We’ve been extensively involved in developing new and testing already existing methods for continuous electrostatics problems solution. Large part of the effort is building and understanding of the so called Generalized Born models. It’s been known for quite a while that although various incarnations of the Poisson Boltzmann equation solvers can provide fast and relatively stable polar solvation energies contributions, General Born models beyond the so called Coulomb field approximation (CFA) are fairly accurate and clearly superior both in speed and stability.
The caveat is the overall scaling of the computational time and resources required for a single solvation energy calculation scales like 
To understand the errors of the approximation and the numerical stability of the solvation energies we attempted the following calculation. We took 90 large molecules and calculated the polar solvation energies using COSMO-like PBE solver (Surface Electrostatics Solver, SES), a series of fast surface based electrostatics methods (FSBE) and the surface charges based GB methods. The energies were calculated at different, though the same for all the molecules values of the surface density (number of the discretization points per unit area of the molecular surface,
). The error, 
for each of the molecules was defined as the difference of the calculated energy 
at a given density minus the saturation value of the solvation energy 
: 
. For any given calculation method and each density of the surface we averaged and plotted the R.M.S. errors 
. The results are represented on the graph above and thus characterize both the stability of the obtained value and the convergence speed.
The results show that all the surface GB methods are pretty stable, demonstrate the error steadily decreasing as the surface density (and thus the computational time) increases. SES though is supposed to be exact, shows a great level of numerical instability and appears to produce unreasonable numbers even at high values of 
Although nearly all of the methods show similar results, SCGB(3) emerges as a clear winner. Among all the surface charges GB methods, SCGB(3) is the most accurate. It is also the only method producing correct surface polarization charge for an arbitrary molecule. The analysis show that surface-based GB models are stable and can be used in practical calculations already at surface densities
.
Related posts:
