Excel solver

  • Thread starter Thread starter Carlson
  • Start date Start date
C

Carlson

I cannot distinguish between the "precision" and "convergence" setting in
the Solver. Anybody know about them. They seem to represent the same thing.

Just increasing the "precision" seems OK.

Thanks
 
Carlson said:
I cannot distinguish between the "precision" and
"convergence" setting in the Solver. Anybody know
about them. They seem to represent the same thing.
Click to expand...

Precision is used when comparing equality. Suppose you set up Solver to
iterate until the Target is equal to a value. If we required exact
equality, the condition would almost never be satisfied because most numbers
with decimal fractions cannot be represented exactly internally, and because
more iterative arithmetic using numbers with decimal fractions rarely
converge to an exact number. So it is quite common to test for "equality"
with the following "fuzzy logic": abs(curValue - targValue) <= precision.

(According to Solver Help, Precision is used for other internal equality
comparisons as well.)

Convergence is used to determine when to stop the iteration by comparing
results of difference stages of the iteration. Suppose you set up Solver to
iterate until the Target is the min or max value. Of course, Solver could
simply iterate for the max number of iterations configured. But that is
wasteful if changes in the variables make an "insignificant" change in the
Target cell value. Convergence defines what is "insignificant". For
example, if Convergence is 0.0001 and the Target values for 5 iterations
(according to Solver Help) are withing +/-0.0001 of each other, Solver will
stop even if the Target equality condition is not met.
 
Thanks, Joe User.

What about "use automatic scaling" ?
It seems unnecessary since the computer today is very fast and efficient.
Running more iterations is feasible to get the solution.

Thanks
 
Carlson said:
What about "use automatic scaling" ?
Click to expand...

I'm not familiar with all the vagaries of Solver. I can only guess based on
the Solver Help description....

When dealing with very large numbers, perhaps Solver's normal algorithms do
not converge fast enough due to relatively small increments. This might be
especially true when the large numbers have more than 15 significant digits.
In that case, it might be possible that for some iterations, the small
increments have no impact on the large numbers due to the limitations of the
internal representation (IEEE 64-bit floating point).

Perhaps -- and this is only a WAG -- the option "use automatic scaling"
allows Solver to tailor its algorithms to account for these situations,
ensuring that the increment is significant relative to the "scale"
(magnitude) of the numbers.


----- original message -----
 
Back
Top