Measurement-Based Quantum Computation

Measurement-based quantum computation (MBQC) is a model of universal quantum computation that runs almost no gates. All of the entanglement is manufactured once, up front, in a large fixed resource — paradigmatically a cluster state on a two-dimensional lattice — and the algorithm is then executed entirely by measuring the qubits of that resource, one at a time, in adaptively chosen single-qubit bases. The measurements both consume the entanglement and perform the logic, so the resource is irreversibly used up as the program runs; for the cluster state this earns the model its other name, the one-way quantum computer. The conceptual jolt is that tearing entanglement down by local measurement can do the same work as building it up with gates. The bridge is quantum teleportation: a measurement in a tilted basis teleports the logical qubit from one site to its neighbour while stamping a single-qubit rotation onto it, fixed by the measurement angle. Chain such steps along a line and a logical qubit is threaded through a programmable sequence of rotations; braid two lines together through the graph's two-dimensional connectivity and the resident controlled-Z becomes an entangling gate. Because outcomes are random, each step also applies a known Pauli — the byproduct — tracked and undone by adapting the bases of measurements not yet made. MBQC is exactly as powerful as the circuit model: both decide BQP. What it changes is the accounting. The entangling resource is Clifford and stabilizer, so it can be built offline, even by lossy heralded linear optics; the non-Clifford content lives only in the list of measurement angles. That clean separation is why MBQC is the native language of photonic quantum computing, of fault tolerance on foliated codes, and of the resource theory of magic.

Key takeaways

MBQC inverts the usual order of quantum computing: entanglement is made once in a fixed resource state (a cluster state), and computation proceeds by an adaptive sequence of single-qubit measurements.
Each measurement teleports a logical qubit to the next site while applying a gate; the randomness of the outcome is corrected by adapting later measurement angles (feedforward).
The canonical resource is the 2D cluster state, and the one-way model is polynomially equivalent in power to the circuit model: BQP = MBQC.
Measuring a bonded qubit in the basis {|0⟩ ± e^(−iα)|1⟩} implements HZ(α) on the propagated logical qubit, up to a known Pauli byproduct.
Flow and gflow conditions on the underlying graph guarantee a deterministic, causally consistent measurement order exists; the measurement calculus formalizes pattern compilation.
Only the angles are adaptive — the entangling resource is Clifford and can be prepared offline, making MBQC natural for photonics, cold atoms, and fault-tolerant hardware.
Key idea: a single-qubit measurement of an entangled qubit is simultaneously a teleportation and a gate — measuring in the basis {|0⟩ ± e^(−iα)|1⟩} teleports the logical state to the next qubit while applying HZ(α), up to a Pauli byproduct that adaptive feedforward removes.

The core insight

A single-qubit measurement of an entangled qubit is simultaneously a teleportation and a gate: measuring in the basis $\{|0\rangle\pm e^{-i\alpha}|1\rangle\}$ teleports the logical state to the next qubit while applying $HZ(\alpha)$ , up to a Pauli byproduct that adaptive feedforward removes.

Where this lives today

MBQC is the operating model of much of photonic quantum computing, where deterministic two-photon gates are hard but measurement is cheap and fast. Photonic machines fuse many small entangled states into a large cluster offline with probabilistic heralded gates, then compute by measuring single photons at feedforward-controlled angles; Xanadu's continuous-variable Borealis and PsiQuantum's fusion-based architecture both descend directly from the one-way model. In fault tolerance, foliating a surface code through time produces a 3D cluster state on which Raussendorf, Harrington and Goyal's topological scheme performs error-corrected logic by measurement — the intellectual root of modern lattice-surgery roadmaps. MBQC also underwrites blind quantum computation: because the resource is problem-independent and all content is classical angles, Broadbent, Fitzsimons and Kashefi's Universal Blind Quantum Computation lets a nearly classical client hide a computation from a quantum server — demonstrated with photons by Barz and coworkers in 2012. And because Clifford MBQC is classically simulable, the model remains the cleanest laboratory for the resource theory of magic.

The physical picture

Picture a long chain of qubits, each prepared in $|+\rangle$ and entangled to its neighbours by controlled- $Z$ . Measure the first qubit in a tilted basis: it collapses, but because of the entanglement it does not simply vanish — it deposits a rotated copy of its state onto the second qubit, exactly as in teleportation, except that the rotation is deliberately baked into the choice of basis. Repeat down the chain and an arbitrary single-qubit unitary has been assembled from nothing but measurements. Each step also flips a known Pauli onto the wire — the byproduct the rest of the machinery exists to tame.

Figure

A linear cluster state as a quantum wire: qubits start in the plus state bonded by controlled-Z, and measuring site j in the α-basis teleports the logical state one step to the right while applying the gate HZ(α), leaving a random but known Pauli byproduct X^s that later feedforward cancels

The two-dimensional layout is what supplies entangling power. Set two chains side by side and let them share a single vertical cluster bond; the controlled- $Z$ already sitting on that rung becomes a logical entangling gate once the wires are measured through. A 1D chain is therefore only a programmable single-qubit rotation, while a 2D cluster state is a universal substrate. The angle you choose at each site is the entire program: $\alpha=0$ is a bare wire, the Clifford angles $\alpha\in\{0,\pi/2,\pi,3\pi/2\}$ keep everything stabilizer and classically simulable, and the single non-Clifford magic angle $\alpha=\pi/4$ injects the $T$ -gate content that lifts the model to universality.

Figure

The X–Y measurement plane. The angle α selects the logical gate HZ(α): α = 0 (X basis) is a Clifford wire, α = π/2 (Y basis) is the Clifford phase gate S, and the magic angle α = π/4 gives HZ(π/4) = HT — the one non-Clifford ingredient that lifts the stabilizer model to BQP universality; random outcomes are undone by α′ = (−1)^s_X α + s_Z π

Think of the byproduct Paulis as a frame the computation drifts in. Clifford gates merely permute the frame, so they can be pushed to the very end; only a non-Clifford angle notices it, and that is exactly the moment you must feed earlier outcomes forward and flip the sign of the angle. The resource state never appears in the algorithm's logic: it is generic, problem-independent entangled hardware, and the whole program is a classical adaptive readout schedule.

The core equations

The resource is a graph state. Assign one qubit to each vertex of a graph $G=(V,E)$ , prepare every qubit in $|+\rangle$ , and apply a controlled- $Z$ across every edge. The cluster state is the special case where $G$ is a regular 2D lattice.

|G\rangle=\Big(\prod_{(i,j)\in E}CZ_{ij}\Big)\,|+\rangle^{\otimes V} \qquad (1)

where: $G=(V,E)$ — graph with vertex set $V$ (one qubit per vertex) and edge set $E$ ; $|+\rangle=\tfrac{1}{\sqrt2}(|0\rangle+|1\rangle)$ , the $+1$ eigenstate of $X$ ; $CZ_{ij}$ — controlled- $Z$ between qubits $i,j$ , with $CZ=\mathrm{diag}(1,1,1,-1)$ . Reading: the state is fixed once the graph is chosen — no algorithmic information has entered yet. Graph states are stabilizer states. Each vertex contributes one generator that pins the state to its $+1$ eigenspace, so $|G\rangle$ is the unique joint $+1$ eigenstate of the $K_i$ .

K_i=X_i\prod_{j\in N(i)}Z_j,\qquad K_i|G\rangle=|G\rangle \qquad (2)

where: $K_i$ — the $i$ -th stabilizer generator; $N(i)$ — the neighbours of vertex $i$ ; $X_i$ — Pauli $X$ on $i$ , $Z_j$ — Pauli $Z$ on each neighbour. Reading: the entire resource is described by $|V|$ commuting Pauli products, which is why it can be prepared, verified, and error-corrected with stabilizer tools. Computation is driven by measuring in the $X$ – $Y$ plane of the Bloch sphere. Measuring qubit $j$ at angle $\alpha$ is operationally "apply $Z(\alpha)$ , then measure in the $X$ basis", so the projector is $\langle+|Z(\alpha)$ and its basis is the rotated pair below.

|\pm_\alpha\rangle=\tfrac{1}{\sqrt2}\big(|0\rangle\pm e^{-i\alpha}|1\rangle\big),\qquad M_j^{\alpha}\equiv\langle+|\,Z(\alpha),\quad Z(\alpha)=\mathrm{diag}(1,e^{i\alpha}) \qquad (3)

where: $\alpha$ — measurement angle; outcome $s_j=0$ selects $|+_\alpha\rangle$ , $s_j=1$ selects $|-_\alpha\rangle$ ; $Z(\alpha)$ — phase rotation about $\hat z$ . Reading: choosing $\alpha$ is choosing the gate; the two outcomes are equally likely and differ only by a Pauli. A single such measurement teleports the logical qubit one step and applies $HZ(\alpha)$ , up to a Pauli that depends on the random outcome. This is the atom of the whole model.

|\psi\rangle\ \longrightarrow\ X^{s}\,H\,Z(\alpha)\,|\psi\rangle \qquad (4)

where: $|\psi\rangle$ — input logical state on the measured qubit; $H$ — Hadamard, produced automatically by the teleportation; $X^{s}$ — byproduct Pauli set by the outcome $s$ . Reading: the gate $HZ(\alpha)$ is deterministic; the $X^{s}$ is the price of using an irreversible measurement. Because the byproduct is random, a later measurement must be pre-compensated. Commuting the accumulated Pauli frame through the next $Z(\alpha)$ shows the angle to actually dial in is $\alpha'$ .

\alpha'=(-1)^{s_X}\,\alpha+s_Z\,\pi \qquad (5)

where: $\alpha'$ — corrected angle; $s_X$ — parity of inherited $X$ -byproducts (flips the sign of $\alpha$ ); $s_Z$ — parity of inherited $Z$ -byproducts (shifts $\alpha$ by $\pi$ ). Reading: this feedforward is what makes the answer deterministic despite genuinely random outcomes. Any pattern can be written in a normal form: prepare all qubits, entangle all edges, measure, then correct. This is the measurement-calculus (EMC) standard form, the basis of pattern compilation.

U_G=\Big(\prod_i C_i[s]\Big)\Big(\prod_i M_i^{\alpha_i}\Big)\Big(\prod_{(i,j)\in E} E_{ij}\Big)\Big(\prod_i N_i\Big) \qquad (6)

where: $N_i$ — prepare qubit $i$ in $|+\rangle$ ; $E_{ij}$ — entangle with $CZ_{ij}$ ; $M_i^{\alpha_i}$ — measure at angle $\alpha_i$ ; $C_i[s]$ — Pauli correction depending on prior outcomes $s$ . Reading: read right-to-left — preparation, then entanglement, then adaptive measurement, then residual corrections. Determinism is not automatic; it needs a geometric condition on the open graph. A causal flow assigns each measured qubit a successor and a partial order compatible with feedforward.

\exists\,f:O^{c}\to I^{c},\ \preceq:\quad (i)\ i\sim f(i),\quad (ii)\ i\prec f(i),\quad (iii)\ j\sim f(i)\Rightarrow j=i\ \text{or}\ i\prec j \qquad (7)

where: $f$ — flow function mapping each non-output qubit to a successor; $I,O$ — input and output vertices, $I^{c},O^{c}$ their complements; $\preceq$ — the induced partial order, i.e. a legal measurement schedule; $\sim$ — adjacency in $G$ . Reading: if a flow exists, a fixed order and feedforward rule make the logical map independent of every outcome.

Derivation AOne measurement is a teleportation and a gate

The goal is to establish the atom of MBQC from first principles: that projecting one qubit of a two-qubit cluster onto the $\alpha$ -basis both teleports the logical state onto its partner and applies the fixed gate $HZ(\alpha)$ , with only a random Pauli left over. We start from a logical qubit $|\psi\rangle=a|0\rangle+b|1\rangle$ on site 1, bond it to a fresh $|+\rangle$ on site 2 with a controlled- $Z$ , project site 1, and read off the surviving state on site 2, tracking the outcome-dependent byproduct exactly.

Step 1. Apply the bond: $CZ_{12}\,(a|0\rangle+b|1\rangle)_1|+\rangle_2=\tfrac{1}{\sqrt2}\big(a|0\rangle(|0\rangle+|1\rangle)+b|1\rangle(|0\rangle-|1\rangle)\big)$ , because $CZ$ flips the sign of the $|1\rangle_2$ branch only when site 1 is $|1\rangle$ . Why: the entanglement correlates the two qubits so that measuring one steers the other — no correlation, no teleportation.

Step 2. Project site 1 with $\langle\pm_\alpha|=\tfrac{1}{\sqrt2}(\langle0|\pm e^{+i\alpha}\langle1|)$ , the bra of the basis in equation (3). Collecting terms on site 2 for outcome $s=0$ gives $\tfrac12\big[(a+b\,e^{i\alpha})|0\rangle+(a-b\,e^{i\alpha})|1\rangle\big]$ . Why: the tilt $e^{i\alpha}$ of the basis is precisely what injects the programmable phase — the measurement angle becomes a gate parameter here.

Step 3. Recognize the gate: that vector is $\tfrac{1}{\sqrt2}H\big(a|0\rangle+b\,e^{i\alpha}|1\rangle\big)=\tfrac{1}{\sqrt2}H\,Z(\alpha)|\psi\rangle$ , since $H$ maps $(c,d)\mapsto\tfrac{1}{\sqrt2}(c+d,\,c-d)$ and $Z(\alpha)=\mathrm{diag}(1,e^{i\alpha})$ . Why: the Hadamard falls out for free from the structure of the bond and the $X$ -basis half of the measurement — it is not something you chose, it is a fixed feature of the wire.

Step 4. Handle the other outcome: $s=1$ carries an extra relative sign, giving $\tfrac{1}{\sqrt2}X\,H\,Z(\alpha)|\psi\rangle$ , so both outcomes are summarized by $\tfrac{1}{\sqrt2}X^{s}HZ(\alpha)|\psi\rangle$ with each outcome equally likely, $p=\tfrac12$ . Why: the randomness is confined to a single known Pauli $X^{s}$ — everything unknown about the measurement is one classical bit, which is exactly what feedforward can absorb. *Projecting onto $|\pm_\alpha\rangle=\tfrac1{\sqrt2}(|0\rangle\pm e^{-i\alpha}|1\rangle)$ yields $\tfrac{1}{\sqrt2}X^{s}HZ(\alpha)|\psi\rangle$ exactly with $p=\tfrac12$ . Note that the sign of $\alpha$ inside $Z$ is a basis convention; the physics — a $Z$ -rotation then $H$ , up to a Pauli — is convention-free.*

Go deeper · Derivation B · 1 minGraph states are stabilizer states

Derivation BGraph states are stabilizer states

The goal is to prove that the graph state is the unique joint $+1$ eigenstate of the operators $K_i=X_i\prod_{j\in N(i)}Z_j$ , which is the fact that makes the entire resource preparable, checkable, and correctable with the stabilizer formalism rather than by tracking $2^{|V|}$ amplitudes. We do this by conjugating the trivial stabilizers of the unentangled product state through the entangling layer, using only the two Clifford identities for how $CZ$ transforms single-qubit Paulis.

Step 1. Before entangling, the product state $|+\rangle^{\otimes V}$ is stabilized by each bare $X_i$ : $X_i|+\rangle^{\otimes V}=|+\rangle^{\otimes V}$ , because $|+\rangle$ is the $+1$ eigenstate of its own $X$ . Why: the unentangled resource has an obvious stabilizer group, and stabilizers transform predictably under Clifford maps — so we can carry them forward instead of recomputing.

Step 2. Conjugate by the entangling layer $U_{CZ}=\prod_{(i,j)\in E}CZ_{ij}$ using $CZ_{ij}\,X_i\,CZ_{ij}=X_iZ_j$ and $CZ_{ij}\,Z\,CZ_{ij}=Z$ . Applying this across every neighbour of $i$ sends $X_i\mapsto X_i\prod_{j\in N(i)}Z_j=K_i$ . Why: $CZ$ is Clifford, so it maps Paulis to Paulis; the $X$ on a vertex sprouts a $Z$ on each bonded neighbour, which is exactly the graph-state generator.

Step 3. Assemble the eigenvalue: $K_i|G\rangle=U_{CZ}X_iU_{CZ}^{\dagger}\,U_{CZ}|+\rangle^{\otimes V}=U_{CZ}X_i|+\rangle^{\otimes V}=U_{CZ}|+\rangle^{\otimes V}=|G\rangle$ . The $|V|$ operators $K_i$ commute and are independent, so they generate an abelian group of order $2^{|V|}$ with a unique $+1$ eigenstate. Why: a full set of commuting independent stabilizers pins down one state — this uniqueness is what lets a verifier certify the resource by measuring $|V|$ commuting checks. *The two key $CZ$ Clifford conjugation relations are $CZ\,X\,CZ=XZ$ and $CZ\,Z\,CZ=Z$ ; together they imply the graph-state generators $K_i=X_i\prod_{j\in N(i)}Z_j$ directly.*

Go deeper · Derivation C · 1 minWhy the corrected angle is α′ = (−1)^s_X α + s_Z π

Derivation CWhy the corrected angle is α′ = (−1)^s_X α + s_Z π

The goal is to derive the feedforward rule that keeps MBQC deterministic: if a qubit arrives carrying an accumulated Pauli byproduct $\Sigma=X^{s_X}Z^{s_Z}$ from earlier random outcomes, measuring it at angle $\alpha$ is not what we want — we must instead measure at a shifted angle $\alpha'$ so that the intended gate $HZ(\alpha)$ is applied to the true logical state. We obtain $\alpha'$ by commuting the byproduct through the measurement basis, using how $X$ and $Z$ act on the rotated eigenstates $|\pm_\alpha\rangle$ .

Step 1. See how $X$ moves through the basis: $X|\pm_\alpha\rangle=e^{-i\alpha}|\pm_{-\alpha}\rangle$ , so an incoming $X$ -byproduct is equivalent to negating the measurement angle, $\alpha\mapsto-\alpha$ . Why: an unknown $X$ on the wire is harmless if we simply measure at $-\alpha$ instead — the sign flip $(-1)^{s_X}$ in the rule is nothing but this identity.

Step 2. See how $Z$ moves through the basis: $Z|\pm_\alpha\rangle=|\mp_\alpha\rangle$ , which relabels the two outcomes, and relabelling is identical to shifting the angle by $\pi$ since $|+_{\alpha+\pi}\rangle=|-_\alpha\rangle$ . Why: an incoming $Z$ -byproduct only swaps which outcome we call $s=0$ , and a $\pi$ shift of $\alpha$ undoes exactly that swap — hence the $s_Z\pi$ term.

Step 3. Compose the two: an inherited frame $\Sigma=X^{s_X}Z^{s_Z}$ requires the physical angle $\alpha'=(-1)^{s_X}\alpha+s_Z\pi$ , after which the intended $HZ(\alpha)$ acts on the correct state and only fresh byproducts remain to be propagated onward. Why: this is the rule that turns a genuinely random sequence of measurements into a deterministic unitary, and it is why non-Clifford angles cannot be deferred — their correction sits before the measurement, not after. *For $\alpha=\pi/4$ the feedforward rule gives $\alpha'=\pi/4,\,5\pi/4,\,-\pi/4,\,3\pi/4$ for $(s_X,s_Z)=(0,0),(0,1),(1,0),(1,1)$ respectively; each corrected angle reproduces the outcome statistics of measuring $\Sigma|\psi\rangle$ at $\alpha$ .*

Limiting cases

Special choices of angle and graph collapse the model onto familiar and instructive limits.

Case	Result	Meaning
Clifford angles only	$\alpha\in\{0,\tfrac{\pi}{2},\pi,\tfrac{3\pi}{2}\}$	Every measurement keeps the state stabilizer; by Gottesman–Knill the whole run is efficiently classically simulable. Non-Clifford angles are necessary for any quantum advantage.
1D chain resource	$G=\text{path graph}$	Only single-qubit unitaries are realizable; with no transverse bonds there is no entangling gate, so the model is classically tractable. Two-dimensional connectivity is required for universality.
Zero angle, $X$ basis	$\alpha=0\Rightarrow HZ(0)=H$	The qubit is teleported forward with a Hadamard and at most a Pauli byproduct — a bare wire that routes and reflects information across the lattice.
Magic angle	$\alpha=\tfrac{\pi}{4}\Rightarrow HZ(\tfrac{\pi}{4})=HT$	Supplies the $T$ -gate content; this one non-Clifford angle promotes the Clifford-only model to universality and is the locus of magic-state resource and gate teleportation.

Worked examples

1 · The T-gate by one measurement, and its correction (fully worked)

Show that the magic angle $\alpha=\pi/4$ realizes the non-Clifford gate $HT$ , and work out the corrected angle for every inherited byproduct.

Step 1. Identify the gate. With $Z(\pi/4)=\mathrm{diag}(1,e^{i\pi/4})=T$ , equation (4) gives the applied gate $HZ(\pi/4)=HT$ . Explicitly $HZ(\pi/4)=\tfrac{1}{\sqrt2}\begin{pmatrix}1&e^{i\pi/4}\\1&-e^{i\pi/4}\end{pmatrix}$ , which is exactly $HT$ — the single non-Clifford ingredient a universal set needs. So one measurement at π/4 applies HT, up to a Pauli.

Step 2. Account for randomness. The outcome is $s\in\{0,1\}$ with $p=\tfrac12$ each, leaving byproduct $X^{s}$ . On its own that is fine, but if this qubit sits downstream in a pattern it inherits a frame $\Sigma=X^{s_X}Z^{s_Z}$ from earlier outcomes, and the raw angle $\pi/4$ would then apply the wrong gate.

Step 3. Correct by feedforward, equation (5), $\alpha'=(-1)^{s_X}(\pi/4)+s_Z\pi$ . Enumerate the four inherited frames: $(s_X,s_Z)=(0,0)\Rightarrow\alpha'=\pi/4$ ; $(0,1)\Rightarrow\alpha'=5\pi/4$ ; $(1,0)\Rightarrow\alpha'=-\pi/4$ ; $(1,1)\Rightarrow\alpha'=3\pi/4$ . Dialling $\alpha'$ before pressing "measure" makes the net logical gate $HT$ deterministic regardless of the coins. The four corrected angles are π/4, 5π/4, −π/4, 3π/4 — and this correction, being non-Clifford, cannot be postponed.

2 · Building an arbitrary single-qubit rotation on a chain

Assemble a general single-qubit unitary out of consecutive measurements on a linear cluster, and read off how many qubits it costs.

Step 1. Each measured site applies $J(\alpha)\equiv HZ(\alpha)$ (equation 4). Two sites compose as $J(\beta)J(\alpha)=HZ(\beta)HZ(\alpha)$ ; note $J(0)=H$ , so measuring a site at $\alpha=0$ is a Hadamard wire, and two such sites give $H^2=I$ — pure teleportation with no logical rotation. A length- $(n{+}1)$ chain applies $n$ successive $J$ -gates.

Step 2. Use the Euler decomposition. Any $U\in SU(2)$ (up to global phase) is $U=HZ(\gamma)\,HZ(\beta)\,HZ(\alpha)\,HZ(0)$ — four $J$ -gates feeding one output. Since $HZ(\gamma)H=e^{i\gamma/2}R_x(\gamma)$ , interleaving $H$ -wires converts $z$ -rotations into $x$ -rotations, and three free angles $(\alpha,\beta,\gamma)$ sweep out the whole Bloch sphere.

Step 3. Count the cost and add feedforward. A universal single-qubit gate needs a 5-qubit chain (input + four measured), each non-initial angle corrected by the running byproduct frame via equation (5). Bring two such chains together with one vertical cluster rung and the resident $CZ$ becomes a logical entangling gate, completing a universal set $\{$ arbitrary single-qubit rotation, $CZ\}$ . Four measured qubits per single-qubit gate; one shared rung per entangling gate — all on a fixed 2D cluster.

Physics of the portrait

MBQC is where teleportation stopped being a curiosity and became an architecture — a computer built with no gates at all.

Origins · 1 minmeasurement-based quantum computation and how this was discovered

measurement-based quantum computation

the one-way model, Raussendorf and Briegel, 2001 · cluster-state quantum computing

The measurement-based model has an unusually sharp origin. In 1999 Daniel Gottesman and Isaac Chuang showed that a universal gate could be applied by teleporting a qubit through an entangled resource and correcting with Pauli operators — gates by teleportation — relocating the hard part of computation from unitary evolution to state preparation and measurement. In 2001 Robert Raussendorf and Hans Briegel took the idea to its limit: starting from a cluster state, a fixed lattice of qubits entangled by controlled- $Z$ , they showed that adaptively chosen single-qubit measurements alone suffice for universal computation. Because each measurement irreversibly consumes part of the resource, they named it the one-way quantum computer. Raussendorf, Dan Browne and Briegel gave the full framework in 2003, deriving the byproduct algebra and the universality proof. The model then acquired a rigorous grammar: Danos, Kashefi and Panangaden's measurement calculus supplied a standardization theorem for compiling patterns, and the flow and gflow conditions pinned down exactly when a pattern is deterministic. A trick for moving qubits became a complete, and in photonics preferred, way to compute.

History

1999

— Gottesman and Chuang realize universal gates by quantum teleportation

2001

— Raussendorf and Briegel propose the one-way quantum computer on a cluster state

2003

— Raussendorf, Browne and Briegel give the full cluster-state MBQC framework

2006

— Danos and Kashefi introduce causal flow and the measurement calculus

2007

— Browne, Kashefi, Mhalla and Perdrix define generalized flow (gflow) and determinism

Going further

Texts. Nielsen & Chuang, Quantum Computation and Quantum Information — teleportation and the stabilizer formalism · Preskill, Lecture Notes on Quantum Computation — cluster states and MBQC · Raussendorf & Briegel, "A One-Way Quantum Computer", Phys. Rev. Lett. 86, 5188 (2001) · Raussendorf, Browne & Briegel, "Measurement-based quantum computation on cluster states", Phys. Rev. A 68, 022312 (2003) · Danos, Kashefi & Panangaden, "The Measurement Calculus", J. ACM 54, 8 (2007) · Browne, Kashefi, Mhalla & Perdrix, "Generalized flow and determinism in MBQC", New J. Phys. 9, 250 (2007) · Watrous, The Theory of Quantum Information · Kitaev, Shen & Vyalyi, Classical and Quantum Computation. Key checkpoints in this article: (1) teleportation amplitudes give $\tfrac{1}{\sqrt2}X^{s}HZ(\alpha)|\psi\rangle$ with $p=\tfrac12$ ; (2) graph-state generators $K_i=X_i\prod_{j\in N(i)}Z_j$ follow from $CZ\,X\,CZ=XZ$ ; (3) the feedforward rule $\alpha'=(-1)^{s_X}\alpha+s_Z\pi$ is derived by commuting Paulis through $Z(\alpha)$ , with the four $\pi/4$ cases giving $\alpha'=\pi/4,\,5\pi/4,\,-\pi/4,\,3\pi/4$ ; (4) Clifford-angle patterns are classically simulable via Gottesman–Knill; (5) $HZ(\pi/4)=HT$ as $2\times2$ matrices. Angle-sign conventions vary between texts; the physics (a $Z$ -rotation then $H$ , up to Pauli) is convention-independent.

Prerequisites. Cluster States and Graph States · Quantum Teleportation · Stabilizer Formalism · Clifford Group and Pauli Operators. Leads to. One-Way Quantum Computer · Measurement Calculus and Pattern Compilation · Flow and Gflow Conditions · Magic State Distillation · Topological Quantum Computation.

Ask the tutor: Ask me to compile a small circuit — say a Bell-state preparation or a single $T$ -gate — into an explicit cluster-state measurement pattern, complete with the flow, the angle list, and the feedforward corrections.

If measurement destroys entanglement, how can it possibly compute?

Because destroying entanglement in the right basis is teleportation, and teleportation moves — and can rotate — a logical qubit. When you measure a bonded qubit in a tilted basis, the entanglement does not merely disappear: it channels the logical state onto the neighbour with a gate stamped on. So each measurement is a controlled trade: you spend one bond of entanglement to advance the computation by one gate. The resource is used up, which is why it is called "one-way", but every bit of entanglement spent buys a definite piece of logic. Building the entanglement first and spending it as gates is mathematically equivalent to applying those gates in a circuit.

Why does a 1D cluster state fail to be universal?

A linear chain only teleports a single logical qubit forward, applying single-qubit rotations $HZ(\alpha)$ as it goes. Universality also requires an entangling gate between two logical qubits, and that needs two wires to physically touch. In a cluster state, logical qubits entangle through the shared bonds where wires run alongside each other — a transverse rung of controlled- $Z$ . A path graph has no transverse bonds, so no two logical wires can ever interact, and a model of independent single-qubit rotations is classically simulable. Two-dimensional connectivity (or any non-planar graph with crossing wires) is the minimum needed to supply the missing entangling gate.

What exactly is the byproduct operator, and where does it come from?

It is the random Pauli left on the wire by the randomness of measurement. A projective measurement has two outcomes; in MBQC both apply the same intended gate but differ by a Pauli — typically $X$ or $Z$ — set by which outcome occurred. You cannot control the outcome, so you cannot control the byproduct, only record it. The accumulated byproducts form the "Pauli frame" the computation drifts in. Clifford gates permute the frame harmlessly, so those byproducts can be collected and applied at the very end; a non-Clifford angle must instead be pre-corrected via feedforward, because the byproduct would otherwise change which gate the measurement performs.

Why must some measurements wait for earlier outcomes?

Because feedforward corrections must be causal. The corrected angle $\alpha'=(-1)^{s_X}\alpha+s_Z\pi$ depends on the parities $s_X,s_Z$ of byproducts inherited from qubits measured earlier. If the pattern demanded that an angle depend on an outcome not yet obtained, the schedule would be impossible to run. The flow (and its generalization gflow) is exactly the geometric condition guaranteeing a measurement order in which every correction depends only on already-known outcomes. Clifford-angle qubits are order-insensitive, but each non-Clifford qubit imposes a genuine timing constraint on the ones feeding it.

Is MBQC actually faster or more powerful than a circuit?

No — and that is the point. MBQC decides the same class, BQP, with only polynomial overhead in either direction: any circuit compiles to a cluster-state pattern and vice versa. What changes is where the effort goes. The quantum-hard part — creating entanglement — is pushed into offline preparation of a generic resource, while the run itself is single-qubit measurement plus classical control. This is a practical gift for platforms where entangling gates are expensive but measurements cheap (notably photonics), and a conceptual gift for theory, because it isolates the non-Clifford "magic" as the sole source of quantum advantage.

How does MBQC connect to fault tolerance?

Through foliation. Stack copies of a 2D surface code in time and entangle them into a single 3D cluster state; measuring that cluster qubit-by-qubit both drives the logical computation and continuously extracts error syndromes as parities of outcomes. Logical gates are performed by carving defects — regions left unmeasured or measured differently — and braiding them. This is the Raussendorf–Harrington–Goyal topological cluster-state scheme, the direct ancestor of today's lattice-surgery approaches to fault-tolerant surface-code computing. So MBQC is a framework in which measurement-based error correction and computation become the same operations.