Parquet Decomposition in Many-Fermion Systems

• Parquet decomposition is a diagrammatic framework that decomposes full two-particle vertices into fully irreducible and channel-reducible parts in interacting systems.
• It employs coupled Bethe–Salpeter equations to maintain crossing symmetry and unbiasedly analyze scattering processes in particle–hole and particle–particle channels.
• Modern implementations using kernel approximations and truncated-unity techniques enable efficient and scalable computations for strongly correlated electron, nuclear, and cold atomic systems.

Parquet decomposition is a systematic and non-perturbative diagrammatic framework that provides an exact reorganization of two-particle correlations in interacting many-fermion systems. It underlies modern approaches to treating strongly correlated electron, nuclear, and cold atomic systems, as well as serves as the diagrammatic backbone for non-local extensions of mean-field theories, functional renormalization group (fRG), and advanced ab initio methods. The central idea is to decompose the full two-particle vertex into contributions that are fully irreducible or that are reducible in exactly one of several two-particle channels, yielding a set of coupled nonlinear integral equations—the parquet equations. Concretely, these equations self-consistently link the fully irreducible vertex, three channel-specific Bethe–Salpeter ladders, and the dynamically screened interactions, rigorously enforcing crossing symmetry and systematically summing all two-particle-reducible diagrams (Eckhardt et al., 2023, Kugler et al., 2017, Li et al., 2015, Bergli et al., 2010).

1. Diagrammatic Foundations and Algebraic Structure

The diagrammatic basis of the parquet decomposition is the classification of the full amputated four-point (two-particle) vertex function, typically denoted as $F(1,2;3,4)$ for incoming legs (1,2) and outgoing legs (3,4) with all internal quantum numbers and frequencies. Every Feynman diagram contributing to the two-particle vertex is either:

• (i) fully two-particle irreducible (cannot be separated by cutting any two fermionic lines),
• (ii) two-particle reducible in exactly one of several channels:
  • particle–particle (pp), longitudinal particle–hole (ph), or transverse particle–hole ($\overline{ph}$),
  • with the channel being defined by the specific momentum–frequency transfer carried by the cut lines.

Formally, the exact decomposition is

$$F = \Lambda + \Phi_{ph} + \Phi_{\overline{ph}} + \Phi_{pp}$$

where $\Lambda$ is the fully irreducible vertex, and each $\Phi_r$ is the sum of diagrams two-particle reducible only in channel $r$. There is no double counting due to explicit non-overlap of these definitions (Eckhardt et al., 2023, Li et al., 2015, Li et al., 2017).

Each channel-reducible part $\Phi_r$ obeys its own Bethe–Salpeter equation (BSE), which for channel $r$ relates the full, irreducible, and reducible vertices by

$\Phi_r = \Gamma_r \star G G \star F_r$

where $\Gamma_r$ is the vertex irreducible in channel $\overline{ph}$0, and the operation “$\overline{ph}$1” denotes the convolution in the appropriate two-particle bubble. Thus, the parquet formalism leads to a self-consistent closed set for $\overline{ph}$2, the $\overline{ph}$3, and the $\overline{ph}$4, given a fully irreducible input $\overline{ph}$5 (Eckhardt et al., 2023, Kugler et al., 2018, Kugler et al., 2017, Li et al., 2015).

2. Channel Structure, Crossing Symmetry, and Physical Interpretation

The channel structure is dictated by the three inequivalent ways of pairing the four external legs. In most condensed-matter contexts:

• Particle–hole (ph) channel: propagation of electron–hole pairs,
• Transverse particle–hole ($\overline{ph}$6): crossed electron–hole pairs,
• Particle–particle (pp) channel: pairing.

For each, the reducible part sums up all diagrams that become disconnected upon appropriate two-line cuts (Valli et al., 2014, Gunnarsson et al., 2016). The parquet equations rigorously encode crossing symmetry—exchange of incoming/outgoing indices maps the equations between channels, providing essential constraints (see also the Hamiltonian derivation in (Green et al., 2022)). Physically, the decomposition provides an unbiased analysis of scattering processes, permitting a channel-wise breakdown of contributions to the self-energy and response functions (e.g., identifying which fluctuations dominate in different regimes, as in the pseudogap or antiferromagnetic instability in the 2D Hubbard model (Gunnarsson et al., 2016)).

3. Bethe–Salpeter Equations and Self-Consistency

Each channel-reducible part $\overline{ph}$7 satisfies a nonlinear integral equation (the Bethe–Salpeter equation in channel $\overline{ph}$8), iteratively building up all ladder- and bubble-type diagrams specific to that channel. Schematically,

$\overline{ph}$9

with similar structure for the other two channels, except with the appropriate channel dependence of momentum/frequency transfer (Li et al., 2015, Kugler et al., 2017, Li et al., 2017, Rohshap et al., 2024). The channel-irreducible vertex $$F = \Lambda + \Phi_{ph} + \Phi_{\overline{ph}} + \Phi_{pp}$$0 is then constructed as

$$F = \Lambda + \Phi_{ph} + \Phi_{\overline{ph}} + \Phi_{pp}$$1

and all blocks are updated iteratively until convergence (Eckhardt et al., 2023).

Self-energy feedback is provided by the Schwinger–Dyson (equation-of-motion) relation, expressing the single-particle self-energy in terms of the fully resolved two-particle vertex, which is necessary for enforcing conserving approximations (Kugler et al., 2017, Li et al., 2015, Valli et al., 2014).

4. Numerical Realizations and Computational Schemes

Standard parquet solvers require the storage and updating of all three-frequency (and possibly momentum-dependent) vertex functions in three channels, leading to $$F = \Lambda + \Phi_{ph} + \Phi_{\overline{ph}} + \Phi_{pp}$$2 memory and CPU scaling, which has traditionally been a major computational bottleneck. Modern efficient implementations employ several strategies:

• The "kernel approximation" reduces the parameter space of the vertex functions by replacing asymptotic (“high-frequency”) regions with analytically controlled kernel functions of reduced arguments, greatly lowering the computational load while maintaining correct high-frequency and boundary behavior (Li et al., 2015, Li et al., 2017).
• The "truncated-unity" (TU) scheme expands all momentum dependences in a small basis of form-factors, achieving linear scaling in the number of momenta and efficient use of fast Fourier transforms in momentum convolutions, with rapid convergence observed for leading instabilities (Eckhardt et al., 2018, 2008.04184).
• Tensor-network methodologies—such as quantics tensor trains (QTT)—exploit the low-rank structure of multivariate tensor representations of frequency/momentum-dependent vertices, allowing exponentially large frequency grids to be handled with only linear memory/CPU growth (Rohshap et al., 2024).
• The Single-Boson-Exchange (SBE) or "Hedin vertex" formalism recasts the four-point problem in terms of three-legged (Hedin) vertices and bosonic propagators, summing all Maki–Thompson diagrams and eliminating the need to explicitly store or invert large four-point kernels, with dramatic reductions in complexity and improved numerical stability (Krien et al., 2019, Al-Eryani, 18 Sep 2025).

The core steps in these implementations involve updating the relevant Green’s functions and vertex/irreducible kernels, iteratively solving the Bethe–Salpeter equations in all channels, and updating the self-energy and polarization functions until convergence (Li et al., 2017, Li et al., 2015, Krien et al., 2019). The crossing symmetry and high-frequency asymptotics are maintained at every stage.

5. Physical Applications and Domain-Specific Adaptations

Parquet decomposition is universally applicable to quantum lattice models (Hubbard model, Anderson impurity model, etc.), ab initio nuclear structure theory, and quantum fluids (Li et al., 2017, Bergli et al., 2010, Krotscheck et al., 2021). In correlated electron systems, parquet solvers have been central in demonstrating dominance of specific fluctuation channels (e.g., spin or charge), characterizing pseudogap phenomena, and identifying the feedback of non-local two-particle correlations on spectral properties (Gunnarsson et al., 2016, Eckhardt et al., 2018).

Ab initio nuclear structure calculations exploit the parquet equations to provide size-extensive, conserving summation of ladder, ring, and vertex corrections, with quantitative agreement to other many-body methods in moderate coupling regimes (Bergli et al., 2010, Krotscheck et al., 2021). In functional renormalization group, the equivalence between multiloop fRG and the parquet summation has been rigorously established, offering one-loop differential equations yielding the same set of summed diagrams as the full parquet theory, with all diagram classes generated systematically by loop order (Kugler et al., 2017, Kugler et al., 2018).

In multibosonic and bosonization approaches, parquet formalism has been exactly recast in terms of screened bosonic propagators and Hedin three-leg vertices, offering both interpretative and technical simplification, as well as clarifying the emergence of the Hohenberg–Mermin–Wagner theorem in two-dimensional systems (Al-Eryani, 18 Sep 2025, Krien et al., 2019, Krien et al., 2020).

The table below illustrates a selection of numerical techniques and their typical scaling properties:

Numerical Scheme	Memory Scaling	CPU Scaling
Standard 3-frequency cube	$$F = \Lambda + \Phi_{ph} + \Phi_{\overline{ph}} + \Phi_{pp}$$3	$$F = \Lambda + \Phi_{ph} + \Phi_{\overline{ph}} + \Phi_{pp}$$4
Kernel approximation (Li et al., 2015)	$$F = \Lambda + \Phi_{ph} + \Phi_{\overline{ph}} + \Phi_{pp}$$5	$$F = \Lambda + \Phi_{ph} + \Phi_{\overline{ph}} + \Phi_{pp}$$6
Truncated-unity (Eckhardt et al., 2018)	$$F = \Lambda + \Phi_{ph} + \Phi_{\overline{ph}} + \Phi_{pp}$$7	$$F = \Lambda + \Phi_{ph} + \Phi_{\overline{ph}} + \Phi_{pp}$$8
Single-Boson-Exchange (Krien et al., 2019)	$$F = \Lambda + \Phi_{ph} + \Phi_{\overline{ph}} + \Phi_{pp}$$9	$\Lambda$0
Quantics Tensor Trains (Rohshap et al., 2024)	$\Lambda$1	$\Lambda$2

6. Generalizations, Theoretical Insights, and Limitations

From the functional-analytic perspective, the parquet decomposition emerges naturally from successive Legendre transforms of the Luttinger–Ward functional, placing all constituent objects—self-energy, irreducible vertices, reducible parts—on precise mathematical footing as functional derivatives (Eckhardt et al., 2023). This ensures that symmetry/diagrammatic combinatorics are fully respected and that generalizations to higher-order vertices (three-particle and beyond) can, at least in principle, be constructed.

A notable universal result is that parquet theory, in the limit of criticality for an O(N) order parameter, reduces to the self-consistent screening approximation for the corresponding bosonic field theory, with implications for critical exponents and the enforcement of constraints such as the Hohenberg–Mermin–Wagner theorem (Al-Eryani, 18 Sep 2025).

While the parquet approach is unbiased and systematic, numerical instabilities and divergences can arise as interaction strength increases, typically reflecting physically meaningful proximity to strong-coupling regimes, resonating-valence-bond correlations, or suppression of specific channels (e.g., charge). In practice, careful frequency parametrization and regularization are necessary for stability at large U, especially for implementations aiming at non-perturbative phenomena (Gunnarsson et al., 2016, Li et al., 2015).

7. Extensions and Outlook

Current research continues to develop numerically tractable parquet-based algorithms on large clusters and multi-orbital systems by combining kernel truncations, form-factor expansions, and tensor-network approaches (Eckhardt et al., 2018, Rohshap et al., 2024). The bosonic reformulations of the parquet equations, particularly based on single-boson-exchange, now provide viable alternatives in regimes previously inaccessible due to vertex divergences or high computational cost (Krien et al., 2019, Al-Eryani, 18 Sep 2025). Unified fRG–parquet frameworks and further functional-analytic generalizations promise systematic progress towards unbiased, conserving approximations including collective modes and criticality in strongly correlated systems (Kugler et al., 2017, Kugler et al., 2018, Eckhardt et al., 2023).

Parquet decomposition remains an indispensable tool for both theoretical and computational quantum many-body physics, providing both the framework for and the benchmark against which new diagrammatic and numerical methods are developed.