Micron-Scale Atomistic Studies of Shock Ejecta Production using BlueGene/L

Los Alamos researchers have recently completed a series of simulations of shock ejecta production in copper, using the entire BlueGene/L machine at Livermore, with 212,992 CPUs after its recent (fall 2007) expansion. The longest such simulation ran continuously for 88 hours (400,000 timesteps, or one nanosecond of simulated time) with nearly 800 million atoms, generating 101 checkpoint dumps (4.18 TB of data) and 3024 images (14.5 GB). The nearly 20 million CPU-hours of this simulation alone are equivalent to more than 2 CPU-millennia, setting a new standard for HPC stability.

These simulations utilized a quasi-2D geometry, 5.7 microns in length and with a 2.23-micron periodic cross-section (but only 1.5 nm in the third direction, also periodic, to preserve 3D equation-of-state and transport properties), with copper atoms described by an embedded atom method (EAM) potential. The free surface opposite to the impact plane is initially machined with a profile matching that measured in recent tin ejecta experiments at LANL*, with an approximately 1:40 length scaling so that the ~1 micron experimental amplitude is ~25 nm. Earlier MD simulations with a single machining groove demonstrated jet formation, with subsequent necking instabilities leading to jet breakup and droplet formation at later times. This fragmentation and atomization process is difficult to study experimentally, and various theories have been proposed; atomistic-level simulations will contribute to the development of physics-based models as part of the LANL “Science@Scale” effort (see Sept 2007 ASC eNews article). In particular, the dependence of ejecta production and transport on shock pressure (e.g. below and above the shock melting transition, to study material strength effects) and background gas (with either a vacuum as in previous simulations, or an inert gas atmosphere) have been studied.

Fig. 1. Series of 8 snapshots, one every 10 ps, for shock ejection from a roughened copper surface

Fig. 2. Snapshot of 300 ps after shock encounters the free surface

Time sequences from two such simulations are shown here, with the initial surface roughness and shock pressure equal in both cases. Color represents the local density in each case. For shock ejection into a vacuum, we observe jet formation from small-scale features at early times (Fig. 1.: Series of 8 snapshots, one every 10 ps, for shock ejection from a roughened copper surface). These jets, surrounded by a cloud of atomic ejecta, expand and subsequently merge or break up over time. Also at later times (Fig. 2. : snapshot 300 ps after shock encounters the free surface), the longer-wavelength machining defects produce a Richtmyer-Meshkov instability, with three “bubbles” of vacuum pushing into the copper surface, and three large “spikes” of copper protruding out, with smaller-scale jets superimposed throughout. The bubbles and spikes are very asymmetric, as one would expect for Atwood number A=1.

Fig. 3. Series of 9 snapshots, one every 50 ps, for Richtmyer-Meshkov instability development as a shock wave is transmitted from copper to a dense (0.5 g/cc) gas

On the other hand, the presence of a background gas (A < 1) inhibits jetting and leads to a more symmetric Richtmyer-Meshkov instability, as seen in the second simulation (Fig. 3.: Series of 9 snapshots, one every 50 ps, for Richtmyer-Meshkov instability development as a shock wave is transmitted from copper to a dense (0.5 g/cc) gas.) In the third and fourth frames, one can clearly see the transmitted gas shock (still nonplanar), as well as a complex interaction of rarefaction fans from the roughened Cu/gas interface.

Contacts: Timothy C. Germann, James E. Hammerberg, and Guy Dimonte.

*M. B. Zellner et al., “Effects of shock-breakout pressure on ejection of micron-scale material from shocked tin surfaces,” J. Appl. Phys. 102, 013522 (2007).