Chemistry here and there: Comparison between the CMZ of the Milky Way and NGC 253

For decades, chemical studies in external galaxies have been limited by sensitivity, resolution, bandwidth or a combination of them. A decade of ALMA operations, made the detailed study of the molecular abundances in nearby galaxies a routine job. We can finally bridge the ISM chemistry from Galactic to extragalactic environments, through one to one comparison, albeit probing significantly different scales. The ALMA large program ALCHEMI imaged the chemistry of the starburst galaxy NGC 253 at an unprecedented combination of sensitivity and resolution, achieving a detailed spatially resolved chemical inventory. ALCHEMI probed molecular differences among regions within its CMZ, and allowed the comparison with Galactic environments ranging from hot cores, quiescent molecular clouds and comets. This comparison can be put in the context of available data towards other extragalactic environments. Globally, the comparisons presented here show an astoundingly good correlation between the chemistry in Galactic GMCs and the central molecular zone of NGC 253, ranging scales from 1 pc up to ∼ 0.5 kpc. The wideband sensitivity upgrade of ALMA will expand similar comparisons towards a wider range of Galactic and extragalactic objects.

The ALCHEMI project was an ALMA large program designed not only to have the most comprehensive molecular study at giant molecular cloud (GMC) scale in an extragalactic starbursting environment, but to enable an astrochemical connection to the Galactic center. The target was the nearby prototypical starburst galaxy NGC 253, one of the most prolific extragalactic molecular emitters, and consisted of an unbiased spectral line imaging survey (84 to 374 GHz, ranging ALMA Bands 3 through 7) of the whole central molecular zone (CMZ, 600 pc × 300 pc) at an unprecedented combination of resolution (1.6′′) and sensitivity (10 − 20 mK).

Of particular relevance is the detection of Phosphorus Nitride (PN) due to its relevance as prebiotic chemistry precursor. The observed subthermal excitation and PN/SiO abundance ratio both follow the trend found towards the Galactic Center molecular clouds, confirming the shock origin of PN in both environments. In fact, the shocks across the whole CMZ of NGC 253 probed through various molecular proxies, and its evolution was studied through the SLED analysis of HNCO and SiO, under the assumption of both species being originated in a single shock event. While SiO traces fast shocks with kinetic temperatures of a few 100 K, HNCO traces denser and cooler slow shocks with Tkin ∼ 100 K.

Various ALCHEMI works made use of different molecular probes and their comparison to state-of-the-art chemical models to derive the pervading cosmic ray ionization rate. All these probes, C2 H abundances, HCO+ /HOC+ ratios, H3 O+ /SO ratio, and HCN/HNC ratio, yield similarly extreme rates of ζ ∼ 103−4 ζ0, where ζ0 ∼ 10−17 s−1 is the Milky Way canonical rate towards the Galactic spiral arms. More importantly, this is similar or even an order of magnitude higher than the rate towards the Galactic center.

However, the largest potential of unbiased spectral line surveys like ALCHEMI results from the study of multiple molecular probes. Tanaka et al. (2024) modeled 11 bright dense molecular tracers to image the physical properties across the CMZ and compared the properties with those found in the GC. NGC 253 showed ten times more high density gas and three times higher dense mass fraction. While this is not enough to justify the 30 times higher star forming rate in NGC 253, the larger fraction of gas with densities above 104−5 cm−3 might be the key for this star formation enhancement.

The work by Harada et al. (2024) embarked into the largest sample of molecular species and unblended transitions (44 and 148, respectively) ever used into a Principal Component Analysis (PCA) for any Galactic or extragalactic source. The results showed the clear differences between the species tracing young and evolved starburst regions, as well as the different morphologies of the low velocity shocks and star formation dominated locations. More importantly, it could be appreciated the evolution of the projection of the individual transitions of a given species into the different principal components as a function of its upper energy level. The follow up work by Kishikawa et al. (2025) showed the potential of the non-negative matrix factorization (NMF) in better disentangling the different structure components, usually appearing combined as positive or negative structures in the principal components in PCA.

The work by López-Gallifa et al. (Poster S27, in Prep.) aimed to perform a one-to-one comparison with similar broadband millimeter spectral surveys existing towards Galactic sources, finally making the bridge between Galactic and extra-galactic astrochemisty. This work analyzed spectra towards the 4 brighter GMCs in the CMZ of NGC 253 to analyze their molecular composition. Although the model included more than 150 species, only 35 of them were used in the comparison with Galactic sources. Abundances were opacity corrected and when available, the optically thinner isotopologues were used to derive the column densities of the brightest, optically thick, species. Also isotopic ratios were calculated for all observed isotopologues, providing accurate estimates of atomic isotopic ratios.

The first result of this work is the astounding homogeneity observed across the CMZ. All analyzed GMCs show a very close correlation of their molecular abundances indicating a homogeneous chemistry. This similarity also holds at different scales when compared the 28 pc resolution data from ALCHEMI, with the 7 m data alone at ∼ 255 pc resolution, or IRAM 30m single dish observations at ∼ 350 − 500 pc scales. This result is not different from what had been previously reported towards the center of the Milky-Way. The study from Requena-Torres et al. (2006) using a sample of 8 organic species towards 40 GC molecular clouds showed uniform abundance ratios both in GC clouds and Galactic hot cores. Moreover it was claimed that this might imply a similar average composition of grain mantles in both types of regions. Similarly, a study using 20 molecular species over a more limited sample of 11 GC clouds further supported this homogeneity. That study showed that HNCO, among all sample species, showed the largest abundance dynamic range, turning this species into one of the best diagnostic probe to trace dense shocked gas with high degree of shielding against the pervading UV radiation from newly formed stars.

Even more interesting than the relative homogeneity across the central molecular zone of NGC 253, is the comparison with the Galactic center GMC G+0.693-0.027 (hereafter G+0.693). This GC source is located in the Sgr B2 complex, north of the B2(N) and B2(M) hot cores. While there is not trace of star formation in this cloud, its chemistry appears to be driven by the low-velocity shocks originated by large-scale cloud-cloud collisions. This source has become one of the most promising laboratories for the search of complex organic molecules in the interstellar medium. In fact, more than 30 species have been detected over the past few years towards G+0.693.

The similarity between the observed chemistry in G+0.693 and NGC 253 over two orders of magnitude in spatial scales implies that the complex chemistry must be dominated by very extended emission and not constrained to the densest star forming cores even in the starburst environment. In fact the single dish observation of the whole Sgr B2 region show that the integrated molecular emission in this complex is mostly extended and only a relatively small fraction of the emission is concentrated in the hot cores and the molecular clouds G+0.693 and G+0.633. As shown in Fig. 3, only ∼ 35% of the integrated emission of the dense gas tracer HC3 N is located within the hot cores and the two GMCs. More interestingly, an even lower percentage (∼ 17%) of the HNCO emission is observed not to be widespread, indicating the vast majority of the chemical complexity in this region is released by shocks over extremely extended regions.

Such extended emission dominating the central molecular zone of NGC 253 is also found to be similar to that observed in other galactic nuclei, either dominated by star formation or by a nuclear active galactic nuclei, since the hot-core or AGN effect on the chemistry is restricted to much smaller scales. As such, the emission towards various types of nearby galactic nuclei appears to be dominated by G+0.693-like GMCs.

The work by López-Gallifa et al. (in Prep.) presented here and the previous Galactic studies mentioned above appear to suggest a homogeneous chemistry driven by a “Universal” dust chemistry, released into the ISM by large galactic scales shocks. This result does not imply that specific local physical conditions cannot be probed by the chemistry of specific species, since the deviation from the unity in Fig. 1 can be of more than an order of magnitude. This result implies that globally, the observed chemistry is dominated by extended shock dominated molecular gas, and it is observed to be statistically similar across a wide range of scales.

The ALMA wideband sensitivity upgrade (WSU) will be yet another turning point in the field of astrochemistry. This upgrade will affect the whole ALMA signal chain from receiver upgrades, digital signal processing, transmission, down correlator and computing infrastructure. This massive upgrade endeavor will provide an increase in the correlated bandwidth and sensitivity which will be crucial for the performance of ALMA in spectral scans over large bandwidths, needed for accurate astrochemical observations. Such upgrade will enable ALCHEMI-like observations towards a large sample of galactic nuclei of different luminosities and activities, and even more, will enable large scale mapping of the Galactic central molecular zones like the ACES project spanning over whole atmospheric windows or across the whole ALMA observable wavelengths.

Improvements to AI Systems:

1. Astrochemical Abundance Prediction Model

• Train a transformer-based model on the ALCHEMI spectral line data (84–374 GHz) and Galactic GMC surveys to predict molecular abundances (e.g., PN, SiO, HNCO, HC3N) from physical conditions (density, temperature, shock velocity, cosmic ray ionization rate).

• The improved system can, given a new galaxy’s integrated spectrum, estimate its star formation efficiency, shock fraction, and cosmic ray flux without requiring full radiative transfer modeling.

2. Multi-Scale Chemical Homogeneity Classifier

• Use the observed homogeneity across 1 pc to 0.5 kpc scales (NGC 253 vs. Galactic center) to build a neural network that classifies unresolved extragalactic spectra into “G+0.693-like” (shock-dominated, extended) vs. “hot-core-dominated” regimes.

• The system can flag whether a galaxy’s chemistry is driven by large-scale shocks or localized star-forming cores, aiding in rapid source classification for future ALMA surveys.

3. Principal Component Analysis (PCA) / NMF Enhancement for Spectral Cubes

• Implement a hybrid PCA–NMF algorithm (as in Harada et al. 2024 and Kishikawa et al. 2025) that automatically separates velocity components, shock tracers, and star formation tracers in 3D spectral line data cubes.

• The improved system can decompose a new galaxy’s CMZ into physically distinct regions (young vs. evolved starburst, low-velocity shocks) without manual line selection, and output maps of each component’s spatial extent.

4. Cosmic Ray Ionization Rate Estimator

• Train a regression model on the multiple molecular probes (C2H, HCO+/HOC+, H3O+/SO, HCN/HNC) and their modeled dependencies on ζ (cosmic ray ionization rate).

• The system can take observed line ratios from any galaxy and output a probability distribution of ζ, including uncertainties, directly comparable to the extreme rates (103–104× Milky Way) found in NGC 253.

5. Isotopic Ratio Calculator with Opacity Correction

• Develop an AI that automatically identifies optically thin isotopologues (e.g., 13C, 18O, 15N) from a given spectrum, applies opacity corrections using machine-learned escape probability models, and derives atomic isotopic ratios (e.g., 12C/13C, 14N/15N) across a galaxy.

• The system can provide accurate isotopic abundances for new ALMA data without manual line-by-line fitting, enabling rapid comparisons of nucleosynthetic history across galaxies.

6. Shock Evolution Simulator (SLED-based)

• Build a generative model that takes SiO and HNCO spectral line energy distributions (SLEDs) and predicts the shock age, temperature, and density evolution (single-event shock assumption).

• The improved system can, for any galaxy with these two tracers, reconstruct the shock timeline (e.g., fast shocks at 100s K vs. slow shocks at 100 K) and map the fraction of gas affected by each shock type.

7. Universal Dust Chemistry Transfer Learner

• Use the “universal” dust chemistry hypothesis (homogeneous abundances across scales) to create a transfer learning framework: train a model on Galactic GMC chemistry (e.g., G+0.693) and fine-tune it on NGC 253 data to predict molecular abundances in other starburst galaxies with limited spectral coverage.

• The system can extrapolate from a few observed lines to full chemical inventories for distant galaxies, reducing observational time requirements for future surveys.

8. Wideband Sensitivity Upgrade (WSU) Data Pipeline Optimizer

• Design an AI-based observation scheduler and data reduction optimizer that, given the ALMA WSU’s increased bandwidth, automatically selects frequency windows to maximize chemical information gain (e.g., detect PN, SiO, HNCO simultaneously) while minimizing noise for a target galaxy.

• The system can propose optimal spectral scan strategies for large samples of galactic nuclei, balancing resolution, sensitivity, and bandwidth to achieve ALCHEMI-like coverage in a fraction of the observing time.

9. Extended Emission Fraction Estimator

• Train a convolutional neural network on the HC3N and HNCO spatial maps (from ALCHEMI and Sgr B2) to predict the fraction of emission that is extended (non-core) vs. compact (hot cores).

• The system can, from a single galaxy’s integrated intensity map, quantify the dominance of shock-driven extended chemistry, which is key to understanding whether AGN or starburst activity dominates the molecular gas.

10. Cross-Scale Chemical Bridge Generator

• Create a generative adversarial network (GAN) that takes low-resolution extragalactic spectra (e.g., 255 pc or 350–500 pc) and produces high-resolution (1.6″) synthetic spectra, trained on the ALCHEMI multi-resolution data (7m array vs. full array vs. single dish).

• The improved system can enhance future observations of distant galaxies by predicting fine-scale chemical structure from coarse data, enabling one-to-one comparisons with Galactic clouds without needing full ALMA resolution.

For decades, chemical studies in external galaxies have been limited by sensitivity, resolution, bandwidth or a combination of them. A decade of ALMA operations, made the detailed study of the molecular abundances in nearby galaxies a routine job. We can finally bridge the ISM chemistry from Galactic to extragalactic environments, through one to one comparison, albeit probing significantly different scales. The ALMA large program ALCHEMI imaged the chemistry of the starburst galaxy NGC 253 at an unprecedented combination of sensitivity and resolution, achieving a detailed spatially resolved chemical inventory. ALCHEMI probed molecular differences among regions within its CMZ, and allowed the comparison with Galactic environments ranging from hot cores, quiescent molecular clouds and comets. This comparison can be put in the context of available data towards other extragalactic environments. Globally, the comparisons presented here show an astoundingly good correlation between the chemistry in Galactic GMCs and the central molecular zone of NGC 253, ranging scales from 1 pc up to 0.5 kpc. The wideband sensitivity upgrade of ALMA will expand similar comparisons towards a wider range of Galactic and extragalactic objects.