📈 #20 PDLP: bridging the gap between traditional solvers and the demands of large-scale optimization

Is Google doing a right job here?

I like writing the issue number at the beginning of the title.

The past issue number was 47, so today should be 48, but it’s 20. Who am I to change the counting rules?

I just realized one thing:

Who am I to break the counting rules? That’s true, no one. So today’s issue number is 20, though the real one is 47.

I hope to enter the heaven of mathematics after my life on Earth anyway.

Remember the other day I wrote the shortest email ever thanks to Google Research?

They developed a new solver for Linear Programming problems called Primal-Dual hybrid gradient enhanced for LP (PDLP).

In the last issue, I told you that LP was mostly solved. So, why should you care about that development?

Today in Feasible:

• Why is PDLP crucial in modern optimization?

• The power and efficiency of this new solver

• When to use PDLP vs. traditional methods?

🎧 Remember you can listen to a podcast made for this issue, at the top of this post 🎧

Ready, set… Let’s dive in! 🪂

🎯 Why is PDLP crucial in modern optimization?

The world is more connected than ever.

Thanks to globalization, there’s an increasing amount of data. We talk about big data, and today’s numbers are intangible, like hundreds of zettabytes per year.

Industries worldwide face increasingly complex problems.

The demand for powerful optimization tools has never been greater.

However, traditional solvers are struggling to keep pace, hindered by memory constraints and inefficient hardware use. We need something to address: