Qutrit Clifford+T Set

• The qutrit Clifford+T set is an extension of the Clifford group by a non-Clifford T gate using ninth roots of unity, enabling canonical representations and uniqueness proofs.
• It supports exact synthesis and normalization algorithms that optimize T-count through cyclotomic arithmetic and least denominator exponent analyses.
• The framework extends to multi-qutrit controlled operations and reversible ternary computation, offering scalable techniques for fault-tolerant quantum circuit design.

The qutrit Clifford+T set is the extension of the qutrit Clifford group by a diagonal non-Clifford gate built from ninth roots of unity. In the single-qutrit setting it supports canonical and normal forms, exact synthesis, uniqueness theorems, and T-count optimality; in the multi-qutrit setting it underlies controlled-gate constructions, exact synthesis over cyclotomic rings, and fault-tolerant resource analyses (Glaudell et al., 2018, Prakash et al., 2018, Yeh et al., 2022, Kalra et al., 2024).

1. Algebraic setting and gate conventions

Let $\omega=e^{2\pi i/3}$ and $\zeta=e^{2\pi i/9}$. In one common $SU(3)$-normalized convention, the single-qutrit Pauli operators are

$$X=\begin{bmatrix} 0&0&1\ 1&0&0\ 0&1&0 \end{bmatrix},\qquad Z=\mathrm{diag}(1,\omega,\omega^2),$$

the Clifford generators are

$$H=\frac{1}{\sqrt{-3}} \begin{bmatrix} 1&1&1\ 1&\omega&\omega^2\ 1&\omega^2&\omega \end{bmatrix},\qquad S=\zeta^8\,\mathrm{diag}(1,1,\omega),$$

and the qutrit $T$ gate is

$T=\mathrm{diag}(1,\zeta,\zeta^8).$

In this convention the Pauli group has $27$ elements, or $9$ if global phases are ignored, and the single-qutrit Clifford group has $648$ elements, or $\zeta=e^{2\pi i/9}$0 up to global phase. The gate $\zeta=e^{2\pi i/9}$1 lies in the third level of the Clifford hierarchy, and the relations

$\zeta=e^{2\pi i/9}$2

are the basic rewriting identities for the canonical-form theory (Glaudell et al., 2018).

The literature also uses alternative but related conventions. One single-qutrit normal-form paper uses $\zeta=e^{2\pi i/9}$3 and the Howard–Vala gate

$\zeta=e^{2\pi i/9}$4

while later hierarchy work presents the diagonal representative

$\zeta=e^{2\pi i/9}$5

and states that it is Clifford-equivalent to a standard qutrit $\zeta=e^{2\pi i/9}$6 representative. A hierarchy classification paper also singles out

$\zeta=e^{2\pi i/9}$7

as a canonical level-3 qutrit gate (Prakash et al., 2018, Yeh, 2023, Silva et al., 14 Jan 2025).

2. Canonical and normal forms for single-qutrit circuits

A central result is the existence of a canonical form for every single-qutrit Clifford+T operator. In the formulation of Glaudell, Ross, and Taylor, the canonical circuits are exactly those matching

$\zeta=e^{2\pi i/9}$8

where $\zeta=e^{2\pi i/9}$9 is an arbitrary Clifford operator. The six allowed non-Clifford syllables are precisely the elements of $SU(3)$0, with $SU(3)$1, $SU(3)$2, and the Clifford decomposition $SU(3)$3. The same paper also gives a channel-form reinterpretation: for $SU(3)$4,

$SU(3)$5

with $SU(3)$6, $SU(3)$7, and $SU(3)$8. Channel forms are in bijection with canonical forms and impose the qutrit analogue of the qubit rule that consecutive rotations must be about different axes (Glaudell et al., 2018).

A distinct normal-form formalism appears in the work of Prakash, Jain, Kapur, and Seth. There the operator is written in

$SU(3)$9

where $$X=\begin{bmatrix} 0&0&1\ 1&0&0\ 0&1&0 \end{bmatrix},\qquad Z=\mathrm{diag}(1,\omega,\omega^2),$$0, $$X=\begin{bmatrix} 0&0&1\ 1&0&0\ 0&1&0 \end{bmatrix},\qquad Z=\mathrm{diag}(1,\omega,\omega^2),$$1 with $$X=\begin{bmatrix} 0&0&1\ 1&0&0\ 0&1&0 \end{bmatrix},\qquad Z=\mathrm{diag}(1,\omega,\omega^2),$$2, $$X=\begin{bmatrix} 0&0&1\ 1&0&0\ 0&1&0 \end{bmatrix},\qquad Z=\mathrm{diag}(1,\omega,\omega^2),$$3, $$X=\begin{bmatrix} 0&0&1\ 1&0&0\ 0&1&0 \end{bmatrix},\qquad Z=\mathrm{diag}(1,\omega,\omega^2),$$4, and $$X=\begin{bmatrix} 0&0&1\ 1&0&0\ 0&1&0 \end{bmatrix},\qquad Z=\mathrm{diag}(1,\omega,\omega^2),$$5. In that convention $$X=\begin{bmatrix} 0&0&1\ 1&0&0\ 0&1&0 \end{bmatrix},\qquad Z=\mathrm{diag}(1,\omega,\omega^2),$$6, $$X=\begin{bmatrix} 0&0&1\ 1&0&0\ 0&1&0 \end{bmatrix},\qquad Z=\mathrm{diag}(1,\omega,\omega^2),$$7, and $$X=\begin{bmatrix} 0&0&1\ 1&0&0\ 0&1&0 \end{bmatrix},\qquad Z=\mathrm{diag}(1,\omega,\omega^2),$$8, so $$X=\begin{bmatrix} 0&0&1\ 1&0&0\ 0&1&0 \end{bmatrix},\qquad Z=\mathrm{diag}(1,\omega,\omega^2),$$9 factors can be bubbled to the right and the $$H=\frac{1}{\sqrt{-3}} \begin{bmatrix} 1&1&1\ 1&\omega&\omega^2\ 1&\omega^2&\omega \end{bmatrix},\qquad S=\zeta^8\,\mathrm{diag}(1,1,\omega),$$0-syllables alternate with $$H=\frac{1}{\sqrt{-3}} \begin{bmatrix} 1&1&1\ 1&\omega&\omega^2\ 1&\omega^2&\omega \end{bmatrix},\qquad S=\zeta^8\,\mathrm{diag}(1,1,\omega),$$1-syllables (Prakash et al., 2018).

Both constructions are presented as qutrit analogues of Matsumoto–Amano normal forms for single-qubit Clifford+T. The differences are structural rather than merely cosmetic: qutrit normal forms use ninth roots of unity, a richer Clifford coset structure, and cyclotomic denominator analyses rather than the dyadic techniques of the qubit case.

3. Uniqueness, T-optimality, and exact synthesis

The canonical-form program establishes three properties simultaneously. First, every single-qutrit Clifford+T operator has a canonical representative. Second, there is a normalization algorithm that runs in time linear in the number of $$H=\frac{1}{\sqrt{-3}} \begin{bmatrix} 1&1&1\ 1&\omega&\omega^2\ 1&\omega^2&\omega \end{bmatrix},\qquad S=\zeta^8\,\mathrm{diag}(1,1,\omega),$$2 gates: after reducing each power $$H=\frac{1}{\sqrt{-3}} \begin{bmatrix} 1&1&1\ 1&\omega&\omega^2\ 1&\omega^2&\omega \end{bmatrix},\qquad S=\zeta^8\,\mathrm{diag}(1,1,\omega),$$3 to a bounded form, right-multiplication by a generator updates only a bounded suffix of the canonical circuit, at most six operators, and never increases T-count. Third, the canonical form is unique. The uniqueness proof uses an $$H=\frac{1}{\sqrt{-3}} \begin{bmatrix} 1&1&1\ 1&\omega&\omega^2\ 1&\omega^2&\omega \end{bmatrix},\qquad S=\zeta^8\,\mathrm{diag}(1,1,\omega),$$4-dimensional real adjoint representation $$H=\frac{1}{\sqrt{-3}} \begin{bmatrix} 1&1&1\ 1&\omega&\omega^2\ 1&\omega^2&\omega \end{bmatrix},\qquad S=\zeta^8\,\mathrm{diag}(1,1,\omega),$$5, cyclotomic rings such as $$H=\frac{1}{\sqrt{-3}} \begin{bmatrix} 1&1&1\ 1&\omega&\omega^2\ 1&\omega^2&\omega \end{bmatrix},\qquad S=\zeta^8\,\mathrm{diag}(1,1,\omega),$$6 and $$H=\frac{1}{\sqrt{-3}} \begin{bmatrix} 1&1&1\ 1&\omega&\omega^2\ 1&\omega^2&\omega \end{bmatrix},\qquad S=\zeta^8\,\mathrm{diag}(1,1,\omega),$$7, a residue map $$H=\frac{1}{\sqrt{-3}} \begin{bmatrix} 1&1&1\ 1&\omega&\omega^2\ 1&\omega^2&\omega \end{bmatrix},\qquad S=\zeta^8\,\mathrm{diag}(1,1,\omega),$$8, and least denominator exponent analysis on quadrants of $$H=\frac{1}{\sqrt{-3}} \begin{bmatrix} 1&1&1\ 1&\omega&\omega^2\ 1&\omega^2&\omega \end{bmatrix},\qquad S=\zeta^8\,\mathrm{diag}(1,1,\omega),$$9. For a canonical form with T-count $T$0, the least denominator exponent of $T$1 is $T$2, and the corresponding residue pattern determines the leftmost syllable unambiguously. Uniqueness together with the non-increasing rewrite algorithm implies T-optimality: among all single-qutrit Clifford+T circuits implementing a fixed operator, the canonical form uses the least number of $T$3 gates (Glaudell et al., 2018).

The parallel exact-synthesis theory of Prakash, Jain, Kapur, and Seth is phrased in the ring $T$4 with $T$5. For $T$6, the least denominator exponent $T$7 with respect to $T$8 controls the synthesis depth. If the $T$9-count $T=\mathrm{diag}(1,\zeta,\zeta^8).$0, then $T=\mathrm{diag}(1,\zeta,\zeta^8).$1; if $T=\mathrm{diag}(1,\zeta,\zeta^8).$2, then the operator is Clifford. Exact synthesis proceeds by finding the unique syllable among $T=\mathrm{diag}(1,\zeta,\zeta^8).$3 whose removal decreases $T=\mathrm{diag}(1,\zeta,\zeta^8).$4 by one, iterating until the remainder lies in $T=\mathrm{diag}(1,\zeta,\zeta^8).$5. This gives a polynomial-time exact synthesis algorithm and another route to uniqueness and T-count optimality (Prakash et al., 2018).

A later arithmetic study enlarges the diagonal resource from $T=\mathrm{diag}(1,\zeta,\zeta^8).$6 to

$T=\mathrm{diag}(1,\zeta,\zeta^8).$7

In that setting, sde $T=\mathrm{diag}(1,\zeta,\zeta^8).$8 gates are exactly Pauli+$T=\mathrm{diag}(1,\zeta,\zeta^8).$9, sde $27$0 gates are exactly $27$1, and an $27$2 step reduces smallest denominator exponent by one iff an explicitly computed obstruction $27$3 satisfies $27$4. This does not replace the canonical single-qutrit Clifford+T theory, but it extends the same synthesis language to a larger cyclotomic family (Kalra et al., 2023).

4. Controlled gates and ternary reversible computation

The multi-qutrit controlled-gate theory is developed explicitly by Yeh and van de Wetering. They prove that any $27$5-controlled qutrit Clifford+T unitary can be constructed exactly, unitarily, and ancilla-free with T-count $27$6. The construction combines singly controlled Clifford primitives, a generalized $27$7–$27$8 trick, the identities $27$9 and $9$0, and the qutrit square-root-style identities $9$1 and $9$2. The resulting toolkit covers multiple-controlled $9$3, $9$4, $9$5, $9$6, and Toffoli-like gates (Yeh et al., 2022).

For the controlled-$9$7 subproblem, one borrowed ancilla improves the asymptotic count to $9$8, whereas the ancilla-free controlled-$9$9 and controlled-$648$0 constructions remain $648$1. The same paper applies these results to classical reversible ternary computation: any reversible function on $648$2 trits can be implemented ancilla-free as a qutrit unitary using $648$3 T gates, and some such functions require at least $648$4 Clifford+T gates. The authors also stress a contrast with the qubit setting, where ancilla-free controlled-$648$5 and ancilla-free three-controlled Toffoli constructions are impossible in Clifford+T (Yeh et al., 2022).

A separate line of work uses $648$6 as the non-Clifford primitive. For qutrits, $648$7 is treated as a T-like gate, and the paper derives a $648$8-controlled $648$9 with qutrit T-count $\zeta=e^{2\pi i/9}$00. On that basis it gives deterministic constructions of qubit $\zeta=e^{2\pi i/9}$01 states of sizes three, six, and powers of three inside the qutrit $\zeta=e^{2\pi i/9}$02 subspace (Yeh, 2023).

5. Multi-qutrit exact synthesis and arithmetic properties

Multi-qutrit exact synthesis sharpens the algebraic class of Clifford+T-representable unitaries. One theorem states that every

$\zeta=e^{2\pi i/9}$03

has an exact Clifford+T realization using at most one ancilla. A stronger equivalence then identifies the full exact class: a $\zeta=e^{2\pi i/9}$04 unitary is exactly representable by Clifford+T with finitely many ancillas iff its entries lie in $\zeta=e^{2\pi i/9}$05, and at most two ancillas suffice. The synthesis factors the target into $\zeta=e^{2\pi i/9}$06-level unitaries, converts each $\zeta=e^{2\pi i/9}$07-level block into multiply-controlled gates by straightening triples on the $\zeta=e^{2\pi i/9}$08-ary grid $\zeta=e^{2\pi i/9}$09, and handles the $\zeta=e^{2\pi i/9}$10-extension by catalytic embedding (Kalra et al., 2024).

A more general framework replaces $\zeta=e^{2\pi i/9}$11 by

$\zeta=e^{2\pi i/9}$12

and studies the gate sets

$\zeta=e^{2\pi i/9}$13

For this Clifford-cyclotomic hierarchy, a $\zeta=e^{2\pi i/9}$14 unitary is exactly representable over $\zeta=e^{2\pi i/9}$15 iff its entries lie in $\zeta=e^{2\pi i/9}$16, and $\zeta=e^{2\pi i/9}$17 fresh ancillae suffice for synthesis. For $\zeta=e^{2\pi i/9}$18, the same work proves that $\zeta=e^{2\pi i/9}$19 is equivalent, up to at most one borrowed ancilla, to qutrit Clifford+$\zeta=e^{2\pi i/9}$20 (Glaudell et al., 2024).

The number-theoretic status of qutrit Clifford+T differs sharply from the qubit case. In $\zeta=e^{2\pi i/9}$21, the group generated by qutrit Clifford+T is not arithmetic but a thin matrix group: it is Zariski-dense and has infinite index in its ambient S-arithmetic group. The extension Clifford+$\zeta=e^{2\pi i/9}$22, by contrast, generates the full S-arithmetic subgroup. For words in $\zeta=e^{2\pi i/9}$23, the same paper proves a Solovay–Kitaev statement with optimal exponent $\zeta=e^{2\pi i/9}$24 and leading-rate loss $\zeta=e^{2\pi i/9}$25 (Evra et al., 2024).

6. Relation to metaplectic gates, hierarchy, benchmarking, and implementations

The relation between qutrit Clifford+T and qutrit metaplectic computation is asymmetric. On systems of at least two qutrits, Clifford+R is a strict subset of Clifford+T: $\zeta=e^{2\pi i/9}$26 has an exact Clifford+T realization with T-count $\zeta=e^{2\pi i/9}$27 using a borrowed ancilla, whereas $\zeta=e^{2\pi i/9}$28 cannot be exactly synthesized in Clifford+R, and there is no single-qutrit Clifford+T decomposition of $\zeta=e^{2\pi i/9}$29 (Glaudell et al., 2022).

From the hierarchy viewpoint, every third-level gate on at most two qutrits is semi-Clifford. Consequently, every one- or two-qutrit $\zeta=e^{2\pi i/9}$30 gate admits a Clifford–diagonal–Clifford decomposition, and efficient gate teleportation can use the magic state $\zeta=e^{2\pi i/9}$31 instead of the $\zeta=e^{2\pi i/9}$32-qudit Gottesman–Chuang resource state. For small-locality non-Clifford qutrit gates, this identifies a structurally privileged subset of the hierarchy (Chen et al., 2023).

For characterization rather than synthesis, a benchmarking protocol based on the hyperdihedral group

$\zeta=e^{2\pi i/9}$33

targets the qutrit $\zeta=e^{2\pi i/9}$34 gate. Its representation range is $\zeta=e^{2\pi i/9}$35, $\zeta=e^{2\pi i/9}$36, and only $\zeta=e^{2\pi i/9}$37 distinct gates are needed modulo global phase. The resulting survival probabilities obey

$\zeta=e^{2\pi i/9}$38

providing a randomized-benchmarking route to complete characterization of a universal qutrit gate set when combined with qutrit Clifford benchmarking (Amaro-Alcalá et al., 2023).

A recent implementation-oriented development represents qutrit Clifford+T gates with spin-$\zeta=e^{2\pi i/9}$39 angular momentum operators. In that model,

$\zeta=e^{2\pi i/9}$40

single-qutrit Cliffords are synthesized from rotations and one-axis twisting, and controlled gates use only linear two-body couplings such as

$\zeta=e^{2\pi i/9}$41

Via the Jordan–Schwinger map, the same constructions extend to two bosonic modes per qutrit using beam-splitter rotations, self-Kerr nonlinearities, and cross-Kerr interactions (Steinhoff, 24 Apr 2026).

Open directions recur across this literature. They include extension of canonical-form methods to higher prime dimensions, systematic optimization of T-depth beyond T-count minimization, ancilla reduction in multi-qutrit exact synthesis, and lower bounds for controlled-gate constructions (Glaudell et al., 2018, Glaudell et al., 2024, Yeh et al., 2022).