Unitary Coupled-Cluster Ansatz

The quality of a VQE result is capped by the ansatz: the trial state can only ever be as good as the family of states the circuit can reach. The Unitary Coupled-Cluster (UCC) ansatz is the chemistry-native choice — it is derived from the classical coupled-cluster method, the long-standing gold standard for molecular energies, and made unitary so a quantum circuit can implement it.

From classical CC to unitary CC

Classical coupled cluster writes the wavefunction as

|\Psi_{\text{CC}}\rangle = e^{\hat T}\,|\Psi_{\text{HF}}\rangle,

where the cluster operator $\hat T = \hat T_1 + \hat T_2 + \cdots$ is a sum of excitation operators that promote electrons from occupied to virtual orbitals. The trouble for a quantum computer is that $e^{\hat T}$ is not unitary ( $\hat T$ is not anti-Hermitian), so it cannot be implemented as a circuit. UCC fixes this by symmetrizing the exponent:

|\Psi_{\text{UCC}}\rangle = e^{\hat T - \hat T^\dagger}\,|\Psi_{\text{HF}}\rangle.

Because $\hat T - \hat T^\dagger$ is anti-Hermitian, $e^{\hat T - \hat T^\dagger}$ is unitary and can be realized as a gate sequence after Jordan–Wigner mapping.

Singles and doubles: UCCSD

Truncating the cluster operator at single and double excitations gives UCCSD, the most common variant:

\hat T_1 = \sum_{i \in \text{occ},\, a \in \text{virt}} \theta_i^a\, a_a^\dagger a_i, \qquad \hat T_2 = \sum_{ij \in \text{occ},\, ab \in \text{virt}} \theta_{ij}^{ab}\, a_a^\dagger a_b^\dagger a_j a_i.

The amplitudes $\{\theta\}$ are the variational parameters VQE optimizes. $\hat T_1$ moves one electron from an occupied orbital $i$ to a virtual orbital $a$ ; $\hat T_2$ moves a pair. For $\text{H}_2$ in the minimal basis there is exactly one double excitation (both electrons from the bonding to the antibonding orbital), so UCCSD reduces to the single-parameter ansatz from the previous lesson — the Givens rotation is the $\text{H}_2$ UCC double excitation.

Trotterizing the exponential

The operator $\hat T - \hat T^\dagger$ is a sum of (generally non-commuting) anti-Hermitian terms, so its exponential is approximated by a Trotter decomposition:

e^{\hat T - \hat T^\dagger} \approx \prod_k e^{\theta_k (\hat\tau_k - \hat\tau_k^\dagger)},

where each $\hat\tau_k$ is a single excitation operator. After Jordan–Wigner each factor becomes a short sequence of CNOTs and a single parametrized rotation — a structure sometimes called an excitation gate or Givens rotation gate. A single Trotter step usually suffices because the parameters are re-optimized variationally, absorbing the Trotter error.

Where UCC fits

| Ansatz | Depth | Symmetry-preserving | Trainability | |---|---|---|---| | Hardware-efficient | shallow | no | barren-plateau risk | | UCCSD | deep ( $O(n^4)$ doubles) | yes | reliable | | ADAPT-VQE (grown UCC) | adaptive | yes | reliable, compact |

Modern practice often uses ADAPT-VQE, which grows the ansatz operator-by-operator from a UCC pool, adding only the excitations that most lower the energy. This keeps the circuit as shallow as the molecule allows while retaining UCC's physical structure. For the small systems in this module, the bare single-parameter UCC ansatz already reaches the exact ground state, as the next lesson's dissociation curve will show.