[FRIAM] recap on Rosen

[Marcus G. Daniels]

phil henshaw wrote:
> Ok, 'find a function' assumes there is one to find, but the problem set is
> running into behavior which has already had major consequences (like
> starvation for 100million people because of an unexpected world food price
> level shift) and the question is what 'function' would you use to not be
> caught flat footed like that.
The caloric requirements of a person are autocorrelated, but probably 
for a lot of models a constant will suffice -- a certain amount of body 
weight decrease, and then the probability of death goes up.   As for 
price fluctuations, that's a matter of modeling the natural resources 
that go in to food, the costs and benefits to motivate farmers, the 
commodity markets, and so on.   Certainly we can try to understand how 
each of these work, and then do what-if scenarios when one or more 
components are perturbed (or destroyed).   It's still a matter of 
finding stories (functions) to fit observables.  The availability and 
accuracy of those observables may be poor, and sometimes all that is 
possible to imagine worst and best cases, run the numbers, and see how 
the result changes.
> Is there some general function to use in
> cases where you have no function and don't even know what the problem
> definition will be?   
>   
I think you do know what the problem could look like, but most details 
remain unspecified.   If you can construct an example that has 
catastrophes of the kind you often talk about, and spell out all of the 
details of your work of fiction (that even may happen to resemble 
reality), such that the what-if scenarios can be reproduced in 
simulations, then others can study the sensitivities.   If there is a 
`forcing structure' that will occur in many, many variant forms, then 
you can demonstrate that.
> I actually have a very good one, but you won't like it because it means
> using the models to understand what they fail to describe rather than the
> usual method of using them to represent other things.
Right.  Model predicts something, it turns out to have some error 
structure and that structure suggests ways to improve the model or make 
a new one.  Paper published. Meanwhile another guy makes a different 
model on the same phenomena and publishes a paper.   Third person reads 
the two papers and has idea that accounts for problems in both.   So she 
makes a new model!

Marcus