ORION-B project results

Seminar/presentation: Slides: ppt Paper: link Poster: poster

5. Quantifying the informativity of emission lines to infer physical conditions in giant molecular clouds. I. Application to model predictions

Observations of spectral lines are performed to improve our understanding of the interstellar medium (ISM). However, assessing quantitatively the potential of a line to constrain the physical conditions of the ISM is complex, even more so for combinations of lines. Observation campaigns thus usually observe as many lines as possible for as long as possible.

In this paper, we propose an approach that quantifies how well observing one or more lines can constrain a physical parameter. To do so, we resort to information theory and in particular mutual information. In fact, this statistic quantifies the average reduction in uncertainty on a given physical parameter resulting from the observation of a given set of lines. We illustrate our approach on synthetic observations — produced from an emulator of the Meudon PDR code and a simulator of IRAM 30m observations — to constrain the cloud visual extinction A_V and the intensity of the incident UV field G₀. We study how mutual information evolves with integration time, which could allow observers to optimize their proposals to achieve the highest constraining power with the minimum time budget. We also propose visualizations, called "information maps", of how mutual information evolves as a function of the physical regime (A_V, G₀) for a given line or set of lines. We show that combining lines provides information in regimes where none of the individual lines is informative, thanks to line complementarities (see figure). Finally, we compare mutual information values for multiple lines and sets of lines to determine which one best constrains a physical parameter. This could allow observers to select the most suitable lines to observe for a given physical environment. Due to line complementarities, the most informative set of lines does not necessarily combine the most informative individual lines.

Our results were obtained using this Python package. For you to perform a similar study in your context, we published the more general InfoVar package. To install it, simply run pip install infovar.

Figure: Comparison of information maps between millimeter lines and visual extinction A_V. The first two columns of each row show the information brought by individual lines. The third column shows the maximum information brought by these lines taken individually, while the fourth column shows the information brought by their combination. Finally, the fifth column shows the difference between these two latter maps that can be interpreted as the gain in information brought about by the combination compared to the more informative individual lines.

4. Bias versus variance when fitting multi-species molecular lines with a non-LTE radiative transfer model: Application to the estimation of the gas temperature and volume density

The technological progress of millimeter-wavelength telescopes will soon make it common to observe multiple molecules over large fields of view, but with only few transitions. Assuming all these lines are emitting from the same element of interstellar gas allows us to constrain the estimation of physical parameters of the gas, such as temperature and volume density. However, when this assumption is not verified, it may generate some important biases. We explore this challenge based on the data of the ORION-B project.

We use a radiative transfer model (RADEX) to estimate the gas characteristics in the Horsehead Nebula, by fitting different sets of lines emitted by CO and HCO⁺ isotopologues and study the theoretical accuracy that can be reached depending on the spectral lines and on the targeted region characteristics. We observe that the column densities of ¹³CO and C¹⁸O can be estimated with a 30% accuracy, while the volume density of H₂ can typically only be obtained within a factor of 2. As shown in the figure, the fitted volume density and the estimation accuracy strongly depend on the considered chemical species. To explain these differences, we demonstrate with simulations that the estimations can be strongly biased in case of an inappropriate chemical assumption.

Figure: Depending on the considered set of species, the map of estimated volume densities in the Horsehead changes dramatically. The pixels for which the estimation accuracy is insufficient have been left in white.

3. Neural network-based emulation of interstellar medium models

The interpretation of line observations from the interstellar medium (ISM) requires comparisons with astrophysical numerical models to infer physical conditions such as thermal pressure, kinetic temperature or visual extinction. Usually, ISM models are too time-consuming for Bayesian inference procedures that require thousands of calls. As a result, they are often replaced by interpolation of a grid of precomputed models, the accuracy of which is rarely assessed.

We propose a new general method to derive faster (from more than ten hours to a few milliseconds for a calculation), lighter and typically two to three times more accurate approximations of the model. These emulators are defined with artificial neural networks. The specificities inherent to ISM models are addressed to design and train adequate neural networks. For example, the problem of anomalies in model outputs is handled by automatically detecting and then ignoring them during training, in order to achieve a more robust version of the model. Other properties such as the important number of outputs or their redundancy are also taken into account, for instance by dividing them into clusters according to their similarity, then jointly training dedicated, simpler networks for each cluster.

We compare the proposed neural networks with standard interpolation methods, such as nearest neighbors, linear, spline, or radial basis function to emulate a representative ISM numerical model, the Meudon PDR code. This work will enable efficient inferences on million-pixel images of multiline observations.

The documentation of the code is available online and it can be easily installed through pip install nnbma.

Figure: Comparison between some interpolation and regression methods on a mathematical function for two different types of training sets (grid and random).

2. Deep learning denoising by dimension reduction: Application to the ORION-B line cubes

The study of stellar formation and galaxy evolutions is being profoundly transformed by the commissioning of new receivers that can probe the detailed atomic and molecular composition of astrophysical media, as well as the kinematics of the emitting gas. Most of the new instruments provide massive data cubes that contain a lot of information spatially and spectrally. However, their large size coupled with an inhomogeneous signal-to-noise ratio are major challenges for consistent analysis and interpretation. Current leading algorithms are unable to extract all the information from noisy data. It is therefore necessary to invent more modern methods at the frontier between astrophysics, signal processing and data science to get the maximum scientific return of these new observations.

Autoencoders neural networks have been successfully used to denoise data Earth remote sensing hyperspectral data. We have taken this type of approach steps further by taking advantage of the statistical properties of astrophysical data cubes. In particular, the absence of redundancy between distant spectral channels and the sparsity of the signal have been used. This results in a new algorithm based on an innovative architecture and loss function.

Our new method leads to an increase in the signal-to-noise ratio in data with a weak signal, allowing us to use faint but very informative signals. At the same time, it preserves the spectral shape of the profiles with a high signal-to-noise ratio, contrary to previous state-of-the-art methods such as ROHSA, a Gaussian profile decomposition algorithm. Indeed, we show that our method distorts the signal less, providing more reliable atomic and molecular line profiles. It has been applied to data from the IRAM 30m large program ORION-B, but is perfectly usable on data cubes from other modern telescopes and instruments (VLT/MUSE, JWST, ALMA, etc).

Figure: Comparison between autoencoder (first row) and ROHSA (second row) denoisings of a ¹³CO (1—0) velocity channel. The last column shows the residuals of the methods and illustrate the fact that previous state-of-the-art algorithm can distort the signal, whereas the autoencoder does not.

1. A dynamically young, gravitationally stable network of filaments in Orion B

Gaseous filaments are a prominent feature of molecular clouds, as they contribute to channelling the diffuse gas from the cloud envelope into the dense cores which will eventually form stars. Their formation, structure and evolution are however not well understood. The ORION-B data allow us to identify and characterise in detail the shape, dynamics and molecular content of an entire network of filaments in the Orion B cloud, the first such large-field study of star-forming filaments. Compared to filaments studied in other clouds, the filaments in Orion B appear to be of typical dimensions (about 0.1 pc diameter) and the more massive ones are, as expected, linked to the presence of star-forming cores (as shown in the figure). However, they also display unusually low masses, which challenges our understanding of their formation: their presence in the entire cloud suggests that they play an major role in the structure of the gas even at low densities where gravity does not dominate over turbulence. Their density and velocity profiles also show that they actively contribute to the dissipation of turbulence. This further indicates how important filaments are in the transition between a turbulent cloud of gas and quiescent, pre-stellar cores.

Figure: the filaments in Orion B rarely reach densities high enough to collapse into star-forming cores. Newly formed stars in the cloud are marked as diamonds, the youngest ones in green — most of them follow closely the position of dense filaments.