1: \documentclass{article}
2: \usepackage{amssymb}
3: \usepackage{amsmath}
4: \usepackage{epsfig}
5: \parskip5pt\begin{document}
6: \numberwithin{equation}{section}
7:
8: \title{Raman solitons in transient SRS\footnote{to appear in Inverse Problems}}
9: \author{M. Boiti$^+$, J-G. Caputo, J. Leon, F. Pempinelli$^+$ \\
10: {\em Physique Math\'ematique et Th\'eorique, CNRS-UMR5825,}\\
11: Universit\'e Montpellier 2, 34095 MONTPELLIER (France)\\
12: $(^+)$ {\em and Dipartimento di Fisica dell' Universit\`a}, Lecce (Italy)}
13: \date{}\maketitle
14:
15: \begin{abstract} We report the observation of Raman solitons on numerical
16: simulations of transient stimulated Raman scattering (TSRS)
17: with small group velocity dispersion. The theory
18: proceeds with the inverse scattering transform (IST) for initial-boundary value
19: problems and it is shown that the explicit theoretical solution obtained
20: by IST for a semi-infinite medium fits strikingly well the numerical solution
21: for a finite medium. We understand this from the rapid decrease
22: of the medium dynamical variable (the potential of the scattering theory).
23: The spectral transform reflection coefficient can be computed directly
24: from the values of the input and output fields and this allows to
25: see the generation of the Raman solitons from the numerical
26: solution. We confirm the presence of these nonlinear modes in
27: the medium dynamical variable by the use of a discrete spectral analysis.
28: \end{abstract}
29:
30: \section{Introduction}\label{sec:basics}
31:
32: Stimulated Raman scattering (SRS) is a 3-wave interaction process with
33: extremely wide application in physics, especially in nonlinear optics
34: \cite{newell}\cite{yariv}. This essentially nonlinear phenomenon couples two
35: electromagnetic waves (pump and Stokes waves) to a two-level medium and is
36: described by a simple system of partial differential equations
37: \cite{chusco}.
38:
39: Such a system applies to different physical situations, depending on a
40: phenomenological damping factor chosen to match the observed Raman linewidth.
41: For instance, the steady state regime occurs when one neglects dynamical
42: effects on the medium. Then the system becomes explicitly solvable in terms of
43: the intensities and the result fits well the situation of strong damping and
44: long pulses, such as in fiber guides \cite{agrawall}. When instead dynamical
45: effects and damping are of same order, very interesting phase effects have been
46: discovered in \cite{druwen}. These have been interpreted as the manifestation
47: of solitons which, in the spectral transform scheme, would be related to
48: discrete eigenvalues. Instead they are related to the continuous
49: spectrum (they are not solitons) and named {\em Raman spikes} \cite{leon}.
50: Therefore the question of the observation of the Raman soliton is open since the
51: original work \cite{chusco} in 1975.
52:
53: When the damping term is much smaller than
54: the dynamical response of the medium, and can be neglected
55: the regime is called hyper-transient and applies to short duration pulses.
56: Here the system possesses a Lax pair \cite{chusco,kaup,steudel} and can be
57: treated by means of the inverse scattering transform (IST) generalized to
58: boundary-value problems \cite{jl,gakhov}. The observation of the Raman soliton
59: is an interesting open problem, especially since the boundary-value problem for
60: TSRS has been completely solved on the finite interval in \cite{fok-men} with
61: the essential result that, as expected from the works
62: \cite{pavel,menyuk,winter}, the field $q(x,t)$ universally evolves toward the
63: self-similar solution.
64:
65: The commonly used TSRS model is obtained as a 3-wave interaction
66: process where the group velocities of the 3 fields involved are considered
67: equal. As shown in \cite{sasha}, this assumption is an asymptotic
68: limit and consequently the group velocity
69: dispersion (GVD) has to be taken into account (through the spectral extension
70: of the input laser fields), leading to a modified SRS system (eq. \eqref{srs}
71: below). The main consequence of this fact is that Raman solitons (discrete
72: eigenvalues in the nonlinear Fourier spectrum) are effectively generated by SRS
73: in the medium \cite{sasha}.
74:
75: The aim of this paper is to show that Raman solitons are indeed created in a
76: finite length medium and that their generation is accompanied with poles in the
77: nonlinear Fourier spectrum (spectral transform). We follow the model
78: derived and solved in \cite{sasha} and establish the following
79: results.
80:
81: 1 - Based on the assumption of a semi-infinite medium, IST gives an
82: explicit solution\footnote{The solution is not explicit for the finite length
83: case: both results of \cite{sasha} and of \cite{fok-men} give the answer by
84: solving a system of Cauchy-Green integral equations.} which will be shown to
85: be an excellent approximation of the finite medium case. We explain
86: this from the fast decay of the {\em potential}
87: $q(x,t)$ for large $x$. Such is not the case in the zero-GVD case for which
88: $q(x,t)$ decreases as $x^{-3/4}$ typical of the self-similar solution
89: \cite{pavel,fok-men}.
90:
91: 2 - The reflection coefficient $\rho$ of the spectral transform
92: (or nonlinear Fourier spectrum) is expressed in
93: terms of the input and output field envelopes, allowing us to check
94: the appearance of Raman solitons (the real valued single poles of
95: $\rho$) on the numerical solutions. This also provides a means
96: to observe the generation of Raman solitons in experimental data.
97:
98: 3 - A recursion formula for computing the spectral transform of a finite set of
99: data has been recently proposed \cite{nft}. We have implemented this
100: {\em discrete spectral transform} and confirmed the creation of these
101: Raman solitons in a simulation on a finite length.
102: A remarkable feature here
103: is that the numerical spectral transform can be quite easily implemented,
104: its computation is faster, and it provides more information than a
105: conventional Fourier transform.
106:
107: After presenting the model in section 2, we report the numerical
108: observation of the Raman solitons in section 3. Section 4 explains
109: this by an analysis of the medium dynamical variable.
110:
111:
112: \section{The model}\label{sec:mod}
113:
114: We present here the extension of the SRS model of \cite{chusco} to the case
115: where the dispersion of the group velocity is not neglected.
116: We consider the classical model as in \cite{yariv}, well adapted
117: to molecular Raman scattering, and for which the medium is schematically
118: represented by a collection $X$ of harmonic oscillators coupled to the electric
119: field $\vec E$ through the polarizability of the medium.
120:
121: \subsection{Basic equations}\label{sub:base}
122:
123: When the polarizability depends on the frequency of the applied field, the
124: group velocity of the electromagnetic waves becomes frequency dependent. While
125: a very small group velocity dispersion (GVD) has no consequence on the evolution
126: of the amplitude, it has nontrivial effects on the phase dynamics. In this
127: context, it has been proved in \cite{sasha}, using {\em multiscale
128: analysis}, that for the electric field
129: \begin{align}\label{electric}
130: E(x,t')=& e^{i(k_1x-\omega_1t')}\int dk\ a (k,x,t)\ e^{-ikx}\notag\\
131: +& e^{i(k_2x-\omega_2 t')}\sqrt{\frac{\omega_2 }{\omega_1}}
132: \int dk\ b (k,x,t)\ e^{ikx}+ c.c.
133: \end{align}
134: and medium dynamical variable
135: $q(x,t) e^{i(Kx-\Omega t')} + c.c.$
136: where $\omega_1 -\omega_2 = \Omega$ and $k_1 - k_2= K$,
137: the resulting model of transient SRS can be written in the {\em retarded time}
138: $t=t'-x/v$
139: \begin{align}\label{srs}
140: &\partial_xa =qb e^{2ikx}\ ,\quad
141: \partial_xb =-\bar qa e^{-2ikx}\ , \notag\\
142: & \partial_tq =-g\int dk\ a \bar be^{-2ikx}\ , \end{align}
143: where the overbar stands everywhere for the complex conjugate.
144: Note the conservation ($x$-independence) of the flux
145: density $|a|^2+|b|^2$.
146:
147: The above model differs from the usual SRS system, corresponding
148: to the zero GVD case
149: \begin{equation}\label{srs-sharp}
150: \partial_xa =q_0 b \ ,\quad
151: \partial_xb =-\bar q_0 a \ , \quad
152: \partial_tq_0 =-g\, a \bar b \ , \end{equation}
153: by the presence of the integral over all possible realizations of the phase
154: mismatch $k$.
155:
156: \subsection{Boundary value problem}\label{sub:bounds}
157:
158: The question of interest is the time evolution of a couple $(a,b)$ of pump and
159: Stokes
160: pulses of time duration $T$ sent into a medium of length $\ell$. Hence the
161: domain of integration is
162: \begin{equation}\label{domain}
163: x\in[0,\ell]\ ,\quad t\in[0,T]\ .\end{equation}
164: and it is necessary to prescribe the values of the fields on the two boundaries
165: $t=0$ and $x=0$. The medium is initially ($t=0$) at rest (all molecules in the
166: fundamental state), hence the initial datum for the field $q$ is
167: \begin{equation}\label{init}
168: q(x,0)=0\ ,\end{equation}
169: independently of the reference frame
170: \footnote{Note that a {\em prepared} medium would
171: correspond to a state $q$ given at {\em physical time} zero, that is on the
172: characteristic $t=-x/v$.}.
173:
174: The input ($x=0$) light pulses are arbitrary functions of time $t$ with a given
175: spectral distribution around $k=0$, namely
176: \begin{equation}\label{input}
177: a(k,0,t)=A(k,t)\ ,\quad
178: b(k,0,t)=B(k,t)\ .\end{equation}
179: We shall be working here with the representative example treated in
180: \cite{sasha} which consists in assuming a common spectral lineshape for $A$ and
181: $B$ as a normalized Lorentzian lineshape for the intensities. More precisely we
182: set
183: \begin{equation}\label{in-spectr}
184: |A(k,t)|^2=|A_0(t)|^2\ \frac1\pi\frac{\kappa}{k^2+\kappa^2}\ ,\quad
185: |B(k,t)|^2=|B_0(t)|^2\ \frac1\pi\frac{\kappa}{k^2+\kappa^2}\ ,
186: \end{equation}
187: where the input pulse profiles $A_0(t)$ and $B_0(t)$ are arbitrary.
188:
189: It is useful also to understand the scale invariance of the SRS
190: system, including its boundary values. By a simple change of variables it can
191: be easily shown that the system \eqref{srs} (characterized by the parameters
192: $g$ and $\ell$) {\em together with} the input boundary values
193: \eqref{input} \eqref{in-spectr}
194: (characterized by the parameter $\kappa$) bears the scale invariance
195: \begin{equation}\label{scale-param}
196: \ell\to\alpha\ell\ ,\quad g\to g/\alpha\ ,\quad\kappa\to\kappa/\alpha\ .
197: \end{equation}
198: This invariance has been checked on numerical simulations as one way to test
199: the code accuracy, and it works so well that we have obtained indiscernible
200: pictures.
201:
202:
203:
204: \section{Laser fields}
205:
206: We describe hereafter the IST solution of the boundary-value problem
207: \eqref{input} for the system \eqref{srs} using the results of \cite{sasha},
208: and compare it with the direct numerical solution in the finite length case.
209:
210: \subsection{Sketch of the direct problem}
211:
212: The direct problem consists in defining the spectral transform from the
213: data of the {\em potential} $q(x,t)$ and the boundary values $A(k,t)$
214: and $B(k,t)$ (as time $t$ appears everywhere as an external parameter,
215: we shall forget it here).
216: This is done by defining the following Jost solutions (for $k\in{\mathbb R}$)
217: \begin{align}\label{jost-sol}
218: &\varphi(k,x)=1-\int_0^xd\xi\ q(\xi)\phi(k,\xi)\ ,\notag\\
219: &\phi(k,x)=\int_0^xd\xi\ \overline{q}(\xi)\varphi(k,\xi)
220: \ e^{2ik(x-\xi)}\ ,
221: \end{align}
222: which are both entire functions of $k$ vanishing as
223: $k\to\infty$ in the upper half-plane.
224: \footnote{The relation with notations of \cite{sasha} is:
225: $\varphi(k)=\varphi_{11}^+(k)=\overline{\varphi_{22}^-}(\bar k)$,
226: $\phi(k)=-\varphi_{21}^+(k)=
227: \overline{\varphi_{12}^-}(\bar k).$}
228:
229: These two functions then allow to define the {\em reflection coefficient}
230: $\rho(k)$ and the {\em transmission coefficient} $\tau(k)$ by taking the
231: limit $x\to\infty$ of $\varphi$ and $\overline{\phi} e^{2ikx}$ for
232: $k\in \mathbb R$, namely
233: \begin{equation}\label{scatt-coeffs}
234: \frac1{\tau}=1-\int_0^\infty d\xi\ q(\xi)\phi(k,\xi)\ ,\quad
235: \frac{\rho}{\tau}=-\int_0^\infty d\xi\ q(\xi)\overline{\varphi}(k,\xi)
236: e^{2ik\xi}\ .
237: \end{equation}
238: The coefficient $1/\tau$ is clearly an entire function of $k$ and one can show
239: \cite{sasha} that the reflection coefficient $\rho(k)$ is meromorphic in the
240: upper half-plane with a finite number of single poles related to the
241: solitonic part of the solution $q$. We have also the following {\em unitarity}
242: relation for $k\in{\mathbb R}$
243: \begin{equation}\label{unit}
244: |\rho|^2=1+|\tau|^2\ .\end{equation}
245:
246: It is easy to prove finally that the vectors $(\varphi,\ -\phi e^{-2ikx})$
247: and $(\overline{\phi} e^{2ikx},\ \overline{\varphi})$ solve the same
248: differential equation as the vector $(a,\ b)$ in \eqref{srs}. Then by
249: comparing their values in $x=0$ we readily obtain from \eqref{input}
250: \begin{equation}\label{lin-rel}
251: a=A\varphi+B\overline{\phi} e^{2ikx}\ ,\quad
252: b=B\overline{\varphi}-A\phi e^{-2ikx}\ .\end{equation}
253:
254:
255: \subsection{Output pump pulse}
256:
257: From \eqref{scatt-coeffs}, \eqref{unit} and \eqref{lin-rel},
258: the output $|a(k,\ell,t)|^2$ is explicitly given for $\ell\to\infty$ by
259: the expression
260: \begin{equation}\label{pump-out-infty}
261: |a (k,\infty,t)|^2=\frac1{1+|\rho|^2}\
262: \left|A-\rho B\right|^2\ ,\end{equation}
263: where $A$ and $B$ are the input data defined in \eqref{input}.
264:
265: The main result of \cite{sasha} is that
266: the function $\rho(k,t)$ is obtained by solving the following Riccati time
267: evolution
268: \begin{equation}\label{rho-evol}
269: \rho_t=-\rho^2\ {\cal C}_k^+[m^*]-2\rho\ {\cal C}_k^+[\phi]-
270: {\cal C}_k^+[m]\ ,\quad \rho(k,0)=0\ ,
271: \end{equation}
272: where the functions $m(k,t)$ and $\phi(k,t)$ are given from the input data by
273: \begin{equation}\label{m-phi}
274: m=\frac{i\pi}{2}g AB^*,\quad \phi=\frac{i\pi}{4}g (|A|^2-|B|^2)\ ,
275: \end{equation}
276: and where ${\cal C}_k^+$ denote the following Cauchy integral
277: \begin{equation}\label{cal-c}
278: {\cal C}_k^+[f]=\frac1\pi\int_{-\infty}^{+\infty}
279: \frac{d\zeta}{\zeta-(k+i0)}\ f(\zeta)\ .
280: \end{equation}
281: Consequently, for given inputs $A(k,t)$ and $B(k,t)$, the expression
282: \eqref{pump-out-infty} gives the explicit asymptotic output pump intensity
283: from the solution of the evolution \eqref{rho-evol}.
284: It is worth remarking that, at any given time, $\rho$ possesses possibly
285: a finite number of simple poles whose time evolution is given by the
286: nonlinearity of the Riccati equation.
287:
288: For practical purpose, we choose in \eqref{input} the Stokes wave input seed as
289: a portion $e^{-\gamma}$ of the pump wave, namely
290: \begin{equation}\label{ratio}
291: B_0(t)=A_0(t)\ e^{-\gamma}\ .\end{equation}
292: In that case the evolution \eqref{rho-evol} can be explicitly solved
293: \cite{sasha}
294: \begin{equation}\label{rho-sol-1}
295: \rho(k,t)=\frac{\sinh\delta(k,t)}{\cosh(\delta(k,t)-\gamma)},
296: \end{equation}
297: \begin{equation}\label{delta}
298: \delta(k,t)=\frac{i T(t)}{k+i\kappa},\quad
299: T(t)=\frac14g(1+e^{-2\gamma})\int_0^td\tau|A_0(\tau)|^2.
300: \end{equation}
301:
302:
303: \subsection{Numerical solution of finite length TSRS}
304:
305: Our purpose is to understand how the above IST-solution can be used to model
306: the solution of the SRS system \eqref{srs} on a finite length.
307: This can be done first in a
308: qualitative way by writing the solution of \eqref{srs} with boundary values
309: \eqref{input} as the equivalent integral form
310: \begin{equation}\label{integral}
311: \left(\begin{array}{c} a(k,x,t) \\ b(k,x,t) \end{array}\right)=
312: \left(\begin{array}{c} A(k,t) \\ B(k,t) \end{array}\right)+
313: \int_0^xd\xi
314: \left(\begin{array}{c} q(\xi,t)b(k,\xi,t)e^{2ik\xi}\\
315: -\overline{q}(\xi,t)a(k,\xi,t)e^{-2ik\xi} \end{array}\right)\ .\end{equation}
316: The output $a(k,\ell,t)$ will not differ much from its asymptotic value
317: $a(k,\infty,t)$ if $q(x,t)$ is sufficiently small for $x>\ell$. Consequently
318: the behavior of $q(x,t)$ at large $x$, is essential for the adequation of the
319: formula \eqref{pump-out-infty} to real situations. We will see in the next
320: section that it is also crucial for the applicability of the spectral method.
321: We now proceed to show that the numerical solution of the TSRS system
322: \eqref{srs} on a finite interval is in excellent agreement with the theoretical
323: expression for the output \eqref{pump-out-infty}.
324:
325: We discretized \eqref{srs}
326: in both $x$ and $t$ using an order 2 Runge-Kutta method and advance
327: via the following algorithm:\\
328: 1 - given $a(k,x,t)$ and $b(k,x,t)$ for all real k, compute $q(x,t)$
329: by integrating the time-evolution of $q$ at $x$ for the initial datum
330: $q(x,t=0)=0$, where the integral is calculated using the trapezoidal rule,\\
331: 2 - advance to $a(k,x+dx,t)$ and $b(k,x+dx,t)$, for all $k$,
332: by integrating the differential equation for $a$ and $b$,\\
333: 3 - go to step 1 with $x=x+dx$.\\
334: The scheme is started at step 1 for $x=0$.
335:
336: The quality of the computation is monitored by evaluating the relative
337: error in the total flux $|a(k,x,t)|^2 + |b(k,x,t)|^2$ which is conserved
338: by \eqref{srs}. In all the runs that are presented it remained
339: smaller than $10^{-5}$.
340:
341: We chose as parameters
342: \begin{equation}\label{param}
343: \ell=80\ ,\quad g=0.5\ ,\quad \gamma=5\ ,\quad \kappa=0.2\ ,\end{equation}
344: and the input pump pulse envelope is the Gaussian
345: \begin{equation}
346: A_0(t)=\exp\left[-\left(\frac{t-50}{30}\right)^2\right]\ .
347: \end{equation}
348: For this choice we found that $dt \leq T/1000$ and $dx \leq \kappa / 2$
349: gave stable results. Another point is that because of the Lorentzian
350: line width we had to take a $k$ interval of width $ \approx 80 \kappa$
351: in order to ensure that the integrals were normalized. To describe
352: the strong oscillations present for the zero GVD system \eqref{srs-sharp}
353: we had to chose $dt \leq T/ 2000$ and $dx \leq \ell / 10000$.
354:
355: A typical run with number of grid points in $x,t$ and $k$ $(n_x=n_t=1000;n_k=
356: 500)$ takes about 2 hours CPU monoprocessor on a RS10000. We used the
357: parallelism of the problem, i.e. the fact that the marching in $x$ (resp. $t$)
358: can be made in parallel for the loops in $t$ and $k$ (resp. $x$) and
359: implemented the code on a Silicon Graphics SGI 10000 using the OpenMP software.
360: This enabled a gain of a factor 8 or 10 in computing time depending on the
361: number of processors used.
362:
363: Figure \ref{f01} show the pump intensity input $|a(k,0,t)|^2$ and output
364: $|a(k,\ell,t)|^2$ computed from the IST expression \eqref{pump-out-infty}
365: together with the numerical solution of the system \eqref{srs} for a length
366: $\ell=80$ and four different values of the parameter $k$, in excellent
367: agreement (a tiny discrepancy is only seen in the last picture).
368:
369: It is worth remarking that one cannot pursue a given numerical experiment to a
370: longer length arbitrarily. We have stopped for instance the above calculation
371: at length $\ell=80$ where the potential $|q(x,t)| < 10^{-5}$. Continuing the
372: run to longer lengths would yield an increase of the potential which would then
373: start to oscillate. This is probably due to the system itself which amplifies
374: the numerical errors in a drastic way at long lengths (note that from the scale
375: invariance \eqref{scale-param} longer length means larger Raman amplification).
376:
377: \subsection{Raman solitons}
378:
379: The function $\rho$ of \eqref{rho-sol-1} has an essential singularity in
380: $k=-i\kappa$ and a set of single poles $k_n$ evolving in time, given by
381: \begin{equation}\label{pole}
382: k_n(t)=-i\kappa+\frac{T(t)}{(n+\frac12)\pi-i\gamma}\ ,
383: \end{equation}
384: for $n\in{\mathbb N}$. At time zero no pole is present and, as $t$ evolves,
385: poles move upward from $-i\kappa$ and eventually reach the real axis at the
386: times $t_n$ defined by ${\rm Im}(k_n(t_n))=0$, i.e. by the implicit expression
387: \begin{equation}
388: T(t_n)=\frac{\kappa}{\gamma}\ [\gamma^2+[(n+\frac12)\pi]^2]\ .\end{equation}
389: Note that $t_n=t_{-n-1}$, hence the poles cross the real axis by pair. The
390: corresponding positions on the real axis are then given by
391: \begin{equation}\label{zetan}
392: k_n(t_n)=-k_{-n-1}(t_n)=\zeta_n\ ,\quad
393: \zeta_n=\frac\kappa\gamma(n+\frac12)\pi\ .\end{equation}
394: We plot in Figure \ref{f02} the imaginary part of the poles $k_n(t)$
395: as a function of time for the parameter values (\ref{param}) and
396: $n=0,1,2$ and 3. We observe that the first three poles cross the
397: real axis at the positions $\zeta_0=6.3 10^{-2}, \zeta_1=0.19, \zeta_2= 0.31$
398: and times $t_0=39.2, t_1=46.3$ and $t_2=59.6$. The pole $k_3$ starts
399: to evolve with $t$ but cannot cross the real axis because of the
400: finite duration of the pump pulse.
401:
402: As soon as these poles move to the upper half-plane, they generate a soliton
403: component in the {\em "potential"} $q(x,t)$. We will prove in the next section
404: that this potential (the medium dynamical variable) is a continuous
405: function of $t$ when a pole crosses the real axis.
406:
407: \subsection{The spectral transform from the output laser pulses}
408:
409: The expressions \eqref{lin-rel} can be inverted to get the following
410: expressions of the Jost solutions in terms of the (physical) fields $a(k,x,t)$
411: and $b(k,x,t)$ and of the input pulses $A(k,t)$ and $B(k,t)$
412: \begin{equation}\label{inv-rel}
413: \varphi =\frac{a \overline{A}+\overline{b} B}{|A|^2+|B|^2} \ ,\quad
414: \overline{\phi}e^{2ikx}=\frac{a \overline{B}-\overline{b} A}{|A|^2+|B|^2} \ .
415: \end{equation}
416: Then the definitions \eqref{scatt-coeffs}
417: lead to the following formula
418: \begin{equation}\label{rho-out}
419: \rho(k,t)=\left.
420: \frac{\overline{b} A-a \overline{B}}{a \overline{A}+\overline{b} B}
421: \right|_{x\to\infty}
422: \end{equation}
423: which gives the spectral transform $\rho(k,t)$ in terms of the output pump and
424: Stokes fields. This function for $k$ real must become singular at the two
425: points $\pm \zeta_n$
426: each time a soliton $k_n$ is created and it is used now to prove
427: the generation of Raman solitons in numerical experiments.
428:
429:
430: For the input pulses given in \eqref{in-spectr} and the particular choice
431: \eqref{ratio}, we readily obtain from the above
432: \begin{equation}\label{rho-out-th}
433: \rho=\left.\frac{\overline{b}-a e^{-\gamma}}{a +\overline{b} e^{-\gamma}}
434: \right|_{x\to\infty}\ .
435: \end{equation}
436: We use this expression to estimate $\rho_\ell$ from the numerical solution
437: $(a,b)$ at $x=\ell < +\infty$ and compare it to the $\rho$ obtained from the
438: IST (\ref{rho-sol-1}) for the parameters (\ref{param}). Notice that
439: $|\rho(-k)|=|\rho(k)|$ so that we will only present positive values of $k$.
440:
441: Figure \ref{f03} shows the function $|\rho|$ of \eqref{rho-sol-1} in full line
442: and the numerical result obtained from \eqref{rho-out-th} in $x=\ell$ the in
443: dashed line, for the 3 values $k=\zeta_0,\zeta_1,\zeta_2$ and $\zeta_3$. Both
444: expressions are very close and as expected $\rho(\zeta_n)$ is singular for
445: $t=t_n$ for $n\leq 2$, while it is regular for $n=3$. In the first picture of
446: figure \ref{f03}, the dashed line actually represents the value at $k=0$
447: instead of $k=\zeta_0$. This is the only noticable discrepancy between
448: analytic asymptotic formula and numerical/experimental expression, resulting
449: from the finiteness in $x$ of the data $q(x)$ (the wave length $\lambda_0
450: \equiv 2 \pi / \zeta_0 \approx 104 > \ell$).
451:
452:
453: \section{Medium }
454:
455: In the previous section we have seen that the finite length solution
456: is in good agreement with the asymptotic behavior
457: (\ref{pump-out-infty}) given
458: by the IST on the semi-infinite line. We now proceed to justify this
459: fact by analyzing the behavior of the medium dynamical variable
460: $q(x,t)$.
461:
462:
463: \subsection{Sketch of the inverse problem}
464:
465: IST furnishes the solution $q(x,t)$, called {\em medium dynamical variable},
466: by the expression \cite{sasha}
467: \begin{equation}\label{qrebuild}
468: q(x,t) =2i\psi^{(1)}(x,t)
469: \end{equation}
470: where $\psi^{(1)}$ is the coefficient of $1/k$ in the Laurent expansion
471: of the function $\psi(k,x,t)$ solution of
472: the following Cauchy-Green coupled system
473: \begin{align}\label{cauchy}
474: \varphi(k)=1+\frac1{2i\pi}\int_{-\infty}^{+\infty} d\lambda\
475: \frac{\bar\rho(\lambda)\psi(\lambda)}{\lambda-k-i0}\ e^{2i\lambda x}-
476: \sum_1^N\frac{\bar\rho(\bar k_n)\psi(\bar k_n)}{k-\bar k_n}\ e^{2i\bar k_n x}
477: \notag\\
478: \psi(k)=\frac1{2i\pi}\int_{-\infty}^{+\infty}
479: d\lambda\ \frac{\rho(\lambda)\varphi(\lambda)}{\lambda-k+i0}
480: e^{-2i\lambda x}+
481: \sum_{n=1}^{N}\frac{\rho_n\varphi(k_n)}{k-k_n}\ e^{-2ik_nx} \ .
482: \end{align}
483: Here $\rho_n(t)$ are the $N$ residues of $\rho(k,t)$ at the poles $k_n(t)$
484: that are in the upper half $k$-plane (if any).
485: Note that the solution $\varphi$ corresponds to the solution of
486: \eqref{jost-sol} and that $\psi(k)=\overline{\phi}(\bar k)$.
487:
488:
489: Apparently the generation of a soliton is followed by the adjunction
490: of a term in the equation for $\psi$, and consequently also in the
491: expression of $q$, leading to a possible discontinuity for $q$.
492: We will now proceed to show that this is not the case.
493: This situation is particular to the spectral problem on the semi-line
494: where the continuous spectrum (value of $\rho(k)$ on the real axis)
495: is not separable from the discrete spectrum (poles of $\rho$ in the upper
496: half complex plane). This is due to the motion of the poles of $\rho$
497: that can cross the real axis. On the contrary, in the full line case, solitons
498: (discrete spectrum) can exist without radiation (continuous spectrum).
499:
500: \subsection{Continuity}
501:
502: From now on when the spectral parameter $k$ is generically complex we shall be
503: using boldface letter ${\bf k}.$ From what precedes, at $t=t_{0}$ a pole
504: crosses the real axis at $k=k_0$ from the lower half plane. We consider here
505: the limits $t\rightarrow t_{0}\pm 0$ and we write for $k\in{\mathbb R}$
506: \begin{equation}
507: \rho_{\pm }(k)= \rho (k,t_{0}\pm 0)=\frac{R_{0}(k)}{k-k_0\mp i0}
508: \label{rho+-}
509: \end{equation}
510: where $R_{0}({\bf k})$ is analytic in a neighborhood of $k_0$.
511: Note that
512: \begin{equation} \label{rho+rho-}
513: \rho _{+}(k)=\rho_{-}(k)+2\pi i\rho_{0}\delta (k-k_0)
514: \end{equation}
515: where $\rho_{0}=R_{0}(k_0)$ is the residue of $\rho$ at the pole
516: $k=k_0$.
517:
518: Assuming for simplicity no other pole, we have
519: from \eqref{cauchy} at $t=t_{0}-0$
520: \begin{equation}\label{cauch1}
521: \psi(k,x,t_{0}-0)=
522: {\frac{1}{2i\pi }}\int_{-\infty }^{+\infty }d\lambda
523: \ \frac{\rho_{-}(\lambda)\varphi(\lambda,x,t_{0}-0)}{\lambda -k+i0}
524: \ e^{-2i\lambda x}
525: \end{equation}
526: which is well defined at $k=k_0$ since the distribution
527: $(\lambda -k_0+i0)^{-2}$ is meaningful.
528:
529: On the contrary at $t=t_{0}+0$, expression \eqref{cauch1} diverges
530: due to the distribution $(\lambda -k_0+i0)^{-1}(\lambda -k_0-i0)^{-1}$.
531: Since now there is a pole in the upper half complex plane, we should
532: take the limit ${\rm Im}({\bf k})\to 0$ in the complete expression in
533: \eqref{cauchy}. i.e.
534: \begin{align}\label{cauch2}
535: \psi({\bf k},x,t_{0}+0) =&
536: {\frac{1}{2i\pi }}\int_{-\infty }^{+\infty }d\lambda
537: \ \frac{\rho_{+}(\lambda)\varphi(\lambda,x,t_{0}+0)}{\lambda-{\bf k}}
538: \ e^{-2i\lambda x}\notag\\
539: &+\frac{\rho _{0}\varphi(k_0,x,t_{0}+0)}{{\bf k}-k_0}e^{2ik_0x}\ ,
540: \end{align}
541: for ${\rm Im}({\bf k})<0$. We obtain
542: \begin{equation}
543: \psi(k,t_{0}-0)=\psi(k,t_{0}+0).
544: \end{equation}
545: and consequently from (\ref{qrebuild})
546: \begin{equation}
547: q(t_{0}-0)=q(t_{0}+0).
548: \end{equation}
549:
550: \subsection{Asymptotic behavior}
551:
552: We consider here for simplicity the case when solitons are not yet present.
553: Then from \eqref{qrebuild} and \eqref{cauchy}, $q$ can be reconstructed in
554: terms of the reflection coefficient $\rho $ and the Jost solution $\varphi$ by
555: means of
556: \begin{equation}\label{qrecons}
557: q(x)=-\frac{1}{\pi }\int_{-\infty}^{+\infty}
558: dk\ \rho(k)\varphi(k,x)e^{-2ikx}\ .
559: \end{equation}
560: This formula yields directly an information on the behavior of $q$
561: at large $x$.
562:
563: In fact, if $\rho (k)\varphi(k,x),$ $\dots$,
564: $\partial_{k}^{(n-1)}\{ \rho (k)\varphi(k,x)\}$ are
565: continuous and tend to $0$ for $k\to\infty $, and if
566: $\partial_{k}^{(n)}\{ \rho(k)\varphi(k,x)\} \in L({\mathbb R})$, we obtain by
567: repeated integration by parts
568: \begin{equation}
569: q(x)=\left( \frac{i}{2\pi x}\right)^{n}\int dk\ \partial _{k}^{(n)}\{
570: \rho (k)\varphi(k,x)\} e^{-2ikx}.
571: \end{equation}
572: Hence finally
573: \begin{equation}
574: x\to \infty\quad\Rightarrow\quad x^{n}q(x)\to 0\ .\end{equation}
575:
576: As mentioned in the introduction, such an asymptotic behavior is not found for
577: the potential $q_0(x,t)$ which would be obtained from TSRS with zero GVD
578: \eqref{srs-sharp}. Indeed, in that case the medium initially at rest evolves
579: universally towards the self-similar solution \cite{fok-men} which behaves as
580: $x^{-3/4}$ as found in \cite{pavel}. More precisely we have
581: \begin{equation}\label{q-behav-sharp}
582: q\sim s(t)\,\xi^{-3/4}\left[\alpha\cos(h\xi^{1/2}+\beta)+{\cal O}(\xi^{-1/2})
583: \right]\ ,\quad \xi=s(t)\,x\ ,\end{equation}
584: where $s(t)$ is a function of $t$ determined by the boundary conditions and
585: where $\alpha$, $\beta$ and $h$ are arbitrary constants.
586:
587: Figure \ref{f04} presents the decay of $|q|$ and $|q_0|$ respectively in
588: dashed line and full line as a function of $x$ for $t=T/2=50$ in
589: a linear-log plot and the parameters \eqref{param}. The exponential
590: decay of $|q|$ can be seen for all values of $t$ and follows approximately
591: $e^{-\kappa x}$ as expected from the analysis. This observation justifies
592: the fact that the finite length evolution of \eqref{srs} is very close to
593: the asymptotic expression \eqref{pump-out-infty} given by IST. \\
594: On the contrary the slow decay of $|q_0|$ as $x^{-3/4}$ makes it impossible
595: for the integrals of \eqref{jost-sol} to exist so that the scattering theory must be
596: revisited in this case.
597:
598:
599: \subsection{Numerical spectral transform and Raman solitons}
600:
601: By numerical integration of the SRS system \eqref{srs} with a spatial grid of
602: dimension $h$, we get, at each value of time, a set of $L$ discrete data
603: $q(n,t)$ (with $x=nh$ and $\ell=Lh$), that we now analyze by means of the
604: spectral transform. This is easily done, using the results of \cite{nft}
605: which, adapted to our notations, read
606: \begin{equation}\label{disr}
607: \rho(k,t)=R_0(\zeta,t)\ ,\end{equation}
608: where $R_0$ is obtained from the following inverse recursion
609: \begin{equation}R_L =\zeta^Lhq(L,t) \ ,\quad
610: R_{m-1}=\frac{R_m-hq(m{-}1,t)\,\zeta^{m-1}}
611: {1+R_mh\bar q(m{-}1,t)\,\zeta^{-m+1}}\ ,
612: \end{equation}
613: and where the parameter $\zeta$ is related to $k$ by
614: \begin{equation}
615: \zeta=e^{2ikh}\ .\end{equation}
616:
617: It is then a simple task to use this recursion relation to compute the
618: numerical spectral transform $R_0(\zeta,t)$ and compare it to the asymptotic
619: theoretical expression \eqref{rho-sol-1} of $\rho(k,t)$. In Figure
620: \ref{f05} we plot $|\rho(k,t)|$ where one can clearly see the three
621: singularities $(\zeta_n,t_n)$ for $n=0,1,2$ corresponding to the three
622: poles of $\rho$.
623:
624:
625: The non linear spectral transform is easier to compute, faster and
626: yields more information than the standard Fourier transform which in
627: particular does not present any singularity (as we checked numerically).
628:
629:
630: \section{Conclusion}
631:
632: This work demonstrates that the IST solution of transient SRS on the
633: semi-infinite line furnishes a very accurate model for the solution on a finite
634: domain. The accuracy stands not only for the output laser pulses intensities
635: but also for their phases as shown in a spectacular way by the generation of
636: the Raman solitons. It is shown that the question of the experimental
637: observation of Raman solitons is solved by a convenient combination of the
638: output (and input) laser profiles. We also performed a numerical {\em
639: nonlinear spectral analysis} which resulted to be not only rich of information
640: but also quite easy to implement and faster than the usual FFT procedure.
641:
642: Finally we mention that it is difficult to compare these results with those
643: obtained in the zero GVD case \eqref{srs-sharp}. The problem is that in the
644: limit $\kappa\to0$, the spectral transform gets an essential singularity in
645: $k=-i0$ which sends poles to the upper half-plane. To discuss the number and
646: time location of such poles would require not only a careful study of the
647: singular limit $\kappa\to0$, but also to reformulate IST on the half-line as we
648: have seen that the potential $q(x)$ does not decrease {\em fast enough} as
649: $x\to\infty$.
650:
651: \section*{Acknowledgments}
652: JGC is on leave from the Laboratoire de Math\'ematiques de
653: l'INSA de Rouen. The authors thank the CRIHAN computing
654: center for the use of their facilities. We thank INTAS
655: for support through grant 96-339 and support from Sezione INFN di Lecce
656: and PRIN 97 "Sintesi".
657:
658:
659:
660: \begin{thebibliography}{aaaaaa}
661:
662: \bibitem{newell} A.C. Newell, J.V. Moloney, {\em Nonlinear Optics},
663: Addison-Wesley (Redwood City CA, 1992)
664:
665: \bibitem{yariv} A. Yariv, {\em Quantum Electronics},
666: J. Wiley (New York 1975)
667:
668: \bibitem{chusco} F.Y.F. Chu, A.C. Scott,
669: Phys. Rev. A {\bf12} (1975) 2060
670:
671: \bibitem{agrawall} G.P. Agrawall, {\em Nonlinear optics}, Academic Press
672: (London 1989)
673:
674: \bibitem{druwen} K. Dr\"uhl, R.G. Wenzel, J.L. Carlsten, Phys. Rev. Lett.,
675: {\bf51} (1983) 1171; R.G. Wenzel, J.L. Carlsten, K. Dr\"uhl,
676: J. Stat. Phys., {\bf 39} (1985) 621
677:
678: \bibitem{leon}C. Claude, J. Leon, Phys. Rev. Lett., {\bf74} (1995) 3479;
679: C. Claude, F. Ginovart, J. Leon, Phys. Rev. A {\bf52} (1995) 767
680:
681: \bibitem{kaup} D.J. Kaup, Physica {\bf6D} (1983) 143
682:
683: \bibitem{steudel} H.Steudel, Quantum Opt., {\bf 2} (1990) 387
684:
685: \bibitem{jl}J. Leon, Phys. Lett. A {\bf170} (1992) 283;
686: Phys. Rev. A{\bf47} (1993) 3264; J. Math. Phys., {\bf 35} (1994) 3054
687:
688: \bibitem{gakhov} D.E. Gakhovich, A.S. Grabchikov, V.A. Orlovich, Optics Comm.,
689: {\bf 102} (1993) 485
690:
691: \bibitem{fok-men}A.S. Fokas and C.R. Menyuk, J. nonlinear Sci., {\bf 9}
692: (1999) 1
693:
694: \bibitem{pavel} C.R. Menyuk, D. Levi, P. Winternitz, Phys. Rev. Lett. {\bf 69}
695: (1992) 3048; D. Levi, C.R. Menyuk, P. Winternitz, Phys. Rev. A {\bf 44}
696: (1991) 6057
697:
698: \bibitem{menyuk}C.R. Menyuk, Phys. Rev. A, {\bf 47} (1993) 2235
699:
700: \bibitem{winter}{\em Self-similarity in stimulated Raman scattering}, eds.
701: D. Levi, C.R. Menyuk, P. Winternitz, Les Publications CRM (Montreal 1994)
702:
703: \bibitem{sasha} J. Leon, A.V. Mikhailov, Phys. Lett. A. {\bf 253} (1999) 33
704:
705: \bibitem{nft}M. Boiti, J. Leon, F. Pempinelli, Phys. Rev. E {\bf 54} (1996)
706: 5739. M. Boiti, F. Pempinelli, Inverse Problems {\bf 13} (1997) 919
707:
708: \end{thebibliography}
709:
710:
711: \begin{figure}
712: \centerline{\psfig{file=fig-leon1.ps,height=12cm,width=18cm}}
713: \caption{ Plot, as functions of time, of the output pump intensity
714: $|a(x=\ell,k,t)|^2$ from the theoretical expression (\ref{pump-out-infty})
715: (full line) and the numerical solution of (\ref{srs}) (dashed line)
716: for four different values of $k$.
717: The large gaussian curves (dashed line) represent the input pump intensities
718: $|a(x=0,k,t)|^2$.}\label{f01}
719: \end{figure}
720:
721:
722: \begin{figure}
723: \centerline{\psfig{file=fig-leon2,height=18cm,width=12cm,angle=-90}}
724: \caption{ Time evolution of the imaginary part of the poles $k_n$
725: of $\rho$ for $n=0,1,2$ and 3. The parameters are the same as in Figure 1}
726: \label{f02}
727: \end{figure}
728:
729: \begin{figure}
730: \centerline{\psfig{file=fig-leon3.ps,height=12cm,width=18cm,angle=0}}
731: \caption{Time evolution of the reflection coefficient $|\rho(\zeta_n)|$
732: for $n=0,1,2$ and 3 obtained from the inverse scattering
733: theory (\ref{rho-sol-1}) (full line) and from the numerical solution of
734: (\ref{srs}) using
735: formula (\ref{rho-out-th}) (dashed line). }
736: \label{f03}
737: \end{figure}
738:
739: \begin{figure}
740: \centerline{\psfig{file=fig-leon4.ps,height=18cm,width=12cm,angle=-90}}
741: \caption{Evolution of $|q|(x,t=50)$ and $|q_0|(x,t=50)$ in linear-log
742: coordinates. }
743: \label{f04}
744: \end{figure}
745:
746: \begin{figure}
747: \centerline{\psfig{file=fig-leon5.ps,height=18cm,width=12cm,angle=-90}}
748: \caption{Evolution of $|\rho|(k,t)$ obtained from the
749: discrete spectral transform (\ref{disr}). }
750: \label{f05}
751: \end{figure}
752: \end{document}
753:
754: