# Roman Krems

#### Relevant Thesis-Based Degree Programs

#### Affiliations to Research Centres, Institutes & Clusters

## Graduate Student Supervision

##### Doctoral Student Supervision

Dissertations completed in 2010 or later are listed below. Please note that there is a 6-12 month delay to add the latest dissertations.

This thesis investigates applications of classical machine learning to quantum problems and the possibilities of combining machine learning and quantum computing to improve algorithms for solving quantum problems. In quantum physics and quantum chemistry, the high dimensionality of quantum problems poses a significant challenge. Due to the increased complexity of such problems, traditional algorithms may not solve them effectively. However, new insights and better computational methods have become possible with the development of machine learning methods. This thesis aims to develop new methods based on classical and quantum machine learning methods applied to quantum problems. The first part of the thesis shows how Bayesian machine learning can be applied to quantum research when the number of calculations is limited. To be more specific, I construct accurate global potential energy surfaces for polyatomic systems by using a small number of energy points and demonstrate methods to improve the accuracy of quantum dynamics approximations with few exact results. The second part of the thesis looks into combining machine learning and quantum computing to improve machine learning algorithms. I demonstrate the first practical application of quantum regression models and use the resulting models to produce accurate global potential energy surfaces for polyatomic molecules. Furthermore, I illustrate the effect of qubit entanglement for the resulting models. In addition, I propose a quantum-enhanced feature mapping algorithm that is proven to have a quantum advantage for specific classically unsolvable classification problems and is more computationally efficient than previous methods. Finally, I highlight the potential for combining machine learning and quantum computing to improve algorithms for solving quantum problems.

View record

This thesis illustrates the use of machine learning algorithms and exact numerical methods to study quantum observables for different systems. The first part of this thesis depicts how to construct accurate potential energy surfaces (PESs) using supervised learning algorithms such as Gaussian Process (GP) regression. PESs have a leading part in quantum chemistry since they are used to study chemical reaction dynamics. Constructing the PES from quantum reactive scattering calculations, as the reaction probability, is known as the inverse scattering problem. Here, we illustrate a possible solution to the inverse scattering problem with a two-tiered GP model one GP model interpolates the PES and the second in Bayesian optimization (BO) algorithm. The end result is an accurate PES constructed with a GP with fewer points than with standard methods previously used for PES. BO is an optimization algorithm for black-box functions that use GP regression as an approximation of the interrogative function. We applied BO to find the optimal parameters of hybrid-density functionals. Quantum observables can differ between phases of matter. GP models with kernel combinations can extrapolate quantum observables such as the polaron dispersion energy between different phases and discover phases of matter. The same algorithm can predict quantum observables where standard numerical techniques lack convergence.In the second half of the dissertation, we studied the evolution of quantum walks in various graphs with Hamiltonians permitting particle number changes. We showed that particle number-changing interactions accelerate quantum walks for any of the graph considered. Quantum simulators to study many-body physics is an active research field. We proposed the use of magnetic atoms trapped in optical lattices to experimentally mimic Bose-Hubbard type models by preparing atoms in different Zeeman states.

View record

This thesis considers (1) novel manifestations and applications of quantum interference in complex systems and (2) development of new approaches to study complex quantum systems.First, I examine a general model of particles with long-range hopping amplitudes. For at least 30 years, it has been widely accepted that these particles do not undergo Anderson localization in 3D lattices. We show that these particles do undergo Anderson localization in 3D lattices if their hopping amplitudes are isotropic. In contrast, particles with anisotropic long-range hopping appear to follow the widely held expectations. We show these results by demonstrating that cooperative shielding extends to 3D cubic lattices with isotropic long-range hopping, but not with anisotropic long-range hopping and by computing the scaling behaviour of participation ratios and energy level statistics.Secondly, I develop a fully quantum mechanical model of molecular surface spin-echo experiments, which study surface properties and dynamics by scattering molecules off the sample surface. This model, based on the transfer matrix method, incorporates molecular hyperfine degrees of freedom, allows for the efficient calculation of the experimental signal given a molecule-surface scattering matrix, and permits us to begin addressing the inverse scattering problem. This fully quantum model is required to properly describe these experiments as the semi-classical methods used to describe experiments using helium-3 atoms do not take magnetic-field induced momentum changes into account. We apply our method to the case of ortho-hydrogen and then apply Bayesian optimization to determine the molecule-surface scattering matrix elements from a calculated signal, for a scattering matrix defined by three parameters. Our work sets the stage for Bayesian optimization to solve the inverse scattering problem for these experiments.Finally, I propose using Bayesian model calibration to improve the convergence of Monte Carlo calculations in regions where the sign problem or critical slowing down are an issue. Specifically, Bayesian model calibration would correct poorly converged Monte Carlo calculations with the information from a small number of well-converged Monte Carlo calculations. As a simple proof of principle demonstration, we apply Bayesian model calibration to a diagrammatic Monte Carlo calculation of the scattering length of a spherical potential barrier.

View record

In this thesis, I investigate the behavior of particles dressed by quantum field excitations and random interactions.First I consider two-carrier states in the Peierls model describing the modulation of the particle hopping due to lattice distortions. I compute the spectral response using the Momentum Average approximation. Combining accurate numerical techniques and analytical arguments, I provide a complete picture of the Peierls bipolarons. It is found that polarons bind into strongly bound yet light bipolarons in the singlet sector, even at large values of the electron-phonon coupling strength. At finite electron fillings, these bipolarons may condense into a high-Tc superconductor. On the other hand, phonons mediate a repulsive interaction in the triplet sector, or equivalently (in one dimension), between two hard-core particles, in which case the ground-state dimers bound by sufficiently attractive bare interactions exhibit two sharp transitions, one of which is the first known example of a self-trapping transition at the two-carrier level. In both situations, phonons mediate pair-hopping effective interactions between the carriers. I further study some aspects of the excited spectrum for the two hard-core particles, a situation relevant to ultracold quantum simulators. It is found that the repulsive phonon-mediated interaction binds a repulsive bipolaron embedded in the excited spectrum.I then turn to the study of quenched randomness in an ultracold molecular plasma. I argue that the quenched ultracold plasma presents an experimental platform for studying quantum many-body physics of disordered systems in the long-time and finite energy-density limits. I analyze an experiment that quenches a plasma of nitric oxide to an ultracold system of Rydberg molecules, ions and electrons that exhibits a long-lived state of arrested relaxation. The qualitative features of this state fail to conform with classical models. I develop a microscopic quantum description for the arrested phase based on an effective many-body spin Hamiltonian that includes both dipole-dipole and van der Waals interactions. This effective model appears to offer a way to envision the essential quantum disordered non-equilibrium physics of this system.This thesis thus examines the quantum many-body response in interacting systems coupled to bosonic fields or in disordered environments.

View record

Two interacting particles in lattices, in the absence of dissipation, can not distin- guish between attractive or repulsive interaction when the range of their tunnelling is limited to nearest neighbor sites. However, we find that, in the case of long-range tunnelling, the particles exhibit different dynamics for different types of interactions of the same strength. The nature of dynamical correlations between particles also becomes significantly different. For weak interactions, particles develop a character in correlation which is in between that of antiwalking and cowalking when the tunnelling is long-range. For strong interactions, particles cowalk independently of their statistics. A few recent experiments have demonstrated such effects of interactions on quantum walk of photons, atoms and spin excitations on various lattice platforms.In disordered lattices the effect of coherent backscattering makes particles localize to their initial position. We find that a weak repulsive interaction reduces localization and a strong interaction enhances localization. We also calculate the correlations between the particles in the disordered 1D and 2D systems. The effect of long-range tunnelling on localization of particles in disordered 1D systems has been explored.For large ordered or disordered lattices, computation of localization parameters becomes difficult. In these cases, an efficient recursive algorithm is used to calculate Green’s functions exactly. We extend such algorithm to disordered systems in both one and two dimensions. We also illustrate that this recursive algorithm maps directly to some graph structures like binary trees. We perform calculations for quantum walk of interacting particles on such graphs. The method is also used to calculate the properties of interacting particles on lattices with gauge fields. For disordered 2D lattices, we introduce and test approximations which produce accurate results and make the calculations more efficient. We examine the localization parameters for a broad range of interaction and disorder strengths and try to find differences among parameters within the range.

View record

This thesis describes four types of cold molecular collision systems with increasing complexity: from simple atom-diatomic molecule to complicated polyatomic molecule-polyatomic molecule. The early thesis work is concerned with three open questions pertaining to collision dynamics of cold molecules. We demonstrate the possibility of controlling collisional decoherence of ultracold molecules by tuning an external magnetic field. We then provide insight into the feasibility of evaporative cooling of molecules, and for the first time incorporate the uncertainty analysis of the potential energy surface (PES) into scattering calculations. In addition, we use classical trajectory methods to study the effects of the interaction strength and the geometry of rigid polyatomic molecules on the formation of long-lived collision complexes at low collision energies. The second half of the thesis work is focused on combining statistical methodology and scattering calculations to address two major problems in molecular dynamics calculations: increasing computational complexity and uncertainties due to inaccuracies of PES. Using a small number of scattering calculations, we show that we can build a Gaussian Process (GP) model to statistically approximate collision outcomes for complex molecules, and then perform the uncertainty analysis and the sensitivity analysis. We also demonstrate that trained by a combination of classical and quantum calculations, a GP model can provide an accurate description of the quantum scattering cross sections, even near scattering resonances.

View record

This thesis describes research on controlling the dynamics of quasiparticles in periodic and disordered lattice potentials. Working with model systems of arrays of atoms and molecules trapped in optical lattices, I focus on, but not limited to, the rotational excitons of polar molecules and propose to use external fields to control the binding and propagation of quasiparticles. First, we study the binding of rotational excitons in a periodic potential. We show that non-linear interactions of such excitons can be controlled by an electric field. The exciton-exciton interactions can be tuned to induce exciton pairing, leading to the formation of biexcitons and three-body bound states of excitons. In addition, we propose a non-optical way to create biexcitons by splitting a high-energy exciton into two low-energy excitons. Second, we present schemes to control the propagation of a collective excited state in ordered and disordered aggregates of coupled particles. We demonstrate that the dynamics of these excitations can be controlled by applying a transient external potential which modifies the phase of the quantum states of the individual particles. The method is based on an interplay of adiabatic and sudden time scales in the quantum evolution of the many-body states. We show that specific phase transformations can be used to accelerate or decelerate quantum energy transfer and spatially focus delocalized excitations onto different parts ofarrays of quantum particles. For the model systems of atoms and molecules trapped in an optical lattice, we consider possible experimental implementations of the proposed technique and study the effect of disorder, due to the presence of impurities, on its fidelity. We further show that the proposed technique can allow control of energy transfer in completely disordered systems. Finally, in an effort to refine the theoretical tools to study dynamics of quasiparticles, I extend calculations of lattice Green's functions to disordered systems. We develop a generic algorithm that can be easily adapted to systems with long-range interactions and high dimensionalities. As an application of the method, we propose to use the Green's function to study the tunneling of biexciton states through impurities.

View record

The research ﬁeld of cold and ultracold atoms and molecules is rapidly growing and expanding into different areas of research such as quantum information science and condensed matter physics. The success of this ﬁeld is due to the possibility of precisely controlling and manipulating atoms and molecules at low temperatures. The progress in this ﬁeld relies on the development of new methods for controlling collisional dynamics and interactions of particles with electromagnetic ﬁelds. This Thesis describes research on modiﬁcation of the collisional dynamics of ultracold atoms and molecules by external laser and microwave ﬁelds as well as new methods for the detection of weak electromagnetic ﬁelds. First, we study the scattering of atoms and molecules conﬁned in a 2D geometry by optical lattices. In particular we develop a theory for scattering in 2D and derive the equations for the threshold dependence of the collision cross sections. We show that inelastic processes and chemical reactions can be suppressed under strong conﬁnement in one dimension and can be controlled by varying the orientation of the external ﬁeld with respect to the plane of conﬁnement. Next, we present a rigorous theory of low-temperature molecular collisions in the presence of a microwave ﬁeld. The microwave ﬁeld can theoretically be used to trap and control polar molecules. The molecular collisions may lead to trap loss and decoherence. We develop a rigorous quantum theory for molecular scattering in the presence of microwave ﬁelds. We study inelastic, spin-changing molecular collisions and Feshbach resonances in the presence of microwave ﬁelds. We demonstrate that inelastic collisions accompanied by absorption of microwave photons can be signiﬁcant. The detection of weak electromagnetic ﬁelds is very important for various applications ranging from fundamental measurements to biomagnetic imaging, and for tests of microwave chips. We present a method for the detection of weak electromagnetic ﬁelds in a wide range of frequencies from sub-kHz to THz with ultracold polar molecules. We show that using ultracold molecules one can achieve the sensitivity of two orders of magnitude larger than with a similar method based on ultracold Rb atoms.

View record

Ultracold molecules are expected to find applications in cold chemistry, quantum phases, precision measurements and quantum information. In this thesis three novel applications of cold molecules are studied.First the thesis presents a general method for coherent control of collisions between non-identical particles. It is shown that by preparing two alkali-metal atoms in a superposition of hyperfine states, the elastic-to-inelastic cross section ratio can be manipulated at ultracold temperatures by tuning laser parameters in the presence of a magnetic field. The static field is needed to induce quantum interference between scattering states. Extensions of this scheme for ultracold molecular reactive scattering are discussed.Second, the thesis describes rotational excitons and polarons in molecular ensembles trapped in optical lattices. Rotational excitons can be manipulated using static electric and magnetic fields. For a one-dimensional molecular array with substitutional impurities any localized exciton state can be delocalized by applying a suitable electric field. The electric field induces correlations between diagonal and off-diagonal disorder. It is also shown that the translational motion of polar molecules in an optical lattice can lead to phonons. The lattice dynamics and the phonon spectrum depend on the strength and orientation of a static electric field. An array of polar molecules in an optical lattice can be described by generalized polaron model with tunable parameters including diagonal and off-diagonal exciton-phonon interactions. It is shown that in a strong electric field the system is described by a generalized Holstein model, and at weak electric fields by the Su-Schrieffer-Heeger (SSH) model. The possibility of observing a sharp polaron transition in the SSH model using polar alkali-metal dimers is discussed.Finally, the thesis presents a method to generate entanglement of polar molecules using strong off-resonant laser pulses. Bipartite entanglement between alkali-metal dimers separated by hundreds of nanometers can be generated. Maximally entangled states can be prepared by tuning the pulse intensity and duration. A scheme is proposed to observe the violation of Bell’s inequality based on molecular orientation correlation measurements. It is shown that using a combination of microwave and off-resonant optical pulses, arbitrary tripartite and many-particle states can be prepared.

View record

This Thesis describes new mechanisms for controlling elastic and inelastic collisions of ultracold atoms and molecules with static electromagnetic and laser fields. The dynamical properties of ultracold atoms are usually tuned in experiments by applying an external magnetic field to induce a Feshbach resonance. The work presented in this Thesis demonstrates the possibility of inducing and manipulating Feshbach resonances with electric fields. We discuss in detail the mechanisms of electric-field-induced resonances in ultracold mixtures of alkali metal atoms and demonstrate that electric fields may shift and split the magnetic resonances. We show that electric fields may spin up the collision complex of ultracold atoms and induce anisotropic scattering which may be exploited in experiments on many-body dynamics of ultracold gaseous mixtures. The mechanisms of electric-field-induced resonances described in this Thesis allow for two-dimensional control of inter-particle interactions, leading to total control over ultracold gases. To guide future experiments, we generate accurate interaction potentials for ultracold Li--Rb mixtures by fitting positions and widths of experimentally measured Feshbach resonances. Ultracold atomic and molecular gases can be confined by laser fields in one or two dimensions which produces an optical lattice of ultracold particles.We develop a multichannel scattering theory for collisions of atoms and molecules in two dimensions and explore the effects of the confining laser potential on inelastic and reactive collisions of ultracold atoms and molecules in a 1D optical lattice. We show that ultracold collisions can be controlled in a quasi-2D geometry by varying the orientation of a magnetic field with respect to the confinement plane normal and demonstrate that the threshold energy dependence of cross sections for inelastic collisions in an optical lattice can be tuned by varying the confining potential and the magnetic field. Our results show that applying laser confinement in one dimension may stabilize ultracold systems with large scattering lengths, which may open up interesting opportunities for studies of ultracold controlled chemistry and might lead to a new research direction of ultracold chemistry in restricted geometries.

View record

##### Master's Student Supervision

Theses completed in 2010 or later are listed below. Please note that there is a 6-12 month delay to add the latest theses.

Quantum Diffraction Universality (QDU) is a law that allows the experimental determination of the thermalized rate coefficient ⟨σₜₒₜv⟩ between an impinging gas and a nearly stationary sensor gas in high vacuum. QDU relies on the insensitivity of the rate coefficient to the short-range interaction potential for the collision partners. This law allows one to bypass time-intensive theoretical scattering calculations for ⟨σₜₒₜv⟩ and additionally leads to the determination of the impinging gas pressure. In this thesis, I describe two projects that further our understanding of QDU. First, I conduct coupled-channel quantum scattering calculations for the collision partners Li+H₂ and Rb+H₂. These calculations for ⟨σₜₒₜv⟩, in combination with experimental measurements, show that Rb+H₂ is a system that cannot be described by QDU. The reason is that Rb+H₂ has a light reduced mass and small C6 coefficient (which characterizes the long-range interaction potential). For these reasons, one can infer that Li+H₂ also deviates from Universality. In the second project, I modify the shortrange interaction potential of Rb+H₂ and analyze how the modifications lead to a change in ⟨σₜₒₜv⟩. Furthermore, I describe how machine learning – specifically Gaussian Process Regression – can be utilized to predict ⟨σₜₒₜv⟩ for different modulations of the short-range interaction potential. This analysis will give an estimate on the error arising from interaction potential uncertainty associated with ⟨σₜₒₜv⟩ for Rb+H₂. Additionally, it serves as a second demonstration of the non-Universal behaviour of the Rb+H₂ system.

View record

This thesis explores the interplay of machine learning and molecular physics, demonstrating how developments in one field lead to advances in the other. We show this using two examples. First, we illustrate how improvements made to a machine learning model of molecular PES and FF can significantly increase its generalization accuracy. Gradient domain machine learning (GDML) models produce accurate results for high-dimensional molecular systems with a small number of ab initio calculations. The present work extends GDML to models with composite kernels built to maximize inference from a small number of molecular geometries. We illustrate that GDML models can be improved by increasing the complexity of underlying kernels through a greedy search algorithm using Bayesian information criterion as the model selection metric. We show that this requires including anisotropy into kernel functions and produces models with significantly smaller generalization errors. The results are presented for ethanol, uracil, malonaldehyde and aspirin. For aspirin, the model with composite kernels trained by forces at 1000 randomly sampled molecular geometries produces a global 57-dimensional PES with the mean absolute error 0.177 kcal/mol (61.9 cm⁻¹) and FFs with the mean absolute error 0.457 kcal/mol Å⁻¹. Second, we propose a procedure to perform quantum computation in the form of quantum annealing using a crossing between Zeeman states of different parity in the rotational and fine structure of open-shell molecules in a Σ electronic state. These crossings become avoided in the presence of an electric field. We propose an algorithm that encodes Ising models into qubits defined by pairs of ²Σ molecules sharing an excitation near these avoided crossings. This can be used to realize a transverse field Ising model tunable by an external electric or magnetic field, suitable for quantum annealing applications. We perform dynamical calculations for several examples with 1D and 2D connectivities. Our results demonstrate that the probability of obtaining valid annealing solutions is high and can be optimized by varying the annealing times.

View record

If this is your researcher profile you can log in to the Faculty & Staff portal to update your details and provide recruitment preferences.