Pith. sign in

REVIEW 4 minor 2 cited by

This paper establishes that systems of causal stochastic differential equations satisfy the σ-separation Markov property and the three do-calculus rules at the level of sample paths, with a stronger d-separation Markov property for additive

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 06:42 UTC pith:NEM6LC6T

load-bearing objection A strong, honest theory paper on causal SDEs whose main Markov/do-calculus results look solid, but the proof of Theorem 17 (Granger/local independence) has a real gap that needs fixing.

arxiv 2607.12140 v2 pith:NEM6LC6T submitted 2026-07-13 math.ST stat.MLstat.TH

Causal Graphs, Markov Properties and Do-calculus for Stochastic Differential Equations

Philip Boeken , Joris M. Mooij This is my paper
classification math.ST stat.MLstat.TH MSC 60H1060G4462M02
keywords causal stochastic differential equationsσ-separationd-separationdo-calculussample-path independenceessential unique solvabilitytime-split systemsGranger causality
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper asks whether the standard toolkit of graphical causal modeling — causal graphs, Markov properties, and the do-calculus — can be rigorously transferred from static or discrete-time systems to continuous-time dynamics described by stochastic differential equations. It answers yes, provided the SDE system satisfies a weak solvability condition called essential unique solvability. Treating each solution path as a single random variable, the paper proves that σ-separation in the system's causal graph implies conditional independence between the corresponding sample paths, and that all three rules of do-calculus hold. For additive-noise SDEs driven by Brownian motion with an invertible diffusion matrix, it proves the stronger d-separation Markov property even in cyclic systems. A direct corollary gives the graph a clean causal reading: if there is no directed path from u to v, then intervening on u does not change the law of v.

Core claim

The paper treats each solution path X_v of an SDE system as a single random variable and develops a structural causal model on path space. Under essential unique solvability — a weak fixed-point condition that also survives marginalisation — the causal graph G(D) correctly encodes conditional independence: any σ-separation in the graph implies conditional independence of the corresponding sample-path variables, and the three rules of do-calculus hold. For additive-noise SDEs driven by Brownian motion with Lipschitz drift and invertible diffusion matrix, this strengthens to d-separation, including cyclic systems, via total-variation convergence of an interpolated Euler scheme. A corollary giv

What carries the argument

The workhorse is a pathwise, Skorokhod-measurable version of the Itô integral (the map Ψ) that lets equations be read as structural equations on path space. On top of that, the paper introduces essential unique solvability, a weak fixed-point condition that ensures well-defined observational and interventional distributions even after marginalising out variables. The σ-separation Markov property is obtained by an acyclification step: each strongly connected component of the graph is replaced by its solution function, producing an acyclic SCM whose d-separation matches σ-separation in the original graph. For the d-separation result, the proof discretizes the SDE into an acyclic Euler scheme,

Load-bearing premise

The d-separation Markov property rests on the assumption that the Brownian noise has full rank (invertible diffusion matrix Σ); when Σ is singular, the Euler scheme and the true solution are mutually singular with respect to the reference measure, so the total-variation argument that transfers conditional independence collapses.

What would settle it

Simulate a cyclic additive-noise SDE with an invertible diffusion matrix and test the claimed conditional independence for a d-separation in its causal graph: any persistent dependence in the sample-path distribution, or any failure of the interpolated Euler law to converge in total variation to the true law, would refute Theorem 11.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If there is no directed path from u to v in the causal graph, then the distribution of X_v is the same with and without an intervention on X_u; the graph's edges are exactly the possible causal routes.
  • Marginalising out a subset of latent processes preserves essential unique solvability and leaves observational and interventional distributions on the remaining processes unchanged, so inference on the observed part is valid.
  • Time-split systems give the σ- and d-separation Markov properties for variables on disjoint intervals, formalising continuous-time Granger non-causality and local independence as graphical statements.
  • Because the Markov properties hold on sample paths, constraint-based discovery algorithms that are sound for cyclic or acyclic graphs (FCI, CCI) apply directly to SDE data, provided conditional independence between paths can be tested consistently.
  • The d-separation Markov property holds for cyclic additive-noise systems, extending the known list of cyclic SCMs that admit the stronger Markov property to a continuous-time class.

Where Pith is reading between the lines

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

  • The same acyclification strategy should transfer to other infinite-dimensional structural models, such as systems driven by SPDEs or rough paths, as long as solution functions and a solvability notion analogous to essential unique solvability are available.
  • The invertible-diffusion assumption in the d-separation proof is likely not fundamental: for singular diffusion, one could try a different reference measure (e.g. a lower-dimensional noise) and hope for a similar total-variation argument in a restricted graph, though the paper leaves this open.
  • The time-split framework suggests that subsampled time-series display conditional independencies that the underlying continuous-time system does not; quantifying this distortion might give practical guidance on sampling rates for causal discovery.
  • For additive-noise SDEs, the d-separation do-calculus is conjectured but not proven; if it holds, causal effect identification for cyclic continuous-time systems would reduce to graphical criteria exactly as in the acyclic SCM case.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

0 major / 4 minor

Summary. The paper develops a graphical causal modelling framework for systems of causal SDEs. It introduces systems of causal SDEs with a pathwise SCM interpretation, defines essential unique solvability, and gives two sufficient model classes (Assumptions 1 and 2). Its main results are the σ-separation Markov property (Theorem 6), the d-separation Markov property for additive-noise SDEs (Theorem 11), the σ-separation do-calculus (Theorem 12), the no-directed-path/no-causal-effect corollary (Theorem 13), and the transfer of these properties to time-split and subsampled systems (Theorems 14–15). The final sections apply the framework to Granger non-causality (Theorem 16) and local independence (Theorem 17, Corollary 1), and discuss constraint-based discovery.

Significance. If correct, this is a substantial contribution: it gives continuous-time stochastic dynamics an explicit structural-causal semantics and lifts the standard SCM toolkit — Markov properties, do-calculus, marginalisation — to sample-path level. The d-separation Markov property for a class of cyclic additive-noise SDEs is particularly notable. The paper is careful about scope, explicitly flagging the global-Lipschitz restriction and the open d-separation do-calculus and marginalised-system problems. The appendix proofs are detailed. I examined the potential gap in the proof of Theorem 17 raised in the review process and found that it does not survive scrutiny: the disputed equality follows by two tower-property collapses plus the conditional-independence assumption applied to X_B on the future interval. The step is, however, too compressed and should be expanded in revision.

minor comments (4)
  1. [Appendix F.1.1 / Theorem 17] The equality E[(α^C_T)^n | F^C_t] = E[(α^C_T)^n | F^{A,C}_t] is justified, but the text's phrase 'by the assumed conditional independence and the tower property' is too terse. For j>i one should first collapse E[X^C_B(t_j)-X^C_B(t_{j-1}) | F^C_t] to E[X_B(t_j)-X_B(t_{j-1}) | F^C_t] by the tower property, use X^C_B=X^{A,C}_B to do the same on the A,C side, and then apply the conditional-independence assumption with s=t and t'=t_j, since both endpoints lie in (t,t_j]. Please spell this out.
  2. [Abstract] Minor typo: 'We considersystems' is missing a space; there are a few similar LaTeX artifacts in author names in the references.
  3. [Section 6, Theorem 12] In Rule 2 the notation P(X_A | do(X_B), X_C) uses do(X_B) as a random argument; this is standard shorthand but should be explicitly defined (e.g., as the kernel evaluated at the random value X_B) to avoid ambiguity.
  4. [Section 5.2 / Theorem 14(ii)] The sentence 'by the same argument as Lemma 7 adapted to the time-split graph' compresses a nontrivial transfer step between the time-split graph and the Euler graph for interval-indexed vertices. A short lemma or a more detailed proof would improve verifiability.

Circularity Check

0 steps flagged

No circularity found: the SDE-specific Markov properties and do-calculus are derived from the solvability and pathwise representation, not assumed from the target conclusions.

full rationale

The paper's central derivation chain is self-contained rather than circular. Theorem 6 is proved by constructing an acyclification from the essential solution functions of the SDE system, proving observational equivalence, and then invoking the d-separation Markov property for acyclic SCMs as a background lemma; the SDE-specific work is the construction and equivalence proof. Theorem 11 is proved by transferring d-separation to an Euler discretization, extending it to a continuous Euler scheme, proving total-variation convergence via Girsanov densities and Scheffé's theorem, and then exporting the conditional independence via Lauritzen's external result. None of these steps defines its conclusion in terms of its inputs, and no equation is constructed so that the target independence is imposed by definition. The do-calculus (Theorem 12) is derived from the σ-separation Markov property and an auxiliary intervention-variable system, with Lemma E.2 establishing the required distributional equivalences by construction; Theorem 13 is then a direct consequence of Rule 3. The self-citations to Forré & Mooij and Bongers et al. supply general SCM lemmas, but the present results do not reduce to those lemmas: the lemmas are applied to a newly constructed path-space SCM and are not the same as the claimed SDE-specific conclusions. There are no fitted parameters, so no fitted-input-called-prediction pattern exists. The questionable equality in the proof of Theorem 17 is a technical correctness concern about the local-independence theorem, not a circular dependency, and it does not feed back into the main Markov-property or do-calculus results. Section 8.2 explicitly flags the open problems (d-separation for marginalised systems and the d-separation do-calculus), which further confirms that the assumptions are load-bearing but are not hidden assumptions of the target theorems.

Axiom & Free-Parameter Ledger

0 free parameters · 9 axioms · 2 invented entities

The paper introduces no fitted free parameters; it is a pure theorem paper. The load-bearing assumptions are the pathwise SCM interpretation (via Przybyłowicz et al. 2024), the perfect-intervention semantics, the essential-unique-solvability framework with global Lipschitz conditions, and—for the d-separation theorem—additive noise with invertible diffusion. The formal 'invented' objects (intervention variables, time-split variables) are bookkeeping devices, not entities with independent empirical content.

axioms (9)
  • standard math Usual conditions on the filtered probability space; Itô integration and semimartingale decompositions (Protter 2005).
    Section 2; foundational tools used throughout.
  • standard math Przybyłowicz et al. (2024) Theorem 3.1/Cor. A.1: existence of a measurable, adapted pathwise solution map Ψ for semimartingale SDEs under Lipschitz/linear growth.
    Invoked in Section 2.1 / Appendix A as the basis of the pathwise SCM interpretation and Theorem 2.
  • standard math Bongers et al. (2021) Theorem 6.3 (d-separation Markov property for acyclic SCMs) and Forré & Mooij (2025) Prop 3.5.2 (d-separation ↔ σ-separation under graphical acyclification).
    Used in the proof of Theorem 6 (acyclification strategy).
  • standard math Lauritzen (2024): total-variation convergence preserves conditional independence (Theorem 10).
    Used to pass from Euler-scheme CI to continuous-time CI in Theorem 11.
  • standard math Domains are Skorokhod spaces D([0,T],R^n) with J1 topology; they are standard Borel spaces.
    Section 2 footnote; needed for measurable SCM structure.
  • domain assumption Exogenous processes X_w (w∈W) are mutually independent semimartingales and integrators are exogenous (non-intervenable).
    Definition 1; essential for causal graph and independence structure.
  • domain assumption Perfect intervention semantics: setting X_v(t)=x_v(t) and zeroing the mechanisms g_v,h_v for v∈S, with no side effects (Definition 3).
    Definition 3; this is the causal semantics the entire do-calculus rests on.
  • domain assumption Assumption 1/2: global Lipschitz + linear growth; SCC-wise α(v)∩S=∅; additive noise with invertible Σ.
    Section 3.1; guarantees essential unique solvability and, for Theorem 11, the Girsanov density argument.
  • ad hoc to paper Admissibility of inputs in time-split systems: intervention paths on left-open intervals must have left limits and concatenate to right-continuous paths.
    After Definition 8; a framework-specific regularity patch needed to keep càdlàg solutions.
invented entities (2)
  • Intervention indicator variables R_v (and exogenous regime variables E_B) no independent evidence
    purpose: Reformulate do-calculus rules 2&3 via an auxiliary system with intervention variables (Definition E.1), avoiding the iSCM formalism.
    Formal bookkeeping device; each R_v carries either a prescribed path or the '⋆' idle marker. It has no physical referent and no falsifiable handle outside the proof.
  • Time-split variables X^I_v for a partition interval I no independent evidence
    purpose: Define causal relations between processes restricted to disjoint time intervals, enabling subsampling, Granger and local-independence statements (Definition 8).
    Deterministic functions of the original processes (restriction and concatenation); not new empirical entities.

pith-pipeline@v1.3.0-alltime-deepseek · 62450 in / 37500 out tokens · 333435 ms · 2026-08-02T06:42:48.483423+00:00 · methodology

0 comments
read the original abstract

Stochastic differential equations (SDEs) are widely used to model continuous-time dynamical systems, but graphical causal models for them are not yet well-understood. We consider systems of causal SDEs that are equipped with an explicit causal semantics. We pose solvability conditions for systems of causal SDEs such that they have well-defined observational and interventional distributions - even after marginalisation - and provide a general class of Lipschitz semimartingale SDEs that satisfies these conditions. As core results we establish the $\sigma$-separation Markov property and the do-calculus in terms of the system's causal graph for probabilistic independence and interventions on the level of sample paths. For a class of additive-noise SDEs we prove a stronger $d$-separation Markov property, even if the system is cyclic. As a corollary of the do-calculus, we obtain an explicit causal interpretation of the graph: that the absence of a directed path implies the absence of a causal effect. We further introduce time-split systems, which consider the causal relations between the processes when evaluated on disjoint intervals or time-points, and use them to reason about subsampled time-series, continuous-time Granger non-causality and local independence. Finally, we discuss how constraint-based causal discovery algorithms (PC, FCI, CCD, CCI) apply directly to SDEs within our framework when conditional independence between sample paths can be consistently tested.

Figures

Figures reproduced from arXiv: 2607.12140 by Joris M. Mooij, Philip Boeken.

Figure 1
Figure 1. Figure 1: The repressilator of the E. coli bacteria, modelled with additive noise. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The effects of interventions on the ‘lacI’ gene of the repressilator. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The marginalisation of the repressilator onto [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The causal graph G(D¯), the time-split causal graph G(D¯E ) with E = {{0},(0, t), {t},(t, T), {T}}, and the subsampled causal graph G(D¯{0,t,T} ) (see Definition 9 below) [PITH_FULL_IMAGE:figures/full_fig_p018_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Repressilator with gene ‘lacI’ knocked out in the first half of the simulation. [PITH_FULL_IMAGE:figures/full_fig_p018_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Outputs of the constraint-based causal discovery algorithms CCI and FCI applied to [PITH_FULL_IMAGE:figures/full_fig_p026_6.png] view at source ↗

Review history (2 revisions) →

discussion (0)

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

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Orca: Neural Operators for Causal Reasoning in Continuous Time

    cs.AI 2026-07 conditional novelty 6.0

    Orca extends structural causal models to continuous time with neural operators, enabling resolution-invariant dose-response and counterfactual trajectories on irregularly sampled cyclic systems.

  2. DoTime: A Synthetic Benchmark Generator for Interventional and Counterfactual Time Series

    cs.LG 2026-07 conditional novelty 6.0

    DoTime generates temporal structural causal models with interventions and counterfactuals, ships four frozen benchmark suites, and reports that interventional pretraining improves causal direction accuracy over observ...

Reference graph

Works this paper leans on

89 extracted references · cited by 2 Pith papers

  1. [1]

    Aalen, O. (1978). Nonparametric Inference for a Family of Counting Processes . The Annals of Statistics , 6(4):701--726

    1978

  2. [2]

    Aalen, O., R ysland, K., Gran, J., Kouyos, R., and Lange, T. (2016). Can we believe the DAGs ? A comment on the relationship between causal DAGs and mechanisms. Statistical Methods in Medical Research , 25(5):2294--2314

    2016

  3. [3]

    O., Borgan, ., and Gjessing, H

    Aalen, O. O., Borgan, ., and Gjessing, H. K. (2008). Survival and Event History Analysis: A Process Point of View . Statistics for Biology and Health. Springer, New York

    2008

  4. [4]

    O., R ysland, K., Gran, J

    Aalen, O. O., R ysland, K., Gran, J. M., and Ledergerber, B. (2012). Causality, mediation and time: A dynamic viewpoint. Journal of the Royal Statistical Society: Series A (Statistics in Society) , 175(4):831--861

    2012

  5. [5]

    Andrews, B., Spirtes, P., and Cooper, G. F. (2020). On the Completeness of Causal Discovery in the Presence of Latent Confounding with Tiered Background Knowledge . In Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics , pages 4002--4011. PMLR

    2020

  6. [6]

    Assaad, C., Devijver, E., and Gaussier, E. (2022). Survey and Evaluation of Causal Discovery Methods for Time Series . Journal of Artificial Intelligence Research , 73:767--819

    2022

  7. [7]

    and Mooij, J

    Blom, T. and Mooij, J. (2023). Causality and independence in perfectly adapted dynamical systems. Journal of Causal Inference , 11(1)

    2023

  8. [8]

    and Mooij, J

    Boeken, P. and Mooij, J. M. (2024). Dynamic Structural Causal Models

    2024

  9. [9]

    Boeken, P., Skapinakis, E., Genin, K., and Mooij, J. M. (2026). Topological Criteria for Hypothesis Testing with Finite-Precision Measurements

    2026

  10. [10]

    Bongers, S., Blom, T., and Mooij, J. (2022). Causal Modeling of Dynamical Systems

    2022

  11. [11]

    Bongers, S., Forr \'e , P., Peters, J., and Mooij, J. (2021). Foundations of structural causal models with cycles and latent variables. The Annals of Statistics , 49(5)

    2021

  12. [12]

    Bowsher, C. G. (2010). Stochastic kinetic models: Dynamic independence, modularity and graphs. The Annals of Statistics , 38(4):2242--2281

    2010

  13. [13]

    and Yor, M

    Br \'e maud, P. and Yor, M. (1978). Changes of filtrations and of probability measures. Zeitschrift f \"u r Wahrscheinlichkeitstheorie und Verwandte Gebiete , 45(4):269--295

    1978

  14. [14]

    Br \"u ck, F., Engelke, S., and Volgushev, S. (2026). Graph structure learning for stable processes

    2026

  15. [15]

    Christgau, A., Petersen, L., and Hansen, N. (2023). Nonparametric conditional local independence testing. The Annals of Statistics , 51(5)

    2023

  16. [16]

    and Renault, E

    Comte, F. and Renault, E. (1996). Noncausality in Continuous Time Models . Econometric Theory , 12(2):215--256

    1996

  17. [17]

    Cooper, G. (1997). A Simple Constraint-Based Algorithm for Efficiently Mining Observational Databases for Causal Relationships . Data Mining and Knowledge Discovery

    1997

  18. [18]

    Dawid, P. (2021). Decision-theoretic foundations for statistical causality. Journal of Causal Inference , 9(1):39--77

    2021

  19. [19]

    and Meyer, P

    Dellacherie, C. and Meyer, P. (1978). Probabilities and Potential . Number 29 in North- Holland Mathematics Studies. Hermann [u.a.], Paris

    1978

  20. [20]

    Didelez, V. (2008). Graphical Models for Marked Point Processes Based on Local Independence . Journal of the Royal Statistical Society. Series B (Statistical Methodology) , 70(1):245--264

    2008

  21. [21]

    Eichler, M. (2012). Graphical modelling of multivariate time series. Probability Theory and Related Fields , 153(1):233--268

    2012

  22. [22]

    and Didelez, V

    Eichler, M. and Didelez, V. (2010). On Granger causality and the effect of interventions in time series. Lifetime Data Analysis , 16(1):3--32

    2010

  23. [23]

    Elowitz, M. B. and Leibler, S. (2000). A synthetic oscillatory network of transcriptional regulators. Nature , 403(6767):335--338

    2000

  24. [24]

    Engelke, S., Ivanovs, J., and Th stesen, J. (2024). L 'evy graphical models

    2024

  25. [25]

    and Assaad, C

    Ferreira, S. and Assaad, C. (2024). Identifying macro conditional independencies and macro total effects in summary causal graphs with latent confounding

    2024

  26. [26]

    and Fougere, D

    Florens, J. and Fougere, D. (1996). Noncausality in Continuous Time . Econometrica , 64(5):1195--1212

    1996

  27. [27]

    and Protter, P

    F \"o llmer, H. and Protter, P. (2011). Local martingales and filtration shrinkage. ESAIM: Probability and Statistics , 15:S25--S38

    2011

  28. [28]

    and Mooij, J

    Forr \'e , P. and Mooij, J. (2017). Markov Properties for Graphical Models with Cycles and Latent Variables

    2017

  29. [29]

    and Mooij, J

    Forr \'e , P. and Mooij, J. (2020). Causal Calculus in the Presence of Cycles , Latent Confounders and Selection Bias . In PMLR , pages 71--80. PMLR

    2020

  30. [30]

    and Mooij, J

    Forr \'e , P. and Mooij, J. (2025). A Mathematical Introduction to Causality

    2025

  31. [31]

    and Robins, J

    Gill, R. and Robins, J. (2001). Causal Inference for Complex Longitudinal Data : The Continuous Case . The Annals of Statistics , 29(6):1785--1811

    2001

  32. [32]

    Granger, C. (1969). Investigating Causal Relations by Econometric Models and Cross-spectral Methods . Econometrica , 37(3):424--438

    1969

  33. [33]

    Granger, C. (1980). Testing for causality: A personal viewpoint. Journal of Economic Dynamics and Control , 2:329--352

    1980

  34. [34]

    Guan, V., Janssen, J., Rahmani, H., Warren, A., Zhang, S., Robeva, E., and Schiebinger, G. (2024). Identifying Drift , Diffusion , and Causal Structure from Temporal Snapshots

    2024

  35. [35]

    and Sokol, A

    Hansen, N. and Sokol, A. (2014). Causal interpretation of stochastic differential equations. Electronic Journal of Probability , 19(none):1--24

    2014

  36. [36]

    and Shiryaev, A

    Jacod, J. and Shiryaev, A. (2003). Limit Theorems for Stochastic Processes , volume 288 of Grundlehren Der Mathematischen Wissenschaften . Springer Berlin Heidelberg, Berlin, Heidelberg

    2003

  37. [37]

    Kallenberg, O. (1996). On the existence of universal functional solutions to classical SDE 's. The Annals of Probability , 24(1)

    1996

  38. [38]

    Kallenberg, O. (2021). Foundations of Modern Probability , volume 99 of Probability Theory and Stochastic Modelling . Springer International Publishing, Cham

    2021

  39. [39]

    Karandikar, R. L. (1995). On pathwise stochastic integration. Stochastic Processes and their Applications , 57(1):11--18

    1995

  40. [40]

    and Shreve, S

    Karatzas, I. and Shreve, S. (1988). Brownian Motion and Stochastic Calculus , volume 113 of Graduate Texts in Mathematics . Springer US, New York, NY

    1988

  41. [41]

    u gelgen , J., Park, J., Sch \

    Laumann, F., von K \"u gelgen , J., Park, J., Sch \"o lkopf, B., and Barahona, M. (2023). Kernel- Based Independence Tests for Causal Structure Learning on Functional Data . Entropy , 25(12):1597

    2023

  42. [42]

    Lauritzen, S. (2024). Total variation convergence preserves conditional independence. Statistics & Probability Letters , 214:110200

    2024

  43. [43]

    and Shiryaev, A

    Liptser, R. and Shiryaev, A. (2001). Statistics of Random Processes: I : General Theory . Number 5 in Applications of Mathematics . Springer, Berlin, Germany ; Heidelberg, Germany

    2001

  44. [44]

    Lok, J. (2008). Statistical Modeling of Causal Effects in Continuous Time . The Annals of Statistics , 36(3):1464--1507

    2008

  45. [45]

    Lundborg, A., Shah, R., and Peters, J. (2022). Conditional Independence Testing in Hilbert Spaces with Applications to Functional Data Analysis . J. R. Stat. Soc. Ser. B Methodol. , 84(5):1821--1850

    2022

  46. [46]

    Lyons, T., Caruana, M., and L \'e vy, T. (2007). Differential Equations Driven by Rough Paths : \'E cole d' \'E t \'e de Probabilit \'e s de Saint-Flour XXXIV - 2004 , volume 1908 of Lecture Notes in Mathematics . Springer Berlin Heidelberg, Berlin, Heidelberg

    2007

  47. [47]

    and Spirtes, P

    Malinsky, D. and Spirtes, P. (2018). Causal Structure Learning from Multivariate Time Series in Settings with Unmeasured Confounding . In Proceedings of 2018 ACM SIGKDD Workshop on Causal Discovery , pages 23--47. PMLR

    2018

  48. [48]

    Mani, S. (2006). A Bayesian Local Causal Discovery Framework . University of Pittsburgh ETD , University of Pittsburgh

    2006

  49. [49]

    Manten, G., Casolo, C., Ferrucci, E., Mogensen, S., Salvi, C., and Kilbertus, N. (2024). Signature Kernel Conditional Independence Tests in Causal Discovery for Stochastic Processes

    2024

  50. [50]

    Manten, G., Casolo, C., Mogensen, S., and Kilbertus, N. (2025). An Asymmetric Independence Model for Causal Discovery on Path Spaces

    2025

  51. [51]

    and Hansen, N

    Mogensen, S. and Hansen, N. (2022). Graphical modeling of stochastic processes driven by correlated noise. Bernoulli , 28(4)

    2022

  52. [52]

    and Hansen, N

    Mogensen, S. and Hansen, N. R. (2020). Markov equivalence of marginalized local independence graphs. The Annals of Statistics , 48(1)

    2020

  53. [53]

    Mogensen, S., Malinsky, D., and Hansen, N. (2018). Causal Learning for Partially Observed Stochastic Dynamical Systems

    2018

  54. [54]

    and Claassen, T

    Mooij, J. and Claassen, T. (2020). Constraint- Based Causal Discovery using Partial Ancestral Graphs in the presence of Cycles . In UAI2020 , pages 1159--1168. PMLR

    2020

  55. [55]

    Mooij, J., Janzing, D., and Sch \"o lkopf, B. (2013). From Ordinary Differential Equations to Structural Causal Models : The deterministic case

    2013

  56. [56]

    Mooij, J., Magliacane, S., and Claassen, T. (2020). Joint causal inference from multiple contexts. The Journal of Machine Learning Research , 21(1):99:3919--99:4026

    2020

  57. [57]

    Nathaniel, J., Roesch, C., Buch, J., DeSantis, D., Rupe, A., Lamb, K., and Gentine, P. (2025). Deep Koopman operator framework for causal discovery in nonlinear dynamical systems

    2025

  58. [58]

    Neal, R. M. (2000). On Deducing Conditional Independence from d- Separation in Causal Graphs with Feedback ( Research Note ). Journal of Artificial Intelligence Research , 12:87--91

    2000

  59. [59]

    Niemiro, W. (2024). Causal graphs, composable stochastic processes and conditional independence. Applicationes Mathematicae , pages 1--22

    2024

  60. [60]

    Pearl, J. (1993). Comment: Graphical Models , Causality and Intervention . Statistical Science , 8(3):266--269

    1993

  61. [61]

    Pearl, J. (1995). Causal Diagrams for Empirical Research . Biometrika , 82(4):669--688

    1995

  62. [62]

    Pearl, J. (2009). Causality . Cambridge University Press

    2009

  63. [63]

    and Dechter, R

    Pearl, J. and Dechter, R. (1996). Identifying independencies in causal graphs with feedback. In Proceedings of the Twelfth International Conference on Uncertainty in Artificial Intelligence , UAI '96, pages 420--426, San Francisco, CA, USA. Morgan Kaufmann Publishers Inc

    1996

  64. [64]

    Peters, J., Bauer, S., and Pfister, N. (2020). Causal models for dynamical systems

    2020

  65. [65]

    Peters, J., Janzing, D., and Sch \"o lkopf, B. (2013). Causal Inference on Time Series using Restricted Structural Equation Models . In Advances in Neural Information Processing Systems , volume 26. Curran Associates, Inc

    2013

  66. [66]

    Protter, P. (2005). Stochastic Integration and Differential Equations , volume 21 of Stochastic Modelling and Applied Probability . Springer Berlin Heidelberg, Berlin, Heidelberg

    2005

  67. [67]

    Przyby owicz, P., Schwarz, V., Steinicke, A., and Sz \"o lgyenyi, M. (2024). A Skorohod measurable universal functional representation of solutions to semimartingale SDEs . Stochastic Analysis and Applications , 42(6):1137--1155

    2024

  68. [68]

    G., Su \'a rez, A., Weichwald, S., and Chambaz, A

    Reisach, A. G., Su \'a rez, A., Weichwald, S., and Chambaz, A. (2025). The Case for Time in Causal DAGs

    2025

  69. [69]

    Reiter, N., Gerhardus, A., Wahl, J., and Runge, J. (2024). Causal Inference on Process Graphs , Part I : The Structural Equation Process Representation

    2024

  70. [70]

    Richardson, T. (1996). A discovery algorithm for directed cyclic graphs. In Proceedings of the Twelfth International Conference on Uncertainty in Artificial Intelligence , UAI '96, pages 454--461, San Francisco, CA, USA. Morgan Kaufmann Publishers Inc

    1996

  71. [71]

    R ysland, K., Ryalen, P., Nyg rd, M., and Didelez, V. (2024). Graphical criteria for the identification of marginal causal effects in continuous-time survival and event-history analyses

    2024

  72. [72]

    Rubenstein, P., Bongers, S., Sch \"o lkopf, B., and Mooij, J. (2018). From Deterministic ODEs to Dynamic Structural Causal Models . In Conference on Uncertainty in Artificial Intelligence

    2018

  73. [73]

    Runge, J., Nowack, P., Kretschmer, M., Flaxman, S., and Sejdinovic, D. (2019). Detecting and quantifying causal associations in large nonlinear time series datasets. Science Advances , 5(11):eaau4996

    2019

  74. [74]

    C., Stensrud, M

    Ryalen, P. C., Stensrud, M. J., and R ysland, K. (2026). On causal inference with marked point process data

    2026

  75. [75]

    Rytgaard, H., Gerds, T., and Van Der Laan, M. (2022). Continuous-time targeted minimum loss-based estimation of intervention-specific mean outcomes. The Annals of Statistics , 50(5)

    2022

  76. [76]

    Scheff\'e, H. (1947). A Useful Convergence Theorem for Probability Distributions . The Annals of Mathematical Statistics , 18(3):434--438

    1947

  77. [77]

    and Drton, M

    Schwank, R. and Drton, M. (2026). Non-parametric recovery of causal diffusion mechanisms from steady-state observations

    2026

  78. [78]

    Schweder, T. (1970). Composable Markov Processes . Journal of Applied Probability , 7(2):400--410

    1970

  79. [79]

    Skorokhod, A. (1956). Limit Theorems for Stochastic Processes . Theory of Probability & Its Applications , 1(3):261--290

    1956

  80. [80]

    Spirtes, P. (1994). Conditional independence in directed cyclic graphical models for feedback. Technical Report CMU-PHIL-54, Carnegie Mellon University

    1994

Showing first 80 references.