Glasses, disorder and glass transition

A glass lacks long-range crystal periodicity

Two pieces of the same glass composition reach the same final temperature, yet have slightly different specific volumes. Must their masses or compositions differ? Not necessarily: their cooling histories can leave different retained structures. Specific volume v means volume divided by mass, here in cubic centimetres per gram (cm³/g); its reciprocal is density in g/cm³.

On cooling a liquid, a glass can form when crystallisation is avoided and structural rearrangement becomes too slow to follow equilibrium. Local bonding remains. Its state depends on composition and thermal history, and it can relax or crystallise under suitable later conditions.

In ordinary silica glass, silicon is approximately fourfold coordinated by oxygen in SiO₄ tetrahedra, linked through shared oxygen atoms. Nearby bonding and preferred distances survive. What is absent is the ordinary repeating arrangement of the whole network found in a silica crystal such as quartz. Sharing four corner oxygens, each between two silicon centres, gives four halves, or two oxygens per silicon: SiO₂ is a composition, not a collection of isolated SiO₂ molecules. This ideal local description does not specify every bond angle, ring or defect.

Disorder is therefore not empty space, missing bonds or a mass of tiny crystals. A broad diffraction pattern can retain evidence of local correlations; undetected small crystals remain a measurement question. “Crystalline, amorphous and molecular arrangements” supplies the structural preparation, and “Microscopy, diffraction and structure evidence” develops the detection limits. Here the task is to follow a glass-forming liquid as its structure ceases to keep up with cooling.

A supercooled liquid, a glass and a crystal take different paths

Consider a notional single-component glass former at fixed pressure. Its equilibrium crystal and liquid coexist at a melting temperature Tₘ. Melting a crystal there involves latent heat, an enthalpy difference between the phases. The two phases can also have different volumes; the sign and size of that volume change are material dependent.

If crystallisation is avoided on cooling below Tₘ, the liquid is supercooled: it retains liquid-like structural rearrangement although a crystal is thermodynamically favoured. Forming a crystal requires nucleation and growth. Avoiding that process is a separate requirement from slowing the liquid’s own rearrangements. Rapid cooling is one route to glass formation, with no one cooling rate suitable for every composition, size and surface condition.

In the original model below, Tₘ = 650 °C is stipulated, not a value for silica or a laboratory recipe. The chosen crystal branch has v = 0.430 + 0.000100(T − 400); the relaxed liquid branch has v = 0.450 + 0.000300(T − 400), with T entered in °C and v in cm³/g. Each coefficient has units cm³/(g·K); a 1 °C temperature difference is a 1 K difference. Above Tₘ the stable branch is liquid; below it the liquid branch is a metastable reference if crystallisation is suppressed.

At 650 °C these branches give 0.455 and 0.525 cm³/g. The chosen crystal-to-liquid change is a jump of 0.070 cm³/g. By contrast, each glass-cooling path below is continuous where its slope changes. A kink in a volume trace is not that melting jump. For multicomponent materials, crystallisation and melting may involve ranges and different phase compositions; a sharp single Tₘ is the stipulated comparison here.

The glass transition has a timescale

Across a glass-transition range, molecular or structural mobility changes enough to alter response strongly. The observed transition depends on the measurement and timescale; it is not the same as a single equilibrium melting point of a crystal. Some polymers contain both glassy amorphous regions and crystals.

Let τ be a characteristic time for the structural rearrangement being probed, and t_obs the time available to observe or permit that rearrangement. Both must refer to the same process and have the same time units. When τ is much shorter than t_obs, the structure can substantially adjust; when τ is much longer, little adjustment is seen. At comparable times, the measured response changes strongly. Atomic vibrations still occur in the glassy state; “frozen” describes the slow structural response on the chosen clock, not absolute immobility.

A cooling scan does not remain at one temperature. For a stated comparison window ΔT, define t_window = ΔT/q, where q is the positive magnitude of the cooling rate in K/s. This is a useful time estimate, not a universal definition of T_g. For the invented 20 K window, slow cooling at q_S = 0.20 K/s gives 100 s, whereas fast cooling at q_F = 2.0 K/s gives 10 s.

The table supplies invented characteristic structural times for the same chosen starting protocol. For this simplified construction only, set the crossover where τ = t_window. The slow path crosses at 470 °C; the fast one at 510 °C, while rearrangements are still faster. Thus faster cooling departs from the relaxed-liquid path at a higher temperature in this model. The table does not establish a universal relaxation law or measured transition onset.

For the actual two-laboratory heating question, retain the prior cooling/holding history as well as the heating rate. A dilatometric slope intersection, a calorimetric onset or midpoint, and a frequency-dependent mechanical response are different definitions and signals. Instrumental temperature lag also needs checking. The cooling construction explains the timescale issue; it does not predict the laboratories’ heating values or justify a universal direction of every onset shift.

Read a retained-structure difference at the same temperature

Represent each cooling path by two straight branches. Above its crossover it follows the common relaxed-liquid line. Below 470 °C, the slow path is v_S = 0.464 + 0.000100(T − 400); below 510 °C, the fast path is v_F = 0.472 + 0.000100(T − 400). All specific volumes are in cm³/g. Each glass branch meets the relaxed-liquid line exactly at its stated crossover. These sharp joins approximate a broad transition; they are neither measured discontinuities nor a numerical solution of changing structure during the scan.

At 400 °C the slow and fast paths have v_S = 0.464 and v_F = 0.472 cm³/g. Equal 10.0 g pieces therefore occupy 4.64 and 4.72 cm³. The retained-state volume difference is 0.08 cm³ at the same temperature and mass. In this illustration slower cooling permits a lower-volume retained state. That ranking is a feature of the supplied branches, not a claim that every glass property increases or decreases with slower cooling.

Within either frozen-structure branch, warming from 400 to 450 °C adds 0.000100 × 50 = 0.00500 cm³/g without the structural adjustment represented by the steeper liquid line. The liquid-reference increase over that interval is 0.0150 cm³/g. Slope has units of specific-volume change per temperature change. A volumetric expansion coefficient would instead divide the slope by the current specific volume; the drawn slope alone is not that coefficient.

The temperature at which a low-temperature line extrapolates to the relaxed-liquid line is sometimes used to describe a retained structural state through a fictive temperature. It is not necessarily a directly reported heating T_g, and one such number need not identify the whole structure or its later response. Different structural processes can have different relaxation times. Do not interpret the two-line model as a complete state description.

A constant temperature can still produce a changing volume

Take the fast-history model specimen rapidly to 470 °C, with its initial structure retained during the transfer. Its initial specific volume at that temperature is 0.479 cm³/g; the metastable relaxed-liquid reference there is 0.471 cm³/g. For a spatially uniform, fixed-pressure hold, assume no crystallisation, composition change or mass exchange. The excess x = v − v∞ starts at x₀ = 0.008 cm³/g, where v∞ = 0.471 cm³/g is the chosen long-time reference.

Adopt a deliberately simple local relaxation model, dx/dt = −x/τ, with constant τ = 100 s. The negative sign makes a positive excess decrease. Separating variables gives dx/x = −dt/τ, so integrating from x₀ at time zero gives ln(x/x₀) = −t/τ and x(t) = x₀ exp(−t/τ). Thus v(t) = 0.471 + 0.008 exp(−t/100), with t in seconds. The exponent is dimensionless. If logarithms are unfamiliar, read the equation as: after every 100 s, the remaining excess is multiplied by exp(−1) ≈ 0.368.

After 300 s, the remaining excess is 0.0003983 cm³/g and v ≈ 0.471398 cm³/g. About 95.0% of the initial excess has relaxed, but only about 1.59% of the initial total specific volume has been lost. “95% relaxed” is not “95% smaller”, nor a transformed crystal fraction. Because temperature stays fixed, the model’s changing volume is structural response, rather than the ordinary expansion term from changing temperature.

Real evidence for a history-dependent property can be obtained by returning differently held samples to the same measurement temperature. In one lead-metasilicate study, refractive-index measurements at common room-temperature conditions followed different isothermal holds; the index changed with the hold history. That is a structural probe, not proof that the specimen must have crystallised. The study used stretched relaxation and found that a viscosity-based shear-relaxation time did not simply equal its measured structural-relaxation time.

The single-time equation here is an original teaching approximation for this stated hold. In a real glass, relaxation can depend on the evolving structure and involve a distribution of times; changing temperature, starting history or observable can require another model. Structural relaxation can alter volume or enthalpy without forming crystals. Crystallisation needs separate evidence for crystal formation, and either process may affect a useful property. A treatment to reduce residual stress also requires its own temperature uniformity and stress-response assessment.

Decide what the measurement has established

A transition report should identify composition, preparation and prior thermal path; local specimen temperature versus time; the measured property and its calibration; the chosen onset, midpoint, intersection or other criterion; and relevant repeatability or uncertainty. Matching furnace settings alone does not show that two specimens followed the same local path.

Use comparisons that isolate the proposed cause. To test a history effect, compare at the same measurement temperature and pressure, with mass and composition controlled. To claim crystallisation, add a structural method with appropriate detection capability. To claim a usable optical, thermal or mechanical response, measure that response under the required conditions: a glass-transition number is not an allowable service temperature or a lifetime guarantee.

“Temperature, time and structural change” and “Equilibrium tendencies and transformation rates” provide the thermal-path preparation. “Metastability and persistence” is a useful return path for distinguishing a retained state from equilibrium. The later “Viscoelasticity, temperature and time” lesson develops mechanical creep and stress relaxation, where the controlled load or deformation is a separate part of the experiment.

Worked examples

Try the reasoning

1. Does amorphous mean bonds are absent?

   Show the answer and reasoning

   No.

   It describes absence of long-range periodic arrangement. In the ideal corner-sharing SiO₄ network, four shared oxygens per silicon count as two oxygens per silicon overall. Local coordination and SiO₂ stoichiometry do not establish a periodically repeated network.

2. Can thermal history affect glass properties?

   Show the answer and reasoning

   Yes.

   Relaxation and retained structure depend on the path. The original cooling model gives 0.464 versus 0.472 cm³/g at 400 °C for equal composition and pressure. This is a concrete possible history effect, not a universal ranking of every glass property.

3. Use the same invented τ table and 20 K window, but cool at 0.50 K/s. Under the stated equality criterion and monotonic τ between tabulated temperatures, bound the crossover. Can the table alone give an exact temperature?

   Show the answer and reasoning

   The window lasts 40 s. Since τ falls from 100 s at 470 °C to 31.62 s at 490 °C, the crossover lies strictly between 470 and 490 °C. No exact value follows without an interpolation or rate law.

   Compare times, not the cooling rate directly with τ. The tabulated endpoints bracket 40 s. A straight line through plotted data would be an additional assumption, not a measured law.

4. On the slow frozen-structure branch, warm from 400 to 450 °C without structural adjustment. Find Δv and the approximate volumetric expansion coefficient at 400 °C. A student reports 0.000100 K⁻¹ merely by copying the slope. Diagnose the units.

   Show the answer and reasoning

   Δv = 0.00500 cm³/g. At 400 °C, (1/v)(dv/dT) = 0.000100/0.464 ≈ 2.16 × 10⁻⁴ K⁻¹. The given slope has units cm³/(g·K), not K⁻¹.

   The coefficient is a fractional volume change per temperature change. Dividing the specific-volume slope by the current specific volume supplies that normalisation. It is a local value for the specified structural branch, not the effective slope while structure is also relaxing.

5. During the constant-τ hold, compare the volume changes from 0 to 100 s and from 100 to 200 s. Does the first interval’s average rate continue unchanged?

   Show the answer and reasoning

   The decreases are about 0.005057 and 0.001860 cm³/g. Their magnitudes per second are about 5.06 × 10⁻⁵ and 1.86 × 10⁻⁵ cm³/(g·s). The first rate does not persist.

   The volumes are 0.479000, 0.473943 and 0.472083 cm³/g. The rate −x/τ decreases in magnitude as the excess decreases. The second interval’s decrease is exp(−1) times the first, even though temperature and τ are constant.

6. For a different independently justified local hold, v∞ = 0.500 cm³/g, v₀ = 0.506 cm³/g and constant τ = 200 s. Use the same exponential form to find v at 400 s and the percentage of the initial excess removed. Can the old 470 °C temperature or its time constant be transferred to this new material?

   Show the answer and reasoning

   v = 0.500 + 0.006 exp(−2) ≈ 0.500812 cm³/g; about 86.5% of the initial excess is removed. The old material’s temperature and τ do not transfer.

   The new ratio t/τ is 2, not 4. Using 100 s from the earlier specimen would impose a different process. The law and new parameters are supplied for this local model; they do not turn 86.5% excess relaxation into a crystal fraction or a percentage total-volume loss.

7. A specimen’s refractive index changes after a hold. Both measurements are made at the same temperature and wavelength; no diffraction measurement is supplied. A colleague calls the index change proof of crystallisation and of a safe service temperature. Which inference is justified?

   Show the answer and reasoning

   There is a history-dependent optical-property change under the stated measurement conditions. Neither crystallisation nor a safe service temperature is established by that observation alone.

   Structural relaxation can change a glass property without producing a crystal. Check composition, surface and measurement alternatives, and use suitable structural evidence for a crystallisation claim. A service endpoint also requires its own loads, environment, duration and response requirements.

What to carry forward

• A glass has structure without long-range periodicity.
• Report the method and timescale of glass-transition measurements.
• Separate avoiding crystallisation from losing the ability to follow the relaxed-liquid structure on the observation clock.
• A continuous volume trace can change slope during glass transition; the stipulated crystal-melting jump is a different event.
• State the observable, thermal history and criterion with T_g. A fictive-state parameter is not automatically a heating onset or a complete structural description.
• Normalise a relaxation fraction to its stated excess; it is neither total volume lost nor a crystal fraction.