Experimental validation of an LLM-hypothesized Fe-Mn-Al-Ni-Si-C alloy: phase stability governs the absence of super-elasticity

Iron-based shape memory alloys (Fe-SMAs) offer a low-cost alternative to Nitinol, but their super-elastic response is highly sensitive to composition and processing. LLM deep-research agents can now propose new alloy compositions in seconds, raising the question of whether such proposals survive contact with a furnace. We report what is, to our knowledge, the first experimental test of an LLM-hypothesized shape memory alloy carried from proposal through processing to mechanical and synchrotron characterization. The LLM-alloy achieved good conventional mechanical properties (0.2% proof stress 1908–519 MPa across the anneal series, elongation peaking at 33%) but showed no super-elasticity under any condition tested: strain recovery on unloading was indistinguishable from elastic springback, against ≈0.5% recoverable transformation strain in the benchmark alloy. Synchrotron diffraction attributes this to a phase-stability mismatch – a stable FCC γ + BCC α duplex microstructure with a minor ordered D03 Fe3Al phase, unchanged by deformation. Equilibrium CALPHAD calculations across three independent databases, validated against the benchmark, locate the cause: the benchmark solution-treats to single-phase BCC α at 1200 ◦ C, whereas the LLM-alloy is two-phase there and reaches single-phase α only at ≈1340 ◦ C, above its accessible processing range; deleting carbon alone from the measured chemistry lowers that field to ≈1150 ◦ C and restores the single-phase state. A carbon addition of ≈0.10 wt% – ten times the benchmark’s 0.010 wt% – is thus sufficient to close the processing window super-elasticity depends on. LLM-proposed compositions should therefore be screened for competing phase stability and interstitial content before synthesis – a check costing minutes with openly available databases.