Ch · Chemistry

How to Choose Methods and Basis Sets for Different Systems

Published By the 9to5Science editorial team

At a glance

  1. 1. Define the target

    Specify the property, system type, and accuracy needed before choosing a method.

  2. 2. Choose a method family

    Use DFT for practical molecular work, coupled cluster for suitable small molecules, and periodic methods for solids.

  3. 3. Match the basis

    Start with at least triple-zeta quality for reliable energies; add diffuse functions for anions and weak interactions.

  4. 4. Test convergence

    Compare methods and basis sizes, and check spin, relativistic, and multireference effects where relevant.

Choosing a method and basis set depends on what you want to predict—not just on the size of the molecule. A method that gives good equilibrium structures may not give accurate reaction energies, weak intermolecular interactions, or excited states. The recommendations below are practical starting points, not guarantees of accuracy.

Start with the property and system

First specify the target: a geometry, relative energy, barrier, vibrational frequency, excitation energy, or another observable. Also identify features that can change the calculation substantially: net charge, open-shell character, transition metals, heavy elements, or a periodic solid.

The method approximates the electronic structure. The basis set provides the functions used to represent it. Increasing basis-set size reduces basis incompleteness, but it does not fix errors in the electronic-structure method. For example, a larger basis does not automatically make a functional reliable for a difficult transition-metal spin-state energy.

Practical starting points by system

System or goalReasonable starting pointImportant caveat
Main-group molecular structuresA modern hybrid or meta-GGA functional with dispersion treatment; def2-TZVPCheck another functional if the energy difference matters.
Noncovalent complexesA dispersion-inclusive functional, such as ωB97X-D or a dispersion-corrected PBE0 variant, with diffuse, at least triple-zeta functionsCheck basis-set superposition error (BSSE), especially with smaller bases.
Small, well-behaved main-group moleculesMP2 for a lower-cost comparison; CCSD(T) with a correlation-consistent basis for a higher-level checkCoupled cluster is most dependable when the electronic state is predominantly single-reference.
Anions, Rydberg states, and weakly bound speciesUse a basis with diffuse functions, such as aug-cc-pVTZ or an appropriate diffuse-augmented def2 basisWithout diffuse functions, electron density can be too confined and energies or properties misleading.
Transition-metal complexesCompare suitable functionals and, where feasible, higher-level calculations; use a basis and relativistic treatment appropriate to the elementSpin states and bonding can be method-sensitive; there is no universally reliable functional.
Periodic solidsA periodic electronic-structure method, often using plane waves and pseudopotentials or an all-electron periodic basisMolecular Gaussian basis recommendations do not transfer directly to periodic calculations.

Names such as B3LYP, PBE0, and PBE refer to different density functionals, not basis sets. Likewise, “triple-zeta” describes a basis-set quality, not a method. For common molecular calculations, Karlsruhe def2-TZVP is a useful triple-zeta starting point; def2-TZVPP adds further polarization flexibility. For correlated wavefunction calculations, the Dunning aug-cc-pVXZ family is widely used, with “aug” indicating diffuse augmentation and X indicating the cardinal basis size.

A practical workflow

  1. Choose a model appropriate to the question. For a molecular geometry, begin with a validated DFT method that includes dispersion if noncovalent effects may matter. For a precise energy difference, plan a higher-level energy calculation or a systematic method comparison.
  2. Use a basis suited to the electrons. Triple-zeta quality is a sensible target for many final molecular calculations. Add diffuse functions for anions, weakly bound complexes, and many excited-state problems. For heavy elements, use a compatible relativistic approach or effective core potential where appropriate.
  3. Optimize, then check. Optimize the geometry and verify that it is the intended stationary point—for example, a minimum has no imaginary vibrational frequencies. If accurate energies are needed, calculate them with a larger basis or a more accurate method on that geometry, while recognizing that geometry and energy errors are distinct.
  4. Test sensitivity. Repeat key calculations with a second reasonable functional or a larger basis. If the scientific conclusion changes, report the spread and investigate rather than presenting one result as definitive.

Worked example: a hydrogen-bonded water dimer

Suppose you want the interaction energy of two water molecules. Optimize the dimer and evaluate each isolated monomer at its frozen geometry in the dimer, using consistent settings. The uncorrected interaction energy is

Eint=Edimer−EA−EB.E_{\mathrm{int}} = E_{\mathrm{dimer}} - E_A - E_B.

Here, EAE_A and EBE_B are the energies of the monomers at their respective geometries in the dimer, each calculated in its own basis. A negative value means the dimer is lower in energy than the separated monomers under this convention. If the monomers are instead separately optimized, the resulting energy difference also includes their deformation contributions and is not the frozen-fragment interaction energy. With a finite basis, each monomer can borrow functions from the other in the dimer calculation, making the interaction appear too favorable. Estimate this BSSE with the counterpoise method, which computes each monomer at its frozen dimer geometry in the full dimer basis using ghost functions on the other monomer. Compare the corrected and uncorrected values, and, if the result is important, test a larger basis or a suitable higher-level method. This example illustrates why basis choice and a consistent energy definition matter as much as the functional name.

Special cases that need extra care

Excited states: Time-dependent DFT can be useful for many excitations, but conventional functionals may perform poorly for long-range charge-transfer states. A range-separated functional is often worth testing. States with substantial double-excitation character are especially challenging for ordinary TD-DFT.

Open-shell and transition-metal systems: Confirm the intended spin state and inspect spin contamination or other indicators of an unstable electronic description. Compare plausible spin states using consistent methods and basis sets. If results differ significantly between reasonable approaches, that disagreement is a limitation to report, not something a larger basis alone will resolve.

Heavy elements: Relativistic effects can be important. Use a scalar-relativistic Hamiltonian or an effective core potential suited to the element and compatible with the basis; include spin-orbit effects when the property requires them.

Common mistakes

  • Treating one functional as universally best. Performance depends on the property and chemical environment; benchmark against relevant data or compare methods.
  • Using a small basis for final energy differences. Small bases can distort relative energies and noncovalent binding. Check basis convergence when the result matters.
  • Leaving out diffuse functions by default. This is risky for anions and weakly bound or spatially extended electron density.
  • Comparing inconsistent calculations. Use matching geometries, charge and spin conventions, basis sets, and energy definitions when subtracting energies.
  • Assuming a converged calculation is accurate. Numerical convergence only shows that the chosen model was solved consistently; it does not establish that the model is physically adequate.

References

  1. Molecular Electronic-Structure Theory — Trygve Helgaker, Poul Jørgensen, and Jeppe Olsen
  2. Introduction to Computational Chemistry — Frank Jensen
  3. Density functional theory: a practical introduction — David Sholl and Janice A. Steckel

Spotted an error? Send a correction. Read how we write and check articles in our editorial policy.