Neural network quantum state tomography.
A Quantum Calculator Used for Many-body Eigenstate Reconstruction.
QuCumber is a program that reconstructs an unknown quantum wavefunction from a set of measurements. The measurements should consist of binary counts; for example, the occupation of an atomic orbital, or the Sz eigenvalue of a qubit. These measurements form a training set, which is used to train a stochastic neural network called a Restricted Boltzmann Machine. Once trained, the neural network is a reconstructed representation of the unknown wavefunction underlying the measurement data. It can be used for generative modelling, i.e. producing new instances of measurements, and to calculate estimators not contained in the original data set.
QuCumber is developed by the Perimeter Institute Quantum Intelligence Lab (PIQuIL). The project is currently in an early-beta, expect some rough edges, bugs, and backward incompatible updates.
QuCumber is licensed under the Apache License Version 2.0.
QuCumber implements unsupervised generative modelling with a two-layer RBM. Each layer is a number of binary stochastic variables (with values 0 or 1). The size of the visible layer corresponds to the input data, i.e. the number of qubits. The size of the hidden layer is varied to systematically control representation error.
Currently the reconstruction can be performed on pure states with a positive-definite wavefunction. Data is thus only required in one basis. Upcoming versions will allow reconstruction of more general wavefunctions and density matrices; in this case tomographyically-complete basis sets may be required in the training data.
Python 3. QuCumber is written in PyTorch, with CPU and GPU support. See https://pytorch.org.
Documentation can be found here.