The Classification of Clifford Gates over Qubits

We examine the following problem: given a collection of Clifford gates, describe the set of unitaries generated by circuits composed of those gates. Specifically, we allow the standard circuit operations of composition and tensor product, as well as ancillary workspace qubits as long as they start and end in states uncorrelated with the input, which rule out common “magic state injection” techniques that make Clifford circuits universal. We show that there are exactly 57 classes of Clifford unitaries and present a full classification characterizing the gate sets which generate them. This is the first attempt at a quantum extension of the classification of reversible classical gates introduced by Aaronson et al., another part of an ambitious program to classify all quantum gate sets. The classification uses, at its center, a reinterpretation of the tableau representation of Clifford gates to give circuit decompositions, from which elementary generators can easily be extracted. The 57 different classes are generated in this way, 30 of which arise from the single-qubit subgroups of the Clifford group. At a high level, the remaining classes are arranged according to the bases they preserve. For instance, the CNOT gate preserves the X and Z bases because it maps X-basis elements to X-basis elements and Z-basis elements to Z-basis elements. The remaining classes are characterized by more subtle tableau invariants; for instance, the T_4 and phase gate generate a proper subclass of Z-preserving gates.

💡 Research Summary

The paper tackles a fundamental question in quantum circuit theory: given an arbitrary set of Clifford gates, what is the full set of unitaries that can be realized using only those gates under a natural circuit model? The model permits composition (serial connection), tensor product (parallel connection), arbitrary qubit swaps, and the use of ancillary qubits that must start and end in a product state uncorrelated with the computational input. Crucially, the model disallows “magic‑state injection” or any non‑Clifford resource that would boost computational power beyond the Clifford group.

The authors’ central technical tool is a refined version of the stabilizer tableau, a binary matrix representation of Clifford operations. By extending the tableau to include a phase vector, they obtain a compact description of how each gate conjugates Pauli strings. This representation makes it possible to define a family of invariants—properties of the tableau that remain unchanged under the allowed circuit operations. The most basic invariants capture whether a gate preserves the X, Y, or Z Pauli bases. For example, the CNOT gate preserves the X‑basis and the Z‑basis but not the Y‑basis. These three basis‑preserving invariants already generate a three‑fold symmetry in the classification.

Beyond the high‑level basis preservation, the authors identify finer invariants that look at specific sub‑structures of the tableau: whether the induced transformation in a given basis is a full linear map, an orthogonal map, or merely a sign‑flip; whether certain off‑diagonal blocks vanish; and whether particular combinations of Pauli operators are mapped to each other up to phase. These subtler invariants distinguish classes that share the same basis‑preserving behavior but differ in how they act on the underlying Pauli group.

Using these invariants, the authors prove that there are exactly 57 distinct Clifford classes. Thirty of them arise from the single‑qubit Clifford subgroups; the remaining 27 are non‑degenerate, multi‑qubit classes. The classes are organized in an inclusion lattice (Figures 1 and 2 in the paper). Nodes are colored according to X‑, Y‑, or Z‑preservation, and edges indicate strict inclusion. Each class is characterized by a small generating set of gates, often a single gate acting on at most four qubits. Remarkably, any set of Clifford gates that generates a given class can be reduced to three gates that already generate the same class, and in many cases a single gate suffices.

A key consequence is that the traditional generating set {CNOT, Hadamard, Phase} is not minimal in this model. The authors show that CNOT together with Hadamard, together with appropriate ancilla preparation and disposal, can synthesize the Phase gate, confirming a known result in a new, systematic way. Moreover, every class can be generated using only a constant number of ancillary states; no class requires arbitrarily large entangled ancillae.

The paper also provides explicit canonical forms for each class, derived from the tableau decomposition. For two‑qubit Clifford circuits, a particularly compact normal form is given in Appendix E. From these normal forms the authors derive exact counting formulas for the number of n‑qubit gates in each class, showing that each class is exponentially smaller than any strictly larger class.

Algorithmically, the classification yields a linear‑time decision procedure: given the tableau of an unknown Clifford gate, one inspects a constant‑size subset of bits to determine which of the 57 classes the gate belongs to. This makes the result immediately applicable to compiler optimizations, error‑correction code analysis, and the study of fault‑tolerant subroutines.

Finally, the authors discuss several “sporadic” classes that arise from intricate interactions of the invariants, such as those generated by the T₄ gate combined with the phase gate. Some of these require a four‑qubit generator, while others admit a three‑qubit generator, illustrating the richness of the Clifford landscape even under severe restrictions.

In summary, the work delivers a complete, mathematically rigorous taxonomy of Clifford gate sets under a realistic circuit model, establishes concrete generation and counting results, and supplies efficient algorithms for class identification. It lays a solid foundation for the broader program of classifying quantum gate sets, bridging the gap between the well‑understood Clifford world and the far more complex landscape of general quantum operations.