Liu et al. [Phys. Rev. B 98, 241109 (2018)] used Monte Carlo sampling of the physical degrees of freedom of a Projected Entangled Pair State (PEPS) type wave function for the S = 1/2 frustrated J1-J2 Heisenberg model on the square lattice and found a non-magnetic state argued to be a gapless spin liquid when the coupling ratio g = J2/J1 is in the range g ∈ [0.42, 0.6]. Here we show that their definition of the order parameter for another candidate ground state within this coupling window-a spontaneously dimerized state-is problematic. The order parameter as defined will not detect dimer order when lattice symmeties are broken due to open boundaries or asymmetries originating from the calculation itself. Thus, a dimerized phase for some range of g cannot be excluded (and is likely based on several other recent works).
I. OVERVIEWIn a recent Rapid Communication [1], Liu et. al. argued that there is a gapless spin liquid phase in the ground state of the S = 1/2 frustrated square-lattice J 1 -J 2 Heisenberg model for g = J 2 /J 1 ∈ [0.42, 0.6]. At variance with other recent works [2-5], they found no spontaneously dimerized valence-bond-solid (VBS) phase within this range of coupling ratios (where other works have roughly placed the VBS at g ∈ [0.52 − 0.61]). They reached their conclusions based on the method of Monte Carlo sampling of gradient-optimized tensor network states [6-8], which they have further refined for the specific case of a tensor network of the Projected Entangled Pair State (PEPS) type. Open-boundary lattices with up to 16 × 16 spins were used, and, taken at face value, the results appear to be well converged and reliable.In this Comment we point out that the definition of the VBS order parameter used by Liu et al. has a potential flaw and may not capture long-range order correctly on the open-boundary lattices considered. The squared columnar VBS order parameters for x and y oriented dimers, m 2 dx and m 2 dy , were defined in Eq. (2) of [1] as follows (up to typographical errors):where α = x, y, B α r = S(r) · S(r +α) is the bond operator along the α direction, and N b is the number of bonds summed over. The wave-vector corresponding to columnar order is q α = (π, 0) and (0, π) for α = x and α = y, respectively. We can rewrite this squared order parameter in the equivalent form:whereThe problem with the definitions is the subtraction of the nonuniform B α r B α r in Eq.(1) or D α 2 in Eq.(2) when longrange order is induced by some symmetry-breaking mechanism, e.g., with certain open lattice boundaries or some imperfection in the method used. In essence, the baby is then thrown out with the bath water.We will demonstrate this problem by considering a columnar VBS state which is four-fold degenerate on periodic L×L lattices with even L. The ground state is uniform in the absence of some symmetry-breaking mechanism, and the subtracted term D α 2 in Eq.(2) vanishes. However, on rectangular lattices with L x × L y spins (even L x and L y ) the ground state is unique and hosts a specific dimer pattern...