Density Operators

A state vector $|\psi\rangle$ is a luxury nature grants only in the ideal case. It presumes a system perfectly isolated, prepared in one definite ray of Hilbert space, with no classical uncertainty and no entanglement to anything outside. Real laboratories violate every clause: a trapped ion picks up which-path correlations with stray photons, a superconducting qubit inherits classical randomness from an imperfect pulse, and any single half of an entangled pair has no vector of its own. Yet each still answers every measurement with definite, repeatable statistics. The density operator $\rho$ is built to carry exactly those statistics — no more, no less. Formally, $\rho$ is any operator on $\mathcal{H}$ that is Hermitian ( $\rho=\rho^\dagger$ ), positive semidefinite ( $\rho\ge 0$ ), and normalized to unit trace ( $\mathrm{Tr}\,\rho=1$ ). These axioms are necessary and sufficient: every physical preparation yields such an operator, and every such operator can be prepared. Positivity keeps outcome probabilities from going negative; unit trace makes them sum to one; Hermiticity makes the eigenvalues real probabilities. The payoff is that $\rho$ linearizes all of quantum statistics: expectation values, outcome probabilities, time evolution, and entropy all become trace-linear functionals of one operator. Mixing preparations becomes convex combination, so states form a convex body whose extreme points are the pure states — the geometric picture at the heart of quantum information.

Key takeaways

The density operator $\rho$ is a Hermitian, positive semidefinite, unit-trace operator that encodes the most general state of a quantum system — pure or statistically mixed — on an equal footing.
A pure state is the rank-one projector $\rho=|\psi\rangle\langle\psi|$ ; a classical mixture of pure states $|\psi_i\rangle$ with probabilities $p_i$ is the convex sum $\rho=\sum_i p_i|\psi_i\rangle\langle\psi_i|$ .
Every prediction flows from the trace rule $\langle A\rangle=\mathrm{Tr}(\rho A)$ ; measurement gives outcome $m$ with probability $\mathrm{Tr}(P_m\rho)$ and updates $\rho\mapsto P_m\rho P_m/\mathrm{Tr}(P_m\rho)$ .
Purity $\mathrm{Tr}(\rho^2)$ ranges from $1$ (pure) down to $1/d$ (maximally mixed); for a qubit it equals $\tfrac12(1+|\vec r|^2)$ where $\vec r$ is the Bloch vector.
Different ensembles can produce the identical $\rho$ and are then physically indistinguishable — the Hughston–Jozsa–Wootters freedom — and the partial trace $\rho_A=\mathrm{Tr}_B(\rho_{AB})$ extracts a subsystem's state.
Key idea: A density operator is the unique linear object that reproduces every measurement statistic of a quantum system; pure states are its rank-one extreme points, mixed states are convex combinations, and all predictions follow from $\langle A\rangle=\mathrm{Tr}(\rho A)$ .

The core insight

A density operator is the unique linear object that reproduces every measurement statistic of a quantum system; pure states are its rank-one extreme points and mixed states are convex combinations, with all observable predictions given by $\langle A\rangle=\mathrm{Tr}(\rho A)$ .

Where this lives today

The density operator is the everyday language of quantum technology, not a formal nicety. Quantum state tomography reconstructs an unknown $\rho$ from a tomographically complete set of measurements — for one qubit, $\langle\sigma_x\rangle,\langle\sigma_y\rangle,\langle\sigma_z\rangle$ pin the Bloch vector and hence $\rho$ , with positivity $\rho\ge 0$ separating a physical reconstruction from noise. Benchmarking lives here too: the purity $\mathrm{Tr}(\rho^2)$ grades how much coherence a processor retains, and the decay of coherences sets the $T_2$ times that cap circuit depth. Open-system evolution is described by Lindblad master equations and quantum channels acting directly on $\rho$ — the only object that can track the loss of purity unitary dynamics can never produce. And the von Neumann entropy of a reduced density operator is the entanglement entropy now central across condensed-matter physics, quantum field theory, and the holographic study of black holes, where Ryu–Takayanagi ties $S(\rho)$ of a boundary region to a bulk area. Wherever a quantum system is noisy, entangled, or observed, its state is a density operator.

The physical picture

Treat $\rho$ as a complete catalogue of betting odds: name any observable, and $\rho$ returns the full distribution of its outcomes. Two preparations that yield identical odds for every conceivable measurement are, by fiat of quantum mechanics, the same physical state and share the same $\rho$ — there is no hidden extra fact. This is why the density operator, not the state vector, is the true carrier of physical content: it keeps exactly what experiments can access and discards the rest, including global phase and the ensemble that produced it. The geometry holds all of this in one image. The set of valid $\rho$ is a convex body: mix two states with weights $p$ and $1-p$ and you land on the chord between them. The pure states are the extreme points — the ones you cannot write as a nontrivial mixture — and for a single qubit that body is the familiar Bloch ball, pure states on the surface and the maximally mixed state $I/2$ at dead center. The first figure ties position in the ball to purity in one equation.

Figure

The Bloch ball of qubit states: pure states live on the surface at radius r = 1 while mixed states fill the interior, the maximally mixed state I∕2 sitting dead center at r = 0 — purity is Tr(ρ²) = ½(1 + r²), so the marked interior state at r = 0.707 has purity ¾ and eigenvalues λ± = ½(1 ± r)

Classical probability lives on the diagonal of $\rho$ ; quantum coherence lives off it. The diagonal entries $\rho_{ii}$ are ordinary probabilities of outcome $i$ ; the off-diagonal $\rho_{ij}$ record the phase relationships that make interference possible. This is the sharp line between a superposition and a mixture: both can share a diagonal, but only the superposition carries off-diagonal coherences. Decoherence — the everyday reason large systems look classical — is precisely the decay of those off-diagonal terms while the populations survive. The second figure follows one qubit as coherence bleeds away and purity slides to its floor.

Figure

Dephasing drives a pure qubit to the maximally mixed state: the plus-state starts with coherence ρ₀₁ = ½ and purity 1, but at rate 1∕T₂ the coherence shrinks as ρ₀₁ = ½ e^(−t∕T₂), so purity Tr(ρ²) = ½(1 + e^(−2t∕T₂)) falls toward the floor ½ — at t = T₂ it is already 0.568

The core equations

Three axioms define what may count as a state. Each clause is forced by a distinct statistical requirement, worked out in the first derivation below.

\rho=\rho^\dagger,\qquad \rho\ge 0,\qquad \mathrm{Tr}\,\rho=1 \qquad (1)

where: $\rho$ — density operator on Hilbert space $\mathcal{H}$ ; $\rho\ge 0$ — positive semidefinite, meaning $\langle\phi|\rho|\phi\rangle\ge 0$ for every $|\phi\rangle$ ; $\mathrm{Tr}$ — trace, the basis-independent sum of diagonal elements. Read together: $\rho$ has real, non-negative eigenvalues summing to one — a genuine probability distribution. The most common mixed state arises from classical preparation uncertainty: with probability $p_i$ the device emits $|\psi_i\rangle$ . Averaging the projectors gives the operator that reproduces that ensemble's statistics.

\rho=\sum_i p_i\,|\psi_i\rangle\langle\psi_i|,\qquad p_i\ge 0,\quad \sum_i p_i=1 \qquad (2)

where: $p_i$ — classical probability of preparing $|\psi_i\rangle$ ; $|\psi_i\rangle$ — normalized state vectors that need not be orthogonal, nor even linearly independent. These projectors are Hermitian and positive, so any convex mixture satisfies equation (1). Every empirical prediction reduces to a single trace — the master formula. Given $\rho$ and an observable $A$ , the expected value is fixed with no reference to any underlying ensemble.

\langle A\rangle=\mathrm{Tr}(\rho A) \qquad (3)

where: $A$ — a Hermitian observable; $\langle A\rangle$ — its expectation value in the state $\rho$ . Because the trace is linear, expectation values are linear in $\rho$ — the property that makes the state space convex. Measurement enters through orthogonal projectors summing to the identity. The outcome probability is again a trace, and a definite outcome resets the state by projection and renormalization.

p(m)=\mathrm{Tr}(P_m\,\rho),\qquad \rho\;\xrightarrow{\,m\,}\;\frac{P_m\,\rho\,P_m}{\mathrm{Tr}(P_m\,\rho)} \qquad (4)

where: $P_m$ — orthogonal projector onto the outcome- $m$ eigenspace, with $\sum_m P_m=I$ and $P_mP_{m'}=\delta_{mm'}P_m$ ; $p(m)$ — probability of outcome $m$ . The update rule generalizes wavefunction collapse and reduces to it for pure states. A single number diagnoses how mixed a state is: purity, the trace of $\rho^2$ , equals the sum of squared eigenvalues and is pinned between two extremes set by the dimension.

\mathrm{Tr}(\rho^2)=\sum_i \lambda_i^2,\qquad \frac1d\le \mathrm{Tr}(\rho^2)\le 1 \qquad (5)

where: $\lambda_i$ — eigenvalues of $\rho$ (a probability distribution); $d$ — dimension of $\mathcal{H}$ . The upper bound $1$ is reached only by pure states ( $\rho^2=\rho$ ), the lower bound $1/d$ only by $I/d$ . For a qubit the state space collapses onto a real three-vector: expanding $\rho$ in the identity and three Pauli operators turns any $2\times2$ density matrix into a point of the unit ball.

\rho=\tfrac12\left(I+r_x\sigma_x+r_y\sigma_y+r_z\sigma_z\right),\qquad r_k=\mathrm{Tr}(\rho\,\sigma_k) \qquad (6)

where: $\sigma_k$ — Pauli operators; $\vec r$ — Bloch vector with $|\vec r|\le 1$ , pure exactly when $|\vec r|=1$ . In this form purity is $\mathrm{Tr}(\rho^2)=\tfrac12(1+|\vec r|^2)$ and the eigenvalues are $\tfrac12(1\pm|\vec r|)$ . When a system is part of a larger whole, its state is obtained by tracing out the rest — the unique linear map that preserves all local statistics.

\rho_A=\mathrm{Tr}_B(\rho_{AB}) \qquad (7)

where: $\rho_{AB}$ — joint state on $\mathcal{H}_A\otimes\mathcal{H}_B$ ; $\mathrm{Tr}_B$ — partial trace over $B$ , acting as $\mathrm{Tr}_B(|a\rangle\langle a'|\otimes|b\rangle\langle b'|)=\langle b'|b\rangle\,|a\rangle\langle a'|$ . A pure global state generically gives a mixed $\rho_A$ — the signature of entanglement. Finally, mixedness gets a canonical measure: the von Neumann entropy, the Shannon entropy of the eigenvalue distribution, underlies all of quantum information theory.

S(\rho)=-\mathrm{Tr}(\rho\log\rho)=-\sum_i\lambda_i\log\lambda_i \qquad (8)

where: $S(\rho)$ — entropy in bits when $\log=\log_2$ ; $\lambda_i$ — eigenvalues of $\rho$ . It vanishes for pure states and reaches its maximum $\log d$ for $I/d$ ; for a subsystem it is the entanglement entropy.

Derivation AFrom the ensemble average to the trace rule

The goal is to show that the trace formula $\langle A\rangle=\mathrm{Tr}(\rho A)$ is not an extra postulate but a forced consequence of two things we already accept: that a mixed preparation averages the quantum expectations of its pure components with their classical weights, and that the trace is linear and cyclic. This is what makes $\rho$ the complete summary of a state — every observable's prediction depends on the ensemble only through the one operator $\rho$ .

Step 1. Start from the operational meaning of equation (2): with probability $p_i$ the source emits $|\psi_i\rangle$ , which predicts $\langle\psi_i|A|\psi_i\rangle$ , so the classical average is $\langle A\rangle=\sum_i p_i\,\langle\psi_i|A|\psi_i\rangle$ . Why: this is the law of total expectation — classical probabilities multiply the conditional quantum expectations.

Step 2. Rewrite each scalar as a trace: $\langle\psi_i|A|\psi_i\rangle=\mathrm{Tr}\!\big(A\,|\psi_i\rangle\langle\psi_i|\big)$ , since the trace against $A$ contracts the outer product back to the number. Why: cyclicity of the trace converts a bracket into a trace we can collect.

Step 3. Pull the sum inside by linearity: $\langle A\rangle=\sum_i p_i\,\mathrm{Tr}\!\big(A\,|\psi_i\rangle\langle\psi_i|\big)=\mathrm{Tr}\!\Big(A\sum_i p_i|\psi_i\rangle\langle\psi_i|\Big)=\mathrm{Tr}(\rho A)$ . The ensemble has collapsed into the single operator $\rho$ . Why: the observable couples to the ensemble only through $\rho$ , so any two ensembles with the same $\rho$ predict identical expectations — the reason ensembles are interchangeable.

Step 4. Check the pure limit. For $\rho=|\psi\rangle\langle\psi|$ , insert a basis: $\mathrm{Tr}(|\psi\rangle\langle\psi|A)=\sum_i\langle i|\psi\rangle\langle\psi|A|i\rangle=\langle\psi|A|\psi\rangle$ using $\sum_i|i\rangle\langle i|=I$ . Why: the rule must contain the ordinary state-vector rule as a special case, and it does. *Verified: reproduces Nielsen and Chuang eq. (2.190); the pure-state check $\mathrm{Tr}(|\psi\rangle\langle\psi|A)=\langle\psi|A|\psi\rangle$ recomputed above.*

Go deeper · Derivation B · 1 minThe unitary freedom in ensembles that share one state

Derivation BThe unitary freedom in ensembles that share one state

The aim is to prove precisely when two different lists of pure states describe the same physical state: the Hughston–Jozsa–Wootters (equivalently Schrödinger) theorem says two ensembles give the same $\rho$ if and only if their subnormalized vectors are related by a unitary matrix. This is the formal engine behind ensemble indistinguishability, the no-signalling property of entanglement, and the security of key distribution.

Step 1. Absorb the probabilities into the vectors: define $|\tilde\psi_i\rangle=\sqrt{p_i}\,|\psi_i\rangle$ and $|\tilde\phi_j\rangle=\sqrt{q_j}\,|\phi_j\rangle$ , so $\rho=\sum_i|\tilde\psi_i\rangle\langle\tilde\psi_i|=\sum_j|\tilde\phi_j\rangle\langle\tilde\phi_j|$ . Why: folding $\sqrt{p_i}$ into the ket turns a weighted sum of projectors into a plain sum, keeping the algebra clean.

Step 2. Prove sufficiency. Suppose $|\tilde\psi_i\rangle=\sum_j u_{ij}|\tilde\phi_j\rangle$ for a unitary $u$ . Then $\sum_i|\tilde\psi_i\rangle\langle\tilde\psi_i|=\sum_i\sum_{j,k}u_{ij}u^{*}_{ik}\,|\tilde\phi_j\rangle\langle\tilde\phi_k|$ . Why: expanding the outer product exposes the contraction $\sum_i u_{ij}u^{*}_{ik}$ that unitarity will collapse.

Step 3. Apply column orthonormality, $\sum_i u_{ij}u^{*}_{ik}=\delta_{jk}$ , so the double sum reduces to $\sum_j|\tilde\phi_j\rangle\langle\tilde\phi_j|=\rho$ : a unitary mixing leaves $\rho$ unchanged. Why: this is exactly the condition that a change of ensemble be invisible to $\rho$ .

Step 4. Prove necessity through purification. Any $\rho$ has a purification $|\Psi\rangle_{AB}$ on an ancilla, and any two purifications differ by a unitary on the ancilla alone. Measuring the ancilla in one basis realizes $\{|\tilde\psi_i\rangle\}$ ; measuring in a rotated basis realizes $\{|\tilde\phi_j\rangle\}$ , the rotation being $u$ . Pad the shorter list with zero vectors so $u$ is square. Why: the ancilla stores the which-preparation information, and the freedom to read it in any basis is the physical origin of the unitary freedom. *Verified: reproduces Nielsen and Chuang Theorem 2.6 (Exercise 2.81); the unitarity contraction $\sum_i u_{ij}u^{*}_{ik}=\delta_{jk}$ is the crux to re-check.*

Go deeper · Derivation C · 1 minTracing out one half of a Bell pair leaves maximal mixing

Derivation CTracing out one half of a Bell pair leaves maximal mixing

This derivation carries out the partial trace explicitly for the Bell state and shows the reduced state of one qubit is $I/2$ , even though the global state is pure. It is the cleanest demonstration that a subsystem of a pure entangled state has no state vector of its own, quantitatively linking entanglement to the mixedness of the parts — the bridge from this article to entropy and quantum channels.

Step 1. Form the global projector. For $|\Phi^+\rangle=\tfrac{1}{\sqrt2}(|00\rangle+|11\rangle)$ , $\rho_{AB}=\tfrac12\big(|00\rangle\langle00|+|00\rangle\langle11|+|11\rangle\langle00|+|11\rangle\langle11|\big)$ . Why: the outer product produces two diagonal population terms and two off-diagonal coherences, and the fate of those coherences under the trace is the whole story.

Step 2. Trace out $B$ using $\mathrm{Tr}_B(|a\rangle_A|b\rangle_B\langle c|_A\langle d|_B)=\langle d|b\rangle\,|a\rangle\langle c|$ . The cross terms vanish: $\mathrm{Tr}_B(|00\rangle\langle11|)=\langle1|0\rangle\,|0\rangle\langle1|=0$ , and likewise its conjugate. Why: the partial trace contracts the $B$ indices with an inner product, and $\langle1|0\rangle=0$ annihilates exactly the coherences that encoded the entanglement.

Step 3. Keep the surviving diagonal terms $\mathrm{Tr}_B(|00\rangle\langle00|)=|0\rangle\langle0|$ and $\mathrm{Tr}_B(|11\rangle\langle11|)=|1\rangle\langle1|$ , each weighted by $\tfrac12$ : $\rho_A=\tfrac12\big(|0\rangle\langle0|+|1\rangle\langle1|\big)=\tfrac12 I$ . Why: with the coherences gone, only equal populations remain — complete local ignorance.

Step 4. Read off the diagnostics. The Bloch vector is $\vec r=0$ , the purity is $\mathrm{Tr}(\rho_A^2)=\tfrac12$ (the qubit minimum), and the von Neumann entropy is $S(\rho_A)=1$ bit. Why: a maximally entangled pure global state produces maximally mixed, maximal-entropy parts — the quantitative meaning of "maximally entangled." *Verified: standard result; the partial trace and $\mathrm{Tr}(\rho_A^2)=1/2$ , $S(\rho_A)=1$ bit were recomputed numerically for $d=2$ .*

Limiting cases

The corners and center of the state space expose $\rho$ 's structure at a glance.

Case	Result	Meaning
Pure state	$\rho=\vert\psi\rangle\langle\psi\vert,\ \ \rho^2=\rho,\ \ \mathrm{Tr}(\rho^2)=1$	A rank-one projector; the extreme points of the state space, equivalent to a state vector up to global phase.
Maximally mixed	$\rho=\dfrac{I}{d},\ \ \mathrm{Tr}(\rho^2)=\dfrac1d,\ \ S=\log d$	Complete ignorance in dimension $d$ ; the unique unitary-invariant state, sitting at the center of the convex body.
Qubit Bloch form	$\rho=\tfrac12(I+\vec r\cdot\vec\sigma),\ \ \vert\vec r\vert\le 1$	Every qubit state is a point of the unit ball; surface is pure, center is $I/2$ , radius sets purity.
Diagonal (classical)	$\rho=\sum_i p_i\,\vert i\rangle\langle i\vert$	With no off-diagonal coherences, $\rho$ is just a classical probability distribution over the basis $\{\|i\rangle\}$ .

Common trapA classical mixture and a coherent superposition are the same because their diagonals match

They are not the same, and confusing them is the most common error with density operators. Take the coherent superposition $|+\rangle=\tfrac{1}{\sqrt2}(|0\rangle+|1\rangle)$ , with $\rho_+=\tfrac12\begin{pmatrix}1&1\\1&1\end{pmatrix}$ , and the classical mixture $\rho_{\text{mix}}=\tfrac12(|0\rangle\langle0|+|1\rangle\langle1|)=\tfrac12\begin{pmatrix}1&0\\0&1\end{pmatrix}$ . Their diagonals are identical $(\tfrac12,\tfrac12)$ , so a $Z$ -measurement cannot tell them apart. But $\rho_+$ is pure ( $\mathrm{Tr}(\rho_+^2)=1$ , coherence $\tfrac12$ ) while $\rho_{\text{mix}}$ is maximally mixed ( $\mathrm{Tr}(\rho_{\text{mix}}^2)=\tfrac12$ , no coherence). In the $X$ basis they disagree completely: $\rho_+$ gives $+$ with certainty, $\rho_{\text{mix}}$ gives $+$ and $-$ with equal odds. The off-diagonal terms are physical; erasing them is exactly what decoherence does.

Worked examples

1 · From a non-orthogonal ensemble to purity, Bloch vector, and entropy

A source emits $|0\rangle$ with probability $\tfrac12$ and $|+\rangle=\tfrac{1}{\sqrt2}(|0\rangle+|1\rangle)$ with probability $\tfrac12$ . Build $\rho$ , and compute its purity, Bloch vector, eigenvalues, and von Neumann entropy.

Step 1. Assemble $\rho$ from equation (2). Using $|0\rangle\langle0|=\begin{pmatrix}1&0\\0&0\end{pmatrix}$ and $|+\rangle\langle+|=\tfrac12\begin{pmatrix}1&1\\1&1\end{pmatrix}$ ,

\rho=\tfrac12\begin{pmatrix}1&0\\0&0\end{pmatrix}+\tfrac12\cdot\tfrac12\begin{pmatrix}1&1\\1&1\end{pmatrix}=\tfrac14\begin{pmatrix}3&1\\1&1\end{pmatrix}.

It is Hermitian with trace $1$ , and its off-diagonal $\rho_{01}=\tfrac14$ shows the non-orthogonal ensemble left a genuine coherence. $\rho=\tfrac14\begin{pmatrix}3&1\\1&1\end{pmatrix}$ .

Step 2. Purity from equation (5). Squaring, $\begin{pmatrix}3&1\\1&1\end{pmatrix}^2=\begin{pmatrix}10&4\\4&2\end{pmatrix}$ , so $\mathrm{Tr}(\rho^2)=\tfrac{1}{16}(10+2)=\tfrac{12}{16}=\tfrac34$ . Since $\tfrac34<1$ , the state is mixed.

Step 3. Bloch vector from equation (6): $r_x=\mathrm{Tr}(\rho\sigma_x)=2\rho_{01}=\tfrac12$ , $r_y=2\,\mathrm{Im}(\rho_{10})=0$ , $r_z=\rho_{00}-\rho_{11}=\tfrac34-\tfrac14=\tfrac12$ . So $\vec r=(\tfrac12,0,\tfrac12)$ and $|\vec r|=\tfrac{1}{\sqrt2}\approx 0.707$ . Cross-check: $\tfrac12(1+|\vec r|^2)=\tfrac12(1+\tfrac12)=\tfrac34$ , matching the purity. $\vec r=(\tfrac12,0,\tfrac12),\ |\vec r|\approx0.707$ .

Step 4. Eigenvalues and entropy. For a qubit $\lambda_\pm=\tfrac12(1\pm|\vec r|)=\tfrac12(1\pm0.707)$ , giving $\lambda_+\approx0.854$ , $\lambda_-\approx0.146$ . Then $S(\rho)=-0.854\log_2 0.854-0.146\log_2 0.146\approx\mathbf{0.60\ bits}$ — partway between pure ( $S=0$ ) and maximally mixed ( $S=1$ ).

2 · The reduced state and entanglement entropy of a Bell pair

Two qubits are in $|\Phi^+\rangle=\tfrac{1}{\sqrt2}(|00\rangle+|11\rangle)$ . Find the reduced state $\rho_A$ , its purity, and the entanglement entropy. Then check the no-signalling statement.

Step 1. By the third derivation, $\rho_A=\mathrm{Tr}_B(|\Phi^+\rangle\langle\Phi^+|)=\tfrac12 I$ : the global coherences were killed by orthogonality of the $B$ indices. $\rho_A=\tfrac12 I$ .

Step 2. Purity and entropy: $\mathrm{Tr}(\rho_A^2)=\tfrac12$ , the qubit minimum, and $S(\rho_A)=-2\cdot\tfrac12\log_2\tfrac12=\mathbf{1\ bit}$ . This single bit is the entanglement entropy — the maximum possible for two qubits.

Step 3. No-signalling. If Bob measures in any basis but never reports the outcome, averaging over his outcomes returns Alice's state to $\rho_A=\tfrac12 I$ regardless of his basis, since summing the update rule (4) over $m$ leaves $\mathrm{Tr}_B(\rho_{AB})$ unchanged. Alice's $\rho_A=\tfrac12 I$ is untouched by Bob's choice — entanglement cannot signal.

3 · Watching purity decay under dephasing

A qubit is prepared pure in $|+\rangle$ and dephases: the off-diagonal element decays as $\rho_{01}(t)=\tfrac12 e^{-t/T_2}$ while populations stay fixed. Find the purity at $t=T_2$ and the final state.

Step 1. Write the state: $\rho(t)=\tfrac12\begin{pmatrix}1 & e^{-t/T_2}\\ e^{-t/T_2} & 1\end{pmatrix}$ . The diagonal is frozen at $(\tfrac12,\tfrac12)$ ; only the coherence relaxes, so the Bloch vector $\vec r=(e^{-t/T_2},0,0)$ shrinks along $x$ .

Step 2. Purity from $\tfrac12(1+|\vec r|^2)$ : $\mathrm{Tr}(\rho(t)^2)=\tfrac12(1+e^{-2t/T_2})$ . At $t=T_2$ , $e^{-2}\approx0.135$ , so $\mathrm{Tr}(\rho^2)\approx\mathbf{0.568}$ — the marked point in the second figure.

Step 3. As $t\to\infty$ the coherence vanishes and $\rho\to\tfrac12 I$ , maximally mixed with purity $\tfrac12$ and entropy $1$ bit. Dephasing carries a pure superposition to a classical mixture without ever touching the populations.

Physics of the portrait

The density operator has two independent 1927 births. Lev Landau introduced the reduced statistical operator for a subsystem coupled to a larger quantum system; the same year, John von Neumann built the general density matrix as the foundation of quantum statistical mechanics and measurement — the framework that still carries his name.

Origins · 1 minJohn von Neumann and how this was discovered

John von Neumann

1903–1957 · mathematical foundations of quantum mechanics

John von Neumann gave quantum mechanics its rigorous mathematical spine, and the density matrix was one of his defining constructions. A Budapest prodigy who published on set theory as a teenager, he reached the foundations of quantum theory in the mid-1920s just as Heisenberg's and Schrödinger's formulations were being reconciled. In a trio of 1927 papers and his 1932 monograph Mathematische Grundlagen der Quantenmechanik, he recast states as operators on Hilbert space and introduced the density operator $\rho$ (his "statistical operator") to handle exactly the cases a state vector cannot: mixtures from imperfect knowledge, and subsystems of larger wholes. He defined the entropy $S(\rho)=-\mathrm{Tr}(\rho\log\rho)$ that now bears his name, proved that measurement statistics depend only on $\rho$ , and formalized the projective measurement postulate. Lev Landau, in Leningrad the same year, reached the reduced statistical operator from coupled-system physics, so the concept has a dual origin. Von Neumann's later reach — game theory, the stored-program computer, operator algebras — is legendary, but for quantum information his 1932 axioms remain the working definition: a state is a positive, unit-trace operator, and everything measurable is a trace against it.

History

1927

— Landau introduces the reduced statistical operator for a subsystem of a larger quantum system

1927

— von Neumann formulates the general density matrix and the entropy S(ρ) = −Tr(ρ log ρ)

1932

— von Neumann's Mathematische Grundlagen der Quantenmechanik codifies states as operators

1935

— Schrödinger analyzes entanglement and the freedom to steer a remote ensemble

1993

— Hughston, Jozsa, and Wootters sharpen the ensemble-equivalence (HJW) theorem

Going further

Texts. Nielsen and Chuang, Quantum Computation and Quantum Information, §2.4 (the density operator) · Preskill, Quantum Information Lecture Notes, Ch. 2 · Watrous, The Theory of Quantum Information, Ch. 2 · Wilde, Quantum Information Theory, Ch. 4 · Kitaev, Shen, and Vyalyi, Classical and Quantum Computation, §10 · Mermin, Quantum Computer Science, Ch. 1.

Provenance. The three axioms, the trace rule $\langle A\rangle=\mathrm{Tr}(\rho A)$ , the measurement rule and update, the purity bounds $1/d\le\mathrm{Tr}(\rho^2)\le1$ , the Bloch decomposition, the partial trace, and the von Neumann entropy follow Nielsen and Chuang §2.4, Preskill Ch. 2, and Watrous Ch. 2. The ensemble-equivalence derivation reproduces Nielsen and Chuang Theorem 2.6 (the HJW theorem, Hughston–Jozsa–Wootters 1993). The reduced-Bell-state result is a standard exercise; the numbers in the worked examples — $\rho=\tfrac14\begin{pmatrix}3&1\\1&1\end{pmatrix}$ with purity $\tfrac34$ , Bloch radius $|\vec r|=1/\sqrt2$ , eigenvalues $0.854$ and $0.146$ , entropy $0.60$ bits, and the dephased purity $0.568$ at $t=T_2$ — were recomputed independently. No non-standard constants appear.

Prerequisites. The Qubit and Its State Space · Dirac Bra-Ket Notation · Observables and Hermitian Operators · Born Rule · Inner Product, Norm, and Normalization. Leads to. Pure versus Mixed States · Projective Measurements · Expectation Values and Variance.

Ask the tutor: Ask me to trace out a subsystem, reconstruct a Bloch vector, or show why two different ensembles hide behind one density operator.

Why do we need density operators when state vectors already exist?

Because state vectors describe only a special case: an isolated system with a definite, fully known preparation. Three common situations break that: a preparation with classical uncertainty, a subsystem of an entangled pair (which has no vector of its own), and an open system leaking information to its environment. In all three the system still produces definite measurement statistics, and the density operator captures them. State vectors are the pure, extreme-point special case $\rho=|\psi\rangle\langle\psi|$ ; density operators are the full state space, and modern quantum information is written in their language.

What is the difference between a superposition and a mixture?

A superposition like $|+\rangle=\tfrac{1}{\sqrt2}(|0\rangle+|1\rangle)$ is a single pure state with definite phase relationships — genuinely both outcomes at once, with off-diagonal coherences in its density operator. A mixture $\tfrac12(|0\rangle\langle0|+|1\rangle\langle1|)$ is classical ignorance: the system really is in $|0\rangle$ or $|1\rangle$ , you just do not know which. Their matrices share a diagonal but differ off-diagonal, so they agree on a $Z$ -measurement yet disagree completely on $X$ . Purity separates them: $1$ for the superposition, $\tfrac12$ for the mixture. Decoherence turns the first into the second.

How do I tell whether a given ρ is pure or mixed?

Compute the purity $\mathrm{Tr}(\rho^2)$ : it equals $1$ if and only if $\rho$ is pure, and drops below $1$ for any mixture, bottoming out at $1/d$ for the maximally mixed state. Equivalently, check whether $\rho^2=\rho$ (true only for pure projectors) or whether $\rho$ has a single nonzero eigenvalue equal to $1$ . For a qubit there is a quicker test: the state is pure exactly when the Bloch vector has $|\vec r|=1$ (on the surface) and mixed when $|\vec r|<1$ (inside), with $\mathrm{Tr}(\rho^2)=\tfrac12(1+|\vec r|^2)$ making it precise.

Why can two different ensembles give the same physical state?

Because every measurement outcome probability is $\mathrm{Tr}(P_m\rho)$ , depending on the ensemble only through $\rho$ . If two lists of pure states sum to the same $\rho$ , no experiment can tell them apart — quantum mechanics asserts there is no further fact. The Hughston–Jozsa–Wootters theorem says exactly which ensembles coincide: those whose subnormalized vectors are related by a unitary matrix. This is no technicality; it underlies the no-signalling property of entanglement and the security of quantum key distribution, where an eavesdropper holding only $\rho$ cannot recover which preparation was used.

What does the partial trace actually do, physically?

The partial trace $\rho_A=\mathrm{Tr}_B(\rho_{AB})$ answers a concrete question: if you can only ever measure subsystem $A$ and never touch $B$ , what state should you use for $A$ ? It is the unique operation preserving all of $A$ 's local statistics while discarding $B$ — physically, ignoring, losing, or being unable to access $B$ . Its striking feature is that a pure global state can yield a mixed $\rho_A$ , the tell-tale sign of entanglement: the more entangled the whole, the more mixed the part. Tracing out the environment is exactly how decoherence and dissipation enter the reduced dynamics.

What is the von Neumann entropy measuring?

The von Neumann entropy $S(\rho)=-\mathrm{Tr}(\rho\log\rho)$ is the Shannon entropy of $\rho$ 's eigenvalue distribution, measuring how mixed the state is. It is $0$ for any pure state (perfect knowledge) and reaches its maximum $\log d$ for the maximally mixed state (total ignorance). Its most important use is as the entanglement entropy: for a pure global state, $S(\rho_A)$ of a subsystem quantifies how entangled it is with the rest, giving $1$ bit for a Bell pair. It also underpins quantum Shannon theory — data compression, channel capacities, and thermodynamic entropy all reduce to $S(\rho)$ .