Newton’s method for nonlinear mappings into vector bundles Part II: Application to Variational Problems
Laura Weigl, Ronny Bergmann, Anton Schiela · 2025
Read Newton’s method for nonlinear mappings into vector bundles Part II: Application to Variational Problems by Laura Weigl, Ronny Bergmann, Anton Schiela (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-13836.txt.
Introduction
Laura Weigl, Ronny Bergmann, and Anton Schiela extend Newton's method to curved spaces in a 2025 paper. Many physics and engineering problems ask for the state that makes an energy smallest, which can be rewritten as finding a root of a mapping between curved spaces. Newton's method, the workhorse for root finding, needs geometry-aware tools in that setting.
This second part of a series builds the needed calculus for mappings into vector bundles and adds a damping strategy that keeps steps safe without coordinates. The method is then applied to several variational problems, with numerical experiments showing it working in practice. The audience is specialists in optimization on manifolds. The payoff is a solver that respects the geometry of problems ranging from computer vision shapes to mechanical systems, rather than flattening them first.
Catalog record
| CREATOR | Laura Weigl, Ronny Bergmann, Anton Schiela |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 86 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-13836
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Newton’s method for nonlinear mappings into vector bundles Part II: Application to Variational Problems Laura Weigl, Ronny Bergmann, Anton Schiela Abstract . We consider the solution of variational equations on manifolds by Newton’s method. These problems can be expressed as root finding problems for mappings from infinite dimensional manifolds into dual vector bundles. We derive the differential geometric tools needed for the realization of Newton’s method, equipped with an affine covariant damping strategy. We apply Newton’s method to a couple of variational problems and show numerical results. 2020 Mathematics Subject Classification 49Q99 1. Introduction Newton’s method is one of the standard algorithms for solving nonlinear problems numerically. It can be formulated for mappings between very general spaces, classically between Banach spaces, and is thus applicable in particular to problems, involving nonlinear partial differential equations (PDEs). Many PDEs are formulated in weak form as variational problems. Here, we are looking for solutions of the equation F(x)=0 , where F\colon X\to E^{*} maps between a Banach space X and a dual space E^{*} . With the help of test functions e\in E this can be written equivalently as (1) \text{find }x\in X\text{ such that}\quad F(x)e=0\quad\text{ for all }e\in E. Some, but not all, variational problems arise from an energy principle of the form \min_{x\in X}f(x) , resulting in a stationarity condition of the form f^{\prime}(x)=0 with f^{\prime}\colon X\to X^{*} . The class of variational problems, however, is not restricted to mappings between linear spaces. Plenty of interesting problems involve nonlinear spaces, in particular, differentiable manifolds. One of the most well known is the problem of finding a geodesic that connects two given points by minimizing its Dirichlet energy among all connecting curves \gamma\colon[0,T]\to\mathcal{M} on a Riemannian manifold \mathcal{M} : (2) \min_{\gamma}\int_{[0,T]}\frac{1}{2}\lVert\dot{\gamma}(t)\rVert^{2}_{\gamma(t)}\,dt,\quad\gamma(0)=\gamma_{0},\gamma(T)=\gamma_{T}. More generally, harmonic maps minimize a similar energy functional of mappings, defined on more general domains (cf. [ 19 ] ). Other important classes of examples are found in continuum mechanics [ 24 ] , including cosserat materials [ 28 ] and models of inextensible solids [ 27 ] , in micro-magnetism, or in density functional theory [ 3 ] , to name just a few, see also [ 6 , 18 ] for an account on the numerical analysis. In this case, an energy principle of the form \min_{x\in\mathcal{X}}f(x) , where \mathcal{X} is a differentiable, possibly infinite dimensional manifold, will lead to the equation f^{\prime}(x)=0 , where f^{\prime}\colon\mathcal{X}\to T^{*}\mathcal{X} maps into the cotangent bundle T^{*}\mathcal{X} of \mathcal{X} . Slightly more generally, we can formulate variational problems via mappings F\colon\mathcal{X}\to\mathcal{E}^{*} , where \mathcal{E} is a vector bundle, and \mathcal{E}^{*} its so-called dual bundle. Then our problem is of the form (which we will elaborate in detail, below): (3) \text{find }x\in\mathcal{X}\text{ such that }\quad F(x)e=0\quad\text{ for all }e\in E_{y(x)}, where E_{y(x)}\subset\mathcal{E} is the appropriate linear space of test functions, which depends on x . In contrast to ( 1 ), determining the correct space of test functions is part of the problem. In our example of a geodesic, \mathcal{X} is a Sobolev manifold of mappings into \mathcal{M} , and the variational equation is written as follows: (4) 0=\int_{[0,T]}\langle\dot{\gamma}(t),\dot{\delta\gamma}(t)\rangle_{\gamma(t)}\,dt\quad\text{ for all }\delta\gamma\in T_{\gamma}\mathcal{X}. In this work we consider the application of Newton’s method to problems of the form ( 3 ). We build upon the results of [ 35 ] , where Newton’s method for mappings from a manifold \mathcal{X} into a vector bundle \mathcal{E} has been introduced and analyzed. One of the main outcomes of this work is that a linear connection on \mathcal{E} is required to render Newton directions well defined. This generalizes the well known insight that covariant derivatives have to be employed in Newton methods for finding zeros of vector fields on Riemannian manifolds (cf. e.g. [ 15 , 4 , 14 , 23 ] ) or one-forms [ 33 , 32 ] . For an account on Newton’s method for vector fields and further literature we refer to [ 2 , Section 6] . In the semismooth context, Newton’s method for vector fields on Riemannian manifolds has already been discussed in [ 13 , 11 , 31 ] . Newton’s method for shape optimization was considered in [ 30 ] . To obtain convergence from remote initial guesses, Newton’s method is typically equipped with a globalization strategy. Especially in Newton’s method applied in optimization on Riemannian manifolds some strategies including damping and trust region methods were introduced, see e.g. [ 9 , 1 ] . In [ 35 , Sec. 6] we presented an affine-covariant damping scheme. Here, a vector transport on \mathcal{E} is needed in addition, which has to be consistent with the connection in a certain sense. In the current paper, we will discuss, how these concepts can be implemented, if the co-domain of our mapping is a dual vector bundle, so that variational equations on manifolds can be solved. Our paper is structured as follows: we start by a brief recapitulation of the main differential geometric concepts, required to define Newton steps for mappings into general vector bundles in Section 3 . We consider the specific case of Newton’s method for mappings into dual vector bundles, and discuss the differential geometric tools, including dual connections, necessary to define Newton’s method for variational equations. This includes an affine covariant damping strategy that we present for dual vector bundles in Section 4 . As an important special case we consider embedded submanifolds in detail. Finally, we turn to a couple of examples in Section 6 , two of which can be seen as a generalization of the geodesic problem, namely elastic geodesics in a force field and with obstacle avoidance. Finally, as an example from continuum mechanics we consider the simulation of an elastic inextensible rod. 2. Preliminaries In this section we recall some notations, definitions and basic properties of differential geometry and from [ 35 ] used throughout the work. Let \mathcal{X} be a Banach manifold , i. e. a topological space \mathcal{X} and a collection of charts (U,\phi) , where each chart \phi\colon U\to\mathbb{X} maps an open subset U of \mathcal{X} homeomorphically into a real Banach space \mathbb{X} with norm \|\cdot\|_{\mathbb{X}} . Unless otherwise noted we will assume that the manifold is of class C^{1} , cf. [ 21 , II, §1] . We call mappings F\colon\mathcal{X}\to\mathcal{M} between two Banach manifolds continuous, differentiable or locally Lipschitz , if their representations in charts, i. e. \phi_{\mathcal{M}}\circ F\circ\phi_{\mathcal{X}}^{-1} , have the respective property. It can be shown that these properties are independent of the choice of charts. The tangent spaces of a manifold \mathcal{X} , denoted by T_{x}\mathcal{X} for x\in\mathcal{X} , can be defined by using charts, see, e.g., [ 21 , II §2] . We will equip each T_{x}\mathcal{X} with a norm that is equivalent to \lVert\cdot\rVert_{\mathbb{X}} and thus renders T_{x}\mathcal{X} a Banach space. In [ 35 , Sec. 5.1] we showed that, if the mapping x\mapsto\lVert\cdot\rVert_{x} is sufficiently regular, a construction similar to the one, performed on Riemannian manifolds, leads to a metric distance on \mathcal{X} that can be used to study convergence of Newton’s method. A map F\colon\mathcal{X}\rightarrow\mathcal{M} between two C^{1} -manifolds \mathcal{X} and \mathcal{M} is called continuously differentiable , if its representation in charts has the same property. Then this mapping induces a