AUSTIN-ELLIS/
austin ~/austin-ellis $ whoami
austin ~/austin-ellis $ cat INCAR
I use first-principles calculations, atomistic simulation, and machine learning to understand how bonding and reactivity shape materials. My projects span superconducting hydrides and electrides under high pressure, defects in 2D materials, and neural networks that predict electronic structure. At Oak Ridge National Laboratory, I’m building automated workflows that use machine-learned interatomic potentials to discover reaction pathways and model their kinetics.
Research and papers: six published, two preprints, five in progress, one pending
austin ~/austin-ellis $ sacct -X --format=JobID,JobName,Comment,Partition,State,End
| JobID | JobName | Comment | Partition | State | End |
|---|---|---|---|---|---|
| 2023_1 | csun-ms | tmdc-electrides | J. Phys. Chem. Lett. | COMPLETED | 2024 |
| 2023_2 | csun-ms | nakh12 | Adv. Funct. Mater. | COMPLETED | 2025 |
| 2023_3 | csun-ms | superhydrides | J. Am. Chem. Soc. | COMPLETED | 2025 |
| 2023_4 | csun-ms | fe-redox | Proc. Natl. Acad. Sci. | COMPLETED | 2025 |
| 2023_5 | csun-ms | h2-electride | J. Phys. Chem. Lett. | COMPLETED | 2026 |
| 2023_6 | csun-ms | elfnet | J. Mater. Inf. | COMPLETED | 2026 |
| 2023_7 | csun-ms | cs-core-reactivity | arXiv | COMPLETING | 2026 |
| 2023_8 | csun-ms | cs-polyoxides | draft | RUNNING | Unknown |
| 2026_1 | cnms-internship | polymerization | in_prep | RUNNING | Unknown |
| 2026_2 | cnms-internship | mlip-benchmark | draft | RUNNING | Unknown |
| 2026_3 | cnms-internship | wisp | ChemRxiv | COMPLETING | 2026 |
| 2026_4 | cnms-internship | oligomerization | in_prep | RUNNING | Unknown |
| 2026_5 | cnms-internship | tmd-vacancies | draft | RUNNING | Unknown |
| 2027 | phd | fall_2027 | PENDING | Unknown |
jobs/elfnet J. Mater. Inf. 2026 · NeurIPS AI4Mat 2025 · APS 2026
A. Ellis, S. Ellis, M. Miao
my roleFirst author. I designed, trained, and tested the network.
A periodic 3D convolutional network maps inexpensive superposed atomic densities to DFT electron localization functions, so bonding and interstitial localization can be screened before a full calculation.
jobs/superhydrides J. Am. Chem. Soc. 2025 · NeurIPS AI4Mat 2025
Y. Sun, A. Ellis, X. Chen, M. Miao
my role2nd of 4 authors. I helped build the training set and ran the template-guided CALYPSO searches and stability screening.
A template-guided machine-learning workflow found 13 new structural prototypes and 31 stable metal superhydrides, 19 of them with predicted Tc above 100 K.
jobs/h2-electride J. Phys. Chem. Lett. 2026
A. Ellis, S. Ellis, A. Pandit, M. Miao
my roleFirst author. I ran the PBE and HSE06 calculations and the analysis of the interstitial states.
As solid hydrogen metallizes under pressure, electrons localize in the interstitial regions between H2 molecules, an electride-like route that sets it apart from other molecular crystals.
jobs/tmdc-electrides J. Phys. Chem. Lett. 2024
Y. Sun, A. Ellis, S. Diaz, W. Li, M. Miao
my role2nd of 5 authors. I analyzed the vacancy states with ELF, charge-density differences, and band structures, and helped with the H-adsorption calculations.
Chalcogen vacancies strengthen the weak electride character of monolayer transition-metal dichalcogenides to the level of known electrides, and the vacancy states help catalyze hydrogen evolution.
jobs/nakh12 Adv. Funct. Mater. 2025
W. Zhao, A. Ellis, D. Duan, H. Wang, Q. Jiang, M. Du, T. Cui, M. Miao
my role2nd of 8 authors. I did the band-structure, density-of-states, crystal-orbital, and ELF analyses.
Interstitial electrons act as chemical templates that assemble H2 units in NaKH12, predicted to superconduct up to 245 K at a moderate 60 GPa.
jobs/fe-redox Proc. Natl. Acad. Sci. 2025
X. Wang, X. Feng, J. Li, Y. Lv, A. Ellis, S. Scott, A. Pandit, D. Khodagholian, R. J. Hemley, M. G. Jackson, F. J. Spera, S. A. T. Redfern, M. Miao
my role5th of 13 authors. I mapped phase stability across Fe–p-block compounds with CALYPSO searches and HSE06 convex hulls, and ran the Bader charge analysis.
Large-scale first-principles calculations show iron turning from electron donor to electron acceptor at core pressures, which changes the elements it can pull into Earth’s core.
jobs/wisp ChemRxiv 2026 · preprint
A. Ellis
my roleSole author.
A network that predicts the electron localization function of both spin channels at any point in a crystal, from the atomic structure and the magnetic moments on the atoms. On 21,922 held-out structures the mean absolute error is 0.011 per channel on the ELF scale of 0 to 1.
jobs/polymerization manuscript in preparation
A. Ellis, S. Bagchi, M. Goswami, J. Jakowski, J. Huang, P. Ganesh, I. Ivanov
my roleFirst author.
Compressed crystalline acetonitrile polymerizes through hydrogen transfers and C–C additions. ReactionFlow picks these steps out of molecular dynamics, and DFT checks their barriers.
jobs/mlip-benchmark manuscript · draft available upon request
A. Ellis, R. Morelock, S. Bagchi, A. M. dos Santos, M. Goswami, J. Jakowski, J. Huang, P. Ganesh, I. Ivanov
my roleFirst author.
Fifteen pretrained machine-learned interatomic potentials compared with DFT energies, forces, and barriers along 9,017 gas-phase reactions.
jobs/cs-core-reactivity arXiv 2026 · preprint
S. Ellis, A. Ellis, L. Wyke, C. Taylor, M. Miao
my role2nd of 5 authors.
High pressure can bring the Cs 5p core electrons into covalent bonding. Structure searches in the Cs–Ba–O system at 150 and 250 GPa put CsBa3O8 and CsBaO4 on the convex hull at 250 GPa, with mild Cs 5p–O 2p bonding and O2 units still intact. Cs2BaO3 bonds more strongly but is metastable.
jobs/tmd-vacancies manuscript · draft available upon request
J. Jeffries, A. Ellis, S. Ellis, E. Martinez, E. Zarkadoula
my role2nd of 5 authors.
A statistical prediction of how chalcogen vacancies form in alloyed transition-metal dichalcogenides, and of how their energies respond to stress.
jobs/oligomerization manuscript in preparation
I. N. Ivanov, A. Ellis, P. Christakopoulos, A. Puretsky, C. Rouleau, J. Huang, J. Jakowski, M. Goswami, P. Ganesh, B. Sumpter
my role2nd of 10 authors.
How the choice of stimulus directs acetonitrile toward oligomers or polymers.
jobs/cs-polyoxides manuscript · draft available upon request
Y. Sun, S. Ellis, D. Khodagholian, L. Wyke, A. Ellis, P. C. Müller, B. N. Schüpp, R. Nelson, R. Dronskowski, M. Miao
my role5th of 10 authors.
First-principles structure searches show the polyoxygen units of Cs–O compounds breaking down under pressure, as the Cs core electrons start to take part in bonding.
slurm-talks.out posters and talks
2026-10 poster, ORNL NEAT Science Symposium 2026-09 invited talk, NTI group meeting, CNMS 2026-08 talk, ORNL Neutron Scattering Division 2026-08 poster, CNMS User Meeting 2026-07 poster, ORNL NEAT initiative 2026-05 M.S. thesis defense, CSUN 2026-04 talk, CSUNposium 2026-03 talk, APS Global Physics Summit 2026-01 2 posters, Gordon Research Conf. 2025-12 2 posters, NeurIPS AI4Mat 2025 Sigma Xi talk; CSUN seminar 2023-06 talk, CECAM workshop, Lausanne
Fun visuals from some projects
2D and under pressure
austin ~/austin-ellis $ VESTA structures/electrides/MoS2_vacancy/ELFCAR_441 &
scf 1.00000000000000 12.589023 -0.000000 0.000000 -6.294512 10.902414 0.000000 0.000000 0.000000 24.942065 Mo S 16 31 Direct 0.084274 0.167635 0.500859 0.086214 0.418107 0.499328 0.084274 0.666639 0.500859 0.083333 0.916667 0.501453 0.333361 0.167635 0.500859 0.336095 0.422190 0.497662 0.336095 0.663905 0.497662 0.333361 0.915726 0.500859 0.582838 0.166419 0.500406 0.581893 0.418107 0.499328 0.577810 0.663905 0.497662 0.581893 0.913786 0.499328 … 35 more positions 108 108 200 (2,332,800 values follow) # VESTA 3.5.8 scene mos2_vacancy.vesta # isosurface ELF = 0.60 # sections ELF 0 → 1, B-G-R # region hexagonal flake, 6 Å apothem, # centred on the vacancy
C2c12_300GPa_PBE_fatband 1.00000000000000 4.939000 0.000000 0.000000 0.000000 2.811000 0.000000 -3.282370 0.000000 2.521383 H 24 Direct 0.226000 0.067200 0.246400 0.726000 0.567200 0.246400 0.774000 0.067200 0.253600 0.274000 0.567200 0.253600 0.774000 0.932800 0.753600 0.274000 0.432800 0.753600 0.226000 0.932800 0.746400 0.726000 0.432800 0.746400 0.344300 0.195800 0.222700 0.844300 0.695800 0.222700 0.655700 0.195800 0.277300 0.155700 0.695800 0.277300 … 12 more positions 64 36 56 (129,024 values follow) # VESTA 3.5.8 scene h2_c2c_b15.vesta # field band 15 at Γ, scaled to its cell mean # isosurface ρ = 2⟨ρ⟩ # sections 0 → 3⟨ρ⟩, B-G-R # region 2 × 3 × 2 of (a+c, b, c)
scf 1.00000000000000 4.796022 0.000000 0.000000 0.000000 4.679341 0.000000 0.000000 0.000000 2.693860 H 48 Direct 0.998954 0.269437 0.317259 0.998954 0.769437 0.817259 0.001046 0.730563 0.317259 0.001046 0.230563 0.817259 0.501046 0.269437 0.317259 0.501046 0.769437 0.817259 0.498954 0.730563 0.317259 0.498954 0.230563 0.817259 0.016047 0.371559 0.948133 0.016047 0.871559 0.448133 0.983953 0.628441 0.948133 0.983953 0.128441 0.448133 … 36 more positions 100 96 56 (537,600 values follow) # VESTA 3.5.8 scene h2_ama2_b25.vesta # field band 25 at R, scaled to its cell mean # isosurface ρ = 2⟨ρ⟩ # sections 0 → 3⟨ρ⟩, B-G-R # region 1 × 1 × 2 cells
P63m_325GPa_PBE_A_parchg 1.00000000000000 3.220273 0.000001 -0.000011 -1.610136 2.788838 0.000002 -0.000008 -0.000003 2.523312 H 16 Direct 0.080285 0.398827 0.249964 0.681460 0.080286 0.749999 0.601172 0.681459 0.250034 0.919715 0.601173 0.750036 0.318540 0.919714 0.250001 0.398828 0.318541 0.749966 0.195092 0.253717 0.250030 0.941376 0.195091 0.750013 0.746283 0.941375 0.249983 0.804908 0.746283 0.749970 0.058624 0.804909 0.249987 0.253717 0.058625 0.750017 … 4 more positions 48 48 32 (73,728 values follow) # VESTA 3.5.8 scene h2_p63m_b9.vesta # field band 9 at A, scaled to its cell mean # isosurface ρ = 2⟨ρ⟩ # sections 0 → 3⟨ρ⟩, B-G-R # region 2 × 2 × 3 cells
P63m_325GPa_PBE_A_parchg 1.00000000000000 3.220273 0.000001 -0.000011 -1.610136 2.788838 0.000002 -0.000008 -0.000003 2.523312 H 16 Direct 0.080285 0.398827 0.249964 0.681460 0.080286 0.749999 0.601172 0.681459 0.250034 0.919715 0.601173 0.750036 0.318540 0.919714 0.250001 0.398828 0.318541 0.749966 0.195092 0.253717 0.250030 0.941376 0.195091 0.750013 0.746283 0.941375 0.249983 0.804908 0.746283 0.749970 0.058624 0.804909 0.249987 0.253717 0.058625 0.750017 … 4 more positions 48 48 32 (73,728 values follow) # VESTA 3.5.8 scene h2_p63m_b11.vesta # field band 11 at A, scaled to its cell mean # isosurface ρ = 1.5⟨ρ⟩ # sections 0 → 3⟨ρ⟩, B-G-R # region 2 × 2 × 3 cells
S Ta 1.00000000000000 13.399793 0.000000 0.000000 -6.699897 11.604561 0.000000 0.000000 0.000000 83.190000 S Ta 31 16 Direct 0.417104 0.085042 0.481196 0.666492 0.082983 0.481750 0.917939 0.085042 0.481196 0.164958 0.332061 0.481196 0.417104 0.332061 0.481196 0.664389 0.332194 0.481347 0.917806 0.332194 0.481347 0.167017 0.583508 0.481750 0.417691 0.585382 0.481040 0.666492 0.583508 0.481750 0.917806 0.585611 0.481347 0.164958 0.832896 0.481196 … 35 more positions # region 4 × 4 supercell, one sheet # vacancy one S, lower layer # view from below
A MoS2 monolayer with one sulfur removed at the center of the flake. Every S carries a lone-pair ELF basin; at the vacancy a smaller basin with no atom inside sits where the sulfur was: the electride-like vacancy state.
C2/c-12 solid hydrogen, band 15 at Γ (Fig. 3a of the paper). The lowest conduction state runs in continuous channels between the H2 molecules: electrons moving into the interstitial space as the gap closes.
Ama2 hydrogen at 425 GPa, band 25 at R (Fig. 3b). Compression pushes this state out of the H–H bonds and into the interstitial space between the molecular layers.
P63/m hydrogen, band 9 at A (Fig. 3c): interstitial maxima stacked in columns along c, between the H2 molecules.
P63/m hydrogen, band 11 at A (Fig. 3d): the next state up also gathers in the interstitial pockets instead of on the H–H bonds.
A TaS2 sheet with one sulfur missing from the lower layer, seen from below. The sulfur above the gap has sunk toward the tantalum plane.
Chemical templates for superhydrides
austin ~/austin-ellis $ VESTA structures/superhydrides/NaKH12_60GPa/ELFCAR &
scf 1.00000000000000 6.083404 0.000000 0.000000 0.000000 6.083404 0.000000 0.000000 0.000000 6.083404 H K Na 48 4 4 Direct 0.429349 0.750000 0.750000 0.250000 0.250000 0.929349 0.070651 0.250000 0.750000 0.250000 0.750000 0.570651 0.250000 0.929349 0.750000 0.250000 0.250000 0.570651 0.250000 0.070651 0.750000 0.250000 0.750000 0.929349 0.070651 0.750000 0.750000 0.250000 0.570651 0.750000 0.429349 0.250000 0.750000 0.250000 0.429349 0.750000 … 44 more positions 54 54 54 (157,464 values follow) # VESTA 3.5.8 scene nakh12_60GPa.vesta # isosurface ELF = 0.75 # sections ELF 0 → 1, B-G-R # region 1 × 1 × 1 cell # radii Na 1.25 Å, K 1.50 Å
scf 1.00000000000000 6.082837 0.000000 0.000000 0.000000 6.082837 0.000000 0.000000 0.000000 6.082837 K Na 4 4 Direct 0.500000 0.000000 0.000000 0.500000 0.500000 0.500000 0.000000 0.000000 0.500000 0.000000 0.500000 0.000000 0.000000 0.000000 0.000000 0.000000 0.500000 0.500000 0.500000 0.000000 0.500000 0.500000 0.500000 0.000000 54 54 54 (157,464 values follow) # VESTA 3.5.8 scene nak_sublattice.vesta # isosurface ELF = 0.55 # sections ELF 0 → 1, B-G-R # region 1 × 1 × 1 cell # radii Na 1.25 Å, K 1.50 Å
Pa10H84_I4m_200GPa 1.00000000000000 7.3252971927448902 -0.0000000002934098 0.0000000000000000 0.0000000002934098 7.3252971927448893 -0.0000000000000000 -0.0000000000000000 0.0000000000000000 4.6482720658846830 H U 84 10 Direct 0.6043816623181978 0.5527704859944376 0.1315448655588674 0.3956183376818020 0.4472294830055636 0.8684551714411322 0.3956183376818020 0.4472294830055636 0.1315448655588674 0.6043816623181978 0.5527704859944376 0.8684551714411322 0.4472294830055636 0.6043816623181978 0.1315448655588674 0.5527704859944376 0.3956183376818020 0.8684551714411322 0.5527704859944376 0.3956183376818020 0.1315448655588674 0.4472294830055636 0.6043816623181978 0.8684551714411322 0.1043816623181980 0.0527704459944340 0.6315448785588649 0.8956183066818032 0.9472294830055634 0.3684551464411336 0.8956183066818032 0.9472294830055634 0.6315448785588649 0.1043816623181980 0.0527704459944340 0.3684551464411336 … 82 more positions # VESTA 3.5.8 scene u10h84_structure.vesta # model ball-and-stick # bonds H–H ≤ 1.35 Å # region 1 × 1 × 2 cells
scf 1.00000000000000 7.487900 0.000000 0.000000 0.000000 7.487900 0.000000 0.000000 0.000000 4.680100 U 10 Direct 0.400540 0.200430 0.000000 0.900540 0.700430 0.500000 0.799570 0.400540 0.000000 0.299570 0.900540 0.500000 0.599460 0.799570 0.000000 0.099460 0.299570 0.500000 0.200430 0.599460 0.000000 0.700430 0.099460 0.500000 0.000000 0.000000 0.000000 0.500000 0.500000 0.500000 60 60 40 (144,000 values follow) # VESTA 3.5.8 scene u10_template.vesta # isosurface ELF = 0.50 # sections ELF 0 → 1, B-G-R # region 1 × 1 × 2 cells
NaKH12 at 60 GPa. Yellow ELF isosurfaces wrap each H2 unit; six of them sit on the faces of every small Na4K4 cube, bonds pointing at its center.
The same cell with the hydrogen taken out. An interstitial ELF basin sits at the center of each small Na4K4 cube: the site the six H2 units gather around in NaKH12.
U10H84 (I4/m, 200 GPa), one of the superhydrides from the template-guided search, with its hydrogen cage network.
The bare U10 metal lattice, step one of the two-step search. At ELF 0.50 the isosurface shows the interstitial basins between the U atoms as well as the shell around each atom. Same view and scale as U10H84.
ELF predicted by ELFNet and WISP
austin ~/austin-ellis $ VESTA structures/ml/CaH6/ELFCAR_Ca_ELFNet &
CaH6 sites on the Ca template 1.00000000000000 3.403440 0.000000 0.000000 -0.000000 3.403440 0.000000 0.000000 0.000000 3.403440 Ca H 2 12 Direct 0.000000 0.000000 0.000000 0.500000 0.500000 0.500000 0.250000 0.500000 0.000000 0.750000 0.500000 0.000000 0.500000 0.250000 0.000000 0.500000 0.750000 0.000000 0.000000 0.250000 0.500000 0.000000 0.750000 0.500000 0.000000 0.500000 0.250000 0.000000 0.500000 0.750000 0.250000 0.000000 0.500000 0.750000 0.000000 0.500000 0.500000 0.000000 0.250000 0.500000 0.000000 0.750000 32 32 32 (32,768 values follow) # field ELF predicted by ELFNet # (no hydrogen included) # isosurface ELF = 0.45 # sections ELF 0 → 0.71, B-G-R # region 1 cell
CaH6 sites on the Ca template 1.00000000000000 3.403440 0.000000 0.000000 -0.000000 3.403440 0.000000 0.000000 0.000000 3.403440 Ca H 2 12 Direct 0.000000 0.000000 0.000000 0.500000 0.500000 0.500000 0.250000 0.500000 0.000000 0.750000 0.500000 0.000000 0.500000 0.250000 0.000000 0.500000 0.750000 0.000000 0.000000 0.250000 0.500000 0.000000 0.750000 0.500000 0.000000 0.500000 0.250000 0.000000 0.500000 0.750000 0.250000 0.000000 0.500000 0.750000 0.000000 0.500000 0.500000 0.000000 0.250000 0.500000 0.000000 0.750000 28 28 28 (21,952 values follow) # field ELF from DFT # (no hydrogen included) # isosurface ELF = 0.45 # sections ELF 0 → 0.71, B-G-R # region 1 cell
CaH6 sites on the Ca template 1.00000000000000 3.403440 0.000000 0.000000 -0.000000 3.403440 0.000000 0.000000 0.000000 3.403440 Ca H 2 12 Direct 0.000000 0.000000 0.000000 0.500000 0.500000 0.500000 0.250000 0.500000 0.000000 0.750000 0.500000 0.000000 0.500000 0.250000 0.000000 0.500000 0.750000 0.000000 0.000000 0.250000 0.500000 0.000000 0.750000 0.500000 0.000000 0.500000 0.250000 0.000000 0.500000 0.750000 0.250000 0.000000 0.500000 0.750000 0.000000 0.500000 0.500000 0.000000 0.250000 0.500000 0.000000 0.750000 # region 1 cell # bonds H–H up to 1.3 Å
MnPd mp-1902 1.00000000000000 3.110125 0.000000 0.000000 0.000000 3.110125 0.000000 0.000000 0.000000 3.110125 Mn Pd 1 1 Direct 0.000000 0.000000 0.000000 0.500000 0.500000 0.500000 28 28 28 × 2 spin channels (21,952 values follow) # field spin-down ELF, WISP # moments Mn 3.95, Pd 0.31 μB # isosurface ELF = 0.30 # sections ELF 0 → 0.45, B-G-R # region half a cell, cut at b = 0.5
MnPd mp-1902 1.00000000000000 3.110125 0.000000 0.000000 0.000000 3.110125 0.000000 0.000000 0.000000 3.110125 Mn Pd 1 1 Direct 0.000000 0.000000 0.000000 0.500000 0.500000 0.500000 28 28 28 × 2 spin channels (21,952 values follow) # field spin-up ELF, WISP # moments Mn 3.95, Pd 0.31 μB # isosurface ELF = 0.30 # sections ELF 0 → 0.45, B-G-R # region half a cell, cut at b = 0.5
MnPd mp-1902 1.00000000000000 3.110125 0.000000 0.000000 0.000000 3.110125 0.000000 0.000000 0.000000 3.110125 Mn Pd 1 1 Direct 0.000000 0.000000 0.000000 0.500000 0.500000 0.500000 28 28 28 × 2 spin channels (21,952 values follow) # field spin-down ELF, DFT # moments Mn 3.95, Pd 0.31 μB # isosurface ELF = 0.30 # sections ELF 0 → 0.45, B-G-R # region half a cell, cut at b = 0.5
MnPd mp-1902 1.00000000000000 3.110125 0.000000 0.000000 0.000000 3.110125 0.000000 0.000000 0.000000 3.110125 Mn Pd 1 1 Direct 0.000000 0.000000 0.000000 0.500000 0.500000 0.500000 28 28 28 × 2 spin channels (21,952 values follow) # field spin-up ELF, DFT # moments Mn 3.95, Pd 0.31 μB # isosurface ELF = 0.30 # sections ELF 0 → 0.45, B-G-R # region half a cell, cut at b = 0.5
Sc5Cl8 mp-542449 1.00000000000000 3.507162 0.001333 0.665304 1.651030 9.327444 0.887435 -0.030616 -0.576538 9.716608 Sc Cl 5 8 Direct 0.000000 0.000000 0.000000 0.512874 0.321990 0.653712 0.487126 0.678010 0.346288 0.117578 0.388273 0.376595 0.882422 0.611727 0.623405 0.691594 0.189231 0.428658 0.308406 0.810769 0.571342 0.367242 0.152401 0.112899 0.632758 0.847599 0.887101 0.671541 0.465677 0.190068 0.328459 0.534323 0.809932 0.974803 0.831954 0.216411 … 1 more positions 30 84 84 × 2 spin channels (211,680 values follow) # field channel-averaged ELF, WISP # isosurface ELF = 0.70 # sections ELF 0 → 1, B-G-R # region half a cell, cut at a = 0.5 # moments predicted from the geometry
Sc5Cl8 mp-542449 1.00000000000000 3.507162 0.001333 0.665304 1.651030 9.327444 0.887435 -0.030616 -0.576538 9.716608 Sc Cl 5 8 Direct 0.000000 0.000000 0.000000 0.512874 0.321990 0.653712 0.487126 0.678010 0.346288 0.117578 0.388273 0.376595 0.882422 0.611727 0.623405 0.691594 0.189231 0.428658 0.308406 0.810769 0.571342 0.367242 0.152401 0.112899 0.632758 0.847599 0.887101 0.671541 0.465677 0.190068 0.328459 0.534323 0.809932 0.974803 0.831954 0.216411 … 1 more positions 30 84 84 × 2 spin channels (211,680 values follow) # field channel-averaged ELF, DFT # isosurface ELF = 0.70 # sections ELF 0 → 1, B-G-R # region half a cell, cut at a = 0.5
The calcium lattice of CaH6, with the hydrogen left out of the calculation. ELFNet predicts this ELF from superposed atomic densities, and the yellow surface encloses the sites where the hydrogens sit in CaH6, drawn as small spheres.
The DFT ELF of the same lattice, for comparison. It is highest at the hydrogen sites.
The full CaH6 structure, a superhydride. Its hydrogens form a cage of squares and hexagons around each calcium.
MnPd, spin-down channel, predicted by WISP from the structure and the magnetic moments on the atoms. At ELF 0.30 a large region surrounds the Pd atom.
The spin-up channel from WISP at the same isovalue. The large region around Pd is absent, and that difference between the two channels is what WISP has to get right.
The spin-down channel from DFT, for comparison (Fig. 2b of the preprint).
The spin-up channel from DFT, for comparison (Fig. 2b of the preprint).
Sc5Cl8, a known electride, from WISP with only the geometry as input. Besides the shells around the atoms, the ELF reaches 0.83 at the center of the cut face, 2.2 Å from the nearest atom: the interstitial maximum of the electride.
The same view from DFT, for comparison (Fig. 5 of the preprint). The interstitial maximum is 0.83 here as well.
Iron and cesium compounds
austin ~/austin-ellis $ VESTA structures/high-pressure/Fe-I/FeI2_0GPa.vasp &
Fe1 I2 1.00000000000000 4.016759 0.000000 -0.000000 -2.008380 3.478616 0.000000 -0.000000 -0.000000 7.444104 Fe I 1 2 Direct -0.000000 0.000000 -0.000000 0.333333 0.666667 0.216454 0.666667 0.333333 0.783546 # region 3 × 3 × 2 cells # bonds Fe–I up to 3.0 Å
1:Pm-3m(221) 1.00000000000000 3.157641 -0.000000 -0.000000 -0.000000 3.157641 0.000000 0.000000 0.000000 3.157641 Fe I 3 1 Direct -0.000000 0.500000 0.500000 0.500000 -0.000000 0.500000 0.500000 0.500000 0.000000 0.000000 -0.000000 0.000000 # region 2 × 2 × 2 cells
CONTCAR_290 1.00000000000000 4.160187 0.000000 0.000000 -2.080094 3.602828 0.000000 -0.000000 0.000000 5.998209 Cs O 4 4 Direct 0.666667 0.333333 0.750000 0.333333 0.666667 0.250000 0.000000 -0.000000 -0.000000 -0.000000 -0.000000 0.500000 0.333333 0.666667 0.867567 0.666667 0.333333 0.132433 0.666667 0.333333 0.367567 0.333333 0.666667 0.632433 # region 2 × 2 × 2 cells # bonds O–O up to 1.7 Å
8:R-3m(166) 1.00000000000000 2.011168 -1.161147 2.744058 0.000000 2.322295 2.744059 -2.011168 -1.161147 2.744058 Cs O 2 2 Direct -0.000000 -0.000000 0.000000 0.500000 0.500000 0.500000 0.234438 0.234438 0.234438 0.765562 0.765562 0.765562 # drawn hexagonal setting, 2 × 2 × 1 # bonds Cs–O up to 2.05 Å
FeI2 at ambient pressure, from the set of iron and p-block structures behind the paper: sheets of iron between layers of iodine.
Fe3I at 300 GPa, from the same set: a cubic structure with iodine at the corners of the cell and iron on its faces.
CsO at 150 GPa. The oxygens sit in pairs, 1.41 Å apart.
CsO at 250 GPa, past the transition near 221 GPa. The pairs have come apart, and each oxygen now sits 1.93 Å from a cesium.
Acetonitrile under pressure and on sodium
austin ~/austin-ellis $ VESTA structures/reactions/ACN_35GPa/CONTCAR &
C H N 1.00000000000000 9.339321 0.000000 0.000000 0.054916 13.092645 0.000000 -0.019215 0.053993 12.778178 C H N 96 144 48 Direct 0.001850 0.009301 0.063695 0.250287 0.176067 0.063657 0.000183 0.324904 0.312278 0.250795 0.158120 0.313322 0.001832 0.008200 0.563036 0.250508 0.175851 0.564262 0.998933 0.328145 0.813804 0.250496 0.158689 0.814234 0.998792 0.345323 0.062493 0.247266 0.514329 0.059596 0.999352 0.659880 0.315665 0.249767 0.491317 0.311827 … 276 more positions # region 1 cell, molecules kept whole # full color the three linked molecules # faded unreacted CH3CN
40 4-amino-2,6-dimethylpyrimidine on Na(100) C -1.2490 2.1950 1.3830 C -0.6260 1.0210 0.6680 C -1.1200 -0.2800 0.8070 C -0.4510 -1.3070 0.1130 C 2.2090 0.5310 -1.6500 C 1.0210 0.2480 -0.7560 N 0.4510 1.2920 -0.1240 N -0.8530 -2.6490 0.2390 N 0.6180 -1.0530 -0.6810 H -2.0660 1.8790 2.0480 H -1.6380 2.9230 0.6510 H -0.4810 2.7180 1.9760 … 28 more atoms # surface Na(100) slab, 22 Na atoms
40 2,4,6-trimethyl-1,3,5-triazine on Na(100) C -2.3020 -0.5810 1.5480 C -1.0850 -0.3060 0.7010 C 2.5710 -0.5810 1.0370 C 1.1980 -0.3300 0.4580 C -0.2500 1.3490 -2.4870 C -0.1240 0.5700 -1.1980 N -1.2640 0.2860 -0.5120 N 0.1230 -0.6610 1.2200 N 1.1320 0.2540 -0.7690 H -3.1890 -0.0820 1.1370 H -2.4940 -1.6720 1.6110 H -2.1240 -0.2520 2.5840 … 28 more atoms # surface Na(100) slab, 22 Na atoms
Crystalline acetonitrile at 35 GPa, 48 molecules in the cell. Three of them have linked into one unit, drawn in full color; the rest are faded.
4-Amino-2,6-dimethylpyrimidine on a sodium (100) surface: three acetonitrile molecules joined into one ring. This is the product seen in experiment.
The triazine, the other ring that three acetonitrile molecules can close into, on the same surface.
Our ELFNet paper was selected for the front cover of the Journal of Materials Informatics, Volume 6, Issue 3.
Took part in the Open Scientific Intelligence (OSI) Hackathon with an ORNL team. We built Electride Detectives, a swarm of LLM agents that reason about what makes a material an electride.
Published my first-author ELFNet paper, Prediction of Electron Localization Functions from Superposed Atomic Densities for Accelerated Superhydride Discovery, in the Journal of Materials Informatics.
Attended my first CNMS User Meeting at Oak Ridge and presented a poster on my internship work.
Published my first-author paper, Emerging Electride Behavior and Metallization in Molecular Hydrogen under High Pressure, in The Journal of Physical Chemistry Letters.
Started a research internship at Oak Ridge National Laboratory’s Center for Nanophase Materials Sciences.
Defended my M.S. thesis, From New Chemistry to New Materials: Computational and Data-Driven Discovery, at CSU Northridge.
Talk at the APS Global Physics Summit in Denver on predicting electron localization functions with a 3D CNN, supported by a DCOMP travel award.
Led the semester’s first Chemistry and Biochemistry Journal Club meeting, on Metallic Oxides and the Overlooked Role of Bandwidth.
Two posters at the Gordon Research Conference on Multifunctional Materials and Structures, Ventura.
NeurIPS AI4Mat 2025, San Diego: two posters and a travel grant.
CSUN Newsroom featured our work on the pressure-driven redox reversal of iron in Earth’s core.
About: the structure to continue from
I’m preparing to apply to a Ph.D. program. I’m broadly interested in quantum and energy-relevant materials, condensed-phase chemical systems, and structure–property relationships across inorganic and materials chemistry.