OctoInformBeta

OpenAI preprint, September 2026 · §4. Transverse configurations and determinant signs · §4

Lemma 4.3 [Concentration of restriction signs]: For uniform QQ and distinct rr-dimensional subspaces (Ui)i∈I(U_i)_{i\in I}, the share fξf_\xi with Q|UiQ|_{U_i} nondegenerate of sign ξ\xi has 𝔼|fξ−1/2|≤r/q+1/|I|\mathbb E|f_\xi-1/2|\le\sqrt{r/q+1/|I|}

Variance bounds for restriction signs of a random symmetric form.

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

Let 𝔽q\mathbb F_q be a finite field of odd characteristic with qq elements, QQ a symmetric bilinear form on a finite-dimensional vector space VV over 𝔽q\mathbb F_q, and UU a subspace of VV of dimension rr. The restriction of QQ to UU is degenerate or nondegenerate, and a nondegenerate form has a sign ±1\pm1. For each QQ, a family of rr-dimensional subspaces then has a share of members on which QQ restricts to a nondegenerate form of sign +1+1, and a share on which it restricts to one of sign −1-1. The question is how close each share is to 1/21/2 when QQ is drawn uniformly at random.

The result stated below is a step in the part of the counterexample to Sidorenko’s conjecture that treats transverse configurations and determinant signs.

Signs of Restricted Forms

The uniform mean 𝔼a∈Ag(a)\mathbb E_{a\in A}\,g(a) of a real function gg on a finite set AA is its average over AA: 𝔼a∈Ag(a)=1|A|∑a∈Ag(a).\mathbb E_{a\in A}\,g(a)=\frac1{|A|}\sum_{a\in A}g(a).(1)✦

The set Sym(V)\mathrm{Sym}(V) of symmetric bilinear forms on a vector space VV over 𝔽q\mathbb F_q consists of the bilinear maps B:V×V→𝔽qB\colon V\times V\to\mathbb F_q with B(x,y)=B(y,x)B(x,y)=B(y,x) for all x,y∈Vx,y\in V.✦ For a natural number DD, the space SDS_D of symmetric matrices of size DD is the subspace of the space of D×DD\times D matrices over 𝔽q\mathbb F_q consisting of the matrices MM with MT=MM^{T}=M. A matrix M∈SDM\in S_D defines the symmetric bilinear form BMB_M on 𝔽qD\mathbb F_q^D given by BM(x,y)=xTMy(x,y∈𝔽qD).B_M(x,y)=x^{T}My\qquad(x,y\in\mathbb F_q^D).(2)✦

For vector spaces VV and WW over 𝔽q\mathbb F_q, the pullback φ*B∈Sym(V)\varphi^*B\in\mathrm{Sym}(V) of a form B∈Sym(W)B\in\mathrm{Sym}(W) along a linear map φ:V→W\varphi\colon V\to W is the form (φ*B)(x,y)=B(φx,φy)(\varphi^*B)(x,y)=B(\varphi x,\varphi y). The restriction B|UB|_U of B∈Sym(V)B\in\mathrm{Sym}(V) to a subspace UU of VV is its pullback along the inclusion of UU in VV.

The radical of B∈Sym(V)B\in\mathrm{Sym}(V) is the subspace R={x∈V:B(x,y)=0 for all y∈V}R=\{x\in V: B(x,y)=0\text{ for all }y\in V\}, and BB is nondegenerate when R=0R=0. The radical quotient form B‾\bar B of BB is the nondegenerate form on V/RV/R given by B‾(x+R,y+R)=B(x,y)\bar B(x+R,y+R)=B(x,y).✦ The sign sgn⁡(B)\operatorname{sgn}(B) of BB is the quadratic character χ\chi of the determinant of the Gram matrix GG of B‾\bar B in a basis of V/RV/R, sgn⁡(B)=χ(det⁡G),\operatorname{sgn}(B)=\chi(\det G),(3)✦ with sgn⁡(B)=1\operatorname{sgn}(B)=1 when BB has rank zero. For ξ∈{1,−1}\xi\in\{1,-1\}, the sign indicator 𝟏ξ(B)\mathbf 1_\xi(B) of BB is 11 when BB is nondegenerate and sgn⁡(B)=ξ\operatorname{sgn}(B)=\xi, and 00 otherwise.✦

The share fξ(Q)f_\xi(Q) of a family (Ui)i∈I(U_i)_{i\in I} of subspaces of VV, indexed by a nonempty finite set II, for a form Q∈Sym(V)Q\in\mathrm{Sym}(V) and a sign ξ∈{1,−1}\xi\in\{1,-1\}, is the fraction of indices ii for which Q|UiQ|_{U_i} is nondegenerate of sign ξ\xi: fξ(Q)=𝔼i∈I𝟏ξ(Q|Ui).f_\xi(Q)=\mathbb E_{i\in I}\,\mathbf 1_\xi\bigl(Q|_{U_i}\bigr).(4)

For a vector space VV over 𝔽q\mathbb F_q and a natural number kk, Grk(V)\mathrm{Gr}_k(V) is the set of subspaces of VV of dimension kk.✦ On 𝔽qD\mathbb F_q^D, two further shares of a matrix Q∈SDQ\in S_D run over subspaces of a fixed dimension rr. For Q∈SDQ\in S_D and ξ∈{1,−1}\xi\in\{1,-1\}, the share fξGr(Q)f^{\mathrm{Gr}}_\xi(Q) over all rr-dimensional subspaces of 𝔽qD\mathbb F_q^D is the average of the sign indicator of BQ|UB_Q|_U over U∈Grr(𝔽qD)U\in\mathrm{Gr}_r(\mathbb F_q^D): fξGr(Q)=𝔼U∈Grr(𝔽qD)𝟏ξ(BQ|U).f^{\mathrm{Gr}}_\xi(Q)=\mathbb E_{U\in\mathrm{Gr}_r(\mathbb F_q^D)}\,\mathbf 1_\xi\bigl(B_Q|_U\bigr).(5) For a bilinear form LL on 𝔽qD\mathbb F_q^D, let ℐr(L)\mathcal I_r(L) be the set of U∈Grr(𝔽qD)U\in\mathrm{Gr}_r(\mathbb F_q^D) with L(x,y)=0L(x,y)=0 for all x,y∈Ux,y\in U, the totally isotropic ones. For Q∈SDQ\in S_D and ξ∈{1,−1}\xi\in\{1,-1\}, the share fξL(Q)f^{L}_\xi(Q) over the totally isotropic rr-dimensional subspaces of LL is the average of the sign indicator of BQ|UB_Q|_U over U∈ℐr(L)U\in\mathcal I_r(L): fξL(Q)=𝔼U∈ℐr(L)𝟏ξ(BQ|U).f^{L}_\xi(Q)=\mathbb E_{U\in\mathcal I_r(L)}\,\mathbf 1_\xi\bigl(B_Q|_U\bigr).(6)✦

The Claim

Theorem 1. Let 𝔽q\mathbb F_q be a finite field of odd characteristic with qq elements, rr a natural number and ξ∈{1,−1}\xi\in\{1,-1\}. Let (Ui)i∈I(U_i)_{i\in I} be a family of pairwise distinct rr-dimensional subspaces of a finite-dimensional vector space VV over 𝔽q\mathbb F_q, indexed by a nonempty finite set II. Then the share fξf_\xi of (4) satisfies 𝔼Q∈Sym(V)(fξ(Q)−12)2≤rq+1|I|\mathbb E_{Q\in\mathrm{Sym}(V)}\bigl(f_\xi(Q)-\tfrac12\bigr)^2\le\frac rq+\frac1{|I|}(7)✦ and (𝔼Q∈Sym(V)|fξ(Q)−12|)2≤rq+1|I|.\Bigl(\mathbb E_{Q\in\mathrm{Sym}(V)}\bigl|f_\xi(Q)-\tfrac12\bigr|\Bigr)^2\le\frac rq+\frac1{|I|}.(8)✦ If moreover V=𝔽qDV=\mathbb F_q^D and the family has |I|≥q|I|\ge q members, then 𝔼Q∈SD|fξ(BQ)−12|≤r+1q.\mathbb E_{Q\in S_D}\bigl|f_\xi(B_Q)-\tfrac12\bigr|\le\sqrt{\frac{r+1}q}.(9)✦ For r≥1r\ge1, the share over all rr-dimensional subspaces of 𝔽q2r\mathbb F_q^{2r} satisfies 𝔼Q∈S2r|fξGr(Q)−12|≤r+1q.\mathbb E_{Q\in S_{2r}}\bigl|f^{\mathrm{Gr}}_\xi(Q)-\tfrac12\bigr|\le\sqrt{\frac{r+1}q}.(10)✦ For r≥2r\ge2 and a nondegenerate symmetric bilinear form LL on 𝔽q2r\mathbb F_q^{2r} of sign sgn⁡(L)=c0=χ((−1)r)\operatorname{sgn}(L)=c_0=\chi\bigl((-1)^r\bigr), the share over the totally isotropic rr-dimensional subspaces of LL satisfies 𝔼Q∈S2r|fξL(Q)−12|≤r+1q.\mathbb E_{Q\in S_{2r}}\bigl|f^{L}_\xi(Q)-\tfrac12\bigr|\le\sqrt{\frac{r+1}q}.(11)✦

In the paper, the result reads: for any nonempty family (Ui)i∈I(U_i)_{i\in I} of distinct rr-dimensional subspaces of 𝔽qD\mathbb F_q^D with r≥1r\ge1 and a uniform symmetric form QQ, the share fξ(Q)f_\xi(Q) of the UiU_i on which QQ is nondegenerate of sign ξ\xi satisfies 𝔼|fξ(Q)−12|=O(q−1/2+|I|−1/2)\mathbb E\,|f_\xi(Q)-\tfrac12|=O\bigl(q^{-1/2}+|I|^{-1/2}\bigr) uniformly, and the same holds conditionally on QQ being nondegenerate, or on QQ and Q−LQ-L being nondegenerate, uniformly in the form LL.

What Is Checked

The bounds (7) to (11) are stated over the uniform mean (1) on Sym(V)\mathrm{Sym}(V) and on SDS_D, the sign (3) taken through the radical quotient form, the sign indicator 𝟏ξ\mathbf 1_\xi, restriction as pullback along an inclusion, the form BQB_Q of a symmetric matrix (2), and the shares (4), (5) and (6) over subspaces of dimension rr.

The library also proves that M↦BMM\mapsto B_M is a bijection from SDS_D onto Sym(𝔽qD)\mathrm{Sym}(\mathbb F_q^D)✦, and that pullback along a linear map φ:V→W\varphi\colon V\to W is a linear map from Sym(W)\mathrm{Sym}(W) to Sym(V)\mathrm{Sym}(V)✦.