Fermi's golden rule

Almost every number in experimental particle physics is a rate: a particle decays so many times per second, two beams collide and produce so many events per second. Fermi's golden rule is the formula that turns quantum mechanics into those rates. It says that the probability current from an initial state $|i\rangle$ into a set of final states $|f\rangle$ factorises cleanly into two independent pieces — a dynamical factor that measures how strongly the interaction couples the states, and a kinematic factor that counts how many final states are on offer. Written compactly, $\Gamma_{fi}=2\pi|T_{fi}|^2\rho(E_i)$ : the transition matrix element squared times the density of final states. That factorisation is not a convenience, it is the deep structure of scattering theory. The matrix element $T_{fi}$ carries all the physics of the specific interaction — electromagnetic, weak, or strong — and when the coupling is weak it unfolds as a perturbation series whose successive terms are exactly the sums over virtual intermediate states that quantum field theory later draws as Feynman diagrams. The density of states $\rho(E_i)$ carries none of the dynamics and all of the counting; energy conservation forces it to be evaluated at the initial energy. This article derives the rule from time-dependent perturbation theory, shows where the sharp energy-conserving delta function comes from, and follows the two factors as they are promoted to the master formulas for decay rates and cross sections used everywhere downstream.

Key takeaways

The rule gives the transition rate as $\Gamma_{fi}=2\pi|T_{fi}|^2\rho(E_i)$ (natural units) — matrix element squared times density of final states.
$T_{fi}$ is the transition matrix element, fixed by the interaction Hamiltonian $\hat H'$ ; for weak coupling it is a perturbation series, not just $\langle f|\hat H'|i\rangle$ .
$\rho(E_i)=(\mathrm{d}n/\mathrm{d}E)_{E_i}$ is the density of final states, evaluated at the initial energy because energy is conserved.
The rate emerges from first-order perturbation theory: a $\mathrm{sinc}^2$ lineshape narrows in time into $2\pi t\,\delta(E_f-E_i)$ , giving a constant rate only in an intermediate time window.
Promoted relativistically — $T_{fi}\to$ Lorentz-invariant $\mathcal M$ , $\rho\to$ Lorentz-invariant phase space — it becomes the decay-rate and cross-section master formulas.
Key idea: A transition rate factorises into how strongly the interaction connects the states (the matrix element squared) times how many final states are available (the density of states).

The core insight

A transition rate factorises into how strongly the interaction connects the states (the matrix element squared) times how many final states are available (the density of states), with energy conservation forcing that density to be counted at the initial energy.

Where this lives today

The golden rule is the root of a family tree that spans modern physics. Promoted relativistically — matrix element into the Lorentz-invariant amplitude $\mathcal M$ , density of states into Lorentz-invariant phase space — it becomes the two master formulas that every collider analysis rests on: the decay rate $\mathrm{d}\Gamma\propto|\mathcal M|^2\,\mathrm{dLIPS}$ and the cross section $\mathrm{d}\sigma\propto|\mathcal M|^2\,\mathrm{dLIPS}/F$ . When the ATLAS and CMS experiments quote a Higgs production rate, or the LHCb collaboration measures a rare $B$ -meson branching fraction, they are evaluating this structure with $\mathcal M$ supplied by Feynman diagrams. The finite-time $\mathrm{sinc}^2$ , matured into a Lorentzian for a genuinely decaying state, is the Breit–Wigner resonance whose width the LEP experiments measured to exquisite precision — $\Gamma_Z=2.4952\pm0.0023\ \mathrm{GeV}$ , from which the number of light neutrino species ( $2.984\pm0.008$ ) was famously read off. Beyond particle physics the same rule sets spontaneous-emission rates in quantum optics and carrier lifetimes in semiconductors. Wherever a quantum system leaks irreversibly into a continuum, Fermi's golden rule is the first formula anyone writes down.

The physical picture

Picture a single sharp energy level — the initial state — sitting beside a dense thicket of final states that spreads across a band of energies. A weak interaction is switched on. It reaches across and mixes the initial state into the thicket, and the system begins to leak irreversibly into the continuum. Two things control how fast it leaks: how strongly the interaction couples $|i\rangle$ to a typical final state (the matrix element $T_{fi}$ ), and how many final states sit at the right energy to receive the amplitude (the density of states $\rho$ ). The first figure makes that two-factor structure literal.

Figure

A single initial level leaks into a continuum of final states. The interaction matrix element T_fi bridges the gap, and the number of final states per unit energy ρ(E) counts how many receivers sit at the conserved energy E_i, giving the rate Γ_fi = 2π T_fi² ρ(E_i)

Where does the rule's most surprising feature — a constant rate, the hallmark of exponential decay — come from? First-order perturbation theory does not hand it to you directly. What it gives is a transition probability that, plotted against the energy mismatch $E_f-E_i$ at a fixed time, is a $\mathrm{sinc}^2$ lineshape: a tall central peak at exact energy conservation, flanked by small side lobes. As time runs on, the central peak grows taller and narrower, its area growing in strict proportion to $t$ . In the limit it becomes a delta function that enforces $E_f=E_i$ exactly, and it is the linear-in-time growth of its area that turns a probability into a rate. The second figure is that emergence — the engine room of the golden rule.

Figure

The transition probability versus energy mismatch is a sinc² lineshape peaked at E_f = E_i. As time grows from t₁ to t₂ ≫ t₁ the central peak sharpens, its area growing as 2πt, approaching the delta function that enforces energy conservation. First zero at ΔE = 2πħ/t, so for t = 1 ns the peak is already only ≈ 4.1 μeV wide — the energy-time uncertainty at work

The core equations

The golden rule itself is a single line. It states that the rate $\Gamma_{fi}$ — a probability per unit time — is the squared transition matrix element multiplied by the density of final states, with a universal factor of $2\pi$ . The relativistic $\hbar$ is restored on the left form and set to one on the right, the convention used throughout particle physics.

\Gamma_{fi}=\frac{2\pi}{\hbar}\,|T_{fi}|^2\,\rho(E_i)\;\xrightarrow{\ \hbar=1\ }\;2\pi\,|T_{fi}|^2\,\rho(E_i) \qquad (1)

where: $\Gamma_{fi}$ — transition rate from $|i\rangle$ to the final continuum, inverse time (or energy in natural units); $T_{fi}$ — transition matrix element, the dynamical factor; $\rho(E_i)$ — density of final states at the initial energy, the kinematic factor. The reading: dynamics and counting sit in separate factors, so the interaction calculation ( $T_{fi}$ ) and the kinematics ( $\rho$ ) are done independently and multiplied. The matrix element is not simply the direct overlap $\langle f|\hat H'|i\rangle$ . For a weak perturbation it is a series in the interaction Hamiltonian $\hat H'$ , whose second term sums over all virtual intermediate states $|j\rangle$ weighted by an energy denominator — the seed of the propagator.

T_{fi}=\langle f|\hat H'|i\rangle+\sum_{j\neq i}\frac{\langle f|\hat H'|j\rangle\langle j|\hat H'|i\rangle}{E_i-E_j}+\cdots \qquad (2)

where: $\hat H'$ — interaction (perturbing) Hamiltonian; $|j\rangle$ — complete set of intermediate eigenstates of $\hat H_0$ ; $E_i-E_j$ — energy denominator, nonzero because $|j\rangle$ is virtual. The reading: the first term is a single vertex, the second is two vertices joined through an off-shell intermediate state — order-by-order this becomes the diagrammatic expansion. The density of states counts momentum eigenstates. In a box of volume $V$ with periodic boundary conditions, states fill momentum space uniformly at $V/(2\pi\hbar)^3$ per unit volume, so the number in a spherical shell divided by $\mathrm{d}E$ gives the density.

\rho(E)=\frac{\mathrm{d}n}{\mathrm{d}E}=\frac{V}{(2\pi\hbar)^3}\,4\pi p^2\,\frac{\mathrm{d}p}{\mathrm{d}E} \qquad (3)

where: $V$ — normalisation volume (it cancels against a $1/V$ in $|T_{fi}|^2$ ); $4\pi p^2\,\mathrm{d}p$ — momentum-space shell volume; $\mathrm{d}p/\mathrm{d}E$ — inverse group velocity, $1/v$ . The reading: $\rho$ is pure kinematics, and $\mathrm{d}p/\mathrm{d}E$ is where non-relativistic ( $E=p^2/2m$ ) and relativistic ( $E^2=p^2+m^2$ ) kinematics differ. Underneath the rule sits the object first-order perturbation theory actually computes: the transition probability at finite time, a $\mathrm{sinc}^2$ in the energy mismatch. This is the pre-limit expression from which the rate is extracted.

P_{fi}(t)=|c_f^{(1)}(t)|^2=\frac{|\langle f|\hat H'|i\rangle|^2}{\hbar^2}\,\frac{4\sin^2(\omega_{fi}t/2)}{\omega_{fi}^2},\qquad \omega_{fi}=\frac{E_f-E_i}{\hbar} \qquad (4)

where: $c_f^{(1)}$ — first-order amplitude to be found in state $|f\rangle$ ; $\omega_{fi}$ — Bohr angular frequency of the transition. The reading: at fixed $t$ this is peaked at $\omega_{fi}=0$ ; the peak height grows as $t^2$ while its width shrinks as $1/t$ , so the area grows only as $t$ . Taking the long-time limit collapses the lineshape into a delta function. Its area is $2\pi t$ , and the surviving delta enforces exact energy conservation, converting probability into a time-independent rate.

\lim_{t\to\infty}\frac{4\sin^2(\omega_{fi}t/2)}{\omega_{fi}^2}=2\pi t\,\delta(\omega_{fi})=2\pi t\hbar\,\delta(E_f-E_i) \qquad (5)

where: $\delta(E_f-E_i)$ — energy-conserving delta that, summed against $\rho(E_f)\,\mathrm{d}E_f$ , sets $E_f=E_i$ . The reading: the linear-in- $t$ area is what makes $\mathrm{d}P/\mathrm{d}t$ constant; combining (4), (5) and $\rho$ assembles equation (1).

Derivation AFrom perturbation theory to a constant transition rate

The goal is to show, starting from nothing but the time-dependent Schrödinger equation and a weak perturbation switched on at $t=0$ , that the probability of finding the system in the final continuum grows linearly in time, and that the constant of proportionality is precisely $2\pi|T_{fi}|^2\rho(E_i)/\hbar$ . The subtle step is the appearance of a sharp energy-conserving delta function out of a perturbation calculation that, at any finite time, respects only an energy-time uncertainty — and understanding that delta is understanding both why the rate is constant and why it is constant only in a window of time.

Step 1. Expand the state in unperturbed eigenstates, $|\psi(t)\rangle=\sum_k c_k(t)\,e^{-iE_k t/\hbar}|k\rangle$ , and insert into $i\hbar\,\partial_t|\psi\rangle=(\hat H_0+\hat H')|\psi\rangle$ . Projecting onto $\langle f|$ leaves $i\hbar\,\dot c_f=\sum_k\langle f|\hat H'|k\rangle e^{i\omega_{fk}t}c_k$ , an exact set of coupled equations for the amplitudes. Why: moving to the interaction picture strips off the trivial free evolution so that only the perturbation drives the coefficients — all the transition physics now lives in the $c_k(t)$ .

Step 2. Impose weak coupling and short times: start in $|i\rangle$ with $c_i\approx1$ and all other $c_k\approx0$ , so the right-hand side keeps only the $k=i$ term. Integrating a constant $\hat H'$ from $0$ to $t$ gives $c_f^{(1)}(t)=\dfrac{\langle f|\hat H'|i\rangle}{i\hbar}\dfrac{e^{i\omega_{fi}t}-1}{i\omega_{fi}}$ . Why: this is first-order perturbation theory — it assumes the initial state is barely depleted, which is exactly the assumption that later limits the rule's validity to a finite time window.

Step 3. Square to a probability. Using $|e^{i\theta}-1|^2=4\sin^2(\theta/2)$ gives equation (4), the $\mathrm{sinc}^2$ lineshape $P_{fi}(t)=\dfrac{|\langle f|\hat H'|i\rangle|^2}{\hbar^2}\dfrac{4\sin^2(\omega_{fi}t/2)}{\omega_{fi}^2}$ . Why: this single function contains the whole story — a central peak of height $\propto t^2$ at $\omega_{fi}=0$ and a width $\propto1/t$ , so it is neither purely a delta nor a smooth function but sharpens with time.

Step 4. Sum over the final continuum. A real detector accepts a band of final states, so multiply by $\rho(E_f)$ and integrate: $P_i(t)=\int P_{fi}(t)\,\rho(E_f)\,\mathrm{d}E_f$ . Because the lineshape is sharply peaked at $E_f=E_i$ , pull the slowly varying $|\langle f|\hat H'|i\rangle|^2\rho(E_f)$ out at $E_f=E_i$ and use $\int\frac{4\sin^2(\omega t/2)}{\omega^2}\,\mathrm{d}\omega=2\pi t$ . Why: the continuum is what makes the decay irreversible — a single discrete final state would just oscillate back and forth (Rabi), and it is the integral over many states that dephases and gives a monotone probability.

Step 5. Read off the rate. The integral leaves $P_i(t)=\dfrac{2\pi t}{\hbar}|\langle f|\hat H'|i\rangle|^2\rho(E_i)$ , linear in $t$ , so $\Gamma=\mathrm{d}P_i/\mathrm{d}t=\dfrac{2\pi}{\hbar}|T_{fi}|^2\rho(E_i)$ , equation (1). Why: the linear-in- $t$ growth is the whole point — only a probability proportional to time defines a time-independent rate, and this holds only while $t$ is long compared with $\hbar/(\text{final-state bandwidth})$ yet short compared with the lifetime $1/\Gamma$ .

Go deeper · Derivation B · 1 minCounting the final states — the density of states factor

Derivation BCounting the final states — the density of states factor

The goal is to compute the kinematic factor $\rho(E)$ explicitly by counting momentum eigenstates in a normalisation box, and to expose the single place — the derivative $\mathrm{d}p/\mathrm{d}E$ — where relativity enters. This factor carries no dynamics whatsoever; it is pure phase-space bookkeeping, and getting it right is what lets the same golden rule serve non-relativistic scattering, nuclear beta decay, and relativistic particle production once the kinematics are swapped in.

Step 1. Quantise momenta in a cube of side $L$ with periodic boundary conditions: $p_x=2\pi\hbar n_x/L$ , and likewise for $y,z$ , with integer $n$ . Each allowed state occupies a momentum-space cell of volume $(2\pi\hbar/L)^3=(2\pi\hbar)^3/V$ , so the number of states per unit momentum-space volume is $V/(2\pi\hbar)^3$ . Why: periodic boundary conditions turn a continuum of momenta into a countable lattice, making "number of states" a well-defined thing to differentiate.

Step 2. Count states in a momentum shell. The number with $|\mathbf p|$ in $[p,p+\mathrm{d}p]$ is $\mathrm{d}n=\dfrac{V}{(2\pi\hbar)^3}\,4\pi p^2\,\mathrm{d}p$ , the shell volume times the density. Why: for an isotropic final state only the magnitude of momentum matters, so the natural counting volume is the spherical shell $4\pi p^2\,\mathrm{d}p$ .

Step 3. Convert to energy with the chain rule: $\rho(E)=\dfrac{\mathrm{d}n}{\mathrm{d}E}=\dfrac{V}{(2\pi\hbar)^3}\,4\pi p^2\,\dfrac{\mathrm{d}p}{\mathrm{d}E}$ , equation (3). Why: the golden rule wants states per unit energy, and $\mathrm{d}p/\mathrm{d}E=1/v$ is the inverse velocity — the one factor that knows whether the particle is slow or relativistic.

Step 4. Put in the kinematics. Non-relativistically $E=p^2/2m$ gives $\mathrm{d}p/\mathrm{d}E=m/p$ , so $\rho=\dfrac{4\pi Vmp}{(2\pi\hbar)^3}=\dfrac{V}{4\pi^2}\Big(\dfrac{2m}{\hbar^2}\Big)^{3/2}\sqrt{E}$ ; relativistically $E^2=p^2c^2+m^2c^4$ gives $\mathrm{d}p/\mathrm{d}E=E/(pc^2)$ , so $\rho=\dfrac{V}{(2\pi\hbar)^3}\dfrac{4\pi pE}{c^2}$ . Why: the same counting formula covers both regimes — the branch point is entirely in $\mathrm{d}p/\mathrm{d}E$ — and it is the relativistic branch, generalised to several final particles, that becomes Lorentz-invariant phase space.

Step 5. Note the volume cancellation. A single-particle final-state normalisation gives $|T_{fi}|^2\propto1/V$ , which cancels the explicit $V$ in $\rho$ , so the physical rate is independent of the fictitious box. Why: $V$ is a calculational scaffold, and its disappearance from every observable is the consistency check that the counting was done correctly.

Go deeper · Derivation C · 1 minThe transition matrix element as a perturbation series

Derivation CThe transition matrix element as a perturbation series

The goal is to justify writing $T_{fi}$ as the full series of equation (2) rather than the bare overlap $\langle f|\hat H'|i\rangle$ , and to show that its second-order term — a sum over virtual intermediate states divided by an energy denominator — is the non-relativistic ancestor of the Feynman propagator. This is where the golden rule stops being a formula about atoms and starts being the organising principle of every scattering amplitude in particle physics.

Step 1. Return to the amplitude equations of the previous derivation but iterate to second order: feed the first-order $c_j^{(1)}(t)$ back into $i\hbar\,\dot c_f=\sum_k\langle f|\hat H'|k\rangle e^{i\omega_{fk}t}c_k$ . The final amplitude now receives a contribution routed through every intermediate state $|j\rangle$ . Why: a single application of $\hat H'$ can only reach states it connects directly to $|i\rangle$ ; two applications reach further, through a stepping-stone state that need not conserve energy on its own.

Step 2. Perform the time integrals in the same long-time limit. The direct path contributes $\langle f|\hat H'|i\rangle$ ; the two-step path contributes $\sum_{j\neq i}\dfrac{\langle f|\hat H'|j\rangle\langle j|\hat H'|i\rangle}{E_i-E_j}$ , with an energy denominator $E_i-E_j$ that does not vanish because $|j\rangle$ is off the energy shell. Why: the denominator is the propagator in disguise — it weights each virtual route by how far off-shell it is, so distant intermediate states contribute little.

Step 3. Identify the effective operator. The series defines a corrected coupling, sometimes written $T_{fi}=\langle f|\hat H'|i\rangle+\sum_j\langle f|\hat H'|j\rangle\frac{1}{E_i-E_j}\langle j|\hat H'|i\rangle+\cdots$ , an operator sandwiched between $\langle f|$ and $|i\rangle$ . Why: this is exactly the structure a Feynman diagram encodes — vertices ( $\hat H'$ ) joined by internal lines (energy denominators / propagators) summed over intermediate states.

Step 4. Recognise the physical content. Processes forbidden at first order (where $\langle f|\hat H'|i\rangle=0$ ) can proceed at second order through virtual states — this is how second-order transitions, Raman scattering, and weak-interaction box diagrams all arise. Why: the perturbation series is not a mathematical nicety, it is the mechanism of virtual-particle exchange, and truncating $T_{fi}$ at the first term silently discards it.

Limiting cases

Case	Result	Meaning
Intermediate-time window	$P\propto t$ , $\Gamma$ constant	the golden-rule regime: $\hbar/\Delta E_{\text{band}}\ll t\ll1/\Gamma$ ; a well-defined rate exists
Very short time	$P\propto t^2$	quadratic growth, no constant rate; repeated measurement freezes decay (quantum Zeno)
Very long time	$P$ saturates, deviates from $e^{-\Gamma t}$	initial state depleted; first-order theory breaks down, non-exponential tails appear
Discrete final state	$P$ oscillates (Rabi)	no continuum to dephase into; transition is reversible, no irreversible rate
Non-relativistic final state	$\rho\propto\sqrt{E}$ , $\mathrm{d}p/\mathrm{d}E=m/p$	slow particles; DOS grows with energy — atomic and nuclear applications
Relativistic final state	$\rho\propto pE$ , $\mathrm{d}p/\mathrm{d}E=E/p$	promotes to Lorentz-invariant phase space; particle production at colliders

Common trap“Fermi's golden rule gives the exact transition probability at all times.”

It does not — it gives a constant rate only inside a window of time, and both edges of that window matter. At very short times the probability grows as $t^2$ , not $t$ , so there is no rate at all; a system watched continuously never decays (the quantum Zeno effect is this statement made experimental). At very long times the derivation's own assumption — that the initial state stays nearly full, $c_i\approx1$ — fails once an appreciable fraction has leaked away, and the true survival probability develops non-exponential tails that first-order theory cannot see. The clean $\Gamma=2\pi|T_{fi}|^2\rho$ holds only for $\hbar/\Delta E_{\text{band}}\ll t\ll1/\Gamma$ : long enough for the $\mathrm{sinc}^2$ to have sharpened into an effective delta over the width of the final-state band, but short enough that the source is not yet drained. A companion trap lives in equation (2): treating $T_{fi}$ as merely $\langle f|\hat H'|i\rangle$ throws away every higher-order term, and with them all virtual-state and second-order processes.

Worked examples

1 · How many final states is a continuum? (fully worked)

A free electron is confined to a normalisation box of volume $V=1\ \mathrm{m}^3$ . Find the density of states $\rho(E)$ at kinetic energy $E=1\ \mathrm{eV}$ , in states per joule and per electron-volt, and comment on why the continuum approximation is so safe.

Step 1. Choose the non-relativistic branch, since $1\ \mathrm{eV}\ll m_ec^2=511\ \mathrm{keV}$ : $\rho(E)=\dfrac{V}{4\pi^2}\Big(\dfrac{2m_e}{\hbar^2}\Big)^{3/2}\sqrt{E}$ . Insert $m_e=9.109\times10^{-31}\ \mathrm{kg}$ , $\hbar=1.055\times10^{-34}\ \mathrm{J\,s}$ , $E=1\ \mathrm{eV}=1.602\times10^{-19}\ \mathrm{J}$ .

Step 2. Evaluate the bracket: $2m_e/\hbar^2=1.637\times10^{38}\ \mathrm{J^{-1}m^{-2}}$ , so $(2m_e/\hbar^2)^{3/2}=2.09\times10^{57}$ . With $\sqrt{E}=4.00\times10^{-10}\ \mathrm{J^{1/2}}$ and $V/4\pi^2=0.0253\ \mathrm{m}^3$ , the product gives $\rho\approx\mathbf{2.13\times10^{46}\ states/J}$ .

Step 3. Convert to per-eV by multiplying by $1.602\times10^{-19}\ \mathrm{J/eV}$ : $\rho\approx\mathbf{3.4\times10^{27}\ states/eV}$ . Cross-checking with the momentum form $\rho=4\pi Vm_ep/(2\pi\hbar)^3$ using $p=\sqrt{2m_eE}$ reproduces $2.13\times10^{46}\ \mathrm{J^{-1}}$ exactly.

Step 4. Interpret. There are of order $10^{27}$ final states per electron-volt in a cubic-metre box, so the level spacing is $\sim10^{-27}\ \mathrm{eV}$ — utterly unresolvable. This is why replacing the sum over final states by an integral against $\rho(E)$ , the move that dephases the $\mathrm{sinc}^2$ into a delta, is overwhelmingly justified for any macroscopic volume. $\rho(1\ \mathrm{eV})\approx2.1\times10^{46}\ \mathrm{J^{-1}}\approx3.4\times10^{27}\ \mathrm{eV^{-1}}$ per cubic metre.

2 · Width and lifetime are the same statement (fully worked)

The golden rule delivers a rate $\Gamma$ , and a state that decays at rate $\Gamma$ has survival probability $e^{-\Gamma t}$ , hence lifetime $\tau=1/\Gamma$ (natural units) or $\tau=\hbar/\Gamma$ in SI. Use this to connect two very different particles: the muon, with $\tau_\mu=2.197\ \mu\mathrm{s}$ , and the Z boson, with total width $\Gamma_Z=2.495\ \mathrm{GeV}$ .

Step 1. Muon, lifetime to width. With $\hbar=6.582\times10^{-16}\ \mathrm{eV\,s}$ , $\Gamma_\mu=\hbar/\tau_\mu=(6.582\times10^{-16})/(2.197\times10^{-6})=\mathbf{3.00\times10^{-10}\ eV}$ , i.e. $3.0\times10^{-19}\ \mathrm{GeV}$ . An extraordinarily narrow width — the muon is long-lived precisely because its weak matrix element is tiny.

Step 2. Z boson, width to lifetime. With $\hbar=6.582\times10^{-25}\ \mathrm{GeV\,s}$ , $\tau_Z=\hbar/\Gamma_Z=(6.582\times10^{-25})/2.495=\mathbf{2.64\times10^{-25}\ s}$ . The Z lives about a quarter of a yoctosecond — 19 orders of magnitude shorter than the muon.

Step 3. Read the physics. Both numbers are the same golden-rule rate wearing different clothing: a rate $\Gamma$ , a lifetime $\tau=\hbar/\Gamma$ , and — because the $\mathrm{sinc}^2$ of a truly decaying state is a Lorentzian — a resonance width $\Gamma$ in the energy spectrum. The relation $\Gamma\tau=\hbar$ is the energy-time uncertainty made exact. $\Gamma_\mu=3.0\times10^{-10}\ \mathrm{eV}$ ; $\tau_Z=2.6\times10^{-25}\ \mathrm{s}$ ; in both, width × lifetime = $\hbar$ .

Physics of the portrait

The name attached to the most-used formula in particle physics belongs to Enrico Fermi, though he was neither the first to derive it nor shy about saying so — he called it, with characteristic dryness, "Golden Rule No. 2."

Origins · 1 minEnrico Fermi and how this was discovered

Enrico Fermi

1901–1954 · quantum transition rates, the theory of β decay, and the naming of the golden rule

The transition-rate formula was first derived by Paul Dirac in 1927 within his new time-dependent perturbation theory of radiation, and applied soon after by Gregor Wentzel to radiationless (Auger) transitions. What made it Fermi's was 1933: his theory of beta decay took the abstract rule, supplied a concrete four-fermion interaction Hamiltonian for the weak force, and computed actual nuclear decay rates and electron spectra from it — the first time the rule's two factors, matrix element and density of states, were put to work predicting a real particle-physics observable. The Fermi coupling constant $G_F$ that still parametrises the weak interaction dates from this calculation. Fermi's own name for the formula appears in his Chicago lecture notes around 1949–1950, where he lists it as the second of two "golden rules" for computing rates, and students propagated the label until it stuck. Dirac wrote the equation, but Fermi showed what it was for — turning a line of perturbation theory into the engine that computes how particles decay and scatter.

History

1927

— Dirac derives the first-order transition rate within time-dependent perturbation theory of radiation

1928

— Wentzel applies the same rate to radiationless Auger transitions

1933

— Fermi's theory of β decay computes nuclear rates and spectra from the rule and a weak Hamiltonian

1949

— Fermi's Chicago lecture notes name it "Golden Rule No. 2"

1958

— the V−A structure of the weak current sharpens the matrix element Fermi first wrote

Going further

Texts. Thomson, Modern Particle Physics, §3.1 and the derivation in §2.3.6 · Griffiths, Introduction to Elementary Particles, 2nd ed., §6.1 (Fermi's golden rule and decay rates) · Halzen & Martin, Quarks and Leptons, §4 (transition rates and phase space) · Sakurai & Napolitano, Modern Quantum Mechanics, §5.7 (time-dependent perturbation theory and the golden rule) · Peskin & Schroeder, An Introduction to Quantum Field Theory, §4.5 (relativistic decay rates and cross sections) · Schwartz, Quantum Field Theory and the Standard Model, Ch. 5. Key results to keep in mind: the $\mathrm{sinc}^2\to2\pi t\,\delta$ limit yields a linear-in- $t$ probability and hence a constant rate; the density of states $\rho=\frac{V}{(2\pi\hbar)^3}4\pi p^2\,\mathrm{d}p/\mathrm{d}E$ reduces to $\frac{V}{4\pi^2}(2m/\hbar^2)^{3/2}\sqrt E$ non-relativistically; numerically, $\rho(1\ \mathrm{eV},1\ \mathrm{m}^3)=2.1\times10^{46}\ \mathrm{J^{-1}}=3.4\times10^{27}\ \mathrm{eV^{-1}}$ ; and the width–lifetime anchor values are $\Gamma_\mu=\hbar/\tau_\mu=3.0\times10^{-10}\ \mathrm{eV}$ and $\tau_Z=\hbar/\Gamma_Z=2.6\times10^{-25}\ \mathrm{s}$ . The relativistic promotion to $|\mathcal M|^2$ and Lorentz-invariant phase space is treated in the companion articles on Lorentz-invariant phase space, particle decay rates, and interaction cross sections.

Prerequisites. Non-relativistic QM review · Natural units. Leads to. Lorentz-invariant phase space · Particle decay rates · Interaction cross sections.

Ask the tutor: Want to see the $\mathrm{sinc}^2$ lineshape sharpen into energy conservation, or walk the golden rule up to the relativistic decay-rate master formula step by step?

Why is it called a "golden rule," and did Fermi discover it?

The formula was derived by Dirac in 1927 as part of his time-dependent perturbation theory. Fermi used it so heavily — most importantly in his 1933 theory of beta decay — and taught it so memorably that his own nickname stuck. In his Chicago lecture notes he listed two workhorse results as "Golden Rule No. 1" and "Golden Rule No. 2"; the second one, the transition rate, is what everyone now means by "Fermi's golden rule." So the name honours the physicist who made it indispensable, not the one who first wrote it down — a common pattern in physics naming.

Where exactly does energy conservation come from if the perturbation acts at all times?

From the long-time limit, not from any input assumption. At finite time the calculation respects only energy-time uncertainty: the $\mathrm{sinc}^2$ lineshape has a nonzero width $\sim\hbar/t$ , so transitions to states slightly off the initial energy are allowed. As $t$ grows the lineshape narrows and its integral becomes $2\pi t\,\delta(E_f-E_i)$ — a strict delta that permits only $E_f=E_i$ . Energy conservation is therefore an emergent, long-time statement, which is why it coexists with virtual intermediate states in $T_{fi}$ that individually violate it.

Is Γ a rate or an energy? The units confuse me.

Both, and that is the point of natural units. Dimensionally a rate is inverse time. Setting $\hbar=1$ makes time and inverse energy the same unit, so a rate can be quoted as an energy — this is why the Z boson's "width" is $2.495\ \mathrm{GeV}$ even though it is fundamentally a decay rate. To recover SI, restore $\hbar$ : the lifetime is $\tau=\hbar/\Gamma$ , and $\Gamma_{\text{[energy]}}\times\tau_{\text{[time]}}=\hbar$ . So a wide resonance is short-lived, a narrow one long-lived.

Why must the final states form a continuum? What if there's only one?

A single final state gives a reversible two-level (Rabi) oscillation: probability sloshes back and forth and never settles into a steady rate. The golden rule needs the sum over many final states, weighted by $\rho(E)$ , so that their slightly different phases wash out and the probability grows monotonically. Physically, the continuum is a "sink" the initial state can leak into irreversibly. This is why the rule applies to decay into free-particle momentum states (a genuine continuum) but not to a two-state atom in a lossless cavity.

How does this non-relativistic formula become the collider master formulas?

By promoting each factor to a Lorentz-invariant form. The matrix element $T_{fi}$ , built from a specific $\hat H'$ , is replaced by the Lorentz-invariant amplitude $\mathcal M$ computed from Feynman rules; the density of states $\rho$ , tied to a normalisation box, is replaced by Lorentz-invariant phase space $\mathrm{dLIPS}$ with its $\frac{\mathrm{d}^3p}{(2\pi)^3 2E}$ per final particle. The result is $\mathrm{d}\Gamma\propto|\mathcal M|^2\,\mathrm{dLIPS}$ for decays and $\mathrm{d}\sigma\propto|\mathcal M|^2\,\mathrm{dLIPS}/F$ for cross sections — the two-factor skeleton intact, only the pieces made relativistic.

What is the quantum Zeno effect, and how is it hidden in the golden rule?

It is the short-time edge of the rule's validity. Before the $\mathrm{sinc}^2$ area has entered its linear-in- $t$ regime, the survival probability falls only as $1-(\text{const})t^2$ , quadratic rather than linear. A measurement that projects the system back to $|i\rangle$ resets this quadratic clock, so frequent enough measurement keeps the decay probability near zero — a watched quantum pot never boils. The golden rule's constant rate assumes you are past this initial quadratic phase, which is why it is a statement about an intermediate time window, not about $t=0$ .