New methods to improve the decontamination of slitless spectra

This paper proposes four new methods, utilizing either local instantaneous or local convolutive approaches, to effectively decontaminate slitless spectra of stars and galaxies by leveraging direct photometric images.

The Gist

Imagine you are at a crowded, noisy cocktail party. You are trying to listen to a single friend tell you a story, but there are dozens of other people around you, all talking at once. Some people are standing right next to you, while others are shouting from across the room. To make matters worse, the room has a weird echo, and occasionally, someone drops a glass, creating a sudden, sharp "pop" of noise.

In this analogy:

• Your friend’s story is the light (spectrum) from a distant galaxy.
• The other guests are "contaminant" stars or galaxies that are overlapping with your target.
• The cocktail party noise is the "contamination" in slitless spectroscopy.
• The echo is the way the telescope's optics blur the light.
• The dropped glass is a "hot pixel"—a glitch in the camera sensor.

The Problem: The "Slitless" Mess

Most telescopes use a "slit"—think of it like a narrow hallway that only lets one person through at a time. This makes it easy to hear one person, but it’s slow because you can only talk to one guest at a time.

The Euclid space mission uses a "slitless" method. It’s like opening the doors to the entire ballroom at once. It’s incredibly fast and lets you observe millions of galaxies simultaneously, but it creates a massive mess. Because there are no "hallways" (slits) to separate the light, the spectra of different galaxies overlap and bleed into each other. It’s a giant, colorful soup of light where everything is mixed together.

The Solution: The "Super-Listeners"

The researchers in this paper have developed four new mathematical "super-listeners" (algorithms) to help astronomers untangle this soup. They use two main strategies:

1. The "Snapshot" Approach (Local Instantaneous)

Imagine trying to identify your friend’s voice by looking at a single, frozen photo of the room. You see where everyone is standing. If you know where the "noisy" guests are located, you can mathematically "subtract" their volume from the total noise.

• The "Beamformer" trick: One of their methods acts like a high-tech hearing aid. It focuses all its energy on the specific "frequency" of your friend's voice and actively tries to cancel out the background chatter.

2. The "Echo-Canceling" Approach (Local Convolutive)

This is the "Gold Standard" method described in the paper. Instead of just looking at a snapshot, this method understands the physics of the echo. It knows that the telescope doesn't just record light; it blurs it.

• It’s like having a computer that knows exactly how the room’s acoustics work. It doesn't just try to subtract the other guests; it actually "reconstructs" the original, sharp sound of your friend's voice by accounting for the blur and the overlap.

Why is this a big deal?

The researchers tested these methods using realistic, messy data that mimics what the Euclid telescope will actually see. They found that:

1. They are fast: Because they work "locally" (focusing on one object at a time), they can be run on many computers at once, like having a thousand tiny assistants working in parallel.
2. They are tough: Even when there are "hot pixels" (those sudden "glass-dropping" noises), these methods can ignore the glitch and find the galaxy underneath.
3. They are accurate: Their best method (called LC-LCMP) was significantly better at cleaning up the "soup" than previous techniques, allowing astronomers to see the true colors and properties of galaxies with much higher clarity.

In short: This paper provides the "noise-canceling headphones" for the Euclid telescope, ensuring that when we look at the deep universe, we hear the music of the galaxies, not just the roar of the crowd.

Technical Summary

Technical Summary: New Methods to Improve the Decontamination of Slitless Spectra

1. Problem Statement

In slitless spectroscopy (used by missions like Euclid, JWST, and HST), the absence of a physical slit means that light from multiple astronomical sources (stars and galaxies) can overlap on the detector. This phenomenon, known as contamination, results in superimposed spectrograms. For missions like Euclid, which aims to map the Universe to understand dark energy, accurate spectroscopic redshifts (specifically identifying the $H\alpha$ emission line) are critical. Contamination by neighboring objects or unmodeled artifacts (like hot pixels) can lead to significant errors in these redshift measurements.

Existing decontamination methods face several limitations:

• Model-subtraction methods (e.g., GRIZLI): Cannot model contaminants that are undetected in reference direct images.
• Global inversion frameworks (e.g., LINEAR): Highly computationally intensive and sensitive to noise.
• Non-negative Matrix Factorization (NMF) approaches: Often suffer from slow convergence, sensitivity to initialization, and do not utilize available photometric direct images.

2. Methodology

The authors propose four new source separation methods based on two distinct local approaches (decontaminating objects one by one to allow for parallelism). These methods simultaneously exploit multiple dispersion directions (four in the case of Euclid) to improve estimation accuracy.

A. Local Instantaneous Approach
This approach uses an approximate linear model where the observed signal at each wavelength is a linear combination of sources at the same detector position.

• Mixing Model: Assumes the object profile is separable into $x$ and $y$ coordinates.
• Methods:
  • LI-LSQ (Least Squares): Minimizes the Frobenius norm of the residual.
  • LI-LSQN (Least Squares with Non-negativity): Adds a constraint that spectra must be non-negative.
  • LI-MPDR (Minimum Power Distortionless Response): A beamforming technique that extracts the target spectrum while minimizing total power under a constraint. It is particularly robust against unmodeled interferences like hot pixels.

B. Local Convolutive Approach
This approach uses a more physically realistic model where the observed signal is the result of the convolution of sources with wavelength-dependent filters.

• Mixing Model: Formulated in the Fourier domain, allowing for simultaneous decontamination and deconvolution.
• Method:
  • LC-LCMP (Linearly Constrained Minimum Power): Uses a modified Orthogonal Matching Pursuit (OMP) algorithm to detect active contaminants at each frequency, followed by a beamformer to estimate the Fourier transform of the target spectrum.

3. Key Contributions

• Multi-directional Integration: Unlike previous methods that process dispersion directions independently, these models explicitly integrate all four available dispersion directions into a single observation matrix.
• Hybrid Information Usage: The methods leverage direct photometric images to estimate the mixing matrix (the spatial profiles of the objects), significantly improving the accuracy of the separation.
• Robustness to Unmodeled Sources: By utilizing beamforming (MPDR and LCMP), the methods can mitigate the effects of "invisible" contaminants or detector defects (hot pixels) that are not present in reference images.
• Computational Efficiency: The local nature of the models allows for high levels of parallelism, making them more scalable than global inversion methods.

4. Results

The methods were tested using realistic, noisy, Euclid-like simulated data across several scenarios:

• Simple Scenario: All methods successfully attenuated contamination. The LC-LCMP method yielded the highest Signal-to-Interference Ratio improvement ($SIR_{imp}$) and lowest error ($RMSE$).
• Hot Pixel Scenarios: In environments with high contamination and detector defects, the beamforming methods (LI-MPDR and LC-LCMP) significantly outperformed the standard LSQ methods.
• Large-scale Testing (100 objects):
  • LC-LCMP was the top performer, showing an average $SIR_{imp}$ improvement of 8.59 dB over the NMF-ALS baseline.
  • LC-LCMP provided the most accurate estimates around emission lines (lowest $LRMSE$) because it performs deconvolution.
  • LI-MPDR was identified as the fastest method in terms of computation time.

5. Significance

This work provides a robust mathematical framework for improving the data quality of upcoming large-scale spectroscopic surveys. By providing methods that are both more accurate (through convolutive modeling) and more resilient (through beamforming), the authors offer a pathway to more precise cosmological measurements. The ability to handle unmodeled contaminants is particularly vital for the success of the Euclid mission, where the density of galaxies in deep fields makes contamination an inevitable challenge.