A botanical extraction company is designing a supercritical CO2 process to extract cannabinoids from hemp biomass. The target: 90%+ extraction efficiency for CBD and CBG while minimizing co-extraction of waxes, chlorophyll, and other undesired compounds. The process engineer uses the Peng-Robinson equation of state with published binary interaction parameters to predict CBD solubility in supercritical CO2 as a function of pressure and temperature. The model predicts 0.8 g/L at 300 bar and 50C. Pilot testing shows 0.3 g/L. The prediction is off by 2.7x.
This is not unusual. Cubic equations of state (Peng-Robinson, Soave-Redlich-Kwong) were developed for hydrocarbon systems where molecules interact primarily through dispersion forces. Supercritical extraction targets (natural products, pharmaceuticals, polymers, lipids) interact through hydrogen bonding, pi-stacking, and dipole-dipole forces that cubic EOS cannot represent. The binary interaction parameter kij is a single adjustable parameter that absorbs all the molecular interaction physics into one number. When the actual interactions are complex, one parameter is not enough.
The consequence is that SFE process design relies heavily on pilot testing. Each combination of pressure, temperature, co-solvent type, and co-solvent concentration requires experimental measurement. A full design-of-experiments for a new feedstock covers 50-100 conditions, takes 3-6 months of pilot time, and costs $100K-500K. If the feedstock changes (different plant cultivar, harvest timing, or preprocessing), the data may not transfer and the campaign must be repeated.
The Molecular Physics of Supercritical Solvation
Supercritical CO2 is a tunable solvent. At the critical point (31.1C, 73.8 bar), CO2 transitions from gas-like to liquid-like density. Above the critical point, density (and therefore solvent power) increases continuously with pressure. At 100 bar and 40C, the density is 0.63 g/cm3 (moderate solvent power). At 300 bar and 40C, it reaches 0.91 g/cm3 (strong solvent power). This tunability is what makes SFE attractive: you can selectively extract compounds by adjusting pressure and temperature.
But density alone does not determine solubility. The solute-CO2 interaction energy is equally important. Nonpolar compounds (terpenes, waxes) dissolve well in supercritical CO2 because CO2 has a significant quadrupole moment that provides dispersion interactions. Polar compounds (phenolic acids, glycosides) dissolve poorly unless a co-solvent (ethanol, methanol) is added to provide hydrogen bonding capacity. Intermediate compounds (cannabinoids, which have both nonpolar terpene moieties and polar hydroxyl groups) have solubilities that depend sensitively on the balance of interactions, exactly the regime where cubic EOS predictions are worst.
Molecular Simulation of Solubility in Supercritical Fluids
Molecular dynamics simulation computes solubility from the free energy of transferring a solute molecule from its pure solid or liquid phase into the supercritical solvent. The calculation does not require experimental solubility data or fitted binary interaction parameters. It requires only the molecular structures of the solute and solvent, which are known from chemistry.
The method works by constructing a simulation box of supercritical CO2 at the target density (determined by pressure and temperature through the CO2 equation of state, which is known to high accuracy). A single solute molecule is inserted, and the solvation free energy is computed using thermodynamic integration or free energy perturbation methods. The solvation free energy, combined with the sublimation or vaporization energy of the pure solute, gives the solubility at that P and T.
The computation captures the specific molecular interactions that cubic EOS misses. For CBD in supercritical CO2, the simulation shows that the pentyl chain inserts into the CO2 bulk (dispersion-driven), the resorcinol hydroxyl groups form weak hydrogen bonds with CO2 (the quadrupole accepts H-bonds poorly), and the cyclohexene ring interacts through pi-quadrupole interactions with CO2. The net solvation free energy is -18.3 kJ/mol, giving a predicted solubility of 0.32 g/L at 300 bar and 50C, within 7% of the experimental value compared to the 170% error from Peng-Robinson.
Co-Solvent Effects: Beyond Binary Interaction Parameters
Adding 5-15% ethanol to supercritical CO2 dramatically increases the solubility of polar compounds. The standard approach is to fit a new kij for the ternary system (solute-CO2-ethanol). But ternary kij values are rarely available in the literature, and fitting requires ternary solubility data that costs as much as the process design campaign.
Molecular simulation predicts co-solvent effects from first principles. Add ethanol molecules to the CO2 simulation box at the desired concentration. The simulation shows how ethanol molecules cluster around the polar functional groups of the solute, forming hydrogen bonds that stabilize the dissolved state. The solvation free energy becomes more negative (more favorable dissolution), and the solubility increases. The magnitude of the increase depends on the number and strength of hydrogen bonds between the solute and co-solvent, which the simulation computes directly.
For CBD, adding 10% ethanol increases the predicted solubility from 0.32 g/L to 2.1 g/L (6.5x enhancement). The simulation reveals that ethanol forms 1.8 hydrogen bonds on average with the two hydroxyl groups of CBD, replacing the weak CO2-hydroxyl interactions with strong ethanol-hydroxyl hydrogen bonds. The predicted enhancement matches experimental data (reported 5-8x enhancement) without any fitted parameters.
This extends to co-solvent screening. The simulation compares ethanol, methanol, isopropanol, ethyl acetate, and water as co-solvents for a given solute, predicting the solubility enhancement of each without experimental testing. For some compounds, ethyl acetate outperforms ethanol despite being less commonly used, a finding that would require months of pilot testing to discover experimentally.
Selectivity: Extracting What You Want, Leaving What You Don't
Selectivity is often more important than yield. In botanical extraction, the target compound (CBD) co-extracts with waxes, chlorophyll, lipids, and other plant metabolites. Post-extraction purification (winterization, chromatography) is expensive and lossy. Maximizing selectivity in the extraction step reduces downstream processing cost and improves overall yield.
Selectivity depends on the differential solubility of target vs. contaminant compounds. If CBD has 10x higher solubility than chlorophyll at a given P, T, and co-solvent concentration, the extract will be enriched 10x in CBD. The optimal conditions maximize the CBD/contaminant solubility ratio, not the absolute CBD solubility.
Molecular simulation computes selectivity by calculating the solubilities of all relevant compounds simultaneously. For a hemp extraction, the simulation computes solubilities for CBD, CBG, THC, plant waxes (C28-C34 alkanes and fatty acids), chlorophyll a and b, and major lipids (oleic and linoleic acids). The selectivity map (a plot of CBD/contaminant solubility ratio as a function of P, T, and co-solvent concentration) identifies the operating window that maximizes extract purity.
This selectivity map takes 1-2 weeks to compute for a full set of target and contaminant compounds. The equivalent experimental campaign (measuring solubilities for each compound at each condition) would take 6-12 months.
Mass Transfer and Extraction Vessel Design
Solubility determines the thermodynamic limit of extraction. Mass transfer determines how fast you reach that limit. In a packed bed extractor, the CO2 flows through the ground biomass, dissolving the target compounds and carrying them out. The extraction rate depends on the diffusion of CO2 into the biomass particles, the dissolution rate of the target compound from the plant matrix, and the transport of dissolved solute through the external fluid film around each particle.
The diffusion coefficient of the solute in supercritical CO2 is a critical transport property. In supercritical fluids, diffusivities are 10-100x higher than in liquids (10⁻⁸ vs 10⁻⁹-10⁻¹⁰ m2/s), which is one of the key advantages of SFE over liquid extraction. Molecular simulation computes the diffusion coefficient from the mean-square displacement of the solute molecule in the CO2 simulation, with no empirical correlation needed.
The computed transport properties feed into a continuum-scale model of the extraction vessel. The model solves the mass balance equations for the packed bed, predicting the extraction curve (yield vs. time) and the optimal flow rate, particle size, and cycle time. By varying the CO2 flow rate in the simulation, you identify the transition from kinetically limited extraction (faster flow increases yield) to solubility-limited extraction (faster flow wastes CO2 without increasing yield).
Process Optimization and Scale-Up
SFE process design involves optimizing multiple coupled variables: extraction pressure (higher = more solubility, more pumping cost), temperature (affects both solubility and selectivity), co-solvent type and concentration (increases polar compound recovery, adds separation step), flow rate (faster = more throughput, more CO2 recycling), particle size (smaller = faster extraction, higher pressure drop), and cycle time (longer = higher yield, lower throughput).
The simulation provides the thermodynamic and transport properties needed to optimize this space mathematically rather than experimentally. With solubility, selectivity, and diffusivity as functions of P, T, and co-solvent concentration, the process model computes the extraction yield, purity, and cost for each operating condition. The optimization algorithm finds the Pareto front, the set of conditions that minimize cost while meeting purity and yield targets.
Scale-up of SFE processes is challenging because the mass transfer characteristics change with vessel diameter and bed height. Channeling, wall effects, and axial dispersion are different at pilot scale (1-10 L) versus production scale (100-1000 L). The continuum model, validated against pilot data, predicts production-scale performance by solving the same mass balance equations at the larger geometry with updated fluid dynamics.
Economics of simulation-guided SFE process design:
- Pilot screening campaign (50-100 conditions): $100K-500K, 3-6 months
- Simulation-guided design (100 virtual + 10 pilot conditions): $20K-80K, 4-6 weeks
- CO2 cost optimization: 15-30% reduction in CO2 consumption
- Yield improvement from optimized P/T/co-solvent: 10-25%
- Faster product changeover (new feedstock): days vs. months of pilot work
For companies processing multiple botanical feedstocks (where each feedstock requires separate optimisation), simulation reduces the per-product development cost from $200K+ to under $50K. The simulation pays for itself on the first product and generates compounding returns for every subsequent feedstock. Explore MolSim for supercritical fluid and extraction simulation, or discuss your SFE process design with our engineering team.