Multiscale II: analysis and efficient simulation of multiscale problems.

Abstracts for the talks

A new approach to variational problems with multiple scales

Giovanni Alberti   (Pisa)
Wednesday, October 08, 1997
Many problems in science involve the creation and interaction of microstructures on multiple spatial or temporal scales. Most mathematical approaches have addressed "infinitely fine" microstructures or the case of a single small scale. In this talk we propose a new idea to deal with multiple scales and illustrate it with some simple examples. This is joint work with S. Müller.

Weak convergence methods and adiabatic results in classical and quantum mechanics

F. Bornemann   (ZIB)
Thursday, October 09, 1997
We will show that weak convergence methods are conveniently suited to explicitly access singular limits of a certain family of mechanical systems with multiple time scales. This family turns out to be characterized by the existence of sufficiently many adiabatic invariants.
The key step is the idenfication of the weak limits of all those quadratic quantities which carry important information for the limit system. This idenfication becomes possible by a weak convergence analogue of the Virial Theorem, resulting in certain matrix commutation relations.
We will address natural mechanical systems with strong constraining potentials and the adiabatic theorem of quantum mechanics. These examples will show that our approach considerably extends the possibility of passing through resonances.

Numerical analysis of the (relaxed) double well problem

Carsten Carstensen   (Kiel)
Thursday, October 09, 1997
The mathematical modelling of phase transitions in alloys, or optimal design of composites, is concerned with non-convex variational problems, where a minimum may or may not be attained. Young's model example, the two-well or double-well problem (P), is considered in this talk. Convexification yields a relaxed problem (RP) related to (P). The loss of information we face in recasting (P) as (RP) is not substantial as long as we are interested in the global displacements, stress fields and Young measures (which describe oscillations creating microstructures); these variables defined by (P) can be computed from solutions of (RP).
Typically, the problem (RP) is convex, but not strictly convex. Hence error estimates for finite element methods are more difficult to prove than for simpler uniformly convex problems. A priori and a posteriori error estimates are presented and illustrated in numerical examples which indicate that it is more reasonable to perform a numerical analysis of (RP) rather than of (P).

The approximation of complicated dynamical behavior

M. Dellnitz   (Bayreuth)
Wednesday, October 08, 1997
Multilevel techniques are presented for the efficient numerical approximation of complicated dynamical behavior. Concretely we develop (adaptive) methods which allow to extract statistical information on the underlying dynamical system. This is done by an approximation of natural invariant measures as well as (almost) cyclic dynamical components. We discuss issues concerning the implementation (e.g. parallelization strategies) and indicate potential applications of these methods (e.g. to the computation of Lyapunov exponents). The results are illustrated by several numerical examples.

Essential molecular dynamics: Progress in a new algorithmic approach

P. Deuflhard   (ZIB, )
Wednesday, October 08, 1997
Recently, the author -- together with his co-authors Michael Dellnitz, Oliver Junge, and Christof Schütte -- developed a new algorithmic approach to molecular dynamics, which is based on the computation of almost invariant sets of Hamiltonian systems. In this approach, only well-conditioned short term subtrajectories in lieu of ill-posed long term trajectories (which are typically used in Monte Carlo simulations) are exploited. The aim is to directly compute averages of physical observables, comformations and conformational changes - informations that are actually desired by computational chemistry. Mathematically speaking, such informations come out of the computation of the invariant measures and sets (corresponding to the dominant eigenvalue 1 and infinite relaxation time) and almost invariant measures and sets (corresponding to eigenmodes for eigenvalues close to 1 and therefore finite, but "large" relaxation times). An adaptive multilevel box method or subdivision technique is presented, which helps to solve the arising stochastic eigenvalue problem "fast". The basic concepts of the new algorithm and recent speed-ups will be presented.

Energy density for the wave equation on thin domains

Gilles Francfort   (Paris)
Wednesday, October 08, 1997
In this joint work with P. Gérard, 3-d/2-d asymptotic analysis of the wave equation is undertaken for cylindrical domains with vanishing thickness. The limit equation is a 2-d wave equation but its energy density is not the weak-limit of its 3-d analogue. The computation of that limit is performed microlocally through the computation of the semi-classical measure associated to the space-time derivatives of the field.

Flow properties for Young-measure solutions of semilinear hyperbolic systems

Alexander Mielke   (Hannover)
Friday, October 10, 1997
We consider a semilinear hyperbolic system of d equations of first order. We consider sequences of initial data which approximate a Young measure and study the Young-measure limit of the associated sequence of solutions. Thus we are able to define the notion of Cauchy problem and semigroups for Young measure solutions, at least for tex2html_wrap_inline8 by using compactness through compensation. For tex2html_wrap_inline10 the subclass of product measure solutions is considered, where each Cauchy problem has a unique solution. Continuous dependence of the associated semigroup on the initial data can be established in the Wasserstein topology and, under additional structural condition on the nonlinearity, also in the weaktex2html_wrap_inline12 topology. The reported research extends previous work of L. Tartar and is partly joint work with Florian Theil.

Adiabatic invariants and their preservation under symplectic discretizations

S. Reich   (ZIB)
Thursday, October 09, 1997
We will apply normal form theory to highly oscillatory Hamiltonian systems and discuss the preservation of adiabatic invariants over exponentially long periods of time. Next, the conservation of adiabatic invariants under a symplectic discretization will be considered. Backward error analysis will provide us the appropriate tool to show that, again, the invariants are preserved over exponentially long periods of time.

The amplitude equation for degenerate subcritical bifurcations in pattern formating systems

G. Schneider   (Hannover)
Friday, October 10, 1997
We consider a model for a pattern formating system on the infinite line which leads to a subcritical bifurcation. For the degenerate case we use multiple scaling analysis to derive


as an amplitude equation for the envelope A of the bifurcating pattern which is modulated slowly in time and in space. We show exact estimates between the approximations obtained via the amplitude equation and true solutions of the original system. Moreover, we show that every small solution of the original system develops in such a way that it can be described after a certain time by the solutions of the amplitude equation. The difficulty is to show the estimates on an tex2html_wrap_inline20-time scale in contrast to tex2html_wrap_inline22 for the classical Ginzburg-Landau equation, if tex2html_wrap_inline24 is the order of the amplitude. This theory allows the description of modulated N-pulses in the original system.

Multiple scales and nonadiabatic effects in molecular quantum dynamics

Christof Schütte   (ZIB)
Thursday, October 09, 1997
The full quantum dynamical (QD) description of molecules includes multiple scales in particular if particles of essentially different masses are included. This makes full QD simulations of larger molecules simply impossible. In real life molecular dynamics one deals with this problem by introducing simplified "mixed quantum-classical" models which describe the heavy particles of the molecule by the means of classical mechanics and only a small portion by the means of a wavefunction. This talk will present two of these simplified models. By using the techniques of the previous talk "Weak Convergence Methods and Adiabatic Results in Classical and Quantum Mechanics", it will be discussed under which circumstances these models gain a suitable approximation of the true QD solution. The discussion will be concentrated on the understanding of the nonadiabatic effects which may cause such simplified descriptions to fail and on the alternatives which may prevent such failure.

H-measures and their variants

Luc Tartar   (CMU/Leipzig)
Wednesday, October 08, 1997
When a sequence converges weakly to 0 in tex2html_wrap_inline12 but not strongly (tex2html_wrap_inline14 being a subset of tex2html_wrap_inline16), one says that it contains oscillations or concentration effects depending upon if the limit of its square has a N-dimensional density or not. One quantitative way to analyze these oscillations or concentration effects is to use H-measures (which I have introduced for a few applications and which Patrick GÉRARD has introduced independently for other ones); they are measures on tex2html_wrap_inline18 which have a quadratic microlocal character (and make more precise the quadratic theorem of Compensated Compactness Theory).

For dealing with problems with one characteristic length, Patrick GÉRARD then introduced a variant, called semi-classical measures, living on tex2html_wrap_inline20 (quite similar to the idea which I had proposed independently to use H-measures after adding one dimension). It was then shown by Pierre-Louis LIONS and Thierry PAUL that the semi-classical measures could be introduced using WIGNER transform. I then found with Patrick GÉRARD how to use correlations instead, and why these objects are not good enough when at least two scales are present.
Some other variants of H-measures can be defined, adapted to different questions, and I will describe the advantages and defects of some of these variants.

Date and Location

October 08 - 10, 1997
Max Planck Institute for Mathematics in the Sciences
Inselstraße 22
04103 Leipzig
see travel instructions

Scientific Organizers

P. Deuflhard

P. Dolzmann

Stefan Müller

05.04.2017, 12:42