try except #2894
-
|
Hi Eventhough it converges for 0.5ns(I checked 0.5ns as the initial time step and it converged), when it comes to except, it does not. Can't I solve it in that order? What would be the reason? Appreciate your comments. If necessary I can upload the code. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 6 replies
-
|
There's not enough information here to really tell. Please publish a minimal failing example. |
Beta Was this translation helpful? Give feedback.
-
|
The following code solved for 1 second under try. Then I need to solve it under except. But it did not solve. from firedrake import *
import time as comp_time
from firedrake.petsc import PETSc
mesh = Mesh('domain.msh')
x = SpatialCoordinate(mesh)
t = 0.0 #STARTING TIME
Dt = (1e-9) #If time step is 1 second , there is no error, solved under try
s = FunctionSpace(mesh, 'CG', 1)
w = VectorFunctionSpace(mesh, 'CG', 1)
d = TrialFunction(s)
de = TestFunction(s)
un = Function(s)
un0 = Function(s)
initial = Function(s)
initial.interpolate( 1e5*( exp( -(1e-20)* ( (x[0])**2 + (x[1]-0.001)**2 ) ) ))
un.assign(initial)
un0.assign(initial)
We=as_vector((5e10,0))
u = Function(w).interpolate(We)
Form_N = x[0]*( (un)*(((un)-(un0))/Dt)*de + ((un)*We[i]) .dx(i)*de )*dx
bc = []
bc.append( DirichletBC(s, 0.0, 12) )
bc.append( DirichletBC(s, 0.0, 11) )
Jacobian = derivative(Form_N,un,d)
try:
solve(Form_N == 0, un, J=Jacobian, bcs=bc, solver_parameters={"nonlinear_solver": "snes",
"snes_solver" : { "linear_solver": "mumps",
"maximum_iterations": 20,
"relative_tolerance": 1E-4,
"absolute_tolerance": 1E-3,
"report": True,
"error_on_nonconvergence": True,
}},form_compiler_parameters={"quadrature_degree":2}
)
#un0.assign(un)
PETSc.Sys.Print('done')
except:
Dt=Dt*1e9
solve(Form_N == 0, un, J=Jacobian, bcs=bc, solver_parameters={"nonlinear_solver": "snes",
"snes_solver" : { "linear_solver": "mumps",
"maximum_iterations": 20,
"relative_tolerance": 1E-4,
"absolute_tolerance": 1E-3,
"report": True,
"error_on_nonconvergence": True,
}},form_compiler_parameters={"quadrature_degree":2}
)Error was: |
Beta Was this translation helpful? Give feedback.
-
|
OK, I think I can tell you what's going on here. However for future reference, there are a couple of issues with the above. First, it's important to format code and output as code in your comment by surrounding it with triple backticks: If you don't do this then GitHub will misinterpret some of the code as format markings and the result won't be legal code. I have edited your comment to correctly format the code (I also put in that the code is in python3 so it get's syntax highlighting, but that's not strictly needed). Next, I tried to run your code and discovered that I couldn't because it depends on Now, in this case your issue is how you defined the Form. It's not directly anything to do with the Dt = (1e-9) #If time step is 1 second , there is no error, solved under tryyou set Dt to the Python float you chose. Then, when you define the Form: because The way to change this is to define Dt = Constant(1.e-9)This will make this constant a parameter of the Form rather than hard-coding its value. You can then change the value of Dt.assign(Dt*1e9)I also note that your solver options are probably not doing what you want them to. The options you are setting do not appear to be PETSc options. I'm not sure where they are from. I think you mean something like: I will note that those are very, very loose solver tolerances, so you might get some surprises when they are applied (currently they're not being applied because you haven't used the right names). As an aside, if you are solving a timestepping problem then you probably want to use a NonlinearVariationalSolver rather than Solve. This will cause a bunch of reuse between timesteps which can be a significant performance improvement. In addition, you probably don't need to explicitly set |
Beta Was this translation helpful? Give feedback.
-
|
Hi, I just used Constant and assign as you have shown above for Dt. Then I printed new Dt value. But the printed value is 'w_0'. Why is it? I will apply and check the other points you have suggested as well. |
Beta Was this translation helpful? Give feedback.
-
|
Printing a print(float(Dt)) |
Beta Was this translation helpful? Give feedback.
-
|
It works now. Thanks! |
Beta Was this translation helpful? Give feedback.
-
|
I have one more question. I modified the code with a for loop to find the correct Dt value instead of giving it directly( According to the below code, in 9 th iteration Dt should be 1 second and answer should be converged. But it did not. Why does it not work with the loop? It always goes to last |
Beta Was this translation helpful? Give feedback.
OK, I think I can tell you what's going on here. However for future reference, there are a couple of issues with the above. First, it's important to format code and output as code in your comment by surrounding it with triple backticks:
If you don't do this then GitHub will misinterpret some of the code as format markings and the result won't be legal code. I have edited your comment to correctly format the code (I also put in that the code is in python3 so it get's syntax highlighting, but that's not strictly needed).
Next, I tried to run your code and discovered that I couldn't because it depends on
domain.msh. When posting a failing example, it's likely that others w…