Penn Engineers' AI Breakthrough Solves Complex Math Puzzle

Scientists just found a way to answer questions they've been stuck on for decades, and it didn't require building bigger computers.

Researchers at the University of Pennsylvania developed a new AI method that solves inverse partial differential equations, some of the toughest puzzles in mathematics. These equations help scientists work backward from what they observe to discover hidden causes, like watching ripples in a pond and figuring out exactly where the pebble landed.

The team, led by professor Vivek Shenoy, realized the problem wasn't about needing more computing power. Instead, they needed better math.

Their solution came from an unexpected source: a concept from the 1940s called "mollifiers," originally developed by mathematician Kurt Otto Friedrichs. The researchers created special "mollifier layers" that smooth out noisy data before AI analyzes it, like cleaning a scratched window before trying to see through it clearly.

The results surprised even the research team. Their method dramatically reduced errors and slashed the computing power needed to solve these complex equations.

Graduate student Vinayak Vinayak explains that modern AI often advances by using more powerful hardware. "But some scientific challenges require better mathematics, not just more compute," he says.

The breakthrough has immediate real-world applications, particularly in understanding how DNA organizes itself inside cells. For years, scientists could see DNA structures but couldn't reliably figure out the chemical processes controlling which genes turn on and off.

"We kept running into the same problem," says Shenoy, who studies chromatin, the folded state of DNA inside cell nuclei. "We could see the structures, but we couldn't infer the epigenetic processes driving the system."

Now they can. The new method lets researchers decode these hidden genetic processes, which could transform how we understand and treat diseases.

The approach works for other scientific challenges too. Weather prediction, understanding how heat moves through materials, and modeling chemical reactions all rely on these types of equations.

Why This Inspires

This story shows how looking backward sometimes moves us forward. Instead of chasing the newest, most powerful technology, the Penn team found their answer in decades-old mathematical theory that just needed the right application.

Their work proves that solving today's hardest problems doesn't always require tomorrow's technology. Sometimes it requires remembering yesterday's wisdom and applying it in new ways.

The research will be presented at the 2026 Conference on Neural Information Processing Systems and was published in Transactions on Machine Learning Research.

Scientists worldwide can now use this method to answer questions that seemed impossibly difficult just months ago, opening doors to discoveries we haven't even imagined yet.