We present a robust scheme to derive effective models non-perturbatively for quantum lattice models when at least one degree of freedom is gapped. A combination of graph theory and the method of continuous unitary transformations (gCUTs) is shown to efficiently capture all zero-temperature fluctuations in a controlled spatial range. The gCUT can be used either for effective quasi-particle descriptions or for effective low-energy descriptions in case of infinitely degenerate subspaces. We illustrate the method for 1d and 2d lattice models yielding convincing results in the thermodynamic limit. We find that the recently discovered spin liquid in the Hubbard model on the honeycomb lattice lies outside the perturbative strong-coupling regime. Various extensions and perspectives of the gCUT are discussed.
We present a robust scheme to derive effective models non-perturbatively for quantum lattice models when at least one degree of freedom is gapped. A combination of graph theory and the method of continuous unitary transformations (gCUTs) is shown to efficiently capture all zero-temperature fluctuations in a controlled spatial range. The gCUT can be used either for effective quasi-particle descriptions or for effective low-energy descriptions in case of infinitely degenerate subspaces. We illustrate the method for 1d and 2d lattice models yielding convincing results in the thermodynamic limit. We find that the recently discovered spin liquid in the Hubbard model on the honeycomb lattice lies outside the perturbative strong-coupling regime. Various extensions and perspectives of the gCUT are discussed.
A fundamental principle in condensed matter physics is the effective description in terms of quasiparticles. The high-energy part of the system is accounted for by renormalized properties of the quasiparticles, providing both, an accurate description and interpretation of the relevant low-energy physics. We consider two classes of methods suited to provide effective quasiparticle descriptions in the thermodynamic limit: (non-perturbative) linked-cluster expansions ((N)LCEs) and continuous unitary transformations (CUTs). We focus specifically on the combination of both methods, providing a different perspective and solutions to existing challenges. LCEs provide high-order series expansions in a perturbation parameter, by combining effective Hamiltonians determined on finite clusters. We introduce a white-graph expansion for the method of perturbative continuous unitary transformations when implemented as an LCE. The essential idea behind an expansion in white graphs is to perform an optimized bookkeeping during the calculation by exploiting the model-independent effective Hamiltonian in second quantization and the associated inherent cluster additivity. This approach is shown to be especially well suited for microscopic models with many coupling constants, since the total number of relevant graphs is drastically reduced. The white-graph expansion is exemplified for a two-dimensional quantum spin model, illustrating its efficiency. In NLCEs, the perturbative treatment of finite clusters is replaced by a numerical exact (block) diagonalization. While LCEs are restricted due to their perturbative nature, the non-perturbative variant is not. In graph-based continuous unitary transformations (gCUTs), the block diagonalization is achieved by a non-perturbative CUT performed on finite clusters. The central objective of this thesis is a modification of the gCUT scheme, allowing to treat two kinds of major issues, denoted by pseudo and genuine decay, occuring for NLCEs due to the non-perturbative treatment of clusters. Indeed, one finds surprising effects caused by the non-perturbative renormalization. In particular, we identify a fundamental challenge for any non-perturbative approach based on finite clusters resulting from the reduced symmetry on graphs, most importantly the breaking of translational symmetry when targeting the properties of excited states. This can be traced back to the appearance of intruder states in the low-energy spectrum, which represent a major obstacle in quasi-degenerate perturbation theory. Here, a generalized notion of cluster additivity is introduced, which is used to formulate an optimized scheme of gCUTs, allowing to solve and to physically understand this major issue. Most remarkably, our improved scheme demands to go beyond the paradigm of using the exact eigenvectors on graphs. We demonstrate that the modified scheme is correct in the non-perturbative regime. Even at quantum criticality, the scheme gives valid results. To determine the critical behavior, one must rely on extrapolation schemes. We introduce a generic approach to extract critical properties from sequences of numerical data which is directly relevant for NLCEs. The scheme is applied to the quantum phase transition between the dimerized and the isotropic spin 1/2 Heisenberg chain. Finally, we investigate a scenario where the gapped quasiparticle excitation is not stable for all momenta, i.e., the one-particle mode merges with the continuum for certain momenta and one observes quasiparticle decay. In this case, intruder states merging with the low-energy spectrum on finite clusters represent genuine physics of the system. Again, a proper renormalization on finite clusters satisfies the generalized cluster additivity. Our adjusted renormalization approach performed on finite clusters does not necessarily lead to a full decoupling of the quasiparticle subspaces and the remaining interactions are part of the effective description. The resulting effective Hamiltonian can be analyzed to describe quasiparticle decay in the thermodynamic limit. The modified scheme is applied to four-leg spin 1/2 Heisenberg ladders, providing insights into the spectral properties. Overall, this thesis presents several important developments for the derivation of effective quasiparticle pictures in quantum spin models via perturbative and non-perturbative LCEs, opening various opportunities for future investigations.
The emergence of complex behavior out of simple elementary building blocks is a typical property of condensed matter systems. While the individual constituents may be well-studied, strongly interacting many-particle systems can give rise to novel collective behavior that can not be understood by considering the individual properties alone. Effective models are a powerful tool to handle this complexity: Instead of aiming at a full description a complex many-particle system on the high-energy scale, an effective model covers only those degrees of freedom which are essential for the low-energy physics. While the high-energy degrees of freedom are integrated out, they influence the low-energy degrees of freedom due to a renormalization of the effective coupling constants. In particular, the concept of the quasi-particle allows for the description of a strongly-interacting many-particle excitations in terms of a simple objects with similar degrees of freedom as a dispersive elementary particle, which gives rise to phenomena like quasi-particle bound states or quasi-particle decay. This thesis aims at the application and methodological improvement of a systematic and well-controlled framework to derive effective low-energy models, denoted as Continuous Unitary Transformations (CUT). In this method, a Hamiltonian is mapped continuously to an effective basis, in which the low-energy degrees of freedom can be interpreted as quasi-particles. Mathematically, the transformation is parametrized by a flow parameter and is represented as a differential equation, denoted as flow equation, while the direction of the transformation is controlled by the generator scheme. As foundation of our studies, we use the sCUT method, in which the types of possible interaction terms are fixed by a truncation scheme. In a common project, we established a new flavor of CUT, called epCUT. It allows us to derive the effective Hamiltonian in a perturbation series. In contrast to the already existent pCUT method, epCUT can handle system with a non-equidistant spectrum as well. Moreover, we discovered the deepCUT method, the non-perturbative twin of epCUT. While it bears large similarities to sCUT, it replaces the cumbersome definition of an appropriate truncation scheme by the order of the flow equation system. Compared to sCUT, deepCUT offers a larger robustness while the ambiguities of the truncation scheme are eliminated. My contribution comprises the computational, parallelized construction of the flow equation system as well. In another project, we investigated the breakdown of the conventional, particle-sorting generator schemes in system with strongly overlapping quasi-particle sub-spaces and identified the emergence of a spectrum unbounded from below due to truncation errors as origin of the divergence. As a testing ground, we used a system of two coupled bosonic degrees of freedom. To overcome these problems, we developed a family of three novel variational generator schemes, denoted as scalar, vectorial and tensorial optimization. After this, we investigated the dimerized, two-dimensional S=1/2 Heisenberg model with an emphasis on the quantum phase transition between a magnetically disordered and a magnetically ordered phase using sCUT. Parametrizing the starting point by a variational parameter, we were able to describe the spontaneous breaking of the SU(2) spin symmetry by CUT for the first time. We were able to calculate the ground state energy and the dispersion of triplons, and magnons, respectively, in both phases. However, we were not able to provide quantitative results for the spontaneous magnetization. Furthermore, we found a way to derive an optimal starting point for the integration of the flow equations that is independent on the particularities of the model and does not require a parametrization by a variational parameter. As a further application, we analyzed the S=1 Heisenberg chain by means of deepCUT. To this end, we mapped the system to an extended S=1/2 Heisenberg ladder with the help of a variational parameter. Using the variational generator schemes developed before, we were able to decouple the low-energy spectrum and to calculate spectral densities. In this investigation, we found an indication for a S=0 bound state. Furthermore, we were able to find some signs of spontaneous quasi-particle decay, but a quantitatively correct description is pending.
In this thesis the method of self-similar continuous unitary transformations (CUTs) in real space is developed. It is applied to various models from the field of strongly correlated solid state physics and optical lattices. The CUT method maps a given Hamiltonian to an effective Hamiltonian which is in some respect simpler. In a self-similar CUT the truncation scheme relies on the structure of the operators. The real space approach put forward in this thesis truncates the operators according to their extension in real space. We mainly use a MKU generator to derive an effective model which is block-diagonal in the number of quasiparticles. We give a detailed description of the method. The transformation of the Hamiltonian and of the observables is explained. The implementation on a computer and the optimization of the programs are extensively discussed. The first application of the method deals with the fermionic Hubbard model. We derive an effective Hamiltonian, a generalized t-J-model, which conserves the number of doubly occupied sites. The effective model includes beyond the well-known Heisenberg term also further two-spin and four-spin terms. In addition, the charge motion and interaction within the effective model is addressed. In the context of the Hubbard model we discuss the question if an effective model that conserves the number of double occupancies can be derived in principle. The antiferromagnetic spin ladder and spin chain are treated from the viewpoint of a dimerized system. For the spin ladder we calculate the dispersion, the two-particle energies including bound states and the multi-particle continua. The discussion of multi-particle continua shows that the order of the energy levels is important for the convergence of the transformation. For the spin chain the dispersion is calculated. In addition, the transformation of an observable yields the corresponding spectral weights. The spectral weights are compared to results from a perturbative CUT. For the bosonic Hubbard model we calculate the dispersion and determine the phase diagram. The spectral weight of a certain observable is calculated in order to explain the outcome of a recent experiment that measures the spectroscopic response of atoms confined in optical traps. The distribution of spectral weight is analyzed also at finite temperatures. We find evidence for a substantial temperature of the size of the microscopic parameters. This finding is supported by purely thermodynamic considerations.
Motivated by the recent discovery of a spin-liquid phase for the Hubbard model on the honeycomb lattice at half-filling (Meng et al 2010 Nature 88 487), we apply both perturbative and non-perturbative techniques to derive effective spin Hamiltonians describing the low-energy physics of the Mott-insulating phase of the system. Exact diagonalizations of the so-derived models on small clusters are performed, in order to assess the quality of the effective low-energy theory in the spin-liquid regime. We show that six-spin interactions on the elementary loop of the honeycomb lattice are the dominant sub-leading effective couplings. A minimal spin model is shown to reproduce most of the energetic properties of the Hubbard model on the honeycomb lattice in its spin-liquid phase. Surprisingly, a more elaborate effective low-energy spin model obtained by a systematic graph expansion rather disagrees beyond a certain point with the numerical results for the Hubbard model at intermediate couplings.
We study the gapped phase of the Kitaev model on the honeycomb lattice using perturbative continuous unitary transformations. The effective low-energy Hamiltonian is found to be an extended toric code with interacting anyons. High-energy excitations are emerging free fermions which are composed of hard-core bosons with an attached string of spin operators. The excitation spectrum is mapped onto that of a single particle hopping on a square lattice in a magnetic field. We also illustrate how to compute correlation functions in this framework. The present approach yields analytical perturbative results in the thermodynamical limit without using the Majorana or the Jordan-Wigner fermionization initially proposed to solve this problem.
Abstract. Motivated by the recent discovery of a spin liquid phase for the Hubbard model on the honeycomb lattice at half-filling [11], we apply both perturbative and non-perturbative techniques to derive effective spin Hamiltonians describing the low-energy physics of the Mott-insulating phase of the system. Exact diagonalizations of the so-derived models on small clusters are performed, in order to assess the quality of the effective low-energy theory in the spin-liquid regime. We show that six-spin interactions on the elementary loop of the honeycomb lattice are the dominant sub-leading effective couplings. A minimal spin model is shown to reproduce most of the energetic properties of the Hubbard model on the honeycomb lattice in its spin-liquid phase. Surprisingly, a more elaborate effective low-energy spin model obtained by a systematic graph expansion rather disagrees beyond a certain point with the numerical results for the Hubbard model at intermediate couplings.