Method for predicting a molecular structure
A method for predicting a molecular structure includes: preparing a learning data set including first learning data including an eigenvector value and a quantum mechanics calculation value for a monoatomic and molecular structural model, a bulk structural model, a slab structural model, and a nanoparticle structural model of a material including a plurality of elements; learning an artificial neural network using the learning data set to obtain a potential value; and predicting a molecular structure of another material by using the potential value.
1 . A method for predicting a molecular structure, the method comprising:
preparing, by a CPU or GPU device, a learning data set including first learning data including eigenvector values and quantum mechanics calculation values for a monoatomic and molecular structural model, a bulk structural model, a slab structural model, a nanoparticle structural model, a defect structural model, and an amorphous structural model of a material including a plurality of elements;
learning, by the CPU or GPU device, an artificial neural network using the learning data set to obtain a potential value, and wherein the potential value is provided as a file;
predicting, by the CPU or GPU device, a molecular structure of another material by using the potential value; and
using the molecular structure of the other material to dissolve, aggregate, or react the other material in the fields of semiconductors, magnetics, energy materials, and electrification materials,
wherein the learning data set further comprises second learning data comprising an eigenvector value and a quantum mechanics calculation value for a structural model obtained by molecular dynamics calculation using an effective medium theory potential value, and
wherein the learning data set further comprises third learning data comprising an eigenvector value and a quantum mechanics calculation value for a structural model in which quantum mechanical calculation values are different from predicted values obtained using the first learning data and the second learning data.
2 . The method of claim 1 , wherein
the learning data set further comprises fourth learning data comprising an eigenvector value and a quantum mechanics calculation value for a structural model generated during stable structural relaxation of an amorphous structural model, a structural model generated during stable structural relaxation of a defect structural model, a structural model generated during stable structural relaxation of the slab structural model, or a combination thereof.
3 . The method of claim 2 , wherein
the learning data set further comprises fifth learning data comprising an eigenvector value and a quantum mechanics calculation value for a structural model generated during stable structural relaxation of a structure obtained by molecular dynamics calculation using an effective medium theoretical potential value or a structural model generated during stable structural relaxation of a structural model of the third learning data.
4 . The method of claim 3 , wherein
the learning data set comprises sixth learning data comprising the first learning data to the fifth learning data.
5 . The method of claim 3 , wherein
the learning data set comprises seventh learning data comprising the first learning data and the fifth learning data.
6 . The method of claim 1 , wherein
in the obtaining of the potential value,
each of the eigenvector values for the monoatomic and molecular structural model, the bulk structural model, the slab structural model, the nanoparticle structural model, the defect structural model, and the amorphous structural model is an input value of the artificial neural network, and
a potential value for the molecular structure of the material is an output value.
7 . The method of claim 6 , wherein
in the obtaining of the potential value, when an error of the output value is greater than or equal to 25 me V/atom, reconstructing the learning data set and then learning the artificial neural network using the reconstructed learning data set to obtain the potential value are repeated.
8 . The method of claim 1 , wherein
the eigenvector values for the monoatomic and molecular structural model, the bulk structural model, the slab structural model, the nanoparticle structural model, the defect structural model, and the amorphous structural model are obtained by Equation 1 or Equation 2,
wherein Equation 1 comprises:
G
i
II
=
∑
j
≠
i
atoms
j
within
R
c
distance
of
atom
1
e
-
η
(
R
ij
-
R
s
)
2
/
R
c
2
f
c
(
R
ij
)
wherein, in Equation 1, R ij is a distance between atom i and atom j, R s and η are variables controlling a shape of a graph, f c is a cut-off function, which is expressed by Equation 3,
wherein Equation 2 comprises:
G
i
IV
=
2
1
-
ς
∑
j
,
k
≠
i
(
j
≠
k
)
atoms
j
,
k
within
R
c
distance
of
atom
i
(
1
+
λcos
θ
ijk
)
ς
e
-
η
(
R
ij
2
+
R
ik
2
+
R
jk
2
)
/
R
c
2
f
c
(
R
ij
)
f
c
(
R
ik
)
f
c
(
R
jk
)
wherein, in Equation 2, R ij is a distance between atom i and atom j, Rik is a distance between atom i and atom k, R jk is a distance between atom j and atom k, R s , η, λ, and ζ are variables controlling the shape of a graph, and f c is a cut-off function, which is expressed by Equation 3,
wherein Equation 3 comprises:
when r≠R c ,
fc
=
0.5
(
1
+
cos
(
π
r
Rc
)
)
when r>R c , f c =0
wherein, in Equation 3, r is a radius of a neighbor atom from a center atom and Rc is a limiting radius for limiting an effect on the center atom.
9 . The method of claim 8 , wherein
the eigenvector values are obtained by changing η, λ, ζ, or a combination thereof in Equation 1 or Equation 2.