Abstract:We propose a method to construct a tensor network representation of partition functions without singular value decompositions nor series expansions. The approach is demonstrated for one- and two-dimensional Ising models and we study the dependence of the tensor renormalization group (TRG) on the form of the initial tensors and their symmetries. We further introduce variants of several tensor renormalization algorithms. Our benchmarks reveal a significant dependence of various TRG algorithms on the choice of initial tensors and their symmetries. However, we show that the boundary TRG technique can eliminate the initial tensor dependence for all TRG methods. The numerical results of TRG calculations can thus be made significantly more robust with only a few changes in the code. Furthermore, we study a three-dimensional $\mathbb{Z}_2$ gauge theory without gauge-fixing and confirm the applicability of the initial tensor construction. Our method can straightforwardly be applied to systems with longer range and multi-site interactions, such as the next-nearest neighbor Ising model.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is about the dependence of the Tensor Renormalization Group (TRG) method on the initial tensor form when dealing with statistical physics problems. Specifically, the author proposes a new method for constructing tensor network representations, which does not rely on singular value decomposition (SVD) or series expansion. Through this method, the author studies the dependence of different TRG algorithms on the initial tensor and its symmetry, and proposes an improved boundary TRG technique to eliminate this dependence, thereby enhancing the robustness of numerical results.
### Main contributions:
1. **New tensor network construction method**: The author proposes a simple decomposition method based on the identity matrix, which avoids using SVD or series expansion and directly constructs a locally - connected tensor network. This method can be applied to systems with long - range and multi - point interactions, such as the next - nearest - neighbor Ising model.
2. **Study of the dependence on the initial tensor**: Through the study of one - dimensional and two - dimensional Ising models, the author finds that the accuracy of different TRG algorithms (such as HOTRG) is highly dependent on the symmetry of the initial tensor. In particular, for asymmetric initial tensors, the accuracy of HOTRG decreases significantly.
3. **Boundary TRG technique**: The author introduces the boundary TRG technique. By using "squeezers" instead of isometries in the coarse - graining step, the dependence on the initial tensor form is eliminated. This makes the TRG algorithm more robust, and high - precision results can be obtained even when using asymmetric initial tensors.
4. **Application to three - dimensional Z2 gauge theory**: The author applies the new tensor network construction method to three - dimensional Z2 gauge theory, verifies its applicability, and calculates the free energy, specific heat, and critical temperature, which are consistent with previous calculation results.
### Research methods:
- **One - dimensional Ising model**: As a simple example, the author shows how to construct a locally - connected tensor network and verifies the effectiveness of the method through exact diagonalization.
- **Two - dimensional Ising model**: By comparing the effects of different initial tensors (such as tensors based on the delta function and Taylor expansion) on the TRG and HOTRG algorithms, the author studies the influence of the symmetry of the initial tensor on the accuracy of the algorithms.
- **Three - dimensional Z2 gauge theory**: The new tensor network construction method is applied to calculate the free energy and specific heat, verifying the applicability of the method in higher - dimensional systems.
### Conclusions:
- **Robustness improvement**: By introducing the boundary TRG technique, the robustness of the TRG algorithm can be significantly improved, and the dependence on the initial tensor form can be eliminated.
- **Wide applicability**: The new tensor network construction method is not only applicable to the Ising model, but can also be extended to other systems with long - range and multi - point interactions.
- **Computational efficiency**: While maintaining high precision, the new method simplifies the construction process of the tensor network and improves computational efficiency.
In summary, this paper solves the problem of the TRG algorithm's dependence on the initial tensor form by proposing a new tensor network construction method and an improved TRG technique, providing more robust and efficient computational tools for complex systems in statistical physics and quantum field theory.