pith. sign in

arxiv: 2606.13568 · v1 · pith:TQXMQONRnew · submitted 2026-06-11 · 💻 cs.LG · math-ph· math.MP

Adjusted Cup-Product Neural Layer

Snigdha Chandan Khilar This is my paper

Pith reviewed 2026-06-27 07:24 UTC · model grok-4.3

classification 💻 cs.LG math-phmath.MP
keywords cup productneural networksgauge invariancehigher gauge theorycochainsquadratic formsgauge transformationsmachine learning
0
0 comments X

The pith

An adjusted cup-product neural layer makes its output on closed cycles depend only on the adjustment coefficient from higher gauge theory.

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

The paper presents a neural primitive that embeds the cup product of cochains together with an adjustment term from higher gauge theory. This construction produces a readout that is gauge invariant by design. On a closed cycle the output value is determined entirely by the adjustment coefficient, and setting the coefficient to zero eliminates the output irrespective of all other parameters. The observable turns out to be a nonzero quadratic form that stays exactly the same under one and two gauge transformations. Readers interested in embedding physical symmetries directly into neural architectures would find this relevant because the invariance does not have to be learned from examples.

Core claim

The adjusted cup product neural layer hard wires the cup product with an adjustment term from higher gauge theory. This creates a readout that is gauge invariant by design. On a closed cycle the output relies entirely on the adjustment coefficient. Setting this coefficient to zero removes the output completely regardless of other parameters. Thus the adjustment is the only source of gauge invariant signal. The observable is a nonzero quadratic form and is exactly invariant under one and two gauge transformations.

What carries the argument

The adjusted cup product, a combination of the standard cup product on cochains with an adjustment term that enforces gauge invariance.

If this is right

  • The output is a nonzero quadratic form.
  • The output is exactly invariant under one gauge transformation.
  • The output is exactly invariant under two gauge transformations.
  • The adjustment coefficient is the sole source of the gauge invariant signal.

Where Pith is reading between the lines

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

  • If the adjustment term preserves the algebraic structure of the cup product, similar layers could be defined for other cohomology operations.
  • Networks using this layer may exhibit improved generalization on data with underlying gauge symmetry because invariance is built in rather than approximated.
  • The quadratic nature of the observable suggests it could be used to define new loss functions that penalize deviations from gauge invariance.

Load-bearing premise

The adjustment term from higher gauge theory can be hard-wired into a neural layer while exactly preserving the algebraic properties of the cup product on cochains.

What would settle it

Evaluate the neural layer on any closed cycle with the adjustment coefficient set to zero; if the output is not identically zero for nonzero inputs, the claim fails.

Figures

Figures reproduced from arXiv: 2606.13568 by Snigdha Chandan Khilar.

Figure 1
Figure 1. Figure 1: Test R 2 vs. training-set size on 3D Chern–Simons. The adjusted layer generalizes from M=64; competent baselines memorize the training set (train R 2→1, not shown) yet stay at or below zero on test. The κ=0 ablation confirms the adjustment is the source of signal. 5.2 An external benchmark: 2D Chern numbers (Haldane bundles) We reproduce the published Haldane-bundle benchmark [18], predicting the integer C… view at source ↗
Figure 2
Figure 2. Figure 2: Left: 2D Chern accuracy vs. grid resolution; the cup rides a discretization ceiling while the fair CNN trails. Right: under a random local gauge scramble the CNN collapses to chance while the gauge-invariant cup is unaffected—the cleanest isolation of the mechanism [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Helicity under corruption (5 seeds, error bars = std). A learned front-end (cup-net or CNN) is far more noise-robust than the raw analytical estimator. The cup-net additionally retains the estimator’s out-of-distribution￾physics generalization, slightly ahead of the CNN; under extreme sensor sparsity the unconstrained CNN wins. The cup-net combines noise- and OOD-robustness, but the margin over a tuned CNN… view at source ↗
read the original abstract

Many important observables in physics and geometry are cup products of cochains. The adjusted cup product neural layer has been introduced in this paper. It is a neural primitive that hard wires the cup product with an adjustment term from higher gauge theory. This creates a readout that is gauge invariant by design. Their main theoretical result shows that on a closed cycle the output relies entirely on the adjustment coefficient. Setting this coefficient to zero removes the output completely regardless of other parameters. Thus the adjustment is the only source of gauge invariant signal. They prove this observable is a nonzero quadratic form and is exactly invariant under one and two gauge transformations.

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.

Referee Report

1 major / 0 minor

Summary. The manuscript introduces the Adjusted Cup-Product Neural Layer, a neural primitive that hard-wires the cup product of cochains together with an adjustment term drawn from higher gauge theory. The central theoretical claim is that, on a closed cycle, the layer output depends only on the adjustment coefficient (vanishing identically when that coefficient is set to zero), that the resulting observable is a nonzero quadratic form, and that it is exactly invariant under one- and two-form gauge transformations.

Significance. If the stated invariance and dependence properties can be rigorously established from the layer definition, the construction would supply a neural primitive that enforces gauge invariance by design rather than through training. This could be of interest for physics-informed or geometric deep-learning architectures that must respect local symmetries. No machine-checked proofs, reproducible code, or parameter-free derivations are referenced in the manuscript.

major comments (1)
  1. [Abstract] Abstract: the claim that the output 'relies entirely on the adjustment coefficient' and 'vanishes when it is set to zero' is asserted without any equations, derivation steps, or explicit verification that the result follows from the layer definition rather than by construction; the math cannot be checked from the supplied text.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the output 'relies entirely on the adjustment coefficient' and 'vanishes when it is set to zero' is asserted without any equations, derivation steps, or explicit verification that the result follows from the layer definition rather than by construction; the math cannot be checked from the supplied text.

    Authors: We agree that the abstract asserts the central invariance property without exhibiting the supporting equations. In the revised version we will expand the abstract by one or two sentences that sketch the key algebraic steps: the layer is defined as the cup product plus the adjustment term drawn from the higher-gauge 3-cochain; on a closed 2-cycle the cup-product contribution cancels identically by the coboundary property, leaving only the adjustment coefficient multiplied by the pairing with the fundamental class. This cancellation is shown explicitly in Theorem 3.2 of the body; the abstract will now point to that theorem and quote the resulting simplified expression. revision: yes

Circularity Check

1 steps flagged

Main invariance result follows by construction from layer definition

specific steps
  1. self definitional [Abstract / main theoretical result]
    "Their main theoretical result shows that on a closed cycle the output relies entirely on the adjustment coefficient. Setting this coefficient to zero removes the output completely regardless of other parameters. Thus the adjustment is the only source of gauge invariant signal. They prove this observable is a nonzero quadratic form and is exactly invariant under one and two gauge transformations."

    The layer is introduced as a neural primitive that hard-wires the cup product together with the adjustment term. The claimed dependence of the output solely on that term (and its vanishing when the term is removed) is therefore true by the explicit construction of the layer; the 'proof' merely restates the definition rather than deriving a non-tautological property.

full rationale

The paper defines the adjusted cup-product layer by hard-wiring an adjustment term taken from higher gauge theory. Its central theorem then states that on a closed cycle the output depends entirely on this coefficient and vanishes when the coefficient is set to zero. This dependence is a direct consequence of the definition rather than an independent derivation; the proof reduces to substituting the defining expression. The nonzero quadratic form and gauge invariance claims are likewise properties built into the construction. No external benchmark or non-self-referential step is exhibited.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 1 invented entities

The central claim depends on the mathematical existence of cup products on cochains, the availability of an adjustment term from higher gauge theory, and the assumption that these can be combined into a neural primitive whose output is governed only by the adjustment coefficient.

free parameters (1)
  • adjustment coefficient
    The scalar that the abstract states completely determines the output on closed cycles; appears to be a free parameter of the layer.
axioms (2)
  • domain assumption Cup products of cochains form important observables in physics and geometry.
    Opening sentence of the abstract.
  • domain assumption Higher gauge theory supplies an adjustment term that can be hard-wired into a neural layer while preserving cup-product properties.
    Required for the layer to be defined as described.
invented entities (1)
  • Adjusted cup-product neural layer no independent evidence
    purpose: Neural primitive that produces gauge-invariant readouts by construction.
    New construct introduced in the paper.

pith-pipeline@v0.9.1-grok · 5620 in / 1420 out tokens · 26233 ms · 2026-06-27T07:24:56.800507+00:00 · methodology

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

22 extracted references · 4 linked inside Pith

  1. [1]

    M´ emoli, A

    F. M´ emoli, A. Stefanou, L. Zhou. Persistent Cup Product Structures and Related Invariants. arXiv:2211.16642, 2022

    arXiv 2022

  2. [2]

    Conti, F

    M. Conti, F. M´ emoli, A. Stefanou. Persistent cup-length.SoCG, 2022. arXiv:2107.01553

    arXiv 2022

  3. [3]

    T. K. Dey, A. Rathod. Cup Product Persistence and Its Efficient Computation.SoCG, 2024. arXiv:2212.01633

    arXiv 2024

  4. [4]

    Cohen, M

    T. Cohen, M. Weiler, B. Kicanaoglu, M. Welling. Gauge Equivariant Convolutional Networks and the Icosahedral CNN.ICML, 2019. arXiv:1902.04615

    Pith/arXiv arXiv 2019

  5. [5]

    Cohen, M

    T. Cohen, M. Geiger, M. Weiler. A General Theory of Equivariant CNNs on Homogeneous Spaces.NeurIPS, 2019

    2019

  6. [6]

    Weiler, P

    M. Weiler, P. Forr´ e, E. Verlinde, M. Welling. Coordinate Independent Convolutional Networks. 2021

    2021

  7. [7]

    arXiv:2305.16657, 2023

    Higher Order Gauge Equivariant CNNs on Riemannian Manifolds. arXiv:2305.16657, 2023

    arXiv 2023

  8. [8]

    Huang, O

    L. Huang, O. Balabanov, H. Linander, M. Granath, D. Persson, J. E. Gerken. Learning Chern Numbers of Topological Insulators with Gauge Equivariant Neural Networks (GEBLNet).NeurIPS, 2025. arXiv:2502.15376

    Pith/arXiv arXiv 2025

  9. [9]

    Favoni, A

    M. Favoni, A. Ipp, D. I. M¨ uller, D. Schuh. Lattice Gauge Equivariant Convolutional Neural Networks.Phys. Rev. Lett.128, 032003, 2022

    2022

  10. [10]

    arXiv:2604.20797, 2025

    Gauge-Equivariant Graph Neural Networks for Lattice Gauge Theories. arXiv:2604.20797, 2025

    Pith/arXiv arXiv 2025

  11. [11]

    S¨ amann, H

    C. S¨ amann, H. Kim. Adjusted Parallel Transport for Higher Gauge Theories.J. Phys. A, 2020. arXiv:1911.06390

    arXiv 2020

  12. [12]

    Palmer, C

    S. Palmer, C. S¨ amann. Adjusting Higher Chern–Simons Theory. arXiv:2507.02082, 2025

    arXiv 2025

  13. [13]

    S. Ebli, M. Defferrard, G. Spreemann. Simplicial Neural Networks.NeurIPS Workshop on TDA and Beyond,

  14. [14]

    T. M. Roddenberry, S. Segarra. Principled Simplicial Neural Networks for Trajectory Prediction.ICML, 2021

    2021

  15. [15]

    Battiloro et al.E(n)-Equivariant Topological Neural Networks

    C. Battiloro et al.E(n)-Equivariant Topological Neural Networks. 2024. arXiv:2405.15429

    arXiv 2024

  16. [16]

    Copresheaf Topological Neural Networks.NeurIPS, 2025

    2025

  17. [17]

    Zhang et al

    Y. Zhang et al. Machine Learning Topological Invariants with Neural Networks.Phys. Rev. Lett.120, 066401, 2018. 9

    2018

  18. [18]

    Tipton et al

    C. Tipton et al. Haldane Bundles: A Dataset for Learning to Predict the Chern Number of Line Bundles on the Torus.NeurIPS, 2023. arXiv:2312.04600

    arXiv 2023

  19. [19]

    Ohtsuki et al

    T. Ohtsuki et al. Deep Learning Topological Invariants of Band Insulators.Phys. Rev. B98, 085402, 2018

    2018

  20. [20]

    J. C. Baez, U. Schreiber. Higher Gauge Theory: 2-Connections on 2-Bundles. arXiv:hep-th/0412325

    Pith/arXiv arXiv

  21. [21]

    Bodnar, F

    C. Bodnar, F. Frasca, Y. G. Wang, et al. Weisfeiler and Lehman Go Topological: Message Passing Simplicial Networks.ICML, 2021. arXiv:2103.03212

    arXiv 2021

  22. [22]

    Fukui, Y

    T. Fukui, Y. Hatsugai, H. Suzuki. Chern Numbers in Discretized Brillouin Zone.J. Phys. Soc. Jpn.74:1674, 2005. 10

    2005