Past exam of the mathematics course of the University of Cambridge / 2017 / iii / Paper 129 (Introduction to Additive Combinatorics)

Use the usual convention that a three-term arithmetic progression must be nonconstant. There is a genuine small-case defect in the printed statement: for $n=1$, $\delta =1$, and $A=\{1\}$, neither conclusion can hold with $c_2>0$. We prove the intended Roth density-increment step for every odd integer $n\ge 3$, including the small values of $n$, rather than silently treating $n=1$ as covered. Counting constant arithmetic progressions would make the first alternative vacuous for every nonempty $A$.

Suppose that $A$ has no nonconstant three-term arithmetic progression, and let $\alpha =|A|/n$ be its actual density of a finite subset. Thus $0<\delta \le \alpha \le 1$. Put $I=[n]$, $N=2n+1$, and embed $I$ in the cyclic group $G=\mathbb{Z}/N\mathbb{Z}$. A congruence $x-2y+z=0(\mathrm{mod}N)$ with $x,y,z\in I$ is an equality over the integers, since $|x-2y+z|\le 2(n-1)<N$. Thus this embedding creates no extra three-term arithmetic progressions. Since $N$ is odd, multiplication by two permutes $G$.

Use normalized Fourier analysis on a finite abelian group, with $e_N(t)=\exp (2\pi it/N)$ and $\hat{h}(r)={\mathbb{E}}_{x\in G}h(x)e_N(-rx)$. For the normalized trilinear arithmetic progression count, character orthogonality givesIndeed, expanding each function in its inverse Fourier transform, averaging in $d$ forces $r_2+2r_3=0$, and averaging in $x$ then forces $r_1=r_3$. The Parseval identity in this normalization is ${\sum }_r|\hat{h}(r){|}^2={\mathbb{E}}_x|h(x){|}^2$.

Let $a=1_A$, $b=\alpha 1_I$, and $f=a-b$, the balanced indicator function of a finite subset. Then $\hat{f}(0)=0$. Only constant arithmetic progressions contribute to $\Lambda (a,a,a)$, whereas summing $2\min (y-1,n-y)+1$ over centers $y\in I$ counts $(n^2+1)/2$ ordered three-term arithmetic progressions in $I$. HenceSet $M={\max }_r|\hat{f}(r)|$. Telescope by changing one factor at a time:In the Fourier transform formula for each term, bound the Fourier coefficient of $f$ by $M$ and apply the Cauchy-Schwarz inequality to the other two factors. The permutation $r\mapsto -2r$ preserves the required squared sums. Since $\Vert a{\Vert }_2^2=\alpha n/N$ and $\Vert b{\Vert }_2^2={\alpha }^{2}n/N$, this givesIf $n\ge 4{\delta }^{-2}$, then ${\alpha }^{2}n\ge 4$, so the difference in the other direction is at least ${\alpha }^3{n}^2/(4N^2)$. Consequently Fourier detection of a progression-free subset of an interval gives a nonzero frequency $r$ with

To turn this into a density increment, take $Q=\lfloor \sqrt{n}\rfloor$. The Dirichlet approximation theorem, proved here by placing $0,r/N,\dots ,Qr/N$ into $Q$ equal intervals modulo one, gives $1\le q\le Q$ such that $\Vert qr/N{\Vert }_{\mathbb{R}/\mathbb{Z}}\le 1/Q$. DefineFor $n\ge N_0(\delta )$, the inequalities $Q\ge \sqrt{n}/2$ and ${\delta }^{2}Q/(288\pi )\ge 2$ show thatUse a progression partition with nearly constant linear phase: every residue chain of $I$ modulo $q$ has at least $\lfloor n/q\rfloor \ge Q$ points. Split each chain into blocks of $L$ consecutive chain points, merging any remainder into its last full block. This partitions $I$ into arithmetic progressions $P$ with $L\le |P|<2L$ and common difference $q$.

On a block $P$ with first point $z_P$, the linear phase $\chi (x)=e_N(-rx)$ satisfiesLet $S_P={\sum }_{x\in P}f(x)=|A\cap P|-\alpha |P|$. Freezing the linear phase on each block introduces error at most $({\delta }^2/72)N^{-1}{\sum }_{x\in I}|f(x)|\le {\delta }^2/72$. ThereforeThe block sums total zero. Their positive part consequently totals at least $N{\delta }^2/144$. Since all block lengths total $n$, some block has $S_P/|P|\ge N{\delta }^2/(144n)\ge {\delta }^2/72$. Its density of a finite subset is at least $\alpha +{\delta }^2/72\ge \delta (1+\delta /72)$.

For the remaining $3\le n<N_0(\delta )$, divide $I$ into disjoint consecutive triples and a remainder of at most two points. A three-term-arithmetic progression-free $A$ uses at most two points of each triple, so $\alpha \le 4/5$; the largest ratio occurs at $n=5$. Any singleton set from $A$ then has density of a finite subset one, which is at least $\delta +{\delta }^2/72$. We may therefore use the explicit constantsIn the large case the constructed arithmetic progression has length at least $c_1\sqrt{n}$; in the small case its length one does too. Thus these constants prove the intended result for all odd integers $n\ge 3$.

Write $d(U,V)=e(U,V)/(|U||V|)$ for the edge density of a bipartite graph between nonempty disjoint vertex sets. A regular pair of vertex sets $(U,V)$ is $\epsilon$-regular if $|d(U^{\prime },V^{\prime })-d(U,V)|\le \epsilon$ whenever $U^{\prime }\subseteq U$, $V^{\prime }\subseteq V$, $|U^{\prime }|\ge \epsilon |U|$, and $|V^{\prime }|\ge \epsilon |V|$.

The Szemerédi regularity lemma says that for every $\epsilon >0$ and positive integer $m_0$ there is $M=M(\epsilon ,m_0)$ such that every finite simple graph on $N\ge m_0$ vertices has a set partitionwhere $V_1,\dots ,V_k$ have equal positive size and at most $\epsilon k^2$ unordered pairs $(V_i,V_j)$, $i<j$, fail to be regular pairs of vertex sets. The version requiring only sufficiently large $N$ is equivalent, since the finitely many smaller admissible sizes can use singleton sets as classes. It suffices to prove this for $0<\epsilon \le 1/2$: reducing a larger parameter to $1/2$ gives stronger conclusions.

Here is a complete energy proof, including restoration of equal class sizes. For a set partition with exceptional set $V_0$, use the following version of equitable regularity energy, omitting pairs incident with $V_0$:For $i=j$, define $d(V_i,V_i)$ using the ordered adjacency indicator function, with diagonal entries zero. Thus the same formula works on every rectangle. The refinement variance identity for regularity energy iswhere $A,B$ run over the refined cells. Expanding the square proves the identity because their weighted mean edge density of a bipartite graph is $d(U,V)$. In particular, refinement never decreases the energy. If a union of refined cells $U^{\prime }\times V^{\prime }$ has discrepancy greater than $\epsilon$, the Cauchy-Schwarz inequality bounds its contribution below by $|U^{\prime }||V^{\prime }|{\epsilon }^2/N^2$. For an irregular equal-class pair of size $L$, witnesses have sizes at least $\epsilon L$, so the increase exceeds ${\epsilon }^4{L}^2/N^2$ in each orientation.

Choose $k_0\ge \max \{m_0,2\}$ so large that $2^{-k_0}\le {\epsilon }^6/100$. Initially partition into $k_0$ equal classes and fewer than $k_0$ exceptional vertices. For sufficiently large $N$ this exceptional set has size at most $\epsilon N/2$. Suppose there are more than $\epsilon k^2$ irregular unordered pairs at a subsequent step, with equal class size $L$. Choose one witnessing pair of subsets for every such pair. Split each class according to membership in all its witness subsets; each class produces at most $2^k$ atoms. The preceding refinement variance identity for regularity energy shows that the energy increase exceedsas long as the exceptional set has size at most $\epsilon N\le N/2$.

Apply equalization with a controlled exceptional set. Set $\ell =\lfloor L/4^k\rfloor$, cut every atom into blocks of size $\ell$, and send its leftover points to $V_0$. Provided $L\ge 2\cdot 4^k$, the new number $k^{\prime }$ of classes satisfiesThe lower bound holds because every old class retains at least one full block: its leftovers total less than $2^k\ell <L$. The discarded fraction is at mostCutting blocks is a further refinement before discarding. Discarding changes only ordered pairs with at least one discarded vertex, so it loses at most $2r$ in the energy. The net increase is therefore at least ${\epsilon }^5/2-2{\epsilon }^6/100\ge {\epsilon }^5/4$.

There can be fewer than $T=\lceil 4{\epsilon }^{-5}\rceil$ such strict energy increases, because the energy stays in $[0,1]$. Through $T$ steps the exceptional fraction is at mostThis also verifies the exceptional-set bound used at every preceding step. Iterating the finite recursion $k\mapsto 2k{4}^k$ at most $T$ times bounds all class counts by some $K=K(\epsilon ,m_0)$. Choosing $N$ sufficiently large in terms of $K$, for instance with $N\ge 4K{4}^K$ and $N\ge 2k_0/\epsilon$, guarantees $L=(N-|V_0|)/k\ge N/(2K)\ge 2\cdot 4^k$ throughout. Thus the process terminates with the desired regular pairs of vertex sets. Choose a finite integer threshold $N_{\ast }$ for these requirements and set $M\ge \max \{K,N_{\ast }\}$. For $m_0\le N<N_{\ast }$, use $N$ singleton sets; every pair is regular because its only admissible nonempty subsets are the whole pair. This completes the Szemerédi regularity lemma for all admissible $N$.

For the half graph, a class of size $s$ has exactly $\lceil s/10\rceil$ elements with fewer than $s/10$ predecessors, and the same number with fewer than $s/10$ successors. Thus its boundary contains at most $2\lceil s/10\rceil$ elements. Let $k_X,k_Y$ be the numbers of classes. The boundary rank obstruction in a half graph gives the union boundbecause $k_X,k_Y\le n/100$. There is therefore an index $m$ away from both boundaries. In its classes $X_i,Y_j$, all four subsets $X^{-}=\{x<m\}$, $X^{+}=\{x>m\}$, $Y^{-}=\{y<m\}$, $Y^{+}=\{y>m\}$ have at least one tenth of their respective class sizes. These subsets are taken within $X_i,Y_j$, which need not be intervals.

In the half graph, $d(X^{-},Y^{+})=1$ and $d(X^{+},Y^{-})=0$. If $(X_i,Y_j)$ were a $1/10$-regular pair of vertex sets, both values would be within $1/10$ of $d(X_i,Y_j)$, implying $1\le 2/10$, a contradiction. At least one class pair is not $1/10$-regular.

Work in positive dimension $n\ge 1$. A cap set has no distinct zero-sum triple. Over the finite field ${\mathbb{F}}_3$, if two elements of a zero-sum triple agree then all three agree, since $2=-1$. Thus a zero-sum triple in a cap set must be diagonal. We prove the Ellenberg–Gijswijt cap-set bound by the polynomial method in combinatorics, keeping the exponential constant explicit and removing the fixed prefactor.

First prove the needed slice rank of a diagonal tensor over any field. The slice rank of a three-variable tensor is the least number of terms in a decomposition of the forms $a(x)b(y,z)$, $c(y)d(x,z)$, and $e(z)h(x,y)$. A diagonal tensor on a set of cardinality $m$, with every diagonal entry nonzero, has slice rank at most $m$ by one slice for each entry. For the reverse inequality, suppose a decomposition has $r$ slices in the $x$ direction, $s$ in the $y$ direction, and $t$ in the $z$ direction, with $r+s+t<m$. LetThe rank-nullity theorem gives $\dim W\ge m-r$. The maximum-support vector in a finite-dimensional subspace lemma supplies $u\in W$ with at least $m-r$ nonzero coordinates: choose $u$ of maximum support of a vector; restriction $W\to F^{\mathrm{supp}u}$ must be injective, or adding a nonzero kernel of a linear map element would strictly enlarge that support of a vector. This argument is valid over finite fields, without any assumption that a generic vector avoids all coordinate hyperplanes.

Contract the proposed tensor decomposition in $x$ against $u$. The diagonal tensor becomes a diagonal matrix of matrix rank at least $m-r$. All $x$ slices vanish. Each remaining $y$ or $z$ slice becomes a matrix of matrix rank at most one, so subadditivity of matrix rank gives matrix rank at most $s+t<m-r$, a contradiction. Therefore the slice rank of a diagonal tensor is exactly $m$.

Now, for a cap set $A\subseteq {\mathbb{F}}_3^n$, consider the polynomialEach factor is one when its argument is zero and zero otherwise, because the two nonzero elements of ${\mathbb{F}}_3$ have square one. Thus $P$ is the indicator function of $x+y+z=0$. Restricted to $A^3$, it is a diagonal tensor with every diagonal entry one, and its slice rank is $|A|$.

Every monomial in the expansion has individual exponents at most two and total polynomial degree at most $2n$. Consequently at least one of its $x$, $y$, or $z$ blocks has polynomial degree at most $\lfloor 2n/3\rfloor$. Assign each monomial to one such block, breaking ties in a fixed way, and collect assigned terms by the monomial in that block. Each collection is a slice. Ifthere are at most $m_n$ slices in each direction. Thus $|A|\le 3m_n$.

The low-degree monomial count for the cap-set bound follows directly from a weighted sum. Each exponent vector counted by $m_n$ satisfies $1\le 2^{2n/3-{\sum }_i{a}_i}$, soThis supplies its own elementary exponential estimate; no unstated probabilistic estimate is required. Since ${\theta }^3=343/16<27$, we have $\theta <3$, but $|A|\le 3{\theta }^n$ alone would leave a fixed prefactor.

Use Cartesian powers of cap sets to remove that prefactor. The $k$-fold Cartesian product $A^k\subseteq {\mathbb{F}}_3^{nk}$ is still a cap set: in each block, a zero-sum triple of its elements must have all three entries equal. Hence $|A{|}^k\le 3{\theta }^{nk}$. Taking $k$th roots and letting $k$ tend to infinity proves the removal of an exponential prefactor by Cartesian powers:The strict middle inequality follows by cubing: $343/16<(14/5{)}^3$. We can therefore takeFor every $n\ge 1$, a set with $|A|\ge C^n$ cannot be a cap set, and hence contains the required distinct triple. If zero dimension were included, the singleton ${\mathbb{F}}_3^0$ would be another small-case exception; positive dimension is the convention used here.

Uniform expectations require nonempty finite sets; assume this throughout the analytic argument. For real-valued functions on $X\times Y$, define the box norm byAll four variables are sampled as independent random variables with replacement, so repeated coordinates are included. The defining fourth power equalsIf it vanishes, every summand vanishes. In particular, the terms $y_0=y_1=y$ give ${\mathbb{E}}_{x}f(x,y{)}^2=0$ for each $y$, so $f=0$. Absolute homogeneity of a norm follows directly: $\Vert af{\Vert }_{\square }=|a|\Vert f{\Vert }_{\square }$.

For the triangle inequality, prove the mixed box Cauchy-Schwarz inequality. WriteSeparating the two $y$ averages and applying the Cauchy-Schwarz inequality in $(x_0,x_1)$ givesExpanding the square and reversing the order of finite sums,by a second Cauchy-Schwarz inequality. Thus $|\Lambda |\le {\prod }_{i,j}\Vert f_{ij}{\Vert }_{\square }$. Expand $\Lambda (f+g,f+g,f+g,f+g)$ into its sixteen multilinear terms and apply this bound to each term. Their upper bounds sum to $(\Vert f{\Vert }_{\square }+\Vert g{\Vert }_{\square }{)}^4$. Taking fourth roots yields the triangle inequality. Together with definiteness of a norm and absolute homogeneity of a norm, this proves the box norm is a norm. Complex functions require conjugates and are outside the printed real-valued convention.

For the bilinear correlation bound for the box norm, use the normalized L2 norm: let $B={\mathbb{E}}_{x,y}f(x,y)u(x)v(y)$ and $\Vert u{\Vert }_2^2={\mathbb{E}}_{x}u(x{)}^2$, $\Vert v{\Vert }_2^2={\mathbb{E}}_{y}v(y{)}^2$. The Cauchy-Schwarz inequality in $y$ givesThe remaining factor is ${\mathbb{E}}_{x,x^{\prime }}u(x)u(x^{\prime }){\mathbb{E}}_{y}f(x,y)f(x^{\prime },y)$. By the Cauchy-Schwarz inequality in $(x,x^{\prime })$, its magnitude is at mostTaking square roots proves

For the tripartite graph, assume its parts are nonempty and write $g_{XY},g_{YZ},g_{XZ}$ for its adjacency indicator functions. Let $f=g_{XY}-\alpha$, so $\Vert f{\Vert }_{\square }\le c$. The normalized triangle count isThe constant-degree hypothesis is exactly ${\mathbb{E}}_y{g}_{YZ}(y,z)=\beta$ for each $z$. Therefore the contribution of the constant $\alpha$ isFor fixed $z$, apply the bilinear correlation bound for the box norm with $u_z(x)=g_{XZ}(x,z)$ and $v_z(y)=g_{YZ}(y,z)$. Since these are indicator functions, $\Vert v_z{\Vert }_2^2=\beta$ and $\Vert u_z{\Vert }_2^2=d_X(z)$, the fraction of $X$ adjacent to $z$. Averaging and applying the Cauchy-Schwarz inequality in $z$ gives the sharper triangle counting with one box-uniform pair and constant opposite degree estimateMultiplying by $|X||Y||Z|$ yieldsThe middle bound also handles $\beta =0$ or $\gamma =0$, when the triangle count is zero.

To connect explicitly with the suggested expansion, put $f_2=g_{YZ}-\beta$, $f_3=g_{XZ}-\gamma$. Both have overall mean zero, and ${\mathbb{E}}_y{f}_2(y,z)=0$ for every $z$. Consequently ${\mathbb{E}}_{x,y,z}{f}_2(y,z)f_3(x,z)=0$; all terms with constant $\alpha$ reduce to $\alpha \beta \gamma$. The terms containing $f$ combine into the single error just estimated. This is where the exact degree assumption is used. If any part is empty the triangle count itself is zero, but the printed uniform expectations and edge density of a bipartite graph values would be undefined; the nonempty-parts convention must therefore be stated rather than dividing by zero.