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

Foulkes' conjecture for the sixth symmetric power

For every integer b≥6b \geq 6 and every finite-dimensional complex vector space VV, there is a GL(V)\mathrm{GL}(V)-equivariant injection Sym6(SymbV)↪Symb(Sym6V)\mathrm{Sym}^6(\mathrm{Sym}^b V) \hookrightarrow \mathrm{Sym}^b(\mathrm{Sym}^6 V), or equivalently hb[h6]−h6[hb]h_b[h_6] - h_6[h_b] is Schur-positive for every b≥6b \geq 6.

For every finite-dimensional complex VV and every b≥30b \geq 30, there is a GL(V)\mathrm{GL}(V)-equivariant injection Sym6(SymbV)↪Symb(Sym6V)\mathrm{Sym}^6(\mathrm{Sym}^b V) \hookrightarrow \mathrm{Sym}^b(\mathrm{Sym}^6 V).

For a≥2a \geq 2, one has Rb=AbR_b = A_b whenever b≥a(a−1)2b \geq a(a-1)^2.

For every integer b≥1b \geq 1 and partition λ\lambda of 6b6b with ∑i≥2λi≤b\sum_{i \geq 2} \lambda_i \leq b, one has [sλ]h6[hb]≤[sλ]hb[h6][s_\lambda]h_6[h_b] \leq [s_\lambda]h_b[h_6].

The function UU is nondecreasing in every gap, and for every nonnegative gap vector dd with 6∣w(d)6 \mid w(d), one has F(d)≤U(d)F(d) \leq U(d).

Suppose Q(x)≥C>0Q(x) \geq C > 0, where CC is an integer and w(x)=6d0w(x) = 6d_0 with d0>0d_0 > 0. Put z=min⁡(18,min⁡{h≥0:(d0+hh)≥C})z = \min\left(18, \min\left\{h \geq 0 : \binom{d_0+h}{h} \geq C\right\}\right). Then for all k≥1k \geq 1, Q(kx)≥f(C,z,k)Q(kx) \geq f(C, z, k).

The complete program in Appendix A verifies the base comparisons in Table 1 (145435145435 exact fallback comparisons, 92729272 equalities, and no negative difference for 6≤b≤256 \leq b \leq 25), has a mode list of 8735487354 points (114314114314 after downward closure), and flags 49696984969698 pairs in 26≤b≤14926 \leq b \leq 149, all with t2(d)≤27t_2(d) \leq 27.

The complete program in Appendix B checks 5706566857065668 coefficients in the small-tail band 26≤b≤14926 \leq b \leq 149, t2≤27t_2 \leq 27; of these, 177134177134 are zero, none are negative, and every integer in (20) is divisible by 720720.

Methods

Realize the canonical Foulkes–Howe map μa,b,V:Symb(SymaV)→Syma(SymbV)\mu_{a,b,V} : \mathrm{Sym}^b(\mathrm{Sym}^a V) \to \mathrm{Sym}^a(\mathrm{Sym}^b V) as multiplication in the Sa\mathfrak{S}_a-invariant subring R=⨁j≥0((Sym V)⊗a)(j,…,j)SaR = \bigoplus_{j \geq 0} ((\mathrm{Sym}\, V)^{\otimes a})_{(j,\dots,j)}^{\mathfrak{S}_a} generated by R1=SymaVR_1 = \mathrm{Sym}^a V.

Source excerpts

“Let A ⊂ R be the graded subalgebra generated by R1 = Syma V . For v ∈ V , put z(v) = v⊗a ∈ R1. Multiplication defines the canonical Foulkes–Howe map μa,b,V : Symb (Syma V ) −→ Rb = Syma (Symb V ), (1) whose image is Ab.” · Show in paper

Use gap coordinates di=λi−λi+1d_i = \lambda_i - \lambda_{i+1} and rectangular highest-weight shifts d+6eid + 6e_i to propagate target multiplicity lower bounds (Q(d+c)≥Q(d)Q(d+c) \geq Q(d) when Q(c)>0Q(c) > 0) and reduce d6≥6d_6 \geq 6 in 26≤b≤14926 \leq b \leq 149 to earlier degrees.

Source excerpts

“Let c, d ∈ Z 6 ≥0 satisfy 6 | w(c) and 6 | w(d). If Q(c) > 0, then Q(d + c) ≥ Q(d). In particular Q(d + 6ei) ≥ Q(d) (1 ≤ i ≤ 6), F(d + 6e6) = F(d).” · Show in paper

Invert z(v)=∏ixi0z(v) = \prod_i x_{i0} to express row-permutation-invariant polynomials in Yi=(xi1/xi0,… )Y_i = (x_{i1}/x_{i0}, \dots) of total degree at most tt as linear combinations of products of at most tt elements of z−1R1z^{-1}R_1, clearing denominators uniformly.

Source excerpts

“For every integer t ≥ 0, every row-permutation-invariant polynomial in the lists Yi of total degree at most t is a linear combination of products of at most t elements of z −1R1.” · Show in paper

Use sl2\mathfrak{sl}_2 polarization operators Di=y⋅∂xiD_i = y \cdot \partial_{x_i} to prove Rj=∑v∈Vz(v)LRj−LR_j = \sum_{v \in V} z(v)^L R_{j-L} for j≥aLj \geq aL and finite generation of RR over AA in degrees below aa.

Source excerpts

“For integers L ≥ 1 and j ≥ aL, Wj = ∑ v∈V z(v) LWj−L, Rj = ∑ v∈V z(v) LRj−L.” · Show in paper

Bound source multiplicities F(d)F(d) from above by a gap-monotone function U(d)U(d) via the cycle formula for h6[hb]h_6[h_b] and the signed strip rule (plethystic Murnaghan–Nakayama determinant), and compute exact target and source tables via Newton's identity.

Source excerpts

“We bound a coefficient of each product by removing its factors one at a time. The signed strip rule restricts the intermediate partitions; the bound below counts their possible coordinates while discarding their signs and some of their constraints.” · Show in paper

“The target tables and the remaining source comparisons use Newton’s identity in place of a bound.” · Show in paper

Amplify known target multiplicities Q(x)≥CQ(x) \geq C to Q(kx)≥f(C,z,k)Q(kx) \geq f(C, z, k) by row-reducing highest-weight polynomials to obtain distinct leading exponents and applying a triangulation-based sumset lower bound.

Source excerpts

“Row reduction with respect to a fixed monomial order gives a basis with distinct leading monomials. Their exponent vectors form a set D of size C, all with nonnegative integer entries summing to d0.” · Show in paper

Compute exact differences 720(Q(λ)−F(λ))720(Q(\lambda) - F(\lambda)) across the small-tail band t2≤27t_2 \leq 27 for 26≤b≤14926 \leq b \leq 149 by constructing seven degree-independent Laurent series numerators Gj(q,y)G_j(q,y) truncated at tail degree 2727.

Source excerpts

“For each fixed tail η, we will construct seven Laurent series independent of b. Their coefficients at B, B − b, . . . , B − 6b will sum to 720(Q(λ) − F(λ)).” · Show in paper