A phase diagram is not a collection of boundary lines drawn empirically. The starting point is the Gibbs free energy G(x,T,p) of each phase. Searching for the lowest free energy at each temperature determines the phase boundaries, and adding interfacial energy and the rate of mass transport to this takes us on to microstructure formation by the phase-field method. On this page we follow this single line of reasoning through equations and figures you can operate. We also cover the glass transition and crystallization (§7), in which the cooling rate decides between crystal and glass, and spinodal decomposition (§9), in which waves of composition grow without the formation of nuclei.
In a system whose temperature T and pressure p are held constant by the surroundings, a spontaneous change proceeds in the direction that lowers the total Gibbs free energy. When a material of composition x0 splits into an α phase and a β phase, the total free energy per 1 mol is
Here, the phase fractions and the mass balance of component B must always be satisfied:
The equilibrium problem is therefore the problem of finding the values of fα, fβ, xα, xβ that minimize Eq. (1) under the constraints of Eq. (2). This minimization will later become visible geometrically as the common tangent.
For 1 mol of a single phase, the total differential of the Gibbs free energy is
If the pressure and the amount of substance are fixed, it follows immediately that
In other words, the slope of the G–T curve is −S, and its curvature is −Cp/T. A liquid usually has a larger entropy than a solid, so the G of the liquid falls steeply with temperature, crosses the G of the solid at the melting point, and lies below that of the solid at higher temperatures.
If H and S at a reference temperature T0 are known, then
gives G(T). If a phase transition occurs along the way, add the transition enthalpy ΔHtr to H and a jump of ΔHtr/Ttr to S. The pure-substance databases of CALPHAD store this thermochemical information as analytical expressions for each phase.
Define ΔGfus = Gliquid − Gsolid. At the melting point Tm the two phases are in equilibrium, so ΔGfus(Tm) = 0, and ΔSfus(Tm) = ΔHm/Tm. If the difference in heat capacity between solid and liquid, ΔCp, is regarded as constant, integration gives
That is Eq. (6). The corresponding phase diagram simulator, to keep the mechanism easy to follow, sets ΔCp = 0 and uses the following linear expression.
For example, Al has Tm = 933.5 K and ΔHm = 10.711 kJ/mol. By Eq. (7), ΔGfus = +1.53 kJ/mol at 800 K, so the solid is stable; at 1000 K it is −0.763 kJ/mol, so the liquid is stable. The sign reverses at the melting point.
Fig. 1. Free energy of melting of a pure substance. Change the substance and the temperature, and check the sign of Eq. (7).
Suppose that components A and B mix substitutionally in a phase φ. Letting x be the mole fraction of B, the free energy per 1 mol can be divided into the “end members,” “ideal mixing,” and “deviation from ideality.”
The number of ways of placing NA A atoms and NB B atoms on N lattice sites is W = N!/(NA!NB!). From S = kB ln W and Stirling’s approximation we obtain
For 0 < x < 1, ΔSmix > 0, so −TΔSmix always stabilizes mixing. Moreover, since its coefficient is T, the temperature effect that the higher the temperature, the more readily the components mix is built in naturally.
In the phase diagram simulator, the deviation from an ideal solution is expressed as
The simplest regular solution has L0 = Ω, L1 = L2 = 0. Ω > 0 is the case in which the energy rises when unlike atoms A–B are neighbors, which promotes phase separation. Ω < 0 promotes mixing or ordering. L1 and the higher terms express the left–right asymmetry of the curve.
Minimizing Eq. (1) subject to Eq. (2) by the method of Lagrange multipliers shows that, between coexisting α and β phases, the chemical potential of each component is equal.
Using the tangent to the molar free energy G(x), we have
If the slope and the intercept of the tangent are the same, Eq. (11) is satisfied simultaneously. Therefore the points where one and the same straight line touches two curves give the compositions of the two coexisting phases. The same holds when the line touches a single curve at two points; this is the case in which the same crystalline phase separates into two phases of different compositions.
Solving Eq. (2) for the phase fractions gives
For example, if xα = 0.20, xβ = 0.80, and the overall composition is x0 = 0.35, then fα = 0.75 and fβ = 0.25. The phase composition and the phase fraction are different quantities.
Fig. 2. Free energy of a regular solution and the lower convex hull. Change the temperature and Ω to see two-phase separation, the common tangent, and the spinodal. The purple line is the lower convex hull.
For the regular solution Gmix = RT[x ln x + (1−x)ln(1−x)] + Ωx(1−x), the local stability is determined by the curvature.
The critical temperature of a symmetric regular solution is Tc = Ω/(2R), and the critical composition is x = 1/2. In Fig. 2, if you move T above Tc, the curve becomes convex everywhere and the common tangent disappears. How the separation proceeds inside the spinodal (which wavelengths of fluctuation grow) is derived in §9.
At a given temperature, draw the common tangent to the liquid GL(x) and the solid Gα(x), and record the points of tangency xL and xα. Repeating the same operation while changing the temperature slightly, xL(T) becomes the liquidus and xα(T) becomes the solidus.
| Type of phase diagram | What happens in the free energy | Typical reaction and microstructure |
|---|---|---|
| Complete solid solution | The liquid competes with a single solid-solution curve. The solid is stable over the whole composition range at high temperature, but if Ω is large and positive, a miscibility gap opens at low temperature | L → α. Composition differences can arise during solidification |
| Eutectic | At the eutectic temperature, a single tangent line touches the three curves L, α, and β simultaneously | L → α + β. Eutectic structures such as lamellar or rod-like ones |
| Peritectic | At the peritectic temperature, too, a single tangent line touches the three curves L, α, and β simultaneously. Unlike the eutectic, the order of the points of tangency is xα < xβ < xL, so the liquid lies outside the two solids | L + α → β. Diffusion through the solid tends to be rate-limiting |
| Monotectic (two-liquid separation) | The Ω of the liquid is a large positive value, the liquid curve itself bulges up in the middle, and a common tangent can be drawn within a single curve (L1 + L2). At the monotectic temperature, a single tangent line touches L1, α, and L2 at three points simultaneously | L1 → α + L2. Bearing alloys such as Cu–Pb |
| Line compound | A phase with no composition range joins the competition as a single point on the G–x diagram | Congruent melting and incongruent melting; multiple eutectics on either side of a compound |
For a binary system at constant pressure, Gibbs’ phase rule gives F = C − P + 1 = 3 − P. With three phases coexisting, F = 0, so the eutectic and peritectic temperatures and the phase compositions cannot be moved freely. This is why they appear on a phase diagram as horizontal invariant reactions.
So far we have assumed that the solid and liquid reach complete equilibrium at every temperature. In real solidification, however, diffusion in the solid is far slower, and the composition of the portion that solidified first remains frozen in. So, as the opposite limit, we make three assumptions: no diffusion occurs at all inside the solid, the liquid is always uniform, and local equilibrium holds only at the solid–liquid interface. When the solid fraction fS increases by dfS, the solute balance is
Here CL is the composition of the liquid and CS is the composition of the solid in equilibrium with it at the interface. Integrating this along the liquidus and solidus, two things happen. One is microsegregation (coring): the center, which solidified first, is rich in the high-melting-point component, and layers pile up in which the later-solidified outer regions are progressively richer in the low-melting-point component. The other is the nonequilibrium eutectic: in a system with a eutectic, since the liquid keeps getting richer in the low-melting-point component, even a composition that would end as a single phase in equilibrium finally reaches the eutectic temperature, and eutectic structure remains. Even in a complete-solid-solution system with no eutectic, the apparent solidus falls below the equilibrium solidus. Equilibrium cooling and Scheil are both extreme idealizations, and real solidification always falls between them. Homogenization heat treatment of cast material is done to smooth out these layers by diffusion.
What a phase diagram tells you is the phases in equilibrium at each temperature and their proportions. The microstructure seen under a microscope adds to that a history: in what order those phases formed. Figs. 2-7 (eutectic system) and 2-8 (peritectic system) of Understanding Inorganic Materials Chemistry (in Japanese) depict how the microstructure changes when a melt is cooled slowly. “On cooling” in simulator ① and the equilibrium cooling in ③ reproduce this way of drawing by calculation. The numerical examples below are the values for the simulator’s “A–B” schematic systems shaped like textbook Figs. 2-7 and 2-8 (D = 18 %, E = 55 %, F = 84 %; for the peritectic system, α end 10 %, P ≈ 50.83 %, end of the liquid ≈ 84.33 %).
(a) Primary crystals and eutectic structure. When a melt of composition X (hypoeutectic), on the A side of the eutectic point E and on the B side of D, is cooled, primary α crystallizes in the liquid where the liquidus is crossed and grows as the temperature falls. At the eutectic temperature TE, the remaining liquid (composition xE) turns simultaneously into α (xD) and β (xF), forming a eutectic structure in which the two phases alternate. The primary crystals remain as they are, so the proportions of the two constituents that can be told apart in the microstructure are determined by the lever rule just above TE.
For a composition X1 = 24 %, the primary α is 83.8 %, the eutectic structure is 16.2 %, and the α inside the eutectic structure is 43.9 %. The overall fraction of α is 83.8 + 16.2 × 0.439 = 90.9 %, which agrees with the value obtained by applying the lever rule between α and β just below TE, (xF − X)/(xF − xD) = 90.9 % (combining the three equations above shows that this agreement holds in general). Note that the fractions of phases (α and β) and the fractions of constituents (primary crystals and eutectic structure) are different things. For a hypereutectic composition (between E and F), the roles of α and β are exchanged, and the microstructure is primary β + eutectic structure.
(b) Precipitation below the solvus. Below the solvus (solid solubility limit line), the amount that can stay in solid solution decreases as the temperature falls, so the part that has become supersaturated precipitates as a second phase. For the composition X2 of Fig. 2-7, β crystallizes in the liquid at T2 and becomes a β polycrystal at T3, and below T4 α precipitates inside the β grains and along the grain boundaries. The fraction of α that precipitates is given by the lever rule at that temperature T, (xβ(T) − X)/(xβ(T) − xα(T)). The same happens inside the hypoeutectic primary α (composition xD), and the fraction of β that precipitates in it is
The fractions of α and β inside the eutectic structure change with temperature in the same way, and when these are added together, the overall fractions of α and β agree with the lever rule at that temperature.
(c) Peritectic reaction. In a peritectic system, when a composition X between the peritectic point P and the end of the liquid (composition xL) is cooled, primary α crystallizes and grows, and then at the peritectic temperature TP the peritectic reaction L + α → β takes place. β forms where α and the liquid are in contact and grows so as to enclose α (this is the origin of the name “peritectic”). With xa the composition of the end of α and xP the composition of the peritectic point, the fractions before and after the reaction are
For the composition X (70.5 %) of textbook Fig. 2-8, there are α 18.6 % and liquid 81.4 % before the reaction, and β 41.3 % and liquid 58.7 % after it, so all the α turns into β, and that β grows from the remaining liquid and becomes a β polycrystal at T2. For compositions on the A side of P, the liquid is used up first, and a microstructure in which β encloses a core of α remains. The simulator draws the fractions in the middle of the reaction (progress ξ) by moving them between the values before and after the reaction in proportion to ξ. The mass balance is preserved regardless of ξ. The actual peritectic reaction proceeds by diffusion through the β shell and is therefore slow, so with fast cooling such as in casting, the α core tends to remain (a nonequilibrium microstructure in which the peritectic reaction does not go to completion).
(d) The horizontal part of the cooling curve. At the eutectic and peritectic temperatures three phases coexist and F = 0, so the temperature stays constant and stops falling until the reaction ends (thermal arrest). This is because the heat released by the reaction cancels the cooling, and the length of the horizontal part is proportional to the amount that reacts (for a eutectic, the fraction of liquid just above TE). This is why the eutectic and peritectic temperatures can be determined from cooling curves and differential thermal analysis (DTA). (In simulator ③, for ease of viewing, the time during which the temperature stops is made “a fixed time + a time proportional to the amount that reacts.”) For a pure substance or for the composition of a congruently melting compound too, the liquid solidifies all at once into the solid without changing composition, so the temperature stops in the same way.
(e) Lamellae or rods? The two phases of a eutectic structure can be either layers (lamellae) or rods. Let f be the volume fraction of the minority phase and compare at the same spacing λ: the interfacial area per unit volume is 2/λ for lamellae and, for rods on a square array (radius r, f = πr2/λ2), 2πr/λ2 = 2√(πf)/λ. If the interfacial energy is isotropic, the smaller interfacial area is favored, so the rule of thumb is rods for f < 1/π (about 0.32) and lamellae for larger f (a commonly used rule of thumb; comparing in the same way with rods on a triangular lattice gives about 0.28). In practice, anisotropy of the interfacial energy and the growth rate also matter, and quite a few systems depart from the rule of thumb (Ag–Cu is a typical lamellar system even though its minority-phase fraction is about 0.32, and the Si in Al–Si forms irregular plates). The simulator draws Pb–Sn, Ag–Cu, and Al–Si, whose shapes are known, as lamellae, and uses this rule of thumb to choose between lamellae and rods for the other systems (mole fractions are substituted for the fractions).
(f) How a eutectic structure grows as layers. At the eutectic temperature, α first nucleates, β forms next to it, and α and β grow side by side (the same idea as Fig. 7 of Tanaka 2008). Since α is rich in A and β is rich in B, when α grows, B is left over in the liquid ahead of it, and when β grows, A is left over. The leftover B diffuses along the interface to the front of the neighboring β and is used there, and the leftover A diffuses to the front of the neighboring α. This side-by-side growth of two phases that exchange components is called cooperative growth.
Since neighboring phases exchange components, the two phases form alternating layers (or rods) that are elongated in the direction of growth. The narrower the layer spacing λ, the shorter the diffusion distance and the faster the growth, but the α/β interface per unit volume increases and the interfacial energy cost goes up. The spacing is determined by this balance, and λ2v is nearly constant for a growth rate v (the faster the cooling, the finer the microstructure), as is known from the analysis of Jackson and Hunt (1966).
(g) Microstructure of a monotectic (two-liquid-separation) system. In a system with a large positive Ω of the liquid, inside the dome below the critical point (the region of two-liquid separation), the liquid splits into two liquids of different composition, L1 (A-rich) and L2 (B-rich), and the minority liquid forms droplets. At the monotectic temperature TM the monotectic reaction L1 → α + L2 proceeds at constant temperature, leaving droplets of L2 inside α (for compositions in which L2 is the majority, conversely, grains of α end up scattered in L2). As it cools, L2 becomes richer in B while crystallizing α, and finally solidifies at the eutectic temperature near the melting point of B. In Cu–Pb, the microstructure consists of Pb grains dispersed in Cu, which is used for plain bearings (Kelmet). In practice, it is also known that the droplets rise or sink because of the density difference between the two liquids, so the microstructure tends to be biased vertically (gravity segregation). For compositions in which primary α appears first, the combination of phases stays “α + liquid” at the monotectic temperature; the composition of the liquid, however, jumps from L1 to L2.
(h) Crystallization and precipitation, eutectoid and peritectoid. The emergence of a solid from a liquid is called crystallization, and the emergence of another solid from inside a solid is called precipitation (Tanaka 2008). When a reaction of the same form as the eutectic or peritectic takes place entirely among solids, it is called eutectoid (example: the eutectoid reaction of steel, γ → α + Fe3C, which gives the microstructure called pearlite) or peritectoid. If you read primary crystals as proeutectoid phases and the eutectic structure as the eutectoid structure, the way of reading the microstructures in this section (the lever rule included) applies.
Even when the equilibrium phases are determined, creating a new phase requires creating an interface. In the simplest classical nucleation theory, in which a spherical nucleus of radius r forms in a uniform parent phase, the free-energy change is
If the new phase is stable, the volumetric free-energy difference is Δgv < 0, but the interfacial energy γ > 0 tries to eliminate small nuclei. The critical radius and the barrier are
Increasing the supercooling increases |Δgv| and makes nucleation easier. Furthermore, the growth rate after a nucleus has formed depends on the diffusion of heat and solute, the interface reaction, crystal orientation, and elastic strain. This is why even materials with the same phase diagram can have greatly different microstructures if their heat-treatment histories differ. How nucleation and growth compete with the cooling rate is treated in detail in §7, as a story of the glass transition and crystallization.
At the end of §6 I noted that even materials with the same phase diagram can change their microstructure with heat-treatment history. The most extreme example is glass. When a melt of SiO2 or B2O3 is cooled, the crystal is stable below the melting point Tm on the phase diagram, yet the melt solidifies without becoming a crystal. In this section we trace through equations what separates “becoming a crystal” from “becoming a glass,” taking up the enthalpy diagram, the viscosity, the rates of nucleation and growth, and the TTT diagram in that order.
Plot the enthalpy H of 1 mol of the substance against temperature. Just as in Eq. (4), at constant pressure dH/dT = Cp, so the slope of a line is the heat capacity. Regarding the heat capacity as constant in each state and taking the enthalpy of the crystal at its melting point as the zero reference,
Here the subscript c denotes the crystal and l the liquid. The line for the liquid extends below Tm unchanged, and this is the supercooled liquid. The supercooled liquid has a higher G than the crystal and so is not stable, but until a crystal nucleus forms it keeps internal equilibrium as a liquid (metastable). On further cooling, at some temperature the rearrangement of the atomic configuration can no longer keep up, and the structure freezes as that of the liquid at that temperature. If the temperature that represents the frozen structure is called the fictive temperature Tf (Tool 1946), the enthalpy of the glass is
In a glass the atomic vibrations remain, but the rearrangement of the atoms (the configurational degrees of freedom) stops, so the heat capacity falls to about that of the crystal, and the slope of the line bends at Tf. The temperature of this bend is the glass transition temperature Tg. On the other hand, if a crystal is born during cooling, H drops by the enthalpy of fusion ΔHm and moves onto the line of the crystal.
Fig. 3. Enthalpy–temperature diagram drawn with the values of B2O3. Changing the cooling rate q moves the temperature Tg at which the curve departs from the line of the supercooled liquid, and changes the enthalpy of the glass on the room-temperature side. The line for the crystal is drawn to show the equilibrium path.
Melting was an equilibrium phase transition that occurs at a definite temperature where Gl and Gc become equal (§2.3). The glass transition is not like that. It is a question of rate: the structure freezes at the point where the time needed for the structure of the liquid to follow a change in temperature (the structural relaxation time) τ becomes longer than the time scale of the cooling. A rough estimate of τ follows from the Maxwell model of viscoelasticity. If a spring (high-frequency shear modulus G∞) and a dashpot (viscosity η) are connected in series, the stress
relaxes in this way. Here σ is the stress (a different quantity from the interfacial energy γ of Section 7.5). When the temperature is lowered at q [K/s], the time it takes to cool through the temperature width δT that the structure has to follow is δT/q. If τ is shorter than this, the liquid stays in equilibrium; if it is longer, the structure freezes. The condition at the boundary is
Using the order of magnitude of the shear modulus of silicate glass, G∞ ≈ 1010 Pa, η = 1012 Pa·s gives τ ≈ 100 s. When cooling at 10 K/min, it takes 60 s to pass through 10 K, so the structure freezes right around here. This is the meaning of the commonly used rule of thumb that “Tg is the temperature at which η ≈ 1012 Pa·s.” This, however, is a conventional definition, and Tg changes by several tens of K depending on the measurement method (DSC, dilatometer) and the cooling and heating rate.
What is important in Eq. (20) is that increasing q lowers the η at which the structure freezes, that is, raises Tg. A glass cooled faster has the structure of a liquid at a higher temperature, and both its H and its volume are larger (Moynihan et al. 1976 measured the cooling-rate dependence of the fictive temperature). Tg is not a material constant but a temperature determined by the cooling rate.
The viscosity of glass-forming liquids often rises more steeply than the Arrhenius type, and is often well described by the Vogel–Fulcher–Tammann (VFT) equation (Fulcher 1925, Tammann & Hesse 1926).
In Eq. (20), take η(Tg) = 1012 Pa·s at 10 K/min (qref = 1/6 K/s). Regarding G∞δT as a constant, η(Tg) ∝ 1/q, so log10η(Tg) = 12 − log10(q/qref). Substituting this into the left-hand side of Eq. (21) and solving for T gives
Inserting the VFT parameters of B2O3 (Table 1), we get Tg ≈ 537, 555, and 574 K for q = 10−3, 1/6, and 10 K/s, so Tg shifts by about 10 K for every order-of-magnitude change in the cooling rate (values in this model).
Another indicator is Angell’s fragility index m, which is the slope of log10η against the temperature normalized by Tg (Angell 1995). Setting s = Tg/T gives T = Tg/s and dT/ds = −Tg/s2, so
Liquids with a small m that are close to the Arrhenius type are called “strong liquids” (SiO2 is the typical example), and liquids that become viscous abruptly near Tg are called “fragile liquids.”
| Quantity | Li2O·2SiO2 (LS2) | SiO2 | B2O3 |
|---|---|---|---|
| Tm | 1307 Ka | 1996 K (β-cristobalite) | 723 Kb |
| ΔHm | 57.4 kJ/molc | 8.9 kJ/mold | 24.07 kJ/molb |
| VFT A, B, T0 | −2.40, 3082 K, 509.6 Ke | −2.94, 12997 K, 610 K (model)f | −0.97, 1580 K, 433 Kg |
| η(Tm) (Eq. 21) | 101.5 Pa·s | 106.4 Pa·s | 104.5 Pa·s |
| Tg(10 K/min) (Eq. 22) | 724 K | 1480 K | 555 K |
| Tg/Tm | 0.55 | 0.74 | 0.77 |
| m (Eq. 23) | 49 | 25 | 59 |
When a crystal forms from a supercooled liquid, the free energy per 1 mol changes by ΔG = Gc − Gl = −ΔGfus. Using Eq. (7) (the approximation ΔCp = 0) and rewriting in terms of the degree of supercooling ΔT = Tm − T, this is −ΔHm(1 − T/Tm) = −ΔHmΔT/Tm. Dividing by the molar volume Vm to get a value per unit volume, we obtain
The greater the supercooling, the larger the driving force. There are also approximations, such as Hoffman’s, that partly incorporate the effect of ΔCp at large supercooling.
The free-energy change when a spherical crystal of radius r forms in a supercooled liquid is, as in Eq. (15), ΔG(r) = 4πr2γ + (4/3)πr3Δgv (γ is the interfacial energy between the crystal and the liquid, and Δgv < 0). The first term goes as r2 and the second as r3, so for small nuclei the interfacial loss wins, and for large nuclei the volume gain wins. We find the top of the hill that marks the boundary between them.
Substituting this back into ΔG(r),
we obtain Eq. (16). Substituting Eq. (24) into it,
As Tm is approached, the barrier becomes infinite and no nuclei can form. The frequency with which critical nuclei form is proportional to the Boltzmann factor exp(−W*/kBT), and the frequency with which atoms attach to a critical nucleus is proportional to the diffusion coefficient. If the diffusion coefficient is related to the viscosity through an Eyring (or Stokes–Einstein) type relation D ≈ kBT/(d0η) (d0 is the size of a structural unit; numerical coefficients are omitted), the steady-state nucleation rate (the number of nuclei per unit volume and unit time) takes the following form (the form used by Cassar et al. 2020).
On the high-temperature side W* is too large, and on the low-temperature side 1/η drops steeply, so I(T) has a sharp peak at a temperature well below Tm (Fig. 4, left). Whether the peak lies above or below Tg depends on the material. In the model of this page, for LS2, which has a small Tg/Tm (it crystallizes easily), the peak lies a little above Tg (783 K; the Tg calculated in simulator ④ for cooling at 10 K/min is about 727 K), while for SiO2 and B2O3, which have a large Tg/Tm (they form glass easily), the peak lies below Tg (in the temperature range where the structure freezes) (1357 K versus about 1459 K for SiO2, and 531 K versus about 559 K for B2O3). If the peak lies below Tg, the motion of the atoms freezes during cooling before nuclei can form, so the material becomes a glass more easily. LS2 is a material often used in studies of homogeneous nucleation; for example, in the undoped sample of Thieme et al. 2016 the maximum of the nucleation rate is 16 mm−3s−1 (1.6×1010 m−3s−1) at about 469 °C. In the model of this page (the values of Table 1 with γ = 0.151 J/m2 fitted), the maximum is 1.0×1011 m−3s−1 at 783 K: the order of magnitude is close, but the temperature comes out higher. It is a much-discussed point that, when classical nucleation theory is combined with the viscosity, the maximum tends to shift to the high-temperature side (Nascimento et al. 2011, Cassar et al. 2020).
A crystal that has formed grows larger by advancing its interface. Consider the difference between the frequency at which structural units on the liquid side of the interface attach to the crystal side and the frequency at which they conversely leave the crystal. Let the distance of a single hop be d0 and the frequency be D/d02, and suppose that hops in the reverse direction are made less likely by the driving force ΔGm = ΔHmΔT/Tm (per mole). Then the growth rate is
where f is the fraction of sites at the interface where atoms can attach. When the supercooling is small, the exponent ε = ΔHmΔT/(RTTm) of Eq. (27) is much smaller than 1, so 1 − exp(−ε) ≈ ε and u ∝ ΔT/η, which is the same form as the result of Wagstaff 1969, who organized the growth of cristobalite in silica in the form u ∝ ΔT/η. The peak of u(T) lies on the higher-temperature side (just below Tm) of the peak of I(T). The maximum growth rate of LS2 is measured to be 6.87×10−5 m/s (at 1210 K; the table of Jiusti et al. 2020), and in the model of this page (fitted with f = 0.23) it is 6.8×10−5 m/s (at 1242 K).
Suppose that nuclei are born at random in the liquid and grow isotropically. The volume fraction of crystals counted while ignoring the collisions between crystals (“allowing them to overlap”) is called the extended volume fraction Xext. The volume fraction dX that actually increases is only the part of dXext that falls on the portion that is still liquid (fraction 1 − X), so
These relations hold (Avrami 1939, 1940, among others). If I and u are constant at constant temperature, a nucleus born at time t′ has radius u(t − t′) at time t, so
We thus obtain Eq. (29) (Avrami exponent 4). When cooling at a constant rate q, I and u change with time, so they are kept inside the integral, and with t = (Tm − T)/q and dt = −dT/q the time integral is converted into an integral over temperature.
The outer dt′ gives a factor 1/q, and the ds inside the brackets gives (1/q)3, so that the extended volume fraction is inversely proportional to the fourth power of the cooling rate (exact as long as I and u are functions of temperature alone). While X is small, X ≈ Xext, so cooling 10 times faster reduces the amount of crystal to about 1/104.
Uhlmann 1972 took “a crystal can be detected once it has formed at a volume fraction X = 10−6” as the criterion and considered the conditions for forming a glass. The time needed to reach X when the sample is held at temperature T is, from Eq. (29) solved for t,
When this is plotted on a temperature–time (logarithmic) diagram, it takes the shape of a “nose” jutting out sideways (Time–Temperature–Transformation, the TTT diagram). This is because I is small at high temperatures and η is large at low temperatures. The critical cooling rate Rc can be defined in two ways: by a simple estimate using the nose temperature Tn and time tn, and by the rate at which Eq. (30) gives X = 10−6.
Here C is the value of the integral in Eq. (30) carried out to a sufficiently low temperature (the temperature at which crystallization practically stops; Tlo in the simulator). Since Rc is a quantity determined by the detection limit X = 10−6, for q > Rc a glass forms, while for q < Rc a detectable amount of crystal begins to appear, and if q is sufficiently smaller than Rc, crystallization proceeds (since X ∝ q−4, reducing q to 1/10 increases X by a factor of about 104). The nose-based estimate counts as if the sample stayed for a long time only at the nose temperature during cooling, so it gives a larger value than the integral for continuous cooling (Weinberg et al. 1989 discuss the comparison of the two).
Fig. 4. Nucleation rate, growth rate, and the TTT diagram. Left: I(T) of Eq. (26) and u(T) of Eq. (27), each divided by its maximum and plotted on a logarithmic scale. Right: the TTT curve for X = 10−6 from Eq. (31), and the cooling curve for the critical cooling rate Rc (continuous cooling). The cooling curve T = Tm − qt looks curved on a logarithmic time axis. The TTT curve gives the time “when the sample is held at that temperature,” so even where the cooling curve crosses the TTT curve, in continuous cooling, which passes through each temperature in a short time, the crystal does not reach X = 10−6 (this is the difference between the two Rc values in Eq. (32)).
| Material | Nose Tn, tn | Rc (nose) | Rc (continuous cooling) | Literature benchmark |
|---|---|---|---|---|
| LS2 | 892 K, 1.8×102 s | 2.3 K/s | 0.19 K/s | Measured 14 ± 2 °C/min (0.23 K/s) without a container and 62 ± 3 °C/min (1.0 K/s) in a container (Ranasinghe et al. 2007; confirmed in the abstract) |
| SiO2 | 1528 K, 5.8×105 s | 8.0×10−4 K/s | 8.6×10−5 K/s | About 1.1×10−4 K/s from a viscosity-based estimate (Sakaguchi 1995; confirmed in the abstract) |
| B2O3 | 571 K, 1.2×108 s | 1.3×10−6 K/s | 7.8×10−8 K/s | Crystallization is practically never observed at ordinary pressure without seed crystals (Zanotto & Cassar 2017) |
Putting Table 1 and Table 2 side by side, one can read empirical rules for how easily glass forms. (i) The larger Tg/Tm is, the more easily the material forms a glass. There is the observation that Tg/Tm ≈ 2/3 is often found for glass-forming substances (Kauzmann 1948, Sakka & Mackenzie 1971), and Turnbull 1969 argued that homogeneous nucleation is suppressed more strongly the larger Tg/Tm is (roughly 2/3 or more). This is because the temperature range from Tm to Tg in which crystallization can proceed is narrow. (ii) The higher the viscosity at the melting point, the harder it is for atoms to move, and both u and I become small (106.4 for SiO2, 104.5 for B2O3, and 101.5 Pa·s for LS2).
Zachariasen 1932 gave the following empirical rules as the conditions under which an oxide AmOn can form a continuous random network without long-range order.
In SiO2, SiO4 tetrahedra are connected by sharing corner oxygens (bridging oxygens), and in B2O3, BO3 triangles are connected by sharing corners. Corner-sharing polyhedra can extend the network without large strain even without periodicity, by changing bond angles such as Si–O–Si. Oxides that form networks in this way (SiO2, B2O3, GeO2, P2O5, and so on) are called network-forming oxides. On the other hand, Li2O and Na2O are modifier oxides that break Si–O–Si bonds and create non-bridging oxygens, lowering the viscosity and making crystallization easier. This is one reason why LS2 is an “example that crystallizes easily.” It also matters that LS2 is a congruently melting compound, so that a crystal of the same composition as the glass can form and homogeneous nucleation can occur without a change in composition, and that Tg/Tm is small, about 0.55 (Table 1). Note that Zachariasen’s rules are empirical rules for network-forming oxides alone, and for glasses containing modifier oxides an extension is needed in which the modifier ions enter the gaps in the network (Zachariasen 1932 himself discussed it in this form). They do not apply to metallic glasses.
In Fig. 4, when LS2 is selected, one can see that the peak of I (just above Tg) and the peak of u (just below Tm) are separated in temperature. So if one first makes a glass and then carries out a two-stage heat treatment in which the glass is held at the temperature of the nucleation peak to create many nuclei, and then heated toward the growth peak to grow the crystals, one obtains a material in which fine crystals are uniformly dispersed throughout the glass. This is the basic idea of crystallized glass (glass-ceramics), and the Li2O–SiO2 system is one of its representative systems (for an overview, see the book by Höland & Beall). That adding a nucleating agent changes Rc has also been investigated for LS2 (Huang et al. 1986).
In simulator ③, pressing “Quench” during cooling freezes the remaining liquid as glass. That was based on the assumption that there is “no time to become a crystal.” In the language of this section, quenching is an operation that makes the cooling rate q sufficiently larger than the critical cooling rate Rc. Rc differs greatly between materials; even the three materials of Table 2 alone differ by more than 6 orders of magnitude (the value for B2O3 is from the model). Some materials, like SiO2, turn into glass even when cooled slowly in a furnace, while others, like metals, need an extremely large cooling rate. The phase diagram (equilibrium) tells which phase is stable, and Rc (kinetics) tells whether the system can get there.
Simulator ④ solves the equations of this section in order along a single cooling (or heating) path, and draws the results simultaneously in three panels: a schematic picture of the atomic arrangement, an enthalpy–temperature diagram, and a TTT diagram. Here we describe which equations and values the numbers on the screen come from, how to read them, and how far they are schematic. The numbers below were checked by evaluating the same functions as the simulator in node.
(1) The model — the equations used
At high temperatures where τ is short, Tf = T (liquid); when τ becomes longer than the cooling time, Tf stops moving (glass). The Tf that freezes when cooling from high temperature at a constant rate q is called Tg(q) in the simulator. Within each temperature step τ is regarded as constant (evaluated at the midpoint of the interval), and the equation is advanced with its exact solution Tf ← T1 − sτ + (Tf − T0 + sτ)exp(−Δt/τ) (s = ±q is the rate of change of temperature).
(2) Material presets
| Quantity | Li2O·2SiO2 (initial display) | SiO2 | B2O3 |
|---|---|---|---|
| Tm, ΔHm, VFT | Same as Table 1 | ||
| M, ρ (crystal) | 150.05 g/mol, about 2.46 g/cm3 | 60.08, about 2.20 | 69.62, about 2.46 |
| σ (= γ) | 0.1514 J/m2 (model) | 0.065 (model) | 0.100 (model) |
| g (= f) | 0.23 (model) | 1 | 1 |
| Cp,c, Cp,l | 224, 280 J/(mol·K) (schematic values) | 75.9, 85.8 (JANAF) | 106.7, 129.7 (JANAF) |
| Temperature range Tlo–Thi | 580–1400 K | 1150–2150 K | 470–800 K |
| q slider (K/s) | 10−5–106 | 10−8–103 | 10−11–104 |
| q for “Cool slowly” | 10−3 K/s | 10−6 K/s | 10−9 K/s |
| Rc (continuous cooling, X = 10−6) | 0.18 K/s | 8.6×10−5 K/s | 7.8×10−8 K/s |
| Tg(10 K/min): simulator | 727 K | 1459 K | 559 K |
| Tg(10 K/min): Eq. (22) (Table 1) | 724 K | 1480 K | 555 K |
σ is a value corresponding to Turnbull’s dimensionless interfacial energy α = 0.345 for LS2, and for SiO2 a value chosen to match the order of magnitude of Rc (the literature benchmark of Table 2). For B2O3 there is no literature Rc to match, so it is a provisional value fitted to nothing (α ≈ 0.326, smaller than that of LS2; B2O3 is hard to crystallize in the model because the viscosity is high and Tm − Tg is narrow. There is no measurement to compare with the model’s Rc = 7.8×10−8 K/s. See the caution in Section 7.8). For the heat capacities, SiO2 uses the values at 1696 K in the NIST-JANAF table (O-039), 75.94 for quartz (high-temperature form) and 85.77 J/(mol·K) for the liquid, and B2O3 uses the values at 723 K from the same table (B-097). The difference for SiO2, ΔCp ≈ 10 J/(mol·K), is the difference between the crystal and the liquid, not a measured difference between glass and liquid. The Cp,c of LS2 is the Dulong–Petit value for 9 atoms (9 × 3R ≈ 224), and Cp,l is a schematic value set at 1.25 times that; the measured values have not been confirmed. The shape of the H plot (the size of the kink at Tg) depends on ΔCp, but Tg and X are not affected.
(3) Numerical method
(4) Controls
(5) How to read the display
(6) Examples of checked values
In the phase-field method, the composition c(r,t) and an order parameter representing the phase, η(r,t), are placed continuously in space. For example, take η = 0 as the liquid phase and η = 1 as the solid phase, and let it change smoothly from 0 to 1 at the interface. We take the total free-energy functional to be
Here fchem is the bulk free energy that gave the phase diagram, and κc|∇c|²/2 and κη|∇η|²/2 are interface terms that penalize rapid spatial variation. As a result the interface acquires a finite thickness, and the interfacial energy appears naturally.
Composition is a conserved quantity that cannot disappear on its own, so with the chemical potential
it flows down its gradient, with flux J = −M∇μ. This is the Cahn–Hilliard equation.
On the other hand, the order parameter η, which represents order or disorder of a crystal or liquid versus solid phase, is a nonconserved quantity that can change locally from 0 to 1. The steepest-descent time evolution is given by
For periodic boundaries or zero-flux boundaries, with M > 0 and L > 0, dℱ/dt ≤ 0 holds, and the system changes its microstructure while dissipating free energy. This is the heart of building a “time evolution that does not violate thermodynamics.”
For the solidification of a binary alloy, one can take, for example,
Here the solid- and liquid-phase free energies of the phase diagram are connected by an interpolation function h(η), and the double well g(η) = η²(1−η)² stabilizes the pure liquid and solid phases. By coupling the temperature field, latent heat, interface anisotropy, and elastic field, the method can be extended to dendrites, eutectic growth, precipitation, grain growth, and so on.
Fig. 5. A two-dimensional Cahn–Hilliard mini-experiment. It uses the dimensionless free energy f = (c²−1)²/4 and periodic boundaries. Watch how concentration variations arise while the average composition is conserved, and then coarsen. At which wavelengths the initial variations grow is found in §9.5 by linearizing Eq. (35).
In §4.2 we saw that inside a two-phase region there are two boundaries, the binodal and the spinodal. In this section we derive spinodal decomposition, which occurs inside the spinodal, following the flow of §8.2 of Kingery, Bowen, and Uhlmann, Introduction to Ceramics, 2nd edition. It is also an account of linearizing the phase-field method of §8 (the Cahn–Hilliard equation) around the uniform state. Next to each equation number we add the corresponding equation number of Kingery.
For simplicity we use the same symmetric regular solution as in §4.2. Let g be the free energy per molecule (per structural unit), and use the critical temperature Tc = Ω/(2R) to write Ω = 2RTc; then Eq. (14) becomes
Here the per-mole Eq. (14) has been divided by NA, using R/NA = kB. Two curves follow from this.
The binodal equation is solved numerically. For example, at T/Tc = 0.9 the binodal is at 0.237 and the spinodal at 0.342, at 0.8 they are at 0.145 and 0.276, and at 0.7 they are at 0.086 and 0.226 (subtracting from 1 gives the values on the opposite side). The two curves touch at the critical point (x = 1/2, T = Tc).
Fig. 6. Temperature variation of the g–x curve of a regular solution, and the binodal and spinodal. Left: g(x) at five temperatures (the vertical scale is changed from curve to curve for visibility). Circles are the points of tangency of the common tangent (binodal), and triangles are the inflection points (spinodal). Right: the phase diagram obtained by stacking them by temperature.
Looking at the temperatures from the top down, (a) for T > Tc the curve is convex downward everywhere (one valley), and a uniform solution is stable at every composition. (b) At T = Tc the bottom of the valley is flat. (c) For T < Tc a hump appears in the middle, and the region between the two points of tangency of the common tangent is the two-phase (immiscible) region. Between the inflection points inside the hump, ∂2g/∂x2 < 0. This is what is drawn in Fig. 8.2 of Kingery (the source is given as Seward 1970).
Eq. (38) is the value for a uniform solution. When the composition varies from place to place, at places where the composition changes sharply the energy of a “nascent interface” is added. Following Cahn & Hilliard 1958, we express it by the square of the composition gradient, and the free energy of a system of volume V is
where C is the composition (the same as the mole fraction x) and Nv is the number of molecules per unit volume, and κ > 0 is the gradient-energy coefficient. This is the same idea as κc|∇c|2/2 of Eq. (33) in §8, and only the way the coefficient is set differs (it corresponds to κc = 2Nvκ).
Suppose that a small composition fluctuation C − C0 has formed in a uniform solution of average composition C0. Expanding g(C) in a Taylor series around C0, we get
Put this into Eq. (40) and subtract G = NvVg(C0) of the uniform solution. The first-order term vanishes, since the amount of substance is conserved and hence ∫(C − C0)dV = 0. (Kingery explains that in an isotropic solution the odd-order terms vanish. For the first-order term, the reason is conservation of the amount of substance.) If the terms of third and higher order are dropped as small, we obtain
The second term is always positive. Therefore, if (∂2g/∂C2)C0 > 0 (outside the spinodal), then ΔG > 0 for any small fluctuation, and the uniform solution is stable against fluctuations. Inside the spinodal the first term becomes negative, and ΔG < 0 for fluctuations that satisfy
that is, fluctuations whose gradient is sufficiently gentle (long wavelength). A solution inside the spinodal is always unstable against some fluctuation.
Take the fluctuation to be a single sine wave and examine which wavelengths are unstable.
β is the wavenumber, and the wavelength is λ = 2π/β. ∇C = −Aβ sin βx, and the average over one wavelength is ⟨cos2βx⟩ = ⟨sin2βx⟩ = 1/2. Putting this into Eq. (41) and dividing by the volume, we get
Hereafter we abbreviate g″ = (∂2g/∂C2)C0. Kingery’s Eq. 8.12 is written in the per-molecule form without the coefficient Nv, but the argument about the sign is the same. For g″ < 0, the bracket is negative for waves with β2 < −g″/(2κ).
All fluctuations with wavelengths longer than λc are unstable, and at shorter wavelengths the gradient energy wins and ΔG > 0. The more negative g″ is (the deeper inside the spinodal), the shorter λc is, and the finer the scale on which decomposition can occur. If the system separates into two phases on a scale finer than λc, it is by nucleation and growth.
From Eq. (40), the chemical potential (per molecule, the driving force of diffusion) obtained by variation is μ = g′(C) − 2κ∇2C (the same operation as Eq. (34) in §8). Taking the flux to be J = −M̃∇μ (M̃ > 0 is the mobility) and conservation of matter to be Nv∂C/∂t = −∇·J, we obtain
Put C = C0 + δC and expand to first order as g′(C) ≈ g′(C0) + g″δC. g′(C0) is a constant and vanishes under ∇2, so we obtain the linear equation
Putting in δC = A(t) cos βx, and using ∇2 cos βx = −β2 cos βx and ∇4 cos βx = β4 cos βx, we get dA/dt = R(β)A, which gives
We call R(β) the amplification factor. Since the content of the bracket is the same as in Eq. (44), a wave that lowers the free energy (ΔG < 0) grows exponentially with R > 0, and a wave that raises it decays with R < 0. R(β) is 0 at β = 0 (it takes time to carry matter far), and becomes 0 again at β = βc, and has a maximum in between. The position of the maximum is found from
so
The maximum value is obtained by putting 2κβm2 = −g″/2 back into Eq. (48):
Defining the interdiffusion coefficient D̃ = M̃g″/Nv, Eq. (47) for κ = 0 is the ordinary diffusion equation ∂δC/∂t = D̃∇2δC. Inside the spinodal g″ < 0, so D̃ < 0, and atoms gather even more toward the denser side (uphill diffusion). Eq. (50) can also be written as
For a regular solution with C0 = 1/2, Eq. (38) gives g″ = −4kB(Tc − T), so λm ∝ (Tc − T)−1/2. The deeper you cool, the finer the microstructure; conversely, as the spinodal is approached, λm becomes longer and growth also becomes slower.
Fig. 7. Amplification factor R(β). Eq. (48) is drawn in nondimensional form (unit of length ℓ = (κ/kBTc)1/2, unit of time τ0 = Nvℓ2/(M̃kBTc)). Vary the temperature and the average composition, and watch the range with R > 0 appear and disappear.
| Nucleation and growth (between the binodal and the spinodal) | Spinodal decomposition (inside the spinodal) | |
|---|---|---|
| How it starts | Small nuclei with a large composition difference are born stochastically, here and there | Waves with a small composition difference grow everywhere at the same time |
| Barrier | Yes (ΔG* in Eq. 16) | None (waves with R > 0 grow from the start) |
| Composition | The nucleus has a near-equilibrium composition from the start | The amplitude grows continuously and eventually approaches the equilibrium composition |
| Scale | Set by the size and number density of the nuclei, and scattered | Initially, waves near the wavelength λm grow together |
| Diffusion | Ordinary diffusion down the concentration gradient (D̃ > 0) | Uphill diffusion (D̃ < 0) |
Experimental clues are the initial stage in which, in small-angle X-ray scattering or light scattering, the position (wavelength) of the scattering peak stays almost constant while the intensity grows, and the difference in behavior inside and outside the spinodal. For example, it has been reported that in TiO2–SnO2, a modulated structure forms inside the coherent spinodal for [001], but no modulation was seen between the coherent spinodal for [001] and the equilibrium curve (Stubican & Schultz 1970). Studies of phase separation in glass by small-angle X-ray scattering go back to Porai-Koshits & Andreyev 1958. Linear theory (Eq. 48) holds only in the initial stage, and in practice the scattering peak often shifts toward lower angles (longer wavelengths) with time, so the effects of thermal fluctuations (Cook 1970) and nonlinear effects have been discussed (Cahn & Hilliard 1971).
After enough time has passed, the composition field becomes a superposition of waves of various directions with wavelengths near λm. Cahn 1965 superposed 100 sinusoidal waves of wavenumber magnitude βm with random directions and phases, and computed a 50:50 microstructure obtained by dividing the field into two phases according to the sign of its value. Looking at successive cross sections, he showed that this is a microstructure in which both phases are interconnected in three dimensions, and that it resembles the microstructures observed in phase-separated glass (Fig. 8.4 of Kingery).
Fig. 8 below is our own two-dimensional calculation of the same idea. If you move the threshold to reduce the volume fraction of the minority phase, the minority phase turns into isolated grains (droplet-like). In BaO–SiO2 glass, Ba-rich spherical grains at low BaO contents, an interconnected two-phase microstructure near the middle of the immiscibility region, and silica-rich spherical grains at high BaO contents have in fact been observed (Seward et al. 1968a).
Fig. 8. Two-phase microstructure made by superposing random composition waves (two dimensions). 100 sinusoidal waves with the same wavenumber magnitude βm are superposed with random directions and phases, and the specified fraction, taken from the large-value side, is shown as the B-rich phase (orange). The number at the upper right is the number of connected clusters of the minority phase.
Once the amplitude has grown to the equilibrium composition difference, linear theory no longer holds. After that, the microstructure becomes coarser so as to reduce the interfacial area (coarsening). In the phase separation of alkali borosilicate glass, the interfacial area decreases in proportion to the −1/2 power of time, and an apparent activation energy of 40 kcal/mol (about 167 kJ/mol) has been reported (Haller 1965).
When the temperature is raised above Tc, Eq. (38) becomes positive at every composition, and R(β) of Eq. (48) becomes negative for every β. All the composition waves decay, and the system returns to a uniform solution. Even below Tc, if the system moves outside the binodal, the two phases eventually dissolve into each other. Since the material has to be carried by diffusion, however, the coarser the microstructure, the longer it takes (the time is of the order of λ2/D̃). Haller et al. 1970 used the temperature at which the milky opalescence of glass (light scattering by phase separation) disappears on heating to determine the range of metastable immiscibility in the Na2O–B2O3–SiO2 system. In glass, on the other hand, diffusion effectively stops below Tg, so a phase-separated microstructure can be carried out to room temperature as it is (§7).
| System | What is known | Source |
|---|---|---|
| Na2O–B2O3–SiO2 glass (Vycor process) | When alkali borosilicate glass is heat-treated, it separates into a silica-rich phase and a sodium-borate-rich phase. The patent lists heat treatments such as 3 days at 525 °C and several hours at 600 °C. If the sodium-borate-rich phase is dissolved away with acid to make the glass porous and then sintered at 900–1000 °C, a glass of about 96 % silica results. The interconnected microstructure of the phase separation is widely believed to result from spinodal decomposition (depending on the composition and heat treatment, a contribution from nucleation is also possible). Because this is a three-component system, a two-component diagram is only an approximation | Hood & Nordberg 1938 (patent), Nordberg 1944, Haller 1965, Haller et al. 1970 |
| BaO–SiO2 glass | Metastable two-liquid immiscibility below the liquidus. The critical temperature is about 1460 °C and the critical composition about 10 mol% BaO. Depending on the composition, the microstructure changes as “Ba-rich spherical particles → interconnected structure → silica-rich spherical particles” | Seward et al. 1968a, b |
| TiO2–SnO2 (rutile-type solid solution) | Immiscibility in the solid phase. Critical temperature 1430 °C, critical composition 47 mol% TiO2. A modulated structure in which TiO2-rich plates and SnO2-rich plates alternate perpendicular to [001]. The smallest wavelength in the early stage of decomposition is reported to be about 11 nm at 925 °C, 17 nm at 1000 °C, and 24 nm at 1150 °C | Stubican & Schultz 1970, Park et al. 1975 |
| (Ti,Al)N hard coatings | Metastable cubic Ti1−xAlxN first decomposes on heating into coherent cubic nanoscale domains, and the hardness increases (age hardening). At higher temperatures it becomes stable TiN and hexagonal AlN | Hörling et al. 2002, Mayrhofer et al. 2003, Mayrhofer et al. 2005 |
| InxGa1−xN | Solid-phase immiscibility with an interaction parameter of 5.98 kcal/mol (25.0 kJ/mol) was predicted by calculation. Phase separation was observed in films grown at 700–750 °C when In > 35 %, in agreement with the prediction of spinodal decomposition | Ho & Stringfellow 1996, Doppalapudi et al. 1998 |
| CoFe2O4–Co3O4 (spinel) | There is an immiscibility region below 860 K. When aged inside the spinodal (500 °C), the composition fluctuations increase the coercivity at room temperature (7 kOe and 4.5 kOe for fully aged samples). The coercivity of a sample that was only quenched is small at room temperature and very large only at low temperatures | Takahashi & Fine 1972 |
The values in the table were checked against the abstracts of the original papers (for the heat-treatment conditions of the Vycor process, the text of the patent). If the InxGa1−xN value is regarded as a symmetric regular solution, Tc = Ω/(2R) ≈ 1.5×103 K, but this is an estimate made on this page, not a value given in the paper. The temperature range of the immiscibility region of the Na2O–B2O3–SiO2 system is not listed because the figure in the original source could not be checked. Simulator ⑤ tentatively uses a critical temperature of 760 °C, a rough guide from secondary sources (Section 9.11).
Simulator ⑤ numerically solves the Cahn–Hilliard equation, Eq. (46), in two dimensions, and draws how the composition field grows in a material cooled from a high temperature (or heat-treated) in four figures at the same time: the atom arrangement, the position on the phase diagram, the free-energy curve, and the amplification factor R(β). The numbers below were checked by computing the same functions as those in the simulator, using node.
(1) The model — free energy
Borosilicate glass and TiO2–SnO2 are the same symmetric regular solutions as in Section 9.1. Because the immiscibility region of BaO–SiO2 is skewed toward about 10 mol% BaO, it is represented as a mixture of two kinds of structural units of different sizes (the same form as the Flory–Huggins type for polymer solutions). Assuming that a SiO2 unit occupies one lattice site and a BaO-containing unit occupies N sites, and writing the volume fraction of BaO-containing units as φ = Nx/[1 + (N − 1)x] (x is the mole fraction), the free energy per lattice site, in units of RTc, is given by
Here χc is chosen so that ∂2g/∂φ2 = ∂3g/∂φ3 = 0 at the critical point (φc, Tc). For N = 1, φ = x and χc = 2, which reduces to the regular solution of Eq. (38) (Ω = 2RTc). For BaO–SiO2, the critical composition xc = 1/(N3/2 + 1) = 0.10 gives N = 92/3 ≈ 4.33. N is a phenomenological parameter for fitting the shape of the immiscibility region, and does not claim to be the size of an actual structural unit. The value per mole is Gm(x) = [1 + (N − 1)x]g(φ), and its curvature is ∂2Gm/∂x2 = N2(∂2g/∂φ2)/[1 + (N − 1)x]3; the free-energy curve on the screen and the readout ∂2g/∂C2 are these per-mole values (kJ/mol). The binodal is found by bisection on the slope of the common tangent, and the spinodal as the two roots of ∂2g/∂φ2 = 0.
(2) The model — time evolution
Taking ℓ = (κ/RTc)1/2 as the unit of length and τ0 = ℓ2/[M(Tc)RTc] as the unit of time (the same nondimensionalization as in Fig. 7, written per mole), Eq. (46) becomes
Here g is in units of RTc, t in units of τ0, and ∇ in units of ℓ. The mobility is of Arrhenius type and decreases as the temperature falls, so the lower the temperature, the more slowly the microstructure changes. ξ is the thermal fluctuation (Cook 1970); to conserve the amount of substance it is added with a variance proportional to the square of the wavenumber, and its strength is 2.5×10−3 × T/Tc (a schematic value). Linearizing this about a uniform state gives, in the same form as Eq. (48), R(β) = −mβ2[g″ + e(θ) + 2β2] (g″ = ∂2g/∂φ2), λc = 2π[−(g″ + e)/2]−1/2, λm = √2 λc, and Rm = m(g″ + e)2/8, and the readouts and the figure of R(β) are calculated with these formulas.
e(θ) is the coherent-strain term, included only for the crystal (TiO2–SnO2) (the box in Section 9.8, Cahn 1962). To match the calculation of Park et al. 1975, in which the coherent spinodal at x = 0.5 is lowered by 105 K along [001] and by 387 K along <100>, we set e1 = 4 × 105 K/Tc ≈ 0.247 ([001], horizontal in the figure) and e2 = 4 × 387 K/Tc ≈ 0.909 ([100], vertical in the figure), and interpolate for directions in between by e(θ) = e2 + (e1 − e2)cos2θ (θ is the angle between the wavevector and [001]). With this interpolation the lowering along <101> comes out as 302 K. For a regular solution, g″ at x = 0.5 is 4(T/Tc − 1), so e = 4ΔT/Tc is the amount that lowers the spinodal by ΔT. In the phase diagram, the coherent spinodal for the softest direction, [001], is drawn as a dotted line, and the region inside it is labeled “Unstable.”
(3) Material presets
| Quantity | Borosilicate glass (initial display) | BaO–SiO2 glass | TiO2–SnO2 solid solution |
|---|---|---|---|
| Model | Regular solution (N = 1). Na2O–B2O3–SiO2 is treated as a pseudo-binary system | Eq. (52), N = 92/3 ≈ 4.33 | Regular solution + coherent strain |
| Critical point | 760 °C (guide from secondary sources), c = 50 % | 1460 °C, 10 mol% BaO | 1430 °C, 50 % (actually 47 mol% TiO2) |
| Ea of the mobility | 167 kJ/mol | 200 kJ/mol (schematic value) | 200 kJ/mol (schematic value) |
| Temperature slider | 450–850 °C | 950–1600 °C | 850–1550 °C |
| Initial holding temperature | 600 °C | 1150 °C | 1000 °C |
| Composition slider (initial value) | Normalized composition c 5–95 % (50 %) | 1–40 mol% BaO (10 %) | 5–95 mol% SnO2 (50 %) |
| Length conversion ℓ | Assumed to be 1 nm | Assumed to be 1 nm | About 1.2 nm (matched to the measured wavelength) |
Ea = 167 kJ/mol for borosilicate glass is the apparent activation energy of coarsening from Haller 1965 (Section 9.8), borrowed for the mobility. The critical temperature of 760 °C is a rough guide from secondary sources; since it could not be checked in the original source (the figure of Haller et al. 1970), it is not listed in the table of Section 9.9. The composition is the normalized composition c between the “silica-rich component” and the “Na–B-rich component,” not the composition of actual oxides. The critical point of BaO–SiO2 and the dependence of the microstructure on composition are based on the values and observations of Seward et al. 1968a, but the high-BaO side of the curve of Eq. (52) has not been checked against measurements. ℓ ≈ 1.2 nm for TiO2–SnO2 is a value chosen so that λm at 1000 °C equals the measured modulation wavelength of 17 nm (Stubican & Schultz 1970).
(4) Numerical method
(5) Controls
(6) How to read the display
(7) Readouts
(8) Examples of checked values
| Quantity | Borosilicate glass, 600 °C, 50 % | BaO–SiO2, 1150 °C, 10 mol% | TiO2–SnO2, 1000 °C, 50 % |
|---|---|---|---|
| T/Tc | 0.845 | 0.821 | 0.748 |
| Binodal | 18.1–81.9 % | 1.5–34.9 mol% | 11.1–88.9 mol% |
| Spinodal | 30.3–69.7 % | 4.1–22.5 mol% | 24.9–75.1 mol% (coherent [001]: 27.5–72.5) |
| ∂2g/∂C2 | −5.32 kJ/mol | −44.7 kJ/mol | −14.3 kJ/mol (−10.8 with strain) |
| λc, λm | 11.3 ℓ, 16.0 ℓ | 14.2 ℓ, 20.1 ℓ | 10.2 ℓ, 14.4 ℓ (17.3 nm) |
| Rm | 1.4×10−3/τ0 | 9.4×10−4/τ0 | 6.2×10−4/τ0 |
| M(T)/M(Tc) | 0.028 | 0.049 | 0.0085 |
The borosilicate glass values lie between the regular-solution values of Section 9.1 (0.237/0.342 at T/Tc = 0.9 and 0.145/0.276 at 0.8) and come from the same formulas. The λm of TiO2–SnO2 is 15.6, 17.3, and 23.5 nm at 925, 1000, and 1150 °C, which agrees with the measured 11, 17, and 24 nm in order of magnitude, but the variation with temperature does not agree completely (the measurement is the “smallest wavelength observed,” which is not necessarily the initial λm itself). Also, 1360 °C is inside the chemical spinodal but outside the coherent spinodal for [001], so the simulator displays “Metastable” and the composition stays uniform. In checks made during development, we confirmed that the 50:50 composition gives a maze-like microstructure, that shifting the composition gives a droplet-like microstructure, that reheating above Tc returns the system to a uniform state, and that in BaO–SiO2 changing the composition 5 → 12 → 20 mol% gives Ba-rich drops → maze-like → silica-rich drops (the trend of Seward et al. 1968a).
| Question | The current phase-diagram simulator | Quantitative phase-field calculation |
|---|---|---|
| Which phase is stable | Found from the lower convex hull | Uses the same bulk free energy as the local driving force |
| Composition and fraction of each phase | Common tangent and lever rule | Found by integrating the local composition over space |
| Segregation during cooling | The two limits of equilibrium and Scheil–Gulliver | Solves directly with finite diffusion coefficients and interface velocities |
| Grain shape and size | A schematic (§5.2) showing grains grown from a small number of nuclei (Johnson–Mehl type) and the layers and precipitates of the eutectic structure, painted in areas that follow the lever-rule fractions | Developed from interfacial energy, anisotropy, nucleation, and elastic fields |
| Prediction of times and lengths | Not treated | Treated by calibrating the mobility, gradient coefficient, and spatial step to physical units |
| Misconception | The correct reading |
|---|---|
| If the free energy is negative, the state is stable | The zero point is arbitrary. When compared at the same T, p, and overall composition, the lower state is the stable one. |
| The point of tangency of the common tangent is the phase fraction | The points of tangency are the compositions of the phases. The phase fractions are obtained from the lever rule using the overall composition. |
| Everything inside the two-phase region is unstable | Between the binodal and the spinodal the state is metastable, and there is a nucleation barrier. |
| The transformation happens the instant you cross a line of the phase diagram | A phase diagram describes equilibrium. In practice there are supercooling, nucleation, and diffusion delays, and the start of the transformation depends on the history. |
| mol% and mass% can simply be read interchangeably | The numbers change depending on the atomic weights. The composition units of the mass balance and the thermodynamic model must be matched. |
| The glass transition temperature is a material constant like the melting point | Tg is the temperature at which structural relaxation freezes, and it changes with the cooling rate and the measurement method (§7.2). It is not an equilibrium transition temperature like the melting point. |
| Glass is a substance that does not crystallize however slowly it is cooled | Whether it crystallizes is a matter of the cooling rate q compared with the critical cooling rate Rc. Rc differs greatly from substance to substance; even the three substances in Table 2 differ by more than six orders of magnitude (§7.8). |
| An interconnected two-phase microstructure means spinodal decomposition | An interconnected structure also forms when grains formed by nucleation coalesce. Judge from the process, such as whether the initial wavelength is uniform and the amplitude grows (§9.7). |
| If a phase-field image resembles an experiment, it is correct | Verification is needed of mass conservation, energy dissipation, interface-width dependence, physical parameters, and the time and length scales. |
Corresponding simulator: Binary phase diagrams.
Related resources: Gibbs free energy and phases,
Entropy and free energy.
Main sources and data sources
§7 Glass transition and crystallization (numbers described in the text as “confirmed in the abstract” were checked in the abstracts of the original papers; the full texts have not been read)
§9 Spinodal decomposition (for Cahn 1961 and Cahn & Hilliard 1958, see the list above)