GENERAL  ARTICLE

Infinite Descent – but not into Hell!

Shailesh A Shirali

I nduct ion and D escent : Com plem ent ar y Pr inci-

ples
Everyone knows t he principle of mat hemat ical induc-
t ion (PMI for short ); it is now st andard fare even at t he
high-school level. Curiously, very few seem t o know it s
close relat ive { t he principle of descent (PD for short );
Shailesh Shirali has
been at the Rishi Valley
curious, because t he two principles are complement ary
School (Krishnamurti t o one anot her. In t his art icle we st udy some typical
Foundation of India), applicat ions of t his principle.
Rishi Valley, Andhra
Pradesh, for more than The PMI st at es t he following. Let S be a subset of t he
ten years and is cur- set of nat ural numbers N wit h t he property t hat (a)
rently the Principal. He 1 2 S, and (b) if k 2 S, t hen k + 1 2 S t oo. Then
has been involved in the
it must be t he case t hat S = N . It can be st at ed in
Mathematical Olympiad
Programme since 1988.
`cont raposit ive form' t oo. Let S be a subset of t he set
of nat ural numbers N wit h t he property t hat (a) 1 2 S,
(b) if k 2= S for some k 2 N , k > 1, t hen k ¡ 1 2= S.
Then it must be t he case t hat S = N . The PMI looks
`obvious' but it possesses surprising power, and a great
many non-t rivial result s owe t heir proofs t o it . For a
survey of some applicat ions, see [1].
The principle of descent , which was ¯ rst enunciat ed and
used by Fermat, may be st at ed in various equivalent
ways, e.g.:
1. I t is not possible to have an in¯ nite, monotonically
decreasing sequence of positive integers.
2. Let S be a non-empty subset of the set of natural
numbers with the property that for any positive integer
k greater than 1, if k 2 S then k ¡ 1 2 S too. Then it
must be the case that 1 2 S.
Keywords
Descent, induction, irrational- 3. Let S be a subset of the set of natural numbers with
ity, regular polygons. the property that 1 6
2 S, and for any positive integer k,

42 RESONANCE  February 2003

 GENERAL  ARTICLE

if k 2 S then k ¡ 1 2 S too. Then S is the empty set. Every rational

number between 0
The close relat ionship of t hese st at ement s t o t he PMI is and 1 can be
easy t o see.
represented as a
Inasmuch as the PMI and t he PD are equivalent , it may sum of distinct
be wondered whet her it is wort h our while making a Egyptian fractions.
separat e st udy of proofs based on t he PD. T he answer
is: most assuredly, yes! { one must not conceal t he
richness of t he part icular under t he cloak of t he general!
Accordingly, we now present various `case st udies' t hat
show t he descent principle in act ion.
T he U nit Fr act ion A lgor it hm
Fract ions wit h unit numerat or are known as unit frac-
tions, or (more exot ically) as Egyptian fractions, because
of t he pract ice in ancient Egypt of represent ing fract ions
as sums of dist inct unit fract ions. Thus, 5=13 would be
writ t en as

5 1 1 1
= + + :
13 3 20 780
It is easy t o show t hat every posit ive rat ional number
between 0 and 1 can be writ t en in t his way, and t he
algorit hm t o do so is part icularly nice; we simply peel
o® t he largest unit fract ion not great er t han t he given
fract ion, t hen do t he same for t he port ion remaining,
and so on, recursively. We shall show via t he PD t hat
t his process necessarily t erminates, and so yields t he
desired represent at ion. (For example, for t he fract ion
5=13, we ¯ rst peel o® 1=3. The port ion left is 2=39, from
which we peel o® 1=20. Now t he port ion left is 1=780,
itself a unit fract ion. So 5=13 = 1=3 + 1=20 + 1=780, as
given above.)
The validity of t he algorit hm may be shown as follows.
Let r be any rational number between 0 and 1, say r =
a=b, where a; b are int egers wit h 0 < a < b. Let t he

RESONANCE  February 2003 43

 GENERAL  ARTICLE

The algorithm posit ive int eger n be (uniquely) ¯ xed by t he condit ion
described here for
expression as a sum
1 a 1
· < :
of Egyptian fractions n b n¡ 1
is called the greedy If equality holds on t he left side, t hen t here is not hing
algorithm. There are more t o be done. If not , let s = r ¡ 1=n; t hen
many algorithms of
this kind to be found a 1 na ¡ b
s= ¡ = :
in mathematics e.g., b n nb
some implemen- By de¯ nit ion 0 < na ¡ b < a, so the numerator of s is
tations of the simplex strictly less than that of r . If s is a unit fract ion, t hen
method used to we are t hrough. If not , we repeat t he process wit h s in
solve linear place of r and obt ain anot her fract ion t wit h a smaller
programs. numerat or; and so we proceed. As t he numerat ors of
r; s; t; : : : form a st rictly decreasing sequence of posit ive
int egers, t he descent must come t o a halt sooner or lat er.
The desired represent at ion is now at hand.
We make t he following remarks before moving on. (a)
While t he above algorit hm cert ainly yields a represent a-
t ion in t he desired form, it does not always do so in t he
most economical way, i.e., wit h t he least number of sum-
mands. (Work out t he algorit hm wit h r = 47=60 and
check for yourself !) (b) The above algorit hm, for ob-
vious reasons, is called a `greedy algorit hm'. There are
many algorit hms of t his kind to be found in mat hemat -
ics (e.g., some implementat ions of t he simplex met hod
used t o solve linear programs). (c) A similar analysis
can be made for t he well-known algorit hm of Euclid's
t hat yields t he GCD of two given whole numbers; here
t he st ep carried out repeat edly is

(a; b) 7
! (b; Rem(a ¥ b)) ;

( Rem(a¥ b) refers t o t he remainder when a is divided by

b). The remainders form a strictly decreasing sequence
of non-negative int egers, which t herefore must reach 0
at some st age; t erminat ion of t he algorit hm is assured
because of t his.

44 RESONANCE  February 2003

 GENERAL  ARTICLE

Pr oofs of I r r at ionalit y of Cer t ain N umb er s

p
The classic proof of t he irrat ionality of 2, due to t he
Pyt hagorean school, is well known; it dependsult imat ely
on t he fundament al theorem of arit hmet ic (`prime fac-
t orizat ion in N is unique'). Much less well known is t he
proof by descent ; it s surprising feat ure is t hat unique
fact orizat ion is not used anywhere. Here are t he det ails.
p p
Assuming t hat 2 is rat ional, let 2 = a=bp where a; b
are posit ive int egers. The equat ion yields b 2 = a, so
it follows t hat t he set
n p o
S = n 2 N : n 2 is an int eger
p
is non-empty. Now observe t hat if b 2 S t hen b 2¡ b 2
S t oo, because
p ³ p ´ p
2 b 2 ¡ b = 2b¡ b 2 = int eger ¡ int eger = int eger;
p p
also, b 2 ¡ b = b( 2 ¡ 1) is less t han b. It follows t hat
for each number in S, t here exists anot her one, smaller
t han it self. Now we invoke pt he PD t o deduce t hat S is
empty, in ot her words t hat 2 is irrat ional.
Ot her square-root irrat ionalit ies may be shown
p in like
manner. Thus, t o show t he irrat ionality
p of K , where
K 2 N is notp square, we let c = [ pK ]; t hen if b 2 N is
such t hat b K 2 N , t he number b K ¡ cb t oo has t he
same property, andp it is a posit ive int eger smaller than
b. So, by t he PD, K is irrat ional.
It may not be t oo obvious how one could show t he ir-
rat ionality of, say, t he cube root of 2 using descent ; but
see t he sect ion on polynomials wit h int egral coe± cient s.
I r r at ionalit y { T hr ough G eomet r y!
In some cases a descent proof of irrat ionality has an
elegant geomet
p ric analogue. Consider for example t he
number 2.

RESONANCE  February 2003 45

 GENERAL  ARTICLE

p
Suppose t hat 2 is ratpional; t hen t here exist posit ive
int egers n such t hat n 2 is int egral, so t he set S of
posit ive int egers n such t hat t he diagonal of a square of
side n has int egral lengt h is non-empty. Let b 2 S, and
let AB CD bea p squarewit h side b; then t he diagonal D B
has lengt h b 2 = a, say (see Figure 1). Locat e a point E
on D B such t hat D E = b; t hen B E = a¡ b. Const ruct a
Figure 1. square B E F G on side B E ; t hen E F = a ¡ b. Triangles
D E F and D CF are congruent, so CF = E F = a ¡ b,
t herefore B F = b ¡ (a ¡ b) = 2b ¡ a. It follows t hat
B E F G is an int eger-sided square whose diagonal has
int egral lengt h, and it is smaller t han t he square we
st arted wit h. Now we invoke t he PD t o conclude t hat p
such a sit uat ion is unt enable; t he irrat ionality of 2
follows.
Here is anot her such proof, even simpler t han p t he one
above (it was feat ured in [3]). Suppose t hat 2 = a=b
where a; b 2 N . Let AB CD be a square of side a.
Draw squares B PQR and D STU each wit h side b inside
square AB CD (see Figure 2). The equat ion a2 = 2b2 im-
plies t hat t he area of square AB CD is t he sum of t he ar-
eas of squares B PQR and D STU. This means t hat t he
area of t he cent ral square, wit h side b¡ (a¡ b) = 2b¡ a,
equals t he sum of t he areas of t hep two small squares at
t he corners (wit h side a ¡ b); so 2 = (2b¡ a)=(a ¡ b).
Now we invoke t he PD as earlier t o reach t he same con-
clusion.
I rrationality of the Golden Ratio

Figure 2. The same idea can be used t o show t he pirrat ionality of

t he `golden rat io' { t he number ¿ = 12 ( 5 ¡ 1). Here
we use t he fact t hat in a regular pent agon, t he rat io of
t he diagonal to t he side is ¿. Suppose t hat ¿ is rat ional,
say ¿ = a=b, where a; b 2 N . Let AB CD E be a regu-
lar pent agon wit h side b; t hen each diagonal has lengt h
a. Let t he diagonals AC; B D ; CE ; D A; E B be drawn.
Each diagonal get s divided int o t hree part s, wit h lengt hs

46 RESONANCE  February 2003

 GENERAL  ARTICLE

x; y; x, say, where y < x (t he lengt h of t he small port ion

in t he middle is y). We see t hat b = x + y and a = 2x + y.
Solving for x; y we get x = a ¡ b and y = 2b ¡ a; so x
and y are int egers! T he (small) inner pent agon is now
seen t o be int eger-sided (wit h side y), and it s diagonal
has int egral lengt h (equal t o x). Now using t his obser-
vat ion recursively, we get an endless sequence of smaller
and smaller int eger-sided regular pent agons, clearly an
impossibility.
R egular Poly gons in L at t ices
We shall show t he following non-exist ence result s in t his
sect ion.
(a) An equilat eral t riangle cannot be imbedded in t he
square lat t ice L = Z2 . In ot her words, it is not possible
t o ¯ nd dist inct lat t ice point s A; B ; C in t he coordinat e
plane R2 such t hat t riangle AB C is equilat eral.
(b) A square similarly cannot be imbedded in an equi-
lat eral lat t ice.
To prove (a), we use t he observat ion t hat a quart er-t urn
cent ered at any point of the square lat t ice L maps t he
lat t ice back ont o it self.
Suppose t hat t here exist s an equilat eral t riangle AB C
wit h A; B ; C dist inct point s of L . Let it s side be s.
Consider t he squares erect ed on sides B C; CA; AB re-
spect ively, each one overlapping wit h t he given t riangle;
let t heir cent ers be P; Q; R (see Figure 3). The vert ices
of t he squares lie at lat t ice point s, so t he coordinat es of
P; Q; R are half-int egers. Triangle PQR is clearly p equi- Figure 3.
1
lat eral. Comput at ions show t hat it s side t is 2 ( 3 ¡ 1).
An enlargement about t he origin by a fact or of 2 now
0 0 0
yields a latp t ice point equilat eral t riangle P Q R , wit h
0
side t = ( 3¡ 1)s ¼ 0:73s. We t hus get anot her lat t ice
point equilat eral t riangle, wit h sides less t han 3=4 of t he
original one. Recursively cont inuing t his process, we get
an in¯ nit e sequence of latt ice point equilat eral t riangles,

RESONANCE  February 2003 47

 GENERAL  ARTICLE

wit h side t ending t o 0. This is clearly not possible. We

conclude t hat such t riangles do not exist at all.
The proof of (b) is similar; now we use t he observat ion
t hat a 60± t urn about any point of an equilat eral lat t ice
E maps t he lat tice back ont o it self.
Suppose t hat AB CD is a square wit h A; B ; C; D 2 E;
Figure 4.
let it s side be s. We locat e point s P; Q; R; S wit hin
t he square such t hat t riangles PAB , QB C, RCD , SD A
are equilat eral (see Figure 4); t hen P; Q; R; S are lat t ice
point s, andpPQRS ispa square. Computat ions show t hat
its side is ( 3¡ 1)s= 2 ¼ 0:52s. Now arguing as we did
earlier, we conclude that a square cannot be imbedded
in E.
We leave it t o t he reader t o ¯ nd a proof by descent
showing t hat t he lat t ice L does not cont ain a regular
n-sided polygon for any n > 4.
In t he t hree-dimensional lat tice Z3, equilat eral t rian-
gles do indeed exist ; e.g., t he t riangle wit h vert ices at
(1; 0; 0), (0; 1; 0) and (0; 0; 1) is equilat eral! The reader
should ¯ nd out why and where t he proof given above
fails in t his case.
D iophant ine Equat ions
There are many diophant ine equat ions which possess no
solut ions in int egers and sometimes t he non-exist ence of
solut ions may be shown in a nice manner via t he PD.
We consider two examples.
(a)The equation x 2 + y2 + z2 = 2xyz possesses no solution
in positive integers.
Suppose t hat t here exist posit ive int egers x; y; z such
t hat x 2 + y2 + z2 = 2xyz. Then eit her just one or all
t hree of x; y; z are even. If just one of t hem were even,
t hen we would have 2xyz ´ 0 (mod 4), x 2 + y2 + z2 ´
2 (mod 4), an absurdity; so all of x; y; z must be even.
Writ e (x; y; z) = 2(a; b; c); t hen a; b; c are posit ive int e-

48 RESONANCE  February 2003

 GENERAL  ARTICLE

gers and 4(a2 + b2 + c2 ) = 16(abc), so a2 + b2 + c2 = 4abc. Any number of the

The argument just given can now be repeat ed verbat im form n1/k, with n, k
t o conclude t hat a; b; c t hemselves are even int egers. Us- ∈ N, is either an
ing this st ep recursively and invoking t he PD, we deduce integer or is
t he st at ed assert ion. irrational; it cannot
be a non-integral
(b) The equation x 2 + y2 = 3z2 possesses no solution in
rational number.
positive integers.
Suppose t hat x; y; z are posit ive int egers sat isfying t he
given relat ion. Since 3 j 3z2, if one of x; y is a mult iple
of 3, t hen t he ot her one must be, t oo. If neit her of x; y
is a mult iple of 3, t hen we get x 2 + y2 ´ 2 (mod 3),
which cannot be; so 3 j x; y and t herefore 9 j x 2 + y2 .
This implies t hat 3 j z, so 3 j x; y; z. Writ ing (x; y; z) =
3(a; b; c), where a; b; c are posit ive int egers, we get a2 +
b2 = 3c2, and now t he same reasoning applies all over
again, verbat im. T he st at ed assert ion follows.
Poly nom ials wit h I nt egr al Coe± cient s
Here is a proposit ion about polynomials over Z t hat we
prove using descent : I f ® is a rational root of a monic
polynomial f with integral coe± cients, then ® is an in-
teger.
Suppose not ; let ® be a rat ional but non-int egral root of
a monic n t h degree polynomial wit h int egral coe± cient s.
The following st at ement s may now be made: (i) ®n and
all higher powers of ® are int egral linear combinat ions of
®i for i = 0, 1, 2, . . . , n ¡ 1; (ii) t here exist s a posit ive
int eger k such t hat k®i is int egral for i = 0, 1, 2, . . . ,
n ¡ 1; (iii) t he set of nat ural numbers k such t hat k®i is
an int eger for all natural numbers i is nonempty; (iv)
for any element k of t his set , t he number k 0 = k(®¡ [®])
is a posit ive int eger smaller t han k, and wit h t he same
property. Now, via t he PD, we are done.
An immediat e corollary of t his result is that any number
of t he form n1=k , wit h n; k 2 N , is eit her an int eger or is
irrat ional; it cannot be a non-int egral rat ional number.

RESONANCE  February 2003 49

 GENERAL  ARTICLE

T wo I M O P r oblem s
We t op o® t his survey of case st udies wit h two challeng-
ing and pret ty problems from t he Int ernat ional Mat h-
emat ical Olympiads of 1986 and 1988 (held in Poland
and Aust ralia, respect ively). The latt er problem was
considered for long t o be t he most di± cult problem ever
t o be asked in an IMO. (However, it has since lost t his
t it le!)
I MO 1986/ 3.
To each vertex of a regular pentagon an integer is as-
signed in such a way that the sum of all the ¯ ve num-
bers is positive. I f three consecutive vertices are assigned
the numbers x; y; z respectively and if y < 0, then the
following operation is allowed: the numbers x; y; z are
replaced by x + y; ¡ y; z + y, respectively. Such an opera-
tion is performed repeatedly as long as at least one of the
¯ ve numbers is negative. Determine whether this proce-
dure necessarily comes to an end after a ¯ nite number
of steps.
We not e ¯ rst ly t hat t he sum s of t he numbers at t he
vert ices st ays t he same all t hrough, because (x + y) +
(¡ y) + (z + y) = x + y + z.
Experiment at ion suggest s t hat t he procedure cannot be
cont inued inde¯ nitely. If we are t o prove t hat t he proce-
dure necessarily t erminat es, we must ¯ nd a non-negat ive
int eger-valued funct ion of t he vert ex numbers t hat st ric-
t ly decreases at each st age. Finding such a function en-
t ails a bit of hit -and-t rial, but in t he end we obt ain t he
following candidat e for t he funct ion, f : if t he numbers
at t he vert ices are x; y; z; u; v, read in cyclic order, t hen
f (x; y; z; u; v) = (x¡ z) 2 + (y¡ u) 2+ (z¡ v) 2 + (u¡ x) 2 + (v¡ y) 2 :
Observe t hat f is non-negat ive and int eger-valued, as
required.
Assuming t hat y < 0, let t he st ep (x; y; z; u; v) 7
! (x +

50 RESONANCE  February 2003

 GENERAL  ARTICLE

y; ¡ y; z + y; u; v) be performed. Let us now see what

e®ect it has on f . We have,

f (x + y; ¡ y; z + y; u; v) = (x ¡ z) 2 + (u + y) 2+

(z + y ¡ v) 2 + (x + y ¡ u) 2 + (v + y)2 :
Simplifying, we get t he following expression for ¢ f , t he
change in f :

¢ f = 2y(x+ y+ z+ u+ v) = 2ys < 0 (since s > 0; y < 0):

It follows t hat t he f -value strictly decreases as a result

of t he operat ion. So the operat ion cannot be performed
in¯ nit ely oft en; sooner or lat er t he numbers at t he ver-
t ices will all be non-negat ive.

R em ar k . What happens if t he numbers placed at t he

st art at t he vert ices are not int egers? So long as t hey are
rat ional, exact ly t he same idea works (we simply scale
up everyt hing by t he LCM of t he denominat ors). But
what happens if some of t he numbers are not rat ional?
It t urns out t hat in t his case t oo t he procedure must
t erminat e; however, t he proof has t o be worded very
di®erent ly now. Here is a possible approach. Let t he
vert ices be numbered 0, 1, 2, 3, 4, let t he label on vert ex
i be x i , and for i = 0, 1, 2, 3, 4 and j > i let t he sums
ai ;j be comput ed as follows:

ai ;j = x i + x i + 1 + ¢¢¢+ x j ¡ 1;

wit h x 5 = x 0 , x 6 = x 1 , and so on. Not e t hat t here are

in¯ nit ely many such sums, but only a ¯ nit e number of
t hem can be negat ive, because on cycling around t he full
set of vert ices we add s t o t he sum, and s is posit ive.
Now suppose t hat x r < 0 and (x r ¡ 1 ; x r ; x r + 1 ) 7
! (x r ¡ 1
+ x r ; ¡ x r ; x r + 1 + x r . Let us check what happens t o t he
various sums ai ;j . For convenience we writ e bi ;j for t he
updat ed sums.

RESONANCE  February 2003 51

 GENERAL  ARTICLE

If all t hree indices r ¡ 1; r; r + 1 are part of t he range

[i; j ], or none of t hem is, t hen bi ;j = ai ;j . Now suppose
t hat eit her one or two of t he indices r ¡ 1; r; r + 1 are
part of t he range [i; j ]. We now get :

bi ;r = ai ;r + x r = ai ;r + 1 ; bi ;r + 1 = ai ;r ;
br ¡ 1;r + 1 = x r ¡ 1 = ar ¡ 1;r ; br ;r + 1 = ¡ x r = ¡ ar;r + 1:
We observe t hat in t he in¯ nit e mult iset of t he sums ai ;j ,
most of t he element s st ay unchanged, some element s
swap places wit h ot hers, and precisely one element is
replaced by it s negat ive. As a negat ive number is re-
placed by a posit ive number each t ime, t he sequence of
operat ions necessarily t erminat es. Indeed, we can say
more: the number of steps needed for termination does
not depend on the order in which the steps are carried
out! { it equals t he number of negat ive element s in t he
init ial mult iset of values of ai ;j . An elegant and pret ty
result indeed.
Is t his a proof by descent ? Oh yes! { but t he descent
principle has been used in a rat her more subt le manner.
I MO 1988/ 6.
Let a and b be positive integers such that the quantity
a2 + b2
c=
ab+ 1
is an integer. Show that c is a square.

Solut ion. We shall prove a st ronger st at ement : Let

a; b; c be positive integers such that 1 · a2 + b2 ¡ abc ·
c + 1; then the quantity a2 + b2 ¡ abc is a square.
Let d denot e t he value of a2 + b2 ¡ abc; t hus 1 · d · c+ 1.
If c = 1 t hen d = 1 or 2. If d = 1, t here is not hing t o
prove; and d = 2 cannot happen at all, for a2 + b2 ¡ ab
is odd if at least one of a, b is odd, and is a mult iple of
4 if a, b are bot h even. If c = 2, t hen t he result follows

52 RESONANCE  February 2003

 GENERAL  ARTICLE

t rivially, wit h d = (a ¡ b) 2 . In t he discussion below we

assume that c > 2.
If d = 1, t here is not hing t o prove. If d > 1, we consider
t he curve ¡ d ½ R2 de¯ ned by

¡ d := f (x; y) : x 2 + y2 ¡ cxy = dg:

This is t he equation of a hyperbola symmetric in t he

lines y = § x. Our int erest lies in t he lat t ice point s on
¡ d , of which (a; b) is one such. The crucial observat ion
t hat we make is t his: from a single lattice point of ¡ d
one can by descent generate in¯ nitely many such points.
The descent is accomplished as follows.
Let A (u; v) be a lat t ice point on t he upper branch of
¡ d ; t hen v > u. We move vert ically down from A t o
t he line y = x (see Figure 5), meet ing it at B (u; u);
t hen we move horizont ally (t o t he left ) from B t o ¡ d ,
meet ing it at C. To ¯ nd the coordinates of C, not e t hat
t he image of A under re° ect ion in t he line y = x is t he
point A 0 (v; u). So one point of int ersect ion of t he curve
and t he line y = u is (v; u); t hat is, one root of t he
quadrat ic equat ion

x 2 ¡ cux + (u2 ¡ d) = 0

is x = v. Since t he sum of t he root s of t he equat ion is

cu, t he ot her root must be x = cu ¡ v, implying t hat
C = (cu ¡ v; u). Since c; u; v are int egers, it follows t hat
C is a lat tice point . Not e t hat C and A lie on t he same Figure 5.
branch of t he curve. y
A
So we have moved from t he latt ice point (u; v) t o t he y=x
lat t ice point (cu¡ v; u); t his is t he descent st ep, which we
C
iterate. Each such st ep carries us from one lat t ice point B
x
of ¡ d t o anot her one, on t he same branch. As t he curve
has posit ive slope at every point, t he movement result s
in a st rict decrease in bot h coordinat es. The descent
result s in an event ual passage int o t he t hird quadrant .

RESONANCE  February 2003 53

 GENERAL  ARTICLE

The proof that the Since d · c+ 1, t he curve cannot have any lat t ice point s
continued fraction in t he int eriors of t he second and fourth quadrant s; for
approach yields all if (x; y) were such a lat t ice point , t hen xy · ¡ 1 and
possible non- x 2 + y2 ¸ 2, and so 2 · x 2 + y2 = cxy + d · 1, which is
negative integral absurd.
solutions to Pell’s
Now it follows from t he nat ure of t he descent t hat we
equation is another cannot jump from t he int erior of t he ¯ rst quadrant t o
example of a task t he int erior of t he t hird quadrant in a single st ep; for
that can be if v > u > 0 t hen t he signs of t he second coordinat es
accomplished in (u; v) and (cu ¡ v; u) are bot h posit ive. Also, we are
elegantly via the PD. barred from ent ering t he int erior of t he second quad-
rant . T herefore in order t o reach the int erior of t he
t hird quadrant , we must st ep upon p bot h t phe x-axis and
t he y-axis, i.e., on t he point s (0; d), (¡ d; 0), which
consequent ly must be lat t ice point s. It follows t hat d is
a square.
Here is anot her (rat her short ) proof of t he above asser-
t ion. We st art wit h nat ural numbers a; b; c wit h a2 +
b2 ¡ abc ( = t; say) between 1 and c + 1.
= b; for a = b gives a2(2¡ c) =
If t is not a square, then a 6
t which gives in t urn c = 1 and t = a2 , a square.
Assume t hat it is possible t o have t non-square; assume
t hat t he corresponding a > b has b t he least possible (of
course, b > 0). Then t he `ot her' root d of t he equat ion
a2 + b2 ¡ abc = t, viewed as a quadrat ic equat ion in
a, sat is¯ es t he relat ions a + d = bc, ad = b2 ¡ t. In
part icular, d is an int eger.
If d < 0, t hen we get , in t urn, db · ¡ 1, dbc · ¡ c,
c · ¡ dbc, c+ 1 · 1¡ dbc. So d2 + b2 ¡ dbc = t · c+ 1 ·
1 ¡ dbc, i.e., d2 + b2 · 1, an impossibility; so d ¸ 0.
If d = 0, t hen we get t = b2 , a cont radict ion; so d > 0.
Now d = bc¡ a. If d ¸ b, t hen we get , in t urn, bc¡ a ¸ b,
bc ¸ a+ b, abc ¸ a2 + ab > a2 + b2 . That is, a2 + b2 ¡ abc <
0, which is not possible. Hence d < b.

54 RESONANCE  February 2003

 GENERAL  ARTICLE

Since 1 · b2 + d2 ¡ dbc = t · c+ 1, we obt ain a `smaller' Suggested Reading

solut ion (b; d) in place of (a; b), t he value of t being non-
[1] B Sury, Mathematical In-
square (it is t he same t). Now PD leads t o a cont radic- duction – an Impresario of
t ion. It follows t hat t is a square. the Infinite, Resonance,
Vol.3, pp. 69-76, 1998.
Family connections [2] G H Hardy and E M Wright,
Introduction to the Theory of
There are many cousins t o IMO 1988/ 6 (in it s original Numbers, Cambridge Uni-
form). T he reader may enjoy trying t o prove t he follow- versity Press, 1960.
ing. (a) I f a; b are positive integers such that the number [3] K Puly, The square root by
c = (a2 + b2 )=(ab ¡ 1) is an integer, then c = 5. (b) inifnite descent, Resonance,
Vol.5, No.8, p.83, 2000.
I f a and b are positive integers such that a2 + b2 ¡ a is
divisible by 2ab, then a is a perfect square.
Concluding R em ar ks
We have cert ainly not exhausted t he list of applicat ions
of t he principle of descent ! Ot her elegant applicat ions
include t he proof by Fermat of his claim t hat t he equa-
t ion x 4 + y4 = z2 has no solut ions in posit ive int egers;
t he proof by Euler of Fermat 's t heorem st at ing t hat any
prime of t he form 1 (mod 4) is a sum of two squares; and
t he proof by Lagrange of his own t heorem that every
prime is a sum of four squares. (Euler knew of t his re-
sult but was not able t o prove it . Ironically, t he met hod
used by Lagrange is pract ically t he same as t hat used
by Euler t o prove Fermat 's claim.) (See [2] for det ails
of bot h proofs.) T he proof that t he continued fract ion
approach yields all possible non-negat ive int egral solu-
t ions t o Pell's equat ion is another example of a t ask t hat
can be accomplished elegant ly via t he PD. More appli-
cat ions could be cat alogued, but we leave t he t ask t o
t he reader.
A cknowledgem ent s
Address for Correspondence
Thanks are due t o my friend and colleague Professor Shailesh A Shirali
B Sury for several helpful suggest ions t hat have been Rishi Valley School
incorporat ed here. Chittoor District
Rishi Valley 517 352
Andhra Pradesh, India.

RESONANCE  February 2003 55