Uncertainty Quantification of Molecular Dynamics Simulations for Crosslinked Polymers

As a key component of carbon fiber composites, crosslinked polymers are an important class of materials that have helped to revolutionize the automotive, boating, and aerospace industries. Such systems derive many of their properties from the fact that unlike metals (i.e. crystals), polymers form random networks with little long-range structure at the microscale. As a result, the bulk properties of crosslinked systems depend strongly on the detailed statistics of their molecular configurations but are effective averages taken over $O(10^{23})$ atoms, i.e. the order of Avogadro's number. Unsurprisingly, this disparity of scales creates considerable problems for modelers.

Motivated by this observation, I will discuss current projects on uncertainty quantification for simulations of crosslinked polymers. Generally speaking, simulations are a tool of choice for representing such materials, since computers can resolve the statistics of many-body systems. Nonetheless, current technological limitations mean that heroic models are still $17$ orders of magnitude shy of bulk limits. Thus, considerable uncertainty accompanies any simulated prediction. To address these issues, we propose a sequence of analyses in which we verify that simulated data is informative before estimating uncertainties. Key steps in the analysis invoke concepts from physics and chemistry. Therefore our current analysis is limited to a case study of the glass transition temperature, although extensions to additional material properties should be straightforward.

This is a collaboration with Andrew Dienstfrey (NIST Boulder) and Steve Christensen, Andrea Browning, and Sam Tucker (Boeing).

Presentation Slides: PDF

Note: Visitors from outside NIST must contact Cathy Graham; (301) 975-3800; at least 24 hours in advance.