The Future of Optical Fibre

An anonymous reader writes "An Australian researcher has come up with a novel way of developing optical fibres. Steven Manos, a researcher at the Optical Fibre Technology Centre in Sydney, Australia has developed a method of using genetic algorithims for discovering optimal designs of optical fibres. An article on his work had this to say "The problem with designing optical fibres is starting with a specific set of criteria and then coming up with a design to fit this. The computer program developed by Manos, which is run on supercomputers, does this by mimicking the process of evolution. The computer program combines two patterns to create a third fibre 'offspring', which Manos described as "similar but a bit different". This process is repeated thousands of times with the 10 designs best suited for the particular application chosen to 'breed' again." Another case of "When in doubt, use brute force"?"

Comments

PDFs from Manos

[antic]

There are some interesting PDFs of papers co-written by Steven Manos available including these two:

• Novel fibre Bragg grating design using multiobjective evolutionary algorithms
• Optical Fibre Design with Evolutionary Strategies: Computational implementation and results

I'm not going to pretend that I know exactly what's going on, but the first of those two is worth looking at if you have even a passing interest. The second looks to be a little more towards the "deep end".

Much Better

[irokie]

This is a much better example of the application of Genetic Algorithms than the story that was on slashdot the other day (can't find a link, the one about Formula One racing).

in this case they have a very specific set of criteria.

it didn't however mention in the article how they're testing the designs (did it?)...
and are they actually manufacturing any of the designs that have come from thiss yet?

Comment removed

[account_deleted]

Comment removed based on user account deletion

Brute force? No way?

[carldot67]

Genetic algorithms are computational shortcuts that are used to very quickly find minima in complex multiparameter functions.

Suppose you wanted to find the lowest value of f(x)=sin(x) where x is from 0-360. (OK we all know its at x=270 but hear me out) - you can do it a couple of ways:

1. calculate sin(x) for all 360 possible values of "x" or
2. calculate sin(x) for (say) 20 values of "x".

Statistics says approach 2 will give you a couple of promising results, for only 1/18th of the effort. Now "breed" another 20 from the 6 values of x for which sin(x) were lowest, say 190, 210, 212, 260, 278, 290. This "next generation" gives sin(x) values whiach are closer to zero. Take the best 6 again. After three generations you are *close* to finding the values for "x" that give you sin(x)=0.

So systematic examination takes 360 tries and the genetic shortcut takes 60 tries - about 17% of the computational effort.

Now imagine a function a bit more complex; some mad multivariate affair like the wave equation. Each variable becomes a "gene" in the above "breeding program". All the time we are looking for parents and offspring that *tend* towards the answer we are looking for. (We also chuck in some unrelated parents too, since inbreeding can be bad - a tip stolen from Monte Carlo techniques [which see]).
The computational savings from GA, GP and MC techniques are potentially huge (as in orders of magnitude) so long as you dont care that:

a) The answer is not 100% exact
b) Some alternative minima are missed