Global electricity demand is climbing fast, and solar energy needs materials that go well beyond what silicon panels can offer today. One family of crystalline compounds — halide perovskites — has drawn intense scientific interest for years, thanks to their exceptional ability to absorb and emit light. But a critical piece of their structural puzzle has stubbornly resisted explanation.
Now, researchers at Chalmers University of Technology in Sweden say they’ve finally filled that gap — using a simulation method that simply wasn’t possible a few years ago.
The missing piece in perovskite research
Halide perovskites have become the material scientists most want to crack for next-generation solar applications. Unlike conventional silicon panels — rigid, heavy, and expensive to manufacture — perovskite-based cells could be made ultra-thin and flexible enough to coat a smartphone case or wrap around an entire building facade.
Within that family, one compound stands out: formamidinium lead iodide, or FAPbI3. Its ability to absorb and emit light is exceptional, making it a strong candidate for highly efficient solar cells. The catch is stability. FAPbI3 degrades quickly under real-world conditions, and researchers have long suspected that a clearer understanding of its internal structure would point toward fixing that problem.
One phase in particular — the structure the material adopts at low temperatures — has resisted clear explanation. Without knowing what’s happening at that level, designing stable mixtures of perovskite compounds remains largely guesswork.
Machine learning meets supercomputer simulations
The Chalmers team’s approach combined conventional computational physics with what are called machine-learned potentials — models trained to predict atomic interactions far more efficiently than traditional methods allow. The gains in computing capacity were substantial.
“By combining our standard methods with machine learning, we’re now able to run simulations that are thousands of times longer than before,” said researcher Sangita Dutta. “And our models can now contain millions of atoms instead of hundreds, which brings them closer to the real world.” That scale matters because perovskite behavior is sensitive to subtle structural details that only emerge across large numbers of atoms over extended time periods — details that smaller, shorter simulations simply miss.
With this expanded capability, the team mapped the low-temperature structure of FAPbI3 in detail. As the material cools, the formamidinium molecules — organic ions at the heart of the crystal — don’t settle cleanly into place. Instead, they become trapped in a semi-stable state, a kind of structural limbo that had previously gone undetected.
From simulation to lab confirmation
Computational findings, however detailed, require experimental validation. The Chalmers researchers partnered with collaborators at the University of Birmingham, who cooled physical samples of FAPbI3 to −200°C and observed the material’s behavior directly. The lab results matched the simulations — and that agreement between two very different methods significantly strengthens confidence in the newly identified structural picture.
The study, authored by Dutta, Erik Fransson, and colleagues from both institutions, was published in the Journal of the American Chemical Society in August 2025. Principal investigator Julia Wiktor, an associate professor at Chalmers, described the findings as a resolution of something the field had been circling for years. “The low-temperature phase of this material has long been a missing piece of the research puzzle,” Dutta noted, “and we’ve now settled a fundamental question about the structure of this phase.”
Why it matters for the future of solar energy
The timing isn’t incidental. The International Energy Agency projects that electricity’s share of global energy consumption could exceed 50 percent within 25 years, up sharply from roughly 20 percent today. Meeting that demand sustainably will require solar technology that’s cheaper, more adaptable, and more durable than what currently dominates the market.
Perovskites are already competitive with silicon in laboratory efficiency measurements. Real-world longevity is what’s still missing. Knowing how FAPbI3 behaves structurally — including what happens during cooling — gives researchers a clearer target when engineering stable perovskite mixtures designed to hold up over time.
The implications extend beyond this single compound. The machine-learned potential methodology the Chalmers team developed could be applied to other complex materials in the perovskite family, many of which remain poorly understood for similar reasons. As Wiktor put it, the field now has “simulation methods that can answer questions that were unresolved just a few years ago.”
What comes next is the harder engineering work: translating structural insights into perovskite blends that perform reliably outside the lab. If that challenge can be met, the path to solar cells thin enough to coat everyday surfaces — and affordable enough to deploy at scale — becomes considerably shorter.