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.

Latest version

Checks complete

Sync OpenAI Math manuscripts from 4eaee87548250b2682affda7d05d947c05554bae

Committed

All papers

Claims

Findings

The periodic spin-one Haldane gap

For the pure antiferromagnetic spin-one Heisenberg chain on even periodic rings HL=∑j=0L−1∑α=x,y,zSjαSj+1αH_L = \sum_{j=0}^{L-1} \sum_{\alpha=x,y,z} S_j^\alpha S_{j+1}^\alpha (SLα=S0αS_L^\alpha = S_0^\alpha), the ground state is unique for every even L≥60L \ge 60, γL>4105log⁡8079\gamma_L > \frac{4}{105}\log\frac{80}{79} for every even L≥60L \ge 60, γL>log⁡20784\gamma_L > \frac{\log 20}{784} for every even L≥2304L \ge 2304, and Δ1=lim inf⁡L→∞,L evenγL≥log⁡20784>0\Delta_1 = \liminf_{L\to\infty, L\text{ even}} \gamma_L \ge \frac{\log 20}{784} > 0; these spectral conclusions also hold for any simultaneously similar spin-one realization satisfying the same commutator and Casimir normalizations.

For each b>0b > 0, there exist a Hilbert space Kb\mathcal{K}_b, a compact self-adjoint Hilbert-Schmidt operator XbX_b, and a unitary representation g↦Ugg \mapsto U_g of the signed permutation group G⊂SO(3)\mathcal{G} \subset SO(3) commuting with XbX_b, such that Zn(b,g)=Tr⁡(XbnUg)Z_n(b, g) = \operatorname{Tr}(X_b^n U_g) for all n≥4n \ge 4 and g∈Gg \in \mathcal{G}, and Zn(b)=∑iλi(b)nZ_n(b) = \sum_i \lambda_i(b)^n with ∑i∣λi(b)∣k<∞\sum_i |\lambda_i(b)|^k < \infty for k≥2k \ge 2.

If a non-scalar finite-dimensional self-adjoint Hamiltonian family HLH_L (L≥4L \ge 4 even) has partition function ZL(b)=Tr⁡e−b(HL+aLI)=∑i∣λi(b)∣LZ_L(b) = \operatorname{Tr} e^{-b(H_L + aLI)} = \sum_i |\lambda_i(b)|^L and satisfies initial purity bounds S(n0,b0)≥1−u0S(n_0, b_0) \ge 1 - u_0 and T(2n0,b0)≥1−u0T(2n_0, b_0) \ge 1 - u_0 under scalar conditions 0<u0<1/60 < u_0 < 1/6, G(u0)≤CG(u_0) \le C, r=Cu0<1r = Cu_0 < 1, and C(1−3u0)>3C(1 - 3u_0) > 3, then for every even L≥n0L \ge n_0 the physical ground state is unique and γL>log⁡(1/r)b0\gamma_L > \frac{\log(1/r)}{b_0}.

At inverse temperature b=105/4b = 105/4, the spatial and physical purities satisfy S(60,b)>7/8S(60, b) > 7/8 and T(120,b)>7/8T(120, b) > 7/8.

At inverse temperature b=784b = 784, the spatial and physical purity defects satisfy 1−S(2304,b)≤84513107<11001 - S(2304, b) \le \frac{84513}{10^7} < \frac{1}{100} and 1−T(4608,b)≤274935⋅106<11001 - T(4608, b) \le \frac{27493}{5 \cdot 10^6} < \frac{1}{100}.

For the self-adjoint Hamiltonians (1.1), the normalized ground states of the even periodic rings have a subsequence converging on every local operator to a state ω\omega, and every such limit satisfies ω(A∗[H,A])≥log⁡20784(ω(A∗A)−∣ω(A)∣2)\omega(A^*[H, A]) \ge \frac{\log 20}{784}(\omega(A^*A) - |\omega(A)|^2) for every local operator AA.

For a=700741/500000a = 700741/500000, the centers in the tables of Lemma 5.1 approximate the shifted partition functions Zn(b,g)Z_n(b, g) (in units 10−610^{-6}) with absolute errors less than 10−510^{-5} at b=21/2b = 21/2 (n=6,8,10n = 6, 8, 10; g=1,P,Cg = 1, P, C) and less than 30⋅10−630 \cdot 10^{-6} at b=49/4b = 49/4 (4≤n≤124 \le n \le 12; g=1,Pg = 1, P).

For the untwisted periodic spin-one Heisenberg Hamiltonian and a=700741/500000a = 700741/500000, E0(72)<−72aE_0(72) < -72a and E0(120)<−120aE_0(120) < -120a, and consequently Z72(b)>1Z_{72}(b) > 1 and Z120(b)>1Z_{120}(b) > 1 for every b>0b > 0.

Methods

Ordered bond expansion and construction of a self-adjoint Hilbert-Schmidt spatial transfer operator XbX_b on bond histories Ωb\Omega_b whose moments reproduce twisted and untwisted thermal partition functions Zn(b,g)=Tr⁡(XbnUg)Z_n(b, g) = \operatorname{Tr}(X_b^n U_g).

Source excerpts

“For two histories x, y, merge their labeled times in increasing order and define k_b(x, y) as the trace of the resulting product of the matrices G_alpha, with increasing time from left to right.” · Show in paper

Coupled spatial-physical purity bootstrap that doubles length and inverse temperature simultaneously using cancellation identities and interpolates across intermediate even lengths via spatial moment inequalities.

Source excerpts

“S(n, 2b) = S(n, b)^2 T(2n, b) / T(n, b)^2, T(4n, b) = T(2n, b)^2 S(2n, 2b) / S(2n, b)^2.” · Show in paper

Rigorous matrix-exponential trace enclosure using conditioned Poisson polynomials in a contraction, symmetry-reduced representative columns, and overflow-free packed or 64-bit signed integer arithmetic.

Source excerpts

“Put Q = 2^56 and q_j = floor(Q p_j). For a representative column r, start at zero and apply, for j = 256, ..., 0, z <- fl(Bz/16) + q_j e_r.” · Show in paper

“For a selected unit basis column e_r, start with X = 0 and perform, in the order j = J, J-1, ..., 0, X <- floor(DX/32) + c_j e_r, c_j = floor(Q p_j).” · Show in paper

Variational ground-state energy bounding using periodic matrix-product state (MPS) trial vectors defined by explicit 26×2626 \times 26 integer matrices and block-decomposed integer contractions.

Source excerpts

“Only arbitrary-precision integer operations are used. For N = 72, the required power is 70. For N = 120, one may use T^118 = T^64 T^32 T^16 T^4 T^2 and T^120 = T^64 T^32 T^16 T^8 within each block.” · Show in paper