Archis S. Joglekar

Inertial confinement fusion implosion design requires simultaneous optimization of strongly coupled target and driver parameters across high-dimensional design spaces. Existing automated design approaches typically rely on nondifferentiable radiation-hydrodynamics codes treated as black boxes, making optimization increasingly expensive as dimensionality grows. In this work, we present a differentiable simulation approach for high-dimensional inverse design of inertial confinement fusion implosions. Automatic differentiation through a differentiable implosion physics model, driven by an external pressure pulse, provides gradients of implosion objectives with respect to design parameters, enabling gradient-based optimization. The framework is applied to 25 kJ OMEGA-scale direct-drive implosions, optimizing 500-parameter laser pulses across sampled target geometries. The optimized pulse recovers a near-isentropic rise to peak power without that structure (i.e., pulse shape) being imposed. Neural-network pulse parametrizations are additionally explored as a means of accelerating design-space exploration. These results establish differentiable implosion modeling as a promising tool for inertial confinement fusion design, while motivating further work on adjoint robustness and higher-fidelity differentiable simulators.

Optical techniques for spatiotemporal control can produce laser pulses with custom amplitude, phase, or polarization structure. In nonlinear optics and plasma physics, the use of structured pulses typically follows a forward design approach, in which the efficacy of a known structure is analyzed for a particular application. Inverse approaches, in contrast, enable the discovery of new structures with the potential for superior performance. Here, an implementation of the unidirectional pulse propagation equation that supports automatic differentiation is combined with gradient-based optimization to design structured pulses with features that are advantageous for a range of nonlinear optical and plasma-based applications: (1) a longitudinally uniform intensity over an extended region, (2) a superluminal intensity peak that travels many Rayleigh ranges with constant duration, spot size, and amplitude, and (3) a laser pulse that ionizes a gas to form a uniform column of plasma. In the final case, optimizing the full spatiotemporal structure improves the performance by a factor of 15 compared to optimizing only spatial or only temporal structure, highlighting the advantage of spatiotemporal control.

The training of neural networks, the workhorses of modern ML, relies on a key, enabling technology called Automatic Differentiation (AD). This talk will show how to leverage AD with physical, hpc-scale simulations in order to solve typical plasma physics research questions. We will motivate and discuss the integration of AD into a 1D1V Vlasov-Fokker-Planck code in order to make it "differentiable". This enables the usage of gradient based techniques to solve inverse/optimization problems. It also puts mathematical modeling on an equal footing with neural networks and enables seamless mixing and matching between the two. We discuss two examples of this. First, we turn the search for new physics from a parameter scan into an inverse problem and find a novel, nonlinear kinetic effect. Next, we model these dynamics using a fluid code by adding a machine-learned closure that is trained by minimizing the error between the ML+fluid and kinetic simulations. The training is performed on a set of single-wavelength simulations and produces stable dynamics that scale to 100x larger geometries in time and space.

Optical techniques for spatiotemporal control can produce laser pulses with custom amplitude, phase, or polarization structure. In nonlinear optics and plasma physics, the use of structured pulses typically follows a forward design approach, in which the efficacy of a known structure is analyzed for a particular application. Inverse approaches, in contrast, enable the discovery of new structures with the potential for superior performance. Here, an implementation of the unidirectional pulse propagation equation that supports automatic differentiation is combined with gradient-based optimization to design structured pulses with features that are advantageous for a range of nonlinear optical and plasma-based applications: (1) a longitudinally uniform intensity over an extended region, (2) a superluminal intensity peak that travels many Rayleigh ranges with constant duration, spot size, and amplitude, and (3) a laser pulse that ionizes a gas to form a uniform column of plasma. In the final case, optimizing the full spatiotemporal structure improves the performance by a factor of 15 compared to optimizing only spatial or only temporal structure, highlighting the advantage of spatiotemporal control.

We present simulations of heat flow relevant to gaspipe experiments on the National Ignition Facility to investigate kinetic effects on transport phenomena. D2 and neopentane (C5H12) filled targets are used to study the laser preheat stage of a MagLIF scheme where an axial magnetic field is sometimes applied to the target. Simulations were done with the radiation-MHD code HYDRA with a collision-dominated fluid model and the SNB nonlocal electron thermal conduction model. Using the SNB model to evolve the electron temperature increased the heat front propagation of neopentane gas targets compared to a local model by limiting radial heat flow. This increases electron temperature near the axis, which decreases laser absorption. We find that the effect of heat flow models on temperature profiles and laser propagation is modest. Beyond the SNB model, we utilize HYDRA to initialize plasma conditions for the Vlasov–Fokker–Planck K2 code. We run K2 until a quasi-steady state is reached and examine the impact of kinetic effects on heat transport. Although axial heat flow is well predicted by fluid models, the fluid model consistently overpredicts radial heat flow up to 150% in regions with the largest temperature gradient of D2 filled gaspipes. On the other hand, the SNB nonlocal electron conduction model is found to be adequate for capturing kinetic heat flow in gaspipes.

The achievement of ignition in the laboratory has renewed interest in defining the requirements for a future high-gain inertial confinement fusion (ICF) facility. Our best chance of predicting future ICF performance is with 3-D radiation hydrodynamic simulations that have been benchmarked against experimental data, but their high computational cost is prohibitive for use in practical design studies. We introduce a hierarchical approach where 3-D simulations are tuned to match experimental measurements and used to train 3-D degradation models in 1-D simulations allowing for accurate predictions over the entire OMEGA direct-drive database. A genetic algorithm was used in combination with the trained 1-D simulations to search for optimal direct-drive implosion designs at driver energies ranging from 20 kJ to 10 MJ. As the fidelity of 3-D codes improves, this approach will provide a viable experimentally informed tool for defining the next ICF facility.

Partial-differential-equation (PDE)-constrained optimization is a well-worn technique for acquiring optimal parameters of systems governed by PDEs. However, this approach is limited to providing a single set of optimal parameters per optimization. Given a differentiable PDE solver, if the free parameters are reparameterized as the output of a neural network, that neural network can be trained to learn a map from a probability distribution to the distribution of optimal parameters. This proves useful in the case where there are many well performing local minima for the PDE. We apply this technique to train a neural network that generates optimal parameters that minimize laser-plasma instabilities relevant to laser fusion and show that the neural network generates many well performing and diverse minima.

Parameter estimation using observables is a fundamental concept in the experimental sciences. Mathematical models that represent the physical processes can enable reconstructions of the experimental observables and greatly assist in parameter estimation by turning it into an optimization problem which can be solved by gradient-free or gradient-based methods. In this work, the recent rise in flexible frameworks for developing differentiable scientific computing programs is leveraged in order to dramatically accelerate data analysis of a common experimental diagnostic relevant to laser–plasma and inertial fusion experiments, Thomson scattering. A differentiable Thomson-scattering data analysis tool is developed that uses reverse-mode automatic differentiation (AD) to calculate gradients. By switching from finite differencing to reverse-mode AD, three distinct outcomes are achieved. First, gradient descent is accelerated dramatically to the extent that it enables near real-time usage in laser–plasma experiments. Second, qualitatively novel quantities which require O ( 10 3 ) parameters can now be included in the analysis of data which enables unprecedented measurements of small-scale laser–plasma phenomena. Third, uncertainty estimation approaches that leverage the value of the Hessian become accurate and efficient because reverse-mode AD can be used for calculating the Hessian.

Solving fluid dynamics equations often requires the use of closure relations that account for missing microphysics. For example, when solving equations related to fluid dynamics for systems with a large Reynolds number, sub-grid effects become important and a turbulence closure is required, and in systems with a large Knudsen number, kinetic effects become important and a kinetic closure is required. By adding an equation governing the growth and transport of the quantity requiring the closure relation, it becomes possible to capture microphysics through the introduction of ‘hidden variables’ that are non-local in space and time. The behavior of the ‘hidden variables’ in response to the fluid conditions can be learned from a higher fidelity or ab-initio model that contains all the microphysics. In our study, a partial differential equation simulator that is end-to-end differentiable is used to train judiciously placed neural networks against ground-truth simulations. We show that this method enables an Euler equation based approach to reproduce non-linear, large Knudsen number plasma physics that can otherwise only be modeled using Boltzmann-like equation simulators such as Vlasov or particle-in-cell modeling.