Full-Bandwidth Anisotropic Migdal-Eliashberg Equations

• The theory is a full-bandwidth approach that retains complete momentum, energy, and band dispersion without reducing to the Fermi surface.
• It captures key phenomena like spectral weight transfer, replica bands, and particle-hole asymmetry through an expanded self-consistency framework.
• Advanced numerical strategies, including non-uniform Matsubara grids and IR basis compression, enable accurate Tc predictions in systems such as FeSe/SrTiO₃ and superhydrides.

Full-bandwidth anisotropic Migdal-Eliashberg equations are the band-, momentum-, and frequency-resolved superconducting self-consistency equations obtained when Migdal-Eliashberg theory is formulated without projecting the electronic problem onto the Fermi surface. In this form, the theory retains the explicit dispersion $\varepsilon_{n\mathbf{k}}$ or $\xi_{n\mathbf{k}}$, the normal and anomalous Nambu self-energy channels, and the contribution of electronic states away from $E_F$. The resulting framework differs qualitatively from the standard Fermi-surface-restricted anisotropic formalism used in many EPW and Wannier-interpolated applications: it can capture particle-hole asymmetry, chemical-potential renormalization, spectral-weight transfer, replica bands, and pairing contributions from states far below or above the Fermi level. Explicit full-bandwidth realizations have been developed for FeSe/SrTiO$_3$, for model non-adiabatic vertex-corrected problems, and more recently within EPW and direct real-axis solvers (Aperis et al., 2017, Schrodi et al., 2019, Lucrezi et al., 2023, Simon et al., 18 Mar 2026).

1. Definition and conceptual lineage

The general anisotropic Eliashberg formalism is naturally written in Nambu-Gor'kov language, with self-energy components $Z(\mathbf{k},i\omega_n)$, $\chi(\mathbf{k},i\omega_n)$, and $\phi(\mathbf{k},i\omega_n)$, and with the gap defined by $\Delta=\phi/Z$. In the foundational Wannier-based anisotropic framework, these equations are written before Fermi-surface reduction as momentum- and frequency-dependent equations that still contain $\epsilon_{\mathbf{k}}$ and $\chi$; in the practical implementation of that work, however, they are then reduced to the standard Fermi-surface form with $\xi_{n\mathbf{k}}$0 and $\xi_{n\mathbf{k}}$1 (Margine et al., 2012).

The defining step toward a genuine full-bandwidth anisotropic formulation is the decision not to make that reduction. In the FeSe/SrTiO$\xi_{n\mathbf{k}}$2 study, superconductivity is formulated on the full ten-band FeSe tight-binding dispersion, over the full Brillouin zone, with full Matsubara and real-frequency dependence and fixed filling; in the EPW full-bandwidth implementation, the same distinction is expressed as retaining scattering processes beyond the Fermi surface, the energy-shift self-energy $\xi_{n\mathbf{k}}$3, and a self-consistent chemical potential $\xi_{n\mathbf{k}}$4 (Aperis et al., 2017, Lucrezi et al., 2023).

The term “full bandwidth” is therefore formal rather than merely numerical. It denotes a theory in which the superconducting problem is posed over the retained electronic structure itself, rather than over a Fermi-surface shell supplemented by a constant-density-of-states approximation. This point is central to later applications in superhydrides and to real-axis treatments that retain an energy-dependent DOS $\xi_{n\mathbf{k}}$5 and particle-hole asymmetry (Lucrezi et al., 2023, Simon et al., 18 Mar 2026).

2. Nambu-space structure and representative equation sets

In full-bandwidth anisotropic Migdal-Eliashberg theory, the inverse Nambu Green’s function is decomposed as

$\xi_{n\mathbf{k}}$6

with

$\xi_{n\mathbf{k}}$7

The corresponding denominator is

$\xi_{n\mathbf{k}}$8

This structure appears explicitly in the EPW full-bandwidth implementation and, in equivalent notation, in the FeSe/SrTiO$\xi_{n\mathbf{k}}$9 full-bandwidth equations and in the beyond-Migdal model treatment (Aperis et al., 2017, Lucrezi et al., 2023, Schrodi et al., 2019).

A representative full-bandwidth anisotropic Matsubara system is

$E_F$0

$E_F$1

$E_F$2

The full-bandwidth formalism is completed by an electron-number equation for $E_F$3, which becomes necessary once states away from $E_F$4 are renormalized (Lucrezi et al., 2023).

In FeSe/SrTiO$E_F$5, the same logic is written in terms of the full momentum dependence over the ten-band dispersion: $E_F$6

$E_F$7

$E_F$8

with

$E_F$9

The formal distinction from Fermi-surface equations is visible immediately in the retained $_3$0 and $_3$1 dependence (Aperis et al., 2017).

3. Full bandwidth versus Fermi-surface restriction

In the literature, the phrase “fully anisotropic” is frequently used for a theory that is fully resolved in band index and momentum on the Fermi surface, but not full-bandwidth. This is the case in the Wannier-EPW anisotropic formalism and in representative applications to heavily $_3$2-doped graphene, MoTe$_3$3, and hexagonal HfRuAs, where the equations explicitly contain $_3$4 or $_3$5, and only states at or near the Fermi level enter the pairing problem (Margine et al., 2014, Paudyal et al., 2021, Reddy et al., 18 May 2026, Margine et al., 2012).

This distinction has methodological consequences. In the Fermi-surface-restricted equations, the denominator is reduced to a Fermi-surface quantity such as

$_3$6

or, in EPW notation,

$_3$7

The energy-shift channel $_3$8 is dropped, the DOS is treated as effectively constant in the pairing window, and the electronic phase space is projected by $_3$9. By contrast, full-bandwidth theories retain

$Z(\mathbf{k},i\omega_n)$0

or its equivalent, and therefore allow off-shell states to contribute directly (Lucrezi et al., 2023, Aperis et al., 2017).

A recurrent misconception is that a large near-$Z(\mathbf{k},i\omega_n)$1 energy window is equivalent to full bandwidth. The papers surveyed here distinguish these cases sharply. The HfRuAs study uses a Fermi-window width of 0.5 eV around $Z(\mathbf{k},i\omega_n)$2, but still formulates the equations through Fermi-surface weights $Z(\mathbf{k},i\omega_n)$3; this is explicitly described as a Fermi-surface-centered anisotropic ME treatment with a finite Fermi window, not a strict full-bandwidth scheme (Reddy et al., 18 May 2026). Similarly, the IR-based anisotropic solver with inner and outer windows of $Z(\mathbf{k},i\omega_n)$4 eV and $Z(\mathbf{k},i\omega_n)$5 eV is broader than conventional Fermi-surface-restricted anisotropic ME, because it keeps high-energy states for Coulomb retardation, but it does not solve the fully anisotropic electron-phonon problem over all bands up to tens of eV (Mori et al., 2024).

The full-bandwidth formulation is thus best understood as a theory in which the retained electronic structure itself participates in the self-consistency. In FeSe/SrTiO$Z(\mathbf{k},i\omega_n)$6, that means the entire ten-band tight-binding dispersion and fixed occupancy; in the EPW superhydride implementation, it means full-BZ sums over states not restricted to the Fermi surface, together with $Z(\mathbf{k},i\omega_n)$7 and, in the FBW+$Z(\mathbf{k},i\omega_n)$8 variant, a self-consistent $Z(\mathbf{k},i\omega_n)$9 (Aperis et al., 2017, Lucrezi et al., 2023).

4. Numerical realization and algorithmic developments

Full-bandwidth anisotropic Migdal-Eliashberg equations are substantially more expensive than their Fermi-surface-restricted counterparts because the self-consistency spans band, momentum, and Matsubara or real-frequency variables, and because the number equation and $\chi(\mathbf{k},i\omega_n)$0 channel introduce additional long-frequency tails. Several numerical strategies have been developed to make this tractable.

In Matsubara space, the analytic-tail scheme assumes that for sufficiently large $\chi(\mathbf{k},i\omega_n)$1, the interacting solution approaches the bare one: $\chi(\mathbf{k},i\omega_n)$2 Instead of discarding the infinite tail, the omitted contributions are added back analytically. Applied to the full-bandwidth, multiband, anisotropic FeSe/SrTiO$\chi(\mathbf{k},i\omega_n)$3 equations, this yields similarly converged results with less than one fifth of the number of frequencies compared to the standard hard-cutoff procedure; for the isotropic equations, the paper reports approximately ninety percent of the complexity spared (Schrodi et al., 2018).

Within EPW, a separate acceleration was introduced through a non-uniform Matsubara grid,

$\chi(\mathbf{k},i\omega_n)$4

which is uniform at low frequencies and progressively sparser at high frequencies. In the superhydride full-bandwidth implementation, the default $\chi(\mathbf{k},i\omega_n)$5 gives about 30% fewer Matsubara frequencies and about 40% lower computational cost, with maximum $\chi(\mathbf{k},i\omega_n)$6 differences of 3 K for D$\chi(\mathbf{k},i\omega_n)$7S and 1 K for the other tested hydrides (Lucrezi et al., 2023).

A more radical compression of Matsubara frequency dependence is obtained by combining Wannier interpolation with the intermediate representation basis. In that approach, fermionic and bosonic Green’s functions are expanded in IR basis functions, and with $\chi(\mathbf{k},i\omega_n)$8 and $\chi(\mathbf{k},i\omega_n)$9, only 96 fermionic and 97 bosonic sampling points are needed; after symmetry reduction, 48 positive fermionic and 49 positive bosonic frequencies are used in practice. The method is explicitly designed to retain Coulomb retardation over a wide outer window, here $\phi(\mathbf{k},i\omega_n)$0 eV, while keeping the electron-phonon anisotropy in an inner $\phi(\mathbf{k},i\omega_n)$1 eV window (Mori et al., 2024).

Real-axis solvers introduce a different set of difficulties. The direct real-axis full-bandwidth method beyond the constant DOS approximation rewrites the problem in terms of an energy-dependent DOS $\phi(\mathbf{k},i\omega_n)$2, an energy-dependent static screened Coulomb interaction $\phi(\mathbf{k},i\omega_n)$3, and a real-frequency kernel

$\phi(\mathbf{k},i\omega_n)$4

Its principal-value integrals are reorganized so that the computationally expensive part depends only on the distinct differences $\phi(\mathbf{k},i\omega_n)$5, reducing the cost to linear scaling in the size of the real-frequency grid (Simon et al., 18 Mar 2026).

These algorithmic developments indicate that full-bandwidth anisotropic ME theory is constrained less by formalism than by representation. This suggests that future implementations will continue to combine Wannier interpolation, sparse frequency representations, and direct real-axis methods rather than relying on a single universal solver (Schrodi et al., 2018, Mori et al., 2024, Simon et al., 18 Mar 2026).

5. Physical consequences and representative applications

The best-known demonstration of what full bandwidth changes physically is monolayer FeSe/SrTiO$\phi(\mathbf{k},i\omega_n)$6. There the only pairing glue is a dispersionless interfacial optical phonon with $\phi(\mathbf{k},i\omega_n)$7 meV and forward-focused coupling

$\phi(\mathbf{k},i\omega_n)$8

In a Fermi-surface-only anisotropic calculation, the coupling is $\phi(\mathbf{k},i\omega_n)$9, $\Delta=\phi/Z$0 K, and replica bands do not appear. When the same two electron bands are treated at full bandwidth, spectral weight is transferred into replica bands, $\Delta=\phi/Z$1 drops to about 0.4, and $\Delta=\phi/Z$2 decreases to about 56.8 K. When the full ten-band structure is restored, deep Fermi-sea Cooper pairing compensates the loss and gives $\Delta=\phi/Z$3 K. The same full-bandwidth treatment yields a plain anisotropic $\Delta=\phi/Z$4-wave gap of about 8–11 meV, explains replica bands, and predicts tunneling features near $\Delta=\phi/Z$5 meV together with higher-energy dip-hump structure (Aperis et al., 2017).

Superhydrides provide a different kind of test: they probe the failure of the constant-DOS approximation near strong DOS structure. In YH$\Delta=\phi/Z$6 and CaH$\Delta=\phi/Z$7, where the DOS near $\Delta=\phi/Z$8 is comparatively smooth, FSR and FBW results are close: for YH$\Delta=\phi/Z$9, $\epsilon_{\mathbf{k}}$0 K in FSR, $\epsilon_{\mathbf{k}}$1 K in FBW, and $\epsilon_{\mathbf{k}}$2 K in FBW+$\epsilon_{\mathbf{k}}$3; for CaH$\epsilon_{\mathbf{k}}$4, $\epsilon_{\mathbf{k}}$5, $\epsilon_{\mathbf{k}}$6, and $\epsilon_{\mathbf{k}}$7 K, respectively. In H$\epsilon_{\mathbf{k}}$8S and D$\epsilon_{\mathbf{k}}$9S, by contrast, $\chi$0 lies on the shoulder of a strong DOS peak. There the full-bandwidth treatment suppresses $\chi$1 more strongly: H$\chi$2S gives $\chi$3 K in FSR, $\chi$4 K in FBW, and $\chi$5 K in FBW+$\chi$6; D$\chi$7S gives $\chi$8, $\chi$9, and $\xi_{n\mathbf{k}}$00 K. The isotope coefficient changes from $\xi_{n\mathbf{k}}$01 in FSR to $\xi_{n\mathbf{k}}$02 in FBW and $\xi_{n\mathbf{k}}$03 in FBW+$\xi_{n\mathbf{k}}$04, closer to the quoted experimental value $\xi_{n\mathbf{k}}$05 at 150 GPa (Lucrezi et al., 2023).

The same superhydride work also uses rigid-band shifts to test a common design heuristic, namely maximizing $\xi_{n\mathbf{k}}$06 by placing $\xi_{n\mathbf{k}}$07 on a DOS peak. In doped H$\xi_{n\mathbf{k}}$08S, the FSR approximation predicts a large enhancement near the DOS maximum, whereas FBW yields only about a 5–10% maximum increase in $\xi_{n\mathbf{k}}$09. In BaSiH$\xi_{n\mathbf{k}}$10, whose DOS is strongly asymmetric and step-like around $\xi_{n\mathbf{k}}$11, the full-bandwidth treatment can instead increase $\xi_{n\mathbf{k}}$12: the undoped FBW value is 87 K, and doping with $\xi_{n\mathbf{k}}$13 eV raises it to about 92 K. This suggests that full-bandwidth theory is not merely a suppression correction; it restores the actual energy-dependent phase space and may raise or lower $\xi_{n\mathbf{k}}$14 depending on DOS shape (Lucrezi et al., 2023).

Direct real-axis full-bandwidth calculations reinforce this point. In H$\xi_{n\mathbf{k}}$15S at 200 GPa, the van-Hove singularity near the Fermi level produces strong particle-hole asymmetry. The full-bandwidth real-axis solution yields $\xi_{n\mathbf{k}}$16 meV, whereas the cDOS+$\xi_{n\mathbf{k}}$17 treatment gives about 75 meV, and the full-bandwidth lineshapes are reported to be closer to experiment. The computed spectral functions $\xi_{n\mathbf{k}}$18, quasiparticle DOS, and occupancies also show the expected asymmetry and fine structure that are difficult to reconstruct reliably from analytic continuation (Simon et al., 18 Mar 2026).

6. Beyond Migdal, limitations, and open questions

The full-bandwidth anisotropic formalism does not, by itself, settle the question of when Migdal-Eliashberg theory remains valid. One line of development extends the full-bandwidth anisotropic equations beyond Migdal’s approximation by retaining the first crossed self-energy diagram in addition to the usual noncrossing term. In the resulting one-band model formulation, the self-energy contains both the standard Migdal contribution and a nested vertex-corrected term, and the corresponding $\xi_{n\mathbf{k}}$19, $\xi_{n\mathbf{k}}$20, and $\xi_{n\mathbf{k}}$21 equations are described as “formally exact up to second order in $\xi_{n\mathbf{k}}$22” within that diagrammatic truncation. The numerical conclusion is that vertex corrections can be positive, negative, or negligible depending on dimensionality, coupling strength, and the non-adiabatic ratio $\xi_{n\mathbf{k}}$23; in 3D the adiabatic full-bandwidth equations often resemble the vertex-corrected results closely, whereas in 2D the corrections can substantially alter both $\xi_{n\mathbf{k}}$24 and $\xi_{n\mathbf{k}}$25 (Schrodi et al., 2019).

A different limitation is more fundamental. The analysis of the breakdown of Migdal-Eliashberg theory in electron-phonon systems argues that the decisive failure can come from polaronic or bipolaronic ground states generated by fermions with energies comparable to the bandwidth $\xi_{n\mathbf{k}}$26, rather than from phonon softening alone. The central claim is that such processes are outside the field of standard MET because the theory is built from $\xi_{n\mathbf{k}}$27, $\xi_{n\mathbf{k}}$28, and $\xi_{n\mathbf{k}}$29, not from the bandwidth as an essential scale. In this view, full-bandwidth inclusion is necessary to see where the important physics lives, but a one-loop Migdal-Eliashberg treatment may still miss the relevant nonperturbative instability (Chubukov et al., 15 Apr 2026).

Several practical limitations recur across current full-bandwidth implementations. Coulomb effects are often treated by a static screened interaction or by a Morel-Anderson-type $\xi_{n\mathbf{k}}$30, not by a fully dynamical momentum-resolved kernel. Many broad-bandwidth solvers retain energy dependence in $\xi_{n\mathbf{k}}$31 but not in $\xi_{n\mathbf{k}}$32 or $\xi_{n\mathbf{k}}$33, or treat high-energy states only in the Coulomb channel. Even the EPW full-bandwidth implementation uses a finite retained electronic window, typically 2 eV in the self-consistency and $\xi_{n\mathbf{k}}$34 eV around the shifted Fermi level in the doping studies, rather than an unrestricted all-band calculation (Lucrezi et al., 2023, Mori et al., 2024, Simon et al., 18 Mar 2026).

The most stable conclusion is therefore methodological. Full-bandwidth anisotropic Migdal-Eliashberg equations are required when the superconducting kernel samples a strongly energy-dependent electronic structure, when replica or shake-off features are visible, when particle-hole asymmetry matters, or when off-Fermi-surface states contribute appreciably to pairing. They are not interchangeable with standard Fermi-surface anisotropic equations, and “fully anisotropic” is not synonymous with “full-bandwidth.” At the same time, current full-bandwidth realizations remain approximations: they extend the electronic phase space and the self-energy structure, but they do not eliminate the underlying assumptions of Migdal-Eliashberg theory itself (Aperis et al., 2017, Lucrezi et al., 2023, Chubukov et al., 15 Apr 2026).