Smart Data Approaches for the Inverse Design of Soft Materials

设计新型智能材料的传统方法需要昂贵的实验来探索大量的参数组合。这个过程可以在“虚拟实验室”中通过计算机模拟来加速，但它需要一种非传统的方法。直到最近，计算机模拟的重点一直是正演问题：给定诸如分子结构、组成等参数的值，找出材料的性质。但是，对于加速新材料的设计和进步来说，在给定一组所需的材料特性的情况下，我们寻求产生这些特性的参数值的反问题更为关键。这个逆设计问题是一个非常具有挑战性的高维全局优化问题，是这个项目的重点之一。设想的数学和计算框架将广泛适用于各种各样的材料，如嵌段共聚物和纳米结构软材料。该项目将对教育产生重大影响，因为来自代表性不足的群体和本科生的学生的参与将会增加。传统上，计算机模拟是对一些受过教育的模型参数选择进行的。每次模拟通常都是昂贵的，因此，即使使用最先进的计算资源，也只有有限数量的模拟对现实模型是可行的。这是一个特别严重的问题，当参数空间是中等到高维，是不可能充分探索该空间。在这个项目中，首席研究员将开发一种概率方法来解决逆向设计问题，为正向问题设计优越的方法，并为构建计算效率更高的粗粒度模型执行统计推断。这是一种智能数据方法，其中使用关于目标函数的先验信念进行优化或推断（基于先验评估）来指导参数的采样将首次在多嵌段共聚物的场理论模型中进行探索。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The traditional approach to the design of new, smart materials requires costly experimentation to explore a vast combination of parameters. This process can be accelerated in a "virtual lab" with computer simulations but it requires a non-traditional approach. Until recently, the focus of computer simulations has been the forward problem: given values of the parameters such as molecular architecture, composition, etc., find the material properties. But the inverse problem in which given a desired set of material properties we seek values of the parameters that give rise to those properties, is more critical for accelerating the design and advancement of new materials. This inverse design problem, which is an extraordinarily challenging, high-dimensional global optimization problem, is one of the main thrusts of this project. The envisioned mathematical and computational framework will be broadly applicable to a wide variety of materials such as block copolymers and nano-structured soft materials. This project will have a substantial impact in education with an increased participation of students from underrepresented groups and undergraduates Traditionally, computer simulations are performed for some educated choices of the model parameters. Each simulation is in general costly so that, even with state of the art computing resources, only a limited number of simulations is feasible for realistic models. This is a particularly serious problem when the parameter space is moderate to high dimensional and is impossible to fully explore that space. In this project, the principal investigator will develop a probabilistic approach to solve the inverse design problem, to design superior methods for the forward problem, and to perform statistical inference for the construction of computationally more efficient, coarse-grained models. This is a smart-data approach, in which the use of prior belief about the objective function to be optimized or to be inferred (based prior evaluations) to guide the sampling of the parameter will be explored for the first time in the context of field-theoretic models of multi-block copolymers.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.