Changes made to allow GRAMM deterministic mode

[JoshLovesFun]

Requesting review of changes. Initial testing shows deterministic results.
See example input file attached with new option on the last line of "IIN.dat" file.
IIN.txt

[MarkusKuntner]

The modifications are plausible and can be compiled without any problems.
GRAMM is not completely deterministic either, because the same numerical problems occur here as with the GRAL wind field. In earlier versions, the solver was even parallelized with Parallel.For, which repeatedly led to aborted calculations in our applications due to numerical problems.
Due to the division into blocks, the synchronization of the edge columns of the blocks (only contained in GRAMM but not in GRAL) at many iterations and the low relaxation factors and the rounding at the end, however, it can look as if the model would run deterministically.
My work in GRAMM was comparatively little and only for technical aspects (corrections of parallization, automatic time step control, performance optimization, solver restructuring).

[JoshLovesFun]

Thanks for explaining. I have not done extensive testing with GRAMM, but I have confirmed the modifications appear to work as expected for a simple test case. If you think these changes can be merged into a new branch of the GRAMM repository, feel free to do so. I am not sure I have permission to create a new branch.

I am not sure if GRAMM is still in active development, or if changes/additions are less encouraged here. Over the coming months, I may have time to look into some aspects of the GRAMM code, and do some refactoring. But if you think this is not something I should spend any time with, let me know.

[MarkusKuntner]

I'm not sure that individual changes will have much effect here. After talking to one of the co-developers, there are several approaches for improvement, but there is no funding at the moment.
According to his tests, GRAMM_SCI is partly an enhancement, but he discovered various new issues.
In my humble opinion, GRAMM is a kit full of physics, but like some of the algorithms it contains, it dates back to the late 80s and early 90s. In addition, the physics are very simplified so that, for example, the sea breeze cannot be calculated correctly. If ECMWF and WRF are used as initialisation for GRAMM, acceptable results can be obtained.
An email conversation may be more appropriate for more in-depth discussions.

[JoshLovesFun]

Okay, thank you for explaining the overview.

…

On Sun, Aug 11, 2024, 12:48 PM Markus ***@***.***> wrote: I'm not sure that individual changes will have much effect here. After talking to one of the co-developers, there are several approaches for improvement, but there is no funding at the moment. According to his tests, GRAMM_SCI is partly an enhancement, but he discovered various new issues. In my humble opinion, GRAMM is a kit full of physics, but like some of the algorithms it contains, it dates back to the late 80s and early 90s. In addition, the physics are very simplified so that, for example, the sea breeze cannot be calculated correctly. If ECMWF and WRF are used as initialisation for GRAMM, acceptable results can be obtained. An email conversation may be more appropriate for more in-depth discussions. — Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BA4S7FFFMDY3JTRUVTUV6TDZQ6IVVAVCNFSM6AAAAABLUQSHAKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEOBSHAZDEMJWHE> . You are receiving this because you authored the thread.Message ID: ***@***.***>