Molecular Modeling Using HPC and Gaussian:
Density-Functional Theory and Noncovalent Interactions
Gino A. DiLabio
National Institute for Nanotechnology
11421 Saskatchewan Drive
Edmonton, Alberta
T5T 5A6
Gino.DiLabio@nrc.ca
www.ualberta.ca/~gdilabio
Westgrid Seminar Series
University of Alberta
Feb. 5, 2014
Geckos have evolved one of the most
versatile and effective adhesives
known. The mechanism of dry
adhesion in the millions of setae on
the toes of geckos has been the focus
of scientific study
for over a century. We provide the first
direct experimental evidence for dry
adhesion of gecko setae by
noncovalent interactions (also called
van der Waals forces)
12252–12256 PNAS September 17, 2002 vol. 99 no. 19
www.pnas.orgcgidoi10.1073pnas.192252799
Why do we care about weak
interactions?
Formation of 1D and 2D organic lines on Si
Nature, 2005, 435, 658-661.
Why do we care about weak
interactions?
Heavy oil/bitumen upgrading
Mackie and DiLabio, Energy and Fuels, 2010, 24, 6468-6475.
Why do we care about weak
interactions?
Organic electronic material
J. Phys. Chem. C, 2010, 114. 10952.
How do we model these systems?
• Large systems necessitate the use of density-functional theory (DFT)
• B3LYP was one of the first hybrid DFT that was implemented in most
computational chemistry programs. It has since found general
acceptance and use and works well for thermochemistry.
• However, most conventional DFT methods, including B3LYP, cannot
accurately treat weak, non-covalently bonded systems – specifically
“dispersion” interactions.
How bad are conventional DFT methods
for weak interactions?
Chem. Phys. Lett. 2006, 419, 333-339
Percent Error in Binding Energy vs. Interaction
Improving the performance of DFT methods:
Dispersion-corrected DFT in Gaussian
•DFT-D/D3: Add-on to the DFT energy an empirical van der Waals
term. Developed for use with many functionals. (Grimme, J. Chem.
Phys. 2010, 132, 154104)
•M06-2X: A DFT method parameterized to reproduce dispersion
binding, among other properties. (Zhao and Truhlar, Theor. Chim.
Acta, 2008, 120, 215.)
•DCP: Dispersion correcting potentials correct the erroneous
noncovalent behaviour of a small number of DFT methods. (Torres
and DiLabio, J. Phys. Chem. Lett. 2012, 3, 1738) – Compute Canada
RAC supported work.
Review: J. Phys. Org. Chem. 2009, 22, 1127-1135.
See also: 2014 version of Reviews in Computational Chemistry
How to incorporate dispersion corrections into your
Gaussian DFT calculations
Grimme “D3” approach
Truhlar M06-2X approach
#B3LYP Gen EmpiricalDispersion=GD3BJ SCF=(Conver=6)
#M062X Gen SCF=(Conver=6)
Water Dimer
Water Dimer
01
O -0.702196054
H -1.022193224
H 0.257521062
O 2.220871067
H 2.597492682
H 2.593135384
01
O -0.702196054
H -1.022193224
H 0.257521062
O 2.220871067
H 2.597492682
H 2.593135384
HO 0
6-31+G(2d,2p)
****
-0.056060256
0.846775782
0.042121496
0.026716792
-0.411663274
-0.449496183
0.009942262
-0.011488714
0.005218999
0.000620476
0.766744858
-0.744782026
HO 0
6-31+G(2d,2p)
****
-0.056060256
0.846775782
0.042121496
0.026716792
-0.411663274
-0.449496183
0.009942262
-0.011488714
0.005218999
0.000620476
0.766744858
-0.744782026
#B3LYP Gen Pseudo=Read SCF=(Conver=6)
Water Dimer
How to use DCPs in Gaussian:
01
O -0.702196054
H -1.022193224
H 0.257521062
O 2.220871067
H 2.597492682
H 2.593135384
-0.056060256
0.846775782
0.042121496
0.026716792
-0.411663274
-0.449496183
0.009942262
-0.011488714
0.005218999
0.000620476
0.766744858
-0.744782026
Conventional
Input
HO 0
6-31+G(2d,2p)
****
Input generating utility for
Gaussian input files at:
www.ualberta.ca/~gdilabio
H0
H10
P an up
3
2 0.120883601
2 0.044528578
2 0.005658790
S-P
1
2 0.174740501
O0
O30
F an up
3
2 0.192168931
2 0.166560549
2 0.016734867
S-F
1
2 0.039337457
P-F
1
2 0.123780982
D-F
1
2 0.061418151
0.000231333
-0.000070677
-0.000000451
-0.000049845
DCPs
0.000242019
0.003000000
-0.000000131
0.000016041
-0.000129441
-0.003520223
How DCPs work
Based on (old) ECP technology:
e.g. Iodine = 1s22s22p6…5s24d105p5  [Potential]
5s24d105p5
In the case of DCPs, the potentials don’t replace any core electrons
but instead are constructed such that the long-range behaviour of
a DFT is corrected.
How DCPs are generated
Compute Canada RAC work
#B3LYP Gen Pseudo=Read SCF=(Conver=6)
Water Dimer
01
O -0.702196054 -0.056060256 0.009942262
H -1.022193224 0.846775782 -0.011488714
H 0.257521062 0.042121496 0.005218999
O 2.220871067 0.026716792 0.000620476
H 2.597492682 -0.411663274 0.766744858
H 2.593135384 -0.449496183 -0.744782026
HO 0
6-31+G(2d,2p)
****
H0
H10
P an up
3
2 0.120883601
2 0.044528578
2 0.005658790
S-P
1
2 0.174740501
O0
O30
F an up
3
2 0.192168931
2 0.166560549
2 0.016734867
S-F
1
2 0.039337457
P-F
1
2 0.123780982
D-F
1
2 0.061418151
0.000231333
-0.000070677
-0.000000451
-0.000049845
0.000242019
0.003000000
-0.000000131
0.000016041
-0.000129441
-0.003520223
H0
H10
P an up
3
2 0.120883601
2 0.044528578
2 0.005658790
S-P
1
2 0.174740501
0.000231333
-0.000070677
-0.000000451
-0.000049845
How DCPs are generated
Compute Canada RAC work
Steps:
1.
Develop an initial guess for a DCP for an atom/DFT method/basis set
combination.
2.
Select the fitting data to which DCPs will be optimized.
3.
Submit optimization script to Grex. The script:
a) Builds Gaussian input files based on the information in 1&2.
b) Builds queue files for Gaussian runs
c) Submits queue files to the queue
d) Monitors status of runs
e) Extracts energies of noncovalently bonded systems and their monomers
f) Adjusts values of Ci and ζi according to some optimization scheme
4.
Select next atom and repeat from step 2.
Past Successes
B3LYP-DCP versus other dispersion-corrected DFT methods
Functional/6-31+G(2d,2p)
25
20
15
10
5
0
-30
-20
-15
-10
-5
0
5
10
15
20
30
40
50
M06-2X
Performance on the S66 benchmark set: Occurrences in %error in BE
Past Successes
B3LYP-DCP versus other dispersion-corrected DFT methods
Functional/6-31+G(2d,2p)
25
20
15
10
5
0
-30
-20
-15
-10
M06-2X
-5
0
5
10
15
20
30
40
50
ωB97XD
Performance on the S66 benchmark set: Occurrences in %error in BE
Past Successes
B3LYP-DCP versus other dispersion-corrected DFT methods
Functional/6-31+G(2d,2p)
25
20
15
10
5
0
-30
-20
-15
-10
M06-2X
-5
0
ωB97XD
5
10
15
20
30
40
50
B97D
Performance on the S66 benchmark set: Occurrences in %error in BE
Past Successes
B3LYP-DCP versus other dispersion-corrected DFT methods
Functional/6-31+G(2d,2p)
25
20
15
10
5
0
-30
-20
-15
-10
M06-2X
-5
0
ωB97XD
5
10
B97D
15
20
30
40
50
B3LYP-DCP
Performance on the S66 benchmark set: Occurrences in %error in BE
Past Successes
More than just noncovalent interactions. DCPs can improve
thermochemical properties: LC-ωPBE/6-31+G(2d,2p)
Bond Dissociation
CO
CN
CC
DCP
D3
OH
unadorned
NH
CH
0
1
2
3
Mean Absolute Error (kcal/mol)
4
Future Efforts on DCPs
1. “Low cost” Density-functional theory methods + various
computational chemistry/physics codes.
2. Thermochemistry, kinetics and noncovalent interactions.
3. Develop a deeper understanding of how DCPs work.
Challenges
1. Optimization scripts are complicated and platform
dependent.
2. Conventional resource allocation scheme is not ideal for
this work.
Thanks to:
Dr. Edmanuel Torres - NINT
Dr. Iain D. Mackie - NINT
Prof. Erin R. Johnson – UC Merced