Multistage optimization allows users to obtain the most accurate energetics for the lowest computational cost by optimizing at one level of theory and obtaining the single point-energy at a higher level of theory. This is controlled via a mode ("Reckless," "Rapid," "Careful," "Meticulous") that sets the level of theory for a desired accuracy. The molecule is optionally pre-optimized with xTB (useful if the molecule was hand drawn) and then optimized with the desired level of theory. Finally a single-point energy is taken. The optimization can take place in solvent, and if desired, frequencies can be computed at the same level of theory as the optimization.
Transition state optimizations can be performed, including for xTB thanks to the use of the geomeTRIC optimizer. A guess Hessian is computed using either a neural network potential or xTB to accelerate the geometry optimization and increase the likelihood of success.
| Mode | Pre-Optimization | Initial Optimization | Final Optimization + Frequencies | Single Point |
|---|---|---|---|---|
| Reckless | GFN-FF | GFN2-xTB | ||
| Rapid | GFN-FF | GFN2-xTB | r²SCAN-3c | |
| Careful | GFN2-xTB | r²SCAN-3c | ωB97X-3c | |
| Meticulous | GFN2-xTB | r²SCAN-3c | ωB97X-3c | ωB97M-D3BJ/def2-TZVPPD |
"Rapid" is recommended for most work, while "Careful" is recommended for publication quality results.