Help in crack propagation

Hello Dr Steve, Dr Axel and everyone.

I am trying to model crack propagation in a nanocomposite. I used NVE + NPT for equilibration and applied tensile displacement on the boundaries.

A] However, my crack fails to propagate and the structure breaks near the boundary.

I suspect the possible reasons and tried debugging them but couldn’t figure out what was wrong, thus seeking your expert guidance.

  1. My fixes are wrong or overlapping
  2. My strain rate is too high. (I tried lowering it but still getting the same behavior)
  3. Interactions are not correctly defined
  4. Equilibrium state is not achieved.

B] For the same simulation once I get everything correct, how do I track the crack opening displacement?

Code Link: https://we.tl/5sb7132zp7

Please assist. Thanks in advance!

Regards,
Jane

Dear Dr Shan,

Thank you for your inputs. After equilibration, the gap between iron and graphene layers have decreased slightly (as expected).

  1. I did not see any warning where more than one atom is integrated than once, however I had a warning which said “building an occasional neighbor list when atoms have have moved too far”

  2. I gave a lot of thought, read several papers and decided to use these parameters. What would you recommend instead of this ?

  3. I increased the equil time and am still waiting for the results.

I am still not able to figure out why the crack isn’t propagating.

Regards,
Jane

Crack wasn’t propagating could be due to your structure and your force field parameters. Is it really physical to have a layer of graphene sheet between two bulk bcc iron? Is the eam/fs you are using the best potential to describe and predict iron cracking? Is your choice of hybridizing eam, lj/cut and airebo the best potential? Just off the top of my head, there is a MEAM potential suitable for Fe-C systems – but effort needs to be done to make sure it is suitable for cracking.

I suggest you read the literature and talk to your advisor, which this mailing list can not replace.

HTH,

I tried running the code only for Fe slab with eam/fs potential , yet the crack fails to propagate… Thus, I believe that something is wrong with my boundary conditions.

I will definitely discuss about the issues you pointed out with my advisor however, I still think there’s something wrong with my code besides interactions.

I have spent weeks in trying to debug but have failed miserably. Any other recommendations?

Thanks.

I tried running the code only for Fe slab with eam/fs potential , yet the
crack fails to propagate.. Thus, I believe that something is wrong with my
boundary conditions.

I will definitely discuss about the issues you pointed out with my advisor
however, I still think there's something wrong with my code besides
interactions.

I have spent weeks in trying to debug but have failed miserably. Any other
recommendations?

jane,

please note that scientific discovery is an activated process and thus
the amount of time you spend on it is (mostly) irrelevant, if you are
not improving your understanding of what is going on, and thus
flattening the barrier that you have to overcome. a statement like "i
think that something is wrong with my input" and then waiting for
somebody else to point it out to you, is not going to get you much
closer. what you will have to develop is a strategy to eliminate
possible causes of problems. one way is to construct simple/trivial
input examples, where the expected problem is *definitely* present (or
not) and then study how it manifests itself.

when i look over you input, it looks to me like you've patched it
together from different pieces you have seen somewhere and you are
using them without really having thought through what they do. this
concerns:

- your force field setup
- your neighbor list setup
- your thermostat choices
- your boundary conditions
- your data analysis computes

some pointers that you need to consider and look into:

- ray already pointed out, that your hybrid setup needlessly uses two
instances of the same force field and also voiced doubts about
combining AIREBO with EAM via lj/cut. i have to support that strongly.
part of EAM is the "embedding energy" term, which is dependent on the
electron density of surrounding atoms. this is fully neglected in your
setup, so the metal at the boundary to graphene will behave as if it
is a free surface (which is not true).

- your neighbor list skin is typical for reduced units and very small
for metal units. also a delay of 10 timesteps is huge. what saves you
is that you have solids.

- temp/rescale is bad, bad, *bad*! strangely enough, you switch to
langevin at a later point. why not use that right away.

- the combination of immobile (and stepwise displaced) boundary atoms
and all periodic boundaries doesn't make much sense. have you
visualized your input with periodic replication of the principal cell?
this should give you a visual hint of why things don't work as
expected.

- why do you use 6x compute reduce, when it would be sufficient to use it once.

some of this is obvious, when you read the documentation for each
command that you use carefully; some of this becomes obvious when you
change your focus from "i'll try anything just to make it work" to "i
need to be certain of every step i do and what each command means";
some if this requires studying the physics of the underlying theory.
if you are new to the discipline this may take significant time (i.e.
of the order of months rather than weeks to grasp sufficiently well).

the fact, that you tried with an iron-only system is the right
approach. you have to make that one work before you can even consider
moving to something more complex.

axel.

Dear Dr Axel,

Thank you for your deep insights. I will look into it again and revert back after having further clarity and implement your recommendations.

Thanks & Regards,
Jane

Dear Dr Axel and Dr Shan,

I am looking at the other details as you both suggested , meanwhile I wanted to show you what is happening at the boundaries . Link to image : https://we.tl/mfW3xiXuIO The boundary group is being separated from the mobile group which is stopping the crack to propagate. I believe this has to do with boundary conditions. [Same thing was happening with only iron simulation so I eliminated the cause of wrongly defined interactions in that case]

I have read about different ways [erate, addforce, displace_atoms, setforce, velocity] on lammps manual and your comments in archive and I established that controlled displacement loading is the best approach for my simulation.

Am I missing something here?

I really appreciate how you all patiently respond to the queries of graduate students and we all couldn’t thank you all enough!

Thanks again,
Jane

Jane,

Have you followed Axel’s advice on changing to the correct boundary condition? Take a look at the YOUR_LAMMPS_DIR/examples/crack/in.crack input file for an example on how to set the correct boundary.

Before experimenting with a complicated Fe-C-Fe trilayer system, please run a crack simulation of pure Fe with your eam potential. Make sure you correctly understand how to set up the simulation and have all analyses and dump commands in place for analysis. Only after then switch to your trilayer system. Note that I still have doubts on your choice of potentials, which you should talk to your advisors about.

Ray

Dear Dr Axel and Dr Shan,

I am looking at the other details as you both suggested , meanwhile I wanted
to show you what is happening at the boundaries . Link to image :
WeTransfer - Send Large Files & Share Photos Online - Up to 2GB Free The boundary group is being separated from the
mobile group which is stopping the crack to propagate. I believe this has to
do with boundary conditions. [Same thing was happening with only iron
simulation so I eliminated the cause of wrongly defined interactions in that
case]

I have read about different ways [erate, addforce, displace_atoms, setforce,
velocity] on lammps manual and your comments in archive and I established
that controlled displacement loading is the best approach for my simulation.

Am I missing something here?

what you are missing is that you are not approaching this problem in a
scientific way. instead you only keep writing about what you
hypothesize might be the cause of the problem, but you are not showing
any systematic effort to prove or disprove your hypothesis. simply
reading about flags isn't enough. you will not magically come across a
paragraph that will tell you exactly what to do for your specific
case. rather, you have to identify the individual components that make
up your simulation and for each of them determine whether you are
using it correctly.

I really appreciate how you all patiently respond to the queries of graduate
students and we all couldn't thank you all enough!

at this point, i have to say that patience is running thing, because
we keep pointing things out to you, but you are not acting on it.
instead you keep just repeating the same old stuff. if you don't see
an obvious solution, simplify the problem until you understand it.

axel.