Projective Measurement versus POVM
Part 1. Name a state that was not sent
The \(X\)–\(Z\) plane of the Bloch sphere. Green: the three states Bob may send. Red: the measurement. In projective mode, drag the end of the axis to turn it.
Score
1. The game
Bob sends one of the trine states \(\ket{e_k} = \cos(2\pi k/3)\ket{0} + \sin(2\pi k/3)\ket{1}\), \(k = 0, 1, 2\), each with probability \(\tfrac13\). You measure it once and then name one of the three states. You win if Bob did not send the state you name.
2. Your measurement
You measure \(\hat n\cdot\vec\sigma\) with \(\hat n = (\sin\alpha, 0, \cos\alpha)\). After each outcome you name the state least likely to give that outcome; no other rule does better for this axis.
\(\Lambda_k = \tfrac23\ket{e_k^\perp}\!\bra{e_k^\perp}\), \(k = 0, 1, 2\), where \(\ket{e_k^\perp}\) is orthogonal to \(\ket{e_k}\). For outcome \(k\), you name \(\ket{e_k}\).
3. Play
Exercises
- In projective mode, turn the axis and find the angle \(\alpha\) with the highest winning probability. Show that the best value is \(1 - \tfrac{2 - \sqrt3}{6} \approx 0.955\).
- Why can no projective measurement on a qubit win every round? Think about how many outcomes it has and how many states must be excluded.
- Switch to the POVM and look at the table. Which entries are zero, and why does that guarantee a win?
- Show that \(\sum_k \tfrac23\ket{e_k^\perp}\!\bra{e_k^\perp} = I\), using the Bloch vectors of \(\ket{e_k^\perp}\).
Part 2. Find the hidden state
Dashed circles: the pure states that agree with your projective data (one circle per measured axis). Red arrow: estimate from projective data. Green arrow: estimate from POVM data. Black arrow: the hidden state, after you reveal it.
1. The task
Many copies of an unknown pure state \(\rho = \tfrac12(I + \vec r\cdot\vec\sigma)\) are available. Estimate \(\vec r\) by measuring copies. Each measurement setting is a separate apparatus, so count how many settings you need.
2. Measure
Projective
A measurement of \(\sigma_i\) has two outcomes and estimates one number, \(r_i = \langle \sigma_i\rangle\).
POVM
\(\Lambda_j = \tfrac12\ket{t_j}\!\bra{t_j}\), \(j = 1, \dots, 4\), with the Bloch vectors \(\hat n_j\) of \(\ket{t_j}\) at the corners of a regular tetrahedron. Then \(P(j) = \tfrac14\left(1 + \vec r\cdot\hat n_j\right)\) and \(\vec r = 3\sum_j P(j)\,\hat n_j\).
3. Estimates
Clears all measurements and hides the state again. The hidden state stays the same, so you can try another strategy on it; use "New hidden state" for a different one.
Exercises
- Measure only \(Z\). Which pure states agree with your data? How many settings do you need before the state is pinned down?
- Starting from a new hidden state, use only the tetrahedral POVM. How many settings does it take?
- Derive \(\vec r = 3\sum_j P(j)\,\hat n_j\) from \(P(j) = \tfrac14(1 + \vec r\cdot\hat n_j)\), \(\sum_j \hat n_j = 0\) and \(\sum_j \hat n_j\hat n_j^{\mathsf T} = \tfrac43 I\).
- Spend the same total number of copies on each strategy and compare the trace distances to the hidden state. Which strategy needs fewer settings for a similar error?
- Why can no single projective measurement on a qubit do what the tetrahedral POVM does?