Molecules have fingerprints

Before any of this, finding out what a molecule looked like was a career rather than a measurement. You began with combustion analysis. Burn a weighed sample, trap and weigh the carbon dioxide and the water, and you have the empirical formula. Liebig’s apparatus of 1831 made this routine, and it is exact, and it tells you nothing about arrangement. C₂H₆O is ethanol, and C₂H₆O is also dimethyl ether, and they are different compounds. Then you took it apart. Reagents whose behaviour was known would break particular bonds; you identified the fragments, and reasoned backwards from the pieces to the whole. This is slow, destructive, and inferential — the answer is a hypothesis consistent with the debris. Then you built it. The proof of a structure was a total synthesis by a route in which every step was understood, ending in a substance whose melting point matched the natural one. That is a genuine proof and it is heroic, and it takes years. How slow? Cholesterol occupied chemistry for most of a century. Heinrich Wieland and Adolf Windaus took the Nobel prize in 1927 and 1928 for work on bile acids and sterols — and the ring structure they had settled on was shown to be wrong in 1932, by Bernal’s X-ray photographs. Two Nobel prizes, and the picture was still not right. None of this was incompetence. It was the best available, and it was the only thing available, and it is the situation that a magnet is about to end.

The instrument that ended it rests on one fact about a hydrogen nucleus. A proton is a small magnet. It has spin ½, and in a magnetic field it has exactly two allowed orientations — loosely, with the field and against it — with slightly different energies. The energy gap is proportional to the field strength. The gap is in the radio band. Put a proton in a field of 1.000 T and the gap corresponds to a photon of 42.577 MHz, which is a frequency somewhere between shortwave broadcasting and FM. Send radio waves in at exactly that frequency and the nuclei absorb them; a fraction of a hertz off and they do not. That sharpness is what makes the technique a measurement rather than an impression. Felix Bloch’s group at Stanford and Edward Purcell’s at Harvard detected this in bulk matter within about six weeks of each other, at the end of 1945 — Purcell in a block of paraffin wax, Bloch in a flask of water. Neither was looking for a chemical instrument. They shared the Nobel Prize in Physics in 1952.

And now the part that turned a physics experiment into the most-used instrument in chemistry, which arrived disguised as a problem. The nucleus does not feel the field the magnet applies. It feels that field minus whatever its own electrons shield off. Electrons circulating in an applied field generate a small field of their own, opposed to it, so the nucleus sits in a slightly weaker field than the pole pieces provide — and therefore resonates at a slightly lower frequency. How much shielding depends on the chemistry. Electron density around a given proton depends on what it is bonded to. Put an oxygen next door and it pulls electron density away; the proton is deshielded, feels more of the applied field, and resonates higher. The size of the effect is a few parts per million. This was found as an annoyance. In 1950 William Proctor and Fu Chun Yu, trying to measure the magnetic moment of the nitrogen nucleus, found that ammonium nitrate gave them two signals instead of one — the ammonium nitrogen and the nitrate nitrogen, in the same bottle, resonating at measurably different frequencies. Warren Dickinson found the same thing with fluorine the same year. For a physicist calibrating a nuclear moment, that is contamination of the measurement. It is also the whole instrument. The quantity that was spoiling the physics is a direct report on the electronic environment of every nucleus in the sample.

The splitting is the part that turns a chart into a structure, so it is worth doing properly. A proton feels its neighbours. Not through space — through the bonding electrons in between. A neighbouring nucleus polarises the electrons in its bond very slightly, that polarisation is passed along the bonds, and the effect reaches the next nucleus. Norman Ramsey and Edward Purcell explained the mechanism in 1952, two years after Herbert Gutowsky and Charles McCall, and Erwin Hahn and Donald Maxwell, had independently found the splitting. One neighbour gives two lines. That neighbour is either aligned with the applied field or against it, with almost equal probability, so half the molecules in the tube have their signal nudged one way and half the other. The line splits in two, of equal height. Two neighbours give three, in the ratio 1:2:1. Both up, both down, or one of each — and there are two ways to get one of each. The intensities are the number of ways. n neighbours give n+1 lines, with the intensities of the nth row of Pascal’s triangle. That is the whole rule, and it follows from counting arrangements of coins. The separation between adjacent lines is the coupling constant, J, quoted in hertz. It is a property of the pair of nuclei and the bonds between them, and — this matters — it does not depend on the magnet. Change the field and every chemical shift scales with it while every J stays exactly where it was. So: count the lines, subtract one, and you have counted the hydrogens on the atom next door. That is a statement about connectivity, which is what a structure is.

Infrared spectroscopy does the other half of the job, and the division of labour between them is worth stating plainly. A bond is a spring. Two masses joined by something with a stiffness will oscillate at a characteristic frequency, and if the oscillation changes the molecule’s dipole moment it will absorb infrared light at that frequency. Stiffer bond, higher frequency; heavier atoms, lower. Which makes group frequencies almost transferable. A carbon–oxygen double bond absorbs near 1700 cm⁻¹ in essentially every ketone, aldehyde, ester and acid, shifted by a few tens of wavenumbers according to what is attached. An O–H stretch gives a broad band above 3200 cm⁻¹, broad precisely because of the hydrogen bonding. A nitrile sits near 2250 cm⁻¹ in a region where almost nothing else absorbs, so its presence is a single glance. William Coblentz measured the infrared spectra of over a hundred compounds by 1905 and established that these frequencies travel from molecule to molecule. And below about 1500 cm⁻¹ the spectrum stops being transferable and starts being unique. That region depends on the vibrations of the skeleton as a whole, which is why it is called the fingerprint region: two spectra that match through it are the same compound, and that identification takes seconds against a library. So the two techniques answer different questions. Infrared names the parts — this has a carbonyl, that has a hydroxyl — in seconds, on a milligram, with no solvent needed. NMR draws the assembly. A chemist checking whether an oxidation has gone runs the infrared, because the O–H disappearing and a C=O appearing is the whole answer. A chemist who does not yet know what they have made runs the NMR.

What this did to the practice of chemistry is easy to underrate, because the result is an absence: things that used to take months stopped being discussed at all. Varian shipped the first commercial spectrometer in 1952, and the A-60 in 1961. The A-60 is the one that mattered. It was a benchtop instrument a graduate student could run without understanding the electronics, and within a few years it was in every serious chemistry department in the world. Structure determination stopped being a project and became a step. The sample survives. Combustion analysis burns it, degradation destroys it, and X-ray work needs a crystal you may not be able to grow. NMR needs a few milligrams in a solvent and gives them back. And it kept going. Richard Ernst and Weston Anderson replaced the slow field sweep with a pulse and a Fourier transform in 1966, which improved the sensitivity by orders of magnitude and made carbon-13 — 1.1% abundant, and correspondingly feeble — routine. Two-dimensional experiments followed in the 1970s, correlating each proton with the ones it is coupled to and turning connectivity into something you look at rather than deduce. Kurt Wüthrich used those to determine protein structures in solution, without a crystal at all, and took the 2002 Nobel prize in chemistry for it. The same magnet, pointed at a person instead of a tube, is an MRI scanner. The word nuclear was dropped from the name for the hospital market, and the physics is unchanged.

I remember, in the winter of our first experiments, just seven years ago, looking on snow with new eyes. There the snow lay around my doorstep — great heaps of protons quietly precessing in the earth’s magnetic field. To see the world for a moment as something rich and strange is the private reward of many a discovery.

1. 1945Purcell in Cambridge, Massachusetts, and Bloch at Stanford detect nuclear magnetic resonance in bulk matter, weeks apart.
2. 1950Proctor and Yu see two nitrogen signals in ammonium nitrate; Dickinson sees the same in fluorine compounds. The chemical shift arrives as an obstacle.
3. 1951Arnold, Dharmatti and Packard resolve ethanol into three lines, areas 3:2:1.
4. 1951Gutowsky and McCall, and Hahn and Maxwell, find spin–spin splitting. Ramsey and Purcell explain it through the bonding electrons the following year.
5. 1952Bloch and Purcell share the Nobel Prize in Physics. Varian ships the first commercial spectrometer.
6. 1956Arnold resolves the fine structure in ethanol: the triplet, the quartet, and the hydroxyl coupling in dried samples.
7. 1961The Varian A-60 puts an NMR spectrometer in the corner of ordinary laboratories.
8. 1966Ernst and Anderson replace the sweep with a pulse and a Fourier transform. Carbon-13 becomes routine.
9. 2002Wüthrich takes the Nobel prize for protein structures in solution — no crystal required.