Advanced
Handle advanced general solver properties.
Syntax
model.sol(sname).feature(solv).create(fname,"Advanced")
model.sol(sname).feature(solv).feature(fname).set(pname,value)
Description
Feature for a number of advanced solver settings and assembly settings.
Table 6-24: Valid General Property/value Pairs for Advanced
Property
Values
Default
Description
autorescale
on | off
on
Automatic rescaling of linear equations (for the Stationary solver).
blocksize
positive integer | auto
auto
Assembly block size.
cachepattern
true | false
false
Reuse sparsity pattern during assembly.
checkmatherr
on | off
off
Check for undefined numerical values after each operation.
complexfun
on | off
off
Use complex-valued functions with real input.
convinfo
on | detailed | off
on
Print info to log.
D, E, K, L, M, N
on | off
 
Manual control of reassembly.
keep
on | off
 
Manual control of reassembly.
logsampling
double, zero or positive
0.005
Minimum time (in seconds) for log sampling of time-dependent solvers.
matherr
on | off
on
Error for undefined operations.
matrixformat
auto | sparse | filled | matrixfree
auto
Matrix format.
orthonormallimit
double
1e7
Limit, as an estimate of the complexity (number of operations), for using the orthonormal method in the automatic null-space function.
nullfun
flnullorth | flspnull | auto|explicitorth|
explicitsp
auto
Null-space function.
rowscale
on | off
on
Equilibrate rows.
storeresidual
off | solving | solvingandoutput
off
Store the latest residual while solving or while solving and in the output.
symmetric
auto | on | off | hermitian
auto
Symmetric matrices.
The Advanced section in the COMSOL Multiphysics Reference Manual, describes the functionality corresponding to the properties blocksize, complexfun, nullfun, and rowscale.
You can use the property symmetric to tell the solver that the model is symmetric or Hermitian, or you can use the automatic feature to find out (see Advanced in the COMSOL Multiphysics Reference Manual).
You can set convinfo=detailed to print more detailed information about the solver process in the log window. For example information about individual linear iterations or the scales per field computed by the automatic scaling algorithm. When convinfo=off, only minimal information about the solution process is printed.
By default, COMSOL Multiphysics gives an error message if the solver encounters an undefined mathematical operation when solving the model, for instance 0/0 or log(0). If you instead want the solver to proceed, put the property matherr=off. Then 0/0=NaN (not a number) and log(0)=-Inf.
The properties keep and D, E, K, L, M, and N allow manual control of reassembly. If keep=on, each of the other properties controls reassembly of a specific matrix or vector. Setting the property value to on, means that the quantity is constant, and therefore can be assembled once and then kept. The letters have the following meaning: E=constant mass, D=constant damping, K=constant Jacobian, L=constant load, M=constant constraint, N=constant constraint Jacobian.
The autorescale property control if the automatically computed scales should be recomputed. This property only affects stationary nonlinear problems and fields that are using the automatic scaling method and for the constant damping technique. The initially computed scales are based on the initial assembled matrix. When autorescale=on the scales are recomputed in each nonlinear iteration based on the current solution.
You can use the property matrixformat to tell the solver which matrix format to store the system matrices in:
•
sparse stores the matrix in a sparsely populated matrix format.
•
filled stores the matrix in a densely populated format.
•
matrixfree does not store matrices. The effects of matrix operations are assembled when needed.
When you specify auto (the default), the format is automatically determined based on the solver used.
By default, COMSOL does not check for undefined numerical values (for example, from numerical overflow) after each numerical operation. Set the property checkmatherr to on to make COMSOL check for such undefined numerical values, which will give more accurate error messages if such undefined numerical values occur.