pith. sign in

arxiv: 2505.00494 · v2 · submitted 2025-05-01 · ❄️ cond-mat.str-el · quant-ph

Accelerating two-dimensional tensor network contractions using QR decompositions

Yining Zhang , Qi Yang , Philippe Corboz This is my paper

Pith reviewed 2026-05-22 17:33 UTC · model grok-4.3

classification ❄️ cond-mat.str-el quant-ph
keywords tensor networksiPEPSCTMRGQR decompositionstrongly correlated systemsHeisenberg modelJ1-J2 modelGPU acceleration
0
0 comments X

The pith

Replacing singular value decompositions with QR decompositions speeds up CTMRG contractions of C4v-symmetric iPEPS by up to two orders of magnitude.

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

This paper sets out to demonstrate that QR decompositions can take the place of slower singular value or eigenvalue decompositions in the renormalization step of the corner transfer matrix renormalization group method. The change is made for tensor networks that have C4v symmetry. A reader would care because the contraction of infinite projected entangled-pair states is a major computational cost in studying two-dimensional strongly correlated systems, and faster methods open the door to more detailed calculations on accessible hardware.

Core claim

The authors introduce a contraction scheme for C4v-symmetric tensor networks that integrates QR decompositions into the CTMRG algorithm. This replaces the standard singular value or eigenvalue decompositions in the renormalization. The result is an up to 100-fold speedup with no loss of accuracy, allowing state-of-the-art computations for the Heisenberg and J1-J2 models to complete in less than one hour on an H100 GPU.

What carries the argument

The key mechanism is the use of QR decompositions to approximate the corner transfer matrices in the CTMRG renormalization step for symmetric tensors.

Load-bearing premise

That the QR decompositions maintain the fixed-point accuracy and convergence properties of the original CTMRG renormalization for these symmetric tensors.

What would settle it

Running the standard CTMRG and the QR version on the Heisenberg model at the same bond dimension and finding a significant difference in the computed ground state energy or order parameter.

Figures

Figures reproduced from arXiv: 2505.00494 by Philippe Corboz, Qi Yang, Yining Zhang.

Figure 1
Figure 1. Figure 1: FIG. 1. (a) Translational invariant iPEPS represented by [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. (a) Energy per site of the 2D Heisenberg model as [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. (a) Relative error of the energy per site of the 2D [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. (a) Comparison of the optimization between stan [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. QR based iPEPS results as a function of [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
read the original abstract

Infinite projected entangled-pair states (iPEPS) provide a powerful tool for studying strongly correlated systems directly in the thermodynamic limit. A core component of the algorithm is the approximate contraction of the iPEPS, where the computational bottleneck typically lies in the singular value or eigenvalue decompositions involved in the renormalization step. This is particularly true on GPUs, where tensor contractions are substantially faster than these decompositions. Here we propose a contraction scheme for $C_{4v}$-symmetric tensor networks based on combining the corner transfer matrix renormalization group (CTMRG) with QR-decompositions which are substantially faster, especially on GPUs. Our approach achieves up to two orders of magnitude speedup compared to standard CTMRG without loss of accuracy and yields state-of-the-art results for the Heisenberg and $J_1$-$J_2$ models in less than 1 h on an H100 GPU.

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 replacing singular value or eigenvalue decompositions with QR decompositions in the renormalization step of CTMRG for C4v-symmetric iPEPS tensor networks. It claims this yields up to two orders of magnitude speedup on GPUs with no loss of accuracy, demonstrated by state-of-the-art energies for the Heisenberg and J1-J2 models obtained in under one hour on an H100 GPU.

Significance. If the accuracy claim holds, the approach would meaningfully accelerate 2D tensor-network simulations by alleviating the decomposition bottleneck on GPUs, enabling higher bond dimensions for strongly correlated systems. The concrete GPU timings and benchmarks on standard models are positive features.

major comments (2)
  1. [QR-based renormalization description] The central claim of preserved accuracy rests on the QR substitution in the CTMRG renormalization step for C4v tensors. A direct comparison of retained singular values, truncation errors, or corner-matrix spectra between the QR scheme and standard SVD (as in §3 or the renormalization subsection) is needed to confirm that deviations do not accumulate and alter the fixed-point environment.
  2. [Numerical results section] Table or figure reporting energies for the Heisenberg model: while state-of-the-art values are stated, the manuscript should include explicit truncation-error metrics or convergence plots comparing QR-CTMRG to SVD-CTMRG at the same bond dimension D and environment dimension χ to substantiate 'no loss of accuracy'.
minor comments (2)
  1. [Abstract and methods] Clarify in the methods whether the QR adaptation uses symmetry block structure or post-selection to mimic SVD truncation, and specify the exact bond dimensions D and environment sizes χ at which the reported speedups were measured.
  2. [Performance benchmarks] Ensure all timing benchmarks explicitly state hardware details, number of iterations, and whether the comparison uses identical convergence criteria.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their positive assessment of our work and for the constructive suggestions that will improve the manuscript. We address each major comment below and have incorporated revisions to provide the requested comparisons.

read point-by-point responses

  1. Referee: The central claim of preserved accuracy rests on the QR substitution in the CTMRG renormalization step for C4v tensors. A direct comparison of retained singular values, truncation errors, or corner-matrix spectra between the QR scheme and standard SVD (as in §3 or the renormalization subsection) is needed to confirm that deviations do not accumulate and alter the fixed-point environment.

    Authors: We agree that explicit comparisons strengthen the accuracy claim. In the revised manuscript we have added a new subsection (Section 3.3) containing a direct comparison of the retained singular values and truncation errors for the QR and SVD renormalization steps at multiple bond dimensions. We also include a figure showing the corner-matrix spectra for both methods at the fixed point; the spectra agree to within 10^{-9} and the truncation errors differ by less than 10^{-8}, confirming that deviations do not accumulate over iterations. revision: yes

  2. Referee: Table or figure reporting energies for the Heisenberg model: while state-of-the-art values are stated, the manuscript should include explicit truncation-error metrics or convergence plots comparing QR-CTMRG to SVD-CTMRG at the same bond dimension D and environment dimension χ to substantiate 'no loss of accuracy'.

    Authors: We thank the referee for this suggestion. The revised manuscript now contains a new figure (Figure 4) that plots the energy per site versus environment dimension χ for both QR-CTMRG and SVD-CTMRG at fixed D = 8 for the Heisenberg model. An accompanying table reports the truncation errors and final energies at several χ values; the two methods agree to better than 10^{-7} in energy and exhibit comparable truncation errors, directly substantiating the claim of no accuracy loss. revision: yes

Circularity Check

0 steps flagged

No circularity: algorithmic substitution validated on external physical benchmarks

full rationale

The paper describes an explicit algorithmic change—replacing SVD/EVD with QR decompositions inside the CTMRG renormalization step for C4v-symmetric iPEPS tensors—and measures its effect via wall-clock speedup and agreement with known ground-state energies of the Heisenberg and J1-J2 models. No derivation step reduces to a fitted parameter that is then relabeled as a prediction, no self-citation supplies a uniqueness theorem that forces the result, and the accuracy claim is tested against independent external data rather than being true by construction. The method is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The work is an algorithmic optimization that rests on the standard validity of the iPEPS ansatz and CTMRG fixed-point equations; no new free parameters, physical axioms, or invented entities are introduced.

axioms (1)
  • domain assumption The iPEPS ansatz and CTMRG renormalization produce accurate approximations to the thermodynamic limit for the models considered.
    Invoked implicitly when claiming that the new scheme yields state-of-the-art results without loss of accuracy.

pith-pipeline@v0.9.0 · 5681 in / 1277 out tokens · 40246 ms · 2026-05-22T17:33:12.660954+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Single-layer framework of variational tensor network states

    cond-mat.str-el 2025-12 unverdicted novelty 6.0

    A single-layer variational tensor network method reduces computational cost by three orders of magnitude in bond dimension for 2D quantum models and confirms an intermediate empty-plaquette valence bond solid phase in...

Reference graph

Works this paper leans on

76 extracted references · 76 canonical work pages · cited by 1 Pith paper · 3 internal anchors

  1. [1]

    S. R. White, Phys. Rev. Lett.69, 2863 (1992)

    work page 1992

  2. [2]

    Schollwöck, Annals of Physics326, 96 (2011)

    U. Schollwöck, Annals of Physics326, 96 (2011)

    work page 2011

  3. [3]

    Renormalization algorithms for Quantum-Many Body Systems in two and higher dimensions

    F. Verstraete and J. I. Cirac, arXiv:cond-mat/0407066 (2004)

    work page internal anchor Pith review Pith/arXiv arXiv 2004

  4. [4]

    Tensor Product Variational Formulation for Quantum Systems

    Y. Nishio, N. Maeshima, A. Gendiar, and T. Nishino, Preprint (2004), arXiv:cond-mat/0401115

    work page internal anchor Pith review Pith/arXiv arXiv 2004

  5. [5]

    Jordan, R

    J. Jordan, R. Orús, G. Vidal, F. Verstraete, and J. I. Cirac, Phys. Rev. Lett.101, 250602 (2008)

    work page 2008

  6. [6]

    Corboz and F

    P. Corboz and F. Mila, Phys. Rev. Lett. 112, 147203 (2014)

    work page 2014

  7. [7]

    H. J. Liao, Z. Y. Xie, J. Chen, Z. Y. Liu, H. D. Xie, R. Z. Huang, B.Normand,andT.Xiang,Phys.Rev.Lett. 118, 137202 (2017)

    work page 2017

  8. [8]

    Niesen and P

    I. Niesen and P. Corboz, SciPost Physics3, 030 (2017)

    work page 2017

  9. [9]

    J.-Y. Chen, L. Vanderstraeten, S. Capponi, and D. Poil- blanc, Phys. Rev. B98, 184409 (2018)

    work page 2018

  10. [10]

    S. S. Jahromi and R. Orús, Phys. Rev. B 98, 155108 (2018)

    work page 2018

  11. [11]

    Niesen and P

    I. Niesen and P. Corboz, Phys. Rev. B97, 245146 (2018)

    work page 2018

  12. [12]

    Yamaguchi, Y

    H. Yamaguchi, Y. Sasaki, T. Okubo, M. Yoshida, T. Kida, M. Hagiwara, Y. Kono, S. Kittaka, T. Sakak- ibara, M. Takigawa, Y. Iwasaki, and Y. Hosokoshi, Phys. Rev. B 98, 094402 (2018)

    work page 2018

  13. [13]

    Kshetrimayum, C

    A. Kshetrimayum, C. Balz, B. Lake, and J. Eisert, arXiv:1904.00028 [cond-mat, physics:quant-ph] (2019)

    work page arXiv 1904

  14. [14]

    S. S. Chung and P. Corboz, Phys. Rev. B100, 035134 (2019)

    work page 2019

  15. [15]

    Kawashima, and Y

    H.-Y.Lee, R.Kaneko, L.E.Chern, T.Okubo, Y.Yamaji, N. Kawashima, and Y. B. Kim, Nat. Comm.11, 1639 (2020)

    work page 2020

  16. [16]

    Gauthé, S

    O. Gauthé, S. Capponi, M. Mambrini, and D. Poilblanc, Phys. Rev. B101, 205144 (2020)

    work page 2020

  17. [17]

    J. L. Jiménez, S. P. G. Crone, E. Fogh, M. E. Za- yed, R. Lortz, E. Pomjakushina, K. Conder, A. M. Läuchli, L. Weber, S. Wessel, A. Honecker, B. Normand, C. Rüegg, P. Corboz, H. M. Rønnow, and F. Mila, Nature 592, 370 (2021)

    work page 2021

  18. [18]

    Czarnik, M

    P. Czarnik, M. M. Rams, P. Corboz, and J. Dziarmaga, Phys. Rev. B103, 075113 (2021)

    work page 2021

  19. [19]

    Hasik, D

    J. Hasik, D. Poilblanc, and F. Becca, SciPost Physics10, 012 (2021)

    work page 2021

  20. [20]

    W.-Y. Liu, J. Hasik, S.-S. Gong, D. Poilblanc, W.-Q. Chen, and Z.-C. Gu, Phys. Rev. X12, 031039 (2022)

    work page 2022

  21. [21]

    Hasik, M

    J. Hasik, M. Van Damme, D. Poilblanc, and L. Vander- straeten, Phys. Rev. Lett.129, 177201 (2022)

    work page 2022

  22. [22]

    Ponsioen, S

    B. Ponsioen, S. S. Chung, and P. Corboz, Phys. Rev. B 108, 205154 (2023)

    work page 2023

  23. [23]

    E. L. Weerda and M. Rizzi, Phys. Rev. B109, L241117 (2024)

    work page 2024

  24. [24]

    Hasik and P

    J. Hasik and P. Corboz, Phys. Rev. Lett.133, 176502 (2024)

    work page 2024

  25. [25]

    Schmoll, J

    P. Schmoll, J. Naumann, J. Eisert, and Y. Iqbal, arXiv:2407.07145 [cond-mat] (2024)

    work page arXiv 2024

  26. [26]

    Quantum spin liquid phase in the Shastry-Sutherland model revealed by high-precision infinite projected entangled-pair states

    P. Corboz, Y. Zhang, B. Ponsioen, and F. Mila, arXiv:2502.14091 [cond-mat] 10.48550/arXiv.2502.14091 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2502.14091 2025

  27. [27]

    Nishino and K

    T. Nishino and K. Okunishi, J. Phys. Soc. Jpn.65, 891 (1996)

    work page 1996

  28. [28]

    Nishino and K

    T. Nishino and K. Okunishi, J. Phys. Soc. Jpn.66, 3040 (1997)

    work page 1997

  29. [29]

    Orús and G

    R. Orús and G. Vidal, Phys. Rev. B80, 094403 (2009)

    work page 2009

  30. [30]

    Vanderstraeten, M

    L. Vanderstraeten, M. Mariën, F. Verstraete, and J. Haegeman, Phys. Rev. B92, 201111 (2015)

    work page 2015

  31. [31]

    Zauner-Stauber, L

    V. Zauner-Stauber, L. Vanderstraeten, M. T. Fishman, F. Verstraete, and J. Haegeman, Phys. Rev. B97, 045145 (2018)

    work page 2018

  32. [32]

    Vanderstraeten, L

    L. Vanderstraeten, L. Burgelman, B. Ponsioen, M. Van Damme, B. Vanhecke, P. Corboz, J. Haegeman, and F. Verstraete, Phys. Rev. B105, 195140 (2022)

    work page 2022

  33. [33]

    Zhang, S

    X.-Y. Zhang, S. Liang, H.-J. Liao, W. Li, and L. Wang, Phys. Rev. B108, 085103

    work page

  34. [34]

    Levin and C

    M. Levin and C. P. Nave, Phys. Rev. Lett.99, 120601 (2007)

    work page 2007

  35. [35]

    H. H. Zhao, Z. Y. Xie, Q. N. Chen, Z. C. Wei, J. W. Cai, and T. Xiang, Phys. Rev. B81, 174411 (2010)

    work page 2010

  36. [36]

    Z. Y. Xie, J. Chen, M. P. Qin, J. W. Zhu, L. P. Yang, and T. Xiang, Phys. Rev. B86, 045139 (2012)

    work page 2012

  37. [37]

    Evenbly and G

    G. Evenbly and G. Vidal, Phys. Rev. Lett.115, 180405 (2015)

    work page 2015

  38. [38]

    Corboz, S

    P. Corboz, S. R. White, G. Vidal, and M. Troyer, Phys. Rev. B 84, 041108(R) (2011)

    work page 2011

  39. [39]

    Corboz, T

    P. Corboz, T. M. Rice, and M. Troyer, Phys. Rev. Lett. 113, 046402 (2014)

    work page 2014

  40. [40]

    M. T. Fishman, L. Vanderstraeten, V. Zauner-Stauber, J. Haegeman, and F. Verstraete, Phys. Rev. B98, 235148 (2018)

    work page 2018

  41. [41]

    Z. Y. Xie, H. J. Liao, R. Z. Huang, H. D. Xie, J. Chen, Z. Y. Liu, and T. Xiang, Phys. Rev. B96, 045128 (2017)

    work page 2017

  42. [42]

    Haghshenas, S.-S

    R. Haghshenas, S.-S. Gong, and D. N. Sheng, Phys. Rev. B 99, 174423 (2019)

    work page 2019

  43. [43]

    Lan and G

    W. Lan and G. Evenbly, arXiv:2306.08212 [quant-ph] (2023)

    work page arXiv 2023

  44. [44]

    Naumann, E

    J. Naumann, E. L. Weerda, J. Eisert, M. Rizzi, and P. Schmoll, arXiv:2502.10298 [cond-mat] 10.48550/arXiv.2502.10298 (2025)

    work page doi:10.48550/arxiv.2502.10298 2025

  45. [45]

    A. D. S. Richards and E. S. Sørensen, arXiv:2503.13900 6 [cond-mat] 10.48550/arXiv.2503.13900 (2025)

    work page doi:10.48550/arxiv.2503.13900 2025

  46. [46]

    Unfried, J

    J. Unfried, J. Hauschild, and F. Pollmann, Phys. Rev. B 107, 155133 (2023)

    work page 2023

  47. [47]

    In the C4v symmetric case, the number of variational parameters is n(D) = d/8(D4 + 2D3 + 3D2 + 2D)

    work page

  48. [48]

    Here we initialize the boundary from the bulk tensors by contracting the corresponding bra- and ket- legs on the open boundary

    work page

  49. [49]

    Another type of reduced-rank approximation was intro- duced in [43] to decrease the computational cost

    work page

  50. [50]

    H. C. Jiang, Z. Y. Weng, and T. Xiang, Phys. Rev. Lett. 101, 090603 (2008)

    work page 2008

  51. [51]

    H. N. Phien, J. A. Bengua, H. D. Tuan, P. Corboz, and R. Orus, Phys. Rev. B92, 035142 (2015)

    work page 2015

  52. [52]

    Corboz, Phys

    P. Corboz, Phys. Rev. B94, 035133 (2016)

    work page 2016

  53. [53]

    Vanderstraeten, J

    L. Vanderstraeten, J. Haegeman, P. Corboz, and F. Ver- straete, Phys. Rev. B94, 155123 (2016)

    work page 2016

  54. [54]

    Liao, J.-G

    H.-J. Liao, J.-G. Liu, L. Wang, and T. Xiang, Phys. Rev. X 9, 031041 (2019)

    work page 2019

  55. [55]

    Francuz, N

    A. Francuz, N. Schuch, and B. Vanhecke, arXiv:2311.11894 (2023)

    work page arXiv 2023

  56. [56]

    A. W. Sandvik and H. G. Evertz, Phys. Rev. B 82, 024407 (2010)

    work page 2010

  57. [57]

    Hasik, jurajHasik/j1j2_ipeps_states (2024), https://github.com/jurajHasik/j1j2_ipeps_states

    J. Hasik, jurajHasik/j1j2_ipeps_states (2024), https://github.com/jurajHasik/j1j2_ipeps_states

    work page 2024

  58. [58]

    W.-J. Hu, F. Becca, A. Parola, and S. Sorella, Phys. Rev. B 88, 060402 (2013)

    work page 2013

  59. [59]

    Richter, R

    J. Richter, R. Zinke, and D. J. J. Farnell, Eur. Phys. J. B 88, 2 (2015)

    work page 2015

  60. [60]

    Wang and A

    L. Wang and A. W. Sandvik, Phys. Rev. Lett. 121, 107202 (2018)

    work page 2018

  61. [61]

    Hering, J

    M. Hering, J. Sonnenschein, Y. Iqbal, and J. Reuther, Phys. Rev. B99, 100405 (2019)

    work page 2019

  62. [62]

    Nomura and M

    Y. Nomura and M. Imada, Phys. Rev. X 11, 031034 (2021)

    work page 2021

  63. [63]

    Liu, S.-S

    W.-Y. Liu, S.-S. Gong, Y.-B. Li, D. Poilblanc, W.-Q. Chen, and Z.-C. Gu, Science Bulletin67, 1034 (2022)

    work page 2022

  64. [64]

    S. Dong, C. Wang, H. Zhang, M. Zhang, and L. He, arXiv:2407.15333 [cond-mat, physics:quant-ph] 10.48550/arXiv.2407.15333 (2024)

    work page doi:10.48550/arxiv.2407.15333 2024

  65. [65]

    Liu, S.-J

    W.-Y. Liu, S.-J. Du, R. Peng, J. Gray, and G. K.-L. Chan, Phys. Rev. Lett.133, 260404 (2024)

    work page 2024

  66. [66]

    X. Zhao, M. Li, Q. Xiao, J. Chen, F. Wang, L. Shen, M. Zhao, W. Wu, H. An, L. He, and X. Liang, inSC22: International Conference for High Performance Comput- ing, Networking, Storage and Analysis (IEEE Computer Society, Los Alamitos, CA, USA, 2022) pp. 1–15

    work page 2022

  67. [67]

    Liang, M

    X. Liang, M. Li, Q. Xiao, J. Chen, C. Yang, H. An, and L. He, Mach. Learn.: Sci. Technol.4, 015035 (2023)

    work page 2023

  68. [68]

    C. Roth, A. Szabó, and A. H. MacDonald, Phys. Rev. B 108, 054410 (2023)

    work page 2023

  69. [69]

    Chen and M

    A. Chen and M. Heyl, Nat. Phys.20, 1476 (2024)

    work page 2024

  70. [70]

    Rende, L

    R. Rende, L. L. Viteritti, L. Bardone, F. Becca, and S. Goldt, Commun Phys7, 1 (2024)

    work page 2024

  71. [71]

    Singh, R

    S. Singh, R. N. C. Pfeifer, and G. Vidal, Phys. Rev. B 83, 115125 (2011)

    work page 2011

  72. [72]

    Bauer, P

    B. Bauer, P. Corboz, R. Orús, and M. Troyer, Phys. Rev. B 83, 125106 (2011)

    work page 2011

  73. [73]

    van Alphen, S

    O. van Alphen, S. V. Kleijweg, J. Hasik, and P. Corboz, Phys. Rev. B111, 045105 (2025)

    work page 2025

  74. [74]

    Czarnik and J

    P. Czarnik and J. Dziarmaga, Phys. Rev. B92, 035152 (2015)

    work page 2015

  75. [75]

    Czarnik, M

    P. Czarnik, M. M. Rams, and J. Dziarmaga, Phys. Rev. B 94, 235142 (2016)

    work page 2016

  76. [76]

    Czarnik, J

    P. Czarnik, J. Dziarmaga, and P. Corboz, Phys. Rev. B 99, 035115 (2019)

    work page 2019