Sneak Previews: CPLEX 2.0

Later in 1991, users can expect a new release (Version 2.0) of the CPLEX Linear Optimizer and CPLEX Callable Library. All CPLEX licensees with current maintenance status will automatically receive copies of the updated products.

What's new in 2.0? A number of new features and algorithmic improvements to make CPLEX faster, easier to use, and more flexible, including:

• CPLEX 2.0 will include the option to perform a dual simplex optimization For some problems, the dual method will improve solving performance significantly. For example, the dual method is up to 4 times faster for large, difficult airline fleet scheduling applications. Within the dual method, users will have multiple parameter options to choose from (such as pricing options) just as are currently offered with the primal method.
• Both Linear Optimizer and Callable Library users will notice that CPLEX 2.0 provides significantly more flexibility and control over output. Users will be able to more directly control the type, degree and destination of results, dialogue, error/warning messages, and various other types of output. Interactive users with batch files, for example, will appreciate the ability to control the level of information going to the screen while a batch file is executing Both Callable Library and interactive users will be able to distinguish different output files for (and if desired, suppress altogether) warning notices, error notices, "state" notices, and results.
• CPLEX 2.0 will incorporate a number of significant new algorithmic improvements that will be "invisible" to users--except that you will certainly notice even better performance, particularly on large, difficult problems. These algorithmic improvements alone have improved the performance of some difficult problems by over 30%.
• Finally, CPLEX 2.0 will incorporate a number of minor features and additions that users have requested since the last release.

All of these improvements will also be included in updated releases of the CPLEX Mixed Integer Optimizer and Mixed Integer Library which will follow soon after the 2.0 Release of the linear optimization products.

On the Move: Office Relocation and New Staff

Most of you know that the CPLEX business offices were recently moved to Incline Village, Nevada. We had rapidly outgrown our Houston office and were pleased to find a pleasant location with a supportive business climate and software community. All sales and support activities are now conducted out of Nevada. Along with the move is the addition of new staff--and our first such welcome addition is Lorrie Harlem, office and license administrator. Please make a note of our new address and phone/fax numbers.

Product Development: New Mixed Integer Program Solvers

CPLEX Optimization is pleased to announce the availability of two new products: The CPLEX Mixed Integer Optimizer and the CPLEX Mixed Integer Library. Both new products are based on the same CPLEX linear optimization engine offered in our original products, but add the capability to incorporate binary and general integer variables into your models.

Mixed integer problems are generally more computationally demanding than "continuous" linear program problems. So the speed and robustness of the core CPLEX solving algorithms are particularly beneficial for mixed integer (MIP) problems.

Like the original CPLEX linear programming products, the mixed integer products are available in two forms:

• The CPLEX Mixed Integer Optimizer is an executable, interactive product that contains all the functionality of the CPLEX Linear Optimizer with the added capability of accepting integer variables. Like the CPLEX Linear Optimizer, the Mixed Integer Optimizer accepts problem files written in MPS or LP format, or, alternatively, problems may be entered interactively. This version features the easy-to-learn and easy-to-use command structure and help menu system that CPLEX Linear Optimizer users are already familiar with.
• The CPLEX Mixed Integer Library is "callable" version for mixed integer problems. The Mixed Integer Library contains all the functionality of the Callable Library, but with additional flexibility to accept general and binary integer variables. The Mixed Integer Library is ideal for developers wishing to "embed" the CPLEX mixed integer solver routines within their own applications. The Mixed Integer Library is designed to allow efficient and seamless integration within custom applications and/or custom interfaces.

Limited-Time MIP Upgrade Offer

Through September, 1991, we are offering current CPLEX licensees an opportunity to upgrade to the CPLEX Mixed Integer products at a special upgrade rate. Call us at (702) 831-7744 for availability and price information for your specific computer platforms.

Hewlett Packard 9000/700 Series Support

We are pleased to announce the availability of CPLEX for the new HP 9000 Series 700 technical workstations. The performance of CPLEX on these new HP workstations is astounding! We would be pleased to provide CPLEX benchmark results on the new HP platform for your problems if you are considering an HP purchase.

Q&A

Q: How can I distribute applications developed using CPLEX? What third-party distribution programs are available?

A: CPLEX offers two distinctly different third-party distribution programs -- the CPLEX Dealer and CPLEX Value Added Reseller (VAR) programs. A CPLEX Dealer buys CPLEX products at a discount for resale to third parties. The dealer then assumes primary responsibility for selling and delivering CPLEX products to his customers. This program is ideal for consultants wishing to deliver CPLEX-based solutions to their clients.

A CPLEX VAR buys a Callable Library product for development and then creates a distinct software product ("derivative work") for resale to third parties. The VAR assumes total responsibility for marketing, licensing and supporting the derivative product. Typically, VAR applications are developed to address a particular vertical market application, such as a financial portfolio optimization product. Royalties are paid whenever copies of the new product are distributed to end-users. The CPLEX Callable Library products are ideal platforms for VAR applications because they can be "embedded" readily and transparently within applications.

If you have been thinking about delivering LP-based software to others, please contact us to learn more about either of our third-party distribution programs. We consider our third party distributors a critical link in bringing complete solutions to end users, and we encourage and support these programs.

Q: I am a user of the CPLEX Linear Optimizer. I use an internally developed model-generator then submit an LP file to CPLEX to solve. Would it make sense for me to upgrade to the callable version of CPLEX (CPLEX Callable Library)?

A: It depends on how you prefer to interface to the optimizer. If generating problem files in either LP or MPS format and reading CPLEX's standard solution files meets your requirements, use the Linear Optimizer. Simple batch command files can be created to automate the operation of CPLEX.

However, if your requirements can not be met using standard CPLEX input and output or if you wish to create a more efficient or user-transparent link, use the Callable Library. Virtually anything is possible with the Callable Library directly linked with your own program. The optimization step is simply a 'function call" inside your program. Data manipulation and I/O remain under your full control.

Applications: An LP-Based Approach to Acid Rain Control

How can the acid rain problem be mitigated equatably and cost-effectively? Dr. Hugh Ellis at John Hopkins University is solving LP models with CPLEX to help answer this question. His models have been used in highly visible settings, including the United Nations in Geneva, to examine acid rain reduction alternatives within the context of scientific, political and economical concerns.

As is true for most political decisions, multiple objectives are possible. One approach is to minimize the amount of reduction to be imposed at the various sources subject to meeting maximum acid rain levels at sensitive deposition targets. Other possible objectives include minimizing average violations or maximum violations, where a violation is defined as a variance above the defined maximum safe deposition level. Dr. Ellis has tackled each of these as well as other approaches.

In the acid rain model, the dispersion of sulfur dioxide (SO2) from multiple emission sources (factories, power plants, etc.) to multiple acid rain deposition sites, or "receptors," is modeled using transfer coefficients which specify the SO2 transport expected between each source-receptor pairing. These coefficients are derived from "Long Range Transport" (LRT) simulation models--several well-known models are documented in the scientific community. A decision variable is introduced to specify the SO2 removal level, as a percentage reduction, to be imposed at each source. The model is formulated with constraints defining the deposition contributed from each source (after the imposed removal level) to each receptor site, as well as any other required economic or political constraints. The model is solved to find the optimal SO2 removal levels to be imposed at each of the various emission sources.

Unfortunately, there is much disagreement regarding the LRT models providing SO2 transfer coefficients--and conflicting models can result in conflicting solutions. Dr. Ellis has invested considerable effort to accommodate multiple LRT models and deal with the associated uncertainty. While multiple (and sometimes conflicting) coefficient models exist, Dr. Ellis argues that they all contain useful information. He concludes that "a lack of consensus regarding which transfer coefficients are better, let alone best, need not preclude their use." While many recommend waiting for additional LRT studies before imposing potentially expensive SO2 reduction requirements, Dr. Ellis hopes that his analysis will provide rational support for earlier action.

The resulting linear problems are only moderately large in terms of number of rows and columns, but close to 100% dense because almost every source potentially contributes to each receptor. For this reason, the models can consume considerable computing power and time. Dr. Ellis replaced his prior solver with CPLEX and achieved a 10X speedup. This improved performance allowed, for the first time, interactive solving. Instead of waiting an hour for problems to solve, with CPLEX, his problems ran in a few minutes. So now, for example, during a political or regulatory discussion on acid rain reduction strategies, a new constraint can be added and the model re-solved to provide immediate feedback on a question or new issue.