PRISM: Platform for Research In Simulation Methods

计算科学是一个多学科的研究工作，跨越应用数学，计算机科学和工程以及跨科学，技术和医学应用领域的投入。先进的仿真方法不仅有可能彻底改变科学研究，而且有可能改变工业经济，为企业提供产品竞争优势、更高的生产率以及创造性探索和创新的环境。计算科学涵盖的大量主题意味着该领域非常广阔，新方法不断被发表。这些方法不仅涉及到核心的仿真技术，而且还涉及到依赖于仿真的问题。这些问题包括量化不确定性（即要求误差条），将模型与数据混合以做出更好的预测，解决逆问题（如果输出是Y，输入X是什么？），以及优化设计（例如，找到最符合空气动力学的车辆形状）。不幸的是，先进的新方法进入应用和工业实践的过程非常缓慢，其中一个原因是将数学算法应用于复杂的仿真模型是非常侵入性的;大多数情况下，它们不能将仿真代码视为"黑匣子"。它们通常需要重写软件，这非常耗时且昂贵。在我们的研究中，我们解决这个问题，通过使用自动生成的计算机代码进行模拟。其关键思想是以某种抽象的方式描述模拟算法（在仔细考虑关键方面是什么之后，看起来尽可能像底层数学），并使用专门的软件工具来自动构建计算机代码。当实现的某些方面需要改变时（例如，正在使用一种新型计算机），这些工具可以用来从抽象描述重建代码。这种灵活性极大地加快了高级算法在现实问题中的应用。考虑优化一级方程式赛车形状以最小化其阻力的例子。优化过程是高度侵入性的：它必须解决辅助问题以学习如何改进设计，并且能够在每次迭代时修改模拟中使用的形状。通常，这种侵入性需要对模拟软件进行大量修改。但是，通过存储空气动力学方程的符号表示，可以在我们的系统中生成优化所需的所有操作，而根本不需要重写或修改空气动力学代码。我们平台的研究目标是研究和推广这种方法，并制作公开可用的，可持续的开源软件，以确保其采用。该平台将使我们能够在软件方法方面取得进步，使我们能够继续在我们所从事的广泛应用领域获得工业和政府资金，包括航空航天和汽车行业，可再生能源，医药和外科手术，环境和制造业。

Computational science is a multidisciplinary research endeavour spanning applied mathematics, computer science and engineering together with input from application areas across science, technology and medicine. Advanced simulation methods have the potential to revolutionise not only scientific research but also to transform the industrial economy, offering companies a competitive advantage in their products, better productivity, and an environment for creative exploration and innovation.The huge range of topics that computational science encapsulates means that the field is vast and new methods are constantly being published. These methods relate not only to the core simulation techniques but also to problems which rely on simulation. These problems include quantifying uncertainty (i.e. asking for error bars), blending models with data to make better predictions, solving inverse problems (if the output is Y, what is the input X?), and optimising designs (e.g. finding a vehicle shape that is the most aerodynamic). Unfortunately, the process through which advanced new methods find their way into applications and industrial practice is very slow.One of the reasons for this is that applying mathematical algorithms to complex simulation models is very intrusive; mostly they cannot treat the simulation code as a "black box". They often require rewriting of the software, which is very time consuming and expensive. In our research we address this problem by using automating the generation of computer code for simulation. The key idea is that the simulation algorithm is described in some abstract way (which looks as much like the underlying mathematics as possible, after thinking carefully about what the key aspects are), and specialised software tools are used to automatically build the computer code. When some aspect of the implementation needs to change (for example a new type of computer is being used) then these tools can be used to rebuild the code from the abstract description. This flexibility dramatically accelerates the application of advanced algorithms to real-world problems.Consider the example of optimising the shape of a Formula 1 car to minimise its drag. The optimisation process is highly invasive: it must solve auxiliary problems to learn how to improve the design, and it be able to modify the shape used in the simulation at each iteration. Typically this invasiveness would require extensive modifications to the simulation software. But by storing a symbolic representation of the aerodynamic equations, all operations necessary for the optimisation can be generated in our system, without needing to rewrite or modify the aerodynamics code at all.The research goal of our platform is to investigate and promote this methodology, and to produce publicly available, sustainable open-source software that ensures its uptake. The platform will allow us to make advances in our software approach that enables us to continue to secure industrial and government funding in the broad range of application areas we work in, including aerospace and automotive sectors, renewable energy, medicine and surgery, the environment, and manufacturing.