Welcome to KryPy’s documentation!

What is KryPy and where is the code?

KryPy is a Krylov subspace methods package for Python. If you’re looking for the source code or bug reports, take a look at KryPy’s github page. These pages provide the documentation of KryPy’s API.

Getting started


KryPy can be installed easily with the Python package installer by issuing pip install krypy. Alternatively, it can be installed by downloading the source from KryPy’s github page and then running python setup.py install.

Solve a linear system

The following code uses MINRES to solves a linear system with an indefinite diagonal matrix:

from numpy import diag, linspace, ones, eye
from krypy.linsys import LinearSystem, Minres

# construct the linear system
A = diag(linspace(1, 2, 20))
A[0, 0] = -1e-5
b = ones(20)
linear_system = LinearSystem(A, b, self_adjoint=True)

# solve the linear system (approximate solution is solver.xk)
solver = Minres(linear_system)


The vector \(e_1\) can be used as a deflation vector to get rid of the small negative eigenvalue \(-10^{-5}\):

from krypy.deflation import DeflatedMinres
dsolver = DeflatedMinres(linear_system, U=eye(20, 1))


The deflation subspace can also be determined automatically with a recycling strategy. Just for illustration, the same linear system is solved twice in the following code:

from krypy.recycling import RecyclingMinres

# get recycling solver with approximate Krylov subspace strategy
rminres = RecyclingMinres(vector_factory='RitzApproxKrylov')

# solve twice
rsolver1 = rminres.solve(linear_system)
rsolver2 = rminres.solve(linear_system)

The convergence histories can be plotted by

from matplotlib.pyplot import semilogy, show, legend
semilogy(solver.resnorms, label='original')
semilogy(dsolver.resnorms, label='exact deflation', ls='dotted')
semilogy(rsolver2.resnorms, label='automatic recycling', ls='dashed')

which results in the following figure.

Indices and tables