Landau–Zener Transitions

Sweep two coupled quantum levels through their crossing point and you face the sharpest question in time-dependent quantum mechanics: does the system follow the bending eigenstate, or jump straight across the gap? Adiabatic Approximation and Validity Conditions identifies exactly this — a closing gap swept at finite rate — as the failure mode no slowness bookkeeping can rescue. Here we answer it quantitatively, with one of the most celebrated closed forms in quantum theory: for diabatic energies crossing linearly at rate $\alpha$ with minimum gap $\Delta$ , the jump probability is exactly $P_{\mathrm{LZ}}=e^{-\pi\Delta^2/(2\hbar|\alpha|)}$ . Four physicists — Landau, Zener, Stückelberg and Majorana — solved this problem independently in a single year, 1932. We build the result three ways: a one-line Gaussian integral giving the fast-sweep limit and the origin of the $\pi$ ; Zener's reduction to the parabolic-cylinder equation, upgrading the perturbative seed to the exact exponential with an honest account of where the rigor lives; and the article's laboratory — direct numerical integration across three decades of sweep rate, landing on $e^{-2\pi\Gamma}$ to better than one part in $10^6$ , Stückelberg interference and all. Along the way we prove why avoided crossings are generic, repair a factor-of-two slip in this article's own source material, and put real numbers on a qubit sweep and a Na–Cl harpoon collision. If the adiabatic theorem is the promise, Landau–Zener is the fine print.

Key takeaways

Levels swept linearly through an avoided crossing of minimum gap $\Delta$ at rate $\alpha=\mathrm{d}(E_2-E_1)/\mathrm{d}t$ jump with probability $P_{\mathrm{LZ}}=e^{-2\pi\Gamma}$ , $\Gamma=\Delta^2/(4\hbar|\alpha|)$ the only knob.
Slow sweep ( $\Gamma\gg1$ ): adiabatic following — rapid passage. Fast ( $\Gamma\ll1$ ): the state stays frozen diabatically and jumps. At $\Gamma=1$ , $P_{\mathrm{LZ}}=e^{-2\pi}\approx1.9\times10^{-3}$ already.
First-order theory gives $1-P_{\mathrm{LZ}}\approx2\pi\Gamma$ from one Fresnel integral; Zener's parabolic-cylinder reduction makes it exact; our numerics confirm to $2\times10^{-7}$ absolute across $\Gamma=0.003$ – $3$ .
Two passages interfere: net transfer oscillates as $4P_{\mathrm{LZ}}(1-P_{\mathrm{LZ}})\sin^2\Phi_{\mathrm{St}}$ — Stückelberg interferometry, measured here with envelope $0.9955$ .
Key idea: project the dynamics onto the two relevant states, model the diabatic energies as crossing linearly in time with a constant coupling, and the entire physics collapses to one number: the adiabaticity $\Gamma=\Delta^2/(4\hbar|\alpha|)$ , with diabatic-survival probability $e^{-2\pi\Gamma}$ .

Where this lives today

Landau–Zener physics is the operating manual for every technology that steers quantum states by tuning energies. Adiabatic rapid passage — sweep slowly, transfer with exponential certainty — prepares states in NMR, cold atoms and trapped ions because it is robust against the calibration errors that ruin resonant pulses. Quantum annealers live on the opposite ledger entry: the runtime doctrine that Adiabatic Approximation and Validity Conditions attaches to the smallest gap along the schedule is, at each near-crossing, this formula demanding $\Gamma\gg1$ — one narrow avoided crossing can silently ruin a schedule. In circuit QED, deliberately repeated fast passages produce Landau–Zener–Stückelberg–Majorana fringes that map a qubit's spectrum without a spectroscopy tone. And in ultracold lattices, a controlled force sweeps quasimomentum across the Brillouin-zone edge, where interband "Zener tunnelling" — the same $e^{-2\pi\Gamma}$ with the band gap for $\Delta$ — either confines atoms to a band (Bloch oscillations) or launches them ballistically, at the experimenter's choice of $\Gamma$ .

The physical picture

Picture the two diabatic energies — the levels with the coupling switched off — as straight lines crossing in time, an X. The coupling $\Delta/2$ forbids the touch: it bends the X into two hyperbolas, the adiabatic curves, with minimum separation $\Delta$ . Sweep slowly and the state turns the corner, riding the lower hyperbola from one diabatic character into the other. Sweep fast and it barrels straight along its diabatic line, leaping the gap as if the coupling weren't there.

Figure

The avoided crossing: diabatic lines, adiabatic hyperbolas, and the two exit routes

The competition is between two clocks: the dwell time $\sim\Delta/|\alpha|$ in the crossing region, and the internal clock at the pinch, period $\sim\hbar/\Delta$ . Their ratio is, up to a factor of 4, the adiabaticity $\Gamma$ : many ticks per dwell and the state rotates with the Hamiltonian; a fraction of a tick and it is frozen — the sudden approximation applied locally, as The Sudden Approximation frames it. Why an exponential rather than a power law? The transition amplitude is an oscillatory integral whose saddle sits at complex time $t_*=\pm i\Delta/|\alpha|$ — where the two adiabatic energies actually touch, off the real axis. The imaginary action out to that branch point is the exponent: jumping a swept gap is tunnelling in time. The payoff of exactness is measured below — our numerical jump probabilities sit on the predicted line across three decades of sweep rate.

Figure

The laboratory: measured jump probability lands on e^(−2πΓ) across three decades, with Stückelberg fringes on the second passage

The two-level stage

Every swept avoided crossing is modelled by the same two-state Hamiltonian. Project onto the two crossing states — the diabatic basis $\{|1\rangle,|2\rangle\}$ — let their uncoupled energy difference pass through zero linearly at rate $\alpha$ , and keep a constant coupling $\Delta/2$ :

H(t)=\frac{1}{2}\begin{pmatrix}\alpha t & \Delta\\ \Delta & -\alpha t\end{pmatrix}=\frac{\alpha t}{2}\,\sigma_z+\frac{\Delta}{2}\,\sigma_x \qquad (1)

where: $\alpha=\mathrm{d}(E_2-E_1)/\mathrm{d}t$ is the sweep rate of the diabatic energy difference, $\Delta$ is twice the coupling matrix element, and $t=0$ marks the would-be crossing. The model is honest because near the crossing any smooth detuning looks linear and any smooth coupling looks constant — and nothing else matters, as the numerics will confirm. Diagonalizing at each instant — eigenvalues of $\vec b\cdot\vec\sigma$ are $\pm|\vec b|$ , Bloch vector $\vec b=\tfrac12(\Delta,0,\alpha t)$ — gives the adiabatic hyperbolas

E_\pm(t)=\pm\tfrac{1}{2}\sqrt{(\alpha t)^2+\Delta^2} \qquad (2)

where: the minimum separation is exactly $\Delta$ , at $t=0$ . Far away the adiabatic and diabatic states coincide — but which diabatic state is which adiabatic branch swaps across the crossing, and that swap is the whole drama. The instantaneous eigenbasis is reached by rotating the diabatic basis by the mixing angle $\theta(t)$ , whose rate of change drives transitions. Two related quantities must not be confused — our source conflated them, and the repair is instructive:

\tan\theta(t)=\frac{\Delta}{\alpha t},\qquad \dot\theta=-\frac{\alpha\Delta}{(\alpha t)^2+\Delta^2},\qquad \Big\langle +\Big|\frac{\mathrm{d}}{\mathrm{d}t}\Big|-\Big\rangle=\frac{\dot\theta}{2} \qquad (3)

where: $\dot\theta$ is the honest derivative of $\arctan(\Delta/\alpha t)$ — a Lorentzian pulse of width $\Delta/|\alpha|$ centred on the crossing — while the matrix element coupling the adiabatic states carries an extra $1/2$ , because eigenvectors of a spin- $1/2$ -like Hamiltonian rotate through half the Bloch angle. The exported base of this article printed the half-angle coupling and labelled it $\dot\theta$ — a factor-of-two slip, machine-checked against the symbolic derivative; downstream results were unaffected. Keep the ledger straight: $\dot\theta$ rotates the basis, $\dot\theta/2$ drives the transition. Everything collapses into one dimensionless number,

\Gamma=\frac{\Delta^2}{4\hbar\,|\alpha|} \qquad (4)

where: $\Gamma\gg1$ is adiabatic, $\Gamma\ll1$ diabatic — (dwell time $\Delta/|\alpha|$ ) × (gap frequency $\Delta/\hbar$ ), up to the factor 4. The claim to be established, then measured:

P_{\mathrm{LZ}}=e^{-2\pi\Gamma}=\exp\!\left(-\frac{\pi\Delta^2}{2\hbar\,|\alpha|}\right),\qquad P_{\mathrm{adiabatic}}=1-e^{-2\pi\Gamma} \qquad (5)

where: $P_{\mathrm{LZ}}$ is the probability, after a complete sweep, of remaining in the initial diabatic state — of failing to follow the eigenstate. At $\Gamma=1$ it is already $e^{-2\pi}=1.87\times10^{-3}$ : near-perfect adiabaticity arrives at modest $\Gamma$ , which is why rapid passage is so forgiving.

Why avoided crossings are the generic case

For a real symmetric two-level Hamiltonian with diagonal $a,b$ and off-diagonal $c$ , the gap obeys $(E_+-E_-)^2=(a-b)^2+4c^2$ — a sum of two squares (machine-verified). Closing it needs both $a=b$ and $c=0$ : two conditions, while a sweep tunes one parameter. Levels therefore generically repel — the von Neumann–Wigner non-crossing rule (1929); complex Hermitian families need three conditions. Whenever an experiment tunes two coupled levels toward degeneracy, Eq. (1) awaits at the bottom — one formula for molecules, spins, qubits and lattice bands. At an exact symmetry-protected degeneracy the rule is evaded because the coupling vanishes identically; the right tool there is a symmetry-adapted basis, as Good Quantum Numbers and Symmetry-Adapted Bases builds it.

Deriving the jump probability

Derivation AThe fast-sweep exponent from one Gaussian integral

The goal is the leading transition probability for a fast sweep, from first-order time-dependent perturbation theory — and with it, the origin of the $\pi$ in the Landau–Zener exponent. The strategy: work in the diabatic basis, where the sweep lives on the diagonal and the constant coupling $\Delta/2$ is the perturbation; the first-order amplitude is then a single integral with a purely quadratic phase, a Fresnel–Gaussian integral in closed form. The answer must match the expansion of the exact $e^{-2\pi\Gamma}$ — a nontrivial lock on the coefficient.

Step 1. First-Order Transition Amplitudes and Probabilities supplies the master integral $c_f^{(1)}=-\tfrac{i}{\hbar}\int V_{fi}(t')\,e^{i\omega_{fi}t'}\mathrm{d}t'$ ; here the matrix element is the constant $\Delta/2$ and the phase generalizes to the running integral of the detuning:

a_2(\infty)=-\frac{i}{\hbar}\int_{-\infty}^{\infty}\frac{\Delta}{2}\,\exp\!\left[-\frac{i}{\hbar}\int_0^{t}\alpha t'\,\mathrm{d}t'\right]\mathrm{d}t=-\frac{i\Delta}{2\hbar}\int_{-\infty}^{\infty}e^{-i\alpha t^2/2\hbar}\,\mathrm{d}t

Why the quadratic phase: the diabatic splitting is $\alpha t$ , so the relative phase grows as $\alpha t^2/2\hbar$ . Fast phase growth away from $t=0$ means only the crossing's neighbourhood contributes — the sweep builds its own stationary-phase window of width $\sqrt{\hbar/|\alpha|}$ .

Step 2. Evaluate the Fresnel–Gaussian integral (contour rotated by $45°$ ; modulus checked symbolically in this article's verification script):

\int_{-\infty}^{\infty}e^{-i\alpha t^2/2\hbar}\,\mathrm{d}t=\sqrt{\frac{2\pi\hbar}{|\alpha|}}\;e^{-i\frac{\pi}{4}\mathrm{sgn}\,\alpha}

Step 3. Modulus squared:

\big|a_2(\infty)\big|^2=\frac{\Delta^2}{4\hbar^2}\cdot\frac{2\pi\hbar}{|\alpha|}=\frac{\pi\Delta^2}{2\hbar|\alpha|}=2\pi\Gamma

— the diabatic-basis transition probability, i.e. the deficit $1-P_{\mathrm{LZ}}$ , valid while small.

Step 4. Lock against the exact answer: $e^{-2\pi\Gamma}=1-2\pi\Gamma+\mathcal{O}(\Gamma^2)$ — the perturbative coefficient is exactly the Gaussian-integral $2\pi$ ; the exact result is the exponentiation of the first-order seed. The complex-time view says why: the saddle sits at $t_*=\pm i\Delta/|\alpha|$ , where $E_+=E_-$ ; with $t=i(\Delta/|\alpha|)s$ the gap integrand becomes $\Delta\sqrt{1-s^2}$ , integral $\pi/4$ , so $\mathrm{Im}\,\phi(t_*)/\hbar=\pi\Delta^2/(4\hbar|\alpha|)$ and $|c|^2\sim e^{-2\,\mathrm{Im}\phi/\hbar}$ reproduces the full exponent. Verified: the Fresnel modulus and $2\pi\Gamma=\pi\Delta^2/(2\hbar|\alpha|)$ confirmed symbolically (sympy); the $\Gamma\to0$ tangency of $e^{-2\pi\Gamma}$ is visible in the laboratory table below.

Go deeper · Derivation B · 1 minThe exact formula by the Zener reduction

Derivation BThe exact formula by the Zener reduction

The goal is Zener's exact result: reduce the coupled two-level equations with a linear sweep to a single second-order ODE — Weber's parabolic-cylinder equation — and read the survival probability off its connection formulae. The strategy is elimination: solve one amplitude equation for $c_1$ , substitute into the other, and recognize the canonical equation. We are explicit about where rigor enters: every algebraic step is elementary and machine-checked, while the asymptotic connection of parabolic-cylinder functions is quoted with its source, because that is genuinely where the theorem lives.

Step 1. Write the diabatic amplitude equations for $|\psi\rangle=c_1|1\rangle+c_2|2\rangle$ under Eq. (1):

i\hbar\,\dot c_1=\frac{\alpha t}{2}c_1+\frac{\Delta}{2}c_2,\qquad i\hbar\,\dot c_2=\frac{\Delta}{2}c_1-\frac{\alpha t}{2}c_2

Step 2. Eliminate $c_1=\tfrac{2}{\Delta}\big(i\hbar\dot c_2+\tfrac{\alpha t}{2}c_2\big)$ from the second equation; substituting into the first yields

\hbar^2\,\ddot c_2+\left(\frac{\alpha^2t^2}{4}+\frac{\Delta^2}{4}-\frac{i\hbar\alpha}{2}\right)c_2=0

Why the imaginary term: differentiating the time-dependent diagonal energy produces $-i\hbar\alpha/2$ — the algebraic fingerprint of the sweep, absent for any static Hamiltonian. The elimination is verified symbolically in the verification script.

Step 3. Rescale to $\tau=\sqrt{|\alpha|/\hbar}\;t$ :

\frac{\mathrm{d}^2c_2}{\mathrm{d}\tau^2}+\left(\frac{\tau^2}{4}+\Gamma-\frac{i}{2}\,\mathrm{sgn}\,\alpha\right)c_2=0

with $\Gamma=\Delta^2/(4\hbar|\alpha|)$ the only surviving parameter — the model's entire physics in one number, already at the level of the ODE.

Step 4. Recognize Weber's equation $y''+(\tfrac{\tau^2}{4}-a)y=0$ with $a=-\Gamma+\tfrac{i}{2}\mathrm{sgn}\,\alpha$ ; after the standard rotation $\tau\to\tau e^{-i\pi/4}$ the solutions are parabolic-cylinder functions $D_\nu$ with $\nu=-i\Gamma$ . Why this is the honest hard step: $D_\nu(z)$ has different asymptotics in different complex sectors (the Stokes phenomenon); connecting the two infinities needs the sector-crossing formula, quoted rather than re-derived (Whittaker–Watson §16.5; Zener 1932): $D_\nu(z)\sim z^\nu e^{-z^2/4}$ plus a subdominant piece weighted by $\sqrt{2\pi}/\Gamma_{\!f}(-\nu)$ in the next sector.

Step 5. Impose $|c_1|=1$ , $c_2=0$ at $\tau\to-\infty$ and propagate. The surviving amplitude's modulus involves $|e^{-\pi\nu/2}|$ continued at $\nu=-i\Gamma$ ; the gamma-function factors contribute only phases, and

|c_1(+\infty)|^2=e^{-2\pi\Gamma}

is the entire answer — all Stokes-sector bookkeeping collapsing to one clean exponential, the special grace of the linear-sweep model. Verified: reproduces Zener (1932), Sakurai & Napolitano §5.6; elimination and rescaling checked symbolically; the formula is confirmed to $1.8\times10^{-7}$ absolute by direct numerical integration below, including $\Gamma=3$ where $P_{\mathrm{LZ}}=6.5\times10^{-9}$ .

The laboratory: formula versus Schrödinger equation

An exact claim invites an exact test. We integrated the Schrödinger equation for Eq. (1) with a high-order adaptive integrator ( $\hbar=\alpha=1$ , window $|t|\le\max(150,60\Delta)$ ), preparing the lower adiabatic eigenstate at the start and projecting on the upper adiabatic branch at the end — the state a perfect diabatic jumper lands on. Seven values of $\Gamma$ across three decades:

$\Gamma$	$\Delta$	$P_{\mathrm{LZ}}$ (TDSE)	$e^{-2\pi\Gamma}$	abs. error
0.003	0.110	0.9813270	0.9813270	$8\times10^{-9}$
0.01	0.200	0.9391013	0.9391014	$2\times10^{-8}$
0.03	0.346	0.8282041	0.8282042	$7\times10^{-8}$
0.1	0.632	0.5334879	0.5334881	$2\times10^{-7}$
0.3	1.095	0.1518360	0.1518358	$2\times10^{-7}$
1.0	2.000	$1.867\times10^{-3}$	$1.867\times10^{-3}$	$5\times10^{-8}$
3.0	3.464	$6.453\times10^{-9}$	$6.512\times10^{-9}$	$6\times10^{-11}$

The worst absolute deviation is $1.8\times10^{-7}$ — integrator tolerance, not physics. Two lessons ride along. First, at finite windows the diabatic-basis survival oscillates before settling (a residual $\sim\Delta/2\alpha t$ mixing with the receding level); projecting on the instantaneous adiabatic basis at the endpoints removes the contamination — how careful experiments define their measurement basis too. Second, the convention audit: fits against $e^{-\pi\Gamma}$ , $e^{-4\pi\Gamma}$ or $e^{-\pi\Gamma/2}$ fail by two orders of magnitude — the numerics pin both the $2\pi$ and the dictionary ( $\Delta$ = full gap, $\alpha$ = diabatic slope), with no room for folklore factor slips.

Two passages: Stückelberg interference

Sweep through the crossing and back, and it becomes a beam splitter traversed twice — a Mach–Zehnder interferometer in energy space. Each passage splits the amplitude; between passages the branches accumulate the relative phase $\Phi_{\mathrm{St}}=\tfrac1\hbar\int(E_+-E_-)\,\mathrm{d}t$ . The round-trip transfer interferes:

P_{\mathrm{net}}=4\,P_{\mathrm{LZ}}\big(1-P_{\mathrm{LZ}}\big)\,\sin^2\!\big(\Phi_{\mathrm{St}}+\varphi_{\mathrm{S}}\big) \qquad (6)

where: the envelope $4P(1-P)$ is the two-path bound — maximal for 50/50 splitting — and $\varphi_{\mathrm{S}}$ is the passage-dependent Stokes phase, which shifts but does not modulate the fringes. In our double-sweep numerics at $\Gamma=0.1$ per passage ( $P_{\mathrm{LZ}}=0.5335$ ), scanning the return time swept the fringe from $0.012$ to $0.994$ — 86-fold contrast — against the envelope prediction $0.9955$ , matched at the crests to 0.2%. Driving a qubit periodically through its crossing turns this into Landau–Zener–Stückelberg–Majorana interferometry: a self-calibrating spectroscopy of the two-level system.

Limiting cases

case	result	meaning
slow sweep, $\Gamma\gg1$	$P_{\mathrm{LZ}}\to0$	adiabatic following — the exponentially forgiving regime of rapid passage
fast sweep, $\Gamma\ll1$	$P_{\mathrm{LZ}}\to1-2\pi\Gamma$	diabatic jump; the deficit is first-order perturbation theory
decoupled levels, $\Delta\to0$	$P_{\mathrm{LZ}}\to1$	no repulsion — levels sail through each other, the bare X restored
frozen sweep, $\alpha\to0$	$P_{\mathrm{LZ}}\to0$	perfect adiabaticity; slow finite sweeps answer instead to the criterion of Adiabatic Approximation and Validity Conditions
double passage	$4P(1-P)\sin^2\Phi_{\mathrm{St}}$	crossings act as beam splitters; fringes replace monotone exponentials

Worked examples

Example 1 — sweeping a superconducting qubit. A flux qubit's avoided crossing has gap $\Delta/h=100\ \mathrm{MHz}$ ; the bias sweeps the diabatic splitting at rate $\mathrm{d}\nu/\mathrm{d}t$ . Where is the adiabatic/diabatic boundary?

Step 1. Convert to lab units. With $\Delta=h\,\Delta\nu$ and $\alpha=h\,(\mathrm{d}\nu/\mathrm{d}t)$ : $\Gamma=\Delta^2/(4\hbar\alpha)=\tfrac{\pi}{2}\,(\Delta\nu)^2/(\mathrm{d}\nu/\mathrm{d}t)$ — frequencies only, no $\hbar$ left.

Step 2. Three regimes. At $10^{14}\ \mathrm{Hz/s}$ : $\Gamma=157$ , $P_{\mathrm{LZ}}\approx10^{-429}$ — perfect passage. At $10^{16}$ : $\Gamma=\pi/2$ , $P_{\mathrm{LZ}}=e^{-\pi^2}=5.2\times10^{-5}$ — still $99.995\%$ transfer. At $10^{17}$ : $\Gamma=0.157$ , $P_{\mathrm{LZ}}=\mathbf{0.373}$ — a beam splitter.

Step 3. Tolerance boundary: $P_{\mathrm{LZ}}\le10^{-4}$ needs $2\pi\Gamma\ge9.21$ , so $\mathrm{d}\nu/\mathrm{d}t\le\tfrac{\pi}{2}(10^8)^2/1.466=\mathbf{1.07\times10^{16}\ Hz/s}$ : a 20 GHz excursion may be swept in about $2\ \mu\mathrm{s}$ and no faster. The exponential makes the boundary razor sharp — 10× faster turns $10^{-4}$ into $0.37$ .

Example 2 — the Na + Cl harpoon crossing. The covalent and ionic curves cross where Coulomb attraction pays the transfer cost: $R_c=e^2/(4\pi\varepsilon_0\,\Delta E_0)$ with $\Delta E_0=5.139-3.612=1.527\ \mathrm{eV}$ , so $R_c=14.40/1.527=\mathbf{9.4\ \text{Å}}$ . Does a thermal collision hop?

Step 1. Sweep rate: the slope difference at $R_c$ is Coulomb-dominated, $14.40/9.43^2=0.162\ \mathrm{eV/\text{Å}}$ ; at $v_{\mathrm{nuc}}\approx1000\ \mathrm{m/s}=0.01\ \text{Å/fs}$ , $\alpha=1.6\times10^{-3}\ \mathrm{eV/fs}$ .

Step 2. Representative coupling at $9\ \text{Å}$ : $\Delta\approx0.03\ \mathrm{eV}$ . With $\hbar=0.658\ \mathrm{eV\cdot fs}$ : $\Gamma=(0.03)^2/(4\times0.658\times1.6\times10^{-3})=\mathbf{0.21}$ .

Step 3. Branching: $P_{\mathrm{LZ}}=e^{-2\pi\times0.21}=\mathbf{0.27}$ — a quarter of trajectories shoot through covalently, three quarters hand off the electron. Honest caveat: $\Delta(R_c)$ varies by an order of magnitude across systems and enters squared in an exponent — the formula's job is the scaling, and the scaling is exact.

Origins · 1 minClarence Zener and how this was discovered

Clarence Zener

1905–1993 · Bristol, Pittsburgh

Zener is the rare physicist whose name marks both a fundamental formula and a component in your power supply. Indianapolis-born, Harvard PhD at 24, he wrote "Non-adiabatic crossing of energy levels" in 1932 as a research fellow at Bristol's H.H. Wills Laboratory — reducing the swept two-level problem to the parabolic-cylinder equation and extracting the exact exponential, while in that same year Landau's complex-plane attack found the exponent, Stückelberg organized the WKB phases, and Majorana solved the spin-flip version. Two years later Zener reused his own exponential for electrons tunnelling across a band gap in a strong field — "Zener breakdown", the mechanism that, harnessed in reverse-biased diodes, made his name a circuit element. Then he walked away from quantum theory: at Westinghouse and Carnegie Tech he founded the modern theory of anelasticity (the "Zener model" of internal friction), invented geometric programming, and consulted on everything from thermoelectrics to desalination. Four fields carry his name because he treated none of them as home.

History

1929

— von Neumann and Wigner prove generic levels avoid crossing: degeneracy demands two conditions, a one-parameter sweep supplies one

1932

— Landau, Zener, Stückelberg and Majorana independently solve the swept crossing; Zener's parabolic-cylinder treatment delivers the exact exponential

1934

— Zener transplants the same exponential to electrons in strong fields: interband "Zener breakdown", later domesticated as the Zener diode

1961

— Adiabatic rapid passage, long an NMR workhorse, is codified in texts as the practical face of the slow-sweep limit

2005

— Strongly driven superconducting qubits show Landau–Zener–Stückelberg fringes, turning repeated crossings into calibration spectroscopy

Going further

References. Landau, L. D., Phys. Z. Sowjetunion 2, 46 (1932) · Zener, C., Proc. R. Soc. London A 137, 696 (1932) · Stückelberg, E. C. G., Helv. Phys. Acta 5, 369 (1932) · Majorana, E., Nuovo Cimento 9, 43 (1932) · Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §5.6 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, complement on level crossings · Shankar, Principles of Quantum Mechanics, 2nd ed. · Wittig, C., J. Phys. Chem. B 109, 8428 (2005) — derivation review

Provenance. All equations are standard Landau–Zener results. A verifier should: (1) diagonalize $H(t)$ to confirm $E_\pm=\pm\tfrac12\sqrt{(\alpha t)^2+\Delta^2}$ ; (2) reduce the coupled amplitude equations to the parabolic-cylinder (Weber) equation and confirm the connection formula yields $|c_1(\infty)|^2=e^{-2\pi\Gamma}$ with $\Gamma=\Delta^2/(4\hbar|\alpha|)$ (Zener 1932; Sakurai & Napolitano §5.6; Cohen-Tannoudji level-crossing complement); (3) verify the Fresnel integral $\int e^{-i\alpha t^2/2\hbar}\mathrm{d}t=\sqrt{2\pi\hbar/|\alpha|}\,e^{-i\pi/4}$ and the perturbative result $2\pi\Gamma=\pi\Delta^2/(2\hbar|\alpha|)$ ; (4) check limiting cases $\Gamma\to0\Rightarrow P_{\mathrm{LZ}}\to1$ and $\Gamma\to\infty\Rightarrow P_{\mathrm{LZ}}\to0$ , and the numeric $e^{-2\pi}\approx1.87\times10^{-3}$ . No fabricated constants are used.

Builds on: The Interaction Picture and Time-Dependent Hamiltonians · Time-Dependent Perturbation Theory: General Formalism · First-Order Transition Amplitudes and Probabilities · Transition Probability for a Periodic Two-Level System Leads to: Semiclassical Theory of Atom–Radiation Interaction · Harmonic (Sinusoidal) Perturbations and Resonance

Is the Landau–Zener formula exact or an approximation?

Both. For the idealized model — two levels, exactly linear diabatic sweep from $-\infty$ to $+\infty$ , constant coupling — it is mathematically exact; our numerics confirm to $2\times10^{-7}$ . For a real system its quality depends only on how well the crossing region matches that model: the transition happens in a window $\sim\max(\Delta/|\alpha|,\sqrt{\hbar/|\alpha|})$ , so local linearity and local constancy are all that matter. Level curvature far away is irrelevant.

Which probability does the formula give — jumping or following?

$P_{\mathrm{LZ}}=e^{-2\pi\Gamma}$ is the probability of ending in the same diabatic state you started in — crossing the gap. In adiabatic language the same event is a failure to follow the eigenstate, so the adiabatic-transfer probability is $1-e^{-2\pi\Gamma}$ . Interchanging these is the most common applied error; fix the basis first, then quote the number.

Where exactly does the transition happen?

Within the crossing window $|t|\lesssim\Delta/|\alpha|$ , where the non-adiabatic coupling $\dot\theta/2$ — a Lorentzian of exactly that width — is appreciable. This locality is why the idealized infinite sweep describes finite experiments: many gaps away, amplitudes only accumulate phase. It is also why finite-window numerics must measure in the instantaneous eigenbasis — the residual mixing angle decays slowly, like $\Delta/(\alpha t)$ , long after transitions cease.

Why does a perturbative calculation produce a non-perturbative exponential?

First-order theory gives $2\pi\Gamma$ ; the exact answer $e^{-2\pi\Gamma}$ expands to exactly that. The exponentiation is special to the linear sweep — higher orders organize into the exponential series with no extra structure. The deeper reason lives in complex time: the amplitude is controlled by the branch point at $t_*=i\Delta/|\alpha|$ where the adiabatic gap closes, and the imaginary action to that point is the exponent. Swept-gap jumps are tunnelling events in the time domain.

What happens at an exact crossing where the gap really is zero?

If symmetry forces the coupling to vanish identically, the levels genuinely cross — von Neumann–Wigner evaded, not violated. Each symmetry sector evolves independently and there is nothing to compute; the right tool is a symmetry-adapted basis, as Good Quantum Numbers and Symmetry-Adapted Bases develops. The Landau–Zener question reappears the moment any perturbation reopens a small gap — and since $\Delta$ enters squared in an exponent, a tiny symmetry-breaking coupling can flip the outcome at slow sweeps.

How is Gamma related to the general adiabaticity criterion?

The general criterion compares $\hbar|\langle m|\dot H|n\rangle|$ to the squared gap — the dimensionless $\eta$ of Adiabatic Approximation and Validity Conditions. For the linear sweep at the crossing, $\langle m|\dot H|n\rangle=\alpha/2$ and the gap is $\Delta$ , so $\eta\sim\hbar\alpha/\Delta^2\sim1/\Gamma$ : the LZ adiabaticity is the reciprocal of the worst-case $\eta$ , evaluated where the criterion is tightest. The exponential is the quantitative payoff the inequality could only gesture at.