Skip to main content

Least action principles for incompressible flows and geodesics between shapes

Abstract

As V. I. Arnold observed in the 1960s, the Euler equations of incompressible fluid flow correspond formally to geodesic equations in a group of volume-preserving diffeomorphisms. Working in an Eulerian framework, we study incompressible flows of shapes as critical paths for action (kinetic energy) along transport paths constrained to have characteristic-function densities. The formal geodesic equations for this problem are Euler equations for incompressible, inviscid potential flow of fluid with zero pressure and surface tension on the free boundary. The problem of minimizing this action exhibits an instability associated with microdroplet formation, with the following outcomes: any two shapes of equal volume can be approximately connected by an Euler spray—a countable superposition of ellipsoidal geodesics. The infimum of the action is the Wasserstein distance squared, and is almost never attained except in dimension 1. Every Wasserstein geodesic between bounded densities of compact support provides a solution of the (compressible) pressureless Euler system that is a weak limit of (incompressible) Euler sprays.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2

References

  1. 1.

    Ambrosio, L., Gigli, N.: A user’s guide to optimal transport. In: Modelling and Optimisation of Flows on Networks, vol. 2062 of Lecture Notes in Mathematics, pp. 1–155. Springer, Heidelberg (2013)

  2. 2.

    Ambrosio, L., Gigli, N., Savaré, G.: Gradient Flows in Metric Spaces and in the Space of Probability Measures, Lectures in Mathematics, 2nd edn. ETH Zürich, Birkhäuser Verlag, Basel (2008)

  3. 3.

    Arnold, V.: Sur la géométrie différentielle des groupes de Lie de dimension infinie et ses applications à l’hydrodynamique des fluides parfaits. Ann. Inst. Fourier (Grenoble) 16, 319–361 (1966)

    MathSciNet  MATH  Article  Google Scholar 

  4. 4.

    Bauschke, H.H., Combettes, P.L.: Convex Analysis and Monotone Operator Theory in Hilbert Spaces, CMS Books in Mathematics/Ouvrages de Mathématiques de la SMC, Springer, Cham, 2nd ed. With a foreword by Hédy Attouch (2017)

  5. 5.

    Beg, M.F., Miller, M.I., Trouvé, A., Younes, L.: Computing large deformation metric mappings via geodesic flows of diffeomorphisms. Int. J. Comput. Vis. 61, 139–157 (2005)

    Article  Google Scholar 

  6. 6.

    Benamou, J.-D., Brenier, Y.: A computational fluid mechanics solution to the Monge–Kantorovich mass transfer problem. Numer. Math. 84, 375–393 (2000)

    MathSciNet  MATH  Article  Google Scholar 

  7. 7.

    Braides, A.: \(\Gamma \)-Convergence for Beginners. Oxford Lecture Series in Mathematics and its Applications, vol. 22. Oxford University Press, Oxford (2002)

  8. 8.

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

    MathSciNet  MATH  Article  Google Scholar 

  9. 9.

    Brenier, Y., Otto, F., Seis, C.: Upper bounds on coarsening rates in demixing binary viscous liquids. SIAM J. Math. Anal. 43, 114–134 (2011)

    MathSciNet  MATH  Article  Google Scholar 

  10. 10.

    Brézis, H.: Opérateurs maximaux monotones et semi-groupes de contractions dans les espaces de Hilbert. North-Holland, Amsterdam; Elsevier, New York. North-Holland Mathematics Studies, No. 5. Notas de Matemática (50) (1973)

  11. 11.

    Bruveris, M., Michor, P.W., Mumford, D.: Geodesic completeness for Sobolev metrics on the space of immersed plane curves. Forum Math. Sigma 2, e19–e38 (2014)

    MathSciNet  MATH  Article  Google Scholar 

  12. 12.

    Bruveris, M., Vialard, F.-X.: On completeness of groups of diffeomorphisms. J. Eur. Math. Soc. 19, 1507–1544 (2017)

    MathSciNet  MATH  Article  Google Scholar 

  13. 13.

    Caffarelli, L.A.: Some regularity properties of solutions of Monge Ampère equation. Comm. Pure Appl. Math. 44, 965–969 (1991)

    MathSciNet  MATH  Article  Google Scholar 

  14. 14.

    Coutand, D., Shkoller, S.: Well-posedness of the free-surface incompressible Euler equations with or without surface tension. J. Am. Math. Soc. 20, 829–930 (2007)

    MathSciNet  MATH  Article  Google Scholar 

  15. 15.

    Coutand, D., Shkoller, S.: A simple proof of well-posedness for the free-surface incompressible Euler equations. Discrete Contin. Dyn. Syst. Ser. S 3, 429–449 (2010)

    MathSciNet  MATH  Article  Google Scholar 

  16. 16.

    Dautray, R., Lions, J.-L.: Mathematical Analysis and Numerical Methods for Science and Technology. Vol. 3, Springer, Berlin: Spectral theory and applications, With the collaboration of Michel Artola and Michel Cessenat. Translated from the French by John C, Amson (1990)

  17. 17.

    De Philippis, G., Figalli, A.: The Monge–Ampère equation and its link to optimal transportation. Bull. Am. Math. Soc. (N.S.) 51, 527–580 (2014)

    MATH  Article  Google Scholar 

  18. 18.

    De Philippis, G., Figalli, A.: Partial regularity for optimal transport maps. Publ. Math. Inst. Hautes Études Sci. 121, 81–112 (2015)

    MathSciNet  MATH  Article  Google Scholar 

  19. 19.

    Dirichlet, P.G.L.: Untersuchungen über ein Problem der Hydrodynamik, vol. 8, Dieterichschen Buchhandlung (1860)

  20. 20.

    Dunford, N., Schwartz, J.T.: Linear Operators. I. General Theory, With the Assistance of W. G. Bade and R. G. Bartle. Pure and Applied Mathematics, Vol. 7. Interscience, New York/London (1958)

  21. 21.

    Dupuis, P., Grenander, U., Miller, M.I.: Variational problems on flows of diffeomorphisms for image matching. Q. Appl. Math. 56, 587–600 (1998)

    MathSciNet  MATH  Article  Google Scholar 

  22. 22.

    Ebin, D.G., Marsden, J.: Groups of diffeomorphisms and the motion of an incompressible fluid. Ann. Math. 92(2), 102–163 (1970)

    MathSciNet  MATH  Article  Google Scholar 

  23. 23.

    Evans, L.C., Gariepy, R.F.: Measure Theory and fine Properties of Functions, Studies in Advanced Mathematics. CRC Press, Boca Raton (1992)

  24. 24.

    Figalli, A.: Regularity properties of optimal maps between nonconvex domains in the plane. Comm. Partial Differ. Equ. 35, 465–479 (2010)

    MathSciNet  MATH  Article  Google Scholar 

  25. 25.

    Figalli, A., Kim, Y.-H.: Partial regularity of Brenier solutions of the Monge–Ampère equation. Discrete Contin. Dyn. Syst. 28, 559–565 (2010)

    MathSciNet  MATH  Article  Google Scholar 

  26. 26.

    Fuchs, M., Jüttler, B., Scherzer, O., Yang, H.: Shape metrics based on elastic deformations. J. Math. Imaging Vis. 35, 86–102 (2009)

    MathSciNet  Article  Google Scholar 

  27. 27.

    Gangbo, W., McCann, R.J.: Shape recognition via Wasserstein distance. Q. Appl. Math. 58, 705–737 (2000)

    MathSciNet  MATH  Article  Google Scholar 

  28. 28.

    García Trillos, N., Slepčev, D.: Continuum limit of total variation on point clouds. Arch. Ration. Mech. Anal. 220, 193–241 (2016)

    MathSciNet  MATH  Article  Google Scholar 

  29. 29.

    Gay-Balmaz, F., Holm, D.D., Ratiu, T.S.: Geometric dynamics of optimization. Commun. Math. Sci. 11, 163–231 (2013)

    MathSciNet  MATH  Article  Google Scholar 

  30. 30.

    Grenander, U., Miller, M.I.: Computational anatomy: an emerging discipline. Q. Appl. Math. 56, 617–694 (1998)

    MathSciNet  MATH  Article  Google Scholar 

  31. 31.

    Haker, S., Zhu, L., Tannembaum, A., Angenent, S.: Optimal mass transport for registration and warping. Int. J. Comput. Vis. 60, 225–240 (2004)

    Article  Google Scholar 

  32. 32.

    Haraux, A.: Nonlinear Evolution Equations–Global Behavior of Solutions. Lecture Notes in Mathematics, vol. 841. Springer, Berlin (1981)

  33. 33.

    Holm, D.D., Trouvé, A., Younes, L.: The Euler–Poincaré theory of metamorphosis. Q. Appl. Math. 67, 661–685 (2009)

    MATH  Article  Google Scholar 

  34. 34.

    Jost, J.: Riemannian Geometry and Geometric Analysis, 6th edn. Universitext, Springer, Heidelberg (2011)

    MATH  Book  Google Scholar 

  35. 35.

    Knott, M., Smith, C.S.: On the optimal mapping of distributions. J. Optim. Theory Appl. 43, 39–49 (1984)

    MathSciNet  MATH  Article  Google Scholar 

  36. 36.

    Lamb, H.: Hydrodynamics, Cambridge Mathematical Library, 6th edn. Cambridge University Press, Cambridge (1993). With a foreword by R. A. Caflisch [Russel E. Caflisch]

  37. 37.

    Lindblad, H.: Well-posedness for the motion of an incompressible liquid with free surface boundary. Ann. Math. 162(2), 109–194 (2005)

    MathSciNet  MATH  Article  Google Scholar 

  38. 38.

    Liu, J.-G., Pego, R.L., Slepčev, D.: Euler sprays and Wasserstein geometry of the space of shapes. arXiv:1604.03387v1

  39. 39.

    Longuet-Higgins, M.S.: A class of exact, time-dependent, free-surface flows. J. Fluid Mech. 55, 529–543 (1972)

    MATH  Article  Google Scholar 

  40. 40.

    McCann, R.J.: A convexity principle for interacting gases. Adv. Math. 128, 153–179 (1997)

    MathSciNet  MATH  Article  Google Scholar 

  41. 41.

    Michor, P.W., Mumford, D.: Vanishing geodesic distance on spaces of submanifolds and diffeomorphisms. Doc. Math. 10, 217–245 (2005)

    MathSciNet  MATH  Google Scholar 

  42. 42.

    Michor, P.W., Mumford, D.: Riemannian geometries on spaces of plane curves. J. Eur. Math. Soc. 8, 1–48 (2006)

    MathSciNet  MATH  Article  Google Scholar 

  43. 43.

    Mignot, F.: Contrôle dans les inéquations variationelles elliptiques. J. Funct. Anal. 22, 130–185 (1976)

    MATH  Article  Google Scholar 

  44. 44.

    Otto, F.: The geometry of dissipative evolution equations: the porous medium equation. Comm. Partial Differ. Equ. 26, 101–174 (2001)

    MathSciNet  MATH  Article  Google Scholar 

  45. 45.

    Rubner, Y., Tomassi, C., Guibas, L.J.: The earth mover’s distance as a metric for image retrieval. Int. J. Comput. Vis. 40, 99–121 (2000)

    MATH  Article  Google Scholar 

  46. 46.

    Rumpf, M., Wirth, B.: Discrete geodesic calculus in shape space and applications in the space of viscous fluidic objects. SIAM J. Imaging Sci. 6, 2581–2602 (2013)

    MathSciNet  MATH  Article  Google Scholar 

  47. 47.

    Santambrogio, F.: Optimal Transport for Applied Mathematicians, Progress in Nonlinear Differential Equations and Their Applications, Vol. 87. Birkhäuser/Springer, Cham (2015); Calculus of Variations, PDEs, and Modeling

  48. 48.

    Schmitzer, B., Schnörr, C.: Contour manifolds and optimal transport. arXiv:1309.2240 (2013)

  49. 49.

    Schmitzer, B., Schnörr, C.: Globally optimal joint image segmentation and shape matching based on Wasserstein modes. J. Math. Imaging Vis. 52, 436–458 (2015)

    MathSciNet  MATH  Article  Google Scholar 

  50. 50.

    Thompson, D.W.: On Growth and Form. Cambridge University Press, Cambridge (1917)

    Google Scholar 

  51. 51.

    Trouvé, A.: Action de groupe de dimension infinie et reconnaissance de formes. C. R. Acad. Sci. Paris Sér. I Math. 321, 1031–1034 (1995)

    MathSciNet  MATH  Google Scholar 

  52. 52.

    Villani, C.: Topics in Optimal Transportation. Graduate Studies in Mathematics, vol. 58. American Mathematical Society, Providence (2003)

  53. 53.

    Villani, C.: Optimal Transport, vol. 338 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer, Berlin (2009); Old and New

  54. 54.

    Wang, W., Ozolek, J.A., Slepčev, D., Lee, A.B., Chen, C., Rohde, G.K.: An optimal transportation approach for nuclear structure-based pathology. IEEE Trans. Med. Imaging 30, 621–631 (2011)

    Article  Google Scholar 

  55. 55.

    Wang, W., Slepčev, D., Basu, S., Ozolek, J.A., Rohde, G.K.: A linear optimal transportation framework for quantifying and visualizing variations in sets of images. Int. J. Comput. Vis. 101, 254–269 (2013)

    MathSciNet  MATH  Article  Google Scholar 

  56. 56.

    Wirth, B., Bar, L., Rumpf, M., Sapiro, G.: A continuum mechanical approach to geodesics in shape space. Int. J. Comput. Vis. 93, 293–318 (2011)

    MathSciNet  MATH  Article  Google Scholar 

  57. 57.

    Wu, S.: Well-posedness in Sobolev spaces of the full water wave problem in 2-D. Invent. Math. 130, 39–72 (1997)

    MathSciNet  MATH  Article  Google Scholar 

  58. 58.

    Wu, S.: Well-posedness in Sobolev spaces of the full water wave problem in 3-D. J. Am. Math. Soc. 12, 445–495 (1999)

    MathSciNet  MATH  Article  Google Scholar 

  59. 59.

    Younes, L.: Computable elastic distances between shapes. SIAM J. Appl. Math. 58, 565–586 (1998). (electronic)

    MathSciNet  MATH  Article  Google Scholar 

  60. 60.

    Younes, L.: Shapes and Diffeomorphisms, Applied Mathematical Sciences, vol. 171. Springer, Berlin (2010)

  61. 61.

    Younes, L., Michor, P.W., Shah, J., Mumford, D.: A metric on shape space with explicit geodesics. Atti Accad. Naz. Lincei Cl. Sci. Fis. Mat. Natur. Rend. Lincei (9) Mat. Appl. 19, 25–57 (2008)

    MathSciNet  MATH  Article  Google Scholar 

Download references

Acknowledgements

The authors thank Yann Brenier for enlightening discussions and generous hospitality. Thanks also to Matt Thorpe for the computation of the optimal transport map appearing in Fig. 1, and to Yue Pu for careful reading and corrections. The authors also appreciate the efforts of several referees whose suggestions led to improvements in this paper. This material is based upon work supported by the National Science Foundation under NSF Research Network Grant No. RNMS 1107444 (KI-Net) and partially supported by the Center for Nonlinear Analysis (CNA) under National Science Foundation PIRE Grant No. OISE 0967140. The first author was partially supported by the National Science Foundation with Grant DMS 1514826. The second author was partially supported by the National Science Foundation with Grants DMS 1211161 and DMS 1515400, and by the Simons Foundation under Grant 395796. The third author was partially supported by the National Science Foundation with Grants CCF 1421502, DMS 1516677 and DMS 1814991.

Author information

Affiliations

Authors

Corresponding author

Correspondence to Dejan Slepčev.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Communicated by L. Ambrosio.

Appendices

Appendix A: Some basic facts about subgradients

For the convenience of readers, we include here proofs of a few facts about subgradients that we use in Sect. 4 for the proof of Theorem 1.1. The proofs are standard and simple but seem not to be easy to extract from monographs on the subject, e.g., see [4, 10, 32].

Proposition A.1

Let H be a Hilbert space, and let \(\varphi :H\rightarrow (-\infty ,\infty ]\) be convex, lower semi-continuous, and proper (i.e., somewhere finite). Let \(S(x) = \frac{1}{2}\Vert x\Vert ^2+\varphi (x)\). Then:

  1. i.

    The subgradient \(\partial \varphi \) is a monotone operator.

  2. ii.

    \(\partial S(x) = x + \partial \varphi (x)\) for all \(x\in H\).

  3. iii.

    The range of \(\partial S\) is all of H. I.e., for all \(y\in H\) there exists \(x\in H\) and \(z\in \partial \varphi (x)\) such that \(y= x+z\).

Proof

i. Given any \(x,{\hat{x}} \in H, z\in \partial \varphi (x)\), \({\hat{z}}\in \partial \varphi ({\hat{x}})\), by the definition of \(\partial \varphi (x)\) and \(\partial \varphi ({\hat{x}})\) respectively we have \( \varphi ({\hat{x}}) - \varphi (x) \ge \langle z,{\hat{x}} - x \rangle \) and \( \varphi (x) - \varphi ({\hat{x}}) \ge \langle {\hat{z}},x -{\hat{x}} \rangle \), whence \( 0\le \langle z-{\hat{z}},x-{\hat{x}} \rangle \). This proves \(\partial \varphi \) is monotone.

ii. 1. Let \(z\in \partial \varphi (x)\). We claim \(x+z\in \partial S(x)\). Indeed, for all \(h\in H\),

$$\begin{aligned} \frac{1}{2}\Vert x+h\Vert ^2 + \varphi (x+h) \ge \frac{1}{2}\Vert x\Vert ^2 + \varphi (x) + \langle z+x,h \rangle . \end{aligned}$$

2. Suppose \(z\notin \partial \varphi (x)\). We claim \(z+x\notin \partial S(x)\). We know there exists \(h\in H\) such that

$$\begin{aligned} t^{-1}(\varphi (x+th)-\varphi (x)) - \langle z,h \rangle < 0 \end{aligned}$$

for \(t=1\), hence for all \(t\in (0,1]\) by convexity. Then for sufficiently small \(t>0\) we can add \(\frac{1}{2} t\Vert h\Vert ^2\) to the left-hand side and conclude that for small \(t>0\),

$$\begin{aligned} \frac{1}{2} \Vert th\Vert ^2 + \varphi (x+th) < \varphi (x) + \langle z,th \rangle , \end{aligned}$$

whence \(z+x\notin \partial S(x)\), since

$$\begin{aligned} \frac{1}{2} \Vert x+th\Vert ^2 + \varphi (x+th) < \frac{1}{2}\Vert x\Vert ^2+\varphi (x) + \langle z+x,th \rangle . \end{aligned}$$

iii. Let \(y\in H\) and define \({\hat{S}}(x)=S(x)-\langle y,x \rangle =\frac{1}{2}\Vert x\Vert ^2+\varphi (x)-\langle y,x \rangle \). Due to our hypotheses, \({\hat{S}}\) has a minimum at some \(x\in H\). This implies that for all \(h\in H\),

$$\begin{aligned} \frac{1}{2}\Vert x+h\Vert ^2 +\varphi (x+h)\ge \frac{1}{2}\Vert x\Vert ^2 +\varphi (x)-\langle y,h \rangle , \end{aligned}$$

which means that \(y\in \partial S(x)=x+\partial \varphi (x)\). \(\square \)

Appendix B: \({T\!L^p}\) stability of Wasserstein geodesics

Here we recall the notion of \({T\!L^p}\) convergence as introduced in [28], which provides a more precise way to compare Wasserstein geodesics than the notion of weak convergence does alone. We establish the \({T\!L^p}\) stability of optimal transport maps in Theorem B.1 and \({T\!L^p}\) stability of the Wasserstein geodesics in Corollary B.2. In Proposition B.4 we recall a basic property of \({T\!L^p}\) convergence and use it to show that the stability in Corollary B.2 holds even if the maps used to couple the relevant measures are not optimal. This technical result is needed in the proofs in Sect. 6.2.

The \({T\!L^p}\) metric provides a natural setting for comparing optimal transport maps between different probability measures. Let \({{\mathcal {P}}}_p({\mathbb {R}}^d)\) be the space of probability measures on \({\mathbb {R}}^d\) with finite p-th moments. On the space \({T\!L^p}({\mathbb {R}}^d)\), consisting of all ordered pairs \((\mu ,g)\) where \( \mu \in {{\mathcal {P}}}_p({\mathbb {R}}^d)\) and \( g \in L^p(\mu )\), the metric is given as follows: For \(1\le p<\infty \),

$$\begin{aligned} d_{{T\!L^p}} (( \mu _0, g_0), ( \mu _1, g_1)) = \inf _{\pi \in \Pi ( \mu _0, \mu _1)}\left( \int |x-y|^p + | g_0(x) - g_1(y)|^p\ d \pi (x,y) \right) ^{1/p},\\ d_{T\!L^\infty }((\mu _0,g_0),(\mu _1,g_1)) = \inf _{\pi \in \Pi ( \mu _0, \mu _1)} \mathop {\mathrm{ess\, sup}}_\pi (|x-y|+|g_0(x)-g_0(y)| ) , \end{aligned}$$

where \( \Pi ( \mu _0, \mu _1)\) is the set of transportation plans (couplings) between \( \mu _0\) and \( \mu _1\).

The following result establishes a (new) \({T\!L^p}\) stability property for optimal transport maps, as a consequence of a known general stability property for optimal plans.

Theorem B.1

(\({T\!L^p}\) stability of transport maps) Let \(\mu ,\mu _k\in {{\mathcal {P}}}_p({\mathbb {R}}^d)\) be probability measures absolutely continuous with respect to Lebesgue measure, and let \(\nu ,\nu _k\in {{\mathcal {P}}}_p({\mathbb {R}}^d)\), for each \(k\in {\mathbb {N}}\). Assume that

$$\begin{aligned} d_p(\mu _k, \mu ) \rightarrow 0\; \text { and } \; d_p(\nu _k,\nu ) \rightarrow 0 \; \text { as } k \rightarrow \infty . \end{aligned}$$

Let \(T_k\) and T be the optimal transportation maps between \(\mu _k\) and \(\nu _k\), and \(\mu \) and \(\nu \), respectively. Then

$$\begin{aligned} (\mu _k,T_k) \overset{{T\!L^p}}{\longrightarrow } (\mu ,T) \text { as } k \rightarrow \infty . \end{aligned}$$

Proof

The measures \(\pi _k= ({{\,\mathrm{id}\,}}\times T_k)_\sharp \mu _k\) and \(\pi = ({{\,\mathrm{id}\,}}\times T)_\sharp \mu \) are the optimal transportation plans between \(\mu _k \) and \(\nu _k\), and \(\mu \) and \(\nu \), respectively. By stability of optimal transport plans (Proposition 7.1.3 of [2] or Theorem 5.20 in [53]) the sequence \(\pi _k\) is precompact with respect to weak convergence and each of its subsequential limits is an optimal transport plan between \(\mu \) and \(\nu \). Since \(\pi \) is the unique optimal transportation plan between \(\mu \) and \(\nu \) the sequence \(\pi _k\) converges to \(\pi \). Furthermore, by Theorem 5.11 of [47] or Remark 7.1.11 of [2],

$$\begin{aligned} \lim _{k \rightarrow \infty } \int |x|^p + |y|^p \ d \pi _k(x,y)&= \lim _{k \rightarrow \infty } \int |x|^p\, d\mu _k + \int |y|^p\, d \nu _k \\&= \int |x|^p\, d\mu + \int |y|^p\, d \nu = \int |x|^p + |y|^p\ d \pi (x,y). \end{aligned}$$

By Lemma 5.1.7 of [2], it follows the \(\pi _k\) have uniformly integrable p-th moments, therefore

$$\begin{aligned} d_p(\pi _k, \pi )\rightarrow 0 \quad \text{ as } k \rightarrow \infty , \end{aligned}$$

by Proposition 7.1.5 in [2]. Hence there exists (optimal) \(\gamma _k \in \Pi (\pi , \pi _k)\) such that

$$\begin{aligned} \int |x - {\tilde{x}}|^p + |y- {\tilde{y}}|^p \ d\gamma _k(x,y,{\tilde{x}}, {\tilde{y}}) \rightarrow 0 \quad \text { as } k \rightarrow \infty . \end{aligned}$$
(B.1)

Since \(\pi \)-almost everywhere \(y = T(x)\) and \(\pi _k\)-almost everywhere \({\tilde{y}} = T_k({\tilde{x}})\) and the support \({{\,\mathrm{supp}\,}}\gamma _k\) of \(\gamma _k\) is contained in \({{\,\mathrm{supp}\,}}\pi \times {{\,\mathrm{supp}\,}}\pi _k\), we conclude that \(\gamma _k\)-almost everywhere \((x,y, {\tilde{x}}, {\tilde{y}}) = (x,T(x), {\tilde{x}}, T_k({\tilde{x}}))\). Therefore

$$\begin{aligned} \int |x - {\tilde{x}}|^p + |T(x)- T_k({\tilde{x}})|^p\ d\gamma _k(x,y,{\tilde{x}}, {\tilde{y}}) \rightarrow 0 \quad \text { as } k \rightarrow \infty . \end{aligned}$$

Finally let \(\theta _k\) be the projection of \(\gamma _k\) to \((x, {\tilde{x}})\) variables. Since \(\theta _k \in \Pi (\mu , \mu _k)\), by above

$$\begin{aligned} \int |x - {\tilde{x}}|^p + |T(x)- T_k({\tilde{x}})|^p\ d\theta _k(x, {\tilde{x}}) \rightarrow 0 \quad \text { as } k \rightarrow \infty . \end{aligned}$$
(B.2)

Thus \((\mu _k, T_k) \overset{{T\!L^p}}{\longrightarrow } (\mu ,T)\). \(\square \)

We now consider the convergence of Wasserstein geodesics between measures \(\mu _k\) and \(\nu _k\) as in the Lemma B.1, treating only the case \(p=2\). We recall that particle paths along these geodesics are given by

$$\begin{aligned} T_{k,t}(x) = (1-t)x + t T_k(x). \end{aligned}$$

The displacement interpolation between \(\mu _k\) and \(\nu _k\), and particle velocities (in Eulerian variables) along the geodesics, are given by (cf. (6.15)–(6.16))

$$\begin{aligned} \mu _{k,t} = {T_{k,t}}_\sharp \mu _k, \quad v_{k,t}= (T_k-{{\,\mathrm{id}\,}})\circ T_{k,t}^{-1}, \quad t\in [0,1). \end{aligned}$$

If \(\nu _k\) is absolutely continuous with respect to Lebesgue measure, then \(t=1\) is allowed. We also recall that

$$\begin{aligned} \int |v_{k,t}(z)|^2 d \mu _{k,t}(z) = \int |v_{k,0}(x)|^2 d \mu _k(x) = d_2^2(\mu _k, \nu _k). \end{aligned}$$

Furthermore it is straightforward to check that \(t \mapsto (\mu _{k,t}, v_{k,t})\) is Lipschitz continuous into \(T\!L^2({\mathbb {R}}^d)\), satisfying for \(0\le s<t<1\)

$$\begin{aligned} (t-s)d_2(\mu _k,\nu _k)=d_2(\mu _{k,t},\mu _{k,s}) \le d_{{T\!L^2}}((\mu _{k,t},v_{k,t}),(\mu _{k,s},v_{k,s})) \le (t-s)d_2(\mu _k,\nu _k).\nonumber \\ \end{aligned}$$
(B.3)

Corollary B.2

(\({T\!L^2}\) stability for displacement interpolants) Under the assumptions of Theorem B.1 for the case \(p=2\), as \(k\rightarrow \infty \) we have

$$\begin{aligned} \sup _{t\in [0,1]} d_2(\mu _{k,t},\mu _t) \rightarrow 0 \quad \text{ and }\quad \sup _{t\in [0,1)} d_{{T\!L^2}}( (\mu _{k,t},v_{k,t}),(\mu _t,v_t)) \rightarrow 0. \end{aligned}$$
(B.4)

If the measures \(\nu _k\) and \(\nu \) are absolutely continuous with respect to Lebesgue measure then the convergence in (B.4) also holds for \(t\in [0,1]\).

Proof

Let \(\pi \in \Pi (\mu , \nu ), \pi _k \in \Pi (\mu _k, \nu _k)\), and \(\gamma _k \in \Pi ( \pi , \pi _k)\) be as in the proof of Theorem B.1. Similarly to \(\theta _k\), we define \(\theta _{k,t} = (z_t \times z_t)_{\sharp } \gamma _k\) where

$$\begin{aligned} z_t(x,y) = (1-t)x+ty \quad \text{ and }\quad (z_t \times z_t)(x,y,{\tilde{x}}, {\tilde{y}}) = (z_t(x,y), z_t({\tilde{x}}, {\tilde{y}})). \end{aligned}$$

We note that \(\theta _{k,t} \in \Pi (\mu _t, \mu _{k,t})\) and hence, for all \(t\in [0,1]\),

$$\begin{aligned} d_2(\mu _t, \mu _{k,t})^2&\le \int |z-{\tilde{z}}|^2 d \theta _{k,t}(z,{\tilde{z}}) \\&= \int |(1-t)(x-{\tilde{x}}) + t(y - {\tilde{y}})|^2 d \gamma _k(x, y, {\tilde{x}}, {\tilde{y}}) \\&\le 2 \int |x - {\tilde{x}}|^2 + |y - {\tilde{y}}|^2 d \gamma _k(x, y, {\tilde{x}}, {\tilde{y}}) , \end{aligned}$$

which by (B.1) converges to 0 as \(k \rightarrow \infty \).

We use the same coupling \(\theta _{k,t}\) to compare the velocities. Using that \(\gamma _k\)-almost everywhere \((x,y, {\tilde{x}}, {\tilde{y}}) = (x,T(x), {\tilde{x}}, T_k({\tilde{x}}))\), for any \(t\in [0,1)\) we obtain

$$\begin{aligned}&\int |v_t(z) - v_{k,t}({\tilde{z}})|^2\ d \theta _{k,t}(z, {\tilde{z}}) \\&\qquad = \int |v_t((1-t)x + ty) - v_{k,t}((1-t) {\tilde{x}} + t {\tilde{y}})|^2\ d \gamma _k(x, y, {\tilde{x}}, {\tilde{y}}) \\&\qquad = \int |v_t(T_t(x)) - v_{k,t}(T_{k,t}({\tilde{x}}))|^2\ d \theta _k(x, {\tilde{x}}) \\&\qquad = \int |v_0(x)-v_{k,0}({\tilde{x}})|^2 \ d \theta _k(x, {\tilde{x}}) \\&\qquad \le 2 \int |x - {\tilde{x}}|^2 + |T(x) - T_k({\tilde{x}})|^2 \ d \theta _k(x, {\tilde{x}}), \end{aligned}$$

which converges to 0 as \(k \rightarrow \infty \), as in (B.2). \(\square \)

Remark B.3

If the target measure \(\nu _k\) is not absolutely continuous with respect to Lebesgue measure, then \(T_k\) may fail to be invertible on the support of \(\nu _k\) and \((\mu _{k,t},v_{k,t})\) may fail to converge as \(t\rightarrow 1\) to some point in \({T\!L^2}({\mathbb {R}}^d)\) due to oscillations in velocity. However, if \(\nu _k\) and \(\nu \) are absolutely continuous with respect to Lebesgue measure, then the curves \(t \mapsto (\mu _{k,t}, v_{k,t}), t\mapsto (\mu _t,v_t)\) extend as continuous maps into \({T\!L^2}\) for all \(t\in [0,1]\), and the uniform convergences in (B.4) holds on [0, 1].

A number of properties of the \({T\!L^p}\) metric are established in Section 3 of [28] for measures supported in a fixed bounded set. One useful characterization of \({T\!L^p}\)-convergence in this case is stated in Proposition 3.12 of [28], which implies the following.

Proposition B.4

(A characterization of \({T\!L^p}\) convergence on bounded domains) Let \(D\subset {\mathbb {R}}^d\) be open and bounded, and let \(\mu \) and \(\mu _k\) (\(k\in {\mathbb {N}}\)) be probability measures on D, and suppose \(\mu \) is absolutely continuous with respect to Lebesgue measure. Call a sequence of transport maps \((S_k)\) that push forward \(\mu \) to \(\mu _k\) (satisfying \({S_k}_\sharp \mu =\mu _k\)) stagnating if

$$\begin{aligned} \lim _{n\rightarrow \infty }\int _D |x-S_k(x)|\,d\mu (x) = 0 . \end{aligned}$$
(B.5)

Then the following are equivalent, for \(1\le p<\infty \).

  1. (i)

    \((\mu _k,f_k)\overset{{T\!L^p}}{\longrightarrow } (\mu ,f)\) as \(k\rightarrow \infty \).

  2. (ii)

    \(\mu _k\) converges weakly to \(\mu \) and there exists a stagnating sequence \((S_k)\) such that

    $$\begin{aligned} \int _D |f(x)-f_k(S_k(x))|^p \,d\mu (x) \rightarrow 0 \quad \text{ as } k\rightarrow \infty . \end{aligned}$$
    (B.6)
  3. (iii)

    \(\mu _k\) converges weakly to \(\mu \) and for every stagnating sequence \((S_k)\) the equality (B.6) holds.

This result together with Proposition B.2 yields the following.

Corollary B.5

(A characterization of \({T\!L^p}\) convergence for displacement interpolants) Make the same assumptions as in Corollary B.2, and assume all measures \(\mu _k, \mu \), \(\nu _k, \nu \) are absolutely continuous with respect to Lebesgue measure and have support in a bounded open set D. Then for any stagnating sequence of transport maps \((S_k)\) that push forward \(\mu \) to \(\mu _k\), with the notation

$$\begin{aligned} S_{k,t} = T_{k,t}\circ S_k\circ T_t^{-1}\end{aligned}$$

the sequence \((S_{k,t})\) pushes forward \(\mu _t\) to \(\mu _{k,t}\) and is stagnating, and as \(k\rightarrow \infty \),

$$\begin{aligned}&\sup _{t\in [0,1]} \int |x-S_{k,t}(x)|^2\,d\mu _t(x) \rightarrow 0, \end{aligned}$$
(B.7)
$$\begin{aligned}&\sup _{t\in [0,1]} \int |v_t(x)-v_{k,t}(S_{k,t}(x))|^2\,d\mu _t(x) \rightarrow 0, \end{aligned}$$
(B.8)

Proof

First we note that indeed

$$\begin{aligned} \mu _{k,t}=(T_{k,t})_\sharp \mu _k = (T_{k,t}\circ S_k)_\sharp \mu = (S_{k,t})_\sharp \mu _t. \end{aligned}$$

Next, fix any \(t\in [0,1]\). Because \(d_2(\mu _{k,t},\mu _t)\rightarrow 0\) by (B.4) and \(T_{k,t}\) is the optimal transport map pushing forward \(\mu _k\) to \(\mu _{k,t}\), by Theorem B.1 we have \(d_2( (\mu _k,T_{k,t}),(\mu ,T_t))\rightarrow 0\). Now by Proposition B.4, because \((T_t)_\sharp \mu =\mu _t\) we have

$$\begin{aligned} \int |x-S_{k,t}(x)|^2 \,d\mu _t(x) = \int |T_t(z)-T_{k,t}(S_k(z))|^2\,d\mu (z)\rightarrow 0. \end{aligned}$$
(B.9)

We infer that \((S_{k,t})\) is stagnating and the convergence in (B.7) holds pointwise in t. But now, the middle quantity in (B.9) is a quadratic function of t, so the uniform convergence in (B.7) holds.

Next, we note that the quantity in (B.8) is actually independent of t. We have

$$\begin{aligned} \int |v_t(x)-v_{k,t}(S_{k,t}(x))|^2\,d\mu _t(x) =\int |v_0(z)-v_{k,0}(S_k(z))|^2\,d\mu (z) \rightarrow 0, \end{aligned}$$

due to Proposition B.4. \(\square \)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Liu, JG., Pego, R.L. & Slepčev, D. Least action principles for incompressible flows and geodesics between shapes. Calc. Var. 58, 179 (2019). https://doi.org/10.1007/s00526-019-1636-7

Download citation

Mathematics Subject Classification

  • 35Q35
  • 65D18
  • 35J96
  • 58E10
  • 53C22