Geometry Optimization Overview

Geometry optimization finds the arrangement of atoms that minimizes (or maximizes) the energy of a molecular system. Starting from an initial guess geometry, the optimization algorithm evaluates the energy and its gradient (the forces on the atoms) at each step and adjusts the atomic positions to move toward a stationary point on the potential energy surface. The default optimization algorithm in Jaguar is BFGS, a quasi-Newton method that builds up an approximate Hessian (matrix of second derivatives) over the course of the optimization.

Most geometry optimizations seek a minimum-energy structure, which represents a stable molecular geometry. Transition state searches seek a first-order saddle point, which has one direction of negative curvature. See Transition-State Optimizations for details.

Local and Global Minima

An optimization converges when the energy gradient and displacement between successive steps fall below a set of convergence thresholds (see SCF and Geometry Convergence). However, the converged structure is only guaranteed to be a local minimum, not the global minimum. For flexible molecules with multiple conformations, the result depends on the starting geometry. A conformational search (e.g., with MacroModel) can help identify the global minimum. To verify that the converged structure is a true minimum rather than a saddle point, perform a vibrational frequency calculation: a minimum has all positive frequencies, while a saddle point has one or more imaginary frequencies.

Effect of Method and Basis Set

The choice of functional and basis set affects the shape of the potential energy surface and therefore the optimized geometry. However, molecular structures are generally less sensitive to the level of theory than energies, conformational energies, or reaction energies. For guidance on basis set selection, see Basis Set Overview. For energy-related properties, a higher level of theory should be used, often as a single-point energy calculation on the optimized structure (see Multi-Level Approaches below).

Multi-Level Approaches

A common strategy in computational chemistry is to optimize the geometry at a computationally efficient level of theory, then compute the final energy at a more accurate level. This multi-level approach exploits the fact that structures are less sensitive to basis set and functional choice than energies. For example:

  • Optimize the geometry with an (m)GGA functional and a DZ basis set (e.g., r2SCAN-D4/def2-SVP or PBE-D4/def2-SVP).

  • Compute the final energy with a hybrid or range-separated hybrid functional and a TZ or QZ basis set (e.g., ωB97M-V/def2-TZVPPD).

This approach can provide results close to those of a full high-level optimization at a fraction of the computational cost. The 3c Composite Methods for Geometry Optimization described below are also well suited as the low-level optimization step in this kind of workflow.

3c Composite Methods for Geometry Optimization

The 3c composite methods are purpose-built for efficient and accurate geometry optimization. They combine a density functional with a small, tailored basis set and empirical corrections for dispersion and basis set errors (geometric counterpoise correction). These methods balance accuracy, robustness, and computational efficiency, making them suitable for routine structure optimizations, including large systems where hybrid functionals would be prohibitively expensive.

Jaguar supports the following 3c composite methods:

  • r2SCAN-3c — meta-GGA composite with def2-mTZVPP basis, gCP, and D4 dispersion. Recommended as a robust general-purpose method for geometry optimization.

  • B97-3c — GGA composite with def2-mTZVP basis, short-range basis correction, and D3(BJ) dispersion.

  • PBEh-3c — hybrid composite with def2-mSVP basis, gCP, and D3(BJ) dispersion. Uses a small DZ basis set for maximum efficiency; recommended for preliminary optimizations or very large systems.

  • ωB97X-3c — range-separated hybrid composite with vDZP basis and D4 dispersion.

  • HF-3c — Hartree-Fock composite with MINIX basis, short-range basis correction, gCP, and D3(BJ) dispersion. The most economical option; recommended for preliminary optimizations or very large systems.

When a 3c method is selected, Jaguar automatically applies the matching basis set and corrections. To use a 3c method, set Composite methods in the gen section of the input file (e.g., dftname=r2SCAN-3c), or select the method from the Theory tab in Maestro.

For further details on recommended methods and a decision matrix for choosing functional and basis set combinations by task, see Bursch et al. [330].

Practical Tips

For practical suggestions on setting up and troubleshooting geometry optimizations in Jaguar, see Jaguar Geometry Optimization Suggestions. For a tutorial exercise, see Introduction to Geometry Optimizations, Functionals, and Basis Sets.