Digital navigation of chemical space for function

材料既使我们今天所依赖的技术成为可能，又推动了科学理解的进步。新材料(例如锂过渡金属氧化物)产生的新科学现象促进了技术(电动汽车)的创造，强调了创造新材料的能力与经济繁荣之间的联系。面对气候变化和资源短缺，新材料提供了一条实现清洁增长的途径，这对社会的未来至关重要。为了利用功能材料的力量实现可持续的未来，我们必须提高识别它们的能力。这是一项艰巨的任务，因为材料是从巨大的、基本上未知的化学和结构耦合空间组装而成的。因此，我们被迫主要通过与已知材料进行类比来识别新材料。从科学和技术的角度来看，这种必然渐进的方法限制了成果的多样性。如果我们要开发材料的潜力来应对社会挑战，我们需要能够设计出超出这种“类似范例”的材料。这个项目将通过融合物理和计算机科学来改变我们获得具有前所未有的化学和结构多样性的功能材料的能力。我们将开发一个数字发现平台，通过创建结构新颖和粘合的新材料类来推进知识前沿，并解决关键应用挑战，将开发的能力集中在明确的科学新颖性和应用性能目标上。发现平台将由识别新材料的需求和应用程序所需的性能决定。这一表现既是由新材料类实现的，也是对新材料类的需求，强调了项目链的相互依赖性质。我们将通过利用与计算机科学(CS)的广泛协同作用来加强尖端物理科学(PS)的能力和思维，以提高物理学家在可能材料空间中导航的能力。计算机可以吸收大型数据库并以与人类专家互补的方式处理多变量复杂性，因此我们将开发模型，将PS的知识和需求与CS关于如何在寻找化学空间中有希望的区域的精度和效率之间取得平衡的见解融合在一起。使用可解释的基于符号AI的自动推理和模型构建方法与机器学习相结合的混合技术的开发只是说明这种机会如何远远超出插值式机器学习的一个例子，插值式机器学习本身就是对我们当前知识的基线评估。通过跨CS/PS接口进行协作，我们可以在PS专业知识的指导下，在PS专业知识的指导下，以数字方式探索未知空间，以转变我们获取破坏性功能材料的能力。只有针对PS新颖性和功能价值的硬约束进行测试，才能推动发现平台达到实现这一目标所需的水平。当我们在未知的空间航行时，我们开发的工具和模型将是专家PS团队的指南针一样的指南，而不是卫星导航一样的导向器。改变材料发现的巨大机遇带来了激烈的国际竞争，来自行业(例如丰田研究院10亿美元)和政府(例如美国能源部2700万美元；日本NIMS的一个新中心，两项都是2019年)的巨额投资。我们的变革性愿景利用了英国最近在自主机器人研究人员和人工智能指导下识别性能优异的功能材料方面的进展，这些材料不是基于类似物的。这一PG的规模和灵活性将确保英国在这一重要领域走在前列。

Materials both enable the technologies we rely on today and drive advances in scientific understanding. The new scientific phenomena produced by novel materials (for example, lithium transition metal oxides) enable the creation of technologies (electric vehicles), emphasising the connection between the capability to create new materials and economic prosperity. New materials offer a route to clean growth that is essential for the future of society in the face of climate change and resource scarcity.To harness the power of functional materials for a sustainable future, we must improve our ability to identify them. This is a daunting task, because materials are assembled from the vast and largely unknown coupled chemical and structural spaces. As a result, we are forced to work mostly by analogy with known materials to identify new ones. This necessarily incremental approach restricts the diversity of outcome from both scientific and technological perspectives. We need to be able to design materials beyond this "paradigm of analogues" if we are to exploit their potential to tackle societal challenges.This project will transform our ability to access functional materials with unprecedented chemical and structural diversity by fusing physical and computer science. We will develop a digital discovery platform that will advance the frontier of knowledge by creating new materials classes with novel structure and bonding and tackle key application challenges, thus focussing the developed capability on well-defined targets of scientific novelty and application performance. The discovery platform will be shaped by the need to identify new materials and by the performance needed in applications. This performance is both enabled by and creates the need for the new materials classes, emphasising the interdependent nature of the project strands.We will strengthen cutting-edge physical science (PS) capability and thinking by exploiting the extensive synergies with computer science (CS), to boost the ability of the physical scientist to navigate the space of possible materials. Computers can assimilate large databases and handle multivariate complexity in a complementary way to human experts, so we will develop models that fuse the knowledge and needs from PS with the insights from CS on how to balance precision and efficiency in the quest for promising regions in chemical space. The development of mixed techniques that use explainable symbolic AI-based automated reasoning and model construction approaches coupled with machine learning is just one example that illustrates how this opportunity goes far beyond interpolative machine learning, itself valuable as a baseline evaluation of our current knowledge.By working collaboratively across the CS/PS interface, we can digitally explore the unknown space, informed and guided by PS expertise, to transform our ability to harvest disruptive functional materials. Only testing against the hard constraints of PS novelty and functional value will drive the discovery platform to the level needed to deliver this aim. As we are navigating uncharted space, the tools and models that we develop will be compass-like guides, rather than satellite navigation-like directors, for the expert PS team. The magnitude of the opportunity to transform materials discovery produces intense international competition with significant investments at pace from industry (e.g., Toyota Research Institute $1bn) and government (e.g., DoE $27m; a new centre at NIMS, Japan, both in 2019). Our transformative vision exploits recent UK advances in autonomous robotic researchers and artificial intelligence-guided identification of outperforming functional materials that are not based on analogues. The scale and flexibility of this PG will ensure the UK is at the forefront of this vital area.