Different ave/chunk setting change the dynamic process?--rnemd setting

Dear all,
I met some questions when i used the reverse-nemd method to compute the thermal conductivity of amorphous polymer.The box is orthogonal.And i use the stable lammps version (12 Dec 2018).
Here i list some commands in my input file:

kspace_style pppm 1e-7

run_style respa 2 20 #dt 2fs
compute ke all ke/atom
variable temp atom c_ke*{kcal2j}/1.5/{kb}
compute layer all chunk/atom bin/1d z lower 0.02 units reduced #chunk

fix 1 all nve
fix 2 all ave/chunk 1 1000 1000 layer v_temp file tmp_running_100_50_0.1_2_1e-7_1_t.profile ave running
variable tdiff equal f_2[26][3]-f_2[1][3]
fix 3 all thermal/conductivity 100 z 50 swap 1 #velocity swap

thermo_style custom step temp press etotal f_3 v_tdiff

run 2500000 #5ns

fix 2 all ave/chunk 1 1000 1000 layer v_temp file tmp_window_100_50_0.1_2_1e-7_1_t.profile ave window 250
fix 3 all thermal/conductivity 100 z 50 swap 1 #velocity swap

run 1250000 #2.5ns

The first question is that when i just changed the command line in the input file from
fix 2 all ave/chunk 1 1000 1000 layer v_temp file tmp_running_100_50_0.1_2_1e-7_0_t.profile ave running
fix 2 all ave/chunk 1 1000 1000 layer v_temp file tmp_window_100_50_0.1_2_1e-7_1_t.profile ave window 250
to
fix 2 all ave/chunk 10 100 1000 layer v_temp file tmp_running_100_50_0.1_2_1e-7_10_t.profile ave running
fix 2 all ave/chunk 10 100 1000 layer v_temp file tmp_window_100_50_0.1_2_1e-7_10_t.profile ave window 250 ,the corresponding log file change from(here only list some steps)

Step Temp Press TotEng f_3 v_tdiff
17500000 296.90211 -337.72528 365.01821 0 0
17501000 300.98308 17.415207 365.7877 37.259272 4.578336
17502000 299.59007 87.718874 364.74202 73.023286 17.922393
17503000 300.26692 -262.06341 363.42155 111.40938 24.945059
17504000 300.12886 -6.3961086 364.82343 148.75104 30.719492
17505000 303.18384 274.6431 364.05325 180.84007 38.305601

to

Step Temp Press TotEng f_3 v_tdiff
17500000 296.90211 -337.72528 365.01821 0 0
17501000 300.9831 17.410152 365.78773 37.259272 3.5861204
17502000 299.53688 75.587576 364.74171 73.01134 17.112288
17503000 304.86453 -123.27919 362.01516 106.88022 25.757288
17504000 297.07845 -510.47247 366.2634 146.19557 29.026208
17505000 301.13917 -324.03912 365.5218 182.10188 30.348852

And it seems that the different setting of the fix ace/cunk command influence the dynamic simulation(thought only a slightly difference of the total energy during the all 7.5ns’s simulation).After carefully checking of the fix ave/chunk manual,i think that the fix ave/chunk command only do some data average and should not have any influence on the dynamic process,am i right?And i noticed that a former email also put out a similar question.
https://sourceforge.net/p/lammps/mailman/message/36448033/

Also there is another peculiar phenomenon that when i set
fix 3 all thermal/conductivity 100 z 50 swap 1 ,pppm 1e-6,the total energy is not well conserved.But when i set
fix 3 all thermal/conductivity 100 z 20 swap 1,the total energy seems to conserve better.Is there some probable reasons for this?

I also have some questions about the energy conservation when using nemd method(rnemd included) to compute the thermal conductivity.After doing some test , i find that the NEMD require a much smaller timestep than the EMD method to meet the energy conservation while the absolute temperature difference when the NEMD simulation reachs the stable state is not that large(just about 100K),and it should not lead to a much smaller timestep for the energy conservation purpose.Are there some reasons for using such a small timestep?
Another question is about the pppm setting.After checking the mail list , i find that the default setting pppm diff ik is not designed for the energy conservation purpose,it only conserve the momentum,and the diff ad method conserves the total energy but can not be applied to the triclinic system . So i think maybe the only way to better conserve the total energy for a triclinic system is to use higher accuracy in the pppm command,and this lead to a unaffordable time consuming (maybe the real space cutoff is not set right while raise the accuracy,and i really do not have any setting experience ).Are there some other ways to conserve the energy better in a triclinic system ?
Best regards,
xiaonengRan

The previous mail message you linked to was a bug which was fixed and should be in the 12Dec18
version. I can’t think of any reason why fix ave/chunk should affect the dynamics.

Re: NEMD energy conservation, I don’t see what that has to do with triclinic boxes. NEMD
for thermal conductivity usually refers to maintaining two temperatures, the heat flow is the NEMD part.
If you can do EMD in an orthogonal box, you can do NEMD in one as well.

Steve