Expression member variable automatic zero default value feature missing

[Gordon Klaus]

In CBC.Twist: fishy.py and flap.py, I get "RuntimeError: expected a default value to all member variables in the Expression." for the force and magnitude Expressions, respectively. The automatic-zero-default-value feature was a nice one, so I suspect it went away by accident and the bug actually lies in Expression; but I don't really know.

[Gordon Klaus (gordon-klaus)]

Forgot to mention, I'm using the development versions of FEniCS and CBC.Solve.

[Harish Narayanan (hnarayanan)]

There are two problems with changing API here. First is with the Expression class, and that is easily fixed.

The second is, without the VariationalProblem class, how do I specify solver parameters to solve(a == L)? e.g., what does the following translate to?

equation = VariationalProblem(L, a, bcu)
equation.parameters["solver"]["newton_solver"]["absolute_tolerance"] = 1e-12
equation.parameters["solver"]["newton_solver"]["relative_tolerance"] = 1e-16
equation.parameters["solver"]["newton_solver"]["maximum_iterations"] = 100

[Anders Logg (logg)]

On Mon, Aug 15, 2011 at 12:32:17PM -0000, Harish Narayanan wrote:
> There are two problems with changing API here. First is with the
> Expression class, and that is easily fixed.
>
> The second is, without the VariationalProblem class, how do I specify
> solver parameters to solve(a == L)? e.g., what does the following
> translate to?
>
> equation = VariationalProblem(L, a, bcu)
> equation.parameters["solver"]["newton_solver"]["absolute_tolerance"] = 1e-12
> equation.parameters["solver"]["newton_solver"]["relative_tolerance"] = 1e-16
> equation.parameters["solver"]["newton_solver"]["maximum_iterations"] = 100

solve(a == L) does not take any parameters.

So we need to use the more "advanced interface" which is

problem = NonlinearVariationalProblem(F, u, bcs)
solver = NonlinearVariationalSolver(problem)
solver.parameters["..."] = ...
solver.parameters["..."] = ...
solver.solve()

--
Anders

[Harish Narayanan (hnarayanan)]

On 8/15/11 2:45 PM, Anders Logg wrote:
> On Mon, Aug 15, 2011 at 12:32:17PM -0000, Harish Narayanan wrote:
>> There are two problems with changing API here. First is with the
>> Expression class, and that is easily fixed.
>>
>> The second is, without the VariationalProblem class, how do I specify
>> solver parameters to solve(a == L)? e.g., what does the following
>> translate to?
>>
>> equation = VariationalProblem(L, a, bcu)
>> equation.parameters["solver"]["newton_solver"]["absolute_tolerance"] = 1e-12
>> equation.parameters["solver"]["newton_solver"]["relative_tolerance"] = 1e-16
>> equation.parameters["solver"]["newton_solver"]["maximum_iterations"] = 100
>
> solve(a == L) does not take any parameters.
>
> So we need to use the more "advanced interface" which is
>
> problem = NonlinearVariationalProblem(F, u, bcs)
> solver = NonlinearVariationalSolver(problem)
> solver.parameters["..."] = ...
> solver.parameters["..."] = ...
> solver.solve()

Will use this instead then. Thanks!

Harish