DVGeoMutli - dIdptTriA - too many indices for array: array is 2-dimensional, but 3 were indexed

[bmlowe1]

Hey Anil (and whomever else),

We've been running into some issues with this particular line in DVGeoMulti:

pygeo/pygeo/parameterization/DVGeoMulti.py

Line 2021 in c889e1c

dIdptTriA = dIdptTriA[:, disp[self.comm.rank] : disp[self.comm.rank + 1], :]

For us, the allreduce just above this line returns a 2-dimensional array (we're using mpi4py 4.0.0), which gives off this error at line 2021:

IndexError: too many indices for array: array is 2-dimensional, but 3 were indexed
    dIdptTriA = dIdptTriA[:, disp[self.comm.rank] : disp[self.comm.rank + 1], :]

As a quick fix, I have found that the following lines fix the problem when placed just before line 2021:

# Make dIdptTriA at least 3D
if len(dIdptTriA.shape) == 2:
    dIdptTriA = np.array([dIdptTriA])

Thank you for your time on this.

[A-CGray]

Hey Brandon

I'm not familiar with DVGeoMulti, but from a quick look at the code I think the issue is here:

pygeo/pygeo/parameterization/DVGeoMulti.py

Line 2014 in c889e1c

dIdptTriA = np.zeros(self.compA.nodes.shape)

This is setting the shape of dIdptTriA to be the same as the shape of the coordinates array, but there should be an additional 1st dimension that is the size of the number of functions, which is already defined earlier in the function:

pygeo/pygeo/parameterization/DVGeoMulti.py

Lines 1971 to 1972 in c889e1c

	# number of functions we have
	N = dIdpt.shape[0]

So the line should actually be:

  # Set the triangulated mesh seeds to all zeros on these procs
  else:
      dIdptTriA = np.zeros((N,) + self.compA.nodes.shape)

Same goes for dIdptTriB just below this section.

We'd welcome a PR fixing this.

Out of interest, are you using numpy 2?

[bmlowe1]

Hey Ali,

Yes, that makes sense, nice find, and a much cleaner fix.

We're using Numpy 1.26.4 currently.

Thanks.

[A-CGray]

We're using Numpy 1.26.4 currently.

Interesting, I thought this might've been caused by some change in behaviour between versions 1 and 2 but I guess not.

[joanibal]

So the line should actually be:

 # Set the triangulated mesh seeds to all zeros on these procs
 else:
     dIdptTriA = np.zeros((N,) + self.compA.nodes.shape)

So do we all agree this is the fix? If so I will go ahead and make the PR

[sabakhshi]

So the line should actually be:

 # Set the triangulated mesh seeds to all zeros on these procs
 else:
     dIdptTriA = np.zeros((N,) + self.compA.nodes.shape)

So do we all agree this is the fix? I so I will go ahead and make the PR

Yes. Please go ahead and create the PR and we will merge.