search

mesh-adaptation / goalie

Goal-oriented error estimation and mesh adaptation for finite element problems solved using Firedrake
Other
2 stars 0 forks source link

Passing adaptation parameters to `fixed_point_iteration` #203

Closed ddundo closed 5 months ago

ddundo commented 5 months ago

At the moment we are passing adaptation parameters (i.e. those in goalie.options) to MeshSeq, which are then used in fixed_point_iteration methods. Since we will introduce the on-the-fly approach (#88) which might have different parameters, I think it would make more sense to pass these adaptation parameters directly to fixed_point_iteration.

What do you think @jwallwork23? I'm happy to take this on

jwallwork23 commented 5 months ago

Hm maybe, but some of them would surely be in common? e.g., the metric-related ones.

ddundo commented 5 months ago

Sorry, it's not clear to me why metric-related parameters would "surely" be the same? :) I'd have though that it's not unreasonable to set different ones since on-the-fly approach cannot utilise space-time normalisation.

Besides, do we even use these metric parameters anywhere? Aren't the default parameters set in Animate and we always modify them with RiemannianMetric.set_parameters() inside the adaptor function?

jwallwork23 commented 5 months ago

Yeah that's a good point. You're right - in light of our discussion the other day it would probably be good to drop these metric parameters anyway and leave that to Animate.