numerics - 2D Heat equation: inconsistent boundary and initial conditions

I'm attempting to use NDSolve on a 2D boundary value problem with initial conditions. Upon running my code, I get the following message:

"NDSolve::ibcinc: Warning: Boundary and initial conditions are inconsistent."

After much head-scratching, I can't seem to find my mistake. It seems to me that my initial condition is consistent with my boundary conditions:

k = 1 / (5*(Pi^2));
soln = NDSolve[
  {
  (* PDE *)

  D[u[x, y, t], t] == k*(D[u[x, y, t], x, x] + D[u[x, y, t], y, y]),

  (* initial condition *)
  u[x, y, 0] == y + Cos[Pi*x] Sin[2*Pi*y],

  (* boundary conditions *)
  u[x, 0, t] == 0,
  u[x, 1, t] == 1,
  (D[u[x, y, t], x] /. x -> 0) == 0,
  (D[u[x, y, t], x] /. x -> 1) == 0

  },
 u,
 {x, 0, 1},
 {y, 0, 1},
 {t, 0, 1}
 ]

Any advice is greatly appreciated,

Rick

Answer

The documentation has a full section dedicated to inconsistent boundary conditions in PDEs.

Quoting it,

Occasionally, NDSolve will issue the NDSolve::ibcinc message warning about inconsistent boundary conditions when they are actually consistent. This happens due to discretization error in approximating Neumann boundary conditions or any boundary condition that involves a spatial derivative. The reason this happens is that spatial error estimates (see "Spatial Error Estimates") used to determine how many points to discretize with are based on the PDE and the initial condition, but not the boundary conditions. The one-sided finite difference formulas that are used to approximate the boundary conditions also have larger error than a centered formula of the same order, leading to additional discretization error at the boundary. Typically this is not a problem, but it is possible to construct examples where it does occur.

Then an example follows, and a possible solution using the Method option's "TensorProductGrid" suboption, which we can also apply to your problem.

When the boundary conditions are consistent, a way to correct this error is to specify that NDSolve use a finer spatial discretization.

k = 1/(5*(Pi^2));

soln = NDSolve[{
   D[u[x, y, t], t] == k*(D[u[x, y, t], x, x] + D[u[x, y, t], y, y]),
   u[x, y, 0] == y + Cos[Pi*x] Sin[2*Pi*y],
   u[x, 0, t] == 0, 
   u[x, 1, t] == 1, 
   (D[u[x, y, t], x] /. x -> 0) == 0, 
   (D[u[x, y, t], x] /. x -> 1) == 0}, 

  u, {x, 0, 1}, {y, 0, 1}, {t, 0, 1}, 


  Method -> {"MethodOfLines", 
               "SpatialDiscretization" -> {"TensorProductGrid", "MinPoints" -> 20}}]

In this instance "MinPoints" -> 20 was sufficient to make the problem go away.

---

The same problem was discussed here. I vaguely remembered it, but it took me a while to find it again ...