By OpenAISourceGitHub

Manuscripts from openai/math

Fork
Clone
git clone https://openversions.com/git/index/openai-math.git
Cite
Version link
https://openversions.com/index/openai-math/commit/dfd8c46870b48511e697cbb43a3c8c91d375fa2f
Version identifier (commit hash)
dfd8c46870b48511e697cbb43a3c8c91d375fa2f

Fork to create your own copy with credit to the source. Clone to get a Git copy on your computer.

All papers

Claims

Findings

Bounded-degree coboundary expanders in every dimension

For every integer d≥3d \ge 3, there exist constants D<∞D < \infty, ε>0\varepsilon > 0, and finite connected pure dd-dimensional simplicial complexes XmX_m, with ∣Xm(0)∣⟶∞|X_m(0)| \longrightarrow \infty, such that every vertex belongs to at most DD faces of dimension dd and, for every f∈Ci(Xm;F2)f \in C^i(X_m; \mathbb{F}_2), ∥δif∥Xm,i+1≥ε distXm,i(f,Bi(Xm))\|\delta_i f\|_{X_m, i+1} \ge \varepsilon \, \mathrm{dist}_{X_m, i}(f, B^i(X_m)) (0≤i<d0 \le i < d), with constants independent of mm, ii, and ff.

For every proper J⊊SJ \subsetneq S, UJ={g:gaa=1,gab∈Vab(J) for a≠b}U_J = \{g : g_{aa} = 1, g_{ab} \in V_{ab}(J) \text{ for } a \neq b\} are finite groups satisfying UJ∩UK=UJ∩KU_J \cap U_K = U_{J \cap K} and ⋂i∈AHi=US∖A\bigcap_{i \in A} H_i = U_{S \setminus A} (∅≠A⊆S\varnothing \neq A \subseteq S), and for m≥2m \ge 2, reduction modulo tmt^m is injective on all UJU_J and preserves all these intersections, with ∣Hi∣=M:=q(N2)|H_i| = M := q^{\binom{N}{2}} and ∣U{i}∣=q|U_{\{i\}}| = q.

For q≥2n−1q \ge 2^{n-1}, the infinite coset complex L=CC(U;(Hi)i∈S)L = \mathrm{CC}(U; (H_i)_{i \in S}) satisfies H~j(L;F2)=0\widetilde{H}_j(L; \mathbb{F}_2) = 0 for 0≤j<n−10 \le j < n - 1.

The simple generators xix_i, i∈Si \in S, generate UU.

For 0<j<n−10 < j < n - 1 and m≥2m \ge 2, Hj(U;F2)=0H_j(U; \mathbb{F}_2) = 0 and Hj(Tm;F2)≅Hj(ΓN(m);F2)H_j(T_m; \mathbb{F}_2) \cong H_j(\Gamma_N(m); \mathbb{F}_2); moreover, TmT_m is connected.

Let h≥0h \ge 0 be an integer. If a≥2h+1a \ge 2h + 1, the inclusion on the first aa coordinates induces surjections Hj(Γa(m);F2)⟶Hj(Γa+1(m);F2)H_j(\Gamma_a(m); \mathbb{F}_2) \longrightarrow H_j(\Gamma_{a+1}(m); \mathbb{F}_2) for 0≤j≤h0 \le j \le h and m≥1m \ge 1.

For 1≤j≤d−11 \le j \le d - 1, conjugation by SLN(R)\mathrm{SL}_N(R) acts trivially on Hj(ΓN(m);F2)H_j(\Gamma_N(m); \mathbb{F}_2), for every m≥1m \ge 1.

For every m≥2m \ge 2, Hj(ΓN(m);F2)=Hj(Tm;F2)=Hj(Tm;F2)=0H_j(\Gamma_N(m); \mathbb{F}_2) = H_j(T_m; \mathbb{F}_2) = H^j(T_m; \mathbb{F}_2) = 0 for 1≤j≤d−11 \le j \le d - 1.

Methods

Congruence coset complex construction: defining U≤SLN(Fq[t])U \le \mathrm{SL}_N(\mathbb{F}_q[t]) as the inverse image of the upper unitriangular subgroup at t=0t = 0, congruence subgroups ΓN(m)\Gamma_N(m), finite quotients Gm=U/ΓN(m)G_m = U/\Gamma_N(m), coset complexes Tm=CC(Gm;(H‾i)i∈S)T_m = \mathrm{CC}(G_m; (\overline{H}_i)_{i \in S}), and dd-skeleta Xm=skeldTmX_m = \mathrm{skel}_d T_m.

Source excerpts

“Let U ≤ SLN(R) be the inverse image, under evaluation at t = 0, of the upper unitriangular subgroup of SLN(k).” · Show in paper

“Gm = U/ΓN(m), Tm = CC(Gm; (Hi)i∈S), Xm = skeld Tm.” · Show in paper

Twin-building codistance filtration: realizing the infinite coset complex L=CC(U;(Hi)i∈S)L = \mathrm{CC}(U; (H_i)_{i \in S}) as the codistance-1 opposition subcomplex in the affine twin building of SLN(k[t,t−1])\mathrm{SL}_N(k[t, t^{-1}]) and filtering the negative building by codistance length.

Source excerpts

“To obtain acyclicity, we now place the coset complex in the classical affine twin building of SLN(k[t, t^−1]).” · Show in paper

“Starting with L, add all chambers of the negative building in increasing order of the length of their codistance.” · Show in paper

Pro-stability to finite-level surjectivity via odd-index transfer: combining Morrow's pro Tor-unitality theorem and Iwasa's homology pro-stability theorem with odd-index transfer maps to obtain surjections Hj(Γa(m);F2)→Hj(Γa+1(m);F2)H_j(\Gamma_a(m); \mathbb{F}_2) \to H_j(\Gamma_{a+1}(m); \mathbb{F}_2) at every finite level m≥1m \ge 1.

Source excerpts

“The available stability theorem concerns inverse systems; odd-index transfer will supply the additional step from that theorem to an ordinary surjection.” · Show in paper

Skeleton weight comparison: bounding the ratio of intrinsic skeleton weights between Ym=skeln−1TmY_m = \mathrm{skel}_{n-1} T_m and Xm=skeldTmX_m = \mathrm{skel}_d T_m using chamber-face incidence counts.

Source excerpts

“The expansion theorem uses the intrinsic weights of Ym, while Theorem 1.1 uses those of Xm. The following elementary observation keeps track of this change of normalization.” · Show in paper