1: \documentclass[superscriptaddress, floatfix, unsortedaddress, aps, prb,
2: amssymb, amsmath, twocolumn]{revtex4} %showpacs
3:
4: \usepackage{amsmath}
5: \usepackage{graphicx}
6:
7: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
8:
9: \begin{document}
10:
11: \bibliographystyle{apsrev}
12:
13: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
14:
15: % user defined commands and short-hands
16: \newcommand{\ord}[1]{\mathcal{O}(#1)}
17: \newcommand{\rl}[1]{#1_{12}}
18: \newcommand{\eps}{\epsilon}
19: \newcommand{\rhat}{\hat{\mathbf{r}}_{12}}
20: \newcommand{\nhat}{\hat{\mathbf{n}}}
21: \newcommand{\rhohat}{\hat{\mathbf{\rho}}}
22: \newcommand{\dr}{\delta_{R}}
23: \newcommand{\dt}{\delta_{T}}
24: \newcommand{\resq}{RE${}^2_{\;\;\;}$}
25: %\newcommand{\resq}{RE-squared~}
26: \newcommand{\toluene}{C$_7$H$_8$}
27: \newcommand{\benzene}{C$_6$H$_6$}
28:
29: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
30:
31: \title{Coarse-grained Interaction Potentials for Anisotropic Molecules}
32: \author{M. Babadi}
33: \address{Sharif University of
34: Technology, Department of Physics, P.O. Box 11365-9161, Tehran,
35: Iran.}
36: \author{R. Everaers}
37: \address{Max-Planck-Institut f\"ur Physik komplexer
38: Systeme, N\"othnitzer Str. 38, 01187 Dresden, Germany}
39: \author{M.R. Ejtehadi}
40: \email{ejtehadi@sharif.edu}
41: \address{Sharif University of
42: Technology, Department of Physics, P.O. Box 11365-9161, Tehran,
43: Iran.}
44: %\address{Institute for studies in Theoretical Physics and Mathematics,
45: %P.O. Box 19395-5531, Tehran, Iran.}
46:
47: \date{\today}
48:
49: \begin{abstract}
50: We have proposed an efficient parameterization method for a recent
51: variant of the Gay-Berne potential for dissimilar and biaxial
52: particles and demonstrated it for a set of small organic molecules.
53: Compared to the previously proposed coarse-grained models, the new
54: potential exhibits a superior performance in close contact and large
55: distant interactions. The repercussions of thermal vibrations and
56: elasticity has been studied through a statistical method. The study
57: justifies that the potential of mean force is representable with the
58: same functional form, extending the application of this
59: coarse-grained description to a broader range of molecules.
60: Moreover, the advantage of employing coarse-grained models over
61: truncated atomistic summations with large distance cutoffs has been
62: briefly studied.
63: \end{abstract}
64:
65: \maketitle
66:
67: \section{Introduction}
68: The development of accurate, reliable and computationally efficient
69: interaction models is the main activity of molecular modeling. The
70: need to attain larger simulated time scales and the excessive
71: complexity of a wide range of molecular systems (e.g. biomolecular)
72: has emphasized the factor of computation efficiency as a dominant
73: deliberation in choosing the appropriate interaction model for
74: molecular simulations. In particular, grouping certain atoms into
75: less detailed interaction sites, known as "coarse-graining", in one
76: way of achieving such efficiency.
77:
78: Various coarse-grained (CG) approaches have been recently developed
79: with such goal in mind~\cite{GB, BFZ98, Shelley, Izvekov}. The
80: implementation of coarse-graining models is usually divided into two
81: distinct stages. The first is a partitioning of the system into the
82: larger structural units while the second stage is the construction
83: of an effective force field to describe the interactions between the
84: CG units. Typically, CG potentials of a pre-defined analytical form
85: are parameterized to produce average structural properties seen in
86: atomistic simulations. Such analytical forms are chosen in a way to
87: describe the governing interaction between the CG
88: units~\cite{Shelley}. The parameterizations are usually based on
89: matching samples of potentials of mean force~\cite{Shelley, Meyer},
90: inverse Monte Carlo data~\cite{Murtola} or certain atomistic
91: potentials characteristics~\cite{BFZ98}. The main concern of the
92: present work is parameterizing a CG force field for the short-range
93: attractive and repulsive interactions between ellipsoidal molecules
94: and groups, based on atomistic potential sampling and potential of
95: mean force.
96:
97: In molecular simulations, short-range attractive and repulsive
98: interactions are typically represented using Lennard-Jones(6-12)
99: potentials~\cite{AllenTildesley, FrenkelSmit}:
100: \begin{equation}\label{eq:ULJ}
101: U_{LJ}(r; i,j) = 4\epsilon_{ij} \left[ \left(\frac{\sigma_{ij}}
102: r\right)^{12} -
103: \left(\frac{\sigma_{ij}} r\right)^{6}
104: \strut\right]
105: \end{equation}
106: where $\sigma_{ij}$ and $\epsilon_{ij}$ are the effective
107: heterogeneous interaction radius and well-depth between particles of
108: type $i$ and~$j$ respectively and $r$ is the inter-particle
109: displacement. While the $r^{-6}$ part has a physical origin in
110: dispersion or van der Waals interactions, the $r^{-12}$ repulsion is
111: chosen for mathematical convenience and is sometimes replaced by
112: exponential terms as well. For large molecules, the exact evaluation of the
113: interaction potential of this type involves a computationally
114: expensive double summation over the respective (atomic) interaction
115: sites:
116: \begin{equation}\label{eq:UMacroSum}
117: U_{int}(\mathcal{M}_1, \mathcal{M}_2) =
118: \sum_{i\in\mathcal{M}_1}\sum_{j\in\mathcal{M}_2} U_a(r_{ij}; i, j)
119: \end{equation}
120: where $\mathcal{M}_1$ and $\mathcal{M}_2$ denote the interacting
121: molecules and $U_a(\cdot)$ is the atomic interaction potential, e.g.
122: Eq.~(\ref{eq:ULJ}). In practice, a large distant interaction cutoff
123: accompanied by a proper tapering is used to reduce the computation
124: cost. More sophisticated and efficient summation methods such as
125: Ewald summation and the Method of Lights are also widely
126: used~\cite{Leach}.
127:
128: As an alternative approach, Gay and Berne~\cite{GB} proposed a more
129: complicated single-site CG interaction potential (in contrast to
130: sophisticated summation techniques) for uniaxial rigid molecules
131: which was generalized to dissimilar and biaxial particles later by
132: Berardi {\it et al} as well~\cite{BFZ98}. We will refer to this
133: potential as the biaxial-GB in the rest of this article.
134:
135: In response to the criticism of the unclear microscopic
136: interpretation of the GB potential~\cite{Perram96}, we have recently
137: used results from colloid science~\cite{Hunter} to derive an
138: approximate interaction potential based on the Hamaker
139: theory~\cite{Hamaker} for mixtures of ellipsoids of arbitrary size
140: and shape, namely the \resq potential~\cite{EE03}. Having a
141: parameter space identical to that of Berardi, Fava and
142: Zannoni~\cite{BFZ98}, the \resq potential agrees significantly
143: better with the numerically evaluated continuum approximation of
144: Eq.~(\ref{eq:UMacroSum}), has no unphysical large distant limit and
145: avoids the introduction of empirical adjustable parameters.
146:
147: In an anisotropic coarse-grained model, a molecule $\mathcal{M}$ is
148: treated like a rigid body. Neglecting the atomic details, each
149: molecule is characterized by a center separation $\mathbf{r}$ and a
150: transformation operator (a unitary matrix $\mathbf{A}$ or a unit
151: quaternion $\mathbf{q}$) describing its orientation.
152:
153: In the first section of the article, we briefly introduce the \resq
154: potential followed by a review of the biaxial-GB potential and
155: Buckingham(exp-6) atomistic model. The Buckingham(exp-6) potential
156: is used in the MM3 force field~\cite{MM3} and will serve as the
157: atomistic model potential for parameterizations. The second section
158: describes a parameterization method which has been demonstrated for
159: a few selected molecules, followed by an exemplar comparison between
160: the \resq and the biaxial-GB potential. We will study the
161: repercussion of internal vibrations, in contrast to the usually
162: assumed proposition of the ideal stiffness~\cite{GB, BFZ98}, and
163: propose an error analysis method to define trust temperature regions
164: for single site potentials. Finally, we will show that the potential
165: of mean force (PMF) is representable with the same functional form
166: for a wide range of temperatures.
167:
168: \section{Atomistic and Single-site Non-bonded Potentials}
169: In the MM3 force field~\cite{MM3}, the van der Waals interaction is
170: described in terms of Buckingham(exp-6) potential which is an exponential
171: repulsive accompanied by a $r^{-6}$ attractive term:
172: \begin{equation}\label{eq:mm3a}
173: U^{MM3}(r_{ij}; i, j) = \epsilon_{ij}\left(A
174: e^{-B\sigma_{ij}/r_{ij}}-C\left(\frac{\sigma_{ij}}{r_{ij}}\right)^6\right)
175: \end{equation}
176: where A, B and C are fixed empirical constants while $\sigma_{ij}$
177: and $\epsilon_{ij}$ are heterogeneous interaction parameters specific
178: to the interacting particles. Usually, Lorenz and Berthelot
179: averaging rules are used to define heterogeneous interaction
180: parameters in terms of the homogeneous ones, i.e. $\sigma_{ij} =
181: (\sigma_{i} + \sigma_{j})/2$ and $\epsilon_{ij} =
182: \sqrt{\epsilon_{i}\epsilon_{j}}$. The hard core repulsion is usually
183: described via a $r^{-12}$ term with an appropriate energy switching:
184: \begin{equation}\label{eq:mm3r}
185: U_{HC}^{MM3}(r_{ij}; i, j) = \gamma
186: \left(\frac{\sigma_{ij}}{r_{ij}}\right)^{12}
187: \end{equation}
188: where $\gamma$ is defined in a way to provide continuity at the
189: switching distance. The interaction energy between two arbitrary molecules
190: is trivially the pairwise double summation over all of the interaction
191: sites, i.e. Eq.~(\ref{eq:UMacroSum}).
192:
193: The dissimilar and biaxial Gay-Berne potential (biaxial-GB) is a
194: widely used single-site model proposed by Berardi et
195: al.~\cite{BFZ98} which is an extension of the original uniaxial
196: description~\cite{GB} to biaxial molecules and heterogeneous
197: interactions. Based on the original Gay and Berne concept, the
198: biaxial-GB is a shifted Lennard-Jones(6-12) interaction between two
199: biaxial Gaussian distribution of interacting sites. In this
200: coarse-grained model, each molecule is described by two diagonal
201: characteristic tensors (in the principal basis of the molecule)
202: $\mathbf{S}$ and~$\mathbf{E}$, representing the half radii of the
203: molecule and the strength of the pole contact interactions,
204: respectively. As mentioned earlier, the orientation of a molecule is
205: described by a center separation vector $\mathbf{r}$ and a unitary
206: operator~$\mathbf{A}$, revolving
207: the lab frame to the principal frame of the molecule.\\
208:
209: The biaxial-GB description for the interaction between two molecules
210: with a center separation of $\rl{\mathbf{r}}= \mathbf{r}_2 -
211: \mathbf{r}_1$ and respective orientation tensors $\mathbf{A}_1$ and
212: $\mathbf{A}_2$ is defined as:
213: \begin{multline}\label{eq:GBdef}
214: U^{GB}_{A,R}(\rl{\mathbf{r}},\mathbf{A}_1,\mathbf{A}_2)=\\
215: 4\epsilon_0\rl{\eta}^{\nu}\rl{\chi}^{\mu}
216: \Bigg[\bigg(\frac{\sigma_c}{\rl{h}+\sigma_c}\bigg)^{12}-
217: \bigg(\frac{\sigma_c}{\rl{h} + \sigma_c}\bigg)^{6}\Bigg]
218: \end{multline}\\
219: where $\epsilon_0$ and $\sigma_c$ are the energy and length scales,
220: $\rl{\eta}$ and $\rl{\chi}$ are purely orientation dependant terms~\cite{BFZ98}
221: and $\rl{h}$ is the the least contact distance between the two
222: ellipsoids which are defined by the diagonal covariance tensor of the assumed
223: Gaussian distributions. The orientation dependant terms ($\rl{\eta}$
224: and $\rl{\chi}$) describe the anisotropy of the molecules.\\
225:
226: We have recently proposed a single-site potential, namely
227: \resq~\cite{EE03} giving the approximate interaction energy between
228: two hard ellipsoids in contrast to the tradition of the Gaussian
229: clouds, initiated by Gay and Berne~\cite{GB}. The orientation
230: dependence of the \resq potential fall at large distances, reducing
231: asymptotically to the interaction energy of two spheres. Moreover,
232: it gives a more realistic intermediate and close contact interaction
233: using a heuristic interpolation of the Deryaguin
234: expansion~\cite{Deryaguin, EE03}. Being a shifted
235: Lennard-Jones(6-12) potential, the biaxial-GB fails to exhibit the
236: correct functional behavior for large molecules~\cite{EE03}. The
237: attractive and repulsive contributions of the \resq potential are
238: respectively:
239: \begin{subequations}\label{eq:re2}
240: \begin{multline}\label{eq:re2A}
241: U_A^{RE^2}(\mathbf{A}_1, \mathbf{A}_2, \rl{\mathbf{r}}) =-\frac{A_{12}}{36}\Big(1+
242: 3\rl{\eta}\rl{\chi}\frac{\sigma_c}{\rl{h}}\Big)\times\\
243: \prod_{i=1}^2\prod_{e=x,y,z}
244: \Bigg(\frac{\sigma^{(i)}_e}{\sigma^{(i)}_e+h_{12}/2}\Bigg)
245: \end{multline}
246: \begin{multline}\label{eq:re2R}
247: U_R^{RE^2}(\mathbf{A}_1, \mathbf{A}_2, \rl{\mathbf{r}})=\frac{A_{12}}{2025}
248: \Big(\frac{\sigma_c}{\rl{h}}\Big)^6\Big(1+
249: \frac{45}{56}\rl{\eta}\rl{\chi}\frac{\sigma_c}{\rl{h}}\Big)\times\\
250: \prod_{i=1}^2 \prod_{e=x,y,z}
251: \Bigg(\frac{\sigma^{(i)}_e}{\sigma^{(i)}_e+h_{12}/60^{\frac{1}{3}}}\Bigg)
252: \end{multline}
253: \end{subequations}\\
254: where $A_{12}$ is the Hamaker constant (the energy scale),
255: $\sigma_c$ is the atomic interaction radius and $\sigma_x^{(i)}$,
256: $\sigma_y^{(i)}$ and $\sigma_z^{(i)}$ are the half-radii of $i$th
257: ellipsoid (i=1,2). The terms $\rl{\eta}$, $\rl{\chi}$ and $\rl{h}$ are
258: defined in parallel to the biaxial-GB model and thus, are described
259: in terms of the same characteristic tensors.\\
260:
261: The structure tensor $\mathbf{S}_i$ and the relative potential well depth
262: tensor $\mathbf{E}_i$ are diagonal in the principal basis of $i$th molecule
263: and are defined as:
264: \begin{subequations}
265: \begin{equation}
266: \mathbf{S}_i = \textrm{diag}\{\sigma_x^{(i)}, \sigma_y^{(i)},
267: \sigma_z^{(i)}\}
268: \end{equation}
269: \begin{equation}
270: \mathbf{E}_i = \textrm{diag}\left\{E_x^{(i)}, E_y^{(i)},
271: E_z^{(i)}\right\}
272: \end{equation}
273: \end{subequations}
274: where $E_x^{(i)}$, $E_y^{(i)}$ and $E_z^{(i)}$ are dimensionless energy scales
275: inversely proportional to the potential well depths of the respective orthogonal
276: configurations of the interacting molecules ({\bf aa}, {\bf bb} and {\bf cc},
277: Table~\ref{tab:ortho}). For large molecules with uniform constructions, it has
278: been shown~\cite{EE03} that the energy parameteres are approximately representable
279: in terms of the local contact curvatures using the Deryaguin expansion~\cite{Deryaguin}:
280: \begin{equation}\label{eq:derya}
281: \mathbf{E}_i = \sigma_c \textrm{diag}\left\{\frac{\sigma_x}{\sigma_y
282: \sigma_z}, \frac{\sigma_y}{\sigma_x \sigma_z},
283: \frac{\sigma_z}{\sigma_x \sigma_y}\right\}
284: \end{equation}
285: The assumptions leading to these estimations are not valid for the
286: studied small organic molecules. Therefore, we will cease to impose
287: further suppositions and take these three scales as independent characteristics
288: of a biaxial molecule. Computable expressions for the orientation dependent factors
289: of the \resq potential ($\rl{\eta}$ and $\rl{\chi}$) among with the Gay-Berne
290: approximation for $\rl{h}$ has been given in the Appendix~(\ref{sec:orientation}).
291:
292: \section{Parameterization for Arbitrary Molecules}
293: \subsection{The Principal Basis and The Effective Center of Interaction}
294: Associating a biaxial ellipsoid to an arbitrary molecule, one must
295: define an appropriate principal basis and a center of interaction for it
296: beforehand, according to the used coarse-grained model. Although there's
297: no trivial solution to this problem, the centroid and the eigenbasis of the
298: geometrical inertia tensor of the molecule are promising candidates and may
299: be taken as suitable initial guesses as they yield to the correct solution at
300: least for the molecules with perfect symmetry. For a molecule consisting of
301: $N$ particles, the centroid is defined as:
302: \begin{equation}
303: \mathbf{r}_c = \frac{\sum_{i=1}^{N}\mathbf{r}_i}{N}
304: \end{equation}
305: and the principal basis is the eigenbasis of the geometrical inertia
306: tensor $\mathbf{I}_g$ given by:
307: \begin{equation}
308: \mathbf{I}_g =
309: \sum_{i=1}^{N}(r_i^2\mathbf{1}-\mathbf{r_i}\otimes\mathbf{r}_i)
310: \end{equation}
311: where $\mathbf{r}_i$ is the position of $i$th atom.\\
312:
313: The most general parameter space of the \resq potential contains the energy and
314: length scales, the characteristic tensors and the parameters specifying the relative
315: orientation of the ellipsoids to the molecules. In order to overcome the degeneracy
316: of the parameter space and to guarantee the rapid convergence of the optimization
317: routines, a two-stage parameterization is proposed. In the first stage, the center and
318: principal frame of the molecule will be fixed at the centroid and the eigenbasis of the
319: inertia tensor. A preliminary optimization in the reduced parameter space yields
320: to an approximate parameterization. In the second stage, the results of the first stage
321: will be taken as the initial guess, followed by an optimization in the unconstrained
322: variable space. This two-stage parameterization will theoretically result in
323: superior results for molecules with imperfect symmetries.
324:
325: \subsection{Sampling and Optimization}
326: Physical and symmetrical considerations lead to the proposition that
327: a sampling of the pole contact interactions between two biaxial particles is
328: essentially sufficient to reproduce the interaction for all configurations.
329: There are 18 different orthogonal approaching configurations (pole contacts)
330: between two dissimilar and biaxial particles (Table~\ref{tab:ortho}).
331: Based on physical grounds, we optimize the parameter space for the important
332: characteristics of the sampled orthogonal energy profiles, i.e. potential well
333: depth, potential well distance, the width of well at half depth and the soft contact
334: distance. This parameterization fashion is guaranteed to produce a satisfactory
335: reconstruction of the most crucial region of interaction.\\
336:
337: The geometry of the molecules were initially optimized using TINKER
338: molecular modeling package~\cite{TINKER} with the MM3 force field. We
339: have used the same force field to sample the interaction energy for
340: the orthogonal configurations. \\
341:
342: Given a parameter tuple $\mathbf{p}$, we denote the potential well depth,
343: well distance, well width at half depth and the soft contact distance for $i$th
344: orthogonal configuration predicted by the \resq potential by $U_m(i;\mathbf{p})$,
345: $R_m(i;\mathbf{p})$, $W(i;\mathbf{p})$ and $R_{sc}(i;\mathbf{p})$ respectively.
346: The same potential well specifications calculated from the atomistic sum is
347: denoted by scripted letters. An appropriate cost function is:
348: \begin{multline}
349: \Omega(\mathbf{p})=\frac{1}{\Omega_0}\sum_{i=1}^{N_{\dagger}}
350: e^{-\beta\mathcal{U}_m(i)}\Bigg[
351: w_{U_m}\bigg(\frac{U_m(i;\mathbf{p})-\mathcal{U}_m(i)}{\mathcal{U}_0}\bigg)^2+\\
352: w_{R_m}\bigg(\frac{R_m(i;\mathbf{p})-\mathcal{R}_m(i)}{\mathcal{R}_0}\bigg)^2
353: +w_W\bigg(\frac{W(i;\mathbf{p})-\mathcal{W}(i)}{\mathcal{W}_0}\bigg)^2+\\
354: w_{R_{sc}}\bigg(\frac{R_{sc}(i;\mathbf{p})-\mathcal{R}_{sc}(i)}{\mathcal{R}_{sc0}}\bigg)^2\Bigg]
355: \end{multline}
356: where $\Omega_0$ is a normalization factor:
357: \begin{equation}
358: \Omega_0 = 4(w_{U_m}^2 + w_{R_m}^2 + w_{W}^2 +
359: w_{R_{sc}}^2)^{\frac{1}{2}}\sum_{i=1}^{N_{\dagger}}e^{-\beta
360: \mathcal{U}_m(i)}.
361: \end{equation}
362: We have chosen $\mathcal{U}_0$ as
363: $\min\left\{|\mathcal{U}_m(i)|\right\}$ and $\mathcal{R}_0$,
364: $\mathcal{W}_0$ and $\mathcal{R}_{sc0}$ as
365: $\min\left\{\mathcal{W}_m(i)\right\}$ based on physical
366: considerations. $N_{\dagger}$ is the number of orthogonal profiles
367: (12 and 18 for homogeneous and heterogeneous interactions
368: respectively) and $(w_U, w_R, w_W, w_{R_{sc}})$ are fixed error
369: partitioning factors for different terms, set to $(1.0, 3.0, 2.5,
370: 1.0)$ in order to emphasize on the structural details. We have also
371: included a fixed error weighting according to the Boltzmann
372: probability of the appearance of the corresponding profiles. One
373: expects higher amplitude of relative appearance for orientations
374: with deeper wells, which justifies the requisite of higher
375: contribution in the cost function. We have also chosen $\beta$ as
376: $1/\left\langle|\mathcal{U}_m(i)|\right\rangle$ in
377: order to avoid deep submergence of the lower energy orientations.\\
378:
379: Further implications such as matching the large distance behavior will be
380: regarded as constraints on the parameter space, leaving the defined cost
381: function unchanged.\\
382:
383: The nonlinear optimization procedure consists of a preliminary Nelder-Mead Simplex
384: search followed by a quasi-Newton search with BFGS Hessian updates~\cite{NumOpt}.
385: The whole parameterization routine is coded in MATLAB/Octave and is freely
386: available~\cite{RE2CLib}. The procedures of sampling and parameterization are purely
387: automated and requires only a Cartesian input file. The interaction parameters for the
388: homogeneous interaction of a set of small prolate and oblate organic molecules has been
389: provided in Table~(\ref{tab:param}). It is noticed that the provided half radii agree
390: significantly better with the molecular dimensions compared to the biaxial-GB
391: parameterizations~\cite{BFZ98}, reflecting the precise microscopic interpretation of
392: the \resq potential.
393:
394: \subsection{Large Distance Analysis}
395: % The \resq potential asymptotically approaches the Lennard-Jones(6-12) large distant
396: %interaction between two spheres.
397: The cost function defined in the previous section focuses on close contact regions only.
398: In order to achieve the correct large distant limit as well, we will constrain the
399: variable space by matching the asymptotic behavior
400: of the \resq potential with the atomistic summation. The asymptotic behavior of the
401: \resq potential is described as:
402: \begin{equation}\label{eq:re2limita}
403: \lim_{\rl{r}\rightarrow\infty}\rl{r}^6U_{RE^2}(\rl{\mathbf{r}},\mathbf{A}_1,
404: \mathbf{A}_2)=-\frac{16}{9}\rl{A}\det[\mathbf{S}_1]\det[\mathbf{S}_2]
405: \end{equation}\\
406: The atomistic summation defined by Eq.~(\ref{eq:UMacroSum}) and
407: Eq.~(\ref{eq:mm3a}) exhibits the same asymptotic behavior, which
408: together with Eq.~(\ref{eq:re2limita}) results in the relation:
409: \begin{equation}\label{eq:re2limit}
410: \rl{A}\det[\mathbf{S}_1]\det[\mathbf{S}_2]=\frac{9}{16}\sum_{i\in
411: \mathrm{A}}\sum_{j\in \mathrm{B}}\epsilon_{ij}\sigma_{ij}^6
412: \end{equation}\\
413: The summation appearing in right hand side is most easily evaluated
414: by a direct force field parameter lookup. Applying such a constraint
415: guarantees the expected large distant behavior while leads to a
416: faster parameterization, reducing the dimensions of the variable
417: space. Unconstrained optimization routines are still applicable as
418: one may solve Eq.~(\ref{eq:re2limit}) for $A_{12}$ explicitly. A graphical
419: comparison between the biaxial-GB and the \resq potential has been given for
420: the homogeneous interaction of the pair Perylene~\cite{Mols} has been sketched
421: in Fig.~\ref{fig:loglog}. The large distance convergence of the \resq potential
422: is noticed in contrast to the divergent behavior of the biaxial-GB potential,
423: which is due to the non-vanishing orientation dependent pre-factors.
424: Although the energy contribution is small at this limit, it is not generally
425: negligible, e.g. the large distant separability of the orientation dependence of
426: the model potential alters the nature of the phase diagram and the long range
427: order of a hard rod fluid in general~\cite{Cuesta}.
428:
429: \subsection{Heterogeneous Interactions}
430: The heterogeneous interaction between two molecules $\mathcal{M}_1$
431: and $\mathcal{M}_2$ is calculable by equations \eqref{eq:re2A} and \eqref{eq:re2R}
432: once the characteristic tensors of each molecule ($\mathbf{S}$ and $\mathbf{E}$)
433: along with the heterogeneous Hamaker constant $A_{\mathcal{M}_1 \mathcal{M}_2}$ and
434: the atomic potential radius $\sigma_{\mathcal{M}_1 \mathcal{M}_2}$ are available.
435: The heterogeneous Hamaker constant may be evaluated directly with a force-field
436: parameter lookup. Moreover, the arithmetic mean of $\sigma_{\mathcal{M}_1
437: \mathcal{M}_1}$ and $\sigma_{\mathcal{M}_2 \mathcal{M}_2}$ is a reasonable
438: estimate for the heterogeneous interaction radius, $\sigma_{\mathcal{M}_1\mathcal{M}_2}$.
439: Therefore, the homogeneous interaction parameters of the molecules $\mathcal{M}_1$
440: and $\mathcal{M}_2$ are sufficient to describe their respective heterogeneous interaction
441: using the \resq potential. Apparently, there is no trivial mixing rule available for
442: the energy scale of the biaxial-GB potential. Inspired by the atomic mixing rules
443: and the theory of the Gay-Berne potential, we have used Berthelot's geometric
444: averaging rule for this purpose. An instance of a heterogeneous interaction has been
445: illustrated in Fig.~(\ref{fig:orthoplot}) for the pair Perylene (oblate) and
446: Sexithiophene (prolate)~\cite{Mols}. The results are quite promising for a coarse-grained
447: model; However, further optimization will theoretically yield to superior results.
448: Concluding from the graphs, the \resq potential performs significantly better at
449: end-to-end and cross interactions compared to the biaxial-GB. The error measures
450: ($\Omega_{RE^2}=6.5\times 10^{-3}, \Omega_{GB}=7.7\times 10^{-3}$) agree with this
451: observation.
452:
453: \subsection{The advantages over practical atomistic implementations}
454: As mentioned before, the atomistic evaluation of long-range
455: interaction potentials involve computationally expensive double
456: summations over the interaction sites, resulting in a quadratic time
457: cost with respect to the average number of interactions sites.
458: However, the average computation time of a single-site potential is
459: intrinsically constant, regardless of the number of interacting
460: atoms. These observations have been quantified in
461: Fig.~(\ref{fig:time}) which is a comparison between the computation
462: time of an exact LJ(6-12) atomistic summation and an efficient
463: implementation of the \resq potential~\cite{RE2CLib}. Concluding
464: from the graph, employing the \resq potential for molecules
465: consisting as low as~$\sim 5$ atoms (or $\sim 25$ overall atomic
466: interactions) is economic.
467:
468: The atomistic summations are practically employed with a proper
469: large distance atomic cutoff in order to reduce the computation
470: time. In the presence of large distance cutoffs, long range
471: correction potential terms~\cite{Leach} are usually used to
472: compensate the submergence of particles beyond the cutoff distance.
473:
474: Considerable errors may be introduced by choosing small atomic
475: cutoff distances compared to the dimensions of the interacting
476: molecules. Therefore, it is expectable that a CG model yield to
477: relatively better results compared to truncated atomistic summations
478: in certain configurations. A figurative situation is the end-to-end
479: interaction of two long prolate molecules. In such configurations,
480: usual atomic cutoffs ($\simeq 2.5\sigma$) can be small enough to
481: dismiss the interaction between the far ends of the molecules.
482: Moreover, long range correction terms are of little application in
483: this case due to the excessive inhomogeneity and the small number of
484: interacting particles.
485:
486: This effect has been illustrated in Fig.~(\ref{fig:err_cut}) for
487: Pentacene molecule~\cite{Mols}. The first panel is a semi-log plot
488: of the relative error for the \resq potential together with three
489: atomistic approximations with different cutoffs (6, 9 and 12 \AA).
490: The discontinuity of the truncated atomistic summations is a result
491: of hard cutoffs. In a proper atomic implementation, tapering
492: functions are used to avoid such discontinuities. Concluding from
493: the graphs, the CG description introduces less error in all ranges
494: of this configuration compared to the truncated atomistic
495: summations, even with unusually large atomic cutoff distance ($12
496: \AA$). Moreover, the evaluation of the CG interaction potential
497: requires a considerably lower computation time.
498:
499: \section{Intermolecular Vibrations and Single-site Potentials}
500: In this section, we study the proposition of ideal rigidity of the
501: molecules, which is widely assumed in single-site approximations of
502: extended molecules, including our own study in the previous
503: sections. The samplings are usually taken from the relative
504: orientations of the unperturbed and geometrically optimized
505: structures. The resulting parameterization will be used in molecular
506: dynamics simulations in which internal vibrations may not be
507: negligible. We will introduce a method to estimate the error
508: introduced by this supposition in the first part of this study. A
509: parameterization based on the Potential of Mean Force (PMF) is
510: probably the best one can achieve with the coarse-grained models,
511: although the samplings are expensive. We will study such
512: parameterizations in the second part.
513:
514: \subsection{Analysis of the Mean Relative Error}\label{subsec:err}
515: The PMF for the interaction of semi-rigid molecules in an arbitrary
516: ensemble may be expressed as an additive correction term to the the
517: interaction potential of the respective rigid molecules. We will
518: show that these correction terms are expressible in terms of
519: statistical geometric properties of a molecule in the ensemble. The
520: PMF between two molecules $\mathcal{M}_1$ and $\mathcal{M}_2$ with a
521: mean center separation of $\rl{\mathbf{r}}= \mathbf{r}_2 -
522: \mathbf{r}_1$ and mean orientation tensors $\mathbf{A}_1$ and
523: $\mathbf{A}_2$ is defined as:
524: \begin{equation}\label{eq:pmf}
525: U_{pmf}(\rl{\mathbf{r}}, \mathbf{A}_1, \mathbf{A}_2)=\left<U\left(\mathcal{M}_1,
526: \mathcal{M}_2\right)\right>
527: \end{equation}
528: where $\langle\cdot\rangle$ denotes the ensemble averaging. The mean
529: location of intermolecular particles are expected to remain
530: unchanged compared to the unperturbed structures for a large range
531: of temperatures as the internal structures of semi-rigid molecules
532: are mainly governed by harmonic bond stretching and angle bending
533: potentials. \\
534:
535: We may assume the location of each particle as a random variable,
536: sharply peaked at its mean value. Therefore, we denote the location of
537: $i$th particle measured in its principal coordinate system by:
538: \begin{equation}
539: \mathbf{r}_i = \bar{\mathbf{r}}_i + \delta\mathbf{r}_i
540: \end{equation}
541: where $\bar{\mathbf{r}}_i = \left<\mathbf{r}_i\right>$ and
542: $\delta\mathbf{r}_i$ is a displacement due to internal vibrations
543: with vanishing average. The PMF of the interaction between the
544: molecules $\mathcal{M}_1$ and $\mathcal{M}_2$ is defined as:
545: \begin{equation}\label{eq:pmfLJ}
546: U_{pmf}(\rl{\mathbf{r}}, \mathbf{A}_1, \mathbf{A}_2)=
547: \left<\sum_{i\in\mathcal{M}_1}\sum_{j\in\mathcal{M}_2}U_a
548: \left(\|\mathbf{r}_i-\mathbf{r}_j\|;i,j\right)\right>
549: \end{equation}
550: where $U_a(\cdot)$ is the atomistic interaction potential. It is easy to
551: show that up to the second moments:
552: \begin{multline}\label{eq:aveR}
553: \left<\|\mathbf{r}_i-\mathbf{r}_j\|\right> \simeq
554: \|\mathbf{r}^0_{ij}\| + \\
555: \frac{1}{2\|\mathbf{r}^0_{ij}\|}\sum_{k=3}^3
556: \mathrm{Var}(\delta\mathbf{r}_i-\delta\mathbf{r}_j).\hat{\mathbf{e}}_k
557: \bigg(1-\frac{\big(\mathbf{r}^0_{ij}.\mathbf{e}_k\big)^2}{\|\mathbf{r}^0_{ij}\|^2}\bigg)
558: \end{multline}
559: where $\mathbf{r}^0_{ij} = \bar{\mathbf{r}}_i-\bar{\mathbf{r}}_j$.
560: We have neglected the covariance between the coordinates.
561: Using the last relation, we reach to a second-order estimate of Eq.~(\ref{eq:pmfLJ}):
562: \begin{multline}\label{eq:pmfLJexpanded}
563: U_{pmf}(\rl{\mathbf{r}}, \mathbf{A}_1, \mathbf{A}_2;
564: \mathcal{M}_1,\mathcal{M}_2) \simeq \\
565: \sum_{i\in\mathcal{M}_1}\sum_{j\in\mathcal{M}_2}U_a\left(\|\bar{\mathbf{r}}_i-
566: \bar{\mathbf{r}}_j\|;i,j\right)+\\
567: \frac{1}{2}\sum_{i\in\mathcal{M}_1}\sum_{j\in\mathcal{M}_2}\Bigg(\sum_{k=1}^3
568: \mathrm{Var}(\delta\mathbf{r}_i-\delta\mathbf{r}_j).\hat{\mathbf{e}}_k\times\\
569: \bigg(1-\frac{\big(\mathbf{r}^0_{ij}.\mathbf{e}_k\big)^2}{\|\mathbf{r}^0_{ij}\|^2}\bigg)
570: \frac{U_a'(\|\mathbf{r}^0_{ij}\|;i,j)}{\|\mathbf{r}^0_{ij}\|}+ \\
571: \sum_{k=1}^3\mathrm{Var}(\delta\mathbf{r}_i-\delta\mathbf{r}_j).\hat{\mathbf{e}}_k
572: \frac{\big(\mathbf{r}^0_{ij}.\mathbf{e}_k\big)^2}{\|\mathbf{r}^0_{ij}\|^2}
573: U_a''(\|\mathbf{r}^0_{ij}\|;i,j)\Bigg)
574: \end{multline}
575: The first term of the right hand side is the interaction energy of
576: the averaged structures, where the remaining terms are second-order
577: corrections. The RMS of the relative error introduced by neglecting
578: the correction terms (e.g. the error in parameterizations based on
579: unperturbed samplings) may be evaluated formally via the following integral:
580: \begin{equation}\label{eq:RMSint}
581: \mathcal{E}(T)=\left(\frac{1}{\mathcal{E}_0(T)}\int_{\omega\in\Omega(T)}
582: \exp{\left(-\frac{U(\omega)}{k_B T}\right)}\left[\frac{\delta
583: U(\omega)}{U(\omega)}\right]^2 dN_{\omega}\right)^{\frac{1}{2}}
584: \end{equation}
585: where $\omega$ is a relative orientation, $dN_{\omega}$ is
586: a differential measure of orientations near $\omega$, $\Omega(T)$
587: being the ensemble and $\delta U(\omega)$ is the
588: second-order correction defined by Eq.~(\ref{eq:pmfLJexpanded}).
589: $\mathcal{E}_0(T)$ is the normalization factor defined as:
590: \begin{equation}
591: \mathcal{E}_0(T) = \int_{\omega\in\Omega(T)}
592: \exp{\left(-\frac{U(\omega)}{k_B T}\right)}dN_{\omega}
593: \end{equation}
594:
595: In practice, the spatial variance of each particle in an ensemble is
596: most easily obtainable through an MD simulation. Once the statistical
597: information are accessible, $\mathcal{E}(T)$ is most easily evaluated by
598: Monte Carlo integration. Neglecting the covariance between the dislocation
599: of the particles, we are implicitly overlooking the stretching and bending
600: of the molecules at close contact configurations. Although our proposed error
601: analysis disregards this phenomenon, it still measures the introduced error
602: due to purely thermal vibrations.\\
603:
604: We have demonstrated this error analysis method for three different
605: molecules in a large range of temperatures. The statistical
606: information was extracted from several MD simulation snapshots with
607: the aid of TINKER molecular modeling package~\cite{TINKER}, each
608: with 32 molecules and with periodic boundary conditions in an NVT
609: ensemble (Fig.~\ref{fig:err}). For each isothermal ensemble, the RMS error
610: has been evaluated using the MC integration of Eq.~(\ref{eq:RMSint}) for $10^5$
611: random orientations. The relation between the RMS error and the temperature is
612: noticeably linear. The linear regression analysis has been given at Table~
613: (\ref{tab:err}). According to the required degree of precision, one
614: can define a trust region for the temperature using diagrams like
615: Fig.~(\ref{fig:err}). For example, a mean relative error of
616: less than 10\% is expected for temperatures less than 1500K in a homogeneous
617: ensemble of Benzene molecules, concluding from the graph. It is also concluded
618: that the studied prolate molecule (Sexithiophene) exhibits a higher relative error
619: due to its considerably higher elasticity, compared to the oblate molecules
620: (Perylene and Benzene).
621:
622: \subsection{Parameterizations based on the Potential of Mean Force}
623: The error introduced by the assumption of ideal rigidity may not be
624: negligible for certain purposes, concluding from the previous
625: analysis. However, a coarse-grained potential which is parameterized
626: on a PMF basis is theoretically advantageous as it is expected to
627: describe the mean behaviors closer to the atomistic model.
628: Phenomenologically speaking, the internal degrees of freedom will
629: soften the repulsions at close contacts while the thermal vibrations
630: are expected to smoothen the orientation and separation
631: dependencies of the interaction.\\
632:
633: The PMF for a given macroscopic orientation may be evaluated through
634: a Constrained Molecular Dynamics simulation (CMD) process with
635: appropriate restrains. We have used harmonic restraining potentials
636: for the center separation vector and on the deviations from the
637: desirable principal basis for each molecule in order to keep them at
638: the desired orientation. In order to reduce the random noise of the
639: evaluated PMF, we applied a fifth-order Savitsky-Golay smoothing
640: filter followed by a piecewise cubic Hermite interpolating
641: polynomial fitting to the PMF samples.\\
642:
643: Fig.~(\ref{fig:pmf}) is a plot of the evaluated PMF between the pair
644: perylene for the cross configuration {\bf bc}. The upper (and
645: interior) plots refer to higher temperatures. The expansion of the
646: potential well width at lower temperatures is related to the
647: tendency of the molecules to bend and stretch and thus,
648: resulting in a softer interaction while the shift of the soft contact and
649: potential well distance along with the elevation of the potential
650: well is associated to the thermal vibrations and hence, the
651: expansion of the effective volume of the molecules. The
652: temperature-dependant parameterizations (based on the evaluated PMF)
653: given at Table~(\ref{tab:param_pery}) justifies these qualitative
654: discussions. One may associate the contraction of the molecule
655: ($\sigma_x$, $\sigma_y$ and $\sigma_z$) and the expansion of the
656: atomic interaction radius at lower temperatures to contraction of
657: the molecule at rough repulsions and widening of the potential well,
658: respectively. Furthermore, the expansion of the molecule volume at
659: higher temperatures reflect the overcoming of thermal vibrations to
660: the flexibility of the molecule. According to
661: tables~(\ref{tab:param}) and (\ref{tab:param_pery}), the overall
662: error measures ($\Omega$) for PMF parameterizations closely match
663: the same measure for the unperturbed structure. Thus, the same
664: functional form may be used to represent the PMF as well.
665:
666: \section{Conclusion}
667: We have proposed and demonstrated a parameterization method for
668: the \resq anisotropic single-site interaction potential which
669: leads to a globally valid description of the attractive and repulsive
670: interaction between arbitrary molecules. Unlike the biaxial-GB~\cite{BFZ98},
671: the \resq potential gives the correct large distant interaction~(Fig.~\ref{fig:loglog})
672: while having a superior performance in the close contact region~(Fig.~\ref{fig:orthoplot}).
673: The Potential of Mean Force is representable with the same functional form
674: of the \resq potential. Compared to the parameterizations
675: given at~\cite{BFZ98}, The structure tensors agree significantly better
676: with the spatial distribution of the intermolecular particles.
677: It has also been shown that the coarse-gained models perform significantly
678: better at certain configurations in comparison to a truncated atomistic
679: summation with large distance cutoffs.
680:
681: \section{Acknowledgment}
682: M. R. Ejtehadi would like to thank Institute for studies in Theoretical
683: Physics and Mathematics for partial supports.
684:
685: \appendix
686:
687: \section{The orientation dependent terms}
688: \label{sec:orientation}
689: We will briefly quote computable expressions for the orientation
690: dependant terms from the original article~\cite{EE03}. The term $\rl{\chi}$
691: quantifies the strength of interaction with respect to the local atomic
692: interaction strength of the molecules and is defined as:
693: \begin{equation}\label{eq:chidef}
694: \rl{\chi}(\mathbf{A}_1, \mathbf{A}_2, \rhat) =
695: 2\rhat^T\rl{\mathbf{B}}^{-1}(\mathbf{A}_1, \mathbf{A}_2)\rhat
696: \end{equation}
697: where $\rl{\mathbf{B}}$ is defined in terms of the orientation tensors
698: $\mathbf{A}_i$ and relative well-depth tensors $\mathbf{E}_i$:
699: \begin{equation}\label{eq:Bdef}
700: \rl{\mathbf{B}}(\mathbf{A}_1, \mathbf{A}_1) = \mathbf{A}_1^T
701: \mathbf{E}_1 \mathbf{A}_1 + \mathbf{A}_2^T \mathbf{E}_2
702: \mathbf{A}_2.
703: \end{equation}
704: The term $\rl{\eta}$ describes the effect of contact curvatures of the
705: molecules in the strength of the interaction and is defined as:
706: \begin{equation}
707: \rl{\eta}(\mathbf{A}_1, \mathbf{A}_2, \rhat) =
708: \frac{\det[\mathbf{S}_1]/\sigma_1^2 +
709: \det[\mathbf{S}_2]/\sigma_2^2}{\big[\det[\rl{\mathbf{H}}]/(\sigma_1+\sigma_2)\big]^{1/2}},
710: \end{equation} \\
711: The projected radius of \textit{i}th ellipsoid along $\rhat$ ($\sigma_i$)
712: and the tensor $\rl{\mathbf{H}}$ are defined respectively as:
713: \begin{equation}\label{eq:sigdef}
714: \sigma_i(\mathbf{A}_i,
715: \rhat)=(\rhat^T\mathbf{A}_i^T\mathbf{S}_i^{-2}\mathbf{A}_i\rhat)^{-1/2}
716: \end{equation}
717: \begin{equation}\label{eq:Hdef}
718: \rl{\mathbf{H}}(\mathbf{A}_1,\mathbf{A}_2,\rhat) =
719: \frac{1}{\sigma_1} \mathbf{A}_1^T \mathbf{S}_1^2 \mathbf{A}_1 +
720: \frac{1}{\sigma_2} \mathbf{A}_2^T \mathbf{S}_2^2 \mathbf{A}_2 .
721: \end{equation}
722: There's no general solution to the least contact distance between
723: two arbitrary ellipsoids ($\rl{h}$). The Gay-Berne approximation~\cite{GB,
724: Perram96} is usually employed due to its low complexity and promising
725: performance:
726: \begin{equation}\label{eq:hgb}
727: \rl{h}^{GB} = \|\rl{\mathbf{r}}\| - \rl{\sigma}
728: \end{equation}
729: The anisotropic distance function $\rl{\sigma}$~\cite{BFZ98} is defined as:
730: \begin{equation}\label{eq:sig12def}
731: \rl{\sigma}=\left(\frac{1}{2}\rhat^T
732: \rl{\mathbf{G}}^{-1} \rhat\right)^{-\frac{1}{2}}
733: \end{equation}
734: where the symmetric overlap tensor $\rl{\mathbf{G}}$ is:
735: \begin{equation}\label{eq:Gdef}
736: \rl{\mathbf{G}} = \mathbf{A}_1^T\mathbf{S}_1^2\mathbf{A}_1 +
737: \mathbf{A}_2^T\mathbf{S}_2^2\mathbf{A}_2
738: \end{equation}
739:
740: \bibliography{man}
741:
742: \newpage
743:
744: \begin{table*}
745: \center \caption{The 18 orthogonal configurations of two
746: dissimilar biaxial particles. A unitary operator ($\mathbf{U}$) followed by a
747: translation is applied to the second particle to reach the desired
748: configuration. We adopt the naming scheme introduced by Berardi {\it
749: et al}~\cite{BFZ98}. The operator $\mathbf{R}_e$ denotes a $\pi/2$ rotation with
750: respect to the axis $e$. A two-letter code is attached to each
751: configuration with respect to the faces perpendicular to
752: connecting vector of the ellipsoids. A prime is added if one or
753: three axes are antiparallel. Italic codes refer to configurations which
754: are degenerate in homogeneous interactions.}
755: \begin{tabular}{p{45pt} p{35pt} p{35pt} p{35pt}}
756: \\
757: $\mathbf{U}$ & $\rl{\mathbf{r}}\|\hat{\mathbf{e}}_{x1}$ &
758: $\rl{\mathbf{r}}\|\hat{\mathbf{e}}_{y1}$ &
759: $\rl{\mathbf{r}}\|\hat{\mathbf{e}}_{z1}$ \\
760: \hline \\
761: $\mathbf{I}$ & aa & bb & cc \\
762: $\mathbf{R}_z$ & ab & \textit{ba}$^\prime$ & cc$^\prime$ \\
763: $\mathbf{R}_y$ & ac$^\prime$ & bb$^\prime$ & \textit{ca}$^\prime$ \\
764: $\mathbf{R}_x\mathbf{R}_y$ & \textit{ac} & \textit{ba} & \textit{cb} \\
765: $\mathbf{R}_z\mathbf{R}_x\mathbf{R}_y$ & aa$^\prime$ & \textit{bc}$^\prime$ & cb$^\prime$ \\
766: $\mathbf{R}_x^T\mathbf{R}_z^T$ & ab & bc & ca \\
767: \end{tabular}
768: \label{tab:ortho}
769: \end{table*}
770:
771: \newpage
772:
773: \begin{table*}
774: \center \caption{\resq potential parameters for homogeneous
775: interactions of selected molecules~\cite{Mols}. The oblate molecules
776: are: (1) Perylene (2) Pyrene (3) Coronene (4) Benzene. The prolate
777: molecules are: (5) Sexithiophene (6) Pentacene (7) Anthracene (8)
778: Naphthalene (9) Toluene.}
779: \begin{ruledtabular}
780: \begin{tabular}{lccccccccc}
781: Mol. No. & $A_{12}$($10^2$ Kcal/mol) & $\sigma_c$(\AA) &
782: $\sigma_x$(\AA) & $\sigma_y$(\AA) & $\sigma_z$(\AA) & $E_x$ & $E_y$
783: & $E_z$ & $\Omega (10^{-3})$\\
784: \hline
785: Oblate: \\
786: $(1)$ & 36.36 & 3.90 & 4.20 & 3.12 & 0.49 & 3.96 & 2.39 & 0.49 & 9.6\\
787: $(2)$ & 28.36 & 3.91 & 4.24 & 3.07 & 0.45 & 3.98 & 2.35 & 0.43 & 9.9\\
788: $(3)$ & 21.01 & 3.83 & 4.27 & 4.26 & 0.54 & 2.84 & 2.85 & 0.35 & 12.5\\
789: $(4)$ & 84.95 & 3.99 & 2.14 & 1.82 & 0.36 & 4.60 & 3.70 & 1.03 & 6.8\\
790: \hline
791: Prolate: \\
792: $(5)$ & 49.44 & 4.07 & 10.96 & 1.99 & 0.46 & 6.30 & 1.16 & 0.35 & 10.7 \\
793: $(6)$ & 37.46 & 3.88 & 6.56 & 2.28 & 0.47 & 5.52 & 1.65 & 0.43 & 9.7\\
794: $(7)$ & 45.51 & 3.85 & 4.19 & 2.25 & 0.44 & 6.83 & 2.48 & 0.62 & 9.2\\
795: $(8)$ & 37.76 & 3.82 & 3.09 & 2.20 & 0.49 & 4.59 & 3.39 & 0.77 & 10.9\\
796: $(9)$ & 23.19 & 3.75 & 2.72 & 2.04 & 0.57 & 4.61 & 3.29 & 1.00 & 10.6\\
797: \end{tabular}
798: \end{ruledtabular}
799: \label{tab:param}
800: \end{table*}
801:
802: %\begin{tabular}{p{80pt} p{100pt} p{30pt} p{30pt} p{30pt} p{30pt} p{30pt}}
803: %\\
804: %Mol. No. & $E_0$($10^4 \mathrm{\AA}^6$ Kcal/mol) & $\sigma_c$(\AA) &
805: %$\sigma_x$(\AA) & $\sigma_y$(\AA) & $\sigma_z$(\AA) & $\Omega$\\
806: %\hline
807: %Oblate: \\
808: %$[1]$ & 15.01 & 3.68 & 3.54 & 3.01 & 1.26 & 0.22\\
809: %$[2]$ & 1.69 & 3.93 & 1.65 & 1.83 & 0.82 & 0.09\\
810: %\hline
811: %Prolate: \\
812: %$[3]$ & 49.85 & 5.12 & 6.69 & 1.42 & 0.78 & 1.18 \\
813: %$[4]$ \\
814: %\hline
815: %Asymmetric: \\
816: %$[5]$ & 2.21 & 3.80 & 2.15 & 2.03 & 1.02 & 0.08\\
817: %\end{tabular}
818: %\label{tab:param}
819: %\end{table}
820:
821: \newpage
822:
823: \begin{table*}
824: \center \caption{Linear regression analysis between $(\delta
825: U/U)_{RMS}$ and T for a few selected molecules~\cite{Mols}. The
826: molecules are: (1) Benzene (2) Perylene (3) Sexithiophene. The
827: linear relationship is defined as $(\delta U/U)_{RMS}$ = AT+B in all
828: cases.}
829: \begin{tabular}{p{50pt} p{70pt} p{60pt} p{60pt}}
830: \\
831: Mol. No. & A($10^{-4}K^{-1}$) & B($10^{-3}$) & $R^2$ \\
832: \hline
833: $(1)$ & 0.66 & 1.0 & 0.987\\
834: $(2)$ & 1.39 & -4.1 & 0.983\\
835: $(3)$ & 2.35 & 1.3 & 0.995
836: \end{tabular}
837: \label{tab:err}
838: \end{table*}
839:
840: \newpage
841:
842: \begin{table*}
843: \center \caption{\resq potential parameters for the homogeneous
844: interactions of the pair perylene at different temperatures.}
845: \begin{ruledtabular}
846: \begin{tabular}{ccccccccccc}
847: Temperature (K) & $A_{12}$($10^4$ Kcal/mol) & $\sigma_c$(\AA) &
848: $\sigma_x$(\AA) & $\sigma_y$(\AA) & $\sigma_z$(\AA) &
849: $E_x$ & $E_y$ & $E_z$ & $\Omega (10^{-3})$\\
850: \hline
851: 100 & 14.59 & 4.83 & 3.77 & 2.69 & 0.10 & 3.56 & 2.13 & 0.41 & 15.2\\
852: 300 & 2.04 & 4.51 & 3.97 & 2.85 & 0.24 & 3.52 & 2.18 & 0.43 & 14.1\\
853: 500 & 1.21 & 4.40 & 4.04 & 2.91 & 0.30 & 3.49 & 2.20 & 0.44 & 12.6\\
854: 700 & 0.99 & 4.33 & 4.09 & 2.97 & 0.32 & 3.53 & 2.19 & 0.44 & 12.4\\
855: 900 & 0.82 & 4.31 & 4.09 & 2.98 & 0.35 & 3.43 & 2.15 & 0.44 & 12.6\\
856: \end{tabular}
857: \end{ruledtabular}
858: \label{tab:param_pery}
859: \end{table*}
860:
861: \newpage
862:
863: \begin{figure*}
864: \center
865: \includegraphics[bb=40 195 560 600, scale=0.4]{fig1.eps}
866: \includegraphics[bb=40 195 560 600, scale=0.4]{fig2.eps}
867: \caption{A comparison between the \resq and biaxial-GB potentials
868: for the homogeneous interaction of the pair perylene in a set of
869: uniform random center separations in the range [5\AA, 50\AA] along
870: with random rotations. The Gay-Berne approximation has been used for
871: the least contact distance. (a) A log-log plot of $U_{RE^2}/U_{MM3}$
872: against $U_{MM3}$ (Mean=-0.002, SD=0.08) (b) A log-log plot of
873: $U_{GB}/U_{MM3}$ against $U_{MM3}$ (Mean=-0.87, SD=0.54)}
874: \label{fig:loglog}
875: \end{figure*}
876:
877: \newpage
878:
879: \begin{figure*}
880: \center
881: \includegraphics[bb=40 195 560 600, scale=0.35]{fig3.eps}
882: \includegraphics[bb=40 195 560 600, scale=0.35]{fig4.eps}
883: \includegraphics[bb=40 195 560 600, scale=0.35]{fig5.eps}
884: \includegraphics[bb=40 195 560 600, scale=0.35]{fig6.eps}
885: \includegraphics[bb=40 195 560 600, scale=0.35]{fig7.eps}
886: \caption{The heterogeneous interaction between the pair perylene
887: (oblate) and sexithiophene (prolate) for the 18 orthogonal
888: configurations. The black thick lines denote the \resq potential,
889: the red dashed lines refer to the biaxial-GB~\cite{BFZ98} and the
890: reference atomistic summation (MM3~\cite{MM3}) is denoted by blue
891: thin lines. A combination of homogeneous interaction parameters
892: (Table~\ref{tab:param}) have been used without further optimization.
893: The error measures are: $\Omega_{RE^2}=6.5\times 10^{-3},
894: \Omega_{GB}=7.7\times 10^{-3}$. The graphs are grouped in five
895: plates as: side-by-side (A), cross (B), T-shaped 1 (C), T-shaped 2
896: (D) and end-to-end (E) interactions and are labeled according to
897: the notation introduced in Table~(\ref{tab:ortho}).}
898: \label{fig:orthoplot}
899: \end{figure*}
900:
901: \newpage
902:
903: \begin{figure*}
904: \center
905: \includegraphics[bb=35 185 560 610, scale=0.4]{fig8.eps}
906: \caption{The average computation time of exact LJ(6-12) atomistic
907: summation with respect to the average number of interacting sites (blue dashed line)
908: and \resq single-site potential (red continuous line).}
909: \label{fig:time}
910: \end{figure*}
911:
912: \newpage
913:
914: \begin{figure*}
915: \center
916: \includegraphics[bb=35 185 560 610, scale=0.4]{fig9.eps}
917: \caption{A comparison between the truncated atomistic descriptions
918: with hard atomic cutoffs and the \resq potential for the end-to-end
919: interaction of the pair Pentacene. Thick lines denote the \resq
920: potential while thin lines represent atomistic summations with
921: different atomic cutoffs (6, 9 and 12 \AA). (A) Logarithmic relative
922: error of the \resq potential and truncated atomistic summations
923: (with respect to the exact atomistic summations) vs. center
924: separation. (B) Time consumption of different approximations vs.
925: center separation.} \label{fig:err_cut}
926: \end{figure*}
927:
928: \newpage
929:
930: \begin{figure*}
931: \center
932: \includegraphics[bb=35 185 560 610, scale=0.4]{fig10.eps}
933: \caption{Relative deviations from the PMF vs. temperature for three different
934: molecules. The signs indicate the MD simulation data. The continuous
935: lines are linear regressions. (1) Plus signs: Benzene (2) Cross
936: signs: Perylene (3) Dots: Sexithiophene.} \label{fig:err}
937: \end{figure*}
938:
939: \newpage
940:
941: \begin{figure*}
942: \center
943: \includegraphics[bb=35 185 560 610, scale=0.4]{fig11.eps}
944: \caption{Potential of Mean Force between the pair perylene for the
945: cross configuration {\bf bc}. The dashed line indicate the
946: interaction potential of the unperturbed structures while the
947: continuous lines, ordered descending with respect to their
948: well-depths, represent the PMF at temperatures 100K, 300K, 500K,
949: 700K and 900K, respectively. } \label{fig:pmf}
950: \end{figure*}
951:
952: \end{document}
953: