Fractional crystallization in silicate melts (magmas) is complex compared to crystallization in chemical systems at constant pressure and composition, because changes in pressure and composition can have dramatic effects on magma evolution. Addition and loss of water, carbon dioxide, hydrogen, and oxygen are among the compositional changes that must be considered. For example, the partial pressure (fugacity) of water in silicate melts can be of prime importance, as in near-solidus crystallization of magmas of granite composition. The crystallization sequence of oxide minerals such as magnetite and ulvospinel is sensitive to the oxygen fugacity of melts, and separation of the oxide phases can be an important control of silica concentration in the evolving magma, and may be important in andesite genesis.
Experiments have provided many examples of the complexities that control which mineral is crystallized first as the melt cools down past the liquidus.
One example concerns crystallization of melts that crystallize to mafic and ultramafic rocks. MgO and SiO2 concentrations in melts are among the variables that determine whether forsterite olivine or enstatite pyroxene is precipitated, but the water content and pressure are also important. In some compositions, at high pressures without water crystallization of enstatite is favored, but in the presence of water at high pressures, olivine is favored.
Granitic magmas provide additional examples of how melts of generally similar composition and temperature but at different pressure may crystallize different minerals. Pressure determines the maximum water content of a magma of granite composition. High-temperature fractional crystallization of relatively water-poor granite magmas may produce single-alkali-feldspar granite, and lower-temperature crystallization of relatively water-rich magma may produce two-feldspar granite.
During the process of fractional crystallization, melts become enriched in incompatible elements. Hence, knowledge of the crystallization sequence is critical in understanding how melt compositions evolve. Textures of rocks provide insights, as documented in the early 1900's by Bowen's reaction series. Experimentally-determined phase diagrams for simple mixtures provide insights into general principles. Numerical calculations with software like that referenced below have become increasing able to simulate natural processes accurately.