Journal of Applied Mathematics and Physics (ZAMP)

Vol. 24, 1973 Birkh~iuser Verlag Basel

Steady-State Temperature Distribution

in a Finite Spherical Cone 1)
By Kanti A. Patel 2) and Walter J. Harrington, Dept. of Mathematics, North Carolina
State University, Raleigh, North Carolina, USA

1. I n t r o d u c t i o n

In I-1] a steady-state temperature problem in a finite sector (or wedge) was

presented to illustrate the use of the Naylor transform (for polar coordinates),

1
X 2 {f(r)} = F2 (s) zx S (re-1 - r-e- 1)f(r) dr. (1)
o
In this paper, an analogous problem in a finite spherical cone is solved by the use of a
Naylor transform for spherical coordinates,
1
N2 {g (r)} = G 2 (s) zx S (re _ r-S - 1) g (r) dr, (2)
o
which has the important operational property

N2{~[r2g'(r)]}=s(s+l)G2(s)-(2s+l)g(1) (3)

and an inversion formula

2nig(r)= ~r -s-1Gz(s)ds , 0<r<l (4)
L
where L denotes a line, Re s = 2, in the appropriate strip of convergence of the integral
defining N2 .
The relationship of the above Naylor transforms with Mellin transforms, their
fundamental operational properties and inversion formulas are presented in [1].
Here, it is sufficient to state that if g is a real-valued function on 0 < r < 1 such that g is
piecewise continuous and of bounded variation in every [a, b], 0 < a < b < 1, and such
1
that for some real fl>0, 5r -p-1 [g(r)l dr is convergent, then G2(s ), defined by (2),
o
is defined and analytic in the strip, - / ~ - 1 < R e s < ft.

1) Acknowledgment. This research was supported in part by Grant AFOSR-69-1779 at North Carolina
State University at Raleigh, Raleigh, North Carolina. 27607.
2) Currently at Department of Mathematics, Shaw University, Raleigh, North Carolina, USA.
Vol.24, 1973 Steady-StateTemperature Distribution in a Finite Spherical Cone 215

A steady-state temperature problem in an infinite cone with a prescribed

temperature T(r, ~) on the conical surface was analyzed by Muki and Sternberg [2]
through the use of Mellin transforms. In particular, they considered the step-function,

T(r'~)={~,~ a<r<O<r<a
}oo

and obtained numerical results for infinite cones of half-angles, c~= ~/6, and c~= r~/2.
The problem analyzed in this paper relates to finite cones with boundary
conditions on the conical surface and on the spherical surface, r = ro . For convenience,
we shall consider the radial variable to have been rescaled so that the spherical
boundary is at r =1.

2. Statement of the Problem

We consider a class of problems of axisymmetric, steady-state temperature
distribution in a finite spherical cone of half-angle cr with its vertex at the origin and
its axis along the principal axis of a spherical coordinate system (r, 0, qS).The boundary
surfaces of the cone are expressed by the conical surface, 0= ~, and the spherical
surface, r = 1.
The temperature function U(r, 0) is independent of ~band, for isotropic, thermally
homogeneous materials, satisfies that Laplace equation

r2Ur'+2rUr-F sinl 0 ~0~ s i n 0 ~ - =0. (5)

We seek the solution of (5) subject to the boundary conditions:

Ur(1, O)+hU(1, O)=O, h>O, (6)

U(r, c~)=rm, m__>0. (7)

With x = cos 0, x o = cos c~,and U (r, 0) replaced by T(r, x), our problem can be expres-
sed as follows:

8r (r2 T~)+ [(1 - x 2) T~] = 0 , 0 < r < 1, Xo<X< 1, (8)

T~(1,x)+h T(1, x)=O, h>O, (9)

r(r, Xo)=r" , m=>0. (10)

It will be assumed that T(r, x) and r T~(r, x) remain bounded as r ~ 0 +.

Because of condition (9), we do not apply the N2-transform directly to the
function T(r, x). Instead we note that the function

w(r, x) a=rT~(r,x)+ h T(r, x) (11)

216 Kanti A. Patel and Walter J. Harrington ZAMP

will also satisfy the Laplace equation in the interior of the spherical cone. This leads
to the equivalent problem:

t ~ - (r2 Wr)+ [(1--X2) Wx]=O, 0<r<l, Xo<X<l, (12)

w(1, x ) = 0 , (13)
w(r, xo)=(m+h)r", m>O, h>O, (14)
with w(r, x) remaining bounded as r ~ 0 +.

3. Solutions in Complex Integral Form

To solve equation (12) subject to conditions (13) and (14), we apply the N2-
transform, letting
1
W(s, x) = N2 {w (r, x); r --+ s} -- 5 (rs - r-S- 1) w (r, x) dr. (15)
o
Because of the boundedness of w(r, x) as r-+ 0 +, we expect to have a strip of con-
vergence, - ~ - 1 < Re s < ~ with ~ > 0.
Using property (3) and condition (13), we obtain

~ [ ( 1 - x 2 ) Wx]+s(s+l)W=O, -~-l<Res<~ (16)

subject to the condition from (14)

W(s, Xo)=(m+h)(2s+l)/(s+m+l)(s-m), -m-l<Res<m. (17)

The solution of (16) and (17) is given by

(m+h)(2s+ 1) P~(x)
W(s, x)= (s+m+ 1)(s-m) P~(xo) ' (18)

where P~is the Legendre function of the first kind of degree s. The inversion formula (4)
yields the formal solution 3),

m+h z+i~ (2s+l)r-~-lp~(x)

w(r,x)= 2hi ~-!oo (s+m+l)(s-m)P~(Xo) ds, (19)

where the line of integration is in an appropriate strip of analyticity of the integrand.

Substitution in (11) yields the differential equation,
1 2+ioo
rT~(r,x)+hT(r,x)=2n i ~ W(s,x)r-S-lds. (20)
2-- ioo

3) In the special case, m = h = 0 , w(r,x ) = 0 and T(r,x)=-1. In the material below, it is assumed that
(m+h)~-O.
Vol. 24, 1973 Steady-State Temperature Distribution in a Finite Spherical Cone 217

4. Evaluation by Residue Expansion

For each x, X o < X < 1, the function W(s, x) in (18) is an analytic function of s
everywhere in the s-plane except at the zeros of the function

G(s, Xo) = (s + m + 1 ) ( s - m) P~(Xo). (21)

It is known [3] that the set of zeros of P~(xo) is denumerable and that the zeros are
real and simple and of the forms, Pk and (--pk-- 1), where 0 < p l < p 2 < P 3 < ... and
pk~O0 as k--. oo. Thus the zeros of G(s, Xo) are of the same forms and are also real and
simple unless m = Pk for some k. If m is an integer, this can occur only if ct = n/2 with
m odd. In this case P k = 2 k - 1 and G(s, Xo) has double zeros at s = m and s = - m - 1 .
Computation of the zeros of P~(xo) is treated in [4] and [5].
From the integral representation (see [3]),

cos [(s q-l) y]

P~(cos0)= r - j dy, 0<0<~ (22)
o ]//cos y - cos 0

it is clear that with 0 < 0 < ct < ~/2, then P~(x) > 0 if - 1 < s < 0. Thus for all values of c~,
0 < ct < ~z/2, we can consider a common strip of analyticity, - 1 < Re s < 0, and in
the inversion integrals of(19) and (20) it will be appropriate to choose 2 with - 1 < 2 < 0.
We consider a sequence of counterclockwise paths of integration { Ck} where,
for each integer k, C k is defined as consisting of the straight line segment
(Re s = 2 , - - R k < I m s<=Rk), where pkq-2+ 1 <Rk<Pk+l + 2 + 1, and the semicircle,
I s - 21 = Rk, to the left of Re s = 2.
The asymptotic expansion, as Isl--,

p~(x)=]/ 2 F(s+l){ }
~sin0 F ( s + 3) c ~ s+l) 0-4]+O(s-i, '
(23)
e~O~-e,e>O

shows that P~(x)/P~(xo) remains bounded as Is] --* oo and that the integrand in (19) is
O(1/Rk) on the semicircular portion of C k as k and R k --.oo. Thus, in terms of W(s, x)
as formulated in (18),

1
w (r, x) = lim - - *~ r - s - 1 W(s, x) ds, 0 < r < 1. (24)
k . ~ 27zi Ck

For each integer m >=0 and each Xo, 0 < x o < 1, the set of poles of W(s, x) in the left-half
plane consists o f ( - m - 1) and the descending ordered array { --Pk-- 1; k = 1, 2, 3 .... }
where the sequence {Pk} is dependent upon x o .
218 Kanti A. Patel and WalterJ. Harrington ZAMP

We note that

P_~_t(x)=P~(x)~ ~ = - ~P~ (25)

=-pk-* ~s ]s=pk"
Thus the application of the residue theorem to the contour integral in (24), in the
cases where all poles are simple, yields the expansion

w(r,x)= (m+h) Pm(x)rm+(m+h)~ (2pk+I)Pp~(X) [dPfix~ -1 r p~. (26)

P.,(Xo) k=l (pk-Fmd-1)(Pk--m) [ ds .Is=pk
Integration of(ll) to obtain the T(r, x) corresponding to this w (r, x) yields, for 0 < r < 1,

P~(x) r'+(m+h) ~ at)

(2Pk+ 1) Ppk(X) I- dP~(xo)-I-a ,*
T(r, x)---
Pm(Xo) k=l (pk+h)~k+m+ l ~ k -m) I _ / - - / d s]s=pkrP~" (27))

The verification that T(r, x) is a solution of the boundary-value problems

specified by (8), (9) and (10) can be carried out in the same way as that of Muki and
Sternberg in [2].
If~ = n/2, two cases arise according as m is an even or odd integer since Pk= 2 k - 1.
Case 1: Xo=0, m = 2 n > 0 . In this case, the result (27) is applicable. Some
simplification is possible since x o = 0 and Pk is an integer. In particular, one can show,
using relations s) given in [7], that

dP,_(O) S=Ek+I -- 1 -- (-- 1)k+l 22k(k!)2 (28)

ds (2k+ 1) P2k(O) (2k+l)!
Thus,
rE. ~ (2k+I)(4k+3)P2k(0)P2k+~(x)
w(r,x)=(m+h) [ z.(O) k=O (2k--2n+1)(2k+2n+2) r2k+l (29)
and
T' " P2"(X)r 2 " - (2k+l)(4k+3)PEk(O)P2k+l(x) r2k+1. (30)
tr, X)=p~.(O) f i (2k+h+l)(2k+2n+2)(2k-2n+l)
(2n+h)k=O

Case 2: x o = 0, m = 2 n + 1 > 1. In this case all poles of W(s, x) in the left half-plane
are simple except for a double P01e at s = " 2 n " 2. As an aid in calculating the residue
at the double pole, one can show that

d2 P*(0) s= d2 P ~ ( 0 ) ~ ( n + 3 ) - ~ ( n + 1)
ds2 --2n--2-- ds2 s=2n+l-- (2n+ 1) P2.(0) (31)

where ~ is the digamma function arising in the derivative of the gamma function.

4) Note that when m=0 with h=#0, T(0;x)= 1; when m>0, T(O.x)=0.
sin~sF( S+l I
Vol. 24, 1973 Steady-State Temperature Distribution in a Finite Spherical Cone 219

Letting ~' denote the sum over all k > 0 except k = n, we obtain from the residue
theorem
~, (2k+ 1)(4k + 3)Pzk(O)P2k+I(X) r2k+1
o0

k=O (n--k)(2k + 2n+ 3)

w(r, x)= 2n +2h + 1 - ( 2 n + 1) P2. (0) ]t~O(n + 3 ) - ~k(n + l) (32)

+ 4n~+2(lnr)}pz.+l(X)+2 dP~(X)dss=2n+lJ]l'2n+l
o0
2n + h + 1 ~, (2k+ 1)(4k + 3) Pzk(O)P2k+t(X) .2k+,
2 *=0 (n-k)(2k+2n+3)(2k+h+l)
(2n+ 1) P2,(O) [{
T(r, x)= 2 ~ ( n + 3 ) - ~ ( n + 1) (33)

4n+3 2
2n+h+l
+2(lnr)}Pzn+l(X)+2~s=E.+l]rzn+l.

It is to be noted that both (32) and (33) involve dP~(x)/dsevaluated at s = 2 n + 1

with x 4: 0. There appears to be no simple formula for this quantity. A half-range
Legendre-Fourier series can be obtained; see (47) in section 6.
It should also be noted that the cases corresponding to results (30) and (33) can
be treated by applying the odd Legendre transform (with x as the transform variable).
The transform is not applicable if ~< re/2. Confirmation that the results obtained
by using the Legendre transform are equivalent to those above is presented briefly
in section 6.

5. Numerical Calculations

The residue expansions of the preceding section were not used directly for our
numerical computations. Instead, in the manner of Muki and Sternberg [2], a real
integral form of the inversion integral for w(r, 1) was approximated directly by
computerized integration routines.
In (19) the choice 2 = - 8 9 together with the change of variable, s = - 8 9
yields the real integral

w(r,x)=-2(m+h) ;~sin(~lnr) P~89 d~. (34)

0 {2+(m+ 89 2 P-~+i~(Xo)
The Legendre functions in (33) are real-valued since
2 cosh({ y) dy, O<O<n, 0__<~<oo, (35)
P_~+i~(x)=~-~ 1/cosy-cos 0
220 Kanti A. Patel and WalterJ. Harrington ZAMP

where x = cos 0. Because of the singularity at y = 0, we use, as in [2], the representation

obtained by integration by parts,
P_~+,~(x)
0
41/~
- n sin 2 0 So{c~176176189176176176 (36)

Equation (23) yields the asymptotic estimate as 4 ~ ~ ,

9 89

P_89162 ( sm~ ] .e_r +O(1/4) ] O<e<O<_n-e. (37)

P ~+i~(Xo) \sin0]
For 0 = 01

P-~+ir ~]/2rc 4 sin ~ e -'r [1 + O (1/4)]. (38)

P_~+i~(Xo)
To compute w(r, 1), formula (36), with 0=0, x = 1 and prescribed values of ~, m,
and h, was employed first to calculate P_89162 0 < 4 < 4 0 , at intervals of 0.1,
where 4o = 15 was chosen so that the truncation integral from 4o to oo in (34) would
be O(10-4).
In particular, in view of (38),

-2(m+h) ~o 4 sin(4 In r) 1
J d4
n r 4 2 +(m+~)1 2 P~+ir
2(m+h)~ e -'e~
[~ sin (4o In r)+ In r cos (4o In r)] (39)
n ]~O((X2 + In 2 r)
+ O (4o ~ e-~r

Then formula (34), with the interval of integration truncated to [0, 15], was used
to calculate w (r, 1) on the interval 0 < r < 1 at 21 points equally spaced at 0.05 intervals.
The differential equation (11), with known T(O, 1) and the computed values
of w(r, 1) was then integrated to obtain T(r, 1). It should be noted that because of
the singular point at r = 0 in (11) special attention was required, particularly for
h < 1, in the computation of T(r, 1) near r = 0. The asymptotic behavior of w (r, x) and
T(r, x) which is known from (26), (27), (30), (32), and (33) was useful9
The computations were made using an I.B.M. 360 electronic computer and a
Q.S.F. integration subroutine9 In certain cases shorter subinterval lengths were
used to obtain comparisons with the calculations employing the subinterval lengths
indicated above9
Computations were made for cones with ct=n/6 and ~ = n / 2 (hemispheres).
Parameter values h= 0, 89and 2 together with m = 0, 1 and 2 were considered9 The
variation of temperature along the axis of the cones (0 = 0), for the different com-
binations of parameter values, is shown in Figures 1, 2, and 3. It can be noted that
Vol. 24, 1973 Steady-State Temperature Distribution in a Finite Spherical Cone 221

1.0 t ~m=0
T(r,1) 9 c~ = ~6
o0~=~/2
&8 ,m=l

,m=2
0.6 'm=l

0.4 ,m=2

0.2
Figure 1
Variation of temperature along axis of cone:
h = 0 ; ~=n/6 and n/2; with T(r, cos c~)= r", O.2 OA 0.6 0.8 1,0
m=0, 1,2. r

T(r,1)
08 .o,=~/6 1---o...
otx=~/2
o~ ff---_~ , , ~ "---""~
o0 m=1 ,....--,~-~ m= 2

-~..~ m=2
0.2
Figure 2
Variation of temperature along axis of cone:
h= 89 ct= n/6 and n/2; with T(r, cos ~)= r", 0 0.2 0.4 0.6 0.8 1.0
/-
m=0, 1,2.

1.0 ~ o ~ ~ ~'~/6
T(r'~ii " ~ ~ i T[/2

Figure 3
Variation of temperature along axis of cone:
h = 2 ; c~=~/6 and n/2; with T(r, cos ct)=r m, O0 0.2 0.4 0,6 0.8 1.0
m=0, 1,2. r
222 Kanti A. Patel and Walter J. Harrington ZAMP

in the cones with c~=~/6, the first pole involved corresponds to p1=4.13 and thus,
from (27), for m = 0, 1, 2

T(r, 1)~r"/Pm(COS T~/6) as r--,O +. (40)

W h e n c~= re/2, Pl = 1, and thus from (30) and (33),

3hr
T(r, 1),-~ 1 if m = 0 ;
2 ( h + 1)
T(r, 1 ) ~ - r In r if m = 1 ; (41)

T(r, 1) 3(h+2) r if m = 2 .
4(h+l)

6. Application of the Odd Legendre Transforms in the Case, 9 = x[2

F o r the hemispherical case, ~ = n/2, the odd Legendre transform (see Churchill
[-8]) is applicable to the b o u n d a r y value p r o b l e m given by (12), (13) and (14) with
x o = 0. The independent variable involved in the integral transform is x rather than
r and the inversion yields a Fourier series in PZk+I(X). Here we shall note only the
solutions obtained for w (r, x) in the two cases, m = 2 n and m = 2 n + 1, n an integer > 0.
Case 1 : F o r m = 2 n, one obtains

where
w(r, x)=(m+h)
{2o
rm Bk(m ) P2k+l(x)--
k=O
Bk(m ) P2k+I(X) r 2k+1
} (42)

(2k+ 1)(4 k + 3) P2k(0)

Bk(m)= ( 2 k - m + 1)(2k+m+2) " (43)

Case 2: F o r m = 2 n + 1, one obtains

- m (ln r) Pro-1 (0) P.~ (x) r m
m ~, o0, (44)
w (r, x) = (m + h) + r ~ B k (m) P2 k+X(X)-- ~ B k (m) P2k+ 1(X) r Ek+l'
k=O k=0

T o show that (42) is equivalent to (29), one can merely verify the Fourier-
Legendre expansion,

P2,(x)=P2,(O) ~ Bk(2n ) PZk+I(X). (45)

k=O

The verification that (44) is equivalent to (32) is a bit m o r e difficult. A limiting

process such as is used by B r o m w i c h in [6] is useful. F o r toe0, ]t]< 1, one can use
Vol. 24, 1973 Steady-State Temperature Distribution in a Finite Spherical Cone 223

the same machinery as in (45) to obtain

of)
~' Bk(2n+ 1 +t)P2k+a(X)= P2"+t+t(x) ~ (2n+ 1)(4n+ 3),02,(0)P2.+a(X)
k=O P2,+l+t(0) t(4n+3+t)
P2n+l+t(x)-Pen+l(x) Pzn+l(x)
,-,(2 n + 1) P2,(0) t 4n+3 + t (46)
1
- ~ P2.+1 +t(x) [~ (n +3) _ ~ (n + 1)].
Letting t ~ 0, one finds that
oo

~' Bk(2n + 1) PZk+I(X)= --89 + 1) P2, (0)

k=0
(47)

as Is=~.+lJ
This establishes the equivalence of (44) and (32).

Acknowledgment
The authors express their gratitude to M.A. Shah for his aid in formulating
the computer programs which were used.

References
[1] W.J. HARRINGTON and K.A. PATEL,Naylor Transforms of Mellin Type, SIAM J. Math. Analysis,
4 (1973).
[2] R. MUKI and E. STERNBERG, Steady-State Heat Conduction in a Circular Cone, Z. angew. Math.
Phys. 9 (1960).
[3] W. MAGNUS and F. OBERrlETTINGER, Formulas and Theorems for the Functions of Mathematical
Physics, Chelsea, New York (1954).
[4] H.M. McDONALD, Zeros of the Spherical Harmonic p m(#) considered as a Function of n, Proc. London
Math. Soc., 31 (1899).
[5] B. PAL, On the Numerical Calculation of the Roots of the Equations P,'(#)=0 and d/d#P,m(p)=O
regarded as Equations in n, Bull. Calcutta Math. Soc., 9 (1917) and (Part II) 10 (1918).
[6] Higher Transcendental Functions, edited by A. ERDELYI, Vol. 1, McGraw-Hill, New York (1953).
[7] I.S. GRADSHTEYNand I.M. RYZHIK, Tables of Integrals, Series and Products, Academic Press, New
York (1965).
1-8] R.V. CI~URCHILL, Operational Mathematics, 3rd Ed., McGraw-Hill, New York (1972).
1-9] T.J.I'A. BROMWlCH, Certain Potential Functions and a New Solution of Laplace's Equation, Proc.
London Math. Soc., 2 (1912).

Abstract
A Naylor transform is employed to determine the axisymmetric steady-state temperature distribu-
tions in finite spherical cones where the boundary conditions include a variety of conditions at the spherical
surface (insulation or heat loss in accordance with Newton's law)and an assigned temperature variation
on the conical surface ( in particular, f(r) = r~', m > 0). Solutions are obtained in terms of inversion integrals
and in Fourier-Legendre expansions. Computations based on the inversion integrals are carried out for
cones of half-angles, c~= n/6 and n/2, and with several combinations of parameters relating to the boundary
conditions.
224 Kanti A. Patel and Walter J. Harrington ZAMP

R~sum~
Une transformation de Naylor est employ6e afin de d&erminer les distributions axisym~triques
stationnaires de la temperature dans les c6nes sph6riques finis off les conditions aux limites eomprennent
une vari6t6 des conditions ~ la surface sph6rique (isolation ou perte de chaleur conform~ment ~t la loi de
Newton) et une variation de la temp6rature assignee fi la surface conique (en particulier f(r)= rm, m >=0).
Les solutions sont obtenues sous forme d'int~grales d'inversion et de d~veloppements en s6rie de Fourier-
Legendre. Les calculs fond6s sur les int6grales d'inversion sont accomplis pour les c6nes de demi-angles,
e=n/6 et n/2, et pour diff6rentes combinaisons de param6tres qui apparaissent dans les conditions aux
limites.

(Received: November 14, 1972)