Simple Algorithm
Quantum Computing · Rydberg State · 28 min

Weak Measurement & the Zeno Effect

Gentle measurements, weak values, and freezing evolution by watching.

Measure gently enough and you learn a little while disturbing even less. Push that idea to its limits and you get continuous monitoring of single quantum systems, "weak values" far outside an observable's spectrum, and the quantum Zeno effect: a watched qubit that never flips.

What you’ll learn

  • Model an unsharp measurement with Kraus operators and quantify its information–disturbance trade-off.
  • Derive the weak value from a von Neumann pointer model, and bound its size by the post-selection probability.
  • Derive the quantum Zeno effect for repeated projective measurements and for continuous dephasing, Γeff≈Ω2/γ\Gamma_{\text{eff}}\approx\Omega^2/\gamma.
  • Prove the gentle measurement lemma and connect it to modern uses.

Unsharp measurements

Between not looking at all and a projective measurement lies a continuum of measurements of adjustable strength. The simplest family measures ZZ on a qubit with sharpness η\eta.

DerivationSignal versus disturbance
  1. p±=Tr⁡(E±ρ)=12(1±η rz)p_\pm = \Tr(E_\pm\rho) = \tfrac12\big(1\pm\eta\,r_z\big)

    The outcome bias is p+−p−=ηrzp_+-p_- = \eta r_z: the signal is first order in η\eta.

  2. ∑±M±ρM±:ρ01 ↦ 21+η21−η2  ρ01=1−η2  ρ01\sum_\pm M_\pm\rho M_\pm:\qquad \rho_{01}\ \mapsto\ 2\sqrt{\tfrac{1+\eta}{2}}\sqrt{\tfrac{1-\eta}{2}}\;\rho_{01} = \sqrt{1-\eta^2}\;\rho_{01}

    Averaged over outcomes, the populations are untouched while the coherence, i.e. the transverse part of the Bloch vector, shrinks by 1−η2\sqrt{1-\eta^2}.

  3. 1−1−η2=η22+O(η4),Frz=η21−η2rz21-\sqrt{1-\eta^2} = \tfrac{\eta^2}{2}+\order{\eta^4},\qquad \mathcal F_{r_z} = \frac{\eta^2}{1-\eta^2r_z^2}

    The disturbance is only second order. But so is the information: the Fisher information about rzr_z per shot is Frz≈η2\mathcal F_{r_z}\approx\eta^2. Weak measurements trade speed for gentleness; they do not evade the information–disturbance trade-off.

Repeat the weak measurement NN times on the same qubit. The net count of ++ versus −- outcomes performs a biased random walk, and once Nη2≫1N\eta^2\gg1 the conditional state has collapsed onto ∣0⟩\ket0 or ∣1⟩\ket1, with Born-rule probabilities. In the continuum limit this becomes continuous measurement:

dρ=−i[H,ρ] dt+κ D[A]ρ dt+κ H[A]ρ dW,dy=2κ ⟨A⟩ dt+dWd\rho = -i[H,\rho]\,dt+\kappa\,\mathcal D[A]\rho\,dt+\sqrt\kappa\,\mathcal H[A]\rho\,dW,\qquad dy = 2\sqrt\kappa\,\expval{A}\,dt+dW
The diffusive stochastic master equation for monitoring a Hermitian observable AA at rate κ\kappa with unit efficiency, in the conventions of Wiseman & Milburn: D[A]ρ=AρA−12{A2,ρ}\mathcal D[A]\rho = A\rho A-\tfrac12\acomm{A^2}{\rho}, H[A]ρ=Aρ+ρA−2⟨A⟩ρ\mathcal H[A]\rho = A\rho+\rho A-2\expval{A}\rho, dWdW a Wiener increment and dydy the measured signal. Averaging over dWdW leaves the Lindblad dephasing equation.

Weak values

Aharonov, Albert and Vaidman (1988) asked: what does a weak measurement of AA report if you keep only the runs in which a later, strong measurement finds the system in ∣ϕ⟩\ket\phi? Model the meter as a pointer with position qq and momentum pp, prepared in a wavepacket of width σ\sigma centered at q=0q = 0, and coupled by e−igA⊗pe^{-igA\otimes p}, which shifts the pointer by gaga on each eigenstate ∣a⟩\ket a of AA (with ℏ=1\hbar = 1).

DerivationThe weak value
  1. ⟨ϕ∣e−igA⊗p∣ψ⟩ ≈ ⟨ϕ∣ψ⟩(1−ig Aw p) ≈ ⟨ϕ∣ψ⟩ e−igAwp\bra\phi e^{-igA\otimes p}\ket\psi\ \approx\ \braket{\phi|\psi}\big(1-ig\,A_w\,p\big)\ \approx\ \braket{\phi|\psi}\,e^{-igA_wp}

    Project the system onto the post-selected ∣ϕ⟩\ket\phi and expand to first order in gg. This is valid when the induced shifts are small compared with the pointer width σ\sigma.

  2. Aw:=⟨ϕ∣A∣ψ⟩⟨ϕ∣ψ⟩A_w := \frac{\bra\phi A\ket\psi}{\braket{\phi|\psi}}

    This is the weak value. In general it is complex.

  3. ⟨q⟩post-selected ≈ g Re⁡Aw\expval{q}_{\text{post-selected}}\ \approx\ g\,\operatorname{Re}A_w

    The factor e−igRe⁡(Aw)pe^{-ig\operatorname{Re}(A_w)p} translates the post-selected pointer by gRe⁡Awg\operatorname{Re}A_w. The imaginary part instead shifts the pointer momentum, by an amount proportional to its momentum variance (Jozsa 2007).

Because the denominator can be small, AwA_w can lie far outside the spectrum of AA. For A=ZA = Z, pre-selection ∣ψ⟩=∣+⟩\ket\psi = \ket+ and post-selection ∣ϕ⟩=cos⁡β∣0⟩−sin⁡β∣1⟩\ket\phi = \cos\beta\ket0-\sin\beta\ket1, one finds Zw=tan⁡(π4+β)Z_w = \tan\big(\tfrac\pi4+\beta\big), which diverges as ∣ϕ⟩→∣−⟩\ket\phi\to\ket-. Averaging AwA_w over a complete set of post-selections, weighted by their probabilities ∣⟨ϕ∣ψ⟩∣2\abs{\braket{\phi|\psi}}^2, gives back ⟨ψ∣A∣ψ⟩\bra\psi A\ket\psi.

Check your understanding

In the example above, how small must the post-selection probability be to obtain Zw=100Z_w = 100?

  1. About 10−110^{-1}
  2. About 10−210^{-2}
  3. About 10−410^{-4}
  4. About 10−810^{-8}
Show the answer

C. About 10−410^{-4}

Write β=π4−δ\beta = \tfrac\pi4-\delta. Then Zw=cot⁡δZ_w = \cot\delta and Ppost=∣⟨ϕ∣ψ⟩∣2=sin⁡2δ=1/(1+Zw2)≈10−4P_{\text{post}} = \abs{\braket{\phi|\psi}}^2 = \sin^2\delta = 1/(1+Z_w^2)\approx10^{-4}, essentially saturating the bound Ppost≤1/Zw2P_{\text{post}}\le1/Z_w^2.

The quantum Zeno effect

Frequent strong measurements can do the opposite: freeze the evolution. Start in ∣ψ⟩\ket\psi and evolve under a Hamiltonian HH for a short time τ\tau (with ℏ=1\hbar = 1):

DerivationWhy watching freezes
  1. Psurvive(τ)=∣⟨ψ∣e−iHτ∣ψ⟩∣2=1−(ΔH)2τ2+O(τ4),(ΔH)2=⟨H2⟩−⟨H⟩2P_{\text{survive}}(\tau) = \abs{\bra\psi e^{-iH\tau}\ket\psi}^2 = 1-(\Delta H)^2\tau^2+\order{\tau^4},\qquad(\Delta H)^2 = \expval{H^2}-\expval{H}^2

    Short-time survival is quadratic in τ\tau: there is no linear term.

  2. PN(T)=[1−(ΔH)2(TN)2+…]N≈exp⁡ ⁣(−(ΔH)2T2N) →N→∞ 1P_N(T) = \Big[1-(\Delta H)^2\big(\tfrac TN\big)^2+\dots\Big]^N\approx\exp\!\Big(-\frac{(\Delta H)^2T^2}{N}\Big)\ \xrightarrow{N\to\infty}\ 1

    Measure NN times at intervals τ=T/N\tau = T/N, projecting onto ∣ψ⟩\ket\psi each time: survival tends to 1 (Misra & Sudarshan 1977).

  3. H=Ω2X,T=πΩ:PN=cos⁡2N ⁣(π2N)≈1−π24NH = \tfrac\Omega2X,\quad T = \frac\pi\Omega:\qquad P_N = \cos^{2N}\!\Big(\frac{\pi}{2N}\Big)\approx1-\frac{\pi^2}{4N}

    A resonant π\pi pulse would flip ∣0⟩→∣1⟩\ket0\to\ket1 with certainty. Interrupted by NN equally spaced ZZ measurements, the qubit is found in ∣0⟩\ket0 every time with probability approaching 1.

Predict before you look

With N=10N = 10 equally spaced ZZ measurements during the π\pi pulse, what is the probability that every measurement finds ∣0⟩\ket0?

Show the answer

0.78

P10=cos⁡20(π/20)=(0.98769)20≈0.78P_{10} = \cos^{20}(\pi/20) = (0.98769)^{20}\approx0.78. Without measurements the qubit would end in ∣1⟩\ket1 with certainty; with N=100N = 100 the survival probability rises to about 0.9760.976.

The same physics appears with continuous monitoring. A measurement that dephases the ZZ basis, so that the coherence decays at rate γ\gamma, competes with a Rabi drive of frequency Ω\Omega:

y˙=−Ωz−γy≈0 ⇒ y≈−Ωγ z,z˙=Ωy≈−Ω2γ z(γ≫Ω)\dot y = -\Omega z-\gamma y\approx0\ \Rightarrow\ y\approx-\frac{\Omega}{\gamma}\,z,\qquad \dot z = \Omega y\approx-\frac{\Omega^2}{\gamma}\,z\qquad(\gamma\gg\Omega)
Bloch equations for H=Ω2XH = \tfrac\Omega2X with dephasing in the ZZ basis. Eliminating the fast variable yy, the population relaxes at Γeff=Ω2/γ\Gamma_{\text{eff}} = \Omega^2/\gamma: the harder you look, the slower the qubit flips.

Gentle measurement and modern uses

DerivationProof: pure states first, then purify
  1. For ρ=∣ψ⟩ ⁣⟨ψ∣\rho = \dyad\psi, let a=⟨ψ∣E∣ψ⟩≥1−εa = \bra\psi E\ket\psi\ge1-\varepsilon and b=⟨ψ∣E∣ψ⟩b = \bra\psi\sqrt E\ket\psi. Since 0≤E≤I0\le E\le\id implies E≥E\sqrt E\ge E, we have b≥ab\ge a.

  2. ∥∣ψ⟩ ⁣⟨ψ∣−E∣ψ⟩ ⁣⟨ψ∣E∥1=(1+a)2−4b2≤(1−a)(1+3a)≤21−a≤2ε\norm{\dyad\psi-\sqrt E\dyad\psi\sqrt E}_1 = \sqrt{(1+a)^2-4b^2}\le\sqrt{(1-a)(1+3a)}\le2\sqrt{1-a}\le2\sqrt\varepsilon

    For vectors u,vu,v, the rank-two operator ∣u⟩ ⁣⟨u∣−∣v⟩ ⁣⟨v∣\dyad u-\dyad v has trace norm (∥u∥2+∥v∥2)2−4∣⟨u∣v⟩∣2\sqrt{\big(\norm u^2+\norm v^2\big)^2-4\abs{\braket{u|v}}^2}. With u=∣ψ⟩u = \ket\psi and v=E∣ψ⟩v = \sqrt E\ket\psi (so ∥v∥2=a\norm v^2 = a and ⟨u∣v⟩=b\braket{u|v} = b):

  3. For mixed ρ\rho, purify to ∣ψ⟩AR\ket\psi_{AR} and apply the pure case to E⊗IE\otimes\id; tracing out RR cannot increase the trace norm. The normalized statement follows from Uhlmann's theorem, as shown in Fidelity & Trace Distance. □\square

The ε\sqrt\varepsilon is optimal: take ∣ψ⟩\ket\psi and EE the projector onto a nearby pure state. The lemma is the engine behind several modern results:

  • Sequential decoding and quantum union bounds: asking many "is it message mm?" questions in sequence disturbs the state only mildly when the answers are nearly certain (Sen 2012; Gao 2015).
  • Shadow tomography and differential privacy: Aaronson (2018) estimates many observables on few copies by measuring gently and reusing the state; Aaronson & Rothblum (2019) showed that gentleness and differential privacy are closely linked.
  • Monitored quantum circuits: repeated weak or sparse measurements compete with entangling unitaries, producing measurement-induced entanglement phase transitions (Li, Chen & Fisher 2018; Skinner, Ruhman & Nahum 2019).

Key takeaways

  • Unsharp measurements produce a signal of order η\eta while disturbing coherences only at order η2\eta^2 per round; repeated, they still collapse the state, so gentleness slows collapse without evading the information–disturbance trade-off.
  • Continuous monitoring is described by a stochastic master equation whose average is a Lindblad dephasing equation.
  • Weak values Aw=⟨ϕ∣A∣ψ⟩/⟨ϕ∣ψ⟩A_w = \bra\phi A\ket\psi/\braket{\phi|\psi} are conditional pointer shifts; anomalous values cost post-selection probability at most ∥A∥∞2/∣Aw∣2\norm{A}_\infty^2/\abs{A_w}^2.
  • Zeno: short-time survival is quadratic in time, so NN measurements give survival 1−O(1/N)1-\order{1/N}; continuous dephasing at rate γ\gamma suppresses transitions to the rate Ω2/γ\Omega^2/\gamma.
  • Gentle measurement: an outcome of probability at least 1−ε1-\varepsilon disturbs the state by at most ε\sqrt\varepsilon in trace distance, a workhorse of quantum information theory.

Check yourself

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

Question 1

A qubit in ∣+⟩\ket+ undergoes one non-selective unsharp ZZ measurement with sharpness η=0.6\eta = 0.6, implemented by Lüders operators. What is its Bloch xx-component afterwards?

  1. 0.40.4
  2. 0.640.64
  3. 0.80.8
  4. 0.60.6
Show the answer

C. 0.80.8

Coherences shrink by 1−η2=0.64=0.8\sqrt{1-\eta^2} = \sqrt{0.64} = 0.8, so rxr_x goes from 11 to 0.80.8, while rz=0r_z = 0 is unchanged.

Question 2

Which statement about a long sequence of weak ZZ measurements on the same qubit is correct?

  1. It never disturbs the qubit, because each measurement is weak
  2. It eventually collapses the qubit onto ∣0⟩\ket0 or ∣1⟩\ket1, with Born-rule probabilities, once Nη2≫1N\eta^2\gg1
  3. It reveals rzr_z exactly while leaving the initial superposition intact
  4. It drives the qubit to the maximally mixed state for every measurement record
Show the answer

B. It eventually collapses the qubit onto ∣0⟩\ket0 or ∣1⟩\ket1, with Born-rule probabilities, once Nη2≫1N\eta^2\gg1

Each round dephases by 1−η2\sqrt{1-\eta^2} and nudges the conditional populations by O(η)\order{\eta}. The record performs a biased random walk, and after N≫1/η2N\gg1/\eta^2 rounds the conditional state is (nearly) an eigenstate. Weak measurement slows collapse; it does not avoid it.

Question 3

For pre-selection ∣ψ⟩\ket\psi, post-selection ∣ϕ⟩\ket\phi and an observable with ∥A∥∞=1\norm{A}_\infty = 1, the weak value satisfies:

  1. ∣Aw∣≤1/∣⟨ϕ∣ψ⟩∣\abs{A_w}\le1/\abs{\braket{\phi|\psi}}, so large weak values require rare post-selection
  2. ∣Aw∣≤1\abs{A_w}\le1 always
  3. AwA_w is always an eigenvalue of AA
  4. Aw=⟨ψ∣A∣ψ⟩A_w = \bra\psi A\ket\psi, whatever ∣ϕ⟩\ket\phi is
Show the answer

A. ∣Aw∣≤1/∣⟨ϕ∣ψ⟩∣\abs{A_w}\le1/\abs{\braket{\phi|\psi}}, so large weak values require rare post-selection

Cauchy–Schwarz bounds the numerator ∣⟨ϕ∣A∣ψ⟩∣\abs{\bra\phi A\ket\psi} by ∥A∥∞=1\norm{A}_\infty = 1. The weak value can exceed the spectrum, but only when ∣⟨ϕ∣ψ⟩∣\abs{\braket{\phi|\psi}} is small.

Question 4

A qubit undergoes a π\pi rotation interrupted by NN equally spaced projective measurements in its initial basis. For large NN, the probability that every measurement finds the initial state is approximately:

  1. 12\tfrac12
  2. 1/N1/N
  3. e−Ne^{-N}
  4. 1−π2/(4N)1-\pi^2/(4N)
Show the answer

D. 1−π2/(4N)1-\pi^2/(4N)

Each interval rotates the state by π/N\pi/N, so the survival probability per step is cos⁡2(π/2N)\cos^2(\pi/2N) and PN=cos⁡2N(π/2N)≈1−π2/(4N)→1P_N = \cos^{2N}(\pi/2N)\approx1-\pi^2/(4N)\to1.

Further reading

  • Y. Aharonov, D. Z. Albert & L. Vaidman, "How the result of a measurement of a component of the spin of a spin-1/2 particle can turn out to be 100", Physical Review Letters 60, 1351 (1988).
  • J. Dressel, M. Malik, F. M. Miatto, A. N. Jordan & R. W. Boyd, "Colloquium: Understanding quantum weak values: Basics and applications", Reviews of Modern Physics 86, 307 (2014).
  • B. Misra & E. C. G. Sudarshan, "The Zeno's paradox in quantum theory", Journal of Mathematical Physics 18, 756 (1977).
  • W. M. Itano, D. J. Heinzen, J. J. Bollinger & D. J. Wineland, "Quantum Zeno effect", Physical Review A 41, 2295 (1990).
  • H. M. Wiseman & G. J. Milburn, Quantum Measurement and Control (Cambridge, 2010).