COSMO-RS explained, without the heavy math

COSMO-RS (Conductor-like Screening Model for Real Solvents) is a method for predicting how molecules behave in liquid mixtures — solubility, partitioning, activity coefficients — starting from quantum chemistry rather than from tables of measured data. It is one of the few predictive thermodynamic methods that needs no solute-specific experimental input, which is why it has become a workhorse for solvent screening and early formulation.

The core idea in one paragraph

Run each molecule once in a virtual conductor to get its σ-profile (its surface-charge fingerprint). Then model a real liquid as an ensemble of surface patches that pair up. Two patches with opposite, matching charge are happy together (favorable); a polar patch forced to sit next to a nonpolar one is not. Sum the energy of all these pairings statistically, and you get the chemical potential of any molecule in any mixture — from σ-profiles alone.

Why 'real solvents' is in the name

Plain COSMO describes a molecule in an idealized conductor. The 'RS' extension corrects that idealization back to a real solvent by accounting for the actual σ-profile of the surrounding molecules and adding physically-motivated terms for misfit electrostatics and hydrogen bonding. The result is a statistical thermodynamics that works across arbitrary solvents and mixtures, not just water.

What COSMO-RS predicts well

• Activity coefficients at infinite dilution — the foundation for most downstream properties.
• Partition coefficients (logP / logD) across solvent pairs, including unusual solvents where empirical logP models have no training data.
• Solubility trends across solvents — especially relative rankings for solvent selection.
• Liquid–liquid and vapor–liquid equilibria for screening separations and anti-solvent crystallization.
• Miscibility and co-solvency behavior.

Where it has limits

• Absolute solid solubility needs a fusion/melting term (the energy to melt the crystal) that COSMO-RS alone does not provide — the crystal lattice is outside the model.
• Strong, specific complexation or ionic-liquid speciation can stretch the pairwise-surface assumption.
• Conformationally floppy molecules need proper ensemble averaging or the input σ-profile is biased.
• Quality is bounded by the σ-profiles fed in — garbage profiles, garbage thermodynamics.

The dependency that people underestimate: the σ-profile recipe

COSMO-RS is often discussed as if the σ-profile is a given. In practice the functional, basis set, and cavity used to generate the profile change the numbers materially. A consistent, validated recipe applied identically to every molecule is what makes a COSMO-RS comparison meaningful. This is the single biggest lever on accuracy that is fully under your control.

How to use it in practice

1. Generate σ-profiles for your solute and your candidate solvents with one fixed recipe.
2. Compute infinite-dilution activity coefficients or partition coefficients for the screen.
3. Rank solvents; use the ranking to cut an experimental matrix from dozens to a handful.
4. For solubility, combine the activity coefficient with a fusion term (from DSC or an estimate).
5. Confirm the top few candidates experimentally.

Used this way, COSMO-RS is not a replacement for the lab — it is a way to spend your lab time on the few experiments most likely to matter.

Starting point

Everything above starts from a good σ-profile. mfsig.com produces σ-profiles from SMILES with a validated, reproducible recipe and full provenance, so the thermodynamics you build on top of them is defensible.