Originally posted by: islaabdi

Hi everyone,

I've got a problem with my optimization file. In fact I'm trying to use a binary variable but I've got some error (Cplex default can't extract the expression). Do I need to initialize the variable ? If yes how ?

Here my data and mod files.

Thanks

Originally posted by: EdKlotz

I don't have time to run your model; in general you may get more responses if you point to the specific lines that cause the

errors.   I suspect they involve the multiplications of a float variable by a binary, such as

allocate1:   (sum(i in Investments) Return[i]*Binary[i]*Allocation[i])>= 0.02;

  // sum of allocations equals amount to be invested

  allocate2: (sum (i in Investments) (Allocation[i]*Binary[i])) == Wealth;

You declared these as

dvar int  Binary[Investments] in 0..1;

dvar float  Allocation[Investments] in FloatRange;  // Investment Level

I don't think any of CPLEX's APIs will directly extract the product of a binary and continuous variable.

You either need to linearize this expression through a series of constraints (you can search this forum and

also www.or-exchange.org and will probably find a description of how to do that if it is unfamiliar, but you might

want to try it yourself first), or you can make use OPL's logical if/then functionality and let OPL/CPLEX do the internal

linearization.   In other words, declare an array of dvars, say z[i], and then specify the logical constraint

if Binary[i] == 1 then z[i] = Allocate[i]

if Binary[i] == 0 then z[i] = 0

Originally posted by: islaabdi

I'm sorry if my question wasn't clear. As you said the problem was from the product between a binary variable and a continuous. The solution was to linearize the expression and adding some constraint about it.

Thanks you for your response.

Hi,

this looks similar to

https://www.ibm.com/developerworks/community/forums/html/topic?id=cee677ca-ee08-488e-82c6-851f26790d1d&ps=25

You could  write a logical constraint like

forall(i in Investments) Binary[i]==(Allocation[i]>=0.000001);

and then you would be able to control the number of investments that are not zero.

regards

Originally posted by: EdKlotz

I agree with both responses that followed mine.   But, the main point here is that while can indeed linearize the product of a binary and general integer yourself, you  can save yourself some work by taking advantage of the more powerful modeling constructs in either the OPL language or CPLEX's object oriented APIs.   If you use CPLEX's C API, then you do indeed need to do the linearization yourself.