add support for numpy min, max and diff in ExecComp

[fzahle]

Summary

Summary of PR.

Related Issues

• Resolves #

Backwards incompatibilities

None

New Dependencies

None

[fzahle]

we also use np.interp here and there, but adding this is more tricky since it doesn't pass the complex step test. Looking at using the SplineComp instead.

[JustinSGray]

wait ... min and max are not differentiable .... hrm....

I guess we had fmax in there, as well as maximum and minumum already. None of those are differentiable... Im not even sure how they behave when use complex step with them...

I guess I can't object to new non-differentiable functions being added when there are already some in there, though it does make me a bit nervous

[fzahle]

if you use complex step for derivatives, it should work since it will find the min/max along the real axis and return that along with the corresponding complex part.

[JustinSGray]

That is only mathematically valid if the location of the max-value in the array isn't changing. Will it work? probably for fairly well behaved functions. Anyway, its there now... so I guess we're stuck with it. So I can't object to adding the similar methods either.

import numpy as np 

a = np.ones(4)*10

b = np.ones(4, dtype='complex')*10
b[0] += complex(0, 1e-50)

print(np.max(a))

print(np.max(b))

b[1] += 1e-10
print(np.max(b))