerate over ${\mathbb{F}}_2$, so $\dim U^{\perp }=n-\dim U$. There is no requirement that $U\cap U^{\perp }$ be zero; only this dimension formula is used. Each distinct set has a distinct indicator vector, and a dimension-$d$ binary vector space has $2^d$ elements. Therefore the even cross-intersection bound isThis is the even cross-intersection bound, including empty families, for which the product is zero.

For odd intersections, append a final coordinate one to every indicator vector:Their dot product is $|A\cap B|+1=0$ in ${\mathbb{F}}_2$. Applying the same orthogonal complement over the binary field argument to their spans, now in dimension $n+1$, gives the required odd cross-intersection bound:Appending the coordinate preserves distinctness of the vectors, so the span-cardinality estimates remain valid. This proves the odd cross-intersection bound directly by reducing the odd condition to an even one in a larger vector space.