A wide bandgap solid is transparent, so the energy must be deposited nonlinearly: multiphoton and tunnel ionization create carriers only where the intensity is highest, and the absorption becomes a strongly superlinear function of the local field. That nonlinearity is what makes the deposition sub-diffraction, since a modest enhancement of the field produces a large enhancement of the absorbed energy.

The enhancement follows the polarization and is asymmetric between the illuminated and the shadowed side of the pore. It is also how the pores appear in the first place: scattering and photoionization localize the energy, cavitation opens nanovoids, and after a few pulses the irradiated volume is a nanoporous material that has manufactured its own antennas.

Crystallization starts about four hundred picoseconds after the pulse, and it starts in several places at once. The nucleation sites are not random: they sit where the pores were, because the collapse leaves behind exactly the combination of density, pressure and temperature that lowers the nucleation barrier. Heterogeneous energy localization has become heterogeneous nucleation.

The phase that grows is stishovite, in which silicon is six-fold coordinated instead of four-fold. The partial radial distribution functions of the final structure match those of pure stishovite at the same temperature for the silicon-oxygen and oxygen-oxygen pairs, and the octahedral coordination is confirmed by the bond-angle distribution.

This figure carries the central argument. The transformation is a race: the pressure produced by the collapse relaxes on its own timescale, and the crystal has to nucleate and grow before it does. In the nanoporous sample the crystallization is fast enough to win. In a homogeneous sample that reaches the same average temperature and density, nucleation is delayed and the pressure has already fallen when it starts, so the high-pressure phase never forms.

The inhomogeneity is therefore not a defect to be tolerated but the ingredient that makes the transition possible.

Silica has an unusually rich high-pressure landscape, and the transient state produced by a pore collapse can land on more than one branch of it. The distinction between displacive transitions, where the network shears without breaking bonds, and diffusive ones, where atoms must move past one another, sets both the accessible phases and the time each needs. It also determines whether the quench preserves the new order or returns the material to a glass.

Stishovite itself has a limit: beyond about fifty gigapascals it converts to a post-stishovite orthorhombic phase, so the window in which it can be formed and kept is bounded on both sides. The same reasoning applies to densification without crystallization, where the network compacts by up to a few percent and stays amorphous.

The simulations would remain a claim without this. Diffraction on irradiated samples, in confined geometries and at a silica interface, gives spacings that match stishovite within a few percent. The absence of a preferred orientation is itself informative: it says the crystallites nucleated independently rather than growing from a single front, which is what the atomistic picture predicts.

The dynamics are governed by the same two timescales. If the excitation leaves the glass long enough above the glass transition, nuclei form and the material devitrifies, losing precisely the disordered structure that gives a metallic glass its properties. If the quench is fast enough, the amorphous state is preserved even though the material was fully molten.

The control parameter is again the deposited energy and the depth over which it is deposited, and the outcome is again a competition rather than a threshold.

What makes this case instructive alongside silica is the reversal of the goal. In the glass one wants to create and keep a crystal; in the metallic glass one usually wants to avoid creating one, since the amorphous state is the functional one. The physics is the same and only the desirable side of the competition changes, which is a good test of whether the description is genuinely predictive.

The comparison also connects this theme to the microstructural one: in a crystalline alloy the same simulations follow melting and recrystallization, and the quantity to control is the grain size rather than the presence of a crystal at all.