Professional Documents
Culture Documents
1. Introduction
The error function defined by
2
erf(u) :=
(1.1)
ex dx
0
is one of the basic non-elementary special functions. The coefficient 2/ is a normalization factor that has the effect of giving erf() = 1 in view of the normal integral
x2
e
dx =
(1.2)
.
2
0
The reader will find in [2] different proofs of this evaluation.
In this paper we produce evaluations of entries in [4] that contain this function.
The methods are elementary. The reader will find in [3] a more advanced approach to
the question of symbolic integration around this function.
2. Elementary integrals
The table [4] contains many integrals involving the error function. This section
contains some elementary examples.
Lemma 2.1. Define
(2.1)
Fn (v) =
tn et dt.
0
Fn (v) = 21 v n1 ev +
n1
2 Fn2 (v),
2
1
1 ev .
erf(v) and F1 (v) =
(2.3)
F0 (v) =
2
2
Proof. This follows simply by integration by parts.
The recurence (2.2) shows that F2n (v) is determined by F0 (v) and F2n+1 (v) by
F1 (v). An explicit formula for the latter is easy to establish.
Lemma 2.2. The function F2n+1 (v) is given by
n
2j
X
2
v
n!
1 ev
.
(2.4)
F2n+1 (v) =
2
j!
j=0
n1 2j
2 X v
(n 1)!
1 2n v2
1 ev
+n
v e
2
2
j!
j=0
n1
X v 2j
1 2n v2 n!
v 2
.
v e
+
1e
2
2
j!
j=0
Now observe that the first term matches the one in the sum for j = n.
(2.5)
(2.6)
F2n (v) =
Pn (v) =
n1
X
k=0
(2n 1)!! 2k 2k
v .
(2k + 1)!!
3. Elementary scaling
Several entries in [4] are obtained from the expressions in the last section via
simple changes of variables. For example, a linear transformation gives
Z v
2 2
1
xn eq x dx = n+1 Fn (qv).
(3.1)
q
0
This section present some examples of this type.
ERROR FUNCTION
Z u
2 2
erf(qu).
eq x dx =
(3.2)
2q
0
Example 3.2. Entry 3.321.3
q 2 x2
e
dx =
(3.3)
2q
0
q 2 u2
,
erf(qu)
que
x2 eq x dx =
2q 3
2
0
Z u
2 2
2 2
1
x3 eq x dx =
1 (1 + q 2 u2 )eq u ,
4
2q
0
Z u
2 2
1
3
3
q 2 u2
2 2
x4 eq x dx =
.
que
erf(qu)
+
q
u
2q 5
4
2
0
Example 3.4. Simple scaling produces other integrals in [4]. For example, starting with
Z
2
ex dx =
(1 erf(a)) ,
(3.4)
2
a
Z
2 2
(1 erf(qa)) .
eq t dt =
(3.5)
2q
a
The integral
p + 2aq 2
,
e
1 erf
(3.6)
2q
a
is now computed by completing the square. The choice q = 1/2 and p = appears
as entry 3.322.1 in [4]:
Z
p
p
u
x2
2
1 erf
.
+
exp
(3.7)
x dx = e
4
2
u
Z
q 2 x2 px
p2 /4q2
dx =
e
2q
In order to minimize the choice of greek letters (clearly a personal choice of the authors)
it is suggested to write this entry using the notation in (3.6).
Z
p2 /4q2
p
q 2 x2 px
e
dx =
(3.8)
.
1 erf
e
2q
2q
0
Z
2
p+2
p2 /4
.
1 erf
e
ex px dx =
(3.9)
2
2
1
Z
q
2 2
exp p x qx dx =
(3.10)
exp
2
p
4p
p2 x2 qx = p2 x q/2p2
(3.11)
2
q2
.
4p2
2 X (1)k u2k+1
(4.1)
erf(u) =
k! (2k + 1)
k=0
that comes from term by term integration of the power series of the integrand. The
second one is more interesting and is given in the next example.
Example 4.1. Entry 3.321.1 states that
2 X
2
2k
(4.2)
erf(u) = eu
u2k+1 .
(2k + 1)!!
k=0
j
X
X
X
(1)r
2
(1)k 2k
u
u2j+1 =
u2r+1 .
k!
(2j
+
1)!!
r!(2r
+
1)
r=0
j=0
k=0
Multiplying the two series on the left, we conclude that the result follows from the
finite sums identity
r
X
(1)r
(1)k 2rk
=
.
(4.3)
k! (2r 2k + 1)!!
r! (2r + 1)
k=0
r
X
1
r
2k
2r + 1
(4)
= 1,
2k + 1
k
k
k
k=0
ERROR FUNCTION
Both terms A and B have natural boundaries, that is, they vanish outside the summation range. Summing from k = to k = + and using the telescoping property
of the right-hand side shows that
r
X
A(r, k)
(4.8)
ar :=
k=0
The first example in this section reproduces a classical integral due to P. Laplace.
Example 5.1. Entry 3.325 states that
r
Z
1 2ab
2
2
e
exp ax bx
dx =
.
(5.1)
2 a
0
To evaluate this, we complete square in the exponent and write
Z
Z
2
2
2
2 ab
e( ax b/x) dx.
exp ax bx
dx = e
(5.2)
0
2
e( ax b/x) dx.
(5.3)
J :=
0
b/ ax produces
Z
2 dt
b
e( at b/t) 2 .
(5.4)
J :=
t
a 0
The average of these two forms for J produces
Z
2
1
(5.5)
J=
a + b/x2 dx.
e( ax b/x)
2 a 0
1
u2
e
du = ,
(5.6)
J=
2 a
2 a
The change of variables t =
The reader will find in [1] details about this transformation and the evaluation of
many related integrals.
Example 5.2. Entry 3.472.1 of [4]:
r
Z
x2
1 2a
2
1
exp a/x 1 e
dx =
(5.8)
e
2
0
(5.10)
I=2
eat
b/t2
dt.
satisfies
In
= In+1 (p).
p
(5.12)
Using this it is an easy induction exercise, with (5.9) as the base case, to verify the
evaluation
n h
Z
i
ERROR FUNCTION
(6.2)
dt =
erf(u).
q
t
0
dt =
(6.3)
erf( qv).
q
t
0
This appears as 3.361.1 in [4].
dt =
(6.4)
.
q
t
0
The change of variables x = t + a produces
r
Z qx
e
aq
.
dx = e
(6.5)
q
x
a
a
The special case a = 1 appears as 3.361.3:
r
Z qx
e
(6.6)
,
dx = e
q
x+1
1
and a = 1 appears as 3.362.1:
r
Z qx
e
dx = e
(6.7)
.
q
x1
1
Z 2 a2 x2
b a2 b2
x e
dx =
e
(1 erf(ab))
(6.11)
2
2
x +b
2a
2
0
is obtained by writing
Z
2 2
a2 b2
a2 b2
x2 ea x dx
= 2ae
Ie
a
0
and the integral I can be evaluated via the change of variables t = ax, to get
2 2
a2 b2
Ie
= 2 ea b .
a
2a
2 2
ea b
a2 b2
b (1 erf(ab)) .
(6.12)
Ie
=
2
a
Now simplify to produce the result.
Example 6.5. The evaluation of entry 3.462.5:
r
Z
2
1
2 /
xex 2x dx =
(6.13)
e
(1 erf(/ ))
2 2
0
can also be obtained in elementary terms. The change of variables t =
by y = t + c with c = / yields
x followed
(6.14)
I=
ec
J
where
(6.15)
J=
(y c)ey dy.
The first integrand is a perfect derivative and the second one can be reduced to twice
the normal integral to complete the evaluation.
Example 6.6. The integral in entry 3.462.6:
r
Z
q
px2 2qx
xe
dx =
(6.16)
exp(q 2 /p)
p p
The first integral is elementary and the second one can be reduced to twice the normal
integral to produce the result.
Example 6.7. Similar arguments as those presented above yield entry 3.462.7:
r
Z
2 2 + 2 /
2 x2 2x
e
(1 erf(/ )),
x e
dx = 2 +
(6.19)
5
2
4
0
and 3.462.8:
(6.20)
2 x2 +2x
x e
1
dx =
2
(1 + 2 2 /)e / .
ERROR FUNCTION
x u qx
qu
e
dx =
e
u(1 erf( qu)).
(7.1)
x
q
u
The evaluation is elementary, but more complicated than those in the previous section.
We first let x = u + t2 to produce I = 2equ J, where
Z
2
t2
eqt dt.
(7.2)
J=
2
t +u
0
J=
(7.4)
so that
(7.5)
qt2
dt u
eqt
dt.
u + t2
/2 q and we let
Z r2
e
K=
dr,
2 + qu
r
0
I=
qu
e
2u qequ K.
q
(7.6)
K= e
.
2
s sv
v
The scaling s = vy reduces the question to the evaluation of
Z
evy
(7.7)
T =
dy.
y y1
1
Observe that
(7.8)
T
=
v
evy
dy =
y1
v
e ,
v
e
dr
T =
r
v
Z
Z v r
r
e
e
dr
dr
=
r
r
0
0
= (1 erf( v)).
10
1
1
xu
xu
(7.9)
x
u
x
xu
and the results of 3.362.2 and 3.363.1 give an evaluation of entry 3.363.2:
Z qx
e
dx
(7.10)
= (1 erf( qu)) .
u
x
x
u
u
8. Differentiation with respect to a parameter
The evaluation of the integral J described in the previous section is an example
of a very powerful technique that is illustrated below.
Example 8.1. The evaluation of 3.466.1:
Z a2 x2
2 2
e
dx
(8.1)
= (1 erf(ab))ea b ,
2
2
x +b
2b
0
is simplified first by the scaling x = bt. This yields the equivalent form
Z c2 t2
2
e
dt
= (1 erf(c))ec ,
(8.2)
2
1
+
t
2
0
with c = ab. Introduce the function
Z c2 (1+t2 )
e
(8.3)
f (c) =
dt
1 + t2
0
and the identity is equivalent to proving
(8.4)
f (c) = (1 erf(c)).
2
Differentiation with respect to c we get
Z
2
2
2
e(ct) dt = ec .
(8.5)
f (c) = 2cec
0
we get
(8.6)
f (c) =
(1 erf(c))
2
as required.
Example 8.2. The evaluation of entry 3.464:
(8.7)
ex ex
is obtained by introducing
(8.8)
f () =
dx
x2
,
ex ex
dx
x2
and differentiating with respect to the parameter we obtain
x2
e
dx = .
(8.9)
f () =
2
0
0
ERROR FUNCTION
11
1
2
e erfc( aq)
L 1 (a, q) =
q
2
These results are special cases of entry 3.382.4, that gives Lb (a, u) in terms of the
incomplete gamma function
Z
et ta1 dt
(9.5)
(a, c) :=
c
in the form
(9.6)
In this paper only the case b = n 12 is considered. Details for the general situation
will be given elsewhere.
Example 9.1. The integral in 3.362.2:
r
Z qx
aq
e
dx
e (1 erf qa)
=
(9.7)
q
x
+
a
0
and scaling by y =
q s yields
(9.9)
2eqa
I=
q
qa
2eqa
I=
q
ey dy
qa
e
0
y 2
dy ,
and now just write this in terms of the error function to get the stated result.
12
Pm (a)
ex dx
(1)m 2m a
e erfc( a) m1 m1/2 ,
=
(9.11)
(2m 1)!!
(x + a)m+1/2
2
a
0
where Pm (a) is a polynomial that satisfies the recurrence
(9.12)
d
Pm1 (a) (2m 3)Pm1 (a)
da
e dx
(9.13)
= ea erfc a.
x+a
0
Now differentiate m times with respect to a and using
m
d
(1)m (2m 1)!!
1
(9.14)
= m
da
2 (x + a)m+1/2
x+a
and the ansatz
m
a
Pm (a)
d
e erfc( a) = ea erfc( a) m1 m1/2
(9.15)
da
2
a
give the recurrence for Pm (a).
It is possible to obtain a simple expression for the polynomial Pm (a). This is given
below.
Corollary 9.2. Define
(9.16)
Then
(9.17)
Rm (a) =
m1
X
j=0
(2j 1)!!am1j .
In summary, the integral considered in this section is given in the next theorem.
Theorem 9.3. Let m N and a > 0. Then
Z
m1
m m
x
j
X
(1) 2
e dx
(1) (2j 1)!!
1
=
.
ea erfc( a)
(2m 1)!!
(2a)j
a j=0
(x + a)m+1/2
0
ERROR FUNCTION
13
1 (a, q)
m+ 2
=
=
eqx dx
(x + a)m+1/2
0
m1
(1)m 2m q m1/2 aq
1 X (1)j (2j 1)!!
.
e erfc( aq)
(2m 1)!!
aq j=0
(2aq)j
The table [4] contains a small number of entries that involve the complementary
error function defined in (9.4). To study these integrals introduce the notation
Z
2
xn erfcm (x)ebx dx.
(10.1)
Hn,m (b) :=
0
The table [4] contains the values H0,2 (b)in 8.258.1, H1,2 (b) in 8.258.2 and 8.258.3
is H3,2 (b). The change of variables x = t yields the form
Z
1 n1
t 2 erfcm ( t)ebt dt.
(10.2)
Hn,m (b) :=
2 0
In this format, entry 8.258.4 contains H3,1 (b) and H2,1 (b) appears as 8.258.5. This
section contains an analysis of this family of integrals.
Example 10.1. The value
(10.3)
H0,0 (b) =
2 b
is elementary.
The next result presents a recurrence for these integrals.
Proposition 10.1. Assume n > 2 and m > 1. The integrals Hn,m (b) satisfy
(10.4)
Hn,m (b) =
n1
m
Hn2,m (b) Hn1,m1 (b + 1).
2b
b
1
Hn,m (b) =
2b
xn1 (erfc x)
d bx2
e
dx.
dx
Note. The family Hn,m (b) is determined by the initial conditions H0,m (b), H1,m (b)
and Hn,0 (b). Each of these are analyzed below.
The family Hn,0 (b) is easy to evaluate.
Lemma 10.2. The integral Hn,0 (b) is given by
(10.6)
Hn,0 (b) =
1 n+1
b 2
2
n+1
2
14
Hn,0 (b) =
xn ebx dx
fn,m (y, b) =
Z
xy
t2
dt
m
ebx dx.
Then
(10.9)
d
fn,m (y, b) = mfn+1,m1 (y, b + y 2 ).
dy
(10.10)
Hn,m (b) = Hn,0 (b) m
fn+1,m1 (y, b + y 2 ) dy.
0
Proof. Integrating (10.9) and using the values
m
m
(10.11)
fn,m (1, b) =
Hn,m (b) and fn,m (0, b) =
Hn,0 (b)
2
2
gives the result.
2 b
0
The next examples are obtained by specific choices of the parameter m.
Example 10.2. The first example deals with m = 1. In this case, Corollary 10.5
reduces to
Z 1
2
(10.12)
H0,1 (b) =
f1,0 (y, b + y 2 ) dy.
2 b
b 0
The value f1,0 (y, r) =
(10.13)
1
2r
produces
tan1 b
.
H0,1 (b) =
b
The computation of H0,2 (b) employs and alternative expression for the integrand
in Corollary 10.5.
ERROR FUNCTION
15
b + 2y 2
(m 1) m/21
( /2)m1
2
.
m1
H0,m2
(10.14)
f1,m1 (y, b + y ) =
2(b + y 2 )
2
(b + y 2 )
y2
Proof. Integration by parts gives
Z
Z
2
2
xe(b+y )x
f1,m1 (y, b + y 2 ) =
0
=
=
xy
et dt
m1
dx
m1
Z
Z
d (b+y2 )x2
1
t2
e
dt
dx
e
2(b + y 2 ) 0 dx
xy
m2
Z
Z
( /2)m1
(m 1) m/21 1
t2
(b+2y 2 )x2
dt
e
dx.
e
2(b + y 2 )
2m1
b + y2 0
xy
This expression for the integrand in Corollary 10.5 gives the next result.
Corollary 10.7. The integral H0,m (b) satifies
Z
m(m 1) H0,m2 (t + 2) dt
m
1
+
H0,m (b) = tan1
.
t1/2 (t + 1)
2 b
b
b
b
b
Example 10.3. The case m = 2 in the previous formula yields the value of H0,2 (b).
The value of H0,0 (b) in (10.3) produces
Z
1
dt
2
1 1
1
1
2
1
1
bx2
b cos
2 tan
.
e
erfc (x) dx =
(10.15) H0,2 (b) :=
1+b
b
0
This appears as entry 8.258.1 in [4].
The family H1,m (b). This is the final piece of the initial conditions.
Proposition 10.8. The integral H1,m (b) satisfies the relation
1
2m
(10.16)
H1,m (b) =
1
H0,m1 (b + 1) .
2b
16
Entry 8.258.2
(10.20)
H1,2 :=
x erfc x e
bx2
1
dx =
2b
4 tan1 ( 1 + b)
1
,
1+b
The change of variables t = px shows that this formula is equivalent to the special
case p = 1. This is an instance of a fake parameter.
To show that
Z
q + 21
2q1
,
(1 erf(t)) t
dt =
(11.2)
2 q
0
q
0
0
The change of variables s = t2 gives the result.
ERROR FUNCTION
17
Example 11.3. Entry 6.282.2, with a minor change from the stated formula in
the table, is
Z
1
1
+1
( + 1)2
erf x + 12 erf 21 ex+ 4 dx = exp
1 erf
.
4
2
0
Integration by parts gives
Z
erf x + 12 erf
(11.6)
0
1
2
1
x+ 4
dx =
ex
(+1)x
dx.
The result follows by completing the square. This entry in [4] has the factor in the
denominator replaced by ( + 1)( + 2). This is incorrect. The formula stated here is
the correct one.
Example 11.4. Entry 6.283.1 states that
Z
e
1 erf( x) dx =
(11.7)
1 .
2 t2 /
ex 1 erf( x) dx =
(11.8)
te
erfc t dt.
0
0
.
p p+q
2 px2 /q
pt
xe
erf x dx.
erf( qt) e dt =
(11.10)
q 0
0
Integration by parts, with erf x the term that will be differentiated, gives the result.
Note. The table [4] contains many other integrals containing the error function with
results involving more advanced special functions. For instance, entry 6.294.1:
Z
2 2 dx
[1 erf(1/x)] e x
(11.11)
= Ei(2),
x
0
where Ei denotes the exponential integral. These will be described in a future publication.
Acknowledgments. Matthew Albano and Erin Beyerstedt were partially supported
as students by NSF-DMS 0713836. The work of the last author was also partially
supported by the same grant.
18
References
[1] T. Amdeberhan, L. Glasser, M. C. Jones, V. Moll, R. Posey, and D. Varela. The CauchySchl
omilch transformation. Submitted for publication, 2010.
[2] G. Boros and V. Moll. Irresistible Integrals. Cambridge University Press, New York, 1st edition,
2004.
[3] G. W. Cherry. Integration in finite terms with special functions: the error function. J. Symb.
Comput., 1:283302, 1985.
[4] I. S. Gradshteyn and I. M. Ryzhik. Table of Integrals, Series, and Products. Edited by A. Jeffrey
and D. Zwillinger. Academic Press, New York, 7th edition, 2007.
[5] M. Petkovsek, H. Wilf, and D. Zeilberger. A=B. A. K. Peters, Ltd., 1st edition, 1996.
Department of Mathematics, Rutgers University at Newark, Newark, NJ 07102
E-mail address: malbano@rutgers.edu
Department of Mathematics, Tulane University, New Orleans, LA 70118
E-mail address: tamdeber@math.tulane.edu
Department of Mathematics, Tulane University, New Orleans, LA 70118
E-mail address: ebeyerst@math.tulane.edu
Department of Mathematics, Tulane University, New Orleans, LA 70118
E-mail address: vhm@math.tulane.edu
Received 23 12 2010, revised 13 04 2011
tica
Departamento de Matema
cnica Federico Santa Mara
Universidad Te
Casilla 110-V,
Valparaso, Chile