---
title: Energy-Selective Quantum Search
url: https://www.emergentmind.com/topics/energy-selective-quantum-search
type: topic
---

# Energy-Selective Quantum Search

Searching arXiv for relevant papers on energy-selective quantum search and closely related formulations.
In the literature surveyed here, energy-selective quantum search encompasses search procedures in which the search target is singled out through energetic bias, spectral phase, or relaxation dynamics rather than through a purely Boolean marked-set oracle. The relevant constructions include a generalized analog search Hamiltonian with a tunable energy ratio $r=E'/E$, non-Hermitian adiabatic search with a finite complex gap, dissipative relaxation of an $N$-level system into an unknown ground state, Hamiltonian phase oracles based on direct Ising evolution $\exp(-iTH)$, and exciton trapping at an impurity site whose energy defect marks the “winner” configuration [1903.11186][1208.4642][2106.02703][2606.03380][1002.4056].

## 1. Conceptual setting and relation to oracle search

A recurring feature of these models is that search is reformulated as dynamics toward energetically distinguished configurations. In the generalized analog Hamiltonian search, the target $\lvert w\rangle$ and source $\lvert s\rangle$ are assigned distinct energy weights through
$$
H = E\,|w\rangle\langle w| + E'\,|s\rangle\langle s| ,
$$
with $r=E'/E$ acting as an additional control parameter. In the non-Hermitian adiabatic construction, the target state is the ground state of $H_0=-|m\rangle\langle m|$, while an auxiliary term proportional to $(g+i\delta)(1-s)H_1$ controls the instantaneous gap. In the dissipative database search, the unique marked element is made the ground state of an $N$-level system and the task is recast as relaxation under a master equation. In the Ising Hamiltonian phase-oracle construction, the native evolution $U_T=\exp(-iTH)$ marks configurations continuously by phases and selects a finite resonance band rather than a preassigned marked set. In the Frenkel-exciton model, a shallow isotopic impurity introduces a site-energy shift $\Delta_p$ that singles out $\lvert w\rangle$ energetically [1903.11186][1208.4642][2106.02703][2606.03380][1002.4056].

These formulations differ materially from the standard Boolean-oracle model. Grover’s algorithm and its unitary Hamiltonian analogue accomplish unstructured search in $\mathcal{O}(\sqrt{N})$ time steps, while the dissipative model is described as incoherent Markov dynamics in $N$ levels with no “oracle calls,” and the Ising phase-oracle model uses a continuous spectral oracle rather than a Boolean marking rule [2106.02703][2606.03380]. This suggests that complexity comparisons must be read together with the assumed oracle, spectral, and control resources.

## 2. Generalized analog search and the energy-ratio parameter

The generalized analog construction operates in an $N$-dimensional Hilbert space with two special normalized states, the target $\lvert w\rangle$ and the source $\lvert s\rangle$, with overlap
$$
x\equiv \langle w|s\rangle,\qquad 0\le |x|<1.
$$
Because the Hamiltonian acts only in $\mathrm{span}\{\lvert w\rangle,\lvert s\rangle\}$, the dynamics reduces to a two-dimensional problem after introducing
$$
|r\rangle \equiv \frac{|s\rangle - x|w\rangle}{\sqrt{1-x^2}},
$$
so that $\lvert s\rangle=x\lvert w\rangle+\sqrt{1-x^2}\lvert r\rangle$. The exact propagator is $U(t)=e^{-iHt/\hbar}$, with eigenfrequency
$$
\Omega=\frac{\sqrt{4x^2EE'+(E'-E)^2}}{2\hbar}.
$$
The success probability is
$$
P(t)=|\langle w|U(t)|s\rangle|^2,
$$
and its peak value is
$$
P_{\max}(x,r)=\frac{x^2(1+r)^2}{4x^2r+(1-r)^2},
$$
while the first time at which the maximum is attained is
$$
T_{\mathrm{opt}}(x,r)=\frac{\pi\hbar}{2E\sqrt{4x^2r+(1-r)^2}}.
$$
For $r=1$, one recovers the original Farhi–Gutmann result, with $P_{\max}=1$ and $T_{\mathrm{FG}}=\pi\hbar/(2Ex)$ [1903.11186].

The central feature of this model is a speed-versus-fidelity trade-off. When $r>1$, the maximum fidelity drops below $1$ but $T_{\mathrm{opt}}$ can be shortened because the denominator $\sqrt{4x^2r+(1-r)^2}$ becomes larger than $x$. The fidelity deficit is parametrized by
$$
P_{\max}(x,r)\equiv \cos^2\delta,
$$
with
$$
\delta(x,r)=\arccos\!\left[\frac{(1+r)x}{\sqrt{(1-r)^2+4rx^2}}\right].
$$
The analysis imposes the condition that the search infidelity not exceed the minimum-error probability for ambiguous discrimination of two pure states,
$$
1-\cos^2\delta \le p_E(\delta).
$$
For symmetric priors $p_w=p_t=1/2$, this reproduces a Fuchs–van-de-Graaf-type bound. Numerically, the advantage region
$$
R=\{(x,r):0<x<1,\ r\ge 1,\ P_{\max}(x,r)\ge P_{\mathrm{th}},\ T_g(P_{\mathrm{th}},x,r)<T_s(P_{\mathrm{th}},x)\}
$$
is non-empty, so the modified algorithm can attain a prescribed near-optimal fidelity in strictly less time than the $r=1$ search [1903.11186].

## 3. Non-Hermitian adiabatic search and engineered finite gaps

The non-Hermitian adiabatic algorithm for Grover-type search begins with
$$
H_0=-|m\rangle\langle m|,\qquad H_1=-|\psi_0\rangle\langle\psi_0|,
$$
where
$$
|\psi_0\rangle=\frac1{\sqrt N}\sum_{i=1}^N |i\rangle.
$$
Using the schedule parameter $s=t/\tau\in[0,1]$ and the complex coupling
$$
h(t)=\gamma(\tau-t),\qquad \gamma=\frac{g+i\delta}{\tau},
$$
the full Hamiltonian is
$$
H(s)=H_0+(g+i\delta)(1-s)H_1 .
$$
Because both $H_0$ and $H_1$ have degeneracy $N-1$ at zero energy, the dynamics reduces exactly to the two-dimensional subspace spanned by $\{|\psi_0\rangle,|m\rangle\}$. The instantaneous eigenvalues are
$$
E_{0,1}(s)=-\frac{\varepsilon(s)}2 \mp \frac12\,\Omega(s),
$$
with
$$
\varepsilon(s)=1+h(s),\qquad
\Omega(s)=\sqrt{h(s)^2-2h(s)\cos\alpha+1},\qquad
\sin\frac\alpha2=\frac1{\sqrt N}.
$$
The complex gap is $\Delta E(s)=\Omega(s)$, and for large $N$ its minimum magnitude is finite:
$$
\min_{s\in[0,1]}|\Delta E(s)|\approx \frac{\delta}{\sqrt{g^2+\delta^2}}+\mathcal{O}\!\left(\frac1N\right).
$$
The minimum occurs near the avoided crossing $s\simeq 1$ [1208.4642].

Because the minimum gap is controlled by $\delta$ rather than by $1/\sqrt N$, the adiabatic run time required for adiabatic following is
$$
\tau \gtrsim \frac{g^2}{\delta}\,\ln N.
$$
In the Hermitian limit $\delta=0$, the minimum gap is $\sim 1/\sqrt N$, giving $\tau\sim N$ for a global schedule. The success probability tends to $1$ already for $\tau\sim (g^2/\delta)\ln N$, whereas the overall norm decays because of the imaginary component:
$$
P_{\mathrm{survive}}(t)=\|\Psi(t)\|^2\approx e^{-\delta t}.
$$
The trade-off is therefore explicit: adding a small imaginary part $\delta$ opens a finite minimum gap and allows an adiabatic run time only logarithmic in $N$, but the system norm decays as $e^{-\delta t}$, so $\delta$ must be chosen small enough to keep $P_{\mathrm{survive}}\sim 1$ over the evolution interval [1208.4642].

## 4. Dissipative relaxation search in logarithmic spectra

The dissipative search paradigm maps an unstructured database onto an $N$-level system weakly coupled to a thermal bath. The marked element corresponds to an unknown index $\ell$ and is made the ground state by
$$
H_0=\epsilon\,|w_\ell\rangle\langle w_\ell|,\qquad \epsilon<0.
$$
A commuting splitting Hamiltonian $H_1$ with nondegenerate spectrum $\{\eta_k\}$ removes the degeneracy of the remaining $N-1$ states, so that
$$
H=H_0+H_1.
$$
A central construction chooses
$$
\eta_k=a\ln k,\qquad k=1,\dots,N,\qquad a>0,
$$
and
$$
\epsilon=-b\ln N,\qquad b>a,
$$
which yields the gap
$$
\Delta_N\equiv \eta_1-(\epsilon+\eta_N)=(b-a)\ln N>0.
$$
The weak-coupling, Born–Markov limit produces a classical master equation for the populations,
$$
\dot p_k=\sum_{\ell=1}^N v_{k\ell}p_\ell-p_k\sum_{\ell=1}^N v_{\ell k},
$$
with detailed balance
$$
v_{k\ell}e^{-\beta\epsilon_\ell}=v_{\ell k}e^{-\beta\epsilon_k}.
$$
An equivalent population-only Lindblad description uses jump operators
$$
L_{k\leftarrow \ell}=\sqrt{v_{k\ell}}\,|w_k\rangle\langle w_\ell|.
$$
The generalized Glauber rates
$$
v_{k\ell}=\frac{v}{\max(n_k,n_\ell)}\,(1+e^{-\beta(\epsilon_\ell-\epsilon_k)})^{-1}
$$
satisfy detailed balance together with the bounded-rate condition
$$
\gamma_k\equiv \sum_\ell v_{\ell k}\lesssim v=O(1)\qquad \forall k
$$
[2106.02703].

The relaxation time is governed by the slowest nonzero eigenvalue $\alpha_2$ of the rate matrix $A$:
$$
\tau_{\mathrm{rlx}}=\frac1{|\alpha_2|}.
$$
Numerically, for $a\beta>1$ and $b>a$,
$$
\tau_{\mathrm{rlx}}\,v \simeq 1.82\,\ln N - \mathrm{const},
$$
so $\tau_{\mathrm{rlx}}=O(\ln N)$. The mechanism is attributed to three simultaneous conditions: at low temperature $\beta(b-a)>1$ the Gibbs weight overwhelmingly concentrates on level $1$; the long-range rates allow direct jumps between any two levels; and the logarithmic spacing keeps upward and downward hops moderately biased while each site’s total exit rate remains bounded. In the trivial case $H_1=0$, by contrast, one finds a purely classical $O(N)$ search. The model is therefore presented as an exponentially better nominal scaling in system size, but with explicit limitations: it requires fine engineering of the many-level energy spectrum $\{\eta_k\}$ and of system–bath couplings, plus cooling to low $T$ so that $\beta\Delta_N\gg 1$, and the resource cost of constructing a $2^n$-level system with log-spaced levels and all-to-all couplings may offset the algorithmic gain [2106.02703].

## 5. Hamiltonian phase oracles and resonance-band amplification in Ising systems

The Ising Hamiltonian phase-oracle primitive uses the native evolution of an $n$-spin Ising Hamiltonian directly:
$$
H=\sum_{i=1}^n h_i Z_i + 2\sum_{i<j}J_{ij}Z_iZ_j,
$$
so that on a basis state $\lvert s\rangle=\lvert s_1,\dots,s_n\rangle$,
$$
H|s\rangle=E_s|s\rangle,\qquad
E_s=\sum_i h_i s_i + 2\sum_{i<j}J_{ij}s_is_j.
$$
The oracle is
$$
U_T=\exp(-iTH),\qquad U_T|s\rangle=e^{-iTE_s}|s\rangle,
$$
and one iteration alternates it with the Grover diffusion operator:
$$
W_T=D_\xi U_T,\qquad D_\xi=2|\xi\rangle\langle \xi|-\openone,
$$
starting from the uniform superposition
$$
|\xi\rangle=\frac1{\sqrt{2^n}}\sum_{s\in\{\pm1\}^n}|s\rangle.
$$
Configurations with $TE_s\approx \pi$ are only partially marked at first, but repeated reflections build a narrow resonance band whose amplitudes grow as in Grover-type amplification. The exact dynamics is organized through the spectral measure
$$
f(E)=\frac1{2^n}\sum_s\delta(E-E_s),
$$
the characteristic-function samples
$$
g_m=\int e^{-imTE}f(E)\,dE,
$$
and generating functions $G(z)=\sum_{m=0}^\infty g_m z^m$ and $S(z)=\sum_{K=0}^\infty s_K z^K$, with the algebraic identity
$$
S(-z)=\frac{G(z)}{2G(z)-1}.
$$
The denominator
$$
C_r(\varphi)\equiv 2G(-re^{i\varphi})-1
$$
acts as the spectral response; its minima in modulus determine resonance center, width, and height [2606.03380].

For an annealed Gaussian density of states with variance
$$
\Sigma_n^2=2n(n-1)\sigma_J^2+n\sigma_h^2,
$$
the characteristic samples are
$$
g_m=\exp\!\left[-\frac12(m\Sigma_nT)^2\right].
$$
Targeting a prescribed energy $E_*$ uses
$$
T_*=\frac{\pi}{E_*},
$$
so that the principal resonance satisfies $T_*E_*=\pi$. In the high-density tail of the spectrum, the resonance contains $M_{\mathrm{peak}}$ levels and the peak iteration count satisfies
$$
K_{\mathrm{peak}}^{(G)}\sim \sqrt{\frac{2^n}{M_{\mathrm{peak}}}},
$$
recovering the Grover-type square-root speedup; the abstract states the same scaling as $\Theta(\sqrt{2^n/M})$ when the resonance contains $M$ configurations. In a true random-Ising ensemble, overlap-induced correlations shift and distort the peak. The resonance condition is shifted by a small $\varphi_0$ with mean zero and variance
$$
\langle \varphi_0^2\rangle = O(n^{-6}),
$$
so $\varphi_0=O(n^{-3})$ in r.m.s. Although this is negligible on the $O(n)$ spectral scale, it is large compared to the exponentially narrow resonance width $\Delta\varphi\sim \gamma=O(n^{3/2-\mu}2^{-n/2})$, so precise energy targeting requires correction. Two correction methods are given. Spectral symmetrization replaces $H$ by $\widetilde H=Z_a\otimes H$, making the relevant characteristic function real and restoring an exact zero at $\varphi=0$. Iterative calibration instead updates $T$ using measured output energies and is a contraction mapping; achieving final phase accuracy $O(2^{-n/2})$ requires only $k_{\mathrm{tune}}=O(n/\ln n)$ calibration steps. Each iterate costs $O(n^2)$ gates for a dense two-body Ising Hamiltonian, and the overall gate count in the Gaussian tail regime is
$$
O\!\bigl(n^2\sqrt{2^n/M}\bigr)
$$
[2606.03380].

## 6. Physical realization by exciton trapping and broader interpretive issues

A concrete physical implementation of energy-selective search is the Frenkel-exciton trapping model in a one-dimensional molecular crystal with one impurity site $w$. The exciton Hamiltonian is
$$
H_{\rm ex}=\sum_l E_e\,B_l^\dagger B_l+\sum_{l\ne m} M_{l,m}B_l^\dagger B_m,
$$
with long-range couplings
$$
M_{l,m}=J/|l-m|^p,\qquad p>1.
$$
The impurity is marked by
$$
H_w=\Delta_p\,B_w^\dagger B_w,
$$
and the oracle-like exciton–phonon interaction is
$$
H_{\rm exp}=B_w^\dagger B_w\sum_q \chi(q)\,(b_q^\dagger+b_{-q}),
$$
with
$$
\chi(q)^2=\hbar\omega(q)\,E_{LR},\qquad E_{LR}=\frac{D^2}{2Iv^2}.
$$
Because $H_{\rm exp}$ contains the projector $B_w^\dagger B_w$, it couples phonons only to the impurity and acts as the Grover oracle. The selectivity is explicitly energy-based: resonant exchange requires
$$
E_0(k)+\Delta_p=\hbar\omega(q),
$$
so only at site $w$ can the exciton emit or absorb an acoustic phonon at low temperature [1002.4056].

The search time is extracted from the damping rate of the Green’s-function exponent and has the form
$$
T_s\approx T_0+T_N,
$$
with
$$
T_0\equiv \frac{3\pi\,\hbar\,E_{LR}}{\Delta_p^3},
\qquad
T_N\approx A(p)'\,T_0\,N^{p-1}
$$
for $1<p<2$. For $p\to 1^+$, one obtains an $N$-independent coherent search time; at the marginal $p=2$, the $N$-term recovers a Grover-like scaling; and for dipole-type coupling $p=3$ one finds $T_s\approx T_0+O(N^{-2})\to T_0$ as $N\to\infty$. Using naphthalene parameters $\Delta_p\approx 50\,\mathrm{cm}^{-1}$, $\hbar\omega_D\approx 90\,\mathrm{cm}^{-1}$, $E_{LR}\approx 0.004\,\mathrm{eV}$, $B\approx 100\,\mathrm{cm}^{-1}$, with $N\sim 10^2$ and $p=3$, the model gives $T_0\approx 10^{-2}\,\mathrm{ps}$ and $T_N\approx 10$–$10^2\,\mathrm{ps}$, while competing decoherence times are much longer at sufficiently low temperature [1002.4056].

Taken together, these proposals motivate several interpretive cautions. The dissipative $O(\ln N)$ construction uses incoherent Markov dynamics in $N$ levels and no “oracle calls,” the Ising primitive uses a Hamiltonian phase oracle that marks configurations continuously by their phases and selects a finite resonance band rather than a preassigned marked set, and the non-Hermitian adiabatic construction exchanges a finite gap for norm loss $e^{-\delta t}$ [2106.02703][2606.03380][1208.4642]. This suggests that the apparent improvements over $\mathcal{O}(\sqrt{N})$ do not operate within a single resource model. The common pattern is not the replacement of one oracle by a faster oracle in the same formal setting, but the introduction of additional spectral, dissipative, or calibration structure: logarithmic level design and long-range bounded rates, fine control of the evolution time $T$, ancilla-based symmetrization with three-body terms, or impurity-specific exciton–phonon coupling. Within those assumptions, energy selectivity becomes the operative mechanism that directs amplitude, probability, or excitation transport toward the sought configuration.

Source: https://www.emergentmind.com/topics/energy-selective-quantum-search