Containments of Tensor Network Varieties

Sofía Garzón Mora, Christian Haase · 2026

Read Containments of Tensor Network Varieties by Sofía Garzón Mora, Christian Haase (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-2601-19805.txt.

Introduction

Sofía Garzón Mora and Christian Haase study the shapes formed by tensor networks. A tensor network writes a large tensor as a web of small ones arranged along a binary tree. The set of all tensors a given network can produce forms a variety, a geometric object. Understanding these varieties is key to knowing what tensor networks can express.

Building on earlier work, the authors study which of these varieties contain others. Containment tells us when one network format can simulate another. The combinatorics of the tree drive the encoding they study. Such results guide the design of tensor methods in physics and machine learning. They also enrich the geometry of tensors, a fast-growing field.

Catalog record

CREATORSofía Garzón Mora, Christian Haase
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE91 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-19805

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Containments of Tensor Network Varieties Sofía Garzón Mora, Christian Haase Abstract . Building upon the work of Buczyńska et al. [ BBM15 ] , we study here tensor formats and their corresponding encoding of tensors via two-fold tensor products determined by the combinatorics of a binary tree. The set of all tensors representable by a given network forms the corresponding tensor network variety. A very basic question asks whether every tensor representable by one network is representable by another network, namely, when one tensor network variety is contained in another. Specific instances of this question became known as the Hackbusch Conjecture. Here, we propose a general framework for this question and take first steps, theoretical as well as experimental, towards a better understanding. In particular, given any two binary trees on n n leaves, we define (and prove existence of) a new measure, the containment exponent , which gauges how much one has to boost the parameters of one network for the containment to hold. We present an algorithm for bounding these containment exponents of tensor network varieties and report on an exhaustive search among trees on up to n = 8 n=8 leaves. 2021 Mathematics Subject Classification Primary: ; Secondary: 1. Introduction Encoding data efficiently using tensors is common practice in many sciences. The approach we will study here makes use of tensor formats . A tensor format encodes an n -way tensor as a sequence of linear subspaces via two-fold tensor products in a way determined by a binary tree with n leaves. The main idea being that the number of parameters needed to determine the tensor is much smaller than the dimension of the ambient space – the tensor product of n vector spaces. Our choice of binary tree, of a family of vector spaces for the leaves and a dimension function for the subspaces on the vertices of the tree determines a tensor network. This, together with a nesting property of these subspaces of the nodes determines the variety of tensor network states, meaning the locus of all possible tensors which can be encoded in this manner. Hence, the natural question which arises is whether one tensor network variety is contained in another. We will make sense of this question and give precise definitions in the next section. The containments of tensor network varieties have been studied for some particular plane trees, namely the so-called hierarchical tree and the train track tree in [ BBM15 ] , where a conjecture initially posed by Hackbusch [ Hac12 ] is resolved. Here, our aim is to generalize this containment question to more generally shaped trees, disregarding a choice of planar embedding. For this, we study the poset structure of the trees and find a combinatorial bound for the containment of their respective tensor network varieties. In particular, given any two binary trees on n leaves, we define (and prove existence of) a new measure, the containment exponent , which gauges how much one has to boost the parameters of one network for the containment to hold. Namely, when we scale the dimension function of the first network, how much do we have to scale the dimension function of the second for the containment to hold? We have implemented an algorithm, which we use to bound the containment exponents among all tensor network varieties that arise from full binary rooted trees with n\leq 8 leaves. Since the bound depends on the labeling of the leaves, in order to cover all possible containment exponents, we take into account possible permutations of the labels of the n leaves. The results for n=4,5,6,7,8 leaves are collected in [ sofiaGitHub ] . Additionally, we expand the previous results from the literature by introducing a combinatorial bound for the containment of varieties arising from arbitrarily shaped plane binary trees into one another. We present examples showing that the obtained bounds are not necessarily sharp and end this paper with some open questions. 2. Tensor Formats for Binary Trees We begin by introducing basic notions regarding binary trees and the associated varieties of network states. Then, we review results from the literature, mainly [ BBM15 ] , which will be crucial in our argument. 2.1. Trees All trees in this article will be full binary, rooted trees. Definition 2.1 . A full binary tree \mathcal{T} is a rooted tree with root \mathbb{r}(\mathcal{T}) such that each vertex has either exactly two children, in which case it is referred to as a node , or it has no descendants at all, so that it is called a leaf . A plane tree is a tree together with a choice of which child of any given node is left and which one is right. We denote the set of nodes (including \mathbb{r}(\mathcal{T}) ) by \mathcal{N}(\mathcal{T}) , the set of leaves by \mathcal{L}(\mathcal{T}) . We omit the tree from the notation if the tree which we are talking about is clear from the context. We will also say that \mathcal{T} is a tree on \mathcal{L} . The set of vertices is the union \mathcal{V}=\mathcal{N}\cup\mathcal{L} . We have |\mathcal{N}|=|\mathcal{L}|-1 . The descendant relation gives a poset structure to \mathcal{V} . We write v\leq_{\mathcal{T}}w if v is a descendant of w . Following [ BBM15 ] , we will denote by \downarrow\negthickspace^{\negthickspace{}^{\mathcal{T}}}\hskip 1.42262ptv ⊆ L thesubsetofdescendantleavesof v ,andby \uparrow\negthickspace_{\negthickspace{}_{\mathcal{T}}}\hskip 1.42262ptv thecomplement L ∖ \downarrow\negthickspace^{\negthickspace{}^{\mathcal{T}}}\hskip 1.42262ptv .Ifitisunambiguouswhichtreeweareworkingwith,wewillomititfromthenotationandonywrite \uparrow\negthickspace_{\negthickspace}\hskip 1.42262ptv respectively \downarrow\negthickspace^{\negthickspace}\hskip 1.42262ptv .\par Twoparticularclassesoftreeswhichhavegottenspecialattentioninapplicationsandintheliteraturearethehierarchicalandthetraintracktrees:\par The\textit{hierarchical tree} H T _k isaperfectbinarytree,i.e.afullbinarytreeinwhicheveryleafhasexactlythesamenumber k ofancestors.Wesaythetree H T _k hasdepth k ,sothat | V |=2^k+1-1 and | L |=2^k .Anytwoplanetreeswiththisunderlyingtreeareisomorphic.\par\begin{figure}\centering\hbox to161.79pt{\vbox to98.57pt{\pgfpicture\makeatletter\hbox{\hskip 80.89337pt\lower-92.16347pt\hbox to0.0pt{\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{stroke} \lxSVG@begingroup@{fill} \lxSVG@setlinewidth{\the\pgflinewidth}\lxSVG@begingroup@{stroke-width} \lx@inpgf@ignorespaces\nullfont\hbox to0.0pt{\lxSVG@begingroup@{_scopebegin} {{}}{{{ {}{}{}}}}{}{}\lx@inpgf@ignorespaces\hbox{\hbox{\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{-9.04169pt}{-3.0754pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{60}\lxSVG@closescope }}} \lxSVG@closescope }}}\hbox{{\lxSVG@begingroup@{_scopebegin} {{}}{{}}{{}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{-47.5252pt}{-44.0819pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{60}\lxSVG@closescope }}} \lxSVG@closescope }}} { {{{ {}{}{}}}}{}{}{ {}{}{}}{}{ {}{}{}} {{{{{}}{ {}{}}{}{}{{}{}}}}}{}{{{{{}}{ {}{}}{}{}{{}{}}}}}{{}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M -9.14 -9.14 L -50.15 -50.15}{fill:none} \lx@inpgf@ignorespaces\hbox{\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}}{{}}{{}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{-77.56036pt}{-88.83046pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{60}\lxSVG@closescope }}} \lxSVG@closescope }}} { { {}{}{}}{}{ {}{}{}} {{{{{}}{ {}{}}{}{}{{}{}}}}}{}{{{{{}}{ {}{}}{}{}{{}{}}}}}{{}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M -63.51 -67.96 L -83.74 -108.42}{fill:none} \lx@inpgf@ignorespaces }\lxSVG@closescope }}}\hbox{{\lxSVG@begingroup@{_scopebegin} {{}}{{}}{{}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{-34.88124pt}{-88.83046pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{60}\lxSVG@closescope }}} \lxSVG@closescope }}} { { {}{}{}}{}{ {}{}{}} {{{{{}}{ {}{}}{}{}{{}{}}}}}{}{{{{{}}{ {}{}}{}{}{{}{}}}}}{{}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M -54.6 -67.96 L -34.37 -108.42}{fill:none} \lx@inpgf@ignorespaces }\lxSVG@closescope }}} }\lxSVG@closescope }}}\hbox{{\lxSVG@begingroup@{_scopebegin} {{}}{{}}{{}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{37.83307pt}{-44.0819pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{60}\lxSVG@closescope }}} \lxSVG@closescope }}} { {{{ {}{}{}}}}{}{}{ {}{}{}}{}{ {}{}{}} {{{{{}}{ {}{}}{}{}{{}{}}}}}{}{{{{{}}{ {}{}}{}{}{{}{}}}}}{{}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M 9.14 -9.14 L 50.15 -50.15}{fill:none} \lx@inpgf@ignorespaces\hbox{\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}}{{}}{{}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{7.79791pt}{-88.83046pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{60}\lxSVG@closescope }}} \lxSVG@closescope }}} { { {}{}{}}{}{ {}{}{}} {{{{{}}{ {}{}}{}{}{{}{}}}}}{}{{{{{}}{ {}{}}{}{}{{}{}}}}}{{}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M 54.6 -67.96 L 34.37 -108.42}{fill:none} \lx@inpgf@ignorespaces }\lxSVG@closescope }}}\hbox{{\lxSVG@begingroup@{_scopebegin} {{}}{{}}{{}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{50.47704pt}{-88.83046pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{60}\lxSVG@closescope }}} \lxSVG@closescope }}} { { {}{}{}}{}{ {}{}{}} {{{{{}}{ {}{}}{}{}{{}{}}}}}{}{{{{{}}{ {}{}}{}{}{{}{}}}}}{{}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M 63.51 -67.96 L 83.74 -108.42}{fill:none} \lx@inpgf@ignorespaces }\lxSVG@closescope }}} }\lxSVG@closescope }}} \lxSVG@closescope {\lx@inpgf@ignorespaces}{\lx@inpgf@ignorespaces}{\lx@inpgf@ignorespaces}\hss}\lxSVG@discardpath\lxSVG@closescope \hss}}\lxSVG@closescope\endpgfpicture}} \@@toccaption{{\lx@tag[ ]{{1}}{Hierarchical tree $H\mathcal{T}_{2}$ of depth 2, with 4 leaves.}}}\@@caption{{\lx@tag[. ]{{\small Figure 1}}{\small Hierarchical tree $H\mathcal{T}_{2}$ of depth 2, with 4 leaves.}}} \@add@centering\end{figure}\par The\textit{train track tree} T T _n with n leavesischaracterizedbythefactthateachnodehasatleastoneleafasachild.\begin{figure}\centering\hbox to204.47pt{\vbox to141.25pt{\pgfpicture\makeatletter\hbox{\hskip 144.91211pt\lower-134.84259pt\hbox to0.0pt{\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{stroke} \lxSVG@begingroup@{fill} \lxSVG@setlinewidth{\the\pgflinewidth}\lxSVG@begingroup@{stroke-width} \lx@inpgf@ignorespaces\nullfont\hbox to0.0pt{\lxSVG@begingroup@{_scopebegin} {{}}{{{ {}{}{}}}}{}{}\lx@inpgf@ignorespaces\hbox{\hbox{\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{-9.04169pt}{-3.0754pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{60}\lxSVG@closescope }}} \lxSVG@closescope }}}\hbox{{\lxSVG@begingroup@{_scopebegin} {{}}{{}}{{}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} {{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lx@inpgf@ignorespaces }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{-47.5252pt}{-44.0819pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{60}\lxSVG@closescope }}} \lxSVG@closescope }}} { {{{ {}{}{}}}}{}{}{ {}{}{}}{}{ {}{}{}} {{{{{}}{ {}{}}{}{}{{}{}}}}}{}{{{{{}}{ {}{}}{}{}{{}{}}}}}{{}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M -9.14 -9.14 L -50.15 -50.15}{fill:none} \lx@inpgf@ignorespaces\hbox{\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin}

Read the complete text