Breaking the Curse of Dimensionality: Solving Configurational Integrals for Crystalline Solids by Tensor Networks

Benjamin Nebgen; Boian S. Alexandrov; Derek DeSantis; Dimiter N. Petsev; Duc P. Truong; Kim {\O}. Rasmussen

REVIEW 1 major objections 1 minor 32 references

Breaking the Curse of Dimensionality: Solving Configurational Integrals for Crystalline Solids by Tensor Networks

T0 review · 1 major / 1 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 A tensor-train method computes configurational integrals for crystals in seconds, reproducing molecular dynamics results for energies, pressures, and the tin phase boundary.

desk verdict The rank-1 TT formula is an Einstein-crystal model in disguise, and the paper overclaims, but the symmetry-adapted TT-cross idea and the rank-2 extension are worth a serious look. read the letter →

arxiv 2505.21826 v1 pith:4BQQOLGC submitted 2025-05-27 cond-mat.stat-mech

Duc P. Truong , Benjamin Nebgen , Derek DeSantis , Dimiter N. Petsev , Kim {O}. Rasmussen , Boian S. Alexandrov This is my paper

classification cond-mat.stat-mech

keywords configurationalintegraltensortraindecompositionTT-crossinterpolationBoltzmannfactorcrystallinesolidsfreeenergycalculationphaseboundarymachinelearningpotentials

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that the 3N-dimensional configurational integral that governs crystal thermodynamics can be evaluated by a tensor-network approximation instead of simulation. The authors reformulate the Boltzmann integrand as a high-dimensional tensor and compress it with tensor-train (TT) cross interpolation, using specially designed rank-1 and rank-2 schemes that exploit the sharply peaked shape of the Boltzmann factor and the symmetry of identical-particle lattices. Applied to copper, argon, and tin with different interatomic potentials, the method reproduces molecular dynamics values for internal energy, pressure, and the α–β tin phase boundary while taking seconds of computation. If the claim holds, equilibrium thermodynamic functions such as Helmholtz free energies and equations of state become directly computable from the potential energy surface without long trajectories.

What carries the argument

The central object is the tensor-train (TT) decomposition, which compresses a high-dimensional tensor into a chain of small three-way cores, together with the TT-cross algorithm that builds this decomposition from a few adaptively chosen fibers using the maxvol principle. For the Boltzmann integrand, the paper replaces generic TT-cross with two symmetry-aware constructions: a rank-1 product of the peak-point fibers (Eq. 7), and a rank-2 construction in which 2×2 intersection matrices $S_k$ of the unfolded tensor are inverted to form the TT cores. The peak location $\bar{q}$ and the lattice symmetry of FCC, BCC, and diamond-cubic crystals make the cores identical across dimensions, so the whole integral reduces to a handful of one-dimensional quadratures.

What would settle it

Compute the rank-1 and rank-2 TT configurational integrals for a crystal with pronounced anharmonicity—for example, near the melting line or for a soft molecular crystal—and compare with a converged high-rank TT-SVD or thermodynamic-integration MD result; if the rank-2 prediction diverges from the reference as temperature increases while the reference stays converged, the central claim that these low ranks capture the integrand is falsified.

Watch

Extended reading notes

Core claim

The central claim is that for crystalline solids, the configurational integral $Z_N = \int e^{-\beta U(\mathbf{q})}\,d\mathbf{q}$ can be evaluated accurately by a tensor-train cross interpolation of the Boltzmann factor $e^{-\beta U}$. Because the integrand is sharply peaked at the minimum-energy configuration, the rank-1 approximation represents it as a product of one-dimensional fibers through the peak, and lattice symmetry reduces the required fiber calculations to a single representative one. The rank-2 scheme captures the leading off-diagonal correlations by selecting two points on the super-diagonal and building each TT core from four fibers, with the off-diagonal entries assumed near zero. The paper reports that these approximations match molecular dynamics for internal energy and pressure–temperature curves for Cu and Ar, and reconstruct the α–β Sn phase boundary, all in seconds of computation.

Load-bearing premise

The load-bearing assumption is that the Boltzmann factor $e^{-\beta U}$ is so sharply peaked that the tensor-train ranks 1 or 2 suffice, i.e., that off-diagonal integrand values are effectively negligible; if atomic displacements are strongly correlated, the approximation loses accuracy and the method would need higher ranks.

Editorial extensions

If this is right

Equilibrium thermodynamic functions of crystalline solids—Helmholtz free energy, internal energy, pressure, equations of state—can be computed directly from a potential energy surface in seconds rather than by equilibrating molecular dynamics runs.
Machine-learned many-body potentials that go beyond pairwise additivity can be used in these integrations, as demonstrated by the HIP-NN argon results, so the accuracy of the potential is no longer bought at the price of an intractable integral.
Solid–solid phase boundaries can be mapped by comparing Gibbs free energies at a small fraction of the molecular-dynamics cost: the tin study used about 5 core hours versus 2560 for MD.
The same TT-cross construction, with more than two points on the super-diagonal, is a direct route to higher-rank approximations if the integrand requires them (a generalization noted at the end of Appendix C).
The method supplies a deterministic, trajectory-free alternative to MD for thermodynamic averages, which may help in settings where statistical convergence of MD is slow.

Reading between the lines

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

Because the rank-1/rank-2 ansatz treats off-diagonal Boltzmann weights as negligible, the method's accuracy should degrade as the crystal softens—near the melting line or for molecular crystals with large-amplitude librations; a systematic rank-versus-temperature study would test this boundary.
The symmetry reduction to a single 1D fiber (Eq. 7) suggests a natural diagnostic: the difference between the rank-1 and rank-2 results for a given system measures the importance of the neglected correlations, and could be used as an adaptive rank-selection criterion.
The approach could plausibly be paired with thermodynamic integration or reweighting to compute free-energy differences between phases without solving the full integral, since the phase-boundary comparison already relies only on $Z_N$ differences.
One testable extension is to dilute alloys or defective crystals where translational symmetry is broken; the rank necessary for a given accuracy would quantify how much symmetry is required for the method to remain practical.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Here's my read. The genuinely new thing is the symmetry-adapted fiber selection for TT-cross on Boltzmann integrands. Skipping the full tensor and sampling only fibers through the super-diagonal peak is a practical trick that makes the computation fast, and applying it with a machine-learning potential (the Ar/HIP-NN test) is a nice stress test. The phase-boundary result for Sn with a MEAM potential is also encouraging, and the reported speed-up (5 core hours vs. 2560 for MD) is real if not fully fleshed out.

But the central claim needs a serious caveat. As you noted, the rank-1 formula in Eq. (4) is exactly an independent-oscillator (Einstein crystal) approximation. It replaces the potential with the sum of single-coordinate fibers and discards all off-diagonal Hessian couplings. The paper never acknowledges this equivalence, and it does not compare against the Einstein crystal or harmonic approximations, which is a significant omission. The agreement with MD for internal energy and pressure is not, by itself, evidence that the full configurational integral is being evaluated; for a harmonic crystal, internal energy is correct by equipartition regardless of the rank-1 truncation, so Fig. 3(a) is weaker evidence than it looks. The rank-2 extension adds only one super-diagonal pivot and still cannot represent the full phonon covariance. No convergence studies in rank, grid spacing, δ, or system size are reported, and the MD comparisons lack error bars. The rank-2 construction in Appendix C also looks typo-ridden (two G_k(2,:,1) entries appear, one likely should be G_k(2,:,2)). No code or data are provided, which blocks reproducibility.

The paper is not circular: it compares against independent MD with the same potentials and has no fitted free parameters beyond convergence tolerances. But the phrase "solving configurational integrals" in the conclusions is too strong. What the paper shows is that a low-rank TT ansatz can reproduce MD thermodynamics for these specific crystals — which is useful, but it is an approximation with an uncontrolled error.

Who benefits? People working on tensor-network methods for statistical mechanics, and maybe materials screening if the errors are characterized. As it stands, the evidence supports a modest claim: a fast approximate alternative to MD for crystalline solids, not a curse-breaking solver. With major revision — add Einstein-crystal/reference comparisons, error bars, convergence tests, fix Appendix C, and release code — it could be a solid methods paper. I’d send it to peer review rather than desk reject, but I would not cite it in its current form.

Referee Report

1 major / 1 minor

Summary. The manuscript proposes a tensor-network (TT-cross) method for evaluating the high-dimensional configurational integral Z_N of identical-particle crystals. The integrand e^{-βU(q)} is discretized on a grid centered at the potential minimum, and the authors introduce rank-1 and rank-2 TT-cross interpolation schemes that exploit the sharp peak and super-diagonal dominance of the Boltzmann factor. The method is tested by computing internal energy and pressure for crystalline Cu (tight-binding potential) and Ar (HIP-NN machine-learning potential), and the α→β phase boundary of Sn (MEAM potential), with comparisons against molecular dynamics simulations. The paper reports good agreement with MD and very large computational speedups.

Significance. If the central accuracy claim were established, the method would be a notable advance in computing free energies and equations of state for solids, bypassing the exponential cost of direct integration. The paper has several concrete strengths: the rank-1 maxvol derivation in Appendix B is mathematically correct; the cubature summation rule in Eq. (3) is valid; the method contains no fitted parameters aimed at matching the target thermochemical quantities; and the numerical experiments compare against independent MD simulations using the same potentials. However, the significance is currently limited by the lack of any validation of the low-rank assumption itself, and by the absence of convergence studies or error estimates. The speed advantage is real but secondary if the accuracy is not quantified. The manuscript is therefore potentially significant but requires substantial additional evidence before the central claim can be accepted.

major comments (1)

[Conclusion and general claim] The conclusion states that the method 'successfully evaluate configurational integrals in crystalline solids from first principles.' This is overstated given the evidence. The method is an approximation with uncontrolled rank truncation, and the tested potentials are empirical or machine-learned, not first-principles. The central claim should be reformulated as: the low-rank TT-cross approximation reproduces certain thermodynamic observables for the tested systems within the accuracy of the presented comparisons, subject to validation of the truncation error.

minor comments (1)

[Appendix C] The rank-2 construction says 'all the matrix M_k, and all TT-cores G_k; k ∈ {2, ..., 3N−1} are identical' for symmetric structures, but this is only true if the fibers are the same for all dimensions; the statement should be justified or qualified. Also, the notation S_k is introduced in the main text but not defined in Appendix C beyond the matrix entries.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: TT-cross integrals are computed from the stated potentials and checked against independent MD using the same potentials; no fitted parameter is renamed as a prediction.

full rationale

The paper's derivation chain is not circular. ZN is obtained by tensor-train cross interpolation of the Boltzmann factor e^{-beta U(q)} built from the stated interatomic potentials (SMATB, HIP-NN, MEAM), and the resulting internal energy, pressure, and phase boundary are compared with independent MD/thermodynamic-integration calculations using the same potentials. No parameter is fitted to the MD target quantities; the only tunables (integration domain, grid size, delta, Delta beta, Delta V) are numerical convergence parameters, and the paper states that Delta beta and Delta V are reduced until the computed quantities become numerically stable. The rank-1 and rank-2 constructions are explicit low-rank approximations (Eqs. 4-7), not exact identities, and the absence of a convergence study in rank, delta, grid spacing, and system size is a correctness/validation limitation rather than circularity. In the harmonic limit the rank-1 factorization reduces to a single-coordinate independent-oscillator ansatz, but this is a modeling approximation, not a circular reduction, because the paper's claims are comparative (TT versus MD) rather than derived from the target data. The self-citation [5] is introductory and not load-bearing; the core TT-cross methodology rests on standard Oseledets/Tyrtyshnikov references [9,12,15]. No load-bearing step reduces by construction to its own inputs, so the circularity score is 0.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The method rests on two key assumptions: the potential is sharply peaked at the lattice configuration, and the integrand is effectively low-rank. The first is reasonable for ordered solids at moderate temperatures; the second is equivalent to an independent-oscillator approximation and is not justified for interacting solids. Several numerical parameters (domain width, grid size, delta) are not specified, which weakens reproducibility.

free parameters (4)

Integration domain width [a_k, b_k] = Not specified
The integration interval for each coordinate is chosen symmetric around the lattice minimum qbar_k, but the width is not stated or justified; the results may depend on it.
Grid size n = Not specified
Number of grid points per dimension is not reported; no convergence study shown.
Rank-2 offset delta = Not specified
Distance between the two super-diagonal points in the rank-2 TT-cross; the paper does not state how delta is chosen or its influence on accuracy.
Derivative step sizes Delta beta and Delta V = Chosen empirically
The paper states these are reduced until results stabilize; this is a numerical convergence parameter, not fitted to target data.

assumptions (5)

domain assumption The Boltzmann factor e^{-beta U(q)} has a unique global maximum at the ideal lattice configuration qbar
Used to select the super-diagonal pivot and the fibers; stated in 'Discretization of the Boltzmann factor'.
domain assumption The Boltzmann factor is sharply peaked and vanishes outside the integration domain
Justifies the localized integration domain and the low-rank approximation; assumed in rank-1 and rank-2 derivations.
ad hoc to paper The integrand is well approximated by rank-1 or rank-2 TT decomposition
The accuracy of this approximation is not derived; it is the central uncontrolled assumption. Appears in Eq. (4) and the rank-2 construction.
ad hoc to paper For the rank-2 scheme, off-diagonal points have near-zero integrand, so |det(S)| approx fbar^2
Stated as 'for a very steep function f' in the Rank-2 TT-cross section; not quantitatively justified.
domain assumption Lattice periodicity makes all particles and x, y, z directions equivalent
Used to reduce 3N one-dimensional integrals to a single representative integral in Eq. (7). Standard for perfect crystals.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Breaking the Curse of Dimensionality: Solving Configurational Integrals for Crystalline Solids by Tensor Networks." pith.science (2026). https://pith.science/paper/4BQQOLGC

@misc{pith2026250521826,
  author       = {Pith},
  title        = {Pith review of: Breaking the Curse of Dimensionality: Solving Configurational Integrals for Crystalline Solids by Tensor Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4BQQOLGC}},
  note         = {Machine review of arXiv:2505.21826}
}

read the original abstract

Accurately evaluating configurational integrals for dense solids remains a central and difficult challenge in the statistical mechanics of condensed systems. Here, we present a novel tensor network approach that reformulates the high-dimensional configurational integral for identical-particle crystals into a sequence of computationally efficient summations. We represent the integrand as a high-dimensional tensor and apply tensor-train (TT) decomposition together with a custom TT-cross interpolation scheme. This approach avoids the need to explicitly construct the full tensor, which would otherwise be computationally intractable. We introduce tailored rank-1 and rank-2 schemes optimized for sharply peaked Boltzmann probability densities, typical in crystalline solids. When applied to the calculation of internal energy and pressure-temperature curves for crystalline copper (Cu) and argon (Ar), as well as the alpha-to-beta phase transition in tin (Sn), our method accurately reproduces molecular dynamics simulation results using tight-binding, machine learning (HIP-NN), and MEAM potentials, all within seconds of computation time.

Figures

Figures reproduced from arXiv: 2505.21826 by the authors.

**Figure 3.** FIG. 3. TT-rank 1 (squares) and TT-rank 2 (stars) [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 5.** FIG. 5. Phase diagram of Tin (Sn) using a MEAM [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

32 extracted references · 30 canonical work pages

[1]

J. C. Maxwell, Phil. Trans. Roy. Soc. (London)157, 49 (1867)

work page
[2]

≈ 0, and we have | det(S)| ≈¯f 2. Hence, the choice of this submatrix is close to the theoretical maximum volume, we need to calculate for the rank-2 TT-cross, and we can formulate the following result: For N identical particles, let the point ¯ qand point ¯ q+ δ are on the super-diagonal. LetSk, for k = 1, . . . ,3N − 1, denote the submatrix of the kth s...

work page
[3]

Boltzmann, Lectures on Gas Theory (Dover, New York, 2011)

L. Boltzmann, Lectures on Gas Theory (Dover, New York, 2011)

work page 2011
[4]

J. W. Gibbs, Elementary Pirnciples in Statistical Mechanics (Cambridge University Press, New York, 2010)

work page 2010
[5]

Bellman, Science153, 34 (1966)

R. Bellman, Science153, 34 (1966)

work page 1966
[6]

Alexandrov, G

B. Alexandrov, G. Manzini, E. W. Skau, P. M. D. Truong, and R. G. Vuchkov, Mathematics11, 534 (2023)

work page 2023
[7]

D. A. McQuarrie,Statistical Mechanics (University Science Books, 2000)

work page 2000
[8]

M. P. Allen and D. J. Tildesley,Computer Simulation of Liquids (Oxford University Press, 1987)

work page 1987

Show all 32 references

[9]

Cichocki, N

A. Cichocki, N. Lee, I. Oseledets, A.-H. Phan, Q. Zhao, D. P. Mandic,et al., Found. Trends Mach. Learn. 9, 249 (2016)

2016
[10]

Oseledets and E

I. Oseledets and E. Tyrtyshnikov, Linear Algebra Appl. 432, 70 (2010)

2010
[11]

Khoromskaia, B

V. Khoromskaia, B. N. Khoromskij, and R. Schnei- der, SIAM J. Sci. Comput.35, A987 (2013)

2013
[12]

Dolgov and D

S. Dolgov and D. Savostyanov, Comput. Phys. Com- mun. 246, 106869 (2020)

2020
[13]

I. V. Oseledets, SIAM J. Sci. Comput. 33, 2295 (2011)

2011
[14]

S. A. Goreinov, E. E. Tyrtyshnikov, and N. L. Za- marashkin, Linear Algebra Appl.261, 1 (1997)

1997
[15]

M. W. Mahoney and P. Drineas, Proc. Natl. Acad. Sci. U.S.A. 106, 697 (2009)

2009
[16]

D. V. Savostyanov, Linear Algebra Appl.458, 217 (2014)

2014
[17]

T. Cui, S. Dolgov, and O. Zahm, J. Comput. Phys. 485, 112103 (2023)

2023
[18]

Cleri and V

F. Cleri and V. Rosato, Phys. Rev. B48, 22 (1993)

1993
[19]

Lubbers, J

N. Lubbers, J. S. Smith, and K. Barros, J. Chem. Phys. 148 (2018)

2018
[20]

A. H. Larsen, J. J. Mortensen, J. Blomqvist, I. E. Castelli, R. Christensen, M. Dułak, J. Friis, M. N. Groves, B. Hammer, C. Hargus, et al. , J. Phys.: Condens. Matter 29, 273002 (2017)

2017
[21]

A. P. Thompson, H. M. Aktulga, R. Berger, D. S. Bolintineanu, W. M. Brown, P. S. Crozier, P. J. in ’t Veld, A. Kohlmeyer, S. G. Moore, T. D. Nguyen, R. Shan, M. J. Stevens, J. Tranchida, C. Trott, and S. J. Plimpton, Comput. Phys. Commun.271, 108171 (2022)

2022
[22]

B.-J. Lee, M. Baskes, H. Kim, and Y. K. Cho, Phys. Rev. B 64, 184102 (2001)

2001
[23]

Ko, D.-H

W.-S. Ko, D.-H. Kim, Y.-J. Kwon, and M. H. Lee, Metals 8, 10.3390/met8110900 (2018)

2018 doi
[24]

Menon, Y

S. Menon, Y. Lysogorskiy, J. Rogal, and R. Drautz, Phys. Rev. Mater.5, 103801 (2021)

2021
[25]

S. A. Goreinov and E. E. Tyrtyshnikov, Contemp. Math. 280, 47 (2001)

2001
[26]

https://github.com/oseledets/TT-Toolbox
[27]

Oseledets, Comput

I. Oseledets, Comput. Methods Appl. Math.11, 382 (2011)

2011
[28]

S. V. Dolgov and D. V. Savostyanov, SIAM J. Sci. Comput. 36, A2248 (2014)

2014
[29]

Rosato, M

V. Rosato, M. Guillope, and B. Legrand, Philos. Mag. A 59, 321 (1989)

1989
[30]

Chigaev, J

M. Chigaev, J. S. Smith, S. Anaya, B. Nebgen, M. Bettencourt, K. Barros, and N. Lubbers, The Journal of Chemical Physics158, 184108 (2023)

2023
[31]

Y. Shao, Z. Gan, E. Epifanovsky, A. T. Gilbert, M. Wormit, J. Kussmann, A. W. Lange, A. Behn, J. Deng, X. Feng, D. Ghosh, M. Goldey, P. R. Horn, L. D. Jacobson, I. Kaliman, R. Z. Khaliullin, T. Kuś, A. Landau, J. Liu, E. I. Proynov, Y. M. Rhee, R. M. Richard, M. A. Rohrdanz, R...

2015
[32]

temperature sweep

D. E. Woon and J. Dunning, Thom H., The Journal of Chemical Physics 98, 1358 (1993), https://pubs.aip.org/aip/jcp/article- pdf/98/2/1358/19124934/1358_1_online.pdf. 7 End Matter Appendix A: TheCUR approximation: The CUR is a matrix decomposition which approximate the matrix A ...

1993

Pith tools

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

[1] [1]

J. C. Maxwell, Phil. Trans. Roy. Soc. (London)157, 49 (1867)

work page

[2] [2]

≈ 0, and we have | det(S)| ≈¯f 2. Hence, the choice of this submatrix is close to the theoretical maximum volume, we need to calculate for the rank-2 TT-cross, and we can formulate the following result: For N identical particles, let the point ¯ qand point ¯ q+ δ are on the super-diagonal. LetSk, for k = 1, . . . ,3N − 1, denote the submatrix of the kth s...

work page

[3] [3]

Boltzmann, Lectures on Gas Theory (Dover, New York, 2011)

L. Boltzmann, Lectures on Gas Theory (Dover, New York, 2011)

work page 2011

[4] [4]

J. W. Gibbs, Elementary Pirnciples in Statistical Mechanics (Cambridge University Press, New York, 2010)

work page 2010

[5] [5]

Bellman, Science153, 34 (1966)

R. Bellman, Science153, 34 (1966)

work page 1966

[6] [6]

Alexandrov, G

B. Alexandrov, G. Manzini, E. W. Skau, P. M. D. Truong, and R. G. Vuchkov, Mathematics11, 534 (2023)

work page 2023

[7] [7]

D. A. McQuarrie,Statistical Mechanics (University Science Books, 2000)

work page 2000

[8] [8]

M. P. Allen and D. J. Tildesley,Computer Simulation of Liquids (Oxford University Press, 1987)

work page 1987