Computational Materials Science 112 (2016) 161–169

Contents lists available at ScienceDirect

Computational Materials Science

journal homepage: www.elsevier.com/locate/commatsci

Study of cohesion and adhesion properties of asphalt concrete with

molecular dynamics simulation
Guangji Xu, Hao Wang ⇑
Department of Civil and Environmental Engineering, Rutgers University, Piscataway, NJ 08854, USA

a r t i c l e i n f o a b s t r a c t

Article history: The objectives of this study is to develop a molecular modeling approach for studying cohesive and
Received 26 June 2015 adhesive properties of asphalt concrete and evaluate the accuracy of modeling through comparisons with
Received in revised form 21 September experimental data. Fully atomistic models were built for molecular dynamics (MD) simulation consider-
2015
ing two representative asphalt models and two types of aggregate mineral. MD simulations were per-
Accepted 20 October 2015
Available online 19 November 2015
formed to study thermodynamic and cohesive properties of asphalt binder, such as density, solubility
parameter, cohesive energy density, and surface free energy. The adhesion properties were investigated
by calculating the interaction energy and the work of adhesion at asphalt–aggregate interface for the first
Keywords:
Asphalt–aggregate interface
time. The bond energy parameters in dry and wet conditions were used to evaluate moisture sensitivity
Molecular dynamics simulation of interface adhesion. The results show that van der Waals force plays critical role for cohesive properties
Interaction energy of asphalt binder; while the adhesion bonding between asphalt to aggregate is largely dependent on the
Work of adhesion type of aggregate mineral (silica or calcite) in both dry and wet surface conditions. The effect of asphalt
Moisture susceptibility type was found significant for the adhesion between asphalt and silica at the relatively small moisture
content. The simulation results agree well with experimental measurements reported in the literature.
This work illustrates MD can help in understanding fundamental chemo-mechanics relationship of
asphalt concrete at an atomistic scale, which can be used as a useful tool for material design and perfor-
mance prediction.
Ó 2015 Elsevier B.V. All rights reserved.

1. Introduction mental theories including mechanical theory, chemical theory,

Asphalt concrete has been widely used for roadway pavements
due to its ability to provide structural capacity and smooth surface.
In asphalt concrete, aggregate forms the skeleton of mixture; while
asphalt binder serves as the primary binding material. However,
several types of pavement distresses, such as premature rutting,
raveling, and cracking are commonly observed during the early life
of asphalt pavements.
It has been found that three factors are responsible for the fail-
ure of asphalt concrete, including cohesion loss within asphalt,
strength reduction of aggregate particles, and breakdown of adhe-
sive bonding between aggregate and asphalt [1]. Cohesion proper-
ties of asphalt binder and adhesion properties of asphalt to
aggregate (i.e. bonding strength) are largely dependent on chemi-
cal compositions of asphalt and aggregate. To better understand
the adhesive bond between asphalt and aggregates, four funda-
⇑ Corresponding author at: Department of Civil and Environmental Engineering,
Rutgers, The State University of New Jersey, Piscataway, NJ 08854, USA.
Tel.: +1 848 4452874.
E-mail address: hwang.cee@rutgers.edu (H. Wang).
weak boundary theory, and thermodynamic theory have been pro-
posed [2]. Meanwhile, moisture effect has been found being the
major driving force causing the loss of adhesion between asphalt
and aggregate and further deterioration [3].
A number of studies have been conducted to evaluate adhesion
properties of the asphalt–aggregate interface and moisture suscep-
tibility of asphalt concrete using bond strength measurements
from experimental and phenomenological engineering aspects
[4–7]. Nanoscale experiments have also been used to measure sur-
face free energy and adhesive/cohesive force–displacement rela-
tionships between asphalt molecular groups and aggregates at
dry and wet conditions [8–11]. These experimental studies have
produced acceptable and reasonable results for better understand-
ing cohesion and adhesion properties of asphalt concrete.
Recently, atomistic modeling with molecular dynamic (MD)
simulation has been employed to study the link between chemical
structures and physical and mechanical behavior of asphalt con-
crete and durability, such as density, thermal expansion coefficient,
and viscosity [12], adhesive strength [13], aging effect [14], curing
[15] and self-healing [16]. Molecular dynamics simulation is a pow-
erful and feasible approach for material design and performance

http://dx.doi.org/10.1016/j.commatsci.2015.10.024
0927-0256/Ó 2015 Elsevier B.V. All rights reserved.
162 G. Xu, H. Wang / Computational Materials Science 112 (2016) 161–169
prediction because the atomistic structures of asphalt and aggre-
gate and their interfaces determine the behavior of asphalt concrete
at the continuum scale. The significant advantage of MD simulation
is to provide fundamental observations in the initiation and evolu-
tion of material damage at the nanoscale that are very difficult to
obtain from the traditional experiments conducted at the
laboratory.
While a number of studies have used MD simulation to study
asphalt concrete from a more fundamental and nature perspective,
little work has been focused on deriving cohesion and adhesion
properties of asphalt concrete from the atomistic scale to the con-
tinuum scale. Limitations exist for in-depth investigation of mois-
ture susceptibility of asphalt concrete at the nanoscale. The
interaction between asphalt and aggregate is governed by various
attractive intermolecular and intramolecular forces across the
interface. This necessitates investigation of the fundamental cohe-
sion and adhesion properties of asphalt concrete using a system-
atic atomistic modeling approach.
2. Objective and scope
The objectives of this study is to develop an atomistic modeling
approach for studying cohesive and adhesive properties of asphalt
concrete and evaluate the accuracy of modeling through compar-
isons with experimental data. In order to achieve this objective,
fully atomistic models were built for MD simulation considering
two representative asphalt models and two types of mineral aggre-
gate. Bulk asphalt models and confined layer models were used to
derive thermodynamic material properties of asphalt binder,
including density, surface free energy, cohesive energy density,
and solubility parameter. The work of adhesion between asphalt
binder and aggregate were derived from interaction energy with
different definitions of surface area. The effects of chemical struc-
tures of asphalt and aggregate and moisture contents on the adhe-
sion between asphalt and aggregate were investigated.
The fact that material failure originates from particle dislocation
and bond rupture at the nanoscale necessitates the importance of
atomistic modeling approach in material design. The physical and
chemical compatibility of asphalt binder and aggregate affects fati-
gue cracking and moisture damage in asphalt concrete. Molecular
dynamics simulations can provide valuable insights in understand-
ing the interaction between asphalt and aggregate at the nanoscale
and relate chemical compositions of asphalt concrete to its macro-
scopic behavior. Currently, understanding how asphalt concrete
deforms and breaks under mechanical and environmental loading
is often limited to phenomenological engineering approaches,
neglecting the underlying failure mechanism in the atomistic struc-
ture. The nanoscale details of material failure are still not clear and
the experiments are time-consuming and expensive. The atomistic
approach developed from this study laid the ground work to predict
failure potential of asphalt concrete in a computational testing
environment. This will lead to better design of asphalt mixture for
road pavements with longer service life and less repair.
3. Molecular dynamics simulation
3.1. Basic principle
Molecular dynamics simulation is a technique for calculating
the equilibrium and transport properties of a classic many-body
system [17], in which the motion of the constituent particles obeys
the laws of classical mechanics. It computes the motions of a num-
ber of atoms and molecules in a system as a function of time. The
basis of MD simulation is Newton’s law of motions and statistical
mechanics, where statistical ensemble averages are equal to time
averages of a system, as shown in Eq. (1). Molecular dynamics sim-
ulations require the definition of a potential function, or a descrip-
tion of the terms by which the particles in the simulations will
interact. The force on an atom i can be directly calculated by the
derivative of the potential energy, E, with respect to the coordinate
ri.
@E @ 2 ri

¼ mi 2 ð1Þ
@ri @t
where mi is the mass of atom i. In a system consisting of N atoms, all
pairs can result in N(N
1)/2 interactions, that is to say, even mod-
erate numbers of atoms N will generate a very large number of
calculations.
3.2. Force field
Force field, sometimes used as potentials, refers to a mathemat-
ical function used to compute the energies of a system of atoms
with various system conformations in the context of molecular
modeling. In general, the force field parameters can be obtained
from the experiments, together with quantum mechanical calcula-
tions. The goal of a force field is to describe entire classes of mole-
cules with reasonable accuracy. Basically, the total potential
energy E can be comprised of a number of bonded and non-
bonded interaction terms, as shown in Eq. (2).
E ¼ Ebonded þ Enon-bond ð2Þ
where the bonded terms, Ebonded is for atoms interactions con-
tributed by covalent bonds including 0062ond stretching, angle
bending and dihedral and improper interaction, etc., while the
non-bond interaction term describes non-covalent contributions
that mainly contain van der Waals energy, Coulomb electrostatic
energy, and hydrogen bond energy. A number of force fields have
been developed over the years. The energy terms describing differ-
ent kinds of deformations were added or changed in Eq. (2) for dif-
ferent force fields. In all cases using an appropriately parameterized
force field is an important issue in molecular simulation.
Molecular dynamics (MD) simulations were performed using
commercially available simulation software, Materials Studio
[18]. COMPASS II (Condensed-Phase Optimized Molecular Poten-
tials for Atomistic Simulation Studies) force field was used for
describing atom-level interactions in molecular models. COMPASS
is the first parameterized and validated ab initio-based force field
that has a broad coverage in covalent molecules including most
common organics, small inorganic molecules, and polymers. COM-
PASS II is a significant development extension to the COMPASS
force field in terms of atom types and force field terms, which
enable us to make accurate predictions material properties for a
wide range of compounds in isolation and in condensed phases
[19]. The more realistic the potential is, the closer the MD simula-
tions results represent the real properties of materials. The amount
of time needed for MD simulations is largely dependent on the
functional complexity of potential.
3.3. Molecule models for asphalt
Asphalt material is a complex chemical mixture of molecules
that are predominantly hydrocarbons with a small amount of
structurally analogous heterocyclic species and functional groups
containing sulfur, nitrogen and oxygen atoms. From solubility
point of view, asphalt materials are composed of three main con-
stituents, i.e., asphaltene that is the most viscous and polar compo-
nents, saturates, consisting of aliphatic molecules and are least
viscous and non-polar, and resin whose properties are in between
of above two. Due to the complex mixture nature of asphalt, many
 G. Xu, H. Wang / Computational Materials Science 112 (2016) 161–169
efforts were made to construct model asphalt mixtures with
reasonable compositions. The average molecular structures for
different asphalt binders were proposed based on Nuclear
Magnetic Resonance (NMR) spectra [20]. These average asphalt
molecular models can achieve proper element concentrations
and aromatic/aliphatic ratios, but it cannot reflect the molecular
packing between different molecular phases and thus accurately
capture the complexity of asphalt composition.
In order to build a model more similar to the real asphalt, the
diversity of chemistries and polarities are required. Since the focus
of this study is not to determine the representation of molecular
structure for asphalt binder, an existing three-component model
asphalt system was used [21]. In the three-component model,
n-C22 molecules were used to represent saturates and 1,7-
dimethylnaphthalene was used to represent resin (naphthene aro-
matics). Two different asphaltene molecules were proposed, one is
a moderate-size aromatic core with very small branches [22], and
the other is smaller aromatic core and much longer alkane side
branches [23]. This relative simple model has been proved accept-
able to predict physical and rheological properties of asphalt bin-
der in a general sense [12]. Fig. 1 shows the molecular structures
of two different asphaltene models. Table 1 shows the molecular
compositions of asphalt mixtures with two different asphaltene
models. The mass fraction was selected to resemble the total car-
bon/hydrogen ratio reported by previous researchers [24,25].
3.4. Construction of bulk asphalt model
Amorphous cell module in Materials Studio was employed to
construct three-dimensional (3D) periodic structures of molecular
Fig. 1. Molecular structures of (a) asphaltene 1; (b) asphaltene 2; (c) 1,7-dimethylnaphthalene; and (d) n-C22 (Carbon atoms are shown in grey, sulfur atoms in yellow, and
hydrogen atoms in white). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
163
liquids and polymeric systems. In this study a cubic cell with an
initial density of 0.1 g/cm3 in 3D periodical condition was built at
first, with a purpose to randomly distribute three types of mole-
cules and prevent the molecule chains twisted with each other.
After a geometry optimization process, an equilibrium run with
constant volume and temperature (NVT) for 200 ps with a time
step of 1 fs was performed to pre-equilibrate the system, which
brought the asphalt system from an initial state to a more equili-
brated state at the target temperature. Then another molecular
dynamic run with isothermal–isobaric (NPT) ensemble was per-
formed for 300 ps of relaxation time to shrink the system volume
and to approach a stable state of the real density. During this pro-
cess, Nose–Hoover–Langevin (NHL) thermostat and Andersen
barostat were applied for system to maintain a target temperature
and pressure, respectively. The relaxation time used in the NVT and
NPT processes was proved to be enough because it was observed
that the temperature and density reached the stable values in
the process. The time step of 1 fs was selected considering the bal-
ance between accuracy of simulation results and simulation time
[21].
After the density results were stabilized, an additional NVT of
100 ps was performed to further equilibrate the model. The final
structure of asphalt model is shown in Fig. 2(a). One important fea-
ture of the asphalt model is that molecules occupy not only within
the cell but also across the cell boundaries due to the periodicity. A
colloidal asphalt structure can be observed in which asphaltene
molecules were dispersed in resins and saturates. Fig. 2(b) shows
the microstructure of asphalt bulk model with distributions of free
and substantial volumes. The dispersion of asphaltene molecules in
the maltene molecular network causes molecular motion that is
164 G. Xu, H. Wang / Computational Materials Science 112 (2016) 161–169
Table 1
Molecular composition of asphalt model (after [21,24,25]).
Asphalt model Asphaltene-1/2
Molecule number Mass fraction
Asphalt-1 5 20.7
Asphalt-2 5 21.1
Fig. 2. Illustration of asphalt-1: (a) 3D periodic molecular model showing arrangements of asphaltene 1 (red), resins (yellow), and saturates (green); and (b) cell volumetric
data showing substantial volumes (gray) and free volumes (blue). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of
this article.)
temperature dependent. The free volumes provide spaces for
molecular movements between different asphalt components and
thus influence thermodynamic properties of asphalt.
3.5. Construction of confined asphalt layer
Confined layers of asphalt binders were built for the
asphalt–aggregate interface model. The only difference between
the bulk asphalt model and the confined asphalt layer lies in the
periodical conditions in the z direction. In the confined layer, the
box boundary in the z direction functions as hard repulsive wall,
and if asphalt atoms move close to the z boundary during model
construction, the atoms will rebound back into the box. Compared
to the bulk asphalt model, confined asphalt layer had a relatively
flat surface in the z direction, which was used for the asphalt–
aggregate interface model as described below. For the purpose of
calculating surface free energy, a free surface was created by arbi-
trarily elongating the dimension of the cell in the z direction and
equilibrating the periodic system in isometric–isothermal (NVT)
conditions. The illustration of bulk asphalt and the confined
asphalt layer with free surface is shown in Fig. 3.
3.6. Molecular models of mineral aggregate
A mineral is defined as an element or chemical compound that
is normally crystalline and resulted from geological processes. A
crystal structure is the orderly geometric spatial arrangement of
atoms in the internal structure. This crystal structure is based on
regular internal atomic or ionic arrangement in the geometric
form. Chemistry and crystal structure together define a mineral.
Crystal structure greatly influences the physical properties of
mineral. Mineral aggregates that are widely used in construction
materials include sand, granite, limestone, quartz, etc. Among all
the mineral aggregate materials used in asphalt concrete, silicon
1,7-Dimethyl naphthalene n-C22
Molecule number Mass fraction Molecule number Mass fraction
27 19.7 41 59.6
30 19.8 45 59.1
dioxide (or silica) and calcium carbonate (or calcite) are two most
common chemical compounds. For example, limestone consists
primarily of calcium carbonate, while silicon dioxide is common
in sand and granite with a high percentage. Therefore, in this study,
silica (SiO2) and calcite (CaCO3) were used to represent ideal aggre-
gate in molecular dynamics simulation.
3.7. Creation of asphalt–aggregate Interface
The confined asphalt layer was attached to the aggregate sub-
strate to form an asphalt–aggregate interface model. At first, a unit
crystalline silica, with lattice parameters of a = b = 4.913 Å,
c = 5.405 Å, a = b = 90°, and c = 120°, was imported from structures
database in Materials Studio. As for the calcite, a unit cell with
dimension of a = b = 4.990 Å, c = 17.061 Å, a = b = 90°, and
c = 120° was used. It is believed that during the simulations, the
orthogonal geometry was preferred for the purposes of model con-
struction and further calculation. Therefore, the unit silica or cal-
cite cell were first cleaved in [0, 0, 1] direction, and then followed
by geometry transformation to have an orthogonal shape with P1
space group, in which a = b = c = 90°. By repeating the unit cell in
x and y directions to generate the same size as asphalt models, a
2-D bulk silica or calcite surface was created. After that, a vacuum
slab was added to form silica or calcite blocks with 3D periodic
boundary conditions. Confined asphalt layers were then placed
over the silica or calcite blocks. A vacuum of 30 Å was added above
to avoid the interaction across the mirror image in the z direction.
The final bi-material interface model between asphalt and aggre-
gate was built as shown in Fig. 4.
Once the interface model was obtained, geometry optimization
for 5000 iterations was carried out followed by a dynamic equili-
bration run of 200 ps with canonical ensemble (NVT), to make sure
that the model configuration was further optimized. An additional
50 ps NVT with 50 frames was performed for a product run with
 G. Xu, H. Wang / Computational Materials Science 112 (2016) 161–169 165

a b

(a) (b)
Fig. 3. Asphalt thin film formation for surface energy calculation: (a) bulk model and (b) confined asphalt layer with free surface.

at the interface and the diffusion of water through asphalt binder

and aggregate adsorption.
In MD simulation, the three-phase interface models were geom-
etry optimized first to eliminate the highest energy overlap and
then followed by a NVT equilibration for 100 ps to dissipate high
energies and optimize the configuration. An additional 50 ps NVT
run with 50 frames was then used for energy calculation. Unless
otherwise stated, all MD simulations for interface interaction were
performed at 298 K with a time step of 1 fs in this study. A cutoff
distance of 10.5 Å was adopted to speed up the computation. This
means that the interactions between two atoms separated by a dis-
tance greater than this pre-defined cutoff distance are ignored.

4. Results and discussion

4.1. Density of asphalt

Density is an important thermodynamic property and it is a

Fig. 4. Asphalt-1/aggregate interface atomistic model: (a) silica and (b) calcite.
(Carbon atoms are grey, sulfur atoms are yellow, hydrogen atoms are white, red is
oxygen, orange is silicon, and green is calcium.) (For interpretation of the references
to color in this figure legend, the reader is referred to the web version of this
article.)
the subsequent calculation purposes. Trajectory files after molecu-
lar dynamic simulations were used for calculation with statistical
mechanics. A modified script based on Perl language was devel-
oped to calculate the interaction energy between asphalt and
aggregate.
direct indicator to see whether the model size and force field could
yield reasonable and accurate simulation results. Therefore, density
values at different temperatures were calculated for asphalt-1 and
asphalt-2 models, respectively. With the simulation process stated
above, the density results of asphalt models are shown in Fig. 5. The
results show that the highest predicted density is 0.92 g/cm3
at
10 °C and the densities reduces to 0.85–0.88 g/cm3 at 60 °C.
As expected, a steady increase of density with decreasing tempera-
ture was observed. The density predictions are slightly smaller than
the reported experiment data from 0.99 to 1.33 g/cm3 at 60 °C [27].
This is properly because the higher hydrogen, more carbon, and less
sulfur were used in the molecular model of asphalt, together with a
lack of heteroatoms besides sulfur [21]. An improved model asphalt

3.8. Consideration of moisture effect

Based on the asphalt–aggregate interface model, the moisture

effect was simulated by adding water molecules between asphalt
and aggregate to form a three-phase interface system. This was
assumed to represent the final stage of the moisture diffusion pro-
cess at the asphalt–aggregate interface, which was regarded as the
major mechanism for moisture-induced damage in asphalt con-
crete [26]. Three different numbers (100, 150, and 200) of water
molecules were considered to represent different levels of mois-
ture contents, which correspond to 7–14% mass fraction of asphalt.
The diffusion process and distribution pattern of water molecules
will affect simulation results and experimental observations.
Future should be conducted to simulate the asphalt–aggregate
interfacial behavior considering the moisture distribution pattern Fig. 5. Comparison of density results for two asphalt models.
166 G. Xu, H. Wang / Computational Materials Science 112 (2016) 161–169
system with 12 chemical components has been proposed recently
to provide more accurate density predictions [28], which will be
used in future study.
4.2. Surface free energy of asphalt
Surface free energy is theoretically defined as the magnitude of
work required to create a unit area of a new surface of the material
under vacuum conditions. From an atomistic point of view, mole-
cules in the bulk material are surrounded by other molecules from
all sides; the surface free energy equals the work needed to extract
the molecules from the bulk and create a new area of surface mole-
cules. Therefore, surface free energy can be used to measure the
disruption of intermolecular interaction that occurs when a surface
is created. Surface free energy has been proposed for selecting the
combination of asphalt binders and aggregate types that have the
necessary compatibility to form strong bonds and resist moisture
damage [29] and it was found related to the cohesive bond energy
of asphalt materials [8].
In this study, the surface free energy was calculated as the dif-
ference between the potential energies of the thin asphalt layer
and its corresponding bulk asphalt divided by the surface area cre-
ated upon formation of the thin layer, as shown in Eq. (3) [30,31].
This definition calculates the contribution of internal potential
energy to the surface free energy but cannot consider the entropic
contribution due to deformation and segregation of molecular
chains to the surface energy. The Connolly surface area was used
to consider the surface unevenness due to molecular agglomera-
tion in asphalt models. The Connolly surface area consist of all
the points of the van der Walls surface at which a solvent sphere
can touch and it is also known as accessible surface area [32].
The work of cohesion (Waa) can be calculated from the surface
free energy (ca) using Eq. (4), which is defined as the work required
to separate a unit area of a liquid or solid into two unite areas. The
work of cohesion is used as an indicator for cohesion properties of
asphalt when the failure within the asphalt is concerned.
ca ¼ ðEfilm
Ebulk Þ=2A ð3Þ
W aa ¼ 2ca ð4Þ
where ca is surface free energy; Waa is work of cohesion; Efilm and
Ebulk are the potential energy of the confined asphalt layer and bulk
asphalt, respectively; and A is area of the new surface to be created.
Table 2 summarizes the calculated surface free energy and work
of cohesion for two asphalt models. The difference between the
dimensional area and the real surface area of asphalt model can
be clearly observed. It can be seen that in general the calculated
Table 2
Surface free energy and work of cohesion of asphalt.
Properties Asphalt-1
Dimensions (Å) 34.11  34.11  34.11
Surface area of new surface (Å2) 2258
Surface free energy ca (mJ/m2) 38.81
Work of cohesion Waa (mJ/m2) 77.62
Table 3
Cohesive energy density and solubility parameter of bulk asphalt.
Properties Asphalt-1
CED (108 J/m3) 3.22
Solubility parameter d ((J/cm3)1/2) 17.95
Electrostatic d ((J/cm3)1/2) 1.11
Van der Waals d ((J/cm3)1/2) 17.92
results agree well with the measured values reported in the litera-
ture, although the simulation results were slightly greater than
most experimental results. The discrepancies could be caused by
the approximation of asphalt molecular models. Previous experi-
mental measurements have shown that the concentration of satu-
rates or asphaltene could lower the value of surface free energy,
while the resins and the ratio of resin/asphaltene could increase
the value of surface free energy [33]. This indicates that further
study is needed to evaluate sensitivity of surface free energy of
asphalt with different chemical compositions and elemental
contents.
4.3. Cohesive energy density and solubility parameter of asphalt
Cohesive energy density (CED) is used as a measure of the
mutual attractiveness of molecules to assess intermolecular inter-
action inside an asphalt molecule model [36]. A solubility parame-
ter can be further defined from the CED using Eq. (5). The solubility
parameter quantifies the solvent–solute interactions to separate
solvent-solvent and solute–solute molecules as a means of mini-
mizing the Gibbs free energy of the entire system [37]. It has been
used to study molecular attractions properties in the asphalt sys-
tem and new asphalt models and evaluate colloidal stability in
asphalt [38,39]. Since the intermolecular interaction within asphalt
is evaluated via van der Waals and electrostatic (including hydro-
gen bond) interactions, solubility parameter can be expressed in
two components, as shown in Eq. (6).
pffiffiffiffiffiffiffiffiffiffi
d¼ CED ð5Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
d¼ d2v dw þ d2ele ð6Þ
where d is solubility parameter; CED is cohesive energy density;
and dvdw and dele are representing the contributions from van der
Waals forces and electrostatic interactions, respectively.
Table 3 shows the calculated CED and solubility parameter of
bulk asphalt model. The values of CED and solubility parameter
were found in good agreement between simulation results and
previous results from experiments or simulations. It was found
that for solubility parameter values, van der Waals component
was much greater than electrostatic component. This means that
cohesion is largely dependent on van der Waals forces which held
the molecules of asphalt binder together. On the other hand, the
results show that the asphalt-1 model has slightly greater values
of CED and solubility parameter than those of the asphalt-2 model.
It indicates that asphalt-1 has stronger mutual attractive interac-
tions among its molecules fractions, and presents better cohesion
Asphalt-2 Experiment values
35.64  35.64  35.64 N/A
2604 N/A
37.01 13–47.6 [33,45]
74.02 27–65 [34,35]
Asphalt-2 Experiment or previous simulation values
3.08 3.19–3.22 [40]
17.56 13.30–22.50 [40,41]
0.90 N/A
17.54 N/A
 G. Xu, H. Wang / Computational Materials Science 112 (2016) 161–169
properties as compared to asphalt-2. This finding is consistent with
the work of cohesion of two asphalt models that is derived from
surface free energy as shown in Table 2.
4.4. Work of adhesion between asphalt and aggregate
The work of adhesion is described as the work required to sepa-
rate asphalt from aggregate at the interface. The work of adhesion
between asphalt and aggregate relates significantly to the resis-
tance of asphalt mixture to fracture-related failure behavior and
durability. The work of adhesion between asphalt and aggregate
can be calculated using Eq. (7), with the interaction energy (Einter)
derived from Eq. (8). This approach has been used in the literature
to successfully calculate work of adhesion between polymer–metal,
polymer–polymer and more general solid–liquid systems [42,43],
which considers the pure physicochemical interaction between
two interacting materials. Based on the definition in Eqs. (7) and
(8), a negative value of Einter or Wadhesion indicates attraction
between these two components, while a positive value of Einter or
Wadhesion indicates repulsion. The calculated Einter or Wadhesion from
MD simulation are negative values and the absolute values were
used in the analysis for indicating the attraction between asphalt
and aggregate.
W adhesion ¼ DEinter
aag =A
DEinter aag ¼ Etotal
ðEasphalt þ Eaggregate Þ
where DEinter_aag is interaction energy between asphalt and aggre-
gate; Etotal is total potential energy of asphalt and aggregate in con-
tact at equilibrium; and Easphalt and Eaggregate are the potential
energies of asphalt and aggregate (i.e., silica and calcite) separated
in vacuum at equilibrium, respectively; Wadhesion is work of adhe-
sion between asphalt and aggregate; and A is the interface contact
area between asphalt and aggregate.
Fig. 6 shows the calculated interaction energy between asphalt
and aggregate and the contributions of non-bond components (van
der Waals and electrostatic energy). The results indicate that the
interaction energy between asphalt and aggregate varies depend-
ing on the combination of material types. The values of interaction
energy were found equal to the summation of van der Waals and
electrostatic components, which indicated that adhesion between
asphalt and aggregate was mainly the non-bond interaction. This
is because no chemical bond was formed during the interaction
and therefore covalent components were negligible. On the other
hand, it was found that van der Waals energy constituted the
major contribution to the non-bond energy terms. The Coulomb
interaction was found negligible in the asphalt–silica interaction
and serving as a minor role in the asphalt–calcite interaction.
Fig. 6. Interaction energy and its component predicted by MD simulation.
167
Although different theories (physical, chemical or mechanical)
have been proposed to explain the adhesion between asphalt bin-
der and aggregate, the actual mechanism is not fully explained. The
Strategic Highway Research Program (SHRP) studies found that the
polar constituents in the asphalt adhere to the aggregate surface
through electrostatic forces, hydrogen bonding, or Van der Waals
interactions. Several different methods of measuring the energy
of adsorption indicated that physisorption rather than chemisorp-
tion is occurring [1].
Fig. 7 shows the calculated work of adhesion between different
types of asphalt and aggregate models. Three definitions of inter-
face areas, including the layer dimensional area, the Connolly area
of asphalt surface, and the Connolly area of aggregate surface, were
used in the calculation. It was found that the selection of interface
contact area could cause significant variations in calculating the
work of adhesion from MD simulation. It is believed that the actual
interface contact area is dependent on the morphological feature of
asphalt and aggregate, which could be larger than the layer dimen-
sional area and stay between the rough surface areas of asphalt and
aggregate. In general, the range of work of adhesion values
calculated using the Connolly area agree well with the experimen-
tal results reported in the literature, which range from 60 to
150 mJ/m2 with different sources of asphalt and aggregate [35,44].
The work of adhesion results show that for the same type of
aggregate, the difference between the work of adhesion of
ð7Þ
asphalt-1 and asphalt-2 models is small. On the other hand, the
effect of aggregate mineral on the work of adhesion is much more
ð8Þ obvious. In particular, for the same type of asphalt, the work of
adhesion between asphalt and calcite is much greater than that
between asphalt and silica. Therefore, calcite is more prone to
adsorb asphalt than silica at dry condition. This is in agreement
with the experimental finding that a mineral with an alkaline
(basic) nature such as calcite, results in a stronger adhesion with
asphalt, compared to quartz (acidic) with a rather acidic nature [5].
4.5. Moisture effect on asphalt–aggregate adhesion
Due to the hydrophobic nature of asphalt binder, water has a
detrimental impact on the bonding between asphalt and aggregate.
The energy associated with the displacement of asphalt by water
from the asphalt–aggregate interface or debonding is referred to
as the work of debonding, as expressed in Eq. (9) [45]. The calcu-
lated values of the work of debonding were found negative that
was consistent with experimental findings. This indicates that
there exists a thermodynamic potential for water to disrupt the
asphalt–aggregate interface. In other words, no external energy is
required to be added into the system in order to separate the
asphalt–aggregate interface due to the water-loving nature of
aggregate.
Fig. 7. Work of adhesion between asphalt and aggregate predicted by MD
simulation.
168 G. Xu, H. Wang / Computational Materials Science 112 (2016) 161–169
W debonding water ¼ ðDEinter aw þ DEinter agw
DEinter aag Þ=A ð9Þ
where Wdebonding_water is work of debonding when water displaces
asphalt from the asphalt–aggregate interface; DEinter_aw is the inter-
action energy between asphalt and water; DEinter_agw is the interac-
tion energy between aggregate and water; DEinter_aag the interaction
energy between asphalt and aggregate; and A is the interface con-
tact area.
The energy ratio (ER) has been used in previous experimental
studies to evaluate moisture effect on the adhesion between
asphalt and aggregate, which is based on the hypothesis that mois-
ture sensitivity is directly proportional to the dry adhesive bond
energy, and inversely proportional to the work of debonding
[8,45]. A higher value of ER is more likely indicating the less sus-
ceptibility to moisture damage. In this study, the ER was calculated
using the ratio of the work of adhesion in dry condition as com-
pared to the work of adhesion in wet condition, as shown in Eq.
(10). It was assumed here that the introduction of water would
not change the interface area between asphalt and aggregate. In
other words, water molecules at the interface were assumed in full
contact with asphalt and aggregate at the interface. Therefore, the
selection of interface contact area would not affect the results of ER
values.
ER ¼ jW adhesion =W debonding water j ð10Þ
where Wadhesion is the work of adhesion between asphalt and aggre-
gate in dry condition, and Wdebonding_water is work of debonding
when water displaces asphalt from the asphalt–aggregate interface.
Fig. 8 shows the calculated ER values for asphalt–silica and
asphalt–calcite models with three scenarios of moisture contents.
The simulation results are toward to the lower ends as compared
to the range of ER values (0.3–30) observed from experiments
[44–46]. The discrepancy is properly because the chemical compo-
sition of aggregate (such as quartz and limestone) are more com-
Fig. 8. Moisture sensitivity of (a) asphalt–silica and (b) asphalt–calcite interaction
with ER values calculated from MD simulation.
plicated than the pure silica and calcite. Previous research has
found that ER values higher than 1.5 indicates that asphalt mixture
has relatively high resistance to moisture damage; while asphalt
mixtures with high susceptibility to moisture damage are related
to ER values smaller than 0.5 [47].
The results show that large differences were observed for the ER
values when the same type of asphalt interacts with silicate and cal-
cite. This indicates that moisture susceptibility is strongly depen-
dent on aggregate chemistry. In particular, silica shows much less
water susceptibility than calcite. It is noted that mixed findings
have been reported for the moisture susceptibility among different
mineral aggregates (such as limestone, granite, sandstone) due to
differences in chemical composition, aggregate adsorption, poros-
ity, contaminants, and surface texture [8,44–46]. It was found that
the ER value between asphalt and silica decreased as the moisture
content increased; while the ER value between asphalt and calcite
were very small regardless of moisture contents. This indicates that
the asphalt–calcite bond could be susceptible to moisture damage
at relatively small moisture content. Previous studies have
measured bond strength between asphalt binder and aggregate at
different moisture conditions and found that the pull-off strength
between limestone and asphalt binder dropped significantly after
6–24 h moisture conditioning [7].
On the other hand, for the two asphalt models interacting with
silica, the ER value was smaller for the asphalt-2 model at the
relatively small moisture content. The ER values of two asphalt
models became close when the moisture content increased. This
indicates that the effect of asphalt composition on moisture sus-
ceptibility with silica may be only significant at the relatively small
moisture content. However, when two asphalt models interact
with calcite, the effect of asphalt on moisture susceptibility is neg-
ligible. This observation should be further studied by changing the
chemical composition of asphalt in different ways, such as
polymer-modified binder or the addition of anti-stripping agent.
It is noted that the classical MD simulations conducted in this
study cannot account for the formation and breakage of bonds in
the reaction. For example, the dissociation of water molecules at
the surface of quartz can form hydrogen bonds and affect the cal-
culation of surface energy and work of debonding. Future study
with the reactive interatomic potential (such as ReaxFF) will be
conducted to conquer this limitation. The MD simulations could
be further improved by considering the effect of aggregate adsorp-
tion (such as hydroxylated silica or calcite surface), porosity, and
surface texture.
5. Conclusions
This study developed a thermodynamics-based molecular mod-
eling approach to investigate cohesion and adhesion properties of
asphalt concrete for the first time. Molecular dynamic simulation
was employed to derive thermodynamic properties of amorphous
asphalt models (such as surface free energy and cohesive energy
density) and work of adhesion between asphalt and aggregate at
dry and wet conditions. The simulation results agree well with
the experimental measurements reported in the literature. The
results show that van der Waals force plays critical role for cohe-
sive properties of asphalt binder; while the adhesion bonding
between asphalt to aggregate is largely dependent on the type of
aggregate mineral (silica or calcite) in both dry and wet surface
conditions. The effect of asphalt type was found significant for
the adhesion between asphalt and silica at the relatively small
moisture content.
The developed approach sheds light on the complex interaction
between different components of asphalt and at the organic–inor-
ganic bi-material interface in asphalt concrete. It promotes devel-
 G. Xu, H. Wang / Computational Materials Science 112 (2016) 161–169 169

opment of computational experiments for material performance
prediction from the atomistic scale to the continuum scale. This
will provide knowledge into design of asphalt concrete using a
‘‘bottom-up” approach to achieve performance requirements of
roadway pavements.
Acknowledgement
The authors would like to acknowledge the support provided by
China Scholarship Council (CSC) to Guangji Xu for study abroad.
References
[1] C.W. Curtis, K. Ensley, J. Epps, Fundamental Properties of Asphalt–Aggregate
Interactions Including Adhesion and Absorption, SHRP-A-341, Washington, DC,
1993.
[2] K. Kanitpong, H.U. Bahia, Asph. Paving Technol. 72 (2003) 502–528.
[3] S. Caro, E. Masad, A. Bhasin, D.N. Little, Int. J. Pavement Eng. 9 (2) (2008) 81–98.
[4] M.J. Khattak, G.Y. Baladi, L.T. Drzal, J. Mater. Civil Eng. 19 (5) (2007) 411–422.
[5] Z.W. Chen, J. Xie, Y. Xiao, J.Y. Chen, S.P. Wu, Constr. Build. Mater. 64 (2014) 60–
66.
[6] F. Canestrari, F. Cardonea, A. Graziania, et al., Road Mater. Pavement Des. 11
(2010) 11–32.
[7] R. Moraes, R. Velasquez, H.U. Bahia, Trans. Res. Rec. 2209 (2011) 70–81.
[8] A. Bhasin, D. Little, K.L. Vasconcelos, E.A. Masad, Trans. Res. Rec. 2007 (2001)
37–45.
[9] R.A. Tarefder, A. Zaman, J. Mater. Civil Eng. 22 (7) (2010) 714–725.
[10] H.R. Fischer, E.C. Dillingh, C.G.M. Hermse, Appl. Surf. Sci. 265 (2013) 495–499.
[11] A.L. Lynea, V. Wallqvistb, B. Birgisson, Fuel 113 (2013) 248–256.
[12] L.Q. Zhang, M.L. Greenfield, J. Chem. Phys. 127 (19) (2007) 194502.
[13] Y. Lu, L. Wang, Int. J. Pavement Eng. 12 (4) (2011) 311–323.
[14] R.A. Tarefder, I. Arisa, Energy Fuels 25 (2011) 2211–2222.
[15] X. Zhou, S. Wu, G. Liu, P. Pan, Mater. Struct. (2014).
[16] A. Bhasin, R. Bommavaram, M.L. Greenfield, D. Little, J. Mater. Civil Eng. 23 (4)
(2011) 485–492.
[17] M.P. Allen, Introduction to molecular dynamics simulation, in: N. Attig, K.
Binder, H. Grubmüller, K. Kremer (Eds.), Computational Soft Matter: From
Synthetic Polymers to Proteins, Lecture Notes, John von Neumann Institute for
Computing, Julich, 2004, pp. 1–28.
[18] Materials Studio, Biovia Software Inc., San Diego, 2015.
[19] H. Sun, J. Phys. Chem. B 102 (1998) 7338–7364.
[20] P.W. Jennings, J.A. Pribanic, M.A. Desando, M.F. Raub, Binder Characterization
and Evaluation by Nuclear Magnetic Resonance Spectroscopy, SHRP-A-341,
Washington, DC, 1993.
[21] L.Q. Zhang, M.L. Greenfield, Energy Fuels 21 (2007) 1712–1716.
[22] L. Artok, Y. Su, Y. Hirose, et al., Energy Fuels 13 (2) (1999) 287–296.
[23] H. Groenzin, O.C. Mullins, Energy Fuels 14 (3) (2000) 677–684.
[24] D.A. Storm, S.J. DeCanio, M.M. DeTar, V.P. Nero, Fuel 69 (6) (1990) 735–738.
[25] D.A. Storm, J.C. Edwards, S.J. DeCanio, E.Y. Sheu, Energy Fuels 8 (1994) 561–
566.
[26] N. Kringos, T. Scarpas, C. Kasbergen, P. Selvadurai, Int. J. Pavement Eng. 9 (2)
(2008) 115–128.
[27] R.E. Robertson, J.F. Branthaver, P.M. Harnsberger, et al., Fundamental
Properties of Asphalt and Modified Asphalts, vol. I, Interpretive Report,
Technical Report FHWA-RD-99-212, McLean, VA, US Department of
Transportation, Federal Highway Administration, 2001.
[28] D.D. Li, M.L. Greenfield, Fuel 115 (2014) 347–356.
[29] D. Little, A. Bhasin, Using surface energy measurements to select materials for
asphalt pavement, in: Final Report for NCHRP Project 9–37, 2006.
[30] M. Deng, V.B.C. Tan, T.E. Tay, Polymer 45 (2004) 6399–6407.
[31] S. Verenich, S. Paul, B. Pourdeyhimi, J. Appl. Polym. Sci. 108 (2008) 2983–2987.
[32] M.L. Connolly, J. Am. Chem. Soc. 107 (5) (1985) 1118–1124.
[33] J. Wei, F. Dong, Y. Li, Y. Zhang, Constr. Build. Mater. 71 (2014) 116–123.
[34] A. Bhasin, J. Howson, E. Masad, et al., Trans. Res. Rec. 2007 (1998) 29–37.
[35] Y. Tan, M. Guo, Constr. Build. Mater. 47 (2013) 254–260.
[36] E.K. Ensley, Thermodynamics of asphalt intermolecular interactions and
asphalt–aggregate interactions, in: G.V. Chilingarian, T.F. Yen (Eds.),
Asphaltenes and Asphalts, first ed., Elsevier, Los Angeles, 1994, pp. 401–426.
[37] A.F.M. Barton, CRC Handbook of Solubility Parameters and Other Cohesion
Parameter, second ed., CRC Press, Boca Raton, Florida, 1991.
[38] J.R. Lin, T.F. Yen, The study of molecular attractions in the asphalt system by
solubility parameter, in: M.K. Sharma, T.F. Yen (Eds.), Asphaltene Particles in
Fossil Fuel Exploration, Recovery, Refining, and Production Processes, Plenum
Press, New York, 1994, pp. 171–184.
[39] P.G. Redelius, Fuel 79 (1) (2000) 27–35.
[40] P. Wang, Z. Dong, Y. Tan, Z. Liu, Energy Fuels 29 (2015) 112–121.
[41] P.C. Painter, The Characterization of Asphalt and Asphalt Recyclability, SHRP-
A-675, Washington, DC, 1993.
[42] S. Kisin, J.B. Vukić, P.G.T. van der Varst, et al., Chem. Mater. 19 (4) (2007) 903–
907.
[43] M.A.A.R. Quddus, O.J. Rojas, M.A. Pasquinelli, Biomacromolecules 15 (4) (2014)
1476–1483.
[44] A.E. Alvarez, E. Ovalles, S. Caro, Constr. Build. Mater. 28 (2012) 599–606.
[45] A. Bhasin, Development of Methods to Qualify Bitumen–Aggregate Adhesion
and Loss of Adhesion Due to Water, Texas A&M University, Texas, 2006.
[46] Y. Liu, A. Apeagyei, N. Ahmad, et al., Int. J. Pavement Eng. 15 (7) (2014) 657–
670.
[47] A. Bhasin, E. Masad, D. Little, R. Lytton, Limits on adhesive bond energy for
improved resistance of hot mix asphalt to moisture damage, Trans. Res. Rec: J.
Trans. Res. Board 1970 (2006) 3–13.