Any thinking in applying evolutionary algorithms to fusion?

[arupel]

Has any thinking be done in applying evolutionary programs (cellular automata with a purpose) to fusion.

It would seem that the current methods that can be applied are few. Evolution, itself, has been described as the dumbest way of realizing the smartest methods of achieving its goals.

Evolutionary programs might be applied in making current methods of obtaining fusion more efficient and exploring other methods, now not known.

It has been used well in design optimization. I see no loss, whether it succeeds or not, at least in making such attempts, if they have not been already applied.

Any thinking on this?

 

[mfb]

What exactly do you want to optimize that way?

You cannot build thousands of prototypes, you are limited to testing the most promising approaches. For a given type of fusion reactor, once you have a proper simulation of it, the parameters usually have optimal operation conditions that can be found by conventional minimization algorithms.

 

[the_wolfman]

Has any thinking be done in applying evolutionary programs (cellular automata with a purpose) to fusion.

It would seem that the current methods that can be applied are few. Evolution, itself, has been described as the dumbest way of realizing the smartest methods of achieving its goals.

Evolutionary programs might be applied in making current methods of obtaining fusion more efficient and exploring other methods, now not known.

It has been used well in design optimization. I see no loss, whether it succeeds or not, at least in making such attempts, if they have not been already applied.

Any thinking on this?

In order to use a genetic algorithm you need a model that can be used to judge the fitness of an individual solution. In fusion research there are many situations where our models are not accurate enough.

I do know of a few cases where fusion scientist have used genetic algorithms in their research. For example, a friend of mine used a genetic algorithm to calculate an optimal set of coils. But this is a simple example where we know how to calculate the magnetic field from currents in a coil.

 

[arupel]

It is not necessary to build thousands of prototypes based upon the evolutionary programs. What it does is offer insights, possibilities, and possible guidance. It is not the end all of design processes. Standard scientific computer programs are then used to study these possibilities and see what information they bring up.

It is an exploratory tool that can reveal factors that would not otherwise be noticed.

An example is that evolutionary programming was used to determine a wing design. Results offered several possibilities, including one in which the wings swept forwards, not backwards. Standard computer simulations showed the validity of many of these designs.

Evolutionary programming needs criteria to evaluate which cells will propagate and which cells will die.
From what I understand we are not sure what these criteria should be in the fusion processes. Still it does not hurt to try in using best guesses and see what comes out. As an iterative process these criteria are can be changed.

I see no harm in trying evolutionary programming, but with initially low expectations.
What is possible is that the methodologies in using evolutionary programs are, in a sense, an evolutionary process. It gets better as you go along. Many efforts will be failures, but some may reveal roads which should be further explored.

What has been said about babies is that they have no practical use at the time they are babies.

There is no harming in trying it.

Just a random thought: use fuzzy logic with fuzzy criteria in evolutionary programming.

 

[ZapperZ]

It is not necessary to build thousands of prototypes based upon the evolutionary programs. What it does offer are insights, possibilities, and possible guidance. It is not the end all of design processes. Standard scientific computer programs are then used to study these possibilities and see what information they bring up.

It is an exploratory tool that can reveal factors that would not otherwise be noticed.

An example is that evolutionary programming was used to determine a wing design. Results offered several possibilities, including one in which the wings swept forwards, not backwards. Standard computer simulations showed the validity of many of these designs.

Evolutionary programming needs criteria to evaluate which cells will propagate and which cells will die.
From what I understand we are not sure what these criteria should be in the fusion processes. Still it does not hurt to try in using best guesses and see what comes out.
As an iterative process these criteria are can be changed.

I see no harm in trying evolutionary programming, but with initially low expectations.
What is possible is that the methodologies in using evolutionary programs is, in a sense, an evolutionary process.
It gets better as you go along.

What has been said about babies is that they have no practical use at the time they are babies.

There is no harming in trying it.

Just a random thought: use fuzzy logic with fuzzy criteria in evolutionary programming.

The problem here is that you seem to be shooting in the dark, and expecting the rest of us to either justify your assertion, or falsify it.

You need to be VERY specific. Give a definite example of a "fusion" problem that is currently solved numerically that you think may be applicable to your "evolutionary programming". Simply saying that it might be applicable to "fusion" is extremely vague and superficial. This is not how it is done in a serious, scientific discussion.

So go pick out a specific problem, figure out how it was tackled, and then show clearly how this "evolutionary programming" is better.

Zz.

 

[mfb]

You can do this with wing geometries because we have good computer simulations of wings. Whatever the program comes up with can be tested to determine the fitness of this particular wing.

You can't do that with plasma geometries because we don't understand different geometries well enough to study them in an automated way in a computer.

There is no harming in trying it.

It takes time.

 

[averow45]

Arupel: thank you for an interesting idea. Not a physicist myself (just a fan) but I am surmising that the current models are still insufficient to apply evolutionary algorithms to (as others have alluded). I would guess that basic research in nuclear fusion is incredibly expensive and difficult to do at this point therefore the models remain incomplete. In my own field of Medicine we are starting to nail down the mysteries of protein folding, genomics, binding etc which is allowing some reasonable modelling to guide 'virtual' drug design. Good discussion by all thank you!

Tony Verow MD

 

[arupel]

Thanks for the limitations on plasma geometries. Perhaps at a later date evolutionary programs might be useful