We discuss two different approaches for splitting the wavefunction of a single-particle-box (SPB) into two equal parts. Adiabatic insertion of a barrier in the center of a SPB in order to make two compartments which each have probability 1/2 to find the particle in it is one of the key steps for a Szilard engine. However, any asymmetry between the volume of the compartments due to an offcenter insertion of the barrier results in a particle that is fully localized in the larger compartment, in the adiabatic limit. We show that rather than exactly splitting the eigenfunctions in half by a symmetric barrier, one can use a non-adiabatic insertion of an asymmetric barrier to induce excitations to the first excited state of the full box. As the barrier height goes to infinity the excited state of the full box becomes the ground state of one of the new boxes. Thus, we can achieve close to exact splitting of the probability between the two compartments using the more realistic non-adiabatic, not perfectly centered barrier, rather than the idealized adiabatic and central barrier normally assumed.

Machine learning techniques based on artificial neural networks have been successfully applied to solve many problems in science. One of the most interesting domains of machine learning, reinforcement learning, has natural applicability for optimization problems in physics. In this work we use deep reinforcement learning and Chopped Random Basis optimization, to solve an optimization problem based on the insertion of an off-center barrier in a quantum Szilard engine. We show that using designed protocols for the time dependence of the barrier strength, we can achieve an equal splitting of the wave function (1/2 probability to find the particle on either side of the barrier) even for an asymmetric Szilard engine in such a way that no information is lost when measuring which side the particle is found. This implies that the asymmetric non-adiabatic Szilard engine can operate with the same efficiency as the traditional Szilard engine, with adiabatic insertion of a central barrier. We compare the two optimization methods, and demonstrate the advantage of reinforcement learning when it comes to constructing robust and noise-resistant protocols.

We have made a simple and natural modification of a recent quantum refrigerator model presented by Cleuren et al. in Phys. Rev, Lett. 108, 120603 (2012). The original model consist of two metal leads acting as heat baths, and a set of quantum dots that allow for electron transport between the baths. It was shown to violate the dynamic third law of thermodynamics (the unattainability principle, which states that cooling to absolute zero in finite time is impossible), but by taking into consideration the finite energy level spacing in metals we restore the third law, while keeping all of the original model's thermodynamic properties intact.

In any general cycle of measurement, feedback, and erasure, the measurement will reduce the entropy of the system when information about the state is obtained, while erasure, according to Landauer's principle, is accompanied by a corresponding increase in entropy due to the compression of logical and physical phase space. The total process can in principle be fully reversible. A measurement error reduces the information obtained and the entropy decrease in the system. The erasure still gives the same increase in entropy, and the total process is irreversible. Another consequence of measurement error is that a bad feedback is applied, which further increases the entropy production if the proper protocol adapted to the expected error rate is not applied. We consider the effect of measurement error on a realistic single-electron box Szilard engine, and we find the optimal protocol for the cycle as a function of the desired power P and error ε.

The rising significance of organic light emitting diodes as lighting devices puts their peripheral devices into focus as well. Here, we present an organic optoelectronic device allowing for multistable luminance and emission color control. The introduced device is monolithically built up from organic resistive switching elements processed directly on top of a polymer light emitting diode (PLED). This realization, representing a serial connection, allows for precise control of the voltage drop across and thus the current density through the PLED resulting in a control of its luminance. Additionally, by using a fluorescence-phosphoresence host-guest blend as the light emitting layer, it is possible to tune the emission color in the same way. Specifically, focus was set on color temperature tuning in a white light emitting diode. Notable, for all different luminance and color states, the driving voltage is constant, enabling, e.g., a conventional battery as power supply.