Simple Algorithm
Quantum Computing · Rydberg State · 26 min

Lüders Rule & Sequential Measurements

Measuring twice, measuring in turn, and disturbance.

Measure a spin along zz, then along xx, then along zz again: the last answer is a coin toss, even though the first one was certain. Sequences of measurements expose exactly how measurement disturbs, and the Lüders rule says precisely what survives.

What you’ll learn

  • Distinguish the von Neumann and Lüders update rules, and apply the Lüders instrument to degenerate measurements.
  • Prove that the Lüders instrument is the unique ideal measurement of a sharp observable.
  • Derive Wigner's formula for sequential outcome probabilities and explain order effects.
  • Prove Lüders' theorem: measuring AA leaves every statistic of BB unchanged iff [A,B]=0\comm{A}{B} = 0.
  • Quantify steering by measurement: NN projective steps carry ∣0⟩\ket0 to ∣1⟩\ket1 with probability cos⁡2Nπ2N≥1−π24N\cos^{2N}\tfrac{\pi}{2N} \ge 1 - \tfrac{\pi^2}{4N}.

Two update rules

For a non-degenerate observable there is nothing to decide: outcome aa leaves the eigenvector ∣a⟩\ket{a}. For a degenerate one, A=∑aaPaA = \sum_a a P_a with rank⁡Pa>1\rank P_a > 1, von Neumann (1932) imagined refining AA into a non-degenerate observable, so that the state ends up in some basis vector of the eigenspace, which depends on an arbitrary choice of refinement. Lüders (1951) proposed instead to simply project:

ρ  ⟼  PaρPaTr⁡(Paρ).\rho \;\longmapsto\; \frac{P_a\rho P_a}{\Tr(P_a\rho)}.

Ideal measurements are Lüders measurements

DerivationProof
  1. Ia(ρ)=∑jKaj ρ Kaj†,∑jKaj†Kaj=Pa\mathcal I_a(\rho) = \sum_j K_{aj}\,\rho\,K_{aj}^\dagger, \qquad \sum_j K_{aj}^\dagger K_{aj} = P_a

    Write each operation in Kraus form. Because the instrument measures AA, the Kraus operators of outcome aa add up to the projector PaP_a.

  2. Kaj(I−Pa)=0K_{aj}(\id - P_a) = 0

    For ∣φ⟩\ket\varphi orthogonal to the range of PaP_a, ∑j∥Kaj∣φ⟩∥2=⟨φ∣Pa∣φ⟩=0\sum_j\norm{K_{aj}\ket\varphi}^2 = \bra\varphi P_a\ket\varphi = 0, so every KajK_{aj} vanishes off the eigenspace.

  3. ∑jKaj∣ψ⟩ ⁣⟨ψ∣Kaj†=∣ψ⟩ ⁣⟨ψ∣  ⟹  Kaj∣ψ⟩=caj(ψ)∣ψ⟩\sum_j K_{aj}\dyad{\psi}K_{aj}^\dagger = \dyad{\psi} \;\Longrightarrow\; K_{aj}\ket\psi = c_{aj}(\psi)\ket\psi

    Apply ideality to a pure state ∣ψ⟩=Pa∣ψ⟩\ket\psi = P_a\ket\psi. A sum of positive rank-one operators equals a rank-one projector, so every term is proportional to it.

  4. Kaj=cajPa,  ∑j∣caj∣2=1  ⟹  Ia(ρ)=∑j∣caj∣2 PaρPa=PaρPaK_{aj} = c_{aj}P_a,\ \ \sum_j\abs{c_{aj}}^2 = 1 \;\Longrightarrow\; \mathcal I_a(\rho) = \sum_j\abs{c_{aj}}^2\,P_a\rho P_a = P_a\rho P_a

    If every vector of a subspace is an eigenvector of KajK_{aj}, the eigenvalue cannot depend on the vector (compare ∣u⟩\ket u, ∣v⟩\ket v and ∣u⟩+∣v⟩\ket u + \ket v). So KajK_{aj} acts on the eigenspace as a scalar.

  5. Tr⁡(Paρ)=1⇒PaρPa=ρ,Tr⁡(Pa PaρPa)Tr⁡(Paρ)=1■\Tr(P_a\rho) = 1 \Rightarrow P_a\rho P_a = \rho, \qquad \frac{\Tr\big(P_a\,P_a\rho P_a\big)}{\Tr(P_a\rho)} = 1 \qquad\blacksquare

    Conversely, a state with a certain outcome is supported on the range of PaP_a, so the Lüders instrument leaves it alone (ideal), and Pa2=PaP_a^2 = P_a makes it repeatable.

Sequences and order effects

Chain Lüders updates, and the probability of a whole sequence of outcomes becomes a single trace, often called Wigner's formula:

p(a1,…,an)=Tr⁡(Pan(n)⋯Pa1(1) ρ Pa1(1)⋯Pan(n))p(a_1,\dots,a_n) = \Tr\Big(P^{(n)}_{a_n}\cdots P^{(1)}_{a_1}\,\rho\,P^{(1)}_{a_1}\cdots P^{(n)}_{a_n}\Big)
The normalizations p(a1)p(a_1), p(a2 ∣ a1)p(a_2\,|\,a_1), … telescope away. The product Pa1(1)⋯Pan(n)P^{(1)}_{a_1}\cdots P^{(n)}_{a_n} is in general not a projector, which is why sequential statistics are not those of any single projective measurement.
DerivationTwo measurements, two asymmetries
  1. p(a,b)=Tr⁡(QbPaρPaQb)=Tr⁡(PaQbPa ρ)p(a, b) = \Tr\big(Q_bP_a\rho P_aQ_b\big) = \Tr\big(P_aQ_bP_a\,\rho\big)

    Measure A=∑aaPaA = \sum_a aP_a, then B=∑bbQbB = \sum_b bQ_b, both with Lüders updates.

  2. ∑bp(a,b)=Tr⁡(PaρPa)=Tr⁡(Paρ)\sum_b p(a,b) = \Tr\big(P_a\rho P_a\big) = \Tr(P_a\rho)

    The future does not disturb the past: summing over the later outcome always restores the earlier statistics.

  3. ∑ap(a,b)=Tr⁡(Qb ΛA(ρ))  ≠  Tr⁡(Qbρ)in general\sum_a p(a,b) = \Tr\big(Q_b\,\Lambda_A(\rho)\big) \;\neq\; \Tr(Q_b\rho)\quad\text{in general}

    The past does disturb the future: an earlier measurement, even with its outcome forgotten, changes later statistics.

When measuring one observable leaves another alone

DerivationProof
  1. Tr⁡(QbΛA(ρ))=Tr⁡(ΛA(Qb) ρ)  ∀ρ  ⟺  ∑aPaQbPa=Qb\Tr\big(Q_b\Lambda_A(\rho)\big) = \Tr\big(\Lambda_A(Q_b)\,\rho\big)\ \ \forall\rho \iff \sum_a P_aQ_bP_a = Q_b

    ΛA\Lambda_A is self-adjoint for the Hilbert–Schmidt inner product, so the condition says that each QbQ_b is a fixed point of the pinching.

  2. PcQb=PcQbPc=QbPcP_cQ_b = P_cQ_bP_c = Q_bP_c

    Multiply the fixed-point equation by PcP_c on the left, and separately on the right. Orthogonality, PcPa=δcaPcP_cP_a = \delta_{ca}P_c, collapses each sum to a single term.

  3. [Pa,Qb]=0  ∀a,b  ⟺  [A,B]=0■\comm{P_a}{Q_b} = 0\ \ \forall a,b \iff \comm{A}{B} = 0 \qquad\blacksquare

    Every QbQ_b commutes with every PaP_a, and spectral projectors commute exactly when the observables do. Conversely, commuting projectors give ∑aPaQbPa=Qb∑aPa=Qb\sum_a P_aQ_bP_a = Q_b\sum_a P_a = Q_b.

In finite dimensions the same conclusion holds for unsharp Lüders operations ρ↦∑kEk ρEk\rho\mapsto\sum_k\sqrt{E_k}\,\rho\sqrt{E_k}, because the fixed points of a unital channel are exactly the operators that commute with all of its Kraus operators. In infinite dimensions this can fail.

Check your understanding

On a qubit, you measure XX (Lüders update, outcome forgotten) and then ZZ. For which initial states are the ZZ statistics the same as without the XX measurement?

  1. All states, because the XX outcome is forgotten
  2. Only ∣+⟩\ket+ and ∣−⟩\ket-
  3. Exactly the states with ⟨Z⟩=0\expval{Z} = 0
  4. No states at all, since XX and ZZ do not commute
Show the answer

C. Exactly the states with ⟨Z⟩=0\expval{Z} = 0

ΛX\Lambda_X maps the Bloch vector (rx,ry,rz)(r_x, r_y, r_z) to (rx,0,0)(r_x, 0, 0), so afterwards ⟨Z⟩=0\expval{Z} = 0. The ZZ statistics are unchanged exactly when rzr_z was already 00: the whole xx–yy disk, not just ∣±⟩\ket\pm. Lüders' theorem only guarantees that some state is disturbed when [X,Z]≠0\comm{X}{Z}\neq0.

Steering by measurement

Disturbance can be put to work. Measure a qubit NN times along axes that turn in equal steps from +z+z to −z-z. Whenever every outcome "passes", the state has been dragged from ∣0⟩\ket0 to ∣1⟩\ket1 without any Hamiltonian at all.

DerivationDragging a state with projective measurements
  1. ∣θk⟩=cos⁡θk2∣0⟩+sin⁡θk2∣1⟩,θk=kπN,k=1,…,N\ket{\theta_k} = \cos\tfrac{\theta_k}{2}\ket0 + \sin\tfrac{\theta_k}{2}\ket1, \qquad \theta_k = \frac{k\pi}{N},\quad k = 1,\dots,N

    Measure {∣θk⟩ ⁣⟨θk∣, I−∣θk⟩ ⁣⟨θk∣}\{\dyad{\theta_k},\ \id-\dyad{\theta_k}\} in turn, starting from ∣θ0⟩=∣0⟩\ket{\theta_0} = \ket0.

  2. ∣⟨θk−1∣θk⟩∣2=cos⁡2π2N\abs{\braket{\theta_{k-1}|\theta_k}}^2 = \cos^2\frac{\pi}{2N}

    Neighboring states differ by a Bloch angle π/N\pi/N, which is half of that in Hilbert space.

  3. Ppass(N)=cos⁡2Nπ2N  ≥  (1−π24N2)N  ≥  1−π24NP_{\text{pass}}(N) = \cos^{2N}\frac{\pi}{2N} \;\ge\; \Big(1-\frac{\pi^2}{4N^2}\Big)^{N} \;\ge\; 1 - \frac{\pi^2}{4N}

    Using cos⁡2x≥1−x2\cos^2x \ge 1 - x^2 and Bernoulli's inequality. The failure probability falls like 1/N1/N.

Predict before you look

With N=10N = 10 dragging measurements, what is the probability, in percent, that every one of them passes?

Show the answer

78 %

cos⁡20(π/20)=cos⁡20(9∘)≈0.781\cos^{20}(\pi/20) = \cos^{20}(9^\circ) \approx 0.781, about 78%. Compare N=1N=1: 00, since ∣0⟩⊥∣1⟩\ket0\perp\ket1; N=2N=2: 14\tfrac14, the three-polarizer effect; N=100N=100: about 97.6%. The bound 1−π2/4N1-\pi^2/4N gives 75% for N=10N=10.

This is the measurement-only cousin of the quantum Zeno effect: frequent measurement along a fixed axis freezes a slowly evolving state, while measurement along a slowly moving axis drags the state along. Weak measurement and the Zeno effect are developed in the lessons on generalized measurements.

Key takeaways

  • The Lüders rule ρ↦PaρPa/p(a)\rho\mapsto P_a\rho P_a/p(a) keeps coherence inside degenerate eigenspaces. It is the unique ideal measurement of a sharp observable, and it is repeatable.
  • Sequential probabilities follow Wigner's formula. Later measurements never change earlier statistics, but earlier ones generally change later ones, so order matters.
  • Lüders' theorem: measuring AA leaves all statistics of BB unchanged, for every state, iff [A,B]=0\comm{A}{B} = 0.
  • Repeated measurements can steer a state: NN dragging steps succeed with probability cos⁡2N(π/2N)≥1−π2/4N\cos^{2N}(\pi/2N) \ge 1 - \pi^2/4N.

Check yourself

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

Question 1

A qutrit in 12(∣1⟩+∣2⟩)\tfrac{1}{\sqrt2}(\ket1+\ket2) undergoes the Lüders measurement {∣0⟩ ⁣⟨0∣, ∣1⟩ ⁣⟨1∣+∣2⟩ ⁣⟨2∣}\{\dyad0,\ \dyad1+\dyad2\}. What is the state after the second outcome (which is certain)?

  1. Unchanged: 12(∣1⟩+∣2⟩)\tfrac{1}{\sqrt2}(\ket1+\ket2)
  2. ∣1⟩\ket1 or ∣2⟩\ket2, each with probability 12\tfrac12
  3. The mixture 12(∣1⟩ ⁣⟨1∣+∣2⟩ ⁣⟨2∣)\tfrac12(\dyad1+\dyad2)
  4. ∣0⟩\ket0
Show the answer

A. Unchanged: 12(∣1⟩+∣2⟩)\tfrac{1}{\sqrt2}(\ket1+\ket2)

The state lies in the range of P=∣1⟩ ⁣⟨1∣+∣2⟩ ⁣⟨2∣P = \dyad1+\dyad2, so PρP=ρP\rho P = \rho. The Lüders rule is ideal and leaves it alone.

Question 2

Start in ∣0⟩\ket0. Measure XX, then ZZ, both projectively with Lüders updates. What is p(X=−, Z=1)p(X = -,\ Z = 1)?

  1. 00
  2. 12\tfrac12
  3. 18\tfrac18
  4. 14\tfrac14
Show the answer

D. 14\tfrac14

p=∣⟨−∣0⟩∣2 ∣⟨1∣−⟩∣2=12⋅12=14p = \abs{\braket{-|0}}^2\,\abs{\braket{1|-}}^2 = \tfrac12\cdot\tfrac12 = \tfrac14. Without the XX measurement, Z=1Z = 1 would be impossible.

Question 3

Which pair of two-qubit observables can be measured one after the other, in either order and for every state, without changing each other's statistics?

  1. Z⊗IZ\otimes\id and X⊗IX\otimes\id
  2. Z⊗ZZ\otimes Z and X⊗XX\otimes X
  3. Z⊗ZZ\otimes Z and X⊗IX\otimes\id
  4. X⊗IX\otimes\id and Y⊗IY\otimes\id
Show the answer

B. Z⊗ZZ\otimes Z and X⊗XX\otimes X

By Lüders' theorem the criterion is commutation. Z⊗ZZ\otimes Z and X⊗XX\otimes X anticommute on both tensor factors, so the signs cancel and they commute.

Question 4

What singles out the Lüders instrument among all instruments that measure a sharp observable AA?

  1. It is the only repeatable one
  2. It is the only one with a single Kraus operator per outcome
  3. It is the only ideal one: it leaves unchanged every state whose outcome was already certain
  4. It is the only trace-preserving one
Show the answer

C. It is the only ideal one: it leaves unchanged every state whose outcome was already certain

Ideal (minimally disturbing) instruments for a sharp observable are exactly the Lüders instrument. Repeatability alone is weaker: fixed-vector and von Neumann-refinement instruments are repeatable but not ideal.

Further reading

  • G. Lüders, "Über die Zustandsänderung durch den Meßprozeß", Annalen der Physik (1951); English translation by K. A. Kirkpatrick, "Concerning the state-change due to the measurement process", Annalen der Physik (2006).
  • J. von Neumann, Mathematische Grundlagen der Quantenmechanik (Springer, 1932); English translation, Mathematical Foundations of Quantum Mechanics (Princeton, 1955).
  • E. P. Wigner, "The problem of measurement", American Journal of Physics 31 (1963).
  • T. Heinosaari & M. Ziman, The Mathematical Language of Quantum Theory: From Uncertainty to Entanglement (Cambridge, 2012).
  • B. Skinner, J. Ruhman & A. Nahum, "Measurement-induced phase transitions in the dynamics of entanglement", Physical Review X 9 (2019).