pith. sign in

arxiv: 1907.08260 · v2 · pith:75OVMJ4Pnew · submitted 2019-07-18 · ⚛️ physics.data-an · stat.AP· stat.CO

A geometric approach to the transport of discontinuous densities

Caroline Moosm\"uller , Felix Dietrich , Ioannis G. Kevrekidis This is my paper

Pith reviewed 2026-05-24 19:36 UTC · model grok-4.3

classification ⚛️ physics.data-an stat.APstat.CO
keywords density transportmanifold reconstructionoptimal transportattractor reconstructiondynamical systemsuncertainty quantificationdiscontinuous densities
0
0 comments X

The pith

Short histories of observations recover the underlying manifold that determines the correct transport between source and target densities

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

The paper claims that relations between input and output distributions, reported as histograms, can be clarified by recovering a hidden smooth manifold of which the sources and targets are partial observations. When direct mappings are not bijective because of folds or marginalizations that create density singularities, standard transport methods become ambiguous. Incorporating short histories of the observation process, drawing on attractor reconstruction methods, allows the manifold to be reconstructed and thus fixes the transport map. A reader would care because this modifies how one identifies the underlying system in uncertainty quantification once distributions are known.

Core claim

We hypothesize that there exists a smooth manifold underlying the relation; the sources and the targets are then partial observations (possibly projections) of this manifold. Knowledge of such a manifold implies knowledge of the relation, and thus of the right transport between source and target observations. When the source-target observations are not bijective, recovery of the manifold is obscured. Using ideas from attractor reconstruction in dynamical systems, additional information in the form of short histories of an observation process can help recover the underlying manifold.

What carries the argument

Short histories of the observation process, used via attractor reconstruction ideas to embed partial observations and recover the hidden manifold that carries the source-target relation.

If this is right

  • The transport map is fixed once the manifold is recovered.
  • This resolves ambiguities arising from density singularities due to folds over the observation spaces.
  • The approach extends optimal transport that uses only the density observations themselves.
  • Limitations remain in cases where the manifold cannot be fully reconstructed from the available histories.

Where Pith is reading between the lines

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

  • This suggests trying the method on time-series data from known dynamical systems where the manifold is already understood.
  • It may connect to delay-embedding techniques already used in nonlinear time-series analysis for state-space reconstruction.
  • Testing whether the recovered manifold improves downstream predictions in uncertainty quantification tasks would be a direct next check.

Load-bearing premise

A smooth manifold exists underneath the source-target observations.

What would settle it

A concrete example with a known folded manifold where adding short histories still leaves multiple possible transports between the observed densities.

Figures

Figures reproduced from arXiv: 1907.08260 by Caroline Moosm\"uller, Felix Dietrich, Ioannis G. Kevrekidis.

Figure 1
Figure 1. Figure 1: The uniform density fµ(x) on [0, 1] is transported by a “folded”, i.e. non-injective trans￾port (black, y = T(x) = −2(1 − x) 3 + 1.5(1 − x) + 0.5) to a density fν(y) which has a jump discontinuity and a singularity. The red transport map is the Wasserstein optimal transport be￾tween fµ(x) and fν(y). In contrast to the black transport, the Wasserstein transport is one-to-one, but it is not C 1 everywhere. i… view at source ↗
Figure 2
Figure 2. Figure 2: Top left: A uniform density fµ(x) on [0, 1] is transported to a discontinuous density fν(y) with a smooth, non-injective transport (denoted by “folded transport” T). The coloring is with respect to the x-axis, and shows how fν(y) arises as the sum of the two branches of the transport map (compare (6)). Top right: The same uniform density fµ(x) is transported to the same discontinuous density fν(y) with a c… view at source ↗
Figure 3
Figure 3. Figure 3: Left: On an unknown manifold (black curve), obtained from an unknown parametrization (uniform on x-axis, blue), we observe consecutive values of y starting from uniformly distributed points in x. Middle: Histogram recorded from y-values. Right: With the knowlegde of two con￾secutive y values, we can reconstruct (up to a diffeomorphism) the original curve using a delay￾embedding. We show the time-delay embe… view at source ↗
Figure 4
Figure 4. Figure 4: Left: Plot shows the curve c(x) = (x, T(x)) in the (x, y, 0)-plane, the discontinuous density fν(y) on y ≡ T(x), and the curve x 7→ (x, T(x), T(x − τ )), all colored by arclength of d(x) = (T(x − τ ), T(x)). Right: Densities are plotted as height over the respective axis: Uniform density on [0, 1] (x-axis), discontinuous density on y-axis, and density over the arclength, plotted on the curve (transport map… view at source ↗
Figure 5
Figure 5. Figure 5: Left: On an unknown manifold (curve), obtained from an unknown parametrization (uni￾form on x-axis, indicated by coloring), we observe consecutive values of y starting from uniformly distributed points in x. Middle: Histogram recorded from y-values. The coloring of the histogram indicates from which part of the curve (left) it has been produced. Right: With the knowlegde of two consecutive y values, we can… view at source ↗
Figure 6
Figure 6. Figure 6: Left: On an unknown manifold (curve), obtained from an unknown parametrization (uniform on its arclength), we observe consecutive values of projections to the y-axis starting from uniformly distributed points on the arclength of the curve. Middle: Histogram recorded from y￾values. The coloring of the histogram indicates from which part of the curve (left) it has been produced. Right: With the knowlegde of … view at source ↗
Figure 7
Figure 7. Figure 7: Top right: The cusp surface, together with its parametrization in (x, β1) and the cusp (black curve), shown as the projection of the folds of the surface on the plane of the two parameters. We treat this surface as the intrinsic, unknown manifold. We observe, for each randomly chosen initial condition (xn, [β1]n), the values [β1]n+1, [β2]n and [β2]n+1, as the observation process moves in β1-direction. Top … view at source ↗
Figure 8
Figure 8. Figure 8: Top right: The same surface as in [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Application of DMAP with Mahalanobis distance to the PCA parametrization of the [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: A manifold (M, g) (in this example M = [0, 1] and g is the Euclidean metric) is mapped to itself through an invertible function S : M → M. This map induces a new metric g 0 = S∗g on M. The two axes indicate how measures, illustrated as red point distributions, are mapped by S: The uniform density of Volg (the measure on M induced by the Lebesgue measure on R, horizontal axis) is mapped by S to a uniform d… view at source ↗
Figure 11
Figure 11. Figure 11: A commutative diagram showing how isometries are represented as orthogonal mappings [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Commutative diagram showing how measure-preserving maps induce measure-preserving [PITH_FULL_IMAGE:figures/full_fig_p016_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Uniformly sampled points xi on the unit square (upper left, blue), are mapped with the mushroom map S (11) to yi (upper right). DMAP with Mahalanobis distance (coming from the Ja￾cobian of S −1 ) is applied to yi , to obtain an embedding φM(yi) (lower right). The points yi are also mapped back to the unit square (˜xi , upper left, red) with the Wasserstein optimal transport (com￾puted from the mushroom-di… view at source ↗
Figure 14
Figure 14. Figure 14: Notation as in Figure 13. We compare the functions that relate the diffusion maps [PITH_FULL_IMAGE:figures/full_fig_p018_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Continuation of the example discussed in Figure 1. We construct a [PITH_FULL_IMAGE:figures/full_fig_p019_15.png] view at source ↗
read the original abstract

Different observations of a relation between inputs ("sources") and outputs ("targets") are often reported in terms of histograms (discretizations of the source and the target densities). Transporting these densities to each other provides insight regarding the underlying relation. In (forward) uncertainty quantification, one typically studies how the distribution of inputs to a system affects the distribution of the system responses. Here, we focus on the identification of the system (the transport map) itself, once the input and output distributions are determined, and suggest a modification of current practice by including data from what we call "an observation process". We hypothesize that there exists a smooth manifold underlying the relation; the sources and the targets are then partial observations (possibly projections) of this manifold. Knowledge of such a manifold implies knowledge of the relation, and thus of "the right" transport between source and target observations. When the source-target observations are not bijective (when the manifold is not the graph of a function over both observation spaces, either because folds over them give rise to density singularities, or because it marginalizes over several observables), recovery of the manifold is obscured. Using ideas from attractor reconstruction in dynamical systems, we demonstrate how additional information in the form of short histories of an observation process can help us recover the underlying manifold. The types of additional information employed and the relation to optimal transport based solely on density observations is illustrated and discussed, along with limitations in the recovery of the true underlying relation.

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 / 1 minor

Summary. The paper claims that hypothesizing a smooth manifold of which source and target densities are partial observations allows recovery of the underlying relation (and thus the correct transport map) by incorporating short histories of an observation process, drawing on attractor reconstruction techniques from dynamical systems; this is proposed to address cases where direct source-target mappings are non-bijective due to folds or marginalization, leading to density singularities.

Significance. If the reconstruction step can be made rigorous, the approach would extend density-transport methods to singular cases by using sequential observation data, offering a geometric alternative to standard optimal transport in uncertainty quantification settings where the transport map itself must be identified.

major comments (2)
  1. [Abstract] Abstract (final paragraph): the claim that short histories recover the manifold rests on an unstated transfer of time-delay embedding results, but the manuscript supplies no analogue of the required conditions (compact invariant set under a diffeomorphism, genericity of the observation map) for an arbitrary static source-target relation; without this, there is no guarantee of injectivity or uniqueness of the recovered transport.
  2. [Abstract] Abstract (hypothesis statement): the assertion that knowledge of the manifold implies knowledge of 'the right' transport is load-bearing, yet the text provides neither a derivation showing how finite-length histories yield an embedding nor an error analysis or validation that the recovered manifold determines a unique map when folds are present.
minor comments (1)
  1. [Abstract] The abstract refers to 'the types of additional information employed' and 'limitations in the recovery' but does not enumerate them; a short explicit list would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and for identifying points where the abstract requires greater precision regarding the connection to embedding results and the implications of manifold recovery. We address the two major comments below.

read point-by-point responses

  1. Referee: [Abstract] Abstract (final paragraph): the claim that short histories recover the manifold rests on an unstated transfer of time-delay embedding results, but the manuscript supplies no analogue of the required conditions (compact invariant set under a diffeomorphism, genericity of the observation map) for an arbitrary static source-target relation; without this, there is no guarantee of injectivity or uniqueness of the recovered transport.

    Authors: The manuscript draws an explicit analogy to attractor reconstruction via time-delay embeddings but applies it to a hypothesized static manifold underlying source-target observations rather than to a dynamical system. We agree that the abstract does not articulate the precise analogue of the compactness, invariance, and genericity conditions required for a rigorous guarantee of injectivity. The approach is presented under the standing hypothesis that such a manifold exists; for arbitrary static relations lacking this structure the recovery is not claimed. In revision we will expand the abstract and add a dedicated paragraph in the introduction stating the minimal assumptions transferred from the dynamical-systems literature and the consequent limitations on uniqueness. revision: yes

  2. Referee: [Abstract] Abstract (hypothesis statement): the assertion that knowledge of the manifold implies knowledge of 'the right' transport is load-bearing, yet the text provides neither a derivation showing how finite-length histories yield an embedding nor an error analysis or validation that the recovered manifold determines a unique map when folds are present.

    Authors: The hypothesis that the manifold encodes the underlying relation (and therefore selects the transport consistent with that relation) is indeed central. The current text illustrates the recovery via short histories in concrete examples but supplies neither a general derivation for finite-length delay vectors nor a quantitative error analysis for the fold case. We will therefore add a methods subsection that sketches how finite histories are assembled into delay coordinates and will include an explicit statement of the conditions under which the recovered manifold yields a unique transport map, together with a brief discussion of residual non-uniqueness when folds remain after embedding. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external dynamical systems concepts

full rationale

The provided abstract and description contain no equations, fitted parameters, or self-citations that reduce any claimed result to an input by construction. The central approach hypothesizes an underlying smooth manifold and invokes ideas from attractor reconstruction as an external method to recover it from short histories, without defining the recovery in terms of the target transport map itself. No load-bearing step matches the enumerated circularity patterns; the argument is framed as importing an independent technique rather than a self-referential fit or renaming. This is the expected non-finding for a paper whose core proposal does not reduce to tautology or internal data fitting.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the existence of an underlying smooth manifold and the utility of short observation histories for its reconstruction; no free parameters, invented entities, or additional axioms are specified in the abstract.

axioms (1)
  • domain assumption There exists a smooth manifold underlying the source-target relation
    Explicitly stated as the hypothesis in the abstract.

pith-pipeline@v0.9.0 · 5799 in / 1057 out tokens · 18731 ms · 2026-05-24T19:36:01.655667+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.

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

37 extracted references · 37 canonical work pages

  1. [1]

    Adler, I.-M

    J. Adler, I.-M. Sintorn, R. Strand, and I. Parmryd. Conventional analysis of movement on non-flat surfaces like the plasma membrane makes Brownian motion appear anomalous. Com- munications Biology, 2(12), 2019. doi: 10.1038/s42003-018-0240-2

    work page doi:10.1038/s42003-018-0240-2 2019

  2. [2]

    D. Aeyels. Generic observability of differentiable systems. SIAM J. Control Optim. , 19(5): 595–603, 1981. doi: 10.1137/0319037

    work page doi:10.1137/0319037 1981

  3. [3]

    Belkin and P

    M. Belkin and P. Niyogi. Laplacian eigenmaps for dimensionality reduction and data repre- sentation. Neural Comput., 15(6):1373–1396, 2003. doi: 10.1162/089976603321780317

    work page doi:10.1162/089976603321780317 2003

  4. [4]

    Berry and T

    T. Berry and T. Sauer. Local kernels and the geometric structure of data. Appl. Comput. Harmon. Anal., 40(3):439–469, 2016. doi: 10.1016/j.acha.2015.03.002

    work page doi:10.1016/j.acha.2015.03.002 2016

  5. [5]

    Y. Brenier. Polar factorization and monotone rearrangement of vector-valued functions.Comm. Pure Appl. Math. , 44(4):375–417, 1991. doi: 10.1002/cpa.3160440402

    work page doi:10.1002/cpa.3160440402 1991

  6. [6]

    Chen and R

    L. Chen and R. K. Mehra. A study of nonlinear filters with particle flow induced by log- homotopy. In I. Kadar, editor, Signal Processing, Sensor Fusion, and Target Recognition XIX. SPIE, 2010. doi: 10.1117/12.853001

    work page doi:10.1117/12.853001 2010

  7. [7]

    R. R. Coifman and S. Lafon. Diffusion maps. Appl. Comput. Harmon. Anal. , 21(1):5–30, 2006. doi: 10.1016/j.acha.2006.04.006. Special Issue: Diffusion Maps and Wavelets

    work page doi:10.1016/j.acha.2006.04.006 2006

  8. [8]

    Courty, R

    N. Courty, R. Flamary, and D. Tuia. Domain adaptation with regularized optimal transport. In Machine Learning and Knowledge Discovery in Databases , pages 274–289. Springer Berlin Heidelberg, 2014. doi: 10.1007/978-3-662-44848-9 \ 18

    work page doi:10.1007/978-3-662-44848-9 2014

  9. [9]

    M. Cuturi. Sinkhorn distances: Lightspeed computation of optimal transport. In C. J. C. Burges, L. Bottou, M. Welling, Z. Ghahramani, and K. Q. Weinberger, editors, Advances in Neural Information Processing Systems 26 , pages 2292–2300. Curran Associates, Inc., 2013

    work page 2013

  10. [10]

    Dietrich, M

    F. Dietrich, M. Kooshkbaghi, E. M. Bollt, and I. G. Kevrekidis. Manifold learning for orga- nizing unstructured sets of process observations. arXiv:1810.12952v3, 2018

    work page arXiv 2018

  11. [11]

    Dsilva, R

    C. Dsilva, R. Talmon, C. Gear, R. Coifman, and I. Kevrekidis. Data-driven reduction for a class of multiscale fast-slow stochastic dynamical systems. SIAM J. Appl. Dyn. Syst. , 15(3): 1327–1351, 2016. doi: 10.1137/151004896

    work page doi:10.1137/151004896 2016

  12. [12]

    F. P. Kemeth, S. W. Haugland, F. Dietrich, T. Bertalan, K. Hohlein, Q. Li, E. M. Bollt, R. Talmon, K. Krischer, and I. G. Kevrekidis. An emergent space for distributed data with hidden internal order through manifold learning. IEEE Access, 2018. doi: 10.1109/access.2018. 2882777

    work page doi:10.1109/access.2018 2018

  13. [13]

    P. G. Kevrekidis, C. I. Siettos, and Y. G. Kevrekidis. To infinity and some glimpses of beyond. Nature Communications, 8(1), 2017. doi: 10.1038/s41467-017-01502-7

    work page doi:10.1038/s41467-017-01502-7 2017

  14. [14]

    Kim and J

    Y.-H. Kim and J. Kitagawa. Prohibiting isolated singularities in optimal transport. Ann. Sc. Norm. Super. Pisa Cl. Sci. , XVI(5):277–290, 2016. 24

    work page 2016

  15. [15]

    J. M. Lee. Introduction to Smooth Manifolds . Springer New York, 2012. doi: 10.1007/ 978-1-4419-9982-5

    work page 2012

  16. [16]

    G. Loeper. On the regularity of solutions of optimal transportation problems. Acta Math., 202 (2):241–283, 2009. doi: 10.1007/s11511-009-0037-8

    work page doi:10.1007/s11511-009-0037-8 2009

  17. [17]

    X.-N. Ma, N. S. Trudinger, and X.-J. Wang. Regularity of potential functions of the optimal transportation problem. Arch. Ration. Mech. Anal. , 177(2):151–183, 2005. doi: 10.1007/ s00205-005-0362-9

    work page 2005

  18. [18]

    Marchenko, D

    D. Marchenko, D. V. Evtushinsky, E. Golias, A. Varykhalov, T. Seyller, and O. Rader. Extremely flat band in bilayer graphene. Science Advances , 4(11):eaau0059, 2018. doi: 10.1126/sciadv.aau0059

    work page doi:10.1126/sciadv.aau0059 2018

  19. [19]

    R. J. McCann. Polar factorization of maps on Riemannian manifolds. Geometric & Functional Analysis GAFA, 11(3):589–608, 2001. doi: 10.1007/PL00001679

    work page doi:10.1007/pl00001679 2001

  20. [20]

    G. Monge. M´ emoire sur la Th´ eorie des D´ eblais et des Remblais.Historie de l’Acad´ emie Royale des Sciences de Paris, avec les M´ emoires de Math´ ematique et Physique pour la Mˆ eme ann´ ee, pages 666–704, 1781

    work page

  21. [21]

    N. H. Packard, J. P. Crutchfield, J. D. Farmer, and R. S. Shaw. Geometry from a time series. Phys. Rev. Lett., 45(9):712–716, 1980

    work page 1980

  22. [22]

    Parno and Y

    M. Parno and Y. M. Marzouk. Transport map accelerated Markov chain Monte Carlo. SIAM/ASA J. Uncertain. Quantif. , 6(2):645–682, 2018. doi: 10.1137/17M1134640

    work page doi:10.1137/17m1134640 2018

  23. [23]

    Parno, T

    M. Parno, T. Moselhy, and Y. M. Marzouk. A multiscale strategy for Bayesian inference using transport maps. SIAM/ASA J. Uncertain. Quantif. , 4(1):1160–1190, 2016. doi: 10.1137/ 15M1032478

    work page 2016

  24. [24]

    J. W. Portegies. Embeddings of Riemannian manifolds with heat kernels and eigenfunctions. Comm. Pure Appl. Math. , 69(3):478–518, 2016. doi: 10.1002/cpa.21565

    work page doi:10.1002/cpa.21565 2016

  25. [25]

    Rosenberg

    S. Rosenberg. The Laplacian on a Riemannian Manifold: An Introduction to Analysis on Manifolds. London Mathematical Society Student Texts. Cambridge University Press, 1997. doi: 10.1017/CBO9780511623783

    work page doi:10.1017/cbo9780511623783 1997

  26. [26]

    Sauer, J

    T. Sauer, J. A. Yorke, and M. Casdagli. Embedology. J. Stat. Phys. , 65(3):579–616, 1991

    work page 1991

  27. [27]

    Singer and R

    A. Singer and R. R. Coifman. Non-linear independent component analysis with diffusion maps. Appl. Comput. Harmon. Anal. , 25(2):226–239, 2008. doi: 10.1016/j.acha.2007.11.001

    work page doi:10.1016/j.acha.2007.11.001 2008

  28. [28]

    Singer, R

    A. Singer, R. Erban, I. G. Kevrekidis, and R. R. Coifman. Detecting intrinsic slow variables in stochastic dynamical systems by anisotropic diffusion maps. Proc. Nat. Acad. Sci. , 106: 16090–16095, 2009. doi: 10.1073/pnas.0905547106

    work page doi:10.1073/pnas.0905547106 2009

  29. [29]

    J. Stark. Delay Embeddings for Forced Systems. I. Deterministic Forcing. J. Nonlinear Sci. , 9(3):255–332, 1999. doi: 10.1007/s003329900072. 25

    work page doi:10.1007/s003329900072 1999

  30. [30]

    Stark, D

    J. Stark, D. Broomhead, M. Davies, and J. Huke. Takens embedding theorems for forced and stochastic systems. Nonlinear Analysis: Theory, Methods & Applications , 30(8):5303–5314,

    work page

  31. [31]

    doi: 10.1016/s0362-546x(96)00149-6

    work page doi:10.1016/s0362-546x(96)00149-6

  32. [32]

    Stark, D

    J. Stark, D. Broomhead, M. Davies, and J. Huke. Delay Embeddings for Forced Systems. II. Stochastic Forcing. J. Nonlinear Sci. , 13(6):519–577, 2003. doi: 10.1007/s00332-003-0534-4

    work page doi:10.1007/s00332-003-0534-4 2003

  33. [33]

    F. Takens. Detecting strange attractors in turbulence. Lecture Notes in Math., pages 366–381,

    work page

  34. [34]

    doi: 10.1007/bfb0091924

    work page doi:10.1007/bfb0091924

  35. [35]

    C. Villani. Optimal Transport . Springer Berlin Heidelberg, 2009. doi: 10.1007/ 978-3-540-71050-9

    work page 2009

  36. [36]

    H. Whitney. Differentiable manifolds. Ann. of Math. , 37(3):645, 1936. doi: 10.2307/1968482

    work page doi:10.2307/1968482 1936

  37. [37]

    O. Yair, M. Ben-Chen, and R. Talmon. Parallel transport on the cone manifold of SPD matrices for domain adaptation. IEEE Transactions on Signal Processing , 67(7):1797–1811, 2019. doi: 10.1109/tsp.2019.2894801. 26

    work page doi:10.1109/tsp.2019.2894801 2019