For nearly a century, the atmosphere has been a persistent obstacle for astronomers. It blurs starlight, smearing a point into a trembling blob, and it has forced generations of researchers to choose between fuzzy images and outrageously expensive workarounds. A team at the University of California has now turned one of the atmosphere's most basic properties, the scattering of light by air molecules, into a lens. Their prototype camera, tested on a single patch of sky, resolved a star's photosphere in a few hours of observing. The same target would have taken years with traditional interferometric methods. The result, posted to the arXiv preprint server in early August 2026, has not yet been peer-reviewed, but it has already stirred quiet excitement among stellar physicists who have waited decades for something like this.

The old way: a star's face, blurred by the sky

Every ground-based telescope, no matter how large, hits the same wall: atmospheric turbulence. Warm and cold pockets of air, constantly shifting, refract light along slightly different paths. The incoming wavefront arrives at the detector wrinkled, and a star that should be a crisp point becomes a dancing, fuzzy patch. The larger the telescope aperture, the more turbulence it integrates, so bigger mirrors do not automatically mean sharper images. This is why the sharpest pictures of stellar surfaces have historically come from space telescopes or from heroic interferometric campaigns that stitch together light from multiple telescopes separated by tens or hundreds of meters.

Interferometry works, but it is slow. To resolve a star's disk, you need to combine observations from many baseline orientations, each requiring careful calibration and long integration times. The first resolved image of a star other than the Sun, Betelgeuse, took years of coordinated effort across multiple observatories and was finally achieved in the 1990s. Even then, the image had only a handful of resolution elements across the disk. A typical star like our Sun, observed from a distance of a few parsecs, subtends an angle of roughly a milliarcsecond; resolving that from the ground demands baselines of hundreds of meters or adaptive optics systems that correct for turbulence in real time.

Adaptive optics has improved dramatically over the past two decades. Modern systems on 8-meter-class telescopes can achieve near-diffraction-limited imaging in the near-infrared, and they have produced remarkable images of stars, including the red supergiant Antares and the giant star R Doradus. But adaptive optics requires a bright guide star or a laser beacon, and it only works well in narrow wavelength bands. The correction is never perfect; residual errors limit the final resolution to a few times the diffraction limit. For stars fainter than about magnitude 10, the technique becomes impractical. The result is that, as of 2026, only a handful of stars have had their surfaces resolved at all, and most of those are giants or supergiants that are both large and relatively close.

The new scattering camera bypasses much of this. It does not fight the atmosphere; it uses it. The key insight, published in a series of theoretical papers over the past decade, is that Rayleigh scattering, the same process that makes the sky blue, preserves information about the incident light's phase. When a photon scatters off an air molecule, its direction changes, but the phase relationship between photons from different parts of the stellar disk is not destroyed; it is merely redistributed across a wide angular pattern on the ground. A camera that records this scattered light, rather than the direct beam, can in principle reconstruct the original wavefront with a computational inversion algorithm.

Rayleigh scattering as a new kind of lens

The idea is counterintuitive. Scattering is usually treated as a loss mechanism, a source of background noise that degrades images. In the new camera, it becomes the signal. The detector is not pointed at the star but at a patch of sky a few degrees away, where the scattered light from the star's own photons, after bouncing off air molecules, creates a faint, extended halo. This halo contains the same information as the direct image, but smeared out in a predictable way. The scattering pattern depends on the wavelength, the altitude of the scattering layer, and the geometry of the observation, all of which can be modeled.

The reconstruction is an inverse problem: given the measured intensity pattern on the detector, recover the brightness distribution on the stellar surface. This is mathematically ill-posed, meaning that many different surface maps could produce the same scattered pattern. The team's breakthrough was to regularize the problem using prior knowledge about stellar surfaces: they are roughly circular, they have limb darkening, and they contain features like spots and granulation that are smooth on small scales. By imposing these constraints, the algorithm can converge to a unique solution.

The prototype camera is small, with an aperture of only 10 centimeters, and it has no adaptive optics. It consists of a fast, low-noise CMOS detector, a narrowband filter centered at 500 nanometers, and a wide-field lens that captures about 10 square degrees of sky. The team pointed it at the bright star Arcturus, a red giant about 37 light-years away, and recorded the scattered halo for four hours. The resulting data set, roughly 200 gigabytes, was fed into the reconstruction algorithm running on a cluster of GPUs. The output, released in the team's preprint, shows a clear disk with a dark spot near the equator and a slight asymmetry that the authors attribute to convection cells.

The resolution is not yet at the level of the best interferometric images, but it is close. The team estimates that their reconstruction has an effective resolution of about 2 milliarcseconds, which is comparable to what a 100-meter baseline interferometer achieves. The difference is time: the interferometric image of Arcturus took several years of observations, while the scattering camera did it in a single night. The technique is also far cheaper; the entire prototype, including the detector and lens, cost under $50,000, compared to the tens of millions of dollars required for a long-baseline interferometric array.

The prototype: what the camera actually sees

On the night of April 22, 2026, the team pointed the prototype at Arcturus, a bright red giant in the constellation Boötes. At first glance, the raw data look like nothing. The detector records a smooth, faint glow across a wide field, with a slight brightening near the star's position. The star itself, if it enters the field, is heavily saturated and must be masked out. The useful signal is the halo, which has a surface brightness of roughly 1% of the sky background. Extracting it requires careful subtraction of the zodiacal light and the airglow, both of which vary over the night. The team developed a calibration procedure using a standard star with a known spectrum to model these backgrounds.

The reconstruction algorithm works in two stages. First, it estimates the phase of the scattered light across the detector, using the known scattering geometry. This is similar to the phase retrieval problem in coherent diffraction imaging, but with a twist: the scattering is not coherent, so the phase must be inferred from the intensity alone, using the regularization constraints. The second stage uses this phase estimate to invert the scattering integral and produce a surface map. The algorithm is iterative, and the team found that it converges to a stable solution after about 50 iterations, taking roughly 10 hours on their GPU cluster.

The team tested the reconstruction on synthetic data first, simulating a star with a known spot pattern and adding realistic noise. They found that the algorithm could recover the spot positions to within 10% of their true locations, and the limb darkening profile matched the input to within 5%. Encouraged, they moved to real data. The Arcturus observation was chosen because the star is bright (magnitude 0), well-studied, and known to have spots from previous interferometric campaigns. The reconstructed surface shows a spot that matches the location of a feature seen in earlier observations, although the spot appears darker and larger in the scattering image.

One limitation is that the technique only works for bright stars. The scattered halo is faint, and for stars fainter than about magnitude 4, the signal-to-noise ratio drops below 10, even with long exposures. The team is confident that this limit can be pushed to magnitude 6 or 7 with a larger detector array and longer integration times, but it will never reach the faint magnitudes accessible to adaptive optics on large telescopes. For the brightest few hundred stars, however, the scattering camera offers a new way to study their surfaces in detail.

Why earlier attempts failed

The idea of using scattered light for imaging is not new. In the 1970s, astronomers noticed that the halo around bright stars was not uniform, and some proposed that it contained information about the star's disk. But the detectors of that era were slow and noisy, and the mathematical tools for solving the inverse problem were in their infancy. Early attempts to invert the scattering pattern produced garbage, and the field moved on.

The key advances came from a different area: speckle interferometry. In the 1980s, astronomers developed techniques to extract diffraction-limited information from short-exposure images that froze the atmospheric turbulence. These methods, based on the work of Antoine Labeyrie and others, showed that the atmosphere does not destroy all information; it merely scrambles it in a way that can be undone statistically. The scattering camera takes this idea to its logical extreme: instead of freezing the turbulence, it uses the scattering itself as the carrier of information.

Another barrier was computational. Inverting the scattering integral is a computationally intensive problem, and it was not until the late 2010s that GPUs became powerful enough to run the iterative algorithms in reasonable time. The team's reconstruction of Arcturus required about 10^15 floating-point operations, which would have taken months on a 2010-era supercomputer. Even with modern hardware, the algorithm is not trivial; it requires careful tuning of the regularization parameters, and the team admits that their first few attempts produced artifacts.

Credit also goes to theoretical work on the statistics of scattered light. In particular, a 2019 paper by researchers at the University of Arizona showed that the mutual coherence function of scattered light is related to the Fourier transform of the source intensity, under certain assumptions. This provided the theoretical foundation for the reconstruction algorithm, and it has been cited by the UC team as a major inspiration. Without that groundwork, the scattering camera would have remained a curiosity.

What this unlocks for stellar physics

The most immediate payoff is the ability to image stellar surfaces routinely. Currently, the census of resolved stellar disks is tiny: as of early 2026, fewer than two dozen stars have been directly imaged with any technique, and most of those are giants or supergiants. A scattering camera on a modest telescope could increase this number by an order of magnitude within a few years, simply by observing the brightest stars in the sky.

This matters for stellar physics because surface features reveal the internal processes that shape a star. Starspots, for example, are caused by magnetic fields that suppress convection. By tracking spots over time, astronomers can measure stellar rotation periods and study magnetic activity cycles, similar to the Sun's 11-year cycle. Convection cells, the granulation pattern, reveal the transport of heat from the interior to the surface, and their size and contrast depend on the star's mass, age, and composition. Limb darkening, the decrease in brightness toward the edge of the disk, is a direct probe of the temperature structure of the atmosphere, and precise measurements can test models of stellar atmospheres.

The scattering camera could also complement asteroseismology. Space telescopes like Kepler and TESS have measured stellar oscillations with extraordinary precision, but they observe the star as a point source. Surface imaging would allow astronomers to see how the oscillations manifest as visible waves on the disk, linking the global oscillation modes to localized features. This is a long-standing goal in the field, and the scattering camera may provide a way to achieve it from the ground.

Not everyone is convinced. Some stellar physicists are skeptical that the reconstruction is truly reliable, given the ill-posed nature of the problem. The team has responded by comparing their Arcturus image to an independent interferometric map, and the agreement is good, but the comparison is limited to a few features. A more rigorous test would be to observe a star with a known surface map, such as the Sun, but the Sun is too bright for the current prototype. The team plans to test the camera on the Moon during a partial eclipse, which would provide a known, high-contrast source.

Beyond stars: scattering cameras for other skies

The technique is not limited to stellar surfaces. The same principle could be applied to other astronomical sources that are too bright for conventional imaging but too small for direct resolution. Planetary atmospheres, for example, scatter light in a way that encodes information about their temperature and composition. A scattering camera pointed at Jupiter during a spacecraft flyby could potentially map the planet's cloud bands with higher resolution than the spacecraft's own camera, by using the scattering from Earth's atmosphere as a lens.

Closer to home, the technique could be used to image the Sun's corona during a total solar eclipse. The corona is a faint, extended source that is normally overwhelmed by the Sun's disk, but its light is scattered by the Earth's atmosphere in a predictable way. A scattering camera could, in principle, reconstruct the coronal structure without needing to block the Sun's disk, simplifying eclipse observations. The team has not tested this yet, but they note that the mathematics is similar.

The most speculative application is exoplanet direct imaging. Exoplanets are extremely faint compared to their host stars, but they scatter a small fraction of the star's light. The team calculates that for the closest hot Jupiters, such as 51 Pegasi b, the scattered light from the planet might be detectable as a slight perturbation in the halo, but the signal would be tiny, on the order of one part in a million of the stellar halo. They caution that this is a rough estimate and that significant technical challenges remain, but they have not ruled it out.

The technique could also breathe new life into small, ground-based observatories. Many universities and amateur observatories have telescopes in the 10- to 50-centimeter range that are currently used for photometry or spectroscopy. Adding a scattering camera would allow them to contribute to high-resolution imaging, a field that has been dominated by a few large facilities. The team has released their reconstruction algorithm as open-source software, and they are planning a workshop to train other groups in its use.

The craft of building a scattering camera

The prototype's success rests on a series of deliberate design choices. The aperture is small, which seems counterintuitive, but a larger aperture would collect more scattered light from the star itself, swamping the faint halo. The detector must have low read noise and a fast frame rate, because the scattering pattern changes as the atmosphere evolves. The team chose a scientific CMOS sensor with 4.2 megapixels and a read noise of 1.5 electrons, running at 30 frames per second. The exposure time is 33 milliseconds, which freezes the atmospheric variations, and the frames are then summed in software, after co-aligning them to correct for field rotation.

Calibration is a major part of the craft. The team observed a standard star, Vega, which is known to have a nearly uniform disk, and used it to measure the point-spread function of the scattering camera. This PSF includes both the scattering and the instrumental response, and it is used as a kernel in the reconstruction algorithm. They also observed a dark field to measure the detector bias and a flat field to correct for pixel-to-pixel sensitivity variations. The entire calibration process takes about an hour per night.

The reconstruction algorithm is the heart of the instrument, and it is where the team expects to see the most improvement. The current version uses a maximum a posteriori approach with a total-variation regularizer, which promotes smoothness while preserving sharp edges. The team is experimenting with more sophisticated priors, such as models of stellar convection, which could improve the resolution further. They are also working on a faster algorithm that can run in real time, which would allow them to guide the observation and adjust the exposure if the reconstruction is not converging.

The team is led by postdoctoral researcher Elena Rodriguez, who conceived the project during her graduate work, and principal investigator Marcus Chen, a professor of astronomy at UC Berkeley. They have received a small grant from the National Science Foundation to build a second-generation camera with a larger detector and a wider field of view. The next steps are to observe a sample of bright stars over several nights to test the reproducibility of the reconstructions, and to compare their images with those from the CHARA array, a long-baseline interferometer that has resolved many stars. If those tests pass, the scattering camera could become a standard tool in stellar astronomy.

Limitations and open questions

Despite the promise, the scattering camera is not without significant limitations. The most obvious is the brightness constraint: the technique currently works only for stars brighter than roughly magnitude 4, which limits the sample to a few hundred objects. This excludes most of the stars that astronomers would like to study, including many red dwarfs and most main-sequence stars beyond a few parsecs. The team hopes to push the limit to magnitude 6 or 7 with a larger detector and longer integrations, but even that would only reach a few thousand stars, a tiny fraction of the Milky Way's stellar population.

Another concern is the reliability of the reconstructions. The inverse problem is ill-posed, and the algorithm relies on strong priors about stellar surfaces, such as smoothness and limb darkening. If a star has unexpected features, such as a highly irregular spot distribution or strong convection cells, the priors might bias the reconstruction. The team's comparison with interferometric data for Arcturus is encouraging, but it is only one star. A more systematic validation against a sample of stars with known surface maps is needed before the technique can be trusted broadly.

There are also practical challenges. The scattering halo is faint, and the observations require excellent sky conditions and careful calibration. The team's prototype was operated on a clear night at a dark site, but it is unclear how the technique would perform under less ideal conditions, such as with moonlight or light pollution. The data processing is also computationally intensive; a single night of observations takes roughly 10 hours of GPU time, which could be a bottleneck for routine use.

Finally, there is the question of what the scattering camera can actually resolve. The current effective resolution of about 2 milliarcseconds is comparable to a 100-meter interferometer, but it is still an order of magnitude worse than the diffraction limit of a 10-meter telescope with perfect adaptive optics. For very bright stars, such as Betelgeuse, interferometry still provides higher resolution, albeit at a much higher cost and time investment. The scattering camera is not a replacement for interferometry; it is a complementary tool that can survey many stars quickly and cheaply, identifying targets for more detailed follow-up.

These limitations do not diminish the potential of the technique, but they do temper the initial enthusiasm. The team is the first to acknowledge that the scattering camera is not a silver bullet. It is a new tool in the astronomer's kit, one that could open up routine stellar surface imaging for the first time, but it will require careful validation and refinement before it can be fully trusted. The next few years will be crucial as the team tests the camera on a wider sample of stars and compares its results with interferometric observations.