ott.math.decomposition.SparseCholeskySolver

class ott.math.decomposition.SparseCholeskySolver(A, beta=0.0, key=None, **kwargs)

Sparse Cholesky solver using jax.experimental.host_callback.call().

Uses the CHOLMOD [Chen et al., 2008] bindings from sksparse.cholmod.

Parameters

• A (TypeVar(T)) – Symmetric positive definite matrix of shape [n, n].

• beta (float) – Decompose \(A + \beta * I\) instead of \(A\).

• key (Optional[Hashable]) – Key used to cache sksparse.cholmod.Factor. This key must be unique to A to achieve correct results. If None, use hash() of this object.

• kwargs (Any) – Keyword arguments for sksparse.cholmod.cholesky().

Methods

clear_factor_cache()	Clear the sksparse.cholmod.Factor cache.
create(A, **kwargs)	Instantiate sparse or dense Cholesky solver.
solve(b)	Solve the linear system \(A * x = b\).

Attributes

A	Symmetric positive definite matrix of shape [n, n].
L	Compute the lower-triangular factor of A and cache the result.