The sound of a violin has loudness and pitch. If you hear only the loudness, the melody is lost. In the same way, an electron in the quantum world is described by two parts: an amplitude (how strong the wave is) and a phase (where the wave sits). In moiré materials, where two layers are slightly twisted, electrons interact in complex ways, and a magnetic field adds even more detail. A conventional neural network trained to describe such systems often sees only the amplitude and misses the phase.
The new Complex Psiformer network is built to take both parts into account. Inside it, the hidden features are complex, and attention works with complex messages. It is as if a musician could hear not only the loudness but also the pitch of every note. Thanks to this, the network tunes the electron wave function more accurately.
The comparison showed that the complex network reaches a lower energy, which means it describes the real behavior of electrons better. This helps scientists model quantum materials and understand how current flows in such systems. Previously it was unclear whether complex numbers offer an advantage, and now it has become clear that they do.
In the 25-cell system, the average gain of Complex Psiformer over Real Psiformer is 1.458 meV per electron, roughly 14 times smaller than the room-temperature thermal energy of 25 meV. Such small energy differences are nevertheless reproducibly stable across five independent runs.
The idea of a neural network that learns to 'see' a quantum phase echoes the image of self-learning quantum computers in Greg Egan's novel Schild's Ladder, where states of matter are computed as variational approximations.