Landau–Zener Transitions: The Quantum Coin-Flip at an Avoided Crossing

The setup: two levels driven toward a collision

Take the simplest non-trivial quantum system: two states, |1⟩ and |2⟩. Left alone their energies are ε₁(t) and ε₂(t), and we drive the system so these two energies sweep past each other linearly in time. Add a fixed coupling Δ/2 between them. The full Hamiltonian is a 2×2 matrix:

• H(t) = ½( αt·σ_z + Δ·σ_x ), where σ_z and σ_x are Pauli matrices.
• The diabatic energies (coupling switched off) are ±½αt — two straight lines that cross at t = 0.
• Here α = d(ε₁−ε₂)/dt is the sweep rate of the energy difference, in units of J/s (or, dividing by ℏ, in rad/s²).

Diagonalizing gives the adiabatic (instantaneous) eigenvalues E±(t) = ±½√((αt)² + Δ²). Instead of crossing, these two curves recoil from each other — the hallmark avoided crossing — with a minimum separation of exactly Δ at t = 0. This gap is a direct consequence of the von Neumann–Wigner non-crossing theorem: any non-zero coupling forbids a true degeneracy for a generic two-level Hamiltonian. The question Landau–Zener answers is: start deep on one branch at t = −∞ and sweep to t = +∞ — with what probability does the system end up on the other diabatic branch?

The formula and what each symbol means

The remarkable fact is that the infinite-time transition probability is exact — not perturbative, valid for any sweep rate. Writing it as the probability P_dia of ending on the opposite adiabatic branch (i.e. of a diabatic 'jump' through the gap):

• P_dia = exp(−2πΓ), with the adiabaticity parameter Γ = (Δ/2)² / (ℏ·|α|) = Δ²/(4ℏ|α|).
• Δ = minimum energy gap (full splitting at closest approach), in joules or eV.
• α = |d(ε₁−ε₂)/dt| = rate the diabatic energies sweep past each other.
• ℏ = 1.055×10⁻³⁴ J·s.
• The probability of staying on the same energy eigenstate (the adiabatic, level-following outcome) is P_adia = 1 − exp(−2πΓ).

Two clean limits fall out. When the sweep is slow (Γ ≫ 1), P_dia → 0: the system faithfully tracks the instantaneous eigenstate and follows the lower branch around the bend — this is the adiabatic theorem in action. When the sweep is fast (Γ ≪ 1), P_dia → 1: the state has no time to react to the coupling and shoots straight through the gap, staying in its original diabatic state. The crossover happens at Γ ≈ 1, i.e. when the sweep rate α ≈ Δ²/(4ℏ).

Why it happens: the ℏ/Δ time budget

The physics is a race between two clocks. The coupling Δ wants to mix the two states, and it does so on a timescale set by the uncertainty principle: τ_gap ≈ ℏ/Δ. That is the time the system needs to 'feel' the avoided crossing and reorganize into the lower eigenstate. Meanwhile the sweep spends only a limited time near the crossing, where the diabatic levels are within Δ of each other: the levels move apart at rate α, so the interaction window is τ_sweep ≈ Δ/α.

• Compare the two: Γ ∝ τ_sweep / τ_gap ∝ (Δ/α)/(ℏ/Δ) = Δ²/(ℏα). This ratio is the adiabaticity parameter, up to the numerical factor.
• If τ_sweep ≫ τ_gap (Γ ≫ 1): plenty of time to reorganize → the system follows the eigenstate adiabatically.
• If τ_sweep ≪ τ_gap (Γ ≪ 1): the crossing flashes by before the coupling can act → a diabatic jump.

Zener's 1932 derivation turns the Schrödinger equation for the two amplitudes into a Weber (parabolic-cylinder) differential equation, whose asymptotic connection formulae yield the exponential exactly. The appearance of 2π is a genuine feature of that special-function analysis, not a hand-wave — which is why the result is quantitatively trusted across a dozen fields.

A worked number: a spin qubit swept in nanoseconds

Put in real numbers. Consider a spin-½ qubit with an avoided crossing produced by a transverse coupling of Δ/h = 100 MHz, so Δ = h × 10⁸ Hz ≈ 6.6×10⁻²⁶ J. We sweep the diabatic energies by ramping a control field; suppose the energy difference sweeps at α/h = 100 MHz per nanosecond = 10¹⁷ Hz/s, so α = h × 10¹⁷ = 6.6×10⁻¹⁷ J/s.

• Γ = Δ²/(4ℏ|α|). Working in frequency units (divide numerator and denominator by h²): Γ = (Δ/h)² / (4·(1/2π)·(α/h)) = 2π·(10⁸)² / (4·10¹⁷) ≈ 0.157.
• P_dia = exp(−2πΓ) = exp(−2π·0.157) ≈ exp(−0.99) ≈ 0.37.

So this 'fast' sweep leaves ~37% of the population jumping the gap and ~63% following adiabatically — squarely in the crossover. To make the transfer adiabatic (say P_dia < 1%, needing 2πΓ > 4.6, so Γ > 0.73), you must slow the sweep by roughly 5×, to about α/h ≈ 2×10¹⁶ Hz/s (20 MHz/ns). To make it maximally diabatic (a clean pass-through), sweep an order of magnitude faster. This exponential sensitivity — a factor-of-5 change in ramp speed swinging the outcome from 37% to <1% — is exactly why the Landau–Zener knob is such a precise tool in qubit control.

Where it shows up: from molecules to qubits to cold atoms

The two-level-crossing template is astonishingly universal:

• Superconducting and spin qubits. Ramping flux or gate voltage through a qubit's avoided crossing is a standard state-preparation and readout primitive; Landau–Zener sets the ramp speed. Driving repeatedly back and forth produces Landau–Zener–Stückelberg–Majorana (LZSM) interferometry — Stückelberg oscillations in population versus drive amplitude, used to measure gaps and coherence times.
• Molecular chemistry. When two electronic potential-energy surfaces approach along a nuclear coordinate, the nuclei act as the sweeping parameter. Landau–Zener gives the probability of a non-adiabatic hop between surfaces — the heart of charge-transfer reactions, photochemistry, and surface hopping simulations.
• Ultracold atoms and molecules. Sweeping a magnetic field across a Feshbach resonance converts free atom pairs into weakly-bound molecules; the association efficiency is a Landau–Zener probability set by the sweep rate through the resonance.
• Neutrinos. The MSW effect — flavor conversion of solar neutrinos as they pass through the Sun's varying electron density — is a Landau–Zener crossing in matter, with the density gradient playing the role of α.
• Rydberg atoms, NV centers, atomic clocks. Any place two levels are tuned through resonance faces the same adiabatic-versus-diabatic choice.

Subtleties, misconceptions, and where the clean formula breaks

Several traps catch newcomers:

• 'Diabatic' does not mean 'nothing happens.' A diabatic pass keeps you in the same diabatic state, but that state is the upper energy eigenstate on the far side of the crossing — so a fast sweep actually delivers energy up the level structure. Adiabatic following, not the jump, is what keeps you in the ground state.
• The formula is asymptotic (t = ±∞). For finite sweeps the survival probability oscillates and rings before settling; corrections scale like the parabolic-cylinder function's subleading terms. Real experiments approximate '∞' by starting many gap-times before the crossing.
• It is a single-crossing, closed-system result. With multiple crossings you get Stückelberg interference between paths; with an environment, decoherence during the ℏ/Δ window degrades the coherent superposition and the clean exponential no longer holds.
• The gap Δ is the full minimum splitting, not the coupling matrix element alone in every convention. Because α, Δ, and factors of 2 and 2π vary between textbooks, always pin down whether Γ uses Δ or Δ/2 — the exponent can be off by 4× if you don't. A safe cross-check: P_dia should → 1 as the sweep gets infinitely fast and → 0 as it gets infinitely slow.

Finally, note the astonishing feature that makes this result so beloved: it is non-perturbative and exact for a linear sweep, spanning the full range from perfectly adiabatic to perfectly sudden with a single closed-form exponential — a rarity in time-dependent quantum mechanics.

The two limits of a Landau–Zener sweep through an avoided crossing

Feature	Adiabatic (slow)	Diabatic (fast)
Sweep rate α	small, Γ = Δ²/(4ℏα) ≫ 1	large, Γ ≪ 1
Transition probability P_dia	exp(−2πΓ) → 0	exp(−2πΓ) → 1
State follows	the energy eigenstate (lower branch)	the diabatic state (crosses the gap)
Physical picture	level smoothly bends and stays	level shoots straight through
Qubit use	adiabatic state transfer / ground-state prep	fast gate leaving spin in original basis
Time near crossing needed	t ≫ ℏ/Δ	t ≪ ℏ/Δ