BRIDGING PLASMA SCALES USING DATA-DRIVEN MODEL IDENTIFICATION
Saved in:
| Title: | BRIDGING PLASMA SCALES USING DATA-DRIVEN MODEL IDENTIFICATION |
|---|---|
| Authors: | Vasey, Gina |
| Committee Members: | O'Shea, Brian; Christlieb, Andrew; Lei, Huan; Anthony, Rebecca |
| Summary: | Plasma is a state of matter studied in many fields in part due to its influence across a widespan of spatial and temporal scales. This can range from plasma effects in star and galaxy evolution in astrophysics to fusion power research. Across these fields, however, common needs and questions arise. Plasma can be described using many different models, each accurate under different assumptions or at specific scales. From this variance in representation arises questions regarding how plasma act under extreme conditions, conditions where the best representative model is ambiguous, or when multiple regimes of behavior are traversed. Each regime requires different representations to accurately model behavior, where some techniques work by modeling the evolution of the particle distribution function while others work with moments requiring simplifying assumptions to be made regarding which effects are significant. Data-driven model identification techniques based on sparse-regression have recently been developed to identify governing partial differential equations (PDEs) from limited data. Such methods do not require a large body of training data and scale well with complex systems with many state variables, making them well-suited to work in problem spaces where limited highfidelity data is available and simplified fluid models are desirable. This dissertation explores how one such method, weak sparse identification of nonlinear dynamics (WSINDy), translates to plasma applications. First, variations of ideal MHD simulations across a variety of initial conditions are considered in relation to Shannon information entropy, providing insight into how WSINDy tends toward reduced representations in cases where the data is low in information. Next, challenges in scale bridging between particle and fluid behavior are addressed by analyzing fluid equation recovery as identified from particle simulations with varying degrees of diffusive behavior. Finally, a real-world application is explored for quantifying hall conductivity of the inner MITL on Sandia’s Z-Machine. |
| URL: | https://doi.org/doi:10.25335/ajph-eg77 |
| Database: | OpenDissertations |
| Abstract: | Plasma is a state of matter studied in many fields in part due to its influence across a widespan of spatial and temporal scales. This can range from plasma effects in star and galaxy evolution in astrophysics to fusion power research. Across these fields, however, common needs and questions arise. Plasma can be described using many different models, each accurate under different assumptions or at specific scales. From this variance in representation arises questions regarding how plasma act under extreme conditions, conditions where the best representative model is ambiguous, or when multiple regimes of behavior are traversed. Each regime requires different representations to accurately model behavior, where some techniques work by modeling the evolution of the particle distribution function while others work with moments requiring simplifying assumptions to be made regarding which effects are significant. Data-driven model identification techniques based on sparse-regression have recently been developed to identify governing partial differential equations (PDEs) from limited data. Such methods do not require a large body of training data and scale well with complex systems with many state variables, making them well-suited to work in problem spaces where limited highfidelity data is available and simplified fluid models are desirable. This dissertation explores how one such method, weak sparse identification of nonlinear dynamics (WSINDy), translates to plasma applications. First, variations of ideal MHD simulations across a variety of initial conditions are considered in relation to Shannon information entropy, providing insight into how WSINDy tends toward reduced representations in cases where the data is low in information. Next, challenges in scale bridging between particle and fluid behavior are addressed by analyzing fluid equation recovery as identified from particle simulations with varying degrees of diffusive behavior. Finally, a real-world application is explored for quantifying hall conductivity of the inner MITL on Sandia’s Z-Machine. |
|---|
