Talks

Whilst tomography has dominated the theory behind reconstructing/approximating quantum objects, such as states or channels, conducting full tomography is often not necessary in practice. If one is interested in learning properties of a quantum system, side-stepping the exponential lower bounds of tomography is then possible. In this talk, we will introduce various learning models for approximating quantum objects, survey the literature of quantum learning theory and explore instances where learning can be fully time- and sample efficient.

Particle physics underpins our understanding of the world at a fundamental level by describing the interplay of matter and forces through gauge theories. Yet, despite their unmatched success, the intrinsic quantum mechanical nature of gauge theories makes important problem classes notoriously difficult to address with classical computational techniques. A promising way to overcome these roadblocks is offered by quantum computers, which are based on the same laws that make the classical computations so difficult. Here, we present a quantum computation of the properties of the basic building block of two-dimensional lattice quantum electrodynamics, involving both gauge fields and matter. This computation is made possible by the use of a trapped-ion qudit quantum processor, where quantum information is encoded in d different states per ion, rather than in two states as in qubits. Qudits are ideally suited for describing gauge fields, which are naturally high-dimensional, leading to a dramatic reduction in the quantum register size and circuit complexity. Using a variational quantum eigensolver we find the ground state of the model and observe the interplay between virtual pair creation and quantized magnetic field effects. The qudit approach further allows us to seamlessly observe the effect of different gauge field truncations by controlling the qudit dimension. Our results open the door for hardware-efficient quantum simulations with qudits in near-term quantum devices.

"Null Polygons" in N=4 SYM theory describe the multi-point correlators of 1/2-BPS local operators with large R-charge, when they approach the vertices of a light-like polygon. The leading UV divergences of null polygons is conjectured to satisfy a hierarchy of coupled Toda field theory equations [E.O., Vieira ’22]. I will present some progress towards the prediction of Null Polygons beyond leading logarithms via the hexagons technique, appropriately truncated in the light-cone regime. The method, still conjectural, relies on a series of weak-coupling derivations performed in the Fishnet limit of the theory, where the hexagon representation is derived in the basis of eigenfunctions of a conformal Heisenberg magnet in the principal series. I will present a number of worked-out examples for multi-point multi-loop Fishnet Feynman integrals and Null Polygons.

---

Zoom link

BosonSampling is one of the leading candidate models for a demonstration of quantum computational advantage. However, there are still important gaps between our best theoretical results and what can be implemented realistically in the laboratory. One of the largest gaps concerns the scaling between the number of modes (m) and number of photons (n) in the experiment. The original proposal by Aaronson and Arkhipov, as well as all subsequent improvements, required m to scale as n^2, whereas most state-of-the-art typically operate in a regime where m is linear in n. In this talk, I will describe how our recent work bridges this gap by providing evidence that BosonSampling remains hard even for m as low as 2n. I will review the template for proofs of computational advantage used in BosonSampling and other proposals, and discuss how we solved the new challenges that appear in this regime.