Fig. 3: Results for the 1D t − XXZ system with 20 sites and J± = 1, Jz = 4 and t = 8.
From: Neural network approach to quasiparticle dispersions in doped antiferromagnets
a Quasiparticle dispersion for a single hole obtained with the recurrent neural network (RNN, red markers), compared to exact energies from exact diagonalization(ED, light red lines) and the combined spinon and holon dispersions from Eq. (2) (gray). We average the RNN energy over the last 100 training iterations, each with 200 samples, with the standard deviation denoted by the error bars. We show the exact low-energy excited states as well. b Relative error \(\Delta \epsilon =\frac{{E}_{{{{{{{{\rm{RNN}}}}}}}}}-{E}_{{{{{{{{\rm{ED}}}}}}}}}}{| {E}_{{{{{{{{\rm{ED}}}}}}}}}| }\) during the ground state training. a and b are obtained using a 1D RNN architecture with a hidden dimension dh = 100.
