 a nice quick break, coffee or something else. So next up we've got a session that where Oling will take us through some of the parameters, the QM parameters, in CPG input files. So actually looking into the CPG input files that are generated by the interface and going through some of the steps that would be useful to know about. Hello everyone and so yeah this section is going to focus on CPG UK parameters. So we've already had a go at using the interface and seeing how when you run the interface a CPG UK input file is generated automatically and we've taken this and then just this has gone in and run CPG UK in the background doing lots of steps. Let's run each FD step. So what we're going to look at now is how would we want to go about changing anything within this file and why we might want to change it. So we've already had a lecture this morning discussing different basis sets and potentials and exchange correlation functionals within CPG UK and the different types that are available. So now kind of going to put these two things together and look at using these within CPG UK and how you might go about changing these and investigating which ones you might want to use and verifying that what you've done is a sensible change. So what we're going to look at is sorry so yeah what we're going to look at in particular is changing the basis set so changing from using the standard basis mollups that we've seen used in in the runs we've been doing so far. I'm looking at using the B3 lip functional so in the CPG inputs we've been using so far they automatically generate with PVE. PVA is great functional but maybe if you want to look at using a hybrid functional with more potentially more accuracy and a greater higher level theory how would we do this within CPG UK and look at adding in a distortion correction so using the grimmer DFT TD3 correction and another point to cover is how we check the convergence of the grid so this involves tuning the cutoff parameter within the CPG UK input file and look at how we check for convergence of the energy and showing that you've chosen a good value. So I'm going to start off by covering just a bit about the CPG UK best practice guide and so this is something we've put together to cover QMM simulations within CPG UK so this is kind of was put together for see using standalone CPG UK with QMM on its own of a lots of it is relevant to using it with in combination with Gromax so here are some kind of the information that you can find within the CPG UK best practice guide and I've put links here stuff that's particularly relevant when you're using it with Gromax so for example if you want to look at understanding the QMM part there's a section that covers the QMM so the importance section within CPG UK it gives descriptions about all the different parameters you might have and why you might choose these and some information there it gives overview of basis sets and the types available and the different descriptions of all the different types so covering the single zeta double zeta and the polarization ones that we've discussed earlier overview of exchange functionals available the different types how you use these within the CPG UK input so the different sections that you would actually add into the file how you might do these overview of more more advanced ones so hybrid methods PB0 and B3 lip also covers some information about pseudo potentials, dispersion corrections and then an overview of important QM parameters so what these can be set to why you choose these values and how important they might be there's a bit about troubleshooting at the end so yeah that covers most of the stuff there there's also overview of how to run CPG UK so setting up what the output actually means they might have looked at this yesterday it prints quite a lot of output what is actually going on here it's useful to understand so yeah it's just something to be aware of and during this exercise we will refer to it and point to the relevant sections so part one of this exercise is going to look at changing the basis set so we're starting off with the input file this will be a CPG UK input file on its own and you'll have this input file and the PDB file which contain the coordinate information so this is kind if you've gone away and run the interface this is typical input file that it might generate and we're going to look at tailoring options within this input file and changing these so the first step is we're going to look at changing the basis set from this basis mall ops basis to the hfx basis so we're doing this here in preparation for the next part of the exercise where we're going to be looking at using B3 lip when you're changing to different functional and look particularly hybrid functions you might want to look at changing the basis set in combination with this so particularly using basis mall opt with hybrid methods it's quite a large basis set and if you try and do this your calculation will be really slow especially if you've got a larger number of qm atoms so yeah this is why you might want to change it to the hfx basis so you can find the download the input files from these locations here and yeah get set up on archer to log in touch to again and we'll start running some of these calculations so key points to change um you'll have to change the basis set file name so this is the file name for for the basis acts and i'll show you where these files have come from and what's in them in a second and then look at changing the basis set option um within these as well so this is given for each element so your key line here is the basis set file name which is here is basis mall opt i want to change this to hfx basis and uh further on down in the input file you'll find a different all these sections for each of the different elements and here your key line is basis set um so here which we want to change to tz vp this tz v to p gth so this is a triple valence potential with uh two polarization functions so here are the instructions um i'll just switch to sharing my terminal now because i think that's gonna be more helpful okay can everyone see my terminal window yep great um so here i have all the files we're going to need for the start of this exercise so if you need any of these you can get them from the commands shouldn't exercise so just first go and grab those um and then the job script i've also placed in the practical instructions so you can just copy and paste this into a text file so yeah this let's just take a look at our input file so at the moment this is the typical one generated but when you run the interface so you can see you have the basis small opt your potential name there um pbe used as the exchange function exchange correlation functional and you put your cell your qm atoms listed already which has been generated nicely and your link atoms and all the information there so yeah the key line you want to change is the basis set file name which is found towards the top and then these within these kind sections you want to change the basis set itself so yeah i was going to explain what are these basis basis set files so they are actual files um if you've built cpgk correctly they're included from the data directory so if you know where cpgk is installed on your computer so in this case it's installed in this directory here um workshared work y07 shared cpgk cpg8.1 there's also a data directory within this and if you ls this directory you can see all the available basis set file names so you have here your basis small opt and your hfx basis and yeah not only basis sets but also the potentials are stored here um some different parameter files so there's also the parameters needed for dftd3 for the dispersion correction so these are all stored here um and if you ever want to see what available basis sets there are for a particular element within a certain file you can do like a a grep search for that element so for example for hydrogen you want to check for hydrogen within these files um or a particular file shared cpg8.1 data and then your basis small opt uh this gives you all the different basis sets found for your hydrogen um and you can see here the dzvp basis small opt that we're currently using and it's the same for usually the same for most of your common elements they'll be the same available ones so for carbon for example and then yeah for our change that we're making in the hfx basis oh it didn't work as i thought um okay maybe i've just searched for the wrong thing but yeah you can have a look at all these files and they're pretty useful to see what basis sets are available yep can you see the questions or can you see the chat i mean oh sorry yeah i'm looking at the wrong place okay uh okay uh yeah so for a copper to use these even people copper do i need to use these uh more like to say the short range one basis set um so for copper let's just have a look at the available ones quickly i've forgotten my elements copper means to see you here yeah so your option there so a lot of the metals only have the short range um marked basis sets available so yeah this is probably your best option so yeah that's our chance for short range um cool so yeah i suggest if you're if there are any other questions um people go on to looking at that exercise so just it shouldn't take you too long just make those changes within your input file and then go ahead and run it so i'm gonna do this here um you should be modifying the basis set file name just changing the basis set at the moment okay so that should be running so yeah just raise your hand once you've submitted the calculation because we'll go through looking at the output together and just see what you care like or cut spit out warning messages when things have gone wrong and conversely a job completed successfully message um so yeah it does spit out warning messages um you have to be quite careful sometimes it spits out warning messages and they're buried within a lot of text and you need to find them so sometimes it will do a lot do a full crash and it'll be very obvious that things have gone wrong um other times yeah there are warning messages buried within quite a lot of text so you can usually do a search for the keyword warning um um when it's completed so we'll look at this at the second but it prints timing report at the bottom um so you can always tell if it's run to full completion but yeah you don't have to be quite careful sometimes if particularly things like scf not converging um can kind of get buried particularly if you're running like quite a long md calculation okay cool yeah so some people have done this um and yeah there's a very small difference in the energy so basically this step is a preparation step towards the next exercise so if you've run this calculation mine's not done what i wanted um oh i've made a typo that's great um what did i miss oh i didn't change the basis to mallot from mallot right did i change the line i did great so this was just potential let's try that again yeah thanks for noticing that's quite easy to mistake mistakes like that which explains why people could work confused about faces set by an arm to potential i know actually what's wondering about that yeah uh okay hopefully okay brilliant this looks like it's worked um so yes we've got our cpga output file that's been produced and we've also got these gromax restarts.wfn files which have also been produced so these are wave function files so these basically can be used to provide uh like a starting guess for an scf run if you're then to run this calculation again um so these files are actually like quite important for what we're going to be doing for the next step when we use to use uh move to use in b3 lip we're going to take these files and use them to provide initial guess and this will be really helpful because they can change the exchange correlation function and these will help really speed up the calculation so reading these files is always a good thing to do um yeah regardless of what you're doing it really helps with the first scf step um so yeah have a quick look at our cpgk output file so here is output um so print out a lot of stuff at the start mainly reading in all your parameters that you've set in the input file you see here we're doing a qm calculation which is good um and yeah so it starts printing stuff about the scf so the convergence of the self consistent functional um you can see here the energy is printed out and this conversion factor slowly getting smaller um so it left the inner loop after reaching 20 steps um this 20 steps is a parameter set in the file so you can modify this for it to do more steps before it um goes around and then does a another outer scf loop um so all this is explained in the best practice guide if i'm going a bit quick here but eventually you'll see this number gets smaller and then it'll print scf run converged in 15 steps so you know it's been successful um and then it'll print out the final energy here so this is your total energy the qm energy this is a printed atomic units so heart trees um which isn't always the most useful unit but this is the default unit used within cpgk so if you want to convert this into a more useful unit um just go and look up the multiplication factor so i've written somewhere in the notes the multiplication factor to convert this into kilojoules primal um then a lot of statistics and the very end here references um it prints a time in report to this always be printed at the very end and you can see the run time for the whole calculation here um so this cpgk is the whole thing it's just the sub routine it's the whole code um and then this will be your run time for the whole the whole calculation and then also prints time reports for all the different sub routines called which is sometimes pretty useful okay so yeah this is what you record if you're looking at the run time and then further up the energy um which i showed so yeah within your notes you can see i put some commands for extracting the energy and the run time from the output so yeah the next step is to just make a note of these um and then also we want to keep this wave function file for the next part of the exercise so to ensure that it's not overwritten we're just going to rename it because what would happen if you ran this calculation again it would overwrite the restart file wave function files the same thing again so we just want to kind of checkpoint this so um we're just going to rename egfp restart wfx and then that is saved um yeah so that's the end of the first part um so yes um raise hands if you're ready to move on to the second part um i'll answer some questions although the qm atoms are defined in the cp2k input file the pdb file has the res student res name column filled with mm and qm file is there a way to instruct through file that info from the pdb file or needs to be redefined in the input file um so it has to be defined in the input file that is the part where cp2k knows that they are qm atoms and the thing is here that everything we're using has been generated by the interface so the pdb file and the input file are kind of generated at the same time based on what the qm atoms are so it's yeah it's kind of all done at the same time so it's when you're producing this pdb file with the qm atoms you're also producing the input file in the interface so yeah it was it's not really a point um but yeah the cp2k needs very much needs um the qm atoms in this format in the input file that you see but yeah hopefully the internet if the interface does this all for you okay yeah good point Arne um yeah gramax restart when function has nothing to do with gramax so yeah this is just a name um this is defined in your input file it's the project name so it's yeah it's just a name you could put anything there and it would work just as well but this is what is output when we run the interface into this file okay so i'm going to move on to the second part um let's just clear that um so yeah second part of this exercise we'll look at adding in a b3 lip and dispersion corrections so we've already looked at this the input we have this uses pbe however we can quite easily change this within the cp2k input file so we're going to look at changing it to b3 lip um so this is a hybrid functional and half of this will be half of it well not half but a portion of it will come from how to focus change and thus thus it's more accurate but you'll find it's more computation expensive than gga methods such as pbe so your area of interest in the input file if you scroll down you'll see this xc section um so this is what it looks like at the moment we just have um some parameters specified here the density gradient cutoff and tau cutoff and we have the xc functional section here which just lists one thing that we're using pbe so what we're going to do is be as we'll change this to use dftd3 and so yeah quite a lot of changes that go into this it looks like quite a lot of text but i'll just quickly explain it um so you can see here um you have the different um contributions which make up b3 lip so yeah these numbers here tell you the amount of this contribution and list it so using 81% of the lip correlation 72% of the back 88 exchange and 19% of lda correlation um so if you look back on the lecture this is actually in the formula for b3 lip so all we're doing is just directly specifying how much of these different parameters we use um so this part comes in the under the xc functional so this is kind of your gga portion um of the hybrid method and then we have this hf method so the hearty hearty fox section um and here we specify 20% of the hearty fuck exchange to be used holy yeah uh just wondering at the moment i'm seeing your terminal is that right oh i forgot to switch okay yeah just just checking yeah i think as soon as you start explaining the the different percentage contributions i couldn't see to reflect it in the in the terminal so okay yeah i'll switch to sharing the different um sorry everyone okay so yes and these are the different parameters i've just been describing um so this is the section that you'll be modified in your file um so yeah we have the hearty fox section here um we're using 20% of the hearty fuck exchange um so yeah all these parameters just basically make up the description of the through lip um so we have this screen in section so this is to help stabilize the um hearty fuck exchange um some parameters here so epf schwarz um this is a parameter you can kind of play with if you find that your calculations are a bit unstable do what you might see is like weird energy jumps when you're running the scf so something look at um and then you have this truncated operator here which um basically prevents self exchange interactions across the neighboring grid neighboring grids so um yeah it's got a cut off which should be less than half the cell to prevent this interaction across neighboring grids uh memory you don't really need to worry about um this is just set to set the maximum memory that you want to consume um it'll depend on your on the machine you're running on and how much memory there is per process and then uh this well the van der Waals potential section so this is for your dftd3 your grimo dispersion interactions um basically you're telling it you're doing a pair potential you want to use the parameters for dftd3 which are in this particular file um this is again in the same directory as the basis sets and potentials and the type and then your reference functional which is speech lip and the cut off um so yeah i think that covers that part um so some other subtle changes you will also need to make um so in the last section we made these uh these wave function restart files um we just need to add a line here after where you specify the potential tell it to read these particular this particular restart file name so it knows the name of it and we're also going to change the potential here um to one that's optimized for belip rather than pbe um so yeah we noticed before um this potential was dth pbe um just changing this to dth belip because it makes more sense to use this in combination with b3 lip than using pbe um so yeah it's just a small change but yeah 50k has a lot of these potentials and they're particularly optimized for using with particular functionals so it's just something to be aware of um so yeah i'll let people do that um and yeah let me know if you have any questions so just yeah make these changes and run again um yeah make sure you've recorded the energy from the last one or you've moved that file to a different place because it will overwrite your cptk.out file uh yes you can also use dispersion correction with pbe so yeah you don't have to go all the way and add in um hybrid functional and pbe um any reason why the cutoff is 16 angstroms um so it's kind of yeah cutoff for to be within the realm of how you want your how far you want your dispersion correction interactions to be um your van der Waals interactions so it's kind of a typical number it's probably something you could play around with but yeah 10 sort 20 angstroms seems to be the norm oh yeah so yeah why one might want to use a restart wave function so yeah it's usually when you run a first your first scf step with without a restart wave function you'll see um it does a way more like scf steps to try and converge um and this will like obviously take more time if you're supplying the restart wave function it's already got like a nice initial guess um to start with and it will just basically speed up the calculation and reduce the number of scf steps um and you'll you'll see this if you when you're running the second part compared to the first part you should see that first there will be a lot less scf steps in the first step it'll still take longer because we're using a more expensive hybrid functional so yeah don't get confused about that but um so we felt long can we calculate this with different functional basis uh yes um well here we've got the same basis set um I think usually using the same basis set is key um but it can be with a different functional um so obviously it won't be a perfect guess but it'll it'll help the count help with speeding up the calculation so obviously your system is still the same like your atoms and your coordinates so yeah has any people done and looked at the energies um so you see there's a slight change in the energy okay so yeah the smart energy here is using b3 lip and then the one below was the first one we did with uh pb0 um so you can see it's roughly 0.04 difference in the energy here this is in heart trees so what's that in kilojoules from all okay yeah so quite a big difference in kilojoules from all so yeah this is just a static energy calculation so you can see there's been a difference however if you were like doing this for real you'd probably want to go away and calculate your actual property of interest um so if you're running like an empty simulation you might want to go away and compare the two or if you're calculating say an energy barrier or something like that um oh yeah the time so how long did this one take 25 seconds how long did my other one take which is upper directory I don't know about the same okay that's interesting um so yeah this time yeah so I noticed this too um I'm guessing it's because we've supplied the scf wave function for this one so in the first one um the first part of the calculation we'd have done we did what so we did 20 scf steps and then another 15 so we did 35 total so we did more at the start and this one it converged quite quickly I believe yeah so only 13 steps so you can see each one of those steps is taking 0.4 and in the other one just hopefully we put it 0.3 so yeah each scf scf step was shorter in the first one um so this is the time per step um shorter and with pb zero but we ended up running more scf steps because we didn't start with the preconverged wave function mine's been stuck for only for 12 minutes oh no um I mean check what's going on inside the output file if it looks like it's converging or not converging it might be that it didn't somehow didn't read the initial wave function file um and that it's it's kind of stuck trying to converge the scf or it might just be something's gone wrong on the compute node I don't know it's a bit worrying yeah double check um double check you've got that that file there the the wave function restart file in the right place and double check the you've supplied that not uh that line which says the name of that file and that the spelling is correct okay yeah so I'll move on to the third exercise now um the third part of the exercise we're going to look at converging the cutoff so usually you have this well you have this integration grid using per map in the electronic density um in cpdk and you want to check that this grid uses fine enough um so you have this cutoff keyword in the input file um that just defines the cutoff in ryberg which kind of relates the finest level of the multigrid so basically the higher you're playing with cutoff the finer the grid um so yeah this is kind of an important step when running a cpdk calculation you always want to check that your cutoff is large enough um otherwise your energies could be wrong um so in the interface the default cutoff written into the cpdk input file is 450 rybergs um so you'll see that uh if you currently open the file that you have at the moment um yeah so this is quite a large fairly large value for the cutoff so this should automatically so it's set to kind of you know be be sure that the cutoff will be large enough regardless of what you're doing because you don't want incorrect results um so yeah it's a reasonable value um so here's for example it's the cutoff total energy versus cutoff for your initial input so the pbe example um you can see here um it's quite a drastic change in energy if you use anything below 150 um and then you can see it stabilizes quite nicely um above that value um there will be some steal some like subtle differences in this value that you can't see in this graph but yeah so 450 uh quite a large value um yeah so cutoff is kind of a funny value um if you yeah you need to use a large enough value if you go way too large it can quite slow down the simulation quite a lot so you have to balance this um but yeah you might want to look into tuning this you definitely don't want to use a value too small maybe sometimes using a really large value could slow down your calculation quite a lot um so this exercise we're going to look at just converging the cutoff um for our new system so we've made changes to the input file we've changed the basis sets from functional the potential with added and dispersion corrections um whenever you make major changes if you were running cbtk on it so and you'd want to go and double check that your cutoff is still reasonable and we converge it um so it's basically what we're going to do here um kind of run in multiple calculations and changing the cutoff each time and recording the total energy yes on to pulling these calculations with the hybrid so taking your input that you've just modified and added in um your b3 lip and dispersion correction um yeah your value is this cutoff here should be quite near the top so just change this example um write it and then run it and record the energy um you're meant to want to make sure again that you're using the restart way function file otherwise you could run into trouble with taking a long time okay i have a question how do we decide the scale c scale x and values in the xe part so these values come from the definition of v3 lip itself um so i don't know if i can reshare the best practice guide quickly okay so for example here with pb zero the definition of pb zero it's comprised of 75 percent of pb and 25 percent of party foc exchange um so this kind of defines how these values are set so you've got 25 percent here for the heart to fuck exchange and then 75 percent of the gg exchange and then 100 percent of the gg correlation you can see the equation for it here b3 lip is more complicated than pb zero so it's basically comprised of an lda part which is of the exchange function um some different parts the heart to fuck exchange another part of the gg exchange and some correlation from lda and that's going off the screen i can't okay make it bigger um yeah some other parts um so this is covered in Emiliano's lecture um so basically these are the different parameters and this is how they are defined in the cpdk input so you can see um we've got this contribution of the lip coloration here which is not going any one and this is set here um in your input file um so yeah it's somewhat a little bit confusing to work out where they all come from some of the mobile straightforward so you can see that's 20 percent of the heart to fuck exchange so 0.2 there um yeah some of them have been come around we have to subtract values from others so you can see yeah your coloration is 90 lda and 81 lip sums to a hole and then if you sum these the exchange parts up as well that sums to 100 percent as well so that's where they come from okay so we have some values already so we have okay one for 50 one for 150 one for 250 for 500 um yeah so we're missing 100 200 300 350 yeah great um yeah if you've done one and you filled it in uh feel free to pick a value in the scratch pad that hasn't been done yet and you can add in another line if you want okay yeah so a big difference for 50 okay um I see that we've got values and most of the cutoffs now which is good I'll let people continue to add those while I just answer some of the questions um so I'm wondering how much the influence will be on the court when I apply a setup when I'm doing um yeah I mean it's difficult to say um so for certainly if you use a very low value of the cutoff your results could be wrong um we actually did a kind of a look at this as well um modifying the cutoff and you see if obviously you can see there's subtle differences in the energy here between different values and then if you go ahead and run like a long MD simulation kind of the effect of that and affecting the energies and the forces which are calculated will sort of be magnified over the course of the run so yeah it can kind of have more of an effect if you're running a long MD simulation um will we see some variation in energy for two calculations of the same cutoff yeah there's a bit of variation um you can run um the same CPGK calculation with exact same parameters twice and the energies will not be exactly the same um so you see the I mean you'd expect them to be the similar have values here for example where two people have done the same thing and yeah that's kind of what I expect so if you look at 250 here um beyond say one to six decimal places um is where you typically start to see variations just yeah this is very small at this point in terms of your actual energy um I was going to quickly share so I've just made a graph I basically took everyone's results and I've just made a little graph here um so you can see this is kind of what the energy is doing with the cutoff so before we kind of said 150 plus it looked like it sort of converged that was difficult to tell here even 200 might not be enough so I'd say like you could possibly say 300 plus 350 should be a suitable value but yeah you can still see some variation going from like 200 to 250 um so yeah that was a quick summarise of that um I will now just cover a bit more of the performance so this is the very last part this is an exercise this will just be me talking for a bit the very last part is just looking at the performance so we've kind of seen how performance can be affected when you change um to use in different functionals so these are some results I got earlier and I did see a difference um from switching from PBE to B3 lip and you could see the runtime had gone up um I don't know what I've done differently this time um so yeah um you expect change in season like a more complex functional such as B3 lip to increase the runtime overall and we only did like one step here so when you're running an MD calculation you're running multiple of these steps you're calling it once every MD step so you do expect the effect to be larger overall um and our first step's a bit weird because we're dealing with sort of doing the first like setup and the SCF calculating the wave function so yeah we had strange results because we're reading in the wave function the second example with B3 lip and not in the first one so but yeah you can see that the time PS SCF step was longer for B3 lip so yeah here's some similar results looking at different functionals so I want to PBE and PBE 0 so this is a the time per MD step this is a lot bigger system so you'll notice time per step is huge here um compared to what we've been looking at before where we only had around 20 qm atoms this has 68 but you can see it's almost a sign of a double in the time taken per step just running on one node of Cirrus here um but yeah you can see it takes much longer but also the the scaling the speed up here is a bit improved so you can kind of justify running on multiple nodes um yeah also some stuff about qm region size here um so I've looked at changing the qm region size so you have 19 and 253 qm atoms but for the same system and you can see the effect of runtime there on runtime there and then some information about using multiple threads so the moment we've just been running on a single thread for these calculations but sometimes it's advantageous to use multiple threads per process so MPI plus open MP so there's some different results here just showing the effect so example in this system using six threads is a lot faster than using one thread so these are just things to be aware of when you're running your own system if you want to look at the performance give you a rough idea of what to expect and kind of what changes might make to help improve the performance um how many cores on nodes you want to run on um so this will all be added to the best practice guide shortly but important thing is just experiment with your own systems cool um I'll have a quick look at the questions but that's all I want to say um could you explain conceptually how these calculations are parallelized um it's yes it's very complicated um it mainly is parallelized over the grid function so when you do multi grid yeah yeah it parallelized over that so you have a grid points and you have a grid grid which consists of a large number of points and they can be split that points could be split along the nodes and each node take some small part of the grid and to wait everything in that grid uh in that part of the grid and then they uh uh sum up all contributions for moments in a way this is a pretty similar algorithm as like PME works in gromax is classical md it's also decompose the system into smaller parts and uh and calculate them separately yes uh fft is also utilized for the grids to pass it to the reciprocal spacing back uh density from it and uh but of course this is a very well known parallel algorithms which are already implemented in let's say fftv library and so on yeah it's also employed actually very funny thing might not many know but in previous p2k at least it's a 7.1 I don't know how it's working 8.1 didn't test it but a sp2k very well liked when the number of mpi uh tasks which are running should be uh some uh square of some number for example 84 a square of 8 why because because of that grid decomposition which it sees so uh the best performance usually which I observed was exactly when your number of mpi tasks not open in peter's but mpi tasks is square of some number uh yes uh don't choose yeah not prime numbers but for course it's better to choose a square of some number yes okay yes I guess it provides but yeah it's very funny it's actually connected to the parallelization scheme over the grid switch cp2k implies it comes to how it splits the grid over the nodes so I think we uh we realize that this is one of the areas where we probably the best practice guide of holly link to the plan to expand with some more guidance on this in future also what I saw from the holly stock that uh if you noticed uh the default cutoff which dromex with the interface set up is 450 readbacks and this is a quite high number in reality why it's done so it done so because it should be suitable for most of the system you can imagine so probably in real uh calculations it always worth to reduce it because it it speed ups the your simulations a lot in reality so yeah but 450 is kind of safe uh cutoff which suitable for 99 percent of the system biological system uh is there are some unparameterized molecules have to be treated in the mem part it's possible to use cp2k because the it seems quite promising because of the f t it charges hash and optimization derivatives and so on so yeah um for parametrization of the molecules usually you should use uh software which not connected to the qm implementation but to the mm so for example finder port field usually one should use uh um pic chamber uh and in that um pic chamber the parametrization procedure is explained i guess and in anti chamber for example you need to generate him uh several uh so you need to generate uh cube files with electron density uh from the several geometries of that molecule which it generates you and usually it should be done with gaussian what they suggest is gaussian uh with some like this relief i guess six three one g star basis but yeah i mean you also can use cp2k in that case yeah so the rest charges is not enough for usually for the force field for force field you need to do a rest charge in several points and then in several geometries because yeah the force should cover more geometries and on the left is that it ever ages through somehow uh yeah that's one of the procedures another if you don't have i mean if you want to use a qm for parametrization it's better to go to the particular force field which you want to use and check how it should be parametrized usually yeah it's indeed parametrized qm like both charm and ember are parametrized qm calculations and basically check uh the procedure it's not a good idea to use another method then the standard user to the force field because then you have inconsistency in your force field because uh your protein for example you parametrized use one method and your ligand will be parametrized in other term which is not the best idea in general yeah but if you want to do a qm mm then it's fine it's completely okay to use whatever mm charges you think will be suitable because anyway you will then switch to qm qm uh uh a description of your ligand for example and qm of course is not dependent on that okay all right well um thanks very much thanks Dimitri uh and thanks Holly uh for useful session so uh now we have a break um lunch or whatever your local time appropriate uh meal is or or not and we'll see everybody back again at uh one o'clock uh central european time um for the next practical