Sie sind auf Seite 1von 8

Chemical Physics Letters 604 (2014) 8996

Contents lists available at ScienceDirect

Chemical Physics Letters


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

FRONTIERS ARTICLE

Structural and electronic properties of aqueous NaCl solutions from


ab initio molecular dynamics simulations with hybrid density functionals
Alex P. Gaiduk a, Cui Zhang b, Franois Gygi c, Giulia Galli a,
a

Institute for Molecular Engineering, The University of Chicago, Chicago, IL 60637, United States
Department of Chemistry, Princeton University, Princeton, NJ 08544, United States
c
Department of Computer Science, University of California, Davis, Davis, CA 95616, United States
b

a r t i c l e

i n f o

Article history:
Received 23 March 2014
In nal form 19 April 2014
Available online 30 April 2014

a b s t r a c t
We present a study of a dilute solution (1 M) of NaCl in water, carried out using ab initio molecular
dynamics with semilocal and hybrid functionals. We showed that the structural and electronic properties
of the solute and the solvent are the same as those obtained in the innite dilution limit, i.e. for aqueous
ions in the presence of a uniform compensating background. Compared to semilocal functionals, simulations with hybrid functionals yield a less structured solution with a smaller number of hydrogen bonds
and a larger coordination number for the Cl anion. In addition, hybrid functionals predict qualitatively
correct positions of the energy levels of the ions with respect to the valence band of water.
2014 Elsevier B.V. All rights reserved.

1. Introduction
Understanding the basic properties of simple salts in water
remains a topic of great interest, from a fundamental standpoint
[18]. In addition, the electronic properties of solvated ions are
attracting much interest for their role in electrochemical processes
occurring at aqueous interfaces, as well as in several biochemical
and biological processes [911].
In spite of extensive experimental and computational research
of simple salts in water, many fundamental questions remain
unanswered, e.g. the extent to which the hydrogen-bond network
and the overall structural properties are perturbed by the presence
of ions. It is commonly accepted that ions affect the properties of
water within their rst coordination shell [1], but whether their
inuence extends further in the liquid is controversial. A traditional view, supported by the interpretation of a number of macroscopic studies of viscosity, diffusion, and hydration entropy, is that
the presence of solvated ions has far-reaching effects on the structure of the liquid, even in dilute solutions. (For a summary, see the
extensive review of Ref. [2].) For example, according to neutron diffraction studies combined with empirical-potential structural
renements [3], the presence of ions in solutions, especially at high
concentrations (25 M), leads to substantial modications of the
water oxygenoxygen radial distribution function (RDF), resembling the effect of high pressure on pure water [12]. Hence Refs.

Corresponding author.
E-mail address: gagalli@uchicago.edu (G. Galli).
http://dx.doi.org/10.1016/j.cplett.2014.04.037
0009-2614/ 2014 Elsevier B.V. All rights reserved.

[3,13] concluded that the ions affect the hydrogen-bond network


of water well beyond their rst solvation shell. Kondoh et al. [8]
drew similar conclusions by studying dielectric relaxation in aqueous 0.61.4 M solutions of LiCl, NaCl, KCl, and CsCl using terahertz
time-domain spectroscopy.
However, recent spectroscopic studies suggest that the effect of
dissolved aqueous ions may be more local than previously thought.
It was established using femtosecond mid-infrared spectroscopy
[4,14,15] that the orientational dynamics of water is not modied
outside the rst solvation shell of dissolved Na+, Mg2+, SO2
4 , and

ClO4 ions (36 M). The authors of Refs. [4,14,15] also proposed that
the previous ndings on measured viscosity values of salt solutions
could be explained without invoking long-ranged ion-induced
hydrogen bonds breaking in the solutions. A Raman spectroscopy
study combined with SPC/E molecular dynamics simulations [6]
concluded that the changes in OH vibrational spectra of 1 M KCl
solutions compared to pure water result from the electric eld of
the ions on adjacent H2O molecules, rather than from overall
changes of the liquid structure. An infrared-spectroscopy study of
0.2 M NaCl solutions [16] corroborated by BLYP ab initio simulations showed that the correlation of water dipoles is largely unaffected outside the rst two solvation shells of Na+ and Cl ions,
again suggesting that the effect of these ions is short-ranged.
Finally, a recent X-ray study carried out for 117 M solutions of
LiCl [7] found that the concentration-dependent features of X-ray
spectra saturate only at very high (11 M) salt concentrations,
implying that in the entire concentration range, Li+ and Cl ions
do not substantially distort the hydrogen bonding of the liquid outside the rst solvation shell.

90

A.P. Gaiduk et al. / Chemical Physics Letters 604 (2014) 8996

Instrumental to understanding the structure of aqueous solutions are molecular dynamics simulations, which provide a valuable source of information about the microscopic structure of
solutions. A number of classical and ab initio simulations have been
carried out to study the structure of water and how it may be
affected by the presence of solutes. Some simulations, mostly done
with classical potentials, reported that the ions signicantly affect
the solvent structure in their second hydration shell and beyond.
Specically, using the CHARMM22 force eld, Obst et al. [17] found
that the number of hydrogen bonds in the second solvation shell of
cations of rst- and second-group elements is above the purewater limit, in 0.11 and 0.45 M solutions. Chandrasekhar et al.
[18] reached similar conclusions by analyzing a 0.45 M solution
of Na+ using MD simulations and the TIPS2 potential; however, a
subsequent study by the same authors [19], which employed a
slightly modied ion-water interaction potential, did not conrm
their previous ndings. Interestingly, the studies of both Refs.
[18,19] did not report differences in the OO radial distribution
function of water with respect to that of the aqueous solutions.
Most ab initio simulations reported to date seem to support
the point of view that the effect of ions on water structure does
not extend beyond their rst solvation shell. In particular, in our
ab initio studies of 1 M Na+ [20], Mg2+ [21], and Ca2+ [22] ions
using the PBE functional, we did not observe any signicant
hydrogen bond breaking in the liquid outside the rst solvation
shell of ions. These results were supported by simulations of
0.87 and 1.73 M solutions of aqueous Li+, Na+, and K+ [23] using
the HCTH functional [24], which concluded that the effect of ions
on the dipole moment of water molecules is conned to the rst
solvation shell. These ndings were also conrmed by ab initio
simulations of 0.43 M solution of NaCl in water using the BLYP
functional [25].
As well known, the quality of ab initio simulations critically
depends on the approximations chosen for the exchange-correlation
potential used to solve the KohnSham equations. To date, most
studies have been performed with semilocal generalized-gradient
approximations such as PBE [26,27] and BLYP [28,29], which suffer
from a number of problems including self-interaction (delocalization) errors [3032]. Hybrid functionals [33] such as PBE0 [34,35]
and B3LYP [28,29,36] correct some of the problems of local density-functional approximations and are appealing candidates for
rst-principles simulations. Unfortunately, the use of hybrid functionals has been limited in this eld due to computational difculties
in evaluating the long-range HartreeFock exchange in periodic
plane-wave-based calculations. Some existing implementations
[37,38] employ approximate schemes for treating exact exchange,
similar in spirit to range-separated hybrids. Simulations carried
out with these implementations have led to controversial results:
Refs. [37,38] reported almost no change in the liquid structure of
water when using PBE0, compared to PBE. Other studies [39], carried
out with an accurate implementation of the PBE0 functional [40],
reported a substantial softening of the PBE0 oxygenoxygen radial
distribution function with respect to PBE and an overall better agreement with experiments, especially in predicting IR spectra.
Hybrid functionals are expected not only to provide a more
accurate description of the structural properties, but also to
improve on the electronic properties of solutions [9,41,4245],
which, at the PBE level, may be seriously affected by self-interaction errors. Encouraging results were recently obtained by Zhang
et al. [46], who found that the use of the PBE0 functional signicantly improves the position of the Cl highest occupied level with
respect to the water valence band maximum (VBM) in 0.87 M solution of aqueous Cl ion. However, in Ref. [46], the solution was
treated within the innite dilution model, i.e. the charge of a dissolved ion was neutralized by a uniform background with an opposite sign, implicitly introduced by the Ewald summation scheme,

and it is yet unclear whether the presence of a counterion in the


solution would affect the computed properties.
It was found computationally that counterions may affect the
solvation shell and stability of ions in solutions [48,49]. The pres
ence of ClO4 increased the coordination number of Cm3+ from 8
to 9 in PBE simulations of 0.87 M solutions of curium salts [48];
however, Cl and Br did not have such an effect on the coordination shell of Cm3+. It was also shown that for inhomogeneous systems such as proteins and membranes in water, the innite
dilution model may erroneously overstabilize ions inside a lower
dielectric by tens to hundreds kilojoules per mole [50].
In this work, we carried out simulations of realistic solutions of
a simple salt, NaCl, where both ions were included in our molecular dynamics simulations. We used hybrid functionals and analyzed several samples. Our results suggest that the presence of
Na+ counterions does not affect the structural and electronic properties of solvated Cl compared to the innite dilution limit. In
addition, we found that the use of the PBE0 functional improves
the description of radial distribution functions and positions of
electronic levels compared to PBE.
The Letter is organized as follows. After a brief discussion of
computational details in Section 2, we assess the recursive subspace bisection algorithm proposed in Ref. [47] for approximate
treatment of the exchange energy, and compare simulations performed in NVT and NVE ensembles. We then discuss the structural
(Section 3) and and electronic (Section 4) properties of solutions,
followed by a brief summary.

2. Methods
We studied three aqueous solutions: A 0.87 M solution of aqueous Cl ion in the absence of any counterions, and two solutions
(0.87 and 1.03 M) of NaCl in water. The 0.87 M solutions were simulated with a periodically repeated unit cell of size 12.42 containing 63 D2O and one Cl (aqueous Cl) or 62 D2O with Na+
and Cl ions (aqueous NaCl). The 1.03 M NaCl solution contained
52 D2O molecules and one NaCl unit in a 11.74 simulation box.
The initial congurations for the 1.03 M NaCl solution were taken
from an existing PBE trajectory for NaCl [51]. We chose heavy
water (D2O) as a solvent so as to use larger simulation time
steps0.363 fs for NaCl with PBE0 and 0.242 fs for all other
simulations.
We carried out ab initio molecular dynamics simulations with
the PBE [26] and PBE0 [34,35] functionals using the Qbox code
[52]. We used HamannSchlterChiangVanderbilt (HSCV)
pseudopotentials (PP) [53,54] for deuterium, oxygen, and chlorine
atoms [55], and TroullierMartins pseudopotential with the 2p
electrons included in the valence subspace for sodium. The PBE0
simulation for 0.83 M Cl solution was done with norm-conserving
Hamann pseudopotentials [53], solely for comparison with previous PBE0 results [39,46], all obtained with Hamann PPs.
We collected MD trajectories in both the canonical (NVT) and
microcanonical (NVE) ensembles. We carried out NVT simulations
with the BussiDonadioParrinello (BDP) thermostat [56] at the
target T 380 K (PBE0) and 400 K (PBE). The thermostat time constant was varied between 60 and 120 fs during equilibration, and
between 120 and 240 fs in the production NVT runs. We chose a
temperature above 300 K because in the absence of proton quantum effects and at the experimental density, water is known to
be overstructured at ambient conditions when using semilocal or
hybrid functionals [39,57,58], although to different extents,
depending on the functionals. Prior to accumulating statistics,
the systems were equilibrated for  34 ps.
For the NaCl solution simulated with the PBE and PBE0
functionals, we performed two runs in parallel, starting with two

91

A.P. Gaiduk et al. / Chemical Physics Letters 604 (2014) 8996

2.1. Accuracy of hybrid functionals computed with the recursive


subspace bisection technique

4
PBE0
PBE0-RSB

g O-O(r)

different uncorrelated samples. In total, we collected over 50 ps


(2  25 ps) of PBE0 data for NaCl in the NVT ensemble and  8
ps in the NVE ensemble. Access to these relatively long simulation
times with the hybrid functionals was possible by using the recursive subspace bisection (RSB) method to compute the exchange
energy [47,59]. In the following subsection we discuss this technique and assess its accuracy.

1
A generalized-gradient approximation (GGA) [61] may be represented as a sum of exchange (X) and correlation (C) functionals
GGA
GGA
EGGA
q;
XC q EX q EC

both of which depend only on the electron density q and its gradient rq. GGA functionals predict reasonably good total energies and
energy differences for broad classes of solids and molecules, but the
eigenvalues (orbital energies) of the KohnSham equations yield a
poor description of excitation energies, in part because semilocal
exchange-correlation potentials do not exhibit the correct Coulombic (1=r) decay.
The quality of KohnSham orbital energies may be improved by
approximating the exchange functional as a linear combination of
terms including HartreeFock (HF) exchange; such functionals are
called hybrid:
HF
GGA
Ehybrid
q 1  aEGGA
q;
X q aEX EC
XC

EHF
X

where
is the HartreeFock exchange energy functional evaluated using self-consistent KohnSham orbitals /i (exact-exchange
functional):

EHF
X 

N
1X
4 i;j

dr

/i r/i r0 /j r/j r0 0


dr :
jr  r0 j

(For simplicity, this expression is written for doubly-occupied orbitals only.) The mixing parameter a may be chosen empirically
[33,36] or derived from the rst principles [34]; it typically varies
between 0.25 and 0.5. The inclusion of the exact-exchange term
partially recovers the correct asymptotic behavior of the
exchange-correlation potential, yielding improved eigenvalues and
total electronic energies. As a result, hybrid functionals outperform
gradient-corrected approximations for most non-metallic systems.
However, the calculation of exact exchange using plane-wave
basis sets is a very demanding task due to a large number of
two-electron integrals needed to compute the energy functional
of Eq. (3). Gygi and Duchemin [47] proposed a method that efciently screens and avoids computing some of these integrals. This
procedure, which involves a recursive bisection of orbitals in subdomains of the unit cell, leads to a 2- to 10-fold acceleration of
ab initio MD simulations of 64-water-molecule cells with hybrid
functionals, depending on the threshold  and the level of recursion chosen [47,59]. Generally, the larger the threshold , the more
computational savings. In addition, the method becomes increasingly efcient as the size of the system increases.
In order to understand the effect of the bisection approximation
on the average properties computed in our MD simulations, we
performed a 14-ps NVE simulation (390  21 K) of aqueous Cl
with the PBE0 functional and bisection (abbreviated as PBE0RSB) with  0:02 and a two-level recursion, and we compared
the results to the published data for the same system performed
without bisection [46]. For a fair comparison, we used a segment
of the PBE0-RSB trajectory that has exactly the same length
(5.6 ps) as the entire PBE0 run from Ref. [46].
Figure 1 shows that PBE0 and PBE0-RSB yield essentially the
same radial distribution functions for water within the simulation

0
2

r []
Figure 1. Oxygenoxygen radial distribution functions [g OO r] of liquid water
obtained in NVE simulations of 0.87 M Cl solution with PBE0 and no approximations in the calculation of HartreeFock exchange (379  18 K) [46], and with PBE0
adopting the bisection method to evaluate exact exchange (PBE0-RSB) [47] with
 0:02 (387  21 K). Each trajectory is 5:6 ps. Shaded areas are estimated error
bars (see Supplementary Information).

error bars. The average number of hydrogen bonds1 obtained in the


two simulations is also similar: 3:53  0:08 per water molecule in
PBE0 simulation and 3:47  0:09 when the RSB technique is used.
The use of bisection does not alter the coordination and the structure
of the solvation shell of Cl ion, as Figure 2 shows.
We further checked the effect of bisection on the electronic
structure of the solutions. Using PBE0-RSB to compute the orbital
energies along a PBE0 trajectory (taken from Ref. [46]) lowers the
solution band gap and position of the Cl highest occupied molecular orbital (HOMO) level with respect to the results obtained
without bisection by only 0.010.02 eV, well within our statistical
error bars (0.030.05 eV for band gaps and relative single particle
energies). The PBE0 band gap of the solution simulated with
PBE0-RSB remains the same, and the relative position of Cl HOMO
level changes by  0:05 eV. Again, these differences fall within our
error bars.2
Although the runs with and without bisection we analyzed
were rather short, our comparative results suggest that the RSB
technique [47] does not have a signicant effect on the structure
of the solutions and on the positions of electronic energy levels.
We now turn to the comparison of NVT and NVE simulations.
2.2. NVT and NVE simulations: Performance of the BussiDonadio
Parrinello thermostat
We compared radial distribution functions obtained from the 8ps NVE trajectory (381  20 K) for aqueous NaCl with those from
our two 25-ps NVT trajectories (382  25 K and 383  24 K) for
the same system. The results, obtained using PBE0-RSB, are
1
We used a geometrical criterion for hydrogen bond formation [39,62,63]: Two
water molecules are considered hydrogen-bonded if the distance between the
oxygens is less than the cutoff distance of 3.35 and the angle O  OD is less than
30. The cutoff distance is usually chosen as the position of the rst minimum in the
OO RDF, and should be different for each method and system. However, for
consistency we used the same value of OO distance for all the trajectories because
the average number of hydrogen bonds is not very sensitive to this distance within
the variations found in our simulations.
2
Note that the positions of levels reported in this work differ from those published
in Ref. [46] (by  0:1 eV) for the same trajectory. This is partially because we
computed positions of electronic levels for a larger number of congurations (64128
in this study compared to 815 in Ref. [46]), and also because we used HSCV
pseudopotentials instead of the Hamann ones to compute the KohnSham
eigenvalues.

92

A.P. Gaiduk et al. / Chemical Physics Letters 604 (2014) 8996

PBE0
PBE0RSB
gClO(r)

gClO(r)

NVT
NVE

3
2
1

0
4

0
2

gNaO(r)

3
2
1

r []

0
Figure 2. Same comparison as Figure 1 but for chlorineoxygen radial distribution
functions [g ClO r].

r []
Figure 4. Same comparison as Figure 3 but for chlorineoxygen [g ClO r] and
sodiumoxygen [g NaO r] radial distribution functions.

gOO(r)

3. Structural properties of solutions

NVT
NVE

2
1

gHH(r)

0
3
2
1

gOH(r)

0
2
1
0
1

r []
Figure 3. Partial radial distribution functions [gr] between oxygen and hydrogen
atoms of liquid water obtained with  8-ps NVE (381  20 K) and the average of
two  25-ps NVT (383  24 K) simulations of 1.03 M NaCl solution. NVT simulations employed the BDP thermostat [56] with a variable time constant of 120
240 fs. Calculations were performed with the PBE0 hybrid functional using the
recursive subspace bisection technique [47],  0:02.

reported in Figures 3 and 4, where it is seen that all radial


distribution functions are almost indistinguishable. We note that
the overall quality of RDFs computed in the NVT ensemble is
higher because the corresponding trajectory is longer.
We also computed the values of the band-gaps and positions of
single-particle electronic levels along the NVT and NVE trajectories. The average band gap in the NVT (NVE) run is 6.42
(6.39) eV, with the difference well within the error bar of 0.03
0.05 eV. The HOMO level of chlorine is on average 0.30 eV above
the water valence band maximum in the NVT simulation, compared to 0.29 eV in the NVE run, with the difference again within
the error bars (0.020.03 eV).
We therefore conclude that the use of the BDP thermostat does
not signicantly alter the structural and electronic properties of
the solution in the molecular-dynamics simulation, and can be
used to generate production-quality data. In the rest of the Letter,
we report results obtained for two 25-ps NVT simulations of the
1.03 M NaCl solution with the PBE0-RSB functional, equivalent to
50 ps of simulation data.

In order to investigate how the solute changes the structure of


water, we analyzed the radial distribution functions of pure water
and of salt solutions, all simulated in similar conditions. We
focused on the oxygenoxygen RDF, which was suggested to be
sensitive to the presence of ions in water [3]. Figure 5 compares
oxygenoxygen radial distribution functions for pure water and
for solutions of Cl and NaCl, simulated at the PBE and PBE0 levels
of theory. The left panel of the gure shows that the RDFs of the
three systems described using PBE are similar, although the salt
solutions have somewhat lower magnitude of peaks in RDFs. The
average number of hydrogen bonds is 3:48  0:12 in pure water
and is 3:29  0:12 and 3:28  0:13 in Cl and NaCl solutions,
respectively. This is consistent with previous theoretical study
[65], which found monotonous decrease of the number of hydrogen bonds as the concentration of dissolved ions increased.
Analogous comparisons for PBE0 results are less straightforward because previous PBE0 simulations [39,46] were carried out
with Hamann, not HSCV, pseudopotentials. Compared to Hamann,
the use of HSCV PPs yields a less structured oxygenoxygen radial
distribution function that is closer to the experimental data (see
Supplementary Information). To assess the effect of the solute on
the structure of the solution, we estimated what the RDFs for water
and aqueous Cl solution would be, had they been obtained from
HSCV, not Hamann PP, calculations. To this end, we computed
the difference between Hamann and HSCV OO RDFs for aqueous
Cl solutions simulated at the PBE level of theory [46]. We then
subtracted this difference from the RDFs of water and Cl solutions
simulated with Hamann pseudopotentials at the PBE0 level of theory (for details, see Supplementary Information). The plots, shown
in the right panel of Figure 5, exhibit the same trends as the PBE
ones. The OO RDFs for solutions of Cl and NaCl are similar to
each other and are slightly less structured compared to pure water,
but this might be an effect of a lower temperature (370 K) and/or
smaller simulation box with 32 D2O molecules used for pure water.
The number of hydrogen bonds in NaCl (3:10  0:16) and Cl
( 3:15) solutions is as well similar.
It follows from the analysis of Figure 5 that adding  1 M of Na+
counterions to the solution containing Cl does not alter the oxygenoxygen radial distribution function. This conclusion is supported by a similar number of hydrogen bonds found in aqueous
Cl and NaCl, both at the PBE and PBE0 levels of theory. Furthermore, we performed separate PBE simulations of 0.87 M Na+

93

A.P. Gaiduk et al. / Chemical Physics Letters 604 (2014) 8996

Cl, PBE
NaCl, PBE

D2O (estimated), PBE0

Cl (estimated), PBE0RSB
NaCl, PBE0RSB

gOO(r)

gOO(r)

D2O, PBE

0
2

r []

r []

Figure 5. (Left panel) Oxygenoxygen radial distribution functions [g OO r] for pure water and 0.87 M solutions of Cl and NaCl described with the PBE functional and HSCV
pseudopotentials for D, O, and Cl. The g OO of water is an average between two RDFs computed in an NVT ensemble over two 15-ps trajectories, at a target temperature of
400 K [60]. The results for the simulations of Cl and NaCl solutions are an average over two 21-ps trajectories obtained in an NVT ensemble at 400 K. (Right panel) Oxygen
oxygen radial distribution functions for pure water, 0.87 M solution of Cl, and 1.03 M solution of NaCl computed with the PBE0 functional (water) and the PBE0-RSB
framework,  0:02 (solutions). The simulation of (NaCl)aq was performed with HSCV PPs for D, O, and Cl, and TroullierMartins PP for Na atom. The RDF for water (Ref. [39])
and Cl were obtained from simulations with Hamann pseudopotentials and corrected to represent the HSCV results (see Supplementary Information). Note that, unlike in
Figure 1, the RDF for Cl solution here is computed using the entire 14-ps PBE0-RSB trajectory.

Cl (aq), PBE0
NaCl (aq), PBE0
NaCl (aq), PBE

gClO(r)

PClO()

0.02

0.01

0.00

0.02

0.01
0.00
0

20

40

60

80

100 120 140 160 180

[degree]

2
1
0
4

gNaO(r)

Cl (aq), PBE0
NaCl (aq), PBE0
NaCl (aq), PBE

Figure 7. Distribution [Ph] of the tilt angle h between the oxygenion position
vector and bisection vectors of water molecules within the rst solvation shell of
each ion. The thermodynamic and simulation conditions are the same as in Figures
5 and 6.

0.03

PNaO()

solutions and found that the number of hydrogen bonds in aqueous Na+ is almost the same (3:42  0:12) as in pure water
(3:48  0:12), indicating that the presence of Na+ does not affect
the hydrogen-bond network as much as the presence of Cl.
Finally, the top panels of Figures 6 and 7 show that the ionoxygen
radial distribution functions and tilt angle distributions of water
molecules exhibit negligible changes upon adding Na+ to Cl solution. Namely, the positions of the rst maximum and minimum in
g ClO r are almost the same in the solutions of Cl and NaCl for
both PBE and PBE0 simulations (Table 1), also resulting in similar
coordination numbers.
In agreement with our previous studies of water with hybrid
functionals [39], we found that using PBE0 leads to a softer
OO radial distribution function (Figure 8). The number of
hydrogen bonds for aqueous NaCl decreases from 3:28  0:13
(PBE) to 3:10  0:16 (PBE0). These observations are consistent with
those of Ref. [37], which found a similar albeit smaller change in
the RDFs when using PBE0, but they disagree with the results of

3
2
1
0
2

r []
Figure 6. Ionoxygen radial distribution functions [g XO r] obtained from the
simulations of NaCl and Cl solutions. The systems and computational details are
same as in Figure 5 (except no Hamann-to-HSCV extrapolation was done). Results
for PBE simulation of the Cl solution are not shown and are similar to all other
curves reported in the top panel.

Ref. [38], which reported no difference between PBE and PBE0 simulations. Note, however, that the authors of Ref. [38] used a modied version of the PBE0 functional with an approximately
screened exchange term (see Ref. [71] of Ref. [39] for more detail),
which might have affected their conclusions.
The use of PBE0 also slightly changes the ionoxygen radial distribution functions with respect to PBE. Figure 6 and the data in
Table 1 show that the PBE0 hybrid extends the rst solvation shell
of Cl ion by 0:050:1 compared to PBE and increases its coordination number by 0.20.5. At the same time, PBE0 leaves the coordination of Na+ ion almost unaltered, with a negligible 0.01
difference in the size of the coordination shell and no effect on
the coordination number. The tilt angle distributions (Figure 7) of
Cl and Na+ ions are very similar at the PBE and PBE0 levels of theory, in both Cl and NaCl solutions. As expected, the solvation shell
of Cl is different from that of Na+: The smaller tilt angle for Cl ion

( 55 ) indicates that the solvating water molecules point towards



Cl with one of the hydrogen atoms. On the other hand, Na+ is

94

A.P. Gaiduk et al. / Chemical Physics Letters 604 (2014) 8996

Table 1
Solvation characteristics of Cl and Na+ ions in aqueous Cl and NaCl solutions. The systems (simulations) are 0.87 M NaCl (PBE), 1.03 M NaCl (PBE0), and 0.87 M Cl (PBE and
PBE0). The distances r max and r min denote, respectively, the positions of the rst maximum and minimum in the radial distribution functions. N coord is a coordination number
computed by counting all water molecules within the rst solvation shell of each ion. All quantities and the corresponding errors were estimated by averaging over 10 segments
of every trajectory, each  4:3-ps long. All simulations were performed with HSCV pseudopotentials except for the PBE0-RSB of Cl, which was performed with Hamann PPs. We
veried that the Hamann and HSCV PPs yield very similar ionoxygen radial distribution functions.
Solution/simulation

ClO

NaCl/PBE0
Cl/PBE0
NaCl/PBE
Cl/PBE
Experiment
a
b
c

NaO

r max ()

r min ()

N coord

r max ()

r min ()

N coord

3:13  0:04
3:13  0:02
3:12  0:04
3:13  0:03
3:11  0:03a
3:16  0:11c

3:82  0:15
3:78  0:08
3:73  0:11
3:73  0:07

6:4  0:7
6:2  0:5
5:9  0:8
6:0  0:5
6:4  1:0a
6:9  1:0c

2:41  0:03

2:42  0:01

2:43  0:02b
2:34  0:14c

3:27  0:07

3:26  0:12

5:3  0:2

5:3  0:5

5:4  0:1b
5:3  0:8c

X-ray diffraction experiment of Ref. [66].


X-ray diffraction experiment of Ref. [67].
Empirical potential structure renement of neutron diffraction experiment of Ref. [13].

g OO(r)

apparent visual modication of the OO RDFs in Ref. [3] occurred


only at salt concentrations greater than 1:4 M, while theoretical
studies employed sub-molar concentrations of the salt. Another
cause of the discrepancy may be the structural renement procedure used in Ref. [3], which relied on empirical potentials.

NaCl, PBE
NaCl, PBE0RSB
Experiment

2
4. Electronic properties of solutions

0
2

r []
Figure 8. Oxygenoxygen radial distribution functions [g OO r] obtained for 0.87
and 1.03 M solutions of NaCl simulated with PBE and PBE0-RSB functionals,
respectively. Both simulations used HSCV pseudopotentials for D, O, and Cl, and
TroullierMartins pseudopotential for Na. Trajectories for the NaCl solution were
obtained in an NVT ensemble at the target temperatures of 380 K (PBE0) and 400 K
(PBE). The RDFs were obtained by averaging over two independent trajectories of
21 ps (PBE) and 25 ps (PBE0). The experimental data at room temperature for pure
water is from Ref. [64].

coordinated by oxygen atoms only, with both hydrogen atoms


pointing away, which results in a large and broad tilt-angle
distribution.
The coordination numbers obtained in our study suggest that
the Na+ and Cl ions are well-separated in all PBE and PBE0 runs
presented here for NaCl [68]. To further verify the ion separation,
we analyzed the distance between the ions along the two 25-ps
PBE0 and 21-ps PBE trajectories. We found that the average distance between Na+ and Cl in the PBE (PBE0) run is 5.31 (5.11) .
The distance between Na+ and Cl in the rock-salt structure is
2.82 , hence in our simulations, the ions appear to be separated
by at least one solvation shell and be practically independent of
each other.
The results of our analysis are in agreement with those of other
ab initio MD simulations of aqueous solutions of ions and salts [20
23,25], which reported that the presence of the solute does not signicantly alter the radial distribution function of solvent. Unlike
Mancinelli et al. [3] who observed signicant visual changes in
the oxygenoxygen radial distribution functions when adding NaCl
to water, our study indicated only a minor softening of the OO
RDF. The difference between theoretical studies [2022] (including
the present work) and Ref. [3] could be due, at least in part, to a
smaller concentration of ions studied theoretically. We note that

We now turn to the comparison of the electronic properties of


solutions. We focus on the band-gaps and positions of the electronic levels of Cl and Na+ with respect to the top occupied level
of water. These quantities are summarized in Table 2 and Figure 9
for the PBE0 and PBE simulations of NaCl and Cl solutions. As
expected, the electronic properties predicted by PBE0 are more
accurate than those computed by PBE, regardless of the system
and the trajectory analyzed, although the agreement with experiment is only qualitative [46]. On average, when the PBE0 functional is used, the band-gaps are increased by 2.22.3 eV. The
relative positions of Cl states with respect to the top water level
also increase, by  0:40 eV. This change is sufcient to ensure a
qualitative agreement with experiment for Cl, according to which

Table 2
Electronic band gaps and positions of Cl and Na+ single-particle energy levels with
respect to the top occupied level of water (valence band maximum, VBM). The
solutions (simulations) are 0.87 M NaCl (PBE), 1.03 M NaCl (PBE0), and 0.87 M Cl
(PBE and PBE0). Electronic levels were assigned by the analysis of maximally localized
Wannier functions. eX denotes the energy of the top occupied level localized on a
Cl or Na+ ion. All energies were computed as an average over 128 snapshots equally
spaced along the trajectories (sampling every 0.10.2 ps). The standard deviation of
the mean for band gaps is 0.030.05 eV, for Cl levels is 0.020.03 eV, and for Na+
levels is above 0.1 eV.
Levels/simulation

Eg , eV

eHOMO X  eVBM D2 O, eV
X = Cl

X = Na+

NaCl
Cl

PBE/PBE

4.10
4.08

0.19
0.24

21.12

NaCl
Cl

PBE/PBE0

4.16
4.20

0.12
0.07

21.01


NaCl
Cl

PBE0/PBE

6.37
6.38

0.22
0.18

21.47

NaCl
Cl

PBE0/PBE0

6.42
6.40

0.30
0.34

21.35

8.7a

1.25b

24.24b

Solution

Experiment
a

Ref. [69].
Relative positions of levels were computed by subtracting the electron binding
energies of either Cl (9:6  0:07 eV) or Na+ (35:4  0:04 eV) in a 3 M NaCl solution
[70] from that of liquid water (11:16  0:04) [71].
b

A.P. Gaiduk et al. / Chemical Physics Letters 604 (2014) 8996

Figure 9. Schematic representation of PBE (left) and PBE0 (right) KohnSham


energy levels computed for a  1 M NaCl solution (see text). Eg denotes the solution
band gap computed as the energy difference of single-particle states. Dashed lines
correspond to the top occupied levels localized primarily on Cl and Na atoms, and
red lled bands depict the positions of the electronic levels of water. The PBE and
PBE0 band structures are not aligned with respect to each other.

the chlorine top occupied level is above the water VBM [70]. The
improvement seems to be more pronounced for the Cl solution
(0.120.17 eV) than for the NaCl solution (0.050.08 eV).
It is easy to explain why PBE0 improves the band-gaps more
than the positions of ionic energy levels. The band-gap is computed
as a difference between the lowest unoccupied and highest occupied KohnSham orbital energies, whereas the relative positions
of ionic and water levels depend solely on occupied orbital energies. Due to the self-interaction error of generalized gradient
approximations such as PBE, virtual orbital energies are particularly inaccurate. The non-local exchange part of the PBE0 functional reduces the self-interaction error, so the functional
improves the virtual orbital energies and yields more accurate
band-gaps.
5. Conclusions
In summary, we carried out ab initio molecular dynamics
simulations of sodium chloride dissolved in water, using both
semilocal and hybrid functionals. For  1 M salt concentration,
we found no signicant effect of the cation presence on the
structural and electronic properties of the anion, compared to
the innite dilution case, both at the PBE and PBE0 levels of
theory. We also found that the presence of the solute does not
signicantly alter the OO radial distribution function of liquid
water at concentrations below or equal to 1 M. Our results are
in agreement with those of other ab initio studies of sub-molar
solutions of salts [16,2023,25].
We observed differences between the average results predicted
by the PBE and PBE0 functionals. The hybrid functional yields a
lower (by 0.150.20) number of hydrogen bonds per water molecule in the solutions studied here. The oxygenoxygen radial distribution functions are softer than the ones obtained with PBE,
indicating a less structured liquid, consistent with our previous
studies performed using hybrid functionals [37,39]. In addition,
PBE0 predicts a water coordination number for Cl in closer agreement with experiments.
Our simulations also showed that addition of Na+ counterions
does not alter the Cl solution band gaps and the positions of occupied electronic states of Cl with respect to the valence band of
waterwe obtained similar energies in both Cl and NaCl-containing solutions. The choice of the functional, on the other hand, has
an important effect on the positions of the electronic energy levels.
Our results are in agreement with the analysis of Zhang et al. [46]
the top energy level of Cl is systematically above the valence band
maximum of water only when PBE0 is used to calculate the
KohnSham eigenvalues along the MD trajectory. These results

95

are in qualitative agreement with experiment [70], although quantitative discrepancies remain. We also found that on average, the
positions of levels are improved when the hybrid functional is used
both for the trajectories and the calculation of KohnSham
eigenvalues.
Finally, the present study demonstrates that sufciently long
ab initio simulations (3050 ps for 64-molecule cells) with hybrid
functionals are now feasible, using an efcient and accurate recursive subspace bisection approximation [47,59]. This technique
accelerates computations with PBE0 by a factor of 210 and does
not alter the structural and electronic properties of the solutions.
Studies of more complex solutions using recursive bisection and
hybrid functionals, including the effects of solutesolvent interactions, are underway.
Acknowledgments
We thank Dr. Eric Schwegler, Dr. Ding Pan, Quan Wan, T. Anh
Pham, and Nicholas Brawand for helpful discussions. Our work
was supported by DOE/BES (Grant No. DE-SC0008938). This
research used resources of the Argonne Leadership Computing
Facility at Argonne National Laboratory, which is supported by
the Ofce of Science of the U.S. Department of Energy under contract DE-AC0206CH11357.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.cplett.2014.04.
037.
References
[1] H. Ohtaki, T. Radnai, Chem. Rev. 93 (3) (1993) 1157, http://dx.doi.org/10.1021/
cr00019a014.
[2] Y. Marcus, Chem. Rev. 109 (3) (2009) 1346, http://dx.doi.org/10.1021/
cr8003828.
[3] R. Mancinelli, A. Botti, F. Bruni, M.A. Ricci, A.K. Soper, Phys. Chem. Chem. Phys.
9 (23) (2007) 2959, http://dx.doi.org/10.1039/b701855j.
[4] A.W. Omta, M.F. Kropman, S. Woutersen, H.J. Bakker, Science 301 (5631)
(2003) 347, http://dx.doi.org/10.1126/science.1084801.
[5] K.J. Tielrooij, N. Garcia-Araez, M. Bonn, H.J. Bakker, Science 328 (5981) (2010)
1006, http://dx.doi.org/10.1126/science.1183512.
[6] J.D. Smith, R.J. Saykally, P.L. Geissler, J. Am. Chem. Soc. 129 (45) (2007) 13847,
http://dx.doi.org/10.1021/ja071933z.
[7] I. Juurinen, T. Pylkknen, K.O. Ruotsalainen, C.J. Sahle, G. Monaco, K.
Hmlinen, S. Huotari, M. Hakala, J. Phys. Chem. B 117 (51) (2013) 16506,
http://dx.doi.org/10.1021/jp409528r.
[8] M. Kondoh, Y. Ohshima, M. Tsubouchi, Chem. Phys. Lett. 591 (2014) 317,
http://dx.doi.org/10.1016/j.cplett.2013.11.055.
[9] J. Cheng, M. Sprik, Phys. Chem. Chem. Phys. 14 (32) (2012) 11245, http://
dx.doi.org/10.1039/c2cp41652b.
[10] Y. Ping, D. Rocca, G. Galli, Chem. Soc. Rev. 42 (6) (2013) 2437, http://dx.doi.org/
10.1039/c3cs00007a.
[11] S. Chen, L.-W. Wang, Chem. Mater. 24 (18) (2012) 3659, http://dx.doi.org/
10.1021/cm302533s.
[12] R. Leberman, A.K. Soper, Nature 378 (6555) (1995) 364, http://dx.doi.org/
10.1038/378364a0.
[13] R. Mancinelli, A. Botti, F. Bruni, M.A. Ricci, A.K. Soper, J. Phys. Chem. B 111 (48)
(2007) 13570, http://dx.doi.org/10.1021/jp075913v.
[14] A.W. Omta, M.F. Kropman, S. Woutersen, H.J. Bakker, J. Chem. Phys. 119 (23)
(2003) 12457, http://dx.doi.org/10.1063/1.1623746.
[15] K.J. Tielrooij, S.T. van der Post, J. Hunger, M. Bonn, H.J. Bakker, J. Phys. Chem. B
115 (43) (2011) 12638, http://dx.doi.org/10.1021/jp206320f.
[16] D.A. Schmidt, R. Scipioni, M. Boero, J. Phys. Chem. A 113 (27) (2009) 7725,
http://dx.doi.org/10.1021/jp9016932.
[17] S. Obst, H. Bradaczek, J. Phys. Chem. 100 (39) (1996) 15677, http://dx.doi.org/
10.1021/jp961384b.
[18] J. Chandrasekhar, W.L. Jorgensen, J. Chem. Phys. 77 (10) (1982) 5080, http://
dx.doi.org/10.1063/1.443682.
[19] J. Chandrasekhar, D.C. Spellmeyer, W.L. Jorgensen, J. Am. Chem. Soc. 106 (4)
(1984) 903, http://dx.doi.org/10.1021/ja00316a012.
[20] J.A. White, E. Schwegler, G. Galli, F. Gygi, J. Chem. Phys. 113 (11) (2000) 4668,
http://dx.doi.org/10.1063/1.1288688.
[21] F.C. Lightstone, E. Schwegler, R.Q. Hood, F. Gygi, G. Galli, Chem. Phys. Lett. 343
(56) (2001) 549, http://dx.doi.org/10.1016/S0009-2614(01)00735-7.

96

A.P. Gaiduk et al. / Chemical Physics Letters 604 (2014) 8996

[22] F.C. Lightstone, E. Schwegler, M. Allesch, F. Gygi, G. Galli, ChemPhysChem 6 (9)


(2005) 1745, http://dx.doi.org/10.1002/cphc.200500053.
[23] T. Ikeda, M. Boero, K. Terakura, J. Chem. Phys. 126 (3) (2007) 034501, http://
dx.doi.org/10.1063/1.2424710.
[24] F.A. Hamprecht, A.J. Cohen, D.J. Tozer, N.C. Handy, J. Chem. Phys. 109 (15)
(1998) 6264, http://dx.doi.org/10.1063/1.477267.
[25] R. Scipioni, D.A. Schmidt, M. Boero, J. Chem. Phys. 130 (2) (2009) 024502,
http://dx.doi.org/10.1063/1.3054197.
[26] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 77 (18) (1996) 3865, http://
dx.doi.org/10.1103/PhysRevLett.77.3865.
[27] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 78 (7) (1997) 1396, http://
dx.doi.org/10.1103/PhysRevLett.78.1396.
[28] A.D. Becke, Phys. Rev. A 38 (6) (1988) 3098, http://dx.doi.org/10.1103/
PhysRevA.38.3098.
[29] C. Lee, W. Yang, R.G. Parr, Phys. Rev. B 37 (2) (1988) 785, http://dx.doi.org/
10.1103/PhysRevB.37.785.
[30] J.P. Perdew, What do the KohnSham orbital energies mean? How do atoms
dissociate?, in: R.M. Dreizler, J. da Providncia (Eds.), Density Functional
Methods in Physics, Plenum, New York, 1985, p. 265, http://dx.doi.org/
10.1007/978-1-4757-0818-9_10.
[31] A.J. Cohen, P. Mori-Snchez, W. Yang, Science 321 (5890) (2008) 792, http://
dx.doi.org/10.1126/science.1158722.
[32] A.J. Cohen, P. Mori-Snchez, W. Yang, Chem. Rev. 112 (2012) 289, http://
dx.doi.org/10.1021/cr200107z.
[33] A.D. Becke, J. Chem. Phys. 98 (1993) 5648, http://dx.doi.org/10.1063/1.464913.
[34] J.P. Perdew, M. Ernzerhof, K. Burke, J. Chem. Phys. 105 (22) (1996) 9982, http://
dx.doi.org/10.1063/1.472933.
[35] C. Adamo, V. Barone, J. Chem. Phys. 110 (13) (1999) 6158, http://dx.doi.org/
10.1063/1.478522.
[36] P.J. Stephens, F.J. Devlin, C.F. Chabalowski, M.J. Frisch, J. Phys. Chem. 98 (45)
(1994) 11623, http://dx.doi.org/10.1021/j100096a001.
[37] T. Todorova, A.P. Seitsonen, J. Hutter, I.-F.W. Kuo, C.J. Mundy, J. Phys. Chem. B
110 (8) (2006) 3685, http://dx.doi.org/10.1021/jp055127v.
[38] M. Guidon, F. Schiffmann, J. Hutter, J. VandeVondele, J. Chem. Phys. 128 (21)
(2008) 214104, http://dx.doi.org/10.1063/1.2931945.
[39] C. Zhang, D. Donadio, F. Gygi, G. Galli, J. Chem. Theory Comput. 7 (5) (2011)
1443, http://dx.doi.org/10.1021/ct2000952.
[40] I. Duchemin, F. Gygi, Comput. Phys. Commun. 181 (5) (2010) 855, http://
dx.doi.org/10.1016/j.cpc.2009.12.021.
[41] C. Adriaanse, J. Cheng, V. Chau, M. Sulpizi, J. VandeVondele, M. Sprik, J. Phys.
Chem. Lett. 3 (23) (2012) 3411, http://dx.doi.org/10.1021/jz3015293.
[42] C.W. Swartz, X. Wu, Phys. Rev. Lett. 111 (8) (2013) 087801, http://dx.doi.org/
10.1103/PhysRevLett. 111.087801.
[43] P. Hunt, M. Sprik, ChemPhysChem 6 (9) (2005) 1805, http://dx.doi.org/
10.1002/cphc.200500006.
[44] D. Prendergast, J.C. Grossman, G. Galli, J. Chem. Phys. 123 (1) (2005) 014501,
http://dx.doi.org/10.1063/1.1940612.
[45] L. Ge, L. Bernasconi, P. Hunt, Phys. Chem. Chem. Phys. 15 (31) (2013) 13169,
http://dx.doi.org/10.1039/c3cp50652e.
[46] C. Zhang, T.A. Pham, F. Gygi, G. Galli, J. Chem. Phys. 138 (18) (2013) 181102,
http://dx.doi.org/10.1063/1.4804621.
[47] F. Gygi, I. Duchemin, J. Chem. Theory Comput. 9 (1) (2013) 582, http://
dx.doi.org/10.1021/ct3007088.
[48] R. Atta-Fynn, E.J. Bylaska, W.A. de Jong, J. Phys. Chem. Lett. 4 (13) (2013) 2166,
http://dx.doi.org/10.1021/jz400887a.
[49] M. Bhl, G. Schreckenbach, N. Sieffert, G. Wipff, Inorg. Chem. 48 (21) (2009)
9977, http://dx.doi.org/10.1021/ic901298q.
[50] J.S. Hub, B.L. de Groot, H. Grubmller, G. Groenhof, J. Chem. Theory Comput. 10
(1) (2014) 381, http://dx.doi.org/10.1021/ct400626b.
[51] H.J. Kulik, N. Marzari, A.A. Correa, D. Prendergast, E. Schwegler, G. Galli, J. Phys.
Chem. B 114 (29) (2010) 9594, http://dx.doi.org/10.1021/jp103526y.
[52] Qbox code: http://eslab.ucdavis.edu/software/qbox/ (retrieved April 13, 2014).
[53] D. Hamann, M. Schlter, C. Chiang, Phys. Rev. Lett. 43 (20) (1979) 1494, http://
dx.doi.org/10.1103/PhysRevLett. 43.1494.
[54] D. Vanderbilt, Phys. Rev. B 32 (12) (1985) 8412, http://dx.doi.org/10.1103/
PhysRevB.32.8412.
[55] HSCV pseudopotential table: http://fpmd.ucdavis.edu/potentials/index.htm
(retrieved April 13, 2014).
[56] G. Bussi, D. Donadio, M. Parrinello, J. Chem. Phys. 126 (1) (2007) 014101,
http://dx.doi.org/10.1063/1.2408420.
[57] J.C. Grossman, E. Schwegler, E.W. Draeger, F. Gygi, G. Galli, J. Chem. Phys. 120
(1) (2004) 300, http://dx.doi.org/10.1063/1.1630560.
[58] E. Schwegler, J.C. Grossman, F. Gygi, G. Galli, J. Chem. Phys. 121 (11) (2004)
5400, http://dx.doi.org/10.1063/1.1782074.
[59] F. Gygi, Phys. Rev. Lett. 102 (16) (2009) 166406, http://dx.doi.org/10.1103/
PhysRevLett. 102.166406.
[60] A.P. Gaiduk, F. Gygi, G. Galli, unpublished.
[61] G.E. Scuseria, V.N. Staroverov, Progress in the development of exchangecorrelation functionals, in: C.E. Dykstra, G. Frenking, K.S. Kim, G.E. Scuseria
(Eds.), Theory and Applications of Computational Chemistry: The First Forty
Years,
Elsevier,
Amsterdam,
2005,
http://dx.doi.org/10.1016/B978044451719-7/50067-6.
[62] A. Luzar, D. Chandler, J. Chem. Phys. 98 (10) (1993) 8160, http://dx.doi.org/
10.1063/1.464521.
[63] D. Prada-Gracia, R. Shevchuk, F. Rao, J. Chem. Phys. 139 (8) (2013) 084501,
http://dx.doi.org/10.1063/1.4818885.

[64] L.B. Skinner, C. Huang, D. Schlesinger, L.G.M. Pettersson, A. Nilsson, C.J.


Benmore, J. Chem. Phys. 138 (7) (2013) 074506, http://dx.doi.org/10.1063/
1.4790861.
[65] A. Chandra, Phys. Rev. Lett. 85 (4) (2000) 768, http://dx.doi.org/10.1103/
PhysRevLett. 85.768.
[66] L.X. Dang, G.K. Schenter, V.-A. Glezakou, J.L. Fulton, J. Phys. Chem. B 110 (47)
(2006) 23644, http://dx.doi.org/10.1021/jp064661f.
[67] T. Megyes, S. Blint, T. Grsz, T. Radnai, I. Bak, P. Sipos, J. Chem. Phys. 128 (4)
(2008) 044501, http://dx.doi.org/10.1063/1.2821956.
[68] J. Timko, D. Bucher, S. Kuyucak, J. Chem. Phys. 132 (11) (2010) 114510, http://
dx.doi.org/10.1063/1.3360310.
[69] A. Bernas, C. Ferradini, J.-P. Jay-Gerin, Chem. Phys. 222 (23) (1997) 151,
http://dx.doi.org/10.1016/S0301-0104(97)00213-9.
[70] B. Winter, R. Weber, I.V. Hertel, M. Faubel, P. Jungwirth, E.C. Brown, S.E.
Bradforth, J. Am. Chem. Soc. 127 (19) (2005) 7203, http://dx.doi.org/10.1021/
ja042908l.
[71] B. Winter, R. Weber, W. Widdra, M. Dittmar, M. Faubel, I.V. Hertel, J. Phys.
Chem. A 108 (14) (2004) 2625, http://dx.doi.org/10.1021/jp030263q.
Alex Gaiduk obtained his B.Sc. from Belarus State
University in 2008 and his Ph.D. from the University of
Western Ontario, Canada in 2013, working with Professor Viktor Staroverov on theory of model KohnSham
potentials. After obtaining doctoral degree, he joined
the group of Professor Giulia Galli at University of
California, Davis as a postdoctoral scholar, to study
aqueous solutions from ab initio molecular dynamics
simulations. In 2014, he moved to the University of
Chicago to continue working in Professor Gallis group
at the Institute for Molecular Engineering. Alex Gaiduk
is a recipient of the 2013 Paul de Mayo Award for
Excellence in Chemical Research.

Cui Zhang received a Ph.D. in Chemistry from the University of California, Davis in 2012, working with Prof.
Galli. In 2013 she joined the group of Prof. Selloni in the
Chemistry Department of Princeton University where
she is working on ab initio simulations of liquids and
nanostructures.

Franois Gygi received a M.Sc. in Physics from the Swiss


Federal Institute of Technology in 1983 and a Ph.D. in
Physics from the same institution in 1988. He then
joined AT&T Bell Laboratories, Murray Hill, NJ (1989
1990) and IBM Research Laboratory, Zurich, Switzerland
(19911992) as a post-doctoral researcher. From 1993
to 1998 he was a senior Researcher at the Swiss Federal
Institute of Technology. In 1998 he moved to LLNL where
he was project leader in the Center for Applied Scientic
Computing. In 2005 he was appointed Professor of
Applied Science at the University of California, Davis,
and in 2009 he was appointed Professor of Computer
Science. He is a fellow of the American Physical Society, the recipient of the LLNL
award of science and technology (2004), and of the Gordon Bell Award (2006).

Giulia Galli is the Liew Family Professor of electronic


structure and simulations in the Institute for Molecular
Engineering, at the University of Chicago, and Senior
Scientist at Argonne National Laboratory. She holds a
Ph.D. in Physics from the International School of
Advanced Studies in Trieste, Italy. Prior to joining the
University of Chicago, she was Professor of Chemistry
and Physics at the University of California at Davis, and
prior to that the head of the Quantum Simulations
Group at the Lawrence Livermore National Laboratory.
She is a Fellow of the American Physical Society and of
the American Academy of Arts and Sciences. She is the
recipient of an award of excellence from the Department of Energy and of the
Science and Technology Award from the Lawrence Livermore National Laboratory.

Das könnte Ihnen auch gefallen