<script type="text/javascript">
<!--
document.write('<div id="oa_widget"></div>');
document.write('<script type="text/javascript" src="https://www.openaire.eu/index.php?option=com_openaire&view=widget&format=raw&projectId=ukri________::3d13f34a31d8cbc13175962d2793a367&type=result"></script>');
-->
</script>
This project will open new horizons in materials science by developing the high-risk techniques of computational geometry that are needed for an efficient design of solid crystalline materials (crystals). Our project team combines the crucial skills in computer science, mathematics, and crystal chemistry to cause the necessary paradigm shift. Every year, pharmaceutical companies refuse many crystalline drugs during the early stages of trials because of their poor solubility in the human body. Billions of pounds and years of work could be saved if the physical properties of a crystal can be guaranteed by design. The first obstacle is the insufficiency of discrete classifications that were established for periodic crystals already in the 19th century. Now, the Cambridge Crystallographic Data Centre, the industry partner in our project, curates the world's largest collection, the Cambridge Structural Database (CSD), of more than 1.1M existing crystals in a conventional form consisting of an elementary pattern (a motif of atoms) and a linear basis generating the same underlying crystal lattice. This conventional form works in the ideal world where all measurements have infinite precision. However, even tiny atomic displacements (e.g., from measurement error) can break the symmetry of a crystal and make it incomparable with its idealised version. As a result, experimental databases keep growing by accepting near-duplicates of known materials because all available comparison tools are slow or require manually chosen parameters. Recent research from our project team revealed five pairs of suspected duplicates even in the well-curated CSD because our new invariants provably distinguish all generic crystals up to isometry preserving rigid forms of crystals. This project will tackle the unresolved challenge of making the invariants invertible in the sense that any set of invariants gives rise to a well-defined periodic crystal, like a blueprint of a new building which is sufficient for full construction. The simple case of triangles illustrates the challenges of invertibility. The list of side lengths of a triangle is an isometry invariant and can be represented as a point in the positive octant of 3-dimensional space. Not all points with positive coordinates can represent a triangle, but simple inequality conditions define those points that do represent triangles. No equivalent conditions are known for isometry invariants of periodic crystals. The second obstacle is the established paradigm of materials discovery based on trial-and-error of mixing components in the lab or on lottery-type searches, when a huge space of parameters is randomly sampled for subsequent slow optimisation without guarantees of success. What if we could locate the most promising spots in this vast space, where we can confidently find all desired crystals? The exciting and disruptive idea of inverse design is to start from a target property and test only a shortlist of potential candidates. For crystals, a key property is their thermodynamic stability, which is not universally defined for all types of crystals and is currently explored using various approximate energy functions tuned for specific compositions. The increasing complexity of energy functions makes their computation slower without reducing the search space. Imagine that the most promising crystals are peaks of mountains on a new planet: the past way to find such highest peaks is to randomly throw millions of 'bugs' that slowly move to the higher ground, and most of them become stuck on the much more numerous small hills (local maxima), rather than the few highest peaks (global maxima). Following this analogy, our radically new method is to push down the atmospheric clouds and watch the highest peaks appear on a global scale. This 'cloud pushing' will be realised by a simple geometric function whose analytically computable local extrema approximate realistic crystals.
This project will open new horizons in materials science by developing the high-risk techniques of computational geometry that are needed for an efficient design of solid crystalline materials (crystals). Our project team combines the crucial skills in computer science, mathematics, and crystal chemistry to cause the necessary paradigm shift. Every year, pharmaceutical companies refuse many crystalline drugs during the early stages of trials because of their poor solubility in the human body. Billions of pounds and years of work could be saved if the physical properties of a crystal can be guaranteed by design. The first obstacle is the insufficiency of discrete classifications that were established for periodic crystals already in the 19th century. Now, the Cambridge Crystallographic Data Centre, the industry partner in our project, curates the world's largest collection, the Cambridge Structural Database (CSD), of more than 1.1M existing crystals in a conventional form consisting of an elementary pattern (a motif of atoms) and a linear basis generating the same underlying crystal lattice. This conventional form works in the ideal world where all measurements have infinite precision. However, even tiny atomic displacements (e.g., from measurement error) can break the symmetry of a crystal and make it incomparable with its idealised version. As a result, experimental databases keep growing by accepting near-duplicates of known materials because all available comparison tools are slow or require manually chosen parameters. Recent research from our project team revealed five pairs of suspected duplicates even in the well-curated CSD because our new invariants provably distinguish all generic crystals up to isometry preserving rigid forms of crystals. This project will tackle the unresolved challenge of making the invariants invertible in the sense that any set of invariants gives rise to a well-defined periodic crystal, like a blueprint of a new building which is sufficient for full construction. The simple case of triangles illustrates the challenges of invertibility. The list of side lengths of a triangle is an isometry invariant and can be represented as a point in the positive octant of 3-dimensional space. Not all points with positive coordinates can represent a triangle, but simple inequality conditions define those points that do represent triangles. No equivalent conditions are known for isometry invariants of periodic crystals. The second obstacle is the established paradigm of materials discovery based on trial-and-error of mixing components in the lab or on lottery-type searches, when a huge space of parameters is randomly sampled for subsequent slow optimisation without guarantees of success. What if we could locate the most promising spots in this vast space, where we can confidently find all desired crystals? The exciting and disruptive idea of inverse design is to start from a target property and test only a shortlist of potential candidates. For crystals, a key property is their thermodynamic stability, which is not universally defined for all types of crystals and is currently explored using various approximate energy functions tuned for specific compositions. The increasing complexity of energy functions makes their computation slower without reducing the search space. Imagine that the most promising crystals are peaks of mountains on a new planet: the past way to find such highest peaks is to randomly throw millions of 'bugs' that slowly move to the higher ground, and most of them become stuck on the much more numerous small hills (local maxima), rather than the few highest peaks (global maxima). Following this analogy, our radically new method is to push down the atmospheric clouds and watch the highest peaks appear on a global scale. This 'cloud pushing' will be realised by a simple geometric function whose analytically computable local extrema approximate realistic crystals.
<script type="text/javascript">
<!--
document.write('<div id="oa_widget"></div>');
document.write('<script type="text/javascript" src="https://www.openaire.eu/index.php?option=com_openaire&view=widget&format=raw&projectId=ukri________::3d13f34a31d8cbc13175962d2793a367&type=result"></script>');
-->
</script>