一直郁闷与这点:在一本书上看到先关内容
是:《nonlinear programming》 second edition by dimitri P.bertsekas (MIT)
在书中309页
给出两个解释:从几何 和 算法角度进行了解释
几何角度:参看书籍
算法角度:1.将不等式 active 约束松弛成gj(x)<=uj,
the optimal cost will tend to decrease because the constraint set will become larger.
therefore the sensitivity interpretation of the lagrange multiplier uj* = -(deta cost due to uj )/uj suggests that uj*>=0
就是说从边际价格的角度来看,随着active 约束的松弛,可行域增大,总得费用会减小,拉格朗日乘子代表的边际价格会增大。