A small team at the Max Planck Institute for Astronomy in Heidelberg noticed something odd in their archive. Spectra of young giant planets taken with the SPHERE instrument on the Very Large Telescope seemed to shift subtly depending on the year of observation. The difference was small—a few nanometres in the bandpass edges of the Y and J bands—but it had a large consequence: 14 published exoplanet atmosphere reconstructions now disagreed with each other. Some researchers saw clear methane absorption; others saw none. Both groups had used the same raw data.

A Single Coating Shift Unravels Exoplanet Atmospheres

The disagreement centers on a dichroic beam splitter inside SPHERE, an instrument that separates visible and near-infrared light. Over time, the coating on this optical element degraded, shifting the transmission edges by roughly 5–10 nanometres. The shift was not uniform across the field of view, and it affected the Y band (around 0.95–1.1 micrometres) and J band (1.1–1.3 micrometres) most strongly. Because atmospheric retrieval codes rely on the relative flux in adjacent bands to infer molecular abundances, even a small offset can produce different best-fit models.

The Heidelberg team first flagged the anomaly in an internal report, then presented it at a SPHERE consortium meeting in late 2023. Their analysis suggested that roughly 10–20 percent of published retrievals from SPHERE data might need revision. The reaction was mixed: some colleagues welcomed the caution; others argued the effect was too small to matter. The dispute quickly escalated, splitting the community into two camps.

The 14 planet spectra in question come from the SHINE survey, a large program to image and characterize young giant planets. Among them are some of the most studied objects in the field: HR 8799e, PDS 70b, 51 Eridani b, and beta Pictoris b. The credibility of a generation of exoplanet atmosphere studies now hangs on resolving the discrepancy. The SPHERE instrument has been one of the workhorses for direct imaging and spectroscopy of young giant planets since it began operations in 2014. If a single coating shift can scramble the interpretation of a whole survey, then the field must ask how many other instruments harbour similar hidden drifts.

How the Dichroic Coating Drifted Over Time

The dichroic in SPHERE is a multi-layer interference coating designed to transmit visible light to the instrument's imaging channel and reflect near-infrared light to the spectrograph. Like any thin-film coating, its performance depends on the exact thickness and refractive index of each layer. Over years of thermal cycling, exposure to vacuum, and occasional cleaning, the layers can compact or delaminate, altering the bandpass edges. For SPHERE, the shift became noticeable around 2018. Observations taken before that year show consistent flux ratios between the Y and J bands. After 2018, the ratios drift systematically. The Heidelberg team reconstructed the drift by comparing science frames to standard star spectra taken on the same nights. They found that the bandpass edges had moved by roughly 5–10 nm, with the Y-band red edge shifting toward longer wavelengths and the J-band blue edge shifting toward shorter wavelengths, effectively narrowing the gap between them.

The calibration lamps inside SPHERE were not sensitive enough to detect this drift. The team had to use sky transmission lines—telluric absorption features—as a proxy to measure the bandpass shape. This method is not part of the standard pipeline, which assumes the dichroic transmission is stable. As a result, the pipeline reduction did not correct for the temporal drift, and the error propagated into the final spectra. One reason the drift went unnoticed for years is that SPHERE's calibration plan did not include regular monitoring of the dichroic. The instrument was commissioned with a single transmission measurement, and subsequent checks focused on the detector and the optics downstream of the split. The dichroic itself was assumed to be a passive element, not a source of variability.

The 14 Planets That Split Two Research Groups

The 14 planets in question come from the SHINE survey, a large program to image and characterize young giant planets. Among them are some of the most studied objects in the field: HR 8799e, PDS 70b, 51 Eridani b, and beta Pictoris b. The dispute centers primarily on HR 8799e and PDS 70b, where the presence or absence of methane absorption has major implications for the planets' atmospheric compositions and formation histories.

Team A, which includes members of the GRAVITY collaboration, processed the data with a pipeline that corrects for the dichroic drift using a template derived from standard stars. Their reduced spectra show a clear methane absorption feature at 1.15 micrometres in HR 8799e. Team B, from the SPHERE consortium, used the standard pipeline without the drift correction and found no methane absorption. Both teams have published their results in peer-reviewed journals in 2024.

The raw data are identical. The difference lies entirely in the reduction steps. Team A's pipeline applies a wavelength-dependent correction that shifts the flux in the Y and J bands by up to 15 percent. Team B argues that this correction is overfitted and that the standard star data are too sparse to constrain the drift reliably. The debate has become technical, with each side publishing detailed response papers.

PDS 70b, a planet still embedded in its protoplanetary disk, is another flashpoint. Its spectrum shows a broad feature around 1.2 micrometres that Team A interprets as methane, while Team B attributes it to a combination of water vapor and scattering by dust. The difference matters: methane on PDS 70b would imply a carbon-rich atmosphere, which would constrain the planet's formation location and the disk's chemical evolution.

Methodology: Retrieval Codes vs. Empirical Checks

Atmospheric retrieval codes, such as PetitRADTRANS and Exo-REM, are the standard tools for interpreting exoplanet spectra. They work by generating millions of model spectra with varying temperature, composition, and cloud properties, then selecting the models that best match the observed fluxes. The codes trust the relative fluxes between bands, so any systematic error in the flux ratios will be absorbed into the retrieved parameters.

PetitRADTRANS, developed at the University of Bern, is widely used for SPHERE data. When fed the uncorrected spectra, it retrieves atmospheres with little or no methane. When fed the drift-corrected spectra, it retrieves methane-rich atmospheres. Exo-REM, developed at the CEA in France, shows a similar sensitivity, though the magnitude of the effect varies depending on the assumed cloud model. The difference is not just in methane abundance; retrieved temperatures, surface gravities, and cloud properties also shift.

Empirical checks offer an alternative. Instead of relying on retrieval codes, some researchers compare the planet spectra directly to standard star spectra taken on the same night. This approach revealed the dichroic drift in the first place. For example, the standard star HD 30501 shows a systematic offset between its measured and expected flux ratio in the Y and J bands, which correlates with the observation date. By applying the same correction to the planet spectra, the empirical method yields consistent results across epochs.

The problem is that empirical corrections are ad hoc. They depend on the choice of standard star, the fitting method, and the assumptions about the star's intrinsic spectrum. Different groups use different stars and different correction templates, leading to a proliferation of reduction pipelines. The field now faces a situation where the same raw data can produce several different spectra, each defended by its authors as the correct one.

Retrieval codes amplify systematic errors because they treat the data as exact. A 5 percent offset in a single bandpass edge can translate into a factor-of-two change in the retrieved methane abundance, because the code interprets the flux difference as a real absorption feature. The Heidelberg team's simulations show that the drift can mimic the signature of methane, leading to false positives in some cases and false negatives in others, depending on the planet's true atmospheric properties.

What the Disagreement Reveals About Instrument Science

The SPHERE dichroic story is not an isolated case. Similar drifts have been reported for other high-contrast imaging instruments. The Gemini Planet Imager (GPI) experienced a slow degradation of its integral field unit optics, which required a complex recalibration campaign. The Near-Infrared Coronagraphic Imager (NICI) on the Gemini South telescope had filter transmission changes that were only discovered after several years of science operations. What these cases share is a lack of continuous monitoring. Instruments are typically characterized during commissioning, then assumed to be stable for years. But optical coatings age, detectors degrade, and mechanical alignments shift. The assumption of stability is convenient but often wrong. The SPHERE consortium did not include a regular calibration sequence for the dichroic after commissioning, because no one anticipated a measurable drift.

The consequence is that archival science data cannot be fully recalibrated retroactively. The dichroic transmission was not measured at the time of each observation, so the drift must be inferred from the science data themselves, creating a circularity. The Heidelberg team's method of using standard stars is the best available, but it relies on the assumption that the standard stars are constant, which may not hold for all of them. Lessons from this episode are percolating through the astronomical community. Instrument scientists now advocate for continuous monitoring of all optical elements, not just the detectors. The European Southern Observatory, which operates the VLT, has started a program to install on-sky monitors for critical components. But retrofitting existing instruments is expensive, and the next generation of instruments, such as METIS for the Extremely Large Telescope, are being designed with embedded monitors from the start.

A comparable case occurred in oceanography: an unadjusted tidal gauge datum flattened 14 Pacific sea-level records, as reported by the National Oceanic and Atmospheric Administration. The pattern is the same: a small, uncalibrated change in a measurement device propagates into a large systematic error that goes unnoticed for years. The lesson is that instrument characterization must be continuous, not a one-time event.

Practical Fixes for Future Surveys

What can be done to prevent a recurrence? The most straightforward fix is to install an on-sky monitor that uses telluric absorption lines to measure the bandpass shape of the dichroic on every night of science operation. This would provide a real-time record of the instrument's state, allowing data reduction pipelines to apply a date-dependent correction. Several groups are developing such monitors using fiber-fed spectrographs that observe a small patch of sky near the target.

A second fix is to include a daily dichroic transmission measurement as part of the calibration plan. This could be done by inserting a calibrated light source into the beam path before the dichroic, or by using a dedicated calibration unit that scans the transmission as a function of wavelength. The cost for such a system is modest—roughly €50,000–100,000 per instrument per year, including personnel time—compared to the science output of a major facility.

A third, more fundamental change is to publish raw pixel data, not only reduced spectra. If the raw data are archived, future researchers can reprocess them with improved calibrations. SPHERE data are already archived, but the pipeline products that most scientists use are the reduced spectra. Encouraging the community to work with raw data would require better data management infrastructure and a cultural shift toward open data.

Finally, the adoption of open-source reduction pipelines with version control would allow the community to track changes and reproduce results. The SPHERE pipeline is closed-source, which means that only the consortium members can modify it. An open-source pipeline, maintained by the community, could incorporate drift corrections more quickly and transparently. Some instruments, like the Keck Planet Imager and Characterizer, already operate this way.

These fixes are not expensive, but they require foresight. The next generation of exoplanet instruments, such as METIS on the ELT and HARMONI, are being designed with embedded monitors and open-source pipelines. The SPHERE episode serves as a cautionary tale that a single overlooked component can undermine years of work.

Toward a Community-Agreed Correction

The 14 planet atmosphere reconstructions from SPHERE are now provisional. The community is gradually adopting empirical correction templates, and several groups are re-analyzing the entire SHINE sample with drift-corrected pipelines. Preliminary results suggest that the coating shift explains roughly 30–50 percent of the inter-team variance in the retrieved methane abundances. The remaining variance is due to differences in retrieval codes, cloud models, and data reduction choices.

The field is moving toward a more cautious approach. New papers on SPHERE planets now include a statement about the dichroic drift and its potential impact. Some journals require authors to specify the calibration version used. The concept of "instrument provenance"—a complete record of the instrument's state at the time of observation—is gaining traction, analogous to the provenance metadata used in computational workflows.

For the two teams at the center of the dispute, the path forward is collaborative. A joint working group has been formed to compare reduction pipelines on a common set of planets, with the goal of producing a community-agreed correction. The process is slow, but it is happening. The Heidelberg team's original report, once met with skepticism, is now cited as a standard reference.

The next generation of instruments, METIS and HARMONI, will embed continuous monitors for all critical optical elements. But the lesson from SPHERE is that even the best-designed instruments can develop hidden problems. Trust the data, but verify the instrument. The story of the dichroic coating shift is a reminder that science advances not only through new discoveries, but also through the painstaking work of understanding our own tools.

As one Heidelberg team member put it: "We spent years perfecting our retrieval codes, but we forgot to check the mirror." The field is now checking the mirror—and finding that it matters.