Strong Correlation meets Materials Modelling: DMFT and GW in CASTEP

许多现代技术的功能是建立在利用材料的特殊物理性质的基础上的。这可能是半导体的电流控制，数据存储的磁性，固态冰箱的珀尔帖效应，或者MRI扫描仪中使用的超高功率磁铁的超导性。未来这些“功能材料”的改进和新材料、新器件的发展，其基础是“材料设计”的理念。计算机模拟方法能够仅根据特定晶体结构中电子的量子力学行为来预测有源特性，这是为未来设备设计“设计材料”的任何尝试的重要组成部分。该基金的提案人是负责编写CASTEP密度泛函理论（DFT）建模代码的人员之一，该代码是上一代HECToR国家高性能计算服务中使用的前几个建模代码之一。CASTEP被授权给近100个英国研究小组，在2013年被用于产生118个研究出版物，全球总数超过4000个。密度泛函理论是大部分现代材料科学、凝聚态物理和固态化学所选择的模拟方法。它可以处理各种各样的材料，包括大块金属、氧化物、半导体、石墨烯等层状材料、表面等等。它被广泛用于预测结构和能量性质，如相变、电子输运性质、光谱性质，包括核磁共振化学位移、非弹性中子和x射线、拉曼和红外、XANES和其他近边电子能谱。然而，它的成功并不是普遍的。交换和相关的基本近似（即LDA， GGAs和杂化）对许多含有开放d和f壳层的材料的预测能力很差，通常被描述为“强相关”。例如，许多过渡金属氧化物被LDA和GGA错误地预测为金属，没有Jahn-Teller结构畸变，也没有重现涉及电子离域的相变。处理强相关系统的方法弥补了半局部DFT的不足，包括动态平均场理论（DMFT）和所谓的GW近似。然而，材料建模社区目前还不容易获得这些，部分原因是计算费用，部分原因是缺乏健壮的、易于部署的建模代码。在此，我们建议在CASTEP中实现LDA+DMFT和GW， CASTEP是英国首屈一指的基于电子结构的材料建模代码。我们的战略目标是大幅扩大英国的强相关性建模社区，与法国或德国相比，英国目前规模较小。计划交付的成果将为CASTEP提供最终用户能力的扩展，同时也为该方法的进一步开发提供一个软件平台。我们的DMFT实施的一个关键特征将是力的可用性和结构优化的能力。这将使具有低位置对称性的复杂材料的现实处理成为可能。我们希望这将刺激这些方法在英国的采用，从开发小组到更广泛的建模社区，就像DFT发生的那样。当与实验研究结合使用时，模拟的力量会发挥最大的作用。英国已经在钻石光源的RIXS和ARPES光束线以及ISIS目标站2中子设施的MERLIN仪器上投入了大量资金，以瞄准“强相关”材料。该提案的一个关键特征是与这两个机构的科学家合作，为科学用户社区选择的实验研究提供一些实际材料的建模能力。我们相信这将使这项工作的影响最大化。

The function of much modern technology is based on exploiting special physical properties of materials. This might be control of electrical current in the case of semiconductors, magnetism for data storage, the Peltier effect for solid-state refrigerators, or superconductivity for ultra high power magnets used in MRI scanners. Underpinning future refinements of such "functional materials" and development of new materials and devices lies the idea of "materials design". Computer simulation methods with the power to predict the active properties based only on the quantum-mechanical behaviour of the electrons in a particular crystal structure are a vital part of any attempt to engineer "designer materials" for future devices.The proposers of this grant are among those responsible for writing the CASTEP density-functional theory (DFT) modelling code, which is among the top few modelling codes used on the previous generation HECToR national high-performance computing service. CASTEP is licensed to nearly 100 UK research groups, was used to generate 118 research publications in 2013 and a worldwide total exceeding 4000.Density-functional theory is the simulation method of choice for a large part of modern materials science, condensed matter physics and solid-state chemistry. It can treat a huge range of materials, including bulk metals, oxides, semiconductors, layered materials such as graphene, surfaces and much more. It is widely used to predict structural and energetic properties such as phase transitions, electronic transport properties, spectroscopic properties including NMR chemical shifts, inelastic neutron and X-Ray, Raman and infrared, XANES and other near-edge electronic spectra.However, its success is not universal. The essential approximations to exchange and correlation (ie the LDA, GGAs and hybrids) have poor predictive power for many materials containing open d- and f-shells, frequently described as "strongly-correlated". For example many transition-metal oxides are erroneously predicted by LDA and GGA to be metallic, Jahn-Teller structural distortions are absent and phase transitions involving electron delocalisation are not reproduced.Methods for treating strongly-correlated systems which remedy the deficiencies of semi-local DFT include dynamical mean-field theory (DMFT) and the so-called GW approximation. However these are currently not readily accessible to the materials modelling community, partly because of computational expense and partly because of the lack of robust, easily deployable modelling codes.Herein we propose to implement both LDA+DMFT and GW within CASTEP, the UK's premier electronic structure-based materials modelling code. Our strategic target is to substantially broaden the strong-correlation modelling community in the UK, which is currently small compared to either France or Germany. The planned deliverables will offer both an extension to CASTEP with end-user capability, but also a software platform for further development of the methodology. A key feature of our DMFT implementation will be the availability of forces, and capability of structure optimization. This will enable a realistic treatment of complex materials with low site symmetry. We hope this will stimulate an growth in the adoption of these methods in the UK from the developer groups to a wider modelling community much as happened with DFT.The power of simulation works best when used hand in hand with experimental studies. The UK has invested heavily in instruments targeted at "strongly-correlated" materials in the RIXS and ARPES beamlines at Diamond Light Source and the MERLIN instruments at ISIS target station 2 neutron facility. A key feature of this proposal is to work with scientists at both facilities to deliver the capability of modelling some of the actual materials selected for experimental study by the scientific user community. We believe this will maximise the impact of the work.