|
April 27th, 2007
Release of Xpress-MP 2007
Artelys and Dash Optimization today announced
the release of Xpress-MP 2007. Xpress-MP continues to provide
state-of-the-art modeling and optimization tools to customers
in a wide range of vertical industries.
Xpress-MP release 2007 includes major revisions
to the Optimizer LP/MIP solver, the Mosel modeling language
and the Kalis constraint programming engine.
Constraint Programming: Xpress-Kalis
Artelys and Dash reinforce their joint
efforts to provide the most advanced constraint programming
functionality in the Mosel environment.
Xpress-Kalis
now includes dedicated high-level functionality to ease the
formulation and solving of scheduling problems:
•
New constraint relations representing disjunctive and cumulative
resource constraints have been introduced;
•
Predefined objects such as tasks and resources provide high-level
modeling support for many scheduling situations;
•
For solving scheduling problems, the user has the choice between
efficient predefined or configurable branching strategies
reasoning on the scheduling objects, and the implementation
of his own search heuristics;
•
The automated result display for scheduling problems (resource
diagrams, Gantt charts) in the development environment IVE
helps analyzing the obtained solutions.
Xpress-Kalis now also handles continuous
variables and arithmetic (linear and non-linear) constraints
over continuous and discrete variables. This means that mixed
integer non-linear problems can now be solved with Xpress-Kalis.
Xpress-Optimizer 18.00.01
Optimization performance has been improved
across a wide range of models. The table below shows the performance
comparison with release 2006A for the Dash standard QA set
of models:
| Algorithme |
Improvement 2007A
/ 2006A |
|
Primal
|
9%
|
|
Barrier
|
13%
|
|
MIP
|
41%
|
•
Presolve
| |
- Presolve has been completely rewritten. It is now
faster, especially on large models, and performs more
reductions;
- Improved tightening for mixed integer models results
in better solutions and reductions in solve times;
- Dualization: The LP now optimizes much faster on models
with far more rows than columns.
|
•
Dual Simplex has been improved
on models where it is hard to achieve dual feasibility;
•
Some numerical problems have
been made more robust with better objective function scaling;
•
MIR cut generation has been improved,
leading to smaller gaps and/or faster solve times for mixed
integer problems;
•
New Network Simplex algorithm
is 10 - 100 times faster than dual on network models. Advanced
feasible starts are supported;
•
New node-to-node presolve leads
to significant improvements to branch-and-bound on mixed integer
models;
•
The local search heuristic now
generates better solutions, especially at the root node;
•
The diving heuristic has been
improved to find better solutions;
•
Improvements have been made to
obtain faster solutions to scheduling models;
•
Solutions are now memory resident
by default.
Xpress-Mosel 2.0.0
The new release features a multitude of
improvements to the most advanced modeling environment available
today:
•
Packages: Mosel can now be used
to create external compiled libraries of functions and variables
which can be imported for use in other models;
•
Records: Mosel now supports user-defined
structure types that can encapsulate several member variables;
•
Complex nested types: Any structured
type (array, set, list or record) can now be used as an elementary
type e.g. array(range) of set of integer
•
A new structure list replaces
tuples, making it possible to declare and manipulate ordered
collections of objects of a given type. Aggregate operators
now accept lists as a source of values for their indices;
•
User types can now be declared
as an alias to a complex type;
•
A new IO driver excel
provides initialization from Excel files without requiring
SQL statements;
•
Mosel 2 includes full support
for Parallel MIP multi-threading.
Back
to news |
