1: \documentclass[12pt]{article}
2: \usepackage[dvips]{epsfig}
3: \usepackage{cite,overcite,amsmath,amssymb,dcolumn}
4: \usepackage{typearea}\typearea{22}
5: \newcommand{\mul}[1]{\multicolumn{1}{c}{#1}}
6: \newcommand{\mult}[1]{\multicolumn{2}{c}{#1}}
7: \newcommand{\opC}{\operatorname{C}}
8: \newcommand{\opH}{\operatorname{H}}
9: \begin{document}
10: \title{Local Structure and Dynamics of {\it Trans}-polyisoprene oligomers}
11: \author{Roland Faller$^{1,}$\footnotemark \and Florian M{\"u}ller-Plathe$^1$
12: \and
13: Manolis Doxastakis$^2$ \and Doros Theodorou$^2$ \\
14: \small $^1$ Max-Planck-Institut f\"ur Polymerforschung, Ackermannweg 10,
15: D-55128 Mainz, Germany\\
16: \small $^2$ Department of Chemical Engineering, University of Patras \\
17: \small and ICE/HT-FORTH , GR-26500 Patras, Greece
18: }
19: \maketitle
20: %\renewcommand{\baselinestretch}{2}
21: \setcounter{footnote}{1}
22: \renewcommand{\thefootnote}{\fnsymbol{footnote}}
23: %For submission to {\it Macromolecules}
24: \footnotetext{Corresponding author, new address: Department of Chemical
25: Engineering, University of Wisconsin, Madison, WI 53706, USA}
26: \abstract{
27: \noindent Mono- and poly-disperse melts of oligomers (average length 10
28: monomers) of {\it trans}-1,4-polyisoprene are simulated in full atomistic
29: detail. The force-field is developed by means of a mixture of {\it ab
30: initio} quantum-chemistry and an automatic generation of empirical
31: parameters. Comparisons to NMR and scattering experiments validate the
32: model. The local reorientation dynamics shows that for C$-$H vectors there is
33: a two-stage process consisting of an initial decay and a late-stage
34: decorrelation originating from overall reorientation. The atomistic model
35: can be successfully mapped onto a simple model including only beads
36: for the monomers with bond springs and bond angle potentials.
37: End-bridging Monte Carlo as an equilibration stage and molecular dynamics
38: as the subsequent simulation method together prove to be a useful method for
39: polymer simulations.
40: }
41: \clearpage
42: %
43: \section{Introduction}
44: %
45: The understanding of polymer materials from the very local to the macroscopic
46: scale is at the focus of theoretical material science. Simulations are a
47: useful means for reaching this goal. Fully detailed simulations incorporating
48: interaction centers for all atoms allow extensive investigations of
49: polymer-specific models to compare directly to high resolution experiments
50: like neutron scattering\cite{moe99}, positronium annihilation
51: spectroscopy\cite{schmitz00a} or nuclear magnetic
52: resonance\cite{mplathe00}. In contrast to simple generic models\cite{grest86},
53: atomistic simulations capture the differences between polymer species. This
54: understanding is a prerequisite for the design of tailormade materials for
55: special applications.
56:
57: The present contribution investigates, as an example, oligomers of {\it
58: trans}-1,4-polyisoprene. We focus on {\it local} structure and dynamics as
59: the local scale is influenced mainly by the specific chemical structure. In
60: this range all-atom models are necessary for a reliable static and dynamic
61: investigation\cite{gunsteren90}. Moreover we compare our results to NMR
62: measurements, which always sample atom-atom vectors. Therefore, it is useful
63: for the analysis to have all atoms present. For the properties under
64: investigation short chains are sufficient, as the connectivity and
65: non-crossability contribute to phenomena on the generic large-scale
66: level\cite{kremer90,faller00sa} which is not of interest here. Several
67: experiments have been performed aimed at the local properties of polyisoprene
68: like reorientation times and the
69: packing\cite{batie89,denault89,zorn92,laupretre93,dlubek98,english98}.
70: Additionally, simulations of pure {\it cis}-polyisoprene are already
71: known\cite{moe96a,moe99}. To our knowledge there are, however, not yet
72: simulations of the {\it trans} conformer. Industrially this conformer is not
73: as important as the {\it cis}-conformer. However, typical polyisoprene
74: materials include several percent {\it trans} content. In order to understand
75: the influence of this content, it is worth investigating pure {\it
76: trans}-polyisoprene and look for differences from the other conformers.
77: %
78: \section{The simulation model and technical details}
79: %
80: \label{sec:ff}
81: %
82: \subsection{Starting structures from quantum chemistry}
83: %
84: Three different chain length mixtures, each containing 100 oligomers (13200
85: atoms) of polyisoprene with an average molecular weight of 682~g/mol
86: (corresponding to 10 monomers) are simulated at temperatures of 300~K and
87: 413~K. One system is mono-disperse (in the following referred to as system~1)
88: and is simulated at constant density of 890~kg/m$^3$ (experimental density of
89: the {\it cis}-conformer: 900~kg/m$^3$)\cite{fetters99a} in an orthorhombic box
90: under periodic boundary conditions with box lengths:
91: $4.97\text{nm}\times5.16\text{nm}\times4.97\text{nm}$. This is about two and
92: a half times the end-to-end distance of the chains, thus artifacts from
93: interactions of chains with their mirror images are absent. The torsional
94: conformations were set up in the equilibrium distribution derived from quantum
95: chemical calculations. Twenty-seven configurations of a dimer were
96: investigated with quantum chemical calculations (for details see
97: Section~\ref{sec:ffex}). Only seven of them are actually at low energy
98: (Table~\ref{tab:relmin}) and only their statistics was used to generate initial
99: torsional distributions. The chains were packed in a simulation box of
100: $9\text{nm}\times9\text{nm}\times9\text{nm}$ and compressed to the final
101: density in steps, using molecular dynamics. During this procedure anisotropic
102: box fluctuations were allowed to better equilibrate stresses in the box
103: leading to the orthorhombic box explained above. At the same time the time-step
104: was increased from 0.01~fs to 1~fs.
105: %
106: \subsection{Starting structure via End-bridging Monte Carlo}
107: %
108: The two polydisperse systems (systems~2 and~3) were additionally equilibrated
109: using the end-bridging Monte Carlo (EBMC) procedure\cite{pant95,mavrantzas99}
110: before the molecular dynamics simulations were performed at a constant
111: pressure of 101.3~kPa to determine the density in a poly-disperse melt.
112: We used two systems with slightly different chain length distributions (see
113: below) in order to increase the statistics and to look for influences of the
114: actual realization of the ensemble.
115:
116: The EBMC was performed in a united-atom model as the positions of the
117: hydrogens are not important at this stage. The non-bonded interaction
118: potential parameters for this procedure are shown in Table~\ref{tab:ebff}. All
119: Monte Carlo steps are accepted according to the Metropolis criterion.
120:
121: The end-bridging steps were performed using a hypothetical chain (see dashed
122: line in Figure~\ref{fig:PI}). The monomers on this chain consist only of the
123: carbons 1, 4, and 5. Unlike the original procedure for
124: polyethylene\cite{pant95,mavrantzas99}, moves can
125: only be performed breaking the bonds between carbon 5 and 1 as the topology of
126: {\it trans}-polyisoprene must not be altered. As the bridging procedure
127: always needs three trimers: one to bridge from, one to bridge to, and one
128: as the bridge, the three carbons of one pseudo-monomer act as such trimers.
129: After the respective move the positions of carbons~2 and~3 have to be
130: recalculated. Since the double bond keeps
131: the whole monomer in plane the positions are defined exactly if bond lengths
132: and angles are left unchanged. In the Monte Carlo procedure no bond length or
133: bond angle was changed, neither for the hypothetical chain, nor for the
134: atomistic chain. Any of these local degrees of freedom were left for the
135: molecular dynamics to equilibrate.
136:
137: \begin{table}
138: \begin{center}
139: \[
140: \begin{array}{lD{.}{.}{-1}D{.}{.}{-1}}
141: \hline
142: \mul{\text{Interaction}} & \mul{\sigma [nm]} &
143: \mul{\epsilon[kJ/mol]}\\
144: \hline
145: \text{C$-$C},\text{C$-$CH},\text{CH$-$CH} & 0.38 & 0.4186 \\
146: \text{C$-$CH}_2,\text{CH$-$CH}_2 & 0.4257 & 0.4249 \\
147: \text{C$-$CH}_3,\text{CH$-$CH}_3 & 0.4257 & 0.6299 \\
148: \text{CH}_2\text{$-$CH}_2 & 0.45 & 0.3918 \\
149: \text{CH}_2\text{$-$CH}_3 & 0.45 & 0.6095 \\
150: \text{CH}_3\text{$-$CH}_3 & 0.45 & 0.9479 \\
151: \hline
152: \end{array}
153: \]
154: \end{center}
155: \caption{Nonbonded parameters (Lennard-Jones 12-6) used in the united atom
156: end-bridging Monte Carlo simulations of {\it trans}-polyisoprene. Bond
157: lengths, angles and torsion potentials are the same as in the all atom
158: simulations. All interactions within a topological distance of up to~3
159: bonds are excluded, as they are taken into account in the torsion
160: potentials.}
161: \label{tab:ebff}
162: \end{table}
163: %
164: Anisotropic box fluctuations were allowed but no shearing. The systems are
165: slightly poly-disperse, as this version of EBMC does not work at constant
166: topology, i.e. chain length distribution. The molecular weight distribution is
167: sharply peaked at $N=10$ (Figure~\ref{fig:distMW}) as we did not intend to
168: equilibrate the chain length distribution but to investigate nearly
169: mono-disperse systems. We started with a monodisperse sample and let the
170: end-bridging procedure run for a limited time with chemical potential $\mu=0$
171: for $5\le N\le15$ and $\mu=-\infty$ elsewhere. Strict monodispersity, however,
172: would require much longer equilibration times as EBMC would be prohibited. On
173: average, every chain underwent 2.8 successful end-bridging moves in addition
174: to many local moves before the molecular dynamics started. The resulting
175: distribution shows two weak side-peaks at $N\approx6$ and $N\approx14$. These
176: can be explained by the procedure. Chain lengths of 9 or 11 are not directly
177: accessible by endbridging moves starting from the initial monodisperse ($N=10$)
178: sample, as no single monomers can be transferred. Since they have to be
179: reached indirectly, they have a low probability.
180:
181:
182: \begin{figure}
183: \begin{center}
184: \includegraphics[angle=-90,width=0.5\linewidth]{molweight.eps}
185: \caption{Molecular weight distributions of systems~2 (filled bars) and~3
186: (open bars)}
187: \label{fig:distMW}
188: \end{center}
189: \end{figure}
190:
191:
192: %
193: \subsection{The force-field}
194: %
195: \label{sec:ffex}
196: %
197: As the torsion potentials are very important for the configurations of the
198: chains, quantum chemical calculations were performed with the packages
199: Gaussian~94\cite{gaussian94} and Gaussian~98\cite{gaussian98}. The energy
200: differences between different minima were calculated with hybrid density
201: functional calculations (B3LYP)\cite{becke93} and the barrier heights are
202: calculated by Hartree-Fock using a 6-31G** basis set on a dimer of
203: trans-polyisoprene. In order to calculate barrier heights, constrained
204: optimizations at fixed bond lengths are applied where the torsions are
205: changed in steps of 15 or 30~degrees. The results were fitted to a Fourier
206: series in the torsion angle with four terms. The constant energy shift is
207: discarded as it does not enter the forces
208: %
209: \begin{equation}
210: V_{\text{tors}}=\sum_{i=1}^{3}k_{\tau}/2\Big[1-\cos[i(\tau-\tau_0)]\Big].
211: \end{equation}
212: %
213: The resulting parameters are shown in Table~\ref{tab:torspot}. The relevant
214: minima are first supposed to occur in {\it trans} and {\it gauche} states.
215: %
216: \begin{table}
217: \begin{center}
218: \begin{tabular}{crcrcc}
219: \hline
220: torsion & \mult{dihedral angle} & \mult{strength} & periodicity \\
221: & \mult{$\tau_0$ [degrees]} & \mult{$k_{\tau}$[kJ/mol]} & $i$ \\
222: \hline
223: 1 & 0 & & 5.2 & & 1 \\
224: 1 & 0 & & -7.4 & & 2 \\
225: 1 & 0 & & 10.0 & & 3 \\
226: 2 & 180 & & 9.7 & & 1 \\
227: 2 & 180 & & 14.1 & & 3 \\
228: 3 & 0 & &-21.1 & & 1 \\
229: 3 & 0 & &-12.3 & & 2 \\
230: 3 & 0 & & 0.5 & & 3 \\
231: \hline
232: \end{tabular}
233: \caption{Force-field parameters for the torsion angles. For the
234: nomenclature see Figure~\ref{fig:PI}.}
235: \label{tab:torspot}
236: \end{center}
237: \end{table}
238: %
239: \begin{table}
240: \begin{center}
241: \[
242: \begin{array}{cD{.}{.}{-1}D{.}{.}{-1}D{.}{.}{-1}D{.}{.}{-1}}
243: \hline
244: \text{Conf.} & \mul{\text{Torsion 1}} & \mul{\text{Torsion 2}} &
245: \mul{\text{Torsion 3}} &
246: \mul{\Delta E [\text{kJ/mol}]}\\
247: & \mul{\text{C$_2-$C$_3-$C$_5-$C$_1$}}
248: & \mul{\text{C$_3-$C$_5-$C$_1-$C$_2$}}
249: & \mul{\text{C$_5-$C$_1-$C$_2-$C$_3$}} & \\
250: \hline
251: 1 & 180.0 & 180.0 & 180.0 & 42.56\\
252: 2 & -111.2 & 176.1 & 115.7 & 0.33\\
253: 3 & 117.4 & -62.9 & 133.5 & 2.87\\
254: 4 & 115.3 & -63.8 & -124.0 & 2.28\\
255: 5 & -109.3 & -177.7 & -115.3 & 0.00\\
256: 6 & -94.9 & -70.5 & 123.6 & 4.91\\
257: 7 & -86.0 & -59.0 & -108.3 & 6.72\\
258: \hline
259: \end{array}
260: \]
261: \caption{Relevant torsion conformations after geometry optimization using
262: the hybrid method 6$-$311G**/B3LYP.}
263: \label{tab:relmin}
264: \end{center}
265: \end{table}
266: %%
267: \begin{figure}
268: \begin{center}
269: \includegraphics[width=0.5\linewidth]{sketch.eps}
270: \caption{Carbons in the dimer of {\it trans}-polyisoprene with definition
271: of the torsions. Additionally, the hypothetical chain for the
272: end-bridging moves is shown as dashed line.}
273: \label{fig:PI}
274: \end{center}
275: \end{figure}
276: %
277: The initial torsion states are set up according to this distribution (before
278: EBMC) in a Markov chain way, i.e. the chains are built monomer by monomer and
279: the triple of torsions connecting two monomers is selected by random numbers
280: according the energy distribution.
281:
282: The bond angle potential and the improper dihedral potential, which keeps the
283: atoms at the double bond in plane, are harmonic with force constants adapted
284: from previous simulations of small molecules
285: (Table~\ref{tab:ffangbnd}).\cite{faller99c,schmitz99a} The harmonic dihedral
286: has a strength of 160~kJ/(mol*rad$^2$) and is applied to the dihedrals
287: C$_1-$C$_2-$C$_3-$C$_4$, H$_{\text{C2}}-$C$_2-$C$_3-$C$_5$, and
288: C$_2-$C$_3-$C$_4-$C$_5$.
289: %
290: \begin{table}
291: \[
292: \begin{array}{crc|cD{.}{.}{-1}}
293: \hline
294: \text{angle} & \mul{\phi_{0}} & k \text{[kJ/(mol*rad}^2)] & \text{bond}
295: & l_{b} \text{[nm]} \\
296: \hline
297: \opC_1-\opC_2-\opC_3 & 128.7 & 250 & \opC_1-\opC_2 & 0.150 \\
298: \opC_2-\opC_3-\opC_4 & 124.4 & 250 & \opC_2=\opC_3 & 0.1338 \\
299: \opC_2-\opC_3-\opC_5 & 120.2 & 250 & \opC_3-\opC_4 & 0.151 \\
300: \opC_4-\opC_3-\opC_5 & 115.4 & 250 & \opC_3-\opC_5 & 0.1515 \\
301: \opC_3-\opC_5-\opC_1 & 114.5 & 250 & \opC_5-\opC_1 & 0.155 \\
302: \opC_5-\opC_1-\opC_2 & 112.7 & 250 & \opC-\opH & 0.109 \\
303: \opC-\opC_{\text{sp3}}-\opH & 109.5 & 250 & & \\
304: \opC_1-\opC_2-\opH & 114.4 & 250 & & \\
305: \opC_3-\opC_2-\opH & 114.4 & 250 & & \\
306: \opH-\opC-\opH & 109.5 & 250 & & \\
307: \hline
308: \end{array}
309: \]
310: \caption{Angles and bond lengths for atomistic simulations. The
311: equilibrium values of the angles $\phi_{0}$ and the bond lengths $l_{b}$
312: are derived from density functional calculations using
313: Gaussian~94\cite{gaussian94} and experimental data\cite{CRC}, $k$ is the
314: force constant for the harmonic bond angle potential. The bond lengths
315: are constrained.}
316: \label{tab:ffangbnd}
317: \end{table}
318: %
319: Also the non-bonded interaction parameters (Table~\ref{tab:ff}) are adapted
320: from simulations of cyclohexene.\cite{schmitz99a} They were derived
321: using the automatic simplex parameterization.\cite{faller99c} The polymer
322: density was not fully satisfactory with the original cyclohexene
323: force-field.\cite{faller00a} Thus, the $\sigma$ value of the hydrogens was
324: slightly increased to reproduce the experimental density at 300~K.
325: Lennard-Jones interactions between unlike atoms were based on the
326: Lorentz-Berthelot mixing rules.\cite{allen87}
327: %
328: \begin{table}
329: \[
330: \begin{array}{cD{.}{.}{-1}D{.}{.}{-1}D{.}{.}{-1}}
331: \hline
332: \text{atom} & \mul{m\text{[amu]}} & \mul{\sigma\text{[nm]}} &
333: \mul{\epsilon\text{[kJ/mol]}}\\
334: \hline
335: \opC_{\text{sp2}} & 12.01 & 0.321 & 0.313\\
336: \opC_{\text{sp3}} & 12.01 & 0.311 & 0.313\\
337: \opH & 1.00782 & 0.24 & 0.2189\\
338: \hline
339: \end{array}
340: \]
341: \caption{Force-field parameters for the non-bonded interactions. $m$ is
342: the atom mass, $\sigma$ the interaction radius, and $\epsilon$ the
343: interaction strength. }
344: \label{tab:ff}
345: \end{table}
346:
347: %%
348: \begin{table}
349: \[
350: \begin{array}{ccrrD{.}{.}{-1}}
351: \hline
352: \text{system} & \text{T[K]} & t_{sim}\text{[ps]}
353: & \frac{M_w}{M_n} & \mul{\rho\text{[kg/m}^3]} \\
354: \hline
355: 1 & 300 & 1184 & 1.00 & 890\\
356: 2 & 300 & 2012 & 1.05 & 917.4\\
357: 3 & 300 & 1737 & 1.05 & 916.8\\
358: 3 & 413 & 792 & 1.05 & 826\\
359: \hline
360: \end{array}
361: \]
362: \caption{Simulation parameters for the three systems under study. Before
363: the above-mentioned simulation time was started a few hundred
364: picoseconds equilibration time was waited for.}
365: \label{tab:simlen}
366: \end{table}
367: %
368:
369: Atoms connected by any bonding potential did not interact by the Lennard-Jones
370: potential. Additionally, the following non-bonded interactions were
371: excluded: all within one monomer, and all C$-$C, C$-$H and H$-$H
372: interactions between the second half of the carbons of one monomer (atoms
373: C$_3$, C$_4$, and C$_5$) and the first half of its following neighbor (C$_1$
374: and C$_2$) including the hydrogens connected to them for system~1. For
375: system~2 and~3 the latter C$-$H and H$-$H interactions were not excluded (only
376: up to ``$1-4$'' interactions), which leads to differences as the different
377: torsion states alter the distances between these atoms so that the effective
378: energies of the torsions are shifted. This happened due to an oversight in the
379: initial simulations. As the simulations are very time consuming we deemed a
380: complete repetition not necessary.
381:
382: Constant temperature and pressure are ensured using Berendsen's
383: method~\cite{berendsen84} with time constants 0.2~ps for temperature and 8~ps
384: (system~2) or 12~ps (system~3) for pressure, respectively. The pressure
385: coupling using a compressibility of $2\times10^{-7}~\text{kPa}^{-1}$ was
386: employed for the three directions independently. All simulations were performed
387: using the YASP molecular simulation package\cite{yasp} with a time-step of 1~fs
388: and a cutoff for the non-bonded interactions at 0.9~nm. Configurations were
389: saved every picosecond. Simulation lengths and polydispersities are shown in
390: Table~\ref{tab:simlen}. Bond lengths are constrained to the desired values of
391: Table~\ref{tab:ffangbnd} using the SHAKE algorithm.\cite{ryckaert77,mplathe91}
392: %
393: \section{Thermodynamic and static structural properties}
394: %
395: \subsection{Density}
396: %
397: The systems simulated under constant pressure conditions arrived at densities
398: of 917.4~kg/mol (system 2) and 916.8~kg/mol (system 3), respectively, which is
399: a discrepancy of less than 2\% relative to the experimental value of
400: 900~kg/mol for chains of 16 monomer length~\cite{fetters99a}. The isochoric
401: simulation of system 1 had a pressure of -500~kPa. In $NVT$ simulations, a
402: negative pressure of this magnitude means that the system is not exactly at
403: the correct density but would like to contract a little further. As all
404: densities are quite close to the experimental value, which itself is not too
405: certain, since it was determined for a mixture of {\it cis}- and {\it
406: trans}-PI, a closer refinement of the force-field parameters was not deemed
407: necessary. The density also depends weakly on the intra-chain part of the
408: potential. If the torsion potentials are switched off, the density increases
409: by about 2\%, as the chains can pack more effectively (System~1).
410: %
411: \subsection{Single chain properties}
412: %
413: The mean squared end-to-end distance of the oligomers was measured between the
414: terminating carbons (C$_1^{\text{mono 1}}$ and C$_5^{\text{mono~}n}$). The
415: respective result of 5.12~nm$^2$ corresponds to 0.0075~nm$^2$mol/g for the
416: monomer-weight-specific end-to-end distance. The experimental value of
417: 0.0060~nm$^2$mol/g per monomer is for a mixture of {\it trans} and {\it cis}
418: polyisoprene. The experiments are performed doing small angle neutron
419: scattering in a melt of 7-mers.\cite{fetters94}
420:
421: The persistence length $l_p$ measures direction correlations of unit vectors
422: along the chain. It is calculated using suitably defined points along the
423: backbone. The function
424: %
425: \begin{equation}
426: \big\langle\vec{u}(\vec{r})\cdot\vec{u}(\vec{r}_0)\big\rangle=
427: e^{-\frac{|\vec{r}-\vec{r}_0|}{l_p}}
428: \end{equation}
429: %
430: is fitted against the bond correlation. There is some freedom in how to define
431: the tangent vector in an atomistic model. The vectors connecting the C$_1$ (or
432: the C$_2$) atoms of adjacent monomers were investigated. Additionally,
433: intra-monomer vectors along the double-bond and vectors connecting the two end
434: carbons of the same monomer (C$_1-$C$_5$) are included. The bond-correlation
435: functions are not really exponential (Figure~\ref{fig:atobc}), as the
436: interactions are complex and the chains are short. Thus, persistence lengths
437: deduced from the fitting procedure can only be taken as estimates.
438: End-effects were not excluded for reasons of statistics. The correlation
439: functions are different for different vectors. The vector representing the
440: shortest connection (the double-bond) has the shortest correlation length.
441: %
442: \begin{figure}
443: \begin{center}
444: \includegraphics[angle=-90,width=0.5\linewidth]{atomisticbondcor.eps}
445: \end{center}
446: \caption{Bond vector correlation functions in the atomistic polyisoprene
447: simulations of system~1. The filled symbols are correlations between
448: direction vectors in the monomers. (diamond: double-bonds, circle
449: C$_1-$C$_5$) The open symbols correspond to vectors connecting atoms
450: belonging to neighboring monomers (squares: C$_2$, triangles C$_1$).
451: }
452: \label{fig:atobc}
453: \end{figure}
454: %%
455: The persistence lengths range between 0.5 (for the double bonds) and 1.5 (for
456: the intermonomer vectors) in monomer diameters ($\approx
457: 0.3$~nm). Polyisoprene is, therefore, rather flexible already on the length
458: scale of a few monomers. The monomer itself is intrinsically stiff, but the
459: three torsions between neighboring monomers provide a flexible link.
460: %
461: \subsection{Local packing of chains}
462: %
463: The local structure in the melt can be characterized by different pair
464: distribution functions. Figure~\ref{fig:rdf} shows inter-chain radial
465: distribution functions of the different atom pairs present in polyisoprene.
466: %
467: \begin{figure}
468: \begin{center}
469: \includegraphics[angle=-90,width=0.45\linewidth]{rdfall.eps}
470: \includegraphics[angle=-90,width=0.45\linewidth]{rdfC-C.eps}
471: \includegraphics[angle=-90,width=0.45\linewidth]{rdfC-H.eps}
472: \includegraphics[angle=-90,width=0.45\linewidth]{rdfH-H.eps}
473: \caption{Interchain radial distributions of the different atomic pairs
474: (full line system~1, broken line system~2) a) all atoms, b) only
475: carbons, c) only carbon hydrogen d) only hydrogen pairs}
476: \label{fig:rdf}
477: \end{center}
478: \end{figure}
479: %
480: The curves between the different systems differ only slightly due to density
481: differences. The conspicuous absence of a distinctive first peak in the
482: all-atom and the hydrogen $g(r)$ at short distances indicates that chains
483: cannot easily interpenetrate. Structure is averaged out in the all-atom $g(r)$,
484: in contrast to, e.g., the carbon-carbon distribution function, which exhibits
485: several clear peaks. However, also these do not rise to values
486: $g_{\text{CC}}(r)>1$.
487:
488: Figure~\ref{fig:chainchainrdf} shows the center-of-mass distribution of whole
489: chains.
490: %
491: \begin{figure}
492: \[
493: \includegraphics[angle=-90,width=0.5\linewidth]{rdfcm.eps}
494: \]
495: \caption{Center of mass radial distribution function of the atomistic
496: polyisoprene chains. A correlation hole on local scales is seen, on bigger
497: length scales the distribution is flat. The dashed line corresponds
498: to system~1, the solid line to system~2, the dotted line to system~3.}
499: \label{fig:chainchainrdf}
500: \end{figure}
501: %
502: In spite of the statistical noise it is clear that they can approach as close
503: as 0.2~nm. The similarity of the RDFs for systems~2 and~3 suggests that the
504: combination of EBMC and MD leads to a reliable structure. At distances of
505: more than 1~nm the distribution is quite flat, at shorter distances a
506: correlation hole is
507: clearly visible. System~1, which was not equilibrated using the end-bridging,
508: exhibits unrealistically sharp peaks. The differences in the molecular weight
509: distribution also contribute to the different structures. However, molecular
510: dynamics alone is not able to equilibrate a simulation of this size. The
511: introduction of EBMC brings us a good step further, although some remnants of
512: the setup may still be present.
513:
514: Carbon$-$carbon radial distribution functions (RDF) resolved according to
515: carbon type allow the study of preferential arrangements between different
516: chains. Partial pair distribution functions between the five different carbons
517: present in polyisoprene were recorded (Figure~\ref{fig:partialrdf}).
518: %
519: \begin{figure}
520: \includegraphics[angle=-90,width=0.5\linewidth]{rdf1.eps}
521: \includegraphics[angle=-90,width=0.5\linewidth]{rdf2.eps}
522: \includegraphics[angle=-90,width=0.5\linewidth]{rdf3.eps}
523: \includegraphics[angle=-90,width=0.5\linewidth]{rdf4.eps}
524: \begin{minipage}{0.5\linewidth}
525: \includegraphics[angle=-90,width=\linewidth]{rdf5.eps}
526: \end{minipage}
527: \begin{minipage}{0.5\linewidth}
528: \[
529: \begin{array}{lrrrrr}
530: & \opC_1 & \opC_2 & \opC_3 & \opC_4 & \opC_5\\
531: \opC_1 & 0.41 & 0.36 & 0.28 & 0.49 & 0.42 \\
532: \opC_2 & 0.36 & 0.31 & 0.25 & 0.43 & 0.36 \\
533: \opC_3 & 0.28 & 0.25 & 0.16 & 0.33 & 0.25 \\
534: \opC_4 & 0.49 & 0.43 & 0.33 & 0.58 & 0.49 \\
535: \opC_5 & 0.42 & 0.36 & 0.25 & 0.49 & 0.40 \\
536: \end{array}
537: \]
538: \end{minipage}
539: \caption{Partial inter-chain radial distribution functions (system~2,
540: T=300~K). a) C$_1$ b) C$_2$ c) C$_3$ d) C$_4$ e) C$_5$. In all figures the
541: definition of line styles in figure~c) applies. Figure f (inset table):
542: Integrated number of neighbors in the innermost shell
543: $r<0.45\text{nm}$, only foreign chains.}
544: \label{fig:partialrdf}
545: \end{figure}
546: %
547: The methyl carbon (C$_4$) is the most exposed and can, therefore, approach
548: closest to the others ($\approx0.4$nm). At this distance there is also
549: a shoulder in the C$_1$, C$_2$ and C$_5$-RDF indicating direct contact. C$_3$
550: is the most ``shielded'' carbon with a slight shoulder at direct contact to
551: C$_1$. It is linked to C$_2$, C$_4$ and
552: C$_5$, thus, it is often found as second contact ($\approx0.55$nm).
553: The two methylene carbons C$_1$ and C$_5$ are coordinated very
554: similarly. C$_2$ is easily accessible, as it
555: has only one hydrogen but not very exposed to other chains, since
556: it is part of the double bond in the backbone leading to weak structure.
557:
558: Integration of the pair distribution function yields the number of neighbors
559: of an atom in a shell of radius $r$. By relating the local number density
560: $\rho_{\text{local}}=\frac{n_{\text{C}}}{4/3\pi r^3},\,n_{\text{C}}$ being the
561: number of carbons, to the overall concentration $\rho(\infty)=8.102$nm$^{-3}$,
562: local enrichment ($x:=\frac{\rho_{\text{local}}}{\rho(\infty)}>1)$ and
563: depletion ($x<1)$ can be resolved.
564: Overall integrated values in Figure~\ref{fig:partialrdf}f are smaller than
565: unity due to the correlation hole, as only foreign chains are taken into
566: account. In the innermost shell ($r<0.45$nm, cf. Figure~\ref{fig:partialrdf}f)
567: of all carbons methyl groups (C$_4$) are enriched, sp$^2$ carbons (C$_2$ and
568: esp. C$_3$) are depleted, whereas the methylenes (C$_1$ and C$_5$) are close
569: in concentration. In the second shell ($0.45<r<0.65$~nm), this is partly
570: reversed, as C$_3$ is enriched and C$_4$ is weakly depleted. At distances
571: larger than 0.65~nm all species are found at bulk concentration. In brief, one
572: can say that the methylene carbons occur at constant density almost
573: everywhere. The methyl and the double bonded carbons (esp. C$_3$) show much
574: more structure. Monomers of different chains thus approach each other
575: typically with their side groups as closest contact. Orientational influences
576: from the double-bond keeping the monomer planar play a role, too (see
577: below). The whole overall structure in the RDFs extends about two monomer
578: sizes ($r<1$~nm), whereas the concentrations of different carbons level out
579: already at 0.65~nm.
580:
581: The local structure is not fully described by the (spherically averaged)
582: radial distribution functions. Mutual orientation is
583: %
584: \begin{figure}
585: \includegraphics[angle=-90,width=0.5\linewidth]{dblbndodf.eps}
586: \includegraphics[angle=-90,width=0.5\linewidth]{dblbndp1.eps}
587: \includegraphics[angle=-90,width=0.5\linewidth]{intermonoodf.eps}
588: \caption{Static {\it inter} chain orientation correlations {\it OCF}
589: (T=300~K, system~1). a) Double-bonds and vectors connecting C$_5$ to
590: C$_1$ of the next monomer between neighboring chains. Dot-dashed:
591: atomistic double bond {\it OCF}; solid: C$_5-$C$_1$ vectors; dashed: {\it
592: OCF} of a simple model with persistence length $l_p=1.5$ monomer
593: units\cite{faller00b} scaled for coincidence at the first maximum to the
594: solid one. b) Same as figure~a) but $P_1$ in order to show the
595: directionality of the correlations. c) Inter-monomer vectors, solid line:
596: C$_1-$C$_1$, dot-dashed line: C$_2-$C$_2$, dashed line: simple model for
597: comparison. Note the different ordinate scale.}
598: \label{fig:atoodf}
599: \end{figure}
600: %
601: measured by the orientation correlation function
602: {\it OCF} (Figure~\ref{fig:atoodf})
603: %
604: \begin{equation}
605: OCF(r)=P_{2}(r)=\frac{1}{2}\Big\langle
606: 3(\vec{u}_i\cdot\vec{u}_j)^2-1\Big\rangle\;.
607: \end{equation}
608: %
609: The distance $r$ is measured between the centers of mass of the respective
610: pairs. Again there are several
611: possible choices for the tangent vectors $\vec{u}_i$.
612: Orientation correlations of the vector connecting the methylenes (C$_1-$C$_5$)
613: extend over several inter-atomic distances (Figure~\ref{fig:atoodf}a). They
614: resemble the packing of model chains consisting of simple
615: beads~\cite{faller99b}. The very few segments that approach closely
616: (cf. Figure~\ref{fig:rdf} and~\ref{fig:partialrdf}) show a perpendicular
617: orientation. A parallel ordering peak is encountered at about 0.4~nm. The
618: first Legendre polynomial ($P_1(r):=\langle\vec{u}_i\cdot\vec{u}_j\rangle$,
619: Figure~\ref{fig:atoodf}b) which carries direction information shows that these
620: inter-chain contacts have a small preference of
621: neighboring chains running in the same direction. At larger distances the
622: explicit atomistic structure is no longer important and the structures become
623: much broader. But there are still structural effects originating from the
624: packing visible up to about 0.7~nm, about two chain diameters.
625:
626: The order of the double bonds between the chains exhibits more atomistic
627: details. As the double bond lead to a planar configuration of the environment,
628: there is parallel orientation of the neighbors especially at short
629: distances ($r\lesssim0.4$~nm) which is visible in the first Legendre
630: polynomial. The bond vectors C$_5-$C$_1$ (between subsequent monomers), which
631: are about the same length as the double bonds, show even less structure than
632: the simple model.
633:
634: The inter-monomer vectors (Figure~\ref{fig:atoodf}c), in contrast, display even
635: more generic features which can also be seen in a simple
636: model~\cite{faller99b,faller00b,mplathe00} (compare
637: Section~\ref{sec:simple}). As they describe larger segments compared to the
638: vectors discussed above, the features are less accentuated. On close contact
639: ($r\lesssim0.4\text{nm}$) an almost perfect perpendicular order is found. At
640: intermediate distances ($0.4\text{nm}<r<0.5\text{nm}$) a preferred parallel
641: alignment, and for the case of the C$_2-$C$_2$ vector a second perpendicular
642: region appears ($r\approx0.8\text{nm}$). The differences between the two
643: different inter-monomer (C$_1-$C$_1$ and C$_2-$C$_2$) vectors are small. So
644: for orientations on length scales as small as monomer sizes, simple models
645: already provide a useful description if the persistence lengths are the same.
646:
647: This shows that the generic packing effects are important for the
648: understanding of the structure of atomistic models. However, the fine
649: structure at short distances, as found in the first Legendre polynomial,
650: cannot be explained by generic arguments as here the detailed chemistry of the
651: polymer is important.
652: %
653: \subsection{Structure of the melt}
654: %
655: Radial distribution functions are quite illustrative in characterizing the
656: structure. However, the experimental observable in scattering experiments is
657: the structure function. The static melt structure factor
658: %
659: \begin{figure}
660: \begin{center}
661: \includegraphics[angle=-90,width=0.4\linewidth]{strfct.eps}
662: \includegraphics[angle=-90,width=0.4\linewidth]{strfctexp.eps}
663: \caption{a) Static structure factor of {\it trans} polyisoprene from this
664: work (solid line: T=300~K, dashed line T=413~K) with the scattering
665: lengths of all atoms taken to be the same in comparison to cis-PI
666: (dotted line, T=413~K, data from ref.~\citen{moe99}). b) Experimental
667: structure factor obtained by neutron scattering at an unknown
668: temperature at molecular weight of 17.200 (data from
669: ref.~\citen{zorn92}).}
670: \label{fig:strfct}
671: \end{center}
672: \end{figure}
673: %
674: \begin{equation}
675: S_{\text{melt}}=
676: \frac{1}{N}\Bigg\langle\left|\sum_{m=1}^{N_{\text{C}}}\sum_{j=1}^{N}
677: \exp(ikr_j^m)\right|^2\Bigg\rangle
678: \end{equation}
679: %
680: is shown in Figure~\ref{fig:strfct} in comparison to simulations of {\it
681: cis}-polyisoprene~\cite{moe99} and experiments on a mixture dominated by the
682: {\it cis}-conformer~\cite{zorn92}. The melt structure factor shows a clear
683: peak at about 15~nm$^{-1}$. In addition, there is some smaller structure,
684: especially higher order peaks. The lower limit of resolution is given by the
685: size of the box corresponding to a minimum
686: $|\vec{k}|$-vector of about 0.4$\pi$~nm$^{-1}$. At higher temperature the
687: overall structure flattens out with less pronounced peaks, as expected.
688:
689: Moe and Ediger performed simulations on pure {\it cis}-polyisoprene at 363~K
690: and 413~K with one chain of 100 monomers~\cite{moe96a,moe99}. This, obviously,
691: reduces the influence of end effects. The densities were much lower than the
692: experimental values (798 vs. 869 and 775 vs. 836~kg/m$^3$, respectively). The
693: structure functions of both simulations are, however, comparable at
694: 413~K. Neither simulation compares really satisfactorily to the experimental
695: structure factor. The double-maximum structure of the first peak is not
696: reproduced, the lower maximum ($\approx$13~nm$^{-1}$) being enhanced, the
697: higher ($\approx18$~nm$^{-1}$) being reduced to a shoulder. The positions of
698: the peaks, however, are in reasonable agreement. As the experimental
699: temperature is
700: not given, one cannot say whether at least part of the discrepancy is a low
701: temperature effect, indicating the formation of a glass, whereas both
702: simulations are performed deep in the melt. Another reason for the
703: disagreement could be the fact that the short chains presented are still
704: oligomers. The alternative would be one long chain, as in refs.~\citen{moe96a}
705: and~\citen{moe99}. However, the periodic boundary replication of a single
706: chain this imposes a strong artificial periodicity to a amorphous melt, which
707: makes it difficult to look at {\it inter}-chain effects.
708: %
709: \section{Dynamical properties}
710: %
711: \subsection{End-to-end vector}
712: %
713: Figure~\ref{fig:end-end} shows for the two polydisperse systems~2 and~3 the
714: reorientation correlation functions (first and second Legendre polynomial)
715: %
716: \begin{eqnarray}
717: P_1(t)&=&\Big\langle(\vec{u}(t)\cdot\vec{u}(0))\Big\rangle\\
718: C_{\text{reor}}(t):=P_2(t)&=&\Big\langle0.5\big[
719: 3(\vec{u}(t)\cdot\vec{u}(0))^2-1\big]\Big\rangle
720: \end{eqnarray}
721: %
722: of the end-to-end vector of the chains of length $N=10$, which is defined as
723: the vector connecting the two terminal carbons C$_1^{\text{mono 1}}$ and
724: C$_5^{\text{mono }n}$. The relaxation time is clearly longer than the time
725: accessible in the simulations. Even system~2, which was simulated for more
726: than 2~ns, did not relax appreciably on this time scale; local vectors relax,
727: of course, much faster. At the higher temperature of 413~K the relaxation time
728: decreases drastically. Still, one has to be cautious discussing length scales
729: of more than a monomer.
730: %
731: \begin{figure}
732: \includegraphics[angle=-90,width=0.5\linewidth]{endendreoratomist.eps}
733: \includegraphics[angle=-90,width=0.5\linewidth]{endendreoratomist413.eps}
734: \caption[Reorientation of atomistic end-to-end vector]
735: {Relaxation of the end-to-end vector. Solid curves are first Legendre
736: polynomials, dashed curves second Legendre polynomials.
737:
738: a) T=300~K: Of both pairs, the upper curve is the correlation
739: function for system~3 and the lower one for system~2.
740:
741: b) For T=413~K (system~3) the relaxation times are much shorter.
742: }
743: \label{fig:end-end}
744: \end{figure}
745: %
746: This figure shows some scatter between the two systems, which may
747: be taken as a rough estimate of the error of the simulations.
748: %
749: \subsection{C$-$H bond reorientation}
750: %
751: Reorientation in polyisoprene melts with 92\% {\it cis}-conformer was measured
752: by the group of Laupr\^etre and Monnerie~\cite{batie89,laupretre93}. Another
753: investigation with a higher {\it trans}
754: content of 22\%~\cite{denault89} focused also on the {\it
755: cis}-conformer. Experimentally, the direct observable is the $T_1$ time.
756: %
757: \begin{equation}
758: \frac{1}{T_{1}}=\frac{\hbar^{2}\gamma_{\opC}^{2}
759: \gamma_{\opH}^{2}}{10r^{2}_{\text{C}-\text{H}}}
760: \Big[J(\omega_{\opH}-\omega_{\opC})+3J(\omega_{\opC})+
761: 6J(\omega_{\opH}+\omega_{\opC})\Big].
762: \end{equation}
763: %
764: The $\gamma_{i}$ are the gyro-magnetic ratios of the respective nuclei and
765: $\omega_{\opC}$ and $\omega_{\opH}$ are the Larmor frequencies, $r$ is the
766: distance between the nuclei. The function $J(\omega)$ is the spectral density,
767: i.e. the Fourier transform of $C_{\text{reor}}$ of the respective C$-$H vector
768: %
769: \begin{equation}
770: J(\omega)=\frac{1}{2}\int_{-\infty}^{\infty}C_{\text{reor}}(t)e^{i\omega t}
771: \text{d}t\;.
772: \end{equation}
773: %
774:
775: In atomistic simulations, $T_{1}$ has been determined for different
776: polymers\cite{moe96a,paul97,antoniadis98}. Moe and Ediger use the limit of
777: {\it extreme narrowing} ($\omega\tau_{\text{reor}}\ll1$) to analyze their {\it
778: cis}-polyisoprene data at $\text{T}=413$~K, as is done in most other atomistic
779: simulations\cite{moe96a}. This has the advantage that $T_{1}$ becomes
780: independent of the Larmor frequencies\cite{kalinowski84}, which
781: cannot be measured in simulations without extrapolation. For long
782: chains in a simple model one sees that extrapolation starting at such high
783: values is very questionable.\cite{faller00b} The spectral density $J(\omega)$
784: is for very short reorientation times independent of $\omega$:
785: %
786: \begin{equation}
787: J(\omega)=B_{\text{local}}^2\frac{2\tau_{\text{reor}}}
788: {1+\omega^2\tau_{\text{reor}}^2}\;.
789: \end{equation}
790: %
791: However, the {\it extreme narrowing} regime is
792: not reached normally by the experiments, as high temperatures are needed to
793: yield correlation times that the results correspond to the limit.
794:
795: The reorientation time $\tau_{\text{reor}}$ is defined by the time integral
796: over the correlation function
797: %
798: \begin{equation}
799: \tau_{\text{reor}}=\int_0^{\infty}C_{\text{reor}}\text{d}t\;,
800: \end{equation}
801: %
802: and in the extreme narrowing limit this is directly linked to the $T_1$ time
803: for a C$-$H vector
804: %
805: \begin{equation}
806: T_1^{-1}=10nK\tau_{\text{reor}}\;,
807: \end{equation}
808: %
809: where $K$ is a constant related to the bond length and $n$ is the number of
810: protons connected to the respective $^{13}$C. Note that a shorter $T_1$
811: corresponds to a longer reorientation time.
812:
813: The C$-$H vector reorientation is followed in the simulations. To minimize
814: chain-end effects only the inner-chain monomers are included in
815: Figure~\ref{fig:CHreor}~b and~c.
816: %
817: \begin{figure}
818: \includegraphics[angle=-90,width=0.5\linewidth]{Catdblnd.eps}
819: \includegraphics[angle=-90,width=0.5\linewidth]{methylenereor.eps}
820: \includegraphics[angle=-90,width=0.5\linewidth]{methylenereor413.eps}
821: \caption{Reorientation of C$-$H vectors in inner monomers of atomistic
822: polyisoprene chains (system~3, chains of length~10,) a) Vinyl C$_2-$H
823: depending on monomer position (1: end monomer, 2: next to end monomer,
824: etc.), b, c) Methylene groups of the central monomer at different
825: temperatures.}
826: \label{fig:CHreor}
827: \end{figure}
828: %
829: Comparing Figures~\ref{fig:dblbndreor}~a and \ref{fig:CHreor}~a, one sees that
830: the hydrogen connected to the backbone at carbon C$_2$ is strongly tied to the
831: backbone. Thus, even such local quantities as bond vectors can be used as
832: observables to monitor the dynamics of intermediate-size chain segments.
833:
834: Figure~\ref{fig:CHreor}~b illustrates that different torsion potentials and the
835: side group have considerable influence on the reorientation of vectors. The
836: reorientations of the two methylene carbons C$_1$ and C$_5$ differ. As C$_5$
837: is vicinal to the methyl
838: group, the steric hindrance for the hydrogens tied to it is more
839: pronounced. Thus, the initial stage of the reorientation, present for
840: the C$_1-$H vector, is considerably smaller for C$_5-$H. This
841: is still visible at 413~K (Figure~\ref{fig:CHreor}~c), although all relaxations
842: are faster. The above-mentioned experiments~\cite{batie89,denault89} on {\it
843: cis}-polyisoprene melts show the same tendenies.
844:
845: Similar to our reorientation correlation function
846: (cf. Figure~\ref{fig:CHreor}) the experimental data~\cite{batie89} provide
847: evidence for a two stage process, the first part being simply exponential
848: followed by a non-exponential long-term stage, described by model correlation
849: function
850: %
851: \begin{equation}
852: C_{\text{reor}}(t) =ae^{-t\tau_0} + (1-a)e^{-t/\tau_2}e^{-t/\tau_1}
853: I_0(t/\tau_1)\;,
854: \label{eq:modelreor1}
855: \end{equation}
856: %
857: with $\tau_0$ the local libration time, $\tau_1$ the time of conformation
858: jumps, and $\tau_2$ connected to damping; $I_0$ is a Bessel function. The
859: separation of time scales for polyisoprene is $\tau_1/\tau_0\ge150$ for the
860: two faster characteristic times~\cite{laupretre93}; the two slow processes
861: ($\tau_1,\,\tau_2$) differ by a factor of 40. A separation of motions was used
862: by Lipari and Szabo as well to analyze NMR data of
863: polymers~\cite{lipari82a,lipari82b}. The correlation function they used has
864: the simpler double exponential shape
865: %
866: \begin{equation}
867: C_{\text{reor}}(t)={\cal S}^2e^{-t/\tau_1}+(1-{\cal S}^2)e^{-t/\tau_2}\;,
868: \label{eq:12}
869: \end{equation}
870: %
871: where the generalized order parameter ${\cal S}$ is related to the parameter
872: $a$ of Laupr\^etre {\it et. al.}~\cite{laupretre93}. The
873: reorientation motion is described by one local and one global reorientation if
874: there is no reptation.
875:
876: Actual numbers for the three times of the model (Eq.~\ref{eq:modelreor1}) are
877: not provided in refs.~\citen{batie89,laupretre93}, but only values for
878: $a$. Although the experimental polymer is mainly {\it cis} the values for $a$
879: for the different vectors are comparable between our simulations and
880: experiment (Table~\ref{tab:avalues}). The simulation data in our case were
881: determined using a fit to Eq.~\ref{eq:12} with a purely exponential form of
882: the second term disregarding the Bessel function ($I_0=1$). The second column
883: in Table~\ref{tab:avalues} shows the values of $a$ estimated from the value of
884: $C_{\text{reor}}$ at 1~ps, which is the shortest time resolved in the
885: simulations ($a_{1\text{ps}}=1-C_{\text{reor}}(1\text{ps})$), coming closer to
886: experimental values. This assumes that the first process is too fast to be
887: resolved here. If the conformational jump time is
888: regarded as the time for torsion rearrangement (see below), and keeping in
889: mind that the two times differ at least by a factor of 150, the guess for
890: $a_{1\text{ps}}$ is probably more realistic.
891: %
892: \begin{table}
893: \[
894: \begin{array}{c*{4}{D{.}{.}{-1}}}
895: \hline
896: \text{vector}
897: & \mul{a_{\text{fit}}^{trans}} & \mul{a_{1\text{ps}}^{trans}} &
898: \mul{a_{\text{sim}}^{cis}}& \mul{a_{\text{exp}}^{cis}}\\
899: \hline
900: & \multicolumn{2}{c}{\text{sim: this work}}
901: & \mul{\text{sim: ref. \citen{moe99}}} &
902: \mul{\text{exp: ref. \citen{batie89}}} \\
903: \hline
904: \operatorname{C}_{1}-\operatorname{H} & 0.75 & 0.42 & 0.28 & 0.40\\
905: \operatorname{C}_{2}-\operatorname{H} & 0.29 & 0.16 & 0.16 & 0.17\\
906: \operatorname{C}_{5}-\operatorname{H} & 0.29 & 0.18 & 0.23 & 0.48\\
907: \hline
908: \end{array}
909: \]
910: \caption{Comparison of the experimental ({\it cis}-PI) and simulation ({\it
911: cis} and {\it trans}-PI, system~2) data for the efficiency of the
912: initial stage of the reorientation process. $a_{\text{fit}}$ originates
913: from an exponential fit of the second stage extrapolated to $t=0$ and
914: $a_{1\text{ps}}$ is the value of $1-C_{\text{reor}}$ at 1ps. In the
915: analysis of the simulations for {\it cis}-polyisoprene a stretched
916: exponential second process was assumed. The experiments used a range of
917: temperature between 283~K and 363~K (ref.~\citen{batie89}). The {\it
918: trans} simulations were at 300~K (this work) and the {\it cis}
919: simulations at 363~K (ref.~\citen{moe99}).}
920: \label{tab:avalues}
921: \end{table}
922: %
923: Still, the simulations underestimate the difference between C$_1$ and C$_2$
924: and overestimate the one between C$_1$ and C$_5$. The data provided here
925: correspond to a sample of pure {\it trans}-polyisoprene
926: oligomers which is presumably quite different from real {\it cis}-polyisoprene
927: with some added {\it trans}-conformer. Moreover, the discrepancy becomes
928: weaker for system~1 (below). Recent simulations on {\it cis}-polyisoprene at
929: higher temperature (T=363~K and T=413~K) were interpreted in terms of a two
930: stage correlation, too. There a separation between the exponential first
931: stage and a stretched exponential second process was deduced~\cite{moe99}. The
932: corresponding $a$-parameters are included in Table~\ref{tab:avalues}. Except
933: for the C$_5-$H vector they are comparable to our data. For
934: C$_5-$H they are even farther from the experimental value.
935:
936: %
937: \begin{figure}
938: \includegraphics[angle=-90,width=0.5\linewidth]{methyshort.eps}
939: \includegraphics[angle=-90,width=0.5\linewidth]{methylong.eps}
940: \caption{a) Short time reorientation of methyl C$-$H vectors, monomer number
941: as in legend (end monomer vs. central monomer,System~3, T=300~K, only
942: 10-mers are included). b) Long time. Clearly, the two stages are
943: separate. A slight difference between the two terminal
944: monomers is visible}
945: \label{fig:methyreor}
946: \end{figure}
947: %
948: In the reorientation of the C$_4-$H vectors of the methyl side group the
949: two-stage reorientation is clearly visible (Figure~\ref{fig:methyreor}~a). The
950: first process is non-exponential on the scale of a few
951: pico-seconds. The monomer index has almost no influence, only the very local
952: surroundings contribute. On this time scale, the vector does not experience the
953: connection to the chain.
954:
955: The long time tail, however, is linked to the overall reorientation. The
956: bonding to the chain leads to a bias in the orientation of the methyl group,
957: which prevents total decorrelation on the short time scale. The second
958: (exponential) process is influenced by chain end effects with decay constants
959: of $\tau_{\text{end}}\approx1~\text{ns}$ and
960: $\tau_{\text{center}}\approx3~\text{ns}$, respectively
961: (Figure~\ref{fig:methyreor}~b). The decay times were determined by an
962: exponential fit of the time region between 500 and 1250~ps.
963:
964: Correlation times for the reorientation of various C$-$H bonds in {\it
965: cis}-polyisoprene melts were calculated by Moe and Ediger, who used one
966: long chain under periodic boundary conditions~\cite{moe96a}.
967: Table~\ref{tab:cmpmoe} compares the results of this work to their data and the
968: extreme narrowing limit of the experiments.
969: %
970: \begin{table}
971: \[
972: \begin{array}{c*{6}{r}*{4}{D{.}{.}{-1}}}
973: \hline
974: \text{Vector} &
975: \multicolumn{3}{c}{\tau_{\text{reor}}^{trans,\text{sys~1}} [\text{ps}]} &
976: \multicolumn{3}{c}{\tau_{\text{reor}}^{trans,\text{sys~2}} [\text{ps}]} &
977: \multicolumn{1}{c}{\tau_{\text{reor}}^{trans,A} [\text{ps}]} &
978: \multicolumn{1}{c}{\tau_{\text{reor}}^{trans,B}[\text{ps}]} &
979: \multicolumn{1}{c}{\tau_{\text{reor}}^{\text{(exp)}} [\text{ps}]} &
980: \multicolumn{1}{c}{\tau_{\text{reor}}^{cis} [\text{ps}]} \\
981: & \multicolumn{3}{c}{300~\text{K}}
982: & \multicolumn{3}{c}{300~\text{K}}
983: & \multicolumn{1}{c}{413~\text{K}}
984: & \multicolumn{1}{c}{413~\text{K}}
985: & \multicolumn{1}{c}{413~\text{K}}
986: & \multicolumn{1}{c}{413~\text{K}}\\
987: \hline
988: \opC=\opC & & 2400 & & & 1900 & & 49 & 51 & - & - \\
989: \opC_1-\opH & & 810 & & & 640 & & 39 & 25 & 22 & 50 \\
990: \opC_2-\opH & & 2400 & & & 1900 & & 50 & 46 & 35 & 75 \\
991: \opC_4-\opH & & - & & & 480 & & 2.1 & 5.8 & 3.6 & 7.6\\
992: \opC_5-\opH & & 1200 & & & 1800 & & 67 & 55 & 26 & 60 \\
993: \hline
994: \end{array}
995: \]
996: \caption{Comparison of simulated reorientation times in {\it
997: trans}-polyisoprene (this work) and {\it cis}-polyisoprene
998: (ref.~\citen{moe96a}) to data determined by extrapolation of
999: experiments at lower temperatures into the extreme narrowing limit.
1000: The {\it cis}-values\cite{moe96a} were determined by numerical
1001: integration of the first 400~ps. For the {\it trans}-polyisoprene only
1002: the two innermost monomers are used. At 413~K the analysis was done
1003: in two ways in order to compare more directly to the Moe and Ediger
1004: simulations. $A$: numerical integration and exponential long-time
1005: tail (see text), $B$: numerical integration to 400~ps. The simulation
1006: errors are estimated to be about 20\% (difference between systems).
1007: The reorientation of the methyl group in the $NVT$ simulation could
1008: not be calculated meaningfully, as there was a force-field problem. (The
1009: hydrogens were connected to the carbon with an additional torsion,
1010: which was too strong.)}
1011: \label{tab:cmpmoe}
1012: \end{table}
1013: %
1014: At 300~K, the correlation times were determined by numerical integration of
1015: $C_{\text{reor}}(t)$ over the first nano-second and an analytical correction
1016: for the exponential tail. For the methyl groups the numerical integration
1017: extended only to 20~ps.
1018:
1019: Heating to 413~K speeds up the simulation. Here the numerical integration was
1020: performed up to 100~ps, again with an analytical correction, except for the
1021: methyl groups (20~ps) and the C$_1-$H (200~ps) (label
1022: $A$ in Table~\ref{tab:cmpmoe}). For the oligomers, however, the results are
1023: very similar to a numerical integration to 400~ps, which was performed
1024: additionally in order to compare directly to the data by Moe and Ediger (label
1025: $B$ in Table~\ref{tab:cmpmoe}).
1026:
1027: The discrepancies between the different systems give a measure of uncertainty
1028: in the results. The reader is reminded that the systems are governed by
1029: slightly differing force-fields and are equilibrated in different manners.
1030:
1031: If the data is compared directly to extrapolated experimental data an overall
1032: discrepancy of about 50\% is found between the {\it trans}-simulations
1033: presented here and the experiments. The integration error in the simulation as
1034: well as the extrapolation of the experiments are sources of error. The systems
1035: are not the same and the simulation model does not reflect reality
1036: perfectly. The experiments themselves are not perfectly reliable. Witt {\it et
1037: al.} showed that NMR experiments for systems as simple as liquid benzene can
1038: result in reorientation times differing by an order of
1039: magnitude~\cite{witt00}.
1040:
1041: In the simulation system~2, the hydrogen at C$_5$ reflects too much how the
1042: backbone reorients on the time scale of the double
1043: bond. Experimentally there is a difference between C$_5$-H and C$_2$-H. This
1044: may result from the interaction with the methyl group which repels the H at
1045: C$_5$ very effectively. In system~1 the difference is more pronounced, as the
1046: non-bonded interactions between the methyl hydrogens and the methylene
1047: hydrogen are switched off ($1-5$ interaction, cf. Section~\ref{sec:ff}).
1048: Comparison with experiment does not allow us to decide which scheme for the
1049: exclusion of nonbonded interactions is more appropriate.
1050:
1051: %
1052: \subsection{Segmental motion}
1053: %
1054: The double bonds reflect the reorientation of local segments.
1055: %
1056: \begin{figure}
1057: \includegraphics[angle=-90,width=0.5\linewidth]{reordblbndafterebonly10.eps}
1058: \includegraphics[angle=-90,width=0.5\linewidth]{dblbndreor1stmono.eps}
1059: \includegraphics[angle=-90,width=0.5\linewidth]{dblbndlonginner.eps}
1060: \caption{Reorientation of double bonds in system~2 at 300~K:
1061: a) Only the chains with ten monomers; results shown for different monomer
1062: distances from chain end. b) Only the first monomer for different chain
1063: lengths. c) Central monomers of the chains of length ten monomers and
1064: more (the first number denotes the chain length, the second the monomer
1065: number) in semilogarithmic representation.}
1066: \label{fig:dblbndreor}
1067: \end{figure}
1068: %
1069: The ends reorient faster than inner
1070: monomers (Figure~\ref{fig:dblbndreor}a); the six innermost monomers are
1071: comparable; chains of ten monomer length have a ``bulk'' inner part.
1072:
1073: There is a two step reorientation. On very short time scales there is a fast
1074: drop due to bond angle vibrations. This is not resolved here. For the
1075: inner monomers, a long-time process on the order of the reorientation of the
1076: whole chain (a few nano-seconds) sets in afterwards.
1077:
1078: There is little difference in the dynamics of chains of different lengths, at
1079: least in the limited range under study here: the end monomers are free to
1080: move regardless of the rest of the chain (Figure~\ref{fig:dblbndreor}~b) and,
1081: for central monomers, the relaxation is the same within the (large) statistical
1082: error (Figure~\ref{fig:dblbndreor}c). As seen in Figure~\ref{fig:distMW} there
1083: are very few chains of any length other than 10 in system~2 and~3.
1084:
1085: Denault {\it et al.} estimated for chains of molecular weights between
1086: $7\times10^3$g/mol and $1.5\times10^5$~g/mol a segmental reorientation time of
1087: 1.0~ns at 303.15~K and of 43~ps at 373.15~K by analyzing their methylene
1088: reorientations~\cite{denault89}. For this, they used the Schaefer
1089: model~\cite{schaefer73} for segmental reorientation, where a
1090: $\chi^2$-distribution of relaxation times is assumed, arising from cooperative
1091: local motion. The values are of the order of magnitude of the reorientation
1092: data presented here for 300~K. However, the chains are clearly longer and the
1093: focus is on the {\it cis} conformer. The chain length dependence is weak. An
1094: Arrhenius plot of the segmental reorientation time of Denault {\it et. al.} in
1095: comparison to the simulated reorientation of the double bond shows a similar
1096: temperature dependence (Figure~\ref{fig:arrhenius}). The activation energies
1097: deduced are $E_A^{\text{(sim)}}\approx33$~kJ/mol and
1098: $E_A^{\text{(exp)}}=65$~kJ/mol at low temperature and
1099: $E_A^{\text{(exp)}}=19$~kJ/mol at high temperature. The effective experimental
1100: activation energy taking
1101: only the lowest and extrapolated highest point into account arrives at
1102: $E_A^{\text{(exp)}}\approx36$~kJ/mol rather close to the simulation value. To
1103: decide if there is a similar behavior as in experiments with two temperature
1104: regimes more simulations at intermediate temperatures would be necessary.
1105: %
1106: \begin{figure}
1107: \[
1108: \includegraphics[angle=-90,width=0.5\linewidth]{arrhenius.eps}
1109: \]
1110: \caption{Arrhenius-plot comparing the C=C reorientation time of
1111: the simulations (filled circles) to the segmental reorientation time
1112: inferred by Denault {\it et al.} from their experiments on {\it cis}-PI at
1113: temperatures between 293~K and 373~K at molecular weights of 7000 to
1114: 130000~\cite{denault89}. For the simulations the values of
1115: table~\ref{tab:cmpmoe} are averaged at 300~K and 413~K respectively. The
1116: solid lines are exponential fits to the curves.}
1117: \label{fig:arrhenius}
1118: \end{figure}
1119: %
1120: \section{Mapping onto a simple model}
1121: %
1122: \label{sec:simple}
1123: %
1124: As atomistic models such as the one presented here are obviously too demanding
1125: in computer time for long chain and/or long time investigations, mapping onto
1126: simpler models is highly desirable. The simple model we choose here is made of
1127: soft spheres (repulsive Lennard-Jones potential) connected by anharmonic
1128: springs and a weak harmonic bond angle potential leading to a persistence
1129: length of 1.5 monomeric units chosen similar to the atomistic model. For
1130: details of the model and results on structure and dynamics, see
1131: refs.~\citen{faller99b,faller00b}.
1132:
1133: We have to choose a mapping of one model onto the other before we can
1134: compare. For the mapping of length scales the natural choice is the mean
1135: end-to-end distance. Thus, melts of 10-mers of the atomistic and the simple
1136: models are simulated, then the lengths are set equal. The atomistic
1137: simulations of system~3b and simulations presented in Ref.~\citen{faller00b}
1138: are used. The convergence of global chain properties in the atomistic case is
1139: shown in Figures~\ref{fig:end-end}b and~\ref{fig:map}. For the mapping of time
1140: scales a dynamic quantity is necessary. We map in Figure~\ref{fig:map} the
1141: center-of-mass mean-square displacements $g_3(t)$ onto each other. Due to
1142: limited simulation times, a corresponding mapping at 300~K could not be
1143: accomplished.
1144: %
1145: \begin{figure}
1146: \[
1147: \includegraphics[angle=-90,width=0.5\linewidth]{CGg3.eps}
1148: \]
1149: \caption{Mean-square displacements of the center-of-mass in the atomistic
1150: simulation at 413~K (dashed line) and in the simple model with persistence
1151: length 1.5 monomer units (solid line). The ordinates are rescaled by the
1152: respective mean-square end-to-end distances. The time scale of the simple
1153: model is adjusted to bring the curves into coincidence for a whole order
1154: of magnitude.}
1155: \label{fig:map}
1156: \end{figure}
1157: %
1158: With this mapping fixed, comparative analyses of the two models can be
1159: employed in order to check whether both models follow the same
1160: dynamics. Figure~\ref{fig:cmp} shows the respective results for the
1161: mean-square-displacement of inner monomers, $g_1(t)$, the first three Rouse
1162: modes and the reorientation of nearest neighbor monomer-connecting
1163: vectors. These vectors are for the simple model just vectors connecting the
1164: beads. For the detailed model, these are vectors connecting the same carbon in
1165: adjacent monomers, i.e. C$_1-$C$_1$ (dotted line) and C$_2-$C$_2$ (dashed
1166: line). For the Rouse mode analysis of the atomistic model the centers-of-mass
1167: of the double bond are taken as monomer positions.
1168: %
1169: \begin{figure}
1170: \includegraphics[angle=-90,width=0.5\linewidth]{CGg1.eps}
1171: \includegraphics[angle=-90,width=0.5\linewidth]{rouse413.eps}
1172: \includegraphics[angle=-90,width=0.5\linewidth]{reor10merscmp.eps}
1173: \caption{Comparative analysis of the simple model with persistence length
1174: $l_p=1.5$ and the atomistic simulation at T=413~K. a) Inner monomer
1175: mean-square displacement (dashed line: atomistic model, center-of-mass of
1176: double bond of central monomer, solid line: simple model), b) Rouse modes
1177: (solid lines: first mode, dashed: second mode, dot-dashed: third mode),
1178: and c) reorientation of nearest neighbor monomer connecting vectors of the
1179: atomistic model at 413~K compared to the simple model.}
1180: \label{fig:cmp}
1181: \end{figure}
1182: %
1183: The first three Rouse modes and the inner-monomer-MSD do not coincide
1184: perfectly within this mapping. Still, they differ by only about a factor of
1185: two to three. The reorientation of the monomer-monomer vector even coincides
1186: without any refinement. The two differently defined vectors in the atomistic
1187: models are almost indistinguishable. Thus, on this scale the atomistic
1188: details are already rather unimportant; they may be incorporated into an
1189: intrinsic chain stiffness. These results show that the local dynamics are not
1190: exactly the same for the atomistic and the simple (coarse-grained) model.
1191: There are striking similarities, however, which are much stronger than one
1192: might expect as the models are completely different. This finding confirms the
1193: concept that the atomistic details on the scale of more than a monomer play
1194: only the role of shaping a persistence length. With the simple model we have
1195: carried out several
1196: investigations\cite{faller99b,faller99d,faller00b,faller00sa}, especially of
1197: local reorientation.
1198:
1199: One might argue, that discrepancies of a factor of three are no small effects.
1200: However, keeping in mind that the two models are drastically different one
1201: would expect on first sight a much stronger discrepancy. We performed the
1202: extreme step from the smallest possible model without quantum effects to a
1203: model where the complete identity of the polymer is only regarded by means of
1204: its stiffness.The successful mapping to the atomistic model is a further
1205: (retrospective) validation of those results. A more detailed investigation of
1206: the mapping will be shown in a separate publication.~\cite{faller00sc}
1207: %
1208: \section{Conclusions}
1209: %
1210: This simulation study is concerned with decamers of {\it trans}-polyisoprene,
1211: which are short in comparison to experimental systems. Yet, the simulations
1212: are able to describe correctly local structural and dynamical features of the
1213: polymer. This can be seen by comparing results like chain extension, structure
1214: functions and correlation times for C$-$H vectors obtained by NMR. Taking into
1215: account that the simulated and experimental systems are not always identical
1216: in composition, temperature and so on, the agreement is quite good. We have
1217: also recently studied the free volume properties of our melts and compared
1218: them to positronium annihilation data\cite{schmitz00s}, and the agreement is
1219: again very good. We thus conclude that our all-atom model provides a faithful
1220: description of this polymer.
1221:
1222: With the model, we analyzed the packing of chains. The detailed analysis of
1223: several inter-chain radial distribution functions shows that monomers approach
1224: each other most strongly with the exposed methyl groups followed by the
1225: methylene groups, whereas the vinyl carbons tend to be less accessible. All
1226: specificity in the interaction is however limited to the first solution
1227: shell. This is also seen in the mutual orientation of tangent vectors of
1228: neighboring chains. Chains have the tendency to be parallel at the first
1229: neighbor distance. The orientational correlation between second neighbors is
1230: already very small. From comparing different choices for the tangent vectors
1231: it is also evident that the orientations of bond vectors (i.e. short vectors)
1232: show some structure arising from atomistic interactions, whereas the
1233: orientation between inter-monomer (long) vectors is less structured and
1234: already close to what is found for a generic bead-spring
1235: model\cite{faller99d,faller99b}.
1236:
1237: The reorientation dynamics of bond (C$-$C and C$-$H) vectors typically follows
1238: a two-stage process, a fast (picosecond) relaxation due to local vibrations
1239: followed by a long-time reorientation characteristic for the reorientation of
1240: the parent polymer segment. The relative contribution of both processes to the
1241: overall reorientation is in good agreement with estimates from NMR
1242: measurements.
1243:
1244: In order to study the non-local structure and dynamics of {\it
1245: trans}-polyisoprene extensive simulations have been performed with a generic
1246: bead-spring model which was only augmented by an intrinsic bending
1247: stiffness\cite{faller00b,faller00sa}. We have successfully mapped this model
1248: onto the present atomistic simulation. After matching length and time scales,
1249: all characteristic time constants then agree to within a factor of $2-3$. This
1250: illustrates that it is possible to develop polymer models at different levels
1251: of detail and have a description of polymer dynamics which smoothly connects
1252: both time scales. Systematic protocols for mapping atomistic to coarse-grained
1253: models and back are therefore being developed in our laboratory also for other
1254: polymer systems\cite{tschoep98a,tschoep98b,meyer00,reith00s}.
1255:
1256: An interesting technical point is the comparison between the polymer samples
1257: equilibrated with end-bridging Monte Carlo (EBMC) and those without. The EBMC
1258: proves to be a useful tool in the initial equilibration of polymer melts as it
1259: offers a way to wander through phase space more efficiently. At a global
1260: structural level, the EBMC provides better equilibrated samples. However, it
1261: is quite surprising that strictly local properties like monomer packing or
1262: bond-vector reorientation do not seem to be affected appreciably by the way
1263: the sample is prepared.
1264: %
1265: \section*{Acknowledgments}
1266: %
1267: We'd like to thank Kurt Kremer and Heiko Schmitz for fruitful
1268: discussions. Financial support from the German ministry of research (BMBF) as
1269: well as the TMR program of the European Union is gratefully acknowledged.
1270: %
1271: \bibliography{standard}
1272: \bibliographystyle{macromolecules}
1273: \end{document}
1274: