Session: K9-11 Phonon Modeling and Machine Learning for Thermal Transport

Paper Number: 108246

108246 - Phonon Scattering and Vibrational Localization for Dense 3d Nanoparticle Composites 

We disclose a message passing interface (MPI) parallelized 3D implementation of the frequency domain perfectly matched layer method (FDPML) method and use it to study scattering of phonon modes in dense nano particulate composites.  The approach allows us to utilize realistic force constants from first-principles calculations and simulate large 3D domains of embedded nanoparticles for systems with at least 100 million atoms.  We demonstrate analogs of our groups recent 2D work on the same topic; in particular we show that the individual scattering regime holds until >10% volume fraction over a wide range of wavelengths.   A consequence of this is that we show much less expensive analytic approaches achieve reasonable accuracy in this regime.  For the case of embedded nanoparticles that have smaller cutoff frequency than the matrix (i.e. heavy particles) we find no localized modes until beyond the cutoff frequency for the heavy particle.  
We also demonstrate a technique for additional parallelization of the FDPML method whereby polarization/wavevector information is lost in favor of a frequency-based description of transport.  This allows the method to be compared with the more popular Green’s function approach, and the transmission functions it calculates - as well as to very efficiently calculate transport coefficients.

Presenting Author: Joseph Feser University of Delaware