Probes, Patterns, and (nano)Particles


Philip Moriarty

This is a guest post by Philip Moriarty, Professor of Physics at the University of Nottingham (and blogger).

“We shape our tools, and thereafter our tools shape us.”

Marshall McLuhan (1911-1980)

My previous posts for Raphael’s blog have focussed on critiquing poor methodology and over-enthusiastic data interpretation when it comes to imaging the surface structure of functionalised nanoparticles. This time round, however, I’m in the much happier position of being able to highlight an example of good practice in resolving (sub-)molecular structure where the authors have carefully and systematically used scanning probe microscopy (SPM), alongside image recognition techniques, to determine the molecular termination of Ag nanoparticles.

For those unfamiliar with SPM, the concept underpinning the operation of the technique is relatively straight-forward. (The experimental implementation rather less so…) Unlike a conventional microscope, there are no lenses, no mirrors, indeed, no optics of any sort [1]. Instead, an atomically or molecularly sharp probe is scanned back and forth across a sample surface (which is preferably atomically flat), interacting with the atoms and molecules below. The probe-sample interaction can arise from the formation of a chemical bond between the atom terminating the probe and its counterpart on the sample surface, or an electrostatic or magnetic force, or dispersion (van der Waals) forces, or, as in scanning tunnelling microscopy (STM), the quantum mechanical tunnelling of electrons. Or, as is generally the case, a combination of a variety of those interactions. (And that’s certainly not an exhaustive list.)

Here’s an example of an STM in action, filmed in our lab at Nottingham for Brady Haran’s Sixty Symbols channel a few years back…

Scanning probe microscopy is my first love in research. The technique’s ability to image and manipulate matter at the single atom/molecule level (and now with individual chemical bond precision) is seen by many as representing the ‘genesis’ of nanoscience and nanotechnology back in the early eighties. But with all of that power to probe the nanoscopic, molecular, and quantum regimes come tremendous pitfalls. It is very easy to acquire artefact-ridden images that look convincing to a scientist with little or no SPM experience but that instead arise from a number of common failings in setting up the instrument, from noise sources, or from a hasty or poorly informed choice of imaging parameters. What’s worse is that even relatively seasoned SPM practitioners (including yours truly) can often be fooled. With SPM, it can look like a duck, waddle like a duck, and quack like a duck. But it can too often be a goose…

That’s why I was delighted when Raphael forwarded me a link to “Real-space imaging with pattern recognition of a ligand-protected Ag374 nanocluster at sub-molecular resolution”, a paper published a few months ago by Qin Zhou and colleagues at Xiamen University (China), the Chinese Academy of Science, Dalian (China), the University of Jyväskylä (Finland), and the Southern University of Science and Technology, Guandong (China). The authors have convincingly imaged the structure of the layer of thiol molecules (specifically, tert-butyl benzene thiol) terminating 5 nm diameter silver nanoparticles.

What distinguishes this work from the stripy nanoparticle oeuvre that has been discussed and dissected at length here at Raphael’s blog (and elsewhere) is the degree of care taken by the authors and, importantly, their focus on image reproducibility. Instead of using offline zooms to “post hoc” select individual particles for analysis (a significant issue with the ‘stripy’ nanoparticle work), Zhou et al. have zoomed in on individual particles in real time and have made certain that the features they see are stable and reproducible from image to image. The images below are taken from the supplementary information for their paper and shows the same nanoparticle imaged four times over, with negligible changes in the sub-particle structure from image to image.

This is SPM 101

This is SPM 101. Actually, it’s Experimental Science 101. If features are not repeatable — or, worse, disappear when a number of consecutive images/spectra are averaged – then we should not make inflated claims (or, indeed, any claims at all) on the basis of a single measurement. Moreover, the data are free of the type of feedback artefacts that plagued the ‘classic’ stripy nanoparticle images and Zhou et al. have worked hard to ensure that the influence of the tip was kept to a minimum.

Given the complexity of the tip-sample interactions, however, I don’t quite share the authors’ confidence in the Tersoff-Hamann approach they use for STM image simulation [2]. I’m also not entirely convinced by their comparison with images of isolated molecular adsorption on single crystal (i.e. planar) gold surfaces because of exactly the convolution effects they point towards elsewhere in their paper. But these are relatively minor points. The imaging and associated analysis are carried out to a very high standard, and their (sub)molecular resolution images are compelling.

As Zhou et al. point out in their paper, STM (or atomic force microscopy) of nanoparticles, as compared to imaging a single crystal metal, semiconductor, or insulator surface, is not at all easy due to the challenging non-planar topography. A number of years back we worked with Marie-Paule Pileni’s group on dynamic force microscopy imaging (and force-distance analysis) of dodecanethiol-passivated Au nanoparticles. We found somewhat similar image instabilities as those observed by Zhou et al…

A-C above are STM data

A-C above are STM data, while D-F are constant height atomic force microscope images [3], of thiol-passivated nanoparticles (synthesised by Nicolas Goubet of Pileni’s group) and acquired at 78 K. (Zhou et al. similarly acquired data at 77K but they also went down to liquid helium temperatures). Note that while we could acquire sub-nanoparticle resolution in D-F (which is a sequence of images where the tip height is systematically lowered), the images lacked the impressive reproducibility achieved by Zhou et al. In fact, we found that even though we were ostensibly in scanning tunnelling microscopy mode for images such as those shown in A-C (and thus, supposedly, not in direct contact with the nanoparticle), the tip was actually penetrating into the terminating molecular layer, as revealed by force-distance spectroscopy in atomic force microscopy mode.

The other exciting aspect of Zhou et al.’s paper is that they use pattern recognition to ‘cross-correlate’ experimental and simulated data. There’s increasingly an exciting overlap between computer science and scanning probe microscopy in the area of image classification/recognition and Zhou and co-workers have helped nudge nanoscience a little more in this direction. Here at Nottingham we’re particularly keen on the machine learning/AI-scanning probe interface, as discussed in a recent Computerphile video…

Given the number of posts over the years at Raphael’s blog regarding a lack of rigour in scanning probe work, I am pleased, and very grateful, to have been invited to write this post to redress the balance just a little. SPM, when applied correctly, is an exceptionally powerful technique. It’s a cornerstone of nanoscience, and the only tool we have that allows both real space imaging and controlled modification right down to the single chemical bond limit. But every tool has its limitations. And the tool shouldn’t be held responsible if it’s misapplied…

[1] Unless we’re talking about scanning near field optical microscopy (SNOM). That’s a whole new universe of experimental pain…

[2] This is the “zeroth” order approach to simulating STM images from a calculated density of states. It’s a good starting point (and for complicated systems like a thiol-terminated Ag374 particle probably also the end point due to computational resource limitations) but it is certainly a major approximation.

[3] Technically, dynamic force microscopy using a qPlus sensor. See this Sixty Symbols video for more information about this technique.


Stripy Nanoparticles Revisited

Challenging published results is an onerous but necessary task. Today, our article entitled Stripy Nanoparticles Revisited has been published in Small, three years after its initial submission to this journal (3/12/09) and about three  and a half years after the first submission (to Nature Materials, 21/07/09).

As its title indicates, the article challenges the evidence for the existence and properties of “stripy” nanoparticles. The stripy nanoparticle hypothesis was first proposed in Nature Materials in 2004 by the group of Professor Stellacci (then at MIT and now at the EPFL). This hypothesis now forms the basis of 23 articles by the same group, mostly published in high impact journals including Nature Materials, Nature Nanotechnology, Nature Communications, Science, Journal of the American Chemical Society, Small, etc: 1234567 , 8910111213141516171819202122 and 23. Our article today is followed by a response from Professor Stellacci.

The stripy hypothesis can be described in simple terms as follows:

  1. take a ball (gold nanoparticle of ~4 nm diameter)
  2. cover it with two types of hairs, some short yellow ones (a small molecule with a thiol to bind to the gold) and some longer red ones (a small -but slightly longer- molecule)
  3. the hair will spontaneously organize to form a stripy ball with red and yellow hairs forming parallel lines (see cartoon below)

Image from Stellacci/Irvine press release reproduced from

This is a remarkable hypothesis. It did not come from a theoretical argument but from an observation. On the nanoscale,  it is extremely difficult to visualize how molecules organize. In the 2004 article, the authors decided to use scanning tunneling microscopy to “watch” molecules on nanoparticles. Scanning tunneling microscopy does allow atomic resolution imaging on very flat surfaces (and its inventors obtained the nobel prize for this reason); it works by scanning the surface with a very sharp tip. It is however a technique which requires particular care in the interpretation as artefacts can easily occur. It works well on very flat surfaces but is not normally used to look with sub-nanometre resolution at very bumpy surfaces such as films of adsorbed particles. In the 2004 article, the authors obtained images and proposed a model to fit these images; in you follow the link (first figure of the 2004 article), you can see some images in the panel a) and b), and the cartoon in panel c).

In Stripy Nanoparticles Revisited, we demonstrate  that the stripy hypothesis is based on an artefact. First, we show that the images are not compatible with the stripy hypothesis for basic geometrical reasons (if stripes are regularly spaced in 3D they cannot be in a 2D projection). Second, we use an image analysis technique called fast Fourier transform to study the periodicity in these images and we prove that the stripes are in fact an imaging artefact related to the scanning direction.

This explains the extremely improbable fact that the 10 “stripy” nanoparticles seen in panel a are all aligned in the same direction; why would they? and can you find what’s wrong in the movie below?

In Stripy Nanoparticles Revisited, we also consider some of the follow-up articles and in particular some of the very special structure-related physico-chemical and biological properties claimed and we fail to corroborate those.

In summary, Stripy Nanoparticles Revisited shows that 23 peer reviewed articles published over the course of 8 years in prestigious journals are based on a simple microscopy artefact. This perhaps suggests a failure of the peer review system, in particular given that ~ 10 of these articles have been published while our attempt to open this discussion into the scientific community was slowly going from submission (3/12/2009) to publication (23/11/2012).

We welcome comments on the scientific argument as well as the broader issue of  the process of establishing scientific knowledge.

Update 1: Responding to response?

Update 2: From above, the view of my Head of Department, Dave Fernig

Update 3: Seeing is believing? Not always… guest post by Philip Moriarty

Update 4: Scientific claims should be supported by experimental evidence, part 1, transmission electron microscopy

Update 5: Scientifc claims should be supported by experimental evidence, part 2, water-soluble stripy nanoparticles

Update 6: Alan Dove comments on the web and peer review in his post entitled Do These Stripes Make My Nanoparticles Look Weird?

Update 7: Scientific claims should be supported by experimental evidence, part 3, non-specific interactions with proteins

Update 8: Simon Hadlington covers the controversy, Chemistry World

Update 9: Gaping holes in the gap; no data to support the existence of the stripes that are supposed to catch toxic ions in the latest Nature Materials installment of the stripy series…

Update 10:  Predrag Djuranovic, former graduate student of Francesco Stellacci, explains how he came to the conclusion that the stripes were an artefact seven years ago.

Update 11: I am not updating here anymore; instead, see the round up post