Optical Lattices

Take the single-particle quantum mechanics of a particle in a periodic potential — Bloch's theorem, energy bands, forbidden gaps — and ask what would happen if the crystal were made not of ions bonded by chemistry but of light. That is an optical lattice. It borrows the conservative dipole force established for far-off-resonance traps: two coherent laser beams interfere, their intensity varies as $\cos^2(kz)$ with period $\lambda/2$ , and through the second-order AC Stark shift this bright-and-dark pattern becomes a periodic potential-energy landscape for a neutral atom's centre of mass. Cool the atoms below the well depth — with optical molasses, evaporation, or by loading a Bose–Einstein condensate — and they localise into a defect-free array whose spacing is set by a wavelength, not by a bond length. The single-atom problem is then exactly Bloch's problem, with one honest energy scale: the recoil energy $E_R=\hbar^2k^2/2m$ , the kinetic energy an atom gains by absorbing one lattice photon. Shallow lattices give nearly free particles with small gaps; deep lattices give tight harmonic wells joined only by exponentially weak tunnelling. Switch on atom–atom interactions and a pretty single-particle problem becomes a paradigm of many-body physics: projected onto the lowest band, the gas obeys the Bose–Hubbard model with hopping $J$ and on-site repulsion $U$ . Because raising the laser power suppresses $J$ exponentially while barely changing $U$ , the single dial $U/J$ is swept across a genuine zero-temperature quantum phase transition — the superfluid-to-Mott-insulator transition, first seen with rubidium in 2002. Nearly every parameter is a knob, which is why lattices are the premier analog quantum simulators of condensed matter.

Key takeaways

An optical lattice is a periodic potential for neutral atoms built from interfering laser beams; the standing-wave intensity acts through the AC Stark (dipole) force, trapping atoms at intensity maxima for red detuning and at minima for blue detuning.
One retro-reflected beam gives a 1D lattice of period $\lambda/2$ and depth $V_0$ set by power and detuning; three orthogonal standing waves build a 3D crystal of light with sites spaced $\lambda/2$ .
Depths are quoted in recoil energies, $s=V_0/E_R$ with $E_R=\hbar^2k^2/2m$ ; the band structure follows from a Mathieu equation that depends on the single parameter $s$ .
Projected onto the lowest band the physics is the Bose–Hubbard model; the dimensionless ratio $U/J$ , tuned through the lattice depth, drives the superfluid–Mott-insulator quantum phase transition.
Because geometry, dimensionality, filling, and (via Feshbach resonances) the interaction are all controllable, optical lattices are the leading platform for quantum simulation of Hubbard-type models with bosons and fermions.
Key idea: a standing wave of light $\to$ a periodic dipole potential $V(z)=V_0\sin^2(kz)$ $\to$ Bloch bands on the recoil scale $E_R$ $\to$ (with interactions) the Bose–Hubbard model, whose tunable $U/J$ realises the superfluid–Mott-insulator transition.

The core insight

A standing wave of light $\to$ a periodic dipole potential $V(z)=V_0\sin^2(kz)$ $\to$ Bloch bands on the recoil-energy scale $E_R=\hbar^2k^2/2m$ $\to$ (with interactions) the Bose–Hubbard model, whose tunable ratio $U/J$ realises the superfluid–Mott-insulator quantum phase transition.

Where this lives today

The most spectacular modern instrument is the quantum-gas microscope: a high-numerical-aperture objective mounted over a two-dimensional lattice that resolves the fluorescence of individual atoms on individual sites, turning the abstract Bose- or Fermi–Hubbard model into a directly readable occupation map. With it, experiments now image antiferromagnetic spin order in the Fermi–Hubbard model, watch charge and spin transport site by site, and prepare bespoke initial states one atom at a time — probing exactly the strongly correlated regimes of the 2D Hubbard model that no classical computer can simulate at scale, the physics believed to underlie high-temperature superconductivity. At the other end of the application space sit optical-lattice clocks, today's most accurate measuring devices of any kind. Strontium or ytterbium atoms are held in a "magic-wavelength" lattice tuned so that the ground and excited clock states feel identical light shifts, cancelling the perturbation on the clock transition, while the tight confinement places the atoms in the Lamb–Dicke regime where first-order Doppler and recoil shifts vanish. The result is fractional frequency uncertainty at the $10^{-18}$ level and below — clocks that would neither gain nor lose a second over the age of the universe, and sensitive enough to measure the gravitational redshift across a millimetre of height. The same crystal of light that simulates a solid also keeps the world's best time.

The physical picture

Picture a corrugated egg-carton made of light. A red-detuned laser pulls atoms toward high intensity, so they settle in the bright fringes; a blue-detuned laser pushes them into the dark nodes — where, as a bonus, the light and hence the heating is minimal. The atom feels the time-averaged intensity, and because the fringe spacing is only $\lambda/2\approx0.5\,\mu\mathrm{m}$ , the crystal is micron-periodic and essentially perfect: no vacancies, no dislocations, no phonons of a host lattice, just atoms sitting in wells of light.

Figure

Two counter-propagating beams of wavelength λ = 1064 nm interfere into a standing wave whose intensity antinodes form a periodic potential V of z equal to V₀ sin²(kz) with lattice constant d = λ/2 = 532 nm; ⁸⁷Rb atoms of recoil energy E_R/h ≈ 2.0 kHz localise one per well in the deep-lattice limit

The recoil energy $E_R$ is the honest yardstick. A lattice a few $E_R$ deep is "shallow": atoms tunnel freely and a condensate stays superfluid. A lattice tens of $E_R$ deep is "deep": each well is a tight harmonic trap, tunnelling is frozen out exponentially, and the atoms number-lock into a Mott insulator. Turning the laser power thus continuously interpolates between free-particle and localised quantum matter — all in the same apparatus, over a few milliseconds. Interactions decide the ground state. With one atom per site, if hopping wins ( $J\gg U$ ) the atoms delocalise into a phase-coherent superfluid; if repulsion wins ( $U\gg J$ ) putting a second atom on an occupied site costs an energy $U$ , so the atoms lock to exactly one per site — an incompressible, gapped Mott insulator. The entire competition is captured by one dimensionless number, $U/J$ , and the second figure shows how it is dialled.

Figure

Superfluid-to-Mott boundary for ⁸⁷Rb at λ = 1064 nm and unit filling: raising the depth s = V₀/E_R makes U/J climb roughly as e to the 2√s, crossing the 3D mean-field threshold of 34.8 at s ≈ 13.5, so a shallow lattice is a superfluid and a deep one a Mott insulator

Everything here is a laboratory dial. Beam power sets the depth; beam geometry sets the crystal structure — cubic, triangular, honeycomb, kagome, superlattice; the number of beam pairs sets the dimensionality — 1D tubes, 2D pancakes, 3D crystal; and a magnetic Feshbach resonance sets the interaction $U$ independently of everything else. That total programmability is precisely what earns optical lattices the name analog quantum simulators.

The core equations

A single retro-reflected beam produces the simplest lattice. Superposing two counter-propagating waves of the same amplitude gives a standing wave whose intensity — the quantity the dipole force responds to — is modulated as $\cos^2(kz)$ , so the potential inherits that shape and repeats every half wavelength. Choosing the origin at a node casts it in the equivalent $\sin^2$ form.

V(z)=V_0\sin^2(kz)=\frac{V_0}{2}\big[1-\cos(2kz)\big] \qquad (1)

where: $V_0$ — lattice depth (peak-to-valley potential), fixed by intensity and detuning; $k=2\pi/\lambda$ — wavenumber of the lattice laser; $z$ — position along the standing-wave axis. Physical reading: the wells recur with period $\pi/k=\lambda/2$ , and because two beams of intensity $I_1$ interfere to give a peak intensity $4I_1$ , the depth is four times a single beam's Stark shift. The natural unit of energy for the atomic motion is not a temperature or a trap frequency but the recoil energy — the kinetic energy imparted by one absorbed lattice photon of momentum $\hbar k$ . Quoting depths as $s=V_0/E_R$ collapses the whole problem to a single dimensionless number.

E_R=\frac{\hbar^2 k^2}{2m}=\frac{h^2}{2m\lambda^2} \qquad (2)

where: $E_R$ — single-photon recoil energy; $m$ — atomic mass; $\hbar k$ — photon momentum; $s=V_0/E_R$ — reduced depth. Physical reading: for $^{87}\mathrm{Rb}$ at $\lambda=1064\,\mathrm{nm}$ , $E_R/h\approx2.0\,\mathrm{kHz}$ and $E_R/k_B\approx0.10\,\mu\mathrm{K}$ , so a "deep" lattice of $s=15$ is still only a fraction of a microkelvin deep. The dipole potential and the spontaneous scattering rate both come from the driven-atom polarisability, but they scale differently with the detuning $\Delta$ . This single fact is what makes conservative lattices possible: the useful trap falls only as $1/\Delta$ while the harmful heating falls as $1/\Delta^2$ .

U_{\mathrm{dip}}(\mathbf r)=\frac{3\pi c^2}{2\omega_0^3}\frac{\Gamma}{\Delta}\,I(\mathbf r),\qquad \Gamma_{\mathrm{sc}}(\mathbf r)=\frac{3\pi c^2}{2\hbar\omega_0^3}\Big(\frac{\Gamma}{\Delta}\Big)^2 I(\mathbf r) \qquad (3)

where: $\Delta=\omega-\omega_0$ — detuning from resonance $\omega_0$ ; $\Gamma$ — natural linewidth; $\Gamma_{\mathrm{sc}}$ — photon scattering (heating) rate. Physical reading: the ratio $\Gamma_{\mathrm{sc}}/(U_{\mathrm{dip}}/\hbar)=\Gamma/\Delta$ shrinks without bound at large detuning, so one works far off resonance and compensates with more power. A single atom in the potential (1) obeys a stationary Schrödinger equation with a sinusoidal potential. This is one of the exactly solvable periodic problems of quantum mechanics — it is Mathieu's equation in disguise — and its solutions organise into energy bands.

-\frac{\hbar^2}{2m}\frac{\mathrm{d}^2\psi}{\mathrm{d}z^2}+V_0\sin^2(kz)\,\psi=E\,\psi \qquad (4)

where: $\psi$ — single-atom centre-of-mass wavefunction; $E$ — energy eigenvalue; the spectrum splits into bands $E_n(q)$ . Physical reading: this is the atomic analogue of an electron in a crystal, but with a potential known to full precision because it is drawn by lasers. Because the potential is periodic with lattice constant $d=\lambda/2$ , Bloch's theorem applies verbatim: every eigenstate is a plane wave modulated by a lattice-periodic envelope, labelled by a band index and a quasimomentum confined to the first Brillouin zone.

\psi_{n,q}(z)=e^{iqz}\,u_{n,q}(z),\qquad u_{n,q}(z+d)=u_{n,q}(z),\ d=\lambda/2 \qquad (5)

where: $q$ — quasimomentum in the zone $[-\pi/d,\pi/d]$ ; $n$ — band index; $u_{n,q}$ — lattice-periodic Bloch function. Physical reading: the maximally localised superpositions of Bloch states in a band are the Wannier functions, one per site, which become the natural single-particle basis once interactions enter. Projecting the interacting many-body Hamiltonian onto the lowest band and the Wannier basis yields the Bose–Hubbard model — the minimal Hamiltonian of interacting lattice bosons, in which only nearest-neighbour hopping and on-site interaction survive.

\hat H=-J\!\!\sum_{\langle i,j\rangle}\! b_i^\dagger b_j+\frac{U}{2}\sum_i \hat n_i(\hat n_i-1)-\mu\sum_i \hat n_i \qquad (6)

where: $b_i^\dagger,b_i$ — bosonic operators for the lowest Wannier state at site $i$ ; $\hat n_i=b_i^\dagger b_i$ ; $J$ — nearest-neighbour hopping; $U$ — on-site two-body interaction; $\mu$ — chemical potential fixing the mean filling. Physical reading: $J$ favours delocalisation and phase coherence, $U$ favours number localisation, and their ratio decides the phase. The two Hubbard parameters are not free — they are integrals of the Wannier functions over the lattice potential and the interaction. Their opposite dependence on depth is the entire mechanism of the transition.

J=-\!\int\! w^*(x-x_i)\Big[-\frac{\hbar^2}{2m}\nabla^2+V(x)\Big]w(x-x_j)\,\mathrm dx,\quad U=\frac{4\pi\hbar^2 a_s}{m}\!\int\!|w(x)|^4\mathrm dx \qquad (7)

where: $w(x)$ — maximally localised Wannier function of the lowest band; $a_s$ — s-wave scattering length. Physical reading: $J$ is an overlap of neighbouring wells, exponentially small in a deep lattice, whereas $U$ tracks the on-site density $|w|^4$ and grows only algebraically — so $U/J$ climbs steeply with depth. The competition has a threshold. In mean field, the superfluid gives way to the $n=1$ Mott insulator when the interaction overwhelms the total kinetic bandwidth $zJ$ set by the coordination number.

\Big(\frac{U}{zJ}\Big)_c\approx 5.8\quad(n=1,\ \text{mean field}),\qquad z=2D \qquad (8)

where: $z$ — coordination number ( $z=6$ in a 3D simple-cubic lattice); $n$ — integer filling of the Mott lobe; $D$ — spatial dimension. Physical reading: for 3D this is $(U/J)_c\approx34.8$ ; quantum Monte Carlo sharpens it to $(U/J)_c\approx29.3$ , the value modern experiments confirm.

Derivation ABand structure from Mathieu's equation and the recoil scale

The goal of this derivation is to show that the single-atom problem in a sinusoidal lattice is not merely similar to a solved problem but is exactly Mathieu's equation, and that when every energy is measured in recoil units the entire band structure collapses to a one-parameter family labelled only by the reduced depth $s=V_0/E_R$ . We rescale position to the lattice wavenumber, insert the recoil energy explicitly, and use a trigonometric identity to bring the Schrödinger equation into canonical Mathieu form, from which the bands and gaps follow.

Step 1. Rescale position by the lattice wavenumber, $\xi=kz$ , so that $\mathrm d^2/\mathrm dz^2=k^2\,\mathrm d^2/\mathrm d\xi^2$ and one full period $\lambda/2$ becomes $\xi\in[0,\pi]$ . Why: measuring length in units of $1/k$ removes the wavelength from the equation, so the geometry enters only through dimensionless combinations.

Step 2. Substitute into equation (4) and pull out the recoil energy $E_R=\hbar^2k^2/2m$ , giving $-E_R\,\mathrm d^2\psi/\mathrm d\xi^2+V_0\sin^2\xi\,\psi=E\psi$ . Why: every term now carries $E_R$ as its natural energy unit, which is exactly why depths are quoted as $s=V_0/E_R$ rather than in joules or kelvin.

Step 3. Use $\sin^2\xi=\tfrac12(1-\cos2\xi)$ and divide through by $E_R$ to reach $\mathrm d^2\psi/\mathrm d\xi^2+\big[E/E_R-s/2+(s/2)\cos2\xi\big]\psi=0$ , the standard Mathieu equation $y''+[a-2q_M\cos2\xi]y=0$ with $a=4E/E_R-2s$ and $q_M=s/2$ . Why: the identification is exact — no approximation has been made — so the lattice inherits the full, tabulated Mathieu spectrum.

Step 4. Read off the physics: the characteristic values of the Mathieu equation are the band edges $E_n(q)$ , with $q$ entering through the Floquet exponent, and since $a$ and $q_M$ depend only on $s$ , one number fixes the whole band structure. Why: this is what makes the lattice universal — two experiments at different wavelengths and masses with the same $s$ have identical band structures in recoil units. Verified: as $s\to0$ the bands reduce to the free-particle parabolas $E=\hbar^2(q+2\ell k)^2/2m$ folded into the zone; as $s\to\infty$ they become harmonic levels $E_n=(n+\tfrac12)\hbar\omega$ with $\hbar\omega=2\sqrt{s}\,E_R$ . Cross-check: Bloch, Dalibard and Zwerger, Rev. Mod. Phys. 80, 885 (2008).

Go deeper · Derivation B · 1 minThe deep-lattice tunnelling amplitude

Derivation BThe deep-lattice tunnelling amplitude

The goal here is to establish the workhorse formula for how strongly atoms hop between neighbouring wells, and in particular to expose the exponential suppression that is the engine of the superfluid–Mott transition. We approximate a deep well as harmonic to fix the on-site oscillator, recognise the lowest-band dispersion as a simple cosine whose full width is $4J$ , and then use a WKB estimate of the tunnelling through the barrier to assemble the closed-form hopping amplitude $J(s)$ .

Step 1. Expand $V_0\sin^2(kz)$ about a minimum: $V(z)\approx V_0k^2z^2=\tfrac12 m\omega^2z^2$ with $\hbar\omega=2\sqrt{s}\,E_R$ , a local harmonic oscillator whose frequency grows as $\sqrt{s}$ . Why: in a deep lattice each atom is nearly a harmonic-trap ground state, and the oscillator length sets the width of the Wannier function that appears in the overlap integral for $J$ .

Step 2. For an isolated low band the exact dispersion is a single cosine, $E_0(q)\approx\bar E_0-2J\cos(qd)$ , so the total bandwidth from zone centre to zone edge is $4J$ . Why: tight-binding on a chain with one orbital per site can only give a cosine band, which lets us read $J$ directly off the band width instead of evaluating the overlap integral by brute force.

Step 3. Estimate the tunnel matrix element between adjacent Gaussian ground states through the barrier of height $V_0$ : the WKB action across the barrier gives the exponential factor $e^{-2\sqrt{s}}$ , while the harmonic normalisation and barrier curvature supply the algebraic prefactor $s^{3/4}$ . Why: the exponential dominance means a modest change in depth changes $J$ by orders of magnitude — the sensitivity that makes $U/J$ a precision knob.

Step 4. Assemble $J=\tfrac14[E_0(0)-E_0(\pi/d)]=\dfrac{4}{\sqrt\pi}\,E_R\,s^{3/4}\,e^{-2\sqrt{s}}$ , identifying $J$ as one quarter of the band width. Why: this closed form is the analytic approximation used throughout the field, accurate to a few percent against exact Mathieu bands once $s\gtrsim15$ and good to tens of percent already by $s\approx10$ . Verified: numerical diagonalisation of the Mathieu bands reproduces the scaling and the leading coefficient; standard result in Bloch–Dalibard–Zwerger, RMP 80, 885 (2008) and Zwerger, J. Opt. B 5, S9 (2003).

Go deeper · Derivation C · 1 minWhy the lattice depth sweeps U over J across the transition

Derivation CWhy the lattice depth sweeps U over J across the transition

The goal of this final derivation is to combine the tunnelling and interaction scalings into the single control parameter $U/J$ and to show that it grows essentially exponentially in $\sqrt{s}$ , so that a smooth, slow ramp of laser power drives the gas through a genuine quantum phase transition at a well-defined depth. We take the depth dependence of each Hubbard parameter, form the ratio, and set it equal to the mean-field threshold to predict the critical depth.

Step 1. From the tunnelling derivation, $J=\dfrac{4}{\sqrt\pi}E_R\,s^{3/4}e^{-2\sqrt{s}}$ ; from the on-site integral of $|w|^4$ over the harmonic ground state, $U\approx\sqrt{8/\pi}\,k a_s\,E_R\,s^{3/4}$ . Why: both carry the same $s^{3/4}$ from the Wannier width, so that algebraic factor cancels in the ratio and only the exponential survives.

Step 2. Form the ratio: $\dfrac{U}{J}=\dfrac{\sqrt2}{2}\,k a_s\,e^{2\sqrt{s}}\propto\dfrac{a_s}{d}\,e^{2\sqrt{s}}$ . Why: the interaction strength enters only through the fixed material ratio $a_s/d$ , while the depth enters through a steep exponential — this is why $U/J$ can be tuned over three orders of magnitude with a factor-of-a-few change in power.

Step 3. Set $U/J$ equal to the 3D mean-field threshold $(U/J)_c=z\times5.8=34.8$ and solve for $s$ ; for $^{87}\mathrm{Rb}$ at $1064\,\mathrm{nm}$ ( $a_s\approx100\,a_0$ , so $ka_s\approx0.031$ ) this gives $s_c\approx13.5$ , and the sharper QMC threshold $29.3$ gives $s_c\approx12.9$ . Why: the predicted critical depth of order ten recoils, insensitive to the exact threshold because of the exponential, is exactly what the first Mott experiment observed. Verified: the exponential ratio and $s_c\approx13$ are consistent with Greiner et al., Nature 415, 39 (2002), who saw the loss of interference near $V_0\approx13\,E_R$ ; QMC threshold from Capogrosso-Sansone et al., Phys. Rev. A 75, 013619 (2007).

Limiting cases

The reduced depth $s=V_0/E_R$ organises every regime, and the interaction ratio $U/J$ selects the many-body ground state within them.

Case	Result	Meaning
Shallow lattice $s\ll1$	$E(q)\approx\dfrac{\hbar^2q^2}{2m}$ , gaps $\sim V_0/2$ at the zone edge	Nearly free particles; small gaps open only at the Brillouin-zone boundary, atoms tunnel easily, and a condensate stays superfluid with coherence across the whole lattice.
Deep lattice $s\gg1$	$J\approx\dfrac{4}{\sqrt\pi}E_R\,s^{3/4}e^{-2\sqrt{s}}$ , $\hbar\omega=2\sqrt{s}\,E_R$	Tight-binding regime: each site is a harmonic trap, tunnelling collapses exponentially, and the physics reduces to the lowest Wannier band.
Non-interacting $U\to0$	$\hat H=-J\!\sum_{\langle i,j\rangle}b_i^\dagger b_j$	Pure tight-binding band; the ground state is a perfect superfluid — a BEC in the $q=0$ Bloch state at any filling, dispersion $-2J\cos(qd)$ per axis.
Strong coupling $U/J\to\infty$	$(U/J)_c\approx34.8$ (3D MF), $\approx29.3$ (3D QMC), $\approx16.7$ (2D QMC)	At integer filling the Mott lobe forms; the threshold is lattice- and dimension-specific, and beyond it the gas is a number-locked incompressible insulator.

Common trapThe lattice period is λ, and the dipole force is the same thing as the scattering force

Both halves are wrong. The period is $\lambda/2$ , not $\lambda$ : two counter-propagating beams of wavelength $\lambda$ interfere to give an intensity varying as $\cos^2(kz)$ , and $\cos^2$ repeats every $\pi/k=\lambda/2$ , so the potential the atoms feel has half the optical wavelength. And the trapping force is the conservative dipole (gradient) force, proportional to $\nabla I/\Delta$ , an entirely different object from the dissipative scattering (radiation-pressure) force that cools and pushes in molasses: the lattice is built by the dipole force, while spontaneous scattering is only a heating nuisance, minimised by working far off resonance. Confusing the two leads to the false expectation that a redder or bluer detuning flips the trap the way it flips cooling — it changes the sign of $U_{\mathrm{dip}}$ through the sign of $\Delta$ , but that is a statement about the conservative potential, not about radiation pressure.

Worked examples

1 · The recoil energy of rubidium (confidence builder)

A $^{87}\mathrm{Rb}$ atom sits in a lattice made from $\lambda=1064\,\mathrm{nm}$ light. Find the lattice constant, the recoil energy in frequency and temperature units, and comment on what "deep" means.

Step 1. Lattice constant: the wells recur every half wavelength, $d=\lambda/2=1064/2=\mathbf{532\ nm}$ , so a centimetre of atomic cloud spans nearly twenty thousand sites — a large, clean crystal.

Step 2. Wavenumber $k=2\pi/\lambda=5.905\times10^{6}\,\mathrm{m^{-1}}$ ; with $m=87\times1.6605\times10^{-27}=1.445\times10^{-25}\,\mathrm{kg}$ , the recoil energy is $E_R=\hbar^2k^2/2m=1.34\times10^{-30}\,\mathrm{J}$ . Dividing by Planck's constant and by Boltzmann's constant gives $E_R/h=\mathbf{2.03\ kHz}$ and $E_R/k_B=\mathbf{97\ nK}$ .

Step 3. "Deep" is measured in these units. A lattice of $s=15$ has depth $V_0=15E_R\approx1.5\,\mu\mathrm{K}\times k_B$ and on-site trap frequency $\hbar\omega=2\sqrt{15}\,E_R\approx7.7E_R$ , i.e. $\omega/2\pi\approx16\,\mathrm{kHz}$ . The whole crystal is thus barely a microkelvin deep and vibrates at kilohertz frequencies — the reason ultracold temperatures below $E_R/k_B$ are a prerequisite for loading it. d = 532 nm, E_R/h = 2.03 kHz, E_R/k_B = 97 nK.

2 · Where does the Mott transition happen?

For the same $^{87}\mathrm{Rb}$ lattice ( $ka_s\approx0.031$ with $a_s=100\,a_0$ ), find the depth $s_c$ at which a unit-filled 3D lattice crosses from superfluid to Mott insulator.

Step 1. Use the depth-controlled ratio from the third derivation, $U/J=(\sqrt2/2)\,k a_s\,e^{2\sqrt{s}}=0.5\times0.031\times e^{2\sqrt{s}}=0.0221\,e^{2\sqrt{s}}$ . The $s^{3/4}$ factors have already cancelled, leaving a clean exponential in $\sqrt{s}$ .

Step 2. Set this equal to the 3D mean-field threshold $(U/J)_c=z\times5.8=6\times5.8=34.8$ : solving $0.0221\,e^{2\sqrt{s}}=34.8$ gives $e^{2\sqrt{s}}=1575$ , so $2\sqrt{s}=\ln1575=7.36$ and $\sqrt{s}=3.68$ , hence $\mathbf{s_c\approx13.5}$ .

Step 3. Cross-check against the sharper quantum Monte Carlo threshold $(U/J)_c=29.3$ : the same algebra gives $s_c\approx12.9$ . Both land near $s\approx13$ because the exponential makes $s_c$ almost insensitive to the exact threshold — and this is precisely the depth ( $V_0\approx13\,E_R$ ) at which Greiner et al. saw the interference peaks vanish in 2002. s_c ≈ 13, robust to the choice of critical ratio.

3 · Hubbard parameters and timescales at s = 15

Evaluate $J$ , $U$ , and $U/J$ for the $^{87}\mathrm{Rb}$ lattice at depth $s=15$ , and state the many-body phase and the relevant timescales.

Step 1. Hopping: $J=(4/\sqrt\pi)E_R\,s^{3/4}e^{-2\sqrt{s}}$ . With $s^{3/4}=15^{0.75}=7.62$ , $\sqrt{s}=3.873$ , and $e^{-7.746}=4.3\times10^{-4}$ , this is $J=2.257\times7.62\times4.3\times10^{-4}\,E_R=7.4\times10^{-3}\,E_R$ , i.e. $J/h\approx\mathbf{15\ Hz}$ .

Step 2. Interaction: $U=\sqrt{8/\pi}\,k a_s\,E_R\,s^{3/4}=1.596\times0.031\times7.62\,E_R=0.38\,E_R$ , i.e. $U/h\approx\mathbf{770\ Hz}$ . Their ratio is $U/J=0.38/0.0074\approx\mathbf{51}$ .

Step 3. Since $U/J\approx51$ exceeds the 3D threshold (about $29$ to $35$ ), the unit-filled gas is a Mott insulator. The tunnelling time $\hbar/J\approx11\,\mathrm{ms}$ is slow — coherence takes many milliseconds to build — while the on-site gap $U/h\approx0.77\,\mathrm{kHz}$ and the band gap $\hbar\omega/h\approx16\,\mathrm{kHz}$ set the fast scales; keeping the ramp adiabatic with respect to $J$ without exciting higher bands is the experimental tightrope. J ≈ 15 Hz, U ≈ 770 Hz, U/J ≈ 51 → Mott insulator.

Physics of the portrait

Optical lattices are where solid-state physics is rebuilt atom by atom, and one name recurs at almost every milestone of that rebuilding — a physicist who took a 1998 theoretical proposal and turned lattices into the standard toolbox of quantum simulation.

Origins · 1 minImmanuel Bloch and how this was discovered

Immanuel Bloch

born 1972 · ultracold atoms in optical lattices and quantum simulation

There is a pleasing coincidence in the name. Felix Bloch proved in 1929 that electrons in any periodic potential organise into energy bands — the theorem that governs every crystal, and every optical lattice. Seventy-three years later a different Bloch, Immanuel, used a crystal of light to make Felix Bloch's bands visible one atom at a time. Working in Theodor Hänsch's group in Munich, Immanuel Bloch and his student Markus Greiner loaded a $^{87}\mathrm{Rb}$ Bose–Einstein condensate into a three-dimensional optical lattice and, in 2002, ramped the depth to drive the superfluid-to-Mott-insulator transition that Jaksch, Zoller and collaborators had predicted in 1998 — watching the sharp matter-wave interference peaks dissolve into the incoherent cloud of a number-locked insulator and then reappear on the way back. It was the first quantum phase transition realised by turning a laser knob, and it opened the era of cold-atom quantum simulation. Bloch went on to co-author, with Dalibard and Zwerger, the field's canonical 2008 review, and his groups (Mainz, then Munich again) built quantum-gas microscopes that image single atoms on single lattice sites, realised artificial gauge fields and Hofstadter and Haldane bands, and probed antiferromagnetic correlations in the Fermi–Hubbard model. In his hands the optical lattice stopped being a curiosity of atomic physics and became a programmable analog computer for the many-body problems that defeat classical simulation.

History

1929

— Felix Bloch proves that electrons in a periodic potential form energy bands

1986

— Chu, Bjorkholm, Ashkin and Cable demonstrate the optical dipole trap for neutral atoms

1998

— Jaksch, Bruder, Cirac, Gardiner and Zoller propose realising the Bose–Hubbard model in an optical lattice

2002

— Greiner, Mandel, Esslinger, Hänsch and Bloch observe the superfluid–Mott-insulator quantum phase transition in ⁸⁷Rb

2009

— Bakr and coworkers build the first quantum-gas microscope, imaging single atoms on single lattice sites

Going further

Texts. Foot, Atomic Physics (2005), Ch. 9–10 (light forces, dipole trapping) · Metcalf and van der Straten, Laser Cooling and Trapping (1999), Ch. 7 · Grimm, Weidemüller and Ovchinnikov, Optical Dipole Traps for Neutral Atoms, Adv. At. Mol. Opt. Phys. 42, 95 (2000) · Bloch, Dalibard and Zwerger, Many-body physics with ultracold gases, Rev. Mod. Phys. 80, 885 (2008) · Jaksch, Bruder, Cirac, Gardiner and Zoller, Phys. Rev. Lett. 81, 3108 (1998) · Greiner, Mandel, Esslinger, Hänsch and Bloch, Nature 415, 39 (2002) · Saleh and Teich, Fundamentals of Photonics (2007), interference and standing waves.

Provenance. The forms of $V(z)$ , $E_R$ , the dipole-versus-scattering scalings, the Mathieu reduction, the tight-binding $J$ , and the Bose–Hubbard Hamiltonian follow Grimm et al. (2000), Bloch–Dalibard–Zwerger (RMP 80, 885, 2008), and Jaksch et al. (PRL 81, 3108, 1998). A verifier should confirm: (i) the standing-wave intensity is $4I_1\cos^2(kz)$ with period $\lambda/2$ ; (ii) the Schrödinger equation reduces to Mathieu form with $a=4E/E_R-2s$ , $q_M=s/2$ ; (iii) the deep-lattice $J=(4/\sqrt\pi)E_R\,s^{3/4}e^{-2\sqrt s}$ and $\hbar\omega=2\sqrt s\,E_R$ ; (iv) for $^{87}\mathrm{Rb}$ at $1064\,\mathrm{nm}$ , $E_R/h=2.03\,\mathrm{kHz}$ , $E_R/k_B=97\,\mathrm{nK}$ , $d=532\,\mathrm{nm}$ ; (v) the SF–Mott critical depth $s_c\approx13$ from $(U/J)_c\approx34.8$ (mean field) or $29.3$ (QMC). Numerical anchors (first Mott experiment 2002; magic-wavelength clock uncertainty near $10^{-18}$ ) are standard and cited.

Prerequisites. Dipole Force and the Optical Dipole Potential · Radiation Pressure and the Scattering Force · Optical Molasses · Bose-Einstein Condensation in Dilute Gases. Leads to. Degenerate Fermi Gases and the BEC–BCS Crossover · Evaporative Cooling.

Ask the tutor: Ask me to walk through the Mathieu band structure at your chosen depth $s$ , or to trace $U/J$ across the superfluid–Mott boundary for your atom, wavelength, and filling.

Why is the lattice period λ/2 and not λ?

The atoms respond to intensity, not to field amplitude, and the intensity of a standing wave repeats twice as fast as the field. Two counter-propagating beams of wavelength $\lambda$ give a field envelope proportional to $\cos(kz)$ , but the time-averaged intensity is proportional to $\cos^2(kz)$ , and $\cos^2$ has period $\pi/k=\lambda/2$ . So the dipole potential, which follows the intensity, has half the optical wavelength as its lattice constant — about $0.4$ to $0.5\,\mu\mathrm{m}$ for near-infrared light. It is a common slip to quote $\lambda$ ; always halve it.

What actually holds the atoms in place — is there a real crystal there?

There is no material crystal at all: the "crystal" is the interference pattern of light itself, and the atoms are held by the AC Stark shift, a second-order energy shift the light imposes on the atomic ground state. Where the intensity is high (red detuning) the ground state is pushed down in energy, so high-intensity regions are potential minima and atoms fall into them. Nothing is touching the atoms; the potential is entirely conservative. Because it is drawn by lasers, it has no defects, no impurities, and no thermal phonons of a host lattice — one of the reasons optical lattices are such clean model systems.

Red or blue detuning — where do the atoms end up?

The sign of the dipole potential follows the sign of the detuning $\Delta=\omega-\omega_0$ . Red detuning ( $\Delta<0$ , laser below resonance) makes $U_{\mathrm{dip}}<0$ at high intensity, so atoms are attracted to the bright antinodes. Blue detuning ( $\Delta>0$ ) reverses this: atoms are repelled from bright regions and sit at the dark nodes. Blue lattices have a practical advantage — trapped atoms live where the light is weakest, so photon scattering and its heating are minimised — but they require an additional envelope trap to provide overall confinement, since the intensity maxima now repel.

Why is the recoil energy, and not temperature, the natural unit?

Because casting the single-atom Schrödinger equation in recoil units reduces it to a Mathieu equation that depends only on $s=V_0/E_R$ — the mass, wavelength, and Planck's constant all disappear into $E_R$ . Two completely different atoms in two different lasers with the same $s$ have identical band structures measured in recoil energies. Temperature is a property of the gas you loaded, not of the lattice; $E_R$ is the intrinsic quantum of the problem, the kinetic energy of a single photon recoil, and every meaningful energy — depth, band gap, hopping, interaction — is naturally expressed as a multiple of it.

Is the critical ratio 5.8, 29, or 35 — which one is right?

They describe the same transition at different levels of rigour. The number $5.8$ is the mean-field (Gutzwiller) result for $(U/zJ)_c$ at unit filling; multiplied by the 3D coordination number $z=6$ it becomes $(U/J)_c\approx34.8$ . Mean field overestimates the ordered phase, and unbiased quantum Monte Carlo sharpens the 3D value to $(U/J)_c\approx29.3$ . In two dimensions the QMC threshold is about $16.7$ , and in one dimension the transition is of Kosterlitz–Thouless type at $(U/J)_c\approx3.4$ . Because $U/J$ depends exponentially on depth, all these thresholds correspond to nearly the same critical depth, which is why the mean-field estimate already predicts the observed $s_c\approx13$ .

How do you read out which bands and quasimomenta are occupied?

The standard tool is adiabatic band mapping. If the lattice is turned off slowly compared with the band gap but fast compared with tunnelling, an atom with quasimomentum $q$ in band $n$ is mapped smoothly onto a free-particle momentum lying in the $n$ -th Brillouin zone. A subsequent time-of-flight expansion converts momentum into position, so a single absorption image shows the population of each band as a distinct momentum square, and the shape within the lowest zone reveals the quasimomentum distribution. Sharp peaks at reciprocal-lattice vectors signal the phase coherence of a superfluid; a smooth incoherent background signals the Mott insulator.

What is the difference between an optical lattice and an optical tweezer array?

Both use the dipole force, but a lattice is an interference pattern — a single standing wave fixes the spacing at $\lambda/2$ , and every site is identical and coherently connected to its neighbours, so tunnelling $J$ is a natural part of the physics. A tweezer array is a set of individually focused, separately addressable dipole traps, typically microns apart, positioned by acousto-optic deflectors or spatial light modulators; there the sites are essentially isolated unless deliberately brought close. Lattices excel at many-body physics with well-defined tunnelling and huge atom numbers; tweezers excel at single-atom control, defect-free assembly, and quantum-computing architectures. Increasingly the two are combined.