- Rating
- 3(+1)
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
The paper adds an auxiliary loss to Kohn–Sham Hamiltonian prediction. The predicted Hamiltonian is diagonalised during training, the occupied eigenvectors are Löwdin-orthonormalised, and the squared Grassmannian geodesic distance to the reference occupied subspace is penalised. A Taylor expansion of arccos$^2$ near , a small-gap truncation of the eigensolver backward pass and an auxiliary orbital-energy loss complete the method ("GrassDist"). With SPHNet and QHNet backbones, the paper reports large reductions in density-matrix and non-self-consistent total-energy errors on QH9, smaller gains on MD17, and much better size extrapolation to QMugs molecules with 10–17 heavy atoms, compared with element-wise Hamiltonian regression and WALoss.
Critical strengths
- Well-motivated target. Element-wise Hamiltonian error is a poor proxy for the quantities users care about, and the occupied subspace is the right gauge-invariant object. The outlier case study (App. C.7) makes this concrete: two molecules have a HOMO/LUMO occupation swap (
d_Gr ≈ pi/2) and energy errors above 1000 kcal/mol. - Large empirical effect on density-derived quantities. On all four QH9 splits and both backbones,
d_Gr, P, andE_totimprove substantially, medianE_toterror drops by 4–12×. The QMugs transfer result (Table 8) is significant: both baselines collapse beyond 13 heavy atoms while GrassDist's quality degrades smoothly.
Critical weaknesses
W1. The paper does not isolate what its title claims: the benefit of the Grassmannian geodesic formulation, as opposed to any occupied-subspace (density-matrix) supervision.
-
The geodesic loss coincides with simpler losses. For small principal angles:
- the geodesic loss is ;
- the chordal loss is . This is a plain Frobenius loss on the density matrix in the Löwdin basis, and needs no SVD, arccos or Taylor patch;
- the Procrustes-aligned coefficient loss is . This is exactly the leading term of Eq. 9.
They differ only at relative order . At the reported
d_Gr ≈ 5×10⁻³(2–7×10⁻² for the baseline), every angle lies deep inside the Taylor branch, and the three losses agree to better than 0.1%. They diverge materially only near , i.e. for rare occupation swaps. -
The gauge-redundancy ablation (Sec. 3.3) is weak.
- Coefficient losses already cut
E_totfrom 4.87 to 0.18–0.19 kcal/mol. Most of the gain therefore comes from any orbital-level supervision.
- Coefficient losses already cut
-
Closely related work is not discussed. This includes:
- direct one-electron density-matrix learning (e.g. Shao et al., 2023);
- self-consistency training (Zhang et al., ICML 2024), cited only for the truncation trick;
- DEQH (NeurIPS 2024);
- prior Grassmann treatments of occupied subspaces in quantum chemistry (Van Voorhis & Head-Gordon, 2002; Polack et al., JCTC 2021).
As it stands, the evidence supports "supervising the density matrix through the eigensolver helps a lot". It does not support "the Grassmannian geodesic helps". A density-matrix (chordal) loss and a Procrustes-aligned coefficient loss, trained to convergence under the main protocol, are the decisive missing baselines.
W2. The headline gains are inflated by the evaluation protocol and by an additional loss term.
-
Two molecules account for the "over 100-fold" improvement.
E_totis evaluated on the first 800 test configurations.- The two Table 11 outliers (2213.94 and 1321.25 kcal/mol) are in that set, as Fig. 1a shows. Together they contribute ≈4.42 of the baseline's 4.87 kcal/mol mean (≈91%). This suggests the 800-configuration subset is not representative.
-
An extra loss term confounds Table 1. All main results use , but appears only in App. B.2. Table 7 shows that does not work on its own without , so its role should be clearly stated.
-
Comparisons are not equally rigorous.
- The default weights appear to be chosen from test-split metrics (Tables 5, 6, 10).
- WALoss uses a weight tuned in its original paper for a different dataset and backbone.
- No training cost is reported for the per-step float64 eigendecomposition and SVD, although the paper claims "scalable".
Questions for authors
- Under the main protocol (residual parameterisation, 200k steps, test split, ≥3 seeds), how do (a) the chordal/density-matrix Frobenius loss in the Löwdin basis, (b) a Procrustes-aligned coefficient loss and (c) the geodesic loss without the Taylor branch compare with GrassDist on all Table 1 metrics and on QMugs? What fraction of principal angles ever takes the
arccosbranch during training? - What are the stable-iid baseline
E_tot, P andd_Grmeans with the two Table 11 molecules excluded? Please report medians, 5/95 percentiles and the fraction of occupation-swap failures (any ) for every metric and split, on a common, randomly drawn evaluation set. How are the "first 800" configurations distributed over molecules in the dynamic splits? - Please add the H + case (1:0:0.25) to Table 1, QMugs, MD17 and the QHNet ablation. Which weights were used on MD17?
- Why does the purely -MAE-trained model have higher MAE than the Hamiltonian-only baseline?
- Were and the coefficient weights selected on validation or test metrics? What exactly is the validation loss used for checkpoint selection for each method? Was the WALoss weight re-tuned on QH9?
- What is the wall-clock time and memory per training step relative to the baseline, and how does it scale with the number of basis functions?
- How is
E_totobtained — a single non-self-consistent energy evaluation at $\hat{P}? - Why are the MD17 and QHNet baselines so much better than the published numbers, and why is ethanol 3–4× worse than the larger molecules?
- The paper discusses QHFlow2, which reports accurate non-self-consistent energies on QH9 from Hamiltonian-only training. How does GrassDist compare, or combine, with such architecture-side improvements?
17 comments
Continuing the response to W1 & Q1
- Comparison with additional loss variants. Following your suggestion, we have further extended the ablation to include chordal loss, Procrustes-aligned coefficient loss, and geodesic loss without the Taylor branch, under the main training protocol (Appendix C.9).
| Subspace loss | H (10^-6 Eh) | d_Gr (10^-3) | P (10^-3) | E_tot (kcal/mol) | eps_occ (10^-6 Eh) | mu (10^-3 a.u.) | S_acc (%) |
|---|---|---|---|---|---|---|---|
| Chordal | 180.29 ± 1.33 | 4.84 ± 0.21 | 0.27 ± 0.02 | 0.04 ± 0.00 | 198.10 ± 9.36 | 6.29 ± 0.39 | 61.69 ± 0.11 |
| Procrustes | 180.84 ± 0.33 | 4.94 ± 0.21 | 0.28 ± 0.03 | 0.05 ± 0.00 | 208.68 ± 20.76 | 6.61 ± 0.53 | 61.60 ± 0.08 |
| Geodesic | 180.85 ± 1.32 | 5.08 ± 0.37 | 0.28 ± 0.02 | 0.05 ± 0.02 | 205.62 ± 12.14 | 6.58 ± 0.59 | 61.73 ± 0.02 |
| Geodesic-Taylor | 182.10 ± 0.16 | 4.72 ± 0.03 | 0.26 ± 0.00 | 0.04 ± 0.00 | 191.93 ± 3.56 | 6.02 ± 0.06 | 61.64 ± 0.12 |
The results show that these occupied-subspace objectives achieve comparable in-distribution performance when training converges, further supporting the effectiveness of subspace supervision. However, removing the Taylor branch introduces greater numerical variance.
- Frequency of the arccos branch. We additionally tracked approximately 8.44 million principal angles across two complete 200k-step training runs using different seeds. Only 0.0066% of angles enter the arccos branch overall, with the fraction decreasing from 1.21% during the first 1k steps to 0.00014% after 100k steps. This confirms that as training progresses, nearly all principal angles fall into the Taylor branch.
Response to W2-1 & Q2: Evaluation-set representativeness and outlier analysis.
We thank the reviewer for raising this concern. We provide additional evaluations to clarify the representativeness of our results and the influence of extreme outliers.
- Evaluation on the full test sets. Evaluating energy and SCF acceleration is computationally expensive, and prior studies typically report SCF acceleration on subsets of 50 configurations [2,3,7]. Our original evaluation of 800 configurations per dataset already provides a relatively large-scale benchmark. We also explicitly reported both mean and median errors to distinguish extreme failures from typical performance, and analyzed the two outliers in Appendix C.12 to illustrate the limitations of conventional Hamiltonian supervision. To directly address your concern, we have now extended the total-energy evaluation to the entire test set across all four QH9 splits, with the results summarized below.
Full-test MAE (kcal/mol):
| Dataset | Baseline E_tot mean (median) | GrassDist E_tot mean (median) | Fold (mean / median) |
|---|---|---|---|
| QH9-stable-iid | 2.8391 (0.2627) | 0.0385 (0.0331) | 73.7× (7.9×) |
| QH9-stable-ood | 0.6812 (0.2264) | 0.0392 (0.0367) | 17.4× (6.2×) |
| QH9-dynamic-geo | 0.8131 (0.2787) | 0.0256 (0.0248) | 31.8× (11.2×) |
| QH9-dynamic-mol | 68.2119 (1.2039) | 0.3805 (0.2296) | 179.3× (5.2×) |
GrassDist maintains substantial improvements across all four full test sets, reducing mean and median energy errors by 73.7-fold and 7.9-fold on stable-iid, and by 179.3-fold and 5.2-fold on dynamic-mol, respectively. Although the two extreme outliers affect the original 800-configuration mean, these full-test results confirm that our main conclusions extend well beyond the original evaluation subsets.
- Results excluding the two outliers. We further recompute the mean errors on the original 800-frame stable-iid evaluation subset (800−2), after excluding the two molecules identified in Table 11: | Metric | Baseline | GrassDist | Fold (×) | |---|---:|---:|---:| | (kcal/mol) | 0.454 | 0.040 | 11.4× | | P () | 1.747 | 0.263 | 6.7× | | () | 18.74 | 5.12 | 3.7× |
Even after removing these outliers, GrassDist reduces the errors by 11.4-fold, 6.7-fold, and 3.7-fold for , , and , respectively. This demonstrates that the improvements are not driven solely by the two extreme cases.
- Error distributions across test splits. We have additionally included five tables in Appendix C.11 reporting the medians and 5th/95th percentiles across all metrics and QH9 splits. For brevity, we present the stable-iid results here.
Median [P5, P95] on QH9 stable-iid (units as in Table 1):
| metric | Baseline | WALoss | GrassDist |
|---|---|---|---|
| H | 38.341 [23.855, 65.520] | 53.803 [34.461, 88.426] | 177.306 [113.908, 265.670] |
| dGr | 14.120 [8.352, 26.647] | 21.623 [13.020, 40.497] | 4.279 [2.739, 6.593] |
| P | 1.198 [0.514, 3.417] | 1.863 [0.774, 5.516] | 0.237 [0.129, 0.433] |
| Etot | 0.263 [0.092, 0.958] | 0.629 [0.223, 2.306] | 0.033 [0.015, 0.073] |
| 830.746 [85.386, 3035.945] | 1208.708 [124.133, 4039.046] | 528.079 [44.871, 2815.422] | |
| 133.681 [68.779, 308.603] | 183.600 [93.129, 418.205] | 163.901 [83.675, 364.641] | |
| 10.511 [3.290, 31.041] | 15.979 [5.047, 45.706] | 4.751 [1.498, 13.505] | |
| Sacc (%) | 66.667 [58.333, 75.000] | 69.231 [63.531, 75.000] | 61.538 [54.545, 66.667] |
GrassDist consistently achieves better median and percentile results for , , , , and , supporting improvements in typical-case performance beyond the extreme outliers. Similar trends are observed across the other QH9 splits (Appendix C.11).
4.Occupation-swap failure rates. Following your suggested criterion (any principal angle ), we further quantify occupation-swap failures across all QH9 splits:
| model | stable-iid | stable-ood | dynamic-geo | dynamic-mol |
|---|---|---|---|---|
| Baseline | 0.16% (21/13084) | 0.03% (3/9335) | 0.03% (3/9990) | 4.12% (416/10100) |
| WALoss | 0.11% (14/13084) | 0.12% (11/9335) | 0.00% (0/9990) | 2.34% (236/10100) |
| GrassDist | 0.00% (0/13084) | 0.00% (0/9335) | 0.00% (0/9990) | 0.00% (0/10100) |
Notably, GrassDist exhibits zero occupation-swap failures across all four splits, whereas Baseline and WALoss exhibit nonzero failure rates on several splits, particularly dynamic-mol. This further highlights the benefit of occupied-subspace supervision in preventing catastrophic prediction failures.
Response to W1 & Q1: Grassmannian geodesic formulation and occupied-subspace supervision.
We thank the reviewer for the comprehensive and constructive feedback. We address your concerns point by point below.
Geodesic-Taylor (GrassDist in main text) loss achieves better performance across d_Gr, P, and E_tot (mean and median). With orbital-energy supervision included, the two losses achieve more comparable in-distribution performance, while geodesic loss retains a clear advantage in QMugs size extrapolation. Specifically, chordal loss yields 3.8%, 13.2%, and 43.8% higher energy MAE for molecules with 13, 15, and 17 heavy atoms, respectively. These findings support the value of the geodesic formulation, particularly for generalization beyond the training distribution.
Gauge-redundancy ablation. We partially agree with the reviewer that occupied-subspace supervision is a major source of improvement. Indeed, demonstrating its importance is a central motivation and conceptual contribution of our work, as occupied-subspace supervision has not yet become a widely adopted principle in Hamiltonian prediction. Nevertheless, the results in Sec. 3.3 further demonstrate the benefits of our geometric formulation. Compared with coefficient-level supervision, GrassDist reduces total-energy MAE from 0.18 to 0.04 kcal/mol and dipole-moment error from 13.91 to 6.04, corresponding to reductions of 77.8% and 56.6%, respectively. Moreover, even with a small geodesic-loss weight and without orbital-energy supervision, GrassDist-lite improves total-energy MAE from 0.18 to 0.12 kcal/mol and dipole-moment error from 13.91 to 9.16. These results provide additional evidence for the benefit of our manifold-based supervision beyond naive coefficient-level objectives.
Discussion of related work. We thank the reviewer for highlighting these relevant studies. We have expanded Sec. 4 and added the earlier quantum-chemistry Grassmann studies to Sec. 2.3. Their distinctions from GrassDist are summarized below: • Shao et al. (2023) directly learn the one-electron density matrix as a surrogate for electronic-structure calculations, whereas GrassDist predicts the Hamiltonian and explicitly supervises its induced occupied subspace. • Zhang et al. (ICML 2024) exploit Hamiltonian self-consistency for training without reference Hamiltonian labels, whereas GrassDist directly aligns predicted and reference occupied subspaces through a Grassmann manifold loss. • DEQH (NeurIPS 2024) incorporates self-consistency through an implicit fixed-point architecture, which is complementary to our loss-level approach. • Van Voorhis & Head-Gordon (2002) and Polack et al. (2021) employ Grassmann geometry for orbital optimization and accelerated MD via density-matrix extrapolation in conventional electronic-structure calculations, respectively. In contrast, GrassDist introduces Grassmannian supervision into equivariant Hamiltonian learning.