A translator at negotiations doesn't listen to every word; they grasp the meaning of whole phrases. Similarly, the new method SBOCF doesn't compare every pixel of a diffraction pattern with a simulation. It cuts the image into nine squares, like a page into paragraphs, and for each square it sums the brightness. Instead of 24,649 pixels, you're left with 11 numbers: nine sums and two correction coefficients.

These 11 numbers are modeled by cheap stand-ins for expensive simulations, called Gaussian processes. A smart search, like an experienced negotiator, chooses the next set of thickness and tilt that promises the greatest improvement. On a synthetic sample with a thickness of 380 Å and a tilt of 1.5 mrad, the method reduced the error by a factor of 290 compared to the old approach, using only 50 simulations.

A translator who hears the meaning of whole phrases works faster and more accurately. Likewise, SBOCF finds the thickness and tilt of a crystal in 50 simulations and helps build precise maps of atoms. The method can already be applied in electron microscopes, and in the future, in X-ray diffraction.