Fig. 2: Unsliced costs, sliced costs and slicing overheads for various tensor networks studied in this paper.
From: Efficient parallelization of tensor network contraction for simulating quantum computation
Each box represents the lower (Q1) to upper (Q3) quartiles of contraction costs over ten independent runs of the algorithm, with a horizontal line that represents the median. The whiskers indicate the highest and lowest contraction costs/overheads that are not outliers, where outliers are defined as data points whose distance to the nearest quartile is larger than 1.5 times the interquartile range. a, Tensor networks for evaluating a batch of 64 amplitudes in Sycamore random circuits. b, Tensor networks corresponding to 2 + 1 rounds of syndrome extraction for the Surface-17 code. c, Tensor networks associated with edges of the Cai–Fürer–Immerman (CFI) graphs. For each graph and each QAOA depth, there are two pairs of unsliced/sliced costs for the two isomorphic classes of edges in that graph: the left pair corresponds to the first class, whereas the right pair corresponds to the second class (see Fig. 4).
