Advancing Rare-Earth Separation by Machine Learning

Abstract

Constituting the bulk of rare-earth elements, lanthanides need to be separated to fully realize their potential as critical materials in many important technologies. The discovery of new ligands for improving rare-earth separations by solvent extraction, the most practical rare-earth separation process, is still largely based on trial and error, a low-throughput and inefficient approach. A predictive model that allows high-throughput screening of ligands is needed to identify suitable ligands to achieve enhanced separation performance. Here, we show that deep neural networks, trained on the available experimental data, can be used to predict accurate distribution coefficients for solvent extraction of lanthanide ions, thereby opening the door to high-throughput screening of ligands for rare-earth separations. One innovative approach that we employed is a combined representation of ligands with both molecular physicochemical descriptors and atomic extended-connectivity fingerprints, which greatly boosts the accuracy of the trained model. More importantly, we synthesized four new ligands and found that the predicted distribution coefficients from our trained machine-learning model match well with the measured values. Therefore, our machine-learning approach paves the way for accelerating the discovery of new ligands for rare-earth separations.

Highly Tunable 4-Phosphoryl Pyrazolone Receptors for Selective Rare-Earth Separation

“…Yield: 3.83 g (48%); mp 72−74 °C; Raman (80 mW, in cm −1 ) ν: 3083 (16), 2983 (26), 2937 (52), 2929 (53), 2873 (18), 1598 (100), 1570 (13), 1527 (15), 1494 (20), 1450 (23), 1427 (14), 1415 (13), 1402 (19), 1369 (21), 1344 (21), 1313 (14), 1284 (13), 1178 (13), 1095 (25), 1053 (24), 837 (17), 671 (22), 73 (56); IR (ATR, in cm −1 ) ν: 2980 (vw), 2933 (vw), 2877 (vw), 1597 (vw), 1587 (vw), 1562 (w), 1520 (w), 1493 (m), 1468 (w), 1450 (vw), 1414 (w), 1387 (w), 1375 (w), 1336 (w), 1282 (vw), 1269 (vw), 1242 (vw), 1178 (w), 1153 (m), 1101 (w), 1053 (vw), 978 (vs), 937 (w), 899 (w), 887 (w), 850 (vw), 829 (m), 775 (m), 748 (m), 704 (vw), 669 (m), 646 (w), 629 (vw), 607 (w), 575 (s), 530 (w), 519 (w), 503 (w), 422 (w); 1 H NMR (CDCl 3 , 300 K, in ppm): δ 1.28 (6H, d, 3 J HH = 6.2 Hz, H6a), 1.40 (6H, d, 3 J HH = 6.1 Hz, H6b), 2.27 (3H, d, 4 J HP = 0.7 Hz, H4), 4.62 (2H, d sept, 3 J HP = 7.7 Hz, 3 J HH = 6.3 Hz, H5), 7.39 (2H, md, 3 J HH = 9.0 Hz, H9/H10), 7.77 (2H, md, 3 J HH = 9.0 Hz, H9/ H10); 13 4.8. General Procedure for the Syntheses of Lanthanum(III), Europium(III), and Ytterbium(III) Complexes.…”

“…Yield: 0.45 g (18%); mp 101−103 °C; Raman (80 mW, in cm −1 ) ν: 3074 (38), 3064 (42), 3035 (15), 2987 (16), 2958 (46), 2923 (42), 2858 (35), 1598 (100), 1550 (9), 1527 (27), 1492 (13), 1438 (45), 1400 (32), 1371 (19), 1325 (22), 1300 (9), 1180 (6), 1161 (20), 1051 (10), 1024 (14), 1010 (11), 999 (61), 850 (15), 715 (9), 642 (18), 615 (15), 436 (6), 266 (6), 247 (6), 148 (11), 117 (13), 75 (74); IR (ATR, in cm −1 ) ν: 2955 (vw), 2856 (vw), 2490 (vw), 1599 (w), 1549 (m), 1531 (m), 1491 (w), 1460 (w), 1437 (w), 1400 (w), 1369 (vw), 1325 (w), 1298 (vw), 1265 (vw), 1188 (s), 1165 (m), 1092 (w), 1070 (w), 1051 (w), 1011 (vs), 924 (vw), 833 (w), 804 (vs), 768 (s), 712 (m), 692 (s), 654 (w), 642 (vw), 607 (m), 552 (vs), 511 (m), 434 (m); 1 H NMR (CDCl 3 , 300 K, in ppm): δ 2.26 (3H, d, 4 J HP = 0.8 Hz, H4), 3.76 (6H, d, 3 J HP = 11.7 Hz, H5), 7.28 (1H, mt, 3 J HH = 7.5 Hz, H10), 7.43 (2H, mt, 3 J HH = 8.0 Hz, H9), 7.77 (2H, md, 3 J HH = 7.6 Hz, H8); 13 C{ 1 H} NMR (CDCl 3 , 300 K, in ppm): δ 14.0 (1C, s, C4), 52.9 (2C, d, 2 J CP = 5 Hz, C5), 82.6 (1C, d, 1 J CP = 219 Hz, C2), 121.5 (2C, s, C8), 126.8 (1C, s, C10), 129.2 (2C, s, C9), 137.8 (1C, s, C7), 149.6 (1C, d, 2 J CP = 10 Hz, C3), 159.8 (1C, d, 2 J CP = 23 Hz, C1); 31 (2-ethylhexyl)(5-hydroxy-3-methyl-1-phenyl-1H-pyrazol-4-yl)phosphonate HL 3 .…”

An assessment of the strategies for the energy-critical elements necessary for the development of sustainable energy sources

“…It is possible to do high-throughput screening of an extended chemical space, and the machine learning algorithm-based model will continuously improve as more data are generated as well. As a result of the increasing popularity of this approach, it has been increasingly used in predicting key equilibrium properties of molecules such as solubility, binding a nity, pKa, adsorption capacities, and partition coe cients (Wang et showing that deep neural networks, which have been trained on the available experimental data, can be applied to the prediction of accurate distribution coe cients of rare earth ions for solvent extraction, opening the door for high-throughput screening of ligands for rare earth separations (Liu et al 2022).…”

Machine learning-guided discovery of ionic polymer electrolytes for lithium metal batteries

“…27 However, the reported high accuracy usually originates from the over tting of the dataset. 28,29 Among the training datasets, the sample size of unique ILs is extremely limited. 27,28 The investigated datasets usually contain the datapoints of the same ILs at varying temperatures, these replicated datapoints will increase the appeal accuracy of the reported models arti cially.…”