- Rating
- 3
- Confidence
- 2
- Informativeness
- 3
- Presentation
- 3
- Soundness
- 3
Read review
Paper summary
This paper introduces Noether World Models, a latent dynamics architecture designed to preserve a family of quadratic invariants. The method combines graph-Laplacian generators with Cayley-transform updates to enforce conservation by construction. The paper also investigates the limitations of imposing hard conservation constraints on existing latent representations and explores relaxed alternatives for prediction and planning.
Critical strengths
- A principled structural contribution. The conservation properties follow from the architecture rather than relying solely on a training penalty. The theoretical analysis explains which quadratic quantities are preserved, and the controlled comparisons help distinguish the contribution of the proposed structure from that of the Cayley transform alone.
- An informative investigation of practical limitations. The paper reports cases where hard conservation constraints perform poorly in planning, rather than presenting conservation as universally beneficial. The capacity controls and relaxation experiments provide useful evidence about the trade-off between structural constraints and predictive flexibility.
Critical weaknesses
- The numerical evidence for conservation precision requires clarification. The supplementary results report substantially different relative drift magnitudes for selected invariants and randomly sampled members of the invariant family. These differences may reflect normalization, small initial quadratic values, or numerical implementation details rather than a failure of the theory. Nevertheless, the claim of double-precision conservation would be more convincing with a consistent account of numerical precision and normalization, accompanied by absolute or appropriately scaled residuals.
- The practical contribution of preserving the invariant family is not sufficiently isolated. Exact conservation establishes a structural property, but does not by itself ensure accurate dynamics or effective planning. A model can preserve quadratic quantities while predicting incorrect transition directions or action responses. The paper includes useful controls and acknowledges failures of hard constraints; however, it remains unclear which downstream improvements can be specifically attributed to invariant preservation. The discussion would benefit from a clearer connection between the preserved quantities and measurable gains in prediction or planning, including the conditions under which conservation helps or hurts.
Questions for authors
- Can the authors reconcile the conservation precision reported for selected invariants with that reported for randomly sampled quadratic forms, including the effects of normalization and numerical precision?
- What concrete benefits does preserving the proposed invariant family provide beyond reducing invariant drift? Could the authors clarify whether, and under what conditions, it improves prediction accuracy, long-horizon rollout quality, or planning performance, and which controlled comparisons support attributing these improvements to invariant preservation?
- Could the authors clarify the guarantees associated with the relaxed variants? In particular, a finite conservation penalty generally encourages approximate conservation but does not guarantee exact conservation at the optimum of the combined objective.
4 comments
Response to Reviewer SE5x (2/3): W2/Q2 on what the family buys, and under which conditions
W2 and Q2. Benefits of preserving the family, and the conditions under which they arise. Beyond lower drift, preserving the family makes prediction robust to a change of force law and slows degradation over long horizons. It leaves in-distribution accuracy unchanged, and it is detrimental where a fixed encoder changes the norm of its latent. Matched controls attribute the gain to the family.
(i) Controlled comparison. Ctrl-cayley shares the coupling network and the Cayley step with NWM and differs only in placing the couplings on a random Hermitian basis instead of a Laplacian, so it holds the norm but not the family (Appendix C.4.2). Ctrl-free is an unconstrained transition with the same parameter count to within 1%. Each sector variant breaks one part of the family, as verified on the trained weights. NWM, its variants and the two controls share the encoder, decoder and training recipe. PSNR in dB at , mean over three seeds (Tables 12 and 17, Appendix D.13):
| Model | Conserved | In distribution | Coulomb |
|---|---|---|---|
| NWM | full family | 18.86 | 20.41 |
| NWM without zero row sums | 18.87 | 20.37 | |
| NWM without block-scalar | and | 18.72 | 20.33 |
| HNN (128K parameters) | learned energy | 18.85 | 20.41 |
| Ctrl-free (same size) | none | 18.67 | 18.58 |
| Ctrl-cayley (same size) | norm only | 17.65 | 17.50 |
Ctrl-cayley is precisely the case the reviewer describes: it preserves a quadratic quantity yet exchanges two objects (Figure 4). NWM matches HNN without assuming Hamiltonian dynamics, with 32 times fewer parameters. Preserving either sector retains the gain, while preserving the norm alone does not.
(ii) Breaking the family by a controlled amount. NWM interpolated with the MLP transition through Eq. 9, three seeds each, one shared recipe, so that retrains NWM itself (PSNR in dB):
| 0 | 0.01 | 0.03 | 0.1 | 0.25 | 0.5 | 1 | |
|---|---|---|---|---|---|---|---|
| In distribution | 18.86 | 18.95 | 18.88 | 18.10 | 17.86 | 18.96 | 18.96 |
| Coulomb | 20.39 | 17.62 | 17.57 | 17.65 | 17.97 | 18.34 | 18.44 |
| Energy rescaling | 21.30 | 21.91 | 21.99 | 20.21 | 20.36 | 21.26 | 21.24 |
Every breaks the family and costs 1.95 to 2.82 dB under the Coulomb law, while and match NWM in distribution, so the loss is not due to a weaker fit. Under breaking the family has no consistent effect. The gain is therefore attributable to the family and arises under the change of force law.
(iii) Conditions.
| Condition | Effect of preserving the family | Evidence |
|---|---|---|
| In distribution, | neutral: the nine designs lie within 1.31 dB, NWM 0.10 dB behind the best | Table 12 |
| In distribution, to 100 | slower degradation: 0.68 dB, against 1.57 and 1.53 dB for Ctrl-free and Ctrl-cayley | Table 16 |
| Coulomb law, exponents 1 to 3 | 1.8 to 3.0 dB over MLP, GRU, Ctrl-free, Ctrl-cayley and UWM, 25 paired tests, each | Appendix D.13 |
| Coulomb law, 10% of the data | 20.20 dB, above every unconstrained or norm-conserving transition trained on all data | Appendix D.13 |
| Fixed encoder whose norm varies | exact operator detrimental (planning at 3.17%), as bounded by Proposition 3 | §4.4 |
| Same, members imposed as a penalty | planning +8.50, +6.50, +7.00 points on Tworoom, where the host drifts most () | Table 2 |
The neutral in-distribution result agrees with the finding that physics priors on pixels do not significantly improve standard models (Botev et al., NeurIPS Datasets and Benchmarks 2021), and the gain under the change of force law agrees with the finding that symmetry built into a dynamics model improves generalization under distributional shift (Wang et al., ICLR 2021). The energy decoded from the NWM latent by a fixed probe also remains the most accurate of the nine models at every horizon up to 100 steps (see our response to Reviewer 7e1N, Q2).
We have added a paragraph to §4.3 that states these conditions and the controls that support each, with the new results in Appendix D.13. All changes in the revised PDF are marked in blue, and line numbers refer to the submitted version.
Response to Reviewer SE5x (3/3): Q3 on the guarantees of the relaxed variants
Q3. Guarantees of the relaxed variants. We agree that a finite penalty encourages approximate conservation and does not guarantee exact conservation at the optimum of the combined objective. The two relaxations carry different guarantees, and the revision states each precisely. With , a family member and a linear member such as the block sum:
| Variant | Used where | Guarantee at every parameter value | Violation after one step |
|---|---|---|---|
| NWM (Theorem 1, Corollary 2) | encoder trained with the operator | exact, for the whole family | zero up to round-off |
| Interpolated operator, Eq. 9 (Proposition 5) | NWM mixed with by weight | explicit bound, controlled by | and |
| Penalty, Eq. 10 | fixed host whose predictor is not replaced | none for a finite | approximate at a minimiser, exact only as |
(i) For the interpolated operator the violation is therefore controlled by , and Theorem 1 is recovered at .
(ii) For the penalty, as for any quadratic penalty, the constraint is met exactly only in the limit of a large weight (Nocedal and Wright, Numerical Optimization, 2006, Chapter 17). The penalty serves a different purpose, namely to impose the invariants on a host whose predictor must remain unchanged (§3.5).
We have replaced "holds at a minimiser of L" in §3.5 by "holds approximately at a minimiser of L, with no exact guarantee for any finite ".
We hope these responses resolve the reviewer's concerns, and we would be grateful if the reviewer reconsidered the score.
Response to Reviewer SE5x: each point of the review and its revision
As the discussion closes soon, we thank you for your positive assessment. We have answered each of your points, and each is listed below with its revision in the PDF, marked in blue.
W1 and Q1. Table 1, Tables 11 and 15 and the corresponding panels of Figures 3 and 5 now report the family ratios in double precision, and Appendix D.11 states the precision of every table. Table 23 adds the scaled residual. In double precision every member of the family, named or sampled, stays at round-off, and the family ratio exceeds , against at most 5.0 for every other transition. The statements on precision in §4.1 and Appendices D.1 and D.2 are corrected accordingly.
W2 and Q2. §4.3 now states what the family contributes and under which conditions, with the new results in Appendix D.13. Matched controls, seven interpolation weights and two sector variants attribute the gain to the family (Tables 19 and 20). Under a change of force law NWM outperforms the matched baselines in all 25 paired comparisons on a newly generated test set (), also when trained on a tenth of the data (Table 21).
Q3. §3.5 now states that the penalty holds only approximately at a minimiser, with no exact guarantee for a finite , and Proposition 5 gives the explicit bound of the interpolated operator for quadratic and linear members of the family.
We would be grateful to hear whether these revisions resolve your concerns. If they do, we would sincerely appreciate your consideration of a higher score. Thank you for your time and your support of the paper.
Response to Reviewer SE5x (1/3): Overview, and W1/Q1 on conservation precision and scaled residuals
We thank the reviewer for recognising that the conservation properties follow from the architecture and that the controls isolate the Laplacian structure. Part 1 shows that the gap between the named invariants and the sampled forms comes from numerical precision and normalisation, since in double precision every member of the family stays at round-off, and reports scaled residuals (W1, Q1). Part 2 shows with matched controls and new ablations that preserving the family makes prediction robust to a change of force law, and states when it helps and when it does not (W2, Q2). Part 3 states the guarantee of each relaxed variant (Q3).
W1 and Q1. Precision of the named invariants and of the sampled forms. The discrepancy stems from numerical precision and normalisation, not from the theory. Table 10 was computed with the whole rollout in double precision. Table 11, the ratios of Tables 1 and 13, and Table 14 were computed from the single-precision rollout of the trained model. Recomputed in double precision, every member of the family, named or sampled, drifts at the level of double-precision round-off. Line 338 and Appendices D.1, D.2 and D.11 state that all of these tables were computed in double precision from the same rollouts, which holds for Table 10 only, and we have corrected these four statements.
(i) Where precision enters. We recomputed all three datasets and seeds in three settings: (A) the published single-precision rollout, where H is assembled and the state is rounded in single precision at every step, (B) a double-precision rollout with H still assembled in single precision, and (C) a fully double-precision rollout, which is the setting of Table 10. The Cayley solve is in double precision in all three, and on the trained checkpoints ∥U†U−I∥F≤2.6×10−14. Single-precision assembly keeps H exactly Hermitian and block-scalar but leaves its row sums nonzero at about 4×10−8 relative, which affects only the block sum and the members with a 11⊤ component. Setting B isolates this term and setting C removes it.
(ii) Scaled residual. The relative drift divides by ∣Q(z0)∣, which for a sampled member can be close to zero: over the in-family forms, ∣Q(z0)∣/(∥B∥2∥z0∥2) falls to 8×10−7, so a mean of relative drifts is set by a few small denominators. We therefore report rB=∣QB(zh)−QB(z0)∣/(∥B∥2∥z0∥2), which does not depend on the scale of the form or on QB(z0). Median at h=29 over trajectories and forms, range over the three datasets and three seeds:
In setting C the largest in-family residual over all trajectories and forms is 1.6×10−15. In the same measure the in-family residuals of the eight alternative transitions lie between 3.7×10−3 and 7.5.
(iii) Growth with the horizon. In setting C the median residual of the norm, the block sum and the in-family forms grows from h=1 to h=100 with a log-log slope between 0.39 and 0.62, close to the square-root growth of round-off in exactly conservative schemes (Hairer et al., BIT Numerical Mathematics 2008), and stays below 6.6×10−15 at h=100. Extending the latent rollout to h=104 on all three datasets, the median in-family residual remains below 3.1×10−14 and its maximum over trajectories below 1.4×10−13 (at most 1.2×10−7 in the published single-precision setting). For the other transitions it is at least 1.2×10−2 and reaches 3.6×108, and GNN diverges before h=103.
(iv) Family ratio. The ratios of Tables 1 and 13 in the submission (4.3e6, 3.6e6, 6.0e6) are single-precision values. Our rerun in setting A gives 5.4e6, 3.1e6 and 9.7e6, with the difference coming from single-precision round-off and one Balls trajectory. In setting C they become 6.3e14, 5.1e14 and 6.3e14, while the largest ratio among the other eight transitions, evaluated in double precision as well, is 5.0. The separation between the family and the other forms is therefore eight orders of magnitude wider in double precision, and the conclusions of §4.1 hold with a wider margin.
The revised PDF reports the family ratios of Table 1, Tables 11 and 15 and the corresponding panels of Figures 3 and 5 in double precision, marks Tables 13 and 14 as single precision, adds the scaled residual (Table 23), and states the precision of every table in Appendix D.11.