1887

Abstract

Full waveform inversion (FWI) is typically a large-scale optimization problem. It generally relies on efficient local optimization methods. The importance of accounting for the inverse Hessian matrix is demonstrated to scale the parameters and accelerate the convergence. However, it is unfeasible to compute directly the Hessian or its inverse due to high computational costs and large memory requirements. Quasi-Newton and Hessian-free inexact Newton (HFN) are usually used to solve these difficulties. But the Hessian information they collected is either limited or wasted. This study interlaces both L-BFGS and HFN methods, designs a new scheme that adjusts dynamically the length of each method, and proposes a hybrid L-BFGS/HFN iterative optimization method. It aims at fully utilizing the Hessian information generated in the two methods. Numerical FWI experiments show that it converges more rapidly than L-BFGS. It also runs faster than HFN and the enriched method.

Loading

Article metrics loading...

/content/papers/10.3997/2214-4609.20130603
2013-06-10
2024-03-29
Loading full text...

Full text loading...

http://instance.metastore.ingenta.com/content/papers/10.3997/2214-4609.20130603
Loading
This is a required field
Please enter a valid email address
Approval was a Success
Invalid data
An Error Occurred
Approval was partially successful, following selected items could not be processed due to error