discretise

Contents

discretise#

hybrid_aif.discretise(A, B, Q, dt, modes=None)[source]#

The exact zero-order-hold discretisation of dx = A x + B u dt + dw, dw ~ N(0, Q dt).

Over a duration dt: F = expm(A dt), Q_d = int_0^dt expm(A s) Q expm(A s)ᵀ ds and G = int_0^dt expm(A s) ds B (a control held over the step), by Van Loan’s matrix exponential. A and Q are (d, d), B is (d, du) or None (the kernel then has no G). dt is one positive number, shared (a kernel whose innovations have the trivial parent shape (1,), or (modes,) copies when modes is given), or a (levels,) array, one duration per level of a control node: the kernel is then per-mode, F, G and the innovation’s parent axis the level, which is how an option’s kernel is written (modes, if given, must equal levels). Design: docs/design/step-durations.md T3.