REVIEW 3 major objections 3 minor 47 references
Persistent path homology is stable: a factor-two bound between path complex distance and bottleneck distance.
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 →
T0 review · deepseek-v4-flash
2026-08-01 00:26 UTC pith:37LXDZNW
load-bearing objection Clean framework, but the main stability theorem for general path complexes has a load-bearing gap in the homotopy argument. the 3 major comments →
Stability of persistent path homology of path complexes
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The central claim is Theorem 3.19: for filtered path complexes P and Q, in every homological degree k, the bottleneck distance between their degree-k persistence diagrams is at most twice the path complex distance between P and Q. The path complex distance is half the infimum, over all vertex correspondences, of the maximum difference between the entry times of corresponding paths. The proof shows that any correspondence whose path-distortion is less than epsilon is an epsilon-path-correspondence, which induces an epsilon-interleaving of the persistence modules, and algebraic stability converts that interleaving into the bottleneck bound.
What carries the argument
The construction that carries the argument is the covering framework: a weight on the generating cells of an object (paths, hyperedges, ordered subsequences, directed edges) is extended to all cells by declaring the weight of a cell to be the cheapest total weight of a covering family, i.e., the algebraic path problem in the min-plus semiring. Lemma 2.3 proves the extension is idempotent, so the sublevel sets form a filtration that is already complete. Instantiating this framework for path complexes, hypergraphs, sequence hypergraphs, and digraphs reduces each stability statement to Theorem 3.19 via a distance comparison.
Load-bearing premise
The proof of the key step (Proposition 3.15) assumes that the target path complex is closed under concatenation of allowed paths along a shared vertex and contains the one-vertex path (x,x) for every vertex; the definition of a path complex guarantees neither, and the paper's own Remark 3.7 notes that path complexes need not be closed under concatenation.
What would settle it
Build two filtered path complexes that are within epsilon in path complex distance but whose persistence diagrams are farther than 2epsilon apart, or, more locally, find a filtered path complex violating the concatenation property and two subordinate maps of an epsilon-path-correspondence that are not homotopic, which would break the interleaving construction. The paper's own example in Remark 3.7 is a natural starting point.
If this is right
- If two weighted hypergraphs are close in hypergraph distance, their persistent path homology persistence diagrams are close with factor 2 (Theorem 4.12).
- The same statement holds for sequence hypergraphs (Theorem 5.10) and for digraphs (Theorem 6.9), recovering the known digraph stability result.
- Any weighted object that admits a covering relation and a functor to path complexes inherits stability automatically, provided a distance comparison can be proved.
- Because the induced weight is idempotent, the resulting filtrations are unchanged by repeating the completion process; no iterative refinement alters them.
Where Pith is reading between the lines
- If the concatenation-closure gap in the homotopy argument (Proposition 3.15) is real, the theorem as stated may fail for arbitrary filtered path complexes; however, the applications use generated filtrations whose sublevel path complexes are closed under concatenation, so the gap may not affect the practical cases. Testing whether Theorem 3.19 holds for all filtered path complexes or only for gene
- The covering construction is the metric completion of a weighted structure in the enriched-category sense, so the same idempotent extension could plausibly yield stability for other hypergraph homology theories or bring Vietoris–Rips filtrations under the same umbrella.
- Since the distance equality d_path = d_N holds for digraphs but not for hypergraphs in general, the factor of 2 is likely not optimal for hypergraphs; sharper bounds may exist for specific densities.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper introduces a covering framework in which a weight on the generating cells of a combinatorial object is extended to all cells, and uses it to prove a stability theorem for persistent path homology of filtered path complexes (Theorem 3.19). From this theorem it derives stability for hypergraphs (Theorem 4.12), sequence hypergraphs (Theorem 5.10), and digraphs (Theorem 6.9), the last recovering a result of Chowdhury and Mémoli. The proof of Theorem 3.19 follows the standard correspondence/interleaving template: an ε-path-correspondence between filtered path complexes yields an ε-interleaving of their path homology persistence modules, and algebraic stability gives the bottleneck bound.
Significance. If Theorem 3.19 were established, the covering framework would be a genuinely useful unification: Lemma 2.3 is clean and correct, the distance comparisons in Sections 4–6 are transparent, and proving stability once for path complexes would indeed cover several widely used objects. The paper is also candid about an important structural fact, Remark 3.7, namely that path complexes need not be closed under concatenation. However, that very remark exposes a fatal gap in the central proof: Proposition 3.15 is false for arbitrary path complexes, so the main theorem is not established. The significance of the paper is therefore conditional on a repair that, as explained below, cannot be achieved by a local correction within the current definitions.
major comments (3)
- [§3, Lemma 3.14] The proof assumes p=(x,x)∈P^δ. Definition 1.1 only guarantees the length-0 path (x), and Definition 3.1 likewise ensures (x)∈P^0. No axiom forces (x,x) into P^δ; the example in Remark 3.7 is a filtered path complex with no loops. Consequently, applying ε-path-multivaluedness to (x) only gives (y)∈Q^{δ+ε}, not (y1,y2)∈Q^{δ+ε}. Concretely, take P^δ={x} and Q^{δ+ε}={a,b} with no 1-paths. Then C(x)={a,b} is ε-path-multivalued in degree 0, but Lemma 3.14 would conclude (a,b)∈Q^{δ+ε}, which is false.
- [§3, Proposition 3.15] The one-step homotopy α is not well-defined for general filtered path complexes. A mixed cylinder path is mapped to the concatenation (f1(x0),...,f1(xk),f2(xk),...,f2(xm)). To show this is an allowed path of Q^{δ+ε}, the proof uses Lemma 3.14 for the transition edge and then says the ε-path-multivalued property 'places the whole image' in Q^{δ+ε}. But that property applies only to paths of P^δ, and the duplicated sequence (x0,...,xk,xk,...,xm) is not generally an allowed path of P^δ. Moreover, even if the two segments and the transition edge are each allowed, their concatenation need not be allowed because path complexes are not closed under concatenation, as Remark 3.7 explicitly states. The statement is also false: with P^δ={x}, Q^{δ+ε}={a,b}, and C={(x,a),(x,b)}, both subordinate maps f1(x)=a and f2(x)=b are path-complex maps, but they induce distinct maps on H0, so no canonical HPath
- [§3, Theorem 3.19 and Corollary 3.20] Since Proposition 3.15 is false, Proposition 3.18 and Theorem 3.19 are unsupported. Corollary 3.20 is the bridge used by all later sections, so the hypergraph and sequence-hypergraph results in Sections 4 and 5 inherit the same obstruction: the associated path complexes need not contain loops or be closed under concatenation. Even if the digraph case in Section 6 can be salvaged, because digraph path complexes built from reflexive relations do contain loops and are concatenation-closed, the general stability theorem for path complexes—and therefore the claimed stability for hypergraphs and sequence hypergraphs—does not follow from the arguments given.
minor comments (3)
- [§3, Definition 3.1] The symbol P^0 is used both for the stage-0 path complex and for the common vertex set. This is confusing; a separate notation such as V_P would clarify the statement and proofs.
- [§3, Definition 3.11] The path distortion is defined as a supremum over extended real differences. If both l_P(π1(σ)) and l_Q(π2(σ)) are ∞, the expression |∞−∞| is undefined. A convention should be stated, or the distance should be restricted to paths with finite entry weight.
- [§4 and §5] The phrase 'when no confusion arises, we denote the generated filtered path complex again by P' is used repeatedly. It would be helpful to make the completion notation G↦G uniform and to state explicitly in each section which objects are completed before stability is applied.
Circularity Check
No circularity: the stability theorem is derived from first principles; the main proof gap is a missing hypothesis, not a circular reduction.
full rationale
The derivation chain is not circular. Theorem 3.19 is proved by the standard CdSO14/BM24 route: a correspondence with small path distortion is shown to be an eta-path-correspondence, Proposition 3.18 turns it into an eta-interleaving, and algebraic stability converts the interleaving into the bottleneck bound. The path-complex distance is defined from entry weights before homology is computed; the bottleneck distance appears only as the target inequality, so the conclusion is not built into the input. The covering framework of Section 2 is an idempotent closure operator proved from the stated axioms (Lemma 2.3), and Sections 4-6 are genuine deductions from Theorem 3.19 via Corollary 3.20 plus explicit distance comparisons (Propositions 4.11, 5.9, 6.8). Self-citations ([CDK+24], [CK24]) are background references and are not load-bearing. There is, however, a serious correctness gap that is not circularity. In Proposition 3.15 the homotopy check uses Lemma 3.14, whose proof assumes "Let p=(x,x) in P^delta," while Definition 3.1 only guarantees length-0 paths in P^0; and the mixed cylinder image requires closure under concatenation, which Remark 3.7 explicitly denies: "a path complex need not be closed under concatenation." This makes Theorem 3.19 unproved as stated for arbitrary filtered path complexes, but it is a missing hypothesis in a proof, not a reduction of the theorem to its own inputs.
Axiom & Free-Parameter Ledger
axioms (5)
- ad hoc to paper Path complexes are closed under concatenation of allowed paths along a shared vertex.
- ad hoc to paper The 1-path (x,x) is allowed in every path complex for every vertex x.
- standard math The isometry theorem between bottleneck and interleaving distances holds for pointwise finite-dimensional persistence modules.
- standard math Algebraic stability: an ε-interleaving implies d_B ≤ ε.
- standard math Pointwise finite-dimensional persistence modules decompose into interval modules.
read the original abstract
We show stability of persistent path homology of path complexes. As a consequence, we deduce the stability of persistent path homology of hypergraphs and of sequence hypergraphs, and recover the known stability result for digraphs, originally due to Chowdhury and M\'emoli.
Reference graph
Works this paper leans on
-
[1]
Yasuhiko Asao, Magnitude homology and path homology, Bull. Lond. Math. Soc. 55 (2023), no. 1, 375--398
2023
-
[2]
Eric Babson, H\' e l\`ene Barcelo, Mark de Longueville, and Reinhard Laubenbacher, Homotopy theory of graphs, J. Algebraic Combin. 24 (2006), no. 1, 31--44, http://dx.doi.org/10.1007/s10801-006-9100-0 doi:10.1007/s10801-006-9100-0 , https://doi.org/10.1007/s10801-006-9100-0
-
[3]
Kepple, Rui Qi, Zhouchun Shang, Yanan Xing, Yanru An, Nannan Zhang, Yong Hou, Tanya L
Katherine Benjamin, Aneesha Bhandari, Jessica D. Kepple, Rui Qi, Zhouchun Shang, Yanan Xing, Yanru An, Nannan Zhang, Yong Hou, Tanya L. Crockford, Oliver McCallion, Fadi Issa, Joanna Hester, Ulrike Tillmann, Heather A. Harrington, and Katherine R. Bull, Multiscale topology classifies cells in subcellular spatial transcriptomics, Nature 630 (2024), no. 801...
2024
-
[4]
Fran c ois Baccelli, Guy Cohen, Geert Jan Olsder, and Jean-Pierre Quadrat, Synchronization and linearity: An algebra for discrete event systems, Wiley, 1992, Introduction to the min-plus (tropical) semiring and shortest-path closure
1992
-
[5]
White, Discrete homology theory for metric spaces, Bull
H\' e l\`ene Barcelo, Valerio Capraro, and Jacob A. White, Discrete homology theory for metric spaces, Bull. Lond. Math. Soc. 46 (2014), no. 5, 889--905, http://dx.doi.org/10.1112/blms/bdu043 doi:10.1112/blms/bdu043 , https://doi.org/10.1112/blms/bdu043
-
[6]
484--490
Ulrich Bauer and Herbert Edelsbrunner, The M orse theory of cech and D elaunay filtrations , Computational geometry ( S o CG '14), ACM, New York, 2014, pp. 484--490
2014
-
[7]
H\' e l\`ene Barcelo, Curtis Greene, Abdul Salam Jarrah, and Volkmar Welker, Discrete cubical and path homologies of graphs, Algebr. Comb. 2 (2019), no. 3, 417--437, http://dx.doi.org/10.5802/alco.49 doi:10.5802/alco.49 , https://doi.org/10.5802/alco.49
doi:10.5802/alco.49 2019
-
[8]
4, 125503
Ulrich Bauer, Michael Kerber, Fabian Roll, and Alexander Rolle, A unified view on the functorial nerve theorem and its variations, Expositiones Mathematicae 41 (2023), no. 4, 125503
2023
-
[9]
3, 479--500
Stephane Bressan, Jingyan Li, Shiquan Ren, and Jie Wu, The embedded homology of hypergraphs and applications, Asian Journal of Mathematics 23 (2019), no. 3, 479--500
2019
-
[10]
Peter Bubenik and Nikola Mili\' c evi\' c , Homotopy, homology, and persistent homology using closure spaces, J. Appl. Comput. Topol. 8 (2024), no. 3, 579--641, http://dx.doi.org/10.1007/s41468-024-00183-8 doi:10.1007/s41468-024-00183-8 , https://doi.org/10.1007/s41468-024-00183-8
-
[11]
Gunnar Carlsson, Topology and data, Bull. Amer. Math. Soc. (N.S.) 46 (2009), no. 2, 255--308, http://dx.doi.org/10.1090/S0273-0979-09-01249-X doi:10.1090/S0273-0979-09-01249-X , https://doi.org/10.1090/S0273-0979-09-01249-X
-
[12]
5, 1550066
William Crawley-Boevey, Decomposition of pointwise finite-dimensional persistence modules, Journal of Algebra and its Applications 14 (2015), no. 5, 1550066
2015
-
[13]
2, 475--514
Daniel Carranza, Brandon Doherty, Chris Kapulkin, Morgan Opie, Maru Sarazola, and Liang Ze Wong, Cofibration category of digraphs for path homology, Algebraic Combinatorics 7 (2024), no. 2, 475--514
2024
-
[14]
Fr \'e d \'e ric Chazal, Vin de Silva, Marc Glisse, and Steve Oudot, The structure and stability of persistence modules, SpringerBriefs in Mathematics, Springer, 2016
2016
-
[15]
1, 193--214
Fr \'e d \'e ric Chazal, Vin de Silva, and Steve Oudot, Persistence stability for geometric complexes, Geometriae Dedicata 173 (2014), no. 1, 193--214
2014
-
[16]
1077--1082
Samir Chowdhury, Thomas Gebhart, Steve Huntsman, and Matvey Yutin, Path homologies of deep feedforward networks, 18th IEEE International Conference on Machine Learning and Applications (ICMLA), 2019, pp. 1077--1082
2019
-
[17]
Thomas Chaplin, Heather A. Harrington, and Ulrike Tillmann, Grounded persistent path homology: a stable, topological descriptor for weighted digraphs, Foundations of Computational Mathematics 25 (2025), 1711--1776
2025
-
[18]
Samir Chowdhury, Steve Huntsman, and Matvey Yutin, Path homologies of motifs and temporal network representations, Applied Network Science 7 (2022), no. 4
2022
-
[19]
Daniel Carranza and Krzysztof Kapulkin, Cubical setting for discrete homotopy theory, revisited, Compos. Math. 160 (2024), no. 12, 2856--2903, http://dx.doi.org/10.1112/S0010437X24007486 doi:10.1112/S0010437X24007486 , https://doi.org/10.1112/S0010437X24007486
-
[20]
1--2, 115--175
Samir Chowdhury and Facundo M \'e moli, A functorial D owker theorem and persistent homology of asymmetric networks , Journal of Applied and Computational Topology 2 (2018), no. 1--2, 115--175
2018
-
[21]
1152--1169, http://dx.doi.org/10.1137/1.9781611975031.75 doi:10.1137/1.9781611975031.75
Samir Chowdhury and Facundo M\'emoli, Persistent path homology of directed networks, Proceedings of the Twenty-Ninth Annual ACM-SIAM Symposium on Discrete Algorithms (SODA), SIAM, 2018, pp. 1152--1169, http://dx.doi.org/10.1137/1.9781611975031.75 doi:10.1137/1.9781611975031.75
-
[22]
2, 243--361
Samir Chowdhury and Facundo M \'e moli, Distances and isomorphism between networks: stability and convergence of network invariants, Journal of Applied and Computational Topology 7 (2023), no. 2, 243--361
2023
-
[23]
Robert J. MacG. Dawson, Homology of weighted simplicial complexes, Cahiers de Topologie et G\'eom\'etrie Diff\'erentielle Cat\'egoriques 31 (1990), no. 3, 229--243
1990
-
[24]
Shaobo Di, Sergei O. Ivanov, Lev Mukoseev, and Mengmeng Zhang, On the path homology of C ayley digraphs and covering digraphs , arXiv preprint arXiv:2305.15683 (2023)
Pith/arXiv arXiv 2023
-
[25]
Dey, Tianqi Li, and Yusu Wang, An efficient algorithm for 1-dimensional (persistent) path homology, Discrete & Computational Geometry 68 (2022), no
Tamal K. Dey, Tianqi Li, and Yusu Wang, An efficient algorithm for 1-dimensional (persistent) path homology, Discrete & Computational Geometry 68 (2022), no. 4, 1102--1132
2022
-
[26]
Herbert Edelsbrunner, David Letscher, and Afra Zomorodian, Topological persistence and simplification, Discrete Comput. Geom. 28 (2002), no. 4, 511--533, Discrete and computational geometry and graph drawing (Columbia, SC, 2001), http://dx.doi.org/10.1007/s00454-002-2885-2 doi:10.1007/s00454-002-2885-2 , https://doi.org/10.1007/s00454-002-2885-2
-
[27]
Xin Fu and Sergei O. Ivanov, Path homology of digraphs without multisquares and its comparison with homology of spaces, arXiv preprint arXiv:2407.17001 (2024)
Pith/arXiv arXiv 2024
-
[28]
Gardner, Erik Hermansen, Marius Pachitariu, Yoram Burak, Nils A
Richard J. Gardner, Erik Hermansen, Marius Pachitariu, Yoram Burak, Nils A. Baas, Benjamin A. Dunn, May-Britt Moser, and Edvard I. Moser, Toroidal topology of population activity in grid cells, Nature 602 (2022), no. 7895, 123--128
2022
-
[29]
20 (2018), no
Alexander Grigor'yan, Rolando Jimenez, Yuri Muranov, and Shing-Tung Yau, On the path homology theory of digraphs and E ilenberg-- S teenrod axioms , Homology Homotopy Appl. 20 (2018), no. 2, 179--205
2018
-
[30]
, Homology of path complexes and hypergraphs, Topology and its Applications 267 (2019), 106877
2019
-
[31]
Alexander Grigor'yan, Yong Lin, Yuri Muranov, and Shing-Tung Yau, Homologies of path complexes and digraphs, arXiv preprint arXiv:1207.2834 (2012)
Pith/arXiv arXiv 2012
-
[32]
, Homotopy theory for digraphs, Pure Appl. Math. Q. 10 (2014), no. 4, 619--674, http://dx.doi.org/10.4310/PAMQ.2014.v10.n4.a2 doi:10.4310/PAMQ.2014.v10.n4.a2 , https://doi.org/10.4310/PAMQ.2014.v10.n4.a2
-
[33]
5, 564--599
, Path complexes and their homologies, Journal of Mathematical Sciences 248 (2020), no. 5, 564--599
2020
-
[34]
16 (2014), no
Alexander Grigor'yan, Yuri Muranov, and Shing-Tung Yau, Graphs associated with simplicial complexes, Homology Homotopy Appl. 16 (2014), no. 1, 295--311
2014
-
[35]
, Homologies of digraphs and K \"unneth formulas , Comm. Anal. Geom. 25 (2017), no. 5, 969--1018
2017
-
[36]
Ellen Gasparovic, Emilie Purvine, Radmila Sazdanovi \'c , Bei Wang, Yusu Wang, and Lori Ziegelmeier, A survey of simplicial, relative, and chain complex homology theories for hypergraphs, Journal of Applied and Computational Topology 10 (2026), no. 7
2026
-
[37]
Richard Hepworth and Emily Roff, Bigraded path homology and the magnitude-path spectral sequence, arXiv preprint arXiv:2404.06689 (2024)
Pith/arXiv arXiv 2024
-
[38]
Richard Hepworth and Simon Willerton, Categorifying the magnitude of a graph, Homology Homotopy Appl. 19 (2017), no. 2, 31--60, http://dx.doi.org/10.4310/HHA.2017.v19.n2.a3 doi:10.4310/HHA.2017.v19.n2.a3 , https://doi.org/10.4310/HHA.2017.v19.n2.a3
-
[39]
Johnstone, Sketches of an elephant: A topos theory compendium, Oxford Logic Guides, vol
Peter T. Johnstone, Sketches of an elephant: A topos theory compendium, Oxford Logic Guides, vol. 43--44, Oxford University Press, 2002
2002
-
[40]
William Lawvere, Metric spaces, generalized logic, and closed categories, Rendiconti del Seminario Matematico e Fisico di Milano 43 (1973), 135--166, Reprinted in Repr
F. William Lawvere, Metric spaces, generalized logic, and closed categories, Rendiconti del Seminario Matematico e Fisico di Milano 43 (1973), 135--166, Reprinted in Repr. Theory Appl. Categ. 1 (2002), 1--37
1973
-
[41]
3, 321--350
Mehryar Mohri, Semiring frameworks and algorithms for shortest-distance problems, Journal of Automata, Languages and Combinatorics 7 (2002), no. 3, 321--350
2002
-
[42]
M. E. J. Newman, The structure and function of complex networks, SIAM Review 45 (2003), no. 2, 167--256
2003
-
[43]
Porter, Ulrike Tillmann, Peter Grindrod, and Heather A
Nina Otter, Mason A. Porter, Ulrike Tillmann, Peter Grindrod, and Heather A. Harrington, A roadmap for the computation of persistent homology, EPJ Data Science 6 (2017), no. 1, 17, http://dx.doi.org/10.1140/epjds/s13688-017-0109-5 doi:10.1140/epjds/s13688-017-0109-5 , https://doi.org/10.1140/epjds/s13688-017-0109-5
-
[44]
8, 2661--2687
Shiquan Ren, Chengyuan Wu, and Jie Wu, Weighted persistent homology, Rocky Mountain Journal of Mathematics 48 (2018), no. 8, 2661--2687
2018
-
[45]
Yuhei Umeda, Time series classification via topological data analysis, Information and Media Technologies 12 (2017), 228--239
2017
-
[46]
Afra Zomorodian and Gunnar Carlsson, Computing persistent homology, Discrete Comput. Geom. 33 (2005), no. 2, 249--274, http://dx.doi.org/10.1007/s00454-004-1146-y doi:10.1007/s00454-004-1146-y , https://doi.org/10.1007/s00454-004-1146-y
-
[47]
Shen Zhang, Stability of persistent path diagrams, arXiv preprint arXiv:2406.11998 (2024)
Pith/arXiv arXiv 2024
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.