# Fileset

[supporting-information.pdf](https://mdr.nims.go.jp/filesets/3a16cf5d-e2dc-4835-bd8a-e8f163deb852/download)

## Creator

[Yukinori Koyama](https://orcid.org/0000-0002-7090-4430), [Yukako Kohriki](https://orcid.org/0000-0002-6858-1273), [Masamichi Harada](https://orcid.org/0000-0002-7321-0733), [Naoto Hirosaki](https://orcid.org/0000-0001-9218-9557), [Takashi Takeda](https://orcid.org/0000-0003-2510-4562)

## Rights

This document is the Accepted Manuscript version of a Published Work that appeared in final form in Chemistry of Materials, copyright © 2024 American Chemical Society after peer review and technical editing by the publisher. To access the final edited and published work see https://doi.org/10.1021/acs.chemmater.4c01981.[In Copyright](http://rightsstatements.org/vocab/InC/1.0/)

## Other metadata

[Accelerating Materials Discovery of Novel Europium(II)-Activated Phosphors through Machine Learning Classification of Europium Valences](https://mdr.nims.go.jp/datasets/6ffc8a71-859d-4299-af90-08cf6a49cc7a)

## Fulltext

1 Accelerating Materials Discovery of Novel Europium(II)-Activated Phosphors through Machine Learning Classification of Europium Valences Yukinori Koyama, *, a Yukako Kohriki,b Masamichi Harada,b Naoto Hirosaki,b Takashi Takeda b a Center for Basic Research on Materials, National Institute for Materials Science, Tsukuba, Ibaraki 305-0047, Japan b Research Center for Electronic and Optical Materials, National Institute for Materials Science, Tsukuba, Ibaraki 305-0044, Japan.     2 S1. Feature importance of the machine learning model This section discusses the importance of the features used in the machine learning model. Figure S1 illustrates the absolute values of the logistic regression coefficients in descending order for the 90 features selected in the preprocessing steps. The top three regression coefficients were significant, and the following six features had coefficients of similar size. The coefficient of the tenth feature was less than one-third of that of the top feature. After that, the coefficients gradually decreased. The top nine features and their coefficients are presented in Table S1. The top two features were both related to electronegativity, indicating that the electronegativity of the constituent elements significantly affects the oxidation state of the Eu ions in the host. The signs of the coefficients were inverse between the arithmetic mean of the top feature and the geometric mean of the second one, suggesting a nonlinear relationship between the electronegativity and the oxidation states of Eu ions. This study used arithmetic, geometric, and harmonic means to average the elemental features. Items with smaller values contribute more in this order. Therefore, the negative coefficient for the arithmetic mean and the positive coefficient for the geometric mean indicate that there was a negative correlation between the electronegativity and the Eu2+-paper ratio for electronegative elements, namely anions and a positive correlation for less electronegative elements, namely cations. The negative correlation for anions corresponds to the small Eu2+-paper ratio in the O- and F-containing compounds. On the other hand, we have not identified specific trends regarding the positive correlation for cations. Similarly, looking at the third and sixth features, which are related to the Mendeleev numbers (MN), there was a positive correlation between the MN and Eu2+-paper ratio for elements of large MN located on the right side of the periodic table, and a negative correlation for elements of small  3 MN located on the left side of the periodic table. The former positive correlation corresponds to the large Eu2+-paper ratio of compounds containing halogens (Cl, Br, and I), and the latter negative correlation corresponds to the small Eu2+-paper ratio of compounds containing rare earth and Group-5 and 6 metals. Because the number of unoccupied valence p states is inversely correlated with the MN, the trend of the Eu2+-paper ratio regarding the number of unoccupied valence p states is almost the same as that regarding the MN. Because this study used general-purpose descriptors, it is impossible to directly measure the effect of elemental content. Thus, the above discussion is only an interpretation of the feature importance. Furthermore, atomic number, ionization energy, and group number appear only once in this table, but the averages of different types are included in the group below the tenth feature. It is necessary to interpret the effect of elemental content with consideration of the related features.     4  Figure S1. Absolute values of the logistic regression coefficients in descending order. Table S1. Top nine features and those coefficients in the logistic regression. Rank Elemental feature Statistics Coefficient 1 Electronegativity arithmetic mean –25.829 2 Electronegativity geometric mean 19.505 3 Mendeleev number geometric mean –16.567 4 Atomic number geometric mean 12.311 5 Ionization energy geometric mean –11.775 6 Mendeleev number arithmetic mean 10.947 7 Number of unoccupied valence p states geometric mean 10.840 8 Number of unoccupied valence p states  harmonic mean –10.552 9 Group in the periodic table geometric mean 10.212  5 S2. Machine learning classification with gradient boosted trees method A machine learning classification model with the gradient boosted trees method was developed to predict the oxidation states of Eu ions in the hosts regarding luminescence based on their host compositions. The machine learning pipeline was the same as that using logistic regression, explained in the main text, and the gradient-boosted-trees method was used as the classifier in the RFE and final steps. The parameters shown in Table S2 were determined by the Bayesian optimization method using the scikit-optimize package [Ref. S1] to maximize the F-score for the validation data of the 10-fold cross-validation. Accuracy, precision, recall, and F-score with the optimum parameters evaluated for the training and validation dataset in 10-fold cross-validation are summarized in Table S3.  Table S2. Parameter ranges and optimum values of the machine learning classification model using the gradient-boosted-trees method. Parameter Candidates Optimum Number of selected features in RFE 10, 20, …, 100 100 Learning rate 10–2 – 0.5 (log-scale) 0.037 Maximum depth of trees 1, 2, 3, 4, 5 3 Number of trees 100, 200, …, 1000 700     6 Table S3. Classification metrics of the machine learning model using the gradient-boosted-trees method with the optimum parameters evaluated for the training and validation datasets in 10-fold cross-validation. Metric Training Validation Accuracy 94.7% (0.1%) 86.9% (1.8%) Precision 92.9% (0.2%) 84.0% (3.7%) Recall 93.6% (0.4%) 81.7% (2.7%) F-score 93.2% (0.2%) 82.8% (2.6%) The scores were averaged among the folds. Standard deviations over folds are shown in parentheses.  References [S1] Scikit-optimize. https://scikit-optimize.github.io/.     7 S3. Synthesis and powder X-ray diffraction analysis Synthesis conditions and powder X-ray diffraction (XRD) analysis results of the samples for which photoluminescence was observed are summarized in this section. The XRD patterns of the powder samples are illustrated with simulated patterns of target and impurity phases. Some samples showed hollow peaks at 10-20 degrees and high backgrounds owing to using an airtight sample holder to protect air-sensitive samples. The XRD patterns were analyzed to identify phases and fractions in the products using the whole powder pattern fitting method (Rigaku, PDXL2) with card peak patterns provided by the Joint Committee on Powder Diffraction Standards (JCPDS) and simulated peak patterns from crystal structure data obtained from the Inorganic Crystal Structure Database (ICSD).     8 S3.1. Rb3Ca2Cl7 Table S4. Starting materials and synthesis conditions of Rb3Ca2Cl7. Starting materials RbCl, CaCl2, EuCl3 Synthesis conditions 600 °C, 5 h, H2(5%)/N2   Figure S2. XRD pattern of the Rb3Ca2Cl7 sample (top) and simulated patterns of the target and impurity phases (bottom). Table S5. XRD analysis results of the Rb3Ca2Cl7 sample. Phase ICDD card number Space group Fraction (wt%) Ca2 Cl7 Rb3 04-022-6840 I 4/m m m (139) 72.1(15) Rb Cl 04-005-4309 F m -3 m (225) 14.9(5) Ca Cl2 01-071-5407 P b c n (60) 13.0(17)     9 S3.2. RbSrCl3 Table S6. Starting materials and synthesis conditions of RbSrCl3. Starting materials RbCl, SrCl2, EuCl3 Synthesis conditions 600 °C, 5 h, H2(5%)/N2   Figure S3. XRD pattern of the RbSrCl3 sample (top) and simulated patterns of the target and impurity phases (bottom). Table S7. XRD analysis results of the RbSrCl3 sample. Phase ICDD card number Space group Fraction (wt%) Rb Sr Cl3 ICSD 138559 P n m a (62) 98.3(8) Rb Cl 04-005-4309 F m -3 m (225) 1.7(8)     10 S3.3. Cs2CaBr4 Table S8. Starting materials and synthesis conditions of Cs2CaBr4. Starting materials CsBr, CaBr2, EuBr2 Synthesis conditions 550 °C, 5 h, H2(5%)/N2   Figure S4. XRD pattern of the Cs2CaBr4 sample (top) and simulated patterns of the target and impurity phases (bottom). Table S9. XRD analysis results of the Cs2CaBr4 sample. Phase ICDD card number Space group Fraction (wt%) Cs2 Ca Br4 04-010-0063 I 4/m m m (139) 90.8(7) Cs Br 01-082-9638 P m -3 m (221) 9.2(7)     11 S3.4. Rb4CaBr6 Table S10. Starting materials and synthesis conditions of Rb4CaBr6. Starting materials RbBr, CaBr2, EuBr2 Synthesis conditions 550 °C, 5 h, H2(5%)/N2   Figure S5. XRD pattern of the Rb4CaBr6 sample (top) and simulated patterns of the target and impurity phases (bottom). Table S11. XRD analysis results of the Rb4CaBr6 sample. Phase ICDD card number Space group Fraction (wt%) Rb4 Ca Br6 04-010-0064 R -3 c (167) 85.4(3) Rb Br 01-073-0384 F m -3 m (225) 13.1(3) Ca Br2 01-071-5406 P 42/m n m (136) 1.45(2)     12 S3.5. BaB2S4 Table S12. Starting materials and synthesis conditions of BaB2S4. Starting materials BaS, B, S, EuS Synthesis conditions 450 °C, 2 h + 700 °C, 6 h, vacuum-sealed quartz ampoule   Figure S6. XRD pattern of the BaB2S4 sample (top) and simulated patterns of the target and impurity phases (bottom). Table S13. XRD analysis results of the BaB2S4 sample. Phase ICDD card number Space group Fraction (wt%) B2 Ba S4 04-010-8537 C c (9) 87(6) B2 S3 03-065-1287 P 21/c (14) 12(5)     13 S3.6. BaLa2ZnS5 Table S14. Starting materials and synthesis conditions of BaLa2ZnS5. Starting materials BaS, La2S3, ZnS, EuS Synthesis conditions 900 °C, 10 h, vacuum-sealed quartz ampoule.   Figure S7. XRD pattern of the BaLa2ZnS5 sample (top) and simulated patterns of the target and impurity phases (bottom). Table S15. XRD analysis results of the BaLa2ZnS5 sample. Phase ICDD card number Space group Fraction (wt%) Ba Zn La2 S5 04-009-1670 I 4/m c m (140) 76.8(8) La2 O2 S 01-078-7509 P -3 m 1 (164) 18.4(6) Zn S 01-072-9271 P 3 m 1 (156) 4.8(6)     14 S3.7. NaLi3BaB6O12 Table S16. Starting materials and synthesis conditions of NaLi3BaB6O12. Starting materials BaCO3, Na2CO3, Li2CO3, H3BO3, Eu2O3 Synthesis conditions 300 °C, 4 h, air + 600 °C, 8 h, air + 600 °C, 5h, H2(4%)/Ar   Figure S8. XRD pattern of the NaLi3BaB6O12 sample (top) and simulated patterns of the target and impurity phases (bottom). Table S17. XRD analysis results of the NaLi3BaB6O12 sample. Phase ICDD card number Space group Fraction (wt%) Na Li3 Ba B6 O12 04-019-4572 R -3 (148) 97.1(2) Li Ba2 B5 O10 04-009-4377 P 21/m (11) 2.9(2)     15 S3.8. KNaCaMg5Si8O22F2 Table S18. Starting materials and synthesis conditions of KNaCaMg5Si8O22F2. Starting materials Na2CO3, K2CO3, SiO2, MgO, CaF2, Eu2O3 Synthesis conditions 1000 °C, 5 h, H2(5%)/N2   Figure S9. XRD pattern of the KNaCaMg5Si8O22F2 sample (top) and simulated patterns of the target and impurity phases (bottom). Table S19. XRD analysis results of the KNaCaMg5Si8O22F2 sample. Phase ICDD card number Space group Fraction (wt%) K Na Ca Mg5 Si8 O22 F2 04-013-2169 C 2/m (12) 95.1(6) Ca Si O3 04-016-5569 P -1 (2) 4.3(5)     16 S3.9. Sr2OI2 Table S20. Starting materials and synthesis conditions of Sr2OI2. Starting materials SrO, SrI2, Eu2O3 Synthesis conditions 675 °C, 5 h, H2(5%)/N2   Figure S10. XRD pattern of the Sr2OI2 sample (top) and simulated patterns of the target and impurity phases (bottom). Table S21. XRD analysis results of the Sr2OI2 sample. Phase ICDD card number Space group Fraction (wt%) Sr2 I2 O 04-014-0889 I b a m (72) 73(2) Sr I ( O H ) 04-011-6332  P n m a (62) 13.5(10) Sr4 I6 O 04-022-6011 P 63 m c (186) 7.9(11) Sr O 01-075-6979 F m -3 m (225) 6(2)   17 S3.10. Sr4OI6 Table S22. Starting materials and synthesis conditions of Sr4OI6. Starting materials SrO, SrI2, Eu2O3 Synthesis conditions 650 °C, 8 h, H2(5%)/N2   Figure S11. XRD pattern of the Sr4OI6 sample (top) and simulated patterns of the target and impurity phases (bottom). Table S23. XRD analysis results of the Sr4OI6 sample. Phase ICDD card number Space group Fraction (wt%) Sr4 I6 O 04-022-6011 P 63 m c (186) 74.5(7) Sr I2 04-007-1983 P b c a (61) 25.5(7)     18 S3.11. Ba4OI6 Table S24. Starting materials and synthesis conditions of Ba4OI6. Starting materials BaO, BaI2, Eu2O3 Synthesis conditions 650 °C, 5 h, H2(5%)/N2 + 650 °C, 2 h, H2(5%)/N2   Figure S12. XRD pattern of the Ba4OI6 sample (top) and simulated patterns of the target and impurity phases (bottom). Table S25. XRD analysis results of the Ba4OI6 sample. Phase ICDD card number Space group Fraction (wt%) Ba4 I6 O 04-011-1893 P 63 m c (186) 95.8(8) Ba2 I2 O 04-016-5407 I b a m (72) 4.2(8)     19 S3.12. Ba2PO4I Table S26. Starting materials and synthesis conditions of Ba2PO4I. Starting materials Ba3(PO4)2, BaI2, Eu2O3 Synthesis conditions 800 °C, 5 h, H2(5%)/N2   Figure S13. XRD pattern of the Ba2PO4I sample (top) and simulated patterns of the target and impurity phases (bottom). Table S27. XRD analysis results of the Ba2PO4I sample. Phase ICDD card number Space group Fraction (wt%) Ba2 ( P O4 ) I 04-021-7746 P 21/c (14) 94(2) Ba O 01-080-3980 F m -3 m (225) 6(2)     20 S3.13. Sr2PO4I Table S28. Starting materials and synthesis conditions of Sr2PO4I. Starting materials Sr3(PO4)2, SrI2, Eu2O3 Synthesis conditions 800 °C, 5 h, H2(5%)/N2   Figure S14. XRD pattern of the Sr2PO4I sample (top) and simulated patterns of the target and impurity phases (bottom). Table S29. XRD analysis results of the Sr2PO4I sample. Phase ICDD card number Space group Fraction (wt%) Sr2 ( P O4 ) I 04-021-7748 P 21/c (14) 97(2) Sr3 ( P O4 )2 01-073-4870 R -3 m (166) 3(2)