 This was really a collective work and without any of them this wouldn't have happened in the same way. So this is a quite long work that took place over the last three years. And the idea is coming from the following facts. If you look at the situation around concerning 3D dimensional crystal structures, they are very well known. You have hundreds of thousands of them that are populating databases. So at least the structural parameters are very well known and even more properties and comparatively 2D materials are still a very fast evolving field but where you don't know so much we know maybe a few dozens of them. So the idea here is to try to fill this knowledge gap and if we imagine now we have such a 2D materials database with a number of quantities inside we could think of doing high super screening, so looking for high mobility materials, ionic conductors, catalysts, topological insulators, piezoelectric, ferroelectric materials, superconductors, pintronics materials or porous membranes. So those are some of them were already beginning to do it in the lab. Others are the state of project. And even you could also build a property database with quantities like thermomechanical properties, stability or electromagnetic magnetic and this in itself also has a big value. So that's what we are trying to do and the method is to try to exfoliate compounds that are known experimentally from the 3D structures where the 3D structure is known. So we begin with external databases like the ICD and COD and we look for structure that looks like this. So with a certain screening algorithm that I will describe, we really want to see gaps if you want between some layers. So when we have found them, we want to put a little bit more physics inside. So we then turn to DFT calculation first to relax those 3D structure and then to compute the energy between those layers to make sure these are really weakly bonded. And from that we can filter even further and get our 2D database which we can populate with properties. So electromagnetic magnetic properties, phonons and eventually topological phases. So let me begin with the algorithm to screen those layers. So the idea is fairly simple. You have an initial structure. I drew it in 2D here but you have to imagine a 3D one. You build a supercell and then you try to identify atoms that are chemically connected. And our definition of chemical connection is just based on interatomic distances. So you look for atoms for which the distance between them is smaller than a certain value that is given by the sum of one diverse radii minus some quantity delta. So the idea behind is this plot here from this reference. Essentially when you look at the distribution of interatomic distance in all kind of structures that contain this diver, you have a peak of chemical bonds here. Most of the time covalent bonds but it can also be other kinds of strong bonds. And you have another peak that is made of Van der Waals bonds. And in between you have most of the time a gap that we call the Van der Waals gap. So the idea here is that you want to be sufficiently further from the Van der Waals peak to be sure that really you are inside a peak of chemically bonded things. So you are sure that they are really chemically connected. So when you have found those groups of connected atoms, you want to know if it's 2D or not. And the idea is fairly simple again. You just look for periodic copies of a given atoms because you are inside a supercell. So this makes an ensemble of vectors and you can just compute the rank of these vectors. And this gives you the dimensionality of your substructure. So it can be 2D, it can also be 1D or even a cluster if you don't find any periodic copy. So we can get any kind of substructure with any dimensionality. So since the algorithm is in essence simple, it can deal with many complex situations because it's very general. So for instance, you can get things that are with a stacking axis, not along a crystallographic axis. You can get such structure where the layer really spans a lot of adjacent unit cells. So you really need this atom here and this one here as well as the unit cell here. You can get also these kind of things where the layers are really interleaved and there is no kind of vacuum you could define here. So you really need to look at individual bones or also such complex structure where you have clusters between layers. When you have done this purely geometrical screening, you want to go a bit further and put a bit more physics. So the idea is to try to split such a 3D structure and see the energy of interaction between the layers. So here you need an ingredient that is not given for granted in DFT calculation which are van der Waals interaction because most of these compounds are holding only through van der Waals interactions. So you need to use van der Waals function. We have chosen two, I will show later, a posteriori that they work well. The two we have chosen are the DFT with C09 exchange and RVV10, but there are many more and actually I doubt that it would change significantly the results. So the other ingredients also that you need are good super tensures and here I would like to insist also on another work that we did in the lab which is to compile a library of pseudo-potential. We did not do them ourselves except for one exception but essentially we collected all the possible pseudo-potential libraries testing them compared to Olexon calculation and testing conversions until we get for each element kind of the best pseudo-potential. So this is the SSSP which is available on the web and it's highly reliable. So a few words about the functionals that we have chosen. If you look at the structural parameters of these compounds computed with DFT, if you do not use a van der Waals function and if you look at the error in the interlayer distance, so the out of plane axis if you want versus the packing ratio, so the lowest packing ratio is more empty is your structure if you want. You see that there is a big spread, there is a big systematic error and a big spread in the error. Now if you instead use the two van der Waals function we have chosen, the spread is much reduced and the systematic error is almost absent. So this gives the first confidence of what we are doing. But you need a bit more because we are going to compute energetics, binding energies. So we also compare with the state-of-the-art reference binding energy from RPA which we did not do ourselves but they are coming from here and on these well-known layered compounds essentially the difference in binding energy is not only it's not very significant but it has a certain magnitude but let's say the compound to compound difference is more or less the same as the RPA with the van der Waals function also. So we are pretty confident that we can use the van der Waals function also for the energetics and that's what we did in the end but only the binding energy is not completely enough because it's a bit arbitrary to fix a threshold below which you will say okay those are van der Waals layers, van der Waals bounded and above they are not. So we try to find another descriptor which we choose to be the difference in interlay distance between a non-van der Waals and a van der Waals computation. So if you want, if you take a function like refPB that is supposedly probably very approximately van der Waals free, you can say that those van der Waals bounded compounds shouldn't bound and the interlay distance should grow indefinitely versus something that contains van der Waals that will bound them much tighter together and that's actually what you see. So for small binding energies which are on the y-axis this difference in percent is very big and grows for that graphite actually. While when the binding energy is much bigger this delta is much smaller. So in the end we have three regions, we have identified kind of three regions. One is quite clearly the region of easily exfoliable materials. Here the yellow one we think that clearly not exfoliable or at least with difficulties and in between there is this intermediate region which we are not really sure of but the binding energy is still relatively small so they may be exfoliable. We think that there is still some possibility to exfoliate them. So let me summarize where we are up to now. So we have done the geometrical selection on more than 100,000 structures from external databases. So this gave 5600 layers structure which for some of them more than half of them we relaxed, the others it's still ongoing or they are really much bigger so it's more complicated to do them. Then we computed the binding energies on those, we exfoliated them, we ended up with 1,800 monolayers and we can go further with these identifying prototypes which I will show doing structural relaxation and computing some properties on the most promising materials. So this is what I will show now. Let's begin with some statistics. So if you look at the distribution of point groups among the layer materials versus those of the standard 3D databases it's quite striking to see that actually there is no real preferred point group compared to the external database except for very few exceptions. I like this way of looking at this where you really see that actually low symmetry point group are very present in layer materials so it's not only hexagonal or trigonal, there are really lots, all kinds of things. Also with the structure we did some prototyping, the 2D structure, so we took about 1,000 of these structures and tried to identify using Pimogen and the structure match of Pimogen some prototypes. So the most common one, those are the 10 most common. The most common one contains 67 structure, it's the cadmium diadide, it's also the 1T phase of MOS2. So that's not surprising if you want. There are others that are more surprising like this trital arid that contains a rare earth or even this quaternary compound that is also quite well represented. Now maybe you wonder if these structures are stable. So to evaluate the mechanical stability which is the first step, we computed the phonons. Now there is a catch here which is for 2D monolayers one has to be careful. Typically one uses 3D periodic boundary condition with a big vacuum between the layers and in some cases it's not good enough even with a very large vacuum because you have long range Coulomb interaction and the periodic images interact. So fortunately there is a 2D version of the DFT code quantum espresso now with the truncated coulomb interaction so this has been developed by Thibault-Soyer. And this allows for instance to get the proper LO-TO splitting or I should say maybe the absence of splitting because so that's boron nitride that's the high part of the phonon dispersion and in 3D you have really a splitting between LO and TO while in 2D the LO mode actually bends down until it reaches the TO and the only thing that is discontinuous is the slope actually. So with this code you can get this right which is quite important. So now how do you deal with the stability? So the idea is to relax the 2D structure to compute the force constant and then to look at possible unstable phonon at gamma so meaning imaginary frequencies and if there is one you try to distort the structure to follow the instability and you use this as an initial guess and you go back here and you do it again. So we use the speedily the library from Atsuhi Togo to perform this kind of things and the symmetry analysis and this allows to go from a situation like this where you have really a highly negative phonon at gamma to something like this where this is well stabilized. So in the end we have 245 phonons actually now we are computing more and let me now make a small parenthesis because up to now I talked about results which is the thing that happens in the project at the very end actually and before this end there is all the technical stuff you have to do and let me say a word about it so all of this has been I would say made possible thanks to the platform AIDA which you have heard about which is one of the flagship code and AIDA is about really dealing with all the calculations you want to put on a cluster that you want to launch on a cluster and most importantly also storing all the data generated by this calculation storing the provenance of this data they are linked together so you have a structure that gives a calculation that would give another structure for instance so you have in the end some graphs, some trees it's also an environment where you have really connection between different kinds of codes for instance ASC, SPG, PIMAGEN you can all use them within AIDA and maybe most importantly the sharing so all that we do in AIDA the workflows we develop, the data we generate can then be easily shared with other people using also AIDA so this is really the key tool in this study and in Marvel at large I would say so let me show you an example of workflow that is the following dispersion workflow so it's a workflow to begin from a structure compute the dynamical matrices and in the end get the following dispersion and in this actually you have lots of sub workflows so the structure relaxation first you need to go through several relaxation until you're sure you have reached a convergence the dynamical matrices you have to do some initialization then you compute for each queue point you compute a separate you use the pH code of quantum espresso and actually all these are also workflows that are really low level workflows that deals with all the restart management on a cluster the fact that maybe your calculation can fail for some external reason and just need to relaunch it all there is the CPU time that was reached and you need just to continue it this kind of thing and even some small logic to change a little bit the parameters to make it work so this is a quite complex workflow but then for the user it's simple to use actually recently with Giovanni Pizzi we have developed a fun and turnkey solution which I can show I hope you can see something but anyway so essentially you choose a structure and you choose if you want to relax it you choose the computer in which you want to compute and so this is really a small hub that you can readily use you choose the pseudo potential library you choose some k-points density mesh there is also some advanced options like the smearing or this kind of thing and then you generate your input and just have to click on a button to run the workflow so after that you'll go to some status app that will show you what has happened so now I skip some of the waiting part and now it's finished and you see the result right away also in the same application so there is the band structure and also the phone on dispersion so now if I tell you that this is so easy to use that a kid could use it maybe you would not believe me or maybe you would say this is just a nice sentence it's actually true because what you've just saw has been performed by ten years old daughter she did it on the computer without even let's say the closest thing from a computer she used to use was a smartphone and she did it very easily so she computed the phone on dispersion of aluminum that way so let me go back a bit to the properties and go to the perspectives of this study so after that we computed the magnetic and electronic properties by scanning possible ground states so both ferro and antiferromagnetic configurations and this is a plot of the absolute magnetization in Bohr magneton per unit cell versus bandgap and those are the materials that we found so I indicated the strongest ones all those that are insulating which are very interesting the red are the antiferromagnets blue are the ferromagnets you see also some chemical trends like the manganese compound the vanadium compounds or the chromium compounds here around also the nickel here so all this is going to be actually published in one week from now on February 5th in this Nature and Technology paper with all the co-authors and also the electronic structures so this is a just 258 electronic band structure those are effective masses the color indicates the thickness of the layer and the size indicates the bandgap that's the work of Davide Campi who is preparing a paper on this and we also searched for topological insulators and actually we found one actually Antimo Marazzo found one and there is this Sarkov paper here so it's based on the recruiting guide which is actually known as everything in this study it's a known 3D bulk compound that has been discovered in Brazil so before concluding let me show you all my co-authors and really I would like to thank all of them because what they've done was really crucial I would also like to thank the founding partners and the computing time from CSCS and now I will leave you with my summary so we found essentially more than 5,000 layers of material with a geometric screening then out of them almost 2,000 easily exfoliable structures not easily but exfoliable structure among which 1,000 should really be easily exfoliable we did all this with AIDA which keeps the provenance of all calculations easily shareable now we are computing phonons, magnetic, electronic topological properties and all this data is going to be in the materials cloud is actually now available in the materials cloud and in this DOI here so as I said there is a nature and technology in present it is going to be out next week thanks a lot for your attention