Solver not resulting in min

[DrBiff]

I am trying to use excel's solver to minimize a very complex function that is subject to six constraints(optimizing the number of fins on a heat sink), but the solver never hits the minimum value, when I plot the function versus the number of fins, there is always a lower value(greater number of fins) than the solver outputs...and yes there is a min to my function, looks like a parabola

has anyone every come across this and hopefully have a solution

 

[electricpete]

I assume you have tried with different initial guesses closer to the minimum that you saw on your plot?

Also you mentioned many constraints, is it possible that the minimum that you saw on your graph is outside of your constraints?

I have seen solver do some strange things that I can't explain. Sometimes it seems like it makes a difference in what order you define the variables to be changed.



=====================================
Eng-tips forums: The best place on the web for engineering discussions.

 

[DrBiff]

ok now its going to the correct min value if I set it to only alter one value, but if I have it change two or more values it goes outside my constraints

how does the solver come up with a solution? does it test values outside the constraints and then filters the results to the constraints or is just ignoring my constraints?

 

[GregLocock]

There's input constraints and output constraints (and mixtures)

I suspect that it will only try solutions that meet input constraints, and will take less or no notice of results that exceed mixed or output constraints (depending on how much).

There is quite a bit of literature around on how the solver works, it is a crippled version of a commercial product.





Cheers

Greg Locock

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[DrBiff]

oh...well crap

 

[COEngineeer]

did you put enough cycles?

 

[DrBiff]

ya it wasn't a cycle problem its that if the guess goes outside a constraint there are #NUM! and division by zero errors....what I ended up doing was changing the initial guesses to something more realistic and the computer seems to not go out side the constraints and hits the minimum value, so problem solved....hopefully I can automate it with out going outside the constraints

 

[IRstuff]

Solvers are generally modelled along the lines of a Newton-Raphson algorithm, so anything with multiple minimas or discontinuities will cause problems if the starting point is in a bad place.

TTFN

Eng-Tips Policies FAQ731-376

 

[GregLocock]

And excel's solver is not robust if there are calculation errors - that's a large part of setting the problem up, making sure that any combination of allowed parameters yields a numerical answer.

Cheers

Greg Locock

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[prost]

Is the problem the solver isn't finding the global minimum or is the solver missing even local minima?

No algorithm exists will find the global minimum in a space of local minima (please refute and provide example if you have one! I'd be extremely interested in the algorithm that is guaranteed to find the global minimum). For instance, imagine if you had a surface in just a 3D space; the surface has lots of dips and peaks, so that there are lots of local minima but only one, global minimum. If you know the function that maps the surface, make an initial guess near a local minima which IS NOT the global minima, than your solver might find that local minima after a few iterations. Or it might not. Either way, just be looking at computer output, you won't know whether your minimum is a local one or the global one.

 

[DrBiff]

your are 100% correct in that it is finding local minima, my solution is based on if statements that make an educated initial guess that results in the correct minimum, where as before I was just sticking in 1 and the answer was 75.........

 

[GregLocock]

prost - there are ways around that, but so far as I am aware the Solver does not use them. I get the impression it finds local minima subject to the constraints arguments.

Of course if you can improve your initial guess then the Solver is a fast solution to find the local optimum.

Typically I'd use GA or just a random exploration, depending on the size of the problem.



Cheers

Greg Locock

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.