[petsc-users] Poor Convergence for Non-Linear Least Squares using NTR

[Tristan Konolige]

Hello All,

I’m trying to optimize a nonlinear least squares problem using PETSc. I’d like to use Levenberg-Marquardt, but, as it is not available, I am using Tao’s NTR with J^T J as the hessian. The NTR is converging very slowly (compared to an LM solver in Ceres — a nonlinear optimization package).

How can I debug this poor performance? Is there something I am missing for solving nonlinear least squares in PETSc?

Thanks,
Tristan Konolige
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20181102/a7f76536/attachment.html>