MOSEK ApS, C/O Symbion Science Park, Fruebjergvej 3, Box 16, 2100 Copenhagen Ø, Denmark

Introduction

MOSEK is a software package for the solution of linear, mixed-integer linear, quadratic, mixed-integer quadratic, quadratically constraint, conic, and semidefinite mathematical optimization problems. MOSEK is particularly well suited for solving large-scale linear, convex quadratically constraint, and conic programs using an extremely efficient interior point algorithm.

These problem classes can be solved using an appropriate optimizer built into MOSEK. All the optimizers available in MOSEK are built for the solution of large-scale sparse problems. Current optimizers include:

• Interior-point optimizers for continuous and conic problems
• Simplex optimizer for linear problems
• Branch-and-cut optimizer for mixed-integer linear, quadratic, and conic problems

Licensing

Licensing of GAMS/MOSEK is similar to other GAMS solvers. MOSEK is licensed in two different ways:

• GAMS/MOSEK:
  All model types.
• GAMS/MOSEK Solver Link:
  Users must have a separate, licensed MOSEK system. For users who wish to use MOSEK within GAMS and also in other environments.

Attention

The free bare-bone link mode (previously GAMS/OSIMOSEK) that allowed to solve LP and MIP when the user had a separate MOSEK license installed has been removed. If you relied on using this bare-bone link option, then do not hesitate to contact sales@mosek.com to arrange for a GAMS/MOSEK Solver Link license.

For information regarding MOSEK standalone or interfacing MOSEK with other applications contact sales@mosek.com.

Solving Problems in Parallel

MOSEK can exploit multiple CPUs (or a CPU with multiple cores) to solve an optimization problem when using the interior-point or the mixed-integer optimizers.

This implies that whenever the MOSEK interior-point optimizer should solve an optimization problem, then it will try to divide the work so each CPU gets a share of the work. The user decides how many CPUs MOSEK should exploit. Unfortunately, it is not always easy to divide the work. Also some of the coordination work must occur in sequential. Therefore, the speed-up obtained when using multiple CPUs is highly problem dependent. However, as a rule of thumb, if the problem solves very quickly, i.e., in less than 60 seconds, then there is no advantage in using the parallel option.

The parameter MSK_IPAR_NUM_THREADS sets the number of threads (and therefore the number of CPU's) that the optimizer will use.

Infeasible/Unbounded Models

Farkas Certificates

MOSEK determines if either the primal or the dual problem is infeasible by means of a Farkas certificate. In such a case MOSEK returns a certificate indicating primal or dual infeasibility.

The primal infeasibility certificate indicates a primal infeasible model. For a minimization problem

\[ \begin{array}{rl} \text{minimize} & \langle c,x\rangle \\ \text{subject to} & Ax = b, \\ & x \ge 0, \\ \end{array} \]

the primal infeasibility certificate is the solution \(y\) satisfying \(A'\,y \le 0\) and \(\langle b,y\rangle > 0\).

A primal infeasibility certificate is reported in the marginal records of the variables and equations. As no primal solution is available in this case, the level values for variables and equations and the objective function value are set to 0 (setting them to NA would be more appropriate, but GAMS does not support this well). At the moment, primal infeasibility certificate are not available for conic programs.

Since GAMS reports all model statuses in the primal space, the notion of dual infeasibility does not exist and GAMS reports a status of unboundedness, which assumes that the primal problem is feasible. Although GAMS reports the primal as unbounded, there is the possibility that both the primal and dual problem are infeasible. To check if this is the case, the user can set appropriate lower and upper bounds on the objective variable, using the (variable).LO and (variable).UP suffixes and resolve.

The dual infeasibility certificate is reported in the level values for the variables. As no dual solution exists, the marginal values for both variables and equations are set to NA.

For more details on primal and dual infeasibility certificates see the MOSEK Modeling Cookbook.

Infeasibility Report

MOSEK has some facilities for diagnosing the cause of a primal or dual infeasibility. They can be turned on using the parameter setting MSK_IPAR_INFEAS_REPORT_AUTO. This causes MOSEK to print a report about an infeasible subset of the constraints, when an infeasibility is encountered. Moreover, the parameter MSK_IPAR_INFEAS_REPORT_LEVEL controls the amount of information presented in the infeasibility report. We will use the TRNSPORT example from the GAMS Model Library with increased demand \((b(j)\leftarrow 1.6\,b(j))\) to make the model infeasible. MOSEK produces the following infeasibility report:

MOSEK PRIMAL INFEASIBILITY REPORT.

Problem status: The problem is primal infeasible

The following constraints are involved in the primal infeasibility.

Index    Name                    Lower bound      Upper bound      Dual lower       Dual upper
1        supply(seattle)         none             3.500000e+002    0.000000e+000    1.000000e+000
2        supply(san-diego)       none             6.000000e+002    0.000000e+000    1.000000e+000
3        demand(new-york)        5.200000e+002    none             1.000000e+000    0.000000e+000
4        demand(chicago)         4.800000e+002    none             1.000000e+000    0.000000e+000
5        demand(topeka)          4.400000e+002    none             1.000000e+000    0.000000e+000

The following bound constraints are involved in the infeasibility.

Index    Name                    Lower bound      Upper bound      Dual lower       Dual upper

The report indicates which constraints and bounds are causing the infeasibility. In this case, the constraints causing infeasibility are supply and demand. The values in the columns Dual lower and Dual upper are also useful, because if the dual lower value is different from zero for a constraint, then it implies that the lower bound on the constraint is important for the infeasibility. Similarly, if the dual upper value is different from zero on a constraint, then this implies the upper bound on the constraint is important for infeasibility.

Conic Programming

MOSEK is well suited for solving generalized linear programs involving certain conic constraints.

For an overview of quadratic conic programming and how these conic constraints are implemented in GAMS, see the Section Conic Programming in the GAMS User's Guide. Note, that for Mosek no variable can appear in more than one conic constraint.

Additionally, the primal power cone, defined as

\[ x_0^\alpha x_1^{(1-\alpha)} \geq \sqrt{\sum_{i=2}^n x_i^2}, \quad x_0, x_1 \geq 0, \]

with \(\alpha \in (0,1)\), and the primal exponential cone, defined as

\[ x_0 \geq x_1 \exp(x_2/x_1), \quad x_0, x_1 \geq 0, \]

are available.

Since GAMS does not offer capabilities to directly restrict a variable to one of these cones, the GAMS/MOSEK link tries to detect the above algebra from a general nonlinear equation. For example, the following GAMS code should work with MOSEK:

Set i / 0*10 /;
Variable x(i);
Equations e1, e2;
Scalar alpha;
e1.. x('0')**alpha * x('1')**(1-alpha) =G= sqrt(sum(i$(ord(i)>2), sqr(x(i))));
e2.. x('0') =G= x('1') * exp(x('2') / x('1'));
x.lo('0') = 0;
x.lo('1') = 0;

See also testlib models powercone1, powercone2, and expcone1.

Solver Options

MOSEK works like other GAMS solvers, and many options can be set in the GAMS model (see GAMS Options). The most relevant GAMS options are reslim, nodlim, iterlim (iteration limit for simplex and interior point algorithms), optca, optcr, and optfile. A description of all available GAMS options can be found in GAMS Options and Solver related options.

We remark that MOSEK contains many complex solver options, many of which require a deep understanding of the algorithms used. For information on how to use a GAMS/Mosek options file, see The Solver Option File. For example, an option file

MSK_IPAR_INTPNT_MAX_ITERATIONS                20
MSK_IPAR_INTPNT_SCALING         MSK_SCALING_NONE

limits the number of interior-point iterations to 20 and disables scaling.

In the following, we summarize the MOSEK options that are available through the GAMS/MOSEK interface. For details, see Detailed Descriptions of MOSEK Options.

General

Problem Data

Presolving

Simplex Optimizer

Interior Point Optimizer and Basis Identification

Mixed Integer Optimizer

Infeasibility Analyser for Continuous Problems

Output

The MOSEK Log File

The MOSEK log output gives much useful information about the current solver progress and individual phases.

Log Using the Interior Point Optimizer

The following is a MOSEK log output from running the TRNSPORT model from the GAMS Model Library:

Optimizer started.
Interior-point optimizer started.
Presolve started.
Linear dependency checker started.
Linear dependency checker terminated.
Eliminator - tries                  : 0                 time                   : 0.00
Lin. dep.  - tries                  : 1                 time                   : 0.00
Lin. dep.  - number                 : 0
Presolve terminated. Time: 0.00
Optimizer  - threads                : 1
Optimizer  - solved problem         : the primal
Optimizer  - Constraints            : 5
Optimizer  - Cones                  : 0
Optimizer  - Scalar variables       : 11                conic                  : 0
Optimizer  - Semi-definite variables: 0                 scalarized             : 0
Factor     - setup time             : 0.00              dense det. time        : 0.00
Factor     - ML order time          : 0.00              GP order time          : 0.00
Factor     - nonzeros before factor : 11                after factor           : 12
Factor     - dense dim.             : 0                 flops                  : 1.80e+02

The first part gives information about the presolve (if used). The main log follows:

ITE PFEAS    DFEAS    GFEAS    PRSTATUS   POBJ              DOBJ              MU       TIME
0   6.0e+02  1.0e+02  1.0e+02  1.00e+02   1.053000000e+00   0.000000000e+00   1.0e+02  0.00
1   5.9e+02  1.3e+02  2.6e+02  0.00e+00   3.063797526e+00   2.650041612e+02   2.4e+02  0.00
2   4.6e+01  1.0e+01  2.0e+01  -9.59e-01  3.650704301e+01   2.594816940e+02   1.9e+01  0.00
3   3.9e-01  8.7e-02  1.7e-01  -4.39e-01  1.604589379e+02   2.276036163e+02   1.6e-01  0.00
4   2.7e-02  6.0e-03  1.2e-02  9.62e-01   1.627664502e+02   1.676438787e+02   1.1e-02  0.00
5   2.2e-03  4.9e-04  9.7e-04  1.04e+00   1.585004810e+02   1.591499235e+02   8.9e-04  0.00
6   3.1e-04  6.9e-05  1.4e-04  1.01e+00   1.546312243e+02   1.547272945e+02   1.2e-04  0.00
7   2.9e-05  6.5e-06  1.3e-05  1.01e+00   1.536906429e+02   1.536999628e+02   1.2e-05  0.00
8   7.6e-08  1.7e-08  3.4e-08  1.00e+00   1.536751995e+02   1.536752387e+02   3.1e-08  0.00
9   7.5e-12  1.7e-12  3.4e-12  1.00e+00   1.536750000e+02   1.536750000e+02   3.1e-12  0.00
Basis identification started.
Primal basis identification phase started.
ITER      TIME
1         0.00
Primal basis identification phase terminated. Time: 0.00
Dual basis identification phase started.
ITER      TIME
0         0.00
Dual basis identification phase terminated. Time: 0.00
Basis identification terminated. Time: 0.00
Interior-point optimizer terminated. Time: 0.00. 

Optimizer terminated. Time: 0.00

Interior-point solution summary
  Problem status  : PRIMAL_AND_DUAL_FEASIBLE
  Solution status : OPTIMAL
  Primal.  obj: 1.5367500002e+02    nrm: 6e+02    Viol.  con: 3e-10    var: 0e+00
  Dual.    obj: 1.5367500002e+02    nrm: 2e-01    Viol.  con: 0e+00    var: 6e-11

Basic solution summary
  Problem status  : PRIMAL_AND_DUAL_FEASIBLE
  Solution status : OPTIMAL
  Primal.  obj: 1.5367500000e+02    nrm: 6e+02    Viol.  con: 0e+00    var: 0e+00
  Dual.    obj: 1.5367500002e+02    nrm: 2e-01    Viol.  con: 0e+00    var: 5e-11

Return code - 0 [MSK_RES_OK]: No error occurred.

The last section gives details about the model and solver status, primal and dual feasibilities, as well as solver resource times. Furthermore, the log gives information about the basis identification phase. Some of this information is listed in the GAMS solve summary in the model listing (.LST) file as well.

The fields in the main MOSEK log output are:

Log Using the Simplex Optimizer

Below is a log output running the TRNSPORT model from the GAMS Model Library using the MOSEK simplex optimizer.

Reading parameter(s) from "mosek.opt"
>>  MSK_IPAR_OPTIMIZER MSK_OPTIMIZER_DUAL_SIMPLEX
Finished reading from "mosek.opt"

Optimizer started.
Simplex optimizer started.
Presolve started.
Linear dependency checker started.
Linear dependency checker terminated.
Eliminator - tries                  : 0                 time                   : 0.00
Lin. dep.  - tries                  : 1                 time                   : 0.00
Lin. dep.  - number                 : 0
Presolve terminated. Time: 0.00
Dual simplex optimizer started.
Dual simplex optimizer setup started.
Dual simplex optimizer setup terminated.
Optimizer  - solved problem         : the primal
Optimizer  - Constraints            : 5
Optimizer  - Scalar variables       : 6                 conic                  : 0
Optimizer  - hotstart               : no
ITER      DEGITER(%)  PFEAS       DFEAS       POBJ                  DOBJ                  TIME      TOTTIME 
0         0.00        NA          0.00e+00    NA                    0.0000000000e+00      0.00      0.00
4         20.00       NA          0.00e+00    NA                    1.5367501014e+02      0.00      0.00
Dual simplex optimizer terminated.
Simplex optimizer terminated. Time: 0.00. 

Optimizer terminated. Time: 0.00

Basic solution summary
  Problem status  : PRIMAL_AND_DUAL_FEASIBLE
  Solution status : OPTIMAL
  Primal.  obj: 1.5367500000e+02    nrm: 6e+02    Viol.  con: 0e+00    var: 0e+00
  Dual.    obj: 1.5367500000e+02    nrm: 2e-01    Viol.  con: 0e+00    var: 0e+00

Return code - 0 [MSK_RES_OK]: No error occurred.

The fields in the main MOSEK log output are:

Log Using the Mixed Integer Optimizer

Below is a log output running the model CUBE from the GAMS model library using the MOSEK mixed-integer optimizer.

Optimizer started.
Mixed integer optimizer started.
Threads used: 1
Presolve started.
Presolve terminated. Time = 0.00
Presolved problem: 76 variables, 99 constraints, 419 non-zeros
Presolved problem: 0 general integer, 27 binary, 49 continuous
Clique table size: 0
BRANCHES RELAXS   ACT_NDS  DEPTH    BEST_INT_OBJ         BEST_RELAX_OBJ       REL_GAP(%)  TIME
0        1        0        0        NA                   0.0000000000e+00     NA          0.0
0        1        0        0        6.0000000000e+00     0.0000000000e+00     100.00      0.0
Cut generation started.
0        2        0        0        6.0000000000e+00     0.0000000000e+00     100.00      0.0
Cut generation terminated. Time = 0.00
15       18       1        0        6.0000000000e+00     0.0000000000e+00     100.00      0.0
31       34       1        0        4.0000000000e+00     0.0000000000e+00     100.00      0.0
53       56       1        0        4.0000000000e+00     0.0000000000e+00     100.00      0.0
83       86       1        0        4.0000000000e+00     0.0000000000e+00     100.00      0.0
98       101      16       8        4.0000000000e+00     0.0000000000e+00     100.00      0.0
114      117      28       9        4.0000000000e+00     0.0000000000e+00     100.00      0.0
142      145      44       6        4.0000000000e+00     0.0000000000e+00     100.00      0.1
175      177      63       7        4.0000000000e+00     0.0000000000e+00     100.00      0.1
208      210      84       12       4.0000000000e+00     0.0000000000e+00     100.00      0.1
245      247      103      4        4.0000000000e+00     0.0000000000e+00     100.00      0.1
278      279      124      5        4.0000000000e+00     0.0000000000e+00     100.00      0.1
309      310      147      14       4.0000000000e+00     0.0000000000e+00     100.00      0.1
347      345      165      10       4.0000000000e+00     3.3333333333e-01     91.67       0.1
A near optimal solution satisfying the absolute gap tolerance of 0.00e+00 has been located.

Objective of best integer solution : 4.000000000000e+00
Best objective bound               : 3.333333333333e-01
Construct solution objective       : Not employed
Construct solution # roundings     : 0
User objective cut value           : 0
Number of cuts generated           : 3
  Number of Gomory cuts            : 3
Number of branches                 : 347
Number of relaxations solved       : 345
Number of interior point iterations: 6
Number of simplex iterations       : 3460
Time spend presolving the root     : 0.00
Time spend in the heuristic        : 0.00
Time spend in the sub optimizers   : 0.00
  Time spend optimizing the root   : 0.00
Mixed integer optimizer terminated. Time: 0.12

Optimizer terminated. Time: 0.12

Integer solution solution summary
  Problem status  : PRIMAL_FEASIBLE
  Solution status : NEAR_INTEGER_OPTIMAL
  Primal.  obj: 4.0000000000e+00    nrm: 1e+01    Viol.  con: 1e+01    var: 1e+00    itg: 3e-16

Return code - 10004 [MSK_RES_TRM_MIO_NEAR_ABS_GAP]: The mixed-integer optimizer terminated because the near optimal absolute gap tolerance was satisfied.

The fields in the main MOSEK log output are:

The log then gives information about solving the model with discrete variables fixed in order to determine marginals. Option SOLVEFINAL can be used to disable this step. The fixed problem is solved as a regular LP with warm start information. So the log looks identical to the MOSEK simplex optimizer for linear programs:

Solving fixed problem...
[...]
Optimizer started.
Simplex optimizer started.
Presolve started.
Eliminator - tries                  : 0                 time                   : 0.00
Lin. dep.  - tries                  : 0                 time                   : 0.00
Lin. dep.  - number                 : 0
Presolve terminated. Time: 0.00
Simplex optimizer terminated. Time: 0.00. 

Optimizer terminated. Time: 0.00

Basic solution summary
  Problem status  : PRIMAL_AND_DUAL_FEASIBLE
  Solution status : OPTIMAL
  Primal.  obj: 4.0000000000e+00    nrm: 1e+01    Viol.  con: 4e-16    var: 0e+00
  Dual.    obj: 4.0000000000e+00    nrm: 5e+00    Viol.  con: 0e+00    var: 0e+00

Return code - 0 [MSK_RES_OK]: No error occurred.

MIP Solution:            4.000000   (3466 iterations, 347 nodes)
Final Solve:             4.000000   (0 iterations)

Best possible:           0.333333
Absolute gap:            3.666667
Relative gap:            0.916667

Semidefinite Programming with GAMS/MOSEK (experimental)

With Version 7, MOSEK introduced a semidefinite programming (SDP) solver into their portfolio. The following introduction is taken from the Mosek Documentation.

Semidefinite programming is a generalization of quadratic conic programming, allowing the use of matrix variables belonging to the convex cone of positive semidefinite matrices

\[ \mathcal{S}_{r}^{+} = \left\{ X \in\mathcal{S}_{r}: z^{T} X z \geq 0, \;\forall z \in\mathbb{R}^{r} \right\}, \]

where \(\mathcal{S}_{r}\) is the set of \(r \times r\) real-valued symmetric matrices. MOSEK can solve semidefinite optimization problems of the form

\[ \tag{SDP} \begin{array}{lcclcll} \displaystyle{} \mbox{minimize}&\displaystyle{} &\displaystyle{} &\displaystyle{} \sum_{j=0}^{n-1} c_{j} x_{j} + \sum_{j=0}^{p-1}\left\langle{}\overline{C}_{j}, \overline{X}_{j}\right\rangle{} + c^{f} &\displaystyle{} &\displaystyle{} &\displaystyle{} \\[0pt] \displaystyle{} \mbox{subject to}&\displaystyle{} l_{i}^{c} &\displaystyle{} \leq{}&\displaystyle{} \sum_{j=0}^{n-1} a_{ij} x_{j} + \sum_{j=0}^{p-1}\left\langle{}\overline{A}_{ij}, \overline{X}_{j}\right\rangle{}&\displaystyle{} \leq{}&\displaystyle{} u_{i}^{c}, &\displaystyle{} i = 0, \ldots{}, m-1, \\[0pt] \displaystyle{} &\displaystyle{} l_{j}^{x} &\displaystyle{} \leq{}&\displaystyle{} x_{j} &\displaystyle{} \leq{}&\displaystyle{} u_{j}^{x}, &\displaystyle{} j = 0, \ldots{}, n-1, \\[0pt] \displaystyle{} &\displaystyle{} &\displaystyle{} &\displaystyle{} x \in{}\mathcal{C}, \overline{X}_{j} \in{}\mathcal{S}_{r_{j}}^{+}, &\displaystyle{} &\displaystyle{} &\displaystyle{} j = 0, \ldots{}, p-1, \\[0pt] \end{array} \]

where the problem has \(p\) symmetric positive semidefinite (PSD) variables \(\overline{X}_{j}\in\mathcal{S}_{r_j}^{+}\) of dimension \(r_j\) with symmetric coefficient matrices \(\overline{C}_{j}\in{}\mathcal{S}_{r_{j}}\) and \(\overline{A}_{i,j}\in{}\mathcal{S}_{r_{j}}\). We use the standard notation for the matrix inner product, i.e., for \(A,B\in{}\mathbb{R}^{m{\times} n}\) we have

\[ \left\langle{}A,B\right\rangle{} := \sum_{i=0}^{m-1}\sum_{j=0}^{n-1} A_{ij} B_{ij}. \]

Example

An example for a mixed semidefinite and conic quadratic programming problem with a 3-dimensional PSD matrix variable is the following:

\begin{align*} \text{minimize}\; & \left\langle \begin{bmatrix} 2 & 1 & 0 \\ 1 & 2 & 1 \\ 0 & 1 & 2 \end{bmatrix}, \bar X \right\rangle + x_0 \\ \text{subject to}\; & \left\langle \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix}, \bar X \right\rangle + x_0 = 1 \\ & \left\langle \begin{bmatrix} 1 & 1 & 1 \\ 1 & 1 & 1 \\ 1 & 1 & 1 \end{bmatrix}, \bar X \right\rangle + x_1 + x_2 = \frac{1}{2} \\ & x_0 \geq \sqrt{x_1^2 + x_2^2} \\ & \bar X \succeq 0 \end{align*}

The GAMS/MOSEK interface offers an experimental interface to MOSEK's SDP solver. It allows to state SDP's of the form (SDP) in GAMS language. For instance, the example problem from above can be formulated as follows (sdp01 in the GAMS Test Library):

We see that the matrix \(\bar X\) is defined via the 2-dimensional GAMS variable barX. Additionally, the keyword PSDMATRIX at the beginning of the descriptive text (!) of the variable is used to indicate that the variables belonging to symbol barX are to be interpreted as a matrix variable with PSD constraint.

The Model shown above can be solved with MOSEK via the statements

As GAMS has no native support for conic programming, the modeltype is specified as QCP.

The solve statement produces the following log output (see also Log Using the Interior Point Optimizer):

    M O S E K       version 7.0.0.75 (Build date: 2013-7-1 19:28:43)
    Copyright (C)   MOSEK ApS, Fruebjergvej 3, Box 16
                    DK-2100 Copenhagen, Denmark
                    http://www.mosek.com

Recognizing SDP variables for symbols barX (1)
Recognized 1 quadratic equation as conic constraint.
Optimizer started.
Conic interior-point optimizer started.
Presolve started.
Linear dependency checker started.
Linear dependency checker terminated.
Eliminator - tries                  : 0                 time                   : 0.00
Eliminator - elim's                 : 0
Lin. dep.  - tries                  : 1                 time                   : 0.00
Lin. dep.  - number                 : 0
Presolve terminated. Time: 0.00
Optimizer  - threads                : 1
Optimizer  - solved problem         : the primal
Optimizer  - Constraints            : 2
Optimizer  - Cones                  : 1
Optimizer  - Scalar variables       : 3                 conic                  : 3
Optimizer  - Semi-definite variables: 1                 scalarized             : 6
Factor     - setup time             : 0.00              dense det. time        : 0.00
Factor     - ML order time          : 0.00              GP order time          : 0.00
Factor     - nonzeros before factor : 3                 after factor           : 3
Factor     - dense dim.             : 0                 flops                  : 2.88e+02
ITE PFEAS    DFEAS    GFEAS    PRSTATUS   POBJ              DOBJ              MU       TIME
0   3.0e+00  1.0e+00  8.0e+00  0.00e+00   7.000000000e+00   0.000000000e+00   1.0e+00  0.00
1   4.3e-01  1.4e-01  1.2e+00  1.67e-01   1.601234178e+00   3.103213067e-01   1.4e-01  0.00
2   6.6e-02  2.2e-02  1.8e-01  1.21e+00   8.534948745e-01   6.725260078e-01   2.2e-02  0.00
3   4.9e-03  1.6e-03  1.3e-02  1.02e+00   7.158760069e-01   7.026959645e-01   1.6e-03  0.00
4   2.8e-04  9.4e-05  7.5e-04  1.00e+00   7.063055286e-01   7.055481573e-01   9.4e-05  0.00
5   1.6e-05  5.2e-06  4.1e-05  1.00e+00   7.057440915e-01   7.057021878e-01   5.2e-06  0.00
6   8.9e-07  3.0e-07  2.4e-06  1.00e+00   7.057124546e-01   7.057100494e-01   3.0e-07  0.00
7   1.6e-08  5.4e-09  4.3e-08  1.00e+00   7.057105296e-01   7.057104862e-01   5.4e-09  0.00
Interior-point optimizer terminated. Time: 0.00. 

Optimizer terminated. Time: 0.00

Interior-point solution summary
  Problem status  : PRIMAL_AND_DUAL_FEASIBLE
  Solution status : OPTIMAL
  Primal.  obj: 7.0571052965e-01  Viol.  con: 2e-08  var: 0e+00  barvar: 0e+00  cones: 0e+00
  Dual.    obj: 7.0571048621e-01  Viol.  con: 0e+00  var: 1e-16  barvar: 0e+00  cones: 0e+00

Return code - 0 [MSK_RES_OK]: No error occurred.

Finally, the optimal value for the matrix \(\bar X\) and the dual values associated with bound constraints on entries of the matrix \(\bar X\) can be displayed in GAMS as usual: display X.l, X.m;

Usage

The general syntax for defining a symmetric matrix or an (indexed) set of symmetric matrices with PSD constraints in GAMS is

Variable X(a,b,c,...,i1,..,ik,i1,...,ik)  "PSDMATRIX_k <explanatory text>";

The number k>0 specifies the number of indices that define the row/column dimension of the matrix. Specifying only the term PSDMATRIX is equivalent to PSDMATRIX_1. For a given k, the 2k-last indices are used to index the rows and columns of the matrix. The dimensions a, b, c, ... are optional and can be used to index a set of matrices. In the GAMS/MOSEK output, a number in parenthesis is used to indicate the number of PSD matrices that have been found for one symbol. For example, the code

generates the output

Recognizing SDP variables for symbols X (42)

if all matrices X(j,.,.) also occur in a model instantiation.

Note

The current syntax for declaring PSD matrix variables via the explanatory text a variable is a temporary solution that will hopefully be replaced by a GAMS language feature in the future.

Variables that were tagged as belonging to a PSD matrix can only occur in linear constraints. Within these constraints, the coefficient matrix for a PSD matrix needs to be specified in symmetric form. That is, if the parameter matrix barAobj in the example above is equivalently specified as

the GAMS/MOSEK interface will quit with the error message

SDP coefficient matrix not symmetric: defObj_z: 2*barX(1,0) != 0*barX(0,1)

Bounds on entries in a PSD matrix variable can be specified as usual with .lo and .up attributes. These bound constraints are translated into linear constraints by the interface. If different bounds are given to symmetric entries of a PSD matrix variable ( \(\bar X_{i,j}\) vs. \(\bar X_{j,i}\)), the stronger bounds are used, which is equivalent to adding constraints for each of the matrix entries. For nonpositive lower bounds on diagonal entries, no extra constraints are added, as they are implied by the PSD constraint.

GAMS variables vs. PSD matrix entries

As GAMS is not aware that a indexed variable will be interpreted as PSD matrix variable, it may not generate variables for all matrix entries when instantiating a model. This is especially critical if the modeler specified bounds on matrix entries that do not appear in any of the model equations, as these bounds would not be visible to the GAMS/MOSEK interface.

As an example, consider the SDP relaxation \(\max\{-\langle W,\bar X\rangle : X_{i,i} = 1 \;\forall i, X\succeq 0\}\) corresponding to the Goemans-Williamson Randomized Approximation Algorithm for MaxCut:

As W(i,i)=0, GAMS will not create any variables for X(i,i) when instantiating the model in a solve statement. As a consequence, the constraints X(i,i)=1 will not be visible to GAMS/MOSEK. However, as MOSEK will compute values for the full matrix \(X\), it will also compute values for the diagonal entries of X.

Further, the GAMS/MOSEK interface can pass primal solution values only for those entries of a PSD matrix variable that have a corresponding GAMS variable. This may make a solution matrix appear not to be PSD in GAMS, because not all entries have been passed back.

To be aware of such problems, the GAMS/MOSEK interface checks that it has a GAMS variable available for every entry of a PSD matrix variable. If not, it will report an error like

ERROR: Have 1600 GAMS variables for entries of 400 x 400 PSD Variable X(,), expected 160000 many.

This check can be disabled by setting the option SDPCHECKVARS to 0. Note, that the check is not able to alarm the user in situations where no GAMS variables were created for all entries of a row and corresponding column.

A simple workaround for this issue is to force all variables to be generated when a model is instantiated. This can be done by adding something like eps*sum((i,j),X(i,j)) to one of the equations. Note, that eps is numerically equal to a 0.0 in GAMS, but has the effect that the term sum((i,j),X(i,j)) is passed to the solver with 0.0-coefficient.

Dual Values for PSD constraints

The PSD constraint on a matrix variable \(\bar X\) is associated with a dual PSD matrix variable \(\bar Y\). As GAMS is not aware of the PSD constraints, it is also not aware of the corresponding dual variables. Thus, there is no native way to pass the duals for the PSD constraints back to GAMS. (Recall, that the marginals for a PSDMATRIX variable X are used to store the dual values associated with the bound constraints on matrix entries.)

To work around this issue, the GAMS/MOSEK interface offers the option SDPSOLUFILE. This option allows to specify the name of a GDX file that stores primal values for all entries of a matrix variable and dual values of the corresponding PSD constraint. For a variable X(i,j), the GDX file stores the primal matrix value for all entries \((i,j)\) (i.e., not just the ones for which GAMS variables were created, therefor offering another workaround for the issue discussed in the previous section) as level values of variable X and the dual matrix for the PSD constraint as marginal values.

Infeasible and Unbounded SDPs

The GAMS/Mosek link currently does not pass on certificates for primal or dual infeasibility from Mosek to GAMS if PSD variables are present.

Detailed Descriptions of MOSEK Options

FIXOPTFILE (string): Name of option file which is read just before solving the fixed problem. ↵

MSK_DPAR_ANA_SOL_INFEAS_TOL (real): If a constraint violates its bound with an amount larger than this value, the constraint name, index and violation will be printed by the solution analyzer. ↵

Default: 1e-06

MSK_DPAR_BASIS_REL_TOL_S (real): Maximum relative dual bound violation allowed in an optimal basic solution. ↵

Default: 1e-12

MSK_DPAR_BASIS_TOL_S (real): Maximum absolute dual bound violation in an optimal basic solution. ↵

Range: [1e-09, ∞]

Default: 1e-06

MSK_DPAR_BASIS_TOL_X (real): Maximum absolute primal bound violation allowed in an optimal basic solution. ↵

Range: [1e-09, ∞]

Default: 1e-06

MSK_DPAR_CHECK_CONVEXITY_REL_TOL (real): Not in use. ↵

Default: 1e-10

MSK_DPAR_DATA_SYM_MAT_TOL (real): Absolute zero tolerance for elements in in symmetric matrices. ↵

If any value in a symmetric matrix is smaller than this parameter in absolute terms MOSEK will treat the values as zero and generate a warning.

Range: [1e-16, 1e-06]

Default: 1e-12

MSK_DPAR_DATA_SYM_MAT_TOL_HUGE (real): An element in a symmetric matrix which is larger than this value in absolute size causes an error. ↵

Default: 1e+20

MSK_DPAR_DATA_SYM_MAT_TOL_LARGE (real): An element in a symmetric matrix which is larger than this value in absolute size causes a warning message to be printed. ↵

Default: 1e+10

MSK_DPAR_DATA_TOL_AIJ_HUGE (real): An element in the constraint matrix which is larger than this value in absolute size causes an error. ↵

Default: 1e+20

MSK_DPAR_DATA_TOL_AIJ_LARGE (real): An element in the constraint matrix which is larger than this value in absolute size causes a warning message. ↵

Default: 1e+10

MSK_DPAR_DATA_TOL_BOUND_INF (real): Any bound which in absolute value is greater than this parameter is considered infinite. ↵

Default: 1e+16

MSK_DPAR_DATA_TOL_BOUND_WRN (real): If a bound value is larger than this value in absolute size, then a warning message is issued. ↵

Default: 1e+08

MSK_DPAR_DATA_TOL_CJ_LARGE (real): A coefficient in the objective function which is larger than this value in absolute terms causes a warning message. ↵

Default: 1e+08

MSK_DPAR_DATA_TOL_C_HUGE (real): A coefficient in the objective function which is larger than the value in absolute terms is considered to be huge and generates an error. ↵

Default: 1e+16

MSK_DPAR_DATA_TOL_QIJ (real): Absolute zero tolerance for coefficients of quadratic terms. ↵

Default: 1e-16

MSK_DPAR_DATA_TOL_X (real): Zero tolerance for constraints and variables i.e. if the distance between the lower and upper bound is less than this value, then the lower and upper bound is considered identical. ↵

Default: 1e-08

MSK_DPAR_INTPNT_CO_TOL_DFEAS (real): Dual feasibility tolerance used by the interior-point optimizer for conic problems. ↵

Range: [0, 1]

Default: 1e-08

See also: MSK_DPAR_INTPNT_CO_TOL_NEAR_REL.

MSK_DPAR_INTPNT_CO_TOL_INFEAS (real): Infeasibility tolerance used by the interior-point optimizer for conic problems. ↵

Controls when the interior-point optimizer declares the model primal or dual infeasible. A small number means the optimizer gets more conservative about declaring the model infeasible.

Range: [0, 1]

Default: 1e-12

MSK_DPAR_INTPNT_CO_TOL_MU_RED (real): Relative complementarity gap tolerance used by the interior-point optimizer for conic problems. ↵

Range: [0, 1]

Default: 1e-08

MSK_DPAR_INTPNT_CO_TOL_NEAR_REL (real): Termination tolerance multiplier that is used if no accurate solution can be found. ↵

If MOSEK cannot compute a solution that has the prescribed accuracy, then it will multiply the termination tolerances with value of this parameter. If the solution then satisfies the termination criteria, then the solution is denoted near optimal, near feasible and so forth.

Range: [1, ∞]

Default: 1

MSK_DPAR_INTPNT_CO_TOL_PFEAS (real): Primal feasibility tolerance used by the interior-point optimizer for conic problems. ↵

Range: [0, 1]

Default: 1e-08

See also: MSK_DPAR_INTPNT_CO_TOL_NEAR_REL.

MSK_DPAR_INTPNT_CO_TOL_REL_GAP (real): Relative gap termination tolerance used by the interior-point optimizer for conic problems. ↵

Range: [0, 1]

Default: 1e-08

See also: MSK_DPAR_INTPNT_CO_TOL_NEAR_REL.

MSK_DPAR_INTPNT_QO_TOL_DFEAS (real): Dual feasibility tolerance used when the interior-point optimizer is applied to a quadratic optimization problem. ↵

Range: [0, 1]

Default: 1e-08

See also: MSK_DPAR_INTPNT_QO_TOL_NEAR_REL.

MSK_DPAR_INTPNT_QO_TOL_INFEAS (real): Infeasibility tolerance used by the interior-point optimizer for quadratic problems. ↵

Controls when the interior-point optimizer declares the model primal or dual infeasible. A small number means the optimizer gets more conservative about declaring the model infeasible.

Range: [0, 1]

Default: 1e-12

MSK_DPAR_INTPNT_QO_TOL_MU_RED (real): Relative complementarity gap tolerance used by the interior-point optimizer for quadratic problems. ↵

Range: [0, 1]

Default: 1e-08

MSK_DPAR_INTPNT_QO_TOL_NEAR_REL (real): Termination tolerance multiplier that is used if no accurate solution can be found. ↵

If MOSEK cannot compute a solution that has the prescribed accuracy, then it will multiply the termination tolerances with value of this parameter. If the solution then satisfies the termination criteria, then the solution is denoted near optimal, near feasible and so forth.

Range: [1, ∞]

Default: 1

MSK_DPAR_INTPNT_QO_TOL_PFEAS (real): Primal feasibility tolerance used by the interior-point optimizer for quadratic problems. ↵

Range: [0, 1]

Default: 1e-08

See also: MSK_DPAR_INTPNT_QO_TOL_NEAR_REL.

MSK_DPAR_INTPNT_QO_TOL_REL_GAP (real): Relative gap termination tolerance used by the interior-point optimizer for quadratic problems. ↵

Range: [0, 1]

Default: 1e-08

See also: MSK_DPAR_INTPNT_QO_TOL_NEAR_REL.

MSK_DPAR_INTPNT_TOL_DFEAS (real): Dual feasibility tolerance used by the interior-point optimizer for linear problems. ↵

Range: [0, 1]

Default: 1e-08

MSK_DPAR_INTPNT_TOL_DSAFE (real): Controls the initial dual starting point used by the interior-point optimizer. ↵

If the interior-point optimizer converges slowly and/or the constraint or variable bounds are very large, then it might be worthwhile to increase this value.

Range: [0.0001, ∞]

Default: 1

MSK_DPAR_INTPNT_TOL_INFEAS (real): Infeasibility tolerance used by the interior-point optimizer for linear problems. ↵

Controls when the interior-point optimizer declares the model primal or dual infeasible. A small number means the optimizer gets more conservative about declaring the model infeasible.

Range: [0, 1]

Default: 1e-10

MSK_DPAR_INTPNT_TOL_MU_RED (real): Relative complementarity gap tolerance used by the interior-point optimizer for linear problems. ↵

Range: [0, 1]

Default: 1e-16

MSK_DPAR_INTPNT_TOL_PATH (real): Controls how close the interior-point optimizer follows the central path. ↵

A large value of this parameter means the central path is followed very closely. On numerically unstable problems it may be worthwhile to increase this parameter.

Range: [0, 0.9999]

Default: 1e-08

MSK_DPAR_INTPNT_TOL_PFEAS (real): Primal feasibility tolerance used by the interior-point optimizer for linear problems. ↵

Range: [0, 1]

Default: 1e-08

MSK_DPAR_INTPNT_TOL_PSAFE (real): Controls the initial primal starting point used by the interior-point optimizer. ↵

If the interior-point optimizer converges slowly and/or the constraint or variable bounds are very large, then it may be worthwhile to increase this value.

Range: [0.0001, ∞]

Default: 1

MSK_DPAR_INTPNT_TOL_REL_GAP (real): Relative gap termination tolerance used by the interior-point optimizer for linear problems. ↵

Range: [1e-14, ∞]

Default: 1e-08

MSK_DPAR_INTPNT_TOL_REL_STEP (real): Relative step size to the boundary for linear and quadratic optimization problems. ↵

Range: [0.0001, 0.999999]

Default: 0.9999

MSK_DPAR_INTPNT_TOL_STEP_SIZE (real): Step size tolerance. ↵

If the step size falls below the value of this parameter, then the interior-point optimizer assumes it is stalled. In other words the interior-point optimizer does not make any progress and therefore it is better stop.

Range: [0, 1]

Default: 1e-06

MSK_DPAR_LOWER_OBJ_CUT (real): Lower objective limit. ↵

If either a primal or dual feasible solution is found proving that the optimal objective value is outside, the interval (MSK_DPAR_LOWER_OBJ_CUT, MSK_DPAR_UPPER_OBJ_CUT), then MOSEK is terminated.

Range: [-∞, ∞]

Default: -1e+30

See also: MSK_DPAR_LOWER_OBJ_CUT_FINITE_TRH.

MSK_DPAR_LOWER_OBJ_CUT_FINITE_TRH (real): Lower objective limit threshold. ↵

If the lower objective cut (MSK_DPAR_LOWER_OBJ_CUT) is less than this value, then it is treated as infinity.

Range: [-∞, ∞]

Default: -5e+29

MSK_DPAR_MIO_MAX_TIME (real): This parameter limits the maximum time spent by the mixed-integer optimizer. ↵

A negative number means infinity.

Range: [-∞, ∞]

Default: -1

MSK_DPAR_MIO_REL_GAP_CONST (real): This value is used to compute the relative gap for the solution to an integer optimization problem. ↵

Range: [1e-15, ∞]

Default: 1e-10

MSK_DPAR_MIO_TOL_ABS_GAP (real): Absolute optimality tolerance employed by the mixed-integer optimizer. ↵

Default: GAMS OptCA

MSK_DPAR_MIO_TOL_ABS_RELAX_INT (real): Absolute relaxation tolerance of the integer constraints. ↵

That means if the fractional part of a discrete variable \((\min(|x|-\lfloor x \rfloor,\lceil x \rceil - |x|))\) is less than the tolerance, then the integer restriction is assumed to be satisfied.

Range: [1e-09, ∞]

Default: 1e-05

MSK_DPAR_MIO_TOL_FEAS (real): Feasibility tolerance for mixed integer solver. ↵

Range: [1e-09, 0.001]

Default: 1e-06

MSK_DPAR_MIO_TOL_REL_DUAL_BOUND_IMPROVEMENT (real): If the relative improvement of the dual bound is smaller than this value, the solver will terminate the root cut generation. ↵

A value of 0.0 means that the value is selected automatically.

Range: [0, 1]

Default: 0

MSK_DPAR_MIO_TOL_REL_GAP (real): Relative optimality tolerance employed by the mixed-integer optimizer. ↵

Default: GAMS OptCR

MSK_DPAR_OPTIMIZER_MAX_TIME (real): Maximum amount of time the optimizer is allowed to spent on the optimization. ↵

A negative number means infinity.

Range: [-∞, ∞]

Default: GAMS ResLim

MSK_DPAR_PRESOLVE_TOL_ABS_LINDEP (real): Absolute tolerance employed by the linear dependency checker. ↵

Default: 1e-06

MSK_DPAR_PRESOLVE_TOL_AIJ (real): Absolute zero tolerance employed for constraint coefficients in presolve. ↵

Range: [1e-15, ∞]

Default: 1e-12

MSK_DPAR_PRESOLVE_TOL_PRIMAL_INFEAS_PERTURBATION (real): The presolve is allowed to perturb a bound on a constraint or variable by this amount if it removes an infeasibility. ↵

Default: 1e-06

MSK_DPAR_PRESOLVE_TOL_REL_LINDEP (real): Relative tolerance employed by the linear dependency checker. ↵

Default: 1e-10

MSK_DPAR_PRESOLVE_TOL_S (real): Absolute zero tolerance employed for dual variables in presolve. ↵

Default: 1e-08

MSK_DPAR_PRESOLVE_TOL_X (real): Absolute zero tolerance employed for primal variables in presolve. ↵

Default: 1e-08

MSK_DPAR_QCQO_REFORMULATE_REL_DROP_TOL (real): This parameter determines when columns are dropped in incomplete Cholesky factorization during reformulation of quadratic problems. ↵

Default: 1e-15

MSK_DPAR_SEMIDEFINITE_TOL_APPROX (real): Tolerance to define a matrix to be positive semidefinite. ↵

Range: [1e-15, ∞]

Default: 1e-10

MSK_DPAR_SIMPLEX_ABS_TOL_PIV (real): Absolute pivot tolerance employed by the simplex optimizers. ↵

Range: [1e-12, ∞]

Default: 1e-07

MSK_DPAR_SIM_LU_TOL_REL_PIV (real): Relative pivot tolerance for LU factorization in simplex optimizers and basis identification. ↵

A value closer to 1.0 generally improves numerical stability but typically also implies an increase in the computational work.

Range: [1e-06, 0.999999]

Default: 0.01

MSK_DPAR_UPPER_OBJ_CUT (real): Upper objective limit. ↵

If either a primal or dual feasible solution is found proving that the optimal objective value is outside the interval (MSK_DPAR_LOWER_OBJ_CUT, MSK_DPAR_UPPER_OBJ_CUT), then MOSEK is terminated.

Range: [-∞, ∞]

Default: 1e+30

See also: MSK_DPAR_UPPER_OBJ_CUT_FINITE_TRH.

MSK_DPAR_UPPER_OBJ_CUT_FINITE_TRH (real): Upper objective limit threshold. ↵

If the upper objective cut MSK_DPAR_UPPER_OBJ_CUT is greater than this value, then it is treated as infinity.

Range: [-∞, ∞]

Default: 5e+29

MSK_IPAR_AUTO_SORT_A_BEFORE_OPT (string): Controls whether the elements in each column of the coefficient matrix are sorted before an optimization is performed. ↵

This is not required but makes the optimization insusceptible to reorderings of variables.

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_BI_CLEAN_OPTIMIZER (string): Controls which simplex optimizer is used in the clean-up phase. ↵

Anything else than primal or dual simplex is equivalent to free simplex.

Default: MSK_OPTIMIZER_FREE

value	meaning
MSK_OPTIMIZER_FREE	The optimizer is chosen automatically.
MSK_OPTIMIZER_PRIMAL_SIMPLEX	The primal simplex optimizer is used.
MSK_OPTIMIZER_DUAL_SIMPLEX	The dual simplex optimizer is used.
MSK_OPTIMIZER_FREE_SIMPLEX	One of the simplex optimizers is used.

MSK_IPAR_BI_IGNORE_MAX_ITER (string): Controls if basis identification is performed under certain conditions. ↵

If the parameter MSK_IPAR_INTPNT_BASIS has the value MSK_BI_NO_ERROR and the interior-point optimizer has terminated due to maximum number of iterations, then basis identification is performed if this parameter has the value MSK_ON.

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_BI_IGNORE_NUM_ERROR (string): Turns on basis identification if interior-point optimizer is terminated due to a numerical problem. ↵

If the parameter MSK_IPAR_INTPNT_BASIS has the value MSK_BI_NO_ERROR and the interior-point optimizer has terminated due to a numerical problem, then basis identification is performed if this parameter has the value MSK_ON.

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_BI_MAX_ITERATIONS (integer): Controls the maximum number of simplex iterations allowed to optimize a basis after the basis identification. ↵

Default: 1000000

MSK_IPAR_INFEAS_PREFER_PRIMAL (string): If both certificates of primal and dual infeasibility are supplied then only the primal is used when this option is turned on. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_INFEAS_REPORT_AUTO (string): Controls whether an infeasibility report is automatically produced after the optimization if the problem is primal or dual infeasible. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_INFEAS_REPORT_LEVEL (integer): Controls the amount of information presented in an infeasibility report. ↵

Higher values imply more information.

Default: 1

MSK_IPAR_INTPNT_BASIS (string): Controls whether the interior-point optimizer also computes an optimal basis. ↵

Default: MSK_BI_ALWAYS

value	meaning
MSK_BI_NEVER	Never do basis identification.
MSK_BI_ALWAYS	Basis identification is always performed even if the interior-point optimizer terminates abnormally.
MSK_BI_NO_ERROR	Basis identification is performed if the interior-point optimizer terminates without an error.
MSK_BI_IF_FEASIBLE	Basis identification is not performed if the interior-point optimizer terminates with a problem status saying that the problem is primal or dual infeasible.

See also: MSK_IPAR_BI_CLEAN_OPTIMIZER, MSK_IPAR_BI_IGNORE_MAX_ITER, MSK_IPAR_BI_IGNORE_NUM_ERROR, MSK_IPAR_BI_MAX_ITERATIONS.

MSK_IPAR_INTPNT_DIFF_STEP (string): Controls whether different step sizes are allowed in the primal and dual space. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_INTPNT_MAX_ITERATIONS (integer): Controls the maximum number of iterations allowed in the interior-point optimizer. ↵

Default: GAMS IterLim

MSK_IPAR_INTPNT_MAX_NUM_COR (integer): Controls the maximum number of correctors allowed by the multiple corrector procedure. ↵

A negative value means that MOSEK is making the choice.

Range: {-1, ..., ∞}

Default: -1

MSK_IPAR_INTPNT_MAX_NUM_REFINEMENT_STEPS (integer): Maximum number of steps to be used by the iterative refinement of the search direction. ↵

A negative value implies that the optimizer chooses the maximum number of iterative refinement steps.

Range: {-∞, ..., ∞}

Default: -1

MSK_IPAR_INTPNT_OFF_COL_TRH (integer): Controls how aggressively offending columns are detected in the Jacobian of the constraint matrix. ↵

0 means no detection, 1 means aggressive detection, and higher values mean less aggressive detection.

Default: 40

MSK_IPAR_INTPNT_ORDER_GP_NUM_SEEDS (integer): The GP ordering is dependent on a random seed. ↵

Therefore, trying several random seeds may lead to a better ordering. This parameter controls the number of random seeds tried. A value of 0 means that MOSEK makes the choice.

Default: 0

MSK_IPAR_INTPNT_ORDER_METHOD (string): Controls the ordering strategy used by the interior-point optimizer when factorizing the Newton equation system. ↵

Default: MSK_ORDER_METHOD_FREE

value	meaning
MSK_ORDER_METHOD_FREE	The ordering method is chosen automatically.
MSK_ORDER_METHOD_APPMINLOC	Approximate minimum local fill-in ordering is employed.
MSK_ORDER_METHOD_TRY_GRAPHPAR	Always try the graph partitioning based ordering.
MSK_ORDER_METHOD_FORCE_GRAPHPAR	Always use the graph partitioning based ordering even if it is worse than the approximate minimum local fill ordering.
MSK_ORDER_METHOD_NONE	No ordering is used. Note using this value almost always leads to a significantly slow down.

MSK_IPAR_INTPNT_REGULARIZATION_USE (string): Controls whether regularization is allowed. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_INTPNT_SCALING (string): Controls how the problem is scaled before the interior-point optimizer is used. ↵

Default: MSK_SCALING_FREE

value	meaning
MSK_SCALING_FREE	The optimizer chooses the scaling heuristic.
MSK_SCALING_NONE	No scaling is performed.

MSK_IPAR_INTPNT_SOLVE_FORM (string): Controls whether the primal or the dual problem is solved. ↵

Default: MSK_SOLVE_FREE

value	meaning
MSK_SOLVE_FREE	The optimizer is free to solve either the primal or the dual problem.
MSK_SOLVE_PRIMAL	The optimizer should solve the primal problem.
MSK_SOLVE_DUAL	The optimizer should solve the dual problem.

MSK_IPAR_INTPNT_STARTING_POINT (string): Starting point used by the interior-point optimizer. ↵

Default: MSK_STARTING_POINT_FREE

value	meaning
MSK_STARTING_POINT_FREE	The starting point is chosen automatically.
MSK_STARTING_POINT_GUESS	The optimizer guesses a starting point.
MSK_STARTING_POINT_CONSTANT	The optimizer constructs a starting point by assigning a constant value to all primal and dual variables. This starting point is normally robust.

MSK_IPAR_LOG (integer): Controls the amount of log information. ↵

The value 0 implies that all log information is suppressed. A higher level implies that more information is logged.

Default: 10

MSK_IPAR_LOG_BI (integer): Controls the amount of output printed by the basis identification procedure. ↵

A higher level implies that more information is logged.

Default: 1

MSK_IPAR_LOG_BI_FREQ (integer): Controls logging frequency of the basis identification ↵

Default: 2500

MSK_IPAR_LOG_FEAS_REPAIR (integer): Controls the amount of output printed when performing feasibility repair. ↵

Default: 1

MSK_IPAR_LOG_INFEAS_ANA (integer): Controls amount of output printed by the infeasibility analyzer procedures. ↵

A higher level implies that more information is logged.

Default: 1

MSK_IPAR_LOG_INTPNT (integer): Controls amount of output printed by the interior-point optimizer. ↵

A higher level implies that more information is logged.

Default: 1

MSK_IPAR_LOG_MIO (integer): Controls the log level for the mixed-integer optimizer. ↵

A higher level implies that more information is logged.

Default: 4

MSK_IPAR_LOG_MIO_FREQ (integer): Controls how frequent the mixed-integer optimizer prints the log line. ↵

It will print a line every time MSK_INTPAR_LOG_MIO_FREQ relaxations have been solved.

Range: {-∞, ..., ∞}

Default: 10

MSK_IPAR_LOG_ORDER (integer): If turned on, then factor lines are added to the log. ↵

Default: 1

MSK_IPAR_LOG_PRESOLVE (integer): Controls amount of output printed by the presolve procedure. ↵

A higher level implies that more information is logged.

Default: 1

MSK_IPAR_LOG_RESPONSE (integer): Controls amount of output printed when response codes are reported. ↵

A higher level implies that more information is logged.

Default: 0

MSK_IPAR_LOG_SIM (integer): Controls amount of output printed by the simplex optimizer. ↵

A higher level implies that more information is logged.

Default: 4

MSK_IPAR_LOG_SIM_FREQ (integer): Controls simplex optimizer logging frequency. ↵

Default: 1000

MSK_IPAR_LOG_STORAGE (integer): When turned on, MOSEK prints messages regarding the storage usage and allocation. ↵

Default: 0

MSK_IPAR_MIO_BRANCH_DIR (string): Controls whether the mixed-integer optimizer is branching up or down by default. ↵

Default: MSK_BRANCH_DIR_FREE

value	meaning
MSK_BRANCH_DIR_FREE	The mixed-integer optimizer decides which branch to choose.
MSK_BRANCH_DIR_UP	The mixed-integer optimizer always chooses the up branch first.
MSK_BRANCH_DIR_DOWN	The mixed-integer optimizer always chooses the down branch first.
MSK_BRANCH_DIR_NEAR	Branch in direction nearest to selected fractional variable.
MSK_BRANCH_DIR_FAR	Branch in direction farthest from selected fractional variable.
MSK_BRANCH_DIR_ROOT_LP	Chose direction based on root lp value of selected variable.
MSK_BRANCH_DIR_GUIDED	Branch in direction of current incumbent.
MSK_BRANCH_DIR_PSEUDOCOST	Branch based on the pseudocost of the variable.

MSK_IPAR_MIO_CONIC_OUTER_APPROXIMATION (string): If this option is turned on outer approximation is used when solving relaxations of conic problems; otherwise interior point is used. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_MIO_CONSTRUCT_SOL (string): Whether to construct an initial solution from starting point ↵

If enabled and all integer variables have been given a value for which a feasible mixed integer solution exists, then MOSEK generates an initial solution to the mixed integer problem by fixing all integer values and solving the remaining problem.

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_MIO_CUT_CLIQUE (string): Controls whether clique cuts should be generated. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_MIO_CUT_CMIR (string): Controls whether mixed integer rounding cuts should be generated. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_MIO_CUT_GMI (string): Controls whether GMI cuts should be generated. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_MIO_CUT_IMPLIED_BOUND (string): Controls whether implied bound cuts should be generated. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_MIO_CUT_KNAPSACK_COVER (string): Controls whether knapsack cover cuts should be generated. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_MIO_CUT_LIPRO (string): Controls whether lift-and-project cuts should be generated. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_MIO_CUT_SELECTION_LEVEL (integer): Controls how aggressively generated cuts are selected to be included in the relaxation. ↵

Default: -1

value	meaning
-1	The optimizer chooses the level of cut selection.
0	Generated cuts less likely to be added to the relaxation.
1	Cuts are more aggressively selected to be included in the relaxation.

MSK_IPAR_MIO_DATA_PERMUTATION_METHOD (string): Controls what problem data permutation method is appplied to mixed-integer problems. ↵

Default: MSK_MIO_DATA_PERMUTATION_METHOD_NONE

value	meaning
MSK_MIO_DATA_PERMUTATION_METHOD_NONE	No problem data permutation is applied.
MSK_MIO_DATA_PERMUTATION_METHOD_CYCLIC_SHIFT	A random cyclic shift is applied to permute the problem data.
MSK_MIO_DATA_PERMUTATION_METHOD_RANDOM	A random permutation is applied to the problem data.

MSK_IPAR_MIO_DUAL_RAY_ANALYSIS_LEVEL (string): Controls the amount of symmetry detection and handling employed by the mixed-integer optimizer in presolve. ↵

Default: -1

value	meaning
-1	The optimizer chooses the level of dual ray analysis employed.
0	Dual ray analysis is disabled.
1	A lower amount of dual ray analysis is employed.
2	A higher amount of dual ray analysis is employed.

MSK_IPAR_MIO_FEASPUMP_LEVEL (string): Controls the way the Feasibility Pump heuristic is employed by the mixed-integer optimizer. ↵

Default: -1

value	meaning
-1	The optimizer chooses how the Feasibility Pump is used.
0	The Feasibility Pump is disabled.
1	The Feasibility Pump is enabled with an effort to improve solution quality.
2	The Feasibility Pump is enabled with an effort to reach feasibility early.

MSK_IPAR_MIO_HEURISTIC_LEVEL (integer): Controls the heuristic employed by the mixed-integer optimizer to locate an initial good integer feasible solution. ↵

A value of zero means the heuristic is not used at all. A larger value than \(0\) means that a gradually more sophisticated heuristic is used which is computationally more expensive. A negative value implies that the optimizer chooses the heuristic. Normally a value around \(3\) to \(5\) should be optimal.

Range: {-∞, ..., ∞}

Default: -1

MSK_IPAR_MIO_MAX_NUM_BRANCHES (integer): Maximum number of branches allowed during the branch and bound search. ↵

A negative value means infinite.

Range: {-∞, ..., ∞}

Default: -1

MSK_IPAR_MIO_MAX_NUM_RELAXS (integer): Maximum number of relaxations allowed during the branch and bound search. ↵

A negative value means infinite.

Range: {-∞, ..., ∞}

Default: GAMS NodLim

MSK_IPAR_MIO_MAX_NUM_RESTARTS (integer): Maximum number of restarts allowed during the branch and bound search. ↵

Default: 0

MSK_IPAR_MIO_MAX_NUM_ROOT_CUT_ROUNDS (integer): Maximum number of cut separation rounds at the root node. ↵

Default: 100

MSK_IPAR_MIO_MAX_NUM_SOLUTIONS (integer): The mixed-integer optimizer can be terminated after a certain number of different feasible solutions has been located. ↵

If this parameter has the value \(n>0\), then the mixed-integer optimizer will be terminated when \(n\) feasible solutions have been located.

Range: {-∞, ..., ∞}

Default: -1

MSK_IPAR_MIO_MEMORY_EMPHASIS_LEVEL (string): Controls how much emphasis is put on reducing memory usage. ↵

Being more conservative about memory usage may come at the cost of decreased solution speed.

Default: 0

value	meaning
0	The optimizer chooses.
1	More emphasis is put on reducing memory usage and less on speed.

MSK_IPAR_MIO_MIN_REL (integer): Number of times a variable must have been branched on for its pseudocost to be considered reliable. ↵

Default: 5

MSK_IPAR_MIO_NODE_OPTIMIZER (string): Controls which optimizer is employed at the non-root nodes in the mixed-integer optimizer. ↵

Default: MSK_OPTIMIZER_FREE

value	meaning
MSK_OPTIMIZER_FREE	The optimizer is chosen automatically.
MSK_OPTIMIZER_INTPNT	The interior-point optimizer is used.
MSK_OPTIMIZER_CONIC	The optimizer for problems having conic constraints.
MSK_OPTIMIZER_PRIMAL_SIMPLEX	The primal simplex optimizer is used.
MSK_OPTIMIZER_DUAL_SIMPLEX	The dual simplex optimizer is used.
MSK_OPTIMIZER_FREE_SIMPLEX	One of the simplex optimizers is used.
MSK_OPTIMIZER_MIXED_INT	The mixed-integer optimizer.

MSK_IPAR_MIO_NODE_SELECTION (string): Controls the node selection strategy employed by the mixed-integer optimizer. ↵

Default: MSK_MIO_NODE_SELECTION_FREE

value	meaning
MSK_MIO_NODE_SELECTION_FREE	The optimizer decides the node selection strategy.
MSK_MIO_NODE_SELECTION_FIRST	The optimizer employs a depth first node selection strategy.
MSK_MIO_NODE_SELECTION_BEST	The optimizer employs a best bound node selection strategy.
MSK_MIO_NODE_SELECTION_PSEUDO	The optimizer employs selects the node based on a pseudo cost estimate.

MSK_IPAR_MIO_NUMERICAL_EMPHASIS_LEVEL (string): Controls how much emphasis is put on reducing numerical problems possibly at the expense of solution speed. ↵

Default: 0

value	meaning
0	The optimizer chooses.
1	More emphasis is put on reducing numerical problems.
2	Even more emphasis.

MSK_IPAR_MIO_PERSPECTIVE_REFORMULATE (string): Enables or disables perspective reformulation in presolve. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_MIO_PRESOLVE_AGGREGATOR_USE (string): Controls if the aggregator should be used. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_MIO_PROBING_LEVEL (integer): Controls the amount of probing employed by the mixed-integer optimizer in presolve. ↵

Default: -1

value	meaning
-1	The optimizer chooses the level of probing employed
0	Probing is disabled
1	A low amount of probing is employed
2	A medium amount of probing is employed
3	A high amount of probing is employed

MSK_IPAR_MIO_PROPAGATE_OBJECTIVE_CONSTRAINT (string): Use objective domain propagation. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_MIO_QCQO_REFORMULATION_METHOD (string): Controls what reformulation method is applied to mixed-integer quadratic problems. ↵

Default: MSK_MIO_QCQO_REFORMULATION_METHOD_FREE

value	meaning
MSK_MIO_QCQO_REFORMULATION_METHOD_FREE	The mixed-integer optimizer decides which reformulation method to apply.
MSK_MIO_QCQO_REFORMULATION_METHOD_NONE	No reformulation method is applied.
MSK_MIO_QCQO_REFORMULATION_METHOD_LINEARIZATION	A reformulation via linearization is applied.
MSK_MIO_QCQO_REFORMULATION_METHOD_EIGEN_VAL_METHOD	The eigenvalue method is applied.
MSK_MIO_QCQO_REFORMULATION_METHOD_DIAG_SDP	A perturbation of matrix diagonals via the solution of SDPs is applied.
MSK_MIO_QCQO_REFORMULATION_METHOD_RELAX_SDP	A Reformulation based on the solution of an SDP-relaxation of the problem is applied.

MSK_IPAR_MIO_RINS_MAX_NODES (integer): Controls the maximum number of nodes allowed in each call to the RINS heuristic. ↵

The default value of -1 means that the value is determined automatically. A value of zero turns off the heuristic.

Range: {-1, ..., ∞}

Default: -1

MSK_IPAR_MIO_ROOT_OPTIMIZER (string): Controls which optimizer is employed at the root node in the mixed-integer optimizer. ↵

Default: MSK_OPTIMIZER_FREE

value	meaning
MSK_OPTIMIZER_FREE	The optimizer is chosen automatically.
MSK_OPTIMIZER_INTPNT	The interior-point optimizer is used.
MSK_OPTIMIZER_CONIC	The optimizer for problems having conic constraints.
MSK_OPTIMIZER_PRIMAL_SIMPLEX	The primal simplex optimizer is used.
MSK_OPTIMIZER_DUAL_SIMPLEX	The dual simplex optimizer is used.
MSK_OPTIMIZER_FREE_SIMPLEX	One of the simplex optimizers is used.

MSK_IPAR_MIO_ROOT_REPEAT_PRESOLVE_LEVEL (integer): Controls whether presolve can be repeated at root node. ↵

Default: -1

value	meaning
-1	The optimizer chooses whether presolve is repeated.
0	Never repeat presolve.
1	Always repeat presolve.

MSK_IPAR_MIO_SEED (integer): Sets the random seed used for randomization in the mixed integer optimizer. ↵

Selecting a different seed can change the path the optimizer takes to the optimal solution.

Default: 42

MSK_IPAR_MIO_SYMMETRY_LEVEL (string): Controls the amount of symmetry detection and handling employed by the mixed-integer optimizer in presolve. ↵

Default: -1

value	meaning
-1	The optimizer chooses the level of symmetry detection and handling employed.
0	Symmetry detection and handling is disabled.
1	A low amount of symmetry detection and handling is employed.
2	A medium amount of symmetry detection and handling is employed.
3	A high amount of symmetry detection and handling is employed .
4	An extremely high amount of symmetry detection and handling is employed.

MSK_IPAR_MIO_VAR_SELECTION (string): Controls the variable selection strategy employed by the mixed-integer optimizer. ↵

Default: MSK_MIO_VAR_SELECTION_FREE

value	meaning
MSK_MIO_VAR_SELECTION_FREE	The optimizer decides the variable selection strategy.
MSK_MIO_VAR_SELECTION_PSEUDOCOST	The optimizer employs pseudocost variable selection.
MSK_MIO_VAR_SELECTION_STRONG	The optimizer employs strong branching variable selection.

MSK_IPAR_MIO_VB_DETECTION_LEVEL (integer): Controls how much effort is put into detecting variable bounds. ↵

Default: -1

value	meaning
-1	The optimizer chooses.
0	No variable bounds are detected.
1	Only detect variable bounds that are directly represented in the problem.
2	Detect variable bounds in probing.

MSK_IPAR_NUM_THREADS (integer): Controls the number of threads employed by the optimizer. ↵

If set to 0 the number of threads used will be equal to the number of cores detected on the machine.

Default: GAMS Threads

MSK_IPAR_OPF_WRITE_HEADER (string): Write a text header with date and MOSEK version in an OPF file. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_OPF_WRITE_HINTS (string): Write a hint section with problem dimensions in the beginning of an OPF file. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_OPF_WRITE_LINE_LENGTH (integer): Aim to keep lines in OPF files not much longer than this. ↵

Default: 80

MSK_IPAR_OPF_WRITE_PARAMETERS (string): Write a parameter section in an OPF file. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_OPF_WRITE_PROBLEM (string): Write objective, constraints, bounds etc. ↵

to an OPF file.

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_OPF_WRITE_SOLUTIONS (string): Enable inclusion of solutions in the OPF files. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_OPF_WRITE_SOL_BAS (string): Whether to include basic solution in OPF files. ↵

If MSK_IPAR_OPF_WRITE_SOLUTIONS is MSK_ON and a basic solution is defined, include the basic solution in OPF files.

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_OPF_WRITE_SOL_ITG (string): Whether to include integer solution in OPF files. ↵

If MSK_IPAR_OPF_WRITE_SOLUTIONS is MSK_ON and an integer solution is defined, write the integer solution to OPF files.

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_OPF_WRITE_SOL_ITR (string): Whether to include interior solution in OPF files. ↵

If MSK_IPAR_OPF_WRITE_SOLUTIONS is MSK_ON and an interior solution is defined, write the interior solution to OPF files.

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_OPTIMIZER (string): The parameter controls which optimizer is used to optimize the task. ↵

Default: MSK_OPTIMIZER_FREE

value	meaning
MSK_OPTIMIZER_FREE	The optimizer is chosen automatically.
MSK_OPTIMIZER_INTPNT	The interior-point optimizer is used.
MSK_OPTIMIZER_CONIC	The optimizer for problems having conic constraints.
MSK_OPTIMIZER_PRIMAL_SIMPLEX	The primal simplex optimizer is used.
MSK_OPTIMIZER_DUAL_SIMPLEX	The dual simplex optimizer is used.
MSK_OPTIMIZER_FREE_SIMPLEX	One of the simplex optimizers is used.
MSK_OPTIMIZER_MIXED_INT	The mixed-integer optimizer.

MSK_IPAR_PRESOLVE_ELIMINATOR_MAX_FILL (integer): Controls the maximum amount of fill-in that can be created by one pivot in the elimination phase of presolve. ↵

A negative value means the parameter value is selected automatically.

Range: {-∞, ..., ∞}

Default: -1

MSK_IPAR_PRESOLVE_ELIMINATOR_MAX_NUM_TRIES (integer): Control the maximum number of times the eliminator is tried. ↵

A negative value implies MOSEK decides.

Range: {-∞, ..., ∞}

Default: -1

MSK_IPAR_PRESOLVE_LINDEP_ABS_WORK_TRH (integer): Controls the linear dependency check, which is potentially computationally expensive. ↵

Range: {-∞, ..., ∞}

Default: 100

MSK_IPAR_PRESOLVE_LINDEP_NEW (string): Controls whether a new experimental linear dependency checker is employed. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_PRESOLVE_LINDEP_REL_WORK_TRH (integer): Controls the linear dependency check, which is potentially computationally expensive. ↵

Range: {-∞, ..., ∞}

Default: 100

MSK_IPAR_PRESOLVE_LINDEP_USE (string): Controls whether the linear constraints are checked for linear dependencies. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_PRESOLVE_MAX_NUM_PASS (integer): Control the maximum number of times presolve passes over the problem. ↵

A negative value implies MOSEK decides.

Range: {-∞, ..., ∞}

Default: -1

MSK_IPAR_PRESOLVE_MAX_NUM_REDUCTIONS (integer): Controls the maximum number of reductions performed by the presolve. ↵

The value of the parameter is normally only changed in connection with debugging. A negative value implies that an infinite number of reductions are allowed.

Range: {-∞, ..., ∞}

Default: -1

MSK_IPAR_PRESOLVE_USE (string): Controls whether the presolve is applied to a problem before it is optimized. ↵

Default: MSK_PRESOLVE_MODE_FREE

value	meaning
MSK_PRESOLVE_MODE_OFF	The problem is not presolved before it is optimized.
MSK_PRESOLVE_MODE_ON	The problem is presolved before it is optimized.
MSK_PRESOLVE_MODE_FREE	It is decided automatically whether to presolve before the problem is optimized.

MSK_IPAR_PTF_WRITE_PARAMETERS (string): If enabled, then the parameters section is written. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_PTF_WRITE_SOLUTIONS (string): If enabled, then the solution section is written if any solutions are available, otherwise solution section is not written even if solutions are available. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_PTF_WRITE_TRANSFORM (string): If enabled, then constraint blocks with identifiable conic slacks are transformed into conic constraints and the slacks are eliminated. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_SIM_BASIS_FACTOR_USE (string): Controls whether an LU factorization of the basis is used in a hot-start. ↵

Forcing a refactorization sometimes improves the stability of the simplex optimizers, but in most cases there is a performance penalty.

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_SIM_DEGEN (string): Controls how aggressively degeneration is handled. ↵

Default: MSK_SIM_DEGEN_FREE

value	meaning
MSK_SIM_DEGEN_NONE	The simplex optimizer should use no degeneration strategy.
MSK_SIM_DEGEN_FREE	The simplex optimizer chooses the degeneration strategy.
MSK_SIM_DEGEN_AGGRESSIVE	The simplex optimizer should use an aggressive degeneration strategy.
MSK_SIM_DEGEN_MODERATE	The simplex optimizer should use a moderate degeneration strategy.
MSK_SIM_DEGEN_MINIMUM	The simplex optimizer should use a minimum degeneration strategy.

MSK_IPAR_SIM_DUAL_CRASH (integer): Controls whether crashing is performed in the dual simplex optimizer. ↵

In general if a basis consists of more than (100*MSK_IPAR_SIM_DUAL_CRASH) percent fixed variables, then a crash will be performed.

Default: 90

MSK_IPAR_SIM_DUAL_RESTRICT_SELECTION (integer): Controls how aggressively a restricted selection/pricing strategy is used to choose the outgoing variable in the dual simplex. ↵

The dual simplex optimizer can use a so-called restricted selection/pricing strategy to chooses the outgoing variable. Hence, if restricted selection is applied, then the dual simplex optimizer first choose a subset of all the potential outgoing variables. Next, for some time it will choose the outgoing variable only among the subset. From time to time the subset is redefined. A larger value of this parameter implies that the optimizer will be more aggressive in its restriction strategy, i.e. a value of 0 implies that the restriction strategy is not applied at all.

Range: {0, ..., 100}

Default: 50

MSK_IPAR_SIM_DUAL_SELECTION (string): Controls the choice of the incoming variable, known as the selection strategy, in the dual simplex optimizer. ↵

Default: MSK_SIM_SELECTION_FREE

value	meaning
MSK_SIM_SELECTION_FREE	The optimizer chooses the pricing strategy.
MSK_SIM_SELECTION_FULL	The optimizer uses full pricing.
MSK_SIM_SELECTION_ASE	The optimizer uses approximate steepest-edge pricing.
MSK_SIM_SELECTION_DEVEX	The optimizer uses devex steepest-edge pricing. If it is not available an approximate steep-edge selection is chosen.
MSK_SIM_SELECTION_SE	The optimizer uses steepest-edge selection. If it is not available an approximate steep-edge selection is chosen.
MSK_SIM_SELECTION_PARTIAL	The optimizer uses a partial selection approach. The approach is usually beneficial if the number of variables is much larger than the number of constraints.

MSK_IPAR_SIM_EXPLOIT_DUPVEC (string): Controls if the simplex optimizers are allowed to exploit duplicated columns. ↵

Default: MSK_SIM_EXPLOIT_DUPVEC_OFF

value	meaning
MSK_SIM_EXPLOIT_DUPVEC_ON	Allow the simplex optimizer to exploit duplicated columns.
MSK_SIM_EXPLOIT_DUPVEC_OFF	Disallow the simplex optimizer to exploit duplicated columns.
MSK_SIM_EXPLOIT_DUPVEC_FREE	The simplex optimizer can choose freely.

MSK_IPAR_SIM_HOTSTART (string): Controls the type of hot-start that the simplex optimizer perform. ↵

Default: MSK_SIM_HOTSTART_FREE

value	meaning
MSK_SIM_HOTSTART_NONE	The simplex optimizer performs a coldstart.
MSK_SIM_HOTSTART_FREE	The simplex optimize chooses the hot-start type.
MSK_SIM_HOTSTART_STATUS_KEYS	Only the status keys of the constraints and variables are used to choose the type of hot-start.

MSK_IPAR_SIM_HOTSTART_LU (string): Determines if the simplex optimizer should exploit the initial factorization. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_SIM_MAX_ITERATIONS (integer): Maximum number of iterations that can be used by a simplex optimizer. ↵

Default: GAMS IterLim

MSK_IPAR_SIM_MAX_NUM_SETBACKS (integer): Controls how many set-backs are allowed within a simplex optimizer. ↵

A set-back is an event where the optimizer moves in the wrong direction. This is impossible in theory but may happen due to numerical problems.

Default: 250

MSK_IPAR_SIM_NON_SINGULAR (string): Controls if the simplex optimizer ensures a non-singular basis, if possible. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_SIM_PRIMAL_CRASH (integer): Controls whether crashing is performed in the primal simplex optimizer. ↵

In general if a basis consists of more than (100*MSK_IPAR_SIM_PRIMAL_CRASH) percent fixed variables, then a crash will be performed.

Default: 90

MSK_IPAR_SIM_PRIMAL_RESTRICT_SELECTION (integer): Controls how aggressively a restricted selection/pricing strategy is used to choose the outgoing variable in the primal simplex. ↵

The primal simplex optimizer can use a so-called restricted selection/pricing strategy to chooses the outgoing variable. Hence, if restricted selection is applied, then the primal simplex optimizer first choose a subset of all the potential incoming variables. Next, for some time it will choose the incoming variable only among the subset. From time to time the subset is redefined. A larger value of this parameter implies that the optimizer will be more aggressive in its restriction strategy, i.e. a value of 0 implies that the restriction strategy is not applied at all.

Range: {0, ..., 100}

Default: 50

MSK_IPAR_SIM_PRIMAL_SELECTION (string): Controls the choice of the incoming variable, known as the selection strategy, in the primal simplex optimizer. ↵

Default: MSK_SIM_SELECTION_FREE

value	meaning
MSK_SIM_SELECTION_FREE	The optimizer chooses the pricing strategy.
MSK_SIM_SELECTION_FULL	The optimizer uses full pricing.
MSK_SIM_SELECTION_ASE	The optimizer uses approximate steepest-edge pricing.
MSK_SIM_SELECTION_DEVEX	The optimizer uses devex steepest-edge pricing. If it is not available an approximate steep-edge selection is chosen.
MSK_SIM_SELECTION_SE	The optimizer uses steepest-edge selection. If it is not available an approximate steep-edge selection is chosen.
MSK_SIM_SELECTION_PARTIAL	The optimizer uses a partial selection approach. The approach is usually beneficial if the number of variables is much larger than the number of constraints.

MSK_IPAR_SIM_REFORMULATION (string): Controls if the simplex optimizers are allowed to reformulate the problem. ↵

Default: MSK_SIM_REFORMULATION_OFF

value	meaning
MSK_SIM_REFORMULATION_ON	Allow the simplex optimizer to reformulate the problem.
MSK_SIM_REFORMULATION_OFF	Disallow the simplex optimizer to reformulate the problem.
MSK_SIM_REFORMULATION_FREE	The simplex optimizer can choose freely.
MSK_SIM_REFORMULATION_AGGRESSIVE	The simplex optimizer should use an aggressive reformulation strategy.

MSK_IPAR_SIM_SAVE_LU (string): Controls if the LU factorization stored should be replaced with the LU factorization corresponding to the initial basis. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_SIM_SCALING (string): Controls how much effort is used in scaling the problem before a simplex optimizer is used. ↵

Default: MSK_SCALING_FREE

value	meaning
MSK_SCALING_FREE	The optimizer chooses the scaling heuristic.
MSK_SCALING_NONE	No scaling is performed.

MSK_IPAR_SIM_SCALING_METHOD (string): Controls how the problem is scaled before a simplex optimizer is used. ↵

Default: MSK_SCALING_METHOD_POW2

value	meaning
MSK_SCALING_METHOD_POW2	Scales only with power of 2 leaving the mantissa untouched.
MSK_SCALING_METHOD_FREE	The optimizer chooses the scaling heuristic.

MSK_IPAR_SIM_SEED (integer): Sets the random seed used for randomization in the simplex optimizers. ↵

Range: {0, ..., 32749}

Default: 23456

MSK_IPAR_SIM_SOLVE_FORM (string): Controls whether the primal or the dual problem is solved by the primal-/dual-simplex optimizer. ↵

Default: MSK_SOLVE_FREE

value	meaning
MSK_SOLVE_FREE	The optimizer is free to solve either the primal or the dual problem.
MSK_SOLVE_PRIMAL	The optimizer should solve the primal problem.
MSK_SOLVE_DUAL	The optimizer should solve the dual problem.

MSK_IPAR_SIM_STABILITY_PRIORITY (integer): Controls how high priority the numerical stability should be given. ↵

Range: {0, ..., 100}

Default: 50

MSK_IPAR_SIM_SWITCH_OPTIMIZER (string): Controls the simplex behavior. ↵

The simplex optimizer sometimes chooses to solve the dual problem instead of the primal problem. This implies that if you have chosen to use the dual simplex optimizer and the problem is dualized, then it actually makes sense to use the primal simplex optimizer instead. If this parameter is on and the problem is dualized and furthermore the simplex optimizer is chosen to be the primal (dual) one, then it is switched to the dual (primal).

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_TIMING_LEVEL (integer): Controls the amount of timing performed inside MOSEK. ↵

Default: 1

MSK_IPAR_WRITE_COMPRESSION (integer): Controls whether the data file is compressed while it is written. ↵

0 means no compression while higher values mean more compression.

Default: 9

MSK_IPAR_WRITE_DATA_PARAM (string): If this option is turned on the parameter settings are written to the data file as parameters. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_WRITE_GENERIC_NAMES (string): Controls whether generic names should be used instead of user-defined names when writing to the data file. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_WRITE_GENERIC_NAMES_IO (integer): Index origin used in generic names. ↵

Default: 1

MSK_IPAR_WRITE_IGNORE_INCOMPATIBLE_ITEMS (string): Controls if the writer ignores incompatible problem items when writing files. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_WRITE_JSON_INDENTATION (string): When set, the JSON task and solution files are written with indentation for better readability. ↵

Default: MSK_OFF

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_WRITE_LP_FULL_OBJ (string): Write all variables, including the ones with 0-coefficients, in the objective. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_WRITE_LP_LINE_WIDTH (integer): Maximum width of line in an LP file written by MOSEK. ↵

Range: {40, ..., ∞}

Default: 80

MSK_IPAR_WRITE_MPS_FORMAT (string): Controls in which format the MPS is written. ↵

Default: MSK_MPS_FORMAT_FREE

value	meaning
MSK_MPS_FORMAT_STRICT	It is assumed that the input file satisfies the MPS format strictly.
MSK_MPS_FORMAT_RELAXED	It is assumed that the input file satisfies a slightly relaxed version of the MPS format.
MSK_MPS_FORMAT_FREE	It is assumed that the input file satisfies the free MPS format. This implies that spaces are not allowed in names. Otherwise the format is free.
MSK_MPS_FORMAT_CPLEX	The CPLEX compatible version of the MPS format is employed.

MSK_IPAR_WRITE_MPS_INT (string): Controls if marker records are written to the MPS file to indicate whether variables are integer restricted. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_WRITE_TASK_INC_SOL (string): Controls whether the solutions are stored in the task file too. ↵

Default: MSK_ON

value	meaning
MSK_ON	Switch the option on.
MSK_OFF	Switch the option off.

MSK_IPAR_WRITE_XML_MODE (string): Controls if linear coefficients should be written by row or column when writing in the XML file format. ↵

Default: MSK_WRITE_XML_MODE_ROW

value	meaning
MSK_WRITE_XML_MODE_ROW	Write in row order.
MSK_WRITE_XML_MODE_COL	Write in column order.

MSK_SPAR_DATA_FILE_NAME (string): If set, problem data is written to this file. File extension determines format. ↵

Synonym: writeprob

MSK_SPAR_PARAM_READ_FILE_NAME (string): Modifications to the parameter database is read from this file. ↵

MSK_SPAR_PARAM_WRITE_FILE_NAME (string): The parameter database is written to this file. ↵

MSK_SPAR_WRITE_LP_GEN_VAR_NAME (string): Sometimes when an LP file is written additional variables must be inserted. ↵

They will have the prefix denoted by this parameter.

Default: "xmskgen"

QEXTRACTALG (integer): Switch to choose extraction algorithm for quadratic equations in GAMS interface. ↵

Default: 0

value	meaning
0	Choose automatically.
1	ThreePass: uses a three-pass forward / backward / forward AD technique to compute function / gradient / Hessian values and a hybrid scheme for storage
2	DoubleForward: uses forward-mode AD to compute and store function, gradient, and Hessian values at each node or stack level as required. The gradients and Hessians are stored in linked lists.
3	Concurrent: Uses ThreePass and DoubleForward in parallel. As soon as one finishes, the other one stops.

SDPCHECKVARS (boolean): Switch to disable checking that for every entry of a PSD matrix variable also a corresponding GAMS variable is present. ↵

Default: 1

SDPSOLUFILE (string): Name of GDX file to write primal solution of all PSD matrix variables and dual solution for PSD constraints. ↵

SOLVEFINAL (boolean): Switch to resolve the problem with fixed discrete variables after the MOSEK optimizer finished. ↵

Default: 1

value	meaning
0	Do not solve the fixed problem.
1	Solve the fixed problem and return duals.

Setting up a GAMS/MOSEK-Link license

To use the GAMS/MOSEK solver with a GAMS/MOSEK-Link license you have save a MOSEK standalone license and point to the MOSEK license file via environment variable MOSEKLM_LICENSE_FILE.