1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: % Submitted to Physical Review B
3: %
4: % First Author Information:
5: % Keun Hyuk Ahn, Ph.D. (ahn@lanl.gov)
6: % T-11, Theoretical Division, MS B262
7: % Los Alamos National Laboratory
8: % Los Alamos, New Mexico 87545, USA
9: % Tel:505-667-8648 Fax:505-665-4063
10: %
11: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
12: %\documentclass[aps,prb,preprint,groupedaddress,showpacs]{revtex4}
13: \documentclass[aps,prb,twocolumn,groupedaddress,showpacs]{revtex4}
14: %\documentclass[aps,prl,twocolumn,groupedaddress,preprintnumbers,showpacs]{revtex4}
15: %\documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
16: %\documentclass[amssymb,amsmath,aps,prb,showpacs,twocolumn,nobalancelastpage]{revtex4}
17:
18: %\usepackage{graphics}% Include figure files
19: \usepackage{epsfig}
20: %\usepackage{times}
21: %\usepackage{bm}
22:
23: \begin{document}
24:
25: %\preprint{LAUR-nnn}
26:
27: \title{Ultrafast quasiparticle relaxation dynamics
28: in normal metals and heavy fermion materials}
29:
30: \author{
31: K. H. Ahn, M. J. Graf, S. A. Trugman,
32: J. Demsar,~\cite{Demsar}
33: R. D. Averitt, J. L. Sarrao, and A. J. Taylor
34: }
35: \affiliation{
36: Los Alamos National Laboratory,
37: Los Alamos, New Mexico 87545}
38:
39: %\date{\today}
40:
41: \begin{abstract}
42: We present a detailed theoretical study of the ultrafast quasiparticle
43: relaxation dynamics observed
44: in normal metals and heavy fermion materials
45: with femtosecond time-resolved optical pump-probe spectroscopy.
46: For normal metals, a nonthermal electron distribution
47: gives rise to a temperature ($T$) independent electron-phonon relaxation time
48: at low temperatures,
49: in contrast to the $T^{-3}$-divergent behavior predicted by the two-temperature model.
50: For heavy fermion compounds, we find that the blocking of electron-phonon
51: scattering for heavy electrons within the density-of-states peak near the Fermi energy
52: is crucial to explain the rapid
53: increase of the electron-phonon relaxation time below the Kondo temperature.
54: We propose the hypothesis that the slower Fermi velocity compared to the sound velocity
55: provides a natural blocking mechanism due to energy and momentum conservation laws.
56: \end{abstract}
57:
58: \pacs{78.47.+p 71.27.+a, 71.10.-w}
59:
60: \maketitle
61:
62: \newpage
63:
64: \section{Introduction}
65:
66: Recently, ultrafast time-resolved optical spectroscopy has been
67: used to reveal the nature of the quasiparticle relaxation dynamics
68: in condensed matter systems.
69: Femtosecond time-resolved pump-probe optical measurements have been
70: carried out in normal metals,~\cite{Groeneveld92,Groeneveld95}
71: conventional~\cite{Brorson90} and
72: high-T$_c$ superconductors,~\cite{Stevens97}
73: and charge-density wave solids.~\cite{Demsar99}
74: Time-resolved terahertz spectroscopy has measured
75: picosecond time-scale transient conductivity
76: in colossal magnetoresistance manganites.~\cite{Averitt01}
77: These experiments show directly in the time domain how
78: constituent degrees of freedom of materials
79: interact with each other,
80: which is important to understand the physics
81: governing the ground state and the low-energy excited states
82: of materials.
83: Understanding the fast dynamics of quasiparticles is also
84: crucial for technological applications of these materials,
85: for example, in the design of very fast switching devices.
86:
87: We recently reported femtosecond time-resolved pump-probe optical
88: measurements on LuAgCu$_4$ and YbAgCu$_4$.~\cite{Demsar03}
89: These two materials are isostructural with
90: a negligible difference
91: in lattice constants~\cite{Pagliuso97} (about 0.3 \%) and
92: about 1 \% difference in atomic mass between Lu and Yb.
93: However, their electronic structures are strikingly different
94: due to the $f$-level occupancies in Lu and Yb.
95: The closed-shell $4f$ levels in Lu have no significant interaction with
96: Cu $3d$ and Ag $4d$ conduction electrons,
97: and LuAgCu$_4$ is a normal metal.~\cite{LuAgCu4}
98: In contrast, the open-shell $4f$ levels in Yb,
99: that is, the localized hole in the $4f^{13}$ configuration,
100: have a strong interaction with conduction electrons,
101: and make YbAgCu$_4$ a heavy-fermion material,
102: which is characterized by a large peak
103: in the electron density of states (DOS)
104: at the Fermi energy ($E_F$) and, equivalently,
105: a large Sommerfeld coefficient $\gamma$.~\cite{Sarrao99,YbAgCu4}
106: Our time resolved optical experiments show
107: very different relaxation dynamics in these two materials.
108: LuAgCu$_4$ shows a relaxation time ($\tau $) versus temperature ($T$) behavior
109: similar to other normal metals, such as Ag and Au.~\cite{Groeneveld92,Groeneveld95}
110: In particular, there is very little $T$ dependence in $\tau $ at low temperatures.
111: However, YbAgCu$_4$ shows approximately a 100-fold increase in $\tau $,
112: as $T$ is decreased from the Kondo temperature ($T_K$) down to 10 K.
113: Kondo temperature is about 100 K in YbAgCu$_4$ and typically characterizes
114: the width of the large DOS near $E_F$.
115:
116: In Ref.~\onlinecite{Demsar03},
117: we presented experimental data showing different relaxation dynamics in
118: LuAgCu$_4$ and YbAgCu$_4$
119: along with the main theoretical ideas and final results of the calculations based on
120: coupled electron and phonon Boltzmann transport equations.
121: In this paper, we report on the details of the theoretical model and analysis,
122: and clarify the underlying physics.
123: The theories that existed before our study are summarized in Sec.~II~A.
124: Sections~II~B and III discuss
125: our calculations of the relaxation dynamics in normal metals and heavy fermion materials,
126: respectively.
127: Section~IV summarizes our results.
128:
129: \section{Normal metals}
130:
131: \subsection{Two-temperature model, its inconsistency with experiments,
132: and the nonthermal electron model}
133:
134: In ultrafast optical pump-probe spectroscopy,
135: an ultrafast laser pulse initially excites
136: the electron system, and
137: the probe pulse monitors the relaxation of the electron system
138: by measuring transient optical properties
139: with subpicosecond time-resolution.
140: Because the diffusion of heat out of the illuminated (probed) region
141: is much slower (tens to hundreds of nanoseconds)
142: than the time scales of interest,
143: the relaxation of the excited electrons is due to
144: the thermalization among electrons and other degrees of freedom, such as phonons,
145: within the illuminated areas.
146:
147: The most commonly used model for the relaxation dynamics
148: of photoexcited electrons in metals
149: is called the two-temperature model (TTM),~\cite{Kaganov56}
150: which assumes much faster relaxation
151: {\it within} each constituent system
152: (e.g., electron system, phonon system)
153: compared to the relaxation {\it between} these constituent systems.
154: In this approximation, the temperature of each system can be defined during relaxation,
155: and the relaxation time $\tau$ between system 1 and 2 is determined by
156: their specific heats, $C_1$ and $C_2$, and
157: the energy transfer rate per temperature difference, $g(T)$:~\cite{TTM.eq}
158: \begin{equation}
159: \frac{1}{\tau}=g(T) \left( \frac{1}{C_1} + \frac{1}{C_2} \right). \label{eq:TTM}
160: \end{equation}
161: Kaganov, Lifshitz, and Tanatarov~\cite{Kaganov56}
162: calculated the energy transfer rate $g(T)$
163: between electrons (e) and phonons (p) in normal metals
164: by solving coupled Boltzmann transport equations
165: for electrons and phonons with thermal equilibrium distributions at different temperatures.
166: Their results predict
167: $\tau \sim T$ at $T > T_{Debye}/5$ and
168: $\tau \sim T^{-3}$ at $T < T_{Debye}/5 $,
169: where $T_{Debye}$ is the Debye temperature.
170:
171: Groeneveld {\it et al.}~\cite{Groeneveld92,Groeneveld95} have
172: measured the relaxation time $\tau$ in Au and Ag
173: as a function of temperature ($T=10-300$ K)
174: and laser-energy density ($U_{laser}=0.3-1.3$ J cm$^{-3}$)
175: using the femtosecond pump-probe technique.
176: Their experimental data was inconsistent
177: with the TTM predictions for
178: $\tau$ versus $T$ and $\tau$ versus $U_{laser}$,
179: which led Groeneveld {\it et al.} to consider the nonthermal electron model (NEM).
180: Using a simple estimate, they pointed out that the relaxation time
181: within the electron system is comparable to the electron-phonon
182: relaxation time.
183: Instead of assuming a thermal equilibrium distribution for the electrons,
184: they numerically solved the Boltzmann equation
185: for {\it electrons}
186: with electron-electron and electron-phonon scattering,
187: starting from an initial nonthermal electron distribution
188: created by a laser pulse.
189: (Based on the fact that the phonon specific heat is much larger
190: than the electron specific heat,
191: phonons were assumed to have a thermal distribution
192: with a time independent $T$.)
193: This simulation could explain experimental data down to about 50 K.
194: Although Groeneveld {\it et al.}'s work proposed the essential idea
195: of a {\it nonthermal electron system},
196: their analysis focused on temperatures above 50~K,
197: and excluded the low-$T$ region where
198: the most striking difference between the TTM ($\tau \sim T^{-3}$)
199: and the experimental data (almost $T$-independent $\tau$) occurs.
200: Therefore, we undertake a more detailed analysis,
201: particularly focused on the low-$T$ region,
202: to understand the difference in relaxation dynamics
203: between thermal and nonthermal electrons in normal metals
204: as well as in heavy fermion metals.
205:
206: \subsection{Relaxation dynamics between nonthermal electrons and phonons}
207:
208: We consider the coupled Boltzmann equations
209: for both {\it electrons and phonons}, with electron-electron and electron-phonon
210: scattering included.~\cite{pp}
211: The Boltzmann equations with momentum indices
212: are transformed into the following equations with energy indices
213: for a model with isotropic Debye phonons
214: and electrons with an isotropic parabolic dispersion relation.
215: In terms of electron and phonon distributions at time $t$,
216: $f_{\epsilon}(t)$ and $b_{\omega}(t)$,
217: and the electron and phonon DOS, $D_e(\epsilon)$ and $D_p(\omega)$,
218: where $\epsilon$ and $\omega$ represent the electron and phonon energies,
219: the coupled Boltzmann equations are~\cite{AM,Landau,Ziman}
220: \begin{eqnarray}
221: \frac{df_{\epsilon}}{dt} &=& \left[ \frac{df_{\epsilon}}{dt} \right]_{ep}
222: + \left[ \frac{df_{\epsilon}}{dt} \right]_{ee}, \label{Boltzmann.e} \\
223: \frac{db_{\omega}}{dt} &=& \left[ \frac{db_{\omega}}{dt} \right]_{ep},
224: \label{Boltzmann.p}
225: \end{eqnarray}
226: with the collision integrals
227: \begin{eqnarray}
228: \left[ \frac{df_{\epsilon}}{dt} \right]_{ep}
229: &=& \int d\omega K_{ep}
230: \{ [f_{\epsilon + \omega} (1-f_{\epsilon}) (b_{\omega}+1)
231: \nonumber \\
232: &-& f_{\epsilon} (1-f_{\epsilon+\omega}) b_{\omega} ]
233: D_{p}(\omega) D_{e}(\epsilon+\omega)
234: \nonumber \\
235: &+&[f_{\epsilon - \omega} (1-f_{\epsilon}) b_{\omega}
236: - f_{\epsilon} (1-f_{\epsilon-\omega}) (b_{\omega}+1) ]
237: \nonumber \\
238: & & D_{p}(\omega) D_{e}(\epsilon-\omega) \}, \label{eq:Boltzmann.1}
239: \end{eqnarray}
240: \begin{eqnarray}
241: \left[\frac{db_{\omega}}{dt}\right]_{ep}
242: &=&\int d\epsilon K_{ep}
243: [ -b_{\omega} f_{\epsilon} (1-f_{\epsilon+\omega}) +
244: \nonumber \\
245: & &(b_{\omega}+1)f_{\epsilon+\omega} (1-f_{\epsilon}) ]
246: D_{e}(\epsilon) D_{e}(\epsilon+\omega), \label{eq:Boltzmann.2}
247: \end{eqnarray}
248: \begin{eqnarray}
249: \left[ \frac{df_{\epsilon}}{dt} \right]_{ee}
250: &=&\int d\epsilon ' d\epsilon ''
251: \frac{1}{2} K_{ee}
252: [
253: -f_{\epsilon} f_{\epsilon '} (1-f_{\epsilon ''}) (1-f_{\epsilon+\epsilon '-\epsilon ''})
254: \nonumber \\
255: &+&(1-f_{\epsilon})(1-f_{\epsilon '}) f_{\epsilon ''} f_{\epsilon+\epsilon '-\epsilon ''}
256: ] \nonumber \\
257: & &D_{e}(\epsilon ') D_{e}(\epsilon '') D_{e}(\epsilon +\epsilon ' - \epsilon '' ).
258: \label{eq:Boltzmann.3}
259: \end{eqnarray}
260: The linear dependence
261: of the electron-phonon scattering rate on the acoustic phonon energy $\omega$
262: [see, e.g., Eq.~(26.42) in Ref.~\onlinecite{AM}]
263: has been canceled out by a $1/\omega$ factor
264: which originates from changing momentum indices to
265: energy indices,
266: so that $K_{ep}$ is an energy independent constant.
267: $K_{ep}$ and $K_{ee}$ include the squares of the scattering matrix elements
268: and all other numerical factors, and
269: have the units of [energy/time].
270:
271: In the remainder of this section,
272: we consider only light electron systems (normal metals).
273: The above Boltzmann equations, with electron and phonon DOS explicitly included,
274: show why the electron-electron relaxation time is comparable
275: to the electron-phonon relaxation time at low temperatures.
276: The relaxation rate depends on the square of the scattering matrix elements
277: (i.e., $K_{ep}$ and $K_{ee}$)
278: and the number of available final states (i.e., $D_{e}$ and $D_{p}$).
279: If we compare Eqs.~(\ref{eq:Boltzmann.1}) and (\ref{eq:Boltzmann.3}),
280: even though $K_{ee}$ is typically much larger than $K_{ep}$,
281: the large disparity between the magnitude of the phonon ($D_p$) and electron ($D_e$) DOS
282: expedites the relaxation of electrons with phonons,
283: and hinders electrons from reaching thermal equilibrium.
284: The electron-electron and electron-phonon relaxation times,
285: $\tau_{ee}$ and $\tau_{ep}$,
286: vary with $T$.
287: In Ref.~\onlinecite{Groeneveld95}, Groeneveld {\it et al.}
288: found $\tau_{ee} \sim T^{-2}$ [Eq.~(16) of Ref.~\onlinecite{Groeneveld95}]
289: from Fermi liquid theory.
290: Figure~\ref{fig:taueeep} schematically
291: compares the temperature dependence of $\tau_{ee}$ (thick line) and
292: $\tau_{ep}$ (thin line), the latter of which is from the TTM results.~\cite{Kaganov56,Groeneveld95}
293: The plot shows that it is likely that
294: at high temperatures
295: the electron system thermalizes within itself faster
296: than the thermalization time with the lattice,
297: whereas at low temperatures
298: electron-electron thermalization
299: becomes slower than the electron-phonon relaxation.
300: More detailed discussion on $T$-dependent
301: thermal versus nonthermal electron distributions
302: will be given later in this section.
303:
304: \begin{figure}
305: \leavevmode
306: \epsfxsize5.5cm\epsfbox{taueeep.eps}
307: \caption
308: {\label{fig:taueeep}
309: Schematic plot for
310: the temperature dependence of
311: electron-electron and electron-phonon relaxation times,
312: $\tau_{ee}$ (thick line) and $\tau_{ep}$ (thin line).
313: $\tau_{ee}$ is from Eq.~(16) in Ref.~\onlinecite{Groeneveld95}.
314: $\tau_{ep}$ is from the TTM calculation,
315: which assumes a {\it thermal} electron distribution.
316: The ratio between $\tau_{ee}$ and $\tau_{ep}$
317: at high $T$ (e.g., $T_{Debye}$ or 300 K)
318: depends on the $K_{ee}/K_{ep}$ ratio.
319: For $K_{ee}/K_{ep}$=700 used in our calculation,
320: $\tau_{ee} < \tau_{ep}$ at $T=T_{Debye}$, as shown in the plot.
321: }
322: \end{figure}
323:
324: We now discuss the methods and parameter values
325: we used to analyze the above Boltzmann equations.
326: The coupled Boltzmann equations are solved in two different ways.
327: In the first method, an initial condition at $t=0$ is set as
328: a nonthermal electron distribution
329: excited by the laser pulse, $f_{\epsilon} (t=0)=f_{FD}(\epsilon,T_i) + \Delta f_{\epsilon}$,
330: and a thermal phonon distribution,
331: $b_{\omega}(t=0)=b_{BE}(\omega,T_i)$,
332: where $T_i$, $f_{FD}(\epsilon,T_i)$, and $b_{BE}(\omega,T_i)$
333: represent the initial temperature, Fermi-Dirac, and Bose-Einstein
334: distribution functions before the photoexcitation.
335: We find $f_{\epsilon}(t)$ and $b_{\omega}(t)$ at subsequent times by solving the differential
336: equations
337: until both $f_{\epsilon}(t)$ and $b_{\omega}(t)$ approach
338: the final thermal distributions at equal $T$'s.
339: A small $\Delta f_{\epsilon}$ has been chosen so that
340: the final $T$ ($T_f$) is less than 3 K higher than $T_i$
341: (even at the lowest $T$),
342: which simulates the low laser intensity used
343: in the experiments.~\cite{Demsar03}
344: In the simulation
345: the total electron energy $E_{e}(t)$
346: is calculated
347: at each time step,
348: and the instantaneous relaxation time $\tau (t)$
349: is evaluated using
350: $E_{e}(t)$ at three consecutive time steps.~\cite{3points}
351:
352: In the second method, we perform a linear stability analysis around the final states,
353: $f_{FD}(\epsilon,T_f)$ and $b_{BE}(\omega,T_f)$,
354: by expanding
355: the coupled Boltzmann equations, Eqs.~(\ref{Boltzmann.e}) and (\ref{Boltzmann.p}),
356: linearly in $\delta f_{\epsilon}(t)$ and $\delta b_{\omega}(t)$, where
357: $f_{\epsilon}(t)=f_{FD}(\epsilon,T_f)+\delta f_{\epsilon}(t)$
358: and $b_{\omega}(t)=b_{BE}(\omega,T_f)+\delta b_{\omega}(t)$.
359: The zeroth order terms of the expansion vanish on both sides of the equations,
360: because they correspond to the final equilibrium state.
361: With
362: $\delta f_{\epsilon}(t)=v_{\epsilon}^e e^{-t/\tau}$,
363: $\delta b_{\omega}(t)=v_{\omega}^p e^{-t/\tau}$,
364: and discretized $N$ electron and $M$ phonon energy levels,
365: the linear differential equations can be cast into
366: an eigenvalue problem of an $(N+M) \times (N+M)$ matrix,
367: the solution of which gives $-1/\tau$ (eigenvalues) and
368: the normal modes of the relaxation.
369: Two of the normal modes have an unphysical infinite relaxation time $\tau$,
370: which originates from
371: total energy and total electron number conservation.
372: The rest of the modes represent all possible relaxation modes of the system.
373: In general, the relaxation of the system
374: can be represented as a linear combination of these modes.
375: By examine the eigenvector of each mode,
376: we find whether the mode predominantly contributes to
377: $e$-$e$, $p$-$p$, or $e$-$p$ relaxation.
378: Since the energy transfer rate
379: from the electron system to the phonon system
380: at time $t$
381: of a specific mode with relaxation time $\tau$ is given by
382: $dE_{e,\tau}(t)/dt=[\int (\epsilon -E_F) (-1/\tau) v^e_{\epsilon} D_e(\epsilon) d \epsilon ] e^{-t/\tau}$,
383: a useful quantity to identify the $e$-$p$ relaxation modes
384: is the following $e$-$p$ energy transfer strength $r_E(i)$:
385: \begin{equation}
386: r_E(i)=\frac{1}{\tau(i)}
387: \sum^{N}_{n=1} (\epsilon_n - E_F) v^e_n(i) D_e(\epsilon_n) \Delta \epsilon, \label{rE}
388: \end{equation}
389: where $\Delta \epsilon$ is the energy step size, and
390: $(v^e_1, v^e_2, ..., v^e_N, v^p_1, ..., v^p_M)_i $
391: is the normalized eigenvector for mode $i$ with proper overall sign.
392: $r_E(i)$ characterizes the effectiveness of mode $i$
393: for electron-phonon energy relaxation.
394: The modes which have large $r_E$ dominate in $e$-$p$ relaxation,
395: and their eigenvectors describe how
396: the relaxation between the electron and phonon systems occurs.
397: The modes with small $r_E(i)$ are either $p$-$p$ or $e$-$e$ relaxation modes,
398: and describe how the relaxation within each system happens.
399:
400: The electron and phonon energy levels are discretized
401: with a step size of 2 meV, which we find
402: small enough that any smaller step size would not
403: change our results even at 10 K (the lowest $T$ of the calculations and experiments).
404: The energy window for the electrons has been chosen to be
405: between -0.15 eV and 0.15 eV with $E_F=0$,
406: which is wide enough in comparison to the highest $T$ of the calculations ($\sim$ 300~K).
407: Since the typical band width of normal metals
408: is of the order of eV,
409: the electron DOS for normal metals is assumed
410: to be constant within this energy window.
411: Fitting the experimental electronic specific heat ($C_e$) data,
412: we obtain $D_e$ = 2.1 eV$^{-1}$ f.u.$^{-1}$ spin$^{-1}$
413: for LuAgCu$_4$,~\cite{LuAgCu4}
414: which has been used for all of the results presented in this section.
415: Phonons are modeled by the Debye phonon model.
416: Since only longitudinal phonon modes couple with electrons
417: in the isotropic electron and phonon model,~\cite{Ziman}
418: we use the longitudinal phonon DOS with
419: a Debye energy $\omega_D$ = 24 meV (Ref.~\onlinecite{Sarrao99})
420: and 6 atoms per unit cell, $D_p (\omega)=18 \omega^2 / \omega_D^3$,
421: in the Boltzmann equations.
422: For most of the calculations, we use $K_{ee}/K_{ep}=700$
423: (see Figs.~\ref{fig:tau.Lu}, \ref{fig:tau.Lu.s.rep0},
424: and related discussion for this choice of $K_{ee}/K_{ep}$ ratio), and
425: $K_{ep}$ has been chosen as $K_{ep}=0.93$ eV/ps
426: for normal metals by fitting experimental data with the results of the calculation
427: (see Fig.~\ref{fig:tau.Lu}).
428:
429: We now discuss the results obtained from the first method.
430: Figure~\ref{fig:two.tau} shows a typical result
431: of normalized excess electron energy,
432: $\Delta E_{e} / \Delta E_{e,max}$ versus time ($t$),
433: and $\tau$ (instantaneous relaxation time) versus $t$
434: at high (solid lines) and low (dotted lines) temperatures,
435: where
436: $\Delta E_{e} (t)=E_{e}(t)-E_{e}(t= \infty )$ and
437: $\Delta E_{e,max} = \Delta E_{e} (t=0)$.
438: Two different stages of relaxation can be identified
439: for high $T$ results.
440: During the intermediate time between 0.3 ps and 2 ps
441: (as mentioned earlier, the timescale in picosecond
442: is obtained by using $K_{ee}/K_{ep}=700$ and $K_{ep}$ = 0.93 eV/ps),
443: the energy transfer from electrons to phonons is fast.
444: After more than 99 percent of the excess electron energy is transfered
445: to phonons,
446: a slow relaxation (about 100 times slower than the intermediate
447: stage) appears.
448: The changes of distribution functions at different time stages
449: show that the fast relaxation during the intermediate time
450: corresponds to electron-phonon relaxation,
451: whereas the later slow relaxation corresponds to
452: phonon-phonon thermalization processes.~\cite{late.tau}
453: The relaxation behavior is independent of the initial conditions,
454: i.e., the form of $\Delta f_{\epsilon}$,
455: except at {\it very} early stages of the relaxation,
456: namely, $t < 0.3$ ps in Fig.~\ref{fig:two.tau}.
457: Since the changes in electron distribution during
458: the late stage are expected to be unobservable due to experimental noise,
459: we identify the intermediate stage $\tau$
460: [$t= 0.3 \sim 2$ ps for the solid line in Fig.~\ref{fig:two.tau}(b)]
461: as the one measured experimentally at high $T$.
462: As $T$ is lowered in the simulation,
463: the well-defined flat intermediate time region in the $\tau$ versus $t$ plot
464: is replaced by a gradually changing $\tau$
465: with a minimum as shown in Fig.~\ref{fig:two.tau}(b).
466: This indicates multiple relaxation times,
467: a sign of a nonthermal electron distribution.
468: As done in the experiments, we find the best fitting single relaxation time
469: through a linear fit of $\log | dE_{e}/dt |$ versus $t$,
470: which is close to the minimum $\tau$
471: of the dotted line in
472: Fig.~\ref{fig:two.tau}(b).
473:
474: \begin{figure}
475: \leavevmode
476: \epsfxsize5.5cm\epsfbox{two.tau.eps}
477: \caption
478: {\label{fig:two.tau}
479: A typical result of (a) normalized excess electron energy
480: versus time and
481: (b) the instantaneous relaxation time versus time
482: at high $T$ (solid lines, $T$ = 18 meV = 208 K) and low $T$ (dotted lines, $T$ = 4 meV = 46 K).
483: }
484: \end{figure}
485:
486: The $\tau$'s obtained at various $T$'s are
487: plotted in Fig.~\ref{fig:tau.Lu} (solid circles)
488: along with the experimental data (open circles) for LuAgCu$_4$
489: and the TTM prediction (dashed line, see Ref.~\onlinecite{Demsar03}).
490: The time unit has been scaled to fit the experimental data at high $T$,
491: which gives $K_{ep}$ = 0.93 eV/ps
492: as mentioned previously.
493: The fit
494: does not uniquely determine $K_{ee}$,
495: as long as $K_{ee}$ is smaller than or similar to $700 \times K_{ep}$.
496: The results show reasonable agreement with experimental data,
497: including the region below 50 K, where
498: the TTM predicts strikingly different
499: $\tau \sim T^{-3}$ behavior (Refs.~\onlinecite{Groeneveld95} and \onlinecite{Kaganov56}).
500: By comparing $df/dt$ and $-df_{FD}/dT$,
501: we examine whether the electron system approaches
502: the final equilibrium state
503: while maintaining a thermal distribution or not, as shown in
504: Fig.~\ref{fig:dfdt.Lu} at high and low $T$'s.
505: At high $T$ [Fig.~\ref{fig:dfdt.Lu}(a)],
506: the two curves coincide with each other [indistinguishable in Fig.~\ref{fig:dfdt.Lu}(a)],
507: implying that the electron system has a thermal distribution.
508: However, at low $T$ [Fig.~\ref{fig:dfdt.Lu}(b)],
509: $df/dt$ (solid line with solid circles) has a width of the order
510: of the Debye temperature ($T_{Debye}$)
511: instead of $T$, which is the width of $-df_{FD}/dT$ (dotted line).~\cite{ee}
512: The results clearly show that at low temperatures the electron system
513: does not have a thermal distribution, which agrees with the previous discussion
514: in relation to Fig.~\ref{fig:taueeep}.
515:
516: \begin{figure}
517: \leavevmode
518: \epsfxsize8.0cm\epsfbox{tau.Lu.eps}
519: \caption
520: {\label{fig:tau.Lu}
521: Relaxation time $\tau$
522: calculated from the coupled Boltzmann equations
523: for LuAgCu$_4$,
524: along with experimental data
525: and the TTM prediction.~\cite{Demsar03}
526: }
527: \end{figure}
528:
529: \begin{figure}
530: \leavevmode
531: \epsfxsize8.5cm\epsfbox{dfdt.Lu.eps}
532: \caption
533: {\label{fig:dfdt.Lu}
534: Normalized $df/dt$ compared with
535: normalized $-df_{FD}/dT$ at
536: (a) high $T$ and
537: (b) low $T$.
538: For high $T$, the two lines coincide.
539: }
540: \end{figure}
541:
542: The effect of a nonthermal electron distribution
543: on the relaxation dynamics becomes even more evident,
544: if we increase $K_{ee}$ to $7000 \times K_{ep}$
545: so that the electron-electron relaxation is faster than
546: the electron-phonon relaxation even in the low-$T$ region.
547: The result (solid circles) is shown in Fig.~\ref{fig:tau.Lu.s.rep0}
548: along with TTM prediction (line).
549: In this case, the electron system has a thermal distribution in the whole $T$ range,
550: and the TTM prediction $\tau \sim T^{-3}$ is recovered at low $T$.
551: It is remarkable that our simulation with a large $K_{ee}$ recovers
552: the TTM prediction
553: in spite of the fact that our Boltzmann equations do not include
554: direct phonon-phonon scattering and, therefore, the phonon distribution is
555: nonthermal except at $t=\infty $.
556: We speculate that the very weak dependence
557: of e-p thermalization dynamics
558: on the phonon distribution
559: (in contrast to the very strong dependence
560: on electron distribution)
561: originates from the
562: characteristics of Bose (versus Fermi)
563: statistics.~\cite{TTM.phonon}
564:
565: \begin{figure}
566: \leavevmode
567: \epsfxsize8.0cm\epsfbox{tau.Lu.s.rep0.eps}
568: \caption
569: {\label{fig:tau.Lu.s.rep0}
570: Solid circles: calculated relaxation time $\tau$ from our model for LuAgCu$_4$
571: for a very large $K_{ee} \geq 7000 \times K_{ep}$.
572: Line: TTM prediction from Fig.~\ref{fig:tau.Lu}.
573: With fast electron-electron relaxation
574: our simulations based on coupled Boltzmann equations recover the TTM prediction
575: $\tau \sim T^{-3}$ at low $T$.
576: }
577: \end{figure}
578:
579: We provide a qualitative reason why the thermal electron
580: distribution gives rise to a slow $\tau \sim T^{-3}$ relaxation behavior
581: and the nonthermal electron distribution gives
582: faster and less-$T$-dependent
583: relaxation behavior at low $T$.
584: As depicted in Fig.~\ref{fig:el.dist}(a),
585: if the electrons have a thermal
586: distribution at $T$ (solid line), which is slightly higher than the final $T$ (dotted line),
587: the electron-phonon scattering important for
588: the relaxation happens
589: within the energy range of the order of $T$ from $E_F$.
590: Therefore, the relaxation rate depends on how many phonon modes
591: exist at $\omega < T$.
592: In the Debye phonon model, since $D_p(\omega) \sim \omega^2$,
593: the relaxation rate $\tau^{-1}$ is proportional to $T^3$.
594: In contrast,
595: if the electron distribution is nonthermal [solid line in Fig.~\ref{fig:el.dist}(b)],
596: then electron-phonon relaxation occurs in
597: over an energy range of the order of Debye energy [see also Fig.~\ref{fig:dfdt.Lu}(b)].
598: This makes the electron-phonon relaxation faster and less $T$-dependent,
599: as indeed observed (Fig.~\ref{fig:tau.Lu}).
600:
601: \begin{figure}
602: \leavevmode
603: \epsfxsize6.0cm\epsfbox{el.dist.eps}
604: \caption
605: {\label{fig:el.dist}
606: Schematic pictures explaining different relaxation dynamics
607: for (a) thermal and (b) non-thermal electron distributions at low temperatures.
608: Dotted lines represent the final thermal electron distribution.
609: }
610: \end{figure}
611:
612: In the second method,
613: we linearize the coupled Boltzmann equations
614: and calculate eigenvalues and eigenvectors of the matrix.
615: This second method supports and clarifies the results obtained
616: earlier by the first method.
617: With the same discretized energy levels as in the first method,
618: a total of 163 energy levels (151 electron and 12 phonon energy levels)
619: exist in our model, which means
620: a matrix of size $163 \times 163$ has to be numerically diagonalized.
621: Two of the 163 eigenmodes are unphysical as discussed earlier.
622: The remaining 161 physical modes have negative eigenvalues, $-1/\tau$,
623: as expected for a stable fixed point,
624: and represent
625: all possible relaxation modes of the system.
626: The $\tau$'s at high and low $T$'s
627: are plotted in Figs.~\ref{fig:FPEA.Lu}(a) and \ref{fig:FPEA.Lu}(c), respectively.
628: The indices of modes $i$ are assigned in descending order of
629: $\tau$.
630: The $e$-$p$ energy transfer strength $r_E$ calculated by Eq.~(\ref{rE})
631: for high and low $T$'s are plotted in
632: Figs.~\ref{fig:FPEA.Lu}(b) and \ref{fig:FPEA.Lu}(d),
633: showing the efficiency of each mode for electron-phonon energy relaxation.
634: At both $T$'s, there exist 11 modes
635: (one less than the number of phonon energy levels)
636: which have a much larger $\tau$ than the rest of the modes.
637: For example, at high $T$ these modes are 2 orders of
638: magnitude slower than the rest of the modes.
639: These slow modes,
640: whose eigenvectors have mostly phonon components and very little electron components,
641: correspond to the late time phonon-phonon
642: relaxation (see Fig.~\ref{fig:two.tau}).
643: For the high $T$ results in Figs.~\ref{fig:FPEA.Lu}(a) and \ref{fig:FPEA.Lu}(b),
644: a single mode with $i=12$ has a large $r_E$ and is very effective in
645: transferring energy from the electron system to the phonon system
646: in comparison to the other modes.
647: This mode has a value of $\tau$
648: identical to the intermediate time $\tau$ in Fig.~\ref{fig:two.tau}
649: found by the first method,
650: and is well separated from other $\tau$ values as seen
651: in Fig.~\ref{fig:FPEA.Lu}(a).~\cite{singular}
652: The electron part of its eigenvector matches $-df_{FD}/dT$,
653: confirming the results
654: obtained by the first method at high $T$ [Fig.~\ref{fig:dfdt.Lu}(a)],
655: that is, the presence of a thermal electron distribution.
656: Modes with $i$=14, 41, and 158 have smaller $r_E$
657: values
658: than the $i$=12 mode.
659: Their eigenvectors show that they participate in
660: $e$-$p$ relaxation, but the electronic parts of
661: the eigenvectors do not match $-df_{FD}/dT$.
662: Their nonthermal behavior seems suppressed in the
663: experiments and in the simulation
664: due to the thermal $i$=12 mode, which has
665: more than 5 times larger energy transfer strength
666: $r_E$.
667: The remainder of modes between $i$=13 $\sim$ 161
668: with very small $r_E$ values are mainly
669: $e$-$e$ thermalization modes.
670: These features change as $T$ is lowered, as shown in
671: Figs.~\ref{fig:FPEA.Lu}(c) and \ref{fig:FPEA.Lu}(d).
672: Many modes (modes with roughly $r_E >$ 0.1 meV/ps)
673: transfer energy from the electrons to the phonons.
674: However, none of their corresponding electronic eigenvectors
675: can be described by $-df_{FD}/dT$,
676: indicating the absence of a well defined electron temperature during
677: relaxation.
678: This again supports the results of the first method,
679: presented in Fig.~\ref{fig:dfdt.Lu}(b).
680: The rest of the modes between $i$=13 $\sim$ 161
681: with $r_E$ very close to zero again correspond
682: to $e$-$e$ relaxation modes.
683: Importantly, these have larger $\tau$ values compared to
684: the high $T$ case,
685: showing the slowing down of
686: $e$-$e$ relaxation at low $T$,
687: consistent with Fig.~\ref{fig:taueeep}.
688:
689: \begin{figure}
690: \leavevmode
691: \epsfxsize8.5cm\epsfbox{FPEA.Lu.eps}
692: \caption
693: {\label{fig:FPEA.Lu}
694: [(a), (c)] Relaxation time $\tau$ and
695: [(b), (d)] $e$-$p$ energy transfer strength $r_E$
696: at
697: [(a), (b)] high $T$ and
698: [(c), (d)] low $T$
699: calculated by the second method,
700: which linearizes the coupled Boltzmann equations.
701: The index $i$ is assigned in descending order of $\tau$.
702: }
703: \end{figure}
704:
705: Both methods show the importance of a nonthermal electron distribution
706: for the electron-phonon relaxation dynamics at low $T$ for normal metals.
707: In the next section, we discuss how the relaxation dynamics
708: of heavy fermion compounds are dramatically different from that
709: of normal metals.
710:
711: \section{Heavy fermion materials}
712:
713: As mentioned in Sec.~I, heavy fermion materials
714: are characterized by a large DOS peak with a width of order $T_K$
715: near $E_F$.
716: YbAgCu$_4$ has $T_K \sim$ 100 K
717: and is a paramagnetic metal down to the
718: lowest $T$ studied.~\cite{Sarrao99}
719: The interaction between localized $f$-electrons and
720: delocalized conduction electrons in heavy fermion materials
721: is a focus of intensive research
722: in strongly correlated electron systems
723: and has not yet been fully understood.
724: In this paper, we consider the simplest model
725: that captures the basic physics of heavy fermions,
726: that is, a peak in the DOS near $E_F$.
727: Even though some theories~\cite{McQueen94}
728: predict a disappearance
729: of the peak DOS above $T_K$, we find that
730: the calculated specific heat
731: [Fig.~\ref{fig:eDOS.Ce.Yb}(b)] and
732: relaxation time (Fig.~\ref{fig:tau.Yb}) have only
733: a weak dependence on the presence or absence of the peak DOS
734: above $T_K$.
735: (Therefore, neither the relaxation time data nor the specific heat data
736: can distinguish whether the peak DOS is present or not above $T_K$.)
737: In the hybridization gap model,~\cite{Hewson}
738: local $f$ levels hybridize with the conduction band
739: and open a gap with DOS peaks above and below the gap.
740: If $E_F$ is located within the peak, not in the gap,
741: the results we will show below have little dependence
742: on whether we use a hybridization gap model or a single peak DOS model.
743: Therefore, in our simple model,
744: we assume a $T$-independent single peak electron DOS
745: as shown in Fig.~\ref{fig:eDOS.Ce.Yb}(a).
746: We further simplify the problem by choosing the $E_F$ at the center of the peak,
747: so that the chemical potential is $T$-independent.
748: The peak DOS is described by a Gaussian function
749: with a constant background DOS:
750: \begin{equation}
751: D_e(\epsilon)=D_{peak} \exp [-(\epsilon/\Delta)^2]
752: +D_0. \label{eq:peakDOS}
753: \end{equation}
754: The discretized energy step size and energy window are identical
755: to the normal metal case in Sec.~II.
756: We determine the DOS parameters by fitting the electronic specific heat data, $C_e$,
757: as shown in Fig.~\ref{fig:eDOS.Ce.Yb}(b), and obtain
758: $D_{peak}=70$ eV$^{-1}$ f.u.$^{-1}$ spin$^{-1}$,
759: $D_0=2.1$ eV$^{-1}$ f.u.$^{-1}$ spin$^{-1}$,
760: and $\Delta = 13$ meV.
761: $D_{peak}$ is directly related to the linear slope of $C_e$ near $T=0$.
762: $\Delta$ is of the order of $T_K$ and determines the temperature
763: of the peak in $C_e$.
764: $D_0$ is identical to
765: the electron DOS at $E_F$ for LuAgCu$_4$,
766: and is related to $C_e$ above $T_K$.
767: Since LuAgCu$_4$ and YbAgCu$_4$ are isostructural with almost identical atomic masses,
768: we use identical phonon DOS for LuAgCu$_4$ and YbAgCu$_4$.
769:
770: \begin{figure}
771: \leavevmode
772: \epsfxsize8.5cm\epsfbox{eDOS.Ce.Yb.eps}
773: \caption
774: {\label{fig:eDOS.Ce.Yb}
775: (a) Model DOS and (b) calculated electron specific heat (solid circles)
776: along with experimental data of $C_e$ (line) for YbAgCu$_4$.
777: }
778: \end{figure}
779:
780: To gain insight into the nature of
781: the scattering between heavy electrons and phonons,
782: we consider a case in which only
783: heavy electrons with an isotropic parabolic dispersion
784: exist.
785: As in the normal metal case,
786: we transform the coupled Boltzmann equations with momentum indices
787: into the equations with energy indices.
788: The result is similar to the normal metal case,
789: Eqs.~(\ref{eq:Boltzmann.1})-(\ref{eq:Boltzmann.3}),
790: except for one important difference:
791: If the Fermi velocity $v_F$ is slower than the sound velocity $v_s $,
792: where $ v_F = ( \partial \epsilon / \partial k)_{\epsilon=E_F} $
793: and $ v_s = ( \partial \omega / \partial k)_{k \rightarrow 0} $,
794: then the phonon integration in the electron-phonon scattering has a
795: lower bound, which represents
796: the blocking of electron-phonon scattering for low energy phonons
797: and has a great influence on electron-phonon relaxation at low $T$.
798: Therefore, if $ v_F > v_s $,
799: Eqs.~(\ref{eq:Boltzmann.1})-(\ref{eq:Boltzmann.3}) with a peak DOS for $D_e$
800: can be used to approximately model the relaxation dynamics for both light and heavy electrons,
801: whereas if $ v_F < v_s $, appropriate blocking of scattering processes
802: should be imposed upon these equations.
803:
804: We first consider the case $v_F > v_s$,
805: for which the important physics is simply
806: the increased DOS near $E_F$ given by
807: Eq.~(\ref{eq:peakDOS}).
808: We use the same value of $K_{ep}$ and $K_{ee}$ as the LuAgCu$_4$ case in Sec. II.
809: The calculated $\tau$ (solid squares) and the experimental data (open circles)
810: shown in Fig.~\ref{fig:tau.Yb.wo.rst}
811: disagree in two respects.
812: First, the calculated $\tau$ at around $T$=300 K
813: is about 60 times less than the observed $\tau$.
814: (Note that, experimentally, both YbAgCu$_4$ and LuAgCu$_4$ have similar $\tau$
815: at $\sim $ 300 K.)
816: This difference is due to the increased electron DOS in the calculation,
817: which enhances the e-p relaxation.
818: An approximately 60 times smaller $K_{ep}$
819: would shift the calculated $\tau$ in the whole $T$ range by 60 times,
820: but it is unlikely that $K_{ep}$ in YbAgCu$_4$
821: would be smaller than $K_{ep}$ in LuAgCu$_4$
822: in such a drastic way.
823: Secondly, and more importantly,
824: the divergence of calculated $\tau$'s at low $T$ is much weaker than the experimental data.
825: This slow divergence in calculated $\tau$ can be understood in the following way.
826: The large electron DOS increases both the electron-electron and the electron-phonon
827: scattering rates due to the increased number of available final states
828: [see Eqs.~(\ref{eq:Boltzmann.1})-(\ref{eq:Boltzmann.3})].
829: However, electron-phonon scattering in Eq.~(\ref{eq:Boltzmann.1})
830: has only one factor of $D_e$,
831: whereas electron-electron scattering in Eq.~(\ref{eq:Boltzmann.3})
832: has three factors of $D_e$.
833: Therefore, the electron-electron scattering rate increases much faster than
834: the electron-phonon scattering rate, which makes the electron system thermal and
835: the TTM description valid.~\cite{OTM}
836: According to Kaganov {\it et al.}'s calculation,~\cite{Kaganov56}
837: $g(T)$ in Eq.~(\ref{eq:TTM})
838: goes as $g(T) \sim T^4$ at low $T$ and $g(T) \sim constant$ at high $T$.
839: For normal metals, since $C_e \ll C_p$ even down to 10 K,
840: the TTM relation Eq.~(\ref{eq:TTM}) gives
841: $\tau \sim T^{-3}$.
842: In the case of heavy fermion compounds,
843: $C_e$ is comparable to or larger than $C_p$ at low $T$,
844: and therefore $C_p \sim T^3$ for the Debye phonons
845: produces $\tau^{-1} \sim T$ as the leading $T$ dependence.
846: Indeed, the low-$T$ divergence of the calculated $\tau$ in Fig.~\ref{fig:tau.Yb.wo.rst}
847: can be well described as $T^{-1}$.
848: Therefore, the discrepancy with experimental data
849: cannot be fixed by simply changing parameters,
850: and it indicates that there is additional physics involved
851: other than the large DOS at the Fermi level to explain
852: a 100-fold increase of $\tau$ below $T_K$.
853:
854: \begin{figure}
855: \leavevmode
856: \epsfxsize8.0cm\epsfbox{tau.Yb.wo.rst.eps}
857: \caption
858: {\label{fig:tau.Yb.wo.rst}
859: Relaxation time calculated
860: with the peak DOS only [Eq.~(\ref{eq:peakDOS}) and Fig.~\ref{fig:eDOS.Ce.Yb}(a)],
861: along with experimental data for YbAgCu$_4$.
862: }
863: \end{figure}
864:
865: If we take the approximation that
866: the difference between $E_F$ and the bottom of the heavy electron band
867: is about $T_K$, then from the isotropic parabolic heavy electron dispersion model
868: and the carrier density of $n$ = 0.85 measured
869: by the Hall effect for YbAgCu$_4$,~\cite{Sarrao99}
870: we can estimate $v_F \approx $ 4 km/sec.
871: Ultrasonic measurements show that
872: the longitudinal sound velocity along the [111] direction is
873: about 4.4 km/sec.~\cite{Zherlitsyn99}
874: This comparison shows that in fact
875: the two velocities are comparable to each other,
876: supporting the possibility of $v_F < v_s$.
877: We should note that experiments like de-Haas van-Alphen
878: are required for
879: a more reliable estimate of $v_F$.
880: Therefore, we hypothesize that
881: the Fermi velocity is lower than the sound velocity,
882: and discuss how this affects the electron-phonon relaxation dynamics.
883:
884: As depicted in Fig.~\ref{fig:vfvs},
885: if the Fermi velocity is slower than the sound velocity [Fig.~\ref{fig:vfvs}(b)],
886: then momentum and energy conservation requirements
887: prohibit scattering between heavy electrons
888: and phonons.
889: In the actual heavy fermion DOS, heavy electron dispersion exists
890: only within the peak DOS around $E_F$,
891: and electrons outside the peak
892: have a regular light mass.
893: Therefore, the above blocking mechanism
894: applies only when both the initial and final
895: electron states are within the peak of the DOS.
896: If one or both of the states are outside the peak,
897: regular electron-phonon scattering is expected.
898: These effects can be simply incorporated into the simulation
899: by blocking electron-phonon scattering within the peak DOS.
900: The results of the simulation are shown in Fig.~\ref{fig:tau.Yb}
901: as solid circles,
902: for which the blocked energy range is from -24 meV to 24 meV,
903: $K_{ep}$ = 0.23 eV/ps, and $K_{ee}=700 \times K_{ep}$.
904: This $K_{ep}$ value is smaller than the value for LuAgCu$_4$ case,
905: by about half in terms of the scattering matrix element.
906: It shows reasonable agreement with experimental data (open circles).
907:
908: \begin{figure}
909: \leavevmode
910: \epsfxsize5.0cm\epsfbox{vfvs.eps}
911: \caption
912: {\label{fig:vfvs}
913: Schematic pictures explaining different electron-phonon scattering processes
914: for (a) $v_F > v_s$ and (b) $v_F < v_s$ cases.
915: If $v_F < v_s$, then the electron dispersion changes
916: more slowly than the phonon dispersion.
917: Therefore, the distance between
918: the two $k$-space spheres with energies $\epsilon $ and $\epsilon - \omega $
919: is larger than the phonon momentum $q$,
920: and no momentum and energy conserving scattering process is possible,
921: as shown in (b) in this figure.
922: }
923: \end{figure}
924:
925: \begin{figure}
926: \leavevmode
927: \epsfxsize8.0cm\epsfbox{tau.Yb.eps}
928: \caption
929: {\label{fig:tau.Yb}
930: Solid circles: relaxation time calculated with peaked DOS
931: and blocking of e-p scattering within the peak DOS due to $v_F < v_s$.
932: Open circles: experimental data for YbAgCu$_4$.
933: Solid line: results obtained in Ref.~\onlinecite{Demsar03} using TTM
934: with partial blocking of the e-p scattering within the peak.
935: }
936: \end{figure}
937:
938: Since the electron system has a thermal distribution,
939: the TTM approach, with a similar blocking of electron-phonon scattering within the peak,
940: can be used to describe the relaxation between
941: the electrons and phonons,
942: which was discussed in Ref.~\onlinecite{Demsar03}.
943: In this TTM approach, we also included the possibility that some electron-phonon
944: scattering within the peak is allowed due to the anisotropy of
945: $v_F(\hat{k})$ and $v_s(\hat{k})$,
946: that is, $v_F > v_s$ along some directions and $v_F < v_s$ along other directions.
947: The results were shown as a solid line in Fig.~3 in Ref.~\onlinecite{Demsar03}
948: and reproduced in Fig.~\ref{fig:tau.Yb} (solid line),
949: which shows good agreement with experimental data.~\cite{light.HF}
950:
951: The physical idea behind the increase of $\tau$ below $T_K$ is simple:
952: As $T$ decreases below $T_K$, or equivalently the width of the DOS peak,
953: since the electron system has a thermal distribution,
954: the main electron-phonon relaxation processes should occur
955: within the peak [see Fig.~\ref{fig:el.dist}(a)],
956: where the momentum and energy conservation laws block
957: the electron-phonon scattering.
958: Therefore, the electron-phonon relaxation time increases very rapidly
959: as $T$ is lowered below $T_K$.~\cite{T.dep.pk}
960: The results show that complete or substantial
961: blocking of electron-phonon scattering processes within the DOS peak
962: is essential to explain the rapid increase of the electron-phonon
963: relaxation time below $T_K$ in YbAgCu$_4$.
964:
965: \section{Summary}
966:
967: We provide a theoretical analysis of the ultrafast relaxation dynamics
968: observed by femtosecond time-resolved optical spectroscopy
969: in isostructural LuAgCu$_4$ and YbAgCu$_4$; the former is a normal metal
970: and the latter is a heavy fermion compound.
971: For normal metals, we find that a nonthermal electron distribution
972: is responsible for
973: a temperature-independent electron-phonon relaxation time
974: at low temperatures,
975: instead of a $T^{-3}$ divergent behavior predicted by the two-temperature model.
976: For heavy fermion compounds, we find that prohibiting electron-phonon
977: scattering within the density-of-states peak near the Fermi energy
978: is crucial to explain the rapid
979: increase of the electron-phonon relaxation time below the Kondo temperature.
980: On the basis of the estimated Fermi velocity and the measured sound velocity,
981: we propose the hypothesis that the slower Fermi velocity compared to the sound velocity
982: provides this blocking mechanism due to energy and momentum conservation laws.
983: We find good agreement between the experimental data and our model
984: for both normal metals and heavy fermion compounds.
985:
986: We thank V. V. Kabanov for useful discussions.
987: This work has been supported by U.S. DOE.
988:
989: \begin{thebibliography}{99}
990:
991: \bibitem[*]{Demsar}
992: Current address :
993: ``J. Stefan'' Institute, Jamova 39, SI-1000, Ljubljana, Slovenia.
994:
995: \bibitem{Groeneveld92}
996: R. H. M. Groeneveld, R. Sprik, and A. Lagendijk,
997: Phys. Rev B {\bf 45}, 5079 (1992).
998:
999: \bibitem{Groeneveld95}
1000: R. H. M. Groeneveld, R. Sprik, and A. Lagendijk,
1001: Phys. Rev B {\bf 51}, 11433 (1995).
1002:
1003: \bibitem{Brorson90}
1004: S. D. Brorson, A. Kazeroonian, J. S. Moodera, D. W. Face, T. K. Cheng, E. P. Ippen, M. S. Dresselhaus, and G. Dresselhaus,
1005: Phys. Rev. Lett. {\bf 64}, 2172 (1990).
1006:
1007: \bibitem{Stevens97}
1008: C. J. Stevens, D. Smith, C. Chen, J. F. Ryan, B. Podobnik, D. Mihailovic,
1009: G. A. Wagner, and J. E. Evetts,
1010: Phys. Rev. Lett. {\bf 78}, 2212 (1997);
1011: J. Demsar, B. Podobnik, V. V. Kabanov, Th. Wolf, and D. Mihailovic,
1012: Phys. Rev. Lett. {\bf 82}, 4918 (1999).
1013:
1014: \bibitem{Demsar99}
1015: J. Demsar, K. Biljakovi\'{c}, and D. Mihailovic,
1016: Phys. Rev. Lett. {\bf 83}, 800 (1999).
1017:
1018: \bibitem{Averitt01}
1019: R. D. Averitt, A. I. Lobad, C. Kwon, S. A. Trugman, V. K. Thorsm\o lle,
1020: and A. J. Taylor,
1021: Phys. Rev. Lett. {\bf 87}, 017401 (2001).
1022:
1023: \bibitem{Demsar03}
1024: J. Demsar, R. D. Averitt, K. H. Ahn, M. J. Graf, S. A. Trugman,
1025: V. V. Kabanov, J. L. Sarrao, and A. J. Taylor,
1026: Phys. Rev. Lett. {\bf 91}, 027401 (2003).
1027:
1028: \bibitem{Pagliuso97}
1029: P. G. Pagliuso, C. Rettori, S. B. Oseroff, J. Sarrao, Z. Fisk,
1030: A. Cornelius, and M. F. Hundley,
1031: Phys. Rev. B {\bf 56}, 8933 (1997).
1032:
1033: \bibitem{LuAgCu4}
1034: LuAgCu$_4$ has
1035: a 15 times larger density of states at the Fermi energy ($E_F$)
1036: compared to Ag
1037: due to the large number of bands at $E_F$,
1038: not due to the hybridization of
1039: local $f$-levels and conduction levels.
1040:
1041: \bibitem{Sarrao99}
1042: J. L. Sarrao, C. D. Immer, Z. Fisk, C. H. Booth,
1043: E. Figueroa, J. M. Lawrence, R. Modler, A. L. Cornelius,
1044: M. F. Hundley, G. H. Kwei, J. D. Thompson, and F. Bridges,
1045: Phys. Rev. B {\bf 59}, 6855 (1999);
1046: J. M. Lawrence, P. S. Riseborough, C. H. Booth,
1047: J. L. Sarrao, J. D. Thompson, and R. Osborn,
1048: Phys. Rev. B {\bf 63}, 054427 (2001).
1049:
1050: \bibitem{YbAgCu4}
1051: The Sommerfeld coefficient $\gamma$ for YbAgCu$_4$ is 210 mJ/mol K$^2$,~\cite{Sarrao99}
1052: which corresponds to a 300 times larger DOS at $E_F$ compared to Ag.
1053:
1054: \bibitem{Kaganov56}
1055: M. I. Kaganov, I. M. Lifshitz, and L. V. Tanatarov,
1056: Zh. Eksp. Teor. Fiz. {\bf 31}, 232 (1956)
1057: [Sov. Phys. JETP {\bf 4}, 173 (1957)];
1058: S. I. Anisimov, B. L. Kapeliovitch, and T. L. Perel'man,
1059: Zh. Eksp. Teor. Fiz. {\bf 66}, 3776 (1974)
1060: [Sov. Phys. JETP {\bf 39}, 375 (1974)];
1061: P. B. Allen, Phys. Rev. Lett. {\bf 59}, 1460 (1987).
1062:
1063: \bibitem{TTM.eq}
1064: The TTM result is found from
1065: $ C_1 \frac{dT_1}{dt} =- C_2 \frac{dT_2}{dt} = -g(T) (T_1 - T_2) $.
1066:
1067: \bibitem{pp}
1068: We have examined the effects of phonon-phonon scattering,
1069: and found that it has
1070: a negligible effect on the electron-phonon relaxation time.
1071:
1072: \bibitem{AM}
1073: N. W. Ashcroft and N. D. Mermin,
1074: {\it Solid State Physics} (Holt, Rinehart, and Winston, New York, 1976),
1075: Chaps. 16 and 26.
1076:
1077: \bibitem{Landau}
1078: L. D. Landau and E. M. Lifshitz,
1079: {\it Physical Kinetics}, Vol. 10
1080: of {\it Course on Theoretical Physics}
1081: (Pergamon, Oxford, 1981).
1082:
1083: \bibitem{Ziman}
1084: J. M. Ziman, {\it Electrons and phonons}
1085: (Oxford University Press, London, 1972).
1086:
1087: \bibitem{3points}
1088: From the relation $ \tau = ( - dE_e/dt ) / ( d^2E_e/dt^2)$ for
1089: $E_e(t)=E_{0}+\Delta E \exp (-t/\tau)$,
1090: we use
1091: $ \tau (t) = - [E_e(t+dt)-E_e(t-dt)] dt / \{ 2 [E_e(t+dt)-2 E_e(t)+ E_e(t-dt)] \}$.
1092:
1093: \bibitem{late.tau}
1094: The phonon-phonon thermalization process occurs via electron-phonon scattering,
1095: since our Boltzmann equations do not include direct phonon-phonon scattering.
1096: For example, a high energy phonon is absorbed by an electron,
1097: which emits two low energy phonons.
1098:
1099: \bibitem{ee}
1100: The presence of the extra feature near $E_F$
1101: depends on details of the parameter values and time at which $df/dt$ is evaluated.
1102: When this feature is present, as shown in Fig.~\ref{fig:dfdt.Lu}(b),
1103: it is partly due to electrons heating up
1104: near $E_F$ through electron-electron scattering.
1105: (Since electron-electron scattering conserves
1106: total electron energy,
1107: $(df/dt)_{ee}$ must cross zero at least three times.)
1108:
1109: \bibitem{TTM.phonon}
1110: Contrary to Groeneveld {\it et al}'s
1111: assumption,~\cite{Groeneveld92,Groeneveld95}
1112: the large phonon specific heat compared
1113: to the electron specific heat may not be the reason why TTM results apply to
1114: nonthermal phonons,
1115: because our simulation
1116: agrees with TTM predictions
1117: even when the electron specific heat is larger than the phonon
1118: specific heat, as shown in Sec.~III
1119: (see the discussion of Fig.~\ref{fig:tau.Yb.wo.rst}).
1120:
1121: \bibitem{singular}
1122: When the step size for energy discretization
1123: is reduced by half,
1124: the gaps in $\tau$ between this single mode and the edges
1125: of the continuum-like faster and slower $\tau$'s
1126: do not change.
1127: This singular behavior deserves future theoretical studies.
1128:
1129: \bibitem{McQueen94}
1130: P. G. McQueen, D. W. Hess, and J. W. Serene,
1131: Phys. Rev. B {\bf 50}, 7304 (1994);
1132: B. N. Bickers, D. L. Cox, and J. H. Wilkins, {\it ibid.} {\bf 36}, 2036 (1987);
1133: B. Horvati\'{c}, D. \v{S}ok\v{c}evi\'{c}, and
1134: V. Zlati\'{c}, {\it ibid.} {\bf 36}, 675 (1987).
1135:
1136: \bibitem{Hewson}
1137: A. C. Hewson, {\it The Kondo problem to heavy fermion},
1138: (Cambridge University Press, Cambridge, 1993), Chap. 10.
1139:
1140: \bibitem{OTM}
1141: We find that the electron thermalization recovers the TTM predictions,
1142: as in normal metals (see Fig.~\ref{fig:tau.Lu.s.rep0}).
1143:
1144: \bibitem{Zherlitsyn99}
1145: S. Zherlitsyn, B. L\"{u}thi, B. Wolf,
1146: J. L. Sarrao, Z. Fisk, and V. Zlatic,
1147: Phys. Rev. B {\bf 60}, 3148 (1999).
1148:
1149: \bibitem{light.HF}
1150: The hypothesis that the relation $v_F < v_s$ provides
1151: the blocking mechanism indicates that
1152: the heavy fermion metals with lighter effecttive masses
1153: may show relaxation dynamics
1154: qualitatively different from that of YbAgCu$_4$.
1155:
1156: \bibitem{T.dep.pk}
1157: Even when the electron DOS is $T$-dependent,
1158: as long as the peak width in the DOS does not
1159: increase faster than $T$,
1160: the results do not change much.
1161:
1162: \end{thebibliography}
1163:
1164: \end{document}
1165: