Pith. sign in

REVIEW 3 major objections 5 minor 86 references

A hierarchical search algorithm, Hiss, automatically finds tensor-network structures that beat fixed Tensor Train and Hierarchical Tucker formats by 2.5× to 100× — up to 1000× on radiation transport data.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 17:09 UTC pith:DKTYAK3K

load-bearing objection Useful structural-search algorithm with strong empirical results, but the stated error-guarantee is unproven and the comparisons are not yet apples-to-apples. the 3 major comments →

arxiv 2603.27856 v2 pith:DKTYAK3K submitted 2026-03-29 cs.CE cs.NAmath.NA

Hierarchical Search of Tree Tensor Networks for High-Dimensional Data

Zheng Guo , Aditya Deshpande , Xinyu Wang , Brian C. Kiedrowski , Alex A. Gorodetsky This is my paper
classification cs.CE cs.NAmath.NA MSC 15A6965F5568W20
keywords tensor network structure searchstructural roundingtree tensor networksindex reshapingentropy-guided clusteringstochastic samplinghigh-dimensional compressiontensor train
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper argues that the bottleneck in compressing high-dimensional data is not just choosing bond ranks but choosing the network topology itself, including how indices are reshaped. It formalizes this as the tensor-network structural rounding problem and presents Hiss, which alternates stochastic sub-network sampling with entropy-guided hierarchical refinement. If its claims hold, Hiss turns structure search into a practical, nearly polynomial-time step that can replace fixed formats in simulation pipelines, and discovered structures transfer across related data instances with small loss. The central empirical claim is compression gains of 2.5× to 100× over standard formats, with the strongest gains on time-evolving physical solutions.

Core claim

The paper introduces the tensor-network structural rounding problem (Problem 2): given a tree tensor network and an error tolerance, find the topology and index reshaping that minimize parameter count while keeping reconstruction error within tolerance. It then proposes Hiss, a two-layer algorithm — stochastic sampling of sub-networks at the global level and deterministic, heuristic-guided recursive search at the local level — that interpolates index clustering, structure search on clustered indices, and recursive refinement with index reshaping. The reported discovery is that this integrated search consistently finds structures with far higher compression than fixed TT and HT formats, and t

What carries the argument

The load-bearing mechanism is the combination of three routines: (1) ClusterIndices, which uses singular-value entropy of all-pair unfoldings to group correlated indices into virtual clusters, reducing dimensionality; ( (2) StructureSearch, an enumerative search over compatible index bi-partitions that estimates network cost via a rank-constrained optimization instead of full decompositions; and (3) TopReshape, a heuristic that factorizes free indices at low-entropy split points. These are orchestrated in RecSearch, where an error budget is split between the clustered search and recursive subnetwork refinements via the quadrature rule ε₁² + ε₂² = ε². This rule is what lets the search be hier

Load-bearing premise

The algorithm assumes that errors from independently optimized sub-network replacements add in quadrature (ε₁² + ε₂² = ε²), so the locally controlled errors guarantee a global error within tolerance; if local truncation errors compound instead, the returned network may violate the promised error bound.

What would settle it

Take a small tensor (e.g., d = 6, modes of size 8) and run Hiss to a nominal ε = 1e-2. Also enumerate all tree topologies and index reshapings with the same budget to get the true minimal network size. If Hiss's output exceeds ε when reconstructed against the exact tensor, or if the exhaustive search finds a structure with, say, 2× fewer parameters at the same error, then the central claim that Hiss finds near-optimal structures within tolerance is false in that regime.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Hiss can serve as a drop-in replacement for TT-rounding in existing tensor-network PDE solvers, reconfiguring the network as solution correlations evolve.
  • The 10% transferability result implies that a structure discovered once on representative data can be reused across a family of related datasets, making the search a one-time offline cost.
  • Entropy-guided index reshaping exposes latent low-rank structure, such as the x/y separability of flattened spatial grids, that static formats miss.
  • Empirical polynomial scaling (even with the exponential terms at the sub-network level) suggests that the approach remains tractable for 20+ dimensional data, unlike prior flat search methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The quadrature error-splitting rule is unproven as a global guarantee; local truncation errors can align, so the final network's error may exceed ε even if each sub-search respects its budget. Testing on adversarial or non-orthogonal data would reveal whether the reported compression ratios are meaningful at the stated tolerance.
  • Because the entropy heuristic is local, it may miss long-range correlations that only appear after multiple regrouping steps; a learned or global entropy surrogate could improve robustness on heterogeneous data.
  • The paper only treats tree (acyclic) networks; the bi-partition correspondence used for structure transformation does not extend to loopy graphs, so extending the approach to general tensor networks would require a different structural representation.
  • The generalization claim, if stable, suggests that one could precompute a library of reusable structures across simulation phases, essentially converting structure search into a dictionary lookup for common physics regimes.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper formalizes the tensor network structural rounding (TNSR) problem, in which a given tree tensor network must be replaced by a smaller-network representation within a prescribed relative error tolerance, and proposes Hiss, a hierarchical search algorithm that combines stochastic sub-network sampling with recursive, entropy-guided structure search and index reshaping. The algorithm is intended to scale polynomially with dimension and to discover structures that outperform fixed tensor train (TT) and hierarchical Tucker (HT) formats. Numerical evidence is reported on analytical functions, thermal radiation transport, neutron diffusion, and compressible Navier–Stokes data, with claimed compression improvements of 2.5–100× over TT/HT, up to 1000× in one application, plus cross-instance generalization of discovered structures.

Significance. If the claims are established, the work addresses an important practical bottleneck: fixed tensor-network topologies can be poorly matched to data, and structure search with index reshaping is a plausible route to substantially better compression. The paper is well written and the empirical suite is broad, including real physics applications and useful ablations against random search and against variants without reshaping. The formal problem definition (Problem 2) is a natural contribution, and the hierarchical decomposition idea is sensible. However, the central error-tolerance guarantee is not proved or verified at the required metric, and the evidence for 'near-optimality' is partly circular. The contribution would be solid if the authors close these gaps; as written, the headline compression ratios are not yet established at the stated tolerance.

major comments (3)
  1. [Algorithms 1–2; §4.1] The output guarantee stated in Algorithm 1 and Problem 2 — relative Frobenius error ≤ ε against the input network — is never proved. RecSearch splits ε into ε1, ε2 with ε1² + ε2² = ε² (Alg. 2, line 3) and assigns ε2/√k to each of k recursive calls; Algorithm 1 calls RecSearch Tθ times with ε/√Tθ. Such a composition is valid only if errors from different subnetwork replacements are orthogonal in the Frobenius inner product. Replaced subnets share boundary bonds and overlapping index sets, so this assumption is not justified in general; coherent errors could sum to a factor of √Tθ above ε. Moreover, the empirical error metric in §4.1 is computed on 3,000 sampled points against the original function, not against R_I(N), and the text concedes that errors 'occasionally exceed' the tolerance. Thus the reported compression ratios are not established at the stated tolerance. The authors should e
  2. [§4.2.1, Fig. 11; §4.2.2, Fig. 14] The evidence for 'near-optimality' is the relative compression ratio CR_i,j / CR_i,i being mostly below 1.0. This statistic is almost tautologically below 1.0 for any search that specializes structures to data: a structure optimized for data j is expected to underperform the structure optimized for data i when applied to data i, regardless of whether the search is near-optimal or merely overfits each tensor. To support the optimality claim, compare against a lower bound on the achievable size for the given error, or run exhaustive search on small subproblems; alternatively, report the same distribution for an independent stochastic baseline. As written, the 'near-optimality' inference is not supported.
  3. [§3.4.3, Eq. (4)] The cost model in Eq. (4) is used to select candidate structures, but the lazy-merge transformation is described only as maximizing reuse of the existing network. No statement or proof shows that the transformation actually realizes the ranks and errors assumed by Eq. (4). If merging, swapping, or splitting introduces truncation in addition to what the cost model accounts for, the final network's error is not controlled by the search. Please state explicitly how the transformation preserves the error budget, or verify the final network against the input network in the experiments.
minor comments (5)
  1. [Abstract and Highlights] The algorithm is called HIST in the abstract/highlights but Hiss in the main text and algorithms. Please standardize the notation.
  2. [Algorithm 2] The function signature lists (N_I, ε, C_θ, ϵ, k_θ) but the body uses d_θ from the enclosing context. Either include d_θ in the signature or clarify that it is a global parameter.
  3. [Algorithm 4, line 8] The 'continue' inside the while loop skips adding an index to the current cluster. This can leave clusters smaller than |I|/d_θ and, in principle, causes repeated iterations when the uniform draw is below ϵ. Please clarify the intended behavior (e.g., redraw until an index is added, or allow variable cluster sizes).
  4. [§3.3] Typo: 'orthognality center' should be 'orthogonality center'. There are also a few other typos ('comprimising', 'simplication') that a proofread pass should catch.
  5. [§4.1, Fig. 7] The reconstruction error plots would be more informative if they showed the exact relative error against the input TT network, not only against the original function on sampled points, since Problem 2 is defined relative to the input network.

Circularity Check

0 steps flagged

No significant circularity; the main issues are an unproven error-composition guarantee and a self-referential optimality diagnostic, neither of which reduces the derivation to its inputs.

full rationale

The paper's chain is an algorithmic construction: Hiss minimizes size (the reported compression numerator) under an error budget, and the empirical comparisons to TT/HT, random-structure, and non-hierarchical baselines are external to the construction. The recursive budget split in Algorithm 2 (Line 3: 'Split ε into ε1, ε2 such that ε1²+ε2²=ε²'; Line 6: 'RecSearch(G_i,j, ε2/√k)') is an unproven quadrature assumption, and Section 4.1 concedes that 'measured errors occasionally exceed the prescribed tolerance ε slightly'; that is a correctness/verification gap, not a circular reduction, because the budget split does not define the final error by construction. The relative-compression diagnostic CR_i,j/CR_i,i used to claim 'near-optimality' is self-referential and therefore weak evidence, but it is a measured ratio, not an input that is later reported as a prediction. The paper does rely on the authors' prior structure-search framework [50] as a subroutine and baseline, but the central contribution (hierarchical sampling + clustering + reshaping) is independently benchmarked, so this is normal inheritance rather than a load-bearing self-citation. No equation reduces to itself or to a fitted parameter renamed as a prediction.

Axiom & Free-Parameter Ledger

7 free parameters · 5 axioms · 0 invented entities

The algorithm introduces no new physical entities; all free parameters are heuristic hyperparameters chosen by hand. The key axiomatic loads are the error-budget composition rule and the entropy-as-cost proxy, both unproven.

free parameters (7)
  • N_θ (subnet node number) = 4
    Chosen by hand for all experiments; controls the size of sampled subnetworks and directly limits the per-subnetwork search effort.
  • d_θ (dimensionality threshold) = 4
    Chosen by hand; caps the number of index clusters at each recursion level, making the per-subnetwork search constant/exponential in d_θ rather than the global dimension.
  • C_θ (number of cluster candidates) = 5
    Number of candidate partitions tried in ClusterIndices; chosen by hand.
  • ϵ (exploration probability) = 0.1
    Probability of random exploration in the ϵ-greedy cluster selection; chosen by hand.
  • k_θ (factorization threshold) = 4
    Maximum number of factors in index reshaping; chosen by hand.
  • Top-K reshape candidates = 10
    Number of reshaping candidates retained per index in TopReshape; chosen by hand.
  • T_θ (sampling iterations) = 5 (default; 3 in neutron diffusion)
    Number of stochastic sampling iterations; tuned per problem (Section 4.3 sets T_θ=3 'for best performance').
axioms (5)
  • domain assumption Tree tensor network unfoldings are valid only for index sets that form a cut; the algorithm restricts to such cuts.
    Section 2.2 defines unfoldings via cuts; the structure search only considers bi-partitions compatible with a tree, which may exclude useful non-cut groupings.
  • ad hoc to paper Singular-value entropy (effective rank, Eq. 3) is a reliable proxy for compression potential of an index grouping.
    Section 3.3.1 introduces H(M) as the clustering cost without theoretical justification; clustering quality is only validated indirectly through final compression ratios.
  • ad hoc to paper Error budget splitting ε_1² + ε_2² = ε² yields a global error below ε.
    Algorithm 2 Line 3 assigns ε_1 to structure search and ε_2 to recursive refinement, assuming errors from independent refinements combine in quadrature; no proof is provided.
  • ad hoc to paper The lazy-merge structure transformation preserves the data within the error budget.
    Section 3.4.3 describes merging conflicting nodes to apply a target split, but offers no error analysis for the transformation; it is assumed not to exceed the allocated ε_1.
  • standard math The bi-partition compatibility criterion (nested/disjoint) is sufficient to generate all valid tree structures.
    Section 3.4.1 relies on the standard correspondence between tree tensor networks and compatible index bi-partitions.

pith-pipeline@v1.3.0-alltime-deepseek · 29997 in / 13605 out tokens · 130861 ms · 2026-08-02T17:09:03.091081+00:00 · methodology

0 comments
read the original abstract

Tensor network methods provide a scalable solution to represent high-dimensional data. However, their efficacy is often limited by static, expert-defined structures that fail to adapt to evolving data correlations. We address this limitation by formalizing the structural rounding problem for tree tensor networks and introducing a hierarchical search algorithm HIST, which automatically identifies optimized structures with index reshaping for input tree tensor networks. To navigate the combinatorial explosion of the structural search space, HIST integrates stochastic sub-network sampling with hierarchical refinement. This approach utilizes entropy-guided index clustering to reduce dimensionality and targeted reshaping to expose latent data correlations. Numerical experiments on analytical functions and real-world physics applications, including thermal radiation transport, neutron diffusion, and computational fluid dynamics, demonstrate that HIST exhibits empirical polynomial scaling with dimensionality relative to the sampling budget, bypassing the scalability barriers in prior work. HIST achieves compression ratios $2.5\times$ to $100\times$ higher than standard fixed formats such as Tensor Trains and Hierarchical Tuckers (peaking at $1000\times$). Furthermore, HIST discovers structures that generalize effectively: applying a structure optimized for one data instance to a related target data typically maintains compression performance within $10\%$ of the result obtained by performing structure search on that target data. These results highlight HIST as a robust, automated tool for adaptive data representation and high-dimensional simulation compression with tensor network methods.

Figures

Figures reproduced from arXiv: 2603.27856 by Aditya Deshpande, Alex A. Gorodetsky, Brian C. Kiedrowski, Xinyu Wang, Zheng Guo.

Figure 1
Figure 1. Figure 1: Traditional tensor network structure search [50] with the extension to support index reshaping. It enumerates di [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Hierarchical tensor network structure search (H [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Cost computation for index pairs. Given a tree network, the process orthonormalizes the nodes relative to [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of singular value computation in tensor networks with clustered indices. The initial tensor network has 6 free indices that forms [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Resolution of structural conflicts via lazy merge during structure transformations. The modifying sub-networks are bolded and highlighted [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Compression ratio over TT versus dimensions. We choose dimensions from 3 to 20, and error tolerance [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Reconstruction error versus dimensions. We choose dimensions from 3 to 20, and error tolerance [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Search time versus dimensions. We choose dimensions from 3 to 20, and error tolerance [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Examples of solutions (top) and discovered structures (bottom) for the line source problem at three timesteps 50, 200, and 350 with error [PITH_FULL_IMAGE:figures/full_fig_p018_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: CR over input TT across timesteps for line source data at [PITH_FULL_IMAGE:figures/full_fig_p019_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Distribution of relative compression ratios ( [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Thermal radiation transport equation solutions (top) and discovered structures (bottom) for the crooked pipe problem [5] at time steps [PITH_FULL_IMAGE:figures/full_fig_p021_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Comparison of compression ratios over TT (left) and compression ratios over data (right) across timesteps for H [PITH_FULL_IMAGE:figures/full_fig_p021_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Distribution of relative compression ratios ( [PITH_FULL_IMAGE:figures/full_fig_p022_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Example core layout (left) and the keff values (right) for three materials. Each color stands for a different material. decomposition [6] for the native indices, with an internal node connecting the spatial and the two angular dimensions. The discrepancy between machine-discovered structures and expert-defined structures suggests the advantages of automatic structure search, as expert intuition might fail… view at source ↗
Figure 16
Figure 16. Figure 16: (a) Search time scales polynomially with the number of materials. (b) Reconstruction error around pre-defined error tolerance. (c, d) [PITH_FULL_IMAGE:figures/full_fig_p024_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: CR over data for error bounds ε = 10−3 (top) and ε = 10−4 (bottom). Compression ratios scale exponentially with the number of materials. Hiss with index reshaping consistently delivers the highest compression efficiency at both error tolerances, while the HT and TT methods provide lower ratios, particularly at higher material counts. standard TT and HT formats plateau or even degrade, and fail to exceed 2… view at source ↗
Figure 18
Figure 18. Figure 18: An example solution of 3-D Navier-Stokes problems [1] [PITH_FULL_IMAGE:figures/full_fig_p026_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Ablation analysis of algorithmic performance, illustrating the trade-o [PITH_FULL_IMAGE:figures/full_fig_p027_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Distribution of relative compression ratios (left: [PITH_FULL_IMAGE:figures/full_fig_p027_20.png] view at source ↗

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

86 extracted references · 7 linked inside Pith

  1. [1]

    Takamoto, T

    M. Takamoto, T. Praditia, R. Leiteritz, D. MacKinlay, F. Alesiani, D. Pflüger, M. Niepert, Pdebench: An exten- sive benchmark for scientific machine learning, Advances in Neural Information Processing Systems 35 (2022) 1596–1611

    2022

  2. [2]

    M. H. Zawawi, A. Saleha, A. Salwa, N. Hassan, N. M. Zahari, M. Z. Ramli, Z. C. Muda, A review: Fundamentals of computational fluid dynamics (cfd), in: AIP conference proceedings, V ol. 2030, AIP Publishing LLC, 2018, p. 020252

    2030

  3. [3]

    Bassett, B

    B. Bassett, B. Kiedrowski, Comparison of meshless and high-order polynomial functions for neutron transport with streamline-upwind petrov-galerkin stabilization, Tech. rep., Lawrence Livermore National Lab.(LLNL), Livermore, CA (United States) (2019). 27

    2019

  4. [4]

    A. M. Cox, S. C. Harris, A. E. Kyprianou, M. Wang, Monte carlo methods for the neutron transport equation, SIAM/ASA Journal on Uncertainty Quantification 10 (2) (2022) 775–825

    2022

  5. [5]

    A. A. Gorodetsky, P. D. Mullen, A. Deshpande, J. C. Dolence, C. D. Meyer, J. M. Miller, L. F. Roberts, Thermal radiation transport with tensor trains, arXiv preprint arXiv:2503.18056 (2025)

    Pith/arXiv arXiv 2025

  6. [6]

    T. G. Kolda, B. W. Bader, Tensor decompositions and applications, SIAM review 51 (3) (2009) 455–500

    2009

  7. [7]

    I. V . Oseledets, E. E. Tyrtyshnikov, Breaking the curse of dimensionality, or how to use svd in many dimensions, SIAM Journal on Scientific Computing 31 (5) (2009) 3744–3759

    2009

  8. [8]

    Altman, M

    N. Altman, M. Krzywinski, The curse (s) of dimensionality, Nat Methods 15 (6) (2018) 399–400

    2018

  9. [9]

    Bachmayr, Low-rank tensor methods for partial differential equations, Acta Numerica 32 (2023) 1–121

    M. Bachmayr, Low-rank tensor methods for partial differential equations, Acta Numerica 32 (2023) 1–121

    2023

  10. [10]

    S. V . Dolgov, Tt-gmres: on solution to a linear system in the structured tensor format, arXiv preprint arXiv:1206.5512 (2012)

    Pith/arXiv arXiv 2012

  11. [11]

    Rodgers, D

    A. Rodgers, D. Venturi, Tensor approximation of functional differential equations, Physical Review E 110 (1) (2024) 015310

    2024

  12. [12]

    D. P. Truong, M. I. Ortega, I. Boureima, G. Manzini, K. Ø. Rasmussen, B. S. Alexandrov, Tensor networks for solving the time-independent boltzmann neutron transport equation, Journal of Computational Physics 507 (2024) 112943

    2024

  13. [13]

    L. R. Tucker, Some mathematical notes on three-mode factor analysis, Psychometrika 31 (3) (1966) 279–311. doi:10.1007/BF02289464

    doi:10.1007/bf02289464 1966

  14. [14]

    I. V . Oseledets, Tensor-train decomposition, SIAM Journal on Scientific Computing 33 (5) (2011) 2295–2317

    2011

  15. [15]

    Grasedyck, Hierarchical singular value decomposition of tensors, SIAM Journal on Matrix Analysis and Applications 31 (4) (2010) 2029–2054.doi:10.1137/090764189

    L. Grasedyck, Hierarchical singular value decomposition of tensors, SIAM Journal on Matrix Analysis and Applications 31 (4) (2010) 2029–2054.doi:10.1137/090764189

    doi:10.1137/090764189 2010

  16. [16]

    Falcó, W

    A. Falcó, W. Hackbusch, A. Nouy, Tree-based tensor formats, SeMA Journal 78 (2) (2021) 159–173

    2021

  17. [17]

    Veeramacheneni, M

    L. Veeramacheneni, M. Wolter, H. Kuehne, J. Gall, Canonical rank adaptation: An efficient fine-tuning strategy for vision transformers, in: Forty-second International Conference on Machine Learning, 2025

    2025

  18. [18]

    Y . Yang, J. Zhou, N. Wong, Z. Zhang, Loretta: Low-rank economic tensor-train adaptation for ultra-low- parameter fine-tuning of large language models, in: Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (V olume 1: Long Papers), 2024, pp. 3161–3176

    2024

  19. [19]

    Ghiasvand, M

    S. Ghiasvand, M. Alizadeh, R. Pedarsani, Decentralized low-rank fine-tuning of large language models, in: Proceedings of the 1st Workshop for Research on Agent Language Models (REALM 2025), 2025, pp. 334–345

    2025

  20. [20]

    Zheng, C

    H. Zheng, C. Zhou, Z. Shi, A. L. de Almeida, Subttd: Doa estimation via sub-nyquist tensor train decomposition, IEEE Signal Processing Letters 29 (2022) 1978–1982

    2022

  21. [21]

    Q. Xie, Z. Wang, F. Wen, J. He, T.-K. Truong, Coarray tensor train decomposition for bistatic mimo radar with uniform planar array, IEEE Transactions on Antennas and Propagation (2025)

    2025

  22. [22]

    M. Wang, D. Hong, Z. Han, J. Li, J. Yao, L. Gao, B. Zhang, J. Chanussot, Tensor decompositions for hyper- spectral data processing in remote sensing: A comprehensive review, IEEE Geoscience and Remote Sensing Magazine 11 (1) (2023) 26–72

    2023

  23. [23]

    Ravishankar, M

    J. Ravishankar, M. Sharma, A hierarchical approach for lossy light field compression with multiple bit rates based on tucker decomposition via random sketching, IEEE Access 10 (2022) 56677–56690. 28

    2022

  24. [24]

    W. A. Sands, W. Guo, J.-M. Qiu, T. Xiong, High-order adaptive rank integrators for multiscale linear kinetic transport equations in the hierarchical tucker format, SIAM Journal on Scientific Computing 47 (6) (2025) A3383–A3412

    2025

  25. [25]

    Ghahremani, H

    B. Ghahremani, H. Babaee, A deim tucker tensor cross algorithm and its application to dynamical low-rank approximation, Computer Methods in Applied Mechanics and Engineering 423 (2024) 116879

    2024

  26. [26]

    Ghahremani, H

    B. Ghahremani, H. Babaee, Cross interpolation for solving high-dimensional dynamical systems on low-rank tucker and tensor train manifolds, Computer Methods in Applied Mechanics and Engineering 432 (2024) 117385

    2024

  27. [27]

    F. Pan, P. Zhang, Simulation of quantum circuits using the big-batch tensor network method, Physical Review Letters 128 (3) (2022) 030501

    2022

  28. [28]

    Kalachev, P

    G. Kalachev, P. Panteleev, M.-H. Yung, Multi-tensor contraction for xeb verification of quantum circuits, arXiv preprint arXiv:2108.05665 (2021)

    Pith/arXiv arXiv 2021

  29. [29]

    Huang, F

    C. Huang, F. Zhang, M. Newman, J. Cai, X. Gao, Z. Tian, J. Wu, H. Xu, H. Yu, B. Yuan, et al., Classical simulation of quantum supremacy circuits, arXiv preprint arXiv:2005.06787 (2020)

    Pith/arXiv arXiv 2005

  30. [30]

    Nakatani, G

    N. Nakatani, G. K. Chan, Efficient tree tensor network states (ttns) for quantum chemistry: Generalizations of the density matrix renormalization group algorithm, The Journal of chemical physics 138 (13) (2013)

    2013

  31. [31]

    V . Murg, F. Verstraete, R. Schneider, P. R. Nagy, O. Legeza, Tree tensor network state with variable tensor order: An efficient multireference method for strongly correlated systems, Journal of Chemical Theory and Computation 11 (3) (2015) 1027–1036

    2015

  32. [32]

    Gunst, F

    K. Gunst, F. Verstraete, D. Van Neck, Three-legged tree tensor networks with su (2) and molecular point group symmetry, Journal of chemical theory and computation 15 (5) (2019) 2996–3007

    2019

  33. [33]

    Ferrari, G

    G. Ferrari, G. Magnifico, S. Montangero, Adaptive-weighted tree tensor networks for disordered quantum many- body systems, Physical Review B 105 (21) (2022) 214201

    2022

  34. [34]

    K. Seki, T. Hikihara, K. Okunishi, Tensor-network strong-disorder renormalization groups for random quantum spin systems in two dimensions, Physical Review B 102 (14) (2020) 144439

    2020

  35. [35]

    Hikihara, H

    T. Hikihara, H. Ueda, K. Okunishi, K. Harada, T. Nishino, Automatic structural optimization of tree tensor networks, Physical Review Research 5 (1) (2023) 013031

    2023

  36. [36]

    Ke, Tree tensor network state approach for solving hierarchical equations of motion, The Journal of Chemical Physics 158 (21) (2023)

    Y . Ke, Tree tensor network state approach for solving hierarchical equations of motion, The Journal of Chemical Physics 158 (21) (2023)

    2023

  37. [37]

    Yang, Z.-C

    S. Yang, Z.-C. Gu, X.-G. Wen, Loop optimization for tensor network renormalization, Physical review letters 118 (11) (2017) 110504

    2017

  38. [38]

    J. Gray, G. K.-L. Chan, Hyperoptimized approximate contraction of tensor networks with arbitrary geometry, Physical Review X 14 (1) (2024) 011009

    2024

  39. [39]

    Y . Gao, H. Zhai, J. Gray, R. Peng, G. Park, W.-Y . Liu, E. F. Kjønstad, G. K.-L. Chan, Fermionic tensor network contraction for arbitrary geometries, Physical Review Research 7 (2) (2025) 023193

    2025

  40. [40]

    Watanabe, H

    R. Watanabe, H. Manabe, T. Hikihara, H. Ueda, Ttnopt: Tree tensor network package for high-rank tensor compression, arXiv preprint arXiv:2505.05908 (2025)

    arXiv 2025

  41. [41]

    Dektor, D

    A. Dektor, D. Venturi, Coordinate-adaptive integration of pdes on tensor manifolds, Communications on Applied Mathematics and Computation 7 (4) (2025) 1562–1579. 29

    2025

  42. [42]

    Dektor, D

    A. Dektor, D. Venturi, Tensor rank reduction via coordinate flows, Journal of Computational Physics 491 (2023) 112378

    2023

  43. [43]

    Rodgers, A

    A. Rodgers, A. Dektor, D. Venturi, Adaptive integration of nonlinear evolution equations on tensor manifolds, Journal of Scientific Computing 92 (2) (2022) 39

    2022

  44. [44]

    Oseledets, E

    I. Oseledets, E. Tyrtyshnikov, Tt-cross approximation for multidimensional arrays, Linear Algebra and its Ap- plications 432 (1) (2010) 70–88

    2010

  45. [45]

    Handschuh, Numerical methods in tensor networks, Ph.D

    S. Handschuh, Numerical methods in tensor networks, Ph.D. thesis, Dissertation, Leipzig, Universität Leipzig, 2015 (2015)

    2015

  46. [46]

    C. Li, Z. Sun, Evolutionary topology search for tensor network decomposition, in: International conference on machine learning, PMLR, 2020, pp. 5947–5957

    2020

  47. [47]

    C. Li, J. Zeng, Z. Tao, Q. Zhao, Permutation search of tensor network structures via local sampling, in: Interna- tional conference on machine learning, PMLR, 2022, pp. 13106–13124

    2022

  48. [48]

    C. Li, J. Zeng, C. Li, C. F. Caiafa, Q. Zhao, Alternating local enumeration (tnale): Solving tensor network structure search with fewer evaluations, in: International conference on machine learning, PMLR, 2023, pp. 20384–20411

    2023

  49. [49]

    Hashemizadeh, M

    M. Hashemizadeh, M. Liu, J. Miller, G. Rabusseau, Adaptive learning of tensor network structures, arXiv preprint arXiv:2008.05437 (2020)

    Pith/arXiv arXiv 2008

  50. [50]

    Z. Guo, A. Deshpande, B. Kiedrowski, X. Wang, A. Gorodetsky, Tensor network structure search with program synthesis, arXiv preprint arXiv:2502.02711 (2025)

    Pith/arXiv arXiv 2025

  51. [51]

    S. Börm, L. Grasedyck, W. Hackbusch, Introduction to hierarchical matrices with applications, Engineering analysis with boundary elements 27 (5) (2003) 405–422

    2003

  52. [52]

    Z. Chen, R. Cai, F. Xie, J. Qiao, A. Wu, Z. Li, Z. Hao, K. Zhang, Learning discrete latent variable structures with tensor rank conditions, Advances in Neural Information Processing Systems 37 (2024) 17398–17427

    2024

  53. [53]

    Anandkumar, R

    A. Anandkumar, R. Ge, D. J. Hsu, S. M. Kakade, M. Telgarsky, et al., Tensor decompositions for learning latent variable models., J. Mach. Learn. Res. 15 (1) (2014) 2773–2832

    2014

  54. [54]

    Linderman, R

    S. Linderman, R. Adams, Discovering latent network structure in point process data, in: International conference on machine learning, PMLR, 2014, pp. 1413–1421

    2014

  55. [55]

    B. N. Khoromskij, O (d log n)-quantics approximation of n-d tensors in high-dimensional numerical modeling, Constructive Approximation 34 (2) (2011) 257–280

    2011

  56. [56]

    C. J. Hillar, L.-H. Lim, Most tensor problems are np-hard, Journal of the ACM (JACM) 60 (6) (2013) 1–39

    2013

  57. [57]

    Zheng, X.-L

    Y .-B. Zheng, X.-L. Zhao, J. Zeng, C. Li, Q. Zhao, H.-C. Li, T.-Z. Huang, Svdinstn: A tensor network paradigm for efficient structure search from regularized modeling perspective, in: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, 2024, pp. 26254–26263

    2024

  58. [58]

    M. Wang, B. Yu, S. Zhang, L. Mi, W. Wang, Y . Wang, P. Jia, X. Wei, Z. Xu, R. Guo, et al., Renormalization group guided tensor network structure search, arXiv preprint arXiv:2512.24663 (2025)

    arXiv 2025

  59. [59]

    Juels, M

    A. Juels, M. Wattenberg, Stochastic hillclimbing as a baseline method for evaluating genetic algorithms, Ad- vances in Neural Information Processing Systems 8 (1995)

    1995

  60. [60]

    G. Evenbly, A practical guide to the numerical implementation of tensor networks i: Contractions, decomposi- tions, and gauge freedom, Frontiers in Applied Mathematics and Statistics 8 (2022) 806549. 30

    2022

  61. [61]

    O. Roy, M. Vetterli, The effective rank: A measure of effective dimensionality, in: 2007 15th European signal processing conference, IEEE, 2007, pp. 606–610

    2007

  62. [62]

    S. R. White, Density matrix renormalization group algorithms with a single center site, Physical Review B—Condensed Matter and Materials Physics 72 (18) (2005) 180403

    2005

  63. [63]

    S. R. White, Density-matrix algorithms for quantum renormalization groups, Physical review b 48 (14) (1993) 10345

    1993

  64. [64]

    Evenbly, Gauge fixing, canonical forms, and optimal truncations in tensor networks with closed loops, Phys- ical Review B 98 (8) (2018) 085155

    G. Evenbly, Gauge fixing, canonical forms, and optimal truncations in tensor networks with closed loops, Phys- ical Review B 98 (8) (2018) 085155

    2018

  65. [65]

    Morrison, S

    D. Morrison, S. Jacobson, J. Sauppe, E. Sewell, Branch-and-bound algorithms: a survey of recent advances in searching, branching, and pruning. discret optim 19: 79–102 (2016)

    2016

  66. [66]

    Ballani, L

    J. Ballani, L. Grasedyck, M. Kluge, Black box approximation of tensors in hierarchical tucker format, Linear algebra and its applications 438 (2) (2013) 639–657

    2013

  67. [67]

    Ryzhakov, A

    G. Ryzhakov, A. Chertkov, A. Basharin, I. Oseledets, Black-box approximation and optimization with hierarchi- cal tucker decomposition, arXiv preprint arXiv:2402.02890 (2024)

    Pith/arXiv arXiv 2024

  68. [68]

    G. B. Rybicki, A. P. Lightman, Radiative processes in astrophysics, John Wiley & Sons, 2024

    2024

  69. [69]

    Y .-F. Jiang, An implicit finite volume scheme to solve the time-dependent radiation transport equation based on discrete ordinates, The Astrophysical Journal Supplement Series 253 (2) (2021) 49

    2021

  70. [70]

    Kurzer-Ogul, B

    K. Kurzer-Ogul, B. M. Haines, D. S. Montgomery, S. Pandolfi, J. P. Sauppe, A. F. Leong, D. Hodge, P. M. Kozlowski, S. Marchesini, E. Cunningham, et al., Radiation and heat transport in divergent shock–bubble inter- actions, Physics of plasmas 31 (3) (2024)

    2024

  71. [71]

    Einkemmer, K

    L. Einkemmer, K. Kormann, J. Kusch, R. G. McClarren, J.-M. Qiu, A review of low-rank methods for time- dependent kinetic simulations, Journal of Computational Physics 538 (2025) 114191

    2025

  72. [72]

    M. K. Bhattacharyya, D. Radice, A finite element method for angular discretization of the radiation transport equation on spherical geodesic grids, Journal of Computational Physics 491 (2023) 112365

    2023

  73. [73]

    B. S. Southworth, S. Olivier, H. Park, T. Buvoli, One-sweep moment-based semi-implicit-explicit integration for gray thermal radiation transport, Journal of Computational Physics 517 (2024) 113349

    2024

  74. [74]

    Z. Wang, Z. He, L. Mu, S. Dong, Parallel algorithm and its convergence of spatial domain decomposition of discrete ordinates method for solving radiation heat transfer problem, Chinese Journal of Aeronautics 28 (1) (2015) 77–85

    2015

  75. [75]

    Brunner, L

    S. Brunner, L. Einkemmer, T. Haut, Domain decomposition dynamical low-rank for multi-dimensional radiative transfer equations, arXiv preprint arXiv:2602.14854 (2026)

    arXiv 2026

  76. [76]

    J. P. Jessee, W. A. Fiveland, L. H. Howell, P. Colella, R. B. Pember, An adaptive mesh refinement algorithm for the radiative transport equation, Journal of computational Physics 139 (2) (1998) 380–398

    1998

  77. [77]

    Velarde, F

    P. Velarde, F. Ogando, Radiation transport in amr, in: Adaptive Mesh Refinement-Theory and Applications: Proceedings of the Chicago Workshop on Adaptive Mesh Refinement Methods, Sept. 3–5, 2003, Springer, 2005, pp. 271–280

    2003

  78. [78]

    Allaire, Y

    G. Allaire, Y . Capdeboscq, Homogenization of a spectral problem in neutronic multigroup diffusion, Computer methods in applied mechanics and engineering 187 (1-2) (2000) 91–117

    2000

  79. [79]

    W. Xiao, X. Liu, J. Zu, X. Chai, H. He, T. Zhang, Operator inference driven data assimilation for high fidelity neutron transport, Computer Methods in Applied Mechanics and Engineering 430 (2024) 117214. 31

    2024

  80. [80]

    Wilson, M

    S. Wilson, M. Eaton, J. Kópházi, A symmetric interior-penalty discontinuous galerkin isogeometric analysis spatial discretization of the self-adjoint angular flux form of the neutron transport equation, Computer Methods in Applied Mechanics and Engineering 432 (2024) 117414

    2024

Showing first 80 references.