A photograph of a molecule

In 2009 an instrument at IBM Zurich resolved the individual bonds of a single pentacene molecule. It is not a photograph — no light is involved, and what it maps is force — but it is one molecule rather than an average over 10¹⁶, and Dalton’s indivisible pieces are in the picture.

1873 – now15 min

By the end you should be able to

Explain why a crystal structure is an average, and what averaging destroys
Say why no optical microscope can ever image a molecule
Describe what the 2009 atomic force microscope actually measured, and what it did not
Judge honestly what “seeing a molecule” does and does not mean

Every structure in this act so far is an average, and it is worth being clear about what that costs. A diffraction pattern is produced by the whole crystal at once. Every molecule in it contributes to every spot, and what comes back out of the transform is the mean electron density over all of them. How many is all of them? For a crystal you can barely see — two-tenths of a millimetre on a side — the answer is around 2 × 10¹⁶ molecules. The problem below works it out. That average is one of the most informative objects in science. It is also not a picture of a molecule. It is a picture of what 10¹⁶ molecules have in common, taken while every one of them is held in the position and orientation the lattice imposes. And averaging fails in a specific way that is worth knowing, because it is invisible in the result. A part that moves does not blur. It disappears. If a side chain or a loop sits differently in different copies, its electron density is spread over many positions and none of them reaches the level at which anything is drawn. In protein crystallography this is routine: regions of the chain simply do not appear in the map. The honest caption is not "this loop is absent" but "this loop is not the same in every copy", and those are very different statements about the molecule. So the question at the end of the last lesson stands. What would it take to look at one?

The obvious answer is to build a better microscope, and the obvious answer has been ruled out since 1873. Ernst Abbe, working for Carl Zeiss in Jena, showed that two points closer together than roughly half a wavelength cannot be resolved by any lens, however perfectly made. It is not a manufacturing limit. The information about their separation is carried in light diffracted at angles so steep that it never enters the objective at all, and you cannot focus what you did not collect. For visible light, the floor is about 200 nanometres. A pentacene molecule is 1.40 nanometres from end to end — 1.21 nm if you measure carbon to carbon and leave the hydrogens out of it — and the bonds inside it are 0.14 nm apart. You are short by a factor of well over a hundred, and no advance in glass, no improvement in engineering, and no amount of money changes it. The super-resolution techniques that won the 2014 Nobel Prize get around Abbe by making individual fluorescent labels blink and locating each one from many frames — which is beautiful, and gets to tens of nanometres, and is still two orders of magnitude away from a bond. If you want to see a molecule, light is not the tool. Something has to touch it.

Individual atoms had been imaged for decades before 2009, and skipping that makes the story sound more sudden than it was. 1955 — the field ion microscope. Erwin Müller, at Penn State, put a hard vacuum and a very high field around a sharpened tungsten needle and imaged the individual atoms at its tip. This is generally reckoned the first image of individual atoms, and it is fifty-four years before the pentacene picture. 1981 — the scanning tunnelling microscope. Gerd Binnig and Heinrich Rohrer, at IBM Zurich, measured the current tunnelling between a sharp tip and a conducting surface, which falls off so steeply with distance that essentially all of it comes from the last atom. Nobel Prize in 1986. But an STM maps electron states near the Fermi level, not atoms — which is why STM images of a molecule show its orbitals as smooth lobes and not its bonds. 1986 — the atomic force microscope. Binnig, Calvin Quate and Christoph Gerber measured the force on the tip instead of the current, which frees you from needing a conducting sample. 1990 — thirty-five xenon atoms. Don Eigler and Erhard Schweizer at IBM Almaden dragged individual atoms across a nickel surface and spelled out I B M. Not an image of a molecule, but proof that a single atom is an object you can pick up and put down. So: atoms, yes, and for a long time. The chemical structure inside a single molecule — this bond here, that ring there — is a harder problem, and it stayed unsolved for another two decades.

The experiment

The chemical structure of a molecule resolved by atomic force microscopy

Leo Gross, Fabian Mohn, Nikolaj Moll, Peter Liljeroth and Gerhard Meyer · Published in Science, 28 August 2009 · IBM Research – Zurich, Rüschlikon

The question: Scanning probe microscopes had imaged single molecules for twenty years and always as blobs or orbital lobes. Can the bonds inside one molecule be resolved — the actual chemical structure, on an individual?
The apparatus: A single pentacene molecule (C₂₂H₁₄) deposited on a copper surface, and on two monolayers of sodium chloride grown on copper — the salt layer electrically decoupling the molecule from the metal beneath it. A non-contact atomic force microscope using a qPlus tuning-fork sensor, run in constant-height mode with an oscillation amplitude smaller than an atom. The whole thing at 5 K in ultra-high vacuum, so the molecule does not move and nothing lands on it. And the essential modification: a single carbon monoxide molecule picked up from the surface onto the apex of the tip, on purpose, to serve as the probe.

Theory predicted

On the experience of the previous two decades, a bright oval. Scanning probes average over whatever the last few atoms of the tip can reach, and a metal tip — atomically sharp by any ordinary standard — is far too blunt for features a tenth of a nanometre apart.

They measured

The five fused rings, individually. The carbon–carbon bonds as bright ridges. The fourteen hydrogens at the periphery as faint outward extensions. A molecular structure diagram, obtained from one molecule.

How sure could they be? The contrast depends steeply on tip height and appears only in the repulsive regime, below roughly 4 Å — deliberately closer than anyone had thought reasonable to go. Above that the tip feels only the smooth van der Waals attraction, which varies over the whole molecule rather than over its bonds, and the image is the featureless oval everyone had been getting.

Why it mattered

The chemical structure of an individual molecule, measured rather than inferred. Everything before it in this act — every structure from Bragg to Hodgkin — is an average over an enormous number of copies. This is one. It does not replace crystallography, which remains far more accurate and works on things that are not flat, rigid and lying on a cold metal surface. It answers a different question: not "what do these molecules have in common" but "what is this one like".

Drag the tip height from 6 Å down to 3 Å and watch the picture change character. Then press "Bare metal tip".

Loading the diffractometer…

Two ways of knowing a molecule, side by side. On the left, the crystal: every copy superimposed, and what comes out is their average — computed, never looked at. On the right, one molecule, scanned line by line. The tip height is not a focus control: far out the instrument feels the smooth van der Waals attraction and returns an oval, and the bonds appear only when you push in close enough for Pauli repulsion to dominate. That is why the result waited until 2009, and why it needed a molecule on the end of the tip.

Force, not light

Be precise about what this instrument does, because the popular version of it is wrong in a way that matters. No light is involved. Nothing is illuminated, reflected or focused. There is no lens in the apparatus. What is measured is a frequency shift. A tiny quartz tuning fork carrying the tip is driven at its resonance. When the tip feels a force gradient, that resonance shifts by a few hertz. The shift is recorded at every point of a raster scan, and the map of those numbers is the image. The bright lines are ridges of repulsion. Where the molecule’s electron density is highest, the CO on the tip is pushed hardest, and the frequency shifts most. Electron density is highest along the bonds — so the map of repulsion looks like a structural formula. It is not a picture of bonds; it is a picture of where the electrons are dense enough to push back, which is the same places. The image is built up, not captured. One point at a time, over minutes. A photograph is an exposure; this is a survey. None of that makes it less of an achievement. It makes it a measurement, which on this site is the higher compliment — a measurement is the kind of thing that can be checked, and this one has been, by other groups, on other molecules, hundreds of times since.

The single most important part of the apparatus is the thing on the end of it. A metal tip is too blunt. Sharpened to a point one atom wide, it still averages over whatever its last few atoms can reach, and it is chemically reactive — it interacts with the molecule underneath in ways that vary from place to place and swamp what you are trying to see. So they put a molecule on it. A single carbon monoxide, picked up from the surface by driving the tip into it, and thereafter carried around as the probe. Three properties make it work: It is inert. CO does not react with the pentacene beneath it, so what the tip feels is repulsion and nothing else. It stands up. The molecule bonds to the metal through its carbon and points its oxygen down at the sample, giving a probe of known and reproducible geometry. It is small. The electron density at the oxygen end is a compact lump, and the resolution of the image is essentially the size of that lump. The instrument that resolved the bonds in a molecule uses a molecule as its sensor. That is a pleasing sentence and it is also the technical heart of the result: the 2009 paper is not the story of a better microscope, it is the story of a better tip, and the microscope had existed since 1986. One more piece of the arrangement is worth noticing. The best images came with the pentacene sitting on two monolayers of sodium chloride grown on the copper, which insulates the molecule electronically from the metal. Rock salt again — the same crystal Bragg solved first, now used as a table.

The part the story usually rounds off

These images are almost always captioned "the first photograph of a molecule", and they are not photographs. That is worth pressing on, because the difference is not pedantry. It is a force map, not an image. Discussed above. Nothing was seen; something was felt, point by point. And the CO on the tip is not rigid. It is attached to the metal by a single bond and it tilts as it is pushed. That lateral relaxation sharpens the apparent edges of everything — which is part of why the images look so crisp — and it distorts apparent distances. Bond lengths measured off an AFM image are systematically wrong in a way that depends on the local environment, and are not comparable to a crystallographic bond length. It has produced at least one real misreading. Images of adjacent molecules showed bright lines running between them, in positions where hydrogen bonds were expected, and these were reported as direct images of hydrogen bonds. Later work — notably Hapala and colleagues in 2014 — showed that similar features appear between atoms that are certainly not bonded, and that most of the effect is the CO bending in the potential rather than a bond being imaged. Some of those papers stand with revised interpretations; the lesson is that a picture is still an instrument reading, and instrument readings have artefacts. So: an extraordinary measurement, of one molecule, showing its real connectivity, with a known distortion in it. That is a more interesting thing to have than a photograph, and it is what we actually have.

Three years later the same group did something that ought to be impossible. They distinguished bond orders. In Science in 2012, Gross and colleagues imaged fullerene C₆₀ and two polycyclic aromatic hydrocarbons, and showed that bonds with more double-bond character come out shorter and brighter than bonds in the same molecule with less. The differences involved are a few hundredths of an ångström. That is a quantity organic chemistry has been reasoning about since Kekulé, and settled with hydrogenation enthalpies and X-ray spacings in the 1920s and 30s — now readable off one individual molecule. And they watched a reaction. In 2013 a group at Berkeley, working with the same technique, imaged an individual molecule before and after a cyclisation, and identified the products from the pictures. Not the average product of a flask. The particular molecules that happened to form. Both of those are, at present, restricted to a narrow class of subject: flat, rigid, thermally stable molecules that will lie down on a clean cold surface and stay put. That is a real limitation and it is not obviously going away. But the direction of travel is clear, and the thing being extended is the ability to interrogate individuals rather than populations — which is the same shift that transformed molecular biology when single-molecule methods arrived.

You might think

The 2009 image is the first time anyone had seen an atom.

Actually

Individual atoms had been imaged for fifty-four years by then. Erwin Müller’s field ion microscope showed the atoms on a tungsten tip in 1955; the scanning tunnelling microscope did it routinely from 1981; Eigler and Schweizer picked up thirty-five xenon atoms and arranged them into letters in 1990. What was new in 2009 was the chemical structure inside a single molecule — which bonds connect which atoms, on one individual, rather than atoms as featureless bumps on a surface. That is a different and harder problem, because it requires resolving features about 0.14 nm apart within an object that is itself only 1.4 nm long, and because the contrast mechanism that does it — Pauli repulsion in the close, repulsive regime — is not the one anybody had been operating in. The honest summary is not "we finally saw an atom" but "we can now read a structural formula off one molecule".

Problem

How many is an average over?

A crystal of pentacene, 0.2 mm on each side, is mounted on a diffractometer. Its unit cell has a volume of 692 Å³ and contains 2 molecules. How many molecules does the resulting structure average over?

Crystal edge: 0.2 mm = 2 × 10⁻⁴ m
Unit cell volume: 692 Å³
Molecules per cell: Z = 2
Conversion: 1 Å = 10⁻¹⁰ m

It would be very easy to make an analysis of any complicated chemical substance; all one would have to do would be to look at it and see where the atoms are.
Richard Feynman"There’s Plenty of Room at the Bottom", a lecture to the American Physical Society at Caltech, 29 December 1959. He was arguing that the electron microscope should be improved by a factor of a hundred, and offered a prize for it. He was right about the goal and wrong about the instrument: what got there was a tuning fork with a carbon monoxide molecule glued to the end.

1808Dalton argues from whole-number ratios that matter comes in indivisible pieces, and expects nobody will ever see one.
1873Abbe shows that no lens can resolve below about half a wavelength. Optical microscopy is finished at 200 nm.
1913Bragg gets coordinates out of a crystal — averaged over every molecule in it.
1955Müller’s field ion microscope images the individual atoms on a tungsten tip.
1981Binnig and Rohrer build the scanning tunnelling microscope. It maps electron states, not atoms.
1986The atomic force microscope: measure the force rather than the current, and the sample need not conduct.
1990Eigler and Schweizer spell IBM in thirty-five xenon atoms, moved one at a time.
2009Gross and colleagues resolve the bonds of a single pentacene molecule with a CO-terminated tip at 5 K.
2012The same group distinguish bond orders from apparent bond length and brightness.
2013A single molecule imaged before and after a reaction, and the products identified from the pictures.
2014The bright lines between molecules are shown to be largely tip bending, not images of hydrogen bonds.

Where this leaves us

The arc that opened with Dalton in 1808 closes here. He argued from whole-number ratios — nothing more than the mass of one thing combining with the mass of another — that matter comes in indivisible pieces. He also said, reasonably, that no one would ever see one. Two centuries later there is a picture of a single molecule with its individual bonds in it, and the honest description of that picture is that it is a map of force made with a molecule on the end of a tuning fork at five degrees above absolute zero. Both of those sentences are true at once, and holding both is the point. So: eight acts, and structure has gone from something argued about, to something inferred from reaction counts, to something computed from diffraction spots, to something an individual example of which can be laid out and looked at. And it is worth ending on what that does not buy. We can determine the structure of almost anything we can crystallise, and read one molecule at a time when it will lie flat. We still cannot reliably do the reverse. Hand a chemist a formula nobody has ever made and ask what shape it will take, what it will react with, and how fast — and the honest answer is that we compute approximations and then go and find out. We can see it. We cannot yet predict it. That gap is where chemistry actually stands, and it is worth looking at squarely.