Ensemble choices for MSD calculation

Dear my friends,
I want to calculate the diffusion coefficient of hydrogen in water at infinite dilution. My method is based on the example in LAMMPS files. First, I use fix NVE and fix langevin to equilibrate the whole system. Then I use fix NVT to control the temperature. And I use fix NVE to compute MSD.
But from the results, in NVE ensemble, the temperature fluctuate a little wildly. For example, when I set 1000K, the temperature may go down most to 940K. And my work is focusing on a series of temperature setting with 700,750,800,850…,1150 and 1200K. What is confusing is that the diffusion coefficient at 1050K is lower than 1000K at same system density. I think this is because the temperature fluctuation in NVE ensemble. Thus I cannot accept this temperature fluctuation. Now I think of two solutions.

  1. Use fix temp/rescale with fix NVE to control the temperature precisely. But I learned from mail-list that this is not usually recommended.
  2. Use fix NVT to calculate MSD rather than fix NVE. It may require a bigger Tdamp to decrease temperature fluctuation.
    Can you give me some advice about these two solutions? Or any other suggestion?
    Thanks a lot.