2 Major changes

Specific information regarding particular APIs, parameters and portability of code from version 8 can be found in the section Interface changes towards the end of the respective manual. This section lists general changes throughout MOSEK.

2.1 New features

Optimizer

  • Introduced the primal exponential cone \(\EXP\) (the constraint \(x\geq y\exp(z/y)\)).
  • Introduced the primal power cone \(\POW\) (the constraint \(x^\alpha y^{1-\alpha}\geq |z|\)).

Presolve

  • Improved presolve. Primary noticeable for particular conic problems.

Interior-point Optimizer

  • Tighten the stopping criteria when solving conic optimization problems.
  • Changed the scaling so better accuracy is obtained in some cases.

Mixed-integer Optimizer

  • Handles exponential and power cones.
  • Introduced an outer approximation method for solving conic mixed integer optimization problems. This can be enabled with the parameter MSK_IPAR_MIO_CONIC_OUTER_APPROXIMATION.

Linear algebra utilities

  • Employs the BLIS library when run on an AMD CPU instead of the Intel MKL library.
  • Updated the Intel MKL BLAS library employed.

License Manager

  • The FLEXlm license system has been updated to version 11.14.1.0.

Interface

  • Added support for constraints of the form \(Fx+g\in \K\) (affine conic constraints) in the Toolbox for MATLAB and in Rmosek.

General

  • The Intel C compiler version 19 is used. This will improve performance of the interior-point optimizer on computers with a CPU that supports AVX-512 instructions. However, for most sparse problems the improvement will be negible.
  • Much improved performance on recent AMD CPUs.
  • Mosek can now read and write Zstandard compressed files.
  • Fixed many 32 bit overflow issues occuring for huge problems.

2.2 Removed features

General nonlinear optimizer

The optimizer for general nonlinear problems in C is removed. Most likely problems using this tool can be expressed in conic form.

Separable convex optimization SCopt

The Separable Convex interface (SCopt) and extensions for exponential and dual geometric optimization are removed. All applications of these interfaces can now be expressed using cones. It is recommended to rewrite these problems in conic form.

Please see your interface documentation for a tutorial on converting SCopt-like problems to conic form. Please ask us if there are questions or you require help adapting your code.

Note that the SCopt interface was relying on the general nonlinear optimizer.

Fusion for MATLAB

Separate Fusion distribution for MATLAB (mosekmatlab.jar) is discontinued. Users are recommended to import the general Java library mosek.jar into MATLAB and adapt the code to 0-based numbering. This applies to all explcitly indexed references to entries in Fusion objects such as variables and expressions. See also https://themosekblog.blogspot.com/2019/02/porting-fusion-for-matlab-code.html

Other

  • Remove the optimial partition sensitivity analysis method.
  • Remove all “near” solution statuses, such as “near optimal”.

2.3 Deprecated features

None.