BundleSolver setup too restrictive

[shuheng-liu]

The following code block assumes each bundle parameter must fall into one of the following categories.

1. a parameter that only appears in the boundary/initial condition (NOT in the equation); or
2. a parameter that only appears in the equation (NOT in the boundary/initial condition)
   

   neurodiffeq/neurodiffeq/solvers.py

   Lines 1318 to 1332 in a3df9fd

   	non_var = []
   	for c in conditions:
   	try:
   	bc = tuple(c.bundle_conditions.values())
   	except Exception:
   	bc = ()
   	for index in bc:
   	non_var.append(index)
   	non_var = list(set(non_var))
   	def non_var_filter(*variables):
   	variables = list(variables)
   	for i, n in enumerate(non_var):
   	del variables[len(conditions) + 1 + n - i]
   	return ode_system(*variables)

However, in theory, there should be a case where the parameter appears both in the equation and in the initial/boundary condition. The var here probably implies a parameter that appears only in the equation. A fix is needed.

[shuheng-liu]

Plus, here's my interpretation of line 1331. It might be helpful to anyone else trying to understand what the non_var_filter is doing.

• len(conditions) is the number of functions to solve for
• + 1 is hard-coded number of free variables/coordinates/inputs to functions. Since this is an ODE solver, there's only 1 input.
• + n is the index of a var (a parameter that appears only in the condition). It's deleted since the var doesn't appear in the equation by assumption.
• - i accounts for the fact that at step i (starting from 0), a total of i vars have been previously deleted. This assumes previously deleted vars appear before the current var (i.e., non_var list is sorted)! Which is not guaranteed by line 1326 and can lead to unnoticeable bugs!