See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/265598610

Method of Inverse Differential Operators applied to certain classes of

nonhomogeneous PDEs and ODEs

Conference Paper · September 2012

DOI: 10.13140/2.1.2716.0966

CITATIONS READS

4 3,161

1 author:

Robert Dr. Kragler

Weingarten University of Applied Sciences, (Hochschule Ravensburg-Weingarten, Weingarten, Germany)
58 PUBLICATIONS   86 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Contour Integration View project

Solid Modeling with Mathematica View project

All content following this page was uploaded by Robert Dr. Kragler on 13 September 2014.

The user has requested enhancement of the downloaded file.

 Method of Inverse Differential Operators
applied to certain classes of nonhomogeneous
PDEs and ODEs
Prof. Dr. Robert Kragler
Weingarten University of Applied Sciences
kragler@hs-weingarten.de
http://portal.hs-weingarten.de/web/kragler/mathematica/

á Introduction
The paper deals with the Method of Inverse Differential Operators (MIDO) which is already well
established for ordinary differential equations (ODE) but has never been thoroughly applied to
nonhomogeneous partial differential equations (PDE). P.K.Kythe, P.Puri and M.R.Schaeferkotter
[1] have extended MIDO stepwise to PDEs but the full implementation into the CAS Mathematica
is new.

General restriction for the differential equations (DE), PDEs or ODEs, under consideration is that

Ú
n1 ,n2 ,…
Lx1 ,x2 ,… = ai, j,… Dx1 i1 Dx2 i2 … is a linear partial differential operator polynomial
i1 ,i2 ,…=0

ΧIDx1 , Dx2 , …M with constant coefficients (a,b,…,Α,Β,…,-2,3,…). Here, Dx1 m = ¶x1 m , Dx2 n = ¶x2 n
etc. represent partial derviatives of order m, n,… In order to faciliate the algebraic manipulation of
the differential operator polynomial Χ an intermediate representation in terms of pseudo
differential operators Dxi is introduced which are fully convertable into each other, e.g.

Dx 3 Dy 2 u(x,y) – ¶x,x,x ¶y,y uHx, yL. Thus, ΧIDx1 , Dx2 , …M uHx1 , x2 , …L constitutes the lhs of
the PDE. The rhs of the DE is either Φ= 0 in case of a homogeneous equation or ΦHx1 , x2 , …L for
a nonhomogeneous one. In principle, the order of the PDE in terms of (pseudo) differential
operators may be quite general.

As to homogeneous PDEs (where Φ = 0) the general solution is denoted as uh . For the method
used it is essential that the differential operator polynomial ΧIDx1 , Dx2 , ...M can be factorized into
linear factors of type L Κ x1 ,x2 ,... xn = IΑ1 Dx1 + Α2 Dx2 + … + Αn Dxn + ΓM Κ with multiplicity Κ for a
subset of n independent variables 8x1 , x2 , ... xn < Î 8t, x, y, z, Ξ, Η, Ζ<. Hence, each linear factor
(Κ =1) gives rise to the following type of solution :

IΑ1 Dx1 + Α2 Dx2 + … + Αn Dxn + ΓM Þ

uh Hx1 , x2 , ... xn L = f1 × ã-Γ x1 ‘Α1

Α1 x2 - Α2 x1 Α1 x3 - Α3 x1 Α1 xn - Αn x1
, , ... ,
Α1 Α1 Α1

If a linear factor, e.g. IΑ1 Dx1 + Α2 Dx2 + Γ M Κ , has multiplicity Κ >1 then the corresponding
solution is :

IΑ1 Dx1 + Α2 Dx2 + Γ M Κ Þ uh Hx1 , x2 L = Ú x1 k fk HΑ1 x2 -Α2 x1 L × ã-Γ x1 ‘Α1

Κ-1

k=0

As to nonhomogeneous PDEs the functional form of the nonhomogenuity ΦHx1 , x2 , …L ¹ 0 is

subject to certain restrictions which are essential to the applicability of MIDO. In this respect the
nonhomogenuity Φ can either be :

(i) an exponential function Φ1 = ãa x + b y + c z+… ,

(ii) a trigonometric functions Φ2 = sin cosHa x + b y + c z + …L ,
(iii) a hyperbolic functions Φ2 = sinh coshHa x + b y + c z + …L or
123
(iv) any multiplicative combination of an exponential Φ1 with trigonometric or hyperbolic
 (i) an exponential function Φ1 = ãa x + b y + c z+… ,
(ii) a trigonometric functions Φ2 = sin cosHa x + b y + c z + …L ,
(iii) a hyperbolic functions Φ2 = sinh coshHa x + b y + c z + …L or
(iv) any multiplicative combination of an exponential Φ1 with trigonometric or hyperbolic
functions Φ2 such that Φ = Φ1 ×Φ2 = ãa x + b y + c z+… × sin cos sinh coshH…L resp. ,

(v) any additive combination Φ2 = Ú Φ2 i with terms

k

i=1
a x + b y + c z+…
Φ2 i = ã × sin cos sinh coshH…L resp. ;

(vi) an arbitrary pure monomial Φ3 = MHx, y, z, …L = Ú Αi x ki y mi zni ×… ,

k

i=1

(vii) a monomial with rational part Φ3 = Ú Αi x ki y mi z¡ni ×… ,

k

i=1

(iix) a monomial multiplied with logarithmic term Φ3 = Ú Αi x ki y mi z¡ni ×logHyL … or

k

i=1

(ix) a monomial multiplied with exponential function Φ1 × Φ3 =ãa x + b y + c z+… ×MHx, y, z, …L ,

(x) a monomial multiplied with trigonometric function

Φ2 × Φ3 = sin cosHa x + b y + c z + …L× MHx, y, z, …L ,

(xi) a monomial multiplied with hyperbolic function

Φ2 × Φ3 = sinh coshHa x + b y + c z + …L×MHx, y, z, …L or

(xii) any sum Φ = Ú Φi with multiplicative combinations of Φ1 , Φ2 and Φ3 (as given above).
k

i=1

The nonhomogenuity Φ is restricted to classes of exponential, trigonometric (sin or cos) and

hyperbolic (sinh or cosh) functions for which only linear combinations of an arbitrary selection of
reference variables from VList are allowed as arguments, e.g. (a x + Β y - 2z +…) but not
(a x 2 + Β y 3 +…). However, for monomials Φ3 any combination of power products xki ymi z¡ni … of
the reference variables is admitted. As regards to trigonometric or hyperbolic functions any terms
containing {tan, cot} resp. {tanh, coth} are excluded simply because only the subgroups {sin,
cos} or {sinh, cosh} are closed under differentiation. This behavior is reflected in certain
substitution rules within MIDO.

á The Method
For short, in order to calculate the particular solution of a DE the essential idea of MIDO is to
move the inverse of the differential polynomial Χ from the lhs to the rhs of the DE and apply it on
the nonhomogenuity Φ :

ΧIDx1 , Dx2 , …M uHx1 , x2 , …L = ΦHx1 , x2 , …L ” uHx1 , x2 , …L = ΧIDx1 , Dx2 , …M-1 ΦHx1 , x2 , …L

For subsequent computation several replacement rules play an important role :

(1) Inversion of differential operator polynomial Χ× Χ-1 Φ = 1 Φ

ΧIDx , Dy M B Φ Hx, yL F = ΦHx, yL

1
ΧIDx ,Dy M

124
(2) Factorization of Χ = Χ1 × Χ2 ×…

Φ Hx, yL =
1 1 1 1 1
Χ1 IDx ,Dy M × Χ2 IDx ,Dy M Χ1 IDx ,Dy M Χ2 IDx ,Dy M Χ2 IDx ,Dy M Χ1 IDx ,Dy M
ΦHx, yL = ΦHx, yL

(3) Exponential nonhomogenuity Φ1 = ãa x+b y+c z+…

ãa x + b y = ãa x + b y with Χ(a , b) ¹ 0 and Φ1 º ãa x + b y

1 1
ΧIDx ,Dy M ΧIa , bM

(4) Trigonometric nonhomogenuity Φ = Φ2 = sin cos H…L

CIDx 2 , Dy 2 M cos Ha x + b yL = CI-a2 , -b2 M cos Ha x + b yL

CIDx 2 , Dy 2 M sin Ha x + b yL = CI-a2 , -b2 M sin Ha x + b yL

(5) Hyperbolic nonhomogenuity Φ = Φ2 = sinh coshH…L

CIDx 2 , Dy 2 M cosh Ha x + b yL = CI a2 , b2 M cosh Ha x + b yL

CIDx 2 , Dy 2 M sinh Ha x + b yL = CI a2 , b2 M sinh Ha x + b yL

(6) Multiplicative nonhomogenuity Φ = Φ1 ×Φ2 = ãH…L × sin cos sinh coshH…L

ãa x + b y Φ2 Hx, yL = ãa x + b y Φ2 Hx, yL =
1 1
ΧIDx ,Dy M ΧIDx +a , Dy +bM

ãb y Φ2 Hx, yL = ãb y ãa x Φ2 Hx, yL ;
1 1
ΧIDx + a , Dy M
= ãa x
ΧIDx , Dy +bM

CIDx , Dy M ãa x + b y Φ2 Hx, yL = ãa x + b y CIDx + a , Dy + bM Φ2 Hx, yL

where C = I Χ-1 Mrationalized is the inverse of the differential polynomial which is "rationalized" such
that the denominator denom(C) is free of Dx , Dy , ...

(7) Monomial nonhomogenuity Φ3 = MHx, y, z, …L = Ú x ki y mi zni … Moreover, it turns out that

k

i=1
-m
monomials with rational terms x y k
z n
and/or with a logarithmic factor xk y± m zn log(z) are
covered by the algorithm as well.

1 1
a0 Dx n I1-R IDy , Dz ,...MM
MHx, y, z, …L = MHx, y, z, …L =
ΧIDx ,Dy ,Dz ...M

Ú
Nmax
Dx -n
1
= RIDy , Dz ...M j MHx, y, z, …L
a0
j= 0

where Χ-1 is expanded into a geometric series with respect to the residual expression
RIDy , Dz ...M and is applied to a monomial M of finite order so that the series of differential
operators Dy , Dz ... is truncated at some order Nmax .

(8) Additive nonhomogenuity Φ = Ú Φi where terms Φi = MH…L ãH…L sin cos sinh coshH…L
k

i=1
125
 (8) Additive nonhomogenuity Φ = Ú Φi where terms Φi = MH…L ãH…L sin cos sinh coshH…L
k

H Ρ1 Φ1 Hx, yL +Ρ2 Φ2 Hx, yL L = Ρ1 ΧID 1,D

i=1

Φ1 Hx, yL + Ρ2 Φ2 Hx, yL
1 1
ΧIDx ,Dy M x yM ΧIDx ,Dy M

á A simple example
For better understanding of the mechanism how the replacement rules previously given are
applied consider the following simple example of a 2nd order PDE with variables {x,y} :

I 3 Dx 2 - 2 Dx Dy - 5 Dy 2 M u Hx, yL = ãx-y + H3 x + yL.

From inspection of the lhs of the PDE the differential operator polynomial Χ turns out to be
Χ = I3 Dx 2 - 2 Dx Dy - 5 Dy 2 M = I3 Dx - 5 Dy M IDx + Dy M, whereas the nonhomogenuity Φ is the
sum of an exponential function Φ1 = ãx-y and a (simple) monomial Φ3 = H3 x + yL.

I3 Dx - 5 Dy M IDx + Dy M it is straightforward that the solution of the homogeneous PDE is

As can be seen from the decomposition of the differential polynomial Χ into the linear factors

uh = f1,0 J H5 x + 3 yLN + f2,0 Hy - xL . According to the second term f2,0 Hy - xL it is obvious that the
1
3

function ãx-y is only a special instance of f2,0 and will satisfy the homogeneous PDE.

In order to calculate the particular solution up1 for the monomial nonhomogenuity Φ3 = H3 x + yL a
series expansion of Χ-1 into a (truncated) geometric series is done.

up1 Hx, yL = Χ-1 @3 x + yD = @3 x + yD =

1

3 D2x - 2 Dx Dy - 5 D2y

@3 x + yD = ¡ … @3 x + yD =
1 1 1 2 Dy 5 Dy 2
= Dx -2 Dx -2 1 + +
3 2 Dy 5 Dy 2 3 3 Dx 3 Dx
1- +
3 Dx 3 Dx

Dx -2 BH3 x + yL + Dx -1 1F = Dx -2 @xD + Dx -2 @1D + Dx -3 @1 D =

1 2 y 2
=
3 3 3 9

3 2 3
x y x 2 x 1 11
= + + = x2 y + x3
6 3 2 9 6 6 54

It should be noted that the inverse differential operator Dx -1 is the antiderivative; thus Dx -n is a n-
th order nested (indefinite) integral with respect to x.

As a consequence of the discussion of the homogeneous solution it turns out that as regards to
Φ1 = ãx-y (after application of replacement rule (6) to Χ-1 ) the expression
Χ-1 IDx , Dy M ãx-y = ãx-y × Χ-1 IDx ® +1, Dy ® -1M @1D is singular. Therefore the naive ansatz
for up2 does not suffice the nonhomogeneous PDE : due to the fact that the perturbing function
Φ1 = ãx-y is already included in the homogeneous solution uh it is essential to multiply the ansatz
with an extra term x Κ where Κ is the multiplicity of the root of the linear factor IDx + Dy M (here Κ =
1). Thus, the general ansatz up2 = Ia0 + a1 x + a2 x2 M ãx-y has to be made with unknown
coefficients 8a0 , a1 , a2 < and substituted into the PDE. Then it turns out that up2 Hx, yL is a
1
particular solution of the PDE for all x only if the coefficients are chosen to be a1 = , a2 = 0 with
8
x-y
arbitrary value for a0 (because any ã already satisfies the homogeneous PDE, therefore a0 = 0
may be chosen). Due to replacement rule (2) the particular solution of the PDE is given as
up = up1 + up2 = J x 2 y + x3N +
1 11 1
x × ãx-y .126
6 54 8
 1). Thus, the general ansatz up2 = Ia0 + a1 x + a2 x2 M ãx-y has to be made with unknown
coefficients 8a0 , a1 , a2 < and substituted into the PDE. Then it turns out that up2 Hx, yL is a
1
particular solution of the PDE for all x only if the coefficients are chosen to be a1 = , a2 = 0 with
8

arbitrary value for a0 (because any ãx-y already satisfies the homogeneous PDE, therefore a0 = 0
may be chosen). Due to replacement rule (2) the particular solution of the PDE is given as
up = up1 + up2 = J x 2 y + x3N +
1 11 1
x × ãx-y .
6 54 8

á Implementation of MIDO
The central procedure for the solution of the DE is DESolve[Χ,Φ,onoff,opt]. It works as a
kind of "black box" : the only imput required is the differential operator polynomial ΧIDx1 , Dx2 , …M
given in terms of pseudo differential operators Dxi and the nonhomogenuity ΦHx1 , x2 , …L . There

DList = 9Dt , Dx , Dy , Dz , DΞ , DΗ , DΖ = corresponding to the independent variables from list

is no loss of generality if the selection of differential operators is restricted to the list

VList = { t, x, y, z, Ξ, Η, Ζ } . Both sets, DList and VList, may be changed if necessary.

The type of DE (ODE or PDE) is determined as regards to the number of variables 8x1 , x2 , … <
being used. The variables are counted from analyzing the number of distinct differential operators
Dxi used in Χ with reference to VList.Thus, if there is only a single variable x1 involved an ODE
is given whereas if several variables x1 , x2 , … occur a PDE has to be treated. It should be

pointed out that the coefficients occuring in Χ and Φ can either be numbers, e.g. :1, -3, 2 , …>
or symbols, e.g. {a,b,…,Α,Β,…}, or a mixture of both types which is a non-trivial problem to
distinguish them from the variables in VList.

The algorithm of MIDO takes the following steps :

(1) Whether the nonhomogenuity Φ is zero or nonzero decides which of the procedures
homogeneousDEsolutions (for Φ = 0) or nonhomogeneousDEsolutions (for Φ ¹ 0) is called
and thus either the homogeneous solution uh or the particular solution up will be calculated.
If Φ = 0 either the procedure homogeneousODEsolutions or homogeneousPDEsolutions is
called depending on the number of variables and the homogeneous solution uh is evaluated.

For Φ ¹ 0 the most general form of the nonhomogenuity is Φ = Φ3 + Φ3 ×Φ1 2 , where Φ3 denotes a
monomial (resp. polynomial for a single variable), Φ1 2 is a non-monomial which is either an
exponential Φ1 = expH…L, a trigonometric Φ2 = sin cosH…L or hyperbolic Φ2 = sinh coshH…L
function, Moreover, Φ1 2 may be multiplied with an additional monomial prefactor Φ3 . The routine
monomialTest[Φ,onoff] which investigates Φ returns the separated components of Φ
(monomial, non-monomial, monomial prefactor) for further treatment. Three different flags
(typeΦ1,typeΦ2,typeΦ3) are set with values T or F according to which the algorithm makes a
distinction of cases between five different combinations.

In the case (F,T,F) where only a non-monomial nonhomogenuity Φ1 2 is present its type is further
analyzed by means of the routine analyzeΦ@Φ, onoffD which returns a parameter typeΦ.
According to its value there will be a switch between different cases : "Exp" (for exponential
functions ), "Trig" (for trigonometric functions Φ2 = sin cos), "Hyp" (for hyperbolic functions
Φ2 = sinh cosh ), "Times" (for products Φ1 ×Φ2 ) and "Plus" (for Φ1 + Φ2 ).

In the case (T,F,F) the nonhomogenuity is a monomial Φ3 = MHx, y, z, …L = Ú x ki y mi zni … .

k

i=1
k -m
The algorithm covers, in addition, as well monomials with rational terms x y zn and/or with a
logarithmic factor xk y± m zn log(z). Due to five different combinations for the flags
(typeΦ1,typeΦ2,typeΦ3) to be considered the computation of up branches into one of the
procedures listed below :

(a) (T,F,F) for pure monomial Φ3 Þ uPMonomial Ä RationalΦ1@Χ, Φ3 , onoffD ;

127
(b) (F,T,F) for non-monomial Φ1 2 Þ uPExpΦ3@Χ, Φ1 , onoffD or
 (a) (T,F,F) for pure monomial Φ3 Þ uPMonomial Ä RationalΦ1@Χ, Φ3 , onoffD ;

(b) (F,T,F) for non-monomial Φ1 2 Þ uPExpΦ3@Χ, Φ1 , onoffD or

uPTrigΦ2@Χ, Φ2 , onoffD or uPHypΦ2@Χ, Φ2 , onoffD or
combinations uPExpΦ1 ÄTrigHypΦ2@Χ, Φ1 × Φ2 , onoffD]
resp. uPnonMonomialΦ21 Å Φ22@Χ, Φ1 + Φ21 + Φ22 …, onoffD ;

(c) (F,T,T) for monomial · non-monomial Φ3 ×Φ2 Þ

uPMonom Ä nonMonomial@Χ, Φ3 × Φ2 , onoffD ;

(d) (T,T,F) for monomial + non-monomial Φ3 + Φ2 Þ

uPMonomÅnonMonomial@Χ, Φ3 + Φ2 , onoffD;

(e) (T,T,T) for monomial + monomial · non-monomial Φ3 + Φ3 ×Φ2 Þ

uPMonom Ä nonMonomial@Χ, Φ3 + Φ3 × Φ2 , onoffD.

(2) The coefficients of the nonhomogenuity Φ are extracted by means of auxiliary routines :
coeffExpΦ[Φ], coeffTrigΦ[Φ], coeffHypΦ[Φ], coeffExpTrigOrHypΦ[Φ],
coeffnonMonomialΦ[Φ], coeffMonomialΦ[Φ], coeffMonomialLogΦ[Φ],
coeffMonom Å nonMonomΦ@ΦD or coeffMonom Ä nonMonomΦ@ΦD and passed through to one
of the special routines:
uPExpΦ1|uPTrigΦ2|uPHypΦ2[Χ,coeffs,Φ,Ρ,onoff],
uPExpΦ1ÄTrigHypΦ2[Χ,coeffs,Φ,Ρ,onoff], uPnonMonomialΦ21ÅΦ22[Χ,Φ,onoff],
uPMonomÄRationalΦ1[Χ,Φ,onoff], uPMonomialΦ1[Χ,Φ,onoff],
uPMonomÅnonMonomΦ[Χ,Φ,onoff], uPMonomÄnonMonomΦ[Χ,Φ,onoff]
which finally determine the solution up .

(3) The replacement rules which are important to deal with functions constituting Φ are generated
by the subsequent procedures :

subD2Coeffs : 8Di ® ci <

1 1
ãa x + b y = ãa x + b y
ΧIDx ,Dy M
rule 3 : ,
ΧIa , bM

:Di n ¦ I-ci 2 M I-ci 2 M Di >

n‘2 en‘2u
subDD2CoeffTrig :
n even n odd

rule 4 : CJDx 2 , Dy 2 N ∂ Ha x + b yL = CJ-a2 , -b2 N ∂ Ha x + b yL

cos cos
,
sin sin

:Di n ¦ I+ci 2 M I+ci 2 M Di >

n‘2 en‘2u
subDD2CoeffHyp :
n even n odd

CJDx 2 , Dy 2 N ∂ Ha x + b yL = CJ a2 , b2 N ∂ Ha x + b yL
cosh cosh
rule 5 : ,
sinh sinh

subD2DplusCoeffΦ1 : 8Di ® HDi + ci L<

rule 6 : CIDx , Dy M Φ2 Hx, yL ãa x + b y = ãa x + b y CIDx + a , Dy + bM Φ2 Hx, yL .

(4) Another crucial routine is rationalizeX[X,subDD,onoff] which 'rationalizes' the

denominator of the inverse differential operator polynomial Χ-1 IDx1 , Dx2 , …M. In analogy to
1 1 a-ä b
complex conjugation z whereby the denominator of a complex number = =
z a+ä b a2 +b2

becomes real, an expression which contains terms linear in (pseudo) differential operators Di ,
1 a-Di subDD a-Di
for example = Þ , is simplified by application of replacement rules 4 and 5.
a+Di a -Di 2
2 a2 ± ci 2

However, if products such as Di ×D j appear in the denominator then the rationalizing process has
to be repeated until all (pseudo) differential operators Di , D j , … become squared and thus can
be replaced by I±ci 2 M, I±c j 2 M , ... .
128
 z a+ä b a2 +b2

becomes real, an expression which contains terms linear in (pseudo) differential operators Di ,
1 a-Di subDD a-Di
for example = Þ , is simplified by application of replacement rules 4 and 5.
a+Di a -Di 2
2 a2 ± ci 2

However, if products such as Di ×D j appear in the denominator then the rationalizing process has
to be repeated until all (pseudo) differential operators Di , D j , … become squared and thus can
be replaced by I±ci 2 M, I±c j 2 M , ... .

(5) In order to verify the correctness of the solutions, whether uh and/or up will satisfy the original
PDE Χ IDx1 , Dx2 , …M uHx1 , x2 , …L = ΦHx1 , x2 , …L , there is another essential central procedure
given: testDE[Χ,Φ,u,PDEtype,onoff,optSimplify] which deduces from the number of
variables resp. differential operators the type of DE to be either an ODE or PDE.
For Φ = 0 which implies a homogeneous DE, the suitable test procedure to be applicable is
testHomDE[Χ,uh,PDEtype,onoff,optSimplify]. (For parameter PDEtype the default
value is "Off", for optSimplify it is Identity). The pseudo differential operators Dxi (which
are used only to faciliate algebraic manipulations of the differential operator polynomial Χ) are
converted into standard differential operators ¶xi n . The execution is done with the procedure
ConvertΧ2PDE[Χ,Φ,onoff]. Auxiliary routines used for the conversion of the pseudo
differential operators Dx n into (proper) differential operators ¶8x,n< ð are ruleD2D,
D2DMultiRule1, concatDRule1|2|3; ruleJ2Int, foldJ2Int, foldJ2IntS;
uvPairs, uvwTriples, uvwqQuadruples, varsNtuples; Duv2Dxy, Duvw2Dxyz,
Duvwq2Dtxyz; pairsDiDj1, triplesDiDjDk1 and quadruplesDiDjDkDl1. In this
way the familiar representation of a DE is reconstructed from Χ as, for example, required by the
built-in procedure DSolve from Mathematica.

(6) Moreover, in addition to the implemented solver DESolve both procedures testDE and
testHomDE investigate whether DSolve[PDEeqn,upVar,ΧVar] is capable finding a solution for
the given DE. There is a time constraint of 180 CPU seconds given; if it is exceeded the
computation will be aborted. It turns out, however, that only in rare cases DSolve can provide a
solution named upDS. Thus, the current implementation of MIDO within Mathematica provides
solutions for a wide range of PDEs which are not generally covered by the built-in DSolve .

(7) For PDEtype="On" the type of a 2nd order PDE (to be hyperbolic|parabolic|elliptic) is
determined from the coefficients of the differential operator polynomial Χ (in analogy to the type of
>0 hyperbolic
a quadric surface). As regards to the value of the discriminant D = = 0 PDE is parabolic;
<0 elliptic
otherwise the discriminant and hence the type of PDE is indeterminate. The procedure which
determines the type of a 2nd order PDE is EquationType and is switched On|Off by the
parameter PDEtype from testDE .

á Interface for switching between different representations of a DE

In order to change between different representations for PDEs within DESolve and DSolve three
useful conversion procedures are provided :

(1) ConvertPDEops2PDEeqn[PDEops,onoff] converts (the lhs of) a PDE given in pure

function operator representation into a PDE given in form of an equation with (dummy)
nonhomogenuity Φ = F suitable as input for DSolve , for example :

PDEops = I a0 ð + a1 ¶y ð + a2 ¶9x,1= 9y,3= ð + a3 ¶9x,2=,9z,2= ð + a4 ¶8y,4< ð M & ”

PDEeqn = a0 u@x, y, zD + a1 uI0,1,0M @x, y, zD +

a2 uI1,3,0M @x, y, zD + a3 uI2,0,2M @x, y, zD + a4 uI0,4,0M @x, y, zD Š F@x, y, zD

(2) ConvertPDEops2Χ[PDEops,onoff] converts a PDE given in pure function operator form

into a differential polynomial Χ with pseudo differential
129 operators Dxi and Φ = F, for example :
 (2) ConvertPDEops2Χ[PDEops,onoff] converts a PDE given in pure function operator form
into a differential polynomial Χ with pseudo differential operators Dxi and Φ = F, for example :

PDEops = I a0 ð + a1 ¶y ð + a2 ¶9x,1= 9y,3= ð + a3 ¶9x,2=,9z,2= ð + a4 ¶8y,4< ð M & ”

Χ = a0 + a1 Dy + a2 Dx D3y + a3 D2x D2z + a4 D4y , Φ=F.

(3) ConvertPDEeqn2PDEops[PDEeqn,vars,onoff] converts a PDE given in form of an

required as input for DSolve@PDEeqn, u, 8x1 , x2 , …<D , for example :

equation into a PDE in pure function operator representation with nonhomogenuity Φ = F

PDEeqn = a0 u@x, y, zD + a1 uI0,1,0M @x, y, zD + ”

a2 uI1,3,0M @x, y, zD + a3 uI2,0,2M @x, y, zD + a4 uI0,4,0M @x, y, zD Š F@x, y, zD
PDEops = I a0 ð + a1 ¶y ð + a2 ¶9x,1= 9y,3= ð + a3 ¶9x,2=,9z,2= ð + a4 ¶8y,4< ð M &

suitable for DSolve@PDEeqn, u, 8x1 , x2 , …<D into a differential polynomial Χ with pseudo
(4) ConvertPDEeqn2Χ[PDEeqn,vars,onoff] converts a PDE given in form of an equation

differential operators Dxi and Φ = F, for example :

PDEeqn = a0 u@x, y, ΖD + a1 uI0,0,1M @x, y, ΖD +

a2 uI1,3,0M @x, y, ΖD + a3 uI2,2,0M @x, y, ΖD + a4 uI4,0,0M @x, y, ΖD = F@x, y, ΖD ”
Χ = a0 + a1 DΖ + a2 Dx D3y + a3 D2x D2y + a4 D4x , Φ = F .

(5) ConvertΧ2PDEeqn[Χ,F,onoff] casts a differential polynomial Χ with pseudo differential

operators Dxi and nonhomogenuity Φ = F into a PDE given in form of an equation suitable as
input for DSolve@PDEeqn, u, 8x1 , x2 , …<D, for example :

Χ = a0 + a1 DΖ + a2 Dx D3y + a3 D2x D2y + a4 D4x , Φ = F ”

PDEeqn = a0 u@x, y, ΖD + a1 uI0,0,1M @x, y, ΖD +

@x, y, ΖD + a3 u @x, y, ΖD + a4 u @x, y, ΖD = F@x, y, ΖD
.
I1,3,0M I2,2,0M I4,0,0M
a2 u

Thus, these three conversion procedures will faciliate the switching between different
representations of a DE.

á Application of MIDO to 12 classes of nonhomogeneous PDEs

(1) exponential nonhomogenuity Φ = Φ1 = ExpH ...L

(i) The 4th order PDE H5 + Dx L2 I1 + Dy M H2 + Dz L uHx, y, zL = 4 ã-5 x-y + z

Χ = H5 + Dx L2 H1 + Dy L H2 + Dz L; Φ = 4 ã -5 x - y + z ;
DESolve[Χ,Φ,"Off"];

It may be noted that in this case DESolve[Χ,Φ,onoff] does not return most general particular
solution for the nonhomogeneous PDE H5 + Dx L2 I1 + Dy M H2 + Dz L uHx, y, zL = 4 ã-5 x-y + z , but

x2 y ã-5 x-y+z only which comprises a monomial part

2 1
up = x y 2 and an exponential function
3 2

ãx+2 y . However, up may rather be supplemented by additional lower order monomial terms
usupp = ã-5 x-y+z IΑ1 + x Α2 + x2 Α3 + y Α4 + x y130
Α5 M which satisfy the PDE too. This is achieved by
the procedure usupp = uPsupplement@u p , onoffD which requires as input only the existing
It may be noted that in this case DESolve[Χ,Φ,onoff] does not return most general particular
solution for the nonhomogeneous PDE H5 + Dx L2 I1 + Dy M H2 + Dz L uHx, y, zL = 4 ã-5 x-y + z , but

x2 y ã-5 x-y+z only which comprises a monomial part

2 1
up = x y 2 and an exponential function
3 2

ãx+2 y . However, up may rather be supplemented by additional lower order monomial terms
usupp = ã-5 x-y+z IΑ1 + x Α2 + x2 Α3 + y Α4 + x y Α5 M which satisfy the PDE too. This is achieved by
the procedure usupp = uPsupplement@u p , onoffD which requires as input only the existing
particular solution up .

usupp = uPsupplement@up , "Off"D;

Thus, the supplemented particular solution turns out to be :

ã-5 x-y+z x 2 y + ã-5 x-y+z IΑ1 + x Α2 + x 2 Α3 + y Α4 + x y Α5 M . Testing the resulting
2
up + usupp =
3
solution with testDE@Χ, Φ, u p + usupp, OffD gives rise to the subsequent typical output :

I 3 Dx 2 - 2 Dx Dy - 5 Dy 2 M u Hx, yL = ãx- y is handled.

(ii) Another example shows how the degenerate solution occuring for :

Χ = I3 Dx 2 - 2 Dx Dy -5 Dy 2 M; Φ := ãx-y ;
uh = DESolve@Χ, 0, OffD;

The PDE I 3 Dx 2 - 2 Dx Dy - 5 Dy 2 M u Hx, yL = ãx- y already possesses the homogeneous

solution uh = f1,0 J H5 x + 3 yLN + f2,0 H-x + yL.

1
3

up = DESolve@Χ, Φ, OffD;

Due to the factorization of Χ = I3 Dx - 5 Dy M IDx + Dy M the replacement rule 9Dx ® 1, Dy ® -1=

due the second factor (which results from interchange of ã x- y with Χ-1 ) causes the inverse
differential polynomial Χ-1 to becomes singular. In order to cope with this degeneracy of the
particular solution up ~ãx-y with one of the homogeneous solutions uh1 = f2,0 H-x + yL the
procedures uPmodifySingular[up,onoff] and optimizeSolution[Χ,Φ,up,onoff]
give rise to the following ansatz Ha1 x + … + aΚ x Κ L ãx-y with multiplicity Κ = 1; thus the resulting
1
solution turns out to be up = x ãx-y . Testing the resulting solution with
8

testDE@Χ, Φ, uh + up , " Off "D verifies the correctness.

(2) trigonometric nonhomogenuity Φ = Φ2 = sin cosH…L

This is a parabolic 2nd order PDE I3 Dx 2 - Dy + 4 Dz M uHx, yL = sin H a x + b y + c zL

Χ=I3 Dx 2 -Dy +4 Dz M; Φ = Sin[a x + b y + c z];

DESolve[Χ, Φ, "Off"];

Hb-4 cL cosHa x+b y+c zL-3 a2 sinHa x+b y+c zL

The particular solution is up = is again verified
9 a4 +Hb-4 cL2
by testDE.

(3) hyperbolic nonhomogenuity Φ = Φ2 = sinh coshH…L

131
For the 4th order PDE : IDx 4 + 3 Dx 2 Dy + 2 Dt M uHx, y, tL = sinh HyL

Χ = IDx 4 + 3 Dx 2 Dy + 2 Dt M; Φ = Sinh[y];
DESolve[Χ, Φ, "Off"];

8Dt n ® 0, Dx n ® 0,
there occurs an exceptional case : in the process of 'rationalizing' Χ-1 the denominator reduces to
Dt and will vanish after applying the appropriate replacement rules

> . However, the routine dTermsException copes with the situation

n = even
Dy n ®
1
Dy n = odd

ΧR = ¥ and instead handles the antiderivative Dt -1 Þ J @tD ž Sinh@yD which gives rise to the
sinh HyL which is supplemented by Α1 sinhHyL which is verified by testDE. The
t
correct result
2
t
particular solution is up = sinhHyL + Α1 sinhHyL which is verified by testDE:
2

(4) multiplicative nonhomogenuity Φ = Φ1 ×Φ2 = ãH…L × sin cos sinh coshH…L

For the 4th order PDE : I3 Dx 4 - Dy + Dz 2 M uHx, y, zL = ãΑ x + Βz

sinhH b x + a yL

Χ=I3 Dx 4 - Dy + Dz 2 M; Φ = ãΑ x + Β z Sinh@b x+a yD;

DESolve[Χ,Φ,"Off"];

there is 9Dx , Dy , Dz = whereas the coefficient list from Φ2 gives only {b, a} (originating from
Φ2 = sinHb x + a yL L. Hence, in this specific case the coefficient list must be extended with the

replacement list will be instead 9Dx ® b , Dy ® a , Dz ® 0= . The correct particular solution is

help of makeListsEqualLength[varD,Φ,onoff] to {b, a, 0} so that the correct

up =
Iãx Α+z Β I3 a Ib4 + 6 b2 Α2 + Α4 M Cosh@b x + a yD + I-12 a b Α Ib2 + Α2 M + Β2 M Sinh@b x + a yDMM’

J-9 a2 Ib2 - Α2 M - 24 a b Α Ib2 + Α2 M Β2 + Β4 N

4

verified by testDE

(5) additive (non-monomial) nonhomogenuity Φ2 = Ú Φ2 i with Φ2 i = exp sin cos sinh cosh
k

i=1

The example given makes essentially use of uPnonMonomialΦ21ÅΦ22 to deal with the sum of
non-monomial terms Φ21 + Φ22 + Φ23 + … .

For the 4th order PDE : I Dx 4 + 3 Dx 2 Dy + 2 Dt M u Hx, yL = Φ were the nonhomogenuity

x-y
Φ= Α ã + sinHx + yL + cosHt - xL + 2 sinhH x - 2 y + 3 tL + 3 coshHx + 2 y - 3 tL is a mixed
sum of exponential, trigonometric and hyperbolic functions

Χ=IDx 4 +3 Dx 2 Dy +2 Dt M;
Φ = Γ ãx-y +Sin@x+yD +Cos@t-xD
+ + 2 Sinh@x-2 y+3 tD +3 Cosh@x+2 y-3 tD;
DESolve[Χ,Φ,"Off"];

Hsin Hx + yL + 3 cos Hx + yLL +

1 1
the particular solution is up = - Γ ãx-y +

H2 sinHt - xL + cosHt - xLL + 2 sinhH3 t + x - 2 yL + 3 coshH3 t - x - 2 yL

2 10
1
+
5
132
and is verified by testDE.
 Hsin Hx + yL + 3 cos Hx + yLL +
1 1
the particular solution is up = - Γ ãx-y +

H2 sinHt - xL + cosHt - xLL + 2 sinhH3 t + x - 2 yL + 3 coshH3 t - x - 2 yL

2 10
1
+
5
and is verified by testDE.

Φ = Φ3 = MHx, y, z, …L = Ú Αi x ki y mi zni ×…
k
(6) monomial nonhomogenuity
i=1

Pure monomial nonhomogenuity

Φ1 = MHx, y, z, …L = Φ11 + Φ12 + Φ13 + … = Α t k1 x m1 y n1 × ... + Β t k2 x m2 y n2 … + … gives rise to

an expansion of Χ-1 into a truncated geometric series. The order iMax of the truncated geometric
series expansion is determined in a heuristical way as sum of leading exponents ni of the
monomial variables in Φ1 , i.e. iMax = k1 + m1 + n1 + … . (In case of a rational term xi -m the
minimum exponent is chosen for iMax = m ). However, this approach sometimes leads to
huge expansion order which has to be corrected by a global positive variable $jMax (where its

truncatedSeries[Χ,leadD,iMax-$jMax,onoff]. E.g. Χ = IDt + 2 Dx + 3 Dy - 7 DΖ M with

default value is 0) to diminish the expansion order. iMax serves as input to the routine

leadD = Dt as leading term gives rise to

Χ-1 = Dt -1 × = Dt -1 × = Ú Dt -Hi+1L × HΡLi

1 1 iMax
3 Dy 7 DΖ 1 -ѐDt
1 -J- 2DDx - Dt
+
Dt
i=0
t
ΡD

Only in cases where the differential operator polynomial ΧIDt , Dx , … ) is not too complicated
and the nonhomogenuity Φ is only a monomial then the built-in procedure DSolve is
(sometimes) able to calculate a solution.

The 1st order PDE in the variables {t,x,y,z} :

IDt + 2 Dx + 3 Dy - 4 Dz M u Ht, x, y, zL = H3 x + yL + 5 x 4 y 5 t 6 + Α x y 2 z3

Χ =HDt +2 Dx +3 Dy -4 Dz L; Φ = H3 x + yL + 5 x4 y5 t6 + Α x y2 z3 ;
$jMax = 14;
DESolve[Χ,Φ,"Off"];

has the solution up where the expansion order is reduced from iMax=6+4+5+3=18 down to
order 4 by subtraction of $jMax=14.

25 920 2160 600 75 5 768

up = t 11 y - t 10 x y 2 + t 9 x2 y3 - t 8 x3 y4 + t 7 x4 y5 - t5 y Α +

+ t 4 I16 x y 2 Α - 144 y z ΑM + t 3 I16 x y 2 z Α - 48 y z2 ΑM +

77 7 7 7 7 5

+t 2 I6 x y 2 z2 Α - 6 y z3 ΑM + t Ix y 2 z3 Α + 3 x + yM

In order to test the efficiency of the implementation of MIDO the following examples have been
investigated :

(i) Χ= IDx + Dy - Dz M with Φ = Hx + y + zLn (n =1,2,15,20,30,50,70,100)

(ii) Χ= IDx k + Dy k - Dz k M with Φ = Hx + y + zL10 (k =1,2,3,5,10)

to iMax=1 so that there results an antiderivative J @xDm with respect to the leading term Dx up to
For case (i) with n =100 one has to choose $jMax=3n-1=299 (!) to reduce the expansion order

order m = 2 , i.e. 1 J @xD - Dy J @xD2 + Dz J @xD2 . The particular solution up consists of 5253
terms,
x101 †1‡
up = + + †200‡ + 133

+x Iy100 + 100 y99 z + 4950 y98 z2 + 161 700 y97 z3 + 3 921 225 y96 z4 + 75 287 520 y95 z5 +
101 101
to iMax=1 so that there results an antiderivative J @xDm with respect to the leading term Dx up to
For case (i) with n =100 one has to choose $jMax=3n-1=299 (!) to reduce the expansion order

order m = 2 , i.e. 1 J @xD - Dy J @xD2 + Dz J @xD2 . The particular solution up consists of 5253
terms,
x101 †1‡
up = + + †200‡ +

+x Iy100 + 100 y99 z + 4950 y98 z2 + 161 700 y97 z3 + 3 921 225 y96 z4 + 75 287 520 y95 z5 +
101 101

+†90‡ + 3 921 225 y4 z96 + 161 700 y3 z97 + 4950 y2 z98 + 100 y z99 + z100 M

results an antiderivative J @xDm with respect to the leading term Dx up to order m=2k=10. The
For case (ii) with k = 5 it requires $jMax = 3n-1 = 29 to reduce the expansion order. There

particular solution is :

up =
x 15 y x 14 z x 14 y 2 x 13 z2 x 13 y z x 13 y 10 z5 y 11 z4 y 12 z3 y 13 z2 y 14 z
+ + + + + + ... + + + + +
360 360 24 024 24 024 3432 3432 1716 120 264 792 3432 24 024

(7) monomial nonhomogenuity with rational part Φ = Φ3 = x ±m y ±n z±k × …

This is an elliptic 2nd order PDE IDx 2 +Dy 2 +Dz 2 M uHx, y, zL = Β x -5 y 4 z with a monomial
-5
containing a rational term x .

Χ =IDx 2 +Dy 2 +Dz 2 M; Φ = Β x-5 y4 z; $jMax = 3;

DESolve[Χ,Φ,"Off"];

I-12 x 4 - 6 x 2 y 2 + y 4 + 12 x 4 log HxLM

zΒ
The particular solution is up =
12 x 3

(8) monomial·log nonhomogenuity Φ = Φ3 = x ±m y ±n z±k ×… logHyL

This is an elliptic 2nd order PDE IDx 2 +Dy 2 +Dz 2 M uHx, y, zL = Β x -5 y 4 z with a monomial
containing in addition a rational term z-4 and a logarthmic term log(z) .

Χ =IDx 2 +Dy 2 +Dz 2 M; Φ = x2 y3 z-4 Log@zD; $jMax = 0;

DESolve[Χ,Φ,"Off",Simplify];

The particular solution is

J5 x2 y2 + 72 z4 + J22 y2 z2 - 24 z4 + 6 x2 Jy2 + 11 z2 NN logHzL + 6 z2 J3 x2 + y2 - 6 z2 N logHzL2 N
y
up =
36 z2

(9) monomial · exponential nonhomogenuity Φ = Φ3 ×expH…L

This is a 3rd order PDE IDx 3 + 3 Dy 2 + Dz - 4M uHxL = Ia x + b y 2 + c z3 M×IΑ + x + y 4 M ãz with the

product of two monomials multiplied with an exponential function.

Χ=IDx 3 +3 Dy 2 +Dz -4M; Φ=Ia x +b y2 +c z3 M IΑ+x+y4 M ãz ;

DESolve[Χ,Φ,"Off",polyForm];

The particular solution is

b y6 1 1 2 c y4 1 2
up = ã z - - c z3 y4 - c z2 y4 - 10 b y4 - - a x y4 - c z y4 - 4 c z3 y2 - 8 c z2 y2 - 120 b y2 -
3 3 3 27 3 9

- - - - - - - - - - -

134
 32 c y2 1 1 1 a x2 1
- 4 a x y2 - b x y2 - 8 c z y2 - b Α y2 - 8 c z3 - c x z3 - - 24 c z2 - c x z2 - 240 b -
9 3 3 3 3 3

160 c 2bx 2cx 2cxz 1 1 2bΑ 2cΑ axΑ 2czΑ

- 8ax- - - 32 c z - - c z3 Α - c z2 Α - - - -
9 3 27 9 3 3 3 27 3 9

(10) monomial · trigonometric nonhomogenuity Φ = Φ3 ×sin cosH…L

This is a 3rd order PDE IDx 3 + 3 Dy 2 - 4M uHx, yL = x y 2 × cosH5 x + 3 yL with a monomial x y2

multiplied with cos(5 x + 3 y)

Χ=IDx 3 +3 Dy 2 -4M; Φ = x y2 Cos@5 x + 3 yD;

DESolve[Χ,Φ,"Off",Simplify];

H -I8293 x I-134 245 548 + 1 156 873 500 y + 2 131 989 319 y 2 M +
The particular solution is
1
up =

+450 I12 178 155 469 + 39 626 027 250 y + 84 041 643 478 y 2 MM cosH5 x + 3 yL +
9 459 684 612 549 602

+I-8293 x I450 837 750 + 2 188 953 936 y + 8 596 731 125 y 2 M +
+75 I15 605 928 750 + 212 466 759 516 y + 266 498 664 875 y 2 MM sinH5 x + 3 yL L
(11) monomial · hyperbolic nonhomogenuity Φ = Φ3 ×sinh coshH…L

This is a 3rd order PDE IDx 3 + 3 Dy 2 - 4M uHx, yL = y ×coshH5 xL with a simple monomial y
multiplied with cosh(5 x) .

Χ =IDx 3 +3 Dy 2 -4M; Φ = y Cosh@5 xD;

DESolve[Χ,Φ,"Off"];

4 y coshH5 xL 125 y sinh@H5 xL

The particular solution is up = +
15 609 15 609

(12) additive nonhomogenuity Φ = Ú Φi with Φi = MH…L exp sin cos sinh coshH…L
k

i=1

This is a 4th order PDE IDt 2 + 2 Dx 4 + 3 Dx 2 Dy - 4 Dz M uHt, x, y, zL = Φ with a sum of a

Φ1 part Φ2 Φ1 = Α x 4 y 5 t 6 + Β x y 2 z 3 ,
Φ2 = H3 x + yL× HΑ ãx-y + sin Hx + yL + cos Ht - xL + 2 sinh H x - 2 y + 3 tL + 3 cosh Hx + 2 y - 3 tL )
monomial part and non-monomial :

Χ =IDt 2 +2 Dx 4 +3 Dx 2 Dy -4 Dz M; $jMax = 0;
Φ1 = Α x4 y5 t6 + Β x y2 z3 ;
Φ2 = H3 x + yL * H Α ãx- y + Sin@x + yD + Cos@t - xD +
+2 Sinh@ x - 2 y + 3 zD + 3 Cosh@x + 2 y - 3 zD L;
DESolve@Χ,Φ1 +Φ2 ,"Off"];

The particular solution up is (collected with respect to powers of t ) :

t 10 y 4 J15 x 2 + 4 yN Α +
1 1
up = t 12 y 3 Α -
154 420

t 8 J15 x 4 y 5 Α + 8 x y 2 ΒN +
1 2 1 4 1 2
t6 x y2 z Β + t x y 2 z2 Β + t x y 2 z3 Β +

-ãx-y H9 + 3 x + yL + H3 + 3 x + yL cos@t - xD + 24 sin@t - xD +

840 15 2 2

H44 + 39 x + 13 yL cos@x + yD + H-393

3 1
135 + 78 x + 26 yL sin@x + yD +

3 H-H1383 + 240 x + 80 yL Cosh@x + 2 y - 3 zD +

169 169
1
View publication stats

The particular solution up is (collected with respect to powers of t ) :

t 10 y 4 J15 x 2 + 4 yN Α +
1 1
up = t 12 y 3 Α -
154 420

t 8 J15 x 4 y 5 Α + 8 x y 2 ΒN +
1 2 1 4 1 2
t6 x y2 z Β + t x y 2 z2 Β + t x y 2 z3 Β +

-ãx-y H9 + 3 x + yL + H3 + 3 x + yL cos@t - xD + 24 sin@t - xD +

840 15 2 2

H44 + 39 x + 13 yL cos@x + yD + H-393 + 78 x + 26 yL sin@x + yD +

3 1

3 H-H1383 + 240 x + 80 yL Cosh@x + 2 y - 3 zD +

169 169
1

3 H-211 + 720 x + 240 yL Sinh@x + 2 y - 3 zDL +

12 800

-3 I229+720 x+240 yM CoshAx-2 y+3 zE+I1137-240 x-80 yM SinhAx-2 y+3 zE

6400

á Summary
As demonstrated for various types of nonhomogenuities Φ the Mathematica implementation of
MIDO can be applied to a wide class of (linear) PDEs for which in most cases the built-in
Mathematica procedure DSolve cannot provide a solution; thus DESolve really is an extension
for solving PDEs with Mathematica. However, there are some restrictions inherent to this method :

(i) the (pseudo) differential operator polynomial Χ is subject to decomposition into linear factors
such that (Α Dt + Β Dx + ... +ΓLΚ where Κ denotes the multiplicity.

(ii) the nonhomogenuity Φ is restricted to a limited class of functions with linear arguments such
as e.g. (a x + b y + ...) allowed only but not, for example, arguments which contain higher powers
of the variables (a x 2 + b y 3 + ... ). The functions admitted for Φ are exponential, trigonometric
(sin or cos) and hyperbolic functions (sinh or cosh) and products resp. sums of these functions.
Other trigonometric or hyperbolic functions such as {tan,cot} resp. {tanh,coth} are excluded
simply because for them the replacement rules which are essential for MIDO do not hold. Only
the subgroup of functions {sin,cos} resp. {sinh,cosh} is closed under differentiation.

(iii) if the nonhomogenuity Φ is a monomial MHx, y, z, …L = Ú

k
Α1 x ki y mi zni ×… (which may
i=1

even contain an additional rational term i.e. y -mi or a logarithmic factor e.g. log(y) ) then any
combination of power products with respect to the reference variables in VList is allowed.

Although due to limitation of space not demonstrated here MIDO covers homogeneous and
inhomogeneous ODEs too; adaptation of the procedures to ODEs required only some minor
modifications with respect to pattern recognition.

References :

[1] P. K. Kythe, P. Puri & M. R. Schäferkotter, "Partial Differential Equations and Boundary
Value Problems with Mathematica", 2nd ed. , Chapman & Hall (2002), Chapt. 3.2-3.4, p. 75 - 81.

[2] V. Gerdt / JINR, Dubna, private communication

[3] Special thanks to J. Harris, Oliver Rübernkönig, L. Shifrin and Michael Trott, all at WRI, for
their valuable contributions and help during the development of the implementation

[4] R. Kragler "Method of Inverse Differential Operators for the Solution of PDEs" published in
Computer Algebra Systems in Teaching and Research, 6th Internat. Workshop, CASTR 2011,
Siedlce, Feb. 2- 6, 2011. Proceedings (vol. Differential Equations, Dynamical Systems and
Celestial Mechanics ). L. Gadomski et al. (Eds.). Siedlce (2011) 79-95

[5] R. Kragler "Solution for certain classes of PDEs with the Method of Inverse Differential
Operators". 14th Computer Algebra Seminar, LIT/JINR, Dubna (June 2, 2011)

136