The paper studies the convergence of a Lagrangian particle-based microphysics scheme (SDM) with the vertical and horizontal resolution, as well as with the number of computational particles. It also compares the the Lagrangian scheme with a 2-moment bulk scheme (SN14). The tests are done using a well known Stratocumulus LES setup DYCOMS-II. The paper is well written and clearly discusses the findings.

I only have some minor questions and suggestions:

- Two ideas are proposed for explaining the differences between the SDM and the SN14 schemes: the numerical diffusion, and the droplet sedimentation. I was wondering if the differences could also be caused by the differences in the representation of collisions (stochastic SDM vs deterministic SM14)? Are the precipitation formation and evaporation rates and locations similar between the two schemes? Do the ql and qr co-vary in a similar way between the two schemes? I'm guessing the precipitation is more "continuous" when simulated by SN14? I realise that the precipitation rates reported in this study are very small, but I was wondering if they could still be affecting the simulations? 

- Figure 3 vs 11. I didn't fully understand from the discussion why the buoyancy production and w variance are so different in the cloud layer between the SDM and SN14 runs? Is it only related to the differences in CC? Can the SDM method match those profiles in simulations with larger ql? Similarly, the integrated tke still looks different for SN14 and SDM without sedimentation on Figure 15? Could the authors comment on why that is the case?

- Figure 13 - Why is the pressure so different between SDM and SN14?             

- dz ~ 2.5 m is a very fine resolution for LES, and yet the simulations do not converge. I was wondering what recommendations the authors have related to that issue. What should be done in cases where due to computational limitations the simulations cannot be run at such high resolutions? Would using stretched grids help? Would using higher order advection schemes help? Any other suggestions?

Technical comments: 

- Table 1 - Seems like most of the dts are the same. Would it improve the presentation to only show the different ones? For example in the last column just say DT_cnd = DT_coa = DT_adv and then just print one number in the column?

- Figure 3 and 11 - Would it be possible to also include a qr plot with the axis limits set to showcase the SDM and SN14 results?

- Caption of Fig 13 - Should be ql and not qt?

This manuscript discusses the convergence characteristics of the stratocumulus cloud simulations (DYCOMS-II-RF02 case) using a Lagrangian (SDM) and an Eulerian bulk (SN14) microphysics scheme. The authors evaluated grid and super-droplet (SD) number convergence and compared the simulation results from the bulk and super-droplet schemes. The article thoroughly investigated this topic and is appropriate for publication in GMD. I have the following minor comments to consider.

The super-droplet simulations show convergence at around 16 SDs/grid for this case. It’s a small SD number. But I wonder if this could apply only to this case where precipitation formation is extremely low. This low super-droplet number per grid box may not be sufficient for cases with significant precipitation formation. It may affect the precipitation formation rate and the spatial structure of the rain and cloud water fields. Similarly, for a polluted case with GCCN, a sufficient number of super-droplets might be needed to appropriately sample the aerosol size spectrum and capture the effect of GCCN on precipitation initiation. I recommend the authors clarify this point at appropriate places in the manuscript or present a convergence test for a precipitating case.

335-340: This argument about a higher droplet concentration for lower SD numbers could be improved. A higher droplet concentration for lower SD numbers may result from a higher multiplicity of SDs and associated statistical fluctuations in the activation process (not a longer phase relation timescale). A lower SD case will have more fluctuations in the phase relaxation timescale, with some grids having extremely short timescales and some with cloud-free conditions. Thus, a higher probability of large positive supersaturation excursions.

Could some of the differences in the cloud field between the SDM and bulk runs be due to the spurious in-cloud activation and the Twomey scheme in the bulk run compared to an explicit activation scheme in SDM?