An historical perspective may prove useful to engineers or scientists who encounter mullite during the course of their work: The earliest interpretations of the material’s behavior may reflect the result of nonequilibrium conditions that often occur in production or experimental situations.
Mullite-based ceramics have been widely used as refractories and in pottery for millennia. Although the technology of mullite is becoming more mature, there are still questions concerning its melting behavior and the shape of mullite phase boundaries in the Al2O3-SiO2 phase diagram. In 1924, Bowen and Grieg  published the first phase diagram to include mullite as a stable phase, but did not indicate a solid solution range. The phase 3Al2O32SiO2 was reported to melt incongruently at 1,810°C. Specimens were prepared from mechanical mixtures of alumina and silica melted and quenched in air. Shears and Archibald  reported the presence of a solid solution range from 3Al2O32SiO2 (3:2 mullite) to 2Al2O3SiO2 (2:1 mullite) in 1954. Their phase diagram depicted a mullite solidus shifting to higher alumina concentrations at temperatures above the silica-mullite eutectic temperature.
In 1958, Toropov and Galakhov  presented a phase diagram where mullite was shown to melt congruently at 1,850°C. Aramaki and Roy  published a phase diagram in 1962 corroborating a congruent melting point for mullite at 1,850°C. Their specimens were prepared from gels for subsolidus heat treatments, while mechanical mixtures of a-Al2O3 and silica glass were prepared for heat treatments above the solidus temperature. Specimens were encapsulated to inhibit silica volatilization. A silica-mullite eutectic temperature of 1,595°C and a mullite-alumina eutectic temperature of 1,840°C were reported. No shift in the mullite solidus phase boundary with temperature was reported in either of these publications.
Over a decade later, Aksay and Pask  presented a different phase diagram depicting incongruent melting for mullite at 1,828°C. Specimens, in the form of diffusion couples between sapphire and aluminosilicate glass, were also encapsulated to inhibit volatilization. Many authors suggest that nucleation and growth of mullite occurs within an amorphous alumina-rich siliceous phase located between the silica and alumina particles [21-24]. On the other hand, Davis and Pask  and later Aksay and Pask observed coherent mullite growth on sapphire in a temperature range from about 1,600 to below 1,800°C, indicating interdiffusion of aluminum and silicon ions through the mullite . Risbud and Pask  later modified the diagram to incorporate metastable phase regions. They showed a stable silica-mullite eutectic temperature of 1,587°C. An immiscibility dome with a spinodal region was reported between approximately 7 and 55 mol% Al2O3. The dome has a central composition of about 35 mol% Al2O3, and complete miscibility occurs near 1,550°C (temperatures below the silica-mullite eutectic temperature). A stable mullite-alumina peritectic was reported at 1,828°C. However, a “metastable” incongruent melting point for mullite was reported at 1,890°C. The “metastable” mullite compositions were shifted toward higher alumina concentration. To account for the metastability, the authors suggested there could be a barrier for alumina precipitation in both melt and mullite, and that mullite could be superheated. Figure 3 portrays this phase diagram showing regions of metastability .
In 1987, Klug et al. published their SiO2-Al2O3 phase diagram . They reported incongruent melting for mullite at 1,890°C, and shifting of both boundaries of the mullite solid solution region toward higher alumina content (2:1 mullite) at temperatures above the eutectic point of 1,587°C. This phase diagram appears to reconcile most of the phenomena observed by other workers on the SiO2-Al2O3 system. Seemingly irreconcilable observations involving phase stability of similarly prepared specimens have been attributed convincingly to nonequilibrium conditions and/or silica volatilization. This phase diagram  is shown in Fig. 4.
Mole % АІ2О3
Fig. 3 The system Al2O3-SiO2 showing metastable regions. The gaps shown with spinodal regions are considered the most probable thermodynamically. From 
The 2:1 mullite appears to be only metastable at room temperature , and very high temperature use or cycling might cause some alumina to precipitate. However, Pask  suggested that discrepancies in the reported behavior of mullite are attributable to the presence or absence of a-Al2O3 in the starting materials. Engineers or scientists are cautioned to use the appropriate phase diagram consistent with their experimental methods and conditions. It should also be noted that at tectonic pressures, SiO2 will exsolve from mullite leaving a compound with a stoichiometry Al2O3*SiO2. Depending on temperature and pressure, the compound will be sillimanite, kyanite, or andalusite.