Building a Framework for Predictive Science (1202.1056v1)

Abstract: Key questions that scientists and engineers typically want to address can be formulated in terms of predictive science. Questions such as: "How well does my computational model represent reality?", "What are the most important parameters in the problem?", and "What is the best next experiment to perform?" are fundamental in solving scientific problems. Mystic is a framework for massively-parallel optimization and rigorous sensitivity analysis that enables these motivating questions to be addressed quantitatively as global optimization problems. Often realistic physics, engineering, and materials models may have hundreds of input parameters, hundreds of constraints, and may require execution times of seconds or longer. In more extreme cases, realistic models may be multi-scale, and require the use of high-performance computing clusters for their evaluation. Predictive calculations, formulated as a global optimization over a potential surface in design parameter space, may require an already prohibitively large simulation to be performed hundreds, if not thousands, of times. The need to prepare, schedule, and monitor thousands of model evaluations, and dynamically explore and analyze results, is a challenging problem that requires a software infrastructure capable of distributing and managing computations on large-scale heterogeneous resources. In this paper, we present the design behind an optimization framework, and also a framework for heterogeneous computing, that when utilized together, can make computationally intractable sensitivity and optimization problems much more tractable.

Overview of Mystical Predictive Science Framework

This paper introduces a comprehensive software framework aimed at addressing key challenges within predictive science, specifically focusing on optimization, uncertainty quantification (UQ), and parallel computing. The framework discussed herein, mystic, provides a robust infrastructure for addressing global optimization problems prevalent in realistic physics, engineering, and materials models, which often require extensive computational resources and leverage high-performance computing clusters.

Mathematical Foundation and Software Design

The conceptual foundation set forth is Optimal Uncertainty Quantification (OUQ), which provides stringent methodologies for determining bounds of uncertainties based on available data and set assumptions. This framework is built on quantified optimization problems where the extreme values of interest pertain primarily to probabilities of system failure. The importance of tight upper bounds is emphasized, as these bounds must correspond closely to the actual Probability of Failure (PoF) of a given system.

The mystic software framework is positioned as an observable solution to solve these OUQ problems by employing global optimization methodologies. It facilitates rigorous sensitivity analyses through new algorithms and support for massively-parallel computations, thereby ensuring that complex optimization tasks become achievable. Moreover, the framework introduces methods for constraint imposition which significantly ease solving highly constrained problems by decoupling optimization and constraint solving processes.

Computational Implementation and Parallelization

The paper details mystic's implementation of comprehensive optimization algorithms that span parallel and distributed computing environments. By embedding pathos infrastructure, mystic ensures that computations leverage the dynamic interactions across heterogeneous resources. This is realized through model and cost function factories which convert scientific models into callable services that support distributed and parallel execution. Such features are expanded by pathos, a complementary framework focusing on job management across diverse computing architectures, enhancing user access to high-performance computations without extensive refactoring of code.

Notably, mystic incorporates advanced parallel map functions, enabling optimization tasks to be carried out concurrently on distributed nodes, significantly enhancing computational efficiency in large-scale models.

Noteworthy Computational Techniques

• Global Search Algorithms: Extended to support parallel execution, providing dynamic evaluations across large-scale resources.
• Parallel Launch of Nested Solvers: Exemplified by Buckshot and Lattice Solvers, facilitating simultaneous optimization across different initial configurations, thereby expanding solution space exploration.
• Probability and Uncertainty Toolkit: Allows integration of uncertainty principles within optimization tasks, aiding in accurate model predictions and robust validations.

Implications and Future Perspectives

The development and deployment of this framework have substantial practical implications in the realms of engineering and scientific computation, particularly where resource-intensive models are the norm. The framework's ability to effectively manage the execution of complex algorithms across multiple compute nodes introduces flexibility and efficiency in conducting UQ and optimizations.

Future developments intended for mystic and pathos include expansion of optimization strategies to handle tasks with multi-dimensional correlations and adaptive constraint adjustments. Additionally, advancements in hierarchical partitioning algorithms promise enhanced capability for identifying high-sensitivity regions within model parameter spaces.

Conclusion

This work represents a pivotal advancement in tackling the complexities inherent in predictive science models. With mystic and its underpinning frameworks, researchers can access a configurable, scalable infrastructure to manage computational tasks with high fidelity and reduced computational cost. The robust execution environment offered by mystic, combined with tightly integrated OUQ principles, positions it as a dependable resource for conducting sophisticated predictive analyses.