189

CHAPTER 13

Laplace Distribution
13.1 INTRODUCTION
This distribution, introduced by Pierre–Simon Laplace (1749–1827) in 1774, has many applications
in geology, hydrology and ocean engineering, economics and finance, quality control, image and
speech recognition, inventory control (especially of slow moving items), scientometrics, pharma-
cokinetics (drug absorption, delivery, depletion from the body), nuclear medicine, telecommunica-
tions, and error modeling (called Laplacian noise) in various fields like astronomy, global navigation
systems, etc. Also called the first law of Laplace, two-tailed exponential, bilateral exponential distribu-
tion, or double-exponential (DE) distribution (because each part on the left and right of the location
parameter is proportional to an exponential density or its mirror image), it is also used in life testing
models with heavy tails, and in machine learning (natural language processing, computer vision,
and intrusion detection). The PDF of classical Laplace distribution (CLD) with two-parameters is
given by f .xI a; b/ D

.1=.2b// exp. jx aj=b/; D .0:50=b/ exp. jx aj=b/; 1 < a; x < 1; b > 0; (13.1)

where “a” is the location parameter (mean) and “b” is the scale parameter (also called diversity). It
is denoted as ₤(a; b), Lp.a; b/, or Laplace(a; b). This can also be written as

1 exp. .a x/=b/ if x < aI
f .xI a; b/ D
2b exp. .x a/=b/ otherwise:

It is symmetric around a, and the right part resembles an exponential or Pareto distribution.
The standard Laplace distribution (SLD) is obtained by putting a D 0; b D 1, with PDF f .z/ D
.1=2/e jzj ; and the zero-mean Laplace distribution by putting a D 0 with PDF f .xI 0; b/ D
.1=2b/ exp. jxj=b/. Apone-parameter (location) unit variance Laplace distribution has PDF
f .xI a; 1/ D p1 exp. 2jx aj/. As the exponent involves absolute values, it is appropri-
2
ate in the study of first order phenomena like low-dose response modeling in medical sciences
(where the log-Laplace distribution discussed below is preferred). Put 1=b D 2c to get another
form f .xI a; c/ D c exp. 2cjxp aj/, 1=b Dpc to get f .xI a; c/ D .c=2/ exp. cjx aj/, and
p
2b D c to get f .xI a; c/ D 1=. 2c/ exp. 2jx aj=c/ (which is the preferred form in astro-
nomical sciences).

R. Chattamvelli et al., Continuous Distributions in Engineering and the Applied Sciences – Part II
© Springer Nature Switzerland AG 2021
190 13. LAPLACE DISTRIBUTION
13.2 RELATION TO OTHER DISTRIBUTIONS
If X  Laplace(a; b), then a change of origin and scale transformation cX C d results in
Laplace(ac C d; bc). The transformation Y D jX aj=b results in the standard exponential distri-
bution. Similarly, Z D .2=b/jX aj has a 2 distribution with 2 DoF. The standard Laplace distri-
bution is the distribution of the difference of two IID standard exponential variates. In general, if Xk ,
k D 1; 2 are two independent EXP() variates, then Y D X1 X2 has a Laplace (0,) distribution.
Conversely, if X  Laplace(0; ) then Y D jXj  EXP(1=). If Xk for k D 1; 2; : : : ; n are IID
P
Laplace(a; b) random variables, Z D .2=b/ nkD1 jXk aj has a 22n distribution with 2n DoF.
From this it follows that the ratio of two IID Laplace variates Laplace(0; b1 ) and Laplace(0; b2 ) is
a central F(2; 2) variate. If X  Laplace(0; b) then Y D p jXj has an EXP(1=b) distribution. If X 
EXP(1) and Z  N(0; 1) are IID, then Y D a C bZ 2X is distributed as Laplace(a; b). If X 
N(0; 1) is independent of Y which is distributed as Rayleigh with PDF f .y/ D y exp. y 2 =2/,
then W D X Y is an SLD. If U1 ; U2 are IID standard uniform (U.0; 1/) random variables, then
Y D ln.U1 =U2 / is an SLD. This property can be used to generate random numbers from Laplace
distribution as U and 1 U are identically distributed. Laplace and Cauchy distributions are re-
lated through characteristic functions, as shown below. If Z1 ; Z2 ; Z3 ; Z4 are IID N(0; 1) random
variables, Z1 Z2 ˙ Z3 Z4 is Laplace(0; 2). The asymmetric Laplace distribution has PDF

1 b exp. cjx aj=b/ for x < aI
f .xI a; b; c/ D
c1Cb 2 exp. bcjx aj/ for x  a:

( Jones (2002, 2019) [106, 110]). The ratio of two IID classical Laplace distributions is called the
double Lomax distribution (DLD). DLD can be obtained by compounding CLD with an exponen-
tial density. Discrete version of SLD has been obtained by Kozubowski and Inusah (2006) [120]
with PMF

f .xI p/ D .1 p/=.1 C p/p jkj ; k D 0; 1; 2; : : : I 0 < p < 1; (13.2)

which arises as the difference of two IID geometric random variables.

13.3 PROPERTIES OF LAPLACE DISTRIBUTION

fx .a; / D .=2/ exp. jx aj/; 1 < x < 1; 1 < a < 1;  > 0: (13.3)

Take log to get ln.f .x// D C jx aj, which is a concave function in x, although the derivative
is nonexistent at x D a.
Due to symmetry, ˇ1 and all odd moments except the first are zeros. The mean, median,
and mode are  D a, and the variance is  2 D 2b 2 . The modal value is 1=.2b/, showing that the