没人在家，又让我自己慢慢扣，好吧，我想着是应该初始值不同造成的

This is because that initial values are selected differently. See attached log below.

You can force them to fixed values. Then the result should be the same.

67       proc nlp;
68         lsq f1 f2;
69         decvar x1 x2;
70         bounds x1-x2<=.5;
71         lincon x1+x2<=.6;
72         f1=10*10*(x2-x1*x1);
73         f2=1-x1;
74         run;

NOTE: Your code contains 2 program statements.
NOTE: Gradient is computed using analytic formulas.
NOTE: Cross product Jacobian is computed using analytic formulas.
NOTE: Initial value of parameter x1 is set randomly to 0.6032013691.
NOTE: Initial value of parameter x2 is set randomly to 0.3147264259.
NOTE: Initial point was changed to be feasible for boundary and linear constraints.
NOTE: ABSGCONV convergence criterion satisfied.
NOTE: PROCEDURE NLP used (Total process time):
      real time           0.68 seconds
      cpu time            0.03 seconds


75       proc nlp;
76         lsq f1 f2;
77         decvar x1 x2;
78         bounds x1-x2<=.5;
79         lincon x1+x2<=.6;
80         f1=10*10*(x2-x1*x1);
81         f2=1-x1;
82         run;

NOTE: Your code contains 2 program statements.
NOTE: Gradient is computed using analytic formulas.
NOTE: Cross product Jacobian is computed using analytic formulas.
NOTE: Initial value of parameter x1 is set randomly to 0.4814047206.
NOTE: Initial value of parameter x2 is set randomly to 0.8789847772.
NOTE: Initial point was changed to be feasible for boundary and linear constraints.
NOTE: GCONV convergence criterion satisfied.
NOTE: PROCEDURE NLP used (Total process time):
      real time           0.06 seconds
      cpu time            0.01 seconds

ouch ,my guess is right,thank you very ,very ,very much!