Fast Real-Axis Eliashberg Calculations: Full-bandwidth solutions beyond the constant density of states approximation

This paper introduces an efficient, linear-scaling numerical method for solving finite-temperature Migdal-Eliashberg equations directly on the real-frequency axis with full-bandwidth electronic structure and particle-hole asymmetry, thereby eliminating the need for unstable analytic continuation and yielding more accurate superconducting properties that align closely with experimental observations.

The Gist

Imagine you are trying to understand how a superconductor works—a material that conducts electricity with zero resistance. To do this, physicists use a complex mathematical framework called Migdal-Eliashberg theory. Think of this theory as the "recipe book" for predicting how electrons dance together to form superconducting pairs.

However, for decades, there was a major problem with how scientists used this recipe book.

The Old Way: The Foggy Mirror

Traditionally, scientists didn't look at the recipe directly. Instead, they looked at it through a foggy, distorted mirror (the "imaginary axis").

• The Problem: To see the actual result (the "real" behavior of the material), they had to take that blurry reflection and try to "un-fog" it using a mathematical trick called analytic continuation.
• The Analogy: Imagine trying to reconstruct a high-definition photo of a face by looking at a very low-resolution, grainy sketch of it and then guessing the details. Sometimes the guess is okay, but often the "un-fogging" process introduces errors, blurs fine details, or creates weird artifacts that aren't real. It's a delicate, error-prone step that gets even harder when the material is very cold.

The New Way: The Direct Line of Sight

This paper introduces a new, faster, and clearer way to solve these equations. The authors, a team from MIT and other institutions, have built a direct line of sight to the real answer.

Here is how they did it, using some everyday analogies:

1. The "Constant vs. Variable" Terrain

In the old simplified models, scientists assumed the "electronic landscape" (where electrons live) was flat and uniform, like a perfectly smooth, endless parking lot. This is called the "Constant Density of States" (cDOS) approximation.

• The Reality: Real materials are more like mountainous terrain. They have deep valleys, sharp peaks, and cliffs. In the material they studied (H₃S, a superconductor made of hydrogen and sulfur), there is a specific "cliff" near the energy level where electrons live, called a van-Hove singularity.
• The Mistake: The old "flat parking lot" model ignored these mountains. It was like trying to drive a car on a mountain road while assuming the road is flat; you'd miss the sharp turns and crashes.
• The Fix: The new method accounts for the full, bumpy terrain. It calculates how the electrons interact with the actual shape of the material's energy landscape.

2. The Speed Boost: From O(N²) to O(N)

The math involved in these calculations is heavy. Usually, to get a clear picture, you need to sample millions of points.

• The Old Bottleneck: If you wanted to double the number of points to get a sharper image, the old methods would take four times as long to compute (like trying to count every grain of sand on a beach by hand, then doing it again for every new grain you add).
• The New Trick: The authors developed a clever numerical shortcut. They realized that instead of calculating every single point individually, they could reuse parts of the calculation.
• The Analogy: Imagine you are painting a mural. The old way required you to mix a fresh bucket of paint for every single brushstroke. The new way is like having a smart paint sprayer that mixes the paint once and applies it efficiently across the whole wall.
• The Result: Their method scales linearly. If you double the points, it only takes twice as long. This makes it fast enough to run on a standard laptop in minutes, rather than taking days on a supercomputer.

Why Does This Matter? (The H₃S Example)

The team tested their new method on H₃S, a material that becomes superconductive at very high pressures.

• The "Cliff" Effect: Because H₃S has that sharp "cliff" (van-Hove singularity) in its energy landscape, the electrons behave very differently on one side of the cliff compared to the other. This is called particle-hole asymmetry.
• The Result:
  • The old "flat parking lot" model predicted a superconducting gap (the energy needed to break the electron pairs) of 75 meV.
  • The new "mountain terrain" model predicted 60 meV.
  • The Truth: Experiments show the gap is actually 60 meV.
  • Conclusion: The new method got it right because it respected the "mountainous" shape of the material, while the old method missed the details.

The Bigger Picture

This isn't just about getting a slightly better number for one material.

1. No More Guessing: Scientists can now see the "real" behavior of materials without the blurry "foggy mirror" step.
2. Speed: Because it's so fast, they can now simulate how superconductors react to rapid changes (like being hit by a laser pulse in a pump-probe experiment). This opens the door to understanding how these materials behave in real-time, non-equilibrium situations.
3. Future Tech: This helps in designing better superconducting devices, sensors, and potentially even quantum computers, by giving engineers a crystal-clear map of how electrons will behave in their designs.

In summary: The authors built a faster, more accurate GPS for superconductors. Instead of driving blindfolded and guessing the route (the old method), they now have a high-definition, real-time map that accounts for every bump and turn in the road, allowing them to predict the journey perfectly.

Technical Summary

1. Problem Statement

The paper addresses two fundamental limitations in the theoretical study of conventional superconductors using Migdal-Eliashberg theory:

• The Analytic Continuation Bottleneck: Experimentally relevant quantities (spectral functions, optical conductivity, transport properties) are defined on the real-frequency axis. However, standard calculations are performed on the imaginary-frequency (Matsubara) axis due to numerical stability. Converting these results to the real axis via analytic continuation is an ill-conditioned inverse problem. This step amplifies numerical errors, obscures fine spectral features, and becomes increasingly unstable at low temperatures.
• The Constant Density of States (cDOS) Approximation: Existing efficient real-axis solvers often assume a constant electronic density of states (cDOS) near the Fermi level. This approximation discards full-bandwidth effects, specifically particle-hole asymmetry and band-structure features (such as van-Hove singularities), which are critical in real materials like high-temperature superconducting hydrides.

2. Methodology

The authors propose a novel framework to solve the finite-temperature Migdal-Eliashberg equations directly on the real-frequency axis while retaining the full energy dependence of the electronic density of states (vDOS) and the screened Coulomb interaction.

Key Numerical Innovations:

• Linear-Scaling Kernel Computation: The most computationally expensive part of real-axis Eliashberg equations is evaluating the integral kernel $K(\omega, \omega')$. Previous methods scaled quadratically ($O(N^2)$) with the number of frequency grid points. The authors decompose the real part of the kernel into functions of the difference $(\omega - \omega')$. By precomputing these difference-dependent integrals, they reduce the computational complexity to linear scaling ($O(N)$), making high-resolution grids feasible.
• Semi-Analytic Spectral Integration: Integrals involving the spectral function and a variable density of states $N(\varepsilon)$ are challenging due to sharp peaks near Green's function poles. The authors employ a piecewise-linear interpolation of $N(\varepsilon)$ (and $\phi(\varepsilon, \omega)$ in the full case). Since the spectral factors have known analytic antiderivatives, the integrals over each sub-interval can be solved analytically. This yields high precision without the instability of purely numerical quadrature.
• Fixed-Point Iteration with Causality Enforcement: The equations are solved via fixed-point iteration. To ensure stability, the authors enforce causality constraints (ensuring the imaginary part of Green's function poles has the correct sign) and use the converged cDOS solution as an initial guess for the more complex vDOS calculation.

3. Key Contributions

• Direct Real-Axis Solver: A practical, efficient algorithm to solve Migdal-Eliashberg equations directly on the real axis, bypassing the need for analytic continuation entirely.
• Full-Bandwidth Treatment: The first efficient implementation that incorporates a variable electronic density of states (vDOS) and energy-dependent screened Coulomb interactions ($W(\varepsilon, \varepsilon')$) in real-axis calculations.
• Computational Efficiency: The method achieves runtimes of milliseconds for cDOS cases and minutes for full vDOS cases on standard hardware, a significant improvement over previous real-axis methods that were computationally prohibitive.
• Time-Dependent Capability: The linear scaling and direct real-axis access enable the simulation of nonequilibrium dynamics (e.g., pump-probe experiments), which are difficult to model using imaginary-axis formulations.

4. Results: Application to H$_3$S

The method was applied to the high-pressure superconductor H$_3$S (at 200 GPa), a material known for a prominent van-Hove singularity (vHS) near the Fermi level, which induces strong particle-hole asymmetry.

• Superconducting Gap:
  • The vDOS solution yielded a superconducting gap of $\approx 60$ meV, matching experimental tunneling spectroscopy data.
  • The traditional cDOS approximation overestimated the gap at 75 meV.
  • This demonstrates that neglecting the vHS and particle-hole asymmetry leads to quantitatively incorrect predictions.
• Spectral Features:
  • The direct real-axis solutions revealed fine structures in the spectral functions ($A(\varepsilon, \omega)$) and quasiparticle density of states ($N(\omega)$) corresponding to the phonon spectrum ($\alpha^2F(\omega)$), which were obscured or distorted in analytically continued imaginary-axis solutions.
  • The vDOS solution correctly captured strong particle-hole asymmetry in the spectral weight and quasiparticle occupancy, features absent in the cDOS model.
• Comparison with Analytic Continuation:
  • When compared to imaginary-axis solutions continued via the Padé approximation, the direct real-axis method showed superior numerical stability. The analytic continuation exhibited unphysical large gradients and peaks, particularly in the renormalization function $Z(\omega)$ at low temperatures ($T=1$ K).

5. Significance

This work bridges the gap between first-principles theory and experimental observables in superconductivity.

• Accuracy: It establishes that full electronic structure details (band structure, vHS) are not just qualitative corrections but are essential for quantitative agreement with experiments in strong-coupling superconductors.
• Accessibility: By removing the analytic continuation bottleneck and reducing computational cost, it makes high-fidelity Eliashberg calculations accessible for routine materials screening.
• Future Directions: The framework opens the door to simulating ultrafast nonequilibrium phenomena (e.g., light-induced superconductivity) and calculating transport coefficients directly from microscopic theory, moving beyond phenomenological models.

In summary, the paper provides a robust, fast, and accurate numerical tool that resolves long-standing issues in the theoretical description of superconductors, specifically enabling the study of materials where band-structure details dictate macroscopic properties.