Another consequence of LRO in network glasses is the quasiperiodic alignment of groups of caged voids associated with the aperiodic network of rings (Figure 7) to which the FSDP at QFSDP (Figure 2) is attributed [7]. In glass formers like SiO2 and B2O3, 2π/QFSDP distances lie around 4 Å. In chalcogenide glasses, such as As2S3 and GeSe2, SRO polyhedra are larger (~4 Å), leading to larger quasiperiodic void separations (~6 Å). In all cases FSDP distances decrease as pressure is applied but also become more dispersed, the FSDP peak widening and decreasing in intensity with increasing glass density [7].
Figure 6 Direct observation of the atomic structures of glasses. (a) Atomic force microscopy image from freshly fractured silica in ultra‐high vacuum, revealing a 2‐D projection of the near‐surface structure and showing SRO and fragments of rings – solid and dotted lines – contributing to LRO [24]. (b) Atomic resolution transmission electron microscope image of a graphene‐supported silica bilayer, showing SRO and extensive LRO network with a variety of ring structures [22], consistent with Zachariasen's CRN [13] (Figure 7a). (c) Nanobeam electron diffraction patterns of Zr0.667Ni0.333 metallic glass [23] (left), with their simulated patterns (right) in terms of the icosohedra shown in (d). The different SRO reflect the variety of icosohedra in Bernal's DRPHS model of liquid metals [25] (Figure 7b).
Source: (a) Reproduced from [24] © (2004) Elsevier; (b) reproduced from [22] © (2012) ACS; (c, d) reproduced from [23] © Nature Publications.
Importantly these changes in the FSDP properties of network glasses with pressure correlate with those of the boson peak [17], where νBP increases and IBP decreases with increasing pressure and therefore density, supporting the view that the BP comprises collective atomic motion of large groups of atoms whose breathing frequency increases as their size shrinks. Furthermore, excess Cp in glasses is attributed to a double‐well vibrational potential, which, in silica, can be modeled through the librational twisting of pairs of tetrahedral units, underscoring how the dynamics of IRO promote buckling of rings across LRO in network structures.
3.2 Metallic Glasses
In metallic glasses (Chapter 7.10) bonding is directionless and SRO comprises clusters of atoms around 3 Å in diameter [11, 12]. Coordination numbers (CN) are between 10 and 11 – much greater than in directionally bonded glasses. With respect to crystalline metals, the CNs of metallic glasses exceed 8, the value for bcc structure (8), but fall short of 10, the CNs for fcc and hcp structures. Atomic cluster units in metallic glasses are around 5 Å apart, similar to interpolyhedral IRO distances in network glasses. The interatomic correlations between neighboring cluster units are identifiable out to around 15 Å (Figure 2), similar to the establishment of LRO in network glasses. In these densely packed metallic structures, however, the geometry of bond angles and dihedral angles and ring topology is absent. The sequence from IRO to LRO is usually collectively described as MRO [4], but is less well understood than in network glasses.
Furthermore, compared to supercooled network systems, where high shear viscosity and low atomic diffusion stem from the existence of open structures, the glass‐forming ability of densely packed metallic melts is imprecisely understood. In searching for melt compositions that are suitably viscous for conventional glass quenching, those associated with deep eutectics can be a guide, but not exclusively so [4, 23]. An overriding requirement, though, resides in achieving the highest atomic packing density in the supercooled state, which is often achieved by “dissolving” smaller atoms. The atomic size ratio for solute to solvent atoms, which yields the most efficient packing, is frequently found to be approximately 0.9 [4].
Figure 7 Simple two‐dimensional models of glass structure created from sparse (a) and dense (b) packing of spheres. (a) The continuous random network (CRN) model of Zachariasen [13] representing a network glass comprising threefold coordinated cations and bridging anions. Different ring sizes (3, 4, 5, 6, 7, 8) perpetuating extended range order are shown. (b) The dense random packing of atomic spheres (DRPHS) model of Bernal [25] showing variations in icosahedral packing, viz. 5, 6, and 7, that promote homogeneous noncrystalline extended range order.
For complicated high‐density geometries, different packing arrangements in metallic glasses are now modeled with computer simulations – mainly RMC, but also ab initio MD – the aim being to analyze the variety and number of different Voronoi polyhedra present [23]. For an atomic size ratio of 0.9, SRO is predominantly icosahedral, the geometry for tessellated quasicrystal structures. In glasses with pronounced chemical order and atomic size ratio lower than 0.9, pentagonal biprisms replace icosahedra as the dominant Voronoi polyhedra. Because they embody fivefold symmetry, icosahedrons and pentagonal biprisms frustrate crystalline close packing and represent the geometric counterparts to odd‐membered rings in directionally bonded glasses with open structures.
Metallic glasses, like network glasses, are less dense than their crystalline counterparts, the additional free volume being a throwback from the configurational diversity of the supercooled liquid. In contrast to that of network glasses, though, the FSDP in metallic glasses (Figure 3) is considered to derive from scattering from voids interspersed within the SRO of atomic clusters (Figure 7). In Ca–Mg–Cu glasses, QFSDP is, for example, about 1.2 Å−1, and the FSDP correlation length about 5 Å (Figure 3). In multicomponent alloys QFSDP is affected by the size distribution of the different metals and its intensity by density. In Ni–Zr–Al, QFSDP is, for example, about 0.9 Å−1 with a FSDP correlation length of about 7 Å.
Also found in metallic glasses, the excess THz modes at the onset of the VDOS and linear low‐temperature specific heat first observed in oxide glasses at THz frequencies [7] appear to have a common origin in the behavior