 Good morning everybody, and welcome to the 51 webinar of the Biaxel webinar series. The talk of today will be a multi-scale QMM simulation, in particular exploring chemical reaction using a novel interface between Gromach and CP2K. The presenter of today is Dimitri Morosov from the University of Ivescola, Finland. I'm Alessandra Villa from the Royal Institute of Technology, and with me is Julian Singh from the University of Edinburgh, and we are hosting this webinar about the presenter of today. So Dimitri received his PhD at the Moscow State University in 2013, and in Moscow he was working on the development application of QMM methods, in particular on biological systems. In 2013 he moved to Finland, in particular to the University of Ivescola, to work in a group of Garrett grown-off. There he was doing a postdoc, and in 2015 he received a grant from the Academy of Finland to go on his postdoctoral research. In 2019 he started to work within Biaxel, and in particular in the implementation of multi-scale methods for performing simulation of chemical and biological systems. Now I will give the word to Dimitri. Good morning everyone, and welcome to my webinar. So today I will talk about the new and developed Gromach's CP2K interface, which I have developed over the last couple of years. And I will show how to use it to make a QMM simulation using Gromach's and CP2K. So, but first let me speak about what is QMM and why do we want to use the QMM in particular. So let's think about the biological systems. So if you want to model a typical biological system, usually it consists of a large number of atoms, usually proteins, something else, much more, including those to the water box around them. They are typically much, much bigger than the 10,000 atoms, let's say, in particular that case it's around 40,000 atoms. And if you want now to model, use a classical micro-dynamics, no problem, you can do that, modern MD packages like Gromach, they can do it easily. But the problem comes when you want to model some reactivity inside the system, chemical reactivity, photo-biological reactivity, whatever. And here comes the problem that for typical system for QMM simulations, which need to be used in that case, because your electronic structure starts to change. It's usually a very small system, so typical maximum number of atoms in current QMM simulations is up to 200, maybe, maximum. And here comes the problem, so how you want to solve that, you want to model a full biological system like that, but with electronic structure methods or quantum mechanics, you can do only small systems like hundreds of atoms. And the solution here is called multi-scale approach of QMM. So the idea here came, like, already in 1980s and 1980s, and in 2013 they received a Nobel Prize. And the solution is the following. You can first divide the system in two particular parts. One will be simulated with in which actual chemical reactivity happens. It could be simulated or calculated with quantum mechanical methods. So, yeah, density functional theory or other initial methods. And the rest of your biological or big, big system will be modeled with a standard molecular mechanical force fields at force to the level of theory. And kind of, of course, in the details, if we look into the details and the question arises, so there should be some interface between those parts. Yeah, and there should be some interaction between these parts to be taken into account. So what these interactions could be, we'll show following. So let us say that interactions within the particular term region, of course, you always can do with straightforward QM description with wave functions, these bases and so on. Also interactions purely within the memory region will be stays the same. So it will still be on the force field level. But actually what is most important here is how you want to complete interaction between QM and them sub sub regions here or between QM and them and particles, in particular. And the answer is, and here you need to solve for two particular things. The first one is how you want to couple. Calculate or calculate the coupling between quantum region and molecular mechanics region. So there is several class of methods first called so called mechanical embedding in that scheme is usually also called subtractive scheme because in that type of schemes. So the total QMM energy is calculated as a full mm energy of full system, minus the mm energy of your QM region. And then you add on top of that just contribution from QM calculation, but in that case your QM will be kind of in a gas phase. So there is the drawback of course is that the QM region does not feel any charges or electrostatic filter from the memory region. So that was introduced as the next type of the coupling scheme is so called electrostatic embedding. And in electrostatic embedding of course the polarization from the molecular mechanical point charges is taken into account polarization of quantum mechanical energies from the mm point charges is evaluated. And this is also called additive scheme. So because now you calculate just mm system with kind of empty space inside it. And then you add on top of that QM calculation in the presence of the mm point charges around. And then you add which is now just rising, it's called polarized embedding, but for that you probably need to use polarizable force fields. And this is not so usual for now nowadays. But then in that case you also take into account polarization of an M part you to the QM part and vice versa. So it's coming to a fully self consistent scheme, but it's not so usable as for now. Okay, this is about the electrostatic. What should happen is electrostatic in QM, but what about the bonded interactions for example, in the following case, you also break in the bonds between there is a broken chemical bonds between the bonds. And of course this cause some consequences. Yeah. So first of all, you need to deal some somehow deal with the force field parameters on that bond broken bond second you need to deal with broken bonds itself. So because for example for QM subsystem, the broken bonds means the dangling electron, which will, which is not so good. And for that there is several schemes again available in literature. So, and the most usable one is the first one. It's so called capping or link atoms. So basically what it means that on the bond connecting on the bond, which, which is, which was broken. We introduce a new quasi atomic center quasi I mean that it's not present to the real system, but it just presented the QM actually system. And usually this hydrogen atom, which just effectively cups this dangling bond. And, and it also fully fully go into the, for example, self consistent field of QM region. So it is present in the QM region but it's, it's influence on the total systems and just subtracted. So the force from that atom, it just goes over these two atoms, which it connects. And there is some other approaches like instead of the real link atoms, you can use also the potentials to cut that valence, or you can use some hybrid orbitals frozen orbitals, but the most general method is usually link atoms. So here, let me present what, what, what have been done by me and my colleagues over the last years, and there is a gromach C2K interface to perform KMM simulations, using, of course, Gromach says the MM driver and C2K SQM driver. I will tell you here a bit. So why using Gromach is of course obvious because we are by Excel. Why we decided to use C2K for that is kind of a good question and the, and the answer is that first, both of them are freely available software. So we want them to be freely available to everyone. And second, C2K, one of the programs one was not so many programs which allows us to perform a fully periodic KMM calculations and KMM simulations. And what this interface basically is doing from the gromach site. The gromach is a main driver. So all integration of the trajectory of the minimization or other methods is done in the gromachs. For all MM interactions within the system. And C2K, what C2K is doing it just first performs QM simulation and secondly performs electrostatic embedding. So basically coupling between QM and MM. And let me go a bit through these terms in the energy, which we want to, which we want to calculate in a typical KMM simulation. So first is a force field, as I said already is done in gromachs and it is standard for any standard force fields you can use. So, yeah, it's comprised of the bonded and not bonded interaction terms and bonded interaction of course is bonds, sand goals, the hydro storage and so on. And non bonded is, of course, linear jobs interaction plus a Columbia interaction between the MM particles only. So the next, what, what, what C2K is doing. So the QM routine of C2K is called quick step. And what it is doing basically is a very interesting thing because it use combined it not use not just a standard ocean basic set like everyone using like 6G 1G or C-CPU disease or something like that. In addition to that basis set, standard basis set, it also uses a reciprocal kind of plane wave basis set, auxiliary kind of thing, which allows that method to calculate periodic system. So like shown on the right slide, so basically now have fully periodic and how it is working it uses so called multi grid real space multi grid approach. So basically it gets initial guess of the density is taken from the Gaussian basis set that oceans is projected onto the multi grids real space multi grids. You can construct your quantum matrix quantum operator and energy functional. You then you do a very efficiently FFT because it can be done very efficiently if you build your grid properly and pass this density into the reciprocal space, in which you can account for long-race electrostatic interaction, basically between the images usually. So that's you can obtain a new density matrix, which you can, which can then can go and be optimized in the CF cycle again and again until your full energies until your energies is converged. So how the CP2K treats the KMM coupling electrostatic coupling, it's also very interesting approach, I can say. So it use so called Jeep or Gaussian expansion electrostatic potential approach. So and what it is basically doing now, you want to include effect of this point charges into your QM region, which is represented by the number of multi grids. It's basically, yeah, it's basically a grids. And instead of doing normal Columbia operator like one over R here with a density, they use so called smearing smeared potential. In reality, it is, it is standard error function over R, like potential, but what is interesting is later it is expanded as a sum of several Gaussians, potentials expanded as a Gaussians, plus some very smooth, very smooth like a remnant function, which is so smooth that it can be efficiently treated in the reciprocal space. So, and of course, our Gaussians can be more directly onto the real space grids. So income, so it's indeed really like quick step is doing but now you are doing this also by including the effect of the MM point charges of the QM region. And, of course, this can be done like periodically, but if you think now that you want to calculate your real system in fully periodic MM box full MM box. And then the problem arises that, of course, two projects, everything come to the multi bit of the QM box. And of course, your, now your quick step can calculate this kind of system. But unless your QM and the main boxes are the same. You have a problem with periodicity because that system which are calculated is not the same as it was initially. Yeah. Unless you're of course boxes are the same. And to deal with that, we are using also very tricky approach. So called a local scheme. So what is basically doing. It is fits into once you convert your function. You fit into that density electrostatic into that density charge density basically your some point charges, which can reproduce the multiple, multiple moments of your system. And then you simply do the evalt like decoupling decoupling basically you first calculate with the evalt like summation. Just the energy of one system, then you shift every system in the position where it should be. And then you recouple them back. So basically places fully classical like electrostatically just by shifting this thing. Okay. And how the simulation flow now looks like in Gromax. So when you think about the normal molecular dynamics, like workflow, simulation flow, it is looking like that. So you read, you want to read your input parameters or current velocities, all that kind of parameters. Then you want to prepare a simulation you want to set up your domain decomposition you want to set up PME and everything else boxes and so on. Then you calculate your energy and forces. Then you integrate your trajectory usually applying sample couplings like temperature pressure and so on. You want to solve constraints. And after the constraints you just basically write any output if you want. Then you can adjust your parameters. I and you are doing that cycle until your simulation will finish. And what is interesting in Gromax that we use so called MD modules framework to make this KMM addition to put additional forces into that additional forces due to KMM and KMM calculation. And this is done like that. So basically what it is do is there is a number of modules available now in Gromax even like for example pooling or density guided simulations. At each step, the MD modules framework, it passes that simulation data to each of the MD module. And then it get back the energy and forces from that. So, and and return it back it adds it back to the whole total energy and total forces. So why why it is good approach because well it's much less dependent and more encapsulated kind of things. And yeah, your main driver is no knowing basically, and as it should be knowing nothing about actual what you're actually doing inside each module. And it's much easier to integrate and maintain. Okay, but now, as I said, like it seems like very simple or maybe not so simple but anyway, now comes the next question. So why then no one is using the KMM or why so low number of researchers using it for as for that we conducted early this year a questionnaire or survey from the high profile scientists. So, and we asked them, first, did they use any given simulation during their research, and only half said that like, yes we use the other said like no, we know, we know but there was some difficult. But it was not successful, or they're considering that, but do not know how to do that. Exactly. So, and the next question was, which issues have you hindered from use KMM. And there was a number of things like most of the people do not know exactly the background in the camp theory because usually to perform came simulations you need a lot of knowledge. And they do not know suitable parameters for particular program, and they want them on tutorials and so on. Every, every, every of these things we can solve is in bioxel. And, yeah, our idea here, we need, we definitely need to make our interface simply to use. So, and I will show you that on several a couple of examples how to use that and it's really simple. Let's first go with a typical workflow for the biochemical KMM modeling. Usually it looks like that. So, first you take some system, usually from the PDB database. Yeah, or whatever you can take some coordinates from crystal from crystalline from criteria structures. You need to fix everything which is missing from your system, residues, protonation states, at box that's all and so on. And then you perform your probably MG equilibration thing. So then you put the box into the group so that box into Gromach sentence to some dynamics and the production run search for the good geometries. And then what you want to calculate for example we want to calculate some reactivity. And here it is very easy without new interface. So, you just need to add. You need to define the group of atoms which will be a QM atoms in the index file of Gromach's. And you need to put a couple of parameters really small number of them. And then you can perform a KMM calculations. So it can be the geometry minimizations ab initio molecular dynamics born upon him or dynamics. You can go do umbrella sampling you can go calculate even absorption spectra. So, yeah, but the thing is that it is very simple in a way, but we still want what we want to achieve we want to achieve that the advanced users, they also want to modify everything yeah usually to adjust parameters standard parameters and so on, and we want them to be able to do so. So, how it is working, basically, on the first step of course when you're going from dissimulations to KMM simulations you need to change really do topology and this is already not so simple usually, but in our interface it is done pretty straightforwardly and automatically. So, when you're doing so what the interface will do for you on the ground even level when you create your TPR file. It will first remove all partial charges from the QM atoms, because you don't need them. You remove all the energy on the interaction between the QM atoms, but you keep them with them and atoms. So, you remove all bonds within the QM region bonded interactions you remove angles, which connects contains two or more. So, for example, this angle. This angle will be cut it will be removed basically you remove all the hydrals which contains three or more QM atoms. You also want to clean up for example if you have water molecule in your QM system you want to remove constraints which usually applied in the conditions. And finally, you want to make a link atoms. And you want to generate input file for the stupid okay, this is done automatically now in the interface. So, there is some features which now supported as a grumpy people level. So, there is sample setup for QM parameters which should be suitable for more many biological systems. Compatibility with most of the simulation techniques, which is now available in gromax, maybe with exception of the FEP, because of energy perturbation you need. Yeah, you have several topologies and there is much more complicated. Compatible also of course with any gromax tools because what you will produce you'll be completely in a gromax format. So, any third party tools or gromax tools can can be used to analyze your trajectories. And it's also of course supports a highly parallelizable simulation methods like umbrella sampling, AWH pulling simulations and so on. Let me show you so how it is set it up. So it's really easy to use. So you take your system which is for classical simulations. Then, in the index value defined the group of atoms. Here is called QM atoms, which will be actual QM part. You said the QM active flag to true, which activates that module. You also can choose some of the preset parameters for some methods. For example, here it is PBE which means PBE with double the databases. There is a number of them will be available and even now available. But you can change it freely, I guess, if you want. And you also need to put charge multiplicity of course of your game system, total charge and total spin state basically of your game system. And what you will get on the output as the output of grom pp will be not one but three files. It will be standard gromax double dot tpr file, which will be your gromax parameter file. And they will also generated CP2k input file and CP2k and PDB file with a split between QM and MMS system as well as this definition of point charges for CP2k. So basically these two files is CP2k input files. In the standard simulation like that, you can throw them out. I mean, yeah, they're also saved within the tpr. So you actually don't need them. But if you want to change something, you are free to do so. And with that three files, you already can perform the calculations you want. Okay, so here is an example of the system. It's really like a very simple system. It is GFP chromophore called HPDI. And it's in the box of waters, T3P waters with sodium ion to neutralize the whole system. So the QM charges that case because it's anionic chromophore. It is minus one multiplicity one. And here how it looks like the input in the gromax. And if you check also what will be in the input file for CP2k, generate during that, it will do like that. So you have here some header and you have here the main sections, first of all. So, and just in case if someone don't know, so this section will set up GIFT section will set up SCF and grid set parameters. QMM section will set up your QM box, small box and GEEP and all that kind link atoms and so on which is connected to QMM. MMM section here will be just defined that CP2k itself should not calculate any electrostatics between MMM atoms because it's done by gromax. And subsist is just a system definition like atom kinds, point charges, which is read from that PDB file by this Gaussian basis and so on. What is in that PDB file generated for you by the interface will be this one so it will be split between two residues QM and MMM. Yeah, the names are in the following. And in the last extended beta field, beta field so called, it will be the partial charge. So for example, here's definitely 3P water standard charge. So let's go and try to first minimize doing energy minimization with that system, which I showed before. And here what it's look like. So it is kind of trajectory and here is step. So yeah, as you can see the energy minimizes. And the second after the minimization was we don't do, of course, we want to do dynamics. And here is a dynamic. So here we use some PT ensemble. So we, we basically want to do it as 300 k is one bar. So here is a dynamics and the graphs of temperature and pressure. So the both temperature pressure couplings is working here working here. Yeah, and here for that system on the 80 cores. So basically two nodes here 80 cores. It was approximately 10 picoseconds per day. Wait, so 10,000 steps, one hundred second type steps, it's 10,000 steps per day. Okay, but if you say that, okay, I'm advanced user, I want to modify myself, can I do that or even I can manage to generate CP to King put myself so can I do that. Yes, sure, you can do that. So for advanced users, we have, you still need to provide, of course, in the index value QM setup. So basically, what is your QM atom defined, which atoms are QM. But then you can put here your own CP to K input using KMM method input and providing input file via this option. And then it can basically be any input, which can evaluate any force you want. Yeah, so what should be in that input only that the force evaluation should be in that input. And, yeah, you can perform that simulation, no problem. And for example, such kind of simulation, there is a big bigger system. This is so called phytochrome protein, it's also in the box of water, I just cut it out of the box, because it really is really big. We wanted to adjust because the system was so really big, we want to adjust some standard to keep it a parameter so it can be more effectively treating the whole box, which is really big. But for that system would calculate that, that reaction of the chromophore inside that protein so it's located here. So we just at some point on the photo cycle, there is a visualization going from that to that configuration. And we want to make the profile of that visualization on the ground state. And we use umbrella sampling for that. So basically this coordinate of the reaction. And here is the profile. Yeah, looks pretty smooth. As for umbrella sampling, here is used approximately 30 picoseconds per one umbrella. And if you think like, okay, I want to know how to use CP2K for how to set up advanced CP2K input options, we also made that within the by itself. So thanks to Arna and Holly, they are made. Oops, sorry, they are made best practice guide on performing KMM with a CP2K. So there is a lot of parameters and explanations and combinations and what each parameter is doing and some troubleshooting equivalent for the biological simulation so thanks to them. And yeah, I think that was almost the last slide. So, my presentation so I want to have the following acknowledgement to Professor Green, David Grunow, and Kristian Blau from KTH is helping with Gromax. And some other people like WifiModi, Arna, Holly and Emiliana as a collaborators who are tested, make a lot of effort in testing and providing feedback. So now I finished my presentation and yeah Q&A session simply starts. Thanks very much for that interesting talk. So we've all we have got a large number of questions so luckily we've got plenty of time to go through them. Our first question is from Adrian. Adrian I have, I do not seem to be able to unmute your microphone so I will ask the question in your stead. Our first question is, as I said from Adrian and is, is the broken symmetry approach accessible in this QMM framework? Do you mean the broken symmetry in the QM region? I suppose that what he wanted to ask. Yes, it is possible so you can use a different spin state symmetry and you even can provide your own CP2K input file. Or you can modify the one which will interface generate for you to perform any kind of simulation. Yeah, it's definitely possible. Okay, thank you very much. Our next question is from Djokvan. And Djokvan asks, can a single QMM simulation take advantage of both parallelization? And if so do both Gromax and CP2K run in parallel using separate resources or how does it work? Yes, so the idea for now. Yeah, of course, there is no point of running simulated Gromax and CP2K. So what's happening? The Gromax is the main driver. It's basically the Gromax simulation, but at some point what will happen inside the Gromax, it just calls the CP2K routine to calculate energy and forces and then CP2K use all available computational resources to do a QM calculation and pass back the energy ingredients. Both of them are working in parallel, in MPI for example environment, and both of them will use the full advantage of the MPI. Great, thank you very much for that question, for that answer. Our next question is from Isabel. Isabel, I do not seem to be able to unmute your microphone so I will ask the question in your stead. There are two questions. Can we do a QMM dynamic in excited states? And is it possible to run these with GPUs? So there is two questions. The first question is, for excited states, CP2K is not the best programming in reality. So it's usually designed to do the more like enzymology studies with ground state. The interface could be extended to do also excited states simulations, but probably not with CP2K. So yeah, in the future we are ourselves doing the excited state simulations, but we are usually doing them not with CP2K, but with some other programs. And this is also regarding to your next questions about the GPUs. So for now, as I know, the only quantum code which can use effectively the GPUs, especially with excited state combination is TeraCAM. Yeah, but the problem with TeraCAM is that it's not a free program, you need to buy it. And that's why it is not considered now in the buy itself also. Yeah, but yeah, definitely for myself, I would be very interested in doing so. So at some point maybe. But for now, the answer to your question is probably no. Sorry. Thank you very much for that answer. The next question we have is from Chen Gong Hui, who, again, I do not seem to be able to unmute anyone today, which I don't know why. But regardless, the question is, well, first a comment, 10 picoseconds a day looks very promising. And the question, how many QL atoms and how many cores, can we get? Yeah, yeah. Yeah, you can, you can, you can. So how many atoms and how many cores are used? How many atoms and how many cores for the simulations. Yeah. Okay, so for a very simple system like I showed like HBI, it was like here. It's really not so big. It is like, how many here 20, 20 something 25 maybe atoms, but the good thing about CP2k it is not very, it is very good. It is very good scales with a number of atoms. So, because the number of atoms increase number of oceans and CP2k is many grid code. So until you not increase the QM box, your CP2k scales almost linearly with a number of atoms inside that QM box. So if you start to expand in your box, it will slow down. So there is, should be a very good compromise between the QM, and in the QM box region. So how big is your QM box. But 10 picoseconds a day it can be achieved and how many cores here it was done on the 80 cores. It's two nodes with 40, 40 cores. Because Intel Xeon something, it is quite new, but still Intel Xeon, basically from last year. I don't remember exactly, but it's 80 cores. It can scale even further. Yeah, definitely until like four nodes I need like it can scale. And yeah, I think I think you very much. Our next question is from Andrea Golovin, who I have been able to unmute so Andrea if you would like to ask your question please go right ahead. Andrea doesn't seem to be able to ask their question so I will ask, I believe they have two questions. The first one is to check. Do I understand right that CP2k is run by Gromax at each time step. The second question is just get the easy answer. Yes. The second question is CP2k write gradients to disk, or does it pass through the memory through memory. Everything is done. So it's another good. Yes, it's another good thing about CP2k. They have a library like implementation so you can, in reality, it will be one executable file. It is basically Gromax linked against CP2k. The whole CP2k is linked into the Gromax so it is just going directly through the memory of one application so you even don't have the different applications. It's one executable file, one application. It is going through the memory of that application. Our next question is from Santosh. Okay, how does a QMMM approach different from a fragment molecular detector? From what? Excuse me. From what? Have you heard of FMO? Fragment molecular orbitals. Yeah, it's completely different. So this is QMM approach. So FMO is fully quantum approach usually. So it means that you split your system into fragments and in each fragment you do a quantum calculation and then you combine everything into one huge quantum system. So this kind of divide and concord methods here. I hear it's QMMM. So here you have only one QMM. I can say fragment in terms of FMO and everything else is just a point charges around. So it's really different thing. And it's much faster. Yeah. Okay. Thank you Dimitri again for that answer. And our next question is from Eleni. Eleni, your microphone is unmuted if you would like to ask a question. No, I would like to ask if CP2K uses a subtractive or additive coupling scheme. The GIP, yeah, the answer is following the GIP which is used by the CP2K. It's an addictive scheme, but without the polarizations for now. Thank you. Thank you again for that answer. Our next question is from Mert. I am not able to unmute Mert's microphone, but it's a very practical question. Is this already on currently implemented in Gromax? The answer is yes, we have an internal branch in the Gromax which is working with that. Yeah, I've showed you previously an example of simulation. So we definitely inside the Gromax already. The thing is that there is some complication with this compilation procedure because we want to also keep it as simple as possible for users. But I suppose that within the release of Gromax 2021, I will make a separate branch on the Gromax GIP Lab, which everyone can download and test where there will be a module activated. So, and everyone who are familiar with compilation procedures who can compile himself, he or she could definitely do that. Yeah, as a plan is a following for now. Great. Thank you very much for that answer. The next question is from Tomasso, who asks, is it possible to use metodynamics on top of Gromax with CP2K? Metodynamics as a definition, so anything which is implemented in the Gromax you can use. As for metodynamic side, well, I'm strongly disagree that someone should use it. It is very low, a slowly converging method in reality. In Gromax, there is much better method called AWH, which is kind of a pretty much very accelerated kind of that approach, bias force approach. And I suggest to use that. But yes, this is possible. I mean, any bias simulation except of FEP, it is possible to do with that interface. Great. Thank you for that. The next question is from Suman. Suman, I do not seem to be able to unmute your microphone, so I will ask in your stead. Is it required to generate the MM parameters for the QM region as well? And if yes, is there a recommended procedure for doing this? So, the answer is a following. If you want just to do QMM simulation, no, you don't need to do so. You can just define it in Gromax as an empty region, empty what I mean, with zero point charges, with no parameters on the bones, dihedral and so on. And it will completely work. My answer here should be also the following, that usually when you do that biological simulations first, before the QMM, you want to do some molecular dynamics to calibrate your system. Because in the crystal, your system is usually not in the same confirmation of the solution. You should be very kind of aware of that. And for that, you will need some molecular mechanical parameters. But the good thing is that for most of the non-standard biological residues like chromophores, for example, very good and suitable parameters could be generated by the Ember tools from the GAF, basically, using the general Ember force field. So, if you Google for the Ember tools, the GAF, general Ember force field, then you can do that. You can generate parameters for your particular molecule and do the MD simulation, basically, initially, to relax your system. But if you want just to do QMM, yes, if you can define in your grommax, if you in your grommax topology you can, you will just simply make an empty bones and empty zero point charges and so on on all QM atoms. You're free to go. Thank you very much for that answer. Our next question is from Peter. Peter, I'm unable to meet your microphone, it seems to be a constant today. Peter asks, starts by saying thank you for your interesting talk. You've mentioned that you used PDB format to provide data for CP2k standard PDB format has a precision of three decimal places. Is this enough, or does it add? Okay, I see, yes, yes, I see the points come from so I will show a bit what he probably talks about this kind of format. For CP2k, this coordinates is used only for the initial setup. Any further coordinates will be passed directly through the memory to the CP2k for the actual calculation. It will be passed through the memory. And it will be a full double precision. So there is no problem with precision of the coordinates. For the point charges, the CP2k can reach them from so called extended beta field, which is not, which is starting in PDB I think at the column number 81, if you check the PDB format, and then it can go indefinitely almost. So the precision and the point charges here it's six digits, but it's already more than enough usually. The coordinates, you should not worry much because on each step of the actual calculation in CP2k the coordinates will be provided directly to CP2k in a full double precision. Yeah. Thank you very much for that answer. Our next question. We're nearly through the list by the way just just in case you But our next question is from how young. And the question is, can CP2k find transition states? CP2k itself can find transition states, the question. If you want to, you do that. Yeah. If you take a bite self can do that. If you're talking about the interface. The gromax have no for now capabilities to do the transition states except of one you can do umbrella sampling in the umbrella sampling you can you can find that. So you can do umbrella sampling on top, for example, or you can do first the pooling simulation to pull your system from react reactance to product states through some kind of transition state and then on top of that pooling simulation in some points you can do also umbrella sampling simulation and thus find the free energy profile of your reaction, including some point near the transition state. But the exact transition state for now for the QMM system know it is not capable to do, but there will be some methods in the future which we allow to do that which we are planning to implement like not just elastic band for example for searching transition states. Yeah. But for now only something. Great. Thank you. So, our penultimate question is to do with where can one find instructions for how to build this coupled gromax EP2k. Yes, it already was such a question where it will be available. So, I probably already answered to that so once there will be a release version of Gromax 2021 I guess I will just make a separate branch on the Gromax GCLAP with interface with active interface, which everyone could download test themselves, and there will be of course the instructions in the review file how to build it. Sure. Yeah. But to do so I can say now that first you will need to be able to compile yourself a CP2k which is already not so straightforward procedure. Thanks. And our final question is to do with a specific system. Neda would like to know about metal ions and ligands such as HEM. Is CP2k a good method or should something else be used to evaluate the biological mechanisms in. Yeah, the answer is the following so CP2k is initially created as a tool for the metals. So I mean it is came from the material science so yes it definitely can manage to to calculate metals for you. The only thing is that for the metals usually you need a big different basis that to be used than the normal basis that we are using now. So if you go to the CP2k CP2k work page and you check. So, if you can do that. But for that most probably to get the correct and conversion results for the metals, you need to provide your own CP2k input with the basis for your metal with the potential for your metal because usually metals use the potential. But yeah, if there's interface is definitely possible, you just need to do additional effort to know which method you want to use for your metal, which method, which basis set, which DFT functional will be best for your particular metal ion in biological system. Perfect, thank you. Before we finish, Arno has asked me to mention that people and anyone who's attended here and attended this talk or listened to this talk and has any questions about using Gromats with CP2k. There is a forum on ask.bioxl.eu where you'll find a section entitled QMM for biomolecular simulations. Please feel free to ask your questions there that's what the forum is for. I also wanted to very briefly mentioned that there are a couple of interesting webinars soon coming soon to the bioxl webinar series. Yeah, so starting next year. First we have the student webinar on the 18th of January, where some of the students attending the winter school will be giving talks about the about the work that they've been doing. And also we wanted to mention that there is currently ongoing as part of bioxl a QMM best practice workshop. And there's a webinar for that on the 10th of December, so that's this Friday, I believe, Thursday, where Professor Maria Ramos will be talking about studies on enzyme catalyzed reactions. And then also come February, there will be more bioxl webinars to come, some of which have already, we already know who the speakers will be but we've not quite set the dates so please follow this space and to find out about those. Finally, I'd like to thank Dimitri again for this very interesting talk. And thank everyone for coming to this talk. And I hope you all have a good day. Thank you.