Quantum Black-Box Reduction: Methods & Limits

• Quantum black-box reduction is a family of techniques that access quantum operations solely via their input-output interfaces without inspecting internal implementations.
• It provides both constructive methods, such as state preparation using fixed-point amplitude amplification, and exposes limitations like the inability to control an unknown unitary's global phase.
• This approach underpins theoretical advances in quantum cryptography and complexity by formalizing oracle models and query-based separations in quantum systems.

Searching arXiv for recent and foundational papers on quantum black-box reductions, oracle models, and related impossibility/constructive results. “Quantum black-box reduction” denotes a family of techniques, models, and impossibility statements in which a quantum algorithm, protocol, or security proof is required to use another object only through its input–output interface, without exploiting its internal implementation. In the literature, this notion appears in several distinct but connected forms: relativized separations for pseudorandom quantum states, oracle-based limitations on controlling unknown unitaries, black-box proofs and impossibility results for proofs of quantumness, oracle/query models for SAT search and approximate counting, black-box state preparation, and fully black-box post-quantum secure computation. Taken together, these works show that black-box access is often sufficient for powerful constructions, but also that quantum mechanics and quantum complexity impose structural barriers absent from the classical setting (Bouaziz--Ermann et al., 2024).

1. Black-box reduction as a quantum methodological principle

In the surveyed literature, a black-box reduction is characterized by the requirement that a construction or security proof use its underlying primitive only as an oracle or subroutine, rather than by inspecting code or implementation details. In cryptography, this distinction is made explicit in work on post-quantum MPC: a protocol has a black-box construction if it “only relies on the input-output behaviour of the underlying cryptographic primitives,” and it has a black-box security proof if the proof uses the adversary only as a black box (Chatterjee et al., 19 Feb 2025). The same operational distinction is central in proofs of quantumness, where black-box reductions are contrasted with white-box or knowledge-based reductions that assume adversary-specific extractors accessing internal randomness or structure (Arabadjieva et al., 2024).

In oracle and query complexity, the black-box viewpoint is formalized through oracle access to a hidden object. For SAT-type tasks, this can be an “NP-like” existential oracle $O_x^\exists$, defined by phase access to the overlap predicate $\mathrm{overlap}(x,z)$, or a succinct SAT oracle $O_\varphi^{\mathrm{NP}}$ answering satisfiability of $\varphi \wedge \psi$ (Gharibian et al., 2024). In quantum algorithms, black-box state preparation assumes oracle access to amplitude data via $U_{\mathrm{amp}}$ or a bitwise phase oracle $U_\omega$, while the cost measure is query complexity plus gate counts (Bausch, 2020). In cloud hardware characterization, a black-box quantum computer is one that exposes only gate-level programming and execution times, while hiding its Hamiltonians and pulse-level details; the reduction then infers hidden energetic quantities from user-visible timing data (Ishida et al., 17 Dec 2025).

A separate line of work uses black-box reduction in a foundational sense: if a quantum operation is given only as an unknown device, one cannot in general coherently wrap it into higher-order control logic. The core impossibility is that a generic black-box unitary cannot be promoted to controlled-$U$, because a procedure accomplishing that would make the global phase of $U$ observable (Thompson et al., 2013). This marks a sharp contrast with classical software modularity, where black-box functions can be placed under arbitrary control flow.

2. Oracle separations and relativized barriers

A major use of quantum black-box reduction is to prove impossibility by relativization. The most explicit example is the separation between long pseudorandom quantum states and their logarithmic-length variants. A keyed family of $n(\lambda)$-qubit pure states $\{\ket{\varphi_k}\}_{k \in \{0,1\}^\lambda}$ is a pseudorandom quantum state (PRS) family if it is efficiently generated and computationally indistinguishable from Haar-random states even given polynomially many copies (Bouaziz--Ermann et al., 2024). A short-PRS is a PRS with $\mathrm{overlap}(x,z)$0, and the paper fixes the regime $\mathrm{overlap}(x,z)$1 with $\mathrm{overlap}(x,z)$2 for the main separation (Bouaziz--Ermann et al., 2024).

The central theorem states that there exists a quantum oracle $\mathrm{overlap}(x,z)$3 such that, relative to $\mathrm{overlap}(x,z)$4, $\mathrm{overlap}(x,z)$5-PRSs exist but $\mathrm{overlap}(x,z)$6-PRSs with $\mathrm{overlap}(x,z)$7 do not (Bouaziz--Ermann et al., 2024). The proof combines three ingredients already present in the data block: Kretschmer’s oracle with $\mathrm{overlap}(x,z)$8 and PRSs relative to $\mathrm{overlap}(x,z)$9; the implication from short-PRSs with $O_\varphi^{\mathrm{NP}}$0 to pseudo-deterministic one-way functions; and the proposition that PD-OWFs imply $O_\varphi^{\mathrm{NP}}$1 (Bouaziz--Ermann et al., 2024). The resulting oracle world contains PRSs but no short-PRSs, so any implication from PRSs to short-PRSs cannot be proved by fully relativizing black-box methods.

This is a paradigmatic black-box reduction barrier. Classically, pseudorandom generators can be stretched and shrunk by black-box manipulations; in the quantum setting, discarding qubits is not pseudorandomness-preserving because reduced states of entangled pure states can become close to maximally mixed rather than Haar-random pure (Bouaziz--Ermann et al., 2024). The separation therefore identifies a genuinely quantum obstruction to classical-style black-box reasoning.

A related phenomenon appears in QMA amplification. Relative to a quantum oracle based on $O_\varphi^{\mathrm{NP}}$2 rotation matrices $O_\varphi^{\mathrm{NP}}$3, no polynomial-resource QMA verification procedure can achieve completeness closer to $O_\varphi^{\mathrm{NP}}$4 than doubly exponential, nor soundness smaller than exponential, by black-box amplification (Aaronson et al., 25 Sep 2025). The completeness bound is derived from rational approximation of $O_\varphi^{\mathrm{NP}}$5, while the soundness bound comes from $O_\varphi^{\mathrm{NP}}$6; together they quantitatively refine Aaronson’s oracle separation between $O_\varphi^{\mathrm{NP}}$7 and $O_\varphi^{\mathrm{NP}}$8 (Aaronson et al., 25 Sep 2025). This suggests that any stronger amplification theorem for finite-dimensional QMA must be non-relativizing.

3. Constructive black-box reductions in quantum algorithms

Black-box reduction is also a positive algorithmic paradigm. In black-box quantum state preparation, the task is to prepare $O_\varphi^{\mathrm{NP}}$9 when amplitudes are available only through an oracle. One line of work studies an amplitude-writing oracle

$\varphi \wedge \psi$0

while another introduces a bitwise phase oracle

$\varphi \wedge \psi$1

and shows how to reduce the latter to the former without changing query complexity (Bausch, 2020). The target state actually prepared is

$\varphi \wedge \psi$2

where $\varphi \wedge \psi$3 is the $\varphi \wedge \psi$4-bit truncation of $\varphi \wedge \psi$5, and the trace-distance error satisfies

$\varphi \wedge \psi$6

(Bausch, 2020).

The main reduction in that work is from state preparation to oracle queries plus fixed-point amplitude amplification. Two Householder-type reflections $\varphi \wedge \psi$7 and $\varphi \wedge \psi$8 act on the span of $\varphi \wedge \psi$9, where $U_{\mathrm{amp}}$0 is an intermediate state whose ancilla amplitudes encode the bits $U_{\mathrm{amp}}$1 (Bausch, 2020). The unoptimized complexity matches prior $U_{\mathrm{amp}}$2 behavior, but a bootstrapped initial ancilla state $U_{\mathrm{amp}}$3, defined from average bit weights $U_{\mathrm{amp}}$4, improves the first-stage overlap to

$U_{\mathrm{amp}}$5

yielding total complexity

$U_{\mathrm{amp}}$6

(Bausch, 2020). For uniform, triangle, sine, and several random or broad distributions, this reduces the number of amplification rounds to $U_{\mathrm{amp}}$7 (Bausch, 2020).

A related state-preparation paper reformulates the task using linear combination of unitaries (LCU), again in a black-box oracle model for amplitude data, and reports improvements in additional qubits and Toffoli gates to $U_{\mathrm{amp}}$8 and $U_{\mathrm{amp}}$9, respectively, in the bit precision $U_\omega$0 (Wang et al., 2021). Across these works, the black-box reduction is from coherent loading of structured amplitude data to a fixed collection of oracle calls, controlled reflections, and ancilla engineering. This suggests that oracle granularity—bitwise phase access versus amplitude writing—can materially change the practical efficiency of the reduction, even when asymptotic query complexity is preserved (Bausch, 2020).

4. Limits of black-box manipulation of unknown quantum operations

One of the most fundamental limits on quantum black-box reduction is that an unknown unitary cannot, in general, be treated as a first-class subroutine in the classical programming sense. If a unitary $U_\omega$1 is available only as a physical process, then no protocol can in general compute its global phase or implement controlled-$U_\omega$2 using only black-box access to that process (Thompson et al., 2013). The obstruction is operational: $U_\omega$3 and $U_\omega$4 are physically identical as channels, but controlled-$U_\omega$5 and controlled-$U_\omega$6 differ by an observable relative phase on the control qubit (Thompson et al., 2013).

This no-go theorem constrains many familiar algorithmic templates. Phase estimation and DQC1 often assume one can coherently add control to a given unitary; that is valid when one has an explicit gate decomposition and can control each gate, but not when the unitary is supplied as a pure black-box process (Thompson et al., 2013). The paper therefore distinguishes between genuine black-box access and scenarios with extra physical structure, such as known invariant subspaces or linear-optical path degrees of freedom, where effective control becomes possible (Thompson et al., 2013).

The same work provides a constructive workaround for phase-insensitive tasks through “black-box DQC1.” Instead of controlled-$U_\omega$7, it uses two maximally mixed registers, a clean control qubit, and controlled-SWAP gates to simulate DQC1 on $U_\omega$8, thereby estimating $U_\omega$9 without ever implementing controlled-$U$0 (Thompson et al., 2013). The resulting protocol is fully black-box with respect to $U$1, since it requires only uncontrolled calls to $U$2 and fixed controlled-SWAP operations (Thompson et al., 2013). A plausible implication is that meaningful quantum black-box reductions often require reformulating the target task so that it is invariant under $U$3.

A later distributed-computing work pushes this idea much further by encoding unknown subroutines as Choi states $U$4 and manipulating them via oblivious quantum teleportation and oblivious quantum control (Xu et al., 20 May 2025). The framework does not implement arbitrary programmable gate arrays, which would violate the no-programming theorem; instead it reproduces measurement statistics of $U$5, compositions $U$6, multiplexers $U$7, DQC1-style trace estimation, and LCU/OAA/QSVT-type transformations, all while respecting no-go theorems (Xu et al., 20 May 2025). This does not contradict the impossibility of controlled-$U$8 for a raw black-box unitary; rather, it circumvents it by changing the representation of the unknown process and weakening the output requirement to measurement-level equivalence (Xu et al., 20 May 2025).

5. Black-box reductions in quantum cryptography and proofs of quantumness

In quantum cryptography, black-box reduction governs both feasibility and impossibility. Proofs of quantumness give a particularly explicit case. A proof of quantumness is an interactive test with an efficient quantum prover strategy and no efficient classical prover strategy above a specified soundness bound (Arabadjieva et al., 2024). For single-round protocols, the paper on knowledge assumptions states that security against classical provers cannot be reduced to a quantumly hard problem like LWE in a black-box manner, because the same reduction would then apply to the honest quantum prover and yield a QPT algorithm breaking the assumption (Arabadjieva et al., 2024).

This leads to a three-way distinction. Multi-round PoQs based on trapdoor claw-free functions can use black-box reductions and rewinding against classical provers, but they require mid-circuit measurements and classical feedback (Arabadjieva et al., 2024). Random-oracle-based single-round PoQs avoid some interaction but inherit the cost of implementing the oracle as a quantum hash function (Arabadjieva et al., 2024). The new single-round PoQs based on DDH or LWE plus knowledge assumptions achieve the same small quantum circuits as their multi-round counterparts, but the reductions are explicitly non-black-box because they rely on adversary-specific extractors guaranteed by KEA or lattice knowledge assumptions (Arabadjieva et al., 2024).

A different line shows that some PoQ functionality can be based on weaker classical assumptions than previously expected. Remote state preparation of $U$9, secure against classical malicious Bob, can be built from classically secure full-domain trapdoor permutations using a coherent version of the NOVY commitment protocol (Morimae et al., 2022). This yields proofs of quantumness from classically secure full-domain trapdoor permutations, without going through collision-resistant hashing, despite known black-box impossibility results for deriving collision-resistant hashing from trapdoor permutations (Morimae et al., 2022). The reduction chain is therefore

$U$0

but only against classical adversaries (Morimae et al., 2022).

Post-quantum secure computation exhibits the same theme at a larger scale. Fully black-box post-quantum 2PC with standard simulation-based security is impossible in constant rounds unless $U$1, but becomes feasible with $U$2-simulation: there are constant-round black-box protocols from post-quantum semi-honest OT, and with quantum communication even from post-quantum OWFs (Chia et al., 2021). Later work extends this landscape by giving the first fully black-box PQ-MPC with full simulation in polynomial rounds from post-quantum semi-honest OT, and constant-round black-box PQ-MPC with $U$3-simulation from post-quantum lossy PKE, linearly homomorphic PKE, or dense cryptosystems (Chatterjee et al., 19 Feb 2025). A central technical ingredient in these constructions is simulatable extraction: black-box extractors that recover committed secrets from QPT adversaries while disturbing their state by at most $U$4 in the relevant output distribution (Chia et al., 2021).

6. Query complexity, inference, and the broader scope of black-box reduction

Black-box reduction also appears in quantum complexity theory in a more literal query sense. For SAT search-to-decision, the relevant oracle is existential rather than standard bit access. The paper “BQP, meet NP” shows that every $U$5 problem lies in $U$6, so a quantum machine can compute an NP witness using only $U$7 NP queries (Gharibian et al., 2024). The upper bound combines approximate counting with Valiant–Vazirani isolation and a Bernstein–Vazirani trick that solves uniquely satisfiable formulas with a single NP query (Gharibian et al., 2024). In the non-succinct “NP-like” black-box oracle model, the same paper proves that $U$8 existential queries are necessary, even quantumly, for both witness extraction and approximate counting (Gharibian et al., 2024). This yields a rare case where the exact quantum black-box complexity of a reduction is pinned down up to constants.

A more interpretive use of black-box reasoning appears in the retrocausal account of oracle speedups. There, the quantum query complexity of an oracle problem is related to the classical query complexity of the same problem under advance knowledge of an $U$9-fraction of the hidden oracle specification, with $n(\lambda)$0 reproducing the known query counts for Deutsch, Deutsch–Jozsa, Grover on $n(\lambda)$1, Simon, and Abelian hidden subgroup examples in the authors’ framework (Castagnoli, 2015). This is not a lower-bound method in the standard adversary-polynomial sense, but it treats quantum speedup as a reduction from ordinary black-box computation to computation with partial prior information (Castagnoli, 2015). A plausible implication is that black-box reduction can serve not only as a proof methodology but also as an explanatory lens on the informational content of quantum query algorithms.

Finally, black-box reduction can mean inferring hidden physical quantities from operational data. In cloud quantum processors, only circuit-level access and execution time are visible. By combining gate-time amplification with the Margolus–Levitin and Mandelstam–Tamm bounds, energy scales of effective gate Hamiltonians can be estimated from execution-time slopes alone (Ishida et al., 17 Dec 2025). For IBM’s ibm_torino, the inferred shortest orthogonalization times are $n(\lambda)$2 ns, $n(\lambda)$3 ns, and $n(\lambda)$4 ns, corresponding to estimated energy scales $n(\lambda)$5, $n(\lambda)$6, and $n(\lambda)$7, respectively (Ishida et al., 17 Dec 2025). This is a black-box reduction from inaccessible Hamiltonian-level information to accessible timing observables.

Across these domains, a common pattern emerges. Black-box reductions are especially effective when the target task depends only on operationally invariant behavior—query responses, measurement statistics, or simulation indistinguishability—but they tend to fail when the task requires access to hidden representations, such as global phase, internal code, witness-specific structure, or non-relativizing complexity properties. This suggests that “quantum black-box reduction” is less a single theorem or definition than a unifying methodological boundary: it marks the point at which quantum information can still be treated abstractly as an oracle resource, and the point beyond which the structure of the implementation becomes essential (Thompson et al., 2013).