What does Curt think? Ask Curt! These remarks are often heard at Carmen, and no wonder! The optimizer Curt has created is thrashing the competition by creating major improvements to problems that were already considered solved. Curt has taken optimization to a higher level and his modest, intuitive ways have made him the undisputed optimization guru at Carmen.

At the Sydney Agifors Symposium you spoke about integration of crew and aircraft planning. Why?
We have done simulations of this on real production data for four different airlines and the results look good. The savings are around 1% on crew cost and the reduction of crew-plane changes can be as dramatic as 70%. This has a big impact on the robustness of the operation.

So why don't airlines do this right away?
The technology is quite new, so most airlines probably don't know about this potential yet. Also, there are some organizational hurdles.

What do you mean?
Crew and aircraft are managed in different departments with different objectives. Nobody is in charge of the combined crew/aircraft problem.

But Carmen still makes different products for crew and aircraft. Where is this integrated crew and aircraft product?
The integration presented at the symposium doesn't require a new product. We can use integration technology like Carmen Connect to provide feedback from pairing to tail assignment. Since our products are flexible with regards to the cost function, this is easy. You can get further savings by incorporating a portion of the aircraft problem into the pairing model. This is done using a plug-in Rave module, and the new constraint features in Carmen 8.

What about the future? Will there be one product solving of what we currently see as separate problems?
Yes, definitely, since integration is one of the main ways to achieve further cost savings. I expect that the borders of our current problems will begin to blur more and more, eventually leading to completely integrated products. Underneath there will be a collection of optimization solvers, working together, but from the user point of view they will see just one optimizer. Right now we have our new Crew Recovery solver that is basically a merger of our pairing and rostering optimizers into one program. Its performance will improve until we reach the point where we don't need pairing and crew rostering, we use just one integrated product which does a better job.

When you started work on the column generator, people were worried about how it could cope with complex rules and costs. How did you achieve this?
Well there was already the Rave system which model-led the rules and costs. But the interface it supplied wasn't what one would normally use in a column gene-rator. When doing my Ph.D. back in Auckland, I had spent a lot of time reading about shortest paths, especially resolving them. I was able to use some of those ideas, together with Rave, to build a flexible column generator.

Isn't there more to the column generator? Why does it work so well?
Well one of the key ideas is to take advantage of problem structure. For the pairing problems it turns out that duties (one working day) are very natural building blocks. But there are billions of different ways to join those duties, too much to compute explicitly. So you start with a relaxed model, and refine it dynamically for the duties you are interested in, the low cost efficient ones. Of course there are also a number of "tweaks" in there as well, based on extensive testing and analysis.

What made you move from New Zealand to Carmen in Sweden?
When I first heard about Carmen it was in connection with Paros, an R&D; project involving Lufthansa and two universities. The opportunity of working at Carmen really excited me since I could develop new optimization ideas with the universities and apply these ideas to real life problems. Since then Carmen has constantly provided me with new challenges to keep me busy.