Martensitic Transformation: A Crystal Shears at the Speed of Sound

The quench: when diffusion runs out of time

Heat a plain carbon steel above about 800 °C and it becomes austenite — face-centred cubic γ-iron with carbon dissolved interstitially in its octahedral holes, up to 2.14 wt% at 1147 °C. Cool it slowly and the carbon walks out over micrometres to build the lamellar ferrite-plus-cementite of pearlite, because body-centred cubic α-iron holds only 0.022 wt% C. Cool it fast enough and that walk never happens.

The arithmetic is brutal. Carbon in ferrite has D ≈ 6×10⁻⁷ exp(−80 kJ mol⁻¹/RT) m² s⁻¹, about 1×10⁻¹³ m² s⁻¹ at 350 °C, and in austenite it is four orders of magnitude slower. Over the ~10⁻⁷ s a martensite plate takes to form, the diffusion length √(Dt) is ~0.1 nm even at the fast ferrite rate — less than one Fe–Fe bond, and far less in the austenite that actually has to move the carbon. The lattice cannot change its composition, so it changes its shape instead.

The onset is the martensite-start temperature, M_s. Andrews' 1965 linear fit remains the workshop standard: M_s(°C) ≈ 539 − 423(%C) − 30.4(%Mn) − 17.7(%Ni) − 12.1(%Cr) − 7.5(%Mo). A 0.4 wt% C steel gives ~370 °C on the carbon term alone and ~350 °C once its 0.7% Mn counts; AISI 4340 drops to ~305 °C, close to measurement. Alloying lowers M_s and pushes the diffusional pearlite and bainite reactions to longer times, so 4340 turns fully martensitic in oil near 10 °C s⁻¹ while plain 1040 needs a violent water quench at 10²–10³ °C s⁻¹. You never speed the martensite up; you only starve its competitors.

Bain's correspondence and the tetragonality that does the hardening

In 1924 Edgar C. Bain pointed out that FCC and BCC are not far apart if you pick the right cell. Draw a body-centred tetragonal cell inside two adjacent FCC cells: its base is a_γ/√2 and its height a_γ, so it already has a centred atom and needs only squashing. With a_γ ≈ 0.357 nm for a 0.4 wt% C austenite and martensite parameters a ≈ 0.286 nm, c ≈ 0.291 nm, the Bain strain is a contraction of ~18% along one ⟨100⟩_γ axis and ~13% expansion along the other two (texts quote 17–20% and 12–13%, since the figures track carbon content). The net volume change is ~3–4% positive — why steel grows as it hardens, why quench cracks form, and why a dilatometer sees M_s as a kink in a cooling curve.

Carbon makes the product tetragonal rather than cubic: the Bain strain maps one of the three sub-lattices of octahedral sites preferentially, so trapped carbon ends up in a single orientation, stretching that axis. Empirically c/a ≈ 1 + 0.045 × (wt% C), giving 1.018 at 0.4% C and 1.045 at 1.0%.

That distortion is the hardness. A carbon atom in the symmetric octahedral hole of FCC austenite makes an almost purely dilatational strain field that barely interacts with a screw dislocation; in the octahedral hole of the BCC lattice the same atom forces a tetragonal distortion whose shear field couples strongly to screws as well as edges — and the quench leaves ten to forty times more carbon sitting in those holes than equilibrium ferrite can hold. As-quenched hardness climbs from ~400 HV at 0.2% C to ~600 HV at 0.4% and ~850 HV near 0.8%, then saturates. Add a dislocation density of 10¹⁵–10¹⁶ m⁻² and the prior-austenite grain → packet → block → lath hierarchy. No carbide is involved. Carbides appear only on tempering, and as they pull carbon out of solution the tetragonality collapses toward c/a = 1 and the steel softens.

Keeping one plane undistorted: shape strain, habit planes and hidden twins

A homogeneous Bain strain would tear a plate out of its surroundings. Nature's solution is that the macroscopic shape change is an invariant-plane strain: one plane — the habit plane — is neither rotated nor distorted, so the plate fits its parent at finite strain energy. On it the shape change is a shear s ≈ 0.2 plus a normal dilatation of ~3%.

This is directly visible. Polish austenite flat and quench it: each plate tilts the free surface into surface relief, measurable by interferometry. Rule a scratch across the surface first and it is kinked where a plate crosses — deflected but continuous. That observation, exploited systematically by Alden Greninger and Alexander Troiano in their 1940s Fe–Ni–C shape-strain work, kills the alternatives: a fracture would sever the scratch, a diffusional reconstruction would leave it undisturbed. Their two-surface trace analysis also fixed the habit plane — near {111}_γ/{557}_γ for low-carbon lath martensite, {225}_γ at intermediate carbon, {259}_γ in high-carbon and Fe–Ni plate martensite — while the lattices obey near-rational orientation relationships: Kurdjumov–Sachs (1930), {111}_γ ∥ {110}_α′, and Nishiyama–Wassermann (1934–35).

The catch: the Bain strain alone leaves no plane undistorted. The plate must add a lattice-invariant shear — a deformation that changes shape without changing structure — by slip in lath martensite, or by internal twins spaced only a few nanometres apart in plate martensite, invisible until transmission electron microscopy in the 1950s and 60s. The phenomenological theory of martensite crystallography, from Wechsler, Lieberman and Read (1953) and independently Bowles and Mackenzie (1954), composes Bain strain, lattice-invariant shear and a rigid rotation into one matrix and predicts habit plane, orientation relationship, shape strain and twin spacing from lattice parameters alone. It worked before anyone had seen the twins it required.

Athermal kinetics: a third of the speed of sound

Diffusional transformations are thermally activated: hold the temperature and they proceed. Martensite is athermal — hold the temperature and it stops; drop it and more forms instantly. The amount depends on how cold you are, not how long you wait, which Koistinen and Marburger captured in 1959 as f = 1 − exp[−0.011(M_s − T)], temperatures in °C. An undercooling of 100 K converts ~67%; 210 K converts 90%. For a 1.0 wt% C steel Andrews puts M_s near 115 °C, so stopping at room temperature predicts a third of the austenite surviving; measured fractions run ~10–30%, an excellent engineering fit rather than a law. It is why 52100 bearing races and gauge blocks get a sub-zero soak in liquid nitrogen.

The speed is the headline. Rointan Bunshah and Robert Mehl measured it in 1953 in Fe–29.5 wt% Ni by watching resistivity pulses on an oscilloscope as plates swept past contacts: ~1100 m s⁻¹, roughly a third of steel's shear-wave velocity c_s = √(G/ρ) = √(80 GPa / 7870 kg m⁻³) ≈ 3200 m s⁻¹. A 100 µm plate therefore exists after ~10⁻⁷ s. Decisively, the velocity was independent of temperature from −20 °C to −196 °C: no activation energy, just an elastic instability propagating at a rate set by the lattice's own wave speeds. In Fe–Ni the plates arrive in autocatalytic bursts, each strain field triggering the next, and the specimen clicks audibly — the events an acoustic-emission sensor counts one by one.

Nucleation remains awkward. Classical homogeneous nucleation of a coherent plate carrying 0.2 shear predicts a barrier of order 10⁴ kT, which would never happen. The accepted picture, from Olson and Cohen (1976), is growth from pre-existing faulted embryos — dislocation arrays that dissociate into the product structure — which explains why finer austenite grains lower M_s, and why M_s sits ~200 K below the equal-free-energy temperature T₀, needing ~1100–1400 J mol⁻¹ of driving force.

How it is observed and measured

Quenching dilatometry is the workhorse: a Bähr DIL 805A or equivalent induction-heats a small rod, gas-quenches it at a programmed rate, and records length to a resolution of a few tens of nanometres. The 3–4% volume expansion appears as a sharp reversal in the cooling curve; that temperature is M_s, and families of such curves build the continuous-cooling-transformation diagrams heat treaters use.

X-ray diffraction reads the crystallography off directly: the cubic {200}_α reflection splits into a {200}/{002} doublet whose separation gives c/a, and comparing integrated γ and α′ peak intensities — the direct-comparison method of ASTM E975, or a full-pattern Rietveld refinement — quantifies retained austenite. Electron backscatter diffraction maps the 24 Kurdjumov–Sachs variants and reconstructs the prior austenite grains the transformation erased; transmission electron microscopy resolves the nanometre transformation twins the theory demanded.

Catching a ~10⁻⁷ s event in flight needs brighter sources. High-energy synchrotron diffraction at ESRF (ID11/ID15), DESY PETRA III (P07) and the Advanced Photon Source takes sub-millisecond patterns through a quench, and three-dimensional X-ray diffraction tracks individual plates inside bulk grains. Time-of-flight neutron diffraction at ISIS ENGIN-X and SNS VULCAN measures phase fractions and residual stresses, while inelastic neutron scattering hunts the soft transverse-acoustic phonon branches that precede the transformation in Ni–Al and NiTi. Two cheap tricks stay unbeaten in the workshop: austenite is paramagnetic and martensite ferromagnetic, so a Ferritescope reads the fraction in seconds; and a scratch on a polished surface still tells you, by kinking rather than breaking, that you are looking at a shear.

The thermoelastic branch: Nitinol, shape memory and ceramic steel

In steel the shape strain is accommodated plastically, the interface is wrecked by dislocation debris, and the change is irreversible. In some alloys the interface stays glissile and runs backwards. At the U.S. Naval Ordnance Laboratory in White Oak, Maryland, William Buehler and Frederick Wang found this in near-equiatomic nickel–titanium between 1959 and 1962 and named it Nitinol (NiTi + NOL). On cooling, cubic B2 austenite shears to monoclinic B19′ martensite, but the 24 variants form in self-accommodating groups whose strains cancel, so the object does not visibly change shape. Bend it cold and you do not deform it plastically — you detwin it, selecting the variant that suits the stress and storing up to ~8% strain in a still-perfect crystal. Heat past the austenite-finish temperature A_f and there is only one way back to B2: the original shape, recovered against restraint of up to ~500 MPa, with ~20–40 K of hysteresis. Transformation temperatures shift ~10 K per 0.1 at% Ni.

Superelasticity is the same transformation run isothermally. Above A_f, austenite is stable unloaded but stress pays the free-energy difference, following Clausius–Clapeyron at dσ/dT ≈ 6 MPa K⁻¹; the alloy yields on a flat plateau near 400–500 MPa and recovers 6–8% on unloading, thousands of times. That is the orthodontic archwire, the self-expanding vascular stent, the kink-proof eyeglass frame, and the Raychem CryoFit couplings that joined hydraulic lines on the Grumman F-14 Tomcat by being expanded in liquid nitrogen and allowed to shrink onto the pipe.

The same physics saves ceramics. Zirconia's tetragonal-to-monoclinic transformation is martensitic, with ~4% volume expansion. In Garvie, Hannink and Pascoe's 1975 Nature paper “Ceramic steel?” the material was calcia partially-stabilised zirconia, and today's yttria- and ceria-stabilised grades work the same way: metastable tetragonal grains are held until the tensile field at a crack tip triggers them, and the expansion clamps the crack shut — fracture toughness rises from ~3 to ~6–15 MPa m^1/2 depending on the stabiliser.

Look-alikes, failures and open questions

Martensite is routinely confused with the reactions above it on the cooling curve. Pearlite is purely diffusional: carbon travels micrometres, it takes seconds to minutes, it leaves no surface relief, and it is soft. Bainite is the hard case — it shows surface relief like martensite yet involves carbon partitioning and an incubation period like a nucleation-and-growth reaction. Whether its ferrite forms displacively and then rejects carbon (Bhadeshia's school) or grows by a ledgewise diffusional mechanism (Aaronson's) has been argued since the 1960s and is unsettled. Separate shape memory from superelasticity too: the first is thermally driven, the second stress-induced and isothermal, and above the temperature M_d slip beats stress-induced martensite so neither works.

The failures instruct. The 3–4% expansion arrives last in the constrained interior of a thick part, so quench cracking is cured by interrupted quenching (martempering), not by faster cooling. Untempered martensite is too brittle to use, so hardened parts are tempered: 150–200 °C precipitates transition ε-carbide, higher temperatures give cementite, and tetragonality is traded for toughness. In zirconia the toughening transformation can run away — moist body fluid at 37 °C nucleates monoclinic grains at the surface (low-temperature degradation), and hundreds of Prozyr femoral heads from late-1990s batches fractured in patients, prompting a 2001 recall.

What is unsolved: there is no first-principles theory of martensite nucleation, and the crystallographic theory that predicts plate martensite beautifully fits the {557}_γ lath martensite of structural steels only with multiple lattice-invariant shears. Whether soft phonon modes control the interface velocity or merely accompany the instability is open, and NiTi's functional fatigue — plasticity that drifts A_f over 10⁶ cycles — caps shape-memory actuators. The engineering has run ahead regardless: press-hardened 22MnB5 boron steel is quenched inside the forming die above its ~27 °C s⁻¹ critical rate to give martensitic 1500 MPa car B-pillars, and TRIP steels deliberately retain austenite so it transforms martensitically during a crash.

Martensite against the transformations it is most often confused with

Transformation	How atoms move	Kinetics	Diagnostic signature
Martensite (γ → α′ in Fe–C)	Cooperative shear; every atom moves < 1 atomic spacing; composition unchanged	Athermal — amount set by temperature, not time; interface at ~10³ m/s	Surface relief that kinks a scratch; BCT doublet in X-ray diffraction; c/a > 1
Bainite	Displacive ferrite plates plus carbon partitioning to carbides (mechanism still debated)	Isothermal, seconds to hours at 250–550 °C	Surface relief present, but carbides and an incubation time appear
Pearlite	Fully diffusional; carbon migrates over microns to build lamellae	Isothermal, nucleation-and-growth at 550–720 °C	Lamellar ferrite + cementite; no shape change, no surface relief
Thermoelastic B2 → B19′ (NiTi)	Same cooperative shear, but the interface stays mobile and reversible	Athermal on cooling, fully reversed on heating past Aᶠ	~20–40 K hysteresis loop; ~8% strain recovered on heating
Tetragonal → monoclinic ZrO₂	Martensitic shear in a ceramic, triggered by crack-tip stress	Athermal; stress-induced at ambient temperature	~4% volume expansion that closes cracks; toughness jumps to ~6–15 MPa·m¹ᐟ², stabiliser-dependent