Polynomial degree unexpected behavior

[mrmundt]

Summary

When terms in the expression cancel, polynomial_degree does not calculate this and reports the wrong answer.

In addition, when variables are fixed to 0 so as to make a 0 coefficient, it is possible to fool polynomial_degree into thinking that the degree is 1 when in fact it is 0. It seems that it should be an option to calculate the degree treating fixed Vars as constants or treating them as variables.

The beginning of this conversation is in the tail of #2271.

Steps to reproduce the issue

@jsiirola mentioned this one, but these examples don't involve fixing variables at all:

m = ConcreteModel()
m.x = Var()
m.e1 = Expression(expr=m.x -m.x)
print(m.e1.polynomial_degree()) # 1
m.e2 = Expression(expr=m.x**2 - m.x**2)
print(m.e2.polynomial_degree()) # 2

But there is also an inconsistency between generate_standard_repn and polynomial_degree when parts of the expression are multiplied by subexpressions that are fixed to 0:

m.y = Var()
m.z = Var()
m.e4 = Expression(expr=(m.x + m.y)*m.z)
print(m.e4.polynomial_degree()) # 2
m.x.fix(0)
m.y.fix(0)
print(m.e4.polynomial_degree()) # 1 (This is objectively wrong... If not 0, then 2 makes more sense.)

m.e5 = Expression(expr=(m.x**2 + m.y)*m.z)
print(m.e5.polynomial_degree()) # 1 (Again, I think 0 and 3 are the only correct answers here.)

# You don't even have to get fancy
m.e6 = Expression(expr=m.x*m.z)
print(m.e6.polynomial_degree()) # 1

Note that generate_standard_repn for m.e4 , m.e5, and m.e6 is 0 for all.

Information on your system

Pyomo version: 6.2 Python version: all Operating system: all How Pyomo was installed (PyPI, conda, source): N/A Solver (if applicable): N/A