## Parallel Fully Automatic hp-Adaptive 3D Finite Element Package, Engineering with Computers 22 (2005)

Venue: | University of Texas at Austin |

Citations: | 6 - 6 self |

### BibTeX

@INPROCEEDINGS{Paszyński05parallelfully,

author = {M. Paszyński and L. Demkowicz},

title = {Parallel Fully Automatic hp-Adaptive 3D Finite Element Package, Engineering with Computers 22},

booktitle = {University of Texas at Austin},

year = {2005},

pages = {3--4}

}

### OpenURL

### Abstract

The paper presents a description of par3Dhp- a 3D, parallel, fully automatic hp-adaptive finite element code for elliptic and Maxwell problems. The parallel implementation is an extension of the sequential code 3Dhp90, which generates, in a fully automatic mode, optimal hp meshes for various boundary value problems. The system constitutes an infrastructure for a class of parallel hp adaptive computations. Its modular structure allows for an independent parallelization of each component of the system. The presented work addresses parallelization of these components, including distributed data structures, load balancing and domain redistribution, parallel (multi-frontal) solver, optimal hp mesh refinements, and a main control module. All components communicate through a distributed data structure, and the control module synchronizes work of all components. The concept of ghost elements has been used to simplify the communication algorithms for parallel mesh refinements. The system has been implemented in Fortran 90 and MPI, and the load balancing is done through an interface with the ZOLTAN library. Numerical results are presented for the model Fichera problem.

### Citations

152 |
The Unified Modeling Language User Guide”. The Addison-Wesley Object Technology Series
- Booch, Jacobson, et al.
- 1998
(Show Context)
Citation Context ...rithm produces a sequence of coarse and corresponding fine meshes that deliver exponential convergence. Fig. 1 presents one step of the hp algorithm using the Unified Modeling Language (UML) notation =-=[3]-=-. The initial mesh is the coarse mesh for the first iteration. The coarse mesh, see Fig.2 is globally refined in both element size h (each hexahedron element is divided into eight element sons in 3D) ... |

15 |
Anisotropic adaptive simulation of transient flows using discontinuous Galerkin methods
- Remacle, Li, et al.
(Show Context)
Citation Context ...or an arbitrary domain partitioning of a current mesh but it does not support anisotropic mesh refinements. The only parallel hp codes that we are aware of, have been developed by Joe Flaherty at RPI =-=[22]-=- in the context of Discontinuous Galerkin (DG) methods, and Abani Patra at SUNY at Buffalo [16, 2, 15]. In our first attempt to develop a parallel hp code, we managed to develop only a parallel two-gr... |

11 |
Parallel, Fully Automatic hp-Adaptive 2D Finite Element Package, The University of Texas at Austin
- Paszynski, Kurtz, et al.
- 2004
(Show Context)
Citation Context ...re redistributed by “cutting off” segments of the space filling curve. Challenges of the 3D implementation. The paper is a continuation and extension of the twodimensional implementation presented in =-=[19]-=-, based on the 2D implementation of the hp strategy described in [7]. Following the experience gained from the three-dimensional implementation of the strategy [21] (elliptic problems), and a two-dime... |

10 |
2d hp-adaptive finite element package (2dhp90) version 2.0
- Demkowicz
- 2002
(Show Context)
Citation Context ...hp meshes and 2D Maxwell equations [1], but we failed miserably with the parallelization of the actual code. This failure motivated us to develop new data structures better suited for parallelization =-=[5, 6]-=-. The presented concept of parallelization is based on the assumption that in the DD based parallel code, each of the processors is executing the sequential code with only minimal upgrades added (to s... |

10 |
A fully automatic hpadaptivity
- Devloo
(Show Context)
Citation Context ... size (number of degrees-of-freedom (d.o.f.)) meeting a prescribed error tolerance criterion. After over a decade of research, a fully automatic, problem independent, hp-strategy has been constructed =-=[7, 8, 9]-=- that delivers a sequence of optimally refined hp meshes and exponential convergence - the error decreases exponentially fast, both in terms of problem size and actual CPU time [7, 8, 9]. The presente... |

9 |
3D hp-Adaptive Finite Element Package (3Dhp90) version 2.0: The Ultimate Data Structure for Three Dimensional, Anisotropic hp Refinitement
- Demkowicz, Pardo, et al.
- 2002
(Show Context)
Citation Context ...hp meshes and 2D Maxwell equations [1], but we failed miserably with the parallelization of the actual code. This failure motivated us to develop new data structures better suited for parallelization =-=[5, 6]-=-. The presented concept of parallelization is based on the assumption that in the DD based parallel code, each of the processors is executing the sequential code with only minimal upgrades added (to s... |

9 |
Fully automatic hp-adaptivity in three dimensions
- Rachowicz, Pardo, et al.
(Show Context)
Citation Context ...nsional implementation presented in [19], based on the 2D implementation of the hp strategy described in [7]. Following the experience gained from the three-dimensional implementation of the strategy =-=[21]-=- (elliptic problems), and a two-dimensional implementation for Maxwell problems [9], a new version of the algorithm along with a 3D implementation of it, has recently been worked out [14]. The present... |

7 |
Designing and Building Parallel Programs", http://www-unix.mcs.anl.gov/dbpp
- Foster
(Show Context)
Citation Context ...ith relative efficiency E = T1 , shown in Fig. 28, and relative speedup pTp S = T1 Tp , in Fig. 29. Here T1 is the total sequential code time, Tp is the total parallel code time for p processors, see =-=[13]-=-. The relative efficiency for 3 processors (plus one master processor for the wire-frame problem) is about 85 %. In other words, execution time for 4 processors is 3.5 times faster than execution time... |

6 |
Patra: Simple data management, scheduling and solution strategies for managing the irregularities in parallel adaptive hp finite element simulations, Parallel Computing 26
- Laszloffy, Long, et al.
- 2000
(Show Context)
Citation Context ...refinements. The only parallel hp codes that we are aware of, have been developed by Joe Flaherty at RPI [22] in the context of Discontinuous Galerkin (DG) methods, and Abani Patra at SUNY at Buffalo =-=[16, 2, 15]-=-. In our first attempt to develop a parallel hp code, we managed to develop only a parallel two-grid solver for hp meshes and 2D Maxwell equations [1], but we failed miserably with the parallelization... |

5 |
Fully automatic hp-adaptivity for Maxwell’s equations
- Demkowicz
- 2005
(Show Context)
Citation Context ... size (number of degrees-of-freedom (d.o.f.)) meeting a prescribed error tolerance criterion. After over a decade of research, a fully automatic, problem independent, hp-strategy has been constructed =-=[7, 8, 9]-=- that delivers a sequence of optimally refined hp meshes and exponential convergence - the error decreases exponentially fast, both in terms of problem size and actual CPU time [7, 8, 9]. The presente... |

4 |
hp-adaptive finite elements for time-harmonic Maxwell equations
- Demkowicz
- 2003
(Show Context)
Citation Context ... size (number of degrees-of-freedom (d.o.f.)) meeting a prescribed error tolerance criterion. After over a decade of research, a fully automatic, problem independent, hp-strategy has been constructed =-=[7, 8, 9]-=- that delivers a sequence of optimally refined hp meshes and exponential convergence - the error decreases exponentially fast, both in terms of problem size and actual CPU time [7, 8, 9]. The presente... |

4 |
Fully Automatic hp-Adaptivity for Acoustic and Electromagnetic Scattering in Three Dimensions
- Kurtz
- 2006
(Show Context)
Citation Context ...the strategy [21] (elliptic problems), and a two-dimensional implementation for Maxwell problems [9], a new version of the algorithm along with a 3D implementation of it, has recently been worked out =-=[14]-=-. The presented parallel code has been developed based on this new implementation. Referring to [14] for technical details, we emphasize here only a few important algorithmical points: • mesh optimiza... |

4 |
A Parallel Multifrontal Solver for hp-Adaptive Finite Elements
- Walsh
- 1999
(Show Context)
Citation Context ...arse mesh and fine mesh problems are solved using a parallel frontal solver. We have implemented an interface with a parallel version of the frontal solver, based on the domain decomposition approach =-=[24]-=-. The logic of the solver is dimension independent. The parallel frontal solver and its interface has been described in detail in the 2D version of the parallel hp adaptive code [19]. Optimal hp refin... |

4 |
Geometrical Modeling Package version 2.0
- Xue, Demkowicz
(Show Context)
Citation Context ...t identifier that remains unchanged through all mesh repartitioning steps. This is accomplished in our work by storing a complete copy of the geometry (supported with our Geometrical Modeling Package =-=[25]-=-) on each processor. Individual nodes and initial mesh elements can then be identified uniquely with their reference coordinates. An alternative technique, better suited for parallel implementation, w... |

3 |
Patra: Robust and efficient domain decomposition preconditioners for adaptive hp finite element approximation for linear elasticity with and without discontinuous coefficients
- Bauer, K
- 2004
(Show Context)
Citation Context ...refinements. The only parallel hp codes that we are aware of, have been developed by Joe Flaherty at RPI [22] in the context of Discontinuous Galerkin (DG) methods, and Abani Patra at SUNY at Buffalo =-=[16, 2, 15]-=-. In our first attempt to develop a parallel hp code, we managed to develop only a parallel two-grid solver for hp meshes and 2D Maxwell equations [1], but we failed miserably with the parallelization... |

3 |
Framework Version 3
- Edwards
(Show Context)
Citation Context ...ored in each processor’s memory. 2sAmong major undertakings to develop a general infrastructure to support DD based parallelization of PDE solvers, one has to list first of all the Sierra Environment =-=[23, 10, 11, 12]-=- developed by Sandia National Labs. Designed to support h-adaptivity, the Sierra framework has been used to parallelize several FE codes developed at Sandia [12]. The Sierra environment allows for an ... |

3 |
Integration of hp-Adaptivity and Multigrid. I. A Two Grid Solver for hp Finite Elements
- Pardo, Demkowicz
- 2002
(Show Context)
Citation Context ...with respect to those presented in 2D version of the code, by using the concept of ghost elements. Moreover, ghost elements will also be essential for a parallel implementation of the two grid solver =-=[18]-=-, where the access to initial mesh elements neighboring interfaces between domains is crucial. The new implementation of the automatic optimal hp refinements package [14] allows for a simple paralleli... |

1 |
A Two-Grid Parallel Solver for Time Harmonic Maxwell’s Equations and hp
- Walsh, Demkowicz
- 2001
(Show Context)
Citation Context ...thods, and Abani Patra at SUNY at Buffalo [16, 2, 15]. In our first attempt to develop a parallel hp code, we managed to develop only a parallel two-grid solver for hp meshes and 2D Maxwell equations =-=[1]-=-, but we failed miserably with the parallelization of the actual code. This failure motivated us to develop new data structures better suited for parallelization [5, 6]. The presented concept of paral... |

1 |
Large-Scale Finite Element Scattering Analysis on Massively Parallel Computers”, Finite Element Software for Microwave Engineering
- D’Angelo, Mayergoyz
- 1996
(Show Context)
Citation Context ...d work is motivated with solutions of large electromagnetic scattering problems involving geometrical singularities (diffraction on edges and points) and scattering from resonating cavities, see e.g. =-=[4]-=-. Resolution of geometric singularities in 2D requires many levels of h-refinements (the ratio of the smallest to largest elements may be 10 −6 ). Additionally, resolution of boundary layers (skin eff... |

1 |
A Software Environment for Developing
- Edwards, Stewart, et al.
- 2001
(Show Context)
Citation Context ...ored in each processor’s memory. 2sAmong major undertakings to develop a general infrastructure to support DD based parallelization of PDE solvers, one has to list first of all the Sierra Environment =-=[23, 10, 11, 12]-=- developed by Sandia National Labs. Designed to support h-adaptivity, the Sierra framework has been used to parallelize several FE codes developed at Sandia [12]. The Sierra environment allows for an ... |

1 |
Parallel HP Adaptive Finite Element Analysis for Viscous Incompressible Fluid Problems” Dissertation
- Patra
- 1995
(Show Context)
Citation Context ...refinements. The only parallel hp codes that we are aware of, have been developed by Joe Flaherty at RPI [22] in the context of Discontinuous Galerkin (DG) methods, and Abani Patra at SUNY at Buffalo =-=[16, 2, 15]-=-. In our first attempt to develop a parallel hp code, we managed to develop only a parallel two-grid solver for hp meshes and 2D Maxwell equations [1], but we failed miserably with the parallelization... |

1 | h-Relation Personalized Communication Strategy For HP-Adaptive Computations - Paszynski, Milfeld - 2004 |