1: \documentclass[12pt]{iopart}

2: \usepackage{graphicx}

3: \eqnobysec

4: 

5: \newcommand{\beq}{\begin{equation}}

6: \newcommand{\beqa}{\begin{eqnarray}}

7: \newcommand{\eeq}{\end{equation}}

8: \newcommand{\eeqa}{\end{eqnarray}}

9: \newcommand{\bfa}{{\bf a}}

10: \newcommand{\bfb}{{\bf b}}

11: \newcommand{\bfd}{{\bf d}}

12: \newcommand{\bfe}{{\bf e}}

13: \newcommand{\bfBB}{{\hskip -.3mm}{\bf B{\hskip -.6mm}B}{\hskip -.3mm}}

14: \newcommand{\bfSS}{{\hskip -.1mm}{\bf S{\hskip -.1mm}S}{\hskip -.1mm}}

15: \newcommand{\bg}{{\rm bg}}

16: \newcommand{\bigmean}[1]{\left\langle#1\right\rangle}

17: \renewcommand{\d}{{\rm d}}

18: \newcommand{\dr}{\delta\rho}

19: \newcommand{\ds}{\displaystyle}

20: \renewcommand{\e}{{\rm e}}

21: \newcommand{\frad}[2]{\displaystyle{\displaystyle#1\over\displaystyle#2}}

22: \newcommand{\gl}{{\rm gl}}

23: \newcommand{\hdr}{\widehat{\delta\rho}}

24: \renewcommand{\i}{{\rm i}}

25: \newcommand{\loc}{{\rm loc}}

26: \renewcommand{\max}{{\rm max}}

27: \newcommand{\mean}[1]{\langle#1\rangle}

28: \newcommand{\prob}{\mathop{\rm Prob}\nolimits}

29: \newcommand{\BB}{{\rm BB}}

30: \newcommand{\BS}{{\rm BS}}

31: \newcommand{\J}{{\bf J}}

32: \newcommand{\N}{{\cal N}}

33: \newcommand{\SB}{{\rm SB}}

34: \renewcommand{\SS}{{\rm SS}}

35: 

36: \begin{document}

37: 

38: \title{Structure of the stationary state of the asymmetric target process}

39: 

40: \author{J M Luck and C Godr\`eche}

41: 

42: \address{Service de Physique Th\'eorique\footnote{URA 2306 of CNRS},

43: CEA Saclay, 91191 Gif-sur-Yvette cedex, France}

44: 

45: \begin{abstract}

46: We introduce a novel migration process, the target process.

47: This process is dual to the zero-range process (ZRP) in the sense that,

48: while for the ZRP the rate of transfer of a particle

49: only depends on the occupation of the departure site,

50: it only depends on the occupation of the arrival site for the target process.

51: More precisely, duality associates to a given ZRP a unique target process,

52: and vice-versa.

53: If the dynamics is symmetric, i.e., in the absence of a bias,

54: both processes have the same stationary-state product measure.

55: In this work we focus our interest on the situation where

56: the latter measure exhibits a continuous condensation transition

57: at some finite critical density $\rho_c$, irrespective of the dimensionality.

58: The novelty comes from the case of asymmetric dynamics,

59: where the target process has a nontrivial fluctuating stationary state,

60: whose characteristics depend on the dimensionality.

61: In one dimension, the system remains homogeneous at any finite density.

62: An alternating scenario however prevails in the high-density regime:

63: typical configurations consist of long

64: alternating sequences of highly occupied and less occupied sites.

65: The local density of the latter is equal to~$\rho_c$

66: and their occupation distribution is critical.

67: In dimension two and above, the asymmetric target process exhibits a phase

68: transition at a threshold density $\rho_0$ much larger than~$\rho_c$.

69: The system is homogeneous at any density below $\rho_0$,

70: whereas for higher densities it exhibits an extended condensate

71: elongated along the direction of the mean current,

72: on top of a critical background with density $\rho_c$.

73: \end{abstract}

74: 

75: \pacs{05.40.-a, 02.50.Ey, 05.70.Ln}

76: 

77: \eads{\mailto{jean-marc.luck@cea.fr},

78: \mailto{claude.godreche@cea.fr}}

79: 

80: \maketitle

81: 

82: \section{Introduction}

83: 

84: In recent years many studies have been devoted to

85: nonequilibrium statistical-mechanical models yielding condensation,

86: such as zero-range processes

87: (ZRP)~\cite{loan,evans1,wis1,cg,gross,evans2,glcond},

88: dynamical urn models~\cite{zeta1,zeta2,zeta3,barc,lux},

89: and mass transport models~\cite{maj}.

90: In all these models the condensate manifests itself by

91: the macroscopic occupation of a single site

92: by a finite fraction of the whole available mass.

93: 

94: The ZRP is the simplest of these models.

95: It is a driven diffusive system with multiple occupations,

96: such that particles hop from site to site on a lattice,

97: with a rate which only depends on the occupation of the departure site.

98: The distribution of the particles among the sites

99: in the stationary state is given by a product measure,

100: which is explicitly known in terms of the rate defining the model,

101: irrespective of the geometry of the system

102: and of the asymmetry of the dynamics~\cite{evans2,spitz,andj}.

103: This property of the stationary-state measure of the ZRP

104: favors the condensation phenomenon.

105: The product structure indeed prevents the condensate

106: from being shared by more than one site~\cite{gross,glcond,maj}.

107: 

108: Dynamical urn models, also called migration processes in the probabilistic

109: li\-te\-ra\-ture~\cite{kelly}, can be viewed as generalizations of the ZRP,

110: where the rate at which a particle is transferred

111: from a departure site to an arrival site

112: now depends on the occupations of both sites.

113: Migration processes do not have a stationary-state product measure in general.

114: 

115: In the present work we introduce a special migration process,

116: the {\it target process}.

117: This process is novel to our knowledge.

118: It has no stationary-state product measure, except in the case

119: of a symmetric dynamics, i.e., in the absence of a bias,

120: leading to an equilibrium stationary state.

121: The class of target processes is dual to the class of ZRP,

122: in the sense that the roles of the departure and arrival sites are interchanged:

123: the rate basically depends on the occupation of the departure (source) site

124: for a ZRP,

125: and on the occupation of the arrival (target) site for a target process.

126: More precisely, to a given ZRP is associated by duality a unique target process,

127: and vice-versa.

128: In the case of a symmetric dynamics,

129: these two dual processes have the same stationary-state product measure.

130: 

131: Our aim is to study the structure of the nonequilibrium stationary state

132: of the asymmetric target process,

133: and especially the fate of the condensation phenomenon.

134: In Section~\ref{stp} we give a detailed definition

135: of the class of target processes.

136: We then focus our attention onto the particular target process

137: which is dual to the `canonical ZRP'

138: studied e.g.~in~\cite{evans1,wis1,cg,gross}.

139: In the case of symmetric dynamics, the target process thus constructed

140: has the same stationary-state product measure as the ZRP,

141: and therefore the same continuous condensation transition.

142: In the asymmetric case, however, the stationary-state measure of the target

143: process is not a product measure,

144: and exhibits non-trivial correlations in general.

145: This basic difference between the ZRP and the target process

146: manifests itself more drastically at high density and in low dimensionality.

147: Sections~\ref{1D1} and~\ref{1D2} are devoted to a thorough theoretical

148: and numerical study of the one-dimensional situation.

149: The system remains homogeneous at any finite density,

150: and presents an alternating structure which is more and more pronounced

151: as the density is increased.

152: The two-dimensional asymmetric target process

153: on the square lattice is the subject of Section~\ref{2D}.

154: It exhibits an unconventional type of condensation phenomenon,

155: with a transition at a threshold density $\rho_0$ much larger than~$\rho_c$,

156: and an extended condensate elongated along the direction of the bias.

157: An analogous scenario presumably generically holds on other lattices

158: and in higher dimension as well.

159: Section~\ref{quatre} contains a Discussion.

160: 

161: \section{The target process}

162: \label{stp}

163: 

164: \subsection{Migration processes and urn models: a reminder}

165: 

166: The definition of a migration process appeared first in the

167: probabilistic literature (see e.g.~\cite{kelly}).

168: For the time being, we restrict the discussion to the one-dimensional case.

169: Consider a system of $N$ particles distributed among~$M$ sites,

170: with periodic boundary conditions.

171: Let $N_m$ be the number of particles at site $m=1,\dots,M$.

172: A migration process (or dynamical urn model) is defined by the following

173: Markovian dynamics:

174: (i) a departure (source) site $d$ is chosen at random,

175: uniformly among the $M$ sites of the system;

176: (ii) an arrival (target) site $a$ is chosen among the neighbors of $d$.

177: To be specific, the right neighbor ($a=d+1$) is chosen with probability $p$,

178: whereas the left neighbor ($a=d-1$) is chosen with

179: the complementary probability $q=1-p$;

180: (iii) a particle is transferred from site $d$ to site $a$

181: at a rate $W_{k,l}$ which only depends on the occupations

182: $k=N_d$ and $l=N_a$ of the two sites involved.

183: Of course, one has

184: \beq

185: W_{0,l}=0,

186: \label{zcons}

187: \eeq

188: since no particle can be removed from an empty site.

189: 

190: A one-dimensional migration process

191: is therefore entirely defined by the bias $p$ and the rate $W_{k,l}$.

192: A natural question is the following:

193: {\it What are the conditions on the bias~$p$ and

194: the rate $W_{k,l}$ for a one-dimensional migration process

195: to have a stationary-state product measure?}

196: This question was first addressed in~\cite{cocozza}

197: (for a review, see~\cite{lux}).

198: 

199: In this context, a stationary-state product measure means

200: that the probability of any configuration of the system

201: in its stationary state has the form

202: \beq

203: P(N_1,\dots,N_M)=\frac{1}{Z_{M,N}}\;p_{N_1}\dots p_{N_M}

204: \;\delta(N_1+\cdots+N_M,N),

205: \label{pm}

206: \eeq

207: where the factors $p_k$ are arbitrary,

208: and the partition function $Z_{M,N}$ reads

209: \beq

210: Z_{M,N}=\sum_{N_1,\dots,N_M}p_{N_1}\dots p_{N_M}\;\delta(N_1+\cdots+N_M,N).

211: \eeq

212: 

213: The answer to the above question is as follows:

214: 

215: \begin{itemize}

216: \item

217: For symmetric dynamics, i.e., when $p=1/2$,

218: the stationary state has a product measure characterized by the factor $p_k$,

219: if the rate $W_{k,l}$ obeys the condition

220: \beq

221: p_{k+1}p_lW_{k+1,l}=p_kp_{l+1}W_{l+1,k}.

222: \label{m1}

223: \eeq

224: The resulting stationary state is an equilibrium state.

225: The relation~(\ref{m1}) expresses detailed balance

226: with respect to this equilibrium state.

227: 

228: \item

229: For asymmetric dynamics, i.e., when $p\neq1/2$,

230: the stationary-state has a product measure

231: if the rate $W_{k,l}$ obeys both~(\ref{m1}) and the following condition:

232: \beq

233: W_{k,l}-W_{k,0}=W_{l,k}-W_{l,0}.

234: \label{m2}

235: \eeq

236: The stationary state is a genuine nonequilibrium steady state.

237: The condition~(\ref{m1}) does not express detailed balance any longer,

238: albeit a weaker property~\cite{lux},

239: referred to as pairwise balance~\cite{pairw}.

240: \end{itemize}

241: 

242: The partition function $Z_{M,N}$ associated with the product measure~(\ref{pm})

243: can be rewritten, using an integral representation of

244: the Kronecker delta function, as

245: \beq

246: Z_{M,N}=\oint\frac{\d z}{2\pi\i z^{N+1}}\,P(z)^M,

247: \label{contour}

248: \eeq

249: where the generating series of the factors $p_k$ reads

250: \beq

251: P(z)=\sum_{k\ge0}p_kz^k.

252: \eeq

253: The product measure~(\ref{pm}) is therefore characterized by $M$, $N$,

254: and the factor $p_k$ or, equivalently, by the generating function $P(z)$.

255: For a homogeneous system in the thermodynamic limit,

256: where $M$ and $N$ are simultaneously large,

257: with a fixed density $\rho=N/M$ of particles per site,

258: the contour integral in~(\ref{contour}) can be evaluated

259: by the saddle-point method.

260: The saddle-point value $z$, which is to be identified with the fugacity

261: in the grand canonical ensemble,

262: is related to the density $\rho$ through the equation

263: \beq

264: \frac{z P'(z)}{P(z)}=\rho.

265: \label{col}

266: \eeq

267: The distribution $f_k=\prob\{N_1=k\}$

268: of the occupations of any given site of the system

269: can be derived by summing the probability~(\ref{pm}) over $N_2,\dots,N_M$.

270: We thus obtain

271: \beq

272: f_k=p_k\,\frac{Z_{M-1,N-k}}{Z_{M,N}}.

273: \eeq

274: In the thermodynamic limit, this expression simplifies to

275: \beq

276: f_k=\frac{p_kz^k}{P(z)}.

277: \label{fdef}

278: \eeq

279: 

280: \subsection{The example of the zero-range process}

281: 

282: In the present context, the ZRP appears as the special

283: case of a migration process where

284: the rate $W_{k,l}$ only depends on the occupation of the departure site:

285: \beq

286: W_{k,l}=u_k,

287: \label{zrpdef}

288: \eeq

289: with $u_0=0$, by virtue of~(\ref{zcons}).

290: The condition~(\ref{m2}) is then automatically satisfied,

291: irrespective of the bias $p$.

292: Equation~(\ref{m1}) yields the following relation between the rate~$u_k$

293: and the factor $p_k$:

294: \beq

295: p_k=u_{k+1}p_{k+1},

296: \label{urel}

297: \eeq

298: up to a multiplicative constant,

299: which we set equal to unity by an appropriate choice of time unit.

300: The corresponding factor $p_k$

301: can be expressed in terms of the rate $u_k$ as follows:

302: \beq

303: p_0=1,\qquad p_k=\frac{1}{u_1\dots u_k}\qquad(k\ge1).

304: \label{pzrp}

305: \eeq

306: Reciprocally, to a given stationary-state product measure

307: characterized by the factor~$p_k$, there corresponds a unique ZRP dynamics

308: (up to a choice of time unit), whose rate reads

309: \beq

310: u_k=\frac{p_{k-1}}{p_k}\qquad(k\ge1).

311: \label{udef}

312: \eeq

313: 

314: \subsection{Characterization of migration processes with stationary-state

315: product measure}

316: 

317: We now give an explicit characterization of the migration processes

318: which admit a stationary-state product measure.

319: 

320: For symmetric dynamics, and for a given factor $p_k$,

321: the most general form of the rate obeying~(\ref{m1}) reads

322: \beq

323: W_{k,l}=\frac{p_{k-1}}{p_k}\,S_{k-1,l}\qquad(k\ge1),

324: \label{wsym}

325: \eeq

326: where the ratio $p_{k-1}/p_k$ is nothing but the rate $u_k$

327: of the corresponding ZRP, given by~(\ref{udef}),

328: whereas~$S_{k,l}$ is a symmetric function of $k$ and~$l$:

329: \beq

330: S_{k,l}=S_{l,k}.

331: \eeq

332: Besides the factor~$p_k$, the rate $W_{k,l}$ depends on an arbitrary

333: symmetric function~$S_{k,l}$ of {\it two} indices.

334: 

335: For asymmetric dynamics, and for a given factor~$p_k$,

336: the most general solution of~(\ref{m1}) and~(\ref{m2}) is determined

337: by the one-dimensional array of rates $\alpha_k=W_{k,0}$~\cite{kls3}.

338: Note that $\alpha_k$ is the rate at which an empty site is refilled,

339: by receiving one particle from a non-empty neighboring site

340: containing $k\ge1$ particles, and that~(\ref{zcons}) implies $\alpha_0=0$.

341: The above property can be shown as follows.

342: Consider all the indices $k$ and $l$ for a fixed value of the sum $k+l=n$,

343: and introduce the quantities

344: \beq

345: A_k=p_kp_{n-k}W_{k,n-k}\qquad(k=0,\dots,n).

346: \eeq

347: Equations~(\ref{m1}) and~(\ref{m2}) respectively become

348: \beq

349: A_k=A_{n+1-k},\qquad A_k-p_kp_{n-k}\alpha_k=A_{n-k}-p_kp_{n-k}\alpha_{n-k}.

350: \eeq

351: Combining these two equations yields

352: \beq

353: A_{k+1}-A_k=p_kp_{n-k}(\alpha_{n-k}-\alpha_k).

354: \eeq

355: The solution of this inhomogeneous difference equation

356: with initial value $A_0=0$ reads

357: \beq

358: A_k=\sum_{m=1}^kp_{k-m}p_{n-k+m}(\alpha_{n-k+m}-\alpha_{k-m}).

359: \eeq

360: We are thus left with the following expression for the rate $W_{k,l}$:

361: \beq

362: W_{k,l}=\frac{1}{p_kp_l}\sum_{m=1}^kp_{k-m}p_{l+m}(\alpha_{l+m}-\alpha_{k-m}).

363: \label{wasym}

364: \eeq

365: Besides the factor $p_k$,

366: the rate $W_{k,l}$ depends on an arbitrary function

367: $\alpha_k$ of {\it one} index.

368: 

369: \subsection{Definition of the target process}

370: 

371: We define the target process as the migration process where the rate

372: \beq

373: W_{k,l}=(1-\delta_{k,0})v_l

374: \label{tpdef}

375: \eeq

376: essentially depends on the occupation $l$ of the arrival (target) site.

377: Note the dual character of this definition with respect to the definition of

378: the ZRP, in that the roles of the departure and arrival sites

379: have been interchanged.

380: There is, however, a key difference between the two models,

381: coming from the presence of the constraint~(\ref{zcons}).

382: The latter,

383: which explicitly enters~(\ref{tpdef}) through the factor $(1-\delta_{k,0})$,

384: implies that the rate of the target process actually also bears some

385: dependence on the occupation $k$ of the departure site,

386: as it is constrained to vanish if $k=0$.

387: For the ZRP the same constraint does not

388: change the fact that the rate~$u_k$ only depends on the departure site;

389: it just imposes $u_0=0$.

390: 

391: For symmetric dynamics,

392: the target process always has a stationary-state product measure,

393: for any choice of the rate $v_k$.

394: Equation~(\ref{m1}) yields the following relation between the rate $v_k$

395: and the factor $p_k$:

396: \beq

397: p_{k+1}=v_kp_k,

398: \label{vrel}

399: \eeq

400: up to a multiplicative constant,

401: which is again set to unity by an appropriate choice of time unit.

402: The factor $p_k$ of the stationary-state measure

403: can thus be expressed in terms of the rate $v_k$ as follows:

404: \beq

405: p_0=1,\qquad p_k=v_0\dots v_{k-1}\qquad(k\ge1).

406: \label{pvdef}

407: \eeq

408: Reciprocally, to a given stationary-state product measure,

409: characterized by the factor~$p_k$,

410: there corresponds a unique symmetric target process

411: (up to a choice of time unit), whose rate reads

412: \beq

413: v_k=\frac{p_{k+1}}{p_k}.

414: \label{vdef}

415: \eeq

416: 

417: Equations~(\ref{urel}) and~(\ref{vrel}) show that the ZRP with rate $u_k$

418: and the target process with rate $v_k$ have the same stationary-state

419: product measure, i.e., the same factor $p_k$, if the rates obey

420: \beq

421: v_k=\frac{1}{u_{k+1}},

422: \label{uvdual}

423: \eeq

424: up to a multiplicative constant.

425: A target process and a ZRP related by this condition

426: are hereafter named {\it dual} to each other.

427: 

428: It is interesting to consider a more general class of migration processes,

429: where the rate has the form

430: \beq

431: W_{k,l}=(1-\delta_{k,0})u_k\,v_l.

432: \label{wgen}

433: \eeq

434: For symmetric dynamics, this model again has a stationary-state product

435: measure for any choice of $u_k$ and $v_l$.

436: Equation~(\ref{m1}) indeed yields

437: \beq

438: u_{k+1}p_{k+1}=v_kp_k,

439: \label{genrel}

440: \eeq

441: up to a multiplicative constant.

442: The factor $p_k$

443: can thus be expressed in terms of the $u_k$ and $v_k$ as follows:

444: \beq

445: p_0=1,\qquad p_k=\frac{v_0\dots v_{k-1}}{u_1\dots u_k}\qquad(k\ge1).

446: \eeq

447: This expression shows that the factor $p_k$

448: only depends on the ratio $v_k/u_{k+1}$.

449: This class of processes interpolates between the ZRP and the target process,

450: which are respectively recovered as the special cases where $v_l=1$ and $u_k=1$.

451: We finally notice that the rate~(\ref{tpdef}) of the target process,

452: and more generally the rate~(\ref{wgen}), is of the form~(\ref{wsym}),~with

453: \beq

454: S_{k,l}=v_kv_l.

455: \eeq

456: 

457: For asymmetric dynamics, the condition~(\ref{m2}) is very stringent.

458: The target process has a stationary-state product measure

459: if and only if the rate $v_k$ only assumes two values,

460: according to whether $k$ is zero or~not:

461: \beq

462: v_k=\left\{\matrix{v_0\hfill&(k=0),\cr v\hfill&(k\ge1).}\right.%}

463: \label{two}

464: \eeq

465: This also holds for the more general process defined by the rate~(\ref{wgen}).

466: 

467: For a generic asymmetric target process,

468: where the rate $v_k$ is not of the form~(\ref{two}),

469: the stationary-state measure is not a product measure.

470: It is not known explicitly, and can be expected to be a non-trivial

471: correlated measure in general.

472: 

473: \subsection{The `canonical target process' considered in this work}

474: 

475: The condensation phenomenon in the ZRP is usually investigated

476: using the rate~\cite{evans1,wis1,cg,gross}:

477: \beq

478: u_k=1+\frac{b}{k}\qquad(k\ge1),

479: \label{ucan}

480: \eeq

481: where the control parameter $b$ is a measure of the strength of interactions.

482: The minimality and exemplarity of this choice of rate suggest to call this

483: model the `canonical ZRP for condensation', or `canonical ZRP', for short.

484: 

485: Throughout the following we focus our attention onto

486: the target process dual to the canonical ZRP.

487: Its rate $v_k$ is therefore related to the rate~(\ref{ucan})

488: by the duality relation~(\ref{uvdual}).

489: We thus obtain

490: \beq

491: v_k=\frac{k+1}{k+b+1}\qquad(k\ge0).

492: \label{vcan}

493: \eeq

494: We name this process the `canonical target process'.

495: 

496: For symmetric dynamics,

497: it has already been shown above that the target process has the same

498: stationary-state product measure as the ZRP.

499: It therefore exhibits the same condensation transition.

500: We now give a brief reminder of the properties of this stationary-state measure.

501: 

502: In the absence of interactions ($b=0$),

503: the rate reads $u_k=v_k=1$, irrespective of $k$.

504: We have therefore $p_k=1$, so that $P(z)=1/(1-z)$.

505: The fugacity $z$ and the density $\rho$ are related by

506: \beq

507: z=\frac{\rho}{\rho+1},\qquad\rho=\frac{z}{1-z}.

508: \eeq

509: The distribution of the occupations~(\ref{fdef}) is a geometric distribution:

510: \beq

511: f_k=(1-z)z^k.

512: \label{fexpo}

513: \eeq

514: 

515: In the situation where $b$ is positive,

516: the rate $u_k$ is a decreasing function of the occupation $k$,

517: so that particles hop less easily out of more occupied sites.

518: Accordingly, the rate $v_k$ is an increasing function of $k$,

519: hence particles hop preferentially towards more occupied sites.

520: The rate~(\ref{ucan}) or~(\ref{vcan}) therefore corresponds to attractive

521: interactions between particles.

522: The model exhibits a trend toward segregation,

523: which leads to a thermodynamical condensation transition

524: if $b$ is strong enough.

525: It turns out that many characteristics of this condensation transition

526: are universal~\cite{loan,evans1,wis1,cg,gross,evans2,glcond}:

527: they only depend on the asymptotic behavior of the rate

528: $u_k$ (or $v_k$) at large $k$, i.e., essentially on the value of $b$.

529: 

530: For the choice of rate~(\ref{ucan}) or~(\ref{vcan}),

531: the factor $p_k$ of the stationary-state product measure

532: (see~(\ref{pzrp}) or~(\ref{pvdef})) reads

533: \beq

534: p_k=\frac{\Gamma(b+1)\,k!}{\Gamma(k+b+1)}=b\int_0^1(1-u)^{b-1}u^k\,\d u,

535: \label{pcan}

536: \eeq

537: so that

538: \beq

539: P(z)=b\int_0^1\frac{(1-u)^{b-1}}{1-zu}\,\d u.

540: \eeq

541: The factor $p_k$ falls off as a power law with exponent $b$ at large $k$:

542: \beq

543: p_k\approx\frac{\Gamma(b+1)}{k^b}.

544: \eeq

545: The canonical ZRP has a condensation transition

546: in the thermodynamic limit whenever the first moment of the factor $p_k$,

547: \beq

548: \rho_c=\sum_{k\ge1}k\,p_k=\frac{P'(1)}{P(1)},

549: \eeq

550: is convergent.

551: We have $P(1)=b/(b-1)$ and $P'(1)=b/((b-1)(b-2))$,

552: so that the critical density $\rho_c$ is finite for $b>2$, and reads

553: \beq

554: \rho_c=\frac{P'(1)}{P(1)}=\frac{1}{b-2}.

555: \eeq

556: This critical density separates a fluid phase and a condensed phase:

557: 

558: \begin{itemize}

559: \item

560: At the {\it critical density} ($\rho=\rho_c$, $z=1$),

561: the occupation distribution reads

562: \beq

563: f_k=\frac{p_k}{P(1)}=\frac{b-1}{b}\,p_k

564: \label{fkc}

565: \eeq

566: (see~(\ref{fdef})).

567: In particular the fraction of empty sites is

568: \beq

569: f_0=\frac{b-1}{b},

570: \label{f0c}

571: \eeq

572: whereas the distribution falls off as a power law for large occupations:

573: \beq

574: f_k\approx\frac{(b-1)\Gamma(b)}{k^b}.

575: \label{fktail}

576: \eeq

577: The statics and the dynamics of the model

578: exhibit many features of critical phenomena, including scaling and universality.

579: 

580: \item

581: In the {\it fluid phase} ($\rho<\rho_c$, $z<1$),

582: the occupation distribution $f_k$ falls off exponentially.

583: 

584: \item

585: In the {\it condensed phase} $(\rho>\rho_c)$, for a large finite system,

586: the particles are arranged so as to form a uniform critical background

587: and a macroscopic condensate, typically occupying one single site

588: and consisting of $N-M\rho_c=M(\rho-\rho_c)$ excess particles.

589: \end{itemize}

590: 

591: The stationary-state measure of the asymmetric canonical target process is not

592: a product measure in the presence of interactions,

593: i.e., for any non-zero value of the parameter~$b$,

594: because the rate~(\ref{vcan}) is not of the form~(\ref{two}).

595: This absence of a product measure also holds for

596: the asymmetric target process on higher-dimensional lattices.

597: The derivation of the conditions~(\ref{m1}),~(\ref{m2})

598: given in~\cite{lux} could indeed easily be extended to the case

599: of a biased one-particle dynamics on any higher-dimensional lattice.

600: The rest of this paper is devoted to a detailed investigation of this model.

601: The study of the one-dimensional totally asymmetric model

602: is presented in Sections~\ref{1D1} and~\ref{1D2},

603: whereas Section~\ref{2D} is devoted to a maximally asymmetric form of the model

604: on the two-dimensional square lattice.

605: 

606: \section{One-dimensional target process: theoretical analysis}

607: \label{1D1}

608: 

609: In this section, we consider the target process defined by the rate~(\ref{vcan})

610: in the totally asymmetric one-dimensional case ($p=1$).

611: The situation of most physical interest will turn out to be the regime

612: of a high density.

613: 

614: We begin the analysis of the model by exploring the consequences of

615: the existence of a conserved current.

616: Consider a large finite system, with periodic boundary conditions.

617: In the stationary state,

618: the mean current of particles through the system is conserved:

619: it assumes the same value $J$ through every bond.

620: The current through the bond between sites $m$ and $m+1$

621: is given by the mean value of the corresponding rate:

622: \beq

623: J=\mean{W_{N_m,N_{m+1}}}.

624: \label{jdef}

625: \eeq

626: The existence of this conserved current is expected to ensure

627: some degree of homogeneity of the stationary state.

628: 

629: To illustrate the method, let us first consider the totally asymmetric ZRP,

630: with rate~(\ref{ucan}).

631: The current therefore reads

632: \beq

633: J=\mean{u_{N_m}}.

634: \label{jzrpdef}

635: \eeq

636: Consider a typical configuration in the condensed phase ($\rho>\rho_c$).

637: The site where the condensate is located contains a macroscopic number of

638: particles, so that~(\ref{jzrpdef}) yields $J=1$

639: for the bond to the right of the condensate,

640: up to a negligible finite-size correction of order~$1/M$.

641: For all the other bonds, the relations~(\ref{fdef}) and~(\ref{urel}) yield

642: \beq

643: J=\sum_{k\ge1}\underbrace{u_k\,f_k}_{\ds zf_{k-1}}=z,

644: \eeq

645: where $z$ is the fugacity.

646: Equating the two above expressions for the current,

647: we recover well-known results for the totally asymmetric ZRP,

648: i.e., $z=z_c=1$ and $J=1$ throughout the condensed phase.

649: 

650: Let us now turn to the totally asymmetric target process,

651: with rate~(\ref{vcan}).

652: Equation~(\ref{jdef}) for the current now reads

653: \beq

654: J=\mean{(1-\delta_{N_m,0})v_{N_{m+1}}}.

655: \label{jtpdef}

656: \eeq

657: At variance with~(\ref{jzrpdef}),

658: this expression involves the occupations of two consecutive sites.

659: 

660: The high-density regime turns out to be the situation of most physical interest.

661: In this regime, at least some of the sites must have large occupations.

662: We are therefore led to distinguish between two types of sites:

663: \begin{itemize}

664: \item

665: B-sites (B for {\it big}), whose occupation is large, of the order of $\rho$.

666: \item

667: S-sites (S for {\it small}), whose occupation is small and fluctuating.

668: \end{itemize}

669: 

670: This distinction will be kept at a heuristic level throughout the following.

671: A typical high-density configuration therefore consists of

672: four types of bonds: BB, BS, SB, and~SS.

673: The typical values of the current in each type of bond

674: obey the following inequalities in the high-density limit:

675: \beq

676: J_\SS<\matrix{J_\SB\cr J_\BS}<J_\BB.

677: \label{jineq}

678: \eeq

679: Each factor in~(\ref{jtpdef}) is indeed less than unity,

680: and approaches unity in the limit where the involved occupation goes to

681: infinity.

682: In the high-density limit we have therefore $J_\BB=1$,

683: whereas $J_\BS=\mean{v_{N_m}}$ and $J_\SB=\mean{1-\delta_{N_m,0}}$,

684: where $m$ is a typical S-site, and~$J_\SS$ is smaller

685: than the last two expressions.

686: 

687: Thanks to the inequalities~(\ref{jineq})

688: the presence of a stable isolated condensate is excluded.

689: This would indeed correspond to a current profile where the two bonds

690: on either side of the condensate carry currents $J=J_\SB$ and $J=J_\BS$

691: which are significantly higher than the background current

692: $J=J_\SS$ of all the other bonds.

693: This non-uniform current distribution with a point defect would have the effect

694: that the condensate would soon dissolve into the background.

695: The inequalities~(\ref{jineq}) actually only leave out two scenarios

696: for typical stationary-state configurations in the high-density limit.

697: Both scenarios, described below and illustrated in Figure~\ref{figunialt},

698: correspond to spatially homogeneous phases.

699: We are therefore led to predict that the asymmetric target process

700: has no condensation transition at any finite density in one dimension.

701: 

702: \begin{figure}[!tb]

703: \begin{center}

704: \includegraphics[angle=90,height=4truecm]{figuni.eps}

705: {\hskip 60pt}

706: \includegraphics[angle=90,height=4truecm]{figalt.eps}

707: \caption{\small

708: Typical configurations in the two possible scenarios

709: for the stationary state of the asymmetric one-dimensional target process

710: at high density.

711: Left: uniform configuration.

712: Right: alternating configuration.}

713: \label{figunialt}

714: \end{center}

715: \end{figure}

716: 

717: \begin{itemize}

718: \item {\it Uniform scenario.}

719: In the uniform scenario, shown in the left panel of Figure~\ref{figunialt},

720: typical configurations are entirely made of B-sites,

721: whose typical occupation is around the density $\rho$.

722: These configurations carry a current $J=J_\BB$, i.e.,

723: \beq

724: J=1

725: \eeq

726: in the high-density limit, up to a correction of order $1/\rho$.

727: 

728: \item {\it Alternating scenario.}

729: In the alternating scenario, shown in the right panel of Figure~\ref{figunialt},

730: typical configurations are alternating, i.e., they have the form BSBSBSBS...

731: 

732: The asymptotic value of the current $J=J_\BS=J_\SB$

733: through a perfect alternating structure

734: in the high-density limit can be evaluated as follows.

735: Most particles belong to B-sites, whose typical occupation is very high,

736: around $2\rho$.

737: Every B-site therefore acts as a reservoir,

738: so that the occupations of the S-sites evolve independently from each other.

739: The master equations for the occupation distribution $f_k(t)$

740: of any given S-site read

741: \beqa

742: \frad{\d f_k}{\d t}&=&f_{k+1}+v_{k-1}f_{k-1}-(1+v_k)f_k\qquad(k\ge1),

743: \nonumber\\

744: \frad{\d f_0}{\d t}&=&f_1-v_0f_0.

745: \eeqa

746: The stationary-state solution of these equations is such that

747: $f_{k+1}=v_kf_k$.

748: It is therefore proportional to the factor $p_k$.

749: The properly normalized solution is given by~(\ref{fkc}).

750: To sum up, the occupations of the S-sites are independent variables,

751: whose distribution coincides with the critical occupation distribution

752: of the dual ZRP or, equivalently, of the symmetric target process.

753: We have

754: \beq

755: J=\sum_{k\ge0}\underbrace{v_k\,f_k}_{\ds f_{k+1}}=1-f_0.

756: \eeq

757: Using (\ref{f0c}), the current in the high-density limit reads

758: \beq

759: J=\frac{1}{b}.

760: \label{jcrit}

761: \eeq

762: This alternating scenario holds a priori whenever $b>1$.

763: The latter condition corresponds to $P(1)$ being finite,

764: so that~(\ref{fkc}) is a properly normalized probability distribution.

765: \end{itemize}

766: 

767: In order to have a hint on which of the two above scenarios is preferred,

768: it is interesting to first consider the simple example of a system of two sites.

769: In this case, there is only one kind of move for the particles,

770: namely from one site to the other, and only one degree of freedom,

771: the occupation $N_1=k$ of site number~1.

772: Indeed the occupation of site number~2 reads $N_2=N-k$.

773: The stationary-state occupation distribution~$f_k$ is

774: clearly equal to the equilibrium product measure

775: \beq

776: f_k=\frac{p_kp_{N-k}}{Z_{2,N}}\approx\frac{C_N}{k^b(N-k)^b}.

777: \eeq

778: When the number $N$ of particles is large,

779: the above distribution exhibits a segregation phenomenon

780: for any positive value of $b$ (see~\cite{glcond} for a more detailed analysis,

781: including an asymptotic analysis of the amplitude $C_N$).

782: The most probable configurations are those where almost all the

783: particles are at one site, i.e., either $k\ll N$ or $N-k\ll N$.

784: This simple example confirms that the target model, just as the ZRP,

785: manifests a trend toward segregation at high density.

786: It therefore suggests that the preferred scenario

787: is that of an alternating structure.

788: 

789: This picture can be corroborated and made more quantitative

790: by means of the following dynamical stability analysis of the uniform situation.

791: For the sake of generality, in this part of the analysis

792: we deal with the partially asymmetric target process with bias $p$.

793: Consider a configuration of the uniform scenario.

794: All the sites have very high local densities $\mean{N_m}=\rho_m(t)$.

795: These local densities obey the exact rate equation

796: \beqa

797: \frac{\d\rho_m}{\d t}

798: &=&p\mean{(1-\delta_{N_{m-1},0})v_{N_m}}

799: +q\mean{(1-\delta_{N_{m+1},0})v_{N_m}}\nonumber\\

800: &-&p\mean{(1-\delta_{N_m,0})v_{N_{m+1}}}

801: -q\mean{(1-\delta_{N_m,0})v_{N_{m-1}}}.

802: \label{rhorat}

803: \eeqa

804: In the high-density regime it is legitimate

805: to simplify the above equation in several respects.

806: The probability that a site is empty is negligible,

807: whereas $\mean{v_{N_m}}\approx v_{\mean{N_m}}\approx1-b/\mean{N_m}$.

808: We are thus left with

809: \beq

810: \frac{\d\rho_m}{\d t}\approx b

811: \left(\frac{p}{\rho_{m+1}}+\frac{q}{\rho_{m-1}}-\frac{1}{\rho_m}\right).

812: \eeq

813: Let us furthermore assume that the density profile is close

814: to being constant, i.e.,

815: \beq

816: \rho_m=\rho+\dr_m,

817: \eeq

818: with $\dr_m\ll\rho$.

819: The rate equation~(\ref{rhorat}) can then be linearized as

820: \beq

821: \frac{\d\,\dr_m}{\d t}\approx\frac{b}{\rho^2}

822: \left(\dr_m-p\,\dr_{m+1}-q\,\dr_{m-1}\right).

823: \eeq

824: The component of the density profile at wavevector $K$

825: therefore grows exponentially in time as $\hdr(K)\sim\exp(\sigma(K)\,t)$,

826: where the characteristic rate $\sigma(K)$ is given by the dispersion relation

827: \beq

828: \sigma(K)=\frac{b}{\rho^2}\left(1-p\,\e^{-\i K}-q\,\e^{\i K}\right).

829: \eeq

830: The growth rate $s(K)$ is given by the real part of $\sigma(K)$, which reads

831: \beq

832: s(K)=\frac{2b}{\rho^2}\;\sin^2\frac{K}{2},

833: \label{grow}

834: \eeq

835: irrespective of the bias $p$.

836: The expression~(\ref{grow}) for the growth rate of fluctuations

837: around the uniform situation is manifestly positive

838: for all values of the wavevector $K$.

839: 

840: The uniform scenario is therefore fully linearly unstable.

841: As a consequence, the alternating scenario is the preferred one.

842: Furthermore, the alternating structure is already appearing

843: as the most favored one within the stability analysis.

844: The most unstable mode indeed corresponds to $K=\pi$,

845: i.e., an alternating density modulation of the form $\dr_m\sim(-1)^m$.

846: A uniform high-density initial configuration

847: is therefore expected to smoothly relax to an alternating one

848: by the dynamics of the asymmetric target process.

849: This observation deserves, however, to be complemented with

850: the following caveat.

851: Consider the symmetric target process, corresponding to $p=1/2$.

852: The stability analysis still leads to the expression~(\ref{grow}),

853: so that the alternating mode is still the most favored one.

854: On the other hand, the stationary state is known to be described

855: by a product measure, and especially to have a single condensate.

856: This suggests that the relaxation dynamics of a uniform

857: high-density initial configuration will exhibit two stages:

858: first, a rather fast relaxation to an intermediate structure with

859: alternating fluctuations,

860: then, a coarsening evolution of the system toward its true fate,

861: by the merging of the excess particles into fewer and fewer condensate

862: precursors.

863: This two-stage relaxation should also hold for the dual ZRP

864: with a uniform high-density initial configuration.

865: The dynamical stability analysis of the uniform situation

866: in the ZRP indeed yields exactly the same expression~(\ref{grow}),

867: again irrespective of the bias $p$.

868: 

869: To close up, we mention that the expression~(\ref{grow})

870: also yields some hints on the time scales involved in the model.

871: The characteristic time of the most unstable mode, $T_\loc=1/s(\pi)$,

872: gives an estimate of the local relaxation time of the alternating structure,

873: at the spatial scale of two consecutive sites.

874: On the other hand,

875: the long-distance behavior of the dynamics is {\it antidiffusive}.

876: We have indeed formally $s(K)\approx -DK^2$ as $K\to0$,

877: with a small negative diffusion coefficient $D=-b/(2\rho^2)$.

878: For a large but finite system made of $M$ sites,

879: with periodic boundary conditions,

880: the longest characteristic time, $T_\gl=1/s(2\pi/M)$,

881: gives an estimate of the global relaxation time of the structure as a whole.

882: In the high-density regime, the characteristic times thus defined scale as

883: \beq

884: T_\loc\approx\frac{\rho^2}{2b},\qquad T_\gl\approx\frac{\rho^2M^2}{2\pi^2b}.

885: \label{times}

886: \eeq

887: The predicted divergence of both characteristic times with density

888: provides an a posteriori confirmation that the high-density regime

889: is indeed the most interesting one.

890: 

891: \section{One-dimensional target process: numerical results}

892: \label{1D2}

893: 

894: In this section we complement our analysis

895: of the target process in the asymmetric one-dimensional case,

896: by means of numerical simulations and scaling arguments.

897: For definiteness we restrict the study to the totally asymmetric

898: situation ($p=1$).

899: Furthermore, we set once for all $b=4$.

900: We successively consider features of the transient dynamics,

901: of the stationary-state measure, and of the stationary-state dynamics.

902: The main focus is on the scaling behavior of quantities of interest

903: in the high-density regime.

904: Two types of initial conditions are considered:

905: \begin{itemize}

906: \item Deterministic initial condition:

907: the occupations of all the sites are set equal to $N_m=\rho$

908: (provided the density $\rho$ is an integer).

909: \item Random initial condition:

910: the occupations $N_m$ are drawn independently at random

911: from the geometric distribution~(\ref{fexpo}) at density $\rho$.

912: \end{itemize}

913: 

914: We start by investigating the early stage of the dynamics,

915: where relatively fast rearrangements of particles

916: bring the system to a locally stationary state.

917: This stage of the relaxation dynamics can be monitored

918: by means of a local probe at one site.

919: We choose the reduced second moment of the occupations:

920: \beq

921: K(t)=\frac{1}{\rho^2}\sum_{k\ge0}k^2f_k(t),

922: \label{kdef}

923: \eeq

924: where $f_k(t)$ is the time-dependent distribution of the occupations.

925: The initial values of this quantity are

926: $K(0)=1$ for the deterministic initial condition,

927: and $K(0)=(2\rho+1)/\rho$ for the random one.

928: 

929: Figure~\ref{figkt} shows a plot of $K(t)$,

930: measured by means of a numerical simulation, against time $t$,

931: for $\rho=10$, and a deterministic and a random initial condition.

932: Each series of data are obtained by averaging over sufficiently many

933: histories in order to obtain a smooth signal

934: ($10^4$ histories of a system of $10^4$ sites in this case).

935: The data exhibit a rather fast rise from their initial values,

936: and converge to a common limiting value, $K\approx2.82$,

937: extracted from data for much longer times, and shown as a dashed line.

938: The data for the deterministic initial condition (lower curve)

939: increase as a monotonic function of time and present a rather sharp shoulder,

940: whereas those for the random initial condition (upper curve)

941: exhibit a non-monotonic behavior with a very flat maximum.

942: 

943: \begin{figure}[!tb]

944: \begin{center}

945: \includegraphics[angle=90,height=6truecm]{figkt.eps}

946: \caption{\small

947: Plot of the reduced second moment $K(t)$ of the occupation distribution

948: in the one-dimensional fully asymmetric target process against time~$t$,

949: for $b=4$ and $\rho=10$.

950: Lower curve: deterministic initial condition.

951: Upper curve: random initial condition.

952: Horizontal dashed line: common limiting value $K\approx2.82$.}

953: \label{figkt}

954: \end{center}

955: \end{figure}

956: 

957: The characteristic time of the rise observed in $K(t)$

958: gives a measure of the local relaxation time.

959: More precisely, we define the local time~$T_\loc$

960: by the condition that~$K(t)$ is near the middle of its rise,

961: i.e., $K(T_\loc)=2$ for the deterministic initial condition,

962: and $K(T_\loc)=2.5$ for the random initial condition.

963: Figure~\ref{figtloc} shows a plot of the numerical values

964: of the local time so defined, divided by $\rho$, against $\rho$,

965: for both types of initial conditions.

966: Here and in subsequent figures,

967: statistical errors are comparable to the symbol size.

968: The least-squares fits of the two series of data suggest a growth of the form

969: \beq

970: T_\loc\approx A\rho^2+B\rho

971: \label{tlinquad}

972: \eeq

973: in the high-density regime of interest.

974: The amplitudes~$A$ and $B$ depend on the initial condition.

975: The least-squares fits shown on the plot

976: yield $A\approx0.095$ for a deterministic initial condition

977: and $A\approx0.039$ for a random initial condition.

978: These numbers, and especially the first one,

979: are comparable to the rough estimate coming from~(\ref{times}),

980: i.e., $A=1/(2b)=1/8=0.125$.

981: 

982: \begin{figure}[!tb]

983: \begin{center}

984: \includegraphics[angle=90,height=6truecm]{figtloc.eps}

985: \caption{\small

986: Plot of the local relaxation time $T_\loc$, divided by $\rho$,

987: against density~$\rho$.

988: Upper data (empty symbols): deterministic initial condition.

989: Lower data (full symbols): random initial condition.

990: Straight lines: least-squares fits with respective slopes 0.095 and 0.039.}

991: \label{figtloc}

992: \end{center}

993: \end{figure}

994: 

995: The initial condition is irrelevant

996: for what concerns later stages of the dynamics,

997: which correspond to the emergence of global features of the stationary state.

998: Hereafter we choose to work with a random initial condition.

999: Before we turn to an analysis of these late stages,

1000: it is worth taking a glance at the spatial structure

1001: of the stationary state in the high-density regime.

1002: 

1003: Figure~\ref{figtrack} shows a typical stationary

1004: occupation profile for $\rho=50$.

1005: In order to better reveal the alternating structure predicted

1006: in Section~\ref{1D1},

1007: we have plotted $(-1)^mN_m$ against the position $m$ of the site.

1008: The alternating structure clearly emerges from this representation:

1009: domains where the signal is positive (resp.~negative)

1010: correspond to domains where the B-sites are the even (resp.~odd) sites.

1011: 

1012: \begin{figure}[!tb]

1013: \begin{center}

1014: \includegraphics[angle=90,height=6truecm]{figtrack.eps}

1015: \caption{\small

1016: Plot of $(-1)^mN_m$ against the position $m$ of the site,

1017: for a typical stationary-state configuration for $\rho=50$,

1018: emphasizing the alternating structure.}

1019: \label{figtrack}

1020: \end{center}

1021: \end{figure}

1022: 

1023: In order to turn this observation into a quantitative measurement,

1024: let us introduce the concept of {\it defects}.

1025: Roughly speaking, a defect is a site around which the structure is not perfectly

1026: of the alternating form BSBSBSBS...

1027: and a domain is any stretch between two consecutive defects.

1028: More precisely, the site~$m$ is considered as a defect whenever

1029: its occupation $N_m$ is neither a maximum nor a minimum of the density profile.

1030: Equivalently,~$N_m$ is between $N_{m-1}$ and $N_{m+1}$,

1031: i.e., the product $(N_m-N_{m-1})(N_m-N_{m+1})$ is negative.

1032: This definition pinpoints 24 defects

1033: in the configuration shown in Figure~\ref{figtrack}.

1034: This number is slightly above the number of domains visible with

1035: the naked eye, i.e.,~18, because some of the domains are microscopic.

1036: Most defects can be viewed either as a BB sequence or as an SS sequence.

1037: We shall return later on to the dynamics of these defects.

1038: The density of defects $R$, i.e., the mean number of defects

1039: per unit length, reads

1040: \beq

1041: R=\bigmean{\Theta\Bigl(-(N_m-N_{m-1})(N_m-N_{m+1})\Bigr)},

1042: \eeq

1043: where the Heaviside step function on the integers is defined as

1044: \beq

1045: \Theta(n)=\left\{\matrix{

1046: 1\quad\hbox{for}\;\;n>0,\hfill\cr

1047: 0\quad\hbox{for}\;\;n\le0.}\right.%}

1048: \eeq

1049: The inverse of the density of defects,

1050: \beq

1051: \xi=\frac{1}{R},

1052: \eeq

1053: is interpreted as the mean size of a domain or, equivalently,

1054: as the {\it coherence length} of the alternating structure.

1055: 

1056: Let us now return to the late stages of the dynamics,

1057: starting from a random initial condition.

1058: In order to characterize the growth of the alternating structure,

1059: we have measured the time dependence of the mean domain size $\xi(t)=1/R(t)$.

1060: Figure~\ref{figd} shows a plot of~$\xi(t)$,

1061: for various values of the density $\rho$.

1062: For a random initial condition, we have $\xi(0)=3$.

1063: Consider indeed the initial values of $N_{m-1}$, $N_m$, and $N_{m+1}$.

1064: The probability that these three independent random numbers

1065: obey either of the inequalities $N_{m-1}<N_m<N_{m+1}$

1066: or $N_{m-1}>N_m>N_{m+1}$ is equal to 1/6 in the high-density regime

1067: (neglecting the fact that these are integer variables,

1068: which may coincide with a small but nonzero probability).

1069: Hence $R(0)=2\times1/6=1/3$ and $\xi(0)=1/R(0)=3$.

1070: The data are plotted against the reduced time variable $(t/T_\loc)^{1/2}$,

1071: for each value of the density~$\rho$,

1072: where $T_\loc$ is taken from the lower data of Figure~\ref{figtloc}.

1073: The observed common initial linear behavior,

1074: shown as a dashed straight line starting from the known value $\xi(0)=3$,

1075: demonstrates that the mean domain size grows according to the coarsening~law

1076: \beq

1077: \xi(t)\sim(t/T_\loc)^{1/2},

1078: \label{coa}

1079: \eeq

1080: before it saturates to a density-dependent

1081: stationary-state value, simply denoted by $\xi$.

1082: The duration of the coarsening process, before the stationary state is reached,

1083: defines the global relaxation time $T_\gl$ of the problem.

1084: By inverting the relation~(\ref{coa}),

1085: we predict that the latter time grows as $T_\gl\sim T_\loc\,\xi^2$.

1086: This relation between $T_\loc$ and $T_\gl$ is in agreement with~(\ref{times}),

1087: where the stationary-state mean domain size $\xi$ plays the role

1088: of the system size $M$.

1089: 

1090: \begin{figure}[!tb]

1091: \begin{center}

1092: \includegraphics[angle=90,height=6truecm]{figd.eps}

1093: \caption{\small

1094: Plot of the mean domain size $\xi(t)$

1095: against the reduced time $(t/T_\loc)^{1/2}$,

1096: for various values of the density $\rho$, indicated on the curves.

1097: The dashed straight line starting from $\xi(0)=3$

1098: illustrates the coarsening law~(\ref{coa}).}

1099: \label{figd}

1100: \end{center}

1101: \end{figure}

1102: 

1103: We now investigate a few characteristic features

1104: of the nonequilibrium stationary state of the asymmetric target process,

1105: emphasizing their scaling behavior at high density.

1106: We start with the mean domain size~$\xi$.

1107: Its scaling behavior at high density can be predicted by the following argument.

1108: Roughly speaking, a defect can be thought of as a B-site whose occupation

1109: is accidentally as small as that of an S-site, i.e., finite.

1110: Anticipating the scaling law~(\ref{bsca}),

1111: and assuming a linear rise for the scaling function~$F(x)$,

1112: we find that the probability of such an event scales as $1/\rho^2$.

1113: This argument leads to an asymptotic

1114: quadratic growth of the mean domain size of the~form

1115: \beq

1116: \xi\approx\Xi\,\rho^2.

1117: \label{xiquad}

1118: \eeq

1119: 

1120: \begin{figure}[!tb]

1121: \begin{center}

1122: \includegraphics[angle=90,height=6truecm]{figxi.eps}

1123: \caption{\small

1124: Plot of the mean domain size $\xi$ in the stationary state

1125: against density~$\rho$.

1126: Full line: second-degree polynomial fit to the data

1127: with leading coefficient $\Xi=416\times10^{-6}$.}

1128: \label{figxi}

1129: \end{center}

1130: \end{figure}

1131: 

1132: Figure~\ref{figxi} shows a plot of the stationary-state value

1133: of $\xi$ against density $\rho$, for densities up to $\rho=200$.

1134: The second-degree polynomial fit to the data is compatible with

1135: our expectation~(\ref{xiquad}).

1136: The numerical value for the prefactor, $\Xi\approx4\times10^{-4}$,

1137: is however found to be very small.

1138: Partly as a consequence of this smallness,

1139: the data exhibit large corrections to the above asymptotic law

1140: for values of the density accessible to numerical simulations.

1141: At variance with the case of $T_\loc$, plotted in Figure~\ref{figtloc},

1142: we found no way to unambiguously characterize these corrections.

1143: The quadratic growth~(\ref{xiquad}) of~$\xi$

1144: corresponds to a very fast growth of the global relaxation time:

1145: \beq

1146: T_\gl\sim T_\loc\,\xi^2\sim\rho^6.

1147: \eeq

1148: 

1149: \begin{figure}[!tb]

1150: \begin{center}

1151: \includegraphics[angle=90,height=6truecm]{figj.eps}

1152: {\hskip 15pt}

1153: \includegraphics[angle=90,height=6truecm]{figk.eps}

1154: \caption{\small

1155: Plot of stationary values of the current $J$ (left) and of the

1156: reduced second moment~$K$ (right),

1157: against the stationary density of defects $R=1/\xi$.

1158: Symbols: data for densities ranging from $\rho=30$ to 200.

1159: Full lines: polynomial fits, of degree 2 for $J$ and 3 for $K$,

1160: yielding the extrapolated values $J_\infty\approx0.250$

1161: and $K_\infty\approx2.98$.

1162: Dashed line: constrained polynomial fit of degree 3 for $K$,

1163: imposing $K_\infty=3$ (see text).}

1164: \label{figjk}

1165: \end{center}

1166: \end{figure}

1167: 

1168: Figure~\ref{figjk} shows a plot of the stationary-state values

1169: of the current $J$ and of the reduced second moment $K$

1170: of the occupation distribution,

1171: against the stationary-state density of defects $R=1/\xi$.

1172: The second-degree polynomial fit for $J$ yields $J_\infty\approx0.250$,

1173: in excellent quantitative agreement with

1174: the limiting value $J=1/b=1/4$ predicted in~(\ref{jcrit}).

1175: A third-degree polynomial fit for $K$ yields the

1176: limiting value $K_\infty\approx2.98$.

1177: This number is very close to the value $K_\infty=3$ corresponding to the

1178: trial scaling function $F_0(x)$ given in~(\ref{ftrial}).

1179: This proximity suggests that the value $K_\infty=3$ could be exact.

1180: Imposing the constraint $K_\infty=3$ indeed hardly alters

1181: the quality of the fit (dashed line).

1182: 

1183: \begin{figure}[!tb]

1184: \begin{center}

1185: \includegraphics[angle=90,height=6truecm]{figf1.eps}

1186: {\hskip 15pt}

1187: \includegraphics[angle=90,height=6truecm]{figf2.eps}

1188: \caption{\small

1189: Stationary-state occupation distribution $f_k$.

1190: The left panel shows data for $k\ll\rho$ (S-sites).

1191: Empty symbols: plot of $\ln f_k$ against $\ln(k+1)$

1192: for $\rho=50$ (upper data) and 100 (lower data).

1193: Full symbols (labeled ZRP): plot of $\ln (f_k/2)$,

1194: where $f_k$ is the critical distribution of the dual ZRP.

1195: The right panel shows data for $k$ comparable to the density $\rho$ (B-sites).

1196: Full lines: scaling plots of the product $\rho f_k$ against $k/\rho$

1197: for $\rho=50$ (upper data) and 100 (lower data).

1198: Symbols: minimum value of $f_k$.

1199: Dashed line: trial scaling function $F_0(x)$ defined in~(\ref{ftrial}).}

1200: \label{figf}

1201: \end{center}

1202: \end{figure}

1203: 

1204: In complete agreement with the alternating scenario

1205: depicted in Section~\ref{1D1},

1206: the occupation distribution $f_k$ in the stationary state

1207: consists of two distinct components with equal weights,

1208: which respectively describe S-sites and B-sites.

1209: These components are emphasized in Figure~\ref{figf},

1210: where the data for $f_k$ for the same values of the density, $\rho=50$ and 100,

1211: are plotted in two different ways.

1212: The left panel shows a plot of $\ln f_k$ against $\ln(k+1)$

1213: for moderate values of the occupation $k$, up to~20.

1214: These are, roughly speaking, the S-sites.

1215: The data for the smaller values of~$k$

1216: are very close to $1/2$ times the critical occupation distribution

1217: of the dual ZRP or, equivalently, of the symmetric target process,

1218: given by~(\ref{fkc}), also shown on the plot (full symbols).

1219: The range of values of $k$ over which the agreement holds

1220: is observed to get larger for larger densities.

1221: This is a convincing confirmation of the prediction made

1222: in Section~\ref{1D1} that the S-sites are critical in the stationary state.

1223: The right panel shows a scaling plot of the product $\rho f_k$ against $k/\rho$,

1224: for larger values of $k$, comparable to the density $\rho$.

1225: These are, roughly speaking, the B-sites.

1226: The data exhibit a scaling law of the form

1227: \beq

1228: f_k\approx\frac{1}{\rho}\;F\!\left(\frac{k}{\rho}\right).

1229: \label{bsca}

1230: \eeq

1231: The scaling function $F(x)$ obeys the sum rules

1232: \beq

1233: \int_0^\infty F(x)\,\d x=1/2,\quad\int_0^\infty xF(x)\,\d x=1,\quad

1234: \int_0^\infty x^2F(x)\,\d x=K_\infty.

1235: \eeq

1236: The first two equalities express that the fraction of B-sites is $1/2$

1237: and that their mean occupation is $2\rho$, whereas

1238: the third one is a rewriting of the definition of~$K_\infty$.

1239: The scaling function is observed to be rather uniformly well approximated by

1240: the trial scaling function

1241: \beq

1242: F_0(x)=\frac12\,x\,\e^{-x},

1243: \label{ftrial}

1244: \eeq

1245: shown on the right panel of Figure~\ref{figf} as a dashed line.

1246: The scaling function~(\ref{ftrial}) corresponds to $K_\infty=3$.

1247: The closeness of this number to the extrapolated value $K_\infty\approx2.98$

1248: opens up the possibility that the scaling function $F(x)$ is exactly given

1249: by $F_0(x)$.

1250: In any case, the linear rise of the trial function $F_0(x)$

1251: seems to be shared by the true scaling function $F(x)$.

1252: 

1253: Both components of the occupation distribution shown in Figure~\ref{figf},

1254: respectively corresponding to S-sites and B-sites,

1255: are separated by a minimum in the occupation distribution.

1256: In the high-density regime

1257: this minimum takes place for a crossover occupation $k_\star$

1258: such that the estimates $1/k^b$ (see~(\ref{fktail})) and $k/\rho^2$

1259: (assuming a linear rise for the scaling function $F(x)$) are comparable.

1260: We thus obtain

1261: \beq

1262: k_\star\sim\rho^{2/(b+1)},\qquad f_{k_\star}\sim\rho^{-2b/(b+1)}.

1263: \label{xover}

1264: \eeq

1265: These estimates make sense as soon as the crossover occupation obeys

1266: $k_\star\ll\rho$.

1267: We thus recover the condition~$b>1$ for the validity

1268: of the alternating scenario.

1269: The mean occupation $\rho_S$ of the S-sites has the finite asymptotic value

1270: $\rho_c$ for $b>2$,

1271: whereas it scales as $\rho_S\sim k_\star^{2-b}\sim\rho^{2(2-b)/(b+1)}$

1272: in the high-density regime for $1<b<2$.

1273: 

1274: We close up this section

1275: with an investigation of the stationary-state dynamics of defects.

1276: For a large but finite density $\rho$,

1277: there is a small density of defects $R=1/\xi\sim1/\rho^2$ (see~(\ref{xiquad})).

1278: These defects cannot stay immobile.

1279: 

1280: \begin{itemize}

1281: \item

1282: Consider indeed a BB defect, made of two consecutive B-sites.

1283: The current between the two B-sites, $J_\BB=1$, exceeds the mean current

1284: $J=1/b$ through the system.

1285: As a consequence, particles flow from the left B-site of the defect

1286: into the right one at a rate $\omega_\BB=J_\BB-J=(b-1)/b$.

1287: After a time of the order of

1288: \beq

1289: \tau_\BB\approx\frac{2\rho}{\omega_\BB}=\frac{2b\rho}{b-1},

1290: \eeq

1291: the left B-site is emptied.

1292: This is the first reaction of~(\ref{bbss}).

1293: An SS defect is thus formed one site to the left of the original BB defect.

1294: 

1295: \item

1296: Consider now an SS defect, made of two consecutive S-sites.

1297: The current between the two S-sites, $J_\SS\approx1/b^2$,

1298: is smaller than the mean current $J$.

1299: As a consequence, particles flow from the B-site to the left of the defect

1300: into the left S-site at a rate $\omega_\SS=J-J_\SS\approx(b-1)/b^2$.

1301: The left S-site is thus soon (i.e., after a time which does not

1302: grow proportionally to $\rho$) turned to a B-site.

1303: This is the second reaction of~(\ref{bbss}).

1304: A BB defect is thus formed one site to the left of the original~SS defect.

1305: \end{itemize}

1306: 

1307: The discussion can be summarized in the form of the following reactions

1308: \beq

1309: \BB\to\SB,\qquad\SS\to\BS.

1310: \label{bbss}

1311: \eeq

1312: The typical history of a single defect therefore looks as follows,

1313: where time runs from bottom to top,

1314: for the sake of consistency with Figure~\ref{figplot}:

1315: \[

1316: t\uparrow\quad\matrix{

1317: \BS\BS\bfBB\SB\SB\SB\SB\SB\SB\SB\cr

1318: \BS\BS{\rm B}\bfSS{\rm B}\SB\SB\SB\SB\SB\SB\cr

1319: \BS\BS\BS\bfBB\SB\SB\SB\SB\SB\SB\cr

1320: \BS\BS\BS{\rm B}\bfSS{\rm B}\SB\SB\SB\SB\SB\cr

1321: \BS\BS\BS\BS\bfBB\SB\SB\SB\SB\SB\cr

1322: \BS\BS\BS\BS{\rm B}\bfSS{\rm B}\SB\SB\SB\SB\cr

1323: \BS\BS\BS\BS\BS\bfBB\SB\SB\SB\SB\cr

1324: \BS\BS\BS\BS\BS{\rm B}\bfSS{\rm B}\SB\SB\SB\cr

1325: n\to}

1326: \]

1327: 

1328: \begin{figure}[!tb]

1329: \begin{center}

1330: \includegraphics[angle=90,height=6truecm]{figplot.eps}

1331: \caption{\small

1332: Space-time plot of the stationary domain pattern for $\rho=100$.

1333: Black (resp.~white) areas show the regions of space-time

1334: where the S-sites are the odd (resp.~the even) sites.

1335: The slope of the straight line in the right part of the plot

1336: yields $V\approx-0.0072$.}

1337: \label{figplot}

1338: \end{center}

1339: \end{figure}

1340: 

1341: The time it takes for a defect to move two sites to the left

1342: is therefore equal to $\tau_\BB$ on average.

1343: As a consequence, the pattern of defects and domains is

1344: advected with an upstream (negative) velocity $V\approx-2/\tau_\BB$, i.e.,

1345: \beq

1346: V\approx-\frac{b-1}{b\rho}.

1347: \label{vbal}

1348: \eeq

1349: This scenario is confirmed by Figure~\ref{figplot},

1350: showing a space-time plot of the stationary dynamics

1351: of the domain pattern for $\rho=100$.

1352: Equation~(\ref{vbal}) predicts $V\approx-0.0075$,

1353: in reasonably good agreement with the observed value $V\approx-0.0072$.

1354: The plot shows that the whole advected domain pattern

1355: behaves more or less as a rigid body over spatial scales

1356: much larger than the mean domain size $\xi$.

1357: 

1358: \section{Two-dimensional target process}

1359: \label{2D}

1360: 

1361: We now consider the canonical target process

1362: defined by the rate~(\ref{vcan}), on the square lattice

1363: with unit vectors $\bfe_1$, $\bfe_2$.

1364: In order to have a genuine two-dimensional model

1365: and to maximize the asymmetry, we choose the following rule:

1366: particles hop either East (displacement $\bfe_1$)

1367: or North (displacement $\bfe_2$) with equal probabilities.

1368: In other words, if the departure site is $\bfd=(m,n)$,

1369: the arrival site is chosen to be either $\bfa=(m+1,n)=\bfd+\bfe_1$,

1370: or $\bfa=(m,n+1)=\bfd+\bfe_2$, with probability $1/2$.

1371: The bias, i.e., the mean displacement proposed

1372: to a particle, $\bfb=(\bfe_1+\bfe_2)/2$, is along the North-East direction.

1373: 

1374: \subsection{Heuristic argument}

1375: 

1376: In the one-dimensional situation, the existence of a conserved current

1377: was instrumental in order to discriminate between possible scenarios

1378: for the stationary state of the model.

1379: In the present situation, however, the current $\J$ is a two-dimensional

1380: vector.

1381: The condition that~$\J$ be conserved in the stationary state

1382: is less stringent than in the one-dimensional case.

1383: On spatial scales much larger than the lattice spacing,

1384: it is reasonable to use the continuum formalism.

1385: Within this framework, the conservation law reads $\nabla\cdot\J=0$.

1386: The symmetry of the dynamical rules implies that $\J$

1387: is aligned with the bias $\bfb$.

1388: Current lines are therefore parallel straight lines along this direction.

1389: The conservation law implies that the magnitude of the current

1390: is constant along each current line,

1391: but may well vary in the transversal direction from one current line to another.

1392: In particular the existence of an {\it extended condensate} is allowed,

1393: in the form of a one-dimensional structure

1394: elongated along the direction of the bias.

1395: This opens up the possibility of having an unconventional

1396: type of condensation transition.

1397: Such a phenomenon is indeed observed in the numerical simulations described

1398: hereafter.

1399: 

1400: \subsection{Numerical results}

1401: 

1402: We choose once for all the value $b=4$ in the numerical simulations

1403: of the two-dimensional target process defined above.

1404: 

1405: In analogy with the one-dimensional case, we begin with the transient dynamics,

1406: starting from a random initial condition.

1407: We monitor the local relaxation by means of the reduced second moment $K(t)$

1408: of the occupation distribution, introduced in~(\ref{kdef}).

1409: Figure~\ref{fig2kt} shows a plot of $K(t)$ against time~$t$.

1410: For all values of the density~$\rho$, $K(t)$ is an increasing function of time,

1411: starting from its initial value $K(0)=(2\rho+1)/\rho$.

1412: The most significant feature to be observed on the data is the following:

1413: $K(t)$ saturates to a finite limiting value, denoted $K$,

1414: for values of the density up to the threshold value $\rho_0\approx6$.

1415: We have~$K_0\approx4.8$ for $\rho=\rho_0$.

1416: To the contrary, $K(t)$ grows indefinitely for larger values of the density.

1417: The right panel of Figure~\ref{fig2kt} demonstrates that

1418: the asymptotic growth law of~$K(t)$ for all~$\rho>\rho_0$ is of the form

1419: \beq

1420: K(t)\approx\frac{Ct}{\rho},

1421: \label{asylaw}

1422: \eeq

1423: where $C\approx9\times10^{-3}$ is taken from the slope

1424: of the parallel dashed lines.

1425: 

1426: \begin{figure}[!tb]

1427: \begin{center}

1428: \includegraphics[angle=90,height=6truecm]{fig2kt.eps}

1429: {\hskip 15pt}

1430: \includegraphics[angle=90,height=6truecm]{fig2klong.eps}

1431: \caption{\small

1432: Plot of the reduced second moment $K(t)$ of the occupation distribution

1433: in the two-dimensional target process against time $t$,

1434: for various values of the density $\rho$, indicated on the curves.

1435: Left: data for moderate values of time.

1436: Right: data for longer values of time, multiplied by density $\rho$.

1437: The parallel straight dashed lines, meant as a guide to the eye,

1438: have a slope $9.1\times10^{-3}$.}

1439: \label{fig2kt}

1440: \end{center}

1441: \end{figure}

1442: 

1443: Pursuing along the lines of our investigation of the one-dimensional case,

1444: we introduce a local relaxation time~$T_\loc$,

1445: defined by the condition $K(T_\loc)=4$.

1446: Figure~\ref{fig2tloc} shows a plot of $T_\loc$ so defined,

1447: divided by~$\rho$, against~$\rho$.

1448: The least-squares fit suggests a growth of the form~(\ref{tlinquad}),

1449: with $A\approx0.63$, in a wide range of values of the density.

1450: Notice that the local relaxation time exhibits no visible singularity

1451: at the threshold density $\rho_0$.

1452: 

1453: \begin{figure}[!tb]

1454: \begin{center}

1455: \includegraphics[angle=90,height=6truecm]{fig2tloc.eps}

1456: \caption{\small

1457: Plot of the local relaxation time $T_\loc$, divided by $\rho$,

1458: against density~$\rho$.

1459: Full straight line: least-squares fit with slope $A\approx0.63$.}

1460: \label{fig2tloc}

1461: \end{center}

1462: \end{figure}

1463: 

1464: We now turn to a more accurate determination of the threshold density $\rho_0$,

1465: using the pre-asymptotic growth of $K(t)$ in the intermediate time regime,

1466: where $K(t)$ has already departed in a significant way

1467: from its plateau value~$K_0$,

1468: but not yet reached the asymptotic linear growth~(\ref{asylaw}).

1469: The right panel of Figure~\ref{fig2kt} demonstrates

1470: that this regime lasts longer and longer

1471: as the threshold density $\rho_0$ is approached.

1472: This observation is turned to a quantitative measurement

1473: by defining the global relaxation time $T_\gl$

1474: by the condition $K(T_\gl)=K_0+\Delta K$,

1475: where we set $K_0=4.8$, whereas the choice $\Delta K=50/\rho$

1476: incorporates the form~(\ref{asylaw}) of the asymptotic growth law.

1477: Figure~\ref{fig2tg} shows a plot of the reciprocal of the global time $T_\gl$

1478: thus defined, against density.

1479: The data convincingly demonstrate that the global time diverges at some

1480: non-trivial threshold density~$\rho_0$.

1481: A crossover of the data toward another type of asymptotic behavior

1482: indeed seems extremely improbable,

1483: in view of the accuracy of the available data.

1484: The second-degree polynomial fit to the data shown on the plot

1485: provides a rather accurate determination of the threshold density,

1486: \beq

1487: \rho_0=6.0\pm0.1,

1488: \label{rhoc}

1489: \eeq

1490: as well as an evidence that the global time

1491: diverges linearly as the threshold density is approached from above, as

1492: \beq

1493: T_\gl\approx T_0\;\frac{\rho_0}{\rho-\rho_0},

1494: \label{tgldiv}

1495: \eeq

1496: with a rather large prefactor $T_0\approx28\,000$.

1497: 

1498: \begin{figure}[!tb]

1499: \begin{center}

1500: \includegraphics[angle=90,height=6truecm]{fig2tg.eps}

1501: \caption{\small

1502: Plot of the reciprocal of the global time $T_\gl$ against density $\rho$.

1503: Full line: second-degree polynomial fit yielding $\rho_0=6.0\pm0.1$.}

1504: \label{fig2tg}

1505: \end{center}

1506: \end{figure}

1507: 

1508: The threshold density $\rho_0$ is the maximal density for which a homogeneous

1509: fluid phase is stable.

1510: At variance with the critical density $\rho_c$ of the symmetric target process

1511: and of the dual ZRP, $\rho_0$ rather appears as a dynamical threshold.

1512: This viewpoint is corroborated by the fact

1513: that the stationary state at density~$\rho_0$ does not exhibit

1514: any critical feature.

1515: The distribution of the site occupations at the threshold density,

1516: shown in Figure~\ref{fig2f}, has an exponential fall-off

1517: of the form $f_k\sim\exp(-\mu k)$, with $\mu\approx0.06$,

1518: at least in the accessible range of values of the occupation.

1519: We checked that the data are not affected

1520: in an appreciable way by finite-size effects in the range considered.

1521: The critical occupation distribution~(\ref{fkc}) of the dual ZRP

1522: is shown on the same plot as a comparison.

1523: The latter distribution has a much smaller density $\rho_c=1/2$,

1524: some 12 times smaller than the observed threshold density~(\ref{rhoc})

1525: of the asymmetric two-dimensional model, but a slower power-law fall-off,

1526: so that the distributions eventually cross each other.

1527: 

1528: \begin{figure}[!tb]

1529: \begin{center}

1530: \includegraphics[angle=90,height=6truecm]{fig2f.eps}

1531: \caption{\small

1532: Logarithmic plot of occupation distributions.

1533: Upper curve: fully asymmetric target process in two dimensions at its threshold

1534: density $\rho_0$.

1535: The dashed line, meant as a guide to the eye, has a slope $-\mu\approx-0.06$.

1536: Lower curve: critical occupation distribution of the dual ZRP,

1537: given by~(\ref{fkc}).}

1538: \label{fig2f}

1539: \end{center}

1540: \end{figure}

1541: 

1542: The progressive emergence of highly occupied coherent structures,

1543: which are strongly elongated along the direction of the bias,

1544: is illustrated in Figure~\ref{fig2plot},

1545: showing two snapshots of the coarsening regime

1546: of a sample of size $100\times100$ at density $\rho=20$.

1547: The filled symbols show the 1\% most occupied sites.

1548: The visible structures clearly are precursors of the extended condensate

1549: mentioned above.

1550: The background density of the fluid phase besides these structures

1551: is found to be much smaller than the threshold density~$\rho_0$,

1552: and comparable to the critical density $\rho_c=1/2$ of the dual ZRP.

1553: We shall return to this point in more detail below (see Figure~\ref{figrhobg}).

1554: 

1555: \begin{figure}[!tb]

1556: \begin{center}

1557: \includegraphics[angle=90,height=6truecm]{fig2plot3.eps}

1558: {\hskip 15pt}

1559: \includegraphics[angle=90,height=6truecm]{fig2plot4.eps}

1560: \caption{\small

1561: Plots of the 100 (i.e.,~1\%) most occupied sites

1562: of a sample of size $100\times100$ at density $\rho=20$

1563: in the coarsening regime.

1564: Left: $t=5\times10^4$.

1565: Right: $t=10^5$.}

1566: \label{fig2plot}

1567: \end{center}

1568: \end{figure}

1569: 

1570: We now study the typical characteristic sizes (width and height)

1571: of the extended condensate in the stationary state of a finite system,

1572: and of its precursors in the coarsening regime of an infinite system.

1573: The width $W$ of a condensate is defined as the number of sites

1574: which take part in the condensate,

1575: whereas its height $H$ is the mean number of particles per site

1576: in the condensate, so that the product~$HW$

1577: gives a measure of the number of particles involved in the condensate.

1578: We first consider the coarsening regime of an infinite system.

1579: The condensate precursors shown in Figure~\ref{fig2plot}

1580: are expected to be characterized by a typical

1581: width $W(t)$ and height $H(t)$, with both scales growing with time.

1582: The contribution of these precursors

1583: to the reduced second moment of the occupation distribution

1584: can be checked to scale as $K(t)\sim H(t)/\rho$,

1585: irrespective of the width $W(t)$.

1586: The growth law~(\ref{asylaw}) therefore implies

1587: that the height of condensate precursors grows linearly in time, according to

1588: \beq

1589: H(t)\approx Ct,

1590: \label{ht}

1591: \eeq

1592: with $C\approx9\times10^{-3}$.

1593: This asymptotic coarsening law is expected to hold

1594: for any density~$\rho>\rho_0$.

1595: The behavior of the width $W$ of condensate precursors can only be investigated

1596: in an indirect way, by means of finite-size scaling.

1597: We therefore consider finite systems,

1598: namely square samples of linear size $L$, with periodic boundary conditions.

1599: Figure~\ref{figfss} shows numerical data concerning

1600: the stationary state of finite systems against their linear size~$L$,

1601: at fixed density $\rho=20$, well above the threshold density~$\rho_0$.

1602: The left panel shows the stationary-state value $K_L$

1603: of the reduced second moment of the occupation distribution.

1604: The right panel shows the characteristic relaxation time $T_L$,

1605: defined by the condition $K(T_L)=(K_L+K_0)/2$, again with $K_0=4.8$.

1606: The data for both quantities clearly exhibit a linear growth with the size $L$.

1607: The first of these growth laws implies $H_L\approx\rho K_L\sim L$.

1608: This is in accord with the expectation

1609: that typical stationary-state configurations

1610: have a single and roughly system-spanning extended condensate,

1611: for which $H_L\sim L$.

1612: Furthermore, as the number of particles involved in the condensate

1613: scales as $H_LW_L\sim L^2$, we have $H_L\sim W_L\sim L$.

1614: Let us now make the finite-size scaling assumption

1615: that $H(t)$ and $W(t)$ become respectively comparable to $H_L$ and $W_L$

1616: for a time $t$ comparable to the relaxation time $T_L$.

1617: This yields the scaling law $H(t)\sim t$,

1618: already known~(see~(\ref{ht})), and the prediction $W(t)\sim t$.

1619: 

1620: \begin{figure}[!tb]

1621: \begin{center}

1622: \includegraphics[angle=90,height=6truecm]{figfss1.eps}

1623: {\hskip 15pt}

1624: \includegraphics[angle=90,height=6truecm]{figfss2.eps}

1625: \caption{\small

1626: Plots of data concerning the stationary state of finite systems

1627: at fixed density $\rho=20$, against their linear size $L$.

1628: Left: stationary-state value $K_L$ of the reduced second moment

1629: of the occupation distribution.

1630: Right: relaxation time $T_L$.

1631: Full straight lines: least-squares fits with respective slopes 1.79 and

1632: 2\,660.}

1633: \label{figfss}

1634: \end{center}

1635: \end{figure}

1636: 

1637: Finally, we have also measured the background density $\rho_L^\bg$ of the fluid

1638: phase in the stationary state of finite samples at density $\rho=20$.

1639: This quantity is algorithmically defined as follows.

1640: For any intercept $k=1,\dots,L$, consider

1641: the total number of particles in the diagonal array

1642: with intercept $k$, i.e., with equation $n=m+k\ (\hbox{mod.}\ L)$:

1643: \beq

1644: \N_k=\sum_{m=1}^LN_{m,m+k}.

1645: \eeq

1646: The largest of these $L$ numbers, $\N_\max$,

1647: corresponds to the diagonal array occupied by the extended condensate.

1648: It is overwhelmingly larger than the others, as it scales as $\N_\max\sim L^2$.

1649: The other $(L-1)$ numbers $\N_k$ represent the fluid phase,

1650: and therefore scale as $\rho_L^\bg L$.

1651: We are thus naturally led to define the background density as

1652: \beq

1653: \rho_L^\bg=\frac{N-\N_\max}{L(L-1)},

1654: \eeq

1655: where $N$ is the total number of particles

1656: in the system.

1657: Figure~\ref{figrhobg} shows a plot of the stationary-state

1658: background density $\rho_L^\bg$ for finite samples of size $L$, against $1/L$.

1659: The data demonstrate that $\rho_L^\bg$ is smaller than unity,

1660: and therefore much smaller than the mean density $\rho=20$, as soon as $L\ge7$.

1661: The segregation phenomenon is therefore already fully at work

1662: for rather small system sizes.

1663: From a quantitative viewpoint, a second-degree polynomial fit to the data yields

1664: the extrapolated value $\rho^\bg\approx0.48$.

1665: The limiting value thus obtained is remarkable close

1666: to the critical density $\rho_c=1/2$ of the dual ZRP.

1667: Imposing the constraint $\rho^\bg=\rho_c=1/2$ indeed hardly changes

1668: the fit (dashed line).

1669: This agreement strongly suggests that the background fluid phase

1670: of the two-dimensional target process above its threshold density

1671: is characterized by the critical occupation distribution of the dual ZRP,

1672: just as the S-sites of the one-dimensional case in the high-density limit.

1673: As a consequence, the density $\rho^\bg$ of the fluid phase:

1674: \beq

1675: \rho^\bg=\left\{\matrix{

1676: \rho\hfill&\hbox{for}\hfill&\rho<\rho_0,\cr

1677: \rho_c\hfill&\hbox{for}\hfill&\rho>\rho_0,}\right.%}

1678: \label{rhojump}

1679: \eeq

1680: has a discontinuous jump at the threshold density $\rho_0$.

1681: 

1682: \begin{figure}[!tb]

1683: \begin{center}

1684: \includegraphics[angle=90,height=6truecm]{figrhobg.eps}

1685: \caption{\small

1686: Plot of the background density $\rho_L^\bg$

1687: in the stationary state of finite systems

1688: at fixed density $\rho=20$, against their reciprocal linear size $1/L$.

1689: Full line: second-degree polynomial fit yielding the extrapolated

1690: value $\rho^\bg\approx0.48$.

1691: Dashed line: constrained polynomial fit of degree 3 imposing

1692: $\rho^\bg=\rho_c=1/2$.}

1693: \label{figrhobg}

1694: \end{center}

1695: \end{figure}

1696: 

1697: \section{Discussion}

1698: \label{quatre}

1699: 

1700: In this work we introduced a novel example

1701: of a migration process, the target process.

1702: We then studied in detail the structure of the nonequilibrium stationary state

1703: of the asymmetric target process, the main focus being on

1704: the fate of the condensation phenomenon.

1705: 

1706: The stationary-state measures of migration processes

1707: do not have a product form in general.

1708: The {\it symmetric} target process, though,

1709: has the same stationary-state product measure as the corresponding dual ZRP.

1710: In particular, the so-called canonical target process, defined by the

1711: rate~(\ref{vcan}) dual to the ZRP with rate~(\ref{ucan}),

1712: has a continuous condensation transition

1713: at a finite critical density $\rho_c$ whenever $b>2$,

1714: with a macroscopic condensate occupying a single site

1715: for densities $\rho>\rho_c$, irrespective of the dimensionality of the system.

1716: The {\it asymmetric} target process has a fluctuating stationary state

1717: with non-trivial spatial and temporal correlations,

1718: whose qualitative features depend on the dimensionality.

1719: Our main effort in the present work consisted in characterizing

1720: this nonequilibrium stationary state,

1721: including its dependence on the dimensionality.

1722: 

1723: We have reached a complete understanding of the one-dimensional target process,

1724: especially in the high-density regime of most interest.

1725: We showed, by exploiting the existence of a conserved current,

1726: that the asymmetric canonical target process has no condensation transition,

1727: and remains homogeneous at any finite density.

1728: In the high-density regime, an alternating scenario prevails for $b>1$:

1729: typical configurations consist of long alternating sequences BSBSBSBS...

1730: of highly occupied B-sites, and less occupied S-sites, whose occupation

1731: distribution coincides with the critical distribution of the dual ZRP.

1732: The coherence length (mean domain size) of this alternating

1733: structure diverges as $\xi\sim\rho^2$.

1734: We also gave a characterization of the scaling behavior of many other

1735: quantities in the vicinity of the `infinite-density fixed point'.

1736: 

1737: For the asymmetric target process in higher dimensions,

1738: we argued that the condensate must be extended

1739: and have the form of a one-dimensional structure

1740: elongated along the direction of the bias.

1741: In the two-dimensional case, numerical simulations performed for $b=4$

1742: show that the model exhibits an unconventional condensation transition

1743: at the density $\rho_0\approx6$.

1744: This density, which is much larger than the critical density $\rho_c=1/2$

1745: of the dual ZRP, appears as a dynamical threshold:

1746: it is the maximal density at which a homogeneous fluid phase

1747: is dynamically stable.

1748: For $\rho>\rho_0$, the predicted extended condensate is observed,

1749: whereas the background fluid phase again appears as critical.

1750: This picture seems to be generic for higher-dimensional systems.

1751: Preliminary numerical simulations

1752: of the asymmetric target process on the three-dimensional cubic lattice

1753: (where the displacement is along either of the unit vectors

1754: $\bfe_1$, $\bfe_2$, $\bfe_3$ with equal probabilities),

1755: indeed show that the overall picture is quite similar

1756: to the two-dimensional one.

1757: The global relaxation time~$T_\gl$ is again found

1758: to diverge according to~(\ref{tgldiv}), with $\rho_0\approx34$ for $b=4$.

1759: 

1760: Let us mention that another mechanism

1761: leading to an extended condensate in a class of mass transport models

1762: in one dimension has been reported recently~\cite{ehm}.

1763: There, the nonequilibrium stationary-state measure is a product

1764: whose factors involve the occupations of two consecutive sites.

1765: In the condensed phase, those models exhibit an extended condensate,

1766: whose height and width scale as $H_N\sim W_N\sim N^{1/2}$

1767: for a finite system of $N$ sites.

1768: These scaling laws are formally identical to those found in the present work.

1769: 

1770: The existence of a threshold density $\rho_0$ at which

1771: the background density has a discontinuous jump (see~(\ref{rhojump}))

1772: is reminiscent of what occurs in the model studied in~\cite{zrp2},

1773: namely a ZRP with two species of particles, and with rates such that

1774: the stationary-state measure does not have a product form.

1775: When the densities $\rho^{(1)}$ and~$\rho^{(2)}$ of the two species are equal,

1776: the behavior of the system is qualitatively the same as that

1777: of the canonical ZRP (with one species).

1778: In particular the system has a continuous phase transition

1779: at some critical density $\rho_c$, from a fluid phase to a

1780: condensed phase with critical background.

1781: The general situation where the two densities are different

1782: however drastically departs from this known scenario.

1783: If either of the two densities ($\rho^{(1)}$, say) is kept fixed

1784: at a value larger than $\rho_c$, on increasing the other density $\rho^{(2)}$,

1785: the system remains homogeneous as long as $\rho^{(2)}$ is less than a

1786: threshold value $\rho^{(2)}_0$ which depends on~$\rho^{(1)}$.

1787: At this threshold the system undergoes a discontinuous transition

1788: from an imbalanced fluid phase,

1789: where both species have densities $\rho^{(1)}$ and $\rho^{(2)}_0$

1790: larger than the critical density, to an imbalanced condensate

1791: coexisting with a balanced critical fluid

1792: with densities $\rho^{(1)}=\rho^{(2)}=\rho_c$.

1793: 

1794: Finally, the observed rapid growth of the threshold density

1795: $\rho_0$ with the di\-men\-sio\-na\-li\-ty raises the question of the behavior

1796: of the asymmetric target process in high dimensions.

1797: At this point let us emphasize that the absence of a stationary-state product

1798: measure for the target process

1799: is a rather subtle effect which needs the conjunction of several ingredients,

1800: and chiefly the presence of a bias.

1801: This feature cannot be present in mean-field geometries

1802: such as the complete graph,

1803: so that the dynamical threshold behavior of the model in high dimensions

1804: is not expected to smoothly converge to a well-defined mean-field limit.

1805: 

1806: \section*{References}

1807: 

1808: \begin{thebibliography}{99}

1809: 

1810: \bibitem{loan}

1811: O'Loan O J, Evans M R and Cates M E, 1998 Phys. Rev. E {\bf 58} 1404

1812: 

1813: \bibitem{evans1}

1814: Evans M R, 2000 Braz. J. Phys. {\bf 30} 42

1815: 

1816: \bibitem{wis1}

1817: Kafri Y, Levine E, Mukamel D, Sch\"utz G M and T\"or\"ok J, 2002 Phys. Rev.

1818: Lett. {\bf 89} 035702

1819: 

1820: \bibitem{cg}

1821: Godr\`eche C, 2003 J. Phys. A {\bf 36} 6313

1822: 

1823: \bibitem{gross}

1824: Grosskinsky S, Sch\"utz G M and Spohn H, 2003 J. Stat. Phys. {\bf 113} 389

1825: 

1826: \bibitem{evans2}

1827: Evans M R and Hanney T, 2005 J. Phys. A {\bf 38} R195

1828: 

1829: \bibitem{glcond}

1830: Godr\`eche C and Luck J M, 2005 J. Phys. A {\bf 38} 7215

1831: 

1832: \bibitem{zeta1}

1833: Bialas P, Burda Z and Johnston D, 1997 Nucl. Phys. B {\bf 493} 505

1834: \nonum

1835: Bialas P, Burda Z and Johnston D, 1999 Nucl. Phys. B {\bf 542} 413

1836: \nonum

1837: Bialas P, Bogacz L, Burda Z and Johnston D, 2000 Nucl. Phys. B {\bf 575} 599

1838: 

1839: \bibitem{zeta2}

1840: Drouffe J M, Godr\`eche C and Camia F, 1998 J. Phys. A {\bf 31} L19

1841: 

1842: \bibitem{zeta3}

1843: Godr\`eche C and Luck J M, 2001 Eur. Phys. J. B {\bf 23} 473

1844: 

1845: \bibitem{barc}

1846: Godr\`eche C and Luck J M, 2002 J. Phys. Cond. Matt. {\bf 14} 1601

1847: 

1848: \bibitem{lux}

1849: Godr\`eche C, 2007 Lect. Notes Phys. {\bf 716} 216

1850: (Berlin: Springer) (cond-mat/0604276)

1851: 

1852: \bibitem{maj}

1853: Majumdar S N, Evans M R and Zia R K P, 2005 Phys. Rev. Lett. {\bf 94} 180601

1854: \nonum

1855: Evans M R, Majumdar S N and Zia R K P, 2006 J. Stat. Phys. {\bf 123} 357

1856: 

1857: \bibitem{spitz}

1858: Spitzer F, 1970 Advances in Math. {\bf 5} 246

1859: 

1860: \bibitem{andj}

1861: Andjel E D, 1982 Ann. Prob. {\bf 10} 525

1862: 

1863: \bibitem{kelly}

1864: Kelly F, 1979 {\it Reversibility and Stochastic Networks} (Chichester: Wiley)

1865: 

1866: \bibitem{cocozza}

1867: Cocozza-Thivent C, 1985 Z. Wahr. {\bf 70} 509

1868: 

1869: \bibitem{pairw}

1870: Sch\"utz G M, Ramaswamy R and Barma M, 1996 J. Phys. A {\bf 29} 837

1871: 

1872: \bibitem{kls3}

1873: Luck J M and Godr\`eche C, 2006 J. Stat. Mech. P08009

1874: 

1875: \bibitem{ehm}

1876: Evans M R, Hanney T and Majumdar S N, 2006 Phys. Rev. Lett. {\bf 97} 010602

1877: 

1878: \bibitem{zrp2}

1879: Godr\`eche C, 2006 J. Phys. A {\bf 39} 9055

1880: 

1881: \end{thebibliography}

1882: \end{document}

1883: