A tilted analyser
★☆☆Atoms prepared in \(\ket{+_z}\) enter a device oriented at 60° from the \(z\) axis, in the \(xz\) plane. What fraction leaves through the \(+\) output?
Chapter 2 · 6 h lectures + 3 h tutorials
A beam of silver atoms, a magnet, and two spots instead of a band. To describe this result, every measurable quantity becomes an operator, and every possible result an eigenvalue.
In nineteen twenty-two, Stern and Gerlach sent silver atoms through a non-uniform magnetic field. Instead of one continuous smear, they found two spots. The magnetic moment is quantised. Only after nineteen twenty-five did physicists understand that this is spin.
Place two devices in a row, one along z and the other along x: each output receives half of the atoms. Add a third device along z, and you find half and half again. The measurement along x has erased the information about z.
In quantum mechanics, every measurable quantity is represented by a Hermitian operator. Its eigenvalues are the only possible results. Its eigenvectors are the states after the measurement. The probability of a result is the squared modulus of the projection of the state onto the corresponding eigenvector.
Two operators that do not commute describe incompatible quantities. The product of their standard deviations is bounded below by half the mean value of their commutator. This bound depends on the state. It is the Robertson inequality, and the Heisenberg relation is a special case of it.
In classical physics, to measure is to read a value that already exists. In 1922, in Frankfurt, Otto Stern and Walther Gerlach sent a beam of silver atoms between the poles of a magnet whose field is not uniform. Each atom carries a small magnetic moment. If these moments could point in every direction, a continuous band would appear on the plate. Two spots are observed.
The silver deposit was too thin to be seen. According to Stern’s account, it appeared when he leaned over the plate while smoking a cheap cigar: the sulphur in the smoke blackened the silver into silver sulphide. Gerlach sent Bohr a postcard with the photograph of the two spots. Both saw in it the “space quantisation” of the old Bohr–Sommerfeld theory. The electron spin was only proposed in 1925, by Uhlenbeck and Goudsmit, and it is the spin that explains the two silver spots.
A measurement can only give certain well-defined values as its result, which we shall call the eigenvalues of the associated operator. Just after the measurement, the system is in the state associated with the result obtained: this is the reduction (or projection) of the state.
Spin or magnetic moment? The lecture notes write “\(\mu_\pm = \pm\hbar/2\)”. What equals \(\pm\hbar/2\) is the projection of the spin \(S_z\), a quantity expressed in J·s. The associated magnetic moment is about \(\mp\mu_B\) (the Bohr magneton, in J/T): for an electron it has the opposite sign.
Discrete, not always. The spin eigenvalues are discrete, but those of position or momentum form a continuum (chapter 5). The principle concerns eigenvalues, not whether they are discrete.
The oven produces a mixture. The atoms leaving the oven are not all in one and the same superposition \(\alpha_+\ket{+} + \alpha_-\ket{-}\). A pure state of this kind points in a definite direction on the Bloch sphere, and would give 100% in a device oriented along that direction. The oven beam gives 50/50 whatever the orientation: it is a statistical mixture, the distinction seen in chapter 1.
A Stern–Gerlach device with one output blocked becomes a filter: it prepares atoms in a known state. We write \(\mathrm{SG}_\theta\) for a device oriented at angle \(\theta\) from the \(z\) axis, in the \(xz\) plane. For an atom prepared in \(\ket{+}_{\theta_1}\), the probability of leaving through the \(+\) output of a device \(\mathrm{SG}_{\theta_2}\) is
Before running the lab: write down your prediction, run it, compare, then explain the gap in one sentence. If the second analyser is oriented along \(x\), what fraction of \(+\hbar/2\) results do you expect at the output of the third?
frequency at the last devicepredicted probability
The dials show the measured axis in the \(xz\) plane (\(z\) upwards, \(x\) to the right). Try Z → X → Z: the \(-_z\) output, although removed by SG₁, reappears in half of the cases. Then remove SG₂ (2 devices, Z → Z): 100% in \(+\). The intermediate measurement along \(x\) has erased the information about \(z\). The oven sends a mixture, so SG₁ lets half of the atoms through whatever its orientation.
| Sequence | Result at the last device | What it shows |
|---|---|---|
| \(\mathrm{SG}_z \to \mathrm{SG}_z\) | 100% in \(+_z\) | A repeated measurement gives the same result again |
| \(\mathrm{SG}_z \to \mathrm{SG}_x\) | 50% / 50% | \(\ket{+_z}\) is not an eigenstate of \(S_x\) |
| \(\mathrm{SG}_z \to \mathrm{SG}_x \to \mathrm{SG}_z\) | 50% / 50% | \(S_x\) and \(S_z\) are incompatible |
| \(\mathrm{SG}_z \to \mathrm{SG}_x \to \mathrm{SG}_\theta\) | \(\tfrac12(1 + \sin\theta)\) in \(+_\theta\) | Born rule with an arbitrary axis |
Atoms prepared in \(\ket{+_z}\) enter a device oriented at 60° from the \(z\) axis, in the \(xz\) plane. What fraction leaves through the \(+\) output?
Set-up Z → X → Z. The \(+\) output is kept after SG₁ and after SG₂. What fraction of the atoms leaving the oven comes out through the \(+_z\) output of SG₃?
Three successive measurements, three factors \(\tfrac12\). The first comes from the fact that the oven produces a mixture, the other two from \(|\braket{+_x|+_z}|^2 = |\braket{+_z|+_x}|^2 = \tfrac12\).
Einstein“The 50/50 can only reflect our ignorance. The atom carries hidden instructions that fix its output in advance.”
Bohr“The property ‘spin along \(z\)’ was not defined while the spin was being measured along \(x\). No memory was erased; there was no value to remember.”
In the long version of the lecture notes, Einstein speaks here of “spooky action at a distance”. That phrase concerns two entangled, distant particles (the EPR paper of 1935, chapter 4). It does not apply to successive measurements on a single atom. Bell-type experiments have since ruled out local hidden variables.
Imagined dialogue: chance in question. The incompatibility of measurements fed the debate between Einstein and Bohr. The lines summarise their positions; they are not quotations (the phrase “God does not play dice” is Einstein’s own).
Einstein: “The atom must know where it is going. This 50/50 can only be a sign of our ignorance: there are hidden instructions. God does not play dice.”
Bohr: “There is no memory to erase: the property ‘spin along \(z\)’ was simply not defined while we were asking about spin along \(x\). This is not ignorance, it is a fundamental indefiniteness.”
Einstein: “Then the theory is incomplete.”
Bohr: “It is complete. It is our language that must adapt, not nature.” Bell’s inequalities (chapter 4) will settle the question experimentally.
Every measurable quantity \(A\) (energy, spin, position…) is associated with a Hermitian linear operator \(\hat A\) acting on the state space. Over a large number of systems prepared in \(\ket\psi\), the mean of the results is
Linearity, \(\hat A(\alpha\ket{\psi_1} + \beta\ket{\psi_2}) = \alpha\hat A\ket{\psi_1} + \beta\hat A\ket{\psi_2}\), is what makes the operator compatible with the superposition principle.
The solutions of \(\hat A\ket{\varphi_i} = a_i\ket{\varphi_i}\) give the possible results \(a_i\) of the measurement and the states \(\ket{\varphi_i}\) of the system just after \(a_i\) has been obtained. The probability of obtaining \(a_i\) is \(P(a_i) = |\braket{\varphi_i|\psi}|^2\) (non-degenerate eigenvalue).
\(\hat Z = \ket0\bra0 - \ket1\bra1\) acts on \(\ket\psi = \alpha\ket0 + \beta\ket1\) to give \(\alpha\ket0 - \beta\ket1\). So \(\hat Z\ket0 = +\ket0\) and \(\hat Z\ket1 = -\ket1\): eigenvalues \(\pm1\), eigenvectors \(\ket0\) and \(\ket1\). In quantum computing, this is the “phase-flip” gate. In an orthonormal basis \(\{\ket i\}\), the operator is described by the matrix \(A_{ij} = \bra i\hat A\ket j\). For the NOT gate, \(\hat X\ket0 = \ket1\) and \(\hat X\ket1 = \ket0\) give \(X_{01} = X_{10} = 1\), \(X_{00} = X_{11} = 0\), that is \(\hat X = \ket0\bra1 + \ket1\bra0\).
For an axis \(\vec n = (\sin\theta\cos\varphi, \sin\theta\sin\varphi, \cos\theta)\), the operator \(\hat\sigma_n = \vec n\cdot\vec{\hat\sigma} = n_x\hat\sigma_x + n_y\hat\sigma_y + n_z\hat\sigma_z\) is
Its eigenvalues are \(\pm1\) for every axis. For a state with Bloch vector \(\vec r\), one can show that \(\langle\hat\sigma_n\rangle = \vec r\cdot\vec n\), hence \(P(+1) = \tfrac12(1 + \vec r\cdot\vec n)\).
The results are always \(+1\) or \(-1\), never an intermediate value: the mean \(\vec r\cdot\vec n\) varies continuously, the results do not. With “Prepare \(\ket\psi = \ket{n_+}\)”, the standard deviation drops to zero: an eigenstate gives a certain result.
A quantity is represented by \(A = \begin{pmatrix}2 & 1-i\\ 1+i & 3\end{pmatrix}\). What is the largest possible measurement result?
\((2-\lambda)(3-\lambda) - |1+i|^2 = \lambda^2 - 5\lambda + 6 - 2 = (\lambda-1)(\lambda-4)\). Check: the trace is \(5 = 1 + 4\) and the determinant \(4 = 1\times4\).
Adjoint: \((A^\dagger)_{ij} = A_{ji}^*\), the conjugate transpose. We have \((AB)^\dagger = B^\dagger A^\dagger\) and \((\lambda A)^\dagger = \lambda^* A^\dagger\). For an expression in Dirac notation: reverse the order, swap kets and bras, conjugate the numbers. Example: \((2\ket0\bra1 - i\ket1\bra0)^\dagger = 2\ket1\bra0 + i\ket0\bra1\).
Hermitian: \(A^\dagger = A\). The diagonal elements are real, and elements placed symmetrically about the diagonal are complex conjugates: \(\begin{pmatrix}a & c\\ c^* & b\end{pmatrix}\), \(a, b\in\mathbb R\).
Unitary: \(U^\dagger U = UU^\dagger = \mathbb 1\). It preserves the norm; the evolution of an isolated system is unitary (chapter 3).
A Hermitian operator has real eigenvalues, and its eigenvectors belonging to distinct eigenvalues are orthogonal. It can then be written as a sum of projectors; this is the spectral decomposition:
The trace \(\mathrm{Tr}\,A = \sum_i A_{ii} = \sum_i a_i\) does not depend on the basis, and \(\mathrm{Tr}(AB) = \mathrm{Tr}(BA)\).
Hermitian and unitary (\(\sigma_j^2 = \mathbb 1\)), traceless, with determinant \(-1\) and eigenvalues \(\pm1\). Commutators: \([\sigma_x, \sigma_y] = 2i\sigma_z\) and cyclic permutations. Anticommutators: \(\{\sigma_j, \sigma_k\} = 2\delta_{jk}\mathbb 1\). The spin follows from \(\hat S_j = \frac\hbar2\sigma_j\), hence \([\hat S_x, \hat S_y] = i\hbar\hat S_z\).
\(\varepsilon_{jkl}\) equals \(+1\) for \((x,y,z)\) and its cyclic permutations, \(-1\) for odd permutations (one exchange of two indices), and \(0\) if two indices are equal.
Which of these matrices can represent a physical quantity?
Compute \([\sigma_z, \sigma_x] = \sigma_z\sigma_x - \sigma_x\sigma_z\).
Let \(\ket\psi = \frac{1}{\sqrt3}\ket0 + i\sqrt{\frac23}\ket1\) and \(\hat P_+ = \ket+\bra+ = \frac12\begin{pmatrix}1&1\\1&1\end{pmatrix}\). Compute \(\bra\psi\hat P_+\ket\psi\), the probability of obtaining \(+1\) when measuring \(\sigma_x\).
Check that \(\hat P_+^2 = \hat P_+\) and that its eigenvalues are \(0\) and \(1\). Here \(\vec r = (0,\ 2\sqrt2/3,\ -1/3)\): the state lies in the \(yz\) plane, hence \(\langle\sigma_x\rangle = 0\). Place it in lab 2.2 (θ ≈ 109.5°, φ = 90°) with n = x.
\([\hat A, \hat B] = \hat A\hat B - \hat B\hat A\). Two observables that commute have a common basis of eigenvectors: there are states in which both results are certain, and measuring \(A\), then \(B\), then \(A\) gives the first result again. They are said to be compatible.
1. Write the state \(\ket\psi\). 2. Identify the operator \(\hat A\). 3. Find its eigenvalues \(a_i\) and eigenvectors \(\ket{\varphi_i}\). 4. Project: \(c_i = \braket{\varphi_i|\psi}\). 5. \(P(a_i) = |c_i|^2\), and check that \(\sum_i P(a_i) = 1\).
Only one of these statements is true. Which one?
The standard deviation of a measurement in the state \(\ket\psi\) is \(\Delta A = \sqrt{\langle A^2\rangle - \langle A\rangle^2}\). For any two observables (Robertson, 1929, generalising Heisenberg, 1927):
For the qubit, \(\sigma_j^2 = \mathbb 1\) gives \(\Delta\sigma_j = \sqrt{1 - r_j^2}\), and the inequality for \(\sigma_x, \sigma_y\) becomes \(\Delta\sigma_x\,\Delta\sigma_y \ge |\langle\sigma_z\rangle| = |r_z|\). For position and momentum, \([\hat x, \hat p] = i\hbar\) gives \(\Delta x\,\Delta p \ge \hbar/2\) (chapter 5).
Δσx·Δσy as a function of θbound |cos θ|
The black curve never enters the forbidden region. It touches the bound when φ = 0°, 90°, 180° or 270°: the inequality is then an equality. On the equator (θ = 90°), the bound is zero, and for φ = 0 we even get \(\Delta\sigma_x = 0\): the state \(\ket+\) is an eigenstate of \(\sigma_x\).
The lecture notes state that, if \([\hat A, \hat B] \neq 0\), “it is impossible to prepare a state in which both spreads are zero”. The inequality does not say this: its bound depends on the state and can vanish. For angular momentum, \([\hat L_x, \hat L_y] = i\hbar\hat L_z \neq 0\), and yet the state \(\ell = 0\) gives \(\Delta L_x = \Delta L_y = 0\). Two operators that do not commute can share a few eigenvectors, but not a complete basis. For \(\sigma_x\) and \(\sigma_y\), they share none: the sum \(\Delta\sigma_x^2 + \Delta\sigma_y^2 = 1 + r_z^2\) is always at least 1.
Another subtlety: the Robertson inequality concerns standard deviations measured on copies of the same state. It does not describe the “disturbance” of one measurement by the next.
For \(\ket\psi = \frac{\sqrt3}{2}\ket0 + \frac12\ket1\), compute the standard deviation \(\Delta\sigma_z\).
This state is at θ = 60°, φ = 0. Check in lab 2.3: \(\Delta\sigma_x = 1/2\), \(\Delta\sigma_y = 1\), product \(1/2 = |\langle\sigma_z\rangle|\). The bound is reached.
Alice sends qubits prepared at random in the Z or X basis; Bob measures each one in a randomly chosen basis. They then compare their bases publicly and keep only the bits where the bases agree. An eavesdropper who measures in the wrong basis projects the state: if Bob then measures in Alice’s basis, he gets the wrong bit half of the time. On the bits that are intercepted and kept, this gives an expected error rate of 25%.
| \([A,B]=0\) | \([A,B]\neq0\) |
|---|---|
| Compatible quantities | Incompatible quantities |
| Measurable together with arbitrary precision | No precise simultaneous measurement |
| Common eigenbasis | No common eigenbasis |
| Measuring \(A\) does not disturb \(B\) | Measuring \(A\) erases information about \(B\) |
| \([S_z,S_z]=0\) | \([S_z,S_x]=i\hbar S_y\) |
Atoms in \(\ket{+_z}\) (already filtered by SG₁) pass through SG₂ oriented at \(\theta = 60°\) (the \(+_\theta\) output is kept), then through SG₃ oriented along \(x\). What fraction of these atoms leaves through the \(+_x\) output of SG₃?
\(\ket{+_z} = \cos\frac\theta2\ket{+_\theta} + \sin\frac\theta2\ket{-_\theta}\), so \(P(+_\theta) = \cos^2\frac\theta2\). Then \(\ket{+_\theta}\) enters \(\mathrm{SG}_x\): \(P(+_x) = \cos^2\frac{90° - \theta}{2} = \frac12(1 + \sin\theta)\). In total, \(\cos^2\frac\theta2\cdot\frac12(1 + \sin\theta)\).
Alice sends (basis, bit): (Z,0) (X,1) (X,0) (Z,1) (Z,1) (X,0) (Z,0) (X,1) (Z,1) (X,1) (Z,0) (X,0). Bob measures in: Z X Z X Z Z X X Z X X Z. Eve intercepts qubits 3, 6 and 11 and measures them in the Z basis. What error rate (QBER, between 0 and 1) do Alice and Bob observe on their sifted key?
The bases agree for qubits 1, 2, 5, 8, 9 and 10. Sifted key: 0 1 1 1 1 1. Qubits 3, 6 and 11 are discarded during sifting, so Eve introduces no visible error: QBER = 0.
Had Eve intercepted qubit 2 (Alice in X, Bob in X), her measurement in Z would have given Bob a wrong bit with probability ½. This is why Alice and Bob sacrifice part of the sifted key to estimate the QBER: an eavesdropper who intercepts everything produces 25% errors on average.
In the statement in the lecture notes (exercise 2.5.4), the three qubits intercepted by Eve are precisely those where Bob did not choose Alice’s basis. The question “what QBER gives Eve away?” therefore has the answer 0: the example does not show detection. To illustrate it, Eve must intercept at least one kept qubit, for example qubit 2 or 8.
In the experiment of Stray et al. (Nature 2022), the 2 m tunnel produces an expected phase shift of 17.5 mrad in the atom interferometer, detected with a signal-to-noise ratio of 8. Take a phase uncertainty \(\delta\varphi = 17.5/8\) mrad. For \(N\) independent atoms the standard quantum limit is \(\delta\varphi = 1/\sqrt N\). How many atoms are needed, in this simplified single-measurement model?
\(\delta\varphi = 2.19\times10^{-3}\) rad, so \(N = 1/\delta\varphi^2 \approx 2.1\times10^{5}\). With correlated atoms (squeezed or entangled states) and suitable measurements, the Heisenberg limit \(\delta\varphi = 1/N\) would in principle need only \(N = 1/\delta\varphi \approx 460\): like \(\Delta x\,\Delta p \ge \hbar/2\), it is a bound that quantum mechanics sets on precision, reachable only with careful state preparation.
Order of magnitude only (the match with a single measurement is fictitious): the real instrument averages many cycles and its noise also includes vibrations and lasers. The standard limit remains the floor that state preparation has to beat.
Analogy: “the future is a set of possibilities; the diviner’s consultation reveals a single path, as measuring \(S_x\) on an atom prepared in state \(\ket{+z}\) reveals one value”. Which assessment is correct?
The state is \(\ket{+x}\) and we measure \(S_z\). What does “\(\Delta S_z \ne 0\)” mean? Discuss in pairs before answering.
We measure \(S_z\) and find \(+\hbar/2\). With no evolution, we measure \(S_z\) again. What do we find? Discuss in pairs before answering.
Can the components \(S_x\) and \(S_z\) of the spin of one electron be measured simultaneously with arbitrary precision? Discuss in pairs before answering.
Choose one analogy from this chapter. With the person’s consent, who may be an elder, a practitioner, a craftsperson or a musician, ask: (1) how they present this notion, in what language and with what words; (2) under what conditions this knowledge is passed on; (3) what, in their view, does not translate into physics. Write down the words in the local language, return your file to the person, then present both systems, each coherent by its own criteria, without concluding that one proves the other.
Main ideas.
Formulas.
Pitfalls. Believing a measurement reveals a hidden pre-existing value; confusing quantum uncertainty (a property of the state) with instrument imprecision; taking the beam of atoms leaving the oven for a superposition: it is a statistical mixture, and only a first Stern-Gerlach prepares a pure state.
| Term | Terme français | Definition |
|---|---|---|
| Commutator | Commutateur | \([A,B]=AB-BA\); if non-zero, the observables are incompatible. |
| Pauli matrices | Matrices de Pauli | Hermitian, unitary \(2\times2\) matrices, \([\sigma_j,\sigma_k]=2i\epsilon_{jkl}\sigma_l\). |
| Observable | Observable | Measurable quantity, represented by a Hermitian operator. |
| Hermitian operator | Opérateur hermitien | Equal to its adjoint; real spectrum. |
| Uncertainty relation | Relation d’indétermination | \(\Delta A\,\Delta B\ge\tfrac12|\langle[A,B]\rangle|\). |
| Spin | Spin | Intrinsic angular momentum with no classical counterpart. |
Quantum fluctuations constrain a sensor’s precision, depending on the prepared state, the resources and the protocol (the standard quantum limit can be beaten with entangled or squeezed states). Diamond NV-centre magnetometers and atomic gravimeters are designed around those limits: quantum noise is a design input, not a curiosity.
The Chinese microsatellite Jinan-1 distributed quantum keys in real time to mobile ground stations of about 100 kg. It enabled encrypted communication between Beijing and Stellenbosch (South Africa), 12,900 km apart. The satellite acts as a trusted relay: it knows the keys, so this is not end-to-end entanglement. The security of BB84 is proven for the ideal protocol; real devices have their own weaknesses.
Li et al., Nature (19 March 2025); press release of the Chinese Academy of SciencesA quantum error-correcting code repeatedly measures mutually commuting operators, chosen to signal an error without reading the stored information. At the end of 2024, Google showed on its Willow chip that the error of a logical qubit (surface code) is divided by about 2.14 each time the code distance is increased. Not to be confused with mere error detection, where faulty runs are discarded.
Nature 638, 920 (2025)NIST’s aluminium-ion clock reaches a relative uncertainty of \(5.5\times10^{-19}\). The Al+ ion is hard to read out directly: its state is transferred to a neighbouring Mg+ ion, which is then measured. A decision to redefine the second is not expected before the 2030 General Conference on Weights and Measures.
Physical Review Letters, July 2025; BIPM