Thermal fluctuation while heating of an alloy

LAMMPS LAMMPS General Discussion
1

I am heating an Al-Si-Mg alloy using npt ensemble from 300 K to 1000 K. For certain composition of the alloy, the temperature fluctuation above 650 K becomes intolerable (~ 0 K - 1800 K). What is the source of such noise? Can you kindly suggest me how to control the temperature? [the input script for LAMMPS simulation is given below]

units metal
atom_style atomic
dimension 3
boundary p p p
read_data system.data
pair_style meam
pair_coeff * * library.meam AlS SiS MgS AlSiMg.meam AlS SiS MgS
timestep 0.0002
thermo 1000
dump 1 all xyz 1000 a*.xyz
velocity all create 300 82557 dist gaussian
compute msd all msd com yes
compute 2 all pe/atom
compute 3 all entropy/atom 0.25 5. local yes
dump 2 all custom 50000 entropy.data x y z c_2 c_3

compute myRDF all rdf 50 1 1 1 2 1 3
fix 3 all vector 1000 c_msd[4]
fix 4 all ave/time 100 1 1000 c_myRDF[*] file tmp1.rdf mode vector
thermo_style custom step temp vol press density ke pe etotal enthalpy c_msd[4]
restart 10000 data.restart

fix 1 all nvt temp 300 300 0.005 drag 0.01
run 10000

unfix 1
fix 1 all npt temp 300 1000 0.001 iso 0 0 2
run 200000

unfix 1
fix 1 all nvt temp 1000 1000 0.005 drag 0.01
run 20000
unfix 1
fix 1 all npt temp 1000 853 .005 iso 0 0 2
run 735000

unfix 1
fix 1 all nvt temp 853 853 0.005 drag 0.01
run 1000000

2

Hi @Imukherjee.

Please, have a careful look at the manual pages of the fix nvt and units commands.

There is a detailed note on the effect of the Tdamp parameter on temperature convergence and the value you use are way out of range of standard rule-of-thumb values compared to your timestep. Please also see the section regarding the tchain parameter for solids. As a side note, the use of the velocity all create command without a velocity all zero linear is very likely to give you a non zero total linear momentum which can also affect thermal/pressure equilibration.

Your issues can stem from other causes in your systems but these are already independent problems to address. I highly doubt your successive NVT/NPT* are actually done on equilibrated systems what so ever. But without more information concerning your output, this last statement is just a guess on my part.

* Please note, brutal ensemble switching between NPT and NVT are not a good way to proceed. Please see the relevant literature concerning such equilibration procedures.

3

There always are (local) fluctuations in finite systems. How large the global fluctuations are depend on multiple factors: choice of force field, timestep, initial geometry, system size.

The most straightforward way to reduce fluctuations is to increase the system size.