pith. sign in

arxiv: 2605.16847 · v1 · pith:2ZL4JJXYnew · submitted 2026-05-16 · 🧮 math.AP

Equivariant nonlinear partial differential operators on constant curvature spaces

Francesco Ballerin , Erlend Grong This is my paper

Pith reviewed 2026-05-19 20:51 UTC · model grok-4.3

classification 🧮 math.AP
keywords equivariant differential operatorsnonlinear PDE operatorsconstant curvature spacesmultigraphsclassifying spacesisometry groupsPDE learning
0
0 comments X

The pith

Nonlinear equivariant differential operators on constant-curvature spaces are classified by a vector space of multigraph equivalence classes.

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

The paper gives a classifying space for nonlinear operators on simply connected constant-curvature spaces that are equivariant under the isometry group. These operators are those expressible as polynomials in linear operators. The classifying space is realized as the vector space spanned by equivalence classes of multigraphs. This allows discovery of non-trivial linear dependence relations among the operators that depend on the manifold dimension. The work also comments on operators under the identity component of the isometry group and on sub-Riemannian model spaces.

Core claim

For nonlinear operators that can be written as a polynomial in linear operators and that are equivariant under the action of the isometry group on simply connected spaces with constant curvature, the classifying space can be realized as the vector space spanned by equivalence-classes of multigraphs. This realization helps discover non-trivial linear dependence relations between nonlinear differential operators that depend on the dimension of the manifold.

What carries the argument

The vector space spanned by equivalence-classes of multigraphs, which realizes the classifying space for the equivariant nonlinear operators.

If this is right

  • Non-trivial linear dependence relations between such operators can be found using the multigraph representation.
  • The classification applies to operators on spheres, Euclidean space, and hyperbolic space of any dimension.
  • Similar classifying spaces may exist for operators equivariant under the connected component of the isometry group.
  • Applications to sub-Riemannian geometry are possible with analogous constructions.

Where Pith is reading between the lines

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

  • The multigraph approach could be used to automate the search for operators in machine learning for PDEs by eliminating redundant ones.
  • Connections to other geometric invariants might emerge from studying these graph equivalences in higher dimensions.
  • Testing the classification on explicit low-dimensional examples could verify the dimension-dependent relations.

Load-bearing premise

The operators under consideration are precisely those that can be expressed as polynomials in linear operators.

What would settle it

A concrete falsifier would be the explicit construction of a nonlinear equivariant operator on a constant-curvature space that cannot be expressed as a polynomial in linear operators and lies outside the multigraph span.

Figures

Figures reproduced from arXiv: 2605.16847 by Erlend Grong, Francesco Ballerin.

Figure 1
Figure 1. Figure 1: Three multigraphs with the same number of vertices. Graphs (a) and (b) are isomorphic, while graph (c) is not isomorphic to them. In the combinatorial study of multigraphs, one natural grading arises from the total number of edges p = |E[γ]|, which we will refer to as the degree of the multigraph (not to be confused with the vertex degree). We will frequently classify multigraphs according to their degree,… view at source ↗
Figure 2
Figure 2. Figure 2: Multigraphs of Γ( ˜ ρ1, [2, 2]) (left) and Γ( ˜ ρ2, [2, 2]) (right), which are not isomorphic. Proposition 2.3. Every multigraph γ with p edges is isomorphic to Γ( ˜ ρ, β) for some degree vector β⃗ ∈ −−→Deg with |β|E = p. Proof. Let w1, . . . , wn be the vertices ordered so that deg(w1) ≤ · · · ≤ deg(wn). Set lI = deg(wI ) and define β(j) := #{I : lI = j}. Then |β|V = n and, by the handshaking lemma, |β|E … view at source ↗
Figure 3
Figure 3. Figure 3: Two alternative representations of the same differential opera￾tor tr(∇ 4,Sym ∗0,∗0,∗1,∗2 f)(∇ 3,Sym ∗1,∗2,∗3 f)(∇∗3 f) as a multigraph (a) and as a weighted graph (b). Example 4.15 (Star multigraph). Let γ be the star multigraph with central vertex v4 and three leaf vertices v1, v2, v3. We allow nj ≥ 0 loops at vertex vj and ej ≥ 1 edges between v4 and vj , for j = 1, 2, 3. In the weighted-graph represent… view at source ↗
Figure 4
Figure 4. Figure 4: Weighted star with four vertices. Node labels nj ≥ 0 denote the number of loops at vertex vj ; edge labels ej ≥ 1 denote the number of edges between the center v4 and leaf vj . The degree of the central vertex is deg(v4) = 2n4 + e1 + e2 + e3, and the degree of leaf vj is deg(vj ) = 2nj + ej , for j = 1, 2, 3. We can contract the corresponding indeces to obtain the opertator Nγf = * (tr∗,∗) n4∇2n4+e1+e2+e3,… view at source ↗
Figure 5
Figure 5. Figure 5: Multigraph representatives for the degree vector β⃗ = [0, 3]. Case β⃗ = [0, 3]. There are three possible corresponding graphs representatives as in [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Multigraph representatives associated to the orbits for the degree vector β⃗ = [2, 2]. Case β⃗ = [2, 2]. There are four possible graphs as in [PITH_FULL_IMAGE:figures/full_fig_p022_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: A graphical way to represent the operators of Example 7.2. We have an edge for every trace, with the red edges going into the vertex ω denoting the traces with the volume form. These graphs are directed so that arrows pointing into ω denotes taking a trace in at an odd argument, while an arrow away from ω denotes a trace in an even input. There must be 0 or 1 more arrows towards ω when respectively d is ev… view at source ↗
read the original abstract

Motivated by PDE-learning, we give a classifying space for nonlinear operators on simply connected spaces with constant curvature which are also equivariant under the action of the isometry group. The nonlinear operators we are considering are those that can be written as a polynomial in linear operators. We show that the classifying space for such operators can be realized as the vector space spanned by equivalence-classes of multigraphs. We also illustrate how this realization can help us discover non-trivial linear dependence relations between nonlinear differential operators relative to the dimension of the manifold. We also give some comments on operators equivariant under the identity component of the isometry group and under isometry groups of sub-Riemannian model spaces.

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

2 major / 3 minor

Summary. The manuscript constructs a classifying space for nonlinear differential operators on simply connected constant-curvature manifolds that are equivariant under the full isometry group. The operators considered are those expressible as polynomials in linear differential operators. The central result realizes this classifying space as the vector space spanned by equivalence classes of multigraphs. The construction is then used to identify dimension-dependent linear dependence relations among such operators, with additional remarks on equivariance under the identity component of the isometry group and on sub-Riemannian model spaces.

Significance. If the claimed isomorphism between the space of polynomial equivariant operators and the span of multigraph equivalence classes holds, the work supplies a concrete combinatorial model for classifying and relating nonlinear geometric operators. This could aid both theoretical analysis of operator algebras on space forms and practical tasks such as automated discovery of identities in PDE learning. The explicit multigraph realization is a methodological strength that permits direct computation of dimensions and relations once the equivalence is fully verified.

major comments (2)
  1. [realization theorem / classifying-space construction] In the section presenting the realization theorem: the claim that the vector space of equivalence classes of multigraphs realizes the classifying space requires an explicit argument that the chosen equivalence relation on multigraphs precisely encodes both the isometry-group action and the algebraic identities satisfied by the curvature tensor (including commutation relations among covariant derivatives that become nontrivial at fixed dimension). A purely combinatorial equivalence (e.g., vertex relabeling or graph isomorphism) would generally miss curvature-induced relations and could therefore produce an incorrect dimension for the spanned space.
  2. [application to linear dependence relations] In the application to linear dependence relations: the manuscript asserts that the multigraph model reveals nontrivial dimension-dependent relations, yet no concrete low-dimensional example is worked out in which the kernel dimension predicted by the graph space is independently verified by direct computation on the manifold. Such a check is load-bearing for the utility claim.
minor comments (3)
  1. [introduction / preliminaries] The abstract and introduction use the phrase 'polynomial in linear operators' without an early formal definition of the precise algebra generated; a short preliminary subsection spelling out the ring structure would improve readability.
  2. [throughout] Notation for the equivalence relation on multigraphs (e.g., the symbol denoting the equivalence) should be introduced once and used consistently; occasional shifts between 'equivalence classes' and 'orbits' are mildly confusing.
  3. [final remarks] The remarks on sub-Riemannian model spaces are brief; if they are intended only as outlook, a single sentence clarifying their scope relative to the main Riemannian results would suffice.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading of our manuscript and for the constructive comments, which will help improve the clarity and strength of our results. We address each major comment below.

read point-by-point responses

  1. Referee: In the section presenting the realization theorem: the claim that the vector space of equivalence classes of multigraphs realizes the classifying space requires an explicit argument that the chosen equivalence relation on multigraphs precisely encodes both the isometry-group action and the algebraic identities satisfied by the curvature tensor (including commutation relations among covariant derivatives that become nontrivial at fixed dimension). A purely combinatorial equivalence (e.g., vertex relabeling or graph isomorphism) would generally miss curvature-induced relations and could therefore produce an incorrect dimension for the spanned space.

    Authors: We agree that the realization theorem would be strengthened by a more explicit argument showing how the equivalence relation on multigraphs encodes the isometry-group action together with the algebraic identities of the curvature tensor, including the dimension-dependent commutation relations among covariant derivatives. In the revised manuscript we will expand the relevant section with a detailed explanation of this encoding, derived directly from the definitions of the polynomial operators and the curvature properties of constant-curvature spaces. revision: yes

  2. Referee: In the application to linear dependence relations: the manuscript asserts that the multigraph model reveals nontrivial dimension-dependent relations, yet no concrete low-dimensional example is worked out in which the kernel dimension predicted by the graph space is independently verified by direct computation on the manifold. Such a check is load-bearing for the utility claim.

    Authors: We acknowledge that an explicit low-dimensional verification would make the utility of the model more convincing. We will add a new subsection containing a concrete example (for instance in dimension 3) in which the dimension of the kernel predicted by the multigraph equivalence classes is computed combinatorially and then independently confirmed by direct calculation on the manifold using the known curvature and covariant-derivative identities. revision: yes

Circularity Check

0 steps flagged

No circularity: classifying space realized via explicit correspondence to multigraphs under equivariance and polynomial structure

full rationale

The paper constructs a classifying space for polynomial nonlinear operators that are equivariant under the isometry group of constant-curvature spaces. The abstract states that this space 'can be realized as the vector space spanned by equivalence-classes of multigraphs,' which is presented as a derived combinatorial model rather than a definitional tautology. No quoted step reduces the central isomorphism or spanning claim to a fitted parameter, a self-referential definition, or a load-bearing self-citation whose content is itself unverified. The polynomial restriction on operators and the equivariance condition supply independent algebraic and representation-theoretic inputs; the multigraph realization is offered as an organizational tool that also reveals dimension-dependent linear dependences. Because the derivation remains self-contained against these external structures and no specific equation or theorem is shown to collapse by construction, the circularity score is 0.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The paper relies on standard assumptions about constant curvature spaces and isometry group actions, plus the restriction to polynomial nonlinear operators; it introduces multigraph equivalence classes as the classifying objects without external independent evidence.

axioms (2)
  • domain assumption The spaces are simply connected with constant sectional curvature.
    Invoked to define the setting for the isometry-equivariant operators.
  • domain assumption The nonlinear operators are polynomials in linear operators.
    Explicitly stated as the class of operators for which the classifying space is constructed.
invented entities (1)
  • Equivalence classes of multigraphs no independent evidence
    purpose: To form a basis for the classifying vector space of the equivariant operators.
    Introduced as the concrete realization of the abstract classifying space.

pith-pipeline@v0.9.0 · 5635 in / 1312 out tokens · 38948 ms · 2026-05-19T20:51:32.991127+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

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/Foundation/AlexanderDuality.lean alexander_duality_circle_linking echoes
    ?
    echoes

    ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

    We show that the classifying space for such operators can be realized as the vector space spanned by equivalence-classes of multigraphs... N: span_R MG → NPDO(M)^G ... bijective when p ≤ d

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

20 extracted references · 20 canonical work pages

  1. [1]

    Andersdotter, D

    E. Andersdotter, D. Persson, and F. Ohlsson. Equivariant manifold neural odes and differential invariants. arXiv preprint arXiv:2401.14131, 2024

    work page arXiv 2024

  2. [2]

    Bella¨ ıche

    A. Bella¨ ıche. The tangent space in sub-riemannian geometry. InSub-Riemannian geometry, pages 1–78. Springer, 1996

    work page 1996

  3. [3]

    Berge and E

    E. Berge and E. Grong. OnG2 and sub-riemannian model spaces of step and rank three.Mathematische Zeitschrift, 298(3):1853–1885, 2021

    work page 2021

  4. [4]

    Bollob´ as.Modern Graph Theory, volume 184 ofGraduate Texts in Mathematics

    B. Bollob´ as.Modern Graph Theory, volume 184 ofGraduate Texts in Mathematics. Springer-Verlag New York, 1998

    work page 1998

  5. [5]

    L. Gao, Y. Du, H. Li, and G. Lin. Roteqnet: Rotation-equivariant network for fluid systems with symmetric high-order tensors.Journal of Computational Physics, 461:111205, 2022

    work page 2022

  6. [6]

    E. Grong. Model spaces in sub-riemannian geometry.Communications in Analysis and Geometry, 29(1):77– 113, 2021

    work page 2021

  7. [7]

    E. Grong. Canonical connections on sub-riemannian manifolds with constant symbol.Analysis and Mathe- matical Physics, 16(2):21, 2026

    work page 2026

  8. [8]

    Grong, H

    E. Grong, H. Z. Munthe-Kaas, and J. Stava. Post-Lie algebra structure of manifolds with constant curvature and torsion.J. Lie Theory, 34(2):339–352, 2024

    work page 2024

  9. [9]

    Harary.Graph Theory

    F. Harary.Graph Theory. Addison-Wesley, Reading, MA, 1969

    work page 1969

  10. [10]

    L. He, Y. Chen, Z. Shen, Y. Yang, and Z. Lin. Neural epdos: Spatially adaptive equivariant partial differential operator based networks. InThe eleventh international conference on learning representations, 2022

    work page 2022

  11. [11]

    Helgason

    S. Helgason. Differential operators on homogeneous spaces.Acta Math., 102:239–299, 1959

    work page 1959

  12. [12]

    N. Iwahori. Some remarks on tensor invariants of O(n), U(n), Sp(n).Journal of the Mathematical Society of Japan, 10(2):145–160, 1958

    work page 1958

  13. [13]

    Kobayashi and K

    S. Kobayashi and K. Nomizu.Foundations of differential geometry, volume 2, volume 2. John Wiley & Sons, 1996

    work page 1996

  14. [14]

    Lagrave and E

    P.-Y. Lagrave and E. Tron. Equivariant neural networks and differential invariants theory for solving partial differential equations. InPhysical Sciences Forum, volume 5, page 13. MDPI, 2022

    work page 2022

  15. [15]

    Laurent and H

    A. Laurent and H. Munthe-Kaas. The universal equivariance properties of exotic aromatic b-series.Founda- tions of Computational Mathematics, 25(5):1595–1626, 2025

    work page 2025

  16. [16]

    J. M. Lee.Introduction to Riemannian manifolds, volume 2. Springer, 2018

    work page 2018

  17. [17]

    Nyholm, O

    E. Nyholm, O. Carlsson, M. Weiler, and D. Persson. Equivariant non-linear maps for neural networks on homogeneous spaces.arXiv preprint arXiv:2504.20974, 2025

    work page arXiv 2025

  18. [18]

    R. W. Sharpe.Differential geometry: Cartan’s generalization of Klein’s Erlangen program, volume 166. Springer Science & Business Media, 2000

    work page 2000

  19. [19]

    Weyl.The classical groups: their invariants and representations, volume 1

    H. Weyl.The classical groups: their invariants and representations, volume 1. Princeton university press, 1946. 28 AppendixA.Tables # edges Graph class Representative Associated operator 0 empty graph∅1 0 isolated vertex ∇0f=f 1 single edge ∥∇f∥2 1 single loop tr(∇2f) = ∆f 2 path of length two ∇2f(∇f,∇f) 2 double edge ∥∇2f∥2 2 loop and edge ⟨∇∆f,∇f⟩ 2 dou...

    work page 1946

  20. [20]

    For ⃗β= (β 1, β2,

    S6 Table 3.All degree vectors ⃗βwith∥ ⃗β∥E = 3, the corresponding graph repre- sentatives, and the symmetry groupS ⃗β ⊂S 6. For ⃗β= (β 1, β2, . . .), the group is ⃗β =Q j≥1 Sβj j ⋊S βj , whereS βj j permutes the slots within theβ j vertices of degreej, andS βj permutes those vertices among themselves

    work page