1: \documentclass[twocolumn,preprintnumbers,amsmath,amssymb]{revtex4}

2: %\usepackage{times}

3: \usepackage{tabularx}

4: \usepackage{rotating}

5: \usepackage{graphicx}

6: \usepackage{epsfig}   

7: \usepackage{amsmath,amssymb,bbm}

8: %\usepackage{doublespace}

9: \usepackage{anysize}

10: \usepackage{subfigure}

11: 

12: 

13: \marginsize{1in}{1in}{1in}{1in}

14: 

15: \newcommand{\parc}[2]{\frac{\partial #1}{\partial #2}} 

16: \newcommand{\units}{\, \mathrm}

17: \renewcommand{\thefootnote}{\fnsymbol{footnote}}

18: 

19: 

20: 

21: 

22: \begin{document}

23: 

24: \title{Adaptive Resolution Molecular Dynamics Simulation:\\ Changing the

25: Degrees of Freedom on the Fly}

26: 

27: 

28: \author{Matej Praprotnik}

29: \altaffiliation{On leave from National Institute of Chemistry, Hajdrihova 19,

30:                    1000 Ljubljana, Slovenia. Electronic Mail: praprot@cmm.ki.si.}

31: \author{Luigi Delle Site}

32: \author{Kurt Kremer}

33: \affiliation{%

34: Max-Planck-Institut f\"ur Polymerforschung, Ackermannweg 10, D-55128 Mainz, Germany

35: }%

36: 

37: 

38: 

39: \date{\today}% It is always \today, today,

40:              %  but any date may be explicitly specified

41: 

42: 

43: \begin{abstract}

44:  We present a new adaptive resolution

45: technique for efficient particle-based multiscale molecular

46: dynamics (MD) simulations. The presented approach is tailor-made

47: for molecular systems where atomistic resolution is required only

48: in spatially localized domains whereas a lower mesoscopic level of

49: detail is sufficient for the rest of the system. Our method allows

50: an on-the-fly interchange between a given molecule's atomic and

51: coarse-grained level of description, enabling us to reach large

52: length and time scales while spatially retaining atomistic details

53: of the system. The new approach is tested on a model system of a

54: liquid of tetrahedral molecules. The simulation box is divided

55: into two regions: one containing only atomistically resolved

56: tetrahedral molecules, the other containing only one particle

57: coarse-grained spherical molecules. The molecules can freely move

58: between the two regions while changing their level of resolution

59: accordingly. The coarse-grained and the atomistically

60: resolved systems have the same statistical properties at the same

61: physical conditions.

62: \end{abstract}

63: 

64: \maketitle

65: 

66: \section{Introduction}

67: Many problems in complex soft matter systems are inherently

68: multiscale in nature, i.e., microscopic interactions are coupled

69: strongly to meso- and macroscopic properties. Despite the

70: increasing computational power and ongoing efforts to enhance the

71: efficiency of molecular dynamics (MD) integration

72: algorithms\cite{Deuflhard:1999,Minary:2004,

73: Praprotnik:2005,Praprotnik:2005:1,Praprotnik:2005:2,Praprotnik:2004},

74: all-atom MD simulations are often incapable to cover the time and

75: length  scales needed in order to reach relaxation in a typical

76: molecular system, such as a polymer solution or melt. In many

77: cases it is also questionable, whether the huge amount of detail

78: information might not even obscure the relevant structural

79: information. On the other hand, details of the chemistry do not

80: affect universal power laws but the prefactors of these power

81: laws, which can vary by several orders of magnitude themselves.

82: Thus even on the more coarse-grained level it is advisable to keep

83: a link to the specific chemistry under

84: investigation\cite{kremer:2001,kremer:2004}. In addition typical

85: soft matter systems can be quite inhomogeneous in a way, that

86: different regions within one system are sufficiently described by

87: more or less detail. A consistent and at the same time highly

88: efficient ansatz to understand modern soft matter systems (both of

89: synthetic as well as biological origin) has to take such

90: considerations into account. One first way to tackle this, was to

91: reduce the number of degrees of freedom by a systematic

92: coarse-graining, which retains only those degrees of freedom that

93: are relevant for the particular property of interest. Examples of

94: molecular systems where the coarse-graining approach has been used

95: with much success are fluids\cite{Ayton:2004}, lipid bilayers

96: \cite{Kranenburg:2003,Nielsen:2004,Marrink:2004,Chang:2005,Cook:2005},

97: and polymer

98: systems\cite{Kremer:1990,Tschop:1998,Tschop:1998:2,Everaers:2004}.

99: 

100: 

101: Since some specific chemical details are usually lost in the

102: coarse-graining procedure, much effort has been devoted recently

103: to the development of multiscale modeling approaches, where

104: different parts of the system are modeled at different levels of

105: detail to account for the local resolution

106: requirement\cite{Chun:2000,Malevanets:2000,Villa:2005,DelleSite:2002,Abrams:2003,DelleSite:2004}.

107: In the dual-resolution modeling approach for studying the behavior

108: of polymers near metal

109: surfaces\cite{DelleSite:2002,Abrams:2003,DelleSite:2004}, for

110: example, polymer chain ends that interact with the metal surface

111: are represented partially atomistically while the remaining parts

112: of the polymers, where lower resolution is adequate, are

113: represented as bead-spring chains. However, the switching between

114: different levels of resolution, i.e, the atomic and mesoscopic, is

115: not allowed during the course of that MD simulation, and therefore

116: the initial level of detail and thereby the number of degrees of

117: freedom in the system remain unchanged. Since the chain ends that

118: interact with the metal typically remain close to the surface and

119: they only contain a small fraction of the whole system, an

120: adaptive on-the-fly change of the molecules' resolution is not

121: strictly required for this class of systems. Another approach

122: reported in the literature concerns the link between quantum

123: mechanical and classical MD simulations. In this QM/MM approach a

124: small subset of the system is defined and considered as quantum

125: mechanical, while the rest is treated by a classical force field

126: simulation. Here the atoms as well as the regions of the

127: different regimes are fixed from the very beginning

128: \cite{Laio:2002}, restricting the application to rather specific

129: cases.

130: 

131: 

132: In contrast, MD simulations, in which the spatially localized

133: atomistic domains frequently exchange particles with the remaining

134: mesoscopic part of the system, would allow for much wider

135: applications. Then it would be possible to define certain areas or

136: develop criteria for certain situations, which ask for a more

137: detailed view, while the rest of the system can be treated on the

138: more coarse-grained level. It is the purpose of this paper to

139: present a first attempt of an MD simulation of that kind. Because

140: of that we here at first restrict ourselves to a most elementary

141: model system. Of course one also could resort to Monte Carlo

142: simulations, where different levels of detail are combined. This

143: actually would be somewhat simpler because of the purely

144: stochastic nature of this simulation method. Since we are

145: eventually aiming at molecular systems, where collective motions

146: are crucial, we decided to stick to the MD approach. Existing

147: hybrid MD methods that concurrently couple different length scales

148: have been developed to study solid state systems, where atomistic

149: MD was either combined with the finite elements

150: method\cite{Rafii:1998,Broughton:1999,Smirnova:1999} or it was

151: linked to a quantum mechanical model\cite{Csanyi:2004}. To our

152: knowledge, however, in pursuit of this objective no adaptive

153: hybrid atomistic/mesoscale particle-based MD method, which would

154: allow to dynamically adjust the level of detail, which means the

155: adjustment of the degrees of freedom in the system, has been

156: developed so far.

157: 

158: 

159: In this paper we present a novel adaptive resolution MD scheme

160: that combines a full atomistic description of a desired region of

161: the system with a mesoscale treatment of the remaining part. The

162: key feature of the new method is that it allows to dynamically

163: adapt the level of a given molecule's resolution according to its

164: position in the system. Hence, the number of degrees of freedom is

165: not a conserved quantity in our MD simulations. Furthermore, the

166: presented method is not restricted to couple only the atomistic

167: and mesoscopic levels of detail but can also be applied to systems

168: with mesoscopic domains that are described at different levels of

169: coarse-graining. Here we present a first test case, showing that

170: such an approach is feasible. Therefore the new approach is tested

171: for a simple model liquid consisting of tetrahedral molecules. The

172: simulation box is divided into two regions: one containing

173: ''atomistically'' resolved tetrahedral molecules, the other

174: containing coarse-grained spherical molecules. Molecules are

175: allowed to freely move between the two regions while changing

176: their level of resolution accordingly. The results show that the

177: statistical properties of the corresponding fully atomistic system

178: are accurately reproduced by using the proposed hybrid scheme. In

179: particular, gradients in the chemical potential across the

180: artificial interface where the resolution changes and

181: corresponding spurious fluxes can be avoided. Although we applied

182: the new method here to a generic test system it should find

183: application in more realistic physical systems. There has been an

184: initial attempt reported in Ref. \cite{Abrams:2004}, where MD

185: simulation of an inhomogeneously coarse-grained system of liquid

186: methane has been described. There the starting point were two

187: established models for methane, a five site atomistic and a one

188: site spherical where the interaction between molecules of

189: different species was derived by standard Lennard-Jones mixing

190: rules with the hydrogens of the atomistic invisible to the

191: spherical molecules. In this case it turned out that an effective

192: flux between the two different regimes occurred. This effect is

193: partially due to the different equilibrium state points described

194: by the two models, but in any case in that approach no effective

195: potential between coarse-grained molecules was derived. Our

196: approach differs from that since here we derive such potentials in

197: such a way that the two regimes are in true thermodynamic

198: equilibrium.

199: 

200: 

201: The organization of the article is as follows. In section 2 the

202: methodology is presented, whereby in the first step a

203: coarse-grained model is derived and parameterized from a fully

204: atomistic system and then, in the second step, the atomic and

205: mesoscopic length scales are systematically coupled in a hybrid

206: atomistic/mesoscopic model. The computational details are given in

207: section 3. The results and discussion are presented in section 4,

208: followed by conclusions in section 5.

209: 

210: \section{Methodology}

211: 

212: In this section, the model systems are described and the

213: methodology for the adaptive multiscale MD simulations is

214: presented.

215: 

216: \subsection*{Models}

217: 

218: \subsubsection*{All-Atom Model}

219: 

220: First, we introduce our reference explicit all-atom (\emph{ex})

221: model. Consider a system of $n$ tetrahedral molecules consisting

222: of $N=4$ atoms of the same mass $m_0$ connected by anharmonic

223: bonds as presented in figure \ref{Fig.1} (a) (consider only the

224: right red molecule). 

225: \begin{figure}[!ht]

226: \centering

227: \subfigure[]{\includegraphics[width=7.5cm]{fig1a.ps}}\\

228: \subfigure[]{\includegraphics[width=7.5cm]{fig1b.ps}}

229: \caption{(a) The on-the-fly interchange between the atomic and coarse-grained

230: levels of description. The middle hybrid molecule is a linear

231:  combination of fully atomistic tetrahedral molecule with an additional center of mass

232:  particle representing the coarse-grained molecule. (b) Snapshot of the hybrid atomistic/mesoscopic

233:  model at $\rho^*=0.1$ and $T^*=1$ (LJ units). The red

234: molecules are the explicit atomistically resolved tetrahedral

235: molecules, the blue molecules are the corresponding one particle

236: coarse-grained molecules.}\label{Fig.1}

237: \end{figure}

238: 

239: All atoms in the system interact according to

240: a purely repulsive shifted $12$-$6$ Lennard-Jones potential with a

241: cutoff at $2^{1/6}\sigma$:

242: \begin{multline}

243:  U_{LJ}^{atom}(r_{i\alpha j\beta})=\\\left\{\begin{array}{rc}

244: 4\varepsilon\bigl[\bigl(\frac{\sigma}{r_{i\alpha

245:       j\beta}}\bigr)^{12}-\bigl(\frac{\sigma}{r_{i\alpha j\beta}}\bigr)^6+\frac{1}{4}\bigr];

246: & r_{i\alpha j\beta}\le 2^{1/6}\sigma\\

247:                           0; & r_{i\alpha j\beta}> 2^{1/6}\sigma

248:                              \end{array}\right.\label{eq.8}

249: \end{multline}

250: where $r_{i\alpha j\beta}$ is the distance between the atom $i\alpha$ of the molecule

251: $\alpha$ and the atom $j\beta$ of the molecule $\beta$. We use

252: $\varepsilon$ as a unit of energy. All atoms have the same excluded

253: volume diameter $\sigma$, where $\sigma$ is the unit of length.

254: The neighboring atoms in a given

255: molecule $\alpha$ are linked via an attractive FENE potential

256: \begin{multline}

257:  U_{FENE}^{atom}(r_{i\alpha j\alpha})=\\\left\{\begin{array}{rc}

258:                   -\frac{1}{2}kR_0^2\ln\bigl[1-\bigl(\frac{r_{i\alpha

259:  j\alpha}}{R_0}\bigl)^2\bigr]; & r_{i\alpha j\alpha}\le R_0\\

260:                        \infty; & r_{i\alpha j\alpha}> R_0

261:                   \end{array}\right.\label{eq.9}

262: \end{multline}

263: with divergence length  $R_0=1.5\sigma$ and stiffness

264: $k=30\varepsilon/\sigma^2$, so that the average bond length is

265: approximately $1.0\sigma$ for $k_BT=\varepsilon$, where $T$ is the

266: temperature of the system and $k_B$ is Boltzmann's constant.

267: The functional form of these potentials and their parameters are

268: the same as usually employed in polymer MD simulations\cite{Kremer:1990}.

269: 

270: \subsubsection*{Coarse-Grained Model}

271: 

272: Next, we map the atomistic model to a coarse-grained (\emph{cg})

273: mesoscopic model. For the latter we have chosen a system composed

274: of $n$ one-particle molecules schematically depicted in figure

275: \ref{Fig.1} (a) (consider only the left blue molecule). A given

276: coarse-grained molecule $\alpha$ in the system has a mass

277: $M_\alpha=4m_0$ equal to the total mass of the explicit

278: tetrahedral molecule. All rotational and vibrational degrees of

279: freedom of atomistically resolved tetrahedral molecules are thus

280: removed, and the number of nonbonded interactions is strongly

281: decreased as well.

282: 

283: We shall now find an effective pair potential between

284: coarse-grained molecules such that the structural properties of

285: the underlying atomistic model are reproduced. There is usually no

286: unique way to coarse-grain to an effective pair potential, which

287: is in general temperature and density dependent

288: \cite{Louis:2002,Curtarolo:2002,Klapp:2004}, just as in

289: statistical mechanics there are different ways to perform a

290: renormalization group step. Here, we extract the effective pair

291: potential from a center-of-mass radial distribution function

292: (RDF$_{cm}$) of the reference atomistic model using the potential

293: of mean force $PMF(r)$ as

294:  \begin{equation}

295:   U^{cm}(r)\approx PMF(r)=-k_BT\log g^{cm}_{ex}(r),\label{eq.0}

296: \end{equation}

297: where $g^{cm}_{ex}(r)$ is the RDF$_{cm}$ of the all-atom system

298: and $U^{cm}(r)$ is the derived effective pair

299: potential\cite{Reith:2003}. The effective potential obtained in

300: this way is correct only in the limit of zero density, where the

301: many-body contributions vanish. For systems with nonzero densities

302: in principle many-body interactions would be needed just as for

303: the classical renormalization group theory in statistical

304: mechanics. In a similar spirit we here resort to the (expected)

305: relevant part in order to gain a significant speed up in our

306: simulations. Because of that we use the PMF as the initial guess

307: for the effective pair potential in systems with nonzero density.

308: Then this is further fine-tuned until the RDF$_{cm}$s and 

309: pressures of the reference atomistic and coarse-grained systems

310: match\cite{Reith:2003}. As it turns out (see the Results and

311: Discussion section) the effective pair potential acting between

312: our coarse-grained molecules is significantly softer than the pair

313: potential between atoms of the resolved molecules in accordance

314: with the results previously found in the

315: literature\cite{Klapp:2004,Bryant:2002}.

316: 

317: 

318: 

319: \subsubsection*{Transition Regime: Hybrid Atomistic/Mesoscopic Model}

320: Let us now introduce a hybrid explicit/coarse-grained

321: (\emph{ex-cg}) model. Consider a box of $n$ molecules where one

322: half of the box is occupied by atomistically resolved four-atom

323: tetrahedral molecules while the other half is filled up with the

324: same number of corresponding coarse-grained one-particle molecules

325: as schematically presented in figure \ref{Fig.1} (b). The two

326: domains exchange molecules which adapt their level of resolution

327: accordingly. To ensure that the transition between the two

328: different levels of description is smooth, i.e., the rotational and

329: vibrational degrees of freedom of a given molecule are gradually

330: 'switched on/off' as it crosses the boundary between the atomistic

331: and coarse-grained domains, we also introduce an interface layer

332: for handshaking between atomistic and mesoscale regions. In this

333: regime also the rotational and vibrational velocities have to be

334: reintroduced in a consistent way. Due to the periodic boundary

335: conditions employed in our simulations there are in fact two such

336: layers as depicted in figure \ref{Fig.1} (b).

337: 

338: Since the total number of molecules $n$ in the system is a

339: conserved quantity in our constant temperature simulations, we

340: sample the phase space from the $nVT$ ensemble. However, the total

341: number of degrees of freedom is not constant in this model.

342: 

343: An alternative way compared to the similarities with

344: renormalization group methods, which probably describes the

345: situation even better, is the comparison to a first order phase

346: transition. The rotational and vibrational part of the free energy

347: then can be viewed as the latent heat at this transition.  At

348: equilibrium, conditions analogous to two phase coexistence,

349: \begin{equation}

350:        \mu_{ex}=\mu_{cg},~~~ p_{ex}=p_{cg},~~~ T_{ex}=T_{cg},\label{eq.1}

351: \end{equation}

352: must be automatically satisfied, where $\mu_{ex}$, $p_{ex}$,

353: $T_{ex}$ and $\mu_{cg}$, $p_{cg}$, $T_{cg}$ are the chemical

354: potentials, pressures, and temperatures of the liquid in the

355: atomistic and coarse-grained domains, respectively. These

356: conditions (\ref{eq.1}) assure that there is no net flux of

357: molecules between the atomistic and coarse-grained regions. To

358: keep this absolute requirement also then defines a central task.

359: This guarantees that the liquid is homogeneous across the box as

360: it is in the reference fully atomistic system. From a

361: molecular point of view, the artificial resolution boundary must

362: be essentially invisible, i.e., the molecules have to cross the border

363: without experiencing any `barrier'. Our approach to reach this

364: objective is presented in the proceeding subsection.

365: 

366: 

367: \subsection*{Adaptive Resolution Scheme}

368: 

369: 

370: To allow a coarse-grained molecule to find an energetically

371: permissible orientation  with respect to its neighboring

372: mo\-le\-cu\-les (when it leaves the coarse-grained domain and is

373: remapped into the atomistically resolved four-atom tetrahedral

374: molecule) we introduce an interface layer between the atomistic

375: and coarse-grained regions, which contains 'hybrid' molecules as

376: presented in figure \ref{Fig.1}. Each hybrid molecule

377: schematically shown in figure \ref{Fig.1}(a) (consider the middle

378: molecule) is composed of a tetrahedral molecule with an additional

379: massless center-of-mass particle serving as an interaction site.

380: This is similar to the flexible TIP4P water

381: model\cite{Lawrence:2003} where apart from the interaction sites

382: on the three atoms of a water molecule an additional interaction

383: site is introduced along the symmetry axis between the hydrogen

384: and oxygen atoms.

385: 

386: Thus, each time a coarse-grained molecule $\alpha$ leaves that

387: domain and enters the interface layer, it is remapped first into a

388: hybrid molecule with the same center-of-mass position and a random

389: orientation in space, where the relative positions of the

390: tetrahedral atoms are taken from a molecular configuration

391: corresponding to a randomly chosen molecule from the atomistic

392: regime. Each of the four explicit tetrahedral atoms in the hybrid

393: molecule gains at this remapping a velocity equal to the velocity

394: of the corresponding coarse-grained molecule to maintain the

395: linear momentum of the molecule. In addition, the tetrahedral

396: atoms are also assigned rotational/vibrational velocities

397: corresponding to atoms of a random molecule from the atomistic

398: region, where we subtract the total linear momentum of the latter

399: molecule. In this way we ensure that the kinetic energy is

400: distributed among all degrees of freedom according to the

401: equipartition principle as $k_BT/2$ of average kinetic energy per

402: quadratic degree of freedom while retaining the linear momentum of

403: the coarse-grained molecule. The center-of-mass interaction site

404: moves obeying the constraints:

405: \begin{eqnarray}

406: {\bf R}_\alpha&=&\frac{\sum_{i\alpha} m_{i\alpha}{\bf

407:     r}_{i\alpha}}{M_\alpha},\label{eq.5}\\

408:  {\bf V}_\alpha&=&\frac{\sum_{i\alpha} m_{i\alpha}{\bf v}_{i\alpha}}{M_\alpha},\label{eq.6}

409: \end{eqnarray}

410: where ${\bf R}_\alpha$ is a center of mass of the molecule $\alpha$,

411: ${\bf r}_{i\alpha}$ is the position vector of the explicit

412: tetrahedral atom $i\alpha$ in the molecule $\alpha$, ${\bf

413: V}_\alpha$ is the center-of-mass velocity of the molecule $\alpha$,

414: ${\bf v}_{i\alpha}$ is the velocity of the explicit tetrahedral

415: atom $i\alpha$, and $M_\alpha=\sum_{i\alpha}m_{i\alpha}$ is the

416: total mass of the molecule $\alpha$. In our case,

417: $m_{i\alpha}=m_0$ and $M_\alpha=4m_0$ for all $i\alpha=1,\dots,4$

418: and $\alpha=1,\dots, n$. Each time a hybrid molecule crosses the

419: boundary into atomistic regime it is remapped into a four-particle

420: tetrahedral molecule with the four tetrahedral atoms retaining

421: their current velocities and positions. In this model also the

422: explicit tetrahedral molecules have the center-of-mass interaction

423: sites, but only for the interactions with the hybrid and

424: coarse-grained molecules. Of course, deep in the atomistic region,

425: where the atomistically resolved molecules do not interact

426: anymore with the hybrid and coarse-grained molecules the

427: center-of-mass interaction site can be omitted.  Every time a

428: four-particle tetrahedral molecule leaves the atomistic region and

429: enters into the transition regime it is mapped into a hybrid

430: molecule with the four tetrahedral atoms retaining instantaneous

431: velocities and positions with the center-of-mass interaction site

432: moving according to Eqs. (\ref{eq.5}) and (\ref{eq.6}). Similarly,

433: as a hybrid molecule crosses a boundary to the coarse-grained

434: region it is mapped into a coarse-grained molecule with a velocity

435: equal to the center-of-mass velocity of the hybrid molecule given by Eq.

436: (\ref{eq.6}).

437: 

438: To couple the atomic and mesoscopic length scales we define in the

439: spirit of thermodynamic perturbation

440: approach\cite{Zwanzig:1954,Leach:2001} the total intermolecular

441: force acting between centers of mass of molecules $\alpha$ and

442: $\beta$ as

443: \begin{multline}

444:  {\bf F}_{\alpha\beta}=\\w(X_\alpha)w(X_\beta){\bf

445:  F}_{\alpha\beta}^{atom}+[1-w(X_\alpha)w(X_\beta)]{\bf

446:  F}_{\alpha\beta}^{cm},\label{eq.4}

447: \end{multline}

448: where

449: \begin{equation}

450: {\bf F}_{\alpha\beta}^{atom}=\sum_{i\alpha, j\beta}{\bf F}_{i\alpha\label{eq.4a}

451:  j\beta}^{atom}

452: \end{equation}

453: is the sum of all pair atom interactions between explicit

454: tetrahedral atoms of the molecule $\alpha$ and explicit tetrahedral

455: atoms of the molecule $\beta$ and

456: \begin{eqnarray}

457:  {\bf F}_{i\alpha j\beta}^{atom}&=&- \parc{U^{atom}}{{\bf r}_{i\alpha j\beta}},\\

458:  {\bf F}_{\alpha\beta}^{cm}&=&-\parc{U^{cm}}{{\bf R}_{\alpha\beta}}.

459: \end{eqnarray}

460: The vector ${\bf r}_{i\alpha j\beta}={\bf r}_{i\alpha}-{\bf

461: r}_{j\beta}$ is the relative position vector of atoms $i\alpha$

462: and $j\beta$, ${\bf R}_{\alpha\beta}={\bf R}_{\alpha}-{\bf

463: R}_{\beta}$ is the relative position vector of the centers of mass

464: of the molecules $\alpha$ and $\beta$, $X_\alpha$ and $X_\beta$ are

465: the $x$ center-of-mass coordinates of the molecules $\alpha$ and

466: $\beta$, respectively, and $w$ is the weighting function that

467: determines the 'identity' of a given molecule. The weighting

468: function $w\in [0,1]$ is defined in such a way that values $0<w<1$

469: correspond to a hybrid molecule with extreme cases $w=1$ and $w=0$

470: corresponding to a four-atom tetrahedral molecule and one-particle

471: coarse-grained molecule, respectively. Hence, as soon as one of

472: the two interacting molecules $\alpha$ and $\beta$ is a

473: coarse-grained molecule with no explicit tetrahedral atoms

474: $w(X_\alpha)w(X_\beta)=0$ and ${\bf F}_{\alpha\beta}={\bf

475: F}_{\alpha\beta}^{cm}$. 

476: 

477: We propose the following functional form of the weighting function

478: $w$:

479: \begin{multline}

480:  w(x)=\\\left\{\begin{array}{rc}

481:                                                    1; & d<x\le \frac{a}{2}-d\\

482:                                                    0; & -\frac{a}{2}+d\le x<-d\\

483:                          \sin^2[\frac{\pi}{4d}(x+d)]; & -d\le x\le d\\

484:              \cos^2[\frac{\pi}{4d}(x-\frac{a}{2}+d)]; & \frac{a}{2}-d<x\le\frac{a}{2}\\

485:              \cos^2[\frac{\pi}{4d}(x+\frac{a}{2}+d)]; & -\frac{a}{2}\le x<-\frac{a}{2}+d

486:             \end{array}\right.\label{eq.7}

487: \end{multline}

488: where $a$ is the box length and $d$ the half-width of the interface

489: layer. The weighting function $w$ is shown in figure \ref{Fig.2}. Our

490: choice, which takes into account the periodic boundary conditions,

491:  is a particularly simple way to ensure an interpolation between

492: $w=0$ and $w=1$ that is monotonic, continuous, differentiable and

493: has zero slope at the boundaries to the atomistic and

494: coarse-grained regions. Apart from these requirements we consider

495: the precise functional form as immaterial.

496: \begin{figure}[!ht]

497: \centering

498: \mbox{\includegraphics[width=7.5cm, angle=0]{fig2.ps}}

499:          \caption{ The weighting function $w(x)\in [0,1]$ defined by

500:   Eq. (\ref{eq.7}). The values $w=1$ and $w=0$ correspond to the atomistic

501:   and coarse-grained regions of the hybrid atomistic/mesoscopic system

502:   with the box length $a$, respectively, while the values $0<w<1$ correspond to

503:   the interface layer. Shown is the example where the half-width $d$ of

504:   the interface layer is $d=a/10$. The vertical lines denote the

505:   boundaries of the interface layers.}\label{Fig.2}

506: \end{figure}

507: 

508: 

509: 

510: Exploiting the analogy with the quantum mechanical mixed state

511: description\cite{Schwabl:1995}, one can consider a hybrid molecule

512: in the interface layer as a normalized linear combination of a

513: four-atom tetrahedral molecule and a corresponding one-particle

514: coarse-grained molecule. As a given molecule moves from the

515: coarse-grained boundary of the interface layer to the atomistic

516: boundary, $w$ is gradually changed from $0$ to $1$ and a

517: coarse-grained molecule with only $3$ translational degrees of

518: freedom gradually turns into an atomistically resolved molecule

519: with additional $3N-3=9$ rotational and vibrational degrees of

520: freedom and a defined spatial orientation. The continuous, i.e.,

521: not instantaneous, 'identity' transition is required since a

522: hybrid molecule is given a random orientation at the

523: coarse-grained boundary, and there can consequently be overlaps of

524: its tetrahedral atoms with the atoms of the neighboring molecules.

525: Since at this boundary $w=0$ and the repulsive potential

526: $U^{atom}$ given in Eq. (\ref{eq.8}) is capped (see the

527: Computational Details section), the forces acting on these atoms

528: cannot diverge, however. While moving towards the atomistic region

529: the hybrid molecule slowly adapts its orientation via the

530: gradually increasing atomistic interactions with the neighboring

531: molecules. Likewise, as presented in figure \ref{Fig.1} (a), as a

532: given molecule moves from the atomistic boundary of the interface

533: layer to the coarse-grained boundary, $w$ is continously changing

534: from $1$ to $0$, the fully atomistically resolved molecule

535: gradually turns into the one particle coarse-grained molecule

536: while omitting all rotational and vibrational degrees of freedom

537: and orientation.

538: 

539: 

540: 

541: To elucidate the definition of the force calculation in our

542: model we rewrite Eq. (\ref{eq.4}) as

543: \begin{multline}

544: {\bf F}_{\alpha\beta}=\\

545: w(X_\alpha)w(X_\beta){\bf F}_{\alpha \beta}^{atom}+[1-w(X_\alpha)][1-w(X_\beta)]{\bf

546:   F}_{\alpha\beta}^{cm}\\+[1-w(X_\alpha)]w(X_\beta){\bf

547:   F}_{\alpha\beta}^{cm}+w(X_\alpha)[1-w(X_\beta)]{\bf F}_{\alpha\beta}^{cm}.\label{eq.3}

548: \end{multline}

549: From Eqs. (\ref{eq.4a}) and (\ref{eq.3}) one can then deduce that

550: the pair force in Eq. (\ref{eq.4}) is defined in such a way that

551: two given atoms $i\alpha$ and $j\beta$ in given explicit molecules

552: $\alpha$ and $\beta$ ($w(X_\alpha)=1$ and $w(X_\beta)=1$) interact

553: via the atomistic potential defined by Eqs. (\ref{eq.8}) and

554: (\ref{eq.9}) while two coarse-grained molecules ($w(X_\alpha)=0$

555: and $w(X_\beta)=0$) interact via the corresponding effective pair

556: potential $U^{cm}$. Furthermore, the coarse-grained molecules

557: 'see' the fully atomistically resolved molecules as coarse-grained

558: molecules. Hence, their intermolecular interaction is defined by

559: the effective pair potential $U^{cm}$. To ensure that the

560: center-of-mass dynamics governed by Eqs. (\ref{eq.5}) and

561: (\ref{eq.6}) of a given explicit or hybrid molecule $\alpha$ is

562: correct, the total intermolecular force ${\bf

563: F}_{\alpha\beta}^{cm}$ between the  atomistically resolved

564: molecule $\alpha$ and a coarse-grained molecule $\beta$ is

565: distributed among the explicit atoms of the molecule $\alpha$ as

566: \begin{equation}

567:  {\bf F}_{i\alpha\beta}=\frac{m_{i\alpha}}{\sum_{i\alpha}m_{i\alpha}}{\bf F}_{\alpha\beta}^{cm},\label{eq.4b}

568: \end{equation}

569: where  ${\bf F}_{i\alpha\beta}$ is the force imposed on the

570: explicit tetrahedral atom $i\alpha$ by the coarse-grained molecule

571: $\beta$. The explicit tetrahedral atoms in a given hybrid molecule

572: interact with other explicit atoms in neighboring explicit and

573: hybrid molecules through atomistic forces, while the then massless

574: center-of-mass particle serves as an effective potential

575: interaction site. The total force on a hybrid molecule is then

576: according to Eq. (\ref{eq.4}) a normalized linear combination of

577: atomistic and effective pair forces. 

578: Using

579: \begin{eqnarray}

580:   {\bf F}_{i\alpha j\beta}^{atom}&=&-{\bf F}_{j\beta i\alpha}^{atom},\\

581:   {\bf F}_{\alpha\beta}^{cm}&=&-{\bf F}_{\beta\alpha}^{cm}

582: \end{eqnarray}

583: we obtain from Eq. (\ref{eq.4})

584: \begin{equation}

585: {\bf F}_{\alpha \beta}=-{\bf F}_{\beta \alpha}.

586: \end{equation}

587: The force definition in Eq. (\ref{eq.4}) hence satisfies Newton's Third

588: Law.

589: 

590: Recall that the effective intermolecular potential is determined in such a way that the 

591: equations of state for the \emph{ex} and \emph{cg} models match around the state

592: point considered. Therefore, following the scheme, as given by Eq.

593: (\ref{eq.4}), Eq. (\ref{eq.1}) is implicitly

594: satisfied, due to which spurious fluxes are avoided at the boundary between

595: the atomistic and coarse-grained regimes.

596: 

597: 

598: To summarize, the new adaptive resolution scheme for the hybrid

599: atomistic/mesoscale MD simulations is a two-stage procedure:

600: \begin{enumerate}

601:  \item{Derive the effective

602: pair potential $U^{cm}$ between coarse-grained molecules on the basis of the

603: reference all-atom system.}

604: \item{Introduce the interface layer containing the hybrid molecules that have

605:   additional interaction sites positioned at their centers of mass. 

606:   Define a weighting function $w$ by Eq. (\ref{eq.7}) and

607:   use Eq. (\ref{eq.4}) for the definition of the intermolecular pair forces.

608:   Allow molecules to adapt their level of resolution according to

609:   their position in the system as explained in the second paragraph of

610:   this subsection.}

611: \end{enumerate}

612: Finally, since the switching of the resolution can be considered as a first

613: order phase transition, the adaptive resolution scheme must

614: necessarily be employed in combination with a thermostat. 

615: Because the latent heat is generated in the

616: transition regime it is important to use a thermostat, which 

617: couples locally to the particle motion, e.g., Langevin or Dissipative Particle Dynamics (DPD) thermostats\cite{Soddemann:2003}.

618: 

619: 

620: \section{Computational Details}

621: 

622: 

623: \subsection*{Temperature Calculation}

624: 

625: In order to treat all $n$ molecules equally regardless of the

626: level of detail, we define, using the equipartition principle, the

627: temperature of a system only from translational degrees of

628: freedom. The 'translational' temperature

629: \begin{equation}

630:  T=\frac{2E_k^{cm}}{3nk_B}=\frac{1}{3nk_B}\sum_\alpha M_\alpha{\bf V}^2_\alpha,\label{eq.12}

631: \end{equation}

632: when averaged gives the temperature of the system

633: \cite{Allen:1987}. Here $E_k^{cm}$ is the total translational

634: kinetic energy of the system.  We also checked the ``particle''

635: temperature in the explicit atomistic region. As to be expected we

636: find the same temperature.

637: 

638: \subsection*{Pressure Calculation}

639: 

640: For the same reason also the pressure calculation is based on the

641: molecular instead of atomic interactions:

642: \begin{equation}

643:  p=\frac{1}{V}\biggl[nk_BT-\frac{1}{3}\sum_{\alpha<\beta}{\bf R}_{\alpha\beta}\cdot{\bf F}_{\alpha\beta}\biggr],\label{eq.12a}

644: \end{equation}

645: where ${\bf F}_{\alpha\beta}$ is given by Eqs.

646: (\ref{eq.4})\cite{Berendsen:1984, Allen:1987}. Moreover, using Eq.

647: (\ref{eq.12a}) for pressure evaluation has two additional

648: advantages compared to the pressure calculation based on atomic

649: interactions:  first, Eq. (\ref{eq.12a}) is also valid in the case

650: that forces on atoms involve internal non-pairwise-additive

651: contributions, and second, even if all interactions are pairwise

652: additive (as in our case), the pressure calculation based on

653: atomic interactions introduces additional fluctuations in the

654: pressure\cite{Berendsen:1984}.

655: 

656: 

657: 

658: 

659: \subsection*{Multiscale Simulation Details}

660: 

661: 

662: For computational convenience during the proof-of-principle stage

663: we replace our \emph{ex-cg} model with a model in which the

664: whole box contains exclusively hybrid molecules with four explicit

665: atoms and a center-of-mass interaction site. This has some

666: technical advantages during the tests of the method. The true

667: level of detail of the molecules is then determined from the value

668: of $w(X_{\alpha})$. For later large scale, however, of course the

669: original \emph{ex-cg} model composed of the explicit,

670: coarse-grained, and hybrid molecules will be employed. 

671: Using Eq. (\ref{eq.7}) in

672: the replacement model for distinguishing between explicit,

673: coarse-grained, and hybrid molecules of our original \emph{ex-cg}

674: model we can capture all definitions of pair forces for all

675: classes of particles in our system in a single expression as

676: \begin{multline}

677:  {\bf F}_{i\alpha j\beta}=\\w(X_\alpha)w(X_\beta){\bf F}_{i\alpha

678:  j\beta}^{atom}+[1-w(X_\alpha)w(X_\beta)]\delta_{\alpha,\beta}{\bf F}_{i\alpha

679:  j\beta}^{atom}\\+[1-w(X_\alpha)w(X_\beta)]\frac{m_{i\alpha}m_{j\beta}}{\sum_{i\alpha}m_{i\alpha}\sum_{j\beta}m_{j\beta}}{\bf F}_{\alpha\beta}^{cm},\label{eq.2}

680: \end{multline}

681: where ${\bf F}_{i\alpha j\beta}$ is the total pair force between

682: the explicit atom $i\alpha$ of the molecule $\alpha$ and the

683: explicit atom $j\beta$ of the molecule $\beta$ and

684: $\delta_{\alpha,\beta}$ is the Kronecker symbol. Summing for

685: $\alpha\ne \beta$

686: \begin{equation}

687: {\bf F}_{\alpha\beta}=\sum_{i\alpha, j\beta}{\bf F}_{i\alpha j\beta}

688: \end{equation}

689: we regain the total force between molecules $\alpha$ and $\beta$

690: given in Eq. (\ref{eq.4}). From Eqs. (\ref{eq.5}), (\ref{eq.6}),

691: and (\ref{eq.2}) follows that in the replacement model with only

692: hybrid molecules a hybrid molecule experiences only translational

693: kicks from other molecules in the coarse-grained region ($w=0$)

694: and hence its center of mass moves exactly as the respective

695: one-particle coarse-grained molecule in the original \emph{ex-cg}

696: model. Similarly, in the explicit region ($w=1$) a hybrid molecule

697: experiences only atomistic forces and hence its explicit atoms

698: move exactly as the explicit atoms in the respective tetrahedral molecule in

699: the original model. Therefore, the model containing only hybrid

700: molecules interacting via the pair force defined by Eq.

701: (\ref{eq.2}) together with the applied Langevin

702: thermostat\cite{Kremer:1990} acting on each particle in the system

703: (to assure that the atom velocities are thermalized in accordance

704: with the equipartition principle) exactly mimics the original

705: \emph{ex-cg} model in which the temperature would also be held

706: constant by the Langevin thermostat. From the methodology

707: development point of view, these two models are therefore

708: identical.

709: 

710: 

711: This yields the Langevin equation of motion

712: \begin{equation}

713:  m_i\frac{d^2{\bf r_i}}{dt^2}={\bf F}_i-m_i\Gamma\frac{d{\bf

714:  r}_i}{dt}+{\bf W}_i(t),

715: \end{equation}

716: where $m_i$ is the mass of particle $i$, ${\bf F}_i$ is the total

717: force acting on the respective particle equal to the sum of pair

718: interactions given by Eq. (\ref{eq.2}), $\Gamma$ is the friction

719: constant, and ${\bf W}$ is the random force of a heat

720: bath\cite{Mann:2004}. We sample the random force from a uniform

721: distribution, since it has been shown that there is no advantage

722: of using Gaussian noise for the Langevin

723: thermostat\cite{Duenweg:1991}.

724: 

725: The value of the friction constant used in our simulations is

726: $\Gamma=0.5\tau^{-1}$ where

727: $\tau=(\varepsilon/m_0\sigma^2)^{-1/2}$. The equations of motion

728: are integrated for each particle of the system using the velocity

729: Verlet algorithm with a $0.005\tau$ time step. Here, again only

730: for the purpose of testing the method, we use only one time step

731: in the whole system. In the coarse-grained regime actually a

732: significantly larger time step could be used. Therefore ultimately

733: one would like to introduce a multiple time step algorithm.

734: Simulations are performed at temperature $T=\varepsilon/k_B$ and

735: number density $\rho=n/V=0.1/\sigma^3$. Here $n=5001$ is the

736: number of molecules in the system, which can be either the

737: explicit, coarse-grained or hybrid. If we roughly estimate the

738: excluded volume diameter of the coarse-grained molecule

739: $\sigma_{CG}$ as the distance, where the repulsive effective pair

740: potential between the coarse-grained molecules in our simulations

741: equals $k_BT$, i.e., $\sigma_{CG}\approx 1.7\sigma$, then the

742: number density

743: $\rho=0.1/\sigma^3=0.1(\sigma_{CG}/\sigma)^3/\sigma_{CG}^3\approx

744: 0.5/\sigma_{CG}^3$ corresponds to a medium dense liquid, which is

745: due to the soft effective repulsive interactions rather weakly

746: correlated\cite{Kremer:2005}.

747: %For

748: %comparison a density of a typical dense Lennard-Jones liquid is $

749: %\rho=0.85/\sigma_{CG}^3$.

750: Periodic boundary conditions and the minimum image

751: convention\cite{Allen:1987} are employed. The interaction range in

752: the system is given by the range of the effective pair potential

753: between molecules  and the geometry of tetrahedral molecules, i.e.,

754: the most outer atoms of two tetrahedral molecules with centers of

755: mass slightly less than $2.31\sigma$ apart still experience the

756: effective potential contribution in Eq. (\ref{eq.2}). Hence, the

757: actual interaction range in the system is approximately

758: $3.5\sigma$.

759: 

760: All molecules are initially randomly placed in a cubic box of size

761: $a=36.845\sigma$. To remove the overlaps between them, a $50\tau$

762: long warm-up run is performed during which the repulsive

763: interparticle potential is capped (see capped interactions in Ref.

764: \cite{Espresso:2005}). Thus, at all interparticle distances,

765: which would lead to larger forces between particles than a

766: prescribed maximal force, the forces defined by the original

767: repulsive pair potential are replaced by repulsive central forces

768: of the maximal force magnitude. The latter is gradually increased

769: from $20\varepsilon/\sigma$ to $110\varepsilon/\sigma$ during this

770: warm-up phase. Afterwards an additional $250\tau$ equilibration

771: run is carried out where we set the maximal force magnitude to

772: $10^{9}\varepsilon/\sigma$, which corresponds to interparticle

773: distance of $0.27\sigma$. The chosen maximal force magnitude value

774: is so high that it has no effect on the dynamics of molecules in

775: the atomistic region because there atoms never come this close

776: together. Therefore, by this force capping we only prevent

777: possible force singularities that could emerge due to overlaps

778: with the neighboring molecules when a given molecule enters the

779: interface layer from the coarse-grained side as explained in the

780: previous section. Production runs with the

781: $10^{9}\varepsilon/\sigma$ force capping are then performed for

782: $7500\tau$, storing configurations at each $5\tau$ time interval

783: for analysis. We performed all our MD simulations using the

784: ESPResSo package\cite{Espresso:2005}, developed at our institute.

785: 

786: 

787: The following reduced Lennard-Jones units\cite{Allen:1987} are

788: used throughout:

789:  $m^*=m/m_0$, $r^*=r/\sigma$, $V^*=V/\sigma^3$,

790:  $T^*=k_BT/\varepsilon$, $U^*=U/\varepsilon$, $p^*=p\sigma^3/\varepsilon$,

791:  $\rho^*=n/V^*$, $t^*=t/\tau$, $D^*=D\sqrt{m_0/\varepsilon}/\sigma$,

792: where $D$ is the self-diffusion constant. Note that in our

793: simulations all atoms have a mass $m^*=1$ while all molecules have

794: a mass $M^*=4$.

795: 

796: 

797: 

798: \section{Results and Discussion}

799: 

800: \subsection*{Determination of the Effective Potential}

801: 

802: We have determined the effective nonbonded pair potential

803: ${U^{cm}}^*$ between coarse-grained molecules illustrated in

804: figure \ref{Fig.3} by using the potential of mean force

805: (PMF$_{ex}$) of the \emph{ex} system, Eq. (\ref{eq.0}), at very

806: low number density $\rho^*=0.0025$ as the initial guess. Then we

807: further adjusted it to obtain the adequate agreement between

808: RDF$_{cm}$s of the \emph{ex} and \emph{cg} systems at the

809: $\rho^*=0.1$.

810: \begin{figure}[!ht]

811: \centering

812: \includegraphics[width=7.5cm, angle=0]{fig3.ps}

813: \caption{ The effective pair potential ${U^{cm}}^*$, Eq.

814: (\ref{eq.11}), between the coarse-grained molecules, where the

815: potential of mean force PMF$_{ex}$ of the explicit system at

816: $\rho^*=0.0025$ and $T^*=1$ was used as the initial guess. The

817: presented function ${U^{cm}}^*$ was determined in such a way that

818: the RDF$_{cm}$s of the explicit (\emph{ex}) and coarse-grained (\emph{cg})

819: systems match at the $\rho^*=0.1$.}\label{Fig.3}

820: \end{figure}

821: We have parameterized the effective potential

822: $U^{cm}$ with the Morse potential

823: \begin{equation}

824:  {U^{cm}}^*(r^*)=\gamma^*\{1-\exp[-\kappa^*(r^*-r^*_0)]\}^2\label{eq.11}

825: \end{equation}

826: with parameters $\gamma^*=\gamma/\varepsilon=0.105$, $\kappa^*=\kappa\sigma=2.4$, and $r^*_0=r_0/\sigma=2.31$.

827: As one can see from figure \ref{Fig.3}, the obtained effective

828: potential is softer than the underlying repulsive interatomic interaction

829: potential given by Eqs. (\ref{eq.8}) and (\ref{eq.9}) since it varies

830: more slowly with the interparticle distance. This is a general feature

831: of effective potentials for polyatomic molecular systems\cite{Klapp:2004}.

832: 

833: The obtained RDF$_{cm}$s of the \emph{ex} and \emph{cg} systems at

834: the temperature $T^*=1$ and number density $0.025\le\rho^*\le

835: 0.175$ using for all number densities the same effective potential

836: given by Eq. (\ref{eq.11}) for the pair interactions between the

837: coarse-grained molecules are depicted in figure \ref{Fig.4}. 

838: \begin{figure}[!ht]

839: \centering

840: \includegraphics[height=10cm,width=7.5cm,angle=0]{fig4.ps}

841: \caption{Center-of-mass radial distribution functions of

842: the explicit (\emph{ex}) and coarse-grained (\emph{cg}) systems at

843: the temperature $T^*=1$ and number density $0.025\le\rho^*\le

844: 2.25$.}\label{Fig.4}

845: \end{figure}

846: The

847: RDF$_{cm}$s are calculated in the range $r^*\in [0,5]$ with

848: $\Delta r^*=0.05$. From results presented in figure \ref{Fig.4}

849: it can be observed that although the effective potential given by

850: Eq. (\ref{eq.11}) was parameterized at the number density

851: $\rho^*=0.1$  the agreement of RDF$_{cm}$s of the \emph{cg} system

852: with the corresponding reference RDF$_{cm}$s of the \emph{ex}

853: system is good also for the lower number densities. In fact, due

854: to weaker many-body interactions the agreement is even better for

855: the systems with lower density. Since the effective potentials are

856: density dependent\cite{Louis:2002}, to obtain a better agreement

857: for higher densities the effective potential should be

858: reparameterized\cite{Reith:2003}. Probably the functional form of

859: the effective potential will also change due to increased

860: contribution from many-body interactions.

861: 

862: As a quantitative measure of accuracy of the evaluated RDF$_{cm}$s

863: we define a penalty function $f_p$ as

864: \begin{equation}

865:  f_p=\int [g(r^*)-g^{cm}_{ex}(r^*)]^2\exp(-r^*)\,dr^*,\label{eq.10}

866: \end{equation}

867: where $g^{cm}_{ex}$, which is taken as a reference, is the

868: RDF$_{cm}$ of the \emph{ex} system. The function $\exp(-r^*)$ is

869: employed to penalize more strongly deviations at small

870: distances\cite{Reith:2003}. The values of $f_p$ for \emph{cg}

871: systems with the number density $0.025\le\rho^*\le 0.175$ are

872: reported in table \ref{Tab.1}.

873: \begin{table}[ht]

874:    \centering

875:   \begin{tabular}{||c|c||}

876:     \hline\hline

877:       $\rho^*$  & $f_p\cdot 10^3$        \\

878:     \hline

879:       $0.025$     & $0.0401$  \\

880:       $0.075$     & $0.1101$  \\

881:       $0.1$       & $0.2067$  \\

882:       $0.125$     & $0.4361$  \\

883:       $0.150$     & $0.9257$  \\

884:       $0.175$     & $1.9025$  \\

885:     \hline\hline

886:    \end{tabular}

887:    \caption{Penalty function $f_p$ defined by Eq. (\ref{eq.10}) as a function of number density

888:    $\rho^*$ for RDF$_{cm}$s $g^{cm}_{cg}(r^*)$ of the coarse-grained

889:    systems in which particles are interacting via the effective

890:    potential ${U^{cm}}^*$ given by

891:    Eq. (\ref{eq.11}). The RDF$_{cm}$s $g^{cm}_{ex}(r^*)$ of all-atom systems at the corresponding $\rho^*$

892:    are taken for the reference RDF$_{cm}$s.} \label{Tab.1}

893:   \end{table}

894:  As expected the $f_p$ grows with the growing density of the system.

895: 

896: 

897: From the RDF$_{cm}$ we can evaluate the average number of

898: neighbors of a given molecule within a sphere with the radius

899: $r^*$ as

900: \begin{equation}

901:   n_n(r^*)=\rho^*\int_0^{r^*} g(r^*)4\pi r^{*^2}\,dr^*.

902: \end{equation}

903: The $n_n(r^*)$ for the \emph{ex} and \emph{cg} systems at the

904: temperature $T^*=1$ and number density $0.025\le\rho^*\le 0.175$

905: are shown in figure \ref{Fig.5}. Despite the deviations between

906: the RDF$_{cm}$s of the \emph{ex} and \emph{cg} systems the

907: corresponding average numbers of neighbors exactly match,

908: indicating, together with the RDF$_{cm}$s presented in figure

909: \ref{Fig.4}, that the \emph{cg} model with the molecules

910: interacting via the effective potential given by Eq. (\ref{eq.11})

911: reproduces well the structure of the underlying \emph{ex} system.

912: \begin{figure}[!ht]

913: \centering

914: \includegraphics[height=10cm,width=7.5cm,angle=0]{fig5.ps}

915: \caption{The average number of neighbors $n_n(r^*)$ of a

916: given molecule as a function of distance for explicit (\emph{ex})

917: and coarse-grained (\emph{cg}) systems at the temperature $T^*=1$

918: and number density $0.025\le\rho^*\le 2.25$.}\label{Fig.5}

919: \end{figure}

920: 

921: To show that the \emph{cg} system with the effective potential

922: from Eq. (\ref{eq.11}) is at the same state point as the original

923: \emph{ex} system at the same temperature and density we also

924: evaluated the pressure in the system. The equations of state for

925: the \emph{ex} and \emph{cg} models are shown in figure

926: \ref{Fig.6}. 

927: \begin{figure}[!ht]

928: \centering

929: \mbox{\includegraphics[width=7.5cm, angle=0]{fig6.ps}}

930:          \caption{Pressure $p^*$ in the explicit (\emph{ex}) and

931: coarse-grained (\emph{cg}) systems at the temperature $T^*=1$ as a

932: function of the number density $\rho^*$ of the system.}\label{Fig.6}

933: \end{figure}

934: The resulting equations of state are similar to the

935: case of colloidal sphere systems\cite{Bryant:2002}. This

936: similarity is to be expected since the tetrahedral molecule as

937: well as the spherical coarse-grained molecule can be considered as

938: spherical colloidal particles with a hard core and a soft coating

939: layer. From figure \ref{Fig.6} we can conclude that the \emph{cg}

940: model with the particles interacting via the effective potential

941: from Eq. (\ref{eq.11}) reproduces the equation of state of the

942: underlying \emph{ex} system up to $\rho^*=0.125$, at which the two

943: pressure curves start deviating from each other, reflecting the

944: fact that the effective potential is density dependent. In order

945: to minimize the artifacts originating from our parameterization of

946: the effective potential while still simulating the liquid state we

947: have decided to perform all our MD simulations of the hybrid

948: atomistic/mesoscopic (\emph{ex-cg}) model at the state point with

949: $T^*=1$ and density $\rho^*=0.1$.

950: 

951: 

952: \subsection*{Statistical Properties}

953: 

954: The new adaptive resolution scheme is tested by comparing the

955: computed statistical properties of the \emph{ex-cg} model with the

956: corresponding properties of the reference fully atomistic \emph{ex}

957: system.

958: 

959: 

960: Figure \ref{Fig.7} (a) displays the RDF$_{cm}$s calculated from

961: center-of-mass positions of all molecules in the box of the

962: \emph{ex} and \emph{ex-cg} systems at $\rho^*=0.1$ and $T^*=1$.

963: \begin{figure}[!ht]

964: \centering

965: \subfigure[]{\includegraphics[width=7.5cm]{fig7a.ps}}\\

966: \subfigure[]{\includegraphics[width=7.5cm]{fig7b.ps}}

967:          \caption{(a) Center-of-mass radial distribution functions

968: for all molecules in the box of the all-atom (\emph{ex}) and hybrid

969: atomistic/mesoscopic (\emph{ex-cg}) systems at $\rho^*=0.1$ and

970: $T^*=1$.  Shown are also the corresponding center-of-mass radial

971: distribution functions for only the explicit molecules from the

972: explicit region (\emph{ex-cg}/\emph{ex}) and for only the

973: coarse-grained molecules from the coarse-grained region

974: (\emph{ex-cg}/\emph{cg}). The width of the interface layer is

975: $2d^*=2.5$. (b) The corresponding average numbers of neighbors

976: $n_n(r^*)$ of a given molecule as functions of distance. The

977: different curves are almost indistinguishable.}\label{Fig.7}

978: \end{figure}

979: Shown is the case with the width of the interface layer

980: $2d^*=2.5$. Depicted are also the corresponding local RDF$_{cm}$s

981: for the explicit  (\emph{ex-cg}/\emph{ex}) and coarse-grained

982: regions (\emph{ex-cg}/\emph{cg}) of the \emph{ex-cg} model. As in

983: all simple liquids, e.g., methane, the RDF$_{cm}$s are zero at

984: short distances between molecules' centers of mass because

985: repulsive forces prevent overlapping of molecules. Then the

986: functions increase rapidly to the first peak. With increasing

987: distance RDF$_{cm}$s reach the limiting value $1$ after few

988: oscillations, indicating that there is no order at long distances.

989: The average numbers of neighbors $n_n(r^*)$ of a given molecule as

990: functions of distance are illustrated in figure \ref{Fig.7} (b).

991: The number of nearest neighbors in the first layer corresponding

992: to the first minimum in the RDF$_{cm}$ is about $11$. For comparison, the corresponding

993: experimental value from X-ray diffraction for liquid methane at

994: $T=92\units{K}$ is approximately $12$\cite{Habenschuss:1981}.

995: 

996: All calculated RDF$_{cm}$s of the \emph{ex-cg} model and

997: $n_n(r^*)$  are in good agreement with the reference RDF$_{cm}$

998: and $n_n(r^*)$ of the \emph{ex} model indicating that the

999: structure of the underlying all-atom system is well reproduced

1000: using the adaptive resolution scheme. This is further confirmed by

1001: the values of the penalty function $f_p$ reported in table

1002: \ref{Tab.2}, evaluated for different interface layer widths. 

1003: \begin{table}[h]

1004:    \centering

1005:    \begin{tabular}{||c|c|c|c||}

1006:     \hline\hline

1007:     \vbox to0pt{\hbox{\lower 0.6\baselineskip\hbox{$2d^*$}}\vss}&

1008:     \multicolumn{3}{c||}{$f_p\cdot 10^3$}\\

1009:     \cline{2-4}

1010:            & ex-cg     & ex-cg/ex   & ex-cg/cg  \\

1011:    \hline

1012:     $2.5$  & 0.0821 &  0.0009 & 0.2109 \\

1013:     $3.0$  & 0.0893 &  0.0008 & 0.2158 \\

1014:     $4.0$  & 0.1058 &  0.0016 & 0.2197 \\

1015:     $5.0$  & 0.1324 &  0.0023 & 0.2315 \\

1016:     $6.0$  & 0.1583 &  0.0033 & 0.2367 \\

1017:     $7.0$  & 0.1944 &  0.0047 & 0.2449 \\

1018:     $8.0$  & 0.2301 &  0.0076 & 0.2616 \\

1019:     $9.0$  & 0.2695 &  0.0082 & 0.2489 \\

1020:    $10.0$  & 0.3159 &  0.0101 & 0.2632 \\

1021:     \hline

1022: 

1023:     \hline\hline

1024:    \end{tabular}

1025:    \caption{Penalty function $f_p$ defined by Eq. (\ref{eq.10}) as a

1026:    function of the interface layer width $2d^*$ for RDF$_{cm}$s

1027:    $g^{cm}_{ex-cg}(r^*)$, $g^{cm}_{ex-cg/ex}(r^*)$,

1028:    $g^{cm}_{ex-cg/cg}(r^*)$ of the hybrid atomistic/mesoscopic model

1029:    at $\rho^*=0.1$ and $T^*=1$.  $g^{cm}_{ex-cg}(r^*)$ is the

1030:    RDF$_{cm}$ of all molecules in the box where all molecules are

1031:    considered indistinguishable, $g^{cm}_{ex-cg/ex}(r^*)$ is the RDF$_{cm}$ of only

1032:    the explicit molecules from the explicit region while

1033:    $g^{cm}_{ex-cg/cg}(r^*)$ is the RDF$_{cm}$ of only

1034:    the coarse-grained molecules from the coarse-grained region. The RDF$_{cm}$ $g^{cm}_{ex}(r^*)$

1035:    of all-atom system at the corresponding $\rho^*$ and $T^*$

1036:    is taken for the reference RDF$_{cm}$.}\label{Tab.2}

1037: \end{table}

1038: From

1039: the results given in table \ref{Tab.2} we can see that the local

1040: structure in the explicit region of the \emph{ex-cg} model is

1041: exactly reproduced independently of the interface layer width. In

1042: contrast, the accuracy of the local structure reproduction in the

1043: coarse-grained region of the \emph{ex-cg} model depends on the

1044: accuracy of the effective potential parameterization in the

1045: \emph{cg} model (see the value of $f_p$ given in table \ref{Tab.1}

1046: for $\rho^*=0.1$). The computed $f_p$s for the total RDF$_{cm}$ of

1047: the \emph{ex-cg} model lie up to $2d^*=8.0$ in between the values

1048: for the local explicit and coarse-grained RDF$_{cm}$s.

1049: 

1050: For narrow interface layers with $2d^*<2.5$ the adaptive

1051: resolution scheme fails to work. The molecules are not given

1052: enough space and time to adapt their orientation to their

1053: environment and the system is not properly equilibrated in this

1054: case. The value $2d^*=2.5$, for which the adaptive resolution

1055: scheme gives the acceptable results,  can be rationalized by the

1056: fact that the interface layer width should at least exceed the

1057: maximal range of interaction, which is the range of the effective

1058: potential, namely $2.31\sigma$.

1059: 

1060: To demonstrate that the thermodynamic properties are correctly

1061: reproduced by the adaptive resolution scheme the temperature and

1062: pressure of the system as a functions of the interface layer width

1063: are given in tables \ref{Tab.3} and \ref{Tab.4}, respectively. 

1064: The calculated temperature profile reported in table \ref{Tab.3} shows

1065: that the system is at the right temperature and that all degrees

1066: of freedom are properly equilibrated in accordance with the

1067: equipartition principle regardless of the interface layer width.

1068: The results given in table \ref{Tab.4} also show that the adaptive

1069: resolution scheme succeeds in reproducing the pressure of the

1070: underlying fully atomistic system.

1071: \begin{table*}[ht]

1072:    \centering

1073:   \begin{tabular}{||c|c|c|c|c|c|c||}

1074:     \hline\hline

1075:       $2d^*$  & $T$           & $T_{ex}$       & $T_{cg}$ & $T_{int}$ & $T_{ex}^{all}$& $T_{int}^{all}$        \\

1076:     \hline

1077:       $0^{ex}$& $1.00\pm 0.01$&  $1.00\pm 0.01$&  $-$     & $-$ &$1.00\pm 0.01$   &       $-$                   \\

1078:       $0^{cg}$&  $1.00\pm 0.01$         &   $-$       &   $1.00\pm 0.01$  &    $-$         &    $-$           &  $-$ \\

1079:       $2.5$   &   $1.00\pm 0.01$         &  $1.00\pm 0.01$          &   $1.00\pm 0.02$        &  $1.00 \pm 0.03$            &   $1.00\pm 0.01$            & $1.00\pm 0.02$  \\

1080:       $3.0$   & $1.00\pm 0.01$ & $1.00 \pm 0.02$ &   $1.00 \pm 0.02$ &   $1.00 \pm 0.03$   &  $1.00 \pm 0.01$ &  $1.00 \pm 0.02$   \\

1081:       $4.0$   &  $1.00\pm 0.01$        &  $1.00\pm 0.01$         &  $1.00 \pm 0.02$        &    $1.00 \pm 0.03$           &    $1.00\pm 0.01$             & $1.00\pm 0.01$   \\

1082:       $5.0$   &  $1.00\pm 0.01$        &   $1.00\pm 0.02$        &  $1.00\pm 0.02$        &    $1.00\pm 0.02$           &     $1.00\pm 0.01$          &  $1.00\pm 0.02$ \\

1083:       $6.0$   &  $1.00\pm 0.01$        &  $1.00\pm 0.02$          &  $1.00\pm 0.02$        &    $1.00\pm 0.02$          &        $1.00\pm 0.02$        &   $1.00\pm 0.02$\\

1084:       $7.0$   &    $1.00\pm 0.01$       &    $1.00\pm 0.02$        &   $1.00\pm 0.02$       &     $1.00\pm 0.02$        &    $1.00\pm 0.02$           &  $1.00\pm 0.01$\\

1085:       $8.0$   &   $1.00\pm 0.01$        & $1.00\pm 0.02$          &   $1.00\pm 0.02$       & $1.00\pm 0.02$            &          $1.00\pm 0.02$     &  $1.00\pm 0.01$\\

1086:       $9.0$   &  $1.00\pm 0.01$        &     $1.00\pm 0.03$      & $1.00\pm 0.03$         &     $1.00\pm 0.02$        &        $1.00\pm 0.01$       &   $1.00\pm 0.01$ \\

1087:       $10.0$  &    $1.00\pm 0.01$      &  $1.00\pm 0.03$         &    $1.00\pm 0.02$        &  $1.00\pm 0.02$           &          $1.00\pm 0.01$        & $1.00\pm 0.02$  \\

1088: 

1089:     \hline\hline

1090:    \end{tabular}

1091:    \caption{Average temperature as a function of the interface

1092:  layer width $2d^*$. $T$, $T_{ex}$, $T_{cg}$, and $T_{int}$ are the average temperatures

1093:  of the total system, the explicit, coarse-grained, and interface

1094:  layer regions, respectively, calculated by Eq. (\ref{eq.12}). $T_{ex}^{all}$ and $T_{int}^{all}$

1095:  are the average temperatures of the explicit and interface layer regions,

1096:  respectively, calculated from total velocities

1097:  (translational+vibrational+rotational) of explicit atoms in

1098:  molecules. $0^{ex}$ and $0^{cg}$ denote the all-atom and

1099:  coarse-grained systems, respectively. } \label{Tab.3}

1100:   \end{table*}

1101: \begin{table}[ht]

1102:    \centering

1103:   \begin{tabular}{||c|c||}

1104:     \hline\hline

1105:       $2d^*$        & $p^*$        \\

1106:     \hline

1107:       $0^{ex}$      &  $0.379\pm 0.009$\\

1108:       $0^{cg}$      &  $0.378\pm 0.004$\\

1109:       $2.5$         &  $0.382\pm 0.007$\\

1110:       $3.0$         &  $0.383\pm 0.006$\\

1111:       $4.0$         &  $0.384\pm 0.006$\\

1112:       $5.0$         &  $0.385\pm 0.007$\\

1113:       $6.0$         &  $0.386\pm 0.006$\\

1114:       $7.0$         &  $0.388\pm 0.004$\\

1115:       $8.0$         &  $0.389\pm 0.006$\\

1116:       $9.0$         &  $0.390\pm 0.005$\\

1117:      $10.0$         &  $0.391\pm 0.006$\\

1118:     \hline\hline

1119:    \end{tabular}

1120:    \caption{Average pressure calculated using Eq. (\ref{eq.12a}) as a function of the interface

1121:  layer width $2d^*$. $0^{ex}$ and $0^{cg}$ denote the all-atom and

1122:  coarse-grained systems, respectively. } \label{Tab.4}

1123:   \end{table}

1124: 

1125: In order to check that the chemical potentials in the atomistic

1126: and coarse-grained regions are equal as required by the condition

1127: (\ref{eq.1}) we report in table \ref{Tab.5} the average number of

1128: molecules in different regions of the system.  In table

1129: \ref{Tab.5} we also give the number of degrees of freedom

1130: $n_{DOF}$ in the system defined as

1131: \begin{equation}

1132:   n_{DOF}=3\sum_\alpha [w_\alpha N+(1-w_\alpha)],\label{eq.13}

1133: \end{equation}

1134: where $N=4$ is the number of explicit tetrahedral atoms in a

1135: molecule and $w_\alpha$ is the value of the weighting function

1136: defined by Eq. (\ref{eq.7}) for the molecule $\alpha$. The

1137: summation in Eq. (\ref{eq.13}) goes over all $n$ molecules of the

1138: system. 

1139: \begin{table}[ht]

1140:    \centering

1141:   \begin{tabular}{||c|c|c|c|c||}

1142:     \hline\hline

1143:       $2d^*$  & $n_{ex}$ & $n_{cg}$ & $n_{int}$ & $n_{DOF}$       \\

1144:     \hline

1145:       $0^{ex}$ &  $5001$       &  $0$          &  $0$         &  $60012$\\

1146:       $0^{cg}$ &    $0$        &  $5001$       &  $0$         &  $15003$\\

1147:        $2.5$   &  $2167\pm 38$ &  $2170\pm 48$ & $663\pm 21$  &  $37446\pm 516$\\

1148:        $3.0$   &  $2100\pm 45$ &  $2103\pm 60$ & $796\pm 37$  &  $37455\pm 493$\\

1149:        $4.0$   &  $1966\pm 58$ &  $1969\pm 47$ & $1064\pm 40$ &  $37467\pm 507$              \\

1150:        $5.0$   &  $1832\pm 47$ &  $1835\pm 56$ & $1332\pm 43$ &  $37470\pm 501$             \\

1151:        $6.0$   &  $1697\pm 57$ &  $1701\pm 54$ & $1601\pm 34$ &  $37467\pm 330$             \\

1152:        $7.0$   &  $1563\pm 45$ &  $1566\pm 29$ & $1871\pm 31$ &  $37479\pm 526$              \\

1153:        $8.0$   &  $1428\pm 48$ &  $1431\pm 27$ & $2141\pm 22$ &  $37479\pm 345$              \\

1154:        $9.0$   &  $1294\pm 22$ &  $1295\pm 23$ & $2411\pm 60$ &  $37488\pm 576$           \\

1155:       $10.0$   &  $1158\pm 33$ &  $1160\pm 16$ & $2682\pm 56$ &  $37482\pm 63$              \\

1156:     \hline\hline

1157:    \end{tabular}

1158:    \caption{Average number of molecules as a function of the interface

1159:    layer width $2d^*$. $n_{ex}$, $n_{cg}$, and $n_{int}$ are the

1160:    average number of molecules in the explicit, coarse-grained, and

1161:    interface layer regions, respectively. $n_{DOF}$ is the average

1162:    number of degrees of freedom defined by Eq. (\ref{eq.13}).

1163:    For orientation: in the system with $2500$ coarse-grained molecules,

1164:    $2500$ four atom explicit molecules, and no hybrid molecules

1165:    $n_{DOF}=37500$. $0^{ex}$ and $0^{cg}$ denote the all-atom and

1166:  coarse-grained systems, respectively.} \label{Tab.5}

1167:   \end{table}

1168: The results show that there is no molecule number bias in

1169: the system and that the \emph{ex-cg} system is homogeneous. Note

1170: that $n_{DOF}$ is greatly reduced employing the adaptive

1171: resolution scheme on the \emph{ex-cg} model in comparison with the

1172: fully atomistic model. The time evolution of the number of

1173: molecules in different regions of the system together with the

1174: $n_{DOF}$ is illustrated in figure \ref{Fig.8}.

1175: \begin{figure}[!ht]

1176: \centering

1177: \mbox{\includegraphics[width=7.5cm, angle=0]{fig8.ps}}

1178:          \caption{ Time evolution of number of molecules in a

1179: explicit ($n_{ex}$) , coarse-grained ($n_{cg}$), and interface

1180: regions ($n_{int}$) in the hybrid atomistic/mesoscopic model with

1181: the $2.5\sigma$ interface layer width. Time evolution of number of

1182: degrees of freedom in the system ($n_{DOF}$) is depicted in the

1183: inset.}\label{Fig.8}

1184: \end{figure}

1185:  The results

1186: clearly demonstrate that the system is in thermodynamical

1187: equilibrium, indicating that the conditions (\ref{eq.1}) are

1188: satisfied by our adaptive resolution scheme.

1189: 

1190: 

1191: Although we have parameterized the effective potential to

1192: reproduce the structural properties of the \emph{ex} system we can

1193: also compare the dynamical properties of the \emph{ex-cg} with the

1194: \emph{ex} model. For that purpose we have computed the

1195: self-diffusion coefficient, which is evaluated from the

1196: center-of-mass displacements using the Einstein relation

1197:  \begin{equation}

1198:   D^*=\frac{1}{6}\lim_{t^*\rightarrow\infty}\frac{\langle|{\bf

1199:   R}^*_\alpha(t^*)-{\bf R}^*_\alpha(0)|^2\rangle}{t^*},\label{eq.14}

1200:  \end{equation}

1201: where ${\bf R}^*_\alpha(t^*)$ is the center-of-mass position of

1202: the molecule $\alpha$ at time $t^*$ and averaging is performed

1203: over all molecules and all choices of time origin. The

1204: self-diffusion coefficient of the \emph{ex} and \emph{cg} models

1205: calculated in the microcanonical ensemble, where the Langevin

1206: thermostat is switched off after the initial warm-up run, are

1207: $0.24$ and $0.30$, respectively. The corresponding values of the

1208: self-diffusion coefficient for the \emph{ex}, \emph{cg}, and

1209: \emph{ex-cg} models with the Langevin thermostat switched on are

1210: $0.12$, $0.14$, and $0.13$, respectively. Since

1211: we use the same time and length scales in all our three models

1212: different values of the self-diffusion coefficient in the

1213: \emph{ex} and \emph{cg} models (with no Langevin thermostat

1214: applied) indicate that the coarse-grained molecules experience a

1215: slightly smaller intermolecular frictional hindrance in their

1216: motion compared to the explicit molecules. This indicates that the

1217: effective pair potential given in Eq. (\ref{eq.11}) introduces an

1218: effective time scale shift in the coarse-grained regime. This is

1219: known from other studies \cite{Tschop:1998,Tschop:1998:2}, where

1220: one actually takes advantage of that in order to reach very long

1221: simulation times\cite{leon:2005}. The apparent self-diffusion

1222: coefficient values of the \emph{ex} and \emph{cg} models are lower

1223: and much closer together when the Langevin thermostat is applied

1224: due to the frictional forces arising from the coupling to the

1225: thermostat unlike to the case of polymers. There typically the

1226: friction of the thermostat is negligible compared to the friction

1227: between monomers. Since the self-diffusion coefficient of the

1228: \emph{ex-cg} model is close to the corresponding values for the

1229: \emph{ex} and \emph{cg} models we can conclude that the

1230: center-of-mass dynamics of the molecules is similar in all three

1231: models.

1232: 

1233: 

1234: As the final test to demonstrate the reliability of the adaptive

1235: resolution scheme we have computed the number density profile of

1236: the \emph{ex-cg} model. 

1237: \begin{figure}[!ht]

1238: \centering

1239: \subfigure[]{\includegraphics[width=7.5cm]{fig9a.ps}}\\

1240: \subfigure[]{\includegraphics[width=7.5cm]{fig9b.ps}}

1241: 

1242:          \caption{(a) Normalized density profile in the $x$

1243: direction of the hybrid atomistic/meso\-scopic model with the

1244: $2.5\sigma$ interface layer width. Vertical lines denote

1245: boundaries between atomistic, coarse-grained and interface regions

1246: of the system. (b) The same as in (a) but for the  $10.0\sigma$

1247: interface layer width.}\label{Fig.9}

1248: \end{figure}

1249: The results for the system with $2d^*=2.5$ and

1250: $2d^*=10.0$ are presented in figures \ref{Fig.9} (a) and (b),

1251: respectively. 

1252: The results in figure \ref{Fig.9} (a) for the case

1253: of $2d^*=2.5$ show that the explicit and coarse-grained regions

1254: have the same homogeneous density as the reference system while

1255: the density in the transition regime undergoes an oscillation

1256: around the reference value $\rho^*_0=0.1$ with a magnitude of

1257: approximately $0.05\rho^*_0$. In the case of $2d^*=10.0$ (figure

1258: \ref{Fig.9} (b)) a $5\%$ drop in the density occurs in the

1259: transition regime, which is compensated by the slight increase of

1260: the density in the explicit and coarse-grained regions.

1261: 

1262: This artifact can be explained by considering the results

1263: displayed in figure \ref{Fig.10} (a), where the pressure of the

1264: system containing only hybrid molecules as a function of the

1265: constant value of the weighting function $w$ (corresponding to the

1266: situation in the interface layer) is illustrated.

1267: \begin{figure}[!ht]

1268: \centering

1269: \subfigure[]{\includegraphics[width=7.5cm]{fig10a.ps}}\\

1270: \subfigure[]{\includegraphics[width=7.5cm]{fig10b.ps}}

1271:          \caption{Artifacts of the adaptive resolution scheme. (a)

1272: Average pressure $p^*$ in the system containing only hybrid

1273: molecules as a function of the constant value of the weighting

1274: function $w$. (b) Center-of-mass radial distribution functions for

1275: the explicit system (ex) and the system containing only hybrid

1276: molecules with $w=const=1/2$ (\emph{ex-cg}(w=const=1/2)) at

1277: $\rho^*=0.0025$ and $T^*=1$. The inset also shows the

1278: corresponding PMF$_{ex}$ and PMF$_{ex-cg(w=const=1/2)}$ determined

1279: from the systems with $\rho^*=0.0025$  using Eq. (\ref{eq.0}).

1280: }\label{Fig.10}

1281: \end{figure}

1282:  The pressure is

1283: increased in comparison to the pressure in the reference system.

1284: Clearly the increase is most prominent for the most 'artificial'

1285: case with $w=1/2$, indicating that there still is a small

1286: 'pressure barrier' in the interface region causing the density

1287: dip. This is also evident from results in figure  \ref{Fig.10}

1288: (b), where the RDF$_{cm}$ and the potential of mean force of the

1289: system with $w=1/2$ are shown. The effective potential in a system

1290: containing only hybrid molecules with constant value of the

1291: weighting function  changes in comparison to the all-atom system.

1292: This means that the hybrid molecules in the interface layers of

1293: the \emph{ex-cg} model experience too strong effective interaction

1294: leading to the pressure variations in the interface layer. Since the

1295: artifact occurs at constant values of $w$ it is an artifact of the

1296: linear combination of forces in Eq. (\ref{eq.4}) and not of the

1297: functional form of the weighting function $w$. It must be

1298: emphasized, however, that this artifact of the proposed adaptive

1299: resolution scheme is within a $5\%$ error and that similar

1300: artifacts, occurring at the boundary of two domains with different

1301: level of detail, are also characteristic for other hybrid

1302: schemes\cite{Cai:2000}.

1303: 

1304: The pressure variations in the interface layer could cause a

1305: spurious reflection of molecules from the boundary. However, the

1306: results presented in figure \ref{Fig.11}, where the time evolution

1307: of two diffusion profiles is monitored for molecules that are

1308: initially localized at the two slabs with $a^*/10$ width

1309: neighboring the interface layer, show that this is not the case.

1310: \begin{figure}[!ht]

1311: \centering

1312: \subfigure[]{\includegraphics[width=7.5cm]{fig11a.ps}}\\

1313: \subfigure[]{\includegraphics[width=7.5cm]{fig11b.ps}}

1314:          \caption{Time evolution of diffusion profiles for the

1315: molecules that are initially, at time $t^*=0$, localized at two

1316: neighboring slabs of the mid interface layer with $2d^*=2.5$ ($n$

1317: is the number of this molecules with the center-of-mass position

1318: at a given coordinate $x^*$). The width of the two slabs is

1319: $a^*/10$. Vertical lines denote boundaries of the interface layer.

1320: (a) The diffusion profile, averaged over $500$ different time

1321: origins, at $t^*=0$, $t^*=10$, and $t^*=50$ for the molecules that

1322: are initially localized at the slab on the coarse-grained side of

1323: the interface region. (b) The same as in (a) but for the molecules

1324: that are initially localized at the slab on the atomistic side of

1325: the interface region.

1326: }\label{Fig.11}

1327: \end{figure}

1328: The molecules initially localized at the two slabs spread out

1329: symmetrically with time. This is because the self-diffusion

1330: coefficients of all models are approximately the same in the case

1331: of the applied Langevin thermostat. Thus, the two distributions

1332: occupy at time $t^*$ regions with mean square radius

1333: \begin{equation}

1334:  \langle |x^*(t^*)-x^*(0)|^2\rangle\simeq 2D^*t^*,

1335: \end{equation}

1336: where $x^*(0)$ is the center of the distribution at time $t^*=0$,

1337: which is equal to $-d^*-a^*/20$  for the left slab (figure

1338: \ref{Fig.11} (a)) and $d^*+a^*/20$ for the right slab (figure

1339: \ref{Fig.11} (b)). Since the diffusion profiles are symmetrical at

1340: any given time we can conclude that the artifact at the interface

1341: layer is too small to have any severe effect on the diffusion of

1342: molecules across the interface layer.

1343: 

1344: \section{Conclusions}

1345: A novel approach for efficient hybrid atomistic/mesoscale

1346: molecular dynamics (MD) simulations has been presented in this

1347: paper. The new adaptive resolution MD simulation scheme

1348: dynamically couples the atomic and mesoscale length scales of the

1349: studied system by allowing an on-the-fly dynamical interchange

1350: between molecules' atomic and mesoscopic levels of description. In

1351: our approach the number of degrees of freedom is allowed to

1352: fluctuate during the course of simulation while the statistical

1353: properties of the underlying all-atom model are properly

1354: reproduced for both levels of detail. Since the purpose of this

1355: paper was to develop the method, we restricted ourselves to a most

1356: simple model system of a molecular liquid with short-range

1357: repulsive interactions. Even on this level a number of questions

1358: still have to be tackled, as there is the application to higher

1359: liquid densities for complex molecular liquids or the proper

1360: analysis and understanding of the the time scale problem, when the

1361: diffusion on the coarse-grained level is faster than on the

1362: all-atom level. For separate runs at the two levels of detail this

1363: is understood, while the occurrence within one simulation box

1364: still poses some conceptual problems. On the other hand the

1365: current density is in between a typical small molecule liquid and

1366: a polymeric fluid. Thus we expect that already this approach can

1367: be generalized and applied to different realistic soft condensed

1368: matter systems where both atomic and mesoscopic length scales have

1369: to be considered. This can be either polymer solutions and melts,

1370: such as a synthetic or biological macromolecule embedded in a

1371: solvent. Similarly our method should also find application for

1372: other polymer systems (same force field), molecular liquids such

1373: as methane (same geometry) or water (tetrahedral clusters), etc.,

1374: enabling to reach much larger length and time scales than for

1375: all-atom MD simulations. In all cases the aim is to treat in a

1376: simulation only as many degrees of freedom as absolutely necessary

1377: for the question considered. In this sense the region of higher

1378: detail can be either given by a geometrical constraint, e.g., close

1379: to a surface, or even be chosen on demand due to specific local

1380: conformations of a (macro-)molecular system. Work along these

1381: lines is underway.

1382: \vspace{0.5cm}

1383: 

1384: \section*{\small ACKNOWLEDGMENTS}

1385: We thank A. Arnold, B.~A. Mann, P. Schravendijk, B. Hess, and N.

1386: van der Vegt for useful discussions. We are also grateful to C.~F.

1387: Abrams for discussions at early stage of this work. This work is

1388: supported in part by the Volkswagen foundation. M.~P. acknowledges

1389: the support of the Ministry of Higher Education, Science and

1390: Technology of Slovenia under grant No. P1-0002.

1391: 

1392: 

1393: 

1394: 

1395: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1396: \begin{thebibliography}{10}

1397: 

1398: \bibitem{Deuflhard:1999}

1399: P.~Deuflhard, J.~Hermans, B.~Leimkuhler, A.~Mark, S.~Reich and R.~Skeel, editors,

1400: \newblock {\em Computational Molecular Dynamics: Challenges, Methods, Ideas},

1401:   volume~4 of {\em Lecture Notes in Computational Science and Engineering},

1402: \newblock Springer Verlag, 1999.

1403: 

1404: \bibitem{Minary:2004}

1405: P.~Minary, M.~E. Tuckerman, and G.~J. Martyna,

1406: \newblock Phys. Rev. Lett. {\bf 93}, 150201 (2004).

1407: 

1408: \bibitem{Praprotnik:2005}

1409: D.~Jane\v{z}i\v{c}, M.~Praprotnik, and F.~Merzel,

1410: \newblock J. Chem. Phys. {\bf 122}, 174101 (2005).

1411: 

1412: \bibitem{Praprotnik:2005:1}

1413: M.~Praprotnik and D.~Jane\v{z}i\v{c},

1414: \newblock J. Chem. Phys. {\bf 122}, 174102 (2005).

1415: 

1416: \bibitem{Praprotnik:2005:2}

1417: M.~Praprotnik and D.~Jane\v{z}i\v{c},

1418: \newblock J. Chem. Phys. {\bf 122}, 174103 (2005).

1419: 

1420: \bibitem{Praprotnik:2004}

1421: M.~Praprotnik, D.~Jane\v{z}i\v{c}, and J.~Mavri,

1422: \newblock J. Phys. Chem. A {\bf 108}, 11056 (2004).

1423: 

1424: \bibitem{kremer:2001}

1425: K.~Kremer, and F.~M\"uller-Plathe,

1426: \newblock MRS Bulletin \textbf{26}, 205 (2001).

1427: 

1428: \bibitem{kremer:2004}

1429: K.~Kremer in "Multiscale  Modelling and Simulation, Lecture Notes

1430: on Computational Science and Engineering, S. Attinger and P.

1431: Koumoutsakis eds, Springer 2004.

1432: 

1433: \bibitem{Ayton:2004}

1434: G.~S. Ayton, H.~L. Tepper, D.~T. Mirijanian, and G.~A. Voth,

1435: \newblock J. Chem. Phys. {\bf 120}, 4074 (2004).

1436: 

1437: \bibitem{Kranenburg:2003}

1438: M.~Kranenburg, M.~Venturoli, and B.~Smit,

1439: \newblock J. Phys. Chem. B {\bf 107}, 11491 (2003).

1440: 

1441: \bibitem{Nielsen:2004}

1442: S.~O. Nielsen, C.~F. Lopez, G.~Srinivas, and M.~L. Klein,

1443: \newblock J. Phys.: Condens. Matter {\bf 16}, R481 (2004).

1444: 

1445: \bibitem{Marrink:2004}

1446: S.~J. Marrink, A.~H. de~Vries, and A.~E. Mark,

1447: \newblock J. Phys. Chem. B {\bf 108}, 750 (2004).

1448: 

1449: \bibitem{Chang:2005}

1450: R.~Chang, G.~S. Ayton, and G.~A. Voth,

1451: \newblock J. Chem. Phys. {\bf 122}, 244716 (2005).

1452: 

1453: \bibitem{Cook:2005}

1454: I.~R. Cooke, K.~Kremer, and M.~Deserno,

1455: \newblock Phys. Rev. E {\bf 72}, 011506 (2005). 

1456: 

1457: \bibitem{Kremer:1990}

1458: K.~Kremer and G.~S. Grest,

1459: \newblock J. Chem. Phys. {\bf 92}, 5057 (1990).

1460: 

1461: \bibitem{Tschop:1998}

1462: W.~Tsch{\" o}p, K.~Kremer, J.~Batoulis, T.~B{\" u}rger, and O.~Hahn,

1463: \newblock Acta Polym. {\bf 49}, 61 (1998).

1464: 

1465: \bibitem{Tschop:1998:2}

1466: W.~Tsch{\" o}p, K.~Kremer, O.~Hahn, J.~Batoulis, and T.~B{\" u}rger,

1467: \newblock Acta Polym. {\bf 49}, 75 (1998).

1468: 

1469: \bibitem{Everaers:2004}

1470: R.~Everaers, S.~K.~Sukumaran, G.~S.~Grest, C.~Svaneborg, A.~Sivasubramanian, and K.~Kremer,

1471: \newblock Science {\bf 303}, 823 (2004).

1472: 

1473: \bibitem{Chun:2000}

1474: H.~M. Chun, C.~E. Padilla, D.~N. Chin, M. Watanabe, V.~I. Karlov,

1475: H.~E. Alper, K. Soosaar, K.~B. Blair, O.~M. Becker, L.~S.~D. Caves,

1476: R. Nagle, D.~N. Haneym, and B.~L. Farmer,

1477: \newblock J. Comput. Chem. {\bf 21}, 159 (2000).

1478: 

1479: \bibitem{Malevanets:2000}

1480: A.~Malevanets and R.~Kapral,

1481: \newblock J. Chem. Phys. {\bf 112}, 7260 (2000).

1482: 

1483: \bibitem{Villa:2005}

1484: E.~Villa, A.~Balaeff, and K.~Schulten,

1485: \newblock PNAS {\bf 102}, 6783 (2005).

1486: 

1487: \bibitem{DelleSite:2002}

1488: L.~Delle Site, C.~F. Abrams, A.~Alavi, and K.~Kremer,

1489: \newblock Phys. Rev. Lett. {\bf 89}, 156103 (2002).

1490: 

1491: \bibitem{Abrams:2003}

1492: C.~F. Abrams, L.~Delle Site, and K.~Kremer,

1493: \newblock Phys. Rev. E {\bf 67}, 021807 (2003).

1494: 

1495: \bibitem{DelleSite:2004}

1496: L.~Delle Site, S.~Leon, and K.~Kremer,

1497: \newblock J. Am. Chem. Soc. {\bf 126}, 2944 (2004).

1498: \bibitem{Laio:2002}

1499: A.~Laio, J.~VandeVondele and U.~Roethlisberger,

1500: \newblock J.Chem.Phys. {\bf 116}, 6941 (2002).

1501: \bibitem{Rafii:1998}

1502: H.~Rafii-Tabar, L.~Hua, and M.~Cross,

1503: \newblock J. Phys.: Condens. Matter {\bf 10}, 2375 (1998).

1504: 

1505: \bibitem{Broughton:1999}

1506: J.~Q. Broughton, F.~F. Abraham, N.~B. Bernstein, and E.~Kaxiras,

1507: \newblock Phys. Rev. B {\bf 60}, 2391 (1999).

1508: 

1509: \bibitem{Smirnova:1999}

1510: J.~A. Smirnova, L.~V. Zhigilei, and B.~J. Garrison,

1511: \newblock Comp. Phys. Commun. {\bf 118}, 11 (1999).

1512: 

1513: \bibitem{Csanyi:2004}

1514: G.~Csanyi, T.~Albaret, M.~C. Payne, and A.~D. Vita,

1515: \newblock Phys. Rev. Lett. {\bf 93}, 175503 (2004).

1516: 

1517: \bibitem{Abrams:2004}

1518: C.~F. Abrams,

1519: \newblock Inhomogenous coarse-graining of polymers and polymer/metal

1520:   interfaces,

1521: \newblock in {\em Computational Soft Matter: {F}rom Synthetic Polymers to

1522:   Proteins}, edited by N.~Attig, K.~Binder, H.~Grubm{\" u}ller, and K.~Kremer,

1523:   volume~23 of {\em {NIC} Series}, pages 275--288, John von {N}eumann

1524:   {I}nstitute for {C}omputing, 2004.

1525: 

1526: \bibitem{Louis:2002}

1527: A.~A. Louis,

1528: \newblock J. Phys.: Condens. Matter {\bf 14}, 9187 (2002).

1529: 

1530: \bibitem{Curtarolo:2002}

1531: S.~Curtarolo and G.~Ceder,

1532: \newblock Phys. Rev. Lett. {\bf 88}, 255504 (2002).

1533: 

1534: \bibitem{Klapp:2004}

1535: S.~H.~L. Klapp, D.~J. Diestler, and M.~Schoen,

1536: \newblock J. Phys.: Condens. Matter {\bf 16}, 7331 (2004).

1537: 

1538: \bibitem{Reith:2003}

1539: D.~Reith, M.~P{\" u}tz, and F.~M{\" u}ller-Plathe,

1540: \newblock J. Comput. Chem. {\bf 24}, 1624 (2003).

1541: 

1542: \bibitem{Bryant:2002}

1543: G.~Bryant, S.~R. Williams, L. Qian, I.~K. Snook, E. Perez, and F. Pincet,

1544: \newblock Phys. Rev. E {\bf 66}, 060501(R) (2002).

1545: 

1546: \bibitem{Lawrence:2003}

1547: C.~P. Lawrence and J.~L. Skinner,

1548: \newblock Chem. Phys. Lett. {\bf 372}, 842 (2003).

1549: 

1550: \bibitem{Zwanzig:1954}

1551: R.~W. Zwanzig,

1552: \newblock J. Chem. Phys. {\bf 22}, 1420 (1954).

1553: 

1554: \bibitem{Leach:2001}

1555: A.~R. Leach,

1556: \newblock {\em Molecular Modelling},

1557: \newblock Pearson Education Limited, Harlow, 2 edition, 2001.

1558: 

1559: \bibitem{Schwabl:1995}

1560: F.~Schwabl,

1561: \newblock {\em Quantum Mechanics},

1562: \newblock Springer-Verlag, Berlin Heidelberg, 2 edition, 1996.

1563: 

1564: 

1565: \bibitem{Soddemann:2003}

1566: T.~Soddemann, B.~D{\" u}nweg, and K. Kremer,

1567: \newblock Phys. Rev. E {\bf 68}, 046702 (2003).

1568: 

1569: \bibitem{Allen:1987}

1570: M.~P. Allen and D.~J. Tildesley,

1571: \newblock {\em Computer Simulation of Liquids},

1572: \newblock Clarendon Press, Oxford, 1987.

1573: 

1574: \bibitem{Berendsen:1984}

1575: H.~Berendsen, J.~Postma, W.~V. Gunsteren, A.~D. Nola, and J.~Haak,

1576: \newblock J. Chem. Phys. {\bf 81}, 3684 (1984).

1577: 

1578: \bibitem{Mann:2004}

1579: B.~A. Mann, R.~Everaers, C.~Holm, and K.~Kremer,

1580: \newblock Europhys. Lett. {\bf 67}, 786 (2004).

1581: 

1582: \bibitem{Duenweg:1991}

1583: B.~D{\" u}nweg and W.~Paul,

1584: \newblock Int. J. Mod. Phys. C {\bf 2}, 817 (1991).

1585: 

1586: 

1587: \bibitem{Kremer:2005}

1588: The physically correct way of course would be

1589: to calculate the second virial coefficient. This here is not that

1590: helpful, because the interaction potential is rather smooth. The

1591: present measure is more appropriate, when the structure of the

1592: liquids is compared in an NVT simulation, as in our case.

1593: 

1594: \bibitem{Espresso:2005}

1595: http://www.espresso.mpg.de.

1596: 

1597: \bibitem{Habenschuss:1981}

1598: A.~Habenschuss, E.~Johnson, and A.~H. Narten,

1599: \newblock J. Chem. Phys. {\bf 74}, 5234 (1981).

1600: 

1601: 

1602: \bibitem{leon:2005}

1603: S.~Leon, N.~van der Vegt, L.~Delle Site, and K.~Kremer,

1604: \newblock Macromolecules in press (2005).

1605: 

1606: \bibitem{Cai:2000}

1607: W.~Cai, M.~de~Koning, V.~V. Bulatov, and S.~Yip,

1608: \newblock Phys. Rev. Lett. {\bf 85}, 3213 (2000).

1609: 

1610: 

1611: 

1612: \end{thebibliography}

1613: 

1614: 

1615: \end{document}

1616: