The course aims at the numerical computation of analytic expansions of invariant objects of Quasiperiodically Forced Skew Products. Due to the usual complicate geometry of these objects we are condemned to use wavelet expansions instead of a more friendly ones. To do this we will develop all the related dynamical and functional analysis machinery.
In some very special cases we can determine theoretically the regularity of the invariant object, thus obtaining a “cotton check” of the quality of our computations: The theoretical and numerical regularity must agree. To be able to do this we will also introduce appropriate notions of regularity and results to compute them from the approximate wavelets coefficients.
Session I (2 hours)
• Motivation and broad perspective of the course
• A forecast of the problems we will encounter and an priori “waving hands” justification of the necessity of the adopted solutions. In particular, the course of big data aka why we need 230 coefficients
• A brief introduction to Quasiperiodically Forced Skew Products in low dimension and their invariant objects
• ANCE’s and why they are horribly complicate
• A naive approach to the semi-analytic approximation of invariant objects: truncated expansions in basis of the ambience space
• Why we do not use Fourier
Session II (2 hours)
• Algebraic reformulation of the problem with the fundamental role of the Invariance Equation
• The easy case: expansions in Haar basis.
• The benefit of pre-conditioners (when we are able to find an easy one)
• An algorithm
• Examples
Session III (2 hours)
• A crash course on wavelets: Multiresolution Analysis
• Compact support or not? This is the question
• Fast Wavelet Transform (FWT) the easy but useless (for us) approach
• Notions of regularity. Hölder, Besov, Sobolev and Calderon-Zygmund spaces
• Theoretical computations of regularity: The case of (upper) semicontinuous functions.
• Wavelet coefficients and regularity. Triebel Theorem.
Session IV (2 hours)
• Expansions in Daubechies Wavelets
• The algebraic reformulation of the problem adapted to Daubechies Wavelets. The Daubechies wavelets matrix and the rotated Daubechies wavelets matrix
• Pre-conditioners are still good but dirty (at the level of errors in the wavelets matrices)
• An algorithm to massively compute wavelets coefficients
• A “waving hands” approach to the solution of the linear systems of Newton Method in super-very-high-dimension. Few words on the Transpose-Free Quasi-Minimal Residual (TFQFMR) algorithms. A justification “Transpose Free” part
Session V (1 hour)
• Computation of the wavelet matrices: a paranoia
• Evaluating Daubechies wavelets at a point (a difficult task): Daubechies-Lagarias Algorithm with a Vidakovich contribution
• A “waving hand” approach to the efficient evaluation of rotated Daubechies wavelets