Fermi's Golden Rule: How Fast a Quantum System Jumps

The equation and what each piece does

Fermi's Golden Rule delivers a transition rate — a probability per unit time that a system in an initial quantum state |i⟩ jumps to a set of final states |f⟩ under a small, steady perturbation Ĥ′. The rate is

• Γᵢ→f = (2π/ℏ) |⟨f|Ĥ′|i⟩|² ρ(E_f)

with three ingredients:

• ⟨f|Ĥ′|i⟩ — the matrix element, the quantum overlap between the perturbation acting on the initial state and the final state. Its square, |M|², has units of energy². It measures how strongly the two states are coupled; forbidden transitions have M = 0.
• ρ(E_f) — the density of final states at the final energy, in states per joule (or states per eV). A jump into a rich continuum goes faster than a jump into a sparse one.
• 2π/ℏ — with ℏ = 1.055 × 10⁻³⁴ J·s, this prefactor is ≈ 5.96 × 10³⁴ (J·s)⁻¹, converting energy²·(states/J) into pure s⁻¹.

Dimensional check: [energy²] × [states/energy] × [1/(energy·time)] = 1/time. The result is a clean rate Γ, and its reciprocal τ = 1/Γ is the state's lifetime. The Golden Rule is thus a machine that turns a coupling strength and a count of destinations into a decay clock.

Deriving it from time-dependent perturbation theory

The rule is not an axiom; it falls out of first-order perturbation theory. Split the Hamiltonian as Ĥ = Ĥ₀ + Ĥ′, where Ĥ₀ has known eigenstates |n⟩ of energy Eₙ and Ĥ′ is a weak, switched-on perturbation. Expand the wavefunction in the |n⟩ basis with time-dependent amplitudes cₙ(t). To first order, the amplitude to be in a final state |f⟩ (starting in |i⟩ at t=0) is

• c_f(t) ≈ −(i/ℏ) ⟨f|Ĥ′|i⟩ ∫₀ᵗ e^{iω_fi t′} dt′, with ℏω_fi = E_f − E_i.
• For a constant perturbation this integrates to give the transition probability P_f(t) = |c_f|² = (|M|²/ℏ²) · [sin²(ω_fi t/2) / (ω_fi/2)²].
• The bracket is a sinc² function of energy mismatch. As t grows, it becomes a tall, narrow spike centered on ω_fi = 0, with area 2πt. Formally, sin²(ω t/2)/(ω/2)² → 2πt · δ(ω), enforcing energy conservation E_f = E_i in the long-time limit.

Because the spike's area grows linearly with t, the probability of landing anywhere in a continuum grows linearly too — so d P/dt is a constant rate. Summing over final states via ∫ dE_f ρ(E_f) and collapsing the delta function yields Γ = (2π/ℏ)|M|²ρ(E_f). The linear-in-t growth (not t², not oscillatory) is precisely what makes decay irreversible and defines a rate at all.

Why a continuum of final states is essential

The single most misunderstood feature is that the Golden Rule needs many densely packed final states, not one. If |f⟩ is a lone discrete level, the atom does not decay — it undergoes reversible Rabi oscillation, sloshing population back and forth at the Rabi frequency Ω = |M|/ℏ. The population goes as sin²(Ωt/2): it returns. Nothing is lost.

Irreversibility requires a bath of destinations. The sinc² spike has a finite width in energy, roughly ΔE ≈ 2πℏ/t. Only when many states fall within that width — i.e. ρ(E) ΔE ≫ 1 — do their contributions add incoherently and phase-scramble, so the system never revives. This is the microscopic origin of exponential decay: population feeds a continuum whose modes dephase, and P(t) = e^{−Γt}.

• An excited atom decays because the electromagnetic vacuum offers a continuum of photon modes — one for every direction, polarization, and frequency.
• A β-emitting nucleus decays because the emitted electron and antineutrino share the released energy in a continuum of momentum partitions — the source of the smooth β-energy spectrum.
• An electron trapped on a defect relaxes into the continuum of conduction-band states, whose ρ(E) sets the scattering rate.

The regime of validity is therefore twofold: the perturbation must be weak (Γ·t stays ≪ 1 over the timescale of the derivation, so P ≪ 1), and the final states must form a quasi-continuum. Outside these bounds you need Rabi dynamics, higher-order terms, or the full non-perturbative Schrödinger equation.

A worked feel: spontaneous emission and the 1.6 ns lifetime

Apply the rule to an excited atom radiating into empty space. The perturbation is the dipole coupling Ĥ′ = −d·E between the atomic dipole d and the vacuum electric field; the final states are the photon modes, whose density in a box scales as ρ(ω) ∝ ω²V/(π²c³). Carrying the mode sum through gives the Einstein A coefficient:

• A = Γ = ω³|⟨f|d̂|i⟩|² / (3πε₀ℏc³)

Read off the scaling. The rate goes as ω³ — the density of photon modes (∝ ω²) times the field-per-photon factor (∝ ω). Blue transitions decay far faster than red ones; UV atomic lines are broad, radio hyperfine lines are agonizingly narrow. Plugging numbers for the hydrogen 2p→1s line (ω ≈ 1.55 × 10¹⁶ rad/s, dipole matrix element ~ea₀ with a₀ = 5.29 × 10⁻¹¹ m) gives Γ ≈ 6.3 × 10⁸ s⁻¹, so τ ≈ 1.6 ns — matching experiment. The same formula explains why the 21 cm hydrogen hyperfine line (ν = 1.42 GHz) has A ≈ 2.9 × 10⁻¹⁵ s⁻¹, a lifetime of about 11 million years: the ω³ suppression at low frequency is brutal.

The controlling variables: matrix element and density of states

Everything a Golden-Rule rate does, it does through its two factors. Tune either and you tune the lifetime.

• Matrix element |M|² encodes selection rules and coupling strength. For electric-dipole radiation it demands Δℓ = ±1, Δm = 0,±1; when these fail, M vanishes at dipole order and the transition is 'forbidden' — it may still proceed via weaker magnetic-dipole or electric-quadrupole terms, but 10⁴–10⁸ times slower. Metastable states (like the astrophysical [O III] green lines) owe their long lives to a tiny M.
• Density of states ρ(E) is pure environment. It is why the Purcell effect works: place an atom inside an optical cavity resonant with its transition and you sharply raise ρ at that frequency, boosting Γ by the Purcell factor F_P = 3Qλ³/(4π²V) — decays speed up by 10× to 1000×. Conversely, a photonic bandgap that removes modes (ρ → 0) can inhibit emission entirely. In solids, ρ(E) inherits the band structure: van Hove singularities where ρ spikes produce fast relaxation channels.

This factorization is the rule's practical power. An engineer changes the matrix element by chemistry and geometry (which orbitals, how big the dipole) and changes ρ(E) by nanophotonics (cavities, waveguides, plasmonic antennas). The rate is the product, so the two levers multiply.

Where the rule runs the world

Fermi's Golden Rule is arguably the most-used equation in applied quantum physics. It quietly sets the numbers behind:

• Nuclear and particle decay. Fermi's own 1934 theory of β decay used it to derive the electron energy spectrum and the coupling now called the Fermi constant, G_F ≈ 1.166 × 10⁻⁵ GeV⁻². The rule connects a nucleus's half-life to the weak matrix element — spanning lifetimes from milliseconds to 10¹⁹ years.
• Optoelectronics. LED and laser-diode brightness, photodetector responsivity, and solar-cell absorption are all Golden-Rule rates: the matrix element between valence and conduction bands times the joint density of states. Semiconductor gain spectra are computed this way.
• Spectroscopy. Absorption and emission line strengths, Raman cross-sections, and X-ray photoemission intensities are matrix-element-squared × density-of-states.
• Chemistry and biology. Marcus theory of electron transfer and Förster resonance energy transfer (FRET), used as a molecular 'ruler' in cell biology, are Golden-Rule expressions where a Franck–Condon-weighted density of states replaces ρ.
• Solid-state transport. Electron–phonon and impurity scattering rates — hence electrical resistivity and carrier mobility — come straight from 1/τ = (2π/ℏ)|M|²ρ, the input to the Drude and Boltzmann transport pictures.

Subtleties, limits, and honest caveats

The rule is powerful but not magic, and several caveats matter:

• It is first-order. If |M|² is large or the process is second-order (two-photon absorption, Raman, Auger via intermediate 'virtual' states), you need the second-order Golden Rule, which sums over virtual states with energy denominators: |Σ_m ⟨f|Ĥ′|m⟩⟨m|Ĥ′|i⟩/(E_i − E_m)|².
• The 'constant rate' is an approximation window. At very short times P grows as t² (the quantum Zeno regime — watch a system continuously and you freeze its decay); at very long times exponential decay develops power-law tails. The linear-in-t rate holds only in the intermediate window, which for atomic decay is astronomically wide but not literally infinite.
• The delta function is an idealization. Energy is conserved only up to the state's own width; the finite lifetime broadens the level by ΔE ≈ ℏΓ (the natural linewidth), a self-consistency that the bare rule ignores but Weisskopf–Wigner theory restores.
• ρ(E_f) must be evaluated at the right energy — the energy the final state actually reaches, E_f = E_i ± ℏω for absorption/emission — not at E_i. Getting the density-of-states argument wrong is the most common error in applying the rule.

A note on the name: the physics of the linear formula was first written down by Paul Dirac in 1927. Enrico Fermi popularized it in his 1949–1950 Chicago lectures, where he labeled it 'Golden Rule No. 2' — and Fermi's name stuck, an example of Stigler's law of eponymy.

Two competing pictures of a quantum transition: coherent Rabi flopping (discrete final state) versus irreversible Golden-Rule decay (continuum of final states).

Property	Rabi oscillation	Fermi's Golden Rule
Final states	Single discrete level	Continuum, density ρ(E_f)
Time dependence	Reversible, sin²(Ωt/2)	Irreversible, linear in t
Population early on	∝ t² then oscillates	∝ t (constant rate Γ)
Governing quantity	Rabi frequency Ω = |M|/ℏ	Rate Γ = (2π/ℏ)|M|²ρ
Energy conservation	Detuning allowed, off-resonant	δ(E_f − E_i), strictly enforced
Example	Two-level atom in laser	Atom emitting into vacuum modes