Wrong empty intersection detection

[Ludox73]

The attached file contains .mat file with:

• a hybrid automaton HA, together with params and options;
• R = reach(HA, params, options);
• 4 sets, A_1, A_2, A_3, A_4, where A_i is R(2).timeInterval.set{21+i};
• a set B_1.
  The plot of the figure is the following (the vertical rectangles are the A_i, the horizontal one is B_1).

Expected behaviour:
launching isIntersecting(A_i,B_1) should return 1 for every i=1,...,4.

Obtained behaviour:
launching isIntersecting(A_i,B_1) returns 1 for i=1,4 and returns 0 for i=2,3.

I know that there may be potential numerical issues with the routines, but the size of the set involved are not that big or small. Is this expected or should it be addressed?

Thank you in advance for any help.

CORA_bug.zip

[toladnertum]

Dear @Ludox73,

Which Matlab and CORA version are you using (type version and CORAVERSION)?
I just tested it with the latest version of CORA and Matlab R2024b and it returns 1 for every i.

Best,
Tobias

[Ludox73]

Hi Tobias,
thank you for checking.
By typing those commands I get
version = '9.13.0.2126072 (R2022b) Update 3'
CORAVERSION = 'CORA v2025.1.0'

I hope I did not mess up any installation.
I will try to reinstall CORA, then to update MATLAB and I will be back to you with the results!

Ludo

[Ludox73]

Hi Tobias,
I still get the same result. I now have:
version = '24.2.0.2863752 (R2024b) Update 5'
CORAVERSION = 'CORA v2025.1.1'

Loading the file I get some warnings, but I think this should not justify the error.

I run MATLAB on Ubuntu 22.04, I use the latest version of YALMIP (Version 22-June-2023) and mosek 10.1.28 (I do not find the toolbox directory in version 11). Do you think this could be the issue?

I think it is really strange that we get inconsistent results.

Ludo

[toladnertum]

Thanks for the updates.
I was not using mosek, can you try to remove mosek from the path?

[Ludox73]

I just tried, and now I get 1 for all the results. Do you think this should be further investigated / is this a mosek issue?

[toladnertum]

It should definitely be further investigated. I will add it to our internal issue tracker and, if it's not a mosek issue, will be resolved in a future release. Thanks a lot for brining this up!

[Ludox73]

No problem. If it is ok for you, I would leave this issue open for a couple of hours: maybe I could need some help in activating mosek for the reachability andalysis and deactivate it for the intersection computation (at the moment I can't make it work this way :) ).

[Ludox73]

Important update:

I managed to use Mosek 11.0 (toolbox is only in 64-bit x86 version, not in ARM - I am sorry, I am not a programmer, maybe this was obvious).

With this version, this specific instance is solved. I will let you know if I have problems with other instances.

Thank you again,

Ludo

[Ludox73]

I was wrong.
The bug is still present. I was not correctly adding mosek to the path when I said it was working with the last version.

To wrap thigs up, I describe again the problem:

Description

Method "isIntersecting" returns 0 on two sets that intersect. It correctly returns 1 on two very similar sets. This happens only if Mosek is added to the path.

Steps to Reproduce the Issue

Load the .mat file inside the zip file and launch isIntersecting(A_2, B1).

Expected Behavior

launching isIntersecting(A_i,B_1) should return 1 for every i=1,...,4.

Current Behavior

launching isIntersecting(A_i,B_1) returns 1 for i=1,4 and returns 0 for i=2,3.

Versions

• CORA version: 'CORA v2025.2.0'
• Matlab version: '24.2.0.2863752 (R2024b) Update 5'
• OS: Ubuntu 24.04
• Mosek version (used using 'mosekdiag'): 11.0.25

The zip file can be found in the initial post.

Please behave as you please with this issue: I reopened it because the reason why I closed it was not valid. If you think this should not be an issue, please close it :).

[toladnertum]

That is unfortunate to hear. We'll look into it!