# Full text of "Global existence for nonlinear wave equations with multiple speeds"

## See other formats

GLOBAL EXISTENCE FOR NONLINEAR WAVE EQUATIONS WITH MULTIPLE SPEEDS CHRISTOPHER D. SOGGE 1. Introduction. We shall be concerned with the Cauchy problem for quasilinear systems in three space dimensions of the form (1.1) dW - cjAu 1 = CH?d c u J d a d h u K + B I J K d a u J d b u K , I = 1, . . . , D. Here we are using the convention of summing repeated indices, and du denotes the space- time gradient, du — {dou, dm, dm, 8311), with do = dt, and dj = d Xj , j = 1, 2, 3. We shall be in the nonrelativistic case where we assume that the wave speeds c& are all positive but not necessarily equal. The main difficulty in the nonrelativistic case is that one can only use a smaller group of commuting vector fields. In particular, since one cannot use the generators of the hyperbolic rotations, due to the different wave speeds, the earlier approach of Klainerman |9 breaks down. This is because the invariant Sobolev inequality that plays a key role in |9 does not hold if one uses a smaller collection of vector fields. The conformal approach of Christodoulou || also does not seem to apply to the nonrelativistic approach. Also, unlike ||, our techniques do not use Morawetz's conformal vector field. In J|, [|, §, and H the (3+l)-dimensional case was handled. The null condition was first identified and shown to lead to global existence of small solutions. Without the null condition, small solutions remain smooth "almost globally" ||, but arbitrarily small compactly supported initial data can develop singularities in finite time Q . We shall assume that the nonlinear terms satisfy a null condition. Let us first assume, for simplicity, that the wave speeds c/, I = 1, . . . , D are distinct. In this case, the null condition only involves the self-interactions of each wave family. First we require that self-interactions among the quasilinear terms satisfy the standard null condition for the various wave-speeds: (1.2) CH J Ubtic = whenever % - £{ - £| - £ 3 2 = 0, /, J = 1, . . . , D. We shall require that the self-interacting part of the semilinear terms satisfy the standard null condition (1.3) B# J &6 = 0, whenever 4 - fi - tl - £3= 0, J, J- !,...,£>. c J The author was supported in part by the NSF. 1 2 CHRISTOPHER D. SOGGE If one allows repeated wave speeds, one must require that interactions of families with the same speeds satisfy a null condition. Specifically, if we let I p = {I : cr = cj , 1 < I < D}, then the above null condition is extended to be (1.4) C^UbCc = whenever k~ - g - $ - $ = 0, (J, e J p x Z p , 1 < I < D, and (1.5) Bi J b K Ub = 0, whenever - £ _ ^ _ ^2 = Qj ^ ^ ) e Xp x J p , 1 < / < ,D. Since we are going to use the energy integral method, we also require that the metric perturbation terms in the system (1.1) are symmetric: (1.6) CH K = Ci J J = C^i 1 , < a, 6, c < 3, 1 < 7, J, K < D. If the symmetry condition and these null conditions hold, we shall show that (1.1) has a global solution, provided that the initial data is small. To prove this result, we shall have to use estimates involving various vector fields. We shall use the generators of translations, d, the generators of Euclidean rotations fl = (fii, SI2, ^3) = x A V x , and the scaling operator (1.7) S = td t + rd r =x a d a . Here X — (xo, x%,X2, X3) — (t,x) denotes a point in R 4 . We shall denote these eight vector fields as (1.8) r = (r ,...,r 7 ) = (d,si,s), and we shall use the multi-index notation r Q = r • • r if a = (ai, . . . , am), for a sequence of indices at £ {0, . . . , 7} of length |a| = m. The D'Alembertian will be the operator (1.9) □ = Diag(D\ . . . , D D ), with a 1 = d 2 t - cjA. We can then write (1.1) as (1.10) Ou = N{du,d 2 u), where (1.11) N z (du, d 2 u) = Ci^d c u J d a d b u K + B'J K d a u J d b u K , I = 1, . . . , D. To describe the solution space we let H m (R 3 ) = {/ e L 2 (R 3 ;R D ) : ((x)V x ) a f € L 2 , \a\ < m}, denote the weighted Sobolev space with norm II/IItw) = E ll((^v,) Q /IU»(R3). |a| <m Here, (x) = (l + lxl 2 ) 1 / 2 . We can now state our global existence theorem for Minkowski space. GLOBAL EXISTENCE FOR NONLINEAR WAVE EQUATIONS WITH MULTIPLE SPEEDS 3 Theorem 1.1. Assume that the nonlinear terms (1.11) satisfy (1.6) as well as the null condition (1.4) and (1.5). Then the initial value problem for ( 1 . 10| ) with initial data du(0, ■) £ r -1 ^ 3 ), m>10, satisfying (1.12) ||Vx«(0, Ollw + IIMO, •) || w . <e, with e > sufficiently small, has a unique global solution satisfying u(t, ■ ) G 7i m (R 3 ) /or even/ i > 0. Sideris and Yu |l4[ proved the special case of this theorem where the semilinear terms vanished identically. Their approach differed from much of the previous work since they did not use any estimates arising from the fundamental solution of the d'Alembertian. A limitation of their approach, though, is that it only leads to good pointwise control of second and higher derivatives, and this explains why they were not able to handle equations with semilinear terms. Our approach is more in line with the original proof of Klainerman B . A key difference, though, is that our main estimates exploit the l/(x) decay of solutions of the wave equation, as opposed to the l/(t) decay, which is much more difficult to obtain. We are able to exploit this weaker decay because of a key pointwise estimate for solutions of the inhomogeneous wave equation that is adapted to the 1 / (x) decay and only uses the vector fields in (1.8). This estimate was proved in an earlier joint paper with Keel and Smith which proved almost global existence for quasilinear equations using only the above vector fields in the Minkowski space setting, as well as for the case of Minkowski space minus star-shaped spatial obstacles. In a future paper, we hope to show how the techniques from this paper can show that in the obstacle setting one has global existence for the above Dirichlet-wave equations when the null condition is satisfied. This would extend results in our earlier joint paper with Keel and Smith , and give a proof that works directly in Minkowski space, instead of relying on the conformal method. This paper is organized as follows. In the next section we shall recall L 2 and pointwise estimates from [|r| and Q that will play an important role in our arguments. After that, we shall see consequences of the null condition and prove some related auxiliary estimates that are adapted to the L 2 and poin twise estimates that we shall use. Then, in the final section we shall prove Theorem |l.l[ The author would like to thank Kunio Hidano for some very helpful conversations, and Mark Keel for useful comments regarding an early draft of this paper. 2. Background: Pointwise estimates and L 2 estimates. As above {£1} = {fijj}, Ojj = Xidj — Xjdi, 1 < i < j < 3, are the Euclidean R 3 rotation operators, and the scaling operator is S = tdt + x ■ V x = tdt + rd r . Then we require the following result from ||. Proposition 2.1. If w G C 4 and {d 2 — A)w — F in [0,t] x R 3 , and the Cauchy data of w are at t = 0, then (2.1) ''' ' dyds a\<3 ' 4 CHRISTOPHER D. SOGGE In H the following variant of of fl2.1|) was actually proved: dyds t\w(t,x)\ < C This estimate of course implies (|2.l|) when F(s, y) = for |y| < 1. One gets the estimate for the case where F(s,y) = for \y\ > 2 by applying the preceding case to a transla- tion such as F(s,yi,y2,y3 + 3). (The translation intro duce s the constant vector fields.) Combining the two cases by a partition of unity yields (2.1) in full generality. In addition to the pointwise estimate ( |2.l| ), we also require an L 2 L 2 estimate that is a simple consequences of the energy inequality and Huygen's principle. Proposition 2.2. Suppose that v solves the wave equation (d 2 — A)v — G in R + x K 3 . Then there is a uniform constant C so that (2.2) ||(1 + r) _1 w'|| L 2 ({(s!:c):0 < ;j < t}) + ||(1 + r)~ 2 v|| L 2 ({(Si:E):0 < s < t}) <C\\v'(0, -)h+C f \\G(s,-)\\ L2m ds. Here, and in what follows, v' denotes the space-time gradient of v, i.e., v' = (dtv, V x v). In addition to this L 2 L 2 estimate we shall also of course need the standard energy estimates for solutions of perturbed wave equations D (2.3) (dl-cjAy+Y, E l IK ^d,d k u K = F T , 1= 1,...,D, K=l 0<j,k<3 which satisfy the symmetry conditions (2.4) 1 IK ' lk = 1 IK ' kl = 1 KI ^ k , < j, k < 3, 1 < I, K < D. The associated energy form then is eo = 53r=i eI ■• where 3 D 3 D u J . k=l J=lfc=0 J=l D (2.5) e J (u,t) = (dou I ) 2 +Y,c 2 I (d k u I ) 2 + 2Y,Y,^ J,0k ^ uIdkuJ ~J2^ Jjk ^ uIdk If we assume that (2-6) E E | 7 ^ fc |<imin(l,c?), I,.J=l j,k=0 it follows that (2.7) ^ min(l, cj)\V t , x u\ 2 <e(u). If we let E(u, t) 2 = J R3 e(u, t) dx be the associated energy, then, assuming (2^) and (|2.6|), we also have the energy inequality (2.8) d t E(u,t)<C\\F(t,-)\\ 2 +CE(u,t) £ \\d a IK ^ k (t, -)Hc 0<j,k,l<3 1<I,K<D where C is an absolute constant (only depending on the wave speeds cj). GLOBAL EXISTENCE FOR NONLINEAR WAVE EQUATIONS WITH MULTIPLE SPEEDS 5 If we use the following commutator relations [0,Z] = 0, when {Z} = {dj, Oy}, and [□, S] = 2D, where, as above, S is the scaling vector field, we see that this implies (2.9) J2 d t E(T a u,t)<C l|r Q F(<, .)l| 2 + £ || [T a , t^^AM*. " )h \a\<M \a\<M \a\<M I,K,j,k + E(T a u,t) \\dn IK ' jk fr OIU, \a\<M 0<j.k,l<3 1<I,K<D 3. Null form bounds and auxiliary estimates. Here we shall prove simple bounds for the null forms. They must involve the weight (ckt — 7'} due to the fact that we are not using the generators of Lorentz rotations. Lemma 3.1. Suppose that the nonlinear form N(du,d 2 u) satisfies the null condition Then (3.1) \Cii J d c nd a d b v\ < C^-^lrul \d 2 v\ + \du\ \dTv\)+C { -^-^\du\ \d 2 v\. {t + r) Also, (3.2) \BH J d a ud b v\ < C(r)- 1 (|r«| \dv\ + \du\ \Tv\) + C^^\du\ \dv\. Proof of Lemma 3.1: Since V x = fd r - ^ AO. So, if we introduce the two operators D ± — \{dt ± cjd r ) and the null vectors Y^ = (1, ±x/cjr) associated to the wave speed cj, we have (ft, V x ) = (Y-D- + Y+D+) — (0, ——o A 0). c,jr Since we can write cjt + r cjt + r the preceding formula can be rewritten as d = Y-D- - ^L^Iy+d- + ^^Y+S - (0, A n). cjt + r cjt + r cjr z Consequently, d = Y-D~ + R, where |iZu| < c^-^rui + ^ Cjt ~ r ' \du\. t ~\~ T Therefore, we have Ci^d c ud a d b v = Cii J (Y-Y b -Y-D-u(D-fv + Rcudabv + Y-D'uRadbV + Y~ D~ uY~ D~ R b v) . Note that (Y -) 2 /c 2 j - (Yf ) 2 - (Y 2 _ ) 2 - {Yff = 0, therefore by Q the first term in the right of the last equation must vanish. Therefore, the bounds for R lead to (3.1). The proof of fl3.2|) is similar. □ Since we shall be proving estimates for scalar functions in the rest of this section, let |2 us abuse notation a bit by letting □ = d 2 — A here 6 CHRISTOPHER D. SOGGE Lemma 3.2. If h e C$°(R + x R 3 ) then (3.3) \\{t-r)Vh'(t, -)h<C £ \\r a h'(t,-)\\ 2 + C\\{t + r)Uh(t, -)\\ 2 . \a\<l Also, if < d < 1/2 is fixed then (3.4) \\h'(t, • )IU 6 (|x|^[(i-5)t,(i+(5)t]) < City 1 ( J2 ll r ^'(*» • ) h + C\\ (t + r)nh(t, ■ )\\ 2 \a\<l Proof of Lemma 3.2: Inequality ( |3.4| ) is a consequence of (3.3) since \\h'(t, ■ )\\L<S(\x\£{(l-6)t,(l+5)t]) <C\\Vh'(t, • )llL 2 (|x|^[(l-25)t,(l+2,5)t]) + C(t) _1 ||/l'(i, -)|| 2 . Inequality (3.3) is essentially in |l3j (see Lemma 2.3 and Lemma 3.1 in |_3[). The first step is to notice that one has the elementary pointwise estimate (t-r)(\dd t h(t,x)\ + \Ah(t,x)\) <C \dT a h(t,x)\ + C(t + r)\Dh(t,x)\, which leads to the I? bounds \\(t - r)(\dd t h(t, -)\ + \Ah(t, ODIU < C £ \\T a h'(t, .)\\ 2 + C\\(t + r)Oh(t, -)h. \a\<l Therefore, to finish, we need to see that (3.5) \\{t-r)Vlh(t, -)h<Cj2 H™(*, ')h + C\\(t + r)Dh(t, \a\<l To see this, we note that if we sum over l<i,j<3 and integrate by parts twice we have / {t-r) 2 y2\d i d j h\ 2 dx= [ (t - r) 2 \Ah\ 2 dx Jm. 3 „- ' Jm 3 dx The first term on the right has already been shown to be dominated by the right side of (3.3), and since an application of Schwarz's inequality shows that the second term is dominated by (/ (t-r} 2 Y,mh\ 2 dx) 1/2 \\h'(t,-)u we conclude that (3.5) must hold, which finishes the proof. □ The following result will be useful for dealing with waves interacting at different speeds. GLOBAL EXISTENCE FOR NONLINEAR WAVE EQUATIONS WITH MULTIPLE SPEEDS 7 Corollary 3.3. Fix C\, c 2 > satisfying C\ 7^ c 2 . Then if u,v £ C^°(R x (3.6) / \u"(t,x)\\v'(t,x)\(x)- 1 dx < C{t)-\ £ ||PV(*, • )|| 2 + || (t + r)(dj c?A)u(f, ■ )|| 2 ) IK^-VCt, |q|<1 + c{t)-^( J2 ll r ° u '(*, -)h + IK* + r >( 9 * ~ C ? A M*> ' )l| 2 ) x ||rV(t, .)l| 2 + ll(t + r)(a ( 2 - c 2AMt, .)|| a ) Proof: Let 26 < \c% — c 2 |. Then if we use Schwarz's inequality and (3.3) we see that we can bound \u"{t,x)\ \v'{t,x)\(x)~ dx cxt— r \>5t by the first term in the right side of ( |3.6| ). For the next step we split the remaining region where |r — c\t\ < 5t into the annuli, Aj, where {at - r) e [2 3 , 2 J+1 ), j = 0, 1, 2, . . . . Assuming that Aj (1 {x : \r - at\ < St} ^ 0, we can use Holder's inequality to find that f \u"(t,x)\\v'(t,x)\(x)- 1 dx J<cit-r)e[23,2J + 1 ) < Ct~ 1/3 2 j / 3 \\u"(t, ■ )||L 2 ((c 1 t-r)e(23,2J + 1 ))ll u '(i I • )lli 6 (<c 1 t-r)6(2J,23 + 1 ))- Since 2(5 < |ci — c 2 |, on the set where \r — c\t\ < 5t we have the lower bound \r — c 2 i| > <5£. Therefore, we can apply (3.3) and fl3.4|) t o se e that the right side is bounded by 2~ 2 ^ 3 times the second term in the right side of ( p.6|) , wh ich, after summing over j, implies that when we restrict the integration in the left side of (3J3) to the the set wh ere |r — c\t\ < 8t the resulting expression is dominated by the second term in the right of (3J3). Therefore, after summing over j, we also have control of the analog of (3.6) where the integration is over the region where \c\t — r\ < S, which completes the proof. □ To handle same-speed interactions, we shall need the following similar result. Corollary 3.4. Let u,v £ Cq The (3.7) x)~ z \d z u{t,x)\ \v(t,x)\dx <C(i)- X (E H rV (*' ^h + W + rpuit, ■)h)\\(x)- 2 v\\, \at\<l + C(t)- 4/3 (E ||PV(f, • )|| a + ||(* + r)n u (t, . )|| a ) \\v%. \a\<l CHRISTOPHER D. SOGGE (3.8) (x)~ 2 \u'(t,x)\ \v'(t,x)\dx KClogp + tm-^Yl \\r a u'(t,-)\\ 2 + \\(t + r)a u (t, oiwik^-vhj |a|<l + C(t)- 2 \\u'(t,-)\\ 2 \\v'(t,-)\\ 2 . (3.9) / { J^l\d 2 u(t,x)\\v'(t,x)\(x)- 1 dx <c(t)~\J2 H r<v (*> ')h + \\(t + r)nu(t, OII^IKx)- 1 ^*, Oils |q|<1 Proof of Corollary |3.4| : To prove (3/7) we first notice that by using Schwarz's inequality and (3.3) we get that |x|jg[t/2,3t/2] x) \d u{t, x)\ \v(t, x)\ dx <c(t)-\Y. ||rMv)ll a + ll<t+r>nu(t, ^hMxy'vh \a\<l If we use (g^) and argue as in the proof of Corollary 3.3 we can estimate the integral over \x\ G [t/2,3i/2]. We do so by noting that for j = 0,1,2,... the integral over {x : \x\ G [t/2, 3t/2}} n{x: (t - r) G [2 j , 2 j+1 )} is dominated by (t)- 2 2-i( \\T a u'(t, ■ )\\ 2 + \\(t + r)Du(t, ■ )\\ 2 )t 2 / 3 2^ 3 \\v(t, • )|| 6 . \a\<2 We conclude that the integral over the region where \x\ G [t/2,3i/2] is dominated by the other term in the right side of (3.7) after summing over j and applying Sobolev's theorem. To prove (3.8) we note that it suffices to show that L x}~ 2 \u'(t,x)\ \v'(t,x)\dx M<(t+l)/2 < cio g (2 + 1)®- 1 ( £ lir Q "U • )ll2 + \\(t + r)au(t, ■ )|| 2 ) ||(x>-V|| ai |q|<1 since clearly the integral over |x| > (t + 1)/2 is dominated by the second term in the right side of fl3.8|). However, if we use Holder's inequality and (3.4), we see that the integral over |x| < (£ + 1) /2 is dominated by log(2 + t)||u'(t, • ) Hi-darK^+ij/a) II <«>~ 1 » / Ila < Clog(2 + t)(t)~ l ( ]T ||FV(i, • )|| 2 + ||<t + r)Du(t, • )|| 2 ) ||(x>-V|| ai |o|<l which handles the remaining part of (3.8) GLOBAL EXISTENCE FOR NONLINEAR WAVE EQUATIONS WITH MULTIPLE SPEEDS 9 To prove (^) we just use Schwarz's inequality and (3.3) to see that its left side is dominated by (*>- 1 |Kt-r)5 2 «(t,.)||2|Kx)-V(t,-)l| a <^)^(E l|rMV)||2 + ||(t + r>n u (t, OlWlK^-V^OIIa. \a\<l which completes the proof. □ We shall also need the following simple result. Lemma 3.5. Let u,v G C£° (R x R 3 ). Then, (3.10) || (t + r)u(t, ■ )v(t, ■ )\\ L *m < C\\u(t, • )|| L 2 (R3) (t)\\v(t, ■ )IU«(R») +<7|iu(i,.)iii a (R») Yl m a d^(t, -Mi^m- \a\ + \/3\<2 Th e proof is simple. We first note that if we replace (t + r) by (t) in the l eft sid e of (3.1C) the resulting quantity is dominated by the first term in the right side of ( 3.1 0| ) . So to finish the proof, it would suffice to show that II "*(*, 'Hi, - )\\ L 2 m <C\\ U (t, - )\\ L 2 {m J2 H "^*. • )lli 2 (K3). \a\+\/3\<2 Since |u(t,a;)| < CJ2\p\<2 ll^x w (*> Olb) we can P rove the analog of this estimate where the norm in the left is taken over \x\ < 2. The L 2 bound over the region |x| > 2 follows from the following Sobolev estimate for K + x S 2 1/2 \v{t,rw)\<C S (/ / \n a df,v \ a \+i<2 KJ \ r -P\< lJs2 (t,pu)\ diodp x\+j<_ since, when expressed in polar coordinates, the standard volume form on R 3 is 4-7T p 2 'dpdui . 4. Proof of Theorem 1.1. We shall ass ume that the data satisfies the smallness condition (1.1 2). We then wish to show that ( 1.10 ) has a global solution if the null condition (1.4)-([Lq) holds. For simplicity, we shall assume that the wave speeds cj, I = 1, . . . , D, are distinct. A simple modification of the arguments to follow handles the general case where this assumption is removed. To proceed, we shall need to use a standard local existence theorem: Theorem 4.1. Suppose that f and g as above satisfy (1.12). Suppose also that th le symmetry condition (1.6) holds. Then there is a T > so that the initial value problem ( |1.10| ) with initial data has a C 2 solution satisfying (4.1) u e L°°([0,T];i? 10 (]R 3 )) nC 04 ([0,T] ;j ff 9 (R 3 )). The supremum of all such T is equal to the supremum of all T such that the initial value problem has a C 2 solution with d a u bounded for \a\ < 2. 10 CHRISTOPHER D. SOGGE This result is essentially Theorem 6.4.11 in gj. The latter result, though, is just for scalar wave equations; however, the same proof, which is based on energy inequalities, yields Theorem 4.1 since we are assuming that the symmetry condition (1.6) and thus one can use the energy- integral method exactly as in H. We now turn to the proof of Theorem 1.1. We let e > be as in ( 1.12 ), and assume that we already have a C 2 solution of our equation for < t < T such that for such t and small e (4.2) (1 + |*|) E \T a u'(t,x)\ <A Q e M<4 (4.3) H rV (*, -)II^(K3) + ll^r 1 r Q «'||L 2 ( {(s , ;c ) e [0, t ]x R 3 } ) a|<9 |o|<8 + E \\{x)- 2 r a u\\ L 2 {{ ^ x)emxm) < A x e{l + t) A ^ Y, H rQu '(°' Olla. |a|<8 |a|<9 Clearly both estimates are valid if T is small. We then let Aq be so large that (4.2) holds with Aq replaced by Aq/3 if u is replaced by the solution of the wave equation Ouq = with Cauchy data (/, g), and □ is as in (1.9). We shall then prove for e smaller than some number depending on A\ and Ai that i) (O) is valid with A replaced by A /2; ii) (13) is a consequence of ( J4.2| ) for suitable Ai, A 2 . By the local existence theorem it will follow that a solution exists for all t > if e is small enough. Pro of of i): Since the Cauchy data of T a u — T a UQ is 0(e 2 ), it suffices by Proposition 2.1 to prove that for small e EE// |r^rV (Siy) |-^<^. \^M^s JJo<s<T 1 + M Using the commutativity relations of the T and □, we can write T a nT a u' as a sum of terms of the form T^Uv! with \a\ < \a\ + \f3\ < 7. Therefore, it suffices to prove that (4.4) E \a\<7 \r°nu>( S ,y)\-^-<Ce 2 . o< s <T 1 + \y\ To do so we note that the I-th. component of T a Du', \a\ < 7 is a linear combination of terms of the form (4.5) E F a d(C I J c K d c u J d a d b u K )+ E r a d(B I J K d a u J d b u K ), \a\<7. a,b,c—0 a, 6—0 GLOBAL EXISTENCE FOR NONLINEAR WAVE EQUATIONS WITH MULTIPLE SPEEDS 11 Let us first handle the contribution to ( |4.4j ) of the terms in (4.5) with J / K. If we use Corollary 3.3, we find that if we fix s £ (0, T) then for J ^ K and \a\ < 7 we have (4.6) ^dici^d^d^)^ + \r a d(Bi J b K d a u J d b u K )\)-^L- i + \y\ ^Cisr^Yl l|r Q U '( S , OII2+ £ ||< S + r>(0 t 2 -4A)rV( S , .)||a) |a|<9 |a|<8 x £ iKr^rV^oih |a|<8 + C( S )- 4 / 3 (£ ||r a u'( a ,.)|| a + £ ||(.s + r)(d 2 - C 2 A)rV( s , .)||a) H<9 |a|<8 x(£ ||rV(a,.)||2+ £ ||(.s + r}(d 2 - C 2 K A)r^( S , .)l| 2 ). a|<9 M<8 Note that |(c\ 2 -c 2 A)L^( s ,y)|<c£ \rf>u'(s, y )\ £ |FV( s ,y)|, H<8. |/3|<4 |/3|<9 Therefore, if we use Lemma 3.5 and our assumptions (4.2)-(4.3) we conclude that £ \\(s + r)(d 2 - c 2 A)r a u J (s, -)\\ 2 < Ce 2 {s) 2A ^. \a\<8 Because of this, if we use (4.2)- (4.3) again, we find that the left side of ( |4.6| ) is dominated by Ce(s)- 1+2A ^ £ IKaO-Vfr ■)\\ 2 + Ce 2 {s)- i / 3 + 2A ^, \a\<8 which means that when J ^ K (4-7) £ |a|<7 0<s<T T {1^0(0^ d c u J d a d b u K )(s,y)\ + \T a d{B J J K d a u J d b u K )\)^- i + \y\ £ 2 {s) -4/3+2A 2 ^ ds (r)" i r a u / || L 2 ({(s ;1 , )e[0 ^ ]xR3}) + Ce 2 , S > 0, -l+2A 2 e £ |]<r)-V( S) -)|| 2 <C/ (e(s) J ° M<8 < C* 5 e £ ||( s )-|+2A 2£ +5/. -i r -„' a|<8 using the Schwarz inequality in the last step. If we assume that e and 8 are small enough \ — 3^2£ — (5 > then we can bound the first term in the right by Ce 2 as well by using (4.3). One first controls the L 2 norms when s lies in a dyadic subinterval of [0,i] and then sums. We therefore conclude that the terms in ( J4.5| ) with J ^ K satisfy the bounds in (4.4). So to finish we have to consider the terms with J = K, in which case we shall need to use the null condition. 12 CHRISTOPHER D. SOGGE We first estimate the contribution of the quasilinear terms satisfying the null condition. If we use Lemma |3.1| we conclude that |q|<7 (4.8) (\^ a d{C I J c J d c u J d a d b u J ) I + \T a d{B I a ( J d a u J d b u J |a|<8 ,(cjs - r) a|<8 a|<8 ' ' |a|<8 |a|<8 If we use (p.7[) we can handle the first term on the right. Indeed, dy (4.9) / J2 \d 2 r a u J (s, y )\ E |r Q « J (*,y)l (1 + M) 2 \a\<8 M<8 <^)^(E lirV(*,-)|| a + E ll^ + '->(5 t 2 -c 2 7 A)r^( s , .)l| 2 a|<9 |a|<8 x E ll<r>- a r°u( a) .)||2 M<8 + c(sr 4/3 (J2 [|rV(*,.)l|2 + E [K« + r>(a?-^A)r«« J ( fl , M<9 l«l<8 x E lirV( s , -)ll2. |a|<8 Using (3.8) we can bound the second term in the right side of (4.S): (4.10) (i + M) 2 <Clog(2 + s){ S )-'(j2 l|rV( S , -)ll2+ E \\{s + r)(d?-$A)r a u J ( S , -)|| 2 |a|<9 l«l<8 x E Urr^u'is, -)\\ 2 \a\<8 + c( s y 2 E lirV( s ,-)III. a|<8 And, similarly, if we use (|3.9| ) we can estimate the last term in (4.11) {VJS ' E \9 2 T a u J (s,y)\ E nVa'is.^-'' (s + r) \y\ \a\<8 |a|<8 <C( S )- X (E ||rV( S) -)|| 2 + E IKs + r)^ 2 -^)^^,.)^) a|<9 |a|<8 x E Urr^u'is,-)]^ \a\<8 GLOBAL EXISTENCE FOR NONLINEAR WAVE EQUATIONS WITH MULTIPLE SPEEDS 13 If we combine ( |4.8| )-(4.11), then the arguments used to prove (4.7) yield // I'lr,, ■ ,, r-<l.l.l •, ■ /' * i * i , ifm-i, ii.l.l ■ , ./■> ./', A ''-'/'^ |q|<7 0<s<T |I~a(C& J c u J o ^u')| + |r«d(i? JJ d aU W)| 7^ 7 1 + Ij/I as desired. This, along with (4/7) yields (14). Therefore, we have completed step i) of the proof. Proof of ii): We need to apply fl2.9| ) with M = 9, and We then have ( |2.6| ), assuming that £ in ( |4.2| ) is small. Since ^ (|r^| + |[r^ 7 /K > ab d Q d b H) < £ |PV| £ |PV|, q|<9 |ct|<4 |a|<9 we conclude from ([h^) that the first two terms in the right side of ( |2.9| ) are < Ceil + ty 1 E{T a u,t). \a\<9 Since (4.2) also implies that this must be the case for the last term in (2.9), we conclude that (4.2) implies that and hence that d t E(T a u,t) < Ce^l + t)- 1 E(T a u,t), \a\<9 l«|<9 E{T a u,t) < (l + t) Cs E(T a u,0). \a\<9 \a\<9 This implies that the first term in (4.3) satisfies the desired bounds. If we take Ql = E E B^daU-'d^ + £ £ CH*d c U J d ah U J,K a, b J,K a,f>,c K then we can apply Proposition 2.2 to bound the other two terms in the left side of (4.3) Indeed, they are controlled by ]T ||r<V(o,.)l|2+ f E lir Q G( s , .)l| 2 ^. M<8 M<8 We can contr ol t he last term if we use (4^) and the fact that we have shown that the first term in (|4.3|) is under control. By doing this we get E ll r " G ( s ' • )h < C W^A', ■ )IU E H rV ( s ' ' )H 2 ^ C'A s 2 ( s )- 1+A ^. M<8 M< 4 M<9 14 CHRISTOPHER D. SOGGE Therefore / V \\r a G( Sl -)IMs<c"4) £ 2 f (s}- 1+A * s ds< Jo ,.,^o Jo \a\<B C'Ap A 2 e(l+t) Since this give the desired bounds for the remaining terms in (4.3) if A2 is large enough, the proof is complete. □ References [1] D. Christodoulou: Solutions globales des equations de champ de Yang-Mills, C.R. Acad. Sci. Paris, 293, Series A, (1981), 481-506. [2] D. Christodoulou: Global solutions of nonlinear hyperbolic equations for small initial data, Comm. Pure Appl. Math. 39 (1986), 267-282. [3] L. Hormander: Lectures on nonlinear hyperbolic equations, Springer- Verlag, Berlin, 1997. [4] F. John: Blow-up for quasilinear wave equations in three space dimensions Comm. Pure Appl. Math. 34, (1981), 29-51. [5] F. John and S. Klainerman: Almost global existence to nonlinear wave equations in three dimen- sions Comm. Pure Appl. Math. 37, (1984), 443-455. [6] M. Keel, H. Smith, and C. D. Sogge: Global existence for a quasilinear wave equation outside of star-shaped domains, J. Funct. Anal., to appear. [7] M. Keel, H. Smith, and C. D. Sogge: Almost global existence for some semilinear wave equations, J. d'Analyse, to appear. [8] M. Keel, H. Smith, and C. D. Sogge: Almost global existence for quasilinear wave equations in three space dimensions, preprint. [9] S. Klainerman: The null condition and global existence to nonlinear wave equations, Nonlinear systems of partial differential equations in applied mathematics, Part 1 (Santa Fe, N.M., 1984), 293-326, Lectures in Appl. Math., 23, Amer. Math. Soc, Providence, R.I., 1986. [10] S. Klainerman: Uniform decay estimates and the Lorentz invariance of the classical wave equation, Comm. Pure Appl. Math. 38(1985), 321-332. [11] S. Klainerman: Remarks on the global Sobolev inequalities in Minkowski space Comm. Pure Appl. Math. 40(1987), 111-117. [12] S. Klainerman: Long time behavior of solutions to nonlinear wave equations, Proceedings of the International Congress of Mathematicians, Warsaw, 1982. [13] S. Klainerman and T. Sideris: On almost global existence for nonrelativistic wave equations in 3d Comm. Pure Appl. Math. 49, (1996), 307-321. [14] T. C. Sideris and S.-Y. Tu: Global existence for systems of nonlinear wave equations in 3D with multiple wave speeds, preprint. [15] C. D. Sogge: Lectures on nonlinear wave equations, International Press, Cambridge, MA, 1995. Department of Mathematics, The Johns Hopkins University, Baltimore, MD 21218