Particle doesnt lose energy.

Hello everyone. I’m trying to simulate particle penetration based on sputter example.
I’m using Tersoff/ZBL potential. I’m inserting Si particle with ~12 keV initial energy. As you can see from the dump file, after penetrating down to 30 nm, particle loses less than 1 keV of energy. But according to MC simulations it should have already stopped, so apparantly something is going wrong. I’ve attached input, Log and dump files to this message.

in1.sputter (990 Bytes)

kin.xyz (6.44 KB)

log.lammps (8.19 KB)

2014-10-23 16:10 GMT+02:00 Rumiantzew Alexander <[email protected]>:

Hello everyone. I'm trying to simulate particle penetration based on
sputter example.
I'm using Tersoff/ZBL potential. I'm inserting Si particle with ~12 keV
initial energy. As you can see from the dump file, after penetrating down
to 30 nm, particle loses less than 1 keV of energy.

judging from you files, the thickness of you slab is only about 300 A (0.3
nm). I don't know the exact physics of your system, but maybe you mixed
your unit conversion up

Kristof

300 A = 30 nm not 0.3 nm, so it must not be the problem .

How can you be sure that MC is right and your MD is wrong? Or the other way around, how can you be sure that the particular potential (Tersoff/zbl) and its parameters can satisfactorily reproduce keV Si ion implantation?

Ray

Because both MC and MD use zbl potential. I’ve chosen Si as projectile example of the problem not to mess up tersoff portion of the potential.
I have the same problem with my real setup (not si on si implantation), but i decided to make separate example to rule out the possibility that i have modified tersoff/zbl potential wrong for different elements (i’ve took parameters from the article) and use the potential file provided with distribution.

Although your script looks good to me, there isn’t any rule that MC and MD results should agree. Maybe you can try an even thicker sample.

Ray

"looking good" is not quite what i would say. it has no gross errors.
but an "equilibration" period of 30fs with a thermostat time constant
of 10ps is not exactly clean.

also, how is it "good" to determine the outcome of a stochastic
process (particle hitting surface at randomized location) from a
single observation?

it is also not clear how comparable the MC reference is to what the MD
input does. how does the MC simulation treat the freely moving
particle?
considering that this MD example has fundamental stochastic issues, it
also casts doubt on the validity of the the MC run.

axel.

Thats not a single observation. I’ve made a dozen runs and problem occured in each. If i extend the sample even further particle still doesnt lose energy. But if i slightly tilt the incident particle direction (or even make the lateral dimensions of sample smaller so that projectile cross the periodic boundary when in the sample), energy is lost in consistent way. I dont know if that “channeling” is expected behaviour (i’ve never seen such references in literature ) or some artifact which is looks like.

Thats not a single observation. I've made a dozen runs and problem occured
in each. If i extend the sample even further particle still doesnt lose

if you don't change the random number seed they are all the same.

energy. But if i slightly tilt the incident particle direction (or even make
the lateral dimensions of sample smaller so that projectile cross the
periodic boundary when in the sample), energy is lost in consistent way. I
dont know if that "channeling" is expected behaviour (i've never seen such
references in literature ) or some artifact which is looks like.

thus i maintain, that what you see is an artifact of your simulation
setup. LAMMPS does what you ask it to do.