Research · Research note · Superconductivity

Published electron-phonon data against the Eliashberg ceiling

The ceiling Tc≤ω2f(λ)T_c\le\omega_2f(\lambda) bounds the Eliashberg transition temperature by the coupling λ\lambda and the rms phonon frequency ω2\omega_2, where Ωf(λ)\Omega f(\lambda) is the TcT_c of a single mode of frequency Ω\Omega (the ceiling note). In the Alexandria deposit of zero-pressure DFT calculations, none of 23,608 isotropic-Eliashberg Tc(μ∗=0)T_c(\mu^*=0) exceeds the certified upper bound of its ceiling from reported moments (largest ratio 0.970), and no such ceiling reaches 300 K; ceilings of the selected stored spectrum do for three entries, but vary with the broadening. None of 201 reported TcT_c values is shown to exceed its ceiling, but at most 75 test the theorem. Moussa and Cohen (citing Allen and Dynes), Kondratev and Siraj stated the ceiling earlier 1, 2, 3, 4, and a public repository 5 posted the 300 K thresholds on ω2\omega_2 at λ=1,2,3\lambda=1,2,3 (Result 2). We add enclosures of ff, these tests and a three-moment supremum.

The question

Electron–phonon theory gives a superconductor's transition temperature TcT_c from its spectrum α2F(ω)≥0\alpha^2F(\omega)\ge0, the phonon density weighted by electron coupling. The question is whether reported values, from a database of computed spectra and from claims of near-room-temperature superconductivity, respect a cap set by two of its moments. Units are kelvin, kB=ℏ=1k_B=\hbar=1; ωmax⁡\omega_{\max} is the highest frequency and

λ=2∫α2Fω dω,ω22=⟨ω2⟩=2λ∫α2F ω dω,ωlog⁡=exp⁡[2λ∫α2Fln⁡ω dωω].\lambda=2\int\frac{\alpha^2F}{\omega}\,d\omega,\qquad \omega_2^2=\langle\omega^2\rangle=\frac{2}{\lambda}\int\alpha^2F\,\omega\,d\omega,\qquad \omega_{\log}=\exp\Big[\frac{2}{\lambda}\int\alpha^2F\ln\omega\,\frac{d\omega}{\omega}\Big].

An Einstein mode is one phonon frequency Ω\Omega; f(λ)=Tc/Ωf(\lambda)=T_c/\Omega for it. The theory is isotropic and one-band, with TcT_c the temperature where the normal state turns linearly unstable. The Coulomb pseudopotential μ∗\mu^* acts on Matsubara frequencies up to a cutoff, on both indices of the pairing kernel (symmetric convention) or the summed index only (standard). At μ∗=0\mu^*=0, Tc≤ω2f(λ)T_c\le\omega_2f(\lambda), with equality only for one Einstein mode 2, 3. The ceiling holds for a constant μ∗≥0\mu^*\ge0; with only ωmax⁡\omega_{\max} reported it is ωmax⁡f(λ)\omega_{\max}f(\lambda), since ω2≤ωmax⁡\omega_2\le\omega_{\max}.

What we found

Data: the Alexandria 3D deposit (zero pressure, PBEsol, snapshot 2025.08.11, 27,180 materials; CC BY 4.0; 6, 7; derived statistics) and the 806 spectra in the BETE-NET file (repository licence Apache-2.0; the paper reports 818 materials) 8. Each Alexandria entry stores ten spectra, one per broadening; we use the positive part of the one with signed λ\lambda closest to the reported λ\lambda.

Result 1 (Screen, certified computation). None of the 23,608 Alexandria entries reporting an isotropic-Eliashberg Tc(μ∗=0)T_c(\mu^*=0) exceeds the certified upper bound of ω2f(λ)\omega_2f(\lambda) from its reported λ\lambda and ω2\omega_2. The largest ratio to that bound is 0.970.

Result 2 (Thresholds, computer-assisted proof). If an isotropic Eliashberg superconductor has Tc(μ∗=0)≥300T_c(\mu^*=0)\ge300 K, then ω2≥2617.3\omega_2\ge2617.3 K at λ=1\lambda=1, ω2≥1416.8\omega_2\ge1416.8 K at λ=2\lambda=2 and ω2≥1075.3\omega_2\ge1075.3 K at λ=3\lambda=3, rounded down.

QuantityValueKind
Largest ceiling from reported moments, 27,175 entries (five report zero or no coupling)292.4 K (B2Be4N2); 290.3 K (H6KPt) without ten entries whose ten stored spectra have no positive λ\lambdaCertified computation
Entries with ceiling above 300 / 150 / 77 K from reported moments, at most0 / 55 / 1,904Certified computation
Largest ceiling, BETE-NET69.5 K (HPd)Certified computation
Entries whose certified lower bound of the ceiling is below the reported TcT_c9, all with λ\lambda from 0.24 to 0.26, by up to 2.6 %; 22 more, with λ\lambda below 0.2, have no lower boundCertified computation
Reported TcT_c above the ceiling of the selected spectrum3 (B2Be4N2, C2Be4, H8Al2Pt2: ten stored spectra with non-positive λ\lambda, 10 to 54 % negative weight); under a second selection rule 3 others (H24Ho8, H24Er8, H24Tm8: 30 to 33 % negative weight). None is above the ceiling from its reported moments, and the theorem needs a positive spectrumExhaustive computation
Largest ceiling of the selected spectra, one broadening each331.7 K (H6PtRb); above 300 K: H6PtRb, H6CsPt and H6KPt, with 5.2 to 7.7 % negative weight, each failing a 2 % test against its own reported momentsCertified computation
H6PtRb, H6CsPt, H6KPt: from reported moments286.4, 277.6, 290.3 KCertified computation
The same, over the ten stored spectra, smallest to largest broadening558 to 226, 603 to 217, 492 to 231 K (factor 2.1 to 2.8)Certified computation
Selected spectra with λ\lambda and ω2\omega_2 raised by 10 % / 30 % (not a DFT error bar)largest ceiling 388 / 510 K; entries above 300 K: 5 / 10Certified computation
Float Tc/ω2f(λ)T_c/\omega_2f(\lambda) of 10,132 selected spectra from a ceiling-selected sample (8,520 with ceiling at least 20 K, 1,612 random from 3 to 20 K), solved in float64 by a separately written solver (truncated operator, so each TcT_c is a lower bound)median 0.64, 99th percentile 0.93, maximum 0.9875 (K4O2), none above 1Measurement
Interval brackets of TcT_c, 257 spectra (Pareto front of (λ,ω2)(\lambda,\omega_2), largest ceilings, random), B2Be4N2 excludedrelative width about 4.0×10−44.0\times10^{-4}; ratio to ceiling 0.174 to 0.988Certified computation
ω2\omega_2 needed for 300 K at μ∗=0\mu^*=0, λ=3.6915\lambda=3.6915, the Einstein-phonon stability limit derived in 10, not examined here and since questioned 11944.9 KCertified computation
Ceilings are certified upper bounds interpolated between enclosures of ff, up to 0.92 % too high for λ≥1\lambda\ge1, 2.6 % for 0.5≤λ<10.5\le\lambda<1 and 9.4 % for 0.2≤λ<0.50.2\le\lambda<0.5. Below λ=0.2\lambda=0.2 the bound is one-sided and can exceed the exact ceiling by one to two orders of magnitude.

Proposition (Three moments, proved). Let λ>0\lambda>0 and 0<ωlog⁡<ω20<\omega_{\log}<\omega_2. Over spectra with no frequency floor and exactly these three moments, the supremum of Tc(μ∗=0)T_c(\mu^*=0) is ω2f(λ)\omega_2f(\lambda), and it is not attained. Reporting ωlog⁡\omega_{\log} alone does not lower the ceiling.

Result 3 (Claims, certified computation). None of 201 reported TcT_c values is shown to exceed its ceiling. Of the 121 that report ω2\omega_2 or ωmax⁡\omega_{\max}, none reaches 0.8 of ω2f(λ)\omega_2f(\lambda) or ωmax⁡f(λ)\omega_{\max}f(\lambda); the largest ratio is 0.78: 287.5 K for ScLuH12 at 190 GPa (Eliashberg, μ∗=0.10\mu^*=0.10, Fan et al. 9; ceiling 367.6 to 369.1 K).

Kind of reported valueRows (141 material-pressure pairs, 16 sources)Below 0.8 of ceilingLog levelNeeds ω2\omega_2
Eliashberg, taken as isotropic (theorem)43131218: three hold for every spectrum, nine need ω2/ωlog⁡\omega_2/\omega_{\log} of 1.016 to 1.175 (it is 1.175 to 1.841 in 9), six report no ωlog⁡\omega_{\log}
Eliashberg, ab initio Coulomb kernel (theorem if the kernel's window matrix is positive semidefinite, unchecked)100100
Calculation, method left to the primary source (theorem if Eliashberg)6600
Measured TcT_c, calculated or tabulated moments (theorem)161600
Allen–Dynes, McMillan or mean-field formula (consistency only; includes two non-refereed postings, a Zenodo record and an arXiv version since withdrawn, both passing)11585282
Anisotropic Eliashberg (outside the theory)100010
MgB₂ measured, two bands (outside the theory)1100
All2011215030
A claim is one reported TcT_c with its inputs. Log level: Tc≤ωlog⁡f(λ)≤ω2f(λ)T_c\le\omega_{\log}f(\lambda)\le\omega_2f(\lambda) by Jensen's inequality for ln⁡\ln, so the claim holds for every spectrum of the reported (λ,ωlog⁡)(\lambda,\omega_{\log}). Needs: the claim holds only if an unreported ω2≥Tc/f(λ)\omega_2\ge T_c/f(\lambda).

Why it is true

Let p(dω)=2α2F dω/(λω)p(d\omega)=2\alpha^2F\,d\omega/(\lambda\omega), a probability measure with second moment ω22\omega_2^2. At μ∗=0\mu^*=0 the largest eigenvalue of the stability operator depends on α2F\alpha^2F through κT(k)=λ∫ω2/(ω2+νk2) p(dω)\kappa_T(k)=\lambda\int\omega^2/(\omega^2+\nu_k^2)\,p(d\omega), νk=2πTk\nu_k=2\pi Tk, and by Theorem 1 of the ceiling note and its proof it increases strictly with κ\kappa on non-increasing kernels. Jensen's inequality for the concave map x↦x/(x+ν2)x\mapsto x/(x+\nu^2) bounds κT(k)\kappa_T(k) by the Einstein kernel with the same λ\lambda and ω2\omega_2, with equality only for a point mass, as in 3 (section 12).

For μ∗≥0\mu^*\ge0 Theorem 10 of the ceiling note covers the symmetric convention; in the standard one an instability needs a zero of 1+2μ∗uT(I−H)−1ρ1+2\mu^*u^{\rm T}(I-H)^{-1}\rho, where I−HI-H is the phonon-only operator at TT and u,ρ≥0u,\rho\ge0 are fixed vectors. Above the phonon-only TcT_c, I−HI-H is a positive definite Z-matrix with entrywise non-negative inverse, so the function stays positive and Tc<Tc(μ∗=0)T_c<T_c(\mu^*=0). This holds on finite truncations of the Matsubara sums, the infinite problem being their limit; Siraj 4 states it for the half-line convention, unchecked by us.

For the proposition, put weight (1−ϵ)λ(1-\epsilon)\lambda at Ωh\Omega_h and ϵλ\epsilon\lambda at ωs\omega_s. By the intermediate value theorem some ωs(ϵ)\omega_s(\epsilon) fixes ωlog⁡\omega_{\log}, with ϵln⁡(ω2/ωs)→ln⁡(ω2/ωlog⁡)\epsilon\ln(\omega_2/\omega_s)\to\ln(\omega_2/\omega_{\log}), so ωs→0\omega_s\to0 and Ωh→ω2\Omega_h\to\omega_2. The kernel dominates that of the hard atom alone, so Tc≥Ωhf((1−ϵ)λ)→ω2f(λ)T_c\ge\Omega_hf((1-\epsilon)\lambda)\to\omega_2f(\lambda). Equality needs one atom, excluded by ωlog⁡<ω2\omega_{\log}<\omega_2.

How it was checked

ff rests on 321 enclosures (643 certificates, one for C∞=lim⁡f/λC_\infty=\lim f/\sqrt\lambda). A certificate puts the largest eigenvalue of the linearised gap operator above 1 at TloT_{\rm lo} or below 1 at ThiT_{\rm hi}, bracketing TcT_c, and a standalone program checks it in interval arithmetic. Separately written code re-ran the checker on all 643, put ff inside every enclosure (float solver) and rebuilt the table from re-downloaded sources, reproducing all 201 classifications.

Controls: one Einstein mode (λ=2\lambda=2, Ω=1000\Omega=1000 K) gives ω2f(λ)\omega_2f(\lambda) to 5×10−85\times10^{-8}, and float64 searches over 3- and 5-atom spectra reach it within rounding (1+3×10−131+3\times10^{-13}). The enclosures rest on one interval method with no second interval-arithmetic implementation. The proofs have not been peer reviewed.

What it does not show

  • Scope: the isotropic one-band model above, a constant μ∗≥0\mu^*\ge0 and a positive spectrum. Nothing follows for vertex corrections, the signed α2F\alpha^2F that DFT integrated, anisotropic or multiband gaps, or odd-frequency or time-reversal-breaking transitions.
  • Fan et al. 9 print a column headed ω2\omega^2 in kelvin, captioned mean square frequency. We read it as the rms frequency, which reproduces all 13 printed Allen–Dynes–McMillan values to 0.13 % with their equations (arXiv version 1).
  • A consistent claim is not a correct one, and Result 2 is necessary only. With a frequency floor the supremum of the proposition may be lower and is not determined here.

Earlier and parallel work

WhoDateWhat they state
Moussa and Cohen 2, citing Allen and Dynes 12006; 1975The Einstein spectrum maximises TcT_c at fixed λ\lambda and ⟨ω2⟩\langle\omega^2\rangle (1 not read by us); Tc≤0.183λ⟨ω2⟩T_c\le0.183\sqrt{\lambda\langle\omega^2\rangle}
Leavens 12, as described in 41977A reduction to Einstein spectra at fixed first moment, and Tc≤0.258(∫α2F ω dω)1/2T_c\le0.258(\int\alpha^2F\,\omega\,d\omega)^{1/2} at μ∗=0\mu^*=0 (not read by us)
Kiessling, Altshuler and Yuzbashyan 13, 142024, revised 2025For λ>0.77\lambda>0.77 an Einstein mode has Tc=Ωf(λ)T_c=\Omega f(\lambda), with lower bounds on ff that converge to it and a loose upper bound; Jensen's inequality for the first lag only (II, Corollary 2, as described in 4)
Kondratev 3, not refereed; its disclosure states that AI systems produced the proof19 September 2026Scalar μ∗=0\mu^*=0 model: a unique onset temperature, kBT<C∗λ⟨ω2⟩k_BT<C_*\sqrt{\lambda\langle\omega^2\rangle} with C∗<0.182727C_*<0.182727, the same extremality, and a 300 K threshold λ⟨ω2⟩>0.141478\sqrt{\lambda\langle\omega^2\rangle}>0.141478 eV, with no value of ff
Siraj 4, not refereed; its record states it is AI-generated; partly machine-checked in Lean 4, not checked by us13 and 14 November 2026Tc≤ω2f(λ)T_c\le\omega_2f(\lambda) for every μ∗≥0\mu^*\ge0, and f(3)/3≤0.16112f(3)/\sqrt3\le0.16112, f(10)/10≤0.176438f(10)/\sqrt{10}\le0.176438; no database, claims or ωlog⁡\omega_{\log}
Public repository 5, one commit, not refereed, certificates unchecked by us1 November 2026Result 2 as λω22≥g(λ)\lambda\omega_2^2\ge g(\lambda), g=λΩE2g=\lambda\Omega_E^2 with ΩE\Omega_E the Einstein frequency giving 300 K, certified on 31 couplings from 0.8 to 100 and infinity; at λ=1,2,3\lambda=1,2,3 it gives our 2617.3, 1416.8 and 1075.3 K
Semenok et al. 102024, 2025Tc<0.18λ⟨ω2⟩T_c<0.18\sqrt{\lambda\langle\omega^2\rangle}; from a stability limit λ≤3.69\lambda\le3.69, Tc<0.32 ωmax⁡T_c<0.32\,\omega_{\max} by the Einstein mode at ωmax⁡\omega_{\max}; 23 materials in Table S1 and 44 rows of λ\lambda and ωlog⁡\omega_{\log} in Table S3
Gao et al. 152025Over 20,000 calculations from the same database: room-temperature conventional superconductivity at ambient pressure is extremely unlikely; no proved ceiling in either version
Sadovskii 162021Five LaH10 and YH10 calculations against a lower bound derived from the linearised gap equation and the Allen–Dynes asymptote 0.18λ Ω0.18\sqrt\lambda\,\Omega, with ωlog⁡=ω2\omega_{\log}=\omega_2
Sharma 17, not refereed30 August 2026The spread of TcT_c among spectra sharing (λ,ωlog⁡,ω2)(\lambda,\omega_{\log},\omega_2), from 806 spectra; no statement of the three-moment supremum
This note16 November 2026Two-sided enclosures of ff at 321 couplings, 0.2 to 10410^4; the tests of Results 1 and 3; the three-moment supremum

We found no earlier application of a proof-backed ceiling to a database or to reported values, or earlier statement of the three-moment supremum, in the sources we searched (arXiv, Zenodo, web search, public GitHub pages; last searched 15 November 2026).

References

  1. 1.P. B. Allen and R. C. Dynes, "Transition temperature of strong-coupled superconductors reanalyzed", Phys. Rev. B 12, 905 (1975). doi.org/10.1103/PhysRevB.12.905
  2. 2.J. E. Moussa and M. L. Cohen, "Two bounds on the maximum phonon-mediated superconducting transition temperature", Phys. Rev. B 74, 094520 (2006). doi.org/10.1103/PhysRevB.74.094520
  3. 3.O. Kondratev, "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, v3 (2026). doi.org/10.5281/zenodo.22848039
  4. 4.A. Siraj, "Sharp T_c bounds and uniqueness of the superconducting state in isotropic Migdal-Eliashberg theory: papers and Lean 4 formalisation (MELean)", Zenodo, v1.0.2 (2026). doi.org/10.5281/zenodo.24234156
  5. 5.PoplerAdmin, "rtsc-inverse-design", GitHub repository, commit 5facda5 (2026), read 15 November 2026. github.com/PoplerAdmin/rtsc-inverse-design
  6. 6.Th. Cavignac et al., "AI-driven expansion and application of the Alexandria database", J. Phys. Mater. 9, 025014 (2026). doi.org/10.1088/2515-7639/ae6620
  7. 7.T. F. T. Cerqueira, Y.-W. Fang, I. Errea, A. Sanna and M. A. L. Marques, "Searching materials space for hydride superconductors at ambient pressure", Adv. Funct. Mater. 34, 2404043 (2024). doi.org/10.1002/adfm.202404043
  8. 8.J. B. Gibson, A. C. Hire, P. M. Dee, O. Barrera, B. Geisler, P. J. Hirschfeld and R. G. Hennig, "Accelerating superconductor discovery through tempered deep learning of the electron-phonon spectral function", npj Comput. Mater. 11, 7 (2025). doi.org/10.1038/s41524-024-01475-4
  9. 9.Y. Fan, B. Li, C. Zhu, J. Cheng, S. Liu and Z. Shi, "Superconductive sodalite-like clathrate hydrides MXH12 with critical temperatures of near 300 K under pressures", arXiv:2408.00234v1 (2024); Phys. Status Solidi B 262, 2400240. arxiv.org/abs/2408.00234
  10. 10.D. V. Semenok, B. L. Altshuler and E. A. Yuzbashyan, "Fundamental limits on the electron-phonon coupling and superconducting T_c", arXiv:2407.12922 (2024); Adv. Mater. 37, 2507013 (2025). arxiv.org/abs/2407.12922
  11. 11.N. Bashan and E. Berg, "Stability and relaxation of electron-phonon systems at strong coupling", arXiv:2611.07014 (2026). arxiv.org/abs/2611.07014
  12. 12.C. R. Leavens, "Important parameters in strong-coupling superconductivity", J. Phys. F 7, 1911 (1977). doi.org/10.1088/0305-4608/7/9/034
  13. 13.M. K.-H. Kiessling, B. L. Altshuler and E. A. Yuzbashyan, "Bounds on T_c in the Eliashberg theory of superconductivity. II: Dispersive phonons", J. Stat. Phys. 192, 94 (2025). arxiv.org/abs/2409.00532
  14. 14.M. K.-H. Kiessling, B. L. Altshuler and E. A. Yuzbashyan, "Bounds on T_c in the Eliashberg theory of superconductivity. III: Einstein phonons", J. Stat. Phys. 192, 93 (2025). arxiv.org/abs/2409.02121
  15. 15.K. Gao, T. F. T. Cerqueira, A. Sanna, Y.-W. Fang, Đ. Dangić, I. Errea, H.-C. Wang, S. Botti and M. A. L. Marques, "The maximum T_c of conventional superconductors at ambient pressure", Nat. Commun. 16, 8253 (2025). doi.org/10.1038/s41467-025-63702-w
  16. 16.M. V. Sadovskii, "Limits of Eliashberg theory and bounds for superconducting transition temperature", arXiv:2106.09948 (2021); Phys. Usp. 65, 724 (2022). arxiv.org/abs/2106.09948
  17. 17.A. Sharma, "Decomposing Allen–Dynes error into learnable and irreducible components", Zenodo preprint (2026). doi.org/10.5281/zenodo.22177654