Lotka-Volterra system.

This kind of stuff really needs a simple application to test the influence of certain factors. For example,

There are tools schools can buy, but a) they suck and b) they are very expensive.

Introducing: "Predator and Prey" by Carsten Niehaus (tm)

As you can see the basic stuff all works. When happens when I start with just on predator but 20 prey but let the predators be more efficient hunters? Look here:

The application is written in PyQt4 using matplotlib. I would really like to express one thing: Python + Qt4

Are you interested in the code or even in helping me writing a cool PP-Simulation? Do you think this should go into KDE EDU? I am hosting everything in GitHub: PP in GitHub.

As you may know I am a biology teacher. Next term I will teach Ecology. One part of that is the relationship between predators and prey (wolves and rabbits, for example). There are quite some mathematical model for these relationships, in school we mostly teach a simplified version of the
This kind of stuff really needs a simple application to test the influence of certain factors. For example,

- what happens if after two years we kill all but 5 predators?
- what happens if after two years we kill all but 5 prey?
- what happens if after 1 year, for whatever reasons, the rabbits become more fertile?

There are tools schools can buy, but a) they suck and b) they are very expensive.

Introducing: "Predator and Prey" by Carsten Niehaus (tm)

As you can see the basic stuff all works. When happens when I start with just on predator but 20 prey but let the predators be more efficient hunters? Look here:

The application is written in PyQt4 using matplotlib. I would really like to express one thing: Python + Qt4

*rocks*. Seriously, the whole application has only 306 lines including comments!Are you interested in the code or even in helping me writing a cool PP-Simulation? Do you think this should go into KDE EDU? I am hosting everything in GitHub: PP in GitHub.

## Comments

cniehausQuickpost this image to Myspace, Digg, Facebook, and others!

(Anonymous)After having a quick look at the sources, I can even explain why:

The integrator you use (the method to integrate the Differential equation over time) is not appropriate for the problem.

The method you are using is known as Euler Method (http://en.wikipedia.org/wiki/Euler_meth

I have an implementation of several integrators in python but they use the numpy package which you probably dont want as dependency. But if you want I can rewrite one of them for your purpose.

I will send you an email with an patch if you are interested.

cniehauscniehaus kde org

cniehaushttp://www01.wolframalpha.com/input/

(Anonymous)The difficult part is to calculate the actual trajectories (the long-term behaviour) from these. As the computer simulates the system using finite time-steps, there will always be errors. These can be minimized by using very small time-steps, but they will never vanish (In your program the size of the timestep is determined by PredatorPreyCalculator.dt and the errors will become smaller the smaller you set this value). But using small timesteps also means that simulating the system takes longer, therefor it is necessary to search for other means to reduce the errors. Such a method is using an integrator that performs better for the same size of timesteps.

The easiest integrator is the euler integrator which is a first order Taylor expansion:

"x_new = x_old + stepsize * dx(x_old,T)"

where dx(x,T) is the differential equation defining the system (aka the derivative of x)

The problem with this approach can be easily seen when you try to calculate the exponential function with it (using the differential equation "dx(x) = x"):

For example with a stepsize of 1 and an x_old=1=exp(0):

"exp(1) = x_new = 1 + 1 * 1 = 2"

Where you would expect x_new = exp(1) = e (eulers number) =~ 2.7

The problem can be lessened by setting the stepsize to 0.5. But the you will need two steps to calculate the value of exp(1) as you have to calculate exp(0.5) first. And still: the value you gain for exp(1) will be to small.

For this reason one usually uses more sophisticated integrators today. One of the still very simple ones is a Runge-Kutta of second order. I have sent you the modified "PredatorPreyCalculator.py" using this integration method per mail, I hope it works.

Sorry for the way too long comment.

Greetings

Stephan

cniehausThanks for your code, it works very well!

(Anonymous)I'm writing an extension on the ode integrators in scipy to have more of the functionality as found in matlab/maple/...:

http://scikits.appspot.com/odes

I'd be glad to add some educational codes like euler, or rk4. If you have a pure python implementation, it is bsd licenced like python/scipy, then send it to me 'Stephan'.

Showing how code must be used in very important, I added the doublependulum with output an mpeg movie in the docs:

http://svn.scipy.org/svn/scikits/trunk/o

An example of the predator prey would be nice too. :-) As this is educational, can you licence it bsd? I am all for gpl, but when it comes to these kind of things, bsd or public domain is the norm.

cniehausThat is pure Python. I will relicence it to BSD, just need to check for the exact licencetext.

Carsten

cniehausfictionpussOh, and PP sims rock, it was one of the first things that interested me about artificial life.

cniehausFor example, look at this commit:

http://github.com/cniehaus/biotools/com

Stephans code works way better, but a math teacher might want to demonstrate why the Euler methode is not the perfect choice here. This way he would just need to uncomment and comment a couple of lines and students could direclty SEE why RK2 is better here.