You may be trying to access this site from a secured browser on the server. Please enable scripts and reload this page.
Turn on more accessible mode
Turn off more accessible mode
Skip Ribbon Commands
Skip to main content
Turn off Animations
Turn on Animations
To navigate through the Ribbon, use standard browser navigation keys. To skip between groups, use Ctrl+LEFT or Ctrl+RIGHT. To jump to the first Ribbon tab use Ctrl+[. To jump to the last selected command use Ctrl+]. To activate a command, use Enter.
Tab 1 of 3.
Tab 2 of 3.
Tab 3 of 3.
For collaborators only
High Performance Computing for Accelerator Design and Optimization
The ComPASS4 research program will combine beam dynamics, loss propagation, and plasma-based acceleration simulation components. Both FACET-II and PIP-II will demand simulations that are simultaneously more complex and more accurate than any simulations to date. To address both of these demands, we are adapting and developing many intertwined pieces of software to efficiently exploit the next generation of DOE HPC architectures. To that end, we have formed a partnership between physics teams at Fermilab and UCLA and applied mathematics and computer science teams at ANL, LBNL, and UCLA to prepare our particle-in-cell (PIC) software to have higher fidelity and to take advantage of new computing hardware. The developed software will incorporate new, scalable implementations of linear algebra algorithms, static and adaptive mesh refinement techniques, and an in situ optimization framework for accelerator simulations. Throughout the research program, the software we develop will be used by project members and the community to simulate and optimize current and future particle accelerators.
We are combining Synergia beam dynamics software with MARS energy deposition software in order to simulate portions of the Fermilab accelerator complex under high-intensity conditions for PIP-II. Our simulations will include collective beam dynamics effects (space charge, wakefields, and electron clouds) combined with the component and tunnel activation arising from subsequent beam losses. FNAL will work with LBNL to develop linear-algebra-based solvers that address the nontrivial boundary conditions needed for accurate loss calculations. These solvers will be optimized for Intel Phi and GPU architectures that will appear in next-generation DOE HPC systems. FNAL will work with ANL to use beam losses and activation results produced by MARS as inputs to a numerical optimization framework that will minimize and/or mitigate the effects of beam losses. These calculations will take advantage of memory hierarchies in Cori, Aurora, and beyond.
We are also expanding plasma-based acceleration simulations using QuickPIC to support future FACET-II experiments. Our work will ensure our software can simulate parameters relevant to linear colliders but that are currently beyond the reach of those planned for FACET-II. UCLA will develop high-fidelity field solvers, PIC data structures, and algorithms that efficiently utilize the Intel Phi and GPU platforms. ComPASS4 team members from UCLA and LBNL will enable the coupling between the UPIC PIC Framework and the AMReX mesh refinement capabilities into the quasi-static solvers; UCLA and ANL team members will identify the best beam and plasma input parameters that optimize the characteristics of the accelerated beam in single and multiple stages. We will use the developed software to support FACET-II and address key physics issues related to future plasma-based linear colliders.
Advances will also be made in cross-cutting technologies based on ASCR-developed software. For adaptive mesh refinement we will build on the AMReX tool base. We will build on PETSc/TAO/APOSMM/POUNDerS to develop numerical optimization solvers for accelerator-focused problems. For numerical linear algebra, we will build on existing direct solvers for linear systems, including symPACK, SuperLU, and STRUMPACK, and we will also consider preconditioning techniques for iterative solvers. These solvers will be utilized extensively within the PIC simulations.
The ComPASS4 collaboration will integrate our physics, applied mathematics, and computer science efforts into a combined plasma-based acceleration/beam dynamics software tool that can model the transport of witness beams between multiple stages of a plasma-based linear collider. These simulations will take advantage of our advances in PIC calculations as well as our numerical optimization framework for tuning beam parameters, setting the stage for simulations in support of a linear collider design.
Milestones & Announcements
Algorithms for Optimizing Particle Accelerator Beam Emittance
Multi-parameter optimization method for minimizing the emittance from laser-plasma accelerators
Optimizing Dipole and Quadrupole Strengths,
Better Accelerators Designed with Numerical Optimizers and Simulations
Multi-parameter design optimization to minimize beta functions in a rapid cycling synchrotron for a potential 2.4 MW beam power upgrade.
Unifying Derivative-Free and
Zeroth-Order Optimization Methods
Survey of derivative-free optimization formulations and algorithms
Mixed-Integer Simulation-Based Optimization
New derivative-free optimization algorithm for simulation-based optimization for unrelaxable integer constraints
Manifold Sampling for Structure-Exploiting Optimization
New manifold sampling algorithm for composite simulation-based optimization; useful for exploiting structure in emittance minimization
Improving performance of sparse direct methods via fill-preserving permutations
Fast implementations of a preprocessing step in sparse matrix factorization to improve dense block structure that will enhance the performance of sparse direct solvers
Evaluation of Sparse Direct and Iterative Finite Difference Field Solvers in the QuickPIC code
Evaluation of sparse direct and iterative solvers in finite difference field solvers in the QuickPIC code
Accelerator Simulations at Scale Using GPUs
Integrate Kokkos Core data structures and parallelization primitives into Synergia to enable code execution on both traditional CPUs and modern hybrid GPU architectures using the same code base.
Beam Simulation Combined with Lost Particle Tracking
Particles are simulated in an accelerator with Synergia. Lost particles are transported through the surrounding with the MARS energy deposition code. Energy deposited in material surrounding the accelerator is enumerated and available for use in an optimizer loop.
Synergia Simulations of Injecting Painting Support PIP-II
The PIP-II project will build an 800 MeV superconducting linac to inject into the Booster replacing the current 400 MeV linac. Injecting at higher energies reduces space charge forces allowing the complex to reach 1.2 MW of delivered beam power. To further reduce space charge, the particle distribution is painted in phase space to spread its spatial extent.
QPAD: A novel quasi-static particle-in-cell code based on an azimuthal Fourier decomposition
Development of a novel quasi-static particle-in-cell code based on an azimuthal Fourier decomposition achieving dramatic speedup over current fully explicit and quasi-static 3D codes.
Generating high quality ultra-brightness electron beams using downramps and evolving drivers