The refrigerator in the basement costs more than a house and cools a copper stage to within a few thousandths of a degree of absolute zero. At that temperature, materials behave in ways that remain, after decades of study, incompletely explained.

The lab studies superconductivity and the exotic magnetic states that appear in thin layers of engineered material. The work is basic research in the strict sense: nobody here is building a product, and the group's principal investigator is direct about the timeline when asked about applications.

"Thirty years, if it works," she said. "Possibly never. That is what basic research means and I would rather say it plainly than dress it up."

Saying it plainly is harder than it used to be. Federal funding for physical sciences has been roughly flat in real terms for years, and the grant cycle rewards proposals that promise deliverables. The result, senior researchers across the university say, is a systematic tilt toward incremental work that can be defended annually.

The lab's graduate students represent the more immediate output. Six are working here now; historically, most such students leave academic research within a few years of finishing, taking their training into industry.

The pipeline into Minnesota's device and instrumentation companies is direct enough that several firms fund fellowships. They are not buying research results. They are buying access to people who know how to keep a dilution refrigerator running.

Quantum computing has changed the funding conversation, mostly for the better and somewhat distortingly. Money has arrived, along with an expectation of applications that the underlying physics does not yet support.

"Every proposal now has a sentence about quantum computing," a postdoctoral researcher said. "Some of them are honest."

The group's most cited recent result concerns an unexpected magnetic ordering in a layered compound. Its practical significance is unknown. Its scientific significance is that an existing theoretical model predicted something else.

That is the actual product of a lab like this one: a model that was wrong, identified as wrong, published as wrong. Everything downstream depends on someone having done it.