Wagner, Holian, Massively Parallel Molecular Dynamics Simulations of Two-Dimensional Materials at High Strain Rates

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

91

Massively Parallel Molecular Dynamics Simulations of


Two-dimensional Materials at High Strain Rates

Norman J. Wagner
Department of Chemical Engineering
University of Delaware
Newark, DE 19716
and
Brad Lee Holian
Los Alamos National Laboratory
Los Alamos, NM 87545

ABSTRACT

Large scale molecular dynamics simulations on a massively parallel computer are performed to
investigate the mechanical behavior of 2-dimensional materials. A model embedded atom many-
body potential is examined, corresponding to "ductile" materials. A parallel MD algorithm is
developed to exploit the architecture of the Connection Machine, enabling simulations of > 106
atoms. A model spallation experiment is performed on a 2-D triagonal crystal with a well-
defined nanocrystalline defect on the spall plane. The process of spallation is modelled as a
uniform adiabatic expansion. The spall strength is shown to be proportional to the logarithm of
the applied strain rate and a dislocation dynamics model is used to explain the results. Good
predictions for the onset of spallation in the computer experiments is found from the simple
model. The nanocrystal defect affects the propagation of the shock front and failure is enhanced
along the grain boundary.

Introduction

Recent attention has focused on understanding the molecular mechanisms that give rise to the
macroscopic failure process [1]. Unlike many equilibrium material properties which can adequately
be represented by suitable simulations of hundreds or thousands of atoms, fracture and failure
involve both microscopic and macroscopic length scales. For example, crack propagation or stress
concentration near a crack tip inherently depends on both the molecular properties of bonding and
atomic arrangement, and the macroscopic properties of the crack length and crack tip geometry.
Further, materials with multiple levels of structure, such as polycrystalline or nanocrystalline
solids, require large ensembles of atoms to accurately represent these mesoscopic structures.
Such considerations have motived the development of a molecular dynamics simulation tool on a
massively parallel computer for investigating material behavior at the molecular level [2].
Through the use of molecular dynamics simulations we hope to elucidate the molecular mech-
anisms underlying material failure under spallation. This is motivated by two considerations:
1) detailed microscopic analyses are difficult to perform during spallation and 2) molecular dy-
namics is well suited to examine processes occurring on these short time scales. Simulations can
also provide detailed information concerning microscopic mechanisms for spallation at high strain
rates, about which little is known [3, 4]. In previous work, our simulation studies motivated the
development of a dislocation dynamics model for the spallation process. The relevant mechanical
process at the spall plane was suitably modeled as an adiabatic expansion. Consideration of
nucleation and aggregation of defects lead to a stress-activated statistical model for the spalla-
tion process, which was in good agreement with the spallation simulations. Preliminary work
on polycrystalline samples indicated that the presence of grain boundaries could affect the shock
propagation and the spallation behavior.
In this work, a molecular dynamics simulation of the spallation of a perfect crystal with a
single nanocrystal defect is performed and analyzed. A model embedded-atom potential, which
Mat. Res. Soc. Symp. Proc. Vol. 291. 01993 Materials Research Society
92

is suitable for metal-like materials, is used [5]. A scaled down version of a typical spallation ex-
periment is simulated, and the relevant molecular and hydrodynamic variables monitored during
the spallation process. The results of these tests are analyzed within the previously proposed
dislocation dynamics model.

Computer Simulation Technique

To address the physics of shock propagation in metals, a molecular dynamics computation


algorithm has been developed for the massively parallel SIMD Connection Machine0 (CM-200).
The details of the numerical calculation are explicitly reviewed in a previous publication [2] so
only a brief overview is given here. The atoms follow Newton's equations of motion, which are
simply integrated forward in time via a half-step Stoermer algorithm [6]. In time-discretized form
the equations of motion are:
Xt = X,-I + UAt (1)
Ut+1 = Ut_1 + FtAt (2)
where the postions X, velocities U, and forces F are evaluated for each particle in parallel.
The interatomic forces are treated as central forces, modeled as a combination of a Lennard-
Jones 6-12 with a spline cutoff and an analytic embedded atom many-body potential, as detailed
previously [7]. This many-body potential has been successfully employed as a model for ductile
metals.
The total potential can be written as:

v(r)+ pi) (3)

where rij is the center-to-center separation distance and Pi is the local density. The W 6-12 part
has the cut-off form:

4
PLj(r) = XE -- ( ,) r < rspt (4)

PLJ(r) = -(L2 rrmax - r2 + a3 (IMa -r rsp < r < rmar (5)


where c is the U well depth, y is a weighting parameter set to - here. a is the point the U
3
potential goes through zero, rspi is the point of zero inflection in the potential, and the other
parameters are given in a previous paper [2].
The many-body embedded atom (EAM) term has the analytic representation:

F[pi] = -d(d + L)e(1 - X)epi lnpi, r < rmax (6)


2
where d is the dimensionality and the local density has the form:
Pi 1-
iv(arij),

w(r) = d(d7 l)_ ax r2 1 r < rma) (7)

with ro- = 21/ 6 a. The parameter value V = 1/3 was taken for the EAM model quoted here, while
X= I returns the pure U material.
The computational algorithm, written in Fortran 8X, was run on the 64K-processor CM-200
at the Advanced Computation Lab of Los Alamos National Laboratory. The computational
performance on the CM was found to be approximately 2 microseconds per timestep per particle
for problems in the range of 105 - 106 particles.
93

Simulation Results

Spallation Experiments

A plane wave impact experiment was simulated to investigate the molecular aspects of spal-
lation. In this experiment, translational momentum is converted into irreversible work consisting
of heat and defect generation. The experimental geometry (Fig. 1) consisted of a flyer plate of
mass M moving with x-velocity 2v, impacting a target plate of mass 2M moving with x-velocity
-v, fixing zero center of mass motion and the particle velocity as U. = 2v. The total number of
actual atoms in the simulations was N 2 30,000. Periodic boundary conditions were used for the
y direction, while imsthe x direction enough free space was allowed around both plates so that no
communication occurred across the x-boundary during the simulation. The plates were cut from
a lattice and equilibrated before the start of the experiment, allowing the free surfaces to relax.
In earlier work, both single crystal and polycrystalline samples were simulated [2]. IHere, in the
interest of understanding the effects of grain boundaries on the spallation process, a spherical
section of a perfect triangular lattice was cut out of the target and rotated 7r/6 radians clockwise.
This was then annealed until an "equilibrium" nanocrystal containing approximately 1700 atoms
was formed as a single, large defect in the otherwise perfect crystal.
The initial plate velocities relative to the target (units: v/e7) were such that the ratios of
particle velocity to sound velocity ( Upl/co) were both in the range where the elastic material
response dominated the shock propagation in previous work [5]. Upon impact, the plates cold-
weld together and shock waves propagate outward from the impact point. These waves reflect
off the free surfaces, returning as a family of release waves that broaden due to nonlinear effects.
Assuming constant wave velocities, these release waves meet at a plane whose distance from the
right side of the target is equal to the flyer plate thickness. The spherical defect was initially
centered on this spall plane. For strong enough impact velocities, the material fails under this
resultant tensile stress and spallation occurs.
Three different initial velocites were chosen to explore the effects of shock strength otl the
failure mechanism in the presence of a nanocrystalline defect. As summarized in Table I, the
shock strengths varied from slightly below the expected strength for the onset of strong plastic
flows to a relatively weak shock. Figure 2 shows the state of the material just after the time
of expected spallation. Three qualitatively different behaviors were observed. For the weakest
shock (C), no significant damage is observed. Only small voids open along the grain boundary.
Increasing the shock strength leads to failure, as shown in experiment B. The voids open along the
spall plane in the bulk material but are "deflected" along the grain boundary when intersecting
the nanocrystal. The last frame (A) shows the material for the highest impact velocity studied
(this corresponds to run A of the previous work [2]). HIere there is damage along the entire spall
plane and damage along the front and back of the grain boundary.
Table I
Model for Spallation
experiment UP Up/co i r B +G , pall
A 0.6 0.12 0.057 0.10 27.3 2.9 > 1.6
B 0.333 0.067 0.057 0.058 27.3 1.6 1.6
C 0.167 0.033 0.057 0.029 27.3 0.8 < 1.6

The dynamics of shock propagation also showed the influence of the nanocrystal. As evident
from a movie, the initial shock wave travelling through the target is slowed as it passes through
the grain boundary relative to propagation in the pure crystal. Release waves are generated as
the shock traverses the grain boundary. This leads to a visible "wake" behind the nanocrystal as
the shock bends slightly. This retardation leads to damage on both sides of the nanocrystal. For
run B, the grain fractures primarily on the forward, or left hand side of the nanocrystal and the
nanocrystal separates with the spall fragment.
94

Analysis

Dislocation Dynamics Limited Failure

A dislocation dynamics model was developed in the previous work [2] to relate the activation
energy for dislocation nucleation and motion to the imposed stress. A criterion relating the strain
rate to the plastic strain resulted in an equation relating spall strength to strain rate in terms of
the elastic constants of the material. This equation not only captures the form of the correlation
observed above, but is also reasonably quantitative. The key criterion for spallation is postulated
to be when the plastic strain rate equals the imposed strain rate. The plastic strain rate is the
product of three terms:
4p = bNv, (8)
the magnitude of the Burgers vector b representing the amount of displacement carried per dis-
location, the dislocation number density N, and the dislocation velocity v. Both the dislocation
number density and the dislocation velocity in our perfect crystal will be taken to be activated
processes. It is supposed that the key internal variable describing the free energy of the system
is the stress. Nucleation must be homogeneous in our perfect crystal and will be an activated
process.
The free energy of a two-dimensional dislocation (minus the core part which drops out of our
analysis) is written as

v
ED Db2 ; D= 7r(BBG
+ G)" (9)

For homogeneous nucleation the major effect of an imposed stress r is to lower the energy penalty
associated with dislocation formation. Further, the force acting on a dislocation is proportional
to rb. Thus, the energy required to move a dislocation some distance I in the material is given
by rbl. As this energy is available due to the imposed strain, the activation energy required for
dislocation motion is lowered by this amount.
Using these basic ideas, a relationship for the spall stress was derived [2] as,
( rok6 r \
T -Io= r,6 2 In-:- (10)
-(1+a)b D l (10

where r, = 2.05, D = 1.61, a = 0.1, and e,, = 1.0 were determined by simulation of adiabatic
expansions and measuring the spall stress as a function of applied strain rate [2]. Additionally,
b = 21/6 was used for the magnitude of the Burgers vector. This relation was used to generate a
master plot for the simulated spall data, as shown in Figure 13 of reference [2].

Analysis of Spallation Simulations

With the above computer experiments, some important details of spallation can be explained.
As noted above, the reflection of the initial shock waves off the free surfaces generates a family of
release waves, which spread as they propagate through the material. The width of this zone bx
controls the strain rate at the spall plane. This phenomenon is understood to be direct result of
inelastic material behavior [8]. Using the Murnagan equation of state, calculating the position of
maximum stress, and employing the Rankine-Hugoniot equations leads to an expression for the
spall zone width as:

,X = (B1 - 1) PU(,)
L~ 8 c2 (1
showing how the zone increases with increasing particle velocity and decreases with increasing
sound velocity. The pressure derivative of the bulk modulus B, is ;; 4.7 for the material studied.
95

I TARGET
V•

L-1 B3

II
A

<2v 2. Grain boundary map after the onset of


spallation. A,B,C refer to Table I.

1. Spallation experiment geometry.

0.2
3. Shock energy in the
sample as a func- 0.1
tion of time after B
impact for runs A,B,
and C of Table I.
0 50 TIME 100 150

The effective strain and strain rate at the spall plane are then estimated by the initial particle
velocity divided by this zone width,

•= ULP, e-2 =c (12)


The maximum stress available for fracturing our material is estimated from elastic wave
mechanics as

a' = (B +G)zU, U.,


I (13)
These relations predict the maximum strain and stress at which spallation occurs in terms
of the initial particle velocity and material properties. The results will be upper bounds on the
strain and stress values, since the use of Up, for the relative particle velocity of the release waves
is clearly an overestimate if any energy is lost during wave propagation, and the use of linear
96

elasticity will also overestimate the resultant stress. The model predicts the gross features of the
observed behavior, as illustrated in Table I. Namely, spallation is expected for a stress at the
spall plane of about 1.6 (from equation 10) and for a strain of approximately 6-8% (see figure 7
of reference [21.) Shown in Table I are the calculated values for the stress, strain rate, and strain
at the spall plane from Eq. 13 and the expected spall stress. As seen, the theoretical model does
a good job of estimating the results of the spall experiment. Namely, the lowest spall velocity
results in no failure while the intermediate one is at the critical value necessary for spallation.
Clearly, the highest impact velocity greatly exceeds that necessary for spallation and significant
damage should occur.
The kinetic energy associated with motion in the horizontal direction is plotted for all three
runs as a function of time after impact in Figure 3. The highest velocity curve corresponds to
results from our previous work [21 except for small differences just before the first minimum.
This is associated with the shock moving through the grain boundary region of the nanocrystal,
resulting in a retardation of the local shock velocity and a nonuniform shock front across the
material.

Conclusions

We have simulated spallation behavior in a model two-dimensional metal containing a nanocrys-


tal defect. The presence of a defect alters the observed void opening and crack formation for
impact velocities near the spall threshold. The failure selectively follows intergranular fracture,
as previously observed for other model experiments [7]. At higher impact velocities both inter-
granular and intragranular fracture along the spall plane are observed, minimizing the effects of
the presence of the nanocrystal. These results were analyzed through a dislocation dynamics
spall model. The predictions of the simple models were found to be in good agreement with the
simulation results.

Acknowledge ments

This work (LAUR # 92-3737) was performed at Los Alamos under the auspices of the Univer-
sity of California, U.S. Department of Energy Contract No. W-7405-ENG-36. NJW acknowledges
the support of the NSF-PYI program. The computational resources, provided by the Advanced
Computational Laboratory of Los Alamos National Laboratory, are gratefully acknowledged.

REFERENCES

1. A. Kelly and N. Macmillan, Strong Solids, Clarendon Press, Oxford, 1986.


2. N.J. Wagner, B.L. Holian, and A.F. Voter, Physical Review A 45, 8457 (1992).
3. J. Harding, editor, Mechanical Properties at High Rates of Strain, 1984, volume 70, The
Institute of Physics, Bristol and London, 1984.
4. P. Follansbee, in Shock Waves in Condensed Matter 1987, edited by S. Schmidt and N. Holmes,
page 249, 1988.

5. B.L. Holiali, in Shock Waves in Condensed Matter 1987, edited by S. Schmidt and N. Holmes,
page 185, 1988.

6. M. Allen and D. Tilsedey, Computer Simulation of Liquids, Clarendon Press, Oxford, 2nd
edition, 1989.

7. B.L. Holian et al., Phys. Rev. A 43, 2655 (1991).

8. A. Keribas and I. Zakharenko, in Mechanical Properties at High Rates of Strain, 1984, edited
by J. Harding, page 277, 19,84.

You might also like