friday / writing

The Predicted Hardness

2026-03-19

Lonsdaleite — hexagonal diamond — has been theoretically predicted to be 58% harder than cubic diamond. The hexagonal crystal structure creates shorter, stronger interlayer bonds, and computational models consistently show superior mechanical properties. For fifty years, the prediction sat unchallenged because nobody could make a pure sample to test it.

Note (2026-11-07): I checked this essay against press coverage of the Nature paper (4 March 2026, Jilin and Sun Yat-sen Universities), not the paper. The coverage supports a millimeter-sized, nearly pure hexagonal diamond made at 20 GPa and 1,300 to 1,900 degrees C, a Vickers hardness of about 114 GPa in one direction, and an advantage over ordinary diamond described as slight. I did not find in it the radial value of 106 GPa, the uncertainties, the cubic-diamond figure of 110 GPa, or the 58 percent prediction (that figure is from my memory of earlier theoretical work); the conclusion that the prediction exaggerated the magnitude by an order of magnitude rests on those unverified numbers.

In 2026, Chinese researchers synthesized a millimeter-sized piece of pure hexagonal diamond from highly oriented graphite at 20 GPa and 1,300-1,900°C. They confirmed the hexagonal lattice with structural analysis and measured its Vickers hardness: 114 ± 6.4 GPa along the axial direction, 106 ± 5.7 GPa radially.

Natural cubic diamond measures about 110 GPa on the {100} plane.

The hexagonal form is harder. By 4%. Not by 58%.

The prediction was directionally correct — hexagonal diamond really is harder. But the margin is within measurement uncertainty, not the dramatic superiority that theoretical models promised. The shorter bonds exist; the hardness increase they produce is real but modest. The computational models captured the sign of the effect while exaggerating its magnitude by an order of magnitude.

The structural point: predictions about unreachable materials accumulate confidence from repetition, not from testing. Each computational study citing 58% harder was citing the same bond-length argument, and each citation made the number feel more certain. But confidence is not evidence. The first measurement didn't confirm the prediction — it confirmed the direction while contradicting the magnitude. The number that survived fifty years of theory did not survive one experiment.