Octeract Engine

Solve large-scale MINLPs in production.

Octeract Engine can solve any problem of the following type to local or global optimality:

\[\min_{x\in X\\y\in Y}f(x,y)\\ \text{s.t.}\\ \tilde{g}(x,y)\leq \tilde{0}\\\tilde{h}(x,y)=\tilde{0}\]

where \(x\in\mathbb{R}^n,y\in\mathbb{Z}^m\). The functions do not need to be continuous nor differentiable. 

Two world records

As of 20 April 2023, Octeract Engine is the first (and only) solver to ever solve 100% of the Mittelmann MINLP benchmark, at a record-breaking unscaled shifted geometric mean of 36.8.

For your convenience, you can find some screenshots of visualisations of these benchmarks taken from this website below. We don’t really keep these tables up to date, so visit the links below for the raw data and more current information.

Third-party Benchmarks

Octeract Engine is a milestone technology in the field of (MI)NLP.

It has made historic achievements, including solving some of the largest known open problems, more benchmark problems than any nonlinear solver, ever, and breaking two world records.

It is also the first nonlinear solver with distributed branch-and-bound.

On 8 February 2023, we got ahead of all MINLP solvers in all third-party benchmarks (388 problems in total) by a considerable margin. We’ve also solved the largest open transmission switching problems in MINLPLIB (this one and this one).

Note that these benchmarks were made with Octeract Engine v4.7, which was most likely our final off-the-shelf build. 

Improving the core technology to the point where it could solve the entire MINLP test set was challenging and intense, but we have hence moved on to more important things.

Nowadays, we work on our Neural technology and solve real-world production problems for various industries.

What about real problems?

We tend to be the people people come to when nothing else works.

Octeract technology is used by consultancies, public companies, academics, and national labs, for R&D and production.

Our off-the-shelf build comes with a lot of options, and is great for R&D.

For production, we produce blazing fast bespoke solutions using a variety of tooling, including Octeract Neural, capable of solving multi-million variable MINLPs every few minutes. This is well beyond the capabilities of vanilla Octeract Engine, or any other solver for that matter.

Solving MINLPs in production used to be impossible, but no longer – it is now merely a function of effort and computing power.

Talk to our experts