Hello everyone,
I’m using the ComponentTransport process to perform non-isothermal HTC simulations over large time scales. Only the staggered scheme is currently supported for this process, and my simulations take several weeks to complete.
I’d like to try a monolithic scheme to see if fewer, more expensive iterations could be faster for this buoyancy-driven case.
Are there any plans to implement a monolithic scheme for non-isothermal ComponentTransport processes, or is there a specific reason it hasn’t been implemented yet?
Thank you for your help.
Hello Rose-Nelly,
could you post ogs’ output over several time steps, so one can take better guesses, would you? And if possible, add the project with meshes, so I can try it out locally.
Reasons are historical ones. And my guess would be that the staggered one would be faster.
There are no plans to change it to monolithic at the moment, but maybe if we understand your model better, we can figure out a better solution.
– d
Could imagine to do HT part monolitically, but the chemistry components are much more difficult…
Hello Dmitri and thank you for your answer,
Here you’ll find all the files you need to run the simulations, i’ve also added some output files over several time steps.
This is the .prj that gives me the “fastest” results. i’ve already tried to increase the relative tolerance for each processes. It is the transport process that takes a lot of time.
Hello Rose-Nelly, not so simple problem. We are getting better performance with Anderson accelerations (depth=3 damping=0.05), not yet merged https://gitlab.opengeosys.org/ogs/ogs/-/merge_requests/5714
Question, are you interested in the accurate transient solution or are you looking for a quasi-steady-state results?
– d