Choosing a DFT recipe for σ-profiles: functional, basis, and cavity

Two labs can compute a σ-profile for the same molecule and get materially different answers — not because anyone made a mistake, but because they made different method choices. Understanding the three knobs that matter, and then fixing them, is what turns σ-profiles from interesting to comparable.

Knob 1: the DFT functional

The functional determines how electron density — and therefore surface polarization — is described. Hybrid and modern meta-GGA functionals generally describe polarization and weak interactions better than older pure functionals. The functional is the most expensive knob to change your mind about later, because it shifts every profile, so it is the one to validate first and then leave alone.

Knob 2: the basis set

The basis set sets how flexibly the wavefunction can respond. Too small a basis underdescribes polar groups and lone pairs, distorting exactly the donor/acceptor wings that matter most for thermodynamics. Larger basis sets are more faithful but cost more; the practical art is choosing the smallest basis that is converged enough for the property you care about. A polarized double-zeta basis such as def2-SVP is a common, well-balanced production choice.

Knob 3: the cavity and radii

The COSMO cavity defines where the molecular surface sits, and the atomic radii that build it scale every σ value. Inconsistent radii are the most insidious source of non-comparability: profiles can look reasonable individually yet be impossible to compare because the surfaces were built differently. Fixing the cavity construction and radii is non-negotiable for a comparable data set.

Why a fixed, validated recipe wins

Once you have chosen functional, basis, and cavity, the single most valuable thing you can do is apply that exact recipe to every molecule, forever, and record it. The benefit is not just reproducibility — it is comparability. A property model trained on profiles from one recipe and applied to profiles from another is comparing apples to oranges, and the error shows up as mysterious, hard-to-debug bias.

The reproducibility checklist

• Functional, basis, and cavity/radii recorded with every profile.
• Geometry source and conformer treatment recorded.
• Code and version pinned — numerical results can shift between library versions.
• Output signed or hashed so it is tamper-evident.

The mfsig approach

mfsig.com fixes these choices to a validated production recipe (v0.91.1: DFT with COSMO implicit solvation on a def2-SVP basis) and ships a provenance record with every reference-grade profile. You get the comparability of a single recipe without having to build and maintain the pipeline yourself.