If you provide your own linear solver, it must be a subtype of
BifurcationKit.jlwill not recognize it. See example just below.
The linear solvers provide a way of inverting the Jacobian
J or solving
J * x = rhs. Such linear solver
linsolve will be called like
sol, success, itnumber = linsolve(J, rhs; kwargs...) throughout the package.
Here is an example of the simplest one (see
src/LinearSolver.jl for the true implementation) to give you an idea, the backslash operator:
struct DefaultLS <: AbstractLinearSolver end function (l::DefaultLS)(J, rhs; k...) return J \ rhs, true, 1 end
Note that for
newton to work, the linear solver must return 3 arguments. The first one is the result, the second one is whether the computation was successful and the third is the number of iterations required to perform the computation.
You can then call it as follows (and it will be called like this in
ls = DefaultLS() J = rand(2, 2) # example of linear operator ls(J, rand(2))
\solver based on
Choleskydepending on the type of the Jacobian. This works for sparse matrices as well. You can create one via
linsolver = DefaultLS().
- GMRES from
IterativeSolvers.jl. You can create one via
linsolver = GMRESIterativeSolvers()and pass appropriate options.
- GMRES from
KrylovKit.jl. You can create one via
linsolver = GMRESKrylovKit()and pass appropriate options.
By tuning the options of
GMRESKrylovKit, you can select CG, GMRES... see KrylovKit.jl.
It is very straightforward to implement the Conjugate Gradients from IterativeSolvers.jl by copying the interface done for
gmres. Same goes for
minres,... Not needing them, I did not implement this.
Preconditioners should be considered when using Matrix Free methods such as GMRES.
GMRESIterativeSolvers provides a very simple interface for using them. For
GMRESKrylovKit, we implemented a left preconditioner. Note that, for
GMRESKrylovKit, you are not restricted to use
Vectors anymore. Finally, here are some packages to use preconditioners
- IncompleteLU.jl an ILU like preconditioner
- AlgebraicMultigrid.jl Algebraic Multigrid (AMG) preconditioners. This works especially well for symmetric positive definite matrices.
- Preconditioners.jl A convenient interface to conveniently called most of the above preconditioners using a single syntax.
- We provide a preconditioner based on deflation of eigenvalues (also called preconditioner based on Leading Invariant Subspaces) using a partial Schur decomposition. There are two ways to define one i.e.
Apart from setting a preconditioner for a linear solver, it can be advantageous to change the preconditioner during computations, e.g. during a call to
newton. This can be achieved by taking advantage of the callbacks to these methods. See the example 2d Ginzburg-Landau equation (finite differences, codim 2, Hopf aBS).