Quantum probabilities obey ordinary probability rules for the recorded outcomes of a specified experiment. The subtlety is that changing the measurement procedure can change the experiment. An unobserved alternative and a measured outcome whose label is later ignored are different physical arrangements.
A preparation and two possible tests
Prepare a qubit in |α⟩ = cos(α/2)|0⟩ + sin(α/2)|1⟩. Restrict attention to real superpositions in one plane of the Bloch sphere. The angles below are Bloch angles; for a polarization implementation they are twice the corresponding angle of linear polarization.
A measurement axis θ has a plus state |θ+⟩ = cos(θ/2)|0⟩ + sin(θ/2)|1⟩ and an orthogonal minus state. An ideal projective measurement returns s = +1 or −1 with probability [1 + s cos(α−θ)]/2, leaving the corresponding eigenstate.
A second ideal measurement along β then returns t = +1 or −1. The probability of the full recorded pair is
p(s,t) = [1 + s cos(α−θ)] [1 + st cos(θ−β)] / 4.
Summing the two rows with final result + gives [1 + cos(α−θ)cos(θ−β)]/2. This is the ordinary sum over the mutually exclusive first outcomes. If the first apparatus is absent instead, direct measurement along β gives [1 + cos(α−β)]/2.
Insert or omit an intermediate measurement
IDEAL QUBIT MODEL · BLOCH ANGLES IN DEGREES
The initial and final axes agree. An intermediate perpendicular measurement changes the final plus probability from one to one half.
| First s | Final t | Joint probability |
|---|---|---|
| + | + | 0.250000 |
| + | − | 0.250000 |
| − | + | 0.250000 |
| − | − | 0.250000 |
Read the default case carefully
At α = β = 0 and θ = 90°, the preparation is |0⟩, the intermediate measurement is in the +,− basis, and the final measurement returns to 0,1. All four sequential outcome pairs have probability one quarter. The final result is therefore equally divided.
Without the intermediate apparatus, the prepared |0⟩ remains unchanged and the final plus result is certain. This difference does not violate the law of total probability. Its two sides refer to different experimental procedures.
Changing θ to 0 makes the intermediate test compatible with the preparation and restores certainty. Other settings can increase rather than decrease a chosen final probability. With α = 0, β = 180° and θ = 90°, the direct plus probability is zero, while inserting the measurement raises it to one half.
Where the interference term goes
In the experiment without an intermediate record, the identity can be expanded in the θ basis while calculating an amplitude. The contributions through θ+ and θ− must be added before taking the squared magnitude. The cross term can be nonzero.
An actual ideal measurement correlates those alternatives with distinguishable records. Ignoring their labels then gives a sum of branch probabilities. Merely changing a basis in a calculation does not create such records or perform a measurement.
The demonstration assumes an ideal projective instrument and no free evolution between tests. A weak measurement, a noisy detector or an intervening unitary operation would require a different calculation. The outcomes alone do not characterize every possible instrument.
Completion must predict sequential experiments
The book explicitly supplies the standard probability prescription in its discussion of composition and probabilities. The calculations here use that prescription. They are benchmarks for an SMM/DTT reconstruction, not an independent derivation of the Born rule.
A completion model should predict when an intermediate interaction creates a distinguishable record and how subsequent measurements depend on it. Matching one final histogram is less restrictive than matching the full four-outcome table across different settings.
Sources and further reading
- Feynman, Leighton and Sands, Probability Amplitudes, for the distinction between combining amplitudes and probabilities.
- Continue with measurement instruments and quantum histories. The angle-controlled calculation is an original teaching example.

