🔭 astrophysics

"Perfect" spectra for ESO's HARPS spectrograph

This paper introduces HARPERFECT, a novel method that reconstructs 1D spectra from HARPS 2D detector data using a forward model based on laser frequency comb observations to achieve superior wavelength calibration and order tracing, thereby providing an exact resolution matrix and independent samples essential for precise measurements of fundamental constants and exoplanet atmospheres.

Original authors: Dinko Milaković, Guido Cupani, Bruce A. Bassett, Stefano Cristiani, Luca Pasquini

Published 2026-07-09

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Dinko Milaković, Guido Cupani, Bruce A. Bassett, Stefano Cristiani, Luca Pasquini

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to take a perfect photograph of a distant, faint star. But there's a problem: your camera lens is a bit warped, the sensor pixels are slightly different sizes, and the light doesn't hit the sensor in a neat, straight line. Instead, the light smears out in a messy, unpredictable blob.

In the world of astronomy, this "messy blob" is called the Point-Spread Function (PSF). For decades, astronomers have used standard methods to "extract" the true image from this mess. These methods work well enough for most things, but they make a big assumption: they assume the blur is perfectly symmetrical and straight, like a clean circle. In reality, the blur is often tilted, stretched, or wobbly. This leads to tiny errors in measuring exactly where a line of light is and how bright it is.

This paper introduces a new, ultra-precise method called "Harperfect" (short for "Harps Perfect") to fix this. Here is how it works, using simple analogies:

1. The Problem: The "Smudged" Photo

Think of the telescope's detector (the camera sensor) as a giant grid of tiny buckets catching rain (light).

Standard Method: The old way of processing data assumes that if a drop of rain falls in one bucket, it only affects that bucket and its immediate neighbors in a predictable, straight line. It's like assuming the rain falls in perfect, straight columns.
The Reality: Because of the telescope's optics, the rain actually splashes diagonally and unevenly. A drop in one bucket might splash heavily into a bucket far away, but lightly into one right next to it. The standard method misses these complex splashes, leading to a slightly blurry or inaccurate final picture.

2. The Solution: The "Laser Comb" Map

To fix this, the researchers needed a perfect map of how the light smears across the camera. They couldn't use a normal star for this because stars are too fuzzy. Instead, they used a Laser Frequency Comb (LFC).

The Analogy: Imagine you have a ruler, but instead of just having marks every inch, you have a ruler with millions of incredibly precise, perfectly spaced lines. This is what the Laser Comb is: a source of light that creates thousands of perfectly known "lines" across the detector.
The Process: Because the laser lines are so sharp and known, the researchers could see exactly how the telescope's "blur" distorted each one. They treated every single laser line as a tiny test subject. By looking at how 21,000 of these lines were stretched and tilted, they built a 3D map of the camera's imperfections.

3. The "Spectral Perfectionism" Engine

Once they had this map, they built a new software engine called Harperfect.

How it works: Instead of guessing how the light is smeared, Harperfect uses the map to run a "reverse engineering" simulation. It asks: "If the light landed here, and we know exactly how the lens smears light, what did the original light look like?"
The Magic: It doesn't just guess; it calculates the exact mathematical relationship between the messy data on the camera and the clean, true spectrum of the star. It creates a "resolution matrix," which is like a master key that unlocks the true shape of the light, removing the smearing effects that other methods leave behind.

4. The Results: Sharper, but with a Catch

The team tested this new method on a famous quasar (a super-bright, distant galaxy) called HE0515−4414.

The Good News: The new method successfully reconstructed the spectrum with a level of detail that standard methods can't guarantee. It provided a "perfect" resolution matrix, meaning the data points are independent and the math is exact. This is crucial for measuring tiny shifts in light that could tell us about the fundamental laws of the universe or the atmosphere of alien planets.
The Catch: The new method produced a signal that was about 1% to 5% noisier (lower signal-to-noise ratio) than the standard method.
- Why? The researchers admit their "map" wasn't 100% perfect. They modeled the blur as a simple "Gaussian" (a smooth bell curve), but the real blur is a bit more complex and lopsided. Because the map wasn't perfect, the math calculated the uncertainty slightly wrong, making the data look a bit "fuzzier" in terms of noise.
The Visuals: When they looked at specific lines of light (like iron or magnesium), the new method showed some lines as slightly sharper and narrower than the standard method, though they didn't do the heavy math to prove it statistically yet.

Summary

Think of this paper as the first time someone built a GPS for a telescope's lens.

Before: Astronomers drove with a rough paper map, assuming the roads were straight. They got to the destination, but sometimes took a wrong turn or missed a detail.
Now: They have a live, high-definition GPS (the Laser Comb map) that tells them exactly how the road curves and twists. They can now navigate the light with extreme precision.
Current Status: The GPS is so new and detailed that the car's computer is still learning how to interpret the data perfectly, so the ride is a tiny bit bumpier (noisier) than the old way. But the path is now mathematically "perfect," paving the way for future upgrades that will make the ride smooth and precise.

The ultimate goal isn't just to take a pretty picture, but to measure the universe with such precision that we can detect the tiniest changes in the laws of physics or the composition of planets light-years away. This paper is the first step in building that ultra-precise ruler.

Technical Summary: "Perfect" Spectra for ESO's HARPS Spectrograph

Problem Statement
Standard spectral extraction for astronomical spectrographs relies on "optimal extraction," a method developed in the 1980s that assumes spectral traces align perfectly with the detector grid and that the Point-Spread Function (PSF) is separable (independent) in the dispersion ( $x$ ) and cross-dispersion ( $y$ ) directions. In practice, this assumption is rarely satisfied, particularly for high-resolution echelle spectrographs like HARPS. When the PSF is non-separable, standard extraction introduces errors in flux estimation, wavelength calibration, and resolution. These errors are detrimental to precision science requiring accurate measurements of line shapes and centers, such as studies of fundamental physical constants and exoplanet atmospheres.

Methodology: Spectral Perfectionism (SP)
The authors present "Harperfect," the first application of "spectral perfectionism" (SP) to a high-resolution echelle spectrograph. SP is a forward-modeling approach that reconstructs the intrinsic 1D spectrum ( $\hat{f}$ ) from 2D detector data ( $p$ ) using a calibration matrix ( $A$ ) and a noise covariance matrix ( $N$ ). The model is defined as $p = Af + n$.

Key methodological components include:

Calibration Matrix Construction: Unlike previous SP implementations that relied on complex instrument models, Harperfect constructs the matrix $A$ entirely from observations of a Laser Frequency Comb (LFC). The LFC provides $\approx 10,000$ lines with known frequencies. Since HARPS cannot resolve individual LFC lines (the lines are broader than the instrument resolution), each observed line serves as a direct sample of the pixel-integrated PSF at a specific detector position and wavelength.
PSF Modeling: The authors fitted 21,148 identified LFC lines using a bivariate (2D) Gaussian function with seven free parameters (amplitude, center, standard deviations $\sigma_x, \sigma_y$ , rotation angle $\theta$ , and offset). This allows for the mapping of PSF shape variations across the detector.
Wavelength and Trace Calibration: Using the known LFC frequencies and the fitted line centers, the authors performed Gaussian Process regression to create a 2D wavelength calibration and to trace the geometric centers of echelle orders.
Resolution Matrix and Deconvolution: To avoid the "covariance ringing" (anti-correlated noise) inherent in direct inversion of the forward model, the authors apply a resolution matrix $R$ . They derive $R$ from the symmetric matrix square root of the inverse covariance matrix ( $C^{-1}$ ), ensuring flux conservation and preserving the native instrumental resolution without the $\sqrt{2}$ broadening penalty associated with simple auto-correlation.
Computational Implementation: To manage computational complexity, the pipeline processes localized "ribbons" (approx. 60 pixels high) of the detector rather than the full frame simultaneously. It utilizes sparse Cholesky decomposition for linear inversion.

Key Contributions

LFC-Based Calibration: The paper demonstrates the use of LFC observations to build a complete calibration matrix (PSF shape, wavelength, and order traces) without relying on traditional arc lamps or flat fields for the core SP parameters.
PSF Mapping: The authors mapped PSF variations across the HARPS detector with unprecedented detail (sampling every $\approx 15$ physical pixels). They quantified the non-separability of the PSF using a correlation coefficient $\rho$ , finding it to be weakly non-separable ( $|\rho| < 0.1$ ) but measurable across the detector.
Order Tracing Accuracy: The order traces derived from LFC line centers were found to better align with the visual geometric center of the orders compared to the standard HARPS Data Reduction Software (DRS) traces, which are derived from tungsten lamp observations.
Software Implementation: The release of "Harperfect," a Python pipeline capable of applying SP to HARPS data, including the generation of mock PSFs at arbitrary positions via interpolation.

Results
The method was applied to 52.5 hours of HARPS data on the quasar HE0515−4414.

Signal-to-Noise (S/N): The S/N of the Harperfect spectra is consistently 1% to 5% lower than that of the standard HARPS DRS optimal extraction. The authors attribute this not to a failure of the SP framework, but to the limitations of their current PSF model. The bivariate Gaussian model does not fully capture the PSF asymmetry and background variations, leading to imperfect uncertainty estimates in the calibration matrix $A$ .
Line Profiles: Visual comparison of six metal absorption transitions showed that Harperfect and DRS spectra agree closely for most lines. However, for two transitions (Fe II $\lambda$ 2600 and Mg II $\lambda$ 2804), the Harperfect profiles appeared visually narrower near the line core. The authors caution that without quantitative Voigt profile fitting, this is a preliminary visual observation and not a confirmed resolution gain.
Calibration Precision: The LFC-based wavelength calibration matches the independent ThAr lamp calibration from the DRS within 250 m s $^{-1}$ . Residuals show a periodic structure ( $\pm 25$ m s $^{-1}$ ) every 512 pixels, corresponding to known pixel size anomalies in the HARPS detector.

Significance and Claims
The paper claims that Harperfect represents a significant step toward "perfect" spectral extraction by providing an exact resolution matrix and independent samples, properties that standard extraction cannot guarantee for non-separable PSFs.

The authors are modest in their claims regarding immediate performance gains. They explicitly state that the current S/N penalty is a consequence of the simplified PSF model (bivariate Gaussian) and that the full benefits of spectral perfectionism will only be realized with more accurate PSF modeling, better background handling, and compensation for pixel size anomalies. The primary value of this work is demonstrated as a proof-of-concept: it validates the feasibility of using LFCs to construct the necessary calibration matrices for SP and establishes a framework for future improvements in measuring fundamental constants and exoplanet atmospheres. The authors note that while the current processing time (approx. 2 hours for a full spectrum) is slower than standard extraction, the potential for increased spectral fidelity justifies the computational cost for precision science.

1. The Problem: The "Smudged" Photo

2. The Solution: The "Laser Comb" Map

3. The "Spectral Perfectionism" Engine

4. The Results: Sharper, but with a Catch

Summary

More like this