OctoInformBeta

OpenAI preprint, September 2026 · Section 3. Finite reflection-positive probability laws

Lemma 3.3: A cut quadratic form controls a constrained second moment uniformly over small mark sets

A lower bound for marked reflections.

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

Square constraints on finite lists give a weighted quadratic form pairing a configuration on one side of a permutation cut with its reflection in a marked block.

Write [d]={0,…,d−1}[d]=\{0,\ldots,d-1\} for a natural number dd. The height hh and block cutoff TT determine the block count mm and block set B=[m]B=[m]. The list base rr, interaction parameter δ\delta, lower interaction scale κ\kappa and half vertex count MM have the fixed values in (1): h=24,T=80,m=2T+1=161,r=h!2,δ=110h2,κ=910δh−1,M=(h!)m2.\begin{gathered} h=24,\qquad T=80,\qquad m=2T+1=161,\qquad r=\frac{h!}{2},\\ \delta=\frac{1}{10h^2},\qquad \kappa=\frac9{10}\delta^{h-1},\qquad M=\frac{(h!)^m}{2}. \end{gathered}(1)

A permutation word vv on a block set CC at height dd is a function assigning each block an element of Sym⁡([d])\operatorname{Sym}([d]).✦ The tuple vertex set VV and the ordered position pairs PdP_d are the sets in (2): V=Sym⁡([h])B,Pd={(a,c)∈[d]2:a<c}.V=\operatorname{Sym}([h])^B,\qquad P_d=\{(a,c)\in[d]^2:a<c\}.(2)

For v∈Vv\in V, b∈Bb\in B and (a,c)∈Ph(a,c)\in P_h, the position transposition vb;a,cv^{b;a,c} is the permutation word obtained from vv by exchanging positions a,ca,c in block bb.✦ For v∈Vv\in V, b∈Bb\in B and α,β∈[h]\alpha,\beta\in[h], the block transposition reflection Rb;α,βvR_{b;\alpha,\beta}v is the permutation word obtained from vv by exchanging symbols α,β\alpha,\beta in block bb.✦ Their compositions are specified by (3), with every other block unchanged: (vb;a,c)(b)=v(b)∘(ac),(Rb;α,βv)(b)=(αβ)∘v(b).(v^{b;a,c})(b)=v(b)\circ(a\ c),\qquad (R_{b;\alpha,\beta}v)(b)=(\alpha\ \beta)\circ v(b).(3)

For a block b∈Bb\in B and symbols α,β∈[h]\alpha,\beta\in[h], the positive cut predicate on a permutation word v∈Vv\in V is the condition that the inverse permutation in block bb places α\alpha before β\beta.✦ The positive cut half V+⊆VV_+\subseteq V at a block b∈Bb\in B and symbols α,β∈[h]\alpha,\beta\in[h], and its cardinality nn, are given in (4): V+={v∈V:v(b)−1(α)<v(b)−1(β)},n=|V+|.V_+=\{v\in V:v(b)^{-1}(\alpha)<v(b)^{-1}(\beta)\},\qquad n=|V_+|.(4)✦

Lists and Weights

Let S⊆BS\subseteq B be the mark set and put s=|S|s=|S|. An inactive-block address uu for a mark set S⊆BS\subseteq B is a choice of a permutation on each block outside SS.✦ The address set USU_S, addressed list index set JSJ_S and addressed list entry set ESE_S are defined in (5); LsL_s is the list length: US=Sym⁡([h])B\S,Ls=rs,JS=[Ls],ES=US×JS.U_S=\operatorname{Sym}([h])^{B\setminus S},\qquad L_s=r^s,\qquad J_S=[L_s],\qquad E_S=U_S\times J_S.(5)

The inactive-block projection uS:V→USu_S:V\to U_S restricts a word to B\SB\setminus S.✦ For a prime pp, list values Y:ES→𝔽pY:E_S\to\mathbb F_p and indices j:V→JSj:V\to J_S, the sampled tuple configuration zS,Y,j:V→𝔽pz_{S,Y,j}:V\to\mathbb F_p is the configuration in (6): uS(v)=v|B\S,zS,Y,j(v)=Y(uS(v),j(v)).u_S(v)=v|_{B\setminus S},\qquad z_{S,Y,j}(v)=Y\bigl(u_S(v),j(v)\bigr).(6)

The cyclic shift τ\tau on [h][h] is the permutation that sends each position to the next and the last to zero.✦ For a finite block set CC, a group GG, an element c∈Gc\in G and a natural threshold DD, threshold shift adjacency is the relation on GCG^C defined by (7): u∼c,Dv⇔|{a∈C:v(a)=u(a)c}|≥D.u\sim_{c,D}v\quad\Longleftrightarrow\quad \bigl|\{a\in C:v(a)=u(a)c\}\bigr|\ge D.(7)✦

The list constraint edges ℰS⊆ES×ES\mathcal E_S\subseteq E_S\times E_S are the ordered pairs specifying the square-difference constraints between list values. The nonzero square indicator σK:K→ℝ\sigma_K:K\to\mathbb R of a field KK, and the square-constraint indicator ISI_S on 𝔽pES\mathbb F_p^{E_S}, are defined in (8): σK(t)=𝟏{t≠0,t is a square in K},IS(Y)=∏(e,f)∈ℰSσ𝔽p(Y(f)−Y(e)).\sigma_K(t)=\mathbf1_{\{t\ne0,\ t\text{ is a square in }K\}},\qquad I_S(Y)=\prod_{(e,f)\in\mathcal E_S}\sigma_{\mathbb F_p}\bigl(Y(f)-Y(e)\bigr).(8)

The transposition interaction HS:V×V→ℝH_S:V\times V\to\mathbb R sums the contributions of position transpositions in marked blocks, as in (9): HS(v,w)=∑b∈S∑(a,c)∈Phδc−a𝟏{w=vb;a,c}.H_S(v,w)=\sum_{b\in S}\sum_{(a,c)\in P_h} \delta^{c-a}\mathbf1_{\{w=v^{b;a,c}\}}.(9)✦

For a finite index set II, a real matrix AA on II, a label set JJ and labelings j,k:I→Jj,k:I\to J, the label coincidence pairing XA(j,k)X_A(j,k) is the counting sum in (10).✦ For a real matrix AA on a finite index set II and a labeling jj of II, the index energy ℋA(j)\mathcal H_A(j) is half the label coincidence self-pairing of jj with respect to AA, and the index Boltzmann weight wA(j)w_A(j) is the exponential of this energy. The tuple Boltzmann weight WSW_S for a mark set S⊆BS\subseteq B is the index Boltzmann weight associated with the transposition interaction HSH_S: XA(j,k)=∑x∈I∑y∈IAx,y𝟏{j(x)=k(y)},ℋA(j)=12XA(j,j),wA(j)=exp⁡(ℋA(j)),WS(j)=wHS(j).\begin{gathered} X_A(j,k)=\sum_{x\in I}\sum_{y\in I}A_{x,y}\mathbf1_{\{j(x)=k(y)\}},\\ \mathcal H_A(j)=\tfrac12X_A(j,j),\qquad w_A(j)=\exp\bigl(\mathcal H_A(j)\bigr),\qquad W_S(j)=w_{H_S}(j). \end{gathered}(10)

The tuple component average 𝒞p,S\mathcal C_{p,S} of a real observable GG on 𝔽pV\mathbb F_p^V is defined in (11). Every expectation is a uniform probability average over the indicated finite set: 𝒞p,S(G)=𝔼Y∈𝔽pES𝔼j∈JSVIS(Y)WS(j)G(zS,Y,j).\mathcal C_{p,S}(G)= \mathbb E_{Y\in\mathbb F_p^{E_S}}\mathbb E_{j\in J_S^V} I_S(Y)W_S(j)G(z_{S,Y,j}).(11)✦

For a real function F:𝔽pV+→ℝF:\mathbb F_p^{V_+}\to\mathbb R, the reflection quadratic form Qp,Sb;α,β(F)Q_{p,S}^{b;\alpha,\beta}(F) is the component average of the product in (12): Qp,Sb;α,β(F)=𝒞p,S(z↦F((z(v))v∈V+)F((z(Rb;α,βv))v∈V+)).Q_{p,S}^{b;\alpha,\beta}(F)= \mathcal C_{p,S}\!\left( z\longmapsto F\bigl((z(v))_{v\in V_+}\bigr) F\bigl((z(R_{b;\alpha,\beta}v))_{v\in V_+}\bigr)\right).(12)✦

The Claim

Theorem 1. Let pp be prime, let S⊆BS\subseteq B, and let b∈Sb\in S. For distinct α,β∈[h]\alpha,\beta\in[h] and every real function F:𝔽pV+→ℝF:\mathbb F_p^{V_+}\to\mathbb R, the marked reflection lower bound is (13): Qp,Sb;α,β(F)≥(exp⁡κ−1)nLsn𝔼Y∈𝔽pES𝔼j+∈JSV+IS(Y)F((Y(uS(v),j+(v)))v∈V+)2.Q_{p,S}^{b;\alpha,\beta}(F)\ge \frac{(\exp\kappa-1)^n}{L_s^n} \mathbb E_{Y\in\mathbb F_p^{E_S}}\mathbb E_{j_+\in J_S^{V_+}} I_S(Y)F\bigl((Y(u_S(v),j_+(v)))_{v\in V_+}\bigr)^2.(13)✦ The marked reflection constant C*C_* is the real number in (14): C*=(exp⁡κ−1)M(rT+1)M.C_*=\frac{(\exp\kappa-1)^M}{(r^{T+1})^M}.(14)✦ It satisfies (15): 0<C*.0<C_*.(15)✦ For any mark set with s≤T+1s\le T+1, any block b∈Bb\in B and distinct α,β∈[h]\alpha,\beta\in[h], the coefficient comparison is (16): C*≤(exp⁡κ−1)nLsn.C_*\le\frac{(\exp\kappa-1)^n}{L_s^n}.(16)✦

In the paper, the result reads: put κ=910δh−1\kappa=\frac9{10}\delta^{h-1}, λ=(exp⁡κ−1)n\lambda=(\exp\kappa-1)^n and c*=λLT+1−n>0c_*=\lambda L_{T+1}^{-n}>0; for s≤T+1s\le T+1, a cut in a marked block b∈Sb\in S and every real F:𝔽pV+→ℝF:\mathbb F_p^{V_+}\to\mathbb R, the estimate (17) holds, with all list values YY and only the indices j+j_+ on V+V_+ averaged uniformly, and its coefficient is at least c*c_* independently of pp and the list values, where LsL_s denotes the paper’s msm_s and TT its t0t_0: Qp,Sb;α,β(F)≥λLs−n𝔼Y∈𝔽pES𝔼j+∈JSV+IS(Y)F((Y(uS(v),j+(v)))v∈V+)2.Q_{p,S}^{b;\alpha,\beta}(F)\ge \lambda L_s^{-n} \mathbb E_{Y\in\mathbb F_p^{E_S}}\mathbb E_{j_+\in J_S^{V_+}} I_S(Y)F\bigl((Y(u_S(v),j_+(v)))_{v\in V_+}\bigr)^2.(17)

What Is Checked

The statements are over the tuple component average (11) and its reflection quadratic form (12) on prime field configurations, with the fixed permutation and list parameters of (1).