Computation and astrophysics of the N-body

problem

Douglas Heggie
School of Mathematics
University of Edinburgh
[email protected]
 2

Course outline (tentative)

1. The N -body problem and its astrophysical settings. Initial

conditions. Units. Introduction to a simple N -body code
(NBODY1). Example - cold collapse. Virialisation and virial
equilibrium.
2. Plummer’s model. Movie of orbital motions. Crossing time
scale. Core collapse - the movie. The relaxation time scale.
The structure of the simplest direct-summation N-body code:
constant, shared, time steps, Euler & Hermite integrators.
3. Quality control & error growth. Complexity. Acceleration of
force computations with software & hardware. GPUs.
Regularisation.
4. Refinements: external effects (tides), internal effects (stellar
evolution, binary stellar evolution, collisions). The NBODY
series of integrators. starlab. Example - cluster in a tidal field
(NBODY6).
 3
References
◮ Stellar Dynamics
1. Galactic Dynamics, 2e James Binney and Scott Tremaine,
Princeton University Press, 2008, 885 pp.
2. Dynamical Evolution of Globular Clusters Lyman J. Spitzer Jr,
Princeton UP, 1988, 196 pp.
3. The Gravitational Million Body Problem, Douglas Heggie, Piet
Hut; Cambridge UP, 2003
◮ N -body codes
1. Gravitational N-body simulations, Sverre J. Aarseth
Cambridge: CUP, 2003, 413pp
2. The Cambridge N-body Lectures, S.J. Aarseth, C.A. Tout, R.A.
Mardling, eds. Springer, LNP760, 2008, 402pp
3. Numerical Methods in Astrophysics: An Introduction, P.
Bodenheimer et al, CRC Press, 2006, 344pp
4. http://www.ast.cam.ac.uk/∼sverre/web/pages/nbody.htm
5. https://github.com/nbodyx/Nbody6ppGPU
6. http://www.ids.ias.edu/∼starlab/ (starlab)
7. http://amusecode.org/ (AMUSE)
 4

Applications of N -Body Schemes in Astrophysics

◮ the Galactic Centre (see http://www.astro.ucla.edu/

∼ghezgroup/gc/pictures/orbitsMovie.shtml )
◮ open clusters (M67 here)
◮ globular clusters (M4 here)
 5

Applications of N -Body Schemes in Astrophysics (continued)

◮ young dense clusters - Magellanic Cloud clusters

◮ galaxy dynamics - the Antennae (requires specialised
software because of large N )
 6

Applications of N -Body Schemes in Astrophysics (continued)

◮ cosmic structure (requires specialised software)

◮ planetary systems (requires specialised software)
 7

The N -body Problem

 7

The N -body Problem

◮ N point masses (this approximation is good while separation

of two stars is much greater than the sum of their radii)
 7

The N -body Problem

◮ N point masses (this approximation is good while separation

of two stars is much greater than the sum of their radii)
◮ classical gravitation and equations of motion (this
approximation is good except close to horizon of black hole, or
for close binaries emitting gravitational waves)
 8

Equations of motion

r
3 r
2

r
1

Force on 1 due to 2 is in the direction of r2 − r1 , i.e. the unit vector

r2 − r1 Gm1 m2
, and has magnitude .
|r2 − r1 | |r2 − r1 |2
 9

Equations of motion (continued)

Gm1 m2 (r2 − r1 )
Therefore force on 1 due to 2 is . Therefore total
|r2 − r1 |3
force on 1 is
j =3
X Gm1 mj (rj − r1 )
m1 r̈1 =
j =1,j ,1
|rj − r1 |3

In the N -body problem the equation of motion for body i is

N
X ri − rj
r̈i = −G mj
j =1,,i
|ri − rj |3
 10

Softening

The equation of motion is

N
X ri − rj
r̈i = −G mj
j =1,,i
|ri − rj |3

There is a singularity if |ri − rj | = 0. To avoid this, the denominator

is sometimes replaced by (|ri − rj |2 + ε2 )3/2 , where ε is a small
constant, the softening parameter. This approximation may be
justifiable if close encounters between particles are unimportant –
for example, in galaxy dynamics (with scaling by N ). Not
necessarily good for modelling star clusters.
 11

Initial Conditions

These are 3N second-order ordinary differential equations, and

hence require 6N initial conditions. Usually one uses the three
cartesian components of position ri and the three components of
velocity ṙi of the N particles.
Example: cold collapse
The initial velocities are zero, and the initial positions are chosen
randomly, with (in this example) a uniform spatial distribution in a
sphere of some radius a .
 12

How to Simulate Cold Collapse

1. Go to the web page

http://www.ast.cam.ac.uk/∼sverre/web/pages/nbody.htm
2. Download nbody1.tar.Z
3. Uncompress it (“gunzip nbody1.tar.Z”)
4. Untar it (“tar xvf nbody1.tar”)
5. Go to the source subdirectory (“cd Real8”)
6. Edit the Makefile as follows
◮ line 2: replace “f77” by your fortran compiler (e.g. “gfortran”)
◮ after line 2, add the line “FC = gfortran” (or your compiler)
7. Make the code (“make”)
8. Go to the test subdirectory (“cd ../test”)
9. Make a copy of the file intest and edit to the following:
1 1.0/25 1 200 1/0.01 0.1 10.0 2.0E-05 0.0/1 0 0 0 0 1 0 0 0 0
0 0 0 0 0/2.0 1.0 1.0/0.0 0.0 0.0 1.0 1.0
 13

How to Simulate Cold Collapse (continued)

(It is best to give the file a new name, such as cc.in, and keep the
old file intest. The changes are in boldface, and lines are here
delimited by “/” to save space, but you should keep the format of
intest.)
1. Run the code (“../Real8/nbody1 < cc.in”)
2. Watch the numbers fly past until the job completes or you
have to kill it (“Ctrl-c”)1
3. To save the output, rerun it with redirection of output
(“../Real8/nbody1 < cc.in > cc.out”)

1
If the run fails or fails to complete in a few minutes, kill it. Then edit cc.in to
change the third number in line 2 of cc.in to some other positive integer, and try
again.
 14
Cold collapse: the output - p1
(Corresponds to slightly different input file.)
 15
Cold collapse: the output - p2
 16

Cold collapse: the Movie

Initial conditions:
◮ All velocities are zero
◮ Particles are distributed uniformly in a sphere
 17

Dynamic Equilibrium
 17

Dynamic Equilibrium

◮ The system “quickly” reaches a “steady state”

 17

Dynamic Equilibrium

◮ The system “quickly” reaches a “steady state”

◮ The steady state is in “dynamic equilibrium”, i.e. there is no
overal expansion or contraction of the system, or other bulk
motion, even though all particles are in motion.
◮ A dynamic equilibrium is also a state of “virial equilbrium”,
which can be analysed in terms of the Virial Theorem.
 18
The Virial Theorem
Define the total kinetic energy T , the total potential energy V , the
total energy E and the total “moment of inertia” I by
N
1X
T = mi vi2 (Kinetic Energy)
2
i =1
N N
G X X mj mi
V = − (Potential Energy)
2 |ri − rj |
i =1 j =1,,i
E = T + V (Total Energy)
XN
I = mi |ri |2 (“Moment of Inertia”).
i =1
Then from the equations of motion we deduce
Ï = 4T + 2V (Virial Theorem)
E = constant (Energy Conservation)
Proofs: Binney and Tremaine, Sec 7.2.1; this lecture, last pages
 19

Virial Equilibrium

In dynamic equilibrium, I is approximately constant, and so Ï ≃ 0.

From the Virial Theorem Ï = 4T + 2V we deduce

2T + V ≃ 0.

Using total energy E = T + V we deduce

T +E ≃0⇒T ≃ −E
and similarly V ≃ 2E .
 20
Mass, length and time scales
 20
Mass, length and time scales
◮ Total mass
N
X
M= mi
i =1
 20
Mass, length and time scales
◮ Total mass
N
X
M= mi
i =1

◮ Characterise system size by “virial radius” R defined by

GM 2
V =− , where M is total mass
2R
 20
Mass, length and time scales
◮ Total mass
N
X
M= mi
i =1

◮ Characterise system size by “virial radius” R defined by

GM 2
V =− , where M is total mass
2R

◮ Characterise speeds by (mass weighted) mean square speed

2T
v2 =
M
 20
Mass, length and time scales
◮ Total mass
N
X
M= mi
i =1

◮ Characterise system size by “virial radius” R defined by

GM 2
V =− , where M is total mass
2R

◮ Characterise speeds by (mass weighted) mean square speed

2T
v2 =
M

◮ Define time scale

2R
tcr = (“Crossing time”)
v
 21

Other useful expressions and definitions

GM
In virial equilibrium V = −2T , and so v 2 = .
2R
T
The virial ratio is defined to be Q = , and is 0.5 in virial
|V |
equilibrium.

Another measure of the size of the system is the half-mass radius,

the radius of a sphere containing the innermost half of the mass,
measured with respect to the “centre” of the system.
More generally, a “lagrangian radius” is the radius of the
sphere containing a given fixed fraction of the mass.

Observational astronomers may prefer the half-light radius, the

radius of a disk containing the innermost half of the light, since it
can be “easily” measured.
 22

Hénon Units (aka N -body units)

This is a conventional system of units in which

G = 1
M = 1
R = 1

These are often used in N -body simulations.

Example
GM
We have v 2 = .
2R
Suppose a star cluster has M = 105 M⊙ , R = 5pc. To r convert a
GM
velocity from the N -body code to km/s, multiply by , where G
R
is expressed in the same units (i.e. km/s, M⊙ , pc), i.e. G ≃ 0.0043.
 23

Hénon Units (continued)

We have

G = 1
M = 1
R = 1

In these units:
GM 1
◮ The characteristic speed v 2 = =
2R 2
2R √
◮ The crossing time tcr = =2 2
v
1 1
◮ The total energy E = − Mv 2 = −
2 4
 24

Significance of the crossing time

 24

Significance of the crossing time

◮ Time scale of cold collapse

 24

Significance of the crossing time

◮ Time scale of cold collapse

◮ Time scale of approach to virial equilibrium
 24

Significance of the crossing time

◮ Time scale of cold collapse

◮ Time scale of approach to virial equilibrium
◮ Time scale of orbital motions in virial equilibrium (Lecture 2)
 25

The code NBODY1: input

The meaning of each input parameter is defined in the file define.f

in the source subdirectory. For illustration we use the file for the
cold collapse simulation:
1 0.5
25 1 200 1
0.01 0.1 10.0 2.0E-05 0.0
100001000000000
2.0 1.0 1.0
0.0 0.0 0.0 1.0 1.0
 25

The code NBODY1: input

The meaning of each input parameter is defined in the file define.f

in the source subdirectory. For illustration we use the file for the
cold collapse simulation:
1 0.5
25 1 200 1
0.01 0.1 10.0 2.0E-05 0.0
100001000000000
2.0 1.0 1.0
0.0 0.0 0.0 1.0 1.0
Line 1: 1 0.5
1. KSTART Control index (1: new run; >1: restart; 3: new
params).
Comment: runs can be restarted following a crash or other
accident
2. TCOMP Maximum computing time in minutes
 26

The code NBODY1: input line 2

Line 2: 25 1 200 1
1. N Total particle number.
2. NFIX Output frequency of data save or binaries (option 3 & 6
– see below).
Comment: There is a basic output interval. You need not
output the data or information on binaries at all such output
times.
3. NRAND Random number sequence skip.
Comment: initialises the random number generator
4. NRUN Run identification index.
 27

The code NBODY1: input line 3

Line 3: 0.01 0.1 10.0 2.0E-05 0.0

1. ETA Time-step parameter for total force polynomial.
Comment: this controls the accuracy of the numerical solution
of the equations of motion (see later)
2. DELTAT Output time interval in units of the crossing
√ time.
Comment: In N -body units the crossing time is 2 2 ≃ 3.
3. TCRIT Termination time in units of the crossing time.
4. QE Energy tolerance (stop if DE/E > 5*QE & KZ(2) ≤ 1).
Comment: the program stops if the relative change in energy
exceeds the stated value and the appropriate option is chosen
(see below)
5. EPS Softening parameter (square saved in EPS2).
 28

The code NBODY1: input line 4

Line 4: 1 0 1 0 0 1 0 0 0 0 0 0 0 0 0
Note: This is the line of options. In general, a zero value indicates
that the option is not selected
1. 1 COMMON save on unit 1 if TCOMP > CPU or if TIME >
TCRIT.
Comment: COMMON is a block of variables which are
enough to restart the run, if desired.
2. 2 COMMON save on unit 2 at output (=1); restart if DE/E >
5*QE (=2).
Comment: if this value is 1, the COMMON variables are saved
every output time; if 2, and the accuracy of the run has
deteriorated, the code attempts to repeat the most recent part
of the run with higher accuracy (and carry on)
 29
The code NBODY1: input line 4 (continued)
Line 4: 1 0 1 0 0 1 0 0 0 0 0 0 0 0 0
3 Basic data written to unit 3 at output time (frequency NFIX).
Comment: if this is positive, complete information on the
particles is saved in a binary file OUT3 at each output time.
4 Initial conditions on unit 4 (=1: output; =2: input).
Comment: if this value is 1 the code dumps the initial
conditions on a file called fort.4; if 2, the initial conditions are
read in from this file (usually called fort.4)
5 Initial conditions (=0: uniform & isotropic; =1: Plummer).
Comment: we used 0. The Plummer model is explained in
Lecture 2
6 Output of significant binaries.
7 Output of movie frames on unit 7.
Comment: you can experiment with this!
8 Generation of two subsystems (merger experiment).
 30
The code NBODY1: input line 4 (continued)
Line 4: 1 0 1 0 0 1 0 0 0 0 0 0 0 0 0
9 Individual bodies printed at output time (MIN(5**KZ9,N)).
Comment: the value is referred to as KZ9, and can be used to
control how many particles are listed
10 No scaling of initial conditions.
Comment: i.e. use units of the input file fort.4, and do not
scale to N -body units
11 Modification of ETA by tolerance QE.
Comment: ETA controls the accuracy of the integration (see
above), and this option lets the code attempt to adjust this by
monitoring the relative change in energy.
12 Initial parameters for binary orbit.
Comment: if non-zero, the code reads the semi-major axis
and eccentricity of a binary formed by the first two particles
13 Escaper removal (R > 2*RTIDE; RTIDE = 10*RSCALE).
Comment: here RSCALE is the virial radius.
 31

The code NBODY1: input line 4 (continued)

Line 4: 1 0 1 0 0 1 0 0 0 0 0 0 0 0 0
14 Adjustment of coordinates & velocities to c.m. condition.
Comment: uses a barycentric coordinate system
15 Ignored
 32

The code NBODY1: input line 5

Line 5: 2.0 1.0 1.0

1. ALPHAS Power-law index for initial mass function (routine
DATA).
Comment: the initial mass function of the stars is a power law,
i.e. f (m) ∝ m−α in the range BODYN < m < BODY 1 (see
below)
2. BODY1 Maximum particle mass before scaling.
3. BODYN Minimum particle mass before scaling.
Comment: choosing BODY1 = BODYN gives equal masses.
The code interprets these as being in solar masses, and then
scales to N -body units internally.
 33

The code NBODY1: input line 6

Line 6: 0.0 0.0 0.0 1.0 1.0

1. Q Virial ratio (routine SCALE; Q = 0.5 for equilibrium).
2. VXROT XY-velocity scaling factor (> 0 for solid-body rotation).
Comment: adds rotation about the z -axis
3. VZROT Z-velocity scaling factor (not used if VXROT = 0).
Comment: if less than 1 the z -components of velocity are
reduced and the systems tends to flatten
4. RBAR Virial radius in pc (for scaling to physical units).
Comment: specifies the unit of length in parsecs
5. ZMBAR Mean mass in solar units.
 34
Cold collapse: the output - p1
The first four lines repeat most of the input parameters
 35

Cold collapse: the output (continued)

line 5: SCALING: SX = 0.44 E = -5.72E-01 M(1) = 4.00E-02 M(N)

= 4.00E-02 <M> = 4.00E-02
Comments:
1. E is the energy before scaling to N -body units
2. M(1),M(N) are the same as BODY1, BODYN (see above), but
now in N -body units.
3. <M> is the mean mass (N -body units)
line 6: SCALING PARAMETERS: R* = 1.00E+00 M* = 2.50E+01
V* = 3.28E-01 T* = 2.99E+00
Comment: scaling factors for length (to parsecs), mass (to solar
masses), velocities (to km/s), time (Myr, approximately)
 36
Cold collapse: the output (continued)
The following gives an example of the output produced at each
output time:
T = 0.9 Q = 0.05 STEPS = 154 DE = 0.000000 E = -0.250000 TC =
0.3
<R> = 1.89 RCM = 0.0000 VCM = 0.0000 AZ = 0.00000 T6 = 2
NRUN = 1
BINARY 1 10 0.040 0.040 -0.1 0.3652 1.3 0.5156 1.54 0.970 0
BINARY 6 15 0.040 0.040 -0.2 0.2525 2.2 0.3785 1.47 0.973 0
BINARY 8 9 0.040 0.040 -0.2 0.2602 2.1 0.4063 1.94 0.989 0
Line 1
1. Time in N -body units
2. Virial ratio
3. Number of integration steps taken (see below)
4. Change in energy (N -body units)
5. Total energy
6. Time in units of the crossing time
 37

Cold collapse: the output (continued)

T = 0.9 Q = 0.05 STEPS = 154 DE = 0.000000 E = -0.250000 TC =

0.3
<R> = 1.89 RCM = 0.0000 VCM = 0.0000 AZ = 0.00000 T6 = 2
NRUN = 1
BINARY 1 10 0.040 0.040 -0.1 0.3652 1.3 0.5156 1.54 0.970 0
BINARY 6 15 0.040 0.040 -0.2 0.2525 2.2 0.3785 1.47 0.973 0
BINARY 8 9 0.040 0.040 -0.2 0.2602 2.1 0.4063 1.94 0.989 0
Line 2
1. Virial radius
2. Distance of centre of mass from origin
3. Velocity of centre of mass
4. Angular momentum about the z -axis
5. Time in Myr (approximately)
6. Run number (as in input)
 38
Cold collapse: the output (continued)
T = 0.9 Q = 0.05 STEPS = 154 DE = 0.000000 E = -0.250000 TC =
0.3
<R> = 1.89 RCM = 0.0000 VCM = 0.0000 AZ = 0.00000 T6 = 2
NRUN = 1
BINARY 1 10 0.040 0.040 -0.1 0.365 1.3 0.5156 1.54 0.970 0
Line 3 (one line for each binary)
1. (2 numbers) Names of the two components
2. (2 numbers) Masses of the components (N -body units)
3. Internal energy of the binary (per unit reduced mass); this
excludes the energy of the centre of mass of the binary
4. Semi-major axis of the binary
5. Angular velocity (mean motion) of the binary
6. Separation of the components
7. Distance of the binary from the “centre” of the system
8. Eccentricity of the binary
9. Number of binary periods since t = 0
 39

Extracting useful information with awk

The typical output:

T = 0.9 Q = 0.05 STEPS = 154 DE = 0.000000 E = -0.250000 TC =
0.3
<R> = 1.89 RCM = 0.0000 VCM = 0.0000 AZ = 0.00000 T6 = 2
NRUN = 1
BINARY 1 10 0.040 0.040 -0.1 0.3652 1.3 0.5156 1.54 0.970 0
(Note that there are only three lines in the output file, but these
may be wrapped in the pdf you are reading.)
We can extract data with the linux command awk.
First, create a text file called (say) q.awk, containing the following
line:
{if ($1==”T”) print $3,$6}
This means that we pick out only those lines of output beginning
with the string “T”, and from such lines print out the third and sixth
fields (fields being separated by spaces), i.e. the values of T , Q .
 40

Extracting useful information with awk (continue)

First rerun the code but collect the output in a text file cc.out.
(Again the run may have to be killed after a short time). The
command is
../Real8/nbody1 < cc.in > cc.out
Then awk is run with
awk -f q.awk cc.out > cc.q
which means that awk is to pick up its instructions from the file
q.awk, act on the file cc.out, and direct its output to a new file cc.q.
Here is an example of cc.q (a table of T , Q values)
0.0 0.00
0.3 0.01
0.6 0.02
0.9 0.05
1.1 0.11
1.4 0.23
 41

Plotting useful information with gnuplot

Start gnuplot (“gnuplot”), and then plot the file just produced:
gnuplot> plot ’cc.q’ w l
Note: “w l” means “with lines”; otherwise you just get points.
This produces a window with the following graph:
0.8
’cc.q’

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 5 10 15 20 25 30
 42

Plotting useful information with gnuplot (continued)

The plot can be improved and made permanent with a sequence

of commands like this:

gnuplot> set xlab ’Time

gnuplot> set ylab ’Virial ratio
gnuplot> set ter post
gnuplot> set out ’q.ps
gnuplot> replot

Then you can print or view the postscript file q.ps

 43
Proof of the Virial Theorem
“Moment of inertia”
N
X
I= mi (ri .ri )2 .
i =1
Hence
N
X
İ = 2mi ri .vi
i =1
N
X
Ï = 2mi (vi .vi + ri .v̇i ).
i =1
The first term gives 4 times the kinetic energy, i.e. 4T . From the
equations of motion (above), the second term gives
N N N N
X X ri − rj X X rj − ri
−G 2mi ri . mj = −G 2mj rj . mi (1)
i =1 j =1,,i
|ri − rj |3 j =1 i =1,,j
|rj − ri |3

(swapping i and j ).
 44

Proof of the Virial Theorem (Contd)

Therefore the second term gives half their sum, i.e.

N X
N X X mi mj
X ri − rj
−G mi mj (ri − rj ). = −G ,
i =1 j =1,,i
|ri − rj |3 |ri − rj |

the sum being over all distinct i , j between 1 and N . This is twice
the potential energy, because each pair is counted twice.
Hence Ï = 4T + 2V .
 45

Exercises from Lecture 1

1. Carry out the procedure of downloading NBODY1, running the

cold collapse simulation, and plotting the time-dependence of
the virial ratio.