Observables and operators
Connect Hermitian operators, eigenvalues, and expectation values to the outcomes of a quantum measurement.
On this page
The question for today
What you’ll understand
- Compute the outcome probabilities and mean of a finite-dimensional observable.
- Explain why an expectation value need not be an allowed individual outcome.
- Use a commutator to test whether the order of two operations matters.
A quantum state tells us about a preparation; an observable specifies a measurable question. Operators connect the two. Their eigenvalues identify possible outcomes in an ideal projective measurement, while the state determines how often those outcomes occur. Learning to keep an individual reading separate from an average prevents a large class of conceptual and numerical mistakes.
An observable is more than its matrix entries
In a finite-dimensional Hilbert space, an observable is represented by a Hermitian operator . Its eigenvalues are real and its eigenspaces provide orthogonal measurement alternatives. A matrix is the representation of that operator in a chosen basis, so a nondiagonal matrix is not a defective observable. Diagonalizing it finds a basis adapted to its outcomes. The eigenvector equation says that applying the operator to an eigenstate only scales that state.
In infinite-dimensional spaces, domains and boundary conditions become part of the operator definition. The technically appropriate condition for an observable is self-adjointness; formal integration-by-parts symmetry alone can be insufficient. The momentum expression is therefore not a complete physical specification until the space and allowed functions are stated. Here we first use finite matrices, where those domain subtleties do not obstruct the calculation.
The mean is a prediction for an ensemble
The expectation value equals the probability-weighted mean of the eigenvalues. It predicts the average across many identically prepared systems measured in the same way. It does not say that any one detector must display that number. A two-outcome apparatus can have a mean between its two readings even though no single trial ever returns that intermediate value.
The variance measures the intrinsic spread of that outcome distribution for the stated preparation. Detector calibration error and technical noise can add further spread, but they are distinct ingredients. If the state is entirely in one eigenspace, the ideal variance vanishes. A zero mean, by contrast, does not imply zero spread: equally likely positive and negative readings can cancel in the average.
Order, commutators, and compatible questions
Operator multiplication represents composition, and generally . The commutator measures this failure. For finite Hermitian matrices, commuting observables admit a common orthonormal eigenbasis. In that basis, both values can be definite together. A nonzero commutator rules out a complete common eigenbasis, although selected special states can still have additional properties that must be checked directly.
Position and momentum illustrate the physical consequences: on an appropriate common domain, . Replacing quantum operators by commuting ordinary numbers would erase this structure. Later, angular momentum and creation operators will let us solve problems through algebra rather than differential equations. At each step, retain the order of factors and ask what space the operators act on; a visually convenient rearrangement is not automatically valid.
Work it through
An average that is never a single outcome
An observable has matrix in an orthonormal basis and the state is . Find its outcome distribution, mean, and standard deviation. Use dimensionless units.
The matrix is already diagonal, so its basis vectors have eigenvalues 1 and 3. The state is normalized because . Its outcome probabilities are therefore and .
Compute the mean: . An individual measurement still returns only 1 or 3.
Square the eigenvalues to obtain . Subtract the squared mean: , so .
The ensemble mean is 1.5 and the standard deviation is about 0.866. Neither quantity adds a new eigenvalue to the apparatus.
The living notebook
Explore the model
Interactive lab · quantum
Weight two observable outcomes
Treat the fixed zero/one basis as the eigenbasis of . Change the mixing angle and compute the mean from the displayed basis probabilities. Changing relative phase leaves this diagonal observable's distribution unchanged. The lab does not diagonalize an arbitrary input operator.
- Probability |0⟩
- 71.1%
- Probability |1⟩
- 28.9%
- Bloch (x, y, z)
- 0.78, 0.45, 0.42
A normalized two-level pure state, not an interacting field simulation. Playback adds one radian per second to the relative phase; basis probabilities stay fixed. Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.
Look forThe spectrum lists possible readings; state-dependent weights determine their statistics.
Pause & reason
Check your understanding
Your turn
Try it yourself
Let and . Apply and separately to . What does this establish?
Take a moment to make a prediction or work it out on paper. Open a hint when you need one.
Give me a hint
The rightmost matrix acts first. Compute the two output columns before subtracting them.
Compare with the solution
Since , . But and , so . Thus : the operators do not commute.
Keep the idea
Take it with you
- Eigenvalues describe ideal individual outcomes; the state supplies their probabilities.
- Expectation and variance summarize an ensemble without changing its possible readings.
- Commutators encode the effect of order and constrain compatible observables.
Read further
Original sources and deeper explanations to follow at your own pace.
Optional progress, saved in this browser when available. Revisit any lesson at any time.
Lesson not marked complete.