Multichannel Dyson Equation Overview

• Multichannel Dyson Equation is a matrix generalization that couples several many-body Green’s functions into an enlarged propagator to treat quasiparticles and satellites on equal footing.
• It employs a block-matrix formulation to explicitly couple different excitation channels (e.g., 1p, 2h1e/2e1h, 2e2h) to enhance the representation of complex many-body interactions.
• The approach is applied to photoemission, neutral excitations, and double ionization, improving spectral fidelity and computational efficiency via effective Hamiltonian mapping.

Searching arXiv for papers on the multichannel Dyson equation and closely related Dyson-equation generalizations. The multichannel Dyson equation is a matrix generalization of the Dyson equation in which several many-body Green’s functions, or several channel sectors of higher-order Green’s functions, are coupled and solved within a single enlarged propagator. In the formulation introduced for photoemission, the one-body Green’s function is coupled to the $2h1e$ and $2e1h$ sectors of the three-body Green’s function, so that quasiparticle poles and satellite structures arise from the same effective problem rather than from a single one-body self-energy alone (Riva et al., 2023). Subsequent developments extended this construction to a broader hierarchy of even- and odd-order Green’s functions, to screened formulations for solids, and to neutral- and double-ionization spectroscopies (Riva et al., 2024, Riva et al., 7 Jan 2025, Romaniello et al., 28 Mar 2026, Sellié et al., 1 Apr 2026).

1. Definition and conceptual scope

In standard many-body perturbation theory, a single-channel Dyson equation relates one interacting Green’s function to a reference propagator and a self-energy,

$$G(\omega)=G_0(\omega)+G_0(\omega)\,\Sigma(\omega)\,G(\omega),$$

or equivalently $G^{-1}=G_0^{-1}-\Sigma$. In that setting, one usually works with the one-body Green’s function, and all many-body structure beyond the independent-particle poles is encoded indirectly in the frequency dependence of a one-body self-energy.

The multichannel Dyson equation replaces that closure on a single propagator by a coupled problem in an enlarged channel space. A “channel” denotes a sector with definite particle-hole content that shares the same final particle number and therefore the same excitation energies. For final states with $M=N\pm1$, the one-body Green’s function and the $2e1h/2h1e$ sectors of the three-body Green’s function have the same poles; for neutral excitations with $M=N$, the electron-hole channel of the two-body Green’s function and the $2e2h$ sector of the four-body Green’s function share the same poles (Riva et al., 7 Jan 2025).

This common pole structure is the central rationale of the formalism. Instead of integrating higher configurations out into a dynamical kernel, the multichannel Dyson equation keeps them explicitly. In the original photoemission application, this means that quasiparticles and satellites are treated on equal footing because both are eigenmodes of the same coupled problem (Riva et al., 2023). In later neutral-excitation and double-ionization formulations, the same logic is used to represent double excitations, biexcitons, and shake-up structures as explicit multichannel configurations rather than as residual structures hidden in a frequency-dependent kernel (Riva et al., 7 Jan 2025, Sellié et al., 1 Apr 2026).

A useful classification introduced in the generalized theory is the $(n,s)$-MCDE notation, where $n$ denotes the highest-order Green’s function retained and $2e1h$0 the change in particle number. Thus $2e1h$1-MCDE targets photoemission and inverse photoemission, $2e1h$2-MCDE targets neutral excitations, and $2e1h$3-MCDE targets double addition or double ionization (Riva et al., 7 Jan 2025, Sellié et al., 1 Apr 2026).

2. Block-matrix formulation

The multichannel Dyson equation is written for a block Green’s function,

$2e1h$4

where $2e1h$5 is block diagonal in channel space and $2e1h$6 contains both intra-channel and inter-channel blocks.

For the original $2e1h$7-MCDE, the noninteracting propagator is

$2e1h$8

where $2e1h$9 is the one-body Green’s function and $$G(\omega)=G_0(\omega)+G_0(\omega)\,\Sigma(\omega)\,G(\omega),$$0 is the three-body Green’s function restricted to the $$G(\omega)=G_0(\omega)+G_0(\omega)\,\Sigma(\omega)\,G(\omega),$$1 sector. The corresponding multichannel self-energy has the block form

$$G(\omega)=G_0(\omega)+G_0(\omega)\,\Sigma(\omega)\,G(\omega),$$2

with $$G(\omega)=G_0(\omega)+G_0(\omega)\,\Sigma(\omega)\,G(\omega),$$3 and $$G(\omega)=G_0(\omega)+G_0(\omega)\,\Sigma(\omega)\,G(\omega),$$4 coupling one-particle and three-particle sectors (Riva et al., 2023, Riva et al., 2024).

For neutral excitations, the $$G(\omega)=G_0(\omega)+G_0(\omega)\,\Sigma(\omega)\,G(\omega),$$5-MCDE couples the electron-hole channel of the two-body Green’s function to the $$G(\omega)=G_0(\omega)+G_0(\omega)\,\Sigma(\omega)\,G(\omega),$$6 channel of the four-body Green’s function. For double ionization, the $$G(\omega)=G_0(\omega)+G_0(\omega)\,\Sigma(\omega)\,G(\omega),$$7-MCDE couples the particle-particle channel of the two-body Green’s function to the $$G(\omega)=G_0(\omega)+G_0(\omega)\,\Sigma(\omega)\,G(\omega),$$8 channel of the four-body Green’s function (Riva et al., 7 Jan 2025, Sellié et al., 1 Apr 2026).

Variant	Coupled channels	Principal target
$$G(\omega)=G_0(\omega)+G_0(\omega)\,\Sigma(\omega)\,G(\omega),$$9-MCDE	$G^{-1}=G_0^{-1}-\Sigma$0 and $G^{-1}=G_0^{-1}-\Sigma$1 sectors of $G^{-1}=G_0^{-1}-\Sigma$2	Photoemission and inverse photoemission
$G^{-1}=G_0^{-1}-\Sigma$3-MCDE	$G^{-1}=G_0^{-1}-\Sigma$4 sector of $G^{-1}=G_0^{-1}-\Sigma$5 and $G^{-1}=G_0^{-1}-\Sigma$6 sector of $G^{-1}=G_0^{-1}-\Sigma$7	Neutral excitations, biexcitons, double excitations
$G^{-1}=G_0^{-1}-\Sigma$8-MCDE	Particle-particle $G^{-1}=G_0^{-1}-\Sigma$9 and $M=N\pm1$0 sector of $M=N\pm1$1	Double ionization and double addition

The physical interpretation of the off-diagonal blocks is direct. In the $M=N\pm1$2 case, they convert a one-particle excitation into a three-particle one and back, i.e. a quasiparticle into a shake-up configuration and back. In the $M=N\pm1$3 case, they convert a single electron-hole pair into a double electron-hole configuration and back. In the $M=N\pm1$4 case, they convert a double-ionization quasiparticle into a configuration with an extra electron-hole pair. This explicit conversion mechanism is what makes satellite and multi-excitation physics appear as ordinary poles of the coupled Green’s function rather than as emergent anomalies of a reduced self-energy (Riva et al., 2024, Riva et al., 7 Jan 2025, Sellié et al., 1 Apr 2026).

3. Multichannel self-energy and generated correlations

A distinctive feature of the MCDE literature is the use of a static multichannel self-energy constructed at first order in the bare Coulomb interaction. In the $M=N\pm1$5-MCDE, the proposed approximation sets

$M=N\pm1$6

with $M=N\pm1$7 the antisymmetrized Coulomb matrix element, while the three-particle block $M=N\pm1$8 is a static first-order expression built from antisymmetrized Coulomb integrals and occupation-number prefactors that separate the $M=N\pm1$9 and $2e1h/2h1e$0 sectors (Riva et al., 2023).

This static construction has two immediate consequences. First, no explicit frequency convolutions appear at the multichannel level. Second, no self-consistency loop is required in the original approximation, because the multichannel self-energy is not taken as a functional of the interacting multichannel Green’s function (Riva et al., 2024). Nevertheless, when the higher channel is downfolded, the effective lower-channel self-energy becomes dynamical. In the screened and unscreened photoemission formulations, the effective one-body correlation self-energy takes the form

$2e1h/2h1e$1

so its frequency dependence is entirely inherited from propagation in the higher channel (Romaniello et al., 28 Mar 2026).

Diagrammatically, the resulting content is substantially richer than the static input might suggest. For the $2e1h/2h1e$2-MCDE, the method is complete up to second order in the one-body self-energy, yields all ten third-order proper skeleton diagrams, includes all $2e1h/2h1e$3 diagrams to infinite order, and additionally generates ladder and vertex-correction diagrams beyond $2e1h/2h1e$4 (Riva et al., 2023). The detailed derivation traces these structures to repeated insertions of the coupling blocks and the three-particle block, with the head of the multichannel Dyson series reproducing the ordinary one-body self-energy expansion (Riva et al., 2024).

The same first-order-in-$2e1h/2h1e$5 strategy is extended systematically to higher-order MCDEs. In the $2e1h/2h1e$6-MCDE, the head block is the standard RPAx Bethe–Salpeter kernel in the electron-hole sector, while the four-particle block and coupling blocks add all first-order direct and exchange interactions within and between $2e1h/2h1e$7 and $2e1h/2h1e$8 configurations (Riva et al., 7 Jan 2025). In the $2e1h/2h1e$9-MCDE, the two-particle head is the pp-RPAx kernel, while the four-particle block and couplings add first-order pairwise Coulomb interactions in the $M=N$0 sector and between that sector and the pp channel (Sellié et al., 1 Apr 2026).

A screened variant modifies this strategy for extended systems. The screened multichannel Dyson equation keeps the block structure of the $M=N$1-MCDE but replaces selected bare Coulomb matrix elements in the three-particle block by statically screened matrix elements $M=N$2, with $M=N$3 evaluated at $M=N$4 from an RPA dielectric function. The motivation is that the unscreened MCDE implicitly screens some, but not all, Coulomb lines, whereas the screened version aims to ensure that all interactions are either explicitly or effectively screened (Romaniello et al., 28 Mar 2026).

4. Effective-Hamiltonian formulation

Because the multichannel self-energy is static in these constructions, the MCDE can be rewritten as a resolvent of an effective Hamiltonian. For the $M=N$5-MCDE,

$M=N$6

with

$M=N$7

where the one-particle block is diagonal in the reference orbital energies and the three-particle block is the noninteracting $M=N$8 Hamiltonian plus $M=N$9 (Riva et al., 2023). The screened MCDE has an analogous Hermitian effective Hamiltonian, with the screened three-particle block $2e2h$0 replacing the unscreened one (Romaniello et al., 28 Mar 2026).

This representation is structurally important. It makes all poles of the coupled Green’s function eigenvalues of a single static operator. Quasiparticles correspond to eigenvectors with large projection on the lowest channel, while satellites or multiple-excitation states correspond to eigenvectors with large projection on higher channels. The physical spectral function is obtained by projecting the resolvent onto the relevant head block; for photoemission, that is the one-particle component, and for neutral excitations it is the two-particle electron-hole component (Riva et al., 2023, Riva et al., 7 Jan 2025).

The effective-Hamiltonian mapping also changes the numerical character of the problem. In the original $2e2h$1-MCDE, direct diagonalization of the effective Hamiltonian scales as $2e2h$2, while iterative methods such as Haydock–Lanczos reduce the scaling to $2e2h$3 (Riva et al., 2023). In the screened silicon application, the effective Hamiltonian is solved with the Lanczos–Haydock method to obtain spectral information without full diagonalization (Romaniello et al., 28 Mar 2026). In the $2e2h$4-MCDE for double ionization, the dominant cost is the construction of the effective Hamiltonian, which scales as $2e2h$5 in the number of electrons or orbitals, and iterative solvers are again proposed as the practical route for large systems (Sellié et al., 1 Apr 2026).

This suggests a general tradeoff that recurs throughout the literature: the MCDE replaces a nonlinear or highly dynamical reduced problem by a larger but static eigenvalue problem. The computational burden moves from frequency-dependent kernels and self-consistency to enlarged configuration spaces and sparse linear algebra.

5. Applications and physical regimes

The first benchmark application was the Hubbard dimer. In the original $2e2h$6-MCDE, coupling the one-body and three-body channels yields the exact spectral function of the Hubbard dimer at both $2e2h$7 and $2e2h$8 filling, while $2e2h$9 fails to reproduce the satellites correctly and yields unphysical extra peaks (Riva et al., 2023). The same work also states that the approximate multichannel self-energy recovers the exact three-body Green’s function $(n,s)$0 for the dimer at those fillings.

A later application to the extended Hubbard dimer used the MCDE to compute the potential energy surface, spectral functions, and HOMO-LUMO gaps. The reported outcome is that the MCDE gives overall very good results for all properties considered and outperforms both $(n,s)$1 and second Born; in particular, it yields the correct ground-state energy and HOMO-LUMO gap in the dissociation limit, contrary to $(n,s)$2 (Paggi et al., 16 Jul 2025). This case is significant because dissociation is precisely the regime in which single-channel perturbative self-energies are known to become qualitatively unreliable.

For realistic solids, the screened MCDE was introduced to describe direct and inverse photoemission in bulk silicon. In that case, $(n,s)$3, the unscreened MCDE, and the second-Born approximation all strongly overestimate the binding energy of the plasmon satellite, placing it around $(n,s)$4 eV, whereas the experimental valence plasmon satellite is about $(n,s)$5 eV below the valence-band maximum. The screened MCDE captures the correct satellite position and also reproduces the main features of the direct and inverse photoemission spectra (Romaniello et al., 28 Mar 2026). The same study interprets this as evidence that properly screening all particle-particle and electron-hole interactions is essential in extended systems.

For neutral excitations, the $(n,s)$6-MCDE was applied to a two-level Helium-like model. Standard static BSE@GW does not produce a genuine double-excitation pole, whereas the MCDE yields both single and double excitations and gives excitation energies in good agreement with exact results (Riva et al., 7 Jan 2025). The analysis also shows that the quantitative position of the double excitation is sensitive to the quasiparticle gap used in the higher-channel block, which indicates that the quality of the underlying one-particle input remains important even when the multichannel coupling is correct.

For double-ionization spectroscopies, the $(n,s)$7-MCDE provides the formal extension beyond pp-RPA by coupling the particle-particle two-body Green’s function to the $(n,s)$8 channels of the four-body Green’s function. In this formulation, double-ionization quasiparticles and satellites are described within the same multichannel framework, and the downfolded two-particle self-energy becomes dynamical through the higher-channel propagator, despite the static first-order input kernel (Sellié et al., 1 Apr 2026).

Across these applications, a common pattern emerges. Whenever the relevant spectroscopy is strongly influenced by shake-up configurations, multiple electron-hole pairs, or dissociation-induced multiconfigurational character, the explicit retention of higher channels improves qualitative fidelity relative to single-channel approximations. This suggests that the principal strength of the MCDE is not merely numerical but representational: it enlarges the propagating space so that physically important configurations appear as primary degrees of freedom.

6. Related meanings, limitations, and extensions

The term “multichannel Dyson equation” is used in at least two related but distinct senses. In the MCDE literature discussed above, it denotes a coupled Dyson equation for different many-body Green’s functions or channel sectors. In another usage, “multichannel” refers to ordinary Dyson equations whose matrix indices already span several coupled physical channels such as momentum, orbital, atom, and spin. The fully self-consistent finite-temperature $(n,s)$9 study of NdNiO$n$0 is an example: there the Dyson equation is solved for $n$1 in the space of momentum-dependent spin-orbitals, and the authors explicitly interpret this as a multichannel problem in $n$2-point, orbital, and spin space. Their homotopy-continuation method is used to track multiple self-consistent Dyson solutions and branch switching in this high-dimensional channel space (Pokhilko et al., 30 Jun 2025). This is mathematically related, but it is not the same formalism as the block-coupled MCDE.

A further related usage appears in non-equilibrium quantum transport with several leads. In a rigorous partition-free formulation, the retarded Green operator acts on the full one-particle Hilbert space of an interacting sample coupled to several non-interacting reservoirs. The resulting Dyson equation is multichannel in the sense that the Green function has sample–lead and lead–lead blocks, while the interaction self-energy is supported only on the sample subspace (Cornean et al., 2019). Again, this is a multichannel Dyson equation in a transport-channel basis, not in the explicit many-body-channel sense of the MCDE.

Within the explicit MCDE program, several limitations are repeatedly emphasized. The first is scaling: enlarging the propagator space to $n$3, $n$4, or higher channels quickly becomes expensive (Riva et al., 2023, Sellié et al., 1 Apr 2026). The second is truncation: a $n$5-MCDE omits $n$6-body channels, a $n$7-MCDE omits $n$8 channels, and so on, so the method is controlled by channel selection rather than by a small parameter alone (Riva et al., 7 Jan 2025). The third is kernel approximation: most current implementations use static first-order kernels, which are exact only through low perturbative order and may require screened interactions or higher-order corrections for quantitative accuracy in general systems (Romaniello et al., 28 Mar 2026, Paggi et al., 16 Jul 2025).

The extension strategy is correspondingly explicit. The general formalism allows odd- and even-order channel towers, dressing of selected Coulomb lines by screened interactions, replacement of Hartree–Fock higher-channel propagators by second Born, $n$9, or $2e1h$00-matrix quasiparticle propagators, and higher-channel constructions such as $2e1h$01-MCDE for double satellites or $2e1h$02-MCDE for Auger and related spectroscopies (Riva et al., 2023, Riva et al., 7 Jan 2025, Romaniello et al., 28 Mar 2026, Sellié et al., 1 Apr 2026). A plausible implication is that the MCDE should be viewed less as a single approximation than as a general architecture for embedding lower-order observables into a systematically enlarged Green’s-function hierarchy.