Article Contents
Article Contents

# Data-driven evolutions of critical points

• * Corresponding author
S.A. acknowledges the support of the DFG Transregio 109 "Discretization in Geometry and Dynamics". M.F. acknowledges the support of the DFG Project "Identification of Energies from Observation of Evolutions" and the DFG SPP 1962 "Non-smooth and Complementarity-based Distributed Parameter Systems: Simulation and Hierarchical Optimization". R.H. acknowledges the support of the DFG-FWF IGDK1754 "Optimization and Numerical Analysis for Partial Differential Equations with Nonsmooth Structures" and the hospitality of TUM during the preparation of this work
• In this paper we are concerned with the learnability of energies from data obtained by observing time evolutions of their critical points starting at random initial equilibria. As a byproduct of our theoretical framework we introduce the novel concept of mean-field limit of critical point evolutions and of their energy balance as a new form of transport. We formulate the energy learning as a variational problem, minimizing the discrepancy of energy competitors from fulfilling the equilibrium condition along any trajectory of critical points originated at random initial equilibria. By $\Gamma$-convergence arguments we prove the convergence of minimal solutions obtained from finite number of observations to the exact energy in a suitable sense. The abstract framework is actually fully constructive and numerically implementable. Hence, the approximation of the energy from a finite number of observations of past evolutions allows one to simulate further evolutions, which are fully data-driven. As we aim at a precise quantitative analysis, and to provide concrete examples of tractable solutions, we present analytic and numerical results on the reconstruction of an elastic energy for a one-dimensional model of thin nonlinear-elastic rod.

Mathematics Subject Classification: Primary: 49N80, 68T05; Secondary: 34A55, 70F17.

 Citation:

• Figure 1.  Reconstructions $\hat a$ (left) and $\hat a'$ (right) in dotted red and true $a$ and $a'$ in blue. The underlying red circles depict the (adaptive) nodes in $\Lambda$. Below you see the distribution $\tilde \eta$ of available data

Figure 2.  Reconstruction of $a'$ with increasing number ${N_e}$ of measurements. The true $a'$ is depicted by the blue curve while the red line depicts the reconstruction $\hat a'$ following (5.23). The red circles at the bottom of the figures depict the position of nodes of the underlying mesh $\Lambda_N$. The histogram below describes the density of the available information encoded by the probability measure $\tilde{\eta}$ for the case $N_e = 60$

Figure 3.  Reconstruction of $a'$ for varying $N$ and $D(N) = 4N$ and distribution $\tilde \eta$ for $N = 60$. Graphic as described in Figure 2

Figure 4.  Reconstruction of $a'$ for varying $N$ and fixed $D(N) = 300$ and distribution $\tilde \eta$. Graphics as described in Figure 2

Figure 5.  Impact of different constants $M_2$ for constraints on $a''$ with $\|a''\| \leq [2,5;20,1000]$ on reconstructs are depicted, showing improved approximations for increasing $M_2$. Graphics as in Figure 2

Figure 6.  Impact of changing constraint $M_1$ bounding the values of $a'$, showing a projection-like behaviour of the reconstruction for small $M_1$. Graphics as in Figure 2

Figure 7.  Left showing $\hat x_\varepsilon(t)$ (dark dashed) and $x_\varepsilon(t)$ (line bright) with data from Section 5.2.2 at times $t = [0.2$, $0.4$, $0.6$, $0.8$, $1.0]$, right with data from Section 5.2.3

•  [1] V. Agostiniani and R. Rossi, Singular vanishing-viscosity limits of gradient flows: The finite-dimensional case, J. Differential Equations, 263 (2017), 7815-7855.  doi: 10.1016/j.jde.2017.08.027. [2] V. Agostiniani, R. Rossi and G. Savaré, On the transversality conditions and their genericity, Rend. Circ. Mat. Palermo, 64 (2015), 101-116.  doi: 10.1007/s12215-014-0184-4. [3] V. Agostiniani, R. Rossi, and G. Savaré, Singular vanishing-viscosity limits of gradient flows in Hilbert spaces, personal communication: in preparation, (2018). [4] V. Albani, U. M. Ascher, X. Yang and J. P. Zubelli, Data driven recovery of local volatility surfaces, Inverse Probl. Imaging, 11 (2017), 799-823.  doi: 10.3934/ipi.2017038. [5] L. Ambrosio, N. Fusco and D. Pallara, Functions of Bounded Variation and Free Discontinuity Problems, Oxford Mathematical Monographs. The Clarendon Press, Oxford University Press, New York, 2000. [6] L. Ambrosio, N. Gigli and G. Savaré, Gradient flows in metric spaces and in the space of probability measures, Lectures in Mathematics ETH Zürich, Birkhäuser Verlag, Basel, second ed., 2008. [7] M. Bongini, M. Fornasier, M. Hansen and M. Maggioni, Inferring interaction rules from observations of evolutive systems Ⅰ: The variational approach, Math. Models Methods Appl. Sci., 27 (2017), 909-951.  doi: 10.1142/S0218202517500208. [8] A. Braides, Γ-Convergence for Beginners, Oxford Lecture Series in Mathematics and its Applications, 22. Oxford University Press, Oxford, 2002. doi: 10.1093/acprof:oso/9780198507840.001.0001. [9] T. Q. Chen, Y. Rubanova, J. Bettencourt and D. K. Duvenaud, Neural ordinary differential equations, Advances in Neural Information Processing Systems, S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi and R. Garnett, eds., Curran Associates, Inc., 31 (2018), 6571–6583. [10] S. Conti, S. Müller and M. Ortiz, Data-driven problems in elasticity, Arch. Ration. Mech. Anal., 229 (2018), 79-123.  doi: 10.1007/s00205-017-1214-0. [11] S. Crépey, Calibration of the local volatility in a generalized Black-Scholes model using Tikhonov regularization, SIAM J. Math. Anal., 34 (2003), 1183-1206.  doi: 10.1137/S0036141001400202. [12] T. S. Cubitt, J. Eisert and M. M. Wolf, Extracting dynamical equations from experimental data is NP hard, Phys. Rev. Lett., 108 (2012), 120503. doi: 10.1103/PhysRevLett.108.120503. [13] G. Dal Maso, An Introduction to $\Gamma$-Convergence, vol. 8 of Progress in Nonlinear Differential Equations and their Applications, Birkhäuser Boston, Inc., Boston, MA, 1993. doi: 10.1007/978-1-4612-0327-8. [14] W. E, A proposal on machine learning via dynamical systems, Commun. Math. Stat., 5 (2017), 1-11.  doi: 10.1007/s40304-017-0103-z. [15] W. E, J. Han and Q. Li, A mean-field optimal control formulation of deep learning, Res. Math. Sci., 6 (2019), No. 10, 41 pp. doi: 10.1007/s40687-018-0172-y. [16] H. Egger and H. W. Engl, Tikhonov regularization applied to the inverse problem of option pricing: Convergence analysis and rates, Inverse Problems, 21 (2005), 1027-1045.  doi: 10.1088/0266-5611/21/3/014. [17] R. Gerstberger and P. Rentrop, Feedforward neural nets as discretization schemes for ODES and DAES, 7th ICCAM 96 Congress., J. Comput. Appl. Math., 82 (1997), 117-128.  doi: 10.1016/S0377-0427(97)00085-X. [18] M. C. Grant and S. P. Boyd, Graph implementations for nonsmooth convex programs, Recent Advances in Learning and Control, V. Blondel, S. Boyd, and H. Kimura, eds., Lecture Notes in Control and Information Sciences, Springer-Verlag Limited, 371 (2008), 95–110. doi: 10.1007/978-1-84800-155-8_7. [19] M. Grant and S. Boyd, CVX: Matlab Software for Disciplined Convex Programming, version 2.1., http://cvxr.com/cvx, Mar. 2014. [20] J. Guckenheimer and P. Holmes, Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields, vol. 42 of Applied Mathematical Sciences, Springer-Verlag, New York, 1983. doi: 10.1007/978-1-4612-1140-2. [21] T. Kirchdoerfer and M. Ortiz, Data-driven computational mechanics, Comput. Methods Appl. Mech. Engrg., 304 (2016), 81-101.  doi: 10.1016/j.cma.2016.02.001. [22] F. Lu, M. Maggioni and S. Tang, Learning interaction kernels in heterogeneous systems of agents from multiple trajectories, arXiv: 1910.04832. [23] F. Lu, M. Zhong, S. Tang and M. Maggioni, Nonparametric inference of interaction laws in systems of agents from trajectory data, Proc. Natl. Acad. Sci. USA, 116 (2019), 14424-14433.  doi: 10.1073/pnas.1822012116. [24] E. Novak and H. Woźniakowski, Approximation of infinitely differentiable multivariate functions is intractable, J. Complexity, 25 (2009), 398-404.  doi: 10.1016/j.jco.2008.11.002. [25] H. Schaeffer, G. Tran and R. Ward, Extracting sparse high-dimensional dynamics from limited data, SIAM J. Appl. Math., 78 (2018), 3279-3295.  doi: 10.1137/18M116798X. [26] H. Schaeffer, G. Tran and R. Ward, Learning dynamical systems and bifurcation via group sparsity, arXiv: 1709.01558. [27] G. Scilla and F. Solombrino, Delayed loss of stability in singularly perturbed finite-dimensional gradient flows, Asymptot. Anal., 110 (2018), 1-19.  doi: 10.3233/ASY-181475. [28] G. Scilla and F. Solombrino, Multiscale analysis of singularly perturbed finite dimensional gradient flows: The minimizing movement approach, Nonlinearity, 31 (2018), 5036-5074.  doi: 10.1088/1361-6544/aad6ac. [29] G. Tran and R. Ward, Exact recovery of chaotic systems from highly corrupted data, Multiscale Model. Simul., 15 (2017), 1108-1129.  doi: 10.1137/16M1086637. [30] C. Zanini, Singular perturbations of finite dimensional gradient flows, Discrete Contin. Dyn. Syst., 18 (2007), 657-675.  doi: 10.3934/dcds.2007.18.657. [31] M. Zhong, J. Miller and M. Maggioni, Data-driven discovery of emergent behaviors in collective dynamics, Phys. D, 411 (2020), 132542, 25 pp. doi: 10.1016/j.physd.2020.132542.

Figures(7)