How Does a Binary Phase Diagram Arise from Free Energy?

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.

0. The Big Picture — Phase Diagrams and Microstructures Start from the Same Free Energy

Pure substancegAφ(T)
Reference for each phase
MixtureGφ(x,T)
Mixing and interaction
EquilibriumLower convex hull, common tangent
Chemical potential
Phase diagramStack the equilibria
of each temperature vertically
Microstructure formationInterfaces, diffusion, time
Phase field
A phase diagram is the “destination”; a microstructure calculation is “the road to get there” What a phase diagram tells you is the equilibrium phases, equilibrium compositions, and phase fractions after a sufficiently long time. The size, shape, and arrangement of grains, and the time a change takes, are not determined by the phase diagram alone. For these you need kinetics: interfacial energy, diffusion coefficients, nucleation, elastic strain, anisotropy, and so on.

1. Why Look for the Minimum of G?

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

G=fα Gα(xα,T) +fβ Gβ(xβ,T) (1)

Here, the phase fractions and the mass balance of component B must always be satisfied:

fα+fβ=1 , x0= fαxα +fβxβ (2)

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.

2. What Determines the Temperature Dependence of the Free Energy?

2.1 The slope is the entropy, and the curvature is the heat capacity

For 1 mol of a single phase, the total differential of the Gibbs free energy is

dG=Vdp −SdT +∑iμidni (3)

If the pressure and the amount of substance are fixed, it follows immediately that

(∂G∂T)p =−S , (∂2G∂T2)p =−CpT (4)

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.

2.2 Integrating the measured Cp gives G(T)

If H and S at a reference temperature T0 are known, then

H(T)=H(T0) +∫T0TCpdT S(T)=S(T0) +∫T0TCpTdT G(T)=H(T)−TS(T) (5)

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.

2.3 Melting as an example

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

ΔGfus(T)= ΔHm(1−TTm) +ΔCp [(T−Tm) −TlnTTm] (6)

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.

ΔGfus(T)= ΔHm(1−TTm) (7) Pure-substance model of the simulator

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).

The linear expression is an “entry point near the melting point,” not a universal formula In reality Cp differs between solid and liquid, and magnetic transitions, structural phase transitions, and electronic excitations also contribute to G(T). In CALPHAD, which treats wide temperature ranges quantitatively, the functions of each phase are evaluated so as to reproduce measured heat capacities, enthalpies of formation, activities, and phase equilibria simultaneously.

3. From a Pure Substance to a Binary Solution

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.”

Gφ(x,T)= (1−x)gAφ(T) +xgBφ(T) +RT[(1−x)ln(1−x) +xlnx] +Gexφ (8)

3.1 The ideal mixing term comes from the “number of arrangements”

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

ΔSmix= −R[(1−x)ln(1−x) +xlnx] (9)

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.

3.2 Expressing the reluctance to mix with a Redlich–Kister polynomial

In the phase diagram simulator, the deviation from an ideal solution is expressed as

Gexφ= x(1−x) [L0+L1(1−2x) +L2(1−2x)2] (10)

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.

In practical CALPHAD the coefficients are also functions of temperature Not only gAφ(T) and gBφ(T) but also Lkφ(T) may have the form a + bT + cT ln T + … Evaluating these coefficients from thermochemical data and phase equilibrium data is the CALPHAD method. The present simulator is a simplified teaching model and is not a CALPHAD thermodynamic assessment.

4. The Common Tangent — What It Means for the Chemical Potentials to Be Equal

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.

μAα=μAβ , μBα=μBβ (11)

Using the tangent to the molar free energy G(x), we have

μA=G−xdGdx , μB=G+(1−x) dGdx (12)

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.

4.1 The lever rule is nothing but the mass balance

Solving Eq. (2) for the phase fractions gives

fα= xβ−x0 xβ−xα , fβ= x0−xα xβ−xα (13)

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.

Gmix Lower convex hull (common tangent) Selected overall composition

4.2 The binodal and the spinodal are not the same

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.

d2Gdx2 =RTx(1−x) −2Ω (14)

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.

5. Stacking the Common Tangents of Each Temperature Gives the Phase Diagram

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 diagramWhat happens in the free energyTypical reaction and microstructure
Complete solid solutionThe 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 temperatureL → α. Composition differences can arise during solidification
EutecticAt the eutectic temperature, a single tangent line touches the three curves L, α, and β simultaneouslyL → α + β. Eutectic structures such as lamellar or rod-like ones
PeritecticAt 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 solidsL + 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 simultaneouslyL1 → α + L2. Bearing alloys such as Cu–Pb
Line compoundA phase with no composition range joins the competition as a single point on the G–x diagramCongruent melting and incongruent melting; multiple eutectics on either side of a compound
Ω alone produces the four basic forms (Fig. 3 of Tanaka 2008) If the melting points at the two ends (and the entropy of fusion) are fixed and only the interaction parameters ΩL and ΩS of the liquid and the solid are varied, all four basic forms appear: complete solid solution, eutectic, peritectic, and monotectic (Pelton and Thompson, Prog. Solid State Chem. 10, 119, 1975). As a rough guide, if Ω is negative for both liquid and solid the diagram is of the complete-solid-solution type; if it is positive for both, of the two-liquid-separation (monotectic) type; and if it is negative for the liquid and positive for the solid, of the eutectic type (Toshihiro Tanaka, “Alloys and Equilibrium Phase Diagrams” (in Japanese), Journal of the Japan Welding Society 77 (6), 576–581, 2008). However, the monotectic type arises only when Ω of the liquid is large enough that the critical temperature of liquid-phase separation (ΩL/2R for a regular solution) rises above the liquidus. If Ω is smaller than that, the separation of the liquid is hidden below the liquidus (metastable), and the diagram becomes eutectic or peritectic. If ΩL is larger still, a syntectic reaction (L1 + L2 → α), in which a solid forms from the two liquids, also appears. In addition, in a complete-solid-solution diagram with no three-phase reaction, if ΩL − ΩS is larger than ΔS(TB − TA) (entropy of fusion × difference between the melting points at the two ends; TB > TA, and the entropy of fusion is assumed to be the same at both ends), the liquidus and solidus have a maximum, and if it is smaller than −ΔS(TB − TA), they have a minimum; in between, a lens shape with no extremum results. Real, complicated phase diagrams, too, can be read as combinations of these basic forms and intermediate phases (compounds). With “Vary Ω (Fig. 3)” in simulator ②, you can check the 20 combinations and the free energy curves.
Whether metal or oxide, the same Ω determines the type The corresponding simulator lets you choose, side by side, metallic systems such as Cu–Ni and Ag–Cu and ceramic systems such as NiO–MgO, CaO–MgO, and Al2O3–Cr2O3. NiO–MgO is a complete solid solution and CaO–MgO is eutectic, yet both have the rock-salt structure. The crystal structure is not what separates the types. This is because the difference in cation radii produces a strain energy in the solid solution, and this strain energy determines the magnitude of L0 = Ω in Eq. (10). The larger Ω is, the more the G–x curve of the solid bulges up in the middle, and a single solid solution can no longer maintain a lower convex hull. The metals Cu–Ni (complete solid solution) and Ag–Cu (eutectic) show exactly the same contrast.

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.

Reading the Pb–Sn eutectic in the language of free energy At 183 °C and 61.9 mass% Sn (about 73.9 mol% Sn), the liquid L changes simultaneously into α (Pb-rich) and β (Sn-rich). At this temperature a single tangent touches GL, Gα, and Gβ, and the chemical potentials of Pb and of Sn each coincide in the three phases.

5.1 If cooling is fast, the lever rule can no longer be used

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

(CL− CS) dfS= (1−fS) dCL Scheil–Gulliver equation (differential form)

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.

5.2 Microstructure on slow cooling — primary crystals, eutectic structure, peritectic reaction, and precipitation

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.

fprimary= xE−X xE−xD , feutectic= X−xD xE−xD , fαin eutectic= xF−xE xF−xD Fractions of primary crystals and eutectic structure

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

fβin primary(T)= xD−xα(T) xβ(T)−xα(T) Precipitation inside the primary crystals

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

Before the reaction (just above TP) fα= xL−X xL−xa After the reaction (xP<X<xL) fβ= xL−X xL−xP ,fα=0 After the reaction (xa<X<xP) fα= xP−X xP−xa ,fL=0 Before and after the peritectic reaction

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.

Liquid L (eutectic composition) α β α β α β BBB AA Growth direction When α (A-rich) grows, B is left over ahead of it → B diffuses to the neighboring β When β (B-rich) grows, A is left over ahead of it → A diffuses to the neighboring α
Fig. 5-1. How a eutectic structure grows as layers (schematic)

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.

Conventions in the simulator’s microstructure pictures (1) A small number of primary-crystal nuclei (16, including ones outside the screen) are placed, and they grow at the same rate with their birth times shifted slightly (the Johnson–Mehl model). Because of this, the grain sizes are not uniform. (2) Assuming that regions solidify in the order in which the growth from the nuclei reaches them, areas are painted in proportion to the lever-rule fractions. The thin liquid film between grains that have collided solidifies last, and that becomes the grain boundary. (3) The eutectic structure changes the direction of the layers from one region of aligned orientation (colony) to the next, and the ratio of layer widths is matched to the fractions of α and β inside the eutectic structure. For systems drawn with rods, the minority phase is drawn as rods arranged on a triangular lattice, and the fraction is matched by the thickness of the rods. (4) Precipitates are placed inside grains (elongated particles in two orientations aligned grain by grain) and along grain boundaries. In every case the area ratios agree with the lever-rule answers (the difference checked in all 15 systems was within 0.5 %), but the shapes, sizes, and spacings of grains are schematic. Also, mole fractions are substituted for the area ratios (the area seen under a microscope is the volume fraction, so it deviates slightly if the molar volumes differ). (5) “On heating” is drawn as the microstructure when a polycrystal is heated above the solidus: in each grain, melting starts from the part far from the nucleus (near the grain boundary), and the original grain boundaries are left as dotted lines (they fade as the liquid fraction approaches 0.8). (6) Bands narrower than 0.8 K in which the same phase appears twice side by side (a minor by-product of the model) are skipped. Genuinely narrow regions, such as the thin L + α region just below the liquidus, are kept.

6. A Phase Diagram Alone Does Not Determine Grain Shape or the Time a Change Takes

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

ΔG(r)=4πr2γ +43πr3Δgv (15)

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

r*=2γ|Δgv| , ΔG*=16πγ3 3Δgv2 (16)

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.

7. Glass Transition and Crystallization — How Fast You Cool Decides Where You End Up

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.

7.1 Distinguishing four states in the enthalpy–temperature diagram

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,

Hc(T)= Cp,c(T−Tm) Hl(T)= ΔHm+ Cp,l(T−Tm) (17)

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

Hg(T)= Hl(Tf)+ Cp,g(T−Tf)Cp,g≈ Cp,c (18)

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.

Liquid and supercooled liquid Glass at the selected q Glass at other q (10−3, 103 K/s) Crystal
Values used in Fig. 3 The values for B2O3 (α-B2O3), Tm = 723 K, ΔHm = 24.07 kJ/mol, Cp,c = 106.69 J/(mol·K) at 723 K, and Cp,l = 129.70 J/(mol·K), are from the NIST-JANAF thermochemical tables. Treating the heat capacities as constants independent of temperature, and setting the heat capacity of the glass equal to that of the crystal, are approximations made for drawing the figure. Tg(q) is calculated by Eq. (22) in Section 7.3. Note that real B2O3 is practically never observed to crystallize at ordinary pressure from a melt without seed crystals, and crystal growth has been studied using seed crystals under GPa-class high pressure (Zanotto & Cassar 2017, Aziz et al. 1985). The line for the crystal in the figure is not a path that can actually be followed by cooling.

7.2 The glass transition is not a phase transition but the “freezing of relaxation”

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

σ(t)=σ0exp(−t/τ)τ=ηG∞ (19)

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

τ(Tg)≈δTq⟹η(Tg)≈G∞δTq (20)

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.

7.3 Temperature dependence of viscosity — the VFT equation and “strong” and “fragile” liquids

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).

log10(η/Pa·s)= A+BT−T0 (21)

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

Tg(q)=T0+ B12−log10(q/qref)−A (22)

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

m=[dlog10ηdT dTds]s=1 =[(−B(T−T0)2) (−Tgs2)]s=1 =BTg(Tg−T0)2 (23)

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.”

Table 1. Values for the three oxides used in this section (values marked “model” were fitted for teaching purposes and are not measured values)
QuantityLi2O·2SiO2 (LS2)SiO2B2O3
Tm1307 Ka1996 K (β-cristobalite)723 Kb
ΔHm57.4 kJ/molc8.9 kJ/mold24.07 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·s106.4 Pa·s104.5 Pa·s
Tg(10 K/min) (Eq. 22)724 K1480 K555 K
Tg/Tm0.550.740.77
m (Eq. 23)492559
Sources of Table 1 and cautions a: values used in Nascimento et al. 2011 and elsewhere. b: NIST-JANAF thermochemical tables (α-B2O3, from the step in enthalpy at 723 K). c: a value cited in the literature as that of Takahashi & Yoshio 1973; the figure in the original paper could not be confirmed. d: Richet et al. 1982 (confirmed in the abstract). e: the value in the table of Jiusti et al. 2020. The temperature at which η = 1012 Pa·s is then 724 K, which is consistent with T12 = 725–730 K for LS2 in Cassar et al. 2020. f: an equation made for teaching purposes to match the activation energies of Doremus 2002 (515 kJ/mol above 1400 °C and 720 kJ/mol below; confirmed in the abstract) and η(1480 K) = 1012 Pa·s. The Tg of silica varies greatly with the OH content. g: the value in the table of Zanotto & Cassar 2017. Calculating from this VFT with Eq. (23) gives m ≈ 59, but this does not agree with the value often cited in the literature for the fragility of B2O3 (about 32). It has not been confirmed which is appropriate, so the numerical values of m are kept here only as a rough guide.

7.4 The driving force for crystallization

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

Δgv=− ΔHmΔTTmVmΔT=Tm−T (24) Turnbull approximation

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.

7.5 Homogeneous nucleation — the barrier W* and the nucleation rate I

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.

dΔGdr= 8πrγ+4πr2Δgv=0⟹r*=−2γΔgv

Substituting this back into ΔG(r),

W*=ΔG(r*)= 4πγ4γ2Δgv2 −43π8γ3Δgv2 =16πγ33Δgv2

we obtain Eq. (16). Substituting Eq. (24) into it,

W*= 16πγ3Tm2Vm2 3ΔHm2(ΔT)2 ∝1(ΔT)2 (25)

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).

I(T)= γkBTd05η(T) exp(−W*kBT) (26)

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).

7.6 Crystal growth rate u

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

u(T)=fDd0 [1−exp(−ΔHmΔTRTTm)]D≈kBTd0η (27) Wilson–Frenkel form

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).

7.7 Crystal volume fraction — the JMAK equation

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

dX=(1−X)dXext⟹X=1−exp(−Xext) (28) Johnson–Mehl–Avrami–Kolmogorov

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

Xext= ∫0tI4π3 u3(t−t′)3dt′ =4π3Iu3t44 =π3Iu3t4 (29)

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.

Xext=4π3 ∫0tI(t′) [∫t′tu(s)ds]3dt′ =Cq4C=4π3 ∫TTmI(T′) [∫TT′u(T″)dT″]3dT′ (30)

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.

7.8 The TTT diagram and the critical cooling rate

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,

t(T)= [3XπI(T)u(T)3]1/4 (31)

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.

Rcnose≈Tm−TntnRccontinuous cooling=(CX)1/4 (32)

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)).

Nucleation rate I Growth rate u TTT curve (X = 10−6) Cooling curve for q = Rc
Table 2. Critical cooling rates calculated with the model, and literature values
MaterialNose Tn, tnRc (nose)Rc (continuous cooling)Literature benchmark
LS2892 K, 1.8×102 s2.3 K/s0.19 K/sMeasured 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)
SiO21528 K, 5.8×105 s8.0×10−4 K/s8.6×10−5 K/sAbout 1.1×10−4 K/s from a viscosity-based estimate (Sakaguchi 1995; confirmed in the abstract)
B2O3571 K, 1.2×108 s1.3×10−6 K/s7.8×10−8 K/sCrystallization is practically never observed at ordinary pressure without seed crystals (Zanotto & Cassar 2017)
Limits of the model — γ is a “fitted value” or a “provisional value” The interfacial energy γ was set to 0.151 for LS2, 0.065 for SiO2, and 0.100 J/m2 for B2O3. Apart from LS2, these are not measured values. The value for SiO2 was chosen to match the order of magnitude of Rc to the literature benchmark (Table 2). For B2O3 there is no literature Rc to compare with, so its value is a provisional one, fitted to nothing (in terms of Turnbull’s dimensionless interfacial energy, α ≈ 0.326, which is smaller than the 0.345 of LS2; B2O3 is hard to crystallize in the model not because of γ but because the viscosity is high and Tm − Tg is narrow). The Rc = 7.8×10−8 K/s of B2O3 in Table 2 is merely a value computed by the model, and there is no measurement to compare it with. Since W* ∝ γ3, changing γ by just 10 % changes Rc by many orders of magnitude. Moreover, the devitrification of real silica occurs mainly through heterogeneous nucleation starting at surfaces, impurities, and OH. That the Rc of LS2 in a container is larger than without a container can also be interpreted as a contribution of crystallization starting from the container walls and the like. It is also known that in deep supercooling the diffusion ceases to agree with the value expected from the viscosity (Nascimento et al. 2011). This model is meant for looking at orders of magnitude and the shapes of the curves.

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).

7.9 Why SiO2 and B2O3 form glass easily — Zachariasen’s rules

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.

  1. Oxygen is not bonded to more than 2 A atoms.
  2. The number of oxygens surrounding A is small (3 or 4).
  3. The oxygen polyhedra share only corners, not edges or faces.
  4. To build a three-dimensional network, each polyhedron shares at least 3 of its corners with other polyhedra.

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.

7.10 Putting crystallization to use — crystallized glass (glass-ceramics)

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).

7.11 Connection with the “Quench” of simulator ③

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.

What kind of state is glass, after all? Glass is not an equilibrium state but a nonequilibrium state in which the structure is frozen. Zanotto & Mauro 2017 define glass, in essence, as “a nonequilibrium, noncrystalline condensed state that shows a glass transition, with a structure resembling that of the original supercooled liquid, that spontaneously relaxes toward the supercooled liquid and, in the limit of infinitely long time, crystallizes.” Also, there is a temperature at which the entropy of a supercooled liquid, extrapolated to low temperature, would become smaller than that of the crystal (Kauzmann 1948), but in reality the glass transition occurs before that and the extrapolation stops. The interpretation of this “Kauzmann paradox” has not yet been settled (Debenedetti & Stillinger 2001).

7.12 What simulator ④ shows

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

dTfdt= T−Tfτ(T)

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

Table 3. Presets of simulator ④ (values marked “model” or “schematic value” were set for teaching purposes and are not measured values)
QuantityLi2O·2SiO2 (initial display)SiO2B2O3
Tm, ΔHm, VFTSame as Table 1
M, ρ (crystal)150.05 g/mol, about 2.46 g/cm360.08, about 2.2069.62, about 2.46
σ (= γ)0.1514 J/m2 (model)0.065 (model)0.100 (model)
g (= f)0.23 (model)11
Cp,c, Cp,l224, 280 J/(mol·K) (schematic values)75.9, 85.8 (JANAF)106.7, 129.7 (JANAF)
Temperature range Tlo–Thi580–1400 K1150–2150 K470–800 K
q slider (K/s)10−5–10610−8–10310−11–104
q for “Cool slowly”10−3 K/s10−6 K/s10−9 K/s
Rc (continuous cooling, X = 10−6)0.18 K/s8.6×10−5 K/s7.8×10−8 K/s
Tg(10 K/min): simulator727 K1459 K559 K
Tg(10 K/min): Eq. (22) (Table 1)724 K1480 K555 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

Where the numbers in the text and in the simulator differ slightly Tg. The Tg in Table 1 and Fig. 3 is the value of Eq. (22), referenced to η(Tg) = 1012 Pa·s (10 K/min). The Tg of the simulator is the temperature at which the fictive temperature freezes in the single-relaxation equation above, so at 10 K/min it comes out a few K higher for LS2, at 727 K (724 K by Eq. 22), and for B2O3, at 559 K (555 K), and about 20 K lower for SiO2, at 1459 K (1480 K). SiO2 is a “strong liquid” whose viscosity changes gently with temperature, so the temperature range over which the relaxation freezes is wide, and the difference between the two definitions becomes large. For SiO2 the difference widens as q increases; at 102 K/s the simulator gives 1634 K against 1679 K from Eq. (22). Both express the same conclusion, “Tg rises with q,” with different approximations. Rc and the nose. The equations and values for nucleation and growth are the same as in Fig. 4, and the nose of the isothermal TTT (892 K and about 180 s for LS2) and the Rc for continuous cooling agree with Table 2 in order of magnitude.
Limits of simulator ④ (i) σ is a value matched to the order of magnitude of Rc for SiO2 and a provisional value fitted to nothing for B2O3, and Rc is extremely sensitive to σ (Section 7.8). The devitrification of real silica occurs mainly through heterogeneous nucleation starting at surfaces and impurities, which is not included in the model. (ii) The relaxation uses a single relaxation time; nonlinear, nonexponential relaxation (Tool–Narayanaswamy–Moynihan type) is not included. For this reason the overshoot of Cp seen on heating in DSC does not appear. (iii) The VFT equation is extrapolated even below Tg, whereas the viscosity of real glass can be lower than this. (iv) Non-steady-state nucleation, surface crystallization, and the decoupling of diffusion and viscosity in deep supercooling (Nascimento et al. 2011) are not included. (v) The crystal is assumed to melt instantaneously at Tm. (vi) An extremely small value such as X = 10−17 is an expectation value of the volume fraction (a value “on average”). The expected number of nuclei is ∫I dt multiplied by the volume of the sample, so it means that for a sample of ordinary size it is highly likely that not a single nucleus forms. (vii) The box of the atom picture is a schematic of a few nm, the hopping rate is compressed logarithmically, and the number of crystal grains is chosen for visibility. This model is meant for looking at the competition between q and Rc, the cooling-rate dependence of Tg, and the shape of the H path.

8. The Phase-Field Method — Holding the Interface Not as a Line but as a Smooth Field

8.1 Adding a gradient energy to the local free energy

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

ℱ[c,η]= ∫V[ fchem(c,η,T) +κc2|∇c|2 +κη2|∇η|2 ]dV (33)

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.

8.2 Conserved and nonconserved quantities obey different evolution equations

Composition is a conserved quantity that cannot disappear on its own, so with the chemical potential

μ=δℱδc =∂fchem∂c −κc∇2c (34)

it flows down its gradient, with flux J = −M∇μ. This is the Cahn–Hilliard equation.

∂c∂t =∇·[M∇μ] =∇·[M∇ (∂fchem∂c −κc∇2c)] (35) Cahn–Hilliard

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

∂η∂t =−Lδℱδη =−L( ∂fchem∂η −κη∇2η) (36) Allen–Cahn

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.”

8.3 Smoothly connecting the free energies of the liquid and solid

For the solidification of a binary alloy, one can take, for example,

fchem= h(η)fS(c,T) +[1−h(η)] fL(c,T) +Wg(η) (37)

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).

A-rich (c < 0) Intermediate B-rich (c > 0)
Fig. 5 is a dimensionless model for looking at shapes To convert to the nm, s, and J/m² of a real material, one must supply the CALPHAD free energy, atomic mobility, interfacial energy, molar volume, and so on in consistent units, and also verify the influence of the artificial interface width. An image that looks plausible is not the same thing as a quantitative prediction.

9. Spinodal Decomposition — Phase Separation in Which Composition Waves Grow Without Forming Nuclei

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.

9.1 The G–x curve and phase diagram of two-liquid immiscibility

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

∂2g∂x2 =kB[Tx(1−x)−4Tc] (38)

Here the per-mole Eq. (14) has been divided by NA, using R/NA = kB. Two curves follow from this.

Spinodal: xs=12[1±1−T/Tc] Binodal: lnxb1−xb= 2TcT(2xb−1) (39)

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.

g(x) Binodal Spinodal

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).

The names of the three regions Outside the binodal: a uniform solution is stable. Between the binodal and the spinodal: a uniform solution is metastable. It is stable against infinitesimal fluctuations, but if a nucleus larger than a certain size forms, it separates into two phases (nucleation and growth). Inside the spinodal: a uniform solution is unstable. Small fluctuations of sufficiently long wavelength grow without a barrier. The “two liquid phases” here lie, in BaO–SiO2 and the like, in a metastable region below the liquidus, and are not necessarily the two-phase region of the equilibrium phase diagram.

9.2 Free energy including the gradient energy

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

G=Nv∫V [g(C)+κ(∇C)2]dV (40) Kingery Eq. 8.8

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).

9.3 Does G go up or down under a small fluctuation?

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

g(C)=g(C0) +(∂g∂C)C0(C−C0) +12(∂2g∂C2)C0(C−C0)2+⋯

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

ΔG=Nv∫V [12(∂2g∂C2)C0 (C−C0)2 +κ(∇C)2]dV (41) Kingery Eq. 8.9

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

12|∂2g∂C2|C0 (C−C0)2 >κ(∇C)2 (42) Kingery Eq. 8.10

that is, fluctuations whose gradient is sufficiently gentle (long wavelength). A solution inside the spinodal is always unstable against some fluctuation.

9.4 A sinusoidal fluctuation and the critical wavelength λc

Take the fluctuation to be a single sine wave and examine which wavelengths are unstable.

C−C0=Acosβx (43) Kingery Eq. 8.11

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

ΔGV=Nv [12g″A22 +κA2β22] =NvA24 [g″+2κβ2] (44) Kingery Eq. 8.12

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).

βc=[−12κg″]1/2λc=2πβc =[−8π2κg″]1/2 (45) Kingery Eq. 8.13a, b

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.

9.5 Which wavelength grows fastest — linearizing the Cahn–Hilliard equation

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

∂C∂t= M~Nv ∇2[g′(C)−2κ∇2C] (46) Cahn–Hilliard (the same form as Eq. 35)

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

∂δC∂t= M~Nv [g″∇2δC−2κ∇4δC] (47)

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

A(β,t)=A(β,0)exp[R(β)t]R(β)=−M~β2Nv [g″+2κβ2] (48) Kingery Eqs. 8.14, 8.15

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

dRdβ= −M~Nv [2βg″+8κβ3]=0⟹βm2=−g″4κ=βc22

so

λm=2λc (49) Kingery Eq. 8.16

The maximum value is obtained by putting 2κβm2 = −g″/2 back into Eq. (48):

Rm=−12M~Nvg″βm2 =2M~Nvκβm4 =M~Nv(g″)28κ (50) Kingery Eq. 8.17
How to read Eq. 8.17 of Kingery Eq. 8.17 of the textbook is printed so that it looks as if two equations, Rm = −½M̃(∂2f/∂C2)C0 and βm2 = 2M̃κβm4, were lined up, but read that way the dimensions of the two sides do not match. As in the derivation above, it is consistent to read it as a single continued equation Rm = −½M̃g″βm2 = 2M̃κβm4 (in a form that, unlike Eq. 8.15, omits 1/Nv). The symbol f is thought to refer to the same quantity as g from Eq. 8.8 onward.

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

Rm=|D~|βm221Rm=λm22π2|D~| (51)

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.

9.6 How it differs from nucleation and growth

Nucleation and growth (between the binodal and the spinodal)Spinodal decomposition (inside the spinodal)
How it startsSmall nuclei with a large composition difference are born stochastically, here and thereWaves with a small composition difference grow everywhere at the same time
BarrierYes (ΔG* in Eq. 16)None (waves with R > 0 grow from the start)
CompositionThe nucleus has a near-equilibrium composition from the startThe amplitude grows continuously and eventually approaches the equilibrium composition
ScaleSet by the size and number density of the nuclei, and scatteredInitially, waves near the wavelength λm grow together
DiffusionOrdinary 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).

9.7 Interconnected microstructures, and the effect of the average composition

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.

An interconnected microstructure alone does not prove “it was spinodal decomposition” In a two-dimensional cross section, a phase that is connected in three dimensions can look like islands. Conversely, an interconnected microstructure also forms when grains born by nucleation coalesce. Seward et al. 1968b heated a thin film of BaO–SiO2 inside an electron microscope and observed isolated grains coalescing into an interconnected structure, and stated that judging the mechanism from the final microstructure alone is insufficient. Haller 1965 also discussed the possibility that an interconnected microstructure can be explained by the overlap of nucleated spheres.

9.8 Subsequent coarsening, and what happens when the temperature is raised

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).

In crystals, strain lowers the spinodal In a crystal whose lattice constant changes with composition, a coherent strain energy is added to the composition wave, and the condition for decomposition changes to the requirement that g″ plus a strain term be negative (Cahn 1962). As a result, the coherent spinodal shifts to the low-temperature side of the chemical spinodal, and waves in the elastically soft direction are selected. In TiO2–SnO2 (tetragonal, rutile type) the lowering is smallest along [001] (105 K in a regular-solution calculation), and a plate-like modulated structure perpendicular to [001] forms (Park et al. 1975, Stubican & Schultz 1970). Isotropic liquids and glasses do not have this term.

9.9 Inorganic materials that show spinodal decomposition

SystemWhat is knownSource
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).

9.10 Applications

9.11 What simulator ⑤ shows

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

g(φ)RTc= TTc[φNlnφ+(1−φ)ln(1−φ)] +χcφ(1−φ) φc=11+N, χc=12(1+1N)2 (52)

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

∂φ∂t= m(T)∇2[g′(φ)−2∇2φ] +(coherent strain)+ξ m(T)=M(T)M(Tc)= exp[−EaR(1T−1Tc)] (53)

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

Table 4. Presets of simulator ⑤ (temperatures and compositions are guides to real systems; times and lengths are teaching conversions)
QuantityBorosilicate glass (initial display)BaO–SiO2 glassTiO2–SnO2 solid solution
ModelRegular solution (N = 1). Na2O–B2O3–SiO2 is treated as a pseudo-binary systemEq. (52), N = 92/3 ≈ 4.33Regular solution + coherent strain
Critical point760 °C (guide from secondary sources), c = 50 %1460 °C, 10 mol% BaO1430 °C, 50 % (actually 47 mol% TiO2)
Ea of the mobility167 kJ/mol200 kJ/mol (schematic value)200 kJ/mol (schematic value)
Temperature slider450–850 °C950–1600 °C850–1550 °C
Initial holding temperature600 °C1150 °C1000 °C
Composition slider (initial value)Normalized composition c 5–95 % (50 %)1–40 mol% BaO (10 %)5–95 mol% SnO2 (50 %)
Length conversion ℓAssumed to be 1 nmAssumed to be 1 nmAbout 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

Table 5. Values at the initial holding temperature and initial composition (linear theory)
QuantityBorosilicate glass, 600 °C, 50 %BaO–SiO2, 1150 °C, 10 mol%TiO2–SnO2, 1000 °C, 50 %
T/Tc0.8450.8210.748
Binodal18.1–81.9 %1.5–34.9 mol%11.1–88.9 mol%
Spinodal30.3–69.7 %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, λm11.3 ℓ, 16.0 ℓ14.2 ℓ, 20.1 ℓ10.2 ℓ, 14.4 ℓ (17.3 nm)
Rm1.4×10−3/τ09.4×10−4/τ06.2×10−4/τ0
M(T)/M(Tc)0.0280.0490.0085

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).

Limits of simulator ⑤ (i) It is a two-dimensional calculation. The “microstructure in which both phases are interconnected in three dimensions” of Cahn 1965 cannot be reproduced in two dimensions (in two dimensions, the two phases cannot both be connected vertically and horizontally at the same time). For this reason the shape of the microstructure is judged using the elongation Q as well. (ii) The time τ0 and length ℓ are teaching conversions. Because measured values of the interdiffusion coefficient and κ have not been confirmed, no conversion to seconds is given. The ℓ = 1 nm for glass is an assumption, and ℓ for TiO2–SnO2 is a value matched to a single measurement. (iii) Borosilicate glass is a schematic that treats a three-component system as pseudo-binary, and effects such as rotation of the tie lines are not included. The critical temperature of 760 °C is a rough guide. (iv) The 200 kJ/mol for Ea, the strength of the fluctuations, and N ≈ 4.33 are schematic or phenomenological values. (v) The coherent strain is approximated by a constant for each direction. The size of a nucleus that grows is a rough guide estimated with the strain eN between those of the softest direction and the direction average (using the fitted coefficient 0.4). In real materials, nuclei that have lost coherency can also form. (vi) λm and Rm of linear theory are quantities for the initial stage only, and the lengthening of the wavelength in the coarsening stage appears only in the numerical calculation. (vii) In the metastable region, nuclei do not form spontaneously. Because the thermal fluctuations are weak and the box is small, within the time range of the screen the compositions outside the spinodal (the metastable region) stay uniform. In real materials, phase separation occurs in the metastable region as well, by (often heterogeneous) nucleation and growth (§8.2 of Kingery). The button that places a nucleus is there to do this on the user’s behalf. (viii) As cautioned in Section 9.7, a resemblance between the microstructure on the screen and a real one is not evidence that spinodal decomposition occurred in the real material.

10. The Division of Roles Between the Phase-Diagram Simulator and the Phase-Field Method

QuestionThe current phase-diagram simulatorQuantitative phase-field calculation
Which phase is stableFound from the lower convex hullUses the same bulk free energy as the local driving force
Composition and fraction of each phaseCommon tangent and lever ruleFound by integrating the local composition over space
Segregation during coolingThe two limits of equilibrium and Scheil–GulliverSolves directly with finite diffusion coefficients and interface velocities
Grain shape and sizeA 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 fractionsDeveloped from interfacial energy, anisotropy, nucleation, and elastic fields
Prediction of times and lengthsNot treatedTreated by calibrating the mobility, gradient coefficient, and spatial step to physical units
The two are not rivals; they connect one after the other CALPHAD / phase-diagram calculation supplies the chemical driving force, “which phase a place locally wants to turn into,” and the phase-field method adds “the cost of making an interface” and “the speed of diffusion and interface motion,” and solves how that driving force turns into a microstructure in space.

11. How to Proceed in an Actual Materials Analysis

  1. Choose a thermodynamic model. Prepare Gφ(x,T) for each phase from experiments, first-principles calculations, or a CALPHAD database.
  2. Verify the equilibrium. Check that the phase boundaries obtained from the common tangent, the eutectic and peritectic temperatures, the activities, and the enthalpies of formation are consistent with known data.
  3. Supply kinetic data. Give the diffusion coefficients or atomic mobilities, the interface mobility, the interfacial energy, and the anisotropy.
  4. Choose the fields. Couple only the degrees of freedom you need, such as the composition c, the phase field η, the crystal orientation, the temperature, and the elastic displacement.
  5. Verify lengths and times. Check that the answer converges when the interface width and lattice spacing are changed, and that mass conservation and the decrease of free energy are satisfied.
  6. Compare with experiment. Compare not only the phase fractions but also the dendrite arm spacing, the precipitate radius, the composition profile, and the growth rate.

12. Common Misconceptions

MisconceptionThe correct reading
If the free energy is negative, the state is stableThe 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 fractionThe 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 unstableBetween 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 diagramA 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 interchangeablyThe 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 pointTg 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 cooledWhether 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 decompositionAn 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 correctVerification is needed of mass conservation, energy dissipation, interface-width dependence, physical parameters, and the time and length scales.

13. Summary

  1. The G(T) of a pure substance is determined by the entropy and the heat capacity. dG/dT = −S and d²G/dT² = −Cp/T.
  2. For a binary solution, add the end members, the entropy of mixing, and the excess free energy. At high temperatures the RT mixing term dominates, and a positive Ω promotes phase separation.
  3. Equilibrium is determined by the lower convex hull. The common tangent is the geometric expression of the equality of the chemical potentials of both components between phases.
  4. Stacking the points of tangency at each temperature gives the phase diagram. The composition of a phase is the point of tangency, and the fraction of a phase is given by the lever rule.
  5. A phase diagram shows the final candidates for the microstructure, but not its shape or the time it takes. Only when interfaces, diffusion, and nucleation are included does it become microstructure formation.
  6. The phase-field method is a theory that extends the same free energy into space. A conserved composition follows Cahn–Hilliard and a nonconserved order parameter follows Allen–Cahn, and both evolve while dissipating the free energy.
  7. The glass transition is the freezing of relaxation. Tg changes with the cooling rate, and whether the liquid becomes a crystal or a glass is decided by the critical cooling rate Rc, obtained from the nucleation rate I and the growth rate u.
  8. Inside the spinodal, composition waves grow without a barrier. The wave with λm = √2 λc, at which the amplification factor R(β) = −(M̃β2/Nv)[g″ + 2κβ2] is maximal, sets the scale of the microstructure.

Corresponding simulator: Binary phase diagrams. Related resources: Gibbs free energy and phases, Entropy and free energy.
Main sources and data sources

  1. O. Redlich and A. T. Kister, “Algebraic Representation of Thermodynamic Properties and the Classification of Solutions,” Ind. Eng. Chem. 40, 345–348 (1948). doi:10.1021/ie50458a036
  2. A. T. Dinsdale, “SGTE Data for Pure Elements,” CALPHAD 15, 317–425 (1991). doi:10.1016/0364-5916(91)90030-N
  3. J. W. Cahn and J. E. Hilliard, “Free Energy of a Nonuniform System. I. Interfacial Free Energy,” J. Chem. Phys. 28, 258–267 (1958). doi:10.1063/1.1744102
  4. J. W. Cahn, “On Spinodal Decomposition,” Acta Metall. 9, 795–801 (1961). doi:10.1016/0001-6160(61)90182-1
  5. S. M. Allen and J. W. Cahn, “A Microscopic Theory for Antiphase Boundary Motion and Its Application to Antiphase Domain Coarsening,” Acta Metall. 27, 1085–1095 (1979). doi:10.1016/0001-6160(79)90196-2
  6. A. A. Wheeler, W. J. Boettinger, and G. B. McFadden, “Phase-Field Model for Isothermal Phase Transitions in Binary Alloys,” Phys. Rev. A 45, 7424–7439 (1992). doi:10.1103/PhysRevA.45.7424

§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)

  1. W. H. Zachariasen, “The Atomic Arrangement in Glass,” J. Am. Chem. Soc. 54, 3841–3851 (1932). doi:10.1021/ja01349a006
  2. G. S. Fulcher, “Analysis of Recent Measurements of the Viscosity of Glasses,” J. Am. Ceram. Soc. 8, 339–355 (1925). doi:10.1111/j.1151-2916.1925.tb16731.x
  3. G. Tammann and W. Hesse, “Die Abhängigkeit der Viscosität von der Temperatur bie unterkühlten Flüssigkeiten,” Z. Anorg. Allg. Chem. 156, 245–257 (1926) (the spelling of the title is as registered in Crossref). doi:10.1002/zaac.19261560121
  4. A. Q. Tool, “Relation Between Inelastic Deformability and Thermal Expansion of Glass in Its Annealing Range,” J. Am. Ceram. Soc. 29, 240–253 (1946). doi:10.1111/j.1151-2916.1946.tb11592.x
  5. W. Kauzmann, “The Nature of the Glassy State and the Behavior of Liquids at Low Temperatures,” Chem. Rev. 43, 219–256 (1948). doi:10.1021/cr60135a002
  6. F. E. Wagstaff, “Crystallization and Melting Kinetics of Cristobalite,” J. Am. Ceram. Soc. 52, 650–654 (1969). doi:10.1111/j.1151-2916.1969.tb16069.x
  7. D. Turnbull, “Under What Conditions Can a Glass Be Formed?,” Contemp. Phys. 10, 473–488 (1969). doi:10.1080/00107516908204405
  8. S. Sakka and J. D. Mackenzie, “Relation Between Apparent Glass Transition Temperature and Liquids Temperature for Inorganic Glasses,” J. Non-Cryst. Solids 6, 145–162 (1971) (the spelling of the title is as registered in Crossref). doi:10.1016/0022-3093(71)90053-6
  9. D. R. Uhlmann, “A Kinetic Treatment of Glass Formation,” J. Non-Cryst. Solids 7, 337–348 (1972). doi:10.1016/0022-3093(72)90269-4
  10. K. Takahashi and T. Yoshio, “Thermodynamic Quantities of Alkali Silicates in the Temperature Range from 25 °C to Melting Point,” J. Ceram. Assoc. Jpn. 81, 524–533 (1973) (in Japanese; bibliographic details checked in Crossref; the ΔHm value could not be confirmed in the original source). doi:10.2109/jcersj1950.81.940_524
  11. C. T. Moynihan, A. J. Easteal, M. A. DeBolt, and J. Tucker, “Dependence of the Fictive Temperature of Glass on Cooling Rate,” J. Am. Ceram. Soc. 59, 12–16 (1976). doi:10.1111/j.1151-2916.1976.tb09376.x
  12. P. Richet, Y. Bottinga, L. Denielou, J. P. Petitet, and C. Tequi, “Thermodynamic Properties of Quartz, Cristobalite and Amorphous SiO2: Drop Calorimetry Measurements Between 1000 and 1800 K and a Review from 0 to 2000 K,” Geochim. Cosmochim. Acta 46, 2639–2658 (1982). doi:10.1016/0016-7037(82)90383-0
  13. M. J. Aziz, E. Nygren, J. F. Hays, and D. Turnbull, “Crystal Growth Kinetics of Boron Oxide under Pressure,” J. Appl. Phys. 57, 2233–2242 (1985). doi:10.1063/1.334368
  14. W. Huang, C. S. Ray, and D. E. Day, “Dependence of the Critical Cooling Rate for Lithium-Silicate Glass on Nucleating Agents,” J. Non-Cryst. Solids 86, 204–212 (1986). doi:10.1016/0022-3093(86)90488-6
  15. M. Avrami, “Kinetics of Phase Change. I General Theory,” J. Chem. Phys. 7, 1103–1112 (1939). doi:10.1063/1.1750380
  16. M. Avrami, “Kinetics of Phase Change. II Transformation-Time Relations for Random Distribution of Nuclei,” J. Chem. Phys. 8, 212–224 (1940). doi:10.1063/1.1750631
  17. M. C. Weinberg, D. R. Uhlmann, and E. D. Zanotto, “‘Nose Method’ of Calculating Critical Cooling Rates for Glass Formation,” J. Am. Ceram. Soc. 72, 2054–2058 (1989). doi:10.1111/j.1151-2916.1989.tb06030.x
  18. C. A. Angell, “Formation of Glasses from Liquids and Biopolymers,” Science 267, 1924–1935 (1995). doi:10.1126/science.267.5206.1924
  19. S. Sakaguchi, “Evaluation of the Critical Cooling Rate in Glass-Forming Materials Based on Viscosity,” J. Non-Cryst. Solids 185, 268–273 (1995). doi:10.1016/0022-3093(94)00682-2
  20. P. G. Debenedetti and F. H. Stillinger, “Supercooled Liquids and the Glass Transition,” Nature 410, 259–267 (2001). doi:10.1038/35065704
  21. R. H. Doremus, “Viscosity of Silica,” J. Appl. Phys. 92, 7619–7629 (2002). doi:10.1063/1.1515132
  22. K. S. Ranasinghe, C. S. Ray, D. E. Day, J. R. Rogers, R. W. Hyers, and T. Rathz, “Containerless Processing of a Lithium Disilicate Glass,” J. Mater. Sci. 42, 4291–4297 (2007). doi:10.1007/s10853-006-1232-z
  23. M. L. F. Nascimento, V. M. Fokin, E. D. Zanotto, and A. S. Abyzov, “Dynamic Processes in a Silicate Liquid from above Melting to below the Glass Transition,” J. Chem. Phys. 135, 194703 (2011). doi:10.1063/1.3656696
  24. W. Höland and G. H. Beall, Glass-Ceramic Technology (Wiley, 2012). doi:10.1002/9781118265987
  25. K. Thieme, I. Avramov, and C. Rüssel, “The Mechanism of Deceleration of Nucleation and Crystal Growth by the Small Addition of Transition Metals to Lithium Disilicate Glasses,” Sci. Rep. 6, 25451 (2016). doi:10.1038/srep25451
  26. E. D. Zanotto and D. R. Cassar, “The Microscopic Origin of the Extreme Glass-Forming Ability of Albite and B2O3,” Sci. Rep. 7, 43022 (2017). doi:10.1038/srep43022
  27. E. D. Zanotto and J. C. Mauro, “The Glassy State of Matter: Its Definition and Ultimate Fate,” J. Non-Cryst. Solids 471, 490–495 (2017). doi:10.1016/j.jnoncrysol.2017.05.019
  28. D. R. Cassar, A. H. Serra, O. Peitl, and E. D. Zanotto, “Critical Assessment of the Alleged Failure of the Classical Nucleation Theory at Low Temperatures,” J. Non-Cryst. Solids 547, 120297 (2020). doi:10.1016/j.jnoncrysol.2020.120297
  29. J. Jiusti, D. R. Cassar, and E. D. Zanotto, “Which Glass Stability Parameters Can Assess the Glass-Forming Ability of Oxide Systems?,” Int. J. Appl. Glass Sci. 11, 612–621 (2020). doi:10.1111/ijag.15416
  30. M. W. Chase Jr., NIST-JANAF Thermochemical Tables, 4th ed., J. Phys. Chem. Ref. Data Monograph 9 (1998). The values for B2O3 (table B-097) and SiO2 (table O-039) were used (not checked against Crossref because there is no DOI).

§9 Spinodal decomposition (for Cahn 1961 and Cahn & Hilliard 1958, see the list above)

  1. W. D. Kingery, H. K. Bowen, and D. R. Uhlmann, Introduction to Ceramics, 2nd ed. (Wiley, New York, 1976), §8.2 Spinodal Decomposition, pp. 323–327 (a book; the source attribution “Seward 1970” of Fig. 8.2 could not be identified in Crossref).
  2. H. P. Hood and M. E. Nordberg, “Treated Borosilicate Glass,” US Patent 2,106,744 (1938) (a patent, so there is no DOI).
  3. M. E. Nordberg, “Properties of Some Vycor-Brand Glasses,” J. Am. Ceram. Soc. 27, 299–305 (1944). doi:10.1111/j.1151-2916.1944.tb14473.x
  4. E. A. Porai-Koshits and N. S. Andreyev, “Low-Angle X-Ray Scattering by Glasses,” Nature 182, 335–336 (1958). doi:10.1038/182335a0
  5. J. W. Cahn, “On Spinodal Decomposition in Cubic Crystals,” Acta Metall. 10, 179–183 (1962). doi:10.1016/0001-6160(62)90114-1
  6. J. W. Cahn, “Phase Separation by Spinodal Decomposition in Isotropic Systems,” J. Chem. Phys. 42, 93–99 (1965). doi:10.1063/1.1695731
  7. W. Haller, “Rearrangement Kinetics of the Liquid–Liquid Immiscible Microphases in Alkali Borosilicate Melts,” J. Chem. Phys. 42, 686–693 (1965). doi:10.1063/1.1695991
  8. T. P. Seward III, D. R. Uhlmann, and D. Turnbull, “Phase Separation in the System BaO–SiO2,” J. Am. Ceram. Soc. 51, 278–285 (1968) (“Seward et al. 1968a” in the text). doi:10.1111/j.1151-2916.1968.tb13858.x
  9. T. P. Seward III, D. R. Uhlmann, and D. Turnbull, “Development of Two-Phase Structure in Glasses, with Special Reference to the System BaO–SiO2,” J. Am. Ceram. Soc. 51, 634–642 (1968) (“Seward et al. 1968b” in the text). doi:10.1111/j.1151-2916.1968.tb12635.x
  10. W. Haller, D. H. Blackburn, F. E. Wagstaff, and R. J. Charles, “Metastable Immiscibility Surface in the System Na2O–B2O3–SiO2,” J. Am. Ceram. Soc. 53, 34–39 (1970). doi:10.1111/j.1151-2916.1970.tb11995.x
  11. V. S. Stubican and A. H. Schultz, “Phase Separation by Spinodal Decomposition in the Tetragonal System,” J. Am. Ceram. Soc. 53, 211–214 (1970). doi:10.1111/j.1151-2916.1970.tb12072.x
  12. H. E. Cook, “Brownian Motion in Spinodal Decomposition,” Acta Metall. 18, 297–306 (1970). doi:10.1016/0001-6160(70)90144-6
  13. J. W. Cahn and J. E. Hilliard, “Spinodal Decomposition: A Reprise,” Acta Metall. 19, 151–161 (1971). doi:10.1016/0001-6160(71)90127-1
  14. M. Takahashi and M. E. Fine, “Magnetic Behavior of Quenched and Aged CoFe2O4–Co3O4 Alloys,” J. Appl. Phys. 43, 4205–4216 (1972). doi:10.1063/1.1660897
  15. M. Park, T. E. Mitchell, and A. H. Heuer, “Subsolidus Equilibria in the TiO2–SnO2 System,” J. Am. Ceram. Soc. 58, 43–47 (1975). doi:10.1111/j.1151-2916.1975.tb18980.x
  16. I. Ho and G. B. Stringfellow, “Solid Phase Immiscibility in GaInN,” Appl. Phys. Lett. 69, 2701–2703 (1996). doi:10.1063/1.117683
  17. D. Doppalapudi, S. N. Basu, K. F. Ludwig, and T. D. Moustakas, “Phase Separation and Ordering in InGaN Alloys Grown by Molecular Beam Epitaxy,” J. Appl. Phys. 84, 1389–1395 (1998). doi:10.1063/1.368251
  18. A. Hörling, L. Hultman, M. Odén, J. Sjölén, and L. Karlsson, “Thermal Stability of Arc Evaporated High Aluminum-Content Ti1−xAlxN Thin Films,” J. Vac. Sci. Technol. A 20, 1815–1823 (2002). doi:10.1116/1.1503784
  19. P. H. Mayrhofer, A. Hörling, L. Karlsson, J. Sjölén, T. Larsson, C. Mitterer, and L. Hultman, “Self-Organized Nanostructures in the Ti–Al–N System,” Appl. Phys. Lett. 83, 2049–2051 (2003). doi:10.1063/1.1608464
  20. P. H. Mayrhofer, C. Mitterer, and H. Clemens, “Self-Organized Nanostructures in Hard Ceramic Coatings,” Adv. Eng. Mater. 7, 1071–1082 (2005). doi:10.1002/adem.200500154