1: \documentstyle[psfig]{espart}

2: \setcounter{topnumber}{3}

3: \setcounter{topnumber}{5} 

4: \setcounter{bottomnumber}{5}

5: \setcounter{totalnumber}{5} 

6: \renewcommand{\topfraction}{0.99}

7: \renewcommand{\bottomfraction}{0.99}

8: \renewcommand{\textfraction}{0.01} 

9: \newcounter{refcaption}

10: \newcounter{nofigs}

11: \setcounter{refcaption}{1}

12: \setcounter{nofigs}{0}

13: \begin{document}

14: \begin{frontmatter}

15: 

16: \title{{\small J. Phys. I France, Vol. 5, p. 1431-1455

17: (1995)}\\[1.5cm]

18: Molecular dynamics of arbitrarily shaped granular particles}

19: 

20: \author{Thorsten P\"oschel\thanksref{bylinetp}}

21: 

22: \address{Arbeitsgruppe Nichtlineare Dynamik, Universit\"at Potsdam, Am

23:   Neuen\\ Palais, D--14415 Potsdam, Germany.}

24: 

25: \author{Volkhard Buchholtz\thanksref{bylinevb}}

26: 

27: \address{Humboldt--Universit\"at zu Berlin, Institut f\"ur

28:   Physik, Unter den Linden 6,\\ D--10099 Berlin, Germany}

29: 

30: \thanks[bylinetp]{E-mail: thorsten@hlrsun.hlrz.kfa--juelich.de}

31: \thanks[bylinevb]{E-mail: volkhard@itp02.physik.hu--berlin.de}

32: \date{16 July 1995}

33: 

34: \maketitle

35: 

36: \begin{abstract}

37:   We propose a new model for the description of complex granular

38:   particles and their interaction in molecular dynamics simulations of

39:   granular material in two dimensions. The grains are composed of

40:   triangles which are connected by deformable beams. Particles are

41:   allowed to be convex or concave. We present first results of

42:   simulations using this particle model.

43: \end{abstract}

44: \end{frontmatter}

45: 

46: \section{Introduction}

47: Molecular dynamics has been proven to be a well suited method to

48: investigate the dynamic and static behavior of granular matter. The

49: concept of molecular dynamics was initially used to simulate the

50: dynamics of atoms and molecules, i.e.~of particles without inner

51: degrees of freedom. Alder and Wainwright who belong to the inventors

52: of this method investigated already 1957 numerically the low density

53: approximation of the entropy $S=-N~k_B~\langle \ln~f(p,t)\rangle$ ($f$

54: is the one--particle probability density in momentum space) in a hard

55: sphere system consisting of $N=100$

56: particles~\cite{AlderWainwright:1957ug}. There are applications of

57: molecular dynamics in many fields of interest and molecular dynamics

58: became one of the standard methods in computational physics. Recent

59: molecular dynamics simulations are of very high algorithmic

60: complexity, and systems consisting of up to $10^9$ particles

61: (with simple interaction forces) have been simulated (see

62: e.g.~\cite{BeazleyLomdahl:1993,BeazleyLomdahlJensenTamayo:1994,BeazleyEtAl:1995}). A

63: very interesting overview on the historical development is given in

64: the first part of Hoover's book~\cite{Hoover:1992}.

65: 

66: In molecular dynamics of granular particles, sometimes called

67: ``granular dynamics'', each of the ``molecules'' is a macroscopic body

68: typically with a diameter of the order $D\approx 0.1\dots 1~mm$ which

69: has its own thermodynamic properties, i.e.~it has internal degrees of

70: freedom and hence it can dissipate mechanical energy when it is

71: subjected to outer forces. In some special cases as in simulations of

72: planetary rings~\cite{Salo:1992,Weidenschilling} the diameter can also

73: be of the order $D\approx 1~mm\dots 10~m$. In the most simple case one

74: can express the inelastic nature of the collisions of the grains by

75: introducing restitution coefficients which describe the relative

76: velocities in normal and tangential direction $\vec{v}_{ij}^{\,N}$ and

77: $\vec v_{ij}^{\,T}$ of two granular particles $i$ and $j$ after a

78: collision as functions of the relative velocities before the collision

79: $\vec V_{ij}^N$ and $\vec V_{ij}^T$

80: %\begin{mathletters}

81: \begin{eqnarray}

82:   \vec v_{ij}^{\,N} &=& -\epsilon_N\cdot \vec V_{ij}^N \hspace{0.5cm}

83:   \left( 0\le \epsilon_N \le 1\right) 

84:   \label{resitutionEQa}\\ 

85:    \vec v_{ij}^{\,T} &=&

86:     -\epsilon_T\cdot \vec V_{ij}^T \hspace{0.5cm} \left( -1\le

87:     \epsilon_T \le 1\right) ~.

88:   \label{resitutionEQb}

89: \end{eqnarray}

90: %\end{mathletters}

91: Here the relative velocity is given by

92: \begin{equation}

93:   \vec{v}_{ij} = \dot{\vec{r}}_i -\dot{\vec{r}}_j 

94:   \label{relvelocity1}

95: \end{equation}

96: or, including the particle rotation with angular velocities $\vec

97: \omega_i$ and $\vec \omega_j$ , by

98: \begin{equation}

99:   \vec v_{ij} = \left( \dot{\vec r}_i + \omega_i\times R_i \left(-\vec{n}\right)

100: \right) - \left( \dot{\vec r}_j + \omega_j\times R_j 

101: \vec{n} \right)

102:   \label{relvelocity2}

103: \end{equation}

104: with the unity vectors

105: %\begin{mathletters}

106:   \begin{eqnarray}

107:     \vec{n} &=& \frac{\vec r_i -\vec r_j }{\left| \vec r_i -\vec r_j

108:     \right|}\\ 

109:     \vec{t} &=& \left( \begin{tabular}{ll} 0 & -1 \\ 1 & 0

110: \end{tabular} \right) \cdot \frac{\vec r_i -\vec r_j }{\left| \vec r_i

111:   -\vec r_j \right|}~.

112: %    \label{unityvectors}

113:   \end{eqnarray}

114: %\end{mathletters}

115: Hence the relative velocities at the point of contact are

116: %\begin{mathletters}

117:   \begin{eqnarray}

118:     \vec v_{ij}^{\,N} &=& \vec n \left(\vec v_{ij} \cdot \vec n \right)\\

119:     \vec v_{ij}^{\,T} &=& \vec t \left(\vec v_{ij} \cdot \vec t \right) ~.

120: %    \label{relvelocities}

121:   \end{eqnarray}

122: %\end{mathletters}

123: In general the normal and tangential restitution coefficients $

124: \epsilon_N $ and $ \epsilon_T $ are themselves functions of the

125: relative impact velocities $\epsilon_N=\epsilon_N\left(

126: V_{ij}^N,~V_{ij}^T\right) $ and $\epsilon_T=\epsilon_T\left(

127: V_{ij}^N,~V_{ij}^T\right) $. For the case that the particles are rough

128: homogeneous spheres the coefficients $\epsilon_N$ and $\epsilon_T$ can

129: be calculated

130: analytically~\cite{BrilliantovSpahnHertzschPoeschel:1994} solving the

131: viscoelastic equations of the particles. Using this model one neglects

132: the complicated interaction between particles during the collision.

133: The approximation is justified if collisions are relatively rare

134: events, i.e. if the mean free path between collisions is long and one

135: can restrict the calculation to two--particle--interactions only. If a

136: system is in this regime one calls it a ``granular gas''. Event driven

137: algorithms have been applied in many papers, recently e.g.

138: in~\cite{GoldhirschZanetti:1993,McNamaraYoung:1992} for an

139: investigation of inelastic clustering of granular particles (using

140: constant restitution coefficients) and by Luding et al.~to simulate

141: vertically shaken

142: material~\cite{LudingClementBlumenRajchenbachDuran:1994DISS,LudingClementBlumenRajchenbachDuran:1994STUDIES}.

143: For sophisticated implementations of such algorithms

144: see~\cite{Rapaport:1980,Lubachevsky:1991}.

145: 

146: In many cases of interest, however, the particles collide very often

147: or they touch each other permanently, i.e. the system reveals static

148: properties. Examples for such systems are flows through pipes and

149: hoppers, ball mills, shaken materials, sand heaps and many others. In

150: the model proposed by Cundall and Strack~\cite{CundallStrack:1979} and

151: Haff and Werner~\cite{HaffWerner:1986} the particles feel restoring

152: forces in normal and tangential direction, and they lose kinetic

153: energy during the collisions. If the particles $i$ and $j$ at the

154: positions $\vec{r}_i$ and $\vec{r}_j$ with radiuses $R_i$ and $R_j$

155: touch each other they feel the force

156: %\begin{mathletters}

157: \begin{equation}

158:   \vec{F}_{ij}= F_{ij}^N\, \vec{n} + F_{ij}^T\, \vec{t}

159: \end{equation}

160: with 

161: \begin{eqnarray}

162:   F_{ij}^N &=& Y \cdot \left(R_i+R_j- \left| \vec{r}_i-\vec{r}_j\right| 

163:   \right)^{\frac{3}{2}} - m_{ij}^{eff} \cdot \gamma_N \cdot 

164:   (\dot{\vec{r}}_i - \dot{\vec{r}}_j) \cdot\vec{n}

165: \label{fnormal}\\ 

166: F_{ij}^T &=& \mbox{sign} (v_{ij}^{rel}) \cdot \min\left( m_{ij}^{eff} \gamma_T \left|

167: v_{ij}^{rel}\right| , \mu \left| F_{ij}^{N}\right| \right) 

168: \label{ftang} \\

169: v_{ij}^{rel} &=& (\dot{\vec{r}}_i - \dot{\vec{r}}_j)  \cdot

170: \vec{t}+ R_i \cdot \Omega_i + R_j \cdot \Omega_j 

171: \label{surfvelocity}\\ 

172: m_{ij}^{eff} &=& \frac{m_i \cdot m_j}{m_i + m_j} ~.

173: \label{meff}

174: \end{eqnarray}

175: %\end{mathletters}

176: $Y$ is the Young modulus, $\gamma_N$ and $\gamma_T$ are the damping

177: coefficients in normal and tangential direction and $\mu$ is the

178: Coulomb friction coefficient. Eq.~(\ref{surfvelocity}) describes the

179: relative velocity of the surfaces of the particles at the point of

180: contact and eq.~(\ref{meff}) gives the effective mass.

181: Eq.~(\ref{ftang}) takes the Coulomb friction law into account, saying

182: that two particles slide on top of each other if the shear force

183: overcomes $\mu$ times the normal force. Eq.~(\ref{fnormal}) comes

184: from the Hertz law \cite{Hertz:1882ug} for the force between two rigid

185: spheres which are in contact with each other.

186: 

187: This model has been successfully applied to simulate the behavior of

188: dry granular material in many works describing various physical

189: phenomena. It developed to the standard method for calculations of

190: granular dynamics. Some of these problems are size segregation and

191: convection in vibrated material in two and three dimensions

192: (e.g.~\cite{GallasHerrmannSokolowski:1992PRL,Taguchi:1992PRL,GallasHerrmannPoeschelSokolowski:1994,PoeschelHerrmann:1994}),

193: the flow in hoppers

194: (e.g.~\cite{RistowHerrmann:1994,FormKohringMelinPuhlTillemans:1994,CampbellPotapov:1993})

195: and pipes (e.g.~\cite{Poeschel:1994,Lee:1994}), the flow in rotating

196: cylinders

197: (e.g.~\cite{Ristow:1994,RistowCantelaubeBideau:1994ug,PoeschelBuchholtz:1993CSF}),

198: the motion of granular material on an inclined

199: surface~\cite{Walton:1991,Poeschel:1993JDP}, sound propagation in

200: granular material~\cite{Melin:1994}, the onset of

201: turbulence~\cite{Taguchi:1992JDP,Taguchi:1994} and many others.  Many

202: experimental results of many authors could be reproduced numerically

203: using this type of molecular dynamics simulations. There have been

204: developed very efficient algorithms for the case of short range

205: interaction with large force gradients as it is typical for granular

206: materials,

207: e.g.~\cite{BuchholtzPoeschel:1993JMPC,FormItoKohring:1993,EveraersKremer:1993}.

208: The model by Cundall and Strack yields reliable results in more

209: dynamic systems, i.e. when the static friction of the particles does

210: not play a major role. When the static properties of the material

211: govern the behavior, however, the model might fail.

212: In~\cite{BuchholtzPoeschel:1994PA} is shown that one cannot build up a

213: stable sand heap with a finite inclination of such particles but the

214: heap dissolves under its own mass and the angle becomes smaller with

215: rising number of particles.  Lee~\cite{Lee:1993} simulated static

216: friction effects using spheres which are connected by springs. When

217: two particles touch each other for a certain time a spring ``grows''

218: between these particles and keeps the particles together when forces

219: are applied which would separate them. When the separating forces

220: become too large the springs break and the particles move freely.

221: Using this model Lee reproduced the finite angle of repose in a sand

222: heap, however, the growing and breaking springs have no direct

223: equivalent in nature, and hence the model seems to be quite

224: artificial.

225: 

226: Another simulation method which corresponds to the previous one for

227: the case of very large damping has been introduced by Visscher and

228: Bolsterli~\cite{VisscherBolsterli:1972ug}. In each iteration step all

229: of the grains are moved according to the motion of a vessel containing

230: the granular material. Then the particles are released one by one,

231: beginning with the lowest one (with respect to the direction of

232: gravity). Each particle moves until it reaches the next local minimum

233: of the potential energy when it hits either the wall or other

234: previously released grains. When a particle reached its local minimum

235: it stays in that position, i.e.~a falling

236: particle cannot cause the motion of another previously released

237: particle. This behavior means that inertia of the moving particles is

238: neglected as soon as they hit another particle or the wall, and hence

239: it corresponds to the case of very large damping. The advantage of

240: this algorithm is its numerical simplicity, i.e.~it is possible to

241: investigate much larger systems than using molecular dynamics.

242: Recently it has been applied in simulations of several problems as

243: size

244: segregation~\cite{JullienMeakinPawlovitch:1992}

245: and the motion of particles in a rotating

246: cylinder~\cite{BaumannJobsWolf:1993,BaumannJanosiWolf:1994}. Obviously

247: the simplification of infinite damping is not valid in each case, and

248: hence one has to carefully investigate whether it is justified to apply

249: this algorithm (see

250: e.g.~\cite{BarkerMehta:1995}).

251: 

252: There are some more simulation methods for the dynamics of granular

253: material which we want to mention only: Peng and Herrmann applied a

254: Lattice Gas Automaton to investigate density waves in a

255: pipe~\cite{PengHerrmann:1994}. Caram and Hong investigated granular 

256: flows using a Random Walk technique~\cite{CaramHong:1991}.

257: 

258: 

259: There are some models to simulate complex shaped grains which are

260: composed of spheres. In the model by Gallas and Soko{\l}owski

261: \cite{GallasSokolowski:1993} the grains consist of two spheres

262: connected with each other by a stiff bar. Walton and Braun applied

263: more complicated particles consisting of four or eight spheres rigidly

264: connected with each other~\cite{WaltonBraun:1993}. Using this model

265: they examined the transition from stationary to sliding flow and the

266: transition from sliding to raining flow in a rotating drum. P\"oschel

267: and Buchholtz~\cite{PoeschelBuchholtz:1993,BuchholtzPoeschel:1994PA}

268: describe grains built up of five spheres where one of them is located

269: in the center of the grain and four identical spheres are at the

270: corners of a square. Each pair of neighboring spheres is connected by

271: a damped spring. The latter model was applied to the rotating cylinder

272: \cite{PoeschelBuchholtz:1993CSF} and it was shown that the simulation

273: results agree much better with the experiment than equivalent

274: calculations with spheres. Especially it was shown that one can

275: reproduce stick--slip motion and avalanches which was not possible in

276: simulations using spheres. The inclination of the surface of the

277: material in the rotating cylinder and the dependency of the

278: inclination on the angular velocity, however, did not agree well with

279: the experiment.

280: 

281: Mustoe and DePorter~\cite{MustoeDePorter:1993} proposed a particle

282: model where the boundary geometry of the particles is defined (in

283: local coordinates) by

284: \begin{equation}

285: f_i\left( x_i,y_i\right) =\left[\frac{\left| x_i \right|}{a_i}\right]^{n_i} + 

286:              \left[\frac{\left| y_i \right|}{b_i}\right]^{n_i} - 1 = 0.

287:  \label{hogue.eq}

288: \end{equation}

289: Varying $n_i$ between 2 and $\infty$ the shape of the $i$th particles

290: varies continuously from elliptical to rectangular. To detect contacts

291: of pairs of particles described by eq.~(\ref{hogue.eq}) one has to

292: solve numerically equations of high order for each pair of possibly

293: touching particles in each iteration step. This requires an iteration

294: technique which converges slowly for higher exponents $n_i$ and which

295: makes hence the algorithm numerically complicated. Hogue and

296: Newland~\cite{HogueNewland:1993} investigated the flow of granular

297: material on an inclined chute and through a hopper using a model where

298: the boundaries of the convex particles are given by polygons with up

299: to 24 vertices each. To detect whether two particles touch each other

300: one has to calculate the intersections between each pair of vertices.

301: During collisions energy is dissipated according to Stronge's energy

302: dissipation hypothesis~\cite{Stronge:1990} in normal direction, whilst

303: Coulomb's friction law models the energy losses also in tangential

304: direction.

305: 

306: Using grains which consist of interconnected spheres or of particles

307: described by eq.~(\ref{hogue.eq}) it is not possible to simulate

308: particles with sharply formed corners. For some effects it seems to be

309: essential to simulate such particles to reproduce the experimental

310: observed effects. This point is discussed in detail

311: in~\cite{BuchholtzPoeschel:1994PA}. Tillemans and

312: Herrmann~\cite{TillemansHerrmann:1994,Tillemans:1994} proposed a two

313: dimensional particle model where the particles are convex polygons.

314: When two particles $i$ and $j$ touch each other, i.e. when there is an

315: overlap, there acts the force

316: %\begin{mathletters}

317: \begin{eqnarray}

318:   \vec{F}_{ij}&=&F_{ij}^N~\vec{e}^{\,N} + F_{ij}^T~\vec{e}^{\,T}\\

319:   F_{ij}^N &=& \frac{Y~A}{L_c} -

320:   m_{ij}^{eff}~\gamma_N~\left( \vec{v}^{\,rel}\cdot \vec{e}^{\,N} \right) \\ 

321:   F_{ij}^T &=& - \min\left( m_{ij}^{eff} \gamma_{T}

322:   \left|\vec{v}_{rel} \cdot \vec{e}^{\,T} \right|, \mu \left|

323:     F_{ij}^{N} \right|\right)~.

324: %  \label{forcetillemans}

325: \end{eqnarray}

326: %\end{mathletters}

327: $A$ is the compression area of the particles

328: (overlap area), Y is the Young module of the material.

329: The effective mass $m_{ij}^{eff}$ is given by eq.~(\ref{meff}), the

330: relative particle velocity is $\vec{v}^{\,rel}= \dot{\vec{r}}_i

331: -\dot{\vec{r}}_j $, where $\vec{r}_i$ points to the center of mass

332: of the particle $i$, and $L_c$ is the characteristic size of the

333: particle. The unit vectors are in the direction of the line that

334: connects the intersection points of the overlapping particles $i$ and

335: $j$ ($\vec{e}^{\,T}$) and perpendicular to this line

336: ($\vec{e}^{\,N}$). Only convex particles are allowed. This model was

337: used e.g.~in simulations of shear cells, earth quakes and flow through

338: hoppers~\cite{Tillemans:1994}. The results differ significantly from

339: similar calculations using spheres. For the case of the hopper

340: simulation they found clogging and arching which could not be found in

341: simulations with spheres. A similar model was used earlier by

342: Handley~\cite{Handley:1993} who investigated the fracture behavior of

343: brittle granular material. Potapov et

344: al.~\cite{PotapovCampbellHopkins:1994aug,PotapovCampbellHopkins:1994bug}

345: investigated a similar model for solid fracture. Initially they

346: subdivide a macroscopic two dimensional body into many small

347: equilateral triangles (elements). The forces in the body are resolved

348: as forces at inter--element contacts of two different types which they

349: call ``glued'' and ``collisional''. Glued contacts are the joints

350: between elements interior to a solid body which can support stresses.

351: When the tensile stresses exceed a certain value, cracks form and the

352: glued bond breaks. Collisional contacts are contacts between the

353: surface elements of the body (which eventually collides with other

354: bodies or walls), and contacts between triangles in the bulk of the

355: body where glued contacts have been broken. In the three latter

356: models~\cite{TillemansHerrmann:1994,Tillemans:1994,Handley:1993,PotapovCampbellHopkins:1994aug,PotapovCampbellHopkins:1994bug}

357: the convex grains have been Voronoi polyhedra in two

358: dimensions~\cite{Finney:1979}.

359: 

360: In this paper we present a new particle model for molecular dynamics

361: of granular material in two dimensions where the particles are

362: simulated using a similar model which has some advantages. In the

363: following section we describe the model, the forces acting between the

364: grains are derived in sections~\ref{trianglesSEC} and

365: \ref{stressesinbeamsSEC}. Some aspects of the implementation and the

366: performance of the algorithm are discussed in

367: section~\ref{implementationSEC} and some sample results based on this

368: particle interaction model will be briefly discussed in

369: section~\ref{resultsSEC}.

370: 

371: \section{The model}

372: In our model the grains are composed of an arbitrary number of

373: ideal elastic triangles which are connected by beams (see

374: fig.~\ref{grainsFIG}). 

375: \begin{figure}[ht]

376:   \ifcase \value{nofigs}

377:   \begin{center}

378:     \begin{minipage}{4cm}

379:       \psfig{figure=figs/grains.eps,width=4cm,angle=270}

380:     \end{minipage}

381:     \hspace{2cm}

382:     \begin{minipage}{2cm}

383:       \psfig{figure=figs/grain5.eps,width=2cm,angle=270} 

384:       \vspace{0.5cm}

385:       

386:       \psfig{figure=figs/grainstern.eps,width=2cm,angle=270}

387:     \end{minipage}

388:   \end{center}

389:   \vspace{0.5cm}

390:   

391:   \fi

392:   \caption{Examples of grains composed of different numbers of triangles. 

393:     The model is not restricted to convex grains.}

394:   \label{grainsFIG}

395: \end{figure}

396: 

397: A beam in our sense is an deformable damped bar

398: which is subjected to forces in the direction of its axis and

399: transverse to its axis, i.e.~to normal and shear forces, and to

400: moments acting on its ends. Triangles which belong to the same grain

401: do not interact with each other. When two triangles of different

402: grains collide, i.e.~if there is an overlap between both, they feel a

403: restoring force. Hence the force acting upon the triangle $i$

404: belonging to the grain $j$ ($j\in

405: \{1,2,\dots,N\}$) is

406: \begin{equation}

407: \vec{F}_i^{\,j} = \sum_{l=1,l\neq j}^N 

408:       \sum_{k=1}^{n_t(l)} \vec{\Gamma}_{ik}^{jl} 

409:    + \sum_{k=1}^{n_b^j(i)} \vec\Lambda_{ik}^{\,j} 

410:    + \vec\Phi(\vec{r}_{i}^{\,j}, \dot{\vec{r}}_i^{\,j})~.

411:   \label{totalforceEQ}

412: \end{equation}

413: The first sum in the first term in eq.~(\ref{totalforceEQ}) runs over

414: all grains $l$ except the $j$th, the second sum runs over all

415: triangles $k\in\{1,2,\dots,n_t(l)\}$, the $l$th grain consists of.

416: $\vec{\Gamma}_{ik}^{jl}$ is the force which acts between the triangles

417: $i$ and $k$ which belong to different grains $j$ and $l$ (see section~\ref{trianglesSEC}). The sum in

418: the second term runs over all beams connecting the triangle $i$ of the

419: grain $j$ with other triangles of the same grain.

420: $\vec\Lambda_{ik}^{\,j}$ is the force induced by the beam $k$ acting

421: on the $i$th triangle of the grain $j$. They originate from the

422: distortion of the beam that connects the triangle $i$ with another one

423: of the same grain. The acting forces and momenta will be discussed

424: below in section~\ref{stressesinbeamsSEC}. A similar model for the

425: beam forces has been introduced earlier by Herrmann et al. for the

426: fracture of disordered lattices~\cite{HerrmannHansenRoux:1989}.  The

427: last term $\vec\Phi(\vec{r}_{i}^{\,j}, \dot{\vec{r}}_i^{\,\,j})$

428: describes the action of an external force, as e.g.~gravity, on the

429: triangles.

430: 

431: During its deformation a beam dissipates energy similar to a linearly

432: damped spring, i.e.~proportional to its deformation rate~(see

433: section~\ref{stressesinbeamsSEC}). To avoid time consuming

434: calculations due to Steiner's law each beam is fixed at both ends in

435: the center of mass points of the triangles it connects. When the beam

436: is not deformed by eventually applied forces (the beam is at rest) it

437: is perpendicular to the neighboring edges of the triangles which it

438: connects.  Fig.~\ref{grainsFIG} shows some examples of grains. The

439: only restriction concerning the number, the shape and the position of

440: the triangles is caused by the condition that the beams are fixed in

441: the center of mass.

442: 

443: The next section~\ref{trianglesSEC} gives a detailed description of the

444: force that acts when triangles collide, in

445: section~\ref{stressesinbeamsSEC} the forces due to deformed beams are

446: derived.

447: 

448: \section{The interaction of the triangles}

449: \label{trianglesSEC}

450: In this section we describe the calculation of the term

451: $\vec{\Gamma}_{ik}^{jl}$ in eq.~(\ref{totalforceEQ}) originating from

452: the compression of two triangles $i$ and $k$ which belong to the

453: grains $j$ and $l$, respectively. When the triangles are compressed

454: there exists a virtual ``overlap area'' which leads to an elastic restoring

455: force $\vec{\Gamma}_{ik}$. For simplicity here and in the following we

456: drop the upper indexes of the variables. Areas of triangles $XYZ$ and

457: quadrangles $WXYZ$ we denote by $\Delta(XYZ)$ and $\Box(WXYZ)$.

458: Fig.~\ref{standard.fig} shows the most frequently found type of a

459: collision between the triangles $A_iB_iC_i$ and $A_kB_kC_k$, i.e.~a

460: corner of one triangle deforms a side of another one. 

461: 

462: \begin{figure}[ht]

463:   \ifcase \value{nofigs}

464: \centerline{\psfig{figure=/figs/standard.eps,width=6cm,angle=270}}

465: \vspace{0.5cm}

466: \fi

467: \caption{The ``standard'' type of a collision 

468:   between two triangles $A_iB_iC_i$ and $A_kB_kC_k$. The forces

469:   $\pm\vec{\Gamma}_{ik}$ are directed perpendicular to the

470:   intersection line $\overline{S_1S_2}$.  Moments act with respect to

471:   the center of mass points $G_i$ and $G_j$.  The absolute value of

472:   the interaction force $\left|\vec\Gamma_{ik}\right|$ is proportional

473:   to the shadowed intersection area $\Delta \left(S_1S_2A_k\right)$.}

474: \label{standard.fig}

475: \end{figure}

476: 

477: The absolute

478: value of the restoring force $\vec{\Gamma}_{ik}$ is given by the

479: shadowed area $\Delta\left(S_1S_2A_k\right)$ of the triangle

480: $S_1S_2A_k$ times the Young module $Y$. Its direction is

481: perpendicular to the intersection line $\overline{S_1S_2}$ (Poisson

482: hypothesis, see e.g.~\cite{FormKohringMelinPuhlTillemans:1994}). Hence

483: resulting momenta acting upon the triangles $A_iB_iC_i$ and

484: $A_kB_kC_k$ read

485: %\begin{mathletters}

486: \begin{eqnarray}

487: \vec{M}\left(A_iB_iC_i\right) &=& \vec{HG_i} \times \vec{\Gamma}_{ik} \label{momentaA}

488: \\

489: \vec{M}\left(A_kB_kC_k\right) &=& \vec{HG_k} \times \vec{\Gamma}_{ki} = 

490:   -\vec{HG_k}\times \vec{\Gamma}_{ik}~.

491: \label{momentaB}

492: \end{eqnarray}

493: %\end{mathletters}

494: The vector $\vec{HG_i}$ points from the middle point of the line

495: $\overline{S_1S_2}$ to the center of mass of the triangle $A_iB_iC_i$.

496: 

497: Although the case shown in fig.~\ref{standard.fig} is the mostly

498: occurring type of collisions there are several other types which

499: treatment has to be discussed below.

500: 

501: Classifying the possible interactions of the triangles we define five

502: different types of collisions which are slightly different from those

503: in~\cite{PotapovCampbellHopkins:1994aug}. The detection of overlapping

504: polygons is a very important problem in computer graphics too, where

505: one frequently has to decide whether two objects cover each other, and

506: which of them is in the foreground or in the background, respectively.

507: Therefore we found it helpful and inspiring first to have a look into

508: advanced methods in computer graphics,

509: e.g.~\cite{BentleyOttmann:1979ug,PreparataShamos:1985ug}. In all

510: figures~\ref{standard.fig}--\ref{schnit5} the overlapping area is

511: drawn extremely exaggerated. Typically the overlapping area of two

512: interacting triangles does not exceed a few tenth of a percent of the area

513: of the triangles.

514: 

515: \begin{enumerate}

516: \begin{figure}[ht]

517:   \ifcase \value{nofigs}

518:   \centerline{\psfig{figure=/figs/schnitt1a.eps,width=6cm,angle=270}\psfig{figure=/figs/schnitt1b.eps,width=6cm,angle=270}}

519:   \vspace{0.5cm}

520: \fi

521: \caption{The first type of collisions: There are two intersection 

522: points $S_1$ and $S_2$. Both lie at the same edge of one of the 

523: triangles.}

524: \label{schnit1}

525: \end{figure}

526: 

527: \item{For the first type (fig.~\ref{schnit1}) there are two

528:     intersection points $S_1$ and $S_2$ lying at the same edge of one

529:     of the interacting triangles. One of the situations drawn in

530:     fig.~\ref{schnit1} corresponds to the standard collision

531:     (fig.~\ref{standard.fig}). We calculate the area $\Delta \left( S_1

532:       S_2 A_i\right)$ (shadowed area) given by the intersection points

533:     $S_1$ and $S_2$ and the included point $A_i$. Because the

534:     overlapping area is definitely smaller than half the area of the

535:     triangles the force is proportional to the minimum between $\Delta

536:     \left( S_1 S_2 A_i\right)$ and $\Delta \left( A_i B_i C_i\right) -

537:     \Delta \left( S_1 S_2 A_i\right)$

538: \begin{equation}

539:   \left|\vec{\Gamma}_{ik}\right| = \Gamma_{ik} = 

540:   Y \cdot \min \left\{\Delta \left( S_1 S_2 A_i\right) , \Delta \left(

541:   A_i B_i C_i\right) - \Delta \left( S_1 S_2 A_i\right) \right\}~.

542: \end{equation}

543: The force acts perpendicular to the line between the intersection

544: points $\overline{S_1 S_2}$.}

545: 

546: \item The second type (fig.~\ref{schnit2}) is characterised by two

547: intersection points lying on different edges for both triangles. We

548: calculate the areas $\Delta \left( S_1 S_2 A_i\right) $ (dark gray)

549: and $\Delta \left( S_1 S_2 A_k\right) $ (light gray). The force is

550: proportional to the overlap

551: \begin{eqnarray}

552: \left|\vec{\Gamma}_{ik}\right| = \Gamma_{ik} = Y \cdot \left[

553: \min \left\{\Delta \left( S_1 S_2 A_i\right) , \Delta 

554: \left( A_i B_i C_i\right) - \Delta \left( S_1 S_2 A_i\right) 

555: \right\} + \right.\nonumber \\

556: \left. \min \{\Delta \left( S_1 S_2 A_k\right) , \Delta 

557: \left( A_k B_k C_k\right) - \Delta \left( S_1 S_2 A_k\right) \}

558: \right]~.

559: \end{eqnarray}

560: It acts perpendicular to the line between the intersection points

561: as in the previous case.

562: \begin{figure}[ht]

563:   \ifcase \value{nofigs}

564: \centerline{\psfig{figure=/figs/schnitt2a.eps,width=6cm,angle=270}\hspace{0.5cm}\psfig{figure=/figs/schnitt2b.eps,width=6cm,angle=270}}

565: \vspace{0.5cm}

566: \fi

567: \caption{The second type of collisions: There are two intersection 

568: points $S_1$ and $S_2$ lying on different edges for both triangles.}

569: \label{schnit2}

570: \end{figure}

571: 

572: 

573: \item {The third type (fig.~\ref{schnit3}) is characterised by four

574:     intersection points, each pair of two points lies on one edge of

575:     both triangles. Therefore the third edge of both interacting

576:     triangles does not intersect the edges of the other triangle.

577:     There are two pairs of forces proportional to the quadrangular

578:     area $\Box \left( S_1 S_2 S_3 S_4\right) $ of the overlap

579: \begin{equation}

580: \left|\vec{\Gamma}_{ik}\right| = \Gamma_{ik}  = 

581: \frac{1}{2} \cdot Y \cdot \Box \left( S_1 S_2 S_3 S_4\right)~, 

582: \end{equation}

583: one acting perpendicular to the line $\overline{S_1 S_2}$ and the

584: other perpendicular to the line $\overline{S_1 S_4}$. The pre--factor

585: $0.5$ is due to two pairs of acting forces instead of one for the first two

586: collision types.}

587: \begin{figure}[ht]

588:   \ifcase \value{nofigs}

589: \centerline{\psfig{figure=/figs/schnitt3.eps,width=6cm,angle=270}}

590: \vspace{0.5cm}

591: \fi

592: \caption{The third type of collisions: There are four intersection 

593: points $S_1$, $S_2$, $S_3$ and $S_4$. Each pair of points lies on 

594: one edge of either the first or the second triangle.}

595: \label{schnit3}

596: \end{figure}

597: 

598: \item{The fourth type (fig.~\ref{schnit4}) is characterised by four

599:     intersection points too, but here one of the triangles has

600:     intersection points at all its edges. Because interactions of this

601:     type occur extremely seldomly we did not implement an exact

602:     calculation for this case, but we calculate two interactions of

603:     type $1$ with the intersection points $S_1$--$S_2$ or $S_3$--$S_4$

604:     respectively, instead of solving the correct problem. In real

605:     simulation the fraction of this type occurs approximately

606:     $10^{-5}$ of the number of collisions.  Nevertheless one has to

607:     deal with these extremely seldom events since otherwise they might

608:     cause problems from which the system cannot recover.  If one does

609:     not care about these events usually the system gains in a few time

610:     steps after the event occurs a huge amount of kinetic energy,

611:     i.e.~the system explodes.}

612: \begin{figure}[ht]

613:   \ifcase \value{nofigs}

614: \centerline{\psfig{figure=/figs/schnitt4.eps,width=6cm,angle=270}}

615: \vspace{0.5cm}

616: \fi

617: \caption{The fourth type of collisions: There are four intersection 

618: points $S_1$, $S_2$, $S_3$ and $S_4$. One of the triangles intersects 

619: with all its three edges.}

620: \label{schnit4}

621: \end{figure}

622: 

623: \item{The fifth and last type of interaction (fig.~\ref{schnit5}) is

624: characterised by six intersection points $S_1$ -- $S_6$. Equivalent to

625: the forth type we do not calculate the exact interaction but

626: substitute this calculation by solving three interactions of the first

627: type with pairs of interaction points $S_1$--$S_2$, $S_3$--$S_4$ and

628: $S_5$--$S_6$}.

629: \end{enumerate}

630: According to the forces $\vec{\Gamma}_{ik}$ we derive forces in

631: parallel to the axes of the Cartesian coordinate system and moments

632: $\vec{M}$ acting with respect to the center of mass points $G_i$ and

633: $G_k$ as described in equations~(\ref{momentaA}, \ref{momentaB}).

634: \begin{figure}[ht]

635:   \ifcase \value{nofigs}

636: \centerline{\psfig{figure=/figs/schnitt5.eps,width=6cm,angle=270}}

637: \fi

638: \caption{The fifth type of collisions: There are six intersection

639: points $S_1$,\dots,$S_6$.\hspace{2cm}}

640: \label{schnit5}

641: \end{figure}

642: 

643: \section{Stresses in beams}

644: \label{stressesinbeamsSEC}

645: When torques and forces are applied to a beam the beam deforms. Since

646: all deformations are assumed to be small compared with the size of the

647: beam one can superpose the deformations originating from different

648: applying forces and

649: torques~\cite{TimoshenkoGere:1972,TimoshenkoLessells:1928}.

650: Fig.~\ref{bending.fig} shows an (infinitesimally) deformed beam with

651: radius of curvature $\rho$. We find the approximation

652: \begin{equation}

653: \tan\left( \frac{d\Theta}{2}\right) =\frac{1}{2 \cdot \rho}\cdot

654: dx \hspace{0.5cm}\mbox{or}\hspace{0.5cm} 

655: \rho=\frac{dx}{d\Theta}

656: \end{equation}

657: with $\tan(d\Theta) = d\Theta$.

658: \begin{figure}[ht]

659:   \ifcase \value{nofigs}

660: \centerline{\psfig{figure=/figs/bending.eps,width=8cm,angle=270}}

661: \vspace{0.5cm}

662: \fi

663: \caption{An infinitesimally deformed beam with radius of curvature 

664: $\rho $. The fibres below or above the neutral fibre $\overline{SS}$ 

665: are in tension or in compression. For a beam with quadratic cross 

666: section $b^2$ the deformation leads to the resulting moment 

667: $M=\frac{E\,b^4}{12\,\rho}$.}

668: \label{bending.fig}

669: \end{figure}

670: 

671: The dashed line $\overline{ss}$ is the neutral surface length of which

672: does not change during the deformation. All other fibres below and

673: above the neutral surface are in tension or in compression,

674: respectively. The length of a fibre in distance $y$ from the neutral

675: surface is $\left( \rho+y\right) d\Theta=\left( 1+y/\rho\right) dx$

676: and hence the strain reads $\epsilon=y/\rho$. With Hooke's law

677: $\sigma=E\cdot\epsilon$ one finds $\sigma=Ey/\rho$. Finally the

678: resulting momentum $M$ is

679: \begin{equation}

680: M= \int_A\sigma y dA = \frac{EI}{\rho}

681: \label{M=EI/rho}

682: \end{equation}

683: where the moment of inertia $I=\int y^2 dA$ depends on the geometrical

684: shape of the cross section of the beam. In the present paper we assume

685: that the beams have quadratic cross section of width $b$ and hence

686: $I=b^4/12$. When we express the radius of curvature $\rho$ by

687: \begin{equation}

688: \frac{1}{\rho} = \frac{d^2\,v}{dx^2}~,

689:   \label{curvature}

690: \end{equation}

691: where $v$ is the deflection of the beam from its initial position

692: (with the approximation $\tan{\Theta\approx \Theta}$).

693: From eqs.~(\ref{M=EI/rho}) and (\ref{curvature}) one finds the basic

694: equation for the deflection of a beam:

695: \begin{equation}

696: \frac{d^2\,v}{dx^2} = v'' = +\frac{M}{E\,I}

697:   \label{basicequation}

698: \end{equation}

699: 

700: This equation has to be solved to find the forces and moments which

701: act when a beam is deformed. The integration constants are used to

702: satisfy the boundary conditions.

703: 

704: There are three different deformation modes for beams: elongation,

705: shearing and bending. In the following three subsections

706: \ref{elongation.subs}, \ref{shearing.subs} and \ref{bending.subs} we

707: will discuss the deformation rules in detail.

708: 

709: \subsection{Elongation of beams}

710: \label{elongation.subs}

711: Since the exact notation of the acting forces and moments as vector

712: functions of the coordinates of the triangles the beam connects, is

713: not very instructive but confusing due to the length of the formulae

714: expressions we discuss for simplicity of notation the deformation of

715: the beams here and in the following in local coordinates. Instead of

716: providing the exact vector notation we rather discuss absolute values.

717: In all cases the directions in which the forces and moments act are

718: very clear, moreover they are given for each case explicitly in the

719: figures~\ref{fig:beamatrest}--\ref{beamdoppelt.fig} and in

720: figure~\ref{shearbend.fig}. For the implementation of the algorithm,

721: however, one should note that it is necessary to transform the given

722: expressions for the forces and momenta into the coordinate system of

723: the triangles. These transformations require a considerable part of

724: the computation time.

725: \begin{figure}[htbp]

726:   \ifcase \value{nofigs}

727: \centerline{\psfig{figure=/figs/beamatrest.eps,width=8cm,angle=270}}

728: \vspace{0.5cm}

729: \fi

730:     \caption{The shape and the location of a beam at rest. 

731: Its length is assumed to be $L$, its ends $A$ and $B$ lie 

732: on the $x$--axis. When the beam is not deformed there do 

733: not act any forces $F_A$, $F_B$ or moments $M_A$, $M_B$. 

734: In the following 

735: \ifcase \value{refcaption}

736: \ref{bending.fig}--\ref{beamdoppelt.fig} 

737: \or

738: figures~8-11

739: \fi

740: and in 

741: \ifcase \value{refcaption}

742: \ref{shearbend.fig} 

743: \or

744: fig.~14

745: \fi

746: the $x$--axis is drawn as a thin horizontal line.}

747: 

748: \label{fig:beamatrest}

749:     

750: \end{figure}

751: 

752: Fig.~\ref{fig:beamatrest} displays the shape and the location of a

753: beam of length $L$ at rest, i.e.~when no deforming forces $F_A$,

754: $F_B$ or moments $M_A$, $M_B$ at the ends $A$ and $B$ of the beam

755: apply. The $x$--axis is drawn with a thin solid line, i.e.~at rest the

756: beam lies on the $x$--axis.

757: 

758: For the elongation deformation we find according to the

759: linear Hooke law the restoring force

760: \begin{equation}

761: \left| F^{el} \right| = \left| \Delta L \right| \cdot E .

762: \end{equation}

763: 

764: \subsection{Shearing}

765: \label{shearing.subs}

766: We want to calculate the moments and forces $M_A$, $M_B$, $F_A$ and

767: $F_B$ of a sheared beam as drawn in Fig.~\ref{beamscherung.fig}.  For

768: a beam of length $L/2$ which is fixed at one end (at $x=0$) and where

769: acts the shear force $F$ (in negative $y$--direction) at the other end

770: (at $x=L/2$) we find for the local moment

771: $M(x)=F\,\left(\frac{L}{2} - x \right)$ and hence

772: eq.~(\ref{basicequation}) reads

773: \begin{equation}

774: v''=\frac{F}{E\,I} \left(\frac{L}{2}-x\right)

775:   \label{shearingeq}

776: \end{equation}

777: with the boundary conditions $v(0)=0$ and $v'(0)=0$.

778: We find for the vertical deviation $\Delta/2$ at the point $x=L/2$

779: \begin{equation}

780: \Delta/2= \frac{FL^3}{24\,EI}

781: \end{equation}

782: force which acts at the free end of the beam.

783: \begin{figure}[ht]

784:   \ifcase \value{nofigs}

785: \centerline{\psfig{figure=/figs/beamscherung.eps,width=8cm,angle=270}}

786: \vspace{0.5cm}

787: \fi

788: \caption{When the beam in 

789: \ifcase \value{refcaption}

790: fig.~\ref{fig:beamatrest} undergoes a shear deformation 

791: $\Delta$ there act the forces $F_A$ and $F_B$ and the moments 

792: $M_A$ and $M_B$. The forces and moments are given in 

793: eqs.~(\ref{forcesshearing.eq}, \ref{momshearing.eq}).

794: %\ref{eq.shearing}).

795: \or 

796: fig.~9 undergoes a shear deformation $\Delta$ there act 

797: the forces $F_A$ and $F_B$ and the moments $M_A$ and $M_B$. 

798: The forces and moments are given in eqs.~(22).

799: \fi

800: }

801: \label{beamscherung.fig}

802: \end{figure}

803: 

804: One can assume that the sheared beam in Fig.~\ref{beamscherung.fig} is

805: built up of two of such beams of length $L/2$ as described above. 

806: Finally one finds

807: \begin{equation}

808: \Delta = \frac{F\,L^3}{12\,EI}

809:   \label{delta}

810: \end{equation}

811: and hence

812: %\begin{mathletters}

813: \begin{eqnarray}

814: F_A &=& -F_B = F = \frac{12 E I}{L^3}\Delta\label{forcesshearing.eq}\\

815: M_A &=& M_B = F \cdot \frac{L}{2} = \frac{6 E I}{L^2} \Delta~.

816: \label{momshearing.eq} 

817: \end{eqnarray}

818: %\end{mathletters}

819: 

820: \subsection{Bending}

821: \label{bending.subs}

822: When a beam undergoes bending deformation

823: (fig.~\ref{beamdoppelt.fig}), the resulting forces and moments can be

824: found by superposing the forces and moments that would act if the beam

825: would be bent at each side separately. Therefore we can restrict

826: ourself on calculating the forces and moments according to the

827: single sided deformation drawn in fig.~\ref{beameinseitig.fig}. 

828: \begin{figure}[ht]

829:   \ifcase \value{nofigs}

830: \centerline{\psfig{figure=/figs/beamdoppelt.eps,width=8cm,angle=270}}

831: \vspace{0.5cm}

832: \fi

833: \caption{A bent beam. The forces $F_A$ and $F_B$ and the moments $M_A$ 

834: and $M_B$ can be calculated by superposing the forces and moments of two 

835: single sided bent beams 

836: \ifcase \value{refcaption}

837: (fig.~\ref{beameinseitig.fig}), one of them bent at the side $A$, 

838: the other one at the side $B$. The results are given in 

839: eqs.~(\ref{momentsbending.eqA}, \ref{momentsbending.eqB}) and (\ref{forcesbending.eq}).

840: \or

841: (fig.~12), one of them bent at the side $A$, the other one at the 

842: side $B$. The results are given in eqs.~(32) and (33).

843: \fi

844: }

845: \label{beamdoppelt.fig}

846: \end{figure}

847: 

848: The deformation in fig.~\ref{beameinseitig.fig}, however, can be

849: understood as a deformation of a beam with free ends according to a

850: single sided moment $M$ (fig.~\ref{auflieger.fig}) superposed by a

851: moment $M_b$ (fig.~\ref{beameinseitig.fig}) that assures that the

852: angle $\Theta_B$ (fig.~\ref{beameinseitig.fig}) results to zero.

853: \begin{figure}[ht]

854:   \ifcase \value{nofigs}

855:   \centerline{\psfig{figure=/figs/beameinseitig.eps,width=8cm,angle=270}}

856: \vspace{0.5cm}

857: \fi

858: \caption{A single sided bent beam. The forces and moments result from the 

859: superposition of the deformation drawn in 

860: \ifcase \value{refcaption}

861: fig.~\ref{auflieger.fig} and the restoring moment $M_B$ 

862: (fig.~\ref{beameinseitig.fig}) which assures the angle 

863: $\Theta_B$ to be zero. The equations for the forces and 

864: moments are given in eqs.~(\ref{onesidebending}) and 

865: (\ref{onesidebendingf}).

866: \or

867: fig.~13 and the restoring moment $M_B$ (fig.~12) which assures the 

868: angle $\Theta_B$ to be zero. The equations for the forces and 

869: moments are given in eqs.~(27) and (29).

870: \fi

871: }

872: \label{beameinseitig.fig}

873: \end{figure}

874: 

875: First we consider one sided bending of the beam

876: (fig.~\ref{beameinseitig.fig}). For a beam as drawn in

877: fig.~\ref{auflieger.fig} where the moment $M$ acts at one of 

878: its both sides one finds easily

879: %\begin{mathletters}

880: \begin{eqnarray}

881: \Theta_A^* &=& \frac{ML}{3EI}

882: \label{thetafree.eqA}

883: \\

884: \Theta_B^* &=& - \frac{ML}{6EI}.

885: \label{thetafree.eqB}

886: \end{eqnarray}

887: %\end{mathletters}

888: \begin{figure}[ht]

889:   \ifcase \value{nofigs}

890:   \centerline{\psfig{figure=/figs/auflieger.eps,width=8cm,angle=270}}

891: \vspace{0.5cm}

892: \fi

893: \caption{If the ends of the beam are not fixed but freely moving, the 

894: applied moment $M$ causes bending due to the angles $\Theta_A$ and 

895: $\Theta_B$ given in 

896: \ifcase \value{refcaption}

897: eqs.~(\ref{thetafree.eqA}, \ref{thetafree.eqB}).

898: \or

899: eqs.~(23).

900: \fi

901: }

902: \label{auflieger.fig}

903: \end{figure}

904: 

905: Since the beam in Fig.~\ref{beameinseitig.fig} has the angle

906: $\Theta_B=0$ we find the moments acting at the ends of the beam in

907: fig.~\ref{beameinseitig.fig} by superposing the moment

908: $M_B$, which causes a virtual angle $\Theta_B^{**}$, fulfilling the

909: condition $\Theta_B^{**}+\Theta_B^{*}=0$:

910: \begin{equation}

911: M_B=\frac{3EI}{L}(- \Theta_B^*)= \frac{M}{2}.

912: \end{equation} 

913: The moment $M_B$ causes the angle 

914: \begin{equation}

915: \Theta_A^{**}=-\frac{\frac{M}{2}L}{6EI}=-\frac{ML}{12EI}

916: \end{equation}

917: and the resulting angle $\Theta_A$ (fig.~\ref{beameinseitig.fig}) is

918: \begin{equation}

919: \Theta_A=\Theta_A^{*}+\Theta_A^{**}=\frac{M_AL}{4EI}~.

920: \end{equation}

921: Hence, finally we find for the different moments and forces for the bending

922: deformation (fig.~\ref{beameinseitig.fig})

923: %\begin{mathletters}

924: \begin{eqnarray}

925: M_A &=& - \frac{4EI}{L} \Theta_A \label{onesidebendingA}

926: \\

927: M_B &=& - \frac{2EI}{L} \Theta_A \label{onesidebendingB}

928: \end{eqnarray}

929: %\end{mathletters}

930: and with 

931: \begin{equation}

932: M_A = - F_BL - M_B

933:   \label{addmoments}

934: \end{equation}

935: \begin{equation}

936: F_B=-F_A=\frac{6EI}{L^2} \Theta_A~.

937: \label{onesidebendingf}

938: \end{equation}

939: (Note that the moments $M_A$ and $M_B$ in eq.~(\ref{addmoments}) have

940: to be added as if they would apply to the same end of the beam.)

941:  

942: If the beam is bent at both ends by angles $\Theta_A$ and $\Theta_B$

943: (Fig.~\ref{beamdoppelt.fig}) the resulting moments $M_A$ and $M_B$ and

944: forces $F_A$ and $F_B$ can be calculated by a superposition of the

945: forces and moments according to two independent bending deformations

946: of the type discussed in the previous case.

947: 

948: The angle $\Theta_A$ generates the moments $M_A^{*}$ and $M_B^{*}$

949: according to eqs.~(\ref{onesidebendingA}, \ref{onesidebendingB})

950: %\begin{mathletters}

951: \begin{eqnarray}

952: M_A^{*} &=& - \frac{4EI}{L} \Theta_A\\

953: M_B^{*} &=& - \frac{2EI}{L} \Theta_A

954: \end{eqnarray}

955: %\label{twosidebendingm}

956: %\end{mathletters}

957: and the angle $\Theta_B$ causes the moments

958: %\begin{mathletters}

959: \begin{eqnarray}

960: M_A^{**} &=& - \frac{2EI}{L} \Theta_B \\

961: M_B^{**} &=& - \frac{4EI}{L} \Theta_B~.

962: \end{eqnarray}

963: %\end{mathletters}

964: Hence we find the resulting moments

965: %\begin{mathletters}

966: \begin{eqnarray}

967: M_A=M_A^{*}+M_A^{**} &=& - \left( \frac{4EI}{L} \Theta_A +

968:   \frac{2EI}{L} \Theta_B \right)

969: \label{momentsbending.eqA}

970: \\

971: M_B=M_B^{*}+M_B^{**} &=& - \left( \frac{2EI}{L} \Theta_A + \frac{4EI}{L} \Theta_B~\right),

972: \label{momentsbending.eqB}

973: \end{eqnarray}

974: %\end{mathletters}

975: and with $M_A+M_B=- F_BL$ the resulting forces

976: \begin{equation}

977: F_B = -F_A =\frac{6EI}{L^2}\left( \Theta_A + \Theta_B \right)~.

978: \label{forcesbending.eq}

979: \end{equation}

980: 

981: Comparing the results from sections \ref{shearing.subs} and

982: \ref{bending.subs} one can show that each deformation of a beam can be

983: expressed by bending and elongation, i.e.~the shear deformation need

984: not to be considered. The proof is given in Appendix \ref{redundanz}.

985: 

986: The deformation of the beams is damped proportionally to the

987: deformation rates

988: %\begin{mathletters}

989: \begin{eqnarray}

990:   M_A^{(d)} &=& -\frac{\gamma~I}{L} ~\dot{\Theta}_A , \\ 

991:   M_B^{(d)} &=& -\frac{\gamma~I}{L} ~\dot{\Theta}_B , \\ 

992:   F_A^{N(d)} &=& -\gamma\left( \vec{v}_A - \vec{v}_B \right)

993:     \cdot \stackrel{\longrightarrow}{AB} ,\\

994:   F_B^{N(d)} &=& \gamma\left( \vec{v}_A - \vec{v}_B \right)

995:     \cdot \stackrel{\longrightarrow}{AB} ,

996: %  \label{dampingmom}

997: \end{eqnarray}

998: %\end{mathletters}

999: where $\gamma$ is the damping coefficient of the beam material.  As

1000: proven in the appendix the deformation of the beam is in linear

1001: approximation completely determined by the angles $\Theta_A$ and

1002: $\Theta_B$ and the length $L$. Hence there is no damping force acting

1003: in shear direction.

1004: 

1005: \section{Implementation of the algorithm}

1006: \label{implementationSEC}

1007: We implemented a molecular dynamics algorithm using the particle model

1008: described in the previous sections in FORTRAN. A Gear predictor

1009: corrector scheme of fourth

1010: order~\cite{AllenTildesley:1987,Gear:1966ug} was applied to integrate

1011: Newton's equations of motion. We have to proceed both the forces

1012: caused by the elastic deformation of the triangles as well as the

1013: forces induced by the beams for each iteration step

1014: (eq.~\ref{totalforceEQ}). Since every particle can interact with each

1015: other one causing a high algorithmic complexity, we applied a Verlet

1016: neighborhood list~\cite{Verlet:1967ug} to decrease the amount of

1017: computer time. As discussed below (see table~\ref{table}) the

1018: calculation of the triangle intersections is the most time intensive

1019: part of the calculation. Hence we applied an advanced version of the

1020: neighborhood list method to keep the fraction of the computation of

1021: the triangle--interactions as small as possible.

1022: 

1023: Our list method acts in two steps. In the first step we prepare

1024: neighborhood lists for each triangle to reduce the number of possible

1025: interactions which has to checked. Because we have to investigate all

1026: possible triangle interactions the time for this step rises as the

1027: square of the total number of triangles $T\sim \left( N \cdot

1028: n_t\right)^2$. (For simplicity we assumed for this estimate that all

1029: $N$ grains consist of $n_t$ triangles each. Hence there are $\left( N

1030: \cdot n_t\right) \cdot \left( \left( N-1\right) \cdot n_t\right) $

1031: possible triangle--interactions.)  In the second step we check for

1032: each entry of the constructed neighborhood list whether the

1033: corresponding triangles interact indeed or not. If they interact,

1034: i.e.~if there is an overlap area, the type of interaction is

1035: determined according to the classification scheme in

1036: section~\ref{trianglesSEC}.

1037: 

1038: Finally we construct three lists for the interactions of the first

1039: three types of the classification scheme. The interactions of the

1040: forth and fifth type are replaced by interactions of the first type as

1041: described above. The computer time for this step rises linearly with

1042: the number of particles $N$ and with the number of triangles $n_t$

1043: belonging to each particle. Different from usual Verlet tables these

1044: lists contain only triangles which definitely do interact, and hence

1045: we call the most time consuming subroutines for the calculation of the

1046: forces only for triangles which do interact in one of the five manners

1047: described above (section~\ref{trianglesSEC}).

1048: 

1049: The calculation of the forces induced by the beam deformations follows straight

1050: forward the formulas described in section~\ref{stressesinbeamsSEC}.

1051: 

1052: In table~\ref{table} we present a detailed performance analysis of

1053: the algorithm. The numbers in the table refer to simulations on a DEC

1054: 3000/700 workstation. As visible from the data most of the time

1055: (approximately $70 \%$) is necessary to construct the interaction

1056: list, because we check here every pair of neighbored triangles for

1057: intersections. The construction of this lists give the possibility to

1058: classify the interactions and therefore to use for each kind of

1059: interaction its own optimised algorithm. This prevents us from

1060: calculating useless intersection points, areas and other data.

1061: 

1062: The computer time needed for the predictor--corrector integration is

1063: very low ($3.7 \%$). Thus it seems to be not useful to optimise the

1064: integration procedure, or to decrease the accuracy of the integration

1065: to safe computer time. The calculation of the beam forces is not

1066: time expensive as well. 

1067: 

1068: \begin{table}[htbp]

1069:   \caption{The performance analysis of the algorithm. The 

1070:     calculation has been done on a DEC--3000/700 workstation

1071:     (explanation in the text).}

1072:   \begin{tabular}{llll}

1073:     Basic algorithm & portion of & subroutine & portion of\\

1074:     &computation&&computation\\

1075:     \hline

1076:     Construction of the lists & 73.7 \% & Verlet list & 2.07 \% \\

1077:     \cline{3-4}

1078:     && Classifying interaction & 70.62 \% \\ \cline{3-4}

1079:    \hline

1080:     Calculation of the & 7.87 \% & TYPE 1 & 7.54 \% \\ \cline{3-4}

1081:     triangle interaction & & TYPE 2 & 0.24 \% \\ \cline{3-4}

1082:     & & TYPE 3 & 0.09 \% \\

1083:     \hline

1084:     Calculation of the & & & 13.33 \% \\

1085:     beam forces & & &\\

1086:     \hline

1087:     integration & 3.71 \% & predictor & 1.16 \% \\ \cline{3-4}

1088:     && corrector & 2.55 \%\\

1089:     \hline

1090:     all others & & & 1.4 \% 

1091:   \end{tabular}

1092:   \label{table}

1093: \end{table}

1094: 

1095: From the data in table~\ref{table} it is obvious that triangle

1096: interactions of the first type dominate. Approximately $98 \%$ of the

1097: occurring interactions are of this type. Interactions of the fourth and

1098: fifth type, where our program yields approximative results only, occur

1099: extremely seldomly. We found them only a few times during all our

1100: simulations.

1101: 

1102: \section{First results}

1103: \label{resultsSEC}

1104: \subsection{The collision of a two particles}

1105: \label{twoparticles}

1106: To demonstrate the non--trivial behavior of colliding non--spherical

1107: particles and to check the correctness of our implemented model we

1108: want to present the results of an experiment where a quadratic grain

1109: collides with another resting one in the absence of gravity. The grains

1110: consist of four equal triangles each, forming a simple quadratic grain

1111: as shown in fig.~\ref{grainsFIG}.  The parameters of the materials

1112: are: $Y=2\cdot 10^{7}\, g/(cm\, sec^2)$, $E = 1\cdot 10^{5}

1113: \,g/(sec^2)$, $I=1\cdot 10^{-4} cm^{3}$ and $\gamma=9$ $g/sec$.

1114: The value of the Young--module $Y$ agrees with most of the simulations

1115: of spheres which can be found in the literature.

1116: 

1117: Figs.~\ref{fig:twoparticles} (left figures) show the time evolution

1118: of the particles as a sequence of stroboscopic snapshots for different

1119: relative velocities of the particles a)~$v_{rel}=10\,cm/sec$,

1120: b)~$v_{rel}=50\,cm/sec$ and c)~$v_{rel}=100\,cm/sec$.

1121: \begin{figure}[htbp]

1122:   \ifcase \value{nofigs}

1123:     \centerline{\psfig{figure=figs/twopart.10.ps,width=4cm}\hspace{0.5cm}\psfig{figure=figs/e10.ps,width=7cm,angle=270}}

1124:     \centerline{\psfig{figure=figs/twopart.50.ps,width=4cm}\hspace{0.5cm}\psfig{figure=figs/e50.ps,width=7cm,angle=270}}

1125:       \centerline{\psfig{figure=figs/twopart.100.ps,width=4cm}\hspace{0.5cm}\psfig{figure=figs/e100.ps,width=7cm,angle=270}}

1126: \vspace{0.5cm}

1127: \fi

1128:   \caption{The collision of a horizontally from left to right 

1129:     moving grain (black drawn) with a resting one (grey) for different

1130:     impact rates $v_{rel}=10 \,cm/sec$ (top), $v_{rel}=50

1131:     \,cm/sec$ (middle), and $v_{rel}=100 \,cm/sec$

1132:     (bottom). The stroboscopic snapshots of the grains are take at

1133:     equidistant time intervals. An animated

1134:     sequence of the collisions can be accessed via World Wide

1135: %???Web~\cite{www}.

1136: Web~[67].

1137:     The right hand figures display the kinetic, rotational and total

1138:     energies of the particle collisions over time.}

1139:   \label{fig:twoparticles}

1140: \end{figure}

1141: 

1142: 

1143: One of the triangles of each grain has been filled to visualise the

1144: rotation of the grains. Initially the velocity of the left particle is

1145: $\vec{v}=\left({v_{rel},0}\right) \,cm/sec$ and the right

1146: particle rests. For low velocities $v<v_{tc}\approx 70 cm/sec$ the

1147: grains collide twice within a very short time. Therefore the traces of

1148: the particles that undergo soft collisions with velocity values

1149: $v<v_{tc}$ differ qualitatively from the traces of particles colliding

1150: with higher relative velocity. Since it is hard to display the

1151: complicated motion of the grains in detail we refer to our World Wide

1152: Web--site URL

1153: http://summa.physik.hu-berlin.de:80/$\sim$thorsten/MDASGP.html~\cite{www}

1154: where one can find animated sequences of various problems mentioned in

1155: the current paper.  The right hand side in fig.~\ref{fig:twoparticles}

1156: shows the rotational and kinetic energies and the total energy as a

1157: function of time. The initial kinetic energy of the translational

1158: motion of the left particle results in kinetic energy of both

1159: particles due to translational motion and rotation. A part of the

1160: mechanical energy is lost due to dissipation in the beams. The

1161: relative amount of rotational energy depends strongly on the impact

1162: velocity as expected from the discussion above. The dependence of the

1163: extinction of the rotational degree of freedom on the impact rate is

1164: demonstrated in fig.~\ref{rotvel} too. This figure shows the time

1165: evolution of the angular velocity of the grains $\omega =

1166: \frac{1}{n_t}\,\left| \sum\limits_{i=1}^{n_t} \left( \vec{v}_i -

1167:     \vec{v}_g \right) \times \vec{s}_i\right| $ over time for

1168: different values of the impact velocity. $\vec{v}_g = \frac{1}{n_t}

1169: \sum\limits_{i=1}^{n_t} \vec{v}_i$ is the velocity of the grain and

1170: $\vec{s}_i$ is the vector pointing from the center of mass point of

1171: the grain to the center of mass point of the triangle $i$. $n_t$ is

1172: the number of triangles the grain consists of. 

1173: \begin{figure}[htbp]

1174:   \ifcase \value{nofigs}

1175:   \centerline{\psfig{figure=figs/angvel.ps,width=8cm,angle=270}}

1176: \vspace{0.5cm}

1177: \fi

1178:   \caption{The angular velocities of the particles over the

1179:     time. For lower velocities the particles collide two times,

1180:     therefore those collisions results in a small angular velocity of

1181:     the grains.}

1182:   \label{rotvel}

1183: \end{figure}

1184: 

1185: For high velocity one

1186: observes an abrupt change of the rotational motion at the time of the

1187: collision and a short damped oscillation according to the excited

1188: inner degrees of freedom of the grain (distortion of beams) and their

1189: relaxation.  After the collision the particles move on with constant

1190: translational and rotational velocities. For low impact rate we find a

1191: quite different behavior: First the particles collide as in the

1192: previous case, resulting in a certain rotational motion. Short time

1193: after the first collision, however, they collide a second time with

1194: different edges. Finally we find for the latter case of slow

1195: collisions a very small resulting angular velocity because the second

1196: collision causes angular moments in opposite directions, hence the

1197: resulting moment is small.

1198: 

1199: Observing the motion of a moving particle colliding with a resting one

1200: (see~\cite{www}), i.e.~the simplest possible contact of particles, one

1201: can remark that even for this system there is a complex behavior.  For

1202: rigid spheres the velocities after a collision as a function of the

1203: impact velocities~(eq.~(\ref{resitutionEQa}, \ref{resitutionEQb})) can be calculated

1204: analytically solving the viscoelastic equations of the

1205: spheres~\cite{BrilliantovSpahnHertzschPoeschel:1994}. We guess that

1206: this will not be possible for the case of our particles.

1207:       

1208: \subsection{Outflow of a hopper}

1209: \label{sec:hopper}

1210: The outflow of a hopper is of high interest not only because of the

1211: technological importance of this process, but also because of the

1212: exciting phenomena as clogging and density waves which have been

1213: discovered recently (e.g.~\cite{BaxterBehringerFagertJohnson:1989}).

1214: Density waves and clogging have been investigated using molecular

1215: dynamics in two and three dimensions in various papers,

1216: e.g.~\cite{FormKohringMelinPuhlTillemans:1994,LangstonTuezuenHeyes:1993,RistowHerrmann:1995}.

1217: There is at least one other interesting and surprising phenomenon,

1218: detected recently by Evesque and

1219: Meftah~\cite{EvesqueMeftah:1992,EvesqueMeftah:JDP}: They found that a

1220: hour glass ``ticks'' much slower if it is subjected to vertical

1221: vibrations $y=A\cos\left(\omega t\right)$. The effect seems to depend

1222: on the frequency $f=\omega/2\,\pi$ and on the acceleration of the

1223: vibration $\Gamma = A\,\omega^2$. For frequencies between $40\,Hz\le f

1224: \le 60\,Hz$ surprisingly they observed for some accelerations that the

1225: flow almost stops.

1226: 

1227: We investigated this effect using different particle models and we

1228: found that the effect could be reproduced with the new particle model

1229: only. The results will be described in detail in a forthcoming

1230: paper~\cite{BuchholtzPoeschel:1995hour}.

1231: 

1232: Fig.~(\ref{fig:hopper}) shows snapshots of the system with $N=400$

1233: complex particles. For an animated sequence we refer to our

1234: WWW--site~\cite{www}. The left figure shows the outflowing hopper, the

1235: flow varies irregularly and the grey scale codes for the particle

1236: velocities. When animating the figures one clearly observes density

1237: waves~\cite{www}. The right hand figure shows the same system a few

1238: moments later. The flow has stopped due to clogging.

1239: \begin{figure}[htbp]

1240:   \ifcase \value{nofigs}

1241: \centerline{\psfig{figure=figs/start.ps,width=4cm,angle=0,bbllx=170bp,bblly=130bp,bburx=430bp,bbury=415bp,clip=}\psfig{figure=figs/end.ps,width=4cm,angle=0,bbllx=170bp,bblly=130bp,bburx=430bp,bbury=415bp,clip=}}

1242: \fi

1243:   \caption{Snapshots of the simulation of an outflowing 

1244:     hopper. The left figure shows regular outflow, the right figure

1245:     show a clogging situation. An animated sequence is available via

1246:     World Wide 

1247: %???Web~\cite{www}.

1248: Web~[67].

1249: }

1250:   \label{fig:hopper}

1251: \end{figure}

1252: %\clearpage

1253: 

1254: \subsection{Granular flow in a rotating cylinder}

1255: \label{sec:drehtrommel}

1256: The flow of granular material in a rotating cylinder is one of the

1257: most popular problems in the field of granular materials. It has been

1258: investigated experimentally and theoretically by many authors using

1259: various techniques (see

1260: e.g.~\cite{Rajchenbach:1990,Rolf:1993KUG,Nakagawa:1994,ZikEtAl:1994,Ristow:1994,BaumannJanosiWolf:1994,PoeschelBuchholtz:1993}

1261: and many others). The results shall not be discussed here. We applied

1262: the algorithm to this problem too. For the detailed description of the

1263: results see~\cite{BuchholtzPoeschelTillemans:1994}. Here we only want

1264: to demonstrate the abilities of the method and to mention the main

1265: results briefly.

1266: 

1267: Fig.~\ref{snap.fig} shows snapshots of the simulation of a slowly

1268: rotating cylinder. The grey scale codes for the velocity of the

1269: grains, black means high velocity, black codes for high velocity.

1270: For low angular velocity of the driven cylinder one finds

1271: experimentally stick--slip flow (e.g.~\cite{Rajchenbach:1990}),

1272: i.e.~the material moves downwards not homogeneously but it forms

1273: avalanches. In the right figure one observes an avalanche on the top

1274: of the material indexed by light grey shadowed grains. The left figure shows

1275: the same system immediately before the avalanche. Avalanches are

1276: relatively seldom events. Most of the time the particles rest with

1277: respect to each other, i.e.~they move only due to the rotation of the

1278: cylinder (left side of fig.~\ref{snap.fig}). When increasing the

1279: angular velocity of the cylinder there is a relatively sharp

1280: transition between the stick--slip flow and the homogeneous regime.

1281: This transition could be found in the simulation using non--spherical

1282: grains~\cite{BuchholtzPoeschelTillemans:1994}. Animated sequences for

1283: both regimes, stick--slip and continuous flow, can be accessed via

1284: World Wide Web~\cite{www}.

1285: \begin{figure}[ht]

1286:   \ifcase \value{nofigs}

1287: \centerline{\psfig{figure=figs/snap.ps,width=8cm,angle=0}}

1288: \fi

1289: \caption{\em Snapshots of the simulation with $N=500$ complex

1290:   particles with $P\in [0.1\,cm; 0.2\,cm]$ in a rotating cylinder of

1291:   diameter $D=4\,cm$. The angular velocity is $\Omega=0.1\,sec^{-1}$.

1292:   The wall particles have not been drawn. The right snapshot shows an

1293:   avalanche, the left one has been taken short time before. The grey

1294:   scale codes for the particle velocity.}

1295: \label{snap.fig}

1296: \end{figure}

1297: 

1298: 

1299: Fig.~\ref{inclination.fig} shows the dependence of the difference

1300: between the inclination of the material and the angle of repose on the

1301: angular velocity of the cylinder. The line displays the behavior found

1302: by Rajchenbach $\Theta-\Theta_c \sim

1303: \Omega^2$~\cite{Rajchenbach:1990}. The simulation data are in good

1304: agreement with this observation. 

1305: \begin{figure}[ht]

1306:   \ifcase \value{nofigs}

1307: \centerline{\psfig{figure=figs/thetaomega.ps,width=10cm,angle=270}}

1308: \vspace{0.5cm}

1309: \fi

1310: \caption{\em The inclination $\Theta$ of the material surface over 

1311:   the angular velocity $\Omega$. The dotted line displays the function

1312:   which has been measured experimentally by

1313: %???  \nobreak{Rajchenbach}~\cite{Rajchenbach:1990}.

1314:   \nobreak{Rajchenbach}~[73].

1315: }

1316: \label{inclination.fig}

1317: \end{figure}

1318: 

1319: Both results mentioned, the sharp transition between the flow regimes

1320: and the inclination as a function of the angular velocity of the

1321: cylinder, agree with experimental observations. We have not been

1322: able to find these effects using simple spheres or using particles

1323: composed of

1324: spheres~\cite{PoeschelBuchholtz:1993,PoeschelBuchholtz:1993CSF} in

1325: molecular dynamics simulations.

1326: 

1327: \section{Conclusion}

1328: \label{ConclusionSEC}

1329: We presented a model for the simulation of the particles of a granular

1330: material. Each particle consists of triangles which are connected by

1331: deformable beams. There are no restrictions concerning the shape of the

1332: grains (convex or concave), nor the number or the shape of the

1333: triangles a grain consists of. To preserve calculation time it is

1334: favourable to chose arrangements of the triangles so that the beams

1335: are fixed at the center of mass points of the triangles. Otherwise one needs

1336: additional computation time caused by the Steiner law. 

1337: 

1338: When two triangles of different grains collide there acts an elastic

1339: restoring force proportionally to the compression of the particles.

1340: During the collision of the triangles the energy is preserved. A

1341: collision of a triangle with another one causes moments and forces

1342: acting on this triangle and hence a deformation of the beams which

1343: connect the triangles to neighboring ones. The deformation of the

1344: beams causes forces and moments acting on the neighboring triangles.

1345: 

1346: Beams can be deformed in axial direction (elongation) and in shear

1347: direction and they can be bent. When a beam is deformed it dissipates

1348: energy proportionally to the deformation rate.

1349: 

1350: In the present paper we described beams that recover completely while

1351: dissipating mechanical energy when the deforming moments and forces

1352: vanish. Our model, however, is not restricted to this case.

1353: Interesting phenomena are plastic deformation and wear which can be

1354: easily simulated by introducing thresholds for the beam forces and

1355: moments or for the deformations. When a force or a moment or a

1356: deformation, respectively, exceeds the threshold the beam breaks or

1357: deforms permanently. Similar simulations with other models have been

1358: reported in the

1359: literature~\cite{Tillemans:1994,Handley:1993,PotapovCampbellHopkins:1994aug}.

1360: 

1361: The proposed algorithm has been implemented in FORTRAN. We have

1362: demonstrated the behavior of a system of grains applying it to three

1363: examples of granular assemblies. In a simple collision simulation of

1364: two grains we discussed the trajectories of the grains as well as the

1365: evolution of the kinetic and rotational energies depending on the

1366: impact rate. Sample results for the flow out of a hopper and the

1367: motion of granular material in a rotating cylinder have been

1368: presented. In the latter case we found quantitative agreement of the

1369: simulation with the experiment that could not be found so far, neither

1370: with spherical grains nor with non--spherical grains of other type.

1371: 

1372: \ack We thank Hans J.~Herrmann, Stefan Schwarzer and

1373: Hans--J\"urgen Tillemans for many helpful discussions and comments.

1374: This work has been supported by the Deutsche Forschungsgemeinschaft

1375: (grant Ro 548/5-1).

1376: 

1377: \begin{appendix}

1378: \section{Shear deformation can be expressed by bending}

1379: \label{redundanz}

1380: The deformation of a beam in the two dimensional space has three

1381: degrees of freedom: the distance of the relative positions of the ends

1382: of the beams $L=\sqrt{\left( x_A-x_B\right) ^2+\left( y_A-y_B\right)

1383:   ^2}$, and the angles $\Theta_A$ and $\Theta_B$. These values are the

1384: parameters in eqs.~(\ref{momentsbending.eqA}, \ref{momentsbending.eqB}) and

1385: (\ref{forcesbending.eq}). The shear parameter $\Delta$ in

1386: eqs.~(\ref{forcesshearing.eq}, \ref{momshearing.eq})

1387: %\ref{eq.shearing}) 

1388: is a fourth

1389: parameter, i.e.~one of them must be redundant. In the following we

1390: will show that shear deformation can be expressed by bending, at least

1391: in our approach of a linear deformation--force relation.

1392: 

1393: \begin{figure}[ht]

1394:   \ifcase \value{nofigs}

1395: \centerline{\psfig{figure=/figs/shearbend.eps,width=8cm,angle=270}}

1396: \vspace{0.5cm}

1397: \fi

1398: \caption{The deformation of a beam in shear direction $\Delta$ (upper 

1399: part) can be represented by bending deformation (lower part). The 

1400: appendix contains the proof that one need not to consider shear 

1401: {\em and} bending deformation in the linear approximation which 

1402: is justified for small deformations.}

1403: \label{shearbend.fig}

1404: \end{figure}

1405: Suppose we have a deformed beam as drawn in Fig.~\ref{shearbend.fig}

1406: (bottom). Then applying eqs.~(\ref{forcesshearing.eq}, \ref{momshearing.eq})

1407: %\ref{eq.shearing}) 

1408: we find for the

1409: forces and moments according to shearing

1410: %\begin{mathletters}

1411: \begin{eqnarray}

1412: M_A^s &=& M_B^s = \frac{6EI}{\left(L^*\right)^2}\, \Delta\\

1413: F_A^s &=& -F_B^s = \frac{12EI}{\left(L^*\right)^3}\, \Delta

1414: %\label{Shearing*}

1415: \end{eqnarray}

1416: %\end{mathletters}

1417: and according to bending (eqs.~(\ref{momentsbending.eqA},\ref{momentsbending.eqB}) and

1418: (\ref{forcesbending.eq}))

1419: %\begin{mathletters}

1420: \begin{eqnarray}

1421: M_A^b &=& -\frac{2EI}{L^*}\,\Theta_B^*\\

1422: M_B^b &=& -\frac{4EI}{L^*}\,\Theta_B^*\\

1423: F_A^b &=& -F_B^b = -\frac{6EI}{\left(L^*\right)^2}\,\Theta_B^*~.

1424: %\label{Bending*}

1425: \end{eqnarray}

1426: %\end{mathletters}

1427: The total forces and moments read

1428: %\begin{mathletters}

1429: \begin{eqnarray}

1430: M_A^*&=& M_A^s + M_A^b = \frac{6EI}{\left(L^*\right)^2}\,\Delta - \frac{2EI}{L^*}\,\Theta_B^* 

1431: \label{Total*A}

1432: \\

1433: M_B^*&=& M_B^s + M_B^b = \frac{6EI}{\left(L^*\right)^2}\,\Delta - \frac{4EI}{L^*}\,\Theta_B^*

1434: \label{Total*B}

1435:  \\

1436: F_A^*&=&-F_B^* = \frac{12EI}{\left(L^*\right)^3}\, \Delta - \frac{6EI}{\left(L^*\right)^2}\,\Theta_B^*~.

1437: \label{Total*C}

1438: \end{eqnarray}

1439: %\end{mathletters}

1440: Turning the entire system by the angle $\Theta_A\approx\tan \Theta_A =

1441: -\frac{\Delta}{L^*}$ we obtain the new system as drawn in the top of

1442: fig.~\ref{shearbend.fig} with the new angles

1443: $\Theta_A=-\frac{\Delta}{L}$ and $\Theta_B=\Theta_B^*-\frac{\Delta}{L}$.

1444: With eqs.~(\ref{momentsbending.eqA}, \ref{momentsbending.eqB}) and (\ref{forcesbending.eq}) we find

1445: %\begin{mathletters}

1446: \begin{eqnarray}

1447: M_A &=& -\frac{4EI}{L}\,\Theta_A- \frac{2EI}{L}\,\Theta_B = \frac{6EI}{L^2}\,\Delta - \frac{2EI}{L}\,\Theta_B^*

1448: \label{TotalA}

1449: \\

1450: M_B &=& -\frac{2EI}{L}\,\Theta_A- \frac{4EI}{L}\,\Theta_B = \frac{6EI}{L^2}\,\Delta - \frac{4EI}{L}\,\Theta_B^*

1451: \label{TotalB}

1452: \\

1453: F_A &=& -F_B = -\frac{6EI}{L^2}\,\left(\Theta_A+\Theta_B \right) = \frac{12EI}{L^3}\,\Delta - \frac{6EI}{L^2}\,\Theta_B^*

1454: \label{TotalC}

1455: \end{eqnarray}

1456: %\end{mathletters}

1457: Approximating $L$ by $\sqrt{L^2+\Delta^2} = L^*$ evidently the results

1458: in eqs.~(\ref{TotalA}-\ref{TotalC}) coincide with eqs.~(\ref{Total*A}-\ref{Total*C}). Thus we do

1459: not need to consider shear deformation in our simulations since it is

1460: redundant here.

1461: \end{appendix}

1462: 

1463: \begin{thebibliography}{10}

1464: 

1465: \bibitem{AlderWainwright:1957ug}

1466: B.~J.~Alder and T.~E.~Wainwright, 

1467: {\em J. Chem. Phys.}, {\bf 27} 1208 (1957).

1468: 

1469: \bibitem{BeazleyLomdahl:1993}

1470: D.~M. Beazley and P.~S. Lomdahl,

1471: {\em Parallel Computing}, {\bf 20}, 173 (1993).

1472: 

1473: \bibitem{BeazleyLomdahlJensenTamayo:1994}

1474: D.~M. Beazley, P.~S. Lomdahl, N.~Gr{\o}nbech--Jensen, and P.~Tamayo,

1475: In {\em Proceedings 8. Intern. Parallel Processing Symposium,

1476:   Canc{\'u}n, Mexico}, pages 800--809, Los Alamos, 1994. IEEE Computer Society.

1477: 

1478: \bibitem{BeazleyEtAl:1995}

1479: D.~M. Beazley, P.~S. Lomdahl, N.~Gr{\o}nbech--Jensen, R.~Giles, 

1480: and P.~Tamayo, In D.~Stauffer, {\em Annual Reviews of Computational Physics}, 

1481: Vol.~III, World Scientific, (Singapore, 1995) (in press).

1482: 

1483: \bibitem{Hoover:1992}

1484: W.~G. Hoover,

1485: {\em Nuclear Physics A}, {\bf 545}, 523c (1992).

1486: 

1487: \bibitem{Salo:1992}

1488: H.~Salo,

1489: {\em Icarus}, {\bf 96}, 85 (1992).

1490: 

1491: \bibitem{Weidenschilling}

1492: S.~J.~Weidenschilling, C.~R.~Chapman, D.~R.~Davis and R.~Greenberg,

1493: In: R.~Greenberg and A.~Brahic (Eds.),{\em Planetary Rings} pages

1494: 367--415, Arizona, 1994.

1495: 

1496: \bibitem{BrilliantovSpahnHertzschPoeschel:1994}

1497: N.~V. Brilliantov, F.~Spahn, J.--M. Hertzsch, and T.~P\"oschel,

1498: {\em preprint HLRZ 17/94} (1994).

1499: 

1500: \bibitem{GoldhirschZanetti:1993}

1501: I.~Goldhirsch and G.~Zanetti,

1502: {\em Phys. Rev. Lett.}, {\bf 70}, 1619 (1993).

1503: 

1504: \bibitem{McNamaraYoung:1992}

1505: S.~McNamara and W.~R. Young,

1506: {\em Phys. Fluids A}, {\bf 4}, 496 (1992).

1507: 

1508: \bibitem{LudingClementBlumenRajchenbachDuran:1994DISS}

1509: S.~Luding, E.~Cl\'ement, A.~Blumen, J.~Rajchenbach, and J.~Duran,

1510: {\em Phys. Rev. E.}, {\bf 50}, 4113 (1994).

1511: 

1512: \bibitem{LudingClementBlumenRajchenbachDuran:1994STUDIES}

1513: S.~Luding, E.~Cl\'ement, A.~Blumen, J.~Rajchenbach, and J.~Duran,

1514: {\em Phys. Rev. E.}, {\bf 49}, 1634 (1994).

1515: 

1516: \bibitem{Rapaport:1980}

1517: D.~C. Rapaport,

1518: {\em J. Comp. Phys.}, {\bf 34}, 184 (1980).

1519: 

1520: \bibitem{Lubachevsky:1991}

1521: B.~D. Lubachevsky,

1522: {\em J. Comp. Phys.}, {\bf 94}, 255 (1991).

1523: 

1524: \bibitem{CundallStrack:1979}

1525: P.~A. Cundall and O.~D.~L. Strack,

1526: {\em G\'eotechnique}, {\bf 29}, 47 (1979).

1527: 

1528: \bibitem{HaffWerner:1986}

1529: P.~K. Haff and B.~T. Werner,

1530: {\em Powder Techn.}, {\bf 48}, 239 (1986).

1531: 

1532: \bibitem{Hertz:1882ug}

1533: H.~Hertz,

1534: {\em J. f. reine u. angewandte Math.}, {\bf 92}, 156 (1882).

1535: 

1536: \bibitem{GallasHerrmannSokolowski:1992PRL}

1537: J.~A.~C. Gallas, H.~J. Herrmann, and S.~Soko{\l}owski,

1538: {\em Phys. Rev. Lett.}, {\bf 69}, 1371 (1992).

1539: 

1540: \bibitem{Taguchi:1992PRL}

1541: Y--h. Taguchi

1542: {\em Phys. Rev. Lett.}, {\bf 69}, 1367 (1992).

1543: 

1544: \bibitem{GallasHerrmannPoeschelSokolowski:1994}

1545: J.~A.~C. Gallas, H.~J. Herrmann, T.~P\"oschel, and S.~Soko{\l}owski,

1546: {\em J. Stat. Phys., in press} (1995).

1547: 

1548: \bibitem{PoeschelHerrmann:1994}

1549: T.~P\"oschel and H.~J. Herrmann,

1550: {\em Europhys. Lett.}, {\bf 29}, 123 (1995).

1551: 

1552: \bibitem{RistowHerrmann:1994}

1553: G.~H. Ristow and H.~J. Herrmann,

1554: {\em Phys. Rev. E.}, {\bf 50}, R5 (1994).

1555: 

1556: \bibitem{FormKohringMelinPuhlTillemans:1994}

1557: W.~Form, G.~A. Kohring, S.~Melin, H.~Puhl, and H.--J. Tillemans,

1558: {\em preprint HLRZ 75/93} (1994).

1559: 

1560: \bibitem{CampbellPotapov:1993}

1561: C.~Campbell and A.~V. Potapov,

1562: {\em in \cite{IinoyaTsujiMasudaHigashitani:1993}}, pages 27--31, (1993).

1563: 

1564: \bibitem{Poeschel:1994}

1565: T.~P\"{o}schel,

1566: {\em J. Physique}, {\bf 4}, 499 (1994).

1567: 

1568: \bibitem{Lee:1994}

1569: J.~Lee,

1570: {\em Phys. Rev. E.}, {\bf 49}, 281 (1994).

1571: 

1572: \bibitem{Ristow:1994}

1573: G.~H. Ristow,

1574: {\em Europhys. Lett.}, {\bf 28}, 97 (1994).

1575: 

1576: \bibitem{RistowCantelaubeBideau:1994ug}

1577: G.~H. Ristow, F.~Cantelaube--Lebec, and D.~Bideau,

1578: {\em preprint} (1994).

1579: 

1580: \bibitem{PoeschelBuchholtz:1993CSF}

1581: T.~P\"{o}schel and V.~Buchholtz,

1582: {\em Chaos, Solitons \& Fractals, in press} (1993).

1583: 

1584: \bibitem{Walton:1991}

1585: O.~R. Walton,

1586: In {\em Second US--Japan Seminar on Micromechanics of Granular

1587:   Materials}, pages 1--9, Potsdam, NY (1991).

1588: 

1589: \bibitem{Poeschel:1993JDP}

1590: T.~P\"oschel,

1591: {\em J. Physique}, {\bf 3}, 27 (1993).

1592: 

1593: \bibitem{Melin:1994}

1594: S.~Melin,

1595: {\em Phys. Rev. E.}, {\bf 49}, 2353 (1994).

1596: 

1597: \bibitem{Taguchi:1992JDP}

1598: Y-h. Taguchi,

1599: {\em J. Physique}, {\bf 2}, 2103 (1992).

1600: 

1601: \bibitem{Taguchi:1994}

1602: Y-h. Taguchi,

1603: {\em preprint TITCMT-94-4} (1994).

1604: 

1605: \bibitem{BuchholtzPoeschel:1993JMPC}

1606: V.~Buchholtz and T.~P\"{o}schel,

1607: {\em J. Mod. Phys. C}, {\bf 4}, 1049 (1993).

1608: 

1609: \bibitem{FormItoKohring:1993}

1610: W.~Form, N.~Ito, and G.~A. Kohring,

1611: {\em J. Mod. Phys. C}, {\bf 4}, 1085 (1993).

1612: 

1613: \bibitem{EveraersKremer:1993}

1614: R.~Everaers, and K.~Kremer,

1615: {\em preprint} (1993).

1616: 

1617: \bibitem{BuchholtzPoeschel:1994PA}

1618: V.~Buchholtz and T.~P\"{o}schel,

1619: {\em Physica A}, {\bf 202}, 390 (1994).

1620: 

1621: \bibitem{Lee:1993}

1622: J.~Lee,

1623: {\em J. Physique I}, {\bf 3}, 2017 (1993).

1624: 

1625: \bibitem{VisscherBolsterli:1972ug}

1626: W.~M. Visscher and M.~Bolsterli,

1627: {\em Nature}, {\bf 239}, 504 (1972).

1628: 

1629: \bibitem{JullienMeakinPawlovitch:1992}

1630: R.~Jullien, P.~Meakin, and A.~Pawlovitch,

1631: {\em Phys. Rev. Lett.}, {\bf 69}, 640 (1992).

1632: 

1633: \bibitem{BaumannJobsWolf:1993}

1634: G.~Baumann, E.~Jobs, and D.~E. Wolf,

1635: In M.~Matsuhita, M.~Shlesinger, and T.~Vicsek, editors, {\em Fractals

1636:   in Natural Sciences, Budapest 1993}, World Scientific (Singapore, 1993).

1637: 

1638: \bibitem{BaumannJanosiWolf:1994}

1639: G.~Baumann, I.~J\'anosi, and D.~E. Wolf,

1640: {\em Europhys. Lett.}, {\bf 27}, 203 (1994).

1641: 

1642: \bibitem{JullienMeakinPawlovitch:1993ug}

1643: R.~Jullien, P.~Meakin, and A.~Pawlovitch,

1644: {\em Europhys. Lett.}, {\bf 22}, 523 (1993).

1645: 

1646: \bibitem{BarkerMehta:1995} 

1647: G.~C. Barker and A.~Mehta, 

1648: Comment on \cite{JullienMeakinPawlovitch:1993ug},  

1649: {\em Europhys. Lett.}, {\bf 29}, 61 (1995); 

1650: R.~Jullien, P.~Meakin, and A.~Pavlovitch, 

1651: Reply to \cite{BarkerMehta:1995}, 

1652: {\em Europhys. Lett.}, {\bf 29}, 63 (1995).

1653: 

1654: \bibitem{PengHerrmann:1994}

1655: G.~Peng and H.~J.~Herrmann,

1656: {\em Phys. Rev. E.}, {\bf 49}, 1796 (1994).

1657: 

1658: \bibitem{CaramHong:1991}

1659: H.~S. Caram and D.~C. Hong,

1660: {\em Phys. Rev. Lett.}, {\bf 67}, 828 (1991).

1661: 

1662: \bibitem{GallasSokolowski:1993}

1663: J.~A.~C. Gallas, and S.~Soko{\l}owski,

1664: {\em J. Mod. Phys. B}, {\bf 7}, 2037 (1993).

1665: 

1666: \bibitem{WaltonBraun:1993}

1667: O.~R. Walton and R.~L. Braun,

1668: {\em Joint DOE/NSF Workshop on flow of particulates and fluids}

1669:   (1993).

1670: 

1671: \bibitem{PoeschelBuchholtz:1993}

1672: T.~P\"{o}schel and V.~Buchholtz,

1673: {\em Phys. Rev. Lett.}, {\bf 71}, 3963 (1993).

1674: 

1675: \bibitem{MustoeDePorter:1993}

1676: G.~G. Mustoe and G.~DePorter,

1677: In C.~Thornton, editor, {\em Powders and Grains'93}, pages

1678:   421--427, Balkema (Rotterdam, 1993).

1679: 

1680: \bibitem{HogueNewland:1993}

1681: C.~Hogue and D.~E. Newland,

1682: In C.~Thornton, editor, {\em Powders and Grains'93}, pages

1683:   413--417, Balkema (Rotterdam, 1993).

1684: 

1685: \bibitem{Stronge:1990}

1686: W.~Stronge,

1687: {\em Proc. Roy. Soc.}, {\bf A431} (1990).

1688: 

1689: \bibitem{TillemansHerrmann:1994}

1690: H.~J. Tillemans and H.~J. Herrmann.

1691: {\em Physica A}, {\bf 217}, 261 (1994).

1692: 

1693: \bibitem{Tillemans:1994}

1694: H.~J. Tillemans,

1695: {\em Molekulardynamik--Simulationen beliebig geformter Teilchen in

1696:   zwei Dimensionen},

1697: PhD thesis, Universit{\"a}t K{\"o}ln (1994).

1698: 

1699: \bibitem{Handley:1993}

1700: M.~F. Handley,

1701: {\em An Investigation onto the Constitutive Behaviour of Brittle

1702:   Granular Media by Numerical Experiment},

1703: PhD thesis, University of Minnesota, Minneapolis (1993).

1704: 

1705: \bibitem{PotapovCampbellHopkins:1994aug}

1706: A.~V. Potapov, C.~S. Campbell, and M.~A. Hopkins,

1707: {\em preprint} (1994).

1708: 

1709: \bibitem{PotapovCampbellHopkins:1994bug}

1710: A.~V. Potapov, C.~S. Campbell, and M.~A. Hopkins,

1711: {\em preprint} (1994).

1712: 

1713: \bibitem{Finney:1979}

1714: J.~L. Finney,

1715: {\em J. Comp. Phys.}, {\bf 32}, 137 (1979).

1716: 

1717: \bibitem{HerrmannHansenRoux:1989}

1718: H.~J. Herrmann, A. Hansen, and S. Roux,

1719: {\em Phys. Rev. B.}, {\bf 39}, 637 (1989).

1720: 

1721: \bibitem{BentleyOttmann:1979ug}

1722: J.~L. Bentley, and T.~Ottmann.

1723: {\em IEEE Trans. Comp. C}, {\bf 28}, 9 (1979).

1724: 

1725: \bibitem{PreparataShamos:1985ug}

1726: F.~P. Preparata and M.~I. Shamos,

1727: {\em Computational Geometry: An Introduction},

1728: Springer--Verlag (Berlin, 1985).

1729: 

1730: \bibitem{TimoshenkoGere:1972}

1731: S.~P. Timoshenko and J.~M. Gere

1732: {\em Mechanics of Materials},

1733: van Nordstrand Reinhold Company (New York, 1972).

1734: 

1735: \bibitem{TimoshenkoLessells:1928}

1736: S.~P. Timoshenko and I.~M. Lessells,

1737: {\em Festigkeitslehre},

1738: Julius Springer (Berlin, 1928).

1739: 

1740: \bibitem{AllenTildesley:1987}

1741: M.~P. Allen and D.~J. Tildesley,

1742: {\em Computer Simulations of Liquids},

1743: Clarendon Press (Oxford, 1987).

1744: 

1745: \bibitem{Gear:1966ug}

1746: C.~W. Gear,

1747: {\em The numerical integration of ordinary differential equations of

1748:   various orders},

1749: Technical Report ANL 7126, Argonne National Laboratory (1966).

1750: 

1751: \bibitem{Verlet:1967ug}

1752: L.~Verlet,

1753: {\em Phys. Rev.}, {\bf 159}, 98 (1967).

1754: 

1755: \bibitem{www} %[67]

1756:   The animated sequences mentioned in this paper can be found via

1757:   world wide web under URL

1758:   http://summa.physik.hu-berlin.de:80/$\sim$thorsten/MDASGP.html~. The

1759:   following 9 ``movies'' can be accessed: a) the collision of two

1760:   dice, one resting, the other one moving with 5 different velocities,

1761:   b) the flow through a hopper, the clogging process, c) the flow of

1762:   granular material in a rotating cylinder in the stick--slip regime,

1763:   and in the continuous regime (1995).

1764: 

1765: \bibitem{BaxterBehringerFagertJohnson:1989}

1766: G.~W. Baxter, R.~P. Behringer, T. Fagert, and G.~A. Johnson,

1767: {\em Phys. Rev. Lett.}, {\bf 62}, 2825 (1989).

1768: 

1769: \bibitem{LangstonTuezuenHeyes:1993}

1770: P.~A. Langston, U.~T\"uz\"un, and D.~M. Heyes,

1771: {\em in \cite{IinoyaTsujiMasudaHigashitani:1993}}, pages 43--48 (1993).

1772: 

1773: \bibitem{RistowHerrmann:1995}

1774: G.~Ristow and H.~J. Herrmann,

1775: {\em Physica A}, {\bf 213}, 474 (1995).

1776: 

1777: \bibitem{EvesqueMeftah:1992}

1778: P.~Evesque and W.~Meftah,

1779: {\em C. R. Acad. Sci. Paris II}, {\bf 314}, 1125 (1992).

1780: 

1781: \bibitem{EvesqueMeftah:JDP}

1782: P.~Evesque and W.~Meftah,

1783: {\em Int.Journ. of Mod. Phys. B}, {\bf 17}, 1799 (1993).

1784: 

1785: \bibitem{BuchholtzPoeschel:1995hour}

1786: V.~Buchholtz and T.~P\"{o}schel,

1787: {\em in preparation} (1995).

1788: 

1789: \bibitem{Rajchenbach:1990}

1790: J.~Rajchenbach,

1791: {\em Phys. Rev. Lett.}, {\bf 65}, 2221 (1990).

1792: 

1793: \bibitem{Rolf:1993KUG}

1794: L.~Rolf,

1795: {\em Zement--Kalk--Gips}, {\bf 46}, 117 (1993).

1796: 

1797: \bibitem{Nakagawa:1994}

1798: M.~Nakagawa,

1799: {\em Chemical Engineering Science}, {\bf 49}, 2540 (1994).

1800: 

1801: \bibitem{ZikEtAl:1994}

1802: O.~Zik, D.~Levine, S.~G. Lipson, S.~Shtrikman, and J.~Stavans,

1803: {\em Phys. Rev. Lett.}, {\bf 73}, 644 (1994).

1804: 

1805: \bibitem{BuchholtzPoeschelTillemans:1994}

1806: V.~Buchholtz, T.~P\"oschel, and H.~J. Tillemans,

1807: {\em Physica A}, {\bf 216}, 199 (1995).

1808: 

1809: \bibitem{IinoyaTsujiMasudaHigashitani:1993}

1810: K.~Iinoya, Y.~Tsuji, H.~Masuda, and K.~Higashitani, editors.

1811: {\em First Nisshin Engineering Particle Technology},

1812: Nisshin Engineering Co. Ltd. (Osaka, 1993).

1813: 

1814: \end{thebibliography}

1815: 

1816: \end{document}

1817: 

1818: