Time-evolving matrix product operators for off-diagonal system-bath coupling

Chu Guo; Ping-Xing Chen; Ruofan Chen; Tian Jiang; Wei Wu; Xiansong Xu

Time-evolving matrix product operators for off-diagonal system-bath coupling

T0 review · 4 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper extends the TEMPO tensor-network method to off-diagonal system-bath couplings, where the system operator may be non-Hermitian, by building the Feynman-Vernon influence functional as a matrix product operator.

desk verdict Solid extension of TEMPO to off-diagonal couplings with sound benchmarks; the main gap is an untested n=20 hybridization fit and a missing analytic benchmark for the flagship application. read the letter →

arxiv 2604.01556 v3 pith:LWUVMYMP submitted 2026-04-02 cond-mat.mes-hall

Chu Guo , Wei Wu , Xiansong Xu , Tian Jiang , Ping-Xing Chen , Ruofan Chen This is my paper · ORCID

classification cond-mat.mes-hall

keywords time-evolvingmatrixproductoperatorsoff-diagonalsystem-bathcouplingFeynman-Vernoninfluencefunctionalprocesstensorspin-bosonmodelJaynes-Cummingssub-ohmicbathsecularapproximation

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to generalize TEMPO, a tensor-network path-integral method for quantum impurity problems, from diagonal system-bath couplings to off-diagonal couplings of the form sum_k (V_k A b_k^dagger + H.c.) with A possibly non-Hermitian. The key move is to treat the discretized Feynman-Vernon influence functional as the thermal state of an effective one-dimensional quantum Hamiltonian and build it as a matrix product operator, so the construction depends only on the Trotter decomposition and the hybridization function, not on the details of the system operator. The authors validate the method on two exactly solvable models and then use it to show that the rotating-wave (secular) approximation can badly fail for a spin coupled to a sub-ohmic bath, even at weak coupling. A sympathetic reader would care because this supplies a unified route to a broad class of bosonic impurity problems, including the type central to bosonic dynamical mean-field theory.

What carries the argument

The central object is the MPO-IF: a matrix product operator representation of the discretized Feynman-Vernon influence functional I[A†,A] = exp(−∫dτ∫dτ′ A†(τ)Δ(τ,τ′)A(τ′)). The construction interprets the discretized exponent as the Hamiltonian of an effective one-dimensional system at inverse temperature 1, builds e^{−H_eff/2^m} with a short-step exponential MPO algorithm, and squares it m times via exponential thermal tensor-network renormalization. A crucial ingredient is the exponential-sum approximation of the hybridization function (n=20 terms in this work), whose parameters enter the MPO site tensor and keep the bond dimension compact. Appendix B supplies the equal-time operator-order

What would settle it

Apply the algorithm to a bath with two widely separated narrow spectral peaks, so the hybridization function clearly needs more than the n=20 exponential terms, and compare the predicted spin dynamics against exact diagonalization; if the results fail to converge as bond dimension and XTRG steps increase, the exponential-sum representation is the cause.

Watch

Extended reading notes

Core claim

The central claim is that the Feynman-Vernon influence functional for off-diagonal system-bath coupling, after QuAPI discretization on the Keldysh contour, is the exponential of an effective long-range quantum Hamiltonian, and therefore can be represented as a matrix product operator using exponential thermal tensor-network techniques. This MPO acts as the process tensor of the impurity, and contracting it with system propagators yields all multi-time correlations. Because the construction rests only on the Trotter decomposition of the influence functional and not on the specifics of the coupling operator, it automatically contains vanilla TEMPO and its diagonal and non-commutative extension

Load-bearing premise

Everything rests on the assumption that a bath's effect can be captured by about twenty exponential decay terms and that the resulting tensor network stays compact; if a real bath needs more terms or generates more entanglement than the examples shown, the method's efficiency and accuracy are not guaranteed.

Editorial extensions

If this is right

Any bosonic impurity with linear coupling to a noninteracting bath, Hermitian or not, can in principle be solved by the same influence-functional MPO, removing the diagonal-coupling restriction of earlier TEMPO variants.
The unified construction provides a practical impurity solver for bosonic dynamical mean-field theory in the normal phase with a scalar hybridization function.
The numerical comparison indicates that the rotating-wave approximation can be quantitatively unreliable for sub-ohmic baths even in the weak-coupling regime, so conclusions drawn from RWA spin-boson models should be rechecked.
The same strategy suggests a direct fermionic extension, since the discretized influence functional for fermions has the same exponential structure, a direction not yet explored in the paper.
The explicit preservation of time-translational invariance opens the door to infinite-time and steady-state versions of the method for bosonic problems.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Editorial inference: The equal-time operator-ordering correction in Appendix B is likely essential for any non-commuting system operator, and its necessity could be tested by benchmarking against pseudo-mode or chain-mapping methods on the same model.
Editorial inference: The method's exponential-sum approximation of the hybridization function suggests that spectral functions with multiple sharp peaks or very low-frequency structure may require more than 20 terms, so the claimed n=20 sufficiency should be checked case by case.
Editorial inference: Because the process tensor encodes the bath's influence on the system, the same MPO-IF could be recycled to study different system Hamiltonians or different initial system states without recomputation, a practical advantage the paper notes but does not demonstrate numerically.
Editorial inference: The demonstrated failure of the secular approximation for sub-ohmic baths suggests that previous spin-boson studies using the rotating-wave approximation in structured or sub-ohmic environments may need to be revisited, though this is an inference beyond the paper's own claims.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Good paper. The extension of TEMPO to off-diagonal system-bath couplings is real, and the construction is more general than anything else in the literature they cite: it works for non-Hermitian A, reduces to the diagonal cases, and is based only on the Trotter decomposition of the Feynman-Vernon IF. The derivation in Appendix A is a standard Gaussian cumulant expansion, done carefully, and the equal-time ordering corrections in Appendix B look right. The two benchmarks — JC model with a single mode and a noninteracting bosonic mode coupled to a sub-ohmic bath — are convincing. The code is on GitHub, so the method is checkable.

The soft spots are proportionate. The n=20 exponential-sum fit to the hybridization (Eq. C1) is a free parameter that is never tested. The δt and χ convergence studies in Figs. 8-9 validate Trotter error and bond truncation, but not the fitting error. For the main physical application, the JC spin-boson model, the paper itself states the single-excitation sector is analytically solvable. Instead of comparing to that solution, Fig. 7 compares to standard TEMPO, which at α=0.12 is not fully converged even at the largest χ and smallest δt they used. That means the observed revivals and the claim that the secular approximation is very poor for structural baths are not pinned down by a controlled benchmark. This is not a fatal flaw: the method is validated on other solvable cases, and the weak-coupling (α=0.01) result is likely robust since both codes converge there. But it is an addressable omission, and the authors should close it before publication. A second, minor point: the generic site tensor in Eq. C2 does not explicitly show the equal-time ordering corrections for arbitrary A; the tested cases avoid the issue because for σ± and a the correction reduces to A†A plus a constant. Worth a sentence.

The reader's conditional verdict is about right. The stress-test note is fair: the n=20 fit is an uncontrolled approximation in the current manuscript. But it does not undercut the method's correctness, given the two clean ED benchmarks and the code release.

This paper deserves a serious referee. It is a real advance for the TEMPO community and a plausible building block for BDMFT impurity solvers. The referee should ask for the single-excitation analytic benchmark and an n-convergence test, not for a rewrite from scratch.

Referee Report

4 major / 4 minor

Summary. The paper extends the time-evolving matrix product operator (TEMPO) method to bosonic quantum impurity problems with off-diagonal system-bath coupling of the form H_hyb = sum_k (V_k A b_k^\dagger + H.c.), where A may be non-Hermitian. The authors derive the Feynman-Vernon influence functional on the Keldysh contour for this general coupling, discretize it, and represent it as an MPO built with the exponential thermal tensor renormalization group (XTRG) algorithm using an exponential-sum approximation of the hybridization function. They validate the method against exact diagonalization in two solvable cases: the Jaynes-Cummings model with a single bosonic mode and a noninteracting bosonic impurity coupled to a sub-ohmic bath. As a physical application, they compute the real-time spin dynamics for a Jaynes-Cummings-type spin-boson model with a sub-ohmic bath and compare it with the standard spin-boson model, concluding that the secular (rotating-wave) approximation can be very poor even at weak coupling. The paper also claims generality over earlier TEMPO variants and suggests a fermionic extension.

Significance. If the central claims hold, this is a useful and fairly general extension of TEMPO: it unifies previously separate treatments of diagonal/commutative and diagonal/non-commutative couplings, and it removes the need to discretize the bath or truncate its local Hilbert space. Two strengths deserve emphasis: (i) the method is validated against external exact-diagonalization benchmarks in two distinct settings, including a continuous sub-ohmic bath; and (ii) the authors provide a public code repository, which makes the numerical results reproducible and the remaining concerns checkable. The potential payoff for bosonic dynamical mean-field theory and for simulating JC-type impurity models gives the work significance beyond a purely incremental algorithmic step. However, as detailed below, some written parts of the derivation and the advertised scope contain inconsistencies that need to be resolved before the paper is publication-ready.

major comments (4)

[Appendix B and Appendix C, Eq. (C2)] The equal-time ordering correction derived in Appendix B (Eqs. B7-B10) replaces the naive term e^{-A†_+ Δ^{++}_{j,j} A_+} by an exponent containing both A†A and A A† with different coefficients. The generic site tensor in Eq. (C2), however, contains only a single local operator Δ0 A†A. For arbitrary non-Hermitian A at finite β, A A† is not proportional to A†A, so the MPO-IF as written does not implement the correction. The examples avoid the issue (β=∞, or low-T with the AA† coefficient exponentially small), but the claim in Sec. III.A that the method applies to any linearly-coupled noninteracting bath is then too strong. Please specify how the two ordering terms are encoded in the MPO, or state the restrictions.
[Appendix B, Eqs. (B5), (B8)] The factor (e^{-βω}-1)^{-1} appears in Δ^{+-}_{j,k} and in the equal-time A A† term. The thermal occupation of the bath is n_B(ω)=(e^{βω}-1)^{-1}, which vanishes at zero temperature. The printed factor instead tends to -1 as β→∞, implying a spurious nonzero contribution from ⟨b†b⟩=0 in the vacuum state. If the formulas as written are used to construct the IF, results at β=∞ would be incorrect; presumably the code uses the correct sign, but the appendix must be corrected or the convention explained.
[Abstract vs. Sec. IV] The abstract states that the paper "study[ies] the imaginary-time evolution of a bosonic impurity with nonzero on-site interaction that is coupled to a sub-ohmic bath." No such imaginary-time simulation appears in Section IV, which contains only real-time results (JC model, free bosonic impurity, JC spin-boson model). Either add the promised imaginary-time study or remove the sentence from the abstract.
[Sec. IV.C, Fig. 7] The JC spin-boson model is stated to be analytically solvable in the single-excitation sector (β=∞, ρ_S=|e⟩⟨e|). Fig. 7 compares only against standard TEMPO, which itself is not fully converged at α=0.12. A direct comparison with the exact single-excitation solution would provide a stronger benchmark for the new MPO-IF and would directly validate the physical conclusion that the secular approximation is very poor. The existing ED benchmarks for the same spectral function in Sec. IV.B reduce but do not eliminate this concern.

minor comments (4)

[Sec. III.D] The phrase "communication relation" should be "commutation relation".
[Appendix C, Eq. (C1)] The exponential-sum approximation with n=20 is asserted as sufficient without a fit-error plot. A small panel showing the maximum relative error of the fitted hybridization function over the relevant time/frequency range would make the fixed n=20 choice transparent and would directly address the n-convergence question.
[Sec. IV.A, Eq. (16)] The JC Hamiltonian uses coefficients 2Ω and 2λ without explanation. A sentence connecting this convention to the more standard notation λ and Ω would help readers.
[Sec. II, Eq. (5)] The symbol ̃Tr for the modified trace of the process tensor is used before being explicitly defined. Please define it at first use.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the method is derived from the Feynman-Vernon influence functional and validated against external exact benchmarks; self-citations and numerical truncation caveats do not make the derivation circular.

full rationale

I found no load-bearing step in which a prediction reduces to an input by construction. The central algorithm (Sec. III.B-C and Appendices A-C) starts from the exact continuous-time Feynman-Vernon influence functional, Eq. (6), for off-diagonal coupling, and constructs the MPO-IF from the discretized effective Hamiltonian, Eq. (9)-(10), using XTRG. The inputs are H_S and the hybridization function Delta, not the target observables. Validation is external and parameter-free: the JC model is compared with exact diagonalization (Fig. 5), and the noninteracting bosonic case is compared with a separately converged Gaussian-state ED calculation (Fig. 6). The claim that the method reduces to vanilla TEMPO is shown algebraically by setting A^dagger = A and recovering the classical-Hamiltonian form of the IF (Sec. III.D), not by citing earlier work. Self-citations, including Ref. [47] for the time-translational invariant IF construction and Ref. [50] for GTEMPO, are algorithmic antecedents but are not used as the justification for the central derivation, which is self-contained and benchmarked. The unquantified numerical truncations — the n=20 exponential-sum approximation of Eq. (C1), the m=7 XTRG step, and the absence of a check against the analytically solvable single-excitation sector in Sec. IV.C — are accuracy/validation limitations, not circularity. Accordingly, the score is 1 rather than 0 only to register the presence of minor self-citations and heuristic truncation choices; there is no fitted-input-called-prediction or definitional reduction.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new physical entities or fitted physical constants. The only hand-chosen parameters are algorithmic truncations (n exponential terms, m XTRG steps) that affect the accuracy of the MPO-IF compression. The central assumptions are the standard QIP conditions: noninteracting Gaussian bath, linear coupling, separable initial state, and first-order Trotter slicing.

free parameters (2)

n (number of exponential terms in hybridization fit) = 20
The hybridization function is approximated as a sum of exponentials (Eq. C1); n=20 is stated to be sufficient for all studied cases, but is an algorithmic truncation chosen by hand rather than derived from the problem.
m (XTRG step halving exponent) = 7
The XTRG construction of the thermal state e^{-H_eff} uses step size 1/2^m; m=7 is chosen for convergence. This is a numerical control parameter that affects the accuracy of the MPO-IF.

assumptions (3)

domain assumption The bath is noninteracting (free bosons) and initially in a thermal state, so the interaction-picture cumulant expansion truncates at second order (Gaussian bath).
Used throughout the Feynman-Vernon derivation in Sec. II.B and Appendix A; if the bath is interacting or non-Gaussian, Eq. (6) is invalid.
domain assumption The system-bath coupling is linear in bath creation/annihilation operators (H_hyb = sum_k (V_k A b_k^dagger + H.c.)).
This is the defining assumption of the QIP class studied; nonlinear couplings would produce a different influence functional.
standard math First-order Trotter decomposition of the total Liouvillian is sufficient; discretization error is controlled by δt and the IF is exact in the continuous-time limit.
Used in Sec. III.C to discretize the Keldysh contour; benchmarks show convergence with decreasing δt.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Time-evolving matrix product operators for off-diagonal system-bath coupling." pith.science (2026). https://pith.science/paper/LWUVMYMP

@misc{pith2026260401556,
  author       = {Pith},
  title        = {Pith review of: Time-evolving matrix product operators for off-diagonal system-bath coupling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LWUVMYMP}},
  note         = {Machine review of arXiv:2604.01556}
}

abstract

The time-evolving matrix product operator (TEMPO) method has proven to be a powerful method to study the long-time dynamics of bosonic impurity problems where a small system is linearly coupled to a noninteracting bosonic bath. However, current developments of TEMPO have mostly focused on the case of diagonal system-bath coupling, i.e., $\sum_k \Aop(V_k \bdop_k + \hc)$, with $\Aop$ a Hermitian operator of the system. Based on the process tensor framework, we extend TEMPO to the more general case of off-diagonal system-bath coupling in the form $\sum_k (V_k\Aop\bdop_k + \hc)$, where $\Aop$ could be non-Hermitian. As applications, we study the real-time dynamics of a spin that is coupled to a sub-ohmic bath via the Jaynes-Cummings-type system-bath coupling and compare it against the standard spin-boson model, where we show that the commonly used rotating-wave approximation could be very poor for this bath. We also study the imaginary-time evolution of a bosonic impurity with nonzero on-site interaction that is coupled to a sub-ohmic bath, to illustrate the flexibility of our method. Our method provides a unified framework to understand different variants of TEMPO, and is a promising building block for an impurity solver in the bosonic dynamical mean field theory for the normal phase with a scalar hybridization function.