Un-reduction

From MaRDI portal



Abstract: This paper provides a full geometric development of a new technique called un-reduction, for dealing with dynamics and optimal control problems posed on spaces that are unwieldy for numerical implementation. The technique, which was originally concieved for an application to image dynamics, uses Lagrangian reduction by symmetry in reverse. A deeper understanding of un-reduction leads to new developments in image matching which serve to illustrate the mathematical power of the technique.


The authors give a complete geometrical descriptions of a new technique called ``un-reduction for solving dynamical and optimal problems posed on spaces that are unwieldy for numerical implementation. Given two smooth manifolds \(M\) and \(N\), the shape space of \(M\)-type submanifolds of \(N\) may be identified with the quotient space \(\Sigma:=\text{Emb}(M,N)/\text{Diff}(M)\), that is, the space of embeddings of \(M\) into \(N\) up to re-parametrization of \(M\). An example of particular interest is the shape space of all closed, simple, planar unparametrized curves, given by \(\Sigma=\text{Emb}(S^{1},\mathbb{R}^{2})/\text{Diff}(S^{1}).\)NEWLINENEWLINEIn order to compare two shapes, a natural way to proceed is to construct a path between them. This is known as the matching problem.NEWLINENEWLINEAs an example, the matching problem for closed, simple planar curves may be stated as follows: Let \(\rho_{0}\) and \(\rho_{1}\) be two smooth submanifolds in \(\mathbb{R}^{2}\) of \(S^{1}\)-type. Find a path of submanifolds of \(S^{1}\)-type, \(\rho(t),\) such that \(\rho(0)=\rho_{0}\) and \(\rho(1)=\rho_{1}.\) The preferred path is selected in such a way that it minimizes the functional \(\displaystyle{\int_{0}^{1}l^{\Sigma}(\rho,\dot{\rho})dt}\) subjected to the boundary conditions \(\rho(0)=\rho_{0}\) and \(\rho(1)=\rho_{1},\) where \(\Sigma\) denotes the space of all submanifolds of \(S^{1}\)-type in the plane and \(l^{\Sigma}:T\Sigma\rightarrow\mathbb{R}\) is a Lagrangian defined on the tangent bundle of \(\Sigma.\)NEWLINENEWLINEThe solution to the matching problem satisfies the Euler-Lagrange equations NEWLINE\[NEWLINE\displaystyle{\frac{d}{dt}\frac{\delta l^{\Sigma}}{\delta\dot{\rho}}-\frac{\delta l^{\Sigma}}{\delta\rho}=0}.NEWLINE\]NEWLINENEWLINENEWLINEIn order to make sense for the relevant Euler-Lagrange equations, \(\Sigma\) must be endowed with a smooth manifold structure.NEWLINENEWLINEThe aim of this article is to investigate alternative formulations of the matching problem in such a way that their solutions are tractable to a practical implementation.











This page was built for publication: Un-reduction

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q420290)