Hi,
I am working on a CGE model in GAMS/MPSGE and have been running into an error I cannot figure out. When increasing the complexity of the model, the solver fails - sometimes as early as timestep 2 of 10, but more commonly in the later timesteps.
In the system log I get a *** System error: Jacobian nonzero allocation, and in the .lst file the iteration count shows -214748 at the point of failure.
I have not been able to find any sources explaining what this indicates, so any help would be greatly appreciated!
Hi @akolstad ,
Can you share code/data that would allow to reproduce the error?
Does this happen with the latest GAMS 53?
Fred
Hi Fred,
Unfortunately, I am not able to due to the size of my files and restrictions from my university. Are there any specific parts that can be helpful? I have attached the last entries of the process log before the error below. Just a clarification from my last post: the solver continues after the Jacobian error for later timesteps and the results converge.
I am currently using GAMS 50.4.0, but will try with the newest version.
From process log:
Major Iteration Log
major minor func grad residual step type prox inorm (label)
109 1467 295 117 5.1853e+05 8.0e-01 SB 0.0e+00 2.5e+05 (PU.FR.Q5)
110 263 306 118 5.1853e+05 1.5e-03 CB 0.0e+00 2.5e+05 (PU.FR.Q5)
Minor Iteration Log
minor t z w v art ckpts enter leave
500 -3.2160e-05 7914 738 0 78 9 w[ 7702] z[ 7691]
1000 8.3518e-04 7915 782 0 33 19 z[ 4197] w[ 2391]
1500 7.4383e-01 7829 877 0 24 29 z[ 2361] a[ 5012]
2000 -1.7561e-01 7974 745 0 11 39 w[ 760] z[ 7352]
2500 -3.0196e-01 7922 797 0 11 48 w[ 7440] z[ 7057]
3000 6.6352e-02 7917 803 0 10 58 z[ 7684] w[ 8115]
3500 7.4548e-01 7828 892 0 10 68 z[ 7698] z[ 3022]
Major Iteration Log
major minor func grad residual step type prox inorm (label)
111 3575 308 119 2.9838e+05 8.0e-01 CB 0.0e+00 1.4e+05 (PU.IT.Q5)
112 352 309 120 2.9927e+05 1.0e+00 CM 0.0e+00 1.4e+05 (PU.IT.Q5)
113 155 310 121 3.4102e+04 1.0e+00 SM 1.0e+01 1.8e+04 (INVB.PL)
114 2 311 122 3.4102e+04 1.0e+00 NM 1.0e+01 1.8e+04 (INVB.PL)
*** MPSGE program error
Last scalar input record 361448
*** System error: Jacobian nonzero allocation.
Thank you in advance.
Aurora
Hello,
There have been some fixes made to MPSGE since GAMS 50, and there is a reasonable chance that these fixes may help in your case. So your plan to try the latest version of GAMS is a good one. BTW, the Release Notes give some detail.
In case the latest version doesn’t address all the issues, you should consider how you can send us a reproducible example. For example, if you contact our tech support team directly, would you be allowed to send the example then? I know we have ways to upload large files and maintain confidentiality. The log output you sent is highly suggestive of an issue that can only be addressed if you send the code that turfs up the bad behavior.
Steve
Hi,
First of all, thank you for the help. For later reference, the issue was unfortunately not resolved with GAMS 53. I have gotten approval to send the code to the tech support team.
Aurora
1 Like