cond-mat0009369/pl.tex
1: \documentstyle[aps,prb,preprint]{revtex}
2: \renewcommand{\baselinestretch}{1}
3: \begin{document}
4: \draft
5: \title{Kinetics of exciton photoluminescence  in type-II 
6: semiconductor superlattices} 
7: 
8: \author{L.~S.~Braginsky\thanks{Electronic address: 
9: brag@isp.nsc.ru},  M.~Yu.~Zaharov, A.~M.~Gilinsky,
10: V.~V.~Preobrazhenskii, M.~A.~Putyato, and K.~S.~Zhuravlev}
11: \address{Institute of Semiconductor Physics, 630090, 
12: Novosibirsk, Russia}
13: 
14: \date{\today}
15: \maketitle
16: \begin{abstract}
17: The  exciton decay rate  at a rough 
18: interface in type-II semiconductor superlattices is 
19: investigated. It is shown that the possibility of recombination 
20: of indirect excitons at a plane interface essentially affects 
21: kinetics of the exciton photoluminescence at a rough interface. 
22: This happens because of  strong correlation between the 
23: exciton recombination at the plane interface and at the 
24: roughness.  Expressions that relate the parameters of the 
25: luminescence kinetics with statistical characteristics of the 
26: rough interface  are obtained.  The mean height and 
27: length of roughnesses in GaAs/AlAs superlattices are estimated 
28: from the experimental data.  
29: \end{abstract} 
30: \pacs{78.66.Fd} 
31: \narrowtext 
32: \section{Introduction}
33: GaAs/AlAs type-II superlattices are the subject of extensive 
34: investigations in the recent decade. Electrons and holes are 
35: separated in these structures: holes are confined in the 
36: $\Gamma$ valley of GaAs, whereas electrons are   in $X$ valleys 
37: of AlAs. Changing the width of AlAs layer during the growth 
38: of the structure, it is possible to confine the electrons 
39: either in the $X_z$ valley ($X$ valley that is directed 
40: along the structure axis [001]) or in the $X_{xy}$ valley 
41: ($X$ valley that is directed along the GaAs/AlAs interface: 
42: [100] or [010]).  The excitons in such structures are indirect 
43: in both real and momentum spaces. 
44: 
45: Kinetics of the exciton luminescence is 
46: investigated by the time-resolved method.
47: The theory by Klein {\it et al. } \cite{Klein} is usually used 
48: to explain the results of such experiments.  The theory 
49: has been developed to consider the no-phonon 
50: radiative decay rates of indirect excitons in  alloy 
51: semiconductors (e.g., Ga$_{1-x}$Al$_x$As).  The recombination of 
52: indirect excitons occurs because of intervalley scattering of 
53: electrons at the potential fluctuations caused by the 
54: compositional disorder. These short-range scatterers are 
55: necessary to compensate the large momentum of the electron in 
56: the $X$ valley. The nonexponential dependence of the decay rate 
57: has been obtained 
58: \begin{equation} 
59: \label{1} 
60: I(t)\propto e^{-w_0t}(1+2w_rt)^{-3/2}.  
61: \end{equation} 
62: Where the value $w_r$ is connected with the compositional 
63: disorder. The exponential factor has been included in 
64: Eq.~(\ref{1}) to consider  different nonstochastic processes of 
65: the exciton recombination (e.g., the phonon-assisted 
66: recombination).  This is possible only in  the absence of 
67: correlation between stochastic and nonstochastic processes.
68: 
69: The possibility to apply the theory \cite{Klein} for 
70: superlattices has been discussed by F.~Minami 
71: {\it et al.}.~\cite{Minami} Authors suppose the short-range 
72: scatterers are distributed along the plane boundary.  This 
73: assumption justifies the application of Eq.~(\ref{1}) for 
74: superlattices; however, it does not allow to relate the 
75: parameter $w_r$ with characteristics of the rough interface, 
76: e.g., the mean height and length of roughnesses. I.  Krivorotov 
77: {\it et al.}\cite{Krivirotov} have showed that nonradiative 
78: decay due to exciton trapping by interfacial defects also 
79: results in nonexponential factor in Eq.~(\ref{1}).  
80: Nevertheless, Eq.~(\ref{1}), wherein the parameters $w_r$ and 
81: $w_0$ are considered as trial, is commonly used for explanation 
82: of the experimental results.\cite{Nagao}
83: 
84: It should be noted that  roughnesses are not necessary  for 
85: the recombination of $X_z$ excitons. Their recombination occurs 
86: even at a plain interface where the normal component of the 
87: electron momentum relaxes. This  important 
88: point also distinguishes the exciton recombination in 
89: superlattices.  The process, however, can not be taken into 
90: account by a simple exponential factor. Indeed, the wave function 
91: of the electron at a rough interface is the 
92: sum of its regular and diffuse components. The first one exists 
93: at a plane interface, whereas the latter is due to the 
94: roughnesses. For this reason the crossed terms arise in 
95: the interband matrix element; so that the probability of the  
96: exciton recombination, which is determined by the squared module 
97: of this matrix element, is no longer   a simple sum of the 
98: probabilities of the  recombination at the plane 
99: interface and at the roughnesses.  This correlation leads to a 
100: more complicated relation than the simple exponential factor in 
101: Eq.~(\ref{1}).
102: 
103: 
104: 
105: In this paper we consider a more realistic model of the rough 
106: interface. We show that Eq.~(\ref{1}) holds for 
107: the decay rate of  $X_{xy}$ excitons and   relate 
108: the $w_r$ value with parameters of the rough interface. We 
109: determine the decay rate of  $X_z$ excitons. In particular,   
110: it is found that this value at  large times behaves roughly as 
111: $I(t)\propto \exp (-w_0t)/t$, rather than $I(t)\propto \exp 
112: (-w_0t)/t^{3/2}$ as it is predicted by Eq.~(\ref{1}). 
113: 
114: Our experiments on the GaAs/AlAs type-II superlattices confirm 
115: these results. We use the experimental data for the radiative 
116: decay rates to estimate the parameters of the rough interface. 
117: The mean height of roughnesses was found to be close to  
118: the lattice constant, whereas their mean length 
119: is about $50\,\AA$. 
120: 
121: \section{Radiative decay rates of indirect excitons in 
122: superlattices. Theory}
123: 
124: Let $z=0$ be the interface between GaAs ($-d_1<z<0$) and AlAs 
125: ($0<z<d_2$), and  $\bbox{\rho}$ be the vector in the $XY$ plane.
126: We consider the exiton recombination at the interface 
127: and write the exciton wave function as 
128: follows:\cite{Ivchenko}
129: \[
130:  \phi({\bf r}_e,{\bf r}_h)=f_e(z_e)f_h(z_h)
131: G(\bbox{\rho}_e-\bbox{\rho}_h,z_e,z_h).
132: \]
133: Where ${\bf r}_e=\{\bbox{\rho}_e,\ z_e\}$ and ${\bf 
134: r}_h=\{\bbox{\rho}_h,\ z_h\}$ are coordinates of  the electron 
135: and the hole,  $f_e(z_e)$ and $f_h(z_h)$ are their wave 
136: functions  in the absence of Coulomb interaction; the 
137: function $G$  takes into account this interaction.  The 
138: probability for the exciton recombination is proportional to 
139: $G^2({\bf 0})$
140: [$G({\bf 0})\equiv G(\bbox{\rho}_e=\bbox{\rho}_h,z_e=z_h=0)$] 
141: and the squared module of the matrix element 
142: \begin{equation} 
143: \label{2} 
144: {\cal P}= \int 
145: f_e(z)\bbox{\nabla}f_h(z)\,dz\,d^2\bbox{\rho}. 
146: \end{equation}
147:  
148: The  functions $f_e(z_e)$ and $f_h(z_h)$ can be expressed via 
149: envelope wave functions of the electron and the hole in the 
150: conduction and valence bands of GaAs and AlAs. To determine the 
151: envelopes, the appropriate boundary conditions at the GaAs/AlAs 
152: interface should be imposed. The roughness of the interface 
153: has an influence on these boundary conditions and, therefore, 
154: affects the envelopes. We shall consider the rough interface 
155: where the mean height of roughnesses is small in comparison 
156: with the electron wavelength (or the exciton size).  This 
157: allows us to use the boundary conditions at the rough 
158: interface\cite{PhysicaE} to consider an influence of 
159: roughnesses on the exciton recombination. 
160: 
161: \subsection{Boundary conditions for the envelope wave functions 
162: at a GaAs/AlAs interface}
163: \subsubsection{Boundary conditions at a plane interface}
164: In general, the boundary conditions for the electron 
165: envelopes can be written as follows:
166: \begin{equation}
167: \label{3}
168: \left(
169: \begin{array}{l}
170: \Psi_\Gamma^r\\
171: \Psi_\Gamma^{r'}\\
172: \Psi_{Xxy}^r\\
173: \Psi_{Xxy}^{r'}\\
174: \Psi_{Xz}^r\\
175: \Psi_{Xz}^{r'}
176: \end{array}
177: \right)
178: =\tilde{T}
179: \left(
180: \begin{array}{l}
181: \Psi_\Gamma^l\\
182: \Psi_\Gamma^{l'}\\
183: \Psi_{Xxy}^l\\
184: \Psi_{Xxy}^{l'}\\
185: \Psi_{Xz}^l\\
186: \Psi_{Xz}^{l'}
187: \end{array}
188: \right).
189: \end{equation}
190: Where  $\Psi_{\Gamma,Xxy,Xz}^{l,r}$ are the envelopes which 
191: correspond to the $\Gamma$, $X_{xy}$, and $X_z$ valleys of GaAs 
192: and AlAs; $\Psi_{\Gamma,Xxy,Xz}^{l,r'}\equiv \partial 
193: \Psi_{\Gamma,Xxy,Xz}^{l,r}/\partial z$ are their normal 
194: derivatives.  The elements $\tilde{t}_{ik}$ of the $6\times 6$ 
195: matrix $\tilde{T}$ are determined by the interface structure. 
196: They are independent of the electron energy. For the GaAs/AlGaAs 
197: interface they have been calculated by Ando {\it et 
198: al.}.\cite{Ando} 
199: 
200: We shall consider the particular cases of $X_z$ and $X_{xy}$ 
201: excitons.   This allows us to simplify Eq.~(\ref{3}).  
202: First, we omit mixing between $X_z$ and $X_{xy}$ valleys.  
203: Second, the energy position of $\Gamma$ minimum in AlAs is 
204: considerably higher than that of $X$ minimum.  For this reason 
205: the wave function $\Psi_\Gamma^r$ decays rapidly apart of the 
206: interface.  We have $\Psi_\Gamma^r\propto \exp(-\gamma^r z),\ 
207: \Psi_\Gamma^{r'}=-\gamma^r\Psi_\Gamma^{r}$, where 
208: $\gamma^r=\sqrt{2m_{\Gamma}^r(E_\Gamma-\varepsilon_e)}$ (here 
209: $m_{\Gamma}^r$ is  effective mass in the $\Gamma$ valley of 
210: AlAs, $\varepsilon_e\approx E_X$ is the electron energy, 
211: $E_\Gamma$ and $E_X$ are energies of bottoms of the $\Gamma$ and 
212: $X$ valleys) can be considered as independent of the electron 
213: energy. By  eliminating of $\Psi_\Gamma^{r}$ from Eq.~(\ref{3}),  
214: for $X_z$ electrons we find:  
215: \begin{mathletters} 
216: \label{bcplane} 
217: \begin{equation}
218: \label{4}
219: \left\{
220: \begin{array}{l}
221: \Psi_{Xz}^r=\Psi_{Xz}^l,\\
222: \Psi_{Xz}^{r'}= 
223: t^z_{41}\Psi_\Gamma^{l}+t^z_{44}\Psi_{Xz}^{l'},\\ 
224: \Psi_\Gamma^l+t^z_{12}\Psi_\Gamma^{l'}+ t^z_{13}\Psi_{Xz}^l=0.
225: \end{array}
226: \right.
227: \end{equation}
228: Where $t^z_{44}\approx m_{Xl}^r/m_{Xl}^l\approx 1$,  this value 
229: takes into account the difference of longitudinal effective 
230: masses in the $X$ valleys of AlAs and GaAs;  $t^z_{41}=t_{\Gamma 
231: X} m^r_{Xl}/(m_ea)$, $t^z_{12}=m^r_\Gamma/(m^l_\Gamma\gamma^r)$, 
232: $t^z_{13}=t_{\Gamma X}m^l_\Gamma/(m_ea\gamma^r)\ll 1$; 
233: $t_{\Gamma X}\approx 1$ is the parameter of $\Gamma$--$X$ 
234: mixing. Other elements of the ${t}^z_{ik}$ matrix   are small; 
235: this is the result of numerical calculations.\cite{Ando} 
236: 
237: Note that the band states in the $X$ valley   result from 
238: interaction of two close-lying bands: lower $X_1$ and upper 
239: $X_3$; meanwhile only the $X_3$ states mix effectively with 
240: $\Gamma$ states.  This means that  $t_{\Gamma X}\approx 1$ is 
241: the upper estimation of $\Gamma$--$X$ mixing.
242: 
243: It is sufficient to consider only $X$ valleys of each 
244: contacted material when $X_{xy}$ electrons are investigated. 
245: Assuming $\Psi_\Gamma^{l'}=\gamma^l\Psi_\Gamma^{l}$, where 
246: $\gamma^l\sim 2\pi/a$ ($a$ is the lattice constant), from 
247: Eq.~(\ref{3}) we find: 
248: \begin{eqnarray} 
249: \label{5} 
250: &&\Psi_{Xxy}^r=t^{xy}_{11}\Psi_{Xxy}^l+t^{xy}_{12}\Psi_{Xxy}^{l'},\\
251: &&\Psi_{Xxy}^{r'}=t^{xy}_{21}\Psi_{Xxy}^l+t^{xy}_{22}\Psi_{Xxy}^{l'}.
252: \nonumber 
253: \end{eqnarray}
254: Where $|t^{xy}_{12}|\ll 1,\ |t^{xy}_{21}|\ll a^{-1},\ 
255: |t^{xy}_{11}|\approx 1$, and $t^{xy}_{22}\approx 
256: m_{Xt}^r/m_{Xt}^l\approx 1$; $m_{Xt}^r$ and $m_{Xt}^l$ are  the
257: transversal effective masses of AlAs and GaAs.
258: 
259: The bands of the light and heavy holes  are splitted due 
260: to the size quantization.  
261: This allows us to consider only the heavy holes in each material 
262: and write the boundary conditions for them as follows: 
263: \begin{eqnarray} 
264: \label{6} 
265: &&\Psi_h^r=t^h_{11}\Psi_h^l+t^h_{12}\Psi_h^{l'},\\
266: &&\Psi_h^{r'}=t^h_{21}\Psi_h^l+t^h_{22}\Psi_h^{l'}.
267: \nonumber 
268: \end{eqnarray}
269: Where $\Psi_h^{l,r}$ are the envelopes for the heavy holes in 
270: each material. 
271: \end{mathletters}
272: 
273: \subsubsection{Boundary conditions at a rough interface}
274: We shall consider the model of a rough interface that is 
275: presented on Fig.~1.  This model is in  agreement with optical 
276: \cite{Lurssen} and structural \cite{Bechstedt}  investigations 
277: of GaAs/AlAs interface.  The interface looks like an array of 
278: the plane areas of the same crystallographic orientation. The 
279: random function $z=\xi(\bbox{\rho})$ of the  coordinates in the 
280: $XY$ plane determines the positions of these areas relative to 
281: $z=0$. 
282: 
283: We  assume the average 
284: height of roughnesses $h$ to be small in comparison with 
285: the electron wavelength. Then it is possible to describe the 
286: rough interface by means of the correlation function $W(\bbox 
287: {\rho '}, \bbox{\rho ''})=\overline{\xi(\bbox{\rho '})\xi(\bbox{\rho 
288: ''})}$.  For the homogeneous  rough interface $W(\bbox{\rho 
289: '}, \bbox{\rho ''})=W(\bbox{\rho '}-\bbox{\rho ''})$, i.e.,   
290: the correlation function is the function of one variable:  
291: $\bbox{\rho}=\bbox{\rho '}-\bbox{\rho ''}$.  There are two 
292: parameters that are most important when the statistical 
293: properties of a rough interface are considered:  $h^2=W(0)$ and 
294: the correlation length $l$ --- the mean attenuation length of 
295: the correlation function.  In our model the correlation length 
296: $l$ can be associated with the mean size of the plane area. 
297: 
298: The special form of the rough interface (Fig.~1) allows  us to 
299: apply the boundary conditions (\ref{bcplane}), which are 
300: applicable at a plane interface, at each plane $z=\xi$. 
301: The inequality $|\xi\Psi'|\sim h/\lambda\ll 1$ ($\lambda$ is the 
302: electron wavelength) allows to rewrite these boundary conditions 
303: at a plane $z=0$.  After some algebra  we obtain:
304: \begin{mathletters} 
305: \label{bcrough}
306: \begin{eqnarray}
307: \label{7}
308: &&\Psi_{Xz}^r=-t^z_{41}\eta(\xi)\xi(\bbox{\rho})\Psi_\Gamma^l
309: +\Psi_{Xz}^l+
310: (1-t^z_{44})\xi(\bbox{\rho})\Psi_{Xz}^{l'},\nonumber\\
311: &&\Psi_{Xz}^{r'}=t^z_{41}\eta(\xi)\Psi_\Gamma^l
312: +t^z_{41}\eta(\xi)\xi(\bbox{\rho})\Psi_\Gamma^{l'}
313: t^z_{44}\Psi_{Xz}^{l'},\\
314: &&\Psi_\Gamma^l+[t^z_{12}+\xi(\bbox{\rho})]\Psi_\Gamma^{l'}
315: +t^z_{13}\eta^*(\xi)\Psi_{Xz}^l
316: +t^z_{13}\eta^*(\xi)\xi(\bbox{\rho})\Psi_{Xz}^{l'},
317: \nonumber 
318: \end{eqnarray}
319: for  electrons in the $X_z$ valley;
320: \begin{eqnarray}
321: \label{8}
322: &&\Psi_{Xxy}^r=\Psi_{Xxy}^l+(1-t^{xy}_{22})\xi(\bbox{\rho})\Psi_{Xxy}^{l'},\\
323: &&\Psi_{Xxy}^{r'}=t^{xy}_{21}\Psi_{Xxy}+t^{xy}_{22}\Psi_{Xxy}^{l'},
324: \nonumber 
325: \end{eqnarray}
326: for  electrons in the $X_{xy}$ valley; and
327: \begin{eqnarray}
328: \label{9}
329: &&\Psi_h^r=\Psi_h^l+(1-t^h_{22})\xi(\bbox{\rho})\Psi_h^{l'},\\
330: &&\Psi_h^{r'}=t^h_{21}\Psi_h+t^h_{22}\Psi_h^{l'},
331: \nonumber 
332: \end{eqnarray}
333: for the holes. Factor $\eta(\xi)=\exp(2\pi i \xi/a)$ in 
334: Eq.~(\ref{7}) takes the two values $\pm 1$ for $\xi=a$ or 
335: $\xi=a/2$. It has been introduced in Ref.~\cite{Ivchenko1} to 
336: take into account the symmetry properies of the Bloch functions 
337: with respect to translation by a single monomolecular layer 
338: ($a/2$) along the $z$ axis.  The Bloch function of the electron 
339: in the $X_z$ valley changes its sigh under this translation 
340: whereas the Bloch function of the electron in the $\Gamma$ 
341: valley does not.  Therefore, the parameter $t_{\Gamma X}$ of 
342: $\Gamma$--$X$ mixing also should change sigh under such 
343: translation. This is not important at a plane interface, but it 
344: must be taken into account when the relative positions of some 
345: interfaces are considered. We assume 
346: $|t^{xy,h}_{21}|\ll a^{-1}$:  this is the result of numerical 
347: calculations.\cite{Ando} 
348: \end{mathletters}
349: 
350: Unlike Eqs.~(\ref{bcplane}) the boundary conditions 
351: (\ref{bcrough}) contain the terms depended on $\xi$. They would 
352: not be important, if $\xi={\rm const}$. Then they relevant to 
353: the phase shift of the wave functions due to the shift of the 
354: interface. However, these terms become important when $\xi$ 
355: depends on $\bbox{\rho}$. Interference     of the electrons 
356: scattered from the neighboring planes in the vicinity of  
357: steps (like point 1 on Fig.~1) results in the diffuse  component 
358: of their wave function. The mean size of the region at the step 
359: where the interference occurs is the parallel-to-interface 
360: component of the electron wavelength. Hence the ratio of this 
361: size to the size of the plane area $l$ characterizes  the 
362: roughnesses influence on the electrons.
363: 
364: We separate  the diffuse components  
365: $\varphi_{\Gamma,X_z,X_{xy}}^{l,r}$ of the envelope wave 
366: functions and write the envelopes as follows: 
367: \cite{Bass}
368: \begin{eqnarray} 
369: \label{10} 
370: &&\Psi_{\Gamma,X_z,X_{xy}}^{l,r}=\Phi_{\Gamma,X_z,X_{xy}}^{l,r}
371: +\varphi_{\Gamma,X_z,X_{xy}}^{l,r}, \\
372: &&\mbox{where} 
373: \quad \overline{\varphi_{\Gamma,X_z,X_{xy}}^{l,r}}=0.
374: \nonumber
375: \end{eqnarray}
376: 
377: \widetext
378: Using the boundary conditions~(\ref{bcrough}), 
379: for the envelopes $\Phi_{\Gamma,X_z,X_{xy}}^{l,r}$ and 
380: $\varphi_{\Gamma,X_z,X_{xy}}^{l,r}$ (see Ref.~\cite{PhysicaE} 
381: for the details) we obtain
382: \begin{eqnarray*}
383: %\label{11}
384: \Phi_\Gamma^l({\bf r})=T_\Gamma e^{-ip_\Gamma z},\quad 
385: &\Phi_{X_z,X_{xy}}^l({\bf r})= 
386: T_{X_z,X_{xy}}e^{\gamma_{X_z,X_{xy}}z},\quad
387: &\Phi_{X_z,X_{xy}}^r({\bf 
388: r})=e^{-iqz}+R_{X_z,X_{xy}}e^{iqz},%\nonumber\\
389: \end{eqnarray*}
390: \begin{eqnarray*}
391: \Phi_h^l({\bf r})=e^{ipz}+R_he^{-ipz},\quad%\nonumber\\
392: &&\Phi_h^r({\bf r})=
393: T_he^{-\gamma_hz},%\\
394: \end{eqnarray*}
395: \begin{eqnarray}
396: \label{11}
397: \varphi_{\Gamma}^l({\bf r})=
398: \frac{2q}{(2\pi)^2}
399: \int_{-\infty}^{\infty} A_{\Gamma}^l(\bbox{k_\parallel})
400: \tilde{\xi}(\bbox{k_\parallel})
401: e^{i(\bbox{\scriptstyle k_\parallel 
402: \rho}-k_\Gamma z)}\, 
403: d\bbox{k_\parallel},\nonumber \\ 
404: %
405: \varphi_{X_z,X_{xy}}^l({\bf r})=
406: \frac{2q}{(2\pi)^2}
407: \int_{-\infty}^{\infty} A_{X_z,X_{xy}}^l(\bbox{k_\parallel})
408: \tilde{\xi}(\bbox{k_\parallel})
409: e^{i\bbox{\scriptstyle k_\parallel 
410: \rho}+\ae_{\Gamma,X_z,X_{xy}}z}\, d\bbox{k_\parallel},\nonumber 
411: \\ 
412: \varphi_{X_z,X_{xy}}^r({\bf r})= \frac{2q}{(2\pi)^2} 
413: \int_{-\infty}^{\infty} A_{\Gamma,X_z,X_{xy}}^r(\bbox{k_\parallel})
414: \tilde{\xi}(\bbox{k_\parallel})
415: e^{i(\bbox{\scriptstyle k_\parallel 
416: \rho}+k_{X_z,X_{xy}}z)}\,
417: d\bbox{k_\parallel}, \nonumber\\
418: \varphi_h^l({\bf r})=
419: \frac{2p}{(2\pi)^2}
420: \int_{-\infty}^{\infty} A_h^l(\bbox{k_\parallel})
421: \tilde{\xi}(\bbox{k_\parallel})
422: e^{i(\bbox{\scriptstyle k_\parallel 
423: \rho}-k_hz)}\,
424: d\bbox{k_\parallel}, \nonumber\\
425: \varphi_h^r({\bf r})=
426: \frac{2p}{(2\pi)^2}
427: \int_{-\infty}^{\infty} A_h^r(\bbox{k_\parallel})
428: \tilde{\xi}(\bbox{k_\parallel})
429: e^{i\bbox{\scriptstyle k_\parallel 
430: \rho}-\ae_hz}\,
431: d\bbox{k_\parallel}. %\nonumber
432: \end{eqnarray}
433: Where
434: \begin{eqnarray*}
435: %\label{12}
436: &&k_\Gamma(\bbox{k_\parallel})=
437: \sqrt{2m_\Gamma(\varepsilon_e-E_\Gamma^l)-\bbox{k_\parallel}^2},
438: \quad
439: \ae_{X_z}(\bbox{k_\parallel})=
440: \sqrt{2m_{Xl}^l(E_{X_z}^l-\varepsilon_e)+\bbox{k_\parallel}^2},\\
441: &&\ae_{X_{xy}}(\bbox{k_\parallel})=
442: \sqrt{2m_{Xt}^l(E_{X_{xy}}^l-\varepsilon_e)+\bbox{k_\parallel}^2},
443: \quad
444: k_{X_z}(\bbox{k_\parallel})=
445: \sqrt{2m_{Xl}^r(\varepsilon_e-E_{X_z}^r)-\bbox{k_\parallel}^2},\\
446: &&k_{X_{xy}}(\bbox{k_\parallel})=
447: \sqrt{2m_{Xt}^r(\varepsilon_e-E_{X_{xy}}^r)-\bbox{k_\parallel}^2},
448: \quad
449: k_h(\bbox{k_\parallel})=
450: \sqrt{2m_{h}^l(E_h^l-\varepsilon_h)-\bbox{k_\parallel}^2},\\
451: &&\ae_h(\bbox{k_\parallel})=
452: \sqrt{2m_{h}^r(\varepsilon_h-E_h^r)+\bbox{k_\parallel}^2},
453: \quad
454: {\rm Im}\,k_{X_z,X_{xy},h}\geq 0,\\
455: %
456: &&T_\Gamma=\frac{2iqt^z_{13}\eta(\xi)}{t^z_{44}\gamma_{X_z}},\;\;\;
457: T_{X_z}=-\frac{2iq}{t^z_{44}\gamma_{X_z}},\;\;\;
458: R_{X_z}=-1-\frac{2iq}{t^z_{44}\gamma_{X_z}},
459: \nonumber\\
460: &&T_{X_{xy}}=\frac{2iq}{t^{xy}_{21}+t^{xy}_{22}\gamma_{X_{xy}}},\;\;\;
461: R_{X_{xy}}=-1+\frac{2iq}{t^{xy}_{21}+t^{xy}_{22}\gamma_{X_{xy}}},
462: \\
463: &&T_h=-\frac{2ip}{-t^h_{21}+t^h_{22}\gamma_h},\;\;\;
464: R_h=-1-\frac{2ip}{-t^h_{21}+t^h_{22}\gamma_h},
465: \nonumber\\
466: &&A_\Gamma^l=\frac{it^z_{13}\eta(\xi)}{t^z_{44}},\quad
467: %\nonumber\\
468: A_{X_z}^r=-\frac{i}{t^z_{44}}
469: \left(\frac{t_{13}^zt_{41}^z}{\gamma_{X_z}}+ 
470: 1-t^z_{44}\right),\quad
471: %\nonumber \\
472: A_{X_z}^l=-\frac{k_{X_z}}{\ae_{X_z}{t^{z}_{44}}^2}
473: \left(\frac{t_{13}^zt_{41}^z}{\gamma_{X_z}}+ 
474: 1-t^z_{44}\right),
475: \nonumber \\
476: &&A_{X_{xy}}^r=i
477: \frac{t^{xy}_{22}\ae_l\ae_{X_{xy}}(1-t^{xy}_{22})}
478: {(t^{xy}_{21}+\ae_{X_{xy}}t^{xy}_{22})
479: (t^{xy}_{22}\ae_l+ik_{X_{xy}})},\qquad
480: %\nonumber\\
481: A_{X_{xy}}^l=k_z
482: \frac{\ae_{X_{xy}}(1-t^{xy}_{22})}
483: {(t^{xy}_{21}+\ae_{X_{xy}}t^{xy}_{22})
484: (t^{xy}_{22}\ae_l+ik_{X_{xy}})},
485: \nonumber\\
486: &&A_h^r=
487: -\frac{k_h(1-t^h_{22})}
488: {t^h_{22}(\ae_ht^h_{22}-t^h_{21})},\qquad
489: %\nonumber\\
490: A_h^l=
491: i\frac{\ae_h(1-t^h_{22})}
492: {\ae_ht^h_{22}-t^h_{21}}.
493: \nonumber
494: \end{eqnarray*}
495: \narrowtext
496: Here 
497: $\tilde{\xi}(\bbox{k_\parallel})=\int\xi(\bbox{\rho})
498: e^{-i\bbox{k_\parallel \rho}}\,d\bbox{\rho}$, 
499: $\gamma_{X_z,X_{xy},h}=\ae_{X_z,X_{xy},h}(0)$, 
500: $p_\Gamma=k_\Gamma(0)$; $E_\Gamma^l$, $E_{X_z}^l$, 
501: $E_{X_{xy}}^l$, $E_{X_z}^r$, $E_{X_{xy}}^r$, $E_h^l$, and 
502: $E_h^r$ are energies of extrema of the appropriate bands.  
503: Integration in Eq.~(\ref{11}) is carried out over the whole 
504: plane because $\xi(\bbox{\rho})$ is not periodical function of 
505: $\bbox{\rho}$.  The values of normal-to-interface components of 
506: the wave vectors of the electrons $q$ and holes $p$  are 
507: determined by the boundary conditions at the interfaces:  
508: $z=-d_1$ for $p$ and $z=d_2$ for $q$ (where $d_1$ and $d_2$ are 
509: widths of GaAs and AlAs layers).  In general, they depends on 
510: the valley under consideration:  $\tan 
511: \frac{qd_2}{2}=-\frac{q}{\gamma_{X_z}}$ for  electrons in the 
512: $X_z$ valley, $\tan 
513: qd_2=-\frac{2q}{t^{xy}_{22}\gamma_{X_{xy}}+t^{xy}_{21}}$ for  
514: electrons in the $X_{xy}$ valley, and $\tan 
515: pd_1=-\frac{2p}{t^h_{22}\gamma_h-t^h_{21}}$ for the holes.  We 
516: assume, however, the  strong confinement of electrons and holes  
517: in the appropriate layers $\gamma_{X_z,X_{xy},h} \gg p,q$, so 
518: that $p\approx \pi/d_1$ and $q\approx \pi/d_2$.
519: 
520: The wave function of the electron in the GaAs $\Gamma$ valley is 
521: small, $p_\Gamma\ll p$; nevertheless, it is real. This 
522: distinguish the short-period GaAs/AlAs superlattices from other 
523: type-II structures, where the electron wave function decays 
524: rapidly from the interface. The electron density is large in 
525: AlAs and small, but almost constant, in GaAs. This small part of 
526: the electron density could be essential for the exiton 
527: recombination would the effective parameter of $\Gamma$--$X$ 
528: mixing $t_{13}^z$ be sufficently large.
529: 
530: 
531: \subsection{Radiative decay rates of indirect excitons at a rough 
532: interface}
533: 
534: To determine the wave functions $f_e(z)$ and $f_h(z)$, we have 
535: to insert the corresponding Bloch amplitudes into expressions 
536: for the envelopes $\Psi_e$ and $\Psi_h$ (\ref{10}). For 
537: instance, for the $X_z$ exciton at a plane interface, we have 
538: 
539: \begin{eqnarray}
540: \label{13}
541: &&f_e({\bf r})=\frac{1}{\sqrt{N_1}}\left\{
542: \begin{array}{ll}
543: T_\Gamma u_\Gamma({\bf r})e^{\gamma_\Gamma z}\\
544: +T_{X_z}u_{X}({\bf 
545: r})e^{\left(\gamma_{X_z}-\frac{2\pi i}{a}\right)z},& z<0,\\ \\ 
546: u_{X}^*({\bf r})e^{i\left(q-\frac{2\pi i}{a}\right)z} \\
547: +R_{X_z}u_{X}({\bf r})e^{-i\left(q-\frac{2\pi i}{a}\right)z},& 
548: z>0,\\ 
549: \end{array} 
550: \right.\nonumber\\ 
551: &&\ \\ 
552: &&f_h({\bf 
553: r})=\frac{1}{\sqrt{N_2}}\left\{ 
554: \begin{array}{ll} v({\bf r})e^{ipz}+R_hv^*({\bf 
555: r})e^{-ipz},& z<0,\\ T_hv({\bf r})e^{-\gamma_hz}, &z>0.  
556: \end{array}
557: \right. \nonumber
558: \end{eqnarray}
559: Where $N_1$ and $N_2$ are numbers of atoms in the AlAs 
560: and GaAs layers, $u_{\Gamma}({\bf r})$, $u_{X}({\bf r})$ and 
561: $v({\bf r})$ are Bloch amplitudes of electrons in the $\Gamma$ 
562: and $X$ valleys, and the holes; we assume these amplitudes to be 
563: periodical functions of ${\bf r}$.
564: 
565: At a rough interface we have to add also the diffuse components 
566: of the wave functions. To do that, we have to multiply  
567: $\varphi({\bf r})$ (\ref{11}) by the corresponding Bloch 
568: amplitudes. Usually the mean size of  Bloch amplitudes is 
569: small in comparison with the lattice constant. This allows us 
570: to assume that $\bbox{\nabla}$-operator in Eq.~(\ref{2}) acts 
571: only on these amplitudes and to separate the integration of them 
572: from integration of the envelopes. Then the matrix element 
573: (\ref{2}) can be written as ${\cal P}={\cal P}_1+{\cal 
574: P}_2+{\cal P}_3$, where 
575: 
576: \begin{eqnarray} 
577: \label{14} 
578: &&{\cal P}_1=\sum_{\Gamma,X}U_{\Gamma,X}\int 
579: \Phi_{\Gamma,X_z,X_{xy}} \Phi_h\,dz\,d\bbox{\rho}, \nonumber \\ 
580: &&{\cal P}_2=\sum_{\Gamma,X}U_{\Gamma,X}
581: \int [\Phi_{\Gamma,X_z,X_{xy}} 
582: \varphi_h+ \Phi_h \varphi_{\Gamma,X_z,X_{xy}}] 
583: \,dz\,d\bbox{\rho},  \\ 
584: &&{\cal P}_3=\sum_{\Gamma,X}U_{\Gamma,X}
585: \int \varphi_{\Gamma,X_z,X_{xy}} \varphi_h
586: \,dz\,d\bbox{\rho}.  \nonumber 
587: \end{eqnarray}
588: Here $\Phi=\Phi^l$, $\varphi=\varphi^l$ if $z<\xi$; 
589: $\Phi=\Phi^r$, $\varphi=\varphi^r$ if $z>\xi$;
590: $U_\Gamma=\Omega_0^{-1}\int_{\Omega_0} u_\Gamma({\bf r}) 
591: \bbox{\nabla} v({\bf r})\,d{\bf r}$,
592: $U_X= \Omega_0^{-1}\int_{\Omega_0}u_{X}({\bf r})
593: \bbox{\nabla} v({\bf r})\,d{\bf r}$,
594: and $\Omega_0$ is the unit cell.
595: 
596: The rate of the exciton recombination is  
597: \begin{eqnarray}
598: \label{w}
599: &&w=\Lambda\left(
600: |{\cal P}_1|^2+{\cal P}_1{\cal P}_2^*+{\cal P}_1^*{\cal P}_2
601: +{\cal P}_1{\cal P}_3^*+{\cal P}_1^*{\cal P}_3
602: +|{\cal P}_2|^2\right), \nonumber\\
603: &&\Lambda=\frac{4\hbar e^2\omega}{3m_e^2c^3}G^2({\bf 0}).
604: \end{eqnarray}
605: Where $\hbar\omega$ is the exciton energy, $e$, $m_e$ and $c$ are 
606: the fundamental constants. 
607: 
608: The luminescence magnitude $I(t)$ is 
609: proportional to the recombination rate $w$ and the number of 
610: excitons at the time $t$. We assume this number to be 
611: proportional to $\exp(-wt)$ (or $\exp[-(w_0+w)t]$, if some 
612: nonstochastic process with the rate $w_0$ occurs).  The $w$ 
613: value is stochastical, since it depends on $\xi$.  Therefore, to 
614: determine the luminescence magnitude, we have to average the 
615: value of $w\exp(-wt)$ over the realization of the random 
616: function $\xi$.  This could be done if we know the distribution 
617: $P(w)$ of the $w$ value:  $\overline{w\exp(-wt)}=\int_0^\infty 
618: w\exp(-wt)P(w)\,dw$.  The distribution $P(w)$ essentially 
619: depends on ${\cal P}_1$, whether or not it vanishes. 
620: 
621: If ${\cal P}_1=0$ (i.e., if 
622: the exciton recombination at a plane interface is forbidden) then 
623: $w$ is proportional to  squared module of 
624: ${\cal P}_2$. The linear dependence between ${\cal P}_2$ and the 
625: random variable $\xi$ follows from Eqs.~(\ref{11}) and 
626:  (\ref{14}).  Therefore, if the distribution of $\xi$ is 
627:  Gaussian, then the distribution of ${\cal P}_2$ is also 
628: Gaussian and the distribution of $w$ is exponential.  This means 
629: applicability of arguments of Refs.~\cite{Klein,Minami}, so that 
630: $I(t)$ is determined by Eq.~(\ref{1}) where $w_r=\Lambda|{\cal 
631: P}_2|^2$.  For the case of $X_{xy}$ exciton we have
632: \begin{eqnarray*}
633: {\cal P}_2&=&\frac{4pqU_X}{\sqrt{N_1N_2}}
634: \sum_{\bf g}\left[
635: \frac{1-t_{22}^h}{\gamma_{X_{xy}}^2} 
636: \tilde{\xi}\left({\bf q}_X+{\bf g}\right)
637: \right.\\
638: &+&\left.\frac{1-t_{22}^{xy}}{\gamma_h^2}
639: \tilde{\xi}^*\left({\bf q}_X+{\bf g}\right)
640: \right],
641: \end{eqnarray*}
642: and
643: \begin{equation}
644: \label{15}
645: w_r=\frac{16\pi^4a^4|U_X|^2\Lambda}{d_1^3d_2^3}
646: \left[
647: \frac{(1-t_{22}^h)^2}{\gamma_{X_{xy}}^4}+\frac{(1-t_{22}^{xy})^2}{\gamma_h^4}
648: \right]
649: \tilde{W}\left(\frac{\pi}{a}\right).
650: \end{equation}
651: Where $\tilde{W}({\bf k})=
652: \int W(\bbox{\rho})e^{-ik\rho}\,d^2\bbox{\rho}$
653: is the Fourier transform of
654: the correlation function; ${\bf q}_X=\left\{2\pi/a,0,0\right\}$ 
655: is the wave number of the $X$ valley, ${\bf g}$ is a 
656: two-dimensional reciprocal lattice vector, it arises here since 
657: integration in Eq.~(\ref{11}) has not been restricted by the 
658: first Brillouin zone. 
659: 
660: If ${\cal P}_1\neq 0$ (i.e., if 
661: the exciton recombination at a plane interface is allowed) then
662: the linear  with respect to ${\cal P}_2$ terms in Eq.~(\ref{w}) 
663: are nonzero.  This allows to omit the terms with  ${\cal P}_3$ 
664: and $|{\cal P}_2|^2$, which are quadratic in $\xi$, or replace 
665: them with their average values.  Then $w$ becomes the linear 
666: function of the random variable $\xi$.  If the distribution of 
667: $\xi$ is Gaussian then the distribution of $w$ is Gaussian too, 
668: i.e., 
669: \[ 
670: P(w)=\frac{1}{\sigma\sqrt{2\pi}}e^{-\frac{(w-\overline{w})^2}{2\sigma^2}}.
671: \] 
672: Where $\overline{w}=\Lambda|{\cal P}_1|^2$, and 
673: $\sigma=\left[\overline{|w-\overline{w}|^2}\right]^{1/2}$.
674: 
675: 
676: 
677: Hence
678: \begin{eqnarray}
679: \label{16}
680: I(t)&=&\frac{e^{-w_0t}}{\sigma\sqrt{2\pi}}
681: \int_0^\infty we^{-wt-\frac{(w-\overline{w})^2}{2\sigma^2}}\,dw
682: \\
683: &=&e^{-(\overline{w}+w_0)t}
684: \left[
685: \frac{\sigma}{\sqrt{2\pi}}+\frac{\overline{w}-\sigma^2t}{2}
686: e^{\frac{\sigma^2t^2}{2}}\mbox{erfc}\left(\frac{\sigma 
687: t}{\sqrt{2}}
688: \right) 
689: \right]. 
690:    \nonumber
691: \end{eqnarray}
692: 
693: If $|{\cal P}_1|\gg |{\cal P}_2|$, then
694: $\sigma^2\simeq
695: 2\Lambda^2
696: |{\cal P}_1|^2 \overline{{\cal P}_2{\cal P}_2^*}
697: $. For the case of $X_z$ exciton we have
698: \begin{eqnarray}
699: \label{16.5}
700: &&{\cal P}_1=
701: \frac{2a^3}{\sqrt{d_1d_2}}
702: \left[
703: 4\left(\frac{d_1}{d_2}\right)
704: \frac{t_{13}^z}{\gamma_{X_z}t_{44}^z}U_\Gamma+
705: \left(\frac{a}{d_1}\right)
706: \frac{1}{\gamma_{h}t_{22}^h}U_X
707: \right],\\
708: &&{\cal P}_2=
709: \frac{2a^2}{\sqrt{d_1d_2}}
710: \left[
711: 8\left(\frac{d_1}{d_2}\right)
712: \frac{{t_{13}^z}}{{t_{44}^z}}U_\Gamma+\right.\nonumber\\
713: &&\phantom{{\cal P}_2=}\left.
714: \frac{2\pi i a}{t_{22}^hd_1}
715: \left(
716: \frac{1}{t_{44}^z\gamma_{X_z}d_2}
717: -\frac{1-t_{22}^h}{t_{22}^h\gamma_hd_1}
718: \right)U_X
719: \right]
720: \eta(\xi)\tilde{\xi}(0),\nonumber
721: \end{eqnarray}
722: so that
723: \begin{eqnarray}
724: \label{17}
725: &&\overline{w}=
726: \frac{4\Lambda a^6}{d_1d_2}
727: \left[
728: 4\left(\frac{d_1}{d_2}\right)
729: \frac{t_{13}^z}{\gamma_{X_z}t_{44}^z}U_\Gamma+
730: \left(\frac{a}{d_1}\right)
731: \frac{1}{\gamma_{h}t_{22}^h}U_X
732: \right]^2,\\
733: &&\sigma^2=
734: \frac{2\Lambda a^4\overline{w}}{d_1d_2}
735: \left[
736: 64\left(\frac{d_1}{d_2}\right)^2
737: \frac{{t_{13}^z}^2}{{t_{44}^z}^2}U_\Gamma^2+
738: \right.
739: \nonumber\\
740: &&\left.\phantom{\sigma^2=}
741: \left(\frac{2\pi a}{t_{22}^hd_1}\right)^2
742: \left(
743: \frac{1}{t_{44}^z\gamma_{X_z}d_2}
744: -\frac{1-t_{22}^h}{t_{22}^h\gamma_hd_1}
745: \right)^2U_X^2
746: \right]
747: \tilde{W}(0).\nonumber
748: \end{eqnarray}
749: The first terms in square brackets can be interpreted as 
750: a electron conversion from the $X$ valley of AlAs to 
751: the $\Gamma$ valley of GaAs followed by the elecron-hole 
752: recombination; they are small, since $t_{13}^z\ll 1$.  The 
753: second ones are due to indirect electron-hole 
754: recombination;\cite{PR98} they occur only at the interface and, 
755: therefore, have a small factor $a/d_{1,2}$. This factor is not 
756: so small in short-period superlattices where $d_{1,2}$ are as 
757: large as a few lattice constants. The indirect electron-hole 
758: recombination prevails in such structures, if $a/d_{1,2}\gg 
759: t_{13}^z$. We omit the terms that contain both these factors or 
760: $|1-t_{44}^z|\ll 1$.
761: 
762: The question arises, how small should be  $|{\cal P}_1|$ in 
763: order to Eq.~(\ref{1}) holds? This is possible if the   
764: deviation of $|{\cal P}_2|^2$  from its average value in 
765: Eq.~(\ref{w}) essentially exceeds $|{\cal P}_1{\cal P}_2^*|$, 
766: i.e., when  $|{\cal P}_1|^2\ll (h/l)^2|{\cal P}_2|^2$ or
767: \begin{equation}
768: \label{criterium}
769: \frac{\overline{w}^2}{\sigma^2}\ll
770: \frac{h^2}{l^2}.
771: \end{equation}
772: 
773: 
774: \section{Kinetics of exciton luminescence in type-II 
775: G\symbol{"61}A\symbol{"73}/A\symbol{"6C}A\symbol{"73}
776: superlattices.  Experiment}
777: 
778: The undoped GaAs/AlAs type-II superlattices used in this study 
779: were grown by molecular-beam-epitaxy  at $600^o$C on a 
780: (100)  GaAs substrate. The sample BP205, where the 
781: $X_z$ excitons were studied, contains 40 periods of 19.8-$\AA$ 
782: GaAs/25.5-$\AA$ AlAs. The $X_{xy}$ excitons were studied in the 
783: sample BP354. It contains 25 periods of 25-$\AA$  
784: GaAs/83.5-$\AA$ AlAs.   
785: 
786: The time-resolved  photoluminescence of $X_z$ excitons was 
787: excited by a YAG:Nd pulse laser with  wavelength $532\, nm$, the 
788: pulse duration was $0.15\,\mu s$. The N$_2$ laser with 
789: wavelength $337\, nm$ and pulse duration $7\,ns$ was used to 
790: investigate the time-resolved photoluminescence of $X_{xy}$ 
791: excitons.
792: 
793: The luminescence was analyzed by a double grating 
794: monochromator equipped with a photomultiplier. Lifetime 
795: measurements were made by the time correlated single-photon 
796: counting technique. The samples were immersed in liquid helium. 
797: 
798: Figures 2 and 3 present the experimental results on the exciton 
799: decay rates in our samples. 
800: Theoretical curves was derived from Eqs.~(\ref{1}), (\ref{16}).
801: The  values of parameters $w_0=320\,c^{-1}$, $w_r=0.002\times 
802: 10^6\,c^{-1}$, $\overline{w}=0.1\times 10^6\,c^{-1}$,  and 
803: $\sigma=0.61\times 10^6\,c^{-1}$ ensure the best fit with the 
804: experiment.  We see that Eq.~(\ref{1}) fits the experimental 
805: data for the decay rates of  $X_{xy}$ excitons in the sample 
806: BP354,  whereas Eq.~(\ref{16}) is more appropriate for $X_{z}$ 
807: excitons in the sample BP205. Note that the value of $w_0$, 
808: which is associated with  the phonon-assisted recombination, is 
809: small in both curves.  That is really the case at a low 
810: temperature.  Recombination of $X_{xy}$ excitons is considerably 
811: slower than that of $X_z$ excitons. This means that the 
812: interfaces in our samples are perfect enough to apply our theory 
813: for interpretation of the experimental data.
814: 
815: Expressions (\ref{15}) -- (\ref{17}) allow to estimate the 
816: function $\tilde{W}(k)$ at the points $k=0$ and $k=2\pi/a$.  
817: This is not sufficient to determine the function. However, it is 
818: possible to estimate the parameters of the rough interface if we 
819: restrict ourself to the particular type of the correlation 
820: function. We  assume the exponential correlation function
821: \begin{equation}
822: \label{18}
823: W(\rho)=h^2\exp(-\rho/l),
824: \end{equation}
825: where $l$ 
826: is the correlation length. This type of the correlation 
827: function is more appropriate to our model of the rough interface 
828: (Fig.~1); it allows to construct the two-position distribution, 
829: so that the distribution of slopes has a $\delta$-singularity, 
830: i.e., the slope is always zero exept a set of points (like point 
831: 1) with measure zero.\cite{Berry}  
832: This is impossible for the Gaussian correlation function 
833: $W(\rho)=h^2\exp(-\rho^2/l^2)$ most frequently employed in 
834: theoretical discussions.\cite{Kosobukin}   Fourier 
835: transform of the exponential function    is
836: \begin{equation} 
837: \label{18a} 
838: \tilde{W}(k)= \frac{2\pi h^2l^2}{(1+k^2l^2)^{3/2}}. 
839: \end{equation} 
840: Unlike the Gaussian function it has not exponential factor, 
841: which is small at a large $k$.    This is also due to the 
842: singular points 1; only in the vicinity of these points the 
843: momentum relaxation of indirect $X_{xy}$ excitons is possible. 
844: 
845: If we assume that correlation functions are equal for the 
846: interfaces of both our samples, then substitution of 
847: Eq.~(\ref{18a}) into Eqs.~(\ref{15}) and (\ref{17}) allows to 
848: find the values of $h$ and $l$.  
849: The decay parameters of the wave functions  
850: $\gamma_{X_z,X_{xy},h}$ in these 
851: expressions are determined by Eq.~(\ref{11}) for the known 
852: energies of the electrons or holes.  As regards to 
853: $U_{\Gamma,X}$, these values can be estimated only from the 
854: band structure calculations for GaAs and AlAs. However,  the 
855: first terms in the expression for $\overline{w}$ and $\sigma^2$ 
856: (\ref{17}) can be omitted.  Indeed, $t^z_{13}=t_{\Gamma 
857: X}m^{\rm GaAs}_\Gamma/(m_ea\gamma^r)<0.06$, whereas 
858: $a/d_{1}=2/7$, i.e., the indirect recombination of 
859: $X_z$ exitons at the interface prevails in  our samples.   Then 
860: the values of ${\sigma^2}/{\overline{w}^2}$ and 
861: $w_r/{\overline{w}}$, which are determined from experimental 
862: data, become independent of $U_{\Gamma,X}$.  For our experiments 
863: this estimation yields 
864: \widetext 
865: \begin{eqnarray} 
866: \label{19} 
867: &&\frac{\sigma^2}{\overline{w}^2}=
868: \frac{1}{2}\left(\frac{2\pi}{a}\right)^2
869: \left(
870: \frac{\gamma_h}{\gamma_{X_z}d_2}-
871: \frac{1-t^h_{22}}{t^h_{22}d_1}
872: \right)^2\tilde{W}(0),
873: \nonumber \\
874: &&\frac{w_r}{\overline{w}}=
875: \frac{4\pi^4\gamma_h^2{t^h_{22}}^2
876: {d_1^{X_{xy}}}^3d_2^{X_{xy}}}
877: {a^4{d_1^{X_z}}^3{d_2^{X_z}}^3}
878: \left[
879: \frac{(1-t_{22}^h)^2}{\tilde{\gamma}_{X_{xy}}^4}+
880: \frac{(1-t_{22}^{xy})^2}{\tilde{\gamma}_h^4}
881: \right]
882: \tilde{W}\left(\frac{2\pi}{a}\right).
883: \end{eqnarray}
884: Here $d_1^{X_z}$, $d_2^{X_z}$, $d_1^{X_{xy}}$, and 
885: $d_2^{X_{xy}}$ are widths of GaAs and AlAs layers in the samples 
886: BP205 ($d_1^{X_z}$, $d_2^{X_z}$) and BP354 ($d_1^{X_{xy}}$,  
887: $d_2^{X_{xy}}$) where $X_z$ and $X_{xy}$ excitons were studied;
888: \begin{eqnarray*}
889: &&\gamma_{X_z}=\frac{1}{\hbar}
890: \sqrt{2m_{Xl}^{GaAs}
891: \left[E_X^{GaAs}-E_X^{AlAs}-\frac{\hbar^2}
892: {2m_{Xl}^{AlAs}(d_2^{X_z})^2} \right]},\\
893: &&\gamma_h=\frac{1}{\hbar}
894: \sqrt{2m_{hh}^{AlAs}
895: \left[E_h^{GaAs}-E_h^{AlAs}-\frac{\hbar^2}
896: {2m_{hh}^{GaAs}(d_1^{X_z})^2} \right]},
897: \quad
898: t_{22}^h=\frac{m_{hh}^{AlAs}}{m_{hh}^{GaAs}},\\
899: &&\tilde{\gamma}_{X_{xy}}=\frac{1}{\hbar}
900: \sqrt{2m_{Xt}^{GaAs}
901: \left[E_X^{GaAs}-E_X^{AlAs}-\frac{\hbar^2}
902: {2m_{Xt}^{AlAs}(d_2^{X_{xy}})^2} \right]},
903: \quad
904: t_{22}^{xy}=\frac{m_{Xt}^{AlAs}}{m_{Xt}^{GaAs}},
905: \\
906: &&\tilde{\gamma}_h=\frac{1}{\hbar}
907: \sqrt{2m_{hh}^{AlAs}
908: \left[E_h^{GaAs}-E_h^{AlAs}-\frac{\hbar^2}
909: {2m_{hh}^{GaAs}(d_1^{X_{xy}})^2} \right]},
910: \quad
911: \gamma^r=\frac{1}{\hbar}
912: \sqrt{2m_{\Gamma}^{AlAs}
913: \left(E_\Gamma^{AlAs}-E_X^{AlAs}\right)};
914: \end{eqnarray*}                             
915: \narrowtext
916: $E_{\Gamma,X,h}^{GaAs,AlAs}$ are positions of the band extrema 
917: in GaAs and AlAs, $m_{\Gamma}^{GaAs}$ is  effective mass of 
918: $\Gamma$ valley of GaAs, $m_{Xl,Xt}^{GaAs,AlAs}$ are longitudial 
919: and transversal effective masses in $X$ valleys of GaAs and 
920: AlAs, $m_{hh}^{GaAs,AlAs}$ are effective masses of heavy holes 
921: in GaAs and AlAs, and $m_e$ is mass of a free electron. We 
922: assume $t_{21}^h\ll\gamma_h\sim 2/a$; this is the result of 
923: calculations. \cite{Ando} 
924: 
925: Eq.~(\ref{19}) estimates the values of 
926:  For 
927: the  height $h$ and diameter $L$ [$L=4l$ for the distribution 
928: (\ref{18})]\cite{Berry} of the roughnesses we find $h\approx 
929: 1.25a$ and $L\approx 9a$.  This is in agreement with structural 
930: reseach of the GaAs/AlAs interface where the steps with the 
931: height $h=a/2$ and the mean length of 40--200$\,\AA$ were 
932: observed (see Ref.~\cite{Bechstedt} for the review). 
933: 
934: Rough estimation of $h$ and $l$ values also can be done if we 
935: assume that criterium (\ref{criterium}) holds. This justifies  
936: Eq.~(\ref{1}) for $X_z$ excitons where $w_0\equiv 
937: w_0^{X_z}=\Lambda|{\cal P}_1|^2$ and $w_r\equiv 
938: w_r^{X_z}=\Lambda|{\cal P}_2|^2$. Using Eq.~(\ref{16.5}), we 
939: find the expressions for $2w_r^{X_z}/w_0^{X_z}$ and 
940: $w_r/w_0^{X_z}$ [unlike $w_r^{X_z}$ the $w_r$ value  
941: correspondes to $X_{xy}$ excitons (Fig.~3) and determined by 
942: Eq.~(\ref{15})]. These expressions accept the form of 
943: Eq.~(\ref{19}) after the substitutions 
944: $\sigma^2/\overline{w}^2\rightarrow 2w_r^{X_z}/w_0^{X_z}$ and 
945: $w_r/\overline{w}\rightarrow w_r/w_0^{X_z}$ of their left sides. 
946: The values of $w_0^{X_z}=0.11\times 10^6\,c^{-1}$ and 
947: $w_r^{X_z}=0.38\times 10^6\,c^{-1}$ ensure the best fit of the 
948: dashed line (Fig.~2) with experiment. This estimation yields 
949: $h=a$, $L=8.8a$, which are close to the values obtained from 
950: Eq.~(\ref{16}). For this reason both theoretical curves (Fig.~2) 
951: fit experimental data at small times.  Nevertheless, 
952: Eq.~(\ref{16}) better fits the experimental data at large times 
953: where it unsure the slower decay of the luminescence:  
954: $I(t)\propto \exp (-\overline{w}t)/t$, instead of $I(t)\propto 
955: \exp (-w_0t)/t^{3/2}$ as it is predicted by Eq.~(\ref{1}). 
956: 
957: 
958: 
959: 
960: 
961: \section{Discussion}
962: 
963: In this paper we investigate the exciton luminescence in type II 
964: GaAs/AlAs superlattices. We use the envelope function 
965: approximation to consider the exciton recombination at an 
966: interface. To justify this approach, we have to note that 
967: envelope function approximation has been used only to find the 
968: reflection and transmission coefficients.  While the Bloch 
969: functions $f_e$ and $f_h$ has been used to find the probability 
970: of the exciton recombination.  The error arises only when we 
971: consider the Bloch amplidudes $u_\Gamma({\bf r})$, $u_X({\bf 
972: r})$, and $v({\bf r})$ as periodical functions at the interface.  
973: Indeed, the deviation of these amplitudes from their bulk values 
974: arises only at a small distance from the interface; this 
975: deviation is especially small for the contacts of similar 
976: materials (e.g., GaAs/AlAs).\cite{Ando,Allmen} 
977: 
978: It seems the boundary conditions  (\ref{bcplane}) connect a 
979: very few valleys of the electron spectrum to consider the 
980: interface influence on the exciton recombination; nevertheless, 
981: it is not the case. Indeed,  the electron wave functions in the 
982: valleys that are not explicitly involved in Eq.~(\ref{bcplane}) are 
983: strongly localized at the interface. This allows to consider 
984: them in terms of the boundary conditions where the parameters 
985: $t_{ik}$ are influenced by these valleys.  This procedure had 
986: been described  when Eq.~(\ref{bcplane}) was derived. The error arises 
987: only when these parameters are considered as independent of the 
988: electron energy; that is possible if the energy difference 
989: between the bottoms of the appropriate valleys considerably 
990: exceeds the exciton energy.  Note that a lot (about 10) of the 
991: electron bands are sometimes taken into account when the 
992: parameters of the interface matrix  
993: are calculated.\cite{Grinyaev}
994: 
995: 
996: We use the boundary conditions for the envelope wave function 
997: to consider $\Gamma-X$ mixing of  electrons at 
998: the interface.  This approach is more general than the kinetic 
999: model proposed in Ref.~\cite{Maaref}. The kinetic equation where 
1000: the electron states in the $\Gamma$ and $X$ valleys are 
1001: considered as independent can be used  for low $\Gamma-X$ 
1002: mixing.  Only in that case it is possible to add the 
1003: probabilities for the electron to be in $\Gamma$ and $X$ 
1004: valleys.  It should be noted that we also assume the small value 
1005: of $\Gamma-X$ mixing ($|t_{13}^z|\ll 1)$. However, this 
1006: approximation is not principal for our consideration; it only 
1007: makes the results [Eqs.~(\ref{11}), (\ref{15}), (\ref{17}), and 
1008: (\ref{19})] not so cumbersome.
1009: 
1010: Influence of a nonstochastic process on the exciton recombination 
1011: in Ref.~\cite{Klein} is considered by the exponential factor 
1012: $e^{-w_0t}$. This factor could be obtained if we 
1013: insert the corresponding term in $\tau$-approximation  
1014: into  the kinetic equation for the exciton density.    If the 
1015: $\tau$-approximation is not applicable for the process, then 
1016: this factor becomes nonexponential.\cite{Krivirotov} Correlation 
1017: between  stochastic and nonstochastic processes changes  the 
1018: second factor in Eq.~(\ref{1}).  In this case the probability of 
1019: the exciton recombination $w$ is not a simple sum of  the 
1020: probabilities of each process.  As the result, the additional 
1021: terms arise  in the expression for $w$ [the second and  third
1022: terms in Eq.~(\ref{w})]. These terms are linear in  the 
1023:  stochastic variable, so that their averages vanish. For this 
1024:  reason they are not important when the mean intensity of the 
1025:  luminescence or the light absorption\cite{PhysicaE} is 
1026: considered. However, they are important for the kinetic 
1027: phenomena, because they determine the mean square of the 
1028: deviation $\sigma$  of the stochastic variable from its mean 
1029: value. The nonexponential behavior of the decay rate 
1030: Eq.~(\ref{16}) valids any time when  linear with respect to 
1031: the stochastic variable terms are main in 
1032: the expression for $w$.   Such a situation can occur also in 
1033: other type-II semiconductor structures where the interface 
1034: influence is essential, e.g., in quantum dots.\cite{Govorov}
1035: 
1036: Expressions (\ref{15}) and (\ref{17}) relate parameters of the 
1037: radiative decay rates ($w_r, \overline{w}$, and $\sigma$) with 
1038: the correlation function of the rough interface. The values of 
1039: the Fourier transform of this function at two particular points, 
1040: $k=0$ and $k=2\pi/a$, are necessary for these relations. This 
1041: allows to estimate the parameters only for  simplest 
1042: functions [like Eq.~\ref{18}]. The real interface might be more 
1043: complicated. In particular,  a few different 
1044: scales could be characteristic for the roughnesses at the interface. 
1045: Expressions (\ref{15}) and (\ref{17}) takes into account all 
1046: these factors; however, it is impossible to determine more than 
1047: two parameters  from the time-resolved luminescence experiments 
1048: only.
1049: 
1050: Comparing the experimental results (Figs. 2 and 3), we see  that 
1051: mean lifetime of $X_{xy}$ excitons essentially exceeds that of 
1052: $X_z$ excitons.  This happens due to  
1053: recombination of  $X_z$ excitons  at a plane interface. 
1054: Meanwhile, influence of the roughnesses, i.e., the 
1055: nonexponential factor in $I(t)$, is more essential for $X_z$ 
1056: excitons. This can be understood from our analysis. Indeed, 
1057: $\sigma\propto \tilde{W}(0)$, whereas $w_r\propto 
1058: \tilde{W}(2\pi/a)$ while $\tilde{W}(2\pi/a)\ll \tilde{W}(0)$. 
1059: The recombination occurs in some region near the step (point 1 
1060: in Fig.~1). The size of this region is of the order of 
1061: $|{\bf q}_\parallel|^{-1}$, where ${\bf q}_\parallel$ is the 
1062: parallel-to-interface component of the electron wave vector.  
1063: This region is large for  $X_z$ electrons  ($|{\bf 
1064: q}_\parallel|\simeq r_B^{-1}$, where $r_B$ is the exciton 
1065: radius) but it is small for $X_{xy}$ electrons ($|{\bf 
1066: q}_\parallel|\simeq 2\pi/a$).  As the result, the small factor 
1067: [of the order of $(a/l)^3$] arises in the expression for $w_r$.
1068: 
1069: In conclusion,   kinetics of the exciton luminescence 
1070: at a rough interface has been considered. The Klein {\it at al.} 
1071: law (\ref{1}) is shown to be valid for the decay rate of 
1072: $X_{xy}$ excitons, whereas the more complicated expression 
1073: (\ref{16}) is applicable for $X_z$ excitons. Expressions  
1074: (\ref{15}) and (\ref{17}), which relate the parameters of the 
1075: exciton kinetics with statistical characteristics of the rough 
1076: interface, allow to estimate some of these characteristics from 
1077: the experimenal data.  The values of the mean height  $7\,\AA$
1078: and length $50\,\AA$ of the roughnesses obtained from 
1079: our experiments are in a good agreement with the results of 
1080: structural investigations of the GaAs/AlAs interface.
1081: 
1082:  \acknowledgments
1083: Authors  wish to thank  
1084: Prof.\ E.\ L.\ Ivchenko  for valuable discussions.   This work 
1085: was supported by the Russian Foundation for the Basic Research, 
1086: Grants No.~99-02-17019, 98-02-17896, 00-02-17658, and the 
1087: Program ''Physics of Solid State Nanostructures'' by the Russian 
1088: Interdisciplinary Scintific and Technical Council, Grant 
1089: No.~99-1133.
1090: 
1091: 
1092: 
1093: \begin{references}
1094: \bibitem{Klein}M.~V.~Klein, M.~D.~Sturge, and E.~Cohen, Phys. 
1095: Rev. B, {\bf 25}, 4331 (1982).
1096: 
1097: \bibitem{Minami}F.~Minami {\it et al.}, Phys. Rev. B, {\bf 36}, 
1098: 2875 (1987).
1099: 
1100: \bibitem{Krivirotov}I.~N.~Krivorotov {\it et al.}, Phys. Rev. B, 
1101: {\bf 58}, 10687 (1998).
1102: 
1103: \bibitem{Nagao}S.~Nagao {\it et al.}, J. of Crystal Growth, 
1104: {\bf 175}, 10687 (1997);
1105: B.~A.~Wilson {\it et al.}, Phys.  Rev.  B, {\bf 40}, 1825   
1106: (1989);
1107: E.~Finkman {\it et al.}, J.~Lumin. {\bf 39}, 57 (1987);
1108: B.~A.~Wilson {\it et al.}, J. Vac. Sci. Technol. B {\bf 6}, 1156 
1109: (1988).  
1110: 
1111: 
1112: 
1113: \bibitem{Ivchenko}E.\ L.\ Ivchenko and G.\ E.\ Pikus, {\it 
1114: Superlattices and Other Heterostructures. Symmetry and Optical 
1115: Phenomena}, second ed. (Springer-Verlag, Berlin, 1997).
1116: 
1117: 
1118: 
1119: \bibitem{PhysicaE}L.~Braginsky, Physica E, {\bf 5}, 142 (1999).
1120: 
1121: \bibitem{Ando}T.~Ando and H.~Akera, 
1122: Phys.  Rev.  B, {\bf 40}, 11619   (1989). 
1123: 
1124: \bibitem{Lurssen}D.~L\"uer\ss en {\it et al.},  Phys.  Rev.  B, {\bf 59}, 
1125: 15862   (1999). 
1126: 
1127: \bibitem{Bechstedt}F.~Bechstedt, R.~Enderlein, {\it 
1128: Semiconductor Surfaces and Interfaces. Their Atomic and 
1129: Electronic Structures},  (Berlin: Akademie-Verlag,  1988).
1130: 
1131: \bibitem{Ivchenko1}I.~L.~Aleiner and E.~L.~Ivchenko, Fiz. Tech. 
1132: Poluprovodn. {\bf 27} 594 (1993) [Sov. Phys. Semicond. {\bf 27} 
1133: (1993)]; Y.~Fu, M.~Willander, E.~L.~Ivchenko, and A.~A.~Kiselev, 
1134: Phys. Rev. B, {\bf 47} 13498 (1993). 
1135: 
1136: 
1137: \bibitem{Bass}F.\ G.\ Bass and I.\ M.\ Fuks, {\it Wave 
1138: Scattering from Statistically Rough Surfaces}, (Pergamon Press, 
1139: 1979).
1140: 
1141: \bibitem{PR98}L.~Braginsky, Phys.  Rev. B, {\bf 57}, R6870 (1998).
1142: 
1143: 
1144: \bibitem{Berry}M.~V.~Berry, Phil. Trans. {\bf A273}, 611 (1973);
1145: J.~M.~Ziman, {\it Models of Disorder. The theoretical physics of 
1146: homogeneously disordered systems}, (Cambrige University Press, 
1147: 1979).
1148: 
1149: \bibitem{Kosobukin}For the exciton problem this approximation 
1150: has been used in V.~A.~Kosobukin, Fiz.  Tverd. Tela 
1151: {\bf 41}, 330 (1999) [ Phys. of Solid States {\bf 41}, XXX 
1152: (1999)]; see also the references therein.
1153: 
1154: 
1155: 
1156: \bibitem{Allmen} P.~von~Allmen,  Phys.  Rev.  B,  {\bf 46}, 15377
1157: (1992).
1158: 
1159: \bibitem{Maaref}M.~Maaref {\it et al.}, Solid State Commun., 
1160: {\bf 81}, 35 (1992).
1161: 
1162: \bibitem{Grinyaev}S.~N.~Grinyaev, private communication; see, 
1163: e.g., S.~N.~Grinyaev and G.~F.~Karavaev, Fiz.  Tverd. Tela 
1164: {\bf 42}, 752 (2000) [ Phys. of Solid States {\bf 42}, XXX 
1165: (2000)].
1166: 
1167: \bibitem{Govorov} A.~O.~Govorov and A.~V.~Chaplik, Sov. Phys. 
1168: JETP 72, 1037 (1991) [Zh. Eks. Teor. Fiz. 99, 1853 (1991)];
1169: A.~V.~Kalameitsev, A.~O.~Govorov, and V.~Kovalev,  JETP Lett.  
1170: 68, 669 (1998) [Pis'ma  Zh. Eks. Teor. Fiz. 68, 634 (1998)].  
1171: 
1172: \end{references}
1173: 
1174: \begin{figure}
1175: \caption{
1176: The model of the rough interface: side view.
1177: }
1178: %\label{}
1179: \end{figure}
1180: 
1181: \begin{figure}
1182: \caption{
1183: Temporal evolution of the $X_z$ exciton emmision.
1184: Theoretical curves (dashed and  solid  lines) was derived from 
1185: Eqs.~(\ref{1}) and (\ref{16})   respectively. Dotts  show 
1186: the experimental data.  } %\label{} 
1187: \end{figure} 
1188: 
1189: \begin{figure} 
1190: \caption{
1191: Temporal evolution of the $X_{xy}$ exciton emmision.
1192: Theoretical curve (solid line)  was derived from 
1193: Eq.~(\ref{1}). Dotts  show the 
1194: experimental data. 
1195: }
1196: %\label{}
1197: \end{figure}
1198: 
1199: 
1200: \end{document}
1201: 
1202: