Lambert W-Function Revisited: Applications in Science

Pushpa N. Rathie1, Prabhata K. Swamee2 and Luan Carlos de S. M. Ozelim3

1
Department of Statistics, University of Brasilia, Brasilia, 70910-900, Brazil,
[email protected] (corresponding author)
2
JM-46 Shankar Nagar, Bhopal, 462 016, Madhya Pradesh, India, [email protected]
(presenting author)
3
Department of Civil and Environmental Engineering, University of Brasilia, Brasilia 70910-
900, Brazil, [email protected]

Abstract: When complex problems arise in science, scientists know there is a good chance
that such complexity is not inherent to the problem itself, but to the way one is trying to solve
it. In other words, simplicity gives the most beautiful – and sometimes unique – solutions to
complex problems in each area of knowledge. Following this idea, Lambert W function is
defined as the solution of one of the most simple equations in Mathematics, which suggests
that it can only be applied to solve simple problems. It is shown that the latter assumption is
far from true by reviewing the application of Lambert W function to a wide class of
situations. A total of eighteen applications in Physics, Civil, Electrical, Control and
Networking Egineering are studied.

Keywords: Lambert W-function; Physics; Civil Engineering; Electrical Engineering; Control

Engineering; Network Engineering.

1. Introduction

Pareto´s principle states that, for the most diversified phenomena, about eighty percent of
the consequences is the result of only twenty percent of the causes. In fact, under the view of
the so-called exact sciences, this logic is easily applied. For example, consider the amount of
time needed to write a paper or a theory. A huge share of such time is spent in thinking the
bases of the new idea and in its description, while relatively less time is needed to textually or
mathematically describe what has come to mind.
More specifically, if the Pareto’s principle is reapplied to the mathematical effort needed
to develop the referred theory, it is expected that simple methodologies would be responsible
for a good part of the solution process. Once again, this is verified in a scientific paper.
This way, consider the following equation:

we w  x (1)

It is intuitive that simple functions might have simple functional equations. For example,
the power function can be easily represented as the solution of the functional equation:

f (a ) f (b)  f (a  b) . (2)

By analysing Eq.(1), it is natural to think that if there exists a function w which satisfies
such equation, this function should be as simple as Eq.(1).
Following the efforts by Euler and Lambert, in fact there is a function, normally named
Lambert W-function, which satisfies Eq.(1). Formally, Lambert W-function of a real variable
x can be defined as (Corless et al., 1996):
W ( x)eW ( x )  x, x  [1/ e, ) . (3)

It is worth noticing that for x  [1/ e, 0] , there are two possible branches of W(x), namely:
W0(x) stands for the branch in which W ( x)  1 , while W-1(x) stand for the branch in which
W ( x)  1 . For simplicity of notation, from now on Lambert W-function is going to be
referred as simply W function. Figure 1 shows the behavior of the function as well as its
branches. The latter are represented by a full line (W0(x)), and a dashed line (W-1(x)).

Figure 1. Lambert W-function (Corless et al., 1996)

The pioneer work by Corless et al. (1996) presented, for the first time, in a systematic way
the W-function to the scientific community. It is undeniable that such paper was and still is
the basis for the study of the W-function and can be considered the first review paper about
the topic.
In the present paper a few applications of W-function are discussed in order to show its
importance in science.

2. Examples of Application in Science

In this section a brief review of where Lambert W-function has been applied is shown. In
order to better structure the paper, the applications are grouped following the area they are
related to.

2.1 Physics

2.1.1 The Moyal Function

As shown in (Veberic, 2010), the Moyal function can be defined as:
  1 
M ( x)  exp   ( x  exp( x))  . (4)
 2 

By means of Lambert W-function it is possible rewrite Eq.(4) as:

M 1 ( x)  W0,1 ( x 2 )  2 ln( x ) . (5)

Moyal function is used for phenomenological recovery of saturated signals form the
photomultipliers.

2.1.2 The Gaisser-Hillas Function

Also in (Veberic, 2010), the Gaisser-Hillas function was defined as:

X max  X 0
 X  X0  
X X .
G ( X ; X 0 , X max ,  )    exp  max  (6)
 X max  X 0    

As pointed out by Veberic (2010), Gaisser-Hillas function is intimately related to the

Lambert W-function. In order to show this, consider the following variable changes:

X  X0 X  X0
x ; xmax  max . (7)
 

By means of Eqs.(6) and (7), it is easy to get:

xmax
 x 
g ( x; xmax )    exp  xmax  x  . (8)
 xmax 

In general, the problems related to Gaisser-Hillas function are the ones concerning its
inverse, i.e.:

g ( x; xmax )  a (9)

in which 0  a  1. In order to invert Eq.(9), consider the following rearrangement:

x  x  1/ xmax 1
 exp     a e . (10)
xmax  xmax 

By comparing Eq.(10) to Eq.(3), the two real solutions to the former equation are:


x1,2   xmaxW0,1 a1/ xmax e1 .  (11)

Note that the branch of W-function in Eq.(11) corresponds to choosing whether one is
dealing with values of x greater or lower than the maximum.
 2.1.3 Diode Ideality Factor
In (Bayhan and Kavasoglu, 2007), it has been established that the dark forward current of
a physical pn junction device can be expressed by the relation:

V  IRs   q (V  IRs )  
I  I 0  exp    1 , (12)
Rsh   nkT  

in which I and V are the terminal current and voltage, I0 is the saturation current, k is the
Boltzmann’s constant, q is the magnitude of the electrical charge on the electron, n is the
junction ideality factor and Rs and Rsh are the series and shunt resistances, respectively.
By means of W-function, Eq.(12) has been analytically solved for I in (Bayhan and
Kavasoglu, 2007) as:

kT  qRs Rsh I 0  qR (V  I 0 Rs )   V  I 0 Rsh

I n W0  exp  sh    . (13)
qRs  nkT ( Rs  Rsh )  nkT ( Rs  Rsh )   Rs  Rsh

2.1.4 The Forward Non-ideal Diode Equation with Series and Shunt Parasitic Resistances
In (Ortiz-Conde et al., 2000), the following equation has been used to describe a lumped
circuit model of a single exponential junction, with parasitic series, Rs, and parasitic parallel
resistances to represent shunt loss paths at the junction (Rp1) and shunt losses at the periphery
(Rp2):

    R   
  q  V 1  s   IRs   
V  IRs V VRs     R p 2   
I    I 0  exp    1
 . (14)
R p1 R p 2 R p1 R p 2   nkT  
   
   

The solution of Eq.(14) for both I and V can be given in terms the W-function as:

kT  qRs R p1I 0  qR (V  I 0 Rs )   V  I 0 R p1 V
I n W0  exp  p1  
qRs  nkT ( Rs  R p1 )  nkT ( Rs  Rp1 )   Rs  Rp1 R p 2 , (15)
  

Rp 2 nkT  q( R  R ) R I  q  IR p1 R p 2  I 0 R p1 ( Rs  R p 2 )   
W0  exp    
s p2 p1 0
V 
Rs  R p 2 q  nkT ( Rs  R p1  R p 2 )  nkT ( Rs  R p1  R p 2 ) 
  
. (16)
Rp 2  R p1R p 2  R p1 R p 2
I  Rs    I0
Rs  R p 2  Rs  R p1  R p 2  Rs  R p1  R p 2

It is worth noticing that the result (13) is a special case of the result (16) when Rp2  . A
few special cases of Eq.(16) have been dicussed in (Ortiz-Conde et al., 2000).

2.1.5 The Two-loop invariant (running) coupling of QCD

In (Magradze, 2000) the following equation related to the running coupling of massless
QCD has been introduced:
 1  1 
(2)
 b ln  1  (2)   ln( x) , (17)
a ( x)  ba ( x) 

in which b is a constant and a(2)(x) is the function of interest. The solution given in
(Magradze, 2000) is:

1
1  x 1/b  
a ( x )   1  W  
(2)
 . (18)
b  e 

The branches of the W-function in Eq.(18) which suit the conditions of interest are further
investigated by Magradze (2000). For example, when only real positive values of x are taken
into account, the correspond branch is -1.

2.1.6 New Analytic Running Coupling (NARC) in QCD

In (Nesterenko, 2001), it has been shown that in order to represent the NARC in a
renorminvariant form and to derive the β function related to the problem in study, one has to
solve the following equation:

z 1
 a, z  0, a 0. (19)
z ln( z )

By means of manipulating Eq.(19), the solution with respect to z is given as:

1
1   1  1a  
z   W  e   . (20)
a   a 


The branches of the W-function in Eq.(19) are carefully analyzed by Nesterenko (2001).

2.1.7 Projectile Launching in a Linear Resisting Medium

Consider an object of mass m launched with initial speed u0 form the origin at an angle of
α (0 < α < π/2 ) over the ground and in a gravitational field g which is constant, in a linear
resisting medium where the retarding force acting upon the object is considered to be
proportional to its velocity. The equations of motion of such an object are (Stewart, 2011):

u0 cos 
x(t ) 

 1  exp   t   , (21)

 g u sin   gt
y (t )   2  0   1  exp   t    , (22)
   

in which t stands for the time past since the launch, and γ is a drag coefficient.
In order to find the time of flight T for the projectile, one has to set y(T) = 0, this way, Eq.
(22) turns to:
u    1  eu  , (23)

in which u = γT and ζ = 1+γv0sinα/g. By means of the W-function, Eq.(23) can be solved for
u and ultimately for T as:

v0 sin     W0 ( e  ) 
T  . (24)
g   1 

By means of Eq.(24), the range R of the projectile can be given by combining Eqs.(21)
and (24) as:

v02 sin 2    W0 ( e  ) 

R  . (25)
2 g   (  1) 

2.1.8 The Parker Solar Wind Problem

It has been derived by Cranmer (2005) that, by means of the mass and momentum
equations of conservation, the velocity of the solar wind u is related to the radial distance r by
means of the following equation:

 u2   u2   r   rc 
 2  1  ln  2   4 ln    4   1  , (26)
 a1   a1   rc   r 

in which a1 and rc are constants.

By means of the W-function, Eq.(26) can be readily solved for u as:

   r 4  
 a1 W0     exp  4 1  c   1  , if r  rc
2 r
  r   r  
  c 
u 
2
. (27)
 2  r 4
  rc   
 a1 W1     exp  4  1    1  , if r  rc
   rc    r  
 

It is worth noticing that the opposite choices must be made to obtain the Bondi accretion
solution.

2.1.9 The Mass Flux Problem

Also in (Cranmer, 2005) the following relation has been derived for the mass flux problem
of solar wind:

n1 AT 2  5ukFc 
Fc  5n1ukT  ln 1  2 , (28)
5uk  n1 AT 

in which n1 is a number density (number of particles per unit of volume); Fc is the heat
conductive flux, κ is the Spitzer-Härm heat conductivity in an ionized plasma, k is the
Boltzmann constant, T is the temperature, A is a constant, and u is the velocity.
The W-function provides the solution of Eq.(28) for n1 as:
 5u0 kFc    25u02 k 2   
n0  1  W1   exp     1  (29)
 AT 2     AT   

in which the subindex 0 stands for the values evaluated at the coronal base.

2.1.10 The Tonks Gas

Caillol (2003), while studying Tonks gases, obtained the following relation which relates
the pressure χ to the length σ and the activity z:

 e   z . (30)

It is clear from the definition of the W-function that Eq.(30) can be solved for χ as:

1
 W0 ( z ) . (31)


2.1.11 The Weak and Long Range Repulsion

Still in (Caillol, 2003), the relaton between the desity  of a gas of particles interacting via
a given pair potential and the activity is shown to be:

 e  a2  z , (32)

in which a2 is a constant.
From the definition of the W-function it is clear that Eq.(32) can be solved for  as:

1
 W0 (a2 z ) , (33)
a2

2.2 Civil Engineering

2.2.1 Green-Ampt Model of Infiltration

Parlange et al. (2002) considered the equation which relates the nondimensional
cumulative infiltration I * to the nondimensional time t * for a Green-Ampt soil as:

t*  I*  ln  1  I*  . (34)

It is possible to rearrange Eq.(34) in order to explicitly solve it for I*. Such solution is
readily obtained by means of Lambert W-fucntion as:

 
I*  1  W1  exp    1  t*   . (35)

2.2.2 Talsma-Parlange Model of Infiltration

Parlange et al. (2002) also considered the equation which relates the nondimensional
cumulative infiltration I * to the nondimensional time t * for a Talsma-Parlange soil as:
t*  I*  exp   I*   1 . (36)

Equation (36) is easily solved for I* by means of the W-function as:

I*  W0   exp(t*  1)   t*  1 . (37)

2.3 Electrical and Control Engineering

2.3.1 Control Process With Time Delay

In a recent paper, Cogan & Paor (2011), while studying control processes with time delay,
considered the following equation:

( s  1 ) 2  k1e  sL  0 , (38)

in which s is the variable of interest and α1, k1 and L are constants. The W-function gives the
solutions of Eq.(38) as:

2 L 1L

s W k1 e 2  1  . (39)
L 2 

Branch issues are further discussed in (Cogan &Paor, 2011).

2.3.2 Continuous Linear Time Invariant Homogeneous Time Delay System

Let one consider the continuous linear time invariant (LTI) homogeneous time delay
system (TDS) (Duan et al., 2011):

dx(t )
 A* x(t )  Ad x(t  h)  0, t 0
dt , (40)
x(0)  x0 , x(t )  g (t ) for t  [h, 0)

in which A* and Ad are m x m coefficient matrices, x(t) is an m x 1 state vector, g(t) is a m x

1 preshape function, t is time and h is a constant scalar time delay.
It is known that the solution to Eq.(40) is given as (Yi et al., 2006):


x(t )  e
k 
Sk t
CkI , (41)

in which:

1
S k  Wk ( Ad hQk )  A* (42)
h

and Qk is solved by the following condition:

Wk ( Ad hQk ) exp A* expWk ( Ad hQk )  A* h     Ad h .

(43)
 It is worth noticing that Sk and Qk are m x m matrices, CkI are m x 1 vectors and are
determined by the preshape function g(t) and x0. In this paper, Wk stands for a matrix W-
function and k for the branch considered and the index of summation.
A few applications of Eqs. (41) to (43) are shown in (Duan et al., 2011).

2.4 Network Engineering

2.4.1 Constrained Resource Allocation in Cooperative Networks

Brah et al. (2011), while considering constrained resource allocation in cooperative
networks, studied the following equation:

k
   n 
ln 1  k   k   0, (44)
 nk  1   k B
(1  k )
nk ln(2)
in which αk, λk and B are constants and nk is the variable of interest. By means of W-function,
Eq.(44) can be solve for nk as:

  
   
 kW   exp  1  
 B


 (1  k )  
  ln(2) 
nk*  , (45)
  
   

1  W  exp  1  
 B


 (1  k )  
  ln(2) 

in which the superscript * denotes the optimal value of nk. Further application of Eq.(45) is
described in Brah et al. (2011).

2.4.2 Sizing Peer-to-Peer Networks

In another application in Network Engineering, Bustos-Jimenez et al. (2008) while
considering the size of peer-to-peer networks, studied the solutions of the following equation:

j  a3 ln( j )  b , (46)

in which a3 and b are constants and j is the variable of interest.

It follows from the definition of the W-function that the solutions to Eq.(46) are given by:

 eb / a3 
j  a3W   . (47)
 a3 

W-function’s branch of interest is discussed in Bustos-Jimenez et al. (2008).

2.4.3 Average Probability of Error over a Log-Normal Fading Channel

 While considering wireless communications, it is common to use the log-normal
probability distribution to model the shadowing and the small scale fading for indoor ultra
wideband communications. This way, Héliot et al. (2008) while studying the approximation
of the average probability of error over a log-normal fading channel came across the
following equation:

vi xˆ f  xˆ f 
2
 exp  (48)
2k k  k 

in which k, vi are constants and xˆ f is the variable of interest.

In order to solve Eq.(48), from the definition of the W-function it is easy to get:

 v 
xˆ f  kW0  i 2  . (49)
 2k 

Both Eqs.(48) and (49) are related to the maximum of a function of interest f(x). The paper
by Héliot et al. (2008) proposes a function g(x) which approximates f(x). While studying the
maximum of g(x), the following equation arose:

 v 
xˆ g  k ln  2 i *  , (50)
 k a0 

in which a0* is a constant.

By setting the maxima of both f(x) and g(x) to occour at the same point, one shall equal
Eq.(49) to Eq.(50), which results in:

 v 
xˆ f  k ln  2 i *  . (51)
 k a0 

By means of Eqs.(51) and (49), the coefficient a0* can be given as:

 v 
a0*  2 xˆ f  2kW0  i 2  . (52)
 2k 

The same procedure has been applied to obtain other coefficients of interest in (Héliot et
al., 2008).

2.5 Statistics, Information Theory and Reliability Theory

Recently, Rathie (2008) and Rathie and Silva (2011) used the W-function to solve a large
variety of problems concerning information theory, reliability analysis and statistical
distributions. Due to the compilation characteristic of both papers, instead of transcribing
such applications, the authors advise the interested reader to take a brief look at them.

5. Conclusions
 It has been shown that despite the apparent simplicity concerning the definition of
Lambert W-function, it can be applied to highly complex problems in a wide class of areas of
Science. A total of eighteen applications have been discussed and shown to brilliantly solve
once thought to be unsolvable equations. Every result is relevant in the development of the
respective field of knowledge, reinforcing the importance of Lambert W-function in the
scientific community.

References
Bayhan, H. and Kavasoglu, A.S. (2007) Exact analytical solution of the diode ideality
factor of a pn junction device using Lambert W-function model. Turkish Journal of Physics,
vol. 31, Issue 1, p.7-10.
Brah, F.; Zaidi, A.; Louveaux, J. and Vandendorpe, L. (2011) On the Lambert-W function
for constrained resource allocation in cooperative networks. EURASIP Journal on Wireless
Comunications and Networking, 19.
Bustos-Jimenez, J.; Bersano, N.; Schaeffer, S.E.; Piquer, J.M.; Iosup, A. and Ciuffoletti,
A. (2008) Estimating The Size Of Peer-To-Peer Networks Using Lambert's W Function, Grid
Computing, p.61-72, doi: 10.1007/978-0-387-09457-1_6.
Caillol, J-M. (2003) Some applications of the Lambert W function to classical statistical
mechanics, J. Phys. A: Math. Gen., 36, 10431, doi:10.1088/0305-4470/36/42/001.
Cogan, B. and Paor, A.M. (2011) Analytic root locus and Lambert W function in control
of a process with time delay. Journal of Electrical Engineering, vol. 62, no. 6, p. 327-334.
Corless, R.M.; Gonnet, G.H.; Hare, D.E.G.; Jeffrey, D.J. and Knuth, D.E. (1996) On the
Lambert W function. Advances in Computational Mathematics, 5, p.329–359.
doi:10.1007/BF02124750.
Cranmer, S.R. (2005) New views of the solar wind with the Lambert W function, Am. J.
Phys., Vol. 72, No. 11, p. 1397-1403.
Duan, S.; Ni, J. and Ulsoy, A.G. (2011) Decay function estimation for linear time delay
systems via the Lambert W function. Journal of Vibration and Control, online 18 october
2011, doi:10.1177/1077546311408168.
Héliot, F.; Ghavami, M. and Nakhai, M.R. (2008) An accurate closed-form approximation
of the average probability of error over a log-normal fading channel, IEEE Transactions on
Wireless Communications, vol. 7, no. 5, p. 1495-1500.
Magradze, B.A. (2000) An Analytic Approach to Perturbative QCD, Int. J. Mod. Phys. A.,
15, p.2715-2734.
Nesterenko, A.V. (2001) Investigation of a new analytic running coupling in QCD,
arXiv:hep-ph/0102203v1.
Ortiz-Conde, A; Sánchez, F.J.G. and Muci, J. (2000) Exact analytical solutions of the
forward non-ideal diode equation with series and shunt parasitic resistances. Solid-State
Electronics, 44, p. 1861-1864.
Parlange, J-Y.; Barry, D.A. and Harverkamp, R. (2002). Explicit infiltration equations and
the Lambert W-function. Advances in Water Resources, 25, p. 1119-1124.
Rathie, P.N.(2008) Lambert W-function in statistical distributions and reliability analysis,
Proceedings of the International Conference on Engineering Optimization, Rio de Janeiro,
Brazil, June 1-5, 30-36.
Rathie, P.N. and Silva, P.H.D (2011) Applications of Lambert W Function, International
Journal of Applied Mathematics & Statistics, vol. 23, no. D11, 1-15.
Stewart, S.M. (2011) Some remarks on the time of flight and range of a projectile in a
linear resisting medium. Journal of Engineering Science and Technology Review, 4 (1), p.
32-34.
 Veberic, D. (2010) Having Fun with Lambert W(x) Function, arXiv:1003.1628v1.
Yi, S.; Ulsoy, A.G. and Nelson, P.W. (2006) Solution of systems of linear delay
differential equations via Laplace transformation. In Proceedings of the 45th IEEE
Conference on Decision and Control, San Diego, CA, p. 2535-2540.