Extremely large or extremely small numerical values might cause computational difficulties for some of the algorithms in PROC OPTNET. For this reason, each algorithm restricts the magnitude of the data values to a particular threshold number. If the user data values exceed this threshold, PROC OPTNET issues an error message. The value of the threshold limit is different for each algorithm and depends on the operating environment. The threshold limits are listed in Table 2.32, where M is defined as the largest absolute value representable in your operating environment.
Table 2.32: Threshold Limits by Statement
Matrix |
Graph Links |
Graph Nodes |
|||||
---|---|---|---|---|---|---|---|
Statement |
weight |
weight2 |
lower |
upper |
weight |
weight2 |
|
|
|
||||||
|
|
||||||
1e15 |
1e15 |
1e15 |
1e15 |
1e15 |
|||
|
|||||||
|
|||||||
|
|
||||||
1e20 |
To obtain these limits, use the SAS function constant. For example, the following DATA step assigns to a variable x
and prints that value to the log:
data _null_; x = constant('SQRTBIG'); put x=; run;
For most of the algorithms in PROC OPTNET, there is no valid interpretation for a missing value. If the user data contain a missing value, PROC OPTNET issues an error message. One exception is for the minimum-cost network flow solver when you are setting the link or node bounds. In this case, a missing value is interpreted as the default bound value, as described in the section Minimum-Cost Network Flow. Another exception is the linear assignment problem when you are using the matrix input format. A missing value in this case defines an invalid assignment between a row and a column of the matrix. An example of this is shown in the section Linear Assignment (Matching).
For certain algorithms in PROC OPTNET, a negative link weight is not allowed. The following algorithm issues an error message if a negative link weight is provided: