Seminar by Kathrin Smetana

Speaker

Kathrin Smetana (Stevens Institute of Technology)

Title

Model order reduction for seismic applications

Date

• March 26, 2024 15:00 CET+0100 (Europe/Rome)

• March 26, 2024 10:00 EDT-0400 (US/Eastern)

• March 26, 2024 09:00 CDT-0500 (US/Central)

• March 26, 2024 07:00 PDT-0700 (US/Pacific)

Abstract

Full waveform inversion to monitor changes in seismicity is a computationally expensive and challenging task. The latter is due to the fact that the discretization of the seismic wave equation can have millions of degrees of freedom. Moreover, aiming at estimating, for instance, the elastic structure at every grid point results in a large parameter space within the inverse problem. Model order reduction (MOR) techniques can help to speed up the computations, using low-dimensional models that capture the original system’s important features. However, for large-scale wave propagation problems, constructing efficient reduced models is challenging as MOR methods can suffer from a slow decay of the Kolmogorov n-width for such problems, thus, requiring a large number of basis functions to reach the desired accuracy. In our work, we address the mentioned challenge as follows: We transform the problem to the Laplace domain, where we can exploit that the output of interest – the seismogram – is band-limited or low-pass filtered, such that we can avoid high frequencies; the latter would require many reduced basis functions for an accurate approximation. By targeting the construction of the reduced model to the seismogram, we obtain a rapidly converging reduced order approximation.

Recording

Watch the recording on our YouTube channel.