Quantum Measurement, Decoherence, and Interpretations of Quantum Mechanics
Quantum Measurement, Decoherence, and Interpretations of Quantum Mechanics
The quantum measurement problem sits at the heart of the foundational controversies of quantum mechanics. When a quantum system in a superposition state interacts with a measuring apparatus, what determines the outcome? The Schrödinger equation predicts the apparatus itself enters a superposition — the infamous Schrödinger's cat paradox. Yet in practice, measurements yield single definite outcomes. This tension between the linear, deterministic Schrödinger dynamics and the apparently discontinuous, probabilistic "collapse" has generated more philosophical debate than any other question in the history of physics.
The von Neumann (projective) measurement scheme provides a formal framework. An observable $\hat{A} = \sum_n a_n |a_n\rangle\langle a_n|$ is measured on a state $|\psi\rangle = \sum_n c_n|a_n\rangle$. The probability of outcome $a_n$ is $P(a_n) = |c_n|^2 = |\langle a_n|\psi\rangle|^2$ (Born rule). Upon obtaining $a_n$, the post-measurement state is $|a_n\rangle$ — the wavefunction "collapses." The expectation value is $\langle\hat{A}\rangle = \langle\psi|\hat{A}|\psi\rangle$. This framework is mathematically clean but physically mysterious: what is the physical mechanism of collapse, and when exactly does it occur?
The density matrix formalism generalizes quantum mechanics to encompass both pure states and statistical mixtures. A pure state $|\psi\rangle$ is represented by $\rho = |\psi\rangle\langle\psi|$ with $\text{Tr}(\rho) = 1$ and $\text{Tr}(\rho^2) = 1$. A mixed state (classical statistical ensemble or subsystem of an entangled system) has $\rho = \sum_i p_i |\psi_i\rangle\langle\psi_i|$ with $\text{Tr}(\rho^2) \leq 1$, with equality iff the state is pure. The von Neumann entropy $S(\rho) = -\text{Tr}(\rho \ln \rho) = -\sum_i \lambda_i \ln \lambda_i$ (where $\lambda_i$ are eigenvalues of $\rho$) vanishes for pure states and is maximized at $\ln d$ for the maximally mixed state $\rho = I/d$.
Decoherence provides the modern understanding of why quantum superpositions are not observed at macroscopic scales. When a quantum system $S$ interacts with its environment $E$, the total state becomes entangled: $|\psi\rangle_S \otimes |\text{env}_0\rangle \to \sum_n c_n |a_n\rangle_S \otimes |e_n\rangle_E$. Tracing over the environment gives a reduced density matrix $\rho_S = \text{Tr}_E(|\Psi\rangle\langle\Psi|) = \sum_n |c_n|^2 |a_n\rangle\langle a_n|$ (when $\langle e_m|e_n\rangle \approx \delta_{mn}$) — the off-diagonal coherences vanish. The preferred basis selected by this process (the "pointer states") are determined by the interaction Hamiltonian.
The quantum Zeno effect demonstrates that frequent measurement can suppress quantum evolution. If a system is measured $N$ times in a short interval $T$, the probability of finding it in its initial state is approximately $(1 - \frac{T^2}{N\tau^2})^N \to e^{-T^2/\tau^2} \to 1$ as $N \to \infty$ for small $T^2/\tau^2$. Paradoxically, sufficiently frequent measurement "freezes" the quantum dynamics. The anti-Zeno effect (frequent measurement accelerating decay) can also occur depending on the spectral density of the environment.
The POVM (Positive Operator-Valued Measure) formalism generalizes projective measurements. A POVM is a set of positive operators $\{E_m\}$ with $\sum_m E_m = I$ (completeness). The probability of outcome $m$ is $P(m) = \text{Tr}(E_m \rho)$. POVMs can distinguish non-orthogonal states optimally, are realized physically by coupling the system to an ancilla and performing a projective measurement on the combined system, and are essential for quantum information tasks like state discrimination and quantum key distribution.
The major interpretations of quantum mechanics each resolve the measurement problem differently. The Copenhagen interpretation (Bohr, Heisenberg) takes the collapse as a primitive, irreducible process and forbids asking what "really" happens between measurements. The many-worlds interpretation (Everett, 1957) denies collapse entirely — the universe splits into branches at each measurement. The de Broglie-Bohm pilot wave theory adds hidden variables (particle positions guided by the wavefunction), recovering determinism at the cost of nonlocality. Quantum Bayesianism (QBism) interprets the wavefunction as an agent's personal probability assignment. No experiment has yet distinguished between these interpretations, though quantum Darwinism (Zurek) attempts to derive the appearance of classical reality from decoherence and redundant information encoding in the environment.
Wigner's friend is a thought experiment that sharpens the measurement problem. Wigner's friend performs a measurement inside a sealed laboratory; Wigner, outside, regards the system+friend as a joint quantum system still in superposition. Does the friend's measurement collapse the wavefunction, or does Wigner's subsequent observation? Recent extensions (Frauchiger-Renner theorem, 2018) suggest that different observers applying the Born rule consistently can derive contradictory predictions if they treat each other as quantum systems — suggesting that quantum mechanics cannot be universally applied without modification.
The density operator (density matrix) $\rho$ is a positive semi-definite Hermitian operator with $\text{Tr}(\rho) = 1$. For a pure state $|\psi\rangle$: $\rho = |\psi\rangle\langle\psi|$, $\text{Tr}(\rho^2) = 1$. For a mixed state: $\rho = \sum_i p_i|\psi_i\rangle\langle\psi_i|$ where $p_i \geq 0$, $\sum_i p_i = 1$, and $\text{Tr}(\rho^2) < 1$. Time evolution: $i\hbar\partial_t\rho = [\hat{H}, \rho]$ (von Neumann equation, the density matrix analogue of Schrödinger). Expectation values: $\langle\hat{A}\rangle = \text{Tr}(\rho\hat{A})$.
The von Neumann entropy of a density matrix $\rho$ with eigenvalues $\{\lambda_i\}$ is:
$$S(\rho) = -\text{Tr}(\rho\ln\rho) = -\sum_i \lambda_i \ln\lambda_i$$
with the convention $0\ln 0 = 0$. Properties: $S(\rho) \geq 0$, with equality iff $\rho$ is a pure state. $S(\rho) \leq \ln d$ for a $d$-dimensional system, with equality iff $\rho = I/d$. The coherence magnitude $|\rho_{01}| \leq \sqrt{\rho_{00}\rho_{11}}$ by the Cauchy-Schwarz inequality. Subadditivity: $S(\rho_{AB}) \leq S(\rho_A) + S(\rho_B)$; strong subadditivity: $S(\rho_{ABC}) + S(\rho_B) \leq S(\rho_{AB}) + S(\rho_{BC})$.
When system $S$ couples to environment $E$ via $H_{SE} = \hat{A}_S \otimes \hat{B}_E$, the eigenstates of $\hat{A}_S$ (the pointer states) are selected as the preferred classical basis. Starting from $|\psi\rangle_S = \sum_n c_n|a_n\rangle$, the joint state evolves to:
$$|\Psi(t)\rangle = \sum_n c_n e^{-iE_n t/\hbar}|a_n\rangle_S|e_n(t)\rangle_E$$
The reduced density matrix is $\rho_S(t) = \sum_{n,m} c_n c_m^* e^{-i(E_n-E_m)t/\hbar}\langle e_m|e_n\rangle|a_n\rangle\langle a_m|$. When $\langle e_m|e_n\rangle \to \delta_{mn}$ (environment states become orthogonal on decoherence timescale $\tau_D$), the off-diagonal elements $\rho_{nm}$ for $n\neq m$ vanish exponentially: $|\rho_{nm}(t)| \propto e^{-t/\tau_D}$.
A Positive Operator-Valued Measure (POVM) is a set of operators $\{E_m\}$ satisfying: (1) $E_m \geq 0$ (positive semi-definite) for all $m$; (2) $\sum_m E_m = I$ (completeness). The measurement probability is $P(m) = \text{Tr}(\rho E_m)$. Unlike projective measurements, POVM elements need not be projectors ($E_m^2 \neq E_m$ in general) and can number more than the Hilbert space dimension. Naimark's theorem: every POVM on $\mathcal{H}$ is realized as a projective measurement on $\mathcal{H} \otimes \mathcal{H}_{\text{ancilla}}$.
A qubit is in state $\rho = \begin{pmatrix}0.7 & 0.4\\0.4 & 0.3\end{pmatrix}$. Compute $\text{Tr}(\rho)$, $\text{Tr}(\rho^2)$, and the von Neumann entropy $S(\rho)$.
Trace: $\text{Tr}(\rho) = 0.7 + 0.3 = 1$. ✓
$\rho^2$: $\rho^2 = \begin{pmatrix}0.7 & 0.4\\0.4 & 0.3\end{pmatrix}^2 = \begin{pmatrix}0.49+0.16 & 0.28+0.12\\0.28+0.12 & 0.16+0.09\end{pmatrix} = \begin{pmatrix}0.65 & 0.40\\0.40 & 0.25\end{pmatrix}$.
$\text{Tr}(\rho^2) = 0.65 + 0.25 = 0.90 < 1$, so this is a mixed state.
Eigenvalues: $\det(\rho - \lambda I) = (0.7-\lambda)(0.3-\lambda) - 0.16 = \lambda^2 - \lambda + 0.21 - 0.16 = \lambda^2 - \lambda + 0.05 = 0$.
$\lambda = \frac{1\pm\sqrt{1-0.20}}{2} = \frac{1\pm\sqrt{0.80}}{2}$. $\lambda_+ = \frac{1+0.894}{2} = 0.947$, $\lambda_- = \frac{1-0.894}{2} = 0.053$.
Von Neumann entropy: $S = -(0.947\ln 0.947 + 0.053\ln 0.053) = -(0.947\times(-0.0545) + 0.053\times(-2.937)) = 0.0516 + 0.1557 = 0.207$ nats.
A system starts in state $|+\rangle = \frac{1}{\sqrt{2}}(|0\rangle+|1\rangle)$. It interacts with an environment, with decoherence timescale $\tau_D = 10^{-12}$ s. Write the reduced density matrix at $t = \tau_D$ and $t = 10\tau_D$. How does coherence decay?
Initial state: $\rho_S(0) = |+\rangle\langle+| = \frac{1}{2}\begin{pmatrix}1&1\\1&1\end{pmatrix}$. Diagonal elements $\rho_{00}=\rho_{11}=\frac{1}{2}$, off-diagonal $\rho_{01}=\rho_{10}=\frac{1}{2}$.
Decoherence model: Off-diagonal elements decay as $\rho_{01}(t) = \rho_{01}(0)e^{-t/\tau_D} = \frac{1}{2}e^{-t/\tau_D}$. Diagonal elements are preserved (dephasing channel).
At $t = \tau_D$: $\rho_{01}(\tau_D) = \frac{1}{2}e^{-1} \approx 0.184$. So $\rho(\tau_D) = \frac{1}{2}\begin{pmatrix}1 & e^{-1}\\e^{-1}&1\end{pmatrix} = \begin{pmatrix}0.5 & 0.184\\0.184 & 0.5\end{pmatrix}$.
At $t = 10\tau_D$: $\rho_{01} = \frac{1}{2}e^{-10} \approx 2.27\times10^{-5} \approx 0$. The density matrix is essentially $\rho \approx \frac{1}{2}I$ — the maximally mixed state. All quantum coherence between $|0\rangle$ and $|1\rangle$ has been destroyed.
Design a POVM for distinguishing the non-orthogonal states $|+\rangle = \frac{1}{\sqrt{2}}(|0\rangle+|1\rangle)$ and $|-\rangle = \frac{1}{\sqrt{2}}(|0\rangle-|1\rangle)$ with minimum error. What is the Helstrom bound?
The two states are sent with equal prior probability $p_1 = p_2 = \frac{1}{2}$. The inner product is $\langle+|-\rangle = \frac{1}{2}(1-1) = 0$ — they are actually orthogonal!
Wait: $\langle+|-\rangle = \frac{1}{\sqrt{2}}\frac{1}{\sqrt{2}}(\langle0|+\langle1|)(|0\rangle-|1\rangle) = \frac{1}{2}(1-1) = 0$. Since they are orthogonal, they can be distinguished perfectly with a projective measurement in the $\{|+\rangle,|-\rangle\}$ basis (same as $\sigma_x$ eigenbasis). Error probability = 0.
Helstrom bound (general): For states $\rho_1, \rho_2$ with priors $p_1, p_2$: $P_{\text{error}} = \frac{1}{2}(1 - \|p_1\rho_1 - p_2\rho_2\|_1/2)$ where $\|\cdot\|_1$ is the trace norm. For equal priors and pure states $|\phi_1\rangle, |\phi_2\rangle$: $P_{\text{error,min}} = \frac{1}{2}(1-\sqrt{1-|\langle\phi_1|\phi_2\rangle|^2})$. For $|\langle+|-\rangle|=0$: $P_{\text{error}} = \frac{1}{2}(1-1) = 0$. ✓
Compare the Copenhagen interpretation and the many-worlds interpretation (MWI) in their treatment of the Schrödinger's cat thought experiment. What does each predict about the state of the cat before the box is opened?
Setup: A radioactive atom ($\frac{1}{2}$ probability of decaying in 1 hour) is coupled to a Geiger counter, which triggers a mechanism that kills a cat. After 1 hour, the combined state before observation is:
$$|\Psi\rangle = \frac{1}{\sqrt{2}}(|\text{decayed}\rangle|\text{dead cat}\rangle + |\text{not decayed}\rangle|\text{live cat}\rangle)$$
Copenhagen: This superposition is not a physical statement about the cat's actual state — the wavefunction is a calculational tool. The cat is either alive or dead (classical object); asking for its state before observation is meaningless. Collapse is instantaneous and irreversible at the classical/quantum boundary (Heisenberg cut), but the location of the cut is not precisely defined.
Many-Worlds (Everett): The wavefunction is physically real and never collapses. The universe branches into two: one in which the atom decayed and the cat is dead, and one in which the atom did not decay and the cat is alive. Both branches are equally real. The observer also branches when they open the box. There is no preferred basis problem in this toy example because decoherence selects the $\{|\text{dead}\rangle, |\text{alive}\rangle\}$ basis. Probability arises from the Born rule applied to branch amplitudes (Deutsch-Wallace theorem attempts to derive this from rationality axioms).
Practice Problems
Show Answer Key
1. $\text{Tr}(\rho^2) = (3/4)^2 + (1/4)^2 = 9/16+1/16 = 10/16 = 5/8 < 1$: mixed state. $S = -(3/4)\ln(3/4) - (1/4)\ln(1/4) = (3/4)\ln(4/3) + (1/4)\ln 4 = 0.2164 + 0.3466 = 0.563$ nats.
2. Any $2\times2$ Hermitian matrix with unit trace can be written as $\rho = \frac{1}{2}(I + r_x\sigma_x + r_y\sigma_y + r_z\sigma_z)$. Positivity requires eigenvalues $\geq 0$: eigenvalues are $\frac{1}{2}(1\pm|\mathbf{r}|)$, so $|\mathbf{r}|\leq 1$. Pure state iff $|\mathbf{r}|=1$ (surface of Bloch sphere).
3. $\rho = \frac{1}{d}I$ has all eigenvalues $1/d$. $S = -d\cdot\frac{1}{d}\ln\frac{1}{d} = \ln d$.
4. Projective measurement of $\sigma_z$: outcomes $+1$ (state $|0\rangle$) with probability $\rho_{00} = 1/2$, and $-1$ (state $|1\rangle$) with probability $\rho_{11} = 1/2$. The off-diagonal elements $\rho_{01} = i/2$ are irrelevant for $\sigma_z$ probabilities.
5. Eigenvalue decomposition: $\rho = \sum_i\lambda_i|\lambda_i\rangle\langle\lambda_i|$ with $\lambda_i\geq0$, $\sum\lambda_i=1$. Then $\rho^2 = \sum_i\lambda_i^2|\lambda_i\rangle\langle\lambda_i|$ and $\text{Tr}(\rho^2) = \sum_i\lambda_i^2 \leq (\sum_i\lambda_i)^2 = 1$ by convexity (equality iff all $\lambda_i = 0$ except one $\lambda_j=1$ — pure state).
6. Survival prob after one interval $T/N$: $p_1 = 1 - \Gamma T/N + O((T/N)^2) \approx (1-\Gamma T/N)$. After $N$ measurements: $P = (1-\Gamma T/N)^N \to e^{-\Gamma T}$ as $N\to\infty$ classically. But quantum mechanically for short times $p_1 = 1 - (\Delta H)^2 T^2/(\hbar^2 N^2)$, so $P_N = (1-(\Delta H)^2 T^2/(\hbar^2 N^2))^N \to 1$ as $N\to\infty$ (Zeno freezing).
7. $\frac{d\rho}{dt} = -\frac{i}{\hbar}[H,\rho] + \sum_k\left(L_k\rho L_k^\dagger - \frac{1}{2}L_k^\dagger L_k\rho - \frac{1}{2}\rho L_k^\dagger L_k\right)$. The first term is unitary evolution; the $L_k\rho L_k^\dagger$ term is the "quantum jump" (non-unitary evolution when the environment registers a quantum); the remaining terms preserve trace and ensure $\rho$ remains positive.
8. The Deutsch-Wallace approach: an agent with preferences satisfying certain rationality axioms (equivalence, dominance, problem continuity) must assign probabilities to measurement outcomes equal to the Born rule weights $|\langle a_n|\psi\rangle|^2$. Alternatively, Zurek's "envariance" (environment-assisted invariance) derives Born rule from symmetry of entangled state under permutation of environment basis.
9. Discord $D(A|B) = I(A:B) - J(A|B)$ where $J$ is classical correlations. Consider $\rho_{AB} = \frac{1}{2}|0\rangle\langle0|\otimes|0\rangle\langle0| + \frac{1}{2}|+\rangle\langle+|\otimes|1\rangle\langle1|$. This is separable (no entanglement), but the two states $|0\rangle$ and $|+\rangle$ for A are nonorthogonal, so any measurement on B disturbs A's state — nonzero discord $D > 0$.
10. Thermal state: $\rho_{\text{th}} = \frac{e^{-\beta\hbar\omega_0\sigma_z/2}}{Z}$ where $Z = 2\cosh(\beta\hbar\omega_0/2)$. In $\{|0\rangle,|1\rangle\}$ basis: $\rho = \begin{pmatrix}p_0&0\\0&p_1\end{pmatrix}$ with $p_0 = e^{\beta\hbar\omega_0/2}/Z$, $p_1 = e^{-\beta\hbar\omega_0/2}/Z$. $S = -p_0\ln p_0 - p_1\ln p_1 = \ln Z - \beta\hbar\omega_0\langle\sigma_z\rangle/2$.
11. Pointer states are eigenstates of $\hat{A}_S$ in $H_{SE} = \hat{A}_S\otimes\hat{B}_E$. They are selected because they are stable under monitoring by the environment: $[H_{SE}, \hat{A}_S\otimes I] = 0$, meaning pointer states do not spread under decoherence. The pointer basis is the eigenbasis of the system observable that appears in the interaction Hamiltonian.
12. In Frauchiger-Renner: Wigner (W) and anti-Wigner (W') observe Wigner's friend (F) and anti-friend (F'). (Q): W can treat F's lab quantum mechanically and predict F's outcomes. (C): F applies Born rule to her own measurement. (S): each measurement has a single outcome. In the specific protocol, W predicts a definite outcome for W's observation, but applying (Q) and (C) simultaneously leads W to assign probability 1 to a statement that F's (C) assigns probability 0 — a contradiction. One of the three assumptions must be false.