Download presentation
1
指導教授:胡維平 (Wei-Ping Hu)
國立中正大學 化學暨生物化學研究所 碩士論文口試 孫翊倫 (Yi-Lun Sun) 指導教授:胡維平 (Wei-Ping Hu) 中華民國97年7月23日
2
Content Chapter 1 Accurate Multi-Level Electronic Structure Methods (MLSE-DFT) for Atomization Energies and Reaction Energy Barriers in neutral system Chapter 2 Accurate Multi-Level Electronic Structure Methods, ML(Cn)-DFT for Atomization Energies and Reaction Energy Barriers Chapter 3 Novel Noble Gas Compound 2017年4月25日 碩士論文口試
3
Abstract We have developed a set of new multi-level electronic structure methods by including energies calculated from several density functional theory methods. The parameterization of the improved methods MLSE-DFT was based on updated databases of 109 atomization energies, 38 hydrogen-transfer barrier heights, and 22 neutral non-hydrogen-transfer reaction barrier heights. The best method, MLSE-TPSS1KCIS, performed impressively on the above three types of energies with mean unsigned errors of 0.62, 0.55, and 0.69 kcal/mol, respectively. We found that the hybrid versions of DFT are not absolutely necessary, and the performance can be improved significantly using two different basis sets in DFT calculation. 2017年4月25日 碩士論文口試
4
Quantum Chemical Calculations
Electron correlation → HF MP MP MP QCISD(T) … Full CI Basis set Type Minimal Split-valence Polarized Diffuse High ang. momentum … … … … … … … HF Limit Schrödinger Equation ∞ 2017年4月25日 碩士論文口試
5
Single Level Methods For example: Deficiencies: MP2/aug-cc-pVDZ
QCISD(T)/aug-cc-pVTZ Deficiencies: Low accuracy MP2/aug-cc-pVDZ: generally more than 5 kcal/mol error. QCISD(T)/aug-cc-pVTZ: generally more than 1 kcal/mol error. Cost expensive The QCISD(T)/aug-cc-pVTZ is more than 100 times the cost of MP2/cc-pVDZ for medium molecules. 2017年4月25日 碩士論文口試
6
Multilevel Methods Base Calculation + Corrections for
Incomplete Basis Set Incomplete Electron Correlation Currently Used Multilevel Methods: G2, G3, G4, CBS HF, MP2, MP4, QCISD(T), empirical HLC 6-31G(d), 6-311G(d,p), G(d,p) 6-311+G(2df,p), G(3df,2p), G3Large Multilevel Methods with Scaled Energies: (Multicoefficient Method) MCG3, G3S, G3X 2017年4月25日 碩士論文口試
7
G1 theory E(G1)= Ebase+ ΔE+ +ΔE2df,p + ΔEQCI + ΔEHLC +EZPE
Geometry:MP2(full)/6-31G(d) Ebase : MP4/6-311G(d,p) ΔE+ : MP4/6-311+G(d,p) – Ebase ΔE2df,p : MP4/6-311G(2df,p) – Ebase ΔEQCI : QCISD(T)/6-311G(d,p) – Ebase ΔEHLC : – nα – nβ EZPE : ZPE(HF /6-31G(d)) × E(G1)= Ebase+ ΔE+ +ΔE2df,p + ΔEQCI + ΔEHLC +EZPE Journal of Chemical Physics, 1990, 93, 2017年4月25日 碩士論文口試
8
G2 theory E(G2) = E(G1)+ ΔE+2df – ΔE+ – ΔE2df + Δ3d2p + ΔEHLC
ΔE+2df :MP2/6-311+G(2df,p) – MP2/6-311G(d,p) Δ+ : MP2/6-311+G(d,p) – MP2/6-311G(d,p) ΔE2df : MP2/6-311G(2df,p) – MP2/6-311G(d,p) Δ3d2p : MP2/6-311+G(3df,2p) – MP2/6-311+G(2df,p) ΔEHLC : nβ E(G2) = E(G1)+ ΔE+2df – ΔE+ – ΔE2df + Δ3d2p + ΔEHLC Journal of Chemical Physics, 1991, 94, 2017年4月25日 碩士論文口試
9
G3 theory Geometry:MP2(full)/6-31G(d) Ebase : MP4/6-31G(d) ΔE+ : MP4/6-31+G(d) - Ebase Δ E2df,p : MP4/6-31G(2df,p) – Ebase Δ EQCI : QCISD(T)/6-31G(d) – Ebase Δ EG3Large : MP2(full)/G3Large – [ MP2/6-31G(2df,p) +MP2/6-31+G(d) – MP2/6-31G(d) ] Δ EHLC : – Anβ – B(nα – nβ) E(G3)= Ebase + ΔE+ + ΔE2df,p + ΔEQCI + ΔEG3Large + ΔEHLC + EZPE Journal of Chemical Physics, 1998, 109, 2017年4月25日 碩士論文口試
10
Multilevel Methods with Scaled Energies
G3S G3X MCG3 MLSEn+d 2017年4月25日 碩士論文口試
11
The MCG3 Method E(MCG3/3) = c0E(HF/6-31G(d)) + c1 E(HF/MG3S | 6-31G(d)) + c2 E(MP2 | HF/6-31G(d)) + c3 E(MP2 | HF/MG3S | 6-31G(d)) + c4 E(MP4SDQ | MP2/6-31G(d)) + c5 E(MP4SDQ | MP2/6-31G(2df,p) | 6-31G(d)) + c6 E(QCISD(T) | MP4SDQ/6-31G(d)) + ESO J. Phys. Chem. A 2003, 107, 3898. 2017年4月25日 碩士論文口試
12
Dunning’s correlation consistent basis sets
Dunning-type basis set Pople-type basis sets cc-pVDZ 6-31G(d) aug-cc-pVDZ 6-31++G(d,p) cc-pVTZ 6-311G(d,p) aug-cc-pVTZ G(2df,p) 2017年4月25日 碩士論文口試
13
The MLSEn+d Method E(MLSEn+d) = CHF × E(HF/cc-pV(D+d)Z) + CHF × [E(HF/cc-pV(T+d)Z )– E(HF/cc-pV(D+d)Z)] + CE2 × [E2/cc-pV(D+d)Z] + CE34 × [E(MP4SDQ/cc-pV(D+d)Z) – E(MP2/cc-pV(D+d)Z)] + CQCI × [E(QCISD(T)/cc-pV(D+d)Z) – E(MP4SDQ/cc-pV(D+d)Z)] + CB × γE2 × [E2/cc-pV(T+d)Z – E2/cc-pV(D+d)Z] + C+ × [E2/aug-cc-pV(D+d)Z – E2/cc-pV(D+d)Z] + ESO Chem. Phys. Lett. 2005, 412, 2017年4月25日 碩士論文口試
14
Density functional theory (DFT)
To obtain energies of molecules and their physical properties without solving wave functions. Common functionals: B3LYP、 MPW1B95 、MPW1PW91、 TPSS1KCIS、B1B95 2017年4月25日 碩士論文口試
15
The MCG3-DFT Method E(MCG3-DFT) = c8{E[HF/Dd] + c1E[MP2|HF/Dd]
+ c2E[MP2/DIDZ|Dd] + c3E[MP2/D2dfp|DIDZ] + c4E[MP2/MG3S|D2dfp] + c5E[MP4SDQ|MP2/Dd] + c6E[MP4SDQ/D2dfp|Dd] + c7E[QCISD(T)|MP4SDQ/Dd]} + (1–c8)E(DFTX/MG3S) + ESO Phys. Chem. Chem. Phys. 2005, 7, 43–52. 2017年4月25日 碩士論文口試
16
Databases Train sets and Test sets
MGAE109 Test Set. The MGAE109 test set consists of 109 atomization energies (AEs). HTBH38/04 Database. The HTBH38/04 database consists of 38 transition state barrier heights for hydrogen transfer (HT) reactions, NHTBH22/04 Database. The NHTBH22/04 database consists of 38 transition state barrier heights for non-hydrogentransfer (NHT) reactions. 2017年4月25日 碩士論文口試
17
The MLSE-DFT Method E(MLSE-DFT) = CWF { E(HF/cc-pV(D+d)Z) +
CHF [E(HF/cc-pV(T+d)Z )– E(HF/cc-pV(D+d)Z)] + CE2 [E2/cc-pV(D+d)Z] + CE34 [E(MP4SDQ/cc-pV(D+d)Z) – E(MP2/cc-pV(D+d)Z)] + CQCI [E(QCISD(T)/cc-pV(D+d)Z) – E(MP4SDQ/cc-pV(D+d)Z)] + CB [E2/cc-pV(T+d)Z – E2/cc-pV(D+d)Z] + CHF+ [E(HF/aug-cc-pV(D+d)Z) – E(HF/cc-pV(D+d)Z]) + CE2+ [E2/aug-cc-pV(D+d)Z – E2/cc-pV(D+d)Z] } + (1 - CWF ) { E(DFTX/cc-pV(D+d)Z) + CB1 [E(DFTX/cc-pV(T+d)Z – DFTX/cc-pV(D+d)Z] } + ESO Chem. Phys. Lett. 2007, 442, 220. 2017年4月25日 碩士論文口試
18
Accuracy 2017年4月25日 碩士論文口試
19
MLSE-DFT Optimized Coefficients
2017年4月25日 碩士論文口試
20
The MLSE-DFT Computational Cost
2017年4月25日 碩士論文口試
21
For charged system In order to perfect multi-level electronic structure methods, we development a new series methods for charged system that are not suitable for MLSE-DFT. These series methods are called MLSE(Cn)-DFT. 2017年4月25日 碩士論文口試
22
Database for charged system
Train sets and Test sets MGAE109 Test Set. The MGAE109 test set consists of 109 atomization energies (AEs). Ionization Potential and Electron Affinity Test Set. These databases are called IP13/3 and EA13/3, respectively. HTBH38/04 Database. The HTBH38/04 database consists of 38 transition state barrier heights for hydrogen transfer (HT) reactions, NHTBH38/04 Database. The HTBH38/04 database consists of 38 transition state barrier heights for non-hydrogentransfer (NHT) reactions. 2017年4月25日 碩士論文口試
23
The MLSE(C1)-DFT Method
E(MLSE(C1)-DFT) = CWF { E(HF/pdz) + C△HF [E(HF/ptz )– E(HF/pdz)] + CE2 [E2/pdz] + CE34 [E(MP4SDQ/pdz) – E(MP2/pdz)] + CQCI [E(QCISD(T)/pdz) – E(MP4SDQ/pdz)] + CB1MP2 [E2/ptz – E2/pdz] + CHF [E(HF/apdz) – E(HF/pdz]) + CE2+ [E2/apdz – E2/pdz] + CHFT+ [E(HF/aptz) - E(HF/apdz)] + CB2MP2 [E2/aptz – E2/apdz] + CB1MP4 [E(MP4D/ptz) - E(MP4D/pdz)] } + (1 - CWF ) { E(DFTX/pdz) + CB1DFT [E(DFTX/ptz – DFTX/pdz] } + ESO 2017年4月25日 碩士論文口試
24
Simplification of MLSE(C1)-DFT
The computational cost of MLSE(C1)-DFT is significantly higher than that of MLSE-DFT because of the expensive MP2/aug-cc-pVTZ calculation. One way to lower the cost would be reducing the size of the aug-cc-pVTZ basis set. We simplify the aug-cc-pVTZ basis sets by omitting the f diffuse functions for the second-row elements, omitting the d,f diffuse functions for the first-row elements, and omitting all diffuse functions for hydrogens. 2017年4月25日 碩士論文口試
25
The MLSE(C2)-DFT Method
E(MLSE(C2)-DFT) = CWF { E(HF/pdz) + C△HF [E(HF/ptz )– E(HF/pdz)] + CE2 [E2/pdz] + CE34 [E(MP4D/pdz) – E(MP2/pdz)] + CQCI [E(QCISD(T)/pdz) – E(MP4D/pdz)] + CB1MP2 [E(MP2/ptz) – E(MP2/pdz)] + CHF+ [E(HF/apdz) – E(HF/pdz]) + CE2+ [E2/apdz – E2/pdz] + CB2MP2 [E(MP2/aptz) – E(MP2/apdz)] + CB1MP4 [E(MP4D/ptz) - E(MP4D/pdz)] } + (1 - CWF ) { E(DFTX/pdz) + CB1DFT [E(DFTX/ptz – DFTX/pdz] } 2017年4月25日 碩士論文口試
26
Simplification of MLSE(C2)-DFT
Two large basis sets, ptz and the simplified aptz, are still used in the MP2 calculation, and the MP4D/ptz calculation is also very expensive. To make the method even more affordable, we eliminate the calculation using the ptz basis set completely in the following MLSE(C3)-DFT method. 2017年4月25日 碩士論文口試
27
The MLSE(C3)-DFT Method
E(MLSE(C3)-DFT) = CWF { E(HF/pdz) + CE2 [E2/pdz] + CE34 [E(MP4SDQ/pdz) – E(MP2/pdz)] + CQCID [E(QCISD/pdz) - E(MP4SDQ/pdz)] + CQCI [E(QCISD(T)/pdz) – E(QCISD/pdz)] + CHF+ [E(HF/apdz) – E(HF/pdz]) + CE2+ [E2/apdz – E2/pdz] + CHFT+ [E(HF/aptzs) - E(HF/apdz)] + CB2MP2 [E(MP2/aptz) – E(MP2/apdz)] + CBMP4+ [E(MP4SDQ/apdz) - E(MP4SDQ/pdz)] } (1 - CWF ) { E(DFTX/pdz) } + ESO 2017年4月25日 碩士論文口試
28
Accuracy AE IP EA HTBH NHTBH NHTBH(C) MUE MLSE(C1)-MPWB 0.645 0.698 0.595 0.473 0.483 0.424 0.581 MLSE(C)1-TS 0.633 0.743 0.637 0.508 0.580 0.536 0.605 MLSE(C)2-MPWB 0.640 0.817 0.790 0.451 0.525 0.445 0.599 MLSE(C)2-TS 0.630 0.890 0.782 0.481 0.578 0.614 0.622 MLSE(C)3-MPWB 0.766 0.709 0.665 0.436 0.337 0.662 MLSE(C)3-TS 0.724 0.677 0.692 0.752 0.347 0.648 MCG3-MPWB 0.75 0.670 0.860 0.54 0.97 0.650 0.729 2017年4月25日 碩士論文口試
29
MLSE(Cn)-DFT Computational Cost
MLSE(C1)-MPWB 7396 MLSE(C2)-MPWB 4059 MLSE(C3)-MPWB 2673 MCG3-MPWB 2334 Total CPU time in seconds to calculate C5H5N, C2Cl4, C4H4O, C4H4S, C4H5N, CF3CN, and SiCl4 using Intel E6600 processer. 2017年4月25日 碩士論文口試
30
Summary Single-Level methods Multilevel Methods
method: QCISD(T)/aug-cc-pVTZ accuracy: >> 1 kcal/mol cost1: several hours to several days Multilevel Methods method: G3 accuracy: 1~2 kcal/mol cost1: several minutes to several hours 1for medium molecules, ~10 heavy atoms. 2017年4月25日 碩士論文口試
31
Summary Scaled Multilevel Methods Scaled Multilevel Methods with DFT
method: MLSE1+d, MCG3 accuracy: 0.9~1.0 kcal/mol cost1: several minutes Scaled Multilevel Methods with DFT method: MLSE-MPWB, MCG3-MPWB accuracy: 0.6~0.7 kcal/mol 1for medium molecules, ~10 heavy atoms. 2017年4月25日 碩士論文口試
32
Concluding Remarks We have developed a set of new multi-level electronic structure methods by including energies calculated from several density functional theory methods, we called it MLSE-DFT method. For neutral system, MLSE-TPSS1KCIS, performed on 169 atomization energies and reaction energy barriers with overall mean unsigned errors(MUE) of 0.61 kcal/mol. We recommend this method for neutral system. Overall MUEs of MLSE(C2)-MPWB is kcal/mol, it’s lower than MCG3-MPWB about 0.13 kcal/mol, cost is also acceptable, so it provides us another choice for charged system. 2017年4月25日 碩士論文口試
33
Acknowledgement Prof. Wei-Ping Hu Our group members.
(Tsung-Hui Li, Jien-Lian Chen et al.) Department of Chemistry & Biochemistry, National Chung Cheng University National Science Council National Center for High-Performance Computing 2017年4月25日 碩士論文口試
34
Thanks for your attention
2017年4月25日 碩士論文口試
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.