Send mail to: mgnet@cs.yale.edu for the digests or bakeoff mgnet-requests@cs.yale.edu for comments or help Anonymous ftp repository: www.mgnet.org (128.163.209.19) Current editor: Craig Douglas douglas-craig@cs.yale.edu World Wide Web: http://www.mgnet.org or http://casper.cs.yale.edu/mgnet/www/mgnet.html or http://www.cerfacs.fr/~douglas/mgnet.html or http://phase.etl.go.jp/mgnet or http://www.nchc.gov.tw/RESEARCH/Math/mgnet/www/mgnet.html Today's editor: Craig Douglas (douglas-craig@cs.yale.edu) Volume 10, Number 6 (approximately June 30, 2000) Today's topics: Mirror Reminder Multigrid and Voronoi Cell Finite Element Method Paper Submission (T.L. Beck) AMG-WS at St. Wolfgang/Strobl 2001-02 NA Year at the Fields Institute in Toronto LBNL-NERSC Workshop on ACTS Conference on Inverse Problems GAMM Seminar on Construction of Grid Generation Algorithms GAMM Workshop on Computational Electromagnetics ------------------------------------------------------- Date: Fri, 30 Jun 2000 25:22:11 -0400 From: Craig C. DouglasSubject: Mirror Reminder I have been asked several times by people in Europe if there is some way to speed up transmission speeds from Kentucky (where http://www.mgnet.org sits). At the top of this newsletter is a list of alternate sites that mirror the Kentucky site. If you are having trouble with transmission speeds, please try a closer site. If you are interested in mirroring MGNet, please contact me directly. ------------------------------------------------------- Date: Fri, 07 Jul 2000 13:30:36 -0400 From: "Paul M. Eder" Subject: Multigrid and Voronoi Cell Finite Element Method My name is Paul Eder. I am a graduate student at The Ohio State University. One of the problems that I am working on involves developing code in Fortran 90 to solve problems involving the deformation and damage of composite materials using the Voronoi Cell Finite Element Method. Currently, there is code that solves the entire problem, but it takes around 2 hours for a simple linear elasticity problem. There are three levels of elements: level 0 = basic quadrilateral elements level 1 = level 0 elements that were adaptively refined around stress concentrations (these are level 0 elements that are divided into smaller elements and represent a general repetition of the microstructure) level 2 = Voronoi cell elements, adaptively created from level 1 elements around stress concentrations, that represent the actual microstructure There is the possibility that all three elements exist simultaneously on different locations of the geometry. My goal is to parallelize the code to run on the SV1 at the Ohio Supercomputer Center and to implement multigrid to solve the problem. Ultimately, it is desired to solve level 2 elements only between level 0 and level 2 wherever level 2 elements exist. I would like to know if you have any suggestions on what multigrid algorithms or multigrid packages would be useful in a situation like this. I have searched through a lot of web sites and am having difficulty with this. Any help that you could provide would be greatly appreciated. Paul M. Eder The Ohio State University Department of Mechanical Engineering (614)846-7779 Editor's Note: If you think that your multigrid code will help Mr. Eder, ------------- please contact him. ------------------------------------------------------- Date: Sat, 24 Jun 2000 15:25:24 -0400 From: "Thomas L. Beck" Subject: Paper Submission (T.L. Beck) I am submitting a postscript file of a paper called "Real-Space Mesh Techniques in Density Functional Theory" to MGNet. This paper will appear in the Jan. 1, 2001 issue of Reviews of Modern Physics. It discusses grid methods for electrostatics and electronic structure calculations (with extensive discussion of multigrid). Thomas L. Beck | http://bessie.che.uc.edu/tlb/ Department of Chemistry | becktl@uc.edu University of Cincinnati | phone: 513-556-4886 Cincinnati, OH 45221-0172 | fax: 513-556-9239 * * * * * Real-space mesh techniques in density functional theory Thomas L. Beck Department of Chemistry, University of Cincinnati, Cincinnati, OH 45221-0172 Abstract This review discusses progress in efflcient solvers which have as their foundation a representation in real space, either through finite-difference or finite-element formulations. The relationship of real-space approaches to linear-scaling electrostatics and electronic structure methods is first discussed. Then the basic aspects of real-space representations are presented. Multigrid techniques for solving the discretized problems are covered; these numerical schemes allow for highly efflcient solution of the grid-based equations. Applications to problems in electrostatics are discussed, in particular numerical solutions of Poisson and Poisson-Boltzmann equations. Next, methods for solving self-consistent eigenvalue problems in real space are presented; these techniques have been extensively applied to solutions of the Hartree-Fock and Kohn-Sham equations of electronic structure, and to eigenvalue problems arising in semiconductor and polymer physics. Finally, real-space methods have found recent application in computations of optical response and excited states in time-dependent density functional theory, and these computational developments are summarized. Multiscale solvers are competitive with the most efflcient available plane-wave techniques in terms of the number of self-consistency steps required to reach the ground state, and they require less work in each self-consistency update on a uniform grid. Besides excellent efficiencies,the decided advantages of the real-space multiscale approach are 1) the near-locality of each function update, 2) the ability to handle global eigenfunction constraints and potential updates on coarse levels, and 3) the ability to incorporate adaptive local mesh refinements without loss of optimal multigrid efflciencies. CONTENTS I. INTRODUCTION 2 II. DENSITY FUNCTIONAL THEORY 5 A. Kohn-Sham equations 5 B. Classical DFT 6 III. LINEAR-SCALING CALCULATIONS 7 A. Classical electrostatics 8 B. Electronic structure 9 IV. REAL-SPACE REPRESENT A TIONS 10 A. Finite differences 11 1. Basic finite-difference representation 11 2. Solution by iterative techniques 12 3. Generation of high-order finite-difference formulas 13 B. Finite elements 14 1. Variational formulation 14 2. Finite-element bases 15 V. MULTIGRID TECHNIQUES 16 A. Essential features of multigrid 16 B. Full approximation scheme multigrid V-cycle 17 C. Full multigrid 18 VI. ELECTROST A TICS CALCULA TIONS 18 A. Poisson solvers 19 1. Illustration of multigrid efficiency 19 2. Mesh-refinement techniques 20 B. Poisson-Boltzmann solvers 21 C. Computations of free energies 23 D. Biophysical applications 25 VII. SOLUTION OF SELF-CONSISTENT EIGENVALUE PROBLEMS 26 A. Fixed-potential eigenvalue problems in real-space 26 1. Algorithms 26 2. Applications 28 B. Finite-difference methods for self-consistent problems 29 C. Multigrid methods 31 D. Finite-difference mesh-refinement techniques 34 E. Finite-element solutions 35 F. Orbital-minimization methods 37 VIII. TIME-DEPENDENT DFT CALCULATIONS IN REAL SPACE 37 A. TDDFT in real time and optical response 38 B. TDDFT calculation of excited states 39 IX. SUMMARY 40 ACKNOWLEDGMENTS 41 Appendix A 41 References 42 Editor's Note: See http://www.mgnet.org/mgnet-papers.html or ------------- http://www.mgnet.org/mgnet/papers/Beck/rmptb.ps.gz ------------------------------------------------------- Date: Wed, 28 Jun 2000 22:25:11 -0400 From: Craig C. Douglas Subject: AMG-WS at St. Wolfgang/Strobl A great two day workshop was held in Austria. Below are the talks that were given. A virtual proceedings is being constructed by Gundolf Haase and me for MGNet. There will be many more details in the July newsletter. Stay tuned. * * * * * Klaus Stueben Some studies of the AMG performance in critical situations Johannes Kraus An Optimal Order Algebraic Multilevel Method Jim E. Jones Algebraic Multigrid for Finite Element Problems (AMGe) Andreas Bollhoefer AMG Preconditioners for Sparse Approximate Inverse Matrices Craig C. Douglas Using a Multilayer Ocean Model on Clusters versus a Traditional Supercomputer Gundolf Haase A principle for constructing parallel AMG and its realization: part I Michael Kuhn A principle for constructing parallel AMG and its realization: part II Claus Bayreuther Construction of homogenization based transfer operators Christian Wagner On the Algebraic Construction of Multilevel Transfer Operators Van Emden Henson AMGe and Element-free AMGe: General algorithms for computing interpolation weights Ferdinand Kickinger Algebraic Multigrid for Stokes Equation Rudolf Beck Algebraic Multigrid for Edge Elements by Component Splitting Stefan Reitzinger Algebraic Multigrid for 3D Magnetic Field Problems Joseph E. Pasciak Iterative techniques for mixed discretizations of Maxwell's equations Participants without a presentation: Ronald Hoppe, Barbara Kaltenbacher, Karl Kunisch, Ulrich Langer, Volker Schulz Editor's Note: See http://www.mgnet.org/mgnet-amg2000-strobl.html (it ------------- may be a few days before this is really available). ------------------------------------------------------- Date: Tue, 6 Jun 2000 14:15:13 -0400 From: Ken Jackson Subject: 2001-02 NA Year at the Fields Institute in Toronto The Fields Institute in Toronto is sponsoring a Thematic Year on "Numerical and Computational Challenges in Science and Engineering" (NCCSE) from August 2001 to July 2002. The main point of this announcement is to inform the scientific computing committee about this event so that any people interested in participating can include it in their plans for 2001-02. A key to the success of this program will be the senior long-term visitors that it attracts. Their research interests will shape many of the events that take place during the year and their participation in the program will attract many junior colleagues, postdocs and graduate students. We are now considering applications from senior researchers to visit the Fields Institute for a month or more, possibly spread out over several shorter visits, such as a couple of weeks around two or more workshops or one day each week for a term. We would particularly welcome visitors for one or both terms of the year. We have some funds to support travel and local expenses for senior researchers, but not to pay their salaries. If you are a senior researcher interested in participating in the program, please e-mail Ken Jackson at krj@cs.utoronto.ca. More information about the Fields Institute in general and the NCCSE Thematic Year in particular can be found at http://www.fields.utoronto.ca and http://www.fields.utoronto.ca/numerical.html respectively. ------------------------------------------------------- Date: Thu, 29 Jun 2000 15:19:56 -0700 From: "Tony Drummond-Lewis" Subject: LBNL-NERSC Workshop on ACTS The US Department of Energy's National Energy Research Scientific Computing Center (NERSC) is now taking applications from graduate students and postdoctoral fellows to participate in a 3-day workshop on the ACTS toolkit, September 28-30 of the current year. The theme of this workshop is: How can ACTS work for you? ACTS stands for Advanced Computational Testing and Simulation, and it is a Department of Energy (DOE) funded project that comprises a set tools developed mainly at DOE laboratories for the support of large scientific and industrial software. These tools deliver solutions for numerical problems, scientific data representations, data manipulation, visualization, program execution and distributed computing (for more information visit the ACTS website at http://acts.nersc.gov). Applicants from any scientific field are encouraged to submit a short abstract describing their work and current or future computational needs. DOE will fully sponsor students and qualified postdoctoral fellows to attend this workshop at Lawrence Berkeley National Laboratory. The deadline for applications is Tuesday, August 1st, 2000, and must be sent to acts-workshop@nersc.gov. Students must submit a short abstract of the current work. A letter of recommendation from the student's advisor with the student's name in the subject line must also be received before the deadline. Others applicants must submit a letter outlining their current work and future plans and needs for computational resources with a list of publications. For more information on the workshop, please visit: http://acts.nersc.gov/wrkshop or contact Tony Drummond at (510) 486-7624 or Osni Marques at (510) 486-5290 ------------------------------------------------------- Date: Thu, 15 Jun 2000 09:50:13 -0400 (EDT) From: Lothar Reichel Subject: Conference on Inverse Problems Preliminary announcement for conference on Applied Inverse Problems: Theoretical and Computational Aspects to be held June 18-22, 2001, at Montecatini Terme, Italy. The scope of the meeting is to bring together scientists and engineers who are working on theoretical and computational aspects of inverse problems. The meeting will consist of longer invited lectures, minisymposia and contributed talks. The following speakers have agreed to present invited lectures: S. Arridge A. Bjorck T. F. Chan L. Elden H. W. Engl G. H. Golub P. C. Hansen J. Nagy Z. Nashed F. Natterer R. J. Plemmons F. Santosa E. Somersalo Organizing and Scientific Committee: M. Bertero, D. Calvetti, T.F. Chan, G.H. Golub, G. Inglese, A. Murli, R.J. Plemmons, L. Reichel, S. Seatzu, F. Sgallari, G. Talenti Further information will be made available in early August. Questions about the conference can be sent to aip2001@mcs.kent.edu. ------------------------------------------------------- From: Jens Burmeister Date: Tue, 27 Jun 2000 08:37:27 +0200 Subject: GAMM Seminar on Construction of Grid Generation Algorithms Dear colleagues, I'm very pleased to announce the 17th GAMM-Seminar Leipzig on Construction of Grid Generation Algorithms February 1st to 3rd, 2001. Chairmanship: Wolfgang Hackbusch (Leipzig) Ulrich Langer (Linz) Location: Max-Planck-Institute for Mathematics in the Sciences, Leipzig, Germany. The first fifteen GAMM-Seminars were held in 1984, 1986-1999 at the Christian-Albrechts-University to Kiel under the title Annual GAMM-Seminar Kiel. In 2000, the sixteenth seminar took place at the Max-Planck-Institute for Mathematics in the Sciences in Leipzig. For more information please visit our website http://www.mis.mpg.de/conferences/gamm/ With best regards Jens Burmeister ------------------------------------------------------- Date: Tue, 4 Jul 2000 13:04:47 +0200 (MET DST) From: Stefan Funken Subject: GAMM Workshop on Computational Electromagnetics GAMM-WORKSHOP ON COMPUTATIONAL ELECTROMAGNETICS Mathematisches Seminar, Christian-Albrechts-University of Kiel, Germany, January 26th to 28th, 2001 Organizers: C. Carstensen (Kiel) S.A. Funken (Kiel) W. Hackbusch (Kiel/Leipzig) R.H.W. Hoppe (Augsburg) P. Monk (Newark, Delaware, U.S.A.) This workshop is intended to reflect the state-of-the-art of the foundations, methods, and algorithmic tools of computational electromagnetics. The spectrum of the topics to be presented and discussed ranges from the mathematical analysis of Maxwell's equations, different discretization schemes and their efficient numerical realizations to the application of optimization and optimal control methods as well as the appropriate treatment of inverse problems. Topics of interest include the following: o Mathematical analysis of Maxwell's equations in the low- and high-frequency range, o Numerical analysis of discretization methods, o Applications of multigrid/multilevel and domain decomposition techniques, o Multipole methods, o A posteriori error analysis, o Adaptivity in space-time discretizations, o Optimal control, structural optimization, o Inverse modeling/parameter identification. Call for papers: Participants wanting to give a talk (20 min) should submit an abstract before September 30, 2000. Notification of acceptance will be given in October, 2000. All correspondence in connection with the workshop, including registration and submission of abstracts, is to be made via e-mail on address: saf@numerik.uni-kiel.de Visit the homepage on http://www.numerik.uni-kiel.de/cc/work01.html where more information on hotel reservation and a registration form is available. ------------------------------ End of MGNet Digest **************************