晶界性质与结构的机器学习。
gblearn的Python项目详细描述
最近,我们提出了一个通用的晶界描述符 具有理想的数学性质,可应用于 任意晶界。使用这个描述符,我们能够 基于局部原子的机器学习特征矩阵的建立 晶界处的环境。除了有用 为了预测晶界能和迁移率,该方法还 允许为每个 财产。
如果您使用此软件包,请引用纸张:
@article{Rosenbrock:2017vd, author = {Rosenbrock, Conrad W and Homer, Eric R and Csanyi, G{\'a}bor and Hart, Gus L W}, title = {{Discovering the building blocks of atomic systems using machine learning: application to grain boundaries}}, journal = {npj Computational Materials}, year = {2017}, volume = {3}, number = {1}, pages = {29} }
您可以为olmsted生成本地环境表示 使用以下代码的数据集。它假设所有的olmsted[1] lammps的转储文件位于/dbs/olmsted。我们告诉框架 将所有表示存储在/gbs/olmsted文件夹中。
# Load the perfect FCC as a seed so the LER can be constructed.# It assumes the the seed file is found at /seeds/"Ni.pissnnl_seed.txt"seed=np.loadtxt("/seeds/Ni.pissnnl_seed.txt")fromgblearn.gbimportGrainBoundaryCollectionasGBColmsted=GBC("olmsted","/dbs/olmsted","/gbs/olmsted",r"ni.p(?P<gbid>\d+).out",seed=seed,padding=6.50)# We explicitly call :meth:`load` to parse the GB files. Then, construct# the SOAP representation for each GB.# As part of the load function, we call it with Z=28 for the nickel database,# and also give it a method and pattern to useolmsted.load(Z=28,method="cna",pattr="c_cna")# Calculate the SOAP representation.# The SOAP representation includes padding around the boundary atoms, so# that each atom in the GB has a full `rcut` of atoms around it.# The "meth: 'soap' auto trims those atoms that don't have full environments.olmsted.soap(rcut=3.25,lmax=12,nmax=12,sigma=0.5)#Now, we can finally construct the LER.olmsted.LER(0.0025)
参考文献
[1]:Olmsted,D.L.,Foiles,S.M.和Holm,E.A.计算机调查 面心立方金属的晶界性质:Ⅰ.晶粒 边界能。实际情况。573694-3703(2009年)。