pith. sign in

arxiv: 2606.24621 · v1 · pith:ILUDS3XVnew · submitted 2026-06-23 · 🧮 math.CT · cs.AI· math.ST· stat.TH

Infinitesimal Causality

Sridhar Mahadevan This is my paper

Pith reviewed 2026-06-25 21:39 UTC · model grok-4.3

classification 🧮 math.CT cs.AImath.STstat.TH
keywords infinitesimal causalityFrobenius Markov categoriestangent bundle semanticscategorical causal sufficiencyintervention distributionsdo-calculusstructural causal modelsLie brackets
0
0 comments X

The pith

Categorical causal sufficiency holds when algebraic Frobenius structure on variables is compatible with geometric involutive closure of interventions.

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

The paper develops a categorical account of infinitesimal causality inside Frobenius Markov categories equipped with tangent-bundle semantics. Interventions appear as tangent vectors that deform the copy, compare, and discard operations encoded by the categorical Frobenius algebra. Categorical causal sufficiency is defined exactly as the compatibility between this algebraic structure and the geometric requirement that the intervention distribution be involutively closed. The framework reformulates Pearl's do-calculus identities as counit invariance, coproduct compatibility under pushforward, and Lie-bracket closure. A reader cares because the approach supplies an algebraic and geometric language for how interventions preserve or break classical information flow at the infinitesimal level.

Core claim

The central claim is that, for structural causal models, infinitesimal causality is formulated most naturally in the slice of deterministic mechanisms over exogenous variables, with visible stochastic kernels recovered only after pushforward. Interventions act as tangent vectors deforming the categorical Frobenius copy/discard operations; their Lie brackets measure whether the deformation preserves classical information-flow structure. Categorical causal sufficiency is the compatibility of the categorical Frobenius algebra on classical variables with the geometric Frobenius integrability condition of involutive closure of the intervention distribution.

What carries the argument

Tangent-bundle semantics in Frobenius Markov categories, in which interventions function as tangent deformations of the categorical Frobenius algebra on classical variables.

If this is right

  • Ignoring irrelevant interventions corresponds to counit invariance of the Frobenius algebra.
  • Action/observation exchange corresponds to coproduct compatibility under pushforward.
  • Independence of interventions corresponds to involutive bracket closure of the visible intervention distribution.
  • Visible stochastic kernels arise only after pushforward from the deterministic slice over exogenous variables.

Where Pith is reading between the lines

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

  • The slice-category formulation separates deterministic mechanism structure from observed stochasticity, which may simplify proofs of causal identifiability.
  • Lie-bracket closure supplies a differential-geometric test that could be discretized for finite causal graphs.
  • The same compatibility condition may extend directly to other categorical models of processes that carry copy/discard structure.

Load-bearing premise

Infinitesimal causality for structural causal models is most naturally formulated in the slice of deterministic mechanisms over exogenous variables, with visible stochastic kernels recovered only after pushforward.

What would settle it

Construct a structural causal model in which two intervention tangent vectors have a nonzero Lie bracket yet the model still satisfies every identity of Pearl's do-calculus; the existence of such a model would falsify the claim that involutive closure is required for categorical causal sufficiency.

Figures

Figures reproduced from arXiv: 2606.24621 by Sridhar Mahadevan.

Figure 1
Figure 1. Figure 1: Stat∞: a Markov model presented by a suffi￾cient statistic. The black nodes are the copied classical data, and the dashed arrows show the tangent score direc￾tion induced by v. T f pU U f t X S v˜ ∈ T U T f(˜v) ∈ T X rij exogenous [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: The three infinitesimal intervention rules as string-diagrammatic equations. Rule 1 says that discarding an [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
read the original abstract

This paper introduces a categorical account of infinitesimal causality in Frobenius Markov categories equipped with tangent-bundle semantics. IDC captures the infinitesimal layer in which interventions act as tangent deformations of copy/discard structure. Two distinct Frobenius structures interact: (1) the categorical Frobenius algebra on classical variables encoding copying, comparing, and discarding; and (2) the geometric Frobenius integrability condition, namely involutive closure of the intervention distribution, distinct from the algebraic Frobenius structure. Categorical causal sufficiency is defined as the compatibility of these two notions. A key observation is that, for structural causal models, infinitesimal causality is most naturally formulated in the slice of deterministic mechanisms over exogenous variables, with visible stochastic kernels obtained only after pushforward. Interventions are tangent vectors that deform the Frobenius copy/discard operations; their Lie brackets measure whether this deformation preserves classical information-flow structure. Pearl's do-calculus is used as a guiding example of intervention identities: ignoring irrelevant interventions corresponds to counit invariance, action/observation exchange to coproduct compatibility with pushforward, and independence to involutive bracket closure of the visible intervention distribution.

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 / 2 minor

Summary. The paper introduces infinitesimal causality (IDC) in Frobenius Markov categories equipped with tangent-bundle semantics. IDC is defined via compatibility of two Frobenius structures: the categorical Frobenius algebra on classical variables (encoding copy/compare/discard) and the geometric Frobenius integrability condition (involutive closure of the intervention distribution). Categorical causal sufficiency is this compatibility. The central claim is that, for structural causal models, IDC is naturally formulated in the slice category of deterministic mechanisms over exogenous variables, with visible stochastic kernels recovered by pushforward; Lie brackets of tangent vectors then measure preservation of classical information-flow structure. Pearl's do-calculus identities are recovered as counit invariance, coproduct compatibility, and involutive bracket closure.

Significance. If the compatibility definition and the pushforward transfer are rigorously established, the work would supply a novel infinitesimal layer for categorical causality that integrates algebraic Frobenius structure with geometric integrability, potentially clarifying intervention semantics in Markov categories. The explicit use of tangent deformations and Lie brackets to track information-flow preservation is a distinctive technical contribution.

major comments (2)
  1. [slice formulation / abstract] The key observation (abstract and the section introducing the slice formulation) states that IDC is most naturally formulated in the slice of deterministic mechanisms over exogenous variables, with stochastic kernels obtained only after pushforward. No derivation is supplied showing that the pushforward functor preserves tangent deformations and Lie-bracket closure in a manner that transfers the involutive-closure condition to the visible-level intervention identities (counit invariance and coproduct compatibility) without additional functoriality hypotheses on the pushforward.
  2. [definition of categorical causal sufficiency] Definition of categorical causal sufficiency as compatibility of the two Frobenius structures (categorical algebraic and geometric integrability) is presented as the central notion, yet the manuscript supplies no independent check or example verifying that this compatibility implies the stated do-calculus identities once the pushforward is applied.
minor comments (2)
  1. Notation for the tangent bundle and the two distinct Frobenius structures should be introduced with explicit comparison to standard references in categorical probability.
  2. The abstract claims that Lie brackets measure preservation of classical information-flow structure; a brief illustrative computation in a simple SCM would clarify this claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful and constructive report. The two major comments correctly identify places where the manuscript would be strengthened by additional explicit derivations and verifications. We respond point-by-point below and will incorporate the requested material in a revised version.

read point-by-point responses

  1. Referee: [slice formulation / abstract] The key observation (abstract and the section introducing the slice formulation) states that IDC is most naturally formulated in the slice of deterministic mechanisms over exogenous variables, with stochastic kernels obtained only after pushforward. No derivation is supplied showing that the pushforward functor preserves tangent deformations and Lie-bracket closure in a manner that transfers the involutive-closure condition to the visible-level intervention identities (counit invariance and coproduct compatibility) without additional functoriality hypotheses on the pushforward.

    Authors: We agree that an explicit derivation of preservation under pushforward is required for rigor. In the revision we will add a dedicated subsection proving that the pushforward, induced by the deterministic mechanism in the slice, is a strict monoidal functor that commutes with the tangent-bundle construction and therefore preserves both tangent deformations and Lie brackets. This transfers the involutive-closure condition directly to the visible-level counit-invariance and coproduct-compatibility identities using only the existing Markov-category axioms. revision: yes

  2. Referee: [definition of categorical causal sufficiency] Definition of categorical causal sufficiency as compatibility of the two Frobenius structures (categorical algebraic and geometric integrability) is presented as the central notion, yet the manuscript supplies no independent check or example verifying that this compatibility implies the stated do-calculus identities once the pushforward is applied.

    Authors: We acknowledge that an independent verification would make the implication clearer. The revision will include a short worked example (a simple chain structural causal model with exogenous noise) that computes the two Frobenius structures in the slice, applies the pushforward, and explicitly checks that compatibility yields counit invariance, coproduct compatibility, and involutive bracket closure, thereby recovering the corresponding do-calculus identities at the visible level. revision: yes

Circularity Check

0 steps flagged

No significant circularity; framework consists of explicit definitions without reduction to inputs

full rationale

The paper introduces IDC as the infinitesimal layer of tangent deformations on copy/discard structure and explicitly defines categorical causal sufficiency as the compatibility of the categorical Frobenius algebra (copy/compare/discard) with the geometric Frobenius integrability condition (involutive closure). The slice formulation for structural causal models is presented as a key observation, with visible kernels obtained by pushforward and Lie brackets measuring preservation; Pearl's do-calculus is invoked only as a guiding example for interpreting counit invariance, coproduct compatibility, and bracket closure. No equations, fitted parameters, or predictions are shown that reduce by construction to prior inputs, no self-citations appear, and no ansatz or uniqueness theorem is smuggled in. The derivation chain is therefore self-contained as a definitional framework rather than a circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The framework rests on domain assumptions from category theory and causal modeling plus ad-hoc definitions of compatibility and deformation; no free parameters or invented entities with independent evidence are stated.

axioms (2)
  • domain assumption Frobenius Markov categories equipped with tangent-bundle semantics exist and are suitable for modeling causality
    Invoked as the setting for the entire account of infinitesimal causality.
  • ad hoc to paper The two Frobenius structures (categorical algebraic and geometric integrability) interact via compatibility to define causal sufficiency
    This compatibility is introduced as the key definition without derivation from prior results.
invented entities (1)
  • Infinitesimal causality (IDC) no independent evidence
    purpose: Captures the infinitesimal layer where interventions act as tangent deformations of copy/discard structure
    New concept introduced to bridge algebraic and geometric Frobenius notions.

pith-pipeline@v0.9.1-grok · 5721 in / 1418 out tokens · 24190 ms · 2026-06-25T21:39:01.487369+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

18 extracted references · 7 canonical work pages

  1. [1]

    [Dob06] Ernst-Erich Doberkat

    K. Cho and B. Jacobs. Disintegration and Bayesian inversion via string diagrams. Mathematical Structures in Computer Science, 29 0 (7): 0 938--971, 2019. doi:10.1017/S0960129518000488. URL https://arxiv.org/abs/1709.00322

    work page doi:10.1017/s0960129518000488 2019

  2. [2]

    J. R. B. Cockett and G. S. H. Cruttwell. Differential structure, tangent structure, and SDG . Applied Categorical Structures, 22: 0 331--417, 2014. doi:10.1007/s10485-013-9312-0

    work page doi:10.1007/s10485-013-9312-0 2014

  3. [3]

    T. Fritz. A synthetic approach to Markov kernels, conditional independence and theorems on sufficient statistics. Advances in Mathematics, 370: 0 107239, 2020. doi:10.1016/j.aim.2020.107239. URL https://arxiv.org/abs/1908.07021

    work page doi:10.1016/j.aim.2020.107239 2020

  4. [4]

    Fritz and A

    T. Fritz and A. Klingler. The d-Separation criterion in categorical probability. Journal of Machine Learning Research, 24 0 (46): 0 1--49, 2023. URL https://jmlr.org/papers/v24/22-0916.html

    2023

  5. [5]

    S. B. Gillispie and M. D. Perlman. Enumerating Markov equivalence classes of acyclic digraph models. In Proceedings of the Seventeenth Conference on Uncertainty in Artificial Intelligence, pages 171--177, 2001. URL https://arxiv.org/abs/1301.2272

    Pith/arXiv arXiv 2001

  6. [6]

    S. Guo, V. T \'o th, B. Sch \"o lkopf, and F. Husz \'a r. Causal de Finetti : On the identification of invariant causal structure in exchangeable data. In Advances in Neural Information Processing Systems, 2022. URL https://arxiv.org/abs/2203.15756

    arXiv 2022

  7. [7]

    S. Guo, C. Zhang, K. Mohan, F. Husz \'a r, and B. Sch \"o lkopf. Do Finetti : On causal effects for exchangeable data. In Advances in Neural Information Processing Systems, 2024. URL https://arxiv.org/abs/2405.18836

    arXiv 2024

  8. [8]

    Jacobs, A

    B. Jacobs, A. Kissinger, and F. Zanasi. Causal inference by string diagram surgery. arXiv preprint arXiv:1811.08338, 2018. URL https://arxiv.org/abs/1811.08338

    Pith/arXiv arXiv 2018

  9. [9]

    Mahadevan

    S. Mahadevan. Causal density functions, 2026 a . URL https://arxiv.org/abs/2606.00754

    Pith/arXiv arXiv 2026

  10. [10]

    Mahadevan

    S. Mahadevan. Latent Confounded Causal Discovery via Lie Bracket Geometry , 2026 b . URL https://arxiv.org/abs/2606.19610

    Pith/arXiv arXiv 2026

  11. [11]

    J. Pearl. Causality: Models, Reasoning, and Inference. Cambridge University Press, 2 edition, 2009

    2009

  12. [12]

    The Annals of Statistics , author =

    T. Richardson and P. Spirtes. Ancestral graph Markov models. The Annals of Statistics, 30 0 (4): 0 962--1030, 2002. doi:10.1214/aos/1031689015

    work page doi:10.1214/aos/1031689015 2002

  13. [13]

    Rosick \'y

    J. Rosick \'y . Abstract tangent functors. Diagrammes, 12: 0 JR1--JR11, 1984

    1984

  14. [14]

    D. B. Rubin. Causal inference using potential outcomes: Design, modeling, decisions. Journal of the American Statistical Association, 100 0 (469): 0 322--331, 2005. doi:10.1198/016214504000001880

    work page doi:10.1198/016214504000001880 2005

  15. [15]

    Schmid and A

    D. Schmid and A. Sly. On the number and size of Markov equivalence classes of random directed acyclic graphs, 2022. URL https://arxiv.org/abs/2209.04395

    arXiv 2022

  16. [16]

    Spirtes, C

    P. Spirtes, C. Glymour, and R. Scheines. Causation, Prediction, and Search. MIT Press, 2 edition, 2000

    2000

  17. [17]

    Studen \'y

    M. Studen \'y . Probabilistic Conditional Independence Structures. Information Science and Statistics. Springer London, 2005. ISBN 978-1-85233-891-6. doi:10.1007/b138557. URL https://link.springer.com/book/10.1007/b138557. Softcover ISBN 978-1-84996-948-2, published 2010

    work page doi:10.1007/b138557 2005

  18. [18]

    J. Zhang. On the completeness of orientation rules for causal discovery in the presence of latent confounders and selection bias. Artificial Intelligence, 172 0 (16--17): 0 1873--1896, 2008. doi:10.1016/j.artint.2008.08.001

    work page doi:10.1016/j.artint.2008.08.001 2008