Non-Negative Least Squares Reweighting and Pruning of Quadrature Grids for Tensor Hypercontraction

Andreas Erbs Hillers-Bendtsen; Lixin Lu; Todd J. Mart\'inez

arxiv: 2604.03899 · v1 · submitted 2026-04-04 · ⚛️ physics.chem-ph

Non-Negative Least Squares Reweighting and Pruning of Quadrature Grids for Tensor Hypercontraction

Andreas Erbs Hillers-Bendtsen , Lixin Lu , Todd J. Mart\'inez This is my paper

Pith reviewed 2026-05-13 16:42 UTC · model grok-4.3

classification ⚛️ physics.chem-ph

keywords tensor hypercontractionquadrature gridsnon-negative least squaresreweightingpruningelectronic structure theorytwo-electron integrals

0 comments

The pith

Non-negative least squares reweighting produces prunable quadrature grids for tensor hypercontraction

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

The authors introduce a reweighting scheme for quadrature grids used in tensor hypercontraction. Grid weights are optimized so that numerical integration accurately reproduces the atomic orbital overlap matrix. Formulating the optimization as a non-negative least-squares problem allows for a black-box approach that also prunes the grid by setting weights of insignificant points to zero. This simplifies the design of compact grids while maintaining accuracy for electronic structure calculations.

Core claim

Casting the fitting task of grid weights to reproduce the atomic orbital overlap matrix as a non-negative least-squares problem yields a methodology that produces robust grids for tensor hypercontraction and prunes them by zeroing quadrature weights for insignificant points.

What carries the argument

Non-negative least-squares optimization of quadrature weights to match the atomic orbital overlap matrix, enabling automatic pruning.

If this is right

Grids become robust without manual tuning for accuracy in tensor hypercontraction.
Pruning reduces the number of grid points, improving efficiency.
The method applies to numerical integration of other integrals besides two-electron repulsion.
Black-box nature facilitates widespread use of tensor hypercontraction in electronic structure methods.

Where Pith is reading between the lines

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

This reweighting could be combined with other grid optimization techniques for even better performance.
It may allow for adaptive grid refinement based on the fitting residuals.
The approach might extend to other integral approximations in quantum chemistry.

Load-bearing premise

Accurate reproduction of the atomic orbital overlap matrix is sufficient to guarantee accurate reproduction of the four-center two-electron repulsion integrals.

What would settle it

Demonstrating that the error in the two-electron integrals remains large even when the overlap matrix is reproduced accurately by the reweighted grid.

read the original abstract

Tensor hypercontraction provides an attractive four-center two-electron repulsion integral format that can lower the scaling of many electronic structure methods while only requiring O(N^2) memory. However, in its grid-based least-squares incarnation, tensor hypercontraction requires the tedious design of compact spatial quadrature grids to achieve efficiency and accuracy, representing a bottleneck for widespread application. To simplify grid generation, we devise a reweighting scheme in which the grid weights are optimized to ensure accurate reproduction of the atomic orbital overlap matrix by numerical integration. By casting this fitting task as a non-negative least-squares problem, we obtain a black-box methodology that not only yields robust grids for tensor hypercontraction as well as numerical integration of other integrals but also prunes the grids by zeroing quadrature weights for insignificant points.

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.

Desk Editor's Note private letter to a colleague

NNLS reweighting to match the AO overlap gives a clean black-box pruning method for THC grids, but the overlap-to-four-center-integral link is the part that needs real checking.

read the letter

The paper's main move is to recast quadrature weight selection for tensor hypercontraction as a non-negative least-squares fit that targets the atomic orbital overlap matrix. This produces weights that can be driven to zero for pruning while keeping the overlap error small. The formulation is straightforward and turns what used to be manual grid tuning into something more automatic. That is the useful piece: a single optimization step that both improves accuracy on the target and removes points without extra heuristics. It sits on top of existing NNLS solvers, so the implementation barrier looks low. The authors are right that this addresses a practical bottleneck in making THC routine for larger molecules. The approach is also general enough that it could apply to other grid-based integrals beyond THC. The soft spot is exactly the one the stress-test note flags. Matching the two-index overlap does not automatically protect the four-index Coulomb integrals that THC actually approximates. Points that matter for the 1/|r-r'| kernel can be zeroed out while the overlap error stays acceptable, and nothing in the construction enforces fidelity on the full (pq|rs) tensor after pruning. The abstract gives no error tables or direct comparisons on the repulsion integrals, so the claim that the grids are robust rests on an untested assumption at this stage. If the full manuscript shows tight errors on actual THC energies or integrals versus reference grids, that would close the gap; otherwise the central promise stays provisional. This is for people already working on THC or grid-based integral factorizations in quantum chemistry. A reader who needs to generate compact, accurate quadrature sets would get a concrete recipe worth trying. I would send it to peer review so the numerical validation and the overlap-to-integral transfer can be examined properly.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes a non-negative least-squares (NNLS) reweighting scheme for quadrature grids in tensor hypercontraction (THC). Grid weights are optimized to reproduce the atomic-orbital overlap matrix by numerical integration; the resulting non-negative solution also prunes the grid by driving insignificant weights to zero. The abstract presents this as a black-box procedure that yields robust grids for THC and for other numerical integrations.

Significance. If the numerical fidelity of the pruned grids for four-center integrals is confirmed, the method would remove a practical bottleneck in grid-based THC implementations by automating the generation of compact, accurate quadrature sets. This could broaden the applicability of THC-based electronic-structure methods.

major comments (2)

[Abstract] Abstract: the central claim that the NNLS-reweighted grids are 'robust' for tensor hypercontraction rests on the unshown assumption that accurate reproduction of the two-index overlap matrix S_ij guarantees accuracy for the four-center integrals (pq|rs) after pruning. No error metrics, THC energy comparisons, or validation data are referenced.
[Method] Method section (grid-reweighting formulation): the optimization is posed solely against the AO overlap matrix; the manuscript must demonstrate, either by direct numerical test or by analysis of the 1/|r-r'| kernel, that points critical to the Coulomb integrals are not erroneously zeroed while the overlap error remains small.

minor comments (1)

[Abstract] Abstract: a single sentence quantifying typical grid-size reduction or observed THC error would help readers assess the practical impact.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment of the work's potential significance and for the constructive major comments. We address each point below and will revise the manuscript to incorporate the requested clarifications and additional validations.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim that the NNLS-reweighted grids are 'robust' for tensor hypercontraction rests on the unshown assumption that accurate reproduction of the two-index overlap matrix S_ij guarantees accuracy for the four-center integrals (pq|rs) after pruning. No error metrics, THC energy comparisons, or validation data are referenced.

Authors: We agree that the abstract's claim of robustness for THC would be strengthened by explicit linkage to four-center integral accuracy. The manuscript demonstrates accurate numerical reproduction of the AO overlap matrix and the resulting grid pruning, but does not currently reference specific THC energy errors or four-center integral metrics. In the revised version we will update the abstract to reference the results section and add direct comparisons of THC energies and (pq|rs) integral errors obtained with the NNLS-reweighted grids against reference values. revision: yes
Referee: [Method] Method section (grid-reweighting formulation): the optimization is posed solely against the AO overlap matrix; the manuscript must demonstrate, either by direct numerical test or by analysis of the 1/|r-r'| kernel, that points critical to the Coulomb integrals are not erroneously zeroed while the overlap error remains small.

Authors: We concur that an explicit check is needed to confirm that pruning driven by the overlap matrix does not inadvertently remove points important for the Coulomb kernel. While the non-negativity constraint in NNLS already tends to retain points that contribute to the fitted quantities, the current text does not provide the requested numerical test or kernel analysis. We will add either a direct numerical comparison of Coulomb integrals on the pruned versus original grids or a supporting analysis of the 1/|r-r'| kernel in the revised method/results sections. revision: yes

Circularity Check

0 steps flagged

No significant circularity; optimization targets independent overlap matrix

full rationale

The core derivation casts grid-weight optimization as a non-negative least-squares problem whose target is the atomic orbital overlap matrix S_ij. This matrix is computed directly from the basis functions and is independent of the quadrature weights being fitted. No step reduces the claimed THC grid quality to the fitted weights by construction, nor does any load-bearing premise rely on a self-citation chain or imported uniqueness theorem. The paper presents the THC application as an empirical outcome of the reweighting procedure rather than a mathematical identity. This is the normal, non-circular case of fitting to an external, verifiable benchmark.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach relies on standard non-negative least-squares optimization and the assumption that overlap-matrix fidelity implies integral accuracy; no new free parameters, ad-hoc axioms, or invented entities are introduced.

axioms (1)

standard math Non-negative least-squares optimization yields a unique or stable solution for grid weights that reproduces the AO overlap matrix
Standard convex optimization result invoked to justify the black-box procedure.

pith-pipeline@v0.9.0 · 5448 in / 1223 out tokens · 65349 ms · 2026-05-13T16:42:05.799162+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

By casting this fitting task as a non-negative least-squares problem, we obtain a black-box methodology that not only yields robust grids for tensor hypercontraction as well as numerical integration of other integrals but also prunes the grids by zeroing quadrature weights for insignificant points.
IndisputableMonolith/Foundation/ArithmeticFromLogic.lean LogicNat.equivNat unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

the optimal w is found from a linear system of equations A w = S

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.

Reference graph

Works this paper leans on

4 extracted references · 4 canonical work pages

[1]

Schwegler, E.; Challacombe, M.; Head-Gordon, M., Linear scaling computation of the Fock matrix. II. Rigorous bounds on exchange integrals and incremental Fock build. J. Chem. Phys. 1997, 106, 9708. 10. Ufimtsev, I. S.; Martinez, T. J., Quantum chemistry on graphical processing units. 1. Strategies for two-electron integral evaluation. J. Chem. Theo. Compu...

work page 1997
[2]

B., Quartic scaling evaluation of canonical scaled opposite spin second-order Møller–Plesset correlation energy using Cholesky decompositions

Aquilante, F.; Pedersen, T. B., Quartic scaling evaluation of canonical scaled opposite spin second-order Møller–Plesset correlation energy using Cholesky decompositions. Chem. Phys. Lett. 2007, 449, 354. 22. Pedersen, T. B.; Sánchez de Merás, A. M.; Koch, H., Polarizability and optical rotation calculated from the approximate coupled cluster singles and ...

work page 2007
[3]

M.; Hohenstein, E

Parrish, R. M.; Hohenstein, E. G.; Schunck, N. F.; Sherrill, C. D.; Martínez, T. J., Exact Tensor Hypercontraction: A Universal Technique for the Resolution of Matrix Elements of Local Finite-Range N-Body Potentials in Many-Body Quantum Problems. Physical Review Letters 2013, 111, 132505. 35. Song, C.; Martínez, T. J., Atomic orbital-based SOS-MP2 with te...

work page arXiv 2013
[4]

M.; Hohenstein, E

Parrish, R. M.; Hohenstein, E. G.; Martínez, T. J.; Sherrill, C. D., Discrete variable representation in electronic structure theory: Quadrature grids for least-squares tensor hypercontraction. J. Chem. Phys. 2013, 138. 47. Kokkila Schumacher, S. I.; Hohenstein, E. G.; Parrish, R. M.; Wang, L.-P.; Martínez, T. J., Tensor hypercontraction second-order Møll...

work page 2013

[1] [1]

Schwegler, E.; Challacombe, M.; Head-Gordon, M., Linear scaling computation of the Fock matrix. II. Rigorous bounds on exchange integrals and incremental Fock build. J. Chem. Phys. 1997, 106, 9708. 10. Ufimtsev, I. S.; Martinez, T. J., Quantum chemistry on graphical processing units. 1. Strategies for two-electron integral evaluation. J. Chem. Theo. Compu...

work page 1997

[2] [2]

B., Quartic scaling evaluation of canonical scaled opposite spin second-order Møller–Plesset correlation energy using Cholesky decompositions

Aquilante, F.; Pedersen, T. B., Quartic scaling evaluation of canonical scaled opposite spin second-order Møller–Plesset correlation energy using Cholesky decompositions. Chem. Phys. Lett. 2007, 449, 354. 22. Pedersen, T. B.; Sánchez de Merás, A. M.; Koch, H., Polarizability and optical rotation calculated from the approximate coupled cluster singles and ...

work page 2007

[3] [3]

M.; Hohenstein, E

Parrish, R. M.; Hohenstein, E. G.; Schunck, N. F.; Sherrill, C. D.; Martínez, T. J., Exact Tensor Hypercontraction: A Universal Technique for the Resolution of Matrix Elements of Local Finite-Range N-Body Potentials in Many-Body Quantum Problems. Physical Review Letters 2013, 111, 132505. 35. Song, C.; Martínez, T. J., Atomic orbital-based SOS-MP2 with te...

work page arXiv 2013

[4] [4]

M.; Hohenstein, E

Parrish, R. M.; Hohenstein, E. G.; Martínez, T. J.; Sherrill, C. D., Discrete variable representation in electronic structure theory: Quadrature grids for least-squares tensor hypercontraction. J. Chem. Phys. 2013, 138. 47. Kokkila Schumacher, S. I.; Hohenstein, E. G.; Parrish, R. M.; Wang, L.-P.; Martínez, T. J., Tensor hypercontraction second-order Møll...

work page 2013