Dear ELECTRODE Package Developer,
I want to know how the ELECTRODE package should deal with the intrinsic charges of semiconductors (such as MXene)? I calculated the charge of the electrode atoms from DFT, but now I don’t know how to add this effect.
Is there any way to achieve this goal?@srtee
Thanks！
Hi there,
I currently have a paper under review, co-authored with some collaborators, where we describe how this can be done together with a per-atom model with some very crude quantum approximations.
The relevant features are coded (but not documented yet) at GitHub - srtee/lammps-USER-CONP2: updated constant potential plugin for LAMMPS under the qinit and ehgo keywords.
Personally, I don’t think these models are ready for prime-time yet, my main objection being that the Lennard-Jones force fields for electrolyte-electrode interactions have never been reparametrised for the (effective) image-charge interaction. I am prioritising my own continued constant-potential research less until I can get around to tackling this problem. If your DFT data would let us properly adjust the short-ranged interactions to account for conductive electrode charging I would be very interested to chat.
Dear stree,
Could you please clarify if, after adding intrinsic charges, the short-range LJ parameters still require refitting? I only fit the partial charge and did not modify the short-range LJ, which comes from the literature.
Thanks！
Currently nobody has refitted LJ potentials to account for charge-equilibration in the constant potential context, as far as I know. This means that if you write a paper using naive LJ potentials, then (1) it won’t be rejected on that basis, because the reviewers who don’t know better, well, don’t know better, and the reviewers who do know better will have to say “well, but I’ve never done it either, so I can’t reject your paper on that basis”; but (2) it will be fundamentally less accurate than a simulation with recalibrated LJ parameters would be, as judged against a hypothetical AIMD simulation of the same system.
1 Like
Additionally, you should know that changing partial charges while keeping short-ranged force fields constant technically creates an all-new force field, because the collective molecular behaviour is a combination of short- and long-range interactions.
Multiple self-consistent combinations of charge+LJ will this satisfy the basic calibration measures (such as density vs temperature of the pure substance as compared to experiment), but using the charges of one parametrisation with the LJ parameters of another is technically incorrect.
One must then remember George Box’s words – all models are wrong, but some are useful. So it can be acceptable to mix de novo charges with prior LJ parameters, but the burden of proof is on you to prove your simulation is meaningful.
1 Like
Dear srtee,
I apologize for my delayed response.Thank you for your comprehensive insight. I truly appreciate and concur with your perspective. I will take another look at this matter. If I proceed with refitting LJ, do you believe I might achieve the initial objective? Then I’d be in a position to begin testing its consistency with AIMD. It would be greatly appreciated if you could suggest a functional version for testing.
I’m currently considering a comparison scheme that draws parallels to recent advancements in machine learning potential fitting. The reliability of the force field is determined by juxtaposing the RDF and density properties between CPM-MD and AIMD.
Thanks！
Dear stree,
I’d like to continue our discussion from last time. According to your statement, the intrinsic charge applied in the MXene simulation is invalid in this paper？（ Effect of Interlayer Spaces and Interfacial Structures on High-Performance MXene/Ionic Liquid Supercapacitors: A Molecular Dynamics Simulation https://pubs.acs.org/doi/10.1021/acs.langmuir.3c03277） Will he be covered by CPM and intrinsic charge not work? So is it possible to add such a function to let the user decide whether to set the intrinsic charge?
Thanks！
I have briefly read the paper you mention ( https://pubs.acs.org/doi/10.1021/acs.langmuir.3c03277 – a hint: whenever pasting URLs, put a space between the last character of the URL and any following punctuation to avoid it being “sucked in” and giving an incorrect URL).
Unfortunately, they have not specified how they did what they did in their paper. The USER-CONP2 package (which is now practically unmaintained by me) does have a “qinit” keyword, but it is undocumented and I don’t know if they used it or not. It looks like they did (given the results they got), but they do not document their Gaussian widths, and without either per-element different Gaussian widths or other metallicity adjustments I am very doubtful of their results. (Hydrogen, oxygen, carbon and titanium must clearly have very different electronic hardnesses!)
Furthermore, they do not include their input scripts. Now, I very sadly couldn’t push my co-authors hard enough to include their input scripts in our own hot-off-the-press MXene HCPM paper – https://pubs.acs.org/doi/10.1021/acs.jctc.3c00940 – but at least we have all our final parameters, including CPM-relevant parameters, and so it should not be hard for anyone to re-use our parameters for HCPM modelling of MXene.
I would still be very, very cautious investing too much time into these methods. Much more calibration needs to be done, frankly without enough funding, and it is possible that – unless people really care about partial charges somehow – ML potentials will make CPM methods obsolete soon, or at least nearly unfundable.
I’ve got too much personal nonsense going on to keep up with the future at this present moment, and my personal dream roadmap still revolves around much more “classic” conductors like metals for now. If we (CPM people) can’t even get water on iron, copper or gold right, we have no business spouting nonsense about MXenes or hBN or whatever crazy materials the AIs start producing (or fail to produce – which will be worse). I also desperately need to write GPU acceleration into the ELECTRODE package – again, there’s no funding or reward for this sort of work.
As Box says, again, it is inappropriate to be concerned about safety from mice when there are tigers abroad. MD papers with esoteric techniques are increasingly publishable without basic replicability, and if a simulation can’t be replicated at all its correctness is irrelevant.
1 Like
Ok, thank you for your insightful reply!
Of course, if you have some progress in this area, it would go a long way towards helping with my deep doubts about calibration.