Question:  Is SPE10 Realistic?

Answer: No, it's not even close

The fine scale 1.1 million cell description given for Model 2 of the Tenth SPE Comparative Solution Project^1,2 exhibits 10 orders of magnitude local variation in gridblock-average values of areal permeability (Kx or Ky), and in areal permeability anisotropy (Ky/Kx).  These variations are many orders of magnitude greater than in any realistic description at the geologic model scale.  In order to obtain this level of disparity in areal gridblock-average permeabilities, gridblock areal dimensions would have to approach the scale of the heterogeneities (vug diameter or fracture or channel width).  In this channelized fluvial carbonate reservoir case with 10 by 20 foot areal blocks, channel width would have to be on the order of feet.  This extreme heterogeneity produces Jacobian matrices that are far more ill-conditioned than in any real case that we have seen.  Many very complex, powerful, and strongly converging solver methods have been developed in the last 8 years that are claimed to represent improvement over conventional methods based on their ability to solve the fine-grid SPE10 problem.  This includes AMG (algebraic multigrid), multi-level preconditioners, and multiscale methods.  We are not aware of any timings for reproducible benchmarks (or other evidence) that have been presented that demonstrate any advantages over conventional solver methods in realistic simulations, except in rare cases where parallel computing may be required due to extreme problem size and insufficient memory to run serially.  In those cases, those techniques that have been shown to provide excellent parallel scalability may be the most efficient solution.

1. http://www.spe.org/csp/

2. Christie, M.A., and Blunt, M.J., "Tenth SPE Comparative Solution Project: A Comparison of Upscaling Techniques", SPE Reservoir Engineering and Evaluation, 4, 308-317, (2001).