Simulation of Mechanical Behaviour of Martensites using Multicore Technology

马氏体-奥氏体相变是冶金中最重要的相变之一，存在于许多材料中。通常，它涉及从高温的延性(奥氏体)相到低温的硬(马氏体)相的对称性破坏。相变涉及应变，在钢铁和超弹性形状记忆合金等实际应用中，两相共存，相变过程快，微观结构为纳米尺度，适合进行分子动力学研究。然而，应变影响相的稳定性，相变相(惯性面)之间存在不合理的界面，需要灵活的初始条件和边界条件。假设边界是完全刚性的，优雅的数学分析得出结论，结构是由最小化边界能量决定的。然而，我们的初步模拟表明这是边界条件的伪影：如果允许软奥氏体变形以适应不断增长的马氏体，则会出现截然不同的结果。描述足够的周围材料以匹配弹性和原子细节，使得这些模拟非常计算密集。并行超级计算机是一种选择，但通过专用多核处理器可以获得更具成本效益的解决方案，这些处理器是我们集团未来几年将推出的标准台式机硬件。充分利用共享内存的多核机器需要对代码进行大量修改：信息需要在处理器之间传输，而要做到这一点需要并行编码专业知识。并行化策略也因原子间相互作用势的类型而异。金属最适合用多体、短程相互作用来描述，这需要一种不同于固定键或离子作用力的势的策略。我们建议重新设计一个模拟程序，用于研究马氏体如何转变，了解奥氏体和马氏体之间的界面采取什么结构，以及材料参数是如何影响这一结构的。为此，我们将利用我们在并行编码、分子动力学和潜在设计方面的经验。

The martensite-austenite phase transformation is one of the most important in metallurgy, existing in many materials. Typically, it involves a symmetry-breaking from a high temperature, ductile (austenite) phase to a low temperature, hard (martensite) phase. The transformation involves strain, and in practical applications such as steels and superelastic shape memory alloys both phases coexist.The transformation process is fast, and the microstructures are nanoscaled, so molecular dynamics study is appropriate. However, strain affects phase stability, and there are irrational interfaces between transforming phases (habit plane) so flexible initial and boundary conditions are required. Elegant mathematical analysis assuming perfectly rigid boundaries concluded that the structure is determined by minimising boundary energy. However, our preliminary simulations shows this to be an artifact of the boundary conditions: if the soft austenite is allowed to deform to accommodate the growing martensite very different results occur.Describing adequate surrounding material to match both elasticity and atomistic detail, makes these simulations very computationally intense. A parallel supercomputer is an option, but a far more cost-effective solution is becoming available through dedicated multicore processors, the standard desktop hardware coming into our group over the next few years. Fully utilising a shared-memory multicore machine requires considerable modification to the code: information needs to be transferred between processors, and doing this efficiently requiresparallel coding expertise. Parallelisation strategies also vary withthe type of interatomic potential. Metals are best described by many-body, short ranged interactions which call for a different strategy from potentials involving fixed bonds or ionic forces.We propose to re-engineer a simulation code suitable for study of how martensites transform, to understand what structure the interfaces between austenite and martensite takes, and how this is affected by material parameter. To do so we will use our experience in parallel coding, molecular dynamics and potential design.