Accelerating Scientific Discovery Through Computation and Visualization II

Abstract

This is the second in a series of articles describing a wide variety of projects at NIST that synergistically combine physical science and information science. It describes, through examples, how the Scientific Applications and Visualization Group (SAVG) at NIST has utilized high performance parallel computing, visualization, and machine learning to accelerate research. The examples include scientific collaborations in the following areas: (1) High Precision Energies for few electron atomic systems, (2) Flows of suspensions, (3) X-ray absorption, (4) Molecular dynamics of fluids, (5) Nanostructures, (6) Dendritic growth in alloys, (7) Screen saver science, (8) genetic programming.

Keywords: FEFF; FeffMPI; Hylleraas-Configuration Interaction; Lennard-Jones; QDPD; discovery science; genetic programming; immersive environments; nanostructures; parallel computing; scientific visualization; screen saver science.

Fig. 1

Abstraction of research loop.

Fig. 2

Cloud example.

Fig. 3

Click to turn on streamlines.

Fig. 4

Flashlight DSO enables the user to shine a flashlight on a spot within the immersive visualization by merely pointing to it.

Fig. 5

Collaboration process.

Fig. 6

Hy-CI Scaling with Cluster Size.

Fig. 7

Motion of a suspension of ellipsoids subject to shear. The single ellipsoid rotation is a well known phenomenon called Jeffery’s Orbits.

Fig. 8

Dense suspension of ellipsoids similar to typical aggregate contribution in concrete. Jeffery’s Orbits are suppressed and the alignment between ellipsoids is enhanced.

Fig. 9

Spherical aggregates subject to a downward applied force. Sphere diameter is approximately one-half the gap spacing between reinforcing rebars. The “jamming” of the aggregates between cylindrical rebars (in black) is observed.

Fig. 10

Spherical aggregates moving around stationary cylindrical rebars (in black). Sphere diameter is one-fifth the gap spacing. No “jamming” is observed.

Fig. 11

Coaxial Rheometer. An inner cylinder (in black) subject to an applied torque rotates, causing the motion of the spheres.

Fig. 12

(a) The 87 atom cluster used to calculate the XANES of crystalline Ge. (b) A similar cluster of 87 atoms of aGe from the CRN displayed with the same length scale.

Fig. 13

The full 519 atom cluster of aGe from the continuous random network with a typical cluster of 87 atoms highlighted in the interior.

Fig. 14

Results of the crystalline Ge calculation (upper solid line), the ensemble average over 20 sites in the aGe CRN (dashed line), and an illustration of the site-to-site variation in the aGe (five offset solid lines).

Fig. 15

Snapshot of initial configuration of a two-component fluidfluid interface system. One component is shown in blue and the other component is shown in yellow.

Fig. 16

Two-component fluid-fluid interface system at equilibrium. As seen in the figure, equilibrium is characterized by two distinct interfaces.

Fig. 17

Plot of the x, y, and z components of the stress tensor. As can be seen from the distinct dips in the curves, the greatest change in these components occurs in the interfacial regions.

Fig. 18

Plot of the particle density. The fact that the unlike particles do not like to mix is evident from the distinct decrease (the dips in the curve) in the particle density in the interfacial regions.

Fig. 19

Pyramid structure.

Fig. 20

A distribution of the computation via layers.

Fig. 21

Execution time versus number of processors for concentric spheres problem.

Fig. 22

A section of a nanostructure (top) with orbitals (bottom).

Fig. 23

A simulated copper-nickel dendrite computed over a uniform 3D grid of 500³ points. Two of the axes have been mirrored resulting in this image of 1000 × 1000 × 500 (x, y, z) points. The physical dimensions of this dendrite are approximately 35 μm by 35 μm by 17.5 μm. The color of the dendrite indicates the concentration of copper in the copper-nickel alloy at each point on the surface.

Fig. 24

A 2D (x, y) slice near the base of the dendrite shown in Fig. 23.

Fig. 25

Several more 2D slices of the dendrite in Fig. 23. Each z value indicates by an integer the position of the slice along the z axis.

Fig. 26

A high-level view of the Screen Saver Science architecture. Clients access a shared object space (JavaSpace) to retrieve tasks to perform, return results, send messages to other clients, broadcast status messages for general use, and exchange other messages as needed. Large data sets will be stored on disk with client access to these files coordinated through entries in the shared object spaces.

Fig. 27

Results of an example regression run. Values of the GP-discovered function are shown at points that were used by the GP system to determine fitness.

Fig. 28

Visualization of a population.

Fig. 29

A population showing a substantial loss of diversity.