Abstract:
We show-to our own surprise-that total electronic energies for a family of m × n rectangular graphene flakes can be very accurately represented by a simple function of the structural parameters m and n with errors not exceeding 1 kcal/mol. The energies of these flakes, usually referred to as multiple zigzag chains Z(m,n), are computed for m, n < 21 at their optimized geometries using the DFTB3 methodology. We have discovered that the structural parameters m and n (and their simple algebraic functions) provide a much better basis for the energy decomposition scheme than the various topological invariants usually used in this context. Most terms appearing in our energy decomposition scheme seem to have simple chemical interpretations. Our observation goes against the well-established knowledge stating that many-body energies are complicated functions of molecular parameters. Our observations might have far-reaching consequences for building accurate machine learning models.
摘要:
令我们惊讶的是,一个m×n矩形石墨烯薄片家族的总电子能量可以通过结构参数m和n的简单函数非常准确地表示,误差不超过1kcal/mol。这些薄片的能量,通常称为多个锯齿形链Z(m,n),是为m计算的,使用DFTB3方法在其优化的几何形状下n<21。我们已经发现,结构参数m和n(及其简单的代数函数)为能量分解方案提供了比通常在这种情况下使用的各种拓扑不变量更好的基础。在我们的能量分解方案中出现的大多数术语似乎都具有简单的化学解释。我们的观察与公认的知识背道而驰,该知识表明许多身体能量是分子参数的复杂函数。我们的观察可能会对构建准确的机器学习模型产生深远的影响。