Statistical properties of BayesCG under the Krylov prior
Statistical properties of BayesCG under the Krylov prior
We analyse the calibration of BayesCG under the Krylov prior. BayesCG is a probabilistic numeric extension of the Conjugate Gradient (CG) method for solving systems of linear equations with real symmetric positive definite coefficient matrix. In addition to the CG solution, BayesCG also returns a posterior distribution over the solution. In this context, a posterior distribution is said to be ‘calibrated’ if the CG error is well-described, in a precise distributional sense, by the posterior spread. Since it is known that BayesCG is not calibrated, we introduce two related weaker notions of calibration, whose departures from exact calibration can be quantified. Numerical experiments confirm that, under low-rank approximate Krylov posteriors, BayesCG is only slightly optimistic and exhibits the characteristics of a calibrated solver, and is computationally competitive with CG.
239-288
Reid, Tim W.
8ab4ae9b-b21e-4fa4-ba9c-8a4bf9bc7cc9
Ipsen, Ilse C.F.
83eae4c2-19d4-4f74-9d16-4146a63d2c4c
Cockayne, Jon
da87c8b2-fafb-4856-938d-50be8f0e4a5b
Oates, Chris J.
3af13c56-dc47-4d2c-867f-e4e933e74619
December 2023
Reid, Tim W.
8ab4ae9b-b21e-4fa4-ba9c-8a4bf9bc7cc9
Ipsen, Ilse C.F.
83eae4c2-19d4-4f74-9d16-4146a63d2c4c
Cockayne, Jon
da87c8b2-fafb-4856-938d-50be8f0e4a5b
Oates, Chris J.
3af13c56-dc47-4d2c-867f-e4e933e74619
Reid, Tim W., Ipsen, Ilse C.F., Cockayne, Jon and Oates, Chris J.
(2023)
Statistical properties of BayesCG under the Krylov prior.
Numerische Mathematik, 155 (3-4), .
(doi:10.1007/s00211-023-01375-7).
Abstract
We analyse the calibration of BayesCG under the Krylov prior. BayesCG is a probabilistic numeric extension of the Conjugate Gradient (CG) method for solving systems of linear equations with real symmetric positive definite coefficient matrix. In addition to the CG solution, BayesCG also returns a posterior distribution over the solution. In this context, a posterior distribution is said to be ‘calibrated’ if the CG error is well-described, in a precise distributional sense, by the posterior spread. Since it is known that BayesCG is not calibrated, we introduce two related weaker notions of calibration, whose departures from exact calibration can be quantified. Numerical experiments confirm that, under low-rank approximate Krylov posteriors, BayesCG is only slightly optimistic and exhibits the characteristics of a calibrated solver, and is computationally competitive with CG.
Text
2208.03885
- Accepted Manuscript
Restricted to Repository staff only until 12 October 2024.
Available under License Other.
Request a copy
More information
Accepted/In Press date: 14 September 2023
e-pub ahead of print date: 12 October 2023
Published date: December 2023
Additional Information:
Funding Information:
The work was supported in part by NSF Grant DMS-1745654 (TWR, ICFI), NSF Grant DMS-1760374 and DOE Grant DE-SC0022085 (ICFI), and the Lloyd’s Register Foundation Programme on Data Centric Engineering at the Alan Turing Institute (CJO)
Identifiers
Local EPrints ID: 483776
URI: http://eprints.soton.ac.uk/id/eprint/483776
ISSN: 0029-599X
PURE UUID: 625d38f0-5e3d-472f-b525-5db0ecdca530
Catalogue record
Date deposited: 06 Nov 2023 17:34
Last modified: 11 May 2024 02:06
Export record
Altmetrics
Contributors
Author:
Tim W. Reid
Author:
Ilse C.F. Ipsen
Author:
Chris J. Oates
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics