Glossary

372 terms, drawn from every chapter's key_terms front matter. Chapter numbers point to where a term is introduced or measured; where the book measured something about a term, the entry says what.

Terms appearing in several chapters usually mean the later chapter complicated the earlier one. Follow the chain — that is often where the interesting result is.

A

abort threshold (38) — The QBER above which no secret key survives privacy amplification. Derived by bisection at 0.110028, not quoted as "11%"; the extra digits change detection counts.

active space (36) — A truncation of a molecular Hamiltonian to a window of orbitals, freezing the rest. Makes chemistry fit a device, and introduced an error 9,870,104× larger than the VQE error it contained.

adiabatic limit (37) — The $p \to \infty$ regime where QAOA becomes adiabatic evolution and is exactly correct. Guaranteed in a limit nobody can reach.

adjoint (15) — Q#'s automatically-derived inverse of a unitary operation.

Aer (7) — Qiskit's high-performance simulator backend.

Aer simulator (2) — The local simulator used throughout the book; Chapter 39 measured it at the same order of magnitude as hardware execution time.

ALAPScheduleAnalysis (31) — The pass that assigns as-late-as-possible start times, and the way to read a circuit's real duration in Qiskit 2.x.

all-to-all connectivity (17) — Every qubit can interact with every other, typical of trapped ions. Removes routing overhead; Chapter 39 measured what that overhead costs elsewhere.

amplitude (3) — The complex coefficient of a basis state. Its squared magnitude is a probability.

amplitude amplification (21) — The generalization of Grover's rotation toward marked states.

amplitude damping (11) — Noise channel modelling energy loss ($T_1$).

amplitude encoding (32) — Packing $N$ classical values into $\log_2 N$ amplitudes. Exponentially compact and the source of the input problem: preparing the state costs what you saved.

analog Hamiltonian simulation (17) — Programming a device by specifying a Hamiltonian rather than a gate sequence; the neutral-atom model.

ancilla (19) — A helper qubit. Chapter 19 measured ancillas cutting an 8-bit oracle from 26,978 T gates to 55.

angle encoding (32) — One feature per rotation angle. Cheap to prepare, linear in qubits.

ansatz (8, 16, 24) — A parameterized circuit whose parameters are optimized. Its structure decides what the optimizer can reach.

API token (2) — The credential authenticating you to a cloud provider.

append (8) — Adding an instruction or sub-circuit to a circuit.

apply_layout (7) — Rewrites an observable to match a transpiled circuit's physical qubits. Skipping it silently measures the wrong qubits.

approximate QFT (22) — A QFT with small controlled rotations dropped. At $n=8$, cutoff 3 kept 97% fidelity for 36% of the rotations.

approximation degree (28) — Qiskit's knob for trading fidelity against gate count. At 0.9 it produced zero two-qubit gates — the optimization deleted the circuit.

approximation ratio (24, 37) — Achieved objective over optimal. Meaningless without its denominator: the same MaxCut assignment is 0.5 of the edges and 0.577 of the optimum.

assignment matrix (13) — The measured confusion matrix between prepared and observed bitstrings, inverted for readout mitigation.

authentication precondition (38) — QKD's classical channel must already be authenticated. Unauthenticated BB84 falls to a man-in-the-middle who shows both parties a clean QBER.

automatic differentiation (16) — Deriving gradients by tracing the computation; PennyLane's core mechanism.

B

backend (2, 12) — The device or simulator a circuit runs on.

barren plateau (16, 24, 32) — Exponentially vanishing gradients as circuit width grows. Measured: variance fell from 1.03e-01 to 1.17e-03 across 2 to 10 qubits, a factor of 88.

barrier (8) — A transpiler directive preventing optimization across a point. Not a physical operation.

basis encoding (32) — One qubit per bit. Simple, and the least compact.

basis gates (10) — The gate set a device physically implements; everything else is synthesized.

basis rotation (5) — Rotating before measurement so a non-$Z$ observable can be read in the computational basis.

basis state (3) — A computational-basis vector, $|0\rangle$ or $|1\rangle$ and their tensor products.

batch (7, 12) / batch mode (39) — Submitting many circuits in one job. The highest-leverage optimization on a cloud platform — worth ~99× for 100 circuits — and not a quantum technique.

batched QNode evaluation (33) — Passing a leading batch axis to a PennyLane QNode. Roughly 100× faster than looping; a sample-by-sample loop timed out at 10 minutes.

BB84 (38) — The Bennett–Brassard 1984 QKD protocol. Measured: 49.0% sifting, QBER 0.0000 clean.

Bell measurement (9) — Measuring in the entangled Bell basis; the core of teleportation.

Bell state (2, 4) — A maximally entangled two-qubit state.

Bernstein–Vazirani (20) — An oracle problem solved in one query where classical needs $n$.

big-endian (14) — Most-significant bit first. Cirq's convention; Qiskit's is the reverse, and Chapter 18 measured what that costs in translation.

binding (8) — Substituting concrete values for circuit parameters.

bit-flip code (25) — The three-qubit code correcting $X$ errors. Blind to $Z$ errors, by construction.

black box (19) — An oracle accessed only by querying; the model in which query-complexity separations are proven.

blind test (27) — A test that passes because it cannot detect the failure, not because there isn't one. Distinct from a flaky test, and worse.

Bloch sphere (3) — The geometric picture of a single qubit's pure states.

blueprint circuit (8) — A parameterized circuit template from Qiskit's library.

bond dimension (11) — The parameter controlling a matrix-product-state simulator's accuracy and cost.

Boolean oracle (19) — An oracle writing $f(x)$ into an output register.

Born rule (5) — $P(x) = |\langle x|\psi\rangle|^2$: the rule that turns a state into outcome probabilities. Three consequences make the second half of Chapter 5 statistics rather than physics — the theory gives probabilities and never outcomes, $N$ shots is one draw from a multinomial rather than $N$ facts, and a global phase cancels in every basis. It is why a quantum program's output type is a histogram.

Braket (17) — AWS's multi-vendor quantum service.

Bravyi–Kitaev (36) — A fermion-to-qubit mapping. Does not reduce the term count — 631 for LiH either way. It reduces Pauli weight: mean 6.16 → 5.62, max 12 → 10.

break-even point (25) — Where a code's logical error rate drops below the physical rate. Below it, error correction makes things worse.

C

calibration data (12) — Per-qubit and per-link error rates and coherence times, refreshed periodically.

calibration staleness (29) — Optimizing a layout against data that has since been refreshed. On a cloud platform this is the gap between submission and execution.

CCSD(T) (36) — Coupled cluster with perturbative triples, $\mathcal{O}(n^7)$. The real classical baseline for quantum chemistry, reaching chemical accuracy at 100+ orbitals — not full CI.

chemical accuracy (24, 36) — $1.6\times10^{-3}$ Hartree ≈ 1 kcal/mol, the threshold at which computed chemistry predicts measured chemistry.

chi-squared test (5) — A goodness-of-fit test for comparing observed counts to expectation.

circuit bisection (26) — Localizing a divergence by comparing prefixes. Chapter 26 measured it finding nothing on 0 of 100 random states — the technique has documented blind spots.

circuit knitting (35) — Cutting a large circuit into device-sized pieces with classical reconstruction, whose cost grows exponentially in the number of cuts.

circuit library (8) — Qiskit's collection of standard parameterized circuits.

Cirq (14) — Google's quantum framework; moment-based scheduling, big-endian.

claim audit (40) — Applying the book's eight questions to a quantitative claim. audit_claim returns failures, not a score.

classical feedforward (9) — Conditioning later gates on mid-circuit measurement results.

classical reduction (23) — The classical part of Shor's algorithm: factoring reduces to order finding.

classical shadows (35) — Estimating many observables from few randomized measurements. ~2.5× less accurate per observable, and wins at equal total budget by 1.5–1.8×.

classical simulation boundary (35) — The ~30–35 qubit point past which state-vector simulation fails. Below it, an advantage claim is checkable and therefore not an advantage.

Clifford twirl (30) — Randomizing over Clifford gates so a complicated error becomes depolarizing; what makes randomized benchmarking interpretable.

Clifford+T (19) — The standard fault-tolerant gate set. T gates dominate the cost: Chapter 15 measured one T gate taking a circuit from 450 to 2,882 physical qubits.

Clopper–Pearson interval (5) — A binomial confidence interval built by inverting the exact binomial rather than a normal approximation. Guaranteed-conservative coverage, wider than necessary. At 0/100 it gives 0.00000–0.03622 where the Wald interval gives 0.00000–0.00000 — a claim no experiment could license. See Wilson interval, rule of three.

CLOPS (30) — Circuit Layer Operations Per Second; the published metric that tries to capture throughput rather than fidelity.

CNOT (4) — The controlled-NOT gate, the standard two-qubit entangler.

code blindness (25) — Errors and states a code cannot detect. Mirrored between the bit-flip and phase-flip codes, and always a documented property.

code distance (15) — The minimum weight of an undetectable logical error; sets the overhead.

coherence budget (31) — Total circuit duration measured against $T_1$ and $T_2$.

CommutativeCancellation (28) — A transpiler pass cancelling gates that commute.

compose (8) — Combining circuits.

computational basis (5) — The $Z$ eigenbasis, in which measurement happens by default.

concurrence (4) — For a two-qubit pure state $a|00\rangle + b|01\rangle + c|10\rangle + d|11\rangle$, the single number $C = 2|ad - bc|$. 0 for every product state, 1 for every Bell state, whatever the phases. Purity, Bloch-vector length and reduced eigenvalues are all functions of it — $\text{purity} = 1 - C^2/2$, $|\vec r| = \sqrt{1-C^2}$ — so §4.4's three measured columns are one fact written three ways. It is also the check Chapter 17 §17.4 needed when a population-only comparison mistook a product state for a Bell state, and the one quantity in Chapter 4 that little-endian ordering cannot break, since $ad - cb = ad - bc$.

confidence interval (5) — A range of true values consistent with what you observed, at a stated coverage. The naive Wald form $\hat p \pm z\sqrt{\hat p(1-\hat p)/N}$ substitutes the estimate into the variance and fails exactly where it matters: it returns negative lower bounds, and at $\hat p = 0$ collapses to zero width. Measured coverage at $p = 0.01$, $N = 100$: Wald 0.6334 against a nominal 0.95, Wilson 0.9206.

continued fractions (23) — The classical post-processing recovering a period from a phase estimate.

controlled (15) — Q#'s automatically-derived controlled version of an operation.

controlled gate (4) — A gate applied conditionally on another qubit's state.

controlled-phase rotation (22) — The QFT's building block; the small ones are what the approximate QFT drops.

convex training (34) — A kernel method's optimization has a unique optimum, unlike a variational circuit's. Chapter 34's real advantage over Chapter 33 — and it bought an $n^2$ Gram matrix.

counting qubits (22) — The precision register in phase estimation; each adds a bit of precision.

counts (2) — The histogram of measured bitstrings.

coupling map (10, 29) — Which qubit pairs can interact. neighbors() is directed — treating it as undirected made 29 qubits appear isolated.

cross-entropy benchmarking (30) — The XEB fidelity measure used in quantum-supremacy experiments.

custom gate (8) — A user-defined unitary added to a circuit.

D

DAG (7, 10) — Directed acyclic graph; the transpiler's internal circuit representation.

data re-uploading (33) — Interleaving data encoding with trainable layers so a single qubit becomes a universal classifier (Pérez-Salinas et al.).

dead link (29) — A coupling-map edge whose error rate is 1.00. Chapter 39 found several on a real 133-qubit device.

debugging blindness (26) — The recurring Part V finding: a debugging technique that cannot detect the class of bug being sought.

decision procedure (12) — A written rule for choosing a backend before you see results, rather than after.

density matrix (11) — The state representation covering mixed states; required for noise.

deployable claim (33) — An accuracy claim supported by enough replicates to survive deployment. ClassifierReport.deployable_claim_supported refuses below MIN_SPLITS = 5.

depolarizing error (11) — Noise replacing a state with the maximally mixed state with some probability.

dequantization (35) — Finding a classical algorithm matching a claimed quantum speedup; it has happened repeatedly in QML.

Deutsch–Jozsa (20) — The first exponential black-box separation.

diffuser (21) — Grover's inversion-about-the-mean operator.

dt (31) — A backend's fundamental time resolution; every duration is an integer multiple.

dynamic circuit (6, 9) — A circuit with mid-circuit measurement and classical control flow.

dynamical decoupling (13, 31) — Pulse sequences suppressing idle-time dephasing. Measured significantly worse: XX at −0.0053 ± 0.0012, 4.4 standard errors in the wrong direction.

E

edge error (29) — The two-qubit error rate on a specific link. Returns 1.0 for dead links.

electronic structure problem (36) — Finding a molecule's ground-state energy; quantum computing's most-cited application.

encoding cost (32) — The qubits and gates needed to load classical data. The $N - \log_2 N - 1$ gap is the input problem.

endianness (18) — Bit-ordering convention. Qiskit and Cirq differ, and it is a leading source of translation bugs.

entanglement (1, 4) — Correlations with no classical description; a computational resource.

entanglement pattern (8) — How an ansatz wires its two-qubit gates: linear, circular, full.

equal-budget comparison (35) — Comparing sampling methods at equal total shots. Chapter 35's shadows comparison inverted when this was fixed.

error correction (13, 40) — Encoding logical qubits redundantly to detect and fix errors. The field's central open problem, and the one that survives the book's scrutiny.

error distribution (30) — The spread of error rates across a device. ErrorDistribution has no .error attribute, deliberately.

error mitigation (13) — Post-processing to reduce noise bias without correcting errors. Buys accuracy with shots.

Estimator (7) — The primitive returning expectation values; owns term grouping and mitigation.

execution metadata (39) — Job ID, timestamp, calibration snapshot, physical qubits. Four fields that come from the provider and are not in your code.

expectation value (5) — The average of an observable over many shots.

exponential separation (20) — A provable exponential gap between quantum and classical query complexity.

F

falsification condition (40) — The result that would show a research programme wrong. A programme without one is a marketing programme.

feature map (34) — The circuit embedding classical data for a quantum kernel. Tuning it moved accuracy from 0.6364 to 0.8500 — a larger swing than quantum-versus-classical.

fidelity (11) — Overlap between an achieved and an intended state. Never quote one without naming its statistic.

flaky test (27) — A test that fails on correct code by chance. Measured at 0.150%, after an initial 1.0% from too few runs.

framework (1) — A quantum SDK. framework selection (18) — Choosing one, on criteria that survive.

framework selection (18) — Choosing a quantum framework by capability rather than by preference. §18.7's procedure is four questions asked in order, stopping at the first yes; every question names something a framework can or cannot do, because a procedure that admits ergonomic terms will be dominated by them. Note the questions compose rather than exclude — PennyLane layers over Qiskit, Q#'s estimator consumes any framework's logical counts, Braket's OpenQASM 3 IR makes it a translation hub. Case Study 18.2's team framed five mutually exclusive options, at least three of which excluded nothing, and spent three weeks on a decision that deserved an afternoon.

full configuration interaction (36) — Exact diagonalization within a basis. Exponential; dies around 20 orbitals. Benchmarking against it is comparing to the method nobody uses.

function / functor (15) — Q#'s deterministic-callable and operation-transforming constructs.

G

gate count vs fidelity (28) — The distinction Chapter 28 turns on: fewer gates is not automatically better, and only fidelity settles it.

gate definition (6) — An OpenQASM gate declaration.

gate duration (31) — Physical execution time, from the backend target. rz is 0.0 ns — a virtual gate.

gate error (12) — Per-gate error probability from calibration.

gate folding (13) — Replacing $G$ with $GG^\dagger G$ to amplify noise for zero-noise extrapolation.

gateset (14) — Cirq's term for a device's native gates.

general number field sieve (23) — The best classical factoring algorithm; the baseline Shor must beat.

GHZ state (4) — A maximally entangled multi-qubit state.

global phase (3, 6) — An unobservable overall phase — except when the state is controlled, when it becomes relative and observable.

Goemans–Williamson (24, 37) — The 1994 SDP-based MaxCut approximation, guaranteeing 0.87856 of the optimum. Beat QAOA 6–0 with 4 ties over ten instances.

GPi (17) — A native IonQ single-qubit gate.

gradient variance (16, 32) — The quantity that vanishes in a barren plateau.

Gram matrix (34) — The $n \times n$ matrix of pairwise kernel values. Must have unit diagonal, be symmetric, and be positive semidefinite.

GridQubit (14) — Cirq's 2D-lattice qubit identifier.

Grover's algorithm (21) — Quadratic-speedup unstructured search. It does not search a database — it queries an oracle you must already be able to build.

H

Hadamard gate (3) — Creates equal superposition; maps $Z$ to $X$.

hardware-aware programming (29) — Choosing layouts and shapes from calibration data. Measured: level 1 hardware-aware (0.9116) beat naive level 3 (0.7720) by +0.1397.

Hartree–Fock (24) — The mean-field reference state VQE starts from.

harvest-don't-recall (40) — Deriving a summary from the source files rather than from memory. A capstone written from recall produces numbers that are plausible and wrong.

heavy-hex (29) — IBM's coupling-map topology: degree ≤ 3, so routing overhead is significant.

hidden subgroup (20) — The abstract problem unifying Simon's, Shor's, and others.

hybrid loop (24) / hybrid loop latency (35) — The quantum-classical optimization cycle, and what re-queuing costs it: 120 iterations as 120 jobs is 10 hours waiting, under a second computing.

I

if_test (9) — Qiskit's classical-conditional construct for dynamic circuits.

inference cost (33) — What a trained quantum model costs to use: 27.8 QPU hours per million predictions at 1,000 shots.

input problem (32) — Loading classical data destroys the speedup that motivated the encoding. The central obstacle in QML.

InsertStrategy (14) — Cirq's control over how operations pack into moments.

interchange format (6) — A representation for moving circuits between tools; OpenQASM's role.

interference (3, 20) — Amplitudes cancelling and reinforcing. The actual mechanism of every quantum speedup.

intermediate representation (6) — A compiler's internal circuit form.

interoperability (18) — Moving work between frameworks, and what it costs.

ISA circuit (7, 10) — A circuit expressed purely in a backend's native instruction set.

Ising mapping (37) — MaxCut as a diagonal Hamiltonian. Exact and free — unlike Chapter 36's active space, no approximation stands between problem and Hamiltonian.

iteration count (21) — Grover's $\approx \frac{\pi}{4}\sqrt{N}$ optimum. Overshooting is severe: at $N=16$, 3 iterations gives 0.9613 and 6 gives 0.0204.

J

job (12) — One submission to a backend.

Jordan–Wigner (36) — The standard fermion-to-qubit mapping. Stores occupation locally and pays in parity: a single excitation becomes an $\mathcal{O}(n)$ Pauli string.

K

kernel concentration (34) — Kernel values collapsing toward a constant as dimension grows. Driven by feature dimension, not qubit count — a prediction Chapter 34 got wrong and corrected.

key expansion (38) — Turning a short shared secret into a long one. What QKD actually does — it cannot establish a key between parties who have never met.

L

latency (9) — Round-trip delay, especially for classical feedback inside a circuit.

layout (7, 10) / layout selection (29) — The map from virtual to physical qubits.

layout roulette (39) — Transpiler seed alone changing 14-qubit fidelity from 0.5755 to 0.7911 — 2.03× the error. Zero variation at 4 qubits, which is the trap.

layout scoring (12) — Ranking candidate layouts by predicted fidelity.

LineQubit (14) — Cirq's 1D qubit identifier.

lingua franca (18) — The role OpenQASM aspires to and partly fills.

LocalSimulator (17) — Braket's local backend.

logical error rate (25) — Error rate of an encoded qubit; the quantity that must fall below the physical rate.

logical qubit (25) — A qubit encoded across many physical ones.

logical vs physical qubit mapping (26) — The distinction that makes transpiled circuits hard to debug.

M

M3 (13) — Matrix-free measurement mitigation, scaling past full assignment-matrix inversion.

magic state (15) — The resource distilled to implement T gates fault-tolerantly. T factories dominate the qubit budget.

marginal distribution (5) — The distribution over a subset of measured qubits.

matrix product state (11) — A simulation method efficient for low-entanglement circuits.

MaxCut (37) — Partition a graph's vertices to maximize crossing edges. The standard QAOA benchmark.

measurement (1, 5) — Collapsing a state to a basis outcome. Chapter 38 turns this from an inconvenience into a security guarantee.

metamorphic testing (27) — Testing relations between outputs when no exact expected value exists.

mid-circuit measurement (9) — Measuring without ending the circuit.

mixer Hamiltonian (37) — QAOA's $\sum_j X_j$ term, driving transitions between solutions.

modality (17) — The physical qubit technology. Chapter 39 measured a 28× per-shot price difference between superconducting and trapped-ion.

modular exponentiation (23) — Shor's dominant subroutine and its real cost.

moment (14) — Cirq's unit of simultaneous execution.

multi-controlled X (19) — A Toffoli generalization; the cost driver in oracle construction.

multinomial distribution (5) — The correct model for a counts dictionary: $N$ shots distributed over outcomes with probabilities summing to 1. Gives the same per-outcome variance as the binomial, $p_i(1-p_i)/N$, plus a covariance the single-outcome formula hides: $\operatorname{Cov}(\hat p_i, \hat p_j) = -p_ip_j/N$. The negative sign is arithmetic, not noise — the fractions sum to exactly 1 in every run — and ignoring it over-estimates the error on anything built from more than one outcome. On a noiseless GHZ fidelity it is the difference between a standard deviation of $\pm 0.011$ and the true 0.000000.

multiple comparisons (5) — The inflation of false-positive rate with the number of questions asked of one dataset. At $\alpha = 0.05$, $P(\text{at least one spurious hit}) = 1 - 0.95^m$: 0.0500 at one test, 0.3366 at eight, 0.6415 at twenty — with every circuit perfect. §5.7's free marginals are a quiet on-ramp, since seven marginals from one run is seven questions. The defences are social rather than statistical: decide which comparison matters before looking, report how many you ran, and tighten $\alpha$ if you must run many.

Mølmer–Sørensen gate (17) — The native trapped-ion two-qubit entangling gate.

N

$n^2$ scaling (34) — A kernel method needs all pairwise values, so training cost is quadratic in dataset size. What convex training cost.

n-squared scaling (34) — The kernel method's distinctive $n^2$ cost shape: training an SVM needs the kernel between every pair of training points, so a Gram matrix for $n$ points is $n(n+1)/2$ unique circuit evaluations. Measured at $n = 201$: 40,401 entries in 51.3 s. It compounds with the shot budget rather than replacing it — each entry is a probability of size $K$ needing $\mathcal{O}(1/K^2)$ shots — which is why §34.7's bill is the third distinct cost shape in Part VI, after Chapter 32's training and Chapter 33's inference.

negative result (40) — A finding that something does not work. The most valuable output of the QML literature, and the most reliable measurements in this book.

neutral atom (17) — A qubit modality using optically trapped atoms.

NISQ (1) — Noisy Intermediate-Scale Quantum; Preskill's term for the current era.

no-cloning theorem (38) — Unknown quantum states cannot be copied. Provides BB84's security and imposes its 240 km range limit — the same theorem does both.

noise model (11) — A simulator's description of device errors.

O

observable (5) — A Hermitian operator whose expectation you estimate.

OpenQASM (6, 18) — The assembly-level interchange language for quantum circuits.

operation (15) — Q#'s callable that may act on qubits.

Operator.from_circuit (26) — Recovers a transpiled circuit's unitary, applying both layout and routing. Limited to ~14 qubits before it becomes impractical.

optimality certificate (37) — A bound proving a solution is near-optimal. GW produces one and QAOA does not — a difference in what the algorithms produce, and it does not close with better hardware.

optimization level (10, 28) — Qiskit's preset transpilation effort, 0–3. Levels 2 and 3 differ in 14 of 40 circuit-seed pairs.

optimizer (16) — The classical routine updating variational parameters.

oracle (19, 21) — A black-box subroutine. Its cost is where speedup claims usually hide.

order finding (23) — Finding the period of $a^x \bmod N$; Shor's quantum core.

over-rotation (21) — Running Grover past the optimum, rotating away from the answer.

P

parameter (8) / ParameterExpression / ParameterVector — Qiskit's symbolic circuit parameters.

parameter-shift rule (16) — Exact analytic gradients of a quantum circuit from two extra circuit evaluations.

ParamResolver (14) — Cirq's parameter-binding mechanism.

partial trace (4) — Discarding a subsystem to get a reduced state.

pass (10) / pass manager (2, 7, 10) / preset pass manager (28) — A transpiler stage, the pipeline of stages, and Qiskit's standard pipelines.

pass introspection (28) — Inspecting which passes actually ran; needs a recursive .tasks flatten.

Pauli operator (5) — $I$, $X$, $Y$, $Z$; the basis for observables and errors.

Pauli weight (36) — Non-identity factors in a Pauli string, setting measurement-circuit depth. What Bravyi–Kitaev actually reduces.

PennyLane (16) — A differentiable-programming quantum framework.

per-minute pricing / per-shot pricing (39) — Two billing models that cannot be compared without the circuit duration. The same run: $50 or $185,542.

period finding (23) — See order finding.

phase estimation (22) — Extracting an eigenvalue's phase. Exact for dyadic phases. Reaches precision $\epsilon$ in $\mathcal{O}(1/\epsilon)$ — the route that survives Chapter 36's arithmetic.

phase gate (3) — A $Z$-axis rotation.

phase kickback (19) — The mechanism putting an oracle's answer into a control qubit's phase.

phase oracle (19) — An oracle marking states by sign rather than by writing an output.

phase-flip code (25) — The three-qubit code correcting $Z$ errors; the bit-flip code conjugated by Hadamards.

physical qubit (6) — An actual device qubit, as opposed to a logical or virtual one.

portability (18) — How well a circuit survives moving between frameworks.

positive semidefinite (34) — A required property of any valid kernel Gram matrix.

post-quantum cryptography (23, 38) — Classical cryptography believed quantum-resistant. ML-KEM-768 measured at 201.7 µs against X25519's 76.6 µs — 2.6× slower, over any existing network.

precision (7) — The Estimator's target standard error.

prefix inspection (26) — Examining the state after the first $k$ instructions.

preflight check (12) — Validating a job before submitting it.

primitive (1, 2, 7) — Sampler and Estimator; the modern execution interface.

privacy amplification (38) — Compressing a reconciled key to the length an eavesdropper provably cannot know.

probabilistic error cancellation (13) — Mitigation by sampling from an inverse noise channel; the sampling overhead grows exponentially.

product state (4) — A multi-qubit state factorizable into single-qubit states — i.e. unentangled.

promise problem (20) — A problem where the input is guaranteed to satisfy a condition. Where most oracle separations live, and why they rarely apply directly.

property-based testing (27) — Testing invariants over generated inputs.

pulse-level control (31) — Programming the analog waveforms beneath gates. qiskit.pulse was removed in Qiskit 2.0.

purity (11) — $\mathrm{Tr}(\rho^2)$; 1 for pure states, lower for mixed.

Q

Q# (15) — Microsoft's quantum language, with a strong resource estimator.

QAOA (24, 37) — The Quantum Approximate Optimization Algorithm. At $p=1$ scores 0.751–0.8086, below a 1994 proof; won 0 of 10 instances against Goemans–Williamson.

QBER (38) — Quantum bit error rate. Intercept-resend gives 0.2398 against a $\frac12 \times \frac12$ prediction.

QFT (22) — Quantum Fourier Transform. Exponentially fewer operations than the classical FFT, and not an exponential speedup, because you cannot read out the amplitudes.

qiskit.pulse removal (31) — Removed in Qiskit 2.0 along with add_calibration, backend.defaults and instruction_schedule_map.

QNode (16) — PennyLane's circuit-plus-device callable.

QPE (22) — See phase estimation.

quadratic speedup (21) — Grover's $\sqrt{N}$. Chapter 21 measured why quadratic is often not enough once constants are counted.

quantum advantage (1) — A quantum computer outperforming the best classical method on a real task. Zero instances in this book's six head-to-head comparisons.

quantum circuit (1) — A sequence of gates and measurements.

quantum cloud platform (39) — The service layer. The queue is the product — at a five-minute queue, utilization is 2.31 × 10⁻⁵.

quantum data (35) — Data that is already a quantum state. The surviving case for QML, because the input problem disappears.

quantum data advantage (40) — Proven exponential separations for learning from quantum experiments.

Quantum Fourier Transform — See QFT.

quantum kernel (34) — A kernel computed as a quantum state overlap. Measured 0.8313 against SVC(rbf)'s 0.8889.

quantum software stack (1) — Application to algorithm to circuit to pulse to control electronics.

Quantum Volume (30) — A single-number device benchmark combining width and depth.

qubit (1) — A two-level quantum system.

qubit release (15) — Returning a qubit to the allocator in Q#; requires it be in a known state.

qubit reuse (9) — Resetting and reusing a qubit mid-circuit.

qubit-wise commuting (5) — The stronger condition two Pauli strings must meet to share one measurement circuit: they must commute on every qubit separately, not merely as operators. $ZZ$ and $XX$ commute globally and still need two circuits. Commuting is not sufficient — this is what decides how many circuits a Hamiltonian actually costs, and why Chapter 36's lighter Bravyi–Kitaev mapping (average Pauli weight 5.62 against Jordan–Wigner's 6.16) groups better.

query complexity (19, 20) — Number of oracle calls needed. Where exponential separations are proven, and it says nothing about the cost of building the oracle.

queue latency (39) — Wait time before execution. Dominates by four to six orders of magnitude.

quoted fidelity (30) — A device fidelity as reported. Requires naming its statistic: one chip supported 0.00750 to 0.07205.

R

randomized benchmarking (30) — Estimating average gate error from Clifford-sequence decay.

readout error (2, 11, 12) — Misidentifying a measured state. Often the largest single error source.

readout mitigation (13) — Correcting readout error by inverting the assignment matrix.

readout problem (22) — Having a result in amplitudes you cannot efficiently extract. Why the QFT's speed does not translate.

reduced state (4) — The state of a subsystem after partial trace.

register (6) — A named array of qubits or classical bits.

relative phase (3) — Phase between amplitudes; observable, unlike global phase.

repeat-until-success (9) — Retrying a probabilistic gate construction until it succeeds.

replication (33) — Repeating a comparison across splits or seeds. compare_models returns INSUFFICIENT_REPLICATES below MIN_SPLITS = 5.

reproducibility (39) — Whether a result can be re-run. A pinned environment and a fixed seed pin the client side of a computation whose variance lives on the server side.

reset (9) — Returning a qubit to $|0\rangle$ mid-circuit.

resource estimation (15) — Computing the physical qubits and runtime a fault-tolerant algorithm needs.

rotation gate (3) — A parameterized single-qubit rotation.

rotation synthesis (22) — Decomposing arbitrary rotations into a discrete fault-tolerant gate set.

round trip (6) — Exporting and re-importing a circuit; the test of a serialization format.

routing (10) — Inserting SWAPs so two-qubit gates act on connected qubits. Chapter 39 measured a 49-to-112 spread in two-qubit gate count from routing alone.

RSA (23) — The public-key cryptosystem Shor's algorithm threatens.

rule of three (5) — Observing zero events in $N$ trials puts a 95% upper bound of roughly $3/N$ on the rate. The one-step answer to "we saw no failures, so the failure rate is 0%." At $N = 100$: $3/N = 3.00\%$ against an exact Clopper–Pearson bound of 3.62%, so the approximation is good and slightly optimistic. Note it bounds a rate and scales as $N^{-1}$, unlike the $N^{-1/2}$ of sampling error.

S

SABRE (10) — Qiskit's default layout-and-routing heuristic.

Sampler (7) / SamplerV2 (2) — The primitive returning measurement outcomes.

sampling error (5) — Statistical error from finite shots; falls as $1/\sqrt{N}$.

sampling overhead (13) — The extra shots a mitigation technique costs.

seam (26) — The boundary between two correct pieces of work, where composition failures live.

second quantization (36) — The creation/annihilation-operator formalism for many-electron systems.

seed variance (28) — Run-to-run variation from transpiler randomness. Grows with circuit size and with QAOA depth: σ from 0.000 at $p=1$ to 0.021 at $p=4$.

seed_transpiler (10) — The transpiler's random seed. Fixing it does not fix the layout if calibration changes.

semidefinite relaxation (37) — Relaxing spins to unit vectors and solving an SDP. Produces GW's solution and its upper bound.

serialization (6) — Writing a circuit to a portable format.

session (7, 12) / session mode (39) — Reserved access for dependent jobs. Pays the queue once instead of once per iteration.

shape factor (5) — The quantity $S = \sum_i \sqrt{p_i(1-p_i)}$ in the shot-noise floor $\mathbb{E}[\mathrm{TVD}] = 0.39894\,S/\sqrt N$, where $0.39894 = 1/\sqrt{2\pi}$. It belongs to the distribution, not to the shot count, which is why the floor cannot be quoted as a single number for all circuits. A noiseless GHZ(3) puts $1/2$ on two outcomes and zero on six, giving $S = 1$ exactly and a predicted 0.01262 at 1,000 shots against a measured 0.01313.

Shor's algorithm (23) — Polynomial-time factoring on a fault-tolerant quantum computer.

shot (5) / shots (1, 2) — One execution of a circuit; the number of repetitions.

shot budget (24) — Total shots a computation needs. Scales as $1/\epsilon^2$ and dominates every variational cost model: 1.91 × 10²⁰ shots at Chapter 36's crossover.

shot-noise floor (27) — The precision limit from finite sampling, ≈ $3/\sqrt{N}$. A tolerance below it makes a test blind.

sifting (38) — Discarding BB84 positions where bases disagreed; keeps ~half. Measured 49.0%.

Simon's algorithm (20) — An exponential separation and the template for Shor's.

simulation method (7, 11) — Statevector, density matrix, matrix product state, stabilizer.

stabilizer (11) / stabilizer code (25) — Efficiently simulable circuits (Gottesman–Knill), and the code family built from stabilizer measurements.

standard error (5) — Uncertainty in an estimated mean. A standard error of exactly zero means it was never estimated.

statevector (3, 11) — The full complex amplitude vector.

statevector inspection (26) — Reading intermediate state in simulation, via Statevector.from_instruction.

statistical power (5) — The probability of detecting a real effect.

statistical significance (40) — Whether a gap survives its uncertainty. Chapter 33: +0.0202 ± 0.0170 is not significant; +0.0626 ± 0.0067 is, at nine standard errors.

stdgates (6) — OpenQASM's standard gate library.

superdense coding (9) — Sending two classical bits with one qubit and prior entanglement.

superposition (1, 3) — A qubit in a combination of basis states.

surface code (25) — The leading planar error-correcting code.

survival probability (30) — The probability a benchmarking sequence returns the initial state.

SWAP (4) / swap insertion (10) — Exchanging two qubits' states, and the transpiler adding them for connectivity.

sweep (14) — Cirq's parameter scan.

Sycamore (14) — Google's superconducting processor.

syndrome extraction (25) — Measuring stabilizers to detect errors without collapsing the logical state. Not free — it costs qubits, time, and its own errors.

T

T factory (15) — The magic-state distillation apparatus. 93% of a small circuit's qubits and 3% of a large one's — the fraction inverts with scale.

T1 / T2 (12) — Energy relaxation and dephasing times. Measured spanning 15.2 to 483.0 µs on one device.

teleportation (9) — Moving a state using entanglement and two classical bits.

template (16) — A reusable PennyLane circuit block.

tensor product (4) — How multi-qubit state spaces compose.

thermal relaxation (11) — Noise combining $T_1$ and $T_2$ effects.

threshold (25) — The physical error rate below which error correction improves with more qubits.

Toffoli gate (4) — The controlled-controlled-NOT.

tolerance (27) — A test's allowed deviation. assert_distribution raises below the shot-noise floor.

total variation distance (27) — Half the $L_1$ distance between distributions; the natural metric for comparing measured outcomes.

train/test split (33) — Partitioning data for evaluation. One split is a draw from a distribution.

translation loss (18) — What is lost moving a circuit between frameworks.

transpilation (1, 10) / transpile (2) — Rewriting a circuit for a target device.

transpilation verification (26) — Checking a transpiled circuit still implements the original.

trapped ion (17) — A qubit modality with all-to-all connectivity and long coherence. 28× the per-shot price.

trusted node (38) — A relay that decrypts and re-encrypts to extend QKD range. Sees the key in the clear — a 2,000 km network has 19 of them.

tuned baseline (40) — A classical method someone actually optimized. is_tuned_baseline rejects "random", "greedy", and "full CI".

twirling (13) — Randomizing to convert coherent errors into stochastic ones.

two error rates (27) — A test's flakiness rate and its blindness rate. Optimizing one alone trades against the other.

U

uncomputation (19) — Reversing intermediate computation to free ancillas without measurement.

unitary (3) — A norm-preserving linear operator; every gate is one.

universality (33) — Data re-uploading making a single qubit a universal function approximator.

unstructured search (21) — Searching with no exploitable structure; Grover's setting, and rarer in practice than it sounds.

usable path (29) — A connected chain of qubits with acceptable error rates.

use block (15) — Q#'s scoped qubit allocation.

utilization (39) — Device time over wall-clock time. 2.31 × 10⁻⁵ at a five-minute queue.

V

variational circuit (16) / variational quantum classifier (32) — A parameterized circuit trained by a classical optimizer, and the classifier built from one.

verbatim box (17) — Braket's directive to run a circuit exactly as written, without compilation.

virtual environment (2) — An isolated Python installation.

virtual Z gate (31) — A $Z$ rotation implemented as a phase-bookkeeping change. Exact, and 0.0 ns.

VQE (24) — The Variational Quantum Eigensolver. Reached chemical accuracy on H₂ from four parameters, and 2.04 × 10⁻⁹ Ha inside an active space wrong by 2.01 × 10⁻² Ha.

W

W state (4) — A multi-qubit entangled state with different structure from GHZ; robust to loss.

Wilson interval (5) — A binomial confidence interval that inverts the test rather than substituting $\hat p$ into the variance. The $z^2/2N$ term pulls the centre off the boundary and the $z^2/4N^2$ under the root keeps the width positive at $\hat p = 0$. At 0/100 it gives 0.00000–0.03699 where Wald gives 0.00000–0.00000; at 50/100 all three methods agree to two decimals, so near $p = 0.5$ the extra machinery buys nothing.

Z

zero-noise extrapolation (13) — Amplifying noise deliberately and extrapolating to the zero-noise limit.