When a nucleus moves between two (sometimes more) chemical states we have Dynamic NMR. Common examples are: the equilibrium between conformers; complexes that can have two different structures, etc. "Chemical Exchange" is the expression that encompasses everything. The principle is so general, that the kind of chemical bonds that are created and destroyed is not relevant. What matters is the rate of exchange between the two states. If the exchange is slow, you see nothing in 1-D, but you see a cross-peak in the EXSY spectrum (another name for the more familiar NOESY). If the exchange is a little faster, you see two broad signals in 1-D. Warm the sample and the exchange becomes faster and faster: you see a single signal, but quite a broad one. At higher rates the single signal is so sharp that we don't mention Dynamic NMR anymore.
More exactly, the appearance of the spectrum depends both on the rate of exchange and on the difference (in Hz) between the two peaks. If we increase the magnetic field, the effect is similar to cooling.
To calculate the rate, we perform Lineshape Analysis, that is we compare a simulated spectrum with the experiment. At the fastest and slowest extremes, even a drastic change in the rate has little effect on the spectrum. At coalescence, instead, even a small change in the rate has a dramatic effect. This is when the signal is the broadest and when the rate of exchange can be calculated with the highest accuracy.
A few years ago I wrote a tutorial on a complex between a ligand with two nitrogens and a platinum ion. The ion could move between the two nitrogens. There were also six hydrogens in the molecule: A exchanging with A', B with B' and C with C'. The very nice thing was that the three frequency differences (A-A', B-B', C-C') had different values. We could therefore see three temperatures of coalescence and measuring the exchange rate was easy. The similarity between the simulation (black) and the experiment (red) was really OK:
In the first days of DNMR, acronym for dynamic NMR, only the singlets were studied. Not only the signals were stronger, but also easy to understand. Simple formulas, valid at coelescence only, were used instead of lineshape analysis. Partly because of this simplicity, results were not consistent. In the late 60s Binsch showed that coupled systems, just because they were very complicated, were also a much more accurate probe than singlets. Binsch also wrote the theory to simulate the coupled systems and the first computer program for the task, called "DNMR".
A great expert, today, is prof. Alex Bain. His suite of programs (open source) is called MEXICO. Another great expert is Hans Reich, who wrote WinDNMR. If you want a cheap alternative, I have written iNMR. You can try the Windows version for free for two months.