subvector assignment

Question

DOLFIN's NonlinearProblem interface passes around vectors instead of functions, so to fill these vectors with meaningful values, one might have to convert them to functions first. No problem:

def F(self, b, x):
    xfun = Function(self.W)
    xfun.vector()[:] = x
    # do something with xfun, get bfun
    b = bfun.vector()

(Are there ways to do this without copying?)

I'm dealing with a block problem now, and it goes like

def F(self, b, x):
    xfun = Function(self.WP)
    xfun.vector()[:] = x
    x0, x1 = xfun.split()
    # do something with x0, x1, get b0, b1

How to appropriately assign b0.vector(), b1.vector() to b?

Comments

If I understand the problem, you may like to encapsulate vector b into Function to be able to use FunctionAssigner.

Answer 1

If self.WP is a VectorFunctionSpace (with the same type of sub space) you should be able to set up two FunctionAssigner objects inside the constructor of your subclassed NonlinearProblem:

def __init__(self, ...)
    ...
    self.full_func = Function(self.WP)
    self.u0 = Function(self.WP.sub(0).collapse())
    self.u1 = Function(self.WP.sub(1).collapse())
    self.to_sub = FunctionAssigner([self.WP.sub(0), self.WP.sub(1)], self.WP)
    self.from_sub = FunctionAssigner(self.WP, [self.WP.sub(0), self.WP.sub(1)])

def F(self, b, x):
    self.full_func.vector()[:] = x
    self.to_sub.assign([self.u0, self.u1], self.full_func)
    ...
    self.from_sub.assign(self.full_func, [self.u0, self.u1])
    b[:] = self.from_sub.vector()