Parallel KS Block-Step Method. Sverre Aarseth. Institute of Astronomy, Cambridge

Similar documents
Post-Newtonian N-body Codes. Sverre Aarseth. Institute of Astronomy, Cambridge

. α β γ δ ε ζ η θ ι κ λ μ Aμ ν(x) ξ ο π ρ ς σ τ υ φ χ ψ ω. Α Β Γ Δ Ε Ζ Η Θ Ι Κ Λ Μ Ν Ξ Ο Π Ρ Σ Τ Υ Φ Χ Ψ Ω. Friday April 1 ± ǁ

and in each case give the range of values of x for which the expansion is valid.

ASSIGNMENT COVER SHEET omicron

. α β γ δ ε ζ η θ ι κ λ μ Aμ ν(x) ξ ο π ρ ς σ τ υ φ χ ψ ω. Α Β Γ Δ Ε Ζ Η Θ Ι Κ Λ Μ Ν Ξ Ο Π Ρ Σ Τ Υ Φ Χ Ψ Ω. Wednesday March 30 ± ǁ

K e sub x e sub n s i sub o o f K.. w ich i sub s.. u ra to the power of m i sub fi ed.. a sub t to the power of a

4sec 2xtan 2x 1ii C3 Differentiation trig

A2 Assignment zeta Cover Sheet. C3 Differentiation all methods. C3 Sketch and find range. C3 Integration by inspection. C3 Rcos(x-a) max and min

Beauty Contests and Iterated Expectations in Asset Markets. Franklin Allen Stephen Morris Hyun Song Shin

Numerical Approximation of Phase Field Models

Further Maths A2 (M2FP2D1) Assignment ψ (psi) A Due w/b 19 th March 18

b e i ga set s oane f dast heco mm on n va ns ing lo c u soft w section

CSE 1400 Applied Discrete Mathematics Definitions

A2 Assignment lambda Cover Sheet. Ready. Done BP. Question. Aa C4 Integration 1 1. C4 Integration 3

License: Creative Commons Attribution Non-commercial No Derivatives

Differential Equations

Rømer Science Mission Plan

Post-Newtonian N-body Dynamics

Numerical Methods in Physics and Astrophysics

Dynamic Asset Allocation - Identifying Regime Shifts in Financial Time Series to Build Robust Portfolios

Y1 Double Maths Assignment λ (lambda) Exam Paper to do Core 1 Solomon C on the VLE. Drill

Mathematics Review Exercises. (answers at end)

Numerical Methods in Physics and Astrophysics

A Method for Solving and Estimating Heterogenous Agent Macro Models (by Thomas Winberry)

PRELIMINARIES FOR GENERAL TOPOLOGY. Contents

Adaptive discretization and first-order methods for nonsmooth inverse problems for PDEs

Lattice Bhatnagar Gross Krook model for the Lorenz attractor

STOP, a i+ 1 is the desired root. )f(a i) > 0. Else If f(a i+ 1. Set a i+1 = a i+ 1 and b i+1 = b Else Set a i+1 = a i and b i+1 = a i+ 1

ECE257 Numerical Methods and Scientific Computing. Ordinary Differential Equations

A Dynamic Model for Investment Strategy

Symbols and dingbats. A 41 Α a 61 α À K cb ➋ à esc. Á g e7 á esc. Â e e5 â. Ã L cc ➌ ã esc ~ Ä esc : ä esc : Å esc * å esc *

Computational Astrophysics

Numerical Solutions to Partial Differential Equations

Nonlinear wave-wave interactions involving gravitational waves

Review Higher Order methods Multistep methods Summary HIGHER ORDER METHODS. P.V. Johnson. School of Mathematics. Semester

Parameterized Expectations Algorithm

AN OVERVIEW. Numerical Methods for ODE Initial Value Problems. 1. One-step methods (Taylor series, Runge-Kutta)

Data Analysis Question Sheet

PowerPoints organized by Dr. Michael R. Gustafson II, Duke University

Asynchronous Training in Wireless Sensor Networks

1 Integration of Rational Functions Using Partial Fractions

Outline. Logic. Definition. Theorem (Gödel s Completeness Theorem) Summary of Previous Week. Undecidability. Unification

PowerPoints organized by Dr. Michael R. Gustafson II, Duke University

Chapter 7: Natural Convection

5.1 Classical Harmonic Oscillator

T m / A. Table C2 Submicroscopic Masses [2] Symbol Meaning Best Value Approximate Value

Nonlinear equations. Norms for R n. Convergence orders for iterative methods

Milnor s Exotic 7-Spheres Jhan-Cyuan Syu ( June, 2017 Introduction

From a Mesoscopic to a Macroscopic Description of Fluid-Particle Interaction

Jim Lambers MAT 772 Fall Semester Lecture 21 Notes

MEMORY EFFECT IN HOMOGENIZATION OF VISCOELASTIC KELVIN-VOIGT MODEL WITH TIME DEPENDENT COEFFICIENTS

(1) u (t) = f(t, u(t)), 0 t a.

Persistence of Plummer-Distributed Small Globular Clusters as a Function of Primordial- Binary Population Size

Wavelets For Computer Graphics

A class of non-convex polytopes that admit no orthonormal basis of exponentials

Hoover Institution, Stanford University and NBER. University of Alicante and Hoover Institution, Stanford University

SHANGHAI JIAO TONG UNIVERSITY LECTURE

2 Post-Keplerian Timing Parameters for General Relativity

Application of the perturbation iteration method to boundary layer type problems

Stellar Dynamics and Structure of Galaxies

Do not turn over until you are told to do so by the Invigilator.

3.4 Numerical orbit integration

QFT Perturbation Theory

Analysis Methods in Atmospheric and Oceanic Science

A Detailed Look at a Discrete Randomw Walk with Spatially Dependent Moments and Its Continuum Limit

VARIATIONAL PRINCIPLE FOR THE ENTROPY

R k. t + 1. n E t+1 = ( 1 χ E) W E t+1. c E t+1 = χ E Wt+1 E. Γ E t+1. ) R E t+1q t K t. W E t+1 = ( 1 Γ E t+1. Π t+1 = P t+1 /P t

Initial value problems for ordinary differential equations

ASSIGNMENT BOOKLET. Numerical Analysis (MTE-10) (Valid from 1 st July, 2011 to 31 st March, 2012)

Numerical Methods - Initial Value Problems for ODEs

Time Evolution of Matter States

CHAPTER 6 : LITERATURE REVIEW

P E R E N C O - C H R I S T M A S P A R T Y

22.615, MHD Theory of Fusion Systems Prof. Freidberg Lecture 14: Formulation of the Stability Problem

Input to state Stability

DELFT UNIVERSITY OF TECHNOLOGY Faculty of Electrical Engineering, Mathematics and Computer Science

(f(x) P 3 (x)) dx. (a) The Lagrange formula for the error is given by

Simulation Method for Solving Stochastic Differential Equations with Constant Diffusion Coefficients

Variations on Backpropagation


Research Article On Existence, Uniform Decay Rates, and Blow-Up for Solutions of a Nonlinear Wave Equation with Dissipative and Source

Periodic orbits high above the ecliptic plane in the solar sail 3-body problem

Numerical Solutions to Partial Differential Equations

MAS114: Exercises. October 26, 2018

Lecture 3 Optimization methods for econometrics models

High-Gain Observers in Nonlinear Feedback Control. Lecture # 3 Regulation

ON ESTIMATION OF TEMPERATURE UNCERTAINTY USING THE SECOND ORDER ADJOINT PROBLEM

Theoretical Statistical Correlation for Biometric Identification Performance

Talk 3: Examples: Star exponential

Numerical Solutions to Partial Differential Equations

Candidates must show on each answer book the type of calculator used. Log Tables, Statistical Tables and Graph Paper are available on request.

Massive star clusters

Online appendix to On the stability of the excess sensitivity of aggregate consumption growth in the US

A sharp diffuse interface tracking method for approximating evolving interfaces

Weak convergence and Brownian Motion. (telegram style notes) P.J.C. Spreij

Electromagnetically Induced Flows in Water

Planet formation (aero)dynamics project

Waveform inversion and time-reversal imaging in attenuative TI media


Mark Gales October y (x) x 1. x 2 y (x) Inputs. Outputs. x d. y (x) Second Output layer layer. layer.

Transcription:

Parallel KS Block-Step Method Sverre Aarseth Institute of Astronomy, Cambridge Code Overview Hermite KS Prediction & Correction Iteration Time-Steps Decision-Making Binary Project

Code Overview Directories Ncode, Docs, Chain, Nchain, ARchain, ARint, Block Standard make nbody6 in dir Ncode nbody6 Block-steps make nbody6 in dir Block nbody6b NBODY7 make -f Makefile7 nbody6 in Ncode nbody7 NBODY7b make -f Makefile7 nbody6 in Block nbody7b GPU make gpu in dir GPU2 produces nbody6.gpu GPUb make -f Makefile6 gpu nbody6b.gpu ARC make -f Makefile7 gpu nbody7.gpu ARCb make -f Makefile8 gpu nbody7b.gpu

Program Flow KS predict Predict u, u to order u (5) Perturbers Predict r, ṙ and obtain F, Ḟ Corrector Temporary correct u,u & h Iteration One or more, same F, Ḟ Final stage Form u (4), u (5), copy to common Time Derivatives up to t (5) or t (6) Time-step Set τ, convert to t & truncate Invert Newton Raphson gives τ bs

Hermite KS Standard KS New notation u h t = 2 hu + 2 R LT F kl = 2u L T F kl = u u P = F kl F u = u Q = L T P, Basic equations F u = 2 hu + 2 RQ h t = 2u Q = u u Hermite F, F formulation F u = 2 hu + 2 RQ F u = 2 (h u + hu + R Q + RQ ) = 2u Q h = 2F u Q + 2u Q t = u u h

Hermite KS with block-steps Predict u, u, h to highest order u pred = u + τu + τ2 2 u + τ3 6 u + τ4 24 u 4 + τ5 20 u u pred = u + τu + τ2 2 u + τ3 6 u 4 + τ4 24 u 5, h pred = h + τh + τ 2 h + τ 6 h + τ 24 h. Force evaluation P = L T uf, P = L T u F + rl T u F. (2) Derivatives for corrector u = (hu + rp), 2 u = 2 (h u + hu + r P + rp ), h = 2(u P), h = 2(u P + u P ). (3)

Fourth-order corrector (Hut, Makino & McMillan 995) u corr = u old + τ 2 (u corr + u old ) τ2 2 (u new u old ), u corr = u old + τ 2 (u new + u old ) τ2 2 (u new u old ), h corr = h old + τ 2 (h new + h old ) τ2 2 (h new h old ). (4) Velocity derivatives for prediction u 4 mid = u new u τ u 5 mid = 2 τ 2 old, ( u new + u old 2 u new u ) old. (5) τ Shifted from midpoint u 4 new = u 4 mid + τ 2 u 5 mid, u 5 new = u 5 mid, h new = h mid + τ h new = h 2 h mid, mid. (6)

Taylor series for physical time t = u u, t = 2(u u ), t = 2(u u + u u ), t 4 = 2(u u ) + 6(u u ), t 5 = 2(u u 4 + 8(u u ) + 6(u u ), t 6 = 2(u u 5 + 0(u u 4 + 20(u u ). (7) Next look-up time t next = t + 6 n= t n τ n. (8) n! Regularized time-step by inversion t + 6 n= t n τ n n! = 0. (9) Newton-Raphson iteration to solve for τ x i+ = x i f(x i) f (x i ).

Initial guess τ = t bs t τ orig, with quantized time-step t bs (= 2 n t orig ). Commensurability condition mod (t sys, t bs ) = 0. Solve equation (9) to obtain τ bs.

0.000 ECC = 0.96 e-05 DT e-06 e-07 0 5e-05 0.000 0.0005 0.0002 0.00025 0.0003 0.00035 0.0004 0.00045 Time

0.03 ECC = 0.96 0.028 0.026 0.024 DTU 0.022 0.02 0.08 0.06 0.04 0 5e-05 0.000 0.0005 0.0002 0.00025 0.0003 0.00035 0.0004 0.00045 Time

subint.f!!!!!! # %&# ()! # () +(, (./0!!!!!!!23456578 9:;58<!=7>!6? 6Α6Β396Χ!!!!!!!!!!!!!() Ε#2!Φ479978ΓΧΗΦ!!!!!! &ΙΙ&)ϑΚ ΛΜ%ϑ!!( Μ +ΚΙΜΝ0!!!!!!! 33!ΟΗ3ΒΗ3>!Β7!:ΠΘ:843!Κ!67Ρ?Β5786!:Β!83Ο! Ρ74; 6Β3ΣΧ!!!!!!(Τ!+)ΛΜ(% ΧΥ Χ/0! ς )!!!!!!!Ι:;3!:!Ρ56Β!7=!:ΡΡ!939 3>6!Π?3!?Σ!Β7!38Π!7=! Ρ74; 6Β3ΣΧ!!!!!!!!!!(.!Ω!.!!!!!!!!!!)Κ!Ω!/!!!!!!!!!!2&!Ξ!(Λ!Ω!. )ΛΜ(%!!!!!!!!!!!!(.!Ω!(.!Ψ!Ξ!!!!!!!!!!!!(Τ!+ /+(.0!Ψ! Λ+(.0ΧΕ Χ Ε& Κ0! ς )!!!!!!!!!!!!!!)Κ!Ω!)Κ!Ψ!.!!!!!!!!!!!!!!Ε( Κ +)Κ0!Ω!(.!!!!!!!!!!!! )2!(Τ!!!!Ξ!!!!! &) ()#!!!!!!!23Β3>9583!Ζ?:8Β5[3Π!Θ:Ρ?3!7=! Ι()Χ!!!!!!!!! Ι()!Ω! Ε& Κ!!!!!!!!!!2&!Γ!Ε!Ω!. )Κ!!!!!!!!!!!!(.!Ω!Ε( Κ +Ε0!!!!!!!!!!!! Ι()!Ω!Ι()+ /+(.0!Ψ! Λ+(.0 Ι()0!!!!Γ!!!!! &) ()#!!!!!!!#ΣΠ:Β3!Β593!:8Π!=7>9!Ρ56Β!7=! 58:>536!Π?3!:Β! Ι()Χ!!!!!!!!!! (Ι!Ω! Ι()!!!!!!!!!!Ε )Υ ς!ω!/!!!!!!!!!!2&!]!ε!ω!. )Κ!!!!!!!!!!!!(.!Ω!Ε( Κ +Ε0!!!!!!!!!!!!(Τ!+ /+(.0!Ψ! Λ+(.0Χ,Χ Ι()0! ς )!!!!!!!!!!!!!!Ε )Υ ς!ω!ε )Υ ς!ψ!.!!!!!!!!!!!!!!) Ν Ε+Ε )Υ ς0!ω!(.!!!!!!!!!!!! )2!(Τ!!!!]!!!!! &) ()#!!!!!!! Η34;!Σ7665 Ρ3!4Η:58!58Β3<>:Β578!78![3>7!Ε )Υ ςχ!!!!!!!!!!(τ!+ε )Υ ςχ,χ/0! ς )!!!!!!!!!!!!Υ&! &!Γ/!!!!!!!!!!! )2!(Τ!!!!!!! Ρ3:8!4?>>38Β!Ρ38<ΒΗ!7=!( Μ Χ!!!!!!!!!!2&!!Ε!Ω!. Ε )Υ ς!!!!!!!!!!!!( Μ +Ε0!Ω!/!!!!!!!!! &) ()#!!!!!!! 3<58!Κ!58Β3<>:Β578!+63Ζ?38Β5:Ρ!7>!Σ:>:ΡΡ3Ρ0Χ!!!!!!!!!!(Τ!+Ε )Υ ςχε Χ)ΕΙΜΝ0! ς )!!!!!!!!!!!!2&!./!Ε(!Ω!. Ε )Υ ς!!!!!!!!!!!!!!(.!ω!) Ν Ε+Ε(0!!!!!!!!!!!!!! ΜΕΕ!Κ () +(. Ε(0!!!./!!!!!!! &) ()#!!!!!!! 3:>4Η!878 [3>7!=Ρ:<6!+( Μ!_!/!93:86!47ΡΡ565780Χ!!!!!!!!!!!!(Τ!+(,Χ,Χ/0! ς )!!!!!!!!!!!!!!2&!.Ξ!Ε(!Ω!. Ε )Υ ς!!!!!!!!!!!!!!!!(τ!+( Μ +Ε(0Χ) Χ/0! ς )!!!!!!!!!!!!!!!!!!(,!Ω!( Μ +Ε(0!!!!!!!!!!!!!!!!!!(./!Ω!) Ν Ε+Ε(0!!!!!!!!!!!!!!!!!!Κ ΛΜ(%!Ω!Κ +(./0!!!!!!!)7Β3!(ΛςΜ!9?6Β! 3!Π3=583Π!=7>!4Η:58Χ!!!!!!!!!!!!!!!!!!(ΛςΜ!Ω!(,!!!!!!!!!!!!!!!!!!Υ&! &!.!!!!!!!!!!!!!!!! )2!(Τ!!!!.Ξ!!!!!!!!! &) ()#!!!!!!!!!!!! )2!(Τ!!!!!!! >3:Β!47ΡΡ56578!3αΣΡ545ΒΡΑ! 3=7>3!Ζ?5ΒΒ58<Χ!!!.!!!!!!!(Τ!+(,ΧΕ Χ/0! ς )

subint.f 2!!!!!!!!!!!!!!(Λ!Ω!Κ +(./0!!!!!!!!!!!!!!Κ ΛΜ(%!Ω!(Λ!!!!!!!!!!!!!!,Λ %(!Ω!%+(Λ0!!!!!!!!!!!!!!(, &ΕΕ!Ω! Ξ!!!!!!!!!!!!!! ΜΕΕ! Ι &2β+,Λ %( Ξ0!!!!!!! 86?>3! Ρ74; 6Β3Σ!Β593!=7>!87>9:Ρ!478Β58?:Β578Χ!!!!!!!!!!!!!! (Ι!Ω! Ε& Κ!!!!!!!)7Β3!>36Β!7=!ΒΗ3! Ρ74;!Ο5ΡΡ! 3!Π783!83αΒ!Β593Χ!!!!!!!!!!!!!!Υ&! &!Γ/!!!!!!!!!!!! )2!(Τ!!!!!!! 79ΣΡ3Β3!:ΡΡ! Ρ74; 6Β3Σ6! 3=7>3!6Σ345:Ρ!Σ>743Π?>3Χ!!!!!!!!!!!!(Τ!+ Ι()ΧΕ Χ Ε& Κ0!Υ&! &!!!!!!!!!!!!! (Ι!Ω! Ε& Κ!!!!!!!!!! Ε!!!!!!!Λ3>=7>9!ΒΗ3!Σ:>:ΡΡ3Ρ!3α34?Β578!Ρ77ΣΧ χδ79σ!σ:>:ρρ3ρ!π7!σ>5θ:β3+ε( (.0!!!!!!!!!!!!2&!Ξ/!Ε(!Ω!. Ε )Υ ς!!!!!!!!!!!!!!(.!ω!) Ν Ε+Ε(0!!!!!!!!!!!!!! ΜΕΕ!Κ () Λ+(. Ε(0!!!Ξ/!!!!!!! &) ()# χδ79σ!38π!σ:>:ρρ3ρ!π7!!!!!!!!!!!!) Λ#!Ω!) Λ#!Ψ!Ε )Υ ς!!!!!!! 3:>4Η!878 [3>7!=Ρ:<6!+( Μ!_!/!=7>!47ΡΡ565780Χ!!!!!!!!!!!!(Τ!+(,Χ,Χ/0! ς )!!!!!!!!!!!!!!2&!Ξ!Ε(!Ω!. Ε )Υ ς!!!!!!!!!!!!!!!!(τ!+( Μ +Ε(0Χ) Χ/0! ς )!!!!!!!!!!!!!!!!!!(,!Ω!( Μ +Ε(0!!!!!!!!!!!!!!!!!!(./!Ω!) Ν Ε+Ε(0!!!!!!!!!!!!!!!!!!(ΛςΜ!Ω!(,!!!!!!!!!!!!!!!!!!Κ ΛΜ(%!Ω!Κ +(./0!!χ!9:Α! 3!833Π3ΠΧ!!!!!!!!!!!!!!!!!!Υ&! &!Ξ]!!!!!!!!!!!!!!!! )2!(Τ!!!!Ξ!!!!!!!!! &) ()#!!!!!!!!!!!! )2!(Τ!!!!!!! >3:Β!47ΡΡ56578!3αΣΡ545ΒΡΑ! 3=7>3!Ζ?5ΒΒ58<Χ!!!Ξ]!!!!!!!(Τ!+(,ΧΕ Χ/0! ς )!!!!!!!!!!!!!!(Λ!Ω!Κ +(./0!!!!!!!!!!!!!!Κ ΛΜ(%!Ω!(Λ!!!!!!!!!!!!!!,Λ %(!Ω!%+(Λ0!!!!!!!!!!!!!!(, &ΕΕ!Ω! Ξ!!!!!!!!!!!!!! ΜΕΕ! Ι &2β+,Λ %( Ξ0!!!!!!!)7Β3!>36Β!7=!ΒΗ3! Ρ74;!Ο5ΡΡ! 3!Π783!83αΒ!Β593Χ!!!!!!!!!!!!!!Υ&! &!Γ/!!!!!!!!!!!! )2!(Τ!!!!!!! Η34;!97>3! Ρ74; 6Β3Σ!Ρ3Θ3Ρ6!?8Ρ366!47ΡΡ56578Χ!!!!!!!!!!!!(Τ!+ Ι()ΧΕ Χ Ε& Κ0!Υ&! &!!!!!!!!!!! )2!(Τ!!!!!!! 7ΣΑ!7>5<58:Ρ! Ρ74;!Β593!:Β!38Π!7=!Κ!Β>3:Β938ΒΧ!!!!!!!!!! (Ι!Ω! Ε& Κ!!!!!! )2!(Τ!!!Γ/!% #%)!!!!!! )2

Massive KS Binary Initial condition e, a, m, m 2, c GR coalescence R coal = 6(m + m 2 ) c 2 PN perturbation PN & 2.5PN Isolated binary e = 0.99, t/t K = 0 5 Time-step τ = η u 2 h /2, N = 2π η u η u = 0., it = 3, E/E = 6 0 3 η u = 0.3, it = 3, E/E = 0 8 η u = 0.3, it = 4, E/E = 3 0

NBODY7 Binary Project Initial condition e, a, m, m 2, c Perturbation PN & 2.5PN in pnpert.f Energy loss Perturbed KS or Peters 964 Accumulated Hermite, previous Ė GR, Ë GR Integration E GR = 2 (Ė 0 + Ė) t + 2 (Ë 0 Ë) t 2 Identity check h < 4 h

0.00 2.5PN PN+2.5PN Peters 0.000 SEMI e-05 e-06 e-07 0 0.2 0.4 0.6 0.8.2.4 Time

Conclusions Performance Significant speed-up Accuracy One or two extra iterations Memory No new variables needed Time-steps Similar to current method Usage Special procedures require attention