Linear-Integer solver:

   + Improved heuristics for general integer programs.

    + Average performance improvement of 2-3% on our standard test set.

   + Improved method for generating all alternative optima to a linear program. 

      The first call to LSgetNextBestSol() creates a prespecified number of corner points  

       via pivoting on the optimal solution set. 

       Subsequent calls will nonredundantly return successive corner points. 

       Standard solution query methods can be used to access primal-dual vectors 

       following each call to LSgetNextBestSol. 

Linearization

     + Support for Indicator constraints, e.g., z = 0 implies x + y <= 0;

     + more expressions can be automatically linearized, so you can now use a fast linear 

        solver where otherwise a much slower (30x?) nonlinear solver.might be required.

     + advanced linearization of QP and Conic models

     + improved linearization of certain IF expressions. 

Nonlinear and Global solver:

      + faster (order of magnitude) solution of linear fractional programs (ratio objectives).

      + improved bound tightening process in preprocessing of nonlinear models.

      + auxiliary variables generated automatically to improve performance with

          complicated expressions. 

      + Support for additional useful but “problematic” functions:

         Power utility function  (x^g-1)/g  and the exponential ratio function (exp(g) – 1)/g,

         are important in some situations modeling consumer behavior.

         LINDO API can now robustly avoid the numerical problems that would otherwise 

         occur when g approaches 0.

Interfaces:

       + Julia/JuMP is now supported officially.

       + Python interface installations now easier via pip (pypi.org)

       + Matlab interface now has two alternative methods for linear and integer

          optimization, LSlinprog and LSintprog. Argument lists follow their counterparts, 

           linprog and intprog, in Matlab’s optimization toolbox.