To enhance the separation selectivity of Mg-MOF-74 towards CO_(2) in a CO_(2)/N_(2) mixture,a series of Mg-MOF-74 and Ni_(x)/Mg_(1-x)-MOF-74 adsorbents were prepared by solvothermal synthesis in this paper.It was foun...To enhance the separation selectivity of Mg-MOF-74 towards CO_(2) in a CO_(2)/N_(2) mixture,a series of Mg-MOF-74 and Ni_(x)/Mg_(1-x)-MOF-74 adsorbents were prepared by solvothermal synthesis in this paper.It was found that the adsorption capacity of Mg-MOF-74 for CO_(2) could be effectively increased by optimizing the amount of acetic acid.On this basis,the bimetal MOF-74 adsorbent was prepared by metal modification.The multi-component dynamic adsorption penetration analysis was utilized to examine the CO_(2) adsorption capacity and CO_(2)/N_(2) selectivity of the diverse adsorbent materials.The results showed that Ni0.11/Mg0.89-MOF-74 showed a CO_(2) adsorption capacity of 7.02 mmol/g under pure CO_(2) atmosphere and had a selectivity of 20.50 for CO_(2)/N_(2) under 15% CO_(2)/85%N_(2) conditions,which was 10.2% and 18.02% higher than that of Mg-MOF-74 respectively.Combining XPS,SEM and N_(2) adsorption-desorption characterization analysis,it was attributed to the effect of the more stable unsaturated metal sites Ni into the Mg-MOF-74 on the pore structure and the synergistic interaction between the two metals.Density Functional Theory(DFT)simulations revealed that the synergistic interaction between modulated the electrostatic potential strength and gradient of the material,which was more favorable for the adsorption of CO_(2) molecules with small diameters and large quadrupole moment.In addition,the Ni0.11/Mg0.89-MOF-74 showed commendable cyclic stability,underscoring its promising potential for practical applications.展开更多
四元数神经网络将实值神经网络推广到了四元数代数中,其在偏振合成孔径雷达奇异点补偿、口语理解、机器人控制等任务中取得了比实值神经网络更高的精度或更快的收敛速度.四元数神经网络的性能在实验中已得到广泛验证,但四元数神经网络...四元数神经网络将实值神经网络推广到了四元数代数中,其在偏振合成孔径雷达奇异点补偿、口语理解、机器人控制等任务中取得了比实值神经网络更高的精度或更快的收敛速度.四元数神经网络的性能在实验中已得到广泛验证,但四元数神经网络的理论性质及其相较于实值神经网络的优势研究较少.从表示能力的角度出发,研究四元数神经网络的理论性质及其相较于实值神经网络的优势.首先,证明了四元数神经网络使用一个非分开激活的修正线性单元(rectified linear unit,ReLU)型激活函数时的通用近似定理.其次,研究了四元数神经网络相较于实值神经网络的逼近优势.针对分开激活的ReLU型激活函数,证明了单隐层实值神经网络需要约4倍参数量才能生成与单隐层四元数神经网络相同的最大凸线性区域数.针对非分开激活的ReLU型激活函数,证明了单隐层四元数神经网络与单隐层实值神经网络间的逼近分离:四元数神经网络可用相同的隐层神经元数量与权重模长表示实值神经网络,而实值神经网络需要指数多个隐层神经元或指数大的参数才可能近似四元数神经网络.最后,模拟实验验证了理论.展开更多
基金supported by National Natural Science Foundation of China(U23A20100)the Strategic Priority Research Program(A)of the Chinese Academy of Sciences(XDA0390404)+5 种基金ICC CAS SCJC-DT-2023-03,the Foundation of State Key Laboratory of Coal Conversion(J24-25-619)Youth Innovation Promotion Association CAS(2018209,2020179)Key R&D Program of Shanxi Province(202102090301008,202202090301013)the special fund for S&T Innovation Team of Shanxi Province(202204051001012)Project of International Cooperation and Exchange NSFC-RFBR(22011530069)Tianjin Science and Technology Plan Project(22YFYSHZ00290)。
文摘To enhance the separation selectivity of Mg-MOF-74 towards CO_(2) in a CO_(2)/N_(2) mixture,a series of Mg-MOF-74 and Ni_(x)/Mg_(1-x)-MOF-74 adsorbents were prepared by solvothermal synthesis in this paper.It was found that the adsorption capacity of Mg-MOF-74 for CO_(2) could be effectively increased by optimizing the amount of acetic acid.On this basis,the bimetal MOF-74 adsorbent was prepared by metal modification.The multi-component dynamic adsorption penetration analysis was utilized to examine the CO_(2) adsorption capacity and CO_(2)/N_(2) selectivity of the diverse adsorbent materials.The results showed that Ni0.11/Mg0.89-MOF-74 showed a CO_(2) adsorption capacity of 7.02 mmol/g under pure CO_(2) atmosphere and had a selectivity of 20.50 for CO_(2)/N_(2) under 15% CO_(2)/85%N_(2) conditions,which was 10.2% and 18.02% higher than that of Mg-MOF-74 respectively.Combining XPS,SEM and N_(2) adsorption-desorption characterization analysis,it was attributed to the effect of the more stable unsaturated metal sites Ni into the Mg-MOF-74 on the pore structure and the synergistic interaction between the two metals.Density Functional Theory(DFT)simulations revealed that the synergistic interaction between modulated the electrostatic potential strength and gradient of the material,which was more favorable for the adsorption of CO_(2) molecules with small diameters and large quadrupole moment.In addition,the Ni0.11/Mg0.89-MOF-74 showed commendable cyclic stability,underscoring its promising potential for practical applications.
文摘四元数神经网络将实值神经网络推广到了四元数代数中,其在偏振合成孔径雷达奇异点补偿、口语理解、机器人控制等任务中取得了比实值神经网络更高的精度或更快的收敛速度.四元数神经网络的性能在实验中已得到广泛验证,但四元数神经网络的理论性质及其相较于实值神经网络的优势研究较少.从表示能力的角度出发,研究四元数神经网络的理论性质及其相较于实值神经网络的优势.首先,证明了四元数神经网络使用一个非分开激活的修正线性单元(rectified linear unit,ReLU)型激活函数时的通用近似定理.其次,研究了四元数神经网络相较于实值神经网络的逼近优势.针对分开激活的ReLU型激活函数,证明了单隐层实值神经网络需要约4倍参数量才能生成与单隐层四元数神经网络相同的最大凸线性区域数.针对非分开激活的ReLU型激活函数,证明了单隐层四元数神经网络与单隐层实值神经网络间的逼近分离:四元数神经网络可用相同的隐层神经元数量与权重模长表示实值神经网络,而实值神经网络需要指数多个隐层神经元或指数大的参数才可能近似四元数神经网络.最后,模拟实验验证了理论.
