## Skeletons from the Treecode Closet (1994)

Venue: | J. Comp. Phys |

Citations: | 40 - 10 self |

### BibTeX

@ARTICLE{Salmon94skeletonsfrom,

author = {John K. Salmon and Michael S. Warren},

title = {Skeletons from the Treecode Closet},

journal = {J. Comp. Phys},

year = {1994},

volume = {111},

pages = {136--155}

}

### Years of Citing Articles

### OpenURL

### Abstract

We consider treecodes (N-body programs which use a tree data structure) from the standpoint of their worst-case behavior. That is, we derive upper bounds on the largest possible errors that are introduced into a calculation by use of various multipole acceptability criteria (MAC). We find that the conventional Barnes-Hut MAC can introduce potentially unbounded errors unless ` ! 1= p 3, and that this behavior while rare, is demonstrable in astrophysically reasonable examples. We consider two other MACs closely related to the BH MAC. While they don't admit the same unbounded errors, they nevertheless require extraordinary amounts of CPU time to guarantee modest levels of accuracy. We derive new error bounds based on some additional, easily computed moments of the mass distribution. These error bounds form the basis for four new MACs which can be used to limit the absolute or relative error introduced by each multipole evaluation, or, with the introduction of some additional data struc...

### Citations

2197 |
The art of computer programming
- Knuth
- 1973
(Show Context)
Citation Context ... computed and inserted back into the priority queue. This process continues until Eq.(15) or Eq.(17) is satisfied. 3 4.4 Optimizations There are a number of possible ways to implement priority queues =-=[23]-=-. Typically the cost of insertion and deletion is logarithmic in the size of the queue. Experience suggests that the queues arising in treecodes will contain up to about one 2 a priority queue [23] is... |

2057 |
Handbook of mathematical functions
- Abramowitz, Stegun
- 1972
(Show Context)
Citation Context ... if the maximum extent, b max is known (or bounded), then upper bounds may be placed on high-order moments by: B (n+k)sb k max B (n) (64) The following formulas, which follow from Holder's inequality =-=[29]-=- are useful for even and odd p respectively: B (p+2)sB 2 (p) B (p\Gamma2) = \Gamma ffi (p=2) M (p) \Delta 2 ffi ((p\Gamma2)=2) M (p\Gamma2) (65) B 2 (p+2)sB 5 (p\Gamma1) B 3 (p\Gamma3) = \Gamma ffi ((... |

686 | A fast algorithm for particle simulations
- Greengard, Rokhlin
- 1987
(Show Context)
Citation Context ...dependently by Appel[1], Jernigan and Porter[2] and Barnes and Hut[3]. Subsequently, a number of other authors have expanded on the theme of multipole approximations and hierarchical data structures. =-=[4, 5, 6, 7, 8]. The comp-=-uter programs described by these authors are known collectively as "treecodes" because the underlying data structure in all cases is a tree. These programs typically can compute the approxim... |

349 |
Computer Simulation Using Particles
- Hockney, Eastwood
(Show Context)
Citation Context ...ntire machinery of treecodes, may be formulated for arbitrary Green's functions. It is possible to apply treecodes to systems with entirely different Green's functions, such as may arise in chemistry =-=[26]-=- or fluid dynamics[27, 28]. The acceleration is obtained from the gradient of the potential: ~a(~r) = \GammarOE(r): (20) The multipole expansion is obtained by selecting a particular point, (usually t... |

339 |
The Rapid Evaluation of Potential Fields in Particle Systems
- Greengard
- 1988
(Show Context)
Citation Context ... ae field. The maximum relative error is thus a monotonically increasing function of b max=d, which is useful primarily for motivating the Bmax MAC. Eq.(69) is similar to results obtained by Greengard=-=[4]-=-, where it is used to select a suitable multipole order, p. Note, however, that the relative error in the acceleration is considerably larger than the relative error in the potential. Thus, if one bou... |

144 |
An efficient program for many-body simulation
- Appel
- 1985
(Show Context)
Citation Context ...ribution of N \Gamma 1 terms in the update of each body. Algorithms which use a hierarchical data structure and an approximate force-law for aggregates of bodies were introduced independently by Appel=-=[1]-=-, Jernigan and Porter[2] and Barnes and Hut[3]. Subsequently, a number of other authors have expanded on the theme of multipole approximations and hierarchical data structures. [4, 5, 6, 7, 8]. The co... |

93 |
A hierarchical o(nlogn) force-calculation algorithm
- Barnes, Hut
- 1986
(Show Context)
Citation Context ...each body. Algorithms which use a hierarchical data structure and an approximate force-law for aggregates of bodies were introduced independently by Appel[1], Jernigan and Porter[2] and Barnes and Hut=-=[3]-=-. Subsequently, a number of other authors have expanded on the theme of multipole approximations and hierarchical data structures. [4, 5, 6, 7, 8]. The computer programs described by these authors are... |

91 |
Parallel hierarchical N-body methods
- Salmon
- 1991
(Show Context)
Citation Context ...mega ` ffi (2) M (6) \Gamma 1 21 ffi\Omega (ffi (3) M (6) ) " : B.1 The error terms. The error term associated with the Plummer potential is expressible in terms of elementary functions and integ=-=rals [11]-=- but the precise form is unwieldy and not terribly useful. The error derived from a Newtonian potential is somewhat more manageable. We assume, without loss of generality, that the center-of-mass, ~r ... |

86 | Astrophysical n-body simulations using hierarchical tree data structures
- Warren, Salmon
- 1992
(Show Context)
Citation Context ... positions of all bodies in the cell are known. The ability to evaluate the MAC before the contents of a cell have been determined makes it attractive for use on parallel distributed memory computers =-=[11, 12]-=-, where one might wish to evaluate the MAC for cells whose data reside in another processor's memory. 2.3 The Bmax MAC. Another MAC is motivated by Eq.(70), according to which the largest possible err... |

80 | An implementation of the fast multipole method without multipoles - Anderson - 1992 |

42 |
An O(N) Algorithm for Three-Dimensional N-Body Simulations
- Zhao
- 1987
(Show Context)
Citation Context ...dependently by Appel[1], Jernigan and Porter[2] and Barnes and Hut[3]. Subsequently, a number of other authors have expanded on the theme of multipole approximations and hierarchical data structures. =-=[4, 5, 6, 7, 8]. The comp-=-uter programs described by these authors are known collectively as "treecodes" because the underlying data structure in all cases is a tree. These programs typically can compute the approxim... |

32 |
Computing three-dimensional incompressible flows with vortex elements
- Leonard
- 1985
(Show Context)
Citation Context ...ecodes, may be formulated for arbitrary Green's functions. It is possible to apply treecodes to systems with entirely different Green's functions, such as may arise in chemistry [26] or fluid dynamics=-=[27, 28]-=-. The acceleration is obtained from the gradient of the potential: ~a(~r) = \GammarOE(r): (20) The multipole expansion is obtained by selecting a particular point, (usually the center-of-mass of the c... |

25 |
TREESPH: a unification of SPH with the hierarchical tree method
- Hernquist, Katz
- 1989
(Show Context)
Citation Context ...tune the errors associated with different bodies if external considerations make it important to achieve high accuracy on only a subset of bodies, e.g., if an individual-timestep integrator is in use =-=[17, 18, 19]. One can -=-also achieve control of relative errors rather than absolute errors. There are two possibilities. If one has access to an estimate of the "exact" acceleration (partial or total), perhaps fro... |

25 |
A modified tree code: Don’t laugh; it runs
- Barnes
- 1990
(Show Context)
Citation Context ...ed so far. 4.2 Optimizations As written, the Absolute MAC requires evaluation of a fairly complicated and costly expression (Eq.(9)) each time it is called. It is likely that techniques like those in =-=[22]-=- can be used to greatly reduce the number of times the MAC is called, at the expense of a few additional interaction evaluations. An additional optimization is extremely powerful if one can arrange th... |

16 |
Dark halos formed via dissipationless collapse. I - Shapes and alignment of angular momentum
- Warren, Quinn, et al.
- 1992
(Show Context)
Citation Context ...ection A. Exact accelerations were computed for 4443 bodies. ffl A single halo model with 54720 bodies. This halo was extracted from from a cosmological simulation with 1.1 million bodies reported in =-=[24]-=-. It is triaxial and highly concentrated toward the middle. Exact accelerations were computed for 4942 bodies. ffl A randomly selected subset of a very large cosmological simulation with 8.8 million b... |

14 |
Performance characteristics of tree codes
- Hernquist
- 1987
(Show Context)
Citation Context ...rs of a given magnitude are acceptable and when are not. We cannot avoid the issue entirely, however, because we must determine the kinds of errors we are talking about. The consensus in astrophysics =-=[13, 14, 10] has been -=-that it is acceptable to measure, e.g., rms errors, in a very small number of model systems. The measured errors are deemed "typical", and further simulations can be performed using the MAC ... |

11 |
Using Parallel Computers for Very Large NBody Simulations: Shell Formation using 180k Particles
- Salmon, Quinn, et al.
- 1990
(Show Context)
Citation Context ...at t = 4:0 is the result of the error introduced by the BH MAC. 5 This is not an entirely contrived example. A very similar system was evolved using the MD MAC (Sec. 2.2), and reported in Salmon et al=-=[25]. �� p 3 P-=- S C x cm Figure 12: Geometry of "Detonating Galaxy" pathological situation for the BH MAC criterion. M ~x 0 ~v 0 r J N primary 1.0 (0.3, 0.3, 0.3) (0,0,0) 1.0 15,000 secondary 0.05 (7.,7.,7... |

10 |
Dynamic mass exchange in doubly degenerate binaries I. 0.9 and 1.2 M⊙ stars, Astrophys
- Benz, Bowers, et al.
- 1990
(Show Context)
Citation Context ...dependently by Appel[1], Jernigan and Porter[2] and Barnes and Hut[3]. Subsequently, a number of other authors have expanded on the theme of multipole approximations and hierarchical data structures. =-=[4, 5, 6, 7, 8]. The comp-=-uter programs described by these authors are known collectively as "treecodes" because the underlying data structure in all cases is a tree. These programs typically can compute the approxim... |

9 |
Topics in vortex methods for the computation of three- and two-dimensional incompressible unsteady flows
- Winckelmans
- 1989
(Show Context)
Citation Context ...ecodes, may be formulated for arbitrary Green's functions. It is possible to apply treecodes to systems with entirely different Green's functions, such as may arise in chemistry [26] or fluid dynamics=-=[27, 28]-=-. The acceleration is obtained from the gradient of the potential: ~a(~r) = \GammarOE(r): (20) The multipole expansion is obtained by selecting a particular point, (usually the center-of-mass of the c... |

8 |
A tree code with logarithmic reduction of force terms, hierarchical regularization of all variables, and explicit accuracy controls
- Jernigan, Porter
- 1989
(Show Context)
Citation Context ...erms in the update of each body. Algorithms which use a hierarchical data structure and an approximate force-law for aggregates of bodies were introduced independently by Appel[1], Jernigan and Porter=-=[2]-=- and Barnes and Hut[3]. Subsequently, a number of other authors have expanded on the theme of multipole approximations and hierarchical data structures. [4, 5, 6, 7, 8]. The computer programs describe... |

7 |
Comparison of two different tree algorithms
- Makino
- 1990
(Show Context)
Citation Context ...with tree-construction in this paper, as it appears to constitute a relatively small fraction of the total expense in using treecodes. Tree construction has been extensively covered in the literature =-=[3, 10, 11, 7]-=-. In this paper we shall assume that trees are made up of cells. A cell represents a bounded region of space, and contains some aggregate information about the bodies that lie within that region. Ofte... |

6 |
Direct methods for N-body simulations. In: Multiple time scales
- Aarseth
- 1985
(Show Context)
Citation Context ...tune the errors associated with different bodies if external considerations make it important to achieve high accuracy on only a subset of bodies, e.g., if an individual-timestep integrator is in use =-=[17, 18, 19]. One can -=-also achieve control of relative errors rather than absolute errors. There are two possibilities. If one has access to an estimate of the "exact" acceleration (partial or total), perhaps fro... |

5 |
Error analysis of a tree code
- Barnes, Hut
- 1989
(Show Context)
Citation Context ...100%. It is curious that the BH MAC can give rise to such large errors when it has been tested so extensively. Indeed, very similar physical systems have been used to verify the accuracy of treecodes =-=[14, 22], but the -=-galaxies in these test case were of equal mass and on an orbit parallel to one of the Cartesian axes. In this configuration, the error does not arise, and the algorithm "passed" the test. A ... |

4 |
Discreteness Noise versus Force Errors in N-Body Simulations
- Hernquist, Hut, et al.
- 1993
(Show Context)
Citation Context ...e figure demonstrates that the bound set by Eq.(9) is very tight. In the mean, the actual error is about 20% of the bound, with occasional partials approaching the bound very closely. Hernquist et al.=-=[20]-=- have asserted that the physics is best preserved when one controls relative partial errors. Their argument rests on the flawed assumption that errors are random and cancel one another out. While this... |

2 |
Cosmological simulations using the hierarchical tree method
- Bouchet, Henquist
- 1988
(Show Context)
Citation Context ...for any (`; p) pair and any MAC. The results are shown in Figure 5. The errors displayed in Figure 5 are an order of magnitude or more worse than the empirical results obtained by a number of authors =-=[13, 16, 14, 10]-=-. We may account for this discrepancy by noting that the other authors report an rms or an average error. Errors approaching the bounds set by Figure 5 are atypical but not impossible. Inspection of F... |

2 |
Techniques and tricks for N-body computation
- Press
- 1986
(Show Context)
Citation Context ...tune the errors associated with different bodies if external considerations make it important to achieve high accuracy on only a subset of bodies, e.g., if an individual-timestep integrator is in use =-=[17, 18, 19]. One can -=-also achieve control of relative errors rather than absolute errors. There are two possibilities. If one has access to an estimate of the "exact" acceleration (partial or total), perhaps fro... |

1 |
Performance analysis of direct N-body simulations
- Makino, Hut
- 1988
(Show Context)
Citation Context ... N bodies in a system in time proportional to N log N or even N . This qualitative change in the time-complexity has led to simulations much larger than were possible with direct-summation algorithms =-=[9]-=-. Obviously, the detailed structure of the tree, and how it is constructed is of great importance in designing treecodes. Nevertheless, except for terminology, we shall not concern ourselves with tree... |

1 |
Error analysis of the GRAPE-1 special purpose N-body machine
- Makino, Ito, et al.
- 1990
(Show Context)
Citation Context ...assumption that errors are random and cancel one another out. While this is true of the N terms that give rise to the total acceleration when using the low-precision arithmetic in the GRAPE processor =-=[21]-=-, it is certainly not true of the O(logN) partials that are added together in a treecode. One erroneous partial in a treecode can introduce errors far in excess of the unavoidable discreteness noise i... |