UCLCHEM: Gas–Grain Astrochemical Code

• UCLCHEM is a time-dependent gas–grain astrochemical code that computes molecular abundances via coupled rate equations from gas-phase and grain-surface processes.
• It employs modular physical workflows for diverse interstellar conditions, from static clouds and collapsing cores to shock fronts and hot-core warm-ups.
• Integrated with radiative transfer and machine learning tools, UCLCHEM enables efficient inference of physical conditions and predictive molecular diagnostics.

Searching arXiv for UCLCHEM and recent related papers. UCLCHEM is a publicly available, open-source, time-dependent gas–grain astrochemical code designed to propagate molecular abundances through prescribed physical histories in interstellar environments where both gas-phase reactions and dust-grain processes are important. In the literature surveyed here, it is described as a gas–grain chemical model written in modern Fortran, extended through modular “physics modules” for static clouds, collapsing cores, hot-core warm-up, C-type shocks, and related scenarios, and later used as the chemical backbone for inference frameworks, neural-network emulators, and machine-learning analyses of molecular diagnostics (Holdship et al., 2017). Across these applications, UCLCHEM functions as a rate-equation solver for coupled chemistry in gas, grain surfaces, and—in three-phase implementations—grain bulk mantles, with outputs typically expressed as fractional abundances relative to hydrogen nuclei and then coupled to radiative-transfer tools or statistical surrogates (O'Donoghue et al., 2022).

1. Definition, architecture, and scope

UCLCHEM is consistently characterized as a time-dependent gas–grain chemical code that computes the evolution of interstellar abundances under user-specified physical conditions rather than solving full MHD or radiation-hydrodynamics. The public code description emphasizes gas-phase reactions, freeze-out, non-thermal and thermal desorption, simple or diffusion-based grain chemistry, and modular physical setups for cold clouds, collapsing cores, hot cores, and C-type shocks (Holdship et al., 2017). Later work further specifies that the code is openly available, written in modern Fortran, and organized around separate chemistry and physics modules, so that a chosen physical history can be coupled to a common chemical solver (O'Donoghue et al., 2022).

In its most general use, UCLCHEM evolves species abundances $n_i$ through coupled ordinary differential equations of the form

$$\frac{dn_i}{dt} = \sum R_{\mathrm{form},i} - \sum R_{\mathrm{dest},i},$$

with reaction rates drawn from gas-phase and gas–grain networks and evaluated under the local density, temperature, radiation field, and ionization environment (O'Donoghue et al., 2022). In some applications, the quantity advanced is the fractional abundance $X(i)=n(i)/n_{\mathrm H}$, explicitly defined relative to hydrogen nuclei (Holdship et al., 2017). This common structure allows the same chemical engine to be reused across dark-cloud, shock, disk, and hot-corino calculations.

The code’s scope is unusually broad within gas–grain astrochemistry. The original release already framed UCLCHEM as suitable for cold molecular clouds and dense cores, protostellar envelopes and hot cores, C-type shocks, and post-processing of hydrodynamic simulations (Holdship et al., 2017). Subsequent studies use it for pre-stellar cores with column-dependent cosmic-ray treatments, outer-Galaxy low-metallicity dark clouds, circumstellar disks, hot-corino COM chemistry, deuterated isotopologue networks, and galaxy-scale molecular-line inference in NGC 1068 and NGC 253 (O'Donoghue et al., 2022). A plausible implication is that UCLCHEM is best understood not as a single environment-specific model, but as a reusable chemical framework whose physical interpretation depends on the selected module and network.

2. Chemical formalism and gas–grain processes

The original code paper presents UCLCHEM as a large reaction-network solver in which two-body gas-phase rates follow the UMIST Arrhenius-type form

$$R_{AB} = \alpha \left( \frac{T}{300\,\mathrm{K}} \right)^{\beta} \exp\left( -\frac{\gamma}{T} \right),$$

with corresponding source and sink terms for each product and reactant (Holdship et al., 2017). Later work on dense-core cosmic rays retains the same reaction-rate logic and explicitly states that UCLCHEM’s default gas-phase network is UMIST RATE12, while grain chemistry is handled by a separate user-defined network including freeze-out, diffusion, surface reactions, and desorption (O'Donoghue et al., 2022). In COM-focused studies, the network is often augmented manually, for example with methyl formate formation and deuteration routes or with alternative methanol-formation channels (Awad et al., 2021).

Freeze-out and desorption are central to the code’s identity. In the original formulation, all gas species can adsorb onto grains, with a freeze-out rate depending on grain size, species mass, temperature, and a freeze-out efficiency; positive ions receive an additional electrostatic enhancement factor (Holdship et al., 2017). Non-thermal desorption mechanisms include cosmic rays, UV photons, and H$_2$-formation-induced desorption, while thermal desorption uses a multi-stage treatment based on laboratory temperature-programmed desorption data, with species grouped into CO-like, H$_2$O-like, and intermediate categories (Holdship et al., 2017). Later benchmarking in the JWST-ice context shows that UCLCHEM is a three-phase model with gas, grain surface, and grain bulk, and that it can include thermal desorption, cosmic-ray heating, UV photodesorption, cosmic-ray-induced UV photodesorption, chemical reactive desorption, and H$_2$-formation-induced desorption (Jiménez-Serra et al., 14 Feb 2025).

Grain chemistry can be treated at multiple levels of sophistication. The original release includes simple hydrogenation-at-freeze-out schemes and optional diffusion-based grain reactions (Holdship et al., 2017). By contrast, some later COM studies deliberately disable UCLCHEM’s temperature-dependent diffusion treatment and replace it with fixed surface reaction rates, showing that the framework can be simplified when diffusion barriers are not well constrained (Awad et al., 2021). This suggests that UCLCHEM’s grain chemistry is not a single immutable model; it is a configurable rate-equation infrastructure whose realism depends on the chosen network, barriers, and desorption prescriptions.

3. Physical modules and standard workflows

A recurring pattern in UCLCHEM applications is the use of sequential phases. In the canonical two-phase workflow, Phase I follows the formation of a dense core from diffuse atomic gas, often through free-fall collapse, while Phase II applies a second physical history such as static dense-core evolution, protostellar warm-up, or passage of a shock (Holdship et al., 2017). This design is explicit in the HCN/HNC shock study of L1157-B1, where UCLCHEM is described as a “time-dependent gas–grain chemical and parametrised shock model” run in two phases: dense-core formation from an atomic medium, then a steady-state C-type shock through that medium (Lefloch et al., 2021). The same logic appears in hot-corino methyl-formate work, where Phase I is a 10 K collapse and Phase II is a 1 $M_\odot$ warm-up described by

$T_{\mathrm{dust}}(t)=10.0 + (A\times t^B),$

following the usual low-mass hot-corino prescription (Awad et al., 2021).

Collapse models are one major branch of this workflow. The original paper includes a Bonnor–Ebert-sphere application to starless cores, with Phase I free-fall to $2\times10^4\,\mathrm{cm^{-3}}$ and Phase II Bonnor–Ebert collapse toward higher density, producing radial abundance profiles for comparison with observed CO, CS, NH$$\frac{dn_i}{dt} = \sum R_{\mathrm{form},i} - \sum R_{\mathrm{dest},i},$$0, and N$$\frac{dn_i}{dt} = \sum R_{\mathrm{form},i} - \sum R_{\mathrm{dest},i},$$1H$$\frac{dn_i}{dt} = \sum R_{\mathrm{form},i} - \sum R_{\mathrm{dest},i},$$2 (Holdship et al., 2017). Dense-core and pre-stellar-core studies similarly begin from diffuse gas, typically $$\frac{dn_i}{dt} = \sum R_{\mathrm{form},i} - \sum R_{\mathrm{dest},i},$$3, and evolve through free-fall collapse to $$\frac{dn_i}{dt} = \sum R_{\mathrm{form},i} - \sum R_{\mathrm{dest},i},$$4–$$\frac{dn_i}{dt} = \sum R_{\mathrm{form},i} - \sum R_{\mathrm{dest},i},$$5, sometimes adding column-density-dependent cosmic-ray ionization and H$$\frac{dn_i}{dt} = \sum R_{\mathrm{form},i} - \sum R_{\mathrm{dest},i},$$6 dissociation during the collapse (O'Donoghue et al., 2022).

Shock modules form another major branch. UCLCHEM’s C-type shock implementation is parameterized rather than fully MHD-coupled; it supplies temperature, density, and ion–neutral drift profiles through the shock and then integrates the chemistry along that flow (Holdship et al., 2017). The L1157-B1 HCN/HNC study uses this machinery to compare $$\frac{dn_i}{dt} = \sum R_{\mathrm{form},i} - \sum R_{\mathrm{dest},i},$$7 and $$\frac{dn_i}{dt} = \sum R_{\mathrm{form},i} - \sum R_{\mathrm{dest},i},$$8 shocks and to identify sputtering plus high-temperature HNC destruction as the origin of the observed rise in the HCN/HNC ratio (Lefloch et al., 2021). A later study extends UCLCHEM with a planar, steady-state J-type shock module parameterized from mhd_vode output, adding explicit J-shock front and post-shock relaxation profiles and enabling direct C-vs-J chemical comparisons in L1157 (James et al., 2019).

Other workflows adapt the same framework to static dark clouds or disks. The outer-Milky-Way metallicity study restricts UCLCHEM to single-zone, isothermal, constant-density dark-cloud models evolved to $$\frac{dn_i}{dt} = \sum R_{\mathrm{form},i} - \sum R_{\mathrm{dest},i},$$9 yr and sampled at $X(i)=n(i)/n_{\mathrm H}$0 yr (Vermariën et al., 13 May 2025). The circumstellar-disk study treats each $X(i)=n(i)/n_{\mathrm H}$1 zone independently and feeds UCLCHEM a spatially varying ionization field from stellar cosmic rays, X-rays, FUV, and radionuclides, thereby turning the code into a disk thermochemical post-processor without modifying its fundamental rate-equation character (Offner et al., 2019).

4. Scientific applications across environments

UCLCHEM has been used repeatedly to connect molecular abundances to physical conditions in shocks. In L1157-B1, it reproduces pre-shock HCN and HNC abundances of order $X(i)=n(i)/n_{\mathrm H}$2, a pre-shock HCN/HNC ratio near unity, and the post-shock rise to HCN/HNC $X(i)=n(i)/n_{\mathrm H}$3 only for a C-type shock with pre-shock density $X(i)=n(i)/n_{\mathrm H}$4, $X(i)=n(i)/n_{\mathrm H}$5, a short pre-shock phase, and a shock age around $X(i)=n(i)/n_{\mathrm H}$6–$X(i)=n(i)/n_{\mathrm H}$7 yr (Lefloch et al., 2021). In the J-vs-C diagnostic study, UCLCHEM finds that molecules such as H$X(i)=n(i)/n_{\mathrm H}$8O and HCN show shock-type-specific behaviour only at low $X(i)=n(i)/n_{\mathrm H}$9 and low $$R_{AB} = \alpha \left( \frac{T}{300\,\mathrm{K}} \right)^{\beta} \exp\left( -\frac{\gamma}{T} \right),$$0, while CH$$R_{AB} = \alpha \left( \frac{T}{300\,\mathrm{K}} \right)^{\beta} \exp\left( -\frac{\gamma}{T} \right),$$1OH is enhanced in both shock types and is therefore a probe of pre-shock density rather than shock type (James et al., 2019).

It is also used to test grain-surface formation pathways for complex organic molecules. In hot-corino models of glycolaldehyde and ethylene glycol, UCLCHEM is run through collapse and warm-up phases to compare different grain routes, finding that no tested scenario perfectly reproduces the observed luminosity trend in the ethylene glycol/glycolaldehyde ratio, though better agreement is found for HCO + HCO recombination followed by hydrogenation, potentially with an added HCO + CH$$R_{AB} = \alpha \left( \frac{T}{300\,\mathrm{K}} \right)^{\beta} \exp\left( -\frac{\gamma}{T} \right),$$2OH contribution (Coutens et al., 2017). In deuterated methyl formate, UCLCHEM indicates that grain radical–radical association is required to reproduce DCOOCH$$R_{AB} = \alpha \left( \frac{T}{300\,\mathrm{K}} \right)^{\beta} \exp\left( -\frac{\gamma}{T} \right),$$3, and that H–D substitution on grains significantly boosts HCOOCHD$$R_{AB} = \alpha \left( \frac{T}{300\,\mathrm{K}} \right)^{\beta} \exp\left( -\frac{\gamma}{T} \right),$$4, DCOOCHD$$R_{AB} = \alpha \left( \frac{T}{300\,\mathrm{K}} \right)^{\beta} \exp\left( -\frac{\gamma}{T} \right),$$5, and HCOOCD$$R_{AB} = \alpha \left( \frac{T}{300\,\mathrm{K}} \right)^{\beta} \exp\left( -\frac{\gamma}{T} \right),$$6 (Awad et al., 2021).

Methanol chemistry provides a further example of the code’s flexibility. A recent study modifies the UCLCHEM grain network to include the radical–molecule abstraction reaction

$$R_{AB} = \alpha \left( \frac{T}{300\,\mathrm{K}} \right)^{\beta} \exp\left( -\frac{\gamma}{T} \right),$$7

and compares “standard” and “current” networks in non-shocked and C-shocked environments (Huang et al., 26 Sep 2025). Within that framework, UCLCHEM shows that this abstraction route is the primary reaction leading to CH$$R_{AB} = \alpha \left( \frac{T}{300\,\mathrm{K}} \right)^{\beta} \exp\left( -\frac{\gamma}{T} \right),$$8OH in the inner layers of interstellar ices under both non-shocked and shocked conditions, and that gas-phase H$$R_{AB} = \alpha \left( \frac{T}{300\,\mathrm{K}} \right)^{\beta} \exp\left( -\frac{\gamma}{T} \right),$$9CO is much more sensitive than CH$_2$0OH to whether the abstraction route is included, motivating the use of the H$_2$1CO/CH$_2$2OH ratio as a diagnostic (Huang et al., 26 Sep 2025).

Dark-cloud and dense-core applications emphasize environmental dependence rather than shock chemistry. The outer-Galaxy study uses a Sobol-sampled UCLCHEM grid over density, temperature, UV field, cosmic-ray ionization rate, and independent carbon and oxygen depletions, then analyzes nine molecular ratios and finds that temperature and density are globally the dominant controls, while CN/HCN and HNC/HCN are particularly sensitive to initial carbon abundance and CS/SO is the only ratio in the set with strong oxygen sensitivity (Vermariën et al., 13 May 2025). The dense-core cosmic-ray study updates UCLCHEM with column-density-dependent $_2$3, column-density-dependent H$_2$4 dissociation, and cosmic-ray-induced excited grain species, showing that the ionization-rate dependence is the dominant effect and can alter predicted abundances by several orders of magnitude (O'Donoghue et al., 2022).

5. Coupling to radiative transfer, inference, and machine learning

A major line of development has turned UCLCHEM from a stand-alone chemical solver into the chemical component of larger inference systems. One approach couples UCLCHEM abundances to RADEX through an on-the-spot conversion from abundance to column density, $_2$5, thereby replacing free molecular columns in line modeling with chemically self-consistent ones (Mijolla et al., 2019). This framework is then embedded in Bayesian sampling, allowing physical parameters such as $_2$6, $_2$7, $_2$8, $_2$9, UV field, and metallicity to be inferred from line intensities rather than treating column densities as independent nuisance parameters (Mijolla et al., 2019).

UCLCHEMCMC generalizes this logic into an MCMC tool that links UCLCHEM and RADEX inside a Bayesian forward model and caches chemical and radiative-transfer evaluations in an SQL database (Keil et al., 2022). The paper reports that, using UCLCHEM and RADEX, the runtime for ten walkers taking one thousand steps drops from $_2$0 s with an empty database to $_2$1 s when nearly all requested models are already stored, an efficiency increase of nearly two orders of magnitude (Keil et al., 2022). The same work also shows, through L1544 case studies, that line selection matters because different species may probe different substructures (Keil et al., 2022).

A second development is direct emulation. A neural-network emulator of UCLCHEM was introduced for chemistry-dependent line modeling, trained on $_2$2 UCLCHEM models to predict equilibrium abundances from $_2$3, with 95% of predictions within 0.05 dex of the underlying UCLCHEM abundance (Mijolla et al., 2019). The broader thermochemical emulator “Chemulator” goes further by modifying UCLCHEM to include heating and cooling, then training a neural network on UCLCHEM outputs; Chemulator reproduces the emulated model with an overall mean squared error of 0.0002 for a single 1000 yr step and is reported to be approximately 50,000 times faster than the time-dependent model it emulates (Holdship et al., 2021).

Recent extragalactic studies use the same emulation strategy operationally. In NGC 253, a neural network trained on UCLCHEM output for HCN and HNC recovers UCLCHEM abundances with about 3 percent error and increases inference speed tenfold, enabling 50 pc-resolution Bayesian maps of CRIR, density, temperature, column density, and beam-filling factor across the CMZ (Behrens et al., 2024). In NGC 1068, a deep emulator trained on $_2$4 UCLCHEM models with quiescent and heating thermal histories reproduces HNC abundances over roughly eight orders of magnitude with RMSE $_2$5 dex, MAE $_2$6 dex, mean relative error 0.35%, and $_2$7, and is then coupled to SpectralRadex inside hierarchical Bayesian inference (Jia et al., 30 Jun 2026).

Machine learning has also been layered on top of large UCLCHEM grids for interpretability rather than direct replacement. In the outer-Milky-Way ratio study, UCLCHEM provides a Sobol-sampled grid of $_2$8 dark-cloud models, after which XGBoost regressors, SHAP values, and UMAP embeddings are used to identify which physical parameters control each line ratio across the parameter space (Vermariën et al., 13 May 2025). A plausible implication is that UCLCHEM has become not only a chemistry code but also a generator of structured synthetic data for statistical learning.

6. Limitations, assumptions, and methodological debates

The literature is explicit that UCLCHEM is powerful but not fully self-consistent physically. Its shock implementations are parameterized rather than simultaneous MHD-plus-chemistry solutions, and the original C-shock module omits grain–grain shattering and vaporization (Holdship et al., 2017). The J-shock extension is likewise planar and steady-state and neglects radiative precursors for the low-velocity regime considered (James et al., 2019). In disk applications, UCLCHEM does not solve the thermal balance; the temperature is prescribed externally, each zone is treated independently, and radiation or cosmic rays entering horizontally through the disk are neglected (Offner et al., 2019).

Radiative transfer is generally outside the code. The HCN/HNC study of L1157-B1 uses LTE analysis in CASSIS alongside UCLCHEM because the chemistry code itself does not convert abundances into line intensities (Lefloch et al., 2021). Extragalactic inference papers therefore couple UCLCHEM to RADEX or SpectralRadex and treat line excitation separately (Mijolla et al., 2019). This division is methodologically important: UCLCHEM supplies chemically plausible abundances, but interpretation of observed spectra still depends on excitation, optical depth, and geometry handled elsewhere.

Grain chemistry is another debated area. The JWST-ice comparison shows that UCLCHEM is a diffusive-only grain-surface code and therefore behaves differently from models that include non-diffusive chemistry, especially below $_2$9 K (Jiménez-Serra et al., 14 Feb 2025). In that benchmark, UCLCHEM predicts more CH$_2$0 than the other compared codes because stronger non-thermal desorption in the translucent phase leaves more atomic C available for freeze-out and hydrogenation later, but it also tends to underproduce H$_2$1O and overproduce CO$_2$2 at $_2$3–$_2$4 K (Jiménez-Serra et al., 14 Feb 2025). This suggests that UCLCHEM’s results can be highly sensitive to desorption yields, diffusion assumptions, and the treatment of low-temperature grain chemistry.

Finally, several papers emphasize network incompleteness or uncertain rates. The glycolaldehyde/ethylene-glycol work stresses that gas-phase destruction routes for these COMs are very poorly known and that only one constrained destruction reaction for glycolaldehyde is included, limiting predictive reliability after desorption (Coutens et al., 2017). The methanol-abstraction study likewise notes missing or uncertain surface reactions and differences between UCLCHEM’s surface/bulk treatment and laboratory or Monte Carlo approaches (Huang et al., 26 Sep 2025). The consistent message is not that UCLCHEM is unreliable, but that it is best used as a configurable research framework whose conclusions are contingent on the adopted network, physical module, and comparison methodology.