Measurement
How quantum information becomes classical, and why you only get one look.
A qubit can be in infinitely many different states, yet when you look at it you get exactly one bit, and the look itself changes the qubit. Measurement is where quantum information becomes classical data, and every quantum algorithm has to plan around it.
What you’ll learn
- Predict measurement probabilities from amplitudes with the Born rule.
- Explain collapse, and why a single copy of a qubit reveals so little about its state.
- Measure in the , and bases and connect the results to directions on the Bloch sphere.
- Show how recording a qubit's value in the middle of a circuit destroys interference.
- Compute an expectation value such as from outcome frequencies.
One look, one bit
The qubit below starts at the north pole, . Press H to swing it onto the equator. It is now the superposition : a perfectly definite state, described by two amplitudes.
Now look at it by pressing Measure. A measurement does not report the amplitudes. It reports a single classical bit, or , and the arrow jumps to the matching pole.
The Born rule
For a qubit in the state , the odds are set by the squared magnitudes of the amplitudes. This is the Born rule:
A qubit sits at polar angle , a third of the way from the north pole to the south pole, so its amplitude on is . You prepare and measure 1000 copies. About how many s do you expect?
Show the answer
250 ones
, so about . An amplitude of means a probability of , not . From run to run the count wobbles by about (one standard deviation is ).
Collapse: you only get one look
After a measurement the qubit is left in the state that matches the outcome: if you saw , if you saw . This update is called collapse. Whatever the amplitudes were before, that information is gone, except for the single bit you recorded.
That makes one qubit a stingy source of information. Measuring it again teaches you nothing new, because it just repeats its answer. And you cannot make backup copies beforehand: the no-cloning theorem forbids copying an unknown quantum state.
Think first A friend hands you one qubit in a state that only they know. You measure it and get . What have you learned about and ?
Almost nothing: only that . Nearly every state can produce a . To estimate you need many identically prepared copies and the fraction of s; to pin down the relative phase you also need other measurement bases (coming up next). Reconstructing a state this way is called state tomography.
| Before the measurement | After the measurement |
|---|---|
| Two continuous amplitudes | One recorded bit |
| Outcome uncertain (unless or is ) | Repeating the measurement gives the same bit |
| Superposition, able to interfere | Replaced by or |
Choosing a basis
"Is it or ?" is a question about the vertical axis of the Bloch sphere. You can ask about any axis. Measuring in the basis asks "is it or ?"; the basis asks "is it or ?". The two possible answers are always a pair of opposite points on the sphere.
Try one state, , with three different questions.
The rule behind these histograms: if the state's arrow makes an angle with the measurement axis, the outcome "along the axis" has probability . For and the axis, and the probability is . For the or axis, and .
A qubit is in the state . In which basis does a measurement give the same outcome every time?
- The basis
- The basis
- The basis
- None: a superposition always gives random outcomes
Show the answer
C. The basis
is one of the two -basis states (the point of the sphere), so a measurement returns "" with certainty. Relative to the and axes its arrow is at , which gives 50/50.
Rotate, then measure
Most quantum hardware can only measure in the basis. To ask a different question, first rotate the state so that the axis you care about points along , then measure.
For the basis the rotation is the Hadamard gate: and . So "apply , then measure" is an measurement, with result meaning and result meaning .
Looking destroys interference
Two Hadamards in a row do nothing: . The first splits the qubit into two branches, and the second recombines them so that the two contributions to cancel. That cancellation is interference.
What if, between the two gates, a record is made of whether the qubit was or ? Nobody even has to read it. The circuit below compares an untouched qubit (top wire) with one whose value is copied onto a third "notebook" qubit by a CNOT gate (the dot joined to the ).
The copy did to the middle qubit exactly what a measurement does: it forced the two branches apart, so they can no longer recombine. Any interaction that leaves a record has this effect, whether it is a detector click, a scattered photon or a stray spin in the chip. When the environment does it uncontrollably, it is called decoherence, the main enemy of quantum computers. It is also why algorithms save their measurements for the very end.
Averages, and what comes next
Physicists often label the outcomes (for ) and (for ). The average of many such results is the expectation value
It runs from (always ) through (a fair coin) to (always ). The same recipe in the and bases gives and , and together the three averages are the coordinates of the qubit's arrow on the Bloch sphere.
A qubit gives in 80% of the shots. What is ?
Show the answer
0.6
. Geometrically, that is the height of the arrow: .
| Excited State lesson | What you will be able to do |
|---|---|
| The Born Rule & Projective Measurement | Use projectors to compute probabilities and post-measurement states, measure one qubit out of many, and prove why overlapping states can never be told apart perfectly. |
| Observables, Bases & Expectation Values | Treat measurable quantities as Hermitian matrices, compute , and measure in any basis. |
| Density Matrices & Mixed States | Describe a qubit you are uncertain about, with states that live inside the Bloch ball. |
| Generalized Measurements (POVMs) | Go beyond projectors: measurements with more outcomes than dimensions, and the best possible ways to distinguish states. |
Key takeaways
- Measuring a qubit returns one classical bit. The Born rule gives the odds: and .
- Measurement collapses the state onto the outcome, so repeating it repeats the answer, and one copy reveals almost nothing about the amplitudes.
- You can measure along any axis of the Bloch sphere. In practice you rotate first and then measure in (apply for the basis).
- A state that is certain in one basis is random in the complementary ones (, , ).
- Any record of a qubit's value, even one nobody reads, destroys interference. That is decoherence, and it is why algorithms measure at the end.
- Averages of outcomes are expectation values, and are the coordinates of the Bloch vector.
Check yourself
In the interactive lesson these are scored with your confidence, and what you miss comes back for review.
A qubit is in the state . What is the probability of measuring ?
Show the answer
C.
. The factor is a phase: it has magnitude and does not change the odds.
Why can't you learn the amplitudes and of a single qubit by measuring it over and over?
- After the first measurement it has collapsed, so later measurements just repeat the first result
- Because every measurement outcome is a 50/50 coin flip
- Because measuring always destroys the qubit
- You can, as long as you measure quickly enough
Show the answer
A. After the first measurement it has collapsed, so later measurements just repeat the first result
The first measurement replaces the state by or . Every later measurement in the same basis reports that same bit, so you learn nothing new. Estimating amplitudes requires many freshly prepared copies.
You measure the state in the basis. What do you get?
- with certainty
- or , each with probability
- with certainty
- with certainty, because the state is
Show the answer
B. or , each with probability
points along , at to the axis, so each outcome has probability . Equivalently, , which a measurement reads as 50/50.
A qubit starts in and goes through , then a measurement, then another , then a final measurement. What is the probability that the final result is ?
- , because does nothing
Show the answer
D.
The middle measurement leaves or . The second turns either one into an equal superposition, so the final result is 50/50. Without the middle measurement the branches would interfere and give every time.
For which of these states is ?
Show the answer
C.
lies on the equator, so and . Every equator state has .
Go deeper
- The Born Rule & Projective Measurement · Probabilities from amplitudes, and the post-measurement state.
- Observables, Bases & Expectation Values · Measuring in any basis, Pauli observables and expectation values.
- Density Matrices & Mixed States · Describing uncertainty about a quantum state: the Bloch ball.
- Generalized Measurements (POVMs) · Measurements beyond projectors: the most general quantum measurement.
Further reading
- M. A. Nielsen & I. L. Chuang, Quantum Computation and Quantum Information, §1.2 and §2.2.3 (Cambridge, 10th anniv. ed., 2010).
- J. Preskill, Caltech Ph219/CS219 lecture notes, Ch. 2 "Foundations I: States and Ensembles" and Ch. 3 "Foundations II: Measurement and Evolution".