Bug in FunctionSpace with bc as dictionary

[Thilo Höschler]

Hi Mikael!

I noticed a small bug in FunctionSpace. When I run the following code

from shenfun import FunctionSpace, Function

A = FunctionSpace(10, family='legendre', bc=(1, 2, 3, 4))
B = FunctionSpace(10, family='legendre', bc={'left': [('D', 1), ('N', 3)],
                                             'right': [('D', 2), ('N', 4)]})

a = Function(A).set_boundary_dofs()
b = Function(B).set_boundary_dofs()

print(a, b)

I get

[0. 0. 0. 0. 0. 0. 1. 2. 3. 4.] [0. 0. 0. 0. 0. 0. 1. 3. 2. 4.]

alhough both functions should be the same since the first two entries in (1, 2, 3, 4) are used to specify the Dirichlet and the last two for the Neumann boundary conditions. I guess the reason for this is the following code:

shenfun/shenfun/forms/arguments.py

Lines 156 to 185 in 33490d8

 elif isinstance(bc, dict): 
     bcs = {k.lower(): list(v) if isinstance(v[0], (tuple, list)) else [v] for k, v in bc.items()} 
     bc = [] 
     key = [] 
     if 'left' in bcs: 
         bcs['left'].sort() 
         for bci in bcs['left']: 
             if bci[0] == 'N': 
                 bc.append(bci[1]/df) 
             elif bci[0] == 'N2': 
                 bc.append(bci[1]/df**2) 
             elif bci[0] == 'N3': 
                 bc.append(bci[1]/df**3) 
             else: 
                 bc.append(bci[1]) 
         key += ['L'+bci[0] for bci in bcs['left']] 
     if 'right' in bcs: 
         bcs['right'].sort() 
         for bci in bcs['right']: 
             if bci[0] == 'N': 
                 bc.append(bci[1]/df) 
             elif bci[0] == 'N2': 
                 bc.append(bci[1]/df**2) 
             elif bci[0] == 'N3': 
                 bc.append(bci[1]/df**3) 
             else: 
                 bc.append(bci[1]) 
         key += ['R'+bci[0] for bci in bcs['right']] 
  
 return key, tuple(bc) 

Herein the boundary conditions are ordered in the wrong way when converting the dict to a tuple.

If you find this to be a bug I can try to fix it myself if you want.

Best regards,
Thilo

[Mikael Mortensen]

Hi Thilo,
You're absolutely right. There is a logical mistake in there. But I'm not sure whether to fix the code above, or to switch the order of bases 2 and 3 here

shenfun/shenfun/legendre/bases.py

Lines 1480 to 1490 in 33490d8

 class BCBiharmonic(BCBase): 
  
     @staticmethod 
     def short_name(): 
         return 'BCB' 
  
     def stencil_matrix(self, N=None): 
         return sp.Rational(1, 30)*np.array([[15, -18, 0, 3], 
                                             [15, 18, 0, -3], 
                                             [5, -3, -5, 3], 
                                             [-5, -3, 5, 3]]) 

Perhaps we need to consider the whole logic to try to make it more robust?

[Thilo Höschler]

Hi Mikael,

switching the order of the bases would definitely be the easiest way of fixing this. Also, I don't think that it is really necessary to change the whole logic since the only problem was the special ordering of the boundary conditions for the ShenBiharmonic basis. So I would simply switch the order in BCBiharmonic.

[Mikael Mortensen]

Ok, great. I will push a fix to this issue then and make sure to update the docstring with the correct order.

[Thilo Höschler]

This was fixed in 4e36d73!