101
Part Three · A · Foundations of quantum information

Measurement: projective, POVM, weak

Someone hands you a qubit prepared in one of two states that are not orthogonal, say ∣0⟩|0\rangle or a state tilted 30° from it, each with equal probability. No measurement can always tell which. But you have choices. You can maximize your chance of guessing right. Or you can insist on never being wrong, at the price of sometimes answering “I don’t know”. These are different measurements, and neither is the textbook projective measurement in a fixed basis. Quantum measurement is richer than Chapter 27 let on.

On the plate: a discrimination game. Choose a strategy and play rounds.

POVMs

The most general measurement assigns to each outcome ii a positive operator EiE_i, with ∑iEi=1\sum_iE_i = \mathbb 1, and the probability of ii is Tr(ρEi)\mathrm{Tr}(\rho E_i). This is a positive operator-valued measure. The EiE_i need not be projectors, need not be orthogonal, and there can be more outcomes than the dimension. Every POVM can be realized as an ordinary projective measurement on a larger system, the qubit plus an auxiliary one, by Naimark’s dilation. So nothing new is assumed; it is projective measurement seen from inside a subsystem. Established

Three strategies

The two states are ∣0⟩|0\rangle and cos⁡θ∣0⟩+sin⁡θ∣1⟩\cos\theta|0\rangle + \sin\theta|1\rangle, overlap cos⁡θ\cos\theta. Set θ=\theta = 30°. Choose: , , , then .

Measuring in z succeeds with probability 12(1+sin⁡2θ)\tfrac12(1 + \sin^2\theta). Carl Helstrom showed in 1969 that the best possible is

Pcorrect=12(1+1−∣⟨ψ∣ϕ⟩∣2)P_{\text{correct}} = \tfrac12\left(1 + \sqrt{1 - |\langle\psi|\phi\rangle|^2}\right)

achieved by measuring in the eigenbasis of the difference of the two density matrices. Unambiguous discrimination, a three-outcome POVM, is never wrong but succeeds only with probability 1−∣⟨ψ∣ϕ⟩∣1 - |\langle\psi|\phi\rangle| and otherwise says “inconclusive”. At 30°: 62.5%, 75%, and 13% with no errors. Established

Weak measurement

Couple a system only slightly to a meter, so each reading is very noisy and barely disturbs the system. Averaged over many runs, a weak measurement gives the expectation value without collapsing the state. Yakir Aharonov, David Albert and Lev Vaidman showed in 1988 that if you also post-select on a later outcome, the average reading is a weak value, ⟨ϕ∣A∣ψ⟩/⟨ϕ∣ψ⟩\langle\phi|A|\psi\rangle/\langle\phi|\psi\rangle, which can lie far outside the eigenvalue range: a spin-½ reading of 100. Weak values were used to map average photon trajectories in a double slit (Chapter 54) and to amplify tiny signals. What they mean physically is argued over. Debated

What you now have

POVMs and Naimark dilation, the Helstrom bound for minimum-error discrimination, unambiguous discrimination with an inconclusive outcome, and weak measurement and weak values.

· · ·

Three questions to carry with you.

Can non-orthogonal states be distinguished perfectly?tap
Not always; but unambiguously some of the time, with an inconclusive answer otherwise.
What does Naimark’s theorem say?tap
Any POVM is a projective measurement on the system plus an ancilla.
Can a weak value exceed the eigenvalues?tap
Yes, with post-selection on nearly orthogonal final states.

And one to work out. Two equally likely pure states have overlap 0.6. What is the Helstrom success probability?

P =

Sources: C. W. Helstrom, J. Stat. Phys. 1, 231 (1969); Quantum Detection and Estimation Theory (1976). I. D. Ivanovic, Phys. Lett. A 123, 257 (1987); D. Dieks, Phys. Lett. A 126, 303 (1988); A. Peres, Phys. Lett. A 128, 19 (1988). Y. Aharonov, D. Z. Albert and L. Vaidman, Phys. Rev. Lett. 60, 1351 (1988). A. Peres, Quantum Theory: Concepts and Methods. J. Watrous, The Theory of Quantum Information, ch. 2–3. Checked by code: at θ = 30°, naive 0.625, Helstrom 0.750 (also by eigen-decomposition), unambiguous 0.134.

Next · Chapter 102
No-cloning, no-deleting, no-broadcasting

If you could copy a qubit, you could signal faster than light.

Turn the page