A Census of Protostellar Outflows in Nearby Molecular Clouds

Duo Xu, Stella S. R. Offner, Robert Gutermuth, Shuo Kong, Héctor G. ArceThe Astrophysical Journal 2022 · DOI

TL;DR

This paper uses a deep learning model (CASI-3D) to create the most comprehensive census of protostellar outflows in nearby molecular clouds to date. They find that outflow mass and energy are directly proportional to the number of young stars, and that these outflows are energetic enough to sustain turbulence within the clouds.

What Problem Does This Paper Address?

Understanding how stars form and evolve requires us to study the energetic outflows that protostars launch. These outflows play a crucial role in regulating star formation, influencing the distribution of stellar masses, and driving turbulence within molecular clouds. However, accurately identifying and quantifying these outflows, especially those deeply embedded within dense gas, has been a significant challenge. Previous methods often relied on visual inspection or simpler automated techniques, leading to uncertainties and potential biases. This paper aims to overcome these limitations by employing a sophisticated deep learning approach to systematically identify and characterize protostellar outflows across multiple nearby molecular clouds.

Methods

The core of this work is the application of CASI-3D (Convolutional Approach to Structure Identification-3D), a deep learning model based on convolutional neural networks (CNNs). CASI-3D is an encoder-decoder architecture that has been trained to identify protostellar outflows in 3D molecular line data (specifically, position-position-velocity or PPV data cubes). The authors use two versions of the model: ME1, which identifies outflow emission, and MF, which estimates the fraction of mass associated with feedback.

The process involves:

  • Data Preprocessing: Taking raw observational data, applying a logarithmic transformation to handle the large dynamic range of CO emission, and then normalizing the data.
  • Chunking: Dividing the large 3D data cubes into smaller, overlapping 3D chunks for the CNN to process.
  • Prediction: Running the CASI-3D models on these chunks to predict the presence and extent of outflows.
  • Mass and Energy Calculations: Estimating outflow mass and energy using both 12CO and 13CO emission, assuming 13CO is optically thin and 12CO is optically thick. They also account for potential uncertainties due to excitation temperature and inclination angles.
  • Spatial Power Spectrum (SPS) Analysis: Investigating the impact of outflows on the structure of the molecular clouds by analyzing the SPS of the integrated intensity maps.

Datasets

The study utilizes observational data from four nearby molecular clouds:

  • Ophiuchus
  • Taurus
  • Perseus
  • Orion

These observations were primarily conducted using the 12CO J=1-0 and 13CO J=1-0 lines from the Five College Radio Astronomy Observatory (FCRAO) Telescope and the Nobeyama Radio Observatory (NRO) 45m telescope.

The authors also use YSO (Young Stellar Object) catalogs for each cloud to validate their outflow identifications:

  • Ophiuchus and Perseus: SESNA (Spitzer Extended Solar Neighborhood Archive) YSO catalog.
  • Taurus: YSO catalog from Rebull et al. (2010).
  • Orion: YSO catalog from Megeath et al. (2012).

Key Insights and Ideas

  • Outflows are ubiquitous and scale with star formation: The study finds a strong linear correlation between the total outflow mass (and momentum/energy) and the number of YSOs in each cloud. This suggests that the amount of outflow material is directly tied to the rate of star formation.
  • Outflows are energetic enough to sustain turbulence: The calculated outflow energies are sufficient to offset the turbulent dissipation within these molecular clouds, implying that outflows are a key mechanism for maintaining the turbulent state of the interstellar medium.
  • Deep learning can uncover hidden structures: CASI-3D's ability to identify outflows even when they are faint or embedded in dense gas is a significant advancement over previous methods. This allows for a more complete census.
  • Spatial Power Spectrum reveals injection scales: The break point observed in the spatial power spectrum of the outflow predictions likely represents the typical scale at which mass and energy are injected into the cloud by outflows.
  • Distinguishing outflows from bubbles is challenging: While CASI-3D can identify feedback structures, differentiating between outflows and stellar wind-driven bubbles can be difficult, especially when viewed along the outflow axis.

Key Results

  • Total Outflow Masses:
    • Ophiuchus: 267 M☉
    • Taurus: 795 M☉
    • Perseus: 1305 M☉
    • Orion: 6332 M☉
  • Total 3D Outflow Energies:
    • Ophiuchus: 9 × 10⁴⁵ ergs
    • Taurus: 6 × 10⁴⁶ ergs
    • Perseus: 1.2 × 10⁴⁷ ergs
    • Orion: 6 × 10⁴⁷ ergs
  • Outflow Mass per YSO: On average, each YSO is associated with approximately one solar mass of outflow material.
  • Mass-weighted LOS Velocity: The outflow velocity is estimated to be around 1 km/s, which is lower than some previous estimates due to the inclusion of mass from lower-velocity gas near the cloud's center.
  • Spatial Power Spectrum Break Point: The break point, indicating typical outflow injection scales, ranges from 0.27 pc (Orion) to 0.65 pc (Perseus).

Limitations and Future Work

  • Uncertainties in Mass Estimates: The mass calculations are subject to uncertainties from the assumed excitation temperature (factor of 2), inclination angle (factor of 2-4 for momentum/energy), and chemical conditions (abundance ratios). The authors estimate an overall uncertainty of a factor of 3 for mass, 3.4 for 3D momentum, and 5 for 3D energy.
  • Contamination from Dense Gas: While the CASI-3D model is designed to exclude contamination from ambient gas near rest-frame velocities, there's a potential for underestimation of outflow contributions from low-velocity gas in complex regions.
  • Distinguishing Outflows and Bubbles: The study acknowledges that models trained to identify outflows can also identify bubble-like structures, and distinguishing between them based solely on spectra can be challenging.
  • Model Limitations: The authors note that machine learning models are not perfect and require careful application and validation. They present examples of potential false detections where non-outflow structures mimic outflow morphology.
  • Future Work: The paper suggests further investigation into the impact of outflows on molecular clouds, including a more detailed comparison with stellar wind feedback and exploring the role of different chemical conditions. The development of more sophisticated models to disentangle outflows from other gas structures and to better account for inclination effects would also be beneficial.

Summary updated

Abstract

Abstract We adopt the deep learning method casi-3d (Convolutional Approach to Structure Identification-3D) to systemically identify protostellar outflows in 12 CO and 13 CO observations of the nearby molecular clouds, Ophiuchus, Taurus, Perseus, and Orion. The total outflow masses are 267 M ⊙ , 795 M ⊙ , 1305 M ⊙ , and 6332 M ⊙ for Ophiuchus, Taurus, Perseus, and Orion, respectively. We show the outflow mass in each cloud is linearly proportional to the total number of young stellar objects. The estimated total 3D deprojected outflow energies are 9 × 10 45 erg, 6 × 10 46 erg, 1.2 × 10 47 erg, and 6 × 10 47 erg for Ophiuchus, Taurus, Perseus, and Orion, respectively. The energy associated with outflows is sufficient to offset turbulent dissipation at the current epoch for all four clouds. All clouds also exhibit a break point in the spatial power spectrum of the outflow prediction map, which likely corresponds to the typical outflow mass and energy injection scale.
BibTeX
@article{xu2022census,
  author = {Duo Xu and Stella S. R. Offner and Robert Gutermuth and Shuo Kong and Héctor G. Arce},
  title = {{A Census of Protostellar Outflows in Nearby Molecular Clouds}},
  journal = {The Astrophysical Journal},
  year = {2022},
  doi = {10.3847/1538-4357/ac39a0},
}
Figure