The particle has energy to travel, yet its cloud stops spreading
A wave can fail to spread through a disordered material even when no individual obstacle is high enough to imprison it classically. This is the surprising content of Anderson localization. The obstruction comes from waves scattered along many paths: crests and troughs add or cancel when the paths meet again. It cannot be understood simply by counting how often a particle hits an impurity.
Anderson’s 1958 paper introduced randomness through the energies of lattice sites and asked whether quantum motion could carry an initially local disturbance away. Under appropriate conditions, the answer was no diffusion. That question was relevant to impurity-band conduction and spin transport, but the underlying wave mechanism is more general. A later cold-atom experiment made the distinction unusually visible: a matter-wave cloud initially expanded and then formed stationary exponential wings in a weak disordered potential. Understanding why that is evidence requires more than noticing that the cloud stopped.
Sources: [1][3]
Random site energies compete with tunnelling
A useful model places one orbital at each lattice site. A wave can tunnel between neighbours, passing through a barrier that would stop a classical particle, but each site has its own energy offset. If all offsets match, waves can extend coherently across the lattice. Random offsets change phases and the resonances connecting different sites; the outcome depends on disorder relative to tunnelling, not merely on whether an imperfection exists.
Both the energy offsets and the hopping strength have energy units, so their ratio measures how uneven the landscape is compared with the ability to move between sites. This model contains no repulsion between particles: disorder and wave interference are enough to obstruct motion. A very deep isolated well can also bind a particle, but through simple confinement. To find Anderson localization, experiments follow a wave through an extended region of weak disorder and look for the interference pattern of many scattered paths.
Sources: [1]
Why a longer random walk is an incomplete picture
In a classical random walk, independent scattering events change direction, and repeated steps usually produce a cloud whose width continues to grow. A quantum amplitude is the wave quantity whose squared magnitude gives probability. Its phase records where the wave is in its cycle. Two paths leading to the same final position must be added before a probability is calculated, so the probabilities cannot always be multiplied as though every scattering event were independent.
Repeated scattering can enhance return and suppress transport. The effect requires phase coherence over the paths being compared, so their relative wave phases remain stable. Inelastic collisions, a fluctuating environment or interactions can change that comparison. This explains why temperature and observation time matter in an electronic experiment. It also explains why a disordered sample’s finite room-temperature resistance is not by itself a demonstration of strong localization. The experiment must distinguish ordinary scattering with slow diffusion from a state that remains spatially localized over the accessible length and time scales.
Sources: [2][3]
A semi-logarithmic plot exposes the length scale
A localized state has an approximately exponential envelope away from its centre. The localization length tells how far its amplitude extends. If the amplitude falls exponentially with one length, the density, which is its squared magnitude, falls twice as rapidly in the exponent. Authors therefore need to specify whether a quoted length belongs to amplitude or density.
Plotting the logarithm of density against distance turns exponential wings into approximately straight lines. In the formula, density varies with distance from the centre, and the Greek letter xi is the amplitude localization length; both distances use the same length unit. This is more informative than the central peak, which can contain interactions, imperfect preparation or a mixture of energies. For an illustrative amplitude localization length of ten micrometres, moving twenty micrometres from the centre reduces density by a factor of about fifty-five under the stated envelope. That numerical consequence distinguishes a localized tail from a broad profile that only appears narrow on a linear colour scale. Fitting must also respect the camera’s noise floor.
Sources: [3]
Cold atoms separated weak disorder from classical trapping
Billy and colleagues used an optical waveguide with a laser-speckle potential in 2008. After releasing the longitudinal confinement, the atomic cloud expanded along the guide. The weak optical potential provided disorder without appreciable dissipative photon scattering. At sufficiently low density in the wings, interactions became negligible, allowing comparison with a single-particle localization theory.
The useful evidence combined several observations: expansion slowed to a stationary profile, exponential wings appeared, and the fitted localization length varied with disorder strength in agreement with the relevant theory and calibration uncertainties. The team chose conditions where the atoms in the analysed wings had kinetic energy exceeding the disorder amplitude, making simple reflection from isolated classical barriers inadequate. They also checked the disorder’s spatial correlations. The speckle contains only a limited range of spatial variations, which affects how efficiently it scatters different wavelengths. Predictions for another energy or another pattern of disorder must include that change.
Sources: [3]
A quasiperiodic lattice is a different controlled test
Roati and colleagues used another route in the same year: a one-dimensional lattice formed with incommensurate optical periods and atoms whose interaction could be tuned close to zero. The site modulation was quasiperiodic, not a collection of independent random numbers. Changing its strength relative to tunnelling crossed from extended behaviour toward localized behaviour.
Transport measurements were supplemented by spatial and momentum distributions. That combination matters because stopped motion, a localized density profile and a changed momentum structure constrain different possible explanations. The quasiperiodic model can possess a finite transition strength that an ordinary one-dimensional uncorrelated random chain does not share. Calling both experiments localization does not erase their different Hamiltonians. Their value is that each creates a potential and interaction strength well enough controlled to compare with its own theoretical prediction.
Sources: [4]
Dimension and sample size change what one can conclude
The 1979 scaling theory asked how conductance changes as a coherent disordered sample is made larger. In its standard noninteracting symmetry setting, the two-dimensional system has no ordinary metallic large-size limit; three dimensions can admit a transition between extended and localized behaviour. Magnetic fields, spin–orbit coupling, interactions and correlations in the disorder require additional care.
A finite experiment may look extended when its localization length exceeds the available apparatus. Likewise, slow expansion during a short observation does not demonstrate localization. Testing several times and sizes helps distinguish those cases. Finite-size numerical calculations should examine transmission, spatial profiles and disorder realizations together. Several sample lengths reveal whether an apparently concentrated wave stays localized as more paths become available.
Sources: [2][4]
Smooth, weak disorder first gathers waves into branches
Disorder does not always hold a wave in place. When the bumps of a landscape are gentle and much wider than the wavelength, a wave crossing them keeps travelling, yet it does not spread into an even haze either. It gathers into narrow bright channels that fork again and again, like the branches of a tree. This is called branched flow. Scanning-probe images showed it in electrons flowing through a semiconductor device, and in 2020 Patsyk and colleagues made the first optical observation, with green laser light guided inside a soap film whose thickness varies naturally from place to place.
Each wide bump acts as a weak lens. Neighbouring parts of the wave pass through nearly the same lenses, so their small bends add up until the paths cross and light piles up along lines called caustics. Kaplan showed in 2002 that the branches grow from these caustics. The formula links three quantities: the distance travelled before the first branches appear, the typical width of the bumps, and their strength compared with the wave's energy, which enters with a power of minus two thirds. Halving the width of the bumps halves the distance, while bumps three times weaker only about double it.
Branched flow is not localization. The light keeps moving forward, and each branch splits again further on; nothing stops the wave from spreading. What changes is how its brightness is shared out. In the soap film, very bright spots were far more common than in the speckle left by uncorrelated disorder, a heavy tail in the intensity statistics. Localization asks whether interference can halt transport; branched flow shows how weak, smooth disorder reorganizes a wave that keeps moving.
Sources: [5][6]
Disorder and repulsion should remain separate questions
Anderson localization made disorder an active organizer of wave motion. Interference could now account for an entire change from spreading to localization. The later atomic experiments made the spatial envelope accessible while retaining explicit control over important alternative mechanisms.
The distinction from classical trapping remains vivid in the atomic experiments. A particle with enough energy to cross each isolated obstacle can still remain confined by the combined wave amplitudes. Watching the wings stop and measuring their decay length makes that collective interference accessible in real space. Repulsion poses a further question: in a periodic landscape, particles can also obstruct one another. Keeping the two mechanisms distinct helps explain why very different materials can both be poor conductors.
Sources: [1][2]