Climate models have a large number of inputs and outputs. In addition, diverse parameters sets can match observations similarly well. These factors make calibrating the models difficult. But as the Earth enters a new climate regime, parameters sets may cease to match observations. History matching is necessary but not sufficient for good predictions. We seek a 'Pareto optimal' ensemble of calibrated parameter sets for the CCSM climate model, in which no individual criteria can be improved without worsening another. One Multi Objective Genetic Algorithm (MOGA) optimization typically requires thousands of simulations but produces an ensemble of Pareto optimal solutions. Our simulation budget of 500-1000 runs allows us to perform the MOGA optimization once, but with far fewer evaluations than normal. We devised an analytic test problem to aid in the selection MOGA settings. The test problem's Pareto set is the surface of a 6 dimensional hypersphere with radius 1 centered at the origin, or rather the portion of it in the [0,1] octant. We also explore starting MOGA from a space-filling Latin Hypercube sample design, specifically Binning Optimal Symmetric Latin Hypercube Sampling (BOSLHS), instead of Monte Carlo (MC). We compare the Pareto sets based on: their number of points, N, larger is better; their RMS distance, d, to the ensemble's center, 0.5553 is optimal; their average radius, {mu}(r), 1 is optimal; their radius standard deviation, {sigma}(r), 0 is optimal. The estimated distributions for these metrics when starting from MC and BOSLHS are shown in Figs. 1 and 2.
Exascale systems will have hundred thousands of compute nodes and millions of components which increases the likelihood of faults. Today, applications use checkpoint/restart to recover from these faults. Even under ideal conditions, applications running on more than 50,000 nodes will spend more than half of their total running time saving checkpoints, restarting, and redoing work that was lost. Redundant computing is a method that allows an application to continue working even when failures occur. Instead of each failure causing an application interrupt, multiple failures can be absorbed by the application until redundancy is exhausted. In this paper we present a method to analyze the benefits of redundant computing, present simulation results of the cost, and compare it to other proposed methods for fault resilience.
While advances in manufacturing enable the fabrication of integrated circuits containing tens-to-hundreds of millions of devices, the time-sensitive modeling and simulation necessary to design these circuits poses a significant computational challenge. This is especially true for mixed-signal integrated circuits where detailed performance analyses are necessary for the individual analog/digital circuit components as well as the full system. When the integrated circuit has millions of devices, performing a full system simulation is practically infeasible using currently available Electrical Design Automation (EDA) tools. The principal reason for this is the time required for the nonlinear solver to compute the solutions of large linearized systems during the simulation of these circuits. The research presented in this report aims to address the computational difficulties introduced by these large linearized systems by using Model Order Reduction (MOR) to (i) generate specialized preconditioners that accelerate the computation of the linear system solution and (ii) reduce the overall dynamical system size. MOR techniques attempt to produce macromodels that capture the desired input-output behavior of larger dynamical systems and enable substantial speedups in simulation time. Several MOR techniques that have been developed under the LDRD on 'Solution Methods for Very Highly Integrated Circuits' will be presented in this report. Among those presented are techniques for linear time-invariant dynamical systems that either extend current approaches or improve the time-domain performance of the reduced model using novel error bounds and a new approach for linear time-varying dynamical systems that guarantees dimension reduction, which has not been proven before. Progress on preconditioning power grid systems using multi-grid techniques will be presented as well as a framework for delivering MOR techniques to the user community using Trilinos and the Xyce circuit simulator, both prominent world-class software tools.
Our ability to field useful, nano-enabled microsystems that capitalize on recent advances in sensor technology is severely limited by the energy density of available power sources. The catalytic nanodiode (reported by Somorjai's group at Berkeley in 2005) was potentially an alternative revolutionary source of micropower. Their first reports claimed that a sizable fraction of the chemical energy may be harvested via hot electrons (a 'chemicurrent') that are created by the catalytic chemical reaction. We fabricated and tested Pt/GaN nanodiodes, which eventually produced currents up to several microamps. Our best reaction yields (electrons/CO{sub 2}) were on the order of 10{sup -3}; well below the 75% values first reported by Somorjai (we note they have also been unable to reproduce their early results). Over the course of this Project we have determined that the whole concept of 'chemicurrent', in fact, may be an illusion. Our results conclusively demonstrate that the current measured from our nanodiodes is derived from a thermoelectric voltage; we have found no credible evidence for true chemicurrent. Unfortunately this means that the catalytic nanodiode has no future as a micropower source.