OPTALG solve crash about optalg

[ttinoco]

This is interesting. It would be helpful if you somehow give me that matrix
you are using so I can recreate the problem, fix the issue and add a
unittest.
Tomas

On Sun, May 8, 2016 at 2:22 PM, Martin Zellner [email protected]
wrote:

OPTALG slove() crashes with the following error (which is not caught as
an Python Exception):

lda must be >= MAX(N,1): lda=5 N=6BLAS error: Parameter number 7 passed to
cblas_dtrsv had an invalid value

Probably there is something odd with my matrix structure, however this
should not lead to a crash without any Python output.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub
#1

[martinzellner]

I hope saving the matrices worked:
crash.npz.zip

Here is what I do:

[martinzellner]

It might be platform specific though (OpenBLAS vs. Apple Accelerate)..

[ttinoco]

There are several odd things in the code above:

1. variables need to be specified before constructing a problem (as an example, see the DCOPF problem construction in GRIDOPT). Your optimization problem has no variables.
2. generator Q participations are not needed since Q should not be variables,
3. generator P participations are needed when there is more than one generator connected to a bus (unless the objective function removes extra degrees of freedom)
4. you are missing the LBOUND constraint to get l <= Gx <= u with G=identity
5. p.gphi is not equal to the g argument of the QP problem. gphi(x) is equal to g + Hx
6. in your problem above, p.l and p.u are meaningless since no inequality constraints have been constructed.

[martinzellner]

Thanks for the remarks. I will change the code accordingly. Mostly your points are taken care of within the MPProblem class at the moment (e.g. variable definition). This needs to be changed, but the problem should not be affected by this.
Concerning 5): Can I get gsomehow?

[ttinoco]

Yes, use the identity given in the previous comment: g + Hx = gphi(x),
solve for g

On Sun, May 8, 2016 at 8:46 PM, Martin Zellner [email protected]
wrote:

Thanks for the remarks. I will change the code accordingly. Mostly your
points are taken care of within the MPProblem class at the moment (e.g.
variable definition). This needs to be changed, but the problem should not
be affected by this.
Concerning 5): Can I get gsomehow?

—
You are receiving this because you commented.
Reply to this email directly or view it on GitHub
#1 (comment)

[martinzellner]

Just as a remark: The crash only occurs when I add the battery energy constraints. So I assume something is odd with my A matrix...

[martinzellner]

I could track it down further. This happens when a all-zero column appears in the A matrix (which shouldn't happen).
I think there should be an Exception complaining about the matrix being rank deficient.

[ttinoco]

It is not as simple as that. Rank of A is not the rank the optimization algorithm cares about, but the rank of the KKT matrix. I suspect the linear solver is throwing the error. The linear solver should be throwing a python exception. If not, then the wrapper has a bug (either superlu or mumps, I am not sure which one you are using). The wrappers are external to optalg. However, I am sure some checks can be added at the beginning of the solve method to check bad data easily. I will take care of this as soon as I have time.