Chapter 23: The Density Matrix and Mixed States: Quantum Mechanics Beyond Pure States

"The density matrix is the most powerful and general description of a quantum state. If you only know kets, you only know half the story." — Adapted from a common sentiment in quantum information theory

"Quantum mechanics has not only been unable to guarantee us a good government — it has so far failed to give us a pure state of the universe." — John von Neumann, paraphrased

Up to this point, every quantum system we have studied has been described by a state vector $|\psi\rangle$ — a ket in Hilbert space. This is sufficient when we know exactly which state the system occupies, and when the system is perfectly isolated from its environment. But physics is rarely so clean. In the laboratory, we routinely face situations where we don't know the exact state, where our system is entangled with degrees of freedom we cannot access, or where thermal fluctuations render a pure-state description inadequate. The density operator formalism — invented by John von Neumann in 1927, the same miraculous year that gave us the Hilbert-space formulation of quantum mechanics — provides the general framework that handles all of these situations.

This chapter introduces you to the most general description of a quantum state: the density operator $\hat{\rho}$. You will learn to distinguish pure states from mixed states, compute expectation values and entropies, perform partial traces to describe subsystems, and understand decoherence — the process by which quantum superpositions dissolve into classical-looking mixtures through environmental interaction.

🏃 Fast Track: If you are comfortable with outer products $|\psi\rangle\langle\psi|$ and the trace operation, skip to Section 23.4 for the key diagnostic ($\text{Tr}(\hat{\rho}^2)$ test), then proceed through Section 23.7 (partial trace) and 23.8 (decoherence), which contain the material most essential for Chapters 24–27.

23.1 Why Pure States Aren't Enough

The Limits of the Ket

Consider a spin-1/2 particle. In the formalism we have developed so far, we describe its state as a ket:

$$|\psi\rangle = \alpha|{+z}\rangle + \beta|{-z}\rangle, \quad |\alpha|^2 + |\beta|^2 = 1$$

This is a pure state — a complete, maximally precise specification of the quantum system. Every property of the system can be computed from $|\psi\rangle$.

But now imagine the following three scenarios, each of which defeats the ket formalism:

Scenario 1: Classical ignorance. You have a box containing spin-1/2 particles. A technician prepared each particle in either $|{+z}\rangle$ or $|{-z}\rangle$, each with probability $1/2$, but she did not tell you which. For any given particle, you know it is in one of these states, but you do not know which. This is classical ignorance — the kind of uncertainty that arises from a coin flip, not from quantum superposition. No single ket describes your knowledge.

⚠️ Common Misconception: It is tempting to describe the classically ignorant mixture as the superposition $\frac{1}{\sqrt{2}}(|{+z}\rangle + |{-z}\rangle) = |{+x}\rangle$. This is wrong. The superposition $|{+x}\rangle$ is a definite, pure quantum state — it has a definite spin along the $x$-axis. The classical mixture of $|{+z}\rangle$ and $|{-z}\rangle$ is something fundamentally different: it has no definite spin along any axis. Confusing these two will lead you astray throughout quantum mechanics.

Scenario 2: Thermal equilibrium. A quantum harmonic oscillator at temperature $T$ occupies the energy eigenstates $|n\rangle$ with Boltzmann probabilities $p_n = e^{-n\hbar\omega/k_BT}/Z$, where $Z$ is the partition function. The system is not in any single energy eigenstate — it is in a thermal mixture. No ket captures this.

Scenario 3: Entangled subsystems. Two spin-1/2 particles are in the singlet state:

$$|\Psi^-\rangle = \frac{1}{\sqrt{2}}\bigl(|{+z}\rangle_A|{-z}\rangle_B - |{-z}\rangle_A|{+z}\rangle_B\bigr)$$

The composite system is in a perfectly well-defined pure state. But what about particle $A$ alone? If you only have access to particle $A$, what state describes your subsystem? It is not $|{+z}\rangle_A$, nor $|{-z}\rangle_A$, nor any superposition. Particle $A$ by itself is in a mixed state, even though the composite system is pure. This is perhaps the deepest reason we need the density matrix: entanglement with inaccessible degrees of freedom forces subsystem descriptions beyond the ket formalism.

💡 Key Insight: These three scenarios — classical ignorance, thermal equilibrium, and entangled subsystems — share a common mathematical description: the density operator. The formalism does not distinguish why the state is mixed, only that it is mixed. This unification is both its power and its limitation.

What We Need

We need a mathematical object that can describe: 1. Pure states (complete knowledge) as a special case 2. Statistical mixtures (partial knowledge) 3. Subsystems of entangled composite systems

That object is the density operator $\hat{\rho}$.

Proper vs. Improper Mixtures

Before we proceed, it is worth highlighting a conceptual distinction that is often glossed over. Physicists recognize two fundamentally different reasons a state might be mixed:

Proper mixtures arise from classical ignorance — Scenario 1 above. Someone knows the exact state; you just don't. In principle, the information exists somewhere. A proper mixture is like a shuffled deck of cards: each card has a definite identity, you just can't see it.

Improper mixtures arise from entanglement — Scenario 3 above. No one, anywhere in the universe, knows the state of subsystem $A$ alone, because the information does not exist locally. The subsystem genuinely has no pure state. An improper mixture is like asking for the color of a single thread in a tangled rope — the question misidentifies what is fundamental.

The density matrix formalism treats both cases identically: the same $2 \times 2$ matrix $\hat{I}/2$ describes both the coin-flip mixture of spin-1/2 states (proper) and the reduced state of a singlet-entangled particle (improper). This operational equivalence is sometimes called the preparation-independent nature of the density operator. No experiment on subsystem $A$ alone can distinguish a proper mixture from an improper one — a fact with deep implications for the foundations of quantum mechanics.

🧪 Thought Experiment: Imagine Alice prepares a spin-1/2 particle in either $|{+z}\rangle$ or $|{-z}\rangle$ by flipping a coin, and hands it to Bob without telling him the result. Bob's density matrix is $\hat{I}/2$. Now imagine instead that Alice prepares a singlet pair, keeps one particle, and gives the other to Bob. Bob's density matrix is also $\hat{I}/2$. Can Bob tell which scenario he is in? No — not by any measurement on his particle alone. He needs to compare results with Alice (classical communication + joint statistics) to discover the entanglement. This is one reason entanglement is so subtle.

23.2 The Density Operator $\hat{\rho}$

Definition: Pure State

For a system in a known pure state $|\psi\rangle$, the density operator is the outer product (projector):

$$\hat{\rho} = |\psi\rangle\langle\psi|$$

This is not just a formal construction — it contains exactly the same information as $|\psi\rangle$, minus an overall phase. Recall that two kets $|\psi\rangle$ and $e^{i\phi}|\psi\rangle$ describe the same physical state. The density operator $|\psi\rangle\langle\psi| = e^{i\phi}|\psi\rangle \, e^{-i\phi}\langle\psi| = |\psi\rangle\langle\psi|$ automatically eliminates this unobservable phase. This is already an advantage.

Example: For a spin-1/2 particle in state $|{+z}\rangle$:

$$\hat{\rho} = |{+z}\rangle\langle{+z}| = \begin{pmatrix} 1 \\ 0 \end{pmatrix} \begin{pmatrix} 1 & 0 \end{pmatrix} = \begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix}$$

For a particle in $|{+x}\rangle = \frac{1}{\sqrt{2}}(|{+z}\rangle + |{-z}\rangle)$:

$$\hat{\rho} = |{+x}\rangle\langle{+x}| = \frac{1}{2}\begin{pmatrix} 1 & 1 \\ 1 & 1 \end{pmatrix}$$

Notice the off-diagonal elements. These encode the quantum coherence — the fact that $|{+x}\rangle$ is a superposition of $|{+z}\rangle$ and $|{-z}\rangle$, not merely a mixture. The off-diagonal terms are the signatures of genuine quantum interference. Remember this — it will be crucial when we discuss decoherence.

Definition: Mixed State (Statistical Ensemble)

When the system is in state $|\psi_k\rangle$ with classical probability $p_k$, the density operator is:

$$\hat{\rho} = \sum_k p_k \, |\psi_k\rangle\langle\psi_k|$$

where: - $p_k \geq 0$ for all $k$ (probabilities are non-negative) - $\sum_k p_k = 1$ (probabilities sum to one) - The $|\psi_k\rangle$ need not be orthogonal (this is important!)

Example: The classical mixture of $|{+z}\rangle$ and $|{-z}\rangle$ with equal probability:

$$\hat{\rho}_{\text{mix}} = \frac{1}{2}|{+z}\rangle\langle{+z}| + \frac{1}{2}|{-z}\rangle\langle{-z}| = \frac{1}{2}\begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix} + \frac{1}{2}\begin{pmatrix} 0 & 0 \\ 0 & 1 \end{pmatrix} = \frac{1}{2}\begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix} = \frac{1}{2}\hat{I}$$

Compare this to the pure superposition state $|{+x}\rangle$:

$$\hat{\rho}_{+x} = \frac{1}{2}\begin{pmatrix} 1 & 1 \\ 1 & 1 \end{pmatrix}$$

Same diagonal elements. Completely different off-diagonal elements. Same probabilities for $z$-measurements. Different probabilities for $x$-measurements. The mixed state $\hat{\rho}_{\text{mix}}$ gives 50/50 for measurements along any axis. The pure state $|{+x}\rangle$ gives 100% spin-up along $x$. The off-diagonal elements — the coherences — make all the difference.

🔵 Historical Note: The density matrix was introduced by John von Neumann in 1927 in a paper titled "Wahrscheinlichkeitstheoretischer Aufbau der Quantenmechanik" ("Probability-Theoretic Construction of Quantum Mechanics"). Lev Landau independently introduced the concept in the same year. Von Neumann immediately recognized its importance for statistical mechanics and the measurement problem, devoting an entire chapter of his 1932 masterwork Mathematische Grundlagen der Quantenmechanik to it.

The Density Matrix in a Basis

In any orthonormal basis $\{|n\rangle\}$, the density operator has matrix elements:

$$\rho_{mn} = \langle m|\hat{\rho}|n\rangle$$

For a pure state $|\psi\rangle = \sum_n c_n |n\rangle$:

$$\rho_{mn} = c_m c_n^*$$

The diagonal elements $\rho_{nn} = |c_n|^2$ are the probabilities of finding the system in state $|n\rangle$. They are sometimes called populations.

The off-diagonal elements $\rho_{mn}$ ($m \neq n$) encode quantum coherence between $|m\rangle$ and $|n\rangle$. They are called coherences. In a mixed state, these coherences are partially or fully suppressed.

💡 Key Insight: The density matrix is the most natural representation of a quantum state for computation. Every prediction quantum mechanics can make — every expectation value, every probability, every measurement outcome distribution — can be extracted from $\hat{\rho}$ alone. The ket $|\psi\rangle$ is a special case, not the general framework.

Non-Uniqueness of Ensemble Decompositions

A crucial subtlety: a given density matrix generally admits many different ensemble decompositions. Consider the maximally mixed qubit $\hat{\rho} = \hat{I}/2$. It can be written as:

$$\hat{\rho} = \frac{1}{2}|{+z}\rangle\langle{+z}| + \frac{1}{2}|{-z}\rangle\langle{-z}|$$

or equally well as:

$$\hat{\rho} = \frac{1}{2}|{+x}\rangle\langle{+x}| + \frac{1}{2}|{-x}\rangle\langle{-x}|$$

or as:

$$\hat{\rho} = \frac{1}{2}|{+n}\rangle\langle{+n}| + \frac{1}{2}|{-n}\rangle\langle{-n}|$$

for any direction $\hat{n}$ on the Bloch sphere. All these ensembles produce exactly the same density matrix and therefore exactly the same predictions for all possible measurements.

This non-uniqueness is not a bug — it is a feature. It reflects the fundamental fact that quantum mechanics cannot distinguish between different preparation procedures that produce the same density operator. The density matrix captures everything that is physically relevant about the state; the particular ensemble used to prepare it is not. This is profoundly different from classical probability, where the identity of the underlying outcomes typically matters.

🔗 Connection: This non-uniqueness is closely related to the purification theorem (Chapter 24): every mixed state $\hat{\rho}_A$ can be obtained as the reduced density matrix of infinitely many different pure entangled states $|\Psi\rangle_{AB}$, related by unitary transformations on the ancilla $B$. The freedom to choose the purification corresponds exactly to the freedom to choose the ensemble decomposition.

Worked Example: Building a Density Matrix from Scratch

Problem: A Stern-Gerlach apparatus prepares spin-1/2 particles as follows. With probability $1/3$, the spin points along $+z$. With probability $1/3$, it points along $+x$. With probability $1/3$, it points along $-y$. Write the density matrix.

Solution: We need the kets in the $\{|{+z}\rangle, |{-z}\rangle\}$ basis:

$$|{+z}\rangle = \begin{pmatrix} 1 \\ 0 \end{pmatrix}, \quad |{+x}\rangle = \frac{1}{\sqrt{2}}\begin{pmatrix} 1 \\ 1 \end{pmatrix}, \quad |{-y}\rangle = \frac{1}{\sqrt{2}}\begin{pmatrix} 1 \\ i \end{pmatrix}$$

The density matrices for each pure state are:

$$|{+z}\rangle\langle{+z}| = \begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix}, \quad |{+x}\rangle\langle{+x}| = \frac{1}{2}\begin{pmatrix} 1 & 1 \\ 1 & 1 \end{pmatrix}, \quad |{-y}\rangle\langle{-y}| = \frac{1}{2}\begin{pmatrix} 1 & -i \\ i & 1 \end{pmatrix}$$

Summing with equal weights $p = 1/3$:

$$\hat{\rho} = \frac{1}{3}\begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix} + \frac{1}{6}\begin{pmatrix} 1 & 1 \\ 1 & 1 \end{pmatrix} + \frac{1}{6}\begin{pmatrix} 1 & -i \\ i & 1 \end{pmatrix} = \frac{1}{6}\begin{pmatrix} 4 & 1-i \\ 1+i & 2 \end{pmatrix}$$

Let us verify: $\text{Tr}(\hat{\rho}) = (4 + 2)/6 = 1$ ✓. The matrix is Hermitian because $\rho_{12} = (1 - i)/6$ and $\rho_{21} = (1 + i)/6 = \rho_{12}^*$ ✓. Eigenvalues can be checked to be positive. This is a valid mixed state.

The Bloch vector is:

$$r_x = \text{Tr}(\hat{\rho}\hat{\sigma}_x) = \frac{2 \text{Re}(\rho_{12})}{1} = \frac{2 \cdot 1/6}{1} = 1/3$$ $$r_y = \text{Tr}(\hat{\rho}\hat{\sigma}_y) = \frac{-2 \text{Im}(\rho_{12})}{1} = \frac{-2 \cdot (-1/6)}{1} = 1/3$$ $$r_z = \rho_{11} - \rho_{22} = 4/6 - 2/6 = 1/3$$

So $\vec{r} = (1/3, 1/3, 1/3)$ and $|\vec{r}| = 1/\sqrt{3} \approx 0.577 < 1$. This is a mixed state, as expected.

23.3 Properties of $\hat{\rho}$: Trace 1, Positive, Hermitian

The density operator is not an arbitrary operator. It satisfies three fundamental properties that constrain its form and guarantee physical consistency.

Property 1: Hermiticity

$$\hat{\rho}^\dagger = \hat{\rho}$$

Proof: For a pure state, $\hat{\rho}^\dagger = (|\psi\rangle\langle\psi|)^\dagger = |\psi\rangle\langle\psi| = \hat{\rho}$. For a mixed state:

$$\hat{\rho}^\dagger = \left(\sum_k p_k |\psi_k\rangle\langle\psi_k|\right)^\dagger = \sum_k p_k^* |\psi_k\rangle\langle\psi_k| = \sum_k p_k |\psi_k\rangle\langle\psi_k| = \hat{\rho}$$

since the $p_k$ are real. $\square$

Consequence: Hermiticity guarantees that $\hat{\rho}$ has real eigenvalues and can be diagonalized by a unitary transformation. This is essential — the eigenvalues of $\hat{\rho}$ will turn out to be probabilities, and probabilities must be real.

Property 2: Unit Trace

$$\text{Tr}(\hat{\rho}) = 1$$

Proof: For a mixed state in any orthonormal basis $\{|n\rangle\}$:

$$\text{Tr}(\hat{\rho}) = \sum_n \langle n|\hat{\rho}|n\rangle = \sum_n \sum_k p_k |\langle n|\psi_k\rangle|^2 = \sum_k p_k \underbrace{\sum_n |\langle n|\psi_k\rangle|^2}_{= 1 \text{ (completeness)}} = \sum_k p_k = 1 \quad \square$$

Physical meaning: The probabilities of all possible outcomes of any measurement must sum to 1. The unit-trace condition encodes normalization.

Property 3: Positive Semi-Definiteness

For any state $|\phi\rangle$:

$$\langle\phi|\hat{\rho}|\phi\rangle \geq 0$$

Proof:

$$\langle\phi|\hat{\rho}|\phi\rangle = \sum_k p_k |\langle\phi|\psi_k\rangle|^2 \geq 0$$

since each term is a product of a non-negative probability $p_k$ and a non-negative squared modulus. $\square$

Consequence: All eigenvalues of $\hat{\rho}$ are non-negative. Combined with unit trace, the eigenvalues satisfy:

$$\lambda_i \geq 0, \quad \sum_i \lambda_i = 1$$

The eigenvalues of $\hat{\rho}$ are a probability distribution. We can always write $\hat{\rho}$ in its spectral decomposition:

$$\hat{\rho} = \sum_i \lambda_i |i\rangle\langle i|$$

where $\{|i\rangle\}$ are the orthonormal eigenvectors of $\hat{\rho}$. This is the unique decomposition into an orthogonal ensemble. (Non-orthogonal decompositions also exist, and are not unique — a single density operator can represent infinitely many different preparation procedures.)

Summary of Properties

✅ Checkpoint: Verify these properties for $\hat{\rho} = \frac{1}{3}|{+z}\rangle\langle{+z}| + \frac{2}{3}|{+x}\rangle\langle{+x}|$. Write out the $2\times 2$ matrix in the $\{|{+z}\rangle, |{-z}\rangle\}$ basis. Confirm it is Hermitian, has trace 1, and has non-negative eigenvalues. (Hint: the answer is $\hat{\rho} = \frac{1}{6}\begin{pmatrix} 5 & 2 \\ 2 & 1 \end{pmatrix}$, which you should verify. Then check: are the eigenvalues positive?)

The von Neumann Equation: Time Evolution

The time evolution of the density operator follows from the Schrodinger equation. Since each $|\psi_k(t)\rangle$ evolves under $\hat{H}$:

$$i\hbar \frac{\partial \hat{\rho}}{\partial t} = [\hat{H}, \hat{\rho}]$$

This is the von Neumann equation, the density-operator analogue of the Schrodinger equation. For a time-independent Hamiltonian, the solution is:

$$\hat{\rho}(t) = e^{-i\hat{H}t/\hbar}\, \hat{\rho}(0)\, e^{i\hat{H}t/\hbar} = \hat{U}(t)\,\hat{\rho}(0)\,\hat{U}^\dagger(t)$$

where $\hat{U}(t) = e^{-i\hat{H}t/\hbar}$ is the unitary time-evolution operator (Chapter 7).

🔗 Connection: Compare the von Neumann equation $i\hbar\dot{\hat{\rho}} = [\hat{H}, \hat{\rho}]$ with the Heisenberg equation of motion for operators $i\hbar\dot{\hat{A}} = [\hat{A}, \hat{H}]$ (Chapter 6). Notice the sign difference: in the Schrodinger picture, $\hat{\rho}$ evolves with a commutator $[\hat{H}, \hat{\rho}]$, while in the Heisenberg picture, observables evolve with $[\hat{A}, \hat{H}]$.

Derivation of the von Neumann equation. Let $\hat{\rho}(t) = \sum_k p_k |\psi_k(t)\rangle\langle\psi_k(t)|$. Taking the time derivative:

$$\frac{\partial \hat{\rho}}{\partial t} = \sum_k p_k \left(\frac{\partial |\psi_k\rangle}{\partial t}\langle\psi_k| + |\psi_k\rangle\frac{\partial \langle\psi_k|}{\partial t}\right)$$

From the Schrodinger equation, $i\hbar\frac{\partial |\psi_k\rangle}{\partial t} = \hat{H}|\psi_k\rangle$, and its adjoint, $-i\hbar\frac{\partial \langle\psi_k|}{\partial t} = \langle\psi_k|\hat{H}$. Substituting:

$$i\hbar\frac{\partial \hat{\rho}}{\partial t} = \sum_k p_k \left(\hat{H}|\psi_k\rangle\langle\psi_k| - |\psi_k\rangle\langle\psi_k|\hat{H}\right) = \hat{H}\hat{\rho} - \hat{\rho}\hat{H} = [\hat{H}, \hat{\rho}] \quad \square$$

Two important consequences follow immediately:

Conservation of trace: $\frac{d}{dt}\text{Tr}(\hat{\rho}) = \text{Tr}(\dot{\hat{\rho}}) = \frac{1}{i\hbar}\text{Tr}([\hat{H}, \hat{\rho}]) = 0$ (the trace of any commutator vanishes). Normalization is preserved.

Conservation of purity: $\frac{d}{dt}\text{Tr}(\hat{\rho}^2) = 2\text{Tr}(\hat{\rho}\dot{\hat{\rho}}) = \frac{2}{i\hbar}\text{Tr}(\hat{\rho}[\hat{H}, \hat{\rho}]) = \frac{2}{i\hbar}\text{Tr}(\hat{\rho}\hat{H}\hat{\rho} - \hat{\rho}^2\hat{H}) = 0$ (by the cyclic property of the trace). This means unitary evolution cannot change a pure state into a mixed state or vice versa. If you start pure, you stay pure. If you start mixed, you stay exactly as mixed as you were.

⚠️ Common Misconception: Students sometimes think that the Schrodinger equation allows entropy to increase during time evolution. It does not. The von Neumann entropy $S = -\text{Tr}(\hat{\rho}\ln\hat{\rho})$ is exactly conserved under unitary evolution, because the eigenvalues of $\hat{\rho}$ do not change. Entropy can only increase when the system interacts with an environment and we trace over the environmental degrees of freedom — this is precisely what decoherence does (Section 23.8).

Stationary States in the Density Matrix Formalism

If $\hat{\rho}$ commutes with $\hat{H}$, i.e., $[\hat{H}, \hat{\rho}] = 0$, then $\dot{\hat{\rho}} = 0$ and the density matrix is stationary. This happens whenever $\hat{\rho}$ is diagonal in the energy eigenbasis:

$$\hat{\rho} = \sum_n p_n |E_n\rangle\langle E_n|$$

The thermal density matrix $\hat{\rho}_{\text{th}} = e^{-\beta\hat{H}}/Z$ is the most important example. It commutes with $\hat{H}$ by construction, confirming that thermal equilibrium is a stationary state — as it must be.

More generally, any density matrix that is a function of $\hat{H}$ alone (i.e., $\hat{\rho} = f(\hat{H})$ for some function $f$) is automatically stationary. This is the quantum analogue of the Jeans theorem in classical statistical mechanics.

23.4 Pure vs. Mixed: The $\text{Tr}(\hat{\rho}^2)$ Test

Given a density matrix, how do you tell whether it represents a pure state or a mixed state? The answer is the purity:

$$\gamma = \text{Tr}(\hat{\rho}^2)$$

The Purity Criterion

• Pure state: $\hat{\rho}^2 = \hat{\rho}$ (the density operator is a projection operator), so $\text{Tr}(\hat{\rho}^2) = \text{Tr}(\hat{\rho}) = 1$.
• Mixed state: $\text{Tr}(\hat{\rho}^2) < 1$.
• Maximally mixed state (complete ignorance in a $d$-dimensional Hilbert space): $\hat{\rho} = \hat{I}/d$, giving $\text{Tr}(\hat{\rho}^2) = 1/d$.

Proof that $\hat{\rho}^2 = \hat{\rho}$ for a pure state:

$$\hat{\rho}^2 = (|\psi\rangle\langle\psi|)(|\psi\rangle\langle\psi|) = |\psi\rangle\underbrace{\langle\psi|\psi\rangle}_{= 1}\langle\psi| = |\psi\rangle\langle\psi| = \hat{\rho}$$

Proof that $\text{Tr}(\hat{\rho}^2) < 1$ for a mixed state:

In the eigenbasis, $\hat{\rho} = \sum_i \lambda_i |i\rangle\langle i|$, so $\hat{\rho}^2 = \sum_i \lambda_i^2 |i\rangle\langle i|$ and:

$$\text{Tr}(\hat{\rho}^2) = \sum_i \lambda_i^2$$

By the Cauchy-Schwarz inequality (or simply by the fact that $\lambda_i^2 \leq \lambda_i$ when $0 \leq \lambda_i \leq 1$, with equality only when $\lambda_i = 0$ or $1$):

$$\sum_i \lambda_i^2 \leq \left(\sum_i \lambda_i\right)^2 = 1$$

with equality if and only if exactly one eigenvalue is 1 and all others are 0 — i.e., a pure state. $\square$

The Purity Spectrum

The purity $\gamma = \text{Tr}(\hat{\rho}^2)$ ranges over:

$$\frac{1}{d} \leq \gamma \leq 1$$

For a qubit ($d = 2$): purity ranges from $1/2$ (maximally mixed, $\hat{\rho} = \hat{I}/2$) to $1$ (pure state, on the Bloch sphere surface).

Bloch Sphere Interpretation

Recall from Chapter 13 that any qubit pure state lives on the surface of the Bloch sphere. The density matrix formalism extends this beautifully. Any $2\times 2$ density matrix can be written:

$$\hat{\rho} = \frac{1}{2}(\hat{I} + \vec{r} \cdot \hat{\vec{\sigma}}) = \frac{1}{2}\begin{pmatrix} 1 + r_z & r_x - ir_y \\ r_x + ir_y & 1 - r_z \end{pmatrix}$$

where $\hat{\vec{\sigma}} = (\hat{\sigma}_x, \hat{\sigma}_y, \hat{\sigma}_z)$ are the Pauli matrices and $\vec{r} = (r_x, r_y, r_z)$ is the Bloch vector.

The purity is:

$$\text{Tr}(\hat{\rho}^2) = \frac{1}{2}(1 + |\vec{r}|^2)$$

So: - $|\vec{r}| = 1$: pure state (on the surface of the Bloch sphere) - $|\vec{r}| < 1$: mixed state (inside the Bloch sphere) - $|\vec{r}| = 0$: maximally mixed state (at the center)

The Bloch ball (interior included) is the space of all qubit states. This is the natural extension of the Bloch sphere (surface only) that you learned in Chapter 13.

💡 Key Insight: The Bloch ball picture gives a beautiful geometric intuition for decoherence: decoherence takes a pure state (on the surface) and drags it toward the center (maximally mixed). The off-diagonal coherences shrink, the Bloch vector shortens, and the state becomes progressively more "classical."

Worked Example: Testing Purity

Problem: Determine whether each of the following density matrices represents a pure or mixed state.

(a) $\hat{\rho}_1 = \begin{pmatrix} 3/4 & \sqrt{3}/4 \\ \sqrt{3}/4 & 1/4 \end{pmatrix}$

(b) $\hat{\rho}_2 = \begin{pmatrix} 3/4 & 0 \\ 0 & 1/4 \end{pmatrix}$

Solution (a):

$$\hat{\rho}_1^2 = \begin{pmatrix} 3/4 & \sqrt{3}/4 \\ \sqrt{3}/4 & 1/4 \end{pmatrix}\begin{pmatrix} 3/4 & \sqrt{3}/4 \\ \sqrt{3}/4 & 1/4 \end{pmatrix} = \begin{pmatrix} 9/16 + 3/16 & 3\sqrt{3}/16 + \sqrt{3}/16 \\ 3\sqrt{3}/16 + \sqrt{3}/16 & 3/16 + 1/16 \end{pmatrix} = \begin{pmatrix} 3/4 & \sqrt{3}/4 \\ \sqrt{3}/4 & 1/4 \end{pmatrix}$$

$\text{Tr}(\hat{\rho}_1^2) = 3/4 + 1/4 = 1$. Pure state. (In fact, this is $|\psi\rangle = \frac{\sqrt{3}}{2}|{+z}\rangle + \frac{1}{2}|{-z}\rangle$, the spin state pointing at $\theta = \pi/3$ on the Bloch sphere.)

Solution (b):

$$\hat{\rho}_2^2 = \begin{pmatrix} 9/16 & 0 \\ 0 & 1/16 \end{pmatrix}, \quad \text{Tr}(\hat{\rho}_2^2) = 9/16 + 1/16 = 10/16 = 5/8 < 1$$

Mixed state. The purity is $5/8$, between the pure value ($1$) and the maximally mixed value ($1/2$).

✅ Checkpoint: For the density matrix $\hat{\rho} = \frac{1}{4}|{+z}\rangle\langle{+z}| + \frac{3}{4}|{+x}\rangle\langle{+x}|$, compute the matrix, the purity, and the Bloch vector. Verify that $|\vec{r}| < 1$.

23.5 Expectation Values: $\langle\hat{A}\rangle = \text{Tr}(\hat{\rho}\hat{A})$

The Trace Formula

The density operator formalism gives a powerful, basis-independent formula for expectation values. For an observable $\hat{A}$:

$$\langle\hat{A}\rangle = \text{Tr}(\hat{\rho}\hat{A})$$

Proof for a mixed state: If the system is in $|\psi_k\rangle$ with probability $p_k$, the expectation value of $\hat{A}$ is the classical average of the quantum expectation values:

$$\langle\hat{A}\rangle = \sum_k p_k \langle\psi_k|\hat{A}|\psi_k\rangle$$

Now, $\langle\psi_k|\hat{A}|\psi_k\rangle = \text{Tr}(|\psi_k\rangle\langle\psi_k|\hat{A})$ (which you can verify by inserting a complete set of states). Therefore:

$$\langle\hat{A}\rangle = \sum_k p_k \,\text{Tr}(|\psi_k\rangle\langle\psi_k|\hat{A}) = \text{Tr}\!\left(\sum_k p_k |\psi_k\rangle\langle\psi_k|\hat{A}\right) = \text{Tr}(\hat{\rho}\hat{A}) \quad \square$$

This formula is remarkably compact. It also works for pure states, since a pure state is just the special case $\hat{\rho} = |\psi\rangle\langle\psi|$.

Measurement Probabilities

The probability of obtaining eigenvalue $a_n$ when measuring $\hat{A}$ is:

$$P(a_n) = \text{Tr}(\hat{\rho}\hat{P}_n)$$

where $\hat{P}_n = |a_n\rangle\langle a_n|$ is the projection operator onto the eigenstate $|a_n\rangle$. For degenerate eigenvalues, $\hat{P}_n$ projects onto the full degenerate subspace.

Worked Example: Spin Measurements on a Mixed State

Problem: A spin-1/2 particle is in the mixed state $\hat{\rho} = \frac{3}{4}|{+z}\rangle\langle{+z}| + \frac{1}{4}|{-z}\rangle\langle{-z}|$. Compute $\langle\hat{S}_z\rangle$, $\langle\hat{S}_x\rangle$, and the probabilities of measuring $\pm\hbar/2$ along $z$ and $x$.

Solution: Write $\hat{\rho} = \begin{pmatrix} 3/4 & 0 \\ 0 & 1/4 \end{pmatrix}$ in the $\{|{+z}\rangle, |{-z}\rangle\}$ basis.

$z$-measurements:

$$\langle\hat{S}_z\rangle = \text{Tr}\!\left(\hat{\rho}\frac{\hbar}{2}\hat{\sigma}_z\right) = \frac{\hbar}{2}\text{Tr}\!\left(\begin{pmatrix} 3/4 & 0 \\ 0 & 1/4 \end{pmatrix}\begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}\right) = \frac{\hbar}{2}\text{Tr}\begin{pmatrix} 3/4 & 0 \\ 0 & -1/4 \end{pmatrix} = \frac{\hbar}{2}\cdot\frac{1}{2} = \frac{\hbar}{4}$$

$P(+\hbar/2, z) = \rho_{11} = 3/4$. $P(-\hbar/2, z) = \rho_{22} = 1/4$.

$x$-measurements: The eigenstates of $\hat{S}_x$ are $|{\pm x}\rangle = \frac{1}{\sqrt{2}}(|{+z}\rangle \pm |{-z}\rangle)$.

$$\langle\hat{S}_x\rangle = \text{Tr}\!\left(\hat{\rho}\frac{\hbar}{2}\hat{\sigma}_x\right) = \frac{\hbar}{2}\text{Tr}\!\left(\begin{pmatrix} 3/4 & 0 \\ 0 & 1/4 \end{pmatrix}\begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}\right) = \frac{\hbar}{2}\text{Tr}\begin{pmatrix} 0 & 3/4 \\ 0 & 1/4 \end{pmatrix}$$

Wait — let us be more careful:

$$\hat{\rho}\hat{\sigma}_x = \begin{pmatrix} 3/4 & 0 \\ 0 & 1/4 \end{pmatrix}\begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix} = \begin{pmatrix} 0 & 3/4 \\ 1/4 & 0 \end{pmatrix}$$

$$\text{Tr}\begin{pmatrix} 0 & 3/4 \\ 1/4 & 0 \end{pmatrix} = 0 + 0 = 0$$

So $\langle\hat{S}_x\rangle = 0$. The mixed state has no preferred direction in the $xy$-plane — all the coherence is gone, and the $x$-expectation value vanishes.

$P(+\hbar/2, x) = \langle{+x}|\hat{\rho}|{+x}\rangle = \frac{1}{2}(3/4 + 1/4) = 1/2$. Similarly $P(-\hbar/2, x) = 1/2$.

⚠️ Common Misconception: A mixed state $\hat{\rho}$ with $\langle\hat{S}_x\rangle = 0$ does not mean the spin is definitely zero along $x$. It means the average is zero — individual measurements still yield $\pm\hbar/2$, each with 50% probability. Do not confuse zero expectation value with a zero eigenvalue.

State Update After Measurement

The density matrix formalism also provides a clean description of how the state changes after a measurement. If we measure an observable $\hat{A}$ and obtain eigenvalue $a_n$ (with projector $\hat{P}_n = |a_n\rangle\langle a_n|$), the post-measurement state is:

$$\hat{\rho}' = \frac{\hat{P}_n \hat{\rho} \hat{P}_n}{\text{Tr}(\hat{P}_n \hat{\rho} \hat{P}_n)}$$

The numerator projects the state onto the eigenspace of $a_n$; the denominator re-normalizes. This is the Luders rule (generalization of the von Neumann projection postulate to density matrices).

If we perform the measurement but do not record the outcome, the state becomes:

$$\hat{\rho}' = \sum_n \hat{P}_n \hat{\rho} \hat{P}_n$$

This non-selective measurement has an important property: it eliminates all coherences between different eigenspaces of $\hat{A}$ while preserving the diagonal elements (populations). In other words, a non-selective measurement in the $\hat{A}$ eigenbasis produces decoherence in that basis — a connection we will exploit in Section 23.8.

Variance and Fluctuations

The variance of observable $\hat{A}$ in a mixed state is:

$$(\Delta A)^2 = \langle\hat{A}^2\rangle - \langle\hat{A}\rangle^2 = \text{Tr}(\hat{\rho}\hat{A}^2) - [\text{Tr}(\hat{\rho}\hat{A})]^2$$

For the mixed state $\hat{\rho} = \frac{3}{4}|{+z}\rangle\langle{+z}| + \frac{1}{4}|{-z}\rangle\langle{-z}|$, the variance of $\hat{S}_z$ is:

$$(\Delta S_z)^2 = \frac{\hbar^2}{4}\text{Tr}(\hat{\rho}) - \left(\frac{\hbar}{4}\right)^2 = \frac{\hbar^2}{4} - \frac{\hbar^2}{16} = \frac{3\hbar^2}{16}$$

Note that this variance contains contributions from two sources: quantum uncertainty (inherent in each pure-state component) and classical uncertainty (from not knowing which component the system is in). For a pure state, only the quantum contribution exists. For a mixed state, both contribute, and they cannot be separated by measurements on the system alone.

23.6 Von Neumann Entropy: $S = -\text{Tr}(\hat{\rho} \ln \hat{\rho})$

Definition

The von Neumann entropy quantifies the amount of uncertainty — or missing information — in a quantum state:

$$S(\hat{\rho}) = -\text{Tr}(\hat{\rho} \ln \hat{\rho})$$

In the eigenbasis where $\hat{\rho} = \sum_i \lambda_i |i\rangle\langle i|$, this becomes:

$$S = -\sum_i \lambda_i \ln \lambda_i$$

with the convention that $0 \ln 0 = 0$ (justified by $\lim_{x\to 0^+} x \ln x = 0$).

🔗 Connection: Compare with the Shannon entropy from classical information theory: $H = -\sum_i p_i \ln p_i$. The von Neumann entropy is the quantum generalization of the Shannon entropy. When we use $\log_2$ instead of $\ln$, entropy is measured in bits (or "qubits" in the quantum case). The relationship is $S_{\text{bits}} = S / \ln 2$.

Properties of von Neumann Entropy

1. Non-negativity: $S(\hat{\rho}) \geq 0$, with equality if and only if $\hat{\rho}$ is a pure state.

Proof: For a pure state, $\hat{\rho} = |i\rangle\langle i|$, so one eigenvalue is 1 and all others are 0. Then $S = -1\ln 1 - 0 = 0$. For any mixed state, at least two eigenvalues are strictly between 0 and 1, and $-\lambda \ln \lambda > 0$ for $0 < \lambda < 1$.

2. Maximum value: $S \leq \ln d$, where $d$ is the dimension of the Hilbert space, with equality for the maximally mixed state $\hat{\rho} = \hat{I}/d$.

Proof: By the method of Lagrange multipliers, the function $-\sum_i \lambda_i \ln \lambda_i$ subject to $\sum_i \lambda_i = 1$ is maximized when all $\lambda_i = 1/d$, giving $S = \ln d$.

3. Concavity: $S\!\left(\sum_k p_k \hat{\rho}_k\right) \geq \sum_k p_k S(\hat{\rho}_k)$

Mixing states increases entropy. This is physically intuitive: mixing introduces additional classical uncertainty.

4. Invariance under unitaries: $S(\hat{U}\hat{\rho}\hat{U}^\dagger) = S(\hat{\rho})$

Unitary evolution preserves the eigenvalues of $\hat{\rho}$ and therefore preserves the entropy. A closed quantum system does not gain or lose information under Schrodinger evolution.

Key Values

Worked Example: Entropy of a Thermal Qubit

Problem: A two-level atom with energy gap $\Delta E$ is in thermal equilibrium at temperature $T$. Find the von Neumann entropy as a function of temperature.

Solution: The thermal density matrix is:

$$\hat{\rho} = \frac{1}{Z}\begin{pmatrix} e^{0} & 0 \\ 0 & e^{-\beta\Delta E} \end{pmatrix} = \begin{pmatrix} p_0 & 0 \\ 0 & p_1 \end{pmatrix}$$

where $\beta = 1/(k_BT)$, $Z = 1 + e^{-\beta\Delta E}$, $p_0 = 1/Z$, and $p_1 = e^{-\beta\Delta E}/Z$.

The von Neumann entropy is:

$$S = -p_0 \ln p_0 - p_1 \ln p_1$$

Let us define $x = \beta\Delta E = \Delta E/(k_BT)$.

$$p_0 = \frac{1}{1 + e^{-x}}, \quad p_1 = \frac{e^{-x}}{1 + e^{-x}}$$

$$S(x) = \ln(1 + e^{-x}) + \frac{x\, e^{-x}}{1 + e^{-x}}$$

Limiting cases: - $T \to 0$ ($x \to \infty$): $p_0 \to 1$, $p_1 \to 0$. $S \to 0$. The system freezes into the ground state (pure). - $T \to \infty$ ($x \to 0$): $p_0, p_1 \to 1/2$. $S \to \ln 2$. The system is maximally mixed.

📊 By the Numbers: For a qubit with $\Delta E = 1$ meV (typical of superconducting quantum circuits): at $T = 50$ mK, $x \approx 0.23$, giving $S \approx 0.56$ nats $\approx 0.81$ bits. At $T = 10$ mK, $x \approx 1.16$, giving $S \approx 0.26$ nats $\approx 0.37$ bits. Quantum computing requires $k_BT \ll \Delta E$ so that the qubit is nearly in a pure ground state.

Subadditivity and the Triangle Inequality

The von Neumann entropy satisfies a remarkable inequality for composite systems. For a bipartite system $AB$:

Subadditivity:

$$S(\hat{\rho}_{AB}) \leq S(\hat{\rho}_A) + S(\hat{\rho}_B)$$

with equality if and only if $\hat{\rho}_{AB} = \hat{\rho}_A \otimes \hat{\rho}_B$ (the subsystems are uncorrelated).

Araki-Lieb triangle inequality:

$$S(\hat{\rho}_{AB}) \geq |S(\hat{\rho}_A) - S(\hat{\rho}_B)|$$

These two inequalities together constrain the entropy of a composite system:

$$|S(\hat{\rho}_A) - S(\hat{\rho}_B)| \leq S(\hat{\rho}_{AB}) \leq S(\hat{\rho}_A) + S(\hat{\rho}_B)$$

The lower bound is saturated for pure composite states (where $S(\hat{\rho}_{AB}) = 0$ and $S(\hat{\rho}_A) = S(\hat{\rho}_B)$). The upper bound is saturated for product states. Everything in between represents varying degrees of classical and quantum correlation.

💡 Key Insight: The subadditivity inequality has a beautiful information-theoretic interpretation. It says that the total uncertainty about a joint system is at most the sum of the uncertainties about its parts. If the parts are correlated (classically or quantumly), knowing about the joint system reduces your total uncertainty. The amount of reduction defines the quantum mutual information $I(A:B) = S(\hat{\rho}_A) + S(\hat{\rho}_B) - S(\hat{\rho}_{AB}) \geq 0$, which quantifies all correlations — both classical and quantum — between $A$ and $B$.

Entropy, Information, and the Second Law

The connection between von Neumann entropy and thermodynamic entropy is deep and precise. For a system in thermal equilibrium at temperature $T$:

$$S_{\text{thermo}} = k_B \, S_{\text{vN}}$$

where $S_{\text{thermo}}$ is the thermodynamic entropy in J/K and $S_{\text{vN}}$ is the von Neumann entropy (dimensionless, in nats). The Boltzmann constant $k_B$ simply converts between natural units and SI units.

The second law of thermodynamics states that the total entropy of an isolated system never decreases. In the density matrix formalism, this becomes: the von Neumann entropy of the total system (system + environment) is conserved under unitary evolution, but the entropy of the subsystem (obtained by partial trace) can increase. This is not a contradiction — it is precisely what happens during decoherence. The missing information does not disappear; it is delocalized into correlations between the system and its environment.

23.7 Reduced Density Matrices and the Partial Trace

The Problem of Subsystems

Suppose we have a composite system of two subsystems $A$ and $B$ (for instance, two spin-1/2 particles, or a system and its environment). The composite state lives in the tensor product space $\mathcal{H} = \mathcal{H}_A \otimes \mathcal{H}_B$ (Chapter 11).

If the composite system is in state $\hat{\rho}_{AB}$, what density matrix describes subsystem $A$ alone?

The answer is the reduced density matrix, obtained by performing a partial trace over subsystem $B$:

$$\hat{\rho}_A = \text{Tr}_B(\hat{\rho}_{AB})$$

Defining the Partial Trace

Let $\{|j\rangle_B\}$ be any orthonormal basis for $\mathcal{H}_B$. The partial trace over $B$ is:

$$\hat{\rho}_A = \text{Tr}_B(\hat{\rho}_{AB}) = \sum_j {}_B\langle j|\hat{\rho}_{AB}|j\rangle_B$$

This sum "traces out" the $B$ degrees of freedom, leaving an operator that acts only on $\mathcal{H}_A$. The result is independent of which basis $\{|j\rangle_B\}$ we choose — exactly as the ordinary trace is basis-independent.

🔴 Warning: The partial trace is not the same as "setting the B part to a specific state." It is a sum over all B states. This distinction is crucial and often confused by beginners.

Worked Example: Partial Trace of an Entangled State

Problem: Two spin-1/2 particles are in the singlet state:

$$|\Psi^-\rangle = \frac{1}{\sqrt{2}}\bigl(|{+z}\rangle_A|{-z}\rangle_B - |{-z}\rangle_A|{+z}\rangle_B\bigr)$$

Compute the reduced density matrix $\hat{\rho}_A$.

Solution: First, write the density matrix of the composite system:

$$\hat{\rho}_{AB} = |\Psi^-\rangle\langle\Psi^-| = \frac{1}{2}\bigl(|{+z}\rangle_A|{-z}\rangle_B - |{-z}\rangle_A|{+z}\rangle_B\bigr)\bigl({}_A\langle{+z}|{}_B\langle{-z}| - {}_A\langle{-z}|{}_B\langle{+z}|\bigr)$$

Expanding the four terms:

$$\hat{\rho}_{AB} = \frac{1}{2}\bigl(|{+z}\rangle_A\langle{+z}| \otimes |{-z}\rangle_B\langle{-z}| \;-\; |{+z}\rangle_A\langle{-z}| \otimes |{-z}\rangle_B\langle{+z}|$$ $$\qquad\qquad -\; |{-z}\rangle_A\langle{+z}| \otimes |{+z}\rangle_B\langle{-z}| \;+\; |{-z}\rangle_A\langle{-z}| \otimes |{+z}\rangle_B\langle{+z}|\bigr)$$

Now take the partial trace over $B$. We sum ${}_B\langle j|\hat{\rho}_{AB}|j\rangle_B$ for $j = {+z}, {-z}$:

$j = {+z}$:

$${}_B\langle{+z}|\hat{\rho}_{AB}|{+z}\rangle_B = \frac{1}{2}\bigl(|{+z}\rangle_A\langle{+z}| \cdot \underbrace{{}_B\langle{+z}|{-z}\rangle_B}_{0}\underbrace{{}_B\langle{-z}|{+z}\rangle_B}_{0} - \ldots + |{-z}\rangle_A\langle{-z}| \cdot \underbrace{{}_B\langle{+z}|{+z}\rangle_B}_{1}\underbrace{{}_B\langle{+z}|{+z}\rangle_B}_{1}\bigr)$$

Working through each term carefully:

$${}_B\langle{+z}|\hat{\rho}_{AB}|{+z}\rangle_B = \frac{1}{2}|{-z}\rangle_A\langle{-z}|$$

$j = {-z}$:

$${}_B\langle{-z}|\hat{\rho}_{AB}|{-z}\rangle_B = \frac{1}{2}|{+z}\rangle_A\langle{+z}|$$

Summing:

$$\hat{\rho}_A = \frac{1}{2}|{+z}\rangle_A\langle{+z}| + \frac{1}{2}|{-z}\rangle_A\langle{-z}| = \frac{1}{2}\hat{I}_A$$

Result: The reduced density matrix of particle $A$ is the maximally mixed state $\hat{I}/2$. Even though the composite system is in a perfectly well-defined pure state, each subsystem individually is maximally uncertain. This is the hallmark of maximal entanglement.

💡 Key Insight: The von Neumann entropy of the reduced density matrix quantifies entanglement. For a pure composite state $|\Psi\rangle_{AB}$, the entanglement entropy is $S(\hat{\rho}_A) = S(\hat{\rho}_B) = -\text{Tr}(\hat{\rho}_A \ln \hat{\rho}_A)$. For the singlet: $S = \ln 2$ (maximum for two qubits). For a product state $|\psi\rangle_A \otimes |\phi\rangle_B$: $S = 0$ (no entanglement). The entanglement entropy is the single most important measure of quantum correlations for pure bipartite states.

Partial Trace: General Procedure

For a composite density matrix written in a product basis $\{|m\rangle_A \otimes |j\rangle_B\}$:

$$\hat{\rho}_{AB} = \sum_{m,n,j,k} \rho_{mj,nk}\; |m\rangle_A\langle n| \otimes |j\rangle_B\langle k|$$

The partial trace over $B$ gives:

$$(\hat{\rho}_A)_{mn} = \sum_j \rho_{mj,nj}$$

In words: sum over the $B$ indices when they match on both sides. This is exactly the matrix operation of "tracing over the second subsystem."

Similarly, the partial trace over $A$:

$$(\hat{\rho}_B)_{jk} = \sum_m \rho_{mj,mk}$$

Example: Partial Trace of a Non-Maximally Entangled State

Consider the state $|\psi\rangle = \sqrt{2/3}\;|0\rangle_A|0\rangle_B + \sqrt{1/3}\;|1\rangle_A|1\rangle_B$. The composite density matrix is:

$$\hat{\rho}_{AB} = \frac{2}{3}|00\rangle\langle 00| + \frac{\sqrt{2}}{3}|00\rangle\langle 11| + \frac{\sqrt{2}}{3}|11\rangle\langle 00| + \frac{1}{3}|11\rangle\langle 11|$$

The partial trace over $B$:

$$\hat{\rho}_A = \frac{2}{3}|0\rangle\langle 0| + \frac{1}{3}|1\rangle\langle 1| = \begin{pmatrix} 2/3 & 0 \\ 0 & 1/3 \end{pmatrix}$$

The entanglement entropy is:

$$S(\hat{\rho}_A) = -\frac{2}{3}\ln\frac{2}{3} - \frac{1}{3}\ln\frac{1}{3} \approx 0.637 \text{ nats}$$

which is less than $\ln 2 \approx 0.693$ nats (the maximum for two qubits), reflecting the fact that this state is entangled but not maximally so.

✅ Checkpoint: Verify that the partial trace of a product state $|\psi\rangle_A|\phi\rangle_B$ gives $\hat{\rho}_A = |\psi\rangle_A\langle\psi|$ — a pure state. The partial trace "does nothing interesting" for product states, confirming that separable states have no entanglement.

Worked Example: Partial Trace of a Three-Qubit State

Problem: Three qubits are in the GHZ state $|\text{GHZ}\rangle = \frac{1}{\sqrt{2}}(|000\rangle + |111\rangle)$. Compute the reduced density matrix of qubit $A$ (the first qubit).

Solution: We need to trace over qubits $B$ and $C$ (the second and third qubits, forming a 4-dimensional subsystem). The density matrix of the composite state is:

$$\hat{\rho}_{ABC} = \frac{1}{2}(|000\rangle + |111\rangle)(\langle 000| + \langle 111|)$$

$$= \frac{1}{2}(|000\rangle\langle 000| + |000\rangle\langle 111| + |111\rangle\langle 000| + |111\rangle\langle 111|)$$

Taking the partial trace over $BC$ (dimension 4, basis $\{|00\rangle, |01\rangle, |10\rangle, |11\rangle\}$):

For $|j\rangle_{BC} = |00\rangle$: ${}_{BC}\langle 00|\hat{\rho}_{ABC}|00\rangle_{BC} = \frac{1}{2}|0\rangle_A\langle 0|$

For $|j\rangle_{BC} = |11\rangle$: ${}_{BC}\langle 11|\hat{\rho}_{ABC}|11\rangle_{BC} = \frac{1}{2}|1\rangle_A\langle 1|$

The $|01\rangle$ and $|10\rangle$ terms contribute zero. The cross terms $|000\rangle\langle 111|$ and $|111\rangle\langle 000|$ vanish because $\langle 00|11\rangle = 0$ and $\langle 11|00\rangle = 0$. Therefore:

$$\hat{\rho}_A = \frac{1}{2}|0\rangle\langle 0| + \frac{1}{2}|1\rangle\langle 1| = \frac{1}{2}\hat{I}$$

Each individual qubit of a GHZ state is maximally mixed — just like the singlet state. But there is an important difference: the GHZ state has genuine three-party entanglement that cannot be decomposed into pairwise entanglement. To detect this, one must look at two-qubit reduced density matrices and higher-order correlation functions.

This example illustrates a general principle: the more parties a state is entangled with, the more mixed each subsystem appears individually. Maximal entanglement of a subsystem with everything else leads to maximal local mixedness.

23.8 Decoherence: From Pure to Mixed

The Central Question

If quantum mechanics allows macroscopic superpositions — Schrodinger's cat alive AND dead — why don't we ever see them? Why does the macroscopic world look classical?

The answer is decoherence: the rapid, unavoidable entanglement of a quantum system with its environment, which suppresses quantum coherences (off-diagonal density matrix elements) on timescales so short that they are unobservable for macroscopic objects.

The Mechanism

Consider a qubit system $S$ initially in a superposition:

$$|\psi_S\rangle = \alpha|0\rangle + \beta|1\rangle$$

and an environment $E$ initially in some reference state $|e_0\rangle$. The initial state of the composite system is:

$$|\Psi(0)\rangle = (\alpha|0\rangle + \beta|1\rangle) \otimes |e_0\rangle$$

Now suppose the system interacts with the environment in a way that correlates (entangles) the system states with distinct environment states. Concretely, suppose:

$$|0\rangle|e_0\rangle \to |0\rangle|e_0\rangle, \quad |1\rangle|e_0\rangle \to |1\rangle|e_1\rangle$$

where $|e_0\rangle$ and $|e_1\rangle$ are orthogonal (or nearly orthogonal) environment states. This is a "which-path" interaction: the environment acquires information about which state the system is in.

By linearity, the composite state evolves to:

$$|\Psi(t)\rangle = \alpha|0\rangle|e_0\rangle + \beta|1\rangle|e_1\rangle$$

The composite state is still pure. No information has been lost from the universe. But now take the partial trace over the environment to find the reduced density matrix of the system:

$$\hat{\rho}_S(t) = \text{Tr}_E(|\Psi(t)\rangle\langle\Psi(t)|)$$

$$= |\alpha|^2 |0\rangle\langle 0| + \alpha\beta^*\underbrace{\langle e_1|e_0\rangle}_{\text{overlap}} |0\rangle\langle 1| + \alpha^*\beta\underbrace{\langle e_0|e_1\rangle}_{\text{overlap}} |1\rangle\langle 0| + |\beta|^2|1\rangle\langle 1|$$

If the environment states become orthogonal, $\langle e_0|e_1\rangle = 0$, and:

$$\hat{\rho}_S(t) = |\alpha|^2 |0\rangle\langle 0| + |\beta|^2|1\rangle\langle 1|$$

The off-diagonal coherences are gone. The system has been transformed from a pure superposition into a classical mixture — not by any mysterious "collapse," but by the perfectly unitary, deterministic entanglement with the environment. The information about the relative phase between $\alpha$ and $\beta$ has not been destroyed; it has been dispersed into the correlations between the system and the environment, where it is effectively unrecoverable.

💡 Key Insight: Decoherence does not require any special "measurement apparatus" or any departure from unitary quantum mechanics. It is the inevitable consequence of any interaction that creates correlations between a system and its environment. The quantum coherence is not destroyed — it is delocalized into the inaccessible environmental degrees of freedom.

Worked Example: Decoherence with a Single Environmental Qubit

Let us work through the decoherence mechanism in complete detail for the simplest possible environment: a single qubit.

Setup: System qubit $S$ starts in $|\psi_S\rangle = \frac{1}{\sqrt{2}}(|0\rangle + |1\rangle)$. Environment qubit $E$ starts in $|0\rangle_E$. The interaction is a controlled-NOT gate: $\hat{U}_{\text{CNOT}}|0\rangle_S|0\rangle_E = |0\rangle_S|0\rangle_E$ and $\hat{U}_{\text{CNOT}}|1\rangle_S|0\rangle_E = |1\rangle_S|1\rangle_E$.

Step 1: Initial state.

$$|\Psi(0)\rangle = \frac{1}{\sqrt{2}}(|0\rangle_S + |1\rangle_S)|0\rangle_E = \frac{1}{\sqrt{2}}(|00\rangle + |10\rangle)$$

The system density matrix is $\hat{\rho}_S(0) = |{+x}\rangle\langle{+x}| = \frac{1}{2}\begin{pmatrix} 1 & 1 \\ 1 & 1 \end{pmatrix}$ — a pure state with full coherence.

Step 2: After the interaction.

$$|\Psi(t)\rangle = \frac{1}{\sqrt{2}}(|0\rangle_S|0\rangle_E + |1\rangle_S|1\rangle_E) = |\Phi^+\rangle$$

This is a Bell state! The system and environment are now maximally entangled.

Step 3: Partial trace over the environment.

$$\hat{\rho}_S(t) = \text{Tr}_E(|\Phi^+\rangle\langle\Phi^+|) = \frac{1}{2}|0\rangle\langle 0| + \frac{1}{2}|1\rangle\langle 1| = \frac{1}{2}\begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix}$$

The off-diagonal coherences are completely gone. The system has gone from a pure superposition to a maximally mixed state in a single step. The purity has dropped from 1 to 1/2. The entropy has jumped from 0 to $\ln 2$.

Where did the coherence go? It did not vanish. The total state $|\Phi^+\rangle$ is still pure (entropy zero). The "missing" coherence is encoded in the correlations between system and environment — the entanglement. If we could bring the environment back and perform a joint measurement on both qubits, we would find perfect correlations that reveal the original superposition. But if we have lost access to the environment — as we always have in practice — the coherence is effectively gone forever.

🧪 Thought Experiment: The step-by-step calculation above is not just a textbook exercise. It is essentially what happens in a which-path experiment (Chapter 7). When a photon passes through a double slit and scatters a "which-path marker" particle, the marker becomes the "environment" that decoheres the photon's path superposition. The interference pattern vanishes — not because of any disturbance to the photon, but because the which-path information is now encoded in the marker particle.

The Decoherence Timescale

In realistic models, the overlap $\langle e_0|e_1(t)\rangle$ decays exponentially:

$$|\langle e_0|e_1(t)\rangle| \approx e^{-t/\tau_d}$$

where $\tau_d$ is the decoherence time. The off-diagonal elements of $\hat{\rho}_S$ decay as:

$$\rho_{01}(t) = \rho_{01}(0)\, e^{-t/\tau_d}$$

For macroscopic objects, $\tau_d$ is absurdly short:

📊 By the Numbers: A dust grain (radius $\sim 10^{-5}$ m) in air at room temperature, placed in a superposition of two positions separated by its own diameter, decoheres in roughly $10^{-31}$ seconds. This is $10^{17}$ times shorter than the Planck time ($\sim 10^{-44}$ s). There is no experiment imaginable that could observe such a superposition. Decoherence is why the macroscopic world appears classical — and it is not a subtle effect. It is brutally fast.

Dephasing Model

The simplest quantitative model of decoherence is pure dephasing. Consider a qubit with density matrix:

$$\hat{\rho}_S(0) = \begin{pmatrix} \rho_{00} & \rho_{01} \\ \rho_{10} & \rho_{11} \end{pmatrix}$$

Under pure dephasing (no energy exchange with the environment, only phase randomization):

$$\hat{\rho}_S(t) = \begin{pmatrix} \rho_{00} & \rho_{01}\,e^{-\Gamma t} \\ \rho_{10}\,e^{-\Gamma t} & \rho_{11} \end{pmatrix}$$

where $\Gamma = 1/\tau_d$ is the dephasing rate and $\tau_d = T_2$ is the dephasing time (a notation borrowed from NMR spectroscopy).

The populations (diagonal elements) are unchanged — the system does not gain or lose energy. Only the coherences decay. The purity decreases monotonically from $\text{Tr}(\hat{\rho}^2)$ at $t = 0$ toward $(|\rho_{00}|^2 + |\rho_{11}|^2)$ as $t \to \infty$.

On the Bloch sphere, dephasing corresponds to the Bloch vector shrinking toward the $z$-axis: the $x$ and $y$ components decay to zero, while $r_z$ is preserved.

Amplitude Damping

A more general decoherence channel is amplitude damping, which models energy relaxation (the system losing energy to the environment, like spontaneous emission). In this case:

$$\rho_{00}(t) \to 1 - (1 - \rho_{00}(0))\,e^{-t/T_1}$$ $$\rho_{11}(t) \to \rho_{11}(0)\,e^{-t/T_1}$$ $$\rho_{01}(t) \to \rho_{01}(0)\,e^{-t/(2T_1)}$$

where $T_1$ is the relaxation time. The populations now change — the excited state decays toward the ground state — and the coherences decay at a rate $1/(2T_1)$.

In general, the dephasing rate $1/T_2$ and the relaxation rate $1/T_1$ are related by:

$$\frac{1}{T_2} = \frac{1}{2T_1} + \frac{1}{T_\phi}$$

where $T_\phi$ is the pure dephasing contribution. This relation, central to NMR spectroscopy and quantum computing, ensures that $T_2 \leq 2T_1$.

🔗 Connection: The $T_1$ and $T_2$ times are the most important figures of merit for quantum computing hardware. Every qubit modality — superconducting, trapped ion, photonic, topological — is ultimately characterized by these timescales. We will return to this in Chapter 25 (Quantum Information) and Chapter 33 (Lindblad master equation).

Pointer States and the Preferred Basis

A deep question remains: decoherence destroys coherences, but in which basis? The answer is that the interaction Hamiltonian $\hat{H}_{\text{int}}$ between system and environment selects a preferred basis — the pointer basis — in which the density matrix becomes diagonal. This is environment-induced superselection (or einselection), a term coined by Wojciech Zurek.

The pointer basis consists of states that are most robust against environmental monitoring. For position-dependent interactions (like scattering of air molecules or photons), the pointer basis is the position basis. This is why macroscopic objects are always found in well-defined positions, not in position superpositions.

⚖️ Interpretation: Decoherence is interpretation-neutral. Copenhagen, many-worlds, Bohmian mechanics, and QBism all accept decoherence as a physical process. What they disagree on is what happens when decoherence is complete: Copenhagen says the state has collapsed; many-worlds says the branches have separated; Bohmian mechanics says the particle always had a definite position. Decoherence explains why we observe definite outcomes, but it does not — by itself — explain why one particular outcome occurs. This is the measurement problem, which persists regardless of decoherence. We return to it in Chapter 28.

23.9 The Quantum-to-Classical Transition (Preview of Chapter 33)

What Decoherence Achieves

Decoherence provides a dynamical explanation for the emergence of classicality from quantum mechanics:

1. Superposition → Mixture: Environmental entanglement converts a pure superposition into a mixed state (in the pointer basis) on the decoherence timescale $\tau_d$.

2. Pointer basis selection: The system-environment interaction selects the preferred basis in which the density matrix diagonalizes.

3. Effective classicality: After decoherence, the reduced density matrix looks exactly like a classical probability distribution over the pointer states. All measurements yield results consistent with classical statistics.

What Decoherence Does Not Achieve

🔴 Warning: Decoherence is necessary but not sufficient to solve the measurement problem. It explains why we don't observe interference between macroscopic alternatives, but it does not explain why we observe one particular outcome. After decoherence, the density matrix is $\hat{\rho} \approx |\alpha|^2|0\rangle\langle 0| + |\beta|^2|1\rangle\langle 1|$. This is a statistical mixture — it looks like a classical coin that has already been flipped but we haven't looked yet. But how and when did the coin "land"? Decoherence is silent on this. Different interpretations provide different answers.

The Hierarchy of Timescales

For a macroscopic object interacting with a thermal environment:

$$\tau_d \ll \tau_{\text{thermal}} \ll \tau_{\text{dynamical}}$$

• $\tau_d$: decoherence time (superposition → mixture)
• $\tau_{\text{thermal}}$: thermalization time (approach to thermal equilibrium)
• $\tau_{\text{dynamical}}$: characteristic time for the object's macroscopic motion

The radical separation between $\tau_d$ and $\tau_{\text{dynamical}}$ is why quantum effects are invisible in everyday life. By the time you blink, decoherence has already acted — not a million times over, but $10^{40}$ times over.

Quantum Darwinism and Redundant Information

An elegant extension of the decoherence program, developed by Zurek, is the idea of quantum Darwinism. Not only does the environment decohere the system into the pointer basis — it also broadcasts redundant copies of the classical information (which pointer state the system is in) into many independent fragments of the environment.

Consider a cat interacting with $N$ photons scattered off its body. Each photon carries partial information about the cat's state (alive or dead). When many photons have scattered, the "alive/dead" information is redundantly encoded in countless environmental fragments. An observer who intercepts any sufficiently large subset of these photons will learn the cat's state. This redundancy is what makes classical reality objective — different observers, accessing different fragments of the environment, all agree on what they see.

In the density matrix formalism, quantum Darwinism is a statement about the mutual information between the system $S$ and fragments $F$ of the environment: $I(S:F) \approx S(\hat{\rho}_S)$ for sufficiently large $F$, and this saturation occurs for many independent, non-overlapping fragments.

Toward the Lindblad Master Equation

The decoherence models we have discussed here — dephasing, amplitude damping — are special cases of a more general framework: the Lindblad master equation (also called the GKSL equation), which governs the time evolution of an open quantum system:

$$\frac{d\hat{\rho}}{dt} = -\frac{i}{\hbar}[\hat{H}, \hat{\rho}] + \sum_k \left(\hat{L}_k \hat{\rho} \hat{L}_k^\dagger - \frac{1}{2}\{\hat{L}_k^\dagger \hat{L}_k, \hat{\rho}\}\right)$$

The first term is the usual unitary (von Neumann) evolution. The second term, involving the Lindblad operators $\hat{L}_k$, captures the non-unitary effects of the environment: dissipation, dephasing, spontaneous emission, and thermal excitation.

We will develop this formalism rigorously in Chapter 33. For now, the essential point is:

💡 Key Insight: The Lindblad equation is to open quantum systems what the Schrodinger equation is to closed ones. It is the most general Markovian evolution equation that preserves the positivity and trace of the density matrix. Every quantum channel — every possible physical process that maps density matrices to density matrices — can be expressed in Lindblad form.

23.10 Summary and Progressive Project

Summary of Key Results

The density operator formalism extends quantum mechanics from pure states to the most general description of a quantum system.

Core definitions: - Pure state: $\hat{\rho} = |\psi\rangle\langle\psi|$ - Mixed state: $\hat{\rho} = \sum_k p_k |\psi_k\rangle\langle\psi_k|$

Three defining properties: $\hat{\rho}^\dagger = \hat{\rho}$, $\text{Tr}(\hat{\rho}) = 1$, $\langle\phi|\hat{\rho}|\phi\rangle \geq 0$

Key formulas:

The decoherence story: 1. System in superposition + environment in reference state 2. System-environment interaction creates entanglement 3. Partial trace over environment → coherences decay 4. System appears classical in the pointer basis 5. Timescale: $\tau_d$ (absurdly fast for macroscopic objects)

The Conceptual Landscape

                    DENSITY OPERATOR
                    ρ = Σ pₖ |ψₖ⟩⟨ψₖ|
                         /          \
                        /            \
                PURE STATE         MIXED STATE
              ρ = |ψ⟩⟨ψ|         Tr(ρ²) < 1
              Tr(ρ²) = 1         S > 0
              S = 0
                                   /           \
                                  /             \
                         CLASSICAL          ENTANGLED
                         IGNORANCE          SUBSYSTEM
                         (coin flip)       (partial trace)
                                                |
                                           DECOHERENCE
                                        pure → mixed via
                                     environmental entanglement

Progressive Project: Density Matrix Module

In this chapter's project checkpoint, you will build the DensityMatrix module for the Quantum Simulation Toolkit. The module should:

1. Construct density matrices from pure states (from_ket) and from statistical ensembles (from_mixture)
2. Verify that a given matrix satisfies the three density operator properties
3. Compute purity ($\text{Tr}(\hat{\rho}^2)$), von Neumann entropy, and expectation values
4. Perform partial traces for bipartite systems
5. Simulate decoherence: dephasing and amplitude damping channels

See code/project-checkpoint.py for the full implementation.

Looking Ahead

The density matrix is the foundation for everything that follows in Part V:

• Chapter 24: Entanglement and Bell's theorem — the partial trace and entanglement entropy from this chapter become essential tools.
• Chapter 25: Quantum information — quantum channels, quantum error correction, and the capacity of noisy channels all require the density operator.
• Chapter 26: Condensed matter — thermal density matrices describe electrons in solids.
• Chapter 27: Quantum optics — coherent states and thermal light are naturally described by density operators.

And looking further ahead:

• Chapter 28: The measurement problem — decoherence and the density matrix are central to every interpretation of quantum mechanics.
• Chapter 33: The Lindblad master equation — the full theory of open quantum systems, building on the decoherence models of this chapter.
• Chapter 35: Quantum error correction — protecting quantum information against decoherence.

You have now graduated from describing quantum mechanics with kets alone to the full generality of density operators. This is not merely a mathematical upgrade — it is a conceptual shift. Quantum mechanics is not just about what we know; it is about what we can know, given that our systems are always entangled with degrees of freedom we cannot access. The density operator is the honest bookkeeping of that fundamental limitation.

"The density matrix is the quantum physicist's conscience — it forces you to be honest about what you don't know." — Anonymous

Key Equations Reference