"ITERATION ON SOLUBILITY FAILED" in restart
Hi All,
I am running TOUGH3 EOS7C for gas production and want to restart the simulation after some time. I rename the SAVE file as INCON (keep the +++ lines as I want to continue with time scale) and modified the TIMES. However, I got the error below (also attached a screenshot).
QQQQQSUBROUTINE QU QQQQQQQQQQQQQ --- [KCYC,ITER] = [ 328, 1]
...ITERATING... AT [ 328, 1] --- DELTEX = 0.100000E+05 MAX. RES. = 0.149866+278 AT ELEMENT 01A68 EQUATION 6
EEEEEEEEEEEEEE#1 ITERATION ON SOLUBILITY FAILED. EEEEEEEE
XMOLE2AQ,XMOLENCG: NaN NaN
XMOLE3AQ,XMOLE: NaN NaN
TOUGH STATUS: Failed in EOS. IGOOD= 2
++++++ REDUCE TIME STEP AT ( 328, 1) ++++++ NEW DELT =0.250000E+04
The time step (DELT) keeps decreasing and the simulation will finally stop.
Can anyone please point me the problem?
Thanks in advance,
Haiyan
7 replies
-
I think the error is related with the big residual "MAX. RES. = 0.149866+278". Any idea how I get so big residual?
Thanks,
Haiyan
-
The log file show two problem (1) MAX RES is extremely large, (2) no convergence in solubility model. Please take a look which happen first. The solubility convergence problem has been over there for a long time. To solve the solubility convergence problem, you may give the small amount for all components through modifying the primary variables. In your case, you may manually modify the initial condition (INCON) of gridblock "01A68" (from the output log we know this gridblock has the maximum residual and cause the problem) .
-
Kenny
Thanks for your reply.
I do have small values for those absent variables in the INCON file. Besides, the grid block "01A68" shows no difference than the neighboring blocks. Please see the attached screenshot.
Haiyan
-
-
The solubility model uses a iterative approach to solve the solubility of gas components. A tiny difference of primary variables may lead to no convergence of the iteration.
-
I tested restart in another model and it works in serial mode but not in parallel mode. Any idea what is causing the problem?
Thanks,
Haiyan
-
In Jacobian matrix setup, the contribution of flux between two elements A and B is needed. If A and B are located at different CPUs, in order to avoid complex communication, we use flux BA replace flux AB for the calculation. Theoretically speaking, they should be the same, but some time they have tiny difference. This could be a reason caused the series and parallel version have different performance.
-
Is there anything we can do to make the parallel restart work?
Thanks,
Haiyan
