J. Phys. Chem.

A 2005, 109, 851-859 851

ReaxFFMgH Reactive Force Field for Magnesium Hydride Systems

Sam Cheung, Wei-Qiao Deng, Adri C. T. van Duin, and William A. Goddard III*
Materials and Process Simulation Center, DiVision of Chemistry and Chemical Engineering,
California Institute of Technology, Pasadena, California 91125
ReceiVed: September 3, 2004; In Final Form: NoVember 5, 2004

We have developed a reactive force field (ReaxFFMgH) for magnesium and magnesium hydride systems. The
parameters for this force field were derived from fitting to quantum chemical (QM) data on magnesium
clusters and on the equations of states for condensed phases of magnesium metal and magnesium hydride
crystal. The force field reproduces the QM-derived cell parameters, density, and the equations of state for
various pure Mg and MgH2 crystal phases as well as and bond dissociation, angle bending, charge distribution,
and reaction energy data for small magnesium hydride clusters. To demonstrate one application of ReaxFFMgH,
we have carried out MD simulations on the hydrogen absorption/desorption process in magnesium hydrides,
focusing particularly on the size effect of MgH2 nanoparticles on H2 desorption kinetics. Our results show a
clear relationship between grain size and heat of formation of MgH2; as the particle size decreases, the heat
of formation increases. Between 0.6 and 2.0 nm, the heat of formation ranges from -16 to -19 kcal/Mg and
diverges toward that of the bulk value (-20.00 kcal/Mg) as the particle diameter increases beyond 2 nm.
Therefore, it is not surprising to find that Mg nanoparticles formed by ball milling (20-100 nm) do not
exhibit any significant change in thermochemical properties.

1. Introduction and aluminum/aluminum oxides.15 This paper presents the new

force field that aims for accurate and consistent predictions of
Metal hydrides such as, LiH, CaH2, and MgH2 have widely covalent systems (MgnH2n) and all phases of magnesium hydride
been studied for hydrogen storage capabilities, and among them, (including high-energy diamond and fluorite MgH2 phases). The
MgH2 has been widely studied because of its comparatively remainder of this paper is organized as follows: Section 2
weaker M-H bonding.1,2 However, the slow sorption kinetics describes the development for the Mg/MgH reactive force field
characteristics for Mg-based materials have so far hampered and theoretical approach. The results and applications are given
commercial usage. An important reason for the slow kinetics in section 3. Section 4 presents our discussion, and concluding
of H desorption/adsorption is the high enthalpy of formation of remarks are given in section 5.
the metal hydride (-17.88 kcal/mol),3 indicating that hydrogen
atoms bind too strongly with the Mg atoms. Consequently, 2. ReaxFFMgH Reactive Force Field for Mg and MgH2
dehydrogenation of magnesium hydrides requires high temper- The development strategies for ReaxFFMgH are in line with
ature (∼552 K at 1 atm). Hence, it is necessary to decrease the those used in generating the reactive force field for hydrocar-
desorption temperature to obtain a suitable hydrogen storage bons,12 Si/SiO2,14 and Al/R-Al2O315 systems. The potential
material. Numerous methods have been focused on improving functions in ReaxFFMgH are the same as those in ReaxFFSiO,
the problematic sorption kinetic, including mechanical ball with the addition of one energy contribution (EMgH) to distin-
milling4,5,6 and mechanical alloying.7,8 However, these methods guish the Mg-H binding energy in MgH and MgH2 (see section
can only improve adsorption and not desorption kinetics,9 2.1). A key to the reactive force field is the formalism that allows
possibly because even the smallest particle sizes obtainable by for accurate description of bond formation and bond breaking
these methods (20 nm) still primarily display bulk desorption by means of a bond order/bond distance relationship similar to
characteristics. To determine whether a further reduction of that incorporated in the Tersoff,16 Brenner,17 and EDIP18
particle size could sufficiently destabilize the MgH2 nanopar- potentials and charge transfer. ReaxFF incorporates the EEM
ticles and thereby lowering the hydrogen desorption temperature charge calculation method19 which allows for a polarizable,
to values workable for mobile applications (100 °C), we have geometry-dependent charge distribution. By calculating non-
developed a reactive force field for magnesium and magnesium bonded van der Waals and Coulomb interactions between all
hydrides (ReaxFFMgH). Although quantum chemical (QM) and atoms (including 1-2 and 1-3 interactions), ReaxFF can be
semiempirical methods can be applied to study reactivity in applied to both covalent and ionic systems.
metal hydride systems,10,11 the computational expense of these In ReaxFF, the total energy expression for Mg and MgH2
methods severely hampers application to large (.100 atoms) ReaxFF is divided into contributions as follows:
high-temperature dynamical systems. Earlier on, we have
demonstrated that the ReaxFF potential can accurately describe Esys ) Ebond + Eover + Eunder + Eval +
reactive processes, including the relative stability of reactive EMgH + EvdWaals + ECoulomb (1)
substrates and products as well as high-energy transition states
and reaction intermediates, for a wide range of chemical systems, The parameters in these energy expressions were obtained from
including hydrocarbons,12 nitramines,13 silicon/silicon oxides,14 fitting to QM derived equations of state (EOS) of pure Mg and
10.1021/jp0460184 CCC: $30.25 © 2005 American Chemical Society
Published on Web 01/13/2005
852 J. Phys. Chem. A, Vol. 109, No. 5, 2005 Cheung et al.

MgH2 condensed phases and to reaction energies, bond dis- and a 6-311G**++ basis set. A smaller basis set without diffuse
sociation profiles, and angle distortions derived from QM functions were used to obtain Mulliken charge distributions.
calculations on small finite clusters. In the hydrogenation- The plane wave code CASTEP21 was used to perform DFT
dehydrogenation process, it is conceivable for (i) the magnesium calculations on the condensed states of pure magnesium and
atom in pure magnesium metal to undergo coordination number magnesium hydride systems. The generalized gradient ap-
change, (ii) phase transformation in both pure Mg and MgH2 proximation (GGA) proposed by Perdew, Burke, and Ernzer-
to occur, or (iii) the coexistence of multiple MgH2 phases. To hof22 was used for the exchange-correlation energy and ultrasoft
ensure that ReaxFFMgH can describe variations in coordination pseudopotentials to replace the core electrons. For both the Mg
for pure Mg, we fit to the following phases: hcp (12-coordinate), and MgH2 crystals, we used the Perdew-Wang implementation
fcc (12-coordinate), bcc (8-coordinate), sc (6-coordinate), and of GGA,23 a kinetic energy cutoff of 380 eV, and the
diamond (4-coordinate). To ensure a good description for MgH2 Monkhorst-Pack scheme24 to generate the k-space grid using
phases, we fit to the γ-MgH2, β-MgH2, -MgH2, CaF2-MgH2, a spacing of 0.1 Å-1. The force field was optimized against
and diamond-MgH2, along with the ground state R-MgH2 these QM data using a successive one-parameter technique.25
structure. The final force field parameters can be found in Tables 5-10;
To provide additional test cases for the reactive potential, an overview of all the potential functions can be obtained from
DFT studies were carried out on angle bending and bond the supplementary material.26
dissociation of small representative magnesium-hydrogen MD Simulations. To generate Mg and MgH2 particle of
clusters. various grain size of interest (0.6-2.2 nm), a bond order cutoff
The EEM parameters were optimized to reproduce Mulliken was used to generate a radial diameter (from a central Mg atom)
charge distributions of small Mg-H clusters calculated from in which spherical clusters can be cut from both the Mg (hcp)
DFT methods. The EEM parameters for H were determined in and R-MgH2 supercell geometry using the Crystal Builder
ReaxFFCH, so we allowed only the parameters for Mg to vary module in Cerius2.27 The supercell geometry was constructed
in our optimization. from quantum optimized bulk structures using GGA lattice
2.1. Bond Energy Function. One modification to the parameters28 when possible. The simulation procedure fol-
potential functions used in ReaxFFMgH is the addition of MgH lows: (a) Each particle was first minimized using low-
undercoordination correction bond energy function (eq 2). In temperature MD. (b) For both the equilibration and annealing
ReaxFFCH, separate functional forms (BOσ, BOπ, and BOππ) runs, we perform NVT MD simulations using the Berendsen
were used to describe single, double, and triple bonds; however, thermostat.29 A time step of 0.25 fs was used in all simulations.
only the single bond was used in determining the overall bond Atomic charges were updated at every MD simulations.
order in ReaxFFMgH. Magnesium is a simple 3s2 metal with no The minimized, equilibrated, and annealed structure for a
d electrons and a valency of two and such prefers to make single representative Mg40H80 (1.2 nm in diameter) particle is depicted
bonds. The parameters for the BOσ functional were determined in Figure 8.
from Mg-Mg and Mg-H bonds
3. Results
EMgH ) kc2(BOij - ∆i - 0.004∆i4 - 1)2
3.1. Bond Dissociation. To optimize bond energy, DFT
if BOij - ∆i - 0.004∆i4 > 1 calculations were carried out for dissociation of single Mg-H
and Mg-Mg bonds in various small clusters. Figure 1a,b shows
EMgH ) 0 if BOij - ∆i - 0.004∆i4 e 1 (2) the Mg-H bond dissociation curves in MgH and MgH2 while
Figure 1c shows the data for the dissociation of two Mg-H
on small magnesium hydride clusters such as MgH, MgH2, single bond in Mg2H4 to give two units of MgH2. Figure 1d
Mg2H2, Mg2H4, and Mg4H8. DFT calculations for the dissocia- shows the Mg-Mg bond dissociation curve in Mg2H2. In each
tion of the Mg-H single bond gave dissociation energies of 72 case, DFT calculations were carried out for both singlet and
kcal/mol for breaking the first Mg-H bond in MgH2 and 33 triplet systems (in the case of MgH, doublet and quartet
kcal/mol for breaking the second Mg-H bond. Dissociation of systems), and bond parameters were optimized to the lowest
the hydrogen atom from the MgH fragment requires removal energy points along the dissociation curve. For every monomer,
of an electron from the sigma antiorbital and leaves Mg in its bond restraints were applied to the internal coordinate of interest
most stable closed-shell configuration. Dissociation of the while the rest of the structure was allowed to relax during the
hydrogen atom from MgH2 leaves an unpaired electron in the minimization.
antibonding orbital, yielding an energetically less stable elec- 3.2. Valence Angle Terms. To optimize the valence angle
tronic configuration. This explains the discrepancy in the two parameters for ReaxFFMgH, we calculated angle bend energies
Mg-H bond dissociation energy and the following trend for from DFT calculations on small clusters of magnesium hydrides.
Mg-H dissociation energy: There are three relevant valence angle (Mg-H-Mg, Mg-Mg-
H, Mg-H-Mg) cases for which we ran DFT calculations on
DHMg-H (72 kcal) , DMg-H (33 kcal) representative monomers. For every monomer, angle restraints
were applied to fix valence angles at various values while the
To properly differentiate between the two Mg-H bond dis- rest of the structure was allowed to relax during minimization.
sociation energies, we have included an additional term in the From these DFT calculations, we obtained a valence angle curve
ReaxFF energy expression (EMgH ). against which the ReaxFFMgH parameters can be optimized.
2.2. Method and Force Field Optimization Procedure. QM Figure 2 shows a comparison between the ReaxFFMgH and DFT
Calculations and Force Field Optimization. Parameter values energies as a function of the H-Mg-H valence angle. Re-
for the potential functions were obtained by optimization against axFFMgH and DFT data for the other two angles are given in
a set of data obtained from quantum chemical methods. For the supplementary material.26
nonperiodic systems, DFT calculations were performed using 3.3. Charge Distributions. In ReaxFFMgH, charge distribu-
the Jaguar program (version 4.01) with the B3LYP functional20 tions are calculated using the EEM19 method. Mulliken charge
Magnesium Hydride Systems J. Phys. Chem. A, Vol. 109, No. 5, 2005 853

Figure 1. Bond dissociation curves for small MgH clusters as calculated by the ReaxFF and DFT methods. Dissociation of the (a) Mg-H bond
in MgH, (b) Mg-H bond in MgH2, (c) two bridging Mg-H bonds in Mg2H4, and (d) Mg-Mg bond in Mg2H2.

3.4. Reaction Energies. To sample all possible covalent

environment of the Mg-H bond, we carried out DFT and
ReaxFFMgH calculations on the reaction energies for small
MgnH2n clusters (Figure 4). Table 1 provides a comparison
between DFT and ReaxFFMgH results for these reactions.
3.5. Condensed Phase Data. One main objective is to apply
ReaxFFMgH to study the magnesium particle size effect on the
thermochemistry of the hydrogen absorption/desorption process.
Hydrogenation of pure magnesium produces MgH2 containing
7.6 mass % hydrogen. However, its formation from bulk
magnesium and gaseous hydrogen is extremely slow and
requires a temperature of 300 °C under a hydrogen pressure of
0.01 MPa. At ambient pressure and low temperatre R-MgH2
crystallizes with the TiO2-rutile-type structure but transforms
Figure 2. Energy curve of MgH2 as a function of the H-Mg-H into orthorhombic γ-MgH2 at higher temperature and pressures.
valence angle. The R- to γ-MgH2 transition pressure has not yet been
experimentally determined. However, in previous theoretical
TABLE 1: Comparison between DFT and ReaxFF Reaction work on pressure-induced transformations in MgH2, Vajeeston
Energies (kcal/mol) for Magnesium Hydride Systems
and co-workers28 found that, at equilibrium, the energy differ-
reaction EDFT EReaxFF ence between R-MgH2 and γ-MgH2 modifications is very small
1
MgH2 f /2Mg2H4 -14.3 -19.8 and concluded that both phases can coexist under a certain field
MgH2 f 1/4Mg4H8 -21.5 -26.5 pressure. Therefore, it is important that ReaxFF be able to
Mg2H4 f 1/2Mg4H8 -7.2 -6.7
MgH4 f MgH2 + H2 -87.7 -64.1 capture all possible modifications of MgH2 crystals and be able
MgH2 + H f MgH + H2 -38.0 -32.5 to predict condensed phase stabilities. For both pure Mg and
MgH + H f Mg + H2 -77.0 -70.1 MgH2 crystal phases, quantum energies were obtained for a wide
Mg3H2 f Mg2H + MgH -36.7 -25.4 range of compression and expansion points (Figures 5 and 6).
Mg2H2 + H2 f Mg2H4 -17.2 -33.5
MgH2 + Mg f Mg2H2 -3.0 -4.4 ReaxFF correctly reproduces the structural parameters (bond
distances and cell dimensions) and the equations of state for
distributions obtained from DFT calculations with the 6-31G**- all phases of pure Mg and MgH2 crystals. For all modifications,
basis set were used to optimize EEM parameters (Table 6). The the cohesive energies and densities obtained from ReaxFF were
comparison between ReaxFF and DFT data on partial charges compared against QC data (Tables 2 and 3). Our results are
for the Mg and H atoms in small MgnH2n clusters are plotted in consistent with those found by Vajeeston and co-workers.28
Figure 3. ReaxFF successfully reproduces charge assignments Furthermore, ReaxFFMgH calculates a heat of formation of
for all clusters. -20.00 kcal/mol for the R-MgH2 phase, which is in good
854 J. Phys. Chem. A, Vol. 109, No. 5, 2005 Cheung et al.

Figure 3. Comparison between Mulliken charge distributions obtained from DFT and ReaxFF charge distributions for small MgnH2n clusters.

Figure 4. MgnH2n clusters obtained from DFT simulations. Black and white balls represent Mg and H atoms, respectively.

Figure 5. Equation of state (compression/expansion) for phases of magnesium as calculated by the DFT and ReaxFF methods.

agreement with QM-derived value (-16.49 kcal/mol). The DFT ended.6,9,30 To determine whether a further reduction of grain
and ReaxFFMgH results for the heat of formations for other size below the 20 nm regime could sufficiently destabilize the
modifications of MgH2 are compared in Table 4. MgH2 nanoparticles, we have investigated the enthalpy of
3.6. Applications. 3.6.1. Heat of Formation from Mg formation of ultrasmall MgH2 particles below the 20 nm regime.
Particles to MgH2 Particles. Although improvements in the We carried out simulations on ultrasmall Mg and MgH2
kinetics of magnesium have been achieved by nanostructuring particles within the 0.62-2.20 nm range. Energies at 0 K for
of particles down to the 20-100 nm regime, the conclusions in each component particle from ReaxFF annealing simulations
the changes in the thermodynamics of these particles with were used to tabulate the heat of formation (Figure 7). The
respect to the bulk magnesium hydride still remain open- minimized, equilibrated, and annealed structures for one rep-
Magnesium Hydride Systems J. Phys. Chem. A, Vol. 109, No. 5, 2005 855

Figure 6. Equation of state (compression/expansion) for phases of magnesium hydride as calculated by the DFT and ReaxFF methods.

TABLE 2: Mg Metal Phase Stabilities (kcal/mol) Relative to P ) 0) are given as follows:

Hcp
crystal phase ∆EDFT ∆EReaxFF µMgnH2n ) energy per MgH2 unit in MgH2 nanocrystal
fcca 1.81 0.00
bcc 1.65 0.38 µMgn ) energy per Mg unit in Mg nanocrystal
simple cubic 10.94 7.01
diamond 19.00 13.26 1
µH2 ) E(H2) + pω0 (5)
a
ReaxFF does not distinguish between the hcp and fcc phases, and 2
therefore the two energies are similar.
In the total energy of H2, we included zero point energy
resentative Mg40H80 (1.2 nm in diameter) particle are depicted correction (ZPE ) 5.33 kcal/mol; heat content ) 2.37 kcal/
in Figure 8. The enthalpy of formation of this reaction mol). The zero-point vibrational energy corrections for the MgH2
clusters were not included; frequency calculations are currently
MgnH2n f Mgn + nH2 (3) under development for ReaxFF. We expect the ZPE correction
to improve the agreement of the heat of formation of bulk MgH2
can be written as with DFT values. From our results, a clear trend is observed
such that the heat of formation becomes more positive with
∆Hf ) µMgnH2n - µMgn - nµH2 (4) decreasing grain size (Figure 10). Beyond 2 nm diameter, the
heat of formation converges to the bulk value of -20.0 kcal/
The chemical potentials of the component systems (at T ) 0, mol.

Figure 7. Final geometries of MgnH2n particles obtained from annealing simulations using ReaxFFMgH. Equilibrated structures (300 K) were used
as starting geometries and annealed to 0 K.
856 J. Phys. Chem. A, Vol. 109, No. 5, 2005 Cheung et al.

Figure 8. Geometries of Mg40H80 particle obtained from ReaxFFMgH simulations after (i) minimization, (ii) equilibration at 300 K for 50 000 steps,
and (iii) annealing from 300 to 0 K (50 000 steps).

Figure 9. Energies of Mg20Hn particles as a function of H2 abstraction as calculated using ReaxFF. Successive abstraction of surface H2 were
perform on Mg20Hn systems until all hydrogen atoms were removed. Particles were equilibrated at 300 K and annealed from 300 to 0 K before each
removal of H2. Severe depletion of surface hydrogen resulted in structures where Mg atoms preferentially occupy the surface sites; thus, the removal
of hydrogen became energetically unfavorable, as evident in slope B.

TABLE 3: Condensed Phase Densities (g/cm3)

phase FDFT FReaxFF
Mg(diamond) 0.949 0.911
Mg(simple cubic) 1.459 1.311
Mg(bcc) 1.623 1.781
Mg(fcc) 1.721 1.760
Mg(hcp) 1.745 1.759
diamond-MgH2 1.167 1.218
R-MgH2 1.425 1.358
γ-MgH2 1.445 1.383
β-MgH2 1.561 1.514
CaF2-MgH2 1.604 1.578
-MgH2 1.736 1.679

TABLE 4: Heat of Formation (kcal/MgH2) of Various MgH2

Phases
DFT ReaxFF
phase ∆H0K ∆H298Ka ∆H0K ∆H298Kb
R-MgH2 -12.04 -16.49 -12.29 -20.00 Figure 10. Variation of heat of formation of magnesium hydride
γ-MgH2 -11.99 -16.44 -11.75 -19.45 particles as a function of grain size calculated using ReaxFFMgH. The
β-MgH2 -9.66 -14.11 -10.21 -17.91 dash line represents the bulk value (-20 kcal/MgH2).
-MgH2 -4.91 -9.36 -3.76 -11.47
CaF2-MgH2 -3.04 -7.49 -3.71 -11.41 TABLE 5: Atom Parameters
diamond-MgH2 -2.26 -6.71 -2.81 -10.52
atom pov/un punder λ11 (val1) pv,5 (val3) pv,6 (val4)
a
With zero-point vibration for both R-MgH2 and H2 (ZPE of R-MgH2 Mg -1.08 38.00 2.37 6.30 2.97
was used to correct for ∆H of the other MgH2 phases.) b With zero- H -10.00 0 2.75 6.30 2.88
point vibration for H2 (ZPE for MgH2 not included, although incorpora-
tion of the ZPE for MgH2 will raise the energy of ∆H and improve the
agreement with DFT results).
from a representative Mg20H40 nanoparticle (eq 6), where n )
38 to 0.

3.6.2. Hydrogen Desorption. To gain better insights into the Mg20Hx+2 f Mg20Hx + H2 (6)
behavior of the crystal during the desorption process, we
simulated the successive abstraction of surface molecular H2 The radial distribution functions gMgH(r) for Mg20Hn systems
Magnesium Hydride Systems J. Phys. Chem. A, Vol. 109, No. 5, 2005 857

Figure 11. Mg-H radial distribution functions and structures for Mg20Hn systems. Lines and balls represent magnesium and hydrogen atoms,
respectively. The RDF(s) calculated by Cerius2 indicates that there are two types of hydrogen atoms (type I hydrogen atoms with an average Mg-H
bond distance of 1.8 Å and type II hydrogen atoms with an average bond distance of 2.3 Å). Depletion of type II hydrogen atoms resulted in
structures where Mg atoms preferentially occupy surface sites. There are no surface hydrogen atoms in the Mg20H6, Mg20H4, and Mg20H2 systems.

TABLE 6: Coulomb and 1-3 Bond Order Correction TABLE 9: Bond Order Parameters
Coulomb parameters 1-3 BO correction bond pbo,1 pbo,2
atom H (eV) χ (eV) γ (Å) λ3 λ4 λ5 Mg-Mg -0.0729 4.6319
Mg-H -0.2692 6.4254
Mg 6.00 0.52 0.40 49.9248 0.3370 0
H 9.88 3.82 0.76 4.0736 2.6883 1.00
TABLE 10: Valence Angle Parameters
TABLE 7: van der Waals and Bond Radius Parameters angle θ0,0 ka kb pv,1 pv,2 λ12
atom rσ (Å) rvdW (Å) ee (kcal/mol) Aa γw (Å) H-Mg-H 0a 49.8261 0.2093 0 2.0870 2.2895
H-Mg-Mg 0 8.7037 0.0827 0 3.5597 1.1198
Mg-Mg 1.832 4.493 0.181 10.92 27.12
Mg-H-Mg 0 0.5047 0.8000 0 0.8933 4.6650
Mg-H 1.294 3.200 0.010 13.29
H-H 0.756 3.132 0.100 8.50 2.00 a
This value leads to an equilibrium angle of 180-0 ) 180° for the
single H-Mg-H valence angle.
TABLE 8: Bond Energy Parameters (Deσ in kcal/mol)
atom Deσ pbe,1 pbe,2 pbe,3 EMgH pMgH centered at 1.7 Å (type I) and 2.3 Å (type II). Figure 9 shows
Mg-Mg 32.3808 -0.0076 4.8726 0.2641 a nonlinear trend in particle stability as a function of molecular
Mg-H 58.6896 -0.0203 8.2136 0.0230 24.4461 0.0736 hydrogen abstraction. The less strongly bound type II hydrogen
atoms are first removed, resulting in the energy trend seen in
and corresponding structures are given in Figure 11. The rdf(s) slope A. Further H2 abstraction (slope B) leads to the depletion
indicates two types of hydrogen corresponding to the peaks, of surface hydrogen atoms, and rearrangement produces particles
858 J. Phys. Chem. A, Vol. 109, No. 5, 2005 Cheung et al.

where Mg atoms preferentially occupy surface sites. Therefore, intend to investigate functionalization of the particle surface with
the last 14 hydrogen atoms are more strongly bound and more Mg-R groups. The R groups should not hinder H2 absorption/
difficult to desorb, as evident in the steepness of slope B and desorption characteristics.
the disappearance of peak II. In future work we aim to apply this method to investigate
the kinetics of H absorption/desorption using different modifica-
4. Discussion tions of MgH2 phases, the process of atomic H adsorption on
the Mg (0001) surface, and the mechanistic pathway of H
We have generated a reactive force field for magnesium and absorption into bulk magnesium.
magnesium hydride system based on the functional forms used
in ReaxFFSiO. We have shown that ReaxFFMgH can reproduce 5. Conclusion
DFT values for the EOS, cell parameter, and density of
condensed systems along with DFT geometry and reaction By optimizing force field parameters against a substantial
energies for small clusters. This overall agreement between the QM-derived training set, containing bond dissociation, angle
QC and ReaxFFMgH data indicates that ReaxFFMgH could play bending, charge distribution, and reaction energy data for small
an important role in helping to elucidate some of the funda- clusters as well as equations of state for a range of magnesium
mental problems that hinder hydrogen absorption/desorption so and magnesium hydride condensed phases, we have developed
that the potential of 7.6 wt % H storage capacity of magnesium a reactive force field for magnesium and magnesium hydrides
can be reached. Specifically, we have used ReaxFFMgH to show (ReaxFFMgH). ReaxFFMgH uses the same functional forms as
a relationship between grain size reduction and decreased previous ReaxFF descriptions, and as such, this method should
structural stability. be straightforwardly extendible to other magnesium-containing
To demonstrate this relationship, we used MD simulations materials.
to generate Mg and MgH2 nanoparticles and calculated their We have used ReaxFFMgH to investigate the trend in
heats of formation as a function of particle size. From Figure thermodynamic destabilization of magnesium hydride particles
10, it is clear that as the nanoparticles size decreases, the heat with grain size. Results indicate a steep increase in heat of
of formation becomes more positive. In particular, there is a formation and decrease in structural stability as particle size
steep increase in the heat of formation within the ultrasmall drop below 1.0 nm. This finding agrees with ball milling
MgH2 particle region (0.62-0.92 nm). The decrease in structural experiments that suggest that there is little change in the heat
stability of magnesium particles in this region can be attributed of formation even when particle size is reduced to the 20-100
to two factors. First, results indicate that, below a critical particle nm regime; our results suggest that the particle size has to be
diameter of 0.92 nm, all Mg and H atoms were exposed to the reduced substantially below the 20 nm range to see a significant
surface. It is at this region that the properties of the material drop in particle stability.
begins to differ from that of the bulk properties. Second, in We believe that these reactive force field methods can provide
ultrasmall MgH2 particles, surface hydrogen atoms are generally a useful computational tool to study the chemistry at surfaces
found to occupy the less stable top and bridge sites vs the more and interfaces and as such could play a pivotal role in designing
stable three-coordinate sites more commonly observed in larger improved hydrogen storage materials.
particles. Consequently, surface Mg atoms in the ultrasmall
regime are left undercoordinated. The hydrogen atoms found References and Notes
in these energetically less favored sites are more easily desorbed (1) Schlapbach, L.; Züttel, A. Nature (London) 2001, 414, 353.
from the surface. We conclude that significant changes in the (2) Noritake, T.; Aoki, M.; Towata, S.; Seno, Y.; Hirose, Y.; Nishibori,
thermochemistry of hydrogen desorption is observed only if the E.; Takata, M.; Sakata, M. Appl. Phys. Lett. 2002, 81, 2008.
(3) Stampfer, J. F.; Holley Jr., C. E.; Suttle, J. F. J. Am. Chem. Soc.
particle grain size is reduced below 2 nm; this is in agreement 1960, 82, 3504.
with ball milling experiments which suggest that there is minor (4) Zaluska, A.; Zaluski, L.; Ström-Olsen, J. O. Appl. Phys. A: Mater.
changes in the heat of formation even for the smallest particle Sci. Process. 2001, 72, 157.
obtainable (20 nm).8 (5) Orimo, S.; Fujii, H.; Ikeda, K. Acta Mater. 1997, 45, 331.
(6) Zaluska, A.; Zaluski, L.; Strom-Olsen, J. O. J. Alloys Compd. 1999,
To study the crystal chemistry of the magnesium hydride 288, 217.
nanoparticles during the desorption process, we ran MD simu- (7) Huot, J.; Liang; G.; Schultz, R. Appl. Phys. A: Mater. Sci. Process.
lations on a series of Mg20Hn systems in which successive ab- 2001, 72, 187.
(8) Huot, J.; Pelletier, J. F.; Lurio, L. B.; Sutton, M.; Schulz, R. J.
stractions of surface H2 from the system were performed. In Alloys Compd. 2003, 348, 319.
our simulations, we did not observe any dramatic changes in (9) Liang, G.; Huot, J.; Boily, S.; Neste, A. V.; Schulz, R. J. Alloys
atom coordination for both Mg and H, nor did we observe any Compd. 1999, 292, 247.
major deviation in Mg-Mg and Mg-H bond lengths from the (10) Vegge, T. Phys. ReV. B 2004, 70, 35412.
(11) Jacobsen, K. W.; Stoltze, P.; Nørskov, J. K. Surf. Sci. 1996, 366,
bulk values. We found that the desorption of the first half of 394.
the hydrogen atoms in the Mg20H40 system occurred with greater (12) van Duin, A. C. T.; Dasgupta, S.; Lorant, F.; Goddard, W. A. Phys.
ease than the last half, as evident from slope A in Figure 9 and ReV. Lett. 2003, 91, 098301.
the disappearance of the type II hydrogen atoms comprising (13) Strachan, A.; van Duin, A. C. T.; Chakraborty, D.; Dasgupta, S.;
Goddard, W. A. J. Phys. Chem. A 2001, 105, 9396.
the peak centered at 2.3 Å for the RDF(s) in Figure 11. This (14) van Duin, A. C. T.; Strachan, A.; Stewman, S.; Zhang, Q.; Xu, X.;
can be attributed to the fact that for systems highly depleted in Goddard, W. A. J. Phys. Chem. A 2003, 105, 3803.
surface hydrogen atoms Mg atoms prefer to occupy surface sites. (15) Zhang, Q.; Çaǧin, T.; van Duin, A. C. T.; Goddard, W. A. Phys.
In fact, Mg20Hn, where n ) 2 and 4, are highly unstable struc- ReV. B 2004, 69, 45423.
(16) Tersoff, J. Phys. ReV. 1988, 61, 2879.
tures with no surface hydrogen atoms, and abstraction of these (17) Brenner, D. W. Phys. ReV. B 1990, 42, 9458.
last four hydrogen atoms would be energetically unfavorable. (18) Bazant, M. Z.; Kaxiras, E. Phys. ReV. Lett. 1996, 77, 4370.
From slope A, we calculate a hydrogen storage capacity of 4.9 (19) Mortier, W. J.; Ghosh, S. K.; Shankar, S. J. Am. Chem. Soc. 1998,
wt %, which is below its theoretical storage value of 7.6 wt %. 120, 2641.
(20) Becke, A. D. J. Chem. Phys. 1993, 98, 5648.
One concern of ultrasmall nanoparticles is rapid aggregation (21) Payne, M. C.; Teter, M. P.; Allan, D. C.; Arias, T. A.; Joannopoulos,
to form larger, more stable particles. To address this issue, we J. D. ReV. Mod. Phys. 1992, 64, 1045.
Magnesium Hydride Systems J. Phys. Chem. A, Vol. 109, No. 5, 2005 859

(22) Perdew, J. P.; Burke, K.; Ernzerhof, M. Phys. ReV. Lett. 1996, 77, (27) Cerius2 package (http://www.accelrys.com).
3865. (28) Vajeeston, P.; Ravindran, P.; Kjekshus, A.; Fjellvåg, H. Phys. ReV.
(23) Perdew, J. P.; Wang, Y. Phys. ReV. B 1992, 46, 6671. Lett. 2002, 89, 175506.
(24) Monkhorst, H. J.; Pack, J. D. Phys. ReV. B 1976, 13, 5188. (29) Berendsen, H. J. C.; Postma, J. P. M.; Van Gunsteren, W. F.; Dinola,
(25) van Duin, A. C. T.; Bass, J. M.; van de Graaf, B. J. Chem. Soc., A.; Haak, J. R. J. Chem. Phys. 1984, 81, 3684.
Faraday Trans. 1994, 90, 2881. (30) Hanada, N.; Ichikawa, T.; Orimo, S.; Fujii, H. J. Alloys Compd.
(26) See our website http://www.wag.caltech.edu/publications/papers/. 2004, 366, 269.