How to set MaxStepSize for spatial variable in NDSolve?

When applying NDSolve to a 1-D transient heat equation, the following code appears to set MaxStepSize for the temporal variable t.

t0 = 0;
t1 = 10;
x0 = 0;
x1 = 1;
nx = 100;
dx = (x1-x0)/nx;
eqs1 = {D[u[x,t],x,x] - D[u[x,t],t] == NeumannValue[1, x==x0 || x == x1], DirichletCondition[u[x,t]==0, t== t0]};
sol1 = NDSolve[eqs1, u, {x, x0, x1}, {t, t0, t1}, Method-> {"MethodOfLines", "TemporalVariable" -> t}, MaxStepSize -> dx]
f = u /. First[sol1];
Dimensions[f["Grid"]]


I would like to apply MaxStepSize to the spatial variable x, rather than to the temporal variable t.

I've tried without success variations involving SpatialDiscretization, such as

NDSolve[eqs1, u, {x, x0, x1}, {t, t0, t1}, Method-> {"MethodOfLines", "TemporalVariable" -> t, "SpatialDiscretization"-> {"TensorProductGrid","MaxStepSize" -> dx}}]


which yields the following error:

NDSolve::moptx: Method option MaxStepSize in {NDSolveFiniteElement,MaxStepSize->1/100} is not one of {ConstraintMethod,BoundaryTolerance,InterpolationOrder,IntegrationOrder,LinearSolveMethod,MeshOptions,PrecomputeGeometryData}. >>

What is the correct approach?

I think you are looking for MaxCellMeasure in case of the finite element method.

t0 = 0;
t1 = 10;
x0 = 0;
x1 = 1;
nx = 100;
dx = (x1 - x0)/nx;
eqs1 = {D[u[x, t], x, x] - D[u[x, t], t] ==
NeumannValue[1, x == x0 || x == x1],
DirichletCondition[u[x, t] == 0, t == t0]};
sol1 = NDSolveValue[eqs1, u, {x, x0, x1}, {t, t0, t1},
Method -> {"MethodOfLines", "TemporalVariable" -> t,
"SpatialDiscretization" -> {"FiniteElement",
"MeshOptions" -> {"MaxCellMeasure" -> 0.01}}}, MaxStepSize -> dx]
Dimensions[sol1["Grid"]]


In case of the "SpatialDiscretization" "TensorProductGrid" you are looking for a "MaxPoints" and/or "MinPoints" option. You'd also need to adjust your equation a bit like so:

t0 = 0;
t1 = 10;
x0 = 0;
x1 = 1;
nx = 100;
dx = (x1 - x0)/nx;
eqs1 = {D[u[x, t], x, x] - D[u[x, t], t] == 0, u[x, 0] == 0
, Derivative[1, 0][u][0, t] == 1
, Derivative[1, 0][u][1, t] == 1
};
sol1 = NDSolveValue[eqs1, u, {x, x0, x1}, {t, t0, t1},
Method -> {"MethodOfLines", "TemporalVariable" -> t,
"SpatialDiscretization" -> {"TensorProductGrid",
"MaxPoints" -> 125, "MinPoints" -> 125}}, MaxStepSize -> dx]
Dimensions[sol1["Grid"]]


This is not quite correct as the initial condition does not match the boundary condition; but that's only possible with more knowledge of what you would like to solve.

• Thanks. The FiniteElement approach worked, and solved my immediate problem. However, Method -> {"MethodOfLines", "TemporalVariable" -> t, "SpatialDiscretization" -> {"TensorProductGrid", "MaxPoints" -> 2000}} yielded the error NDSolveValue::moptx: Method option MaxPoints in {NDSolveFiniteElement,MaxPoints->2000} is not one of {ConstraintMethod,BoundaryTolerance,InterpolationOrder,IntegrationOrder,LinearSolveMethod,MeshOptions,PrecomputeGeometryData}. >> Can you advise the correct syntax for the latter approach? Commented May 17, 2016 at 7:25
• @RonnenLevinson, see update. Commented May 17, 2016 at 9:34
• The sample system models heat conduction following the sudden application of constant surface heat flux to a body at initial temperature zero. As originally written, it defines the BCs by adding a NeumannValue[] term to the PDE, rather than by providing separate BC equations, to avoid the warning from NDSolve[] about inconsistent boundary and initial conditions. I have been unable get the TensorProductGrid flavor of SpatialDiscretization to work with a NeumannValue[]-specified BC. That, coupled with the need to control grid spacing, prompted my original question. Commented May 17, 2016 at 17:20
• @RonnenLevinson, NeumannValue will always trigger FEM - as NeumannValue is FEM specific. If you want TensorProductGrid you need to express the derivatives with Derivative or D. Commented May 17, 2016 at 19:54