The authors present a comprehensive and methodologically robust study assessing the global contribution of RO₂ + RO₂ accretion reactions to secondary organic aerosol (SOA) formation using the GEOS-Chem chemical transport model. By integrating updated theoretical understanding and incorporating new accretion mechanisms via GECKO-AP, the work offers a timely and valuable addition to the literature on aerosol formation, especially in tropical and low- NOx environments. The manuscript is technically rigorous, and the authors demonstrate clear model measurement comparisons using data from SOAS and GOAMAZON. It is suitable for publication in ACP. However, I have several suggestions to improve the clarity, balance, and robustness of the conclusions. 

Major Comments:



- Without directly flagging modeling assumptions the abstract and conclusions suggest that RO₂ accretion products make up 30–50% of PM₂.₅ in some regions. While the model results support this, the claim should be qualified with reference to underlying assumptions (e.g., product yield estimates, volatility assumptions, possible double counting). Use phrases like “may contribute” or “model results suggest up to...” to moderate certainty. 



-While the text does mention uncertainties (e.g., calibration, partitioning), these could be more systematically discussed. I suggest including a short paragraph in the discussion or conclusion explicitly listing key uncertainties, such as lack of isomer-resolved detection and structural diversity limitations in GECKO-AP. 



-There is some ambiguity about whether accretion product mass is being added to or replacing existing parameterized SOA mass in the model. Summarize the key point earlier: that empirically derived SOA yields may already include some fraction of accretion product mass. Provide a clearer summary of how this potential overlap was handled and consider including a schematic or table (perhaps in the SI) for clarity.



-GECKO-AP only considers peroxide formation and alkoxy decomposition channels. This is a major limitation that undermines structural diversity of products (e.g., imines, carbonates, or hydroperoxy derivatives). You should explicitly discuss what classes of real-world accretion products are likely being omitted. Quantify how sensitive your PM₂.₅ results might be to that structural simplification. Acknowledge this limitation and briefly discuss how it may impact modeled volatility and SOA mass.  



-The manuscript lacks a sensitivity test where accretion products are assumed to have lower yields (i.e., uncertainty in GECKO-AP branching), evaporate faster (i.e., higher volatility), decompose photochemically. Include at least one sensitivity simulation testing either a lower dimer yield (e.g., 50% reduction), or increased loss rate (photolysis / fragmentation surrogate), and assess how much PM₂.₅ mass this removes globally. This will add robustness and credibility to the 30–50% claim.



-“Mean model/measurement ratio was 4.6...” This is substantial overprediction. The explanation (sensitivity-based calibration and fragmentation losses) is valid but not quantified. Provide a range of plausible “true” concentrations using a spectrum of calibration sensitivities (e.g., ± order of magnitude). Consider reporting normalized root mean square error (NRMSE) or similar metrics. 



-The OA radiative effect changes are described, but without clear error bars or sensitivity runs to support confidence in the conclusion. Add uncertainty estimates (e.g., based on ±25% OA mass) to the TOA forcing calculations. Even just a bounding box would help. 



- Recent literature has shown that peroxy radicals can react on aqueous or organic surfaces (e.g., aerosol interfaces or freshly nucleated particles). These reactions could either enhance or compete with gas-phase dimer formation. Please add a brief discussion addressing surface-phase RO₂ chemistry as a competing or complementary pathway. Additionally, could you comment on how including this mechanism might affect your global estimates?

Minor Comments 

- The title could include “SOA” or “secondary organic aerosol” for better visibility.



- In the Abstract, consider briefly mentioning the distinction between peroxide and non-peroxide products.



- Clarify what is meant by “DILVOC” in the radiative effect section and its role in the simulations. Its role in radiative effect calculations is a bit unclear, reiterate its inclusion/exclusion in relevant figures/tables



- Some figures (e.g., model-measurement comparison plots) would benefit from more detailed legends or axis labels to improve standalone readability without flipping back to text.



-Consistently use “RO₂ accretion products” or “RO₂ dimers” throughout for clarity. Sometimes the manuscript says “accretion products,” sometimes “RO₂ dimers,” sometimes “non-peroxide dimers.” Clarify earlythat you're referring to peroxide, ester, and ether dimers as the dominant species, and consistently use a single term throughout (e.g., “RO₂ accretion products”). 

- Must quantify or bound NO biases to ensure accurate RO₂ fate modeling.



- Need to address potential double-counting of OA when adding new chemistry to existing parameterizations.



- More clarification is needed on how RO₂ categories like “Small VOCs” and “Mixed VOCs” are chemically defined.



- The assumption that nearly 100% of OA is from accretion in some regions is likely an overestimate, needs better constraint or alternative explanations.



-You should acknowledge how autooxidation might shift results.

This article reports the development of chemical models and their implementation into the global EOS-chem model to estimate the importance of RO2 recombination reactions in the formation of low-vapor pressure compounds in the gas and their contribution to SOA globally. It is an interesting study but, in my opinion probably a bit premature because of the scarcity of the experimental data characterizing the reaction pathways of interest here. But I do understand the interest of putting some figures on the global implications of these reactions, to decide (or justify) to invest more time and resources in this topic.

The results predicting such a global importance for RO2 recombination reactions, even with non-negligible NO concentrations (for instance in the validation in Fig.2) is somewhat puzzling. It would be very interesting to know the rate coefficients used for these reactions in the models. Unfortunately, I could not find this information in the manuscript. This and a few other points regarding the chemical models have prevented me from fully understanding the model results and their implications. These points are listed below, I hope that the authors can clarify them in order to go forward with this paper.

Detailed comments

1. Perhaps the first confusing point is the use of the term “accretion” referring to a range of compounds without a clear definition. According to the IUPAC gold book this term refers to all the compounds contributing to aerosol growth. If this is so, and therefore this term includes all the products discussed in this paper, this should be stated in the introduction. Otherwise, since the discussion in this paper refers to specific mechanisms, it helps to be as specific as possible when referring to the products (covalent dimers, ester, ethers...) and/or mechanism.
2. The text mentions that the full modified mechanisms are available in the Supplementary Material, but I can not find this anywhere. The SI only includes a list of non-accretion reactions. Did I miss something ?
3. I am not sure to understand the statement li. 94-97 p.4 “However, its prediction of the product distribution is highly uncertain ... For this reason GECKO-AP only considers peroxide formation and alkoxy decomposition channels”... Do the authors mean that the channels in Fig. 1 are the ONLY ones of RO2+RO2 (or RO2 + R’O2) taken into account in GECKO-AP ? Or that these are the only added channels because the others are already in GECKO-A ? Since I could not find the full mechanisms, I could not check which one it is.

The first case would obviously lead to large overestimations of the recombination products. This would also explain the “excessive and unrealistic build-up of accretion products” by the model mentioned several times in the paper. If a model tends to grossly overestimate some products, it probably means that it is not realistic. Could the authors comment on why they expect the model to overestimate these products unrealistically ? 

In any case, if a choice has been made to determine upper limits for these recombination products, this should be clearly stated and remined when discussing the model results.

1. Figure 1 is a bit confusing for readers who are not expert in RO2 chemistry. Perhaps the “classical” channels of RO2+RO2 could be reminded (even very briefly) to clarify that only the “third” or other channels are detailed here. Also, numbering some of the pathways (for instance “1” for dimerization, “2” for ether/ester...) for would simplify (and clarify) some discussions in the text.
2. In the second part of the work, the study of the distribution between peroxides and ester/ether products, a list of the VOC or RO2 involved is missing. Was it the VOC mix for the SOAS campaign ? According to the Peräkylä et al., 2023 paper the formation of esters/ethers is specific to RO2 with a carbonyl group in β of the RO2 group. Beside some terpenes, this is not a general feature in RO2, thus the results in Fig. 10 would be highly dependent on the VOC mixture chosen.
3. Although the above questions prevented me from fully appreciating the modelling results, I have a few questions on these results:

- In the validation presented in Fig. 2, the “dip” in the products between 5 and 10 am (especially in the gas-phase products) is clearly correlated with the NO peak presented in Fig. S1 and underestimated by GEOS-Chem. But, at all the other times, measured and predicted NO agree well. Yet, the GEOS-Chem simulation overestimates the products over all these times (for instance around 15:00 for the gas-phase ones). Thus, shouldn’t there be other explanations for this overestimation of the products by GEOS-chem than the discrepancies between the measured and predicted NO ?

As mentioned above, the substantial amount of recombination products predicted by the model, even over 10 - 17 h when the NO concentration is ~ 0.05 ppb is puzzling. It seems that RO2 + NO should be much faster than RO2+RO2 under these conditions. But, of course, it depends on the rate coefficients assumed for RO2+RO2, which brings back to question 2 above.

- The results in Fig.8 and 9 showing major losses of RO2 at very high (fig. 8d) and very low (Fig. 8e, 9b, 9d) latitudes are rather surprising and need to be discussed. How can there be major RO2 losses at these high latitudes, where the RO2 concentrations are expected to be small anyway (few VOC emitted + lack of light for 6 month of the year) ?