The Monte Carlo Method
Written by Mike James   
Friday, 01 November 2024
Article Index
The Monte Carlo Method
Simulations
Pi Program
Working out the matrix product
Page 4 of 4

Working out the matrix product

To work out the matrix product we have to form sums involving the matrix elements a(i,j) and the vector elements b(k) to be precise

y(i)=Σj a(i,j)*b(j)

or in other words

y(i)=sum over all j of a(i,j)*b(j)

This isn’t the place to be teaching matrix algebra, but it comes down to multiplying the elements row j in the matrix by the corresponding elements of vector and adding them up - usually said as "row times column".

 

matrix

How to multiply a vector by a matrix

 

Computing this very non-random value using random numbers seems a very unlikely prospect but it is fairly easy.

For the moment let's suppose that all the values in the matrix are positive and less than one and that each row adds up to one. If this isn't the case then we can modify the matrix to make it true and the solution of the new matrix is related to the solution to the original.

This condition allows us to treat the rows as probabilities of something happening.

Focus for a moment on a single row - its entries give you the probability of some event occurring.

To simulate these events  all you do is set up a number of random number generators R(i) one for each row of the matrix. Each random number generator is set up so that it returns b(j) i.e a value in the B vector, with probability a(i,j) .

For example, if the matrix A was

A= (0.5 0.3 0.2)
   (0.2 0.1 0.7)

and the vector b was

    (4.3)
b=  (6.2)
    (3.2)

Then for the first row of the matrix we would create a random number generator R(1)  that implemented the distribution of the first row. That is R(1) produces  4.3 with probability 0.5, 6.2 with probability 0.3 and 3.2 with probability 0.2. For the second row we would create a random number generator R(2) that produced 4.3 with probability 0.2, 6.2 with probability 0.1 and 3.2 with probability 0.7.

That is each random number generator returns the elements of the vector with probabilities given by the rows of the matrix – this can be achieved using the simple methods described earlier.

Now we can forget about matrix multiplication and simply run the random number generators - one for each row. We run the simulation and work out the average value produced by each of the random number generators. The resulting vector of averages one from each random number generator estimates the vector we want i.e. y.

In terms of our earlier example once we have R(1) and R(2) we would run them and work out the average value they produced. This average is an estimate of the result of multiplying the vector b by the matrix A.

That is:

y=(y(1))= (mean of R(1)) = Ab
  (y(2))  (mean of R(2))

I don’t know if you’ll be impressed by this demonstration of how randomness can be used to get an estimate of an answer that seems to have nothing to do with chance  - but you should be.

Let's look at a simple JavaScript program to implement this algorithm to see how easy it really is. First we need two random number generators R1 and R1:

function R1() {
 var r = Math.random();
 if (r < 0.5) return 4.3;
 if (r < 0.5 + 0.3) return 6.2;
 return 3.2
}
function R2() {
 var r = Math.random();
 if (r < 0.2) return 4.3;
 if (r < 0.2 + 0.1) return 6.2;
 return 3.2;
}

These use the "size of interval" method described in the introduction to generate random numbers with the specified probabilities. Now all we have to do is to write a loop that generates the appropriate random numbers using R1 and R2 and work out the mean:

var num = 100;
var r1mean = 0;
var r2mean = 0;
var total = 0;
for (var i = 1; i <= num; i++) {
 r1mean = r1mean + R1();
 r2mean = r2mean + R2();
 total++;
 Text1.value = r1mean / total;
 Text2.value = r2mean / total;
}

If you try this out using Firefox say you should get results something like:

n       R1     R2100     4.669  3.63799
1000    4.6696 3.75449
10000   4.6552 3.71641
100000  4.6479 3.71656
1000000 4.6510 3.72015

which should be compared to the exact result R1=4.65 R2=3.72

The complete program as a simple HTML page is:

<!DOCTYPE html>
<html lang="en">
 <head>
  <meta charset="utf-8" />
  <title>Matrix Multiply</title>
 </head>
 <body>
  <input type="text" id="Text1"/>
  <br/>
  <input type="text" id="Text2"/>
  <script>
   function R1() {
    var r = Math.random();
    if (r < 0.5) return 4.3;
    if (r < 0.5 + 0.3) return 6.2;
    return 3.2
   }
   function R2() {
    var r = Math.random();
    if (r < 0.2) return 4.3;
    if (r < 0.2 + 0.1) return 6.2;
    return 3.2;
  }
  var num = 100;
  var r1mean = 0;
  var r2mean = 0;
  var total = 0;
  for (var i = 1; i <= num; i++) {
   r1mean = r1mean + R1();
   r2mean = r2mean + R2();
   total++;
   Text1.value = r1mean / total;
   Text2.value = r2mean / total;
  }
  </script>
 </body>
</html>

 

This is the beginning of a wide range of computing applications that use random numbers to get at answers that would otherwise be too difficult to compute exactly.

Who needs quantum computers when you have Monte Carlo to play with…

Related Articles

Inside Random Numbers
Mod function
Advanced Hashing
How not to shuffle - the Knuth Fisher-Yates algorithm
The Programmer's Guide to Chaos
The Monty Hall Problem
John Von Neumann The Great Polymath
 

espbook

 

Comments




or email your comment to: comments@i-programmer.info

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin.

Banner

Principles Of Execution - The CPU

The real complexity of any computer system resides in the processor, but do you know how it works? I mean how it really works? How does the code that you write turn into something that does something? [ ... ]


+ Full Article
IP Addressing and Routing

Every programmer should understand how the Internet works and this means understanding IP addressing and routing. It's a good time to find out about such things with DOS attacks on the rise and IPv6 t [ ... ]


+ Full Article
Other Articles

<ASIN:352740760X>

<ASIN:0812975219>

<ASIN:0387942823>


<< Prev - Next
Last Updated ( Friday, 01 November 2024 )