1: \documentclass[letterpaper]{epl2}

2: 

3: % \newif\ifpdf 

4: % \ifx\pdfoutput\undefined

5: % \pdffalse % we are not running PDFLaTeX

6: % \else

7: % \pdfoutput=1 % we are running PDFLaTeX

8: % \pdfcompresslevel=9

9: % \pdftrue

10: % \fi

11: 

12: %\usepackage{axodraw} 

13: \usepackage{amsmath}

14: \usepackage{graphicx}

15: \usepackage{bm}

16: \newcommand{\<}{\langle}

17: \renewcommand{\>}{\rangle}

18: \newcommand{\rr}{{\bf r}} 

19: \newcommand{\uu}{{\bf u}}

20: \newcommand{\kk}{{\bf k}}

21: \newcommand{\RR}{{\bf R}}

22: \newcommand{\qq}{{\bf q}}

23: \newcommand{\QQ}{{\bf Q}}

24: \newcommand{\GG}{{\bf G}}

25: \newcommand{\dd}{{\vec{\bf \delta}}}

26: \newcommand{\cc}{{\hat c}}

27: \newcommand{\ah}{{\hat a}}

28: \newcommand{\bb}{{\hat b}}

29: \newcommand{\znpk}{z_{n'k}}

30: \newcommand{\znk}{z_{nk}}

31: \newcommand{\ccd}{{\hat c^\dagger}}

32: \newcommand{\ahd}{{\hat a^\dagger}}

33: \newcommand{\bbd}{{\hat b^\dagger}}

34: %\def\KeyWord#1{$\backslash$\IfColor{$\!\!$\textRed{#1}\textBlack}{#1}$\!\!$}

35: \newcommand{\la}{\langle}

36: \newcommand{\ra}{\rangle}

37: \newcommand{\up}{\uparrow}

38: \newcommand{\dn}{\downarrow}

39: \newcommand{\rar}{\rightarrow}

40: \def  \bsig    {\mbox{\boldmath$\sigma $}}

41: \def  \btau    {\mbox{\boldmath$\tau $}}

42: 

43: 

44: \title{The Berry phase and the pump flux in stochastic chemical

45:   kinetics}

46: \shorttitle{The Berry phase in chemical kinetics}

47: 

48: \author{N.A. Sinitsyn \inst{1,2} \and Ilya Nemenman \inst{2}}

49: \shortauthor{N.A. Sinitsyn and Ilya Nemenman}

50: 

51: \institute{

52:   \inst{1}Center for Nonlinear Studies and \\

53:   \inst{2}Computer, Computational and Statistical Sciences Division,

54:   Los Alamos National Laboratory, Los Alamos, NM 87545 USA

55: }

56: \pacs{82.20.-w}{Chemical kinetics and dynamics}

57: \pacs{03.65.Vf}{Phases: geometric; dynamic or topological}

58: \pacs{05.10.Gg}{Stochastic analysis methods (Fokker-Planck, Langevin, etc.)}

59: \pacs{05.40.Ca}{Noise}

60: \pacs{87.15.Ya}{Biological and medical physics: Fluctuations}

61: 

62: \abstract{ We study a classical two-state stochastic system in a sea

63:   of substrates and products (absorbing states), which can be

64:   interpreted as a single Michaelis-Menten catalyzing enzyme or as a

65:   channel on a cell surface. We introduce a novel general method and

66:   use it to derive the expression for the full counting statistics of

67:   transitions among the absorbing states. For the evolution of the

68:   system under a periodic perturbation of the kinetic rates, the

69:   latter contains a term with a purely geometrical (the Berry phase)

70:   interpretation. This term gives rise to a pump current between the

71:   absorbing states, which is due entirely to the stochastic nature of

72:   the system. We calculate the first two cumulants of this current,

73:   and we argue that it is observable experimentally.  }

74: 

75: 

76: \begin{document}

77: 

78: \maketitle

79: 

80: 

81: %\section{Introduction}

82: 

83: Single molecule experiments \cite{english-etal-06} have led to a

84: resurgence of work on the stochastic version of the classical

85: Michaelis-Menten (MM) enzymatic mechanism that describes the

86: enzyme-driven catalytic conversion of a substrate into a product

87: \cite{MM}. A lot of progress (analytical and numerical) has been made

88: under various assumptions and for various internal enzyme structures

89: \cite{qian-elson-02,rao-arkin-03,qian-xie-06,gopich-szabo-06,xue-etal-06}. Still,

90: many questions linger. Specifically, linear signal transduction

91: properties of biochemical networks, such as frequency dependent gains,

92: are of a high interest

93: \cite{detwiler-etal-00,samoilov-etal-02,samoilov-etal-05}. These are

94: studied by probing responses to periodic perturbations in the input

95: signals (in our case, kinetic rates) due to fluctuations in chemical

96: concentrations, in temperature, or due to other signals.  However,

97: such approaches usually disregard a phenomena known from the theory of

98: stochastic ratchets \cite{reimann-02}, especially in the context of

99: biological transport

100: \cite{julicher-etal-02,parrondo-02,qian-98,astumian-bio98}. Namely, a

101: system's symmetry may break, and, under an influence of a periodic or

102: random zero-mean perturbation, the system may respond with a finite

103: flux in a preferred direction. This, indeed, happens for the MM

104: reaction \cite{astumian-03}, and we study the phenomenon in detail in

105: this work.

106: 

107: 

108: The ratchet or pump effect manifests itself during periodic driving of

109: simple classical systems, such as a channel in a cell membrane, which

110: is formally equivalent to the MM enzyme

111: \cite{westerhoff-86,chen-87,astumian-87,tsong-88,astumian-pra89,astumian-jchph89,robertson-90,robertson-91,markin-91,astumian-03,tsong-chang-02}. The

112: prevalence of the transport terminology over the chemical one in the

113: literature is grounded in experiments, where driving is achieved by

114: application of periodic electric field or nonequilibrium noise that

115: modulate barrier heights for different channel conformations. The

116: resulting cross-membrane flux has contributions that have no analog in

117: stationary conditions.

118:  

119: In a related system, a quantum pump

120: \cite{brouwer-98,makhlin-mirlin-01,moskalets-buttiker-02}, pump fluxes

121: admit a purely geometrical Berry phase interpretation

122: \cite{pump_berry}. A similar interpretation has been introduced for

123: some special classical cases \cite{shi,hannay-85}.  However, the Berry

124: phase has not yet been derived for classical stochastic Markov chains

125: that model chemical kinetics, even though recent developments strongly

126: hint at the possibility

127: \cite{hill-book,qian1-98,qian-00,astumian-03,ao-04,kwon-05}.  For

128: example, such systems admit introduction of a vector potential for the

129: fluxes to characterize circular motion \cite{qian1-98}, allow

130: reformulation of the Langevin dynamics in the form that strongly

131: resembles wave packet equations in quantum mecanics with a nontrivial

132: Berry curvature of Bloch bands \cite{ao-04,kwon-05,niu}, and result in

133: pump currents that depend on contour integrals over the the evolution

134: of the system's parameters \cite{astumian-03}.  Although currents in

135: most analyzed models were produced by the lack of the detailed balance

136: rather than by external time-dependent perturbations, a dual

137: description in terms of an external noise induced ratchet effect is

138: usually possible.  Conversely, time-dependent perturbations can break

139: the detailed balance and induce the catalytic cycle \cite{chen-87}.

140: 

141: In the present work we demonstrate that a theory based on the Berry

142: phase can be constructed for a purely classical adiabatically slowly

143: driven stochastic dynamics. The theory leads to an elegant

144: interpretation of prior results, and it makes new predictions even for

145: the minimal model of the MM reaction. Additionally, the theory fills

146: in an important gap in the current state of the field by providing a

147: simple, yet general recipe for calculation of mean particle fluxes and

148: their fluctuations.

149: 

150: In a standard minimal model, a single MM enzyme catalytic reaction

151: consists of an enzyme-substrate complex formation and its decay into

152: the enzyme and the product.  Both reactions may be reversible

153: \cite{qian-elson-02}.  We assume no multiple internal enzyme states

154: (see \cite{english-etal-06,qian-xie-06} for the complementing

155: discussion). The reaction can be viewed as a single bin (the enzyme)

156: that can contain either zero or one substrate particles in it. The

157: particle can escape from the bin by jumping into one of the two

158: absorbing states, either returning to the substrate -- the Left, or

159: converting into the product -- the Right, cf.~Fig.~\ref{system}.  In

160: turn, if the bin is empty, either of the absorbing states can emit a

161: new particle into it.  All transition times are exponentially

162: distributed with the rates defined as in Fig.~\ref{system}.  We

163: investigate the mean particle current $J$ from $L$ into $R$ on time

164: scales much larger than the typical transition times between any two

165: states. Our main goal is to understand the effect of periodic driving,

166: i.\ e., periodic rate changes.

167: 

168: 

169: We emphasize again that this system is equivalent to a cross-membrane

170: transport problem, where the L/R states correspond to compartments on

171: different sides of the membrane, and the bin is a membrane channel

172: complex.  Additionally, this system is relevant to transport through a

173: quantum dot in the Coulomb blockade regime \cite{nazarovFCS}.

174:  

175: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

176: 

177: \begin{figure}[t]

178: \centerline{\includegraphics[width=6 cm]{system}}

179: \caption{\label{system}Transition rates into and out of the absorbing

180:   states $L$ (substrate) and $R$ (product). Note that the rates $k_1$

181:   and $k_{-2}$ denote total rates (rather than per mole of

182:   substrate/product), and they can be modulated, for example, by

183:   changing concentrations of the substrate and the product,

184:   respectively.}

185: \end{figure}

186: 

187: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%         

188: 

189: 

190: 

191: Let $P_{e}$ and $P_{f}$ be probabilities of an empty/filled bin

192: (unbound/bound enzyme). Neglecting particle discreteness and assuming

193: that all rates change slowly, so that $P_{e/f}$ are always in

194: equilibrium, one derives for the instantaneous $L\to R$ current

195: \begin{equation}

196:   j_{\rm cl} (t)=\frac{\kappa_+(t)-\kappa_-(t)}{K(t)},\,\; K\equiv \sum_{m}

197:   k_m,\,\; \kappa_{\pm}=k_{\pm1}k_{\pm2},

198: \label{jcl}

199: \end{equation} 

200: and the summation is over $m=-2,-1,1,2$.  This is the classical (MM)

201: current. Notice that, even if $\<k_1k_2\>=\<k_{-1}k_{-2}\>$ (where

202: $\<\dots\>$ denotes time averaging), $J_{\rm cl}\equiv\<j_{\rm cl}\>\neq0$

203: due to nonlinearities of the MM reaction.

204: 

205: How is this result affected by the particle discreteness? The master

206: equation for the system's evolution is:

207: \begin{equation}

208: \frac{d}{dt} \left[ \begin{array}{l}

209: P_{e}\\

210: P_{f}

211: \end{array} \right] = - \left[ 

212: \begin{array}{ll}

213: \,\,\,\,k_1+k_{-2}   &  -k_{-1}-k_2\\

214:  -k_1-k_{-2}    &\,\,\,\, k_{-1}+k_2

215: \end{array} \right]

216: \left[ \begin{array}{l}

217: P_{e}\\

218: P_{f}

219: \end{array} \right],

220: \label{ev1}

221: \end{equation}

222: and we assume the adiabatic approximation, i.\ e., the rates $k_m$

223: oscillate with the frequency $\omega\ll \min_m k_m$. Since

224: $P_e=1-P_f$, (\ref{ev1}) is equivalent to a formally solvable first

225: order linear differential equation. However, expressing the solution

226: in a simple form requires approximations (e.g., small fluctuations)

227: even for a harmonic dynamics of the parameters

228: \cite{astumian-03}.  Additionally, the solution provides little

229: information about the fluxes beyond their mean values, and it is not

230: easy to generalize to the case of more complicated, multi-state

231: stochastic systems. Thus here we pursue a different analysis

232: technique.

233:  

234: The formal solution of (\ref{ev1}) is

235: \begin{equation}

236: {\bf p}(t)=\hat{T}\left(e^{-\int_{t_0}^{t}\hat{H}(t)dt}\right){\bf p}(t_0).

237: \label{ev12}

238: \end{equation}

239: where $\hat{T}$ stands for the time-ordering operator, ${\bf

240:   p}(t)\equiv [ P_{e}(t),P_{f}(t) ]^T$, and

241: \begin{equation}

242: \hat{H}= \left[ 

243: \begin{array}{ll}

244: \,\,\,\, k_1 +k_{-2}   & -k_{-1}-k_2 \\

245: -k_1-k_{-2}   & \,\,\,\, k_{-1}+k_2

246: \end{array} \right].

247: \label{h0}

248: \end{equation}

249: 

250: It is useful to separate the parts of $\hat{H}$ that are directly

251: responsible for the transition into or out of the $R$-state, namely

252: $\hat{H}\equiv\hat{H}_0 - \hat{V}_{+}-\hat{V}_{-}$, where 

253: \begin{eqnarray}

254: &&\hat{H}_0= \left[

255: \begin{array}{cc}

256: k_1+k_{-2}   & -k_{-1} \\

257: -k_1   &  k_{-1}+k_2

258: \end{array} \right],

259: \label{h00}\\

260: &&\hat{V}_+= \left[ 

261: \begin{array}{cc}

262: 0   & k_2 \\

263: 0   &   0

264: \end{array} \right], \,\,\,\,\,\,\,\,

265: \hat{V}_-= \left[ 

266: \begin{array}{cc}

267: 0   & 0 \\

268: k_{-2}   &   0

269: \end{array} \right].

270: \label{VV}

271: \end{eqnarray}

272: With this, we can write a formal expression for the probability $P_n$ to

273: have $n$ net transitions from the bin into $R$ during time $t$. This

274: is similar to \cite{gopich-szabo-06}, but we also allow $k_{-2}\neq0$,

275: counting transitions from $R$ into the bin with the negative sign. For

276: example,

277: \begin{align}

278: P_{n=1}&= {\bf 1}^+\left[ \int_{t_0}^{ t} dt_1 U_0(t,t_1)  

279: \hat {V}_{+}(t_1) U_0(t_1,t_0) \right.+ \nonumber\\

280: +&\int_{t_0}^{t} dt_3 U_0(t,t_3)  

281: \hat {V}_{+}(t_3)

282: \int_{t_0}^{t3} dt_2 U_0(t_3,t_2)  

283: \hat {V}_{-}(t_2)\times \nonumber\\

284: \times&\left.\int_{t_0}^{t2} dt_1 U_0(t_2,t_1)\hat {V}_{+}(t_1)

285:   U_0(t_1,t_0) +\cdots\right] {\bf p}(t_0). \label{p1}

286: \end{align}

287: Here ${\bf 1}$ is the vector with all unit entries and

288: \begin{equation}

289: U_0(t,t') = \hat{T}\left(e^{-\int_{t'}^t\hat{H}_0(t)dt}\right).

290: \label{u0}

291: \end{equation}

292: We introduce the probability generating function

293: (pgf) for the number of transitions from the bin into R,

294: \begin{equation}

295: Z(\chi)=e^{S(\chi)}=

296: \sum_{s=-\infty}^{\infty} P_{n=s}e^{is\chi}.

297: \label{pgf1}

298: \end{equation}

299: $\chi$ is called {\em the counting field}, $S(\chi)$ is {\em the full

300:   counting statistics}, and its derivatives give cumulants (or

301: connected correlation functions) of $P_n$, e.g. $\<n\>=-i \partial

302: S(\chi)/\partial \chi |_{\chi=0} $, $\<\delta^2

303: n\>\equiv\<n^2\>-\<n\>^2 = (-i)^2 \partial^2 S(\chi)/\partial \chi^2

304: |_{\chi=0}$, etc.

305: 

306: We derive the full counting statistics similarly to

307: \cite{gopich-szabo-06,nazarovFCS}, but we extend the method by

308: allowing for an explicit time-dependence of $\hat{H}$.  First notice

309: that (\ref{pgf1}) with the expansion of probabilities as in (\ref{p1})

310: is equivalent to an expansion of an evolution operator with a

311: $\chi$-dependent Hamiltonian:

312: \begin{equation}

313: Z= {\bf 1}^+\hat{T}\left(e^{-\int_{t_0}^{t}\hat{H}(\chi,t) dt}\right) {\bf p}(t_0),

314: \label{pdf2}

315: \end{equation}

316: where 

317: \begin{align}

318: \hat{H}(\chi,t)&=\hat{H}_0(t)-\hat{V}_+(t)e^{i\chi}-

319: \hat{V}_{-}(t)e^{-i\chi}\nonumber\\

320: =&\left[ 

321: \begin{array}{cc}

322: k_1 + k_{-2}   & -k_{-1}-k_2 e^{i\chi} \\

323: -k_1-k_{-2}e^{-i\chi}   & k_{-1}+k_2

324: \end{array} \right].

325: \label{hchi}

326: \end{align}

327: We define the two instantaneous eigenstates of $\hat {H}(\chi,t)$ as

328: $|u_0\rangle$ and $|u_1\rangle$. There are also the left eigenstates $

329: \langle u_0|$ and $\langle u_1|$ with the same eigenvalues; we

330: normalize them so that $ \langle u_{n}(t)|u_m (t)\rangle=\delta_{nm}$.

331: 

332: 

333: 

334: Let's introduce an intermediate time scale $\delta t$, $ 1/\omega \gg

335: \delta t \gg \max_m (1 /k_m)$.  During this time, typically, many

336: transitions happen but $k_m$'s are approximately constant.  Consider

337: time-points $t_j=t_0+j\delta t$, $j=1\dots N$, and $t_N\equiv t$.  In

338: the adiabatic limit, one can approximately rewrite the expression for

339: the time ordered exponent in (\ref{pdf2}) as a product of evolution

340: operators over time intervals of the size $\delta t$ with parameters

341: set to constants at times $t_j$, i.\ e.,

342: \begin{equation}

343: Z\approx  {\bf 1}^+ e^{-\hat{H}(\chi,t_N)\delta t}

344: e^{-\hat{H}(\chi,t_{N-1})\delta t} \dots e^{-\hat{H}(\chi,t_0)\delta t} {\bf p}(t_0).

345: \label{ZZ}

346: \end{equation}

347: Now we insert the resolution of the identity

348: \begin{equation}

349: \hat{1}= |u_0 (t_i)\rangle \langle u_0(t_i)|

350: + |u_1 (t_i)\rangle\langle u_1(t_i)|

351: \label{one}

352: \end{equation}

353: after every exponent at $t_i$. We define $|u_0(t_i) \rangle$ as the

354: eigenstate of $\hat{H}(\chi,t_i)$ corresponding to

355: the eigenvalue $\lambda_0$ with the smaller real

356: part. Since $\delta t\gg1/k_i$, $\vert \exp\left[-\lambda_0(\chi,t_i)\delta

357:   t\right]\vert \gg \vert \exp\left[-\lambda_1(\chi,t_i)\delta t\right]\vert$, and

358: terms containing $|u_1(t)\rangle$ can be neglected. Moreover, after

359: long evolution, any information about the initial and the final states

360: is lost. Thus we rewrite the pgf as

361: \begin{equation}

362: Z\approx e^{-\lambda_{min}(\chi,t_0)\delta t} \prod_{i=1}^Ne^{-\lambda_0(\chi,t_i)\delta t} 

363: \langle u_0(t_i)|u_0(t_{i-1})\rangle.

364: \label{ZZ1}

365: \end{equation}

366: Finally, we approximate $\langle u_0(t_{i})|u_0(t_{i-1})\rangle

367: \approx \exp[-\delta t \langle u_0(t_{i})

368:   | \partial_t|u_0(t_{i-1})\rangle]$, which allows us to write

369: \begin{equation}

370:   S(\chi) \approx S_{\rm geom}+S_{\rm cl}=-\frac{T}{T_0}\int_{0}^{T_0}dt\, [

371:   \langle u_0|\partial_t|u_0\rangle+\lambda_0(\chi,t) ],

372: \label{fcs2}

373: \end{equation}

374: where $T_0=2\pi / \omega$ is the period of the rate oscillations and

375: $T\gg T_0$ is the total measurement time.  The last term in

376: (\ref{fcs2}) would be the same even for a time-independent problem,

377: and it has been discussed before \cite{nazarovFCS}. Diagonalizing

378: $\hat{H}$, and denoting $e_{\pm\chi}\equiv e^{\pm i\chi}-1$, we get for this term

379: \begin{equation}

380:   S_{\rm cl}=\frac{-T}{2T_0} \int_0^{T_0} dt \left[ K-

381:   \sqrt{K^2+4(\kappa_+e_{+\chi}+\kappa_-e_{-\chi})}  \right].

382: \label{Scl}

383: \end{equation}

384:  

385: The first term in (\ref{fcs2}) is more interesting. Since it depends

386: only on the choice of the contour in the ${\bf k}$ space, but not on

387: the rate of motion along the contour (at least in the adiabatic

388: approximation), it has a geometric interpretation. We write

389: \begin{equation}

390:   S_{\rm geom}=-\frac{T}{T_0}\oint_{{\bf c}} {\bf A} \cdot d{\bf

391:     k},\;\,

392:   A_m = \langle u_0({\bf k})|\partial_{k_m}|u_0({\bf k})\rangle,

393: \label{Sgeom}

394: \end{equation}

395: where the integral is over the contour in the 4-dimentional 

396: parameter space (the ${\bf k}$-space) drawn

397: during one cycle.  

398: 

399: To estimate the error in our results, we note that (\ref{pdf2}) is

400: exact, and then assumptions followed. First, we neglected the initial

401: nonequilibrium relaxation on a time scale of $\sim k_m^{-1}$. Since

402: the $S(\chi)\sim T$, this introduces an error in $S$ of

403: $\sim1/(k_{m}T)$, which vanishes for long observation times. Second,

404: we projected the evolution only to the subspace of states with the

405: smallest real part of the eigenvalue. The resulting error is

406: exponentially suppressed in the adiabatic limit by

407: $\exp(-(\lambda_{1}-\lambda_0)/\omega)$. Third, there is the coarse

408: graining in (\ref{ZZ}).  To the lowest order, it introduces errors in

409: $S$ in the form of commutators, such as $[H(\chi,t_i),H(\chi,t_j)]$, $|t_i-t_j|

410: < \delta t$. Since $\langle u_0(t+\delta

411: t)|[H(\chi,t_i),H(\chi,t_j)]|u_0(t)\rangle \sim O([\omega \delta t]^2)$ for $t

412: <t_i,t_j<t+\delta t$, this error is less significant than the

413: $O(\omega)$ contribution from the geometric term in $S$. Finally, the

414: error due to the approximation of $\<u_0(t_i)|u_0(t_{i-1})\>$ in

415: (\ref{ZZ1},~\ref{fcs2}) is of the same order.

416: 

417: In a chemical system, the rates $k_m$ can be changed by many means,

418: e.g., by varying the system's temperature. However, the simplest

419: scenario is to couple the substrate and the product to particle baths

420: and to vary the corresponding chemical potentials for both

421: species. Since the rates $k_1$ and $k_{-2}$ are proportional to the

422: particle numbers, they will oscillate as well. Thus in what follows we

423: assume that $k_1$ and $k_{-2}$ are time dependent, while $k_2$ and

424: $k_{-1}$ are constants.  Then, using Stokes theorem, we write

425: \begin{equation}

426: \oint {\bf A} \cdot d{\bf k} = \iint_{{\bf s_c}} dk_{1}dk_{-2} F_{k_1,k_{-2}},

427: \label{stokes}

428: \end{equation}

429: where the integration is over the surface ${\bf s_c}$ enclosed by the

430: contour ${\bf c}$, and

431: \begin{equation}

432: F_{k_1,k_{-2}}=\frac{\partial A_{{-2}}}{\partial k_{1}}-

433: \frac{\partial A_{1}}{\partial k_{-2}}.

434: \label{berry}

435: \end{equation}

436: We will call $F_{k_1,k_{-2}}$ the {\em Berry curvature} by analogy

437: with similar definitions in quantum mechanics.  The advantage of

438: working with $F$ rather than with the {\em potentials} $A_m$ is that

439: the Berry curvature is gauge invariant, i.e., it does not depend on an

440: arbitrary ${\bf k}$-dependent normalization of $|u_0({\bf

441:   k})\rangle$. It is a truly measurable quantity. In our case,

442: \begin{equation}

443: F_{k_1,k_{-2}}=\frac{e_{-\chi}(e^{i\chi}k_2+k_{-1})}{[4\kappa_+e_{+\chi}+

444: 4\kappa_-e_{-\chi}+K^2]^{3/2}}.

445: \label{berry2}

446: \end{equation}

447: Note that, with $\chi=0$, the Berry curvature is zero, so it is the

448: counting field $\chi$ that introduces a nontrivial topology in the

449: phase space of the eigenstates of $\hat{H}(\chi,t)$. More generally, a

450: normal Markovian evolution corresponds to $\chi=0$, where the

451: normalization of $P$ ensures that $F|_{\chi=0}=0$. This may be the

452: reason why nobody discussed the Berry phases in the context of Markov

453: chains.

454: 

455: Knowing the full counting statistics, one can study both the particle

456: flux and its fluctuations. Importantly, from (\ref{fcs2},

457: \ref{Sgeom}), all cumulants will have the geometric and the classical

458: terms. In particular, for the mean $L\to R$ flux per unit time, we get

459: \begin{equation}

460:   J=J_{\rm pump}+J_{\rm cl}=\iint_{{\bf  s_c}} d^2k 

461:   \frac{k_2+k_{-1}}{T_0K^3}+

462:   \int_0^{T_0} dt\, \frac{j_{\rm cl}(t)}{T_0},

463: \label{JJ}

464: \end{equation}

465: where $j_{\rm cl}$ is the classical current defined in (\ref{jcl}).

466: Generally, the ratio of geometric and classical terms is $\sim\omega

467: /k_m \ll 1$.

468: 

469: For the large observation time $T$, the total expected $L\to R$

470: particle flux is $\<n(T)\>=JT$. Similarly, its fluctuations are given

471: by $\<\delta^2n(T)\>=J^{(2)}T$, where

472: \begin{align}

473: &J^{(2)}=-\frac{1}{T}\left.\frac{\partial^2 \left(  S_{\rm geom} + S_{\rm cl}\right)}{\partial \chi^2}\right|_{\chi=0}=J_{\rm pump}^{(2)}+J_{\rm cl}^{(2)}

474: ,\label{noise}\\

475: &J^{(2)}_{\rm pump}=\iint_{{\bf  s_c}} d^2k 

476:  \left[ \frac{k_2-k_{-1}}{T_0K^3}- 

477:  \frac{12(k_2+k_{-1})(\kappa_+-\kappa_-)}{T_0K^5} \right],

478: \label{pump_noise}\\

479: &J^{(2)}_{\rm cl}=\frac{1}{T_0}\int_0^{T_0}dt 

480:  \left[

481: \frac{\kappa_++\kappa_-}{K}-\frac{(\kappa_+-\kappa_-)^2}{K^3} \right].

482: \label{cl-noise}

483: \end{align}

484: The ratio of the pump and the classical noises is again $\sim \omega/

485: k_m$. Importantly, we see that, in general, expectations of the total,

486: the classical, and the pump currents are not equal to their variances.

487: Thus none of the currents in the MM problem is Poissonian. This is a

488: general property of complex, multi-step reactions, which is often

489: neglected in computational studies of biochemical reaction

490: networks. Also note that $J^{(2)}_{\rm pump}$ in (\ref{pump_noise}) is

491: not necessarily positive and can, in fact, {\em decrease} the total

492: noise.  However, the total noise variance $J^{(2)}$ is strictly positive.

493: 

494: While the pump current in this system has been analyzed,

495: cf.~\cite{astumian-03}, to our knowledge, an expression for $J$

496: parameterized by the rates, rather than by internal enzyme parameters

497: has not been available. Even more importantly, expressions for the

498: noise (either classical or pump) for variable kinetic rates have not

499: existed either.

500: 

501: The smallness of the pump effect compared to the classical current

502: complicates its observation. However, several opportunities exist.

503: One is the dependence of $J_{\rm pump}$ on the frequency of the

504: perturbation, while $J_{\rm cl}\neq J_{\rm cl}(\omega)$. The second

505: possibility is to vary the rates along a contour with $J_{\rm

506:   cl}=0$. However, even in this case the noise may still be dominated

507: by the classical contribution.

508: 

509: To test our predictions, we choose one particular such contour with

510: $J_{\rm cl}=0$, $J_{\rm pump}\neq 0$ for a numerical analysis:

511: \begin{equation}

512: k_{1}=A+R\cos(\omega t),\; k_{-2} =A+R\sin(\omega t),\; k_{-1}=k_2=1.

513: \label{timeev}

514: \end{equation}

515: To estimate $J^{(2)}$ numerically, we implemented a Gillespie-like

516: scheme \cite{gillespie-77} that admits an explicit time dependence of

517: the rates. However, since $J_{\rm pump}\ll \sqrt{J^{(2)}_{\rm cl}}$,

518: such Monte-Carlo estimation of $J$ would take too long.  Instead, we

519: numerically solved the master equation (\ref{ev1}) for $P_f(t)$ with a

520: time discretization $dt\ll \min_m (1/k_m)$. This gave a better

521: precision than an analytical result of \cite{astumian-03}, which

522: assumes $R\to0$. Knowing $P_f$,

523: $j(t)=k_2P_f-k_{-2}(1-P_f)$. Fig.~\ref{Jpump} shows an excellent

524: agreement of our theory with the numerical results for both $J$ and

525: $J^{(2)}$.

526: 

527: 

528: 

529: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

530: 

531: \begin{figure}[t]

532: \centerline{{\bf(a)}}

533: %\ifpdf 

534: \includegraphics[width=8cm]{mean_data} 

535: %\else

536: %  \includegraphics[height=8cm, angle=270]{mean_data} \fi

537: \centerline{{\bf(b)}}

538: % \ifpdf 

539: \includegraphics[width=8cm]{var_data} 

540: %\else \includegraphics[height=8cm, angle=270]{var_data} \fi

541: \centering

542: \caption{Comparison of analytical predictions and numerical results

543:   for a system defined by (\ref{timeev}) with $A=1.5$. Here the theory

544:   gives: $J_{\rm cl}=0$, $J_{\rm pump}= \omega R^2[2(2-R^2+4A+2A^2)

545:   ]^{-3/2}$, $J_{\rm cl}^{(2)}=1-(50-2R^2)(25-2R^2)^{-3/2}$, and

546:   $J^{(2)}_{\rm pump}=0$. Points with error bars mark numerical

547:   results, while lines correspond to theoretical predictions. {\bf

548:     (a)} Mean $L\to R$ particle flux $J$ obtained by integrating

549:   (\ref{ev1}) with $dt=0.001$ and averaged over $\sim 100$ rate

550:   oscillations. Error bars are negligible, and so is the computational

551:   time to generate the data.  {\bf (b)} Mean $L\to R$ flux variance

552:   $J^{(2)}$ for $\omega=0.03$ Hz. Each point is a result of simulating

553:   $3\cdot 10^6$ elementary particle hops, and covers about 6000 rate

554:   oscillations. Error bars denote one standard deviation, as estimated

555:   from the posterior variance of $J^{(2)}$. These simulations took

556:   less than a day on a modern PC using Octave. }

557: \label{Jpump}

558: \end{figure}

559: 

560: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%    

561: 

562: Intuitively, the flux cannot keep increasing linearly as $\omega$

563: grows and the adiabatic approximation fails.  To understand the

564: behavior at large frequencies, we consider the case of $\max k_m \ll

565: \omega$. We look for the probabilities in the form

566: $P_{e/f}=\bar{P}_{e/f} +\delta P_{e/f}$, where $\bar{P}_{e/f}$ are

567: calculated for the time-averaged parameter values, namely, $\bar{P}_e

568: = (\<k_2\>+\<k_{-1}\>)/\<K\>$, $\bar{P}_f=1-\bar{P}_e$, and $\delta

569: P_{e/f}$ are small and oscillate fast.  The latter can be found from

570: (\ref{ev1}). For example, when $k_1=\<k_1\> +\delta k_1 \cos(\omega

571: t)$, $k_{-2}=\<k_{-2}\>+\delta k_{-2} \sin(\omega t)$, and $k_2$,

572: $k_{-1}$ are constants, we find $\delta P_{e} = -(\delta k_1 \bar{P}_e

573: /\omega) \sin(\omega t) +(\delta k_{-2} \bar{P}_f/\omega )\cos(\omega

574: t)$. Then the current averaged over the oscillation period is

575: $J=J_{\rm cl}(\{\<k_m\>\})+J_{\rm pump}$, where $J_{\rm pump} =

576: \delta k_1 \delta k_{-2}\bar{P}_e/(2\omega)$ is the correction due to

577: the fast rate oscillations. For the parameters as in (\ref{timeev}),

578: we again get $J_{\rm cl}=0$, but $J_{\rm pump}\neq0$. However, the

579: dependency of $J_{\rm pump}$ on $\omega$ has changed from $\sim\omega$

580: in the adiabatic case to $\sim 1/\omega$ for fast oscillations. This

581: agrees with \cite{astumian-03}, which used a different approximation

582: and showed that there is a single maximum in the pump current at

583: $\omega \sim k_m$.

584: 

585: The phenomenon of $J_{\rm pump}\neq 0$ can be explained in simple

586: terms. Since a particle spends a finite time bound to the enzyme, the

587: values of $k_1$ and $k_{-2}$ cannot influence the system during the

588: mean unbinding time following a binding event. If, during an

589: oscillation, the left binding rate is higher than the right one during

590: the upramp of the cycle, then $k_1$ ``shields'' growing values of

591: $k_{-2}$ from having an effect, while $k_{-2}$ shields decreasing

592: values of $k_1$ during the downramp. This leads to a phase-dependent

593: asymmetry that is the source of the pump flux. Larger frequencies lead

594: to more shielding, hence the linear $\omega$ dependence. However, for

595: $\omega\to\infty$, the information about the phase of the oscillations

596: is lost while a particle is bound, decreasing the asymmetry and the

597: flux.

598: 

599: 

600: 

601: Is the pump flux observable experimentally in biochemical (rather than

602: channel transport) experiments?  Exact zeroing of $J_{\rm cl}$, as in

603: the numerical example above, which would make $J_{\rm pump}$ the

604: leading effect, may be difficult to achieve. However, the classical

605: current is also small near the classical steady state,

606: $\<k\>_1k_2=k_{-1}\<k_{-2}\>$. In this case, if

607: $k_{1,-2}=\<k_{1,-2}\>+\delta k_{1,-2}\cos(\omega t-\phi_{1,-2})$, and

608: if the oscillations are small, $\delta k_{1,-2}/k_{1,-2}\ll 1$, then

609: we can disregard variation of the Berry curvature inside the contour, and

610: \begin{align}

611: J_{\rm pump}&\approx  \omega \frac{k_2+k_{-1}}{2\<K\>^3 }\delta k_1 \delta

612: k_{-2}\sin\phi,\\

613: J_{\rm pump}^{(2)}&

614: \approx  \omega \frac{k_2-k_{-1}}{2\<K\>^3 }\delta k_1 \delta

615: k_{-2}\sin\phi,

616: \end{align}

617: where $\phi=\phi_{-2}-\phi_1$.  This should be compared to the

618: classical contributions in the same limit

619: \begin{align}

620: &J_{\rm cl} \approx \frac{k_2 \delta k_1^2-  

621: k_{-1} \delta k_{-2}^2 + (k_2-k_{-1})

622: \delta k_1\delta k_{-2} \cos \phi}{2\<K\>^2},\\

623: &J_{\rm cl}^{(2)} \approx \frac{2\<k_1\>k_2}{\<K\>}.

624: \end{align}

625: We see that, while $J_{\rm pump}/J_{\rm cl} \sim \omega /\<K\>$, the

626: ratio of the variances is further suppressed, $J^{(2)}_{\rm

627:   pump}/J_{\rm cl}^{(2)}\sim (\omega/\<K\>)\delta k_1\delta

628: k_{-2}/(k_1k_{-2})$. Thus overcoming the classical variance is the

629: biggest concern for a successful experiment.

630: 

631: 

632: Our model has a single substrate and a single product. Thus the enzyme

633: we are discussing is an EC 5 enzyme. Properties of such enzymes vary

634: dramatically depending on a reaction, a biological species, and

635: mutations in protein sequences \cite{brenda}. A typical range of $k_2$

636: {\em in vitro} is $10^{-2}\dots10^4$ s$^{-1}$. Similarly, the

637: Michaelis constant, $K_M=(k_{-1}+k_2)[S]/k_1$ in our notation, varies

638: between $0.01$ and $10$ mM (here $[S]$ is the substrate

639: concentration). For our analysis, we take $k_2\sim 10$ s$^{-1}$, and

640: $K_M=1$ mM. While little is known about $k_{-1}$ and $k_2$ separately,

641: it's reasonable to assume $k_{-1}\sim k_2$. Similarly, we take

642: $k_{-2}\sim k_{1}[P]/[S]$ since both rates are often dominated by the

643: mean particle-enzyme collision time \cite{qian-elson-02} (for example,

644: for triose phosphate isomerase reaction, $k_{-2}/k_1\approx 2 [P]/[S]$

645: \cite{knowles-albery-77}).  Many enzymes with similar parameters have

646: been characterized in the BRENDA database \cite{brenda}.  Then, with

647: $[S]\sim [P]\sim 1$ mM and with $\delta k_{1,-2}/k_{1,-2}\sim10$\%, we get

648: $J_{\rm pump}\sim J^{(2)}_{\rm pump}\sim 10^{-2}/T_0$, $J_{\rm cl}\sim

649: 10^{-1}$, and $J_{\rm cl}^{(2)}\sim 10$ particles per

650: second. Oscillation periods of $\sim 1$ s are attainable (and still

651: satisfy $\omega\ll k_m$), which puts the flux ratio at $\sim

652: 10^{-1}$. Different dependence on $\phi$ can further improve

653: detectability of $J_{\rm pump}$ compared to $J_{\rm cl}$. At these

654: parameters, the pump flux becomes equal to the total flux standard

655: deviation for an experiment lasting a few days. Alternatively, working

656: with, say, $\sim10^6$ enzymes, $J_{\rm pump}$ and $J^{(2)}_{\rm pump}$

657: become observable in less than a minute. While real experiments will

658: certainly have additional complications, it is clear that our

659: predictions should be experimentally testable, at least in principle.

660: 

661: 

662: 

663: In conclusion, we constructed the Berry phase theory of a purely

664: classical adiabatic stochastic pump in a simple Michaelis-Menten

665: enzymatic mechanism. Our approach allowed calculation of the particle

666: flux and its variance, including the classical and the pump effect

667: contributions.  We believe that these predictions can be checked

668: experimentally, and it should be interesting to consider their

669: importance in the context of enzymatic signal processing.

670: 

671: While we analyzed only one specific model system, the Berry phase

672: approach is general and can be employed for many processes that can be

673: reduced to slowly driven Markov dynamics.  Examples range from charge

674: transport in quantum dots at strong decoherence, to particle transport

675: through cell membrane channels, and to various biochemical reaction

676: systems. For these problems, our method has several advantages

677: compared to already known techniques.  First, it provides a formal

678: recipe for an analysis of an arbitrary Markov chain: one should

679: construct a Hamiltonian with counting fields, find its eigenvalue with

680: the lowest real part and the Berry curvature from the corresponding

681: eigenstate, and then the counting statistics is given by (\ref{fcs2},

682: \ref{Sgeom}). Second, the method provides a solution not only for

683: average fluxes, but also for the full counting statistics and thus for

684: all flux cumulants. This is important in the context of elimination of

685: fast degrees of freedom for construction of coarse-grained biochemical

686: reaction models (e.~g., MM or Hill phenomenological laws), where a

687: correct treatment of noise in the remaining degrees of freedom has

688: always been a point of contention. Third, the method allows to

689: transfer many results of the well-developed Berry phase theory for

690: dissipative quantum dynamics \cite{garrison-88,vertex-1,vertex-2} to

691: the field of chemical kinetics. In particular, techniques exist to

692: extend our work and to calculate flux cumulants to all orders in

693: $\omega/ k_{min}$ \cite{Garanin}.

694: 

695: 

696: \begin{acknowledgments}

697:   We thank W.\ Hlavacek, F.\ Mu, M.\ Wall, C.\ Unkefer and F.\

698:   Alexander for useful discussions and critical reading of the

699:   manuscript. We are grateful to our two anonymous referees for

700:   insightful comments, which substantially improved this letter. Our

701:   work was funded in part by DOE under Contract No.\

702:   DE-AC52-06NA25396.  IN was further supported by NSF Grant No.\

703:   ECS-0425850.

704: \end{acknowledgments}

705:  

706: \begin{thebibliography}{0}

707: 

708: \bibitem{english-etal-06}

709:   \Name{English B.~P.,  et al.}

710:   \REVIEW{Nat.\ Chem.\ Biol.}{2}{2006}{87}.

711: 

712: \bibitem{MM}

713:   \Name{Michaelis L. \and Menten M.~L.},

714:   \REVIEW{Biochem. Z.}{49}{1913}{333}.

715: 

716: 

717: 

718: \bibitem{qian-elson-02}

719:   \Name{Qian H. \and Elson E.~L.},

720:   \REVIEW{Biophys.\ Chem.}{101–102}{2002}{565}.

721: 

722: \bibitem{rao-arkin-03}

723:   \Name{Rao C.~V. \and Arkin A.~P.}

724:   \REVIEW{J. Chem. Phys.}{118}{2003}{4999}.

725: 

726: \bibitem{qian-xie-06}

727:   \Name{Qian H. \and Xie X.~S.},

728:   \REVIEW{Phys. Rev. E.}{74}{2006}{010902(R)}.

729: 

730: \bibitem{gopich-szabo-06}

731:   \Name{Gopich I.V. \and Szabo A.}

732:   \REVIEW{J. Chem. Phys.}{124}{2006}{154712}.

733: 

734: \bibitem{xue-etal-06}

735:   \Name{Xue X., Liu F. \and Ou-Yang Z.-c.}

736:   \REVIEW{Phys. Rev. E}{74}{2006}{030902(R)}.

737: 

738: \bibitem{detwiler-etal-00}

739:   \Name{Detwiler P., et al.}

740:   \REVIEW{Biophys.\ J.}{79}{2000}{2801}.

741: 

742: \bibitem{samoilov-etal-02}

743:   \Name{Samoilov M., Arkin A.~P. \and Ross J.}

744:   \REVIEW{J.  Phys. Chem. A}{106}{2002}{10205}.

745: 

746: \bibitem{samoilov-etal-05}

747:   \Name{Samoilov M., Plyasunov S. \and Arkin A.~P.}

748:   \REVIEW{Proc. Natl. Acad. Sci. (USA)}{102}{2005}{2310}.

749: 

750: \bibitem{reimann-02}

751:   \Name{Reimann P.}

752:   \REVIEW{Phys. Rep.}{361}{2002}{57}.

753: 

754: \bibitem{julicher-etal-02}

755:   \Name{J\"{u}licher F., Ajdari A. \and Prost J.}

756:   \REVIEW{Rev. Mod. Phys.}{69}{1997}{1269}.

757: 

758: \bibitem{parrondo-02}

759:   \Name{Parrondo J.~M.~R. \and de~Cisneros B.~J.}

760:   \REVIEW{Appl. Phys. A}{75}{2002}{179}.

761: 

762: \bibitem{qian-98}

763:   \Name{Qian H.}

764:   \REVIEW{Phys. Rev. Lett.}{81}{1998}{3063}.

765: 

766: \bibitem{astumian-bio98}

767:   \Name{Astumian A.~D. \and Derenyl I.}

768:   \REVIEW{Eur. Biophys. J.}{27}{1998}{474}.

769: 

770: \bibitem{astumian-03}

771:   \Name{Astumian A.~D.}

772:   \REVIEW{Phys. Rev. Lett.}{91}{2003}{118102}.

773: 

774: 

775: \bibitem{westerhoff-86}

776:   \Name{Westerhoff H.~V., et al.}

777:   \REVIEW{Proc. Natl. Acad. Sci. U.S.A.}{83}{1986}{4734}.

778: 

779: \bibitem{chen-87}

780:   \Name{Chen Y-D}

781:   \REVIEW{Proc. Natl. Acad. Sci. U.S.A.}{84}{1987}{729}.

782: 

783: \bibitem{astumian-87}

784:   \Name{Astumian A.~D., et al.}

785:   \REVIEW{Proc. Natl. Acad. Sci. U.S.A.}{84}{1987}{434}.

786: 

787: \bibitem{tsong-88}

788:   \Name{Tsong T.~Y. \and Astumian R.~D.}

789:   \REVIEW{Ann. Rev. Physiol.}{50}{1988}{273}.

790: 

791: \bibitem{astumian-pra89}

792:   \Name{Astumian A.~D. \and Chock P.~B.}

793:   \REVIEW{Phys. Rev. A.}{39}{1989}{6416}.

794: 

795: \bibitem{astumian-jchph89}

796:   \Name{Astumian A.~D. \and Robertson B.}

797:   \REVIEW{J. Chem. Phys.}{91}{1989}{4891}.

798: 

799: \bibitem{robertson-90}

800:   \Name{Robertson B. \and Astumian A.~D.}

801:   \REVIEW{Biophys. J.}{58}{1990}{969}.

802: 

803: \bibitem{robertson-91}

804:   \Name{Robertson B. \and Astumian A.~D.}

805:   \REVIEW{J. Chem. Phys.}{94}{1991}{7414}.

806: 

807: \bibitem{markin-91}

808:   \Name{Markin V.~S. \and Astumian A.~D.}

809:   \REVIEW{Biophys. J.}{59}{1991}{1317}.

810: 

811: \bibitem{tsong-chang-02}

812:   \Name{Tsong T.~Y. \and Chang C.~H.},

813:   \REVIEW{AAAPS Bulletin}{13}{2003}{12}.

814: 

815: \bibitem{brouwer-98}

816:   \Name{Brouwer P.~W.}

817:   \REVIEW{Phys. Rev. B}{58}{1998}{R10135}.

818: 

819: \bibitem{makhlin-mirlin-01}

820:   \Name{Makhlin Y. \and Mirlin A.~D.}

821:   \REVIEW{Phys. Rev. Lett.}{87}{2001}{276803}.

822: 

823: 

824: \bibitem{moskalets-buttiker-02}

825:   \Name{Moskalets M. \and B\"{u}ttiker M.}

826:   \REVIEW{Phys. Rev. B}{66}{2002}{035306}.

827: 

828: \bibitem{pump_berry}

829:   \Name{Avron J.~E., et al.}

830:   \REVIEW{Phys. Rev. B}{62}{2000}{R10618}.

831: 

832: \bibitem{shi}

833:   \Name{Shi Y. \and Niu Q.}.

834:   \REVIEW{Europhys. Lett.}{59}{2002}{324}.

835:  

836: \bibitem{hannay-85}

837:   \Name{Hannay J.~H.}

838:   \REVIEW{J. Phys. A.}{18}{1985}{221}.

839: 

840: \bibitem{hill-book}

841:   \Name{Hill T.~L.}

842:   \Book{Free Energy Transduction and Biochemical Cycle Kinetics}

843:   \Publ{Springer}

844:   \Year{1995}. 

845: 

846: \bibitem{qian1-98} 

847:   \Name{Qian H.}

848:   \REVIEW{Phys. Rev. Lett}{81}{1998}{3063}.

849: 

850: 

851: \bibitem{qian-00} 

852:   \Name{Qian H., Qian M.}

853:   \REVIEW{Phys. Rev. Lett}{84}{2000}{2271}.

854: 

855: \bibitem{ao-04}

856:   \Name{P. Ao}

857:   \REVIEW{J. Phys. A: Math. Gen.}{37}{2004}{L25}.

858: 

859: \bibitem{kwon-05}

860:   \Name{Kwon C., Ao P., Thouless D.~J.}

861:   \REVIEW{Proc. Natl. Acad. Sci. U.S.A.}{102}{2005}{13029}.

862: 

863: \bibitem{niu}

864:   \Name{Niu Q., Sundaram G.}

865:   \REVIEW{Phys. Rev. B}{59}{1999}{14915}.

866: 

867: \bibitem{nazarovFCS}

868:   \Name{Bagrets D.~A. \and Nazarov Y.~V.}

869:   \REVIEW{Phys. Rev. B}{67}{2003}{085316}.

870: 

871: \bibitem{gillespie-77}

872:     \Name{Gillespie G.~T.}

873:     \REVIEW{J. Phys. Chem}{81}{1977}{2340}.

874: 

875: \bibitem{brenda}

876:   \Name{Pharkya P., Nikolaev E.~V. \and Maranas C.~D.}

877:   \REVIEW{Metab. Eng.}{5}{2003}{71}. Database release 2006.2.

878: 

879: \bibitem{knowles-albery-77}

880:   \Name{Knowles J.~R. \and Albery W.~J.}

881:   \REVIEW{Accounts Chem. Res.}{10}{1977}{105}.

882: 

883: 

884: \bibitem{garrison-88}

885:   \Name{Garrison J.~C. \and Wright E.~M.}

886:   \REVIEW{J. Phys. A: Math. Gen.}{39}{1988}{13727}.

887: 

888: \bibitem{vertex-1}

889:   \Name{Ao P. \and Thouless D.~J.}

890:   \REVIEW{Phys. Rev. Lett.}{76}{1993}{2158}.

891: 

892: \bibitem{vertex-2}

893:   \Name{Thouless D.~J., Ao P. \and Niu Q.}

894:   \REVIEW{Phys. Rev. Lett.}{76}{1996}{3758}.

895: 

896: \bibitem{Garanin}

897:   \Name{Schilling R., Vogelsberger M. \and Garanin D.~A.}

898:   \REVIEW{J. Phys. A: Math. Gen.}{39}{2006}{13727}.

899: 

900: \end{thebibliography}

901: 

902: 

903: \end{document}

904: 

905:  

906: 

907: 

908: 

909:  

910: 

911: