0 large enough, there exists a radially symmetric positive function U,,(X)E C(n) such that quenching occurs at some time T < +co for the corresponding solution of (1 .l)-(1.4). The case p = 1, N = 1 seems more difficult to handle and is not considered in this paper. Theorems 1.1 and 1.2 are proved in Sections 2 and 3, respectively. 2. PROOF OF THEOREM
1.1
LEMMA 2.1. Assume that (1.5) holds. Then there exists a 6 > 0 such that there exists a unique, positive classical solution of (1 . l)-( 1.4) in the cylinder Qs = a x (0,6).
Proof.
It is convenient to change the independent
variables as follows (2.la)
Then, exp(p jID(s)ds)
{n v(x, 0-p&
= 1
(2.lb)
in fi x IR+,
(2.2a)
and u solves v, = ~Au - (I
u-~,>-‘~~~J”u-~
au
(2.2b)
in Sz,
u(-%0) = %(X)
(2.2c)
for x E %2 and t 2 0.
-_=()
an
Let us denote by S(t) the semigroup associated to the heat equation with homogeneous Neumann conditions. By assumption, there exists u > 0 such that uO(x) 2 r in Sz. We now consider the operator (Tu)(+,t)
= S(t)u,
-
[:S(t
- s)(fQ
v”(~,s)dr)ll+p)‘pKp(.,s)d.s
(2.3a)
in the space x, =
v, E C(b x [O, S]): 5 5 9 I 211UJm
(2.3b)
1
t
endowed with the norm (/(~,11) = s~p,,~~~~JIv)(~,s)I),, where 6 > 0 is to be selected presently. is a complete metric space. Then Clearly, (X8, IIIIll)
IIw) - mollcu + 0
as t -+ 0,
339
Adiabatic shear band formation
where I is the identity operator. On the other hand, if u E X,, then vbp 5 (~/2)-~ and (So u-’ dr) -(l+p)‘p I (211uoll_)l+p. It follows that
whence
Ilmoc, 0 - u,C)ll, 5 IIW) - I)u,ll, +
2p$&+pt
5;
provided that t is small enough. This shows that T(X,) c X, if 6 is small enough. Moreover, for some constant L, , JTu(*, t) - Tti(*, t)J - s) / (~nzFp(y,~)dy)-‘l+p)‘p~-p(~,s)
5 j:S(t t I L,
w
-
- (~~6-p(y,s)dy)-o+p~‘pb”(~,s)
1ds
~)llv(*,~) - a(-,s)ll,~
0
Therefore, T is a strict contraction in X, for 6 > 0 sufficiently small. This yields the existence of a unique solution v of (2.2) for small times. By standard results, u E C”(Q&, where Qa = Sz x (0,6) and v > 0 in Q6 = fi x [0, 6). This in turn determines [by (2.1)] a unique solution of (1 . l)-( 1.4) in Qa . l The following result is a key point in the proof of theorem 1.1.
LEMMA 2.2. Let u(x, t) be a positive solution of (2.2) in some cylinder QT = a x (0, t), which is positive in the closure QT. Then, if N = 1 and p > 1, there exist constants C, 0, depending on k, p, u. and a but not on T, such that
ID
for 0 I t I T.
V(X,t)-2p dx 5 0 ect
(2.4)
Set w = v-2p. Then w satisfies
Proof. wt
=
k
AW
_
k’2p2i
‘)
. !?f
+ W
dW
an=0
zp(f
WI/2
&)-‘1+p)‘pWl/2~+3/2
in
QT,
(2.5)
0
for x E 82,
OstsT.
(2.6)
J. J. L. VEL~QUEZ
340
By averaging (2.5) over Q we obtain d WdX dt (f fl > + 2p({c
wl’zdx~o+p”p
{ w1’2p+3’2
for 0 < t < T.
(2.7)
We now claim that
where here and henceforth we shall denote by C a generic contant which possibly depends on k, p, u. and Q, but not on T. We readily see that (2.7) and (2.8) yield
s
w(.x,t)dxl(fnw(x,O)dx)ecf
61
so that the proof is concluded as soon as (2.8) is shown. To prove this inequality, we make use of the well-known Gagliardo-Nirenberg inequality
IlfIIt”(n)5 Cllf IlLglfIlaw’J(*)t
(2.9)
where 1 I CY,p I 00, r > N, and a = (l//3 - l/(r)(l/p + l/N - l/r)-‘, 0 5 a I 1. N = 1, r = 2, Q = l/p + 3 and p = 1 gives that Application of (2.9) with f = CD= We'd,
c
@l/p+3
dx
Jnc(.i.odx>(“5p)‘3p((~.~“~) +(1.
lvwd+)(I+2p~‘3p
I
I
c(
jn
Q, dx)(I+511/p(
So (D2 &T+~P)/~P
+
c(
s.
~ d,)(l+Q’)/*(
JQ Iv~,2
tir+2p)‘3p
(2.10) Since p > 1, (1 + Sp)/3p > (1 + p)/p. (~*~tiyl+“‘/3p
~
Therefore,
by Schwartz’s inequality,
(~~~~~+p)‘p(f.~2~~-l)‘3p
whence (~*@~)o+5p)‘3p(~c~2c+o+2p~‘3p
5 (f~+o+~~“(~*~~d%j.
On the other hand, using Young’s inequality: ab I euy + CEbY’, where
3p >I
y=1+2p
+Y, ’
Y-l
(2.11)
341
Adiabatic shear band formation
and E > 0 is to be selected later, we see that
Substituting (2.11) into (2.10) and taking E = k(2p + 1)/4p2, (2.8) follows.
n
LEMMA 2.3. Let V(X,t) be a solution of (2.2) in some cylinder Qr = Sz x (0, T), 0 < T I 00. Then, if v > 0 in Qr, the following inequalities are satisfied
for 0 < t < T,
(2.12a)
(2.12b)
for 0 < t < T.
(2.12b) follows at once from Jensen’s inequality, since p(t) = teP is convex whenever and t > 0. On the other hand, averaging (2.2a) over Sz yields
Proof. p > 0
-$(~*u(x,t,dx) =-(~nu(x,t)-pdX)(l+p~‘p(~*u(X,t)-~dX) whence (2.12a). LEMMA 2.4.
10
(2.13)
n
Assume that N = 1. Then the following estimates hold Ilf(*, 011L2cn)
5
for 0 I t 5 T,
Cr eelr
(2.14)
where -o f(x, t) =
(S n
(a,
O)-P dx
>
+PvP
(4% O)-P9
for any q E (1,m) and any 6 E (0, T) 5
C2 ee2’
for 6 c t < T,
(2.15)
.wQ)
where Cr and 13~(respectively, C, and 0,) depend on k, p, u. and Sz (respectively, k, p, u,,, Sz, and 6) but not on T.
q
J. J. L. VELAZQUEZ
342
Proof. Inequality (2.14) is a direct consequence of (2.4) and (2.12). To derive (2.15) we may take fi = (-R, R) for some R > 0. We extend u to the whole line as a periodic, even function, symmetric with respect to x = -R, x = +R so that the extended function 6(x, t) satisfies: fit = kc,, + K(x, t)
O
for --oo < x < co,
fi(x, t) = i&(x)
for --Q) < x < +co.
Notice that, since h”(x, t) I 0, w, Now
t) 5 IIQ.llm
for any t E (0, T).
ts
fix t E (6, T), and set t,, = maxlO, t - 1). By standard results, we have that E(x, t - s) *
6(x, t) = E(x, t - t(J * 6(x, to) +
h(+, s) ds,
to
where * denotes Therefore,
convolution
in the space variable,
gcx,%(x,t t) =
to) *
qx, to)
and E(x, s) = (47~s)-“~ exp(-x2/4s).
fax I aE -(x,t
+
-s)*~(*,s)ds.
to
Let X,(x) = 1 when x E &2and zero otherwise. Then
(2.16) We now observe that, for any g E L2(&2)and any q E [l, 00)
x, gc5 I/ ( >li 5 t1*go
Ct-3'411gllL2(n)
(2.17)
Lwa
for some C = C(q, R), whenever t E (0, l), where g is a periodic extension of g symmetric with respect to x = +R, x = -R. Let us assume (2.17) for the moment. Then
5
C i
i
Sincet-t,,r6>OandG
1(t -
C(t -
s)-~‘~eeTsds
t,)-3’4
eeFt.
substituting (2.18) into (2.16), (2.15) follows.
(2.18)
Adiabatic shear band formation
343
It remains yet to show (2.17). This can be done by means of classical (and rather cumbersome) caloric estimates; we shall sketch the corresponding argument for completeness. Clearly,
I/(
I
Q
g(-,t)*go
xfl
Lq(w
R+2nR
X
l/2
q
d-x
_R+2nR k(O/2dt >> -q/2
1
=
C-1 4J7t
1: (1x1 ([:I:% ~-3q’211glle~,n,
lx -
r12e-‘““z’2~dy)l’2~dr
= S.
Notice that, for IuI I 2, . . . Ri2nR _R+2nR
Ix
-
e-(x-t)z’2’
d<
whereas,
ym
<
s
Ix
-
e-(X-t)2’2t
dr
=
ct3’2,
s
for InI > 2, if r E (-R + 2nR, R + 2nR) we have that I<[ 2 5R. Therefore, if we deduce that Ix -
s R+2nR
-R+2nR
d<5 &$ Ix- rl 2e-(X-t)2/2t
r2 e-W/WE2/W
Furthermore,
= t(&t)
e-(8/25)(t2/‘)
Notice also that, if 0 < t I 1, and y1 = -(R there holds Ri2nR
(2e-W/WR2/2’) &Z.
-R+2nR
5 Ct for some universal constant C > 0. y2 = (R + 2nR)/dt, and InI > 2,
+ 2nR)/Jt,
+Yzct1 e-@/25)A2/‘)
i
Ri2nR
dr
-R+2nR
=
Jt
e-W25)t2d( I Jt s Yl(f)
Summing up our previous results, we conclude that
and the proof is concluded.
n
As a last step, we prove the following lemma.
e-@/W2 d<. sk
2 nR
344
J. J. L. VEL.&QUEZ
LEMMA 2.5. Let u(x, t) be as lemma 2.2. Then there exist positive on k, p, i2 and u0 but not on T, such that v(x, t) 5 MemL*
for x E G
and
M and L, depending
constants
for any T* < T.
t E ]O, T*],
(2.19)
Proof. By lemma 2.1, u(x, t) 1 q > 0 for some x E fi and t 5 6 where 6 > 0 is small enough. Assume
now that v(x,, , t,) < MeWLtn
for some sequence
(x,,
t,)
c
fi
where M > 0 and L > 0 will be selected later. By (2.15) and Morrey’s have that
x
[a,
T*],
imbedding
(2.20) theorem,
1u(x, t) - z&F,t)J 5 C, e e3*l~ - Xl* for any x, X E fi, cx E (0, 1) and t E (6, T), on k, p, Q, uO, CYand 6.
where C, and 6$ depend
It then follows from (2.20) and (2.21) that there exist [x, - D,, x,] with D, L (M/(C,) e-(L+e3)*n)1’asuch that v(x, t,) c 2MeeLtn qx, Q-P
k
2
(2.21)
Z,, = [x,, x,, + D,]
or
if x E Z,
“-zLtn (g
?R On the other hand,
intervals
we
e-CL+‘ya.
(2.22)
3
by (2.4) l/2
v(x,
i n Comparing contradiction
t,p
dx I
v(x, tp 11
&
5 p
e(C’2)*” 5 p
ec6/2e
>
(2.22) and (2.23), we readily see that for any fixed M > 0, we arrive by selecting CY> l/p and L > 0 large enough, whence the result. n
at a
End of the proof of theorem 1.1. Let us denote by T the supremum of the times for which v(x, t) exists. Assume that T < +a. By lemma 2.5, u(x, t) 2 MemL* in Sz x [0, T), and we may extend u by continuity up to t = T, so that the previous bound continues to hold. By lemma 2.5, we may then continue u(x, t) to some time interval (T, T + a), but this contradicts the definition of T, thus showing that T = co. H 3. THE PROOF OF THEOREM
1.2
In this section we shall establish the quenching result by adapting a well-known comparison technique which was introduced in [7] in the context of blow-up problems, and has been applied successfully since for extinction and quenching problems (cf. for instance [S, 91). For simplicity, we shall assume that k = 1 in (1.1). (This may be done without loss of generality by resealing the spatial variables.) Set -o+P)/P v(x, t)ydx , y(t) = (3.1) n >
345
Adiabatic shear band formation
where Sz = BR(0) in rdv. When v is a radially symmetric solution, equation (2.2a) reads v, = v, +
N-
1 r
v, - j$t)v-P.
(3.2)
First, we show that as long as V(X,t) > 0, for suitable initial values U,,(X)and R, y(t) 2 2. Consider now the auxiliary function J(r, t) = ?+i(z& - A(r)F(v)),
(3.3)
where functions A and F will be selected in the sequel. As long as v is positive 1
NLJ = Jt - J, + r
J, - (p~-~-‘y(t)
+ 2X(r)F’(v))J
N- 1 N1 A’/(r) + - r A’(r) - -L(r) r2
= rN-l
+ A(r)(puP-‘y(t)F(v)
+ y(t)v-PF’(v)
F(v) + 2A(r)/‘(r)F(v)F’(v)
+ F”(v)v;)
We take
1 .
(3.4)
forOIr
A(r)=f(l+cos(~))=Rh($
(3.5)
It is easy to check that NA”(r) + -I
1
- N+
r
A(r)
whenever 0 I r < R for some constant o > 0 depending only on N. Suppose now that F’(v) I 0. Then, since I’(r)=t(l+cos())-&sin(F)sI we have that A’(r)(-F’(v)) would obtain that
I -F’(v).
LJ L rN-‘l(r)
F(v) (
-G
forrE[O,R],
If in addition F(v) 2 0 and F”(v) 2 0 for r E [0, r], we
+ y(t)(pv-P-lF(v)
+ 2F’(v)
.
+ vmpF’(v))
>
(
>
We want to show that the right-hand side of this equation is nonnegative for a suitable choice of F, u,, . To this end we consider the equation F(v)
-$ (
+ p~-~-‘F(v)
+ 2F’(v)
+ v-~F’(v)
= 0.
>
Invariance under suitable resealing will play an important role in the analysis of (3.7).
(3.7)
J. J. L. VEL~QUEZ
346
LEMMA3.1. There exists a solution F(u) of (3.7) that can be written in the form F(u) = 6G(6%)
= 6G(r)
where 6 = (oR-2)P’@+1) with o as in (3.6).
(3.8a)
Moreover, F(v) 2 0, F’(u) 5 0 and F”(u) 2 0 provided that u I (pR2a-‘)1’(P+1)
(3.8b)
where o is as in (3.6). Proof. In view of (3.7), (3.8), we need G to satisfy 2G(r)G’(r)
+ p<-‘-PC(<)
+ T-pG’(<) - G(r) = 0.
(3.9
Notice that, if G(r) solves (3.9), then G’(P”@+~)) = 0, G’(r) < 0 We now take G(r) = G,(r)
for < < I?“(~+‘).
as the solution of (3.9) which satisfies
G(PwP+l)) =
m,
(3.10)
where m > 0 will be selected later. Clearly, a unique solution of (3.9), (3.10) exists for all r > 0. With such a choice of G = G, we, thus, have that, if F(s) is given by (3.8) F(u) L 0, F’(u) I 0
for u 5 (pRzo-‘)l’(p+l)
(3.11a)
and using (3.9) we readily check that F”(U) 2 0 which concludes the proof.
for u 5 (pR20-1)“(p+1)
(3.11b)
n
We shall presently use the following estimate G(r) 5 mpP’cpi “< -p
for any r E (O,pl’(p+lJ).
(3.12)
To derive (3.12), consider the barrier W(r) = m,lop’(p+l)<-p with ml > m in the interval (0, P l’(p+l)). By (3.10) we have G,,, < W in (cr,p “(p+l)) for some (YL 0. Assume that G,(&) = II%) for some To E (0, P l’(p+l)). By (3.9) we have GA(&,) > W’(&). This readily implies that G,(r) > W(t) for all [ E (p, pl’(p+l) ) where p is the first point where G, and W agree, but this contradicts the fact that G, < Win (a,p”@+‘)); thus, G,(r) 5 ml@‘(p+‘)&j-p for all < E (0,~ “(p+l)). Then letting ml --t m, (3.12) follows. Let now Q(r) be defined as the solution of B’(r) = A(r)F(Q(r))
for r > 0,
Q(O) = s, > 0,
(3.13a) (3.13b)
where F(s) is as in lemma 3.1, and S, is a small positive constant to be fixed later on. If we set Q(r) = R2’(l+P)v f 0
s j72/(1+PJ~(~),
(3.14)
Adiabatic shear band formation
347
we then see that I&) satisfies w’(y) = aP”l+P’h(y)G,(al’o+P’~(y))
forO
1,
(3.15a)
~(0) = R2’(1+P)&,
(3.15b)
where h and A are related through (3.5). Recalling (3.12), we now integrate (3.15) to obtain that v/(y)l+P 5 (R-2(1+J'+&,)1+p + (1 + ~)pP'(~+p)m
forOIy5
1.
(3.16)
We are now in a position to show the following lemma. LEMMA
3.2. There exists uO(x) E C(B, (0)) such that uO(x) > 0 in B, (0) satisfies (3.13a) in BR (0)
and z&x) I (pR*a- ‘)l’(’ +p).
(3.17)
Proof. In view of (3.13)-(3.16), it suffices to take u,,(x) = Q(r) to be a solution of (3.13), and then select S, and m small enough so that w(y) I (pa-‘)l’(l+P’ for y E [0, 11. n End of the proof of theorem 1.2. Let u(x, t) be a solution of (l.l)-(1.3), lemma 3.2, and let u(x, t) be given by (2.1). Since
v, I Au
where uO(x) is as in
as far as u > 0
it follows from (3.17) and the maximum principle that u(x, t) I (pR2a-‘)1’(1+P)
as far as v > 0.
(3.18)
Let y(t) be the function given in (3.1). By (3.14), we have that y(0) = (fQ @D(r)-Pdr)l”“’
= r-I(iIYI
I,,,
r~(??y
dx)(l’pi’p
= KR*
for some K = K(p, m, N) > 0. We now select R > 0 large enough so that y(O) = 2, in which case y(t) L 1 for t E [0, r] and some r > 0. Recalling (3.4), (3.8) and (3.18), we then have that LJ 2 0 in QR,, = BR (0) x (0, t). Notice that the coefficient of J in LJ is smooth in QR,7 since u > 0 there. It then turns out that J 2 0 in QR,r, whence u(r, t) L R2’(1+p)QY(y)
in QR,T,
where I@) satisfies (3.13a) with initial value W(O)= 0. We now remark that f@(y) 2 qy*'(l+P)+&
for any E > 0, and some constant K, = K,(p, N, E).
(3.19)
Inequality (3.19) is easily arrived at by standard ODE methods. Indeed, if G(r) solves (3.9), it is easy to see that H(T) = TpG(r) satisfies ((2HH’ + H’) = 2pH2 + rp”H, a dominated balance argument shows then that H(r) - -GP
log 0-l
as c + 0.
348
Therefore,
J. J. L. VELASQUEZ
G(r) - -cp(2p log <)-’ as < + 0. Recalling (3.15a), we then have that W’(Y) - -MY)(lS/(Y))-P(2P log P(Y))_’
as y + 0,
integrating this relation, we finally obtain W(y)
-
(f+)““+p’(_2C~“1+p’
as y +
0
whence the desired result. If E > 0 is small enough, it then follows from (3.19) that (V(Y)>-’
dy < +m
for N L 2 and any p > 0
BR(o)
(respectively, for 0 < p < 1 if N = 1) and, therefore, y(t) L 2 in QR,7. Iterating this argument, we obtain that y(t) L 2 for any t > 0 such that u (and u) remain positive up to such time. If Qr = BR(0) x (0, T) is such a cylinder, we should have there v, I Au - 2~-~. The fact that T cannot be arbitrarily (3.20) by comparison. n
(3.20)
large under our current hypotheses follows now from
Acknowledgements-The
author is very thankful to Drs A. A. Lacey, H. Ockendon and J. R. Ockendon for several interesting discussions during the preparation of the manuscript, and to A. Friedman for helpful suggestions in the presentation of the results. REFERENCES 1. WRIGHT T. W. & WALTER J. W., On stress collapse in adiabatic shear bands, J. Mech. phys. Solids 35(6), 701-720 (1987). 2. KAWARADA H., On solutions of the initial boundary value problem for r+ = U, + l/(1 - u), RIMS Kyoto U. 10, 729-736 (1975). 3. ACKER A. & WALTER W., The quenching problem for nonlinear parabolic equations, Lecture Notes in Mathematics, Vol. 564, pp. l-12. Springer, Berlin (1976). 4. Guo J. S., On the quenching behaviour of the solution of a semilinear parabolic equation, J. math. Analysis Applic.
151, 58-79 (1990). 5. LEVINE H. A., Advances in quenching, to appear in Proc. Conf. on Reaction-Diffusion Equations and their Equilibrium States. Birkauser, Stuttgart. 6. LEVINE H. A., Quenching, nonquenching and beyond quenching for solutions of some parabolic equations, Annali Mat. pura appl. CLV, 243-250 (1989). I. FRIEDMAN A. & MCLEOD J. B., Blow-up of positive solutions of semilinear heat equations, Indiana Univ. math. J. 34, 425-447 (1985). 8. FRIEDMAN A. & HERRERO M. A., Extinction properties of semilinear heat equations with strong absorption, J. math. Analysis Applic. 124, 530-546 (1987). 9. DENG K. & LEVINE H. A., On the blow up of U, at quenching, Proc. Am. math. Sot. 106(4), 1049-1056 (1989).