Why a Shape-Memory Alloy Has No Memory: The Geometry of Martensite — Epoche B2
Introduction: a misleading metaphor A bent wire of nickel–titanium straightens itself when heated, and the popular explanation is that the metal "remembers" the shape it was made in. This essay argues that the metaphor is wrong in an instructive way: nothing is stored and nothing is recalled. The recovery of shape is a direct geometric consequence of a phase transformation, and once we see the cause, the effect — and its limits — follow automatically. The test of the account is that it should predict not only that the wire straightens but the precise circumstances in which it will fail to. Cause: a transformation in which no atom changes neighbours A metal is a crystal: its atoms sit at the points of a pattern that repeats in three directions, the smallest repeating block being the unit cell . Above a critical temperature nickel–titanium exists as austenite , whose unit cell is a cube with one species at the corners and the other at the centre — a highly symmetric arrangement. On cooling it transforms into martensite , whose cell is monoclinic: the cube has been stretched along one axis, squashed along another and sheared so that one pair of faces is no longer perpendicular to the rest. Symmetry has been lost. The transformation is diffusionless , meaning that no atom migrates from one lattice site to another; every atom shifts by a fraction of a lattice spacing while keeping the same neighbours it had before. This is not a convenient idealisation but a consequence of rates. Diffusion in a solid proceeds by atoms hopping over an energy barrier, at a rate proportional to the Arrhenius factor $\exp(-Q/RT)$, where $Q$ is the activation energy for a hop, $R$ the gas constant and $T$ the absolute temperature; near room temperature this factor is so small that a hop is a rare event on a laboratory timescale. The martensitic front, by contrast, sweeps through a grain at a speed of the order of the speed of sound in the solid. There is simply no time for any atom to go anywhere, and no need for it to: the whole change is a coordinated shear. Two consequences follow immediately, and everything else in the essay rests on them. The composition of every region is unchanged, so no chemical bookkeeping is needed. And because neighbours are preserved, there is a fixed one-to-one correspondence between the sites of the martensite lattice and those of the austenite lattice it came from. Twelve variants: where the number comes from The distorted cell can be produced from the cube in more than one way, because the cube's symmetry offers several equivalent axes along which to stretch and shear. How many distinct orientations of martensite — variants — can grow inside a single crystal of austenite? The count is a piece of elementary symmetry arithmetic. The cube has $48$ symmetry operations: $24$ rotations that carry it onto itself, and each of those combined with inversion through the centre. The monoclinic martensite retains only $4$ of these — the identity, a two-fold rotation about its unique axis, a mirror plane perpendicular to that axis, and inversion. Each variant is thus left unchanged by $4$ of the parent's operations, and the remaining operations carry it to other variants. Distinct orientations therefore number $$\frac{48}{4} = 12.$$ The exponent-style reasoning here is worth spelling out: the denominator is $4$ and not $1$ precisely because the operations that martensite still possesses do not generate new orientations, so they must be divided out. Twelve is the number of lattice correspondence variants for this transformation, and it is the reason a cooled crystal is not a single distorted block but a fine mosaic of plates. The mosaic is what preserves the shape. Each individual variant carries a substantial shear, yet a cooled wire does not visibly deform. The reason is that the variants form in self-accommodating groups: because the twelve are generated from one another by the cube's own symmetry operations, for any variant that shears in one direction there is a partner, related by a lost symmetry operation, whose shear is the mirror image. Stacked as alternating thin plates in equal proportion — a twinned arrangement, in which the lattice on one side of a boundary is the mirror of the lattice on the other — their strains average to zero. The crystal transforms internally while remaining externally the same shape. Why the transformation lags its own equilibrium temperature Which phase is stable is settled by the Gibbs free energy $G = H - TS$, where $H$ is enthalpy (roughly, the bonding energy plus the pressure–volume term) and $S$ entropy (the disorder, including the vibrational freedom of the atoms). At fixed temperature and pressure a system moves towards lower $G$. Austenite, being more symmetric and vibrationally softer, has the higher entropy; martensite has the lower enthalpy. At high $T$ the term $-TS$ dominates and austenite wins; at low $T$ the enthalpy term wins. The two are equal when $$\Delta G = \Delta H - T\,\Delta S = 0 \quad\Rightarrow\quad T_0 = \frac{\Delta H}{\Delta S},$$ with $\Delta H$ and $\Delta S$ the enthalpy and entropy differences between austenite and martensite. In practice the transformation does not begin at $T_0$ but at a lower temperature, the martensite start temperature $M_s$, and the reason can be quantified. Growing a martensite plate costs energy that the free-energy comparison above ignores: new interfaces must be created, the surrounding austenite must be elastically strained to accommodate the plate, and the interfaces must be dragged against friction. Call the total of these non-chemical costs $\Delta G_{\text{extra}}$. Transformation begins only when the chemical driving force exceeds it. Near $T_0$ the driving force is approximately $\Delta S\,(T_0 - T)$, so the required undercooling is $$T_0 - M_s \approx \frac{\Delta G_{\text{extra}}}{\Delta S}.$$ The same argument run backwards explains the reverse transformation: on heating, the austenite fini