Why a metal will not glass the way silica does
Guide 4 in this rung made silica into a glass with a beautifully specific picture: stiff, directional covalent bonds build corner-sharing tetrahedra into a floppy, disordered web — the continuous random network — and because those bonds are strong and hard to reshuffle, the liquid seizes up into a solid the moment you cool it a little past melting. The directionality of the bonds is doing the heavy lifting: it is what makes the network rigid enough to freeze without ever finding the crystal. Now try to do the same trick with a plain metal — copper, gold, iron — and the picture collapses, because a metal has no such bonds to freeze.
Metallic bonding is non-directional: an atom in a liquid metal does not care which way its neighbours sit, only that it is surrounded on all sides by a sea of shared electrons. Strip away directionality and an atom behaves almost like a smooth hard ball, and a molten metal becomes a heap of such balls jostling to pack as tightly as they can. That has two consequences that shape this whole guide. First, the frozen structure of a metallic glass cannot be a network of bonds — it must be some kind of dense random packing, and we will need a fresh model for it. Second, and more painfully, a heap of hard balls slots into a crystal with insulting ease: cool a normal molten metal and its atoms drop straight into the grocer's-orange close packing you met long ago (packing fraction 0.74, coordination 12), so making a metallic glass at all is a fight against a liquid that desperately wants to crystallize.
Random close packing: shaking a bag of ball bearings
The founding experiment is disarmingly physical. In the 1960s J. D. Bernal poured thousands of identical steel ball bearings into a rubber bladder, kneaded and squeezed it to settle them as densely as random jostling allows, then ran in wax or paint to freeze the arrangement, cut it open, and painstakingly measured every ball's neighbours. What he found is the model at the heart of this guide: a reproducible, dense, but thoroughly DISORDERED packing — random close packing — that fills about 0.64 of space. That is distinctly short of the 0.74 a crystalline close packing reaches, yet far above the roughly 0.55 of a loose random pour. It is the densest a heap of equal hard spheres will go without organizing itself into a lattice.
The crucial thing to notice is that this globally disordered packing is, locally, almost crystal-like. Count the balls in the first neighbour shell around a typical ball and you get roughly 12 to 13 — essentially the crystalline 12. Every atom is still snugly, fully surrounded; the nearest-neighbour distance is sharp and definite. So a metallic glass carries genuine, crisp short-range order (a well-defined first shell, near-crystalline coordination) while carrying no long-range order at all — which is precisely the central distinction this whole rung has been circling. The missing 10 percentage points of packing (0.64 against 0.74) is exactly where the disorder is hiding: it lives not in the tight first shell but in the shells beyond it, which smear out and lose all register.
Reading the glass in g(r): the tell-tale split second peak
How do we actually see random close packing in a real sample, without dissecting it in wax? Through the tool guide 3 built: the radial distribution function g(r), which counts how the density of atoms rises and falls as you walk outward from a typical atom. A perfect crystal answers with a picket fence of razor-sharp spikes, each at an exact interplanar distance; a gas answers with a flat line at 1, no structure at any range. A metallic glass answers in between, and its answer is unmistakable. There is a tall, sharp FIRST peak — the well-defined nearest-neighbour shell, the visible face of the short-range order — whose area integrates to that coordination number of about 12 to 13. Then a few damped oscillations that die away within a nanometre or so and settle to 1: the literal end of order, the distance beyond which the glass forgets itself.
Here is the signature that gives a metallic glass away at a glance. Its SECOND peak is SPLIT into two sub-peaks — a shoulder near 1.73 (the square root of 3) times the nearest-neighbour distance, and a second near 2.0 times it. That split is the fingerprint of dense random close packing: it arises from two particular favoured ways that neighbours-of-neighbours arrange themselves (roughly, near-collinear rows of three and near-equilateral triangles of three), and it is real medium-range order, structure reaching past the first shell that no simple crystal produces in that shape. In a diffraction experiment the same story arrives as the structure factor S(Q): broad, diffuse HALOS rather than the sharp Bragg spots of a crystal. See split-second-peak g(r) and diffuse halos, and you know at once you are holding a metallic glass, not a fine crystalline powder.
RADIAL DISTRIBUTION g(r): atom density vs distance r
(r in units of r1, the nearest-neighbour distance)
g | ||
| || <- 1st peak: sharp nearest-neighbour shell
| || area under it = coordination ~ 12-13 (SRO)
| ||
1 |-------||----./\.--./\.------------------------ = 1
| || / \ / \.__
| _||__/ v '~~~~~~ damped out by ~1 nm
0 |_____|________________________________________ r
0 1 1.73 2.0 -> long range: featureless
^ ^
SPLIT 2nd peak (sub-peaks near sqrt(3) and 2 x r1)
= fingerprint of random close packing / metallic glass
crystal -> picket fence of sharp spikes at exact spacings
gas -> flat line at g = 1 (no order at any range)
glass -> sharp 1st peak, SPLIT 2nd, then damps to 1Frustration: the favourite cluster that cannot tile space
We can now face the deep question this rung has been saving. Random close packing tells us what a metallic glass looks like once frozen — but why would any metal liquid resist crystallizing long enough to be frozen at all, when hard balls slot into a crystal so eagerly? The answer is geometric frustration, and it starts with an innocent puzzle: what arrangement packs a central atom most tightly with 12 neighbours? Intuition says a fragment of the close-packed crystal. Intuition is wrong. In 1952 Charles Frank showed the tightest, lowest-energy cluster of 12-around-1 is not a piece of any crystal but an ICOSAHEDRON — twelve atoms at the corners of that soccer-ball-like solid, packed a touch denser and bound a touch more strongly than the crystalline shell. A cooling liquid, seeking the lowest energy locally, wants to build these icosahedral clusters everywhere.
And here the trap springs shut. The icosahedron carries 5-fold symmetry axes — and the crystallographic restriction theorem from the symmetry rung is absolute: no periodic lattice can possess a 5-fold axis, so icosahedra cannot be stacked to tile three-dimensional space. The very cluster the liquid most wants to make is geometrically forbidden from growing into a crystal. THAT clash is frustration: the locally preferred order and the only available long-range order are incompatible, so the liquid is caught between them. To nucleate a crystal, the melt must first tear apart the icosahedral clusters it has already formed and prefers — an energy cost that stalls crystallization. So a frustrated liquid can be supercooled far below its melting point, and if you cool it fast enough it freezes solid, as a glass, before any crystal manages to nucleate and grow.
The glass transition: freezing a liquid without crystallizing it
Follow that supercooled liquid down in temperature and watch what actually freezes it. It does not crystallize — it thickens. Its viscosity climbs steeply, by more than ten orders of magnitude over a fairly narrow band of temperature, until at the glass transition temperature Tg the viscosity reaches roughly 10^12 pascal-seconds and the atoms simply can no longer rearrange within any time you are willing to wait. The structure stops flowing; it is frozen. The vital honesty here is that this is a KINETIC freezing, not a thermodynamic phase transition like melting. Nothing sharp happens to the structure itself at Tg — no latent heat of the melting kind, no sudden reordering; the atoms are merely caught mid-shuffle. The frozen structure of the glass simply IS the structure of the liquid at Tg, held motionless forever.
Because the freezing is kinetic, Tg is not a fixed material constant the way a melting point Tm is. Cool faster and the atoms fall out of equilibrium sooner, at a higher temperature, so Tg rises; cool slower and they keep up longer, so Tg drops. Melting has one sharp temperature; the glass transition is a smear that slides with your cooling rate — a fact worth holding onto, because it is often glossed over. A simple and vivid way to picture the trapping is the free volume idea: for an atom to move, it needs a scrap of empty space nearby to move INTO. As temperature falls the atoms crowd closer and that spare room — the free volume — shrinks; below Tg there is too little room left for any atom to shuffle past its neighbours, and flow stops. Free volume is a simplified model rather than the final word on the glass transition, but it captures the cage-and-trap intuition unforgettably.
Making and using a metallic glass
Put the pieces together and you see the problem in making a metallic glass. A simple metal is only weakly frustrated — its atoms crystallize eagerly — so to reach Tg before crystals nucleate you must cool brutally fast. The classic route is rapid quenching by melt spinning: squirt a jet of molten alloy onto the rim of a fast-spinning, water-cooled copper wheel, quenching it at roughly 10^5 to 10^6 kelvin per second and flinging off a solid RIBBON only tens of micrometres thick. It has to be thin, because only a thin layer can shed heat that fast; the middle of anything thicker cools too slowly and crystallizes. The very first metallic glass, an alloy of about 75 percent gold and 25 percent silicon, was made just this way by Pol Duwez at Caltech in 1960 — and for decades metallic glasses were stuck as thin ribbons and wires.
The modern breakthrough is the bulk metallic glass, and its trick is chemistry, not just speed. Mix four or five elements of very different atomic sizes — the famous 'Vitreloy' family blends zirconium, titanium, copper, nickel, and beryllium — and crystallization becomes sluggish. To crystallize, such a melt would have to sort a chemically confusing jumble of mismatched atoms into an ordered lattice, and that sorting is slow; this is nicknamed the 'confusion principle'. The size mismatch also deepens a eutectic that lowers the melting point Tm down toward Tg, so the reduced glass transition temperature Trg = Tg/Tm climbs above about 0.6. A high Trg means the dangerous window between Tm and Tg — where crystals can nucleate and grow — is narrow, so the critical cooling rate needed to skip it plunges from a million kelvin per second to around 1 kelvin per second. That is slow enough to cast glassy rods centimetres thick. Multicomponent confusion, not raw quench speed, is what buys you a bulk glass.
Why go to all this trouble? For a structure-property payoff that comes precisely from HAVING no crystal. With no lattice, a metallic glass has no dislocations at all — recall from the dislocation rung that a dislocation is a defect OF a lattice, so no lattice means no dislocation — and no grain boundaries either. Real crystalline metals are weak because dislocations let slip creep along one atom-row at a time; erase the lattice and that easy path vanishes, so metallic glasses reach near-theoretical strengths and springy, near-2 percent elastic limits. With no grain boundaries and no magnetocrystalline easy axes to fight, iron-based glasses are superb soft magnets for low-loss transformer cores — the same goal the grain-oriented steel of the texture rung chased, but reached by ERASING structure rather than aligning it. The honest cost is ductility: with no dislocations to spread plastic flow evenly, deformation collapses into a few narrow shear bands, so a bulk metallic glass is tremendously strong yet can snap with little warning in tension.
- Melt the alloy well above its melting point Tm in a crucible, holding it as a clean, fully liquid melt.
- Eject a jet of the melt through a fine nozzle onto the rim of a fast-spinning, water-cooled copper wheel.
- On contact the melt quenches at about 10^5 to 10^6 kelvin per second — far faster than crystals can nucleate and grow — so it freezes as a glass, not a crystal.
- It solidifies as a ribbon only tens of micrometres thick, flung off the wheel; thin, because only a thin layer can lose heat that fast (bulk glasses instead lean on multicomponent chemistry to allow slower cooling).
- Confirm it really is glassy: an X-ray pattern of broad diffuse halos with no sharp Bragg peaks, and a g(r) showing the sharp first peak and split second peak — not crystalline spots.