Measurement and Wavefunction Collapse

Between measurements, quantum mechanics is a model of decorum. The state vector glides through Hilbert space under the Schrödinger equation — smooth, deterministic, reversible, and (as the dynamics chapter will formalize) unitary. Then somebody looks. In that instant the theory switches rulebooks: the outcome is random, chosen from the measured observable's eigenvalue menu with Born-rule odds, and the state is rewritten — projected onto the eigenspace of the value actually found and renormalized, all other branches deleted. This second rulebook is the collapse postulate, and this article is its owner's manual: not which outcome occurs (no one knows a mechanism), but exactly what the state looks like the moment after. The update rule earns its keep in three theorems' worth of consequences. It is the unique prescription that makes measurements repeatable — ask the same question twice in quick succession and nature answers the same way, with certainty. In its degenerate-safe form (the Lüders rule) it projects onto the whole eigenspace rather than jumping to some arbitrary eigenvector, quietly preserving every superposition the measurement did not actually resolve — a distinction with computable, testable consequences. And it is demonstrably not a unitary process: a two-line linearity argument shows no Schrödinger-style evolution can do collapse's job, which is why the postulate list needs a separate axiom at all. You will run the rule on sequential Stern–Gerlach experiments — where measuring one spin component visibly re-randomizes another, numbers you can check — and on degenerate spectra, where Lüders and "naive projection" disagree. What collapse means is a live question parked, honestly and by name, with the interpretations articles at this track's end. What collapse does is settled arithmetic, and it is yours by the end of the page.

Key takeaways

Measuring $\hat A$ on $|\psi\rangle$ yields eigenvalue $a_n$ with Born probability $P(a_n) = \langle\psi|\hat P_n|\psi\rangle$ ; the state then updates to $|\psi'\rangle = \hat P_n|\psi\rangle / \|\hat P_n\psi\|$ — project, then renormalize.
The normalization constant is exactly $\sqrt{P(a_n)}$ : likelier outcomes need less rescaling.
Repeatability: an immediate re-measurement of $\hat A$ returns the same $a_n$ with probability 1 — collapse is what makes measured values stick.
Lüders rule (degenerate-safe): project onto the whole eigenspace; superpositions within the eigenspace survive untouched. Jumping to a single eigenvector is a different (wrong-by-default) update.
Collapse is not unitary: a linear deterministic map that fixes every eigenstate would have to send superpositions to superpositions — it cannot select outcomes.
Sequential incompatible measurements re-randomize each other: in the Stern–Gerlach chain $z^+ \to x \to z$ , the final $z^-$ appears with probability $1/4$ per selected branch, though it was impossible before the middle magnet.
Key idea: measurement = project onto an eigenspace (probability set by the Born rule) and renormalize — one non-unitary update that both selects an eigenvalue and rewrites the state.

The core insight

Measurement = project onto an eigenspace (probability set by the Born rule) and renormalize; this single non-unitary update both selects an eigenvalue and rewrites the state.

Where this lives today

Equation (1) is executed millions of times a second in working quantum hardware. Superconducting-qubit processors implement mid-circuit measurement with explicit Lüders semantics: readout projects a multi-qubit register onto the measured qubit's outcome eigenspace only, leaving coherences among unmeasured qubits intact — the degenerate-safe rule is a design specification, stress-tested in error-correction cycles where syndrome measurements must not disturb the encoded logical superposition. Quantum-optics laboratories realize the "measured but unread" row daily: heralded single-photon sources condition on a detector click — collapse as filtering. And the postulate's boundary is itself an industry: weak measurements trade information for gentleness on a quantitative curve (POVMs, the entanglement chapter's generalization), while decoherence theory explains why macroscopic apparatus behave as if rule (1) fired even in fully unitary models. The rule you learned on three levels is the one the machines run.

The physical picture

Picture the state as a chord on a strange piano. Before measurement, $|\psi\rangle$ sounds several eigen-notes of $\hat A$ at once, each with its amplitude. Measurement is not listening to the chord — it is the piano abruptly deciding to play exactly one note, with odds set by the loudness of each component, and then becoming that note: every other string is damped to silence. Play again immediately and you hear the same note, guaranteed. That final clause — the same note, guaranteed — is the physical content of collapse, and it is an experimental fact about the world (filtered atomic beams stay filtered) before it is an axiom.

Figure

Collapse as state surgery — project onto the outcome's eigenspace, renormalize, replay

Why a projection, of all operations? Because of what the formalism chapter has already certified. Outcomes are eigenvalues of a Hermitian operator, and eigenstates of distinct outcomes are orthogonal (Hermitian Operators and Observables proved both). If the measured value $a_n$ is to be reproducible, the post-measurement state must live entirely inside the $a_n$ -eigenspace — zero amplitude on every rival outcome. The gentlest map that does this, the one that deletes forbidden components while leaving everything compatible with the answer untouched, is orthogonal projection: the very operator $\hat P_n$ whose theory Projection Operators built and whose expectation, by the Born rule article, already computes $P(a_n)$ . One object, both jobs — probability and update. The picture also fixes what collapse is not. It is not a physical wave crumpling in space; $|\psi\rangle$ is a vector in state space, not a water wave. It is not the resolution of mere ignorance — the chord genuinely contained several notes, as interference experiments certify (a point sharpened below and in the density-operator article). And it is not dynamics: no time parameter appears in the update. The postulate is silent about how long collapse takes and what enacts it — that silence is the measurement problem, owned by the entanglement chapter's The Quantum Measurement Problem, not by this page.

The update rule, degenerate-safe from the start

State the rule once, in the form that survives every case. Let $\hat A = \sum_n a_n \hat P_n$ be the measured observable, with $\hat P_n$ the orthogonal projector onto the (possibly many-dimensional) eigenspace of $a_n$ — the spectral packaging assembled by the completeness and projector articles. If the measurement returns $a_n$ , then

|\psi\rangle \;\longrightarrow\; |\psi'\rangle = \frac{\hat P_n|\psi\rangle}{\sqrt{\langle\psi|\hat P_n|\psi\rangle}} \qquad (1)

where: the numerator amputates every component of $|\psi\rangle$ outside the winning eigenspace, and the denominator — exactly $\|\hat P_n\psi\| = \sqrt{P(a_n)}$ , the square root of the Born probability — restores unit norm so that tomorrow's probabilities also sum to one. Note the two readings of the same projector: sandwiched, it gives the odds ( $P(a_n) = \langle\psi|\hat P_n|\psi\rangle$ , the Born Rule article's degenerate form); applied, it performs the update. Improbable outcomes need violent renormalization (dividing by a small $\sqrt{P}$ ); certain ones ( $\hat P_n|\psi\rangle = |\psi\rangle$ ) change nothing at all. Equation (1) is the Lüders rule — the modern, degenerate-safe statement of what older texts call the projection postulate. For a non-degenerate outcome the eigenspace is a line, $\hat P_n = |a_n\rangle\langle a_n|$ , and (1) collapses (pun intended) to the familiar "the state becomes $|a_n\rangle$ ", global phase irrelevant as always. The interesting physics hides in the degenerate case, where "becomes the eigenstate" is ambiguous — which eigenstate? — and (1) answers: the one your state already was, as seen from inside the eigenspace. The measurement resolves only what it measures.

Derivation AThe Lüders update — and why repeatability forces the square root

We derive the update rule's structure from two operational demands: (i) an immediate re-measurement of $\hat A$ must return the same value $a_n$ with certainty (repeatability — the experimental fact the postulate encodes), and (ii) the post-measurement state must remain a legal, unit-norm state that carries no more disturbance than the answer requires. The strategy: let repeatability pin the subspace, minimal disturbance pin the map, and normalization pin the constant. Watch how little freedom survives.

Step 1. Repeatability confines the posterior. A second measurement returns $a_n$ with certainty iff the post-state $|\psi'\rangle$ satisfies $\hat P_n|\psi'\rangle = |\psi'\rangle$ — i.e. $|\psi'\rangle$ lies entirely inside the $a_n$ -eigenspace. (Any component in a rival eigenspace gives, by the Born rule, a nonzero chance of a different answer — contradiction.) Why this uses orthogonality: distinct-eigenvalue eigenspaces are orthogonal by the Hermiticity theorems, so "no rival amplitude" is a clean geometric statement.

Step 2. Minimal disturbance selects the projection. Among all maps sending $|\psi\rangle$ into the eigenspace, orthogonal projection $\hat P_n|\psi\rangle$ is the one that leaves every intra-eigenspace component exactly as it was — it is the closest point in the subspace to the original state (the best-approximation property from Projection Operators, recalled not re-proved). Any other choice would alter amplitudes the measurement never interrogated, injecting disturbance with no informational license.

Step 3. Normalization fixes the constant. $\|\hat P_n\psi\|^2 = \langle\psi|\hat P_n^\dagger\hat P_n|\psi\rangle = \langle\psi|\hat P_n|\psi\rangle = P(a_n)$ , using idempotence and Hermiticity of projectors. Dividing by $\sqrt{P(a_n)}$ is therefore exactly what unit norm demands — the Born probability reappears as the renormalization, a consistency lock between the two halves of the measurement postulate.

Step 4. Check the certainty limit: if $|\psi\rangle$ is already an $a_n$ -eigenstate, $\hat P_n|\psi\rangle = |\psi\rangle$ , $P(a_n) = 1$ , and (1) returns $|\psi\rangle$ unchanged — measuring a value the state already possesses is undisturbing, the quiet clause behind every "prepare by filtering" protocol in the laboratory. Verified: numerically on a 3-level system (projector onto a 2-dim eigenspace): posterior unit-norm; intra-eigenspace amplitude ratio preserved exactly; re-measurement probability $= 1$ ; $\|\hat P_n\psi\| = \sqrt{P(a_n)}$ to machine precision; $\hat P_n^2 = \hat P_n = \hat P_n^\dagger$ and $\sum_n \hat P_n = \mathbb 1$ with probabilities summing to 1.

The degenerate case deserves its own exhibit, because here Lüders and folklore part ways. Take a three-level system where $\hat A$ has eigenvalue $a$ on the two-dimensional span of $|1\rangle, |2\rangle$ and eigenvalue $b$ on $|3\rangle$ , and prepare

|\psi\rangle = \frac{3|1\rangle + 4i|2\rangle + 5|3\rangle}{\sqrt{50}} \qquad (2)

where: the numbers are chosen so both outcomes are equally likely: $P(a) = (9+16)/50 = 1/2$ and $P(b) = 25/50 = 1/2$ (machine-checked). If the measurement returns $a$ , rule (1) gives $|\psi'\rangle = (3|1\rangle + 4i|2\rangle)/5$ — the internal superposition survives, complex phase and all, merely rescaled. A "naive projection" that jumped to $|1\rangle$ or $|2\rangle$ (or to some apparatus-chosen combination) would be a different physical prediction: a subsequent measurement distinguishing $(3|1\rangle+4i|2\rangle)/5$ from $|1\rangle$ would expose it. Coarse measurements make coarse cuts — that is the Lüders discipline, and it is testable.

Figure

Lüders vs naive projection — a degenerate outcome resolves only what it measured

Collapse cannot be unitary — the no-go

The dynamics between measurements is linear: it acts on superpositions component by component. Could collapse secretly be more of the same — some fast, complicated, but ultimately Schrödinger-style evolution? A short argument says no, and it is worth running carefully because it explains why the postulate list needs a separate measurement axiom at all.

Go deeper · Derivation B · 1 minNo linear deterministic map can perform collapse

Derivation BNo linear deterministic map can perform collapse

We prove that collapse cannot be any fixed linear, deterministic evolution — unitary or otherwise. The strategy is proof by contradiction: assume a single linear map $\hat U$ implements "measure $S_z$ " on every input state, extract what repeatability forces on eigenstates, and let linearity compute the fate of a superposition. The two requirements collide in one line. Watch that the argument uses nothing about mechanisms — only linearity and the definition of a measurement having outcomes.

Step 1. Repeatability pins the eigenstates. A state that already has the value must keep it: measuring $S_z$ on $|{+z}\rangle$ must return $+\hbar/2$ and leave the state in the $+z$ eigenspace, so $\hat U|{+z}\rangle = e^{i\alpha}|{+z}\rangle$ , and likewise $\hat U|{-z}\rangle = e^{i\beta}|{-z}\rangle$ . Why phases are the only freedom: rays, not vectors, are states — the Hilbert-space article's global-phase doctrine.

Step 2. Linearity computes the superposition's fate:

\hat U\,|{+x}\rangle = \hat U\,\frac{|{+z}\rangle + |{-z}\rangle}{\sqrt 2} = \frac{e^{i\alpha}|{+z}\rangle + e^{i\beta}|{-z}\rangle}{\sqrt 2} \qquad (3)

Step 3. Read the contradiction: the right side is a fifty-fifty superposition of both outcomes for every choice of phases — never an eigenstate of $S_z$ . But a measurement of $S_z$ on $|{+x}\rangle$ must end with the system in one $z$ -eigenstate (that is what having an outcome means). No linear deterministic $\hat U$ can do both jobs.

Step 4. Name what must give. Actual collapse keeps repeatability and outcome-definiteness and sacrifices deterministic linearity: the update (1) is nonlinear in $|\psi\rangle$ (the normalization depends on the state) and stochastic (which projector fires is random). Every program that tries to recover collapse from pure unitary dynamics — decoherence, many-worlds, objective-collapse models — must correspondingly add randomness, add nonlinearity, or reinterpret "outcome"; the scorecard lives in the entanglement chapter's measurement-problem article. Verified: numerically — a linear map fixing both $z$ -eigenstates maps $|{+x}\rangle$ to a state with equal-magnitude components on both outcomes (never an eigenstate), for arbitrary phase assignments.

The moral is structural, not mechanical: determinism plus linearity is exactly what collapse must violate. That is why measurement stands beside — not inside — the dynamics in The Postulates of Quantum Mechanics. One more consequence is worth owning now. If the measurement is performed but the result not read, the correct description is no single post-state but the ensemble $\{P(a_n), |\psi'_n\rangle\}$ — which the density-operator article (three serials on) packages as $\rho' = \sum_n \hat P_n\rho\hat P_n$ . The unread measurement erases coherence between eigenspaces while preserving it within them (verified numerically on state (2): the $|1\rangle\!-\!|2\rangle$ coherence survives, both $|1\rangle\!-\!|3\rangle$ and $|2\rangle\!-\!|3\rangle$ die, trace still 1). Collapse for the reader, dephasing for the non-reader — same postulate, two bookkeepings.

Limiting cases

case	result	meaning
state already an eigenstate	$P = 1$ , state unchanged	measuring a possessed value is undisturbing — filtering = preparation
non-degenerate outcome	posterior $= \|a_n\rangle$	the familiar textbook collapse; phase irrelevant
degenerate outcome	posterior $= \hat P_n\|\psi\rangle/\sqrt{P}$ , intra-eigenspace structure intact	Lüders: resolve only what was measured
outcome with $P(a_n) \to 0$	update undefined	you cannot condition on an impossible result — no division by zero
measurement performed, result unread	$\rho' = \sum_n \hat P_n\rho\hat P_n$	dephasing between eigenspaces; density-operator language
compatible observable measured next	statistics of the first value undisturbed	shared eigenbasis — the commutator article's doctrine
incompatible observable measured next	first value re-randomized	sequential Stern–Gerlach arithmetic below

Common trapCollapse just reveals a value the particle secretly had all along

Tempting, classical, and wrong. If measurement merely disclosed a pre-existing $S_z$ and a pre-existing $S_x$ , the sequential Stern–Gerlach chain below could not resurrect $z^-$ atoms from a beam from which $z^-$ had been perfectly removed — yet it does, with probability $1/4$ per selected branch, exactly as amplitude arithmetic predicts. Superpositions are not ignorance: the interference the double-slit article exhibits, and the no-go theorems catalogued at the track's end (Kochen–Specker, Bell), close every naive "hidden ledger" account. What survives honestly: measurement outcomes are created in context, the update rule (1) tells you the state after the answer exists, and "the particle had $a_n$ all along" is safe to say only for states already in an eigenspace — where collapse does nothing anyway.

Worked examples

Example 1 — the Stern–Gerlach chain (confidence builder). A beam of silver atoms is prepared spin-up along $z$ ( $|{+z}\rangle$ , the $z^-$ port verified dark). The beam passes an $x$ -oriented magnet; the $x^+$ output is selected; a final $z$ -magnet analyzes it. What fraction of the original beam lands in the final $z^-$ port — the "impossible" outcome?

Step 1. First magnet: $P(x^+) = |\langle{+x}|{+z}\rangle|^2 = 1/2$ , since $|{+x}\rangle = (|{+z}\rangle + |{-z}\rangle)/\sqrt2$ . Collapse: the selected atoms are now $|{+x}\rangle$ — the update rule (1) with a one-dimensional projector.

Step 2. Final magnet on $|{+x}\rangle$ : $P(z^-|x^+) = |\langle{-z}|{+x}\rangle|^2 = 1/2$ . The $z$ -information carried by the original preparation is gone — overwritten by the $x$ -collapse. Incompatible measurements re-randomize each other.

Step 3. Chain the branches: $P(z^+\!\to x^+\!\to z^-) = \tfrac12 \times \tfrac12 = \boxed{1/4}$ of the original beam — atoms exiting a port that was empty before the middle magnet was inserted (all numbers machine-verified). Remove the $x$ -magnet and the $z^-$ port goes dark again ( $P = 0$ ).

Step 4. Variant with both $x$ -beams kept but unread: sum the two paths, $\tfrac12\cdot\tfrac12 + \tfrac12\cdot\tfrac12 = 1/2$ in $z^-$ — the dephasing bookkeeping, no interference between distinguishable paths. Inserting a measurement changes the physics even when nobody looks at its result.

Example 2 — a degenerate collapse, run to the numbers (exam level). In the three-level system of equation (2) — $\hat A$ with eigenvalue $a$ on $\mathrm{span}\{|1\rangle,|2\rangle\}$ , $b$ on $|3\rangle$ — the measurement returns $a$ . A second observable $\hat B$ has eigenvector $|u\rangle = (|1\rangle + |2\rangle)/\sqrt2$ with eigenvalue $+1$ . Find the probability of measuring $\hat B = +1$ after the collapse, and compare with the naive-jump prediction.

Step 1. Collapse by rule (1): $P(a) = 1/2$ , posterior $|\psi'\rangle = (3|1\rangle + 4i|2\rangle)/5$ — verified normalized, the $i$ intact.

Step 2. Born rule on the posterior: $\langle u|\psi'\rangle = (3 + 4i)/(5\sqrt2)$ , so $P(B{=}+1) = |3+4i|^2/50 = 25/50 = \boxed{1/2}$ .

Step 3. Naive jump to $|1\rangle$ would predict $P = |\langle u|1\rangle|^2 = 1/2$ as well — a coincidence of this particular $|u\rangle$ . Sharpen with $|v\rangle = (|1\rangle - i|2\rangle)/\sqrt2$ , whose bra conjugates the $-i$ to $+i$ : $\langle v|\psi'\rangle = \tfrac{1}{\sqrt2}\bigl(\tfrac{3}{5} + i\cdot\tfrac{4i}{5}\bigr) = \tfrac{1}{\sqrt2}\cdot\tfrac{3-4}{5} = -\tfrac{1}{5\sqrt2}$ , so Lüders predicts $P = 1/50 = \boxed{0.02}$ — near-perfect destructive interference of the surviving amplitudes — while the jump-to- $|1\rangle$ story predicts $|\langle v|1\rangle|^2 = 1/2$ . One further measurement separates the two update rules by a factor of 25 — Lüders is not pedantry, it is physics.

Step 4. Moral: the surviving intra-eigenspace phase ( $4i$ against $3$ ) is measurable downstream. The measurement resolved $a$ versus $b$ and nothing else; equation (1) is the exact bookkeeping of that "nothing else."

The pragmatist of the projection postulate

Origins · 1 minAsher Peres and how this was discovered

Asher Peres

1934–2005 · Technion, Haifa

Born Aristide Pressman in a French village to Polish parents, renamed in Israel, Asher Peres became quantum mechanics' great operationalist — the physicist who insisted the theory is exactly what this article practices: a set of rules connecting state preparations to measurement statistics, no more and no less. His maxim "unperformed experiments have no results" (1978) is the sharpest one-line summary of why the warning box above is wrong-headed as classical intuition; his textbook Quantum Theory: Concepts and Methods (1993) rebuilt the subject with the measurement postulate — projectors, Lüders updates, repeatability — as the load-bearing wall rather than an embarrassing appendix. Peres treated collapse neither as a physical wave crumpling nor as a paradox to be anguished over, but as the update of a probability catalogue when new data arrives — while proving, with theorems, that this catalogue can never be a classical ledger of pre-existing values. The same operational rigor later made him a founder of quantum information: co-inventor of quantum teleportation (1993) and author of separability criteria that carry his name in the entanglement chapter, where this track meets him again.

History

1926

— Born's probability interpretation attaches odds to

|\psi|^2

; the "reduction" of the wave packet enters informal usage

1930

— Dirac's Principles codifies measurement as projection onto the observed eigenstate

1932

— von Neumann's Grundlagen formalizes the two evolution types (process 1: measurement; process 2: Schrödinger) and models the apparatus chain

1951

— Gerhart Lüders publishes the degenerate-safe update rule, correcting the naive projection for coarse measurements

1978

— Peres: "unperformed experiments have no results" — the operational reading of contextual outcomes

1990

s — decoherence program (Zurek and others) explains the appearance of collapse for unread measurements via entanglement with environments

Going further

References. Griffiths & Schroeter, Introduction to Quantum Mechanics, 3rd ed., Ch. 1 & 3 · Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., Ch. 1 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. I, Ch. III · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 4 · Townsend, A Modern Approach to Quantum Mechanics, Ch. 1–3 · Lüders, Ann. Phys. 8, 322 (1951); English translation Ann. Phys. 15, 663 (2006) · Peres, Quantum Theory: Concepts and Methods (Kluwer, 1993).

Provenance. Projection/Lüders rule and Born rule cross-checked against Sakurai & Napolitano Ch.1, Cohen-Tannoudji Vol.1 Ch.III, Shankar Ch.4, Griffiths Ch.1&3, Townsend Ch.1-3. Verifier should recompute: idempotence/Hermiticity of $\hat P_n$ ; $\sum_n P(a_n)=1$ via completeness; normalization constant $\sqrt{P(a_n)}$ ; the Stern-Gerlach probabilities $|\langle\pm x|+z\rangle|^2=1/2$ ; and the dephasing result $\rho'=\sum_n\hat P_n\rho\hat P_n$ with $\mathrm{Tr}\,\rho'=1$ . For this edition all five were machine-executed (numpy), plus this article's additions: the Lüders three-level example (equation 2: $P(a)=P(b)=1/2$ , posterior $(3|1\rangle+4i|2\rangle)/5$ with phase preserved and repeatability $P=1$ ), the sequential Stern–Gerlach chain ( $1/4$ selected, $1/2$ unread, $0$ without the middle magnet), the downstream Lüders-vs-jump discrimination ( $P(B{=}+1) = 1/2$ on $|u\rangle$ ; $0.02$ vs $0.5$ on $|v\rangle$ ), and the linearity no-go exhibit. Per the chapter ownership map: the postulate list is owned by The Postulates of Quantum Mechanics, outcome probabilities by Born Rule and Measurement Probabilities, projector theory by Projection Operators, and the apparatus/premeasurement chain by Von Neumann Measurement and the Projection Postulate in the entanglement chapter — this article owns the state-update rule itself.

Builds on: The Postulates of Quantum Mechanics · Born Rule and Measurement Probabilities · Projection Operators · Hermitian Operators and Observables · Eigenvalues and Eigenstates · Completeness Relation and Resolution of Identity

Leads to: Expectation Values and Variance · Density Operator and Mixed States · Quantum Decoherence

Is collapse a physical process that takes time, or just bookkeeping?

The postulate itself is silent — it specifies the state after an outcome exists, with no dynamics, duration, or mechanism. Operationally (Peres's stance) it is conditioning: you learned $a_n$ , you update the catalogue. Physically, decoherence theory shows the appearance of collapse — suppression of interference between outcome branches — develops continuously and very fast through entanglement with the environment, but decoherence alone never selects one outcome (it is unitary; see the linearity no-go). Whether something extra selects outcomes is the measurement problem, treated honestly in the entanglement chapter. For every calculation in this track, rule (1) plus the Born rule is complete and unambiguous.

Does collapse violate conservation of energy?

Measuring an observable incompatible with $\hat H$ genuinely changes the energy distribution: an energy eigenstate hit by a position measurement becomes a localized packet with energy spread. No paradox — the measured system is not closed during measurement; it interacted with an apparatus, and a careful ledger including the apparatus balances. What the postulate does guarantee: measuring $\hat H$ itself, or any observable commuting with it, leaves energy statistics untouched (the compatible row of the limiting-cases table).

Why do we need the Lüders form — when would I ever meet a degenerate measurement?

Constantly. Any measurement of part of a system is degenerate on the whole: reading one qubit of a register projects onto eigenspaces of dimension $2^{N-1}$ , and the Lüders rule is precisely why the unmeasured qubits keep their superpositions (error correction depends on it). Coarse-grained detectors ("the energy is in this band"), parity measurements, and any observable with symmetry all have degenerate spectra. The non-degenerate textbook case is the exception in the laboratory.

What happens if I measure but the detector fails to record which outcome occurred?

Then no single ket describes your situation; the honest object is the ensemble $\{P(a_n),\, \hat P_n|\psi\rangle/\sqrt{P(a_n)}\}$ , packaged as the density operator $\rho' = \sum_n\hat P_n\rho\hat P_n$ . This map preserves the trace, kills coherence between different eigenspaces, and keeps it within each one — verified numerically in this article. The distinction read/unread is not philosophy: the two situations give different downstream interference statistics, as Example 1's Step 4 computes.

After a position measurement, is the particle really in a delta-function state?

Idealized position measurement, yes — and that "state" is not normalizable, which is the honest signal that exact position measurement is an idealization. Real detectors have resolution $\Delta$ : the projector is onto a finite bin, the posterior a legitimate normalizable packet of width $\sim\Delta$ . The delta-function limit and the machinery to handle it are exactly the business of Continuous Spectra and Dirac Delta Normalization later in this chapter; the physics point survives unbruised — finer measurement, harsher momentum disturbance, per the uncertainty article.

Where do POVMs and weak measurements fit relative to this postulate?

Rule (1) covers ideal projective measurements — maximal information, maximal update. Generalized measurements (POVMs) relax both: outcomes correspond to positive operators that need not be projectors, updates use Kraus operators $\hat M_n$ with $|\psi'\rangle = \hat M_n|\psi\rangle/\|\hat M_n\psi\|$ , and projective measurement is the special case $\hat M_n = \hat P_n$ . Weak measurements sit at the gentle end: little information, little collapse. The general theory is the entanglement chapter's Generalized Measurements and POVMs; everything there reduces to this page when the measurement is sharp.