Professional Documents
Culture Documents
Jack Xin
where t 1, h(x) is a real C 2 (R1 ) function that has an absolute maximum at x = 0, w.l.o.g h(0) = 0, h(x) < 0 for x = 0 such that h(x) b, if |x| c, for some positive numbers b, c. Suppose also h(x) fast enough as x to ensure convergence of F for t = 1; h (0) < 0. Question: what is the large t behavior of F ? Intuition: the integrand decays exponentially fast if x is away from the peak of h(x), hence it suces to look at the integral over a small neighborhood of x = 0 where we approximate h(x) h (0)x2 /2. Lemma 1.1 (Laplace Lemma) As t , to leading order: F (t) (2)1/2 (th (0))1/2 , if h(0) = 0, then: F (t) (2)1/2 (th (0))1/2 eth(0) .
(1.2)
(1.3)
Department of Mathematics and TICAM, University of Texas at Austin, Austin, TX 78712. Email: jxin@math.utexas.edu.
Sketch of Proof: It follows from our assumption that for each > 0, there is () > 0 such that: h(x) (), if |x| . Note that for t > 1:
|x|
e(t1)( )
R1
eh(x) dx,
For any (0, |h (0)|/3), we can nd = ( ) > 0 so that if |x| , we have: 1 |h(x) x2 h (0)| x2 . 2 This is due to h(x) 1 x2 h (0) C 2 , (0) = (0) = (0) = 0, so (x) = o(x); then 2 by mean value theorem, (x) = o(x2 ). It follows:
[ , ]
exp{
tx2 (h (0) 2 )} dx 2
[ , ]
exp{th(x)} dx
[ , ] R1
exp{
Each of the three integrals diers from the corresponding independent of t. Using the fact:
(1.4)
with n = 0 to evaluate
R1
R1
to nd:
eth(x) dx (2)1/2 (h (0) 2 )1/2 t1/2 + O(et ) (2)1/2 (h (0) 3 )1/2 t1/2 . being arbitrary, we have shown (1.2).
A similar lower bound holds. The number Corollary 1.1 If g(x) C ( R1 ), and
R1 R1
Question: What if the maximum of h(0) occurs at the boundary ? Practice Problem 1: Suppose that h(x) attains maximum at x = 0, h C 1 ((0, )), h (0) < 0; h(x) < h(0) for x > 0, h(x) as x ; and 0 eh(x) dx converges. Show that to leading order: eth(x) dx (th (0))1 eth(0) ,
0
as t . 2
Remark 1.1 Results above hold for complex t as long as |arg t| < 0 < /2. Also R1 or 1 R+ can be replaced by any interval, x = 0 replaced by an interior or boundary point. Example 1: Find the large t behavior of function: (t + 1) =
0
eu ut du.
Solution: Notice that the integrand is not in standard form, in fact its peak location is u = t, changing with t. Making the change of variables u = t(1 + x), then: (t + 1) = et tt+1
1
The function h(x) satises the assumptions of Laplace Lemma with h (0) = 1, and so: (t + 1) et tt+ 2 (2)1/2 , as t , known as the Stirlings formula. The extension of Laplace method to higher order terms is called: Lemma 1.2 (Watsons Lemma) Suppose that g(x) is analytic in a neighborhood of t = 0,
g(x) =
k=0
ak xk , |x| R,
n
and that: |g(x)| c1 ec2 x , x [R, X], for positive constants c1 , c2 , positive integer n, then:
X 0
etx g(x) dx
(1.6)
as t in the sector |arg t| 0 < /2. The proof can be adapted from that of Laplace Lemma with the help of (1.4) for n 1. Example 2: Consider ex F (t) = dx, x+t 0 3
for complex t. Again this is not in the standard form, so make the change of variables: x = |t|s to nd: e|t|s F (t) = |t|/t ds. 1 + |t|s/t 0
1 The function g(s) = 1+|t|s/t is analytic if s (0, 1), and bounded for s 1. Let z with its phase xed, so |z|/z = ei , [0, 2) xed. We apply Watsons lemma.
g(s) = and
(1)k eik sk ,
k=0
F (t) ei
m=1
exp{th(x)} dx,
and assume that h(0) = 0, h(x) < 0 for x [1, 1]d Q and x = 0, h C 2 (Q), and the maximum is of elliptic type at x = 0. That is: 1 h(x) = xT Ax + o(|x|2 ), 2 A is positive denite. Then: F (t) (2)d/2 (det(A))1/2 td/2 . Practice Problem 2: Show that there is asymptotic expansion:
0 0
(1.7)
(1.8)
en(x
2 +y 2 )
as n .
PDE Applications
ut + uux = uxx , > 0, 4 (2.1)
with initial data u(x, 0) = F (x). Write u = 2x 1 , the Hopf-Cole transform, then t = xx . Solution is: x G/(2) e d 1 u(x, t) = R t G/(2) , (2.2) d R1 e where: (x )2 G = G(; x, t) = F ( ) d + . (2.3) 2t 0 In the limit 0+ , inviscid limit, we encounter asymptotics of integrals with 1 playing the role of t. If the initial data is such that G has a unique minimum point = , then Laplace Lemma applies. We have:
R1
(2.7)
which is the solution formula of the inviscid Burgers equation ut + uux = 0 with initial data F (x) via the method of characteristics [4]. The Laplace method also has a stochastic version when G is a random process, arising from random initial condition F (x), see [3] for an application of Laplace method on front solutions in noisy Burgers equation.
References
[1] N. De Bruijn, Asymptotic Methods in Analysis, Dover, NY, 1981. [2] J. Keener, Principles of Applied Mathematics, Addison-Wesley, 1988. [3] J. Wehr, J. Xin, White Noise Perturbation of the Viscous Shock Fronts of the Burgers Equation, Comm. Math. Phys., 181, pp 183-203, 1996. [4] G. Whitham, Linear and Nonlinear Waves, John-Wiley and Sons, 1974. 5