Solver

L

LCC

Hello! Friends

Relating to the EXCEL/SOLVER.

Could someone advise how I can set a variable(cell) NOT
EQUAL to zero as a CONSTRAINT?

Thanks.

LCC
 
D

duane

take the absolute value of your number in another cell, and put a
constraint that this is >0.
 
L

LCC

Thanks Duane and Martin.

Setting it to absolute value is a brilliant idea. However,
there is no "strictly greater" than option in the
CONSTRAINT box.

Apart from doing what Martin suggested, that is to set the
absolute value >= to a minute number, is there another way
to get around this limitation?

Thanks.
 
D

Dana DeLouis

Just some thoughts. The reason there is no "strictly greater" is that it
becomes very hard to arrive at a solution. In other words, at what value of
precision is your solution correct? If 0.0001 is correct, there is always
another "smaller" number then this that is still greater than zero. It's
hard to come up with a problem that says 0.00000000001 is a correct answer
because it is greater than zero (or 1E-100 for that matter) , but 0 is not.
And remember, there is always another "smaller" number no matter what you
come up with. This is why there is usually only >=, or <=. There has to be
some type of limit for a constraint.
Just some other ideas to consider. The idea of using $E$6>=1E-30 is a good
idea, but I think you will find that Solver will run into it's limits on
Precision and Convergence before it reaches a constraint of this small a
number.
Using ABS is a good idea, but my experience is that this will not work.
Solver can not resolve the problem of the jump in direction with functions
like ABS. The function ABS is not quite a discontinuous function, but it
does give Solver a hard time.
Are you trying to work with an integer amount that you want not equal to
zero? Are you working with possible negative values?
Again, just some thoughts.
 
L

LCC

Thanks for your input.

I fully agree with you that there is always a smaller
number than the previous and the ABS is a discontinous
function. And I also appreciate the convergence issue that
the Solver algorithm has.

I am working on some MATRIX problems that often called for
NON-TRIVIAL (non-zero) solution that can be +ve or -ve.
Even in Optimisation problems, there are always "strictly
(or <)" requirements.
Click to expand...

Therefore, if the SOLVER has a "NOT EQUAL" option, then
the "strictly" constraint can be resolved by inserting
BOTH "NOT EQUAL" and ">= (or <=)" constraints.

Your valuable comments, please.
 
Top