dispersion estimates for third order equations in two...
TRANSCRIPT
COMMUNICATIONS IN PARTIAL DIFFERENTIAL EQUATIONS
Vol. 28, Nos. 11 & 12, pp. 1943–1974, 2003
Dispersion Estimates for Third Order Equations
in Two Dimensions
Matania Ben-Artzi,1,* Herbert Koch,2 and Jean-Claude Saut3
1Institute of Mathematics, Hebrew University, Jerusalem, Israel2Fachbereich Mathematik, Universitat Dortmund,
Dortmund, Germany3UMR de Mathematiques, Bat, Universite de Paris-Sud,
Orsay, France
ABSTRACT
Two-dimensional deep water waves and some problems in nonlinear optics can be
described by various third order dispersive equations, modifying and generalizing
the KdV as well as nonlinear Schrodinger equations. We classify all third order
polynomials up to certain transformations and study the pointwise decay for the
fundamental solutions,ZR
2eitpð�Þþix�� d�
for all third order polynomials p in two variables. We deduce the corresponding
Strichartz estimates. These estimates imply global existence and uniqueness for
the Shrira system.
*Correspondence: Matania Ben-Artzi, Institute of Mathematics, Hebrew University,
Jerusalem 91904, Israel; E-mail: [email protected].
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
120025491_PDE28_11-12_R1_082903
1943
DOI: 10.1081/PDE-120025491 0360-5302 (Print); 1532-4133 (Online)
Copyright & 2003 by Marcel Dekker, Inc. www.dekker.com
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1943] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
1. INTRODUCTION
In has been known since 1968 (Zakhrov, 1968) that the stability of finite ampli-tude gravity waves of deep water is governed by a nonlinear Schrodinger equation(NLS). This is found by a perturbation analysis to O("3) in the wave-steepness"¼ ka� 1, where a is a typical wave amplitude and k the modulus of the meanwave number.
Taking perturbation analysis one step further to O("4), Dysthe (1979) hasderived a system which improves significantly the results relating the stability offinite amplitude waves. One of the dominant new effects is the wave-induced meanflow with potential �. Solving the equation for � in terms of the complex amplitudeof the wave packet allows to put the Dysthe system (in dimensionless variables) inthe following form, see Ghidaglia and Saut (1993):
2i@A
@tþ1
2
@A
@x
� �þ1
2
@2A
@y2�1
4
@2A
@x2� AjAj2
¼i
8
@3
@x3� 6
@3
@x@y2
!Aþ
i
2A A
@ �AA
@x� �AA
@A
@x
� ��5i
2jAj2
@A
@xþ AR1
@
@xjAj2 ð1:1Þ
where R1 is the Riesz transform in R2, that is
FðR1 Þ ¼ i�1j�j :
Here F denotes the Fourier transform.The usual NLS is obtained by neglecting the right hand side of Eq. (1.1), which
is of order "4 in the dimensional variables.A similar derivation of the fourth order (in ") evolution equations for the
amplitude of a train of nonlinear gravity-capillary waves on the surface of an idealfluid of infinite depth was performed by Hogan (1985). The equation reads
2i@A
@tþ cg
@A
@x
� �þ p
@2A
@x2þ q
@2A
@y2� �jAj2A
¼ �is@3A
@x@y2� ir
@3A
@x3� iujAj2
@ �AA
@xþ ivjAj2
@A
@xþ AR1
@
@xjAj2 ð1:2Þ
where cg is the group velocity and �, p, q, s, r, u, and v are real parameters dependingon the surface tension parameter. Note that q and s are strictly positive, while p canachieve both signs (in particular it is negative for purely gravity waves as in theDysthe system and positive for purely capillary waves).
A similar system has been obtained by Lo and Mei (1985) and Hara and Mei(1994) for finite depth gravity waves. Another example of a nonlinear Schrodingertype equation involving third order derivatives has been proposed by Shrira (1981)to describe the evolution of a three dimensional packet of weakly nonlinear internalgravity waves propagating obliquely at an arbitrary angle � to the vertical. If thedependence of the wave packet amplitude A on the transversal coordinate y is much
1944 Ben-Artzi, Koch, and Saut
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1944] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
slower than that on the x and z directions, one obtains the equation
i@A
@tþ!k
2
@2A
@x2þ!nn
2
@A
@z2þ !nk
@2A
@x@z
� i!kkk
6
@3A
@x3þ!kkn
2
@3A
@x2@zþ!knn
2
@3A
@x@z2þ!nnn
6
@3A
@z3
" #
þ i�A A@ �AA
@s� �AA
@A
@s
� �¼ 0 ð1:3Þ
Here @=@s ¼ cðsin�ð@=@xÞ � cos �ð@=@zÞÞ. The coefficients of the linear terms inEq. (1.3) can be computed explicitly as function of �. For nonzero constants �and � we have
!kkn ¼ � sin� cos�ð3� 5 cos2 �Þ
!nnn ¼ � sin� cos�ð3� 5 sin2 �Þ
!nnk ¼ ��ð5 sin2 � cos2 �� 2=3Þ
!kkk ¼ � sin2 �ð4� 5 sin2 �Þ
!kk ¼ �3� sin2 �
!kn ¼ � tan�ð2� 3 sin2 �Þ
!nn ¼ ��ð3 sin2 �� 1Þ
see Shrira (1981) and Ghidaglia and Saut (1993, Chap. 3). In particular !kkn and !knn
cannot vanish simultaneously.Similar problems occur in nonlinear optics (see for instance Zozulya (1999)), in
particular in the modeling of the dynamics of femtosecond laser pulses in a nonlinearmedia with normal dispersion. The evolution of the complex envelope E(x, y, z, t) ofthe field is described by the third order NLS
i@
@zE þ ð1� i"1@tÞ�?E �
@2E
@t2� i"2
@3E
@t3þ ð1þ i"1@tÞgðjEj
2ÞE ¼ 0 ð1:4Þ
where �? ¼ @2=@x2 þ @2=@y2 is the transverse Laplacian and "2 2 R, "1>0. As usualin nonlinear optics the evolution variable (which plays mathematically the role oftime) is z. The transverse Laplacian accounts for diffraction, while the second andthird time derivatives describe group velocity and third over dispersion.
Very little is known concerning the Cauchy problem for Eqs. (1.1)–(1.4). Theonly results available are local existence for analytic Cauchy data for Eqs. (1.1) and(1.2) by de Bouard (1993) and global existence for a very special case of Eq. (1.3), seeGhidaglia and Saut (1993).
The aim of the present article, as a first step towards solving the Cauchy problemin Sobolev classes, is to derive dispersion estimates for the linear group associated tothe linearized equation at 0. Those estimates have an independent interest and seem
Dispersion Estimates for Third Order Equations 1945
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1945] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
to be new for this class of third order symbols in 2 dimensions. On the other hand itis well known that Strichartz estimates are an important tool to solve the Cauchyproblem by a fixed point argument.
Thus we study linear third order equations, more precisely the decay ofthe fundamental solutions in x and t. This requires first a classification, second acalculation of several Fourier transforms, and third several applications of themethod of stationary phases.
The short time behavior is dominated by the homogeneous part (unlessit is strongly nongeneric) whereas the long time behavior is determined by thedegeneracy at the strongest singularity. For generic two-dimensional third-orderpolynomials the long time decay is t�3/4 and the generic short time bound is ct�2/3.
Stationary phase relates the map �! rpð�Þ with the decay of the fundamentalsolution. As side product we analyze the pointwise decay of the unique singularity(up to transformations) of codimension two (the singularity of codimension one hasnormal form �31, where the counting of the codimension is different from the one insingularity theory since linear terms do not change the decay). The decay estimatesimply Strichartz estimates, which in turn allow to prove global well-posedness for theCauchy problem for the Shrira system (1.3).
The article is organized as follows. In Sec. 2 we introduce notation. Section 3 isdevoted to a classification of third degree polynomials up to affine transformationsof coordinates and the addition of affine terms. This extends the work of Dzhuraevand Popelek, (Dzhuraev and Popelek, 1989 and 1991).
In Sec. 4 we give the inverse Fourier transform of eitp(�) for those polynomials forwhich the inverse Fourier transform can be computed in terms of exponentials,trigonometric functions and the Airy functions. The next section, Sec. 5, introducesthe Pearcey integral, which occurs for one of the phase functions, and studies onemore oscillatory integral, where a change of the contour of integration allows toobtain good estimates.
All this relies on the fact that we deal with very specific phase functions.The last two oscillatory integrals require natural but much deeper tools: We
have to study the contribution from general singularities of fold type and cusp type.In Sec. 6 we study local changes of coordinates and local normal forms for
nondegenerate points, folds, and cusps. Here Mather’s theory of stable mapsenters crucially. The reduction to normal forms allows to establish decay estimatesfor compactly supported amplitudes, which is done in Sec. 7. The next section, Sec. 8applies these results to obtain decay estimates for the inverse Fourier transforms ofeitp(�) for the remaining two polynomials.
Section 9 is standard: the decay estimates imply Strichartz estimates, some ofthem with smoothing. Well-posedness for the Shrira system is an easy consequence.We plan to return to the study of the other systems.
It should be noted that there are at least two motivations for the problems athand. First we establish sharp estimates for cusps—here the Pearcey integral plays asimilar role as the Airy function for holds. This is a natural step in the study ofoscillatory integrals. Even though the tools are available we are not aware of similarestimates in the literature, despite their importance for degenerate dispersive inte-grals. Secondly systems like the Shrira system have a strong motivation from physics,which makes a good analytical understanding highly desirable.
1946 Ben-Artzi, Koch, and Saut
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1946] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
2. NOTATION
We denote the Fourier transform by F and its inverse by F�1:
Ff ð�Þ ¼ ff ð�Þ ¼ ð2�Þ�n=2
Ze�ix�f ðxÞ dx
and
F�1ð�Þx ¼ ���ðxÞ ¼ 2ð�Þ�n=2
Zein��ð�Þ dx:
For x in Rn we set ~xx ¼ ðx1, . . . , xn�1Þ. If A � R
n is measurable we denote its volumeby |A|. The Lebesgue spaces are denoted by Lp.
We define Lpw by the quasinorm
k f kLpw¼ sup
�>0�jfx : j f ðxÞj > �gj1=p:
These spaces are Banach spaces if 1 < p <1. They occur in the context of theweak Young inequality (see Stein and Weiss (1971)):
Lemma 2.1. We have
k f � gkLr � ck f kLpkgkLqw
provided 1 < p, q, r <1 and
1
pþ1
q¼ 1þ
1
r:
In what follows, we shall derive space-time estimates for the oscillatory integral
Iðx, tÞ ¼
Zeitpð�Þþix��d�:
The real polynomial p (in � 2 R2) is a general cubic polynomial. However, for
the estimates the constants clearly play no role while the linear terms can bedispensed with by a suitable shift x to xþ at. Thus, we shall assume that p containsonly third and second order homogeneous parts. In the following section we shallreduce it to one of a class of normal forms (modulo linear terms), by suitable lineartransformations of �.
3. NORMAL FORMS
Homogeneous second degree polynomials are quadratic forms. There are onlytwo invariants: rank and signature. There are three equivalence classes, representedby �21, �
21 þ �
22, and �1�2.
Dispersion Estimates for Third Order Equations 1947
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1947] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
Note let p be homogeneous of degree three. Assuming p(1, 0) 6¼ 0 (otherwise werotate our coordinates) we can scale the �1 direction to obtain
pð�1, �2Þ ¼ �31 þ C�21�2 þ A�1�22 þ B�32:
We will keep this normalization for some time. We change coordinates to~��1 ¼ �1 þ ðC=3Þ�2 so that
pð�1, �2Þ ¼ ~�� 31 þ ~AA ~��1�22 þ ~BB�32 ð3:1Þ
with
~AA ¼ A�C2
3and ~BB ¼ B�
AC
3þ2C3
27:
We drop the tilde in the sequel and suppose that p is a polynomial of the form (3.1).Given a regular matrix D ¼ d11 d12
d21 d22
� �we consider
pDð�Þ ¼ pðD�Þ ¼ �1�31 þ �2�1�
22 þ �3�
21�2 þ �4�
32
where �i¼ �i(D). This defines a smooth map from the space of regular matrices toð�iÞ1�i�4 2 R
4. An immediate computation gives
pDð�Þ ¼ ðd311 þ Ad11d
221 þ Bd3
21Þ�31 þ ð3d2
11d12 þ Aðd12d221 þ 2d11d21d22Þ
þ 3Bd221d22Þ�
21�2 þ ð3d11d
212 þ Aðd11d
222 þ 2d12d21d22Þ þ 3Bd21d
222Þ�1�
22
� ðd312 þ Ad12d
222 þ Bd3
22Þ�32: ð3:2Þ
The derivative at the identity is gives by the matrix
3 0 0 00 3 2A 0A 0 3B 2A0 A 0 3B
0BB@
1CCA ð3:3Þ
which has determinant �ðA,BÞ :¼ 3ð27B2þ 4A3
Þ. This is zero if and only ifA/3¼�(B/2)2/3. Suppose that �(A,B) 6¼ 0. Using the implicit function theorem wecan find a nonsingular matrix if ð �AA, �BBÞ is close to (A,B) such that
pDð�Þ ¼ �31 þ �AA�1�22 þ
�BB�32:
A simple covering argument shows that the same is true for each ð �AA, �BBÞ in thepathcomponent of �(A,B) 6¼ 0. Obviously there are two path connected opencomponents ��(A,B)>0.
We choose two representative polynomials one for each component, therebychanging coefficient of �31 : ð1=6Þ�
31 � ð1=2Þ�1�
22 and ð1=6Þð�31 þ �
32Þ.
Now we consider the case A/3¼�(B/2)2/3. If A¼ 0 then the polynomial is �31.If not we may scale �2 so that A¼ 3. Then B¼�2 and changing �2 to ��2 if necessarywe arrive at
pð�Þ ¼ �31 � 3�1�22 þ 2�32: ð3:4Þ
1948 Ben-Artzi, Koch, and Saut
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1948] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
In summary, we have four representatives for the homogeneous third degreepolynomial. A short check shows that they are classified by the shape of the zero setof the determinant of the Hessian �! detD2pð�Þ, which is either the whole space, aline, the union of two transversal lines or a point. The shape is invariant under theoperations we used above. This consideration allows us to use the following set ofrepresentatives:
1
6�31,
1
2�1�
22,
1
6ð�31 þ �
32Þ and
1
6�31 �
1
2�1�
22: ð3:5Þ
Note that the second of those four representatives is replacing the one in Eq. (3.4).For notational simplicity we introduced coefficients different from 1. In what followswe refer to the four cases in Eq. (3.5) by Cases I–IV.
We now want to classify all polynomials with cubic and quadratic terms and tofind normal forms for each class. We shall use affine changes and the addition oflinear and constant terms, as long as they leave the cubic terms in Eq. (3.5) invariant.As has been observed earlier, the transformed polynomials are considered moduloaffine terms.
Thus, let p be a polynomial with cubic and quadratic terms. We set
d ¼ degree detðD2pÞ and S ¼ f� : detðD2pð�ÞÞ ¼ 0g
and consider the four cases separately.
Case I. If the coefficient of �22 is nonzero we may suppose that it is one. Then we havethe polynomial
1
6�31 þ b�21 þ a�1�2 þ �
22 ¼
1
6�31 þ ða�1=2þ �2Þ
2þ ðb� a2=4Þ�21: ð3:6Þ
We set �¼ �2þ a�1/2, plug it into the formula and rename � to �2. Then
pð�Þ ¼1
6�31 þ �
22 þ c�21 ð3:7Þ
and we may assume that c¼ 0 since the term can be eliminated by a shift in �1.This creates affine terms, which we neglect in our classification. It is easily seenthat d¼ 1.
Now we consider the case that the coefficient of �22 is zero and that of �1�2 is not.Then we may assume that it is one and we are left with ð1=6Þ�31 þ �1�2. Then d¼ 0 anddet(D2p) is constant but not zero. If both coefficients are zero then det(D2p)¼ 0.We conclude that Case I leads to three normal forms:
1
6�31,
1
6�31 þ
1
2�22, and
1
6�31 þ �1�2:
Case II. We shift �1 so that the coefficient of �22 is zero. Shifting �2 we achieve thesame for the coefficient of �1�2. We can normalize the coefficient of �21 and henceobtain the two representative polynomials ð1=2Þ�1�
22 and ð1=2Þ�1�
22 þ ð1=2Þ�21.
Dispersion Estimates for Third Order Equations 1949
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1949] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
Then d¼ 2 and S is a line or a parabola. The second case provides a normal form fora cusp. The notion of a cusp here is motivated by the fact that the image of S underthe map �! rpð�Þ is a geometric cusp. The inverse Fourier transform can beexpressed by the Pearcey integral (5.1).
Case III. The same arguments show that there are only two classes with normalpolynomials
pð�Þ ¼1
6ð�31 þ �
32Þ and pð�Þ ¼
1
6ð�31 þ �
32Þ þ �1�2:
Here d¼ 2 and S consists of a hyperbola or two lines.
Case IV. There are two classes with normal forms
1
6�31 �
1
2�1�
22 þ
1
2ð�21 þ �
22Þ, and
1
6�31 �
1
2�1�
22:
Here d¼ 2 and S is either a circle or a point. There cannot be any further reductionwhich can be seen by checking the zero sets of the determinant of the Hessian.
This completes the classification: The third order polynomials:
1
6�31 �
1
2�1�
22 þ
1
2�21 þ
1
2�22, ðIVÞ
1
6�31 þ
1
6�32 þ �1�2 ðIIIÞ, ð3:8Þ
will be discussed in Sec. 8. Their study is the most demanding and the most impor-tant part of this article. We recall that S is a circle (the first polynomial) resp. ahyperbola. The leading part of the asymptotic expansion of the oscillatory integral
Iðx, tÞ ¼
ZR
2eitpð�Þþix� d�
is expressed through the Airy function near points (x, t) for which there aredegenerate critical �c satisfying
xþ trpð�cÞ ¼ 0
for which the null space of D2p(�c) is transversal to S—this corresponds to values ofx/t lying in the image of S 3! �rpð�Þ, where the image is smooth. There exist,however, three cusps �0, �1 and �2 resp. one cusp �0. The leading part of theTaylorexpansion of tp(�) þ x� at these values is, in suitable coordinates,ðt=2Þð�21�2 þ �
22Þ þ x � �, which is the second polynomial in the list below. The leading
part of the expansion of I(x, t) close to directions
x=t ¼ �rpð�jÞ
is expressed using the Pearcey integral. Proving this fact requires deeper notionsand arguments from singularity theory. Even though the technique for getting theestimates is more or less standard it requires considerable work to carry it throughfor the specific case of the cusp.
1950 Ben-Artzi, Koch, and Saut
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1950] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
The polynomials
1
6�31 �
1
2�1�
22, ðIVÞ and
1
2ð�21 þ �1�
22Þ ðIIÞ ð3:9Þ
are discussed in Sec. 5. The oscillatory integral Ix,t with the second polynomial istransformed to the Pearcey integral. It is the model for the cusp singularity.
The inverse Fourier transforms of eitp with p from the lists
1
6�31 þ
1
6�32 ðIIIÞ,
1
2�1�
22 ðIIÞ,
1
6�31 þ �1�2 ðIÞ, ð3:10Þ
1
6�31 þ
1
2�22,
1
6�31 ðIÞ, �1�2
1
2ð�21 þ �
22Þ and
1
2�21 ð3:11Þ
will be given in the next section.
4. SOME FOURIER TRANSFORMS
Here we give the (inverse) Fourier transforms of several functions. For simplicitywe always assume t>0.
(1) Quadratic phase functions: Letffiffi�
pdenote the square root in the slit domain
C=R� and x, � 2 R. Then F�1ðeðit=2Þ�
2
ÞðxÞ ¼ ð1=ffiffiffiit
pÞe�ðx2=2tÞi. We define
KðxÞ ¼1ffiffiffiffiffiffiffi2�i
p eði=2Þx2
:
(2) The Airy function: The Airy function is defined by
AiðxÞ ¼ ð2�Þ�1=2F�1ðei�
3=3ÞðxÞ:
It is the unique bounded solution to
Ai00ðxÞ � xAiðxÞ ¼ 0 Aið0Þ ¼ 3�1=6 �ð1=3Þ
2�:
We have the estimates
jeffiffiffiffiffi�x
p 3
j ð1þ jxj1=4ÞjAiðxÞj þ ð1þ jxjÞ�1=4jAi0ðxÞj
� �� c
which can be seen by the WKB method or standard calculations.(3) Let x, � 2 R
2. Then
F�1ðexpðit�31=6ÞÞðxÞ ¼ ðt=2Þ�1=30ðx2ÞAiðx1=ðt=2Þ
1=3Þ:
(4)
F�1ðexpðitð�31=6þ �
22=2ÞÞÞðxÞ ¼
ffiffiffiffiffiffiffi2�i
pð2tÞ�5=6Aiðx1=ðt=2
1=3ÞÞKðx=
ffiffit
pÞ:
(5) F�1ðexpðitð�31 þ �1�2ÞÞ ¼ t�1eið�xð3=2Þ=t2�x1x2=tÞ, hence
F�1 exp it1
6�31 þ �1�2
� �� �� �¼
ffiffiffiffiffi6t
p �1eið�
ffiffi6
px32=t
2�6x1x2=tÞ:
Dispersion Estimates for Third Order Equations 1951
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1951] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
(6) The inverse Fourier transform of expðitð1=2Þ�1�22Þ is
F�1 exp it1
2�1�
22
� �� �¼
0 if x1 01
�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�tx1=2
p cosðx2ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�2x1=tÞ
pif x1 < 0
8<:
See also Fedoryuk (1977).(7) F�1
ðexpðit16ð�31 þ �
32ÞÞÞ ¼ 2�ðt=2Þ�2=3Aiðx1=ðt=2Þ
1=3ÞAiðx2=ðt=2Þ
1=3Þ:
5. TWO OTHER CASES
We recall the lemma of van der Corput in the form of Stein (1993).
Proposition 5.1. Suppose that k 2, f 2 Ck and f (k) 1 on R. Let 2 C10ðRÞ. ThenZ
R
eitf ð�Þ ð�Þ d�
�������� � ct�1=k
k 0kL1 :
This can be extended to noncompactly supported functions. In that case the integralhas to be understood as suitable limit of truncated integrals.
We begin with the prototype of a phase function where we have to contentourselves with estimates for large x. Let
Bð y, zÞ :¼
ZR
eið1=24s4þ1=2ys2þzsÞ ds ð5:1Þ
for y, z 2 R. This integral was introduced by Pearcey (1946). The level lines of itsmodulus are shown in Arnold et al. (1999). In contrast to the previous cases there isan explicit formula for the Fourier transform. We collect estimates for B below:
Lemma 5.2. We have
jBð y, zÞj � cð1þ jzj2 þ j yj3Þ�1=18½1þ ð1þ jzj2 þ j yj3Þ�5=9
jð3zÞ2 þ ð2yÞ3j�1=4:
Proof. Set
pðsÞ ¼1
24s4 þ
1
2ys2 þ zs:
Since p(4)¼ 1 the van der Corput lemma implies |B( y, z)|�C. It suffices there-fore to take
� :¼ ð3zÞ2 þ ð2j yjÞ3 1:
We normalize the integral by
~yy ¼ ��1=3y, ~zz ¼ ��1=2z; and ~ss ¼ ��1=6s,
1952 Ben-Artzi, Koch, and Saut
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1952] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
so that, with
~ppð~ssÞ ¼1
24~ss 4 þ
1
2~yy~ss2 þ ~zz ~ss,
Bð y, zÞ ¼ �1=6ZR
ei�2=3 ~ppð ~ss Þ d ~ss
¼ �1=6Z�ð~ssÞei�
2=3 ~ppð~ss Þ d ~ssþ
Zð1� �ð~ss Þei�
2=3 ~ppð~ssÞ d ~ss
¼ �1=6½B1ð ~yy, ~zzÞ þ B2ð ~yy, ~zz Þ
where � 2 C10 is identically 1 for all arguments j~ssj � 10. Then, since the derivative of
~pp is bounded from below outside [�5, 5],
jB2ð ~yy, ~zzÞj � xk��k forall k 2 N uniformly in ~yy, ~zz:
The cubic equation
~pp0ð~ss Þ ¼1
6~ss3 þ ~yy~ssþ ~zz ¼ 0
has (for given ð ~yy, ~zzÞ 6¼ ð0, 0Þ) either one, two or three roots, depending on whetherthe discriminant
D ¼ ð3 ~zzÞ2 þ ð2 ~yyÞ3
is positive, zero, or negative. If D then,
j ~pp0ð~ss Þj þ j ~pp00ð~ss Þj cðÞ
and, by the method of stationary phase
jB1ð ~yy, ~zz Þj � c��1=3:
The third derivative of ~ff is uniformly bounded from below at zeroes of ~pp00. Hence
jB1ð ~yy, ~zzÞj � c��2=9
for all y and z. We need however a more precise estimate if D< . Arguing as forfold singularities (see Hormander (1983) and Sec. 7.5 below) we obtain
jB1ð ~yy, ~zzÞj � c��1=18ð1þ �4=9jDjÞ
�1=4
hence
jBð y, zÞj � ��1=18½1þ ��5=9
ðj3zÞ2 þ ð2yÞ3jÞ�1=4:
Remark 5.3. It is not hard to prove the estimates of Lemma 5.2 by elementarycalculation and the van der Corput lemma. The key is the algebraic estimate
j ~pp0j þ j ~pp00j cjDj1=2, ð5:2Þ
Dispersion Estimates for Third Order Equations 1953
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1953] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
which implies the estimate by the van der Corput lemma. It suffices to verify Eq. (5.2)at critical points. There the second derivative of ~pp is of size of the square root of thediscriminant, which implies Eq. (5.2)
The cusp singularity. Here we consider pð�Þ ¼ 1=2ð�1�22 þ �
21Þ. The phase function
tpð�Þ þ x�
has degenerate critical points for x/t in a cusp. Figure 1 (a) shows the setS¼ {�|rkD2p(�)<2} together with the kernel of the Hessian D2
�p and Fig. 1 (b)shows its image and the image of close by level curves of det(D2p) under the map� ! �rpð�Þ.
Let
ItðxÞ ¼ ð2�Þ�1
ZR
2eiðt=2Þð�1�
22þ�
21Þþi�x d�:
Then
Itðx1, x2Þ ¼ t�3=4I1ðx1t�1=2, x2t
�1=4Þ ð5:3Þ
and, by performing the integration with respect to �1,ZR
2ei1=2ð�
21þ�1�
22Þþix� d� ¼ 2��1=2
ZR
ei1=2ð1=2�22þx1Þ
2þix2�2 d�2
¼ ceix21=2Bð3�1=2x1, 3
�1=4x2Þ
where B is the Pearcey integral (5.1). The estimate
kItkL1 � ct�3=4
follows from Lemma 5.2.
The homogeneous integral. pð�Þ ¼1
6�31 �
1
2�1�
22. Let
ItðxÞ ¼
Zeitpð�Þþix� d�: ð5:4Þ
where pð�Þ ¼ 1=6�31 � 1=2�1�22. Then scaling shows
ItðxÞ ¼ t�2=3I1ðxt�1=3
Þ
which motivates the definition
CðxÞ ¼ I1ðxÞ:
Proposition 5.4. The following estimates hold:
jCðxÞj � cð1þ jxjÞ�1=2
jrxCðxÞj � c:
F1
1954 Ben-Artzi, Koch, and Saut
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1954] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
−0.5 0 0.5 1 1.5 2 2.5 3−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
xi
XI
Figure 1ðaÞ. The set S and the kernel of D2�p.
−3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5−3
−2
−1
0
1
2
3
x
y
Figure 1ðbÞ. The image of S 3 �! �rpð�Þ.
Dispersion Estimates for Third Order Equations 1955
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1955] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
The following corollary is an immediate consequence.
Corollary 5.5. We have
jItðxÞ � ct�2=3ð1þ jxjt�1=3
Þ�1=2
and
jrxItðxÞj � ct�1
Proof of Proposition 5.4. There are many ways to prove the estimate. We choose achange of the contour of integration, because this argument carries over to theinhomogeneous phase function and large x/t with only minor changes. For thesame reason we do not make use of scaling here. Let
�ð�Þ ¼1
2j�jð�21 � �
22Þ, �
1
j�j�1�2
� �:
Then j�j� ¼ rp and
j�j2 ¼1
j�j21
4ð�21 � �
22Þ
2þ �21�
22
� �¼
1
4j�j2
and
Im pð� þ i�ð�ÞÞ ¼1
j�jjrpj2 � pð�Þ
1
4j�j3 �
1
6j�j3
1
5j�j3:
We change the contour of integration:
CðxÞ ¼
Zexpðipð� þ i�Þ þ ixð� þ i�ÞÞ d�
hence for all x
jCðxÞj
Ze�1=5j�j3þ1=2jxk�jd� <1:
We want to show that C(x) decays when x is large. Let x 6¼ 0. We identify R2 with C
and define
�xð�Þ :¼�1 � i�2 �
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi2ðx1 þ ix2Þ
p� ��1 � i�2 þ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi2ðx1 þ ix2Þ
p� �2
ffiffiffiffiffiffijxj
pþ j�j
� � ¼r�ð pð�Þ � x�Þffiffiffiffiffiffi
jxjp
þ j�j:
Let �c be one of the two critical points of �! p(�)þ x�. Then we have j�cj ¼ 2ffiffiffiffiffiffijxj
p
and
Im pð� þ i�Þ j�2 � �2c j
2
8ðffiffiffiffiffiffijxj
pþ j�jÞ
ffiffiffiffiffiffijxj
pminfj� � �cj
2, j� þ �cj2g:
Moving the domain of integration as above gives
jCðxÞj � cð1þ jxjÞ�1=2:
1956 Ben-Artzi, Koch, and Saut
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1956] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
The estimate for the gradient is straightforward: there is an additional factor 2ffiffiffiffiffiffijxj
p.
We obtain the following result when we combine Lemma 2.1 with the previousestimates to u(t)¼S(t)u0 where S is the group defined by the differential equation(where we plug in i@xj for D)
@tu� ipðDÞu ¼ 0:
Theorem 5.6. Consider the integral (5.4), where p is the polynomial 1=6�31 � 1=2�1�22.
1. Let 1� r� q�1. The following estimates hold:
kuðtÞkLq � ct�2=3ð1=r�1=qÞku0kLr
if
1
r�1
q
3
4:
If the difference is 3=4 we require that 1< r<q<1.2. Moreover
krukL1 � cjtj�1kuð0ÞkL1 :
6. SOME STABLE MAPS
It remains to study the two polynomials p1ð�Þ ¼ 1=6�31 � 1=2�1�22þ 1=2ð�21 þ �
22Þ
and p2ð�Þ ¼ 1=6�31 þ 1=6�32 þ �1�2. Let Jið�Þ ¼ detD2�pið�Þ, i¼ 1.2. Then
J1ð�Þ ¼ 1� j�j2, J2ð�Þ ¼ �1�2 � 1
which vanish respectively on a circle or on a hyperbola. Let Si be the set of pointsx 2 R
2 where thee is a degenerate critical point of �! pið�Þ þ x�. The set Si is onedimensional. It has three cusps if i¼ 1 and it consists of two unbounded curves ifi¼ 2, one of which contains one cusp. The leading part in suitable coordinates isgiven by the cusp singularity discussed above. It will turn out that the asymptoticbehavior of the Pearcey integral (5.1) determines the asymptotic behavior of theFourier cotransforms of eitp(�), t>0, but this requires some singularity theory aswell as general results about oscillatory integrals.
More generally we will be concerned with the problem of bounding oscillatoryintegrals of the type
It, ðxÞ ¼
ZR
n ð�Þeitð f ð�Þþx��Þ d�
where 2 C10 ðR
nÞ and where f is a smooth function. Suppose that
f ð�Þ þ x � � ¼ f0ð y, �Þ þ �ð yÞ ð6:1Þ
Dispersion Estimates for Third Order Equations 1957
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1957] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
where x¼ x( y) and �¼ �( y, �) in the support of . Then
It, ðxÞ ¼ eit�ð yÞZR
n ð�ð y, �ÞÞdet
@�
@�
� �eitf0ð y, �Þ d�:
This observation decomposes the study of oscillatory integrals into two parts:a classification of the relevant normal forms for the phase function, and anestimation of oscillatory integrals with these phase functions. In this section weclassify the phase functions which are relevant. We follow the work ofDuistermaat (1974) and Mather (1968, 1969). From the discussion above it isclear that we should (and do) replace f(�)þ x�� by more general functions f(x, �).
Definition 6.1. A function f0(x, �) is called stable at (x0, �0) if there exists ", k, and such that for every smooth function f with
k f � f0kCkðBðx0, �0ÞÞ� "
there exist local diffeomorphisms ( y, n) – (x( y), �( y, �)) (which, together with itslocal inverse, are defined in B(x0)�B(�0) and � 2 C1
ðBðx0ÞÞ satisfying (6.1) in aneighborhood of (x0, �0).
Definition 6.2. We call f0 infinitesimally stable at (x0, �0) if for every functionf 2 C1
ðRn� R
nÞ there exist smooth functions gj, hj, and � with
f ðx, �Þ ¼Xnj¼1
gjðx, �Þ@�j f0 þXnj¼1
hjðxÞ@xj f0 þ �ðxÞ
for x and � in a neighborhood of (x0, �0).Given a function f we define the set
Sf0 ¼ fðx, �Þj@�j f ðx, �Þ ¼ 0 for 1 � j � ng
Proposition 6.3. (Theorem 2.1.5 of Duistermaat (1974)). Suppose f0 is smooth near(x0, �0) and that the projection S� 3 ðx, �Þ ! x is proper. Then f0 is stable at (x0, �0) ifand only if it is infinitesimally stable at (x0, �0).
Proof. The implication stable implies infinitesimally stable is obvious. The otherdirection has been proven (in much greter generality) by Mather (1968 and 1969),and closer in the form needed here, by Duistermaat (1974). œ
It is instructive to look at examples. Let A be a symmetric nondegenerate realmatrix. Then
f0ðx, �Þ ¼1
2�tA� þ x � �
is stable for every point (x, �). To see this we assume that f is close to f0 and we haveto find g¼ (g1, . . . , gn), h¼ (h1, . . . ,hn) and �
f ðx, �Þ ¼ gtðx, �ÞðA� þ xÞ þ ht� þ �ðxÞ
1958 Ben-Artzi, Koch, and Saut
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1958] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
in a neighborhood of (x0, �0). This is trivial if
A�0 þ x0 6¼ 0
and we restrict ourselves to A�0¼�x0 in the sequel. Let �¼ �� �0 and y¼ x� x0.Then the problem reduces to the same problem near (0, 0). For simplicity we assumethat A is the identity (otherwise we first change coordinates so that A is a diagonalwith entries � 1 in the diagonal). Let
~ff ðx, �Þ ¼ f ðx, �� xÞ:
Then it suffices to fine ~ggðx, �Þ and � such that
~ff ðx, �Þ ¼ ~ggtðx, �Þ � �þ �ðxÞ:
We set
�ðxÞ ¼ ~ff ðx, 0Þ
and
~ggðx, �Þ ¼
Z 1
0
r� f ðx, t�Þ dt:
The same arguments show that f ð�Þ þ x � � is stable at (x0, �0) if D2f(�0))ij isnondegenerate.
Now let A be a symmetric nonsingular real (n� 1)� (n � 1) matrix. Let ~�� ¼ð�1, . . . , �n�1Þ. We claim that the normal form for a fold
f0ðx, �Þ ¼1
6�3n þ
1
2~��tA ~�� þ x � �
is stable at (0, 0). We again restrict ourselves to the case A ¼ idR
n�1 .We want to verify infinitesimal stability which amounts to finding g, h, and � so
that (with f a given function)
f ðx, �Þ ¼Xn�1
j¼1
gjðx, �Þð�j þ xjÞ þ gnðx, �Þ1
2�2n þ xn
� �þ htðxÞ � � þ �ðxÞ:
We begin with the one-dimensional case. Given ~ff we want to find ~gg, ~hh, and ~�� with
~ff ðs, tÞ ¼ ~ggðs, tÞ1
2t2 þ s
� �þ ~hhðsÞtþ ~��ðsÞ
near (0, 0). This cane be done by the Malgrange preparation theorem, which is theappropriate tool in general. Here we may also use a more elementary construction.It is not hard to see that
~ff �1
2t2, t
� �¼ ~hh �
1
2t2
� �tþ ~�� �
1
2t2
� �
hence
~�� �1
2t2
� �¼
1
2~ff �
1
2t2, t
� �þ ~ff �
1
2t2, � t
� �� �
Dispersion Estimates for Third Order Equations 1959
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1959] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
and
~hh �1
2t2
� �¼
1
2t~ff �
1
2t2, t
� �� ~ff �
1
2t2, � t
� �� �:
We extend both functions smoothly to positive arguments. The definition of ~gg isstraight forward.
We can do the same construction in the higher dimensional case as long aswe restrict ourselves to the subspace where �1. . . �n�1¼ 0. Then we repeat theconstruction for quadratic phase functions above to obtain the remaining functions.
It is clear that the same arguments imply stability for (with 3< k� nþ 1)
f0ðx, �Þ ¼1
2~��tA ~�� þ �kn þ x � � þ
Xk�3
j¼1
xj�jþ1n :
This is a normal form for a cusp if k¼ 4.Moreover, if f is stable at (x0, �0) and we add a nondegenerate phase function in
additional variables, then we obtain a stable singularity.Finally we consider
f0ðx, �Þ ¼1
2�1�
22 þ
1
2�21 þ x�:
We claim that f0 is stable at (0, 0). To see this we change variables: let
�1 ¼ �1 þ1
2�22 and �2 ¼ �2:
Then
f0ðx, �Þ ¼ ~ff ðx, �Þ ¼1
2�21 �
1
8�42 þ x1�1 �
1
2x1�
22 þ x2�2:
This function is infinitesimally stable near (0, 0). The properness assumption isclearly satisfied. Hence f0 is stable.
We collect the results:
Lemma 6.4. The following phase functions are stable at (0, 0):
(1) f ðx, �Þ ¼ �ta� þ x � � (if A is symmetric and invertible),(2) f ðx, �Þ ¼ 1=6�31 þ ���tA� þ x � � (where A is a(n� 1)� (n� 1) matrix,
symmetric and invertible),(3) f ðx, �Þ ¼ �41 � �
22 þ x2�
21 þ x � � and
(4) f ðx, �Þ ¼ 1=2�21�2 þ 1=2�22 þ x � �.
7. ESTIMATES FOR OSCILLATORY INTEGRALS
In this section we prove estimates for oscillatory integrals with degenerate phasefunctions (not necessarily polynomials).
1960 Ben-Artzi, Koch, and Saut
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1960] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
We introduce some notation. Let U � Rn be open and f 2 C1
ðUÞ.
Definition 7.1. A (Nondegenerate). We say that f is nondegenerate at �0 if D2f isnondegenerate at �0, or, equivalently, if the map
� ! rf ð�Þ
is nondegenerate near �0. Let
J ¼ detðD2f ð�ÞÞ:
Then f is nondegenerate at �0 iff J(�0) 6¼ 0.At a degenerate point �0 we assume that J has a simple zero. As a consequence,
its zero level-set is a smooth hypersurface S, and the null-space of D2f is one-dimen-sional at all points � 2 S. Let X 2 kerðD2f Þ be a nontrivial vectorfield, defined alongS i.e., at every point in S, but not necessarily in TS.
B (Fold). We say that f has a fold at the degenerate point �0 if X is transversal to S at�0., i.e., if X � rJ 6¼ 0 at �0.
C (Cusp). We say that f has a cusp at the degenerate point �0 if
(1) X � rJ has a simple zero at �0 (as a function on S ) and:(2) Let S1 be the zero level-set of X � rJ in S. Then X is transversal to S1.
D (Global). We say that f is nondegenerate if it nondegenerate at every point. It hasat most folds if at every point it is either nondegenerate or it has a fold. It has at mostcusps if at every point it is either nondegenerate, or it has a fold or a cusp.
Remark 7.2. Nondegeneracy at �0 is a condition on the Taylor expansion up tosecond order, existence of a fold at �0 is decided by the Taylor expansion up toorder 3 and the conditions for a cusp require a Taylor expansions up to order 4.Nondegeneracy implies that the map
� ! rf ð�Þ
is invertible in a neighborhood of �0. If �0 is a fold then
S 3 �! rf ð�Þ
parameterizes a smooth hypersurface (since the derivative of (rf )|S has full rank)and the image of
� ! rf ð�Þ
lies locally on one side of S—which is a consequence of the results below. If �0 is acusp then
S1 3 � ! rf ð�Þ
parameterizes a smooth submanifold of codimension two and the image ofS 3 � ! rf ð�Þ is a geometric cusp near rf(�0), which is again a consequence ofthe considerations below.
A fold corresponds to the singularity A2 and a cusp to A3 in the notation ofArnol’d et al. (1985), Sec. 11.1.
Dispersion Estimates for Third Order Equations 1961
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1961] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
Remark 7.3. We will apply the estimates below only to functions with two dimen-sional � variable. In that case S1 will consist of isolated point.
Remark 7.4. We shall see that the functions above are closely related to the onesdiscussed in the last section.
Theorem 7.5. Let 2 C10 ðUÞ and
It, ðxÞ ¼
ZU
ð�Þeitf ð�Þþix� d�:
1. Suppose that f is non-degenerate in the support of . Then
kIt, kL1ðRnÞ � ct�n=2 and kIt, kL1ðRnÞ � ctn=2:
2. Suppose that f has at most folds in the support of . Then
kIt, kL1ðRnÞ � ct�n=2þ1=6, kIt, kL4wðR
nÞ � ct�n=4, and kIt, kL1ðRnÞ � ctn=2:
3. Suppose that f has at most cusps in the support of . Then
kIt, kL1ðRnÞ � ct�n=2þ1=4, kIt, kL4wðR
nÞ � ct�n=4, and kIt, kL1ðRnÞ � ctn=2:
Remark 7.6. Estimates for other values of p follow by interpolation. Then for p<4the estimate for kIt, kLp are the same in all cases.
Proof. We begin with the observation that we have for every integer
jIt, ðxÞj � cNð1þ jtj þ jxjÞ�Nð7:1Þ
for x with dist(�x/t,D) 1 where D ¼ frf ð�Þ : � 2 supp g.A standard application of stationary phase shows that for the nondegenerate
phase function
jIt, ðxÞj � ct�n=2:
This implies the first statement.Using a suitable partition of unity it suffices to prove the statements for ampli-
tudes supported in a small neighbourhood of a given point. The proof of the secondstatement will be achieved by a (\,finite) sequence of lemmas.
Step 2 folds. Let �0 be a point where f has a fold. We change coordinates so that�0¼ 0, S¼ {� : J(�)¼ 0} is tangential to R
n�1 and ð0R
n�1 , 1Þ is in the null space ofD2f(0). We may and do assume that is supported in B"(0) for a small number "and f(0)¼�f(0)¼ 0. Then, as above,
jIt, ðxÞj � cN, "ð1þ jtj þ jxjÞ�N
if |x| 2"|t|.
1962 Ben-Artzi, Koch, and Saut
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1962] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
After a linear change of coordinates there exists a symmetric nondegenerate(n� 1)� (n� 1) matrix A such that
f ð�Þ ¼1
2~��tA ~�� þ
1
6�3n þOðj ~��j2j�j þ j�nj
4Þ:
Here ~�� denotes that first n� 1 components of the vector. Now we define�� ¼ ð�1=2 ~��, �1=3�nÞ and f�(�)¼ �
�1f�(��). Then
k f�ð�Þ �1
2~��tA ~�� �
1
6�3nkCN ðB1ð0ÞÞ
� c�1=3:
for all N 2 N. The function f0 ¼ 1=2 ~��tA ~�� þ 1=6�3n þ x� is stable. Thus, if � is small wecan change coordinates according to the last section so that we have the phasefunction f0 and an amplitude depending on x/t and �. Then estimate follows fromLemma 7.7 below.
Lemma 7.7. Let A be a nondegenerate symmetric (n� 1)� (n� 1) matrix and let 2 C1
0 ðB1ð0Þ � B"ð0ÞÞ be a smooth compactly supported function. Then
It, ðxÞ : ¼
ZR
n ðx=t, �Þeitð1=2
~��tA ~��þ1=6�3nÞþix� d�
¼ t�n=2þ1=6 ðx=t, � ½A�1 ~xx=t, 0Þeð ~xxtA�1 ~xxÞ=2ti
ffiffiffiffiffiffiffiffiffiffiffiffidet iA
p �1Aiðxn=ð2tÞ
1=3Þ
þOðt�n=2Þ:
where ½A�1 ~xx=t, 0 denotes the obvious vector in Rn.
Proof. We may assume that x/t is small. We haveZR
n�1 ðx=t, �Þeitð1=2Þ
~��tA ~��þi ~xx� ~�� d ~��
¼ ð4�tÞð1�nÞ=2ðdetðiAÞÞ�1=2 ðx=t, � ~xx=t, �nÞe
ið ~xxtA�1 ~xxÞ=2t
þ
Z½ ðx=t, �Þ � ðx=t, � ~xx=t, �nÞe
it=ð1=2Þ ~��tA ~��þi ~xx ~��d ~��:
The second term in the right hand side is O(t�n/2). This calculation reduces theproblem to the one dimensional one:
Lemma 7.8.ZR
ðs, Þeitð1=6s3þsÞ ds ¼ ð0, 0Þt�1=3Aiðt2=3Þ þOðt�2=3
Þ
if 0 and, for <0,
ZR
ðs, Þeitð1=6s3þsÞ ds ¼
ðffiffiffiffiffiffiffiffiffiffiffiffi�=3
p, 0þ ð�
ffiffiffiffiffiffiffiffiffiffiffiffi�=3
p, 0Þ
2t�1=3Aiðt2=3Þ þOðt�2=3
Þ:
Dispersion Estimates for Third Order Equations 1963
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1963] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
The proof of this estimate is straight forward. See also Theorem 7.7.18(Hormander, 1983) for related estimates. œ
Now, if ct�n/2� �� ct�n/2þ1/6
fx : jIt, ðxÞj > �g � fx : jxj � ct, jxnj � c��4t1�2ng
hence, since |It, (x)|� ct�n/2þ1/6 and since the estimate is obvious for smaller �,
jfx : jIt, ðxÞj > �gj � c��4t�n
and
kIt, k4L4w¼ sup �4jfx : jIt, ðxÞj > �gj � ct�n:
This is the L4w estimate. The L1 estimate is much simpler and we omit it.
Step 3, cusps. We suppose that f has a cusp at �0. We shift �0 to zero. Without loss ofgenerality we assume that f ð0Þ ¼ rf ð0Þ ¼ 0. By assumption J has a simple zero at 0.We make a linear change of coordinates so that X¼ en at �¼ 0 and rJð0Þ is amultiple of e1. Thus the tangent space of S at �¼ 0 is orthogonal to e1. LetA¼D2f(0) with components aij and ~AA ¼ ðaijÞ1�i, j�n�1. The Taylor expansion is
f ð�Þ ¼1
2aij�i�j þ
1
6cijk�i�j�k þ
1
24dijkl�i�j�k�l þOðj�j5Þ
where cijk and d i jkl are symmetric in all coefficients, the summation convention isused, and, by assumption, ain¼ ani¼ 0 for 1� i� n and ~AA is invertible. Then
J ¼ cnni�i det ~AAþ oðj�jÞ
and hence cnni¼ 0 for 2� i� n and cnni 6¼ 0. After scaling �1 we may and do assumethat cnn1¼ 1. Then
ðD2f Þij ¼ aij þ cijk�k þ1
2dijkl�k�l þOðj�j3Þ
and, since by definition on S
0 ¼ ðD2f ÞijXj ¼ aijXj þ cijk�kXj þOðj�j2Þ
and since we may set Xn¼ 1 we have (on S) for 1� i� n� 1
Xi ¼ � ~AA�1ij cnjk�k þOðj�j2Þ:
We compute
J ¼ detð ~AAÞ �1 þ1
2d mnij�i�j � ~AA�1
ij cnik�kcnjm�m þ ~AA�1
ij �k�1
� �þOðj�j3Þ
where the sum runs from 1 to n� 1 if ~AA is involved. Thus
rJ � X ¼ detð ~AAÞðd nnni�i � 3 ~AA�11j c
njk�kÞ þOðj�j2Þ:
1964 Ben-Artzi, Koch, and Saut
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1964] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
The singularity is a cusp provided
d nnnn6¼ 3 ~AA�1
11 :
We set
�i ¼ �i þ1
2~AA�11i �
2n, �n ¼ �n
which gives
f ð�Þ ¼1
2aij�i�j þ
1
24ðd nnnn
� 3 ~AA�111 Þ�
4n þOðj�j5 þ j ~��j2j�jÞ
and
f ð�Þ þx
t� � ¼
1
2aij�i�j þ
1
24ðd nnnn
� 3 ~AA�111 Þ�
4n þ
x
t� ��
1
4�2n ~AA�1
1i
xitþOðj�j5 þ j ~��j2j�jÞ:
For notational simplicity we restrict to ~AA ¼ 1R
n�1 and ðd nnnn� 3A�1
11 Þ ¼ 1. The lead-ing part is stable by Lemma 6.4. Scaling ~��i ¼ �1=2�i for 1� i� n� 1 and ~��n ¼ �1=4�nwe see as above that, at the expense of having an amplitude � dependingon y ¼ �ðx=tÞ as well, we may suppose that we have the phase function
f ð�Þ þ x=t � � ¼1
2j ~��j2 þ
1
24�4n þ
1
2y1�
2n þ y � �:
We integrate over ~�� and obtain for a suitable function �
It, ðxÞ ¼ ðitÞ�ðn�1Þ=2eitj ~yyj2=2
ZR
�ð y, ½ � ~yy, sÞeitð1=24s4þ1=2y1s
2þynsÞ dsþOðt�n=2
Þ:
We continue with the proof of Theorem 7.5, Step 3. Let
Mt, � ¼ fx : jIt, j > �g:
By Eq. (7.1) we have for n 2, 1� p� 8, and some c>0
Z ðctÞ�1
�p�1jMt, � \ ðR
nnBctð0Þjc� � cNt
�N :
where |.| denotes the volume. Clearly Mt,� is empty if �>> t�n/2þ1/4.The L4
w estimate follows from
jMt, � \ Btð0Þj � ct�n��4: ð7:2Þ
This estimate is trivial for �� t�n/2 and it remains verify Eq. (7.2) for
c�1t�n=2� � � ct�n=2þ1=4: ð7:3Þ
According to Duistermaat (1974) there exists a smooth function � with compactsupport and new coordinates y ¼ �ðx=tÞ with
det@y
@x
� �� t�n
Dispersion Estimates for Third Order Equations 1965
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1965] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
such that locally
It, ðxÞ ¼ t�n=2þ1=4�ð ~xx=tÞeit ~yyt ~AA ~yyBðt1=2y1, t
3=4ynÞ þOðt�n=2Þ:
where B is the Pearcey integral (5.1). The O(t�n/2) term is easily dominated.After localization and a change of coordinates it suffices to show for
~MMt, � ¼ f y 2 R2k yj � 1, jBðt1=2y1, t
3=4y2jÞ > t2n�1=4�g:
with
t�n=2� � � t�n=2þ1=4
the estimate
j ~MMt, �j � ct�2n��4: ð7:4Þ
We set R¼ z2þ |y|3 and T¼ z2þ y3. By Lemma 5.2 it suffices to prove
kfð y, zÞ 2 B1ð0ÞjR�1=18
½t�2=3þ R�5=9
jT j�1=4 > �gj � c��4:
for
1 << � � t1=4:
This set is contained in
A1 ¼ fð y, zÞkz2 þ y3j � ��4R1=3g, and in A2 ¼ fð y, zÞjðz2 þ j yj3Þ < ��24=5
g:
The second set has size
jA2j � ��4
and, if ð y, zÞ 2 A1, for fixed |z| 1/2��2, then y lies in an interval of size
�y � ��4z�2=3:
This implies Eq. (7.4), and hence the L4w estimate. The L1 and L8 estimate are simple
consequences. œ
8. THE REMAINING CASES
The classification of third order polynomials has led to the list (3.8)–(3.11). Wehave seen that there are explicit formulas for the oscillatory integrals of Sec. 2 for pas in Eqs. (3.10) and (3.11). The polynomial in list (3.9) have been studied above.It remains to study the oscillatory integrals with p in the list (3.8).
A. The case pð�Þ ¼ 1=6�31 � 1=2�1�22 þ 1=2ð�21 þ �
22Þ: This is a second order
perturbation of Eq. (5.4). The second order term changes the long term asymptotics.
1966 Ben-Artzi, Koch, and Saut
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1966] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
The degenerate singularity of the previous example is broke up into three cusps,see Fig. 1 (c) for S and the null space D2
�p and (d) for the image of �rp|s andcloseby lines.
We begin with a discussion of the mapping properties of � ! rp and of degen-erate points. We have detðD2
�pÞ ¼ 1� j�j2, which vanishes on the circle of radius 1.There are three points where the kernal of D2
�p is tangent to that curve. Those are themost degenerate sets. They are mapped to three points (�1, 0), ð1=2,
ffiffiffi3
p=2Þ, and
ð1=2, �ffiffiffi3
p=2Þby � ! �rp in the x space, which are connected through the image
of the arcs of the circle. Every point outside has two preimages (by mapping degreearguments) and every point inside has four preimages. The map of the circle to thatcurve is a homeomorphism. This is discussed in detail in Arnol’d et al. (1985).
Suppose that |x|/t� 3. Let � 2 C10 ðR
2Þ, �ð�Þ ¼ 1 for |�|� 10 and
It,�ðxÞ ¼
Z�ð�Þeitpð�Þþx�� d�:
Then by Theorem 7.5
kIt,�kL1 � cð1þ tÞ�3=4
and
kIt,�kL4!� cð1þ tÞ�1=2:
−1.5 −1 −0.5 0 0.5 1 1.5−1.5
−1
−0.5
0
0.5
1
1.5
xi
XI
Figure 1ðcÞ. The set S and the kernel of D2�p:
Dispersion Estimates for Third Order Equations 1967
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1967] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
We claim that for the points x under considerationZð1� �ð�ÞÞeitpð�Þþx�� d�
�������� � ct�N :
Indeed, let (�) be supported in B(0, 20) and (�)¼ 1 for |�|� 10. We define
� ¼ ð1� ð�ÞÞrpð�Þ
1þ j�j
and deform the contour of integration. We decompose the integral into one where wemultiply the integrand by (./2) and one where we multiply it with 1� (./2). Thereis no critical point of the phase function in the support of the amplitude. Hence thisintegral decays fast. The other integrand is bounded by
ce�tðj�jþ1Þ
which decays exponentially in t.If |x|/ 3 we use the same arguments as in the previous section, replacing
�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffix1 þ ix2
pby the roots of rp(�)¼�x/t. Here we mix real and complex notation.
We collect all estimates:
Proposition 8.1. The following estimates hold:
jItðxÞj � cminft�3=4, t�2=3g,
jrxItðxÞj � cmax½t�1, t�3=4g:
−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5
−2
−1.5
1
−0.5
0
0.5
1
1.5
2
x
y
Figure 1ðd Þ. The image under S 3 �! �rpð�Þ:
1968 Ben-Artzi, Koch, and Saut
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1968] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
If |x|/t 3 then
jItðxÞj � ct�1=2jxj�1=2:
We have
kItkL4w� ct�ð1=2Þ:
Theorem 8.2. Part 1 of Theorem 5.6 holds here. Moreover
kruðtÞkL1 � cðjtj�1þ jtj�3=4
Þkuð0ÞkL1 :
B. The polynomial ð1=6Þð�31 þ �32Þ þ �1�2. This is a second order perturbation of
ð1=6Þ�31 þ ð1=6Þ�22. The second order terms lead to a cusp, see Fig. 1(e) and 1(f ).
ItðxÞ ¼1
2�
Zeitðð1=6Þ�
31þð1=6Þ�32þ�1�2Þþix� d�:
The map � ! rpð�Þ is degenerate at the hyperbola, �1�2¼ 1. The kernel of D2�p is
tangential to the hyperbola only at (1, 1). The hyperbola is mapped to two curves inthe x space with a singularity at (�1.5,�1.5). We denote these curves by ��.
−5 −4 −3 −2 −1 0 1 2 3 4 5−5
−4
−3
−2
−1
0
1
2
3
4
5
xi
XI
Figure 1ðeÞ. The set S and the kernel of D2�p.
Dispersion Estimates for Third Order Equations 1969
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1969] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
Proposition 8.3. The following estimates hold:
kItkL1 � cminft�2=3, t�3=4g:
Moreover, for all p>4
kItkLp � ct�ð2=3Þþ2=3p:
Proof. We consider first the case |x|/t� 5. Then
It,�ðxÞ ¼1
2�
Z�ð�Þeitðð1=6Þ�
31þð1=6Þ�32þ�1�2Þþix� d�
is controlled by Theorem 7.5. Now
1
2�
Zð1� �ð�ÞÞeitðð1=6Þ�
31þð1=6Þ�32þ�1�2Þþix�d�
¼ �1
2�t
Zeitðð1=6Þ�
31þð1=6Þ�32þ�1�2Þþix�
r �ð1� �ð�ÞÞð�21, �
22Þ
12�41 þ
12�42 þ �1�
22 ¼ �2�
21 þ x1�
11 þ x2�
22
d�:
It is not hard to see that the integrand is integrable, which yields the decay t�1.The argument could be iterated to yield cNt
�N. For t� 1 we apply a similarargument, but with cutoff function
�ð�Þ ¼ �ð�t1=3Þ:
−6 −4 −2 0 2 4
−6
−4
−2
0
2
4
x
y
Figure 1ðf Þ. The image under S 3 �! �rpð�Þ.
1970 Ben-Artzi, Koch, and Saut
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1970] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
Then the argument gives
jI j � ct�2=3:
It remains to study the decay if |x| 5t. We may assume that |x1| |x2|.Let R¼ |x|/t, y¼ x/R, � ¼ �=
ffiffiffiffiR
p. Then it suffices to consider
It,�ðxÞ ¼ R
Z�ð�ÞeitR
3=2ð�31þ�
22þR�1=2�1�2þy��Þ d�
since the integral with amplitude 1� �ð�Þ can be estimated by cjxj�1=2t�1=2 by thesame arguments as above. The estimate follows now from Theorem 7.5 by a tediouscomputation. œ
Theorem 8.4. The estimate
kuðtÞkLq � ct�ð2=3Þð1=p�1=qÞkuð0ÞkLp
holds for p and q satisfying the strict inequalities of Theorem 5.6.
9. STRICHARTZ ESTIMATES AND CONSEQUENCES
The (local) Strichartz estimates follow by standard arguments (complex inter-polation, TT * arguments and the Hardy-Littlewood-Sobolev inequality) from thelocal estimate
jItðxÞj � ct�� for x 2 R2, t 2 ð0,T : ð9:1Þ
A pair (q, r) is admissible if (for Rn)
2 � r <1,1
q¼ �
1
2�1
r
� �:
If � denotes the operator
�ð f ÞðtÞ ¼
Z t
0
Sðt� sÞ f ðsÞ ds for 0 < t � T
then for any admissible pairs (q, r) and ð �qq, �rrÞ one has
sup0<t�T
k�f ðtÞkL2 þ k�f kLqð½0,T ;LrðR2ÞÞ � ck f kL �qq0 ð½0,T ;L�rr0 ðR2ÞÞ
where �qq0 and �rr0 are the Holder conjugate exponents of �qq and �rr, and
kSð�Þ�kLqð½0,T ;LrðR2ÞÞÞ � ck�kL2ðR2Þ:
For instance estimate (9.1) holds with �¼ 2/3 if p(�) is one of the followingpolynomials
1
6�31 �
1
2�1�
22,
1
6ð�31 þ �
32Þ,
1
6�31 �
1
2�1�
22 þ
1
2ð�21 þ �
22Þ,
1
6ð�31 þ �
32Þ þ �1�2: ð9:2Þ
Dispersion Estimates for Third Order Equations 1971
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1971] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
It holds with �¼ 3/4 if pð�Þ ¼ ð1=2Þð�1�22 þ �
21Þ, with �¼ 5/6 if pð�Þ ¼ �31=6þ �
22 and
with �¼ 1 if pð�Þ ¼ ð1=6Þ�31 þ �1�2. For these polynomials we obtain the admissiblepairs and the corresponding Strichartz estimates by the procedure described above.
We now give a simple application of the Strichartz estimates to the Cauchyproblem for the Shrira system (1.3). A partial result has been obtained in Ghidagliaand Saut (1993). The idea was to use the conservation laws (see Proposition 3.6in Ghidaglia and Saut (1993))
d
dt
ZR
2jAj2 dx dz ¼ 0 and
d
dt
ZR
2jAsj
2 dx dz ¼ 0
together with the Strichartz estimates.In Ghidaglia and Saut (1993) only two very special cases were considered,
corresponding to
pð�Þ ¼ �31 þ �32, and pð�Þ ¼
1
6�31 þ
1
2�22:
We obtain
Theorem 9.1. Let
L ¼!kk
2
@2
@x2þ !nn
@2
@z2þ!nk
2
@2
@x@z
� i!kkk
6
@3
@x3þ!kkn
2
@3
@x2@zþ!knn
2
@3
@x@z2þ!nnn
6
@3
@z3
" #
and assume that the symbol is equivalent to a phase p with � 2/3.Then, if A0 2 L2
ðR2Þ and @sA0 2 L2
ðR2Þ there exists a unique solution of Eq. (1.3)
with initial data A0 linear part L such that
A 2 CðRþ;L2ðR
2ÞÞ \ L�locðR
þ,L6ðR
2ÞÞ;
@A
@s2 CðRþ;L
2ðR
2ÞÞ,
where
� ¼9
2if � ¼
2
3, � ¼
18
5if � ¼
5
6, � ¼ 4 if � ¼
3
4, � ¼ 3 if � ¼ 1: ð9:3Þ
Proof. It results immediately from the proof of Theorem 3.8 in Ghidaglia and Saut(1993) by using the Strichartz estimates explained above. œ
Remark 9.2. A straight forward calculation shows that the second order part of J is
ð!kkk!knn � !2kknÞ�
21 þ ð!kkk!nnn � !kkn!knnÞ�1�2 þ ð!knn!nnn � !
2knn�
22:
1972 Ben-Artzi, Koch, and Saut
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1972] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
This is a quadratic form with determinant
D ¼ ð!kkk!knn � !2kknÞð!knn!nnn � !
2knnÞ � ð!kkk!nnn � !kkn!knnÞ:
The number D is a function of �. If D 6¼ 0 then the polynomial is equivalent toone of the four polynomial with �¼ 2/3. Numerically one sees that there are twoangles (<�/2) where D vanishes: � ¼ 0, and � � 7:405. In both cases the normalform for the polynomial is
1
2ð�21 þ �1�
22Þ:
Thus we obtain global existence for all cases of the Shrira system.
ACKNOWLEDGMENT
The work of M. Ben-Artzi was partially supported by a grant from the IsraelScience Foundation – Basic research.
REFERENCES
Arnold, V. I., Afrajmovich, V. S., Il’yashenko, Yu. S., Shil’nikov, L. P. (1999).Bifurcation Theory and Catastrophe Theory. Berlin: Springer-Verlag.
Arnol’d, V. I., Gusejn-Zade, S. M., Varchenko, A. N. (1985). Singularities ofdifferentiable maps. volume I: the classification of critical points, caustics andwave fronts. Arnol’d, V. I., ed. Transl. from the Russian by Ian Porteous, volume82 of Monographs in Mathematics. Boston-Basel-Stuttgart: Birkhaeuser.
de Bouard, A. (1993). Analytic solutions to nonelliptic nonlinear Schroedingerequations. J. Differ. Equations 104(1):196–213.
Duistermaat, J. J. (1974). Oscillatory integrals, Lagrange immersions and unfoldingof singularities. Comm. Pure. Appl. Math 27:207–281.
Dysthe, K. B. (1979). Note on a modification to the nonlinear Schrodinger equationfor applications to deep water waves. Proc. R. Soc. Lond. A 369:105–114.
Dzhuraev, T. D., Popelek, Ya. (1989). Canonical forms of third-order partialdifferential equations. Uspekhi Mat. Nauk 44(4(268)):237–238.
Dzhuraev, T. D., Popelek, Ya. (1991). Classification and reduction to canonicalform of third-order partial differential equations. Differentsial nye Uravneniya27(10):1734–1745, 1836.
Fedoryuk, M.V. (1977). The Saddle-Point Method. (Metod Perevala). Moskva:‘‘Nauka’’. 368 p. R. 1.70.
Ghidaglia, J.-M., Saut, J.-C. (1993). Nonelliptic Schroedinger equations. J.Nonlinear Sci. 3(2):169–195.
Hara, T., Mei, C. C. (1994). Wind effects on the nonlinear evolution of slowlyvarying gravity-capillary waves. J. Fluid Mech. 267:221–250.
Dispersion Estimates for Third Order Equations 1973
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1973] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903
Hogan, S. J. (1985). The fourth-order evolution equation for deep-water gravity-capillary waves. Proc. R. Soc. Lond. A 405:359–372.
Hormander, L. (1983). The analysis of linear partial differential operators I, volume256 of Grundlehren der mathematischen Wissenschaften. Springer Verlag.
Lo, E., Mei, C. C. (1985). A numerical study of water-wave modulation based onhigher-order nonlinear Schrodinger equations. J. Fluid Mech. 150:395–416.
John N. Mather (1968). Stability of C1 mappings. I. The division theorem. Ann. ofMath. (2), 87:89–104.
John N. Mather (1969). Stability of C1 mappings. II. Infinitesimal stability impliesstability. Ann. of Math. (2), 89:254–291.
Pearcey, T. (1946). The structure of an electromagnetic field in the neighborhood ofa cusp of the caustic. Phil. Mag. 37:311–317.
Shrira, V. I. (1981). On the propagation of three-dimensional packet of weaklynonlinear internal gravity waves. Int. J. Nonlinear Mechanics 16(2):129–138.
Stein, E. M. (1993). Harmonic Analysis. Volume 43 of Princeton MathematicalSeries. Princeton University Press.
Stein, E.M., Weiss, G. (1971). Introduction to Fourier Analysis in Euclidean Spaces.Volume 32 of Princeton Mathematical Series. Princeton University Press.
Zakharov, V. E. (1968). Stability of periodic waves of finite amplitude on the surfaceof a deep fluid. J. Appl. Mech. Tech. Phys. 9:86–94.
Zozulya, A. A., Diddams, S. A., Van Engen, A. G., Clement, T. S. (1999).Propagation dynamics of intense femtosecond pulses: multiple splittings,coalescense and continuum generation. Phys. Rev. Lett. 82(7):1430–1439.
Received July 2002Revised May 2003
1974 Ben-Artzi, Koch, and Saut
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
+ [29.8.2003–10:16am] [1943–1974] [Page No. 1974] F:/MDI/Pde/28(11&12)/120025491_PDE_028_011-012_R1.3d Communications in Partial Differential Equations (PDE)
120025491_PDE28_11-12_R1_082903