Development of wide-ranging functionality in ONETEP

量子力学对科学技术产生了深远而普遍的影响。本质上是量子力学的现象，如磁性、半导体中的电子传输和材料中杂质原子的效应，几乎位于工业的每一个分支的核心。从“第一原理”对性质和过程的量子力学计算能够做出准确的定量预测，但需要求解薛定谔方程，这是极其困难的，只能使用强大的计算机才能完成。相比之下，经验建模方法相对便宜，但缺乏第一原理方法的预测能力(第一原理方法不需要参数，只接受组成原子的原子序数作为输入)。预测能力是至关重要的，以便在实验数据不足的新的和具有挑战性的问题上取得快速进展，并产生有用的经验方法，甚至在存在这些方法时检查其可靠性。在第一性原理方法中，密度泛函理论(DFT)是一种非常成功的方法，因为它结合了高精度和中等计算成本。然而，使用传统的DFT方法进行计算的计算工作量随着原子数量的立方而增加，这使得它们即使在现代超级计算机上也无法处理原子超过数百个的问题。自从诺贝尔奖获得者沃尔特·科恩的开创性工作以来，人们已经知道，有可能重新表述密度泛函，使其成线性扩展，原则上这将允许对数千个甚至数百万个原子进行计算。然而，在一种像传统的立方标度DFT方法一样稳健和准确的方法中，这一点的实际实现是极其困难的。这项提案的申请者多年来制定的ONETEP方法就是实现了这一点。ONETEP处于第一性原理计算发展的前沿。然而，虽然用线性定标成本进行精确的第一性原理计算的基本困难已经得到解决，但目前ONETEP只有一小部分功能可用，这阻碍了它的广泛应用。在这三所大学之间的合作项目中，ONETEP的原始开发人员将领导一项雄心勃勃的工作计划，从而使代码的功能得到迅速和显著的丰富。ONETEP的代码开发道德，即软件在当前和未来的高性能计算技术上是健壮的、用户友好的、模块化的、可移植的和高效的，这将是至关重要的，并将通过本提案的三个机构之间的严格相互核对而得到进一步加强。从理论的角度来看，这些发展也是具有挑战性的，因为它们需要在ONETEP的线性缩放框架内，使用其在现场优化的局部函数方面的高度非平凡的DFT公式。工作方案提供了很大的附加值，因为在最初阶段将开发的几项基本使能技术将为随后的许多功能能力奠定基础。其结果将是一种能够以前所未有的精度在纳米级对材料进行全新水平模拟的工具。它将立即应用于分子生物学、纳米结构和材料的模拟，这些模拟为解决当前紧迫的问题提供了基础，如能源、环境和健康。随着ONETEP的商业和学术用户和开发人员数量的增加，这一新工具将在世界范围内迅速传播和广泛使用；最后，不断扩大的ONETEP开发人员小组将为该软件的未来维护和开发协调最佳战略。

Quantum mechanics has had a profound and pervasive influence on science and technology. Phenomena that are intrinsically quantum mechanical, such as magnetism, electron transport in semiconductors, and the effect of impurity atoms in materials, lie at the heart of almost every branch of industry. Quantum mechanical calculations of properties and processes from ``first-principles'' are capable of making accurate quantitative predictions but require solving the Schrdinger equation which is extremely difficult and can only be done using powerful computers. In contrast, empirical modelling approaches are relatively cheap but lack the predictive power of first-principles methods (which are parameter-free and take as input only the atomic numbers of the constituent atoms). The predictive capability is essential, in order to make rapid progress on new and challenging problems where there is insufficient experimental data and to also generate useful empirical approaches or even to check their reliability when these exist. Within the class of first-principles methods, one approach that has been outstandingly successful is the Density Functional Theory (DFT) as it combines high accuracy with moderate computational cost. Nevertheless, the computational effort of performing calculations with conventional DFT approaches increases as the cube of the number of atoms, making them unable to tackle problems with more than a few hundred atoms even on modern supercomputers. Since the pioneering work of the Nobel laureate Walter Kohn, it has been known that it is possible to reformulate DFT so that it scales linearly, which would in principle allow calculations with many thousands or even millions of atoms. The practical realisation of this however, in a method which is as robust and accurate as conventional cubic-scaling DFT approaches has been extremely difficult. The ONETEP approach developed over many years by the applicants of this proposal has achieved just that. ONETEP is at the cutting edge of developments in first principles calculations. However, while the fundamental difficulties of performing accurate first-principles calculations with linear-scaling cost have been solved, only a small core of functionality is currently available in ONETEP which prevents its wide application. In this collaborative project between three Universities, the original developers of ONETEP will lead an ambitious workplan whereby the functionality of the code will be rapidly and significantly enriched. The code development ethic of ONETEP, namely that software is robust, user-friendly, modular, portable and highly efficient on current and future HPC technologies will be of fundamental importance and will be further strengthened by rigorous cross-checking between the three institutions of this proposal. The developments are also challenging from a theoretical point of view as they need to be within the linear-scaling framework of ONETEP, using its highly non-trivial formulation of DFT in terms of in situ optimised localised functions. The program of work provides much added value as the few fundamental enabling technologies that will be developed in its first stages will then underpin many of the functional capabilities that will follow. The result will be a tool capable of a whole new level of materials simulation at the nanoscale with unprecedented accuracy. It will find immediate application in simulations in molecular biology, nanostructures and materials, which underpin solutions in urgent current problems such as energy, environment and health. Through the increasing number of commercial and academic users and developers of ONETEP, the worldwide dissemination and wide use of this novel tool will be rapid; finally the expanding ONETEP Developers' Group will coordinate the best strategies for the future maintenance and development of the software.