V. Cabinet of Curiosities

A small, growing collection of individually interesting materials drawn from Alexandria, one application at a time.

By the numbers

Alexandria has been growing for the better part of a decade, and the shape of that growth is itself part of the story. It began in 2017 as a brute-force search through every possible perovskite that would fit in a small, symmetric unit cell, moved on to prototype-specific machine learning models in 2018, and by 2021 was training crystal graph neural networks on millions of data points. Each generation made it easier to guess, before ever running an expensive calculation, which imagined crystal might actually turn out to be stable: the hit rate for turning a candidate structure into something thermodynamically real climbed from roughly 1 in 1,000 in the early prototype-search era, to about 9 in 100 once chemical-similarity-guided methods arrived, to 99 in 100 with the generative models used today.

The database that resulted currently holds 5.8 million calculated structures, 175,000 of them stable enough to sit on the convex hull. The most recent expansion alone generated 119 million candidate crystals computationally, screened them cheaply with machine-learned shortcuts, and confirmed 1.3 million of them with full first-principles calculations — adding 74,000 newly discovered stable materials in a single pass. Alongside it sits sAlex25, a companion set of 14 million snapshots of atoms still mid-relaxation, not yet settled into their final resting positions, used to teach force-field models what a crystal does when disturbed rather than just what it looks like at rest.

Extremes

Sift through Alexandria's quarter of a million most stable compounds and some genuine extremes turn up. Among ordinary, everyday solids, lithium is the featherweight record holder — so light that a block of it floats on water — while osmium anchors the other end, twice as dense as lead: a golf-ball-sized lump would weigh nearly as much as a bowling ball. The best ordinary insulator is lithium fluoride, whose band gap is wide enough that it is used to make windows for ultraviolet light. Push into the exotic, though, and even these records fall: freeze helium solid and its band gap widens further still, while frozen hydrogen is barely an eighth as dense as water — states of matter that exist only under conditions no household will ever see.

Electrical conductivity

Reference arXiv:2605.22167

Electricity moves through a metal at the mercy of its electrons: the fewer bumps in the road, the better the conductor. Silver is the reigning champion, better than anything else known, natural or man-made — gold and copper trail just behind, which is why they wire our world. No one has ever managed to beat silver at its own game, but chemists have come tantalisingly close: LiBePt₂, a compound that exists nowhere in nature, conducts almost as well as household aluminium foil. Stranger still is NbW₂ — neither of its parent metals, niobium or tungsten, is anything special on its own, yet fused together into one crystal they conduct electricity better than either manages alone.

Superconductivity

Reference 10.1038/s41467-025-63702-w

Some materials, chilled cold enough, carry electric current with no resistance at all — forever, with no losses. MgB₂ holds the real-world record for this trick at ordinary pressure, switching on at 39 degrees above absolute zero. Nb₃Sn needs to be colder still, but it is tough enough to wind into wire, which is why it forms the giant magnet coils inside the Large Hadron Collider at CERN and ITER, the fusion reactor being built in France. Neither has been beaten in the lab — but on paper, two hydrogen-rich compounds that have never been made, Li₂AgH₆ and Li₂AuH₆, are predicted to go superconducting at over 90 degrees — warm enough to cool with cheap liquid nitrogen instead of expensive liquid helium.

Dielectrics

References 10.1103/PhysRevMaterials.8.015201, 10.1103/dg72-zpqn

Every camera lens, antireflective coating, and touchscreen depends on how strongly a material bends light, and how wide a window of that light it lets through untouched. Ordinary titanium dioxide is already good enough at this to end up in sunscreen and white paint. Screening Alexandria for compounds that could rival it turned up CaMgTe₂, a calcium-magnesium telluride sharing none of titanium dioxide's elements yet matching its performance almost exactly. Even more familiar is HfZrO₄, a close cousin of hafnium oxide — the material that has sat inside the transistors of virtually every computer chip made since 2007 — and, by these calculations, does the job slightly better still.