NVT simulation and energy balance

I have a question about the energy balance in the case of NVT simulation.

I am simulating fracture of graphene using the REBO potential on a sheet of about 70 000 atoms under periodic boundary conditions in the two in-plane directions and fixed in the third direction. time step is 1fs, temperature 0.1K. units are metal.
The simulation consist in successively increasing the x size of the simulation box by 0.01% and running an NVT simulation for 20 000 time steps. (Doing so, I simulate an monoaxial strain tensile test at a strain rate of 5e^-9 fs^-1.

My question concern the energy balance under NVT conditions.
I customized the thermo style with f_NVT so that the column “etotal” (total energy) includes the energy given to the thermostat. Moreover I know that by computing the difference “etotal-f_NVT” I get the total energy (internal energy) of the system regardless of what have been given to the thermostat.

— Interesting lines
variable TimeStep equal 0.001 #timestep 1fs
variable Temp equal 0.1 #temperature
variable Tdamp equal 0.1 #Tdamp de Nose-Hoover 100fs

NVT

fix NVT all nvt temp {Temp} {Temp} ${Tdamp}
fix_modify NVT energy yes

#Define thermo parameters in log file
thermo ${ThermalFreq}
thermo_style custom step time lx ly temp pe ke etotal pxx pyy pxy pzz f_NVT

Dear Laurent,

Energy nonconservation is very possible at the fracture simulations because of rapid structural change and change in particle motion spectra.

This happens even when no thermostatting is applied.

Integration with a shorter timestep may improve the energy conservation.

Regards,

Vasily Pisarev