7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
http://www.lasphub.com
http://homepage.fudan.edu.cn/fdzpliu/lasp/
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP 1
1.
2. LASP r
3.
4. VASP U
5.
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
1.
r
×
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
1.
SIESTA VASP
LAMMPS DFT-D3
H
…… NVE NVT NPT
CP2K Gaussian
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
Ø ”lasp.in“ potential8 v
( NN vasp siesta lammps d3 gaussian cp2k )( potential NN vasp siesta)
Ø ”lasp.in“ explore_type8v e
(npt nvt nve sswØ SSWv RUN_TYPE8v SSW
u RUN_TYPE=1 ”
u RUN_TYPE=2 ”
u RUN_TYPE=5 ”
u RUN_TYPE=11 ”
u RUN_TYPE=12 ”
u RUN_TYPE=15 ”
2. LASP r
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
2. LASP r
、
Ø lasp.in input
Ø lasp.str input.arc
= r
NN
VASP INCAR POTCAR KPOINTS (POSCAR lasp.str
Gaussian gaussian.inp.pre=gaussian.inp.after
allkey.log=lasp.out=arc =
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
2. LASP r
rexample (lasp.in)
=
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
2. LASP r
rexample (lasp.str)
Ø Material Studio
Ø
l PBC a b c α β γl Element name x y z
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
2. LASP r
example(lasp.out,all.arc,allstr.arc,allfor.arc,allkey.log)
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
2. LASP r
example(lasp.out,all.arc,allstr.arc,allfor.arc,allkey.log)
all.arc: MS
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
3.
Ø NN FULL = = 、
Ø NSLV = VASP 、 vasp
xxx/Libraries/1.vasplib/genso.sh)
Ø INTER = 、 。 =
(xxx/Libraries/1.vasplib/genlib.sh)
��
(xxx/Src/makefile)
echo $MKLROOT /home7/intel/compilers_and_libraries_2018.2.199/linux/mkl
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
3. 1.vasplib
echo $MKLROOT
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
3. INTERFACE version
version.shversion.preversion.f90
Readkey.F90 v
otherpes_init.F90otherpot.F90
othermove.F90
module_pes.F90 v
Makefile v
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
3. INTERFACE versionRead_key.F90 module_pes.F90
get_intget_realget_string Lget_logicalget_int_arrayget_logical_arrayget_real_arrayget_string_array Lget_block Block
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
15
3. INTERFACE versionRead_key.F90 module_pes.F90
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
16
3. INTERFACE versionOtherpes_init.F90
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
17
3. INTERFACE version
Other_pot.F90
” (3,3) (3,n)
)#)
r = =”= =
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
18
3. INTERFACE version
Othermove.F90
+
converge=.true.
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
19
3. INTERFACE versionMakefile
MKLROOTmkllibFuncObj
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
20
4. VASP U
1 INCARIBRION = 22ISYM = 0NSW = 2000000ISIF = 3
2 ” 、 reset VASP
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
21
4. VASP U
3 siesta.fdf
r
MD.TypeOfRun Forces
DivideAndConquer False
INPUT_DEBUG p lasp.str
4 Gaussian
L lasp=
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP.1 3-0, Large-scale Atomic Simulation with neural network Potential
22
5.
Ø
Ø r
Ø
1
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
http://homepage.fudan.edu.cn/fdzpliu/publication/
1.2.3. LASP NN4. NN S
5. NN
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP 2
•••
����PES��
CO
H2Catalyst
����
����
v
������
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
J.BehlerandM.Parrinello,Phys.Rev.Lett.,2007,98,146401.
High-DimensionalNN(HDNN)
1. “
2.
4.
3. p
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
Rotation-invariantSymmfunc
Behler
Radial function---Gauss
Angular function---
Cutoff function--- tan3h
=
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
LASP
Radial function--- Power
Angular function---
Cutoff function--- tan3h
= =
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
LASP NN
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
LASP NN
(a) Speed-up comparison of a 28-atom Boron crystal between NN and DFT using Plane wave basis set (PW DFT). The x axis is the logarithm of time in the unit of seconds. The inset shows the structure of the lattice. (b) Speed-up of Boron crystals on 560 cores. The x axis ($%) is the number of processors and the y axis is the speed-up.
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASPSSW + NN == NN
1. VASP DFT + PBE +U
2. SSWMD
3. = SSW•• DFT
RMS
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
http://homepage.fudan.edu.cn/fdzpliu/publication/
1.2.3. LASP NN4. NN S
5. NN
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
release_AB(_x).pot
A B:p
X: (
1. TIO2= Ti2O3= Ti
2. =
1.v 、 DFT =
2. = 、
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
NN
RMS
147-50-50-1
=NN “
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP 2
min Scale ---- max Scale
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
1.2.3. LASP NN4. NN S
5. NN
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP 2
LASPExampleswww.lasphub.com
p
LASP r lasp.in( input
NN v
potentialNN
PrintChg 0 0/1 debug=ExSymR 5 5
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASPLASP lasp.outNN lasp.out
Badstr.arc
LASP
NN 8 arc
1.
2. SSW.Safe_hardcurv150.0000
ExSymR 5
3. “NN
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
http://homepage.fudan.edu.cn/fdzpliu/publication/
1.2.3. LASP NN4. NN S
5. NN
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
1. NN
=
2. NN VASP DFT/PBED3 vdW
“30 = ==、
= “
MD
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
3. NN / PES =
MD
biased MD=metadynamics=MC=
4. NN NN= NN
=
ZPE
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
1.2.3. LASP NN4. NN S
5. NN
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP 2LASP r 5
allstr.arc =
SSWgrepEnergyallstr.arc
AlloptClimb
Constrained opt
SSW.printeveryTSSW.printselect1SSW.printdelay-1
v 1 NN SSW =
/ / SSW.printeveryT allstr.arc
allfor.arc
/
/
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
43
SSW.printeveryTSSW.printselect6SSW.printdelay -1
SSW.printselect=SSW
= =
1.2. (
“ =
= “
,= SSW.printselect6
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
v 2
• ” = “、 a
• Badstr.arc SSW
• DFT = SSW
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
45
v 3 ----
SSW
20 SSW = 100 SSWstepsSSW.SSWsteps100)
20*100*200*0.4= 16B5000-10000 =
DFT 、 LASP NNVASP setup =INCAR=KPOINTS=POTCAR
LASP =
2
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
46
3
1.2. ” ”
3.4.
5.
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
47
A s
a
W
l
oW
l
ve. , 0
e1,
g O
663 2639
C
nW
Y gO
EP S
XLi L
4278U
3
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
SSW
PES=
Monte Carlo “
J. Chem. Theory Comput, 2013=9, 1838
3
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
49
3
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
50
=
==
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
DESW
==
IS FS
=
TS
J. Chem. Theory Comput., 2013, 9, 5745
3
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
52
CBD
TS
J. Chem. Theory Comput, 2010, 6, 1136
J. Chem. Theory Comput, 2012=8, 2215
3
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
53
1R real NV V V= +
( ) ( )2 2
2 2N init t inita aV R= - × • = - × D × •é ùë û1 0R -R N N N
( )
0 1
20 1
( )
( )
rot e N
N t
tt init
C C C
R
aR
= +- •
=D•
= - × •D
F F -F N
F -F N N N
Biaseddimer rotation R1R0
25%25%
25%
25%
N0
64%20%
5%
4%
Nb
40%
11%
20%
17%
Nub
42%30%
13%
11%
Ninit
J. Chem. Theory Comput, 2010, 6, 1136
3
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
54
3
1.2. ” ”
3.4.
5.
NN-1...NN-12
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
55
4
1.
2.
3.
4.
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
56
1.
4
LASP_Example/NN/solid-TS-VCDESW
ØTSmode.arc
Ø SSWPath.arc DESW pseudo-pathway
LASP_Example/NN/molecule-ConfirmPathway-ExtrapolatefromTS
Øallstr.arc all.arc SSW Opt
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
57
4
2.
Steinhardt-type
Used in SSW
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
58
4
2.LASP_Example/LAMMPS/varcell-ssw-eam/lasp.out
$grep found lasp.out
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
Q659
43.
Ø QE.sh
Ø dos2D.f90
LASP_Example/LAMMPS/varcell-ssw-eam/lasp.out
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
Q6
60
43.
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
61
43.
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
62
44.
Phys. Chem. Chem. Phys., 2017, 19, 32125-32131
7/24/18
.1 3-0, Large-scale Atomic Simulation with neural network Potential
( ( ( 2
LASP
63
4.
Chem. Sci., 2017,8, 6327-6337
&'-I'