Controllability on landmark manifolds for shapes and neural ODEs

Grong, Erlend and Vega-Molino, Sylvie, 2026


abstract: "<jats:title>Abstract</jats:title>\n <jats:p>Landmark manifolds consist of finite collections of distinct points in an underlying space, referred to as landmark configurations. The dynamics of these configurations can be used to represent flows, such as solutions to ODEs or shape deformations. In this work, we consider landmark configurations in Euclidean space and study how they can be connected via flows of vector fields. For dimensions greater than or equal to two, we explicitly construct two vector fields whose flows can connect any pair of landmark configurations with the same cardinality. This property is known as exact universal interpolation. In dimension one, we show that the same result holds for pairs of landmark configurations that share the same relative ordering. In all dimensions, controllability is achieved using one constant vector field and one polynomial vector field of degree three.</jats:p>" author: Grong, Erlend and Vega-Molino, Sylvie author_list: - affiliation: [] family: Grong given: Erlend - affiliation: [] family: Vega-Molino given: Sylvie citations: - author: A Agrachev doi: 10.1051/cocv/2016029 first-page: '921' issue: '4' journal-title: ESAIM Control Optim Calc Var unstructured: Agrachev A, Baryshnikov Y, Sarychev A (2016) Ensemble controllability by Lie algebraic methods. ESAIM Control Optim Calc Var 22(4):921–938 volume: '22' year: '2016' - author: A Agrachev doi: 10.1007/s10883-021-09561-2 first-page: '989' issue: '4' journal-title: J Dyn Control Syst unstructured: Agrachev A, Sarychev A (2022) Control on the manifolds of mappings with a view to the deep learning. J Dyn Control Syst 28(4):989–1008 volume: '28' year: '2022' - unstructured: Agrachev AA, Caponigro M (2009) Controllability on the group of diffeomorphisms. Ann I H Poincaré - doi: 10.1007/978-3-662-06404-7 unstructured: Agrachev AA, Sachkov YL (2004) Control theory from the geometric viewpoint, volume 87 of Encyclopaedia of Mathematical Sciences. Springer, Berlin. Control Theory and Optimization, II - author: A Álvarez López doi: 10.1016/j.neunet.2024.106640 first-page: '106640' journal-title: Neural Netw unstructured: Álvarez López A, Slimane AH, Zuazua E (2024) Interplay between depth and width for interpolation in neural odes. Neural Netw 180:106640 volume: '180' year: '2024' - unstructured: Arguillère S (2015) The general setting for shape analysis. arXiv:1504.01767 - author: S Arguillère doi: 10.1017/S1474748015000249 first-page: '745' journal-title: J Inst Math Jussieu unstructured: Arguillère S, Trélat E (2014) Sub-Riemannian structures on groups of diffeomorphisms. J Inst Math Jussieu 16:745–785 volume: '16' year: '2014' - author: S Arguillère doi: 10.1016/j.matpur.2015.02.004 first-page: '139' issue: '1' journal-title: J Math Pures Appl unstructured: Arguillère S, Trélat E, Trouvé A, Younes L (2015) Shape deformation analysis from the optimal control viewpoint. J Math Pures Appl 104(1):139–178 volume: '104' year: '2015' - author: A Arnaudon doi: 10.1007/s10208-018-9394-z first-page: '653' journal-title: Found Comput Math unstructured: Arnaudon A, Holm DD, Sommer S (2018) A geometric framework for stochastic shape analysis. Found Comput Math 19:653–701 volume: '19' year: '2018' - author: M Bauer first-page: 09 journal-title: J Geom Mech unstructured: Bauer M, Harms P, Michor P (2010) Sobolev metrics on shape space of surfaces. J Geom Mech 3:09 volume: '3' year: '2010' - doi: 10.1007/978-3-030-18921-1_4 unstructured: 'Boscain U, Sigalotti M (2019) Introduction to controllability of nonlinear systems. In: Contemporary research in elliptic PDEs and related topics. pp 203—-219' - unstructured: 'Chen RTQ, Rubanova Y, Bettencourt J, Duvenaud DK (2018) Neural ordinary differential equations. In: Bengio S, Wallach H, Larochelle H, Grauman K, Cesa-Bianchi N, Garnett R (eds) Advances in neural information processing systems, volume 31. Curran Associates Inc' - author: X Chen doi: 10.1007/s00498-019-0237-5 first-page: '7' issue: '2' journal-title: Math Control Signals Syst unstructured: Chen X (2019) Structure theory for ensemble controllability, observability, and duality. Math Control Signals Syst 31(2):7–40 volume: '31' year: '2019' - doi: 10.1137/23M1599744 unstructured: Cheng J, Li Q, Lin T, Shen Z (2025) Interpolation approximation, and controllability of deep neural networks. SIAM J Control Optim 63(1):625–649 - author: W-L Chow doi: 10.1007/BF01450011 first-page: '98' journal-title: Math Ann unstructured: Chow W-L (1939) Über Systeme von linearen partiellen Differentialgleichungen erster Ordnung. Math Ann 117:98–105 volume: '117' year: '1939' - author: C Cuchiero doi: 10.1137/19M1284117 first-page: '901' issue: '3' journal-title: SIAM J Math Data Sci unstructured: Cuchiero C, Larsson M, Teichmann J (2020) Deep neural networks, generic universal interpolation, and controlled odes. SIAM J Math Data Sci 2(3):901–919 volume: '2' year: '2020' - doi: 10.1109/TAC.2025.3569123 unstructured: Duan Y, Cai Y (2025) A minimal control family of dynamical systems for universal approximation. IEEE Trans Autom Control - unstructured: Glöckner H, Grong E, Schmeding A (2024) Boundary values of diffeomorphisms of simple polytopes, and controllability. arXiv:2407.05444 - author: B Gris doi: 10.1137/16M1076733 first-page: '802' journal-title: SIAM J Imaging Sci unstructured: Gris B, Durrleman S, Trouvé A (2018) A sub-Riemannian modular framework for diffeomorphism-based analysis of shape ensembles. SIAM J Imaging Sci 11:802–833 volume: '11' year: '2018' - author: E Grong first-page: '1291' issue: '4' journal-title: Forum Math unstructured: Grong E, Schmeding A (2025) Controllability and diffeomorphism groups on manifolds with boundary. Forum Math 37(4):1291–1308 volume: '37' year: '2025' - unstructured: Grong E, Sommer S (2022) Most probable paths for developed processes. arXiv:2211.15168 - author: P Harms doi: 10.1016/j.difgeo.2023.102029 journal-title: Differ Geom Appl unstructured: Harms P, Michor PW, Pennec X, Sommer S (2023) Geometry of sample spaces. Differ Geom Appl 90:102029 volume: '90' year: '2023' - author: SC Joshi doi: 10.1109/83.855431 first-page: '1357' issue: '8' journal-title: IEEE Trans Image Process Publ IEEE Signal Process Soc unstructured: Joshi SC, Miller MI (2000) Landmark matching via large deformation diffeomorphisms. IEEE Trans Image Process Publ IEEE Signal Process Soc 9(8):1357–70 volume: '9' year: '2000' - doi: 10.1090/surv/053 unstructured: Kriegl A, Michor P (1997) The convenient setting of global analysis, volume 53 of mathematical surveys and monographs. American Mathematical Society 10 - author: C Lobry doi: 10.21136/CMJ.1972.101094 first-page: '230' issue: '97' journal-title: Czechoslovak Math J unstructured: Lobry C (1972) Une propriété générique des couples de champs de vecteurs. Czechoslovak Math J 22(97):230–237 volume: '22' year: '1972' - doi: 10.1137/10081678X unstructured: Micheli M, Michor PW, Mumford D (2012) Sectional curvature in terms of the cometric, with applications to the Riemannian manifolds of landmarks. SIAM J Imag Sci 5(1):394–433 - author: PW Michor doi: 10.4171/jems/37 first-page: '1' issue: '1' journal-title: J Eur Math Soc unstructured: Michor PW, Mumford D (2006) Riemannian geometries on spaces of plane curves. J Eur Math Soc 8(1):1–48 volume: '8' year: '2006' - author: PW Michor doi: 10.1016/j.acha.2006.07.004 first-page: '74' issue: '1' journal-title: Appl Comput Harmon Anal unstructured: Michor PW, Mumford D (2007) An overview of the Riemannian metrics on spaces of curves using the Hamiltonian approach. Appl Comput Harmon Anal 23(1):74–113 volume: '23' year: '2007' - author: R Montgomery doi: 10.1006/jdeq.1993.1056 first-page: '387' issue: '2' journal-title: J Differ Equ unstructured: Montgomery R (1993) Generic distributions and Lie algebras of vector fields. J Differ Equ 103(2):387–393 volume: '103' year: '1993' - author: P Rashevsky first-page: '83' journal-title: Uch Zapiski Ped Inst Libknechta unstructured: Rashevsky P (1938) About connecting two points of a completely nonholonomic space by admissible curve. Uch Zapiski Ped Inst Libknechta 2:83–94 volume: '2' year: '1938' - doi: 10.1137/21M1411433 unstructured: Ruiz-Balet D, Zuazua E (2023) Neural ode control for classification, approximation, and transport. SIAM Rev 65(3):735–773 - unstructured: Sussmann H (1990) Nonlinear controllability and optimal control, 1st edn. Routledge - author: P Tabuada doi: 10.1109/TAC.2022.3190051 first-page: '2715' issue: '5' journal-title: IEEE Trans Autom Control unstructured: Tabuada P, Gharesifard B (2022) Universal approximation power of deep residual neural networks through the lens of control. IEEE Trans Autom Control 68(5):2715–2728 volume: '68' year: '2022' - doi: 10.1007/978-3-642-12055-8 unstructured: 'Younes L (2010) Shapes and diffeomorphisms. In: Applied Mathematical Sciences. Springer' - author: L Younes doi: 10.1007/s10208-011-9108-2 first-page: '295' journal-title: Found Comput Math unstructured: Younes L (2012) Constrained diffeomorphic shape evolution. Found Comput Math 12:295–325 volume: '12' year: '2012' doc_url: https://link.springer.com/content/pdf/10.1007/s00498-026-00435-1.pdf doi: 10.1007/s00498-026-00435-1 files: - s00498-026-00435-1.pdf issue: '2' journal: Mathematics of Control, Signals, and Systems language: en month: 6 pages: 275--289 papis_id: b3e254b5696b1e8e21ce3bacfff79973 publisher: Springer Science and Business Media LLC ref: Controllability_Grong_2026 time-added: 2026-08-19-22:34:42 title: Controllability on landmark manifolds for shapes and neural ODEs type: article url: https://doi.org/10.1007/s00498-026-00435-1 volume: '38' year: 2026

@article{Controllability_Grong_2026, abstract = {<jats:title>Abstract</jats:title> <jats:p>Landmark manifolds consist of finite collections of distinct points in an underlying space, referred to as landmark configurations. The dynamics of these configurations can be used to represent flows, such as solutions to ODEs or shape deformations. In this work, we consider landmark configurations in Euclidean space and study how they can be connected via flows of vector fields. For dimensions greater than or equal to two, we explicitly construct two vector fields whose flows can connect any pair of landmark configurations with the same cardinality. This property is known as exact universal interpolation. In dimension one, we show that the same result holds for pairs of landmark configurations that share the same relative ordering. In all dimensions, controllability is achieved using one constant vector field and one polynomial vector field of degree three.</jats:p>}, author = {Grong, Erlend and Vega-Molino, Sylvie}, doi = {10.1007/s00498-026-00435-1}, issue = {2}, journal = {Mathematics of Control, Signals, and Systems}, language = {en}, month = {6}, pages = {275--289}, publisher = {Springer Science and Business Media LLC}, title = {Controllability on landmark manifolds for shapes and neural ODEs}, url = {https://doi.org/10.1007/s00498-026-00435-1}, volume = {38}, year = {2026}, }

ads
Provided by ads
ads
Provided by ads