Truncating loopy tensor networks by zero-mode gauge fixing
Ihor Sokolov, Yintai Zhang, Jacek Dziarmaga · 2025
Read Truncating loopy tensor networks by zero-mode gauge fixing by Ihor Sokolov, Yintai Zhang, Jacek Dziarmaga (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-2508-00338.txt.
Introduction
Ihor Sokolov, Yintai Zhang, and Jacek Dziarmaga present a better way to shrink loopy tensor networks, a tool used to compute quantum states. Their 2025 paper explains that tensor networks with loops hold inner correlations that make compression wasteful. Standard cuts throw away more accuracy than needed.
Their fix works locally on one bond of the network. Cutting the bond defines a set of states, and where those states depend on each other, redundant parts can be removed through what they call zero mode gauge fixing. Tests on two standard setups, a paired entangled state method and a renormalization step, gave smaller starting errors than the usual initialization in every example. Sharper compression means bigger quantum systems can be modeled on the same hardware.
Catalog record
| CREATOR | Ihor Sokolov, Yintai Zhang, Jacek Dziarmaga |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 46 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-00338
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Truncating loopy tensor networks by zero-mode gauge fixing Ihor Sokolov, Yintai Zhang, Jacek Dziarmaga Abstract Loopy tensor networks have internal correlations that often make their compression inefficient. We show that even local bond optimization can make better use of the insight it has locally into relevant loop correlations. By cutting the bond, we define a set of states whose linear dependence can be used to truncate the bond dimension. The linear dependence is eliminated with zero modes of the states’ metric tensor. The method is illustrated by a series of examples for the infinite pair-entangled projected state (iPEPS) and for the periodic matrix product state (pMPS) that occurs in the tensor renormalization group (TRG) step. In all examples, it provides better initial truncation errors than standard initialization. I Introduction Understanding strongly correlated quantum many-body systems is a long-standing problem, especially in two spatial dimensions (2D), where exact diagonalization is limited to small system sizes and quantum Monte Carlo is hampered by the notorious sign problem. The problem is bypassed by tensor networks (TN) that provide an efficient representation for typical ground states of quantum many-body systems Verstraete et al. 2008 ; Orús 2014 ; Okunishi et al. 2022 . They include the matrix product states (MPS) in one dimension (1D) Fannes et al. 1992 ; Schöllwock 2011 , the projected entangled pair state (PEPS) in 2D Nishino et al. 2001 ; Gendiar et al. 2003 ; Verstraete and Cirac 2004 and 3D Vlaar and Corboz 2021 ; Lukin and Sotnikov 2024 , and the multi-scale entanglement renormalization ansatz (MERA) Vidal 2007 ; Vidal 2008 ; Evenbly and Vidal 2014a ; Evenbly and Vidal 2014b . MPS that is so powerfull in 1D, thanks to its canonical structure, in 2D is limited to small system sizes. This limitation does not apply to PEPS Nishino et al. 2001 ; Gendiar et al. 2003 ; Verstraete and Cirac 2004 ; Murg et al. 2007 ; Jordan et al. 2008 ; Jiang et al. 2008 ; Orús and Vidal 2009 ; Phien et al. 2015 ; Lubasch et al. 2014 ; Corboz 2016a ; Vanderstraeten et al. 2016 ; Fishman et al. 2018 ; Xie et al. 2017 ; Corboz 2016b ; Corboz et al. 2018 ; Rader and Läuchli 2018 ; Rams et al. 2018 that is its 2D generalization but, for lack of a canonical form that could be treated efficiently Haghshenas et al. 2019 , its expressive power may remain largely unused. PEPS has closed loops that makes local optimization of its tensors less effective, as it does not fully account for correlations around the loops. In this paper we show that even local optimization can make better use of the insight it has locally into relevant loop correlations. The nature of the problem can be illustrated by the example in Fig. 1 . There is a virtual entanglement loop around a plaquette that is decoupled from physical indices but, nevertheless, parasites bond dimensions along the plaquette’s edges. Admittedly, this clear-cut example can probably be fixed in a number of ways but in practice virtual loops are more elusive, as they are neither quite decoupled nor even well defined as loops. Therefore, rather than chasing after a precise definition, we assume a more pragmatic approach. We pick a bond in a tensor network that needs to be truncated, open it for inspection and find an optimal way to truncate its bond dimension by eliminating linear dependence of quantum states constituting the TN. Much of the virtual entanglement is removed along the way. Figure 1: Cartoon virtual loop entanglement — The top panel shows a four-tensor plaquette in a larger tensor network (TN). The four tensors are contracted by the black bond indices with dimension D . Additionally, there is a blue loop carrying a virtual index j that is decoupled from any physical index. The TN state is a sum over j=1...d , \mbox{$|\rm TN\rangle$}=\sum_{j=1}^{d}\mbox{$|\psi_{j}\rangle$} , where each state |\psi_{j}\rangle is the same and proportional to the TN state. The bottom panel shows the same plaquette after the indices i and j were merged into a single index k=1...Dd . Its bond dimension is d times bigger than necessary to represent the TN state. Any single state |\psi_{j}\rangle with the smaller bond dimension D would suffice to represent the same state: \mbox{$|\rm TN\rangle$}\propto\mbox{$|\psi_{j}\rangle$} . The paper is organized as follows. In Sec. II we introduce a simple truncation method, directly inspired by Fig. 1 , where the states |\psi_{j}\rangle are not linearly independent and their linear dependence implies zero modes that can be used to reduce the bond dimension. In Sec. III a toy example is provided to compare the zero mode truncation with a standard method based on the pseudoinverse. In Sec. IV the truncated bond is opened to define a more general form of linear dependence and more general zero modes that allow for more accurate truncation. In Sec. V we define loopiness of the network, as perceived by the truncated bond, and put the zero-mode truncation (ZMT) in the context of the environment assisted truncation (EAT) Sinha et al. 2022 . EAT is the optimal truncation for a bond that perceives the TN as non-loopy. This Section completes our general considerations, the following Sections are a series of examples. In Sec. VI we simulate unitary time evolution after a sudden quench in the 2D transverse field quantum Ising model with the iPEPS TN. Even such an unsuspecting set-up can accumulate significant loopiness with evolution time. With the loopiness, the more general ZMT provides a better initial truncation than EAT. In Sec. VII we consider imaginary time evolution of a purification of the Gibbs thermal state in the Heisenberg model. The purification is represented by iPEPS. The loopiness increases as the temperature is lowered. The initial error after ZMT is found to lie mid way between the initial error after EAT and the final error after the following variational optimization. In Sec. VIII we consider again a sudden quench but this time for the Z_{2} gauge field, where the plaquette terms in the Kogut-Susskind Hamiltonian generate loop entanglement in the iPEPS TN. EAT, to an even greater extent than a simple initial SVD truncation, fails to manage the entanglement around the plaquettes. They provide poor initial truncation that cannot be compensated by the following variational optimization. ZMT is a better initialization with a better final error after the optimization. In Sec. IX we simulate a sudden quench in the fermionic t - J model with a U(1)\times U(1) -symmetric iPEPS. ZMT yields an initial truncation error better than mid way between the initial EAT error and the final error after the optimization. In Sec. X , we leave the iPEPS realm and consider several initial truncation methods in the tensor renormalization group (TRG) for the classical Ising model. TRG is notoriously plagued by the loop entanglement in a periodic MPS that appears in every coarse-graining step. ZMT is shown to make better initial truncation than the standard one employing the Vidal gauge. Finally, we conclude in Sec. XI . Figure 2: Bond zero modes. — In (a) the grey semi-ellipsis contains a tensor network (TN). The red lines are