SAMG-MODFLOW

SAMG for groundwater simulation

Mesh-Agnostic yet fast

SAMG is a very efficient linear solver library based on algebraic multigrid (AMG) specifically developed for industrial applications. SAMG supports both serial and multi-core computations on single PCs, workstations or compute nodes.

LMG is an interface linking the SAMG library to MODFLOW. There are two versions of LMG: LMG-2kx supports MODFLOW-2000 and MODFLOW-2005 whereas LMG-USG supports MODFLOW-USG.

Key Features of SAMG:

State-of-the-art AMG performance – applicable for structured and unstructured meshes.
Optimized solver control including automated-setup-reuse and switching between different solvers.
Supports both multi-core and multi-node systems.

Advantages for MODFLOW:

„Big Models“: Allows increased modeling accuracy for both structured and unstructured models (as in MODFLOW-USG), while keeping compute times manageable.
Handles strongly heterogeneous and unstructured models where standard solvers have problems. No additional knowledge about solvers needed.

Model courtesy of ESI, simulating the development of mining in a western groundwater basin (in seconds).

The two graphs show compute times for solving a single linear system from a steady-state stress period. Both the MODFLOW-USG and the MODFLOW-2005 model have the same numerical grid with roughly 1 Mio. grid-cells. In MODFLOW-2005 the PCG solver was able to use information on the matrix structure resulting from the known structure of the numerical grid, which is no longer possible for MODFLOW-USG. For this particular model PCG did not converge in a reasonable amount of time. The xMD solver converged, although it was much slower than the PCG solver for the MODFLOW-2005 model.

Hence, LMG’s capability to handle unstructured grids natively results in an even larger speed-up compared with the standard MODFLOW solvers in MODFLOW-USG than it did in MODFLOW-2005. For the full simulation, LMG achieves a speed-up of 4.5 on a single core compared with xMD.

User Comments

“...Without SAMG as our solver engine, it would have been impossible to solve [transport models with up to 30 million variables] in a practicable time.” – DHI-WASY GmbH

“[SAMG] was able to reduce the compute time of [one of] our simulations from 12 hours to 3 hours, so our engineers can now run multiple simulations a day.” – ITASCA DENVER Inc.

“In general, the SAMG solver is the fastest MODFLOW solver that [we] have used.” – Aquaveo LLC

“[SAMG] provided optimal runtime speed compared to public-domain solvers evaluated....” – GeoHydroScience LLC

“In general, the SAMG solver is the fastest MODFLOW solver that [we] have used. I have seen multiple models where the SAMG solver was not only faster but more stable than the standard MODFLOW solvers” – Aquaveo LLC