Eigenproblem Parameters
From WikiROMS
Jump to navigationJump to searchEigenproblem Parameters
- NEV
- Number of eigenvalues to compute for the Lanczos/Arnoldi problem. Notice that the model memory requirement increases substantially as NEV increases. The GST requires NEV+1 copies of the model state vector. The memory requirements are decreased in distributed-memory applications.
- option =
- routine = mod_storage.F
- keyword = NEV
- input = roms.in
- NCV
- Number of eigenvectors to compute for the Lanczos/Arnoldi problem. NCV must be greater than NEV.
- option =
- routine = mod_storage.F
- keyword = NCV
- input = roms.in
Notice: At present, there is no a-priori analysis to guide the selection of NCV relative to NEV. The only formal requirement is that NCV > NEV. However in optimal perturbations, it is recommended to have NCV ≥ 2*NEV. In Finite Time Eigenmodes (FTE) and Adjoint Finite Time Eigenmodes (AFTE) the requirement is to have NCV ≥ 2*NEV+1. The efficiency of calculations depends critically on the combination of NEV and NCV. If NEV is large (greater than 10 say), you can use NCV=2*NEV+1 but for NEV small (less than 6) it will be inefficient to use NCV=2*NEV+1. In complicated applications, you can start with NEV=2 and NCV=10. Otherwise, it will iterate for very long time.