Incorrect boundary padded vector with composed PeriodicBC on 2d data

[vigeesh]

I'm trying to pad my vector u using the boundary condition operator Q. It works fine for 1d data, but gives me incorrect padded vector for 2d data when I use PeriodicBC.

Code for reproduction

1-D

#left-plot
plot(data1d)
#right-plot
Q1 = Dirichlet0BC(Float64)
plot(Q1*data1d,label="Q1*u (Dirichlet)")
Q2 = PeriodicBC(Float64)
plot!(Q2*data1d,label="Q2*u (Periodic)")

2-D

#left-plot
heatmap(data2d)
#right-plot
Q3 = compose(Dirichlet0BC(Float64, (6,6))...)
heatmap(Q3*data2d)

But, when I use PeriodicBC,

#left-plot
Q4 = compose(PeriodicBC(Float64, (6,6))...)
heatmap(Q4*data2d)
#right-plot
heatmap(periodic_expected)

I get a padded data as shown in the left plot below, but I expect to get the one shown in the right,

Am I doing something wrong in composing the BC's in 2-D?

[xtalax]

I will take a look at this

[xtalax]

In the meantime, you should be aware that MethodOfLines.jl now has support for periodic BCs. If this doesn't help you solve your problem we'd love to know why.

[xtalax]

There's something strange going on, here the first testset passes, wheras the second testset displays the issue you described though both contain periodic BCs

[vigeesh]

Thanks @xtalax , I'll have a look at MethodOfLines.jl

here the first testset passes,

But it looks like it's not returning a correctly padded data.

I'm not sure if I understand this line:

BCx = vcat(fill(q1, div(m,2)), fill(q2, m-div(m,2)))  #The size of BCx has to be all size components *except* for x
BCy = vcat(fill(q2, div(n,2)), fill(q1, n-div(n,2)))

Why is it not,

BCx = fill(q1,m)
BCy = fill(q2,n)

But, even in the latter case, it gives me wrong results when I use PeriodicBC in both dims, but strangely gives me correct padding when I use a combination of PeriodicBC (along one dim) and Dirichlet/Neumann/RobinBC (for the other dim).

[xtalax]

This was to test the constructor which supports different bcs along a boundary to facilitate space dependant robin.

But, even in the latter case, it gives me wrong results when I use PeriodicBC in both dims, but strangely gives me correct padding when I use a combination of PeriodicBC (along one dim) and Dirichlet/Neumann/RobinBC (for the other dim).

How are you applying the BCs, with compose? I wonder if its due to some compiler strangeness simplifying the filled periodic as a single object.

[vigeesh]

How are you applying the BCs, with compose?

yes, like in the test

BCx = fill(q1,m)
BCy = fill(q2,n)
Qx = MultiDimBC{1}(BCx)
Qy = MultiDimBC{2}(BCy)
Q = compose(BCx,BCy)