search

uwoseis / zephyr

Open-source seismic waveform modelling and inversion code written in Python
https://zephyr.space
MIT License
18 stars 8 forks source link

Correcting discretization for Eurus #8

Closed bsmithyman closed 8 years ago

bsmithyman commented 8 years ago

Issue by bsmithyman Tuesday Nov 17, 2015 at 19:51 GMT Originally opened as https://github.com/uwoseis/anemoi/issues/8

Issue created to discuss Eurus testing & corrections.

bsmithyman commented 8 years ago

Comment by bsmithyman Tuesday Nov 17, 2015 at 19:54 GMT

At #7, @ShaunHadden said:

I've made the change to the 'Dz' calculation that was wrong on the timedomain branch for eurus.py. This still needs more work so perhaps we should wait to merge it into master?

bsmithyman commented 8 years ago

Comment by ShaunHadden Tuesday Nov 17, 2015 at 19:55 GMT

Sorry!

bsmithyman commented 8 years ago

Comment by ShaunHadden Tuesday Nov 17, 2015 at 19:56 GMT

I am currently going through the paper Gerhard sent us and comparing it to our discretization to see if I can spot the error that's causing the amplitudes to be unstable

bsmithyman commented 8 years ago

Comment by bsmithyman Tuesday Nov 17, 2015 at 19:56 GMT

So yeah, I think that the change on the Dz term in the stencil could probably be merged in now. It's a bugfix anyway, so it's not like it's going to break anything outside Eurus. However, if it's hard to separate it from the other commits we can wait, I guess. Is it amenable to being cherry-picked? I seem to recall that there are a number of other changes to different files on that branch, and I think we're going to need a more complex merge.

bsmithyman commented 8 years ago

Comment by bsmithyman Tuesday Nov 17, 2015 at 19:57 GMT

Sorry! No worries :)

Okay, good. Why don't you work on that, and then think about a good way to make an atomic change vs. master (i.e., don't drag in all the other stuff from the timedomain branch yet).

bsmithyman commented 8 years ago

Comment by ShaunHadden Tuesday Nov 17, 2015 at 19:58 GMT

feel free to just 'cherrypick' the one change, I don't think anything else is required just yet

bsmithyman commented 8 years ago

@ShaunHadden I ran a script to import this issue from anemoi

bsmithyman commented 8 years ago

To summarize: In meeting with @ShaunHadden and Gerhard, we found some mistakes in Operto (2009) that resulted in amplitude errors when modelling TTI. This was resolved in 9c0d4b641abcf80e5363abce3cc7b2a3cb83897d.

bsmithyman commented 8 years ago

Outstanding issue: velocity dispersion exists in Eurus solutions when compared to the analytical Helmholtz equation in stretched coordinates.

See Compare Solutions - Homogenous (Eurus)

This makes quite a noticeable difference for gradient computation; cf. MiniZephyr and Eurus.

bsmithyman commented 8 years ago

Eurus problems were dealt with; see e.g., 161af4e