Non-Collapsing Measurement Oracle

• The non-collapsing measurement oracle is defined as a mechanism that extracts measurement outcomes from a quantum state without causing state collapse, preserving the Born-rule distribution.
• It is central to complexity classes like PDQP and variants of QMA, offering improved trade-offs in query and measurement resources for problems such as unstructured search and collision detection.
• The oracle plays dual roles: it challenges standard cryptographic constructions when available to adversaries and inspires nondestructive physical readout schemes that bypass typical ancilla overhead.

A non-collapsing measurement oracle is a hypothetical resource that returns measurement outcomes from a quantum state without disturbing the state that generated them. In the complexity-theoretic literature, it appears as an oracle augmenting quantum computation beyond standard BQP; in collapse-free or passive variants of quantum theory, it corresponds to retaining Born-rule outcome statistics while removing the projection postulate. Across these settings, the same core departure from standard quantum mechanics recurs: measurement outcomes occur probabilistically, but the post-measurement state is not updated by collapse (Miloschewsky et al., 2024, Fiorentino et al., 2023).

1. Formal oracle definitions

In formal oracle models, the input is typically a classical description of a quantum circuit of the form

$C=(U_1,M_1,\ldots,U_T,M_T),$

where the $U_t$ are unitaries and the $M_t$ are projective measurements. One explicit definition initializes $|\psi_0\rangle:=|0^\ell\rangle$, evolves through the circuit, and after each standard measurement $M_t$ with outcome $u_t$ defines the corresponding post-measurement state

$$|\psi_t\rangle=\frac{(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle}{\left\|(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle\right\|}.$$

The non-collapsing measurement oracle then samples

$$v_t\in\{0,1\}^\ell \quad\text{with probability}\quad |\langle v_t|\psi_t\rangle|^2$$

and outputs the tuple $(v_1,\ldots,v_T)$, while leaving the relevant state unchanged for the purpose of further non-collapsing access (Morimae et al., 6 Oct 2025). A related formulation in PDQP uses an oracle $\mathcal{Q}_P$ or $U_t$0 acting on a circuit $U_t$1, with an equivalent explicit circuit-based construction using parallel tensor-product “snapshots,” written as

$U_t$2

so that the full process builds a joint state $U_t$3 while preserving the desired marginals (Miloschewsky et al., 2024).

This oracle is therefore not merely a measurement primitive in the ordinary sense. It is a black-box sampling mechanism that exposes the Born-rule distribution of an intermediate or final quantum state without the usual disturbance associated with projective measurement. In standard quantum mechanics, repeated measurements in a basis after the first collapse would cease to provide fresh information; in the oracle model, repeated access continues to provide new samples from the same unchanged distribution.

2. Collapse-free and algebraic formulations

A related but distinct line of work studies quantum measurement without collapse at the level of physical or interpretational formalism. In passive quantum theory, observables remain Hermitian operators on a Hilbert space, states remain rays or density operators, and dynamics remain unitary, but measurement does not update the state. If

$U_t$4

then the probability of outcome $U_t$5 is still

$U_t$6

yet the state-update rule becomes

$U_t$7

rather than projection onto an eigenspace (Fiorentino et al., 2023).

Pavel Bóna’s “Quantum Measurement Without Collapse” formulates an allied idea in algebraic language. The key restriction is that any “human observer” can observe simultaneously just mutually compatible observables, formalized by restricting attention to an abelian $U_t$8-algebra $U_t$9. For an entangled system–apparatus state

$M_t$0

the expectation of any accessible observable $M_t$1 is

$M_t$2

which is mathematically equivalent, on the observer-accessible algebra, to a proper classical mixture over pointer states with weights $M_t$3 (Bóna, 2021).

These formulations replace collapse by restriction. In pQT, this has strong informational consequences: a tomographically complete set of observables can be repeatedly measured on a single system, so a single system suffices for state reconstruction, non-orthogonal pure states become perfectly distinguishable from one copy, and measurement-based cloning becomes possible after the state is learned (Fiorentino et al., 2023). In Bóna’s framework, the same restriction to commuting observables is presented as yielding a trivial solution of the Schrödinger cat paradox: the observer never accesses the off-diagonal terms corresponding to “dead and alive,” only the classical mixture over compatible macroscopic pointer observables (Bóna, 2021).

3. PDQP and the complexity of non-collapsing access

The complexity class PDQP was introduced as an alteration of BQP enhanced with the ability to obtain non-collapsing measurements, described as samples of quantum states without collapsing them. The class satisfies

$M_t$4

and it is known that

$M_t$5

At the same time, PostBQP is much more powerful, with $M_t$6, so PDQP is characterized as “just above” BQP rather than comparable to full postselection power (Miloschewsky et al., 2024).

Because unrestricted non-collapsing measurements would trivialize several tasks, the literature measures power using both the number of black-box queries $M_t$7 and the number of non-collapsing measurements $M_t$8. One general lower bound takes the form

$M_t$9

with $|\psi_0\rangle:=|0^\ell\rangle$0 derived from a weight scheme in the adversary method; under non-adaptive queries, the bound becomes

$|\psi_0\rangle:=|0^\ell\rangle$1

This framework yields tighter lower bounds and explicit query–measurement trade-offs across canonical black-box problems (Miloschewsky et al., 2024).

Problem	Standard BQP	PDQP
Unstructured Search	$|\psi_0\rangle:=|0^\ell\rangle$2	$|\psi_0\rangle:=|0^\ell\rangle$3 for $|\psi_0\rangle:=|0^\ell\rangle$4; $|\psi_0\rangle:=|0^\ell\rangle$5 for $|\psi_0\rangle:=|0^\ell\rangle$6
Collision	$|\psi_0\rangle:=|0^\ell\rangle$7	$|\psi_0\rangle:=|0^\ell\rangle$8
Majority	$|\psi_0\rangle:=|0^\ell\rangle$9	Between $M_t$0 and $M_t$1; tight $M_t$2 for $M_t$3
Element Distinctness	$M_t$4	$M_t$5, $M_t$6; tight $M_t$7 for $M_t$8

A central outcome is the tight $M_t$9 bound for unstructured search in PDQP, improving over the earlier $u_t$0 lower bound. The same work emphasizes that non-collapsing measurements break the deferred-measurement principle: information can be extracted “in the middle” of a computation without disturbing the state, which is precisely why problems such as collision become solvable with constant queries and measurements (Miloschewsky et al., 2024).

4. QMA variants, superposition detection, and NEXP

The addition of even a single non-collapsing measurement to QMA has much stronger consequences than in the bare decision setting. One result proves that QMA where the verifier may make a single non-collapsing measurement is equal to NEXP. The proof is organized around a property-testing primitive, superposition detection: given a $u_t$1-qubit state $u_t$2, the detector accepts with probability $u_t$3 on computational basis states, while states with

$u_t$4

are accepted with probability at most $u_t$5. A non-collapsing measurement implements such a detector efficiently on a constant-size register by first performing a non-collapsing measurement of the value register, then a standard collapsing measurement, and outputting $u_t$6 iff the two outcomes match (Bassirian et al., 2024).

This property-testing perspective is coupled to succinct gapped CSPs, rigidity arguments, and the earlier proof pattern behind $u_t$7. The result is that the power conferred by one non-collapsing measurement is exactly sufficient to enforce essentially classical behavior in the relevant post-measurement register, collapsing the verification problem upward to NEXP (Bassirian et al., 2024).

A companion result defines PDQMA, where Arthur is a PDQP verifier receiving a quantum witness, and DQMA, where Arthur can inspect the entire history of a hidden variable under mild assumptions on the hidden-variable theory. The main theorem is

$u_t$8

The proof uses the PCP theorem and the equivalence $u_t$9. Honest witnesses encode prover responses as multilinear extensions,

$$|\psi_t\rangle=\frac{(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle}{\left\|(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle\right\|}.$$0

and repeated non-collapsing measurements after projection onto affine lines provide enough evaluations for polynomial interpolation (Aaronson et al., 2024).

5. Cryptographic role: impossibility and hardness

In quantum black-box cryptography, non-collapsing measurement oracles are used as adversarial hyper-oracles to separate primitive classes. One paper studies a classical oracle $$|\psi_t\rangle=\frac{(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle}{\left\|(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle\right\|}.$$1 relative to which sub-exponentially secure indistinguishability obfuscation and one-way permutations exist even against adversaries that make quantum queries to a non-collapsing measurement oracle $$|\psi_t\rangle=\frac{(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle}{\left\|(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle\right\|}.$$2. Very roughly, $$|\psi_t\rangle=\frac{(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle}{\left\|(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle\right\|}.$$3 outputs the result of multiple non-collapsing measurements on the output of any quantum $$|\psi_t\rangle=\frac{(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle}{\left\|(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle\right\|}.$$4-aided circuit. This yields black-box separation results ruling out fully black-box quantum constructions of

$$|\psi_t\rangle=\frac{(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle}{\left\|(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle\right\|}.$$5

from

$$|\psi_t\rangle=\frac{(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle}{\left\|(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle\right\|}.$$6

The same work distinguishes $$|\psi_t\rangle=\frac{(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle}{\left\|(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle\right\|}.$$7 from a collision-finding oracle $$|\psi_t\rangle=\frac{(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle}{\left\|(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle\right\|}.$$8 and from a cloning unitary $$|\psi_t\rangle=\frac{(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle}{\left\|(P_{u_t}\otimes I)U_t|\psi_{t-1}\rangle\right\|}.$$9, the last of which is used to rule out fully black-box constructions of quantum lightning from public-key quantum money (Cakan et al., 6 Oct 2025).

A different cryptographic direction treats the inability to realize non-collapsing measurements efficiently as a hardness assumption. The class $$v_t\in\{0,1\}^\ell \quad\text{with probability}\quad |\langle v_t|\psi_t\rangle|^2$$0 is introduced as a sampling analogue of PDQP: it consists of sampling problems solvable by a deterministic classical polynomial-time algorithm making a single query to a non-collapsing measurement oracle. If $$v_t\in\{0,1\}^\ell \quad\text{with probability}\quad |\langle v_t|\psi_t\rangle|^2$$1 is hard on average for quantum polynomial time, then one-way puzzles exist. The same work introduces distributional collision-resistant puzzles (dCRPuzzs), proves that dCRPuzzs imply average-case hardness of $$v_t\in\{0,1\}^\ell \quad\text{with probability}\quad |\langle v_t|\psi_t\rangle|^2$$2, and shows that two-message honest-statistically-hiding commitments with classical communication and one-shot signatures imply dCRPuzzs (Morimae et al., 6 Oct 2025).

Taken together, these results place non-collapsing measurement in two opposite cryptographic roles. When granted directly to the adversary, it destroys black-box prospects for collision-resistance-type primitives; when assumed to be hard to simulate, it can serve as a basis for minimal quantum cryptographic assumptions. This suggests that the oracle is useful less as a realizable primitive than as a boundary marker for what black-box quantum cryptography can and cannot derive.

6. Physical readout, operational limits, and recurring misconceptions

The abstract oracle is unphysical in the complexity literature, but platform-specific work has explored non-collapsing or nondestructive readout mechanisms. In a quantum-dot Josephson junction, an Andreev qubit with arbitrary superposition

$$v_t\in\{0,1\}^\ell \quad\text{with probability}\quad |\langle v_t|\psi_t\rangle|^2$$3

can be probed through the intrinsic time-dependent oscillatory supercurrent. The measured current decomposes as

$$v_t\in\{0,1\}^\ell \quad\text{with probability}\quad |\langle v_t|\psi_t\rangle|^2$$4

with

$$v_t\in\{0,1\}^\ell \quad\text{with probability}\quad |\langle v_t|\psi_t\rangle|^2$$5

and

$$v_t\in\{0,1\}^\ell \quad\text{with probability}\quad |\langle v_t|\psi_t\rangle|^2$$6

In this proposal, $$v_t\in\{0,1\}^\ell \quad\text{with probability}\quad |\langle v_t|\psi_t\rangle|^2$$7 is extracted from the static component, $$v_t\in\{0,1\}^\ell \quad\text{with probability}\quad |\langle v_t|\psi_t\rangle|^2$$8 from the phase and amplitude of the oscillations, and the qubit frequency from the oscillation period; the readout is described as nondestructive and non-collapsing, avoids ancilla qubits, and removes the need for repetitive qubit resetting (Zhang et al., 2024).

One common misconception is to identify every nondemolition protocol with a non-collapsing measurement oracle. The abstract oracle returns repeated samples from an unchanged state distribution and is explicitly treated as “magical” or highly unphysical in PDQP and SampPDQP. By contrast, ancilla-based QND protocols preserve encoded information by transferring it to an ancilla and still incur substantial overhead from ancilla preparation, transfer, measurement, and reset; the Andreev-qubit protocol is notable precisely because it aims to avoid those ancilla steps (Zhang et al., 2024).

Another misconception is that non-collapsing access would automatically provide simultaneous information about incompatible observables. In the algebraic measurement-without-collapse framework, observer access is restricted to mutually compatible observables and is formalized by an abelian $$v_t\in\{0,1\}^\ell \quad\text{with probability}\quad |\langle v_t|\psi_t\rangle|^2$$9-algebra; joint distributions for noncommuting observables are not defined (Bóna, 2021). In the computational setting, the limitation takes another form: unrestricted non-collapsing measurements would trivialize search, enable quantum cloning and constant communication complexity, and collapse intended complexity separations, which is why $(v_1,\ldots,v_T)$0 and $(v_1,\ldots,v_T)$1 must be counted as separate resources (Miloschewsky et al., 2024).