Hello, I have added 8 new reactions based on the CO2 absorption example. As for the code, I only added the new reaction rates calculation, but did not change much.Finally, the output rates become very very slow. Why does the running speed get slower and slower?
Thanks for your help!
case.zip (83.0 KB)
Hi @Chen_Boyan
I ran your case for a few hours, here’s a plot of simulation time vs. real time:
and here’s a plot of timestep (dt) vs simulation time:
The timestep fluctuates a bit, and as you can see the slope of the simulation time vs real time does change. But I don’t see a drastic slowdown. How long did you run the case before observing a slowdown?
When processes suddenly become extremely slow, this is often due to running out of memory and accessing disk (swap) instead. However, watching with ‘top’, I did not observe the memory usage increasing at all over the time span that I ran this.
– Charles
Hi Charles,
Thanks for your help!
yesterday,I ran this case again,This is the output time.You can see the speed is getting slower and slower. This is even more obvious after ALL_20 (simulating the real time of 10s).It even takes 3 hours to compute 1s at last. I used 32 processes in parallel.
I am calculating the CO2 absorption example ，It is also getting slower and slower
Please check the console output. Does the time step drop significantly? Do you spend a lot of time in the stiff chemistry solver?
Plotting ‘dt’ and ‘simulation time’ is also useful (see above)
Hi Charles,
This is the plot of my simulation.I used 28 processes in parallel.I only have 9480 cells,and I also tried 2 processes、4processes、and 8 processs,the case of 28 processes is more quickly.But as you can see from the plot,it’s still going to get slower and slower.Based on my tests, it’s going to get really slow. How can I solve this problem?
Please attach all the .OUT and .LOG files.
The .LOG file contains a lot of useful info. I extracted the lines matching
Time = 0.10000E-03 Dt = 0.10000E-03
as well as the lines reporting CPU time used by the Eulerian stiff chemistry solver.
Here are some plots:
dT is steadily decreasing during your run, which is why it’s slowing down. The time used by the chemistry solver also seems to increase after about 10.5 s, I’m not sure of the exact reason for this but I don’t think it’s the main reason for the slowdown.
MFiX will decrease the timestep (dT) when the simulation is failing to converge.
Notice that the log file contains a large number of messages of the form:
t= 11.5780 Dt= 0.1501E-03 NIT= 80MbErr%=-0.2580E-03: Run diverged/stalled :-(
that :-( means that the solver is not happy.
There are several common reasons for non-convergence, including poor mesh quality. See the FAQ section:
https://mfix.netl.doe.gov/doc/mfix/22.2/html/faq.html#what-do-i-do-if-a-run-does-not-converge
Hi Charles,
I recently turned off the reaction and found that it still slows down.I use TFM, but you know the entry can’t be 100% solid, it has to be both fluid and solid.When I set the feed style to point source, it also slows down.When it’s only fluid(gas), it doesn’t slow down.
Therefore, I guess that after the gas-liquid mixture enters, the gas moves fast, and it escapes, leaving only solid, so it is difficult to converge.
Do I have a reasonable guess? How can I solve this problem?
Yes, I think you are on the right track. As we suspected it’s not the reactions that are causing this, it’s the “:-(” unhappy solver failing to converge.
Have you consulted the documentation referenced? This type of problem is often due to poor mesh quality.
Hi Charles,
Thanks for your help!
What I want to simulate is the liquid entering from the bottom inlet. But as you know, we don’t allow 100%solid, so I also plugged in some gas. I would like to ask whether the reason for the disconvergence is that when the gas and liquid enter, the gas quickly escapes, and the remaining solid volume fraction becomes unreasonable, and the gas-solid coupling is unreasonable.
I have read some papers and some cases calculated with MFIX. They usually have 100% GAS entering from the bottom inlet. Can MFIX be used to simulate the liquid entering from the bottom?
Hi Jeff,
Thanks for your help!
I have some questions.
1.In the original CO2 absorption example, there was a packed, and the packed(solid2) was closed the momentum equation. At present I have not removed the packed,I’m just going to make the volume fraction very small, while keeping the momentum equation closed, will the presence of the packed affect my flow?Could this be the reason for the slow computation?
At the same time, in the presence of the packed, one of my reaction products is solid2, can it calculate normally?
2.I want to enter the liquid from the bottom, the bottom inlet and the point source, which feeding way is better?What is the appropriate division of fluid and solid volume fractions?