Site Basis Excitation Ansatz for Matrix Product States
Steven R. White · 2025
Read Site Basis Excitation Ansatz for Matrix Product States by Steven R. White (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-06241.txt.
Introduction
Excitations of a one-dimensional quantum magnet are computed cheaply with a site basis ansatz in a 2025 paper by Steven R. White. Matrix product states describe such ground states compactly, and tangent-space methods extract their excitation spectra, but at cost. White's variation builds a small excitation basis from one diagonalization resembling a single-site DMRG step, gathered for several states at once.
After overlap and Hamiltonian elements are computed, any momentum's excitation needs only a tiny matrix diagonalization. A simple alternative to variational uniform matrix product states, based on finite-system DMRG, supplies the ground state. On the spin-one Heisenberg chain the one-magnon dispersion comes out with high accuracy. Gauge choices matter: leaving the basis nonorthogonal is crucial, while forcing left-orthonormal form badly slows convergence.
Catalog record
| CREATOR | Steven R. White |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-06241
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Site Basis Excitation Ansatz for Matrix Product States Steven R. White Abstract We introduce a simple and efficient variation of the tangent-space excitation ansatz used to compute elementary excitation spectra of one-dimensional quantum lattice systems using matrix product states (MPS). A small basis for the excitation tensors is formed based on a single diagonalization analogous to a single site DMRG step but for multiple states. Once overlap and Hamiltonian matrix elements are found, obtaining the excitation for any momentum only requires diagonalization of a tiny matrix, akin to a non-orthogonal band-theory diagonalization. The approach is based on an infinite MPS description of the ground state, and we introduce an extremely simple alternative to variational uniform matrix product states (VUMPS) based on finite system DMRG. For the S = 1 S=1 Heisenberg chain, our method—site basis excitation ansatz (SBEA)—efficiently produces the one-magnon dispersion with high accuracy. We also examine the role of MPS gauge choices, finding that not imposing a gauge condition—leaving the basis nonorthogonal—is crucial for the approach, whereas imposing a left-orthonormal gauge (as in prior work) severely hampers convergence. We also show how one can construct Wannier excitations, analogous to the Wannier functions of band theory, where one Wannier excitation, translated to all sites, can reconstruct the single magnon modes exactly for all momenta. The spectrum of elementary excitations of a quantum system is one of its most important properties. For a one-dimensional system, conventional density matrix renormalization group (DMRG) White 1992 ; White 1993 methods allow one to obtain a limited number of excited states to be obtained, but obtaining a full dispersion relation is difficult. Dynamical algorithms, such as dynamical DMRG Jeckelmann 2002 , which works in frequency space, or those that use real-time evolution Vidal 2004 ; Daley et al. 2004 ; White and Feiguin 2004 ; Haegeman et al. 2016 ; Schollwöck 2011 ; Paeckel et al. 2019 , offer an alternative White and Affleck 2008 , but a dramatic increase in efficiency arose with the introduction of the matrix product state (MPS) version of the single-mode approximation, called the excitation ansatz (EA) Haegeman et al. 2012 ; Vanderstraeten et al. 2019 . In EA, the excitation is a plane-wave superposition of MPS, where in each MPS the tensor on one site is different from that of the ground state and creates a local excitation near that site. Remarkably, the accuracy of this simple ansatz for simple gapped systems often rivals that of DMRG itself. On a technical level, this approach is not without difficulties. The standard method for obtaining a uniform infinite MPS, the variational uniform matrix product states Zauner-Stauber et al. 2018a (VUMPS) method, can fail to converge depending on the initial matrices, and the original formulation involved a complicated set of diagrams even for near-neighbor interactions Haegeman et al. 2012 . Although EA is very efficient compared to previous approaches, each different k requires a separate full diagonalization. Our site basis excitation ansatz (SBEA) addresses each of these shortcomings, increasing the efficiency of an already-efficient method. Perhaps just as important, our approach is arguably simpler, lowering the threshold for adoption. We will test our algorithms using the spin S=1 antiferromagnetic Heisenberg chain, with Hamiltonian H=J\sum_{j}\mathbf{S}_{j}\cdot\mathbf{S}_{j+1}, (1) where \mathbf{S}_{j} are the spin-1 operators at site j and J=1 . This model is gapped Haldane 1983a ; Haldane 1983b , and the lowest excitations are triplet magnons Affleck et al. 1987 ; Affleck 1989 ; White and Huse 1993 . Two of its features, being an integer spin chain and one with a finite gap and correlation length ( \xi\approx 6 ), are important for the algorithm used. The finite correlation length means that finite system DMRG can reach the bulk in the center of a finite system. The \ldots AAA\ldots single-site infinite MPS form is appropriate for integer spin chains. (Here and below, we will frequently use this simplified notation for uniform MPS, which indicates the unit cell.) Note that we will make use of the finite (and modest) correlation length to utilize certain algorithmic simplifications—shortcuts, which are not central to SBEA, but which may be very convenient in avoiding some technical complications particularly for the non-expert. The uniform infinite ground state MPS.— The first step is to obtain a uniform or infinite MPS (iMPS) representation of the ground state. Substantial progress has been made in this area, and two powerful methods are infinite DMRG (iDMRG) McCulloch 2008 and the variational uniform MPS algorithm (VUMPS) Zauner-Stauber et al. 2018a . Here our main goal in looking for an alternative was simplicity, rather than pure efficiency. Finite system DMRG White 1992 ; White 1993 is well understood, efficient, and highly reliable, and the finite-size gaps of the system help speed convergence. If we take finite system DMRG as a black box, can we get uniform infinite MPS? It turns out that answer is yes, and the method is almost trivially simple. It also does not appear to be costly compared to existing infinite methods. It can be considered a simplification of iDMRG, but without utilizing recursion. We start by performing finite system DMRG on a long open chain ( N\gg\xi , e.g. N=150 ), where a matrix product operator (MPO) form for the Hamiltonian is convenient, particularly since MPOs can be generated automatically Fishman et al. 2022 . For N\gg\xi the center is representative of the infinite system in its local properties, but the central tensors have random gauges, making them unsuitable for making an infinite uniform MPS. Specifically, we can say the states of the links on each side of each tensor are different, whereas for the \ldots AAA\ldots form they should be the same. To make the links the same, we insert a new site in the middle, with a new tensor. (Inserting sites in the middle goes all the way back to the original infinite-system DMRG White 1992 ; White 1993 .) To accommodate the new N+1 sites, we simply reconstruct the Hamiltonian MPO for N+1 sites. Picking a central link on the N -site system, we put the orthogonality center there, and obtain a \tilde{\lambda} diagonal matrix of Schmidt coefficients for that link. (For example, to put it between tensors CD , we contract these two tensors and then split them apart with a singular value decomposition, obtaining C^{\prime}\tilde{\lambda}D^{\prime} .) The half-MPS on either side of the link form orthogonal bases for \tilde{\lambda} . We add a new site at that link, replacing \tilde{\lambda} with a new three leg tensor \tilde{D} , which we initialize randomly. We then perform a Lanczos diagonalization to optimize the energy, to high accuracy, equivalent to a single step of single site finite-size DMRG, but without a good initial state. Since this is done only once, the extra cost of the poor initialization and high accuracy is negligible. As in the original infinite system DMRG, the overall state obtained (assuming a single site unit cell is appropriate) is just as accurate for N+1 sites as the state it started from was on N sites. And, importantly, \tilde{D} has identical left and right indices, so it can immediately make an infinite MPS. The canonical form of an infinite MPS looks like \ldots\lambda\Gamma\lambda\Gamma\ldots , where \Gamma have