Adiabatic Quantum Computation

Adiabatic quantum computation (AQC) represents a finite-dimensional closed-system computation by a Hamiltonian path. An initial Hamiltonian has a ground state or isolated ground band that can be prepared; a problem Hamiltonian has an accepted ground subspace whose measurement and classical decoder answer the task; and a schedule connects the two. A computational claim is valid only when the path, schedule, spectral isolation, error certificate, output success, and resource normalization are all declared. This page develops that record and the ideal polynomial-equivalence interface with the circuit model. It stops before generic adiabatic-theorem derivations, infinite-time Landau–Zener scattering, propagator-implementation algorithms, open-system annealing, device control, optimization evidence, fault tolerance, and computational-advantage claims.

Required background. Adiabatic Approximation supplies instantaneous eigenspaces, phases, nonadiabatic couplings, schedule dependence, leakage, and the theorem and Landau–Zener handoff. Circuit Model supplies registers, input–output semantics, measurements, uniform circuit families, and logical resource conventions.

Hamiltonian Paths as Quantum Computations

For an encoded instance $x$ , an AQC specification begins with a finite-dimensional Hilbert space and a Hermitian path

H_x(s), \qquad 0\leq s\leq1.

A monotone schedule $s(t)$ turns the path parameter into physical time:

i\hbar\frac{\partial}{\partial t}|\psi_x(t)\rangle =H_x(s(t))|\psi_x(t)\rangle, \qquad s(0)=0, \qquad s(T)=1.

The common linear interpolation

H_x(s)=(1-s)H_{{\rm i},x}+sH_{{\rm f},x}

is one path choice, not the definition of AQC. Nonlinear paths, catalysts, extra penalty terms, and local schedules can change gaps, derivative norms, control requirements, and runtime certificates without changing the endpoints.

The exact propagator is

U_x(T) = \mathcal T \exp\!\left[ -\frac{i}{\hbar} \int_0^T H_x(s(t))\,dt \right].

The time-ordering symbol may be dropped only when Hamiltonians at distinct times commute or when another exact reduction has been proved. Endpoint ground states alone do not determine this propagator, and the phrase “vary slowly” is not a runtime certificate.

A complete computational specification therefore joins four layers:

an instance-to-Hamiltonian encoding and a prepared initial state;
a path, schedule, and closed-system evolution model;
an isolated target subspace, measurement, and classical decoder; and
a theorem, exact solution, or converged numerical result tied to an explicit error and resource ledger.

These layers separate an abstract computation model from a physical proposal. A mathematical path can be well defined even when no device-native realization, compiled simulation, calibration model, or noise evidence is available.

The Ten-Field AQC Computation Record

Use the following record for a proposed path, theorem application, exact audit, numerical result, or implementation claim. Every field receives a value or an explicit reason that it is not applicable.

AQC task, instance family, and licensed claim — State the input family, promise, output task, fixed-instance or asymptotic status, and exactly what the record licenses.
Hilbert space, encoding, basis, and promises — Give register dimensions, computational basis, logical encoding, input promises, locality convention, and instance parameterization.
Initial Hamiltonian, prepared state, and preparation cost — Give $H_{\mathrm i}$ , its accepted initial subspace, the actual prepared state, the preparation method or assumption, and its separately priced cost.
Problem Hamiltonian, target subspace, and output decoder — Give $H_{\mathrm f}$ , the accepted ground-space projector, degeneracy, measurement, accepted outcomes, and classical decoder.
Interpolation path, schedule, and endpoint conventions — Give $H(s)$ , $s(t)$ , runtime, continuity or differentiability class, endpoint values, and the units or value of $\hbar$ .
Spectrum, external gap, and degeneracy convention — Give the relevant eigenvalues or bands, isolated projector, target-to-complement gap, $g_{\min}$ , and why internal degeneracy is or is not counted.
Runtime certificate, error budget, and regularity assumptions — Name the theorem, exact solution, or numerical evolution used; state its hypotheses, norm, constants, endpoint terms, error metric, and unsupported runtime claims.
Evolution model, controls, and implementation assumptions — State closed- or open-system status, available Hamiltonian terms, coefficient and schedule control, native-versus-simulated status, and every unavailable implementation capability.
Measurement, success metric, tolerance, and uncertainty — State the POVM or projector, success probability or channel metric, numerical tolerance, roundoff, sampling uncertainty or its N/A justification, and reproducible evaluator details.
Resource ledger, conclusion, and canonical handoff — Separate qubits, locality, term count, norm and energy range, coefficient and path precision, preparation, runtime, readout, repeats, logical conclusion, stopping point, and the next canonical owner.

The record is deliberately broader than a gap calculation. It prevents an endpoint encoding from being mistaken for an algorithm, an ideal runtime from being mistaken for wall-clock cost, and a fixed-instance computation from being mistaken for an efficient uniform family.

Problem Hamiltonians, Ground Spaces, and Decoders

The problem Hamiltonian must encode an output task, not merely have an interesting spectrum. For a classical cost function $C_x(z)$ on bit strings, a diagonal encoding can be written as

H_{{\rm f},x} = \sum_z C_x(z)|z\rangle\langle z|,

possibly with penalties enforcing constraints. A measurement in the computational basis then produces a candidate $z$ , but the record must still define which outcomes are accepted and how a classical decoder converts them into the reported answer. Constant energy offsets do not change the minimizers, while coefficient rescalings change the physical action and cannot be treated as free.

More general encodings need not be diagonal. Let $P_x(1)$ project onto the accepted final ground space. Success may mean projection into that whole subspace, recovery of one logical state encoded inside it, or a declared distribution after a further measurement. These are different claims. When the final ground space is degenerate, the page must say whether every vector in it is accepted, whether internal information is protected, and whether small splittings alter the decoder.

Penalty terms also require a scale contract. A coefficient large enough to separate invalid states for one fixed instance may grow exponentially across a family. An encoding is not efficient merely because it uses few formal terms; the term descriptions, locality, coefficient magnitudes, precision, and classical construction cost must all scale acceptably.

Initial-state promises deserve the same care. An easy-looking $H_{\mathrm i}$ does not make its ground state free if state preparation itself hides the hard part of the task. Conversely, a product-state ground state may be a reasonable abstract input while its physical initialization time remains outside the closed-system model.

Spectral Gaps and Target-Subspace Conventions

Let $P_x(s)$ be the spectral projector onto the instantaneous target state or isolated target band, and define

Q_x(s)=I-P_x(s).

The relevant external gap is the distance from the target spectrum to its complement:

g_x(s) = \operatorname{dist}\!\left( \operatorname{spec}\!\left( H_x(s)\big|_{\operatorname{Ran}P_x(s)} \right), \operatorname{spec}\!\left( H_x(s)\big|_{\operatorname{Ran}Q_x(s)} \right) \right),

with

g_{\min}=\min_{s\in[0,1]}g_x(s).

For a rank-one ground state this reduces to the ordinary ground-to-first-excited gap. For an intentionally degenerate accepted ground space, the sorted difference $E_1-E_0$ can vanish even while the accepted band is separated from every rejected state. In that case the external target-to-complement gap, not an internal zero splitting, enters the subspace statement.

Symmetry can further divide the Hilbert space into invariant sectors. A prepared state confined to an exactly preserved sector may encounter a larger sector-restricted gap than the full-space gap. That larger value is licensed only after proving that the complete path preserves the sector. A symmetry-breaking perturbation, control error, or environmental coupling can restore the smaller full-space gap as the safe certificate.

The minimum gap is necessary information, but it is not the whole adiabatic problem. The path derivatives, nonadiabatic matrix elements, endpoint behavior, regularity class, and theorem-specific constants can all matter. A positive gap for one finite matrix says nothing by itself about the scaling of $g_{\min}(n)$ for a uniform instance family.

Schedules, Runtime Certificates, and Adiabatic Error

For a nondegenerate two-level path, a useful local diagnostic is

\eta(t) = \frac{ \hbar|\dot s(t)| \left| \langle1(s)|H'(s)|0(s)\rangle \right| }{ \Delta(s)^2 }.

A small value can guide schedule design, but it is not a universal theorem. A rigorous runtime statement must name the theorem and expose the differentiability, endpoint, spectral-isolation, derivative-norm, constant, and error-metric assumptions it actually uses. There is no assumption-free law $T\sim g_{\min}^{-2}$ .

M(s)=\left|\langle1(s)|H'(s)|0(s)\rangle\right|,

then holding this diagnostic at a chosen value $\epsilon$ suggests the local schedule

\dot s = \frac{\epsilon\Delta(s)^2}{\hbar M(s)},

and diagnostic runtime

T = \frac{\hbar}{\epsilon} \int_0^1 \frac{M(s)}{\Delta(s)^2}\,ds.

This construction slows near difficult parts of the path and speeds up elsewhere. It still requires a separate theorem or direct evolution check before it becomes an error guarantee.

Path and schedule must remain distinct. A smooth curve $H(s)$ can be traversed with a discontinuous or poorly resolved $s(t)$ , while two schedules on the same curve can have different runtime and endpoint error. Any stated runtime also requires energy units: the dimensionless action is set by energy times time divided by $\hbar$ .

Closed-System Evolution and Output Success

For an initial state supported in $P_x(0)$ , an exact worst-case leakage measure is

\epsilon_{\mathrm{ad}}(T) = \left\| Q_x(1)U_x(T)P_x(0) \right\|.

This subspace norm is not automatically the computational failure probability. The latter depends on the actual prepared state, final measurement, accepted outcomes, and decoder. For initial density operator $\rho_{0,x}$ , accepted classical outcomes $\mathcal A_x$ , and POVM elements $M_y$ ,

p_{\mathrm{succ}}(x) = \sum_{y\in\mathcal A_x} \operatorname{Tr}\!\left[ M_yU_x(T)\rho_{0,x}U_x(T)^\dagger \right].

If every state in the target band is accepted and the initial state lies in the corresponding initial band, leakage controls failure. If only one logical vector or one decoded subset is accepted, population inside the band may still give the wrong answer.

Global phase drops out of an isolated projective measurement, but relative phases inside a degenerate code space or a coherent output register may matter. A record must therefore state whether phase is discarded, tracked, or verified interferometrically.

For independent repetitions with per-run success $0<p_{\mathrm{succ}}<1$ , the number needed to reach confidence $\eta$ is

R_\eta = \left\lceil \frac{\ln(1-\eta)}{\ln(1-p_{\mathrm{succ}})} \right\rceil.

Repeats multiply preparation, coherent evolution, readout, and reset costs. Parallel throughput can be claimed only when parallel resources and scheduling are included.

Circuit Equivalence and History-State Encodings

A circuit can be embedded into the ground state of a Hamiltonian by adding a clock register. For a circuit $U_L\cdots U_1$ and input $|\psi_{\mathrm{in}}\rangle$ , define

|\eta\rangle = \frac{1}{\sqrt{L+1}} \sum_{\ell=0}^{L} U_\ell\cdots U_1|\psi_{\mathrm{in}}\rangle \otimes|\ell\rangle_c, \qquad U_0=I.

The propagation kernel is

\begin{aligned} H_{\mathrm{prop}} =\frac12\sum_{\ell=1}^{L} \bigl[ &I\otimes \left( |\ell\rangle\langle\ell| +|\ell-1\rangle\langle\ell-1| \right)\\ &-U_\ell\otimes|\ell\rangle\langle\ell-1| -U_\ell^\dagger\otimes|\ell-1\rangle\langle\ell| \bigr]. \end{aligned}

Each summand compares adjacent clock slices and is minimized by consistent propagation. A complete construction also needs an input penalty, a valid clock subspace, an initial Hamiltonian whose ground state is preparable, a gapped interpolation, and an output decoder. Measuring the clock of the unpadded history state returns the final circuit time with probability $1/(L+1)$ ; appending identity gates increases the fraction of clock positions that carry the final data without changing the circuit output.

The polynomial equivalence between ideal AQC and the circuit model is a family-level statement. It requires a uniform, efficiently specified, bounded-local Hamiltonian construction together with controlled norms, coefficient precision, a sufficiently large gap, and polynomial overhead. It does not provide an equal-runtime compiler, make the interactions device native, or equate the two models’ energy, precision, noise, control, and fault-tolerance costs. Universal Gate Sets retains gate-synthesis criteria, while Quantum Complexity Classes retains BQP, QMA, Local Hamiltonian completeness, and promise-problem claims.

Universality, Locality, and Resource Accounting

An asymptotic AQC claim concerns a uniform family

x\longmapsto \bigl(H_x(s),\rho_{0,x},s_x(t),\{M_y\},\mathcal A_x\bigr),

not an isolated finite matrix. A classical description procedure must generate the register sizes, terms, coefficients, schedule, measurement, and decoder using resources polynomial in the instance length. The ledger should expose at least:

data, ancilla, and clock width;
interaction locality, geometry, and term count;
coefficient range, required precision, $\max_s\|H_x(s)\|$ , and spectral range;
path regularity, schedule precision, and coherent runtime;
initial-state preparation, gap and error certification, measurement, decoding, and repeats; and
any unavailable native controls, simulation access, noise model, fault-tolerant construction, or wall-clock currency.

Energy normalization is essential. For

\widetilde H(s)=aH(s)+b(s)I, \qquad a>0,

the eigenprojectors are unchanged and the gaps scale by $a$ . To reproduce an original schedule $s(t)$ on $0\leq t\leq T$ in duration $T/a$ , use

\widetilde s(t')=s(at'), \qquad 0\leq t'\leq\frac{T}{a}.

Evolution then follows the same ray, with scalar phase

\exp\!\left[ -\frac{i}{\hbar} \int_0^{T/a} b\!\left(\widetilde s(t')\right)\,dt' \right].

The claim is false for a general unchanged physical-time schedule. Multiplying energies by $a$ is not a free speedup: the Hamiltonian norm, spectral range, coefficient precision, control amplitude, and compressed schedule must be priced.

Noise Boundaries, Verification, and Canonical Handoffs

Three kinds of certificate must not be relabeled as one another:

a theorem bound follows from stated mathematical hypotheses and bounds an explicit error metric;
an exact solution proves one declared model and convention; and
a numerical evolution approximates one finite instance with discretization, roundoff, convergence, and reproducibility evidence.

For numerical propagation, record the basis order, arithmetic, runtime version, integrator or exponential method, step count, norm defect, convergence comparison, metric, and tolerance. Sampling uncertainty is N/A only when no measurement samples were generated. A small integration residual does not certify that the encoded Hamiltonian represents the intended application or that a physical device implements it.

Time-Dependent Hamiltonians owns generic propagators and time ordering. The Adiabatic Theorem owns rigorous hypotheses and bounds, and Landau–Zener Transition owns the exact infinite linear sweep. Hamiltonian Simulation owns algorithms that implement evolution from a declared access model; access to a path does not automatically supply a block encoding, controlled evolution, or device-native generator. Quantum Phase Estimation owns spectral readout and its controlled-power costs.

Analog Quantum Simulation owns target-to-device correspondence, calibration, leakage, and observable validation. Optimization Case Studies owns objective evidence, comparators, time-to-solution, and speedup vocabulary, while Algorithmic Benchmarking owns end-to-end workload and advantage ledgers. Quantum Annealing owns the device-agnostic finite-time annealing record across declared closed, open, thermal, nonadiabatic, paused, reversed, and heuristic regimes; this page retains ideal closed-system computation paths and licensed adiabatic certificates.

Quantum Algorithms for Optimization compares the closed-system adiabatic route with circuit, walk-and-tree, variational, annealing, and convex routes after the objective and output are fixed; this page retains Hamiltonian-path semantics, gap and schedule certificates, and model equivalence.

Noise in Quantum Information and the Lindblad–GKSL Equation own noise and open-system dynamics. Resource Estimation owns physical and fault-tolerant overhead, Control, Readout, and Calibration owns device evidence, and Claims, Hype, and Evidence Standards owns advantage language. Intrinsic suppression in one closed model is not fault tolerance, and ideal polynomial model equivalence is not a practical speedup claim.

Worked Audit: A One-Qubit Linear Interpolation

This audit follows one ideal qubit from an $X$ -ground state to a $Z$ -ground state. It declares $\hbar=1$ and preserves the identity offsets in the Hamiltonian, so the reported amplitudes have a fixed phase convention rather than only a projective meaning.

AQC task, instance family, and licensed claim — Audit one fixed ideal qubit under a linear interpolation for runtime $T=8$ . The record licenses the exact spectral data and independently reproduced endpoint success. It licenses no asymptotic runtime theorem, implementation, or speedup.
Hilbert space, encoding, basis, and promises — Use one qubit in computational order $(|0\rangle,|1\rangle)$ . The accepted output is bit zero. There is no hidden instance ensemble, input distribution, or promise.
Initial Hamiltonian, prepared state, and preparation cost — Use
$H_B=\frac{I-X}{2},$
whose unique ground state is $|+\rangle$ . Exact preparation of $|+\rangle$ is assumed and priced separately from the evolution; physical initialization time is N/A in this abstract record.
Problem Hamiltonian, target subspace, and output decoder — Use
$H_P=\frac{I-Z}{2},$
whose unique ground state is $|0\rangle$ . The accepted projector is $|0\rangle\langle0|$ ; measure in the computational basis and decode outcome zero as success.
Interpolation path, schedule, and endpoint conventions — Use
$H(s)=(1-s)H_B+sH_P, \qquad s=\frac{t}{8}, \qquad 0\leq t\leq8.$
The path and schedule are smooth on the closed interval, endpoints are included, and energy and time are dimensionless with $\hbar=1$ .
Spectrum, external gap, and degeneracy convention — The eigenvalues are
$E_\pm(s) = \frac{1\pm\Delta(s)}{2}, \qquad \Delta(s)=\sqrt{(1-s)^2+s^2}.$
The target is rank one, so the external gap is $\Delta(s)$ . It reaches
$\Delta_{\min}=\frac{1}{\sqrt2}$
at $s=1/2$ ; there is no internal degeneracy or symmetry-sector qualification.
Runtime certificate, error budget, and regularity assumptions — Direct finite-time Schrödinger evolution, not an adiabatic theorem, is the certificate. The off-diagonal derivative coupling is
$M(s) = \left|\langle1(s)|H'(s)|0(s)\rangle\right| = \frac{1}{2\Delta(s)},$
so the maximum local diagnostic for $\dot s=1/8$ is
$\max_s\frac{|\dot s|M(s)}{\Delta(s)^2} = \frac{\sqrt2}{8} =0.1767766953.$
This value is not used as a theorem. A 100,000-step RK4 evolution and an independent 100,000-step product of exact exponentials evaluated at interval midpoints agree in every reported amplitude component and probability within $10^{-10}$ .
Evolution model, controls, and implementation assumptions — Use an ideal closed qubit with continuously available $X$ and $Z$ coefficients. Native controls, bandwidth, calibration, compiled gates, noise, thermalization, fault tolerance, and hardware timing are N/A and are not inferred from the mathematical path.
Measurement, success metric, tolerance, and uncertainty — In the declared basis, the RK4 endpoint is
$\begin{aligned} |\psi(8)\rangle \approx{}&(0.804636102344-0.581531139378i)|0\rangle\\ &+(-0.078388950073+0.090760394695i)|1\rangle. \end{aligned}$
Therefore
$p_0\approx0.9856177233, \qquad p_1\approx0.0143822767.$
Node.js 26.4.0 with binary64 arithmetic gives a maximum cross-method amplitude-component difference $6.15\times10^{-11}$ and probability difference $5.1\times10^{-12}$ . The RK4 Euclidean norm defect is below $10^{-15}$ . For the midpoint product, the squared-norm defect is $|\langle\psi|\psi\rangle-1|\approx8.8702\times10^{-12}$ , corresponding to Euclidean norm defect $|\|\psi\|-1|\approx4.4351\times10^{-12}$ ; both arise from accumulated roundoff. Sampling uncertainty is N/A because no shots were drawn.
Resource ledger, conclusion, and canonical handoff — Count one qubit, locality one, two nonidentity Pauli terms plus an identity offset, maximum Hamiltonian norm one, spectral range at most one, exact mathematical coefficients, one assumed $|+\rangle$ preparation, one ideal evolution of duration eight, and one computational-basis readout. No repeat-confidence target is requested, so repeat cost is N/A. This fixed-instance audit establishes the stated endpoint probability only. Adiabatic Theorem owns general error scaling; simulation, control, hardware, noise, and benchmarking owners receive implementation claims.

The calculation illustrates why a gap diagnostic and a direct success calculation should both appear in the record. The first describes local spectral difficulty; the second answers the declared finite-time computational question.

Worked Audit: A One-Gate History Hamiltonian

The second audit isolates the history-state kernel for the one-gate circuit $U=X$ . It verifies the endpoint and full interpolation spectrum exactly, but deliberately supplies no schedule or runtime.

AQC task, instance family, and licensed claim — Audit the endpoint and linear-path spectrum for the history Hamiltonian of the one-gate circuit $U=X$ . The record licenses the exact history-state kernel and conditional decoder, not a general universality, preparation, runtime, or speedup claim.
Hilbert space, encoding, basis, and promises — Use data qubit $d$ followed by clock qubit $c$ in order
$(|00\rangle,|01\rangle,|10\rangle,|11\rangle).$
The promised input is data zero at clock zero. The model has no further instance parameter.
Initial Hamiltonian, prepared state, and preparation cost — Use
$H_{\mathrm{start}} =I-|00\rangle\langle00|,$
with exactly prepared unique ground state $|00\rangle$ . Its abstract product-state preparation is assumed; physical preparation time is N/A.
Problem Hamiltonian, target subspace, and output decoder — Use
$H_{\mathrm{hist}} = \frac12(I-X_dX_c) +|10\rangle\langle10|.$
Its unique ground state is the normalized history state
$|\eta\rangle = \frac{|00\rangle+|11\rangle}{\sqrt2}.$
Measure the clock. Conditioned on clock outcome one, read data outcome one; this is the output of applying $X$ to input zero.
Interpolation path, schedule, and endpoint conventions — Audit the formal linear path
$H(s) =(1-s)H_{\mathrm{start}}+sH_{\mathrm{hist}}, \qquad 0\leq s\leq1,$
with $\hbar=1$ . No physical-time schedule $s(t)$ , runtime, or endpoint switching prescription is supplied, so those quantities are N/A.
Spectrum, external gap, and degeneracy convention — In the declared basis,
$H_{\mathrm{hist}} = \begin{pmatrix} 1/2&0&0&-1/2\\ 0&1/2&-1/2&0\\ 0&-1/2&3/2&0\\ -1/2&0&0&1/2 \end{pmatrix},$
with spectrum
$\left\{ 0, 1-\frac{1}{\sqrt2}, 1, 1+\frac{1}{\sqrt2} \right\}.$
Along the path, the two invariant blocks have eigenvalues
$\lambda_{A,\pm}(s) = \frac{1\pm\sqrt{(1-s)^2+s^2}}{2}, \qquad \lambda_{B,\pm}(s) = 1\pm\frac{s}{\sqrt2}.$
The ground state is nondegenerate. The full-Hilbert-space external minimum gap is
$g_{\min}^{\mathrm{full}} = 1-\frac{1}{\sqrt2} =0.2928932188$
at $s=1$ . The path commutes with $Z_dZ_c$ , and $|00\rangle$ remains in the even-parity $A$ block. In that exactly invariant block the gap is $\sqrt{(1-s)^2+s^2}$ , with minimum $1/\sqrt2$ at $s=1/2$ . This larger sector gap is licensed only by exact symmetry; a symmetry-breaking perturbation restores the full-space gap as the safe value.
Runtime certificate, error budget, and regularity assumptions — Exact block diagonalization certifies the spectrum and endpoint ground state. Runtime, leakage, and adiabatic error are N/A because no schedule or adiabatic theorem is invoked. The audit does not infer finite-time success from its endpoint spectrum.
Evolution model, controls, and implementation assumptions — Treat both two-qubit Hamiltonians as abstract closed-system operators. Native $X_dX_c$ , input-penalty implementation, coefficient precision, continuous path control, compiled simulation, noise, and hardware cost are unavailable and therefore not licensed.
Measurement, success metric, tolerance, and uncertainty — Exact substitution gives
$H_{\mathrm{hist}}|\eta\rangle=0.$
At the ideal endpoint, clock one occurs with probability $1/2$ , and conditioned on that outcome data one occurs with probability one. A symbolic characteristic-polynomial calculation and Node.js 26.4.0 binary64 diagonalization reproduce every eigenvalue within $10^{-12}$ . No random samples are generated, so sampling uncertainty is N/A.
Resource ledger, conclusion, and canonical handoff — Count two qubits, locality two, one clock transition, four nonidentity Pauli terms across the path $(X_dX_c,Z_d,Z_c,Z_dZ_c)$ plus identity offsets, maximum norm and spectral range $1+1/\sqrt2$ , one assumed $|00\rangle$ preparation, and one conditional clock-and-data readout. Coherent runtime and schedule precision are N/A. The endpoint decoder has raw clock acceptance $1/2$ ; no repeat-confidence target is requested. This record verifies one history kernel, not an efficient many-gate construction, runtime theorem, physical implementation, or speedup. Circuit Model and Quantum Complexity Classes own the broader model and complexity statements.

The full-space and sector-restricted gaps answer different questions. Exact parity makes the larger sector value relevant to this prepared ideal path; robustness to parity-breaking terms must use the smaller full-space value.

Common Failure Modes and Ownership Boundaries

Calling a path adiabatic by inspection. Smoothness or a visually slow schedule does not establish small error. State a theorem with its hypotheses, an exact solution, or a converged numerical evolution tied to the claimed metric.

Reporting only the minimum gap. A gap without derivative norms, couplings, schedule, endpoints, regularity, theorem constants, and energy normalization is not a runtime certificate. The gap-squared ratio is a diagnostic unless a specific theorem says otherwise.

Using the wrong gap for a degenerate target. The sorted value $E_1-E_0$ vanishes inside an intentionally degenerate accepted ground space. Use the distance from the accepted spectral band to its complement, and state whether internal splitting matters to the decoder.

Using a sector gap without a symmetry proof. A sector-restricted gap is relevant only when the full path and prepared state preserve that invariant sector. Report the full-space gap as the safe value against symmetry-breaking perturbations.

Turning a finite example into an asymptotic algorithm. A positive gap for one fixed matrix does not establish an inverse-polynomial family gap, uniform construction, scalable coefficient precision, or efficient certification.

Hiding speed in the energy scale. Multiplying $H$ by a large constant can shorten ideal coherent time only by increasing norms and control amplitudes and compressing the schedule. Comparisons require a common energy, coefficient, and control normalization.

Confusing ground overlap with decoded success. The accepted projector, final POVM, classical decoder, conditional clock outcome, and repeat rule must be explicit. Low expected energy can coexist with appreciable failure probability.

Substituting the Landau–Zener formula. A finite path with different endpoints and time dependence is not the exact unbounded linear sweep. Use the canonical transition model only when its assumptions and conventions match.

Treating path access as propagator access. A symbolic $H(s)$ does not automatically provide native evolution, local exponentials, sparse oracles, block encodings, controlled powers, or calibrated hardware controls. Those are separate simulation and implementation interfaces.

Conflating AQC with quantum annealing. This page is closed-system and computational-path specific. Thermalization, dissipation, freeze-out, heuristic schedules, stoquastic restrictions, and device evidence belong to Quantum Annealing and open-system pages.

Upgrading robustness to fault tolerance. Suppression caused by one gap or encoding is not a threshold theorem, encoded fault-tolerant construction, or recovery protocol. Resource Estimation and specialist fault-tolerance pages own those claims.

Overreading circuit equivalence. Polynomial equivalence of ideal uniform models does not imply equal runtime, term locality, coefficient precision, energy, noise behavior, hardware feasibility, or speedup over a classical comparator.

Omitting full cost. Preparation, gap certification, coefficient precision, coherent runtime, readout, decoding, resets, rejected runs, and repeats are separate currencies. A statement about only qubit count or path duration is incomplete.

Exercises

Encode a Two-Bit Cost Function

For bits $x_1,x_2\in\{0,1\}$ , encode

C(x_1,x_2)=2x_1+x_2+x_1x_2

as a diagonal two-qubit problem Hamiltonian using $x_j=(1-Z_j)/2$ . Verify every computational-basis energy and state exactly what the calculation does not establish.

Solution

Substitution gives

\begin{aligned} H_P ={}&2\frac{I-Z_1}{2} +\frac{I-Z_2}{2} +\frac{(I-Z_1)(I-Z_2)}{4}\\ ={}& \frac74I -\frac54Z_1 -\frac34Z_2 +\frac14Z_1Z_2. \end{aligned}

In computational order:

$|00\rangle$ has $(Z_1,Z_2)=(+1,+1)$ and energy $0$ ;
$|01\rangle$ has $(+1,-1)$ and energy $1$ ;
$|10\rangle$ has $(-1,+1)$ and energy $2$ ; and
$|11\rangle$ has $(-1,-1)$ and energy $4$ .

Thus the ordered spectrum is $\{0,1,2,4\}$ , matching the four classical costs. The calculation establishes an endpoint encoding only. It supplies no initial Hamiltonian, path, gap, runtime, output-confidence analysis, or speedup.

Audit Hamiltonian Rescaling

Let

\widetilde H(s)=aH(s)+b(s)I, \qquad a>0.

Given an original schedule $s(t)$ on $0\leq t\leq T$ , find the schedule that reproduces the same final ray in time $T/a$ , derive the extra phase, and identify the resources changed by the transformation.

Solution

Set $t'=t/a$ and compress the schedule as

\widetilde s(t')=s(at'), \qquad 0\leq t'\leq\frac{T}{a}.

If $|\psi(t)\rangle$ solves the original equation, then $|\psi(at')\rangle$ obeys

i\hbar\frac{d}{dt'}|\psi(at')\rangle = aH(s(at'))|\psi(at')\rangle.

The scalar term commutes with every operator and contributes only

\exp\!\left[ -\frac{i}{\hbar} \int_0^{T/a} b\!\left(\widetilde s(t')\right)\,dt' \right].

Therefore the final ray agrees with the original one. The statement is false for a general unchanged physical-time schedule, because the two evolutions then visit different path points at corresponding times.

The projectors are unchanged, but all gaps, the Hamiltonian norm, spectral range, and nonidentity coefficients scale by $a$ . The compressed schedule requires proportionally faster control. These energy, coefficient-precision, bandwidth, and timing resources must be counted; $b(s)$ is phase-only for an isolated evolution but may matter under controlled evolution or phase-sensitive comparison.

Derive the Linear One-Qubit Gap

For the Hamiltonian in the first worked audit, derive the eigenvalue gap, its minimum, the derivative coupling $|\langle1|H'|0\rangle|$ , and the maximum local diagnostic at $T=8$ .

Solution

The Hamiltonian is

H(s) = \frac12 \left[ I-(1-s)X-sZ \right].

Its nonidentity Bloch vector has length

\Delta(s)=\sqrt{(1-s)^2+s^2},

E_\pm(s)=\frac{1\pm\Delta(s)}{2}.

Minimizing $\Delta(s)^2=1-2s+2s^2$ gives $s=1/2$ and

\Delta_{\min}=\frac{1}{\sqrt2}.

For a two-level Bloch Hamiltonian, the off-diagonal derivative component is the part of the derivative perpendicular to the instantaneous Bloch vector. Here this gives

\left|\langle1(s)|H'(s)|0(s)\rangle\right| = \frac{1}{2\Delta(s)}.

With $\dot s=1/8$ , the diagnostic is

\frac{|\dot s|}{2\Delta(s)^3}.

It is largest where the gap is smallest, so

\eta_{\max} = \frac18 \frac{1}{2(1/\sqrt2)^3} = \frac{\sqrt2}{8}.

This is a local diagnostic, not a theorem-level error bound.

Compare Linear and Local Schedules

For the same one-qubit path, evaluate

\int_0^1 \frac{|\langle1|H'|0\rangle|}{\Delta^2}\,ds

and compare the diagnostic runtimes for a constant-local-diagnostic schedule and a linear schedule at the same maximum diagnostic $\epsilon$ .

Solution

Because $|\langle1|H'|0\rangle|=1/(2\Delta)$ ,

\int_0^1 \frac{|\langle1|H'|0\rangle|}{\Delta^2}\,ds = \int_0^1\frac{ds}{2\Delta^3}.

Let $u=2s-1$ . Then $ds=du/2$ and $\Delta^2=(1+u^2)/2$ , so

\begin{aligned} \int_0^1\frac{ds}{2\Delta^3} &= \frac{1}{\sqrt2} \int_{-1}^{1} \frac{du}{(1+u^2)^{3/2}}\\ &= \frac{1}{\sqrt2} \left[ \frac{u}{\sqrt{1+u^2}} \right]_{-1}^{1} =1. \end{aligned}

Holding the local diagnostic at $\epsilon$ gives

\dot s = \frac{\epsilon\Delta^2}{M} =2\epsilon\Delta^3,

and therefore

T_{\mathrm{local}} = \int_0^1\frac{ds}{2\epsilon\Delta^3} = \frac{1}{\epsilon}.

For a linear schedule, $\dot s=1/T$ and the maximum diagnostic is $\sqrt2/T$ , so

T_{\mathrm{linear}}=\frac{\sqrt2}{\epsilon}.

This compares two uses of the same diagnostic. Neither equality is a general theorem relating runtime to actual adiabatic error.

Use an External Gap for a Degenerate Target

Consider

H_P=\frac{I+Z_1Z_2}{2}.

Identify its ground space, compare the sorted $E_1-E_0$ value with the external target-to-complement gap, and explain which one belongs in a subspace adiabatic record.

Solution

$Z_1Z_2$ has eigenvalue $+1$ on even-parity states $|00\rangle,|11\rangle$ and $-1$ on odd-parity states $|01\rangle,|10\rangle$ . Hence

H_P|00\rangle=|00\rangle, \qquad H_P|11\rangle=|11\rangle,

while

H_P|01\rangle=0, \qquad H_P|10\rangle=0.

The accepted ground projector is

P = |01\rangle\langle01| +|10\rangle\langle10|.

The ordered spectrum is $\{0,0,1,1\}$ . Consequently the naive sorted gap is

E_1-E_0=0,

because it measures an internal degeneracy within the accepted band. The distance from that band to the rejected even-parity space is instead

g_{\mathrm{external}}=1.

The external value belongs in a theorem or leakage statement that transports the whole accepted subspace. If a task distinguishes coherent vectors inside the odd-parity space, internal splitting and logical transport require additional analysis.

Pad a History State

A circuit contains $L=3$ nonidentity gates and is followed by $r=5$ identity gates. For the uniform history state, find the probability that a clock measurement lands at a time whose data register already contains the final circuit output. Separate this clock probability from the circuit’s own success probability.

Solution

The padded clock labels are

\ell=0,1,\ldots,L+r,

so there are

L+r+1=3+5+1=9

equally weighted clock positions. The final circuit output first appears after gate $L$ , at $\ell=3$ , and persists through $\ell=8$ . There are therefore

r+1=6

accepted clock positions, giving

p_{\mathrm{clock\ output}} = \frac{r+1}{L+r+1} = \frac69 = \frac23.

This is only the probability of reading a clock position that carries the circuit’s final data state. If the circuit itself succeeds with probability $p_{\mathrm{circuit}}$ , the joint accepted probability is $(2/3)p_{\mathrm{circuit}}$ unless a different decoder or correlation changes the rule.

Count Repeated Runs and Wall-Clock Cost

An AQC record has independent per-run success $p=0.6$ . How many serial attempts are required for success confidence at least $\eta=0.99$ ? Each attempt uses $2\,\mu\mathrm{s}$ for preparation, $8\,\mu\mathrm{s}$ for coherent evolution, and $3\,\mu\mathrm{s}$ for readout and reset. Report both coherent-evolution time and total serial time.

Solution

The required attempt count is

R_{0.99} = \left\lceil \frac{\ln(0.01)}{\ln(0.4)} \right\rceil =6.

Indeed, five failures have probability

0.4^5=0.01024>0.01,

whereas six failures have probability

0.4^6=0.004096<0.01.

The coherent-evolution budget is

6(8\,\mu\mathrm{s})=48\,\mu\mathrm{s}.

The full serial time is

6(2+8+3)\,\mu\mathrm{s} =78\,\mu\mathrm{s}.

The $48\,\mu\mathrm{s}$ value omits preparation and readout; it must not be reported as wall-clock time. Parallel execution or throughput gains are N/A because the problem licenses only serial independent attempts.

Complete a Ten-Field Rotating-Path Record

Consider the one-qubit path

H(s) = -\frac12 \left[ \cos\!\left(\frac{\pi s}{2}\right)Z +\sin\!\left(\frac{\pi s}{2}\right)X \right], \qquad s=\frac{t}{T},

with $T=\pi\sqrt3/2$ and $\hbar=1$ . Starting in $|0\rangle$ , complete the ten-field record, solve the evolution exactly in a rotating frame, and find the final target probability for $|+\rangle$ .

Solution

AQC task, instance family, and licensed claim — Audit one ideal constant-gap rotating one-qubit path at
$T=\frac{\pi\sqrt3}{2}=2.720699046\ldots.$
The record licenses its exact endpoint state, success probability, and abstract Pauli ledger only. It is one fixed path, not an instance family or speedup claim.
Hilbert space, encoding, basis, and promises — Use one qubit in computational order $(|0\rangle,|1\rangle)$ , with no input ensemble, hidden promise, ancilla, or encoded subspace.
Initial Hamiltonian, prepared state, and preparation cost — Use
$H_{\mathrm i}=-\frac{Z}{2},$
whose unique ground state is $|0\rangle$ . Exact preparation is assumed and separate; physical preparation cost is otherwise N/A.
Problem Hamiltonian, target subspace, and output decoder — Use
$H_{\mathrm f}=-\frac{X}{2},$
whose unique ground state is $|+\rangle$ . The success measurement is the projector $|+\rangle\langle+|$ ; no further classical decoding is needed.
Interpolation path, schedule, and endpoint conventions — With $\hbar=1$ , use the stated smooth path and linear schedule $s=t/T$ on the inclusive interval $0\leq t\leq T$ . The path rotates the Hamiltonian axis through angle
$\theta(t)=\frac{\pi t}{2T}$
from $Z$ to $X$ .
Spectrum, external gap, and degeneracy convention — The Bloch vector has constant length one. The eigenvalues are $-1/2$ and $+1/2$ for every $s$ , so the rank-one external gap is exactly
$\Delta(s)=g_{\min}=1.$
There is no degeneracy or sector restriction.
Runtime certificate, error budget, and regularity assumptions — Let
$R_y(\theta)=e^{-i\theta Y/2}.$
Then
$H(t)=R_y(\theta)\left(-\frac Z2\right)R_y(\theta)^\dagger.$
Writing $|\psi(t)\rangle=R_y(\theta)|\phi(t)\rangle$ gives the constant rotating-frame Hamiltonian
$H_{\mathrm{rot}} = -\frac12 \left( Z+\dot\theta Y \right), \qquad \dot\theta=\frac{1}{\sqrt3}.$
Its Bloch-vector length is $2/\sqrt3$ , and $T(2/\sqrt3)/2=\pi/2$ , so exact exponentiation determines the endpoint. This is an exact fixed-runtime solution, not an adiabatic theorem. Since $|\langle1|H'(s)|0\rangle|=\pi/4$ , the local diagnostic is the constant
$\frac{1}{T}\frac{\pi}{4} = \frac{1}{2\sqrt3}.$
Evolution model, controls, and implementation assumptions — Use an ideal closed qubit with continuously commanded $X$ and $Z$ coefficients. Native realization, coefficient error, bandwidth, calibration, noise, thermalization, compiled simulation, fault tolerance, and hardware timing are N/A.
Measurement, success metric, tolerance, and uncertainty — Rotating back to the laboratory basis gives
$|\psi(T)\rangle = \frac{ (1+i\sqrt3)|0\rangle +(-1+i\sqrt3)|1\rangle }{2\sqrt2}.$
Therefore
$|\langle+|\psi(T)\rangle|^2 = \frac34, \qquad |\langle-|\psi(T)\rangle|^2 = \frac14.$
Exact algebra is normative. A 100,000-step midpoint product reproduces every displayed amplitude component within $2.8\times10^{-11}$ . Sampling uncertainty is N/A because no shots are drawn.
Resource ledger, conclusion, and canonical handoff — Count one qubit, locality one, two nonidentity Pauli coefficients, norm $1/2$ , spectral range and constant gap one, exact mathematical coefficients, one assumed $|0\rangle$ preparation, one ideal evolution of duration $\pi\sqrt3/2$ , and one $|+\rangle$ projector readout. No repeat-confidence target is requested. This record proves no family scaling, speedup, native implementation, robustness, or hardware claim. Adiabatic Approximation owns the generic rotating-Hamiltonian analysis; control, noise, and benchmarking owners receive physical claims.

References

D. Aharonov, W. van Dam, J. Kempe, Z. Landau, S. Lloyd, and O. Regev, “Adiabatic quantum computation is equivalent to standard quantum computation,” SIAM Journal on Computing 37(1), 166–194 (2007), doi:10.1137/S0097539705447323.
T. Albash and D. A. Lidar, “Adiabatic quantum computation,” Reviews of Modern Physics 90, 015002 (2018), doi:10.1103/RevModPhys.90.015002.
A. M. Childs, E. Farhi, and J. Preskill, “Robustness of adiabatic quantum computation,” Physical Review A 65, 012322 (2001), doi:10.1103/PhysRevA.65.012322.
E. Farhi, J. Goldstone, S. Gutmann, and M. Sipser, “Quantum computation by adiabatic evolution,” arXiv:quant-ph/0001106 (2000), arXiv:quant-ph/0001106.
E. Farhi, J. Goldstone, S. Gutmann, J. Lapan, A. Lundgren, and D. Preda, “A quantum adiabatic evolution algorithm applied to random instances of an NP-complete problem,” Science 292, 472–475 (2001), doi:10.1126/science.1057726.
S. Jansen, M.-B. Ruskai, and R. Seiler, “Bounds for the adiabatic approximation with applications to quantum computation,” Journal of Mathematical Physics 48, 102111 (2007), doi:10.1063/1.2798382.
D. A. Lidar, A. T. Rezakhani, and A. Hamma, “Adiabatic approximation with exponential accuracy for many-body systems and quantum computation,” Journal of Mathematical Physics 50, 102106 (2009), doi:10.1063/1.3236685.
A. Mizel, D. A. Lidar, and M. Mitchell, “Simple proof of equivalence between adiabatic quantum computation and the circuit model,” Physical Review Letters 99, 070502 (2007), doi:10.1103/PhysRevLett.99.070502; erratum, Physical Review Letters 127, 139901 (2021), doi:10.1103/PhysRevLett.127.139901.
B. W. Reichardt, “The quantum adiabatic optimization algorithm and local minima,” in Proceedings of the 36th Annual ACM Symposium on Theory of Computing, 502–510 (2004), doi:10.1145/1007352.1007428.
J. Roland and N. J. Cerf, “Quantum search by local adiabatic evolution,” Physical Review A 65, 042308 (2002), doi:10.1103/PhysRevA.65.042308.
W. van Dam, M. Mosca, and U. Vazirani, “How powerful is adiabatic quantum computation?” in Proceedings of the 42nd IEEE Symposium on Foundations of Computer Science, 279–287 (2001), doi:10.1109/SFCS.2001.959902.
K. C. Young, M. Sarovar, and R. Blume-Kohout, “Error suppression and error correction in adiabatic quantum computation: techniques and challenges,” Physical Review X 3, 041013 (2013), doi:10.1103/PhysRevX.3.041013.