Using custom linear solver

[Oscar Molina]

Hi,

I'm trying to implement my own linear solver with MRST. However,  I'm getting the following error when launching the simulation: "Solver Failure (...) Failure reason: Linear solver failure: MATLAB:TooManyInputs: Too many input arguments." 

Based on the example below, can you tell if I'm doing something wrong?

Thanks in advance for your help.

Oscar

%%%%%%%%%%%%%%%%%%%%%%%%%%

% Example solver function

%%%%%%%%%%%%%%%%%%%%%%%%%%

function x = gpuGMRESwithILU(A, b)

% ILU on CPU, solve on GPU

setup.type = 'ilutp';

setup.droptol = 1e-4;

[L, U] = ilu(A, setup);

M1_gpu = gpuArray(full(L));

M2_gpu = gpuArray(full(U));

A_gpu = gpuArray(A);

b_gpu = gpuArray(b);

x_gpu = gmres(A_gpu, b_gpu, 50, 1e-8, 10, M1_gpu, M2_gpu);

x = gather(x_gpu);

end

% Define function handle

ls = @(A, b) gpuGMRESwithILU(A, b);

linsolver = HandleLinearSolverAD(ls);

% Incorporate linear solver into schedule simulator

[wellSols, states, report] = simulateScheduleAD(initState, model, schedule, 'linearsolver', linsolver);

[Bård Skaflestad]

Hi,

Would you be able to give a little more context for the diagnostic, please? In particular, I think it would be useful to have a full backtrace for the "Solver Failure...Too many input arguments" message as that might identify the location that triggers the problem.

Best Regards,

Bård Skaflestad

SINTEF Digital, Mathematics & Cybernetics

Applied Computational Science group

---

From: sinte...@googlegroups.com <sinte...@googlegroups.com> on behalf of Oscar Molina <ommoli...@gmail.com>
Sent: Thursday, May 07, 2026 15:53
To: MRST-users: The Matlab Reservoir Simulation Toolbox User Group <sinte...@googlegroups.com>
Subject: [MRST Users] Using custom linear solver

--
You received this message because you are subscribed to the Google Groups "MRST-users: The Matlab Reservoir Simulation Toolbox User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sintef-mrst...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/sintef-mrst/ffa2eb23-6307-4dc9-8dfa-7f5d376a92f0n%40googlegroups.com.

[Oscar Molina]

Hi Bard,

Apologies for the delayed response. It seems like the HandleLinearSolverAD class either cannot be used directly with the schedule simulator or need additional set-up work (superclass initialization, parametrization, etc.) Is my observation correct?

Below is an example custom solver class that I put together for testing purposes; my goal is to extend this approach to GPU-based solvers.

classdef CustomSolver < LinearSolverAD

properties

options = struct("type","ilutp","droptol",1e-6);

end

methods

function solver = CustomSolver(varargin)

solver@LinearSolverAD();

solver = merge_options(solver, varargin{:});

end

function [result, report] = solveLinearSystem(solver, A, b, varargin)

timer = tic();

try

setup.type = 'ilutp';

setup.droptol = max(1e-3, solver.options.droptol); % larger drop tolerance -> cheaper

setup.udiag = 1; % keep unit diagonal if appropriate

[L,U] = ilu(sparse(A), setup);

catch

d = diag(A);

d(abs(d) < eps) = 1; % avoid division by zero

Dinv = spdiags(1./d, 0, size(A,1), size(A,2));

L = speye(size(A,1)); % treat as M1 = I

U = Dinv; % treat as M2 = D^{-1} (applied on right in bicgstab/gmres calls)

end

result = bicgstab(A,b,1e-4,100,L,U);

t_solve = toc(timer);

report = solver.getSolveReport('LinearSolutionTime', t_solve);

end

function [d, sn] = getDescription(solver)

sn = 'mldivide';

sn = [sn, solver.id];

d = 'Matlab \ operator (mldivide direct solver)';

end

end

end

Thanks,

Oscar