v0.9.43

PETScDiffEq v0.9.43

Diff since v0.9.42

Fixes

• The history of a discrete parameter is saved as OrdinaryDiffEq saves it, so sol.ps[:d] returns it; save_discretes = false turns that off (#180, fixes #170).

v0.9.42

PETScDiffEq v0.9.42

Diff since v0.9.41

Fixes

• BrownFullBasicInit and ShampineCollocationInit on a DM over a communicator use the problem's jac when it has one, as they already did on one rank (#178).

v0.9.41

PETScDiffEq v0.9.41

Diff since v0.9.40

Fixes

• With a dm and no jac, an implicit solve on a 2-D DMDA with a periodic axis no longer stops with PetscError(77) when the axis is not a multiple of the colouring's period; PETSc then colours the DM's matrix itself (#177, fixes #168).

v0.9.40

PETScDiffEq v0.9.40

Diff since v0.9.39

Features

• PETScAdjoint takes a DynamicalODEProblem or SecondOrderODEProblem on a communicator, with TSRK, TSARKIMEX and the theta methods (#167).

v0.9.39

PETScDiffEq v0.9.39

Diff since v0.9.38

Features

• PETScAdjoint takes a SplitODEProblem with TSARKIMEX on a communicator. A missing jac or paramjac of either part is built by ForwardDiff under autodiff = AutoForwardDiff() (#165).

Merged pull requests:

• Run PETScAdjoint on a SplitODEProblem over a communicator (#165) (@singhharsh1708)
• Run PETScAdjoint on a partitioned problem over a communicator (#167) (@singhharsh1708)
• Colour the DM's matrix where the DM's colours do not fit it (#177) (@singhharsh1708)
• Initialize a DM on a communicator with its jac (#178) (@singhharsh1708)

Closed issues:

• Default Jacobian with a periodic 2-D DMDA fails with PetscError(77) for most grid sizes (#168)

v0.9.38

PETScDiffEq v0.9.38

Diff since v0.9.37

Features

• Zygote differentiates solve(prob, alg; saveat, sensealg = PETScAdjoint()) for u0 and p. The extension now also loads on ChainRulesCore, which SciMLSensitivity already requires (#164).

v0.9.37

PETScDiffEq v0.9.37

Diff since v0.9.36

Features

• With a DMDA, autodiff = AutoForwardDiff() builds the Jacobian of TSImplicit, TSRosW and TSARKIMEX by ForwardDiff. The default with a dm stays PETSc's finite-difference colouring (#163).

v0.9.36

PETScDiffEq v0.9.36

Diff since v0.9.35

Features

• A serial ODEProblem or SplitODEProblem takes an array-shaped u0: f, jac, callbacks and the solution see arrays of its size. A DAEProblem, a comm, a dm and PETScAdjoint still need a vector (#162).

v0.9.35

PETScDiffEq v0.9.35

Diff since v0.9.34

Features

• autodiff = AutoForwardDiff() works for a SecondOrderODEProblem or DynamicalODEProblem on a communicator (#161).

v0.9.34

PETScDiffEq v0.9.34

Diff since v0.9.33

Features

• PETScAdjoint takes cost times anywhere inside the span of an adaptive solve: they are given to PETSc as its time span and a step ends on each one (#160).