Tensor Networks: from Simulations to Holography III

The suitability of tensor network ansatzes for the description of physically relevant states in one dimensional lattice gauge theories (LGT) has been demonstrated in the last years by a large amount of systematic studies, including abelian and non-abelian LGTs, and including scenarios where traditional Monte Carlo approaches fail due to a sign problem. While this establishes a solid motivation to extend the program to higher dimensions, a similar systematic study in two dimensions using PEPS requires dealing with specific considerations. Besides a larger computational costs associated to the higher spatial dimension, the presence of plaquette terms in LGTs hinders the efficiency of the most up-to-date PEPS algorithms. With a newly developed update strategy, nevertheless, such terms can be treated by the most efficient techniques. We have used this method to perform the first ab initio iPEPS study of a LGT in 2+1 dimensions: a Z3 invariant model, for which we have determined the phase diagram.

In this talk I will give an overview of tensor network approaches to critical systems. I will discuss entanglement scaling laws, show how PEPS can simulate systems with Fermi surfaces, and present some results for simulating systems in the continuum.

"AdS/CFT endows gravity in anti-de Sitter (AdS) spacetime with a dual description in certain conformal field theories (CFTs) with matching symmetries. Tensor networks on regular discretizations of AdS space provide natural toy models of AdS/CFT, but break the continuous bulk symmetries. In this talk, we discuss several aspects of such toy models based on tensor networks. We show that this produces a quasiregular conformal field theory (qCFT) on the boundary and rigorously compute its symmetries, entanglement properties, and central charge bounds, applicable to a wide range of existing models. An explicit AdS/qCFT model with exact fractional central charges is given by holographic quantum error correcting codes based on Majorana dimers. These models also realize the strong disorder renormalization group, resulting in new connections between critical condensed-matter models, exact quantum error correction, and holography. If time allows, we will briefly review other recent group research on using tensor network models in quantum many-body physics including many-body localization and time crystals as well as in probabilistic modelling. Based on arXiv:2004.04173, Phys. Rev. A 102, 042407 (2020), Phys. Rev. Research 1, 033079 (2019), Science Advances 5, eaaw0092 (2019)."

There is now significant experimental evidence that the physics of the underdoped cuprates is controlled by a metallic state with a Fermi surface whose volume does not equal the Luttinger value. However, there has been no proposed wavefunction for such a state for electrons in a single band. I will describe a wavefunction which involves tracing over 2 layers of ancilla qubits. The proposal also leads to a gauge theory for the transition to the conventional Fermi liquid state found at large doping.

Quantum Cellular Automata are unitary maps that preserve locality and respect causality. I will show that in one spatial dimension they correspond to matrix product unitary operators, and that one can classify them in the presence of symmetries, giving rise to phenomenon analogous to symmetry protection. I will then show that in higher dimensions, they correspond to other tensor networks that fulfill an extra condition and whose bond dimension does not grow with the system size. As a result, they satisfy an area law for the entanglement entropy they can create. I will also define other classes of non-unitary maps, the so-called quantum channels, that either respect causality or preserve locality and show that, whereas the latter obey an area law for the amount of quantum correlations they can create, as measured by the quantum mutual information, theformer may violate it. Additionally, neither of them can be expressed as tensor networks with a bond dimension that is independent of the system size.

"I will introduce a tensor-network based language for classifying topological phases via fixed-point models. The "models" will be tensor networks formalizing a discrete Euclidean path integral living in a topological space-time, and can be obtained from Hamiltonian models by Trotterizing the imaginary time evolution. Topological fixed-point models are invariant under topology-preserving space-time deformations. Space-time manifolds and homeomorphisms can be combinatorially represented by graph-like "networks", which together with "moves" form a "liquid". The networks can be interpreted as tensor networks, and the moves as equations which determine the fixed-point models. Different combinatorial representations of the same space-times yield new kinds of fixed-point models. Given the limited time, I will stick to very simple examples in 1+1 dimensions for this talk."

We examine holographic complexity in the doubly holographic model to study quantum extremal islands. We focus on the holographic complexity=volume (CV) proposal for boundary subregions in the island phase. Exploiting the Fefferman-Graham expansion of the metric and other geometric quantities near the brane, we derive the leading contributions to the complexity and interpret these in terms of the generalized volume of the island derived from the higher curvature action for the brane gravity. Motivated by these results, we propose a generalization of the CV proposal for higher curvature theories of gravity. Further, we provide two consistency checks of our proposal by studying Gauss-Bonnet gravity and f(R) gravity in the bulk.

Multi-scale tensor networks offer a way to efficiently represent ground states of critical systems and may be adapted for state-preparation on a quantum computer. The tensor network for a single scale specifies a quantum channel whose fixed-point is a subregion of the approximate critical ground state. The fixed-point of a noisy channel is perturbed linearly in the noise parameter from the ideal state, making local observables stable against errors for these iterative algorithms. We consider the wavelet-designed circuit for the 1+1D critical Ising ground state as a concrete example to numerically test the noise robustness against our error models and compare the smallest instance case with an implementation on a present-day ion-trap quantum computer.

The search for applications of quantum computers has highlighted the field of quantum chemistry, where one can also apply tensor network methods. There are several challenges in getting useful results for molecules compared to simulating a model Hamiltonian in condensed matter physics. The first issue is in descretizing continuum space to get a finite Hamiltonian which is amenable to tensor network techniques. Another is the need for high accuracy, particularly in energies, to compare with experiments. I will give an overview of the approaches used in this field, and then focus on our work using grid and wavelet-based discretizations coupled with DMRG methods.

The success of the Ryu-Takayanagi formula suggests a profound connection between the AdS/CFT correspondence and tensor networks. There are since many works on constructing examples, although it is very difficult to make them explicit and quantitative. We will discuss some new progress in the toy example of p-adic CFT where its tensor network dual was previously constructed explicitly [ arXiv:1703.05445 , arXiv:1812.06059, arXiv:1902.01411], and how some analogue of Einstein equation on the graph emerges as we consider RG flow of these CFTs. These progresses inspire us to find a way to construct explicit holographic tensor networks of more realistic CFTs based on their connection with topological models with one higher dimension, at least in CFT2. We will present some preliminary results and works in progress.