Fig. 3: Gradient variance scaling for SU compound layers, observable not in Lie algebra.
From: Characterizing barren plateaus in quantum ansätze with the adjoint representation
The setup is identical to the Lie algebra supported ansatz (LASA) case, except that here the measurement operator is a single σz/4. We show the analytical prediction derived from the LASA case as explained in the text, and therefore we see a disagreement with numerics, implying that the covariance term is nonzero. Still, the scaling is similar, and additionally, the numerics converge to the prediction at larger system sizes. The error bars are too small to plot. Additional information on the numerics is in the Supplementary Information.
