Most chemical products start their lives as oil. And most of the conversion processes used to turn the black stuff into plastics, fuels and the rest rely on catalysts. Given the sensitivity of catalysts and Earth’s dwindling supplies of oil, you might think that these reactions would be among the most studied and the best understood in the chemist’s cookbook.

Unfortunately not. In fact, for many chemists and chemical engineers — those who work with bucketloads of reactants rather than the contents of pipettes — what goes on inside an industrial reactor is something of a mystery. It’s a black box. Indeed, when some textbooks and academic papers on the subject show flow charts of chemical processes, they actually represent the reactor, the beating heart of our industrial society, as a black box. If process engineers want to know what happens inside — and so how to make it more efficient, safer or more environmentally friendly — they measure what comes out, compare it with what goes in, and make an educated guess.

As computing power has grown, this educated guesswork has been renamed ‘modelling’. Reconstructions of the catalytic processes that occur in reactors use complex mathematics to represent the relationship between reactants, products and everything in between. Heat transfer, fluid dynamics and surface-reaction kinetics all offer a theoretical platform for such models, but, like all models, they rely on observations from the real world to make them realistic. Which takes us back to the black box and, often, to the most basic of questions — just how hot is it in there?

Anyone who has cooked a soufflé will know that the temperature, and how it fluctuates inside the oven, has a crucial bearing on the result. They know that the temperature selected and that the oven reaches can disagree. And they know that, even with the best temperature circulation, cool spots can lurk between lower shelves or above a baking tray. Now imagine that your precious pudding relies on the random collisions of a fizzing tempest of high-pressure gas and ageing, unpredictable catalysts. And that you are being asked to deliver 3,000 puddings an hour.

A reliable temperature map of the guts of a working chemical reactor would be valuable. People have tried to achieve this, most often by placing sensors at strategic points. The problem is the age-old paradox that the measurement disturbs what is being measured.

On page 537 of this issue, chemists offer a solution. Nanette Jarenwattananon at the University of California, Los Angeles, and her colleagues describe how they use the magnetic field of an nuclear magnetic resonance (NMR) scanner to accurately infer the hot and cold spots of a reactor carrying out the hydrogenation of propylene. And they report that, under the right conditions, hotter parts of the reactor signal narrower peaks on the NMR spectra.

There is a pleasing symmetry here. In the 1970s, NMR was handed to biologists and renamed magnetic resonance imaging (MRI). The biologists worked out a way to use MRI to sense the temperature inside the human body remotely. Now the chemists have reclaimed both the tool and the function. It is a proof of concept at this stage, but it does go some way towards opening that mysterious black box.