Pith. sign in

REVIEW 3 major objections 6 minor 43 references

Time-nonlocal versus time-local long-time extrapolation of non-Markovian quantum dynamics

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper argues that long-time extrapolation of non-Markovian open quantum systems can be done with a simple, stationary time-local map instead of the memory-laden transfer tensor method, and that the simpler scheme converges at least…

desk verdict Useful numerical comparison with an overstrong abstract: time-local extrapolation often beats TTM, but the paper's own sub-ohmic data contradict 'invariably.' read the letter →

arxiv 2505.21017 v1 pith:5MIMBUEN submitted 2025-05-27 quant-ph

Moritz Cygorek , Erik M. Gauger This is my paper
classification quant-ph
keywords openquantumsystemsnon-Markoviandynamicstransfertensormethodtime-localdynamicalmapslong-timeextrapolationspin-bosonmodeltime-convolutionlessmasterequationsPT-MPO
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

Simulating a small quantum system coupled to a structured environment is expensive because the environment remembers the past, so researchers often learn the short-time dynamics and then extrapolate to long times. The transfer tensor method is the standard tool for this, building a time-nonlocal propagator from short-time dynamical maps. This paper claims that a much simpler time-local recipe works at least as well: invert the short-time dynamical maps to obtain a step-to-step map, wait until that map stops changing, and then apply the fixed map repeatedly. The authors test this on the spin-boson model with sub-ohmic, ohmic, and super-ohmic spectral densities and find that the time-local extrapolation converges at least as fast as the transfer tensor method, sometimes orders of magnitude better. If this is right, practitioners can skip the memory-kernel machinery entirely and use one stationary local map.

What carries the argument

The central object is the time-local dynamical map $\mathcal{E}_{t+\Delta t,t} = \mathcal{E}_{t+\Delta t,t_0} \mathcal{E}_{t,t_0}^{-1}$, obtained by inverting the short-time propagator from the initial time. The argument is that, for a time-independent Markovian embedding of the open system, all but $M \le D^2$ eigenmode contributions decay or interfere away after a transient, so this map becomes time-independent as $\mathcal{E}_{t+\Delta t,t} \to \mathcal{E}_s$ while the reduced density matrix is still evolving. The extrapolation then applies this fixed map indefinitely, with the cutoff $\tau_c$ chosen after the map becomes nearly stationary and after any singularities of $\mathcal{E}_{t,t_0}$ have passed.

What would settle it

Compute the Frobenius norm $\|\mathcal{E}_{t+\Delta t,t} - \mathcal{E}_{t,t-\Delta t}\|/\Delta t$ for a strong-coupling sub-ohmic spin-boson model with a small cutoff frequency $\omega_c$; if this norm has not decayed well below its short-time values by the time the observable $\langle \sigma_z \rangle$ stops changing, the claimed stationarity-before-equilibrium window does not exist.

Watch

Extended reading notes

Core claim

The central claim is that time-dependent time-local dynamical maps become stationary long before the open quantum system itself reaches its steady state, so long-time dynamics can be extrapolated by repeated application of a constant local map. In the spin-boson model, the time-local extrapolation defined by $\rho_{t_{n+1}} = \mathcal{E}_s \rho_{t_n}$ with $\mathcal{E}_s$ taken from the short-time propagators converges at least as quickly as the time-nonlocal transfer tensor method. The paper presents this as evidence that time-nonlocality is not a prerequisite for accurate and efficient long-time extrapolation of non-Markovian quantum dynamics.

Load-bearing premise

The whole scheme rests on the assumption that, after a short transient, the step-to-step map of the open system stops changing while the system itself is still far from equilibrium, and that the map inversion does not run into a singularity before that happens.

Editorial extensions

If this is right

  • For common spin-boson models, short-time propagation up to the memory time suffices to extrapolate accurately to times far beyond the cutoff, with errors comparable to or smaller than those of the transfer tensor method.
  • The recipe for safe use is concrete: examine the singular values of the short-time dynamical maps, avoid cutoffs near singularities, and monitor the difference $\|\mathcal{E}_{t+\Delta t,t} - \mathcal{E}_{t,t-\Delta t}\|$ to verify that the map has become stationary.
  • Time-local dynamical maps give direct access to canonical Lindblad rates, and persistent negative stationary rates indicate that the system remains non-Markovian even at long times.
  • The transfer tensor method remains useful for other tasks such as tomography, classification of open systems, and reconstruction of memory kernels, even if it is not needed for simple long-time extrapolation.
  • If a time-local map becomes stationary but the system has not equilibrated, the same data can be used to build time-local master equations and to diagnose the flow of information between system and environment.

Reading between the lines

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

  • A testable extension would be to apply the same stationarity check to time-dependent driving: if the map settles to a periodic rather than constant form, a Floquet-like extension of the time-local extrapolation could handle driven systems.
  • The stationarity-before-equilibrium property suggests a general diagnostic for any open-system solver: compute the difference between consecutive time-local maps and use its decay as a convergence criterion, independent of the specific extrapolation method.
  • If the stationarity window is generic, then any numerically exact short-time method, not just process-tensor solvers, could be upgraded to a long-time solver by a trivial postprocessing step that requires no memory-kernel fitting.
  • The observed failure near singularities of $\mathcal{E}_{t,t_0}$ suggests a practical rule of thumb: extrapolation should begin only after the last singularity in the singular-value spectrum, since the time-local map is otherwise ill-defined.
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.

Referee Report

3 major / 6 minor

Summary. This paper compares two schemes for extrapolating short-time non-Markovian quantum dynamics to long times: the time-nonlocal transfer tensor method (TTM) and a time-local scheme based on the observation that time-dependent time-local dynamical maps become stationary well before the system reaches steady state. The authors derive the time-local extrapolation from an assumed Markovian embedding of the system, and then test both methods on driven spin-boson models with sub-ohmic, ohmic, and super-ohmic spectral densities, using the ACE process-tensor code as a numerically exact benchmark. The central claim, stated in the abstract and repeated in the introduction, is that the time-local extrapolation 'invariably converges at least as fast as' TTM, and that time-nonlocality is not a prerequisite for accurate long-time extrapolation. The numerical comparisons use identical short-time data for both methods and validate against an independent solver, so the comparisons are fair; however, the extent to which the stated claim is supported by the results is limited by the sub-ohmic example, where time-local extrapolation is erratic for cutoff times near a singularity of the dynamical map.

Significance. If the findings are robust, they have clear practical value: practitioners using TTM-style extrapolation could instead apply a stationary time-local map once the map has settled, avoiding the need to construct and store a sequence of transfer tensors. The paper is commendable for its clean numerical methodology: both extrapolations start from the same short-time dynamical maps, no free parameters are fitted, and the long-time benchmark is an independent numerically exact calculation. The claim that time-local extrapolation can outperform TTM in certain regimes is well supported by the examples. However, the manuscript's headline claim of 'invariably converges at least as fast' is not supported by the presented evidence, and the theoretical justification for stationarity rests on a number of unverified assumptions. The paper therefore contains a useful observation and a practical proposal, but the central claim requires substantial qualification before publication.

major comments (3)
  1. [Abstract and Sec. III.A, Fig. 1(b)] The abstract's claim that time-local extrapolation 'invariably converges at least as fast as time-nonlocal extrapolation' is contradicted by the paper's own sub-ohmic results. In Fig. 1(b), for cutoff times around τc ≈ 12, the time-local extrapolation is 'erratic and unreliable' and its error is orders of magnitude larger than that of TTM; TTM gives a finite, moderate error while the time-local error becomes very large or divergent. The unconditional claim is therefore false for finite cutoff times. The paper should either replace 'invariably' with a qualified statement, e.g. for cutoff times beyond the last singularity of E_{t,0}, or explicitly state the asymptotic interpretation τc → ∞. This is a load-bearing issue because the abstract and Sec. IV present the unqualified statement as the main conclusion.
  2. [Sec. II.B, Eqs. (4)-(6)] The stationarity argument relies on several assumptions that are not established for the systems treated: the existence of a time-independent Markovian embedding, diagonalizability of the extended Liouvillian, and linear independence of the M projected dual vectors with M ≤ D². Footnote 34 correctly allows for exceptional points via generalized eigenvectors, but the more serious problem is that Eq. (4) requires E_{t_n,t_0} to be non-singular, and Fig. 1(c) shows that this condition is violated at isolated intermediate times in the sub-ohmic example. The paper notes this and advises checking for singularities, but the theoretical derivation in Sec. II.B does not incorporate this limitation. The manuscript should state explicitly that Eq. (4) and the stationarity argument apply only when E_{t_n,t_0} is invertible, and that the derivation does not prove stationarity before the last singularity.
  3. [Sec. III.A, Fig. 1(d) and Sec. IV] The paper's own recommended operational criterion is that time-local extrapolation should be applied only when the dynamical map is nearly stationary, monitored by |E_{t+Δt,t} - E_{t,t-Δt}|, and that cutoff times near singularities should be avoided. This is a sensible practical guideline, but it is not the same as the abstract's 'invariably' claim. The manuscript should reconcile the practical recommendation with the abstract wording. In particular, the statement in Sec. IV that time-local extrapolation 'can be recommended unreservedly for cutoff times beyond which the dynamical maps are nearly stationary' is conditional and should be reflected in the abstract and introduction.
minor comments (6)
  1. [Abstract and Sec. I] There are several typos: 'prerequiste' should be 'prerequisite', 'simulation open qunatum systems dynamics' should be 'simulate open quantum system dynamics', and 'indepedent' should be 'independent'. The introduction also uses 'typically' where the abstract uses 'invariably'; the wording should be made consistent throughout.
  2. [Sec. III.A, Fig. 1 caption] The caption for Fig. 1(b) states 'Difference between extrapolated (from cutoff time τc) and exact value of ⟨σz⟩ at time t = 80.' This is clear, but the vertical axis label in the figure is given as '|extr.−exact|' without stating the observable; please clarify in the caption that the error is for ⟨σz⟩.
  3. [Sec. III.B, Fig. 3 caption] The caption says 'The extrapolation error in (b) is evaluated with respect to reference time t = 100 ps^{-1}.' Since the horizontal axis of panel (b) is the cutoff time and the reference time is a time, the unit should be ps, not ps^{-1}. Please correct the unit.
  4. [Sec. III.B, Eq. (13)] The expression for J_QD(ω) has an unbalanced parenthesis: it reads ω^3 ( c_ee^{-ω^2/ω_e^2} - c_he^{-ω^2/ω_h^2} ) but the opening parenthesis appears before c_e and the closing after the second exponential, which is consistent, but the notation is ambiguous because the ω^3 is outside the parenthesis yet no multiplication sign is shown. Please add parentheses or multiplication dots for clarity.
  5. [Sec. II.B, Eq. (6)] The definition of the projected stationary map in Eq. (6) omits an explicit projection onto the system subspace on the right-hand side: the right side uses Pv_j and ṽ_j^† P, but the left side E_{t+Δt,t} acts on system density matrices. It would be clearer to write E_{t+Δt,t} = P [ sum_j e^{(-λ_j+iω_j)Δt} v_j ṽ_j^† ] P and to state that the maps are understood as restricted to the system subspace.
  6. [Sec. IV] The discussion proposes constructing transfer tensors from dynamical maps E_{τ_c, τ_c - nΔt} around the cutoff time, but does not give a concrete test or estimate of when this would outperform time-local extrapolation. This is a reasonable outlook, but it would strengthen the paper to include at least one numerical example or a clear argument for why such a scheme could be advantageous.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation: both extrapolation schemes are constructed from short-time dynamical maps and validated against an independent exact solver; the abstract's 'invariably' overstates the paper's own data but is not a circular step.

full rationale

The paper's derivation chain is self-contained. Both the transfer tensor method and the time-local extrapolation scheme are defined from the same short-time dynamical maps E_{t_n,t0}; no parameter is fitted to long-time data, and the long-time predictions are compared with an independent numerically exact ACE/PT-MPO reference. Equation (4), E_{t_{n+1},t_n} = E_{t_{n+1},t0} E_{t_n,t0}^{-1}, is an exact identity for nonsingular maps, and the stationarity assumption in Sec. II B is justified by a Markovian-embedding argument involving a time-independent diagonalizable Liouvillian and the eigenexpansion in Eqs. (5)-(6), rather than by assuming the target conclusion. The ACE code is cited from the authors' prior work, but it is used as a black-box solver for both the short-time input maps and the exact long-time benchmark; it does not encode the comparison result, so the self-citations are not load-bearing in a circular sense. The abstract's claim that time-local extrapolation 'invariably converges at least as fast' is contradicted by the paper's own Fig. 1(b), where TL extrapolation is erratic near tau_c ~ 12 while TTM gives a finite error; however, that is an overgeneralization of an empirical finding, not a reduction of the prediction to its inputs. No circular step can be exhibited from the paper's equations.

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

No free parameters are fitted; physical constants and model parameters are taken from the literature. The central assumptions are the invertibility of dynamical maps and the existence of a stationary regime for the local maps, plus trust in the ACE solver. No new physical entities are postulated.

assumptions (3)
  • domain assumption The dynamical map E_{t_n,t_0} is invertible for relevant times tn so that Eq. (4) is well-defined.
    Sec. II B, Eq. (4). The paper acknowledges singularities in Sec. III A and shows they can be detected, so this is a necessary condition rather than a fully justified fact.
  • domain assumption There exists a time-independent Markovian extension with a diagonalizable (or Jordan-form) Liouvillian such that the projected local maps become stationary for t > τc (Eqs. (5)-(6)).
    Sec. II B and footnote [34]; this is the key theoretical premise for the time-local scheme, not proven for general systems.
  • domain assumption The ACE/PT-MPO simulations are numerically exact within stated convergence parameters, providing the reference dynamics and the short-time maps.
    Sec. III: They rely on ACE with Δt and ϵ; no independent cross-check is shown against other solvers, though ACE is published and benchmarked.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Time-nonlocal versus time-local long-time extrapolation of non-Markovian quantum dynamics." pith.science (2026). https://pith.science/paper/5MIMBUEN

@misc{pith2026250521017,
  author       = {Pith},
  title        = {Pith review of: Time-nonlocal versus time-local long-time extrapolation of non-Markovian quantum dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5MIMBUEN}},
  note         = {Machine review of arXiv:2505.21017}
}
read the original abstract

The high numerical demands for simulating non-Markovian open quantum systems motivate a line of research where short-time dynamical maps are extrapolated to predict long-time behavior. The transfer tensor method (TTM) has emerged as a powerful and versatile paradigm for such scenarios. It relies on a systematic construction of a converging sequence of time-nonlocal corrections to a time-constant local dynamical map. Here, we show that the same objective can be achieved with time-local extrapolation based on the observation that time-dependent time-local dynamical maps become stationary. Surprisingly, the maps become stationary long before the open quantum system reaches its steady state. Comparing both approaches numerically on examples of the canonical spin-boson model with sub-ohmic, ohmic, and super-ohmic spectral density, respectively, we find that, while both approaches eventually converge with increasing length of short-time propagation, our simple time-local extrapolation invariably converges at least as fast as time-nonlocal extrapolation. These results suggest that, perhaps counter-intuitively, time-nonlocality is not in fact a prerequiste for accurate and efficient long-time extrapolation of non-Markovian quantum dynamics.

Figures

Figures reproduced from arXiv: 2505.21017 by the authors.

Figure 1
Figure 1. FIG. 1. (a) Dynamics of a driven sub-ohmic spin-boson model including extrapolated dynamics starting from various cutoff [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Ohmic spin-boson model with Drude-Lorentz spec [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

43 extracted references · 34 canonical work pages

  1. [1]

    Lindblad, Communications in Mathematical Physics 48, 119 (1976)

    G. Lindblad, Communications in Mathematical Physics 48, 119 (1976)

    1976

  2. [2]

    Gorini, A

    V. Gorini, A. Kossakowski, and E. C. G. Sudarshan, Journal of Mathematical Physics 17, 821 (1976). 7

    work page 1976

  3. [3]

    White, F

    G. White, F. Pollock, L. Hollenberg, K. Modi, and C. Hill, PRX Quantum 3, 020344 (2022)

    work page 2022

  4. [4]

    de Vega and D

    I. de Vega and D. Alonso, Rev. Mod. Phys. 89, 015001 (2017)

    2017

  5. [5]

    Cosacchi, F

    M. Cosacchi, F. Ungar, M. Cygorek, A. Vagov, and V. M. Axt, Phys. Rev. Lett. 123, 017403 (2019)

    work page 2019

  6. [6]

    Nazir and D

    A. Nazir and D. P. S. McCutcheon, Journal of Physics: Condensed Matter 28, 103002 (2016)

    work page 2016

  7. [7]

    Caruso, A

    F. Caruso, A. W. Chin, A. Datta, S. F. Huelga, and M. B. Plenio, The Journal of Chemical Physics 131, 105106 (2009)

    work page 2009

  8. [8]

    Lorenzoni, N

    N. Lorenzoni, N. Cho, J. Lim, D. Tamascelli, S. F. Huelga, and M. B. Plenio, Phys. Rev. Lett. 132, 100403 (2024)

    work page 2024

Show all 43 references
  1. [9]

    M. P. Woods and M. B. Plenio, Journal of Mathematical Physics 57, 022105 (2016)

    2016

  2. [10]

    Cygorek, J

    M. Cygorek, J. Keeling, B. W. Lovett, and E. M. Gauger, Phys. Rev. X 14, 011010 (2024)

    2024

  3. [11]

    Link, H.-H

    V. Link, H.-H. Tu, and W. T. Strunz, Phys. Rev. Lett. 132, 200403 (2024)

    2024

  4. [12]

    Makri, Journal of Chemical Theory and Computation 17, 1 (2021)

    N. Makri, Journal of Chemical Theory and Computation 17, 1 (2021)

    2021

  5. [13]

    Cerrillo and J

    J. Cerrillo and J. Cao, Phys. Rev. Lett. 112, 110401 (2014)

    2014

  6. [14]

    A. A. Kananenka, C.-Y. Hsieh, J. Cao, and E. Geva, The Journal of Physical Chemistry Letters 7, 4809 (2016)

    2016

  7. [15]

    Gelzinis, E

    A. Gelzinis, E. Rybakovas, and L. Valkunas, The Journal of Chemical Physics 147, 234108 (2017)

    2017

  8. [16]

    Rosenbach, J

    R. Rosenbach, J. Cerrillo, S. F. Huelga, J. Cao, and M. B. Plenio, New Journal of Physics 18, 023035 (2016)

    2016

  9. [17]

    Buser, J

    M. Buser, J. Cerrillo, G. Schaller, and J. Cao, Phys. Rev. A 96, 062122 (2017)

    2017

  10. [18]

    A. Wu, J. Cerrillo, and J. Cao, Nanophotonics 13, 2575–2590 (2024)

    2024

  11. [19]

    F. A. Pollock and K. Modi, Quantum 2, 76 (2018)

    2018

  12. [20]

    Chen, K.-L

    Y.-Q. Chen, K.-L. Ma, Y.-C. Zheng, J. Allcock, S. Zhang, and C.-Y. Hsieh, Phys. Rev. Appl. 13, 034045 (2020)

    2020

  13. [21]

    Chen, Y.-C

    Y.-Q. Chen, Y.-C. Zheng, S. Zhang, and C.-Y. Hsieh, Phys. Rev. Appl. 17, 064007 (2022)

    2022

  14. [22]

    Andersson, J

    E. Andersson, J. D. Cresser, and M. J. W. Hall, Journal of Modern Optics 54, 1695 (2007)

    2007

  15. [23]

    M. J. W. Hall, J. D. Cresser, L. Li, and E. Andersson, Phys. Rev. A 89, 042120 (2014)

    2014

  16. [24]

    D. J. Strachan, A. Purkayastha, and S. R. Clark, The Journal of Chemical Physics 161, 154105 (2024)

    2024

  17. [25]

    F. A. Pollock, C. Rodr ´ ıguez-Rosario, T. Frauenheim, M. Paternostro, and K. Modi, Phys. Rev. A 97, 012127 (2018)

    2018

  18. [26]

    M. R. Jørgensen and F. A. Pollock, Phys. Rev. Lett. 123, 240602 (2019)

    2019

  19. [27]

    Cygorek and E

    M. Cygorek and E. M. Gauger, SciPost Phys. 18, 024 (2025)

    2025

  20. [28]

    Cygorek, M

    M. Cygorek, M. Cosacchi, A. Vagov, V. M. Axt, B. W. Lovett, J. Keeling, and E. M. Gauger, Nature Physics 18, 662 (2022)

    2022

  21. [29]

    Cygorek and E

    M. Cygorek and E. M. Gauger, The Journal of Chemical Physics 161, 074111 (2024)

    2024

  22. [30]

    Anto-Sztrikacs and D

    N. Anto-Sztrikacs and D. Segal, Phys. Rev. A 104, 052617 (2021)

    2021

  23. [31]

    Prior, A

    J. Prior, A. W. Chin, S. F. Huelga, and M. B. Plenio, Phys. Rev. Lett. 105, 050404 (2010)

    2010

  24. [32]

    Pleasance, B

    G. Pleasance, B. M. Garraway, and F. Petruccione, Phys. Rev. Res. 2, 043058 (2020)

    2020

  25. [33]

    Makri and D

    N. Makri and D. E. Makarov, The Journal of Chemical Physics 102, 4600 (1995)

    1995

  26. [34]

    Then, in Eqs

    If the Liouvillian has exceptional points, one can use as basis vectors vj the generalized eigenvectors that bring the Liouvillian into Jordan normal form J. Then, in Eqs. (5) and (6), e(−λj +iωj )tvj should be replaced by the more general expression eJ tvj, which can give ris...

  27. [35]

    Tokuyama and H

    M. Tokuyama and H. Mori, Progress of Theoretical Physics 55, 411 (1976)

    1976

  28. [36]

    Hashitsumae, F

    N. Hashitsumae, F. Shibata, and M. Shing ¯u, Journal of Statistical Physics 17, 155 (1977)

    1977

  29. [37]

    By contrast, the orthogonality condition Tr( L† jLj′ ) = δj,j′ would correspond to fixing the Frobenius norm to 1

  30. [38]

    A. J. Leggett, S. Chakravarty, A. T. Dorsey, M. P. A. Fisher, A. Garg, and W. Zwerger, Rev. Mod. Phys. 59, 1 (1987)

    1987

  31. [39]

    Kahlert, V

    F. Kahlert, V. Link, R. Hartmann, and W. T. Strunz, The Journal of Chemical Physics 161, 184108 (2024)

    2024

  32. [40]

    Krummheuer, V

    B. Krummheuer, V. M. Axt, T. Kuhn, I. D’Amico, and F. Rossi, Phys. Rev. B 71, 235329 (2005)

    2005

  33. [41]

    Cygorek, B

    M. Cygorek, B. W. Lovett, J. Keeling, and E. M. Gauger, Phys. Rev. Res. 6, 043203 (2024)

    2024

  34. [42]

    Gherardini, A

    S. Gherardini, A. Smirne, S. F. Huelga, and F. Caruso, Quantum Science and Technology 7, 025005 (2022)

    2022

  35. [43]

    Minganti, A

    F. Minganti, A. Biella, N. Bartolo, and C. Ciuti, Phys. Rev. A 98, 042118 (2018)

    2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.