Dynamic light scattering is the most widely used sizing method in nanotechnology, and it is popular for good reasons. It is fast, it needs a few hundred microlitres, it does not require a vacuum or a stain or a grid, and it works in the liquid the particles actually live in. A paper on a liposome, a polymer nanoparticle, an exosome preparation or a protein aggregate will almost always carry a DLS number.
How dominant is it? A 2021 paper from the National Institute of Standards and Technology opens by noting that in recent years more than half of the drug products containing nanomaterials submitted to the FDA's Center for Drug Evaluation and Research relied on DLS for size characterisation. That figure is theirs, citing a 2017 survey I have not opened, so treat it as reported rather than independently checked. But it sets the stakes: this is the number regulators see.
What the instrument records
Shine a laser into a suspension and the scattered light from all the particles interferes, producing a speckle pattern. The particles are jiggling with Brownian motion, so the speckle flickers. Small particles jiggle fast and the flicker is fast. Large particles jiggle slowly and the flicker is slow.
The instrument measures the intensity autocorrelation function: how similar the signal is to itself after a delay. From the decay of that function you extract a diffusion coefficient, and from the diffusion coefficient you get a diameter through the Stokes-Einstein relation, given the temperature and the viscosity of the liquid.
So the chain is: flicker rate, to decay rate, to diffusion coefficient, to a diameter that a perfect hard sphere would have had if it diffused that fast. Everything downstream of the correlator is inference, and the European Nanomedicine Characterisation Laboratory's standard operating procedure for the technique states the flip side of that plainly: DLS measurements are based on first principles and hence no calibration is required. No calibration curve, because nothing is being compared to a length. That is a strength and it is also the whole problem, because a quantity nobody calibrates is a quantity nobody audits.
The standard's own definition is the story
ISO 22412:2017 is the international standard for DLS particle size analysis. Its definitions clause, which is in the freely published preview, defines the quantity like this:
average hydrodynamic diameter xDLS hydrodynamic diameter that reflects the central value of the underlying particle size distribution
Central value. Not the mean, not the median, not the mode. And then Note 1 to that entry finishes the thought:
The average particle diameter is either directly determined without calculation of the particle size distribution, or calculated from the computed intensity-, volume- or number-weighted particle size distribution or from its fitted (transformed) density function. The exact nature of the average particle diameter depends on the evaluation algorithm.
Read that again with a methods section in mind. The standard is telling you that the defined quantity is algorithm-dependent, and the methods section is reporting it as a scalar with two decimal places and no algorithm named.
The notes keep going and each one adds a dependency. Note 2 says the cumulants method yields a scattered light intensity-weighted harmonic mean particle diameter, sometimes referred to as the z-average. Note 4 says that mean values calculated from density functions on a linear abscissa and on a logarithmic abscissa may significantly differ, which means the plotting convention changes the number. Note 5 says xDLS also depends on the particle shape and on the scattering vector, and therefore on the angle of observation.
Algorithm, weighting, abscissa, shape, angle. Five dependencies, all in the definition, all in the free preview, and none of them typically stated next to the number.
The sixth power
Here is the piece of physics that does the most damage. Scattered intensity from small particles scales very steeply with size. The NIST paper puts the consequence directly:
The deviation between DLS and AFM measurements would be expected to increase with increase of the actual polydispersity of the sample because the DLS-derived mean, originating from an intensity-weighted measurement which scales as the sixth power of the particle radius, is extremely sensitive to and skewed by the positive tail of the distribution.
The EUNCL protocol, written independently and for a different purpose, states the same scaling and draws the practical conclusion from it:
The intensity of light scattered by nanosize particles is proportional to the sixth power of the particle diameter; thus larger dust particles will scatter much more light than smaller ones.
Sixth power means a particle twice the diameter of its neighbours contributes sixty-four times the signal. One aggregate per ten thousand good particles can move the reported average. This is why the EUNCL SOP spends pages on filtering dispersion media and keeping cuvettes closed, and why it notes that round-robin studies have found DLS to be very sensitive to small numbers of impurities. A dusty result and a genuinely aggregated sample look similar from the outside.
It also explains the single most common misuse of the technique. Software will happily convert the intensity-weighted distribution into volume- or number-weighted versions, and the number-weighted one looks reassuringly like microscopy, so people report it. The NIST authors traced what that conversion does and found it systematically undershoots. They cite a published liposome study in which the number-weighted arithmetic mean underestimates the corresponding z-average by as much as 35 percent, and they note that the current ISO and ASTM revisions specifically deprecate the use of number distributions for methods involving smoothing.
Two algorithms, two answers, same data
There are two standard ways to get from the autocorrelation function to a size, and they do not agree.
The cumulants method fits the log of the correlation function with a low-order polynomial and returns two numbers: the z-average and the polydispersity index. It is well behaved. The NIST paper says the cumulant method provides deterministic and reliable size estimates for samples with a coefficient of variation under three percent in an appropriate solution environment. Narrow distributions, in other words.
The other route is to invert the correlation function into a full distribution, usually by a non-negatively constrained least-squares method. This is where it gets uncomfortable, because the inversion is mathematically ill-posed: many different distributions produce indistinguishable correlation functions once a little noise is present. Software handles this with a regulariser that controls how smooth the answer is, and the smoothing does not solve the indeterminacy, it just picks one answer from the family. The NIST measurements show the cost:
The resulting errors are typically larger than 15% depending on the polydispersity index (PDI) of the samples.
Fifteen percent, between two analyses of the same raw data, driven by a parameter most users never touch and few report. The EUNCL protocol reaches the same place independently: it notes that there are many different methods used for the data analysis, but no standardized algorithm, and that different methods of data analysis can give different particle-size distributions. It also records the boundary that matters for compliance work, which is that only the cumulants method is included in the International Standard.
The polydispersity index is not the safety net
Most people who know all of the above still lean on the PDI: report the z-average, report the PDI, and if the PDI is low the sample must be narrow and the average must be trustworthy. That is the assumption the NIST group set out to test, by measuring gold and polymer nanoparticles by both DLS and atomic force microscopy and comparing the DLS polydispersity descriptors against the actual width measured particle by particle.
Importantly, we investigate the extent to which the DLS polydispersity descriptors are representative of the distributional quality and find them to be completely unreliable and misleading, both for reference materials and biomedical nanoparticles.
Completely unreliable, for certified reference materials as well as for real formulations. Their explanation is that the width of the distribution is the part the inversion is worst at recovering, and the error is almost always in the direction of too wide, which then poisons any conversion between weightings.
Their conclusion is a procedural recommendation rather than a plea to stop using DLS, and it is the sentence I would pin above the instrument:
Cross-examination by a sizing method for individual particles is therefore a must for credible reporting of nanoparticle size measurements (both mean size and distribution), and for reversing the alarming practice of the exclusive reliance on DLS-only in measurement-critical application fields.
An individual-particle method means electron microscopy, atomic force microscopy, or particle tracking: something that sizes objects one at a time. Nothing about ensemble scattering can tell you whether the ensemble is what you think it is.
Why this one is personal
My own published research is in microwave spectroscopy. I have never run a DLS instrument, and nothing in my work involved this technique, so I am reading these documents the way any careful outsider would.
What makes it familiar is the shape of the error rather than the field. Microwave spectroscopy also measures a response and infers a property through a model, and the same discipline applies: name the model, state what it assumed, and do not report the inferred quantity in the units of the thing you wish you had measured. I made the same argument about Brillouin microscopy in Report 145, about Raman in Report 013, and about XPS charge referencing in Report 122. It is the recurring failure of this beat: an instrument's output gets relabelled as the quantity people wanted, and the relabelling survives peer review because everyone in the room does it.
There is a second reason I care, and it deserves stating plainly. I formulate supplements, and Shroombiosis is a company I run. "Liposomal," "nano-encapsulated" and "enhanced bioavailability" are common claims in that market, and where such a claim is backed by any characterisation at all, it is usually a single DLS number on a certificate of analysis. Everything above applies to those numbers, including to any that my own industry produces. A z-average on a COA is not evidence that a delivery system exists, it is evidence that something in the tube diffused at a certain rate. This report is not about any product, mine or anyone else's, and nothing here should be read as a claim about one.
What I could not confirm
I did not read the full ISO 22412:2017. It is a paid standard. What I opened and read is the publicly posted preview PDF, which contains the front matter and the terms and definitions clause, and every ISO quotation above is from clause 3.2 and its notes as they appear there. I have not read the standard's measurement or reporting clauses, so I cannot say what additional guidance they give, and the NIST authors' broader criticism of the standard is their assessment of the full document, not mine.
Two cited figures are secondhand and labelled as such. The FDA submissions statistic comes from the NIST paper citing D'Mello et al. (2017), which I did not open. The 35 percent liposome discrepancy comes from the NIST paper citing Barberio et al. (2020), which I also did not open. In both cases I am reporting what one source I read says about another source I did not.
Standards I did not open. ASTM E2490-09:2015 and ASTM E3247-20:2020 are referenced in the NIST paper, including for the deprecation of number distributions, and both are paid documents that I did not read. ISO 9276-1 and ISO 9276-2, on the representation and calculation of means from particle size distributions, are cited in the ISO 22412 notes and were likewise not opened.
Vintage and scope. The NIST paper is from 2021 and the EUNCL SOP is version 1.0, approved 31 January 2016, and refers in places to ISO 22412:2008 and to the then-draft ISO/DIS 22412; its instrument-specific guidance names one manufacturer's software. Neither document is fresh, and DLS software has continued to change. I reproduced no measurement here and ran no instrument. The reasoning about why the sixth-power weighting makes a single aggregate consequential is standard, but the specific arithmetic (a particle of twice the diameter contributing sixty-four times the signal) is mine, following directly from the scaling both sources state.
The signal
If you are reading a paper or a certificate of analysis, a DLS size is worth very little on its own. Four things make it worth something: which analysis produced it (cumulants or an inversion, and which regulariser), which weighting it is reported under (intensity, and if anything else, why), what the correlation function and count rate looked like, and whether anyone has looked at the sample with a method that sizes particles one at a time.
If you are producing the number, report the intensity-weighted result, keep the z-average and PDI for batch-to-batch comparison under fixed conditions, which is what they are genuinely good at, and do not convert to a number-weighted distribution to make the figure agree with microscopy. If you want to agree with microscopy, do microscopy.
The standard told us all of this in its own definitions clause, in a free preview, in five notes. It is one of the more unusual cases in this beat: nobody hid anything, and the disclosure is sitting in the first three pages of the document that defines the quantity.
Sources
- Natalia Farkas and John A. Kramar (Theiss Research and National Institute of Standards and Technology), "Dynamic light scattering distributions by any means," Journal of Nanoparticle Research 23, article 120 (2021), published 21 May 2021, DOI 10.1007/s11051-021-05220-6. Open access. (PRIMARY, full published article opened and read; the NIST-hosted accepted manuscript was also downloaded and read as a cross-check, and all quotations above are taken from the published version. Source for: the FDA / CDER submissions figure, attributed in the paper to D'Mello et al. 2017; the criticism of the ISO 22412 "central value" wording as ambiguous; the sixth-power sensitivity passage, quoted verbatim; the statement that cumulant analysis is deterministic and reliable below a 3% coefficient of variation; the ill-posed nature of the NNLS inversion and the role of the regulariser; the greater-than-15% discrepancy between default inversion output and the cumulant z-average, quoted verbatim; the finding that DLS polydispersity descriptors are completely unreliable and misleading, quoted verbatim; the 35% number-weighted underestimate reported for liposomes, attributed in the paper to Barberio et al. 2020; the note that current ASTM and ISO revisions deprecate number distributions; the identification of the intensity-weighted harmonic mean as the justified central value; and the concluding recommendation on cross-examination by an individual-particle method, quoted verbatim.)
- International Organization for Standardization, ISO 22412:2017, "Particle size analysis — Dynamic light scattering (DLS)," second edition, February 2017. (PRIMARY for the definition only. The full standard is a paid document and was NOT read. What was opened and read is the publicly posted 13-page preview PDF containing the front matter and the terms and definitions clause. Source for: clause 3.2, the definition of average hydrodynamic diameter xDLS, quoted verbatim; Note 1 on the average being algorithm-dependent, quoted verbatim; Note 2 identifying the cumulants result as a scattered light intensity-weighted harmonic mean also called the z-average diameter; Note 3 referring mean calculations to ISO 9276-2; Note 4 that means from linear and logarithmic density functions may significantly differ; and Note 5 that xDLS depends on particle shape and on the scattering vector and therefore the angle of observation.)
- Luigi Calzolai (authored), Fanny Caputo and Matthias Roesslein (reviewed), "Measuring the Size of Nanoparticles Using Batch Mode Dynamic Light Scattering," EUNCL-PCC-001, version 1.0, European Nanomedicine Characterisation Laboratory, approved 31 January 2016. (PRIMARY, full 12-page PDF opened and read locally. Used as an independent practitioner cross-check on the physics and the analysis ambiguity. Source for: the statement that DLS is based on first principles and requires no calibration, with instrument verification against a NIST-traceable standard instead; the sixth-power scattering statement, quoted verbatim, which independently confirms the NIST paper's scaling; the description of the Laplace inversion as ill-posed with several possible solutions; the statement that there is no standardized algorithm and that different analysis methods can give different particle-size distributions; the note that only the cumulants method is included in the International Standard; the recommended verification tolerance of within 2% on a polystyrene latex standard with PDI below 0.1; the acceptance criterion of 10% relative standard deviation on the z-average across at least three measurements; and the observation that round-robin studies found DLS very sensitive to small numbers of impurities.)
- Onur Oncer, "What a Brillouin microscope actually measures," The Signal Report 145; "The 284.8 eV you didn't measure," The Signal Report 122; and "What a Raman spectrum can't tell you," The Signal Report 013. (Earlier reports in this beat on instrument outputs being relabelled as the physical quantity readers assume was measured.)
Scope note: this report describes what the size value produced by dynamic light scattering represents and what its published standard and protocols say about its limits. No measurements were performed for it, no instrument or software product is assessed or recommended, and nothing here is a claim about any specific material, formulation or product. Two figures are explicitly secondhand, reported from a source that was read about sources that were not, and are labelled above. Disclosure: the author formulates supplements and runs Shroombiosis, a company referenced in the body of this report; no product of that company or any other is characterised, evaluated or recommended here.
Onur Oncer
U.S. Army combat veteran (Counter-IED / Electronic Warfare), peer-reviewed researcher in microwave spectroscopy, and founder & CEO of Shroombiosis. Consults on laboratory operations, AI, and supplement formulation.