How Science Works | How Excitons Move Energy Through a Material

When a semiconductor absorbs suitable light, an electron can be promoted out of an occupied electronic state. The missing electron in the lower-energy states is described as a hole. It behaves as a positively charged quasiparticle in the material’s electronic description, rather than a literal empty cavity in the crystal.

The electron and hole interact. Their attraction can produce a bound state whose energy lies below the corresponding unbound electron–hole continuum. That correlated excitation is an exciton. Its binding and spatial extent depend on screening, electronic structure and the surrounding material.

Some excitons are relatively extended across many lattice spacings. Others are strongly localized on molecular or atomic-scale units. These are useful limiting pictures, not a rule that every actual excitation fits neatly into one cartoon.

An exciton is electrically neutral overall when it contains one electron and one hole. Moving its centre therefore does not transport the same net charge as moving one free electron. Nevertheless, an exciton can carry energy, interact with fields through its internal structure and eventually separate into charge carriers.

A diagram should show these distinctions explicitly. Draw an excitation created near the left of a film, its correlated electron–hole character while it moves, and three possible endings: photon emission, non-radiative loss or charge separation at an interface. Do not draw the hole as a tunnel through which the electron rolls.

The useful object is a collective electronic excitation embedded in matter. Its identity and dynamics come from that environment, not from treating it as a tiny isolated atom with a universal size and lifetime.

Contents · Next chapter

In a molecular solid, one unit can lose its excitation while a neighbouring unit gains one. The excitation has moved even when describing the event as one identifiable electron travelling all the way across the film would be misleading.

At suitable separations, electromagnetic coupling between transition dipoles can transfer excitation without emission of a freely propagating photon. At closer range, exchange-related mechanisms may matter. Which process dominates depends on wavefunction overlap, geometry, spectra and the material environment.

If interactions and environmental fluctuations repeatedly destroy the phase relationships between sites, transport can often be approximated as a sequence of incoherent hops. The waiting times and probabilities of those hops then determine the large-scale spreading.

In other regimes, electronic coupling supports an excitation spread coherently across several sites or a band-like description. Scattering, disorder and vibrations influence how long that description remains useful. “Quantum” does not automatically mean that energy crosses an entire macroscopic sample without scattering.

A relay analogy helps with the first distinction: the location of a message can move even though no single messenger completes the whole journey. But a literal relay is not the microscopic mechanism. Excitation transfer is governed by quantum states and their coupling, not by people handing over a physical packet.

Neither hopping nor coherent motion by itself guarantees long-range delivery. An excitation may become trapped or recombine before reaching the target. To predict useful transport, we need both a movement model and a survival model.

Contents · Previous chapter · Next chapter

A simple low-density model combines ordinary diffusion with a uniform first-order loss. For exciton number density n, write ∂n/∂t = D∇²n − n/τ. Here D is the diffusion coefficient and τ is the population lifetime.

The diffusion term redistributes surviving excitations in space. The loss term reduces their total number. With no boundaries or additional sources, the total population follows N(t) = N₀ exp(−t/τ).

For an initially Gaussian distribution in a homogeneous material, one Cartesian coordinate has variance σ_x²(t) = σ_x²(0) + 2Dt. In an isotropic two-dimensional plane, the added mean squared radial displacement is 4Dt. Those factors differ because the second expression includes two coordinates.

Our original example assigns D = 0.020 square micrometres per nanosecond, τ = 5.0 nanoseconds and an initial one-coordinate standard deviation of 0.20 micrometres. The initial variance is therefore 0.040 square micrometres.

At 1, 3 and 5 nanoseconds, that variance becomes 0.080, 0.160 and 0.240 square micrometres. The corresponding standard deviations are approximately 0.283, 0.400 and 0.490 micrometres. Meanwhile, a starting population of 1000 falls to about 819, 549 and 368.

The surviving cloud becomes broader while containing fewer excitations. A normalized spatial plot can display the first effect while hiding the second. That is why a transport report should preserve both spatial profiles and total signal, with the relationship between signal and population made explicit.

Contents · Previous chapter · Next chapter

A frequently used characteristic length is L = √(Dτ). For our assigned parameters it is √0.10 = 0.316 micrometres. It combines a rate of spreading with the time available before loss.

This is not a distance every exciton travels. Diffusion produces a distribution of displacements and path histories. The path length along a wandering trajectory also differs from its net displacement.

After a fixed elapsed time equal to τ, the additional one-coordinate root mean square displacement is √(2Dτ), approximately 0.447 micrometres. In two dimensions the radial value is √(4Dτ), approximately 0.632 micrometres. None of these numbers contradicts L = 0.316 micrometres; they are different defined quantities.

An author may use “diffusion length” with a geometric prefactor appropriate to their model. Before comparing papers, read the equation, dimensionality and lifetime convention. A factor-of-two discrepancy in a reported length may be partly definitional rather than a material discovery.

Now compare two changes. Doubling D while keeping τ fixed multiplies L by √2. Doubling τ while keeping D fixed does exactly the same to this particular length. Measuring L alone therefore cannot tell us whether an improvement came from faster spreading or longer survival.

This distinction matters for design. Reducing traps could lengthen survival without increasing microscopic hopping speed. Improving electronic coupling could change motion while leaving the dominant recombination channel nearly unchanged. Time-resolved spatial measurements help separate the two possibilities.

Contents · Previous chapter · Next chapter

Consider a one-dimensional half-space with a perfectly absorbing useful interface at x = 0. An exciton starts a distance x away, diffuses with constant D and independently disappears at rate 1/τ. Assume no drift, other boundaries or traps.

The probability P(x) of reaching the interface before disappearing satisfies D P″ − P/τ = 0, with P(0) = 1 and P tending to zero far away. The bounded solution is P(x) = exp(−x/L), where L = √(Dτ).

Using L = 0.316 micrometres, an exciton starting 0.10 micrometres away has a collection probability of about 0.729. Starting 0.50 micrometres away lowers it to about 0.206. At exactly one L, the probability is 1/e, about 0.368.

This is a first-arrival model, not a straight-line travel-time calculation. Some trajectories reach the interface quickly, some wander away and some disappear before arriving. Dividing starting distance by an invented fixed exciton speed would answer a different problem.

A real interface is seldom a perfect useful sink. Arrival may lead to charge separation, transfer to another state, reflection or loss. If useful conversion after arrival has an independent constant probability of 0.60, our model’s overall useful yields become approximately 0.437 and 0.123 for the two starting distances.

The extra factor is an assumption, not a universal law. If interface outcomes depend on arrival energy, position or excitation history, a richer model is needed. Separating arrival from successful conversion prevents poor interface chemistry from being mislabelled as slow bulk transport.

Contents · Previous chapter · Next chapter

Photoluminescence records emitted photons. It often samples excitons at the moment they recombine radiatively. A spatial image therefore depends on the population, the radiative rate, collection efficiency, spectral selection and the microscope response.

If the emission rate per exciton is uniform and constant, a normalized image may trace the normalized population well. If that rate varies with position, time or state, brightness no longer supplies a direct population map without additional modelling.

Even a perfect proportional readout is blurred by the instrument. For a Gaussian point-spread function and Gaussian source, one-coordinate variances add. Suppose the instrument variance is 0.090 square micrometres and is constant throughout our diffusion example.

The measured variances at 0, 1, 3 and 5 nanoseconds are then 0.130, 0.170, 0.250 and 0.330 square micrometres. Subtracting the initial measured variance still gives 2Dt because the constant blur cancels. Taking the raw width as the true source width would not.

That cancellation fails if focus drifts, the spectral response changes or the point-spread function changes between measurements. Background subtraction and a finite field of view can also alter fitted widths. A Gaussian fit is a useful summary only when residuals and competing profile shapes are checked.

Photon emission followed by reabsorption elsewhere can additionally move optical excitation over a distance. Such photon recycling is physically different from non-radiative exciton migration. A distant glow alone cannot determine which route carried the energy there.

Contents · Previous chapter · Next chapter

Here is an original counterexample to the claim that increasing width always proves diffusion. Imagine two stationary, co-centred Gaussian excitation populations. Neither moves. Initially they contribute equal total optical signal, and each has a constant emission factor within its own population.

The narrow population has one-coordinate variance 0.040 square micrometres and lifetime 1 nanosecond. The broad population has variance 0.360 square micrometres and lifetime 10 nanoseconds. Their weights decay exponentially at these different rates.

Initially the combined normalized second moment is the equal-weight mean, 0.200 square micrometres. After 2 nanoseconds, the relative weights are exp(−2) and exp(−0.2). The second moment becomes approximately 0.315 square micrometres.

The apparent distribution has widened because the narrow component disappeared faster. There was no exciton displacement at all. The sum is generally not itself a perfect Gaussian, so its directly calculated second moment is the defined quantity in this example, rather than an assumed fitted Gaussian width.

This does not make diffusion measurements meaningless. It identifies a competing explanation that can be tested using spectral separation, lifetime maps, low-density measurements and explicit multi-population models. Different measurements constrain different parts of the inference.

Density-dependent annihilation offers another complication. Preferential loss in a dense centre can change a profile even while genuine transport also occurs. An analysis that assigns every width change to D may then overestimate or otherwise mischaracterize transport. Measuring both time and space, and checking how results depend on excitation density, is more informative than fitting an intensity decay alone.

Contents · Previous chapter · Next chapter

Zhang and colleagues’ 2019 photovoltaic study addressed a measurement ambiguity: interfacial recombination losses can interfere with extracting an intrinsic exciton diffusion length from photocurrent. Their ratio-based approach used donor and acceptor responses to cancel a shared interface-loss factor under the method’s assumptions. This is an experimental counterpart of separating arrival from conversion in our model. Read the original photovoltaic transport study.

Seitz and colleagues reported spatially resolved transport in layered perovskites in 2020. Their observations included an early regime consistent with normal diffusion and later slower spreading associated with trapping. A single constant D across every time interval would miss that change of regime. Read the original layered-perovskite study.

Dark excitons add a readout challenge. “Dark” means that direct optical recombination is restricted by the relevant state-selection rules; it does not mean the excitation contains no energy or can never be detected. Chand and colleagues’ 2023 study used phonon-mediated optical readout and found long-range, density-dependent transport of dark excitons in monolayer WS₂. Their interpretation involved mutual repulsion, beyond our independent-particle diffusion model. Read the original dark-exciton study.

These examples should not be merged into one universal transport law. They involve different materials, excitonic states and experimental questions. Their shared lesson is methodological: identify the excitation, measure its spatial behaviour and test whether the assumed loss and interaction model matches the actual regime.

Contents · Previous chapter · Next chapter

An exciton is neutral overall, yet its electron and hole still experience their environment. Screening changes their interaction; disorder creates an uneven energy landscape; strain and interfaces can shift available states. Neutrality does not imply immunity to material structure.

In a January 2026 study, Liu and colleagues combined ultrafast transient-absorption imaging with a WSe₂ exciton sensor near a twisted WS₂ moiré system. They reported that correlated electronic states in the neighbouring system altered the dielectric environment and suppressed both exciton lifetime and diffusion in the studied regime. The result concerns a specific coupled structure, not every exciton in every semiconductor. Read the original 2026 imaging study.

This reinforces the need to measure D and τ separately. A shorter observed spreading distance might reflect slower movement, faster loss or both. An optical change can also alter how much of the population the chosen measurement sees.

For a device claim, ask an additional question: which destination is useful? In an energy-harvesting architecture, reaching a charge-separating interface may be desirable. In a spatially resolved light-emitting or imaging application, excessive spreading can blur where excitation was created. More transport is not automatically better for every purpose.

None of the numerical models here establishes a product efficiency or manufacturing outcome. Those require measured material parameters, interface behaviour, optical absorption, collection and stability under operating conditions. The article’s task is to make the transport evidence readable before those larger claims are assessed.

Contents · Previous chapter · Next chapter

Draw four panels for the original diffusion example at 0, 1, 3 and 5 nanoseconds. Keep the horizontal spatial scale fixed. First show each curve normalized to unit area to compare widths; then show decreasing total populations of 1000, about 819, 549 and 368.

On a separate graph plot one-coordinate variance against time. Its slope is 0.040 square micrometres per nanosecond, so D is half that value. Explain why using the two-dimensional radial factor of four on this one-coordinate graph would give the wrong result.

Next compare collection from 0.10 and 0.50 micrometres away. The assigned ideal-interface probabilities are about 0.729 and 0.206. State why these do not yet establish an electrical yield: arrival and successful charge separation are separate events.

Finally, repeat the stationary two-population calculation. Its increasing second moment shows that selective survival can imitate spreading. Propose a discriminating measurement, such as resolving the two populations spectrally and checking whether each one’s own spatial distribution broadens.

All of these tasks use supplied numbers and drawings. They require no nanoparticle handling, material synthesis or intense optical source. The central scientific habit is to distinguish where the excitation is, how long it survives and how the instrument makes it visible.

Return to Condensed Matter Physics for the wider electronic landscape, or explore the How Science Works master guide for related examples of collective behaviour and measurement.

Contents · Previous chapter · Continue to Condensed Matter Physics