1: \begin{abstract}

2:   

3:    We use tensor network techniques to obtain high order perturbative

4:    diagrammatic expansions for the quantum many-body problem at very high

5:    precision.  The approach is based on a tensor train parsimonious

6:    representation  of the sum of all Feynman diagrams, obtained in a controlled

7:    and accurate way with the tensor cross interpolation algorithm. 

8:    %This representation is an effective separation of variables and therefore enables

9:    %a direct calculation of the high dimensional integrals.  It also yields, at

10:    %no additional computational cost, 

11:    It yields the full time evolution of physical

12:    quantities in the presence of any arbitrary time dependent interaction.  Our

13:    benchmarks on the Anderson quantum impurity problem, within the real time

14:    non-equilibrium Schwinger-Keldysh formalism, demonstrate that this technique

15:    supersedes diagrammatic Quantum Monte Carlo by orders of magnitude in

16:    precision and speed, with convergence rates $1/N^2$ or faster, where $N$ is

17:    the number of function evaluations.  The method also works in parameter

18:    regimes characterized by strongly oscillatory integrals in high dimension,

19:    which suffer from a catastrophic sign problem in Quantum Monte-Carlo.

20:    Finally, we also present two exploratory studies showing that the technique

21:    generalizes to more complex situations: a double quantum dot and a single

22:    impurity embedded in a two dimensional lattice. 

23: 

24: \end{abstract}

25: