Quantum Chemistry Benchmarks

A quantum-chemistry benchmark is a specified Hamiltonian, representation, state, observable, and acceptance criterion. A molecule name and an energy are not enough. Small exact fixtures should test implementation before larger molecular comparisons test basis and correlation approximations.

Helpful background. Molecular integrals, Hartree–Fock computation, and configuration interaction identify the main numerical layers.

What must be held fixed

Record nuclear coordinates and units, charges and spin, orbital basis data and angular conventions, electron count, frozen cores and active orbitals, integral approximations, occupation restrictions, and whether nuclear repulsion is included. Identify the software version and numerical thresholds. A basis label can hide normalization or pure-versus-Cartesian conventions; compare overlap and one-electron integrals when two implementations disagree.

Use absolute energy tolerances in stated units and separate tolerances for residuals, densities, forces, and energy differences. Set them from the scientific question and expected numerical error. Printing twelve digits does not imply twelve accurate digits.

A hierarchy of checks is more informative than one reference number: analytic integrals, one-electron cancellation, small determinant diagonalization, a molecular geometry point, and a scan or separated-fragment limit.

An analytic helium variational fixture

For a fixed helium nucleus with charge $Z=2$, take two opposite-spin electrons in the normalized spatial orbital

$$\phi_\zeta(\mathbf r) =(\zeta^3/\pi)^{1/2}e^{-\zeta r}.$$

In atomic units the single-determinant energy is

$$E(\zeta)=\zeta^2-2Z\zeta+\frac58\zeta.$$

The first two terms are the total kinetic and nuclear attraction energies; $5\zeta/8$ is the Coulomb repulsion for this exponential orbital. Minimization gives

$$\zeta=Z-\frac5{16}=\frac{27}{16},\qquad E_{\min}=-\left(\frac{27}{16}\right)^2=-2.84765625.$$

This is an exact result for this one-parameter ansatz, not the exact helium energy and not the complete-basis Hartree–Fock limit. A numerical quadrature or variational code using this very ansatz should reproduce it. A Gaussian-basis code solves a different trial problem and should not be required to reproduce the number exactly.

A two-determinant algebraic fixture

Choose a dimensionless model matrix

$$H=\begin{pmatrix}-1&0.1\\0.1&0\end{pmatrix}.$$

Its energies are

$$E_\pm=\frac{-1\pm\sqrt{1.04}}2.$$

The lower value is approximately $-1.0099019514$. Compare both eigenvalues, normalized coefficients up to phase, residuals, and the energy reconstructed from reduced densities. The fixture is a numerical test Hamiltonian, not a claimed molecular dataset.

Two noninteracting copies then test energy additivity. Full CI over the product space must give twice the ground energy. A truncated excitation space may fail this check for structural reasons; that failure should be identified as size inconsistency rather than hidden by a loose tolerance.

Molecular tests with clear interpretation

For molecular hydrogen, compare restricted Hartree–Fock and full CI in the same small basis at a fixed internuclear distance, then scan toward dissociation. The finite-basis full-CI result separates correlation error from the eigensolver error; enlarging the orbital basis addresses another limit. Restricted Hartree–Fock dissociation failure is a useful physical diagnostic even if every SCF iteration converges.

For forces, compare analytic derivatives with centered differences over a step-size sweep. For vibrational Hessians, require tighter electronic convergence because second differences amplify noise. A vibrational-frequency comparison must also use the same isotope masses and frequency convention.

Keep inputs and expected mathematical identities alongside a result. Independent implementations are most useful when their conventions have been reconciled; two solvers using the same erroneous integrals can agree while both give the wrong answer.

Exercises

At the helium optimum, evaluate the kinetic and nuclear-attraction contributions and check the stationary scaling relation.

Solution

At $\zeta=27/16$, the kinetic energy is $T=\zeta^2=2.84765625$ and the nuclear attraction is $V_{\rm nuc}=-4\zeta=-6.75$ hartree. Adding the electron repulsion $5\zeta/8=1.0546875$ gives the full potential energy $V=-2\zeta^2$. Thus $2T+V=0$ and $E=-T$. This checks optimization with respect to uniform orbital scaling within the ansatz.

References

• Bartlett, R. J., and M. Musiał, “Coupled-cluster theory in quantum chemistry,” Reviews of Modern Physics 79, 291–352 (2007), doi:10.1103/RevModPhys.79.291 — correlation hierarchies and separability.
• Szabo, A., and N. S. Ostlund, Modern Quantum Chemistry: Introduction to Advanced Electronic Structure Theory, Dover, 1996 — molecular benchmarks and finite-basis comparisons.