🔭 astrophysics

DiskMINT-GARDEN: Self-consistent Models to Estimate Disk Masses

This paper introduces DiskMINT-GARDEN, a self-consistent modeling grid and machine-learning tool that accurately estimates protoplanetary disk masses from ALMA observations and demonstrates that grain-surface chemistry, rather than large-scale elemental depletion, explains discrepancies with previous chemical models.

Original authors: Dingshan Deng, Uma Gorti, Ilaria Pascucci, Maxime Ruaud

Published 2026-06-26

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Dingshan Deng, Uma Gorti, Ilaria Pascucci, Maxime Ruaud

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 a protoplanetary disk as a giant, swirling nursery where planets are born. These nurseries are made mostly of invisible gas (like hydrogen) and a sprinkle of dust. To understand how big a family of planets might grow, astronomers need to know exactly how much "stuff" (mass) is in the nursery. But here's the problem: the gas is invisible to our eyes and most telescopes.

This paper introduces a new tool called DiskMINT-GARDEN. Think of it as a sophisticated "translation app" for astronomers. It helps them figure out how much gas is in a disk by looking at the things they can see.

Here is how it works, broken down into simple concepts:

1. The Problem: The "Invisible" Gas

Astronomers usually try to weigh these gas nurseries by looking at Carbon Monoxide (CO), a molecule that acts like a glowing signpost in the gas. However, for years, the CO signs looked much dimmer than expected.

The Old Theory: Scientists thought the gas must be missing. They believed the carbon and oxygen had been "stolen" or locked away into rocks (planetesimals) or turned into other chemicals, making the gas seem lighter than it really was.
The New Idea: This paper suggests the gas isn't missing; we just weren't looking at it correctly. The "dimness" might just be because the gas is cold, clumpy, or hiding in the middle of the disk, not because it's gone.

2. The Solution: A Massive Library of Simulations

The authors built a giant digital library called DiskMINT-GARDEN.

The Library: They ran thousands of computer simulations of different disks. They changed the size of the star, the amount of gas, the amount of dust, and the size of the disk to cover almost every possible scenario.
The Physics: Unlike older models that made guesses about how the disk was shaped, this tool solves the physics "all at once." It calculates how gravity pulls the gas down, how heat rises, and how dust settles, ensuring the simulation is physically realistic.
The Chemistry: It also tracks what happens to the CO molecules. It knows that in the cold, deep parts of the disk, CO freezes onto dust grains and can turn into CO2 (like dry ice). This process changes how the CO glows, and the model accounts for this naturally.

3. The "Magic" Translator: Machine Learning

Running a complex physics simulation for every single new star takes too long. So, the authors trained a machine learning robot (a type of AI) on their library of 480 simulations.

How it works: You give the robot three pieces of data that astronomers can easily measure with the ALMA telescope:
1. How bright the dust is (the "glow" of the nursery).
2. How bright the CO gas is.
3. How big the disk looks.
The Result: The robot instantly looks up its library and says, "Based on these three numbers, this disk has this much gas, this much dust, and is this big." It does this in seconds, whereas the old way took days or weeks.

4. Testing the Tool

The team tested their new tool on 34 real disks that astronomers had already observed.

The Check: They compared their results with two other ways of weighing disks:
1. Dynamical Weighing: Watching how the gas moves (like watching water swirl in a drain to guess its weight). This only works for very heavy disks.
2. HD Tracing: Looking for a rare molecule called Hydrogen Deuteride (HD), which is a direct gas tracer but is very hard to see.
The Verdict: DiskMINT-GARDEN's estimates matched the other methods very well (within a factor of two). This proves the tool is reliable.

5. The Big Discovery: The Gas Wasn't Missing

The most exciting finding is about the "missing" gas.

When they compared their results to older models (like one called DALI), they found that the older models needed to assume that huge amounts of gas were missing (depleted) to make the numbers work.
DiskMINT-GARDEN showed that you don't need to assume the gas is missing. If you properly account for the fact that CO freezes onto dust and turns into CO2, the numbers work perfectly without "stealing" any gas.
Conclusion: This suggests that the chemistry in these disks hasn't changed as much as we thought. The carbon and oxygen are still there, just hiding in the cold, deep layers or converted into CO2 ice, not locked away in giant rocks.

Summary

DiskMINT-GARDEN is a fast, open-source tool that lets astronomers quickly and accurately weigh the gas in planet-forming disks. It uses a massive library of realistic physics simulations and a smart AI to translate simple telescope observations into a precise mass. Most importantly, it tells us that the gas in these cosmic nurseries is likely still there, just harder to see than we thought, and we don't need to invent "missing gas" to explain what we observe.

Technical Summary: DiskMINT-GARDEN: Self-consistent Models to Estimate Disk Masses

Problem Statement
Determining the gas mass of protoplanetary disks is fundamental to understanding planet formation, yet it remains challenging. Carbon monoxide (CO) is the primary tracer for disk gas, but early thermochemical models frequently found that observed CO isotopologue emission was significantly fainter than predicted. This discrepancy was often interpreted as evidence for substantial depletion of gas-phase CO (by up to two orders of magnitude) due to sequestration into planetesimals or chemical processing. However, such depletion factors introduce significant uncertainty into mass estimates ( $M_{\text{gas}}$ ). Furthermore, existing models often rely on simplified assumptions, such as vertically isothermal density structures or fixed depletion factors, which may not self-consistently account for the coupling between disk structure, temperature, and chemistry. While independent constraints exist via dynamical measurements (limited to massive disks) or Hydrogen Deuteride (HD) observations (limited by sensitivity and facility availability), a rapid, reproducible, and self-consistent method to estimate disk masses from commonly available ALMA data is needed.

Methodology
The authors present DiskMINT-GARDEN (Grid of Astrochemical Radiative Disk EmissioN), a framework combining a comprehensive grid of self-consistent models with a machine-learning inference tool.

Model Grid Construction: The grid is built upon the DiskMINT framework, which couples hydrostatic disk structure, continuum/line radiative transfer, and a reduced chemical network. Key physical components include:
- Self-consistent Vertical Structure: The model solves for vertical hydrostatic equilibrium, linking the midplane mass to the emitting surface layers, rather than assuming a fixed isothermal density profile.
- Chemistry: The network includes CO freeze-out, grain-surface conversion (specifically CO to CO $_2$ ), and isotope-selective photodissociation. Notably, it does not assume ad hoc elemental depletion factors.
- Parameter Space: The grid spans stellar masses ( $0.1 - 2.0 M_\odot$ ), gas disk masses ( $10^{-5} - 10^{-1} M_\star$ ), dust-to-gas ratios ($0.003 - 0.1$), and characteristic radii ($10 - 300$ au).
- Synthetic Observables: For each model, the framework computes millimeter continuum luminosity ( $L_{\text{mm}}$ at $\sim 1.3$ mm), C $^{18}$ O (2–1) and (3–2) line luminosities, and dust size metrics ( $R_{\text{dust,90}}$ ).
Inference Tool: A supervised machine-learning regression model based on gradient-boosted decision trees (XGBoost) is trained on the model grid. This tool maps the observable vector $\mathcal{O} = (L_{\text{C}^{18}\text{O}}, L_{\text{mm}}, R_{90,\text{dust}})$ to the physical parameter vector $\Theta = (M_{\text{gas}}, \epsilon, R_c)$ for a given stellar mass. The tool is designed for fast, deterministic predictions within the grid domain.
Validation and Uncertainty: Statistical uncertainties are estimated by propagating observational errors and interpolation scatter (typically $\lesssim 25\%$ ). Systematic uncertainties arising from underlying model assumptions (structure, chemistry) are conservatively estimated as a factor of $\sim 2$ . The framework includes domain checks to flag targets lying outside the training grid boundaries.

Key Results
The authors applied DiskMINT-GARDEN to an archival sample of 34 protoplanetary disks with ALMA continuum and C $^{18}$ O detections.

Agreement with Independent Constraints: The inferred gas masses show good agreement (within a factor of $\lesssim 2$ ) with independent dynamical mass estimates for massive disks and with HD-based mass estimates for the few disks where HD is detected. This validates the framework's ability to reproduce masses in regimes where dynamical constraints are robust.
Comparison with DALI Models: When compared to masses derived using the DALI thermochemical framework (which typically requires large-scale CO depletion factors to match N $_2$ $_{2}$ H $^+$ $^{+}$ and CO observations), DiskMINT-GARDEN yields similar total gas masses without invoking any depletion factors.
- In DALI, the lack of grain-surface chemistry and a fixed isothermal density structure necessitates a depletion factor (often $\sim 10\times$ ) to match observations.
- In DiskMINT, the inclusion of grain-surface chemistry (converting CO ice to CO $_2$ ) and a self-consistent vertical structure (placing the CO snowline at higher temperatures, $\sim 35$ K) naturally reproduces the observed C $^{18}$ O emission and N $_2$ H $^+$ constraints without ad hoc depletion.
Implications for Disk Evolution: The fact that DiskMINT-GARDEN recovers masses consistent with DALI without depletion suggests that current data do not require significant chemical processing (such as CO sequestration into planetesimals) to explain the observed emission. The "missing" CO in previous interpretations may be accounted for by the physical structure and grain-surface chemistry implemented in DiskMINT.

Significance and Claims
The paper claims that DiskMINT-GARDEN provides a rapid, open-source, and homogeneous method for estimating disk gas masses from widely available ALMA data (continuum and C $^{18}$ O), bypassing the need for time-intensive observations of additional tracers like N $_2$ H $^+$ .

Efficiency: It enables population-level studies of disk evolution across large samples, extending mass estimates to lower-mass disks where dynamical constraints are infeasible.
Physical Consistency: By demonstrating that self-consistent models with grain-surface chemistry can reproduce observations without large depletion factors, the work challenges the necessity of assuming significant CO depletion in disk evolutionary models.
Scalability: The framework serves as a first-order inference tool that can guide detailed, spatially resolved modeling of individual targets. The authors note that while the current grid assumes fixed dust properties, the framework can be extended to incorporate additional tracers (e.g., HD from future facilities) and multi-band continuum data to further constrain disk parameters.

The authors conclude that the extant data are consistent with little chemical processing due to disk evolutionary processes, and that the discrepancies in previous mass estimates often stem from differences in how disk structure and chemistry are coupled rather than from actual elemental depletion.