StrictUpperTriQSM

StrictUpperTriQSM#

class tinygp.solvers.quasisep.core.StrictUpperTriQSM(p: JAXArray, q: JAXArray, a: JAXArray)[source]#

Bases: QSM

A strictly upper triangular order m quasiseparable matrix

The notation here is somewhat different from that in Eidelman & Gohberg (1999), because we wanted to map StrictLowerTriQSM.transpose() -> StrictUpperTriQSM while retaining the same names for each component. Therefore, our p is their h, and our a is their b.T.

Parameters:

  • p (n, m) – The right quasiseparable elements.

  • q (n, m) – The left quasiseparable elements.

  • a (n, m, m) – The transition matrices.

matmul(x: tinygp.helpers.JAXArray, *, parallel: bool = False) → tinygp.helpers.JAXArray[source]#

The dot product of this matrix with a dense vector or matrix

Parameters:

  • x (n, ...) – A matrix or vector with leading dimension matching this matrix.

  • parallel – If True, use a parallel associative-scan algorithm instead of the default sequential scan.

scale(other: tinygp.helpers.JAXArray) → StrictUpperTriQSM[source]#

The multiplication of this matrix times a scalar, as a QSM

self_add(other: StrictUpperTriQSM) → StrictUpperTriQSM[source]#

The sum of two StrictUpperTriQSM matrices

self_mul(other: StrictUpperTriQSM) → StrictUpperTriQSM[source]#

The elementwise product of two StrictUpperTriQSM matrices

property shape: tuple[int, int]#

The shape of the matrix

to_dense() → tinygp.helpers.JAXArray#

Render this representation to a dense matrix

This implementation is not optimized and should really only ever be used for testing purposes.

transpose() → StrictLowerTriQSM[source]#

The matrix transpose as a QSM