Magic Unitaries in Quantum Computation

• Magic unitaries are defined as non-Clifford operations that, when used with stabilizer protocols and magic states, enable universal quantum computation.
• They are quantified using measures like robustness of magic and second-order stabilizer Rényi entropy, which frame their resource requirements in fault-tolerant architectures.
• Studies reveal that non-Clifford magic unitaries require consumable (one-copy-per-use) resource states, setting strict computational complexity constraints.

A magic unitary is a Clifford-orbit non-stabilizer unitary—i.e., a unitary transformation not contained in the Clifford group, which, when combined with Clifford operations and stabilizer ancillas, enables universal quantum computation. Physically, magic unitaries are associated with the injection of “magic states,” resource states that do not admit efficient classical simulation by stabilizer methods. The synthesis and reuse properties of magic unitaries, their quantification via resource theories such as robustness of magic and stabilizer Rényi entropy, and the scaling laws for their generation in random circuits, are central to the theory of quantum resources and fault-tolerant quantum computation (Anderson, 2012, Howard et al., 2016, Szombathy et al., 2024). The boundary between Clifford (classically simulable) and non-Clifford (resource) domains imposes strict no-go results for the reusable application of non-Clifford unitaries.

1. Formalism and Definition

Let $\mathcal{H}_S$ denote the “data” register (possibly many logical qubits) and $\mathcal{H}_M$ an ancillary magic register. A pure state $\left|M_U\right\rangle_M \in \mathcal{H}_M$ is a reusable magic state for $U \in \mathrm{U}(\mathcal{H}_S)$ if there exists a Clifford operator $\mathcal{C}$ such that for all $\left|\psi\right\rangle_S$,

$$\mathcal{C} \big( \left|\psi\right\rangle_S \otimes \left|M_U\right\rangle_M \big) = (U \left|\psi\right\rangle_S) \otimes \left|M_U\right\rangle_M.$$

In density operator form,

$$\mathcal{C} [ \rho_S \otimes \left|M_U\right\rangle \!\left\langle M_U \right|_M ] \mathcal{C}^\dagger = (U \rho_S U^\dagger) \otimes \left|M_U\right\rangle\!\left\langle M_U\right|_M, \quad \forall \rho_S.$$

This definition implies that, post-application, the magic register is left unchanged and may be reused for subsequent applications.

For stabilizer characterizations, let $G = \{g_j\}$ stabilize $\left|M_U\right\rangle$ ($\mathcal{H}_M$0). Then,

$\mathcal{H}_M$1

$\mathcal{H}_M$2

Resource protocol: apply $\mathcal{H}_M$3 to $\mathcal{H}_M$4, optionally measure $\mathcal{H}_M$5 in a Pauli basis with Clifford feed-forward correction, and recover $\mathcal{H}_M$6 (Anderson, 2012).

2. Existence and Complexity Constraints

The existence of reusable magic states is sharply restricted by computational complexity:

• If $\mathcal{H}_M$7 is a Clifford operator, reusable gadgets are sometimes possible (e.g., for the $\mathcal{H}_M$8 gate, using $\mathcal{H}_M$9).
• For non-Clifford $\left|M_U\right\rangle_M \in \mathcal{H}_M$0, no fixed-size reusable magic state exists unless BQP = P. If such reusable gadgets existed for any non-Clifford unitary, all polynomial-size quantum circuits from $\left|M_U\right\rangle_M \in \mathcal{H}_M$1 could be simulated efficiently classically, contradicting the widely held belief that BQP $\left|M_U\right\rangle_M \in \mathcal{H}_M$2 P (Anderson, 2012).

Sketch of the argument:

1. Decompose a target circuit $\left|M_U\right\rangle_M \in \mathcal{H}_M$3 as …$\left|M_U\right\rangle_M \in \mathcal{H}_M$4–$\left|M_U\right\rangle_M \in \mathcal{H}_M$5–$\left|M_U\right\rangle_M \in \mathcal{H}_M$6–…–$\left|M_U\right\rangle_M \in \mathcal{H}_M$7–$\left|M_U\right\rangle_M \in \mathcal{H}_M$8, $\left|M_U\right\rangle_M \in \mathcal{H}_M$9 Clifford, $U \in \mathrm{U}(\mathcal{H}_S)$0.
2. Replace each $U \in \mathrm{U}(\mathcal{H}_S)$1 by a Clifford+ancilla gadget using a single, fixed magic state.
3. Combine all magic gadgets and rearrange computation into a convex combination of efficiently-simulable stabilizer circuits.
4. The magic register never entangles with the data, and the entire computation is a polynomial mixture over stabilizer evolutions. Thus, no polynomial-size reusable magic state exists for non-Clifford $U \in \mathrm{U}(\mathcal{H}_S)$2.

3. Magic Quantification and Resource Theory

The non-stabilizerness of unitaries and states underpinning magic unitaries is quantified by several monotones:

• Second-order stabilizer Rényi entropy $U \in \mathrm{U}(\mathcal{H}_S)$3: $U \in \mathrm{U}(\mathcal{H}_S)$4 where $U \in \mathrm{U}(\mathcal{H}_S)$5, $U \in \mathrm{U}(\mathcal{H}_S)$6. $U \in \mathrm{U}(\mathcal{H}_S)$7 iff $U \in \mathrm{U}(\mathcal{H}_S)$8 is a stabilizer state, and $U \in \mathrm{U}(\mathcal{H}_S)$9 otherwise (Szombathy et al., 2024).
• Non-stabilizing power (second-magic moment) $\mathcal{C}$0: $\mathcal{C}$1 This quantifies how much magic is generated on average when $\mathcal{C}$2 is applied to stabilizer inputs.
• Robustness of magic ($\mathcal{C}$3): $\mathcal{C}$4 Robustness is monotonic under stabilizer operations and submultiplicative under tensor product. For $\mathcal{C}$5 copies, the overhead for classical simulation scales as $\mathcal{C}$6 (Howard et al., 2016).

These measures provide primary tools for evaluating the synthesis cost and classical intractability associated with implementing target unitaries.

4. Magic Unitaries in Clifford+$\mathcal{C}$7 Circuits

In random Clifford+$\mathcal{C}$8 circuits (“$\mathcal{C}$9-doped Clifford circuits”), the injection of $\left|\psi\right\rangle_S$0-gates drives a transition from classically simulable (Clifford) circuits to truly quantum, random-circuit-like behavior. Key scaling results:

• In the dilute regime ($\left|\psi\right\rangle_S$1), each $\left|\psi\right\rangle_S$2 contributes on average $\left|\psi\right\rangle_S$3 bits of magic.
• Magic $\left|\psi\right\rangle_S$4 grows linearly: $\left|\psi\right\rangle_S$5, magic density $\left|\psi\right\rangle_S$6 for $\left|\psi\right\rangle_S$7.
• For $\left|\psi\right\rangle_S$8, circuits approach the Haar-random unitary magic density limit $\left|\psi\right\rangle_S$9, with a phase-like transition at $$\mathcal{C} \big( \left|\psi\right\rangle_S \otimes \left|M_U\right\rangle_M \big) = (U \left|\psi\right\rangle_S) \otimes \left|M_U\right\rangle_M.$$0.
• In contrast, the spectral transition (random-matrix-like statistics) occurs at merely $$\mathcal{C} \big( \left|\psi\right\rangle_S \otimes \left|M_U\right\rangle_M \big) = (U \left|\psi\right\rangle_S) \otimes \left|M_U\right\rangle_M.$$1 $$\mathcal{C} \big( \left|\psi\right\rangle_S \otimes \left|M_U\right\rangle_M \big) = (U \left|\psi\right\rangle_S) \otimes \left|M_U\right\rangle_M.$$2-gates, while saturating the magic density requires $$\mathcal{C} \big( \left|\psi\right\rangle_S \otimes \left|M_U\right\rangle_M \big) = (U \left|\psi\right\rangle_S) \otimes \left|M_U\right\rangle_M.$$3 $$\mathcal{C} \big( \left|\psi\right\rangle_S \otimes \left|M_U\right\rangle_M \big) = (U \left|\psi\right\rangle_S) \otimes \left|M_U\right\rangle_M.$$4-gates (Szombathy et al., 2024).

The following table summarizes the scaling behavior:

Regime	$$\mathcal{C} \big( \left|\psi\right\rangle_S \otimes \left|M_U\right\rangle_M \big) = (U \left|\psi\right\rangle_S) \otimes \left|M_U\right\rangle_M.$$5 (number of $$\mathcal{C} \big( \left|\psi\right\rangle_S \otimes \left|M_U\right\rangle_M \big) = (U \left|\psi\right\rangle_S) \otimes \left|M_U\right\rangle_M.$$6-gates)	Magic behavior
Dilute	$$\mathcal{C} \big( \left|\psi\right\rangle_S \otimes \left|M_U\right\rangle_M \big) = (U \left|\psi\right\rangle_S) \otimes \left|M_U\right\rangle_M.$$7	$$\mathcal{C} \big( \left|\psi\right\rangle_S \otimes \left|M_U\right\rangle_M \big) = (U \left|\psi\right\rangle_S) \otimes \left|M_U\right\rangle_M.$$8
Intermediate	$$\mathcal{C} \big( \left|\psi\right\rangle_S \otimes \left|M_U\right\rangle_M \big) = (U \left|\psi\right\rangle_S) \otimes \left|M_U\right\rangle_M.$$9	Quasi-continuous $$\mathcal{C} [ \rho_S \otimes \left|M_U\right\rangle \!\left\langle M_U \right|_M ] \mathcal{C}^\dagger = (U \rho_S U^\dagger) \otimes \left|M_U\right\rangle\!\left\langle M_U\right|_M, \quad \forall \rho_S.$$0
Dense	$$\mathcal{C} [ \rho_S \otimes \left|M_U\right\rangle \!\left\langle M_U \right|_M ] \mathcal{C}^\dagger = (U \rho_S U^\dagger) \otimes \left|M_U\right\rangle\!\left\langle M_U\right|_M, \quad \forall \rho_S.$$1	$$\mathcal{C} [ \rho_S \otimes \left|M_U\right\rangle \!\left\langle M_U \right|_M ] \mathcal{C}^\dagger = (U \rho_S U^\dagger) \otimes \left|M_U\right\rangle\!\left\langle M_U\right|_M, \quad \forall \rho_S.$$2 (Haar limit)

5. Gate Synthesis and Quantum Compiling

The cost of synthesizing target unitaries from Clifford+$$\mathcal{C} [ \rho_S \otimes \left|M_U\right\rangle \!\left\langle M_U \right|_M ] \mathcal{C}^\dagger = (U \rho_S U^\dagger) \otimes \left|M_U\right\rangle\!\left\langle M_U\right|_M, \quad \forall \rho_S.$$3 operations is tightly linked to the robustness of the corresponding resource states:

• For a non-Clifford gate $$\mathcal{C} [ \rho_S \otimes \left|M_U\right\rangle \!\left\langle M_U \right|_M ] \mathcal{C}^\dagger = (U \rho_S U^\dagger) \otimes \left|M_U\right\rangle\!\left\langle M_U\right|_M, \quad \forall \rho_S.$$4 (e.g., control-$$\mathcal{C} [ \rho_S \otimes \left|M_U\right\rangle \!\left\langle M_U \right|_M ] \mathcal{C}^\dagger = (U \rho_S U^\dagger) \otimes \left|M_U\right\rangle\!\left\langle M_U\right|_M, \quad \forall \rho_S.$$5 or $$\mathcal{C} [ \rho_S \otimes \left|M_U\right\rangle \!\left\langle M_U \right|_M ] \mathcal{C}^\dagger = (U \rho_S U^\dagger) \otimes \left|M_U\right\rangle\!\left\langle M_U\right|_M, \quad \forall \rho_S.$$6), prepare $$\mathcal{C} [ \rho_S \otimes \left|M_U\right\rangle \!\left\langle M_U \right|_M ] \mathcal{C}^\dagger = (U \rho_S U^\dagger) \otimes \left|M_U\right\rangle\!\left\langle M_U\right|_M, \quad \forall \rho_S.$$7 and use Clifford gadgets to teleport $$\mathcal{C} [ \rho_S \otimes \left|M_U\right\rangle \!\left\langle M_U \right|_M ] \mathcal{C}^\dagger = (U \rho_S U^\dagger) \otimes \left|M_U\right\rangle\!\left\langle M_U\right|_M, \quad \forall \rho_S.$$8 onto data.
• Minimum $$\mathcal{C} [ \rho_S \otimes \left|M_U\right\rangle \!\left\langle M_U \right|_M ] \mathcal{C}^\dagger = (U \rho_S U^\dagger) \otimes \left|M_U\right\rangle\!\left\langle M_U\right|_M, \quad \forall \rho_S.$$9-state cost $G = \{g_j\}$0 satisfies

$G = \{g_j\}$1

where $G = \{g_j\}$2 is the single-qubit $G = \{g_j\}$3-magic state.

• Explicit: $G = \{g_j\}$4 requires 3 $G = \{g_j\}$5 gates, $G = \{g_j\}$6 requires 4 $G = \{g_j\}$7 gates (Howard et al., 2016).

Table: Example resource costs

Gate	$G = \{g_j\}$8	$G = \{g_j\}$9-gate lower bound
$\left|M_U\right\rangle$0	$\left|M_U\right\rangle$1	3
$\left|M_U\right\rangle$2	$\left|M_U\right\rangle$3	4

This resource-theoretic bound directly informs quantum compilers and magic state factory architectures.

6. Practical Constraints and Circuit Protocols

The lack of reusable gadgets for non-Clifford unitaries demands one-copy-per-use (“consumption”) protocols for non-Clifford magic states. Reusable gadgets are only possible for Clifford-level unitaries (e.g., $\left|M_U\right\rangle$4), and even then, only in some code architectures. Standard circuits for small gates:

• S gate (Clifford): 1-qubit reusable state $\left|M_U\right\rangle$5; Clifford circuit, no feed-forward, magic state returned intact.
• T gate (non-Clifford): Teleportation gadget consumes $\left|M_U\right\rangle$6, no reusable implementation.
• CCZ (non-Clifford): Three-qubit resource $\left|M_U\right\rangle$7 used in a single-shot injection; consumed by projective measurement (Anderson, 2012).

7. Implications for Quantum Resource Optimization

These results demarcate the fundamental boundary between Clifford and non-Clifford resource requirements and set hard lower bounds for fault-tolerant quantum architectures. Synthesizing generic high-magic unitaries for universal computation inevitably requires $\left|M_U\right\rangle$8 non-Clifford resources for $\left|M_U\right\rangle$9-qubit data, as circuits with sublinear $\mathcal{H}_M$00-count remain classically tractable. This scaling law underpins the practical design of distillation factories, architectural choices, and classical simulation strategies (Anderson, 2012, Howard et al., 2016, Szombathy et al., 2024).