Time-dependent density-matrix renormalization-group using adaptive effective Hilbert spaces

From MaRDI portal



Abstract: An algorithm for the simulation of the evolution of slightly entangled quantum states has been recently proposed as a tool to study time-dependent phenomena in one-dimensional quantum systems. Its key feature is a time-evolving block-decimation (TEBD) procedure to identify and dynamically update the relevant, conveniently small subregion of the otherwise exponentially large Hilbert space. Potential applications of the TEBD algorithm are the simulation of time-dependent Hamiltonians, transport in quantum systems far from equilibrium and dissipative quantum mechanics. In this paper we translate the TEBD algorithm into the language of matrix product states in order to both highlight and exploit its resemblances to the widely used density-matrix renormalization-group (DMRG) algorithms. The TEBD algorithm being based on updating a matrix product state in time, it is very accessible to the DMRG community and it can be enhanced by using well-known DMRG techniques, for instance in the event of good quantum numbers. More importantly, we show how it can be simply incorporated into existing DMRG implementations to produce a remarkably effective and versatile ``adaptive time-dependent DMRG variant, that we also test and compare to previous proposals.


One of the authors proposed a time evolving block-decimation (TEBD) algorithm, an algorithm for the simulation of slightly entangled quantum states [\textit{G. Vidal}, Efficient simulation of one-dimensional quantum many-body system, Phys. Rev. Lett. 93, 040502 (2004)]. There already exists density-matrix renormalization group (DMRG) algorithms as a tool to study time-dependent phenomenona in one-dimensional quantum systems and widely used (cf. \textit{S. G. White}, Density matrix formulation for quantum renormalization groups, Phys. Rev Lett. 69, 2863 (1992)]. Since-TEBD algorithm was motivated from quantum information science, its description was done by using a slightly difierent language used in the DMRG algorithm. In this paper, the TEBD algorithm is described by using the common language that are used to describe the DMRG algorithm. The authors say both algorithms search for an approximation to the true wavefunction within a restricted class of wavefunctions, which can be identified as matrix product states (or finitely correlated states). According to this argument, the authors integrate these two algorithms by using White's prediction algorithm [\textit{S. G. White}, Phys. Rev. Lett. 77, 3633 (1996)]. It provides the adaptive time-dependent DMRG algorithm, which is said to be extremely powerful by the authors. The outline of the paper is as follows: In \S1, beside stating motivations and outlines of the paper, the following brief exposition of the DMRG algorithm is given: Let the quantum state of a one-dimensional system be \[ |\psi\rangle= \sum_i\sum_j\psi_{ij}| i\rangle| j\rangle, \] where a partition of the system into two blocks \(S\) and \(E\) is considered. \(\{| i\rangle \}\) and \(\{| j\rangle\}\) are orthonormal basis of \(S\) and \(E\), respectively. The DMRG decimation procedure consists of projecting \(|\psi\rangle\) on the Hilbert spaces for \(S\) and \(E\) spanned by the \(M\) eigenvectors \(| w_\alpha^S \rangle\) and \(|\psi_\alpha^E \rangle\) corresponding to the largest eigenvalues \(\lambda^2_\alpha\) of the reduced density matrices \[ \widehat\rho_S= \text{Tr}_E|\rangle \langle\psi|,\quad\widehat\rho_E=\text{Tr}_S|\psi\rangle\langle\psi|, \] such that \(\widehat\rho_S| wS_\alpha\rangle= \lambda 2_\alpha| wS_\alpha\) and \(\widehat\rho_E| wE_\alpha\rangle= \lambda 2_\alpha| wE_\alpha\rangle\). In \S2, simulations of the time-dependent Bose-Hubbard Hamiltonian \[ \widehat H_{\text{BH}}(t) =-J\sum^{L-1}_{i=1}b^\dag_{i+1}b_i+b_i^\dag b_{i+1}+\frac{U(t)} {2}\sum^L_{i-!}n_i(n_i-1), \] by using DMRG methods with different instantaneous change of interaction strength, are done. They show even relatively robust numerical quantities such as nearest-neighbour correlations can be qualitatively and quantitatively improved by the additional targeting of states. Therefore it is desired to have a modified DMRG algorithm which, for all time \(t\), selects Hilbert spaces of dimension \(M\) such that \(|\psi(t)\rangle\) is represented optimally in the DMRG sense. The authors say the presentation of such an algorithm is the purpose of the paper. \S3 reviews matrix product states (finitely correlated states. \textit{M. Fannes, B. Nachtergaele} and \textit{R. F. Werner}, Commun. Math. Phys., 144, 443 (1992; Zbl 0755.46039)]. Let \({\mathcal H}\) be divided local Hilbert spaces \({\mathcal H}_i\); \({\mathcal H}=\prod^L_{i=1} {\mathcal H}_i\). Then a product state is expressed as \[ |\sigma\rangle=\sigma_1 \rangle \otimes|\sigma_2\rangle\otimes\cdots\otimes|\sigma_L\rangle. \] A general state is expressed as \(\psi\rangle=\sum_\sigma\psi_\sigma \rangle\). To define a matrix product state, operators \(\widehat A_i [\sigma_i]\); \[ \widehat A_i [\sigma_i]=\sum_{\alpha,\beta} A^i_{\alpha,\beta}[\sigma_i] |\alpha \rangle\langle \beta|, \] are introduced. Then an unnormalized matrix product state in a form that is useful for Hamiltonians with open boundary condition is defined as \[ | \overline \psi\rangle=\sum_\sigma\left(\langle \varphi_L| \prod^L_{i=1}\widehat A_i[\sigma_i]|\varphi_R \rangle\right)| \sigma\rangle, \] where \(|\varphi_L\rangle\) and \(|\varphi_R \rangle\) are the left and right boundary states. By using this language, TEBD and DMRG algorithms are described in \S4 and \S5. The TEBD algorithm described in \S4 excludes the use of periodic boundary condition (the authors say if a periodic boundary condition is included, the formulation becomes complicated). A DMRG-style truncation to keep the most relevant density matrix eigenstates is carried out at each step in the TEBD simulation algorithm. This is in contrast with time-dependent DMRG methods used up to now. The authors say this is the key point about the TEBD simulation algorithm. Since the actual decomposition and truncation procedure in the DMRG and the TEBD simulation algorithm are identical, authors propose to use the finite-time algorithm to carry out the sequence of local time evolutions. According to this proposal, the adaptive time-dependent DMRG algorithm, which incorporates the TEBD simulation algorithm in the DMRG framework, is presented in \S6. This is the heart of this paper. Evaluation of computing time is also given. In \S7, this algorithm is adapted to the simulation of the time-dependent Bose-Hubbard model. It is observed that the Trotter error is inversely proportional to the calculation time spent. The authors conclude this paper saying that the adaptive time-dependent DMRG can be applied not only to problems with explicitly time-dependent Hamiltonians, but also to problems where the quantum state changes strongly in time, such as in systems where the initial quantum state is far from equilibrium. At the end of this paper, the authors remark that there is a closely related paper [\textit{S. R. White} and \textit{A. E. Feiguin}, Real-time evolution using density matrix renormalization group, Phys. Rev. Lett 93, 076401 (2004)].




Cited in
(66)








This page was built for publication: Time-dependent density-matrix renormalization-group using adaptive effective Hilbert spaces

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3160031)