To map a tiny magnet, researchers made the sensor’s distance part of the answer

For a tiny magnet, the gap between a specimen and its sensor is part of the picture. A small error in that distance can distort the magnetic pattern inferred from the measurement. Researchers led by the University of Vienna have made the separation something their reconstruction software estimates alongside the magnet itself.[2]

Their instrument uses an atomic defect in diamond as a tiny magnetic-field probe. It scans the field extending above the sample; the arrangement of magnetisation inside has to be inferred. The experiment combined magnetic material made in Singapore with measurements in Dresden and a computational method developed in Vienna.[2]

A picture assembled backwards

Think of a magnetisation map as arrows showing magnetic orientation across a sample. A simulation ordinarily starts with a material and a proposed arrangement of those arrows, then calculates what that arrangement would produce. Reconstruction reverses the question: which arrangement could have produced the observation? NeuralMag, the open software underlying this work, turns that search into an optimisation problem.[3]

Its useful connection to AI is automatic differentiation. The software keeps track of how a calculated result changes when an input changes, allowing an optimiser to adjust many unknowns together. The same computational machinery is used to train neural networks. Here it helps steer a physical model towards a measurement. There is no requirement to collect a library of labelled magnetic images first.[3][4]

The reconstruction code combines two demands: reproduce the measured field and favour a magnetisation pattern with plausible physical energy. A candidate can earn a better field match while becoming physically less convincing. The optimiser therefore has to negotiate between the two demands, instead of treating resemblance to the measured image as the whole objective.[4]

This distinction matters whenever a picture is assembled from indirect evidence. A highly detailed answer can be less informative than a simpler one if its detail mainly accommodates noise. Physical knowledge gives the search a reason to prefer one explanation over another; the observations still have to constrain that preference.

What the 81-nanometre result means

For the experimental reconstruction, the team used a 100-by-100-pixel field scan covering an 800-by-800-nanometre patch of a 100-nanometre-thick Fe₃−xGaTe₂ magnet. At the chosen balance between field agreement and energy, the inferred effective sensor distance settled near 81 nanometres. This effective separation includes the sensor defect’s depth in the diamond and intervening material, as well as the physical gap.[1]

The paper reports a relative field mismatch of about 14.5% in unmasked regions. That is a measure of disagreement between field maps, not an error rate for the recovered magnetisation. Giving the energy term too little weight produced implausible fine structure; giving it too much weakened the match to the experiment.[1]

There is also a physical limit to what any such scan can reveal. Some magnetic arrangements produce no external stray field, and fine features become harder to detect with distance. In this experiment, aspects of the narrow domain walls were consequently determined by the assumed physics. The resulting map is a physically supported reconstruction, rather than a unique photograph of an otherwise invisible state.[1]

A reconstruction other laboratories can examine

The accompanying repository contains the reconstruction workflow and links to the measured maps and archived results. It saves the assumed configuration, recovered magnetisation and calculated field, so another researcher can examine how an answer was obtained.[4]

That makes the distance estimate useful beyond its headline number. A laboratory can investigate how its reconstruction changes when it changes an assumption, and whether the improvement comes from a better fit or a more plausible magnetic state. The computational result becomes an explanation that can be inspected and challenged.

The achievement is a closer relationship between the instrument and its interpretation. An uncertain piece of the measurement geometry has become part of the scientific question—and part of the answer.

Report an error or suggest a correction