Cluster D-TRILEX Framework

• The paper presents a cluster extension of D-TRILEX that replaces the single-site impurity with a cluster-based approach, ensuring an exact treatment of short-range correlations.
• It leverages dual fermion–boson techniques built on three-point vertices to efficiently incorporate long-range charge and spin fluctuation corrections.
• Benchmark results on the Hubbard model show that cluster D-TRILEX closely reproduces QMC self-energy with low error and reduces periodization ambiguities compared to traditional CDMFT.

Searching arXiv for the cited D-TRILEX cluster-extension paper and closely related identifiers. The cluster extension of D-TRILEX is a diagrammatic framework for correlated electronic systems in which the dual TRILEX expansion is built around a cluster reference system rather than a single-site one. In the formulation introduced in "Cluster-diagrammatic D-TRILEX approach to non-local electronic correlations" (Fossati et al., 8 Jul 2025), the method is designed to combine the exact treatment of short-range correlation effects within the cluster with an efficient diagrammatic description of long-range charge and spin collective fluctuations beyond the cluster. The construction is presented for the Hubbard model, benchmarked on the one-dimensional nano-ring Hubbard model, and analyzed in relation to periodization and translational-symmetry breaking.

1. Definition and conceptual setting

D-TRILEX is extended by replacing the single-site reference problem with a cluster impurity problem. The resulting scheme starts from cluster dynamical mean-field theory (CDMFT), but does not stop at the cluster solution: off-diagonal and longer-ranged correlations are generated afterwards by dual diagrammatics. In this sense, the method is a cluster-diagrammatic extension in which the cluster provides the nonperturbative short-distance input and the dual fermion–boson formalism supplies non-local corrections (Fossati et al., 8 Jul 2025).

The central objective is to retain an exact numerical treatment of intra-cluster physics while restoring, at least partially, information lost in a purely cluster-local construction. The paper formulates this objective in terms of combining "the exact treatment of short-range correlation effects within the cluster" with "an efficient diagrammatic description of the long-range charge and spin collective fluctuations beyond the cluster" (Fossati et al., 8 Jul 2025). This suggests that the cluster extension is intended to interpolate between strongly local impurity solvers and explicitly momentum-sensitive many-body approximations.

A distinguishing feature of the construction is that it is organized around three-point vertices rather than four-point vertices. The data explicitly notes that the computational effort "remains far cheaper than handling four-point vertices" (Fossati et al., 8 Jul 2025). A plausible implication is that the method targets a compromise between non-local accuracy and computational tractability.

2. Cluster reference system

The starting point is the Hubbard model on the original lattice, written as an action

${\cal S}[c^*,c] = -\sum_{k\sigma ll'}c_{k\sigma l}^{*}\bigl[(i\nu+\mu)\,\delta_{ll'}-\varepsilon^{ll'}_{k}\bigr]\,c_{k\sigma l'} \;+\;\frac12\!\sum_{\substack{k,k',q\l_1\!-\!l_4,\sigma\sigma'}} U^{\,l_1l_2l_3l_4}\, c_{k\sigma l_1}^{*}c_{k+q,\sigma l_2}^{\phantom{*}} c_{k'+q,\sigma' l_4}^{*}c_{k'\sigma' l_3}^{\phantom{*}}\,.$

Within CDMFT, the lattice is tiled into identical clusters of $N_{\rm c}$ sites and represented by an impurity action

${\cal S}_{\rm imp}[\bar c^*,\bar c] = \sum_{\nu,\sigma,ll'}\bar c_{\nu\sigma l}^*\bigl[(i\nu+\mu)\,\delta_{ll'}-\Delta^{ll'}_{\nu}\bigr]\bar c_{\nu\sigma l'} \;+\; \tfrac12\!\sum_{\substack{\nu,\nu',\omega\l_1\!-\!l_4,\sigma\sigma'}} U_{\,l_1l_2l_3l_4}\, \bar c_{\nu\sigma l_1}^* \bar c_{\nu+\omega,\sigma l_2}^{\phantom{*}} \bar c_{\nu'+\omega,\sigma' l_4}^* \bar c_{\nu'\sigma' l_3}^{\phantom{*}}\,.$

Here $\Delta^{ll'}_{\nu}$ is the hybridization matrix, which "in general has off-diagonal cluster indices" (Fossati et al., 8 Jul 2025). The numerical setup then rotates to a basis $\mathcal R$ that diagonalizes the local part of the cluster Hamiltonian,

$$\overline\varepsilon_{K} \;=\;\mathcal R\,\varepsilon_K\,\mathcal R^\dagger, \qquad \overline\Delta_\nu \;=\;\mathcal R\,\Delta_\nu\,\mathcal R^\dagger,$$

and chooses as reference system the diagonal hybridization $\Delta^{ll}_\nu$. The off-diagonal terms are not solved directly at the impurity level; instead, "off-diagonal terms are then generated later by the dual diagrammatics" (Fossati et al., 8 Jul 2025).

This separation of roles is one of the defining structural choices of the method. The impurity problem supplies a cluster-resolved but diagonally hybridized starting point, while the subsequent expansion reconstructs the missing non-local structure.

3. Dual fermion–boson construction beyond the cluster

The extension beyond the cluster is built by subtracting and adding back the diagonal $\Delta_\nu$, introducing dual fermions $\tilde f,\tilde f^*$ and Hubbard–Stratonovich bosons $\phi^{\varsigma}$ with $N_{\rm c}$0, and decoupling charge and spin channels. The dual action takes the form

$N_{\rm c}$1

where $N_{\rm c}$2 and $N_{\rm c}$3 are the impurity Green’s function and renormalized interaction, and $N_{\rm c}$4 is expressed through the three-point vertices $N_{\rm c}$5 (Fossati et al., 8 Jul 2025).

From the converged cluster impurity problem one computes the three-point vertex

$N_{\rm c}$6

together with the cluster susceptibilities $N_{\rm c}$7. These objects determine the dual propagators and dual self-energies.

At the one-particle level, the cluster self-energy is

$N_{\rm c}$8

with $N_{\rm c}$9. After summing dual diagrams, one obtains a non-local correction ${\cal S}_{\rm imp}[\bar c^*,\bar c] = \sum_{\nu,\sigma,ll'}\bar c_{\nu\sigma l}^*\bigl[(i\nu+\mu)\,\delta_{ll'}-\Delta^{ll'}_{\nu}\bigr]\bar c_{\nu\sigma l'} \;+\; \tfrac12\!\sum_{\substack{\nu,\nu',\omega\l_1\!-\!l_4,\sigma\sigma'}} U_{\,l_1l_2l_3l_4}\, \bar c_{\nu\sigma l_1}^* \bar c_{\nu+\omega,\sigma l_2}^{\phantom{*}} \bar c_{\nu'+\omega,\sigma' l_4}^* \bar c_{\nu'\sigma' l_3}^{\phantom{*}}\,.$0, and the full lattice self-energy is written as

${\cal S}_{\rm imp}[\bar c^*,\bar c] = \sum_{\nu,\sigma,ll'}\bar c_{\nu\sigma l}^*\bigl[(i\nu+\mu)\,\delta_{ll'}-\Delta^{ll'}_{\nu}\bigr]\bar c_{\nu\sigma l'} \;+\; \tfrac12\!\sum_{\substack{\nu,\nu',\omega\l_1\!-\!l_4,\sigma\sigma'}} U_{\,l_1l_2l_3l_4}\, \bar c_{\nu\sigma l_1}^* \bar c_{\nu+\omega,\sigma l_2}^{\phantom{*}} \bar c_{\nu'+\omega,\sigma' l_4}^* \bar c_{\nu'\sigma' l_3}^{\phantom{*}}\,.$1

The leading D-TRILEX correction has a ${\cal S}_{\rm imp}[\bar c^*,\bar c] = \sum_{\nu,\sigma,ll'}\bar c_{\nu\sigma l}^*\bigl[(i\nu+\mu)\,\delta_{ll'}-\Delta^{ll'}_{\nu}\bigr]\bar c_{\nu\sigma l'} \;+\; \tfrac12\!\sum_{\substack{\nu,\nu',\omega\l_1\!-\!l_4,\sigma\sigma'}} U_{\,l_1l_2l_3l_4}\, \bar c_{\nu\sigma l_1}^* \bar c_{\nu+\omega,\sigma l_2}^{\phantom{*}} \bar c_{\nu'+\omega,\sigma' l_4}^* \bar c_{\nu'\sigma' l_3}^{\phantom{*}}\,.$2-like form in each channel ${\cal S}_{\rm imp}[\bar c^*,\bar c] = \sum_{\nu,\sigma,ll'}\bar c_{\nu\sigma l}^*\bigl[(i\nu+\mu)\,\delta_{ll'}-\Delta^{ll'}_{\nu}\bigr]\bar c_{\nu\sigma l'} \;+\; \tfrac12\!\sum_{\substack{\nu,\nu',\omega\l_1\!-\!l_4,\sigma\sigma'}} U_{\,l_1l_2l_3l_4}\, \bar c_{\nu\sigma l_1}^* \bar c_{\nu+\omega,\sigma l_2}^{\phantom{*}} \bar c_{\nu'+\omega,\sigma' l_4}^* \bar c_{\nu'\sigma' l_3}^{\phantom{*}}\,.$3: ${\cal S}_{\rm imp}[\bar c^*,\bar c] = \sum_{\nu,\sigma,ll'}\bar c_{\nu\sigma l}^*\bigl[(i\nu+\mu)\,\delta_{ll'}-\Delta^{ll'}_{\nu}\bigr]\bar c_{\nu\sigma l'} \;+\; \tfrac12\!\sum_{\substack{\nu,\nu',\omega\l_1\!-\!l_4,\sigma\sigma'}} U_{\,l_1l_2l_3l_4}\, \bar c_{\nu\sigma l_1}^* \bar c_{\nu+\omega,\sigma l_2}^{\phantom{*}} \bar c_{\nu'+\omega,\sigma' l_4}^* \bar c_{\nu'\sigma' l_3}^{\phantom{*}}\,.$4 The paper also states that the Bethe–Salpeter equation for the full lattice susceptibility is built from the cluster vertex and the lattice dual propagators (Fossati et al., 8 Jul 2025).

Taken together, these formulas show that the cluster extension preserves the TRILEX-type coupling between fermionic propagation and bosonic collective modes, but embeds that coupling in a cluster-resolved dual formalism.

4. Self-consistent computational scheme

The computational workflow is organized into a cluster DMFT stage, a self-consistent dual loop, and a lattice reconstruction stage (Fossati et al., 8 Jul 2025).

In the cluster DMFT step, one diagonalizes the local cluster Hamiltonian to obtain ${\cal S}_{\rm imp}[\bar c^*,\bar c] = \sum_{\nu,\sigma,ll'}\bar c_{\nu\sigma l}^*\bigl[(i\nu+\mu)\,\delta_{ll'}-\Delta^{ll'}_{\nu}\bigr]\bar c_{\nu\sigma l'} \;+\; \tfrac12\!\sum_{\substack{\nu,\nu',\omega\l_1\!-\!l_4,\sigma\sigma'}} U_{\,l_1l_2l_3l_4}\, \bar c_{\nu\sigma l_1}^* \bar c_{\nu+\omega,\sigma l_2}^{\phantom{*}} \bar c_{\nu'+\omega,\sigma' l_4}^* \bar c_{\nu'\sigma' l_3}^{\phantom{*}}\,.$5, transforms the lattice problem to that basis, solves CDMFT with diagonal ${\cal S}_{\rm imp}[\bar c^*,\bar c] = \sum_{\nu,\sigma,ll'}\bar c_{\nu\sigma l}^*\bigl[(i\nu+\mu)\,\delta_{ll'}-\Delta^{ll'}_{\nu}\bigr]\bar c_{\nu\sigma l'} \;+\; \tfrac12\!\sum_{\substack{\nu,\nu',\omega\l_1\!-\!l_4,\sigma\sigma'}} U_{\,l_1l_2l_3l_4}\, \bar c_{\nu\sigma l_1}^* \bar c_{\nu+\omega,\sigma l_2}^{\phantom{*}} \bar c_{\nu'+\omega,\sigma' l_4}^* \bar c_{\nu'\sigma' l_3}^{\phantom{*}}\,.$6, and extracts the impurity quantities ${\cal S}_{\rm imp}[\bar c^*,\bar c] = \sum_{\nu,\sigma,ll'}\bar c_{\nu\sigma l}^*\bigl[(i\nu+\mu)\,\delta_{ll'}-\Delta^{ll'}_{\nu}\bigr]\bar c_{\nu\sigma l'} \;+\; \tfrac12\!\sum_{\substack{\nu,\nu',\omega\l_1\!-\!l_4,\sigma\sigma'}} U_{\,l_1l_2l_3l_4}\, \bar c_{\nu\sigma l_1}^* \bar c_{\nu+\omega,\sigma l_2}^{\phantom{*}} \bar c_{\nu'+\omega,\sigma' l_4}^* \bar c_{\nu'\sigma' l_3}^{\phantom{*}}\,.$7.

The self-consistent dual loop proceeds through the sequence listed in the paper:

1. Build bare dual propagators ${\cal S}_{\rm imp}[\bar c^*,\bar c] = \sum_{\nu,\sigma,ll'}\bar c_{\nu\sigma l}^*\bigl[(i\nu+\mu)\,\delta_{ll'}-\Delta^{ll'}_{\nu}\bigr]\bar c_{\nu\sigma l'} \;+\; \tfrac12\!\sum_{\substack{\nu,\nu',\omega\l_1\!-\!l_4,\sigma\sigma'}} U_{\,l_1l_2l_3l_4}\, \bar c_{\nu\sigma l_1}^* \bar c_{\nu+\omega,\sigma l_2}^{\phantom{*}} \bar c_{\nu'+\omega,\sigma' l_4}^* \bar c_{\nu'\sigma' l_3}^{\phantom{*}}\,.$8 and ${\cal S}_{\rm imp}[\bar c^*,\bar c] = \sum_{\nu,\sigma,ll'}\bar c_{\nu\sigma l}^*\bigl[(i\nu+\mu)\,\delta_{ll'}-\Delta^{ll'}_{\nu}\bigr]\bar c_{\nu\sigma l'} \;+\; \tfrac12\!\sum_{\substack{\nu,\nu',\omega\l_1\!-\!l_4,\sigma\sigma'}} U_{\,l_1l_2l_3l_4}\, \bar c_{\nu\sigma l_1}^* \bar c_{\nu+\omega,\sigma l_2}^{\phantom{*}} \bar c_{\nu'+\omega,\sigma' l_4}^* \bar c_{\nu'\sigma' l_3}^{\phantom{*}}\,.$9.
2. Compute the dual polarization $\Delta^{ll'}_{\nu}$0 via the two-$\Delta^{ll'}_{\nu}$1–two-$\Delta^{ll'}_{\nu}$2 bubble.
3. Dress the dual interaction through $\Delta^{ll'}_{\nu}$3.
4. Compute the dual self-energy $\Delta^{ll'}_{\nu}$4 via the single-boson exchange.
5. Update the dual Green’s function according to $\Delta^{ll'}_{\nu}$5.
6. Iterate until $\Delta^{ll'}_{\nu}$6 converge.

The lattice reconstruction then uses

$\Delta^{ll'}_{\nu}$7

followed by

$\Delta^{ll'}_{\nu}$8

and finally a rotation back to the original site-orbital basis,

$\Delta^{ll'}_{\nu}$9

with an analogous transformation for $\mathcal R$0.

Convergence is declared when changes in $\mathcal R$1 and $\mathcal R$2 between iterations fall below a chosen threshold, with $\mathcal R$3 given as an example. The cost per dual iteration scales as

$\mathcal R$4

and the paper emphasizes that this remains much less demanding than schemes requiring four-point vertices (Fossati et al., 8 Jul 2025).

5. Periodization and translational symmetry

A central issue in cluster methods is the reconstruction of lattice quantities from cluster-resolved objects. Once $\mathcal R$5 is known in the reduced Brillouin zone, translational invariance is imposed through

$\mathcal R$6

where $\mathcal R$7 denotes either $\mathcal R$8 or $\mathcal R$9 (Fossati et al., 8 Jul 2025).

For a dimer this becomes

$$\overline\varepsilon_{K} \;=\;\mathcal R\,\varepsilon_K\,\mathcal R^\dagger, \qquad \overline\Delta_\nu \;=\;\mathcal R\,\Delta_\nu\,\mathcal R^\dagger,$$0

The paper contrasts pure CDMFT with cluster D-TRILEX on this point. In pure CDMFT, $$\overline\varepsilon_{K} \;=\;\mathcal R\,\varepsilon_K\,\mathcal R^\dagger, \qquad \overline\Delta_\nu \;=\;\mathcal R\,\Delta_\nu\,\mathcal R^\dagger,$$1, and "the periodized $$\overline\varepsilon_{K} \;=\;\mathcal R\,\varepsilon_K\,\mathcal R^\dagger, \qquad \overline\Delta_\nu \;=\;\mathcal R\,\Delta_\nu\,\mathcal R^\dagger,$$2 is far from the intra-cluster value" (Fossati et al., 8 Jul 2025). In cluster D-TRILEX, the diagrammatic correction $$\overline\varepsilon_{K} \;=\;\mathcal R\,\varepsilon_K\,\mathcal R^\dagger, \qquad \overline\Delta_\nu \;=\;\mathcal R\,\Delta_\nu\,\mathcal R^\dagger,$$3 generates a nonzero inter-cluster $$\overline\varepsilon_{K} \;=\;\mathcal R\,\varepsilon_K\,\mathcal R^\dagger, \qquad \overline\Delta_\nu \;=\;\mathcal R\,\Delta_\nu\,\mathcal R^\dagger,$$4, and the difference between intra-, inter-, and periodized $$\overline\varepsilon_{K} \;=\;\mathcal R\,\varepsilon_K\,\mathcal R^\dagger, \qquad \overline\Delta_\nu \;=\;\mathcal R\,\Delta_\nu\,\mathcal R^\dagger,$$5 is reported to be "drastically reduced compared to CDMFT alone" (Fossati et al., 8 Jul 2025). The paper interprets this as a partial restoration of translational symmetry by non-local diagrams.

The source of the remaining mismatch is identified explicitly: "the CDMFT impurity problem" is described as "the main source of the translational-symmetry breaking" (Fossati et al., 8 Jul 2025). A proposed remedy is an outer self-consistency loop in which the dual-corrected $$\overline\varepsilon_{K} \;=\;\mathcal R\,\varepsilon_K\,\mathcal R^\dagger, \qquad \overline\Delta_\nu \;=\;\mathcal R\,\Delta_\nu\,\mathcal R^\dagger,$$6 or hybridization $$\overline\varepsilon_{K} \;=\;\mathcal R\,\varepsilon_K\,\mathcal R^\dagger, \qquad \overline\Delta_\nu \;=\;\mathcal R\,\Delta_\nu\,\mathcal R^\dagger,$$7 is fed back into the impurity solver. The stated aim is that this would enforce $$\overline\varepsilon_{K} \;=\;\mathcal R\,\varepsilon_K\,\mathcal R^\dagger, \qquad \overline\Delta_\nu \;=\;\mathcal R\,\Delta_\nu\,\mathcal R^\dagger,$$8 exactly and thereby restore full translational invariance. Since the paper presents this as a proposal, it is most accurately described as an intended computational scheme rather than an implemented result.

6. Benchmarks and comparative performance

The numerical demonstration is carried out for the one-dimensional nanoring Hubbard model, with chains tiled as $$\overline\varepsilon_{K} \;=\;\mathcal R\,\varepsilon_K\,\mathcal R^\dagger, \qquad \overline\Delta_\nu \;=\;\mathcal R\,\Delta_\nu\,\mathcal R^\dagger,$$9 using two-site clusters, and all results reported at $\Delta^{ll}_\nu$0, $\Delta^{ll}_\nu$1 (Fossati et al., 8 Jul 2025). The principal observable is the imaginary part of the self-energy at the Fermi momentum $\Delta^{ll}_\nu$2.

The benchmark comparisons given in the paper can be organized as follows.

Method	Reported behavior
Exact Hirsch–Fye QMC	Shows an insulating divergence $\Delta^{ll}_\nu$3 as $\Delta^{ll}_\nu$4
Parquet DΓA	Fails to diverge at $\Delta^{ll}_\nu$5 and remains essentially at the DMFT level
Ladder DΓA	Yields a divergence but is quantitatively less accurate
Single-site D-TRILEX	Captures an insulating branch, but slightly overshoots at low $\Delta^{ll}_\nu$6
Two-site cluster D-TRILEX	Lies almost on top of QMC for both $\Delta^{ll}_\nu$7 and $\Delta^{ll}_\nu$8

To quantify the error at the lowest Matsubara point $\Delta^{ll}_\nu$9, the paper defines

$\Delta_\nu$0

The reported values are $\Delta_\nu$1 for cluster D-TRILEX, $\Delta_\nu$2–$\Delta_\nu$3 for ladder DΓA, and $\Delta_\nu$4 for parquet DΓA at $\Delta_\nu$5 (Fossati et al., 8 Jul 2025).

These benchmarks are used to support two claims made explicitly in the paper: first, that the cluster extension of D-TRILEX "accurately reproduces the electronic self-energy at momenta corresponding to the Fermi energy, in good agreement with the numerically exact quantum Monte Carlo solution" (Fossati et al., 8 Jul 2025); second, that it "outperforms significantly more computationally demanding approach based on the parquet approximation" (Fossati et al., 8 Jul 2025). Within the scope of the presented data, the most prominent advantage is therefore momentum-resolved self-energy accuracy near the Fermi surface together with a marked reduction of periodization ambiguity.

7. Scope, interpretation, and open direction

The method is framed as a cluster-diagrammatic approach that extends CDMFT by non-local fermion–boson diagrams. Its short-range content comes from an exact cluster solution, while its long-range content comes from dual corrections built from three-point vertices and collective charge and spin fluctuations (Fossati et al., 8 Jul 2025). This division of labor is the main organizing principle of the formalism.

A frequent misconception in the cluster-method setting is that periodization ambiguities are only a post-processing artifact. The presentation here argues for a more specific diagnosis: the explicit translational-symmetry breaking originates primarily in the CDMFT impurity construction, where intra-cluster bonds are treated exactly and inter-cluster bonds only at the DMFT level, so that $\Delta_\nu$6 in the reference problem (Fossati et al., 8 Jul 2025). The cluster D-TRILEX correction does not eliminate this source completely, but it does generate $\Delta_\nu$7 and brings intra-, inter-, and periodized self-energies closer together.

The open direction identified in the paper is therefore an outer self-consistent reformulation in which the impurity reference itself is updated using the dual-corrected quantities. The paper states that implementing such an outer loop is "a natural next step," while noting the price of recalculating two-particle vertices at each update (Fossati et al., 8 Jul 2025). This suggests a research program aimed at a fully translationally invariant cluster-diagrammatic scheme that preserves the same basic fermion–boson architecture.