2 Major changes¶
Specific information regarding particular APIs, parameters and portability of code from version 11 can be found in the section Interface changes towards the end of the respective manual. This section lists general changes throughout MOSEK.
2.1 Release notes for 12.0¶
2.1.1 New features and improvements¶
Preprocessing
Improved handling of slightly infeasible problems in presolve.
For linear problems folding is employed if deemed worthwhile even if a basic solution is required.
New simplex optimizers
The simplex optimizers have been completely rewritten to improve speed and robustness.
The free simplex optimizer will run dual and primal simplex concurrently in a deterministic mode by default in multithreaded mode. If only one thread is available, then dual simplex is employed.
New linear interior-point optimizer and basis identification
The interior-point optimizer for linear problems has been rewritten.
The basis identification procedure has been rewritten from scratch to improve speed and stability. In particular the speed improvement can be significant for large and massively degenerate linear problems.
The basis reoptimization step employs the parametric primal-dual simplex algorithm which improves robustness vastly when the initial interior solution is inaccurate. In particular, it is less sensitive to degeneration than using using either the primal or dual simplex optimizer.
Mixed-integer optimizer
Improved version of conflict analysis and better conflict handling in general.
Improved symmetry reduction through better propagation algorithms and in some special cases the detection of orbitopes.
Improved space and time efficiency of symmetry handling.
Added heuristics that are not dependent on the existence of a solution to the root node relaxation.
Improved detection of implied integer variables.
Packaging
A cross-platform Maven Java artifact is available from https://central.sonatype.com/artifact/com.mosek/mosek. See the installation instructions for the Java Optimizer API or Fusion API for details.
Interfaces
A simple API for constructing a problem with fixed integer variables, useful for computing shadow prices for mixed-integer models, is introduced in the Optimizer, Fusion and MATLAB APIs.
2.1.2 Major breaking changes¶
Optimizer
The interior-point optimizer for linear problems will by default only return the basic solution, as opposed to the previous versions where both the interior-point and basic solution were returned. To get the old behavior the parameter
MSK_IPAR_REQUEST_INTPNTmust be set toMSK_ON.
Java
The MOSEK package is now
com.mosek.mosek, uniformly in the Maven artifact and the JAR file contained in the distribution.
2.1.3 Deprecations¶
Conic constraints restricted to \(x\in\K\) for a variable \(x\) are deprecated and will be removed in a future major version. Use affine conic constraints instead. This affects mainly the Optimizer API.
The OPF file format for conic problems is deprecated in favor of PTF.
The old MOSEK Optimization Toolbox for MATLAB remains supported, but will eventually be phased out and replaced by the new API.
The
condapackage is deprecated and may be dropped in a future release. Considerpipinstead.The optimizers with DEPRECATED in their name will be removed in version 13.