OctoInformBeta

OpenAI preprint, September 2026 · §4. The rank-one comparison · §4

Lemma 4.3: Entrywise formula for a>0a > 0, plus Gaussian unitary averages for v,h≥0v, h \ge 0 and a Gamma mixture for σ<1\sigma < 1

Positive-parameter tangent Schur multipliers are trace self-adjoint and unital.

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

In the rank-one comparison for one-dimensional Lieb-Thirring inequalities, with spectral exponent γ∈(1/2,3/2)\gamma \in (1/2, 3/2) and σ=γ−1/2∈(0,1)\sigma = \gamma - 1/2 \in (0, 1), one works in an eigenbasis of an N×NN \times N real symmetric matrix B∈Sym⁡NB \in \operatorname{Sym} N with real eigenvalues z1,…,zNz_1, \dots, z_N. The comparison uses Schur multipliers on complex matrices built from rational-power kernels in the eigenvalue differences zi−zjz_i - z_j (1≤i,j≤N1 \le i, j \le N).

Two questions arise for these Schur multipliers: how the averaged power multiplier acts entrywise and behaves with respect to the identity and the matrix trace, and how Gaussian kernels and power factors express as unitary-conjugation averages and Gamma mixtures.

Schur Multipliers and Mixing Density

Fix a finite index set Λ\Lambda; write ℂΛ×Λ\mathbb{C}^{\Lambda \times \Lambda} for the algebra of complex matrices indexed by Λ\Lambda and ℝΛ\mathbb{R}^\Lambda for the space of real tuples z=(zi)i∈Λz = (z_i)_{i \in \Lambda}. Given z∈ℝΛz \in \mathbb{R}^\Lambda and t∈ℝt \in \mathbb{R}, the phase unitary Uz(t)U_z(t) in the unitary group of ℂΛ×Λ\mathbb{C}^{\Lambda \times \Lambda} is the diagonal matrix Uz(t)=diag⁡(eitzi)i∈Λ.U_z(t) = \operatorname{diag}\bigl(e^{i t z_i}\bigr)_{i \in \Lambda}.(1)✦

The Gaussian Schur multiplier 𝒢z,v,h\mathcal{G}_{z, v, h} associated with z∈ℝΛz \in \mathbb{R}^\Lambda and v,h∈ℝv, h \in \mathbb{R} is the map on ℂΛ×Λ\mathbb{C}^{\Lambda \times \Lambda} sending each matrix D∈ℂΛ×ΛD \in \mathbb{C}^{\Lambda \times \Lambda} to the matrix 𝒢z,v,h(D)∈ℂΛ×Λ\mathcal{G}_{z, v, h}(D) \in \mathbb{C}^{\Lambda \times \Lambda} with entries (𝒢z,v,h(D))ij=e−vh(zi−zj)2Dij(i,j∈Λ).\bigl(\mathcal{G}_{z, v, h}(D)\bigr)_{ij} = e^{-v h (z_i - z_j)^2} D_{ij} \qquad (i, j \in \Lambda).(2)✦

For σ∈ℝ\sigma \in \mathbb{R}, the Gamma mixing density wσ:ℝ→ℝw_\sigma : \mathbb{R} \to \mathbb{R} is the function defined for v∈ℝv \in \mathbb{R} by wσ(v)=Γ(1−σ)−1v−σe−v.w_\sigma(v) = \Gamma(1 - \sigma)^{-1} v^{-\sigma} e^{-v}.(3)✦

With σ,h∈ℝ\sigma, h \in \mathbb{R} and z∈ℝΛz \in \mathbb{R}^\Lambda, the power Schur multiplier 𝒮σ,z,h\mathcal{S}_{\sigma, z, h} is the map on ℂΛ×Λ\mathbb{C}^{\Lambda \times \Lambda} sending each D∈ℂΛ×ΛD \in \mathbb{C}^{\Lambda \times \Lambda} to the matrix 𝒮σ,z,h(D)∈ℂΛ×Λ\mathcal{S}_{\sigma, z, h}(D) \in \mathbb{C}^{\Lambda \times \Lambda} with entries (𝒮σ,z,h(D))ij=(1+h(zi−zj)2)σ−1Dij(i,j∈Λ).\bigl(\mathcal{S}_{\sigma, z, h}(D)\bigr)_{ij} = \bigl(1 + h (z_i - z_j)^2\bigr)^{\sigma - 1} D_{ij} \qquad (i, j \in \Lambda).(4)✦

Averaging the power Schur multiplier (4) yields, for σ,a∈ℝ\sigma, a \in \mathbb{R} and z∈ℝΛz \in \mathbb{R}^\Lambda, the tangent Schur multiplier 𝒫σ,a,z\mathcal{P}_{\sigma, a, z}, which is the map on ℂΛ×Λ\mathbb{C}^{\Lambda \times \Lambda} sending each D∈ℂΛ×ΛD \in \mathbb{C}^{\Lambda \times \Lambda} to 𝒫σ,a,z(D)=∫01𝒮σ,z,τ(1−τ)/a(D)dτ.\mathcal{P}_{\sigma, a, z}(D) = \int_0^1 \mathcal{S}_{\sigma, z, \tau(1 - \tau)/a}(D) \, d\tau.(5)✦

The Claim

Theorem 1. Let Λ\Lambda be a finite index set, let z∈ℝΛz \in \mathbb{R}^\Lambda, and let σ∈ℝ\sigma \in \mathbb{R}. When a>0a > 0, any matrix Z∈ℂΛ×ΛZ \in \mathbb{C}^{\Lambda \times \Lambda} and indices i,j∈Λi, j \in \Lambda satisfy (𝒫σ,a,z(Z))ij=(∫01(1+τ(1−τ)a(zi−zj)2)σ−1dτ)Zij.\bigl(\mathcal{P}_{\sigma, a, z}(Z)\bigr)_{ij} = \left( \int_0^1 \left(1 + \frac{\tau(1 - \tau)}{a}(z_i - z_j)^2\right)^{\sigma - 1} d\tau \right) Z_{ij}.(6)✦ Given D∈ℂΛ×ΛD \in \mathbb{C}^{\Lambda \times \Lambda} and v,h≥0v, h \ge 0, let 𝒩(0,2vh)\mathcal{N}(0, 2vh) be the centered real Gaussian probability measure of variance 2vh2vh; then 𝒢z,v,h(D)=∫ℝUz(t)DUz(t)*d𝒩(0,2vh)(t).\mathcal{G}_{z, v, h}(D) = \int_{\mathbb{R}} U_z(t) D U_z(t)^* \, d\mathcal{N}(0, 2vh)(t).(7)✦ If σ<1\sigma < 1, h≥0h \ge 0, and x∈ℝx \in \mathbb{R}, then ∫0∞wσ(v)e−vhx2dv=(1+hx2)σ−1.\int_0^\infty w_\sigma(v) e^{-v h x^2} \, dv = (1 + h x^2)^{\sigma - 1}.(8)✦ With I∈ℂΛ×ΛI \in \mathbb{C}^{\Lambda \times \Lambda} denoting the identity matrix, for every a∈ℝa \in \mathbb{R} one has 𝒫σ,a,z(I)=I,\mathcal{P}_{\sigma, a, z}(I) = I,(9)✦ and whenever a>0a > 0 and D,M∈ℂΛ×ΛD, M \in \mathbb{C}^{\Lambda \times \Lambda}, tr⁡(M𝒫σ,a,z(D))=tr⁡(𝒫σ,a,z(M)D).\operatorname{tr}\bigl(M \mathcal{P}_{\sigma, a, z}(D)\bigr) = \operatorname{tr}\bigl(\mathcal{P}_{\sigma, a, z}(M) D\bigr).(10)✦

In the paper, the result reads: in an eigenbasis of BB with eigenvalues ziz_i (taking Λ={1,…,N}\Lambda = \{1, \dots, N\}, with aa standing for d+δd + \delta, 𝒫σ,a,z\mathcal{P}_{\sigma, a, z} for 𝒫\mathcal{P}, and Uz(t)U_z(t) for eitBe^{itB}), the map 𝒫\mathcal{P} is (𝒫Z)ij=PijZij(\mathcal{P} Z)_{ij} = P_{ij} Z_{ij} with Pij=∫01(1+τ(1−τ)(zi−zj)2/(d+δ))σ−1dτP_{ij} = \int_0^1 (1 + \tau(1 - \tau)(z_i - z_j)^2 / (d + \delta))^{\sigma - 1} \, d\tau, there exists an even probability measure ν\nu on ℝ\mathbb{R} such that 𝒫(Z)=∫ℝeitBZe−itBdν(t)\mathcal{P}(Z) = \int_{\mathbb{R}} e^{itB} Z e^{-itB} \, d\nu(t), and in particular 𝒫(I)=I\mathcal{P}(I) = I and 𝒫\mathcal{P} is self-adjoint for the trace pairing on Hermitian matrices.

In the library, the matrix statements take the real vector zz, parameter aa, and arbitrary matrices in ℂΛ×Λ\mathbb{C}^{\Lambda \times \Lambda} as inputs, while the Gaussian conjugation average (7) and the scalar Gamma mixture (8) appear as individual identities.

What Is Checked

The five parts in the claim are stated over a finite index set Λ\Lambda, a real vector z∈ℝΛz \in \mathbb{R}^\Lambda, real parameters σ\sigma, aa, vv, hh, and xx (with a>0a > 0 in (6) and (10), v,h≥0v, h \ge 0 in (7), and σ<1\sigma < 1 in (8)), complex matrices Z,D,M∈ℂΛ×ΛZ, D, M \in \mathbb{C}^{\Lambda \times \Lambda}, the identity matrix II, the matrix trace, the centered real Gaussian probability measure 𝒩(0,2vh)\mathcal{N}(0, 2vh), the phase unitary Uz(t)U_z(t), the Gamma mixing density wσw_\sigma, and the Schur multipliers 𝒢z,v,h\mathcal{G}_{z, v, h}, 𝒮σ,z,h\mathcal{S}_{\sigma, z, h}, and 𝒫σ,a,z\mathcal{P}_{\sigma, a, z}.