1: \documentstyle[12pt,aps,preprint,prbbib]{revtex}
2: %\documentstyle[aps,preprint]{revtex}
3: \newcommand{\be}{\begin{eqnarray}}
4: \newcommand{\ee}{\end{eqnarray}}
5: \newcommand{\la}{\langle}
6: \newcommand{\ra}{\rangle}
7: \newcommand{\p}{\partial}
8: \newcommand{\w}{{\omega}}
9: \newcommand{\D}{V}
10: \newcommand{\cl}{{\cal L}}
11: \newcommand{\no}{\nonumber}
12: \newcommand{\bmath}{\begin{mathletters}}
13: \newcommand{\emath}{\end{mathletters}}
14:
15: \renewcommand{\baselinestretch}{1.8}
16: \draft
17: \begin{document}
18:
19: \preprint{ \scriptsize SUBMITTED TO BRIAN HEAD SPECIAL ISSUE OF JPC\hspace{4pt}}
20:
21: \title{Electronic coherence in mixed-valence systems:\\ Spectral analysis}
22:
23: \author{Younjoon Jung, Robert J. Silbey,
24: and Jianshu Cao\footnote{To whom correspondence should be addressed. Electronic mail: jianshu@mit.edu}}
25: \address{ Department of Chemistry, Massachusetts Institute of Technology \\
26: Cambridge, MA 01239}
27:
28: \date{\today}
29: \maketitle
30: \vspace{-0.5truein}
31:
32: \begin{abstract}
33:
34: The electron transfer kinetics of mixed-valence systems
35: is studied via solving the eigen-structure of the two-state non-adiabatic
36: diffusion operator for a wide range of
37: electronic coupling constants and energy bias constants.
38: The calculated spectral structure consists of three branches in
39: the eigen-diagram, a real branch corresponding to exponential
40: or multi-exponential decay and two symmetric branches
41: corresponding to population oscillations between donor and acceptor states.
42: The observed electronic coherence is shown as a result of
43: underdamped Rabi oscillations in an overdamped solvent environment.
44: The time-evolution of electron population is calculated by applying
45: the propagator constructed from the eigen-solution to the non-equilibrium
46: initial preparation, and it agrees perfectly with the result of
47: a direct numerical propagation of the density matrix.
48: The resulting population dynamics confirms that
49: increasing the energy bias destroys electronic coherence.
50:
51: \end{abstract}
52:
53: \newpage
54:
55: \section{INTRODUCTION}
56: Quantum coherence in the dynamics of condensed phase systems
57: has become a subject of recent experimental and theoretical studies.
58: A central issue is the observability of electronic coherence
59: in electron transfer systems given the fast dephasing time
60: in many-body quantum systems.
61: Experimentally, with the advance in ultrafast laser technology,
62: oscillations in electronic dynamics have been observed in photo-synthetic
63: reaction centers and other electron transfer systems
64: and are believed to arise from vibrational and/or electronic
65: coherence.\cite{vos-nat-93,jonas-jpc-95,arnett-jacs-95}
66: Accurate measurements on photo-induced electron transfer
67: in mixed-valence compounds have demonstrated oscillations
68: in electronic populations on the femtosecond time-scale.\cite{vos-nat-93,reid-jpc-95}
69: Theoretically, detailed path-integral simulations
70: suggest that such oscillations take place in
71: electron transfer systems with large electronic coupling constants
72: and are sensitive to the initial preparation of the bath modes
73: associated with the transfer processes.
74: Lucke \textit{et al.}\cite{lucke-jcp-97} extended the non-interacting blip approximation
75: to incorporate the non-equilibrium initial preparation
76: and carried out extensive path-integral
77: quantum dynamics simulations
78: for electron transfer reactions. According to their findings,
79: large-amplitude oscillations are most likely to be observed
80: in symmetric mixed-valence systems
81: that are nearly adiabatic and with initial configurations
82: that are centered in the Landau-Zener crossing region.
83: Using the transfer matrix technique,\cite{makarov-cpl-94}
84: Evans, Nitzan, and Ratner\cite{evans-jcp-98}
85: calculated short-time evolution for the photo-induced
86: electron transfer reaction in ${\rm (NH_3)_5 Fe^{II}(CN)Ru^{III}(CN)_5}$.
87: Their results show fast oscillations in the electronic
88: population on the short time-scale(20 fs) followed by a slower population
89: relaxation on the long time-scale(100 fs).
90: They pointed out that these fast oscillations arise
91: as the wave-function oscillates coherently between the donor and acceptor
92: states. The calculated long-time decay rate is considerably smaller
93: than the prediction by the golden-rule formulae,\cite{coalson-jcp-94,cho-jcp-95}
94: confirming the inadequacy of non-adiabatic rate theory
95: in studying mixed-valence systems.
96:
97:
98: In fact, a simple classical argument helps understand the nature
99: of the observed oscillations.
100: As a function of the ratio between $\lambda$
101: (the bath reorganization energy) and $\D$ (the electronic coupling constant),
102: there is a thermodynamic transition from the localized
103: electronic state in a double-well potential
104: to the delocalized electronic state in a single well
105: potential.\cite{harris-jcp-83,silbey-jcp-84,carmeli-jcp-85,chandler-liquid-91,leggett-rmp-87}
106: (i) In the localized regime ($ \lambda \gg \D$),
107: the large reorganization energy destroys electronic coherence;
108: hence, electron transfer is an incoherent rate process,
109: which can be described by the non-interacting
110: blip approximation or golden-rule rate in the non-adiabatic limit
111: and by transition state theory in the adiabatic limit.\cite{marcus-bba-85,newton-arpc-84,bader-jcp-90}
112: (ii) In the delocalized regime ($ \lambda\le \D $), the electronic
113: wave function extends to both the donor and acceptor states
114: and electronic coherence persists over several oscillations.\cite{harris-jcp-83}
115: For mixed-valence compounds, the electronic coupling constant
116: is estimated to be in the range of $10^3 \mbox{cm}^{-1}$,
117: which is in the same order as the reorganization energy.\cite{vos-nat-93,evans-jcp-98}
118: Therefore, the observed oscillations and relaxation in mixed-valence
119: systems are the consequence of a highly non-equilibrium
120: coherence transfer process.
121:
122: Due to the delocalization nature of electronic states,
123: an adiabatic picture\cite{cao-cpl-99}
124: is more useful than the diabatic representation
125: for analyzing the short-time dynamics in strongly-coupled systems.
126: In this picture, electronic coherence arises from Rabi oscillations
127: between two adiabatic surfaces and decays because of electronic dephasing.
128: Further, initial preparation and wave-packet dynamics can
129: modulate Rabi oscillations and the overall electronic dynamics.
130: Thus, the adiabatic representation provides a simple picture for
131: mixed-valence systems as well as a simple analytical method
132: to model fast electron dynamics initiated by laser pulses.
133:
134: As a general approach to describe condensed phase dynamics,
135: we recently proposed a spectral analysis method,\cite{cao-jcp-00}
136: which is based on eigen-structures of dissipative
137: systems instead of dynamic trajectories.
138: An important application of the approach is to analyze
139: a set of two-state diffusion equations,
140: which was first used by Zusman to treat solvent effects on electron transfer
141: in the non-adiabatic limit.
142: The analysis allows us to characterize
143: multiple time-scales in electron transfer processes
144: including vibrational relaxation, electronic coherence,
145: activated curve crossing or barrier crossing.
146: With this unified approach, the observed rate behavior, bi-exponential and
147: multi-exponential decay, and population
148: oscillations are different components of the same kinetic spectrum.
149: Thus, several existing theoretical models, developed for
150: limited cases of electron transfer, can be analyzed, tested, and extended.
151: In particular, rate constants extracted from the analysis bridge smoothly
152: between the adiabatic and non-adiabatic limits, and the kinetic
153: spectrum in the large coupling regime reveals the nature of
154: the localization-delocalization transition as the consequence of
155: two competing mechanisms.
156:
157:
158: In this paper, the spectral analysis approach developed
159: in Ref.\onlinecite{cao-jcp-00} is employed to study
160: the electron transfer dynamics in mixed-valence systems.
161: We invoke the non-adiabatic diffusion equation proposed by Zusman
162: to describe the electron transfer process in the over-damped
163: solvent regime.
164: As discussed earlier, electron transfer in mixed-valence systems takes place
165: in a different kinetic regime from the thermal activated regime described
166: by Marcus theory. Thus, the time-scale separation is not satisfied, and
167: multi-exponential decay and oscillations are
168: intrinsic nature of electron transfer kinetics.
169: As a result, the kinetic spectra exhibit
170: bifurcation, coalescence, and other complicated patterns.
171: Careful examination of these patterns
172: reveals the underlying mechanisms in mixed-valence systems.
173:
174: The rest of the paper is organized as follows:
175: The spectral structure of the non-adiabatic diffusion equation is
176: formulated in Sec.~II.
177: Numerical examples of the spectral structure of
178: strongly mixed electron transfer systems are presented and discussed
179: in Sec.~III and concluding remarks are given in Sec.~IV.
180:
181:
182: \section{THEORY}
183: There have been extensive studies of the solvent effect on
184: electron transfer dynamics in literature with various
185: approaches.\cite{zusman-cp-80,calef-jpc-83,hynes-jpc-86,garg-jcp-85,sparpaglione-jcp-88}
186: One of the most extensively studied models for quantum dissipation
187: is the spin-boson Hamiltonian,\cite{leggett-rmp-87,garg-jcp-85}
188: \be
189: H_{SB}={\epsilon \over 2}\sigma_{z}+V\sigma_{x}+
190: \sum_{\alpha}{p_{\alpha}^2\over 2m_{\alpha}}
191: +\sum_{\alpha}{1\over 2}m_{\alpha}\omega_{\alpha}^2
192: {\left(x_{\alpha}-\sigma_{z}
193: {c_{\alpha}\over{m_{\alpha}\omega_{\alpha}^2}}\right)}^2,
194: \label{sb}
195: \ee
196: where $\epsilon$ is the energy bias between the two electronic
197: states, $\D$ is the electronic coupling constant,
198: $\sigma_{z}$ and $\sigma_{x}$ are the usual Pauli matrices, and
199: $\{x_{\alpha}$,$p_{\alpha}\}$ represents the bath degree of freedom
200: with mass $m_{\alpha}$,
201: frequency $\omega_{\alpha}$, and the coupling constant $c_{\alpha}$.
202: In this model effects of the bath modes on the dynamics of the system
203: can be described via the spectral density defined by,
204: \be
205: J(\omega)={\pi\over2}\sum_{\alpha}{{c_{\alpha}^2}\over{m_{\alpha}\omega_{\alpha}}}\delta(\omega-\omega_{\alpha}). \label{spec}
206: \ee
207:
208: Equivalently, the spin-boson Hamiltonian in Eq.~(\ref{sb})
209: can be separated into the electronic two-level part $H_{TLS}$ and
210: the nuclear bath part $H_{B}$,
211: \be
212: H_{SB}= H_{TLS} + H_B.
213: \label{two}
214: \ee
215: The two-level part of the Hamiltonian can be explicitly written as
216: \be
217: H_{TLS}(E)
218: =U_{1}(E)|1\ra\la 1|+U_{2}(E)|2\ra\la 2|+V(|1\ra\la 2|+|2\ra\la 1|),
219: \ee
220: where the diabatic energy surfaces $U_1(E)$ and $U_2(E)$
221: are functions of the stochastic variable $E$,
222: which represents the polarization energy
223: for a given solvent configuration.\cite{zusman-cp-80}
224: The transformation from the spin-boson Hamiltonian to the two-level
225: system Hamiltonian has been shown in the literature\cite{garg-jcp-85,cao-jcp-97}
226: by the identity,
227: \be
228: E(\{x_{\alpha}\})=\sum_{\alpha}c_{\alpha}x_{\alpha}.
229: \ee
230: It is worthwhile to mention that the polarization energy $E$ was
231: recognized as the reaction coordinate by Marcus in formulating
232: non-adiabatic electron transfer theory.\cite{marcus-bba-85}
233: Since the electron transfer process involves the collective motion of
234: a large number of solvent degrees of freedom
235: and the two-level system is linearly coupled to the harmonic bath modes
236: in the spin-boson Hamiltonian in Eq.~(\ref{sb}),
237: the functional form for the free energy surface is harmonic,\cite{onuchic-jcp-93}
238: thus giving
239: \be
240: U_1(E)&=& {(E+\lambda)^2 \over {4 \lambda}}, \label{5a} \\
241: U_2(E)&=& {(E-\lambda)^2 \over {4 \lambda}} + \epsilon, \label{5b}
242: \ee
243: where $\lambda$ is the reorganization energy, which is related to the
244: parameters in Eq.~(\ref{sb}),
245: \be
246: \lambda=\sum_{\alpha}{{c_{\alpha}^2}\over{2m_{\alpha}\omega_{\alpha}^2}}
247: ={1\over\pi}\int d\omega{J(\omega)\over\omega}. \label{lambda}
248: \ee
249:
250: Considering the fact that electron transfer processes
251: are usually probed at room temperature in polar solvents,
252: we can treat the bath degrees of freedom in $H_{B}$ classically.
253: Then, the spin-boson Hamiltonian in Eq.~(\ref{two}) can
254: be used to derive a two-level classical equation of motion,
255: \be
256: i{\p \over {\p t}}\rho(t)=\cl\rho(t)=(\cl_B+ \cl_{TLS})\rho(t), \label{1}
257: \ee
258: where $i\cl_{B}=\{H_{B},\}$ is the Poisson operator for the classical bath and
259: $\cl_{TLS}=[H_{TLS},\ ]/\hbar$
260: is the Liouville operator for the two level system.
261: Explicitly, we express Eq.~(\ref{1}) in terms of the density matrix elements,
262: \bmath
263: \be
264: \dot\rho_{1} &=& \cl_{1} \rho_{1} + i V (\rho_{12} - \rho_{21}), \label{2a} \\
265: \dot\rho_{2} &=& \cl_{2} \rho_{2} - i V (\rho_{12} - \rho_{21}), \label{2b} \\
266: \dot\rho_{12} &=& \cl_{12} \rho_{12} - i \omega_{12} \rho_{12} +
267: i V (\rho_{1}-\rho_{2}), \label{2c} \\
268: \dot\rho_{21} &=& \cl_{21} \rho_{21} + i \omega_{12} \rho_{21} -
269: i V (\rho_{1}-\rho_{2}), \label{2d}
270: \ee
271: \emath
272: where the Planck constant $\hbar$ is set to unity for simplicity,
273: $\rho_{i}$ is the diagonal matrix element for electronic population,
274: and $\rho_{ij}$ is the off-diagonal matrix element for electronic coherence.
275: Here, $\cl$ describes the relaxation process of classical bath,
276: with $\cl_{i}$ defined on the free energy surface for the $i$th
277: electronic state, and with $\cl_{12}$ and $\cl_{21}$ defined on the
278: averaged free energy surface.
279: This set of semi-classical two-state equations has
280: been previously derived in different context by several authors.
281: \cite{zusman-cp-80,garg-jcp-85,yang-jcp-89} It should be mentioned that the mapping from
282: the spin-boson Hamiltonian into the Zusman model requires the Lorentzian
283: form of the spectral density,
284: \be
285: J(\omega)=2\lambda{{\omega\omega_c}\over{\omega^2+\omega_c^2}}. \label{lorentz}
286: \ee
287:
288: Furthermore, we note that many chemically and biologically
289: important electron transfer processes
290: take place in the over-damped solvent environment.
291: Therefore, to describe the density matrix evolution in the
292: electron transfer kinetics in the mixed-valence system, we
293: invoke the non-adiabatic diffusion equation proposed by Zusman.\cite{zusman-cp-80}
294: Then, the bath relaxation operators in Eq.~(\ref{1}) are
295: one-dimensional Fokker-Planck operators $\cl_{ij}$,
296: \be
297: &&\cl_{i} = D_E {\p \over \p E} \left ({\p \over \p E} + \beta {\p U_i(E)
298: \over \p E} \right ), \label{3a} \\
299: &&\cl_{12} = \cl_{21} = {{\cl_{11}+\cl_{22}} \over 2}=D_E {\p \over \p E}
300: \left ({\p \over \p E} + \beta {\p \bar{U}(E) \over \p E} \right) . \label{3b}
301: \ee
302: where $\beta={1/k_BT}$,
303: $\bar{U}$ and $\omega_{12}$ are the average and the difference
304: of the two free energy surfaces, respectively,
305: \be
306: \bar{U}(E)={{U_1(E)+U_2(E)}\over{2}},\\ \label{4a}
307: \omega_{12}(E)=U_1(E)-U_2(E). \label{4b}
308: \ee
309: The energy diffusion constant $D_E$ is defined as
310: \be
311: D_E=\Omega\Delta_{E}^2, \label{7}
312: \ee
313: where $\Delta_{E}^2$ is the mean square fluctuation
314: of the solvent polarization energy
315: $$
316: \Delta_{E}^2={\la E^2\ra}=2\lambda k_BT, \label{6}
317: $$
318: and $\tau_D=1/\Omega$ is the
319: the characteristic timescale of the Debye solvent.
320: The correlation function of the solvent polarization energy is given by
321: \be
322: C(t)=\la E(t)E(0)\ra=\Delta_E^2 \exp(-\Omega t).
323: \ee
324: Note that since the nuclear dynamics is modeled by the Fokker-Planck
325: operator, the possibility of the vibrational coherence
326: is excluded in this model of electron transfer dynamics.
327: It is worthwhile to mention that one can obtain
328: the non-adiabatic diffusion equation starting from the spin-boson Hamiltonian,
329: by first deriving the evolution equation for the quantum dissipative dynamics,
330: and then taking the semi-classical limit using the Wigner distribution
331: functions, and finally assuming the over-damped diffusion limit.\cite{garg-jcp-85}
332:
333:
334: We investigate the spectral structure of the non-adiabatic diffusion
335: operator by calculating the eigenvalues $\{-Z_{\nu}\}$ and the corresponding
336: eigen-functions $\{|\psi_{\nu}\ra\}$. Hereafter we use Greek indices to denote
337: the eigenstates and Latin indices to denote the basis states of
338: the non-adiabatic diffusion operator.
339: Because the non-adiabatic Liouville operator is non-Hermitian,
340: the eigenvalues are generally given by complex values, and the right and left
341: eigen-functions corresponding to the same eigenvalue are not simply the
342: Hermitian conjugate to each other.\cite{simons-cp-73}
343: For a given eigen-value $Z_{\nu}$,
344: the right and left eigen-functions of the non-adiabatic
345: diffusion operator are obtained from
346: \be
347: \cl |\psi_{\nu}^R \ra &=& -Z_{\nu}|\psi_{\nu}^R \ra,
348: \label{8a}\\
349: \la \psi_{\nu}^L| \cl &=& -Z_{\nu}\la \psi_{\nu}^L |.
350: \label{8b}
351: \ee
352: The method of eigenfunction solution is well known for the diffusion
353: process on the harmonic potential energy surface.\cite{risken-fpe-84}
354: For a single quadratic potential $U(x)=\frac{1}{2} m\omega^2 x^2$,
355: the one-dimensional Fokker-Planck operator
356: $\cl_{FP}=D({{\p^2}\over{\p x^2}}+{\beta {\p \over {\p x}} U'})$
357: can be transformed into the quantum mechanical Hamiltonian in imaginary time,
358: \be
359: H_s =-e^{\beta U(x)/2}\cl_{FP}e^{-\beta U(x)/ 2}
360: =-{1 \over {2\mu}} {\p ^2 \over {\p x^2}} + V_s(x), \label{11}
361: \ee
362: where $\mu^{-1}=2D$, and the quadratic potential is
363: \be
364: V_s(x) = D \left[{1 \over 4} (\beta U^{'}(x))^2 -{1 \over 2} \beta U^{''}(x) \right]
365: = {1 \over 2} {\mu \gamma^2 x^2} - {\gamma \over 2}, \label{12}
366: \ee
367: with $\gamma=D m \omega^2/k_BT$.
368: Since the transformed potential in Eq.~(\ref{12}) is just the same form as for
369: a simple harmonic oscillator with zero point energy compensation,
370: the eigenvalues and the eigen-functions for the original Fokker-Planck
371: operator can be constructed immediately
372: from the eigen-solutions of the harmonic oscillator Hamiltonian.
373: Unlike the diffusion problem on the single potential energy surface, there
374: have been limited studies on the non-adiabatic diffusion problem involving
375: more than one potential energy surface.
376: In this aspect, Cukier and co-workers have calculated
377: the electron transfer rate by calculating the lowest eigenvalue of
378: the non-adiabatic diffusion equation; however,
379: their calculation was limited to the weak-coupling regime
380: where the Zusman rate is applicable.\cite{yang-jcp-89}
381:
382: An important issue in solving the non-adiabatic diffusion equation for
383: electron transfer is the choice of the basis functions
384: since three different free energy surfaces are involved in Eq.~(\ref{1}):
385: two diabatic surfaces for the population density matrix elements and one
386: averaged surface for the coherence density matrix element.
387: In this paper, the eigen-functions of $\cl_{12}$ are used as our basis set
388: to represent the non-adiabatic diffusion equation. In principle, one could
389: have chosen the eigen-functions of $\cl_{1}$ or $\cl_{2}$ as basis functions,
390: however, in that case one has to evaluate appropriate Franck-Condon factors
391: when calculating the coupling matrix elements even with
392: the Condon approximation.
393: The Fokker-Planck operator $\cl_{12}$ is defined on the averaged
394: harmonic potential centered at $E=0$, and its eigen-solutions are
395: \be
396: \cl_{12}| \phi^R_n \ra&=&-n \Omega | \phi^R_n \ra, \label{13a} \\
397: \la \phi^L_n|\cl_{12} &=&-n \Omega \la \phi^L_n |, \label{13b}
398: \ee
399: where the right and left eigen functions are
400: \be
401: \phi^R_n(E)={1\over{(2^n n!)^{1\over2}(2\pi\Delta_E^2)^{1\over4}}}
402: \exp\left(-{{E^2}\over{2 \Delta_E^2}}\right) H_n\left({E\over {\sqrt{2} \Delta_E}}
403: \right), \label{14}
404: \ee
405: and
406: \be
407: \phi^L_n(E)={1\over{(2^n n!)^{1\over2}(2\pi\Delta_E^2)^{1\over4}}}
408: H_n\left({E\over {\sqrt{2} \Delta_E}}\right), \label{15}
409: \ee
410: where $H_n$ is the $n$th order Hermite polynomial.
411: As shown below, this choice of the basis set is convenient
412: for our purpose.
413:
414: To be consistent with the $\cl_{12}$ basis set, we
415: separate the real and imaginary parts of the coherence density matrix, namely,
416: $u=$Re$\rho_{12}$ and $v=$Im$\rho_{12}$, and rewrite Eq.~(\ref{1}) as
417: \bmath
418: \be
419: \dot\rho_{1} &=& (\cl_{12}+\delta\cl) \rho_1 - 2 V v, \label{9a} \\
420: \dot\rho_{2} &=& (\cl_{12}-\delta\cl) \rho_2 + 2 V v, \label{9b} \\
421: \dot u &=& \cl_{12} u + \omega_{12} v, \label{9c} \\
422: \dot v &=& \cl_{12} v - \omega_{12} u + V (\rho_{1}-\rho_{2}), \label{9d}
423: \ee
424: \emath
425: where we have defined $\delta\cl$ as
426: \be
427: \delta\cl\ = {{\cl_{11}-\cl_{22}} \over 2}. \label{10}
428: \ee
429: Then, all the relevant operators in Eqs.~(\ref{9a})-(\ref{9d}) can be evaluated
430: in terms of the right and left eigen-functions of $\cl_{12}$, giving
431: \be
432: \la \phi_{n}^{L} | \cl_{12} | \phi_{m}^{R} \ra&=&-n\Omega\delta_{nm}, \label{16a} \\
433: \la \phi_{n}^{L} | \delta\cl | \phi_{m}^{R} \ra&=&-\Omega\sqrt{{\lambda}
434: \over{2k_BT}}\sqrt{m+1}\delta_{n,m+1}, \label{16b} \\
435: \la \phi_{n}^{L} | \omega_{12} | \phi_{m}^{R} \ra&=&\sqrt{2\lambda k_BT}
436: (\sqrt{m}\delta_{n,m-1}+\sqrt{m+1} \delta_{n,m+1})-\epsilon\delta_{nm}, \label{16c} \\
437: \la \phi_{n}^{L} | V| \phi_{m}^{R} \ra&=&V\delta_{nm}, \label{16d}
438: \ee
439: where we assume the Condon approximation, i.e., the
440: electronic coupling matrix element is independent
441: of the solvent degrees of freedom.
442: With the basis set, we can expand the density matrix elements as
443: \bmath
444: \be
445: \rho_{1}(E,t)&=&\sum_{n=0}^{\infty} a_{n}(t)\phi_{n}^{R}(E), \label{17a} \\
446: \rho_{2}(E,t)&=&\sum_{n=0}^{\infty} b_{n}(t)\phi_{n}^{R}(E), \label{17b} \\
447: u(E,t)&=&\sum_{n=0}^{\infty} c_{n}(t)\phi_{n}^{R}(E), \label{17c} \\
448: v(E,t)&=&\sum_{n=0}^{\infty} d_{n}(t)\phi_{n}^{R}(E). \label{17d}
449: \ee
450: \emath
451: Substituting Eqs.~(\ref{17a})-(\ref{17d})
452: into the eigenvalue equation Eq.~(\ref{8a}),
453: we have the following coupled linear equations
454: \bmath
455: \be
456: -Z_{\nu} a_{n}&=&-n\Omega a_{n}-\Omega\sqrt{\lambda\over{2k_BT}}\sqrt{n}a_{n-1}
457: -2 V d_{n}, \label{18a} \\
458: -Z_{\nu} b_{n}&=&-n\Omega b_{n}+\Omega\sqrt{\lambda\over{2k_BT}}\sqrt{n}b_{n-1}
459: +2 V d_{n}, \label{18b} \\
460: -Z_{\nu} c_{n}&=&-n\Omega c_{n}+\sqrt{2\lambda k_BT}(\sqrt{n+1}d_{n+1}+\sqrt{n}d_{n-1})
461: -\epsilon d_{n}, \label{18c} \\
462: -Z_{\nu} d_{n}&=&-n\Omega d_{n}-\sqrt{2\lambda k_BT}(\sqrt{n+1}c_{n+1}+\sqrt{n}c_{n-1})+\epsilon c_{n} +V (a_{n}-b_{n}), \label{18d}
463: \ee
464: \emath
465: which is an explicit basis set representation for
466: the two-state diffusion operator in Eq.~(\ref{1}).
467: The linear equations for the left eigen-solution as defined by Eq.~(\ref{8b})
468: can be written by the transpose of Eqs.~(\ref{18a})-(\ref{18d}).
469: Diagonalizing the $4N\times 4N$ matrix ($N$ = number of basis functions)
470: defined in Eqs.~(\ref{18a})-(\ref{18d}),
471: we obtain the eigenvalues $-Z_{\nu}$ and the corresponding eigenvectors
472: of the non-adiabatic diffusion operator,
473: \be
474: |\psi_{\nu}^{R}\ra&=&\sum_{n} R_{n\nu}|\phi_{n}^{R}\ra, \label{eig1} \\
475: \la\psi_{\nu}^{L}|&=&\sum_{n} L_{\nu n}\la \phi_{n}^{L}|, \label{eig2}
476: \ee
477: where $R_{n\nu}$ and $L_{\nu n}$ are elements of the
478: transformation matrices.
479:
480: In general,
481: due to the non-Hermitian nature of the non-adiabatic diffusion operator,
482: the right and left eigen-functions do not form an orthogonal set by themselves.
483: However, when the eigenvalues are all non-degenerate,
484: the left and right eigen-functions form an orthogonal and complete set in
485: dual Hilbert space.\cite{dahmen-condmat-99}
486: Explicitly, we have
487: \be
488: \sum_{n=0}L_{\nu n}R_{n\nu^{'}}=\delta_{\nu\nu^{'}},
489: \label{19}
490: \ee
491: for the orthogonality and
492: \be
493: \sum_{\nu} R_{n\nu}L_{\nu m}=\delta_{nm},
494: \label{20}
495: \ee
496: for the completeness. Using these properties,
497: we can construct the real time propagator for the operator $\cl$ as
498: \be
499: G(t)=\sum_{\nu}|\psi_{\nu}^{R}\ra \la\psi_{\nu}^{L}|
500: e^{-Z_{\nu}t}, \label{21}
501: \ee
502: and express the time evolution of the density matrix
503: by projecting a given initial distribution onto the eigenstates, giving
504: \be
505: |\rho(t)\ra=G(t)|\rho(0)\ra=\sum_{\nu}|\psi_{\nu}^{R}\ra
506: \la\psi_{\nu}^{L}|\rho(0)\ra e^{-Z_{\nu}t}. \label{22}
507: \ee
508: Hence, the eigen-solution to the two-state non-adiabatic diffusion equation
509: leads to a complete description of electron transfer dynamics.
510:
511: \section{RESULTS AND DISCUSSIONS}
512:
513: In the section, we present the spectral
514: structure of the non-adiabatic diffusion operator by diagonalizing
515: its matrix representation in Eqs.~(\ref{18a})-(\ref{18d}).
516: In principle, we need infinite
517: number of basis functions to diagonalize the non-adiabatic diffusion operator,
518: however, in practice, we have to truncate our basis set at some finite number.
519: In all the calculations below,
520: we have used $N=50-200$ to diagonalize the $4N\times 4N$ matrix
521: and the effect of finite number basis on the spectral structure
522: has been carefully examined.
523:
524: \subsection{Spectral Structure}
525: \subsubsection{Mixed-valence systems}
526: In the mixed-valence compounds, the electronic coupling constant
527: has the same order of magnitude as the reorganization energy
528: and the electron transfer dynamics
529: is usually probed experimentally at room temperature in polar solvents.
530: To study this process, Evans, Nitzan, and Ratner\cite{evans-jcp-98}
531: carried out real time path-integral simulations for the photo-induced
532: electron transfer reaction in ${\rm (NH_3)_5 Fe^{II}(CN)Ru^{III}(CN)_5}$.
533: Based on their model,
534: we chose the parameters for the calculation shown in Fig.~1
535: as $\beta\Omega=0.6716$, $\beta\lambda=18.225$, $\beta\D=11.99$, and
536: $\beta\epsilon=18.705$. As mentioned in the introduction
537: the mapping between the spin-boson Hamiltonian and
538: the semi-classical Zusman equation is not rigorously defined.
539: For example, for the non-adiabatic diffusion equation, the solvation energy
540: correlation function takes an exponential form with the rate $\Omega$,
541: whereas, for the spin-boson model Hamiltonian,
542: it depends on the functional form of the spectral density.
543: It can be shown that the Ohmic spectral density with an exponential
544: cut-off $\omega_{c}$
545: \be
546: J(\omega)=\eta\omega\exp(-\omega/\omega_{c}),
547: \ee
548: used in the calculation of Evans \textit{et al.}, leads to
549: an energy correlation function with a Lorentzian form
550: at high temperature,\cite{garg-jcp-85}
551: \be
552: C_{SB}(t)\approx {{2\eta\omega_{c} k_BT}\over{\pi}}{1\over{1+(\omega_{c}t)^2}}.
553: \ee
554: Then, the relaxation rate $\Omega$ used in our calculation
555: is taken as the inverse of the mean survival time of $C_{SB}(t)$,
556: which is $\Omega=2\omega_{c}/{\pi}$.
557:
558: In Fig.~1 the spectral structure of the non-adiabatic operator is shown in
559: complex space. We have used $N=200(4N=800)$ basis functions to calculate the
560: eigenvalues. To remove the effect of finite basis set from the resulting
561: spectral structure,
562: we only show the first 400 eigenvalues in the complex plane.
563: Since the non-adiabatic diffusion operator is non-Hermitian,
564: the resulting spectrum shows complex conjugate paired eigenvalues
565: as well as real eigenvalues, giving rise to the tree structure
566: with three major branches (which we will call the \textit{eigen-tree}).
567: In Fig~.1, we separate the real and imaginary parts of eigenvalue by
568: \be
569: -Z_{\nu}=-k_{\nu}-i\omega_{\nu}. \label{23}
570: \ee
571: Obviously, the real part, $k_{\nu}$, is always negative as all
572: non-equilibrium physical quantities decay to zero at time infinity,
573: and it scales linearly with the index $\nu$ since the relaxation rate
574: corresponding to the $n$th basis state ${\phi_{n}}$ is proportional to $n$.
575: In general, the relative magnitudes of real and imaginary parts of
576: eigenvalues determine the time-evolution of the density matrix:
577: the real eigenvalues correspond to the simple exponential decay components
578: and the complex conjugate paired eigenvalues correspond to
579: the damped oscillation components.
580:
581: To classify the eigenvalues quantitatively
582: according to their dynamic behavior,
583: we introduce the dimensionless quantity $\theta_{\nu}$
584: \be
585: \theta_{\nu} \equiv { {2\pi k_{\nu}} \over {|{\omega_{\nu}}|} }, \label{24}
586: \ee
587: where $k_{\nu}$ is the decay rate and $2\pi/\omega_{\nu}$
588: is the oscillation period. The time-evolution of the density matrix
589: component associated with the eigenvalue
590: $Z_{\nu}$ is an exponential decay if $\theta_{\nu} =\infty$,
591: an under-damped oscillation if $\theta_{\nu} >1$,
592: and a damped oscillation if $\theta_{\nu}\le 1$.
593: The relative amplitude of the each component
594: depends on the overlap matrix element between the initial
595: density matrix and the eigenstate.
596: As an approximate criterion for the classification of the eigenvalues,
597: the slope corresponding to $\theta_{\nu}=1$ is shown
598: in the eigen-tree diagram in Fig.~1.
599: There are a few eigenstates around and below the $\theta_{\nu}=1$ line,
600: with a typical rate of $\beta k_{\nu}\approx 5$.
601: For the parameters used in the calculation,
602: $\beta$ corresponds to $\sim$170 fs in real time, and,
603: therefore, these eigenstates exhibits damped oscillations with
604: a period and a decay time in the femtosecond regime.
605: In their real-time path integral simulations,
606: Evans \textit{et al.} showed that the population in the acceptor state
607: oscillates with a few femtosecond period and these oscillation decays
608: in within 20 femtoseconds.
609: Thus, qualitative features of the electron transfer dynamics
610: can be predicted and understood
611: from a careful examination of the spectral structure.
612: Since the spectral analysis presented here
613: is based on the semi-classical diffusion equation while the
614: path-integral study is based on the quantum mechanical spin-boson Hamiltonian,
615: the comparison between the two approaches
616: is expected to be qualitative.
617: In the following subsection, further analysis
618: reveals the nature of these oscillations.
619:
620: \subsubsection{Dependence on the coupling constant $\D$}
621: To examine the underlying spectral structure in more details,
622: eigenvalues of the non-adiabatic diffusion operator are plotted
623: as functions of the electronic coupling constant in Fig.~2.
624: All the parameters except for the electronic coupling constant
625: are the same as used in Fig.~1.
626:
627: In Fig.~2(a), the real parts of the first 20 eigenvalues are shown
628: as functions of the electronic coupling constant.
629: Note that eigenvalues corresponding to complex conjugate pairs
630: have the same real part, thus they coalesce in the real eigenvalue diagram.
631: When the coupling constant is very small $(\beta\D \ll 1)$,
632: the real part of the first non-zero eigenvalue is very well separated
633: from the eigenvalues of excited states,
634: so the dynamics of electron transfer can be considered as a
635: incoherent rate process with a well-defined rate constant, $k_1$.
636: When the coupling constant is larger $(\beta\D \approx 1)$,
637: the first excited state becomes close to the second excited state, and
638: they start to merge into a complex conjugate pair. If the
639: coupling constant increases further, eigen-values show a
640: bifurcation behavior at $\beta\D \approx 10$.
641: Therefore, in this regime, the electron transfer kinetics
642: show multiple time-scale relaxation as well as coherent oscillation.
643: The complicated behavior of
644: coalescence and bifurcation in the real eigenvalue
645: appears more frequently at higher states.
646:
647:
648: Another interesting feature of the real eigenvalue diagram
649: is that a set of real eigenvalues decreases consistently
650: as the coupling constant increases from zero.
651: It turns out that these eigenstates take on large imaginary parts,
652: which are responsible for the onset
653: of the imaginary branches of the eigen-tree.
654: In Fig.~2(b), the imaginary parts of the lowest 30 eigenvalues are plotted
655: as functions of the coupling constant. Interestingly,
656: the imaginary part of the eigenvalue increases approximately
657: linearly with the coupling constant at large coupling regime.
658: In fact, the dependence on the coupling constant is similar to that of
659: the Rabi frequency for the two-level system,
660: \be
661: \Omega_{R}=\sqrt{\epsilon^2+4\D^2},
662: \ee
663: which is shown in Fig.~2(b).
664: As pointed out in a recent paper,\cite{cao-cpl-99}
665: electronic coherence in mixed-valence systems arises from
666: Rabi oscillations between two adiabatic surfaces
667: and decays because of dephasing.
668:
669: To demonstrate
670: the correlation of the real and imaginary parts of the eigenvalues as
671: functions of the coupling constant, we present a three dimensional plot of
672: the spectral structure in Fig.~2(c). For clarity,
673: only the positive branches of the imaginary eigenvalues are shown.
674: If we compare Fig.~2(c) with Fig.~2(a),
675: the very rapidly decaying states shown in Fig.~2(a)
676: take on large imaginary parts corresponding to the Rabi oscillations
677: as the coupling constant increases, and
678: these states are responsible for the onset
679: of the imaginary branches in the eigen-tree for
680: the mixed-valence system shown in Fig.~1.
681:
682: \subsection{Density Matrix Propagation}
683:
684: To check the validity of the spectral analysis as a density matrix propagation
685: scheme, we calculated the time-evolution of the density matrix by applying the
686: propagator defined by Eq.~(\ref{21}) to the initial density matrix for various energy biases.
687: Although it may seem straightforward to use the spectral method as a propagation scheme, the case for a non-Hermitian operator is not trivial and has not been explored. The main reason is that though the left and right
688: eigen-functions of a non-Hermitian operator can be shown to form a
689: bi-orthogonal set for the non-degenerate eigenvalue case,
690: numerically these eigen-functions may not be stable enough to be used as a
691: complete orthonormal basis for the density matrix propagation, especially
692: in the nearly degenerate eigenvalue case.
693: We can understand the situation as follows: When the two nearly
694: degenerate eigenvalues $Z_{1}$ and $Z_{2}$ are
695: obtained from a non-Hermitian operator,
696: the orthogonality implies that $\la L_{2}|$ and $|R_{1}\ra$
697: are orthogonal to each other as well as $\la L_{1}|$ and $|R_{2}\ra$. When two
698: eigenvalues become very close to each other, unlike the Hermitian operator case,
699: $\la L_{1}|$ and $\la L_{2}|$ almost coincide and so do $|R_{1}\ra$ and $|R_{2}\ra$,
700: so that $\la L_{1}|$ and $|R_{1}\ra$ become almost orthogonal to each other. To still
701: satisfy the normalization condition $\la L_{n}|R_{n}\ra=1$ in this case,
702: the eigenfunction should be scaled up, thus making the spectral structure
703: very sensitive to the numerical error involved in the calculation of eigenfunctions.
704: For an interesting discussion on this point,
705: one may refer to the work by Nelson and co-workers.\cite{dahmen-condmat-99} Due to this
706: numerical instability, the use of the spectral method as a density matrix
707: propagation scheme is not without limitation.
708:
709: Figure 3a shows the spectral structure and the time-evolution of the
710: density matrix propagation for the case of
711: $\beta\Omega=1$, $\beta\lambda=15$, $\beta\D=12$, and $\beta\epsilon=5$.
712: Generally, when the energy bias is small ($\beta\epsilon\le 5$),
713: the left and right eigenfunctions can form a complete orthonormal basis set,
714: so the spectral method is stable and can be used
715: as a numerical propagation method for the density matrix.
716: With a large energy bias, however, the calculated eigenfunctions
717: may not form a complete orthonormal basis.
718: To model for the photo-induced back electron transfer
719: experiment in the mixed-valence compounds the initial density matrix is chosen
720: as a thermal equilibrium distribution of the donor state(i.e. 1-state)
721: pumped to the acceptor state(i.e. 2-state),\cite{reid-jpc-95,lucke-jcp-97,evans-jcp-98}
722: \bmath
723: \be
724: \rho_{i}(E,0)&=&{1 \over {{\sqrt{2\pi}} \Delta_{E}}} \exp \left(-{{(E+\lambda)^2}
725: \over {2\Delta_E^2}}\right)\delta_{i2}, \label{28} \\
726: \rho_{12}(E,0)&=&\rho_{21}(E,0)=0. \label{29}
727: \ee
728: \emath
729: It would be straightforward to calculate the spatial distribution of the density
730: matrix in time $\rho(E,t)$ by applying the propagator in Eq.~(\ref{21}) to the
731: above initial density matrix; however, to demonstrate the overall temporal
732: behavior only the time evolution
733: of the total population in the acceptor state is calculated,
734: \be
735: P_{2}(t)=\int dE\rho_{2}(E,t). \label{30}
736: \ee
737: In order to check the validity of the spectral method as a propagation
738: scheme in this case, we also calculated the time evolution of the density
739: matrix by directly solving the $4N$ differential equations
740: for the expansion coefficients of the
741: density matrix using the Bulirsh-Stoer algorithm,\cite{press-numrecipes-92} and the comparison
742: in Fig.~3(a) shows a perfect agreement. If only the transient behavior is
743: concerned with, the direct propagation method would be preferred over the
744: spectral method, however, the spectral propagation has the advantage when
745: calculating the long time behavior once the complete spectrum is known.
746: Overall, the computational costs for two method are comparable to each other.
747: As expected from the spectral structure shown in the previous section
748: the population in the acceptor state shows an underdamped
749: coherent oscillation behavior at initial times
750: followed by a damped oscillation behavior at later times.
751:
752: Further, we have also studied the density matrix propagation for different energy biases to
753: examine the electronic dephasing effect. As seen from Fig.~4(a),
754: the increase in energy bias
755: destroys the electronic coherence dramatically. Another interesting
756: observation is the phase shift in the population dynamics
757: as the energy bias is varied,
758: and it is because the Rabi oscillation frequency increases with energy bias.
759: We can confirm the temporal behavior of the density matrix propagation by
760: examining the spectral structure shown in Fig.~4(b).
761: The period of the initial coherence is estimated to be $\tau_{osc}\approx 0.25\beta$
762: from Fig.~4(a). In comparison, the Rabi frequency for the corresponding adiabatic
763: two-level system is given by
764: $\Omega_{R}=\sqrt{\epsilon^2+4V^2}\approx 25{\beta}^{-1}$,
765: which can also be obtained from the onset of imaginary
766: branches in the eigen-tree shown in Fig.~4(b),
767: and the estimation is consistent with the oscillation period observed in the dynamics
768: since $\tau_{osc}\approx 2\pi/{\Omega_{R}}$.
769: The real eigenvalues of the lowest excited states in the the imaginary branches are
770: estimated to be $\beta k\approx 1-2$, and they agree with the decay time
771: of the oscillation amplitude in Fig.~4(a), confirming the validity of the
772: spectral method as a density matrix propagation scheme.
773: Even though it has been well known in the literature that
774: the damping of population is enhanced with
775: increased energy asymmetry,\cite{leggett-rmp-87} we have also confirmed
776: this through the spectral analysis method.
777:
778: As an example of the eigenfunction responsible for the coherent
779: oscillation behavior observed in Fig.~4 (b), we show
780: the left and right eigenfunctions corresponding to a
781: complex eigenvalue $\beta Z=2.6228\pm i 26.394$ for a symmetric case
782: $(\beta\epsilon=0)$ and $\beta Z=2.8057\pm i 26.466$ for
783: an asymmetric case$(\beta\epsilon=5)$ in Figs.~5 and 6.
784: The eigenfunctions corresponding to a complex conjugate pair
785: of eigenvalues are also complex conjugate to each other;
786: therefore, the frequency spectrum of the density matrix evolution is
787: proportional to the norm of wavefunction.
788: We note that the left eigenfunction is more extended than
789: the right eigenfunction. Although the population distribution
790: in the donor and acceptor states corresponding to coherent oscillation
791: is inverted with respect to the Boltzmann distribution,
792: it does not contribute to the steady-state population
793: distribution due to the transient nature.
794:
795:
796: \section{CONCLUDING REMARKS}
797:
798: In this paper we have applied the spectral analysis method to
799: the non-adiabatic two-state diffusion equation,
800: that describes electron transfer dynamics in Debye solvents.
801: In particular, we have examined electronic coherence
802: in mixed-valence compounds, and
803: demonstrated that underdamped Rabi oscillations
804: are observed in an overdamped solvent environment.
805: Detailed study of the spectral structure of the non-adiabatic operator
806: for various energy biases and coupling constants allows us
807: to determine the underlying mechanisms of electron transfer kinetics.
808: Eigenvalues form three branches in the eigen-diagram: a single branch of
809: real eigenvalues and two symmetric branches of complex conjugate eigenvalues.
810: In strongly coupled systems, all three branches have a similar order of
811: magnitude, indicating that both multiple-exponential decay and coherent
812: oscillations can be observed experimentally.
813:
814: We have investigated the dependence of the spectral structure on
815: the coupling constant. In the very weak coupling regime,
816: the lowest excited state is well separated from higher
817: states, which makes the electron transfer dynamics a well-defined
818: rate process. In the strong coupling regime, however, the eigenvalue
819: diagram shows coalescence/bifurcation behavior in the complex plane.
820: We have used the spectral method to calculate the time-evolution
821: of the density matrix, and indeed, observed
822: electronic coherence in the temporal
823: behavior of population in the acceptor state for
824: non-equilibrium initial distributions.
825: We also found a good agreement between results of
826: the spectral propagation method
827: and of the numerical propagation method for small energy bias cases.
828: Due to non-Hermitianity of the non-adiabatic operator, the
829: spectral propagation method was not numerically stable
830: for large energy bias cases.
831:
832: For an isolated quantum system,
833: the eigen-solution to the Schr\"odinger equation completely
834: determines its dynamics. In a similar fashion,
835: the eigen-solution to the non-adiabatic diffusion operator
836: completely characterizes the dynamics of a dissipative system
837: and thus provides a powerful tool to analyze dissipative dynamics.
838: It is well known that quantum dynamics comes from the underlying
839: spectra, especially in gas-phase chemical systems;\cite{field-acp-97}
840: however, the spectral aspect of condensed phase dissipative systems
841: has not been well recognized yet and deserves further investigation.
842: Though the analysis presented here is restricted to
843: semi-classical dissipative systems, it may also be applied to quantum
844: dissipative dynamics. In principle,
845: we can derive the evolution equation for quantum
846: dissipative systems either from first principles or
847: through numerical reduction,
848: and then pose the quantum dissipative equation of motion
849: as an eigen-value problem.
850: Along this line, the dissipative dynamics of the spin-boson Hamiltonian,
851: which has been studied mostly as a dynamic problem,\cite{makarov-cpl-94,wang-jcp-110}
852: can also be explored as a spectral problem in the future.
853:
854: \section*{ACKNOWLEDGMENTS}
855: The authors would like to thank NSF for financial support.
856: One of us (YJ) would like to thank the Korean Foundations for Advanced Studies
857: for financial support.
858:
859: \begin{thebibliography}{10}
860:
861: \bibitem{vos-nat-93}
862: Vos, M.~H. ; Rappaport, F. ; Lambry, J.-C. ; Breton, J. ; Martin, J.-L.
863: \newblock {\em Nature} {\bf 1993} {\em 363}, 320.
864:
865: \bibitem{jonas-jpc-95}
866: Jonas, D. ; Bradford, S. ; Passino, S. ; Fleming, G.
867: \newblock {\em J. Phys. Chem.} {\bf 1995} {\em 99}, 2554.
868:
869: \bibitem{arnett-jacs-95}
870: Arnett, D.~C. ; Vohringer, P. ; Scherer, N.~F.
871: \newblock {\em J. Am. Chem. Soc.} {\bf 1995} {\em 117}, 12262.
872:
873: \bibitem{reid-jpc-95}
874: Reid, P.~J. ; Silva, C. ; Barbara, P.~F. ; Karki, L. ; Hupp, J.~T.
875: \newblock {\em J. Phys. Chem.} {\bf 1995} {\em 99}, 2609.
876:
877: \bibitem{lucke-jcp-97}
878: Lucke, A. ; Mak, C.~H. ; Egger, R. ; Ankerhold, J. ; Stockburger, J. ; Grabert,
879: H.
880: \newblock {\em J. Chem. Phys.} {\bf 1997} {\em 107}, 8397.
881:
882: \bibitem{makarov-cpl-94}
883: Makarov, D. ; Makri, N.
884: \newblock {\em Chem. Phys. Lett.} {\bf 1994} {\em 221}, 482.
885:
886: \bibitem{evans-jcp-98}
887: Evans, D.~G. ; Nitzan, A. ; Ratner, M.~A.
888: \newblock {\em J. Chem. Phys.} {\bf 1998} {\em 108}, 6387.
889:
890: \bibitem{coalson-jcp-94}
891: Coalson, R.~D. ; Evans, D.~G. ; Nitzan, A.
892: \newblock {\em J. Chem. Phys.} {\bf 1994} {\em 101}, 486.
893:
894: \bibitem{cho-jcp-95}
895: Cho, M. ; Silbey, R.~J.
896: \newblock {\em J. Chem. Phys.} {\bf 1995} {\em 103}, 595.
897:
898: \bibitem{harris-jcp-83}
899: Harris, R.~A. ; Silbey, R.
900: \newblock {\em J. Chem. Phys.} {\bf 1983} {\em 78}, 7330.
901:
902: \bibitem{silbey-jcp-84}
903: Silbey, R. ; Harris, R.~A.
904: \newblock {\em J. Chem. Phys.} {\bf 1984} {\em 80}, 2615.
905:
906: \bibitem{carmeli-jcp-85}
907: Carmeli, B. ; Chandler, D.
908: \newblock {\em J. Chem. Phys.} {\bf 1985} {\em 82}, 3401.
909:
910: \bibitem{chandler-liquid-91}
911: Chandler, D.
912: \newblock in {\em Liquides, Cristallisation et Transition Vitreuse, Les
913: Houches, Session LI}, edited by Levesque, D. ; Hansen, J. ; Zinn-Justin, J. ,
914: Elsevier, New York, 1991.
915:
916: \bibitem{leggett-rmp-87}
917: Leggett, A.~J. ; Chakravarty, S. ; Dorsey, A.~T. ; Fisher, M. P.~A. ; Garg, A.
918: ; Zwerger, W.
919: \newblock {\em Rev. Mod. Phys.} {\bf 1987} {\em 59}, 1.
920:
921: \bibitem{marcus-bba-85}
922: Marcus, R.~A. ; Sutin, N.
923: \newblock {\em Biochim. Biophys. Acta.} {\bf 1985} {\em 811}, 265.
924:
925: \bibitem{newton-arpc-84}
926: Newton, M.~D. ; Sutin, N.
927: \newblock {\em Annu. Rev. Phys. Chem.} {\bf 1984} {\em 35}, 437.
928:
929: \bibitem{bader-jcp-90}
930: Bader, J.~S. ; Kuharski, R.~A. ; Chandler, D.
931: \newblock {\em J. Chem. Phys.} {\bf 1990} {\em 93}, 230.
932:
933: \bibitem{cao-cpl-99}
934: Cao, J.
935: \newblock {\em Chem. Phys. Lett.} {\bf 1999} {\em 312}, 606.
936:
937:
938: \bibitem{cao-jcp-00}
939: Cao, J.; Jung, Y.
940: \newblock {\em J. Chem. Phys.} {\bf 2000}
941: \newblock in press.
942:
943: \bibitem{zusman-cp-80}
944: Zusman, L.~D.
945: \newblock {\em Chem. Phys.} {\bf 1980} {\em 49}, 295.
946:
947: \bibitem{calef-jpc-83}
948: Calef, D.~F. ; Wolynes, P.~G.
949: \newblock {\em J. Phys. Chem.} {\bf 1983} {\em 87}, 3387.
950:
951: \bibitem{hynes-jpc-86}
952: Hynes, J.~T.
953: \newblock {\em J. Phys. Chem.} {\bf 1986} {\em 90}, 3701.
954:
955: \bibitem{garg-jcp-85}
956: Garg, A. ; Onuchic, J.~N. ; Ambegaokar, V.
957: \newblock {\em J. Chem. Phys.} {\bf 1985} {\em 83}, 4491.
958:
959: \bibitem{sparpaglione-jcp-88}
960: Sparpaglione, M. ; Mukamel, S.
961: \newblock {\em J. Chem. Phys.} {\bf 1988} {\em 88}, 3263.
962:
963: \bibitem{cao-jcp-97}
964: Cao, J. ; Voth, G.~A.
965: \newblock {\em J. Chem. Phys.} {\bf 1997} {\em 106}, 1769.
966:
967: \bibitem{onuchic-jcp-93}
968: Onuchic, J.~N. ; Wolynes, P.~G.
969: \newblock {\em J. Chem. Phys.} {\bf 1993} {\em 98}, 2218.
970:
971: \bibitem{yang-jcp-89}
972: Yang, D.~Y. ; Cukier, R.~I.
973: \newblock {\em J. Chem. Phys.} {\bf 1989} {\em 91}, 281.
974:
975: \bibitem{simons-cp-73}
976: Simons, J.
977: \newblock {\em Chem. Phys.} {\bf 1973} {\em 2}, 27.
978:
979: \bibitem{risken-fpe-84}
980: Risken, H.
981: \newblock {\em The Fokker-Planck Equation} (Springer-Verlag
982: \newblock New York, 1984).
983:
984: \bibitem{dahmen-condmat-99}
985: Dahmen, K.~A. ; Nelson, D.~R. ; Shnerb, N.~M.
986: \newblock {\em cond-mat/9903276} {\bf 1999} .
987:
988: \bibitem{press-numrecipes-92}
989: Press, W.~H. ; Teukolsky, S.~A. ; Vetterling, W.~T. ; Flannery, B.~P.
990: \newblock {\em Numerical Recipes in FORTRAN}
991: \newblock 2nd edition (Cambridge University Press, Cambridge, 1992).
992:
993: \bibitem{field-acp-97}
994: Field, R.~W. ; O'Brien, J.~P. ; Jacobson, M.~P. ; Solina, S. A.~B. ; Pollik,
995: W.~F. ; Ishikawa, H.
996: \newblock {\em Adv. Chem. Phys.} {\bf 1997} {\em 101}, 463.
997:
998: \bibitem{wang-jcp-110}
999: Wang, H. ; Song, X. ; Chandler, D. ; Miller, W.~H.
1000: \newblock {\em J. Chem. Phys.} {\bf 1999} {\em 110}, 4828.
1001:
1002: \end{thebibliography}
1003:
1004: \newpage
1005: \begin{figure}
1006: \caption{
1007: A plot of the lowest 400 eigenvalues for the non-adiabatic operator in a mixed-valence system.
1008: The parameters are : $\beta\Omega=0.6716$, $\beta\lambda=18.225$, $\beta\D=11.99$, and
1009: $\beta\epsilon=18.705$. The dot-dashed line is for the case $k=\omega/2\pi$.
1010: }
1011: \end{figure}
1012:
1013: \begin{figure}
1014: \caption{
1015: Plots of (a) real and (b) imaginary parts of the lowest 30 eigenvalues as a function of
1016: the coupling constant, $\D$. Except for the coupling constant, all the other parameters are
1017: set equal to those used in Fig.~1. In Fig. ~2(b), open circles correspond
1018: to the Rabi frequency $\Omega_{R}=\sqrt{\epsilon^2+4V^2}$. Figure 3(c) shows the three
1019: dimensional plot of eigenvalues as a function of the coupling constant.
1020: }
1021: \end{figure}
1022:
1023: \begin{figure}
1024: \caption{
1025: Comparison between the result of direct numerical propagation and
1026: spectral propagation. The parameters are chosen as
1027: $\beta\Omega=1$, $\beta\lambda=15$, $\beta\D=12$, and $\beta\epsilon=3$.
1028: }
1029: \end{figure}
1030:
1031: \begin{figure}
1032: \caption{
1033: Comparison of (a) the dynamics and (b) the spectra in the mixed-valence
1034: system for three different energy biases. Except for the energy bias,
1035: all the other parameters are set equal to those used in Fig.~3. Agreements
1036: between the results of numerical and spectral propagation
1037: have been checked in these cases.
1038: }
1039: \end{figure}
1040:
1041: \begin{figure}
1042: \caption{
1043: (a) Right and (b) left eigenfunctions
1044: with an eigenvalue $\beta Z=2.6228\pm i 26.394$
1045: for a symmetric bias case.$(\beta\epsilon=0)$
1046: All the other parameters are set equal to those used
1047: in Fig.~3 except for the energy bias.
1048: Each line corresponds to $\rho_1$(solid), $\rho_2$(dashed),
1049: $u$(dot-dashed), and $v$(dotted), respectively.
1050: }
1051: \end{figure}
1052:
1053: \begin{figure}
1054: \caption{
1055: (a) Right and (b) left eigenfunctions
1056: with an eigenvalue $\beta Z=2.8057\pm i 26.466$
1057: for an asymmetric bias case.$(\beta\epsilon=5)$
1058: All the other parameters are set equal to those used
1059: in Fig.~3 except for the energy bias.
1060: Each line corresponds to $\rho_1$(solid), $\rho_2$(dashed),
1061: $u$(dot-dashed), and $v$(dotted), respectively.
1062: }
1063: \end{figure}
1064:
1065:
1066: \end{document}
1067:
1068:
1069:
1070:
1071:
1072:
1073:
1074:
1075:
1076: