1: \begin{abstract}

2: In this work, we benchmark the well-controlled and numerically

3: accurate exponential thermal tensor renormalization group (XTRG)

4: in the simulation of interacting spin models in two dimensions.

5: Finite temperature introduces a thermal correlation length,

6: which justifies the analysis of finite system size for the sake

7: of numerical efficiency. In this paper we focus on the square

8: lattice Heisenberg antiferromagnet (SLH) and quantum Ising

9: models (QIM) on open and cylindrical geometries up to width

10: $W=10$. We explore various one-dimensional mapping paths in the

11: matrix product operator (MPO) representation, whose performance

12: is clearly shown to be geometry dependent. We benchmark against

13: quantum Monte Carlo (QMC) data, yet also the series-expansion

14: thermal tensor network results. Thermal properties including

15: the internal energy, specific heat, and spin structure factors,

16: etc., are computed with high precision, obtaining excellent

17: agreement with QMC results. XTRG also allows us to reach

18: remarkably low temperatures. For SLH we obtain at low

19: temperature an energy per site $u_g^*\simeq -0.6694(4)$ and a

20: spontaneous magnetization $\mS^*\simeq0.30(1)$, which is

21: already consistent with the ground state properties. We extract an

22: exponential divergence vs. $T$ of the structure factor $S(M)$,

23: as well as the correlation length $\xi$, at the ordering wave

24: vector $M=(\pi,\pi)$, which represents the renormalized

25: classical behavior and can be observed over a narrow but

26: appreciable temperature window, by analysing the finite-size

27: data by XTRG simulations. For the QIM with a

28: finite-temperature phase transition, we employ several thermal

29: quantities, including the specific heat, Binder ratio, as well

30: as the MPO entanglement to determine the critical temperature

31: $T_c$.

32: \end{abstract}

33: