LFHT is possible without fully estimating distributions when m ≫ 1/ε2, revealing a trade-off between sample sizes n and m.