Born, then projection
★☆☆A system is in \(\ket\psi = \sqrt{1/3}\,\ket0 + \sqrt{2/3}\,\ket1\), with \(\hat A\ket0 = a_0\ket0\) and \(\hat A\ket1 = a_1\ket1\). What is the probability of obtaining \(a_1\)? What state is the system in just afterwards?
Chapter 3 · Lectures 4 h + Tutorials 4 h
Chapter 2 took a snapshot: one state, one question, one answer. Here we roll the film. Between two measurements the state evolves deterministically; at the measurement, chance takes over again.
The postulates set the rules. The state is a normalised vector. Measurable quantities are Hermitian operators. The results are their eigenvalues, with probabilities given by the Born rule. After a measurement, the state is projected. Between two measurements, it evolves according to the Schrödinger equation.
This evolution is unitary: it preserves the norm, and it can be undone. A measurement, on the other hand, cannot be undone.
An eigenstate of the Hamiltonian only picks up a global phase. Nothing observable changes, so we call it a stationary state. For a spin in a magnetic field, the Bloch vector turns around the field. This is Larmor precession.
Now shine an oscillating field at the right frequency on a two-level atom. The probability of finding it excited oscillates between zero and one. These are Rabi oscillations. Their frequency is proportional to the amplitude of the field, not to its intensity. Away from resonance, the oscillation becomes faster and less complete.
These pulses are the logic gates of quantum computers, and the working principle of atomic clocks.
Postulates are not proved. They sum up what chapters 1 and 2 established, and a century of experiments has found no situation that contradicts them. We sort them into two groups: the description of the system, then the measurement.
P1, state. The state is a normalised vector \(\ket\psi\) in a Hilbert space: \(\ket\psi = \sum_n \alpha_n\ket{\varphi_n}\), \(\alpha_n = \braket{\varphi_n|\psi}\), \(\braket{\psi|\psi} = 1\).
P2, observables. Every measurable quantity \(A\) corresponds to a Hermitian operator \(\hat A\).
P3, results. A measurement of \(A\) can only give an eigenvalue \(a_n\): \(\hat A\ket{\varphi_n} = a_n\ket{\varphi_n}\), and \(\hat A = \sum_n a_n\ket{\varphi_n}\bra{\varphi_n}\).
P4, Born rule. \(P(a_n) = |\braket{\varphi_n|\psi}|^2\). On average, \(\langle A\rangle_\psi = \sum_n P(a_n)\,a_n = \bra\psi\hat A\ket\psi\).
P5, projection. After the result \(a_n\), the state becomes
The sixth postulate, the Schrödinger equation, tells how the state changes between two measurements (section 3.2). The lab below runs through the three acts of an experiment on a qubit.
Before running the lab: write down your prediction, run it, compare, then explain the gap in one sentence. Choose a starting state and an evolution time: what probability of measuring \(\ket1\) do you predict?
Three things to check. Prepare \(\ket0\) with \(\hat H \propto \hat\sigma_z\): α picks up a phase, but the vector does not move (stationary state). After an evolution, U† brings back exactly the initial state: the evolution is reversible. After a measurement, U† no longer brings anything back: the projection has erased the information. Finally, measure twice in a row in the same basis: the second result repeats the first (P5).
A system is in \(\ket\psi = \sqrt{1/3}\,\ket0 + \sqrt{2/3}\,\ket1\), with \(\hat A\ket0 = a_0\ket0\) and \(\hat A\ket1 = a_1\ket1\). What is the probability of obtaining \(a_1\)? What state is the system in just afterwards?
“Wave-packet reduction” or “quantum jump” does not describe an object moving instantly through space. It is a rule for updating the state: after a measurement result, the system is described by the normalised projection \(P_n\ket\psi/\sqrt{p_n}\). The formalism does not say whether this update reflects a physical process or a gain of information: interpretations differ, predictions do not.
Between two measurements, the state of an isolated system evolves according to
where \(\hat H\), the Hamiltonian, is the Hermitian operator associated with the energy. The equation is first order in time: \(\ket{\psi(t_0)}\) fixes \(\ket{\psi(t)}\) for all \(t > t_0\). Chance enters only at the measurement.
If \(\hat H\) does not depend on time, the solution is written with the time-evolution operator:
Unitarity guarantees \(\braket{\psi(t)|\psi(t)} = \braket{\psi(t_0)|\psi(t_0)} = 1\): probabilities stay normalised at all times. It also guarantees reversibility, since \(\hat U^\dagger(t,t_0) = \hat U(t_0,t)\): we “rewind” the evolution by applying \(\hat U^\dagger\).
If \(\hat H\ket{\psi_n} = E_n\ket{\psi_n}\), then \(\ket{\psi_n(t)} = e^{-iE_nt/\hbar}\ket{\psi_n(0)}\). Only a global phase turns: no measurement probability changes. A superposition of states with different energies does evolve, at the rate set by the gaps \((E_n - E_m)/\hbar\).
Differentiating \(\langle A\rangle = \bra{\psi(t)}\hat A\ket{\psi(t)}\) and using the Schrödinger equation and its conjugate, we obtain, for a time-independent \(\hat A\):
If \([\hat A,\hat H] = 0\), \(\langle A\rangle\) is constant: \(A\) is a conserved quantity. With \(\hat A = \hat H\), we recover the conservation of the mean energy of an isolated system.
The lecture notes call this relation the “Heisenberg equation”. It is rather the Ehrenfest theorem: it concerns expectation values. The Heisenberg equation, \(\mathrm d\hat A_H/\mathrm dt = \tfrac{1}{i\hbar}[\hat A_H,\hat H]\), concerns the operators themselves, in the picture where they are the ones that evolve.
The summary of the lecture notes also states that “only energy is immutable”. This is too restrictive: any quantity that commutes with \(\hat H\) is conserved (momentum of a free particle, \(\hat\sigma_z\) in lab 3.2…). Likewise, “only energy eigenstates are stationary” assumes that any superposition of states with the same energy (a degenerate level) counts as an eigenstate.
Take \(\hat H = \tfrac{\hbar\omega}{2}\hat\sigma_z\) and \(\ket{\psi(0)} = \cos\tfrac\theta2\ket0 + e^{i\varphi_0}\sin\tfrac\theta2\ket1\). The two components pick up opposite phases \(e^{\mp i\omega t/2}\); up to a global phase,
θ does not change and φ grows at rate ω: the Bloch vector turns around z. It follows that \(\langle\sigma_x\rangle(t) = \sin\theta\cos(\varphi_0+\omega t)\), \(\langle\sigma_y\rangle(t) = \sin\theta\sin(\varphi_0+\omega t)\) and \(\langle\sigma_z\rangle = \cos\theta\), constant since \([\hat\sigma_z,\hat H] = 0\).
⟨σx⟩(t)⟨σy⟩(t)⟨σz⟩ (constant)
The blue arrow is the rotation vector \(\omega\,\hat z\): the Bloch vector turns anticlockwise around it. With θ = 0 or 180°, nothing moves any more: \(\ket0\) and \(\ket1\) are the stationary states. Mind the sign: for the Hamiltonian of the lecture notes, \(E_0\ket0\bra0 + E_1\ket1\bra1\) with \(E_1 > E_0\), we have \(\omega = -(E_1-E_0)/\hbar < 0\) and the rotation is clockwise. The animation time is arbitrary: a proton in a 1.5 T MRI scanner precesses at about 64 MHz.
Which of these matrices can be a time-evolution operator \(\hat U(t)\)?
At \(t = 0\), the system is in an eigenstate \(\ket{\psi_1}\) of \(\hat H\), with energy \(E_1\). The energy is measured at time \(t\). What is the probability of obtaining \(E_1\)?
With \(\hat H = \tfrac{\hbar\omega}{2}\hat\sigma_z\) and \(\ket{\psi(0)} = \ket+\), which of these expectation values stays constant?
Measuring a system often can slow down its evolution. For a short time, the probability of leaving the initial state grows as \(t^2\), not as \(t\). If the interval is cut into \(N\) pieces separated by measurements, each piece contributes \((t/N)^2\), that is \(N(t/N)^2 = t^2/N\) in total: the leakage tends to zero as \(N\) grows.
The lecture notes illustrate the effect with an atom that has “a 50% chance of decaying in 1 s”, measured every microsecond. Two caveats. First, a decay follows an exponential law: the \(t^2\) regime exists only at the very first instants, far shorter than a microsecond, and measurements every microsecond do not change the half-life. Second, the state \(\tfrac{1}{\sqrt2}(\ket{\text{intact}} + \ket{\text{decayed}})\) is a simplification: the decay product becomes entangled with the emitted particle. The reference experiment (Itano et al., NIST, 1990) used a driven Rabi transition in trapped ions, exactly the situation of exercise 3.5.
Two levels \(\ket A, \ket B\), with \(\bra A\hat H\ket A = \bra B\hat H\ket B = 0\) and \(\bra B\hat H\ket A = V\) real. We start from \(\ket A\) and use \(\hat U(\delta t) \approx \mathbb 1 - \tfrac{i}{\hbar}\hat H\delta t\). \(P_1\) is the probability of finding \(\ket B\) after \(\delta t\) without measurement; \(P_2\) is the same probability if we measure in \(\{\ket A, \ket B\}\) at \(\delta t/2\). What is \(P_2/P_1\) to the lowest non-zero order?
\(\hat U(\delta t)\ket A \approx \ket A - \tfrac{iV\delta t}{\hbar}\ket B\), so \(P_1 = (V\delta t/\hbar)^2\). With the measurement: at \(\delta t/2\), we find \(\ket B\) with probability \(p = (V\delta t/2\hbar)^2\) and stay there to first order; otherwise we restart from \(\ket A\) and gain another \(p\). Total: \(P_2 \approx 2p = P_1/2\). With \(N\) measurements, \(P_N \approx P_1/N\).
A qubit has two levels, \(\ket0\) (ground) and \(\ket1\) (excited), separated by \(\hbar\omega_0\): \(\hat H_0 = E_0\ket0\bra0 + E_1\ket1\bra1\). An oscillating field of angular frequency \(\omega\) close to \(\omega_0\) (microwave, laser, radio frequency) adds a coupling \(\hat V(t)\) between the two levels. Moving to the frame that rotates with the field, and neglecting the terms that oscillate at \(\omega + \omega_0\), we obtain a time-independent Hamiltonian:
\(\Omega\) is the Rabi angular frequency (set by the field amplitude), \(\Delta\) the detuning \(\pm(\omega-\omega_0)\), whose sign depends on the convention. As for Larmor, the Bloch vector turns, this time around the axis \((\Omega, 0, \Delta)\), at angular frequency \(\Omega'\).
Starting from \(\ket0\) at \(t = 0\),
At resonance (\(\Delta = 0\)), the transfer is complete. A π pulse (\(t = \pi/\Omega\)) sends \(\ket0\) to \(-i\ket1\): it is an X gate up to a global phase. A π/2 pulse (\(t = \pi/2\Omega\)) gives \(\tfrac{1}{\sqrt2}(\ket0 - i\ket1) = \ket{-i}\).
predicted \(P_1(t)\)frequency over 200 measurements
The blue arrow is the axis \((\Omega, 0, \Delta)\). At resonance, it is horizontal and the vector passes through the south pole. Add some detuning: the axis tilts upwards, the circle traced out shrinks and \(P_1\) levels off at \(\Omega^2/\Omega'^2\), but the oscillation speeds up. The pulse buttons use the durations calculated at resonance; off resonance, the π pulse no longer flips the qubit. A real set-up never sees the curve: it rebuilds it point by point by measuring many qubits, like the “200 qubits” button.
The Rabi frequency is proportional to the amplitude of the field (\(\Omega = \gamma B_1\) for a spin, \(\Omega = dE_0/\hbar\) for an electric dipole), not to its intensity, which varies as the square of the amplitude.
The lecture notes list errors in pulse duration and frequency among the sources of “decoherence”. These are coherent errors: a rotation that lasts slightly too long is still a unitary evolution, reversible and correctable by calibration. Decoherence refers to the loss of information to the environment, which no pulse can undo.
For \(\tilde H = \tfrac\hbar2(\Omega\hat\sigma_x + \Delta\hat\sigma_z)\) with \(\Omega/2\pi = 3\) MHz and \(\Delta/2\pi = 4\) MHz, compute the gap \((E_+ - E_-)/h\) in MHz.
A superconducting qubit is driven at resonance with \(\Omega/2\pi = 10\) MHz. How long must the pulse last to create the superposition \(\tfrac{1}{\sqrt2}(\ket0 - i\ket1)\) from \(\ket0\)? (in ns)
The qubit is driven with a detuning \(\Delta = 3\Omega\). What is the maximum probability of finding the qubit in \(\ket1\)?
An atomic clock locks an oscillator to the frequency \(\omega_0\) of a transition. It needs a resonance that is as narrow as possible. With a single π pulse, the width is of order \(\Omega\). The Ramsey method separates two π/2 pulses by a free precession of duration \(T\): the accumulated phase \(\Delta T\) gives \(P_1 \approx \cos^2(\Delta T/2)\), fringes of width \(1/(2T)\) in frequency: the longer \(T\), the finer the fringes.
one π pulseπ/2 — T — π/2
Both curves are computed exactly, detuning included during the pulses. Lengthen T: the central Ramsey fringe narrows as \(1/(2T)\) without touching the pulse power. It stays slightly narrower than \(1/(2T)\), because the phase \(\Delta t\) also accumulates during the pulses. This is the fringe that caesium fountains track. The widths shown are full widths at half maximum.
The lecture notes state that “the second is defined as 9,192,631,770 Rabi oscillations of caesium”. This is incorrect. Since 1967, the second has been the duration of 9,192,631,770 periods of the radiation corresponding to the hyperfine transition of the ground state of caesium 133. The Rabi frequency, by contrast, depends on the amplitude of the applied field: it cannot serve as a standard. Rabi oscillations (and above all Ramsey fringes) serve only to locate the resonance.
Exercise 3.3.3 of the lecture notes also gives \(\Delta E = 4.135\,667\times10^{-6}\) eV, which leads to 1 GHz and not 9.19 GHz (see exercise 3.9). The correct value is \(\Delta E \approx 3.80\times10^{-5}\) eV.
For MRI (exercise 3.3.2 of the lecture notes), note that at 1.5 T and body temperature, the excess of “aligned” protons is only a few parts per million. The signal comes from the precession of the magnetisation of this small excess, detected by induction in a coil; it is not a projective measurement of individual spins.
The lecture notes give \(\Delta E = 4.135\,667\times10^{-6}\) eV for the caesium clock transition. Compute \(f_0 = \Delta E/h\) in GHz. Is it compatible with the definition of the second (9,192,631,770 Hz)?
A spin in a static field \(B_0\) evolves under \(\hat H = -\tfrac{\hbar\omega_0}{2}\hat\sigma_z\), with \(\omega_0/2\pi = 2\) MHz. For how long must the field act to realise, up to a global phase, the gate \(Z : \ket k \mapsto (-1)^k\ket k\)? (in ns, shortest non-zero duration)
The spins start in \(\ket0\). We apply \(R_y(\pi/2) = \tfrac{1}{\sqrt2}(\mathbb 1 - i\hat Y)\), with \(\hat Y = \begin{pmatrix}0 & -i\\ i & 0\end{pmatrix}\). (a) Compute the final state and place it on the Bloch sphere. (b) What are the probabilities of obtaining \(\pm\hbar/2\) for \(S_z\)? (c) Same question for \(S_x\), whose eigenstates are \(\ket\pm\).
\(-i\hat Y = \begin{pmatrix}0 & -1\\ 1 & 0\end{pmatrix}\) is real: no complex number will appear.
(a) \(R_y(\pi/2)\ket0 = \tfrac{1}{\sqrt2}\binom{1}{1} = \ket+\), on the equator, along +x. (b) 50% and 50%. (c) \(+\hbar/2\) with certainty, since the state is \(\ket+\) itself. Measurements in Z cannot tell \(\ket+\) from a 50/50 mixture; a measurement in X can. In lab 3.1, prepare \(\ket0\), choose \(\hat\sigma_x\) and 90°: you get \(\ket{-i}\), because a rotation about x and a rotation about y do not lead to the same point on the equator.
Analogy: “a distant drum vibrates when you strike yours at the right frequency, as an atom changes level under a field at the resonance frequency”. Which assessment is correct?
A qubit with Hamiltonian \(\hat H = \hbar\omega\sigma_z/2\) (\(\omega\ne0\)) is in the state \((\ket0+\ket1)/\sqrt2\). What can be said? Discuss in pairs before answering.
On resonance, the field is applied for a duration such that \(\Omega t = \pi\), starting from \(\ket0\). What is the final state (up to a phase)? Discuss in pairs before answering.
At exact resonance the transition probability is \(P_{0\to1}(t)=\sin^2(\Omega_R t/2)\). Which pulse duration flips \(\ket0\) into \(\ket1\) with certainty (an \(X\) gate)? 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. Confusing Ehrenfest’s theorem (on the mean value) with the Heisenberg equation (on the operator); thinking evolution between measurements is random; confusing a unitary pulse-calibration error with decoherence (\(T_1\), \(T_2\)), which is irreversible; defining the second through Rabi oscillations (it is defined by the caesium-133 hyperfine transition, probed with Ramsey fringes); believing MRI measures a single spin (it detects a macroscopic magnetisation, polarised to a few ppm).
| Term | Terme français | Definition |
|---|---|---|
| Schrödinger equation | Équation de Schrödinger | \(i\hbar\,\mathrm d_t\ket\psi=H\ket\psi\), continuous unitary evolution. |
| Hamiltonian | Hamiltonien | Total-energy observable, generator of time evolution. |
| Rabi oscillations | Oscillations de Rabi | Coherent oscillation between two levels at angular frequency \(\Omega=\sqrt{\Omega_R^2+\delta^2}\). |
| Postulates of QM | Postulats quantiques | Six axioms: Hilbert space, observables, spectrum, Born, projection, Schrödinger. |
| Born rule | Règle de Born | \(P(a_n)=|\braket{\varphi_n|\psi}|^2\) for a non-degenerate spectrum. |
| Wave-function collapse | Réduction du paquet d’ondes | Projection \(\ket\psi\to P_n\ket\psi/\sqrt{P(a_n)}\) after a measurement. |
The same principles (Rabi transitions, Ramsey interrogation) serve atomic clocks (satellite positioning, network timing) and quantum-computer gates. MRI relies on the Larmor precession of the proton spin.
In October 2025, Google published in Nature a measurement of out-of-time-order correlators on 103 qubits of its Willow chip. The protocol evolves the system “forwards” with \(\hat U\), perturbs one qubit, then makes it “come back” with \(\hat U^\dagger\): exactly the “Rewind” button of lab 3.1, on a much larger machine. Google reports a computation 13,000 times faster than the best known classical algorithm. This factor is a claim relative to current algorithms: it may shrink if better classical methods appear.
Google Research, 22 October 2025 — research.google/blog/a-verifiable-quantum-advantageNIST’s single aluminium-ion clock reaches a relative systematic uncertainty of \(5.5\times10^{-19}\) (2025). On 7 October 2026, two teams (TU Wien and a Chinese team) published the first nuclear clocks in Nature: a vacuum-ultraviolet laser, near 148 nm, is locked to a transition of the thorium-229 nucleus embedded in a crystal. Their stability is still below that of the best atomic clocks. Optical clocks already outperform caesium fountains, but the redefinition of the second is not expected before the 2030 General Conference on Weights and Measures.
NIST, July 2025; Nature, 7 October 2026; BIPM, FAQ on the redefinition of the secondOptically pumped atomic magnetometers measure a magnetic field through the precession of polarised atomic spins, the phenomenon of lab 3.2. They make wearable magnetoencephalography possible, without the cryogenic cooling that SQUIDs require.
2025 review of OPM-MEG — pmc.ncbi.nlm.nih.gov/articles/PMC12320009The 2026 Nobel Prize in Physics rewards Francis Halzen for the IceCube neutrino observatory. It is not a quantum computing prize, but it touches this chapter: a neutrino created with a given “flavour” is a superposition of states with different masses, hence different energies. This state is not stationary, and its flavour oscillates as it propagates.
Nobel Committee press release, 6 October 2026 — nobelprize.org