## N-body Simulation I: Fast Algorithms for Potential Field Evaluation and Trummer's Problem (1996)

Citations: | 7 - 5 self |

### BibTeX

@TECHREPORT{Reif96n-bodysimulation,

author = {John H. Reif and Stephen R. Tate},

title = {N-body Simulation I: Fast Algorithms for Potential Field Evaluation and Trummer's Problem},

institution = {},

year = {1996}

}

### OpenURL

### Abstract

In this paper, we describe a new approximation algorithm for the n-body problem. The algorithm is a non-trivial modification of the fast multipole method that works in both two and three dimensions. Due to the equivalence between the two-dimensional n-body problem and Trummer's problem, our algorithm also gives the fastest known approximation algorithm for Trummer's problem. Let A be the sum of the absolute values of the particle charges in the n-body problem under consideration (or the sum of the masses if the simulation is gravitational). To approximate the particle potentials with error bound ffl, we let p = dlog(A=ffl)e and give complexity bounds in terms of p. Note that, under reasonable assumptions on the particle charges, if we desire the output to be accurate to b bits, then p = \Theta(b). In two dimensions, our algorithm runs in time O(n log 2 p), which is a substantial improvement over the previous best algorithm which requires \Theta(np log p) time. We also apply our new ...

### Citations

2564 |
h~ Design and Analysis of Computer Algorithms
- Hopcroft, Ullman
- 1974
(Show Context)
Citation Context ...f the series. Evaluating either a truncated multipole expansion or a truncated Taylor series is simply a multipoint polynomial evaluation, so using standard methods (originally from [27, 22], but see =-=[1]-=- or [4] for a textbook description) we can evaluate a p term polynomial at ssp points in time O(s log 2 p). The corresponding parallel algorithm runs in time O(log 2 p log p) with O(s= log p) processo... |

794 | A fast algorithm for particle simulations
- Greengard, Rohklin
- 1987
(Show Context)
Citation Context ... some basic facts from complex analysis are recalled. The following discussion is a standard encoding of two dimensional n-body problems in the complex plane, which can also be found, for example, in =-=[12]-=-. In particular, the force equation (1) is induced by a potential field OE, which at point pm is OE(p m ) = n X i=1 i6=m \Gammaq i ln kp i \Gamma pm k: (2) When the p j 's are two dimensional, this eq... |

426 |
Computer Simulation Using Particles
- Hockney, Eastwood
- 1988
(Show Context)
Citation Context ...set of n charged particles, where the particles interact under a self-induced electrostatic or gravitational potential field. Generally, these simulations are done by time stepping (see, for example, =-=[3, 15, 17, 19, 28, 37]-=- for details). These simulations are one of the heaviest users of supercomputer cycles, and are widely used by astronomers, chemists, and biochemists, and to a lesser degree physicists 2 . For example... |

367 | The Rapid Evaluation of Potential Fields in Particle Systems - GREENGARD - 1988 |

249 | A decomposition of multidimensional point sets with applications to k-nearest-neighbors and n-body potential fields
- Callahan, Kosaraju
- 1995
(Show Context)
Citation Context ...ltipole algorithm, and devised a framework for spatial decomposition that proved to be useful not only for the n-body problem, but also for many other interesting problems from computational geometry =-=[6]-=-. These decomposition methods did not improve the running time of the multipole algorithm, but did remove the assumptions about the input distribution that were required by Greengard and Rokhlin's alg... |

215 |
Rapid solution of integral equations of classical potential theory
- Rokhlin
- 1985
(Show Context)
Citation Context ...cations of Trummer's Problem The further applications of Trummer's problem are quite extensive. One is the solution of 2D PDEs by Potential Methods, as pioneered by Rokhlin. An early paper of Rokhlin =-=[39]-=- showed that the solution of Laplace's equation in 2D with irregular boundary conditions (a difficult, fundamental problem in numerical analysis) reduced to Trummer's problem. Another major applicatio... |

171 |
FastCap: A multipole accelerated 3-D capacitance extraction program
- Nabors, White
- 1991
(Show Context)
Citation Context ...set of n charged particles, where the particles interact under a self-induced electrostatic or gravitational potential field. Generally, these simulations are done by time stepping (see, for example, =-=[3, 15, 17, 19, 28, 37]-=- for details). These simulations are one of the heaviest users of supercomputer cycles, and are widely used by astronomers, chemists, and biochemists, and to a lesser degree physicists 2 . For example... |

150 |
editor. Synthesis of Parallel Algorithms
- Reif
- 1993
(Show Context)
Citation Context ...putation, and algorithms (such as the one in this paper) designed for an EREW PRAM can be easily used with less restrictive models. For more information on models of parallel computation, see [18] or =-=[35]-=-. Consider the problem of evaluating expression trees, that is, trees with the leaves labeled with numerical values and internal nodes labeled with arithmetic operations. For example, x + y is represe... |

139 |
Numerical study of slightly viscous flow
- Chorin
- 1973
(Show Context)
Citation Context ...onditions (a difficult, fundamental problem in numerical analysis) reduced to Trummer's problem. Another major application is in fluid mechanics; particularly the vortex method of Anderson and Chorin =-=[2, 8]-=-, which is used in 2D turbulent flow simulations and plasma simulations. In particular Greengard [11] showed that an important case of 2D fluid flow through a channel reduces to a Poisson problem and ... |

137 | A fast adaptive multipole algorithm for particle simulations - Carrier, Greengard, et al. - 1988 |

49 |
Optimal parallel evaluation of tree-structured computations by raking
- Kosaraju, Delcher
- 1988
(Show Context)
Citation Context ...hm was designed to solve precisely this problem. While the original work of Miller and Reif [26] produced a randomized algorithm for tree contraction, later work (such as that by Kosaraju and Delcher =-=[21]-=-) gave purely deterministic algorithms for tree contraction. It is these later algorithms that we use in this paper, so our algorithms are deterministic and the time bounds are true worst-case bounds.... |

47 | Fast algorithms for multiple evaluations of the Riemann zeta function
- Odlyzko, Schönhage
- 1988
(Show Context)
Citation Context ...hesis to be verified for large sets of zeros. Odlyzko and Schonhage have shown that multiple evaluations 2 of the zeta function can be done very efficiently, if some time is allowed for preprocessing =-=[30]-=-. The dominating computation in the preprocessing is the evaluation of the function n X k=1 a k z \Gamma b k ; (4) for values a 1 ; a 2 ; \Delta \Delta \Delta ; a n , and b 1 ; b 2 ; \Delta \Delta \De... |

46 |
On computations with dense structured matrices
- Pan
- 1990
(Show Context)
Citation Context ...pute qM by performing ds=pe row-vector times Vandermondematrix products. Due to the structure of the Vandermonde matrix, Pan has shown that each of these products can be computed in O(p log 2 p) time =-=[31]-=-, giving a total complexity of O(s log 2 p). Simple parallelization of the vector-times-Vandermonde algorithm given by Pan (who only describes the sequential version) [31], along with the observation ... |

44 |
A method of local corrections for computing the velocity field due to a distribution of vortex blobs
- Anderson
- 1986
(Show Context)
Citation Context ...onditions (a difficult, fundamental problem in numerical analysis) reduced to Trummer's problem. Another major application is in fluid mechanics; particularly the vortex method of Anderson and Chorin =-=[2, 8]-=-, which is used in 2D turbulent flow simulations and plasma simulations. In particular Greengard [11] showed that an important case of 2D fluid flow through a channel reduces to a Poisson problem and ... |

42 |
Implementing the fast multipole method in three dimensions
- Schmidt, Lee
- 1991
(Show Context)
Citation Context ...time O(np 2 ), which was later improved to O(np log p) time [14]. The fast multipole algorithm attracted a lot of attention, and numerous studies and implementations have been made (see, for example, =-=[5, 20, 23, 25, 24, 40]-=-). The running time's dependence on p can be considerable, since n-body simulations for molecular dynamics typically require values of p of at least 16 (and often much more for high accuracy planetary... |

34 | Computational structure of the N-body problem
- Katzenelson
- 1989
(Show Context)
Citation Context ...time O(np 2 ), which was later improved to O(np log p) time [14]. The fast multipole algorithm attracted a lot of attention, and numerous studies and implementations have been made (see, for example, =-=[5, 20, 23, 25, 24, 40]-=-). The running time's dependence on p can be considerable, since n-body simulations for molecular dynamics typically require values of p of at least 16 (and often much more for high accuracy planetary... |

32 |
Rapid evaluation of potential Fields in three dimensions, in Vortex Methods, edited by C. Anderson and C
- Greengard, Rokhlin
- 1988
(Show Context)
Citation Context ...-body problem at about the same time, Greengard and Rokhlin gave provably good approximation algorithms (known as the fast multipole algorithm) for the n-body problem in both two and three dimensions =-=[12, 13, 14]-=-. In order to understand the complexity of their algorithm, we define A = P n i=1 jq i j to be the sum of the absolute particle charges, and if we desire the output to meet error bound ffl we define p... |

30 |
survey of parallel algorithms in numerical linear algebra
- Heller
- 1978
(Show Context)
Citation Context ...reation and translation operators are dominated by a computation of p terms of p different linear recurrences, which can be done in O(log p) time with O(p 2 = log p) processors by standard techniques =-=[16]-=-. Sequential Parallel Parallel Operation Time Processors Time Reference T mc (p; s) O(sp 2 ) O(sp 2 = log p) O(log p) [10, Chapter 3] Tmm (p) O(p 2 log p) O(p 2 ) O(log p) [14] Tm\Gammap (p) O(p 2 log... |

28 | Fast modular transforms
- Borodin, Moenck
- 1974
(Show Context)
Citation Context ...near in the size of the series. Evaluating either a truncated multipole expansion or a truncated Taylor series is simply a multipoint polynomial evaluation, so using standard methods (originally from =-=[27, 22]-=-, but see [1] or [4] for a textbook description) we can evaluate a p term polynomial at ssp points in time O(s log 2 p). The corresponding parallel algorithm runs in time O(log 2 p log p) with O(s= lo... |

25 |
The Parallel Fast Multipole Algorithm in Three Dimensions
- Leathrum, Board
- 1992
(Show Context)
Citation Context ...time O(np 2 ), which was later improved to O(np log p) time [14]. The fast multipole algorithm attracted a lot of attention, and numerous studies and implementations have been made (see, for example, =-=[5, 20, 23, 25, 24, 40]-=-). The running time's dependence on p can be considerable, since n-body simulations for molecular dynamics typically require values of p of at least 16 (and often much more for high accuracy planetary... |

23 |
A fast algorithm for the multiplication of generalized Hilbert matrices with vectors
- Gerasoulis
- 1988
(Show Context)
Citation Context ...o-dimensional n-body problem and Trummer's problem, previous work comes from both these areas. Two years after Golub published Trummer's problem as a challenge, a solution was published by Gerasoulis =-=[9]-=-, who gave an O(n log 2 n) time algebraic algorithm that solved Trummer's problem. Unfortunately, Gerasoulis' algorithm turned out to be numerically unstable for large n. Looking at the n-body problem... |

21 |
On the efficient implementation of the fast multipole algorithm
- Greengard, Rokhlin
- 1988
(Show Context)
Citation Context ...-body problem at about the same time, Greengard and Rokhlin gave provably good approximation algorithms (known as the fast multipole algorithm) for the n-body problem in both two and three dimensions =-=[12, 13, 14]-=-. In order to understand the complexity of their algorithm, we define A = P n i=1 jq i j to be the sum of the absolute particle charges, and if we desire the output to meet error bound ffl we define p... |

21 | Problems on Algorithms - Parberry - 1995 |

19 |
Já Já , An Introduction to Parallel Algorithms
- unknown authors
- 1992
(Show Context)
Citation Context ...llel computation, and algorithms (such as the one in this paper) designed for an EREW PRAM can be easily used with less restrictive models. For more information on models of parallel computation, see =-=[18]-=- or [35]. Consider the problem of evaluating expression trees, that is, trees with the leaves labeled with numerical values and internal nodes labeled with arithmetic operations. For example, x + y is... |

19 |
Parallel tree contraction, Part 1: Fundamentals
- Miller, Reif
- 1989
(Show Context)
Citation Context ...rallel this problem poses a challenge if the tree is not balanced. The widely known tree contraction algorithm was designed to solve precisely this problem. While the original work of Miller and Reif =-=[26]-=- produced a randomized algorithm for tree contraction, later work (such as that by Kosaraju and Delcher [21]) gave purely deterministic algorithms for tree contraction. It is these later algorithms th... |

17 |
List ranking and parallel tree contraction
- Reid-Miller, Miller, et al.
- 1993
(Show Context)
Citation Context ...r, Miller, and Modugno demonstrate this technique on a similar problem: given a tree in which each node contains an integer label, for each node compute the maximum value held in any of its ancestors =-=[34]-=-. Given box/local-expansion pairs as values, our algorithm may use the same process for step 4, giving a parallel complexity of O(T (p) log jDT j) time and O(P (p)jDT j= log jDT j) processors, where T... |

16 |
The 10 th zero of the Riemann zeta function and 70 million of its neighbors
- Odlyzko
- 1989
(Show Context)
Citation Context ...elta \Delta \Delta ; b n , at z equal to the n nth roots of unity. The algorithm of [30] takes time O(np c ) for p bit accuracy, for some moderate constant c, which is a considerable improvement (see =-=[29]-=-) over the best previously known n 2 naive algorithm for this problem for the ranges of n of interest (Odlyzko's computations ranged over 1,000,000). The summation shown in (4) can easily be reduced t... |

15 |
Molecular dynamics simulations in biology. Nature
- Karplus, Petsko
- 1990
(Show Context)
Citation Context ...set of n charged particles, where the particles interact under a self-induced electrostatic or gravitational potential field. Generally, these simulations are done by time stepping (see, for example, =-=[3, 15, 17, 19, 28, 37]-=- for details). These simulations are one of the heaviest users of supercomputer cycles, and are widely used by astronomers, chemists, and biochemists, and to a lesser degree physicists 2 . For example... |

12 |
The power of combining the techniques of algebraic and numerical computing: improved approximate multipoint polynomial evaluation and improved multipole algorithms
- Pan, Reif, et al.
- 1992
(Show Context)
Citation Context ...parallel algorithms AMS subject classifications: 68Q25, 65F30, 70F10 Abbreviated title: N-body Simulation I: Potential Field Evaluation A preliminary version of these results was published in FOCS 92 =-=[32]-=-. y Supported by NSF Grant NSF-IRI-91-00681, Rome Labs Contracts F30602-94-C-0037, ARPA/SISTO contracts N00014-91-J-1985, and N00014-92-C-0182 under subcontract KI-92-01-0182. z Supported in part by N... |

9 |
Fast evaluation and interpolation
- Kung
- 1973
(Show Context)
Citation Context ...near in the size of the series. Evaluating either a truncated multipole expansion or a truncated Taylor series is simply a multipoint polynomial evaluation, so using standard methods (originally from =-=[27, 22]-=-, but see [1] or [4] for a textbook description) we can evaluate a p term polynomial at ssp points in time O(s log 2 p). The corresponding parallel algorithm runs in time O(log 2 p log p) with O(s= lo... |

6 | Fast Spatial Decomposition and Closest Pair Computation for Limited Precision Input
- Reif, Tate
(Show Context)
Citation Context ...trivial solution which may be applied to other geometric problems such as closest pair problems, and the authors have described the general limited precision spatial decomposition in a separate paper =-=[38]-=-. The decomposition algorithm from that paper is summarized in the following Lemma (based on the Theorem 2.1 and Lemma 3.1 from [38]). Lemma 2.1 Given n particles in d dimensions (d a constant), where... |

5 | Potential flow in channels
- Greengard
- 1990
(Show Context)
Citation Context ...r major application is in fluid mechanics; particularly the vortex method of Anderson and Chorin [2, 8], which is used in 2D turbulent flow simulations and plasma simulations. In particular Greengard =-=[11]-=- showed that an important case of 2D fluid flow through a channel reduces to a Poisson problem and hence to variants of Trummer's problem. 1.2 Motivation and History The goal of molecular dynamics stu... |

3 |
Parallelization of the fast multipole algorithm using the b012 transputer network
- Leathrum, Board
- 1991
(Show Context)
Citation Context |

2 |
N-body simulation I: Potential field evaluation
- Reif, Tate
- 1992
(Show Context)
Citation Context ...resented here improved the running time of the multipole al4 gorithm to O(n log 2 p) by using methods similar to those presented in this paper, but using a much more complicated spatial decomposition =-=[36]-=-. A summary of these results was subsequently published [32], but since that time the algorithms (as presented here) have completely changed. In one further important piece of previous work, Callahan ... |

1 |
Eds. Computer Simulation of Chemical and
- Beveridge, Jorgensen
- 1986
(Show Context)
Citation Context |

1 |
High performance implementations of the fast multipole algorithm
- Board, Batchelor, et al.
- 1990
(Show Context)
Citation Context |

1 |
Mapping the adaptive multipole algorithm onto mind systems
- Leathrum, Board
- 1992
(Show Context)
Citation Context |

1 |
N-body simulation II: Simulation of moving particles
- Reif, Tate
- 1995
(Show Context)
Citation Context ...e in the current paper --- in a companion paper, we examine the related problem of simulating the motion of particles over time, giving both lower and upper bounds on the complexity required to do so =-=[37]-=-. 1.1 Trummer's problem Trummer's problem is the problem of multiplying a particular structured matrix T times an arbitrary vector ~y. The matrix T is completely specified by n complex values c 1 ; \D... |