1: \begin{abstract}

2: A prerequisite to the successful development of quantum computers and simulators is precise understanding of  physical processes occurring therein, which can be achieved by measuring the quantum states they produce. 

3: However, the resources required for traditional quantum-state estimation  scale exponentially with the system size, highlighting the need for alternative approaches. Here we demonstrate an efficient method for complete characterization of complex multi-qubit quantum states. 

4: Using a variational version of the matrix product state ansatz, we perform the full tomography of quantum states produced in a 20-qubit trapped-ion Ising-type quantum simulator, using the data acquired in only 27 bases with 1000 measurements in each basis. 

5: We observe superior state reconstruction quality and faster convergence compared to the methods based on neural network quantum state representations: 

6: restricted Boltzmann machines and feedforward neural networks with autoregressive architecture. 

7: Our results pave the way towards efficient experimental characterization of complex states produced by the dynamics of many-body quantum systems.

8: \end{abstract}

9: