pith. sign in

arxiv: 2605.01494 · v1 · submitted 2026-05-02 · 🧮 math.NA · cs.NA· quant-ph

Completely Positive and Trace Preserving Schemes with Tensor Train Compression for the Lindblad Equation

Peter DelMastro , Daniel Appel\"o , Yingda Cheng This is my paper

Pith reviewed 2026-05-09 17:41 UTC · model grok-4.3

classification 🧮 math.NA cs.NAquant-ph
keywords Lindblad equationtensor traincompletely positivetrace preservinglow-rank approximationopen quantum systemsnumerical methods for PDEs
0
0 comments X

The pith

A two-level low-rank format with tensor-train compression on matrix columns produces completely positive and trace-preserving integrators for the Lindblad equation.

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

The paper constructs numerical schemes for the Lindblad master equation that governs the evolution of open quantum systems. It represents the density matrix as a product of tall-skinny matrices whose columns are themselves stored in tensor-train format. All arithmetic steps required by the underlying Kraus-map integrator are then carried out directly in this compressed representation, with special attention to how rank truncation is performed. Extensive tests show that the resulting methods converge and can handle quantum systems containing up to 10^19 degrees of freedom on modest hardware.

Core claim

The central claim is that the two-level low-rank representation—tall-skinny factorization of the density matrix followed by tensor-train compression of its columns—permits every column-wise operation of the Kraus-is-King scheme to be executed efficiently in the compressed format, while rank truncation at each step approximately preserves the completely positive and trace-preserving properties of the map.

What carries the argument

The two-level low-rank format consisting of a factorization of the density matrix into tall-skinny matrices whose columns are compressed in tensor-train (matrix-product-state) form, which supports direct implementation of the Kraus-map arithmetic.

Load-bearing premise

Rank truncation and tensor-train compression can be applied at each step while exactly or approximately preserving the completely positive and trace-preserving character of the Kraus map without uncontrolled accumulation of error in long-time dynamics.

What would settle it

A moderate-sized Lindblad problem with a known exact solution in which the numerical trace deviates from one or the eigenvalues of the density matrix become negative after many steps, beyond the stated truncation tolerance.

Figures

Figures reproduced from arXiv: 2605.01494 by Daniel Appel\"o, Peter DelMastro, Yingda Cheng.

Figure 1
Figure 1. Figure 1: Two-level low-rank decomposition of the density matrix view at source ↗
Figure 2
Figure 2. Figure 2: Convergence of the TT/MPS-based integrator applied to the dissipative 1D view at source ↗
Figure 3
Figure 3. Figure 3: Low-rank dynamics of the dissipative 1D Heisenberg model with view at source ↗
Figure 4
Figure 4. Figure 4: 25-qubit portion of a larger heavy-hex lattice. Subfigure view at source ↗
Figure 5
Figure 5. Figure 5: Dynamics of the 25-qubit heavy-hex lattice mock quantum circuit simulation. view at source ↗
Figure 6
Figure 6. Figure 6: Run statistics of the heavy-hex lattice mock quantum circuit simulation: den view at source ↗
Figure 7
Figure 7. Figure 7: Chain of alternating qudits (4-level systems) and resonators (10-level systems), view at source ↗
Figure 8
Figure 8. Figure 8: Low-rank dynamics of six qudits (4-level systems) and five resonators (10- view at source ↗
Figure 9
Figure 9. Figure 9: Run statistics of the heavy-hex lattice mock quantum circuit simulation. view at source ↗
read the original abstract

We propose a family of low-rank, completely positive and trace preserving schemes for the Lindblad equation, a common model for open quantum systems. Low-rank representation is employed at two levels: the density matrix is factorized into the product of tall-skinny matrices, and the columns of these matrices are further represented using the tensor train (TT) format, also know as matrix product states (MPS). This two-level low-rank format fits naturally into our existing Kraus is King scheme (arXiv:2409.08898v2 [math.NA]) for the Lindblad equation, whose underlying operations are arithmetic on the columns of the tall-skinny matrices. We show how these operations can be performed efficiently in the TT/MPS format, with particular emphasis on density matrix rank-truncation. We conclude with extensive numerical experiments demonstrating the convergence of this scheme and its efficiency in simulating systems with up to $10^{19}$ degrees of freedom using only modest compute resources.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a family of low-rank completely positive and trace-preserving (CPTP) schemes for the Lindblad equation. It employs a two-level representation in which the density matrix is factored into tall-skinny matrices whose columns are further compressed in tensor-train (TT/MPS) format. The approach extends the authors' prior Kraus-is-King arithmetic (arXiv:2409.08898) to perform all operations, including density-matrix rank truncation, directly in the TT format. Numerical experiments are presented to demonstrate convergence and computational efficiency on systems with up to 10^19 degrees of freedom.

Significance. If the truncation errors can be controlled so that the approximate CPTP property remains stable over long times, the method would enable scalable simulations of open quantum systems far beyond current capabilities while retaining physical fidelity. The reported ability to handle 10^19-dimensional systems on modest hardware, together with the explicit integration of TT compression into an existing structure-preserving framework, represents a practical advance for quantum dynamics.

major comments (2)
  1. [TT truncation procedure] The section describing TT rank truncation (following the extension of column-wise Kraus operations): the untruncated scheme is exactly CPTP by construction from the prior work, yet the manuscript provides no analytic bound on the deviation of the Choi-matrix eigenvalues or on per-step trace drift after TT-SVD or cross-approximation truncation. Without such bounds it is impossible to guarantee that positivity or trace violations do not accumulate secularly over the long integration intervals that matter for open-system dynamics.
  2. [Numerical experiments] Numerical experiments section: while convergence is shown for systems up to 10^19 dof, the experiments do not quantify the growth of CPTP violations (e.g., minimum Choi eigenvalue or trace deviation) as a function of truncation rank or integration time, nor do they compare against exact solutions on moderate-sized systems to calibrate the truncation-induced error.
minor comments (2)
  1. [Abstract] The abstract and introduction would benefit from an explicit statement of the precise sense in which the truncated scheme is claimed to be CPTP (exact, approximately, or in a weak sense).
  2. [Preliminaries] Notation for the two-level low-rank format (tall-skinny factors plus TT columns) should be introduced once and used consistently to avoid ambiguity when describing arithmetic operations.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the careful reading of the manuscript and the constructive comments. We address each major point below and outline the revisions we will make to strengthen the presentation of the method's properties.

read point-by-point responses

  1. Referee: The section describing TT rank truncation (following the extension of column-wise Kraus operations): the untruncated scheme is exactly CPTP by construction from the prior work, yet the manuscript provides no analytic bound on the deviation of the Choi-matrix eigenvalues or on per-step trace drift after TT-SVD or cross-approximation truncation. Without such bounds it is impossible to guarantee that positivity or trace violations do not accumulate secularly over the long integration intervals that matter for open-system dynamics.

    Authors: We agree that rigorous analytic bounds on the post-truncation deviation of Choi-matrix eigenvalues (or per-step trace drift) would be ideal for guaranteeing long-term stability. Deriving such bounds is, however, technically challenging: the TT-SVD or cross-approximation error is controlled in the Frobenius norm of the tensor factors, but translating this into a uniform bound on the spectrum of the reconstructed Choi operator requires additional assumptions on the singular-value decay and the structure of the Kraus operators that are not generally available. In the revised manuscript we will therefore add a dedicated numerical section that monitors the minimum Choi eigenvalue and the cumulative trace deviation as functions of both truncation tolerance and integration time for representative systems. This will provide practical evidence that violations remain small and do not exhibit secular growth under the tolerances used in the large-scale experiments. revision: partial

  2. Referee: Numerical experiments section: while convergence is shown for systems up to 10^19 dof, the experiments do not quantify the growth of CPTP violations (e.g., minimum Choi eigenvalue or trace deviation) as a function of truncation rank or integration time, nor do they compare against exact solutions on moderate-sized systems to calibrate the truncation-induced error.

    Authors: We accept that the current numerical section does not explicitly quantify CPTP violations or provide calibration against exact solutions. In the revised version we will augment the experiments with two additions: (i) direct comparisons against fully resolved (non-TT) solutions on moderate-dimensional systems (up to 2^10–2^12 states) for several truncation ranks, reporting both global error and the induced CPTP deviation; (ii) plots and tables showing the evolution of the minimum Choi eigenvalue and trace deviation over long integration intervals for the larger test cases, parameterized by the TT truncation tolerance. These additions will calibrate the truncation error and demonstrate its practical control. revision: yes

standing simulated objections not resolved
  • Derivation of analytic bounds on the deviation of Choi-matrix eigenvalues after TT truncation

Circularity Check

1 steps flagged

CPTP property inherited via self-citation to prior Kraus-is-King scheme; TT truncation adds independent implementation but lacks separate preservation proof

specific steps
  1. self citation load bearing [Abstract]
    "This two-level low-rank format fits naturally into our existing Kraus is King scheme (arXiv:2409.08898v2 [math.NA]) for the Lindblad equation, whose underlying operations are arithmetic on the columns of the tall-skinny matrices. We show how these operations can be performed efficiently in the TT/MPS format, with particular emphasis on density matrix rank-truncation."

    The completely-positive and trace-preserving character is asserted because the column-wise operations inherit the structure of the self-cited prior scheme. The TT compression itself is new, but the load-bearing CPTP property is not re-derived or independently verified within this manuscript.

full rationale

The paper's core claim of producing CPTP schemes rests on the new TT/MPS operations fitting into the arithmetic structure of the authors' earlier work (arXiv:2409.08898). This is a self-citation load-bearing step for the preservation guarantee, yet the TT-specific algorithms, rank-truncation handling, and large-scale numerical tests constitute substantial independent content. No self-definitional loops, fitted inputs renamed as predictions, or ansatz smuggling appear in the provided text. The result is therefore only moderately circular.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Assessment limited to abstract; full paper would be needed to identify any fitted truncation thresholds or additional assumptions.

axioms (2)
  • domain assumption The Lindblad master equation is the appropriate model for the open quantum systems under consideration.
    Standard modeling assumption invoked by the choice of equation.
  • domain assumption Low-rank and tensor-train approximations can be constructed so that the resulting maps remain completely positive and trace-preserving.
    Central to the scheme design and truncation procedure.

pith-pipeline@v0.9.0 · 5476 in / 1325 out tokens · 58697 ms · 2026-05-09T17:41:16.941466+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages

  1. [1]

    Structure-preserving numerical schemes for

    Cao, Yu and Lu, Jianfeng , journal=. Structure-preserving numerical schemes for. 2025 , publisher=

    work page 2025

  2. [2]

    50th International Colloquium on Automata, Languages, and Programming (ICALP 2023) , series =

    Xiantao Li and Chunhao Wang , title =. 50th International Colloquium on Automata, Languages, and Programming (ICALP 2023) , series =. 2023 , doi =

    work page 2023

  3. [3]

    Unconditionally Stable Time Discretization of

    R. Unconditionally Stable Time Discretization of. 2025 , eprint =

    work page 2025

  4. [4]

    Adaptive low-rank approximation and denoised

    Le Bris, Claude and Rouchon, Pierre and Roussel, Julien , journal=. Adaptive low-rank approximation and denoised. 2015 , publisher=

    work page 2015

  5. [5]

    The Theory of Open Quantum Systems , isbn =

    Breuer, Heinz-Peter and Petruccione, Francesco , year =. The Theory of Open Quantum Systems , isbn =. The Theory of Open Quantum Systems , doi =

    work page

  6. [6]

    Communications in Mathematical Physics , year=

    On the generators of quantum dynamical semigroups , author=. Communications in Mathematical Physics , year=

    work page

  7. [7]

    AIP Advances , volume =

    Manzano, Daniel , title =. AIP Advances , volume =. 2020 , month =

    work page 2020

  8. [8]

    Journal of Computational Physics , volume =

    Kraus is. Journal of Computational Physics , volume =. 2025 , issn =. doi:https://doi.org/10.1016/j.jcp.2025.114036 , author =

    work page doi:10.1016/j.jcp.2025.114036 2025

  9. [9]

    Full- and low-rank exponential midpoint schemes for forward and adjoint

    Hao Chen and Alfio Borz. Full- and low-rank exponential midpoint schemes for forward and adjoint. ArXiv , year=

    work page

  10. [10]

    Arbitrary High Order Low-rank Completely Positive and Trace Preserving (

    Jiuhua Hu and Daniel Appel\". Arbitrary High Order Low-rank Completely Positive and Trace Preserving (. ArXiv , year=

    work page

  11. [11]

    , author=

    Wave-function approach to dissipative processes in quantum optics. , author=. Physical review letters , year=

    work page

  12. [12]

    Ralph Dum and Peter Zoller and Helmut Ritsch , journal=. Monte. 1992 , volume=

    work page 1992

  13. [13]

    Oseledets , journal=

    I. Oseledets , journal=. Tensor-. 2011 , volume=

    work page 2011

  14. [14]

    Schollw \"o ck ,\ 10.1016/j.aop.2010.09.012 journal journal Ann

    The density-matrix renormalization group in the age of matrix product states , journal =. 2011 , note =. doi:https://doi.org/10.1016/j.aop.2010.09.012 , author =

    work page doi:10.1016/j.aop.2010.09.012 2011

  15. [15]

    , title =

    Chan, Garnet Kin-Lic and Keselman, Anna and Nakatani, Naoki and Li, Zhendong and White, Steven R. , title =. The Journal of Chemical Physics , volume =. 2016 , month =

    work page 2016

  16. [16]

    Manmana and Ulrich Schollwöck and Claudius Hubig , journal =

    Time-evolution methods for matrix-product states , journal =. 2019 , issn =. doi:https://doi.org/10.1016/j.aop.2019.167998 , author =

    work page doi:10.1016/j.aop.2019.167998 2019

  17. [17]

    Efficient Simulation of One-Dimensional Quantum Many-Body Systems , author =. Phys. Rev. Lett. , volume =. 2004 , month =

    work page 2004

  18. [18]

    Matrix Product Density Operators: Simulation of Finite-Temperature and Dissipative Systems , author =. Phys. Rev. Lett. , volume =. 2004 , month =

    work page 2004

  19. [19]

    Computation of extreme eigenvalues in higher dimensions using block tensor train format , author=. Comput. Phys. Commun. , year=

    work page

  20. [20]

    Low-Rank Tensor Methods with Subspace Correction for Symmetric Eigenvalue Problems , author=. SIAM J. Sci. Comput. , year=

    work page

  21. [21]

    Reid and Arvind K

    Hussam Al Daas and Grey Ballard and Paul Cazeaux and Eric Hallman and Agnieszka Miedlar and Mirjeta Pasha and Tim W. Reid and Arvind K. Saibaba , journal=. Randomized algorithms for rounding in the. 2021 , volume=

    work page 2021

  22. [22]

    Adaptive Randomized

    Hussam Al Daas and Grey Ballard and Laura Grigori and Mariana Martinez Aguilar and Arvind Krishna Saibaba and Bhisham Dev Verma , journal=. Adaptive Randomized. 2025 , volume=

    work page 2025

  23. [23]

    Oseledets and Eugene E

    I. Oseledets and Eugene E. Tyrtyshnikov , journal=. 2010 , volume=

    work page 2010

  24. [24]

    The 2011 International Workshop on Multidimensional (nD) Systems , year=

    Fast adaptive interpolation of multi-dimensional arrays in tensor train format , author=. The 2011 International Workshop on Multidimensional (nD) Systems , year=

    work page 2011

  25. [25]

    , author=

    Density matrix formulation for quantum renormalization groups. , author=. Physical review letters , year=

    work page

  26. [26]

    and Dolgov, S

    Oseledets, I. and Dolgov, S. , title =. 2024 , note =

    work page 2024

  27. [27]

    J'erome Houdayer and Gr

    work page

  28. [28]

    Rapid Quantum Ground State Preparation via Dissipative Dynamics , author =. Phys. Rev. X , volume =. 2026 , month =. doi:10.1103/wzb3-dbg9 , url =

    work page doi:10.1103/wzb3-dbg9 2026

  29. [29]

    and Rodrigo-Albert, Daniel and Rath, Yannic and Rungger, Ivan , title =

    Lindoy, Lachlan P. and Rodrigo-Albert, Daniel and Rath, Yannic and Rungger, Ivan , title =. The Journal of Chemical Physics , volume =. 2025 , month =

    work page 2025

  30. [30]

    SIAM Journal on Scientific Computing , volume =

    Ceruti, Gianluca and Kressner, Daniel and Sulz, Dominik , title =. SIAM Journal on Scientific Computing , volume =. 2025 , doi =

    work page 2025

  31. [31]

    , author=

    Matrix product density operators: simulation of finite-temperature and dissipative systems. , author=. Physical review letters , year=

    work page

  32. [32]

    , author=

    Positive Tensor Network Approach for Simulating Open Quantum Many-Body Systems. , author=. Physical review letters , year=

    work page

  33. [33]

    and Bañuls, Mari-Carmen , year =

    Cui, Jian and Cirac, J. and Bañuls, Mari-Carmen , year =. Variational Matrix Product Operators for the Steady State of Dissipative Quantum Systems , volume =. Physical Review Letters , doi =

    work page

  34. [34]

    Quantum Science and Technology , year=

    One-dimensional many-body entangled open quantum systems with tensor network methods , author=. Quantum Science and Technology , year=

    work page

  35. [35]

    Finding purifications with minimal entanglement , author =. Phys. Rev. B , volume =. 2018 , month =

    work page 2018

  36. [36]

    Density matrix renormalization group algorithm for mixed quantum states , volume =

    Guo, Chu , year =. Density matrix renormalization group algorithm for mixed quantum states , volume =. Physical Review B , doi =

    work page

  37. [37]

    SciPost Physics Core , year=

    Nonlocal correlations in noisy multiqubit systems simulated using matrix product operators , author=. SciPost Physics Core , year=

    work page

  38. [38]

    Enabling large-depth simulation of noisy quantum cir- 10 cuits with positive tensor networks.arXiv preprint arXiv:2403.00152, 2024

    M. Enabling large-depth simulation of noisy quantum circuits with positive tensor networks. 2024. arXiv:2403.00152

    work page arXiv 2024

  39. [39]

    2024 , school =

    Xianrui Yin , title =. 2024 , school =

    work page 2024

  40. [40]

    Gauge-fixing quantum density operators at scale , author =. Phys. Rev. Res. , volume =. 2025 , month =

    work page 2025

  41. [41]

    Fux and Dainius Kilda and Brendon W

    Gerald E. Fux and Dainius Kilda and Brendon W. Lovett and Jonathan Keeling , journal=. Tensor network simulation of chains of non-

    work page

  42. [42]

    Anders Petersson and Fortino Garcia , journal=

    N. Anders Petersson and Fortino Garcia , journal=. Optimal Control of Closed Quantum Systems via. 2021 , volume=

    work page 2021

  43. [43]

    Journal of Computational Physics , volume =

    High-order. Journal of Computational Physics , volume =. 2026 , issn =. doi:https://doi.org/10.1016/j.jcp.2026.114697 , author =

    work page doi:10.1016/j.jcp.2026.114697 2026

  44. [44]

    Multi-target density matrix renormalization group

    Gonz. Multi-target density matrix renormalization group. arXiv preprint arXiv:2506.24109 , year=

    work page arXiv