IterativeParams Class

class plask.IterativeParams

Iterative matrix parameters

This class holds parameters for iterative matrix used by solvers implementing Finite Element Method. PLaSK uses NSPCG package for performing iterations. Please refer to its documentation for explanation of most of the settings.

Attributes

accelerator	Solver accelerator
converged	True if the solver converged
err	Residual error in the last run
iters	Number of iterations in the last run
ltrunc	Truncation level
lvfill	Fill-in level
maxerr	Maximum allowed residual iteration
maxit	Maximum number of iterations
ndeg	Polynomial degree
nfact	Frequency of partial factorization
noconv	Desired behavior if the iterative solver does not converge.
nrestart	Restart frequency
nsave	Saved vectors number
omega	Relaxation parameter
preconditioner	Solver preconditioner

Descriptions

Attribute Details

IterativeParams.accelerator = <property object>

Solver accelerator

This is current iterative matrix solver acceleration algorithm.

Possible choices:

cg	Conjugate Gradient acceleration (default)
si	Chebyshev acceleration or Semi-Iteration
sor	Successive Overrelaxation (can use only SOR preconditioner)
srcg	Symmetric Successive Overrelaxation Conjugate Gradient Algorithm (can use only SSOR preconditioner)
srsi	Symmetric Successive Overrelaxation Semi-Iteration Algorithm (can use only SSOR preconditioner)
basic	Basic Iterative Method
me	Minimal Error Algorithm
cgnr	Conjugate Gradient applied to the Normal Equations
lsqr	Least Squares Algorithm
odir	ORTHODIR, a truncated/restarted method useful for nonsymmetric systems of equations
omin	ORTHOMIN, a common truncated/restarted method used for nonsymmetric systems
ores	ORTHORES, another truncated/restarted method for nonsymmetric systems
iom	Incomplete Orthogonalization Method
gmres	Generalized Minimal Residual Method
usymlq	Unsymmetric LQ
usymqr	Unsymmetric QR
landir	Lanczos/ORTHODIR
lanmin	Lanczos/ORTHOMIN or Biconjugate Gradient Method
lanres	Lanczos/ORTHORES or “two-sided” Lanczos Method
cgcr	Constrained Generalized Conjugate Residual Method
bcgs	Biconjugate Gradient Squared Method

IterativeParams.converged = <property object>

True if the solver converged

IterativeParams.err = <property object>

Residual error in the last run

IterativeParams.iters = <property object>

Number of iterations in the last run

IterativeParams.ltrunc = <property object>

Truncation level

Truncation bandwidth to be used when approximating the inverses of matrices with dense banded matrices. An increase in this value means a more accurate factorization at the expense of increased storage.

IterativeParams.lvfill = <property object>

Fill-in level

Level of fill-in for incomplete Cholesky preconditioners. Increasing this value will result in more accurate factorizations at the expense of increased memory usage and factorization time.

IterativeParams.maxerr = <property object>

Maximum allowed residual iteration

IterativeParams.maxit = <property object>

Maximum number of iterations

IterativeParams.ndeg = <property object>

Polynomial degree

Degree of the polynomial to be used for the polynomial preconditioners.

IterativeParams.nfact = <property object>

Frequency of partial factorization

This number initializes the frequency of partial factorizations. It specifies the number of linear system evaluatations between factorizations. The default value is 1, which means that a factorization is performed at every iteration.

IterativeParams.noconv = <property object>

Desired behavior if the iterative solver does not converge.

Possible choices are: error, warning, continue

IterativeParams.nrestart = <property object>

Restart frequency

The number of iterations between restarts for restarted acceleration methods.

IterativeParams.nsave = <property object>

Saved vectors number

The number of old vectors to be saved for the truncated acceleration methods.

IterativeParams.omega = <property object>

Relaxation parameter

IterativeParams.preconditioner = <property object>

Solver preconditioner

This is current preconditioner used for iterative matrix solver.

Possible choices:

rich	Richardson’s method
jac	Jacobi method
ljac	Line Jacobi method
ljacx	Line Jacobi method (approx. inverse)
sor	Successive Overrelaxation (can be used only with SOR accelerator)
ssor	Symmetric SOR
ic	Incomplete Cholesky (default)
mic	Modified Incomplete Cholesky
lsp	Least Squares Polynomial
neu	Neumann Polynomial
lsor	Line SOR
lssor	Line SSOR
llsp	Line Least Squares Polynomial
lneu	Line Neumann Polynomial
bic	Block Incomplete Cholesky (ver. 1)
bicx	Block Incomplete Cholesky (ver. 2)
mbic	Modified Block Incomplete Cholesky (ver. 1)
mbicx	Modified Block Incomplete Cholesky (ver. 2)