Experiment Humphry

Experiment Humphry, named after Humphry Davy, the discoverer of sodium, is a dual-species apparatus which studies the physics of ultracold (bosonic) sodium and (fermionic) lithium atoms.

Recently, the research has been focused on strongly-interacting Fermi superfluids, in particular, in the crossover regime between the Bose-Einstein condensation (BEC) of dimers and Cooper pairing (BCS regime). Primarily, we are interested in the non-equilibrium physics that emerge after perturbing the superfluids, which could pave the way for a better understanding of rapidly switchable superconductors.

Of special interest is the unitary region of strong interactions that connects these two boundaries, as it presents the highest critical temperature for superfluidity (measured in units of the Fermi temperature) for all known materials.

© AG Quantum

Experimental Methods

Interaction Control in Atomic Sample

The atomic samples are prepared from a dual-species oven and Zeeman-slower in a magneto-optical trap, followed by evaporative cooling to quantum degeneracy. In this process, we use the sodium atoms as a refrigerator for the lithium atoms. 

Once we reach nanokelvin temperatures, we employ Feshbach resonances for tuning the interactions in an atomic gas by applying an external magnetic field of several hundreds of Gauss. Feshbach resonances provide us with the means of studying strongly- or weakly-interacting gases at will and also to create molecules from pairs of atoms. A particular feature of our experiment is that we are able to change the magnetic field by up to 36 G in 3 µs – faster than any internal dynamics of our quantum gas.

[Research article: Rev. Sci. Instrum. 92, 093202(2021)]

Machine Learning

For data analysis we employ methods of artificial intelligence and machine learning. We work, for example, with artificial neural networks for image processing and train both supervised and unsupervised networks for the detection of phase transitions. The combination of ultracold atom experiments and machine learning has greatly improved experimental capabilities since it facilitates the extraction of weak signals in noisy environments.

Research Highlights

Higgs mode in a Fermionic Superfluid

© Martin Link

The Higgs and Goldstone modes are collective modes which can be observed in an order parameter after breaking a continuous symmetry. In the BEC-BCS crossover, the symmetry is broken when transitioning through the critical point from a thermal gas to a superfluid phase. While the Goldstone mode corresponds to changes of the phase of the order parameter along the valley of the energy landscape, the Higgs mode leads to variations of the amplitude of the order parameter. In this work, the Higgs mode has been excited with a radio frequency field, which modulates the order parameter by coupling to an unoccupied third spin state. Varying the RF-frequency led to the spectroscopic observation of the mode. [Nature Physics 14, 781-785(2018)]

Implementing the new coil for fast magnetic quenches, we aim to observe the Higgs mode in the time domain to characterize its decay on the BEC side, where it is not protected by particle-hole symmetry. Moreover it will be interesting to reveal the fate of the Higgs mode in the crossover to the quasi-2D regime, due to the absence of true long range order.

Machine Learning Based Detection of Phase Transitions

Determining the critical temperature of the superfluid transition in the BEC-BCS crossover regime is very challenging, as strong interactions greatly affect the observables of the system and also make numerical simulations very difficult. Experimentally, the situation is complicated since the celebrated bimodal momentum distribution of the pairs, that would signal the occupation of the ground state, cannot be observed for weakly bound pairs which break during ballistics. So far, one way around this problem was to convert weakly bound pairs to tightly bound molecules before time-of-flight. The effects of this conversion, however, might present a systematic disturbance for measurements that are very sensitive to interactions. In our work, we aim to improve our understanding of the BEC-BCS crossover and our detection capabilities by applying novel data analysis methods based on machine learning and deep neural networks.

In order to detect the condensate fraction within the momentum distribution of an ultracold Fermi gas, we train a deep convolutional neural network. As input, the network receives a time-of-flight picture of the gas, which, in principal, should contain features of the pairing imprinted onto the momentum distribution. After training, the network allows us to predict the critical temperature of a sample from the momentum distribution of the fermions rather than the Cooper pairs.

Phase Diagram Machine Learning
© AG Quantum
© AG Quantum

Student Projects

Optical Trap for Ultracold 2D Fermi Gas, Masters Thesis by Andreas Kell

A gas can be effectively two-dimensional if the atoms have not enough (e.g. thermal) energy to occupy any excited state along one axis, because then all dynamics will take place only in the perpendicular plane. This can be realized by applying a much stronger confinement of the atoms along that axis. To implement this, a repulsive laser beam in a TEM01 mode is generated with a phase plate. Such a beam has no intensity along the horizontal plane, but creates a harmonic potential in vertical direction. The image below shows the intensity of the green laser in the focus and a trapped atom cloud in red (this is a false color overlay of two separate images of the laser and the atom cloud). The weaker confinement in radial direction stems from the curvature of an applied magnetic field.

© Andreas Kell

After comparison to other approaches and consultation with the team, the TEM01 beam was generated and characterized in a test setup. Shortly after the end of the project the setup has been incorporated to the main experiment and the transition to the quasi-2D regime has been observed with a cold cloud of fermions.

Acousto-Optically Sculptured Potentials, Masters Thesis by Valentin Jonas

Acousto-optic deflection is a phenomenon in which a laser beam traversing an acoustic wave in a crystal gets partially deflected due to the absorption of phonons. In acousto-optical deflectors (AODs) this acoustic wave is created by feeding RF signals to piezoelectric elements positioned at one side of the crystal. As the deflection angle of the beam corresponds to the momentum of the phonons, it can be controlled by changing the frequency of the RF signal. Furthermore, the deflected intensity can be altered via the applied RF power. In this project, two orthogonally placed AODs are used to achieve 2D spatial control of the deflected beam(s). Using an arbitrary wavefunction generator, two multitone RF signals are programmed and fed into the AODs. The amplitudes, phases and frequencies of the signals can be controlled, which enables the construction of static and dynamic optical potentials, like arbitrarily spaced 2D optical lattices with movable rows and columns or the drawing of different 2D shapes like circles or Lissajous figures. One of the challenges in doing so are non-linear effects in the AOD crystal and the amplifier leading to intensity loss and irregularities in the deflected intensities. The programming of the function generator, the investigation of these effects as well as the intensity stabilization of the deflected beams were done as Bachelor and part of a Master thesis. In the future, the creation of optical tweezer arrays which allow for tunneling between the individual lattice sites might allow for the simulation of 2D superconducting layers in high temperature superconductors (like the CuO-planes in YBCO) with a high degree of control of lattice defects and spacing. Another application might be the measurement of the critical velocities of the created superfluids in the experiment via laser stirring as a means to further characterize the state of the quantum gas.

© Valentin Jonas

Design and Construction of a Multi-Species Effusive Atomic Oven for Lithium and Sodium, Masters Thesis by Daniel Eberz

To perform research on ultracold atomic gas experiments, an atom source is required. A popular choice for said source is an atom oven, which contains a sample of the desired atom species in heated reservoirs. In our case this oven must provide simultaneous flux for both lithium and sodium, which is accomplished in a dual-species design (see picture). 

The goal of this masters thesis was to design, build and characterize an improved version of the oven chamber. Major improvements on the oven comprise a new design of the water-cooled aperture, the rotary beam shutter, a new cup to accumulate residue and several new windows. In total, the new design allows for a much lower rate of maintenance and better accessibility for troubleshooting, while maintaining a high flux and compatibility with the subsequent experiment.

In the end, the atomic oven yields a collimated beam of both Na and Li (see picture), which are guided towards the experimental chamber. The shape and flux of the beam is determined by the geometry of the nozzles and the distinct temperatures of both reservoirs. To guarantee proper functionality of the oven, an absorption spectroscopy (see picture) measurement has been done solely on sodium at the exit of the chamber. This measurement confirms the excepted flux and profile of the beam.

© Daniel Eberz


Avatar Eberz

Daniel Eberz

Wird geladen