Output details

An adjoint for likelihood maximization

Significance of output:

The process of likelihood maximization can be found in many areas of computational modeling and the methods in this article make this process faster by orders of magnitude. Likelihood maximization requires the solution of many expensive matrix factorizations. This article derives an adjoint formulation that yields the gradients of the likelihood at a cost almost independent of the number of parameters in the model and so methods previously limited to tens of parameters can now be applied to problems with dimensions many orders of magnitude larger. The method has been integrated into Rolls-Royce plc design software as part of the £17million CFMS project (contact: shahrokh.shahpar@rolls-royce.com).