EOIS: 2892_chap02 2004/9/28 19:18 page 55 #1

2
Unusual Closed-Form Solutions in
Column Buckling
This chapter deals with obtaining closed-form solutions for inhomogeneous
columns under various load conditions and boundary conditions. Among
others, Engessers (1893) and Duncans (1937) closed-form solutions are gen-
eralized to a wide class of problems, including the generalized Euler problem,
that of a column under its own weight, or under polynomially varying distrib-
uted axial loading. In these problems we postulate the buckling mode as the
fourth-order polynomial.
2.1 New Closed-Form Solutions for Buckling of
a Variable Flexural Rigidity Column
2.1.1 Introductory Remarks
The bucklingof uniformcolumns under various loadingandboundarycondi-
tions is a well-studied topic. As far as the columns with variable cross section
are concerned, several exact solutions are available, in terms of logarithmic
and trigonometric (Bleich, 1952; Dinnik, 1955a,b), Bessel (Dinnik, 1928, 1932)
and Lommel (Willers, 1941; Engelhardt, 1954) functions. An exact solution in
terms of series for buckling load, for variable cross-section columns with vari-
able axial forces was furnished by Eisenberger (1991ad). The closed-form
solutions are extremely rare. Two cases will be described. For the column
(Engesser, 1893) that is pinned at both its ends and possesses the following
exural rigidity:
D(x) = 4x(L x)D
0
/L
2
(2.1)
where L is the length and x the axial coordinate, the governing differential
equation reads
_
4x(L x)D
0
/L
2
_
d
2
w
dx
2
+Pw = 0 (2.2)
55
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 56 #2
56 Eigenvalues of Inhomogeneous Structures
where w(x) is a displacement. Substitution of the mode
w(x) = A[4x(L x)/L
2
] (2.3)
where A is a constant, in Eq. (2.2) results in
4x(L x)D
0
/L
2
(8/L
2
) +P4x(L x)/L
2
= 0 (2.4)
Since for buckling A = 0, one obtains the buckling load
P
cl
= 8D
0
/L
2
(2.5)
Here, the exural rigidity varies as
D(x) =
_
1
3
7
(x/L)
2
_
D
0
(2.6)
so that the governing differential equation is
D
0
_
1
3
7
(x/L)
2
_
d
2
w
dx
2
+Pw = 0 (2.7)
The buckling mode was guessed by Duncan (1937) as
w(x) = A[7(x/L) 10(x/L)
3
+3(x/L)
5
] (2.8)
Substituting Eq. (2.8) into Eq. (2.7), the classical buckling load becomes
P
cl
=
60
7
D
0
/L
2
(2.9)
The author is unaware of other closed-form solutions for columns with
variable exural rigidity. Obtaining such solutions is worthwhile, since
closed-form solutions could serve as benchmark solutions for the purpose
of contrasting various approximate solutions with them. The results derived
by Elishakoff and Rollot (1999) are reported. In what follows, one generalizes
the two closed-form solutions discussed above.
2.1.2 Formulation of the Problem
The column buckling is governed by the differential equation
D(x)
d
2
w
dx
2
+Pw = 0 (2.10)
2005 by Issac Elishakoff
(see also Rzhanitsyn, 1955). A second example belongs to Duncan (1937).
EOIS: 2892_chap02 2004/9/28 19:18 page 57 #3
Unusual Closed-Form Solutions in Column Buckling 57
where D(x) is dened as
D = D
0
r(z) (2.11)
where D
0
is a constant and z is a non-dimensional coordinate dened as
z = x/L (2.12)
The governing differential equation, Eq. (2.10), can be rewritten as
r
d
2
w
dz
2
+k
2
w = 0 (2.13)
where k
2
is a constant dened as
k
2
= PL
2
/D
0
(2.14)
One deduces the buckling load from Eq. (2.14):
P = k
2
D
0
/L
2
(2.15)
In this study, r(z) is assumed to be a polynomial of second degree. Three dif-
ferent variations for r(z)are discussed that lead to new closed-form solutions
for the buckling load:
r = z z
2
r = 1 +z z
2
r = 1 z
2
(2.16)
In this section, the displacement w is assumed to be a polynomial function
that satises the differential equation and all boundary conditions. One nds
new closed-form solutions for some particular choices of

and .
2.1.3 Uncovered Closed-Form Solutions
Case 1: r = z z
2
The variation of D(x) is given by
D = D
0
(z z
2
) (2.17)
and the displacement is a polynomial of degree 2:
w = az +bz
2
(2.18)
The boundary conditions for a pinned column are:
w(0) = 0 Dw

(0) = 0 w(1) = 0 Dw

(1) = 0 (2.19ad)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 58 #4
58 Eigenvalues of Inhomogeneous Structures
Equations (2.19a) and (2.19b) are always satised; Eqs. (2.19c) and (2.19d)
lead to
b = a (2.20)
and
= (2.21)
Taking into account the boundary conditions in Eq. (2.19), one denes
r = (z z
2
) (2.22)
and
w = a(z z
2
) (2.23)
w has to satisfy the differential Eq. (2.13) for any z. This problem is solvable
with the aid of the Mathematica

command SolveAways (Wolfram, 1996).

Althoughfor solvingthe problemthe use of symbolic algebra is not absolutely
necessary, it is an extremely convenient tool. SolveAways yields parameter
values for which the given equation or system of equations which depend on
a set of parameters is valid for all variable values. The result of SolveAways
is given in the form of a list of all possible sets of values. SolveAways works
primarily with linear and polynomial equations.
For this case two sets are obtained. The rst one leads to a trivial solution
with a = 0. The second set leads to
k
2
= 2 (2.24)
Finally, using Eq. (2.15) one deduces the buckling load,
P = 2D
0
/L
2
(2.25)
The buckling mode reads as
w = a(z z
2
) (2.26)
This corresponds to the following denition of the exural rigidity:
D = D
0
(z z
2
) (2.27)
One can now relate to the rst example described in the introduction for =4
(see Eq. (2.1)). One obtains the same buckling load,
P = 8D
0
/L
2
(2.28)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 59 #5
Unusual Closed-Form Solutions in Column Buckling 59
For a = 4, the same buckling mode
w = 4(z z
2
) (2.29)
is found as in Eq. (2.3). Equation (2.25) allows one to optimize the column in
the presence of the buckling constraint, demanding that the column does not
buckle prior to a load level

P:
P
cl


P (2.30)
This yields the admissible region of variation of the parameter ,


PL
2
/2D
0
(2.31)
so that the buckling load of the column will satisfy the inequality (2.30).
If one assumes that the displacement is a polynomial of higher degree, one
can nd higher buckling loads. The method proposed below is base of the
logarithm of the Mathematica

function SolveAways. One assumes that the

displacement is of the form
w = w
0
N

j=1
a
j
z
j
(2.32)
This displacement equals zero at z = 0 so it satises the rst boundary
condition. The second derivative of w reads as
w

= w
0
N

j=2
a
j
j(j 1)z
j2
(2.33)
One denes the exural rigidity as
D = D
0
(z z
2
) (2.34)
This denition is very interesting because the exural rigidity equals zero at
both ends. So the bending moment Dw

equals zero at the ends identically,

irrespective of the denition of the displacement. The differential equation
becomes
(z z
2
)w

+k
2
w = 0 (2.35)
Taking into account Eqs. (2.32) and (2.33), one obtains
(z z
2
)
N

j=2
a
j
j(j 1)z
j2
+k
2
N

j=1
a
j
z
j
= 0 (2.36)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 60 #6
60 Eigenvalues of Inhomogeneous Structures
Now, the equation is modied to have only z
j
terms:
N1

j=1
a
j+1
j(j +1)z
j

N1

j=2
a
j
j(j 1)z
j
a
N
N(N 1)z
N
+k
2
a
1
z +
N

j=1
k
2
a
j
z
j
+k
2
a
N
z
N
= 0 (2.37)
Regrouping the terms of the same degree yields
(k
2
a
1
+2a
2
)z +
N1

j=2
[a
j+1
j(j +1) a
j
j(j 1)]z
j
[k
2
N(N 1)]a
N
z
N
= 0 (2.38)
This equation must equal zero for any z. For z
N
,
k
2
= N(N 1) (2.39)
and one deduces that
P = N(N 1)D
0
/L
2
(2.40)
The rst buckling load is for N = 2:
P
cl
= 2D
0
/L
2
(2.41)
As a result, a polynomial of degree N leads to mth buckling load with
N = m+1 (2.42)
The other terms of the polynomial lead to the global denition of the
coefcients. For the rst power of z one obtains
a
2
= m(m+1)a
1
/2 (2.43)
whereas for jth power of z
a
j+1
= {[j(j 1) m(m+1)]/j(j +1)}a
j
(2.44)
One notices that for j = 1, Eq. (2.44) reduces to Eq. (2.43). Hence, Eq. (2.44) is
valid for any a
j
. One chooses a
1
= 1. One can verify via Mathematica

that
the sum of the coefcients a
j
equals zero, and thus the boundary condition
of the displacement at the end z = 1, is satised.
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 61 #7
Unusual Closed-Form Solutions in Column Buckling 61
Now, one can consider that one has a column clamped beam at z = 0
and it has arbitrary boundary conditions at z = 1. If one assumes that w is
a polynomial, as in Eq. (2.32), then the coefcient a
1
must vanish to have a
slope equal to zero at z = 0. One has to take into account Eqs. (2.43) and (2.44).
These relations imply that all a
j
must equal zero. Finally, one arrives at a
trivial solution. One concludes that the present method is in applicable for
the column that has a clamped end at z = 0.
Case 2: r = 1 +z z
2
In this case, the exural rigidity is given by
D = D
0
(1 +z z
2
) (2.45)
The displacement is a polynomial of degree 5,
w = az +bz
2
+cz
3
+dz
4
+ez
5
(2.46)
After simplications, the boundary conditions (2.19) lead to the conditions
b = 0 a +c +d +e = 0 d =
3
a c = 2
2
a e = 0 (2.47)
Mathematica

, leads to three different sets of solutions which respect the

differential equation. The secondandthirdone yieldthe same bucklingmode.
Hence, one has to consider two sets which correspond to different buckling
loads and modes. The rst set is dened as
k
2
= 12
2
=
2
d =
3
a c = 2
2
a e = 0 (2.48)
This set of solutions implies the following denition of the exural rigidity
D = D
0
(1 +z z
2
) (2.49)
with attendant buckling load
P = 12
2
D
0
/L
2
(2.50)
and buckling mode
w = a(z 2
2
z
3
+
3
z
4
) (2.51)
These coefcients have to satisfy the boundary conditions (2.19). From this
one nds three values of that satisfy the above conditions:

1
= 1
2
= (1

5)/2 and
3
= (1 +

5)/2 (2.52)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 62 #8
62 Eigenvalues of Inhomogeneous Structures
For the particular case
1
= 1, the exural rigidity is dened by
D = D
0
(1 +z z
2
) (2.53)
so that the buckling load equals
w = a(z 2z
3
+z
4
) (2.54)
One notices that this function is a Duncan polynomial (1937). For the
particular case
2
= (1

5)/2, the exural rigidity is dened by

D = D
0
[1 +(1

5)z/2 (1

5)
2
z
2
/4] (2.55)
The buckling load equals
P = 3(

5 1)
2
D
0
/L
2
(2.56)
The buckling mode is
w = a{z [(1

5)
2
/2]z
3
+[(1

5)
3
/8]z
4
} (2.57)
The second set is dened as
k
2
= 20 =
_
3 d = [5
3/2
/

3]a c = (10/3)a e = (2
2
/3)a
(2.58)
This set of solutions implies the denition of the exural rigidity
D = D
0
(1
_
3z z
2
) (2.59)
with the buckling load
P = 20D
0
/L
2
(2.60)
and the buckling mode
w = d[z (10/3)z
3
(5
3/2
/

3)z
4
+(2
2
/3)z
5
] (2.61)
These coefcients have to satisfy the boundary conditions (2.19). From there
one nds a single value of that satises these equations: = (5

21)/2.
Then, the exural rigidity becomes
D = D
0
_
1 z
_
[3(5

21)/2] [5

21)/2]z
2
_
(2.62)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 63 #9
Unusual Closed-Form Solutions in Column Buckling 63
The buckling mode becomes
w = a
_
z [5(5

21)/3]z
3
[5(5

21)
3/2
/2

6]z
4
+[(5

21)
2
/6]z
5
_
(2.63)
Case 3: r = 1 z
2
In this last part, one has
D = D
0
(1 z
2
) (2.64)
We look for the buckling mode in the form of a fth-order polynomial:
w = az +bz
2
+cz
3
+dz
4
+ez
5
(2.65)
After simplication, the boundary conditions (2.19) lead to the conditions
b = 0 a +c +d +e = 0 (3c +6d +10e)(1 ) = 0 (2.66)
The solution of the governing differential equation, Eq. (2.13), by
Mathematica

leads to two different sets of solutions, which correspond to

different buckling loads and modes. The rst set is given by
k
2
= 6 c = a d = e = 0 (2.67)
This set of solutions implies the following denition of the buckling load:
P = 6D
0
/L
2
(2.68)
The buckling mode is a cubic polynomial,
w = a(z z
3
) (2.69)
These coefcients have to satisfy the boundary conditions (2.39). Thus, one
nds a single value of that satises these equations: = 1. Therefore, the
exural rigidity is dened by
D = D
0
(1 z
2
) (2.70)
The buckling load is given by
P = 6D
0
/L
2
(2.71)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 64 #10
64 Eigenvalues of Inhomogeneous Structures
The buckling mode reads as
w = a(z z
3
) (2.72)
The second set is given by
k
2
= 20 c = (10/3)a e = (7
2
/3)a d = 0 (2.73)
This set of solutions implies the following denition of the buckling load:
P = 20D
0
/L
2
(2.74)
whereas the buckling mode is
w = a[z (10/3)z
3
+(7
2
/3)z
5
] (2.75)
These coefcients have to satisfy the boundary conditions (2.19). This allows
us to nd two values of that satisfy these equations: =
3
7
or = 1. For the
particular case = 1, the exural rigidity is dened by
D = D
0
(1 z
2
) (2.76)
so the buckling load equals
P = 20D
0
/L
2
(2.77)
The buckling mode is
w = a
_
z
_
10
3
_
z
3
+
_
7
3
_
z
5
_
(2.78)
It is noticeable that this exural rigidity is the same as that for the rst set of
the third case (Eqs. (2.78)(2.80)), but a greater buckling load has been found
implying that one determines the higher buckling loads by increasing the
degree of freedom of the displacement in this case. For the particular case
= 3/7, the exural rigidity is dened by Eq. (2.6), so that the expression for
the buckling load in Eq. (2.82) reduces to that in Eq. (2.9), and the buckling
mode is
w = a
_
z
_
10
7
_
z
3
+
_
3
7
_
z
5
_
(2.79)
By choosing a = 7A, the displacement reduces to Eq. (2.8). This result and the
buckling load match those of Duncan (1937).
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 65 #11

65
2.1.4 Concluding Remarks
In this section Engessers (1893) example is rst generalized to a family of
columns with a variable moment of inertia. Using the proposed approach,
one also determines new closed-form solutions of columns with variable ex-
ural rigidity, including the generalization of Duncans (1937) solution. Then,
a design criterion is discussed, so that the buckling load is required to exceed
any prescribed value. Note that if one uses polynomials of higher degree,
more degrees of freedom are allowed for the displacement. In such circum-
stances, the method leads to higher buckling loads. It appears remarkable
that the closed-form solutions obtained are simpler than the exact solutions
for many problems involving uniform columns. This unusual property will
re-appear again and again throughout the monograph.
2.2 Inverse Buckling Problem for Inhomogeneous Columns
2.2.1 Introductory Remarks
The exact solutions for the buckling load of uniform beams are treated in
almost any textbook on the mechanics of solids. Exact solutions for non-
uniform columns is the subject of several works (see, e.g., the classical
of Bessel or Lommel functions, or some other, elementary or transcendental
functions. As far as the closed-form solutions are concerned the results are
much more restricted. All existing closed-form solutions are apparently lis-
ted, along with new solutions, in the recent study by Elishakoff and Rollot
(1999) and in Section 2.1.
This sectionis devotedtoobtainingadditional closed-formsolutions, which
are posed as inverse problems. The formulation of the problem is as follows:
Find the polynomial distribution of the Young modulus E(x) of an inhomo-
geneous columnof uniformcross-sectionwithspeciedboundaryconditions,
so that the buckling mode will be a pre-selected polynomial function. It turns
out that this seemingly simple formulation allows us to derive new closed-
form solutions. The solutions obtained appear to be of much importance, since
once the technology to construct columns with given variation of the modu-
lus of elasticity is available, any pre-selected buckling load can be achieved
for the appropriate design of the structure.
2.2.2 Formulation of the Problem
The differential equation that governs the buckling of the column under an
axial load P, reads
d
2
dx
2

D(x)
d
2
w
dx
2

+P
d
2
w
dx
2
= 0 (2.80)
2005 by Issac Elishakoff
textbookbyTimoshenkoandGere, 1961). These solutions are derivedinterms
Unusual Closed-Form Solutions in Column Buckling
EOIS: 2892_chap02 2004/9/28 19:18 page 66 #12
66 Eigenvalues of Inhomogeneous Structures
where w(x) is the transverse dispacement, D(x) = E(x)I(x) is the exural
rigidity, E(x) is the modulus of elasticity, I(x) the moment of inertia, and x
the axial coordinate. We consider four sets of boundary conditions. For the
column that is pinned at both its ends the simplest polynomial that satises
the boundary conditions at x = 0 and x = L, with L being the length of the
column, reads
() = 2
3
+
4
= x/L (2.81)
We pose the following question: Is there a column with polynomial vari-
ation of E() that possesses the function in Eq. (2.81) as its fundamental
buckling mode? Indeed, if the sought for solution exists, it corresponds to
the fundamental buckling load, since () in Eq. (2.81) does not have internal
nodes. This problem differs from the direct buckling problem, which presup-
of the mode w(x) and the buckling load P. Here, we are looking for the cause,
i.e., the distributionof the exural rigidity, knowingthe effect, i.e., the buckling
mode.
We are looking for the exural rigidity D() represented as follows:
D() = b
0
+b
1
+b
2

2
(2.82)
where b
0
, b
1
and b
2
are the constants to be determined. The inverse prob-
lems may have no solution, multiple solutions or a unique solution. It
turns out that in the case under study, an innite number of solutions
exists for reconstructing the column, which possess the function in Eq. (2.81)
as their buckling mode. We now consider the different sets of boundary
conditions.
2.2.3 Column Pinned at Both Ends
Using the non-dimensional axial coordinate , dened in Eq. (2.81), the
governing equation, Eq. (2.80), reduces to
d
2
d
2

D()
d
2

d
2

+PL
2
d
2

d
2
= 0 (2.83)
With the buckling mode postulated in Eq. (2.81) we have for the term in
Eq. (2.83)
PL
2
d
2

d
2
= qL
2
(12 +12
2
) (2.84)
2005 by Issac Elishakoff
poses the knowledge of the exural rigidity Dand requires the determination
EOIS: 2892_chap02 2004/9/28 19:18 page 67 #13
Unusual Closed-Form Solutions in Column Buckling 67
whereas the rst differential expression in Eq. (2.84) reads as
d
2
d
2
_
D()
d
2

d
2
_
= 12[2(b
1
b
0
) +6(b
2
b
1
) 12b
2

2
] (2.85)
The sum of the expressions in the right-hand sides in Eqs. (2.84) and (2.85)
must vanish, due to Eq. (2.83). Since the above sum must equal zero
identically, for any value of , we get the following expressions:
2(b
1
b
0
) = 0 (2.86)
72(b
2
b
1
) 12PL
2
= 0 (2.87)
144b
2
+12PL
2
= 0 (2.88)
The solution of Eq. (2.88) yields
P = 12b
2
/L
2
(2.89)
In order for the load P to remain compressive, it must be positive. We con-
clude, therefore, that the coefcient b
2
must be negative. Equations (2.86) and
(2.87) lead, together with Eq. (2.89), to
b
0
= b
1
= b
2
(2.90)
Thus, the exural rigidity is dened up to the coefcient b
2
:
D() = (1 +
2
)b
2
(2.91)
We already established that b
2
must take a negative value. Hence Eq. (2.91)
can be rewritten as
D() = (1 +
2
)|b
2
| (2.92)
This function is in agreement with the physical realizability condition,
namely, with the requirement of non-negativity of the the function D() in
We have thus found the function D() that corresponds to the postulated
buckling mode in Eq. (2.90). Since D() depends on the arbitrary negative
constant b
2
, we conclude that the solution of the posed problem is innity in
the class of polynomially varying exural rigidities. Note that Eq. (2.97) coin-
cides with Eq. (2.56) in the paper by Elishakoff and Rollot (1999). It pertains
to the column that is pinned at both its ends. We will show that Eq. (2.97) is
valid for two other sets of boundary conditions.
2005 by Issac Elishakoff
the interval [0; 1]. It is depicted in Figure 2.1.
EOIS: 2892_chap02 2004/9/28 19:18 page 68 #14
68 Eigenvalues of Inhomogeneous Structures
1.24
1.22
1.2
1.18
1.16
1.14
1.12
1.1
D
(

)
/

b
2

1.08
1.06
1.04
1.02
1
0 0.2 0.4 0.6 0.8 1

FIGURE 2.1
Variation of D()/|b
2
|
2.2.4 Column Clamped at Both Ends
The boundary conditions
w = 0 dw/d = 0 at = 0 = 1 (2.93)
are satised for the following polynomial function:
() =
2
2
3
+
4
(2.94)
We are interested in establishing if this polynomial function may serve as
a buckling shape of any inhomogeneous column. The expression for PL
2

reads, with primes denoting differentiation with respect to , as follows:

PL
2

= qL
2
(2 12 +12
2
) (2.95)
whereas the expression for (D

is
(D

= 2(2b
2
12b
1
+12b
0
) +6(12b
2
+12b
1
) +144b
2

2
(2.96)
We demand that the sum of the expressions in Eqs. (2.95) and (2.96) vanish
for any . This requirement leads to the following three equations:
2(2b
2
12b
1
+12b
0
) +2PL
2
= 0 (2.97)
6(12b
2
+12b
1
) 12PL
2
= 0 (2.98)
144b
2
+12PL
2
= 0 (2.99)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 69 #15
Unusual Closed-Form Solutions in Column Buckling 69
From Eq. (2.99) we get the buckling load
P = 12b
2
/L
2
(2.100)
which, remarkably, coincides with Eq. (2.89). Equations (2.97) and (2.98)
lead to
b
0
= b
2
/6 b
1
= b
2
(2.101)
Hence, we obtain the sought for variation of the exural rigidity
D() =
_
1
6
+
2
_
|b
2
| (2.102)
which takes a positive value throughout the columns axis, [0, 1], as seen
from Figure 2.2.
2.2.5 Column Clamped at One End and Pinned at the Other
The boundary conditions read
w = 0 D()
d
2
w
d
2
= 0 at = 0
w = 0
dw
d
= 0 at = 1
(2.103)
D
(

)
/

b
2

0.18
0.2
0.22
0.24
0.26
0.28
0.3
0.32
0.34
0.36
0.38
0.4
0.42
0 0.2 0.4 0.6 0.8 1

FIGURE 2.2
Variation of D()/|b
2
|
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 70 #16
70 Eigenvalues of Inhomogeneous Structures
The boundary conditions are satised by the following polynomial function:
() = 3
2
5
3
+2
4
(2.104)
Substitution of this expression into the governing differential equation, in
conjunction with the postulated expression for the exural rigidity results in
[2(6b
2
30b
1
+24b
0
) +6(30b
2
+24b
1
) +288b
2

2
]
+PL
2
(6 30 +24
2
) = 0 (2.105)
Since Eq. (2.105) is valid for any , we get the following three equations:
from
0
: 6b
2
18b
1
+24b
0
= 0 (2.106)
from
1
: 6(30b
2
+24b
1
) 30PL
2
= 0 (2.107)
from
2
: 288b
2
+24PL
2
= 0 (2.108)
We arrive at three equations for the four unknowns: b
0
, b
1
, b
2
one of the parameters to be arbitrary, namely b
2
. Then, Eq. (2.108) yields the
same buckling load as in Eqs. (2.89) and (2.100):
P = 12b
2
/L
2
(2.109)
Equations (2.106)(2.108) yield the following interrelation between the coef-
cients describing the exural rigidity variation:
b
0
=
5
16
b
2
b
1
=
5
4
b
2
(2.110)
Substituting into Eq. (2.82) results in the variation of the exural rigidity:
D() =
_
5
16
+
5
4

2
_
|b
2
| (2.111)
whichis a positive functionwithinthe lengthof the column. The functionD()
conditions were studiedby Elishakoff (2000a). These are not reproducedhere,
to save space.
2.2.6 Concluding Remarks
The following conclusions appear to be relevant:
1. Inverse buckling problems with postulated polynomial buckling modes,
as given in Eqs. (2.81), (2.94) or (2.104) for corresponding boundary condi-
tions have closed-form solutions; namely, the variations of exural rigidity
2005 by Issac Elishakoff
for this case is depicted in Figure 2.3. Note that two other sets of boundary
andP. We choose
EOIS: 2892_chap02 2004/9/28 19:18 page 71 #17
Unusual Closed-Form Solutions in Column Buckling 71
D
(

)
/

b
2

0.4
0 0.2 0.4 0.6 0.8 1
0.5
0.6
0.7

FIGURE 2.3
Variation of D()/|b
2
|
corresponding to the above mode shapes are given in Eqs. (2.92), (2.102)
and (2.111), respectively.
2. For three sets of boundary conditions, the fundamental buckling load
is given by the same expression (see Eqs. (2.89), (2.100) and (2.109)). This
conclusionmay appear to be a paradoxical one at the rst glance. To resolve it,
let us consider the correspondingcase of homogeneous anduniformcolumns.
The fundamental buckling loads are
P
PP
=

2
D
1
L
2
P
CC
=
4
2
D
2
L
2
P
PC
=
4.493
2
D
3
L
2

2
2
D
3
L
2
(2.112)
where D
1
is the exural rigidity of the column that is pinned at both its ends,
D
2
corresponds tothecolumnthat is clampedat bothends andD
3
is associated
with the column that is pinned at x = 0 and clamped at x = L. These three
columns possess the same cross-sections and have the same lengths. Now, if
D
1
= 4D
2
2D
3
(2.113)
then all three columns have coalescing buckling loads. This implies that only
those columns with different exural rigidities if they are under different
boundary conditions, but have the same lengths and cross-sections may
share the same fundamental buckling load. The same phenomenon takes
place in the case of our study: the columns with three different sets of bound-
ary conditions share the same fundamental buckling load, since their moduli
of elasticity are different.
3. Still, it appears to be intriguing that the search for the solution of the
inverse buckling problem in the class of polynomial functions leads to the
coincidence of buckling loads for pinnedpinned, clampedclamped and
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 72 #18
72 Eigenvalues of Inhomogeneous Structures
pinnedclamped boundary conditions. Here, we reported the case when the
fundamental buckling loads are sharedby columns under different boundary
problems].
To compare the results for the buckling loads, let us calculate the average
exural rigidity in each of the three cases. The average exural rigidity is
dened as
D
av
=
_
1
0
D() d (2.114)
Thus, for the pinned column
D
av,SS
=
7
6
|b
2
| (2.115)
For the clamped column
D
av,CC
=
1
3
|b
2
| (2.116)
For the column that is pinned at one end and clamped at the other
D
av,SC
=
29
48
|b
2
| (2.117)
Thus, the buckling loads can be put in the following forms, by rst expressing
|b
2
| via D
av
in Eqs. (2.115)(2.117):
P
PP
= 12|b
2
|/L
2
=
72
7
D
av
/L
2
(2.118)
P
CC
= 12|b
2
|/L
2
= 36D
av
/L
2
(2.119)
P
PC
= 12|b
2
|/L
2
=
576
29
D
av
/L
2
(2.120)
If the average exural rigidities of these columns are chosen to be the same,
then, the buckling loads of the inhomogeneous columns are in the proportion
72
7
: 36 :
576
29
(2.121)
or 1 : 3.5 : 1.93, vs. the corresponding proportion 1 : 4 : (2) for the uniform
columns.
4. The calculations of the buckling loads have been performed for
columns possessing the exural rigidity variations given in Eqs. (2.92), (2.102)
and (2.111), respectively. The exural rigidity in Eq. (2.92) is denoted as
D
PP
(). The exural rigidity function in Eq. (2.102) is designated as D
CC
(),
whereas the exural rigidity variation in Eq. (2.111) is denoted as D
CP
().
For each exural rigidity function three different sets of boundary condi-
tions have been examined. For the pinned column with the exural rigidity
2005 by Issac Elishakoff
conditions [see also Gottlieb (1989) Gladwell and Morassi (1995) for vibration
EOIS: 2892_chap02 2004/9/28 19:18 page 73 #19
Unusual Closed-Form Solutions in Column Buckling 73
D
PP
(), the closed-form solution is reported in Eq. (2.89) as 12|b
2
|/L
2
. For
the column with exural rigidity D
PP
, but under boundary conditions dif-
fering from both the pinned ends, the approximate values of the buckling
loads are furnished by many approximate methods. Due to its simplicity, we
employed the BoobnovGalerkin method here. For consistency, the buckling
loadof the pinnedpinnedinhomogeneous column was also evaluatedby the
BoobnovGalerkin method, with the comparison function sin(), yielding
12.0145, the value that is just 0.12%above the closed-formexpression. For the
column with exural rigidity D
PP
() but with clampedpinned boundary
conditions, the buckling mode of the associated uniformcolumn was utilized
as the comparisonfunction. The bucklingloadequals 23.7181|b
2
|/L
2
, whereas
for the clampedclamped beam the buckling load derived by using the func-
tion sin
2
() as the comparison function, equals 45.5582|b
2
|/L
2
. As is seen,
for the columns with identical exural rigidity, but under different boundary
conditions, the buckling loads differ, as it should be. An analogous conclusion
is reached for the columns with exural rigidity equal to either D
CP
() or
D
CC
(). The results are summarized in Table 2.1.
5. For uniform columns the polynomial expressions of the buckling
mode are usually utilized to facilitate the approximate solutions, via the
BoobnovGalerkin or the RayleighRitz methods. For example, Chajes
(1974) uses the function w(x) = xL
3
3x
3
L + 2x
4
, as a comparison
function in the BoobnovGalerkin method for the column that is pinned
at x = 0 and clamped at x = L. It is interesting that the same
polynomial function (coincident with Eq. (2.104) once is replaced by
1 ) turned out to be an exact buckling mode of the inhomogeneous
column.
TABLE 2.1
Comparison of Exact and Approximate Solutions
P
cr
L
2
/|b
2
|
Flexural rigidity
Boundary
conditions
Comparison function in
the BoobnovGalerkin
method
Exact
solution
Approximate
solution Error (%)
D
PP
() =
(1 +
2
)|b
2
| PP sin() 12 12.0145 0.12
CP
1 + +cos(4.493)
sin(4.493)/4.493
23.7181
CC sin
2
() 45.5582
D
CP
() =
(
5
16
+
5
4

2
)|b
2
| PP sin() 6.46289
CP
1 + +cos(4.493)
sin(4.493)/4.493
12 12.363 3.03
CC sin
2
() 23.3515
D
CC
() =
(
1
6
+
2
)|b
2
| PP sin() 3.78987
CP
1 + +cos(4.493)
sin(4.493)/4.493
6.89555
CC sin
2
() 12 12.6595 5.50
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 74 #20
74 Eigenvalues of Inhomogeneous Structures
Likewise, in his collection of problems, Volmir (1984) posed the ques-
tion as to whether the Rayleighs method as well as the BoobnovGalerkin
method could be used for the approximate estimation of the buckling load
of a uniform column, pinned at both its ends, by utilizing the trial function
w(x) = xL
3
2x
3
L + x
4
. In our case, the same buckling mode (coincid-
ent with Eq. (2.81)) serves as an exact expression for the inhomogeneous
column.
6. The question arises as to the generality of the proposed method. In
order for the method to be acceptable it should lead to a positive buckling
load, corresponding to the case of a compressive force, and a modulus of
elasticity that is a positive function. The formulation of some general condi-
tions upon fulllment of which the problem is amenable to a solution, is of
interest.
7. The buckling loads turn out to depend on only a single coefcient b
2
,
once the function of modulus variation is obtained. By a suitable choice of
b
2
, the buckling load can be made arbitrarily large. This conclusion is valid
within the context of elastic buckling that was presupposed in this study.
8. The solution of the buckling problems of uniform columns leads to
irrational values of the buckling loads; for example, the buckling load
of the pinned uniform column is written in terms of
2
(see Eq. (2.112)).
So is the buckling load, found by Euler (1759) in the case of the
variable exural rigidity column, where the buckling load is
2
(D
0
/L
2
)a
2

(a +b)
2
for the column with the exural rigidity D() = D
0
(a + b)
4
. Here,
in the case of inhomogeneous and/or non-uniform columns, the solution turns
out to be in terms of rational numbers. Other solutions of this kind have been
9. Note that the exural rigidity variation of the clampedclamped
column in Eq. (2.102) can be obtained from that of the pinnedpinned
column in Eq. (2.92), by simple translation (Cali, 2000) of the
curve.
2.3.1 Introductory Remarks
The motivation for this section is as follows. In recent decades numerous stud-
ies have been published on the so-called stochastic nite element method,
which deals with stochastically inhomogeneous structures by the nite
element method. Among various methods, the authors utilize the perturb-
ation techniques of various orders, spectral nite element method, Neumann
expansions, etc. For a review of some of these studies the reader can consult
the review papers by Shinozuka and Yamazaki (1988) and Elishakoff et al.
(1995). One of the techniques discussed by them is the nite element method
combined with the Monte Carlo method. In the context of this method, this
2005 by Issac Elishakoff
reported by Elishakoff and Rollot (1999) (see Section 2.1).
2.3 Closed-Form Solution for the Generalized Euler Problem
EOIS: 2892_chap02 2004/9/28 19:18 page 75 #21
Unusual Closed-Form Solutions in Column Buckling 75
stochastic variables or elds are rst simulated. Thus, a stochastic ensemble
of a large amount of deterministic realizations is created. Each of the struc-
tures is solved by a deterministic FEM, and results are statistically analyzed.
In order to elucidate the efcacy of such a technique, the closed-formdeterm-
inistic problems will be of great help. However, even in the deterministic
setting, the number of exact solutions, let alone the closed-form ones, is
extremely limited [for exact solution in the stochastic setting one can consult
the paper by Ellishakoff et al. (1995)]. It is important, therefore, to invest effort
togenerate closed-formsolutions inthe deterministic setting. Obviously, such
solutions are interesting in their own right, for the community of research-
ers and engineers who occupy themselves with deterministic theoretical and
applied mechanics.
The buckling of columns under their own weight is a problem that goes
back to Leonhard Euler (1778), who looked for its solution in terms of power
series with respect to the axial coordinate. For the clampedfree column of
uniform exural rigidity, the exact solution is given in terms of Bessel func-
tions (Timoshenko and Gere, 1961) with the attendant critical load intensity
q
cr
equal to
q
CF
= 7.837D/L
3
(2.122)
where D is the exural rigidity, and L the length.
The column under its own weight was considered by Euler in three famous
papers in 1778. For the fascinating and instructive story of these three papers
the reader is directed to a most instructive book by Panovko and Gubanova
(1965) on numerous fallacies and mistakes in applied mechanics.
In the rst paper, Euler (1778a) arrived at a conclusion that a pinned
column under its own weight does not buckle. In the second paper (1778b)
he expressed a suspicion about the above conclusion, which he called a para-
doxical one. In the third paper, Euler (1778c) found a mistake made in the
rst paper, and concluded that such a column is susceptible to buckling.
Unfortunately, however, he made a numerical mistake and instead of the
rst buckling load calculated the second one. Acorrect numerical value was
calculated by Dinnick (1912) some 134 years after Euler:
q
PP
= 18.6D/L
3
(2.123)
Engelhardt (1954) reported results for the column that is clamped at one end
and pinned at the other,
q
CS
= 52.50D/L
3
(2.124)
as well as for the column that is clamped at both ends,
q
CC
= 74.6D/L
3
(2.125)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 76 #22
76 Eigenvalues of Inhomogeneous Structures
Other exact solutions were furnished by Greenhill (1881), Willers (1941)
and Frisch-Fay (1966). Dinnik (1912), Engelhardt (1954) and Frisch-Fay
(1966) utilized the power series method, whereas Willers (1941) presented
a solution in terms of Bessel and Lommel functions. The solution in terms
of Bessel and Lommel functions was communicated to this writer by
Pger (1975).
The exact nite element method in the context of the buckling was
developed by Eisenberger (1991ad). Both the columns, exural rigidity
and the axial load were considered to vary as polynomial functions. The
buckling load was determined as the load that makes the determinant of
the stiffness matrix zero (Eisenberger, 1991ad). The latter mentions that in
his work exact buckling loads (up to the accuracy of the computer) are
given and that only one element is need for the solution. Waszczyszyn
and Pieczara (1990) also reported several results via the exact nite element
method.
In the present section we obtain the closed-form solutions for the general-
ized Eulers problem, namely, columns with variable exural rigidity under
their own weight, following Elishakoff (2000a). It appears to be remarkable
that for the columnwith uniformexural rigiditytranscendental equations are
obtained for the buckling load determination, whereas for the more complic-
atedcase of the columnwith variable exural rigidity, the closed-formsolution
turns out to be available. This seemingly curious fact is perhaps explainable
by the observation that variability in exural rigidity introduces additional
degrees of freedom that enable us to obtain solutions in terms of polynomial
functions.
2.3.2 Formulation of the Problem
The differential equation that governs the buckling of the non-uniform
column under its own weight reads as
d
2
dx
2
_
D(x)
d
2
w
dx
2
_
+
d
dx
_
q(L x)
dw
dx
_
= 0 (2.126)
where w(x) is the buckling mode, D(x) = E(x)I(x) is the exural rigidity,
E the modulus of elasticity, I the moment of inertia, q the intensity of the
buckling load, L the length and x the axial coordinate.
The differential equation (2.134) is accompanied by appropriate bound-
ary conditions. For the column that is pinned at its ends, the boundary
conditions are
w = 0
d
2
w
dx
2
= 0 at x = 0 x = L (2.127)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 77 #23
Unusual Closed-Form Solutions in Column Buckling 77
For simplicity, we introduce the non-dimensional coordinate = x/L. The
differential equation (2.126) then becomes
d
2
d
2
_
D()
d
2
w
d
2
_
+
d
d
_
qL
3
(1 )
dw
d
_
= 0 (2.128)
We are looking for a polynomial representation of the exural rigidity D()
in Eq. (2.134). Since the buckling mode w() is also required sought as a
polynomial, the left-hand side of Eq. (2.134) turns out to be a polynomial in
terms of the non-dimensional axial coordinate . To achieve the same order of
the polynomial expressions inboththe rst andthe secondterminEq. (2.134),
the exural rigidity must be a third-order polynomial:
D() = b
0
+b
1
+b
2

2
+b
3

3
(2.129)
The buckling mode, denoted by (), is chosen as a polynomial of fourth
degree that satises all boundary conditions and can be written as
() = 2
3
+
4
(2.130)
The rst term in the differential equation (2.128) becomes upon substitution
of Eq. (2.129)
d
2
d
2
_
D()
d
2

d
2
_
= 12[2(b
1
b
0
) +6(b
2
b
1
) +12
2
(b
3
b
2
) 20b
3

3
]
(2.131)
whereas the second term in Eq. (2.128) becomes
d
d
_
qL
3
(1 )
d
d
_
= qL
3
(1 12 +30
2
16
3
) (2.132)
The sum of the expressions in Eqs. (2.131) and (2.132) must vanish for every
. This leads to the following conditions:
24(b
1
b
0
) qL
3
= 0
72(b
2
b
1
) 12qL
3
= 0
144(b
3
b
2
) +30qL
3
= 0
240b
3
16qL
3
= 0
(2.133)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 78 #24
78 Eigenvalues of Inhomogeneous Structures
We arrive at four equations with ve unknowns, which is solvable up to an
arbitrary constant, which is taken here as the coefcient b
3
. The buckling
intensity equals
q
cr
= 15b
3
/L
3
(2.134)
The other coefcients are:
b
0
= b
3
b
1
=
3
8
b
3
b
2
=
17
8
b
3
(2.135)
so that the attendant exural rigidity reads as
D() =
_
1 +
3
8

17
8

2
+
3
_
b
3
(2.136)
Figure 2.4 depicts D()/b
3
. It is instructive to calculate the average exural
rigidity
D
av
=
_
1
0
D() d =
35
48
b
3
(2.137)
The buckling intensity of the column with the average exural rigidity is
obtained by substituting D
av
into Eq. (2.123), once it is expressed in terms of
b
3
in Eq. (2.137):
q
cr
= 18.6D
av
/L
3
= 13.56b
3
/L
3
(2.138)
D
(

)
/

b
3

0.3
0 0.2 0.4 0.6 0.8 1
0.4
0.5
0.6
0.7
0.8
0.9
1

FIGURE 2.4
Variation of D()/b
3
for the column pinned at both its ends
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 79 #25
Unusual Closed-Form Solutions in Column Buckling 79
which turns out to be 1.10 times smaller than its counterpart in Eq. (2.134) for
the variable exural rigidity column.
2.3.3 Column Clamped at Both Ends
The postulated buckling mode reads
() =
2
2
3
+
4
(2.139)
The equations that are counterparts of Eq. (2.133) are
2(12b
0
12b
1
+2b
2
) +2qL
3
= 0
6(12b
1
12b
2
+2b
3
) 16qL
3
= 0
12(12b
2
12b
3
) +30qL
3
= 0
240b
3
16qL
3
= 0
(2.140)
resulting in the buckling load
q
cr
= 15b
3
/L
3
(2.141)
and the exural rigidity
D() =
_
7
48
+
25
24

17
8

2
+
3
_
b
3
(2.142)
ence of D()/b
3
on . As in the previous case we calculate the average exural
rigidity, which equals
D
av
=
5
24
b
3
(2.143)
The column with constant exural rigidity equal to D
av
has a buckling load
obtained from Eq. (2.125):
q
cr
= 74.6D
av
/L
3
= 15.54b
3
/L
3
(2.144)
which is about 4% larger than the column with variable exural rigidity.
2.3.4 Column Pinned at One End and Clamped at the Other
The buckling mode is stipulated to be as follows:
() = 3
3
+2
4
(2.145)
2005 by Issac Elishakoff
Note that Eqs. (2.141) and (2.134) coincide. Figure 2.5 portrays the depend-
EOIS: 2892_chap02 2004/9/28 19:18 page 80 #26
80 Eigenvalues of Inhomogeneous Structures
D
(

)
/

b
3

0.08
0.06
0 0.2 0.4 0.6 0.8 1
0.1
0.12
0.14
0.16
0.18
0.2
0.22
0.24
0.26
0.28

FIGURE 2.5
Variation of D()/b
3
for the column clamped at both its ends
The set of equations in terms of b
0
, b
1
, b
2
, b
3
and q reads as follows:
2(24b
0
18b
1
) qL
3
= 0
6(24b
1
18b
2
) 18qL
3
= 0
12(24b
2
18b
3
) +51qL
3
= 0
480b
3
32qL
3
= 0
(2.146)
The buckling intensity again equals
q
cr
= 15b
3
/L
3
(2.147)
whereas the exural rigidity variation associated with the buckling mode
(2.145) and buckling intensity (2.147) is obtained as follows:
D() =
_
331
512
+
57
128

61
32

2
+
3
_
b
3
(2.148)
3
In order to contrast the results for the equivalent column with average
exural rigidity,
D
av
=
743
1536
b
3
(2.149)
we substitute D
av
into Eq. (2.124) to arrive at
q
cr
= 52.50D
av
/L
3
= 25.39b
3
/L
3
(2.150)
2005 by Issac Elishakoff
Figure 2.6 shows a plot of the ratio D()/b .
EOIS: 2892_chap02 2004/9/28 19:18 page 81 #27
Unusual Closed-Form Solutions in Column Buckling 81
D
(

)
/

b
3

0.2
0 0.2 0.4 0.6 0.8 1
0.3
0.4
0.5
0.6

FIGURE 2.6
Variation of D()/b
3
for the pinnedclamped column
which is about 70% larger than its counterpart for the variable exural
rigidity.
2.3.5 Column Clamped at One End and Free at the Other
The buckling mode postulated for this case reads
() = 6
2
4
3
+
4
(2.151)
Substitution of this expression into the differential equation (2.126) yields
2(12b
0
24b
1
+12b
2
) +6(12b
3
24b
2
+12b
1
) +12(12b
2
24b
3
)
2
+360b
3

3
+qL
3
(12 48 +48
2
16
3
) = 0 (2.152)
Since this equation must constitute an indentity for any value of , we get the
following equations:
2b
0
4b
1
+2b
2
+qL
3
= 0
3b
3
6b
2
+3b
1
2qL
3
= 0 (2.153)
3b
2
6b
3
+qL
3
= 0
240b
3
16pL
3
= 0
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 82 #28
82 Eigenvalues of Inhomogeneous Structures
The last equation yields the expression for the critical parameter
q
cr
= 15b
3
/L
3
(2.154)
where b
3
is an arbitrary constant. We also obtain
b
0
=
3
2
b
3
b
1
= 3b
3
b
2
= 3b
3
(2.155)
leading to the following expression for the columns exural rigidity:
D() =
_
3
2
+3 3
2
+
3
_
b
3
(2.156)
The average exural rigidity is
D
av
=
9
4
b
3
(2.157)
Substituting into Eq. (2.122) we get
q
cr
= 7.837D
av
/L
3
= 17.63b
3
/L
3
(2.158)
whichis about 17%larger thanits counterpart for the variable exural rigidity.
Figure 2.7 depicts the variation of D()/b
3
in terms of . The value of
D() at the origin and at the columns end are 1.5b
3
and 2.5b
3
, respect-
ively. The derivative of the function D() vanishes at = 1, implying
D
(

)
/

b
3

0.8
0 0.2 0.4 0.6 0.8 1
1
1.2
1.4
1.6

FIGURE 2.7
Variation of D()/b
3
for the clampedfree column
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 83 #29
Unusual Closed-Form Solutions in Column Buckling 83
that the exural rigidity is a positive-valued function, so long as b
3
is
positive. Positiveness of b
3
stems also from the expression for the buck-
ling intensity in Eq. (2.154). Thus, these two reasons for positivity of the
exural rigidity are compatible with each other. Note that two other sets
of boundary conditions were studied by Elishakoff (2000f) and are not
included here.
2.3.6 Concluding Remarks
Bulson (1970) noted in the preface to his monograph: In many technical
libraries the supports of bookshelves containing works on the stability of
structure are in danger of buckling under the weight of literature. Can
another work on this subject be justied?. It appears that the answer
to this question is in the afrmative. Indeed, in the present work, we
present the closed-form solutions for buckling of columns under their own
weight, since the topic was initiated by Euler in 1778. In contrast to Euler
and subsequent authors, we considered the column of variable exural
rigidity and derived the closed-form solutions for four sets of bound-
ary conditions. The most signicant conclusion derived in this section is
the fact that, irrespective of boundary conditions, the buckling load in
all four cases is given by the same analytical expression. Naturally, in
these cases, the exural rigidity variations along the columns axis are
different.
In order to compare the buckling loads associated with the four
uncovered columns of specic exural rigidity, given respectively in
Eqs. (2.136), (2.142), (2.148) and (2.156) we undertook the following
calculations, to elucidate the results obtained. For example, the column with
the exural rigidity given in Eq. (2.136), under pinnedpinned boundary
conditions, has the buckling load whose exact expression is q
cr
= 15b
3
/L
3
as given in Eq. (2.134). The question that arises naturally is: What are
the buckling loads of the same column under other boundary condi-
tions? At present there seem to be no exact solutions in these cases.
Hence, an evaluation of approximate estimates was conducted, based
on the BoobnovGalerkin method. The associated comparison functions
As is seen, the single-term approximation is
extremely good. For the PP column, the closed-form solution is q
cr
L
3
/b
3
=
15, whereas the approximation gives 15.02, with an attendant error
of 0.133%.
As is clearly seen from Tables 2.22.5, for each of the four exural rigid-
ity distributions the buckling loads of columns under different boundary
conditions differ from each other, as expected. The uncovered buckling
load that is represented by a simple rational expression can be used as a
benchmark solution for various numerical evaluation techniques. It is hoped
that other researchers will be inspired to derive new closed-form solutions
for more complicated and challenging problems, since even over 220 years
after Leonhard Euler the eld of buckling turns out to be inexhaustible.
2005 by Issac Elishakoff
are listed in Tables 2.22.5.
EOIS: 2892_chap02 2004/9/28 19:18 page 84 #30
84 Eigenvalues of Inhomogeneous Structures
TABLE 2.2
Comparison of Exact and Approximate Solutions
Flexural rigidity
Boundary
conditions Comparison function q
cr
L
3
/b
3
D
PP
() =
_
1 +
3
8

17
8

2
+
3
_
b
3
PP 2
3
+
4
15
sin() 5(7
2
+3)/24 15.02

1
() = sin(),

2
() = sin(2)
15.017

1
() = sin(),

2
() = sin(2)
15.001
PC
3
2

5
2

3
+
4 523
12
= 43.58

1
() = sin(

2)

2 cos(

2)
+

2(1 )
41.83

1
() as above,

2
() = sin(7.7252)
7.7252 cos(7.7252)
+7.7252(1 )
41.69

1
() and
2
() as above,

3
() = homogeneous columns
third buckling mode
41.65
CC
2
2
3
+
4 235
4
= 58.75

1
= homogeneous columns
rst buckling mode
5(28
2
3)/24 56.49

1
= as above

2
= homogeneous columns
second buckling mode
56.77

1
and
2
= as above

3
= homogeneous columns
third buckling mode
56.75
CF 6
2
4
3
+
4 47
6
= 7.83
2.4 Some Closed-Form Solutions for the Buckling of
Inhomogeneous Columns under Distributed Variable
Loading
2.4.1 Introductory Remarks
The study of buckling of columns under their own weight was pioneered
by Euler (1778ac). Subsequent contributions were made by Heim (1838),
Greenhill (1881), Jasinskii (1894) and others. Often it is referred to as the
Greenhills problemrather than Eulers problem(Love, 1944). The case when
the axial distributed load is constant was studied, in addition to the above
authors, by Dinnik (1912), Engelhardt (1954) and others. The notion of the
variable intensity of the axial load was apparently introduced by Hauger
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 85 #31
Unusual Closed-Form Solutions in Column Buckling 85
TABLE 2.3
Comparison of Exact and Approximate Solutions
Flexural rigidity
Boundary
conditions Comparison function q
cr
L
3
/b
3
D
PC
() =
_
105
512
+
153
128

72
32

2
+
3
_
b
3
PP 2
3
+
4 3351
544
= 6.16
sin()
139
2
+216
256
= 6.20

1
() = sin(),

2
() = sin(2)
6.188

1
() = sin(),

2
() = sin(2)
6.114
PC
3
2

5
2

3
+
4
15

n
= homogeneous compressed
columns nth mode shape, =
1
15.384
Two terms 15.112
CC
2
2
3
+
4 2478
128
= 19.36

1
() = sin
2
()
139
2
54
64
20.59

1
() = sin
2
(),
2
() = sin
2
(2) 19.73

i
() = sin
2
(), i = 1, 2, 3 19.54
CF 6
2
4
3
+
4 159
64
= 2.48
TABLE 2.4
Comparison with the BoobnovGalerkin Method
Flexural rigidity
Boundary
conditions Comparison function q
cr
L
3
/b
3
D
CC
() = PP 2
3
+
4 80
17
= 4.71
_
7
48
+
25
24

17
8

2
+
3
_
b
3

1
() = sin()
5(2
2
+3)
24
= 4.74[2pt]

1
() = sin(),

2
() = sin(2)
4.73

1
() = sin(),

2
() = sin(2)
4.68
PC
3
2

5
2

3
+
4 407
36
= 11.31

n
= compressed columns nth
buckling mode (one term
approximation)
11.63
Two terms 11.43
Three terms 11.30
CC
2
2
3
+
4 235
4
= 58.75

1
() = sin
2
() 15

1
() = sin
2
(),

2
() = sin
2
(2)
15.21

i
() = sin
2
(i), i = 1, 2, 3 15.078
CF 6
2
4
3
+
4 17
9
= 1.89
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 86 #32
86 Eigenvalues of Inhomogeneous Structures
TABLE 2.5
Comparison with the BoobnovGalerkin Method
Flexural rigidity
Boundary
conditions Comparison function q
cr
L
3
/b
3
D
CF
() = PP 2
3
+
4 780
17
= 45.88
_
3
2
+3 3
2
+
3
_
b
3

1
() = sin()
3(3
2
+1)
2
= 45.9

1
() = sin(),

2
() = sin(2)
41.53

1
() = sin(),

2
() = sin(2)
41.29
PC
3
2

5
2

3
+
4
120

n
= homogeneous compressed
columns nth buckling mode
(one term approximation)
131.59
Two terms 109.76
Three terms 109.00
CC
2
2
3
+
4
183

1
() = sin
2
()
3(12
2
1)
2
= 176.153

1
() = sin
2
(),

2
() = sin
2
(2)
175.82

i
() = sin
2
(i), i = 1, 2, 3 175.79
CF 6
2
4
3
+
4
15
(1966), who considered the non-conservative case as an analytical exten-
sion of the Becks (1952) column that is acted upon the so-called follower
the follower forces was rightfully criticized by Koiter (1996) due to the fact
that no direct and pure experimental verication has been provided up
to now of Becks (1952) tangentially loaded column, although this is dis-
puted by Sugiyama et al. (1999). We will limit ourselves to the consideration
of the conservative case of the polynomially varying intensity of the axial
loading.
We consider, in this study, the intensity of loading to be a polynomial
function of the non-dimensional axial coordinate
= x/L (2.159)
where x is the axial coordinate and L the length. The variation of the axial
intensity is taken as follows:
q() = q
0
+q
1
+ +q
m

m
(2.160)
2005 by Issac Elishakoff
forces (see also Leipholz and Bhalla, 1977). However, the very notion of
EOIS: 2892_chap02 2004/9/28 19:18 page 87 #33
87
where q
j
(j = 0, 1, . . . , m) are given coefcients. The case with only q
0
present,
and the rest of the coefcients vanishing, i.e., q
j
= 0 (j = 1, 2, . . . , m) was
studied by Elishakoff (2000a).
We are interested in closed-form solutions for the buckling load of the
column. To this end, it is more convenient to pose the inverse problem of
a special kind. We will postulate the prior knowledge of the buckling mode
of the structure. We pose the problem as follows: nd a special variation of the
exural rigidity, such that the buckling mode coincides with the preselected
polynomial function that satises all boundary conditions.
2.4.2 Basic Equations
The differential equation that governs the buckling of the inhomogeneous
column reads (Brush and Almroth, 1975) as follows:
d
2
dx
2
_
D(x)
d
2
w
dx
2
_

d
dx
_
N(x)
dw
dx
_
= 0 (2.161)
where D(x)is the exural rigidity, N(x)the axial force, w(x)the buckling mode
and xthe axial coordinate. With the non-dimensional coordinate , Eq. (2.169)
becomes
d
2
d
2
_
D()
d
2
w
d
2
_
L
2
d
d
_
N()
dw
d
_
= 0 (2.162)
Consider rst the column that is clamped at = 0 and free at = 1. The axial
load reads
N() =
_
L
x
q(t) dt = L
_
1

q(t) dt (2.163)
Here
q() =
m

i=0
q
i

i
(2.164)
leading to
N() = L
m

i=0
q
i
i +1
(1
i+1
) (2.165)
2005 by Issac Elishakoff
Unusual Closed-Form Solutions in Column Buckling
EOIS: 2892_chap02 2004/9/28 19:18 page 88 #34
88 Eigenvalues of Inhomogeneous Structures
We are interested in nding the critical value of q
0
, when the ratios between
the load parameters
g
i
= q
i
/q
0
i = 1, 2, . . . , m (2.166)
are specied. We restrict our considerations to the case of positive values of
the ratios g
i
.
The differential equation (2.162) reads, in conjunction with Eqs. (2.165) and
(2.166), as follows:
d
2
d
2
_
D()
d
2
w
d
2
_
+q
0
L
3
d
d
_
m

i=0
g
i
i +1
(1
i+1
)
dw
d
_
= 0 (2.167)
Moreover, g
0
= 1. The simplest polynomial function that satises the
boundary conditions of the clamped free column is given by
() = 6
2
4
3
+
4
(2.168)
We substitute () into Eq. (2.167) and demand that the resulting equation
is satised identically for every value of . This procedure leads to the set of
algebraic equations for the coefcients b
i
representing the exural rigidity:
D() =
m+3

i=0
b
i

i
(2.169)
Since the case m = 0 was considered in Section 2.3, it will not be recapitu-
lated here. We rst consider the particular case m=1 prior to turning to the
general case.
For m = 1, the result of substitution of Eqs. (2.165) and (2.166) into
Eq. (2.167) is the fourth-order polynomial, whose coefcients must vanish.
This condition results in the following set of equations:
24b
2
48b
1
+24b
0
+q
0
L
3
(12 +6g
1
) = 0
72b
3
144b
2
+72b
1
+q
0
L
3
(48 12g
1
) = 0
144b
4
288b
3
+144b
2
+q
0
L
3
(48 12g
1
) = 0
280b
4
+240b
3
+q
0
L
3
(16 +24g
1
) = 0
360b
4
10q
0
L
3
g
1
= 0
(2.170)
In Eqs. (2.170) we have ve equations for six unknowns: b
0
, b
1
, b
2
, b
3
, b
4
andq
0
.
Thus, one of the coefcients can be chosen arbitrarily. We choose b
4
as an
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 89 #35
89
arbitrary constant, and the last equation leads to the buckling load q
0,cr
:
q
0,cr
= 36b
4
/L
3
g
1
(2.171)
The coefcients dening the exural rigidity are
b
0
= (13g
1
+18)b
4
/5g
1
b
1
= 2(13g
1
+18)b
4
/5g
1
b
2
= 6(g
1
+6)b
4
/5g
1
b
3
= 4(2g
1
3)b
4
/5g
1
(2.172)
Hence, the sought for exural rigidity D() reads as follows:
D() = [(13g
1
+18)b
4
+2(13g
1
+18)b
4
6(g
1
+6)b
4

2
4(2g
1
3)b
4

3
+5g
1

4
]/5g
1
(2.173)
In order for the obtained expression for the exural rigidity to be physically
realizable, it should be non-negative for any positive value of the parameter
g
1
. To check the non-negativity of D(), let us consider the following function
f
1
(, g
1
) = 5g
1
D()/b
4
,
f
1
(, g
1
) = (13g
1
+18) +2(13g
1
+18) 6(g
1
+6)
2
4(2g
1
3)
3
+5g
1

4
(2.174)
It is convenient to treat f
1
(, g
1
) as a function of the independent variable g
1
,
while will be treated as a parameter. We get, therefore, the linear equation
whose slope f
1
/g
1
depends on :
f
1
/g
1
= 13 +26 6
2
8
3
+5
4
(2.175)
[0; 1]. We also note that at g
1
= 0, f
1
(, 0) = 18 +36 36
2
+12
3
, which
takes larger values than 18 +12
3
, for [0; 1]. Since 18 +12
2
is positive in
the same interval, we conclude that for any positive g
1
, f
1
(, g
1
) is positive.
Hence the exural rigidity D() is positive. The dependence D() is depicted
1
We now turn to the general case of the mth order polynomial for the
clampedfree column. We substitute Eqs. (2.176) and (2.177) into (2.175). The
result is the set of m + 4 linear algebraic equations for m + 5 unknowns,
namely m+4 coefcients in Eq. (2.169) for D() and the critical value of q
0
in
2005 by Issac Elishakoff
and is depicted in Figure 2.8. We conclude that the slope remains positive for
in Figure 2.9 as a function of and g .
Unusual Closed-Form Solutions in Column Buckling
EOIS: 2892_chap02 2004/9/28 19:18 page 90 #36
90 Eigenvalues of Inhomogeneous Structures
f
1
g
1
14
0 0.2 0.4 0.6 0.8 1
16
18
20
22
24
26
28
30

FIGURE 2.8
Variation of f
1
/g
1
for the clampedfree column
0
1
0.8
0.6
0.4
0.2
0
10,000
20,000
30,000
40,000
50,000
2
4
6
8
10
D
(

)
/
b
4
g
1

FIGURE 2.9
Variation of D()/b
4
Eq. (2.167):
from
0
: 24b
2
48b
1
+24b
0
+12q
0
L
3
G = 0 (2.176)
from
1
: 72b
3
144b
2
+72b
1
2q
0
L
3
(12G+12) = 0 (2.177)
from
2
: 144b
4
288b
3
+144b
2
+3q
0
L
3
(4G+12 6g
1
) = 0 (2.178)
from
i
, 3 i m+1:
12(i +2)(b
i+2
2b
i+1
+b
i
) q
0
L
3
_
12
g
i1
i
12
g
i2
i 1
+4
g
i3
i 2
_
= 0
(2.179)
from
m+2
: 12(m+4)(2b
m+3
+b
m+2
) +q
0
L
3
_
12
g
m
m+1
4
g
m1
m
_
= 0
(2.180)
from
m+3
: 12(m+5)b
m+3
4q
0
L
3
g
m
m+1
= 0 (2.181)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 91 #37
91
where
g
0
= 1 G =
m

i=0
g
i
i +1
(2.182)
It should be noted that the case m=1 does not follow from Eqs. (2.176)
(2.181). In fact, Eq. (2.176) does not hold, for it is valid for 3 i m + 1,
and when m = 1 the upper value of variation of i turns out to be smal-
ler than the lower level, set at 3. On the other hand, the case m = 2 is
included in the general formulation, for in this limiting case the upper value
of variation of i, namely m + 1 turns out to be equal to the lower value of
variation of i, namely 3. Thus, we have only one special case, associated
with m = 1, while all other cases are covered by the general formulation in
Eqs. (2.176)(2.181).
Note that the set (2.176)(2.181) is a triangular systemof equations, allowing
the determination of m+4 unknowns once one of the unknowns is taken as
an arbitrary constant. It is convenient to choose b
m+3
as such a constant, since
it appears in the last equation in the set (2.176)(2.181).
The critical value of the load intensity q
0
is obtained as follows:
q
0,cr
= 3(m+1)(m+5)b
m+3
/g
m
L
3
(2.183)
Note that although the case m = 1 is a special one, the buckling intensity
associated with it follows fromEq. (2.25) by formally substituting in it m = 1.
The coefcients dening D() read as follows:
b
m+2
=
_
(m+1)(m+5)g
m1
(m+4)mg
m

m+7
m+4
_
b
m+3
(2.184)
b
i
= (m+1)(m+5)
_
g
i3
i
2
4

3g
i2
(i 1)(i +2)
+
3g
i1
i(i +2)
_
b
m+3
g
m
+2b
i+1
b
i+2
for 3 i m+1 (2.185)
b
2
= (m+1)(m+5)(3g
1
2G6)b
m+3
/8g
m
b
4
+2b
3
(2.186)
b
1
= (m+1)(m+5)(3g
1
+2G2)b
m+3
/4g
m
2b
4
+3b
3
(2.187)
b
0
= (m+1)(m+5)(9g
1
2G2)b
m+3
/8g
m
3b
4
+4b
3
(2.188)
We observe that the critical bucklingloaddepends onthe coefcient g
m
only.
This does not mean that the other g
i
coefcients do not affect the columns
behavior, for all values of g
i
dene the exural rigidity of the column. This
conclusion may appear paradoxical, at rst glance, since g
i
are the ratios
of the load coefcients, and it may seem that they should not inuence the
exural rigidity. The resolution of this predicament is that the closed-form
solutions are available only for such specic cases where the exural rigidity
andthe other parameters deningthe system, are intimatelyrelated. Since the
coefcients g
i
were xed at the outset of the solution of the inverse problem,
2005 by Issac Elishakoff
Unusual Closed-Form Solutions in Column Buckling
EOIS: 2892_chap02 2004/9/28 19:18 page 92 #38
92 Eigenvalues of Inhomogeneous Structures
naturally, the sought for variation of D() may, in some manner, depend on
the input parameters g
i
.
2.4.3 Column Pinned at Both Ends
The mode shape is taken as
() = 2
3
+
4
(2.189)
The exural rigidity is again represented by Eq. (2.169). For the particular
case m = 1, the system of equations (2.170) is replaced by the following:
24b
1
+24b
0
q
0
L
3
= 0
72b
2
+72b
1
q
0
L
3
(72 +7g
1
) = 0
144b
3
+144b
2
+q
0
L
3
(30 +6g
1
) = 0
240b
4
+240b
3
+q
0
L
3
(16 +12g
1
) = 0
360b
4
10q
0
L
3
g
1
= 0
(2.190)
We have the same six unknowns as in the clampedfree case. We choose b
4
as an arbitrary constant to obtain q
0,cr
:
q
0,cr
= 36b
4
/L
3
g
1
(2.191)
Note that remarkably Eq. (2.191) coincides with Eq. (2.171). The coefcients
determining the boundary exural rigidity D() are:
b
0
= 6(g
1
+2)b
4
/5g
1
b
1
= 3(4g
1
+3)b
4
/10g
1
,
b
2
= (23g
1
+51)b
4
/10g
1
b
3
= 4(g
1
3)b
4
/5g
1
.
(2.192)
Hence, the exural rigidity D() reads:
D() = [12(g
1
+2)b
4
+3(4g
1
+3)b
4
(23g
1
+51)b
4

2
8(g
1
3)b
4

3
+10g
1

4
]/10g
1
(2.193)
The conditions of physical realizability are positivity of the exural rigidity
as well as the critical load intensity q
0,cr
. The positivity of q
0,cr
is obvious
when both b
4
and g
1
are positive. Concerning the exural rigidity, let us look
at the function f
2
(, g
1
) = 10g
1
D()/b
4
, which is proportional to the exural
rigidity,
f
2
(, g
1
) = (12g
1
+24) +3(4g
1
+3) (23g
1
+51)
2
8(g
1
3)
3
+10g
1

4
(2.194)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 93 #39
Unusual Closed-Form Solutions in Column Buckling 93
f
2
g
1
4
0 0.2 0.4 0.6 0.8 1
6
8
10
12

FIGURE 2.10
Variation of f
2
/g
1
for the column pinned at both ends
As for the clampedfree column we treat f
2
(, g
1
) as a function of the
independent variable g
1
, while will be treated as a parameter. We get,
therefore, the linear equation whose slope f
2
/g
1
depends on :
f
2
/g
1
= 12 +12 23
2
8
3
+10
4
(2.195)
The variation of f
2
/g
1
is portrayed in Figure 2.10. We observe that the slope
remains positive, for [0; 1]. The function f
2
(, g
1
) is increasing with g
1
.
The value of the function f
2
(, g
1
) at g
1
= 0 is f
2
(, 0) = 24 +9 51
2
+24
3
,
1 2 1
and so is the exural rigidity D(). The variation of D() as a function of is
given as a function of variables and g
1
For the case m = 2, the set (2.179) is replaced by
24b
1
+24b
0
q
0
L
3
= 0
72b
2
+72b
1
q
0
L
3
(72 +7g
1
+4g
2
) = 0
144b
3
+144b
2
+q
0
L
3
(30 +6g
1
+3g
2
) = 0
240b
4
+240b
3
+q
0
L
3
(16 +12g
1
) = 0
360b
5
+360b
4
+q
0
L
3
(10g
1
+10g
2
) = 0
504b
5
8q
0
L
3
g
2
= 0
(2.196)
leading to the critical value of q
0,cr
q
0,cr
= 63b
5
/L
3
g
2
(2.197)
2005 by Issac Elishakoff
which is shown in Figure 2.11. Hence, for any positive g , f (, g ) is positive
in Figure 2.12.
EOIS: 2892_chap02 2004/9/28 19:18 page 94 #40
94 Eigenvalues of Inhomogeneous Structures
f
2
(

,
0
)
8
0 0.2 0.4 0.6 0.8 1
10
12
14
16
18
20
22
24
6

FIGURE 2.11
Variation of f
2
(, 0) for the column pinned at both ends
0
1
0.8
0.6
0.4
0.2
0
10,000
20,000
30,000
40,000
50,000
2
4
6
8
10
D
(

)
/
b
4
g
1

FIGURE 2.12
Variation of D()/b
4
for the column pinned at both ends
and the coefcients in D():
b
0
= (115g
2
+168g
1
+336)b
5
/80g
2
b
1
= (115g
2
+168g
1
+126)b
5
/80g
2
b
2
= (165g
2
+322g
1
+714)b
5
/80g
2
b
3
= (15g
2
+28g
1
84)b
5
/20g
2
b
4
= (3g
2
+7g
1
)b
5
/4g
2
(2.198)
In order that the expression for the exural rigidity is physically accept-
able, we should check the positivity of D() in the interval [0; 1]
for any positive value of the parameter g
1
. We consider the function
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 95 #41
Unusual Closed-Form Solutions in Column Buckling 95
f
3
(, g
1
, g
2
) = 80g
2
D()/b
4
, which is proportional to D():
f
3
(, g
1
, g
2
) = (115g
2
+168g
1
+336) +(115g
2
+168g
1
+126)
(165g
2
+322g
1
+714)
2
4(15g
2
+28g
1
84)
3
+20(3g
2
+7g
1
)
4
+80g
2

5
(2.199)
We will describe f
3
(, g
1
, g
2
) as a function of g
1
and g
2
, and take as a
parameter. Hence, the function can be expressed as follows:
f
3
(, g
1
, g
2
) = a()g
2
+b()g
1
+c() (2.200)
where
a() = 115 +115 165
2
60
3
60
4
+80
5
b() = 168 +168 322
2
112
3
+140
4
c() = 336 +126 714
2
+336
3
(2.201)
Figure 2.13 shows the dependence of a(), b() and c() in terms of . We
observe that they are all positive for [0; 1]. This means that f
3
(, g
1
, g
2
) is
positive when g
1
and g
2
are both positive. We conclude, therefore, that the
exural rigidity is also positive when g
1
and g
2
are positive.
The general case for the pinned column is available for m 3. As in the
general case of the clampedfree column, we arrive at a triangular system of
D
(

)
/
b
3
c()
b()
a()
0 0.2 0.4 0.6 0.8 1
20
60
100
140
180
220
260
300
340

FIGURE 2.13
Variation of a(), b() and c()
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 96 #42
96 Eigenvalues of Inhomogeneous Structures
m+4 equations for m+5 unknowns, that are the coefcients in D() and q
0
.
from
0
: 24b
1
+24b
0
q
0
L
3
= 0 (2.202)
from
1
: 72b
2
+72b
1
q
0
L
3
(12G+g
1
) = 0 (2.203)
from
2
: 144b
3
+144b
2
+q
0
L
3
(12G+18 g
2
) = 0 (2.204)
from
i
, 3 i m:
12(i +2)(b
i+1
+b
i
) q
0
L
3
_
g
i
i +1
6
g
i2
i 1
+4
g
i3
i 2
_
= 0 (2.205)
from
m+1
: 12(m+3)(b
m+2
+b
m+1
) +q
0
L
3
_
6
g
m1
m
4
g
m2
m1
_
= 0
(2.206)
from
m+2
: 12(m+4)
_
b
m+3
+b
m+2
_
+q
0
L
3
_
6
g
m
m+1
4
g
m1
m
_
= 0
(2.207)
from
m+3
: 12(m+5)b
m+3
4q
0
L
3
g
m
m+1
= 0 (2.208)
We observe from Eq. (2.206) that the term m1 appears in the denominator.
Therefore, it is necessary to separately consider the case m = 1. Moreover,
in Eq. (2.205) m 3. This implies that additional particular cases, associated
respectively with m=1 and m=2, exist, as they were considered above. We
take b
m+3
to be an arbitrary constant. Equation (2.208) then yieds the value
of the critical load intensity:
q
0,cr
= 3
(m+1)(m+5)b
m+3
g
m
L
3
(2.209)
Note that this expression coincides with Eq. (2.183) for the clampedfree
column. The remaining equations yield the coefcients in D():
b
m+2
=
1
2
_
(2m
2
+12m+10)g
m1
(m+4)mg
m

m+7
m+4
_
b
m+3
(2.210)
b
m+1
=
1
2
_
(m+1)(m+5)(m+6)g
m1
(m+4)(m+3)mg
m
(m+1)(m+5)g
m2
(m1)(m+3)g
m
+
m+7
m+4
_
b
m+3
(2.211)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 97 #43
Unusual Closed-Form Solutions in Column Buckling 97
b
i
=
1
4
(m+1)(m+5)
_
4g
i3
i
2
4

6g
i2
(i 1)(i +2)
+
g
i
(i 1)(i +2)
_
b
m+3
g
m
+
1
4
b
i+1
for 3 i m (2.212)
b
2
= (m+1)(m+5)(g
2
12G18)b
m+3
/48g
m
+b
3
(2.213)
b
1
= (m+1)(m+5)(g
2
+2g
1
+12G18)b
m+3
/48g
m
+b
3
(2.214)
b
0
= (m+1)(m+5)(g
2
+2g
1
+12G12)b
m+3
/48g
m
+b
3
(2.215)
2.4.4 Column Clamped at Both Ends
The preselected buckling mode that fullls the boundary conditions is
() =
2
2
3
+
4
(2.216)
For this column we should rst consider the particular case m = 1. For
this value, the governing differential equation (2.12) yields the following
equations:
4b
2
24b
1
+24b
0
+q
0
L
3
(2 +g
1
) = 0
12b
3
72b
2
+72b
1
q
0
L
3
(16 +6g
1
) = 0
24b
4
144b
3
+144b
2
+q
0
L
3
(30 +3g
1
) = 0
240b
4
+240b
3
+q
0
L
3
(16 +12g
1
) = 0
360b
4
10q
0
L
3
g
1
= 0
(2.217)
In perfect analogy with the two previous cases, we obtain the critical value
of q
0
,
q
0,cr
= 36b
4
L
3
g
1
(2.218)
and the coefcients in D(),
b
0
= (73g
1
+126)b
4
/360g
1
b
1
= (17g
1
+30)b
4
/12g
1
b
2
= (103g
1
+306)b
4
/60g
1
b
3
= 4(g
1
3)b
4
/5g
1
(2.219)
The validity of the obtained exural rigidity imposes the condition D() 0,
for [0; 1]. We introduce again f
4
(, g
1
) = 360g
1
D()/b
4
, which we
treat as a function of the independent variable g
1
, while will be treated
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 98 #44
98 Eigenvalues of Inhomogeneous Structures
as a parameter. We get, therefore, the linear equation whose slope f
4
/g
1
depends on :
f
4
/g
1
= 73 +510 618
2
288
3
+360
4
(2.220)
The slope f
4
/g
1
remains positive for [0; 1] (see Figure 2.14, which
depicts the variation of the slope as a function of ). The function f
4
(, g
1
) is
an increasing function of g
1
. Let us calculate f
4
(, g
1
) = 126+900 1836
2
+
864
3
, which is positive in the interval [0; 1] (see Figure 2.15). Hence, for any
positive g
1
, the exural rigidity remains positive for [0; 1]; the variation
of D() in terms of and g
1
0 0.2 0.4 0.6 0.8 1
40
60
80
100
120
140
160

f
4
g
1
FIGURE 2.14
Variation of f
4
/g
1
for the column clamped at both ends
0 0.2 0.4 0.6 0.8 1
60
80
100
120
140
160
180
200
220
240

f
4
(

)
FIGURE 2.15
Variation of f
4
() for the column clamped at both ends
2005 by Issac Elishakoff
is shown in Figure 2.16.
EOIS: 2892_chap02 2004/9/28 19:18 page 99 #45
Unusual Closed-Form Solutions in Column Buckling 99
0
1
0.8
0.6
0.4
0.2
0
10,000
20,000
30,000
40,000
50,000
2
4
6
8
10
D
(

)
/
b
4
g
1

FIGURE 2.16
Variation of D()/b
4
for the column clamped at both ends
When m 2, we have to satisfy the following set of equations:
from
0
: 4b
2
24b
1
+24b
0
+2q
0
L
3
G = 0 (2.221)
from
1
: 12b
3
72b
2
+72b
1
2q
0
L
3
(6G+2) = 0 (2.222)
from
2
: 24b
4
144b
3
+144b
2
+3q
0
L
3
(4G+6 g
1
) = 0 (2.223)
from
i
3 i m+1:
(i +2)(2b
i+2
12b
i+1
+12b
i
) q
0
L
3
_
2
g
i1
i
6
g
i2
i 1
+4
g
i3
i 2
_
= 0
(2.224)
from
m+2
: 12(m+4)(b
m+3
+b
m+2
) +q
0
L
3
_
6
g
m
m+1
4
g
m1
m
_
= 0
(2.225)
from
m+3
: 12(m+5)b
m+3
4q
0
L
3
g
m
/(m+1) = 0 (2.226)
Since this is again a triangular systemof m+4 equations for m+5 unknowns
(the coefcients in D() and q
0
), we take one unknown as an arbitrary con-
stant to determine the others. The judicious choice for this one is again
b
m+3
and Eq. (2.226) leads to the critical value of q
0
:
q
0,cr
= 3(m+1)(m+5)b
m+3
/g
m
L
3
(2.227)
Each of the remaining equations (2.221)(2.225) lead, respectively, to the
coefcients in D():
b
m+2
=
_
(m+5)(m+1)g
m1
(m+4)mg
m

1
2
m+7
m+4
_
b
m+3
(2.228)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 100 #46
100 Eigenvalues of Inhomogeneous Structures
b
i
= (m+1)(m+5)
_
6g
i3
i
2
4

9g
i2
(i 1)(i +2)
+
3g
i1
i(i +2)
_
b
m+3
g
m
+b
i+1

1
6
b
i+2
for 3 i m+1 (2.229)
b
2
= (m+1)(m+5)(3g
1
12G18)b
m+3
/48g
m

1
6
b
4
+b
3
(2.230)
b
1
= (m+1)(m+5)(3g
1
+12G10)b
m+3
/48g
m

1
6
b
4
+
5
6
b
3
(2.231)
b
0
= (m+1)(m+5)(15g
1
+12G42)b
m+3
/288g
m

5
36
b
4
+
2
3
b
3
(2.232)
2.4.5 Column that is Pinned at One End and Clamped at the Other
The polynomial function that satises the boundary conditions is chosen as
() = 3
3
+2
4
(2.233)
In this case, we have to consider rst of all the particular cases m = 1 and
m = 2.
For m = 1, the governingdifferential equation(2.162) imposes the following
equations:
36b
1
+48b
0
q
0
L
3
= 0
108b
2
+144b
1
q
0
L
3
(18 +10g
1
) = 0
216b
3
+288b
2
+q
0
L
3
(51 +12g
1
) = 0
360b
4
+480b
3
+q
0
L
3
(32 +18g
1
) = 0
720b
4
20q
0
L
3
g
1
= 0
(2.234)
The critical value of q
0
is derived from the last one:
q
0,cr
= 36b
4
/L
3
g
1
(2.235)
We obtain for the coefcients in D():
b
0
= 3(166g
1
+331)b
4
/640g
1
b
1
= (166g
1
+171)b
4
/160g
1
b
2
= 3(26g
1
+61)b
4
/40g
1
b
3
= 3(g
1
4)b
4
/5g
1
(2.236)
We will now verify the positivity of D() in the interval [0; 1] for any positive
g
1
, in order to validate the solution. We dene, f
5
(, g
1
) = 640g
1
D()/b
4
:
f
5
(, g
1
) = 498g
1
+993 +4(166g
1
+171) 48(26g
1
+61)
2
384(g
1
4)
3
+640g
1

4
(2.237)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 101 #47
Unusual Closed-Form Solutions in Column Buckling 101
0 0.2 0.4 0.6 0.8 1
200
300
400
500

f
5
g
1
FIGURE 2.17
Variation of f
5
/g
1
for the pinnedclamped column
Described as a function of the independent variable g
1
, f
5
(, g
1
) is a linear
function whose slope is f
5
/g
1
is
f
5
/g
1
= 498 +664 1248
2
384
3
+640
4
(2.238)
The dependence of f
5
/g
1
is shown in Figure 2.17. We observe that it is
positive, for [0; 1], the function f
5
(, g
1
) is an increasing function of g
1
. Let
us calculate, f
5
(, 0) = 993+684 2928
2
+1536
3
exural rigidity is thus positive for positive values of g
1
. The dependence of
D() on and g
1
Concerning the case m = 2, we have the following set of equations,
36b
1
+48b
0
q
0
L
3
= 0
108b
2
+144b
1
q
0
L
3
(18 +10g
1
+6g
2
) = 0
216b
3
+288b
2
+q
0
L
3
(51 +12g
1
+7g
2
) = 0
360b
4
+480b
3
+q
0
L
3
(32 +18g
1
) = 0
540b
5
+720b
4
+q
0
L
3
(20g
1
+15g
2
) = 0
1008b
5
16q
0
L
3
g
2
= 0
(2.239)
with the last one yielding the critical value of q
0
,
q
0,cr
= 63b
5
/L
3
g
2
(2.240)
2005 by Issac Elishakoff
> 0 (see Figure 2.18). The
is presented in Figure 2.19.
EOIS: 2892_chap02 2004/9/28 19:18 page 102 #48
102 Eigenvalues of Inhomogeneous Structures
0 0.2 0.4 0.6 0.8 1
300
400

f
5
(

)
500
600
700
800
900
1000
FIGURE 2.18
Variation of f
5
() for the pinnedclamped column
0
1
0.8
0.6
0.4
0.2
0
10,000
20,000
30,000
40,000
50,000
2
4
6
8
10
D
(

)
/
b
4
g
1

FIGURE 2.19
Variation of D()/b
4
for the pinnedclamped column
The rest of the equations result in the coefcients in D():
b
0
=
3(6,345g
2
+9,296g
1
+18 536)b
5
20,480g
2
b
1
=
(6,345g
2
+9,296g
1
+9,576)b
5
5,120g
2
b
2
=
(2,365g
2
+4,368g
1
+10,248)b
5
320g
2
b
3
=
3(45g
2
+112g
1
448)b
5
320g
2
b
4
= (9g
2
+28g
1
)b
5
/16g
2
(2.241)
To validate this solution, the exural rigidity D() must be positive in the
interval [0; 1] for any positive value of g
1
and g
2
. As for the pinned column
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 103 #49
Unusual Closed-Form Solutions in Column Buckling 103
0 0.2 0.4
a()
b()
c()
0.6 0.8 1
0.6
0.8
1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8

D
(

)
/
b
3
FIGURE 2.20
Variation of a(), b() and c()
we consider the function f
6
(, g
1
, g
2
) = g
2
D()/b
4
, which is a linear function
of the independent variables g
1
and g
2
, when is taken as a parameter. We,
therefore, describe the function f
6
as follows:
f
6
(, g
1
, g
2
) = a()g
2
+b()g
1
+c() (2.242)
where
a() =
3807
4096
+
1269
1024

473
256

27
64

3
+
9
16

4
+
5
b() =
1743
1280
+
581
320

273
80

2

21
20

3
+
7
4

4
c() =
6951
2560
+
1197
640

1281
160

2
+
21
5

3
(2.243)
Figure 2.20 shows the dependence of a(), b() and c(). We observe that the
function b() takes negative values, especially b(1) < 0, while the other func-
tion a() and c() remain positive. Then, we will calculate f
6
( = 1, g
1
, g
2
) =
a(1)g
2
+b(1)g
1
+c(1).
Since b(1) < 0, for any g
2
, we can nd a positive g
1
that will give
f
6
( = 1, g
1
, g
2
) < 0. We only have to take
g
1
> [a(1)g
2
+c(1)]/b(1) > 0 (2.244)
We prove, thus, that the exural rigidity can take negative values for some
particular positive value of the parameters g
1
and g
2
, which implies that this
solution is not physically acceptable for any positive g
1
and g
2
.
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 104 #50
104 Eigenvalues of Inhomogeneous Structures
We now consider the general case for m3. As above we get a triangular
set of m+4 equations for m+5 unknowns:
from
0
: 36b
1
+48b
0
q
0
L
3
= 0 (2.245)
from
1
: 108b
2
+144b
1
q
0
L
3
(18G+g
1
) = 0 (2.246)
from
2
: 216b
3
+288b
2
+q
0
L
3
(24G+27 g
2
) = 0 (2.247)
from
i
3 i m:
6(i +2)(3b
i+1
+4b
i
) q
0
L
3
_
g
i
i +1
9
g
i2
i 1
+8
g
i3
i 2
_
= 0 (2.248)
from
m+1
: 6(m+3)(3b
m+2
+4b
m+1
) +q
0
L
3
_
9
g
m1
m
8
g
m2
m1
_
= 0
(2.249)
from
m+2
: 6(m+4)(3b
m+3
+4b
m+2
) +q
0
L
3
_
9
g
m
m+1
8
g
m1
m
_
= 0
(2.250)
from
m+3
: 24(m+5)b
m+3
8q
0
L
3
g
m
/(m+1) = 0 (2.251)
and we choose again b
m+3
as an arbitrary constant. The critical value of q
0
reads
q
0,cr
= 3(m+1)(m+5)b
m+3
/g
m
L
3
(2.252)
The coefcients in D() are
b
m+2
=
_
(m+1)(m+5)g
m1
(m+4)mg
m

3
8
m+7
m+4
_
b
m+3
(2.253)
b
m+1
=
_

3
8
(m+1)(m+5)(m+6)g
m1
(m+4)(m+3)mg
m
+
(m
3
+10m
2
+29m20)g
m2
(m1)(m+3)(m+4)g
m

9
32
m+7
m+4
_
b
m+3
(2.254)
b
i
= (m+1)(m+5)
_
g
i3
i
2
4

9
8
g
i2
(i 1)(i +2)
+
1
8
g
i
(i +1)(i +2)
_
b
m+3
g
m
+
3
4
b
i+1
for 3 i m (2.255)
b
2
= (m+1)(m+5)(g
2
24G27)b
m+3
/96g
m
+
3
4
b
3
(2.256)
b
1
= (m+1)(m+5)(3g
2
+8g
1
+72G81)b
m+3
/384g
m
+
9
16
b
3
(2.257)
b
0
= (m+1)(m+5)(3g
2
+8g
1
+72G49)b
m+3
/512g
m
+
27
64
b
3
(2.258)
2005 by Issac Elishakoff
EOIS: 2892_chap02 2004/9/28 19:18 page 105 #51
Unusual Closed-Form Solutions in Column Buckling 105
2.4.6 Concluding Remarks
As in the particular case m = 0, studied by Elishakoff (2000a), here too,
for the different sets of boundary conditions the buckling intensities coin-
cide. This pertains to the fundamental buckling load, since in all cases the
postulated polynomial expressions of the buckling mode do not possess the
internal nodes. It should be noted that whereas for the case m=0 the physical
realizability condition, namely that the exural rigidity must be positive for
0 1, can be easily veried, it is not so for the variable intensity. In this
case the coefcients in the exural rigidity depend upon the ratios g
i
. Only
such ratios are acceptable that lead to a positive exural rigidity.
Since mcan be any positive integer, we conclude that the number of closed-
form solutions that are derived in this section, is a countable innity.
Moreover, the buckling load depends solely on a single coefcient in
the exural rigidity variation and does not depend on other coefcients.
This implies that an innite number of columns, with different coefcients
b
0
, b
1
, . . . , b
m+2
, canhave the same bucklingload, if onlythe sharedcoefcient
b
m+3
is xed.
2005 by Issac Elishakoff