NobleBlocks

Research index

A Memory Efficient Adjoint Method to Enable Billion Parameter Optimization on a Single GPU in Dynamic Problems [Software]

Herrmann, Bürchner, Kudela and 1 other

Zenodo (CERN European Organization for Nuclear Research)19 Sep 2025Open access

NobleIDark:/48914/w1/20260525/583D25C0DOI10.5281/zenodo.17157435

Free to read at the publisher

You can read the abstract here, or the full text at the publisher.

Read at the publisher

Abstract

This is the representative code used in the following publication: Herrmann, L., Bürchner, T., Kudela, L, Kollmannsberger, S., A Memory Efficient Adjoint Method to Enable Billion Parameter Optimization on a Single GPU in Dynamic Problems, in review, September 2025 Driver files are provided for a full waveform inversion (FWI) example:- createFWISquare2D.py defines the specific problem: the problem size can be modified by changing the factor- generateDataFWISquare2D.py computes the measurement data at the sensor locations with a higher resolution (avoiding the inverse crime): set by inverseCrimeFactor- invertFWISquare2D.py performs the full waveform inversion using Adam or L-BFGS-B as optimizer- referenceGradientFWISquare2D.py computes the gradient with standard adjoint sensitivity analysis (can be compared to the gradient in invertFWISquare2D.py)and a transient acoustic topology optimization (TATO) example- createTATOSquare2D.py defines the specific problem- optimizeTATOSquare2D.py performs the optimization using Adam as optimizer- simulateWaveFieldTATOSquare2D.py computes the transient forward solution given an optimized design- referenceGradientTATOSquare2D.py computes the gradient with standard adjoint sensitivity analysis (can be compared to the gradient in simulateWaveFieldTATOSquare2D.py) Helper functions are contained in the helper subfolder, which also contain the CUDA kernels (.cu) in the kernel subfolder. Before running the drivers a measurement folder must be created.

Discussion

Topics

More

Suggest a correction or report