Efficient stochastic Hessian estimation for full waveform inversion

In this abstract we present a method that allows arbitrary elements of the approximate Hessian to be estimated simultaneously. Preliminary theoretical and numerical investigations suggest that the number of forward models required for this procedure does not increase with the number of shots. As the...

Full description

Bibliographic Details
Main Authors: Willemsen, Lucas Abraham, Malcolm, Alison E., Hewett, Russell J.
Other Authors: Massachusetts Institute of Technology. Earth Resources Laboratory
Format: Technical Report
Language:en_US
Published: Massachusetts Institute of Technology. Earth Resources Laboratory 2014
Subjects:
FWI
Online Access:http://hdl.handle.net/1721.1/90528
_version_ 1826201370502365184
author Willemsen, Lucas Abraham
Malcolm, Alison E.
Hewett, Russell J.
author2 Massachusetts Institute of Technology. Earth Resources Laboratory
author_facet Massachusetts Institute of Technology. Earth Resources Laboratory
Willemsen, Lucas Abraham
Malcolm, Alison E.
Hewett, Russell J.
author_sort Willemsen, Lucas Abraham
collection MIT
description In this abstract we present a method that allows arbitrary elements of the approximate Hessian to be estimated simultaneously. Preliminary theoretical and numerical investigations suggest that the number of forward models required for this procedure does not increase with the number of shots. As the number of shots increases this means that the cost of estimating these approximate Hessian entries becomes negligible relative to the cost of calculating the gradient. The most obvious application would be to estimate the diagonal of the approximate hessian. This can then be used as a very inexpensive preconditioner for optimization procedures, such as the truncated Newton method.
first_indexed 2024-09-23T11:50:56Z
format Technical Report
id mit-1721.1/90528
institution Massachusetts Institute of Technology
language en_US
last_indexed 2024-09-23T11:50:56Z
publishDate 2014
publisher Massachusetts Institute of Technology. Earth Resources Laboratory
record_format dspace
spelling mit-1721.1/905282019-04-12T22:15:05Z Efficient stochastic Hessian estimation for full waveform inversion Willemsen, Lucas Abraham Malcolm, Alison E. Hewett, Russell J. Massachusetts Institute of Technology. Earth Resources Laboratory FWI In this abstract we present a method that allows arbitrary elements of the approximate Hessian to be estimated simultaneously. Preliminary theoretical and numerical investigations suggest that the number of forward models required for this procedure does not increase with the number of shots. As the number of shots increases this means that the cost of estimating these approximate Hessian entries becomes negligible relative to the cost of calculating the gradient. The most obvious application would be to estimate the diagonal of the approximate hessian. This can then be used as a very inexpensive preconditioner for optimization procedures, such as the truncated Newton method. 2014-10-02T14:47:44Z 2014-10-02T14:47:44Z 2013 Technical Report http://hdl.handle.net/1721.1/90528 en_US Earth Resources Laboratory Industry Consortia Annual Report;2013-34 application/pdf Massachusetts Institute of Technology. Earth Resources Laboratory
spellingShingle FWI
Willemsen, Lucas Abraham
Malcolm, Alison E.
Hewett, Russell J.
Efficient stochastic Hessian estimation for full waveform inversion
title Efficient stochastic Hessian estimation for full waveform inversion
title_full Efficient stochastic Hessian estimation for full waveform inversion
title_fullStr Efficient stochastic Hessian estimation for full waveform inversion
title_full_unstemmed Efficient stochastic Hessian estimation for full waveform inversion
title_short Efficient stochastic Hessian estimation for full waveform inversion
title_sort efficient stochastic hessian estimation for full waveform inversion
topic FWI
url http://hdl.handle.net/1721.1/90528
work_keys_str_mv AT willemsenlucasabraham efficientstochastichessianestimationforfullwaveforminversion
AT malcolmalisone efficientstochastichessianestimationforfullwaveforminversion
AT hewettrussellj efficientstochastichessianestimationforfullwaveforminversion