-
Notifications
You must be signed in to change notification settings - Fork 8
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[WIP] Splice multilevel #112
base: master
Are you sure you want to change the base?
Conversation
As of 856f1b0, this is what the result of splicing a field |
Note for the future: when converting from delta space to whitenoise and vice-versa, we loose the fundamental mode (which is set to 0). This result in the spliced field to always have a mean value of 0 (for each grid) and causes it to differ by a constant from the target field. A workaround would be to prevent the conversion to happen in the first place, i.e. to do the splicing at the location where the conversion from whitenoise to delta is done. |
Shouldn't the filtering make this irrelevant? All long-wavelength modes should be coming from the parent grid, not the zoom |
My understanding is that, in non-spliced situation, the fields are in whitenoise space until a call to For splicing, we solve for Overall, this leads to a small offset between the spliced field and the target field in the zoomed region, as can be seen below. The difference should be of the order of 1e-14, but is instead exactly the mean value of the field |
My point is that the k=0 should not be contributing, because it should be filtered in from the higher level grids. So I don't quite understand this issue. I am also nervous about doing splicing after everything else is complete. I have a sense that this will lead to confusing behaviour -- e.g. what if you try to modify something that you spliced? Ideally we just rewrite the splicing operation to apply in noise space, as I did in the original one-level case. |
This allows for the matrix to be non definite-positive (but still symmetric)
(Tipsy file)
The MINRES iterations are however still not converging
This allows the splicing to output a field directly in delta space without needing to convert back into whitenoise.
The accuracy in the zoomed window is not perfect yet (order of 10^-3 only)
d8c7511
to
d6c50a3
Compare
This is a work in progress to implement the splicing operation for multi-level setup.