| Abstract [eng] |
The unusual properties of quantum materials, such as superconductivity, arise from how particles behave together. Calculating this collective behaviour can overwhelm even the most powerful computers. Neural networks offer a possible shortcut by learning compact descriptions of quantum systems. This thesis puts that promise to the test. It investigates when these descriptions succeed, why they fail, and how to improve them. It also develops two new methods that improve how neural networks learn and how the computer picks which particle arrangements to check, enabling accurate results in cases where existing approaches stall. Together, these findings help researchers judge when neural-network calculations can be trusted and build more reliable tools for understanding quantum matter. |
