Use FLT_MIN as padded value for the smoothing length to prevent fake particles being in range when vec_fmax(hig2,hjg2) is used to set the smoothing length.