BARON: *** User dit not provide appropriate variable bounds ***

Dear gamsworld,

I am working with a non-linear model using BARON (with MINOS and
CPLEX) as solver. I keep getting this message:
*** User dit not provide appropriate variable bounds ***
*** Globality is therefore not guaranteed ***

Does this mean that the solution obtained is only the first local
optimum found? How should I get around this problem? (Obviously I
should try working with the bounds, but…)

I have made (physically sound, but generally not very tight) lower and
upper bounds for every variable in the model, but still I get the same
message. I have also been looking at the output file sum.dat, which
tell me which of the variables that miss bounds (and I have checked
that they are bounded). Even when I set the upper and lower bound to
the same value (the solution from last run) for the variables with
missing bounds (according to sum.dat) I get the same message (and the
same information in sum.dat). This seems very strange to me.

If I set the bounds close to the solution form last run for all
variables, the message disappears.

The problem seems to relate to equations on the form eq … (k1 + x1) *
x2 * x3 =e= k2 +k3 * sum(i, k4(i) * x4(i) * sum(j, k5(j) *
x5(i)**ord(j))), where xN are variables, and kN are constants, and j
contains 5 elements (card(j) is 5). May such equations be too tough to
handle for BARON?

I really hope that someone can help me, because this has been
bothering me for quite some time now, and I have no clue on what to do
with it.

Best regards,
Erik Lien Johnsen

\

Fix: The error message does not occur anymore after I scaled the
relevant variables to a range close to 1. However, with different
scaling factors GAMS/BARON returns different solutions, even though
there is no error message. May this happen because of numerics, or
does BARON only return local optima? The model status is 2 - locally
optimal, but there is no globally optimal status in GAMS for NLPs and
MINLPs, right?

Erik

\