Simple Algorithm
Quantum Computing · Excited State · 22 min

The Born Rule & Projective Measurement

Probabilities from amplitudes, and the post-measurement state.

Every quantum experiment ends with a detector click. The measurement postulate says which clicks are possible, how likely each one is, and what state is left behind. Those three rules drive everything from reading out an algorithm's answer to error correction.

What you’ll learn

  • State the projective measurement postulate: probabilities ⟨ψ∣Pm∣ψ⟩\bra{\psi}P_m\ket{\psi} and post-measurement states Pm∣ψ⟩/p(m)P_m\ket{\psi}/\sqrt{p(m)}.
  • Prove that the rule yields a valid, phase-independent and repeatable probability distribution.
  • Compute probabilities and post-measurement states when only part of a multi-qubit system is measured.
  • Prove that non-orthogonal states cannot be distinguished with certainty.
  • Estimate Born probabilities from a finite number of shots, with error bars.

The measurement postulate

A projective measurement is specified by its possible outcomes mm and, for each one, a projector PmP_m onto the subspace of states that "give mm". The projectors must be mutually orthogonal and must cover the whole Hilbert space:

Pm†=Pm,PmPn=δmnPm,∑mPm=I.P_m^\dagger = P_m, \qquad P_m P_n = \delta_{mn} P_m, \qquad \sum_m P_m = \id.

The most common case is a measurement in an orthonormal basis {∣m⟩}\{\ket{m}\}. Then Pm=∣m⟩ ⁣⟨m∣P_m = \dyad{m}, the Born rule reads p(m)=∣⟨m∣ψ⟩∣2p(m) = \abs{\braket{m|\psi}}^2, and the state after the measurement is ∣m⟩\ket{m} (up to a phase).

Projectors of rank greater than one are just as legal. Then outcome mm only tells you which subspace the state is in, and the post-measurement state keeps any superposition inside that subspace.

Check your understanding

Continuing the example: you now measure ∣ψb⟩\ket{\psi_b} in the full basis {∣0⟩,∣1⟩,∣2⟩}\{\ket0,\ket1,\ket2\}. What are the outcome probabilities?

  1. 13\tfrac13 each
  2. 00, 12\tfrac12, 12\tfrac12
  3. Either 0,1,00, 1, 0 or 0,0,10, 0, 1, depending on what the first measurement secretly picked
  4. 00, 23\tfrac23, 13\tfrac13
Show the answer

B. 00, 12\tfrac12, 12\tfrac12

∣ψb⟩=12(∣1⟩+∣2⟩)\ket{\psi_b} = \tfrac{1}{\sqrt2}(\ket1+\ket2), so the probabilities are 00, 12\tfrac12, 12\tfrac12. The coarse measurement removed the ∣0⟩\ket0 component but did not decide between ∣1⟩\ket1 and ∣2⟩\ket2.

Why the rule is consistent

The postulate is short, but it must deliver a genuine probability distribution that ignores unobservable phases and agrees with itself when a measurement is repeated. Each property takes one line.

DerivationFour properties in four lines
  1. p(m)=⟨ψ∣Pm∣ψ⟩=⟨ψ∣Pm†Pm∣ψ⟩=∥Pm∣ψ⟩∥2≥0p(m) = \bra{\psi}P_m\ket{\psi} = \bra{\psi}P_m^\dagger P_m\ket{\psi} = \norm{P_m\ket{\psi}}^2 \ge 0

    Positivity. A projector satisfies Pm=Pm†=Pm2P_m = P_m^\dagger = P_m^2.

  2. ∑mp(m)=⟨ψ∣(∑mPm)∣ψ⟩=⟨ψ∣ψ⟩=1\sum_m p(m) = \bra{\psi}\Big(\sum_m P_m\Big)\ket{\psi} = \braket{\psi|\psi} = 1

    Normalization follows from completeness, ∑mPm=I\sum_m P_m = \id.

  3. (eiγ∣ψ⟩)†Pm(eiγ∣ψ⟩)=e−iγeiγ⟨ψ∣Pm∣ψ⟩=⟨ψ∣Pm∣ψ⟩\big(e^{i\gamma}\ket{\psi}\big)^\dagger P_m \big(e^{i\gamma}\ket{\psi}\big) = e^{-i\gamma}e^{i\gamma}\bra{\psi}P_m\ket{\psi} = \bra{\psi}P_m\ket{\psi}

    Phase independence. A global phase cancels between bra and ket, so ∣ψ⟩\ket\psi and eiγ∣ψ⟩e^{i\gamma}\ket\psi have identical statistics.

  4. p(m′ ∣ m)=⟨ψm∣Pm′∣ψm⟩=⟨ψ∣PmPm′Pm∣ψ⟩p(m)=δmm′p(m'\,|\,m) = \bra{\psi_m}P_{m'}\ket{\psi_m} = \frac{\bra{\psi}P_m P_{m'} P_m\ket{\psi}}{p(m)} = \delta_{mm'}

    Repeatability. An immediate second measurement confirms the first with certainty, because PmPm′Pm=δmm′PmP_mP_{m'}P_m = \delta_{mm'}P_m.

From probabilities to shot counts

You never observe p(m)p(m) directly, only frequencies. Repeat a two-outcome measurement on NN fresh copies and the number n1n_1 of outcomes 11 is binomially distributed. The estimate p^=n1/N\hat p = n_1/N is unbiased, with standard error

sd⁡(p^)=p(1−p)N.\operatorname{sd}(\hat p) = \sqrt{\frac{p(1-p)}{N}}.

The error shrinks only like 1/N1/\sqrt{N}: one more decimal digit of precision costs a hundred times more shots.

Predict before you look

You measure a state with p(1)=0.3p(1) = 0.3 a total of N=1000N = 1000 times. The count of 11s fluctuates around 300300. What is its standard deviation?

Show the answer

14.5 counts

The count is binomial with variance Np(1−p)=1000×0.3×0.7=210Np(1-p) = 1000 \times 0.3 \times 0.7 = 210, so its standard deviation is 210≈14.5\sqrt{210} \approx 14.5. About 95% of runs land within 300±29300 \pm 29.

Measuring part of a system

Measuring only qubit 0 of a two-qubit register is a projective measurement with two rank-2 projectors, ∣i⟩ ⁣⟨i∣⊗I\dyad{i}\otimes\id for i∈{0,1}i \in \{0, 1\}. Write the state as ∣ψ⟩=∑i,jcij∣i⟩∣j⟩\ket{\psi} = \sum_{i,j} c_{ij}\ket{i}\ket{j}.

DerivationPartial measurement
  1. p(i)=⟨ψ∣(∣i⟩ ⁣⟨i∣⊗I)∣ψ⟩=∑j∣cij∣2p(i) = \bra{\psi}\big(\dyad{i}\otimes\id\big)\ket{\psi} = \sum_j \abs{c_{ij}}^2

    Add up the Born probabilities of every bit string that starts with ii.

  2. ∣ψi⟩=(∣i⟩ ⁣⟨i∣⊗I)∣ψ⟩p(i)=∣i⟩⊗∑jcij∣j⟩p(i)\ket{\psi_i} = \frac{(\dyad{i}\otimes\id)\ket{\psi}}{\sqrt{p(i)}} = \ket{i}\otimes\frac{\sum_j c_{ij}\ket{j}}{\sqrt{p(i)}}

    The unmeasured qubit is left in a pure state: the normalized "row ii" of the amplitude table cijc_{ij}.

  3. ∣ψ⟩=12∣00⟩+12∣01⟩+12∣11⟩  ⟹  {p(0)=12,∣ψ0⟩=∣0⟩∣+⟩,p(1)=12,∣ψ1⟩=∣1⟩∣1⟩.\ket\psi = \tfrac12\ket{00}+\tfrac12\ket{01}+\tfrac{1}{\sqrt2}\ket{11} \;\Longrightarrow\; \begin{cases} p(0)=\tfrac12, & \ket{\psi_0}=\ket{0}\ket{+},\\[2pt] p(1)=\tfrac12, & \ket{\psi_1}=\ket{1}\ket{1}.\end{cases}

    A worked example, which the circuit below lets you test.

One look cannot tell overlapping states apart

Someone hands you either ∣ψ⟩\ket\psi or ∣ϕ⟩\ket\phi, and you must say which. If ⟨ψ∣ϕ⟩=0\braket{\psi|\phi} = 0, measuring {∣ψ⟩ ⁣⟨ψ∣, I−∣ψ⟩ ⁣⟨ψ∣}\{\dyad{\psi},\ \id-\dyad{\psi}\} settles it with certainty. If the two states overlap at all, nothing does.

How well can you do? For two equally likely pure states the optimum is 12(1+1−∣⟨ψ∣ϕ⟩∣2)\tfrac12\big(1+\sqrt{1-\abs{\braket{\psi|\phi}}^2}\big), the Helstrom bound; for ∣0⟩\ket0 versus ∣+⟩\ket+ it is cos⁡2π8≈85%\cos^2\tfrac{\pi}{8}\approx 85\%. This is why quantum key distribution works: an eavesdropper facing non-orthogonal signals cannot read them without sometimes getting them wrong, and her errors betray her.

Check your understanding

You receive ∣0⟩\ket0 or ∣+⟩\ket+ with equal probability. You measure in the ZZ basis and guess "∣0⟩\ket0" on outcome 00, "∣+⟩\ket+" on outcome 11. What is your probability of guessing correctly?

  1. 12\tfrac12
  2. 34\tfrac34
  3. cos⁡2π8≈0.854\cos^2\tfrac{\pi}{8}\approx 0.854
  4. 11
Show the answer

B. 34\tfrac34

If the state is ∣0⟩\ket0 you always see 00 and guess right. If it is ∣+⟩\ket+ you see 11 only half the time. Overall: 12⋅1+12⋅12=34\tfrac12\cdot 1 + \tfrac12\cdot\tfrac12 = \tfrac34. A measurement in a basis tilted symmetrically between the two states reaches the optimum cos⁡2π8\cos^2\tfrac{\pi}{8}.

Questions the postulate leaves open

The postulate tells you how to calculate. It is silent on three deeper questions, and each has its own Rydberg-level lesson.

Think first Is collapse a physical process, or just an update of your knowledge?

The postulate does not say, and no experiment so far has told the readings apart. If ∣ψ⟩\ket\psi is a state of knowledge, collapse is unremarkable, but you must say whose knowledge and knowledge of what. If ∣ψ⟩\ket\psi is physical, collapse is a real process that the Schrödinger equation does not contain. That tension is the measurement problem.

Key takeaways

  • A projective measurement is a complete set of orthogonal projectors {Pm}\{P_m\}. Outcome mm occurs with probability ⟨ψ∣Pm∣ψ⟩\bra\psi P_m\ket\psi and leaves the state Pm∣ψ⟩/p(m)P_m\ket\psi/\sqrt{p(m)}.
  • The rule is positive, normalized, phase-independent and repeatable. Projectors of rank greater than one preserve superpositions inside their subspace.
  • Measuring one subsystem uses Pi⊗IP_i\otimes\id. The rest is left in the normalized conditional state, and local operations elsewhere cannot change the statistics.
  • Only orthogonal states can be distinguished with certainty; for overlapping states every strategy sometimes fails.
  • Probabilities are estimated from shots, with standard error p(1−p)/N\sqrt{p(1-p)/N}.

Check yourself

In the interactive lesson these are scored with your confidence, and what you miss comes back for review.

Question 1

For ∣ψ⟩=15(∣0⟩+2i∣1⟩)\ket\psi = \tfrac{1}{\sqrt5}(\ket0 + 2i\ket1), what is the probability of outcome 11 in the computational basis?

  1. 25\tfrac{2}{\sqrt5}
  2. 45\tfrac45
  3. −45-\tfrac45
  4. 15\tfrac15
Show the answer

B. 45\tfrac45

p(1)=∣⟨1∣ψ⟩∣2=∣2i/5∣2=45p(1) = \abs{\braket{1|\psi}}^2 = \abs{2i/\sqrt5}^2 = \tfrac45.

Question 2

You measure qubit 0 of 13(∣00⟩+∣01⟩+∣10⟩)\tfrac{1}{\sqrt3}(\ket{00}+\ket{01}+\ket{10}) and get 00. What was the probability of this outcome, and what is the state afterward?

  1. 13\tfrac13; then ∣00⟩\ket{00}
  2. 12\tfrac12; then ∣0⟩∣+⟩\ket{0}\ket{+}
  3. 23\tfrac23; then ∣0⟩∣+⟩\ket{0}\ket{+}
  4. 23\tfrac23; then ∣00⟩\ket{00} or ∣01⟩\ket{01} at random
Show the answer

C. 23\tfrac23; then ∣0⟩∣+⟩\ket{0}\ket{+}

p(0)=∣c00∣2+∣c01∣2=23p(0) = \abs{c_{00}}^2+\abs{c_{01}}^2 = \tfrac23, and the post-measurement state is (∣00⟩+∣01⟩)/2=∣0⟩∣+⟩(\ket{00}+\ket{01})/\sqrt2 = \ket0\ket+. Qubit 1 is left in a superposition, not in a randomly chosen basis state.

Question 3

Which of these is a valid projective measurement on a qubit?

  1. {∣+⟩ ⁣⟨+∣, ∣−⟩ ⁣⟨−∣}\{\dyad{+},\ \dyad{-}\}
  2. {∣0⟩ ⁣⟨0∣, ∣+⟩ ⁣⟨+∣}\{\dyad{0},\ \dyad{+}\}
  3. {12I, 12I}\{\tfrac12\id,\ \tfrac12\id\}
  4. {∣0⟩ ⁣⟨0∣, ∣1⟩ ⁣⟨1∣, ∣+⟩ ⁣⟨+∣}\{\dyad{0},\ \dyad{1},\ \dyad{+}\}
Show the answer

A. {∣+⟩ ⁣⟨+∣, ∣−⟩ ⁣⟨−∣}\{\dyad{+},\ \dyad{-}\}

∣+⟩\ket+ and ∣−⟩\ket- are orthonormal, so their projectors are orthogonal and sum to I\id.

Question 4

Which pair of states can be distinguished with certainty by a single measurement?

  1. ∣0⟩\ket0 and ∣+⟩\ket+
  2. ∣+⟩\ket+ and ∣+i⟩\ket{+i}
  3. ∣0⟩\ket0 and eiπ/3∣0⟩e^{i\pi/3}\ket0
  4. ∣+i⟩\ket{+i} and ∣−i⟩\ket{-i}
Show the answer

D. ∣+i⟩\ket{+i} and ∣−i⟩\ket{-i}

⟨+i∣−i⟩=0\braket{+i|-i} = 0: the states are orthogonal, so a YY-basis measurement tells them apart every time.

Go deeper

Further reading

  • M. A. Nielsen & I. L. Chuang, Quantum Computation and Quantum Information, §2.2.3–2.2.5 and Box 2.3 (Cambridge, 10th anniv. ed., 2010).
  • J. Preskill, Caltech Ph219/CS219 lecture notes, Ch. 2 "Foundations I: States and Ensembles" and Ch. 3 "Foundations II: Measurement and Evolution".
  • M. Born, "Zur Quantenmechanik der Stoßvorgänge", Zeitschrift für Physik 37 (1926).