Tree Tensor Networks Methods for Efficient Calculation of Molecular Vibrational Spectra
Shuo Sun, Richard M. Milbradt, Stefan Knecht, Chandan Kumar, Christian B. Mendl · 2025
Read Tree Tensor Networks Methods for Efficient Calculation of Molecular Vibrational Spectra by Shuo Sun, Richard M. Milbradt, Stefan Knecht, Chandan Kumar, Christian B. Mendl (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-2512-15875.txt.
Introduction
Shuo Sun and co authors use tree tensor networks to compute molecular vibrational spectra in a 2025 paper. Vibrational spectra reveal how molecules move internally, but coupled vibrations in many dimensions defeat direct calculation. Tensor networks compress the needed quantum states into linked pieces whose shape, or topology, can be tuned.
The team tests tree shapes from simple matrix product chains to general trees where every node carries a physical leg, as in multilayer time dependent Hartree methods. Two eigensolvers are implemented. On a 64 dimensional oscillator model and on acetonitrile, all shapes reach high accuracy. Inverse iteration refinement brings all 84 acetonitrile states below 1 wavenumber of error, and a comb like tree balances cost and accuracy best.
Catalog record
| CREATOR | Shuo Sun, Richard M. Milbradt, Stefan Knecht, Chandan Kumar, Christian B. Mendl |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 50 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-15875
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Tree Tensor Networks Methods for Efficient Calculation of Molecular Vibrational Spectra Shuo Sun, Richard M. Milbradt, Stefan Knecht, Chandan Kumar, Christian B. Mendl Abstract We develop and employ general Tree Tensor Networks (TTNs) to compute the vibrational spectra for two model systems: a set of 64-dimensional coupled oscillators and acetonitrile. We explore various tree architectures, ranging from the simple linear structure of Matrix Product States (MPS), to trees where only the leaf nodes carry a physical leg—as seen in the underlying ansatz of the Multilayer Multiconfiguration Time-Dependent Hartree (ML-MCTDH) method—and further to more general trees in which all nodes are allowed to possess a physical leg. In addition, we implement Locally Optimal Block Preconditioned Conjugate Gradient (LOBPCG) methods and Inverse Iteration methods as eigensolvers. By means of comprehensive benchmarking of runtime and accuracy, we demonstrate that sub-wavenumber accuracy in vibrational spectra is achievable with all TTN structures. MPS and three-legged tree tensor network states (T3NS) have similar runtimes, whereas leaf-only trees require significantly more time. All numerical simulations were performed using PyTreeNet, a Python package designed for flexible tensor network computations. I Introduction Vibrational spectroscopies, such as IR and Raman spectroscopies, are widely used in chemistry to characterize molecules. However, interpreting such spectra is not always straightforward, as the vibrational degrees of freedom scale as 3N-6 (where N is the number of nuclei), making numerical simulations crucial for structure and property determination. In recent years, numerous methods have been developed to obtain vibrational spectra, such as VSCF 6 ; 10 ; 45 ; 35 , VCI 34 ; 5 ; 30 ; 27 , VCCSD 11 ; 37 ; 38 ; 43 , neural network methods 46 , and quantum simulation 31 ; 26 . VSCF is a mean-field theory and often lacks the accuracy required for the desired applications. Although VCI, VCCSD, and neural networks can provide accurate calculations, they do so at a high computational cost. Quantum simulation methods also incur high costs when simulated classically or require quantum hardware beyond current near-term device capabilities. As a promising addition to the available methods, tensor network methods have made significant progress in vibrational spectra calculations in recent years 33 ; 21 ; 2 ; 17 ; 16 ; 20 ; 22 ; 23 . However, most implementations have used matrix product operators (MPOs) to represent the Hamiltonian. This approach limits the full potential of tensor network methods, as the molecules’ normal modes exhibit long-range interactions, while the MPO formalism primarily favors nearest-neighbor interactions. This structural limitation is particularly problematic when focusing on higher-lying excited eigenstates, where energy degeneracy becomes significant, making accurate eigenstate computation even more challenging. To this end, we propose in this paper an approach that utilizes tree tensor network operators (TTNOs) 15 ; 29 to represent the Hamiltonian and likewise tree tensor network states (TTNSs) 39 ; 18 to represent the state function for eigenvalue problem calculations, thereby fully leveraging tree structure connectivity. We implement the LOBPCG method and inverse iteration method within the tree tensor network framework to solve the eigenvalue problem. We demonstrate sub-wavenumber accuracy on two benchmark systems — 64-dimensional coupled oscillators and acetonitrile 9 ; 24 — and compare runtime and accuracy across different tree structures (MPS, T3NS 18 , and leaf-only trees). All computations were performed with PyTreeNet 28 . II Methods In this section, we first briefly introduce tree tensor networks in subsection II.1 . We then discuss two well-known numerical eigenvalue algorithms in subsection II.2 . Finally, we combine these two aspects and introduce our implementation of these numerical algorithms within the tensor network framework in subsection II.3 . II.1 Tree Tensor Network Operations II.1.1 Introduction to Tensor Networks The field of tensor networks has seen significant research activity over the last two decades. It has been especially successful when applied to the numerical simulation of quantum systems. We will restrict this introduction to the concepts relevant to the present work. For a deeper and pedagogical introduction, refer to 36 ; 14 ; 7 ; 40 , and for a general overview, see the recent reviews 32 ; 12 ; 3 . In general, a tensor is an element A\in\mathbb{C}^{\times_{i=1}^{N}D_{i}} , i.e., a multi-dimensional array whose entries are identified by a string of N indices, where the i th index can take values in \{1,\dots,D_{i}\} . Intuitively, this is a generalization of vectors and matrices, where the entries are identified by one and two indices, respectively. We will call N the degree of a tensor, while we refer to the different indices on a tensor as legs . If a leg i of a tensor A and a leg i^{\prime} of a different tensor B have the same dimension D_{i}=D_{i^{\prime}} , we can contract over these legs by summing over the products of the tensor entries with the same index for the contracted legs. Let us emphasize this with an example, where A\in\mathbb{C}^{D_{1}\times D_{2}\times D_{3}} and B\in\mathbb{C}^{D_{2}\times D_{4}\times D_{5}} . Here we can contract the second leg of A with the first leg of B , leading to a new tensor C\in\mathbb{C}^{D_{1}\times D_{3}\times D_{4}\times D_{5}} . The entries of C are defined by C_{jk\ell m}=\sum_{n=1}^{D_{2}}A_{jnk}B_{n\ell m}. (1) Now, a tensor network is simply a collection of tensors and a record of the legs that are contracted. To facilitate easier readability, a graphical notation for tensor networks is available. One can build this representation by depicting every tensor as a geometrical shape and attaching one line per leg. Then, all lines representing contracted legs are connected to their corresponding partners. For example, Eq. (1) can be depicted as \raisebox{-14.22636pt}{ \hbox to142.87pt{\vbox to57.31pt{\pgfpicture\makeatletter\hbox{\hskip 14.62933pt\lower-28.65276pt\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{{\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{fill} \lxSVG@setlinewidth{\the\pgflinewidth}\lxSVG@begingroup@{stroke-width} \lx@inpgf@ignorespaces{{}{}{{ {}{}}}{ {}{}} {{}{{\lx@inpgf@ignorespaces}}}{{}{\lx@inpgf@ignorespaces}}{}{{}{\lx@inpgf@ignorespaces}} {\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{fill} \lxSVG@setlinewidth{\the\pgflinewidth}\lxSVG@begingroup@{stroke-width} \lx@inpgf@ignorespaces{}\lxSVG@fillstroke\lxSVG@drawpath@unclipped{M -13.31 -9.34 h 26.63 v 18.68 h -26.63 Z}{} \lx@inpgf@ignorespaces \lxSVG@closescope }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{-3.93124pt}{-3.41666pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\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}{3.34521pt}{14.74641pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\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}{14.98848pt}{2.21825pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\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}{-11.29633pt}{-20.54404pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\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}{6.99086pt}{-19.2246pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {}{{}}{}{ {}{}{}} {{{{{}}{ {}{}}{}{}{{}{}}}}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M 0 39.37 L 0 10.17}{fill:none} \lx@inpgf@ignorespaces {}{{}}{}{ {}{}{}} {{{{{}}{ {}{}}{}{}{{}{}}}}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M -7.87 -39.37 L -2.03 -10.17}{fill:none} \lx@inpgf@ignorespaces {}{{}}{}{ {}{}{}} {{{{{}}{ {}{}}{}{}{{}{}}}}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M 7.87 -39.37 L 2.03 -10.17}{fill:none} \lx@inpgf@ignorespaces {}{{}}{}{ {}{}{}} {{{{{}}{ {}{}}{}{}{{}{}}}}}{}{}{}{}\lxSVG@stroke\lxSVG@drawpath@unclipped{M 39.37 0 L 14.14 0}{fill:none} \lx@inpgf@ignorespaces {{}}\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}{33.09976pt}{-1.83437pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{fill} \lxSVG@setlinewidth{\the\pgflinewidth}\lxSVG@begingroup@{stroke-width} \lx@inpgf@ignorespaces{{}{{{}}}{{}}{}{}{\lx@inpgf@ignorespaces}{\lx@inpgf@ignorespaces}{}{}{}{}{}{\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{fill} \lxSVG@setlinewidth{\the\pgflinewidth}\lxSVG@begingroup@{stroke-width} \lx@inpgf@ignorespaces{}\lxSVG@fillstroke\lxSVG@drawpath@unclipped{M 92.23 0 C 92.23 7.45 86.19 13.49 78.74 13.49 C 71.29 13.49 65.25 7.45 65.25 0 C 65.25 -7.45 71.29 -13.49 78.74 -13.49 C 86.19 -13.49 92.23 -7.45 92.23 0 Z M 78.74 0}{} \lx@inpgf@ignorespaces \lxSVG@closescope }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{53.15552pt}{-3.41666pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\lx@inpgf@ignorespaces\hbox{\hbox{{\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{fill} \lxSVG@setlinewidth{\the\pgflinewidth}\lxSVG@begingroup@{stroke-width} \lx@inpgf@ignorespaces{{}{{{}}}{{}}{}{}{\lx@inpgf@ignorespaces}{\lx@inpgf@ignorespaces}{}{}{}{}{}{\lxSVG@begingroup@{_scopebegin} \lxSVG@begingroup@{fill} \lxSVG@setlinewidth{\the\pgflinewidth}\lxSVG@begingroup@{stroke-width} \lx@inpgf@ignorespaces{}\lxSVG@fillstroke\lxSVG@drawpath@unclipped{M 151.62 0 C 151.62 7.64 145.43 13.82 137.8 13.82 C 130.16 13.82 123.97 7.64 123.97 0 C 123.97 -7.64 130.16 -13.82 137.8 -13.82 C 145.43 -13.82 151.62 -7.64 151.62 0 Z M 137.8 0}{} \lx@inpgf@ignorespaces \lxSVG@closescope }{{{{\lx@inpgf@ignorespaces}}\lxSVG@begingroup@{_scopebegin} \lxSVG@transformcm{1.0}{0.0}{0.0}{1.0}{95.54126pt}{-3.41666pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\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}{48.8698pt}{14.74641pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\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}{73.82118pt}{3.53769pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope }}} {{}}\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}{48.45462pt}{-20.54404pt}\lxSVG@begingroup@{transform} \pgfsys@hbox{66}\lxSVG@closescope }}} \lxSVG@closescope