A multilevel approach for computing the limited-memory Hessian and its inverse in variational data assimilation

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Use of data assimilation techniques is becoming increasingly common across many application areas. The inverse Hessian (and its square root) plays an important role in several different aspects of these processes. In geophysical and engineering applications, the Hessian-vector product is typically defined by sequential solution of a tangent linear and adjoint problem; for the inverse Hessian, however, no such definition is possible. Frequently, the requirement to work in a matrix-free environment means that compact representation schemes are employed. In this paper, we propose an enhanced approach based on a new algorithm for constructing a multilevel eigenvalue decomposition of a given operator, which results in a much more efficient compact representation of the inverse Hessian (and its square root). After introducing these multilevel approximations, we investigate their accuracy and demonstrate their efficiency (in terms of reducing memory requirements and/or computational time) using the example of preconditioning a Gauss-Newton minimisation procedure.
LanguageEnglish
PagesA2934-A2963
Number of pages30
JournalSIAM Journal on Scientific Computing
Volume38
Issue number5
DOIs
Publication statusPublished - 28 Sep 2016

Fingerprint

Data Assimilation
Data storage equipment
Computing
Square root
Eigenvalue Decomposition
Decomposition
Gauss-Newton
Adjoint Problem
Cross product
Requirements
Preconditioning
Engineering Application
Tangent line
Approximation
Operator
Demonstrate

Keywords

  • data assimilation
  • inverse Hessian
  • limited memory
  • preconditioning
  • multigrid

Cite this

@article{3228452cb3dc497ea7d2cc67e7ec8a3d,
title = "A multilevel approach for computing the limited-memory Hessian and its inverse in variational data assimilation",
abstract = "Use of data assimilation techniques is becoming increasingly common across many application areas. The inverse Hessian (and its square root) plays an important role in several different aspects of these processes. In geophysical and engineering applications, the Hessian-vector product is typically defined by sequential solution of a tangent linear and adjoint problem; for the inverse Hessian, however, no such definition is possible. Frequently, the requirement to work in a matrix-free environment means that compact representation schemes are employed. In this paper, we propose an enhanced approach based on a new algorithm for constructing a multilevel eigenvalue decomposition of a given operator, which results in a much more efficient compact representation of the inverse Hessian (and its square root). After introducing these multilevel approximations, we investigate their accuracy and demonstrate their efficiency (in terms of reducing memory requirements and/or computational time) using the example of preconditioning a Gauss-Newton minimisation procedure.",
keywords = "data assimilation, inverse Hessian, limited memory, preconditioning, multigrid",
author = "Brown, {Kirsty L.} and Igor Gejadze and Alison Ramage",
year = "2016",
month = "9",
day = "28",
doi = "10.1137/15M1041407",
language = "English",
volume = "38",
pages = "A2934--A2963",
journal = "SIAM Journal on Scientific Computing",
issn = "1064-8275",
number = "5",

}

A multilevel approach for computing the limited-memory Hessian and its inverse in variational data assimilation. / Brown, Kirsty L.; Gejadze, Igor; Ramage, Alison.

In: SIAM Journal on Scientific Computing, Vol. 38, No. 5, 28.09.2016, p. A2934-A2963.

Research output: Contribution to journalArticle

TY - JOUR

T1 - A multilevel approach for computing the limited-memory Hessian and its inverse in variational data assimilation

AU - Brown, Kirsty L.

AU - Gejadze, Igor

AU - Ramage, Alison

PY - 2016/9/28

Y1 - 2016/9/28

N2 - Use of data assimilation techniques is becoming increasingly common across many application areas. The inverse Hessian (and its square root) plays an important role in several different aspects of these processes. In geophysical and engineering applications, the Hessian-vector product is typically defined by sequential solution of a tangent linear and adjoint problem; for the inverse Hessian, however, no such definition is possible. Frequently, the requirement to work in a matrix-free environment means that compact representation schemes are employed. In this paper, we propose an enhanced approach based on a new algorithm for constructing a multilevel eigenvalue decomposition of a given operator, which results in a much more efficient compact representation of the inverse Hessian (and its square root). After introducing these multilevel approximations, we investigate their accuracy and demonstrate their efficiency (in terms of reducing memory requirements and/or computational time) using the example of preconditioning a Gauss-Newton minimisation procedure.

AB - Use of data assimilation techniques is becoming increasingly common across many application areas. The inverse Hessian (and its square root) plays an important role in several different aspects of these processes. In geophysical and engineering applications, the Hessian-vector product is typically defined by sequential solution of a tangent linear and adjoint problem; for the inverse Hessian, however, no such definition is possible. Frequently, the requirement to work in a matrix-free environment means that compact representation schemes are employed. In this paper, we propose an enhanced approach based on a new algorithm for constructing a multilevel eigenvalue decomposition of a given operator, which results in a much more efficient compact representation of the inverse Hessian (and its square root). After introducing these multilevel approximations, we investigate their accuracy and demonstrate their efficiency (in terms of reducing memory requirements and/or computational time) using the example of preconditioning a Gauss-Newton minimisation procedure.

KW - data assimilation

KW - inverse Hessian

KW - limited memory

KW - preconditioning

KW - multigrid

UR - https://www.siam.org/journals/sisc.php

U2 - 10.1137/15M1041407

DO - 10.1137/15M1041407

M3 - Article

VL - 38

SP - A2934-A2963

JO - SIAM Journal on Scientific Computing

T2 - SIAM Journal on Scientific Computing

JF - SIAM Journal on Scientific Computing

SN - 1064-8275

IS - 5

ER -