--- title: 'D-TRILEX: Dual Triply Irreducible Local Expansion' url: https://www.emergentmind.com/topics/d-trilex type: topic --- # D-TRILEX: Dual Triply Irreducible Local Expansion Searching arXiv for D-TRILEX and related TRILEX papers to ground the article in the current literature. arXiv_search(query="D-TRILEX TRILEX dual triply irreducible local expansion", max_results=10, sort_by="relevance") to=arxiv_search code D-TRILEX, short for dual triply irreducible local expansion, is a self-consistent diagrammatic extension of dynamical mean-field theory built in a dual fermion-boson framework and organized around a local three-leg fermion-boson vertex rather than a local self-energy or a full local four-point vertex [2102.05425; 2204.06426]. In the literature covered here, the term denotes a family of beyond-DMFT methods that start from a local or cluster reference problem, treat strong local or short-range correlations nonperturbatively, and reconstruct nonlocal electronic self-energies and collective fluctuations through Hedin-like equations involving dual fermionic and bosonic propagators [2102.05425; 2507.06015]. The designation should be distinguished from several adjacent usages in the TRILEX literature. One paper on triangular-lattice superconductivity uses single-site TRILEX, mostly in a simplified TRILEX approximation with \(\Lambda^{\mathrm{imp},\eta}\approx 1\), and explicitly does not introduce a “D-TRILEX” formalism [1710.03467]. Another paper develops a superconducting extension of TRILEX for \(d\)-wave pairing, but does not formally adopt “D-TRILEX” as the method name; there the phrase is at most an inferred shorthand for a \(d\)-wave superconducting TRILEX extension [1705.08332]. ## 1. Conceptual definition and place within beyond-DMFT methods D-TRILEX is presented as a particularly simple yet consistent diagrammatic extension of DMFT that combines nonperturbative local physics with nonlocal charge and spin fluctuations while avoiding the explicit manipulation of the full local four-point vertex required in dual fermion, dual boson, or \(D\Gamma A\)-type approaches [2102.05425; 2204.06426]. In this construction, locality is imposed not on the lattice self-energy \(\Sigma(k,i\omega_n)\) itself, but on higher-order impurity objects such as the local three-leg vertex \(\Lambda^\varsigma_{\nu\omega}\), local susceptibilities, and local screened bosonic propagators [2211.13663; 2204.06426]. This design allows momentum-dependent self-energies and polarizations to be reconstructed diagrammatically from a local reference system and nonlocal propagators [2211.13663; 2511.14614]. Within the method landscape, D-TRILEX occupies an intermediate position. Like DMFT, it begins from a self-consistent impurity model and retains a nonperturbative description of local correlations [2102.05425; 2204.06426]. Like dual fermion and dual boson, it reformulates the problem in terms of auxiliary dual fields so that the remaining theory describes only nonlocal corrections beyond the local reference system [2102.05425]. Like TRILEX, it is organized around a three-leg fermion-boson vertex and a partially bosonized treatment of collective modes [2204.06426]. Compared with ladder \(D\Gamma A\), it avoids propagating the full local four-point vertex through ladder resummations in all frequencies and momenta, which is why the partially bosonized three-point formulation is repeatedly described as computationally cheaper [2511.14614; 2507.06015]. A further point emphasized in the dual derivation is that D-TRILEX is advertised as avoiding the Fierz-ambiguity problem that affects some channel-decomposed approaches [2507.06015; 2204.06426]. A recurring source of confusion is terminological overlap with TRILEX proper. The triangular-lattice superconductivity study on Si(111)-motivated adatom systems uses single-site TRILEX with a simplified vertex \(\Lambda^{\mathrm{imp},\eta}\approx 1\), which the authors describe as “a GW+EDMFT like scheme,” but it is not a D-TRILEX paper [1710.03467]. The superconducting TRILEX paper on the two-dimensional Hubbard model generalizes TRILEX to Nambu space and obtains \(d\)-wave superconductivity with a single-site impurity model, yet it likewise does not formally rename the method D-TRILEX [1705.08332]. By contrast, “D-TRILEX” is explicitly used as the central method name in the dual-space Hubbard-model analysis of van-Hove-driven ferromagnetic fluctuations and in the multiband dual formalism [2511.14614; 2204.06426]. ## 2. Dual partially bosonized formalism In the dual formulation, the lattice action is rewritten around a local impurity reference problem, after which Hubbard-Stratonovich transformations generate dual fermionic and bosonic fields [2102.05425]. The dual action contains bare dual propagators that are differences between lattice quantities dressed only by local impurity quantities and the impurity quantities themselves. In the single-band derivation this structure is written as \[ \tilde{\cal G}_{k\sigma} = \check G_{k\sigma}-g_{\nu\sigma}, \qquad \tilde{\cal W}^{\vartheta}_q = \check W^\vartheta_q - w^\vartheta_\omega, \] with \[ \check G^{-1}_{k\sigma} = {\cal G}^{-1}_{k\sigma} - \Sigma^{\rm imp}_{\nu\sigma}, \] \[ [\check W^\vartheta_q]^{-1} = (U^\vartheta + V^\vartheta_q)^{-1} - \Pi^{\rm imp,\vartheta}_\omega \] [2102.05425]. In the multiband implementation, the same idea is expressed in matrix form, \[ \tilde{\cal G}^{\,\sigma\sigma'}_{k,\,ll'} = \left[ \left( (\varepsilon_k - \Delta_\nu)^{-1} - g_\nu \right)^{-1} \right]^{\sigma\sigma'}_{ll'}, \] and \[ \tilde{\cal W}^{rr'}_{q,\, l_1 l_2,\, l_3 l_4} = {\cal W}^{rr'}_{q,\, l_1 l_2,\, l_3 l_4} - \bar u^{r}_{l_1 l_2,\, l_3 l_4}\delta_{rr'} \] [2204.06426]. The defining approximation of D-TRILEX is a partially bosonized representation of the local four-point impurity vertex in terms of local three-leg vertices and local screened interactions [2102.05425; 2204.06426]. In the dual derivation this is encoded in the channel-resolved decomposition \[ M^\vartheta_{\nu\nu'\omega} = \Lambda^\vartheta_{\nu\omega}\,\bar w^\vartheta_\omega\,\Lambda^{*\,\vartheta}_{\nu'\omega}, \] which is then used to approximate the exact local four-point vertex \(\Gamma\) by boson-exchange contributions in density, magnetic, and singlet channels [2102.05425]. The multiband paper gives the corresponding orbital-resolved decomposition, \[ \left[M^{\varsigma\varsigma'}_{\nu\nu'\omega}\right]^{\sigma_1\sigma_2\sigma_3\sigma_4}_{l_1 l_2 l_3 l_4} = \sum_{\{l'\} \Lambda^{\sigma_1\sigma_2\varsigma}_{\nu\omega,\,l_1,\, l_2,\, l'_1 l'_2} \, \bar w^{\varsigma\varsigma'}_{\omega,\,l'_1 l'_2,\, l'_3 l'_4} \, \Lambda^{\sigma_4\sigma_3\varsigma'}_{\nu'+\omega,-\omega,\,l_4,\, l_3,\, l'_4 l'_3} \] [2204.06426]. This replacement is the precise sense in which the method is “triply irreducible”: the basic local object retained for the nonlocal expansion is the fermion-boson three-leg vertex rather than the full four-point vertex [2102.05425; 2204.06426]. The operational D-TRILEX equations then take the form of a self-consistent fermion-boson theory in dual space. In the square-lattice D-TRILEX analysis of ferromagnetic fluctuations, the central equations are \[ \tilde{\Sigma}_{\mathbf{k}\nu} = - \sum_{\mathbf{q},\omega,\varsigma} \Lambda^{\varsigma}_{\nu\omega} \, \tilde{G}_{\mathbf{q}+\mathbf{k},\nu+\omega,\sigma} \, \tilde W_{\mathbf{q}, \omega}^{\varsigma} \, \Lambda^{\varsigma}_{\nu+\omega,-\omega}, \] \[ \tilde{\Pi}_{ \mathbf{q} \omega}^\varsigma = 2 \sum_{\mathbf{k},\nu} \Lambda_{\nu+\omega,-\omega}^\varsigma \, \tilde{G}_{\mathbf{k}\nu} \, \tilde{G}_{\mathbf{k}+\mathbf{q}, \nu+\omega} \, \Lambda_{\nu\omega}^\varsigma \] [2511.14614]. The dual Dyson equations are \[ \tilde{G}_{\mathbf{k}\nu}=(\tilde{\mathcal{G}_{\mathbf{k}\nu}^{-1}-\tilde{\Sigma}_{\mathbf{k}\nu})^{-1}, \] \[ \tilde W_{\mathbf{q}\omega}^{\varsigma}=(\tilde{\mathcal{W}_{\mathbf{q}\omega}^{\varsigma\, -1}-\tilde{\Pi}^\varsigma_{\mathbf{q}\omega})^{-1} \] [2511.14614]. The multiband version adds orbital and site indices to these same structures [2204.06426]. ## 3. Reference systems, impurity quantities, and self-consistency The impurity or reference problem is central because it supplies the local Green’s function \(g_\nu\), local self-energy \(\Sigma^{\rm imp}\), local susceptibilities, local polarization, and especially the local three-leg vertex \(\Lambda\) [2204.06426; 2511.14614]. In the multiband formulation the reference action is written with fermionic hybridization \(\Delta_{\nu,ll'}\) and optionally bosonic hybridizations \(Y^r_\omega\), so the framework can be based on a DMFT or an EDMFT-type local problem [2204.06426]. In practice, many of the calculations discussed in the paper use a DMFT impurity as reference [2204.06426]. The self-consistency loop in D-TRILEX is diagrammatic but local-input-driven. Starting from impurity quantities, one constructs the bare dual propagators, computes the dual polarization from a \(\Lambda \tilde G \tilde G \Lambda\) bubble, obtains the dressed dual interaction, computes the dual self-energy from exchange of dressed dual bosons, updates the dressed dual Green’s function, and iterates to convergence [2204.06426; 2511.14614]. The multiband paper states the convergence criterion as the relative Frobenius norm \[ F = \frac{\|\tilde G_n-\tilde G_{n-1}\|}{\|\tilde G_{n-1}\|}, \] with convergence reached when \(F<\delta\) [2204.06426]. Mixing is used for numerical stabilization in both fermionic and bosonic sectors [2204.06426]. Physical lattice quantities are reconstructed from the dual ones through exact dual-to-lattice relations. For the lattice self-energy, the single-band dual derivation gives \[ \Sigma^{\rm latt}_{k\sigma} = \Sigma^{\rm imp}_{\nu\sigma} + \frac{\tilde\Sigma_{k\sigma}}{1+g_{\nu\sigma}\tilde\Sigma_{k\sigma}}, \] while the multiband paper uses the matrix form \[ \Sigma_{k, ll'} = \Sigma^{\rm imp}_{\nu,ll'} + \sum_{l_1} \tilde{\Sigma}_{k,l l_1} \left[\left(\mathbb{1} + g_\nu \cdot \tilde{\Sigma}_k \right)^{-1} \right]_{l_1 l'} \] [2102.05425; 2204.06426]. The physical susceptibility is reconstructed from the lattice polarization through \[ \left[\left(X^{\varsigma}_{q}\right)^{-1}\right]_{l_1 l_2,\, l_3 l_4} = \left[\left(\Pi^{\varsigma}_{q}\right)^{-1}\right]_{l_1 l_2,\, l_3 l_4} - \left[U^{\varsigma}+V^{\varsigma}_{q}\right]_{l_1 l_2,\, l_3 l_4} \] [2204.06426]. One practical implication stressed in the multiband paper is that the divergence of the physical susceptibility coincides with the divergence of the dual renormalized interaction \(\tilde W_q^\varsigma\), which is why critical fluctuations can feed back strongly onto the electronic self-energy [2204.06426]. A methodological subtlety appears in the choice of reference problem. In the self-consistent D-TRILEX analysis of the Hubbard model, updating the impurity hybridization to satisfy the standard dual condition \(\sum_{\mathbf k}\tilde G_{\mathbf k\nu}=0\) can substantially improve results when DMFT is not the best local starting point [2102.05425]. In the cluster extension, a further issue is translational symmetry breaking inherited from CDMFT-like reference problems; there the authors argue that the reference cluster itself is the main source of periodization ambiguity and propose a rotated-basis cluster reference with diagonal hybridization to improve the starting point for the dual expansion [2507.06015]. ## 4. Single-site, multiband, and cluster realizations The literature represented here includes several realizations of the same core idea. The multiband paper generalizes D-TRILEX to systems with multiple orbitals and several atoms in the unit cell, with local Coulomb tensors, channel-dependent long-ranged interactions, and matrix-valued Green’s functions, self-energies, susceptibilities, polarizations, and three-leg vertices [2204.06426]. The combined orbital-site index \(l\) and the channel-resolved interaction tensors \(U^d\), \(U^m\), \(U^s\), and \(U^t\) define the formal architecture that makes the method applicable to extended Hubbard, Hubbard-Kanamori, and bilayer models [2204.06426]. This implementation is explicitly able to account for frequency- and channel-dependent long-ranged electronic interactions [2204.06426]. The cluster-diagrammatic extension replaces the single-site reference problem by a cluster reference system so that short-range correlations are treated exactly within the cluster and long-range collective fluctuations are added diagrammatically beyond it [2507.06015]. In the benchmark one-dimensional nano-ring study, the authors use a dimer cluster and transform to a bonding-antibonding basis that diagonalizes the local part of the single-particle Hamiltonian, \[ \mathcal{R} = \frac{1}{\sqrt{2}} \begin{pmatrix} 1 & 1\\ 1 & -1 \end{pmatrix}, \] thereby allowing a diagonal hybridization function in the impurity problem and generating the off-diagonal self-energy diagrammatically [2507.06015]. This rotated-basis construction is justified because the local average of the dispersion provides the leading contribution to the rotated hybridization [2507.06015]. The cluster formalism retains the same dual structure, with \[ \tilde{\mathcal G}^{ll'}_{K\nu} = \hat G^{ll'}_{K\nu} - \delta_{ll'} g^{ll}_\nu \] and a Hedin-like dual polarization \[ \tilde{\Pi}_{Q,\omega,\varsigma}^{l_1 l_2;\, l_7 l_8} = 2\sum_{\bf k,\nu,\{l\}} \Lambda^{\,l_4,l_3;\,l_2l_1}_{\nu+\omega,-\omega,\varsigma} \, \tilde G_{K,\nu}^{l_3l_5} \, \tilde G_{K+Q,\nu+\omega}^{l_6l_4} \, \Lambda^{\,l_5,l_6;\,l_7l_8}_{\nu,\omega,\varsigma} \] [2507.06015]. The cluster study is also relevant because it sharpens what D-TRILEX is not. It is not merely cluster DMFT with periodization, because inter-cluster self-energy is generated diagrammatically rather than set to zero [2507.06015]. It is likewise not a parquet method, because it still avoids explicit four-point-vertex parquet machinery and Bethe-Salpeter equations in frequency space [2507.06015]. The authors therefore describe it as a hybrid between cluster DMFT and a dual boson/TRILEX-type Hedin-like extension [2507.06015]. ## 5. Benchmarking, accuracy, and physical regimes A central contribution of the dual-theory analysis of collective fluctuations is to clarify why D-TRILEX can work well despite discarding large parts of the full local four-point vertex [2102.05425]. By comparing ladder dual fermion, exact dual diagrammatic Monte Carlo, diagrammatic Monte Carlo for the partially bosonized dual theory, and D-TRILEX, the paper concludes that the components of the local four-point vertex not represented by the partially bosonized approximation have only a minor effect on the electronic self-energy in a broad range of parameters [2102.05425]. At \(\beta=2\) for the half-filled square-lattice Hubbard model, the normalized deviation from the dual DiagMC reference is only \(\delta=2\%\) at \(U=2\) and \(\delta=3\%\) at \(U=4\) for D-TRILEX, while the largest discrepancy appears around \(U=8\), where \(\delta \approx 18\%\) and antiferromagnetic fluctuations are strongest [2102.05425]. The interpretation offered is that in regimes where ladder dual fermion is accurate, the self-energy is dominated by longitudinal bosonic modes, whereas transverse particle-hole and particle-particle contributions largely cancel [2102.05425]. This picture also identifies the method’s main limitation. In the weak-coupling Slater regime close to antiferromagnetic instability, the neglected transverse and anharmonic fluctuation contributions become important, and D-TRILEX can miss pseudogap formation or place it at too low a temperature [2102.05425]. At \(U=2\), the estimated antinodal pseudogap onset temperature is \(T_*^{AN}=0.065\) in exact DiagMC, \(0.059\) in ladder dual fermion, and only \(0.050\) in D-TRILEX [2102.05425]. At \(U=4\), however, the agreement improves and the self-consistent version captures the nodal/antinodal differentiation much more accurately [2102.05425]. This suggests that the method becomes more reliable when magnetism is more local and harmonic rather than weak-coupling and strongly itinerant [2102.05425]. The multiband benchmarks reinforce this pattern. For the Hubbard-Kanamori dimer, D-TRILEX nearly coincides with exact diagonalization for the Green’s function and yields an average energy error of about \(1.1\%\) at \(J=0\), decreasing with \(J\) [2204.06426]. At half filling and moderate interaction, where DMFT fails because the self-energy develops strong nonlocal real parts, D-TRILEX remains close to exact diagonalization [2204.06426]. In the extended Hubbard model on the square lattice, comparison with DiagMC@DB shows almost perfect agreement at \(U=2\), good agreement at \(U=4\), and a clear deterioration at \(U=6\), especially in \(\mathrm{Im}\,\Sigma\), again consistent with the breakdown of ladder-like approximations in strongly fluctuating regimes [2204.06426]. The cluster nano-ring benchmark further illustrates the method’s selectivity. At the Fermi momentum \(k=\pi/2\) in metallic \(N_c=4\) and \(N_c=8\) rings, exact QMC shows insulating low-frequency behavior, single-site and cluster DMFT remain too metallic, parquet \(D\Gamma A\) performs surprisingly poorly for \(N_c=4\), and D-TRILEX gives the best low-frequency self-energy among the approximate methods [2507.06015]. Away from the Fermi energy at \(k=0\), by contrast, parquet \(D\Gamma A\) is more accurate, and D-TRILEX can even show noncausal low-frequency behavior for the smallest rings, which the authors attribute to the omission of particle-particle correlations and other missing diagram classes [2507.06015]. This suggests that D-TRILEX is especially adapted to low-energy particle-hole fluctuation physics near the Fermi surface, but less complete away from that regime [2507.06015]. ## 6. Applications and physical phenomena The method has been applied to a broad range of correlated-electron problems. In pyrochlore iridates, TRILEX rather than D-TRILEX is used, but the study is still instructive because it demonstrates the core philosophy of reconstructing nonlocal self-energy effects from a local three-leg vertex [2211.13663]. For \(\mathrm{Y_2Ir_2O_7}\), single-site DMFT gives a direct transition from a paramagnetic metal to an all-in/all-out antiferromagnetic insulator, whereas TRILEX produces strong momentum-dependent self-energy structure near the transition and yields evidence for a Weyl semimetal or at least a Weyl metal phase [2211.13663]. The critical interaction is estimated around \(U\approx 1.55\) eV in TRILEX versus around \(U\sim 1.6\) eV in single-site DMFT, and the momentum variation of the self-energy becomes two orders of magnitude larger between \(U=1.2\) eV and \(U=1.4\) eV [2211.13663]. The implication is that local-vertex diagrammatic extensions can qualitatively alter topological low-energy physics [2211.13663]. The 2025 D-TRILEX study of the square-lattice Hubbard model with \(t'/t=-0.45\) applies the fully self-consistent dual formalism to the regime \(0.4