The Applied Partial Differential Equations Integrated Software Infrastructure Center APDEC

1
The Applied Partial Differential Equations
Integrated Software Infrastructure Center (APDEC)

• Goal Develop algorithms and software for
  simulating multiscale problems on structured
  grids, based on block-structured adaptive mesh
  refinement (AMR) and embedded boundary (EB)
  representations of irregular boundaries.
• Applications-driven approach end-to-end
  development of software tools to meet the needs
  of specific OSC scientific problems.

2
ParticipantsAPDEC Principal Investigators P.
Colella, J. Bell (LBNL), L. Diachin (LLNL), R.
Samtaney (PPPL), M. Berger (NYU), R. Leveque
(Univ. of Washington), M. Minion
(Univ. of North Carolina), E. G. Puckett (Univ.
of California-Davis), C. Rutland (Univ. of
Wisconsin).Science Collaborators Magnetic
Fusion S. Jardin (PPPL). Combustion R.Cheng
(LBNL) C. Rutland is funded by both
APDEC and BES SciDAC. Accelerator
Modeling R. Ryne, W. Leemans, E. Esarey
(LBNL).SciDAC ISIC Collaborators Multigrid
solvers R. Falgout (LLNL / TOPS). CCA J. Ray
(SNL-CA / CCTTSS).
3
Applications Development, FY 2001 - 2004
Combustion developed AMR algorithm for turbulent
combustion with detailed chemistry and transport,
and performed high-resolution simulations of
laboratory-scale hydrocarbon flames. Magnetic
Fusion developed AMR capability for MHD, and
applied it to obtain new scientific results in
the areas of pellet injection for tokomaks,
magnetic reconnection, and MHD shock
refraction. Accelerator Modeling developed
preliminary version of AMR-PIC for
Vlasov-Poisson, and coupled it to MaryLie/Impact
(ML/I) code for beam dynamics.
4
General-purpose AMR solvers for elliptic,
parabolic and hyperbolic PDEs, with easy-to-use
APIs for physics-dependent information.EB AMR
finite-volume solvers for elliptic, parabolic and
hyperbolic problems in complex geometries and a
variety of grid-generation tools for
embedded-boundary grids.Support infrastructure
for particle methods on AMR grids. Node-centered
Poisson solvers for AMR-PIC, including support
for complex geometries based on the
Shortley-Weller method.Interoperability tools,
including a framework-neutral data interface for
AMR data, interfaces to hypre and other solver
packages, and a prototype CCA componentization of
AMR elliptic solvers. AMR visualization and data
analysis package.
Software Development, FY 2001 - 2004
5
Application Goals, FY 2004 - 2006
Magnetic fusion The overall goal is to increase
the fidelity of the AMR MHD codes in for problems
in magnetic fusion. The capabilities we propose
developing are driven by important issues for the
design of ITER, i.e. fueling a tokomak and
stability of burning plasmas. Also, developing
AMR algorithms and software to deal with the
strong anisotropies present in these problems
represents a substantial extension of our current
capabilities. Pellet injection anisotropic
diffusion, high-fidelity and efficient treatment
of plasma boundary. Magnetic reconnection
2-fluid model, including Hall term implicit
treatment of magnetosonic, Alfven waves.
6
Applications Goals (contd)
Combustion develop a simulation capability for
an ultra-low NOx burner, combining the existing
AMR combustion capability with an EB AMR
algorithm for viscous incompressible flow in the
nozzle. It is difficult to obtain and interpret
multidimensional data in flows with swirl, while
such flows are the most common way to stabilize
flames. Combining combustion and complex geometry
in AMR represents a substantial increment in
capability over what is currently available.
Accelerator modeling AMR-PIC for Vlasov-Poisson
enhances a core capability for accelerator
modeling, and will be completed early in the
renewal period. Fluid simulation tools for
laser-driven plasma-wakefield accelerators meets
critical need in an area in which there are few
tools available. Beam dynamics complete the
development of a robust AMR-PIC capability for
AST electrostatic PIC codes, including support
for complex geometries. Laser-driven
plasma-wakefield accelerators develop fluid
dynamics simulation capabilities for gas jet
injection, laser-driven shock dynamics.
7
Software Goals, FY 2004 -
2006Performance tuning of embedded boundary
software new aggregate irregular stencil
operations for serial efficiency, and new
load-balancing approaches for improving parallel
scalability (needed for combustion, gas jet
problem).Extension of the APDEC infrastructure
to support AMR on mapped grids,parabolic solvers
for problems with strong anisotropies (needed for
MHD modeling).Increased interoperability
within the APDEC software suite (e.g. mixing
spatial dimensions) and between APDEC software
and other packages (e.g. framework-neutral data
alias for particles) (needed for all
applications).
8
Software Goals
(cont'd)Implement new analysis-based AMR
algorithms for constant-coefficient elliptic
solvers. Combination of FFTs, multipole methods
and Andersons method of local corrections leads
to more efficient algorithms with increased
robustness, vastly reduced parallel communication
requirements, and better coupling to PIC (needed
for all applications).
3D MLC calculation of Poisson's equation, with
2563 mesh broken into 643 blocks. Image on
right shows solution error through slice at
mid-plane (joint work with S. Baden, G. Balls,
UCSD.)
9
Advanced Algorithm Development

• Goal to investigate issues in algorithm design
  with a potential impact on
• SciDAC applications in the 3-5 year time horizon.
• Higher order methods spectral deferred
  corrections (UNC).
• Sharp interface algorithms for compressible jets
  (UCD).
• Higher-order spatial discretizations for embedded
  boundaries (NYU, LLNL, Univ. Washington).
• Improved methods for generating surface data from
  CAD data (LLNL).
• Novel discretization methods for mapped grids.
  (Univ. Washington).
• Improved model fidelity for spray combustion
  (Univ. Wisconsin).

10
Staffing Profile for APDEC
Current staffing profile Applications
Development 2.65 FTE (2 FTE at LBNL, .65 FTE at
PPPL). An additional .35 FTE has been provided to
PPPL by the SAPP program, for which a separate
proposal will be submitted for the
renewal. Software Development 5.66 FTE (4.66 FTE
at LBNL, 1 FTE at LLNL). Advanced Algorithm
Development 1 FTE at LLNL. University
researchers contribute to both software
development and advanced algorithm development
efforts (primarily the latter). For the renewal
period, there will be a shift of 3 FTE of effort
from software development to applications
development. These will mainly consist of the
current members of the software development team
their intimate knowledge of the algorithms and
software will be essential to the successful
development of the new applications capabilities.
The remaining software development efforts will
focus on delivering specific capabilities that
are required by the various applications efforts,
as well as supporting the existing software base.
11
Broader Impacts
Other applications projects that are using APDEC
software.
Cosmology galaxy formation (F. Miniati,
MPA-Garching) semi-local strings (J. Borrill,
LBNL)
NASA CT Program multiphase flow in microgravity
environments star formation.
Flame propagation in type 1A supernovae (S.
Woosley, UCSC).
Time-dependent Ginzburg-Landau models for
phase-field dynamics (F. Alexander, LANL).
AMR-PIC for heavy-ion fusion simulations (A.
Friedman, D. Grote, J.-L. Vay, LBNL / LLNL).
12
Broader Impacts (cont.)
Flow in San Francisco bay (M. Barad, UC Davis E.
Ateljevich, California DWR).
Mesoscale atmospheric modeling (C. Bono, LBNL).
Micro-fluidic simulation for BioMEMS devices (D.
Trebotich, LLNL).
Large-deformation solid-fluid interactions. (G.
Miller, UC Davis).
Cell modeling (A. Arkin, P. Schwartz, LBNL D.
Adalsteinsson, Univ. of North Carolina).