Numeric Tolerance and Precision
ReducedTol
The optimal or reduced cost tolerance. The simplex method looks for a variable to enter the basis that has a negative reduced cost. Decision variables whose reduced cost is less than the negative of this tolerance are candidates for entering the basis during the simplex search.
Default: 10-5
Allowed range: 10-9 to 10-4
PivotTol
During the simplex algorithm, elements in the solution matrix must have an absolute value greater than this value to be candidates for pivoting.
Default: 10-5
Allowed range: 10-9 to 10-4
Precision
This value specifies how closely the calculated values on the left-hand side of constraints must match the right-hand sides in order for the constraint to be satisfied. Because of the finite precision arithmetic, a left-hand side that would ideally evaluate to 7.0 might compute as 6.9999999. With a precision of 10-6, the constraint A1 >= 7 would be considered satisfied in this case.
When using the LSGRG solver engine (an add-on engine), Precision is a relative amount rather than an absolute. So, for example, with a constraint
f(x) <= 1000
and a precision of 1e-6, the slack would be >= 1e-3. With the other engines that use an absolute precision, the slack would be >=1e-6.
Default: 10-6
Allowed range: 10-9 to 10-4
PrimalTolerance
The maximum amount by which the constraints can be violated and still considered feasible.
Engine: LP/Quadratic
Default: 10-7
Allowed range: 0 to 1
DualTolerance
The maximum amount by which the dual constraints and still considered feasible.
Engine: LP/Quadratic
Default: 10-7
Allowed range: 0 to 1
See Also
Enable comment auto-refresher