CVX solver MOSEK fatal error

Matlab Version 2024b
CVX Version 2.2
Mosek Version 10.2.6

Dear expert
When I call Mosek solver (version 10.2.6) in CVX, the code has a problem as follows:

MOSEK fatal error stopenv.
Version : 10.2.6
Platform: Windows/64-X86
File : src\optimize\optconic.c
Line : 1952
Msg. : info.resp: 23 canonical->solsta: 0

Please report this message to support@mosek.com.

Fatal MOSEK error in src\optimize\optconic.c line 1952.

How can I deal with this problem?

Save the task file

cvx_solver_settings('write', 'dump.task.gz')

and send an email with this task to Mosek support as mentioned in the error message.

Hello!
This error warning occurred during the Monte Carlo simulation process. After I ran the code again, the error did not appear. Is this error accidental?

It is a bug in Mosek caused by some specific problem data. To fix it we need the exact reproducible data. The code indicates an internal error.

My guess is that an eigenvalue computation fails, but I’m not sure. If we should do anything about it, we really need an example that reproduces the issue.

Can you start a new MATLAB session, restart the random number generator, or something, so that you can reproduce the same random number stream as the instance in which the error occurred? Perhaps you can reproduce all the random numbers generation leading up to that error. or just run the simulation a bunch of times with the same input and see if the error occurs. Make sure to write the task file each time in case that simulation run has the error, and make sure the task file is not overwritten before you can save the one which had the error.

Thank you!
This rarely occurs. After I ran the code again, this error did not occur. If this error occurs frequently during code execution, I will seek advice from experts.

Hello!
I have encountered the same problem again:
MOSEK fatal error stopenv.
Version : 10.2.6
Platform: Windows/64-X86
File : src\optimize\optconic.c
Line : 1952
Msg. : info.resp: 23 canonical->solsta: 0

Please report this message to support@mosek.com.

{错误使用 cvx_mosek
Fatal MOSEK error in src\optimize\optconic.c line 1952.

Matlab Version 2024b
CVX Version 2.2
Mosek Version 10.2.6

The occurrence time of this error is different from the previous occurrence time of this error.
If I choose different versions of Mosek, can it solve this problem?(MOSEK Fatal error - #2 by Michal_Adamaszek
Now, I try to use Mosek 9.1.9 bundled with CVX and see if that solves the problem?
Has this internal error in Mosek v10.2.6 been fixed in Mosek v11 ?

Hello!
My computer operating system is Windows11.
I have installed Matlab 2022a and Matlab 2024b on my computer.
I add CVX2.2 and Mosek 10.2.6 to Matlab 2024b.
Matlab 2024b:

Also, I add CVX2.2 and Mosek 10.2.6 to Matlab 2022a.
Matlab 2022a:

Is it possible for this case to cause the error as follows?
MOSEK fatal error stopenv.
Version : 10.2.6
Platform: Windows/64-X86
File : src\optimize\optconic.c
Line : 1952
Msg. : info.resp: 23 canonical->solsta: 0

Please report this message to support@mosek.com.

{错误使用 cvx_mosek
Fatal MOSEK error in src\optimize\optconic.c line 1952.

Matlab Version 2024b

In the link (MOSEK Fatal error - #2 by Michal_Adamaszek), you suggest that users use different versions of Mosek for solving and observe if the error can be resolved.
Then, if I want to install ‘Mosek Beta Version 11’ or other versions, I need to install it in a new folder, which is different from the folder of Moske v10.2.6, and then add the path of ‘Mosek Beta Version 11’ to Matlab. Is my operation correct?

Do I need to uninstall the old version ‘Mosek v10.2.6’ before installing the new version? Is there a way to completely uninstall it?

When I restarted the installation program of Mosek without uninstalling Mosek, I received the following prompt:

We will release a new version 11 later today that will remove fatal error. Rather Mosek will return with an error and hence it might be easier to catch the task file.

Thank you!
Does fatal error occur frequently? When I searched for this issue on the CVX forum, it seemed that this issue rarely appeared.
What is the difference between ‘Stable Versions’ and ’ Beta Versions’ ?

Ideally you should do as Mark suggested, that is write a task file with every solve, and then the last task file you get just before it crashes is the one we need to get. Everything else, like changing versions is potentially just moving the problem away a bit but not solving it at its core. We would also appreciate having that example and being able to fix the problem.

Version 11 instructions are here Mosek - Beta: MOSEK 11

Thank you!
Now, I am using Mosek 9.1.9 bundled with CVX and the fatal error has not yet occurred.
After the code ends, I will add the command ‘cvx_solver_settings(‘write’, ‘dump.task.gz’)’ to record the fatal error.
As I mentioned earlier, I add the path of CVX2.2 and Mosek 10.2.6 to Matlab 2022a.


I add the path of CVX2.2 and Mosek 10.2.6 to Matlab 2024ab.

Could it cause the fatal error?

In the link (MOSEK Fatal error - #2 by Michal_Adamaszek), you suggest that users use different versions of Mosek for solving and observe if the error can be resolved.
Then, if I want to install ‘Mosek Beta Version 11’ or other versions, I need to install it in a new folder, which is different from the folder of Moske v10.2.6, and then add the path of ‘Mosek Beta Version 11’ to Matlab. Is my operation correct?

Do I need to uninstall the old version ‘Mosek v10.2.6’ before installing the new version? Is there a way to completely uninstall it?

You can have many Mosek versions installed, cvx_setup will list them and then you can choose one for solving as discussed here Error in CVX mosek: Unrecognized field name 'sol' - #11 by Michal_Adamaszek

Test linear solve: Error 1002 (MSK_RES_ERR_LICENSE_VERSION)
Error message : Feature PTS version 11.0 is not supported by the license file.

Please obtain a new license file. There are instructions in Mosek - Beta: MOSEK 11

I already have used the new license file for Mosek Beta Version 11. However, such error occurs.

I restart the Matlab and this error is corrected, as mentioned in the email:
image