Clifford-Cyclotomic Circuits & Exact Synthesis

• Clifford-cyclotomic circuits are quantum circuits that combine Clifford operations with cyclotomic phase gates to enable exact, resource-optimal synthesis.
• They use canonical forms, ring characterizations, and efficient algorithms to decompose unitaries with minimal non-Clifford rotations.
• Ancilla-assisted and catalytic embeddings extend the synthesis techniques to multiqubit and multiqutrit systems, overcoming arithmetic obstructions.

Clifford-cyclotomic circuits are quantum circuits generated by the Clifford group together with a cyclotomic phase gate, typically a finite-order $z$-rotation in the qubit setting or its odd-prime analogue in the qutrit setting. The subject is organized around exact synthesis: given a unitary that is exactly representable by such a gate set, one seeks a decomposition that is both constructive and resource-optimal, with resources measured either by the number of non-Clifford cyclotomic rotations or by ancilla count. The literature develops this program at several levels: canonical forms and optimal ancilla-free synthesis for single-qubit families, ancilla-assisted ring-theoretic characterizations for multiqubit families, and a qutrit analogue over $\mathbb{Z}[1/3,\omega_k]$ (Forest et al., 2015, Amy et al., 2023, Glaudell et al., 2024, Dinh et al., 20 Aug 2025).

1. Definitions and gate-set families

For each positive even integer $n$, the single-qubit Clifford-cyclotomic gate set is

$$\mathcal{G}_n=\langle \mathcal{C}, U_z(\pi/n)\rangle, \qquad U_z(\theta)= \begin{pmatrix} 1 & 0\ 0 & e^{i\theta} \end{pmatrix},$$

where $\mathcal{C}$ is the single-qubit Clifford group. Since $n$ is even, one has $S=U_z(\pi/2)=(U_z(\pi/n))^{n/2}$, so $\mathcal{G}_n=\langle H_0, U_z(\pi/n)\rangle$. The special case $n=4$ is exactly Clifford+$T$, because $\mathbb{Z}[1/3,\omega_k]$0 (Forest et al., 2015).

In the multiqubit setting, the same terminology refers to extending Clifford generators by the phase gate

$\mathbb{Z}[1/3,\omega_k]$1

One formulation writes the gate set as $\mathbb{Z}[1/3,\omega_k]$2, with $\mathbb{Z}[1/3,\omega_k]$3 the smallest subring of $\mathbb{Z}[1/3,\omega_k]$4 containing $\mathbb{Z}[1/3,\omega_k]$5 and $\mathbb{Z}[1/3,\omega_k]$6. For $\mathbb{Z}[1/3,\omega_k]$7 divisible by $\mathbb{Z}[1/3,\omega_k]$8, the entries of the generators lie in $\mathbb{Z}[1/3,\omega_k]$9; more generally, the multiqubit program asks when every unitary with entries in the relevant ring has an exact $n$0-circuit (Amy et al., 2023, Dinh et al., 20 Aug 2025).

A qutrit analogue is defined for degree $n$1 by

$n$2

where $n$3. The hierarchy satisfies $n$4. The paper states that $n$5 is equivalent to the qutrit Toffoli+Hadamard gate set, while for $n$6, $n$7 is equivalent, up to one borrowed ancilla, to qutrit Clifford+$n$8 (Glaudell et al., 2024).

Across these variants, the cyclotomic gate is the ingredient that introduces roots of unity into the synthesis problem and links circuit representability to arithmetic structure in localization rings such as $n$9 and $$\mathcal{G}_n=\langle \mathcal{C}, U_z(\pi/n)\rangle, \qquad U_z(\theta)= \begin{pmatrix} 1 & 0\ 0 & e^{i\theta} \end{pmatrix},$$0.

2. Single-qubit exact synthesis and canonical form

The single-qubit theory generalizes the optimal exact synthesis program known for Clifford+$$\mathcal{G}_n=\langle \mathcal{C}, U_z(\pi/n)\rangle, \qquad U_z(\theta)= \begin{pmatrix} 1 & 0\ 0 & e^{i\theta} \end{pmatrix},$$1. A decomposition of a unitary $$\mathcal{G}_n=\langle \mathcal{C}, U_z(\pi/n)\rangle, \qquad U_z(\theta)= \begin{pmatrix} 1 & 0\ 0 & e^{i\theta} \end{pmatrix},$$2 is written as

$$\mathcal{G}_n=\langle \mathcal{C}, U_z(\pi/n)\rangle, \qquad U_z(\theta)= \begin{pmatrix} 1 & 0\ 0 & e^{i\theta} \end{pmatrix},$$3

with $$\mathcal{G}_n=\langle \mathcal{C}, U_z(\pi/n)\rangle, \qquad U_z(\theta)= \begin{pmatrix} 1 & 0\ 0 & e^{i\theta} \end{pmatrix},$$4, and its cost is the total number of non-Clifford rotations $$\mathcal{G}_n=\langle \mathcal{C}, U_z(\pi/n)\rangle, \qquad U_z(\theta)= \begin{pmatrix} 1 & 0\ 0 & e^{i\theta} \end{pmatrix},$$5. The quantity $$\mathcal{G}_n=\langle \mathcal{C}, U_z(\pi/n)\rangle, \qquad U_z(\theta)= \begin{pmatrix} 1 & 0\ 0 & e^{i\theta} \end{pmatrix},$$6 is the minimum such cost, allowing an arbitrary global phase. The exact synthesis algorithm is optimal in the strong sense that it returns a decomposition achieving $$\mathcal{G}_n=\langle \mathcal{C}, U_z(\pi/n)\rangle, \qquad U_z(\theta)= \begin{pmatrix} 1 & 0\ 0 & e^{i\theta} \end{pmatrix},$$7 (Forest et al., 2015).

The structural basis of the algorithm is a canonical form. Every $$\mathcal{G}_n=\langle \mathcal{C}, U_z(\pi/n)\rangle, \qquad U_z(\theta)= \begin{pmatrix} 1 & 0\ 0 & e^{i\theta} \end{pmatrix},$$8 can be written as

$$\mathcal{G}_n=\langle \mathcal{C}, U_z(\pi/n)\rangle, \qquad U_z(\theta)= \begin{pmatrix} 1 & 0\ 0 & e^{i\theta} \end{pmatrix},$$9

where $\mathcal{C}$0, adjacent axes satisfy $\mathcal{C}$1, each $\mathcal{C}$2 obeys $\mathcal{C}$3, and $\mathcal{C}$4 is a Clifford up to global phase. Here

$\mathcal{C}$5

for $\mathcal{C}$6. The relations

$\mathcal{C}$7

allow sign normalization and rotation-count reduction. From this canonical form,

$\mathcal{C}$8

so the canonical form directly encodes the minimum number of $\mathcal{C}$9 rotations needed (Forest et al., 2015).

The reconstruction algorithm proceeds through the Bloch-sphere image $n$0, where the generators become rotations $n$1 by $n$2 around coordinate axes. The product structure mirrors the canonical form: $n$3 with $n$4 a signed permutation matrix. The denominator-exponent theorem then identifies the first factor from the arithmetic of matrix entries. Writing $n$5 with $n$6 odd, the paper defines

$n$7

and proves that each nonzero entry of $n$8 can be written as $n$9 with $S=U_z(\pi/2)=(U_z(\pi/n))^{n/2}$0 an algebraic integer not divisible by $S=U_z(\pi/2)=(U_z(\pi/n))^{n/2}$1. If the canonical form has parameters $S=U_z(\pi/2)=(U_z(\pi/n))^{n/2}$2, the maximum denominator exponent is

$S=U_z(\pi/2)=(U_z(\pi/n))^{n/2}$3

exactly two rows attain exponent $S=U_z(\pi/2)=(U_z(\pi/n))^{n/2}$4, and the remaining row has maximal exponent $S=U_z(\pi/2)=(U_z(\pi/n))^{n/2}$5, which reveals the first axis $S=U_z(\pi/2)=(U_z(\pi/n))^{n/2}$6. The algorithm repeatedly searches over $S=U_z(\pi/2)=(U_z(\pi/n))^{n/2}$7 and $S=U_z(\pi/2)=(U_z(\pi/n))^{n/2}$8, chooses the pair minimizing the relevant denominator-exponent statistic after left-multiplying by $S=U_z(\pi/2)=(U_z(\pi/n))^{n/2}$9, and peels off one factor at a time. Divisibility by $\mathcal{G}_n=\langle H_0, U_z(\pi/n)\rangle$0 is testable in polynomial time using basis representations of algebraic integers, so the procedure is efficient in bit complexity (Forest et al., 2015).

3. Ring characterizations and arithmetic obstructions

The ring-theoretic formulation asks whether exact circuit representability coincides with entrywise membership in a cyclotomic localization ring. In the single-qubit qubit setting, the relevant ring is

$\mathcal{G}_n=\langle H_0, U_z(\pi/n)\rangle$1

For $\mathcal{G}_n=\langle H_0, U_z(\pi/n)\rangle$2, the known Clifford+$\mathcal{G}_n=\langle H_0, U_z(\pi/n)\rangle$3 characterization is $\mathcal{G}_n=\langle H_0, U_z(\pi/n)\rangle$4. The same equality is proved for $\mathcal{G}_n=\langle H_0, U_z(\pi/n)\rangle$5, with $\mathcal{G}_n=\langle H_0, U_z(\pi/n)\rangle$6 and $\mathcal{G}_n=\langle H_0, U_z(\pi/n)\rangle$7 identified as new cases. More generally, however, the equality fails for almost all even $\mathcal{G}_n=\langle H_0, U_z(\pi/n)\rangle$8: the fraction of even integers $\mathcal{G}_n=\langle H_0, U_z(\pi/n)\rangle$9 for which $n=4$0 tends to zero as $n=4$1 (Forest et al., 2015).

A particularly sharp intermediate result concerns the $n=4$2-axis subgroup. The subgroup of $n=4$3-rotations inside $n=4$4 is exactly

$n=4$5

If

$n=4$6

then $n=4$7 if and only if, writing $n=4$8 with $n=4$9 odd, there exists a positive integer $T$0 such that

$T$1

Equivalently, the only unit-modulus elements of $T$2 are the roots of unity in $T$3,

$T$4

exactly when that congruence holds; otherwise $T$5 contains an element of infinite order. This is the arithmetic obstruction behind the generic failure of $T$6 (Forest et al., 2015).

A common misconception is that the familiar Clifford+$T$7 ring characterization extends uniformly across the entire single-qubit cyclotomic family. The single-qubit theory shows the opposite: the ring characterization is exceptional rather than generic. By contrast, in the multiqubit setting the limitations that apply to single-qubit unitaries, for which the correspondence between Clifford-cyclotomic operators and matrices over $T$8 fails for all but finitely many values of $T$9, can be overcome through the use of ancillas (Amy et al., 2023).

4. Multiqubit exact synthesis and ancilla-assisted catalytic embeddings

For multiqubit qubit circuits of degree $\mathbb{Z}[1/3,\omega_k]$00, the central theorem states that a $\mathbb{Z}[1/3,\omega_k]$01 unitary $\mathbb{Z}[1/3,\omega_k]$02 can be exactly represented by an $\mathbb{Z}[1/3,\omega_k]$03-qubit circuit over $\mathbb{Z}[1/3,\omega_k]$04 if and only if all entries of $\mathbb{Z}[1/3,\omega_k]$05 lie in the ring $\mathbb{Z}[1/3,\omega_k]$06. The same line of work gives explicit ancilla bounds: one ancilla suffices when $\mathbb{Z}[1/3,\omega_k]$07, and $\mathbb{Z}[1/3,\omega_k]$08 ancillas suffice when $\mathbb{Z}[1/3,\omega_k]$09 (Amy et al., 2023).

The constructive mechanism is a catalytic embedding that lowers the cyclotomic degree. In the later refinement for $\mathbb{Z}[1/3,\omega_k]$10, if $\mathbb{Z}[1/3,\omega_k]$11 with $\mathbb{Z}[1/3,\omega_k]$12, then

$\mathbb{Z}[1/3,\omega_k]$13

where

$\mathbb{Z}[1/3,\omega_k]$14

This $\mathbb{Z}[1/3,\omega_k]$15 is a 2-dimensional catalytic embedding: $\mathbb{Z}[1/3,\omega_k]$16. A determinant compatibility lemma shows that $\mathbb{Z}[1/3,\omega_k]$17 respects determinants in the number-theoretic sense of relative norms, which is used to reduce synthesis to the determinant-one case (Dinh et al., 20 Aug 2025).

The ancilla complexity was subsequently improved. For $\mathbb{Z}[1/3,\omega_k]$18, every $\mathbb{Z}[1/3,\omega_k]$19 unitary with entries in $\mathbb{Z}[1/3,\omega_k]$20 can be exactly represented over $\mathbb{Z}[1/3,\omega_k]$21 using only one ancilla, and this is minimal in the worst case. By induction, for every $\mathbb{Z}[1/3,\omega_k]$22, every $\mathbb{Z}[1/3,\omega_k]$23 unitary in $\mathbb{Z}[1/3,\omega_k]$24 is exactly representable over $\mathbb{Z}[1/3,\omega_k]$25 using $\mathbb{Z}[1/3,\omega_k]$26 ancillas, improving the earlier $\mathbb{Z}[1/3,\omega_k]$27 bound; for $\mathbb{Z}[1/3,\omega_k]$28 this bound is minimal (Dinh et al., 20 Aug 2025).

The same paper extends exact synthesis beyond powers of two. For the base case $\mathbb{Z}[1/3,\omega_k]$29,

$\mathbb{Z}[1/3,\omega_k]$30

with 2 ancillas sufficient. By induction, for all $\mathbb{Z}[1/3,\omega_k]$31,

$\mathbb{Z}[1/3,\omega_k]$32

with $\mathbb{Z}[1/3,\omega_k]$33 ancillas sufficient. The $\mathbb{Z}[1/3,\omega_k]$34-case relies on a constructive decomposition theorem for unitaries over $\mathbb{Z}[1/3,\omega_k]$35, based on the least denominator exponent, residue analysis modulo $\mathbb{Z}[1/3,\omega_k]$36, and the operator

$\mathbb{Z}[1/3,\omega_k]$37

followed by a lift from $\mathbb{Z}[1/3,\omega_k]$38 to $\mathbb{Z}[1/3,\omega_k]$39 via a catalytic embedding (Dinh et al., 20 Aug 2025).

5. Multiqutrit Clifford-cyclotomic circuits

The qutrit theory establishes an analogous exact-synthesis correspondence for odd prime local dimension. For fixed $\mathbb{Z}[1/3,\omega_k]$40, the relevant ring is

$\mathbb{Z}[1/3,\omega_k]$41

denoted $\mathbb{Z}[1/3,\omega_k]$42, and the main theorem states that a $\mathbb{Z}[1/3,\omega_k]$43 unitary matrix $\mathbb{Z}[1/3,\omega_k]$44 can be represented by an $\mathbb{Z}[1/3,\omega_k]$45-qutrit circuit over $\mathbb{Z}[1/3,\omega_k]$46 if and only if $\mathbb{Z}[1/3,\omega_k]$47, where $\mathbb{Z}[1/3,\omega_k]$48 is the group of unitary matrices with entries in $\mathbb{Z}[1/3,\omega_k]$49. Moreover, $\mathbb{Z}[1/3,\omega_k]$50 ancillae are always sufficient to construct a circuit for $\mathbb{Z}[1/3,\omega_k]$51 (Glaudell et al., 2024).

The proof is constructive and splits into a base case and an induction step. For $\mathbb{Z}[1/3,\omega_k]$52, exact synthesis for $\mathbb{Z}[1/3,\omega_k]$53 is obtained by showing that $\mathbb{Z}[1/3,\omega_k]$54 is generated by a finite set of level matrices

$\mathbb{Z}[1/3,\omega_k]$55

The argument uses the least $\mathbb{Z}[1/3,\omega_k]$56-denominator exponent, where $\mathbb{Z}[1/3,\omega_k]$57. A vector with least denominator exponent $\mathbb{Z}[1/3,\omega_k]$58 must be a basis vector up to a phase, while a positive denominator exponent can be reduced by suitable generators. A crucial “Hadamard reduction” lemma says that for any triple of non-$\mathbb{Z}[1/3,\omega_k]$59-divisible entries, a suitable combination of local phases and $\mathbb{Z}[1/3,\omega_k]$60 sends them to entries all divisible by $\mathbb{Z}[1/3,\omega_k]$61. The appendix then gives explicit circuit decompositions showing that each generator in $\mathbb{Z}[1/3,\omega_k]$62 is realizable over $\mathbb{Z}[1/3,\omega_k]$63, with at most 2 borrowed ancillae (Glaudell et al., 2024).

For $\mathbb{Z}[1/3,\omega_k]$64, the proof uses catalytic embeddings. Every element of $\mathbb{Z}[1/3,\omega_k]$65 can be uniquely written as

$\mathbb{Z}[1/3,\omega_k]$66

and similarly every matrix over $\mathbb{Z}[1/3,\omega_k]$67 decomposes as $\mathbb{Z}[1/3,\omega_k]$68. The embedding is

$\mathbb{Z}[1/3,\omega_k]$69

where

$\mathbb{Z}[1/3,\omega_k]$70

and $\mathbb{Z}[1/3,\omega_k]$71. This yields a 3-dimensional catalytic embedding $\mathbb{Z}[1/3,\omega_k]$72; composing the embeddings produces a $\mathbb{Z}[1/3,\omega_k]$73-dimensional catalytic embedding into the base-ring case. The synthesis strategy is then to embed a $\mathbb{Z}[1/3,\omega_k]$74-valued unitary into a larger $\mathbb{Z}[1/3,\omega_k]$75-valued unitary, synthesize the larger unitary using the $\mathbb{Z}[1/3,\omega_k]$76 algorithm, and uncompute the catalyst registers using explicit preparations based on $\mathbb{Z}[1/3,\omega_k]$77 (Glaudell et al., 2024).

6. Structural, rewrite-theoretic, and complexity-theoretic context

Clifford-cyclotomic circuit theory sits on top of structural results for the Clifford group itself. A foundational result gives a complete normal form theory for Clifford circuits on $\mathbb{Z}[1/3,\omega_k]$78 qubits: every Clifford operator has a unique normal form, and a finite terminating rewrite system reduces any Clifford circuit to that form. The paper presents the Clifford groupoid by generators $\mathbb{Z}[1/3,\omega_k]$79 and relations, and shows that no axioms involving 4 or more qubits are needed. This rewrite-theoretic template is directly relevant to later canonical-form and exact-synthesis work for Clifford-cyclotomic extensions (Selinger, 2013).

A different line of work studies the computational power of restricted Clifford-derived models. In conjugated Clifford circuits, every Clifford gate is conjugated by the same one-qubit unitary $\mathbb{Z}[1/3,\omega_k]$80. The classification theorem states that $\mathbb{Z}[1/3,\omega_k]$81-CCCs are efficiently weakly simulable if and only if

$\mathbb{Z}[1/3,\omega_k]$82

for some single-qubit Clifford $\mathbb{Z}[1/3,\omega_k]$83 and some angle $\mathbb{Z}[1/3,\omega_k]$84; otherwise, assuming $\mathbb{Z}[1/3,\omega_k]$85 is infinite, the family is $\mathbb{Z}[1/3,\omega_k]$86-supreme. The paper explicitly notes that many familiar Clifford-plus-$\mathbb{Z}[1/3,\omega_k]$87 or Clifford-cyclotomic constructions fit inside this viewpoint because they are generated by Clifford operations plus a special single-qubit phase gate, often a cyclotomic rotation (Bouland et al., 2017).

There is also a topological classification of loops of Clifford circuits on translation-invariant lattices of prime-$\mathbb{Z}[1/3,\omega_k]$88 qudits. A loop is a periodic one-parameter family of Clifford dynamics, formalized as a path of Lagrangian submodules over a polynomial extension. Using Hermitian $\mathbb{Z}[1/3,\omega_k]$89-theory and a generalized Maslov index, the homotopy classes satisfy

$\mathbb{Z}[1/3,\omega_k]$90

while

$\mathbb{Z}[1/3,\omega_k]$91

Although not focused on cyclotomic compilation, this work is relevant to a broader Clifford-cyclotomic discussion because roots of unity enter through the Pauli commutation relation $\mathbb{Z}[1/3,\omega_k]$92, and the classification is controlled by algebraic invariants over Laurent polynomial rings (Geiko et al., 2023).

Taken together, these results show that Clifford-cyclotomic circuits are not a single theorem but a research program linking exact synthesis, canonical forms, rewrite systems, arithmetic of cyclotomic localization rings, ancilla-assisted embeddings, and, in adjacent directions, complexity classification and algebraic topology. The most persistent theme is the tension between entrywise ring characterizations and circuit realizability: in the single-qubit case the correspondence is exceptional, whereas in multiqubit and multiqutrit settings ancillas and catalytic constructions recover exact representability over broad infinite families (Forest et al., 2015, Amy et al., 2023, Glaudell et al., 2024, Dinh et al., 20 Aug 2025).