Gaussian integral

2.5
Area=sqrt(pi)
e^(-x^2)
∫ +∞
e−x dx
2
2

−∞
1.5
can be evaluated.
1 The Gaussian integral is encountered very often in
physics and numerous generalizations of the integral are
0.5 encountered in quantum field theory.

1 Computation
-0.5
-2 -1 0 1 2

1.1 By polar coordinates

−x2
A graph of ƒ(x) = e and the area between the function and
the x-axis, which is equal to √π . A standard way to compute the Gaussian integral, the idea
of which goes back to Poisson,[2] is
The Gaussian integral, also known as the Euler–
2 2 2
Poisson integral[1] is the integral of the Gaussian func- • consider the function e−(x + y ) = e−r on the plane
2
tion e−x over the entire real line. It is named after R2 , and compute its integral two ways:
the German mathematician and physicist Carl Friedrich
Gauss. The integral is: 1. on the one hand, by double integration in the
Cartesian coordinate system, its integral is a
square:
∫ (∫ )2
+∞ √
e−x dx =
2
π −x2
−∞
e dx ;

This integral has a wide range of applications. For exam-

2. on the other hand, by shell integration (a case
ple, with a slight change of variables it is used to compute
of double integration in polar coordinates), its
the normalizing constant of the normal distribution. The
integral is computed to be π.
same integral with finite limits is closely related both to
the error function and the cumulative distribution func-
tion of the normal distribution. In physics this type of Comparing these two computations yields the integral,
integral appears frequently, for example, in quantum me- though one should take care about the improper integrals
chanics, to find the probability density of the ground state involved.
∫∫ ∫ 2π ∫ ∞
of the harmonic oscillator, also in the path integral for- −(x2 +y 2 )
e−r r dr dθ
2

mulation, and to find the propagator of the harmonic os- e d(x, y) =

R2
cillator, we make use of this integral.
0
∫ ∞0
re−r dr
2
= 2π
Although no elementary function exists for the error func- 0
tion, as can be proven by the Risch algorithm, the Gaus- ∫ 0
1 s
sian integral can be solved analytically through the meth- On the other hand, = 2π 2 e ds s=
−∞
ods of multivariable calculus. That is, there is no elemen- ∫ 0
tary indefinite integral for =π es ds
−∞

∫ = π(e0 − e−∞ )
e−x dx,
2
= π,
where the factor of r comes from the transform to polar
but the definite integral coordinates (r dr dθ is the standard measure on the plane,

1
2 1 COMPUTATION

expressed in polar coordinates ), and the substitution in- Taking the square of I(a) yields
volves taking s = −r2 , so ds = −2r dr.
Combining these yields (∫ a ) (∫ a )
2 −x2 −y 2
I(a) = e dx e dy
−a −a
(∫ ∞ )2 ∫ a (∫ a )
e−y dy e−x dx
2 2
e−x dx
2
= π, =
−∞ −a −a
∫ a ∫ a
−(x2 +y 2 )
so = e dx dy.
−a −a

∫ ∞ Using Fubini’s theorem, the above double integral can be

√
e−x dx =
2
π seen as an area integral
−∞

∫
1.1.1 Careful proof e−(x
2
+y 2 )
d(x, y),

To justify the improper double integrals and equating the

taken over a square with vertices {(−a, a), (a, a), (a, −a),
two expressions, we begin with an approximating func-
(−a, −a)} on the xy-plane.
tion:
Since the exponential function is greater than 0 for all real
numbers, it then follows that the integral taken over the
∫ a
square’s incircle must be less than I(a)2 , and similarly
e−x dx.
2
I(a) =
−a
the integral taken over the square’s circumcircle must be
greater than I(a)2 . The integrals over the two disks can
If the integral easily be computed by switching from cartesian coordi-
nates to polar coordinates:
∫ ∞
e−x dx
2

−∞ x = r cos θ
were absolutely convergent we would have that its Cauchy y = r sin θ
principal value, that is, the limit d(x, y) = r d(r, θ).
∫ ∫ ∫ ∫ √
2π a 2π a 2
−r 2
re−r dr dθ.
2
2
re dr dθ < I (a) <
lim I(a) 0 0 0 0
a→∞
(See to polar coordinates from Cartesian coordinates for
would coincide with help with polar transformation.)
Integrating,
∫ ∞
e−x dx.
2

−∞
π(1 − e−a ) < I 2 (a) < π(1 − e−2a ).
2 2

To see that this is the case, consider that

By the squeeze theorem, this gives the Gaussian integral
∫ ∞ ∫ −1 ∫ 1 ∫ ∞
|e −x2
| dx < −x2
−xe dx+ e −x2
dx+ ∫ −x
xe
2
∞ dx < ∞. √
e−x dx = π.
2
−∞ −∞ −1 1
−∞
so we can compute

∫ 1.2 By Cartesian coordinates

∞
−x2
e dx
−∞ A different technique, which goes back to Laplace
(1812),[3] is the following. Let
by just taking the limit

y = xs
lim I(a) dy = x ds.
a→∞
 3

Since the limits on s as y → ±∞ depend on the sign of 3 Generalizations

2
x, it simplifies the calculation to use the fact that e−x is
an even function, and, therefore, the integral over all real 3.1 The integral of a Gaussian function
numbers is just twice the integral from zero to infinity.
That is, Main article: Integral of a Gaussian function

∫ ∞ ∫ ∞
e −x2
dx = 2 e−x dx.
2 The integral of an arbitrary Gaussian function is
−∞ 0

Thus, over the range of integration, x ≥ 0, and the vari- ∫ ∞

√
ables y and s have the same limits. This yields: −a(x+b)2 π
e dx = .
−∞ a
∫ ∞∫ ∞ An alternative form is
e−(x
2
+y 2 )
I2 = 4 dy dx
∫0 ∞ (∫0
∞ )
−(x2 +y 2 ) ∫ ∞
√
=4 e dy dx −ax2 +bx+c π b2 +c
0 0 e dx = e 4a ,
∫ ∞ (∫ ∞ ) −∞ a
−x2 (1+s2 )
=4 e x ds dx
0 0 This form is very useful in calculating mathematical ex-
∫ ∞ (∫ ∞ ) pectations of some continuous probability distributions
−x2 (1+s2 )
=4 e x dx ds concerning normal distribution.
0 0
∫ ∞[ ]x=∞
See, for example, the expectation of the log-normal dis-
1 −x2 (1+s2 )
=4 e ds tribution.
0 −2(1 + s2 )
( ∫ ∞ ) x=0
ds
= 4 12
0 1 + s2 3.2 n-dimensional and functional general-
[ ]∞
= 2 arctan s ization
0
=π Main article: multivariate normal distribution
√
Therefore, I = π , as expected.
Suppose A is a symmetric positive-definite (hence invert-
ible) n×n covariance matrix. Then,
1.3 Proof by complex integral
  √
A proof also exists using Cauchy’s integral theorem. ∫ ∞ ∑n ∫ ∞ ( )
 1  1 T
exp − n
Aij xi xj d x = exp − x Ax d x =
n
−∞ 2 i,j=1 −∞ 2
2 Relation to the gamma function
where the integral is understood to be over Rn . This fact
is applied in the study of the multivariate normal distri-
The integrand is an even function,
bution.
∫ ∞ ∫ ∞ Also,
−x2 −x2
e dx = 2 e dx
−∞ 0
√   √
∫ ∑n ∑
Thus, after the change of variable x = t , this turns into 1 (2π)n 1
xk1 · · · xk2N 
exp −  n
Aij xi xj d x =
the Euler integral 2 i,j=1 det A 2 N !
N
σ∈S2N

∫ ∞ ∫ ∞ ( ) where σ is a permutation of {1, ..., 2N} and the extra fac-

1 −t − 1 1 √
e−x dx = 2
2
2 e t 2 dt = Γ = π tor on the right-hand side is the sum over all combinatorial
0 0 2 2
pairings of {1, ..., 2N} of N copies of A−1 .
where Γ is the gamma function. This shows why√the
factorial of a half-integer is a rational multiple of π . Alternatively,
More generally,
  √ 
∫ ∑
( ) 1 ∑ −1
n n
∫ ∞  1  (2π)n

Γ 1
f (⃗
x ) exp − A x x d n
x = exp (A )ij
e−ax dx =
b ij i j
b
1 2 i,j=1 det A 2 i,j=1
0 ba b
4 5 REFERENCES

for some analytic function f, provided it satisfies some ap-

propriate bounds on its growth and some other technical ∫ ∞ ∞ ( 3n+2m+p
1 ∑ bn cm dp Γ
criteria. (It works for some functions and fails for others. 4 3 2
eax +bx +cx +dx+f dx = ef 4
Polynomials are fine.) The exponential over a differential −∞ 2 n,m,p=0
n! m! p! (−a) 3n+2m+
4

operator is understood as a power series. n+p=0 mod 2

While functional integrals have no rigorous definition (or The n + p = 0 mod 2 requirement is because the integral
even a nonrigorous computational one in most cases), we from −∞ to 0 contributes a factor of (−1)n+p /2 to each
can define a Gaussian functional integral in analogy to term, while the integral from 0 to +∞ contributes a factor
the finite-dimensional case. There is still the problem, of 1/2 to each term. These integrals turn up in subjects
though, that (2π)∞ is infinite and also, the functional de- such as quantum field theory. For multivariate quartic
terminant would also be infinite in general. This can be Gaussian integrals is:
taken care of if we only consider ratios:
 
∫ ∞∑
∫ ∫∫ n
( 1 Γ( 1 ))n
f (x1 ) · · · f (x2N )e− exp − Df Aijkl1xi xj x∑  −1
1 d d
2 A(x2N +1 ,x2N +2 )f (x2N +1 )f (x2N +2 )d x2N +1 d x2N +2
k xl dn x = 2 4 1/4
∫ − ∫∫ 1 A(x −∞ = N , xσ(2) ) · · · A−1 (xσ(2N −1
A (xσ(1)det(A)
2N +1 ,x2N +2 )f (x2N +1 )f (x2N +2 )d x2N +1 d x2N +2 Df
d d
e 2 i,j,k,l=1 2 N!
σ∈S2N

where det(A) is a hyperdeterminant of the 4-tensor A

In the DeWitt notation, the equation looks identical to the
(which is simply the contraction of A with Levi Cevita
finite-dimensional case.
symbols as in the quadratic case). There is no simple way
to extend this to add 3-tensor and 2-tensor terms without
3.3 n-dimensional with linear term the use of graphs or diagrams similar to Feynman dia-
grams.
If A is again a symmetric positive-definite matrix, then
(assuming all are column vectors)
4 See also
∫ ∫ √
x n•
∑n ∑n List of integrals
(2π) n 1 Tof −1
Gaussian functions
− 12 i,j=1 Aij xi xj + i=1 Bi xi n − 12 ⃗
xT A⃗ ⃗T⃗
x+B ⃗ ⃗
e d x= e d x= e 2 B A B.
det A
• Common integrals in quantum field theory
• Normal distribution
3.4 Integrals of similar form
∫ • List of integrals of exponential functions
∞
2n − x
2 √ (2n − 1)!! 2n+1 √ (2n)! ( a )2n+1
x e a2 dx = π a = π
0 2n+1 n! 2 • Error function
∫ ∞
x2 n! 2n+2
x2n+1 e− a2 dx = a
2
0 5 References
∫ ∞
Γ( (n+1)
2 )
xn e−a x dx =
2
(n+1) [1] Пуассона интегралБСЭ
0 2a 2

∫ ∞
√ [2] http://www.york.ac.uk/depts/maths/histstat/normal_
(2n − 1)!! π
x2n e−ax dx =
2
history.pdf
0 an 2n+1 a
(n positive integer) [3] http://www.york.ac.uk/depts/maths/histstat/normal_
history.pdf
An easy way to derive these is by parameter differentia-
tion.
• Weisstein, Eric W., “Gaussian Integral”,
MathWorld.
∫ ∞ ∫ ∞ ∫ ∞ √
∂ n −αx2 n ∂•nDavid Griffiths. √
Introduction π (2n − 1)!!
∂ n − 1 Mechan-
to Quantum
x2n e−αx dx = (−1) −αx2
2 n n
e dx = (−1) e dx = π (−1) α 2 =

−∞ −∞ ∂αn ∂αnics.−∞
2nd Edition back cover. ∂αn α (2α)n
• Abramowitz, M. and Stegun, I. A. Handbook of
3.5 Higher-order polynomials Mathematical Functions, Dover Publications, Inc.
New York
Exponentials of other even polynomials can easily be
solved using series. For example the solution to the in-
tegral of the exponential of a quartic polynomial is
 5

6 Text and image sources, contributors, and licenses

6.1 Text
• Gaussian integral Source: http://en.wikipedia.org/wiki/Gaussian%20integral?oldid=642015744 Contributors: Michael Hardy, Cyp,
Charles Matthews, Phys, Aleph4, Fredrik, Benwing, Jleedev, Tobias Bergemann, Giftlite, Jason Quinn, CryptoDerk, Sreyan, LucasVB,
Abdull, TheObtuseAngleOfDoom, Skal, KittySaturn, Pt, Spoon!, Danski14, Mattpickman, LOL, Isnow, YurikBot, Lunch, SmackBot,
RDBury, Njerseyguy, Complexica, Nbarth, Duneatreides, Autopilot, Disavian, NNemec, Bilboq, Stikonas, Dicklyon, JacekW, JRSpriggs,
Markjoseph125, Pbond, NickFr, HenningThielemann, Mct mht, Scott.medling, ArgentTurquoise, Zeitnot, Vvidetta, MikeLynch, .ana-
condabot, DRHagen, EulerGamma, Seleucus, Drspintronic, Shoofle, Pagw, Treyd500, STBotD, Gogobera, Alexqc, Yakeyglee, Jluu, Yan-
nis1962, Tamorlan, SieBot, Gamesguru2, This, that and the other, ClueBot, PipepBot, Ideal gas equation, Alexbot, Estirabot, Qwfp,
Uribe27, Addbot, FerrousTigrus, Jasper Deng, Norman21, Vitrone20, Luckas-bot, Yobot, AnomieBOT, Void2258, Jim1138, Drilnoth,
Raffamaiden, Raulshc, Zaglabarg, FrescoBot, Weetoddid, Quaternion k, Prcnarhet arthas, Stpasha, Khinchin’s constant, Wisapi, AltaDe-
Saxa, Ibra86, ZéroBot, Bomazi, Junhogan, Stutts, Syed Ahsan Kamal, Sbacallado, Cowlicks, AHusain314, JohannesBuchner, Yardimsever,
EFZR090440, Feynmanliang, 28shivamsharma, BethNaught and Anonymous: 115

6.2 Images
• File:E\char"005E\relax{}(-x\char"005E\relax{}2).svg Source: http://upload.wikimedia.org/wikipedia/commons/2/2f/E%5E%28-x%
5E2%29.svg License: CC BY-SA 3.0 Contributors: Own work Original artist: Autopilot

6.3 Content license

• Creative Commons Attribution-Share Alike 3.0