Skip to content
LumenaryAcademy
Structure 70 min

Glasses and nanomaterials

Solids without long-range order, the glass transition, silicate and metallic glasses, and what changes when a material is only a few nanometres across.

Code tested · 6AI-draftedAwaiting review by a materials scientist

Your guides: Chloe · The Materials Scientist, Margaret · The Tester, Gideon · The Failure Analyst

MotivationH

Order lost, and order shrunk

Every lesson so far has relied on order: atoms on a lattice, repeating for billions of spacings. This lesson looks at two ways of breaking that assumption.

The first is to lose the long-range order altogether. Window glass, the cover of a phone, an optical fibre carrying data under an ocean, and a transformer core made of an iron-based alloy ribbon are all glasses: solids whose atoms are as jumbled as in a liquid, frozen in place. They have no grains, no grain boundaries and no dislocations of the usual kind, which gives them properties no crystal has.

The second is to keep the order but make the piece tiny. In a gold particle 2 nm across, more than half the atoms are on the surface. Such particles are not gold-coloured but red or purple, melt hundreds of degrees below bulk gold, and catalyse reactions that bulk gold will not. Roman glassmakers in the fourth century, without knowing why, used gold and silver nanoparticles to make the Lycurgus Cup in the British Museum, which looks green when lit from the front and red when lit from behind. These are nanomaterials, and with them come carbon nanotubes, graphene and quantum dots, and serious questions about safety that this lesson takes seriously.

IntuitionH

The structure of a glass

A crystal has long-range order: know the positions of a few atoms and you can predict where every other atom is, millions of spacings away. A glass has only short-range order. Each atom has roughly the right number of neighbours at roughly the right distance (each silicon in silica glass still has four oxygens around it, as in quartz), but beyond a few neighbours the arrangement becomes unpredictable.

Definition— glass, amorphous solid

An amorphous solid has no long-range order in the positions of its atoms. A glass is an amorphous solid that has been formed by cooling a liquid without crystallising, and that shows a glass transition (below) when reheated.

Definition— radial distribution function

The radial distribution function g(r)g(r) is the average density of atoms at distance rr from a given atom, divided by the average density of the whole material. g(r)=1g(r) = 1 means "as many neighbours as in a random arrangement". It can be measured by X-ray or neutron scattering, from the full pattern including the diffuse scattering between Bragg peaks.

The program builds two two-dimensional solids at the same packing: a crystal of equal disks, and a glass of two sizes of disk, mixed and melted then suddenly cooled. Mixing two sizes is a standard trick: the sizes do not fit any simple lattice, so the mixture fails to crystallise, just as many metallic glasses are made from several elements of different sizes. Both are made with the Metropolis Monte Carlo method: pick a disk, try a small random move, always accept it if it lowers the energy, and accept it with probability e−ΔE/kBTe^{-\Delta E/k_BT} if it raises it. Repeated many times, this samples the arrangements a system visits at temperature TT.

python
import numpy as np

rng = np.random.default_rng(2026)

def monte_carlo(pos, sigma, box, T, sweeps, step=0.12):
    """Metropolis Monte Carlo for soft disks, energy (sigma_ij / r)^12, in a periodic box."""
    n = len(pos)
    for _ in range(sweeps):
        for i in rng.permutation(n):
            others = np.arange(n) != i
            s2 = ((sigma[i] + sigma[others]) / 2) ** 2
            def energy(p):
                d = pos[others] - p
                d -= box * np.round(d / box)          # nearest periodic image
                return np.sum((s2 / np.sum(d * d, axis=1)) ** 6)
            trial = pos[i] + step * (rng.random(2) - 0.5)
            dE = energy(trial) - energy(pos[i])
            if dE < 0 or rng.random() < np.exp(-dE / T):
                pos[i] = trial % box
    return pos

def g_of_r(pos, box, r_max=4.0, nbins=40):
    """Radial distribution function: neighbours at distance r, relative to a random arrangement."""
    d = pos[:, None, :] - pos[None, :, :]
    d -= box * np.round(d / box)
    r = np.sqrt(np.sum(d * d, axis=2))[np.triu_indices(len(pos), 1)]
    hist, edges = np.histogram(r, bins=nbins, range=(0, r_max))
    shell = np.pi * (edges[1:] ** 2 - edges[:-1] ** 2)
    expected = shell * len(pos) * (len(pos) - 1) / 2 / (box[0] * box[1])
    return (edges[1:] + edges[:-1]) / 2, hist / expected

def averaged_g(pos, sigma, box, T, snapshots=8, gap=4):
    total = 0
    for _ in range(snapshots):
        pos = monte_carlo(pos, sigma, box, T, gap)
        r, g = g_of_r(pos, box)
        total = total + g
    return r, total / snapshots

n_side = 12
N = n_side**2
a = 1.1                                           # crystal: equal disks of diameter 1, spacing 1.1
box_c = np.array([n_side * a, n_side * a * np.sqrt(3) / 2])
crystal = np.array([((i + 0.5 * (j % 2)) * a, j * a * np.sqrt(3) / 2) for j in range(n_side) for i in range(n_side)])
crystal = monte_carlo(crystal, np.ones(N), box_c, T=0.05, sweeps=20)

# glass: half small (diameter 1) and half large (1.4) disks at the same area fraction,
# mixed at random on a square grid, melted hot, then quenched cold
sigma = rng.permutation(np.where(np.arange(N) < N // 2, 1.0, 1.4))
area_fraction = N * np.pi / 4 / (box_c[0] * box_c[1])
side = np.sqrt(np.sum(np.pi * sigma**2 / 4) / area_fraction)
box_g = np.array([side, side])
glass = np.array([(i + 0.5, j + 0.5) for j in range(n_side) for i in range(n_side)]) * side / n_side
glass = monte_carlo(glass, sigma, box_g, T=5.0, sweeps=60, step=0.4)     # a hot liquid
glass = monte_carlo(glass, sigma, box_g, T=0.05, sweeps=30)              # quenched

r, g_c = averaged_g(crystal, np.ones(N), box_c, 0.05)
_, g_g = averaged_g(glass, sigma, box_g, 0.05)
print(f"{N} disks each, area fraction {area_fraction:.3f}; g = 1 means 'as many neighbours as at random'")
print("   r    g crystal               g glass")
for x, gc, gg in zip(r, g_c, g_g):
    if 0.75 < x < 3.8:
        print(f" {x:4.2f}  {gc:5.2f} {'#' * int(round(3 * gc)):18} {gg:5.2f} {'#' * int(round(3 * gg))}")
far = (r > 2.5) & (r < 3.8)
print(f"\nfor 2.5 < r < 3.8: g varies by +-{np.std(g_c[far]):.2f} in the crystal, +-{np.std(g_g[far]):.2f} in the glass")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.
crystalglass
Snapshots of the two solids made by the program, each of 144 disks at the same packing. The crystal (left) has every disk in a hexagonal pattern that repeats across the box. The glass (right), a mixture of two sizes, is dense and every disk is caged by touching neighbours, but there is no repeating pattern.

The crystal's g(r)g(r) is a series of sharp spikes at the lattice distances (1.1, 1.9, 2.2, 2.9, 3.3, …) with nothing in between, and the spikes continue as far as you look. The glass's g(r)g(r) has a broad first peak near the contact distances (1.0 to 1.4, since there are three kinds of pair), a weaker second peak, and then settles towards 1: beyond about three diameters, its fluctuations are a tenth of the crystal's. That is short-range order without long-range order, and it is exactly what X-ray and neutron measurements of real glasses show. In 1932 William Zachariasen proposed that oxide glasses are continuous random networks: silica glass is made of the same SiO₄ tetrahedra as quartz, joined at corners, but with Si–O–Si angles that vary, so rings of different sizes form and the network never repeats.

DefinitionsH

The glass transition

Cool a liquid and, normally, it crystallises at its melting point TmT_m, with a sudden drop in volume. But crystallising takes time: atoms must find the crystal arrangement and a nucleus must form (Phase transformations). If the liquid is cooled fast enough, or its structure makes crystallisation slow, it passes below TmT_m as a supercooled liquid. Its viscosity rises enormously as it cools, until the atoms cannot rearrange on the time scale of the experiment. The structure freezes: it has become a glass.

liquidsupercooled liquidfast-cooled glassslow-cooled glasscrystalTmTg fastTg slowtemperaturevolume
Volume against temperature on cooling (a schematic). A liquid that crystallises shrinks suddenly at Tm. A liquid that avoids crystallising continues along the supercooled-liquid line until its atoms can no longer keep up with the cooling, and then follows a shallower glass line. The faster the cooling, the higher the temperature at which it falls out of equilibrium, so Tg depends on cooling rate and a fast-cooled glass is less dense.

Definition— glass transition temperature

The glass transition temperature TgT_g is the temperature at which a supercooled liquid, on cooling at a given rate, falls out of equilibrium and becomes a glass. It shows up as a change in the slope of volume or enthalpy against temperature, and as a step in the heat capacity measured by differential scanning calorimetry (Spectroscopy and thermal analysis). A common practical definition is the temperature at which the viscosity reaches about 101210^{12} Pa s, roughly 101510^{15} times that of water.

The viscosity of many liquids near TgT_g rises faster than the Arrhenius law η∝eE/kBT\eta \propto e^{E/k_BT} that describes ordinary activated processes; it is often fitted by the Vogel–Fulcher–Tammann form η=η0 eB/(T−T0)\eta = \eta_0\,e^{B/(T - T_0)}, which would diverge at a temperature T0T_0 below TgT_g. Liquids that follow Arrhenius closely, such as silica, are called strong; those whose viscosity shoots up steeply near TgT_g, such as many polymers and molecular liquids, are fragile (a classification due to Austen Angell). The glass-blower's craft depends on silicate glasses being strong: they soften gradually over hundreds of degrees, so there is a wide working range.

ExplanationH

Silicate glasses

Pure silica, SiO₂, makes an excellent glass (fused silica): transparent from the ultraviolet to the infrared, with tiny thermal expansion, about 0.5×10−60.5\times10^{-6} per kelvin, and resistant to almost everything. But its TgT_g is around 1200 °C and it must be worked well above that, which is expensive. Most glass is therefore silica modified by other oxides. In the language of Zachariasen's network, the oxides are of three kinds:

  • Network formers (SiO₂, B₂O₃, P₂O₅) build the corner-linked network.
  • Network modifiers (Na₂O, K₂O, CaO, MgO) break it: each Na₂O adds an oxygen that turns one bridging Si–O–Si link into two non-bridging Si–O⁻ ends, with Na⁺ ions sitting nearby. Fewer links mean a lower viscosity and a lower working temperature, but also higher expansion and lower chemical durability.
  • Intermediates (Al₂O₃, TiO₂) can play either role.

Ordinary soda–lime–silica glass, used for windows and bottles, is roughly three-quarters SiO₂ with soda and lime; borosilicate glass, with boron oxide, expands about a third as much as soda–lime glass and survives sudden heating and cooling, which is why laboratory glassware and ovenware are made of it. The chemistry of silicate structures is in the solid state and main-group chemistry.

Glass is strong in compression and in theory very strong in tension, but in practice breaks from tiny surface flaws (Fracture and toughness). Two treatments put the surface into compression so that flaws stay closed: thermal tempering, rapid air cooling that freezes the surface while the inside is still soft, and chemical strengthening, in which glass is soaked in molten potassium salt so that larger K⁺ ions replace Na⁺ ions in the surface and wedge it into compression. Chemically strengthened aluminosilicate glasses from several makers, including Corning in the United States, AGC in Japan and SCHOTT in Germany, are used for the cover glass of most smartphones. How glass is melted, formed and treated is in Ceramics and glasses.

ExplanationH

Metallic glasses

Metals crystallise very easily: the atoms are nearly spherical and simply need to stack. To make a glassy metal, the liquid must be cooled faster than the crystals can form. In 1960 Pol Duwez and his students at Caltech did it by splatting droplets of a gold–silicon alloy onto a cold copper surface, at roughly a million kelvin per second. For decades metallic glasses could only be made as thin ribbons, foils or powders.

The breakthrough came from alloy design. Alloys of three or more elements with very different atomic sizes and a deep eutectic (a low melting point for the mixture; see Phase diagrams) crystallise sluggishly, because the atoms must sort themselves out before a crystal can grow. Akihisa Inoue's group in Japan found many such compositions from the late 1980s, and in 1993 Atakan Peker and William Johnson at Caltech reported a zirconium–titanium–copper–nickel–beryllium alloy that forms a glass at cooling rates of only a few kelvin per second, in rods centimetres thick. These are bulk metallic glasses.

Without grains or ordinary dislocations, metallic glasses can be very strong, often around 2 GPa for zirconium-based glasses, and can stretch elastically by about 2 per cent before yielding, several times more than a crystalline alloy, so they store and return elastic energy well. They are also corrosion-resistant and can be moulded like plastics when warmed above TgT_g. Their weakness is that plastic deformation concentrates in thin shear bands, so under tension a bulk metallic glass usually fails suddenly with little ductility. Iron-based glassy ribbons are soft magnets with very low losses and are used in efficient distribution-transformer cores (Magnetic materials); bulk glasses appear in precision springs, small gears, hinges and sports equipment.

ExplanationH

Nanomaterials: when the surface takes over

Definition— nanomaterial

A nanomaterial has at least one external dimension, or an internal structure, in the nanoscale, about 1 to 100 nm (the usage in the ISO 80004 vocabulary). Regulators add details: the European Union's 2022 recommendation counts a material as a nanomaterial if at least half of its particles, by number, have an external dimension in that range.

Two things change at this scale. The fraction of atoms at the surface becomes large, and when sizes approach the wavelength of electrons or light, quantum and optical size effects appear. The program counts the surface atoms of magic-number clusters: cuboctahedra of FCC atoms built shell by shell around one central atom (13, 55, 147, 309, … atoms). It finds each atom's coordination number, and estimates how much each cluster's melting point should drop by "bond counting", assuming that what holds the cluster together is proportional to its average number of neighbours.

python
import numpy as np
from itertools import product

def cuboctahedron(shells):
    """FCC atoms of a cuboctahedral cluster with the given number of shells around a central atom.
    Coordinates are integers with x + y + z even; nearest neighbours are sqrt(2) apart."""
    m = shells
    pts = [p for p in product(range(-m, m + 1), repeat=3)
           if sum(p) % 2 == 0 and sum(map(abs, p)) <= 2 * m]     # square faces |x| = m, triangles |x|+|y|+|z| = 2m
    return np.array(pts, float)

def coordination(pts):
    d2 = np.sum((pts[:, None, :] - pts[None, :, :]) ** 2, axis=2)
    return np.sum(np.abs(d2 - 2) < 1e-9, axis=1)             # neighbours at distance sqrt(2)

d_nn = 0.2884                                                  # gold nearest-neighbour distance, nm
print("shells  atoms   diameter/nm   on surface   mean CN   bond-counting Tm/Tm(bulk)   Tm estimate for gold/K")
for m in [1, 2, 3, 4, 5, 6, 8, 10]:
    pts = cuboctahedron(m)
    cn = coordination(pts)
    surface = np.mean(cn < 12)
    ratio = cn.mean() / 12
    print(f"  {m:3}   {len(pts):5}     {2 * m * d_nn:6.2f}       {100 * surface:5.1f} %    {cn.mean():6.2f}"
          f"           {ratio:.3f}                  {ratio * 1337:6.0f}")

print("\nexact formula check: atoms (10m^3 + 15m^2 + 11m + 3)/3 and surface atoms 10m^2 + 2")
for m in [1, 2, 3, 10]:
    print(f"  m = {m}: {(10 * m**3 + 15 * m**2 + 11 * m + 3) // 3} atoms, {10 * m**2 + 2} on the surface")

# A sphere of diameter D with a surface shell one atom thick (atom diameter delta): surface fraction
delta = d_nn
print("\nsphere diameter/nm   fraction of atoms in the outer layer")
for D in [1, 2, 5, 10, 20, 100, 1000]:
    print(f"   {D:7}               {1 - (1 - 2 * delta / D) ** 3:7.3f}")

# Surface area per gram of spheres of density rho and diameter D: 6 / (rho D)
rho_gold = 19.3e6                                              # g per m^3
for D in [10, 100]:
    print(f"gold spheres {D} nm across: {6 / (rho_gold * D * 1e-9):.1f} m^2 per gram")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

A 55-atom gold cluster, about 1.2 nm across, has three-quarters of its atoms on the surface; a 3871-atom cluster, 5.8 nm across, still has a quarter. For spheres the outer layer holds 16 per cent of the atoms at 10 nm and under 2 per cent at 100 nm. Surface atoms have fewer neighbours, so they are bound less tightly, which is why nanoparticles are such good catalysts (gold, inert in bulk, catalyses the oxidation of carbon monoxide when finely divided) and why they melt at lower temperatures. Bond counting predicts that a 2.3 nm gold particle melts about 260 K below bulk gold's 1337 K. Measurements, starting with Philippe Buffat and Jean-Pierre Borel's electron-diffraction study in 1976, show the same strong trend, with gold particles a few nanometres across melting hundreds of kelvin below the bulk; simple models like this one capture the trend, not the exact numbers.

ExplanationH

Why gold nanoparticles are red

Bulk gold reflects red and yellow light and absorbs blue, which is why it looks golden. A suspension of gold particles 10 to 50 nm across is ruby red, and Michael Faraday, who made such "colloidal gold" at the Royal Institution in the 1850s, realised the colour came from gold in a finely divided state. The explanation is a localised surface plasmon: light's electric field pushes the free electrons of a small metal sphere to one side, the exposed positive ions pull them back, and the electron cloud oscillates. When the light's frequency matches this oscillation, the particle absorbs and scatters strongly.

For a sphere much smaller than the wavelength, the resonance happens where the metal's permittivity satisfies εmetal(ω)=−2εmedium\varepsilon_{\text{metal}}(\omega) = -2\varepsilon_{\text{medium}}. The program uses the simplest model of a metal's permittivity, the Drude model, ε(ω)=εb−ωp2/ω2\varepsilon(\omega) = \varepsilon_b - \omega_p^2/\omega^2, where the plasma frequency ωp=ne2/ε0me\omega_p = \sqrt{ne^2/\varepsilon_0 m_e} comes from the density nn of free electrons (one per atom, from the lattice parameter) and εb\varepsilon_b accounts for the bound d electrons. It then predicts how the colour shifts when the surroundings change.

python
import numpy as np

e, m_e, eps0 = 1.602176634e-19, 9.1093837e-31, 8.8541878128e-12
hbar_eV = 6.582119569e-16                  # eV s
hc = 1239.84198                            # eV nm

def plasma_energy(a_nm, electrons_per_atom=1, atoms_per_cell=4):
    """hbar * omega_p for free electrons in an FCC metal with lattice parameter a."""
    n = electrons_per_atom * atoms_per_cell / (a_nm * 1e-9) ** 3
    return hbar_eV * np.sqrt(n * e**2 / (eps0 * m_e)), n

# A small metal sphere absorbs most strongly where eps_metal = -2 eps_medium (the Froehlich condition).
# With a Drude metal, eps = eps_bound - (omega_p / omega)^2, this happens at omega_p / sqrt(eps_bound + 2 eps_medium).
for metal, a in [("gold", 0.40782), ("silver", 0.40862)]:
    Ep, n = plasma_energy(a)
    E_free = Ep / np.sqrt(1 + 2 * 1.0)        # free electrons only, sphere in vacuum
    print(f"{metal}: n = {n:.2e} electrons/m^3, hbar*omega_p = {Ep:.2f} eV")
    print(f"   free electrons only, in vacuum: resonance at {E_free:.2f} eV = {hc / E_free:.0f} nm (ultraviolet)")

# The bound d electrons screen the free ones. Measured: small gold spheres in water absorb most
# strongly near 520 nm. Work back to the background permittivity this implies:
Ep_gold, _ = plasma_energy(0.40782)
eps_water = 1.333**2
E_res = hc / 520
eps_bound = (Ep_gold / E_res) ** 2 - 2 * eps_water
print(f"\ngold in water at 520 nm ({E_res:.2f} eV) implies eps_bound = {eps_bound:.1f}")

# Now predict how the colour shifts with the surroundings (the basis of plasmonic sensors)
print("medium         refractive index   resonance/nm")
for medium, n_med in [("air", 1.000), ("water", 1.333), ("protein layer", 1.45), ("glass", 1.50)]:
    E = Ep_gold / np.sqrt(eps_bound + 2 * n_med**2)
    print(f"  {medium:14}     {n_med:5.3f}          {hc / E:5.0f}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Free electrons alone would put the resonance of both gold and silver spheres in the ultraviolet, near 240 nm, and both colloids would be colourless. What makes the difference is the bound d electrons. In gold they lie close in energy to the free electrons and screen them strongly, a background permittivity of about 11 in this model, pulling the resonance down to green light near 520 nm; gold colloids absorb green and transmit red. In silver the d electrons lie deeper, screening is weaker, and silver spheres absorb near 400 nm, giving yellow colloids. The same model shows the resonance moving to longer wavelengths as the surrounding refractive index rises: from about 520 nm in water to about 537 nm in glass. This shift is the principle of plasmonic sensors, and when gold particles clump together their resonances couple and the colour turns from red to purple and blue. The test line of many lateral-flow tests, including home pregnancy tests and the rapid antigen tests used during the COVID-19 pandemic, is coloured by gold nanoparticles caught on the line.

Semiconductor nanocrystals show a different size effect. In a crystal of cadmium selenide or indium phosphide a few nanometres across, the electrons and holes are confined, and the smaller the crystal, the larger the band gap and the bluer the light it emits. These quantum dots, whose discovery and synthesis earned Moungi Bawendi, Louis Brus and Alexei Ekimov the 2023 Nobel Prize in Chemistry, colour the pixels of some displays; they return in Optical materials.

ExplanationH

Carbon nanotubes and graphene, in brief

Roll a sheet of graphene, the single layer of graphite, into a seamless cylinder and you have a carbon nanotube. Multi-walled tubes were identified by Sumio Iijima in 1991 in the soot of carbon arcs, single-walled tubes in 1993. A tube is specified by the indices (n,m)(n, m) of the vector around its circumference, measured in graphene's lattice vectors; they fix its diameter, its twist (chiral angle), and remarkably, whether it conducts like a metal or a semiconductor. In 2004 Andre Geim, Konstantin Novoselov and colleagues in Manchester isolated graphene itself by peeling graphite with adhesive tape and showed that its electrons move exceptionally freely; they shared the 2010 Nobel Prize in Physics. In 2008 James Hone's group at Columbia pressed a diamond tip into suspended graphene and measured a Young's modulus of about 1 TPa and an intrinsic strength of about 130 GPa, the strongest material ever measured.

python
import numpy as np

a = 0.246                       # graphene lattice constant, nm (carbon-carbon bond 0.142 nm)

def nanotube(n, m):
    """Diameter (nm), chiral angle (degrees) and electronic type of an (n, m) carbon nanotube."""
    d = a * np.sqrt(n * n + n * m + m * m) / np.pi
    angle = np.degrees(np.arctan2(np.sqrt(3) * m, 2 * n + m))
    kind = "metallic" if (n - m) % 3 == 0 else "semiconducting"
    return d, angle, kind

print(" (n, m)    diameter/nm   chiral angle   type")
for n, m in [(5, 5), (10, 10), (9, 0), (10, 0), (12, 0), (6, 5), (7, 5), (8, 4), (13, 0)]:
    d, ang, kind = nanotube(n, m)
    print(f" ({n:2},{m:2})      {d:6.3f}        {ang:5.1f}       {kind}")

# Of all tubes with diameters between 0.7 and 2.0 nm, what fraction is metallic?
tubes = [(n, m) for n in range(1, 40) for m in range(0, n + 1) if 0.7 <= nanotube(n, m)[0] <= 2.0]
metallic = sum(nanotube(n, m)[2] == "metallic" for n, m in tubes)
print(f"\n{len(tubes)} distinct tubes between 0.7 and 2.0 nm; metallic: {metallic} ({100 * metallic / len(tubes):.0f} %)")

# Graphene's specific surface area: 2 atoms per hexagonal cell of area (sqrt(3)/2) a^2, counting both sides
NA, M_C = 6.02214076e23, 12.011
area_per_atom = np.sqrt(3) / 2 * (a * 1e-9) ** 2 / 2
print(f"graphene: {2 * area_per_atom * NA / M_C:.0f} m^2 per gram (both faces)")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The rule in the program, metallic when n−mn - m is a multiple of 3, comes from asking whether the allowed electron wavelengths around the circumference include the special points in graphene's band structure where its conduction and valence bands touch (Electronic structure and bands). About a third of tubes are metallic, 35 per cent of the 187 tubes between 0.7 and 2.0 nm, and the rest are semiconductors with gaps that shrink as the diameter grows. Strictly, only the "armchair" (n,n)(n, n) tubes are truly metallic: in the others that pass the rule, the curvature of the wall opens a tiny gap. Because growth produces a mixture, sorting tubes by type, or growing only one type, has been a central problem for nanotube electronics. Today nanotubes are used mainly as conductive additives in lithium-ion battery electrodes and in composites, and graphene-family materials in coatings, thermal films and sensors; their single-layer cousins are the subject of Two-dimensional and quantum materials.

ExplanationH

Nanoparticles and safety

The same features that make nanomaterials useful raise real questions about health and the environment, and they deserve a careful answer rather than either alarm or dismissal.

Why might "nano" matter for safety? Small particles have a large surface area per gram, and surface is where chemistry happens, so a dose measured in grams can be much more reactive as nanoparticles than as a lump. Inhaled particles a few tens of nanometres across reach the deepest parts of the lungs. And shape matters: in a 2008 pilot study in mice, Craig Poland and colleagues found that long, rigid multi-walled carbon nanotubes caused inflammation resembling that caused by asbestos fibres, while short or tangled ones did not. In 2014 the International Agency for Research on Cancer classified one type of multi-walled nanotube as possibly carcinogenic to humans (Group 2B), and the US National Institute for Occupational Safety and Health recommends keeping workplace exposure to carbon nanotubes below one microgram per cubic metre of air.

Equally, "nano" does not automatically mean dangerous. Risk depends on hazard and exposure: nanoparticles bound in a solid composite, a coating or a glass (like the gold in medieval ruby glass) are not released in ordinary use, while a dry powder handled openly can be inhaled. Assessments therefore look at a specific material in a specific use. In the European Union, chemical regulation (REACH) has required separate information for nanoforms of substances since 2020, and titanium dioxide was withdrawn as a food colouring (E171) in 2022 because the European Food Safety Authority could no longer rule out concerns about genetic damage; other uses of the same substance, such as in sunscreens and paints, are assessed separately. The OECD coordinates international test guidelines so that results from different countries can be compared.

ExplanationH

The professional view

Glass structure is measured by total scattering, X-ray or neutron diffraction analysed as a pair distribution function (the three-dimensional cousin of the g(r)g(r) above), and by solid-state NMR, which counts bridging oxygens. Compositions for displays, fibres, pharmaceutical vials and nuclear-waste glasses are designed with models fitted to large property databases and, increasingly as of 2026, with machine learning and molecular dynamics (Atomistic simulation).

Nanomaterials are characterised by transmission electron microscopy (size, shape and crystal structure of individual particles), X-ray peak broadening (the Scherrer analysis of X-ray diffraction), dynamic light scattering (sizes in suspension), BET gas adsorption (surface area) and optical spectroscopy (plasmon and quantum-dot resonances). Reference nanomaterials with certified sizes are supplied by national metrology institutes such as NIST in the United States and the European Commission's Joint Research Centre, so that measurements in different laboratories can be trusted.

Common misconceptionsH

Common misconceptions

Misconception

"Old cathedral windows are thicker at the bottom because glass flows slowly." Calculations of glass viscosity at room temperature give flow times far longer than the age of the universe. Medieval panes are uneven because the crown and cylinder glass-making methods of the time produced sheets of uneven thickness, and glaziers often, but not always, put the thick edge at the bottom.

Misconception

"A glass is a liquid." Below TgT_g a glass is a solid by every mechanical test: it has a shear modulus and does not flow on human time scales. It shares a liquid's structure, not its behaviour. "Amorphous solid" is the accurate description.

Misconception

"Nanoparticles are just small pieces of the bulk material." Below a few tens of nanometres, surface atoms, confinement of electrons and plasmon resonances change colour, melting point, catalytic activity and electronic behaviour. A 5 nm gold particle is, for many purposes, a different material from a gold bar.

Misconception

"Smaller grains always make a metal stronger." The Hall–Petch strengthening of Defects and dislocations works down to grain sizes of roughly 10 to 20 nm. Below that, grains are too small to hold piled-up dislocations, deformation shifts to the grain boundaries themselves, and the metal can get weaker: the inverse Hall–Petch effect.

Exercises

Exercises

1. Glasses and nanomaterials

Warm-upHuman

Silica glass and quartz are both made of SiO₄ tetrahedra joined at their corners. What is the essential difference between them?

Choose one

2. Glasses and nanomaterials

CoreHuman

The same glass-forming liquid is cooled twice: once quickly, once slowly. Compared with the slowly cooled glass, the quickly cooled one has…

Choose one

3. Glasses and nanomaterials

CoreHuman

The fifth magic-number cuboctahedral cluster has 561 atoms. How many of them are on its surface?

4. Glasses and nanomaterials

CoreHuman

Is a (12,3)(12, 3) carbon nanotube metallic or semiconducting, according to the simple zone-folding rule? Answer with one word.

5. Glasses and nanomaterials

CoreHuman

Small gold nanoparticles in water absorb most strongly near 520 nm and look red. Antibodies bind to their surface, adding a layer of higher refractive index. What happens to the resonance?

Choose one

6. Glasses and nanomaterials

CoreHuman

Which of these statements about the safety of nanomaterials are accurate? Select all that apply.

Choose all that apply

7. Glasses and nanomaterials

CoreComputational

Write nanotube(n, m) that returns a tuple (diameter, kind) for an (n,m)(n, m) carbon nanotube: the diameter in nm, d=an2+nm+m2/πd = a\sqrt{n^2 + nm + m^2}/\pi with a=0.246a = 0.246 nm, and kind, the string "metallic" if n−mn - m is divisible by 3 and "semiconducting" otherwise. The tests also count the metallic tubes among all (n,m)(n, m) with 1≤n≤301 \le n \le 30 and 0≤m≤n0 \le m \le n.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

8. Glasses and nanomaterials

StretchComputational

A dilatometer measures the specific volume VV of a glass at a series of temperatures TT. Below TgT_g the points lie on one straight line (the glass), above it on a steeper one (the supercooled liquid). Write glass_transition(T, V) that:

  1. sorts the points by temperature;
  2. tries every split into a low-temperature group and a high-temperature group with at least three points in each, fits a straight line to each group with np.polyfit, and keeps the split with the smallest total squared error;
  3. returns the temperature where the two best lines cross.

The tests use exact and noisy simulated data, in increasing and decreasing order, as arrays and as lists.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Further reading and resources