[ODEs] Caching matrix and vector when operator cannot be evaluated at t = 0.0

[ericneiva]

Dear @oriolcg,

I'd say an ODE solver caches (allocates) the matrix and vector, by evaluating the FE operator at t = 0.0.

For instance, the Affine Theta Method:

Gridap.jl/src/ODEs/ODETools/AffineThetaMethod.jl

Lines 12 to 20 in be0dc01

	if cache === nothing
	ode_cache = allocate_cache(op)
	vθ = similar(u0)
	vθ .= 0.0
	l_cache = nothing
	A, b = _allocate_matrix_and_vector(op,u0,ode_cache)
	else
	ode_cache, vθ, A, b, l_cache = cache
	end

Eventually visits these two through _allocate_matrix_and_vector:

Gridap.jl/src/ODEs/TransientFETools/TransientFEOperators.jl

Line 243 in be0dc01

vecdata = collect_cell_vector(V,op.res(0.0,xh,v))

Gridap.jl/src/ODEs/TransientFETools/TransientFEOperators.jl

Line 303 in be0dc01

_matdata = (_matdata...,_matdata_jacobian(op,0.0,xh,i,0.0))

I am trying to solve a transient problem whose operator cannot be evaluated at t = 0.0, so you can imagine I am getting an error when caching the ODE system...

Shall we pass the initial time to all _allocate_matrix_and_vector methods?

[oriolcg]

Good point @ericneiva . Until now we assumed that the operator allocation was independent of the time and "hard coded" its evaluation at t=0 for allocation purposes. Passing t0 would solve the issue and make it more general, without overhead (other than having an argument more in the function).