betaFeedback

WorksPhysics and Astronomy

Sharp bounds on the Eliashberg critical temperature: a comparison principle, convexity in the spectrum, and the Kiessling-Altshuler-Yuzbashyan conjectures

Abstract

In Eliashberg theory of phonon-mediated superconductivity with vanishing Coulomb pseudopotential, the critical temperature TcT_c marks the linear instability of the normal state. Kiessling, Altshuler and Yuzbashyan (KAY) characterized it through the top eigenvalue of a compact operator on Matsubara-frequency sequences. They proved that TcT_c is unique at strong coupling and conjectured Tc<C(λ⟨ω2⟩)1/2T_c<C(λ\langleω^2\rangle)^{1/2} for every coupling λλ, with ⟨ω2⟩\langleω^2\rangle the second moment of the normalized phonon spectrum and C≈0.1827C\approx0.1827 sharp, a bound Leavens stated in 1977 on partly numerical grounds. A preprint by Kondratev, not yet peer reviewed, gives a proof of this bound and of all-coupling uniqueness. We give an independent proof, for every spectrum with finite second moment, based on a comparison principle that is also Kondratev's central step and was proved to first order by Bergmann and Rainer: strengthening the pairing interaction at every Matsubara transfer, strictly at the smallest one, strictly raises the top eigenvalue. At fixed λλ and ⟨ω2⟩\langleω^2\rangle a single Einstein mode maximizes TcT_c. Convexity of the top eigenvalue in the spectrum gives sharp bounds on TcT_c by the moments ∫ωnα2Fdω\intω^nα^2F dω, 0≤n≤10\le n\le1, including Leavens's zeroth-moment bound Tc≤c0∫α2FdωT_c\le c_0\intα^2F dω, c0≈0.231c_0\approx0.231 numerically, optimal at moderate coupling. All bounds persist under a Coulomb pseudopotential with symmetric cutoff. A prefix decomposition of non-increasing vectors proves a third KAY conjecture, not treated in the preprint. The finite-dimensional core, including that conjecture, is formalized in Lean 4, and two computer-assisted proofs independent of the comparison principle confirm the Einstein bound.

MSC Classification: 82D55, 47A75, 15B48, 47B65, 68V15

Introduction

Background

Eliashberg theory [1] is the standard framework for computing the critical temperature TcT_c of phonon-mediated superconductors [2, 3]. Its input is the Eliashberg spectral function α2F(ω)\alpha^2F(\omega), which measures how strongly electrons couple to lattice vibrations of frequency ω\omega. Two numbers summarize it: the dimensionless coupling constant λ=2∫0∞α2F(ω)ω−1dω\lambda= 2\int_0^\infty\alpha^2F(\omega)\omega^{-1}\mathrm{d}\omega, and the second moment ⟨ω2⟩\langle\omega^2\rangle of the normalized spectral measure P(dω)=2α2F(ω)/(λω)dωP(\mathrm{d}\omega) = 2\alpha^2F(\omega)/(\lambda\omega)\mathrm{d}\omega.

For weak coupling TcT_c is exponentially small in 1/λ1/\lambda. For strong coupling, Allen and Dynes [4] found that it grows without bound but only like a square root,

Tc≃Cλ⟨ω2⟩(λ→∞),C≈0.18,(1)T_c \simeq C\sqrt{\lambda\langle\omega^2\rangle}\qquad(\lambda\to\infty),\qquad C\approx0.18, \tag*{(1)}

in units with ℏ=kB=1\hbar= k_B = 1. The right-hand side of (1) has long been regarded as an upper bound on TcT_c at every coupling. Moussa and Cohen state it as one, Tc≤0.183λ⟨ω2⟩T_c \leq0.183\sqrt{\lambda\langle\omega^2\rangle} for μ∗=0\mu^* = 0, citing Allen and Dynes and Leavens [4, 5, 6]; see also [7].

Allen and Dynes also argued that a single sharp phonon line (an Einstein spectrum) maximizes TcT_c at fixed λ\lambda and ⟨ω2⟩\langle\omega^2\rangle: they proved that the Einstein spectrum maximizes each coefficient λ(n)\lambda(n) of their linearized gap equation at fixed λ\lambda and ⟨ω2⟩\langle\omega^2\rangle, and noted that the top eigenvalue, and with it TcT_c, is maximized by making all λ(n)\lambda(n) as large as possible [4 Sect. II and App. A]. Bergmann and Rainer had shown that for μ∗=0\mu^* = 0 added spectral weight at any frequency raises TcT_c to first order, that is, that the functional derivative δTc/δα2F(ω)\delta T_c/\delta\alpha^2F(\omega) is positive [8].

Leavens stated the bound itself as a rigorous least upper bound for μ∗=0\mu^* = 0, in the form Tc≤0.258(∫0∞α2F(ω) dω)1/2T_c \leq0.258\left(\int_0^\infty\alpha^2F(\omega)\,\mathrm{d}\omega\right)^{1/2} [6 eq. (20)]. The first moment equals λ⟨ω2⟩/2\lambda\langle\omega^2\rangle/2, so his constant is 0.258/2≈0.1820.258/\sqrt{2}\approx0.182 in the units of (1). His derivation has two steps. A first-order variational argument based on the positivity of Bergmann and Rainer shows that a spectrum maximizing TcT_c at fixed first moment can only be a single Einstein mode. This presupposes that a maximizer exists, whereas at fixed first moment the supremum is approached only by a mode of diverging weight at vanishing frequency, as he notes. The maximum over Einstein spectra is then read off from numerical solutions of the Eliashberg equations together with the strong-coupling asymptote [6 Sect. 2 and Fig. 1]; this second step is the content of Conjecture 1 of Part III below. Two years earlier he had stated, as rigorously shown, the zeroth-moment bound Tc≤0.2309∫0∞α2F(ω) dωT_c \leq0.2309\int_0^\infty\alpha^2F(\omega)\,\mathrm{d}\omega, with equality for the Einstein spectrum at frequency 1.750∫0∞α2F(ω) dω1.750\int_0^\infty\alpha^2F(\omega)\,\mathrm{d}\omega, that is at coupling λ=1.143\lambda= 1.143 [9 eqs. (13)–(14)]. Its derivation has the same two steps: the variational argument, which presupposes an optimal spectrum, followed by a numerical search over Einstein spectra [9 Sect. 2]; the 1977 paper restates the bound [6 eqs. (18)–(19)]. In both papers Leavens argued that the bounds persist for μ∗>0\mu^{*} > 0 because the Coulomb repulsion lowers TcT_c [6, 9 Sect. 3].

In a series of three papers [10, 11, 12], Kiessling, Altshuler and Yuzbashyan (KAY) made these questions precise. Building on the spin-chain formulation of Yuzbashyan and Altshuler [13], they showed that the normal state of the linearized, isotropic Eliashberg model with Coulomb pseudopotential μ∗=0\mu^{*} = 0 is linearly unstable precisely when λ\lambda times the top eigenvalue of an explicit compact self-adjoint operator exceeds one. They identified the constant in (1) as C=g(2)2π=0.1827262477…C = \frac{\sqrt{g(2)}}{2\pi} = 0.1827262477\ldots, where g(2)g(2) is the top eigenvalue of the operator of their γ\gamma-model at γ=2\gamma= 2 [10], and they stated the following conjectures.

  • Conjecture 1 of Part III [12] (Einstein phonons of frequency Ω\Omega): for all Ω≥0\Omega\ge0 and λ>0\lambda> 0 there is a critical temperature Tc(λ,Ω)>0T_c(\lambda, \Omega) > 0, and Tc(λ,Ω)<Ω2πg(2)λT_c(\lambda, \Omega) < \frac{\Omega}{2\pi}\sqrt{g(2)}\lambda.

  • Conjecture 2 of Part II [11] (general spectra PP): for all admissible PP and λ>0\lambda> 0 there is a critical temperature Tc(λ,P)>0T_c(\lambda, P) > 0, and Tc(λ,P)<12πg(2)⟨ω2⟩T_c(\lambda, P) < \frac{1}{2\pi}\sqrt{g(2)}\langle\omega^{2}\rangle.

  • Conjecture 3 of Part II [11]: a sharp lower bound, attained exactly by constant sequences, for a finite-dimensional Rayleigh quotient on non-negative non-increasing sequences.

KAY proved that TcT_c is unique when λ\lambda exceeds a spectrum-dependent threshold (for Einstein phonons, λ>0.7670…\lambda> 0.7670\ldots [12 Thm. 5]), that Tc/λ⟨ω2⟩T_c/\lambda\sqrt{\langle\omega^{2}\rangle} has the limit g(2)/2π\sqrt{g(2)}/2\pi as λ→∞\lambda\to\infty, and an explicit upper bound on TcT_c whose constant is about 2.03 times the conjectured one [11, 12]. They noted that a proof of uniqueness for all λ>0\lambda> 0, which the thermodynamic picture of the transition presupposes, would be mathematically desirable [11, 12]. Their monotonicity argument needs the dimensionless phonon frequency ϖ=Ω/(2πT)\varpi= \Omega/(2\pi T) to satisfy ϖ≤2\varpi\le\sqrt{2} [11 Lemma 3], and they remarked that proving monotonicity for all ϖ\varpi “would require a different strategy” [12]. Elezaby and Abanov sharpened rigorous bounds for the γ\gamma-model [14].

In September 2026 Kondratev posted a preprint [15], not yet peer reviewed, that gives a proof of Conjecture 2 of Part II, and with it Conjecture 1 of Part III, for every non-zero phonon measure with finite coupling and second moment. It shows that the top eigenvalue decreases strictly with temperature on all of (0,∞)(0, \infty), so that the onset temperature is unique for every λ>0\lambda> 0, and it also derives the optimality of Einstein spectra at fixed λ\lambda and ⟨ω⟩\langle\omega\rangle [15 Sect. 12]. Its central step, a coefficient-by-coefficient comparison of two kernels on the ordered top eigenvector [15 Prop. 1], is in substance our Theorem 3. We found the argument presented here without knowledge of that preprint and learned of it while preparing this manuscript. § 1.4 compares the two treatments.

Results

We prove all three conjectures, together with uniqueness of TcT_c at every coupling (Theorems 5, 10 and 4). Results 1–3 below are also obtained in [15]; results 4 and 5 are not treated there. Result 2 is the bound stated by Leavens, whose derivation is described above [6], and result 3 was argued by Allen and Dynes, with the monotonicity of the top eigenvalue in the coefficients read off from the equations rather than proved [4].

  1. Uniqueness. For every phonon spectrum with finite second moment and every λ>0\lambda> 0 there is exactly one critical temperature: λk(P,T)>1\lambda_k(P,T) > 1 for T<TcT < T_c and λk(P,T)<1\lambda_k(P,T) < 1 for T>TcT > T_c, so for KAY-admissible PP the linearized problem has a superconducting instability below TcT_c and none above. TcT_c increases continuously and strictly with λ\lambda.

  1. The bound. Tc<12πg(2)λ⟨ω2⟩≈0.1827λ⟨ω2⟩T_c < \frac{1}{2\pi}\sqrt{g(2)\lambda\langle\omega^2\rangle} \approx0.1827\sqrt{\lambda\langle\omega^2\rangle} for every λ>0\lambda> 0 and every spectrum. The constant cannot be improved.

  1. Einstein spectra are optimal. At fixed λ\lambda and ⟨ω2⟩\langle\omega^2\rangle, a single Einstein mode at frequency ⟨ω2⟩1/2\langle\omega^2\rangle^{1/2} has a strictly higher TcT_c than any other spectrum.

  1. Conjecture 3 of Part II holds.

  1. Moment bounds. For 0≤n≤10 \le n \le1, Tc≤12π(2Hn∫0∞ωnα2F(ω) dω)1/(n+1)T_c \le\frac{1}{2\pi}\left(2H_n\int_0^\infty\omega^n\alpha^2F(\omega)\,d\omega\right)^{1/(n+1)}, where HnH_n is a constant defined by the Einstein problem, and the bound is the supremum of TcT_c over Einstein spectra with the same moment (Theorem 8). For n=1n = 1 this is result 2; for n=0n = 0 it is Leavens’s bound Tc≤(H0/π)∫0∞α2F(ω) dωT_c \le(H_0/\pi)\int_0^\infty\alpha^2F(\omega)\,d\omega [9], where H0/π≈0.2309H_0/\pi\approx0.2309 in floating point.

The first three follow from a single statement, the comparison principle of Theorem 3: if the pairing interaction is increased at every Matsubara frequency transfer, and strictly at the smallest transfer, then the top eigenvalue strictly increases. Conjecture 3 is proved separately in § 6, by decomposing non-increasing vectors into prefix indicators. Result 5 rests on a second, elementary observation: the top eigenvalue is a supremum of functionals that are linear in the spectral measure, hence convex in it, so it is at most the average of the Einstein eigenvalues over the spectrum (Lemma 15), and at a fixed moment no spectrum does better than the supremum over Einstein spectra. This makes Leavens’s reduction to Einstein spectra rigorous without assuming that a maximizing spectrum exists. Combined with the computer-assisted proofs of the Einstein inequality, it also proves that the normal state is stable above the bound of result 2 without using the comparison principle. Uniqueness also yields these bounds for every instability temperature when a Coulomb pseudopotential with a symmetric frequency cutoff is added (Corollary 9), which makes Leavens’s remark on μ∗>0\mu^* > 0 rigorous in that model. Apart from this corollary, these are theorems about the linearized, isotropic Eliashberg model with μ∗=0\mu^* = 0; § 8 states what they do not cover.

The idea of the proof

The interaction between Matsubara frequencies ωn\omega_n and ωm\omega_m depends only on the transfer ∣n−m∣|n-m|, so the KAY operator is a sum K=∑j≥1ajMjK = \sum_{j\geq1} a_j M_j: one fixed operator MjM_j per transfer jj, weighted by the interaction strength aja_j at that transfer. Heuristically, raising aja_j changes the top eigenvalue, to first order, by ⟨ξ,Mjξ⟩\langle\xi,M_j\xi\rangle, where ξ\xi is the optimizer. Each MjM_j has two competing parts (Figure 1): a pairing part, which couples frequencies a distance jj apart, directly or after reflecting one of them to −ωm-\omega_m, and a negative diagonal part, which comes from the mass renormalization ZnZ_n and acts as pair breaking. On the general vector either part can win. On the optimizer the pairing part always wins: Lemma 11 writes ⟨ξ,Mjξ⟩\langle\xi,M_j\xi\rangle as a sum of terms yn+j(yn−yn+j)y_{n+j}(y_n-y_{n+j}) and ynymy_ny_m, where yn∝Δn/ωny_n \propto\Delta_n/\omega_n is the gap divided by the frequency, and Lemma 12 shows that yy is non-negative and non-increasing. Every term is then non-negative. KAY instead differentiated with respect to ϖ2\varpi^{2}, grouping the transfers with weights that decrease in jj only when ϖ≤2\varpi\le\sqrt{2} [11 Lemma 3]. The transfer decomposition removes this restriction. For the γ\gamma-model, KAY had already expanded their Rayleigh quotient over transfers, with coefficients that are non-negative on non-increasing vectors, to prove that the quotient is positive [11 eqs. (72)–(73)]; here the same expansion is used to compare different kernels. The first-order version of the whole argument goes back to Bergmann and Rainer [8 App. A]. Working with the pair-breaking eigenvalue of the linearized equations at a finite frequency cutoff, they obtained δTc/δα2F(ω)\delta T_{\mathrm{c}}/\delta\alpha^{2}F(\omega) from the Hellmann–Feynman theorem and showed that the coefficient of every transfer in it is non-negative, using that for μ∗=0\mu^{*}=0 the gap function is positive and decreasing in ∣ωn∣|\omega_{n}|. Theorem 3 is the corresponding statement for finite changes of the kernel: it needs no cutoff, no differentiability of the top eigenvalue and no monotonicity of the larger kernel, and the ordering of the optimizer is proved in full (Lemma 12). It also supplies the step in the Allen–Dynes argument for Einstein optimality that they read off from their equations: the top eigenvalue increases when every λ(n)\lambda(n) does, although the pair-breaking diagonal terms grow with them. Likewise, Theorem 5 replaces both steps of Leavens’s derivation: it needs no maximizing spectrum, and the Einstein inequality that he read off from numerical solutions is proved.

Figure 1. Sign pattern of the transfer operator MjM_j in (8) for j=3j=3 on the first ten Matsubara frequencies. The panels show the pairing terms TjT_j and HjH_j, the diagonal term −Dj-D_j, and their sum. Positive entries are solid blue and negative entries are hatched orange. The operator MjM_j is indefinite, but Lemma 11 shows that it is non-negative on the cone containing the optimizer.

Relation to Kondratev’s preprint

The proof of the bound in [15] and the one given here have the same structure. The expansion of the quadratic form over transfers [15 eqs. (9)–(10)] is our Lemma 11, the kernel comparison on the ordered top eigenvector [15 Prop. 1] is our Theorem 3, and both treatments obtain Einstein optimality from Jensen’s inequality, as Allen and Dynes did for the coefficients λ(n)\lambda(n) [4 App. A]. Both rest on the transfer-by-transfer positivity that Bergmann and Rainer used at first order [8]. The two treatments differ in four respects. First, the ordering of the top eigenvector is proved there by Perron–Frobenius theory and power iteration, and here by an extremal argument that avoids the Perron–Frobenius theorem (Lemma 12). Second, the passage to infinitely many frequencies uses norm convergence of the truncated eigenvectors there and weak compactness here. Third, the preprint encloses the constant, 1.318139<g<1.318151.318139 < g < 1.31815, in exact arithmetic, whereas our proofs use gg only through its definition as a top eigenvalue; our certificates instead prove the Einstein inequality h(ϖ)<gϖ2h(\varpi) < g\varpi^{2} itself without the comparison principle (Appendix § 10). Fourth, the preprint proves sharpness along a family of Einstein spectra, while Theorem 5(c) also gives the limit λ→∞\lambda\to\infty for each fixed spectrum with finite fourth moment. Conjecture 3 of Part II, the moment bounds of Theorem 8, Corollary 9 and the formalization of § 7 have no counterpart in [15].

Verification and organization

For every finite truncation, the cone lemma, transfer decomposition, comparison principle, Einstein bound, monotonicity and non-strict Einstein optimality for discrete spectra are formalized and machine-checked in Lean 4 with Mathlib [16, 17], as is Conjecture 3 in full (§ 7). The text supplies the compactness argument needed for infinitely many frequencies. Two further computer-assisted proofs of the Einstein bound, independent of the comparison principle, use interval and exact rational arithmetic (Appendix A).

Section § 2 derives the KAY operator from the linearized Eliashberg equations and collects its basic properties. Section § 3 states the results. Sections § 4–§ 6 contain the proofs. Section § 7 describes the formalization and Section § 8 the scope of the results and some open problems.

The linearized Eliashberg problem

From the Eliashberg equations to an eigenvalue problem

Let T>0T > 0 and ωn=πT(2n+1)\omega_n = \pi T(2n + 1), n∈Zn \in\mathbb{Z}, be the fermionic Matsubara frequencies. The isotropic Eliashberg equations with μ∗=0\mu^* = 0, linearized in the gap function Δn\Delta_n, read [2, 3, 18]

Znωn=ωn+πT∑m∈Zλ(ωn−ωm)sgn⁡ωm(2)Z_n \omega_n = \omega_n + \pi T \sum_{m \in\mathbb{Z}} \lambda(\omega_n - \omega_m) \operatorname{sgn}\omega_m \tag*{(2)}
ZnΔn=πT∑m∈Zλ(ωn−ωm)Δm∣ωm∣(3)Z_n \Delta_n = \pi T \sum_{m \in\mathbb{Z}} \lambda(\omega_n - \omega_m) \frac{\Delta_m}{\lvert\omega_m\rvert} \tag*{(3)}

with the interaction

λ(ν)=2∫0∞ωα2F(ω)ω2+ν2 dω=λ∫0∞ω2ω2+ν2 P(dω),λ=λ(0).\lambda(\nu) = 2 \int_{0}^{\infty} \frac{\omega\alpha^2 F(\omega)}{\omega^2 + \nu^2}\,\mathrm{d}\omega= \lambda\int_{0}^{\infty} \frac{\omega^2}{\omega^2 + \nu^2}\,P(\mathrm{d}\omega), \qquad\lambda= \lambda(0).

Since ωn−ωm=2πT(n−m)\omega_n - \omega_m = 2\pi T(n - m), the interaction depends only on the transfer ∣n−m∣\lvert n - m\rvert. We write

λ(ωn−ωm)=λa∣n−m∣,aj=aj(P,T):=∫0∞ω2ω2+(2πTj)2 P(dω),(4)\lambda(\omega_n - \omega_m) = \lambda a_{\lvert n-m\rvert}, \qquad a_j = a_j(P,T) := \int_{0}^{\infty} \frac{\omega^2}{\omega^2 + (2\pi Tj)^2}\,P(\mathrm{d}\omega), \tag*{(4)}

so that a0=1a_0 = 1 and 1>a1≥a2≥⋯>01 > a_1 \ge a_2 \ge\cdots> 0. Put wn:=2n+1w_n := 2n + 1 and An:=∑j=1najA_n := \sum_{j=1}^{n} a_j.

We look for even solutions, Δ−n−1=Δn\Delta_{-n-1} = \Delta_n, and restrict to n≥0n \ge0. Splitting the sum in (2) into m≥0m \ge0 and m<0m < 0 and telescoping gives ∑m∈Za∣n−m∣sgn⁡ωm=∑m≥0(a∣n−m∣−an+m+1)=1+2An\sum_{m \in\mathbb{Z}} a_{\lvert n-m\rvert}\operatorname{sgn}\omega_m = \sum_{m \ge0}(a_{\lvert n-m\rvert} - a_{n+m+1}) = 1 + 2A_n, hence

Zn=1+λ(1+2An)wn.(5)Z_n = 1 + \frac{\lambda(1 + 2A_n)}{w_n}. \tag*{(5)}

In (3) the same splitting gives ZnΔn=λ∑m≥0(a∣n−m∣+an+m+1)Δm/wmZ_n\Delta_n = \lambda\sum_{m \ge0}(a_{\lvert n-m\rvert} + a_{n+m+1})\Delta_m/w_m. Insert (5) and set yn:=Δn/wny_n := \Delta_n/w_n. The term m=nm = n of the first sum contributes λa0yn=λyn\lambda a_0 y_n = \lambda y_n on the right, which cancels the term λyn\lambda y_n coming from the 11 in 1+2An1 + 2A_n on the left. What remains is

wnyn=λ(∑m≥0[(1−δnm)a∣n−m∣+an+m+1]ym−2Anyn).w_n y_n=\lambda\left(\sum_{m\geq0}\left[(1-\delta_{nm})a_{\lvert n-m\rvert}+a_{n+m+1}\right]y_m-2A_n y_n\right).

Dividing by wn\sqrt{w_n} and writing ξn:=wn yn=Δn/wn\xi_n:=\sqrt{w_n}\,y_n=\Delta_n/\sqrt{w_n} turns this into the symmetric eigenvalue problem

ξ=λK[a] ξ,(6)\xi=\lambda K[a]\,\xi, \tag*{(6)}

where, for any summable sequence a=(aj)j≥1a=(a_j)_{j\geq1}, K[a]K[a] is the operator on real ℓ2(N0)\ell^2(\mathbb{N}_0) with matrix

K[a]nm=(1−δnm)a∣n−m∣+an+m+1wnwm−δnm2Anwn.(7)K[a]_{nm}=\frac{(1-\delta_{nm})a_{\lvert n-m\rvert}+a_{n+m+1}}{\sqrt{w_nw_m}}-\delta_{nm}\frac{2A_n}{w_n}. \tag*{(7)}

The value a0a_0 does not enter. This is the operator K(P,T)K(P,T) of KAY [11 eqs. (39)–(42)], whose quadratic form ⟨ξ,(I−λK)ξ⟩\langle\xi,(I-\lambda K)\xi\rangle agrees, up to a positive overall factor, with the second variation of their spin-chain free energy at the normal state. KAY proved [11 Thms. 1, 1+ and 3] that the normal state is linearly unstable against superconducting perturbations if and only if λk(P,T)>1\lambda k(P,T)>1, where

k(P,T):=sup⁡spec⁡K[a(P,T)]k(P,T):=\sup\operatorname{spec} K[a(P,T)]

is the largest eigenvalue. The critical temperature is therefore the temperature at which λk(P,T)=1\lambda k(P,T)=1.

In KAY’s spin-chain formulation, yn=ξn/wn=Δn/wny_n=\xi_n/\sqrt{w_n}=\Delta_n/w_n is, to linear order, πT\pi T times the spin angle θn\theta_n of [11, 13]. It is proportional to Δn/ωn\Delta_n/\omega_n rather than to the gap itself. The condition ξ∈ℓ2\xi\in\ell^2 means ∑nΔn2/wn<∞\sum_n\Delta_n^2/w_n<\infty.

Kernels

Definition 1. A kernel is a sequence a=(aj)j≥1a=(a_j)_{j\geq1} of positive numbers that is non-increasing in jj and summable.

We use three families of kernels.

  • General spectra. Let P2\mathcal{P}_2 be the set of probability measures on (0,∞)(0,\infty) with ⟨ω2⟩:=∫ω2P(dω)<∞\langle\omega^2\rangle:=\int\omega^2P(\mathrm{d}\omega)<\infty. For P∈P2P\in\mathcal{P}_2 and T>0T>0, a(P,T)a(P,T) in (4) is a kernel, since aj(P,T)<⟨ω2⟩/(2πTj)2a_j(P,T)<\langle\omega^2\rangle/(2\pi Tj)^2. KAY’s admissible class [11], spectra supported in [0,Ω(P)][0,\Omega(P)] with density proportional to ω\omega near 00, is contained in P2\mathcal{P}_2.

  • Einstein phonons. For P=δΩP=\delta_\Omega the kernel depends only on ϖ:=Ω/(2πT)\varpi:=\Omega/(2\pi T), namely aj=ϖ2/(ϖ2+j2)a_j=\varpi^2/(\varpi^2+j^2). We write h(ϖ):=sup⁡spec⁡K[(ϖ2/(ϖ2+j2))j]h(\varpi):=\sup\operatorname{spec}K[(\varpi^2/(\varpi^2+j^2))_{j}]; this is the top eigenvalue of KAY’s Einstein operator [12].

  • The γ\gamma-model at γ=2\gamma=2. The kernel (j−2)j≥1(j^{-2})_{j\geq1} gives KAY’s γ\gamma-model operator G0:=K[(j−2)j]G_0:=K[(j^{-2})_j] at γ=2\gamma=2 [10], and

g:=g(2)=sup⁡spec⁡G0=1.31814…g:=g(2)=\sup\operatorname{spec}G_0=1.31814\ldots

This operator describes the strong-coupling regime: as T→∞T\to\infty one has aj(P,T)≈⟨ω2⟩(2πT)−2j−2a_j(P,T)\approx\langle\omega^2\rangle(2\pi T)^{-2}j^{-2}, hence k(P,T)≈g⟨ω2⟩/(2πT)2k(P,T)\approx g\langle\omega^2\rangle/(2\pi T)^2.

Basic properties of the operator

Write [⋅][\cdot] for the indicator of a statement. Every K[a]K[a] is a weighted sum of fixed operators, one per transfer:

K[a]=∑j≥1ajMj,Mj=Tj+Hj−Dj,(8)K[a]=\sum_{j\geq1}a_jM_j,\qquad M_j=T_j+H_j-D_j, \tag*{(8)}
(Tj)nm=[ ∣n−m∣=j ]wnwm,(Hj)nm=[ n+m+1=j ]wnwm,(Dj)nm=δnm2[n≥j]wn.(T_j)_{nm}=\frac{[\,|n-m|=j\,]}{\sqrt{w_nw_m}},\qquad(H_j)_{nm}=\frac{[\,n+m+1=j\,]}{\sqrt{w_nw_m}},\qquad(D_j)_{nm}=\delta_{nm}\frac{2[n\geq j]}{w_n}.

Here TjT_j (Toeplitz) couples frequencies at distance jj, HjH_j (Hankel) couples ωn\omega_n to the reflected frequency −ωm-\omega_m at distance jj, and DjD_j is the contribution of the transfer jj to the mass renormalization (5): aja_j enters ZnZ_n for every n≥jn\geq j. Figure 1 shows the pattern.

Lemma 2 Let a,ba,b be summable sequences of real numbers.

(a) ∥Tj∥≤2/2j+1\lVert T_j\rVert\leq2/\sqrt{2j+1}, ∥Hj∥≤1/2j−1\lVert H_j\rVert\leq1/\sqrt{2j-1} and ∥Dj∥=2/(2j+1)\lVert D_j\rVert=2/(2j+1); hence ∥Mj∥<3\lVert M_j\rVert<3, the series (8) converges in operator norm, and ∥K[a]−K[b]∥≤3∑j∣aj−bj∣\lVert K[a]-K[b]\rVert\leq3\sum_j|a_j-b_j|.

(b) K[a]K[a] is compact and self-adjoint, and K[sa]=sK[a]K[sa]=sK[a] for s∈Rs\in\mathbb{R}.

(c) If a1>0a_1>0, then κ[a]:=sup⁡spec⁡K[a]≥K[a]00=a1>0\kappa[a]:=\sup\operatorname{spec}K[a]\geq K[a]_{00}=a_1>0, so κ[a]\kappa[a] is the largest eigenvalue of K[a]K[a].

(d) Let K(N)[a]K^{(N)}[a] be the N×NN\times N upper-left block of K[a]K[a] (the truncation to the first NN positive Matsubara frequencies) and κ(N)[a]\kappa^{(N)}[a] its largest eigenvalue. Then κ(N)[a]≤κ(N+1)[a]\kappa^{(N)}[a]\leq\kappa^{(N+1)}[a], and κ(N)[a]→κ[a]\kappa^{(N)}[a]\to\kappa[a] as N→∞N\to\infty.

Proof. (a) Tj=Sj+SjTT_j=S_j+S_j^{\mathsf{T}}, where SjS_j maps en+j↦en/wnwn+je_{n+j}\mapsto e_n/\sqrt{w_nw_{n+j}} and has norm sup⁡n(wnwn+j)−1/2=(2j+1)−1/2\sup_n(w_nw_{n+j})^{-1/2}=(2j+1)^{-1/2}. HjH_j is a finite anti-diagonal matrix with entries (wnwj−1−n)−1/2≤(2j−1)−1/2(w_nw_{j-1-n})^{-1/2}\leq(2j-1)^{-1/2}. DjD_j is diagonal with largest entry $2/w_j$. For j=1j=1 the three bounds sum to 2/3+1+2/3<32/\sqrt{3}+1+2/3<3, and all three decrease in jj. (b) SjS_j is a weighted shift with weights tending to 00, HjH_j has finite rank and DjD_j is diagonal with entries tending to 00; so each MjM_j is compact, and so is the norm-convergent sum. (c) K[a]00=a1K[a]_{00}=a_1 because $A_0=0. A compact self-adjoint operator whose spectrum contains points above 0 has its supremum as an eigenvalue. (d) K(N)K^{(N)} is a compression of K(N+1)K^{(N+1)}, so κ(N)\kappa^{(N)} is non-decreasing in NN and κ(N)≤sup⁡spec⁡K\kappa^{(N)} \le\sup\operatorname{spec} K. Every finitely supported unit vector lies in some truncation, and such vectors are dense in the unit sphere; since the quadratic form of the bounded operator KK is continuous, sup⁡Nκ(N)[a]=sup⁡spec⁡K[a]\sup_N \kappa^{(N)}[a] = \sup\operatorname{spec} K[a]. □

Main results

Theorem 3 (Comparison principle) Let aa be a kernel and b=(bj)j≥1b = (b_j)_{j \ge1} a sequence with bj≥ajb_j \ge a_j for every j≥1j \ge1, b1>a1b_1 > a_1, and ∑j(bj−aj)<∞\sum_j (b_j-a_j) < \infty. Then

κ[b]>κ[a],andκ(N)[b]>κ(N)[a]for every N≥1.\kappa[b] > \kappa[a], \qquad\text{and} \qquad\kappa^{(N)}[b] > \kappa^{(N)}[a] \quad\text{for every } N \ge1.

In Theorem 3 only aa needs to be non-increasing; bb need not be monotone. Monotonicity of aa is used only to locate the optimizer of K[a]K[a] (Lemma 12). We write k(N)(P,T):=κ(N)[a(P,T)]k^{(N)}(P,T) := \kappa^{(N)}[a(P,T)] and h(N)(ϖ):=κ(N)[(ϖ2/(ϖ2+j2))j]h^{(N)}(\varpi) := \kappa^{(N)}[(\varpi^2/(\varpi^2+j^2))_j] for the corresponding truncated eigenvalues.

Theorem 4 (Uniqueness of TcT_c at every coupling)

(a) The Einstein eigenvalue ϖ↦h(ϖ)\varpi\mapsto h(\varpi) is continuous and strictly increasing on (0,∞)(0,\infty), with h(0+)=0h(0+) = 0 and h(ϖ)→∞h(\varpi) \to\infty as ϖ→∞\varpi\to\infty.

(b) For every P∈P2P \in\mathcal{P}_2, T↦k(P,T)T \mapsto k(P,T) is continuous and strictly decreasing on (0,∞)(0,\infty), with k(P,T)→∞k(P,T) \to\infty as T→0T \to0 and k(P,T)→0k(P,T) \to0 as T→∞T \to\infty.

(c) Consequently, for every P∈P2P \in\mathcal{P}_2 and every λ>0\lambda> 0 there is exactly one Tc(λ,P)>0T_c(\lambda,P) > 0 with λk(P,Tc)=1\lambda k(P,T_c) = 1. One has λk(P,T)>1\lambda k(P,T) > 1 for T<TcT < T_c and λk(P,T)<1\lambda k(P,T) < 1 for T>TcT > T_c, so for KAY-admissible PP the normal state is linearly unstable exactly when T<TcT < T_c. The map λ↦Tc(λ,P)\lambda\mapsto T_c(\lambda,P) is a continuous, strictly increasing bijection of (0,∞)(0,\infty) onto itself.

(d) For Einstein phonons of frequency Ω\Omega, Tc(λ,Ω)=Ω/(2πh−1(1/λ))T_c(\lambda,\Omega) = \Omega/(2\pi h^{-1}(1/\lambda)).

For every truncation N≥1N \ge1, the maps ϖ↦h(N)(ϖ)\varpi\mapsto h^{(N)}(\varpi) and T↦k(N)(P,T)T \mapsto k^{(N)}(P,T) are likewise continuous and, respectively, strictly increasing and strictly decreasing.

Theorem 5 (The sharp bound; Conjecture 1 of Part III and Conjecture 2 of Part II)

(a) h(ϖ)<gϖ2h(\varpi) < g\varpi^2 for every ϖ>0\varpi> 0, and k(P,T)<g⟨ω2⟩/(2πT)2k(P,T) < g\langle\omega^2\rangle/(2\pi T)^2 for every P∈P2P \in\mathcal{P}_2 and T>0T > 0.

(b) Consequently, for every λ>0\lambda> 0,

Tc(λ,Ω)<Ω2πgλ,Tc(λ,P)<12πgλ⟨ω2⟩=0.182726…λ⟨ω2⟩.T_c(\lambda,\Omega) < \frac{\Omega}{2\pi}\sqrt{g\lambda}, \qquad T_c(\lambda,P) < \frac{1}{2\pi}\sqrt{g\lambda\langle\omega^2\rangle} = 0.182726\ldots\sqrt{\lambda\langle\omega^2\rangle}.

(c) The constant is sharp: h(ϖ)/(gϖ2)→1h(\varpi)/(g\varpi^2) \to1 as ϖ→0\varpi\to0, hence Tc(λ,Ω)/(Ω2πgλ)→1T_c(\lambda,\Omega)/(\frac{\Omega}{2\pi}\sqrt{g\lambda}) \to1 as λ→∞\lambda\to\infty. The same limit holds for every P∈P2P \in\mathcal{P}_2 with ∫ω4P(dω)<∞\int\omega^4P(d\omega) < \infty.

Theorem 6 (Einstein spectra are optimal). Let P∈P2P \in\mathcal{P}_2 and Ω2:=⟨ω2⟩1/2\Omega_2 := \langle\omega^2 \rangle^{1/2}. Then k(P,T)≤h(Ω2/2πT)k(P,T) \le h(\Omega_2/2\pi T) for every T>0T > 0, with strict inequality unless P=δΩ2P = \delta_{\Omega_2}. Hence Tc(λ,P)<Tc(λ,Ω2)T_c(\lambda,P) < T_c(\lambda,\Omega_2) for every λ>0\lambda> 0 unless P=δΩ2P = \delta_{\Omega_2}.

Theorems 4 and 5 together prove Conjecture 1 of [12] and Conjecture 2 of [11], which KAY stated for their admissible class; we obtain them, as does [15], on the larger class P2\mathcal{P}_2. Theorem 6 says that at fixed λ\lambda and ⟨ω2⟩\langle\omega^2 \rangle the Einstein spectrum at frequency Ω2=⟨ω2⟩1/2\Omega_2 = \langle\omega^2 \rangle^{1/2} has the largest TcT_c. Figure 2 illustrates the Einstein case.

Figure 2. Floating-point illustration for Einstein phonons, with truncation N≥800N \ge800. The curves are not used in the proofs. Left: h(ϖ)/(gϖ2)h(\varpi)/(g\varpi^2), which Theorem 5 proves is below 1 and tends to 1 as ϖ→0\varpi\to0 (strong coupling). Right: Tc/ΩT_c/\Omega against λ\lambda, with the ceiling gλ/2π\sqrt{g\lambda}/2\pi.

Remark 7 (Adding spectral weight). Theorem 3 also gives the global form of the result of Bergmann and Rainer [8]. Write ν(dω)=λP(dω)=2α2F(ω)ω−1dω\nu(d\omega) = \lambda P(d\omega) = 2\alpha^2F(\omega)\omega^{-1}d\omega, so that λaj(P,T)=∫ω2(ω2+(2πTj)2)−1ν(dω)\lambda_{aj}(P,T) = \int\omega^2(\omega^2 + (2\pi Tj)^2)^{-1}\nu(d\omega) and, since K[a]K[a] is linear in aa, λk(P,T)=κ[λa(P,T)]\lambda_k(P,T) = \kappa[\lambda_a(P,T)]. Let ν′=ν+η\nu' = \nu+\eta, where η≠0\eta\ne0 is a finite measure on (0,∞)(0,\infty) with ∫ω2η(dω)<∞\int\omega^2\eta(d\omega) < \infty, and let λ′\lambda' and P′P' be the coupling and normalized spectrum of ν′\nu'. At every T>0T > 0 each coefficient increases strictly, by at most ∫ω2η(dω)/(2πTj)2\int\omega^2\eta(d\omega)/(2\pi Tj)^2, so Theorem 3 gives λk(P′,T)>λk(P,T)\lambda_k(P',T) > \lambda_k(P,T), and Theorem 4(b) gives Tc(λ′,P′)>Tc(λ,P)T_c(\lambda',P') > T_c(\lambda,P). Adding phonon spectral weight at any frequencies strictly raises TcT_c.

Theorem 8 (Sharp moment bounds). For 0≤n≤10 \le n \le1 let

Mn:=∫0∞ωnα2F(ω) dω=λ2∫ωn+1P(dω),Hn:=sup⁡ϖ>0h(ϖ)ϖn+1.M_n := \int_0^\infty\omega^n\alpha^2F(\omega)\,d\omega= \frac{\lambda}{2}\int\omega^{n+1}P(d\omega), \qquad H_n := \sup_{\varpi>0}\frac{h(\varpi)}{\varpi^{n+1}}.

Then Hn≤max⁡{g,3π/2}H_n \le\max\{g,3\pi/2\} and H1=gH_1 = g. For every P∈P2P \in\mathcal{P}_2, λ>0\lambda> 0 and T>0T > 0,

λk(P,T)≤2HnMn(2πT)n+1,henceTc(λ,P)≤12π(2HnMn)1/(n+1).\lambda k(P,T) \le\frac{2H_nM_n}{(2\pi T)^{n+1}}, \qquad\text{hence}\qquad T_c(\lambda,P) \le\frac{1}{2\pi}(2H_nM_n)^{1/(n+1)}.

The constant is sharp: the right-hand side is the supremum of TcT_c over Einstein spectra with the same MnM_n, the coupling λ\lambda being free.

For n=1n = 1, 2M1=λ⟨ω2⟩2M_1 = \lambda\langle\omega^2\rangle and Theorem 8 is the bound of Theorem 5, without strict inequality. For n=0n = 0 it is Leavens’s zeroth-moment bound [9 eq. (13)]. A floating-point evaluation of hh on truncations with N=400N = 400 and N=800N = 800, which agree to 10−1010^{-10}, gives H0≈0.72529H_0 \approx0.72529, attained at ϖ0≈1.2056\varpi_0 \approx1.2056, so that

Tc≤H0π∫0∞α2F(ω) dω≈0.23087∫0∞α2F(ω) dω,T_c \le\frac{H_0}{\pi}\int_0^\infty\alpha^2F(\omega)\,d\omega\approx0.23087\int_0^\infty\alpha^2F(\omega)\,d\omega,

with equality, to the accuracy of these values, for the Einstein spectrum of weight M0M_0 at frequency Ω=2πTcϖ0≈1.749M0\Omega= 2\pi T_c\varpi_0 \approx1.749M_0, that is λ=2M0/Ω≈1.144\lambda= 2M_0/\Omega\approx1.144, close to Leavens’s numerical values 0.2309, 1.750 and 1.143 [9 Sect. 2]. These numerical values of H0H_0 and ϖ0\varpi_0 are not certified; the inequality holds with H0H_0 defined as the supremum. Unlike the first-moment bound, which is approached only as λ→∞\lambda\to\infty, the zeroth-moment bound is, numerically, attained at moderate coupling, where Migdal–Eliashberg theory is usually applied. For n>1n > 1 no such bound exists, since h(ϖ)/(gϖ2)→1h(\varpi)/(g\varpi^2) \to1 as ϖ→0\varpi\to0 forces Hn=∞H_n = \infty, as Leavens observed [6 eq. (17)]; for −1<n<0-1 < n < 0 finiteness of HnH_n would need a growth bound on hh that we do not prove.

Corollary 9 (Coulomb pseudopotential with a symmetric cutoff) Let P∈P2P \in\mathcal{P}_2, λ>0\lambda> 0, μ∗≥0\mu^* \ge0 and ωc>0\omega_c > 0. Replace λ(ωn−ωm)\lambda(\omega_n-\omega_m) in (3) by λ(ωn−ωm)−μ∗1{∣ωn∣<ωc}1{∣ωm∣<ωc}\lambda(\omega_n-\omega_m)-\mu^*1_{\{|\omega_n|<\omega_c\}}1_{\{|\omega_m|<\omega_c\}} and leave (2) unchanged. The cutoff is placed symmetrically on both frequencies, so that the Coulomb correction is self-adjoint; as in the usual pseudopotential treatment, the Coulomb term does not enter the mass renormalization [2, 4]. If at some T>0T>0 the resulting equations have a non-zero even solution with ξ∈ℓ2\xi\in\ell^2, then

T≤Tc(λ,P)<12πgλ⟨ω2⟩,T \le T_c(\lambda,P) < \frac{1}{2\pi}\sqrt{g\lambda\langle\omega^2\rangle},

and TT obeys the bounds of Theorem 8. The same holds for the truncation to the first NN positive frequencies, and for any repulsive term whose matrix in the variables ξ\xi is positive semidefinite.

Proof. For even solutions the Coulomb term adds −2μ∗1{ωn<ωc}∑m≥0, ωm<ωcym-2\mu^*1_{\{\omega_n<\omega_c\}}\sum_{m\ge0,\ \omega_m<\omega_c}y_m to the right-hand side of the equation for ωnyn\omega_n y_n in Section 2, so (6) becomes ξ=(λK[a]−2μ∗vvT)ξ\xi=(\lambda K[a]-2\mu^*vv^{\mathsf{T}})\xi with vn=1{ωn<ωc}/ωnv_n=1_{\{\omega_n<\omega_c\}}/\sqrt{\omega_n}, a vector with finitely many non-zero entries. The inner product with ξ\xi gives λ⟨ξ,K[a]ξ⟩=∥ξ∥2+2μ∗⟨v,ξ⟩2≥∥ξ∥2\lambda\langle\xi,K[a]\xi\rangle=\|\xi\|^2+2\mu^*\langle v,\xi\rangle^2\ge\|\xi\|^2, hence λk(P,T)≥1\lambda k(P,T)\ge1. By Theorem 4(c) this excludes T>Tc(λ,P)T>T_c(\lambda,P), and Theorem 5(b) bounds Tc(λ,P)T_c(\lambda,P). For the truncation the same computation gives λk(N)(P,T)≥1\lambda k^{(N)}(P,T)\ge1, and k(N)≤kk^{(N)}\le k by Lemma 2(d). For a general positive semidefinite repulsion the inner product argument is unchanged. □

Corollary 9 makes rigorous, in this model, Leavens’s remark that the bounds for μ∗=0\mu^*=0 also apply for μ∗>0\mu^*>0 [6, 9 Sect. 3]. It says nothing about uniqueness of the instability temperature when μ∗>0\mu^*>0: the Coulomb term is negative and frequency independent, and the optimizer need not lie in the cone CC.

The last result concerns the quadratic forms behind KAY’s finite-dimensional analysis. For N≥1N\ge1, a sequence aa and θ∈RN\theta\in\mathbb{R}^N put

GN[a](θ):=∑n,m<Nθn[(1−δnm)a∣n−m∣+an+m+1]θm−∑n<N2Anθn2,G_N[a](\theta):=\sum_{n,m<N}\theta_n\left[(1-\delta_{nm})a_{\lvert n-m\rvert}+a_{n+m+1}\right]\theta_m-\sum_{n<N}2A_n\theta_n^2,
DN(θ):=∑n<Nwnθn2.D_N(\theta):=\sum_{n<N}w_n\theta_n^2.

With ξn=wn θn\xi_n=\sqrt{w_n}\,\theta_n these are ⟨ξ,K(N)[a]ξ⟩\langle\xi,K^{(N)}[a]\xi\rangle and ∥ξ∥2\lVert\xi\rVert^2, so GN/DNG_N/D_N is the Rayleigh quotient of K(N)[a]K^{(N)}[a] written in the variables θ=y\theta=y, which are proportional to KAY’s linearized spin angles (§ 2). For aj=j−γa_j=j^{-\gamma} they are the numerator and denominator of [11 l.h.s. of (70)]. KAY proved the quotient positive on non-zero, non-negative, non-increasing θ\theta, showed that constant sequences are local minimizers, and conjectured that they are global minimizers [11 Conjecture 3].

Theorem 10 (Conjecture 3 of Part II) Let N≥1N\ge1 and let aa be positive and strictly decreasing, for instance aj=j−γa_j=j^{-\gamma} with γ>0\gamma>0. Put

cN[a]:=1N2∑n,m<Nan+m+1=1N2∑k=12N−1min⁡{k,2N−k}ak.c_N[a]:=\frac{1}{N^2}\sum_{n,m<N}a_{n+m+1}=\frac{1}{N^2}\sum_{k=1}^{2N-1}\min\{k,2N-k\}a_k.

Then for every non-zero, non-negative, non-increasing θ∈RN\theta\in\mathbb{R}^N,

GN[a](θ)≥cN[a]DN(θ),G_N[a](\theta)\ge c_N[a]D_N(\theta),

with equality if and only if θ\theta is constant.

Since the optimizer of K(N)[a]K^{(N)}[a] lies in the same cone (Lemma 12), Theorem 10 says that on this cone the Rayleigh quotient ranges between cN[a]c_N[a], attained by constants, and κ(N)[a]\kappa^{(N)}[a], attained by the optimizer.

Proof of the comparison principle

For ξ∈ℓ2(N0)\xi\in\ell^2(\mathbb{N}_0) write yn:=ξn/wny_n:=\xi_n/\sqrt{w_n}, the gap divided by the frequency, up to normalization (§ 2). Let

C:={ξ∈ℓ2(N0):ξn≥0 for all n, and y is non-increasing}\mathcal{C}:=\{\xi\in\ell^2(\mathbb{N}_0):\xi_n\ge0\text{ for all }n,\text{ and }y\text{ is non-increasing}\}

and let CN⊂RN\mathcal{C}_N\subset\mathbb{R}^N be its finite-dimensional analogue. In physical terms, ξ∈C\xi\in\mathcal{C} means Δn≥0\Delta_n\ge0 with Δn/ωn\Delta_n/\omega_n non-increasing.

Each transfer operator is non-negative on the cone

Lemma 11 If ξ∈C\xi\in\mathcal{C}, then for every j≥1j\ge1

μj(ξ):=⟨ξ,Mjξ⟩=2∑n≥0yn+j(yn−yn+j)+∑n+m+1=jynym≥0,(9)\mu_j(\xi):=\langle\xi,M_j\xi\rangle=2\sum_{n\ge0}y_{n+j}(y_n-y_{n+j})+\sum_{n+m+1=j}y_ny_m\ge0, \tag*{(9)}

and μ1(ξ)≥y02\mu_1(\xi)\ge y_0^2. The same holds for ξ∈CN\xi\in\mathcal{C}_N and the truncated forms μj(N)(ξ):=⟨ξ,K(N)[ej]ξ⟩\mu_j^{(N)}(\xi):=\langle\xi,K^{(N)}[e_j]\xi\rangle, with the sums restricted to indices below NN.

Proof From (8), ⟨ξ,Tjξ⟩=2∑nynyn+j\langle\xi,T_j\xi\rangle=2\sum_n y_ny_{n+j}, ⟨ξ,Djξ⟩=2∑n≥jyn2=2∑nyn+j2\langle\xi,D_j\xi\rangle=2\sum_{n\ge j}y_n^2=2\sum_n y_{n+j}^2 and ⟨ξ,Hjξ⟩=∑n+m+1=jynym\langle\xi,H_j\xi\rangle=\sum_{n+m+1=j}y_ny_m. All series converge absolutely because ∑nyn2≤∥ξ∥2\sum_n y_n^2\le\lVert\xi\rVert^2. Subtracting gives (9). On C\mathcal{C} every summand is non-negative, and for j=1j=1 the Hankel sum is y02y_0^2. In the truncated case the Toeplitz and diagonal sums run over n+j<Nn+j<N, and the same pairing applies. □

In (9), the pair-breaking term −2yn+j2-2y_{n+j}^{2} is dominated by the pairing term 2ynyn+j2y_ny_{n+j} because yn≥yn+jy_n \ge y_{n+j}. For aj=j−γa_j=j^{-\gamma} and finite NN, the sum ∑jajμj(N)\sum_j a_j\mu_j^{(N)} is KAY’s expansion [11 eqs. (72)–(73)]; the same identity is [15 eqs. (9)–(10)].

The optimizer lies in the cone

Lemma 12. Let aa be a kernel. Then K[a]K[a] has a unit eigenvector ξ∈C\xi\in\mathcal{C} for κ[a]\kappa[a], and y0>0y_0 > 0. The same holds for every truncation K(N)[a]K^{(N)}[a], with ξ∈CN\xi\in\mathcal{C}_N an eigenvector for κ(N)[a]\kappa^{(N)}[a] whose entries are all positive.

For finite NN this is [11 Prop. 8]. We give a self-contained proof that avoids the Perron–Frobenius theorem; it is the proof formalized in Lean (§ 7).

Proof. Finite NN. Write K=K(N)[a]K=K^{(N)}[a] and κ=κ(N)[a]\kappa=\kappa^{(N)}[a]; the case N=1N=1 is trivial, so let N≥2N\ge2. The unit sphere is compact, so the quadratic form ⟨x,Kx⟩\langle x,Kx\rangle attains its maximum κ\kappa on it. Since all off-diagonal entries of KK are positive, ⟨∣x∣,K∣x∣⟩≥⟨x,Kx⟩\langle|x|,K|x|\rangle\ge\langle x,Kx\rangle, so there is a maximizer z≥0z\ge0. A maximizer is an eigenvector: the function t↦⟨z+tu,K(z+tu)⟩−κ∥z+tu∥2t\mapsto\langle z+tu,K(z+tu)\rangle-\kappa\|z+tu\|^2 is non-positive and vanishes at t=0t=0, so its derivative there, 2⟨u,Kz−κz⟩2\langle u,Kz-\kappa z\rangle, vanishes for every uu. If zn=0z_n=0 for some nn, the nn-th row of Kz=κzKz=\kappa z reads ∑m≠nKnmzm=0\sum_{m\ne n}K_{nm}z_m=0, which forces z=0z=0; hence zn>0z_n>0 for all nn.

Put yn:=zn/wn>0y_n:=z_n/\sqrt{w_n}>0, a0:=a1a_0:=a_1 and Cnm:=a∣n−m∣+an+m+1C_{nm}:=a_{|n-m|}+a_{n+m+1} for n,m<Nn,m<N. Multiplying the nn-th row of Kz=κzKz=\kappa z by wn\sqrt{w_n} and adding (a0+2An)yn(a_0+2A_n)y_n to both sides gives

(Cy)n=dnyn,dn:=κwn+a0+2An>0,(Cy)_n=d_ny_n,\qquad d_n:=\kappa w_n+a_0+2A_n>0,

and dnd_n is increasing in nn. So the matrix Q:=diag⁡(d)−1CQ:=\operatorname{diag}(d)^{-1}C has positive entries and satisfies Qy=yQy=y.

QQ maps the cone CN′\mathcal{C}'_N of non-negative non-increasing vectors into itself. The cone consists of non-negative combinations of the prefix indicators ulu_l, 1≤l≤N1\le l\le N, where ulu_l is the indicator of {0,…,l−1}\{0,\ldots,l-1\}, and

(Cul)n=∑i=n−l+1n+la∣i∣=:Fl(n),Fl(n+1)−Fl(n)=an+l+1−a∣n−l+1∣≤0,(Cu_l)_n=\sum_{i=n-l+1}^{n+l}a_{|i|}=:F_l(n),\qquad F_l(n+1)-F_l(n)=a_{n+l+1}-a_{|n-l+1|}\le0,

because n+l+1>∣n−l+1∣n+l+1>|n-l+1| and aa is non-increasing on N0\mathbb{N}_0 (recall a0=a1a_0=a_1). So CCN′⊂CN′C\mathcal{C}'_N\subset\mathcal{C}'_N, and multiplication by the positive non-increasing sequence 1/d1/d preserves CN′\mathcal{C}'_N.

Now let α(v):=min⁡nvn/yn\alpha(v):=\min_n v_n/y_n and maximize α\alpha over the compact set S:={v∈CN′:v≤y}S:=\{v\in\mathcal{C}'_N:v\le y\}. The set SS is mapped into itself by QQ, since Qv≤Qy=yQv\le Qy=y. The maximum α∗\alpha^* is positive, since a small constant vector lies in SS. Let v∗∈Sv^*\in S attain it, so v∗≥α∗yv^*\ge\alpha^*y. If v∗≠α∗yv^*\ne\alpha^*y, then Q(v∗−α∗y)Q(v^*-\alpha^*y) has all entries positive, so α(Qv∗)>α∗\alpha(Qv^*)>\alpha^*, a contradiction. Hence y=v∗/α∗y=v^*/\alpha^* is non-increasing, that is, z∈CNz\in\mathcal{C}_N.

N=∞N=\infty. By Lemma 2(d), κ(N)→κ\kappa^{(N)}\to\kappa. Let ξN∈CN\xi^N\in\mathcal{C}_N be the unit eigenvectors above, padded by zeros, and choose a subsequence with ξN⇀ξ\xi^N\rightharpoonup\xi weakly. Since KK is compact, ⟨ξN,KξN⟩→⟨ξ,Kξ⟩\langle\xi^N,K\xi^N\rangle\to\langle\xi,K\xi\rangle, and ⟨ξN,K(N)ξN⟩=κ(N)→κ\langle\xi^N,K^{(N)}\xi^N\rangle=\kappa^{(N)}\to\kappa; so ⟨ξ,Kξ⟩=κ>0\langle\xi,K\xi\rangle=\kappa>0. Thus ξ≠0\xi\ne0, and as ∥ξ∥≤1\|\xi\|\le1 its Rayleigh quotient is at least κ\kappa, which forces ∥ξ∥=1\|\xi\|=1 and Kξ=κξK\xi=\kappa\xi. Weak convergence implies convergence of each coordinate, which preserves ξn≥0\xi_n\ge0 and the monotonicity of yy. Finally y0>0y_0>0, since a non-zero non-negative non-increasing sequence has a positive first entry. □\square

Proof of Theorem 3

Proof. Take ξ\xi from Lemma 12 for the kernel aa. By (8) and Lemma 11,

κ[b]≥⟨ξ,K[b]ξ⟩=κ[a]+∑j≥1(bj−aj)μj(ξ)≥κ[a]+(b1−a1)y02>κ[a].\kappa[b] \ge\langle\xi, K[b]\xi\rangle= \kappa[a] + \sum_{j\ge1}(b_j-a_j)\mu_j(\xi) \ge\kappa[a] + (b_1-a_1)y_0^2 > \kappa[a].

The rearrangement is justified by ∑j(bj−aj)∥Mj∥<3∑j(bj−aj)<∞\sum_j(b_j-a_j)\lVert M_j\rVert< 3\sum_j(b_j-a_j) < \infty. For a truncation the same computation applies, with μj(N)\mu_j^{(N)} and the eigenvector of K(N)[a]K^{(N)}[a]; there only j≤2N−1j\le2N-1 occur.

Remark 13 Where hh is differentiable, the same computation gives ∂h/∂(ϖ2)=∑jj2(ϖ2+j2)−2μj(ξ)>0\partial h/\partial(\varpi^2)=\sum_j j^2(\varpi^2+j^2)^{-2}\mu_j(\xi)>0. KAY grouped this derivative by the weights j2/(ϖ2+j2)2j^2/(\varpi^2+j^2)^2, which requires them to decrease in jj; that holds if and only if ϖ≤2\varpi\le\sqrt{2} [11 Lemma 3]. Grouping by the transfer jj needs no condition on the weights.

Proofs of uniqueness, the sharp bound and Einstein optimality

We use one elementary estimate twice.

Lemma 14 (Zero-temperature divergence) Let N≥1N\ge1, 0≤ε≤10\le\varepsilon\le1, and let aa be a summable real sequence with 1−ε≤aj≤11-\varepsilon\le a_j\le1 for 1≤j≤2N−11\le j\le2N-1. Then κ[a]≥κ(N)[a]≥2SN−1−ε(2SN+1)\kappa[a]\ge\kappa^{(N)}[a]\ge2S_N-1-\varepsilon(2S_N+1), where SN:=∑n<N1/wnS_N:=\sum_{n<N}1/w_n.

Proof. Test K(N)[a]K^{(N)}[a] with ξn=1/wn\xi_n=1/\sqrt{w_n}, n<Nn<N, that is yn=1/wny_n=1/w_n. The reference matrix K(N)[1]K^{(N)}[1] is defined directly by the finite formula (7); its Rayleigh quotient at this vector is

2(∑nyn)2−∑nyn2−∑n2nyn2∑nwnyn2=2SN2−SNSN=2SN−1.\frac{2\left(\sum_n y_n\right)^2-\sum_n y_n^2-\sum_n 2n y_n^2}{\sum_n w_n y_n^2}=\frac{2S_N^2-S_N}{S_N}=2S_N-1.

Write aj=1−εja_j=1-\varepsilon_j with 0≤εj≤ε0\le\varepsilon_j\le\varepsilon. The off-diagonal entries of K(N)[a]−K(N)[1]K^{(N)}[a]-K^{(N)}[1] are −(ε∣n−m∣+εn+m+1)/wnwm-(\varepsilon_{\lvert n-m\rvert}+\varepsilon_{n+m+1})/\sqrt{w_nw_m}, of absolute value at most 2ε/wnwm2\varepsilon/\sqrt{w_nw_m}; the diagonal entries are (2∑j≤nεj−ε2n+1)/wn(2\sum_{j\le n}\varepsilon_j-\varepsilon_{2n+1})/w_n, of absolute value at most ε\varepsilon. At the test vector the quadratic form of the difference is therefore at most 2εSN2+εSN2\varepsilon S_N^2+\varepsilon S_N in absolute value, and dividing by ∥ξ∥2=SN\lVert\xi\rVert^2=S_N gives the claim. The first inequality is Lemma 2(d).

Since SN→∞S_N\to\infty, the right-hand side can be made as large as we like.

Proof of Theorem 4 Monotonicity. If ϖ<ϖ′\varpi<\varpi', then ϖ2/(ϖ2+j2)<ϖ′2/(ϖ′2+j2)\varpi^2/(\varpi^2+j^2)<\varpi'^2/(\varpi'^2+j^2) for every jj, the differences are summable, and the Einstein kernels are kernels; Theorem 3 gives h(ϖ)<h(ϖ′)h(\varpi)<h(\varpi'). If T′<TT'<T, then aj(P,T′)>aj(P,T)a_j(P,T')>a_j(P,T) for every jj because PP gives no mass to {0}\{0\}, and ∑jaj(P,T′)<∞\sum_j a_j(P,T')<\infty; so k(P,T′)>k(P,T)k(P,T')>k(P,T).

Continuity. By Lemma 2(a) and Weyl’s inequality, ∣κ[a]−κ[b]∣≤3∑j∣aj−bj∣\lvert\kappa[a]-\kappa[b]\rvert\le3\sum_j\lvert a_j-b_j\rvert. For T,T′≥T0>0T,T'\ge T_0>0, each ∣aj(P,T)−aj(P,T′)∣\lvert a_j(P,T)-a_j(P,T')\rvert tends to 00 as T′→TT'\to T by dominated convergence, and is at most 2⟨ω⟩/(2πT0j)22\langle\omega\rangle/(2\pi T_0j)^2, which is summable in jj. Hence T↦k(P,T)T\mapsto k(P,T) is continuous. The Einstein case is the special case P=δΩP=\delta_\Omega.

Limits. As T→∞T \to\infty, Theorem 5(a), whose proof uses only Theorem 3, gives 0<k(P,T)<g(ω2)/(2πT)2→00 < k(P,T) < g(\omega^{2})/(2\pi T)^{2} \to0; likewise h(ϖ)<gϖ2→0h(\varpi) < g\varpi^{2} \to0 as ϖ→0\varpi\to0. As T→0T \to0, aj(P,T)→1a_j(P,T) \to1 for every jj by dominated convergence, so Lemma 14 gives lim inf⁡T→0k(P,T)≥2SN−1\liminf_{T\to0} k(P,T) \ge2S_N-1 for every NN, and SN→∞S_N \to\infty. The same argument gives h(ϖ)→∞h(\varpi) \to\infty as ϖ→∞\varpi\to\infty.

Conclusion. Thus T↦k(P,T)T \mapsto k(P,T) is a decreasing homeomorphism of (0,∞)(0,\infty) onto itself, and Tc(λ,P)T_c(\lambda,P) is the unique solution of k(P,T)=1/λk(P,T)=1/\lambda; it increases continuously and strictly with λ\lambda. For KAY-admissible PP the stability statement is [11 Thm. 1+]. For Einstein phonons, k(δΩ,T)=h(Ω/2πT)k(\delta_{\Omega},T)=h(\Omega/2\pi T) gives Theorem 4(d).

Truncations. The same arguments, with the truncated version of Theorem 3 and the bound ∥K(N)[a]−K(N)[b]∥≤3∑j∣aj−bj∣\lVert K^{(N)}[a]-K^{(N)}[b]\rVert\le3\sum_j |a_j-b_j| (Lemma 2(a) for the N×NN\times N blocks), prove the monotonicity and continuity of h(N)h^{(N)} and k(N)k^{(N)}. Their limits are finite: k(N)(P,T)→κ(N)([1]j)k^{(N)}(P,T)\to\kappa^{(N)}([1]_j) as T→0T\to0 (compare [11 Thm. 5]). □

Proof of Theorem 5 (a) For ϖ>0\varpi>0 the Einstein kernel satisfies ϖ2/(ϖ2+j2)<ϖ2/j2\varpi^2/(\varpi^2+j^2)<\varpi^2/j^2 for every jj, and Theorem 3 with a=(ϖ2/(ϖ2+j2))ja=(\varpi^2/(\varpi^2+j^2))_j and b=(ϖ2j−2)jb=(\varpi^2j^{-2})_j gives h(ϖ)<κ[(ϖ2(j−2))]=ϖ2gh(\varpi)<\kappa[(\varpi^2(j^{-2}))]=\varpi^2g by Lemma 2(b). For general P∈P2P\in\mathcal{P}_2,

aj(P,T)=∫ω2ω2+(2πTj)2 P(dω)<⟨ω2⟩(2πT)2j−2a_j(P,T)=\int\frac{\omega^2}{\omega^2+(2\pi Tj)^2}\,P(d\omega)<\frac{\langle\omega^2\rangle}{(2\pi T)^2}j^{-2}

for every jj, and the same argument gives k(P,T)<g⟨ω2⟩/(2πT)2k(P,T)<g\langle\omega^2\rangle/(2\pi T)^2.

(b) At T=TcT=T_c we have 1=λk(P,Tc)<λg⟨ω2⟩/(2πTc)21=\lambda k(P,T_c)<\lambda g\langle\omega^2\rangle/(2\pi T_c)^2, which rearranges to the claim; the Einstein case is P=δΩP=\delta_{\Omega}.

(c) Let bj:=⟨ω2⟩(2πT)−2j−2b_j:=\langle\omega^2\rangle(2\pi T)^{-2}j^{-2}. Then

0<bj−aj(P,T)=∫ω4(2πTj)2(ω2+(2πTj)2) P(dω)≤⟨ω4⟩(2πT)4j4,0<b_j-a_j(P,T)=\int\frac{\omega^4}{(2\pi Tj)^2\bigl(\omega^2+(2\pi Tj)^2\bigr)}\,P(d\omega)\le\frac{\langle\omega^4\rangle}{(2\pi T)^4j^4},

so by Lemma 2(a), ∣k(P,T)−g⟨ω2⟩/(2πT)2∣≤3ζ(4)⟨ω4⟩/(2πT)4|k(P,T)-g\langle\omega^2\rangle/(2\pi T)^2|\le3\zeta(4)\langle\omega^4\rangle/(2\pi T)^4. Dividing by g⟨ω2⟩/(2πT)2g\langle\omega^2\rangle/(2\pi T)^2 shows that the ratio tends to 11 as T→∞T\to\infty. As λ→∞\lambda\to\infty, Tc(λ,P)→∞T_c(\lambda,P)\to\infty by Theorem 4, and λk(P,Tc)=1\lambda k(P,T_c)=1 then gives (2πTc)2/(gλ⟨ω2⟩)→1(2\pi T_c)^2/(g\lambda\langle\omega^2\rangle)\to1. For Einstein phonons ⟨ω4⟩=Ω4\langle\omega^4\rangle=\Omega^4. □

Proof of Theorem 6 For fixed TT and jj the function fj(s)=s/(s+(2πTj)2)f_j(s)=s/(s+(2\pi Tj)^2) is strictly concave on s≥0s\ge0. By Jensen’s inequality,

aj(P,T)=∫fj(ω2) P(dω)≤fj(⟨ω2⟩)=Ω22Ω22+(2πTj)2,a_j(P,T)=\int f_j(\omega^2)\,P(d\omega)\le f_j(\langle\omega^2\rangle)=\frac{\Omega_2^2}{\Omega_2^2+(2\pi Tj)^2},

which is the Einstein kernel at frequency Ω2\Omega_2, i.e. at ϖ=Ω2/(2πT)\varpi=\Omega_2/(2\pi T). Equality holds for some (equivalently every) jj only if ω2\omega^2 is PP-almost surely constant, i.e. P=δΩ2P=\delta_{\Omega_2}. Otherwise the inequality is strict for every jj, and Theorem 3 gives k(P,T)<h(Ω2/2πT)k(P,T)<h(\Omega_2/2\pi T). At T=Tc(λ,P)T=T_c(\lambda,P) this gives λh(Ω2/2πT)>1\lambda h(\Omega_2/2\pi T)>1, so by Theorem 4 the Einstein system at frequency Ω2\Omega_2 is still below its critical temperature: Tc(λ,P)<Tc(λ,Ω2)T_c(\lambda,P)<T_c(\lambda,\Omega_2). □

Lemma 15 (Convexity in the spectrum) For every P∈P2P\in\mathcal{P}_2 and T>0T>0, k(P,T)≤∫h(ω/(2πT)) P(dω)k(P,T)\le\int h(\omega/(2\pi T))\,P(d\omega).

Proof. Write e(ϖ):=(ϖ2/(ϖ2+j2))j≥1e(\varpi):=(\varpi^2/(\varpi^2+j^2))_{j\ge1} for the Einstein kernel, so that aj(P,T)=∫ej(ω/(2πT)) P(dω)a_j(P,T)=\int e_j(\omega/(2\pi T))\,P(\mathrm{d}\omega) and h(ϖ)=κ[e(ϖ)]h(\varpi)=\kappa[e(\varpi)]. The function hh is continuous (Theorem 4; the continuity part of its proof uses only Lemma 2(a)) and satisfies 0<h(ϖ)≤3∑j≥1ej(ϖ)≤3πϖ/20<h(\varpi)\le3\sum_{j\ge1}e_j(\varpi)\le3\pi\varpi/2 by Lemma 2(a) with b=0b=0 and ∑j≥1ϖ2/(ϖ2+j2)≤∫0∞ϖ2/(ϖ2+x2) dx\sum_{j\ge1}\varpi^2/(\varpi^2+j^2)\le\int_0^\infty\varpi^2/(\varpi^2+x^2)\,\mathrm{d}x; since P∈P2P\in\mathcal{P}_2 has a finite first moment, the integral is finite. Let ξ\xi have finitely many non-zero entries. Then ⟨ξ,K[a]ξ⟩\langle\xi,K[a]\xi\rangle is a finite linear combination of the aja_j, so

⟨ξ,K[a(P,T)]ξ⟩=∫⟨ξ,K[e(ω/2πT)]ξ⟩ P(dω)≤∥ξ∥2∫h(ω/(2πT)) P(dω).\langle\xi,K[a(P,T)]\xi\rangle=\int\langle\xi,K[e(\omega/2\pi T)]\xi\rangle\,P(\mathrm{d}\omega)\le\|\xi\|^2\int h(\omega/(2\pi T))\,P(\mathrm{d}\omega).

Such vectors are dense in ℓ2\ell^2 and K[a(P,T)]K[a(P,T)] is bounded, so taking the supremum of the Rayleigh quotient over them gives the claim. □

Proof of Theorem 8. For ϖ≤1\varpi\le1, h(ϖ)/ϖn+1<gϖ1−n≤gh(\varpi)/\varpi^{n+1}<g\varpi^{1-n}\le g by Theorem 5(a), and for ϖ≥1\varpi\ge1, h(ϖ)/ϖn+1≤(3π/2)ϖ−n≤3π/2h(\varpi)/\varpi^{n+1}\le(3\pi/2)\varpi^{-n}\le3\pi/2 by the estimate in the proof of Lemma 15. Theorem 5(a) and (c) give H1=gH_1=g. By Lemma 15 and h(ϖ)≤Hnϖn+1h(\varpi)\le H_n\varpi^{n+1},

λκ(P,T)≤λHn∫(ω2πT)n+1P(dω)=2HnMn(2πT)n+1.\lambda\kappa(P,T)\le\lambda H_n\int\left(\frac{\omega}{2\pi T}\right)^{n+1}P(\mathrm{d}\omega)=\frac{2H_nM_n}{(2\pi T)^{n+1}}.

At T=Tc(λ,P)T=T_c(\lambda,P) the left-hand side equals 11 by Theorem 4, which gives the bound on TcT_c. For sharpness, fix Mn>0M_n>0 and ϖ>0\varpi>0, put Tϖ:=12π(2Mnh(ϖ)/ϖn+1)1/(n+1)T_\varpi:=\frac{1}{2\pi}(2M_nh(\varpi)/\varpi^{n+1})^{1/(n+1)} and Ω:=2πTϖϖ\Omega:=2\pi T_\varpi\varpi, and take the Einstein spectrum α2F=MnΩ−nδΩ\alpha^2F=M_n\Omega^{-n}\delta_\Omega, whose nn-th moment is MnM_n and whose coupling is λ=2MnΩ−n−1\lambda=2M_n\Omega^{-n-1}. Then λh(Ω/2πTϖ)=2Mnh(ϖ)/(2πTϖϖ)n+1=1\lambda h(\Omega/2\pi T_\varpi)=2M_nh(\varpi)/(2\pi T_\varpi\varpi)^{n+1}=1, so Tϖ=Tc(λ,Ω)T_\varpi=T_c(\lambda,\Omega) by Theorem 4(d). The supremum of TϖT_\varpi over ϖ>0\varpi>0 is 12π(2HnMn)1/(n+1)\frac{1}{2\pi}(2H_nM_n)^{1/(n+1)}. □

Proof of Theorem 10

The proof decomposes a non-increasing vector into the prefix indicators ulu_l of § 4 and has three steps: the form on a single prefix, the cross terms between nested prefixes, and the strict decrease of cl[a]c_l[a] in ll.

Proof. Write cl:=cl[a]c_l:=c_l[a]. For 1≤l≤N1\le l\le N let ul∈RNu_l\in\mathbb{R}^N be the indicator of {0,…,l−1}\{0,\ldots,l-1\}, and let B\mathcal{B} be the symmetric bilinear form of GN[a]−cNDN\mathcal{G}_N[a]-c_N\mathcal{D}_N.

Prefixes. On a single prefix the diagonal subtraction cancels the Toeplitz part exactly, because ∑n≠m<la∣n−m∣=2∑n<lAn\sum_{n\ne m<l}a_{\lvert n-m\rvert}=2\sum_{n<l}A_n. Hence GN[a](ul)=∑n,m<lan+m+1=l2cl\mathcal{G}_N[a](u_l)=\sum_{n,m<l}a_{n+m+1}=l^2c_l, and DN(ul)=∑n<lwn=l2\mathcal{D}_N(u_l)=\sum_{n<l}w_n=l^2. So B(ul,ul)=l2(cl−cN)\mathcal{B}(u_l,u_l)=l^2(c_l-c_N).

Cross terms. For l≤l′l\le l', the vector ul′−ulu_{l'}-u_l is supported on {l,…,l′−1}\{l,\ldots,l'-1\}, disjoint from the support of ulu_l. All matrix entries of GN\mathcal{G}_N that couple the two supports are off-diagonal and positive, and DN\mathcal{D}_N is diagonal. Therefore B(ul,ul′)≥B(ul,ul)=l2(cl−cN)\mathcal{B}(u_l,u_{l'})\ge\mathcal{B}(u_l,u_l)=l^2(c_l-c_N).

clc_l is strictly decreasing. Let U,VU,V be independent and uniform on [0,1)[0,1). Then ⌊lU⌋\lfloor lU\rfloor is uniform on {0,…,l−1}\{0,\ldots,l-1\}, so cl=E a⌊lU⌋+⌊lV⌋+1c_l=\mathbb{E}\,a_{\lfloor lU\rfloor+\lfloor lV\rfloor+1}. For l<l′l<l', ⌊lU⌋≤⌊l′U⌋\lfloor lU\rfloor\le\lfloor l'U\rfloor pointwise, with strict inequality when U>1−1/l′U>1-1/l'. Since aa is strictly decreasing, cl>cl′c_l>c_{l'}.

Conclusion. Every non-negative non-increasing θ\theta is θ=∑l=1Nxlul\theta=\sum_{l=1}^N x_lu_l with xl=θl−1−θl≥0x_l=\theta_{l-1}-\theta_l\ge0 (θN:=0\theta_N:=0). Then

B(θ,θ)=∑l,l′xlxl′B(ul,ul′)≥∑l,l′xlxl′min⁡{l,l′}2(cmin⁡{l,l′}−cN)≥0.\mathcal{B}(\theta,\theta)=\sum_{l,l'}x_lx_{l'}\mathcal{B}(u_l,u_{l'})\ge\sum_{l,l'}x_lx_{l'}\min\{l,l'\}^2(c_{\min\{l,l'\}}-c_N)\ge0.

The diagonal terms with l<Nl<N are xl2l2(cl−cN)x_l^2l^2(c_l-c_N), which are positive unless xl=0x_l=0. So equality forces xl=0x_l=0 for all l<Nl<N, i.e. θ\theta constant. Conversely, a constant θ=xNuN\theta=x_Nu_N gives B(θ,θ)=xN2B(uN,uN)=0\mathcal{B}(\theta,\theta)=x_N^2\mathcal{B}(u_N,u_N)=0. □

Machine-checked and computer-assisted verification

Lean formalization

The finite-dimensional core of the argument is formalized in Lean 4 (v4.35.0) with Mathlib [16, 17]. For N≥1N \ge1 and an arbitrary real sequence aa, the development defines the matrix K(N)[a]K^{(N)}[a] of (7) on Fin NN entry by entry, its quadratic form, and κ(N)[a]\kappa^{(N)}[a] as the supremum of the quadratic form over the unit sphere. A kernel (IsKernel) is a sequence that is positive and non-increasing for j≥1j \ge1; summability plays no role at finite NN. The cone InCone is CNC_N. Discrete phonon spectra are finite families of frequencies ωi>0\omega_i > 0 with non-negative weights summing to one. Table 1 lists the formalized theorems. Each depends only on Lean’s three standard axioms (propext, Classical.choice, Quot.sound), as reported by #print axioms; the development contains no sorry, no added axioms and no native decide.

Lean nameStatement
exists_max, qf_le_kappaκ(N)[a]\kappa^{(N)}[a] is attained on the unit sphere
exists_cone_maximizerLemma 12 for finite NN: a maximizer lies in CNC_N
qf_sub, mu_nonneg, mu_one_gedecomposition (8) and Lemma 11
comparisonTheorem 3 for κ(N)\kappa^{(N)}
kappa_posκ(N)[a]≥a1\kappa^{(N)}[a] \ge a_1 for a kernel aa, Lemma 2(c)
kappa_smulκ(N)[sa]=sκ(N)[a]\kappa^{(N)}[sa] = s\kappa^{(N)}[a] for s>0s > 0
einstein_lth(ϖ)<ϖ2κ(N)[(j−2)]h(\varpi) < \varpi^2\kappa^{(N)}[(j^{-2})] for ϖ>0\varpi> 0, Theorem 5(a)
einstein_strictMonoOnϖ↦h(N)(ϖ)\varpi\mapsto h^{(N)}(\varpi) is strictly increasing on ϖ>0\varpi> 0
disp_ltTheorem 5(a) for discrete spectra
disp_strictAntiOnT↦k(N)(P,T)T \mapsto k^{(N)}(P,T) is strictly decreasing on T>0T > 0, discrete PP
disp_le_einsteink(N)(P,T)≤h(N)(Ω2/2πT)k^{(N)}(P,T) \le h^{(N)}(\Omega_2/2\pi T) for T>0T > 0, discrete PP
conj3, conj3_eq_iffTheorem 10, inequality and equality case

Table 1. Lean theorems and the statements they formalize, all for the truncation to N≥1N \ge1 frequencies.

What is not formalized. (i) The passage N→∞N \to\infty: Lemma 2(d) and the case N=∞N = \infty of Lemma 12 (its first assertion); (ii) continuity and the limits in Theorem 4, and the construction and properties of TcT_c in its parts (c)–(d); (iii) the conversion of the spectral bound to a TcT_c bound in Theorem 5(b), and the sharpness assertion in its part (c); (iv) spectra that are not finite sums of point masses, and the strict form of Theorem 6; (v) the derivation of Section 2, Lemma 15, Theorem 8 and Corollary 9; (vi) KAY’s identification of the instability with λk>1\lambda_k > 1. The steps in (i)–(v) use standard facts about compact operators, dominated convergence and Jensen’s inequality and are proved in full in the text; step (vi) is the cited result of KAY [11 Thm. 1+] and is not reproved here. In particular, the formalized Einstein bound has the truncated constant κ(N)[(j−2)]≤g\kappa^{(N)}[(j^{-2})] \le g; the statement h(ϖ)<gϖ2h(\varpi) < g\varpi^2 for the full operator needs the case N=∞N = \infty of Lemma 12.

Computer-assisted proofs

Before the comparison principle was found, the Einstein inequality h(ϖ)<gϖ2h(\varpi)<g\varpi^2 was proved using ball arithmetic. A second implementation, written from the list of lemmas alone, proves it in exact rational arithmetic. Neither uses Theorem 3; together they provide independent checks of Theorem 5 for Einstein phonons. Lemma 15 does not use Theorem 3 either, and with h(ϖ)<gϖ2h(\varpi)<g\varpi^2 it gives λk(P,T)<gλ⟨ω2⟩(2πT)2\lambda_k(P,T)<\frac{g\lambda\langle\omega^2\rangle}{(2\pi T)^2} for every P∈P2P\in\mathcal{P}_2. So the certificates also prove, independently of the comparison principle, that λk(P,T)<1\lambda_k(P,T)<1 for every T≥12πgλ⟨ω2⟩T\ge\frac{1}{2\pi}\sqrt{g\lambda\langle\omega^2\rangle}: the normal state is linearly stable above the bound for every spectrum. Appendix § 10 gives the details.

Use of large language models

This research used large language models (LLMs) extensively, and we describe their role as the journal’s policy requires. The author selected the problem, directed the research throughout, decided which approaches to pursue and which to abandon, reviewed the statements and proofs of all results in full, and is responsible for the content of this paper. LLM-based coding agents (Anthropic Claude and OpenAI GPT models, used through an agent harness) assisted, under the author’s direction, with the literature search, the exploration of proof strategies and the drafting of proofs, the certificate programs and the Lean formalization, adversarial audits of the arguments by models of different families, and drafting and editing the text. No output was accepted on the authority of a model. Every new result in the paper is proved in the text and checked by the author; its finite-dimensional core is in addition machine-checked in Lean, and the Einstein bound is confirmed by the certificates of Appendix § 10. No LLM is an author of this paper.

Scope and open problems

Apart from Corollary 9, all results are theorems about the linearized, isotropic Eliashberg model with μ∗=0\mu^*=0, in the formulation of KAY [11, 12], where the critical temperature is the linear-stability boundary of the normal state. They do not address the nonlinear gap equations or the order of the transition, a Coulomb pseudopotential μ∗>0\mu^*>0 beyond the bound of Corollary 9, anisotropic or multiband pairing, vertex corrections, or the breakdown of Migdal–Eliashberg theory at large λ\lambda [19, 20]. That breakdown motivates bounds on TcT_c of a different kind [21]. The bound is a statement about this model, not a prediction for any material.

Four questions seem natural. (i) With μ∗>0\mu^*>0 the kernel acquires a negative, frequency-independent part and the cone argument does not apply as it stands. Corollary 9 bounds every instability temperature, but whether that temperature is unique, and whether a comparison principle survives, is open. (ii) The comparison principle concerns the linear problem. Whether the full nonlinear gap equations have a non-trivial solution exactly for T<TcT<T_c, so that the linear and thermodynamic transition temperatures coincide for all λ\lambda, is part of KAY’s broader programme [11]. (iii) The cone C\mathcal{C} seems to be the natural home of physical gap functions in this model. Other monotonicity questions, for instance in the γ\gamma-model with general γ\gamma [10, 14], may yield to the same decomposition by transfer. (iv) Theorem 8 leaves the value of H0H_0 uncertified, which would need an upper bound on h(ϖ)/ϖh(\varpi)/\varpi uniformly for large ϖ\varpi, and the moments −1<n<0-1<n<0 open.

Supplementary information. The supplementary archive contains the Lean project (source and build instructions), the two certificate programs with their receipts, and the scripts that produce the figures, the values quoted in Appendix A and the floating-point value of H0H_0 after Theorem 8.

Declarations

Funding. This work was carried out at, and funded by, SuperMatics Inc. It received no external funding.

Competing interests. The author is a founder of SuperMatics Inc., which develops computational methods for materials discovery. The author declares no other competing interests.

Ethics approval and consent to participate. Not applicable.

Consent for publication. Not applicable.

Data availability. No experimental data were generated. The supplementary archive contains the programs that produce the paper’s own numerical results, together with the certificate receipts.

Materials availability. Not applicable.

Code availability. The Lean formalization, the certificate programs and the figure scripts are included in the supplementary archive, which is also distributed as the ancillary files of the arXiv version of this paper.

Author contribution. D.R. selected the problem, directed the research, reviewed and checked all statements and proofs, and wrote the paper, with the assistance of large language models described in § 7.3.

Computer-assisted proofs of the Einstein inequality

We prove sup⁡spec⁡G(ϖ)<g\sup\operatorname{spec} G(\varpi)<g for every ϖ>0\varpi>0, where cj(ϖ):=1/(ϖ2+j2)c_j(\varpi):=1/(\varpi^2+j^2) and G(ϖ):=K[c(ϖ)]G(\varpi):=K[c(\varpi)]; by Lemma 2(b), h(ϖ)=ϖ2sup⁡spec⁡G(ϖ)h(\varpi)=\varpi^2\sup\operatorname{spec} G(\varpi), so this is Theorem 5(a) for Einstein phonons. Note that G0=G(0)G_0=G(0). Decompose G(ϖ)=P(ϖ)−D(ϖ)G(\varpi)=P(\varpi)-D(\varpi) with P(ϖ)=∑jcj(ϖ)(Tj+Hj)≥0P(\varpi)=\sum_j c_j(\varpi)(T_j+H_j)\geq0 entrywise and D(ϖ)=∑jcj(ϖ)Dj≥0D(\varpi)=\sum_j c_j(\varpi)D_j\geq0 diagonal, and let S(ϖ)=∑k≥1ck(ϖ)=πϖcoth⁡(πϖ)−12ϖ2S(\varpi)=\sum_{k\geq1}c_k(\varpi)=\frac{\pi\varpi\coth(\pi\varpi)-1}{2\varpi^2}, a decreasing function.

Lemma 16 (a) (Envelope.) If 0≤ϖl≤ϖ≤ϖu0\leq\varpi_l\leq\varpi\leq\varpi_u, then sup⁡spec⁡G(ϖ)≤sup⁡spec⁡U\sup\operatorname{spec} G(\varpi)\leq\sup\operatorname{spec} U with U:=P(ϖl)−D(ϖu)U:=P(\varpi_l)-D(\varpi_u).

(b) (Block.) If U=(ABBTC)U = \begin{pmatrix} A & B \\ B^{\mathsf{T}} & C \end{pmatrix} with sup⁡spec⁡A≤α\sup\operatorname{spec} A \le\alpha, sup⁡spec⁡C≤γ\sup\operatorname{spec} C \le\gamma and ∥B∥≤β\lVert B\rVert\le\beta, then

sup⁡spec⁡U≤α+γ2+(α−γ2)2+β2.\sup\operatorname{spec} U \le\frac{\alpha+\gamma}{2}+\sqrt{\left(\frac{\alpha-\gamma}{2}\right)^2+\beta^2}.

(c) (Cholesky certificate.) If LL is an exactly representable lower-triangular matrix and τ>0\tau>0 with ∥(sI−A)−τI−LLT∥F<τ\lVert(sI-A)-\tau I-LL^{\mathsf{T}}\rVert_F<\tau, then sup⁡spec⁡A<s\sup\operatorname{spec} A<s.

(d) (Tails.) On the coordinates n≥Nn\ge N, sup⁡spec⁡C≤3S(ϖl)/wN\sup\operatorname{spec} C\le3S(\varpi_l)/w_N, and

∥B∥F2≤∑n<N∑m≥N2wnwN[T4(N−n)+T4(N+n+1)],\lVert B\rVert_F^2 \le\sum_{n<N}\sum_{m\ge N}\frac{2}{w_nw_N}\left[T_4(N-n)+T_4(N+n+1)\right],
T4(m):=m−4+13m3≥∑k≥mk−4.T_4(m):=m^{-4}+\frac{1}{3m^3}\ge\sum_{k\ge m}k^{-4}.

(e) (Large ϖ\varpi.) sup⁡spec⁡G(ϖ)≤∥P(ϖ)∥≤3S(ϖ)\sup\operatorname{spec}G(\varpi)\le\lVert P(\varpi)\rVert\le3S(\varpi).

Proof. (a) ⟨x,G(ϖ)x⟩≤⟨∣x∣,P(ϖ)∣x∣⟩−⟨∣x∣,D(ϖ)∣x∣⟩≤⟨∣x∣,U∣x∣⟩\langle x,G(\varpi)x\rangle\le\langle\lvert x\rvert,P(\varpi)\lvert x\rvert\rangle-\langle\lvert x\rvert,D(\varpi)\lvert x\rvert\rangle\le\langle\lvert x\rvert,U\lvert x\rvert\rangle, because each cjc_j decreases in ϖ\varpi. (b) ⟨x,Ux⟩≤α∥x1∥2+2β∥x1∥∥x2∥+γ∥x2∥2\langle x,Ux\rangle\le\alpha\lVert x_1\rVert^2+2\beta\lVert x_1\rVert\lVert x_2\rVert+\gamma\lVert x_2\rVert^2. (c) sI−A=LLT+τI+EsI-A=LL^{\mathsf{T}}+\tau I+E with ∥E∥≤∥E∥F<τ\lVert E\rVert\le\lVert E\rVert_F<\tau. (d) CC is bounded above by the tail of P(ϖl)P(\varpi_l); the Schur test with weights wn−1/2w_n^{-1/2} bounds its norm by sup⁡n≥N∑m≥N[(1−δnm)c∣n−m∣+cn+m+1]/wm≤3S/wN\sup_{n\ge N}\sum_{m\ge N}[(1-\delta_{nm})c_{\lvert n-m\rvert}+c_{n+m+1}]/w_m\le3S/w_N. The Frobenius bound uses ck≤k−2c_k\le k^{-2} and wm≥wNw_m\ge w_N. (e) The same Schur test over all indices. □

Primary certificate (ball arithmetic).

A floating-point eigenvector of the N=256N=256 truncation of G0G_0, rounded to an exact dyadic vector and then normalized in Arb to a unit vector ψ~\widetilde{\psi}, gives the rigorous lower bound q0≥1.318140211994q_0\ge1.318140211994 on gg and the residual ρ0≤0.000015596\rho_0\le0.000015596. By min–max and Lemma 16(b)–(d) applied to G0−tψ~ψ~TG_0-t\widetilde{\psi}\widetilde{\psi}^{\mathsf{T}}, the second eigenvalue of G0G_0 is at most g2≤0.010963g_2\le0.010963 (its floating-point value is 5.3485×10−45.3485\times10^{-4}), and the Davis–Kahan theorem [22] bounds the distance of ψ~\widetilde{\psi} to the unit top eigenvector ψ\psi of G0G_0, with its sign chosen so that ⟨ψ,ψ~⟩≥0\langle\psi,\widetilde{\psi}\rangle\ge0, by δ≤2ρ0/(q0−g2)\delta\le\sqrt{2\rho_0/(q_0-g_2)}. The half-line ϖ>0\varpi>0 is split into three regions.

  • Region I, 0<ϖ≤1/40<\varpi\le1/4. Here G0−G(ϖ)=ϖ2R(ϖ)G_0-G(\varpi)=\varpi^2R(\varpi) with R(ϖ)=∑jMjj2(ϖ2+j2)R(\varpi)=\sum_j\frac{M_j}{j^2(\varpi^2+j^2)} and ∥R(ϖ)∥≤ρ≤2.96523\lVert R(\varpi)\rVert\le\rho\le2.96523. For a unit vector x=aψ+bϕx=a\psi+b\phi with ϕ⊥ψ\phi\perp\psi and ∥ϕ∥=1\lVert\phi\rVert=1, ⟨x,(g−G(ϖ))x⟩≥(g−g2)b2+ϖ2(ra2−2ρ∣a∣∣b∣−ρb2)\langle x,(g-G(\varpi))x\rangle\ge(g-g_2)b^2+\varpi^2(ra^2-2\rho\lvert a\rvert\lvert b\rvert-\rho b^2) with r=inf⁡ϖ⟨ψ,R(ϖ)ψ⟩≥1.12441r=\inf_\varpi\langle\psi,R(\varpi)\psi\rangle\ge1.12441, and the resulting positivity criterion r(q0−g2−ϖa2ρ)−ϖa2ρr(q_0-g_2-\varpi_a^2\rho)-\varpi_a^2\rho, with ϖa=1/4\varpi_a=1/4, is at least 0.71189>00.71189>0.

  • Region II, 1/4≤ϖ≤41/4\le\varpi\le4, is divided into 48 cells with dyadic endpoints, obtained by rounding a geometric grid of ratio 1.06. In each cell, Lemma 16(a)–(d) with N=160N=160 and a Cholesky certificate bound sup⁡spec⁡G(ϖ)\sup\operatorname{spec}G(\varpi) below q0q_0. The smallest relative margin is 3.96%, on [1/4,277873/1048576][1/4,277873/1048576].

  • Region III, ϖ≥4\varpi\ge4, is controlled by sup⁡spec⁡G(ϖ)≤3S(4)≤1.084348<q0\sup\operatorname{spec}G(\varpi)\le3S(4)\le1.084348<q_0.

All bounds are evaluated in Arb ball arithmetic [23] at 200 bits, through python-flint 0.9.0 [24] on FLINT 3.6.0 [25]. Floating point only proposes trial objects (the unnormalized trial vector and the Cholesky factors LL), which are then treated as exact dyadic data; the trial vector is normalized in Arb. The run takes about 2.1 s on a laptop. Table A1 lists the certified quantities.

QuantityCertified boundRole
q0q_0 (Rayleigh quotient, N=256N = 256)≥1.318140211994\ge1.318140211994lower bound on gg
ρ0\rho_0 (residual)≤0.000015596\le0.000015596eigenvector error
g2g_2 (second eigenvalue of G0G_0)≤0.010963\le0.010963spectral gap
ρ\rho (sup⁡ϖ∥R(ϖ)∥\sup_{\varpi}\lVert R(\varpi)\rVert)≤2.96523\le2.96523Region I
rr (inf⁡ϖ⟨ψ,R(ϖ)ψ⟩\inf_{\varpi}\langle\psi,R(\varpi)\psi\rangle)≥1.12441\ge1.12441Region I
Region I criterion0.71189>00.71189 > 0Region I
Region II worst margin3.96%3.96\% (48 cells)Region II
3S(4)3S(4)≤1.084348\le1.084348Region III

Table A1. Bounds from the receipt of the primary certificate.

Independent certificate (exact rationals).

A second implementation works after the congruence W1/2(⋅)W1/2W^{1/2}(\cdot)W^{1/2}, W=diag⁡(wn)W = \operatorname{diag}(w_n), in which all entries are rational, and checks every inequality in exact rational arithmetic (FLINT fmpq). It uses N=40N = 40, the regions (0,1/8](0,1/8], [1/8,5][1/8,5] (17 geometric cells of ratio 5/45/4) and [5,∞)[5,\infty), the rational lower bound g≥659/500g \ge659/500, and rational bounds on the integrals in place of coth. It certifies g−sup⁡spec⁡G(ϖ)≥0.8048ϖ2g-\sup\operatorname{spec}G(\varpi) \ge0.8048\varpi^2 on the first region, a gap of at least 0.008260.00826 below 659/500659/500 on the second and of at least 0.43530.4353 on the third, in 0.25 s.

Provenance.

The plan of the primary certificate was fixed before it was run (SHA-256 197786ca6b2fb0676ffde142bf6c7bbec1b5c97f8e43d988bfcf53bd2decbaec). Primary certificate script cd41aff1b564efe08fa23eea1849b709fe1324661d20b26d55652bcd15df4d9d, receipt 9ce805079f1f3a1abce66b757210adba40ee647f656fddf4588387664d893bfe; independent certificate script 7242083ff94a7c1f939a39ea6df55f18b4398ae220fd7b87b9c28646356f8036, receipt 4a502d96d2cc311ba5948e67ebd861b36e913650778af7ab1fc46192c86c3e03.

References

  1. [1]Eliashberg, G.M.: Interactions between electrons and lattice vibrations in a superconductor. Soviet Physics JETP 11(3), 696–702 (1960)
  2. [2]Carbotte, J.P.: Properties of boson-exchange superconductors. Reviews of Modern Physics 62(4), 1027–1157 (1990) https://doi.org/10.1103/RevModPhys.62.1027
  3. [3]Marsiglio, F.: Eliashberg theory: A short review. Annals of Physics 417, 168102 (2020) https://doi.org/10.1016/j.aop.2020.168102
  4. [4]Allen, P.B., Dynes, R.C.: Transition temperature of strong-coupled superconductors reanalyzed. Physical Review B 12(3), 905–922 (1975) https://doi.org/10.1103/PhysRevB.12.905
  5. [5]Moussa, J.E., Cohen, M.L.: Two bounds on the maximum phonon-mediated superconducting transition temperature. Physical Review B 74(9), 094520 (2006) https://doi.org/10.1103/PhysRevB.74.094520
  6. [6]Leavens, C.R.: Important parameters in strong-coupling superconductivity. Journal of Physics F: Metal Physics 7, 1911–1922 (1977) https://doi.org/10.1088/0305-4608/7/9/034
  7. [7]Semenok, D.V., Altshuler, B.L., Yuzbashyan, E.A.: Fundamental limits on the electron–phonon coupling and superconducting T_c. Advanced Materials 37(40), 2507013 (2025) https://doi.org/10.1002/adma.202507013
  8. [8]Bergmann, G., Rainer, D.: The sensitivity of the transition temperature to changes in α^2F(ω). Zeitschrift für Physik 263, 59–68 (1973) https://doi.org/10.1007/BF02351862
  9. [9]Leavens, C.R.: A least upper bound on the superconducting transition temperature. Solid State Communications 17, 1499–1504 (1975) https://doi.org/10.1016/0038-1098(75)90982-5
  10. [10]Kiessling, M.K.-H., Altshuler, B.L., Yuzbashyan, E.A.: Bounds on T_c in the Eliashberg theory of superconductivity. I: The γ-model. Journal of Statistical Physics 192(5), 69 (2025) https://doi.org/10.1007/s10955-025-03446-5
  11. [11]Kiessling, M.K.-H., Altshuler, B.L., Yuzbashyan, E.A.: Bounds on T_c in the Eliashberg theory of superconductivity. II: Dispersive phonons. Journal of Statistical Physics 192(7), 94 (2025) https://doi.org/10.1007/s10955-025-03468-z
  12. [12]Kiessling, M.K.-H., Altshuler, B.L., Yuzbashyan, E.A.: Bounds on T_c in the Eliashberg theory of superconductivity. III: Einstein phonons. Journal of Statistical Physics 192(7), 93 (2025) https://doi.org/10.1007/s10955-025-03469-y
  13. [13]Yuzbashyan, E.A., Altshuler, B.L.: Migdal–Eliashberg theory as a classical spin chain. Physical Review B 106(1), 014512 (2022) https://doi.org/10.1103/PhysRevB.106.014512
  14. [14]Elezaby, A., Abanov, A.: Superconductivity near a quantum critical point: Bounds on the transition temperature in the γ-model. Physical Review B 114(2), 024512 (2026) https://doi.org/10.1103/wpn8-8j2t
  15. [15]Kondratev, O.: A proof of Conjecture 2 of Kiessling, Altshuler and Yuzbashyan: the sharp spectral-moment bound on the linear-stability onset temperature in Eliashberg theory. Zenodo preprint, not peer reviewed. Latest version, record 22848039, 19 September 2026 (2026). https://doi.org/10.5281/zenodo.22845696
  16. [16]de Moura, L., Ullrich, S.: The Lean 4 theorem prover and programming language. In: Automated Deduction – CADE 28. Lecture Notes in Computer Science, vol. 12699, pp. 625–635. Springer, Cham (2021). https://doi.org/10.1007/978-3-030-79876-5_37
  17. [17]The mathlib Community: The Lean mathematical library. In: Proceedings of the 9th ACM SIGPLAN International Conference on Certified Programs and Proofs (CPP 2020), pp. 367–381. ACM, New York (2020). https://doi.org/10.1145/3372885.3373824
  18. [18]Allen, P.B., Mitrović, B.: Theory of superconducting T_c. In: Ehrenreich, H., Seitz, F., Turnbull, D. (eds.) Solid State Physics vol. 37, pp. 1–92. Academic Press, New York (1983). https://doi.org/10.1016/S0081-1947(08)60665-7
  19. [19]Yuzbashyan, E.A., Altshuler, B.L.: Breakdown of the Migdal–Eliashberg theory and a theory of lattice–fermionic superfluidity. Physical Review B 106(5), 054518 (2022) https://doi.org/10.1103/PhysRevB.106.054518
  20. [20]Esterlis, I., Nosarzewski, B., Huang, E.W., Moritz, B., Devereaux, T.P., Scalapino, D.J., Kivelson, S.A.: Breakdown of the Migdal–Eliashberg theory: A determinant quantum Monte Carlo study. Physical Review B 97(14), 140501 (2018) https://doi.org/10.1103/PhysRevB.97.140501
  21. [21]Esterlis, I., Kivelson, S.A., Scalapino, D.J.: A bound on the superconducting transition temperature. npj Quantum Materials 3, 59 (2018) https://doi.org/10.1038/s41535-018-0133-0
  22. [22]Davis, C., Kahan, W.M.: The rotation of eigenvectors by a perturbation. III. SIAM Journal on Numerical Analysis 7(1), 1–46 (1970) https://doi.org/10.1137/0707001
  23. [23]Johansson, F.: Arb: Efficient arbitrary-precision midpoint-radius interval arithmetic. IEEE Transactions on Computers 66(8), 1281–1292 (2017) https://doi.org/10.1109/TC.2017.2690633
  24. [24]Johansson, F., Benjamin, O., et al.: python-flint: Python bindings for FLINT, version 0.9.0. https://github.com/flintlib/python-flint (2026)
  25. [25]The FLINT team: FLINT: Fast Library for Number Theory, version 3.6.0. https://flintlib.org (2026)

Paper details

vlnc:W_m8qpeahp
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Anthropic Claude, GPT (multiple versions)

View statement
Looking for related papers…
Contents