1: % mn2esample.tex
2: %
3: % v2.1 released 22nd May 2002 (G. Hutton)
4: %
5: % The mnsample.tex file has been amended to highlight
6: % the proper use of LaTeX2e code with the class file
7: % and using natbib cross-referencing. These changes
8: % do not reflect the original paper by A. V. Raveendran.
9: %
10: % Previous versions of this sample document were
11: % compatible with the LaTeX 2.09 style file mn.sty
12: % v1.2 released 5th September 1994 (M. Reed)
13: % v1.1 released 18th July 1994
14: % v1.0 released 28th January 1994
15:
16: \documentclass[useAMS,usenatbib,onecolumn,usegraphicx]{mn2e}
17: %\usepackage[dvips]{color}
18: %\documentclass[useAMS,usenatbib,usegraphicx]{mn2e}
19: % If your system does not have the AMS fonts version 2.0 installed, then
20: % remove the useAMS option.
21: %
22: % useAMS allows you to obtain upright Greek characters.
23: % e.g. \umu, \upi etc. See the section on "Upright Greek characters" in
24: % this guide for further information.
25: %
26: % If you are using AMS 2.0 fonts, bold math letters/symbols are available
27: % at a larger range of sizes for NFSS release 1 and 2 (using \boldmath or
28: % preferably \bmath).
29: %
30: % The usenatbib command allows the use of Patrick Daly's natbib.sty for
31: % cross-referencing.
32: %
33: % If you wish to typeset the paper in Times font (if you do not have the
34: % PostScript Type 1 Computer Modern fonts you will need to do this to get
35: % smoother fonts in a PDF file) then uncomment the next line
36: % \usepackage{Times}
37:
38: %%%%% AUTHORS - PLACE YOUR OWN MACROS HERE %%%%%
39:
40: %\linespread{2.5}
41: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
42:
43: \title[Evolution of initially localized perturbations in stratified ionized disks]{Evolution
44: of initially localized perturbations in stratified ionized disks}
45: \author[Edward Liverts and Michael Mond]{Edward Liverts\thanks{E-mail:
46: eliverts@bgu.ac.il (EL); mond@bgu.ac.il (MM)} and Michael Mond \\
47: Department of Mechanical Engineering, Ben-Gurion
48: University of the Negev, \\ P.O. Box 653, Beer-Sheva 84105,
49: Israel}
50: \begin{document}
51:
52: \date{Accepted ---. Received ----; in original form ----}
53:
54: \pagerange{\pageref{firstpage}--\pageref{lastpage}} \pubyear{}
55:
56: \maketitle
57:
58: \label{firstpage}
59:
60: \begin{abstract}
61: A detailed solution of an initial value problem of a vertically localized
62: initial perturbation in rotating magnetized vertically stratified disk is
63: presented. The appropriate linearized MHD equations are solved by
64: employing the WKB approximation and the results are verified numerically.
65: The eigenfrequencies as well as eigenfunctions are explicitly
66: obtained. It is demonstrated that the initial perturbation remains confined within
67: the disk. It is further shown that thin enough disks are stable but as their
68: thickness grows increasing number of unstable modes participate in the
69: solution of the initial value problem. However it is demonstrated that due to
70: the localization of the initial perturbation the growth time of the
71: instability is significantly longer than the calculated inverse growth rate
72: of the individual unstable eigenfunctions.
73: \end{abstract}
74:
75: \begin{keywords}
76: accretion, accretion discs, MHD, MRI, WKB solution.
77: \end{keywords}
78:
79: \section{Introduction}
80:
81: Magnetohydrodynamic (MHD) instabilities play a major role in a great
82: variety of astrophysical and space applications. Their importance is
83: epitomized by the magneto-rotational instability (MRI) that has been
84: discovered by \cite{velikhov} and \cite{chandra} for infinite cylinders and
85: rediscovered by \cite{bh} in astrophysical context.
86: It is widely believed to be one of the prime candidates to provide a viable clue
87: to solving the age old puzzle of the outwards transfer of angular momentum in a
88: plethora of astrophysical disk configurations.
89:
90: However, for a case of finite size system, the common practice in stability analysis
91: to expand the state vector of the linearized dynamical system
92: (the MHD system of equations in our case) in plane waves as:
93: ${\bf u}(z,t)={\bf A}e^{-i\omega t+ikz}$,
94: where $z$ is the spatial coordinate (for simplicity only one spatial
95: coordinate will be considered), $k$ is the wave number, and $\omega$
96: is the natural frequency of the system, is not applicable especially for
97: inhomogeneous systems. Instead, for such cases the appropriate boundary value
98: problem (BVP) for a given set of initial conditions (IC) should be solved
99: (see for example the well-known work of \cite{landaupol}). The solution of such
100: problems is facilitated by obtaining the natural frequencies of the bounded system
101: $\omega_n$ for which the state vector of the linearized dynamical system may be written as
102: ${\bf u_n}(z,t)={\bf A}_n(z)e^{-i\omega_n t}$, where ${\bf A}_n(z)$
103: are the eigenfunctions of the BVP subjected to specific boundary conditions (BC)
104: (see also \cite{sm,cc,ck}). The main purpose of the current work is to employ such
105: approach in order to investigate the dynamical evolution of localized initial
106: perturbations in rotating magnetized disks of finite thickness.
107:
108: In reality, small perturbations, especially in a system of finite
109: dimensions, do not have the form of a single monochromatic wave, but
110: are rather a superposition of individual waves, i.e., wave packets.
111: Furthermore, the asymptotic behavior at large times of the wave
112: packet may significantly differ from that of any one of its
113: individual components. Thus, under certain circumstance, even though
114: some of the components of the wave packet may individually grow with
115: time without bound, the wave packet as a whole can remain bounded at
116: a given place and even decay to zero as the packet is convected
117: away. In such cases the initial perturbations give rise to what is
118: defined as convective instabilities. In other cases, the
119: exponentially growing components of the wave packets may indeed
120: cause the perturbation to grow without bound at each place. In such
121: cases, the perturbations are defined as absolute instabilities.
122: Furthermore, the BC could significantly change the
123: results of the stability analysis (global stability). It is therefore of utmost importance
124: to investigate the asymptotic (in time) development of the solutions
125: of the appropriate Cauchy problems in order to determine whether the
126: system is stable or not.
127: %Ascertaining that some of the individual
128: %monochromatic waves grow without bound provides a necessary
129: %condition for instability but, as we shall see, certainly not a
130: %sufficient one.
131:
132:
133: Various sets of rules in order to distinguish between convective and
134: absolute instabilities have been given by
135: \cite{landau,sturrock,fks,akhiezer,lp,drazin}, while \cite{Huerre
136: Monkewitz} have reviewed more recent developments of the theory
137: pertaining to hydrodynamic stability. The importance of studying
138: the influence of a finite system size on stability analysis
139: has been demonstrated by \cite{budker,sturrock} who reported that the size of system
140: could play a stabilizing role for the two stream instability.
141:
142: %The importance of the
143: %distinction between convective and absolute instability has been
144: %demonstrated in the Riga dynamo experiment in which convective
145: %instabilities were converted into absolute instabilities by an
146: %appropriate redesign of the basic flow \cite{Gailitis}.
147: One should note also that such
148: understanding may play an important role in recent attempts to
149: observe the MRI in the laboratory \citep{Colgate,Rudiger,Goodman}.
150:
151: \section{The magneto rotational instability}
152:
153: \subsection{Formulation of the problem and formal solution}
154: The effect of the finite size of the system is investigated by re-examining
155: the MRI for the case of rotating disks with finite thickness.
156: The basic MHD equations that describe the dynamical development of
157: the system are:
158: \begin{equation}
159: \rho\frac{d\vec{V}}{dt}=-\vec{\nabla}P+\frac{1}{c}\vec{J}\times\vec{B}+\rho\vec{G},
160: \quad \vec{\nabla}\cdot\vec{V}=0,
161: \label{ind}
162: \end{equation}
163: \begin{equation}
164: \frac{\partial\vec{B}}{\partial t}=-c\vec{\nabla}\times\vec{E},
165: \quad \vec{\nabla}\times\vec{B}=\frac{4\pi}{c}\vec{J},
166: \quad \vec{\nabla}\cdot\vec{B}=0,
167: \label{mom}
168: \end{equation}
169: where $\vec{G}$ is acceleration due to gravity, and $c$ is speed of light and
170: the rest of the variables have their usual meanings.
171: The expression for electric field $\vec{E}$ in ideal plasmas is given by:
172: \begin{equation}
173: \vec{E}=-\frac{1}{c}\vec{V}\times\vec{B}.
174: \label{ohm}
175: \end{equation}
176: In order to simplify the
177: calculations (and with no loss of generality of the results) we
178: assume that the Brunt-V\"{a}is\"{a}l\"{a} frequency is small in
179: comparison to all other characteristic frequencies in the system and
180: can hence be neglected. Thus, the linearized MHD system of equations
181: that describe radially independent perturbations in a Keplerian disk
182: under the influence of a constant axial magnetic field, may be represented
183: in the following way:
184:
185: \begin{equation}
186: \frac{\partial {\bf u}}{\partial t}-P\frac{\partial {\bf
187: u}}{\partial z}+Q{\bf u}=0, \label{linearmhd}
188: \end{equation}
189: where ${\bf u}(z,t)= ( v_r, v_{\varphi}, b_r, b_{\varphi} )^T$,
190: \begin{equation}
191: P=\left [
192: \begin{array}{cccc}0& 0 &\frac{\rho_0}{\beta\rho(z)} &0
193: \\0 &0 &0& \frac{\rho_0}{\beta\rho(z)} \\ 1 &0& 0& 0 \\ 0& 1& 0& 0 \end{array} \right ],\;\;\;
194: Q=\left [ \begin{array}{cccc}0& -2 &0 &0
195: \\\frac{1}{2} &0 &0& 0 \\ 0 &0& 0& 0 \\ 0& 0& \frac{3}{2}& 0 \end{array} \right
196: ],
197: \end{equation}%
198:
199: where $\rho(z)$ is the steady state density profile, and $v_r, v_{\varphi}, b_r$, and
200: $b_{\varphi}$ are the perturbed
201: radial and azimuthal velocities ($ v_r, v_{\varphi}\ll \Omega r)$, and the radial and azimuthal
202: components of the magnetic field ($ b_r, b_{\varphi}\ll B_z)$, respectively. In eq.
203: (\ref{linearmhd}) the density is normalized to its value at the disk midplane $\rho_0=\rho(z=0)$,
204: the velocities are normalized to $\sqrt{2}c_s$ ($c_s$ is the sound velocity), the time is normalized
205: to the inverse steady state angular velocity $\Omega$, the magnetic field is scaled with the steady state
206: axial magnetic field $B_z$. Finally, lengths are scaled by $\sqrt{\beta}V_A/\Omega$ where
207: $V_A=B_z/\sqrt{4\pi\rho_0}$ is the Alfv\'{e}n velocity and $\beta$ is the familiar plasma
208: parameter given by $\beta=2c^2_s/V^2_A$.
209:
210: As a specific example consider a thin isothermal Keplerian disk.
211: In order to model the finite thickness of the disk the following profile
212: $$\rho(r,z)=\rho_0(r)\exp(-z^2)$$
213: is assumed for the density. Due to the independence of the perturbations on the radial direction
214: $r$ is from now on merely a parameter. This is tantamount to the local approximation in $r$ with $k_r=0$,
215: which according to \cite{bh} is the most unstable case. For that case, assuming
216: that the IC are ${\bf u}(z,0) = ( v_r(z,0), 0, 0, 0)$,
217: and employing the Laplace transform, the set of eqs.(\ref{linearmhd})
218: is reduced to the following single inhomogeneous ordinary differential equation (ODE):
219: \begin{equation}
220: L[\widetilde{b}_{\varphi}]=2i\omega\beta^2 v'_r(z,0)
221: \label{bphieinh}
222: \end{equation}
223: where
224:
225: \begin{equation}
226: L[\widetilde{b}_{\varphi}]=\frac{d}{dz}\left[e^{z^2}\frac{d^2}{dz^2}\left(e^{z^2}\frac{d\widetilde{b}_{\varphi}}{dz}\right)\right]+
227: (3\beta+2\lambda^2)\frac{d}{dz}\left(e^{z^2}\frac{d\widetilde{b}_{\varphi}}{dz}\right)
228: +\lambda^2(\lambda^2-\beta)\widetilde{b}_{\varphi},
229: %\label{bphieinh}
230: \end{equation}
231:
232: $\lambda=\sqrt{\beta}\omega$ represents the re-normalized spectral parameter $\omega$, and
233: $\widetilde{b}_{\varphi}$ is the Laplace transform of the azimuthal component of the perturbed
234: magnetic field. The solution of eq.(\ref{bphieinh}) should satisfy the following BC:
235:
236: \begin{equation}
237: \widetilde{b}_{\varphi}(\pm\infty)=0, \quad
238: (e^{z^2}\widetilde{b}'_{\varphi})'_{|_{z=\pm\infty}}=-2i\sqrt{\beta}\lambda\widetilde{b}_r(\pm\infty)-
239: \lambda^2\widetilde{b}_{\varphi}(\pm\infty) = 0.
240: \label{mhdbcinh}
241: \end{equation}
242:
243: Boundary conditions (\ref{mhdbcinh}) are obtained due to the requirement that at $z\rightarrow\pm\infty$,
244: where $\rho\rightarrow 0$ and $B_z=const$, the energy flux of the perturbation is finite.
245: %One should note here again that due to the density inhomogeneity the eigenfunctions of this
246: %problem are not plane waves $exp[i(\omega t-kz)]$.
247:
248: The solutions of the BVP (\ref{bphieinh})-(\ref{mhdbcinh}) can be constructed
249: by the countable set of the solutions of the homogeneous equation $L[\widetilde{b}_{\varphi}]=0$,
250: subject to homogeneous boundary conditions. The solutions of that problem are termed eigenfunctions
251: of the homogeneous equation.
252: Assuming completeness and orthogonal property of the eigenfunctions allow to express the solution
253: of eq.(\ref{bphieinh}) in a simple way for an arbitrary IC by using Green's function (see appendixes
254: A and B for proof of orthogonality, and construction of the Green's function). The latter
255: is represented by an appropriate set of eigenfunction $\widetilde{b}_{\varphi, n}(z)$. Thus,
256: carrying out the inverse Laplace transform, the solution of eq.(\ref{linearmhd}) for $t\geq 0$
257: is given by the following expression:
258:
259: \begin{equation}
260: b_{\varphi}(z,t)=4\sum_nd_n\frac{\widetilde{b}_{\varphi, n}(z)}{\omega^2_{-}(n)-\omega^2_{+}(n)}
261: \sin(\frac{\omega_{+}(n)-\omega_{-}(n)}{2}t)\sin(\frac{\omega_{+}(n)+\omega_{-}(n)}{2}t)
262: \label{disksol}
263: \end{equation}
264:
265: where $\widetilde{b}_{\varphi, n}(z)$ are the eigenfunctions of the homogeneous equation,
266: $d_n$ are the coefficients of the expansion of $v'_r(z,0)$
267: in the complete set $\widetilde{b}_{\varphi, n}(z)$, and $\omega_\pm(n)$
268: are the natural frequencies that correspond to the eigenfunctions $\widetilde{b}_{\varphi, n}(z)$.
269:
270:
271: %We start to solve the BVP (\ref{bphieinh})-(\ref{mhdbcinh})
272: %by looking for asymptotic solutions of eq.(\ref{bphieinh})
273: %for $\lambda\gg1$. First we find the
274: %eigenfunctions of the homogenous part of the ODE given by eq.(\ref{bphieinh}).
275: %Under the assumption $\lambda\gg1$
276: \subsection{WKB solution}
277:
278: We turn now to obtaining asymptotic solutions for $\widetilde{b}_{\varphi, n}(z)$ in the
279: limit $\lambda\gg1$ [see \citep{em} as an example of employing the WKB approximation for some problems
280: in inhomogeneous plasmas]. In that case the eigenfunctions may be represented by
281: the following version of the WKB approximation:
282:
283: \begin{equation}
284: \widetilde{b}_{\varphi,k}=e^{-z^2}\phi_k(z)=e^{-z^2}e^{S_k(z)},\quad k=1,\ldots,4.
285: \label{asympt}
286: \end{equation}
287:
288: Inserting expression (\ref{asympt}) into the homogenous part of eq.(\ref{bphieinh}) yields:
289:
290: \begin{equation}
291: S'^4+\lambda^2\left\{3P(z,\lambda)S'^2+2Q(z,\lambda)S'+R(z,\lambda)+3S''\left[P(z,\lambda)+
292: \frac{2S'^2+S''}{\lambda^2}\right]\right\}+4S'S'''+S^{IV}=0
293: \label{eikonal}
294: \end{equation}
295:
296: where
297:
298: $$P(z,\lambda)=\frac{1}{3}\left[e^{-z^2}(2+3\beta/\lambda^2)-(6+4z^2)/\lambda^2\right],\quad Q(z,\lambda)=
299: -z\left[e^{-z^2}(2+3\beta/\lambda^2)+4/\lambda^2\right]$$
300: and
301: $$R(z,\lambda)=e^{-2z^2}(\lambda^2-\beta)-2e^{-z^2}(2+3\beta/\lambda^2).$$
302: By keeping the leading terms in $\lambda$, eq.(\ref{eikonal}) can be written as
303:
304: \begin{equation}
305: S'^4+\lambda^2\left[3P(z,\lambda)S'^2+2Q(z,\lambda)S'+R(z,\lambda)\right]=0
306: \label{eikonal0}
307: \end{equation}
308:
309: The solutions of eq.(\ref{eikonal0}) provide valid approximations for the eigenfunctions
310: of $L$ throughout the disk except in the vicinity of the turning points.
311: Assuming that the latter occur at $|z_0|\gg 1$, it can be shown that the last term
312: in eq.(\ref{eikonal0}) becomes exponentially small in comparison with other terms
313: (this assumption will be validated later on). In that case the equation that determines the turning points
314: is given by
315: \begin{equation}
316: \frac{\textsf{D}(z_0,\lambda)}{\lambda^4}=\lambda^2P^3(z_0,\lambda)+Q^2(z_0,\lambda)=0.
317: \label{discr}
318: \end{equation}
319: Asymptotic solution of eq.(\ref{discr}) for large $\lambda$ yields the following expression for the turning points:
320: \begin{equation}
321: z_0=\pm\sqrt{\ln \frac{2\lambda^2}{\ln 2\lambda^2}} + O\left(z_0^{-3}\right),
322: \end{equation}
323: as well as the following form of the discriminant in the vicinity of the turning points:
324: \begin{equation}
325: \textsf{D}(z,\lambda)\approx K(z-z_0), \quad K=\frac{d D(z,\lambda)}{dz}|_{z=z_0}
326: \approx-6z^7_0(1+5z^{-2}_0)
327: \label{expan}
328: \end{equation}
329: In the outer regions namely $z < -z_0$ and $z > z_0$
330: all four functions $S_k(z)$ are real, whereas in the inner region $-z_0 < z < z_0$ the solutions for $S_k(z)$
331: acquire non zero imaginary part. Thus, the turning points separate
332: the inner range $-z_0 < z < z_0$ where the eigenfunctions are oscillatory from the outer regions $z < -z_0$
333: and $z > z_0$ where the eigenfunctions decrease exponentially away from $z_0$ and
334: have no zeros at a finite distance $z$. The behavior of the various solutions of eq.(\ref{eikonal0})
335: as well as the turning points may be seen in Fig.1.
336:
337: \begin{figure*}
338: %\vspace*{174pt}
339: \includegraphics[width=120mm]{fig1.eps}
340: \caption{The solutions of eq.(\ref{eikonal0}), obtained numerically
341: for $\beta=10,\lambda=25$. Each of the dashed lines ($A,B$) represents the same real part of two complex
342: conjugate solutions.}
343: \label{fig1}
344: \end{figure*}
345:
346: It is obvious from assumption (\ref{asympt}) and BC (\ref{mhdbcinh}) that only solution 1 is admissible.
347: The turning points are obtained as the merging of solution 1 and 2. In the region between the turning points
348: two solutions are admissible, that are complex conjugates of each other. Again, due to BC (\ref{mhdbcinh})
349: those solutions are marked by A that depicts their common real part.
350:
351: In the vicinity of the turning points
352: the solutions assumed in eq.(\ref{asympt}) are invalid and instead an appropriately re-scaled eq.(\ref{bphieinh})
353: is solved. That solution is then matched asymptotically to the outer solution given
354: by (\ref{asympt}) and (\ref{eikonal0}) where $S'$ is obtained from branch 1 in Fig.1.
355:
356: In order to obtain
357: an asymptotic expression for $S'$ in branch 1 it is noticed again that $R(z,\lambda)$ is asymptotically
358: much smaller than the rest of the terms that multiply $\lambda$ in eq.(\ref{eikonal0}). As a result,
359: one of the roots is asymptotically zero (branch 3) while the rest three roots are obtained from
360: the Cardano solution of the reduced eq.(\ref{eikonal0}). Thus, the root that is represented by branch 1 (the only
361: admissible root) gives rise to the following solution in the outer region close to the turning point:
362: \begin{equation}
363: \phi(z)=\frac{C}{2\sqrt[4]{|K|}}\frac{1}{(z-z_0)^{\frac{1}{4}}}\exp{[-\int_{z_0}^z\sqrt{K(z_0-\xi)}d\xi]}
364: \label{right}
365: \end{equation}
366: In the inner region the solution near the turning points is given by
367: \begin{equation}
368: \frac{1}{\sqrt[4]{|K|}(z_0-z)^{\frac{1}{4}}}\left\{C_1\exp{[i\int_{z_0}^z\sqrt{K(\xi-z_0)}d\xi]}+
369: C_2\exp{[-i\int_{z_0}^z\sqrt{K(\xi-z_0)}d\xi]}\right\}.
370: \label{left}
371: \end{equation}
372: The coefficients ($C_{1,2}$) are determined by the connection formulas that express the asymptotic matching
373: of the solution found to the left (\ref{left}) and to the right (\ref{right}) of a turning point $z_0$. Following
374: \cite{landauQM} the connecting relations are:
375: $$C_{1,2}=\frac{C}{2}\exp{(\pm i\frac{\pi}{4})}.$$
376: Applying the same rule to the region close
377: to the left turning point located at $-z_0$ and requireing that the two expressions are the same
378: throughout the region $-z_0<z<z_0$ (the sum of their phase must be multiple of $\pi$)
379: results in the following expression for the eigenfunction:
380: \begin{equation}
381: \widetilde{b}_{\varphi, n}(z,\omega_\pm(\beta,n))=\cos\left[\kappa\Phi(z)-\frac{\pi}{4}\right]
382: e^{-\frac{z^2}{4}},
383: \label{wkbsol}
384: \end{equation}
385: where $\Phi(z)=\int^{z_0}_z e^{-\frac{\zeta^2}{2}}d\zeta$. The analytical expression (\ref{wkbsol}) for the solution
386: in the inner region has been obtained by noticing that away from the turning points $S'Q(z,\lambda)$ is smaller
387: than the rest of the terms in eq.(\ref{eikonal0}) and hence to leading order the latter is a biquadratic equation.
388: The discrete set of eigenvalues $\omega$ is now determined from the Bohr-Sommerfeld relation that is given by:
389: \begin{equation}
390: \kappa=\sqrt{\frac{\beta}{2}}\sqrt{3+2\omega_\pm(\beta,n)^2+\sqrt{9+16\omega_\pm(\beta,n)^2}}=
391: \frac{\pi}{\Phi(-z_0)}\left(n+\frac{1}{2}\right),
392: \label{BZ}
393: \end{equation}
394: The integer $n$ in the last expression describes the number of zeros of the solution
395: within the inner region [see eq.(\ref{wkbsol})], whereas $\omega_\pm(\beta,n)$ are the natural frequencies of the system
396: that correspond to such eigenfunctions. Strictly speaking expression (\ref{BZ}) is valid for $n >> 1$.
397: Nevertheless, as is often the case for WKB-type solutions, numerical calculations
398: indicate that expression (\ref{wkbsol}) provides a close approximation also for $n$'s as low as 1 (see Fig.2).
399: It is also evident from Fig.2 that the wave length of the perturbation depends on $z$ and approaches the disk thickness
400: close to the turning points.
401:
402:
403: \begin{figure*}
404: %\vspace*{174pt}
405: \includegraphics[width=74mm]{fig2a.eps}(a)
406: \includegraphics[width=74mm]{fig2b.eps}(b)
407: \caption{Comparison between the WKB solutions [eq.(\ref{wkbsol}), dashed line] and the numerical solution (full line)
408: of the homogeneous part of eq.(\ref{bphieinh}); (a) for the first three eigenfunctions ($n=0,1,2$) and
409: (b) - $n=37$.}
410: \label{fig2}
411: \end{figure*}
412:
413: As $n=0$ is the first excited unstable mode, it is easy to show that the disk is stable for $\beta\leq\beta_*$
414: where $\beta_*=\pi/(24[Erf(\frac{z_0}{\sqrt{2}})]^2)$ ($n=1,\beta_*=3\pi/(8[Erf(\frac{z_0}{\sqrt{2}})]^2)$).
415:
416: \subsection{Numerical solution}
417:
418: To further follow the development of a localized initial perturbation governed by eq.~(\ref{linearmhd}),
419: consider for example an IC given as:
420: \begin{equation}
421: v_r(z,0)=e^{-z^2/\Delta ^2},
422: \label{gauss}
423: \end{equation}
424: and zero for the rest of the initial values of the perturbed physical variables.
425: It is easy to see that the perturbation is localized within the disk,
426: if $\Delta << 1$. The dynamical development of the localized initial perturbation that
427: is described by eq.(\ref{disksol}) depends on the disk "thickness" defined by the plasma
428: parameter $\beta$ (notice that now velocities are scaled by the Alfv\'{e}n velocity which means that the disk thickness is $2\sqrt{\beta}$ ).
429: If the disk is thin enough such that $\beta \leq \beta_*$ all the natural frequencies $\omega_\pm(n)$
430: that are obtained from eq.~(\ref{bphieinh}) are real and hence the disk is stable. In that case following
431: (\ref{disksol}) the profile of the perturbed azimuthal magnetic field represented by
432: two identical wave packets that move in opposite directions namely up and down. Upon reaching the upper
433: and lower turning points (located at $\pm z_0$) the two packets are reflected and continue their motion through
434: the disk. The amplitudes of the identical wave packets change in time like $\propto \sin(t)$. It should be noted
435: however that this picture is valid if the initial perturbation is localized enough. As example for such behavior,
436: $b_{\varphi}(z,t)$, which has been obtained by using eq.(\ref{disksol})
437: $\Delta =0.1$, and $\beta=3\pi/8$
438: is depicted in Fig. \ref{fig3}. In this particular case all eigenfunctions have real
439: natural frequencies [$\omega_\pm(1)=1.54,3.64, \omega_\pm(3)=3.99,6.01, ...$]
440: and there are no growing modes at all (the wave packets indeed
441: do not grow in time, just move back and forth inside the disk with local Alfv\'{e}n velocity).
442:
443: \begin{figure*}
444: %\vspace*{174pt}
445: \includegraphics[width=120mm]{fig3.eps}
446: \caption{Time evolution of the profile of the perturbed azimuthal component of the
447: magnetic field. The different profiles were calculated at times
448: [0.3, 1.0, 1.5] for $\beta=3\pi/8$, $\Delta=0.1$ .}
449: \label{fig3}
450: \end{figure*}
451:
452: For "thick" disk ($\beta > \beta_*$) some of the natural frequencies have an imaginary part. As the thickness
453: of the disk increases so does the number of unstable modes.
454: Thus for long time $b_{\varphi}(z,t)\sim\widetilde{b}_{\varphi, n}(z)exp(\mid\Im\omega_{-}(n)\mid t)$
455: where $n$ corresponds to natural frequency with a biggest imaginary part. However,
456: for sufficiently localized initial perturbation the
457: fastest growing mode has sufficiently small initial amplitude. The perturbation then
458: keeps it's original shape for a long time until a growing mode emerges out of the wide initial spectrum
459: and is of order of the amplitude of initial wave packet. Raising the value of $\beta$ to $10$, one unstable mode,
460: with growth rate $\gamma=0.749$ enters the spectrum of the eigenmodes. Employing again eq.(\ref{disksol}) together with
461: eigenfunctions $\widetilde{b}_{\varphi, n}(z)$ results in Fig.4. It is indeed seen that
462: after long time the unstable mode dominates the perturbation. Thus, the latter eventhough starting as a well
463: confined perturbation develops into a global instability. It should be noted however, that the time it takes
464: the unstable mode to emerge from the wide spectrum that makes up the localized initial perturbation is about
465: 4 times longer than the predicted growth time (inverse growth rate). Indeed, as the initial perturbation
466: is more localized so the growth time of the most unstable mode is increased relative to its linear predicted value.
467:
468: \begin{figure*}
469: %\vspace*{174pt}[!ht]
470: \includegraphics[width=120mm]{fig4.eps}
471: \caption{Long time evolution of the profile of the perturbed azimuthal component of the
472: magnetic field for $\beta=10$, $\Delta=0.05$. The different profiles were calculated at times
473: [0.3, 1.0, 6.0].}
474: \label{fig4}
475: \end{figure*}
476: %\begin{figure}[!ht]
477: %\epsscale{.80}
478: %\plotone{fig4_05.eps} \caption{Long time evolution of the profile of the perturbed azimuthal component of the
479: %magnetic field for $\beta=10$, $\Delta=0.05$. The different profiles were calculated at time
480: %[0.3, 1.0, 6.0]
481: %\label{fig4}}
482: %\end{figure}
483: It is finally instructive to plot the wave number $n^*$ of the most unstable mode as a function of $\beta$ (Fig.5).
484: Asymptotic estimation reveals that $n^*$ is proportional to $\sqrt{\beta}$. This implies
485: that the wavelength of the most unstable mode is of the order of $h/n^* \sim V_A/\Omega$ (where $h$
486: is the thickness of the disk) which to leading order does not depend on $\beta$.
487:
488: \begin{figure*}
489: %\vspace*{174pt}
490: \includegraphics[width=120mm]{fig5.eps}
491: \caption{Axial standing wave number of the most unstable
492: mode as a function of the plasma beta.}
493: \label{fig5b}
494: \end{figure*}
495: %\begin{figure}[!ht]
496: %\epsscale{.80}
497: %\plottwo{fig5_a.eps}{fig5_b.eps}\caption{Growth rate $\gamma$ and number of zeros for the most unstable
498: %standing wave vs plasma beta parameter.
499: %\label{fig5}
500: %}
501: %\end{figure}
502:
503: \subsection{Limit of infinite "thickness"}
504: It is instructive to examine the limit of thick disk such that effect of boundaries may be ignored.
505: Such an assumption is justified when the Alfv\'{e}n traverse time is longer than the time of interest
506: (say inverse growth rate). Renormalizing therefore lengths to $V_A/\Omega$ solution is sought for
507: $z\ll\sqrt{\beta}$, where $2\sqrt{\beta}$ is now the disk's thickness. It is readily seen that for such case
508: $\widetilde{b}_{\varphi, n}(z)\rightarrow \widetilde{b}_\varphi(k,z)=\cos kz$ for odd initial perturbations
509: [see eq.(\ref{wkbsol}) where $k$ is defined as $\sqrt{\pi/2\beta}(n+1/2)$] so that the eigenfunctions
510: can be approximated by plane waves. Thus, due to $1/\sqrt{\beta}\rightarrow0$, the sum (\ref{disksol})
511: is replaced by an integral over $k$ and eq.(\ref{disksol}) can be written as
512: \begin{equation}
513: b_{\varphi}(z,t)\rightarrow
514: \frac{1}{4\pi^2}\int_Ce^{-i \omega t}d\omega\int_{-\infty}^{\infty}\frac{{\bf A}(k)e^{i kz}dk}{D(k,\omega)}
515: \label{aa}
516: \end{equation}
517: where $C$ is the contour of integration in the complex $\omega$ plane that is a straight line parallel to the real
518: axis and passing above all singular points of the integrand, ${\bf A}(k)$ is the Fourier transform of the right hand side of
519: eq. (\ref{bphieinh}) and the denominator of the integrand is given by
520: \begin{equation}
521: D(\omega ,k)=(\omega ^2-k^2)^2-\omega ^2-3k^2. \label{mri}
522: \end{equation}
523: It is important to notice now that since the initial perturbation is well localized in space ${\bf A}(k)$
524: in contrast is close to a constant that is proportional to the localization length.
525: As a result it can be shown now that $b_{\varphi}(z,t)$ obtained by using (\ref{aa}) is not zero only if the
526: integrand has a singularities of order two (branch points) (see in details in \cite{fks,akhiezer}).
527: Thus, examining the equation $D(\omega ,k)=0$ [the classical MRI dispersion equation \citep{bh}]
528: it is clear that there is just one branch point
529: that gives rise to instability, which is $\omega=3i/4,k=\sqrt{15}/4$. Consequently, the long time
530: behavior of the perturbation is given by
531: \begin{equation}
532: b_{\varphi}(t)\sim e^{3t/4}/\sqrt{t}.
533: \label{longtime}
534: \end{equation}
535: As remarked above this result is due to the fact that the initial perturbation is localized in space.
536: If, however, (a not too physical) monochromatic perturbation is considered, ${\bf A}(k)$ is a delta
537: function and the familiar purely exponential growth is recovered.
538:
539: \section{Conclusions}
540: The importance of solving the initial value problem with some appropriate boundary conditions
541: is highlighted.
542: In the classical works of \cite{velikhov,chandra}, and \cite{bh},
543: an infinite cylinder has been considered and the effects of the boundary conditions were neglected.
544: Therefore, in those works, naturally, the thickness of the disk does not play any role and
545: consequently cannot influence the extent of the domains of instability. In spite of that, physical
546: intuition and insight have led \cite{bh} to conclude that $\beta\approx 1$ is the lower limit for
547: the disk thickness for the occurrence of the classical MRI in disks as for smaller values of $\beta$
548: the thickness of the disk is smaller than the wave length of the dominant unstable mode.
549: In the current work the stabilizing effects of the boundaries are taken explicitly into account and hence
550: the threshold thickness of the disk may be calculated easily from the WKB solutions obtained in Sec.2.
551: Furthermore, the number of unstable modes as a function of $\beta $ may be estimated with the aid of that solution.
552: Thus, for example, it may be shown that there are only three unstable modes within the disk as long as the its beta value is less than $ \sim 15 $.
553: Such information is significant for the study of the consequent nonlinear development of the instability.
554: In addition, the shape of the unstable (as well as the stable) perturbations has been obtained explicitly
555: and as may be seen by its expression (\ref{wkbsol}) may differ significantly from the plain waves
556: assumed in the classical works on MRI's. It is finally interesting to note, that going to
557: the limit of thick disks (see Section 2.4) and assuming a localized perturbation results in a
558: reduced rate of growth in comparison to the less realistic monochromatic classical result [see Eq.(\ref{longtime})].
559: Recently, \cite{cc}, and \cite{ck} have considered the effects of the axial localization of the
560: perturbations by studying axisymmetric ballooning modes in finite disks. Such modes are characterized by
561: finite values of the radial wave vector $k_r$. The appropriate turning points within the disk were
562: found and consequently a discrete spectrum of eigenfunctions and eigenvalues was obtained.
563: The growth rates were found to be smaller than their "long-cylinder" counterparts. In particular,
564: it is shown there that the $k_r\rightarrow 0$ limit cannot be obtained from their scalings and asymptotics,
565: as no turning points exist within the disk in that limit. In that sense the present work is complementary to
566: \citep{ck} as it analyzes that very limit $k_r\rightarrow 0$. Indeed, under the current
567: scaling and asymptotic expansion turning points are found within the disk and the corresponding
568: discrete spectrum is obtained.
569:
570: The linear MHD equations have been employed in order to study the
571: stability and time behavior of a rotating stratified Keplerian disks whose
572: density decreases with height. The full solution of the dynamical evolution in time of an localized Gaussian
573: wave packet is explicitly derived and it's long
574: time behavior is discussed. It is proven analytically (WKB) that MRI can be suppressed
575: in sufficiently thin disk $\beta < \beta_*\approx \pi/11$ (\cite{ms} observed from
576: numerical simulations that if $\beta\approx 1$ the disk becomes MRI stable). Numerical solutions
577: of the Cauchy problem however, indicates that $\beta_*$ is of order 1. This result is consistent
578: with values obtained by \cite{sm}.
579: For thicker disks, the number of discrete unstable mode
580: increases with the thickness of the disk.
581: However, due to the localization in real space the initial amplitude of
582: the unstable mode is diminishingly small and the time it outgrows the original wave packet is
583: significantly longer than its predicted inverse growth rate. Thus, it has been demonstrated
584: numerically the growth time of a perturbation that is initially localized within the inner 5\%
585: of the disk, may be an order of magnitude longer than the inverse growth rate of the fastest
586: growing unstable mode.
587:
588: The considering of the boundary effects may have a large impact on the design
589: of laboratory experiments to model MRI, where a magnetic field have to be
590: quite strong and devices cannot be very high.
591:
592:
593: \section*{Acknowledgments}
594:
595: The authors are greatly indebted to Oded Regev and Orkan Umurhan for fruitful discussions and insights,
596: as well as for their encouragement.
597:
598: \begin{thebibliography}{99}
599:
600: \bibitem[\protect\citeauthoryear{Akhiezer \& Polovin}{1971}]{akhiezer}
601: Akhiezer, A.I., Polovin, R.V. 1971, Usp.Fiz.Nauk, {\bf
602: 104}, 185 [1971, Soviet Physics Uspekhi, {\bf
603: 104}, 278]
604:
605: \bibitem[\protect\citeauthoryear{Balbus \& Hawley}{1991}]{bh}
606: Balbus, S.A., Hawley, J.F., 1991, ApJ, {\bf376}, 214
607:
608: \bibitem[\protect\citeauthoryear{Budker}{1956}]{budker}
609: Budker, G.I., 1956, Atomic Energy, {\bf 1}, issue 5, 673
610:
611: \bibitem[\protect\citeauthoryear{Chandrasekhar}{1960}]{chandra}
612: Chandrasekhar, S. 1960, Proc. Nat. Acad. Sci., A{\bf 46}, 223
613:
614:
615: \bibitem[\protect\citeauthoryear{Coppi \& Coppi}{2001}]{cc}
616: Coppi, B., Coppi, P. S., 2001, Phys. Rev. Lett., 87, 051101
617:
618: \bibitem[\protect\citeauthoryear{Coppi \& Keyes}{2003}]{ck}
619: Coppi, B.; Keyes, E. A., 2003, ApJ, 595, 1000
620:
621: \bibitem[\protect\citeauthoryear{Drazin}{2002}]{drazin}
622: Drazin, P.G. 2002, \emph{Introduction to Hydrodynamic Stability},
623: (Cambridge: University Press)
624:
625: \bibitem[\protect\citeauthoryear{Erokhin \& Moiseev}{1973}]{em}
626: Erokhin, N.S., Moiseev, S.S 1973, \emph{Voprosy teorii plazmy}, v.7, p.146
627: (Mosow: Atomizdat, 1973) [\emph{Reviews of plasma physics},
628: Edited by Acad. M.A. Leontovich, v.7, 181, (New York: Consultants Bureau, 1979)]
629:
630: \bibitem[\protect\citeauthoryear{Fainberg et al.}{1961}]{fks}
631: Fainberg, Ya.B., Kurilko, V.I., \& Shapiro, V.D. 1961, Zhurnal
632: Tekhnicheskoi Fiziki, {\bf 31}, 633 [1961, \emph{Soviet
633: Physics-Technical Physics}, {\bf 6}, 459].
634:
635: \bibitem[\protect\citeauthoryear{Gailitis et al.}{2004}]{Gailitis}
636: Gailitis, A., Lielausis, O., Platacis, E., Gerbeth, G., \& Stefani, F.
637: 2004, Phys. Plasmas, {\bf 11}, 2838
638:
639: \bibitem[\protect\citeauthoryear{Huerre \& Monkewitz}{1990}]{Huerre Monkewitz}
640: Huerre, P., Monkewitz, P. A. 1990, Annu. Rev. Fluid Mech.,
641: {\bf 22}, 473
642:
643: \bibitem[\protect\citeauthoryear{Ji et al.}{2006}]{Goodman}
644: Ji, H.T., Burin, M., Schartman, E., \& Goodman, J. 2006, Nat, {\bf 444},
645: 343
646:
647: \bibitem[\protect\citeauthoryear{Landau}{1946}]{landaupol}
648: Landau, L.D., 1946, Zh. Eksp. Teor. Fiz., {\bf 16}, 574
649: [1946, {\em J. Phys. USSR}, {\bf 10},25]
650:
651: \bibitem[\protect\citeauthoryear{Landau \& Lifshitz}{1959}]{landau}
652: Landau, L.D., Lifshitz, E.M. 1959, \emph{Fluid Mechanics}, (Oxford:
653: Pergamon Press)
654:
655: \bibitem[\protect\citeauthoryear{Landau \& Lifshitz}{1977}]{landauQM}
656: Landau, L.D., Lifshitz, E.M. 1977, \emph{Quantum mechanics :
657: non-relativistic theory}, (Oxford: Pergamon Press)
658:
659: \bibitem[\protect\citeauthoryear{Lifshitz \& Pitaevskii}{1981}]{lp}
660: Lifshitz, E.M., Pitaevskii, L.P. 1981, \emph{Physical kinetics}, (Oxford: Pergamon Press)
661:
662: \bibitem[\protect\citeauthoryear{Miller \& Stone}{2000}]{ms}
663: Miller \& Stone, 2000, ApJ, 534, 398
664:
665: \bibitem[\protect\citeauthoryear{Noguchi et al.}{2002}]{Colgate}
666: Noguchi, K., Pariev, V.I., Colgate, S.A., Beckley, H.F., \&
667: Nordhaus, J. 2002, ApJ, {\bf 575}, 1151
668:
669: \bibitem[\protect\citeauthoryear{Rudiger et al.}{2006}]{Rudiger}
670: R\"{u}diger, G., Hollerbach, R., Stefani, F., Gundrum, T., Gerbeth, G.,
671: \& Rosner, R. 2006, ApJ, {\bf 649}, L145
672:
673: \bibitem[\protect\citeauthoryear{Sano \& Miyama}{1999}]{sm}
674: Sano T., Miyama S. M., 1999, ApJ, 515, 776
675:
676: \bibitem[\protect\citeauthoryear{Sturrock}{1958}]{sturrock}
677: Sturrock, P. A. 1958, Phys.Rev. A, {\bf 112}, 1488
678:
679: \bibitem[\protect\citeauthoryear{Velikhov}{1959}]{velikhov}
680: Velikhov, E.P. 1959, Zh. Eksp. Teor. Fiz., {\bf 36}, 1398
681: [1959, {\em Sov.Phys. JETP}, {\bf 9},995]
682:
683: %\bibitem[\protect\citeauthoryear{Baird}{1981}]{b1} Baird S.R., 1981,
684: %ApJ, 245, 208
685:
686: \end{thebibliography}
687:
688: \appendix
689:
690: \section[]{Orthogonality}
691: Both sides of the homogenous equation $L[\widetilde{b}_{\varphi, n}(z)]=0$ are first multiplied by
692: $\widetilde{b}_{\varphi, m}(z)$ and then integrated over the interval $(-\infty,\infty)$.
693: Integrating by parts and setting the boundary terms to zero due the homogeneous BC (\ref{mhdbcinh})
694: yields
695: \begin{equation}
696: \int^\infty_{-\infty}\widetilde{b}_{\varphi, n}\left[\left(e^{z^2}\left(e^{z^2}
697: \widetilde{b}'_{\varphi,m}\right)''\right)'+3\beta\left(e^{z^2}\widetilde{b}'_{\varphi, m}\right)'\right]dz + 2\lambda^2_n
698: \int^\infty_{-\infty}\widetilde{b}_{\varphi, m}\left(e^{z^2}\widetilde{b}'_{\varphi, n}\right)'dz +
699: \lambda^2_n(\lambda^2_n-\beta)\int^\infty_{-\infty}\widetilde{b}_{\varphi, m}\widetilde{b}_{\varphi,n}dz=0
700: \label{a1}
701: \end{equation}
702: Defining
703: $$
704: Q_{1,m,n}\equiv\int^\infty_{-\infty}\widetilde{b}_{\varphi, m}\widetilde{b}_{\varphi, n}dz, \quad
705: Q_{2,m,n}\equiv\int^\infty_{-\infty}\widetilde{b}_{\varphi,m}\left(e^{z^2}\widetilde{b}'_{\varphi,n}\right)'dz=
706: \int^\infty_{-\infty}\widetilde{b}_{\varphi, n}\left(e^{z^2}\widetilde{b}'_{\varphi, m}\right)'dz,
707: $$
708: and noticing that to each eigenfunction $\widetilde{b}_{\varphi, n}$ correspond two eigenvalues $\lambda^2_{n,1,2}$
709: allows rewriting (\ref{a1}) in the form of the following set of homogenous linear equations:
710: \begin{equation}
711: A{\bf Q}=0, \quad {\bf Q}\equiv (Q_{1,m,n}, Q_{2,m,n})^T, \quad A\equiv\left (\begin{array}{cc}
712: \lambda^2_{n,1}(\lambda^2_{n,1}-\beta)-\lambda^2_{m}(\lambda^2_{m}-\beta) & 2(\lambda^2_{n,1}-\lambda^2_{m})
713: \\
714: \lambda^2_{n,2}(\lambda^2_{n,2}-\beta)-\lambda^2_{m}(\lambda^2_{m}-\beta)&2(\lambda^2_{n,2}-\lambda^2_{m})
715: \end{array} \right )
716: \end{equation}
717: Implying that different eigenfunctions ($n\neq m$) have different eigenvalues results in:
718: \begin{equation}
719: det(A)=2(\lambda^2_{n,1}-\lambda^2_{n,2})(\lambda^2_{n,1}-\lambda^2_{m})(\lambda^2_{n,2}-\lambda^2_{m})\neq 0.
720: \end{equation}%
721: This implies that
722: \begin{equation}
723: Q_{1,m,n}=Q_{2,m,n}=0.
724: \label{a4}
725: \end{equation}
726: In the opposite case of identical eigenfunctions ($m=n$) the equality $det(A)=0$ means that $Q_{1,n,n}\neq 0$.
727: Combining the latter with eq.(\ref{a4}) gives
728: \begin{equation}
729: \langle\widetilde{b}_{\varphi,m},\widetilde{b}_{\varphi,n}\rangle\equiv Q_{1,m,n}\equiv\int^\infty_{-\infty}\widetilde{b}_{\varphi, m}
730: \widetilde{b}_{\varphi,n}dz=\|\widetilde{b}_{\varphi,n}\|^2\delta_{m,n}.
731: \end{equation}%
732:
733: \section[]{Green's function}
734: To obtain the expansion coefficients $c_n$ of a solution $\widetilde{b}_{\varphi}(z)$ for the inhomogeneous
735: equation $L[\widetilde{b}_{\varphi}(z)]=-f(z)$ in terms of the eigenfunctions $\widetilde{b}_{\varphi,n}(z)$
736: the solution expansion $\widetilde{b}_{\varphi}(z)=\sum_l c_l \widetilde{b}_{\varphi,l}(z)$ and expansion of
737: the r.h.s of the inhomogeneous equation $f(z)=\sum_m d_m \widetilde{b}_{\varphi,m}(z)$ is substituted into
738: the inhomogeneous equation. The result is multiplied by $\widetilde{b}_{\varphi, n}(z)$ and integrated over the
739: interval $(-\infty,\infty)$. The result is
740: \begin{equation}
741: \sum_lc_l\int^\infty_{-\infty}\widetilde{b}_{\varphi,n}L[\widetilde{b}_{\varphi,l}]=-d_n\|\widetilde{b}_{\varphi,n}\|^2.
742: \label{b1}
743: \end{equation}%
744: Defining in addition to $Q_{1,n,l}$ and $Q_{2,n,l}$
745: $$
746: Q_{3,n,l}=\int^\infty_{-\infty}\widetilde{b}_{\varphi,n}\left(e^{z^2}\left(e^{z^2}
747: \widetilde{b}'_{\varphi,l}\right)''\right)'dz,
748: $$
749: and employing the homogenous equation $L[\widetilde{b}_{\varphi,l}]=0$ multiplied by $\widetilde{b}_{\varphi, n}(z)$
750: and integrated over the interval $(-\infty,\infty)$ yields:
751: \begin{equation}
752: Q_{1,n,l}=\|\widetilde{b}_{\varphi,l}\|^2\delta_{n,l}, \quad
753: Q_{2,n,l}=-\frac{\|\widetilde{b}_{\varphi,l}\|^2}{2}(\lambda^2_{l,1}+\lambda^2_{l,2}-\beta)\delta_{n,l}, \quad
754: Q_{3,n,l}=\frac{\|\widetilde{b}_{\varphi,l}\|^2}{2}\left[3\beta(\lambda^2_{l,1}-
755: \beta)+\lambda^2_{l,2}(2\lambda^2_{l,1}+3\beta)\right]\delta_{n,l}.
756: \end{equation}%
757: Thus eq.(\ref{b1}) can be rewritten as
758: \begin{equation}
759: c_n(\lambda^2_{n,1}-\lambda^2)(\lambda^2_{n,2}-\lambda^2)=-d_n,
760: \end{equation}%
761: and a solution of the inhomogeneous equation with r.h.s $-f(z)$ has a form:
762: \begin{equation}
763: \widetilde{b}_{\varphi}(z)=-\sum_n\frac{d_n\widetilde{b}_{\varphi, n}(z)}{(\lambda^2_{n,1}-\lambda^2)(\lambda^2_{n,2}-\lambda^2)}.
764: \label{b4}
765: \end{equation}%
766: Finally noticing that $d_n=\frac{1}{\|\widetilde{b}_{\varphi,n}\|^2}\int^\infty_{-\infty}f(z)\widetilde{b}_{\varphi, n}(z)dz$,
767: a solution of inhomogeneous equation (\ref{b4}) can be rewritten as
768: \begin{equation}
769: \widetilde{b}_{\varphi}(z)=-\int^\infty_{-\infty}G(z,\zeta)f(\zeta)d\zeta,
770: \end{equation}%
771: where
772: $$
773: G(z,\zeta)=\sum_n\frac{\widetilde{b}_{\varphi,n}(z)\widetilde{b}_{\varphi,n}(\zeta)}{(\lambda^2_{n,1}-\lambda^2)
774: (\lambda^2_{n,2}-\lambda^2)\|\widetilde{b}_{\varphi,n}\|^2}
775: $$
776: is a Green's function written in terms of an eigenfunctions $\widetilde{b}_{\varphi,n}(z)$.
777:
778: \bsp
779: \label{lastpage}
780: \end{document}
781: