0
$\begingroup$

In my problem, I have a lattice with a stochastic cellular automaton. In order to simplify a bit, let's say it is 1D. In my system, each node can be type A, B or C. A way to represent the system and operate fast over it is using an integer 64-bit variable, working bitwise.

Let me give an example of implementation for a system with 4 nodes. Let's say I have that nodes 0 and 3 are in state A, node 1 is in B, and 2 is in C, then:

A -> 1001 -> A = 9
B -> 0100 -> B = 4
C -> 0010 -> C = 2

You see that 3 integers can represent the whole state of the system. I can update the system with a single bit-wise operation. This is used a lot in cellular automatons.

But, now, suppose I have reactions such as

$$A+B\overset{\mu}{\rightarrow}2B$$ $$B\overset{\nu}{\rightarrow}C$$

I want to simulate this reactions in a synchronous way using a discrete time $\Delta t$. I can obtain the number of B individuals in the next timestep with:

B = A && r

where $r$ is a random number such that the bit at position $j$ is 1 with probability $p_j=1-\exp(-\mu\Delta t\cdot b_j)$, where $b_j$ is the number of B neighbours of individual $j$. This is, it is 1 with the probability of executing reaction 1.

My question is: is there an efficient method to obtain $r$?

$\endgroup$
1
$\begingroup$

Not sure if I understood correctly, but I think that if your parameters µ and $\Delta t$ are constant during all the simulations and for all nodes, you can generate a list of possible $p_j$ values in the beginning of the simulations. Because you have a limited possiblity of $b_j$ values (you have maximum 8 neighbors in a 2D grid if you look at first-neighborhood, so you can do an array of 9 possibles $p_j$ values).

Then you generate a random number with uniform law each time and compare if it's smaller than its corresponding $p_j$ value from the list.

$\endgroup$
  • $\begingroup$ Nice solution for constant rates. Any ideas if this is not the case? $\endgroup$ – VictorSeven Oct 2 '17 at 13:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.