An implementation of the Lanczos algorithm for the exact diagonalization of the two dimensional Hubbard model on a 4×4 square lattice on the Connection Machine CM-2 system is described. The CM-2 is a massively parallel machine with distributed memory. The program is written in C/PARIS. This implementation minimizes memory usage by generating the matrix elements as needed instead of storing them. The Lanczos vectors are stored across the local memory of the processors. Using translational symmetry only, the dimension of the Hilbert space at half filling is more than 10 million. A speed of about 2.4 min per iteration is achieved on a 64K CM-2. This implementation is scalable. Running it on a bigger machine with more processors speeds up the process. The performance analysis of this implementation is shown and discuss its advantages and disadvantages are discussed.
An implementation of the Lanczos algorithm for the exact diagonalization of the two dimensional Hubbard model on a 4×4 square lattice on the Connection Machine CM-2 system is described. The CM-2 is a massively parallel machine with distributed memory. The program is written in C/PARIS. This implementation minimizes memory usage by generating the matrix elements as needed instead of storing them. The Lanczos vectors are stored across the local memory of the processors. Using translational symmetry only, the dimension of the Hilbert space at half filling is more than 10 million. A speed of about 2.4 min per iteration is achieved on a 64K CM-2. This implementation is scalable. Running it on a bigger machine with more processors speeds up the process. The performance analysis of this implementation is shown and discuss its advantages and disadvantages are discussed.
The physics of strongly correlated materials poses one of the most challenging problems in condensed-matter sciences. Standard approximations applicable to wide classes of materials such as the local density approximation fail, due to the importance of the Coulomb repulsion between localized electrons. Instead, we resort to non-perturbative many-body methods. The calculations are, however, only feasible for rather small model systems. The full Hamiltonian of a real material is approximated by a model Hamiltonian comprising only the most important electronic degrees of freedom, while the effect of all other electrons is included in an average way by renormalizing the parameters. Realistic calculations of strongly correlated materials need to include sufficiently many of these electronic degrees of freedom. The new generation of massively parallel supercomputers allows for these realistic calculations. However, exploiting their computational power requires newly devised algorithms. As a solver we use the Lanczos method, which needs the full many-body state of the correlated system. It is thus limited by the available main memory. The foremost problem for a distributed-memory implementation is that the multiplication of the Hamiltonian to the many-body state leads to highly non-local memory access patterns. A simple yet important observation leads to an efficient solution: in the kinetic term of the Hamiltonian the electron-spin is conserved. Thus, writing the many-body vector as a matrix v(iup,idn), where the indices label spin-configurations, we find that the hopping term only connects vector elements that differ in one index. Hence, storing entire slices v(iup,:) on one node, the kinetic term for the spin-down electrons is local to that process. After transposing v, the same is true for the hopping of the spin-up electrons. With an efficient matrix transpose, implemented via MPI collectives, we thus obtain a version of the Lanczos method that efficiently well on JUBL, Jülich's Blue Gene/L. Using the code we treated two-orbital Hubbard Models which show interesting Mott-band insulator transitions. Kohn's criterium which determines the phase of a system by ground-state properties alone, namely the response of the ground-state energies to a vector potential, was employed. This vector potential can be equivalently regarded as a change in boundary conditions, requiring complex wave vectors. We found that half-filled systems become metallic in the region between a band and Mott insulating phase. To extrapolate to infinite system sizes and therefore getting access to an arbitrarily high resolution in reciprocal space, we use cluster perturbation theory (CPT). It works by diagonalizing a finite cluster exactly and then treating the hopping between identical clusters perturbatively in strong coupling perturbation theory. This leads to an effectively infinite chain. Due to our efficient Lanczos implementation it is possible to calculate relatively large clusters, minimizing finite-size effects. The combination of our Lanczos code and CPT allowed for a treatment of the the organic metal TTF-TCNQ. We were able to solve a long-standing problem in the interpretation of its experimental spectra. Being an interesting one-dimensional metal TTF-TCNQ has been studied thoroughly for more than 30 years. Theoretical estimates as well as experimental measurements suggest values for the parameters of the Hubbard model t and U. With these values Hubbard model calculations show qualitatively the same features as APRES experiments, namely the signatures of spin-charge separation at the Gamma-point. But in order to fit the experimental data to the numerical calculations t has to be doubled, which translates into a broadening of the spectra. This ad-hoc adjustment is not only unsatisfactory but also the temperature dependence of the spectral function is incorrectly reproduced. We resolved this problem by properly including nearest-neighbor Coulomb interactions. Recent DFT calculations show that indeed the commonly used values of t and U are correct. But Coulomb interactions between electrons on neighboring sites are also significant and must not be neglected. Thus, we studied the effect of the next-neighbor interaction parameter V being at about U/2. Including V in our Lanczos calculations for obtaining the angular-integrated spectral function shows that its effect is to broaden the spectra, comparable to increasing the value of t. Moreover, we found that doubling t indeed mimics the effect of Vapprox U/2. Surprisingly, the broadening of the spectrum as a result of V can be understood in first-order Rayleigh-Schrödinger perturbation theory.
We present two methods to study model Hamiltonians defined on clusters. One is the symmetrization of the Hilbert space basis according to irreducible representations of the symmetry group of the Hamiltonian. The other one adapts the Lanczos diagonalization method directly to the second quantized form of the Hamiltonian studied. By the symmetrization of the basis, one takes advantage of the symmetries of the Hamiltonian, and uses the group theory to find the invariant subspaces of the Hamiltonian in the complete Hilbert space. Then one needs only to work within the invariant subspaces. This method cuts down substantially the size of the matrix representation of the Hamiltonian to be diagonalized. Therefore, one could study relatively larger systems. As a result, all the eigenvectors and eigenvalues are obtained. Consequently, one could calculate thermodynamical quantities and various correlation functions, such as, the spin correlation, the charge correlation, the superconductivity pairing correlation, and the photoemission spectra. One the other hand, the Lanczos method, mentioned here in particular, is a specially developed computer program for the study of the model Hamiltonians. This program uses the key feature of the Lanczos method in diagonalization, allows one to use the second quantized form of the Hamiltonian directly without pregeneration or storing the matrix elements of the Hamiltonian. Two methods could be combined to take the full advantage of the symmetries of the Hamiltonian by input of the symmetrized wavefunction as the starting trial wavefunction in the Lanczos method. Here we illustrate the above two methods by applying to the Hubbard, and extended Hubbard models on a simple cubic cluster, and the lattice Anderson model on the two dimensional square lattice. Other interesting results for different systems are also briefly discussed here.
We study the basic features of the two-dimensional quantum Hubbard Model at half-filling by means of the Lüscher algorithm and the algorithm based on direct update of the determinant of the fermionic matrix. We implement the Lüscher idea employing the transfer matrix formalism which allows to formulate the problem on the lattice in (2+1) dimensions. We discuss the numerical complexity of the Lüscher technique, systematic errors introduced by polynomial approximation and introduce some improvements which reduce long autocorrelations. In particular we show that preconditioning of the fermionic matrix speeds up the algorithm and extends the available range of parameters. We investigate the magnetic and the one-particle properties of the Hubbard Model at half-filling and show that they are in qualitative
The Lanczos algorithm has been applied to the Hubbard model on a 4\ifmmode\times\else\texttimes\fi{}4 square lattice and several equal-time correlation functions have been evaluated. This computation has been made possible by a classification of all the irreducible representations of the space symmetry group, which allowed a considerable reduction of the dimension of the Hilbert space of the system. Such an analysis is quite general and might be useful in other physical applications. Our results shed light on the effects of doping on the magnetic ordering, suggest a spin-charge decoupling at large U, and show a remarkable reduction of the quasiparticle weight at strong coupling.
Exact diagonalization is a well-established method for simulating small quantum systems. Its applicability is limited by the exponential growth of the Hamiltonian matrix that needs to be diagonalized. Physical symmetries are usually utilized to reduce the matrix dimension, and distributed-memory parallelism is employed to explore larger systems. This paper focuses on an implementation of the core distributed algorithms, with a special emphasis on the matrix-vector product. Instead of the conventional MPI+X paradigm, Chapel is chosen as the language in this work.
We present algorithmic improvements for fast and memory-efficient use of discrete spatial symmetries in exact diagonalization computations of quantum many-body systems. These techniques allow us to work flexibly in the reduced basis of symmetry-adapted wave functions. Moreover, a parallelization scheme for the Hamiltonian-vector multiplication in the Lanczos procedure for distributed memory machines avoiding load-balancing problems is proposed. We demonstrate that using these methods low-energy properties of systems of up to 50 spin-1/2 particles can be successfully determined.