OctoInformBeta

OpenAI preprint, September 2026 · §7. The singular contribution · §7

Lemma 7.3 [Coercivity]: Q≥(P2+T2+δ2)/342+(c−|δ|)(d+3δ)/12Q \ge (P^2+T^2+\delta^2)/342 + (c-|\delta|)(d+3\delta)/12 under linear constraints, for real parameters and for feasible tail tuples in dimension DD

The tail quadratic part is coercive on feasible tuples.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

Sidorenko’s conjecture asserts that t(H,G)≥p(G)|E(H)|t(H,G) \ge p(G)^{|E(H)|} for every bipartite graph HH and every graph GG, where t(H,G)t(H,G) is the homomorphism density of HH in GG and p(G)p(G) is the edge density of GG. The counterexample to this conjecture takes for HH the bipartite graph joining each of 1313 points to the 2222 faces containing it, a face being a triple of points. Section 7, on the singular contribution, works with a quadratic polynomial QQ in real parameters, the quadratic part of the tail exponent of that contribution, defined below.

The parameters range over a region cut out by linear inequalities, and the question is how small QQ can be on that region. The library answers it twice: for real parameters, and for tail tuples, vectors of natural numbers bounded in terms of a dimension DD from which the parameters are computed.

The Quadratic Part and Tail Tuples

For real triples p=(p0,p1,p2)p = (p_0,p_1,p_2) and t=(t0,t1,t2)t = (t_0,t_1,t_2), write P=p0+p1+p2P = p_0+p_1+p_2 and T=t0+t1+t2T = t_0+t_1+t_2. The quadratic part of the tail exponent is the real function of real numbers cc, dd, δ\delta, vv and real triples pp, tt given by Q(c,d,δ,v,p,t)=c(d+δ)−δd+2δ24+16∑i=02(2pi−ti)2+δP−c(P+T)6+(d−3δ2−P)v+v22.\begin{split} Q(c,d,\delta,v,p,t) = {}& \frac{c(d+\delta)-\delta d+2\delta^2}{4} + \frac16\sum_{i=0}^{2}(2p_i-t_i)^2 + \delta P - \frac{c(P+T)}{6} \\ & + \Bigl(\frac{d-3\delta}{2}-P\Bigr)v + \frac{v^2}{2}. \end{split}(1)✦

A tail tuple a=(c,u,v,p,t)a = (c,u,v,p,t) consists of three natural numbers cc, uu, vv and two triples p=(p0,p1,p2)p = (p_0,p_1,p_2) and t=(t0,t1,t2)t = (t_0,t_1,t_2) of natural numbers.✦ The tail sizes of a tail tuple a=(c,u,v,p,t)a = (c,u,v,p,t) are the numbers si=pi+tis_i = p_i+t_i for i∈{0,1,2}i \in \{0,1,2\}, its tail sums PP and TT are the sums of pp and tt, and its tail offset is the real number δ=2u−c\delta = 2u-c✦.

Let DD be a natural number. A tail tuple aa satisfies the size constraints in dimension DD if c≤Dc \le D and s0+s1+s2≤D−cs_0+s_1+s_2 \le D-c; it is feasible in dimension DD if it satisfies the size constraints and, moreover, u≤cu \le c, ti≤pi+δt_i \le p_i+\delta and v≤pi+δv \le p_i+\delta for every ii, and 12(D−c−3δ)−P+v≥0\tfrac12(D-c-3\delta)-P+v \ge 0.✦

The quadratic part of a tail tuple aa in dimension DD is QD(a)=Q(c,D−c,δ,v,p,t)Q_D(a) = Q(c, D-c, \delta, v, p, t)✦, and its tail energy in dimension DD is the real number ED(a)=P2+T2+δ2342+(c−|δ|)(D−c+3δ)12.E_D(a) = \frac{P^2+T^2+\delta^2}{342} + \frac{(c-|\delta|)(D-c+3\delta)}{12}.(2)✦

The Claim

Theorem 1 (Coercivity). Let cc, dd, δ\delta, vv be real numbers and pp, tt real triples with sums PP and TT. Suppose that |δ|≤c|\delta| \le c and P+T≤dP+T \le d, that vv and every pip_i and tit_i are nonnegative, that ti≤pi+δt_i \le p_i+\delta and v≤pi+δv \le p_i+\delta for every ii, and that w=12(d−3δ)−P+vw = \tfrac12(d-3\delta)-P+v is nonnegative. Then Q(c,d,δ,v,p,t)≥P2+T2+δ2342+(c−|δ|)(d+3δ)12.Q(c,d,\delta,v,p,t) \ge \frac{P^2+T^2+\delta^2}{342} + \frac{(c-|\delta|)(d+3\delta)}{12}.(3)✦ Moreover, for every natural number DD, every tail tuple aa feasible in dimension DD satisfies QD(a)≥ED(a).Q_D(a) \ge E_D(a).(4)✦

In the paper, the result reads: every tuple satisfying the face constraints c≥|δ|c \ge |\delta|, P+T≤dP+T \le d, pi,ti,v≥0p_i, t_i, v \ge 0, ti,v≤pi+δt_i, v \le p_i+\delta and w≥0w \ge 0 satisfies Q≥(P2+T2+δ2)/342Q \ge (P^2+T^2+\delta^2)/342.

What Is Checked

The claim is stated over the quadratic part QQ of (1) at real parameters satisfying the hypotheses of the theorem, and over tail tuples feasible in dimension DD, with their tail sizes, tail sums and tail offset, their quadratic part QDQ_D and their tail energy EDE_D of (2).