Pure versus Mixed States

A ket $|\psi\rangle$ is the whole story only when you have prepared a system with maximal knowledge and kept it perfectly isolated. Reality rarely obliges, and two situations break the pure-ket picture. First, you may know only that a preparation machine emitted $|\psi_i\rangle$ with classical probability $p_i$ — a coin flip you did not see. Second, and more deeply quantum, your qubit may be one half of an entangled pair whose partner has wandered into an environment you cannot touch. In neither case does a single vector reproduce the statistics, and quantum mechanics forces you up a level to the density operator $\rho$ . The pure-versus-mixed distinction is the founding dichotomy of that language. A pure state $\rho=|\psi\rangle\langle\psi|$ carries the maximal information a quantum system can hold; it is a rank-one projector and passes the idempotency test $\rho^2=\rho$ . A mixed state carries an irreducible probabilistic spread — eigenvalues strictly between $0$ and $1$ , $\rho^2\neq\rho$ . The scalar that grades them is the purity $\mathrm{Tr}(\rho^2)$ : exactly $1$ for pure states, dropping toward the floor $1/d$ as the state becomes more mixed. Two subtleties give the topic its depth. First, ensemble ambiguity: many different preparation recipes produce the identical $\rho$ , and since every prediction depends on $\rho$ alone, they are physically indistinguishable — the density operator, not the recipe, is the state. Second, the two origins of mixedness — honest classical ignorance versus entanglement with an inaccessible partner — leave exactly the same $\rho$ behind, so no local measurement tells them apart. This article builds the test, the Bloch geometry, the ensemble structure, and both origins in density-operator language throughout.

Key takeaways

A quantum state is a density operator $\rho$ : Hermitian, positive semidefinite, unit trace ( $\mathrm{Tr}\,\rho=1$ ) — the object that reproduces all measurement statistics via the Born rule.
A state is pure iff $\rho=|\psi\rangle\langle\psi|$ for a unit ket, equivalently iff $\rho^2=\rho$ (a rank-one projector); every other density operator is mixed.
The purity $\mathrm{Tr}(\rho^2)\in[1/d,\,1]$ is the scalar that separates them ( $=1$ pure, $<1$ mixed, $=1/d$ at the floor $\rho=I/d$ ); for a qubit it is $\tfrac12(1+|\vec r|^2)$ , so pure states sit on the Bloch surface and mixed ones strictly inside.
A mixed state is a classical ensemble $\rho=\sum_i p_i|\psi_i\rangle\langle\psi_i|$ of pure states — but the ensemble producing a given $\rho$ is not unique, and only $\rho$ is physical.
Mixedness has two origins no measurement distinguishes: classical ignorance of the preparation, and entanglement with an inaccessible environment (the reduced state via partial trace).
Key idea: pure $\iff \rho^2=\rho \iff \mathrm{Tr}(\rho^2)=1 \iff \rho$ has a single nonzero eigenvalue equal to $1$ ; mixed is everything else, and its spread is measured by the purity $\mathrm{Tr}(\rho^2)<1$ .

The core insight

Pure $\iff \rho^2=\rho \iff \mathrm{Tr}(\rho^2)=1 \iff \rho$ has a single nonzero eigenvalue (which equals $1$ ); mixed is everything else, with the spread measured by the purity $\mathrm{Tr}(\rho^2)<1$ .

Where this lives today

The pure/mixed boundary is the working diagnostic of every quantum-hardware lab. On superconducting and trapped-ion processors, quantum state tomography reconstructs $\rho$ , and the drop of the purity $\mathrm{Tr}(\rho^2)$ — the shrink of the Bloch radius away from $1$ — is how decoherence and $T_2$ dephasing are quantified as a single number; a "state fidelity" report is a purity statement in disguise. In entanglement science, reduced purity is an entanglement witness: for a pure bipartite state $\mathrm{Tr}(\rho_A^2)<1$ certifies entanglement, and the linear entropy $1-\mathrm{Tr}(\rho_A^2)$ (concurrence-related for two qubits) turns an abstract correlation into a local measurement. Ensemble ambiguity is load-bearing, not a curiosity: because a distant party cannot change your reduced $\rho$ by its basis choice, it is the no-signaling principle, and the same freedom (the HJW theorem) underlies quantum steering and entanglement-based QKD security. Even mixedness-as-resource has a home — the DQC1 "one clean qubit" model computes with a nearly maximally mixed register plus one pure control qubit — a reminder that purity is a budget, quantified by exactly the entropy von Neumann wrote down.

The physical picture

Read $\rho$ as a joint statement of what you know and how much you do not. A pure state is a fully specified preparation — a unit-length arrow; a mixed state reports residual probability, the arrow retracted. The purity $\mathrm{Tr}(\rho^2)$ is a sharpness dial: turn it to $1$ and the qubit collapses to one point on the Bloch surface; turn it down and the Bloch vector shrinks toward the centre, where at $|\vec r|=0$ you hold total ignorance, $\rho=I/2$ . The first figure makes this geometry concrete, with the pure-mixed boundary as the sphere's skin.

Figure

Qubit states live in the Bloch ball: pure states on the surface where the Bloch radius r = 1 and the purity Tr(ρ²) = 1, generic mixed states strictly inside, and the maximally mixed state I/2 at the centre where r = 0 and Tr(ρ²) = 0.5, all tied together by Tr(ρ²) = ½(1 + r²)

The same $\rho$ can arise two ways no laboratory can pull apart. Either someone flipped a fair coin and handed you $|0\rangle$ or $|1\rangle$ (classical ignorance), or your qubit is half of a Bell pair whose partner was carried into the environment (entanglement, decoherence). The density operator is precisely the data these stories share — the view from inside the system, blind to how the mixedness was produced. And a mixture is less than its pure parts in coherence yet equal in linear statistics: $\rho=\sum_i p_i\rho_i$ gives $\langle A\rangle=\sum_i p_i\langle A\rangle_i$ . Mixing erases the off-diagonal coherences that carry interference while preserving linear averages — which is why a mixed state behaves like a classical probability distribution over quantum states — as the second figure isolates on the smallest example.

The core equations

A quantum state, in full generality, is a density operator. The three axioms below single out exactly the operators that can serve as states: Hermiticity makes statistics real, positivity keeps probabilities non-negative, unit trace normalizes them.

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

where: $\rho$ — density operator on a $d$ -dimensional Hilbert space; $\rho\ge 0$ — positive semidefinite, $\langle\phi|\rho|\phi\rangle\ge0$ for every $|\phi\rangle$ ; $\mathrm{Tr}(\rho)=1$ — normalizes total probability. Reading: these three are jointly necessary and sufficient for $\rho$ to represent a physical preparation. The pure states are the extreme points of this set — those not writable as a nontrivial average of others. Four equivalent characterizations pin them down, each a practical test on a matrix.

\rho\ \text{pure}\iff \rho=|\psi\rangle\langle\psi|\iff \rho^2=\rho\iff \mathrm{Tr}(\rho^2)=1 \qquad (2)

where: $|\psi\rangle$ — a normalized vector, $\langle\psi|\psi\rangle=1$ ; $\rho^2=\rho$ — idempotency, the defining property of a projector; $\mathrm{Tr}(\rho^2)$ — the purity. Reading: to decide pure vs mixed, square the matrix and trace — no eigen-decomposition needed. Every mixed state can be realized as a classical ensemble of pure states with weights $p_i$ — how mixedness enters from honest ignorance of a preparation.

\rho=\sum_i p_i\,|\psi_i\rangle\langle\psi_i|,\qquad p_i\ge0,\ \sum_i p_i=1 \qquad (3)

where: $p_i$ — classical probability of preparing $|\psi_i\rangle$ ; $|\psi_i\rangle$ — pure states in the ensemble, need not be orthogonal, may outnumber $d$ . Reading: this is a recipe for $\rho$ , not a property of it — the same $\rho$ has infinitely many recipes. Whatever the recipe, all observable content is squeezed through $\rho$ — expectation values and outcome probabilities are traces against it.

\langle A\rangle=\mathrm{Tr}(\rho A),\qquad \Pr(m)=\mathrm{Tr}(\rho P_m) \qquad (4)

where: $A$ — an observable (Hermitian operator); $P_m$ — the projector onto outcome $m$ , $\sum_m P_m=I$ . Reading: because every prediction is linear in $\rho$ , two preparations with equal $\rho$ are operationally identical — the root of ensemble ambiguity. For a single qubit the state set is the solid Bloch ball, where purity becomes geometry: it increases with the Bloch radius, so the surface is exactly the pure locus.

\rho=\tfrac12\big(I+\vec r\cdot\vec\sigma\big),\qquad \mathrm{Tr}(\rho^2)=\tfrac12\big(1+|\vec r|^2\big) \qquad (5)

S(\rho)=-\mathrm{Tr}(\rho\log_2\rho)=-\sum_k\lambda_k\log_2\lambda_k \qquad (6)

where: $\lambda_k$ — the eigenvalues of $\rho$ ; $S=0$ for a pure state; $S=\log_2 d$ for the maximally mixed state. Reading: $S$ vanishes exactly when $\rho$ is pure and is maximized by $\rho=I/d$ , mirroring the purity ordering but weighting the spectrum logarithmically. The second, deeper origin of mixedness is not ignorance but entanglement: tracing an inaccessible subsystem out of a pure joint state generally leaves a mixed reduced state.

\rho_A=\mathrm{Tr}_B\big(|\Psi\rangle_{AB}\langle\Psi|\big),\qquad \frac{1}{d}\le \mathrm{Tr}(\rho^2)\le 1 \qquad (7)

where: $|\Psi\rangle_{AB}$ — a pure state of the joint system; $\mathrm{Tr}_B$ — partial trace over the unobserved subsystem $B$ ; $\rho_A$ — reduced state of $A$ , mixed iff $|\Psi\rangle_{AB}$ is entangled. Reading: the purity is bounded by $1/d$ (saturated by $I/d$ ) below and $1$ (pure states) above — the inequality organizing every limiting case.

Derivation AWhy idempotency, rank-one, and unit purity are the same condition

The purity test $\mathrm{Tr}(\rho^2)=1$ , the projector test $\rho^2=\rho$ , and the statement " $\rho$ has exactly one nonzero eigenvalue" look like three different claims, yet they cut out precisely the same density operators. The goal is to prove all three equivalent from one fact — that a density operator's eigenvalues form a probability distribution — so you may use whichever test is cheapest.

Step 1. Because $\rho$ is Hermitian and positive with unit trace, the spectral theorem gives $\rho=\sum_k\lambda_k|k\rangle\langle k|$ with orthonormal eigenvectors, $\lambda_k\ge0$ , $\sum_k\lambda_k=1$ . Why: the eigenvalues literally form a probability distribution — the constraint that forces the pure case to be an all-or-nothing spectrum.

Step 2. Squaring on the orthonormal eigenbasis kills every cross term, leaving $\rho^2=\sum_k\lambda_k^2|k\rangle\langle k|$ . Why: this reduces operator statements about $\rho$ to statements about the numbers $\lambda_k$ .

Step 3. Idempotency $\rho^2=\rho$ holds iff $\lambda_k^2=\lambda_k$ , i.e. each $\lambda_k\in\{0,1\}$ ; with $\sum_k\lambda_k=1$ , exactly one eigenvalue is $1$ and the rest vanish, so $\rho=|\psi\rangle\langle\psi|$ is a rank-one projector. Why: the probability constraint turns $\lambda^2=\lambda$ into uniqueness — one and only one occupied eigenvector.

Step 4. For the purity, $\mathrm{Tr}(\rho^2)=\sum_k\lambda_k^2\le\big(\sum_k\lambda_k\big)^2=1$ , since the cross terms $\sum_{k\neq l}\lambda_k\lambda_l\ge0$ ; equality forces all but one $\lambda_k$ to zero — the same rank-one condition. Why: $\mathrm{Tr}(\rho^2)=1$ is not a weaker proxy but an exact restatement of purity, so $<1$ certifies mixedness. *Verified: eigenvalues $(0.8,0.2)$ give $\mathrm{Tr}(\rho^2)=0.68<1$ , $\rho^2\neq\rho$ ; $(1,0)$ give $\mathrm{Tr}(\rho^2)=1$ , $\rho^2=\rho$ (N&C §2.4).*

Go deeper · Derivation B · 1 minMany ensembles, one density operator

Derivation BMany ensembles, one density operator

The ensemble $\{p_i,|\psi_i\rangle\}$ is a story about how $\rho$ was made, and quantum mechanics erases that story on contact: distinct ensembles produce the identical $\rho$ and are indistinguishable by any measurement. The goal is to exhibit this on the maximally mixed qubit, then state exactly which ensembles collapse to one $\rho$ — the freedom underlying no-signaling and quantum steering.

Step 2. Take the diagonal ensemble instead: a fair coin selects $|+\rangle$ or $|-\rangle$ . With $|\pm\rangle\langle\pm|=\tfrac12\!\left(\begin{smallmatrix}1&\pm1\\\pm1&1\end{smallmatrix}\right)$ , the $\pm\tfrac12$ coherences cancel on averaging and $\tfrac12|+\rangle\langle+|+\tfrac12|-\rangle\langle-|=\tfrac12 I$ . Why: two ensembles from non-overlapping bases give the same operator — the preparation basis is not encoded in $\rho$ .

Step 3. Conclude operationally: since $\Pr(m)=\mathrm{Tr}(\rho P_m)$ depends only on $\rho$ , every measurement gives identical statistics — no experiment reveals which coin was flipped. Why: the precise sense in which the ensemble is not part of the state; only $\rho$ is.

Step 4. State the general rule (Hughston–Jozsa–Wootters): two ensembles of subnormalized vectors $\{\sqrt{p_i}\,|\psi_i\rangle\}$ and $\{\sqrt{q_j}\,|\phi_j\rangle\}$ give the same $\rho$ iff they are related by a rectangular isometry (a unitary after zero-padding). Why: this turns "infinitely many recipes" into a checkable equivalence, and is the theorem behind an EPR party steering a distant ensemble by its local basis choice. *Verified: both ensembles sum entrywise to $\left(\begin{smallmatrix}\frac12&0\\0&\frac12\end{smallmatrix}\right)=I/2$ ; HJW unitary freedom is standard (N&C §2.4).*

Go deeper · Derivation C · 1 minEntanglement makes the local state mixed

Derivation CEntanglement makes the local state mixed

The deepest source of mixedness is not classical: a subsystem of a pure entangled whole is itself mixed, its mixedness measuring the entanglement. The goal is to trace one qubit out of a Bell pair and watch a globally pure, perfectly known state give a locally maximally mixed reduced state — coherences dying under the partial trace.

Step 1. Start from the Bell state $|\Phi^+\rangle=\tfrac1{\sqrt2}(|00\rangle+|11\rangle)$ , whose joint density operator $|\Phi^+\rangle\langle\Phi^+|=\tfrac12(|00\rangle+|11\rangle)(\langle00|+\langle11|)$ is pure, $\mathrm{Tr}(\rho_{AB}^2)=1$ . Why: we begin with maximal knowledge of the whole, so any mixedness cannot be blamed on ignorance of the global state.

Step 3. Partial-trace over $B$ via $\mathrm{Tr}_B(|ij\rangle\langle kl|)=|i\rangle\langle k|\,\langle l|j\rangle$ ; the coherence terms carry $\langle0|1\rangle=0$ and vanish, leaving $\rho_A=\tfrac12(|0\rangle\langle0|+|1\rangle\langle1|)=\tfrac{I}{2}$ . Why: the inaccessible partner is an environment whose orthogonal records decohere $A$ into a classical-looking mixture.

Step 4. Compare purities: $\mathrm{Tr}(\rho_A^2)=\tfrac12<1=\mathrm{Tr}(\rho_{AB}^2)$ . Local mixedness with global purity is the signature of entanglement, and the linear entropy $1-\tfrac12=\tfrac12$ is maximal. Why: the bridge to entanglement detection — reduced purity below one certifies correlation with the outside, no access to $B$ needed. *Verified: recomputing $\mathrm{Tr}_B|\Phi^+\rangle\langle\Phi^+|$ gives $I/2$ with reduced purity $1/2$ , the qubit minimum (N&C §2.4.3).*

Limiting cases

The single inequality $1/d\le\mathrm{Tr}(\rho^2)\le1$ organizes every state, from the sharpest ket to total ignorance.

Case	Result	Meaning
Pure state	$\mathrm{Tr}(\rho^2)=1,\ \ \rho=\vert\psi\rangle\langle\psi\vert,\ \rho^2=\rho$	Single eigenvalue $1$ ; rank-one projector; Bloch radius $\vert\vec r\vert=1$ ; von Neumann entropy $S=0$ .
Maximally mixed	$\rho=\tfrac{I}{d},\ \ \mathrm{Tr}(\rho^2)=\tfrac1d$	Flat spectrum $1/d$ ; total ignorance; Bloch radius $0$ ; $S=\log_2 d$ (one bit for a qubit).
Generic mixed qubit	$\mathrm{Tr}(\rho^2)=\tfrac12(1+\vert\vec r\vert^2),\ 0<\vert\vec r\vert<1$	Strictly inside the Bloch ball; eigenvalues $\tfrac{1\pm\vert\vec r\vert}{2}$ , both nonzero; $0<S<1$ .
Reduced state of a pure pair	$\rho_A=\mathrm{Tr}_B\vert\Psi\rangle\langle\Psi\vert$	Mixed iff $A$ and $B$ are entangled; $\mathrm{Tr}(\rho_A^2)<1$ certifies entanglement of a globally pure state.
Convex combination	$\rho=\sum_i p_i\rho_i\ \Rightarrow\ \mathrm{Tr}(\rho^2)\le\sum_i p_i\,\mathrm{Tr}(\rho_i^2)$	Mixing never increases purity; strict inequality unless all $\rho_i$ coincide.

Common trapA superposition is the same as a mixture because both are "half and half".

This is the most common error in the subject, and the second figure refutes it. The superposition $|+\rangle=\tfrac1{\sqrt2}(|0\rangle+|1\rangle)$ is a pure state: $\rho_+^2=\rho_+$ , on the Bloch surface at $|\vec r|=1$ along $x$ , giving the deterministic result $+1$ for $\sigma_x$ . The mixture $\tfrac12|0\rangle\langle0|+\tfrac12|1\rangle\langle1|=I/2$ is maximally mixed, at the Bloch centre $|\vec r|=0$ , giving a random $\pm1$ for $\sigma_x$ . They share the diagonal $\{\tfrac12,\tfrac12\}$ , so $\sigma_z$ cannot tell them apart — but they differ entirely in the off-diagonal coherences, which any $x$ -basis measurement exposes. A related trap: the ensemble $\{p_i,|\psi_i\rangle\}$ is not physical data — many ensembles give one $\rho$ , and only $\rho$ is the state.

Worked examples

1 · Is this qubit pure or mixed — and how mixed? (fully worked)

A qubit is prepared as $\rho=\tfrac12\big(I+0.6\,\sigma_z\big)$ . Classify it, and compute its purity, eigenvalues, von Neumann entropy, and linear entropy.

Step 1. Read off the Bloch vector. Comparing with $\rho=\tfrac12(I+\vec r\cdot\vec\sigma)$ gives $\vec r=(0,0,0.6)$ , so $|\vec r|=0.6<1$ . Already this places the state strictly inside the Bloch ball, hence mixed — no further work needed to classify it.

Step 2. Purity from the Bloch formula (equation 5): $\mathrm{Tr}(\rho^2)=\tfrac12(1+|\vec r|^2)=\tfrac12(1+0.36)=\mathbf{0.68}$ . Since $0.68<1$ , this confirms mixed and quantifies it.

Step 3. Eigenvalues: $\lambda_\pm=\tfrac{1\pm|\vec r|}{2}=\tfrac{1\pm0.6}{2}=\{\mathbf{0.8,\,0.2}\}$ , both nonzero, so $\rho$ is not a projector. Cross-check: $\lambda_+^2+\lambda_-^2=0.64+0.04=0.68$ , matching the purity.

Step 4. von Neumann entropy: $S=-(0.8\log_2 0.8+0.2\log_2 0.2)=\mathbf{0.7219}$ bits; linear entropy $1-0.68=\mathbf{0.32}$ . Both lie strictly between the pure ( $0$ ) and maximally mixed ( $1$ bit / $0.5$ ) values, placing this state firmly in the generic-mixed regime. Verdict: mixed, purity $0.68$ , spectrum $\{0.8,0.2\}$ , $S\approx0.72$ bit.

2 · Coherence is the difference (superposition vs mixture)

Show numerically that $|+\rangle\langle+|$ and $I/2$ are told apart by a $\sigma_x$ measurement but not by a $\sigma_z$ measurement.

Step 1. Write the matrices in the computational basis: $|+\rangle\langle+|=\tfrac12\!\left(\begin{smallmatrix}1&1\\1&1\end{smallmatrix}\right)$ and $I/2=\tfrac12\!\left(\begin{smallmatrix}1&0\\0&1\end{smallmatrix}\right)$ . They have the same diagonal $\{\tfrac12,\tfrac12\}$ but differ in the off-diagonal coherences $\tfrac12$ versus $0$ .

Step 2. Measure $\sigma_z$ (projectors $|0\rangle\langle0|,|1\rangle\langle1|$ ): for both states $\Pr(0)=\mathrm{Tr}(\rho|0\rangle\langle0|)=\tfrac12$ and $\Pr(1)=\tfrac12$ . $\sigma_z$ cannot distinguish them — it only sees the (identical) diagonal.

Step 3. Measure $\sigma_x$ (projectors $|+\rangle\langle+|,|-\rangle\langle-|$ ). For the pure state, $\Pr(+)=\mathrm{Tr}(|+\rangle\langle+|\,|+\rangle\langle+|)=\mathbf{1}$ , a certain outcome. For the mixture, $\Pr(+)=\mathrm{Tr}(\tfrac12 I\,|+\rangle\langle+|)=\tfrac12$ , a coin flip.

Step 4. Confirm with purity: $\mathrm{Tr}\big[(|+\rangle\langle+|)^2\big]=1$ (pure) versus $\mathrm{Tr}\big[(I/2)^2\big]=\tfrac12$ (maximally mixed). The coherences $\sigma_z$ ignored are exactly what make one state pure and deterministic under $\sigma_x$ . Same diagonal, opposite $\sigma_x$ statistics — coherence is physical.

3 · Reduced purity gauges partial entanglement

A two-qubit pure state $|\psi\rangle=\cos\theta\,|00\rangle+\sin\theta\,|11\rangle$ has $\cos^2\theta=0.85$ . Find $\rho_A$ , its purity and entropy, and interpret.

Step 1. Partial-trace over $B$ : since $\{|0\rangle,|1\rangle\}$ are orthogonal, cross terms drop and $\rho_A=\mathrm{diag}(0.85,0.15)$ — already diagonal, so these are the eigenvalues. Purity $\mathrm{Tr}(\rho_A^2)=0.85^2+0.15^2=\mathbf{0.745}$ , entropy $S=-(0.85\log_2 0.85+0.15\log_2 0.15)=\mathbf{0.610}$ bits.

Step 2. Interpret. Since $0.5<0.745<1$ the pair is entangled but not maximally. Dialling $\theta\to\pi/4$ sends $\rho_A\to I/2$ , purity $\to0.5$ , $S\to1$ bit (maximal); $\theta\to0$ sends $\rho_A\to|0\rangle\langle0|$ , purity $\to1$ , $S\to0$ (product, none). Reduced purity is a continuous entanglement gauge. $\rho_A=\mathrm{diag}(0.85,0.15)$ , purity $0.745$ , $S\approx0.61$ bit.

Physics of the portrait

The pure-versus-mixed distinction was engineered, deliberately, by one person who wanted quantum mechanics to have the rigor of measure theory. The density operator and its entropy are his, and they are what let us speak of "the state of a subsystem" at all.

Origins · 1 minJohn von Neumann and how this was discovered

John von Neumann

1903–1957 · Göttingen · Berlin · Princeton

John von Neumann arrived in Göttingen in the mid-1920s to find quantum mechanics split between Heisenberg's matrices and Schrödinger's waves and patched together by Dirac's brilliant but unrigorous formalism. In a burst of work from 1927 he rebuilt the theory on Hilbert space and self-adjoint operators, and in doing so introduced the object of this article: the statistical operator (density operator) $\rho$ , together with the entropy $S(\rho)=-\mathrm{Tr}(\rho\log\rho)$ that now bears his name. It was von Neumann who drew the sharp line between a pure state — a one-dimensional projector carrying maximal information — and a mixture, whose spectrum spreads across several eigenvalues, and who saw that the same $\rho$ governs both a classically uncertain preparation and the reduced description of an entangled subsystem. His 1932 treatise Mathematische Grundlagen der Quantenmechanik codified all of this, including the measurement postulates and the entropy as the quantum analogue of Gibbs' thermodynamic entropy. Lev Landau, independently in that same year 1927, introduced the density matrix specifically to describe a quantum subsystem, so the tool has two fathers — but the systematic pure/mixed framework, and the entropy that grades it, are von Neumann's. The whole modern edifice of quantum information is written in the language he built.

History

1927

— Landau introduces the density matrix to describe a quantum subsystem of a larger whole

1927

— von Neumann defines the statistical operator and the entropy S(ρ) = −Tr(ρ log ρ)

1932

— von Neumann's Mathematische Grundlagen codifies pure states as projectors versus general mixed density operators

1935

— Schrödinger names entanglement and shows a subsystem's ensemble can be resteered by a distant measurement

1993

— Hughston, Jozsa and Wootters pin down exactly which ensembles share a single reduced density operator

Going further

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

Provenance. A verifier should recompute: (1) the spectral proof $\rho^2=\rho\iff$ rank-one projector $\iff\mathrm{Tr}(\rho^2)=1$ (N&C §2.4); (2) the Bloch purity identity $\mathrm{Tr}(\rho^2)=\tfrac12(1+|\vec r|^2)$ from $(\vec r\cdot\vec\sigma)^2=|\vec r|^2 I$ (Preskill Ch. 2); (3) the Bell-state partial trace $\rho_A=I/2$ , reduced purity $1/2$ (N&C §2.4.3); (4) the ensemble equality $\tfrac12|0\rangle\langle0|+\tfrac12|1\rangle\langle1|=\tfrac12|+\rangle\langle+|+\tfrac12|-\rangle\langle-|=I/2$ and the HJW unitary freedom. All numbers — bounds $[1/d,1]$ , eigenvalues $(1\pm|\vec r|)/2$ , $S(0.8,0.2)=0.7219$ bit, $S(I/2)=1$ bit, partial-pair purity $0.745$ with $S=0.610$ bit — are standard or directly checkable.

Builds on. Density Operators · The Qubit and Its State Space · Dirac Bra-Ket Notation · Born Rule · Inner Product, Norm, and Normalization. Leads to. Expectation Values and Variance · Projective Measurements · Pauli Operators · Global and Relative Phase.

Ask the tutor: Hand me any density matrix and I will classify it pure or mixed, read off its Bloch vector, purity, and entropy, and tell you whether its mixedness is classical ignorance or hidden entanglement.

What exactly makes a state "mixed" rather than "pure"?

Purity is the litmus test: compute $\mathrm{Tr}(\rho^2)$ . If it equals $1$ the state is pure, $\rho=|\psi\rangle\langle\psi|$ for a single ket; if below $1$ it is mixed and no single ket reproduces it. Equivalently, a pure state has one eigenvalue $1$ and the rest zero (a rank-one projector, $\rho^2=\rho$ ), a mixed state at least two nonzero eigenvalues. Physically, a pure state is maximal knowledge — everything predictable packed into one vector — while a mixed state carries an irreducible spread, from classical ignorance or from entanglement with something you cannot see.

Is a superposition a mixture?

No, and conflating them is the classic mistake. The superposition $|+\rangle=\tfrac1{\sqrt2}(|0\rangle+|1\rangle)$ is a pure state with definite phase relations; it gives a certain outcome ( $+1$ ) under $\sigma_x$ , and its matrix has nonzero off-diagonal coherences. The mixture $\tfrac12|0\rangle\langle0|+\tfrac12|1\rangle\langle1|=I/2$ is a classical "either/or"; $\sigma_x$ gives a random result and its off-diagonal entries are zero. They share the diagonal $\{\tfrac12,\tfrac12\}$ , so $\sigma_z$ won't separate them — but a rotated-basis measurement will, because it probes the coherences only the superposition has.

If many ensembles give the same ρ, which one is "real"?

None is more real than any other — that is the point. The density operator $\rho$ is the complete physical state; the ensemble $\{p_i,|\psi_i\rangle\}$ is one story about how $\rho$ might have been prepared, and quantum mechanics gives no way to recover it. The maximally mixed qubit $I/2$ is equally a coin flip between $|0\rangle,|1\rangle$ and a coin flip between $|+\rangle,|-\rangle$ , and no measurement distinguishes them because every probability $\mathrm{Tr}(\rho P_m)$ depends on $\rho$ alone. The Hughston–Jozsa–Wootters theorem says exactly which ensembles collapse to a given $\rho$ : those related by a unitary on the purifying system.

How can a subsystem be mixed if the whole system is in a definite, pure state?

This is entanglement, the most quantum way to be mixed. If two qubits are in a Bell state $\tfrac1{\sqrt2}(|00\rangle+|11\rangle)$ , the pair is perfectly definite and pure — you know everything about the joint system. But ask only about one qubit, tracing out the other, and you get $\rho_A=I/2$ : maximally mixed. The information is not lost; it lives in the correlations between the qubits, which a description of one alone cannot capture. So local mixedness with global purity is the signature of entanglement, and the reduced entropy measures how entangled the pair is.

What is the difference between purity and von Neumann entropy?

Both grade mixedness, at different resolutions. The purity $\mathrm{Tr}(\rho^2)=\sum_k\lambda_k^2$ is a quadratic in the eigenvalues, cheap to compute (square and trace) and ranging from $1/d$ to $1$ . The von Neumann entropy $S=-\sum_k\lambda_k\log_2\lambda_k$ weights the spectrum logarithmically, ranges from $0$ to $\log_2 d$ , and counts the qubits of uncertainty — the quantity in coding theorems and entanglement measures. They agree at the extremes and order states alike, but the entropy is the fundamental one; the purity is its convenient, differentiable proxy.

Does decoherence turn a pure state into a mixed one?

Yes — that is essentially the definition of decoherence. A qubit coupled to an environment becomes entangled with it, and once you (necessarily) ignore the environment, the qubit's reduced state is mixed: its Bloch vector shrinks, purity drops below $1$ , coherences decay. This is the Bell-state partial trace played out continuously as the environment carries away which-path information. It is why real qubits have finite coherence times $T_2$ , why purity is the standard figure of merit for how "quantum" a device still is, and why error correction — the fight to keep the working state pure — is the central engineering problem of quantum computing.