Density-Matrix Expectation Values

Formula

For a density operator $\rho$ and observable $A$,

$$\langle A\rangle_\rho = \operatorname{Tr}(\rho A).$$

A physical density operator satisfies

$$\rho\geq0, \qquad \rho^\dagger=\rho, \qquad \operatorname{Tr}\rho=1.$$

For a measurement effect $E_a$,

$$p(a) = \operatorname{Tr}(\rho E_a).$$

The expectation and probability rules are the same state–operator pairing. An observable labels outcomes numerically; an effect represents one event.

At a glance

Setting	Trace rule
Observable expectation	$\langle A\rangle_\rho=\operatorname{Tr}(\rho A)$
Projective outcome	$p(a)=\operatorname{Tr}(\rho P_a)$
General measurement effect	$p(a)=\operatorname{Tr}(\rho E_a)$
Function or moment	$\langle f(A)\rangle_\rho=\operatorname{Tr}[\rho f(A)]$
Variance	$(\Delta_\rho A)^2=\operatorname{Tr}(\rho A^2)-[\operatorname{Tr}(\rho A)]^2$
Matrix components	$\langle A\rangle_\rho=\sum_{i,j}\rho_{ij}A_{ji}$
Ensemble representation	$\langle A\rangle_\rho=\sum_kp_k\langle\psi_k\rvert A\lvert\psi_k\rangle$
Pure state	$\operatorname{Tr}(\lvert\psi\rangle\langle\psi\rvert A)=\langle\psi\rvert A\lvert\psi\rangle$
Local observable	$\operatorname{Tr}_{AB}[\rho_{AB}(A\otimes I_B)]=\operatorname{Tr}_A(\rho_AA)$
Conditional branch	$\langle A\rangle_a=\operatorname{Tr}(\widetilde\rho_aA)/\operatorname{Tr}\widetilde\rho_a$
Qubit	$\rho=(I+\mathbf r\cdot\boldsymbol\sigma)/2$, $A=a_0I+\mathbf a\cdot\boldsymbol\sigma$ gives $\langle A\rangle=a_0+\mathbf r\cdot\mathbf a$

Meaning

The density operator contains exactly the information needed to predict all measurements on the represented system. If two preparation procedures produce the same $\rho$, then

$$\operatorname{Tr}(\rho A)$$

is the same for every observable $A$, and

$$\operatorname{Tr}(\rho E)$$

is the same for every effect $E$. Measurements on that system alone cannot distinguish those preparations.

The trace rule applies to:

• pure states;
• classical random mixtures of preparations;
• reduced states of entangled systems;
• thermal states;
• states produced by noisy dynamics or unrecorded measurements.

It does not reveal which ensemble decomposition or environment produced the state. A density operator is an operational state, not a unique hidden list of pure states.

Trace essentials

In a finite-dimensional Hilbert space,

$$\operatorname{Tr}X = \sum_n \langle n\rvert X\lvert n\rangle$$

in any orthonormal basis. The result is basis independent.

Useful identities include

$$\operatorname{Tr}(XY) = \operatorname{Tr}(YX),$$ $$\operatorname{Tr}(XYZ) = \operatorname{Tr}(ZXY) = \operatorname{Tr}(YZX),$$

and

$$\operatorname{Tr} \left( \lvert u\rangle\langle v\rvert A \right) = \langle v\rvert A\lvert u\rangle.$$

Cyclicity permits cyclic rotation, not arbitrary reordering. In general,

$$\operatorname{Tr}(XYZ) \neq \operatorname{Tr}(XZY).$$

In infinite dimensions, cyclicity requires the products to lie in classes for which the traces exist. Formal symbol rearrangement is not a substitute for that condition.

Pure states as a special case

For

$$\rho_\psi = \lvert\psi\rangle\langle\psi\rvert,$$

the rank-one trace identity gives

$$\begin{aligned} \operatorname{Tr}(\rho_\psi A) &= \operatorname{Tr} \left( \lvert\psi\rangle \langle\psi\rvert A \right)\\ &= \langle\psi\rvert A\lvert\psi\rangle. \end{aligned}$$

For a rank-one projector

$$P_\phi = \lvert\phi\rangle\langle\phi\rvert,$$ $$\operatorname{Tr}(\rho_\psi P_\phi) = \lvert \langle\phi\rvert\psi\rangle \rvert^2.$$

The density-operator formulation therefore extends rather than replaces the pure-state Born and expectation formulas.

Matrix-component formula

In any orthonormal basis,

$$\begin{aligned} \operatorname{Tr}(\rho A) &= \sum_i (\rho A)_{ii}\\ &= \sum_{i,j} \rho_{ij}A_{ji}. \end{aligned}$$

Off-diagonal entries of $\rho$ contribute whenever $A$ has matching off-diagonal entries. Reading probabilities from the diagonal of $\rho$ is valid only after specifying the measurement basis.

If the chosen basis diagonalizes a nondegenerate $A$,

$$A = \sum_a a\lvert a\rangle\langle a\rvert,$$

then

$$\langle A\rangle_\rho = \sum_a a\,\rho_{aa},$$

where

$$\rho_{aa} = \langle a\rvert\rho\lvert a\rangle$$

is the probability of outcome $a$. In another basis, the same diagonal entries need not be the probabilities for measuring $A$.

For degenerate eigenvalues, use the full spectral projector:

$$p(a) = \operatorname{Tr}(\rho P_a).$$

Ensemble form

If

$$\rho = \sum_k p_k \lvert\psi_k\rangle \langle\psi_k\rvert, \qquad p_k\geq0, \qquad \sum_kp_k=1,$$

then

$$\begin{aligned} \operatorname{Tr}(\rho A) &= \sum_kp_k \operatorname{Tr} \left( \lvert\psi_k\rangle \langle\psi_k\rvert A \right)\\ &= \sum_kp_k \langle\psi_k\rvert A\lvert\psi_k\rangle. \end{aligned}$$

This is the classical average of the component quantum expectations.

The decomposition of $\rho$ is generally not unique. Any two ensembles that produce the same operator give the same result for every $A$. Do not assign physical uniqueness to one convenient decomposition unless the preparation record supplies that additional information.

Probabilities, moments, and variance

For a POVM $\{E_a\}$,

$$E_a\geq0, \qquad \sum_aE_a=I,$$

and

$$p(a) = \operatorname{Tr}(\rho E_a).$$

Positivity gives

$$p(a)\geq0,$$

while completeness gives

$$\sum_ap(a) = \operatorname{Tr}\rho = 1.$$

For a self-adjoint observable $A$ and suitable function $f$,

$$\langle f(A)\rangle_\rho = \operatorname{Tr} \left[ \rho f(A) \right].$$

In particular,

$$\langle A^n\rangle_\rho = \operatorname{Tr}(\rho A^n)$$

when the moment exists, and

$$(\Delta_\rho A)^2 = \operatorname{Tr}(\rho A^2) - \left[ \operatorname{Tr}(\rho A) \right]^2.$$

An effect determines outcome probability but not the conditional state. Different measurement instruments can have the same effects and different post-measurement maps.

Reality, positivity, and spectral bounds

If $\rho$ is a valid state and $A$ is self-adjoint,

$$\operatorname{Tr}(\rho A) \in \mathbb R.$$

If $A\geq0$, then

$$\operatorname{Tr}(\rho A)\geq0.$$

For bounded self-adjoint $A$ with spectrum in $[a_{\min},a_{\max}]$,

$$a_{\min} \leq \operatorname{Tr}(\rho A) \leq a_{\max}.$$

For an effect $0\leq E\leq I$,

$$0 \leq \operatorname{Tr}(\rho E) \leq 1.$$

These are powerful diagnostics. A complex expectation for a Hermitian matrix or probability outside $[0,1]$ usually signals invalid input, inconsistent bases, incorrect conjugation, or numerical error.

Local observables and reduced states

Let $\rho_{AB}$ describe a bipartite system and let $A$ act only on subsystem $A$. The full-system observable is

$$A\otimes I_B.$$

Its expectation is

$$\langle A\rangle = \operatorname{Tr}_{AB} \left[ \rho_{AB} (A\otimes I_B) \right].$$

Define

$$\rho_A = \operatorname{Tr}_B\rho_{AB}.$$

Then

$$\operatorname{Tr}_{AB} \left[ \rho_{AB} (A\otimes I_B) \right] = \operatorname{Tr}_A(\rho_AA).$$

The reduced state reproduces every local prediction on $A$, whether $\rho_{AB}$ is separable or entangled.

For a product observable $A\otimes B$,

$$\langle A\otimes B\rangle = \operatorname{Tr}_{AB} \left[ \rho_{AB} (A\otimes B) \right].$$

This generally cannot be computed from $\rho_A$ and $\rho_B$ alone because those marginals do not determine correlations.

Qubit shortcut

Every qubit state can be written as

$$\rho = \frac12 \left( I+\mathbf r\cdot\boldsymbol{\sigma} \right), \qquad \lVert\mathbf r\rVert\leq1.$$

Every Hermitian qubit observable has the form

$$A = a_0I + \mathbf a\cdot\boldsymbol{\sigma},$$

with real $a_0$ and $\mathbf a$. Using

$$\operatorname{Tr}\sigma_i=0, \qquad \operatorname{Tr}(\sigma_i\sigma_j)=2\delta_{ij},$$

one obtains

$$\langle A\rangle_\rho = a_0+\mathbf r\cdot\mathbf a.$$

For a spin measurement along unit vector $\mathbf n$, the projectors are

$$P_\pm = \frac12 \left( I\pm\mathbf n\cdot\boldsymbol{\sigma} \right),$$

so

$$p(\pm) = \frac12 \left( 1\pm\mathbf r\cdot\mathbf n \right).$$

The Bloch vector encodes exactly the three Pauli expectation values:

$$r_i = \operatorname{Tr}(\rho\sigma_i).$$

Unnormalized and conditional states

A positive operator $\widetilde\rho$ with

$$0< \operatorname{Tr}\widetilde\rho <\infty$$

can be normalized by

$$\rho = \frac{\widetilde\rho} {\operatorname{Tr}\widetilde\rho}.$$

Then

$$\langle A\rangle = \frac{ \operatorname{Tr}(\widetilde\rho A) }{ \operatorname{Tr}\widetilde\rho }.$$

For a measurement branch

$$\widetilde\rho_a = \mathcal I_a(\rho),$$

its trace is the outcome probability:

$$p(a) = \operatorname{Tr}\widetilde\rho_a.$$

If $p(a)>0$, the conditional expectation is

$$\langle A\rangle_a = \frac{ \operatorname{Tr}(\widetilde\rho_aA) }{ p(a) }.$$

Do not normalize a branch before recording its trace if the branch probability is needed.

Changes of basis

Under a unitary basis change,

$$\rho' = U\rho U^\dagger, \qquad A' = UAU^\dagger.$$

Then

$$\begin{aligned} \operatorname{Tr}(\rho'A') &= \operatorname{Tr} \left( U\rho A U^\dagger \right)\\ &= \operatorname{Tr}(\rho A). \end{aligned}$$

Transforming both objects is a passive representation change. Transforming only the state or only the observable generally describes a physical change, not merely new coordinates.

Infinite-dimensional conditions

If $\rho$ is trace class and $A$ is bounded, then $\rho A$ is trace class and

$$\operatorname{Tr}(\rho A)$$

is well defined.

For an unbounded self-adjoint $A$, a standard absolute-integrability condition is

$$\operatorname{Tr} \left( \rho\lvert A\rvert \right) <\infty.$$

Equivalently, the first absolute moment of the Born spectral measure is finite. A finite variance requires the second moment:

$$\operatorname{Tr}(\rho A^2)<\infty$$

in the positive spectral sense.

Writing matrix elements in an arbitrary basis and rearranging infinite sums can fail when absolute convergence is absent. Use the spectral measure or trace-class formulation.

Assumptions

• $\rho$ is positive, trace class, and trace one, or the displayed normalization denominator is included.
• $A$, $P_a$, or $E_a$ acts on the same Hilbert space as $\rho$.
• Observables are self-adjoint.
• POVM effects are positive and complete for the modeled outcome set.
• The trace pairing and required moments exist.
• Local subsystem formulas use an explicitly specified tensor-product decomposition and tensor ordering.
• Matrix component formulas use one common orthonormal basis for both operators.

Validity and limitations

The trace rule is exact within standard quantum mechanics. It applies to pure and mixed states and to reduced states of open subsystems.

The rule does not:

• choose a unique ensemble decomposition of $\rho$;
• determine dynamics without a Hamiltonian, channel, or master equation;
• determine post-measurement states from effects alone;
• make nonpositive trace-one matrices physical;
• guarantee finite moments for unbounded observables;
• reconstruct correlations from reduced states alone.

Calculation checks

• Verify $\rho=\rho^\dagger$, $\rho\geq0$, and $\operatorname{Tr}\rho=1$.
• Verify dimensions and basis ordering of $\rho$ and $A$ match.
• For Hermitian $A$, the result must be real within numerical tolerance.
• For positive $A$, the result must be nonnegative.
• For $A=I$, the result must equal one.
• For an effect, the result must lie in $[0,1]$.
• A complete POVM must give probabilities summing to one.
• A bounded-observable expectation must lie in the spectral range.
• A consistent unitary basis change must preserve the result.
• For a pure state, compare against $\langle\psi\rvert A\lvert\psi\rangle$.
• For a local observable, compare full and reduced-state calculations.

Minimal worked uses

Coherence-sensitive qubit expectation

Let

$$\rho = \frac12 \begin{pmatrix} 1&c\\ c^*&1 \end{pmatrix}, \qquad \lvert c\rvert\leq1.$$

For

$$\sigma_x = \begin{pmatrix} 0&1\\ 1&0 \end{pmatrix},$$ $$\langle\sigma_x\rangle = \operatorname{Tr}(\rho\sigma_x) = \operatorname{Re}c.$$

The diagonal entries alone do not determine this expectation. The off-diagonal coherence contributes.

Local state of a Bell pair

For

$$\lvert\Phi^+\rangle = \frac{ \lvert00\rangle+\lvert11\rangle }{ \sqrt2 },$$

the reduced state of either qubit is

$$\rho_A = \frac{I}{2}.$$

Therefore

$$\langle\sigma_i\otimes I\rangle = \operatorname{Tr} \left( \frac{I}{2}\sigma_i \right) = 0$$

for $i=x,y,z$, even though the joint state has nontrivial correlations.

Derivation and canonical home

Trace Rule for Expectation Values owns the derivation, component formulas, ensemble interpretation, bounds, POVM probabilities, subsystem reduction, and infinite-dimensional qualification.

Density Operators owns state validity and preparation meaning. Reduced Density Matrices and Partial Trace own the subsystem construction.

Worked examples

• Density-Matrix Examples
• Expectation-Value Examples
• Bloch Sphere
• Subsystems and Local Observables

Common mistakes

• Treating the diagonal of $\rho$ as a basis-independent probability table.
• Ignoring off-diagonal terms in $\sum_{i,j}\rho_{ij}A_{ji}$.
• Using matrix entries of $\rho$ and $A$ from different bases.
• Forgetting positivity because Hermiticity and trace one happen to hold.
• Reordering noncommuting factors inside a trace rather than cycling them.
• Using an effect $E_a$ as though it uniquely specified state update.
• Assuming an ensemble decomposition of $\rho$ is unique.
• Suppressing the identity in $A\otimes I_B$ until subsystem order becomes ambiguous.
• Trying to compute correlations from reduced states alone.
• Normalizing a conditional branch before saving its probability.
• Applying finite-dimensional trace manipulations to unbounded operators without checking existence.
• Inferring a Hamiltonian or dynamics from the instantaneous state.

Related formulas

• Expectation Value
• Born Rule
• Variance and Standard Deviation
• Partial Trace
• Kraus Map
• Lindblad Equation
• Von Neumann Entropy
• Bloch Vector

References

• L. E. Ballentine, Quantum Mechanics: A Modern Development, 2nd ed., World Scientific, 2014, chs. 2 and 3.
• J. J. Sakurai and J. Napolitano, Modern Quantum Mechanics, 3rd ed., Cambridge University Press, 2020, ch. 1.
• M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information, 10th anniversary ed., Cambridge University Press, 2010, chs. 2 and 8.
• A. S. Holevo, Probabilistic and Statistical Aspects of Quantum Theory, 2nd ed., Edizioni della Normale, 2011, chs. 1 and 2.

Exercises

Exercise 1: off-diagonal contribution

Let

$$\rho = \begin{pmatrix} 3/4&1/4\\ 1/4&1/4 \end{pmatrix}, \qquad A = \begin{pmatrix} 1&2\\ 2&-1 \end{pmatrix}.$$

Compute $\langle A\rangle_\rho$ and identify the diagonal and off-diagonal contributions.

Solution

Using

$$\operatorname{Tr}(\rho A) = \sum_{i,j} \rho_{ij}A_{ji},$$

the diagonal contribution is

$$\frac34(1) + \frac14(-1) = \frac12.$$

The off-diagonal contribution is

$$\frac14(2) + \frac14(2) = 1.$$

Therefore

$$\langle A\rangle_\rho = \frac32.$$

The matrix $A$ has eigenvalues $\pm\sqrt5$, so the answer lies inside its spectral range as required.

Exercise 2: qubit along an arbitrary axis

A qubit has Bloch vector $\mathbf r$ and is measured with

$$\sigma_{\mathbf n} = \mathbf n\cdot\boldsymbol{\sigma}, \qquad \lVert\mathbf n\rVert=1.$$

Find the expectation and the two outcome probabilities.

Solution

The qubit trace shortcut gives

$$\langle\sigma_{\mathbf n}\rangle = \mathbf r\cdot\mathbf n.$$

The effects are

$$P_\pm = \frac12 \left( I\pm\sigma_{\mathbf n} \right),$$

so

$$p(\pm) = \operatorname{Tr}(\rho P_\pm) = \frac12 \left( 1\pm\mathbf r\cdot\mathbf n \right).$$

Their difference is the expectation and their sum is one.

Exercise 3: conditional expectation

An unnormalized measurement branch is

$$\widetilde\rho_a = \begin{pmatrix} 0.30&0\\ 0&0.10 \end{pmatrix}.$$

Find the outcome probability, normalized conditional state, and conditional expectation of $\sigma_z$.

Solution

The outcome probability is

$$p(a) = \operatorname{Tr}\widetilde\rho_a = 0.40.$$

The conditional state is

$$\rho_a = \frac{\widetilde\rho_a}{0.40} = \begin{pmatrix} 3/4&0\\ 0&1/4 \end{pmatrix}.$$

Therefore

$$\langle\sigma_z\rangle_a = \operatorname{Tr}(\rho_a\sigma_z) = \frac34-\frac14 = \frac12.$$

Equivalently,

$$\langle\sigma_z\rangle_a = \frac{ \operatorname{Tr} (\widetilde\rho_a\sigma_z) }{ \operatorname{Tr}\widetilde\rho_a }.$$