Correlation between Photocatalytic Activities and Structural and ...

Correlation between Photocatalytic Activities and Structural and Physical Properties of Titanium(IV) Oxide Powders Orlando-Omar Prieto-Mahaney,1 Naoya Murakami,1 Ryu Abe,1;2 and Bunsho Ohtani
1;2 1 Graduate School of Environmental Science, Hokkaido University, Sapporo 060-0810 2 Catalysis Research Center, Hokkaido University, Sapporo 001-0021 (Received November 14, 2008; CL-081076; E-mail: [email protected])

predicted photocatalytic activity (standardized)

1

4Ag+ + 2H2O 4Ag + O 2 + 4H+

0.8

(Reaction a) 0.6 0.4 0.2 0 0

0.2 0.4 0.6 0.8 1 observed photocatalytic activity (standardized)

Correlation between structural and physical properties and photocatalytic activities for five kinds of reactions of 35 commercial and home-made samples of pristine titanium(IV) oxide (titania) was obtained through multivariable analyses: photocatalytic activities were empirically reproduced by a linear combination of six properties with fair reliability. An example for oxygen liberation is shown here. While a part of results could be interpreted using the conventional mechanism, significant activity dependence on properties, not disclosed yet, was suggested.

REPRINTED FROM

Vol.38 No.3 2009 p:238{239 CMLTAG March 5, 2009

The Chemical Society of Japan Published on the web (Advance View) January 31, 2009; doi:10.1246/cl.2009.238

238

Chemistry Letters Vol.38, No.3 (2009)

Correlation between Photocatalytic Activities and Structural and Physical Properties of Titanium(IV) Oxide Powders Orlando-Omar Prieto-Mahaney,1 Naoya Murakami,1 Ryu Abe,1;2 and Bunsho Ohtani
1;2 1 Graduate School of Environmental Science, Hokkaido University, Sapporo 060-0810 2 Catalysis Research Center, Hokkaido University, Sapporo 001-0021 (Received November 14, 2008; CL-081076; E-mail: [email protected]) Correlation between structural and physical properties and photocatalytic activities for five kinds of reactions of 35 titania samples was obtained through multivariable analyses: photocatalytic activities were empirically reproduced by a linear combination of six properties with fair reliability. While a portion of results could be interpreted using a conventional mechanism, significant activity dependence on properties, not disclosed yet, was suggested.

Photocatalytic reaction is induced by photoexcited electrons (e ) and positive holes (hþ ) generated in a solid photocatalyst followed by redox reaction with surface-adsorbed substrates. Titanium(IV) oxide (titania) is one of the most promising photocatalysts because of its high chemical stability, sufficient energy of its e and hþ to drive various photocatalytic reactions, negligible toxicity, inexpensiveness, and ease of preparation. Only one plausible weakness of the titania photocatalyst is that it absorbs only ultraviolet light, though such transparency is preferable when titania is coated on colored materials. It is known that the photocatalytic activity, i.e., rate of photocatalytic reaction, depends on the structural and physical properties of titania. It is believed that there is a relationship between properties and activities, i.e., a structure–activity correlation. It has often been claimed that the smaller the particle size, i.e., the larger the specific surface area, the higher the photocatalytic activity or that anatase is better than rutile.1 However, such discussion on the correlations has been limited to a certain series of samples prepared in a similar way or small number of commercial samples,2 and there seem no comprehensive correlations. The present study aims at obtaining the structure–activity correlations for a large number of titania photocatalysts covering commercially available titania samples in Japan (see Table S1)14 by statistical multivariable analysis. Five representative reactions were chosen, and their relative rates were analyzed using six properties,3 specific surface area (BET), density of lattice defects (DEF), primary (PPS) and secondary (SPS) particle size, and existence of anatase (ANA) and rutile (RUT) phases, to obtain intrinsic dependence of photocatalytic activities on the properties. Five test photocatalytic reactions were as follows: a, Oxygen (O2 ) liberation and silver deposition from a deaerated aqueous silver sulfate solution; b, Dehydrogenation of methanol in a deaerated aqueous solution; c, Oxidative decomposition of acetic acid to liberate carbon dioxide (CO2 ) from an aerated aqueous solution; d, Decomposition of acetaldehyde into CO2 in air; and e, Synthesis of pipecolinic acid from L-lysine in a deaerated aqueous solution (see SI). First, the rates of reactions were compared with each structural property. In general, there were no significant relations between them, while, in some cases, dependence of the rate on

a parameter was observed (see SI, Tables S2 and S4). This seems reasonable since the titania samples used in this study were of a wide range and synthesized or treated in different ways. Then, the data were analyzed statistically by solving the following matrix equation, ½rate351 ¼ ½property356  ½coefficient61 , for each reaction. The experimental results, rates and properties, were standardized using mean of data and standard deviation in order to make the calculated coefficients have the same weight being independent of properties, i.e., enabling direct comparison of partial regression coefficients (k). Table 1 shows the summary of results of analyses. Squared multiple correlation coefficient, R2 , was also listed. It was found that the best R2 values were obtained when ANA and RUT were defined to be 1 or 0 (dummy variables) when the most intensive peak of anatase and rutile was detected in the XRD patterns and otherwise, respectively. Although it is known that increase in the number of variables improves the fitting, i.e., makes R2 to be close to unity, no other appropriate properties have been found to be added to the list. As a general trend, a and e gave relatively larger R2 s, i.e., higher reproducibility of the results fitting to a linear combination of properties, while those for the others were also fairly high.4 Figure 1 shows this linearity for a as an example; the calculated (predicted) relative activities of each sample using obtained coefficients (k) were plotted against the observed activity. Another significant feature is that coefficient of kANA has large positive value in all the cases except for a. This is the first example, within the authors’ knowledge, of support for a general understanding in this field that anatase is more active than rutile, since previously reported data were of only a limited number of samples and neglected the influence of the other properties. On the other hand, kRUT was relatively small and it was negative in d suggesting that rutile phase in the photocatalysts is rather inert compared with anatase. Although large positive and negative kBET and kDEF , respectively, are expected since large surface area is advantageous to adsorb reaction substrate(s) and defective sites accelerate recombination of e and hþ , large positive kBET was observed only for b and even large positive kDEF was obtained for c, d, and e as discussed later. kPPS and kSPS depended on the type of reactions. Table 1. Squared multiple correlation coefficient (R2 ) and partial regression coefficients (k) Coefficient R2 kBET kDEF kPPS kSPS kRUT kANA

a 0.86 0:01 0:15 0.12 0.57 0.14 0.04

b 0.52 0.43 0:25 0:20 0.08 0.28 0.40

Copyright Ó 2009 The Chemical Society of Japan

c 0.58 0:09 0.19 0:18 0:20 0.11 0.57

d 0.60 0.13 0.43 0:20 0:04 0:06 0.55

e 0.85 0.19 0.32 0:52 0:07 0.02 0.63

Chemistry Letters Vol.38, No.3 (2009)

calculated photocatalytic activity for reaction a

1 0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 0.8 1 standardized observed photocatalytic activity for reaction a

Figure 1. Relation between standardized photocatalytic activities for reaction a observed and calculated using obtained partial regression coefficients in Table 1. This linearity indicates that the activity of samples can be estimated from their physical properties.

For a, kANA was small but positive, which means that anatase does not prohibit the reaction, while kRUT was not so large, suggesting that it is not indispensable to be rutile for higher activity. It seems that large secondary particles (large positive kSPS ) composed of large primary particles of less crystalline defects (negative kDEF ) are preferable, which agrees with the general recognition reported so far.5 A significant point, which has been revealed in this study first, is that the reported higher activity of rutile powders6 for a is mainly attributable to their secondary particle size7 but not to rutile crystal, and it is expected that anatase would be active when its secondary particle size is large. Reaction b was performed using in situ platinized titania particles, since bare samples show negligible activities.4 The positive kRUT and kANA coefficients seem reasonable, since the bottom of the conduction band (CB) of anatase (0:20 V vs. NHE) and rutile (+0.04 V) crystals is reported8 to be almost the same or a little negative compared with the standard electrode potential of hydrogen evolution (Hþ /H2 ; 0 V). kANA , a little larger than kRUT , might be related to a slightly more negative CB level of anatase. Comparable activity of rutile phase has been shown previously through action spectrum analyses.9 It has been shown for hydrogenation of 2-propanol that the rate depends on the amount of surface-adsorbed alcohol.10 Large positive kBET , as well as negative kPPS , and large negative kDEF suggested that this reaction requires both a large amount of adsorbed methanol and less probability of e –hþ recombination. Reactions c and d, exhibiting a similar trend of coefficients, were conducted under aerated conditions, and their mechanism must contain O2 reduction by e . Preference of anatase crystallites (large positive kANA ) may be caused by the above-mentioned CB position. The potential of one-electron reduction of O2 , O2  /O2 (0:05 V) or HO2 /O2 (0:28 V) is a little more negative than that for H2 evolution, and thereby small difference (ca. 0.16 V) in CB position may be decisive. Another feature of these oxidative decompositions was a relatively large positive kDEF , which must accelerate the e –hþ recombination. A possible interpretation for this inconsistency is that the defective sites act as an adsorption site for organic compounds and/or O2 and the influence of enhanced e –hþ recombination on the rate was small, presumably owing to efficient radical chain reaction.11 Taking into account negative kDEF for b, in which organic compound, methanol, is also oxidized, enhanced adsorption of O2 at the defective sites might occur, though at present we have no evidence supporting this hypothesis. Positive kBET for d, in con-

239 trast to the negative value for c, suggests that this gas–solid phase reaction requires diffusion of substrates, acetaldehyde and O2 , and, therefore, the larger the surface area, the faster the diffusion for d. Reaction e has been known to proceed through redox-combined mechanisms; Lys is oxidized by hþ and hydrolyzed and cyclized to a Schiff base intermediate which undergoes reduction, at surface-loaded Pt deposits, by e to yield PCA.12 Similarity of the trend with b suggests that the reaction rate is also governed by the first oxidation step. The positive kDEF is attributable to the preferable adsorption of Lys at the defective sites. The larger and smaller kANA and kRUT , respectively, compared with b might reflect a difference in reduction process between b and e; reduction of the Schiff base intermediate13 is thought to require a little more negative potential of e than that for H2 liberation. In conclusion, statistical multivariable analyses of photocatalytic activities depending on the properties of titania powders supported the conventional wisdom, e.g., anatase is more active than rutile especially for photocatalytic oxidative decomposition of organic compounds, and lattice defects reduce activity in reactions conducted under deaerated conditions. On the other hand, unexpected results were also observed; e.g., lattice defects are beneficial especially for reactions under aerated conditions, and the activity for O2 liberation is governed by secondary particle size rather than crystal form. Since each property changes depending on each other to some extent (see Tables S2 and S3), e.g., rutile crystallites are often large compared with anatase, intrinsic dependence of activity on each property has been ambiguous. The intrinsic dependence of activity as shown in this study could be demonstrated, for the first time, by statistical multivariable analyses. References and Notes 1 Problems appearing in such discussions have been pointed out in a recent article: B. Ohtani, Chem. Lett. 2008, 37, 216. 2 A recent paper has reported the activity of 6 samples depending on the type of organic compounds to be degraded in aerated aqueous solutions: J. Ryu, W. Choi, Environ. Sci. Technol. 2008, 42, 294. 3 Some of the properties have been already reported: N. Murakami, O. O. Prieto-Mahaney, R. Abe, T. Torimoto, B. Ohtani, J. Phys. Chem. C 2007, 111, 11927. 4 Relatively small R2 for reaction b might be due to possible differences in size and morphologies of platinum deposits and this will be discussed elsewhere. 5 For example: S. Nishimoto, B. Ohtani, H. Kajiwara, T. Kagiya, J. Chem. Soc., Faraday Trans. 1 1983, 79, 2685. 6 For example: S. Nishimoto, B. Ohtani, H. Kajiwara, T. Kagiya, J. Chem. Soc., Faraday Trans. 1 1985, 81, 61. 7 Nosaka et al. reported a relation between secondary particle size and rate of photocatalytic superoxide anion radical production, but the reason has not been fully clarified yet: Y. Nosaka, M. Nakamura, T. Hirakawa, Phys. Chem. Chem. Phys. 2002, 4, 1088. 8 G. Rothenberger, J. Moser, M. Gra¨tzel, N. Serpone, D. K. Sharma, J. Am. Chem. Soc. 1985, 107, 8054. 9 T. Torimoto, N. Nakamura, S. Ikeda, B. Ohtani, Phys. Chem. Chem. Phys. 2002, 4, 5910. 10 S. Nishimoto, B. Ohtani, T. Kagiya, J. Chem. Soc., Faraday Trans. 1 1985, 81, 2467. 11 B. Ohtani, Y. Nohara, R. Abe, Electrochemistry 2008, 76, 147. 12 B. Ohtani, S. Tsuru, S. Nishimoto, T. Kagiya, K. Izawa, J. Org. Chem. 1990, 55, 5551. 13 B. Ohtani, K. Iwai, S. Nishimoto, S. Sato, J. Phys. Chem. B 1997, 101, 3349. 14 Supporting Information is available electronically on the CSJ-Journal Web site, http://www.csj.jp/journals/chem-lett/index.html.

Published on the web (Advance View) January 31, 2009; doi:10.1246/cl.2009.238

Supporting Information

Correlation between Photocatalytic Activities and Structural and Physical Properties of Titanium(IV) Oxide Powders Orlando-Omar Prieto-Mahaney,1 Naoya Murakami,1 Ryu Abe,1,2 and Bunsho Ohtani*1,2 1 Graduate School of Environmental Science, Hokkaido University, Sapporo 060-0810 2 Catalysis Research Center, Hokkaido University, Sapporo 001-0021 (Received November 14, 2008; CL-081076; E-mail: [email protected])

Copyright © The Chemical Society of Japan

Supporting Information

Experimental Details Samples: Samples were supplied by several sources, mainly Catalysis Society of Japan (JRC-TIO-1 to TIO-13), Showa Denko (Supertitania series, F1-F5, G2), and Ishihara Sangyo (CR-EL, ST-21, ST-01). Other samples were named after their suppliers' name (Merck, Kanto, etc.) and finally a home made sample (TUF-01). Degussa P25 was provided by Nippon Aerosil Co. All samples were used as received, without further processing. Although JRC-TIO-4 is to be identical to Degussa P25, their measured properties were a little different. This may be caused by possible heterogeneity of this sample i and they were handled as independent samples. Similarly, another set of samples, ST-01 (Ishihara) and JRC-TIO-8, was considered independent. Photocatalytic reaction: A standard amount of the powder photocatalyst (50 mg) was suspended in an aqueous solution (5.0 mL) in a borosilicate glass tube (transparent for the wavelength > 290 nm, 18 mm in inner diameter and 180 mm in length) containing silver fluoride (50 mmol L-1), methanol (50 vol%) and chloroplatinic acid (H2PtCl6·6H2O; corresponding to 2-wt% platinum (Pt) loading), acetic acid (5.0 vol%) in reactions a, b, and c, respectively. Air was purged off from the systems by passing argon (Ar) through the suspensions for at least 15 min for reactions a, b, and e and tightly sealing the sample tubes using a double-capped rubber septum and a sheet of Parafilm to prevent leakage of gas and/or contamination during the measurements for the liquid-solid phase reactions (a-c, e). In the case of reaction e, Pt-preloaded (2 wt%) titania powders were used alongside a neutralized solution of L-lysine hydrochloride (Lys; 20 mmol L-1) as a substrate. The samples were irradiated by a 400-W high-pressure mercury arc (Eiko-sha) at 298 K under vigorous magnetic stirring (1000 rpm). The reactions were monitored for 60 min (reactions a, b, and e) and 90 min (reaction c) by analyzing liberation of oxygen (O2), hydrogen (H2), a deaminocyclized product, pipecolinic acid (PCA), and carbon dioxide (CO2), respectively. The gas phase products and PCA were analyzed by gas chromatography (a Shimadzu GC-8A gas chromatograph equipped with a TCD and columns of molecular sieve 5A for H2 and O2 and Porapak Q for CO2) and HPLC (Shimadzu LC-6A; a Sumichiral A6000 column with 2.0 mmol L-1 copper sulfate eluent; detected by photoabsorption at 254 nm), respectively. Photocatalytic oxidative decomposition of gaseous acetaldehyde in air (reaction d) was carried out in a cylindrical glass vessel with a volume of 660 mL. Photocatalyst powder (50 mg) was uniformly spread on a glass plate (1.0 cm × 1.0 cm). The plate was placed on the bottom of the cylindrical vessel. Gaseous acetaldehyde (0.66 mL corresponding to ca. 0.027 mmol) was injected into the vessel filled with ambient air to adjust the initial concentration to be 1000ppm. Concentrations of acetaldehyde and CO2 were measured by an Agilent 3000 MicroGC. After adsorption of acetaldehyde had reached an equilibrium in the dark, photoirradiation (>290 nm) through a top window of the vessel was performed using a 300-W xenon lamp (ILC Technology CERMAX-LX300F). Principle and details of these photocatalytic activity tests have been reported elsewhere. a: silver metal deposition and O2 liberation.ii,iii b: methanol dehydrogenation. iv c: oxidative decomposition of acetic acid. v , vi d: gas-phase oxidative decomposition of acetaldehyde.vii e: conversion of Lys to PCA.viii Photocatalytic activity, the rate of reaction, was evaluated by the rate of monitored products in an irradiation time range where the constant rate was observed except for reaction e. The yield of PCA, a sum of stereoisomers, by the photoirradiation for 60 min was used for the calculation of activity in reaction e. Raw data were listed in Table S1.

Structural and physical properties: Specific surface area (BET) of the photocatalysts was evaluated according to a Brunauer-Emmett-Teller equation with data of nitrogen (N2) adsorption isotherms at 77 K on a Yuasa-Quantachrome NOVA 1200e surface and pore size analyzer. Adsorption cross section of N2, 0.162 nm2, was used for calculation. At least 6 data points in the relative pressure range between 0.05 and 0.3 were collected. Density of lattice defects (DEF) was estimated through a quantitative analysis of trivalent titanium species (Ti3+) formed in titania suspended in a deaerated aqueous triethanolamine solution (TEOA, 10 vol%). After the irradiation, a deaerated aqueous solution of methyl viologen (MV2+) was injected to the suspension and the resulting pale blue supernatant involving cation radical of MV2+ was separated and its photoabsorption was measured by a spectrophotometer (Agilent 8453). ix Alternatively, double-beam photoacoustic spectroscopy (DB-PAS) in which titania powders are continuously irradiated by a light-emitting diode (LED) with an emission peak at 365 nm (Nichia NCCU033) and monitored by modulated visible light at 530 nm using light-emitting diode (Nichia NSPG500S) after passing a controlled N2 flow containing methanol vapor. Since no absolute molar amount of defects could be obtained by DB-PAS, the collected data were converted using a correlation curve which has been confirmed in the previous studies.x,xi Primary particle size (PPS) was determined through X-ray diffraction measurements using Cu-Kα radiation in a Rigaku RINT 2500 equipped with a carbon monochromator. Photocatalysts were loaded in a glass sample holder of ca. 0.5-mm depth. The crystallite size was estimated by Scherrer equation using the corrected (line-broadening by Cu-Kα2 radiation and emanation in the optical path of a diffractometer) full width at half maximum (FWHM) of the most intense XRD peaks of anatase or rutile at ca. 25.3 and 27.4 deg, respectively. As a constant in the Scherrer equation, 0.89 was used.xii Using XRD patterns, presence and absence of crystalline phases of anatase and rutile (ANA and RUT, respectively) was checked. The borderline between presence and absence is the detection limit of XRD peak, ca. 1% peak height compared with that of single-phase well-crystallized samples. One of the reasons why dummy values (1 or 0) were used for ANA and RUT was that it seems that there are no incredible methods for crystalline composition analysis, especially for the samples of small crystallitesxii and heterogeneity of crystalline composition as described above. Another reason is that the activity can not be a linear combination of activities of anatase and rutile (and possibly amorphous), e.g., it has been clarified that in certain reaction systems one of the crystalline phase shows rather higher activity even when the composition is small.xiii Secondary particle size (SPS) was determined by a laser diffraction particle analyzer (Shimadzu SALD7000) as a volume-average particle size. The sample was suspended in water by ultrasonication before the measurement. Part of data used in this study have been reported elsewhere.x

Multivariable Analyses Data standardization: All the data, including ANA and RUT, were standardized to give standard deviation, v, to be unity using a equation, zi = (xi - xave)/v, where zi, xi, and xave are standardized datum, raw datum, and average, so that direct comparison is possible for calculated partial regression coefficients. The standard deviations were calculated using n, number of data. Check of multicollinearity: When there are strong correlations betweens variables, properties, multivariable analysis should be made using one of variable out of strongly correlated variables. As shown in Table S3, judging from the calculated correlation coefficient between a set of properties, there were some significant correlation, but actual coefficients are at most < 0.6. We considered that there was no multicollinearity in the data used in

this study. As described in the text, BET and PPS should be strongly related each other assuming spherical particles of the same diameter. However, the corresponding correlation coefficient was not so large, suggesting that they behave independently. Dependence of photocatalytic activity on each property: Table S4 shows the correlation between photocatalytic activities with each one of properties. A few rather strong correlations were observed: kPPS and kSPS for reaction a and kPPS for reaction e. The kSPS coefficient for reaction a and the kPPS coefficient for reaction e were also evident in the multivariable analyses (see Table 1), reflecting this dependence. The reason of a relatively small kPPS coefficient for reaction a in Table 1, in spite of the large correlation coefficient in this table was ambiguous at present. Calculation: Multivariable analyses were performed with the above-mentioned standardized data using least-square method, to minimize sum of residues (square of deviation of calculated value from actual value) on Excel 2003 (Microsoft) and MATLAB (The MathWorks).

i ii iii iv v vi vii viii ix x xi xii xiii

O. O. Prieto-Mahaney, R. Abe, D. Li, Y. Azuma, B. Ohtani, to be submitted. S.-i. Nishimoto, B. Ohtani, H. Kajiwara, T. Kagiya, J. Chem. Soc., Faraday Trans.1 1983, 79, 2685. B. Ohtani, Y. Okugawa, S.-i. Nishimoto, T. Kagiya, J. Phys. Chem. 1987, 91, 3550. S.-i. Nishimoto, B. Ohtani, T. Kagiya, J. Chem. Soc., Faraday Trans. 1 1985, 81, 2467. S. Ikeda, H. Kobayashi, Y. Ikoma, T. Harada, T. Torimoto, B. Ohtani, M. Matsumura, Phys. Chem. Chem. Phys. 2007, 9, 6319. X. Yan, T. Ohno, K. Nishijima, R. Abe, B. Ohtani, Chem. Phys. Lett. 2006, 429, 606. F. Amano, K. Nogami, R. Abe, B. Ohtani, J. Phys. Chem. C 2008, 112, 9320. B. Pal, S. Ikeda, H. Kominami, Y. Kera, B. Ohtani, J. Catal. 2003, 217, 152. S. Ikeda, N. Sugiyama, S.-y. Murakami, H. Kominami, Y. Kera, H. Noguchi, K. Uosaki, T. Torimoto, B. Ohtani, Phys. Chem. Chem. Phys. 2003, 5, 778. N. Murakami, O. O. Prieto-Mahaney, R. Abe, T. Torimoto, B. Ohtani, J. Phys. Chem. C 2007, 111, 11927. N. Murakami, R. Abe, O. O. Prieto-Mahaney, T. Torimoto, B. Ohtani, Stud. Surf. Sci. Catal. 2007, 172, 429. B. Ohtani, Chem. Lett. 2008, 37, 216. T. Torimoto, N. Nakamura, S. Ikeda, B. Ohtani, Phys. Chem. Chem. Phys. 2002, 4, 5910.

Table S1. Actual reaction rates obtained for the five photocatalytic systems. Products monitored for the calculation of rates are indicated. n

Photocatalyst

a O2/mmol h-1

b H2/mmol h-1

c CO2/mmol h-1

d CO2/mmol h-1

e PCA/mmol h-1

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35

TIO-1 TIO-2 TIO-3 TIO-4 TIO-5 TIO-6 TIO-7 TIO-8 TIO-9 TIO-10 TIO-11 TIO-12 TIO-13 ST-G2 ST-F1 ST-F2 ST-F3 ST-F4 ST-F5 VP-P90 TUF-01 PC-101 PC-102 Hombikat Merck(A) Aldrich(R) Aldrich(A) Kanto Wako CR-EL ST-01 ST-21 TKP-101 TKP-102 P25

0.003 0.003 0.021 0.030 0.131 0.001 0.004 0.005 0.003 0.008 0.011 0.004 0.017 0.043 0.005 0.029 0.034 0.024 0.017 0.015 0.017 0.012 0.011 0.006 0.014 0.083 0.019 0.020 0.018 0.106 0.006 0.009 0.006 0.009 0.030

0.180 0.086 0.187 0.778 0.505 0.243 0.593 0.635 0.788 0.699 0.520 0.648 0.538 0.674 0.428 0.490 0.563 0.592 0.487 0.601 0.209 0.726 0.575 0.605 0.190 0.675 0.175 0.223 0.927 0.545 0.704 0.465 0.420 0.603 0.859

0.016 0.025 0.007 0.038 0.015 0.007 0.031 0.020 0.028 0.026 0.042 0.028 0.047 0.029 0.027 0.032 0.038 0.051 0.034 0.043 0.013 0.033 0.032 0.030 0.018 0.029 0.028 0.026 0.023 0.017 0.027 0.035 0.024 0.035 0.053

0.220 0.159 0.063 0.272 0.127 0.049 0.257 0.366 0.235 0.291 0.289 0.261 0.323 0.156 0.111 0.129 0.152 0.235 0.224 0.336 0.193 0.271 0.308 0.261 0.143 0.123 0.125 0.231 0.263 0.067 0.339 0.283 0.195 0.301 0.261

0.036 0.009 0.006 0.044 0.012 0.017 0.052 0.052 0.055 0.054 0.054 0.055 0.048 0.015 0.028 0.039 0.041 0.045 0.049 0.059 0.018 0.048 0.048 0.045 0.016 0.015 0.026 0.038 0.042 0.008 0.049 0.050 0.050 0.048 0.041

Table S2. Actual experimental predictor values of titania powders. The variables ANA and RUT represent the presence (1) or absence (0) of the indicated crystalline phase. n

Photocatalyst

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35

TIO-1 TIO-2 TIO-3 TIO-4 TIO-5 TIO-6 TIO-7 TIO-8 TIO-9 TIO-10 TIO-11 TIO-12 TIO-13 ST-G2 ST-F1 ST-F2 ST-F3 ST-F4 ST-F5 VP-P90 TUF-01 PC-101 PC-102 Hombikat Merck(A) Aldrich(R) Aldrich(A) Kanto Wako CR-EL ST-01 ST-21 TKP-101 TKP-102 P25

BET/m2g-1

DEF/μmol g-1

PPS/nm

SPS/μm

73 18 40 50 3 100 270 338 300 100 97 290 59 3.4 19 29 37 39 73 97 114 301 157 300 11 4 8 11 59 7 298 67 306 114 59

109 28 48 50 14 242 119 118 105 106 156 111 72 43 84 112 108 134 186 140 344 111 118 98 25 18 17 38 21 21 84 101 86 94 50

21 400 40 21 570 15 8 4 9.5 15 15 6 30 500 90 60 50 20 20 5 5 8 12 9 169 517 217 169 29 200 8 25 9 20 28

5.61 0.51 2.4 0.6 18.1 2.5 1.6 0.24 0.92 1.38 0.29 0.79 1.02 0.86 0.41 0.36 0.37 0.67 0.39 1.04 2.3 1.6 1.44 1.39 0.44 14.03 3 0.43 0.46 15.42 5.86 6.89 1.39 0.96 0.94

RUT

ANA

0 0 1 1 1 1 0 0 0 0 1 0 0 1 1 1 1 1 1 1 1 0 0 0 0 1 1 0 1 1 0 0 0 0 1

1 1 0 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1

Table S3. Correlation coefficients for 6 properties of titania photocatalysts. property

BET

DEF

PPS

SPS

RUT

ANA

BET

1

0.291

–0.4982

–0.2358

–0.5930

0.0705

DEF

—

1

-0.5194

–0.2980

0.0734

–0.5385

PPS

—

—

1

0.5791

0.2535

0.1471

SPS

—

—

—

1

0.1908

0.0255

RUT

—

—

—

—

1

–0.2976

ANA

—

—

—

—

—

1

Table S4. Correlation coefficients of each property for five kinds of photocatalytic reactions. property

a

b

c

d

e

BET

–0.46

0.34

–0.02

0.52

0.59

DEF

–0.40

–0.14

–0.06

0.15

0.24

PPS

0.69

–0.16

–0.21

–0.49

–0.68

SPS

0.83

–0.01

–0.32

–0.36

–0.46

ANA

0.10

0.44

0.56

0.42

0.46

RUT

0.48

0.04

0.04

–0.51

–0.41