krypy.recycling.evaluators - evaluators for deflation vector candidates

class krypy.recycling.evaluators.RitzApproxKrylov(mode='extrapolate', tol=None, pseudospectra=False, bound_pseudo_kwargs=None, deflweight=1.0)

Bases: krypy.recycling.evaluators._RitzSubsetEvaluator

Evaluates a choice of Ritz vectors with a tailored approximate Krylov subspace method.

Parameters:
  • mode –

    (optional) determines how the number of iterations is estimated. Must be one of the following:

    • extrapolate (default): use the iteration count where the extrapolation of the smallest residual reduction over all steps drops below the tolerance.
    • direct: use the iteration count where the predicted residual bound drops below the tolerance. May result in severe underestimation if pseudospectra==False.
  • pseudospectra – (optional) should pseudospectra be computed for the given problem? With pseudospectra=True, a prediction may not be possible due to unfulfilled assumptions for the computation of the pseudospectral bound.
  • bound_pseudo_kwargs – (optional) a dictionary with arguments that are passed to bound_pseudo().
  • deflweight – (optional) see estimate_time(). Defaults to 1.
evaluate(ritz, subset)

Returns a list of subsets with indices of Ritz vectors that are considered for deflation.

class krypy.recycling.evaluators.RitzApriori(Bound, tol=None, strategy='simple', deflweight=1.0)

Bases: krypy.recycling.evaluators._RitzSubsetEvaluator

Evaluates a choice of Ritz vectors with an a-priori bound for self-adjoint problems.

Parameters:
  • Bound – the a-priori bound which is used for estimating the convergence behavior.
  • tol – (optional) the tolerance for the stopping criterion, see _KrylovSolver. If None is provided (default), then the tolerance is retrieved from ritz._deflated_solver.tol in the call to evaluate().
  • strategy –

    (optional) the following strategies are available

    • simple: (default) uses the Ritz values that are complementary to the deflated ones for the evaluation of the bound.
    • intervals: uses intervals around the Ritz values that are considered with simple. The intervals incorporate possible changes in the operators.
evaluate(ritz, subset)

Returns a list of subsets with indices of Ritz vectors that are considered for deflation.