AbstractThis paper presents a simplified approach to neural optimization in the presence of linear equality constraints. In contrast to the standard Lagrangian approach, the constraints simplify the final neural circuit instead of complicating it. the number of elements used is also significantly reduced. Instead ofn+tintegrators we need onlyn–t. There is also a similar saving in the number of preprocessing non‐linear devices. Elimination of the constraints allows a large speed‐up of the sol