October 2016

Abstracts of the QSIT Lunch Seminar, Thursday, October 6, 2016

Neural-Network Quantum States: Teaching Quantum Mechanics to an Artificial Intelligence

Giuseppe Carleo - Institute for Theoretical Physics, ETH Zurich

Providing a reliable solution of the quantum many-body problem is a central conundrum in many areas of Physics, with the ultimate goal of describing the behavior of strongly-interacting matter ranging from Condensed-Matter to Ultracold atoms. Central to this problem is the exponential complexity of the exact wave-function, which only for very specific geometries and systems can be either compressed (with
Tensor Networks) or sampled efficiently (with Quantum Monte Carlo). In this Seminar I will show how this problem can be attacked from a completely different perspective, devising artificial neural networks learning and improving themselves with the goal of solving Schrödinger’s equation.

In particular, I will review our recent work [Carleo & Troyer, arXiv:1606.02318] where we introduce, for the first time, a representation of the many-body wave function in terms of a suitable artificial neural network. We have devised a machine-learning approach in which the network is adjusted in order to best represent either the ground-state or the dynamics of the quantum many-body system in exam. This approach is demonstrated on prototypical spin models in one and two dimensions. In particular, we have obtained new state-of-the-art results for the description of the ground-state properties of the two-dimensional Heisenberg model, and we have also obtained numerically exact dynamical properties of interacting spins in one dimension. This is an unprecedented result for a stochastic method, which
are traditionally affected by the infamous sign/phase problem and cannot access quantum dynamics with standard approaches.
Implications for Quantum Information theory as well as the recent exact construction of our quantum states for long-range entangled topological models [Deng, Li, Das Sarma, arXiv:1609.09060] will be also discussed.


Trapping Ions in an optical lattice for quantum simulation

Matthew Grau - Trapped Ion Quantum Information  Group, ETH Zurich

Quantum many-body spin Hamiltonians are important tools for describing condensed matter systems, with important implications for the fields of physics, chemistry, materials science, and biology. Finding numerical solutions to such spin models is difficult on conventional computers due to the exponential scaling of Hilbert space with system size, however quantum simulation of these models by a precisely controlled experimental system is a powerful alternative technique for studying these systems. Arrays of trapped ions are attractive platform for quantum simulation due to the high level of single particle control that is possible, as well as the existence of a long range Coulomb interaction that can be used to engineer tunable spin-spin couplings. To that end we are designing a new apparatus to trap arrays of ions in optical lattices for the purpose of quantum simulation. Optical lattices have the capability of trapping ions in many more interesting geometric structures than is possible with conventional ion traps, including arbitrarily complex two dimensional crystals. We will describe the design of the device, which will include a MOT as a reservoir of cold neutral atoms, a high finesse optical cavity to enhance intensity of the optical lattice light to enable trapping ion at close distances, and a cryo-cooled experiment chamber to suppress background gas collisions.

JavaScript has been disabled in your browser