Modelling the Three Body Problem in Classical Mechanics using Python

As it turns out, something much worse.

We are fortunate to be living in a solar system with only one major star since that makes the orbit of the star (our sun) predictable.

Increase the number of stars to two and the system still remains stable.

It has, what we call, an analytical solution — that is, we can solve the equations describing it and get a function that gives the time evolution of the system from 1 second to a million years accurately.

However, when you add a third body, something extraordinary happens.

The system becomes chaotic and highly unpredictable.

It has no analytical solution (except for a few special cases) and its equations can only be solved numerically on a computer.

They can turn abruptly from stable to unstable and vice versa.

The Trisolarans living in such a chaotic world developed the ability to “dehydrate” themselves and hibernate during the “Chaotic Eras” and awake and live peacefully during the “Stable Eras”.

The intriguing visualization of the star system in the book inspired me to read up on the n-body class of problems in gravitation and the numerical methods used to solve them.

This article touches upon a few core concepts of gravitation required to understand the problem and the numerical methods required to solve the equations describing the systems.

Through this article, you will get to read about the implementation of the following tools and concepts:Solving differential equations in Python using the odeint function in the Scipy module.

Non-dimensionalizing an equationMaking 3D plots in Matplotlib2.

Basics of Gravitation2.

1 Newton’s Law of GravitationNewton’s Law of Gravitation says that any two point masses have an attractive force between them (called the gravitational force), the magnitude of which is directly proportional to the product of their masses and inversely proportional to the square of the distance between them.

The equation below represents this law in vector form.

Here, G is the universal gravitational constant, m₁ and m₂ are the masses of the two objects and r is the distance between them.

The unit vector points away from the body m₁ towards m₂ and the force too acts in the same direction.


2 Equation of MotionAccording to Newton’s second law of motion, the net force on an object produces a net change in momentum of the object — in simple terms, force is mass times acceleration.

So, applying the above equation to the body having mass m₁, we get the following differential equation of motion for the body.

Note here that we have unraveled the unit vector as the vector r divided by its magnitude |r|, thus increasing the power of the r term in the denominator to 3.

Now, we have a second-order differential equation that describes the interaction between two bodies due to gravity.

To simplify its solution, we can break it down into two first order differential equations.

The acceleration of an object is the change in velocity of the object with time so the second order differential of position can be replaced with a first order differential of velocity.

Similarly, the velocity can be expressed as a first order differential of the position.

The index i is for the body whose position and velocity is to be calculated whereas the index j is for the other body which is interacting with body i.

Thus, for a two-body system, we will be solving two sets of these two equations.


3 Centre of MassAnother useful concept to keep in mind is the centre of mass of a system.

The centre of mass is a point where the the sum of the all the mass moments of the system is zero — in simple terms, you can imagine it as the point where the whole mass of the system is balanced.

There is a simple formula to find the centre of mass of a system and its velocity.

It involves taking mass-weighted averages of the position and velocity vectors.

Before modelling a three-body system, let us first model a two-body system, observe its behaviour and then extend the code to work for three bodies.


The Two-Body Model3.

1 Alpha Centauri Star SystemA famous real-world example of a two-body system is perhaps the Alpha Centauri star system.

It contains three stars — Alpha Centauri A, Alpha Centauri B and Alpha Centauri C (commonly called Proxima Centauri).

However, since Proxima Centauri has a mass that is negligibly small as compared to the other two stars, Alpha Centauri is considered to be a binary star system.

An important point to note here is that the bodies considered in an n-body system all have similar masses.

So, Sun-Earth-Moon is not a three body system since they do not have equivalent masses and the Earth and Moon do not significantly influence the Sun’s path.

The Alpha Centauri binary star system captured from the Paranal Observatory in Chile by John Colosimo3.

2 Non-DimensionalizationBefore we start solving these equations, we have to first non-dimensionalize them.

What does that mean?.We convert all the quantities in the equation (like position, velocity, mass and so on) that have dimensions (like m, m/s, kg respectively) to non-dimensional quantities that have magnitudes close to unity.

The reasons for doing so are:In the differential equation, different terms may have different orders of magnitude (from 0.

1 to 10³⁰).

Such a huge disparity may lead to slow convergence of numerical methods.

If the magnitude of all terms becomes close to unity, all the calculations will become computationally cheaper than if the magnitudes were asymmetrically large or small.

You will get a reference point with respect to the scale.

For example, if I give you a quantity say 4×10³⁰ kg, you may not be able to figure out whether it is small or large on the cosmic scale.

However, if I say 2 times the mass of the sun, you will easily be able to grasp the significance of that quantity.

To non-dimensionalize the equations, divide each quantity by a fixed reference quantity.

For example, divide the mass terms by the mass of the sun, position (or distance) terms with the distance between the two stars in the Alpha Centauri system, time term with the orbital period of Alpha Centauri and velocity term with the relative velocity of the earth around the sun.

When you divide each term by the reference quantity, you will also need to multiply it to avoid changing the equation.

All these terms along with G can be clubbed into a constant, say K₁ for equation 1 and K₂ for equation 2.

Thus, the non-dimensionalized equations are as follows:The bar over the terms indicates that the terms are non-dimensional.

So these are the final equations that we’ll be using in our simulation.


3 CodeLet us begin by importing all the required modules for the simulation.

#Import scipyimport scipy as sci#Import matplotlib and associated modules for 3D and animationsimport matplotlib.

pyplot as pltfrom mpl_toolkits.

mplot3d import Axes3Dfrom matplotlib import animationNext, let us define the constants and the reference quantities that are used to non-dimensionalize the equations as well as the net constants K₁ and K₂.

#Define universal gravitation constantG=6.

67408e-11 #N-m2/kg2#Reference quantitiesm_nd=1.

989e+30 #kg #mass of the sunr_nd=5.

326e+12 #m #distance between stars in Alpha Centauriv_nd=30000 #m/s #relative velocity of earth around the sunt_nd=79.


51 #s #orbital period of Alpha Centauri#Net constantsK1=G*t_nd*m_nd/(r_nd**2*v_nd)K2=v_nd*t_nd/r_ndIt is time to define some parameters that define the two stars that we are trying to simulate — their masses, initial positions and initial velocities.

Note that these parameters are non-dimensional, so the mass of Alpha Centauri A is defined as 1.

1 (indicating 1.

1 times the mass of the sun, which is our reference quantity).

The velocities are arbitrarily defined in a way such that none of the bodies escape each others’ gravitational pull.

#Define massesm1=1.

1 #Alpha Centauri Am2=0.

907 #Alpha Centauri B#Define initial position vectorsr1=[-0.

5,0,0] #mr2=[0.

5,0,0] #m#Convert pos vectors to arraysr1=sci.


array(r2,dtype="float64")#Find Centre of Massr_com=(m1*r1+m2*r2)/(m1+m2)#Define initial velocitiesv1=[0.


01,0] #m/sv2=[-0.


1] #m/s#Convert velocity vectors to arraysv1=sci.


array(v2,dtype="float64")#Find velocity of COMv_com=(m1*v1+m2*v2)/(m1+m2)We have now defined most of the major quantities required for our simulation.

We can now move on to preparing the odeint solver in scipy to solve our system of equations.

To solve any ODE, you require the equations (of course!), a set of initial conditions and the time span for which the equations are to be solved.

The odeint solver also requires these primary three things.

The equations are defined through the means of a function.

The function takes in an array containing all the dependent variables (here position and velocity) and an array containing all the independent variables (here time) in that order.

It returns the values of all the differentials in an array.

#A function defining the equations of motion def TwoBodyEquations(w,t,G,m1,m2): r1=w[:3] r2=w[3:6] v1=w[6:9] v2=w[9:12] r=sci.


norm(r2-r1) #Calculate magnitude or norm of vector dv1bydt=K1*m2*(r2-r1)/r**3 dv2bydt=K1*m1*(r1-r2)/r**3 dr1bydt=K2*v1 dr2bydt=K2*v2 r_derivs=sci.

concatenate((dr1bydt,dr2bydt)) derivs=sci.

concatenate((r_derivs,dv1bydt,dv2bydt)) return derivsFrom the code snippet, you may be able to identify the differential equations quite easily.

What are the other odds and ends?.Remember that we are solving the equation for 3 dimensions, so each position and velocity vector will have 3 components.

Now, if you consider the two vector differential equations given in the previous section, they need to be solved for all the 3 components of the vectors.

So, for a single body you need to solve 6 scalar differential equations.

For two bodies, you got it, 12 scalar differential equations.

So we make an array w having size 12 that stores the position and velocity coordinates of the two bodies in question.

At the end of the function, we concatenate or join all the different derivatives and return an array derivs of size 12.

The difficult job is now done!.All that remains is to input the function, initial conditions and time span into the odeint function.

#Package initial parametersinit_params=sci.

array([r1,r2,v1,v2]) #create array of initial paramsinit_params=init_params.

flatten() #flatten array to make it 1Dtime_span=sci.

linspace(0,8,500) #8 orbital periods and 500 points#Run the ODE solverimport scipy.



odeint(TwoBodyEquations,init_params,time_span,args=(G,m1,m2))The variable two_body_sol contains all the information about the two-body system including the position vectors and velocity vectors.

To create our plots and animations, we only need the position vectors so let us extract them to two different variables.

r1_sol=two_body_sol[:,:3]r2_sol=two_body_sol[:,3:6]It’s time to plot!.This is where we will utilize the 3D plotting capabilities of Matplotlib.

#Create figurefig=plt.

figure(figsize=(15,15))#Create 3D axesax=fig.

add_subplot(111,projection="3d")#Plot the orbitsax.


plot(r2_sol[:,0],r2_sol[:,1],r2_sol[:,2],color="tab:red")#Plot the final positions of the starsax.

scatter(r1_sol[-1,0],r1_sol[-1,1],r1_sol[-1,2],color="darkblue",marker="o",s=100,label="Alpha Centauri A")ax.

scatter(r2_sol[-1,0],r2_sol[-1,1],r2_sol[-1,2],color="tab:red",marker="o",s=100,label="Alpha Centauri B")#Add a few more bells and whistlesax.




set_title("Visualization of orbits of stars in a two-body system.",fontsize=14)ax.

legend(loc="upper left",fontsize=14)The final plot makes it pretty clear that the orbits follow a predictable pattern, as is expected of a solution of a two-body problem.

A Matplotlib plot showing the time evolution of the orbits of the two starsHere’s an animation that shows the step-by-step evolution of the orbits.

An animation made in Matplotlib that shows the time evolution step-by-step (code not given in article)There is one more visualization that we can make, and that is from the frame of reference of the centre of mass.

The above visualization is from some arbitrary stationary point in space but if we observe the motion of the two bodies from the centre of mass of the system, we will see an even more visible pattern.

So, first let us find the position of the centre of mass at every time step and then subtract that vector from the position vectors of the two bodies to find their locations relative to the centre of mass.

#Find location of COMrcom_sol=(m1*r1_sol+m2*r2_sol)/(m1+m2)#Find location of Alpha Centauri A w.


t COMr1com_sol=r1_sol-rcom_sol#Find location of Alpha Centauri B w.


t COMr2com_sol=r2_sol-rcom_solFinally, we can use the code utilized for plotting the previous visual with a change in variables to plot the following visual.

A Matplotlib plot showing the time evolution of the orbits of the two stars as seen from the COMIf you were to sit at the COM and observe the two bodies, you would see the above orbits.

It is not clear from this simulation since the timescale is very small but even these orbits keep rotating ever so slightly.

It is quite clear now that they follow very predictable paths and that you can use a function — perhaps the equation of an ellipsoid — to describe their motion in space, as expected of a two-body system.


The Three-Body Model4.

1 CodeNow to extend our previous code to a three-body system, we have to make a few additions to the parameters — add mass, position and velocity vectors of the third body.

Let us consider the Third Star to have a mass equal to that of the sun.

#Mass of the Third Starm3=1.

0 #Third Star#Position of the Third Starr3=[0,1,0] #mr3=sci.

array(r3,dtype="float64")#Velocity of the Third Starv3=[0,-0.


array(v3,dtype="float64")We need to update the formulas of centre of mass and velocity of the centre of mass in the code.

#Update COM formular_com=(m1*r1+m2*r2+m3*r3)/(m1+m2+m3)#Update velocity of COM formulav_com=(m1*v1+m2*v2+m3*v3)/(m1+m2+m3)For a three-body system, we will need to modify the equations of motion to include the extra gravitational force exerted by the presence of another body.

Thus, we need to add a force term on the RHS for every other body exerting a force on the body in question.

In the case of a three-body system, one body will be affected by the forces exerted by the two remaining bodies and hence two force terms will appear on the RHS.

It can be represented mathematically as.

To reflect these changes in the code, we’ll need to create a new function to supply to the odeint solver.

def ThreeBodyEquations(w,t,G,m1,m2,m3): r1=w[:3] r2=w[3:6] r3=w[6:9] v1=w[9:12] v2=w[12:15] v3=w[15:18] r12=sci.


norm(r2-r1) r13=sci.


norm(r3-r1) r23=sci.


norm(r3-r2) dv1bydt=K1*m2*(r2-r1)/r12**3+K1*m3*(r3-r1)/r13**3 dv2bydt=K1*m1*(r1-r2)/r12**3+K1*m3*(r3-r2)/r23**3 dv3bydt=K1*m1*(r1-r3)/r13**3+K1*m2*(r2-r3)/r23**3 dr1bydt=K2*v1 dr2bydt=K2*v2 dr3bydt=K2*v3 r12_derivs=sci.

concatenate((dr1bydt,dr2bydt)) r_derivs=sci.

concatenate((r12_derivs,dr3bydt)) v12_derivs=sci.

concatenate((dv1bydt,dv2bydt)) v_derivs=sci.

concatenate((v12_derivs,dv3bydt)) derivs=sci.

concatenate((r_derivs,v_derivs)) return derivsFinally, we need to call the odeint function and supply the above function as well as initial conditions to it.

#Package initial parametersinit_params=sci.

array([r1,r2,r3,v1,v2,v3]) #Initial parametersinit_params=init_params.

flatten() #Flatten to make 1D arraytime_span=sci.

linspace(0,20,500) #20 orbital periods and 500 points#Run the ODE solverimport scipy.



odeint(ThreeBodyEquations,init_params,time_span,args=(G,m1,m2,m3))As with the two-body simulation, we need to extract the position coordinates for all the three bodies for plotting.

r1_sol=three_body_sol[:,:3]r2_sol=three_body_sol[:,3:6]r3_sol=three_body_sol[:,6:9]The final plot can be made using the code given in the previous section with a few changes.

The orbits have no predictable pattern, as you can observe from the mess of a plot below.

A Matplotlib plot showing the time evolution of the orbits of the three starsAn animation will make it easier to make sense of the messy plot.

An animation made in Matplotlib that shows the time evolution step-by-step (code not given in article)Here’s a solution to another initial configuration in which you can observe that the solution seems to be stable initially but then abruptly becomes unstable.

An animation made in Matplotlib that shows the time evolution step-by-step (code not given in article)You can try playing around with the initial conditions to see different kinds of solutions.

In recent years, many interesting solutions to the three-body problem have been discovered due to the availability of greater computational power, some of which appear to be periodic — like the figure-8 solution in which all the three bodies move in a planar figure-8 path.

Some references for further reading:A small essay on the mathematical description of the three-body problem.

A thesis on the study of planar restricted three-body problem solutions (includes figure-8 and Hill’s solutions).

I haven’t included the code for the animations in this article.

If you want to know more, you can email me or contact me on Twitter.


. More details

Leave a Reply