Computational design of ductile additively manufactured tungsten-based refractory alloys
Kareem Abdelmaqsoud, Daniel Sinclair, Venkata Satya Surya Amaranth Karra, S. Mohadeseh Taheri-Mousavi, Michael Widom, Bryan A. Webler, and John R. Kitchin, “Computational design of ductile additively manufactured tungsten-based refractory alloys”, Computational Materials Science 270, 114735 (2026). https://doi.org/10.1016/j.commatsci.2026.114735
This summary was written by a language model from the paper and has not been checked line by line. Rely on the published paper for details.
Question
Tungsten tolerates high temperatures and radiation, but it is brittle at room temperature, so parts made by additive manufacturing (AM) crack under the thermal stresses of printing. The authors argue that printability, rather than high-temperature strength, is the first property to optimize when designing tungsten-based refractory multi-principal element alloys. They ask whether a computed ductility indicator, the Pugh ratio of bulk to shear modulus, actually tracks cracking in printed alloys, and whether a machine-learned interatomic potential (MLIP) can stand in for DFT when screening a large composition space. The test case is the W–Ta–Nb system, where Nb is added to soften the alloy (Abdelmaqsoud et al. 2026).
Approach
Random BCC solid solutions were modeled with 10-atom special quasirandom structures (generated with ICET) for every W–Ta–Nb composition expressible in ten atoms; the supplement checks that 50-atom cells and ordered structures give similar trends. Structures were relaxed with the OMAT head of the UMA universal potential, chosen over MACE-OMAT because it reproduced the DFT ordering of pure-element ductility (Nb > Ta > W). Elastic tensors were fit with Pymatgen from normal strains of up to 1% and shear strains of up to 6%, and bulk and shear moduli came from Voigt–Reuss–Hill averaging. Melting points were estimated as CALPHAD solidus temperatures from Thermo-Calc (TCHEA8 database). A Pareto front of Pugh ratio against solidus temperature was built, with a 3300 K solidus floor set by a target service temperature near 1500 °C. Compositions on the front were then recomputed with VASP (PBE, 520 eV cutoff, finite-difference elastic constants) (Abdelmaqsoud et al. 2026).
For validation, blended W, Ta and Nb powders were printed by powder-blown directed energy deposition in an argon chamber below 10 ppm oxygen. Seven alloys with 20 to 80 at.% W and about 10 at.% Nb were built and six were analyzed (one deviated too far from its target composition), with compositions checked by EDS. Intergranular crack density (crack length per image area, mm/mm2) was measured in the homogenized top layer of each wall from stitched backscatter SEM images that were thresholded and skeletonized (Abdelmaqsoud et al. 2026).
Key findings
The UMA screening showed a clear trade-off: more W raises the solidus but lowers the Pugh ratio. Every Pareto-optimal composition above 3300 K contained about 10 at.% Nb, giving the series Wx Ta(90-x) Nb10. Along this series UMA predicted a Pugh ratio that falls steadily with W content, whereas DFT gave a non-monotonic trend that rises up to 40 to 50% W and then drops, in line with earlier reports on binary W–Ta. In the DFT results the bulk modulus grows linearly with W, but the averaged shear modulus plateaus below 50% W because the two BCC shear constants diverge there: C44 turns upward while the tetragonal shear C’ keeps falling (Abdelmaqsoud et al. 2026).
Since tungsten fails by cleavage tied to the tetragonal deformation path, the authors redefine the Pugh ratio as K/C’ and use it as the ductility descriptor. This ratio drops sharply between 50 and 60% W. The drop matches the electronic density of states at the Fermi level: above 50% W the Fermi level sits in a pseudogap (low DOS, less ductile), while at 50% W or below the pseudogap lies above it. A rigid-band analysis of B2 TaW attributes the jump in C’ to t2g bands, two nearly flat and one dispersive, whose filling makes C12 saturate near half filling so that C11 must grow faster (Abdelmaqsoud et al. 2026).
In the printed walls, lack of fusion, porosity and unmelted W and Ta particles dominated the macroscopic defects, especially in lower layers and higher-W alloys. Grain sizes were statistically similar except for one alloy. Intergranular microcracking was the main microscopic defect. Crack density rose with W content and tracked the inverse Pugh ratio C’/K, with the measurements lying between the ordered-structure and SQS predictions but closer to SQS. The two lowest-W alloys (nominally W20Ta70Nb10 and W30Ta60Nb10) showed no visible intergranular cracking (Abdelmaqsoud et al. 2026).
A small inconsistency appears in the extracted text: the laser speed is given as 500 mm/s in the methods and 300 mm/min in the results.
Significance
The work provides an experimental check, which is rarely done for alloys, that a DFT ductility descriptor ranks the cracking of additively manufactured refractory alloys. It also shows the limits of a universal MLIP: UMA was good enough to locate the Pareto-optimal region but missed the non-monotonic elastic trend that DFT captured, so the two were used in sequence. The authors suggest extending K/C’ screening to quaternary and quinary compositions and plan to strengthen the two crack-free alloys by removing processing defects and adding oxide dispersions (Abdelmaqsoud et al. 2026).