27,2 (2000), pp. 203–218
J. R E N C L A W O W I C Z (Warszawa)
GLOBAL EXISTENCE AND BLOW-UP
FOR A COMPLETELY COUPLED FUJITA TYPE SYSTEM
Abstract. The Fujita type global existence and blow-up theorems are proved for a reaction-diffusion system of m equations (m > 1) in the form
u
it= ∆u
i+ u
pi+1i, i = 1, . . . , m − 1, u
mt= ∆u
m+ u
p1m, x ∈ R
N, t > 0,
with nonnegative, bounded, continuous initial values and positive num- bers p
i. The dependence on p
iof the length of existence time (its finiteness or infinitude) is established.
1. Introduction. The main concern of this paper is to examine the following system of reaction-diffusion equations:
(1.1) u
it− ∆u
i= u
pi+1i, i = 1, . . . , m − 1, u
mt− ∆u
m= u
p1m,
for x ∈ R
N, t > 0, p
i> 0; and
u
i(0, x) = u
0i(x), i = 1, . . . , m, x ∈ R
N, where u
0iare nonnegative, continuous, bounded functions.
We consider the behaviour of classical nonnegative solutions for (1.1) as regards their maximal existence time. We describe the cases of global existence and finite blow-up time for every solution of (1.1) (in the second situation, a nontrivial solution) in terms of p
iand N .
This work generalizes results of Fujita [F1], [F2] (for the scalar Cauchy problem), Escobedo and Herrero [EH] (for the system (1.1) with m = 2) and the author [R] (for the system (1.1) with m = 3). We want to mention that the method used here is strictly connected with the special form of the
2000 Mathematics Subject Classification: 35B30, 35B50, 35K55, 35K57.
Key words and phrases : blow-up, global existence, reaction-diffusion system.
Research supported by KBN grant no. 2PO3A 038 16.
[203]
system (1.1). In the case of weakly coupled systems proofs are much more complicated (we refer to [EL] for the case of two equations).
We introduce some notations to formulate the main theorems. Let A
mbe a matrix of the form
(1.2) A
m=
0 p
10 p
20 . ..
. ..
0 p
m−1p
m0
.
We denote by α = (α
1, . . . , α
m) the unique solution of (1.3) (A
m− I)α
t= (1, . . . , 1)
t. Let
(1.4) δ = det(A
m− I) = (−1)
m+1Y
mj=1
p
j− 1 . Whenever δ 6= 0, we have
α
1= 1 + p
1+ p
1p
2+ . . . + Q
m−1 j=1p
jQ
mj=1
p
j− 1 , (1.5)
α
i= 1 + p
i+ p
ip
i+1+ . . . + Q
j6=i−1
p
jQ
mj=1
p
j− 1 , i = 2, . . . , m.
Theorem 1.1. Suppose det(A
m− I) = 0. Then every solution of (1.1) is global.
Theorem 1.2. Suppose det(A
m− I) 6= 0 and max
i=1,...,mα
i< 0. Then all the solutions of (1.1) are global.
Theorem 1.3. Suppose det(A
m− I) 6= 0 and max
i=1,...,mα
i≥ N/2.
Then every nontrivial solution of (1.1) blows up in a finite time.
The plan of this paper is the following: we prove some auxiliary lemmas in Section 2, global existence theorems is the content of the last section, whereas the global nonexistence theorem is proved in Section 3.
2. Preliminaries. Let S(t) be the semigroup operator for the heat equation, i.e.
S(t)ξ
0(x) =
\
RN
(4πt)
−N/2exp
− |x − y|
24t
ξ
0(y) dy.
A classical solution of (1.1) satisfies the following variation of constants formulas:
(2.1)
u
i(t) = S(t)u
0i+
t
\
0
S(t − s)u
pi+1i(s) ds, i = 1, . . . , m − 1,
u
m(t) = S(t)u
0m+
t
\
0
S(t − s)u
p1m(s) ds.
It follows that
u
i(τ ) = S(τ − t
i)u
i(t
i) +
τ −t\i
0
S(τ − t
i− s)u
pi+1i(s) ds
≥ S(τ − t
i)u
i(t
i), i = 1, . . . , m − 1, u
m(τ ) = S(τ − t
m)u
m(t
m) +
τ −tm
\
0
S(τ − t
m− s)u
p1m(s) ds
≥ S(τ − t
m)u
m(t
m).
We remark that if (x
i, t
i) are such that u
i(x
i, t
i) > 0, i = 1, . . . , m, then by positivity of S(t), u
i(τ ) > 0 for τ > t
i. Taking t
0= max
i=1,...,mt
iwe have u
i(x, τ ) > 0 for x ∈ R
N, τ > t
0, i = 1, . . . , m.
Lemma 2.1. Let u = (u
1, . . . , u
m) with u
0= (u
01, . . . , u
0m) 6≡ 0 be a solution of (1.1). Then we can choose τ = τ (u
01, . . . , u
0m) and some constants c > 0, a > 0 such that min
i(u
i(τ )) ≥ ce
−a|x|2.
P r o o f. We can assume that for instance u
016= 0. Then there exists R > 0 satisfying
ν
1= inf{u
01(x) : |x| < R} > 0.
By formula (2.1) we have
u
1(t) ≥ S(t)u
01≥ ν
1(4πt)
−N/2exp −|x|
24t
\|y|≤R
exp −|y|
24t
dy.
We put
u
1(t) = u
1(t + τ
0) for some τ
0> 0, a
1= 1
4τ
0, c
1= ν
1(4πτ
0)
−N/2\
|y|≤R
exp −|y|
24τ
0dy.
Hence
u
1(0) = u
1(τ
0) > c
1exp(−a
1|x|
2).
To obtain the assertion we use an inductive argument. By (2.1), u
i(τ
0) ≥
τ\0
0
S(τ
0− s)(S(s)u
0(i+1))
pids.
If p
i≥ 1 then u
i(τ ) ≥
τ\
0
(S(τ − s)S(s)u
0(i+1))
pids = τ (S(τ )u
0(i+1))
pi. Otherwise,
u
i(τ ) ≥
τ
\
0
S(τ − s)S(s)(u
0(i+1))
pids = τ S(τ )u
p0(i+1)i.
It is clear that if we assume that u
0(i+1)satisfies u
0(i+1)≥ c exp(−a|x|
2) for some constants c and a, then an analogous estimate holds true for u
i(τ ).
Replacing u
0(i+1)by u
i+1(τ
0) for some τ
0if necessary, we get the conclu- sion.
To fix ideas, we assume henceforth that max
iα
i= α
1if
m
Y
j=1
p
j> 1.
Lemma 2.2. Let (u
1(t), . . . , u
m(t)) be a bounded solution of (1.1) in some strip [0, T ) × R
N, 0 < T ≤ ∞. Assume that Q
mj=1
p
j> 1. Let n be an integer such that
(2.2) p
i≥ 1 for 1 ≤ i ≤ n, p
i< 1 for n < i ≤ m.
Then there exists a positive constant C, depending on p
ionly, such that (2.3) t
Qmj=n+1pjα1kS(t)u
Qm j=n+1pj
01
k
∞≤ C if n < m,
t
α1kS(t)u
01k
∞≤ C if n = m, t ∈ [0, T ).
P r o o f. Using (2.1) for i = 1 in (2.1) for i = m we get u
m(t) ≥
t
\
0
S(t − s)(S(s)u
01)
pmds.
Consider first n = m. Then by the Jensen inequality for p
m≥ 1 we have
(2.4) u
m(t) ≥
t
\
0
(S(t)u
01)
pmds = t(S(t)u
01)
pm.
Using (2.4) in (2.1) for i = m − 1 and the Jensen inequality for p
mp
m−1≥ 1 we get
u
m−1(t) ≥
t
\
0
S(t − s)(s(S(s)u
01)
pm)
pm−1ds
≥
t
\
0
s
pm−1(S(t)u
01)
pmpm−1ds
≥ 1
p
m−1+ 1 (S(t)u
01)
pmpm−1t
pm−1+1.
We now reason by induction. Assume that
(2.5) u
m−j(t) ≥ c
j(S(t)u
01)
πjt
rj, where
(2.6)
π
j= p
m−jπ
j−1, π
0= p
m, r
j= p
m−jr
j−1+ 1, r
0= 1, c
j= 1
r
jc
pj−1m−j, c
0= 1.
Substituting (2.5) in (2.1) for i = m − j − 1, by the Jensen inequality we obtain
u
m−(j+1)(t) ≥
t
\
0
S(t − s)[c
j(S(s)u
01)
πjs
rj]
pm−(j+1)ds
≥ c
pjm−(j+1)t
\
0
s
rjpm−j−1(S(t)u
01)
pm−j−1πjds
≥ c
pjm−j−1r
jp
m−j−1+ 1 (S(t)u
01)
pm−j−1πjt
rjpm−j−1+1, whence, by (2.6),
u
m−j−1(t) ≥ c
j+1(S(t)u
01)
πj+1t
rj+1and for j = m − 1, (2.5) gives
(2.7) u
1(t) ≥ c
m−1(S(t)u
01)
πm−1t
rm−1, where
(2.8)
π
m−1=
m
Y
j=1
p
j,
r
m−1= (. . . (((p
m−1+ 1)p
m−2+ 1)p
m−3+ 1) . . .)p
1+ 1
= α
1Y
mj=1
p
j− 1 ,
c
−1m−1= (p
m−1+ 1)
p1...pm−2(p
m−2(p
m−1+ 1) + 1)
p1...pm−3. . . . . . [(. . . ((p
m−1+ 1)p
m−2+ 1) . . .)p
1+ 1].
Iterating this scheme and setting p = π
m−1, r = r
m−1, we obtain an estimate for u
1(t) from below:
(2.9) u
1(t) ≥
m
Y
i=1
A
ki(S(t)u
01)
pkt
r(1+p+p2+...+pk−1), where
1/A
k1= r
pk−1[r(p + 1)]
pk−2[r(p
2+ p + 1)]
pk−3. . . (2.10)
. . . [r(p
k−1+ . . . + p
2+ p + 1)],
1/A
km= (p
mr + 1)
pk−1/pm[p
mr(p + 1) + 1]
pk−2/pm× [p
mr(p
2+ p + 1) + 1]
pk−3/pm. . .
. . . [p
mr(p
k−2+ . . . + p
2+ p + 1) + 1]
p/pm,
1/A
km−1= (p
m−1+ 1)
pk/(pm−1pm)(p
m−1p
mr + p
m−1+ 1)
pk−1/(pm−1pm)× [p
m−1p
mr(p + 1) + p
m−1+ 1]
pk−2/(pm−1pm). . . (2.10)
[cont.]
. . . [p
m−1p
mr(p
k−2+. . .+p
2+p+1)+p
m−1+1]
p/(pm−1pm), 1/A
km−i= [p
m−i(. . . (p
m−1+ 1) + . . . + 1)]
pk/(pm−i...pm)× [p
m−i(p
m−i+1(. . . (p
mr + 1)
+ . . . + 1) + 1) + 1]
pk−1/(pm−i...pm)× [p
m−i(. . . (p
mr(p+1)+1)+. . .+1)+1]
pk−2/(pm−i...pm). . . . . . [p
m−i(. . . (p
mr(p
k−2+ . . . + p + 1) + 1)
+ . . . + 1) + 1]
p/(pm−i...pm). We can rewrite 1/A
k1in (2.10) as
(2.11) 1/A
k1= r
(pk−1)/(p−1)k−1
Y
j=1
p
j+1− 1 p − 1
pk−j−1.
By assumption and p > 1, we have max α
i= α
1, so the following inequalities remain true for j ≥ 0:
r(1 + p + . . . + p
j) = α
1(p − 1)(1 + p + . . . + p
j) > 1, r(1 + p + . . . + p
j) > 1 + p
m−1,
r(1 + p + . . . + p
j) > 1 + p
m−2p
m−1+ p
m−2, . . . ,
r(1 + p + . . . + p
j) > 1 + p
m−i. . . p
m−1+ p
m−i. . . p
m−2+ . . . + p
m−i, for i = 1, . . . , m − 2.
Using this in (2.10) for A
k2to A
kmwe obtain
1/A
km≤ [r(p
m+ 1)]
pk−1/pm[r(p + 1)(p
m+ 1)]
pk−2/pm. . . (2.12)
. . . [r(p
k−2+ . . . + p
2+ p + 1)(p
m+ 1)]
p/pm= [r(p
m+ 1)]
pm1 pk −pp−1 k−2Y
j=1
p
j+1− 1 p − 1
pk−j−11/pm, 1/A
km−1≤ [r(p
m−1p
m+ 1)]
pk−1/(pm−1pm)× [r(p + 1)(p
m−1p
m+ 1)]
pk−2/(pm−1pm). . . . . . [r(p
k−2+ . . . + p
2+ p + 1)
× (p
m−1p
m+ 1)]
p/(pm−1pm)r
pk/(pm−1pm)= r
pk/(pm−1pm)[r(p
m−1p
m+ 1)]
pm−1pm1pk −p p−1
×
k−2Y
j=1
p
j+1− 1 p − 1
pk−j−11/(pm−1pm), . . . , (2.13) (A
km−i)
−Qij=0pm−j≤ h Y
ij=0
p
m−j+ 1
r i
pk−1h
r(p + 1)
iY
j=0
p
m−j+ 1 i
pk−2. . .
. . . h
r(p
k−2+ . . . + p + 1) Y
ij=0
p
m−j+ 1 i
pr
pk= r
pkh Y
ij=0
p
m−j+ 1 i
pk −pp−1 k−2Y
j=1
p
j+1− 1 p − 1
pk−j−1, for i = 1, . . . , m − 2.
Substituting (2.11), (2.12) and (2.13) for i = 1, . . . , m − 2 into (2.9) we infer
(2.14) (S(t)u
01)
pkt
rpk −1p−1≤ r
pk −1
p−1[1+pmp + p
pm−1pm+...+Qm−2p j=0 pm−j]
×
m−2
Y
i=0
h
1 +
i
Y
j=0
p
m−j pk −pp−1(Qij=0pm−j)−1
i u(t)
×
k−1Y
j=1
p
j+1− 1 p − 1
pk−j−11+pm1 + 1pm−1pm+...+(Qm−2
j=0 pm−j)−1
. Using (1.5) and (2.2) we observe that
1 + p p
m+ . . . + p p
m. . . p
2= 1 + p
1+ p
1p
2+ . . . + p
1. . . p
m−1= r, 1 + 1
p
m+ . . . + 1
p
m. . . p
2= 1 + r − 1 p ,
i
Y
j=0
p
m−j≤
m−1
Y
j=0
p
m−j= p for all i = 0, 1, . . . , m − 2, whence by (2.14) we obtain
(2.15) (S(t)u
01)
pkt
rpk −1p−1≤ r
pk −1p−1r(1 + p)
pk −pp−1 r−1p k−1Y
j=1
p
j+1− 1 p − 1
k−j−11+r−1pu(t).
Now, we have to estimate B
k=
k−1
Y
j=1
p
j+1− 1 p − 1
k−j−1=
k
Y
j=1
p
j− 1 p − 1
k−j. Thus
ln B
k=
k
X
j=1
p
k−j[ln(p
j− 1) − ln(p − 1)]
(2.16)
≤ p ln p p
k− 1
(p − 1)
2− ln(p − 1) p
k− 1
p − 1 = p
k− 1
p − 1 ln p
p−1pp − 1
= ln p
p−1pp − 1
pk −1p−1= ln[b
pk −1p−1], where b = p
p/(p−1)/(p − 1) > 0.
Employing this in (2.15) we finally have S(t)u
01t
rpk −1 pk(p−1)
≤ r
(pk−1)r
pk(p−1)
(1 + p)
pk −p pk(p−1)
r−1 p
× b
pk −1
pk(p−1)(1+r−1p )
ku(t)k
1/p∞ k. Using ku(t)k
∞< ∞ and letting k → ∞, we get
(2.17) t
r/(p−1)kS(t)u
01k
∞≤ c < ∞, where c = c(p, r, b) = c(p
i), i = 1, . . . , m.
Noting r = α
1(p − 1) we get the assertion.
Now we assume that n < m, i.e. p
i< 1 for n < i ≤ m. Then instead of (2.4) we get, using the Jensen inequality for p
m< 1,
u
m(t) ≥
t
\
0
S(t)u
p01mds = tS(t)u
p01mand, analogously, for p
m−j< 1 (i.e. for m − j > n) we can prove inductively, exactly as (2.5),
(2.18) u
m−j(t) ≥ c
jS(t)u
π01jt
rj, where c
j, π
j, r
jare given by (2.6).
Using (2.18) for j = m−n−1 in (2.1) for i = n, and the Jensen inequality for p
n≥ 1, we get
u
n(t) ≥ c
m−n(S(t)u
π01m−n−1)
pnt
rm−nwhere π
m−n−1=
m
Y
j=n+1
p
j. Again, arguing by induction as in the proof of formula (2.5), we conclude
u
m−j(t) ≥ c
j(S(t)u
π01m−n−1)
pn...pm−jt
rjfor j ≥ m − n, and
u
1(t) ≥ c
m−1(S(t)u
π01m−n−1)
pn...p1t
rm−1(2.19)
= c
m−1(S(t)u
π01m−n−1)
πm−1/πm−n−1t
rm−1, where π
m−n−1= Q
mi=n+1
p
iby (2.6) and π
m−1, r
m−1, c
m−1are given by (2.8).
Let us remark that π
m−1/π
m−n−1= Q
ni=1
p
i≥ 1 and since π
m−1> 1, also Q
li=1
p
i≥ π
m−1> 1 for any n ≤ l ≤ m. It follows that we apply hence- forth the Jensen inequality with an exponent greater than 1. Therefore, repeating the above considerations, we get instead of (2.9),
(2.20) u
1(t) ≥
m
Y
i=1
A
ki(S(t)u
π01m−n−1)
pk/πm−n−1t
rpk −1p−1, where the A
kiare defined by (2.10). To estimate Q
mi=1
A
ki, we proceed in the same way as before (because we have not used n there), whence
kS(t)u
π01m−n−1k
∞t
πm−n−1α1≤ c < ∞ after letting k → ∞. Thus the proof is complete.
Lemma 2.3. Let the assumptions of Lemma 2.2 be satisfied. Then we can find a constant c > 0, c = c(p
i), i = 1, . . . , m, such that for t > 0, (2.21) kS(t)u
π1m−n−1k
∞t
πm−n−1α1≤ c < ∞.
P r o o f. For τ, t ≥ 0 we have, by (2.1) for i = 1, u
1(t + τ ) = S(t + τ )u
01+
t+τ
\
0
S(t + τ − s)u
p21(s) ds
= S(t)u
1(τ ) +
t
\
0
S(t − s)(u
2(τ + s))
p1ds.
It now follows that we can replace u
01by u
1(τ ) in (2.3), whence we have the conclusion by Lemma 2.1 and setting t = τ .
3. Proof of Theorem 1.3. We prove Theorem 1.3 by contradiction.
Assuming that (2.2) holds and max α
i≥ N/2 we see by (1.5) that α
1≥ N/2 and p = Q
mj=1
p
j> 1. We shall derive a lower bound for a solution which contradicts the bounds obtained in Lemmas 2.2 and 2.3.
By Lemma 2.1 we have, for nontrivial initial data of (1.1), u
01(x) ≥ c
∗e
−a|x|2for some constants c
∗> 0, a > 0. Employing this in (2.1) for i = 1, since S(t)e
−a|x|2= (1 + 4at)
−N/2exp −a|x|
21 + 4at
,
we get
(3.1) u
1(t) ≥ S(t)u
01≥ c
∗(1 + 4at)
−N/2exp −a|x|
21 + 4at
. Using this bound in (2.1) for i = m, we have
u
m(t) ≥
t
\
0
S(t − s)(u
1(s))
pmds (3.2)
≥ c
p∗mt
\
0
S(t − s)(1 + 4as)
−N pm/2exp −ap
m|x|
21 + 4as
ds
= c
p∗mt
\
0
(1 + 4as)
−N (pm−1)/2(1 + 4as + 4ap
m(t − s))
−N/2× exp
−ap
m|x|
21 + 4as + 4ap
m(t − s)
ds.
Recall the definition of the number n in Lemma 2.2. By induction, we prove that the following estimate holds for m − j > n and some constant C:
(3.3) u
m−j(t) ≥ c(1 + 4at)
−N/2(4at − ̺
j)
βjexp −aπ
j|x|
21 + 4aπ
jt
, where ̺
j= ̺
2j−1for j > 0, ̺
0= 0, ̺ > 1,
(3.4)
π
j= p
m−jπ
j−1=
m
Y
k=m−j
p
j, π
0= p
m,
β
j= p
m−jβ
j−1+ N
2 (1 − p
m−j) + 1, β
−1= 0, and t > ̺
j/(4a).
Let n < m. Then p
m< 1, p
n+1< 1 and p
n≥ 1 by definition, and for the function f
m(s) = 1 + 4as + 4ap
m(t − s), we have f
′(s) = 4a(1 − p
m) > 0.
Thus f
m(0) ≤ f
m(s) ≤ f
m(t) for s ∈ [0, t]. Using this, we estimate (3.2) as follows:
u
m(t) ≥ c
p∗m(1 + 4at)
−N/2exp −ap
m|x|
21 + 4ap
mt
t\0
(1 + 4as)
−N (pm−1)/2ds
≥ c(1 + 4at)
−N/2(4at)
1−N (pm−1)/2exp −ap
m|x|
21 + 4ap
mt
,
so we have (3.3) for j = 0.
Substituting (3.3) in (2.1) for i = m − j − 1, we have u
m−(j+1)(t)
≥
t
\
̺j/(4a)
S(t − s)(u
m−j(s))
pm−j−1ds
≥ c
t
\
̺j/(4a)
S(t − s)(1 + 4as)
−N pm−j−1/2(4as − ̺
j)
pm−j−1βj× exp −ap
m−j−1π
j|x|
21 + 4aπ
js
ds
= c
t
\
̺j/(4a)
(1 + 4as)
−N pm−j−1/2(1 + 4aπ
js)
N/2× (1 + 4aπ
js + 4aπ
jp
m−j−1(t − s))
−N/2× exp
−ap
m−j−1π
j|x|
21 + 4aπ
js + 4aπ
jp
m−j−1(t − s)
(4as − ̺
j)
pm−j−1βjds.
Consider f
m−j−1(s) = 1+4aπ
js+4aπ
j+1(t−s). From f
m−j−1′(s) = 4aπ
j(1−
p
m−j−1) > 0, we can conclude u
m−(j+1)(t)
≥ c
pjm−j−1(1 + 4aπ
jt)
−N/2exp −aπ
j+1|x|
21 + 4aπ
j+1t
×
t
\
̺j/(4a)
(1 + 4as)
−N pm−j−1/2(4as − ̺
j)
pm−j−1βj(1 + 4aπ
js)
N/2ds.
To estimate the integral, notice that
(1 + 4aπ
jt)
N/2≥ π
jN/2(1 + 4at)
N/2since π
j< 1, (3.5)
(1 + 4aπ
jt)
−N/2≥ (1 + 4aπ
m−1t)
−N/2≥ π
−N/2m−1(1 + 4at)
−N/2, (3.6)
and
(3.7) 4at − ̺
j≥ c
0(1 + 4at) for t > ̺
2j/(4a) with c
0= ̺
j̺
j− 1
̺
2j+ 1 . Thus, integrating we get
u
m−j−1(t) ≥ c(1 + 4at)
−N/2exp −aπ
j+1|x|
21 + 4aπ
j+1t
× (4at − ̺
2j)
1+N (1−pm−j−1)/2+pm−j−1βjfor t > ̺
2j/(4a). This, in view of (3.4), gives (3.3) with j replaced by j + 1.
Analogously, for m − j ≤ n we have the following estimate for any t > ̺
j/(4a):
(3.8) u
m−j(t) ≥ c(1 + 4at)
−N/2(4at − ̺
j)
βjexp
−aπ
j|x|
21 + 4aπ
m−n−1t
, with ̺
j, β
j, π
jdefined in (3.4).
We proceed in a similar way. Assuming (3.8) for some j ≥ m − n and using (2.1) for i = m − (j + 1) we obtain
u
m−j−1(t) ≥ c
t
\
̺j/(4a)
(1 + 4as)
−N pm−j−1/2(4as − ̺
j)
pm−j−1βj× (1 + 4aπ
m−n−1s)
N/2× (1 + 4aπ
m−n−1s + 4aπ
jp
m−j−1(t − s))
−N/2× exp
−ap
m−j−1π
j|x|
21 + 4aπ
m−n−1s + 4aπ
jp
m−j−1(t − s)
ds.
Therefore, we have g
m−j−1(s) = 1 + 4a(π
m−n−1s + π
jp
m−j−1(t − s)) so g
′m−j−1(s) = 4aπ
m−n−11 − π
j+1π
m−n−1= 4aπ
m−n−11 −
n
Y
k=m−j−1
p
kand since p
k≥ 1 for k ≤ n we get g
m−j−1′(s) ≤ 0. Thus u
m−j−1(t) ≥ c(1 + 4aπ
j+1t)
−N/2exp
−aπ
j+1|x|
21 + 4aπ
m−n−1t
×
t
\
̺j/(4a)
(1 + 4as)
−N pm−j−1/2(4as − ̺
j)
pm−j−1βj(1 + 4aπ
m−n−1s)
N/2ds.
Using (3.5) with π
jreplaced by π
m−n−1< 1, (3.6) with π
j+1and integrating we obtain (3.8) with j replaced by j + 1 and t > ̺
j+1/(4a).
We need a lower bound for u
1(t). By assumption Q
mj=1
p
j> 1, whence by (2.2), p
1> 1, but we can have p
2< 1 or p
2≥ 1. In the former case we have m − j = 2 > n = 1, therefore (3.3) gives
(3.9) u
2(t) ≥ c(1 + 4at)
−N/2(4at − ̺
m−2)
βm−2exp −aπ
m−2|x|
21 + 4aπ
m−2t
. Otherwise, from m − j = 2 ≤ n and (3.8) we get
(3.10) u
2(t) ≥ c(1 + 4at)
−N/2(4at − ̺
m−2)
βm−2exp
−aπ
m−2|x|
21 + 4aπ
m−n−1t
.
Defining M = π
m−n−1(with π
m−2= M for n = 1), we conclude (3.11) u
2(t) ≥ c(1 + 4at)
−N/2(4at − ̺
m−2)
βm−2exp −aπ
m−2|x|
21 + 4aM t
. Using (2.1) for i = 1 we obtain, as p
1> 1,
(3.12) u
1(t) ≥
t
\
̺m−2/(4a)
S(t − s)(u
2(s))
p1ds
≥ c
t
\
̺m−2/(4a)
(1 + 4as)
−N p1/2(1 + 4aM s)
N/2× (1 + 4aM s + 4aπ
m−1(t − s))
−N/2× exp
−aπ
m−1|x|
21 + 4aM s + 4aπ
m−1(t − s)
(4as − ̺
m−2)
p1βm−2ds
≥ c(1 + 4aπ
m−1t)
−N/2exp −aπ
m−1|x|
21 + 4aM t
×
t
\
̺m−2/(4a)
(4as − ̺
m−2)
p1βm−2(1 + 4aM s)
N/2(1 + 4as)
−N p1/2ds
≥ c(1 + 4at)
−N/2exp −aπ
m−1|x|
21 + 4aM t
\t̺m−1/(4a)
(1 + 4as)
βm−1−1ds.
By recursive definition (3.4) and a routine inductive calculation we get β
j= 1 + p
m−j+ p
m−jp
m−j−1+ . . .
. . . + p
m−j. . . p
m−1+ N
2 (1 − p
m−j. . . p
m), therefore, using the definition of p, r, α
1(see (1.5), (2.8)), we obtain
β
m−1= 1 + p
1+ p
1p
2+ . . . + p
1. . . p
m−1+ N
2 (1 − p
1. . . p
m)
= r + N
2 (1 − p).
Since α
1≥ N/2, p > 1 and r = α
1(p − 1) we see that β
m−1≥ 0, so (3.12) gives
u
1(t) ≥ (1 + 4at)
−N/2exp
−ap|x|
21 + 4aM t
log
1 + 4at 1 + ̺
m−1for t > ̺
m−1/(4a). It follows that S(t)(u
1(t))
M≥ (1 + 4at)
−N M /2log
1 + 4at 1 + ̺
m−1 MS(t) exp −apM |x|
21 + 4aM t
= c(1 + 4at)
−N M /2log
1 + 4at 1 + ̺
m−1 M× (1 + 4aM (p + 1)t)
−N/2(1 + 4aM t)
N/2exp
−apM |x|
21 + 4aM (p + 1)t
≥ c(1 + 4at)
−N M /21 + 4aM t (1 + 4aM t)(p + 1)
N/2exp
−apM |x|
21 + 4aM (p + 1)t
×
log
1 + 4at 1 + ̺
m−1 M.
Putting x = 0 in the last estimate, we have (1 + 4at)
N M /2S(t)(u
1(t, 0))
M≥ c
log
1 + 4at 1 + ̺
m−1 M.
Therefore, for t > max{1, ̺
m−1/(4a)} we obtain, using α
1≥ N/2 and M = π
m−n−1,
(3.13) t
πm−n−1α1S(t)(u
1(t, 0))
πm−n−1≥ c
log
1 + 4at 1 + ̺
m−1πm−n−1
. It is clear that for t large enough (3.13) is incompatible with the bound (2.21). This implies that u
1(t) blows up in a finite time, which concludes the proof of the theorem in this case.
If n = m, i.e. p
i≥ 1, i = 1, . . . , m, then instead of (3.8) we obtain (3.14) u
m−j(t) ≥ c(1 + 4at)
−N/2(4at − ̺
j)
βjexp −aπ
j|x|
21 + 4at
for t > ̺
j/(4a) and with ̺
j, β
j, π
jdefined as above. Then we can repeat our considerations concerning the lower bound for u
1(t) in the case n < m, simply using (3.14) instead of (3.3) and (3.8) and, starting from (3.11), replacing everywhere M = π
m−n−1by 1. Finally, we conclude that for t > max{1, ̺
m−1/(4a)}
(3.15) t
α1S(t)u
1(t, 0) ≥ c log
1 + 4at 1 + ̺
m−1.
This contradicts (2.21) in the case n = m and the proof is complete.
4. Proofs of Theorems 1.1 and 1.2 Proof of Theorem 1.1. By assumption, Q
mj=1
p
j= 1. We are looking for a global supersolution to (1.1) of the form
(4.1)
u
1.. . u
m
=
A
1e
θ1t.. . A
me
θmt
.
We choose A
i, i = 1, . . . , m, so large that ku
0ik
L∞≤ A
i. Then (4.1) is a supersolution to (1.1) if for all t ≥ 0,
(4.2) u
it− ∆u
i≥ u
pi+1i, i = 1, . . . , m − 1, u
mt− ∆u
m≥ u
p1m.
This is equivalent to
(4.3) θ
i> A
−1iA
pi+1iexp((p
iθ
i+1− θ
i)t), i = 1, . . . , m − 1, θ
m> A
−1mA
p1mexp((p
mθ
1− θ
m)t).
If we take θ
i+1= θ
i/p
i, i = 1, . . . , m − 1, then from Q
mj=1
p
j= 1 we get θ
m= θ
1/ Q
m−1j=1
p
j= θ
1p
m, so (4.3) holds for θ
1large enough. Thus, the solution of (1.1) is global.
Proof of Theorem 1.2. In this case, from p
j≥ 0, (1.4), (1.5), we have Q
mj=1