Clifford gates are unitaries that map Pauli operators to Pauli operators under conjugation: if $U$ is Clifford and $P$ is a Pauli, then $U P U^\dagger$ is also a Pauli (up to a phase). This defines a natural partition in quantum gates with profound implications for simulation, error correction, and computational universality.

• Examples: Pauli X, Y, Z; Hadamard H; Phase gates S and $S^\dagger$; CNOT, CZ, SWAP and all permutations
• Generation: Any product of Clifford gates is Clifford. The Clifford group on $n$ qubits can be generated by H, S, and CNOT
• Simulation: Clifford circuits can be simulated classically in polynomial time using the stabilizer formalism (Gottesman-Knill theorem)
• Expressivity: Clifford gates are “easy” classically but powerful enough to create entanglement, prepare stabilizer states, and implement error correction; they form the backbone of quantum error-correcting codes

• Examples: T gate, $T^\dagger = T^{-1}$, phase rotations like $R_Z(\theta)$ for non-multiples of $\pi/2$, $R_X(\theta)$ and $R_Y(\theta)$ with arbitrary angles
• Property: Non-Clifford gates take Pauli operators outside the Pauli group under conjugation; a single T gate makes a circuit impossible to simulate classically
• Universality: The T gate is the minimal non-Clifford addition needed for universal quantum computation; Clifford + T is universal, while Clifford alone is not
• Implementation cost: Each T gate is expensive to implement fault-tolerantly in quantum error-correcting codes; T-count (number of T gates) is a primary optimization target in quantum circuit synthesis

Stabilizer simulators can efficiently simulate Clifford circuits with up to ~1000 qubits. Adding even one non-Clifford gate layer requires exponential resources.

Transversal implementations of Clifford gates are typically straightforward in surface codes and CSS codes. Non-Clifford gates (especially T) require magic state distillation—a resource-intensive procedure.

Quantum compilers often separate circuits into Clifford+T form, then minimize T-count. Techniques like T-teleportation and optimized T-factories reduce the T-gate overhead.

VQE and QAOA can use Clifford ansätze for certain problems, reducing classical overhead, though they sacrifice expressivity compared to full qubit rotations.

Clifford operations preserve the stabilizer group structure: if a state has stabilizers $\{S_1, \ldots, S_k\}$, applying a Clifford $U$ gives stabilizers $\{U S_1 U^\dagger, \ldots, U S_k U^\dagger\}$, which are still Paulis. Non-Clifford gates can map stabilizer states out of the stabilizer subspace.

• Clifford only: classically simulable, limited expressivity
• Clifford + T: universal for quantum computation, requires magic states
• Clifford + arbitrary rotations: also universal, but harder to implement fault-tolerantly
• All single-qubit + CNOT: most flexible, standard in theory