 Hi, I'm Leandro Liborio from the theoretical and computational physics group at the scientific computing department in the science and technology facilities council of the United Kingdom. I am currently in charge of the MUNA spectroscopy computational project where we develop software tools to the interpretation of new experiments and In this talk I would like to present the method that we use for finding the MUNA stop-insight in crystalline materials and also I'd like to briefly discuss the theoretical basis for this method and also to show how this methodology has been implemented in the web-based workflow management system and graphical user interface called the Galaxy. The tools developed by the MUNA spectroscopy computational project are shown here, and in this talk I will focus on PyMUNE suit, which is the tool we use to find the stopping sites of MUNE in crystalline materials. And in particular, I will focus on how to use the implementation of PyMUNE suit in the Galaxy platform. Now, let me start with some of the key or perhaps the key theoretical concept behind the method, which is the use of the unperturbed electrostatic potential of the host material. In this paper here at the bottom there is a detailed description of how the method works and how it is implemented. But here I will focus on the main ideas that differentiate our, let's say, variant of the unperturbed electrostatic potential method from the classical unperturbed electrostatic potential method or UEP method. The classical UEP method scans the electrostatic potential looking for minimals in it, which are associated to the stopping sites of other variables. Our method does, let's say, a further classification of those minima by estimating its attractor side. So if you look at this picture here on the top left, the two minima have a similar values, but the capacity to attract the MUNE is greater for the minima on the left, which makes it more likely to be a potential MUNE stopping site. And we estimate the probability of a given stopping site by calculating the size of the clusters that form around its associated minima. This method is quite efficient for scanning the electrostatic potential in a sample and relies on only one density functional theory simulation for the host material. And we have tested this method in many systems, but we also recommend that the users exercise scientific criteria when analyzing the results from this method. And this method is implemented via a workflow that comprises essentially six steps. Step number one, we run a density functional theory relaxation on the pure host material. Once we have that in a step number two, we populate the interstitial space of the host material with randomly distributed MUNE. And the advantage here is just for visualization purposes, because what we do in reality, what the code does in reality is to generate a set of unit cells, each one of them with a MUNE in a different place. And then we place all of those MUNE structures in folders. And then on each one of these structures, we calculate the electrostatic forces on the MUNE and move the MUNE in a direction that minimizes those those forces. And finally, we create clusters using the relax structures. The size of the clusters and the minimum energy structure in each cluster, we identify potential MUNE stopping sites. And then all of them once all of this was done and assuming that the MUNE does not perturb the hope so sorry, we do all this assuming that the MUNE doesn't perturb the host. But once the MUNE stopping sites were identified, then we perform a further relaxation that is the function of theory relaxation to estimate the host distortion. If that's what we are interested in, I mean if you want to know the way in which the MUNE distorts the sample. And this is essentially the workflow that you use to apply our method to find the stopping sites in crystalline materials. Now, if you want to run the method using its command line version. You first run the corresponding density function of theory simulation using the casted code. This is done outside in a normally in a computer cluster and we can provide access to the SCARF cluster of the scientific computing department to users of the MUNE source. And once you do that, then you sequentially apply these commands to obtain the potential stopping sites and then relax the final structure to estimate the lattice distortion again in a supercomputer. And each of these commands are applied, each of these commands are applied using certain parameters that I will describe in the next slides in the context of MUNE Galaxy. So that was the previous slides show what you have to do if you run your method using the command line tools. But if the method is run in Galaxy, well, then the steps of creating and relaxing the random structures are condensed in one Galaxy tool, which is called PyMUNE suit errors UEP optimize. And then we run another tool for doing the clustering, which is this one in step number four. And then we do the density functional theory simulations just as we did before with casted outside of Galaxy. So, in the next slide, I will describe in more detail each one of the steps in the workflow. And then I will run this workflow in Galaxy. The first step is the relaxation of the pure host material using the casted code. I should say that you can try not relaxing the something just running what we call a single point calculation and see which MUNE stopping site you find and see if that helps you with interpretation of the MUNE experiment but in any case, a geometry relaxation is one of the easiest calculations that you can run using the density functional theory code. I should mention also that we in the scientific computing department can give users of the MUNE source access to the scientific computing department cluster called SCARF and also to the casted code. So we can provide some basic support for casted. It's not difficult. I mean, as I said, the geometry relaxation is one of the standard types of density functional theory simulations. However, if you run copper, like the example that I'm going to show you in the next slide, that's that's easy. But if you run something like this, well, relaxing this may be a little bit more difficult. So, as I said, the example that I'm going to show here is based on copper, but you will probably have to deal with more complex materials, especially if you do superconductors or complex magnetic materials. So when you run a casted relaxation in your system and once the relaxation is finished, then we need just three of the files that casted generates to apply our methodology for finding the MUNE stock inside. Those files are the main casted outpost, which is a dot casted file. The file containing the charge density, which we will use for estimating the electrostatic potential dot then FMT. And then the relaxed structural file for the host, which is the out dot cell file. So that the next step is the creation of the randomly populated neonated structures. And the main parameters for this step are connected to how the meals are distributed in the interstitial spaces of the host material. Essentially, we have three parameters, the Poisson radius, which controls the interatomic distance between the muons that you put on the interstitial spaces. The Vander Waals scale, scaling factor, which controls the atomic distance between the atoms of the host and the muons. There's something called the Gaussian width factor, which controls, in a way, the size of the atomic nuclei. We have a 0.5 default value for this Gaussian width factor and it's normally okay, so it's better to try to change this order to values. If you do this population of muons in random places of the host material, so better not to touch the Gaussian width factors. And the values of these parameters essentially control how many muonated structures are created and how expensive running this system will be and you can see for example if you change the Poisson radius from 0.5 to 0.6 you can see they have more muonated structures here. Once you have all your muonated structures, the next step is to calculate the analytical electrostatic forces on each of the muons in your structures and move the muons in a direction that minimizes the forces on the muons. There are two parameters to choose here when you do the geometry relaxation. One is the maximum number of geometry optimization steps, and the other one is the optimization tolerance to the forces on the muons. Once everything in random relaxation, and which is the most expensive computational from the computational point of view, most expensive calculation in this process. And once everything is relaxed, we have to proceed with the clustering. And for this, for the clustering, the key parameter is the parameter T, which is a sort of cutoff parameters for performing hierarchical clustering in your system. So the larger the T, the larger the cluster form. And then it is possible to save the structural files with the minimum energy of the muonated structures for each cluster, and these are the files that represent the potential stopping site. And then finally, once we relaxed and clustered everything, we can create and save a structural supercell file in the casted format, for example, in the casted structural format for so that we can then later use it to perform a density functional simulation to calculate the distortions associated to the muon and or other properties associated with the muonated material like, I don't know, the magnetic moment at the muon side, the hyperfine coupling constant for muonium if muonium is what we have inside and things like that. And with this, we have all the information necessary to perform this procedure in muon galaxy. And in this link here, you can find a galaxy, muon galaxy, and the tools that you need for finding the muon stopping site using the galaxy platform. And when you click on the link, you will see the main webpage for the materials galaxy. So, as I said, you see the main webpage of materials galaxy. And on the left, the recent list of all the tools available. And when you choose one tool, like the UP optimized tool. In the second partner you will have all the parameters that the tool requires. And the first thing you need to do is you need to load the casted files resulting from the initial custom simulations. And what I'm doing here in this example is you're loading the relaxed structural file, the file with the charge density and the main casted output file. And you can see that on the right side of the panel, you have a history of the process. Every step that you do is collected in this history and you can see when things are running and if they run okay, meaning you load your file properly, then each step of the history becomes green. And then once you loaded all the files, then you need to choose the parameters for performing the creation and relaxation of the muon structure. So, you choose a Poisson radius, the van der Waal scaling factors, and then the optimization parameters and finally you run your tool. And again in the history, you see what happens when things are running. It's impossible to choose a poor format of the file containing all the generated structures, which can be used to assess, let's say, how well we are covering these traditional spaces for the for the host material. Well, like I was on lunch, we need to wait until this relaxation finishes. This is the most expensive part of the process here I'm using a simple example, which will finish soon, but it may take a while, depending on the size of your system. And then, in the history you see a that two new items. Okay. So, we have green, so we have all the relaxed structures and the file containing the position of all the new ones. And the next step is to load all the relaxed, the relaxed structures in the in the classroom to choose the t parameter for hierarchical crusted in. And then a format to the minimum energy structures and the size of the supercell that we will require for running the subsequent density functional theory simulations, and then we run the tool. And here, you can see that the main output file of the process, the main output file of the process. And we chose the characteristics of all the clusters found, you see, this is the clustering report here you found two cluster each cluster has 20 structures you have the minimum and the average energies for those clusters and something that I'd like to point out is that this is a simple here, a simple system here sorry, that runs very fast, but as I mentioned before calculations might take a while, a while, a little bit longer. Because they are running but I mean it doesn't this is not really a problem for you because these calculations are running on a server so you launch them and you can. Just check them periodically to make sure that they are they to check if they convert or not but they don't occupy space on your computer or anything like that. In my experience, the longest that I spent running a relaxation and placid and proceeding galaxy was overnight. And finally, once the process is over. Well, you can save the entire workflow in in galaxy. You can create the workflow save it and then you can visualize it in a in a graphical form and see how the different tools are connected, and which are the parameters that you that you choose for running each one of the tools. And also what easy that you did with each tools and this workflow is easy to share among galaxy users and also to run again. So, for instance, you can include this workflow in a paper to show exactly the process that you use for finding the neon stop insights. So you can create and we will show this in the tutorial you can create what we call a research object with this workflow. So you can create a DOI for that visual object save it in a place like the nodo for example, and then attach it to a paper. And you have a very transparent way of informing your readers of what you, what you did for finding them in stopping sites with our methodology. So that's it. I hope this talk, sorry, gives you a good overview of what our method is and how it works. As I said, we've tested the method in several systems with satisfactory results. And it is a method that can be run from the command line or in the galaxy platform. It requires one external density functional theory simulation. And in the full tutorial online, of which this video is part of, we will include more detailed explanations and also many examples that you can play with. Okay, and I would like to finish by thanking all the people involved in the Mewan spectroscopy computational project and thank you for your attention I hope you find this video and the whole tutorial useful.