Physics 214 Winter 2017

The Riemann-Lebesgue Lemma

The Riemann Lebesgue Lemma is one of the most important results of Fourier anal-
ysis and asymptotic analysis. It has many physics applications, especially in studies
of wave phenomena. In this short note, I will provide a simple proof of the Riemann-
Lebesgue lemma which will be adequate for most cases that arise in physical applica-
tions.
The simplest form of the Riemann-Lebesgue lemma states that for a function f (x)
for which the integral
Z b
|f (x)| dx < ∞ , (1)
a

where a and b are real numbers, we have

Z b
lim f (x) eikx dt = 0 . (2)
k→∞ a

If we further assume that f (x) has certain “nice” properties [e.g., a sufficient (but not
necessary) condition is that f (x) is continuously differentiable for a ≤ x ≤ b], then it
follows that
Z b  
ikx 1
f (x) e dx = O , as k → ∞. (3)
a k
Moreover, eqs. (2) and (3) continue to hold if a → −∞ and/or b → ∞, assuming that
eq. (1) holds over the infinite interval.
We will present a proof of eq. (2) under the assumption that f (x) is continuous
over the closed interval a ≤ x ≤ b. The origin of eq. (2) in this case is not too difficult
to understand. In the limit of k → ∞, the factor eikx oscillates faster and faster such
that f (x) eikx averages out to zero over any finite region of x inside the interval.
If we can assume that f (x) is N-times differentiable in the region a ≤ x ≤ b, one
can derive eq. (3) simply by a repeated integration by parts. Namely,
Z b b
f (x) ikx  1 b ′
Z
ikx
f (x) e dx = e  − f (x) eikx dx
a ik a ik a
b Z b
eikb f (b) − eika f (a) f ′ (x) ikx  1
= − 2
e  + 2
f ′′ (x) eikx dx
ik (ik) a (ik) a
N −1
eikb f (n) (b) − eika f (n) (a)
 
X
n 1
= (−1) n+1
+O , (4)
n=0
(ik) k N +1

1
where f (0) (x) ≡ f (x) and f (n) (x) ≡ dn f /dxn . For example f ′ (b) is equal to the first
derivative of f (x) evaluated at x = b, etc. Taking N = 1 in eq. (4), we see that the
leading term that survives is of O(1/k) as asserted by eq. (3).
More generally, we can prove eq. (2) without the assumption f (x) is differentiable
in the interval. We do this by writing the integral
Z b
I(k) = f (x) eikx dx ,
a

in two different but equivalent ways:1

Z a+π/k Z b
ikx
I(k) = f (x) e dx + f (x) eikx dx (5)
a a+π/k

and
Z b−π/k Z b
ikx
I(k) = f (x) e dx + f (x) eikx dx . (6)
a b−π/k

By a change of variables, x′ = x − π/k, it is straightforward to verify that

Z b Z b−π/k 
ikx π  ikx
f (x) e dx = − f x+ e dx , (7)
a+π/k a k

after using e−iπ = −1 and dropping the primes from the x in the second integral. Thus,
writing I as one half the sum of eqs. (5) and (6), and employing eq. (7), it follows that
Z a+π/k Z b
ikx
I(k) = 21
f (x) e dx + 2 1
f (x) eikx dx
a b−π/k

b−π/k
π i ikx
Z h 
1
+ 2
f (x) − f x + e dx . (8)
a k
We not take the limit of k → ∞. The mean value theorem of for integrals states
that if f (x) is continuous and bounded over a closed interval, a ≤ x ≤ b, then
Z b
f (x) dx = f (c)(b − a) ,
a

for some real number c that lies in the interval a ≤ c ≤ b. Applying this to the first
two integrals in eq. (8), we immediately conclude that
Z a+π/k   Z b  
ikx 1 ikx 1
f (t) e dx = O , f (k) e dx = O ,
a k b−π/k k
which vanish in the limit of k → ∞.
1
This proof is taken from Brian Davies, Integral Transforms and Their Applications, 3rd edition
(Springer-Verlag, New York, 2002) pp. 39–40.

2
 Finally, under the assumption that f (x) is continuous at all points in the closed
interval a ≤ x ≤ b, it follows that
b−π/tk
π i ikx
Z h 
lim f (x) − f x + e dx = 0 . (9)
k→∞ a k
This is true because a function that is continuous at all points in a closed, bounded
interval is uniformly continuous over the interval.2 Hence, one can make the integrand
in eq. (9) arbitrarily small by choosing k sufficiently large. The limit of eq. (9) is thus
established, and the Riemann-Lebesgue lemma stated in eq. (2) is proven.
Note that the argument above does not necessarily imply that
Z b−π/k h  
 π i ikx 1
f (x) − f x + e =O , (10)
a k k

as k → ∞. However, if the function f (x) is continuously differentiable in the interval,

then we can employ the mean value theorem for differentiable functions, which states
that
f (b) − f (a) = f ′ (c)(b − a) , for some c between a and b .
It follows that
 π π π
f (x) − f x + = − f ′ (x + c) , for 0 ≤ c ≤ .
k k k
Hence, in this case eq. (10) does hold, in which case eq. (3) is satisfied.3
The extension to cases where a → −∞ and/or b → ∞ is straightforward. For
example, suppose that
Z ∞
|f (x)|dx < ∞ . (11)
a

Then, noting that one can write

Z ∞ Z b
ikx
f (x)e dx = f (x)eikx dx + ǫ ,
a a

where Z ∞
ǫ< |f (x)|dx ,
b
2
For further details, see e.g. David Bressoud, A Radical Approach to Real Analysis, 2nd edition
(The Mathematics Association of America, Washington, DC, 2007) pp. 228–229.
3
We also obtain eq. (10) under slightly weaker conditions in which the function f (x) satisfies the
so-called Lipschitz condition, |f (x) − f (y)| ≤ M |x − y| for all x and y in the interval for some positive
finite bound M . Indeed a Lipschitz continuous function is uniformly continuous (although the converse
is not necessarily true). A Lipschitz continuous function need not be differentiable. On the other hand
a differentiable function whose derivative is bounded on the interval satisfies the Lipschitz condition.
Thus, to establish eq. (10) is is sufficient to require that f (k) is Lipschitz continuous in the interval.

3
it follows [in light of eq. (11)] that it is always possible to make ǫ arbitrarily small by
a suitable (finite) choice of b. Hence, we can use eq. (2) to conclude that
Z ∞
lim f (x) eikx dx = 0 . (12)
k→∞ a

Finally, it should be noted that taking the real and imaginary parts of eq. (2) under
the assumption that f (x) is a real valued function yields,
Z b
lim f (x) sin kx dx = 0 , (13)
k→∞ a
Z b
lim f (x) cos kx dx = 0 , (14)
k→∞ a

with a similar extension to cases where a → −∞ and/or b → ∞. If we further assume

that f (x) has certain “nice” properties [cf. the comments above eq. (3)], then the two
integrals above behave as O(1/k) as k → ∞.
By consulting a table of Fourier transforms,4 one can see many examples of functions
that satisfy eq. (1). In all cases, you will find that the corresponding Fourier transform
satisfies eq. (2). It is interesting to look for cases that satisfy eq. (2) and not eq. (3).
For example,
Z ∞
Γ(ν)
xν−1 e−ax eikx dx = , for a > 0 and Re ν > 0 .
0 (a − ik)ν
Indeed, eqs. (1) and (2) are satisfied for all Re ν > 0, whereas eq. (3) is only satisfied
when Re ν ≥ 1.
Finally, we provide the following interesting example that contradicts eq. (13).
Choosing f (x) = sin(x2 ) and consulting I.S. Gradshteynn and I.M. Ryzhik, Table
of Integrals, Series and Products, 8th edition, edited by Daniel Zwillinger (Academic
Press, Elsevier, Amsterdam, 2015), we note the following result on p. 419 [formula
3.691 number 5],
 2
√
Z ∞ 
2 1 k +π
sin(x ) sin kx dx = 2 π cos ,
0 4
which does not vanish as k → ∞. However, this is not surprising in light of the fact
that eq. (11) [with a = 0] is not satisfied. Indeed, even though
Z ∞ r
2 1 π
sin(x ) dx = ,
0 2 2
is finite, it turns out that Z ∞
| sin(x2 )|dx
0
diverges. Thus, one of the key assumptions underlying the Riemann-Lebesgue lemma
does not hold in this case.
4
See e.g., F. Oberhettinger, Table of Fourier Transforms and Fourier Transforms of Distributions
(Springer-Verlag, Berlin, 1990).