A graph displaying a counterexample to the Dinitz-Garg-Goemans conjecture
Dmitry Rybin
A longstanding mathematical conundrum has been solved by ChatGPT in a number of hours, with just a few easy prompts.
The Dinitz-Garg-Goemans conjecture is a 30-year-old query in graph idea, however a counterexample posted on X by Dmitry Rybin, co-founder at AI startup Autokernel, has proven that it’s false.
Rybin entered simply 4 prompts into ChatGPT 5.6 Professional: an preliminary one instructing the AI to “do a breakthrough and discover a structured counterexample”, after which three others merely urging it to proceed looking. All 4 added as much as fewer than 60 phrases, and the AI took a complete of 5 and a half hours to crack the issue.
Graph idea is the examine of networks made up of nodes, or vertices. The Dinitz-Garg-Goemans conjecture will be considered a logistical problem: think about shipments from a warehouse to a number of places will be break up into a lot smaller deliveries that may be despatched on completely different routes. The conjecture states that this state of affairs will be transformed into one other the place shipments can’t be break up, and that the overall price of delivery is not going to enhance.
Rybin didn’t reply to a request for remark, however said on X: “I do know counterexamples to outdated conjectures have gotten a meme at this level. However I actually cared about this downside and spent many weeks desirous about it.”
Chris Bowman-Scargill on the College of York, UK, says there’s a operating joke in arithmetic that each conjecture in graph idea is fake, simply because the Dinitz-Garg-Goemans conjecture has now been proved to be.
“In fields like quantity idea or algebra, patterns that maintain for small rank [simple situations] typically maintain for a very long time,” says Bownman-Scargill. “Whereas in graph idea, structural behaviour can shift dramatically when you add only one or two vertices… which is how these conjectures proceed to be made. You possibly can see how folks miss this stuff.”
AI has made speedy advances in arithmetic in latest months. In Might, an OpenAI mannequin cracked a decades-old conjecture by Paul Erdős, inflicting a stir in mathematical circles. Earlier this week, an AI found a counterexample to the Jacobian conjecture, which had stood for practically a century. In the present day, different AI customers declare to have solved a Graffiti conjecture and a second graph theory problem. A web site has even sprung as much as catalogue the AI findings and listing them by the mannequin that was used.
Abhishek Saha at Queen Mary College of London says present AI fashions appear notably well-suited to issues just like the Dinitz-Garg-Goemans conjecture, however there are limits to what’s at present attainable – and the issues solved by AI to date are of restricted complexity.
“AI is absolutely good, and at the very least at some mathematical duties, already superhuman,” says Saha. “There’s a good bit of low-hanging fruit on the market. Some conjectures can now be proved or disproved by AI with little or no human enter; the principle problem is just pointing the system in the appropriate route. However, I don’t suppose AI is but at a spot the place it might probably construct the speculation wanted to show a number of the deepest open conjectures folks care about.”
However there are indicators that AI is right here to remain and can change into a significant instrument for mathematicians. Alexander Yong on the College of Illinois Urbana-Champaign says AI’s rising position in arithmetic will empower researchers to discard useless ends and push in promising instructions as an alternative.
“Counterexamples to outdated conjectures are by no means fairly as spectacular as discovering a sequence of interlocking arguments that represent a proof,” says Yong. “Nevertheless, I’d count on that AI will quickly show many conjectures by a mixture of their inherent superhuman vitality in understanding the literature and attempting many issues at a immediate. These conjectures that survive AI scrutiny would be the real objectives for human innovation.”
