Hamiltonian Phase Oracle

• Hamiltonian phase oracle is an oracle construction that uses Hamiltonian evolution to imprint computationally meaningful phases instead of relying on abstract black-box unitaries.
• It implements basis-dependent phase marking through continuous-time dynamics, enabling both discrete and spectral phase encoding via adiabatic transitions and ancilla coupling.
• It upholds standard query complexity bounds while revealing sensitivity to noise, which influences practical implementations in Grover-type searches and quantum circuit simulations.

Searching arXiv for recent and foundational papers on Hamiltonian phase oracles, continuous-time oracle models, and related implementations. A Hamiltonian phase oracle is an oracle construction in which the oracle’s action is generated by Hamiltonian dynamics rather than postulated as an abstract black-box unitary. In the standard discrete setting, the phase oracle acts as $\hat{O}_f|m\rangle = (-1)^{f(m)}|m\rangle$, with the marked basis state acquiring a sign flip. In Hamiltonian formulations, the same logical role is played by continuous-time evolution under an oracle Hamiltonian $H_Q(x)$, a problem Hamiltonian $H$, or an ancilla-coupled adiabatic Hamiltonian, so that phase information is produced physically by $e^{-i t H}$ or by more general time-dependent evolution (Yan et al., 2022, Yonge-Mallo, 2011, Plyashechnik et al., 2 Jun 2026). Across the literature, the term encompasses several closely related notions: continuous-time query models, diagonal Hamiltonians that imprint basis-dependent phases, and constructions in which the marked sector is defined either sharply, as in Grover search, or spectrally, as an energy resonance band.

1. Formal definition and oracle models

The canonical discrete phase oracle for search is

$$\hat{O}_f|m\rangle = (-1)^{f(m)}|m\rangle, \qquad f(m)= \begin{cases} 1 & m=\omega,\ 0 & m\neq \omega. \end{cases}$$

For an arbitrary database state $|\phi\rangle=\sum_m c_m |m\rangle$, this gives

$$\hat{O}_f|\phi\rangle=\sum_{m\neq \omega} c_m|m\rangle - c_\omega|\omega\rangle.$$

A Hamiltonian phase oracle is any construction that realizes this basis-dependent phase transformation through Hamiltonian evolution rather than by assuming direct access to a unitary query (Yan et al., 2022).

In the continuous-time Hamiltonian oracle model, the state evolves by the Schrödinger equation

$$i\frac{d}{dt} |\psi\rangle = H_x(t) |\psi\rangle, \qquad H_x(t)=g(t)H_Q(x)+H_D(t), \qquad |g(t)|\le 1.$$

Here $H_Q(x)$ is the oracle Hamiltonian, $H_D(t)$ is an input-independent driver Hamiltonian, and query complexity is defined as the minimum evolution time $H_Q(x)$0 needed to compute a Boolean function with bounded error (Yonge-Mallo, 2011). The oracle Hamiltonian has the decomposition

$H_Q(x)$1

with each $H_Q(x)$2 acting on an orthogonal subspace $H_Q(x)$3.

This model is the continuous-time limit of fractional queries. The discrete fractional query

$H_Q(x)$4

is recovered by choosing $H_Q(x)$5 diagonal with $H_Q(x)$6 in the $H_Q(x)$7-th diagonal entry, setting $H_Q(x)$8, $H_Q(x)$9, and evolving for time $H$0. In this sense, Hamiltonian phase oracles are not merely analogous to discrete phase oracles; they explicitly contain them as a special case (Yonge-Mallo, 2011).

A distinct but related usage appears when the oracle is taken to be direct physical evolution under a diagonal Hamiltonian,

$H$1

so that each computational basis state is assigned a phase determined by its energy. In that setting, the oracle need not encode a Boolean predicate at all; it may instead implement a continuous phase labeling of the spectrum (Plyashechnik et al., 2 Jun 2026).

2. Adiabatic realization of the Grover phase oracle

A concrete physical realization of the Grover oracle is given by an adiabatic phase gate generated from a search Hamiltonian $H$2 coupled to an ancilla (Yan et al., 2022). The database is represented by a diagonal Hermitian operator whose eigenbasis is the computational basis,

$H$3

with the constant $H$4 chosen so that the marked state is the unique negative-energy eigenstate:

• $H$5 for all unmarked states,
• $H$6 for the marked state.

The oracle Hamiltonian is

$H$7

where $H$8 and $H$9 are spin-1 operators, and the initial state is

$e^{-i t H}$0

Because $e^{-i t H}$1 is diagonal, the evolution decomposes into independent sectors labeled by computational basis states: $e^{-i t H}$2 In each sector, the ancilla sees an effective magnetic field

$e^{-i t H}$3

The marked state is distinguished by the sign of its eigenvalue rather than by a direct circuit oracle. For $e^{-i t H}$4 and $e^{-i t H}$5, the ancilla follows two topologically distinct adiabatic paths, and the phase difference is

$e^{-i t H}$6

After the annealing process, up to global phase,

$e^{-i t H}$7

which is exactly the Grover phase flip.

A central feature of this construction is the absence of unwanted dynamical phase. For the adiabatic eigenstate with zero ancillary spin projection, the Hamiltonian expectation value vanishes in every sector, so the dynamical phase is identically zero; the relative sign is purely geometric/topological. This is why the construction functions as a phase oracle rather than merely as a state-preparation protocol.

The schedule

$e^{-i t H}$8

suppresses diabatic transitions exponentially near the endpoints. The resulting oracle error is exponentially small in the total annealing time $e^{-i t H}$9, in contrast to the power-law suppression obtained from a linear schedule. For fixed overall error after the $$\hat{O}_f|m\rangle = (-1)^{f(m)}|m\rangle, \qquad f(m)= \begin{cases} 1 & m=\omega,\ 0 & m\neq \omega. \end{cases}$$0 oracle calls used by Grover search, the paper states that one requires

$$\hat{O}_f|m\rangle = (-1)^{f(m)}|m\rangle, \qquad f(m)= \begin{cases} 1 & m=\omega,\ 0 & m\neq \omega. \end{cases}$$1

The appendix further gives equivalent realizations using two spin-$$\hat{O}_f|m\rangle = (-1)^{f(m)}|m\rangle, \qquad f(m)= \begin{cases} 1 & m=\omega,\ 0 & m\neq \omega. \end{cases}$$2 ancillas, including a deterministic Bell-state protocol in which opposite adiabatic phases cancel except for the desired sign on the marked state (Yan et al., 2022).

3. Query complexity, adversary bounds, and noise sensitivity

Continuous-time Hamiltonian phase oracles obey the same adversary lower bounds as the standard discrete query model (Yonge-Mallo, 2011). For a Boolean function $$\hat{O}_f|m\rangle = (-1)^{f(m)}|m\rangle, \qquad f(m)= \begin{cases} 1 & m=\omega,\ 0 & m\neq \omega. \end{cases}$$3 with adversary matrix $$\hat{O}_f|m\rangle = (-1)^{f(m)}|m\rangle, \qquad f(m)= \begin{cases} 1 & m=\omega,\ 0 & m\neq \omega. \end{cases}$$4, the bounded-error query complexity in the Hamiltonian oracle model satisfies

$$\hat{O}_f|m\rangle = (-1)^{f(m)}|m\rangle, \qquad f(m)= \begin{cases} 1 & m=\omega,\ 0 & m\neq \omega. \end{cases}$$5

and the same conclusion extends to the general adversary method with negative weights.

The proof is a continuous-time adversary argument. The progress measure is built from weighted overlaps of the algorithm states for different inputs. A decisive structural fact is that the driver Hamiltonian cancels in the difference

$$\hat{O}_f|m\rangle = (-1)^{f(m)}|m\rangle, \qquad f(m)= \begin{cases} 1 & m=\omega,\ 0 & m\neq \omega. \end{cases}$$6

so the lower bound depends only on the oracle structure, not on the input-independent dynamics. The rate of change of the progress measure satisfies

$$\hat{O}_f|m\rangle = (-1)^{f(m)}|m\rangle, \qquad f(m)= \begin{cases} 1 & m=\omega,\ 0 & m\neq \omega. \end{cases}$$7

and integration over total time yields the same adversary expression as in the discrete model. The immediate consequence is that Hamiltonian phase oracles do not evade standard information-theoretic query lower bounds.

Noise changes the situation qualitatively. For unstructured search with oracle Hamiltonian

$$\hat{O}_f|m\rangle = (-1)^{f(m)}|m\rangle, \qquad f(m)= \begin{cases} 1 & m=\omega,\ 0 & m\neq \omega. \end{cases}$$8

a faulty oracle subject to dephasing in the marked-state basis is described by the Lindblad master equation

$$\hat{O}_f|m\rangle = (-1)^{f(m)}|m\rangle, \qquad f(m)= \begin{cases} 1 & m=\omega,\ 0 & m\neq \omega. \end{cases}$$9

This is the continuous-time analogue of a faulty phase oracle. The lower bound proved for this model is

$|\phi\rangle=\sum_m c_m |m\rangle$0

for success probability $|\phi\rangle=\sum_m c_m |m\rangle$1. Hence, for constant dephasing rate $|\phi\rangle=\sum_m c_m |m\rangle$2, the runtime is

$|\phi\rangle=\sum_m c_m |m\rangle$3

so the quadratic Grover speedup disappears (Temme, 2014).

A common misconception is that continuous-time oracle access might soften discrete lower bounds or improve noise robustness. The literature supports the opposite conclusion: the adversary method survives intact in the Hamiltonian model, and constant dephasing of the oracle phase collapses the ideal $|\phi\rangle=\sum_m c_m |m\rangle$4 behavior to classical scaling (Yonge-Mallo, 2011, Temme, 2014).

4. Spectral and energy-selective Hamiltonian phase oracles

A more general Hamiltonian phase oracle need not define a Boolean marked set at all. In the Ising setting, direct physical evolution

$|\phi\rangle=\sum_m c_m |m\rangle$5

acts as a phase oracle because

$|\phi\rangle=\sum_m c_m |m\rangle$6

Configurations with

$|\phi\rangle=\sum_m c_m |m\rangle$7

acquire phase near $|\phi\rangle=\sum_m c_m |m\rangle$8, so the oracle “marks configurations continuously by their phases” and selects a finite resonance band rather than a preassigned set (Plyashechnik et al., 2 Jun 2026).

Alternating this oracle with the Grover diffusion operator

$|\phi\rangle=\sum_m c_m |m\rangle$9

produces a Grover-type amplification peak even though the marking is not Boolean. Writing the empirical density of states as

$$\hat{O}_f|\phi\rangle=\sum_{m\neq \omega} c_m|m\rangle - c_\omega|\omega\rangle.$$0

and the energy-grouped amplitudes as

$$\hat{O}_f|\phi\rangle=\sum_{m\neq \omega} c_m|m\rangle - c_\omega|\omega\rangle.$$1

the dynamics satisfy the exact recurrence

$$\hat{O}_f|\phi\rangle=\sum_{m\neq \omega} c_m|m\rangle - c_\omega|\omega\rangle.$$2

The characteristic-function moments

$$\hat{O}_f|\phi\rangle=\sum_{m\neq \omega} c_m|m\rangle - c_\omega|\omega\rangle.$$3

and their generating function determine the peak position, width, and height.

For an annealed Gaussian density of states, the paper derives the asymptotic query complexity

$$\hat{O}_f|\phi\rangle=\sum_{m\neq \omega} c_m|m\rangle - c_\omega|\omega\rangle.$$4

when the resonance band contains $$\hat{O}_f|\phi\rangle=\sum_{m\neq \omega} c_m|m\rangle - c_\omega|\omega\rangle.$$5 configurations. In the special case $$\hat{O}_f|\phi\rangle=\sum_{m\neq \omega} c_m|m\rangle - c_\omega|\omega\rangle.$$6, this reproduces the usual Grover scaling $$\hat{O}_f|\phi\rangle=\sum_{m\neq \omega} c_m|m\rangle - c_\omega|\omega\rangle.$$7. For correlated random Ising spectra, overlap-induced correlations shift and distort the peak. The phase displacement scales as

$$\hat{O}_f|\phi\rangle=\sum_{m\neq \omega} c_m|m\rangle - c_\omega|\omega\rangle.$$8

which translates into an energy shift

$$\hat{O}_f|\phi\rangle=\sum_{m\neq \omega} c_m|m\rangle - c_\omega|\omega\rangle.$$9

Although small on the full spectral scale, this detuning is large relative to the exponentially narrow resonance width and is therefore algorithmically relevant. The paper proposes two correction strategies: spectral symmetrization via an ancilla and iterative calibration of the evolution time $$i\frac{d}{dt} |\psi\rangle = H_x(t) |\psi\rangle, \qquad H_x(t)=g(t)H_Q(x)+H_D(t), \qquad |g(t)|\le 1.$$0 (Plyashechnik et al., 2 Jun 2026).

This suggests a broader interpretation of Hamiltonian phase oracles: the oracle may be a spectral filter rather than a Boolean marker. The distinction is operationally significant. In the Boolean case the oracle identifies a sharply defined subspace; in the spectral case it induces an effective marked band whose width, detuning, and amplification profile are dynamical quantities.

5. Circuit realizations and programmable diagonal Hamiltonians

Several recent constructions treat diagonal Hamiltonians themselves as programmable phase oracles. One such approach learns a diagonal Hamiltonian

$$i\frac{d}{dt} |\psi\rangle = H_x(t) |\psi\rangle, \qquad H_x(t)=g(t)H_Q(x)+H_D(t), \qquad |g(t)|\le 1.$$1

whose evolution acts as

$$i\frac{d}{dt} |\psi\rangle = H_x(t) |\psi\rangle, \qquad H_x(t)=g(t)H_Q(x)+H_D(t), \qquad |g(t)|\le 1.$$2

The learned phases are combined with Hadamard layers in the fixed-depth architecture

$$i\frac{d}{dt} |\psi\rangle = H_x(t) |\psi\rangle, \qquad H_x(t)=g(t)H_Q(x)+H_D(t), \qquad |g(t)|\le 1.$$3

with

$$i\frac{d}{dt} |\psi\rangle = H_x(t) |\psi\rangle, \qquad H_x(t)=g(t)H_Q(x)+H_D(t), \qquad |g(t)|\le 1.$$4

so that the final state is

$$i\frac{d}{dt} |\psi\rangle = H_x(t) |\psi\rangle, \qquad H_x(t)=g(t)H_Q(x)+H_D(t), \qquad |g(t)|\le 1.$$5

The claim is that, with oracle access to the learned Hamiltonian parameters, $$i\frac{d}{dt} |\psi\rangle = H_x(t) |\psi\rangle, \qquad H_x(t)=g(t)H_Q(x)+H_D(t), \qquad |g(t)|\le 1.$$6 classical data values can be encoded into $$i\frac{d}{dt} |\psi\rangle = H_x(t) |\psi\rangle, \qquad H_x(t)=g(t)H_Q(x)+H_D(t), \qquad |g(t)|\le 1.$$7 qubits using $$i\frac{d}{dt} |\psi\rangle = H_x(t) |\psi\rangle, \qquad H_x(t)=g(t)H_Q(x)+H_D(t), \qquad |g(t)|\le 1.$$8 quantum queries after an $$i\frac{d}{dt} |\psi\rangle = H_x(t) |\psi\rangle, \qquad H_x(t)=g(t)H_Q(x)+H_D(t), \qquad |g(t)|\le 1.$$9 classical preprocessing stage. For structured datasets, the diagonal Hamiltonian may be rewritten in a Walsh basis and truncated to one-local and two-local terms, eliminating black-box oracle access and yielding hardware-efficient circuits with no Trotter error because the retained Walsh operators commute (Ramezani et al., 22 Dec 2025).

At the circuit-synthesis level, low-depth phase oracles of the form

$H_Q(x)$0

can be built from piecewise linear approximations $H_Q(x)$1. A parallel piecewise circuit fans out the input register, computes segment flags, and applies all elementary rotations simultaneously, achieving rotation depth one for the large rotation block. For an $H_Q(x)$2-qubit register and an $H_Q(x)$3-segment approximation, the paper states a depth as low as

$H_Q(x)$4

Its stated applications include diagonal Hamiltonian terms such as the Coulomb interaction in grid-based many-body simulation (Sun et al., 2024).

A further line of work uses time evolution on a block-encoded Hamiltonian as the phase-imprinting resource in a qubit-efficient form of quantum phase estimation. The relevant coupling is

$H_Q(x)$5

where a pointer register initialized in a uniform superposition acquires an eigenvalue-dependent phase and an inverse QFT yields an estimate of the eigenvalue. The result is a phase-measurement procedure in which the Hamiltonian functions as a phase-imprinting oracle through simulated evolution rather than through a primitive discrete query (Skelton, 4 Sep 2025). Closely related ancilla-assisted constructions also embed a shifted and scaled Hamiltonian

$H_Q(x)$6

into a larger unitary so that phase estimation can access Hamiltonian eigenvalues without explicit construction of $H_Q(x)$7 (Daskin et al., 2017).

6. Related constructions, scope, and terminological boundaries

The literature also contains several constructions that are phase-oracle-like but should be distinguished from Hamiltonian phase oracles in the strict sense. A canonical oracle-construction pattern starts from an algebraic expression or Ising model that maps all selected basis states to a common value, then applies a simple matching oracle to that value: $H_Q(x)$8 This produces an effective phase-flip oracle suitable for generalized Grover iteration,

$H_Q(x)$9

The method applies in particular to quadratic Hamiltonians such as Ising models, but its defining mechanism is compute–match–uncompute rather than continuous-time Hamiltonian evolution (Gilliam et al., 2020).

Constraint-checking oracles for Gauss’s law in Abelian lattice gauge theories form another adjacent category. There the projector

$H_D(t)$0

defines the sign oracle

$H_D(t)$1

Operationally, however, the implementation is an ancilla-based projective error-detection oracle: it computes the local Gauss-law predicate, applies phase kickback, uncomputes, and measures a query bit. It is therefore best understood as a projector oracle or constraint oracle rather than as a Hamiltonian phase oracle in the Grover or continuous-time query sense (Stryker, 2018).

Another important boundary concerns quantum signal processing. The phase factors $H_D(t)$2 used in QSP satisfy

$H_D(t)$3

and serve as compiler parameters for matrix-polynomial transformations. They are not themselves phase oracles. The relevant contribution in that literature is stable classical synthesis of these angles, not construction of a Hamiltonian phase oracle (Dong et al., 2020).

Finally, some work explicitly removes oracle assumptions. For tridiagonal Hamiltonians, one can derive a structured Pauli decomposition, partition the terms into $H_D(t)$4 commuting subsets, and simulate

$H_D(t)$5

by Clifford diagonalization plus Trotterization without sparse-access or entry-value oracles. This replaces black-box oracle access with explicit matrix structure (Arseniev et al., 2023). A plausible implication is that “Hamiltonian phase oracle” is best treated as a model-dependent term: in some contexts it denotes an actual physical oracle mechanism, while in others it denotes an abstraction whose practical relevance depends on whether the oracle itself is efficiently realizable.

In aggregate, the term refers to a family of constructions unified by one principle: phase marking is produced by Hamiltonian structure. The marked information may be Boolean, geometric, spectral, or learned; the implementation may be adiabatic, continuous-time, block-encoded, or diagonal; but the central operation is the same—the oracle acts by attaching computationally meaningful phases through Hamiltonian evolution rather than by assuming an opaque black-box unitary.