math0402367/heat.tex
1: %\documentstyle[a4,psfig]{article}
2: 
3: %\documentclass{article}
4: \documentclass[12pt]{article}   % for thesis reduction 
5: 
6: \usepackage{a4,psfig,amssymb,verbatim,amsmath}
7: 
8: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
9: % kak bilo ranishe  
10: %
11: %\documentstyle[psfig]{article}
12: %\pagestyle{empty}
13: %\textwidth 11.8 cm  %\textheight 16.5 cm
14: %\textheight 17 cm
15: %\voffset -2.5 truecm
16: %\hoffset -2.5 truecm
17: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
18: 
19: 
20: \begin{document}
21: \newcommand{\ddt}{\partial \over \partial t}
22: \newcommand{\ddx}{\partial \over \partial x}
23: \newcommand{\ddu}{\partial \over \partial u}
24: \def\dhp#1{ \mathop{#1}\limits_{h}}
25: \def\dphh#1{ \mathop{#1}\limits_{h \bar h}}
26: \def\da#1{ \mathop{#1}\limits_{+\tau}}
27: \def\db#1{ \mathop{#1}\limits_{-\tau}}
28: \def\dc#1{ \mathop{#1}\limits_{\pm \tau}}
29: \def\dd#1{ \mathop{#1}\limits_{+h}}
30: \def\df#1{ \mathop{#1}\limits_{-h}}
31: \def\dpm#1{ \mathop{#1}\limits_{\pm h}}
32: \def\dg#1{ \mathop{#1}\limits_{\pm h}}
33: \def\dh#1{ \mathop{#1}\limits_ h}
34: 
35: 
36: \begin{center}
37: {\Large {\bf Symmetry--preserving discrete schemes for \\
38: \medskip
39: some heat transfer equations}.}
40: \end{center}
41: 
42: \bigskip
43: 
44: \begin{center}
45: 
46: {\large 
47: %{\fbox{Bakirova, M.I.,}
48:  {Bakirova, M.I.}$^{*}$,  Dorodnitsyn, V.A.$^{*}$ and 
49: %\footnote{E-mail address: dorod@spp.Keldysh.ru}
50: Kozlov, R.V.$^{\dag}$ } 
51: %\footnote{E-mail address: kozlov@math.ntnu.no}
52: 
53: \bigskip
54: \bigskip
55: 
56: ${}^{*}$
57: Keldysh Institute of Applied Mathematics of Russian Academy of Science, \\
58: Miusskaya Pl.~4, Moscow, 125047, Russia\\
59: 
60: ${}^{\dag}$  Department of Mathematical Sciences, NTNU, N-7491,
61: Trondheim, Norway \\
62: 
63: \end{center}
64: 
65: 
66: \vspace{1.0 ex}
67: 
68: %\noindent Keldysh Institute of Applied Mathematics,
69: %\noindent Miusskaya Pl.4, Moscow, 125047, Russia}
70: 
71: \bigskip
72: 
73: \begin {quotation}
74: \noindent {\bf Abstract.} Lie group analysis of differential 
75: equations is a
76: generally recognized method, which provides invariant solutions,
77: integrability, conservation laws etc. 
78: In this paper we present three
79: characteristic  examples of the construction of invariant difference
80: equations and meshes, where the original continuous symmetries are
81: preserved in discrete models. 
82: Conservation of symmetries 
83: in difference modeling helps to retain qualitative properties of 
84: the differential 
85: equations in their difference counterparts. 
86: 
87: 
88: 
89: \end{quotation}
90: 
91: 
92: \bigskip
93: 
94: \noindent {\bf 1. Introduction}
95: 
96: \medskip
97: 
98: Symmetries are intrinsic and fundamental features of the differential 
99: equations of mathematical physics. Consequently, they should be 
100: retained when discrete analogs of such equations are constructed. 
101: %Symmetries
102: %of differential equations of mathematical physics are
103: %their intrinsic and fundamental feature, and, consequently,
104: %they should be involved when discrete analogs of these equations
105: %are constructed.
106: 
107: 
108: The group properties of a heat transfer equation with a source
109: \begin{equation} \label{eq0}
110: u_t = \left( K(u) u_{x} \right) _{x} + Q(u),
111: \end{equation}
112: were considered in [4], and all choices of $K(u)$ and $Q(u)$ which extend the
113: symmetry group admitted by the general case of equation (1) were identified. 
114: In this
115: paper we consider two partial cases of nonlinearities from the
116: classification in [4] :
117: \begin{equation} \label{eq1I}
118: u_t = \left(  u^{\sigma} u_x \right) _x \pm u^{n},
119: \ \ \sigma ,\  n =const,
120: \end{equation}
121: \begin{equation} \label{eq2I}
122: u_{t} =  u_{xx}  + \delta u \ln u , \ \ \delta = \pm 1,
123: \end{equation}
124: together with linear case
125: \begin{equation} \label{eq3I}
126: u_t = u_{xx} ,
127: \end{equation}
128: whose group properties were known by S.~Lie.
129: For all cases we construct difference equations
130: and meshes (lattices) that admit the same Lie group of point
131: transformations as their continuous limits.
132: \par
133: We recall that Lie point symmetries yield a number of useful
134: properties of differential equations [13,16,10]:
135: \par
136: a) A group action transforms the complete set of solutions into itself;
137: so it is possible to obtain new solutions from a given one.
138: \par
139: b) There exists a standard procedure to obtain the whole set of invariants
140: and differential invariants for a symmetry group; it yields the invariant
141: representation of the differential equations and the forms of invariant
142: solutions in which they could be found (symmetry reduction of PDE).
143: \par
144: c) For ODEs the known symmetry yields the reduction of the order;
145: if the dimension of symmetry is equal to (or greater then) the order of ODE,
146: then we have a complete integrability.
147: \par
148: d) The invariance of PDEs is a necessary condition for the application
149: of Noether's theorem on variational problems to obtain conservation laws.
150: \par
151: e) It should be noticed that Lie point transformations have a clear
152: geometrical interpretation  and one can construct the orbits of a group in a
153: finite
154: dimensional space of independent and dependent variables.
155: 
156: \par
157: 
158: 
159: 
160: The structure of the admitted group
161: essentially effects the construction of equations
162: and grids. Group transformations can break the geometric structure of the
163: mesh that influences the approximation and other
164: properties of a difference equation. Early contributions to the construction 
165: of the
166: difference grids based on the symmetries of the initial difference
167: model are [6,8]. Classes of transformations that
168: conserve uniformity, orthogonality, and other properties of meshes 
169: will be defined below.
170: 
171: \par
172: In accordance with equation (\ref{eq0}) we consider Lie point
173: transformations in a space with two independent variables: $t$ and
174: $x$. Let
175: \begin{equation} \label{operator}
176: X = \xi ^{t} {\partial \over \partial t} + \xi ^{x} {\partial \over \partial x}
177: +\eta {\partial \over \partial u} + \cdots  ,
178: \end{equation}
179: \noindent be an operator of a one-parameter transformation group. Dots denote
180: prolongation of the operator on other variables used in the given
181: differential equation:
182: \begin{equation} \label{neq6}
183: F(x,t,u_t,u,u_x,u_{xx}) = 0 .
184: \end{equation}
185: The group generated by (\ref{operator}) transforms a point
186: $(x,t,u,u_t,u_x,u_{xx})$ to a new one 
187: $ (x^*, t^*, u^*, u^*_t$, $u^*_x, u^*_{xx})$
188: together with equation (\ref{neq6}). This situation changes when 
189: applying 
190: Lie points transformations to difference equations. Let
191: \begin{equation} \label{neq7}
192: F(z) = 0
193: \end{equation}
194: be a difference equation defined on some finite set of points
195: $z^1,z^2,\ldots $ (difference
196: stencil) on a mesh.
197: In contrast to the point $(x,t,u,u_t,u_x,u_{xx})$, the "difference point" -
198: the difference
199: stencil has its own geometrical structure.
200: %Group transformations can destroy the geometric structure of a
201: %difference mesh, which influences on approximation and other
202: %properties of a difference equation.
203: \par
204: Let
205: \begin{equation} \label{neq8}
206: \Omega(z,h) = 0
207: \end{equation}
208: be an equation that define a difference stencil and a  mesh. As an example it will  be a uniform mesh if the left
209: step (spacing) equals the right step:
210: \begin{equation} \label{neq9}
211: h^+=h^- .
212: \end{equation}
213: \par
214: The invariance of the difference equation (\ref{neq7}) depends on the
215: invariance of
216: (\ref{neq8}), since the latter must be included 
217: in the general condition of     invariance:
218: \begin{equation} \label{neq10}
219: \left\{
220: \begin{array}{rcl}
221: XF(z)_{|_{(7)(8)}}& =& 0,\\
222: X\Omega (z,h)_{|_{(7)(8)}}& =& 0.\\
223: \end{array}
224: \right.
225: \end{equation}
226: \par
227: Relations between the two conditions in  (\ref{neq10}) depends on whether 
228: $ \Omega$, $\xi^t$, $\xi^x$ depend on solution 
229: or not. If $ \Omega_u=\xi^t_u=\xi^x_u=0$, then the conditions (\ref{neq10})
230: could be considered independently.
231: \par
232: Thus, what makes our approach [5-8] special is 
233: the inclusion of the second equation of (\ref{neq10}) in the 
234: conditions of invariance,
235: admitting the whole set of properties a)--e) stated for
236: equations (\ref{neq7}) and (\ref{neq8}).
237: \par
238: There exist a few ways to avoid transformations of the difference stencil
239: and, consequently, transformations of a mesh. One way is to restrict
240: transformations to the case when
241: independent variables are not changed: $\xi^t=\xi^x=0$, 
242: yielding any mesh invariant (see [12]). But this restriction is 
243: very strong and would exclude most symmetries of 
244: physical problems.
245: \par
246: Another approach is connected to evolutionary vector fields. It is known
247: [10,16] that the symmetry operator    could be represented as
248: \begin{equation} \label{neq11}
249: {\bar X} = \left(\eta -  \xi^{t} u_{t}  - \xi^{x}u_{x} \right) {\partial\over
250: \partial u} + \cdots
251: \end{equation}
252: which could be viewed as representative of a factor
253: algebra by the ideal
254: \begin{equation}
255: \xi^t(z) D_t + \xi^x(z) D_x,
256: \end{equation}
257: where $ D_t ={\displaystyle \frac{\partial}{\partial t}}+
258: {\displaystyle u_t \frac{\partial}{\partial u}} + \ldots,\quad
259: {\displaystyle D_x =\frac{\partial}{\partial x}}+
260: {\displaystyle u_x\frac{\partial}{\partial u}}\ldots$,
261: both admitted by all equations (\ref{neq6}). For differential equations
262: this approach gives an equivalent result but with some losses in points
263:  a)--e).  In particular we lose the geometric sense even
264: for point transformations due to the realization of group transformations 
265: in 
266: infinite dimensional spaces 
267: $(t,x,u_t,u_x,u_{tx},u_{xx},u_{tt},\ldots)$, there
268: is no procedure
269: for calculating invariants of 
270: operators of the type given by (\ref{neq11}), etc.
271: \par
272: As it was shown in [5],  the exact representation of the operator of total
273: differentiation $D_x$  in a space of difference variables is given by the
274: following operators (the same is also true for $D_t$):
275: \begin{equation} \label{eq13n}
276: \begin{array}{l}
277: D^+ ={\displaystyle  {\partial \over \partial x}} +
278: \tilde {\dd{D}}(u) {\displaystyle {\partial \over \partial u}} + \cdots, 
279: \qquad 
280: {\displaystyle \tilde{\dd D}}={\displaystyle \sum_{n=1}^{\infty}
281: \frac{(-h)^{n-1}}{n} {\dd D}^n} , \\
282: \\
283: D^- = {\displaystyle \frac{\partial}{\partial x}} + \tilde{\df D}(u)
284: {\displaystyle \frac{\partial}{\partial u}} + \cdots, 
285: \qquad 
286:  {\displaystyle \tilde{\df D}}={\displaystyle \sum_{n=1}^{\infty}
287: \frac{h^{n-1}}{n} {\df D}^n} , 
288: \end{array}
289: \end{equation}
290: where 
291: % ${\displaystyle \tilde{\dd D}}={\displaystyle \sum_{n=1}^{\infty}
292: % \frac{(-h)^{n-1}}{n} {\dd D}^n}, \quad
293: % {\displaystyle \tilde{\df D}}={\displaystyle \sum_{n=1}^{\infty}
294: % \frac{h^{n-1}}{n} {\df D}^n}, $ 
295: $\dd D $ and $  \df D$ are right and left 
296: difference derivative operators on a uniform mesh.
297: As difference operators are related to corresponding shift operators
298: $\dd S, \df S$:
299: $$
300:      \dg D = \pm \frac{(\dg S -1)}{h},
301: $$
302: we obtain another representation for $\tilde{\dg D}$:
303: \begin{equation} \label{eqnewt}
304: \tilde{\dd {D}}=\sum_{n=1}^{\infty}\frac{(- h)^{n-1}}{n} {\dd D}^n=
305: \sum_{n=1}^{\infty}{\frac{(-1)^{n-1}}{nh}(\dd {S}-1)^n},
306: \end{equation}
307: \begin{equation}
308: \tilde{\df {D}}=\sum_{n=1}^{\infty}{\frac{(1-\df{S})^n}{nh}}.
309: \end{equation}
310: 
311: \par
312: The group transformation for operators (\ref{eq13n}) can be obtained by
313: the exponential mapping or by means of the so called Newton series (see [5]).
314: It is
315: important to notice that every difference equation on a regular  mesh  admits
316: the operators $D^+$ and $D^-$, which do not change a mesh. It was shown [5], that
317: a family of operators $\xi(z) D^{\pm}$ forms an ideal in the Lie algebra of
318: operators (\ref{operator}),
319: since it is possible to rewrite these operators as evolutionary vector fields
320: \begin{equation} \label{eqn12}
321: \bar X = \left( \eta - {\xi}^t  D^+_{t}(u) - \xi^x D^+_x (u) \right)
322: {\partial \over \partial u} + \cdots . 
323: \end{equation}
324: (We follow here the right semiaxis representation $D^+$. The same formulation
325: can be done with help of left semiaxis representation $D^-$.)
326: 
327: It is important to notice that the representation (\ref{eq13n}) is true 
328: only for regular (uniform) meshes. Thus, the evolutionary vector fields 
329: (\ref{eqn12}) are applicable only for
330: the groups which do not change uniformity of a mesh. So, one can
331: not apply them for modern numerical methods with moving meshes, self adaptive
332: meshes or   multi-grid methods etc.
333: 
334: 
335: Let us consider a one-parameter transformation group which is generated 
336: by the operator 
337: \begin{equation} \label{eq17n}
338: X  = \xi \frac{\partial}{\partial x} + 
339: \eta {\partial \over \partial u} + \cdots .  
340: \end{equation}
341: We prolong (\ref{eq17n}) on the right and left steps $h^{+}, h^{-}$ 
342: by means of relations $h^{+} = x^{+}- x$ and $h^{-} = x - x^{-}$, where
343: $f^{ \pm} \equiv { \dpm S} (f)$:
344: \begin{equation} \label{eq17p}
345: X  = \xi \frac{\partial}{\partial x} + 
346: \eta {\partial \over \partial u} + \cdots +
347: [\dd{S}(\xi) - \xi]{\partial \over {\partial h^+}} + 
348: [\xi - \df{S}(\xi)]\frac{\partial}{\partial h^-}  . 
349: \end{equation}
350: 
351: 
352: 
353:    From (\ref{eq17p}) it is easy to obtain the invariance condition for 
354: a uniform mesh
355: in a given direction. Let (\ref{neq9}) be invariant with
356: respect to (\ref{eq17p}), then
357: \begin{equation} \label{eqn13}
358: \dd{S}(\xi) - 2\xi +\df{S}(\xi) =0 \quad {\mbox{or}} \quad \dd{D}\df{D}(\xi)=0.
359: \end{equation}
360: 
361: \par
362: Condition (\ref{eqn13}) is a strong limitation on the admitted group.
363: In addition
364: the coefficients of (\ref{eqn12}) are the  power  series  of
365: $\dg D$ or $\dg S$, since one should consider the whole set of 
366: mesh points and not a stencil only.
367: \par
368: Let us illustrate the above with a simple example. The ODE
369: \begin{equation} \label{eqn14}
370: u_{xx}=u^2,
371: \end{equation}
372: can be viewed as the stationary case of equation (\ref{eq1I})
373: with $\sigma=0$, $n=2$. Equation
374: (\ref{eqn14}) has the following Lie point
375: symmetries:
376: \begin{equation} \label{eqn15}
377: X_1=\frac{\partial}{\partial x}; \quad
378: X_2=x\frac{\partial}{\partial x} -2u \frac{\partial}{\partial u} .
379: \end{equation}
380: As a difference analog of (\ref{eqn14}) we consider
381: \begin{equation} \label{eqn16}
382: \frac{u^+-2u+u^-}{h^2} = u^2.
383: \end{equation}
384: on a uniform mesh
385: \begin{equation} \label{eqn17}
386: h^+=h^-,
387: \end{equation}
388: where $u^+=\dd{S}(u), \quad u^-=\df{S}(u)$.
389: 
390: The equations (\ref{eqn16}) and (\ref{eqn17}) use a three-point
391: stencil or subspace $(x, x^+, x^-, u, u^+, u^-)$ and 
392: the operators (\ref{eqn15}) have the following prolongation for the shifted
393: points
394: of the difference stencil:
395: \begin{equation} \label{eqn23}
396: \begin{array}{c}
397: {  \displaystyle 
398: X_1=\frac{\partial}{\partial x}+ \frac{\partial}{\partial {x^+}}+
399: \frac{\partial}{\partial {x^-}}; 
400: \qquad 
401: X_2= x\frac{\partial}{\partial x} + x^+
402: \frac{\partial}{\partial {x^+}}  +
403:  x^- \frac{{\partial}}{{\partial} {x^-}}  } \\
404: \\
405: {  \displaystyle 
406:  -2u\frac{\partial}{{\partial}{u}} -
407: 2u^+ \frac{\partial}{\partial {u^+}} -2u^-\frac{\partial}{\partial {u^-}} +
408:  h^+{\partial \over {\partial h^+}} +
409: h^- \frac{\partial}{\partial h^-}. } 
410: \end{array}
411: \end{equation}
412: 
413: 
414: The symmetry algebra (\ref{eqn15}) acts on the space $(t,x)$,
415: so the coefficients
416: of (\ref{eqn23}) have the same form in different points of the stencil. The
417: prolongation
418: forms for $h^+$ and $h^-$ are easily derived from the relations
419: $ h^+ = x^+ - x,\quad h^-= x^- - x$.
420: \par
421: It is easy to verify that the  equations (\ref{eqn16}) and (\ref{eqn17}) are
422: invariant under
423: the action of (\ref{eqn23}):
424: \begin{equation} \label{eqn18}
425: \begin{array}{rcl}
426: X_2 \left({\displaystyle \frac{u^+ -2u + u^-}{h^2}}
427:  - u^2 \right)_{|_{(\ref{eqn16})}}& =&0, \\
428: \\
429: X_2 \left( h^+ -h^- \right)_{|_{(\ref{eqn17})}}& =&0,
430: \end{array}
431: \end{equation}
432: (Operator $X_1$  leaves (\ref{eqn16}) and (\ref{eqn17}) unchanged).
433: \par
434: It follows that the system (\ref{eqn16}) and (\ref{eqn17}) 
435: has the same Lie point symmetry as its continuous limit.
436: Notice that the invariance conditions (\ref{eqn18}) are mutually independent.
437: \par
438: As for the continuous case one can easily  calculate the finite difference
439: invariants for (\ref{eqn23}) by solving the standard system:
440: \begin{equation} \label{eqn19}
441: X_i \  I (x,h^+,h^-,u,u^+,u^-)=0,\qquad i=1,2. 
442: \end{equation}
443: The solution of (\ref{eqn19}) yields the whole set of difference invariants
444: \begin{equation} \label{eqn20}
445: I^1=\frac{h^+}{h^-},\quad
446: I^2=\frac{u^+}{u},\quad
447: I^3=\frac{u^-}{u},\quad
448: I^4=(h^+)^2 u;
449: \end{equation}
450: It follows that the difference model (\ref{eqn16}) and (\ref{eqn17}) 
451: can be represented by
452: means of the invariants (\ref{eqn20}) as $I^2 + I^3 - 2 = I^4$ and $I^1 = 1$.
453: 
454: \par
455: Let us now consider the evolutionary vector fields for the difference
456: equation  (\ref{eqn16}) 
457: (we consider the right semiaxis representation $D^+$). 
458: We prolong the operator (\ref{eq13n}) on all points of a given stencil 
459: $(x,x^{+},x^{-},u,u^{+},u^{-})$:
460: \begin{equation} \label{eqn21n}
461: D^{+} = 
462: {\ddx} + 
463: {\partial \over \partial x^{+} } + 
464: {\partial \over \partial x^{-} } + 
465: u_{x} {\ddu }  + 
466: u_{x}^{+}  {\partial \over \partial u^{+} } + 
467: u_{x}^{-}  {\partial \over \partial u^{-} } ,
468: \end{equation}
469: where 
470: $$
471: u_{x} \equiv 
472: {\displaystyle \sum_{n=1}^{\infty}\frac{(-h)^{n-1}}{n}{\dd D}^n (u) }, 
473: \quad 
474: u_{x}^{+}  \equiv 
475: {\displaystyle \sum_{n=1}^{\infty}\frac{(-h)^{n-1}}{n}{\dd D}^n (u^{+}) } , 
476: \quad 
477: u_{x}^{-} \equiv 
478: {\displaystyle \sum_{n=1}^{\infty}\frac{(-h)^{n-1}}{n}{\dd D}^n (u^{-}) } .
479: $$  
480: The evolution vector fields (\ref{eqn12}) 
481: will have the following forms:
482: \begin{equation} \label{eqn21p}
483: \begin{array}{lcl}
484:   { \dh  {\bar{X}} { }_1   } &   = &  - X_{1} + D^{+}  = 
485: {\displaystyle 
486: u_{x} {\ddu} + 
487: u_{x}^{+}  {\partial \over \partial u^{+} } + 
488: u_{x}^{-}  {\partial \over \partial u^{-} }  }; \\
489:   \\
490:  {  \dh   {\bar{X}} { }_2 }  &  =  & - X_{2} + x D^{+} = 
491: {\displaystyle 
492: - h^{+} {\partial \over \partial h^{+} } - 
493: h^{-} {\partial \over \partial h^{-} }  } \\
494: \\
495: & & 
496: {\displaystyle 
497: + (2u + x u_{x}) {\ddu} + 
498: (2u^{+}  + x u_{x}^{+}) {\partial \over \partial u^{+} }+ 
499: (2u^{-}  + x u_{x}^{-}) {\partial \over \partial u^{-} } }. 
500: \end{array}
501: \end{equation}
502: 
503: 
504: 
505: 
506: 
507: 
508: %\begin{equation} \label{eqn21}
509: %\begin{array}{rcl}
510: %     \bar{ \dhp X_1} =       { {\displaystyle \sum_{n \ge 1}}
511: %              {\displaystyle \frac{(-h)^{n-1}}{n }
512: %                                 \dd {D^n}(u)       }}{\displaystyle
513: %\frac{\partial}{\partial u}}       + \ldots&  =&\\
514: %      \left{{\displaystyle \sum_{n \ge 1} \frac{(-1)^{n-1}}{hn }}
515: %     (\dd S-1)^n(u)
516: %\right } {\displaystyle \frac{\partial}{\partial u} } + \ldots,\\
517: %
518: %     \bar{ \dhp X_2} = (2u + x {\displaystyle \sum_{n \ge 1}
519: %\frac{(-h)^{n-1}}{n }
520: %     \dd {D^n}(u) ) \frac{\partial}{\partial u}} + \ldots &  = &\\
521: %      \left{{\displaystyle(2u+ x \sum_{n \ge 1} \frac{(-1)^{n-1}}{hn}}
522: %     (\dd S-1)^n(u))
523: %     \right } {\displaystyle \frac{\partial}{\partial u} } +\ldots.
524: %\end{array}
525: %\end{equation}
526: 
527: 
528: 
529: \par
530: It is  not  easy  to  check  the  invariance  conditions  of  equation
531: (\ref{eqn16})  for the operators
532: (\ref{eqn21p}) because one should use not only the equation (\ref{eqn16}),
533: but all its sequences
534: obtained by shifting to the right. A harder question is how to
535: produce the finite difference invariants (\ref{eqn20}) by means of
536: ({\ref{eqn21p}). 
537: That is why we prefer the first
538: classical way for Lie point symmetries and include a mesh in the invariance
539: condition ({\ref{neq10}) (a very similar approach for the semi-discretized
540: nonlinear
541: heat equation was introduced recently  in [3]). An additional convincing reason
542: to apply the classical representation of Lie point symmetries springs from
543: comparison of unsuitable evolutionary vector fields approaches to invariant
544: variational problems, developed in [5], and a clear classical way to construct
545: Noether type theorems for difference equations [7].
546: \par
547: Another approach to the symmetry of difference equations 
548: on a fixed uniform mesh was introduced 
549: in [9]. However, that way is only applicable to linear equations; 
550: moreover, it requires
551: knowing a complete set of solutions of difference equations. The newly
552: introduced
553: approach [11] deals with evolutionary vector fields on a uniform mesh. The
554: advantage of the last two approaches seems to be in the potential of finding
555: non-point symmetries of difference equations which are not available in 
556: their continuous limits.
557: 
558: \par
559: Returning to equation (\ref{eq0}), we notice that a transformation defined by
560: (\ref{operator})
561: conserves uniformity of a grid in $t$ and $x$ directions, if
562: % corresponding conditions
563: \begin{equation} \label{ct}
564: { \da{D} }{ \db{D}  }( \xi ^{t} ) = 0 ,
565: \end{equation}
566: \begin{equation} \label{ch}
567: { \dd{D} }{ \df{D} }  ( \xi ^{x} ) = 0,
568: \end{equation}
569: where $\da D$ and $\db D$ are right and left difference operators in $t$ direction.
570: 
571: %, are hold on solutions of (\ref{eqs}) (see [3]).
572: 
573: \par
574: Conditions (\ref{ct}) and (\ref{ch}) are not sufficient to describe 
575: the invariance 
576: of an orthogonal mesh. For an orthogonal mesh to conserve its orthogonality
577: under the
578: transformation, it is necessary and sufficient that [6,8]:
579: \begin{equation} \label{cht}
580: {\dd{D} }( \xi ^{t} ) = -{ \da{D}} ( \xi ^{x} ) .
581: \end{equation}
582: \par
583: When condition (\ref{cht}) is not satisfied for a given group, the
584: flatness of the layer of a grid in some direction is rather important.
585: For evolution equations it is significant to have flat time layers, since
586: otherwise, after transformations, some domains of a space could be
587: in the future, while others in the past. We have a simple
588: criterion [6,8] for the preserving the flatness of the layer of a grid
589: in the time direction under the action of a given operator
590: (\ref{operator}):
591: \begin{equation} \label{cht2}
592: {\dd{D}} { \da{D}} ( \xi ^{t} ) = 0.
593: \end{equation}
594: 
595: \par
596: 
597: So, the conditions (\ref{ct})--(\ref{cht2}) provide a geometry of grids
598: that is based on the Lie group symmetry. These conditions will be used
599: in what follows.
600: 
601: \bigskip
602: 
603: 
604: \noindent{\bf 2. Invariant model for the equation $u_{t} =
605: (u^{\sigma}u_{x})_{x} \pm u^{n}$  }
606: 
607: \medskip
608: \noindent The equation
609: \begin{equation} \label{eq33}
610: u_{t}=(u^{\sigma}u_{x})_{x}  \pm u^{n},\quad \sigma,n=const,
611: \end{equation}
612: admits a 3-parameter symmetry group. This group can be represented by the
613: following infinitesimal operators [4]:
614: \begin{equation} \label{op33}
615: X_{1}= {\ddt};\quad X_{2}={\ddx};\quad X_{3}= 2(n-1)t{\ddt}+(n-\sigma
616: -1)x{\ddx} -2u{\ddu} .
617: \end{equation}
618: The set (\ref{op33}) satisfies all conditions (\ref{ct})--(\ref{cht2}).
619: So, we can use an orthogonal grid that is uniform in the $t$ and $x$
620: directions.
621: Let us consider the set of operators (\ref{op33}) in the space
622: $( t,  \hat{t},  x,  h^{+},  h^{-},  u,  u_{+}, 
623: u_{-},   \hat{u},  \hat{u}_{+},  \hat{u}_{-})$ that
624: corresponds to the stencil shown in Fig. 1.
625: 
626: 
627: 
628: %\input gridsr.tex
629: 
630: 
631: \begin{figure}[h!]  \label{grort}
632: %\documentstyle{progart}
633: 
634: %\begin{document}
635: 
636: 
637: 
638: 
639: 
640: \begin{picture}(300,100)
641:             %45
642: %\put(0,0){\line(1,0){300}}
643: 
644: %\put(0,120){\line(1,0){300}}
645: 
646: %\put(0,0){\line(0,1){120}}
647: 
648: %\put(300,0){\line(0,1){120}}
649: 
650: %
651: 
652: \put(105,0){\begin{picture}(200,100)%
653: 
654: \put(30,30){\line(1,0){140}}
655: 
656: \put(30,70){\line(1,0){140}}
657: 
658: \put(100,30){\line(0,1){40}}
659: 
660: %
661: 
662: \put(30,30){\circle*{5}}
663: 
664: \put(30,70){\circle*{5}}
665: 
666: \put(170,70){\circle*{5}}
667: 
668: \put(170,30){\circle*{5}}
669: 
670: \put(100,30){\circle*{5}}
671: 
672: \put(100,70){\circle*{5}}
673: 
674: %
675: 
676: \put(5,75){$(x-h,\hat{t},\hat{u}_{-})$}
677: \put(85,75){$(x,\hat{t},\hat{u})$}
678: \put(145,75){$(x+h,\hat{t},\hat{u}_{+})$}
679: \put(5,20){$(x-h,t,u_{-})$}
680: \put(85,20){$(x,t,u)$}
681: \put(145,20){$(x+h,t,u_{+})$}
682: 
683: 
684: %\put(0,75){$(x-h,t+ \tau)$}
685: 
686: %\put(80,75){$(x,t+\tau)$}
687: 
688: %\put(140,75){$(x+h,t+\tau)$}
689: 
690: %\put(10,20){$(x-h,t)$}
691: 
692: %\put(89,20){$(x,t)$}
693: 
694: %\put(152,20){$(x+h,t)$}
695: 
696: 
697: 
698: %\put(25,60){$\hat{u}_{-}$}
699: 
700: %\put(90,60){$\hat{u}$}
701: 
702: %\put(165,60){$\hat{u}_{+}$}
703: 
704: %\put(25,35){$u_{-}$}
705: 
706: %\put(90,35){$u$}
707: 
708: %\put(165,35){$u_{+}$}
709: 
710: 
711: 
712: %
713: 
714: \end{picture}}
715: 
716: %
717: 
718: %\put(170,10){Fig. 1}
719: 
720: %
721: 
722: \end{picture}
723: 
724: 
725: 
726: 
727: 
728: %\end{document}
729: \caption{The stencil of the orthogonal mesh.}
730: \end{figure}
731: 
732: 
733: 
734: 
735: 
736: There are 7 difference invariants of the Lie algebra (\ref{op33}):
737: \begin{equation} \label{in33}
738: {\tau ^{ n -\sigma -1 \over 2(n-1) } \over h };
739: \quad \tau u^{n-1} ; \quad { \hat{u} \over u} ;
740: \quad {u_{+} \over u}; \quad { u_{-} \over u} ;
741: \quad {\hat{u}_{+} \over \hat{u}}; \quad { \hat{u}_{-} \over \hat{u}} .
742: \end{equation}
743: The small number of symmetry operators (\ref{op33}) provides us with a large
744: number of
745: difference invariants (\ref{in33}). Thus we are left with some additional 
746: degrees of freedom in invariant difference 
747: modeling of (\ref{eq33}). By means of the invariants (\ref{in33}), we could
748: write
749: the following explicit scheme for (\ref{eq33}):
750: \begin{equation} \label{sh33}
751: { \hat{u} - u \over \tau } =
752: { 1  \over h} \left(  \left({u_{+} +u \over 2} \right)^{\sigma} \dh  u { }_{x}
753: - \left({u+ u_{-}\over 2 }\right)^{\sigma} \dh u { }_{\bar x}  \right)  \pm
754:   u^{n},
755: \end{equation}
756: where $\dh u { } _{x}  = {\displaystyle {u_{+} - u \over h }}$, $\dh u { }_{ \bar x}  =
757: {\displaystyle {u - u_{-} \over h}}$.
758: \medskip
759: 
760: 
761: 
762: \bigskip
763: This scheme is not unique and one could construct another form of invariant
764: difference equation. For example an implicit scheme could be as follows:
765: \begin{equation} \label{shsh}
766: { \hat{u} - u  \over \tau } =
767: { 1\over h^{2}}
768: ( \hat{u}_{+}^{\sigma +1 } -
769: 2 \hat{u}^{\sigma +1 } +
770:  \hat{u}_{-}^{\sigma +1 } ) +
771:  \hat{u}^{n}
772: \end{equation}
773: Notice, that continuous limit of the last difference equation
774: \begin{equation} \label{eq33a}
775: u_{t} =
776: ( {u}^{\sigma +1 } )_{xx} +  u^{n}.
777: \end{equation}
778: is equivalent to the equation (\ref{eq33})
779: up to the scaling of $x$. But scheme (\ref{shsh}) is not equivalent to the
780: scheme
781: (\ref{sh33}), because there is no point transformation that relates them.
782: In [18] Samarskii { \it et all}  considered
783: the case $n = \sigma + 1$, $\sigma >0$ and found a finite-difference blow-up
784: solution for the equation
785: (\ref{shsh}) that is invariant with respect to the operator
786: \begin{equation} \label{operat}
787: X^*_3 = ( t - T  ) { \partial \over \partial t } -
788: { 1 \over \sigma  } u { \partial \over \partial u },
789: \end{equation}
790: where $T$ is constant. The operator (\ref{operat}) defines
791: a subgroup which
792: is equivalent to the self-similar subgroup with the operator
793: $X_{3}$ of the set (\ref{op33}). Let us find the solution of the
794: problem (\ref{shsh}) in the invariant form:
795: $$
796: u = \left( 1 - { t \over T  } \right) ^{  \displaystyle  - { 1 \over
797: \sigma }}  \theta (x).
798: $$
799: %This solution is sought ([18]) on the set of infinite number of time
800: %intervals on $[0,T ]$:
801: %\begin{equation} \label{m1}
802: %\tau _{j} = T  { \rho ^{ \sigma } + 1  \over \rho ^{\sigma} }
803: %\rho ^{- \sigma j },
804: %\end{equation}
805: %where $\rho > 1$ is constant.
806: This solution is sought ([18]) on the time mesh with 
807: the infinite number of decreasing steps 
808: \begin{equation} \label{m1}
809: \tau _{j} =   \sigma   { \rho - 1  \over \rho ^{\sigma +1} } 
810: \rho ^{- \sigma j }, \qquad j = 0,1,2,...,
811: \end{equation}
812: where $\rho > 1$ is constant. This mesh corresponds to the blow-up time 
813: $$
814:  T  = {  \sigma \over \rho }  { \rho - 1  \over \rho ^{\sigma } -1  } . 
815: $$ 
816: For the function $\theta(x)$ we have the equation
817: \begin{equation} \label{m3}
818: ( \theta ^{\sigma + 1} ) _{ \bar{x} x } +
819:  \theta ^{\sigma + 1}  =
820: { 1 \over \sigma  } \theta .
821: \end{equation}
822: The solution of the problem (\ref{m3})
823: was found in [18] for the case $\sigma = 2$. Let us fix an arbitrary $M > 0$
824: and let $h = 2 \sin \left( { \displaystyle { 3 \pi \over 2(M+1) } } \right)$.
825: In this case the localization length equals
826: $$
827: l_{h} = { 3 \pi h \over 2 } \left( \arcsin {h\over 2 } \right) ^{-1} , \ \ 0 <
828: h \leq 2
829: $$
830: (see [18]). Then, one can verify that the solution of the problem
831: (\ref{m3}) in the points $x_k=kh$ has the form
832: \begin{equation} \label{s2}
833: \theta _{kh} =  \sqrt{2} \left(  3 \left( 1 - { 4\over h^{2}} \sin^{2} {a_{h}
834: h\over 2} \right) \right) ^{ -{1/2} }
835: \sin ( a_{h} kh ),\ \
836: k= 0, 1, ...,M+1,
837: \end{equation}
838: where $a_{h} = \pi / l_{h} $.
839: 
840: The obtained function $u$ gives the blow-up
841: solution of the problem in the case $\sigma =  2$, $n = 3$, $l= l_{h}$.
842: This solution tends to infinity in all points of the space grid,
843: conserving the structure.
844: As $h \rightarrow 0$, the difference solution (\ref{s2}) tends
845: to the solution of the ordinary differential equation:
846: $$
847: \theta (x) = \left( { 3 \over 4 } \right) ^{1/2} 
848: \sin \left( { x \over 3 } \right) , \qquad  0< x < l_{0} = 3 \pi .
849: $$
850: 
851: 
852: \bigskip
853: 
854: 
855: \noindent{
856: \bf
857: 3. Invariant difference model for the semilinear heat transfer equation }
858: 
859: \medskip
860: \noindent The semilinear heat transfer
861: equation
862: \begin{equation} \label{eq53}
863: u_{t}=u_{xx} + \delta  u \ln u, \ \  \delta = \pm 1,
864: \end{equation}
865: admits the 4-parameter Lie symmetry group of point
866: transformations [4], that corresponds to the following
867: infinitesimal operators:
868: \begin{equation} \label{op53}
869: X_{1}= {\ddt}; \quad X_{2} = {\ddx}; \quad  X_{3} = 2 {\Large e}^{\displaystyle
870:  \delta t } {\ddx} - \delta {\Large e}^{\displaystyle  \delta t} x u{\ddu};
871: \quad X_{4} = {\Large e}^{\displaystyle  \delta t } u {\ddu} .
872: \end{equation}
873: Before constructing a difference equation and a grid that
874: approximate (\ref{eq53}) and inherit the whole Lie algebra (\ref{op53}),
875: we should first check condition (\ref{cht}) for
876: the invariance of orthogonality. The operators $X_{1}$, $X_{2}$
877: and $X_{4}$ conserve the orthogonality, but $X_{3}$ does not: condition
878: (\ref{cht}) is not true for the last operator. Consequently an orthogonal 
879: mesh cannot be used for the invariant modeling of (\ref{eq53}).
880: 
881: 
882: The conditions (\ref{cht2}) are true for the complete set (\ref{op53}), so 
883: it is possible to use a nonorthogonal grid with flat time layers. An example 
884: of a grid with with flat time layer is shown in Fig. 2.
885: 
886: 
887: 
888: \begin{figure}[h!] 
889: 
890: 
891: %\documentstyle{progart}
892: 
893: %\begin{document}
894: 
895: \begin{picture}(300,160)
896: 
897: %\put(0,0){\line(1,0){300}}
898: 
899: %\put(0,180){\line(1,0){300}}
900: 
901: %\put(0,0){\line(0,1){180}}
902: 
903: %\put(300,0){\line(0,1){180}}
904: 
905: %
906: 
907: \put(120,0){\begin{picture}(200,150)%
908:         %+40
909: { \thicklines
910: 
911: \put(10,20){\vector(1,0){180}}
912: 
913: \put(10,20){\vector(0,1){120}}
914: 
915: }
916: 
917: \put(10,50){\line(1,0){170}}
918: 
919: \put(10,80){\line(1,0){160}}
920: 
921: \put(10,110){\line(1,0){155}}
922: 
923: %
924: 
925: \put(10,20){\line(1,3){10}}
926: 
927: \put(40,20){\line(1,3){10}}
928: 
929: \put(80,20){\line(1,6){5}}
930: 
931: \put(140,20){\line(-1,6){5}}
932: 
933: \put(180,20){\line(0,1){30}}
934: 
935: %
936: 
937: \put(20,50){\line(1,3){10}}
938: 
939: {\thicklines
940: 
941: \put(50,50){\line(1,2){15}}
942: 
943: \put(85,50){\line(0,1){30}}
944: 
945: \put(135,50){\line(-1,3){10}}
946: 
947: \put(50,50){\line(1,0){85}}
948: 
949: \put(65,80){\line(1,0){60}}
950: 
951: }
952: 
953: \put(180,50){\line(-1,3){10}}
954: 
955: %
956: 
957: \put(30,80){\line(1,2){15}}
958: 
959: \put(65,80){\line(1,4){7,5}}
960: 
961: \put(85,80){\line(1,6){5}}
962: 
963: \put(125,80){\line(-1,6){5}}
964: 
965: \put(170,80){\line(-1,6){5}}
966: 
967: %
968: 
969: \put(50,50){\circle*{5}}
970: 
971: \put(85,50){\circle*{5}}
972: 
973: \put(135,50){\circle*{5}}
974: 
975: \put(65,80){\circle*{5}}
976: 
977: \put(85,80){\circle*{5}}
978: 
979: \put(125,80){\circle*{5}}
980: 
981: %
982: 
983: \put(0,135){$t$}
984: 
985: \put(200,20){$x$}
986: 
987: %
988: 
989: \end{picture}}
990: %
991: %\put(170,10){Fig. 3}
992: %
993: \end{picture}
994: 
995: 
996: 
997: %\end{document}
998: \caption{An evolutionary mesh with flat time-layers.}
999: \end{figure}
1000: 
1001: 
1002: 
1003: A possible reformulation of equation (\ref{eq53}) by using the four 
1004: differential invariants in the subspace 
1005: $(t, x, u, u_{x}, u_{xx}, dt, dx, du)$:
1006: $$
1007: J^{1} = dt; \quad     J^{2} = \left( { u_{x} \over u } \right) ^{2} - { u_{xx}
1008: \over u } ;
1009: \quad J^{3} = 2 { u_{x} \over u } + { dx \over dt } ; 
1010: \quad    
1011: J^{4} = {du \over u dt } - \delta  \ln u + { 1\over 4} \left( {dx \over dt}
1012: \right)^{2} ;
1013: $$
1014: is given by the system:
1015: $$
1016: \left\{
1017: \begin{array}{rcl}
1018: J^{3}& =& 0 ; \\
1019: J^{4}& =& J^{2} ;\\
1020: \end{array}
1021: \right.
1022: $$
1023: or
1024: \begin{equation} \label{sys53}
1025: \left\{
1026: \begin{array}{l}
1027: {\displaystyle {dx\over dt}  = -2 {u_{x} \over u }    } ;\\
1028: \\
1029: {\displaystyle {du\over dt}  = u_{xx} + \delta u \ln u - 2 { u_{x}^{2} \over u
1030: }  }.\\
1031: \end{array}
1032: \right.
1033: \end{equation}
1034: So, the structure of the group (\ref{op53}) suggests the use of two
1035: evolution equations.
1036: \par
1037: As the next step, we will find difference invariants for the set
1038: $X_{1}$--$X_{4}$ of the group (\ref{op53}). These invariants are necessary
1039: for the approximation of the system (\ref{sys53}). We will use
1040: the 6-point difference stencil, as shown in Fig. 3
1041: 
1042: 
1043: 
1044: \begin{figure}[h!]   \label{grnonor}
1045: %\documentstyle{progart}
1046: 
1047: %\begin{document}
1048: 
1049: 
1050: 
1051: 
1052: 
1053: \begin{picture}(300,100)
1054: 
1055: %\put(0,0){\line(1,0){300}}
1056: 
1057: %\put(0,120){\line(1,0){300}}
1058: 
1059: %\put(0,0){\line(0,1){120}}
1060: 
1061: %\put(300,0){\line(0,1){120}}
1062: 
1063: %
1064: 
1065: \put(110,0){\begin{picture}(200,100)%
1066:    %+40
1067: \put(20,30){\line(1,0){160}}
1068: 
1069: \put(30,70){\line(1,0){140}}
1070: 
1071: \put(80,30){\line(1,2){20}}
1072: 
1073: %
1074: 
1075: \put(20,30){\circle*{5}}
1076: 
1077: \put(30,70){\circle*{5}}
1078: 
1079: \put(170,70){\circle*{5}}
1080: 
1081: \put(180,30){\circle*{5}}
1082: 
1083: \put(80,30){\circle*{5}}
1084: 
1085: \put(100,70){\circle*{5}}
1086: 
1087: %
1088: 
1089: \put(0,75){$(\hat{x}-\hat{h}^{-},\hat{t},\hat{u}_{-})$}
1090: \put(85,75){$(\hat{x},\hat{t},\hat{u})$}
1091: \put(140,75){$(\hat{x}+\hat{h}^{+},\hat{t},\hat{u}_{+})$}
1092: \put(65,20){$(x,t,u)$}
1093: \put(150,20){$(x+h^{+},t,u_{+})$}
1094: %
1095: %\put(10,75){$(\hat{x}-\hat{h}^{-},\hat{t})$}
1096: 
1097: %\put(90,75){$(\hat{x},\hat{t})$}
1098: 
1099: %\put(150,75){$(\hat{x}+\hat{h}^{+},\hat{t})$}
1100: 
1101: %\put(0,20){$(x-h^{-},t)$}
1102: 
1103: %\put(70,20){$(x,t)$}
1104: 
1105: %\put(160,20){$(x+h^{+},t)$}
1106: 
1107: %
1108: 
1109: \end{picture}}
1110: 
1111: %
1112: 
1113: %\put(170,10){Fig. 4}
1114: 
1115: \put(90,20){$(x-h^{-},t,u_{-})$}
1116: 
1117: %
1118: \end{picture}
1119: 
1120: 
1121: %\end{document} 
1122: \caption{The stencil of the evolutionary mesh.}
1123: \end{figure}
1124: 
1125: 
1126: 
1127: \noindent on which we will approximate the system (\ref{sys53}). The stencil 
1128: defines the difference subspace $( t, \hat{t}, x, \hat{x}, h^{+}, h^{-},
1129: \hat{h}^{+},
1130: \hat{h}^{-}, u,$$ u_{+},$$ u_{-},\hat{u},$$ \hat{u}_{+},$$ \hat{u}_{-} )$.
1131: The group (\ref{op53}) has the following difference invariants in this
1132: subspace:
1133: $$
1134: I^{1} =  \tau; \quad I^{2} =  {h}^{+}; \quad I^{3} = {h}^{-};
1135: \quad I^{4} = \hat{h}^{+}; \quad I^{5} = \hat{h}^{-};
1136: $$
1137: $$
1138: I^{6} = ( \ln u) _{x}  - ( \ln u )_{\bar{x}} ;
1139: \qquad I^{7} = ( \ln \hat{u} ) _{x}  - ( \ln \hat{u} )_{\bar{x}} ;
1140: $$
1141: $$
1142: I^{8} = \delta \Delta x + 2 ({\Large e}^{\displaystyle \delta \tau} - 1)
1143: \left( { h^{-} \over h^{+} + h^{-} } ( \ln u )_{x}  +
1144:     { h^{+} \over h^{+} + h^{-} } (  \ln u )_{\bar{x}}  \right) ;
1145: $$
1146: $$
1147: I^{9} = \delta \Delta x + 2 (1 - {\Large e}^{\displaystyle - \delta \tau} )
1148: \left( { \hat{h}^{-} \over \hat{h}^{+} + \hat{h}^{-} } ( \ln \hat{u} )_{x} +
1149:    {\hat{h}^{+} \over \hat{h}^{+} + \hat{h}^{-} } ( \ln \hat{u} )_{\bar{x}}
1150: \right) ;
1151: $$
1152: $$
1153: I^{10} = \delta (\Delta x) ^{2} + 4 (1 - {\Large e}^{\displaystyle - \delta
1154: \tau} )
1155: \left( \ln \hat{u} - {\Large e}^{\displaystyle  \delta \tau} \ln u \right) ;
1156: $$
1157: where ${\displaystyle \Delta x = \hat{x}-x }$,
1158: ${\displaystyle  (\ln u)_{x}={ \ln u_{+} - \ln u \over h^{+}} }$ ,
1159: ${ \displaystyle ( \ln u ) _{\bar x} = { \ln u - \ln u_{-} \over h^{-}} }$ .
1160: 
1161: \medskip
1162: 
1163: \par
1164: \noindent To obtain an invariant difference model, it is natural 
1165: to use the difference invariants. An explicit model is given by
1166: $$
1167: \left\{
1168: \begin{array}{l}
1169: { \displaystyle I ^{8} = 0 }  ;\\
1170: \\
1171: { \displaystyle I^{10} =  {8 \over \delta} 
1172: {  ({\Large e}^{\displaystyle \delta I^{1} } - 1)^{2} \over
1173: I^{2} + I^{3} } I^{6} } ; \\
1174: \end{array}
1175: \right.
1176: $$
1177: or
1178: \begin{equation} \label{sh53}
1179: \left\{
1180: \begin{array}{l}
1181: { \displaystyle {\delta \Delta x + 2 ({\Large e}^{\displaystyle \delta \tau} -
1182: 1)
1183: \left( { h^{-} \over h^{+} + h^{-} } ( \ln u) _{x}  +
1184:     { h^{+} \over h^{+} + h^{-} } ( \ln u )_{\bar{x}}  \right) = 0 } };\\
1185: \\
1186: { \displaystyle {\delta (\Delta x) ^{2} + 4 (1 - {\Large e}^{\displaystyle -
1187: \delta \tau} )
1188: \left( \ln \hat{u} - {\Large e}^{\displaystyle  \delta \tau} \ln u \right) =
1189: {8 \over \delta} {  ({\Large e}^{\displaystyle \delta \tau} - 1)^{2} \over
1190: h^{+} + h^{-} }
1191:            \left[ (\ln u) _{x} - ( \ln u)_{\bar{x}} \right] } } .
1192: \end{array}
1193: \right.
1194: \end{equation}
1195: 
1196: 
1197: \medskip
1198: 
1199: 
1200: \par
1201: As in the continuous case, there is a reduction 
1202: in the difference case. 
1203: When we consider an invariant solution, we have the reduction of the
1204: equation-grid
1205: system to a system of ordinary difference equations. 
1206: One being the difference model for the considered equation, the other 
1207: for the evolution of the grid.
1208: 
1209:   Let us find the solution of the difference model (\ref{sh53}) which is
1210: invariant with respect to the operator
1211: \begin{equation} \label{oper53}
1212: 2 \alpha X_{2} + X_{3}, \ \ \alpha = const.
1213: \end{equation}
1214: ${ \displaystyle { u \exp \left(  { \delta {\Large e}^{\displaystyle  \delta t
1215: }
1216: \over
1217: \alpha + {\Large e}^{\displaystyle  \delta t } }
1218: { x^{2}  \over 4 } \right)} } $,
1219: $\left( {\displaystyle {\Delta x\over
1220: {\Large e}^{\displaystyle  \delta t }
1221:  ( {\Large e}^{\displaystyle  \delta \tau  } -1 )  } -
1222: { x \over
1223:  \alpha + {\Large e}^{\displaystyle  \delta t } } } \right) $
1224: and $t$ are all the invariants with respect to (\ref{oper53}). 
1225: Therefore
1226: we will seek an invariant solution in the form:
1227: $$
1228: \left\{
1229: \begin{array}{l}
1230: { \displaystyle { u(x,t)\ \  =\ \   \exp \left( -
1231:  { \delta {\Large e}^{\displaystyle  \delta t } \over
1232: \alpha + {\Large e}^{\displaystyle  \delta t } }
1233: { x^{2} \ \over 4 } \right)  {\Large e}^{\displaystyle  f(t)  } } } ;\\
1234: \\
1235: {\displaystyle {\Delta x\over
1236: {\Large e}^{\displaystyle  \delta t }
1237:  ( {\Large e}^{\displaystyle  \delta \tau  } -1 )  } \ \ = \ \
1238: { x \over
1239:  \alpha + {\Large e}^{\displaystyle  \delta t } } + g(t)  } .\\
1240: \end{array}
1241: \right.
1242: $$
1243: \noindent Substituting this form of the solution into 
1244: (\ref{sh53}), we obtain a system of ordinary difference equations to
1245: determine $f(t)$ and $g(t)$:
1246: $$
1247: \left\{
1248: \begin{array}{l}
1249: { \displaystyle { f(t+\tau) - {\Large e}^{\displaystyle  \delta \tau } f(t)
1250: \over
1251: {\Large e}^{\displaystyle  \delta \tau  }
1252:  ( {\Large e}^{\displaystyle  \delta \tau  } -1 )  }  } \ \ = \ \
1253: {\displaystyle {  - {1\over 2}
1254: {  {\Large e}^{\displaystyle  \delta t }  \over
1255: \alpha + {\Large e}^{\displaystyle  \delta t  } } } };\\
1256: \\
1257: g(t)\ \  =\ \  0.\\
1258: \end{array}
1259: \right.
1260: $$
1261: \par
1262: \noindent The solution of this system yields the solution of the difference
1263: equation (\ref{sh53}):
1264: $$
1265: u(x,t) = \exp \left(   {\Large e}^{\displaystyle  \delta t  }
1266: \left( f(0) - {   {\Large e}^{\displaystyle  \delta \tau  }  - 1 \over 2 }
1267: \sum_{j=1}^{n-1}   { {\Large e}^{\displaystyle -  \delta t_{j} } \over
1268:  1+ \alpha  {\Large e}^{\displaystyle -  \delta t_{j}   } } \right) -
1269:  { \delta {\Large e}^{\displaystyle  \delta t } \over
1270: \alpha + {\Large e}^{\displaystyle  \delta t } }
1271: { x^{2} \ \over 4 } \right)   .
1272: $$
1273: \noindent and the grid
1274: $$
1275: x = x^{0}  { {\Large e}^{\displaystyle  \delta t  } + \alpha
1276: \over 1 + \alpha }.
1277: $$
1278: \noindent Here $x = x^{j}_{i} = x_{i}(t_{j})$ and $ t = t_{j}$.
1279: For $t=0$ the grid can be arbitrary, but if a regular grid is used 
1280: the grid will be regular on every time layer.
1281: 
1282: 
1283: \par
1284: The obtained solution is the solution of the Cauchy problem with
1285: initial conditions:
1286: $$
1287: u(x,0) =    \exp \left( f(0) -
1288:  { \delta {\Large e}^{\displaystyle  \delta t } \over
1289: \alpha + {\Large e}^{\displaystyle  \delta t } }
1290: { x^{2} \ \over 4 } \right)   .
1291: $$
1292: 
1293: 
1294: \bigskip
1295: 
1296: \noindent {\bf 4. Invariant discrete version of the linear heat equation }
1297: 
1298: \medskip
1299: \noindent The linear heat transfer equation
1300: \begin{equation} \label{eq54}
1301: u_{t} = u_{xx}
1302: \end{equation}
1303: admits a 6-parameter Lie symmetry group of point transformations, 
1304: corresponding to the following infinitesimal operators:
1305: \begin{equation} \label{op54}
1306: \begin{array}{c} 
1307: { \displaystyle 
1308: X_{1} = {{\partial \over \partial t}} ;\qquad X_{2} = {{\partial \over \partial
1309: x}} ; \qquad X_{3} = {2t} {{\partial \over \partial x}} {-} xu {{\partial \over
1310: \partial u}} ; } \\
1311: \\
1312: { \displaystyle 
1313: X_{4} = {2t} {{\partial \over \partial t}} + {x} {{\partial \over \partial x}}
1314: ; \quad X_{5} = {4t^{2}} {{\partial \over \partial t}} + {4tx} {{\partial \over
1315: \partial x}} - ( {x^{2}} + {2t} {)u} {{\partial \over \partial u}} ;
1316: \quad X_{6} = {u} {{\partial \over \partial u}} ; } 
1317: \end{array} 
1318: \end{equation}
1319: \noindent and an infinite-dimensional symmetry obtained from 
1320: the linearity of the equation (\ref{eq54}):
1321: $$
1322: X^{*} = {a(x,t)} \ {{\partial \over \partial u}} ,
1323: $$
1324: where $a(t,x)$ in an arbitrary solution of the equation (\ref{eq54}).
1325: 
1326: Now we are in a position to show 
1327: that the invariant difference model for the linear
1328: heat
1329: transfer equation cannot be constructed on an orthogonal grid. The model
1330: \begin{equation}  \label{eq55a}
1331: { \hat{u} - u \over \tau } = { u_{+} -2u + u_{-} \over  h
1332: ^{2} }
1333: \end{equation}
1334: on the regular orthogonal mesh, which is used as an 
1335: invariant model in [2], (see also
1336: [1], p.~363), 
1337: actually does not admit operators $X_{3}$ and $X_{5}$.
1338: 
1339: Let us check for example the symmetry that is described by $X_{3}$.
1340: The operator $X_{3}$ generates the following transformations
1341: \begin{equation}  \label{tr}
1342: \begin{array}{l}
1343: x^{*} = x + 2t \alpha, \\
1344: t^{*} = t , \\
1345: u^{*} = u {\Large e }^{\displaystyle - x \alpha - t \alpha ^{2} }.
1346: \end{array}
1347: \end{equation}
1348: This transformation destroys the orthogonality of the mesh as 
1349: shown in Figure 4.
1350: 
1351: 
1352: \begin{figure}[h!]
1353: 
1354: \begin{picture}(300,120)
1355: %
1356: \put(60,10){\begin{picture}(130,100)%
1357: 
1358: \put(10,10){\vector(1,0){120}}
1359: \put(10,10){\vector(0,1){90}}
1360: \put(10,40){\line(1,0){100}}
1361: \put(10,70){\line(1,0){100}}
1362: \put(40,10){\line(0,1){75}}
1363: \put(70,10){\line(0,1){75}}
1364: \put(100,10){\line(0,1){75}}
1365: 
1366: \put(15,100){$t$}
1367: \put(120,15){$x$}
1368: 
1369: \put(13,30){$x-h$}
1370: \put(75,30){$x$}
1371: \put(105,30){$x+h$}
1372: \put(75,75){$ \hat{x}$}
1373: 
1374: \put(40,40){\circle*{5}}
1375: \put(70,40){\circle*{5}}
1376: \put(100,40){\circle*{5}}
1377: \put(70,70){\circle*{5}}
1378: %
1379: %\put(60,0){Fig.5a}
1380: %
1381: \end{picture} }
1382: %
1383: \put(225,10){\begin{picture}(130,100)%
1384: \put(10,10){\vector(1,0){120}}
1385: \put(10,10){\vector(0,1){90}}
1386: \put(10,40){\line(1,0){110}}
1387: \put(10,70){\line(1,0){110}}
1388: %
1389: %
1390: \put(25,10){\line(1,2){37}}
1391: \put(55,10){\line(1,2){37}}
1392: \put(85,10){\line(1,2){37}}
1393: %
1394: \put(15,100){$t$}
1395: \put(120,15){$x$}
1396: %
1397: %
1398: \put(40,40){\circle*{5}}
1399: \put(70,40){\circle*{5}}
1400: \put(100,40){\circle*{5}}
1401: \put(85,70){\circle*{5}}
1402: %
1403: \put(15,30){$x^{*} - h^{*}$}
1404: \put(73,30){$x^{*}$}
1405: \put(103,30){$x^{*} + h^{*}$}
1406: \put(93,75){$ \hat{x}^{*}$}
1407: %
1408: \multiput(85,70)(0,-9){4}{\line(0,-1){6}}
1409: %
1410: %\put(60,0){Fig.5b}
1411: %
1412: \end{picture} }
1413: %
1414: %
1415: %
1416: \end{picture}
1417: %
1418: \caption{Deformation of the orthogonal mesh.}
1419: \end{figure}  
1420: 
1421: 
1422: 
1423: The transformation (\ref{tr}) transforms 
1424: the finite-difference equation (\ref{eq55a}) 
1425: into the following:
1426: \begin{equation} \label{eqor}
1427: { \hat{u} {\Large e} ^{\displaystyle  - \tau \alpha^{2} }
1428: - u \over \tau } =
1429: { u_{+} {\Large e}^{ \displaystyle - h \alpha } - 2u +
1430: u_{-} {\Large e } ^{\displaystyle h \alpha }   \over h^{2} },
1431: \end{equation}
1432: which explicitly depends on a group parameter $\alpha$. 
1433: The first differential approximation of the equation (\ref{eqor})
1434: $$
1435: u_{t} = u_{xx} - 4 \alpha u_{x}     + 2 u \alpha ^{2} + O(\tau +
1436: h )
1437: $$
1438: shows explicitly the absence of invariance for the equation
1439: (\ref{eq55a}) on an orthogonal mesh.
1440: 
1441: Consequently, we have to construct a difference model for (\ref{eq54}) on a
1442: moving mesh. 
1443: With help of the differential invariants in the space 
1444: $( t, x, u, u_{x}, u_{xx}, dt, dx, du)$:
1445: $$
1446: J^{1} = { { dx + 2 {\displaystyle { u_{x} \over {u} } } dt } \over { dt^{1/2}}}
1447: ;
1448: \qquad J^{2} = {{du\over u}} + {{1\over 4}} {{dx\over dt}}^{2} + \left( -
1449: {{u_{xx}\over u}} + {u_{x}^{2}\over u^{2}} \right) dt ;
1450: $$
1451: \noindent we can represent the heat equation (\ref{eq54})
1452: as the system:
1453: $$
1454: \left\{
1455: \begin{array}{rcl}
1456: J^{1} &=& 0;  \\
1457: J^{2} &=& 0;  \\
1458: \end{array}
1459: \right.
1460: $$
1461: \noindent or
1462: \begin{equation} \label{sys54}
1463: \left\{
1464: \begin{array}{rcl}
1465: {\displaystyle {dx\over dt} }&=&   -2 {\displaystyle {u_{x}\over u } }; \\
1466: \\
1467: {\displaystyle {du\over dt} }&=& u_{xx} - {2}{\displaystyle {u_{x}^{2}\over u}
1468: }. \\
1469: \end{array}
1470: \right.
1471: \end{equation}
1472: The system (\ref{sys54}) inherits the set of operators $X_{1}$--$X_{6}$,
1473: $X^{*}$.
1474: Implying that it entirely inherits the Lie
1475: symmetry group admitted by the linear heat  equation (\ref{eq54}).
1476: \par
1477: For the difference modeling of the system (\ref{sys54}) we need the whole set
1478: of difference invariants of the Lie symmetry group (\ref{op54}) in the
1479: difference space, corresponding to the chosen stencil $( t, \hat{t}, x,
1480: \hat{x}, h^{+}, h^{-},
1481: \hat{h}^{+}, \hat{h}^{-},u, \hat{u}, u_{+}, u_{-}, 
1482: \hat{u}_{+}, \hat{u}_{-} )$:
1483: $$
1484: I^{1} = { {h^{+}} \over {h^{-}} } ;
1485: \quad I^{2} = { {\hat h^{+}} \over {\hat h^{-}}} ;
1486: \quad I^{3} = { { {\hat h^{+}} { h^{+}}} \over \tau } ;
1487: \quad I^{4} = { \tau^{1/2} \over h^{+} }  {{\hat{u}\over u}} {\exp } \left(
1488: {{1\over 4}} {{(\Delta x)^{2}\over \tau}} \right) ;
1489: $$
1490: $$
1491: I^{5} = {{1\over 4}} {{h^{+2}\over \tau}} - {{h^{+2}\over h^{+} + h^{-}}}
1492: \left( {{1\over h^{+}}} {\ln }  {{u_{+}\over u}}  + {{1\over h^{-}}} {\ln }
1493:  {{u_{-}\over u}} \right) ;
1494: $$
1495: $$
1496: I^{6} = {{1\over 4}} {{\hat{h}^{+2} \over \tau}} + {\hat{h}^{+2} \over
1497: \hat{h}^{+} + \hat{h}^{-}} \left( {1\over \hat{h}^{+}} {\ln } { \hat u_{+}
1498: \over \hat u} + {1\over \hat{h}^{-}} {\ln } {\hat u_{-} \over \hat u} \right) ;
1499: $$
1500: $$
1501: I^{7} = {\Delta xh^{+} \over \tau}  + {2h^{+} \over h^{+} + h^{-}}  \left(
1502: {h^{-} \over h^{+}} {\ln }  {u_{+} \over u}  - {h^{+} \over h^{-}} {\ln }
1503:  {u_{-} \over u}  \right) ;
1504: $$
1505: $$
1506: I^{8} = {{\Delta x\hat{h}^{+}\over \tau}} + {{2\hat{h}^{+}\over \hat{h}^{+} +
1507: \hat{h}^{-}}} \left( {{\hat{h}^{-}\over \hat{h}^{+}}} {\ln } {\hat u_{+} \over
1508: \hat u}  - {{\hat{h}^{+}\over \hat{h}^{-}}} {\ln } {\hat u_{-} \over \hat u}
1509: \right) .
1510: $$
1511: \par
1512: Approximating the system (\ref{sys54}) by these invariants as 
1513: it was done for the semilinear heat equation, we obtain a system
1514: of difference evolution equations. As an example, we present here
1515: an invariant difference model that has explicit equations for the solution
1516: and the trajectory of $x$:
1517: \begin{equation} \label{sh54}
1518: \left\{
1519: \begin{array}{l}
1520: {\Delta x}\ \  =\ \  {\displaystyle {{2\tau\over h^{+} + h^{-}}} }\left( -
1521: {\displaystyle {{h^{-}\over h^{+}}} } {\ln } {\displaystyle {{u_{+}\over u}} }
1522: + {\displaystyle {{h^{+}\over h^{-}}} } {\ln } {\displaystyle {{u_{-}\over u}}
1523: } \right) ;\\
1524: \\
1525: {\left( {\displaystyle {{u\over \hat{u}}} } \right) }^{2} {\exp } \left( - {
1526: \displaystyle {{1\over 2}} {{ (\Delta x)^{2}\over \tau}} } \right)\ \  =\ \   1
1527: -  {\displaystyle {{4\tau\over h^{+} + h^{-}}} \left( {{1\over h^{+}}} {\ln }
1528: {{u_{+}\over
1529: u}} + {
1530: {1\over h^{-}}} {\ln } {{u_{-}\over u}} \right) }.\\
1531: \end{array}
1532: \right.
1533: \end{equation}
1534: 
1535: \bigskip
1536: 
1537: \noindent {\bf 5. Example of an exact solution.}
1538: 
1539: \medskip
1540: \noindent Let us find the solution of the difference model (\ref{sh54}) for the
1541: heat equation that is invariant with respect to the operator
1542: \begin{equation} \label{oper54}
1543: 2 \alpha X_{2} + X_{3}, \ \ \alpha = const.
1544: \end{equation}
1545: 
1546: 
1547: The operator (\ref{oper54}) has three invariants: $t$ and the expressions  
1548: $u \exp ( {\textstyle {x^{2}\over 4(t + \alpha )} } )$ and
1549: $({\textstyle {\Delta x\over \tau} - {x\over t + \alpha } } )$. So, we will
1550: seek the invariant solution in the form:
1551: $$
1552: \left\{
1553: \begin{array}{l}
1554: u(x,t)\ \  =\ \  f(t) \exp \left( - {\displaystyle {x^{2}\over 4(t + \alpha )}
1555: } \right)  ;\\
1556: \\
1557: {\displaystyle {\Delta x\over \tau} }\ \  =\ \  g(t) + {\displaystyle {x\over t
1558: + \alpha } }.\\
1559: \end{array}
1560: \right.
1561: $$
1562: \noindent Substituting this form of the solution to the system (\ref{sh54}),
1563: we obtain ordinary difference equations for $f(t)$ and $g(t)$:
1564: \begin{equation}   \label{eq55}
1565: \left\{
1566: \begin{array}{l}
1567: f(t+\tau)\ \  =\ \  \left( {\displaystyle {t + \alpha \over t + \tau +
1568: \alpha } } \right) ^{1/2} f(t) ;\\
1569: \\
1570: g(t)\ \  =\ \  0 .\\
1571: \end{array}
1572: \right.
1573: \end{equation}
1574: \par
1575: \noindent Solving this system, we find the solution of the
1576: difference equation
1577: $$            %\begin{equation} \label{eq57}
1578: u(x,t) = f(0) \left( {\alpha \over t + \alpha } \right) ^{1/2} \exp \left( -
1579: {x^{2} \over 4(t + \alpha )} \right),
1580: $$            %\end{equation}
1581: \noindent and the solution for evolution of a grid:
1582: $$
1583: x = x^{0} \left( {t + \alpha \over \alpha } \right) .
1584: $$
1585: \noindent The obtained solution is the solution of the Cauchy problem
1586: with the invariant initial condition:
1587: $$
1588: u(x,0) = f(0) \exp \left(  - {x^{2}\over 4\alpha } \right) .
1589: $$
1590: \noindent If $\alpha = 0$, the fundamental solution of the heat equation
1591: \begin{equation}  \label{eq56}
1592: u(x,t) = C \left( {1\over t} \right) ^{1/2} \exp \left( - {x^{2} \over 4t }
1593: \right) .
1594: \end{equation}
1595: \noindent is a solution of the difference model. This solution holds on the
1596: grid:
1597: $$
1598: \Delta x = {\tau \over t} x .
1599: $$
1600: \par
1601: In all cases listed above the difference mesh is arbitrary at the initial
1602: point, $t = 0$. In this case it will not be uniform on other time layers.
1603: If the grid is uniform in $x$--direction at $t = 0$
1604: ($h_{+} = h_{-} = h$), the steps of the grid in the $x$--direction will be
1605: equal
1606: each other on every time layer, but differ from steps on the previous
1607: time layer.
1608: 
1609: \par
1610: It is necessary to mention that the obtained difference invariant solution
1611: is the solution of the corresponding differential equation that is
1612: invariant with respect to the operator (\ref{oper54}). As in the 
1613: differential case the above reduction
1614: procedure can be applied for every subalgebra of the algebra (\ref{op54}), 
1615: and then
1616: one obtain different moving meshes which are self-adaptive to every solution.
1617: 
1618: \par
1619: Thus, the above difference models inherit both the groups of the 
1620: differential equations and the potential to be integrated on a subgroup.
1621: 
1622: \bigskip
1623: 
1624: 
1625: 
1626: \noindent {\bf 6. Numerical calculations }
1627: 
1628: \medskip
1629: \noindent
1630: Here we do not discuss the questions of stability and convergence of the
1631: developed
1632: schemes. These are hard questions for nonlinear schemes but one of the
1633: ways to check them is by computing the numerical solutions to the 
1634: exact solutions of the original
1635: differential equations. Below we present the results of numerical calculation
1636: of the invariant solution (\ref{eq56}) by means of the invariant model
1637: (\ref{sh54}).
1638: It is necessary to note that the calculations were not done for the 
1639: equations (\ref{eq55}) reduced on the subgroup, but for the nonstationary
1640: equations (\ref{sh54}). Initial data correspond to the solution
1641: (\ref{eq56}) with $t = 10$. In Fig. 5 we present the evolution of $u$
1642: from invariant initial data by the invariant scheme (\ref{sh54}).
1643: 
1644: %\vskip 7 cm
1645: 
1646: \begin{figure}[h!]    \label{Figinsol}
1647: \centerline{\psfig{figure=Figinsol.ps,angle=0,width=9cm}}
1648: \caption{Solution of the invariant model.}
1649: \end{figure}
1650: 
1651: 
1652: Let us note that the difference model (\ref{sh54}) gives us 
1653: practically the exact solution for equation (\ref{eq54}). 
1654: There is only round-off error of computations.  
1655: %Let us note that the difference model (\ref{sh54}) gives us 
1656: %practically the exact solution for equation (\ref{eq54}). The
1657: %difference between the solution of the model (\ref{sh54}) and the
1658: %exact solution of the equation (\ref{eq54}) is shown in
1659: %Fig. 6.
1660: %\eject
1661: %{ $_{ }$}
1662: %\vskip   7 cm
1663: %\begin{center}
1664: %{Fig. 6}
1665: %\end{center}
1666: In the Fig. 6 the evolution of the grid in the plane $(t,x)$
1667: for the calculation of the solution (\ref{eq56}) is shown.
1668: 
1669: %\vskip  7 cm
1670: 
1671: \begin{figure}[h!]   \label{Figgird}
1672: \centerline{\psfig{figure=Figgird.ps,angle=0,width=9cm}}
1673: \caption{Evolution of the mesh.}
1674: \end{figure}
1675: 
1676: 
1677: %     lowered
1678: %The same calculations for the difference equation (\ref{eq55a}) on the
1679: %orthogonal grid (as that in Fig. 1) gives a similar picture (see Fig. 7).
1680: 
1681: %\vskip 7 cm
1682: 
1683: 
1684: \begin{figure}[h!]
1685: \centerline{\psfig{figure=Figsolod.ps,angle=0,width=9cm}}
1686: \caption{Solution of the noninvariant scheme.}
1687: \end{figure}
1688: 
1689: 
1690: 
1691: The same calculations for the difference equation (\ref{eq55a}) on the
1692: orthogonal grid (as that in Fig. 1) gives a similar picture (see Fig. 7). 
1693: In this case the solution does not coincide with exact
1694: solution of
1695: the equation (\ref{eq54}). The difference between exact solution and
1696: numerical solution is shown in Fig. 8.
1697: 
1698: %\vskip  7 cm
1699: 
1700: \begin{figure}[h!]
1701: \centerline{\psfig{figure=Figerod.ps,angle=0,width=9cm}}
1702: \caption{The error of the noninvariant scheme.}
1703: \end{figure}
1704: 
1705: 
1706: We should note that for the calculation on the model (\ref{sh54}) we
1707: defined the boundary values $u$ on the moving ends of
1708: the space interval. For the difference
1709: equation (\ref{eq55a}) we defined the boundary values of $u$ on the ends of the
1710: fixed orthogonal grid in accordance with the same solution.
1711: The comparison of the two different models shows that for the
1712: invariant model even
1713: on the decreasing number of the points of the grid on the initial
1714: space interval we have greater accuracy than for the noninvariant
1715: one.
1716: 
1717: 
1718: \bigskip
1719: \medskip
1720: 
1721: 
1722: 
1723: \noindent {\bf Remarks}
1724: 
1725: \medskip
1726: \noindent Following the above technique for the Burgers equation for
1727: the potential
1728: $$
1729: w_{t}  + {1 \over 2} w_{x}^{2} = w_{xx},
1730: $$
1731: we obtain the finite-difference model for this equation
1732: on a moving mesh
1733: $$
1734: \left\{
1735: \begin{array}{l}
1736: \Delta x\ \  =\ \  \tau {\displaystyle { h^{-} \dh w { } _{x}  + h^{+}
1737: \dh w { } _{\bar x} \over h^{+} + h^{-} } };\\
1738: \\
1739: \exp \left( \hat{w} - w - { \displaystyle {\Delta x^{2} \over 2\tau} } \right)\
1740: \  =\ \  1 + \tau \dh w { } _{x \bar x} ;\\
1741: \end{array}
1742: \right.
1743: $$
1744: where $\dh w { } _{x \bar{x}}  ={2 \over h^{-} + h^{+} } ( \dh w { }_{x}
1745: - \dh  w { } _{\bar x} )$.
1746: \par
1747: It is
1748: interesting to note that this model is connected with model (\ref{sh54})
1749: by the same Hopf transformation
1750: $$
1751: w = - 2 \ln u 
1752: $$
1753: as their continuous counterparts.
1754: 
1755: It is important to notice that in all cases the moving in $(x,t)$-
1756: plane meshes can be stopped by the new coordinates of Lagrange's type
1757: with one additional dependent variable (for involving those coordinates see,
1758: for example, [14]).
1759: 
1760: 
1761: 
1762: \bigskip
1763: \noindent { \it  Acknowledgments.} The authors' research was partly
1764: supported
1765: by The Norwegian Research Council under contract no. 111038/410, through 
1766: the SYNODE project, and  Russian Fund for Base Research.
1767: 
1768: 
1769: 
1770: 
1771: 
1772: \begin{thebibliography}{99}
1773: \bibitem{[1]} Ames W F, Anderson R L, Dorodnitsyn V A,
1774: Ferapontov E V,
1775: Gazizov R K, Ibragimov N H and Svirshevskii S R  1994  {\it CRC Hand-book of
1776: Lie Group
1777: Analysis of Differential Equations, Volume I: Symmetries,
1778: Exact Solutions and Conservation Laws}  (CRC Press)
1779: %\par
1780: \bibitem{[2]} Ames W F  1977  {\it Numerical Methods for Partial Differential
1781: Equations,
1782: 2nd ed.}  (New York: Academic Press)
1783: %\par
1784: \bibitem{[3]} Budd C and Collins G An invariant moving mesh scheme for the
1785: nonlinear diffusion equation, to appear in Applied Numerical Mathematics.
1786: %\par
1787: \bibitem{[4]} Dorodnitsyn V A  1982  On invariant solutions of a
1788: nonlinear heat transfer equation with a source
1789: {\it Zh. Vychisl. Mat. i Mat. Fiz.}   {\bf 22}  1393  (in Russian)
1790: %\par
1791: \bibitem{[5]} Dorodnitsyn V A 1987 Taylor's group and transformations,
1792: conserving finite differences, Preprint of Keldysh Institute of Applied
1793: Mathematics, N67, Moscow. \\
1794: Dorodnitsyn V A 1988 Newton's group and commutative properties of
1795: Lie-Backlund operators in finite difference space, Preprint of Keldysh
1796: Institute of Applied Mathematics N175, Moscow. \\
1797: Dorodnitsyn V A  1991  Transformation groups in mesh spaces
1798: {\it J. Sov. Math.}  {\bf 55}  N1  1490
1799: %\par
1800: \bibitem{[6]} Dorodnitsyn V A  1993  Finite--difference models entirely
1801: inheriting symmetry of original differential equations  {\it Modern Group
1802: Analysis: Advanced
1803: Analytical and Computational Methods  in
1804: Mathematical Physics} (Kluwer Academic Publishers)  191
1805: %\par
1806: \bibitem{[7]} Dorodnitsyn V A  1993  Finite--difference analog of
1807: Noether's
1808: theorem  {\it Dokl. Akad. Nauk S.S.S.R.}  {\bf 328}   N6  678  (in Russian)
1809: %\par
1810: \bibitem{[8]} Dorodnitsyn V  1994  Invariant discrete models for the
1811: Korteweg-de Vries
1812: equation  {\it CRM--2187}  (Universite de Monreal)
1813: %\par
1814: \bibitem{[9]} Floreanini R and Vinet L  1995  Lie symmetries of
1815: finite difference
1816: equations  {\it J. Math. Phys.}  {\bf 36  (12)}  7024
1817: %\par
1818: \bibitem{[10]} Ibragimov N H  1995  {\it Transformation groups applied to
1819: mathematical physics}  (Dovdrecht: D. Reidel)
1820: %\par
1821: \bibitem{[11]}Levi D, Vinet L and Winternitz P 1997 Lie group formalism for
1822: difference equations, J. Phys. A: Math. Gen. { \bf 30}, 633--649
1823: %\par
1824: \bibitem {[12]} Maeda S 1987 The similarity method for difference equations,
1825: J. Inst. Math. Appl., {\bf 38}, 129
1826: %\par
1827: \bibitem{[13]} Ovsiannikov L V  1982  Group analysis of differential
1828: equations  (New York: Acadimic Press)
1829: %\par
1830: \bibitem{[14]} Ovsiannikov L V  1981  {\it Lections on gas dynamics}  (Moscow:
1831: Science)  (in Russian)
1832: %\par
1833: \bibitem{[15]} Ovsiannikov L V  1959  Group properties of a nonlinear heat
1834: equation
1835: {\it Dokl. Akad. Nauk S.S.S.R.}  {\bf 125}  N3  492  (in Russian)
1836: %\par
1837: \bibitem{[16]} Olver P J  1986  {\it Application of Lie groups to the
1838: differential equations}  (New York: Springer--Verlag)
1839: %\par
1840: \bibitem{[17]} Samarskii A A and Sobol I M  1963  Examples of numerical
1841: solutions of temperature waves  {\it Zh. Vychisl. Mat. i Mat. Fiz.}
1842: {\bf 3}  N4  702  (in Russian)
1843: %\par
1844: \bibitem{[18]} Samarskii A A, Galactionov V A, Kurdiumov S P and
1845: Mikhailov A P  1994  {\it Blow-up in problems for quasilinear parabolic
1846: equations}
1847: (Berlin: Walter de Gruyter)
1848: \end{thebibliography}
1849: 
1850: 
1851: \end{document}
1852: 
1853: