Computational screening of doping schemes for LiTi2(PO4)3 as cathode coating materials

In all-solid-state lithium batteries,the impedance at the cathode/electrolyte interface shows close relationship with the cycle performance.Cathode coatings are helpful to reduce the impedance and increase the stability at the interface effectively.LiTi2(PO4)3(LTP),a fast ion conductor with high ionic conductivity approaching 10^-3S·cm^-1,is adopted as the coating materials in this study.The crystal and electronic structures,as well as the Li^+ion migration properties are evaluated for LTP and its doped derivatives based on density functional theory(DFT)and bond valence(BV)method.Substituting part of Ti sites with element Mn,Fe,or Mg in LTP can improve the electronic conductivity of LTP while does not decrease its high ionic conductivity.In this way,the coating materials with both high ionic conductivities and electronic conductivities can be prepared for all-solid-state lithium batteries to improve the ion and electron transport properties at the interface.     

局部信息和全局信息相结合的点云处理网络

针对当前主流点云处理网络仅依靠局部邻域进行特征聚合导致特征提取能力不足，以及使用最大值池化造成信息损失的问题，提出了一种基于注意力的局部信息和全局信息相结合的点云处理网络。首先提出了基于通道自注意力进行局部特征聚合的方法，减少了信息的损失；然后为捕获点的远程依赖信息，设计了一种动态学习关键点的方法获取全局信息；最后构建了一种基于空间注意力的特征融合模块，使每个点均能学习全局上下文信息。在几个常用点云数据集上对方法进行了实验验证，在ModelNet40分类任务上实现了94.0%的总体分类精度、91.7%的平均分类精度；在ScanObjectNN分类任务上实现了81.5%的总体分类精度、78.1%的平均分类精度；在ShapeNet分割任务上实现了86.5%的平均交并比。表明提出的点云处理网络在分类、分割等任务中的精度均较PointNet、PointNet++、DGCNN等经典网络有显著提升，较其他点云处理网络也有不同程度的提高。