Transitory phase transformations during { 10 1 ¯ 2 } twinning in titanium

$\left(10\overline{1}2\right)\left(10\overline{1}\overline{1}\right)$ twinning presents a variety of properties that entirely depart from the classical twinning behavior. In this work, deformation of a single crystal Ti under uniaxial compression perpendicular to the c-axis was investigated by using atomistic simulations. Very interesting phase transformations between body centered cubic (BCC), hexagonal close-packed (HCP), face centered cubic (FCC) structures were observed. These phase transformations are only transitory. Before twin nucleation, $HC{P}_{parent}\to BCC$ transformation occurred. The lattice transformation can be described as: ${\left(10\overline{1}0\right)}_P\to {\left(110\right)}_{BCC}$, and ${\left(0002\right)}_P\to {\left(1\overline{1}0\right)}_{BCC}$. Then the BCC phase transformed to HCP which satisfies $\left(10\overline{1}2\right)\left(10\overline{1}\overline{1}\right)$ twin relationship with the HCP parent. This transformation is accomplished by atomic shuffles and no shear is involved. After the BCC phase was entirely transformed, the deformation was dominated by twin growth. At the twin boundary (TB), a FCC phase was present, separating the parent from the twin. The FCC structure was transformed from the HCP parent. This transformation is also transitory and involves both shear and shuffle. Shuffling transforms the double-layered ${\left(10\overline{1}0\right)}_P$ plane of parent into a close-packed ${\left(111\right)}_{FCC}$ plane, and then a shear with a Burgers vector of half a Shockley partial dislocation on the ${\left(111\right)}_{FCC}$ plane generates the correct FCC structure. It was observed that the formation of FCC phase at the TB hinders twin growth. But as the strain increases, the FCC phase transformed to the HCP twin and this process involves Shockley partials on the interface. Strain accommodation was analyzed to explain the formation of the FCC phase at the TB. These transitory phase transformations during deformation twinning further corroborate that $\left(10\overline{1}2\right)\left(10\overline{1}\overline{1}\right)$ twinning completely departs from the classical twinning behavior.