ACTA ARITHMETICA

LXV.4 (1993)

The diophantine equation x2 + C = y n

by

J. H. E. Cohn (London)

1. Introduction. Many special cases of the equation of the title, where

x and y are positive integers and n ≥ 3, have been considered over the years,
but most results for general n are of fairly recent origin. The earliest reference
seems to be an assertion by Fermat that he had shown that when C = 2,
n = 3, the only solution is given by x = 5, y = 3; a proof was published
by Euler [14]. The first result for general n is due to V. A. Lebesgue [18]
who proved that when C = 1 there are no solutions. Nagell [29] proved
that there are no solutions for C = 3 and 5, but did not complete a proof
for C = 2. Ljunggren [20] generalised Fermat’s result and proved that for
C = 2 the equation has no solution other than x = 5, a result rediscovered
by Nagell [31], who also showed [32] that when C = 4 the only solutions are
x = 2 and x = 11. Chao Ko [6] proved that x = 3 is the only solution for
C = −1, a result which had been sought for many years as a special case of
the Catalan conjecture.
It follows from [36, Theorem 12.2], itself an extension of a deep analytical
result [35], that for any given C there are but finitely many solutions, which
are effectively computable in the usual sense, viz., that it is possible to find
them all by considering all values of x up to a bound K(C) which can be
explicitly calculated. In practice, the power of that method is limited by
the huge size of the K that arises, but it provides a theoretical method for
solving such problems.
The case in which n is even is easily treated, since then C is to be
expressed as the difference of two integer squares; for n odd there is no loss
of generality in considering only odd primes p, which we shall assume in what
follows. We shall only consider positive values of C; in theory, much of the
following applies also to the case C < 0, but the reason for the restriction to
positive C will become evident below. It clearly suffices to find the possible
values of x.
The special case p = 3 has received extensive treatment, and some results
are known for the special cases p = 5 [2, 37] and p = 7 [3].
368 J. H. E. Cohn

The method for C = 1, 2 and 4 consists √ of two√parts. Firstly, using

unique prime factorisation in the fields Q[ −1] and Q[ −2], it is shown √ that
y = a2 + C for some a. Then the fundamental unit in the field Q[ a2 + C]
is expressed simply in terms of a. For other values of C, even if the first step
can be followed, the second cannot, and a different method is required to
complete the proof. Nagell [33] found such a method for C = 8, and proved
that there are no solutions.
As the list of references shows, there are quite a few results for general p.
Many of these are not so well known, perhaps because they are published
in journals which are not so readily available. There are numerous cases of
duplication of known results. Thus for example the case C = 2, first proved
by Ljunggren [20], was duplicated by Nagell [31]. For C = 3, the result first
proved by Nagell [29] was duplicated by Brown [4], and subsequently, I am
ashamed to say, by Cohn [10]. The early papers mainly use ad hoc methods
for the special values of C considered. Ljunggren’s result in [19] is incomplete
as he himself points out in [21]; in particular, the case C = 25 remained and
remains unsolved. However, a series of papers by Korhonen [15, 16, 17] deals
with numerous values, and the author wishes to thank Professor Schinzel
for drawing his attention to them.
This note, which attempts to be self-contained, attempts to collect to-
gether some of the known results, and to prove some new ones using some
new techniques. The examples in Section 4 are expository in nature, and it is
not suggested that all the results are new; Section 5 contains all the results
we have managed to prove for values of C under 100, and gives complete
solutions in 77 of these cases, of which more than half appear to be new.
We have attempted to list the authors of previously known results, although
in view of the duplication mentioned earlier inadvertent omission is quite
possible.
The author wishes to thank the referee most sincerely for suggesting
some improvements in the presentation of Section 2.

2. Preliminaries. The obvious starting √ point is to factorise

√ the √
equation
in the imaginary quadratic field in which −C lies, (x+ −C)(x− −C) =
y p , and then it is clear that if for integers a and b,
√ √
(A) ±x + −C = (a + b −C)p
then indeed x will be a solution with y = a2 + b2 C. It is not of course
suggested that (A) is necessary for a solution; this would only be the case if
(1) C 6≡ 3 (mod 4);
(2) a question of units did not arise;
(3) the field had unique factorisation;
 The diophantine equation x2 + C = y n 369

(4) C were square-free;

√ and
(5) the factors ±x + −C had no common factor,
but what is true in any event is that (A) is a sufficient condition for a
solution. Since this possibility must be considered for every C, we prove
some results concerning it in the next section. Before doing so, however, we
consider the above questions.
Firstly, if C ≡ 3 (mod 4), we obtain in addition to (A) the sufficient
condition
√ √
(B) ±x + −C = ( 12 (A + B −C))p , A ≡ B ≡ 1 (mod 2)
with y = 14 (A2 +B 2 C). This can occur only if p = 3. For, equating imaginary
parts in (B) yields
(p−1)/2  
X p
p
2 =B Ap−2r−1 (−B 2 C)r ,
r=0
2r + 1
and so since B is odd, B = ±1. Thus
±2 ≡ ±2p ≡ (−C)(p−1)/2 ≡ (−C | p) ≡ 0, ±1 (mod p) ,
which implies that p = 3, and then ±8 = 3A2 − C. Thus we have
Lemma 1. Case (B) occurs if and only if C = 3A2 ± 8 and gives only the
solution p = 3, x = A3 ± 3A.
We next consider the question of units. If p 6= 3, the units in the field
can be absorbed into the power in (A) whatever the value of C. Even if
p = 3, this is still usually the case, since unless C = 3d2 the only units are
±1 and possibly ±i. However, if C = 3d2 , there are six units, ±1, ±ω and
±ω 2 . Apart from (A) or (B), this leads to
√ √
(C) ±x + d −3 = ω( 12 (A + Bd −3))3 , A ≡ B (mod 2) .
Lemma 2. Case (C) occurs if and only if C = 48D6 , x = 4D3 , p = 3.
P r o o f. Equating imaginary parts gives 16d = A3 − 9AB 2 d2 − 3A2 Bd +
3B d and so if r = A − Bd, then 16d = r3 − 12rB 2 d2 − 8B 3 d3 , whence
3 3

r = 2s and so 2d = s3 − 3sB 2 d2 − B 3 d3 . Since the right hand side is to be

even, both s and Bd are even, and so the right hand side is actually divisible
by 8. Then with d = 4δ, s = 2σ we obtain δ = σ 3 −12σB 2 δ 2 −8B 3 δ 3 , whence
σ 3 = δ(1 + 12σB 2 δ + 8B 3 δ 2 ), the factors on the right being coprime. Thus
δ = D3 , σ = Dτ , and τ 3 = 1+12τ B 2 D4 +8B 3 D6 , whence 1 = τ 3 −3τ %2 −%3
with % = 2BD2 . It is known [28, Chap. 23, Th. 8] that this equation has just
the six solutions in integers, (τ, %) = (−3, 2), (−1, 1), (1, 0), (0, −1), (1, −3)
and (2, 1). Since % = 2BD2 must be even and τ > 0, it follows that the √ only
possibility is B = 0, d = 4D3 , A = 4D, C = 48D6 and so ±x + 4D3 −3 =
8ωD3 , whence x = 4D3 , which concludes the proof.
370 J. H. E. Cohn

If we were to allow C < 0, then the question of units would be very much
more complicated, and this explains our restriction to C > 0.
The next question concerns the uniqueness of factorisation in the field.
There are only nine imaginary quadratic fields in which this holds, so that
it might be thought that our discussion was of limited application. √ This,
however, is not the case. Consider the example C = 6; here the field Q[ −6]
has class number h = 2. For any possible solution of the equation,√ (x, 6) = 1
and p is odd. Then the principal ideal π generated by x + −6 and its
conjugate π 0 are coprime with product√[y]p . Thus for √some ideal ξ, π = ξ p ,
p
but the conclusion we desire, that x + −6 = (a + b −6) , does not follow
immediately from this, since ξ is not known to be principal.
√ However, since
2 2
h = 2, it follows that ξ is principal and hence (x + −6) must equal the
pth power
√ of an element in the field. Since p is odd, it then follows that
x + −6 itself is such a power. It will then follow from Lemmas 4 and 5
below that the equation has no solutions at all if C = 6.
The same argument can be applied in other cases; what is important is
not that h = 1, but that p - h. This will hold for all odd primes if h happens
to be a power of 2. Even for other values of h, it will hold for all but finitely
many primes, and only these have to be treated by other methods. Thus for
example, for C = 26, h = 6, and so the above method applies unless p = 3,
which must be considered separately.
Next, if C is not square-free, say C = cd2 where c is square-free, then
we obtain in addition to (A) also
√ √
(D) ±x + d −c = (a + b −c)p
and if c ≡ 3 (mod 4) also
√ √
(E) ±x + d −c = ( 12 (A + B −c))p , A ≡ B ≡ 1 (mod 2) .
Here we obtain from (D) equating imaginary parts
(p−1)/2  
X p
d=b ap−2r−1 (−b2 c)r ,
r=0
2r + 1
from which it follows that b divides d, and then
d/b ≡ (−b2 c)(p−1)/2 ≡ (−b2 c | p) ≡ 0, 1 or − 1 (mod p) .
Here b = ±d leads back to (A) and otherwise p is limited to a finite set for
each such b, and there is then no difficulty is solving the resulting polyno-
mial equations
√ for a. Treatment
√ of (E) is even simpler for we observe that
( 21 (A + B −c))p ∈ Z[ −c] only if c ≡ 3 (mod 8) and p = 3 and then solu-
tions arise only for B | d and 8d/B = 3A2 − B 2 c with 8x = |A3 − 3AB 2 c|.
The most serious difficulties arise from possible common factors, and
here no method seems to be available. Thus for example if C = 7, and x
 The diophantine equation x2 + C = y n 371
√ √
is odd, the factors ±x + −7 are both divisible by 12 (1 + −7) and by
1
√
2 (1 − −7), which leads to great difficulties. The special case, y = 2, of the
equation proposed by Ramanujan [34] was solved by Nagell [30] thirty five
years later. It is now known as the Ramanujan–Nagell equation, and there
is an excellent survey article [9] concerning it.
Possible common factors√arise in one of two ways. As above, there are
the prime factors of 2 in Q[ −C] if C ≡ 7 (mod 8). Secondly, if C = cd2
say, where c is square-free, then it sometimes happens√that x shares a factor
with d which splits into distinct primes in the field Q[ −c]. Thus if C = 25,
then if 5 - x the methods outlined in Section 3 can be applied, but if x = 5X,
y = 5Y then (X + i)(X − i) = 5p−2 Y p . Since 5 = (1 + 2i)(1 − 2i), this can
occur;
√ in fact, it does with x = 10. However, if no prime factor of d splits in
Q[ −c], this cannot happen; for example if C = 9 we cannot have 3 | x since
otherwise x = 3X, y = 3Y , and X 2 + 1 would be divisible by 3. √ It is easily
seen that if q is a prime dividing (x, C) and q does not split in Q[ −c], then
p | κ where q κ is the highest power√of q dividing C; in particular, if C is free
of prime factors which split in Q[ −c] then (x, C) = 1.
Summarising the above, we obtain the following
Theorem 1. Let C > 0, C = cd2 , c square-free, c 6≡ 7 (mod 8). If p is
an odd prime and x2 + C = y p for coprime positive integers x , y then either
√
(a)√there exist integers a, b with b | d, y = a2 + b2 c and ±x + d −c =
(a + b −c)p ; or
(b) c ≡ 3 (mod 8), p √ = 3 and there exist
√ odd integers A, B with B | d,
y = 14 (A2 + B 2 c), ±x + d −c = 81 (A + B −c)3 ; or
√
(c) p | h, the class number of the field Q[ −c]; or
(d) C = 3A2 ± 8, p = 3, x = A3 ± 3A; or
(e) C = 48D6 , p = 3, x = 4D3 .

3. Treatment of (A). From (A) equating imaginary parts gives

(p−1)/2  
X p
1=b ap−2r−1 (−b2 C)r
r=0
2r + 1
and so b = ±1. Thus
(p−1)/2  
X p
(1) ±1 = ap−2r−1 (−C)r
r=0
2r + 1
and it is from this that the remaining conclusions will follow. It is clear from
(1) that a and C must have opposite parity, for if a and C were both even
then the right hand side of (1) would be even, whereas if they were both
372 J. H. E. Cohn

odd then we should obtain

(p−1)/2  
X p
±1 ≡ = 2p−1 ≡ 0 (mod 2) .
r=0
2r + 1

Lemma 3. Let q be any odd prime dividing a, satisfying (1). Then C q−1 ≡
1 (mod q 2 ); unless p = q = 3, q α k a implies q 2α k(C q−1 − 1).
P r o o f. From (1) we see that (−C)(p−1)/2 ≡ ±1 (mod q 2 ). If now
q k(p − 1) with γ ≥ 0, let p − 1 = Hq γ . Then every term except the
γ

last on the right of (1) is divisible by q γ+2 , and so C p−1 ≡ 1 (mod q γ+2 ),
whence C H ≡ 1 (mod q 2 ). But by Fermat’s theorem C q−1 ≡ 1 (mod q) and
thus if K = (H, q − 1) then C K ≡ 1 (mod q). But C H ≡ 1 (mod q 2 ) and
since H is a multiple of K, but not of q, it follows that C K ≡ 1 (mod q 2 ).
Since q − 1 is also a multiple of K, C q−1 ≡ 1 (mod q 2 ).
Finally, unless p = q = 3 we find if q α k a that (−C)(p−1)/2 ∓ 1 is divis-
ible by q 2α+γ and by no higher power of q, and then repeating the above
argument yields q 2α k(C q−1 − 1) without difficulty.
Lemma 4. The minus sign in (1) can occur only if p ≡ 3 (mod 4), and
the following conditions are both satisfied :
1. either
(a) C ≡ 1 or 13 (mod 16), or
(b) C ≡ 0 (mod 8), and p ≡ 7 (mod 8), or
(c) C ≡ 4 (mod 8), and p ≡ 3 (mod 8) ;
2. either
(a) C ≡ 1 (mod 9), or
(b) C ≡ 0 (mod 3) and p ≡ 2 (mod 3), or
(c) C ≡ 4 or 7 (mod 9), and p ≡ 3 (mod 8) .

P r o o f. If C is odd, then a is even, and −1 ≡ (−C)(p−1)/2 (mod a2 ).

So C ≡ 1 (mod 4) and p ≡ 3 (mod 4). If 4 |
a, then C ≡ 1 (mod 16). If
2 k a, then we find with p = 4k + 3 that p2 ≡ 2k + 3 (mod 8) and so
C 2k+1 ≡ 1 + 4(2k + 3) (mod 16), whence C ≡ 13 (mod 16).
2
If C is even then
a is odd. If 8 | C then p ≡ 7 (mod 8); if 2 k C then p ≡ 3
p
(mod 4) and so 3 is odd, whence −1 ≡
p + 4 (mod 8), i.e. p ≡ 3 (mod 8).
Finally, 2 k C would imply −1 ≡ p + 2 p3 (mod 4) or p + 1 ≡ p(p − 1)(p − 2)
(mod 4), which is impossible.
Next if 3 | a then −1 ≡ (−C)(p−1)/2 (mod 9), and so C ≡ 1 (mod 3)
since p ≡ 3 (mod 4). Then by Lemma 3, C ≡ 1 (mod 9).
 The diophantine equation x2 + C = y n 373

If 3 - a then a2 ≡ 1 (mod 3) and C ≡ −1 (mod 3) is impossible since it

would imply
(p−1)/2  
X p
−1 ≡ = 2p−1 (mod 3) ;
r=0
2r + 1
C ≡ 0 (mod 3) yields −1 ≡ pap−1 ≡ p (mod 3) and C ≡ 1 (mod 3) gives
−1 ≡ {(1 + i)p − (1 − i)p }/(2i) (mod 3), which implies p ≡ 3 or 5 (mod 8),
the latter of which we have already seen to be impossible. This concludes
the proof.
Lemma 5. The plus sign in (1) can occur only if the following conditions
are both satisfied :
1. either
(a) C ≡ 3 (mod 4), or
(b) C ≡ 1 (mod 4), 22γ k(C − 1) and p ≡ 1 (mod 4), or
(c) C ≡ 2 (mod 8), and p ≡ 3 (mod 8), or
(d) C ≡ 6 (mod 8), and p ≡ 7 (mod 8);
2. either
(a) C ≡ 2 (mod 3), or
(b) C ≡ 4 or 7 (mod 9), and p ≡ 1 or 7 (mod 8), or
(c) C ≡ 1 (mod 9), and p 6≡ 3 (mod 8), or
(d) C ≡ 21 (mod 27), and p ≡ 1 (mod 3) .
P r o o f. If 4 | C then a is odd and 1 ≡ pap−1 ≡ p (mod 4). Let 2% k(p−1).
Then
   
p−1 p p−3 p
1 = pa − Ca + C 2 ap−5 + . . . ≡ pap−1 (mod 2%+1 ) .
3 5
But now 1 ≡ (p − 1)ap−1 + ap−1 (mod 2%+1 ), and so 2% k(ap−1 − 1), while
%
since 2%+2 | (a2 − 1), 2%+2 | (ap−1 − 1), which is impossible.
If 2 k C, then a is odd, and so
   
p p
1≡p−C +4 (mod 8) .
3 5
If C ≡ 2 (mod 8) this gives
0 ≡ (p − 1){1 − (p2 − 2p)/3 + (p2 − 2p)(p2 − 7p + 12)/30} (mod 8)
≡ (p − 1){1 + 1 − 2p + (1 − 2p)(p − 3)/30} (mod 8)
≡ (p − 1)(p − 3)(1 − 2p)/30 (mod 8)
374 J. H. E. Cohn

and so p ≡ 1 or 3 (mod 8). In exactly the same way we find that if C ≡ 6

(mod 8), then p ≡ 1 or 7 (mod 8). However, in either case we now show that

p ≡ 1 (mod 8) is impossible.

For if 2% k(p−1) with % ≥ 3 then 2% k Cap−3 p3
% 2 p−3 p %+1
and 2 k C a 5 whereas 2 divides all subsequent terms. Thus again
p−1 p−1 p−1
1 ≡ pa ≡ (p − 1)a +a (mod 2%+1 ), which is impossible as before.
If C ≡ 1 (mod 4) then a is even, and 1 ≡ (−C)(p−1)/2 (mod 4) yields
p ≡ 1 (mod 4). Suppose that 2% k(p − 1) where % ≥ 2, that 2α k a with α > 0
and that C = 1 + k · 22γ+3 where k is odd and γ ≥ 0. Then (1) would give
 
(p−1)/2 (p−3)/2 p
1 − (−C) = (−C) a2 + . . .
2
and here the left is divisible by precisely 2%+2γ+2 whereas the right is divis-
ible by precisely 2%+2α−1 , which is clearly impossible.
Next consider the equation modulo powers of 3. If 3 | a then by Lemma 3,
C 2 ≡ 1 (mod 9); if here C ≡ 1 (mod 9), we see from (1) that (p−1)/2 must
P(p−1)/2 p
r
be even, i.e. p ≡ 1 or 5 (mod 8). Otherwise 3 - a and 1 ≡ r=0 2r+1 (−C)
p p
(mod 3). If C ≡ 1 (mod 3), then 1 ≡ {(1 + i) − (1 − i) }/(2i) (mod 3)
whence p ≡ ±1 (mod 8).
Finally, if 3 | C, then 3 - a and so 1 ≡ pap−1 (mod 3). Thus p ≡ 1
(mod 3), and so ap−1 ≡ 1 (mod 9). Then 9 | C is impossible, since it would
imply p ≡ 1 (mod 9), and then if 3% k(p − 1) that 1 ≡ pap−1 ≡ (p − 1)ap−1 +
ap−1 (mod 3%+1 ), whence
3% k(ap−1 − 1), which is impossible. Next if C ≡ 6
(mod 9), then 1 ≡ p − 6 p3 (mod 9). Thus 9 divides (p − 1){1 − p(p − 2)} =
(p − 1){2
− (p − 1)2 }, and so p ≡ 1 (mod 9). Let p − 1 = 3% λ, where 3 - λ.
Then C p3 ≡ −3% λ ≡ 1 − p (mod 3%+1 ) and so (1) yields
1 ≡ pap−1 + p − 1 ≡ (p − 1)(ap−1 + 1) + ap−1 (mod 3%+1 ) ,
which is impossible as before. The last case is C ≡ 3 (mod 9); suppose here
that 3% k(p − 1). Then we use an identity proved in [10],
(p−1)/2   (p−1)/2
X p X
p−2r−1 r p−1
a (−3) = 2 + Ar (a2 − 1)r
r=0
2r + 1 r=2

where for each r, Ar is divisible by 3%+2−r . Inserting this into (1) we obtain
(p−1)/2  
X p
p−1
2 −1≡ ap−2r−1 (−1)r−1 {C r − 3r } (mod 3%+2 ) .
r=0
2r + 1
On the right hand side

of this congruence, the term with r = 0 disappears
and so 2p−1 − 1 ≡ p3 ap−3 (C − 3) (mod 3%+2 ), since as is easily seen, all the
terms with are divisible by 3%+2 . Now let p − 1 = 2k · 3% where 3 - k.

r ≥ 2 %−1
p
Then 3 ≡ −k · 3 (mod 3% ), (C − 3) is divisible by 9, ap−3 ≡ 1 (mod 3)
%
and 2p−1 = (1 + 3)k·3 ≡ 1 + k · 3%+1 (mod 3%+2 ). Substituting these yields
 The diophantine equation x2 + C = y n 375

k · 3%+1 ≡ −k · 3%−1 (C − 3) (mod 3%+2 ) whence cancelling k · 3%−1 yields

C ≡ 21 (mod 27) as required, concluding the proof.
Lemma 6. Let P denote any odd prime dividing C. Then the plus sign
in (1) implies pap−1 ≡ 1 (mod P ), (p | P ) = 1 and if P 6= 3 then p 6≡ 1
(mod P ), whereas the minus sign implies pap−1 ≡ −1 (mod P ), (−p | P )
= 1, and p 6≡ 1 (mod P ).
P r o o f. We have ±1 ≡ pap−1 (mod P ).
Then the upper sign gives (p | P ) = 1. Also if
P > 3 and p ≡ 1
(mod P ), suppose that P % k(p − 1); then P % divides p3 . Thus 1 ≡ pap−1 ≡
(p−1)ap−1 +ap−1 (mod P %+1 ) and so P % k(ap−1 −1). But this is impossible
since as is easily seen if ap−1 − 1 is divisible by P at all, it is divisible by
P %+1 .
The lower sign gives (−p | P ) = 1. The last part follows as above, except
that the condition P > 3 can be eliminated in view of Lemma 4.
Lemma 7. In (1), p = 3 if and only if C = 3a2 ∓ 1, x = 8a3 ∓ 3a, p = 5
if and only if C = 19, x = 22, 434 or C = 341 and x = 2, 759, 646 and p = 7
does not occur.
P r o o f. If p = 3 then ±1 = 3a2 − C, yielding C = 3a2 ∓ 1, y = 4a2 ∓ 1
and then x = 8a3 ∓ 3a.
The result for p = 5 is due to Wren [37], but a very simple proof is
available. From (1) we obtain (C − 5a2 )2 = 20a4 ± 1. The lower sign is
impossible, and it is shown in [13] that the upper sign occurs only for a = 6,
yielding the result.
The result for p = 7 is due to Blass and Steiner [3].
Although the results proved to date are useful, only occasionally, e.g. if
C = 6, do they suffice to dismiss both cases of (1). We now consider how
we might treat other cases. As will be seen, a fair amount of computation
is often involved, and there seems no a priori guarantee of success in any
particular example. Nevertheless, the technique does seem to work with
sufficient perseverence.
Define for integers a and m ≥ 0 the integer function
√ √
(a + −C)m − (a − −C)m
(2) fm (a) = √ .
2 −C
Then (1) takes the form fp (a) = ±1. We consider one odd prime q not
dividing C at a time, and shall attempt to locate any possible solutions or
to prove that there are none, by showing (2) to be impossible modulo q for
some values of p not already excluded by one of the lemmas above. We first
observe that as a function of m, the sequence {fm (a)} is periodic modulo q,
the period being a factor of Q where Q = q − 1 or q 2 − 1 according as
376 J. H. E. Cohn

(−C | q) = 1 or −1. For,

√ √ √
(a + −C)q ≡ aq + (−C)(q−1)/2 −C ≡ a + (−C | q) −C (mod q) ,
and similarly for the complex conjugate; if (−C | q) = −1 then repeating
√ 2 √ √
the process gives (a + −C)q ≡ (a − −C)q ≡ (a + −C) (mod q), and
so fm+Q (a) ≡ fm (a) (mod q) in either case.
From (2) we see that f0 (a) = 0, f1 (a) = 1 and for m ≥ 0,
(3) fm+2 (a) = 2afm+1 (a) − (a2 + C)fm (a)
and so as a function of a, fm (a) is a polynomial and contains only even
powers of a if m is odd. Thus to solve fp (a) ≡ ±1 (mod q) it suffices to
consider only p ≡ m (mod Q) with 1 ≤ m ≤ Q−1 and values of a satisfying
0 ≤ a < q/2. Using Lemma 3, the value a = 0 can be excluded, provided
we check that q 2 - (C q−1 − 1). It is a routine computation to check that
this does not occur for any of the primes q we wish to use. Of course, this
condition is occasionally violated, and in such cases we have to select other
primes q.
The procedure therefore is as follows. For a given C, we select some odd
primes q - C such that Lemma 3 ensures that q - a. For each one of these, we
calculate the residue of fm (a) modulo q using (3) for each a in the range
1 ≤ a < q/2 and each odd m in the range 1 ≤ m ≤ Q − 1. In this way, we
compile a list of all those m in the range for which the congruence can hold
for any a, and so prove a result which forces p to belong to one of a set of
residue classes modulo Q. From this list we may delete any possible residue
which would prevent p being a prime; thus the value 15 can be removed
from the list if Q = 66. All this is easily automated. The object of the
exercise is to find sufficiently many such congruence conditions, together
with any information available from Lemmas 4 or 5 to complete a proof. It
may appear surprising that this should ever work, but it often does.
In the case of the plus sign in (1), it might be objected that the above
method can never succeed, for since f1 (a) = 1, it follows that the possibility
p ≡ 1 (mod Q) can never be excluded by the reasoning of the previous
paragraph, and accordingly, unless p ≡ 1 (mod 4) be excluded by Lemma 5,
no matter for how many different primes q the method be repeated, all that
will be achieved will be to show that p − 1 must have a very large number of
different factors. However, if C has any prime factor P ≥ 5, then choosing
some suitable q with P dividing Q, there remains the prospect of proving
that p ≡ 1 (mod P ), and this is excluded by Lemma 6. This can sometimes
be done directly; sometimes, for example if C = 21, the plus sign can be
eliminated by proving that p − 1 is divisible by P − 1, where P is a prime
exceeding 3 which divides C and the conclusion then follows by Lemma 6;
occasionally, for example if C = 17, a combination of these two ideas is
 The diophantine equation x2 + C = y n 377

required. A few examples may make all this much clearer. The plus sign in
cases in which C has only prime factors 2 or 3 does not arise in view of
Lemma 5.

4. Some examples
Result 1. There are no solutions when C = 6.
For, in this case there are no solutions with n even, since 6 is not the
difference of two squares.
√ So n = p, and (x, 6) = 1. The principal ideals
generated by ±x + −6 are coprime, and since the class number of the field
is 2, it follows that we need only consider (A). But now by Lemmas 4 and 5,
both signs are impossible.
Result 2. There are no solutions when C = 5.
Again, we see without difficulty that we cannot have n even, and as
again h = 2, we obtain only (A). This time the minus sign in (1) is again
excluded by Lemma 4, but with the plus sign, the only information provided
by Lemma 5 is p ≡ 1 (mod 4). Then by Lemma 6, (p | 5) = 1, p 6≡ 1 (mod 5)
gives p ≡ 4 (mod 5) and combining these yields p ≡ 9 (mod 20).
Now consider q = 61 and 601 in turn, for each of which it is easily
seen that 5q−1 6≡ 1 (mod q 2 ), for 560 ≡ 1 + 38 · 61 (mod 612 ) and 5600 ≡
1 + 405 · 601 (mod 6012 ). Since (−5 | 61) = 1 we obtain Q = 60 and find
only p ≡ 49 (mod 60), and in particular p ≡ 1 (mod 3). Then with q = 601,
we obtain Q = 600 and we find that modulo Q all the possible residues have
p ≡ 2 (mod 3) which is impossible.
Result 3. When C = 11, the only solutions are x = 4 or 58.
Again, n cannot be even. Here h = 1, and we obtain case (A) or case (B),
the latter of which gives just x = 4, by Lemma 1. By Lemma 4, the minus
sign does not occur, but Lemma 5 gives no information about (1) with the
plus sign. By Lemma 7, when p = 3 we get just one more solution x = 58,
and so we may assume that p ≥ 5. By Lemma 3, we then find that none
of the primes q = 23, 67, 89 or 397 divides a, since a calculation reveals
the residues of 11q−1 modulo q 2 to be respectively 1 + 7q, 1 + 43q, 1 + 72q
and 1 + 82q. We then apply the procedure outlined above, and obtain from
q = 23, p ≡ 3 or 15 (mod 22). Using this with results from q = 67 then gives
p ≡ 25 or 37 (mod 66), since we can now reject p ≡ 3 (mod 66). Similarly,
from q = 89 we obtain p ≡ 3, 15, 47 or 59 (mod 88), and so certainly p ≡ 3
(mod 4). But the result from q = 397 taken together with the one modulo 66
then gives p ≡ 37 or 289 (mod 396) each of which implies p ≡ 1 (mod 4),
and so there is no further solution.
It may be observed that this result generalizes [8] which proves that
x = 4 is the sole solution of x2 + 11 = 3n .
378 J. H. E. Cohn

Result 4. There are no solutions when C = 21.

Here if n were even, y n/2 = (21 + 1)/2 or (7 + 3)/2 neither of which

gives a solution. Thus assume n = p. Since h = 4, we need only consider
case (A). Then the minus sign is impossible by Lemma 4, and for the plus
sign Lemma 5 yields both p ≡ 1 (mod 4) and p ≡ 1 (mod 3). Thus p ≡ 1
(mod 6), and this is impossible by Lemma 6 with P = 7.

Result 5. When C = 17, the only solution is x = 8.

Here we find for n even y n/2 = 9, yielding the solution x = 8. For n

odd we need only consider (A) since h = 4, and exclude the minus sign
by Lemma 4. By Lemma 5, the plus sign would imply p ≡ 1 (mod 4), and
then by Lemma 6, p must be a quartic residue, but not ≡ 1 (mod 17). Thus
p ≡ −1, 4, or −4 (mod 17).

Now with q = 7, 137 and 409 we obtain respectively Q = 6, 136 and

408 and in each case q - a by Lemma 3. Then we obtain p ≡ 1 (mod 6)
from q = 7, and p ≡ 101 (mod 136) using p ≡ −1, 4 or −4 (mod 17) from
q = 137. Finally, using q = 409 and all the above congruences, we find no
possible residue modulo 408.

Result 6. When C = 40, the only solution is x = 52.

Here we find no solution√for n even. Suppose

√ first that x is odd. Then
since h = 2, we have ±x + 2 −10 = (a + b −10)p , as in (D), where a must
be odd. But now
(p−1)/2  
X p
2=b ap−2r−1 (−10b2 )r
r=0
2r + 1

and since the second factor is odd, b = ±2. Thus we arrive at (1), and now
both sings can be eliminated by Lemmas 4 and 5. It follows therefore that
x must be even. Then 2 k x is impossible since it would imply 22 k y p and if
4 | x then 23 k y p , whence p = 3. We then obtain only x = 52 as in [33].

5. Statement of results. We have considered in detail all values of

C ≤ 100, and have completed the solution for 77 of these values, quoting
the known results for p = 3 [7] in a few cases, as follows:

There are no solutions at all for 46 values of C, viz.: 1, 3, 5, 6, 8, 9, 10,

14, 21, 22, 24, 27, 29, 30, 33, 34, 36, 37, 38, 41, 42, 43, 46, 50, 51, 52, 57, 58,
59, 62, 66, 68, 69, 70, 73, 75, 78, 82, 84, 85, 88, 90, 91, 93, 94, and 98.
 The diophantine equation x2 + C = y n 379

For the following 31 values of C, the only solutions for x are:

C x C x C x C x
2 5 20 14 53 26, 156 77 2
4 2, 11 26 1, 207 54 17 80 1
11 4, 58 32 7, 88 56 5, 76 81 46
12 2 35 36 61 8 83 140
13 70 40 52 64 8 89 6
16 4 44 9 65 4 96 23
17 8 48 4, 148 67 110 97 48
19 18, 22434 49 24, 524 76 7, 1015

Two values C = 74 and 86 for which the class number is divisible by 5

may have other solutions with p = 5 but none with p 6= 5, apart from the
known solutions x = 13, 985 when C = 74.
The remaining √ 21 cases seem to be very difficult indeed because the
ideals π = [x + −C] and π 0 can have common factors; these values are the
twelve for which C ≡ 7 (mod 8) and nine others, viz., 18, 25, 28, 45, 60,
72, 92, 99 and 100. In some cases there are partial results which avoid the
difficulties, e.g. that there are no solutions with x even if C = 7, nor with x
odd if C = 28.

As mentioned in Section 1, by no means all of the above results are

new. We are aware of previous results, explicit or implicit, for each of the
following 33 values of C for which a complete solution, in some cases in
conjunction with [7], has been obtained previously, although in the case of
C = 40, the solution x = 52 is inadvertently omitted. We hope that the list
is complete and that in each case the reference is to the first such solution:

1 [14] 14 [16] 56 [15]

2 [20] 19 [11] 57 [21]
3 [29] 20 [1] 58 [16]
4 [32] 21 [22] 73 [21]
5 [29] 32 [12] 76 [1]
6 [16] 33 [21] 78 [16]
8 [33] 34 [16] 82 [16]
9 [19] 40 [15] 88 [15]
10 [16] 41 [21] 89 [21]
12 [26] 42 [16] 94 [16]
13 [22] 44 [1] 97 [21]
380 J. H. E. Cohn

6. Conclusion. As indicated, there are a number of open questions,

principally that of dealing with the difficult cases, exemplified by C = 7.
The author would like to throw out a challenge to prove the
Conjecture. The equation x2 + 7 = y n has only the solutions given by
x = 1, 3, 5, 11 and 181.
or, if it be false, to solve it completely. What is known about this equation
is that there are no other solutions if y is odd [25], nor if n is even nor if
3 | n [24] nor if y is a power of 2 [30].
Secondly, it would be of interest to complete the solution for the values 74
and 86. These may be considerably less difficult questions.
Finally, Lemma 7 leads to a rather peculiar result, capable of various
extensions, but which we state in its simplest form as
Theorem 2. For any positive integers A, B with A odd , with C = A7 −B 2
positive
√ and square-free, the class number of the imaginary quadratic field
Q[ −C] is divisible by 7.
A similar result applies with 7 replaced by 5 throughout with two excep-
tions C = 19 and 341, and with 7 replaced by 3 if we exclude the cases in
which C ± 1 or C ± 8 is of the form 3a2 .

References

[1] A. A i g n e r, Die diophantische Gleichung x2 + 4D = y p im Zusammenhang mit

Klassenzahlen, Monatsh. Math. 72 (1968), 1–5.
[2] J. B l a s s, A note on diophantine equation Y 2 + k = X 5 , Math. Comp. 30 (1976),
638–640.
[3] J. B l a s s and R. S t e i n e r, On the equation y 2 + k = x7 , Utilitas Math. 13 (1978),
293–297.
[4] E. B r o w n, Diophantine equations of the form x2 + D = y n , J. Reine Angew. Math.
274/275 (1975), 385–389.
[5] —, Diophantine equations of the form ax2 + Db2 = y p , ibid. 291 (1977), 118–127.
[6] K. C h a o, On the diophantine equation x2 = y n + 1, xy 6= 0, Sci. Sinica (Notes) 14
(1964), 457–460.
[7] F. B. C o g h l a n and N. M. S t e p h e n s, The diophantine equation x3 − y 2 = k, in:
Computers in Number Theory, Academic Press, London, 1971, 199–205.
[8] E. L. C o h e n, Sur l’équation diophantienne x2 + 11 = 3k , C. R. Acad. Sci. Paris
Sér. A 275 (1972), 5–7.
[9] —, On the Ramanujan–Nagell equation and its generalizations, in: Proc. First Con-
ference of the Canadian Number Theory Association, Banff, Alberta, 1988, de
Gruyter, 1990, 81–92.
[10] J. H. E. C o h n, The Diophantine equation x2 +3 = y n , Glasgow Math. J. 35 (1993),
203–206.
[11] —, The diophantine equation x2 + 19 = y n , Acta Arith. 61 (1992), 193–197.
[12] —, The diophantine equation x2 +2k = y n , Arch. Math. (Basel) 59 (1992), 341–344.
 The diophantine equation x2 + C = y n 381

[13] J. H. E. C o h n, Lucas and Fibonacci numbers and some Diophantine equations,

Proc. Glasgow Math. Assoc. 7 (1965), 24–28.
[14] L. E u l e r, Algebra, Vol. 2.
[15] O. K o r h o n e n, On the Diophantine equation Ax2 + 8B = y n , Acta Univ. Oulu.
Ser. A Sci. Rerum Natur. Math. 16 (1979).
[16] —, On the Diophantine equation Ax2 + 2B = y n , ibid. 17 (1979).
[17] —, On the Diophantine equation Cx2 + D = y n , ibid. 25 (1981).
[18] V. A. L e b e s g u e, Sur l’impossibilité en nombres entiers de l’équation xm = y 2 + 1,
Nouvelles Annales des Mathématiques (1) 9 (1850), 178–181.
[19] W. L j u n g g r e n, On the diophantine equation x2 + p2 = y n , Norske Vid. Selsk.
Forh. Trondheim 16 (8) (1943), 27–30.
[20] —, Über einige Arcustangensgleichungen die auf interessante unbestimmte Glei-
chungen führen, Ark. Mat. Astr. Fys. 29A (1943), no. 13.
[21] —, On the diophantine equation x2 + D = y n , Norske Vid. Selsk. Forh. Trondheim
17 (23) (1944), 93–96.
[22] —, On a diophantine equation, ibid. 18 (32) (1945), 125–128.
[23] —, New theorems concerning the diophantine equation Cx2 + D = y n , ibid. 29 (1)
(1956), 1–4.
[24] —, On the diophantine equation y 2 − k = x3 , Acta Arith. 8 (1963), 451–463.
[25] —, On the diophantine equation Cx2 +D = y n , Pacific J. Math. 14 (1964), 585–596.
[26] —, On the diophantine equation Cx2 + D = 2y n , Math. Scand. 18 (1966), 69–86.
[27] —, New theorems concerning the diophantine equation x2 + D = 4y q , Acta Arith.
21 (1972), 183–191.
[28] L. J. M o r d e l l, Diophantine Equations, Academic Press, London, 1969.
[29] T. N a g e l l, Sur l’impossibilité de quelques équations à deux indéterminées, Norsk.
Mat. Forensings Skrifter No. 13 (1923), 65–82.
[30] —, Løsning til oppgave nr 2 , 1943 , s. 29 , Norske Mat. Tidsskrift 30 (1948), 62–64.
[31] —, Verallgemeinerung eines Fermatschen Satzes, Arch. Math. (Basel) 5 (1954),
153–159.
[32] —, Contributions to the theory of a category of diophantine equations of the second
degree with two unknowns, Nova Acta Regiae Soc. Sci. Upsaliensis (4) 16 (2) (1955).
[33] —, On the Diophantine equation x2 + 8D = y n , Ark. Mat. 3 (1954), 103–112.
[34] S. R a m a n u j a n, Question 464 , J. Indian Math. Soc. 5 (1913), 120.
[35] T. N. S h o r e y, A. J. v a n d e r P o o r t e n, R. T i j d e m a n and A. S c h i n z e l, Appli-
cations of the Gel’fond–Baker method to diophantine equations, in: Transcendence
Theory: Advances and Applications, Academic Press, London, 1977, 59–77.
[36] T. N. S h o r e y and R. T i j d e m a n, Exponential Diophantine Equations, Cambridge
University Press, Cambridge, 1986.
[37] B. M. E. W r e n, y 2 + D = x5 , Eureka 36 (1973), 37–38.

DEPARTMENT OF MATHEMATICS
ROYAL HOLLOWAY AND BEDFORD NEW COLLEGE
EGHAM, SURREY TW20 OEX, ENGLAND
E-mail: UHAH206@UK.AC.RHBNC

Received on 27.11.1992
and in revised form on 25.6.1993 (2347)