Li et al. present a box model with a new photochemical solver, ROMAC v1.0, that aims to be flexible and computationally efficient through an unique adaptive dynamic optimization module. It also improves over traditional box models with better quantification of physical effects. A new chemical solver and box model development is exciting news for the field and well fit for the scope of GMD. I'm happy to recommend this manuscript for publication in GMD.

Major comments:

1. An important development for a highly efficient chemical solver is to move beyond the box model and improve the efficiency of 3-D chemical transport models. Could the authors elaborate on ROMAC's extensibility to be used, eventually, as a chemical solver component in 3-D models?

2. In the intro around L51, authors discuss simplified chemical mechanisms. These approaches have a long history. The authors mainly talk about "fixed" reductions of the chemical mechanism, where a larger mechanism (e.g., MCM) is processed down to fewer species beforehand and used. But there are general methods for reduction (e.g., Young & Boris, 1977; Djouad & Sportisse, 2002) and their on-line implementations (e.g., Sander et al., 2019; Shen et al. 2022; Lin et al. 2023) that provide stability and efficiency at some cost of error, and would be useful to include in the introduction.

More generally, I think that saying simplified mechanisms lead to bias to simulation results is not a fully fair statement to make. There is no one true chemical mechanism that gives 100% accurate answers - so is the bias defined against the MCM, or against the observations? Reduced mechanisms generally have a focus on getting particular parts of chemistry that are of interest correct, and thus introduce some degree of bias in species that are not fully represented (and it is unavoidable to have some bias in fast-cycling radicals).

3. The software is currently limited-access to reviewers. Will the software be open in the future / upon publication to GMD? That is very important to the community and I believe in line with GMD policy.

Minor comments:

1. L71: The authors say that "Since the ROMAC model is computationally efficient, accurate and stable, users can dynamically optimize the influence of physical processes on pollutant concentration, and overcome the shortcomings of the lack of physical processes in the traditional box models." I'm not sure I follow here. What is the main difficulty in incorporating physical processes in other box models, is it a deliberate choice to focus on chemistry, or limited by efficiency, or (as the authors imply) affected by the stability of the solver? Incorporation of physical processes in a box model is a large part of the paper and I think the introduction would be better if more could be elaborated here, e.g., on why these processes weren't fully implemented and their effects.

2. L109-110: "...explicit methods ... are difficult to solve these problems." can be worded more clearly, "...explicit methods ... cannot achieve a stable solution without using a timestep shorter than all lifetimes in the system, which is computationally infeasible."

3. L130: The ROMAC model uses a Diagonal-Simplified-Newton (DSN) method which approximates the inverse of the Jacobian. Is there a quantified estimate of how much error this will introduce and effects on stability?

4. L255: Specify the OS version, compilers & versions used.

5. In Section 2.4 authors show the accuracy of the model as expressed in maximum relative error %. Does the % grow over time throughout the integration? It would be useful to show a time series plot.

6. In the abstract authors claim a 96% improvement in computational efficiency in ROMAC compared to "other box models". It may be useful to say which, and at what expense in error (which is small but worth mentioning).

Specific corrections:

1. L132 "specie" -> "species"

References:

Djouad, R., & Sportisse, B. (2002). Partitioning techniques and lumping computation for reducing chemical kinetics. APLA: An automatic partitioning and lumping algorithm. Applied Numerical Mathematics, 43(4), 383–398. https://doi.org/10.1016/S0168-9274(02)00111-3

Lin, H., Long, M. S., Sander, R., Sandu, A., Yantosca, R. M., Estrada, L. A., et al. (2023). An adaptive auto-reduction solver for speeding up integration of chemical kinetics in atmospheric chemistry models: Implementation and evaluation in the Kinetic Pre-Processor (KPP) version 3.0.0. Journal of Advances in Modeling Earth Systems, 15, e2022MS003293. https://doi.org/10.1029/2022MS003293

Sander, R., Baumgaertner, A., Cabrera-Perez, D., Frank, F., Gromov, S., Grooß, J.-U., et al. (2019). The community atmospheric chemistry box model CAABA/MECCA-4.0. Geoscientific Model Development, 12(4), 1365–1385. https://doi.org/10.5194/gmd-12-1365-2019

Shen, L., Jacob, D. J., Santillana, M., Bates, K., Zhuang, J., & Chen, W. (2022). A machine-learning-guided adaptive algorithm to reduce the computational cost of integrating kinetics in global atmospheric chemistry models: Application to GEOS-chem versions 12.0.0 and 12.9.1. Geoscientific Model Development, 15(4), 1677–1687. https://doi.org/10.5194/gmd-15-1677-2022

Young, T. R., & Boris, J. P. (1977). A numerical technique for solving stiff ordinary differential equations associated with the chemical kinetics of reactive-flow problems. Journal of Physical Chemistry, 81(25), 2424–2427. https://doi.org/10.1021/j100540a018