Derivative Polynomials for Tangent and Secant

Author(s): Michael E. Hoffman

Source: The American Mathematical Monthly, Vol. 102, No. 1 (Jan., 1995), pp. 23-30
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2974853
Accessed: 16-01-2016 00:29 UTC
REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/2974853?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact [email protected].

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to The American
Mathematical Monthly.

http://www.jstor.org

This content downloaded from 193.52.245.246 on Sat, 16 Jan 2016 00:29:05 UTC
All use subject to JSTOR Terms and Conditions

DerivativePolynomials For Tangent

and Secant
Michael E. Hoffman

1. INTRODUCTION.Sometimesproblemsnaturallyoccur in pairs, and it's best

to tackle both at the same time. For instance,considerthe problemof findingthe
nth derivativeof tan x. It's not hard-tosee that there are polynomialsPnOfdegree
n + 1 for n = 0,1,... so that
dn

sX n tan x = Pn(tan

x).

This problemhas a naturalcompanion:find the nth derivativeof sec x. Here there

are polynomialsQnof degree n so that
dn

dXnsec x = Qn(tanx)sec x.
The Pn and Qnare different sequences of polynomials,but they are evidently
related. The numbers Pn(O)/n!
and Qn(O)/n!
are the coefficients of xn in the
Maclaurin series for tan x and sec x respectively,and their computation is a
classicalproblem.
Here's another problem: for positive integer n and 0 < a < 1, what is the
improperintegral
oo

X ne

eX_ldX?

It has a naturalcompanionproblemwhere the denominatoris replacedby ex + 1.

As we shall see, this turns out to be essentially the same pair of problems as
consideredin the previousparagraph.
This paper has two main parts.First,in 2 and 3 we obtain the polynomialsPn
and Qnas instances of 'derivativepolynomials'associatedwith functions f such
that f'(x) is a polynomialfunctionof f(x). Then in 4 we applythis theoryto the
computation of improper integrals and infinite series, followed by concluding
remaFksin 5.
2. DERIVATIVE
POLYNOMIALS.Suppose f is a functionwhose derivativeis a
polynomialin f, i.e. f'(x) = P(f(x)) for some polynomialfunction P. Then all the
higher derivativesof f are also polynomials in f, so we have a sequence of
polynomialsPndefined by
f

(x)

Pn(f(x)),

n 2 O.

In fact, the polynomialsPnare determinedby the conditions

Po(a) = u,
1995]

Pn+l(u) = Pn(u)P(u)

for n 2 1.

DERIVATIVE POLYNOMIALS FOR TANGENT AND SECANT

This content downloaded from 193.52.245.246 on Sat, 16 Jan 2016 00:29:05 UTC
All use subject to JSTOR Terms and Conditions

(1)
23

If we form the generatingfunction

= L t Pn(a),
n=0
n

F(u, t)

tn

then the equations(1) are equivalentto

Ft(u, t) = P(u)F"(u, t).

F(u, O) = u,

(2)

Also, f has a Ccompanion'

function g = exp(lJ(x) dx) satisfyingg'(x) = f(x)g(x),
and there are polynomialsQn
sOthat
gfn)(x) = Qn(t(X))g(X),

n > 0.

It is easy to see that the Qn

are determinedby the conditions

Q0(u)1

forn 2 0.
Qn+l(a)
= uQn(a)
+ Qn(a)P(a)

If we define a second generatingfunction G(u, t) by replacingt

of F(u, t) with Qn,
then (3) is equivalentto
G(u,O) = 1,

(3)

in the definition

+ uG(u,t).

Gt(u,t)-P(u)Gu(u,t)

(4)

Our first result gives explicitformulasfor the generatingfunctions F and G.

generatingfunctions
Theorem2.1. Forfunctsns f and g as above, the corresponding
for theirderivativepolynomialsare givenby
F(u,t) =t(t-1(u)

+ t)

and G(u,t) =

( (f_1( )) ) .

Proof: Let u = f(x). If F(u, t) = f(x + t) we have F(u, 0) = f(x) = u, and

du
dx
P(u)FIl(u, t) = dxf (x + t) du =f'(x + t) = Ft(u, t),
so F(u, t) = f(x + t) satisfies (2) above. This establishes the first formula;the
proof of the second formula(using (4)) is similar.
In view of (2) and (4) above, the generatingfunctions F and G are determined
by P alone. Indeed y = f(x) is a solutionof the differentialequationy'(x) = P(y)
so P determinesf up to a constant(i.e., f(x) can be replacedby f(x + c)). Here
are two examples.
Example1. Let f(x) = kdx SQ that P(u) = k. The generatingfunction F(uSt) is
just u + kt. The companionof f iS g(X) = ekX /2, and from the theorem G(u, t)
is exp(tu + kt2/2). Expandthis out to get

Q(a)=
In the case k =-1

vE ( n )(2i-1)(2i-3)

*-* lkiun-2i

the Qn
are (one variantof) the Hermite polynomials.

Example2. The main exampleof this paper, of course, is the case t(x) = tan x,
g(x) = sec x (i.e. P(u) = u2 + 1). Here the generatingfunctionsare
F(u t)=
'
24

u + tan t
sin t + u cost
=
1 - utant
cost - usint

DERIVATIVE POLtYNOMIALSFOR TANGENT AND SECANT

This content downloaded from 193.52.245.246 on Sat, 16 Jan 2016 00:29:05 UTC
All use subject to JSTOR Terms and Conditions

[January

and (cf. [10])

cos(tan-l

u)

cos t - u sin t
cos(tan-1 u)cos t - sin(tan-1 u)sin t
( ' )
Until now the Pn and Qnhave been treated in parallel,but separately.The next
result bringsthem together.
Theorem2.2. Thegeneratingfinctions F and G satisfy (and are determinedby) the
conditions
and Gt =FG.
Ft =P(F),
G(u,O) = 1,
F(u,O) = u,
Proof: Using 2.1, we have
Ft(u, t) = f'( f-l(u) + t)-P( f( f-l(u) + t)) P(F)
and the second equationis similar.
This result is useful for obtainingrecurrences.For instance,in Example1 above
the equation Gt = FG leads to the recurrence

Qn+1(u) = Po(u) Qn(u) + nPl( u) Qn- 1(u) = uQn(u) + nkQn- 1(u)

sincePk(u)= 0 for k 2 2. Compare this with (3) to get Q'n(u)= nQn_1(u).
It follows that y(x) = Qn(x) is a solutionof the differentialequation ky"+ ' ny - O.(Of course, this is Hermite'sequationif k = -1.)
In Example2, Theorem2.2 leads to the pair of recurrences
Pn+1(U) =
=

Qn+l(a)

( g ) Pi(U)Pn-i( u) + bOn and

* )Pi(u)Qn-i(U)

VALUES.Henceforthwe specialize to the case

3. COMPUTINGPARTICULAR
P(u) = u2 + 1 (Note this applies equally to f(x) = tan x, g(x) = sec x and to
f(x) = -cot x, g(x) = csc x). In this section we show how to find particular
withoutcomputingthe polynomialsthemselves.
values of Pn and Qn
those at zero are of the most
Of all the values of the polynomialsPn and Qn,
interest:as noted in the introduction,they give the coefficients of the Maclaurin
series for tangent and secant. These numberscan be computed as follows. From
isodd for n
(1) and (3) it is evidentthat Pnis an odd functionif n is even, and Qn
odd. Then using (5):
=

Pl(0)

P3(0)

11

= ( 1 )P1(0) = 2,

( 1 ) P1(0)p3(0) + ( 2 ) P3(0)p1(0) = 16;

s( )
and thus

= 1,
= (0)Q0(0)Pl(0)
= 1, Q2()
Q0(0)
P3(0)
(0 )Q()

P1(0) = 5
2 ) Q2()

= 61.
+ ( 4)Q4(0)Pl(0)
P3(0)
2) Q2()
Othervalues can be obtainedfrom these via the followingfunctionalequation.
()

1995]

5(

DERIVATIVE POLYNOMIALS FOR TANGENT AND SECANT

This content downloaded from 193.52.245.246 on Sat, 16 Jan 2016 00:29:05 UTC
All use subject to JSTOR Terms and Conditions

25

Theorem3.1. It u + 0, then
-

/ u2-1

Pn(u) = 2 Pnt

2u

u2

/ U2-1

2u

2u

Proof: It is enough to show

u2-1

F(u,t)=F

2u

12t} +

u2 +

/ U2-1

2u Gt

,2tJ

for u + 0. Let u = -cot x and apply 2.1 to get F(u,t) = -cot(x + t) and
csc xG(u, t) = csc(x + t). From the half-angleformulafor tangent,
F(u, t) = -cot(2x + 2t)-csc(2x
+ 2t)
= Ff-cot 2x, 2t)-csc 2xG(-cot 2x, 2t),
from which the result follows.
Putting u = 1 in Theorem 3.1, we have

P(1)= 2n(P(0)+ Qn())

= (2np(0)

n odd

To get the Qn(1)we need the followingrecurrence,which complements(5).

Theorem3.2. For n 2 O,
(U

1)

)Qi(a)Qn-i(U)

Pn+l(U)

Proof: It sufficesto provethe identity(U2 + l)G(u, t)2 = Ft(u, t), whose right-hand
side is P(F) = F(u, t)2+ 1 by Theorem2.2. But by 2.1,
F(u, t)2 + 1 = tan2(tan-1(u) + t) + 1 = sec2(tan-1(u) + t)
= seC2(tan-l(a))G(u,

t)2

Given the Pn(l) the Qn(1) can be obtained recursivelyby setting u= 1 in

Theorem3.2. At this point the readermay find it instructiveto compute Q4(1).
Remark.There are more efficient ways to compute the numbersPn(O)
and Qn()
than that outlined here: see [3] and [9]. For the Qn(1)see [8].
4. IMPROPERINTEGRALSAND SERIES. We now use the polynomialsPnand
Qnto expresssome improperintegralsand infiniteseries, startingwith the former.
Theorem4.1. Let O < a < 1. Thenfor integern 2 O,
X x e
|
x 1 dx= Trn+1cscaTrQn(-cotarr).
Proof: The onlypole of the meromorphicfunction eaZ/(eZ+ 1) inside the rectangle with vertices -R, R, R + 2 Tri,and -R + 2 Triis at Tri,where it has residue
-evia.
Integrateit aroundthis rectangle and take limits as R > oo to obtain the
result for n = O.Then differentiaten times with respect to a, noting that the nth
derivativeof csc x is csc xQn(- cot x).
DERIVATIVE POLYNOMIALS FOR TANGENT AND SECANT

26
This content downloaded from 193.52.245.246 on Sat, 16 Jan 2016 00:29:05 UTC
All use subject to JSTOR Terms and Conditions

[January

If the denominatorof the integrandis replaced by ex _ 1, we have to modify

the numeratorin the case n = 0 to get a convergentintegral.
Theorem4.2. Let O < a

1. Then

<

X x e

(a)

x _ 1 dx

= XTn+lPn( -

ooe ax
|

(b)

cot a Tr) for integern 2 1, and

_ e(l-a)x

ex_ 1

dx= -2rcotaTr.

Proof: To prove (a), considerf defined by

zeaz

t()=1.

eZ_1'Z+;

Then f is analyticinside the rectanglewithvertices -R, R, R + Tri,and -R - Tri,

so we can applyCauchy'stheorem and let R > oo to get
xe aX
| Xex_ 1 dX

. X ( x + Tri ) e
t
ex + 1

dx

Now use 4.1 to computethe right-handside, and (a) followsfor n = 1. Differentiation with respect to a then gives the generalcase. For (b), note that the integrand
has no poles within the rectangle used to prove (a): integrate it around this
contour,take limits as R > oo, and simplifyusing 4.1.
Next we pass from integralsto series.
Theorem4.3. For real O < a

<

1 and integer n 2 O,

(a)
and
(b)
E

k-O

(k

(k

)n+l

+ a)n+l

(k

1)

+ (

rrn + 1

1)

n!

1 _ a)n+l

nl csc a7rQn(cota7r).

(k + 1 -a)n+l

Proof: For n 2 1, let

Xx e
In(a)=|O

ex_ldX

Since
lo

and Pn(-X)

x e

= (-l)n+lPn(x)

I (1 - a)

czx
=(_l)n+In(l-a)

( )

4.2(a) can be restated as

+ (-l)n+

I (a)

= vn+lPn(cot aTr).

On the other hand,

n (a-l)x

In(a) = |

1995]

-x dx = ,

nt

oo

oo

| x e (

) dx =

(k

+ 1

DERIVATIVE POLYNOMIALS FOR TANGENT AND SECANT

This content downloaded from 193.52.245.246 on Sat, 16 Jan 2016 00:29:05 UTC
All use subject to JSTOR Terms and Conditions

)n+l

27

Z C is periodic modq if

and so (a) follows for n

>

= O and +(n

+(q)

For n

1.

eaX _ e(l-a)x

q)

CtX

21o

ex -

and

Z,

note that

O,

0 eaX _ e(l-a)x

1 OO ex-1

for all

+(n)

0O
dS

kEO

1
-

and use 4.2(b). The proof of (b) using 4.1 and the series expansion

lo

xneaX

O0

n !

1)

k=o(k+1-a)

+1

is entirelyanalogous(but simpler,since no special argumentis requiredfor n

O).

Equation 4.3(a), for rational a, appears implicitly in 171 of Euler's

algebratext [6]. In two earlierpapers[4], [5] he relates the left-handsides of 4.3(a)
and 4.3(b) (for rational a) to improperintegralsof certain rationalfunctions.
From Theorem 4.3 we can derive a quite useful result on series of powers of
reciprocalsof integers. We shall need the following definitions. A function +:

Remark.

alternatingmodq if +(q) = O and +(n + q) = -+(n) for all n E Z. If + is a

periodic or alternatingfunction modq, then it is even if +(q -j) = +(j) for all
1 < j < q, and odd if +(q - j) = -+(j) for all 1 < j < q.
Theorem4.4.

If

(a)
where

mod

q,

is

even

and

p/q

and

jn+l

odd,

is

mod q, then

function

2qn+ln!

or

is

n(

pE1E(P)

odd

and

even.

If

is

an

alternating

function

are

2q

either

both

n ! p=

even

or

both

odd.

As the proofs of (a) and (b) are similar, we give only the former. Set
in 4.3(a) and multiplyboth sides by lE(p)/qn+1;
then sum over p to get

s1 X
E

E1

J=1 j

Proof:
a

a periodic

then

(b)
when

is

( _ l)n+lt/ff )

p)

,i(

+p)n+l

(qk

+ (qk

q-1

Trn+l

Pm

= qn+ln! E +(P)PntxCotq

q-p)n+l

If n and + have opposite parity,(- l)n+llE(p)

= IR(q -

p)

i
^

and the left-handside

1S
q-l

ti}( p)

E E
p=l k=O

q-l

n+l +
(qk

p)

E
p=1

tyJ}(q

E
k=O

p)

t/>(j)

n+l 2 E
(qk

-P)

j=l

n+l E
J

Here are some examples:for the first two, it's helpful to recall 3.
Example

1.

Since the periodicfunctionmod2 with +(1) = 1 is even, we have

1
1
1
Trn+
1 + 3n+l + sn+l + 7n+l + *5 = 2n+2n!Pn(O)

28

DERIVATIVE POLYNOMIALS FOR TANGENT AND SECANT

This content downloaded from 193.52.245.246 on Sat, 16 Jan 2016 00:29:05 UTC
All use subject to JSTOR Terms and Conditions

[January

for n odd, from which it follows that

1+

,rrn+lp(0)

+ n+l + 4n+1+

2(2n+l-l)n!

(6)

for such n. Similarly,usingthe alternatingfunctionmod2with +(1) - 1,we obtain

1

3n+1 sn+l

Trn+l

7n+l +

2n+2n!Qn()

(7)

for even n.

Example2. By use of alternatingfunctionsmod4, we have

1
1+

- - 5n+1- 7n+l + gn+l + lln+l -n

3n+1

477n+l
4n+ln!t18
KnvJ

if n is efiren,and
1

1- 3n+l - 5n+l + 7n+l + gn+l

lln+l

n+1
4n+ln! n( )

l/;5

Tr

if n is odd. Cf. [7], [8].

Example3. Dirichlet L-series are defined as follows. Let X be a homomorphism

from the units of Z/kZ to the nonzero complexnumbers.Extend X to a function
on Z (called a Dirichlet charactermod k) by defining x(n) to be X(kZ + n) if
kZ + n is a unit of Z/kZ (i.e. (n, k) = 1), and 0 otherwise(so X is periodicmod k
as defined above).For complex s, the Dirichlet L-series correspondingto X is

L(ssX) = 5, X( ) = [l (1 _ X(P) )
n=1
pprime
p
If X is a Dirichlet charactermod k, then either x(k - 1) = 1 or x(k - 1) = -1:
X is called even in the first case and odd in the second (which agrees with our
previous terminology).By Theorem 4.4, for X a Dirichlet charactermod k and
n > 1 an integer,

L(n,x) = 2kn(n-1)!

X(j)Pn-1(Ct k )

if X and n are both even or both odd. Cf. [1], [11].

5. -CONCLUSION.We have emphasizedthroughoutthe symmetrybetween the Pn
and the Qn.Mostly they are developed in parallel, but in the recurrencesand
functionalequationof 3 they intertwinein an essentialway.
It is more usual to state the Maclallrinseries for tangent and secant, and the
closed form for the series in (6) and (7), in terms of Bernoulliand Euler numbers
(see e.g. [2]).The knowledgeablereadermaybe wonderinghow the Pnand Qnare
relatedto the Bernolllliand Euler polynomials.It turnsout that the rationalvalues
of the Bernoulli and Euler polynomialscan be expressedin terms of the Pnand
Qn,but the relationis not a simpleone. For the questionsconsideredhere, Pnand
Qnseem more natural.
1995]

DERIVATIVE POLYNOMIALS FOR TANGENT AND SECANT

This content downloaded from 193.52.245.246 on Sat, 16 Jan 2016 00:29:05 UTC
All use subject to JSTOR Terms and Conditions

29

REFERENCES
1. Tom M. Apostol, Dirichlet L-functions and character power sums, J. NumberTheory2 (1970).
2. M. Ambramowitz and I. A. Stegun, Handhookof Mathematica!Functions,National Bureau of
Standards, Washington, D.C., 1964.
3. M. D. Atkinson, How to compute the series expansions of sec x and tan x, Am. Math.Monthly93
(1986), 387-389.
4. L. Euler, De inventione integralium si post integrationom variabili quantitati determinatus valor
tribuatur, Misc. Berolin.7 (1743), 129-171, Reprinted in OperaOmnia, ser. I, vol. 17, B. G.
Tuebner, Berlin, 1914, pp. 35-69.
5. L. Euler, De summis serierum reciprocarum ex potestatibus numerorum naturalium ortarum
dissertatio altera in qua eaedem summationes ex fonte maxime diverso derivantur, Misc.Berolin.7
(1743), 172-192, Reprinted in Oper Omnia, ser. I, vol. 14, B. G. Tuebner, Berlin, 1924 pp.
138-155.
6. L. Euler, Introductioin AnalysinInfinitorum,Lausanne, 1748, Reprinted in OperaOmnia,ser. I,
vol. 8, B. G. Tuebner, Berlin, 1922.
7. J. W. L. Glashier, On the Bernoullian function, Quarterly
J. of P>reand Appl. Math.29 (1898),
1-168.
8. J. W. L. Glashier, On the coefficients in the expansions of cosx/cos2x
and sinx/cos2x,
QuarterlyJ. of PureandAppl. Math.45 (1914), 187-222.
9. Donald E. Knuth and Thomas J. Backhotz, Computation of Tangent, Euler, and Bernoulli
Numbers, Mathematics
of Computation21 (1967), 663-688.
10. C. Krishnamacharyand M. Bhimasena Rao, On a table for calculating Eulerian numbers based on
a new method, Proc. LondonMath. Soc. (2)22 (1923), 73-80.
11. I. J. Zucker and M. M. Robertson, Some properties of Dirichlet L-series, J. Phys. A: Math.Gen.
9 (1976), 1207-1214.

Departmentof the Navy

UnitedStatesNaval Academy
Annapolis,MD 21402-5000
[email protected]

Wandering Thoughts
A century later, JonathanEdwardswould be
keeping strict accountsof his spirituallifc in a
journal he begins... when he is ninotcen. The
journalseems to have grown out of a scries of
resolutionsEdwardsmade, and againstwhich he
would sometimesdovetailthc recordof his actual
conduct.... He canberatehimselfforfecling"dulls
dry,and dead"orl a givenday,and althoughhe is
awareof the dangersto one's healthfromcxcessivc
self-mortification,
he has enough resolutenessto
commit himself to an occasionalcold shcowerof
mathematics.s<WhenI am violently beset with
temptations,or cannotrid myselfof evil thoughts,"
he resolves,on July 27 1723,"to do some Arithmotic, or Geometry,or some other study,which
necessarily
engagesall mythoughts,andunavoidably
keepsthemfromwandering."
From A B(ok of One's()wn by ThomasMallon,
pp. 106-1077Ticknor& Ficlds,Ncw York, 1984

30

DERIVATIVE POLYNOMIALSFOR TANGENT AND SECANT

This content downloaded from 193.52.245.246 on Sat, 16 Jan 2016 00:29:05 UTC
All use subject to JSTOR Terms and Conditions

[January