On the Optimal Layout of Two-Dimensional Lattices for Density Matrix Renormalization Group
Antonello Scardicchio · 2026
Read On the Optimal Layout of Two-Dimensional Lattices for Density Matrix Renormalization Group by Antonello Scardicchio (2026) 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-2601-13762.txt.
Introduction
Antonello Scardicchio asks how best to order a 2D lattice for a key quantum algorithm in 2026. The method is the density matrix renormalization group, or DMRG. It solves quantum spin models by sweeping through sites in some order. For a flat lattice, the order of sites, called the layout, is a free choice.
Scardicchio seeks the layout that gives the best results at fixed cost. Best here means the lowest variational energy found. It also means low maximum entropy and low truncation error. A better layout gets more truth from the same compute budget.
Catalog record
| CREATOR | Antonello Scardicchio |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-13762
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
On the Optimal Layout of Two-Dimensional Lattices for Density Matrix Renormalization Group Antonello Scardicchio Abstract For quantum spin models defined on a two-dimensional lattice, we look for the best numbering of the lattice sites (a layout) that, at fixed bond dimension and other parameters of the density matrix renormalization group (DMRG) algorithm, gives the lowest value of the variational energy, maximum entropy and truncation error. We consider the conjecture that the optimal layout is a Hamiltonian path, and that it optimizes a simply computable geometric cost function. Finding the minimum of such a function, which is a variant of the minimum linear arrangement problem, provides the DMRG with an efficient layout of the lattice and improves both accuracy and convergence time. Applications to antiferromagnetic or spin glass spin- 1 / 2 1/2 models on the square and triangular lattices are studied. I Introduction Two-dimensional quantum lattice models –such as the Hubbard model, the Heisenberg antiferromagnet, various Ising or Heisenberg spin glasses, and lattice gauge theories– occupy a critical position in condensed matter physics 11 ; 9 ; 2 ; 24 ; 18 . Two dimensions is both experimentally accessible and a theoretical sweet spot: Much interesting quantum physics happens in two dimensions, far from the mean-field predictions but at the same time not too far from one dimension, so that modern numerical methods which are best suited for one dimension, like density matrix renormalization group (DMRG) 25 ; 19 , matrix product states or operators (MPS-MPO) 23 or projected entangled pairs states (PEPS) 22 can still be applied with a certain degree of success. DMRG and Tensor Network 15 methods are, in fact, naturally formulated for one-dimensional chains but can be adapted with some degree of success to higher-dimensional models 21 , by mapping them to a one-dimensional chain with long-range interactions. Mapping a higher-dimensional lattice to a one-dimensional one is an instance of graph layout 5 . A graph layout trivially exists for any graph, see Fig. 1 , however deciding whether the layout has a particular property or finding the layout that, for a given graph, is optimum with respect to some measure, is a hard combinatorial problem, often in the NP-hard or NP-complete class. Graph layout problems have received a lot of attention in very diverse fields. In computer science, the problem appears in circuit design and computer network optimization, in bioinformatics, graph layout algorithms are used to project high-dimensional data, such as protein-protein interaction networks and single-cell gene expression profiles, into interpretable visualizations that reveal functional clusters and cellular relationships 26 . In this sense, the problem we consider in this paper follows a long tradition of incarnations of the graph layout problem. We want to investigate the graph layout for mapping two-dimensional lattices into one-dimensional spin chains, and in particular we want to optimize for the performance of the algorithm. Recently, there have been several heuristic solutions for an optimal layout for the DMRG algorithm 4 ; 10 ; 1 ; 3 ; 16 which showed promise to improve both the ground state detection and the time evolution of initial states for quantum simulations. In those papers, there is already the idea that the optimal layout should be a fractal path, which wanders in a local patch of sites, before leaping to a neighboring patch and doing that again. Here we start from this generic consideration and go further in various directions trying to answer the final question what is the optimal layout for running the DMRG algorithm? Figure 1 : A square lattice with a simple layout. The layout is the numbering of the vertices, which induces the graph on the bottom. One should first consider two fundamental questions: 1) Is the optimal graph layout for DMRG a Hamiltonian path? We have not investigated this question systematically but from a few instances, it seems that the short answer is yes. This is not obvious a priori since the optimal layout depends strongly on the cost function we need to optimize. For example, it is known for some natural cost functions, that the answer is negative 12 . The optimal layout is not a path, and big jumps are necessary to optimize the cost function. However, for the best performance of the DMRG it seems one can restrict to Hamiltonian paths instead of general layout. The investigation of this problem has been hampered by the high cost of running DMRG for every proposed layout. This brings us to a second problem: 2) Is there a geometric cost function which is so strongly correlated with the performance of DMRG that we can use it as a proxy of the latter? This is necessary, as we said, for a comprehensive study of the optimal layout/path since in an optimization algorithm, evaluating the cost function is something which is done at every proposed change in the proposal. Instead, if we find a geometric cost function that can be evaluated efficiently for the given layout and only call DMRG on the optimal path, one can get a significant speedup. Let us introduce some fundamental concepts in the next section and then dwell in the optimization problem in the following one. II Layouts and Optimal Paths II.1 General considerations We follow 5 defining the graph G(V,E) where V=\{1,...,N\}\subset\mathbb{N} is the set of vertices, and E=\{uv|u,v\in V\} is the collection of the edges. A layout is a bijection of the vertices \varphi:V\ni i\mapsto j\in\{1,...,N\}. A Hamiltonian path is a particular layout in which any two consecutive vertices j,j+1 are joined on the original graph \varphi^{-1}(j)=u,\varphi^{-1}(j+1)=v,\ uv\in E . Given a layout \varphi there are a plethora of cost functions one can define, the optimization of which can lead to complicated problems. We are interested in one of them. Define the layout distance between two vertices u,v\in V , \lambda(u,v)=|\varphi(u)-\varphi(v)| . In principle, this distance can go up to N-1 . If uv\in E (the two vertices are connected on G ), then \lambda(u,v) is the called the length of uv in the layout \varphi . The linear arrangement \mathrm{LA} is defined as \mathrm{LA}(\varphi,G)=\sum_{uv\in E}\lambda(u,v). (1) This leads to the problem \textsc{MinLA}(G)=\min_{\varphi}\textsc{LA}(\varphi,G) originally introduced in 6 and solved for hypercubes in 12 : the optimal layout is not a Hamiltonian path. One can generalize the cost function as follows \mathrm{LA}_{q}(\varphi,G)=\sum_{uv\in E}\lambda(u,v)^{q}, (2) with q<1 and the corresponding \textsc{MinLA}_{q}(G) . This has been already considered in 12 where several lower bounds are given for the optimal layout. We will see that the cost function which is best correlated with the performance of the DMRG algorithm is \textsc{LA}_{1/2} , which, as already recongized in 12 is a particular case, separating the 1\geq q>1/2 and the q<1/2 cases which have a simpler behavior of the optimal layouts. II.2 Cost function for DMRG The layout problem in DMRG appears for the following reason 25 . Consider a system of N spin- 1/2 degrees of freedom, whose Hamiltonian has interactions which define some particular graph G . The MPS ansatz with a given bond dimension \chi for a given wave function \ket{\Psi} is \ket{\Psi}=\sum_{\{s\}}A^{1}_{s_{1}}\cdot A^{2}_{s_{2}}\cdot...\cdot A^{N}_{s_{N}}\ket{s_{1},...,s_{N}} (3) where each A^{i}_{s_{i}} is a \chi\times\chi matrix indexed by the spin- z value s_{1}=\pm 1/2 . This is independent of