uh okay good afternoon everyone thanks for coming back we have a couple of very interesting talks so please Alexander hey um everyone yeah thank you so much for being here today this afternoon at the end of the long week I was expecting the room to be empty and I'm sure many of you could have just gone for a long weekend and enjoyed the sun I'd like to thank organizer for bringing us together and also giving me this opportunity to uh present today I think it's uh we're just discussing about opportunity and challenges right bringing different people from different community and figure out ways how we can work together and um we were asked to somewhat introduce you to our fill so that there's a kind of common understanding of what we're doing and then uh more or less left the technical details to the end so of course there's a lot of ideas that will be introduced and if any of those are interesting to you please make sure to drop an email we're located at University of wallu which is about an hour and a half drive away from here so if you ever want to visit you should always feel free to welcome especially at Toronto students so we can organize group tours and I know uh cqi QC and similar Department of physics at University of Toronto I think there could be a significantly more synergies between iqc and Toronto and i' be willing to to help us get there and um on that um yeah let me start my name is Alex cery I'm in the department of physic and astronomy as well as the iqc at the University of watero and today I'm going to tell you about Quantum control and Quantum information with redberg Adam arrays so uh just quickly introducing you about our platform here's what a redberg Adam array Quantum simulator looks like it's a it's a room siiz scientific apparatus and uh thanks to my student they've taken those very nice shots and uh this is a platform that came online about the last five seven years it's been a tremendous Advance on the experimental capability side and I'll I'll describe some of that and um currently we're working with ridium historically ridium was some of the Workhorse of atomic physics and over the years we've learned how to control other alkaline atoms as well as other alkaline earth atoms and now the field is already at a point where we work with mixed species um either bringing different alkal atoms or Alkali and alkaline earth atoms and uh there's tremendous uh growth happening in the field right now uh yet working with ruidum system even though you could argue more classical um plain vanilla type system there's still plenty of physics that can be done uh so it's not always true that more complex brings more uh complexity in physics so our approach here was like still keep it simple and now you see that well it's a lot of optic still in absolute a very small system as compared to a BC type experiment um maybe a decade ago um now just quickly I'll share some few features about our platform um what we do is effectively we have the ability to cool trap in image individual atoms and what look like in practice is we have a source of redium uh we bring those rubidium into this glass cell here and we build a microscope around this glass cell so we can create a cloud of millions of atoms and we can deliver light tightly focused laser light into that cloud of atoms to produce this Optical trap so effectively you can think just a very high intensity laser beam and into that laser beam and atoms will be attract actually individual atoms and we can image those on a sensitive camera and typically it look like this if you uh if you see an atom uh you will you will see something that look like that if you have no atoms you will see just something blue and we can uh discriminate between the presence in absence of an atom so we have sight seleced readout of individual atoms and something nice is we actually have the ability to Multiplex that system so you take the laser bin Multiplex in one dimension to create this ond chain of atoms uh and typically what you see in the lab is something that look like this so some traps will be occupied boms you'll see a a bright spot and some traps will not be occupied which will be effectively a defect and this is a problem so historically we would do um emble average and uh your initial State change from shot to shot which is a problem uh now luckily thanks to Quantum control we have the ability to correct for those defect so here's where it looks in in practice we have the ability to prepare defect configuration of atoms so we take an image identify where are the atoms where are no atoms and given what we want we go about moving the atom in real space and um uh we actually work really really hard on solving that problem so this problem of um preparing a defect free compact configuration of atom in a ond chain is a is a problem that can be solved exactly and the algorithm is actually linear and we have the source score available on GitHub if that's interesting to you to your colleague please make sure to you have a look um and something we've worked really hard is in trying to say okay uh and something important for you theorist in the room is having an efficient algorithm is not sufficient you need an efficient implementation of that algorithm and typically try to work really really hard so that it can run really fast in the experiment and uh here's typical experiment we've worked really hard and we can actually solve this problem in few tens of microsc really pushing state-of-the-art um we thought that we might gain an advantage with gpus and here's actually something else that I've learned is uh even though in theory you would expect greater um capability with gpus it typically take time to initiate the core when solving problems and here uh what we see is that for the problem size that we're dealing with uh handling CPUs is actually better uh We've also worked on the problem of solving reconfiguration problems in 2D U and now this is the typical problems in Quantum control so when you operate those Quantum simulators of course you like to think about those implementing Quantum algorithm but the simple fact of operating the system also involves serving a lot of hard classical optimization problem and here's one example um now depending on which objection function you set the problem might either be easy or hard so if you ask um find the uh minimize the total distance displaced by all atoms it will typically be a problem you can solve with the Hungarian algorithm in N to the cube complexity however if you seek to minimize the number of atom displaced uh it will typically be an NP hard problem and we actually work with graphist uh to solve that problem and uh interestingly graphist uh I've been working on reconfiguration for years so many of you when you were kids might have played this game you have a a board from 1 to 15 in a square of 16 and then you Shuffle the number around and you say okay how can I go back to original configuration that's a typical reconfiguration problem and those graph theorists when they write their paper they said oh here's our Theory typically represented by those game played by kids uh we came to them and said look we have a new problems and their eyes were open up oh thank uh finally a Newfield and actually it led to the establishment of Newfield uh and has been very productive collaboration working with these U these scientists U and uh once again we've worked really hard on implementing those on both CPU and GPU and turn out that for the problem size that we're dealing with once again the CPU provide greater advantages uh now a third feature is not only can we create Cham where we can also scale up to larger Dimension so going from 1D to 2D here you see a grade of 10,000 trap and it's about uh state-of-the-art so now I'm not claiming that each of those trap will trap individual atoms we timately limited by how much power we can bring to a system and it's just a limitation of the technology we can buy off the shelf uh state-of-the-art is about trapping up to 6,000 atoms uh and the nice thing is we're not only limited to uh lates but we can also do arbitrary uh geometry so for example here is a honeycom luses if you're interested in simulating honeycom K model or realizing Quantum spin liquids Etc so there's a lot of flexibility and of course you could go about introducing spatial disorder you could work on sublates or arbitrary geometry whatever you can think of um yes so uh fourth features is that uh we can actually realize a lot of spin models with large red redberg interactions I've showed you those nice 2D configuration each trap can have an atom or not but at PRI those atoms are in their hyperfine ground state they're Micron apart from one another and uh they don't really care about one another they don't see one another they don't interact so bring interaction we actually excite those atom to redber State this is a principal quantum number large n on the order of 30 up to 70 and these redberg atoms have very large electric dipole moment which indu strong interaction uh and the way you can think about that is if you have two spins uh two uh two Lev system uh and you look at the The Joint bases you have these four state ground ground gr redberg redberg ground and redberg redberg and you can just WR the symmetric and anti- symmetric State and what happen is if you have two atoms in the red brick State uh you experience an energy shift on the order of v and what is unique as compared to S State powerform this V this interaction strength between two ROM is typically much much greater than typical um control scale like Omega the RAB frequency so effectively what happen is if you were to undergo rabi oscillation you were to try Drive the system um the system start on the ground introduce an excit patients uh but effectively because this the tuning is much much smaller than the rabi frequency uh this uh transition is forbidden and this is what is called the red bird blockade so effectively what you'll have is that the evolution will be constrained to the Subspace here and you will see an enhancement into the Rabid frequency of the square root of two and whether or not you see that enhancement will depend on the distance right so if the atoms are really close to one another the energyc is very very large and the blockade is very significant and as you bring the atoms towards Infinity now the atoms stop interacting one another and they act as two independent level system right so that's why we typically say they're large interaction but programmable and they prog mod because we can uh tune the interatomic distance uh and typically this is the mechanism that is used for engineering two Cubit gate as well as to realize L pin models so both for Quantum Computing Quantum uh Quantum simulation uh so now typically what is realized is a transverse realizing model so uh you can just think of that as a spin and the Pres of an external magnetic field and when uh two spins are in the same direction they experience an energy penalty V uh and now here we have this driving term and you see here that I right this driving term is sight selective and uh how we go about doing that in the experiment is uh typically that's what we're currently building on is on the one hand we have a reservoir region and on the other hand we have a control region where we put the atoms on which we want to perform Quantum gate and then we uh can apply these uh sight selective control PSE so effectively we use spatial light modulator to create these flat top beams which allow us to very precisely engineered two Cubit gate so uh this is kind of a novel approach right historically you would say okay I have my register here this is where I start my tubit and if I want to perform a control operation I'll shine a laser line onto it um but that's a problem right because in the process doing so you might corrupt Quantum information and code it into neighbor so now we have effectively like what was done his with trapon the ability to move the atom in space while maintaining the cerence encoded in their internal degrees of freedom and therefore we can uh just store our Quantum information here and only move it to that region whenever we want to perform a Quantum operation it's a is a quite uh quite um quite novel and impactful now uh the challenge somehow is in term of quantum control is realizing Quantum control at scale for useful qip applications right so over um if you look at the history of the field of quantum control we've start like having the ability to uh learn about Quantum systems if you think about solar State NMR bulk system we do spectroscopy we'll learn about their Dynamics in the field of Co atoms we have create BCS uh bulk system and then gradually we learn um for example with the work of ser rash uh can we study a single particles and its its uh its physics and now we're going backwards bottom up assembling large Quantum systems and now the challenge is how can we use the control techniques that we demonstrated for a few Cubit and make them applicable at Scales right and we've heard this morning the kind of challenge that involve a lot of engineering um and now of course you could say okay let make it the goal let me create a largest minibody and Tangle system ever created by mankind uh and maybe some of us will gain pleasure in that but I think there's huge demand especially from the Canadian government um and uh our community to demonstrate useful qip appli ations uh and I guess that's what this conference about Quantum control and Quantum information processing uh so uh just to move along in that direction I'd like to uh quickly share with you how I think about Quantum control in the current context of our experiment so by definition Quantum control enables Precision precise manipulation of quantum system so you typically start with a Quantum system or a collection of quantum system identical Quantum system and these could be either a collection and temporal emble or a spatial Ensemble of identical quum system described by hamiltonian and its state and we have a classical controller applying some control field C oft and now we have a this time dependent hamiltonian and typically uh it's convenient to assume that this time independent hamiltonian generat some uh time average Evolution for example using aonian theory of second order perturbation Theory and this anter generate the evolution effectively uh time evolving this um this state we have a measurement apparatus on which we go about performing projective measurement onto the system which leads to these bit strings information about the system and we typically feed this B string to a classical problem processor which is now solving somewhat of a learning and feedback problem right it's learning about the system and uh I think at least it's meaningful and I know that many of you are doing that and I saw some poster of thinking of a Quantum control as a Quantum feedback control so saying that no do we want to learn about the system or we want to use that information in a meaning meaningful way either to bring the system towards a desirable state or stabilize it against uh fluctuations um and now typically what that uh feedback control system does it extract information update the state of knowledge if you think in the basian framework and then some compute the feedback so how should I react um to to a change in the environment and go about feedback classical controller right so this is kind of a maybe uh kind of classic Quantum feedback control now the feel is moving towards right so over the next DEC what we're going to work on is not so much this representation but saying I have a Quantum system a Quantum simulator you have a Quantum sensor how can I use my Quantum uh simulator as a controller of a Quantum system where now the controller itself is quantum exchanging Quantum information I think for this context we can think okay classical U controller updating activating a a Quantum system and now there's of course a lot of things that can go wrong it can be control errors nonary Evolution du to coupling to the environment approximation errors when the hilon the effective hamiltonian that you generate is not exactly what you hope it was there temporal drift and fluctuation spatial inhomogeneity measurement apparatus might be noisy and on the classical side you might just have not enough information to reconstruct a full state of knowledge limited resource in Computing or memory uh hidden models you might not know exactly what your system is and uh typically hard optimization as we've seen uh and more importantly latency you need to be able to process and feedback fast um and more or less this is what the field is about right pick one star here and go about solving it and you've made a contribution and I think this unified us and uh for us we're actually indeed like tackling all these problems uh working in optimal control theory dynamical decoupling doing realtime stabilization perform low noise measurement doing efficient processing and decoding on gpus uh implementing hoton learning methods implementing efficient algorithm as well as their efficient implementation of low lensy feedback system so I'll just quickly highlight some of the work that we're doing in optimal control theory so this is the work of arim you might have met he was here the whole week he presented the poster he just raised his hand now um make sure to drop by uh so Aram has been working on the problem of synthesizing fast High Fidelity and robust tic control PS for red Adam arrays and this might sound somewhat uh counterintuitive because you would say okay well if I want a control PA I want adiabatic right so typically we know that adiabaticity allows to reach really high fidelity but we wouldn't really think of a diabetic as fast right uh we might actually think of a diabetic as robust but not necessarily as fast and here what I mean is we actually play some trick right so if you have the a system and you want to go adiabatically from one state to the other imagine you have an anti- level crossing you'd like to in ground set at all time now you know that as you cross the Gap you need to slow down right as a time scale on the order of one over the energy gap and as the energy go down you need to go exponentially uh slow uh but here you can play tricks to call like super diabetic protocol where you allow the system to be excited and then you play tricks with the fiz of your pulse such as to bring back the information at the end into the ground state so here we kind of play those tricks and what we see for example for a single Cubit gate uh the problem of exciting atom to Red B State we can actually do it with much greater Fidelity and much smaller pulse duration that we would with any other methods and this is allowed by um using optimal control theory we can also doing uh that for two Cubit gate and here we see like this is kind of the standard protocol for entangling two rber gate and we see that using optimal control theory and the vum formalism we can achieve better Fidelity and much faster post time so these are the kind of problem you can do offline ideally you'd like to do online solving this optimization problem in real time this is where gpus might help you U now the nice thing about using adiabetic pulse is that they embed robustness uh just naturally so if you look at uh the how that Fidelity respond with respect to a change in parameter for example a change in the tuning or a change in frequency you see that you're much less uh sensitive so um this is the kind of PS that we're developing and we're implementing in the lab and just like a tool in our toolbox uh and therefore if you're expert in optimal control uh and you want to help us scale that right now we want to do Quantum control scale how do I optimize optimal qu Theory to go across a phase transition involving hundreds of atoms and it becomes much more difficult because you cannot solve exactly that problem on your on your laptop another problem we've work on is efficient algorithm I've already talked about that another one is a low latency feedback system so a lot of those system out there are typically based on fpga which is great uh provide real-time uh control however fpga uh are not easily integratable with GPU and also they uh they tend to be hard to program right so uh good luck finding a student that know how to code a uh fpga if you find one good luck convincing him or her to uh build a program and then good luck maintaining capability when that student graduate so effectively you want to stay away from that and our approach we're just buying a gaming workstation which should have be very good and then just populating with different cards at will to build capability and this is kind of the work that was done with Co-op students here like Amir for example um apologies for the uh the pixelated version but effectively uh the problem is we go about taking images of those systems and based on the information that we extract we uh we go applying feedback control for example to move atoms in space uh but the nice thing is that this problem that I said that we can move space can be reused for example to realize adaptive protocols and this is relevant in sensing if you want to beat the standard Quantum limit uh or to do uh error decoding in a Quantum correction so some kind of a versatile tool and uh what we went about doing is just doing the best job and just tabulating all the numbers like and making sure that each step is doing as good as it can and ultimately now what we're Limited at is the hardware so it's just like how fast can we extract information out of a system is proportional to how many Photon can we scatter out of the atoms how many Photon can we collect and uh now luckily the technology of cameras has evolved and already the community has adop new camera technology uh the point I want to make here is that a lot of the quantum control requires not only the development of algorithm but also Hardware software and algorithm right and that is we need to keep updated about what is happening out there in electric engineering community because this is what allow leap in U in development capability right so uh here effectively this is a stateof the so we can do everything as fast as as as possible in a few tens of microsc But ultimately acquiring information and um playing that these control feedback is the limiting factor and once again like this whole thing is available on GitHub so if if you're interested in using it please feel free to uh to take headit so now in term of using these tool Quantum feedback control for application Quantum information processing you can think of different ideas and I think these are like the core ideas learning sensing Metrology Computing simulation and I've personally contribut to uh many of those fields over the years and I'll just quickly highlight so Shameless self-promotion um but in particular I'd like to emphasize Alex was here earlier this week on Wednesday present his work so actually Alex took over my work he's a third generation bad student and he's been doing amazing continuing pushing state-ofthe-art um and something that is interesting is now what we call hamiltonian learning I've been for a decade called spectroscopy and a few decades ago and I was a glass student with system uh identification and now it's h learning but very much like the tools are very much the same mod some new ideas effec we use these techniques to identify two new defect in diamond in particularly one defect which is a hydrogen related electronicle spin defect which is quite exciting because never been seen before for and we get to give it a name which is quite interesting and allow us to effectively scale up those solid States St defects instead of working with a single sensor in a diamond you can now actually think about scaling up uh but these system are really somewhat frustrating to work on is that you want to create an tangle state of three spin you need to work extremely hard and even then uh its fragility is not guaranteed thanks to its interaction to the environment and this is uh once again where you see the the advantages provided by redber Adam away because now we're not talking about a few but we're talking about hundreds of atoms uh in sensing we measure time dependent V magnetic field related to some of the work we heard earlier today with compress sensing effectively sensing both the terministic and stochastic field using um Walsh functions and we also demonstrated quance magnetometry using mangle state of spins um in Metrology we've shown a atom array Optical clock which is a new way of a PR Optical clock where you stabilize an external oscillator which is a laser to Atomic transition and uh the nice thing is that if you have an array each atom itself is a tiny clock and uh the nice thing is that even in in a ond chain which is 10 micr long this atom on the left tick at a different rate than this atom on the right so now you can start thinking about measuring gravity or doing gravitation measurement at the the Micron scale and uh it really pushed the field of um rology uh now Computing we've demonstrated High Fidelity to kbit gate uh this is work done at CCH when I was a post do and was just very recently beaten by the Princeton team and now what we do in watero is uh trying to advance the field of quantum simulation uh this is the team including artm and this is our system you've already seen it and us our website is live so please have a look tell us what you think um so for those of you that are not familiar just to set the stage Quantum simulation is the method by which an accessible quantum mechanical system is used to study the properties and dynamics of another quantum mechanical system right so think for example if I were interested in learning about graphine but I don't have the ability to synthesize graphine or I don't have the ability to bring that graphine flake into the parameter regime uh that I wish uh I might want to do calculation s on my laptop but if I get into the regime where I'm interested in correlated electron system or highly in tangle State my laptop will fail to do so and this is where Quantum simulation simulator becomes useful and U basically the idea is the following you have an inaccessible Quantum system undergo some unknown Quantum Evolution and effectively what you go about doing it is you map this model onto the DAT model realized by an accessible quum simulator something like a red gon array uh you let that system evolve using the tools of quantum control and then by performing measurements you can learn back about the Dynamics of the Desir system and this is kind of this idea have been around since finem Mon and we've heard from John and we're still being developed and now it's important to recognize and I think that's something that is not set enough we hear a lot of quantum simulation being done out there but typically the quantum simulation that is done is the Quantum simulation of native model right so I've said our system realize transvers sizing model if I study the physics of transverse iing model I haven't done simulation yes or no right I've just like studied the Dynamics of my own system so what would be more interesting if I could use my system transversal ising model ltis spin model to study let's say the Dynamics of interacting electron fironic system but if not this is a very very hard problem right because it requires doing non-trivial mapping effectively uh mapping fir on to Cubit and this is one a very much challenge in the FI of quantum simulation going Beyond native models um another challenge is not only Bridging the Gap between simulator and materials but then Bridging the Gap from materials at small scale towards real Quantum devices right towards applications so imagine once again you have a Quantum spin model you're interested in simulating magnetic materials let's say like like skons uh you might want to use those for spin tronic devices and now we're talking like larger scales and these ponic device could be used in energy efficient processor and to P that Gap we actually set up a company the company's called upscale Quantum solution and I've been out there trying to pitch the idea to investors and here's uh why it's very difficult because realizing that Vision using Quantum simulator to deliver application requires being successful in two different fields right one field is building a Quantum simulator or quantum computer that can solve a meaningful problem at scale and the second thing is uh making sure that whatever we learn from that simulator can inform on real application for example in the semiconductor industry um I think I'm running out of time I actually I have like another 30 minutes of slides but hopefully you had a good time um so I'll take one minute and I'll I'll flash through uh what I wanted to go through it is consider the problem I in 1D chain where you want to entangle one and L and uh I was proposing to study three different approaches the analog approach the digital approach as well as the dynamic approach where the dynamic approach involved moving the atom space and somewhat going about uh comparing which method is best and in which condition would you use one and then a lot of like slid showing you how we went about studying these these different things and um I think there's a lot going on uh but just in conclusion what I want to say is like a few of the challenges that remain realizing Quantum control at scale solving mapping problem for simulation Beyond native model as well as connecting with application and conents matter Material Science so I would like to you to write an email visit us and visit our website thank you very much thanks that was beautiful question