a multi-dimensional quenching problem due to a concentrated nonlinear source in
TRANSCRIPT
Nonlinear Analysis 69 (2008) 1494–1514www.elsevier.com/locate/na
A multi-dimensional quenching problem due to a concentratednonlinear source in RN
C.Y. Chan∗, P. Tragoonsirisak
Department of Mathematics, University of Louisiana at Lafayette, Lafayette, LA 70504-1010, USA
Received 9 May 2007; accepted 6 July 2007
Abstract
Let B be a N -dimensional ball x ∈ RN: |x | < R centered at the origin with a radius R, B be its closure, and ∂B be its
boundary. Also, let ν(x) denote the unit inward normal at x ∈ ∂B, and χB(x) be the characteristic function, which is 1 for x ∈ B,and 0 for x ∈ RN
\B. This article studies the following multi-dimensional semilinear parabolic first initial-boundary value problemwith a concentrated nonlinear source on ∂B:
ut − 4u = α∂χB(x)
∂νf (u) in RN
× (0, T ],
u(x, 0) = 0 for x ∈ RN , u(x, t) → 0 as |x | → ∞ for 0 < t ≤ T,
where α and T are positive numbers, f is a given function such that limu→c− f (u) = ∞ for some positive constant c, and f (u)and its derivatives f ′(u) and f ′′(u) are positive for 0 ≤ u < c. It is shown that the problem has a unique nonnegative continuoussolution before quenching occurs, and if u quenches in a finite time, then it quenches everywhere on ∂B only. It is proved that ualways quenches in a finite time for N ≤ 2. For N ≥ 3, it is shown that there exists a unique number α∗ such that u exists globallyfor α ≤ α∗ and quenches in a finite time for α > α∗. Thus, quenching does not occur in infinite time. A formula for computing α∗
is given. A computational method for finding the quenching time is devised.c© 2007 Elsevier Ltd. All rights reserved.
MSC: primary 35K60; 35K57; 35B35
Keywords: Multi-dimensional quenching; Concentrated nonlinear source; Unbounded domain; Existence; Uniqueness; Quenching; Computationalmethod; Quenching time
1. Introduction
Let H = ∂/∂t − 4, T be a positive real number, x = (x1, x2, . . . , xN ) be a point in the N -dimensional Euclidianspace RN , Ω = RN
× (0, T ], B be a N -dimensional ball x ∈ RN:∣∣x − b
∣∣ < R centered at a given point b with aradius R, ∂B be the boundary of B, ν(x) denote the unit inward normal at x ∈ ∂B, and
χB(x) =
1 for x ∈ B,0 for x ∈ RN
\B,
∗ Corresponding author. Tel.: +1 337 482 5288; fax: +1 337 482 5346.E-mail addresses: [email protected] (C.Y. Chan), [email protected] (P. Tragoonsirisak).
0362-546X/$ - see front matter c© 2007 Elsevier Ltd. All rights reserved.doi:10.1016/j.na.2007.07.001
C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514 1495
be the characteristic function. We would like to study the following multi-dimensional semilinear parabolic first initial-boundary value problem with a source on the surface of the ball:
Hu = α∂χB(x)
∂νf (u) in Ω ,
u(x, 0) = 0 for x ∈ RN , u(x, t) → 0 as |x | → ∞ for 0 < t ≤ T,
(1.1)
where α is a positive number. Without loss of generality, let b be the origin. This model is motivated by a N -dimensional ball B having a radius R and situated in RN ; on the surface ∂B of the ball, there is a nonlinearheat source of strength α f (u), where u(x, t) in Ω is the unknown temperature to be determined. We assume thatlimu→c− f (u) = ∞ for some positive constant c, and f (u) and its derivatives f ′(u) and f ′′(u) are positive for0 ≤ u < c. A solution u is said to quench if there exists an extended real number tq ∈ (0,∞] such that
sup
u(x, t) : x ∈ RN
→ c− as t → tq .
If tq < ∞, then u is said to quench in a finite time. If tq = ∞, then u quenches in infinite time. We note that whetherquenching for the heat equation with a non-concentrated source in an unbounded domain occurs in a finite time wasstudied by Dai and Gu [1], and Dai and Zeng [2].
In Section 2, we show that the nonlinear integral equation corresponding to the problem (1.1) has a uniquenonnegative continuous solution u, which is a strictly increasing function of t . We then prove that u is the uniquesolution of the problem (1.1). If tq is finite, we show that u quenches everywhere on ∂B only. In Section 3, we provethat for N ≤ 2, u always quenches in a finite time. This behavior is completely different from that for N ≥ 3. InSection 4, we show that for N ≥ 3, there exists a unique number α∗ such that u exists globally for α ≤ α∗ andquenches in a finite time for α > α∗. This rules out the possibility of quenching in infinite time. We also derive aformula for computing α∗. In Section 5, we give a computational method for determining the finite quenching time tq .
2. Existence, uniqueness, and quenching
Green’s function g(x, t; ξ, τ ) (cf. Stakgold [3, p. 198]) corresponding to the problem (1.1) is determined by thefollowing system:
Hg(x, t; ξ, τ ) = δ(x − ξ)δ(t − τ) for x, ξ ∈ RN and t, τ ∈ (−∞,∞),
g(x, t; ξ, τ ) = 0 for x, ξ ∈ RN , and t < τ,
g(x, t; ξ, τ ) → 0 as |x | → ∞ for ξ ∈ RN and t, τ ∈ (−∞,∞).
It is given by
g(x, t; ξ, τ ) =
1
[4π(t − τ)]N/2 exp
(−
|x − ξ |2
4(t − τ)
), t > τ,
0, t < τ.
To derive the integral equation from the problem (1.1), let us consider the adjoint operator (−∂/∂t − 4) of H . UsingGreen’s second identity, we obtain
u(x, t) = α
∫ t
0
∫RN
g(x, t; ξ, τ )∂χB(ξ)
∂νf (u(ξ, τ ))dξdτ.
Let R denote a positive number larger than R. Then,∫RN
g(x, t; ξ, τ )∂χB(ξ)
∂νf (u(ξ, τ ))dξ = lim
R→∞
∫|ξ |<R
g(x, t; ξ, τ ) f (u(ξ, τ )) [ν(ξ) · ∇χB(ξ)] dξ.
Since χB(ξ) = 0 for ξ ∈ RN\B, it follows from an integration by parts that
limR→∞
∫|ξ |<R
g(x, t; ξ, τ ) f (u(ξ, τ )) [ν(ξ) · ∇χB(ξ)] dξ
1496 C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514
= limR→∞
[−
∫|ξ |<R
χB(ξ)
N∑i=1
∂
∂ξi(g(x, t; ξ, τ ) f (u(ξ, τ ))νi (ξ)) dξ
]
= limR→∞
[−
∫B
∇ · (g(x, t; ξ, τ ) f (u(ξ, τ ))ν(ξ)) dξ].
By the Divergence Theorem,
limR→∞
[−
∫B
∇ · (g(x, t; ξ, τ ) f (u(ξ, τ ))ν(ξ)) dξ]
= limR→∞
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξ
=
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξ .
Thus,
u(x, t) = α
∫ t
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ. (2.1)
Let Ω denote the closure of Ω .
Lemma 2.1. Let r(x, t) ∈ C(Ω). On Ω ,
∫ t0
∫∂B g(x, t; ξ, τ )r(ξ, τ )dSξdτ is continuous.
Proof. Let η = (ξi − xi ) /(2√
t − τ). Using∫
∞
−∞e−η2
dη =√π , we have∫
∂Bg(x, t; ξ, τ )r(ξ, τ )dSξ ≤
K
2√π(t − τ)1/2
, (2.2)
where K = max(ξ,τ )∈∂B×[0,T ] r(ξ, τ ).Let ωN denote the surface area of a unit sphere. For t > τ , g(x, t; ξ, τ ) is continuous. Thus for any x and x0 in
RN , t > τ and t0 > τ , we have, for any given positive number ε, there exists some positive number δ such that|g(x, t; ξ, τ )− g(x0, t0; ξ, τ )| < ε/(2ωN RN−1t0 K ) whenever |(x, t)− (x0, t0)| < δ. Since∫ t
0
∫∂B
g(x, t; ξ, τ )r(ξ, τ )dSξdτ = limσ→0
∫ t−σ
0
∫∂B
g(x, t; ξ, τ )r(ξ, τ )dSξdτ,
we let
F(x, t) =
∫ t−σ
0
∫∂B
g(x, t; ξ, τ )r(ξ, τ )dSξdτ.
To show that F(x, t) is continuous, we show that for any given ε, there exists some positive number δ1 such that|F(x, t)− F(x0, t0)| < ε whenever |(x, t)− (x0, t0)| < δ1. To achieve this, let us choose δ1 ≤ min
δ, ε2π/(4K 2)
.
Without loss of generality, we assume that t ≥ t0. Then,∣∣∣∣∫ t−σ
0
∫∂B
g(x, t; ξ, τ )r(ξ, τ )dSξdτ −
∫ t0−σ
0
∫∂B
g(x0, t0; ξ, τ )r(ξ, τ )dSξdτ
∣∣∣∣≤
∣∣∣∣∫ t−σ
t0−σ
∫∂B
g(x, t; ξ, τ )r(ξ, τ )dSξdτ
∣∣∣∣+ ∫ t0−σ
0
∫∂B
|g(x, t; ξ, τ )− g(x0, t0; ξ, τ )| |r(ξ, τ )| dSξdτ
≤K
2√π
∫ t−σ
t0−σ
1√
t − τdτ +
K ε
2ωN RN−1t0 K
∫ t0−σ
0
∫∂B
dSξdτ
≤K
√π
(√t − t0 + σ −
√σ)+ε (t0 − σ)
2t0
≤K
√π
(√t − t0 +
√σ −
√σ)+ε (t0 − σ)
2t0
<K
√π
√δ1 +
ε
2≤ ε.
C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514 1497
The lemma then follows by letting σ → 0.
We modify the techniques used in proving Theorems 3 and 4 of Chan and Tian [4] for a blow-up problem toestablish the next two results.
Theorem 2.1. There exists some tq such that for 0 ≤ t < tq , the integral equation (2.1) has a unique continuousnonnegative solution u, and u is a strictly increasing function of t . If tq is finite, then u quenches at tq .
Proof. Let us construct a sequence un in Ω by u0(x, t) = 0, and for n = 0, 1, 2, . . . ,
Hun+1 = α∂χB(x)
∂νf (un) in Ω ,
un+1(x, 0) = 0 for x ∈ RN , un+1(x, t) → 0 as |x | → ∞ for 0 < t ≤ T .
From (2.1),
un+1(x, t) = α
∫ t
0
∫∂B
g(x, t; ξ, τ ) f (un(ξ, τ ))dSξdτ. (2.3)
Since
H(u1 − u0) = α∂χB(x)
∂νf (0) in Ω ,
(u1 − u0)(x, 0) = 0 for x ∈ RN , (u1 − u0)(x, t) → 0 as |x | → ∞ for 0 < t ≤ T,
f (0) > 0, and g(x, t; ξ, τ ) > 0 in(x, t; ξ, τ ) : x and ξ are in RN , T ≥ t > τ ≥ 0
, it follows from (2.1) that
u1(x, t) > u0(x, t) in Ω . Let us assume that for some positive integer j , 0 < u1 < u2 < u3 < · · · < u j−1 < u j inΩ . We have
H(u j+1 − u j ) = α∂χB(x)
∂ν
(f (u j )− f (u j−1)
)in Ω ,
(u j+1 − u j )(x, 0) = 0 for x ∈ RN , (u j+1 − u j )(x, t) → 0 as |x | → ∞ for 0 < t ≤ T .
Since f is a strictly increasing function and u j > u j−1, we have f (u j ) − f (u j−1) > 0. It follows from (2.1) thatu j+1 > u j . By the principle of mathematical induction, 0 < u1 < u2 < · · · < un−1 < un in Ω for any positiveinteger n.
To show that each un is an increasing function of t , let us construct a sequence wn such that for n = 0, 1,2, . . . , wn(x, t) = un(x, t + h)− un(x, t), where h is any positive number less than T . Then, w0(x, t) = 0. By (2.3),we have
w1(x, t) = α f (0)(∫ t+h
0
∫∂B
g(x, t + h; ξ, τ )dSξdτ −
∫ t
0
∫∂B
g(x, t; ξ, τ )dSξdτ).
Let σ = τ − h. Then,∫ t+h
0
∫∂B
g(x, t + h; ξ, τ )dSξdτ =
∫ h
0
∫∂B
g(x, t + h; ξ, τ )dSξdτ +
∫ t
0
∫∂B
g(x, t + h; ξ, σ + h)dSξdσ
=
∫ h
0
∫∂B
g(x, t + h; ξ, τ )dSξdτ +
∫ t
0
∫∂B
g(x, t; ξ, σ )dSξdσ
since g(x, t + h; ξ, σ + h) = g(x, t; ξ, σ ). Thus,
w1(x, t) = α f (0)∫ h
0
∫∂B
g(x, t + h; ξ, τ )dSξdτ,
which is positive for 0 < t ≤ T − h. Let us assume that for some positive integer j , w j > 0 for 0 < t ≤ T − h. Then,
w j+1(x, t) = α
(∫ t+h
0
∫∂B
g(x, t + h; ξ, τ ) f (u j (ξ, τ ))dSξdτ −
∫ t
0
∫∂B
g(x, t; ξ, τ ) f (u j (ξ, τ ))dSξdτ).
1498 C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514
Let σ = τ − h. We have∫ t+h
0
∫∂B
g(x, t + h; ξ, τ ) f (u j (ξ, τ ))dSξdτ
=
∫ h
0
∫∂B
g(x, t + h; ξ, τ ) f (u j (ξ, τ ))dSξdτ +
∫ t
0
∫∂B
g(x, t + h; ξ, σ + h) f (u j (ξ, σ + h))dSξdσ
=
∫ h
0
∫∂B
g(x, t + h; ξ, τ ) f (u j (ξ, τ ))dSξdτ +
∫ t
0
∫∂B
g(x, t; ξ, σ ) f (u j (ξ, σ + h))dSξdσ
>
∫ h
0
∫∂B
g(x, t + h; ξ, τ ) f (u j (ξ, τ ))dSξdτ +
∫ t
0
∫∂B
g(x, t; ξ, σ ) f (u j (ξ, σ ))dSξdσ.
Thus,
w j+1(x, t) > α
∫ h
0
∫∂B
g(x, t + h; ξ, τ ) f (u j (ξ, τ ))dSξdτ > 0.
By the principle of mathematical induction, wn > 0 for 0 < t ≤ T − h and all positive integers n. Thus, each un is anincreasing function of t .
By (2.2),∫∂B g(x, t; ξ, τ )dSξ is integrable with respect to τ . Thus for any given positive constant M (<c), it follows
from (2.3) and un being an increasing function of t that there exists some t1 such that un+1 ≤ M for 0 ≤ t ≤ t1 andn = 0, 1, 2, . . . . In fact, t1 satisfies
un+1(x, t1) ≤ α f (M)∫ t1
0
∫∂B
g(x, t1; ξ, τ )dSξdτ ≤ M.
Let u denote limn→∞ un . From (2.3) and the Monotone Convergence Theorem (cf. Stromberg [5, p. 266]), we have(2.1) for 0 ≤ t ≤ t1.
To prove that u is unique, let us assume that the integral equation (2.1) has two distinct solutions u and u on theinterval [0, t1]. From (2.1),
u(x, t)− u(x, t) = α
∫ t
0
∫∂B
g(x, t; ξ, τ ) ( f (u(ξ, τ ))− f (u(ξ, τ ))) dSξdτ. (2.4)
Since f ′′(u) > 0 for u ∈ [0, c), it follows from the Mean Value Theorem that | f (u)− f (u)| ≤ f ′(M) |u − u|. From(2.4),
|u(x, t)− u(x, t)| ≤ α f ′(M)∫ t
0
∫∂B
g(x, t; ξ, τ ) |u(ξ, τ )− u(ξ, τ )| dSξdτ.
By Lemma 2.1, there exists some t2 (≤t1) such that
α f ′(M) supRN ×[0,t2]
(∫ t
0
∫∂B
g(x, t; ξ, τ )dSξdτ)< 1. (2.5)
Let Θ = supRN ×[0,t2] |u − u|. Then,
Θ ≤ α f ′(M) supRN ×[0,t2]
(∫ t
0
∫∂B
g(x, t; ξ, τ )dSξdτ)
Θ .
By (2.5), this gives a contradiction. Thus, we have uniqueness of the solution for 0 ≤ t ≤ t2.If t2 < t1, then for t2 ≤ t ≤ t1,
u(x, t) =
∫RN
g(x, t; ξ, t2)u (ξ, t2) dξ + α
∫ t
t2
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ (2.6)
(cf. Chan and Tian [4]). Thus for t2 ≤ t ≤ t1,
u(x, t)− u(x, t) = α
∫ t
t2
∫∂B
g(x, t; ξ, τ ) ( f (u(ξ, τ ))− f (u(ξ, τ ))) dSξdτ.
C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514 1499
Let Θ = supRN ×[t2,min2t2,t1] |u − u|. Then,
Θ ≤ α f ′(M) supRN ×[t2,min2t2,t1]
(∫ t
t2
∫∂B
g(x, t; ξ, τ )dSξdτ)
Θ .
For t ∈ [t2,min2t2, t1],
α f ′(M) supRN ×[t2,min2t2,t1]
(∫ t
t2
∫∂B
g(x, t; ξ, τ )dSξdτ)
= α f ′(M) supRN ×[t2,min2t2,t1]
(∫ t−t2
0
∫∂B
g(x, t; ξ, σ + t2)dSξdσ)
= α f ′(M) supRN ×[t2,min2t2,t1]
(∫ t−t2
0
∫∂B
g(x, t − t2; ξ, σ )dSξdσ)< 1 (2.7)
by (2.5). This gives a contradiction. Hence, we have uniqueness of the solution for 0 ≤ t ≤ min2t2, t1. By proceedingin this way, the integral equation (2.1) has a unique solution u for 0 ≤ t ≤ t1.
To prove that u is continuous on RN× [0, t1], we note that f (un(ξ, τ )) is bounded (by f (M)). It follows from
(2.3), Lemma 2.1 and f being continuous that for n = 0, 1, 2, . . . , un+1(x, t) is continuous on RN× [0, t1]. From
(2.3),
un+1(x, t)− un(x, t) = α
∫ t
0
∫∂B
g(x, t; ξ, τ ) ( f (un(ξ, τ ))− f (un−1(ξ, τ ))) dSξdτ. (2.8)
Using the Mean Value Theorem, we have
f (un)− f (un−1) ≤ f ′(M) (un − un−1) .
Let Λn = supRN ×[0,t2] (un − un−1). From (2.8),
Λn+1 ≤ α f ′(M) supRN ×[0,t2]
(∫ t
0
∫∂B
g(x, t; ξ, τ )dSξdτ)
Λn .
By (2.5), the sequence un(x, t) converges uniformly to u(x, t) for 0 ≤ t ≤ t2, and hence, u is continuous there.If t2 < t1, then from (2.6),
un+1(x, t) =
∫RN
g(x, t; ξ, t2)u (ξ, t2) dξ + α
∫ t
t2
∫∂B
g(x, t; ξ, τ ) f (un(ξ, τ ))dSξdτ.
Let Λn = supRN ×[t2,min2t2,t1] (un − un−1). Then for t2 ≤ t ≤ t1,
Λn+1 ≤ α f ′(M) supRN ×[t2,min2t2,t1]
(∫ t
t2
∫∂B
g(x, t; ξ, τ )dSξdτ)
Λn .
It follows from (2.7) that for t ∈ [t2,min2t2, t1], u is continuous there. By proceeding in this way, the integralequation (2.1) has a unique continuous solution u for 0 ≤ t ≤ t1.
Let tq be the supremum of the intervals for which the integral equation (2.1) has a unique continuous solution u.If tq is finite, and u does not reach c− at tq , then for any positive constant between supRN u(x, tq) and c, a proofsimilar to the above shows that there exists some t3 (>tq) such that the integral equation (2.1) has a unique continuoussolution u for 0 ≤ t ≤ t3. This contradicts the definition of tq . Hence, if tq is finite, then u reaches c− somewhere inRN at tq .
Since un is an increasing function of t , we have for any positive number h such that t + h < tq ,
u(x, t + h)− u(x, t)
= α
(∫ t+h
0
∫∂B
g(x, t + h; ξ, τ ) f (u(ξ, τ ))dSξdτ −
∫ t
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ).
1500 C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514
As before,∫ t+h
0
∫∂B
g(x, t + h; ξ, τ ) f (u(ξ, τ ))dSξdτ
≥
∫ h
0
∫∂B
g(x, t + h; ξ, τ ) f (u(ξ, τ ))dSξdτ +
∫ t
0
∫∂B
g(x, t; ξ, σ ) f (u(ξ, σ ))dSξdσ.
Hence,
u(x, t + h)− u(x, t) ≥ α
∫ h
0
∫∂B
g(x, t + h; ξ, τ ) f (u(ξ, τ ))dSξdτ > 0,
which shows that u is a strictly increasing function of t .
The next result shows that the solution of the integral equation (2.1) is the solution of the problem (1.1).
Theorem 2.2. The problem (1.1) has a unique solution u for 0 ≤ t < tq .
Proof. Since∫ t
0
∫∂B g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ exists for x ∈ RN , and t in any compact subset of [0, tq), it follows
that for any t4 ∈ (0, t),∫ t
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ = limm→∞
∫ t−1/m
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ
= limm→∞
∫ t
t4
∂
∂ζ
(∫ ζ−1/m
0
∫∂B
g(x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξdτ
)dζ
+ limm→∞
∫ t4−1/m
0
∫∂B
g(x, t4; ξ, τ ) f (u(ξ, τ ))dSξdτ.
Since ∣∣gζ (x, ζ ; ξ, τ ) f (u(ξ, τ ))∣∣ =
∣∣∣∣∣ 1
[4π(ζ − τ)]N/2
(|x − ξ |2
4(ζ − τ)2
)exp
(−
|x − ξ |2
4(ζ − τ)
)
+1
[4π ]N/2
(−
N
2
)(ζ − τ)−N/2−1 exp
(−
|x − ξ |2
4(ζ − τ)
)∣∣∣∣∣ f (u(ξ, τ ))
is bounded for (ξ, τ ) ∈ ∂B × (0, ζ − 1/m), and hence integrable, it follows from the Leibnitz Rule (cf. Stromberg[5, p. 380]) that
∂
∂ζ
∫∂B
g(x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξ =
∫∂B
gζ (x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξ ,
which is continuous and bounded for each τ ∈ (0, ζ − 1/m). By the Leibnitz Rule,
∂
∂ζ
(∫ ζ−1/m
0
∫∂B
g(x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξdτ
)
=
∫∂B
g(x, ζ ; ξ, ζ − 1/m) f (u(ξ, ζ − 1/m))dSξ +
∫ ζ−1/m
0
∂
∂ζ
∫∂B
g(x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξdτ
=
∫∂B
g(x, 1/m; ξ, 0) f (u(ξ, ζ − 1/m))dSξ +
∫ ζ−1/m
0
∫∂B
gζ (x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξdτ.
Since g(x, 1/m; ξ, 0) is independent of ζ,
limm→∞
∫ t
t4
∫∂B
g
(x,
1m
; ξ, 0)
dSξdζ =
∫ t
t4lim
m→∞
∫∂B
g
(x,
1m
; ξ, 0)
dSξdζ.
C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514 1501
For x 6= ξ , g(x, 1/m; ξ, 0) converges to zero uniformly with respect to ξ on ∂B as m → ∞. Thus,
limm→∞
∫∂B
g
(x,
1m
; ξ, 0)
dSξ =
∫∂B
limm→∞
g
(x,
1m
; ξ, 0)
dSξ . (2.9)
For x = ξ , g(x, 1/m; ξ, 0) increases as m increases. By the Monotone Convergence Theorem, (2.9) holds. Thus,
limm→∞
∫ t
t4
∫∂B
g
(x,
1m
; ξ, 0)
dSξdζ =
∫ t
t4
∫∂B
limm→∞
g
(x,
1m
; ξ, 0)
dSξdζ
=
∫ t
t4
∫∂Bδ(x − ξ)dSξdζ.
Since f is bounded, it follows from the Dominated Convergence Theorem (cf. Stromberg [5, p. 268]) that
limm→∞
∫ t
t4
∫∂B
g
(x,
1m
; ξ, 0)
f
(u
(ξ, ζ −
1m
))dSξdζ =
∫ t
t4
∫∂Bδ(x − ξ) f (u(ξ, ζ ))dSξdζ
=
∫ t
t4
∫RNδ(x − ξ)
∂χB(ξ)
∂νf (u(ξ, ζ ))dξdζ
=
∫ t
t4
∂χB(x)
∂νf (u(x, ζ ))dζ.
Hence,∫ t
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ = limm→∞
∫ t
t4
∫∂B
g
(x,
1m
; ξ, 0)
f
(u
(ξ, ζ −
1m
))dSξdζ
+ limm→∞
∫ t
t4
∫ ζ−1/m
0
∫∂B
gζ (x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξdτdζ
+ limm→∞
∫ t4−1/m
0
∫∂B
g(x, t4; ξ, τ ) f (u(ξ, τ ))dSξdτ
=
∫ t
t4
∂χB(x)
∂νf (u(x, ζ ))dζ + lim
m→∞
∫ t
t4
∫ ζ−1/m
0
∫∂B
gζ (x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξdτdζ
+
∫ t4
0
∫∂B
g(x, t4; ξ, τ ) f (u(ξ, τ ))dSξdτ.
Let
hm(x, ζ ) =
∫ ζ−1/m
0
∫∂B
gζ (x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξdτ.
Without loss of generality, let m > l. We have
hm(x, ζ )− hl(x, ζ ) =
∫ ζ−1/m
ζ−1/ l
∫∂B
gζ (x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξdτ.
The integrand is integrable. It follows from the Fubini Theorem (cf. Stromberg [5, p. 352]) that
hm(x, ζ )− hl(x, ζ ) =
∫∂B
∫ ζ−1/m
ζ−1/ lgζ (x, ζ ; ξ, τ ) f (u(ξ, τ ))dτdSξ .
Since f (u(ξ, τ )) is a monotone function of τ , it follows from the Second Mean Value Theorem (cf. Stromberg[5, p. 328]) that there exists some number γ with ζ − γ ∈ [ζ − 1/ l, ζ − 1/m] such that
1502 C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514∫ ζ−1/m
ζ−1/ lgζ (x, ζ ; ξ, τ ) f (u(ξ, τ ))dτ
= f
(u
(ξ, ζ −
1l
))∫ ζ−γ
ζ−1/ lgζ (x, ζ ; ξ, τ )dτ + f
(u
(ξ, ζ −
1m
))∫ ζ−1/m
ζ−γ
gζ (x, ζ ; ξ, τ )dτ.
Hence,
hm(x, ζ )− hl(x, ζ ) =
∫∂B
f
(u
(ξ, ζ −
1l
))∫ ζ−γ
ζ−1/ lgζ (x, ζ ; ξ, τ )dτdSξ
+
∫∂B
f
(u
(ξ, ζ −
1m
))∫ ζ−1/m
ζ−γ
gζ (x, ζ ; ξ, τ )dτdSξ .
From gζ (x, ζ ; ξ, τ ) = −gτ (x, ζ ; ξ, τ ), we have
hm(x, ζ )− hl(x, ζ ) =
∫∂B
f
(u
(ξ, ζ −
1l
))(g
(x, ζ ; ξ, ζ −
1l
)− g(x, ζ ; ξ, ζ − γ )
)dSξ
+
∫∂B
f
(u
(ξ, ζ −
1m
))(g(x, ζ ; ξ, ζ − γ )− g
(x, ζ ; ξ, ζ −
1m
))dSξ
=
∫∂B
f
(u
(ξ, ζ −
1l
))(g
(x,
1l; ξ, 0
)− g (x, γ ; ξ, 0)
)dSξ
+
∫∂B
f
(u
(ξ, ζ −
1m
))(g(x, γ ; ξ, 0)− g
(x,
1m
; ξ, 0))
dSξ ,
which can be made as small as we wish by choosing l and m sufficiently large since f (u) is bounded, and g(x, ε; ξ, 0)is continuous for any ε > 0. Thus, hm is a Cauchy sequence converging uniformly with respect to ζ in any compactsubset of (0, tq). Hence,
limm→∞
∫ t
t4
∫ ζ−1/m
0
∫∂B
gζ (x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξdτdζ =
∫ t
t4
∫ ζ
0
∫∂B
gζ (x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξdτdζ.
Therefore,∫ t
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ
=
∫ t
t4
∂χB(x)
∂νf (u(x, ζ ))dζ +
∫ t
t4
∫ ζ
0
∫∂B
gζ (x, ζ ; ξ, τ ) f (u(ξ, τ ))dSξdτdζ
+
∫ t4
0
∫∂B
g(x, t4; ξ, τ ) f (u(ξ, τ ))dSξdτ.
Thus,
∂
∂t
∫ t
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ =∂χB(x)
∂νf (u(x, t))+
∫ t
0
∫∂B
gt (x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ.
Since
gxi (x, t; ξ, τ ) =1
[4π(t − τ)]N/2
[−(xi − ξi )
2(t − τ)
]exp
(−
|x − ξ |2
4(t − τ)
)
is bounded for (ξ, τ ) ∈ ∂B × (0, t − ε), it follows from the Leibniz Rule that for any x in RN and t in any compactsubset of (0, tq),
∂
∂xi
∫ t−ε
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ =
∫ t−ε
0
∫∂B
gxi (x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ.
C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514 1503
For any X i = (x1, x2, . . . , xi−1, xi , xi+1, . . . , xN ) and Zi = (x1, x2, . . . , xi−1, ςi , xi+1, . . . , xN ), we have
limε→0
∫ t−ε
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ
= limε→0
∫ xi
xi
∂
∂ςi
(∫ t−ε
0
∫∂B
g(Zi , t; ξ, τ ) f (u(ξ, τ ))dSξdτ)
dςi
+ limε→0
∫ t−ε
0
∫∂B
g(X i , t; ξ, τ ) f (u(ξ, τ ))dSξdτ
= limε→0
∫ xi
xi
∫ t−ε
0
∫∂B
gςi (Zi , t; ξ, τ ) f (u(ξ, τ ))dSξdτdςi +
∫ t
0
∫∂B
g(X i , t; ξ, τ ) f (u(ξ, τ ))dSξdτ. (2.10)
By the Fubini Theorem,∫ xi
xi
∫ t−ε
0
∫∂B
gςi (Zi , t; ξ, τ ) f (u(ξ, τ ))dSξdτdςi
=
∫ t−ε
0
∫ xi
xi
∫∂B
gςi (Zi , t; ξ, τ ) f (u(ξ, τ ))dSξdςi dτ
=
∫ t−ε
0
∫ xi
xi
∂
∂ςi
∫∂B
g(Zi , t; ξ, τ ) f (u(ξ, τ ))dSξdςi dτ
=
∫ t−ε
0
∫∂B
(g(x, t; ξ, τ )− g(X i , t; ξ, τ )
)f (u(ξ, τ ))dSξdτ.
Thus,
limε→0
∫ xi
xi
∫ t−ε
0
∫∂B
gςi (Zi , t; ξ, τ ) f (u(ξ, τ ))dSξdτdςi
=
∫ t
0
∫∂B
(g(x, t; ξ, τ )− g(X i , t; ξ, τ )
)f (u(ξ, τ ))dSξdτ,
which exists by Lemma 2.1 since f is continuous and bounded. Therefore, by using the Fubini Theorem,
limε→0
∫ xi
xi
∫ t−ε
0
∫∂B
gςi (Zi , t; ξ, τ ) f (u(ξ, τ ))dSξdτdςi
= limε→0
∫ t−ε
0
∫ xi
xi
∫∂B
gςi (Zi , t; ξ, τ ) f (u(ξ, τ ))dSξdςi dτ
=
∫ t
0
∫ xi
xi
∫∂B
gςi (Zi , t; ξ, τ ) f (u(ξ, τ ))dSξdςi dτ
=
∫ xi
xi
∫ t
0
∫∂B
gςi (Zi , t; ξ, τ ) f (u(ξ, τ ))dSξdτdςi .
From (2.10),∫ t
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ
=
∫ xi
xi
∫ t
0
∫∂B
gςi (Zi , t; ξ, τ ) f (u(ξ, τ ))dSξdτdςi +
∫ t
0
∫∂B
g(X i , t; ξ, τ ) f (u(ξ, τ ))dSξdτ,
which gives
∂
∂xi
∫ t
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ =
∫ t
0
∫∂B
gxi (x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ. (2.11)
1504 C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514
We note that
gxi xi (x, t; ξ, τ ) =1
[4π(t − τ)]N/2
(−(xi − ξi )
2(t − τ)
)2
exp
(−
|x − ξ |2
4(t − τ)
)
+1
[4π(t − τ)]N/2
(−
12(t − τ)
)exp
(−
|x − ξ |2
4(t − τ)
)is bounded for (ξ, τ ) ∈ ∂B × (0, t − ε). An argument similar to the above in proving (2.11) shows that for any x inRN and t in any compact subset of
(0, tq
),
∂
∂xi
∫ t
0
∫∂B
gxi (x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ =
∫ t
0
∫∂B
gxi xi (x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ.
From (2.11),
∂2
∂x2i
∫ t
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ =
∫ t
0
∫∂B
gxi xi (x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ.
From the integral equation (2.1), we have for x ∈ RN and t ∈ (0, tq),
Hu = α∂χB(x)
∂νf (u(x, t))+ α
∫ t
0
∫∂B
Hg(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ
= α∂χB(x)
∂νf (u(x, t))+ α lim
ε→0
∫ t−ε
0
∫∂Bδ(x − ξ)δ(t − τ) f (u(ξ, τ ))dSξdτ
= α∂χB(x)
∂νf (u(x, t)).
From (2.1), limt→0 u(x, t) = 0 for x ∈ RN , and u → 0 as |x | → ∞ since u is continuous. Thus, the nonnegativecontinuous solution of the integral equation (2.1) is a solution of the problem (1.1). Since a solution of the problem(1.1) is a solution of the integral equation (2.1), which has a unique solution before quenching occurs, u is the solutionof the problem (1.1), and the theorem is proved.
Let
χu<c =
1 if u < c,0 if u ≥ c.
What happens after quenching occurs in a finite time? Physically, the source disappears at the point where u reaches cat tq . Thus to study beyond quenching (cf. Chan and Kong [6], and Chan and Yang [7]), we extend the problem (1.1)as follows:
Hu = α∂χB(x)
∂νf (u)χu<c in RN
× (0,∞) ,
u(x, 0) = 0 for x ∈ RN , u(x, t) → 0 as |x | → ∞ for 0 < t < ∞.
(2.12)
With this, we show in our next result that at tq , quenching occurs on the surface of the ball only.
Theorem 2.3. If tq is finite, then at tq , u quenches everywhere on ∂B only.
Proof. By Theorems 2.1 and 2.2, there exists some tq such that for 0 ≤ t < tq , the problem (1.1) has a uniquenonnegative continuous solution u, which is a strictly increasing function of t . Since u(x, t) on ∂B × (0, tq) is known,let us denote it by g(x, t), and rewrite the problem (1.1) as two initial-boundary value problems:
Hu = 0 in B × (0, tq),u(x, 0) = 0 on B, u(x, t) = g(x, t) on ∂B × (0, tq).
(2.13)
Hu = 0 in (RN\B)× (0, tq),
u(x, 0) = 0 on RN\B, u (x, t) = g(x, t) on ∂B × (0, tq),
u(x, t) → 0 as |x | → ∞ for 0 < t < tq .
(2.14)
C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514 1505
To show that u cannot attain its maximum outside B for t > 0, let us prove that (x,5u) < 0 for x ∈ RN\B. For
x ∈ RN\B, it follows from (2.11) that
∂u(x, t)
∂xi= α
∫ t
0
∫∂B
1
[4π(t − τ)]N/2
exp
[−
|x − ξ |2
4(t − τ)
]f (u(ξ, τ ))
[−(xi − ξi )
2(t − τ)
]dSξdτ,
which gives
(x,∇u) = −α
∫ t
0
∫∂B
1
2N+1πN/2(t − τ)1+N/2
exp
[−
|x − ξ |2
4(t − τ)
]f (u(ξ, τ ))
(|x |
2− (x, ξ)
)dSξdτ.
By the Cauchy–Schwarz inequality,
|x |2− (x, ξ) ≥ |x |
2− |x | |ξ | = |x | (|x | − R) > 0.
Thus, (x,∇u) < 0 for x ∈ RN\B.
Let us consider the problem (2.13). It follows from the strong maximum principle (cf. Friedman [8, p. 34]) that uattains its maximum on ∂B × (0, tq). Since u is a strictly increasing function of t , we have, for each given ρ ∈
(0, tq
),
that u attains its maximum for 0 ≤ t ≤ ρ somewhere on ∂B × ρ. Suppose that there exists a smallest positive valueof t , say t0, and some y 6∈ ∂B such that u(y, t0) = minx∈∂B u(x, t0). We claim that for x ∈ ∂B, u(x, t0) = u(y, t0). Ifthis is not true, then there exists some x ∈ ∂B such that u(x, t0) > minx∈∂B u(x, t0). Since u is continuous, there existssome point (y, t0) in a neighborhood of (x, t0) such that y 6∈ ∂B and u(y, t0) > minx∈∂B u(x, t0). This contradictsthe definition of t0. Thus, u attains its maximum at (y, t0) for 0 ≤ t ≤ t0. By the strong maximum principle and thecontinuity of u, we have u ≡ u(y, t0) on B × [0, t0]. This contradicts u(x, 0) = 0. Thus for any t > 0,
u(x, t) > u(y, t) for any x ∈ ∂B, and any y 6∈ ∂B. (2.15)
We claim that for each t > 0, u attains the same value for x ∈ ∂B. If this is not true, then for some t > 0, there existssome x ∈ ∂B such that u(x, t) > minx∈∂B u(x, t). By continuity, there exists some point (y, t) in a neighborhood of(x, t) such that y 6∈ ∂B and u(y, t) > minx∈∂B u(x, t). This contradicts (2.15). Hence for any t > 0,
u(x, t) = M(t) for x ∈ ∂B,M(t) > u(y, t) for any y 6∈ ∂B, (2.16)
where M(t) denotes supx∈RN u(x, t).From (2.16), u quenches everywhere on ∂B as t → tq . Since tq is finite, it follows from (2.12) that the problem
(2.13) holds also at tq with g(x, tq) = c for x ∈ ∂B. Thus, if u quenches somewhere at (z, tq) where z ∈ B, then bythe strong maximum principle and the continuity of u, we have u ≡ c on B × [0, tq ]. This gives a contradiction sinceu(x, 0) = 0. Hence, if u quenches, then it quenches on ∂B only. For the problem (2.14), if u quenches somewhere at(Z , tq) for some Z ∈ RN
\B, then let B be a ball centered at the origin such that Z ∈ B. Since u quenches at (Z , tq),it follows that u ≡ c on (B\B)× [0, tq ]. This contradicts u(x, 0) = 0, and hence, u quenches only on ∂B at tq . Thisproves the theorem.
3. N ≤ 2
Let
I (x, t) =
∫∂B
g(x, t; ξ, 0)dSξ .
Lemma 3.1. For t ≥ 1 and for any b ∈ B,
(4π)−N/2e−R2ωN RN−1t−N/2
≤ I (b, t) ≤ (4π)−N/2ωN RN−1t−N/2.
Proof. For t ≥ 1,
e−R2≤ exp
(−
|b − ξ |2
4t
)≤ 1,
1506 C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514
where ξ ∈ ∂B. Thus,
e−R2(4π t)−N/2
∫∂B
dSξ ≤ I (b, t) ≤ (4π t)−N/2∫∂B
dSξ .
The surface area ωN of a unit sphere is given by NπN/2/Γ (N/2 + 1) (cf. Evans [9, p. 615]). Using Γ (N/2 + 1)= (N/2)Γ (N/2), we have ωN = 2πN/2/Γ (N/2). Since
∫∂B dSξ = ωN RN−1, the lemma then follows.
Let us denote the (positive) constants (4π)−N/2e−R2ωN RN−1 and (4π)−N/2ωN RN−1 by c1 and c2 respectively.
Then from Lemma 3.1,
c1t−N/2≤ I (b, t) ≤ c2t−N/2.
Theorem 3.1. For N ≤ 2, the solution u of the problem (1.1) (for any α) always quenches everywhere in a finite timeon ∂B only.
Proof. Since u ≥ 0 and f ′(u) > 0, we have f (u) ≥ f (0) > 0. For any t ∈ (1,∞), it follows from (2.1) that
u(x, t) ≥ α f (0)∫ t
0
∫∂B
g(x, t; ξ, τ )dSξdτ
> α f (0)∫ t−1
0
∫∂B
g(x, t; ξ, τ )dSξdτ
= α f (0)∫ t−1
0
∫∂B
g(x, t − τ ; ξ, 0)dSξdτ
= α f (0)∫ t−1
0I (x, t − τ)dτ
= α f (0)∫ t
1I (x, θ)dθ.
For each t (>0), it follows from (2.16) that u attains its (absolute) maximum M(t) at any b ∈ ∂B. Thus for anyt ∈ (1,∞),
u(b, t) > α f (0)∫ t
1I (b, θ)dθ.
By Lemma 3.1,∫ t
1I (b, θ)dθ ≥ c1
∫ t
1θ−N/2dθ =
2c1
(t1/2
− 1)
if N = 1,
c1 ln t if N = 2.
Thus, u(b, t) quenches in a finite time. The theorem then follows from Theorem 2.3.
4. N ≥ 3
In this section, we show that the quenching behavior for N ≥ 3 is completely different from that for N ≤ 2.
Theorem 4.1. (i) For N ≥ 3, the solution u of the problem (1.1) exists globally for α sufficiently small.(ii) For N ≥ 3, the solution u of the problem (1.1) quenches in a finite time for α sufficiently large.
Proof. (i) Since f ′(u) > 0, we have f (u) ≤ f (c/2) for u(x, t) ≤ c/2. Thus for u(x, t) ≤ c/2, it follows from (2.1)that
u(x, t) ≤ α f( c
2
) ∫ t
0
∫∂B
g(x, t; ξ, τ )dSξdτ.
For 0 < t ≤ 1, it follows from (2.2) that
u(x, t) ≤ α f( c
2
) ∫ t
0
dτ
2√π(t − τ)1/2
=α
√π
f( c
2
)t1/2
≤α
√π
f( c
2
).
C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514 1507
For t > 1, it follows from (2.2) and (2.16) that for any b ∈ ∂B,
u(x, t) ≤ α f( c
2
)(∫ t−1
0I (b, t − τ)dτ +
∫ t
t−1I (b, t − τ)dτ
)
≤ α f( c
2
)(∫ t
1I (b, θ)dθ +
∫ t
t−1
dτ
2√π(t − τ)1/2
)< α f
( c
2
)(∫ ∞
1I (b, θ)dθ +
1√π
)since I (b, θ) is positive. By Lemma 3.1,∫
∞
1I (b, θ)dθ ≤ c2
∫∞
1θ−N/2dθ =
c2
N/2 − 1< ∞ for N ≥ 3. (4.1)
Thus for α sufficiently small, u(x, t) ≤ c/2 for (x, t) ∈ RN× (0,∞).
(ii) For any b ∈ ∂B and t ∈ (1,∞),
u(b, t) ≥ α f (0)∫ t
0
∫∂B
g(b, t; ξ, τ )dSξdτ > α f (0)∫ t−1
0I (b, t − τ)dτ = α f (0)
∫ t
1I (b, θ)dθ.
By (4.1), (0<)∫ t
1 I (b, θ)dθ < ∞. We can choose α sufficiently large such that u(b, t) > c. This proves thetheorem.
The fundamental solution (cf. Evans [9, pp. 22 and 615]) of the Laplace equation for N ≥ 3 is given by
G(x) =Γ( N
2 + 1)
N (N − 2)πN/2
1
|x |N−2 . (4.2)
Let us modify the proof of Theorem 3 of Chan and Kaper [10] for a one-dimensional problem with a non-concentratedsource in a bounded domain to obtain the following result.
Theorem 4.2. If u(x, t) ≤ C for some constant C ∈ (0, c), then u(x, t) converges from below to a solutionU (x) = limt→∞ u(x, t) of the nonlinear integral equation,
U (x) = α
∫∂B
G(x − ξ) f (U (ξ))dSξ . (4.3)
Proof. Let
F(x, t) =
∫RN
G(x − ξ)u(ξ, t)dξ for (x, t) ∈ RN× (0,∞). (4.4)
For any point x , we can always choose a number R such that |x | < R. Using Green’s second identity, we have∫RN
G(x − ξ)4 u(ξ, t)dξ = limR→∞
∫|ξ |<R
G(x − ξ)4 u(ξ, t)dξ
= limR→∞
∫|ξ |<R
u(ξ, t)4 G(x − ξ)dξ
= limR→∞
∫|ξ |<R
−u(ξ, t)δ(x − ξ)dξ
= −u(x, t).
Since u is the solution of the problem (1.1), F(x, t) may be regarded as a distribution. Thus,
1508 C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514
Ft (x, t) =
∫RN
G(x − ξ)ut (ξ, t)dξ
=
∫RN
G(x − ξ)
(4u(ξ, t)+ α
∂χB(ξ)
∂νf (u(ξ, t))
)dξ
= −u(x, t)+ α
∫RN
G(x − ξ)∂χB(ξ)
∂νf (u(ξ, t))dξ.
From Theorem 2.1, u is a strictly increasing function of t for x ∈ RN . Since f is increasing, the integrand of the secondterm on the right-hand side is monotone increasing with respect to t . It follows from the Monotone ConvergenceTheorem and the continuity of f that
limt→∞
Ft (x, t) = − limt→∞
u(x, t)+ α
∫RN
G(x − ξ)∂χB(ξ)
∂νf ( lim
t→∞u(ξ, t))dξ
= − limt→∞
u(x, t)+ α
∫∂B
G(x − ξ) f ( limt→∞
u(ξ, t))dSξ , (4.5)
which exists since u ≤ C . We note from (4.4) that F is a strictly increasing function of t for x ∈ RN . Iflimt→∞ Ft (x, t) > 0 at some point x , then F(x, t) would increase without bound as t tends to infinity. Since Fis bounded, limt→∞ Ft (x, t) = 0. The theorem follows from (4.5).
The next result shows that there exists a critical value for α.
Theorem 4.3. For N ≥ 3, there exists a unique α∗ such that u exists globally for α < α∗, and u quenches in a finitetime for α > α∗.
Proof. Let us consider the sequence un given by u0(x, t) = 0, and for n = 0, 1, 2, . . . ,
un+1(x, t) = α
∫ t
0
∫∂B
g(x, t; ξ, τ ) f (un(ξ, τ ))dSξdτ.
Since un is an increasing sequence as n increases, it follows from the Monotone Convergence Theorem that
u(x, t) = α
∫ t
0
∫∂B
g(x, t; ξ, τ ) f (u(ξ, τ ))dSξdτ,
where limn→∞ un(x, t) = u(x, t). To show that the larger the α, the larger the solution u, let α > β, and consider thesequence vn given by v0(x, t) = 0, and for n = 0, 1, 2, . . . ,
vn+1(x, t) = β
∫ t
0
∫∂B
g(x, t; ξ, τ ) f (vn(ξ, τ ))dSξdτ.
Similarly,
v(x, t) = limn→∞
vn(x, t) = β
∫ t
0
∫∂B
g(x, t; ξ, τ ) f (v(ξ, τ ))dSξdτ.
Since un > vn for n = 1, 2, 3, . . . , we have u ≥ v. Hence, the solution u is a nondecreasing function of α. It followsfrom Theorem 4.1 that there exists a unique α∗ such that u exists globally for α < α∗, and u quenches in a finite timefor α > α∗.
We note that the critical value α∗ is determined as the supremum of all positive values α for which a solution Uof (4.3) exists. The proof of the next result, showing that the solution u exists globally when α = α∗, is analogous tothat of Theorem 7 of Chan and Jiang [11] for a degenerate one-dimensional problem in a bounded domain.
Theorem 4.4. For N ≥ 3, the solution u does not quench in infinite time.
Proof. From (2.16), U (x) = limt→∞ u(x, t) attains its maximum at b ∈ ∂B. From (4.3),
U (b) = α
∫∂B
G(b − ξ) f (U (b))dSξ .
C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514 1509
Thus,
α∗=
(1∫
∂B G(b − ξ)dSξ
)max
0≤U (b)≤c
(U (b)
f (U (b))
). (4.6)
Let us consider the function ψ(s) = s/ f (s). Since ψ(s) > 0 for 0 < s < c, and ψ(0) = 0 = lims→c− ψ(s), a directcomputation shows that ψ(s) attains its maximum when ψ(s) = 1/ f ′(s), where s ∈ (0, c) by the Rolle Theorem.Thus, max0≤U (b)≤c (U (b)/ f (U (b))) occurs when
U (b)
f (U (b))=
1f ′(U (b))
with 0 < U (b) < c. This implies that U (x) exists when α = α∗. Hence for α ≤ α∗, u exists globally and is uniformlybounded away from c. Since u quenches in a finite time for α > α∗, u does not quench in infinite time.
Our next result gives a formula for α∗.
Theorem 4.5. For N ≥ 3,
α∗=
(N − 2)π (N−3)/2
RΓ(
N−12
) N−3∏i=1
(∫ π0 sini ϕdϕ
) max0≤U (b)≤c
(U (b)
f (U (b))
), (4.7)
where for N = 3,∏N−3
i=1
(∫ π0 sini ϕdϕ
)= 1.
Proof. Using cos2 µ+ sin2 µ = 1, we have
(1 − cosϕ1)2+ (sinϕ1 cosϕ2)
2+ (sinϕ1 sinϕ2 cosϕ3)
2+ · · ·
+ (sinϕ1 · · · sinϕN−2 cos θ)2 + (sinϕ1 · · · sinϕN−2 sin θ)2 = 2 (1 − cosϕ1) . (4.8)
Since∫∂B |b − ξ |2−N dSξ is independent of where b is on ∂B, we may let b = (R, 0, 0, 0, . . . , 0) ∈ RN . Using the
spherical coordinates (cf. Stromberg [5, pp. 369-370]), we have
∫∂B
dSξ|b − ξ |N−2 =
RN−1
RN−2
∫ π
−π
∫ π
0. . .
(N−2 times)
∫ π
0
N−2∏i=1
sinN−i−1 ϕi dϕ1dϕ2 . . . dϕN−2dθ
[2 (1 − cosϕ1)](N−2)/2
=2πR
2(N−2)/2
∫ π
0. . .
(N−2 times)
∫ π
0
N−2∏i=1
sinN−i−1 ϕi dϕ1dϕ2 . . . dϕN−2
(1 − cosϕ1)(N−2)/2
.
Since 1 − cosϕ1 = 2 sin2(ϕ1/2), and sinϕ1 = 2 sin(ϕ1/2) cos(ϕ1/2), we have∫∂B
dSξ|b − ξ |N−2 = 2πR
(∫ π
0cosN−2
(ϕ1
2
)dϕ1
) N−2∏i=2
∫ π
0sinN−i−1 ϕi dϕi ,
where for N = 3,∏N−2
i=2
∫ π0 sinN−i−1 ϕi dϕi = 1. By using Mathematica version 5.2, we get
∫ π
0cosN−2
(ϕ1
2
)dϕ1 =
√πΓ
(N−1
2
)Γ( N
2
) .
Therefore,∫∂B
dSξ|b − ξ |N−2 = 2π3/2 R
Γ(
N−12
)Γ( N
2
) N−3∏i=1
(∫ π
0sini ϕN−i−1dϕN−i−1
),
1510 C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514
where for N = 3,∏N−3
i=1
(∫ π0 sini ϕN−i−1dϕN−i−1
)= 1. From (4.2), and Γ ((N/2)+ 1) = (N/2)Γ (N/2),∫
∂BG(b − ξ)dSξ =
Γ( N
2 + 1)
N (N − 2)πN/2
∫∂B
dSξ|b − ξ |N−2
=
RΓ(
N−12
)(N − 2)π (N−3)/2
N−3∏i=1
(∫ π
0sini ϕdϕ
),
where for N = 3,∏N−3
i=1
(∫ π0 sini ϕdϕ
)= 1.
From (4.6), we have (4.7).
For illustration, let f (u) = 1/(1−u). A direct computation shows that U (b) (1 − U (b)) attains its maximum whenU (b) = 1/2. Hence,
α∗=
(N − 2)π (N−3)/2
4RΓ(
N−12
) N−3∏i=1
(∫ π0 sini ϕdϕ
) .
5. Quenching time tq
To determine the quenching time tq , it follows from (2.16) that it is sufficient to consider
M(t) = α
∫ t
0
∫∂B
g(b, t; ξ, τ ) f (M(τ ))dSξdτ,
where b is any point on ∂B. Since u is an increasing function of t , we have f (M(τ )) < f (M(t)) for t > τ . To find alower bound tl of tq , we may use
M(t) < α f (M(t))∫ t
0
∫∂B
g(b, t; ξ, τ )dSξdτ =α f (M(t))
(4π)N/2
∫ t
0
1
(t − τ)N/2
∫∂B
exp
[−
|b − ξ |2
4(t − τ)
]dSξdτ.
(5.1)
Without loss of generality, let b = (R, 0) for N = 2 and b = (R, 0, 0, 0, . . . , 0) for N ≥ 3.For N = 2, we have∫
∂Bexp
[−
|b − ξ |2
4(t − τ)
]dSξ = R
∫ π
−π
exp[−(R − R cos θ)2 + (R sin θ)2
4(t − τ)
]dθ
= R∫ π
−π
exp[−
R2 (1 − cos θ)2(t − τ)
]dθ
= R
exp
[−
R2
2(t − τ)
] ∫ π
−π
exp[
R2 cos θ2(t − τ)
]dθ.
From (5.1),
M(t) <αR f (M(t))
4π
∫ t
0
1(t − τ)
exp
[−
R2
2(t − τ)
] ∫ π
−π
exp[
R2 cos θ2(t − τ)
]dθdτ.
Let η = R/√
2(t − τ). We have dη = R−2η3dτ , and
M(t) <αR f (M(t))
2π
∫∞
R/√
2t
exp(−η2
)η
∫ π
−π
exp(η2 cos θ
)dθdη,
which gives∫∞
R/√
2t
exp(−η2
)η
∫ π
−π
exp(η2 cos θ
)dθdη >
2παR
M(t)
f (M(t)).
C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514 1511
Thus, a lower bound tl for the quenching time is given by solving∫∞
R/√
2tl
exp(−η2
)η
∫ π
−π
exp(η2 cos θ
)dθdη =
2παR
max0≤M(t)≤c
(M(t)
f (M(t))
). (5.2)
For N ≥ 3 and α > α∗, it follows from (4.8) that∫∂B
exp
[−
|b − ξ |2
4(t − τ)
]dSξ
= RN−1∫ π
−π
∫ π
0. . .
(N−2 times)
∫ π
0
exp
[−
R2 (1 − cosϕ1)
2(t − τ)
] N−2∏i=1
sinN−i−1 ϕi dϕ1dϕ2 . . . dϕN−2dθ
= 2πRN−1
exp[−
R2
2(t − τ)
](N−3∏i=1
∫ π
0sini ϕdϕ
)∫ π
0
exp
[R2 cosϕ1
2(t − τ)
]sinN−2 ϕ1dϕ1.
From (5.1),
M(t) < 2πRN−1α f (M(t))
(4π)N/2
(N−3∏i=1
∫ π
0sini ϕdϕ
)
×
∫ t
0
1
(t − τ)N/2
exp
[−
R2
2(t − τ)
] ∫ π
0
exp
[R2 cosϕ1
2(t − τ)
]sinN−2 ϕ1dϕ1dτ.
Let η = R/√
2(t − τ). We have dη = R−2η3dτ , and
M(t) <αR f (M(t))
(2π)(N−2)/2
(N−3∏i=1
∫ π
0sini ϕdϕ
)
×
∫∞
R/√
2tηN−3
[exp
(−η2
)] ∫ π
0
[exp
(η2 cosϕ1
)]sinN−2 ϕ1dϕ1dη,
which gives∫∞
R/√
2tηN−3
[exp
(−η2
)] ∫ π
0
[exp
(η2 cosϕ1
)]sinN−2 ϕ1dϕ1dη >
(2π)(N−2)/2
αRN−3∏i=1
∫ π0 sini ϕdϕ
M(t)
f (M(t)).
Thus, a lower bound tl for the quenching time is given by solving∫∞
R/√
2tlηN−3
[exp
(−η2
)] ∫ π
0
[exp
(η2 cosϕ1
)]sinN−2 ϕ1dϕ1dη
=(2π)(N−2)/2
αRN−3∏i=1
∫ π0 sini ϕdϕ
max0≤M(t)≤c
(M(t)
f (M(t))
). (5.3)
From (2.3) and (2.16), we have for N = 2, M (0)(t) = 0, M (k) (0) = 0, and for k = 1, 2, 3, . . . ,
M (k)(t) = α
∫ t
0
∫∂B
g(x, t; ξ, τ ) f (M (k−1)(τ ))dSξdτ
=αR
2π
∫∞
R/√
2t
exp(−η2
)η
f
(M (k−1)
(t −
R2
2η2
))∫ π
−π
[exp
(η2 cos θ
)]dθdη. (5.4)
1512 C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514
Similarly for N ≥ 3, M (0)(t) = 0, M (k) (0) = 0, and for k = 1, 2, 3, . . . ,
M (k)(t) =αR
(2π)(N−2)/2
(N−3∏i=1
∫ π
0sini ϕdϕ
)
×
∫∞
R/√
2tηN−3
[exp
(−η2
)]f
(M (k−1)
(t −
R2
2η2
))∫ π
0
[exp
(η2 cosϕ1
)]sinN−2 ϕ1dϕ1dη.
(5.5)
To compute the quenching time tq , we use Mathematica version 5.2. We obtain its lower bound tl by solving (5.2)for N = 2 or (5.3) for N ≥ 3. We then let t = 2tl in (5.4) for N = 2 or (5.5) for N ≥ 3, and check the convergenceof the sequence
M (k)(t)
. If it diverges, then we use 2tl to be an upper bound of tq and tl to be a lower bound. If
M (k)(t)
converges, then we check the convergence for t = 3tl . If it diverges, then we use 3tl to be an upper boundand 2tl to be a lower bound. By this iterative process, we obtain an upper bound j tl for some positive integer j forwhich
M (k)(t)
diverges, and a lower bound ( j − 1)tl for which
M (k)(t)
converges. We then use the bisection
procedure to compute tq .We give below the steps to compute tq :
Step 1. We input the values of R, c and α, and the function f (u). Let j = 2.Step 2. We find tl from (5.2) for N = 2, or from (5.3) for N ≥ 3.Step 3. Let t = j ∗ tl .Step 4. We replace t by t in (5.4) for N = 2, or in (5.5) for N ≥ 3, and let M (0)(τ ) = 0.Step 5. Let h = t/m, where m denotes the number of subdivisions of equal length h. For k = 1, 2, 3, . . . , it follows
from (5.4) that for N = 2,
M (k)(rh) =αR
2π
∫∞
R/√
2rh
exp(−η2
)η
f
(M (k−1)
(rh −
R2
2η2
))∫ π
−π
[exp
(η2 cos θ
)]dθdη, (5.6)
and from (5.5) that for N ≥ 3,
M (k)(rh) =αR
(2π)(N−2)/2
(N−3∏i=1
∫ π
0sini ϕdϕ
)∫∞
R/√
2rhηN−3
[exp
(−η2
)]× f
(M (k−1)
(rh −
R2
2η2
))∫ π
0
[exp
(η2 cosϕ1
)]sinN−2 ϕ1dϕ1dη, (5.7)
where r = 1, 2, 3, . . . ,m with M (k)(0) = 0.Step 6. At the kth iteration, if for some r , M (k)(rh) ≥ c, then an upper bound t is found; otherwise for a given
tolerance δ, if maxr=0,1,2,...,m∣∣M (k)(rh)− M (k−1)(rh)
∣∣ < δ, then the sequence
M (k)(t)
converges; in thiscase, let j = j + 1 and repeat Steps 3–6. However, if maxr=0,1,2,...,m
∣∣M (k)(rh)− M (k−1)(rh)∣∣ ≥ δ, then
we use the interpolation
M (k)(τ ) = Interpolation [rh,M (k)(rh)r=0,...,m]
to approximate M (k)(τ ) and continue the iterative process for the (k + 1)th iteration. In this way, we obtainthe upper bound t (to be j ∗ tl ) and the lower bound t (to be ( j − 1) ∗ tl ) for tq .
Step 7. Let the lower and upper bounds t and t from Step 6 be the initial estimates of a lower bound t (0) and an upperbound t (0) of tq respectively.
Step 8. Let t (n) and t (n) be the (n + 1)th estimates of lower and upper bounds of tq respectively, t = (t (n) + t (n))/2,and h = t/m. With M (0)(τ ) = 0, and M (k) (0) = 0, we use (5.6) for N = 2, or (5.7) for N ≥ 3.
Step 9. At the kth iteration, if for some r , M (k)(rh) ≥ c, then t (n+1)= t (n), and t (n+1)
= t . We go to Steps8 and 9. If for the given tolerance δ, maxr=0,1,2,...,m
∣∣M (k)(rh)− M (k−1)(rh)∣∣ < δ, then the sequence
M (k)(t)
converges; in this case, let t (n+1)= t, t (n+1)
= t (n) and repeat Steps 8 and 9. However, ifmaxr=0,1,2,...,m
∣∣M (k)(rh)− M (k−1)(rh)∣∣ ≥ δ, then we use the interpolation to approximate M (k)(τ ) and
C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514 1513
continue the iterative process for the (k + 1)th iteration to obtain an upper bound t and a lower bound t , andgo to Steps 8 and 9.
Step 10. After n iterations, if |t (n) − t (n)| < ε for a given tolerance ε, then t (n)q = (t (n) + t (n))/2 is accepted as thefinal estimate of tq .
For illustration, let N = 3, R = 1, and f (u) = 1/(1 − u). We note that max0≤M(t)≤c (M(t)/ f (M(t))) = 1/4.From (4.7), α∗
= 1/4. From (5.3),∫∞
1/√
2tlexp
(−η2
) ∫ π
0
[exp
(η2 cosϕ1
)]sinϕ1dϕ1dη =
(2π)1/2
α
(14
).
Using Mathematica version 5.2 to do the integration, we obtain
√2π +
√2tl −
√2tle−1/tl −
√2π erf(1/
√tl)−
√2π
4α= 0.
For each given α > α∗, we use
Plot
[√
2π +√
2t −√
2te−1/t−
√2πErf[1/
√t] −
√2π
4α, t, 0, 10,PlotRange- > All
]to find an approximate value z for tl . With this, we use
FindRoot
[√
2π +√
2t −√
2te−1/t−
√2πErf[1/
√t] −
√2π
4α== 0, t, z
]to find tl . Using Steps 1–10 with ε = 10−4, δ = 10−6, and m = 40, we obtain the following table for tq (to foursignificant figures):
α tl from (5.3) tq for N = 3
1.5α∗ 2.525 5.8592α∗ .9250 2.0462.5α∗ .5283 1.1203α∗ .3542 .7303
Similarly, we obtain the following results:
α tl from (5.2) tq for N = 2
.3750 1.168 2.414
.5000 .6797 1.365
.6250 .4531 .8964
.7500 .3247 .6389
We note that in either case, the quenching time tq is a decreasing function of α.
References
[1] Q.Y. Dai, Y.G. Gu, A short note on quenching phenomena for semilinear parabolic equations, J. Differential Equations 137 (1997) 240–250.[2] Q.Y. Dai, X.Z. Zeng, The quenching phenomena for the Cauchy problem of semilinear parabolic equations, J. Differential Equations 175
(2001) 163–174.[3] I. Stakgold, Boundary Value Problems of Mathematical Physics, vol. II, Macmillan, New York, NY, 1968, p. 198.[4] C.Y. Chan, H.Y. Tian, Multi-dimensional explosion due to a concentrated nonlinear source, J. Math. Anal. Appl. 295 (2004) 174–190.[5] K.R. Stromberg, An Introduction to Classical Real Analysis, Wadsworth, Belmont, CA, 1981, pp. 266, 268, 328, 352, 369–370, 380.[6] C.Y. Chan, P.C. Kong, Solution profiles beyond quenching for degenerate reaction–diffusion problems, Nonlinear Anal. 24 (1995) 1755–1763.[7] C.Y. Chan, J. Yang, Beyond quenching for degenerate singular semilinear parabolic equations, Appl. Math. Comput. 121 (2001) 185–201.
1514 C.Y. Chan, P. Tragoonsirisak / Nonlinear Analysis 69 (2008) 1494–1514
[8] A. Friedman, Partial Differential Equations of Parabolic Type, Prentice Hall, Englewood Cliffs, NJ, 1964, p. 34.[9] L.C. Evans, Partial Differential Equations, Graduate Studies in Mathematics, vol. 19, American Mathematical Society, Providence, RI, 1998,
pp. 22, 615.[10] C.Y. Chan, H.G. Kaper, Quenching for semilinear singular parabolic problems, SIAM J. Math. Anal. 20 (1989) 558–566.[11] C.Y. Chan, X.O. Jiang, Quenching for a degenerate parabolic problem due to a concentrated nonlinear source, Quart. Appl. Math. 62 (2004)
553–568.