Wavefunction-Based Excited-State Methods

Introduction

When a predicted electronic spectrum is sensitive to the density functional, a wavefunction method can provide a different route to the excited states. Configuration interaction and equation-of-motion coupled-cluster methods build states from controlled changes in electronic configurations rather than only a density response. They can clarify the character and energy of transitions, but their accuracy depends on excitation rank, basis set and whether one reference determinant is adequate. The most expensive calculation is not automatically the best answer if the molecule has strong multireference character or the experimental comparison is poorly defined.

Core explanation

A configuration describes how electrons occupy spin orbitals while respecting antisymmetry. Starting from a Hartree–Fock reference determinant, a singly excited determinant replaces one occupied spin orbital with one virtual spin orbital. A doubly excited determinant replaces two. Configuration interaction (CI) combines determinants linearly, with coefficients determined by diagonalizing the electronic Hamiltonian in the chosen space. Full CI within a finite orbital basis would be exact for that Hamiltonian and basis, but its determinant count grows rapidly. Practical truncated CI includes selected excitation ranks and thus introduces method error in addition to basis error.

Configuration interaction singles, CIS, includes only the reference and single substitutions for excited-state description. It can provide a qualitative map of states dominated by one-electron promotion at relatively modest cost, but it omits much dynamical electron correlation. Its excitation energies commonly have systematic errors, so a plausible state ordering does not establish precise band positions. CIS also struggles when a target state has substantial double-excitation character. Primary work assessing CIS variants documents its ground-reference bias and excitation-energy limitations. Adding perturbative or more extensive correlation improves some cases but increases computational cost and does not remove the need to inspect state character.

Equation-of-motion coupled-cluster (EOM-CC) begins with a correlated coupled-cluster reference, such as CCSD, and applies an excitation operator to it. In EOM-CCSD the operator includes singles and doubles, and solving a non-Hermitian eigenvalue problem gives excitation energies relative to the correlated ground state. Unlike a simple CIS calculation on an uncorrelated reference, the ground and target states benefit from a substantial correlation treatment. For suitable single-reference molecules, EOM-CCSD often provides a useful benchmark for valence excited states; higher excitation ranks such as triples can improve demanding cases at considerably greater cost. Primary benchmark research on medium-sized molecules compared EOM and related wavefunction methods against high-level theoretical best estimates for excitation energies and oscillator strengths.

Method names do not define an observable by themselves. A vertical excitation energy fixes nuclear coordinates, typically at the ground-state minimum. An adiabatic excitation or 0–0 energy incorporates different optimized structures and zero-point contributions. Oscillator strength describes transition intensity and needs a transition-property calculation, not only two total energies. A spectrum recorded in solvent or a solid adds environmental shifts and broadening. Benchmark a calculated quantity against the corresponding experimental or high-level reference quantity rather than comparing a vertical gas-phase number with an arbitrary solution band maximum.

Basis sets matter particularly for diffuse Rydberg states, whose excited electron density extends far from the molecular core. Charge-transfer states also challenge methods and may require a basis and reference capable of describing spatial separation. States with substantial double-excitation character can require higher-rank EOM operators or multireference treatment; EOM-CCSD is not a universal cure for every TDDFT difficulty. At a stretched bond, a single Hartree–Fock determinant can become a poor ground reference, and then even a sophisticated excitation operator sits on the wrong foundation. Compare diagnostics, orbital occupations and state-character analyses before assigning a benchmark label. Primary tests of EOM-CCSD variants include valence, Rydberg and doubly excited examples.

Step-by-step reasoning

1. Specify whether the target is a vertical absorption energy, adiabatic gap, emission energy or oscillator strength. 2. Inspect the ground-state electronic character to judge whether one reference determinant is reasonable. 3. Select a CI or EOM method with excitation rank appropriate to the expected state character. 4. Choose a basis that resolves valence and, where relevant, diffuse excited density; test convergence. 5. Calculate enough roots and analyze configurations or transition densities so similar states are tracked correctly. 6. Compare with a matching experimental or higher-level quantity, including geometry and environment limits.

Visual explanation

Draw a reference determinant as filled occupied boxes below empty virtual boxes. One upward arrow moves one electron for a single excitation; two arrows show a double. Next draw a tree: CIS retains the reference and single branches, while a richer CI adds double branches. EOM-CC begins from a correlated ground-state trunk and applies excitation operators. The diagram is schematic: each state is generally a mixture, not one isolated orbital arrow.

Real-world analogy

Describing an orchestra with only one instrument changed at a time can capture many variations but misses arrangements that require two simultaneous changes. CIS is analogous to that restricted description. EOM-CC starts from a more accurately tuned full ensemble before generating variations. The analogy emphasizes model space; it does not imply that electronic excitations are literally instruments being swapped.

Real-world example

A computational group studies a small chromophore whose first bright absorption is predicted at noticeably different energies by two functionals. The ground state is well described by one determinant, and state analysis suggests an ordinary valence single excitation. The group computes EOM-CCSD vertical energies with a converged basis as an independent benchmark and compares oscillator strengths as well as positions. If a second state instead has strong double-excitation character, the group flags EOM-CCSD uncertainty and seeks a higher-rank or multireference reference rather than declaring the first benchmark sufficient for every state.

Why?

Why can adding more determinants improve an excited-state wavefunction? Electron configurations beyond one promoted electron allow correlated rearrangements of the remaining electrons. The Hamiltonian mixes configurations that have compatible symmetry, and the resulting eigenstate can lower its energy and acquire more realistic transition properties. However, an incomplete determinant selection can improve some states more than others, changing an excitation difference unpredictably. Both the reference and excited-state errors must be considered.

Common misconception

“One HOMO-to-LUMO arrow is the excited-state wavefunction.” It may be only a dominant contribution to a mixture. “EOM-CCSD is exact because it includes doubles” ignores higher excitations and basis limits. “A better ground-state total energy guarantees a better excitation energy” is false if errors cancel differently between states. “Agreement with one absorption maximum validates every excited state” confuses distinct states and experimental observables.

Worked example

At a fixed ground-state geometry, suppose a hypothetical method gives E0 = −500.000 hartree and E1 = −499.900 hartree. The vertical difference is 0.100 hartree, about 2.72 eV using 1 hartree ≈ 27.21 eV. A second method may shift both total energies downward substantially but change their difference only to 2.60 eV; total-energy lowering alone does not show which excitation is better. If the first state is largely singly excited but another state is approximately an even mixture of two-electron promotions, their method uncertainties can differ. To compare with an experiment reported as a 0–0 band in solution, one would also need excited-state relaxation, vibrational and solvent effects.

Quick check

1. What configurations does CIS primarily add to a reference determinant? Answer: Determinants formed by single occupied-to-virtual substitutions. 2. Does an EOM-CCSD vertical energy include excited-state geometry relaxation automatically? Answer: No. It is evaluated at the specified fixed geometry unless a separate excited-state optimization is performed.

Exam focus

Explain the determinant-expansion idea, the difference between singles and doubles, and why full CI becomes expensive. State what EOM-CCSD adds relative to a simple CIS picture, while naming its single-reference and excitation-rank limits. Convert an energy difference from hartree to eV correctly. For any proposed benchmark, identify the target state character and whether the compared experimental quantity is vertical, adiabatic or a broadened band.

Advanced insight

Truncated CI is not generally size extensive, while coupled-cluster constructions have favorable connected-cluster behavior for many ground-state applications; this matters when comparing systems of different size. Excited-state calculations also face root tracking: two states may cross or mix as geometry changes, so their order in an output list need not preserve identity. Natural transition orbitals, transition densities and dominant configuration weights help track character, but strong multireference states may need methods built around multiple important determinants. High-level theoretical benchmarks carry basis and model uncertainty too, which should be stated rather than treated as exact data.

Summary

Wavefunction excited-state methods represent electronic states through mixtures of configurations. CIS is an affordable qualitative starting point, while EOM-CCSD adds substantial correlated ground-state and excitation treatment for suitable single-reference states. Accuracy still depends on basis, excitation rank and state character, particularly for Rydberg, double-excitation and bond-breaking cases. A trustworthy spectral comparison specifies geometry, environment and whether energy or intensity is being assessed.

Practice questions

1. What does a double excitation change relative to a reference determinant? Answer: It replaces two occupied spin orbitals with two virtual spin orbitals. 2. A vertical gap is 0.080 hartree. Approximately how many eV is that? Answer: 0.080 × 27.21 ≈ 2.18 eV. 3. Why can CIS miss a state dominated by a two-electron promotion? Answer: Its configuration space lacks the necessary double-excitation determinants. 4. Why should an EOM-CCSD result be questioned near a strongly stretched bond? Answer: The single-reference ground state may be inadequate and the target can require stronger multireference character.