The Measurement Postulate

Every absorption or emission spectrum recorded in a chemistry lab is, at bottom, an application of the measurement postulate: transition probabilities computed as $|\langle f|\hat\mu|i\rangle|^2$ (Fermi's golden rule machinery built on exactly this Born-rule logic) predict line intensities in UV-Vis, IR, and Raman spectroscopy, while the discrete detection events at a photomultiplier or CCD pixel are individual collapse events whose statistics reproduce the predicted intensities only after many photons. In NMR and MRI, the entire technology of population-difference detection — the tiny excess of spins aligned with versus against the field, amplified into a detectable bulk magnetization — is a direct macroscopic manifestation of Born-rule statistics applied to enormous numbers of independently collapsing nuclear spins, exactly as worked through in Example 2 above but scaled to Avogadro-sized ensembles. Single-molecule spectroscopy and quantum-dot photoluminescence make the discreteness of collapse dramatically visible: a single fluorescent molecule or quantum dot "blinks" between bright and dark states, and each photon detection event is a real-time collapse of the electronic state, not an artifact of measurement noise. Photoelectron spectroscopy (XPS, UPS) similarly relies on collapse: an incident photon ejects an electron whose kinetic energy is measured, projecting the ion left behind into one specific final electronic state out of a superposition of possible ionic configurations, with probabilities given by Dyson orbital overlaps — squared coefficients in exactly the sense developed here. In quantum computing and quantum information science — increasingly relevant to chemistry through quantum simulation of molecular electronic structure — the measurement postulate is the literal readout mechanism: a qubit register in superposition is measured in a chosen basis, collapsing to a bitstring with probability given by squared amplitudes, and molecular energies computed via variational quantum eigensolvers are extracted precisely by repeated measurement and Born-rule statistics, echoing the Expectation Values formalism used throughout ordinary quantum chemistry. Understanding this postulate is also the necessary stepping stone to appreciating why position and momentum, or energy and time, cannot be simultaneously sharp — the deep content of The Uncertainty Principle, which follows once you accept that measuring one observable can collapse a state into something that is not an eigenstate of a second, non-commuting observable.

Ask the tutor: If two observables don't commute, how does collapsing onto an eigenstate of one destroy certainty in the other — and how does that lead directly to the uncertainty principle?

Before a measurement, a quantum system is described by a state vector $\Psi$ that is, in general, a superposition of eigenfunctions of the operator $\hat A$ representing whatever we intend to measure. Superposition itself is not mysterious once you accept that the Schrödinger equation is linear: any sum of solutions is itself a solution, so a general state is built from a basis of special ones. What is mysterious — genuinely, irreducibly so — is what happens when you look. The measurement postulate answers this in three parts. First, the only possible outcomes of measuring $\hat A$ are the eigenvalues $a_n$ satisfying $\hat A \phi_n = a_n \phi_n$; you can never measure a value that isn't an eigenvalue, no matter how skillfully you engineer the experiment. Second, if the system is in state $\Psi = \sum_n c_n \phi_n$, the probability of obtaining outcome $a_n$ on a single measurement is $|c_n|^2$ (assuming $\Psi$ and the $\phi_n$ are normalized). Third — and this is the part every introductory course underlines in red — immediately after a measurement yields $a_n$, the system's state is no longer $\Psi$; it has collapsed to $\phi_n$. All information about the other components $c_m \phi_m$ ($m \ne n$) is gone. This is why measurement is not a passive readout: it actively and irreversibly changes the state (except in the special case where $\Psi$ was already an eigenstate). Chemists meet this constantly in spectroscopy. Consider a molecule prepared in a coherent superposition of two vibrational levels by a short laser pulse — this is exactly how vibrational wave-packet experiments work. If you then measure the vibrational energy (say, via a probe pulse that only interacts resonantly with one specific level), you will get one of the two energies, with probability set by the population of that level in the superposition, and the wave packet — previously a moving, breathing superposition — collapses into a stationary eigenstate. The interference fringes seen in pump-probe spectroscopy are a direct signature of the coefficients $c_n$ before collapse; the discrete, sudden jump seen in single-molecule fluorescence blinking is the collapse itself.

A second essential picture concerns repeatability. Once collapse has occurred and the state is $\phi_n$, a second, immediate measurement of the same observable $\hat A$ must return $a_n$ again — with probability 1, since $\phi_n$ is already an eigenstate with $|c_n|^2=1$ and all other coefficients zero. This is what makes measurement "objective" in the limited sense that consecutive identical measurements agree; it is also what allows spectroscopists to talk about a molecule being definitively "in the ground state" after de-excitation, rather than perpetually smeared out. The third picture worth internalizing is what happens with a continuous spectrum, such as position $\hat x$. Here there is no discrete sum; instead $\Psi(x)$ itself already is the expansion coefficient function in the position eigenbasis, and $|\Psi(x)|^2\,dx$ is the probability of finding the particle in $[x, x+dx]$. Collapse in this case means the wavefunction becomes sharply peaked at the measured position (idealized as a Dirac delta in the limit of perfect resolution), a picture directly relevant to how STM tips or ionization experiments localize an electron.

The postulate rests on the expansion theorem: because $\hat A$ is Hermitian, its eigenfunctions $\{\phi_n\}$ form a complete, orthonormal set, so any physically admissible state can be written

where: — $\Psi$ is the (normalized) state of the system just before measurement, — $\phi_n$ are eigenfunctions of $\hat A$ with $\hat A \phi_n = a_n \phi_n$, — $c_n$ are complex expansion coefficients obtained by projecting $\Psi$ onto each $\phi_n$.

Reading: this is just Fourier-style decomposition in the eigenbasis of whatever you're about to measure; it says nothing yet about outcomes — it is bookkeeping that makes the postulate expressible. The Born rule for probability is

where: — $P(a_n)$ is the probability that a single measurement of $\hat A$ yields eigenvalue $a_n$, — the eigenfunctions are taken orthonormal so that $\langle \phi_m|\phi_n\rangle = \delta_{mn}$.

Reading: the probability is not $c_n$ itself (which can be complex or negative-real) but its squared modulus — this is the empirical content of quantum mechanics, verified in every diffraction pattern and every spectral line intensity ever measured. Normalization of $\Psi$ guarantees the probabilities sum to one:

where: — the sum runs over the complete set of eigenstates, — this is simply Parseval's theorem applied to quantum states.

Reading: if your calculated $|c_n|^2$ don't sum to 1, either $\Psi$ wasn't normalized or your basis wasn't complete — a built-in consistency check chemists use routinely when computing state populations. The collapse (projection) postulate states that immediately after a measurement returns $a_n$ (non-degenerate case),

where: — the arrow denotes an instantaneous, discontinuous update of the state, not a unitary time evolution, — all coefficients $c_m$ for $m\neq n$ are erased; the system's subsequent evolution starts fresh from $\phi_n$.

Reading: collapse is not "the wavefunction get closer to $\phi_n$" — it is total and immediate, and this is precisely what makes quantum measurement irreversible and non-unitary, unlike the Schrödinger equation's own smooth evolution. For a degenerate eigenvalue $a_n$ shared by $g$ orthonormal eigenfunctions $\phi_{n,1},\ldots,\phi_{n,g}$, the rule generalizes:

where: — $\hat P_n = \sum_k |\phi_{n,k}\rangle\langle\phi_{n,k}|$ is the projector onto the degenerate eigenspace, — the denominator renormalizes the projected state to unit norm.

Reading: you sum probabilities over all states sharing that eigenvalue (they're experimentally indistinguishable outcomes), and the collapsed state is whatever combination of those degenerate eigenfunctions was already present in $\Psi$ — not an arbitrary single one of them. For an observable with a continuous spectrum such as position,

where: — $|\Psi(x)|^2$ plays the role of $|c_n|^2$ per unit length — a probability density, not a probability, — the total probability over all space integrates to 1 for a normalized $\Psi$.

Reading: you cannot ask for the probability of an exact position (it's zero, since it's a density), only for a probability within some finite interval — the resolution the measuring apparatus can actually achieve. Finally, the expectation value formula from Expectation Values is recovered exactly by averaging the Born-rule outcomes:

where: — $\langle A\rangle$ is the mean value obtained from many repetitions of the measurement on identically prepared systems, — it equals the direct operator sandwich, showing the two formulations (statistical average over eigenvalues, versus operator expectation value) are the same object viewed two ways.

Reading: this equation is the bridge between the measurement postulate (discrete outcomes, probabilistic) and the more familiar expectation-value machinery — one is the microscopic mechanism, the other the macroscopic average it produces.

Example 1 — Foundational: energy measurement in an asymmetric superposition, particle in a box. Take a particle of mass $m$ in a one-dimensional box of length $L$, with normalized stationary states

Suppose at $t=0$ the (normalized) state is the specific superposition

chosen with deliberately unequal, non-round coefficients so no symmetry hides an error. Check normalization first:

so $\Psi$ is properly normalized, consistent with $\sum_n |c_n|^2 = 1$. A measurement of energy is now performed. By the Born rule, the possible outcomes and probabilities are

Taking $L = 2.00\ \text{Å} = 2.00\times10^{-10}\,\text{m}$ and an electron mass $m = 9.109\times10^{-31}\,\text{kg}$, the base energy unit is

So $E_1 = 1.51\times10^{-18}$ J, $E_2 = 4\times1.507\times10^{-18} = 6.03\times10^{-18}$ J, $E_3 = 9\times1.507\times10^{-18} = 1.356\times10^{-17}$ J. If the measurement returns $E_2$ (the most likely outcome, at 55%), the state instantly collapses:

A repeated, immediate energy measurement on this collapsed state now returns $E_2$ with probability 1 — no trace of the original $\phi_1$ or $\phi_3$ components remains. The expectation value before the measurement, using $\langle E\rangle = \sum_n P(E_n) E_n$, is

matching what $\langle \Psi|\hat H|\Psi\rangle$ would give directly — a good consistency check, and one an examiner would expect you to run.

Limiting case check: if instead $c_1 = 1$ and $c_2 = c_3 = 0$ (the system is already in an eigenstate), the Born rule gives $P(E_1) = 1$ and collapse is trivial — measurement changes nothing, as it must, since there is nothing to "choose" between.

Example 2 — Applied: spin measurement in NMR-relevant two-level system. Consider a spin-$\tfrac12$ nucleus (like $^1$H) in a magnetic field along $z$, with eigenstates $|\alpha\rangle$ (spin up, eigenvalue $+\hbar/2$) and $|\beta\rangle$ (spin down, eigenvalue $-\hbar/2$) of the operator $\hat S_z$. A radiofrequency pulse tips the spin into the superposition

with an arbitrary phase $\theta$ (phases never affect measurement probabilities of the operator whose eigenbasis you're using, since $|c_n|^2$ discards phase). Take $\theta = 37°$ purely to emphasize that phase is present but irrelevant here. A measurement of $\hat S_z$ (equivalent, physically, to detecting which energy level — spin-up or spin-down in the field — the nucleus occupies) gives outcomes

These sum to 1.00, confirming normalization. Suppose the measurement yields $-\hbar/2$ (the 65% outcome). The state collapses:

and the phase factor $e^{i37°}$ present before the measurement has vanished entirely from the physical predictions (up to the unobservable global phase noted in the derivation above). Now compute the expectation value of $\hat S_z$ in the original superposition, to connect this to a bulk NMR signal, which measures a macroscopic ensemble average rather than a single spin:

This negative average (net excess of spin-down population, 65% vs 35%) is what a bulk magnetization measurement reports — it is the ensemble average over many single-spin collapses, each individual event being all-or-nothing ($+\hbar/2$ or $-\hbar/2$, never $-0.15\hbar$ itself). This is the crucial conceptual point for chemists using NMR: no single nuclear spin is ever observed with magnetization $-0.15\hbar$; only the population-weighted statistics of many identical, independently collapsing spins produce that number.

Limiting case check: if the pulse fully inverts the population so $c_\alpha = 0, c_\beta = 1$, then $P(-\hbar/2)=1$ and $\langle S_z\rangle = -\hbar/2$ exactly — full population transfer, the basis of population-inversion experiments, recovered correctly from the general formula by setting the appropriate coefficient to zero.

Example 3 — Exam/competition level: degenerate eigenvalue and position measurement in a hydrogenic superposition. A hydrogen atom electron is prepared (by an appropriately shaped ultrafast pulse) in the normalized superposition of degenerate $n=2$ states,

where all four states share the same energy eigenvalue $E_2 = -\dfrac{13.6\,\text{eV}}{4} = -3.40$ eV (the $n=2$ degeneracy, with $g=4$ once spin is ignored for this illustration). Check normalization:

Part (a): energy measurement. Since all four basis states are degenerate eigenstates of $\hat H$ with the same eigenvalue $E_2$, the probability of measuring $E_2$ is, by the degenerate Born rule,

This must be 1 regardless of the coefficients (any normalized combination of degenerate eigenstates is still an eigenstate of $\hat H$ with that eigenvalue) — a useful sanity check: measuring energy here is not informative about which orbital the electron is in, only that $n=2$. After this measurement, the state is completely unchanged (up to an overall phase), since $\hat P_{E_2}\Psi = \Psi$ already.

Part (b): angular momentum measurement. Now measure $\hat L_z$ instead. The states $\psi_{200}$ and $\psi_{210}$ both have $m_\ell = 0$ (eigenvalue $L_z = 0$); $\psi_{211}$ has $m_\ell=+1$ ($L_z = +\hbar$); $\psi_{21,-1}$ has $m_\ell=-1$ ($L_z=-\hbar$). Because $\hat L_z$ has a two-fold degenerate eigenvalue at $L_z=0$ (shared by $\psi_{200}$ and $\psi_{210}$), the Born rule for that eigenvalue sums both contributing probabilities:

These sum to $0.55+0.30+0.15 = 1.00$, as required. If the measurement returns $L_z = 0$, the state collapses — using the projector formula from the derivation above — onto the renormalized combination of exactly the degenerate states sharing that eigenvalue, preserving their original relative weight:

Numerically, $\sqrt{0.10/0.55} = \sqrt{0.1818} = 0.4264$ and $\sqrt{0.45/0.55} = \sqrt{0.8182} = 0.9046$, and indeed $0.4264^2 + 0.9046^2 = 0.1818+0.8182=1.000$, confirming the renormalized state is properly normalized — this is exactly the subtlety students most often get wrong: collapse onto a degenerate eigenvalue does not pick a single basis function arbitrarily, it projects and renormalizes onto whatever superposition of the degenerate states was already present.

Limiting case check: if only $\psi_{210}$ had nonzero weight among the $m_\ell=0$ states (i.e., $c_{200}\to 0$), the collapsed state after an $L_z=0$ result reduces cleanly to $\psi_{210}$ alone, matching the simple non-degenerate rule.

The measurement postulate as a distinct, sharply stated axiom owes its clearest formulation to John von Neumann, though its physical content was argued into existence collectively by the Copenhagen physicists — Bohr, Heisenberg, Born, and Pauli — through the late 1920s.