Chapter 2 A New Set of Accurate Multi-level Methods Including Parameterization for Heavy Elements Abstract We have developed a new series of multi-coefficient electronic structure methods that including parameterization for heavy elements. The training set was taken from our MLSE(Cn)-DFT method with 10 additional atomization energies of Br- and Icontaining molecules (Br2, I2, HI, IBr, HBr, ICl, NOBr CH3I, CH3Br, C2H5I ), and ionization potentials and electron affinities of Br and I atoms. Several methods have been developed, we called them MLSE(HAn) methods. The most important new correction term was SCS-MP2(spin component scaled MP2) correction. The best method MLSE(HA-1) gave an average mean unsigned error (MUE) 0.58 kcal/mol on 225 thermochemical kinetics data. It also gave average error less than 1.0 kcal/mol for 10 AEs of Br- and I-containing molecules. In comparison, the MLSE(C1)-DFT gave an MUE of 1.84 kcal/mol on the 10 AEs. The new method MLSE(HA-1) cost approximately 60% more computer time than our previous MLSE(C1)-DFT method. The new set of methods is suitable for thermochemical kinetics study on systems containing the first- and second-row elements as well as on systems containing heavy halogens. 27 Introduction Over past thirty years, the advancement of computer processing capabilities has allowed for more accurate quantum chemical methods to be routinely applied to all areas of chemical studies. The accuracy of quantum chemical calculation is very important in comparing data with experiment and to make quantitative predictions . To approach the high accuracy, one can use very high-level theories, such as QCISD(T) and CCSD(T), with large basis sets, such as G3Large, aug-cc-pVTZ, aug-cc-pVQZ, or even larger basis sets. However, the costs of these methods are prohibitively high except for very small molecules. Finding high accuracy and economic methods is one of the most important goals in the quantum chemical calculation research. In the past two decades, this goal had been realized by the so-called multi-level methods or multi-coefficient methods.1-18 Many of these methods have achieved the traditional goal, that is achieving the so-called “chemical accuracy” of ~1 kcal/mol. In the multi-level methods, additive corrections are applied to a base energy to account for the incomplete treatment of the correlation energies and the incompleteness of the basis-set sizes. One of the first multi-level methods was published at 1985 by Truhlar et al.1 Various methods were published in the following decades, such as Gaussian-n,6-8 CBS,9,10 and Wn methods16-18 and their various variants. After, a series study by Truhlar et al.3,4,13,14 discovered that by using scaled energy components in the multi-level methods, higher accuracy and sometimes higher efficiencies can be achieved. In these so-called multi-coefficient (MC) methods, such as G3S/3,4 G3SX,15 and those in MCCM/3 suite,4 all the scaling factors for various energy components were optimized against experimental dates or high-level theoretical energies. Our group also has developed a series multi-coefficient method, which are called MLSEn5 and MLSEn+d19 in the last decade. However, different 28 with the Pople-type basis sets used in the MCG3 and G3S methods, our methods was developed by using Dunning’s correlation consistent basis sets.20-22 Dunning’s correlation consistent basis sets provide a well-defined hierarchy of basis sets and they were designed for methods that consider high-level electron correlation and for basis-set extrapolation. For neutral systems, our MLSE4+d method gives a mean unsigned error (MUE) of 0.70 kcal/mol for a set of 109 AEs, 0.87 kcal/mol for a set of 38 hydrogen-transfer BHs, and 0.69 for a set of 22 non-hydrogen-transfer BHs. This is compared with the MCG3/3 results of 1.04, 0.90 and 1.05 kcal/mol, respectively. Recently, Truhlar et al. found that incorporating hybrid DFT energies into the multi-coefficient methods would substantially increase the accuracy with only small increases in the computational cost.23 Thus, we also improved our MLSEn+d methods by combining hybrid DFT energies, and the resulting methods were called MLSE-DFT.24 For example, the mean unsigned error (MUE) to a thermochemical kinetics data set of 169 can be improved form 0.91 kcal/mol using MLSE1+d method to 0.61 kcal/mol using MLSE-DFT(MLSE-TPSS1KCIS) method. However, MLSE-DFT methods were also designed for neutral systems. Although MLSE-DFT method can be applied for most chemical reactions that in the neutral conditions, in order to be widely applicable, we also developed another set of MLSE(Cn)-DFT methods25 for charged systems. Augmented with the MP2/aug-cc-pVTZ and MP4SDQ/cc-pVTZ calculations, the MLSE(Cn)-DFT methods performed well both on charged and neutral systems. The efficiency and accuracy of MLSE(Cn)-DFT methods are better than the most existing multi-coefficient methods. The MUE on 211 thermochemical kinetics data was 0.56 kcal/mol using the best method MLSE(C1)-M06-2X. A simplified method MLSE(C2)-M06-2X can achieve similar accuracy (MUE =0.59 ) at 54% of the cost. The most economic method 29 MLSE(C3)-B3LYP gave MUE of 0.62 kcal/mol with 36% of cost only. This is compared with the MUE of 0.73 kcal/mol by the MCG3-MPWB method4,23,24 at cost similar to that of MLSE(C3)- B3LYP. However, all the current multi-coefficient methods were designed and tested for light atoms (from the first- to the third-row elements). There are many important chemical reactions involving heavier atoms on the fourth and fifth rows of the periodic table where the most important ones are the heavy halogens (bromine and iodine). Some of the Pople-type basis sets, that were used in the MCG3 and related methods do not contain basis functions for Br and I atoms. On the other hand, all the halogens are now supported by Dunning’s correlation consistent basis sets, in particular, the aug-cc-pVnZ-pp basis sets. Currently, to our knowledge, there are no multi-level methods available to calculate accurate thermochemical kinetics data involving these heavy halogens. Thus, as a further step to extend the applicability of our methods, in the current study we developed a new set of multi-coefficient methods, called MLSE(HA-n), which includes parameterization for the heavy halogens (Br and I). Recently, a new method called spin-component scaling MP2 (SCS-MP2) was developed by Grimme et al,27-29 which used different scaling factors to the “same spin” and “opposite spin” perturbational terms in the MP2 calculation. The “same spin” energy term consist of long-range, nondynamical effects, and the opposite spin consist of short-range, dynamical correlation. The SCS concept has recently been applied in the double-hybrid DFT method DSD-PLYP by Martin et al.30 By using SCS-MP2 correction term, this new functional achieve very significant improvement on the accuracy over the previous B2GP-PLYP-D31 functional. For example, the root-mean square deviation (RMSD) of the W4-08 database can be improved from 30 3.17 kcal/mol using B2GP-PLYP-D functional to 2.66 kcal/mol using the DSD-PLYP functional.30 The SCS concept has also been extended to higher level ab initio method such as SCS-MP332-34 and SCS-CCSD.35,36 Compared to 48 CCSD(T)/cc-pVQZ reaction energies, the SCS-CCSD performed mean absolute deviation of 1.1 kcal/mol batter than CCSD method of 1.9 kcal/mol.35 Thus, in our most recent development, we also included the SCS-MP2 type corrections in our new methods. In the original work of Grimme, the values for cs and co were fix to 6/5 and 1/3, respectively. In our new methods, the second-order energies (E2) obtained using the cc-pV(D+d)Z, /cc-pV(T+d)Z, aug-cc-pV(D+d)Z, and aug-cc-pV(T+d)Z basis sets were each corrected by two scaling factors (for same-spin and opposite-spin components) which were optimized against our training sets which consist data. 31 experimental of high-level Method To simplify the notation, we first make the following appreviations for Dunning’s correlation-consistent basis sets: pdz cc-pV(D+d)Z apdz aug-cc-pV(D+d)Z ptz cc-pV(D+d)Z aptz aug-cc-p(V+d)Z The “+d” signifies that an additional set of d functions are added to the original correlation-consistent basis sets for the second-row elements.37 We used the MLSE(C1)-M06-2X method we developed previously25 as the starting point for our improved methods: E(MLSE(C1)-M06-2X) = CWF { E(HF/pdz) + C2 [E2/pdz] + C34 [E(MP4SDQ/pdz) – E(MP2/pdz)] + CQCID [E(QCISD/pdz) - E(MP4SDQ/pdZ)] + CQCI [E(QCISD(T)/pdz) – E(MP4SDQ/pdz)] + CB1E2 [MP2/ptz – MP2/pdz] + CHF+ [E(HF/apdz) – E(HF/pdz]) + CE2 [E2/apdz – E2/pdz] + CB2E2 [MP2/aptz – MP2/apdz] + CB1MP4 [E(MP4D/ptz) - E(MP4D/pdz)] } + (1 CWF ) { E(DFT/pdz) + CDFT+ [E(DFT/apdz – DFT/pdz] } (1) Where E(theory/basis set) denotes the single-point Born-Oppenheimer energy 32 calculated with the combination of the particular theory and basis set, and E2 set denotes the second-order energy correction calculated at the MP2 theory using the particular basis set. The DFTX denotes a particular hybrid DFT method with X% of Hartree-Fock exchange energy. The MP4D theory was used instead of the MP4SDQ in the CB1MP4 term because our tests showed that the MP4D theory provided better overall performance at a slightly lower computational cost. The explicit treatment of the spin-orbital coupling was found unnecessary for this method. For the DFTX terms, several hybrid DFT methods, including MPW1B95,38,39 TPSS1KCIS,40,41 MPW1PW91,38 B1B95,39,42 M06-2X43 and B3LYP44,45 were tested in the method. And the M06-2X functional performed best in this method. The coefficients were determined by minimizing the MUEs with respect to a set of 211 accurate thermochemical kinetics data (the “training set”) that listed in Table 1. These data include 109 main-group atomization energies (AEs) from the MGAE109/05 database,47 38 hydrogen-transfer barrier heights (HTBHs), 38 non-hydrogen-transfer barrier heights (NHTBHs) from the HTBH38/04,24 13 ionization potential (IPs) energies and 13 electron affinity (EAs) energies from the IP13/3 and EA13/3 databases,47 respectively. The NHTBH38/04 database values were determined using the W1 theory. All other database values were derived from reliable experimental measurement.48 These databases were compiled by Truhlar and coworkers and were previously used to determine the MCG3DFT and MLSEDFT coefficients.23,24 In the current study, we prepared a new database for training set. The new database consist of the 211 themochemical energies that used in the MLSE(Cn)-DFT25 method and 14 new energies including bromine and iodine elements. The new database called HA-225 database, totally 225 energies. We 33 searched the new data comprehensively in any themochemical database. The experimental errors of new data we will use should less than 1 kJ/mol or 0.24 kcal/mol, which is necessary for used in our highly accurate methods. However, molecules containing 4th and 5th rows elements are mostly difficult to vaporize, others also come with large experimental errors. Therefore, the final data we used are 14 themochemical energies including bromine and iodine elements. These new data are 10 atomization energies of Br and I containing molecules, Br2, I2, HI, IBr, HBr, ICl, NOBr CH3I, CH3Br, and C2H5I , ionization potentials and electron affinities of Br and I. The dissociation energies of Br2 and I2, ionization potentials, and electron affinities of Br and I at 0 K are taken directly from the NIST-JANAF Themochemical Table.49 The energies of the other eight molecules were derived from the 0 experimental ∆H𝑓,298 values26 by removing the zero-point and thermal energies using the same treatment in a previous study26. The new data are listed in Table 2. In the MLSE(Cn)-DFT methods, the structures of training set were obtained using QCISD/MG3S4,50 method. However, the MG3S basis set does not support for Br and I atoms. The structures of Br2, I2, HI, IBr, HBr, ICl, NOBr CH3I, CH3Br, and C2H5I in this study were obtained using QCISD/aug-cc-pVTZ method. According to our experience, the accuracy did not significantly influence by the structures. A pseudo-potential (described as -pp) was used to replace the core electrons of I atom, including cc-pVDZ-pp, cc-pVTZ-pp, aug-cc-pVDZ-pp and aug-cc-pVTZ-pp. We started by testing our previous MLSE(C1)-M06-2X method on the new HA-225 database. The correction of spin-orbital coupling (ESO term) was not used in the original MLSE(C1)-M06-2X method. However, the large spin-orbital coupling effect on these heavy halogen atoms cannot be ignored. The ESO values are 3.5 and 34 7.2 kcal/mol for Br and I atoms, respectively. Thus, the ESO term was added to the MLSE(C1)-M06-2X method and the coefficients have also been readjusted against the same training set. The resulting method is called MLSE(C1)-M06-2X (Eso). Then, we optimized coefficients of the MLSE(C1)-M06-2X (Eso) method by minimizing the overall MUE with respect to HA-225 database and the resulting new method is called MLSE(C1)-M06-2X-HA. However, both MLSE(C1)-M06-2X (Eso) and MLSE(C1)-M06-2X-HA methods performed unsatisfactorily on the 10 heavy halogen containing atomization energies. We also tested other DFT functionals in these methods, but with no significantly improvement. The SCS-MP2 was a new scaling method that was not used in our previous methods. Here we improved the MLSE(C1)-M06-2X method by using different scaling factors to the same spin and opposite spin perturbational terms in the MP2 calculation (cs and co, respectively) , and the resulting method is called the MLSE(C1S)-M06-2X method. The MLSE(C1S)-M06-2X method was performed on the original database (211 themochemical energies), for comparison with MLSE(C1)-M06-2X method, and thus the ESO term was not used in this method. To test the practical limit in accuracy our MLSE methods can achieve, we also developed a new set of methods that include the accurate but also expensive QCISD(T)/aptz energies. The QCISD(T)/aptz method is considered as a reliable method to calculate relative energies in the most chemical reactions. Here we call the new methods the “full” type methods because they include all the energies available obtained after a QCISD(T) calculation using the pdz, apdz, ptz, and aptz basis sets. Thus, these methods used all the energy components obtained from QCISD(T)/pdz, QCISD(T)/ptz, QCISD(T)/apdz and QCISD(T)/aptz calculations. Each calculation 35 give 6 energy components that were HF, MP2, MP4D, MP4SDQ, QCISD and QCISD(T) with a particular basis set. Two methods MLSE(full) and MLSE(full)-M06-2X has been developed. The MLSE(full) method consisted of 24 ab initio energies that consisted of HF, MP2, MP4D, MP4SDQ, QCISD and QCISD(T) methods with pdz, apdz, ptz and aptz basis set. The MLSE(full)-M06-2X method was base on MLSE(full) method with two additional DFT term, M06-2X/pdz and M06-2X/apdz. The formula of MLSE(full) and MLSE(full)-M06-2X methods are: E(MLSE(full) and MLSE(full)-M06-2X) = CWF { E(HF/pdz) + CHF+ [E(HF/apdz) – E(HF/pdz]) + CB1HF [E(HF/ptz) – E(HF/pdz]) + CB2HF [E(HF/aptz) – E(HF/ptz]) + CE2 [E2/pdz] + CE2 [E2/apdz] + CB1E2 [E2/ptz] + CB2E2 [E2/aptz] + CMP4D [E(MP4D/pdz) – E(MP2/pdz)] } + CMP4D+ [E(MP4D/apdz) – E(MP4D/pdz)] } + CB1MP4D [E(MP4D/ptz) – E(MP4D/pdz)] } + CB2MP4D [E(MP4D/aptz) – E(MP4D/apdz)] } + CMP4 [E(MP4SDQ/pdz) – E(MP4D/pdz)] } + CMP4+ [E(MP4SDQ/apdz) – E(MP4SDQ/pdz)] } + CB1MP4 [E(MP4MP4SDQ/ptz) – E(MP4SDQ/pdz)] } + 36 CB2MP4 [E(MP4MP4SDQ/aptz) – E(MP4SDQ/apdz)] } + CQCID [E(QCISD/pdz) – E(MP4SDQ/pdz)] + CQCID+ [E(QCISD/apdz) – E(QCISD /pdz)] + CB1QCID [E(QCISD/ptz) – E(QCISD /pdz)] + CB2QCID [E(QCISD/aptz) – E(QCISD /apdz)] + CQCI [E(QCISD(T)/pdz) – E(QCISD /pdz)] + CQCI+ [E(QCISD(T)/apdz) – E(QCISD(T) /pdz)] + CB1QCI [E(QCISD(T)/ptz) – E(QCISD(T) /pdz)] + CB2QCI [E(QCISD(T)/aptz) – E(QCISD(T) /apdz)] + (1 CWF) CDFT+ { E(M06-2X/pdz) + [E(M06-2X /apdz –M06-2X /pdz] } (2) In the MLSE(full) method, the CWF was fixed to 1, and the DFT terms make no contribution. The coefficients of MLSE(full) and MLSE(full)-M06-2X method were determined by minimizing the MUEs with respect to a set of 211 accurate thermochemical kinetics data (Table 1), the same treatment in MLSE(C1)-M06-2X. Furthermore, we modified these two methods by using SCS-MP2 method. The resulting methods are called MLSE(fullS) and MLSE(fullS)-M06-2X. The formula of MLSE(fullS) and MLSE(fullS)-M06-2X methods are: E(MLSE(fullS) and MLSE(fullS)-M06-2X) = CWF { E(HF/pdz) + CHF+ [E(HF/apdz) – E(HF/pdz]) + CB1HF [E(HF/ptz) – E(HF/pdz]) + 37 CB2HF [E(HF/aptz) – E(HF/ptz]) + CE2S [(E2aa+E2bb)/pdz] + CE2O [(E2ab)/pdz] + CE2+S [(E2aa+E2bb)/apdz] + CE2+O [(E2ab)/apdz] + CB1E2S [(E2aa+E2bb)/ptz] + CB1E2O [(E2ab)/ptz] + CB2E2S [(E2aa+E2bb)/aptz] + CB2E2O [(E2ab)/aptz] + CMP4D [E(MP4D/pdz) – E(MP2/pdz)] } + CMP4D+ [E(MP4D/apdz) – E(MP4D/pdz)] } + CB1MP4D [E(MP4D/ptz) – E(MP4D/pdz)] } + CB2MP4D [E(MP4D/aptz) – E(MP4D/apdz)] } + CMP4 [E(MP4SDQ/pdz) – E(MP4D/pdz)] } + CMP4+ [E(MP4SDQ/apdz) – E(MP4SDQ/pdz)] } + CB1MP4 [E(MP4MP4SDQ/ptz) – E(MP4SDQ/pdz)] } + CB2MP4 [E(MP4MP4SDQ/aptz) – E(MP4SDQ/apdz)] } + CQCID [E(QCISD/pdz) – E(MP4SDQ/pdz)] + CQCID+ [E(QCISD/apdz) – E(QCISD /pdz)] + CB1QCID [E(QCISD/ptz) – E(QCISD /pdz)] + CB2QCID [E(QCISD/aptz) – E(QCISD /apdz)] + CQCI [E(QCISD(T)/pdz) – E(QCISD /pdz)] + CQCI+ [E(QCISD(T)/apdz) – E(QCISD(T) /pdz)] + CB1QCI [E(QCISD(T)/ptz) – E(QCISD(T) /pdz)] + 38 CB2QCI [E(QCISD(T)/aptz) – E(QCISD(T) /apdz)] + (1 - CWF) CDFT+ { E(M06-2X/pdz) + [E(M06-2X /apdz –M06-2X /pdz] } (3) where the CWF was fixed to 1 in the MLSE(fullS) method. The E2aa/basis denotes the alpha-alpha spin perturbational energy correction calculated at the MP2 theory using the particular basis set, and E2ab and E2bb energy terms signify the alpha-beta spin and beta-beta spin respectively. The coefficients of above four methods (MLSE(full), MLSE(full)-M06-2X, MLSE(fullS) and MLSE(fullS)-M06-2X) were determined by minimizing the MUEs with respect to a set of 211 accurate thermochemical kinetics data (Table 1). In order to develop new methods for heavy halogens, the coefficients of these four methods have also been adjusted for the HA-225 database. The simplest modification was the addition of correction energy of spin-orbital coupling (ESO). The resulting methods are called MLSE(full) (Eso), MLSE(full)-M06-2X (Eso), MLSE(fullS) (Eso) and MLSE(fullS)-M06-2X (Eso). The coefficients of these methods were still readjusted against 211 accurate thermochemical kinetics data (Table 1). Then, we optimized the coefficients of these methods against to the HA-225 database. The resulting methods are called MLSE(HA-full), MLSE(HA-full)-M06-2X, MLSE(HA-fullS) and MLSE(HA-fullS)-M06-2X, with the names including “HA”. A similar method, MLSE(HA-fullS)-MPW1, which uses the MPW1PW91 functional were also presented for comparison. The formula for the MLSE(full) (Eso), MLSE(full)-M06-2X (Eso), MLSE(HA-full) and MLSE(HA-full)-M06-2X methods are the same as Eq.(2), and the formula for the MLSE(fullS) (Eso), MLSE(fullS)-M06-2X (Eso), MLSE(HA-fullS) and MLSE(HA-fullS)-M06-2X methods are the same as Eq.(3). 39 The computational cost of MLSE(HA-fullS)-MPW1 method or similar methods that including QCISD(T)/aptz calculation are very large and can only be applied to small systems. The economical, but reasonably accurate methods were needed for larger systems. We tried various combinations of relatively economical calculations, such as MP4/apdz, MP4/ptz and QCISD(T)/apdz. The first version we developed in the current study, called the MLSE(HA-1), was developed by adding the MP4/apdz, QCISD/apdz, QCISD(T)/apdz and SCS-MP2 correction terms to MLSE(C1)-M06-2X method.25 MLSE(HA-1) method used all the energy components obtained from QCISD(T)/pdz, QCISD(T)/apdz, MP4SDQ/ptz and MP2/aptz. The formula of MLSE(HA-1) method is: E(MLSE(HA-1)) = CWF { E(HF/pdz) + CHF+ [E(HF/apdz) – E(HF/pdz]) + CB1HF [E(HF/ptz) – E(HF/pdz]) + CB2HF [E(HF/aptz) – E(HF/ptz]) + CE2S [(E2aa+E2bb)/pdz] + CE2O [(E2ab)/pdz] + CE2+S [(E2aa+E2bb)/apdz] + CE2+O [(E2ab)/apdz] + CB1E2S [(E2aa+E2bb)/ptz] + CB1E2O [(E2ab)/ptz] + CB2E2S [(E2aa+E2bb)/aptz] + CB2E2O [(E2ab)/aptz] + CMP4D [E(MP4D/pdz) – E(MP2/pdz)] } + 40 CMP4D+ [E(MP4D/apdz) – E(MP4D/pdz)] } + CB1MP4D [E(MP4D/ptz) – E(MP4D/pdz)] } + CMP4 [E(MP4SDQ/pdz) – E(MP4D/pdz)] } + CMP4+ [E(MP4SDQ/apdz) – E(MP4SDQ/pdz)] } + CB1MP4D [E(MP4MP4SDQ/ptz) – E(MP4SDQ/pdz)] } + CQCID [E(QCISD/pdz) – E(MP4SDQ/pdz)] + CQCID+ [E(QCISD/apdz) – E(QCISD/pdz)] + CQCI [E(QCISD(T)/pdz) – E(QCISD/pdz)] + CQCI+ [E(QCISD(T)/apdz) – E(QCISD(T)/pdz)] + (1 - CWF) { E(DFT/pdz) + CDFT+ [E(DFT/aptz – DFT/apdz] } + ESO (4) E2aa, E2ab and E2bb signify alpha- alpha spin, alpha-beta spin and beta-beta spin perturbational energy correction terms calculated at the MP2 theory respectively. The treatment of the spin-orbital coupling is explicit (ESO) for selected open-shell species.26 For the DFT terms, several hybrid DFT functionals, including MPW1PW91,38 M06-2X43 and B3LYP44,45 were tested in this method. The MPW1PW91 functional with the apdz/aptz basis set combination was found to provide the best results. The computational cost of MLSE(HA-1) is significantly higher than that of MLSE(C1)-M06-2X because of the expensive QCISD(T)/apdz and MPW1PW91/aptz calculations. One way to lower the cost is to use the MP4SDQ/apdz calculation instead of QCISD(T)/apdz calculation. On the other hand, getting accurate results for the ten heavy halogen-containing atomization energies is still one of the most 41 important goals in the current study. It turns out that using the MP4SDQ/apdz calculation instead of QCISD(T)/apdz calculation was difficult to achieve that goal. Therefore, we gave a weight of three for the MUE of 10 heavy halogens atomization energies in the training set during optimizing the coefficients for the second method. The simplified MLSE(HA-2) method is very similar to MLSE(HA-1). We replaced the QCISD(T)/apdz term with the MP4SDQ/apdz, and replaced the B3LYP term with the pdz/apdz basis set combination. The formula of MLSE(HA-2) method is: E(MLSE(HA-2)) = CWF { E(HF/pdz) + CHF+ [E(HF/apdz) – E(HF/pdz]) + CB1HF [E(HF/ptz) – E(HF/pdz]) + CB2HF [E(HF/aptz) – E(HF/ptz]) + CE2S [(E2aa+E2bb)/pdz] + CE2O [(E2ab)/pdz] + CE2+S [(E2aa+E2bb)/apdz] + CE2+O [(E2ab)/apdz] + CB1E2S [(E2aa+E2bb)/ptz] + CB1E2O [(E2ab)/ptz] + CB2E2S [(E2aa+E2bb)/aptz] + CB2E2O [(E2ab)/aptz] + CMP4D [E(MP4D/pdz) – E(MP2/pdz)] } + CMP4D+ [E(MP4D/apdz) – E(MP4D/pdz)] } + CB1MP4D [E(MP4D/ptz) – E(MP4D/pdz)] } + CMP4 [E(MP4SDQ/pdz) – E(MP4D/pdz)] } + 42 CMP4+ [E(MP4SDQ/apdz) – E(MP4SDQ/pdz)] } + CB1MP4 [E(MP4MP4SDQ/ptz) – E(MP4SDQ/pdz)] } + CQCID [E(QCISD/pdz) – E(MP4SDQ/pdz)] + CQCI [E(QCISD(T)/pdz) – E(QCISD/pdz)] + (1 - CWF) CDFT+ { E(DFT/pdz) + [E(DFT/apdz – DFT/pdz] } + ESO (5) The computational cost of MLSE(HA-2) method was just slightly more than the MLSE(C1)-M06-2X method. The coefficients in Eqs. (45) were determined by minimizing the MUEs with respect to the HA-225 database that consisted of 225 accurate thermochemical kinetics data, which listed in Table 1 and 2. All the calculations were performed using the Gaussian 09 program.51 43 Result and Discussion (a) MLSE(C1S)-M06-2X method In the MLSE(C1S)-M06-2X method, SCS-MP2 correction did not improve the accuracy significantly as compared to the MLSE(C1)-M06-2X method. The overall MUE of MLSE(C1S)-M06-2X method only ~0.01 kcal/mol lower than that of MLSE(C1)-M06-2X method. The MUEs obtained by the MLSE(C1S)-M06-2X were listed in Table 3. The results of MLSE(Cn)-DFT methods also listed for comparison. (b) “full” type methods for 211 thermochemical energies The MLSE(full) and MLSE(full)-M06-2X methods gave very low overall MUEs of 0.55 and 0.50 kcal/mol on the training set (211 energies). These two methods performed best especially for the EAs and NHTBHs, with errors below 0.4 kcal/mol. Adding DFT terms to MLSE(full), gave small improvement of 0.05 kcal/mol. The SCS-MP2 correction did not give significant improvement in MLSE(C1S)-M06-2X method. However, SCS-MP2 played a more important role in the MLSE(fullS) and MLSE(fullS)-M06-2X methods. Compared to the MLSE(full) method (overall MUE = 0.55 kcal/mol), the MLSE(fullS) method predicted more accurate atomization energies and significantly more accurate non-hydrogen-transfer barrier heights. The overall MUE of MLSE(fullS) method is 0.50 kcal/mol and is ~0.05 kcal/mol less than that of MLSE(full) method. The added DFT term in the MLSE(fullS)-M06-2X method also gave a 0.04 kcal/mol improvement on the overall MUE. The overall MUE of MLSE(fullS)-M06-2X method was 0.46 kcal/mol, that is also the best result of these “full” type methods. From the results of these “full” methods, it is reasonable to assume that the best accuracy that can be achieved using this kind methods is ~0.45 kcal/mol. The results of above four methods were also listed in the Table 3. 44 (c) The “full” type methods with ESO term for HA-225 database For the HA-225 database, the overall MUE of 225 energies and 5 types of energy components obtained by the various methods in the current study were listed in Table 4. That were the MUE of 119 AEs, 15 IPs, 15EAs, 38 HTBHs and 38 NHTBHs. Compared to the original training set (211 energies), the added energies were 10 heavy halogen AEs (HHAEs), 2 heavy halogen IPs (HHIPs) and 2 heavy halogen EAs (HHEAs), the MUEs of added energies (HHAE(10), HHIP(2) and HHEA(2)) were also listed in this table. The MLSE(C1)-M06-2X method performed poorly on the HHAE(10) with MUE of of 5.53 kcal/mol. Because the spin-orbital coupling energiest of the open-shell bromine and iodine atoms were not considered in the MLSE(C1)-M06-2X method. The MLSE(C1)-M06-2X (ESO) method performed much better on the HHAE(10) with MUE of 1.84 kcal/mol. The overall MUE of MLSE(C1)-M06-2X (ESO) method is 0.66 kcal/mol, also better than the MLSE(C1)-M06-2X method (MUE = 0.80 kcal/mol). The four “full” type methods MLSE(full), MLSE(full)-M06-2X, MLSE(fullS) and MLSE(fullS)-M06-2X also been tested for HA-225 database. The error of these four methods for the HHAE(10) were 6.73-8.70 kcal/mol before corrected with spin-orbital coupling energies. After corrected for the spin-orbital coupling energies, they gave MUE of 1.50-2.71 kcal/mol on the HHAE(10). The MLSE(fullS)-M06-2X (ESO) method provided the lowest overall MUE of 0.54 kcal/mol on the HA-225 database. The MLSE(full) (ESO), MLSE(full)-M06-2X (ESO) and MLSE(fullS) (ESO) method gave overall MUEs of 0.56-0.60 kcal/mol on the HA-225 database. (d) The re-optimized “full” type methods against HA-225 database 45 Although these methods provided excellent overall MUE on the HA-225 database, the treatment on the HHAE(10), HHIP(2) and HHEA(2) terms was still unsatisfactory, and four methods gave MUEs of 1.50-2.71 kcal/mol on the HHAE(10). One possible reason is that most of the molecules in the training set do not contain heavy halogens. Thus, we re-optimized the coefficients in these methods against the HA-225 database, as described in the previous section. The results were listed in Table 5. Five methods MLSE(C1)-M06-2X-HA, MLSE(HA-full), MLSE(HA-full)-M06-2X, MLSE(HA-fullS) and MLSE(HA-fullS)-M06-2X methods are presented. The treatment of the spin-orbital coupling (ESO) in these methods is explicit. The best MLSE(HA-fullS)-M06-2X method gave 0.51 kcal/mol on the overall MUE(225). Compare to the MLSE(fullS)-M06-2X (Eso) method, the MUE of 10 HHAEs significantly decrease to 1.17 kcal/mol by the MLSE(HA-fullS)-M06-2X method. The overall MUE of MLSE(HA-fullS)-MPW1 method has no significant improvement to the MLSE(HA-fullS)-M06-2X method. However, the MUE of 10 HHAEs decreased from 1.17 kcal/mol by the MLSE(HA-fullS)-M06-2X method to the 0.95 kcal/mol by the MLSE(HA-fullS)-MPW1 method. We have also tested other DFT functionals for HHAEs, but the MPW1PW91 performed best. Therefore, MPW1PW91 functional was used in our new MLSE(HA-1) method. (e) MLSE(HA-1) and MLSE(HA-2) methods The MUEs obtained by the current MLSE(HA-1) and MLSE(HA-2) methods were listed in Table 5. Among the choices of the density functionals, for MLSE(HA-1) method, the methods using the MPW1PW91 performed best, while for MLSE(HA-2) method, the method using the B3LYP functional performed best. The MLSE(HA-1) 46 method provided the lowest overall MUE of 0.58 kcal/mol on the training set. Compared to the MLSE(C1)-M06-2X-HA method, the MLSE(HA-1) method predicted more accurate HHAE and significantly more accurate HHIP and HHEA. The MUE of MLSE(HA-1) method on the 119 AEs, 15 IPs, 15EAs, 38 HTBHs and 38 NHTBHs were 0.57, 0.71, 0.79, 0.46 and 0.65 kcal/mol, respectively. It is remarkable that MLSE(HA-1) method impressively predicted the MUE of 0.87 kcal/mol on the HHAE, even better than the extremely time-consuming MLSE(HA-fullS)-M06-2X and MLSE(HA-fullS)-MPW1 methods. The MLSE(HA-2) method, which is a simplified version of the MLSE(HA-1) method described in the previous section, also gave a satisfactory overall MUE of 0.64 kcal/mol. Compared to the MLSE(HA-1) method, the MLSE(HA-2) method gave similar MUEs on barrier heights, but it gave higher MUE on atomization energies. However, the errors in the IPs and EAs are ~0.1 kcal/mol lower than the MLSE(HA-1) method. Based on the weighting treatment on 10 heavy halogen AEs in optimizing coefficients, MLSE(HA-2) method also provided MUE of 0.98 kcal/mol on the HHAE. In order to further evaluate the new methods developed in the current study, we tested MLSE(HA-1) method on the total noble-gas bond energy (TNGBE) database our laboratory developed recently.52 The database consists of 31 energies corresponding to the Born-Oppenheimer energy of the reaction XNgY X + Ng + Y. This is a very harsh test since the bonding in the noble gas molecules is very special, and none of the energies is included in the training set used in the current study. The MLSE(HA-1) method gave an MUE of 2.71 kcal/mol on the TNGBE database. In comparison, the MP2 and MPW1PW91 methods with aug-cc-pVTZ basis set gave MUEs of 7.4 and 2.9 kcal/mol, respectively. Thus, the new MLSE(HA-1) method 47 developed in this study seem to be robust enough to treat noble gas with reasonable accuracy. (f) Coefficients The optimized coefficients of the MLSE(HA-1) and MLSE(HA-2) method are shown in Table 6. For the two methods, the coefficients of the E2S (same spin) are obviously large than the E2O (opposite spin), the result is consistent with the previous study by Grimme, where the values for cs and co were fix to 6/5 and 1/3, respectively. The DFT energy only contributed to approximately 20% of the total energy in both two methods, that is similar with MLSE(C1)-M06-2X method. The optimized coefficients of other “full” type methods in Table 4 and Table 5 are also shown in Table 7-9. (g) Computational costs Table 10 compares the computational cost and MUEs from various multi-coefficient methods in this study and from recently developed other multi-coefficient methods. The computational cost of MLSE(C1)-M06-2X is equal to the MLSE(C1)-M06-2X-HA method. MLSE(HA-1) method take ~62% more computer time than the MLSE(C1)-M06-2X-HA method. But the MUE of HHAE obtained by MLSE(HA-1) method is remarkably better than the MLSE(C1)-M06-2X-HA method . The MLSE(HA-1) method also provided slightly better accuracy on the original 211 themochemical energies by ~0.03 kcal/mol than the MLSE(C1)-M06-2X-HA method. However, the MLSE(HA-1) method achieved similar accuracy on HHAE but required only 10% of the computational cost of the MLSE(HA-fullS)-MPW1 and similar methods. The MLSE(HA-2) method performed 48 0.98 kcal/mol on the HHAE(10) and required only 4% more cost than the MLSE(C1)-M06-2X-HA method. The MLSE(HA-2) method also gave a satisfactory overall MUE of 0.64 kcal/mol for HA-225 database, better than the MLSE(C1)-M06-2X-HA method. The MLSE(HA-2) method is comparable in cost to the MLSE(C1)-M06-2X-HA method but achieves higher accuracy. Consequently, the MLSE(HA-2) method is another economical alternative to the MLSE(HA-1) method, especially for larger systems. 49 Concluding Remarks In this chapter, we prepared a new database, called HA-225 database. They were including 211 energies that were used in our previous MLSE(Cn)-DFT method and with 10 additional atomization energies of Br- and I- containing molecules (Br2, I2, HI, IBr, HBr, ICl, NOBr CH3I, CH3Br, C2H5I ), and ionization potentials and electron affinities of Br and I atoms. Two sets of methods have been developed. In order to test the practical limit in accuracy our MLSE methods can achieve, we developed a new set of methods that include the accurate but also expensive QCISD(T)/aptz energies, called the “full” methods. From the results of these “full” methods, it is reasonable to assume that the best accuracy that can be achieved using this kind methods is ~0.45 kcal/mol. In addition, we developed two more economical methods, MLSE(HA-1) and MLSE(HA-2). The MLSE(HA-1) method gave an average mean unsigned error (MUE) 0.58 kcal/mol on HA-225 database. It also gave average error less than 1.0 kcal/mol for 10 AEs of Br- and I-containing molecules. In comparison, the MLSE(C1)-DFT gave an MUE of 1.84 kcal/mol on the 10 AEs. The new method MLSE(HA-1) cost approximately 60% more computer time than our previous MLSE(C1)-DFT method. However, the second method, MLSE(HA-2) method, performed 0.98 kcal/mol on the 10 AEs and required only 4% more cost than the MLSE(C1)-M06-2X-HA method. The MLSE(HA-2) method also gave a satisfactory overall MUE of 0.64 kcal/mol for HA-225 database. We expect that the new methods can easily be applied to many types of interesting chemical systems containing heavy halogens, and they will be invaluable for accurate study of thermochemistry and kinetics. 50 Acknowledgements This work is supported by the National Science Council of Taiwan, grant number NSC 100-2113-M-194-007. We are grateful to the National Center for High-Performance Computing (NCHC) for providing part of computation resources. 51 References (1) Brown, F.B.; Truhlar, D.G. Chem. Phys. Lett. 1985, 117, 307. (2) Gordon, M.S.; Truhlar, D.G. J. Am. Chem. Soc. 1986, 108, 5412. (3) Fast, P.L.; Corchado, J.C.; Sánchez, M.L.; Truhlar, D.G. J. Phys. Chem. A 1999, 103, 5129. (4) Lynch, B.J.; Truhlar, D.G. J. Phys. Chem. A 2003,107, 3898. (5) Li, T.-H.; Mou, C.-H.; Hu, W.-P. Chem. Phys. Lett. 2004, 397, 364. (6) Curtiss, L.A.; Raghavachari, K.; Trucks, G.W.; Pople, J.A. J. Chem. Phys. 1991, 94, 7221. (7) Curtiss, L.A.; Raghavachari, K.; Redfern, P.C.; Rasslov, V.; Pople, J.A. J. Chem. Phys. 1998, 109, 7764. (8) Curtiss, L.A.; Redfern, P.C.; Raghavachari, K. J. Chem. Phys. 2007, 126, 84108. (9) Montgomery, J.A., Jr.; Frisch, M. J.; Ochterski, J.W.; Petersson, G.A. J. Chem. Phys. 1999, 110, 2822. (10) Montgomery, J.A., Jr.; Frisch, M. J.; Ochterski, J.W.; Petersson, G.A. J. Chem. Phys. 2000, 112, 6532. (11) Parthiban, S.; Martin, J.M.L. J. Chem. Phys. 2001, 114, 6014. (12) Boese, A.D.; Oren, M.; Atasoylu, O.; Martin, J.M.L. J. Chem. Phys. 2004, 120, 4129. (13) Tratz, C.M.; Fast, P.L.; Truhlar, D.G. Phys. Chem. Comm. 1999, 2, (article 14). (14) Fast, P.L.; Truhlar, D.G.; J. Phys. Chem. A 2000, 104, 6111. (15) Curtiss, L.A.; Redfern, P.C.; Raghavachari, K.; Pople, J.A. J. Chem. Phys. 2001, 114, 108. (16) Martin, J.M.L.; de Oliveira, G. J. Chem. Phys. 1999, 111, 1843. 52 (17) Karton, A.; Rabinovich, E.; Martin, J.M.L.; Ruscic, B. J. Chem. Phys. 2006, 125, 144108. (18) Karton, A.; Taylor, P.R.; Martin, J.M.L. J. Chem. Phys. 2007, 127, 064104. (19) Li, T.-H. ; Chen, H.-R.; Hu, W.-P. Chem. Phys. Lett. 2005, 412, 430. (20) Dunning, T.H., Jr. J. Chem. Phys. 1989, 90, 1007. (21) Kendall, R.A.; Dunning, T.H., Jr.; Harrison, R.J. J. Chem. Phys. 1992, 96, 6796. (22) Woon, D.E.; Dunning, T.H., Jr. J. Chem. Phys. 1993, 98, 1358. (23) Zhao, Y.; Lynch, B.J.; Truhlar, D.G. Phys. Chem. Chem. Phys. 2005, 7, 43. (24) Sun, Y.-L.; Li, T.-H.; Chen, J.-L.; Wu, K.-J.; Hu, W.-P. Chem. Phys. Lett. 2007, 442, 220. (25) Sun, Y.-L.; Li, T.-H.; Chen, J.-L.; Wu, K.-J.; Hu, W.-P. Chem. Phys. Lett. 2009, 475, 141. (26) Fast, P.L.; Corchado, J.; Sanchez, M.L.; Truhlar, D.G. J. Phys. Chem. A 1999, 103, 3139. (27) Schwabe, T.; Grimme, S. Acc. Chem. Res. 2008, 41, 569. (28) Grimme, S. J. Chem. Phys. 2003, 118, 9095. (29) Szabados, A. J. Chem. Phys. 2006, 125, 214105. (30) Kozuch,S.; Gruzman, D.; Martin, J.M.L. J. Phys. Chem. C 2010, 114, 20801. (31) Karton, A.; Tarnopolsky, A.; Lamere, J.-F.; Schatz, G.C.; Martin, J.M.L. J. Phys. Chem. A 2008, 112, 12868. (32) Silva, V.D.; Dias, R.P.; Rocha, W.R. Chem. Phys. Lett. 2007, 439, 69. (33) Grimme, S. J. Comput. Chem. 2003, 24, 1529. (34) Hyla-Kryspin, I.; Grimme, S. Organometallics 2004, 23, 5581. (35) Takatani, T.; Hohenstein, E.G.; Sherrill, C.D. J. Chem. Phys. 2008, 128, 53 124111. (36) Hellweg, A.; Grün, S. A.; Hättig, C. Phys. Chem. Chem. Phys. 2008, 10, 4119. (37) Dunning, T.H., Jr.; Peterson, K.A.; Wilson, A.K. J. Chem. Phys. 2001, 114, 9244. (38) Adamo, C.; Barone, V. J. Chem. Phys. 1998, 108, 664. (39) Becke, A.D. J. Chem. Phys. 1996, 104, 1040. (40) Tao, J.; Perdew, J.P.; Staroverov, V.N.; Scuseria, G.E. Phys. Rev. Lett. 2003, 91, 146401. (41) Toulouse, T.; Savin, A.; Adamo, C. J. Chem. Phys. 2002, 117, 10465. (42) Becke, A.D. Phys. Rev. A 1988, 38, 3098. (43) Zhao, Y.; Truhlar, D.G. Theor. Chem. Acc. 2008, 120, 215. (44) Becke, A.D. J. Chem. Phys. 1993, 98, 1372. (45) Becke, A.D. J. Chem. Phys. 1993, 98, 5648. (46) Zhao, Y.; Schultz, N.E.; Truhlar, D.G. J. Chem. Theory Comput. 2006, 2, 364. (47) Lynch, B.J.; Zhao, Y.; Truhlar, D.G. J. Phys. Chem. A 2003, 107, 1384. (48) Fast, P.L.; Sánchez, M.L.; Corchado, J.C.; Truhlar, D.G. J. Chem. Phys. 1999, 110, 11679. (49) Chase, Jr., M.W. NIST-JANAF Themochemical Tables, Fourth Edition. (50) Zhao, Y.; González-García, N.; Truhlar, D.G. J. Phys. Chem. A 109,2005, 2012. (51) Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb, M. A.; Cheeseman, J. R.; Montgomery, J. A., Jr.; Vreven, T.; Kudin, K. N.; Burant, J. C.; Millam, J. M.; Iyengar, S. S.; Tomasi, J.; Barone, V.; Mennucci, B.; Cossi, M.; Scalmani, G.; Rega, N.; Petersson, G. A.; Nakatsuji, H.; Hada, M.; Ehara, M.; Toyota, K.; Fukuda, R.; Hasegawa, J.; Ishida, M.; Nakajima, T.; Honda, Y.; Kitao, O.; Nakai, H.; Klene, M.; Li, X.; Knox, J. E.; Hratchian, H. P.; Cross, J. 54 B.; Bakken, V.; Adamo, C.; Jaramillo, J.; Gomperts, R.; Stratmann, R. E.; Yazyev, O.; Austin, A. J.; Cammi, R.; Pomelli, C.; Ochterski, J. W.; Ayala, P. Y.; Morokuma, K.; Voth, G. A.; Salvador, P.; Dannenberg, J. J.; Zakrzewski, V. G.; Dapprich, S.; Daniels, A. D.; Strain, M. C.; Farkas, O.; Malick, D. K.; Rabuck, A. D.; Raghavachari, K.; Foresman, J. B.; Ortiz, J. V.; Cui, Q.; Baboul, A. G.; Clifford, S.; Cioslowski, J.; Stefanov, B. B.; Liu, G.; Liashenko, A.; Piskorz, P.; Komaromi, I.; Martin, R. L.; Fox, D. J.; Keith, T.; Al-Laham, M. A.; Peng, C. Y.; Nanayakkara, A.; Challacombe, M.; Gill, P. M. W.; Johnson, B.; Chen, W.; Wong, M. W.; Gonzalez, C.; Pople, J. A. Gaussian03, revision D02; Gaussian, Inc.: Wallingford, CT, 2004. (52) Lai, T.-Y.; Yang, C.-Y.; Lin, H.-J.; Yang, C.-Y.; Hu, W.-P. J. Chem. Phys. 2011, 134, 244110 (53) Zheng, J.; Zhao, Y.; Truhlar, D.G. J. Chem. Theory Comput. 2007, 3, 569. 55 Table 1 211 thermochemical kinetics data for the training sets of MLSE(Cn)-DFT methods (kcal/mol) a. AEs CH 84.0 H2C2H 445.8 CH3Cl 394.6 CH3SH 473.8 HCOOH 501.0 C2H4O 650.7 OH 107.1 SH 87.0 F2 38.2 HOCl 164.4 NF3 204.5 CH3OCH3 798.1 NO 152.1 C2H5 603.8 N2 228.5 SO2 257.9 PF3 363.9 H3CCH2OH 810.4 CN 180.6 CH3CO 581.6 PH3 242.6 C2Cl4 466.3 BCl3 322.9 C3H4_a 704.8 ClO 64.5 (CH3)3C 1199.3 SC 171.3 C2F4 589.4 BF3 470.0 H3CCOOH 803.0 O2 120.0 (CH3)2CH 900.8 CH4 420.1 C4H4O 993.7 AlCl3 306.3 H3CCOCH3 978.0 SO 125.0 C2H5O 698.6 NH3 297.9 C4H6_d 987.2 SiCl4 384.9 H3CCHCH2 860.6 S2 101.7 FH 141.1 H2O 232.6 C4H6_e 1001.6 AlF3 426.5 C2H5OCH3 1095.1 NH 83.7 ClH 106.5 SH2 182.7 CCl4 312.7 SiF4 574.4 C4H10_h 1303.0 CH2 (T) 190.7 Cl2 58.0 CH2 181.4 CF3CN 639.9 C4H4S 962.7 C4H10_i 1301.3 CH3 307.5 CO2 389.1 SiH2 151.8 CF4 476.3 C4H5N 1071.6 C4H8_j 1149.0 56 (continue) NH2 181.9 HCN 313.2 C2H2 405.4 CH3CN 615.8 C5H5N 1237.7 C4H8_k 1158.6 SiH2 (T) 131.1 ClF 61.4 C2H6 712.8 CH3NH2 582.6 C3H4_g 682.7 C5H8_l 1284.3 SiH3 227.4 SiH4 322.4 CO 259.3 CH3NO2 601.3 ClF3_2 125.3 C6H6 1367.6 PH2 153.2 H2O2 268.6 H2CO 373.7 CHCl3 343.2 C4H6_b 1012.4 H2CCO 532.3 HCO 278.4 P2 117.1 CH3OH 512.8 CHF3 457.5 C4H6_c 1004.1 C3H4_f 703.2 Si2_3 75.0 SiO 192.1 N2H4 438.6 H2 109.5 HCOCOH 633.4 C3H8 1006.9 C2H 265.1 C2H4 563.5 Si2H6 530.8 HCOOCH3 785.3 CH3CHO 677.0 C3H6 853.4 CH2OH 409.8 57 b. IPs c. EAs C 259.7 C 29.1 S 238.9 S 47.9 SH 238.9 SH 53.3 Cl 299.1 Cl 83.4 Cl2 265.3 Cl2 55.6 OH 299.1 OH 42.1 O 313.9 O 33.7 O2 278.9 O2 10.8 P 241.9 P 17.2 PH 234.1 PH 23.2 PH2 226.3 PH2 29.4 S2 216.0 S2 38.5 Si 187.9 Si 31.9 Br 272.4 Br 77.6 I 241.0 I 70.5 58 d. HTBHs R1FBH 8.7 R10RBH 13.7 R1RBH 5.7 R11FBH 3.1 R2FBH 5.1 R11RBH 24.2 R2RBH 21.2 R12FBH 10.7 R3FBH 12.1 R12RBH 13.1 R3RBH 15.3 R13FBH 3.5 R4FBH 6.7 R13RBH 17.3 R4RBH 19.6 R14FBH 9.8 R5FBH 9.6 R14RBH 10.4 R5RBH 9.6 R15FBH 22.4 R6FBH 3.2 R15RBH 8.0 R6RBH 12.7 R16FBH 18.3 R7FBH 1.7 R16RBH 7.5 R7RBH 7.9 R17FBH 10.4 R8FBH 3.4 R17RBH 17.4 R8RBH 19.9 R18FBH 14.5 R9FBH 1.8 R18RBH 17.8 R9RBH 33.4 R19FBH 38.4 R10FBH 8.1 R19RBH 38.4 59 e. NHTBHs H + N2O→OH + N2 18.1 HCN→HNC 48.2 REV. BH 83.2 REV. BH 33.1 H + FH→HF+H 42.2 F + CH3F→FCH3 + F -0.34 REV. BH 42.2 REV. BH -0.34 H + ClH→HCl +H 18.0 F...CH3F→FCH3 ... F 13.38 REV. BH 18.0 REV. BH 13.38 H + FCH3→HF+ CH3 30.4 Cl + CH3Cl→ClCH3 + Cl 3.1 REV. BH 57.0 REV. BH 3.1 H + F2→HF+F 2.3 Cl...CH3Cl→ClCH3...Cl 13.61 REV. BH 105.3 REV. BH 13.61 CH3 + FCl→CH3F+ Cl 7.4 F + CH3Cl→FCH3 + Cl -12.54 REV. BH 60.2 REV. BH 20.11 H + N2→HN2 14.7 F...CH3Cl→FCH3...Cl 2.89 REV. BH 10.7 REV. BH 29.62 H + CO→HCO 3.2 OH + CH3F→HOCH3 + F -2.78 REV. BH 22.7 REV. BH 17.33 H + C2H4→CH3CH2 1.7 OH...CH3F→HOCH3...F 10.96 REV. BH 41.8 REV. BH 47.2 CH3 + C2H4→CH3CH2CH2 6.9 REV. BH 33.0 60 Table 2 New data in the HA-225 database (kcal/mol) AE I2 35.87 HI 73.79 IBr 42.27 ICl 50.19 Br2 45.90 HBr 90.51 NOBr 181.64 CH3I 369.12 CH3Br 380.94 C2H5I 662.69 IP I 241.01 Br 272.43 EA I 70.54 Br 77.60 61 Table 3 Mean unsigned errors (kcal/mol) for the original 211 energies IP EA MLSE(full) MLSE(full)-M06-2X MLSE(fullS) MLSE(fullS)-M06-2X 0.62 0.62 0.65 0.57 0.57 0.52 0.55 0.54 0.75 0.59 0.77 0.55 0.63 0.48 0.34 0.37 0.36 0.29 0.47 0.5 0.46 0.45 0.46 0.48 0.43 0.43 0.39 0.37 0.31 0.30 0.56 0.55 0.55 0.50 0.50 0.46 MLSE-TSa MCG3/3b MCG3-MPWBb G3SXb 0.62 1.04 0.75 0.85 0.95 0.67 1.07 0.92 0.86 1.06 0.55 0.84 0.54 0.67 0.69 1.00 0.84 0.60 0.61 0.98 0.73 0.80 MLSE(C1)-M06-2X MLSE(C1S)-M06-2X aFor neutral systems only. bObtained from ref. [4,24,53]. 62 HTBH NHTBH Overall MUE AE Table 4 Mean unsigned errors (kcal/mol) of several methods for the HA-225 database AE(119) IP(15) EA(15) HTBH(38) NHTBH(38) MUE(225) HHAE(10) MLSE(C1)-M06-2X 1.03 0.55 0.97 0.47 0.44 0.80 5.53 MLSE(C1)-M06-2X (Eso) 0.74 0.68 0.82 0.52 0.49 0.66 1.84 HHIP(2) HHEA(2) 0.61 1.22 3.07 2.21 MLSE(full) MLSE(full) (Eso) MLSE(full)-M06-2X MLSE(full)-M06-2X (Eso) MLSE(fullS) MLSE(fullS) (Eso) MLSE(fullS)-M06-2X 1.30 0.70 1.25 0.75 1.23 0.65 1.04 0.84 0.72 0.63 0.51 0.85 0.70 0.61 1.24 0.81 0.98 0.69 1.24 0.79 1.00 0.46 0.45 0.45 0.41 0.46 0.44 0.48 0.39 0.42 0.37 0.38 0.31 0.37 0.30 0.95 0.60 0.89 0.60 0.90 0.56 0.77 8.43 1.64 8.70 2.71 8.51 1.50 6.73 1.47 0.62 0.90 0.43 1.41 0.55 0.96 4.42 1.36 3.55 1.79 4.32 1.16 3.90 MLSE(fullS)-M06-2X (Eso) 0.65 0.52 0.68 0.44 0.32 0.54 1.85 0.37 1.59 63 Table 5 Mean unsigned errors (kcal/mol) of two new methods and several related methods for the HA-225 database AE(119) IP(15) EA(15) HTBH(38) NHTBH(38) MUE(225) HHAE(10) MLSE(C1)-M06-2X-HA 0.72 0.68 0.86 0.54 0.49 0.66 1.66 HHIP(2) HHEA(2) 2.30 1.44 1.64 1.16 1.53 1.06 1.07 1.04 MLSE(HA-full) 0.65 0.78 0.88 0.47 0.44 0.59 1.35 1.21 0.64 MLSE(HA-full)-M06-2X MLSE(HA-fullS) MLSE(HA-fullS)-M06-2X MLSE(HA-fullS)-MPW1 MLSE(HA-1) MLSE(HA-2) 0.61 0.65 0.58 0.57 0.57 0.65 0.59 0.71 0.51 0.70 0.71 0.61 0.75 0.79 0.66 0.77 0.79 0.68 0.45 0.44 0.47 0.40 0.46 0.49 0.44 0.36 0.33 0.41 0.65 0.69 0.55 0.56 0.51 0.51 0.58 0.64 1.17 1.38 1.17 0.95 0.87 0.98 0.47 0.54 0.35 0.48 0.49 0.48 64 Table 6 Optimized coefficients of the MLSE(HA-n) methods MLSE(HA-1) MLSE(HA-2) CWF CHF+ CB1HF CB2HF CE2S CE2O CE2+S CE2+O 0.786928 1.459069 0.589013 -1.005416 1.957262 0.314312 -2.013948 0.422561 0.829308 1.069346 1.325074 -1.108107 1.951836 0.431388 -2.370384 0.201030 CB1E2S CB1E2O CB2E2S -1.876080 0.659927 2.686884 -1.722345 0.385396 3.004871 CB2E2O CMP4D CMP4D+ CB1MP4D CMP4 CMP4+ -0.276516 0.597872 -1.813549 2.685780 0.904651 0.095762 0.025068 0.685108 -0.661122 1.773237 0.705375 0.575352 CB1MP4 CQCID CQCID+ CQCI CQCI+ CDFT+ -1.619272 1.447285 -0.922835 0.667287 2.295483 3.333852 -0.616310 1.302820 0.898955 1.379385 65 Table 7 Optimized coefficients of the MLSE(full) and related methods MLSE(full) MLSE(full) (Eso) MLSE(full) -M06-2X MLSE(full) -M06-2X (Eso) CWF CHF+ CB1HF CB2HF CE2 CE2+ 1 1.037829 0.493843 -1.126519 1.003176 0.000245 1 1.070880 0.520758 -0.823689 0.819161 -0.137225 0.870925 0.650580 0.675219 -1.140078 1.072170 0.126935 0.878519 0.541834 0.528563 -0.675190 0.991038 -0.147319 CB1E2 CB2E2 CMP4D -0.844154 -0.844154 0.829441 -0.761518 -0.761518 1.049052 -0.875105 0.677745 0.899126 1.037470 -0.887498 0.892608 CMP4D+ CB1MP4D CB2MP4D CMP4 CMP4+ CB1MP4 -0.108677 0.571477 -0.168946 1.010427 -0.544262 0.475003 0.204636 0.341153 -0.160930 1.007510 -0.826130 0.076156 -0.109476 0.801761 -0.183723 1.039671 -0.245242 0.721351 0.320543 0.894578 -0.138093 0.973623 -0.639717 0.271239 CB2MP4 CQCID CQCID+ CB1QCID CB2QCID CQCI CQCI+ CB1QCI CB2QCI CDFT+ -2.391712 1.111177 3.031734 -0.068526 0.043619 1.171872 -2.883487 0.770581 2.740099 -2.704330 1.081755 2.809705 -0.265190 0.017438 1.197801 -2.691017 1.296029 3.012421 -2.154875 1.061348 2.828960 -0.355317 0.127422 1.352000 -2.919135 0.731192 2.313722 2.874405 -2.821769 1.110156 2.539067 -0.627058 0.108841 1.141174 -2.642808 1.018256 3.089372 2.602210 66 Table 8 Optimized coefficients of the MLSE(fullS) and related methods MLSE(fullS) MLSE(fullS) (Eso) MLSE(fullS) -M06-2X MLSE(fullS) -M06-2X (Eso) CWF CHF+ CB1HF CB2HF CE2S CE2O CE2+S 1 0.883401 0.703184 -0.913475 1.382839 0.596671 -0.275642 1 0.612080 0.698557 -0.600850 1.951113 0.423185 -1.470842 0.883179 0.416649 0.674666 -0.733691 2.131216 0.440984 -1.306096 0.879725 0.418571 0.649687 -0.592647 2.041226 0.436953 -1.762987 CE2+O CB1E2S CB1E2O 0.032581 -1.173935 -0.305314 -0.066521 -2.123908 0.331396 0.268186 -2.924380 0.368213 0.262996 -2.580876 0.248282 CB2E2S CB2E2O CMP4D CMP4D+ CB1MP4D CB2MP4D 0.996106 0.645705 0.830052 0.048356 -0.161826 -0.285691 2.535541 0.286811 0.814974 0.150407 0.009809 -0.147972 2.962292 -0.075049 0.816386 0.323032 -0.302436 -0.239473 3.134308 0.072893 0.826959 0.371982 0.066923 -0.125712 CMP4 CMP4+ CB1MP4 CB2MP4 CQCID CQCID+ CB1QCID CB2QCID CQCI CQCI+ 0.900704 -0.111807 -0.057520 -2.251636 1.039860 2.813779 -0.554685 0.052197 1.209001 -3.023357 0.879761 0.073514 -0.748149 -2.118263 1.066183 2.299143 -0.837736 -0.027041 1.104021 -2.488856 0.947535 0.124038 0.034873 -2.162555 1.052334 2.465793 -0.767259 0.014413 1.199952 -3.034517 0.927961 0.018081 -0.350364 -2.452403 1.085032 2.306737 -1.010055 0.065826 1.126697 -2.872572 CB1QCI CB2QCI CDFT+ 2.200251 2.521614 2.618276 2.439316 2.413633 2.509004 2.220362 2.595462 2.767826 2.263888 67 Table 9 Optimized coefficients of the MLSE(HA-full), MLSE(HA-fullS) and related methods MLSE (HA-full) MLSE (HA-full) -M06-2X MLSE (HA-fullS) MLSE (HA-fullS) -M06-2X MLSE (HA-fullS) -MPW1 1 1.266598 0.254205 -0.728768 0.898200 0.980842 0.486492 -0.711772 1 0.650849 0.710478 -0.619468 0.875028 0.565250 0.823825 -0.673838 0.847181 0.805937 0.567011 -0.712333 0.579623 0.584063 1.910358 1.870446 2.315592 0.390446 0.361555 0.176064 0.298863 0.258466 -1.384052 -1.648329 -2.132401 -0.005331 0.196881 0.675334 0.609823 0.582610 -2.076050 -2.413309 -2.192308 0.382770 0.385576 0.488093 -0.525417 -0.448739 2.436055 3.007931 2.750775 0.209358 0.066486 -0.256456 CMP4D CMP4D+ CB1MP4D 0.909915 -0.166268 0.720742 0.946587 0.011374 0.960885 0.814464 0.104189 -0.014380 0.833752 0.139293 0.305951 0.740696 -0.390761 0.882816 CB2MP4D CMP4 CMP4+ CB1MP4 CB2MP4 CQCID CQCID+ CB1QCID CB2QCID CQCI -0.187253 1.069257 -1.046107 -0.247101 -2.333597 1.021001 2.919915 -0.547524 0.013474 1.300890 -0.153381 1.156688 -0.863085 -0.305178 -2.448408 1.004261 2.701222 -0.743730 0.023410 1.305167 -0.150273 0.887121 0.085833 -0.828743 -2.080947 1.059439 2.280464 -0.769102 -0.036968 1.126684 -0.154377 0.991543 0.046032 -0.466325 -2.472926 1.050220 2.631438 -1.504552 0.019759 1.167994 -0.080397 0.740535 0.747883 -1.203428 -2.219091 1.177723 1.722150 -1.774898 -0.010091 0.810339 CQCI+ CB1QCI CB2QCI CDFT+ -2.641468 1.706159 2.902419 -2.632944 1.612230 2.897620 2.797512 -2.459164 2.659502 2.405594 -2.956299 2.843523 2.698097 2.323216 -2.187778 3.159057 2.522848 2.936087 CWF CHF+ CB1HF CB2HF CE2S CE2O CE2+S CE2+O CB1E2S CB1E2O CB2E2S CB2E2O 68 Table 10 Computational costa costb (sec.) MUE(211) MUE(225) HHAE(10) MLSE(C1)-M06-2X-HA MCG3-MPWB MCG3/3 MLSE(HA-fullS) MLSE(HA-fullS)-M06-2X MLSE(HA-fullS)-MPW1 MLSE(HA-1) MLSE(HA-2) 1276 514 392 19271 19323 19599 2071 1324 aDetermined 0.59 0.73c 0.98c 0.51 0.47 0.49 0.56 0.62 0.66 N/A N/A 0.59 0.51 0.51 0.58 0.64 1.66 N/A N/A 1.38 1.17 0.95 0.87 0.98 using computers with an Intel i7 2600K CPU at 4.5GHz and a version of Gaussian 03 program optimized for the hardware architecture. b Total CPU time obtained from the calculation for C H N, C Cl , C H O, C H S, 5 5 2 4 4 4 4 4 C4H5N, CF3CN, and SiCl4. cObtained from ref. [4,24]. 69