Exact Solutions of Nonlinear Partial Differential Equations

The Tanh/Sech Method

Willy Hereman
Department of Mathematical and Computer Sciences
Colorado School of Mines
Golden, CO 80401-1887
http://www.mines.edu/fs home/whereman/
whereman@mines.edu

Mathematica Visiting Scholar Grant Program

Wolfram Research Inc., Champaign, Illinois
October 25November 11, 2000

Collaborators: Unal
Goktas (WRI), Doug Baldwin (CSM)
Ryan Martino, Joel Miller, Linda Hong (REUNSF 99)
Steve Formenac, Andrew Menz (REUNSF 00)
Research supported in part by NSF under Grant CCR-9901929
1

OUTLINE
Purpose & Motivation
Typical Examples
Algorithm for Tanh Solutions
Algorithm for Sech Solutions
Extension: Tanh Solutions for Differential-difference Equations (DDEs)
Solving/Analyzing Systems of Algebraic Equations with Parameters
Implementation Issues Mathematica Package
Future Work

Purpose & Motivation

Develop and implement symbolic algorithms to compute exact solutions of nonlinear (systems) of partial differential equations (PDEs)
and differential-difference equations (DDEs, lattices).
Solutions of tanh or sech type model solitary waves in fluid dynamics, plasmas, electrical circuits, optical fibers, bio-genetics, etc.
Class of nonlinear PDEs and DDEs solvable with the tanh/sech
method includes famous evolution and wave equations.
Typical examples: Korteweg-de Vries, Fisher and Boussinesq PDEs,
Toda and Volterra lattices (DDEs).
Research aspect: Design a high-quality application package for
the computation of exact solitary wave solutions of large classes of
nonlinear evolution and wave equations.
Educational aspect: Software as course ware for courses in nonlinear PDEs, theory of nonlinear waves, integrability, dynamical systems, and modeling with symbolic software.
Users: scientists working on nonlinear wave phenomena in fluid dynamics, nonlinear networks, elastic media, chemical kinetics, material science, bio-sciences, plasma physics, and nonlinear optics.

Typical Examples of Single PDEs and Systems of PDEs

The Korteweg-de Vries (KdV) equation:
ut + uux + u3x = 0.
Solitary wave solution:
8c31 c2 2c21
u(x, t) =

tanh2 [c1x + c2t + ] ,

6c1

or, equivalently,
4c31 + c2 2c21
sech2 [c1x + c2t + ] .
u(x, t) =
+
6c1

The modified Korteweg-de Vries (mKdV) equation:

ut + u2ux + u3x = 0.
Solitary wave solution:
v
u
u
u
t



6
3
c1 sech c1x c1t + .
u(x, t) =

The Fisher equation:

ut uxx u (1 u) = 0.
Solitary wave solution:
u(x, t) =
with

1 1
1
tanh + tanh2,
4 2
4

1
5
= x t + .
12
2 6
4

 The generalized Kuramoto-Sivashinski equation:

ut + uux + uxx + u3x + u4x = 0.
Solitary wave solutions:
For = 4 :
u(x, t) = 9 2c2 15 tanh (1 tanh tanh2),
with =

x
2

+ c2t + .

For = 0 :
v
u
u
u
t

u(x, t) = 2
with =

1
2

11
19

v
u
u
u
t

v
u
u
u
t

135 11
165 11
19
c2
tanh +
tanh3,
11
19 19
19 19

x + c2t + .

For = 12/ 47 :

47 (45 + 4418c2)
45
u(x, t) =
+ tanh
2209
47 47
45
15
tanh2 + tanh3,
+
47 47
47 47
with =

1
2 47

x + c2t + .

For = 16/ 73 :

2 73 (30 + 5329c2)
75
u(x, t) =
+ tanh
5329
73 73
60
15
tanh2 + tanh3,
+
73 73
73 73
with =

1
2 73

x + c2t + .

 Three-dimensional modified Korteweg-de Vries equation:

ut + 6u2ux + uxyz = 0.
Solitary wave solution:

u(x, y, z, t) = c2c3 sech [c1x + c2y + c3z c1c2c3t + ] .

The Boussinesq (wave) equation:
utt u2x + 3uu2x + 3ux2 + u4x = 0,
or written as a first-order system (v auxiliary variable):
ut + vx = 0,
vt + ux 3uux u3x = 0.
Solitary wave solution:
c21 c22 + 8c41
u(x, t) =
4c21 tanh2 [c1x + c2t + ]
2
3c1
v(x, t) = b0 + 4c1c2 tanh2 [c1x + c2t + ] .
The Broer-Kaup system:
uty + 2(uux)y + 2vxx uxxy = 0,
vt + 2(uv)x + vxx = 0.
Solitary wave solution:
u(x, t) =

c3
+ c1 tanh [c1x + c2y + c3t + ]
2c1

v(x, t) = c1c2 c1c2 tanh2 [c1x + c2y + c3t + ]

Typical Examples of DDEs (lattices)

The Toda lattice:
un = (1 + u n) (un1 2un + un+1) .
Solitary wave solution:
un(t) = a0 sinh(c1) tanh [c1n sinh(c1) t + ] .
The Volterra lattice:
u n = un(vn vn1)
v n = vn(un+1 un).
Solitary wave solution:
un(t) = c2 coth(c1) + c2 tanh [c1n + c2t + ]
vn(t) = c2 coth(c1) c2 tanh [c1n + c2t + ] .
The Relativistic Toda lattice:
u n = (1 + un)(vn vn1)
v n = vn(un+1 un + vn+1 vn1).
Solitary wave solution:
un(t) = c2 coth(c1)
vn(t) =

1
+ c2 tanh [c1n + c2t + ]

c2 coth(c1) c2
tanh [c1n + c2t + ] .

Algorithm for Tanh Solutions for PDE system

Given is a system of PDEs of order n
(u(x), u0(x), u00(x), u(n)(x)) = 0.
Dependent variable u has components ui (or u, v, w, ...)
Independent variable x has components xi (or x, y, z, t)
Step 1:
Seek solution of the form ui(x) = Ui(T ), with
T = tanh [c1x + c2y + c3z + c4t] = tanh .
Observe cosh2 sinh2 = 1, ( tanh)0 = 1 tanh2 or T 0 = 1T 2.
Repeatedly apply the operator rule

d
ci(1 T 2)
xi
dT
This produces a coupled system of Legendre equations of type
(n)

P(T, Ui, Ui0, . . . , Ui ) = 0

for Ui(T ).
Example: For Boussinesq system
ut + vx = 0
vt + ux 3uux u3x = 0,
we obtain after cancelling common factors 1 T 2
c2 U 0 + c1 V 0 = 0
c2V 0 + c1U 0 3c1U U 0


+c31 2(1 3T 2)U 0 + 6T (1 T 2)U 00 (1 T 2)2U 000 = 0
8

Step 2:
Seek polynomial solutions
Ui(T ) =

M
Xi
j=0

aij T j

Balance the highest power terms in T to determine Mi.

Example: Powers for Boussinesq system
M1 1 = M2 1,

2M1 1 = M1 + 1

gives M1 = M2 = 2.
Hence, U1(T ) = a10 + a11T + a12T 2,

U2(T ) = a20 + a21T + a22T 2.

Step 3:
Determine the algebraic system for the unknown coefficients aij by
balancing the coefficients of the various powers of T.
Example: Boussinesq system
a11 c1 (3a12 + 2 c21) = 0
a12 c1 (a12 + 4 c21) = 0
a21 c1 + a11 c2 = 0
a22 c1 + a12 c2 = 0
a11 c1 3a10 a11 c1 + 2a11 c31 + a21 c2 = 0
3a211 c1 + 2 a12 c1 6a10 a12 c1 + 16 a12 c31 + 2a22 c2 = 0.

Step 4:
Solve the nonlinear algebraic system with parameters.
Reject complex solutions? Test the solutions.
Example: Solution for Boussinesq case
a10 =
a11
a12
a20
a21
a22

=
=
=
=
=

c21 c22 + 8c41

3c21
0
4c21
free
0
4c1c2.

Step 5:
Return to the original variables.
Test the final solution in the original equations
Example: Solitary wave solution for Boussinesq system:
c21 c22 + 8c41
2
2
[c1x + c2t + ]
tanh

4c
u(x, t) =
1
3c21
v(x, t) = a20 + 4c1c2 tanh2 [c1x + c2t + ] .

10

Algorithm for Sech Solutions for PDE system

Given is a system of PDEs of order n
(u(x), u0(x), u00(x), u(n)(x)) = 0.
Dependent variable u has components ui (or u, v, w, ...)
Independent variable x has components xi (or x, y, z, t)
Step 1:
Seek solution of the form ui(x) = Ui(S), with
S = sech [c1x + c2y + c3z + c4t] = tanh .

Observe ( sech ) = tanh sech or S = T S = 1 S 2 S.

0

Also, cosh2 sinh2 = 1, hence, T 2 + S 2 = 1 and

dT
dS

= TS .

Repeatedly apply the operator rule

d
ci 1 S 2S
xi
dS
This produces a coupled system of Legendre type equations of type

(m)
(n)
0
P(S, Ui, Ui , . . . , Ui ) + 1 S 2 Q(S, Ui, Ui0, . . . , Ui ) = 0
for Ui(S).
For every equation one must have
Pi = 0 or Qi = 0. Only odd
derivatives produce the extra factor 1 S 2.
Conclusion: The total number of derivatives in each term in the
given system should be either even or odd. No mismatch is allowed.
Example: For the 3D mKdV equation
ut + 6u2ux + uxyz = 0.
11


we obtain after cancelling a common factor 1 S 2 S
c4U 0+6c1U 2U 0+c1c2c3[(16S 2)U 0+3S(12S 2)U 00+S 2(1S 2)U 000] = 0
Step 2:
Seek polynomial solutions
Ui(S) =

M
Xi
j=0

aij S j

Balance the highest power terms in S to determine Mi.

Example: Powers for the 3D mKdV case
3M1 1 = M1 + 1
gives M1 = 1. Hence, U (S) = a10 + a11S.
Step 3:
Determine the algebraic system for the unknown coefficients aij by
balancing the coefficients of the various powers of S.
Example: System for 3D mKdV case
a11c1 (a211 c2 c3) = 0
a11 (6a210 c1 + c1 c2 c3 + c4) = 0
a10 a211 c1 == 0

12

Step 4:
Solve the nonlinear algebraic system with parameters.
Reject complex solutions? Test the solutions.
Example: Solution for 3D mKdV case
a10 = 0

a11 = c1 c3
c4 = c1 c2 c3
Step 5:
Return to the original variables.
Test the final solution in the original equations
Example: Solitary wave solution for the 3D mKdV equation

u(x, y, z, t) = c2 c3 sech(c1 x + c2 y + c3 z c1c2c3 t).

13

Extension: Tanh Solutions for DDE system

Given is a system of differential-difference equations (DDEs) of order n
(..., un1, un, un+1, ..., u n, ..., u(m)
n , ...) = 0.
Dependent variable un has components ui,n (or un, vn, wn, ...)
Independent variable x has components xi (or n, t).
No derivatives on shifted variables are allowed!
Step 1:
Seek solution of the form ui,n(x) = Ui,n(T (n)), with
T (n) = tanh [c1n + c2t + ] = tanh .
Note that the argument T depends on n. Complicates matters.
Repeatedly apply the operator rule on ui,n
d

c2(1 T 2)
t
dT
This produces a coupled system of Legendre equations of type
0
, . . .) = 0
P(T, Ui,n, Ui,n

for Ui,n(T ).
Example: Toda lattice
un = (1 + u n) (un1 2un + un+1) .
transforms into
c22(1


T )

2T Un0
2

(1 T


)Un00

+ 1 + c2(1 T )Un0 [Un1 2Un + Un+1] = 0

14

Step 2:
Seek polynomial solutions
Ui,n(T (n)) =

M
Xi
j=0

aij T (n)j

For Un+p, p 6= 0, there is a phase shift:

T (n) tanh(pc1)

Ui,np (T (n p) =
ai,j [T (n + p)]j =
ai,j
1

T
(n)
tanh(pc
)
j=0
j=0
1
M
Xi

M
Xi

Balance the highest power terms in T (n) to determine Mi.

Example: Powers for Toda lattice
2M1 1 = M1 + 1
gives M1 = 1.
Hence,
Un(T (n)) = a10 + a11T (n)
T (n) tanh(c1)
1 T (n) tanh(c1)
T (n) + tanh(c1)
Un+1(T (n + 1)) = a10 + a11T (n + 1) = a10 + a11
.
1 + T (n) tanh(c1)

Un1(T (n 1)) = a10 + a11T (n 1) = a10 + a11

Step 3:
Determine the algebraic system for the unknown coefficients aij by
balancing the coefficients of the various powers of T (n).
Example: Algebraic system for Toda lattice
c22 tanh2(c1) a11c2 tanh2(c1) = 0,
15

c2 a11 = 0

Step 4:
Solve the nonlinear algebraic system with parameters.
Reject complex solutions? Test the solutions.
Example: Solution of algebraic system for Toda lattice
a10 = free,

a11 = c2 = sinh(c1)

Step 5:
Return to the original variables.
Test the final solution in the original equations
Example: Solitary wave solution for Toda lattice:
un(t) = a0 sinh(c1) tanh [c1n sinh(c1) t + ] .

16

Example: System of DDEs: Relativistic Toda lattice

u n = (1 + un)(vn vn1)
v n = vn(un+1 un + vn+1 vn1).
Step 1: Change of variables
un(x, t) = Un(T (n)), vn(x, t) = Vn(T (n)),
with
T (n) = tanh [c1n + c2t + ] = tanh .
gives
c2(1 T 2)Un0 (1 + Un)(Vn Vn1) = 0
c2(1 T 2)Vn0 Vn(Un+1 Un + Vn+1 Vn1) = 0.
Step 2: Seek polynomial solutions
Un(T (n)) =
Vn(T (n)) =

M
X1
j=0
M
X2
j=0

a1j T (n)j

a2j T (n)j .

Balance the highest power terms in T (n) to determine M1, and M2 :

M1 + 1 = M1 + M2 ,

M2 + 1 = M 1 + M 2

gives M1 = M2 = 1.
Hence,
Un = a10 + a11T (n),

Vn = a20 + a21T (n).

17

Step 3: Algebraic system for aij :

a11 c2 + a21 tanh(c1) + a10 a21 tanh(c1) = 0
a11 tanh(c1) ( a21 + c2) = 0
a21 c2 + a11 a20 tanh(c1) + 2 a20 a21 tanh(c1) = 0
tanh(c1) (a11 a21 + 2 a221 a11 a20 tanh(c1)) = 0
a21 tanh2(c1) (c2 a11) = 0
Step 4: Solution of the algebraic system
c2 coth(c1)
c2
1
, a21 = .
a10 = c2 coth(c1) , a11 = c2, a20 =

Step 5: Solitary wave solution in original variables:

un(t) = c2 coth(c1)
vn(t) =

1
+ c2 tanh [c1n + c2t + ]

c2 coth(c1) c2
tanh [c1n + c2t + ] .

18

Solving/Analyzing Systems of Algebraic Equations with Parameters

Class of fifth-order evolution equations with parameters:
ut + 2u2ux + uxu2x + uu3x + u5x = 0.
Well-Known Special cases
Lax case: =

3
10 ,

= 2, = 10. Two solutions:

u(x, t) =

4c21

6c21

tanh c1x

56c51t

2(15a20c1

40a0c31

and
u(x, t) = a0

2c21

tanh c1x

28c51)t

where a0 is arbitrary.
Sawada-Kotera case: = 51 , = 1, = 5. Two solutions:
u(x, t) =

8c21

12c21

tanh c1x

16c51t

and
u(x, t) = a0

6c21

tanh c1x

(5a20c1

40a0c31

76c51)t

where a0 is arbitrary.
Kaup-Kupershmidt case: = 15 , = 52 , = 10. Two solutions:


3 2
2
2
5
u(x, t) = c1 c1 tanh c1x c1t +
2
and


2
2
2
5
u(x, t) = 8c1 12c tanh c1x 176c1t + ,
no free constants!
Ito case: = 92 , = 2, = 3. One solution:
u(x, t) =

20c21

30c21

tanh c1x
19

96c51t

+ .

What about the General case?

Q1: Can we retrieve the special solutions?
Q2: What are the condition(s) on the parameters , , for solutions
of tanh-type to exist?
Tanh solutions:
u(x, t) = a0 + a1 tanh [c1x + c2t + ] + a2 tanh2 [c1x + c2t + ] .
Nonlinear algebraic system must be analyzed, solved (or reduced!):
a1( 2a22 + 6a2c21 + 2a2c21 + 24c41) = 0
a1( 2a21 + 6 2a0a2 + 6a0c21 18a2c21 12a2c21 120c41) = 0
2a22 + 12a2c21 + 6a2c21 + 360c41 = 0
2 2a21a2 + 2 2a0a22 + 3a21c21 + a21c21 + 12a0a2c21
8a22c21 8a22c21 480a2c41 = 0
a1( 2a20c1 2a0c31 + 2a2c31 + 16c51 + c2) = 0
2a0a21c1 + 2a20a2c1 a21c31 a21c31 8a0a2c31 + 2a22c31
+136a2c51 + a2c2 = 0
Unknowns: a0, a1, a2.
Parameters: c1, c2, , , .
Solve or Reduce should be used on pieces, not on whole system.
20

Strategy to Solve/Reduce Nonlinear Systems

Assumptions:
All ci 6= 0
Parameters (, , , ...) are nonzero. Otherwise the highest powers
Mi may change.
All aj Mi 6= 0. Coefficients of highest power in Ui are present.
Solve for aij , then ci, then find conditions on parameters.
Strategy followed by hand:
Solve all linear equations in aij first (cost: branching). Start with
the ones without parameters. Capture constraints in the process.
Solve linear equations in ci if they are free of aij .
Solve linear equations in parameters if they free of aij , ci.
Solve quasi-linear equations for aij , ci, parameters.
Solve quadratic equations for aij , ci, parameters.
Eliminate cubic terms for aij , ci, parameters, without solving.
Show remaining equations, if any.
Alternatives:
Use (adapted) Grobner Basis Techniques.
Use combinatorics on coefficients aij = 0 or aij 6= 0.

21

Actual Solution: Two major cases:

CASE 1: a1 = 0, two subcases
Subcase 1-a:
3
a2 = a0
2
c2 = c31(24c21 a0)
where a0 is one of the two roots of the quadratic equation:
2a20 8a0c21 4a0c21 + 160c41 = 0.
Subcase 1-b: If = 10 1, then
a2 =

6 2
c
1

1
c2 = (2 2a20c1 8a0c31 + 12c51 + 16c51)

where a0 is arbitrary.
CASE 2: a1 6= 0, then
=

1
(39 + 38 + 8 2)
392

and
a2 =

168
c21
(3 + 2)

provided is one of the roots of

(104 2 + 886 + 1487)(520 3 + 2158 2 1103 8871) = 0

22

1
Subcase 2-a: If 2 = 104
(886 + 1487), then

2 + 5
26

49c21(9983 + 4378)
a0 =
26(8 + 3)(3 + 2)2
336c21
a1 =
(3 + 2)
168c21
a2 =
(3 + 2)
364 c51 (3851 + 1634)
c2 =
.
6715 + 2946
Subcase 2-b: If 3 =

1
520 (8871

+ 1103 2158 2), then

39 + 38 + 8 2
=
392
28 c21 (6483 + 5529 + 1066 2)
a0 =
(3 + 2)(23 + 6)(81 + 26)
a21

28224 c41 (4 1)(26 17)

=
(3 + 2)2(23 + 6)(81 + 26) 2

168c21
a2 =
(3 + 2)
8 c51 (1792261977 + 1161063881 + 188900114 2)
c2 =
.
959833473 + 632954969 + 105176786 2

23

Implementation Issues Mathematica Package

Demonstration of prototype package
Demonstration of current package without solver
Future Work
Look at other types of explicit solutions involving
hyperbolic functions sinh, cosh, tanh, ...
other special functions.
complex exponentials combined with sech or tanh.
Example: Set of ODEs from quantum field theory
uxx = u + u3 + auv 2
vxx = bv + cv 3 + av(u2 1).
Try solutions
ui(x, t) =

M
Xi
j=0

aij tanh [c1x + c2t + ] +

Ni
X

bij sech2j+1[c1x + c2t + ].

j=0

or
ui(x, t) =

M
Xi

(aij + bij sech[c1x + c2t + ]) tanhj[c1x + c2t + ].

j=0

Obviously, for ODEs c2 = 0.

Solitary wave solutions:
v
u
u
u
u
t

a2 c
u = tanh[
x + ]
2(a c)
v
u
u
u
u
t

v
u
u
u
t

1a
a2 c
v =
sech[
x + ],
ac
2(a c)
s

provided b =

a2 c
2(ac) .
24

 Example: Nonlinear Schrodinger equation (focusing/defocusing):

i ut + uxx |u|2u = 0.
Bright soliton solution (+ sign):
c
c2
k
2
u(x, t) = exp[i ( x + (k )t)] sech[k(x ct x0)]
2
4
2
Dark soliton solution ( sign):
u(x, t) =

1
c2
2
2
exp[i(Kx (2k + 3K 2Kc + )t)]
2
2
)
(
c
k tanh[k(x ct x0)] i(K ) .
2

Example: Nonlinear sine-Gordon equation (light cone coordinates):

uxt = sin u.
Setting = ux, = cos(u) 1, gives
xt = 0
2 + 2 + 2t = 0.
Solitary wave solution (kink):
1
1
= ux =
sech[ (x ct) + ],
c
c
1
= cos(u) 1 = 1 2 sech2[ (x ct) + ],
c
in final form:
1
u(x, t) = 4 arctan exp( (x ct) + ) .
c

25

 Example: Coupled nonlinear Schrodinger equations:

i ut = uxx + u(|u|2 + h|v|2)
i vt = vxx + v(|v|2 + h|u|2)
Seek particular solutions
u(x, t) = a tanh(x) exp(iAt)
v(x, t) = b sech(x) exp(iBt).
Seek solutions u(x, t) = U (F ()), where derivatives of F () are
polynomial in F.
Now,
F 0() = 1 F 2() F = tanh().
Other choices are possible.
Add the constraining differential equations to the system of PDEs
directly.
Why are tanh and sech solutions so prevalent?
Other applications:
Computation of conservation laws, symmetries, first integrals, etc.
leading to linear parameterized systems for unknowns coefficients
(see InvariantsSymmetries by Goktas and Hereman).

26