Density matrix and pure / mixed states

Intuition Beginner

In unit 12.02.01 a quantum state was a single vector — a ket $|\psi ⟩$ in a Hilbert space. That picture works when you know the state exactly. But two everyday situations break it, and both are completely ordinary in a real laboratory.

The first is ignorance. Suppose a machine prepares a particle in state $|{\psi }_1⟩$ half the time and $|{\psi }_2⟩$ the other half, and it does not tell you which. You do not have one ket. You have a list of kets, each with a probability. A single vector cannot record "fifty-fifty between these two".

The second is more surprising. Take two particles prepared together so that their fates are linked. Even if the pair as a whole sits in one definite joint state, one particle on its own has no state vector of its own. Look at it alone and it behaves as if it were a probabilistic mixture.

The density matrix (also called the density operator) is the object that handles both cases with one piece of machinery. Write it $\hat{\rho }$. For a known pure state it is built from the ket as $\hat{\rho }=|\psi ⟩⟨\psi |$ — a way of packaging the same information. For a mixture it is a probability-weighted sum, $\hat{\rho }=p_1|{\psi }_1⟩⟨{\psi }_1|+p_2|{\psi }_2⟩⟨{\psi }_2|+\cdots$, where the numbers $p_i$ add to one.

Why is one object enough? Because everything you can ever measure is an average, and the density matrix is exactly the thing that produces averages. The recipe is short: to predict the average reading of any instrument $\hat{A}$, take the matrix product $\hat{\rho }\hat{A}$ and add up its diagonal. That sum-of-the-diagonal operation is called the trace, written $\mathrm{tr}$. So the one rule you need is the average equals $\mathrm{tr}(\hat{\rho }\hat{A})$.

There is a clean test that separates the two situations. A state is pure — a single honest ket — when $\hat{\rho }$ acts like a projector: applying it twice gives back the same thing, ${\hat{\rho }}^2=\hat{\rho }$. A state is mixed when applying it twice shrinks it, so the diagonal sum of ${\hat{\rho }}^2$ drops below one. The number $\mathrm{tr}({\hat{\rho }}^2)$ is called the purity, and it equals one for a genuine pure state and less than one for any real blend.

Think of it like a weather forecast. A pure state is "tomorrow is sunny, certainly". A mixed state is "seventy percent sunny, thirty percent rain". Both are legitimate descriptions of your knowledge, and you act on the forecast by computing expected outcomes. The density matrix is the forecast written in the language quantum mechanics actually uses.

Visual Beginner

For a single spin (a qubit) every density matrix is a point in a solid ball, the Bloch ball. The centre is the completely mixed state: every measurement comes out fifty-fifty. The surface is the set of pure states, the honest single kets. The length of the arrow from the centre measures how polarised the spin is; a half-length arrow is a partial mixture with purity between one half and one.

Worked example Beginner

Compare two spin-1/2 ensembles, both giving fifty-fifty for a measurement along the $z$-axis, but for opposite reasons.

Ensemble one: an unpolarised beam. Half the spins point up, half point down, and you do not know which is which. The density matrix is

$${\hat{\rho }}_{\text{mix}}=\frac{1}{2}|\uparrow ⟩⟨\uparrow |+\frac{1}{2}|\downarrow ⟩⟨\downarrow |=\left(\begin{array}{cc}1/2& 0\\ 0& 1/2\end{array}\right).$$

The purity is the diagonal sum of the square. Squaring gives $\left(\begin{array}{cc}1/4& 0\\ 0& 1/4\end{array}\right)$, and the diagonal sum is $1/4+1/4=1/2$. Purity $1/2$, so this is mixed.

Ensemble two: a polarised beam. Every spin points along $+x$, the single ket $|+{⟩}_x=\frac{1}{\sqrt{2}}(|\uparrow ⟩+|\downarrow ⟩)$. Its density matrix is

$${\hat{\rho }}_{\text{pure}}=|+{⟩}_x⟨+{|}_x=\left(\begin{array}{cc}1/2& 1/2\\ 1/2& 1/2\end{array}\right).$$

Squaring this matrix returns the same matrix (check the top-left entry: $1/2\cdot 1/2+1/2\cdot 1/2=1/2$). So the diagonal sum of the square is $1/2+1/2=1$. Purity $1$, so this is pure.

Both have the same diagonal — both read fifty-fifty along $z$. The off-diagonal entries tell them apart. The polarised beam has off-diagonal terms of $1/2$; measure it along $x$ and every spin reads $+x$ with certainty. The unpolarised beam has zero off-diagonal terms; measure it along any axis and you still get fifty-fifty.

What this tells us: the diagonal alone does not fix a quantum state. The off-diagonal entries, the coherences, carry the difference between a definite superposition and a genuine blend, and the purity number captures that difference in a single value.

Check your understanding Beginner

Formal definition Intermediate+

Let $\mathcal{H}$ be a Hilbert space. A density operator (density matrix) on $\mathcal{H}$ is a linear operator $\hat{\rho }:\mathcal{H}\to \mathcal{H}$ satisfying three conditions:

1. Hermitian: ${\hat{\rho }}^{†}=\hat{\rho }$.
2. Positive semidefinite: $⟨\varphi |\hat{\rho }|\varphi ⟩\ge 0$ for every $|\varphi ⟩\in \mathcal{H}$. Written $\hat{\rho }\ge 0$.
3. Unit trace: $\mathrm{tr}\hat{\rho }=1$.

The standard construction realises any such operator as a statistical ensemble. Given normalised states $|{\psi }_i⟩$ and probabilities $p_i\ge 0$ with ${\sum }_i{p}_i=1$, set

$$\hat{\rho }=\sum _i{p}_i|{\psi }_i⟩⟨{\psi }_i|.$$

This $\hat{\rho }$ satisfies all three conditions, and the spectral theorem (unit 12.02.02) shows every operator meeting the three conditions arises this way: its eigenvalues are the probabilities and its eigenvectors the ensemble members. The ensemble realising a given $\hat{\rho }$ is not unique; only the operator is physically meaningful, a point made precise in the Master tier.

A state is pure when $\hat{\rho }$ is a rank-one projector, $\hat{\rho }=|\psi ⟩⟨\psi |$ for some unit vector $|\psi ⟩$, equivalently ${\hat{\rho }}^2=\hat{\rho }$. Otherwise the state is mixed. The purity is the functional

$$\gamma (\hat{\rho }):=\mathrm{tr}({\hat{\rho }}^2),$$

bounded by $1/d\le \gamma (\hat{\rho })\le 1$ in dimension $d=\dim \mathcal{H}$, with $\gamma =1$ for pure states and $\gamma =1/d$ for the maximally mixed state $\hat{\rho }=1/d$.

Expectation values. For an observable $\hat{A}$ (a self-adjoint operator), the expected measurement value in state $\hat{\rho }$ is

$$⟨\hat{A}⟩=\mathrm{tr}(\hat{\rho }\hat{A}).$$

For a pure state this reduces to the familiar $⟨\psi |\hat{A}|\psi ⟩$, since $\mathrm{tr}(|\psi ⟩⟨\psi |\hat{A})=⟨\psi |\hat{A}|\psi ⟩$. The trace formula extends the Born rule from kets to ensembles without any new postulate: it is the probability-weighted average of the pure-state expectation values, ${\sum }_i{p}_i⟨{\psi }_i|\hat{A}|{\psi }_i⟩$.

The sign convention here is the physicist's: probabilities are eigenvalues of $\hat{\rho }$ and are non-negative, the trace is unit, and time evolution carries a factor $-i/\hslash$ (fixed in the Key derivation). A reader switching to a $+i/\hslash$ convention must flip the sign of every commutator in the dynamics.

Counterexamples to common slips

• Same diagonal, different state. The matrices $\left(\begin{array}{cc}1/2& 0\\ 0& 1/2\end{array}\right)$ and $\left(\begin{array}{cc}1/2& 1/2\\ 1/2& 1/2\end{array}\right)$ share a diagonal. The first is maximally mixed (purity $1/2$); the second is the pure state $|+{⟩}_x$ (purity $1$). The diagonal alone never determines purity.

• A mixture is not a superposition. The ensemble $\frac{1}{2}|\uparrow ⟩⟨\uparrow |+\frac{1}{2}|\downarrow ⟩⟨\downarrow |$ is not the pure superposition $\frac{1}{\sqrt{2}}(|\uparrow ⟩+|\downarrow ⟩)$. The first has no off-diagonal coherences; the second does. Writing a mixture as a single ket is the most common error in the subject.

• Different ensembles, one operator. The qubit operator $1/2$ is realised by the up/down ensemble and equally by the left/right ensemble $\frac{1}{2}|+{⟩}_x⟨+{|}_x+\frac{1}{2}|-{⟩}_x⟨-{|}_x$. No measurement separates them, because they are the same operator.

Key derivation Intermediate+

Theorem (Purity criterion). Let $\hat{\rho }$ be a density operator on a finite-dimensional Hilbert space $\mathcal{H}$. Then $\mathrm{tr}({\hat{\rho }}^2)\le 1$, with equality if and only if $\hat{\rho }$ is pure, i.e. $\hat{\rho }=|\psi ⟩⟨\psi |$ for some unit vector $|\psi ⟩$.

Proof. Since $\hat{\rho }$ is Hermitian and positive semidefinite with unit trace, the spectral theorem furnishes an orthonormal eigenbasis $\{|n⟩\}$ and eigenvalues ${\lambda }_n$ with

$$\hat{\rho }=\sum _n{\lambda }_n|n⟩⟨n|,{\lambda }_n\ge 0,\sum _n{\lambda }_n=1.$$

The eigenvalues form a probability distribution. From the spectral form, ${\hat{\rho }}^2={\sum }_n{\lambda }_n^2|n⟩⟨n|$, so

$$\mathrm{tr}({\hat{\rho }}^2)=\sum _n{\lambda }_n^2.$$

For any probability distribution, ${\sum }_n{\lambda }_n^2\le ({\sum }_n{\lambda }_n{)}^2=1$, because each ${\lambda }_n\in [0,1]$ forces ${\lambda }_n^2\le {\lambda }_n$, and summing gives ${\sum }_n{\lambda }_n^2\le {\sum }_n{\lambda }_n=1$. This establishes the bound.

Equality ${\sum }_n{\lambda }_n^2={\sum }_n{\lambda }_n$ requires ${\lambda }_n^2={\lambda }_n$ for every $n$, hence each ${\lambda }_n\in \{0,1\}$. With the constraint ${\sum }_n{\lambda }_n=1$, exactly one eigenvalue equals $1$ and the rest vanish. Then $\hat{\rho }=|n_0⟩⟨n_0|$ is a rank-one projector — a pure state. Conversely, a pure state has eigenvalues $(1,0,\dots ,0)$, giving $\mathrm{tr}({\hat{\rho }}^2)=1$. $\square$

Corollary (expectation reproduces the Born rule). Writing $\hat{\rho }={\sum }_i{p}_i|{\psi }_i⟩⟨{\psi }_i|$, the trace formula gives $⟨\hat{A}⟩=\mathrm{tr}(\hat{\rho }\hat{A})={\sum }_i{p}_i⟨{\psi }_i|\hat{A}|{\psi }_i⟩$, the classical average of quantum expectation values. The two layers of probability — classical ignorance ($p_i$) and quantum amplitude ($|{\psi }_i⟩$) — combine in one trace.

Bridge. This criterion is exactly the statement that the purity is a basis-independent fingerprint of how spread an ensemble is, and it builds toward the entropy that quantifies that spread in the Master tier. The foundational reason the trace formula needs no new postulate is that $\mathrm{tr}(\hat{\rho }\hat{A})$ generalises $⟨\psi |\hat{A}|\psi ⟩$ by linearity in $\hat{\rho }$: putting these together, the convex set of density operators has the pure states as its extreme points, and every mixed state is an average of pure ones. The bridge is the eigenvalue distribution $\{{\lambda }_n\}$, which appears again in the von Neumann entropy $-{\sum }_n{\lambda }_n\ln {\lambda }_n$, where the same numbers that measure purity measure information. The central insight is that classical probability and quantum amplitude are different axes of one object, and the density operator is where they meet.

Exercises Intermediate+

Lean formalization Intermediate+

A Lean statement of the purity criterion sits just outside current Mathlib. The three defining properties can be stated with Matrix.IsHermitian, Matrix.PosSemidef, and Matrix.trace, but the packaging and the spectral argument for the purity criterion are not yet assembled. The intended statement, with the proof gap recorded in Mathlib gap analysis, reads:

-- Intended statement; not yet in Mathlib (see lean_mathlib_gap).
-- A density matrix: Hermitian, positive semidefinite, unit trace.
structure DensityMatrix (n : ℕ) where
  ρ      : Matrix (Fin n) (Fin n) ℂ
  herm   : ρ.IsHermitian
  posdef : ρ.PosSemidef
  tr_one : ρ.trace = 1

-- Purity criterion: tr(ρ²) ≤ 1, equality iff ρ is a rank-one projector.
theorem purity_le_one {n : ℕ} (d : DensityMatrix n) :
    (d.ρ * d.ρ).trace.re ≤ 1 := by
  sorry  -- spectral theorem + ∑ λₙ² ≤ (∑ λₙ)² for a probability vector

The proof reduces to the eigenvalue inequality ${\sum }_n{\lambda }_n^2\le 1$ for a probability vector, which Mathlib can express but does not currently bundle with the density-matrix structure.

Advanced results Master

The reduced density matrix and the partial trace. Let the total system live on ${\mathcal{H}}_A\otimes {\mathcal{H}}_B$ with state ${\hat{\rho }}_{AB}$. The partial trace over $B$ is the linear map ${\mathrm{tr}}_B$ defined on product operators by ${\mathrm{tr}}_B(\hat{X}\otimes \hat{Y})=\hat{X}\mathrm{tr}(\hat{Y})$ and extended by linearity. The reduced density matrix ${\hat{\rho }}_A:={\mathrm{tr}}_B{\hat{\rho }}_{AB}$ is the unique operator on ${\mathcal{H}}_A$ reproducing all local expectation values: ${\mathrm{tr}}_A({\hat{\rho }}_A\hat{A})={\mathrm{tr}}_{AB}({\hat{\rho }}_{AB}(\hat{A}\otimes 1_B))$ for every observable $\hat{A}$ on $A$. This uniqueness is what makes ${\hat{\rho }}_A$ the honest state of subsystem $A$: it is the only object an experimenter confined to $A$ can ever access.

A pure global state can have a mixed reduced state, and this is the formal seat of entanglement. For the Bell state $|{\Phi }^{+}⟩=\frac{1}{\sqrt{2}}(|\uparrow \uparrow ⟩+|\downarrow \downarrow ⟩)$, the reduced state is

$${\hat{\rho }}_A={\mathrm{tr}}_B|{\Phi }^{+}⟩⟨{\Phi }^{+}|=\frac{1}{2}|\uparrow ⟩⟨\uparrow |+\frac{1}{2}|\downarrow ⟩⟨\downarrow |=\frac{1}{2}1,$$

maximally mixed despite the pair being in a definite pure state. Knowing the whole with certainty and knowing each part only statistically coexist without contradiction.

The von Neumann (Liouville) equation. If the closed system evolves by the Schrödinger equation $i\hslash {\partial }_t|\psi ⟩=\hat{H}|\psi ⟩$, then $\hat{\rho }={\sum }_i{p}_i|{\psi }_i⟩⟨{\psi }_i|$ evolves by

$$i\hslash \frac{d\hat{\rho }}{dt}=[\hat{H},\hat{\rho }].$$

The sign is opposite to the Heisenberg equation of motion for observables, the quantum echo of the classical Liouville equation ${\partial }_t\rho =\{H,\rho {\}}_{\text{PB}}$ with the Poisson bracket replaced by $[\cdot ,\cdot ]/i\hslash$. The solution is conjugation by the unitary propagator, $\hat{\rho }(t)=\hat{U}(t)\hat{\rho }(0)\hat{U}(t{)}^{†}$ with $\hat{U}(t)=e^{-i\hat{H}t/\hslash }$.

Von Neumann entropy. The information content of a state is

$$S(\hat{\rho }):=-\mathrm{tr}(\hat{\rho }\ln \hat{\rho })=-\sum _n{\lambda }_n\ln {\lambda }_n,$$

the Shannon entropy of the eigenvalue distribution. It vanishes for pure states and reaches $\ln d$ for the maximally mixed state. Under unitary evolution $S(\hat{U}\hat{\rho }{\hat{U}}^{†})=S(\hat{\rho })$, because conjugation permutes nothing in the spectrum — closed-system entropy is conserved, and entropy increase is a statement about coarse-graining or tracing out, not about the microscopic dynamics.

Purification. Every mixed state is the shadow of a pure one. Given ${\hat{\rho }}_A={\sum }_k{p}_k|k⟩⟨k|$ on ${\mathcal{H}}_A$, introduce an auxiliary space ${\mathcal{H}}_R$ of dimension at least $\mathrm{rank}({\hat{\rho }}_A)$ and define $|\Psi ⟩={\sum }_k\sqrt{p_k}|k{⟩}_A\otimes |k{⟩}_R$. Then ${\mathrm{tr}}_R|\Psi ⟩⟨\Psi |={\hat{\rho }}_A$. Any two purifications of the same ${\hat{\rho }}_A$ differ only by a unitary on the auxiliary system.

The qubit Bloch ball. Every qubit density matrix is $\hat{\rho }=\frac{1}{2}(1+\mathbf{a}\cdot \mathbf{\sigma })$ with $\mathbf{a}=⟨\mathbf{\sigma }⟩\in {\mathbb{R}}^3$ and $|\mathbf{a}|\le 1$. Purity is $\frac{1}{2}(1+|\mathbf{a}{|}^2)$, so pure states sit on the sphere $|\mathbf{a}|=1$ and the centre $\mathbf{a}=0$ is maximally mixed. Unitary evolution rotates $\mathbf{a}$ rigidly; decoherence shrinks it toward the centre.

Synthesis. Putting these together, the density operator is the single object on which statistics, dynamics, and information all act compatibly. The foundational reason a subsystem of a pure whole looks mixed is that the partial trace discards the correlations, and this is exactly the content of the Bell-pair computation: the reduced state $\frac{1}{2}1$ is dual to the maximal entanglement of the global state. The von Neumann equation generalises Schrödinger evolution from kets to operators, and its unitary solution is precisely what keeps the entropy constant — the central insight that microscopic reversibility and the second law live at different levels of description. Purification identifies every mixed state with a pure state on a larger space, so the bridge between the open-system picture and the closed-system picture is a partial trace in one direction and an enlargement in the other. The Bloch-ball geometry makes the convex structure visible: pure states on the boundary, mixed states inside, and the same eigenvalue distribution $\{{\lambda }_n\}$ that fixes purity also fixes entropy, the recurring thread that this pattern recurs throughout quantum information.

Full proof set Master

Proposition 1 (Partial trace is well-defined and trace-preserving). The map ${\mathrm{tr}}_B:\mathcal{B}({\mathcal{H}}_A\otimes {\mathcal{H}}_B)\to \mathcal{B}({\mathcal{H}}_A)$ defined on a fixed orthonormal basis $\{|j{⟩}_B\}$ by ${\mathrm{tr}}_B(\hat{M})={\sum }_j(1_A\otimes ⟨j{|}_B)\hat{M}(1_A\otimes |j{⟩}_B)$ is independent of the basis and satisfies ${\mathrm{tr}}_A({\mathrm{tr}}_B\hat{M})={\mathrm{tr}}_{AB}\hat{M}$.

Proof. Let $\{|j{⟩}_B\}$ and $\{|j^{\prime }{⟩}_B\}$ be two orthonormal bases related by a unitary $U$, $|j^{\prime }⟩={\sum }_j{U}_{j{j}^{\prime }}|j⟩$. Substituting,

$$\sum _{j^{\prime }}(1\otimes ⟨j^{\prime }|)\hat{M}(1\otimes |j^{\prime }⟩)=\sum _{j^{\prime },j,k}\overline{U_{j{j}^{\prime }}}{U}_{k{j}^{\prime }}(1\otimes ⟨j|)\hat{M}(1\otimes |k⟩).$$

The unitarity relation ${\sum }_{j^{\prime }}\overline{U_{j{j}^{\prime }}}{U}_{k{j}^{\prime }}={\delta }_{jk}$ collapses the sum to ${\sum }_j(1\otimes ⟨j|)\hat{M}(1\otimes |j⟩)$, the original expression. For trace preservation, take a basis $\{|i{⟩}_A\}$ of ${\mathcal{H}}_A$; then ${\mathrm{tr}}_A({\mathrm{tr}}_B\hat{M})={\sum }_{i,j}⟨i|⟨j|\hat{M}|i⟩|j⟩={\mathrm{tr}}_{AB}\hat{M}$, since $\{|i⟩\otimes |j⟩\}$ is an orthonormal basis of the tensor product. $\square$

Proposition 2 (Reduced state of a maximally entangled pair is maximally mixed). For $|{\Phi }^{+}⟩=\frac{1}{\sqrt{2}}(|\uparrow \uparrow ⟩+|\downarrow \downarrow ⟩)$, the reduced state is ${\hat{\rho }}_A=\frac{1}{2}1$.

Proof. Form $|{\Phi }^{+}⟩⟨{\Phi }^{+}|=\frac{1}{2}(|\uparrow \uparrow ⟩+|\downarrow \downarrow ⟩)(⟨\uparrow \uparrow |+⟨\downarrow \downarrow |)$, expanding into four terms. Apply ${\mathrm{tr}}_B={\sum }_{j\in \{\uparrow ,\downarrow \}}(1_A\otimes ⟨j{|}_B)\cdot (1_A\otimes |j{⟩}_B)$. The cross terms $|\uparrow \uparrow ⟩⟨\downarrow \downarrow |$ and its conjugate contribute $⟨j|\uparrow ⟩⟨\downarrow |j⟩=0$ for both $j$, so they drop. The surviving diagonal terms give

$${\hat{\rho }}_A=\frac{1}{2}(|\uparrow ⟩⟨\uparrow |⟨\uparrow |\uparrow ⟩+|\downarrow ⟩⟨\downarrow |⟨\downarrow |\downarrow ⟩)=\frac{1}{2}(|\uparrow ⟩⟨\uparrow |+|\downarrow ⟩⟨\downarrow |)=\frac{1}{2}1.$$

Maximal entanglement of the pair becomes maximal ignorance of the part. $\square$

Proposition 3 (Unitary invariance of von Neumann entropy). For any unitary $\hat{U}$, $S(\hat{U}\hat{\rho }{\hat{U}}^{†})=S(\hat{\rho })$.

Proof. The operator $\hat{U}\hat{\rho }{\hat{U}}^{†}$ has the same eigenvalues as $\hat{\rho }$: if $\hat{\rho }|n⟩={\lambda }_n|n⟩$, then $\hat{U}\hat{\rho }{\hat{U}}^{†}(\hat{U}|n⟩)=\hat{U}\hat{\rho }|n⟩={\lambda }_n(\hat{U}|n⟩)$, so $\hat{U}|n⟩$ is an eigenvector with the same eigenvalue ${\lambda }_n$. Since $S(\hat{\rho })=-{\sum }_n{\lambda }_n\ln {\lambda }_n$ depends only on the multiset $\{{\lambda }_n\}$, and that multiset is preserved, the entropy is unchanged. As a corollary, von Neumann evolution $\hat{\rho }(t)=\hat{U}(t)\hat{\rho }(0)\hat{U}(t{)}^{†}$ conserves entropy for all $t$. $\square$

Proposition 4 (Solution of the von Neumann equation). The unique solution of $i\hslash \dot{\hat{\rho }}=[\hat{H},\hat{\rho }]$ with time-independent $\hat{H}$ and initial value $\hat{\rho }(0)$ is $\hat{\rho }(t)=\hat{U}(t)\hat{\rho }(0)\hat{U}(t{)}^{†}$, $\hat{U}(t)=e^{-i\hat{H}t/\hslash }$.

Proof. Differentiate the proposed solution: $\dot{\hat{\rho }}=\dot{\hat{U}}\hat{\rho }(0){\hat{U}}^{†}+\hat{U}\hat{\rho }(0){\dot{\hat{U}}}^{†}$. With $\dot{\hat{U}}=-\frac{i}{\hslash }\hat{H}\hat{U}$ and ${\dot{\hat{U}}}^{†}=\frac{i}{\hslash }{\hat{U}}^{†}\hat{H}$,

$$\dot{\hat{\rho }}=-\frac{i}{\hslash }\hat{H}\hat{U}\hat{\rho }(0){\hat{U}}^{†}+\frac{i}{\hslash }\hat{U}\hat{\rho }(0){\hat{U}}^{†}\hat{H}=-\frac{i}{\hslash }(\hat{H}\hat{\rho }-\hat{\rho }\hat{H})=-\frac{i}{\hslash }[\hat{H},\hat{\rho }],$$

which rearranges to $i\hslash \dot{\hat{\rho }}=[\hat{H},\hat{\rho }]$. Uniqueness follows because the equation is a linear first-order ODE in the finite-dimensional space of operators with a Lipschitz right-hand side, so the Picard-Lindelöf theorem applies. $\square$

Connections Master

The reduced density matrix and partial trace introduced here are the exact apparatus on which the Schmidt decomposition and entanglement entropy of unit 12.17.02 are built; the eigenvalues $\{p_k\}$ of ${\hat{\rho }}_A$ are the squared Schmidt coefficients, and the entanglement entropy is the von Neumann entropy of that reduced state.

The quantum-information treatment of the density matrix in unit 12.17.01 specialises this foundational formalism to quantum channels, the data-processing inequality, and the operational tasks of state discrimination; that unit takes the purity criterion and the trace formula proved here as given and pushes toward Kraus maps and accessible information.

The von Neumann equation derived here is the statistical-ensemble companion of the Schrödinger and Heisenberg pictures in unit 12.03.01: the opposite commutator sign relative to the Heisenberg equation of motion is precisely the distinction between evolving states and evolving observables, and the trace formula $⟨\hat{A}⟩=\mathrm{tr}(\hat{\rho }\hat{A})$ is invariant under moving the time dependence between the two.

The spectral theorem of unit 12.02.02 supplies the eigenbasis that turns every abstract density operator into a probability distribution over orthonormal states, which is the single step that makes the purity criterion and the entropy functional computable; without Hermiticity and positivity the trace formulas would not yield real, bounded expectation values.

The Bloch-ball parametrisation connects forward to the angular-momentum and spin formalism of unit 12.05.01, where the vector $\mathbf{a}=⟨\mathbf{\sigma }⟩$ is the expectation of the spin operator and unitary evolution acts as a rigid $\mathrm{SO}(3)$ rotation generated by the $\mathfrak{su}(2)$ algebra.

Historical & philosophical context Master

The density matrix has a genuinely dual origin in 1927. Lev Landau, then nineteen, introduced it in his first published paper on radiation damping, where the description of a quantum subsystem interacting with a field forced an object beyond the wavefunction ^{[Landau 1927]}. Independently and in the same year, John von Neumann constructed the same operator from the thermodynamic side, seeking a quantum analogue of the Gibbs ensemble and a definition of entropy for quantum states ^{[von Neumann 1927]}. Von Neumann's 1932 monograph gave the formalism its now-standard axiomatic shape, fixing the three defining properties and the trace rule for expectation values ^{[von Neumann 1932]}.

The operator-technique synthesis that physicists use today, including the multipole and Bloch-vector expansions, was systematised by Ugo Fano in 1957, whose review consolidated the density-matrix and statistical-tensor methods scattered across atomic and nuclear physics ^{[Fano 1957]}. Landau and Lifshitz present the formalism compactly in §14 of their quantum mechanics volume, deferring the fuller entropy treatment to their statistical-physics volume ^{[Landau-Lifshitz 1977]}, a division that mirrors the historical split between Landau's dynamical motivation and von Neumann's thermodynamic one.

The distinction between a proper mixture (classical ignorance about which pure state was prepared) and an improper mixture (the reduced state of an entangled whole) became a fault line in the philosophy of measurement. Both are described by identical density operators, and no local measurement separates them, yet they encode different physical situations. This indistinguishability is the lever decoherence theory uses, and it is the reason the reduced density matrix sits at the centre of the modern measurement-problem literature.

Bibliography Master

@article{landau1927daempfung,
  author  = {Landau, Lev D.},
  title   = {Das D{\"a}mpfungsproblem in der Wellenmechanik},
  journal = {Zeitschrift f{\"u}r Physik},
  volume  = {45},
  pages   = {430--441},
  year    = {1927}
}

@article{vonneumann1927thermo,
  author  = {von Neumann, John},
  title   = {Thermodynamik quantenmechanischer Gesamtheiten},
  journal = {Nachrichten von der Gesellschaft der Wissenschaften zu G{\"o}ttingen, Mathematisch-Physikalische Klasse},
  pages   = {273--291},
  year    = {1927}
}

@book{vonneumann1932grundlagen,
  author    = {von Neumann, John},
  title     = {Mathematische Grundlagen der Quantenmechanik},
  publisher = {Springer},
  year      = {1932},
  note      = {English translation: Mathematical Foundations of Quantum Mechanics, Princeton University Press, 1955, Ch. IV}
}

@article{fano1957density,
  author  = {Fano, Ugo},
  title   = {Description of States in Quantum Mechanics by Density Matrix and Operator Techniques},
  journal = {Reviews of Modern Physics},
  volume  = {29},
  pages   = {74--93},
  year    = {1957}
}

@book{landau1977quantum,
  author    = {Landau, Lev D. and Lifshitz, Evgeny M.},
  title     = {Quantum Mechanics: Non-Relativistic Theory},
  edition   = {3rd},
  publisher = {Pergamon Press},
  year      = {1977},
  note      = {Course of Theoretical Physics, Vol. 3, \S 14}
}

@book{nielsen2010quantum,
  author    = {Nielsen, Michael A. and Chuang, Isaac L.},
  title     = {Quantum Computation and Quantum Information},
  edition   = {10th Anniversary},
  publisher = {Cambridge University Press},
  year      = {2010},
  note      = {\S 2.4, density operators and partial trace}
}