Wave Function Cloning Algorithms

• Wave Function Cloning Algorithms are specialized quantum procedures that achieve approximate state duplication by restricting the input ensemble in accordance with the no-cloning theorem.
• They encompass varied methods including variational, mirror, phase-covariant optical, and channel cloning approaches to optimize performance under defined fidelity benchmarks.
• Recent implementations leverage hybrid quantum-classical optimization and hardware-adapted circuits to enhance efficiency and address computational complexity in practical settings.

“Wave function cloning algorithm” is not a single standardized algorithmic primitive in quantum information theory. In the arXiv literature, the term refers to several distinct but related constructions: hybrid variational procedures that learn an approximate cloner for a restricted ensemble of states, symmetry-assisted exact cloning on a specially structured two-state family, physical phase-covariant optical cloners and the numerical methods used to model them, and broader channel-cloning frameworks in which state cloning appears as a special case. Across these usages, the common feature is that the objective is never universal exact copying of an arbitrary unknown quantum state; rather, each construction either respects the no-cloning theorem by restricting the input family or by allowing only approximate cloning, or else recasts cloning as a computational or higher-order process problem (Hoch et al., 2024).

1. Scope of the term in the literature

The literature uses “wave function cloning” for heterogeneous tasks that share only a family resemblance. In one line of work, a parameterized quantum circuit or programmable interferometer is trained to maximize cloning fidelity for a chosen state ensemble. In another, a structured memory state is “perfectly cloned up to global mirroring,” which is operationally sufficient only because the downstream associative-retrieval task identifies mirrored and original memories up to a final inversion. In yet another, phase-covariant optical amplification is treated as a physical cloning transformation for equatorial polarization qubits, while the algorithmic novelty lies in numerically evaluating the resulting multiphoton states. A more recent abstraction embeds state cloning into deterministic cloning of quantum channels via trash-and-replace channels, whereas complexity-theoretic work addresses the opposite question: whether cloning may remain computationally intractable even when a verifier description is known (Coyle et al., 2020).

Usage	Object cloned	Characterization
Variational quantum cloning	Restricted state family	Hybrid quantum-classical approximate cloner
Mirror modular cloning	$|M\rangle$ or $|\bar M\rangle$	Exact only on a two-state, symmetry-related family
Phase-covariant optical cloning	Equatorial polarization qubits	Physical restricted cloner, not universal copying
Channel cloning reformulation	Trash-and-replace channels $\mathcal T_\rho$	State cloning as a special case of channel cloning
Witness-cloning hardness	Accepted states of a verifier	Conditional impossibility framework, not a constructive cloner

This multiplicity of meanings matters because apparently similar claims can refer to very different operational tasks. “Optimal,” “perfect,” and even “cloning” are task-relative terms in this area: optimality is usually fidelity-optimality for a specified ensemble; perfection may hold only up to a symmetry; and a cloning algorithm may mean a trainable search procedure, a fixed physical transformation, or an optimization framework rather than a closed-form circuit.

2. Information-theoretic constraints and fidelity benchmarks

The foundational constraint is the no-cloning theorem. In the formulation recalled in experimental variational cloning, the forbidden ideal process would require a unitary satisfying

$$U\ket{\psi}_1 \ket{0}_2\ket{\mathcal{M}}_\mathcal{M} \rightarrow \ket{\psi}_1\ket{\psi}_2\ket{\mathcal{M}'}_\mathcal{M} \quad \forall \ket{\psi},$$

which cannot exist for all unknown $\ket{\psi}$. A generic qubit is written as

$$\ket{\psi} = \cos(\theta)\ket{0} + \sin(\theta)e^{i\phi}\ket{1}.$$

Accordingly, legitimate cloning problems replace universal exact copying by approximate deterministic cloning or by exact cloning on a restricted state set (Hoch et al., 2024).

The canonical universal symmetric $1\rightarrow 2$ qubit benchmark is the Bužek–Hillery cloner, for which the reduced clone states satisfy

$\rho_1=\rho_2=\frac{5}{6}\dyad{\psi}+\frac{1}{6}\dyad{\psi_\bot},$

with fidelity

$F_{1,2}= \Tr\!\left(\sqrt{\sqrt{\sigma}\rho_{1,2}\sqrt{\sigma}}\right)^2 = \frac{5}{6}, \qquad \sigma=\dyad{\psi}.$

The value $5/6$ is the optimal universal symmetric $|\bar M\rangle$0 qubit fidelity benchmark. Restricted ensembles admit higher fidelities. For symmetric phase-covariant $|\bar M\rangle$1 cloning on the equator of the Bloch sphere,

$|\bar M\rangle$2

the quoted optimal fidelity is

$|\bar M\rangle$3

to be compared with the semiclassical measure-and-prepare benchmark

$|\bar M\rangle$4

For state-dependent cloning of only two known nonorthogonal states, the achievable fidelity can exceed the phase-covariant value because the ensemble is smaller (Hoch et al., 2024).

Variational quantum cloning work formalizes several figures of merit. For an $|\bar M\rangle$5 task with output state $|\bar M\rangle$6, the reduced state of clone $|\bar M\rangle$7 is $|\bar M\rangle$8. Local clone fidelity is

$|\bar M\rangle$9

while global fidelity is

$\mathcal T_\rho$0

The paper distinguishes local, squared-local, global, and asymmetric objectives, and proves the general inequality

$\mathcal T_\rho$1

while stressing that this does not imply global and local optima always coincide (Coyle et al., 2020).

At the asymptotic level, the channel-cloning formulation gives a sharp restriction for deterministic state cloning of any continuous set of states: the replication rate is at most linear. In that framework, deterministic super-linear replication of continuous state families obeys a strong converse; if one attempts $\mathcal T_\rho$2 with $\mathcal T_\rho$3, then asymptotically

$\mathcal T_\rho$4

This situates approximate wave-function cloning within a stricter asymptotic landscape than the finite-$\mathcal T_\rho$5 fidelity benchmarks usually emphasized in circuit-design work (Sekatski et al., 9 Sep 2025).

3. Variational quantum cloning as a hybrid learning algorithm

Variational quantum cloning (VQC) is a hybrid quantum-classical algorithm that learns a short-depth parameterized circuit implementing an approximate cloner for a specified family of states. The central object is not a universal exact cloner, but a family-specific circuit $\mathcal T_\rho$6 acting on $\mathcal T_\rho$7 input copies, $\mathcal T_\rho$8 blanks, and optional ancilla. The optimization problem is written as

$\mathcal T_\rho$9

where $$U\ket{\psi}_1 \ket{0}_2\ket{\mathcal{M}}_\mathcal{M} \rightarrow \ket{\psi}_1\ket{\psi}_2\ket{\mathcal{M}'}_\mathcal{M} \quad \forall \ket{\psi},$$0 are continuous parameters and $$U\ket{\psi}_1 \ket{0}_2\ket{\mathcal{M}}_\mathcal{M} \rightarrow \ket{\psi}_1\ket{\psi}_2\ket{\mathcal{M}'}_\mathcal{M} \quad \forall \ket{\psi},$$1 is a discrete circuit structure chosen from a gate pool. The algorithm samples training states from a target family $$U\ket{\psi}_1 \ket{0}_2\ket{\mathcal{M}}_\mathcal{M} \rightarrow \ket{\psi}_1\ket{\psi}_2\ket{\mathcal{M}'}_\mathcal{M} \quad \forall \ket{\psi},$$2, estimates local or global fidelities on quantum hardware, and updates both parameters and, optionally, circuit structure on a classical processor. The continuous optimization uses analytic gradients via the parameter-shift rule, while the structure-learning component perturbs a gate sequence and reoptimizes it. The paper further gives a sample-complexity estimate

$$U\ket{\psi}_1 \ket{0}_2\ket{\mathcal{M}}_\mathcal{M} \rightarrow \ket{\psi}_1\ket{\psi}_2\ket{\mathcal{M}'}_\mathcal{M} \quad \forall \ket{\psi},$$3

and argues that sufficiently shallow alternating layered ansätze avoid barren plateaus for the local cloning cost (Coyle et al., 2020).

This framework is explicitly motivated by quantum cryptanalysis. The 2020 VQC study treats approximate cloning as a resource for attacks on quantum coin-flipping protocols and emphasizes hardware-efficient short-depth circuits rather than analytically optimal but hardware-unfriendly constructions. For phase-covariant $$U\ket{\psi}_1 \ket{0}_2\ket{\mathcal{M}}_\mathcal{M} \rightarrow \ket{\psi}_1\ket{\psi}_2\ket{\mathcal{M}'}_\mathcal{M} \quad \forall \ket{\psi},$$4 cloning, it recalls the optimal local fidelity

$$U\ket{\psi}_1 \ket{0}_2\ket{\mathcal{M}}_\mathcal{M} \rightarrow \ket{\psi}_1\ket{\psi}_2\ket{\mathcal{M}'}_\mathcal{M} \quad \forall \ket{\psi},$$5

and the optimal global fidelity

$$U\ket{\psi}_1 \ket{0}_2\ket{\mathcal{M}}_\mathcal{M} \rightarrow \ket{\psi}_1\ket{\psi}_2\ket{\mathcal{M}'}_\mathcal{M} \quad \forall \ket{\psi},$$6

A learned circuit achieves nearly optimal local fidelity,

$$U\ket{\psi}_1 \ket{0}_2\ket{\mathcal{M}}_\mathcal{M} \rightarrow \ket{\psi}_1\ket{\psi}_2\ket{\mathcal{M}'}_\mathcal{M} \quad \forall \ket{\psi},$$7

but with lower global fidelity,

$$U\ket{\psi}_1 \ket{0}_2\ket{\mathcal{M}}_\mathcal{M} \rightarrow \ket{\psi}_1\ket{\psi}_2\ket{\mathcal{M}'}_\mathcal{M} \quad \forall \ket{\psi},$$8

On Rigetti Aspen-8, the learned hardware-adapted circuit outperforms the textbook ideal circuit by about $$U\ket{\psi}_1 \ket{0}_2\ket{\mathcal{M}}_\mathcal{M} \rightarrow \ket{\psi}_1\ket{\psi}_2\ket{\mathcal{M}'}_\mathcal{M} \quad \forall \ket{\psi},$$9 in observed cloning fidelity because it uses fewer entangling gates and fits the device topology better. The same study reports application-specific attack biases, including $\ket{\psi}$0 for a VQC-based attack on $\ket{\psi}$1, and $\ket{\psi}$2, $\ket{\psi}$3 for two attack models on $\ket{\psi}$4 (Coyle et al., 2020).

A 2024 photonic implementation realizes the same general idea in integrated optics. There, two dual-rail photonic qubits are used: one carries the state to be cloned, and the other is a blank target register initialized to $\ket{\psi}$5. The input state is encoded in modes $\ket{\psi}$6 and $\ket{\psi}$7, the blank in modes $\ket{\psi}$8 and $\ket{\psi}$9, and photon pairs are injected into modes $$\ket{\psi} = \cos(\theta)\ket{0} + \sin(\theta)e^{i\phi}\ket{1}.$$0 and $$\ket{\psi} = \cos(\theta)\ket{0} + \sin(\theta)e^{i\phi}\ket{1}.$$1. The trainable ansatz is a $$\ket{\psi} = \cos(\theta)\ket{0} + \sin(\theta)e^{i\phi}\ket{1}.$$2-mode universal interferometric mesh embedded in a $$\ket{\psi} = \cos(\theta)\ket{0} + \sin(\theta)e^{i\phi}\ket{1}.$$3-mode universal photonic integrated circuit; the orange variational region optimizes $$\ket{\psi} = \cos(\theta)\ket{0} + \sin(\theta)e^{i\phi}\ket{1}.$$4 internal phases. Output clones are identified as dual-rail qubits on modes $$\ket{\psi} = \cos(\theta)\ket{0} + \sin(\theta)e^{i\phi}\ket{1}.$$5 and $$\ket{\psi} = \cos(\theta)\ket{0} + \sin(\theta)e^{i\phi}\ket{1}.$$6. Rather than performing full tomography at each iteration, the chip is configured to project each clone onto the target input-state basis, and fidelities are estimated from post-selected two-fold coincidence events (Hoch et al., 2024).

For phase-covariant cloning in that device, the cost function is

$$\ket{\psi} = \cos(\theta)\ket{0} + \sin(\theta)e^{i\phi}\ket{1}.$$7

Because the cost contains at most second moments of the fidelities, the continuous equatorial average is replaced exactly by a minimal planar $$\ket{\psi} = \cos(\theta)\ket{0} + \sin(\theta)e^{i\phi}\ket{1}.$$8-design with four states,

$$\ket{\psi} = \cos(\theta)\ket{0} + \sin(\theta)e^{i\phi}\ket{1}.$$9

namely the eigenstates of Pauli $1\rightarrow 2$0 and $1\rightarrow 2$1. For state-dependent cloning, an additional regularization term involving post-selection probabilities $1\rightarrow 2$2 is added,

$1\rightarrow 2$3

to suppress pathological solutions with high conditional fidelity but vanishing success probability. The classical optimizer is the gradient-free Nelder–Mead simplex method over $1\rightarrow 2$4, augmented by a reboot strategy that enlarges the simplex around the best point found so far if noise and finite sampling induce premature collapse (Hoch et al., 2024).

The photonic experiment demonstrates near-optimal rather than exact learning. For phase-covariant cloning, training on the four-state design and validating on $1\rightarrow 2$5 evenly spaced equatorial test states leads to convergence after roughly $1\rightarrow 2$6 iterations; the paper states that the machine reaches an almost optimal circuit in under $1\rightarrow 2$7 iterations. The best measured average fidelities on the $1\rightarrow 2$8-state test set are

$1\rightarrow 2$9

which are below $\rho_1=\rho_2=\frac{5}{6}\dyad{\psi}+\frac{1}{6}\dyad{\psi_\bot},$0 but above the semiclassical threshold $\rho_1=\rho_2=\frac{5}{6}\dyad{\psi}+\frac{1}{6}\dyad{\psi_\bot},$1. The reported asymmetry indicates only approximate symmetry of the learned solution, despite the symmetry penalty in the cost. The authors attribute the residual gap mainly to incomplete convergence induced by experimental noise and finite counting statistics (Hoch et al., 2024).

4. Symmetry-restricted and physical cloning constructions

A conceptually different use of “wave function cloning” appears in mirror modular cloning for quantum associative memory. The memory state is

$\rho_1=\rho_2=\frac{5}{6}\dyad{\psi}+\frac{1}{6}\dyad{\psi_\bot},$2

and its mirrored version is

$\rho_1=\rho_2=\frac{5}{6}\dyad{\psi}+\frac{1}{6}\dyad{\psi_\bot},$3

The proposed state-dependent unitary acts only on the two-state set spanned by $\rho_1=\rho_2=\frac{5}{6}\dyad{\psi}+\frac{1}{6}\dyad{\psi_\bot},$4 and $\rho_1=\rho_2=\frac{5}{6}\dyad{\psi}+\frac{1}{6}\dyad{\psi_\bot},$5, with amplitudes determined by the overlap $\rho_1=\rho_2=\frac{5}{6}\dyad{\psi}+\frac{1}{6}\dyad{\psi_\bot},$6. On input $\rho_1=\rho_2=\frac{5}{6}\dyad{\psi}+\frac{1}{6}\dyad{\psi_\bot},$7, the unitary produces a flagged superposition whose ancilla indicates whether the output branch carries $\rho_1=\rho_2=\frac{5}{6}\dyad{\psi}+\frac{1}{6}\dyad{\psi_\bot},$8 or $\rho_1=\rho_2=\frac{5}{6}\dyad{\psi}+\frac{1}{6}\dyad{\psi_\bot},$9. Because retrieval from mirrored memory is equivalent to retrieval from the original memory followed by a global bitwise inversion, the paper describes this as “perfectly cloned up to global mirroring.” The caveat is essential: this is not a universal wave-function cloner, but a restricted symmetry-assisted transformation defined on a highly structured family of states (Diamantini et al., 2022).

The same paper couples this cloning step to a content-addressable retrieval algorithm. Using an input register $F_{1,2}= \Tr\!\left(\sqrt{\sqrt{\sigma}\rho_{1,2}\sqrt{\sigma}}\right)^2 = \frac{5}{6}, \qquad \sigma=\dyad{\psi}.$0, a cloned memory register, and $F_{1,2}= \Tr\!\left(\sqrt{\sqrt{\sigma}\rho_{1,2}\sqrt{\sigma}}\right)^2 = \frac{5}{6}, \qquad \sigma=\dyad{\psi}.$1 control qubits, it computes Hamming distances and amplifies patterns according to

$F_{1,2}= \Tr\!\left(\sqrt{\sqrt{\sigma}\rho_{1,2}\sqrt{\sigma}}\right)^2 = \frac{5}{6}, \qquad \sigma=\dyad{\psi}.$2

The number of amplitude-amplification iterations is

$F_{1,2}= \Tr\!\left(\sqrt{\sqrt{\sigma}\rho_{1,2}\sqrt{\sigma}}\right)^2 = \frac{5}{6}, \qquad \sigma=\dyad{\psi}.$3

and under an approximately uniform-pattern assumption the paper estimates

$F_{1,2}= \Tr\!\left(\sqrt{\sqrt{\sigma}\rho_{1,2}\sqrt{\sigma}}\right)^2 = \frac{5}{6}, \qquad \sigma=\dyad{\psi}.$4

Its claim of “perfect cloning” is therefore inseparable from associative-memory semantics: the mirrored branch is acceptable only because the downstream task treats $F_{1,2}= \Tr\!\left(\sqrt{\sqrt{\sigma}\rho_{1,2}\sqrt{\sigma}}\right)^2 = \frac{5}{6}, \qquad \sigma=\dyad{\psi}.$5 and $F_{1,2}= \Tr\!\left(\sqrt{\sqrt{\sigma}\rho_{1,2}\sqrt{\sigma}}\right)^2 = \frac{5}{6}, \qquad \sigma=\dyad{\psi}.$6 as operationally equivalent up to a final inversion (Diamantini et al., 2022).

An older but still relevant strand concerns phase-covariant optical cloning of polarization qubits. There the physical transformation is generated by

$F_{1,2}= \Tr\!\left(\sqrt{\sqrt{\sigma}\rho_{1,2}\sqrt{\sigma}}\right)^2 = \frac{5}{6}, \qquad \sigma=\dyad{\psi}.$7

with

$F_{1,2}= \Tr\!\left(\sqrt{\sqrt{\sigma}\rho_{1,2}\sqrt{\sigma}}\right)^2 = \frac{5}{6}, \qquad \sigma=\dyad{\psi}.$8

The input family is the equator of the Poincaré sphere, expressed through

$F_{1,2}= \Tr\!\left(\sqrt{\sqrt{\sigma}\rho_{1,2}\sqrt{\sigma}}\right)^2 = \frac{5}{6}, \qquad \sigma=\dyad{\psi}.$9

Amplification of one photon of a polarization singlet yields macroscopic states

$5/6$0

and a micro-macro singlet

$5/6$1

The associated numerical paper makes clear that the algorithmic contribution is primarily the simulation of these macrostates, not a general cloning algorithm (Buraczewski et al., 2011).

That numerical work addresses slowly convergent hypergeometric sums governing probabilities, mean photon number, variance, and distinguishability. Its methods include logarithmic evaluation of factorial expressions, dynamic cutoff estimation, reordering of nested sums, switching between complementary summation formulas depending on threshold size, and tile-based parallelization with OpenMP. The reported implementation uses about $5/6$2 MB RAM for a $5/6$3 tile and requires $5/6$4–$5/6$5 hours runtime per such tile. In this context, “wave function cloning algorithm” can denote the computational machinery used to model the output of a restricted optical cloner rather than the cloning map itself (Buraczewski et al., 2011).

5. Reformulation as channel cloning

A more general and conceptually unifying reformulation treats state cloning as a special case of deterministic cloning of quantum channels. Given a state family $5/6$6, define the corresponding trash-and-replace channels

$5/6$7

The central equivalence statement is

$5/6$8

where $5/6$9, and any state-cloning map $|\bar M\rangle$00 induces a channel-cloning process $|\bar M\rangle$01 through

$|\bar M\rangle$02

For trash-and-replace channels, the paper further proves that parallel, sequential, and non-causal higher-order processes collapse to the same power as an ordinary CPTP state-cloning map. In this sense, wave-function cloning is exactly the trash-and-replace corner of a broader theory of channel replication (Sekatski et al., 9 Sep 2025).

The channel framework replaces ordinary circuits by higher-order quantum operations, or processes, that take $|\bar M\rangle$03 queries of an unknown channel $|\bar M\rangle$04 and return a channel approximating $|\bar M\rangle$05. The quality criteria are worst-case Choi–Jamiołkowski fidelity,

$|\bar M\rangle$06

and worst-case diamond fidelity,

$|\bar M\rangle$07

The optimal worst-case channel-cloning fidelity is then

$|\bar M\rangle$08

This gives a single formalism for state cloning, unitary-gate replication, noisy-channel cloning, and super-replication questions (Sekatski et al., 9 Sep 2025).

Within that framework, the paper provides an SDP-based method to search for optimal cloning processes. Fidelity is represented by the semidefinite characterization

$|\bar M\rangle$09

A finite net of target channels is chosen, a process Choi operator is optimized under complete-positivity and process constraints, and measure-and-prepare protocols can be approximately isolated by adding a PPT condition. The paper also draws a direct connection between optimal channel cloning and Bayesian channel estimation, and shows that measure-and-prepare super-replication can already achieve the optimal quadratic scaling for some unitary-channel families, while deterministic state cloning remains limited to linear rate (Sekatski et al., 9 Sep 2025).

6. Computational hardness, limitations, and recurrent misconceptions

A distinct line of work asks whether cloning may remain hard even when the cloner is given classical side information specifying a verifier circuit. This is not the standard no-cloning setting. The task is: given one copy of a witness state $|\bar M\rangle$10 and a classical description of a polynomial-size verification circuit $|\bar M\rangle$11, output two valid copies, or more generally map the accepting space $|\bar M\rangle$12 of $|\bar M\rangle$13 to $|\bar M\rangle$14. The paper develops a conditional hardness framework using hidden maximally entangled states over representation-theoretically defined subspaces. For a verifier $|\bar M\rangle$15 uniquely accepting a hidden-EPR state $|\bar M\rangle$16 with completeness $|\bar M\rangle$17 and soundness $|\bar M\rangle$18, it conjectures that if a uniform polynomial-time algorithm performs

$|\bar M\rangle$19

then there must also exist a polynomial-time algorithm that outputs some state in $|\bar M\rangle$20. From this, and assuming $|\bar M\rangle$21, the paper derives a conditional theorem excluding efficient cloning for a representation-theoretic family of unique-witness verifiers in the case $|\bar M\rangle$22 (Havlíček et al., 2024).

This hardness result is computational rather than information-theoretic. It does not say that exact cloning is newly forbidden by linearity; instead, it argues that even when a classical verifier description weakens the usual “unknown state” premise, efficient cloning may still be impossible because witness generation or hidden-subspace reconstruction is computationally hard. The construction relies on projectors

$|\bar M\rangle$23

whose dimensions encode Kronecker multiplicities $|\bar M\rangle$24, and on hidden maximally entangled states

$|\bar M\rangle$25

The article’s relevance to “wave function cloning algorithms” is therefore negative but important: it delineates a regime in which algorithmic cloning may fail for reasons of complexity even after the strict information-theoretic unknown-state barrier has been softened (Havlíček et al., 2024).

Several misconceptions recur across the literature. First, no paper surveyed here provides a loophole around the no-cloning theorem for arbitrary unknown states. Second, “perfect cloning” can mean exactness only within a restricted equivalence class, as in mirror modular cloning up to global bitwise complementation. Third, “optimal cloning” usually means attaining the maximum fidelity allowed by quantum mechanics for a chosen ensemble, not fidelity $|\bar M\rangle$26. Fourth, some papers concern the search for a cloner or the simulation of a physical cloner rather than a standalone closed-form cloning circuit. A plausible implication is that “wave function cloning algorithm” is best understood as an umbrella label for restricted, task-specific, and often hardware- or application-adapted procedures, rather than as the name of a single universally accepted algorithmic object.