Dynamic Light Scattering

The flicker carries the size

Shine a focused laser into a clear dispersion of nanoparticles and look — through a pinhole, at one fixed angle — at the light that bounces off. You won't see a steady glow. You'll see a single bright speckle that flickers: it brightens, dims, sparkles, never settling. That flicker is not noise. It is the entire measurement.

The speckle exists because light scattered from every particle in the illuminated volume interferes at the detector. When the particles sit still, the interference is frozen and the speckle is constant. But particles in a fluid never sit still — they are kicked around by solvent molecules in Brownian motion. Each kick shifts a particle by a fraction of a wavelength, changes its contribution to the interference, and the speckle reshuffles.

Here is the key insight: how fast the speckle reshuffles depends on how fast the particles move, and how fast they move depends on how big they are. A small particle gets shoved around quickly and the speckle blurs in microseconds. A large particle lumbers along and the speckle holds its pattern for milliseconds. Time the flicker and you have timed the diffusion — and diffusion is a ruler for size.

That is dynamic light scattering. No staining, no labels, no drying, no electron beam. Just a laser, a detector fast enough to count individual photons, and a clock.

From Brownian motion to a correlation function

DLS does not stare at the speckle and try to describe it. Instead it asks a sharper statistical question: how similar is the intensity now to the intensity a tiny moment τ ago? That similarity is quantified by the intensity autocorrelation function:

g₂(τ) = ⟨ I(t) · I(t + τ) ⟩ / ⟨ I(t) ⟩²

At τ = 0 the intensity is perfectly correlated with itself, so g₂ is at its maximum. As τ grows, the particles have had time to rearrange, the speckle has changed, and the correlation decays toward 1 (fully scrambled). The rate at which g₂ falls off is the signal.

The measured intensity correlation connects to the underlying electric-field correlation g₁(τ) through the Siegert relation:

g₂(τ) = 1 + β · |g₁(τ)|²

where β (the coherence factor, ~0.2–0.95) depends on the optics. For a single population of identical spheres, the field correlation is a clean single exponential:

g₁(τ) = exp(−Γ · τ)

Γ (units s⁻¹) is the decay rate — the only thing the instrument really has to extract. It is tied to the diffusion coefficient D and the scattering vector q by:

Γ = D · q²

The scattering vector q sets the length scale the experiment probes. It depends on the laser wavelength in vacuum λ₀, the solvent refractive index n, and the detection angle θ:

q = (4π·n / λ₀) · sin(θ / 2)

So the recipe is: measure g₂(τ), subtract the baseline, take the square root to get g₁(τ), fit the exponential for Γ, divide by q² for D, and convert D to size with Stokes–Einstein. Every commercial DLS instrument runs exactly this chain in real time.

The Stokes–Einstein bridge to size

The diffusion coefficient by itself isn't a size — it's a mobility. The link to radius is the Stokes–Einstein equation, which says a sphere's mobility is set by thermal energy fighting viscous drag:

D = k_B · T / (6 · π · η · R_h)

where k_B = 1.381 × 10⁻²³ J/K is Boltzmann's constant, T is absolute temperature (K), η is the solvent viscosity (Pa·s), and R_h is the hydrodynamic radius. Rearranged to give the answer DLS actually reports:

R_h = k_B · T / (6 · π · η · D)

Three things fall straight out of this equation. First, D scales as 1/R_h: double the size, halve the diffusion, halve the decay rate Γ. Second, the answer depends on viscosity and temperature — get η wrong (e.g. assume water at 25 °C when the buffer is 40% glycerol) and the size is wrong by the same factor. Third, R_h is a hydrodynamic radius: it is the radius of the equivalent hard sphere that drags the same amount of solvent, so it always includes the solvation shell and any surface coating.

For polydisperse samples the single exponential breaks down. The standard fix is the method of cumulants, which expands the log of g₁:

ln g₁(τ) = −Γ̄·τ + (µ₂/2)·τ² − …

z-average diameter  ← from Γ̄  (mean decay rate)
PDI = µ₂ / Γ̄²       ← from the second cumulant (distribution width)

Γ̄ gives the intensity-weighted harmonic-mean size called the z-average; the second cumulant µ₂ gives the polydispersity index (PDI). For broader or multimodal distributions, regularized inversion algorithms (CONTIN, NNLS) fit a full distribution of decay rates instead of just two moments.

Worked example: a 100 nm liposome in water

Take a liposome of hydrodynamic radius R_h = 50 nm (100 nm diameter) in water at 25 °C, probed by a red HeNe laser at λ₀ = 633 nm and a backscatter angle θ = 173°. Water: η = 0.890 × 10⁻³ Pa·s, n = 1.332.

Step 1 — diffusion coefficient.

D = k_B·T / (6πη·R_h)
  = (1.381×10⁻²³ × 298.15) / (6π × 0.890×10⁻³ × 50×10⁻⁹)
  = 4.117×10⁻²¹ / (8.388×10⁻¹⁰)
  ≈ 4.9×10⁻¹² m²/s   (4.9 µm²/s)

Step 2 — scattering vector at 173°.

q = (4π·n / λ₀)·sin(θ/2)
  = (4π × 1.332 / 633×10⁻⁹) × sin(86.5°)
  = (2.644×10⁷) × 0.99813
  ≈ 2.64×10⁷ m⁻¹

Step 3 — decay rate Γ and correlation time.

Γ = D·q² = 4.9×10⁻¹² × (2.64×10⁷)²
  = 4.9×10⁻¹² × 6.97×10¹⁴
  ≈ 3.4×10³ s⁻¹
correlation time  1/Γ ≈ 290 µs

So this liposome's speckle decorrelates in roughly 0.3 milliseconds. Shrink the particle to a 5 nm protein (R_h = 2.5 nm) and D jumps to ~98 µm²/s, Γ rises 20-fold, and the correlation time drops to ~15 µs. That 20:1 spread in timescale across the nanoscale is exactly the leverage DLS exploits, and it is why correlators must span lag times from ~100 ns to several seconds.

Sample requirements, optics, and scope

DLS is forgiving in some ways and ruthless in others. What it needs:

• A transparent, dilute dispersion. Too concentrated and multiply-scattered photons corrupt the single-scattering model; too dilute and the count rate (kcps) is too low for good statistics. A typical target is 100–500 kcps. Backscatter detection (173°) tolerates higher concentrations because the light penetrates less.
• Scrupulous cleanliness. Because intensity scales as d⁶ (see below), one airborne dust speck can outshine the sample. Buffers are filtered through 0.2 µm (or 0.02 µm) membranes and cuvettes are rinsed dust-free.
• Known viscosity and refractive index. These go straight into Stokes–Einstein and q. Sucrose, glycerol, or high salt change η; the software must be told.
• Temperature control. D depends on T directly and through η(T). A Peltier stage holds T to ±0.1 °C; a 1 °C error shifts water viscosity ~2% and the reported size with it.

The working range runs from about 0.5 nm (small proteins, surfactant micelles) up to roughly 1 µm, beyond which sedimentation and the breakdown of pure Brownian motion (gravity, number fluctuations) end the technique. It works on proteins, micelles, microemulsions, polymers, colloidal metals and oxides, quantum dots, viruses, exosomes, and protein aggregates — anything that scatters and diffuses freely in a low-viscosity medium.

DLS vs other particle-sizing methods

	DLS	Static light scattering (SLS)	Electron microscopy (TEM/SEM)	Nanoparticle tracking (NTA)
What it returns	Hydrodynamic radius, z-average, PDI	Molecular weight, Rg, A₂	Dry geometric size + shape	Number-weighted size + count
Size range	~0.5 nm – 1 µm	~10 nm – µm (Rg)	~1 nm – µm	~30 nm – 1 µm
Weighting	Intensity (∝ d⁶) — biased to large	Intensity / mass	Number (count grains)	Number — each particle once
Sample state	Native, in solution	Native, in solution	Dried, fixed, under vacuum	Native, in solution
Resolves two populations?	Poorly (need ≥3× size ratio)	No	Yes — you see them	Yes (single-particle)
Speed per sample	~1–2 min	~5–20 min (conc. series)	Hours (prep + imaging)	~5 min
Cost / accessibility	Benchtop, ~$50–100k	Often same instrument	Facility, $$$, vacuum	Benchtop, $$
Best at	Fast monodisperse QC, aggregation onset	Absolute mass of polymers/proteins	True shape, hard sizing	Counting + low-conc. mixtures

The headline difference is weighting. DLS reports an intensity-weighted size, EM and NTA report a number-weighted size, and for any real polydisperse sample these disagree — often by a factor of two or more. A DLS z-average of 120 nm and a TEM number-mean of 80 nm for the same gold sol are not a contradiction; they are two honest answers to two different questions.

The d⁶ trap: why one aggregate ruins everything

For particles much smaller than the wavelength, Rayleigh theory says the scattered intensity scales as the sixth power of diameter:

I_scattered  ∝  d⁶   (Rayleigh regime, d ≪ λ)

The arithmetic is brutal. A single 100 nm particle scatters (100/10)⁶ = 10⁶ times as much light as one 10 nm particle. So if 0.0001% of your particles by number have aggregated into 100 nm clumps, those clumps can still contribute as much signal as the entire 10 nm population. The z-average drifts upward, a phantom large peak appears, and the PDI balloons. This is why:

• Samples are filtered religiously and centrifuged before measurement.
• The intensity-weighted distribution is the raw result; converting to volume- or number-weighted distributions requires Mie theory and an assumed refractive index, and amplifies any error.
• DLS is exquisitely sensitive as an early-warning detector of aggregation — a biopharma stability assay can catch a 1% aggregated fraction long before it's visible by other means. The same sensitivity that makes it noisy makes it the gold standard for spotting the first signs of protein aggregation.

Where DLS earns its keep

• Biopharmaceutical QC. Therapeutic monoclonal antibodies (R_h ≈ 5.5 nm) must stay monomeric. DLS screens every formulation buffer for aggregation, reporting PDI and the fraction of high-R_h species. A PDI creeping past 0.1 flags a stability problem before clinical lots are released.
• Lipid nanoparticle (LNP) vaccines. The mRNA-LNPs in COVID-19 vaccines are specified at roughly 80–100 nm diameter with PDI < 0.2 — and DLS is the routine release assay that confirms it, batch after batch.
• Colloidal gold and quantum dots. Nanoparticle synthesis is tuned in real time by tracking R_h; citrate-capped 15 nm gold sols report R_h ≈ 8–10 nm including the double layer, and PEGylation is confirmed by a measurable size increase.
• Polymers and micelles. Block-copolymer micelles, surfactant aggregates above the critical micelle concentration, and protein–polymer conjugates are all sized in their native solvated state, which EM cannot do.
• Exosomes and viruses. Extracellular vesicles (~30–150 nm) and viral particles are sized non-destructively, often alongside NTA for the number-weighted count.

Common misconceptions and pitfalls

• "DLS gives the real particle size." It gives the hydrodynamic radius — the equivalent hard-sphere size including solvation and coatings — and it is intensity-weighted. Expect it to read larger than a TEM dry size, and larger still for charged or PEGylated particles.
• "DLS resolves a mixture of sizes." Only if the populations differ by roughly a factor of three or more in size. Two peaks at 50 nm and 70 nm collapse into one broad z-average; the instrument cannot pull them apart from a single decay curve.
• "Use the volume or number distribution — it looks cleaner." Those are computed from the raw intensity distribution using Mie theory and an assumed refractive index. They are model-dependent transformations, not measurements, and small intensity-tail errors blow up when divided by d⁶.
• "Higher concentration gives a better signal." Past a limit it causes multiple scattering and particle–particle interactions, both of which distort the apparent size. Backscatter optics help, but dilution is often the right answer.
• "The viscosity setting doesn't matter much." It enters Stokes–Einstein linearly. Forgetting that a 50% sucrose buffer is ~15× more viscous than water inflates the reported size ~15×.
• "A non-spherical particle has a well-defined R_h." Stokes–Einstein assumes a sphere. A rod or a worm-like micelle still diffuses and still yields a D, but the R_h is an equivalent-sphere value that conflates translational and rotational motion — interpret with care.