OctoInformBeta

OpenAI preprint, September 2026 · §3. A regularized matrix field · §3

Lemma 3.2: Pointwise cutoff convergence and local Lipschitz continuity on Hermitian matrices, Fréchet and path derivatives for parameters σ<1\sigma < 1, δ>0\delta > 0

Matrix field regularity and differentiation of its scalar primitive.

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 study of sharp one-dimensional Lieb-Thirring constants 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), for a natural number NN one considers a regularized matrix field MM on the space SymN\mathrm{Sym}_N of N×NN \times N real symmetric matrices together with its scalar primitive JJ. Both constructions are formulated for general real parameters σ<1\sigma < 1 and δ>0\delta > 0 with a vector k∈ℝNk \in \mathbb{R}^N, and admit extensions to complex Hermitian matrices of arbitrary finite size.

The question concerns convergence of the symmetric-cutoff integral and local Lipschitz regularity of the field, together with differentiation of both JJ and its complex Hermitian potential counterpart against the matrix trace pairing.

Definitions

The field normalization constant for a real parameter σ\sigma is the real number cσc_\sigma defined by (1): cσ=σ∫ℝ(1+s2)σ−3/2ds.c_\sigma = \frac{\sigma}{\int_{\mathbb{R}} (1 + s^2)^{\sigma - 3/2}\,ds}.(1)✦

For real parameters σ\sigma and δ\delta, writing t+=max⁡(t,0)t_+ = \max(t, 0), the regularized scalar primitive is the real-valued function fσ,δf_{\sigma,\delta} on ℝ\mathbb{R} given by (2), and the second scalar primitive is the real-valued function gσ,δg_{\sigma,\delta} on ℝ\mathbb{R} given by (3): fσ,δ(y)=∫0y(t++δ)σ−3/2dt,f_{\sigma,\delta}(y) = \int_0^y (t_+ + \delta)^{\sigma - 3/2}\,dt,(2)✦ gσ,δ(y)=∫0yfσ,δ(t)dt.g_{\sigma,\delta}(y) = \int_0^y f_{\sigma,\delta}(t)\,dt.(3)✦

Given a natural number NN, a vector k∈ℝNk \in \mathbb{R}^N, an N×NN \times N real matrix BB, and s∈ℝs \in \mathbb{R}, the real matrix pencil is the N×NN \times N real matrix Xk,B(s)X_{k,B}(s) defined by (4), where II is the N×NN \times N identity matrix and D(k)D(k) is the diagonal matrix whose ii-th diagonal entry is ki2k_i^2 for each index ii: Xk,B(s)=D(k)−2sB+s2I.X_{k,B}(s) = D(k) - 2sB + s^2 I.(4)✦

With σ,δ∈ℝ\sigma, \delta \in \mathbb{R}, a natural number NN, and k∈ℝNk \in \mathbb{R}^N, the truncated matrix field at an N×NN \times N real matrix BB and real cutoff RR is the N×NN \times N real matrix MR(B)M_R(B) given by (5), where fσ,δ(Xk,B(s))f_{\sigma,\delta}(X_{k,B}(s)) is evaluated via the continuous functional calculus: MR(B)=cσ∫−RR(fσ,δ(Xk,B(s))−fσ,δ(s2)I)ds.M_R(B) = c_\sigma \int_{-R}^{R} \bigl(f_{\sigma,\delta}(X_{k,B}(s)) - f_{\sigma,\delta}(s^2) I\bigr)\,ds.(5) Passing to R→+∞R \to +\infty, the matrix field for σ,δ∈ℝ\sigma, \delta \in \mathbb{R}, k∈ℝNk \in \mathbb{R}^N, and an N×NN \times N real matrix BB is the N×NN \times N real matrix M(B)M(B) defined by (6): M(B)=limR→+∞MR(B).M(B) = \lim_{R \to +\infty} M_R(B).(6)✦ Integrating the trace pairing along t↦tBt \mapsto tB defines the matrix field primitive J(B)∈ℝJ(B) \in \mathbb{R} for σ,δ∈ℝ\sigma, \delta \in \mathbb{R}, k∈ℝNk \in \mathbb{R}^N, and an N×NN \times N real matrix BB by (7): J(B)=−∫01tr⁡(BM(tB))dt.J(B) = -\int_0^1 \operatorname{tr}\bigl(B\, M(tB)\bigr)\,dt.(7)✦

Write ℋn\mathcal{H}_n for the real vector space of n×nn \times n complex Hermitian matrices over a finite index set nn; the matrix complexification map sends each n×nn \times n real matrix AA to the n×nn \times n complex matrix AℂA_{\mathbb{C}} obtained by entrywise inclusion of ℝ\mathbb{R} into ℂ\mathbb{C}, as a real **-algebra homomorphism.✦ Restricting complexification to SymN\mathrm{Sym}_N for a natural number NN defines the Hermitian complexification map ιℂ:SymN→ℋN\iota_{\mathbb{C}} : \mathrm{Sym}_N \to \mathcal{H}_N, B↦BℂB \mapsto B_{\mathbb{C}}, and evaluating it on D(k)D(k) for k∈ℝNk \in \mathbb{R}^N yields the Hermitian diagonal matrix Dℂ(k)∈ℋND_{\mathbb{C}}(k) \in \mathcal{H}_N.

Over a finite index set nn, the complex matrix pencil for n×nn \times n complex matrices CC and BB and s∈ℝs \in \mathbb{R} is the n×nn \times n complex matrix XC,Bℂ(s)X^{\mathbb{C}}_{C,B}(s) given by (8), where II is the n×nn \times n identity matrix: XC,Bℂ(s)=s2I+(C−2sB).X^{\mathbb{C}}_{C,B}(s) = s^2 I + (C - 2sB).(8)✦ Evaluating fσ,δ(XC,Bℂ(±s))f_{\sigma,\delta}(X^{\mathbb{C}}_{C,B}(\pm s)) via the continuous functional calculus for σ,δ∈ℝ\sigma, \delta \in \mathbb{R}, n×nn \times n complex matrices CC and BB, and s∈ℝs \in \mathbb{R} defines the matrix field integrand Fσ,δ(C,B,s)F_{\sigma,\delta}(C, B, s) in (9) and the paired field integrand F̃σ,δ(C,B,s)\widetilde{F}_{\sigma,\delta}(C, B, s) in (10), both n×nn \times n complex matrices: Fσ,δ(C,B,s)=fσ,δ(XC,Bℂ(s))−fσ,δ(s2)I,F_{\sigma,\delta}(C, B, s) = f_{\sigma,\delta}\bigl(X^{\mathbb{C}}_{C,B}(s)\bigr) - f_{\sigma,\delta}(s^2) I,(9)✦ F̃σ,δ(C,B,s)=fσ,δ(XC,Bℂ(s))+fσ,δ(XC,Bℂ(−s))−2fσ,δ(s2)I.\widetilde{F}_{\sigma,\delta}(C, B, s) = f_{\sigma,\delta}\bigl(X^{\mathbb{C}}_{C,B}(s)\bigr) + f_{\sigma,\delta}\bigl(X^{\mathbb{C}}_{C,B}(-s)\bigr) - 2 f_{\sigma,\delta}(s^2) I.(10)✦ Integrating the paired integrand defines the Hermitian matrix field for σ,δ∈ℝ\sigma, \delta \in \mathbb{R} and n×nn \times n complex matrices CC and BB as the n×nn \times n complex matrix MHerm(σ,δ,C,B)M_{\mathrm{Herm}}(\sigma, \delta, C, B) in (11): MHerm(σ,δ,C,B)=cσ2∫ℝF̃σ,δ(C,B,s)ds.M_{\mathrm{Herm}}(\sigma, \delta, C, B) = \frac{c_\sigma}{2} \int_{\mathbb{R}} \widetilde{F}_{\sigma,\delta}(C, B, s)\,ds.(11)✦

When C,B∈ℋnC, B \in \mathcal{H}_n are Hermitian matrices over a finite index set nn and σ,δ∈ℝ\sigma, \delta \in \mathbb{R}, the paired trace potential is the real number ϕσ,δ(C,B,s)\phi_{\sigma,\delta}(C, B, s) at s∈ℝs \in \mathbb{R} given for s≠0s \neq 0 by (12) and equal to 00 at s=0s = 0, with gσ,δ(XC,Bℂ(±s))g_{\sigma,\delta}(X^{\mathbb{C}}_{C,B}(\pm s)) evaluated via the continuous functional calculus: ϕσ,δ(C,B,s)=−Re⁡tr⁡(gσ,δ(XC,Bℂ(s)))−Re⁡tr⁡(gσ,δ(XC,Bℂ(−s)))2s−2fσ,δ(s2)Re⁡tr⁡(B).\phi_{\sigma,\delta}(C, B, s) = -\frac{\operatorname{Re}\operatorname{tr}\bigl(g_{\sigma,\delta}(X^{\mathbb{C}}_{C,B}(s))\bigr) - \operatorname{Re}\operatorname{tr}\bigl(g_{\sigma,\delta}(X^{\mathbb{C}}_{C,B}(-s))\bigr)}{2s} - 2 f_{\sigma,\delta}(s^2) \operatorname{Re}\operatorname{tr}(B).(12)✦ Associated with σ,δ∈ℝ\sigma, \delta \in \mathbb{R} and C,B∈ℋnC, B \in \mathcal{H}_n, the Hermitian matrix potential is the real number Φσ,δ(C,B)\Phi_{\sigma,\delta}(C, B) defined by (13): Φσ,δ(C,B)=cσ2∫ℝϕσ,δ(C,B,s)ds.\Phi_{\sigma,\delta}(C, B) = \frac{c_\sigma}{2} \int_{\mathbb{R}} \phi_{\sigma,\delta}(C, B, s)\,ds.(13)✦

Each n×nn \times n complex matrix AA over a finite index set nn determines the real trace pairing functional τA\tau_A, the continuous real-linear functional on n×nn \times n complex matrices given by τA(X)=Re⁡tr⁡(AX)\tau_A(X) = \operatorname{Re}\operatorname{tr}(A X).✦ Composing τA\tau_A with the canonical inclusion of ℋn\mathcal{H}_n into all n×nn \times n complex matrices defines the continuous real-linear trace gradient functional ℓA\ell_A on ℋn\mathcal{H}_n, and the trace gradient map A↦ℓAA \mapsto \ell_A is continuous and real-linear.✦

The Claim

Theorem 1. Let σ,δ∈ℝ\sigma, \delta \in \mathbb{R} satisfy σ<1\sigma < 1 and δ>0\delta > 0. For any finite index set nn and any C,B∈ℋnC, B \in \mathcal{H}_n, the cutoff limit (14) holds: limR→+∞cσ∫−RRFσ,δ(C,B,s)ds=MHerm(σ,δ,C,B).\lim_{R \to +\infty} c_\sigma \int_{-R}^{R} F_{\sigma,\delta}(C, B, s)\,ds = M_{\mathrm{Herm}}(\sigma, \delta, C, B).(14)✦ For any finite index set nn and any C∈ℋnC \in \mathcal{H}_n, the Hermitian field satisfies (15): B↦MHerm(σ,δ,C,B)is locally Lipschitz on ℋn,B \longmapsto M_{\mathrm{Herm}}(\sigma, \delta, C, B) \quad \text{is locally Lipschitz on } \mathcal{H}_n,(15)✦ and for any B∈ℋnB \in \mathcal{H}_n, the map Φσ,δ(C,⋅)\Phi_{\sigma,\delta}(C, \cdot) is Fréchet differentiable at BB with Fréchet derivative (16): DBΦσ,δ(C,B)=ℓMHerm(σ,δ,C,B).D_B \Phi_{\sigma,\delta}(C, B) = \ell_{M_{\mathrm{Herm}}(\sigma,\delta,C,B)}.(16)✦ Let NN be a natural number and k∈ℝNk \in \mathbb{R}^N. For any B∈SymNB \in \mathrm{Sym}_N, the primitive JJ is Fréchet differentiable at BB with Fréchet derivative (17): DJ(B)=−ℓM(B)ℂ∘ιℂ.D J(B) = -\ell_{M(B)_{\mathbb{C}}} \circ \iota_{\mathbb{C}}.(17)✦ For any path Y:ℝ→SymNY : \mathbb{R} \to \mathrm{Sym}_N differentiable at x∈ℝx \in \mathbb{R} with derivative Y′∈SymNY' \in \mathrm{Sym}_N, the composite t↦J(Y(t))t \mapsto J(Y(t)) is differentiable at xx with derivative (18): ddtJ(Y(t))|t=x=−tr⁡(M(Y(x))Y′).\left.\frac{d}{dt} J\bigl(Y(t)\bigr)\right|_{t=x} = -\operatorname{tr}\bigl(M(Y(x))\, Y'\bigr).(18)✦ For any B∈SymNB \in \mathrm{Sym}_N, the primitive JJ satisfies (19): J(B)=−Φσ,δ(Dℂ(k),ιℂ(B)).J(B) = -\Phi_{\sigma,\delta}\bigl(D_{\mathbb{C}}(k), \iota_{\mathbb{C}}(B)\bigr).(19)✦

In the paper, the result reads: the cutoff limit MR→MM_R \to M as R→+∞R \to +\infty is locally uniform, the resulting map MM is locally Lipschitz, and J(B)=−∫01tr⁡(BM(tB))dtJ(B) = -\int_0^1 \operatorname{tr}(B M(tB))\,dt is C1C^1 with dJ(B)[H]=−tr⁡(M(B)H)dJ(B)[H] = -\operatorname{tr}(M(B) H) for B,H∈SymNB, H \in \mathrm{Sym}_N.

What Is Checked

The theorem is stated for real parameters σ<1\sigma < 1 and δ>0\delta > 0 over the normalization cσc_\sigma, the integrand Fσ,δF_{\sigma,\delta}, the Hermitian field MHermM_{\mathrm{Herm}} and potential Φσ,δ\Phi_{\sigma,\delta} on ℋn\mathcal{H}_n for any finite index set nn and C,B∈ℋnC, B \in \mathcal{H}_n, the trace gradient map A↦ℓAA \mapsto \ell_A, and, for any natural number NN and k∈ℝNk \in \mathbb{R}^N, the matrix field MM and primitive JJ on SymN\mathrm{Sym}_N (along a path Y:ℝ→SymNY : \mathbb{R} \to \mathrm{Sym}_N differentiable at x∈ℝx \in \mathbb{R} with derivative Y′∈SymNY' \in \mathrm{Sym}_N), together with matrix complexification A↦AℂA \mapsto A_{\mathbb{C}}, Hermitian complexification ιℂ\iota_{\mathbb{C}}, and Dℂ(k)D_{\mathbb{C}}(k).

Hermitian complexification ιℂ\iota_{\mathbb{C}} on SymN\mathrm{Sym}_N is continuous and preserves real scalar multiplication.✦