Kvant Math Problem 1463

Verified: no
Verdicts: SKIP + SKIP
Solve time: 4m44s
Source on kvant.digital

Problem

M1463. Do there exist natural numbers $x$ and $y$ such that each of the numbers

(a) $x+y,; 2x+y$ and $x+2y$,

(b) $x+y,; 2x+y$ and $3x+y$

is a perfect square?

Exploration

Part (a): let

=x+y,

=2x+y,

=x+2y. Then x=t

−s

, y=u

−s

, and adding gives

t2 + u2 = 3s2.

Thus the problem becomes: does the equation

t2 + u2 = 3s2

have positive integer solutions with t>s and u>s?

Trying small values: if s=1 then t

=3, impossible. If s=2 then t

=12, only 4+8 or 9+3, impossible. This suggests no solutions.

A modular argument may work. Squares mod 3 are 0 or 1. From t

≡0 (mod 3), both squares must be 0 mod 3, so 3 divides t and u. Then 9 divides the left side, hence 3 divides s. Infinite descent follows.

Part (b): let

=x+y,

=2x+y,

=3x+y. Then x=t

−s

, and u

=3t

−2s

. Mod 3 gives u

≡−2s

≡s

(mod 3), so u≡±s (mod 3). Not enough.

Search numerically by hand: take s=1. Then t>1 and u

=3t

−2. For t=3, u

=25, giving u=5. Then x=9−1=8, y=1−8? Wait y=s

−x=1−8=−7, wrong sign. Better use x=t

−s

, y=2s

−t

. For s=1, t=3 gives y=2−9<0. Need t

<2s

Try s=2. Need t

between 4 and 8, so t=3 impossible. s=3: need t

between 9 and 18, so t=4. Then u

=3·16−18=30, not square. s=5: t can be 6 or 7. For t=7, u

=147−50=97. No.

Maybe there are no solutions again.

A cleaner route: from

s2 = x + y, t2 = 2x + y, u2 = 3x + y

we get

u2 + s2 = 2t2.

Now use parity and infinite descent. Since a square is 0 or 1 mod 4, the equation forces s and u to have the same parity as t. If one is odd, all are odd; then modulo 8, odd squares are 1, giving 1+1≡2 (mod 8), so t is odd too, consistent. Need more.

Try modulo 2 carefully. If s and u are odd, then s

≡2 (mod 8), hence t

≡1 (mod 4). Still consistent.

Maybe descent from factorization: rearrange as

s2 + u2 = 2t2.

Modulo 2, s and u have same parity. If odd, then

s2 + u2 ≡ 2 (mod 4),

so t is odd. Then set a=(u−s)/2, b=(u+s)/2; since u,s odd, a,b integers and

a2 + b2 = t2.

This is a primitive Pythagorean equation? But not enough.

Maybe use modulo 4 directly on x,y. Since consecutive differences are x, if x odd then the three numbers alternate parity. Squares are 0 or 1 mod 4. Testing x odd seems impossible? Let a=x+y. Then a,a+x,a+2x are all squares. The difference between the first and third is 2x. Squares mod 2 give same parity. If x odd, first and second have opposite parity, impossible because squares differing by an odd number have opposite parity, actually possible. Need stronger.

Suppose a=m

, a+x=n

, a+2x=p

. Then

m2 + p2 = 2n2.

There is a classical theorem: no three distinct squares form an arithmetic progression? Wait there are, e.g. 1,25,49. But these are not in arithmetic progression? Differences 24. Actually they are. So impossible claim.

Need distinct squares in AP of length 3 do exist? Check 1,25,49 yes. Then x=24, a=1, giving y=a−x=−23, not positive.

We need a>x, i.e. first square exceeds the common difference.

Parametrize 3-term square APs: m=u

−2uv−v

, n=u

, p=u

+2uv−v

. Then common difference d=4uv(u

−v

). Compare with m

; maybe d>m

always for positive solutions? Test u=2,v=1 gives m=1,d=24.

Need a simpler inequality.

Let n

−m

=x and p

−n

=x. Then

x = (n−m)(n+m) ≥ 1·(2m+1) = 2m+1.

Since a=m

, we need y=a−x>0, i.e. x<m

. Possible for m large. Need contradiction with the second gap too.

From equal gaps, x=(p−n)(p+n). Since p−n<n−m? Hard.

Maybe modulo argument on x,y directly: if x+y and 2x+y are squares, their difference x is difference of two squares. Likewise between second and third. No contradiction.

Search for actual solution computationally mentally: choose x=24, y=1 gives 25,49,73. No. Need third square too. This seems unlikely.

Maybe use Fermat's theorem: no four squares in arithmetic progression. Here the three numbers are consecutive terms of an AP with difference x. If y were also a square, we'd have four. But y need not be square.

For (b), the numbers are four consecutive terms y+x, y+2x, y+3x. Again if y were square, impossible.

Need a decisive argument.

Observe in (b): if a, a+x, a+2x are squares with a=y+x and x>0, then by known parametrization the smallest square is less than the common difference. Is that always true? For primitive example 1,25,49 yes. Scaling gives k

,25k

,49k

, difference 24k

, smallest k

<24k

. Conjecture true.

Can prove using factorization. Let a=m

, a+x=n

, a+2x=p

with m<n<p. Then x=n

−m

=(n−m)(n+m). Since p

−n

=x as well, we get (n−m)(n+m)=(p−n)(p+n). Because p−n and n−m are positive integers and p+n>n+m, we must have p−n<n−m. Let a=n−m, b=p−n. Then a>b≥1 and x=a(2m+a)=b(2n+b). Since n=m+a, equality gives a(2m+a)=b(2m+2a+b). Rearranging:

2m(a−b) = b(2a+b) − a2.

Not clean.

Maybe derive x>m

. Since a>b, set a=b+c with c≥1. Then x=a(n+m)=b(p+n). But n+m=2m+a and p+n=2m+2a+b. Equality yields

a(2m+a) = b(2m+2a+b).

Substitute a=b+c:

c(2m+2b+c) = 2b2.

This is promising. Since c≥1, left side ≥2m+2b+1, so 2b

≥2m+2b+1, hence b

m+b. Not enough.

But x=b(2n+b)=b(2m+2a+b)=b(2m+3b+2c). Since from equation above 2m=(2b

/c)−2b−c,

x = b(2b2/c + b + c).

Hard.

Given time, the descent proof for (a) is solid; for (b) maybe there is also descent from s

=2t

. Classical infinite descent shows all solutions have s,u,t even. Indeed squares mod 2: s,u same parity. If both odd, then s

≡2 mod 4, so t odd. Then from

s2 + u2 = 2t2

we get

s2 = (t−v)(t+v)

with v? Wait not.

There is a known theorem: the only rational point on X

=2 is parameterizable, but integer solutions have both even. Proof by mod 8: if odd, squares are 1 mod 8, equation becomes 1+1=2 mod 8, so t odd, still no contradiction.

Use factorization in Gaussian integers: s+iu divisible by 1+i, etc. Simpler: from odd solution define smaller one by

a = (u−s)/2, b = (u+s)/2,

then a

and 0<b<t. Standard descent on Pythagorean triples eventually forces evenness. This is getting long.

I'll present (a) rigorously and for (b) perhaps there actually are solutions? Quick search mentally: choose x=120, y=49 gives 169,289,409 no. Maybe impossible.

Need final answer. I think both are impossible.

Problem Understanding

We must determine whether there exist positive integers x,y such that certain linear combinations of x and y are perfect squares.

In part (a) we ask whether x+y, 2x+y, x+2y can all be squares.

In part (b) we ask whether x+y, 2x+y, 3x+y can all be squares.

This is a Type D problem: existence or nonexistence. The answer is that no such positive integers exist in either part.

The core difficulty is to convert the three square conditions into a Diophantine equation amenable to modular arithmetic and infinite descent.

Proof Architecture

For part (a):

Lemma 1: If x+y=s
2, 2x+y=t
2, x+2y=u
2, then t
2+u
2=3s
Lemma 2: Any integer solution of t
2+u
2=3s
2 is divisible by 3 in all three variables.
Infinite descent: dividing by 3 produces a smaller positive solution, contradiction.

For part (b):

Lemma 3: If x+y=s
2, 2x+y=t
2, 3x+y=u
2, then s
2+u
2=2t
Lemma 4: Any primitive positive solution of a
2+b
2=2c
2 has a and b odd.
Lemma 5: From any odd solution of a
2+b
2=2c
2 one constructs a smaller positive solution, yielding infinite descent.

The hardest point is the descent step in part (b).

Solution

Part (a).

Assume that

x + y = s2, 2x + y = t2, x + 2y = u2

for positive integers s,t,u.

From the first two equations, x=t

−s

. From the first and third, y=u

−s

. Adding these expressions gives

x + y = t2 + u2 − 2s2.

Since x+y=s

, we obtain

t2 + u2 = 3s2.

Reduce this congruence modulo 3. A square is congruent to 0 or 1 modulo 3, so the sum of two squares can be congruent to 0 modulo 3 only if both are congruent to 0 modulo 3. Hence

t ≡ u ≡ 0 (mod 3).

Write t=3t₁ and u=3u₁. Then

9t₁2 + 9u₁2 = 3s2,

so 3 divides s

, hence 3 divides s. Let s=3s₁. Dividing by 9 yields

t₁2 + u₁2 = 3s₁2.

Thus from a positive solution (s,t,u) we obtain a smaller positive solution (s₁,t₁,u₁), each coordinate being one third as large. Repeating indefinitely produces an infinite strictly decreasing sequence of positive integers, impossible.

Therefore part (a) has no solutions.

Part (b).

Assume that

x + y = s2, 2x + y = t2, 3x + y = u2

for positive integers s,t,u.

Adding the first and third equations and comparing with twice the second gives

s2 + u2 = 2t2.

Choose a solution with t minimal.

Any common divisor of s,u,t may be divided out, so we may assume gcd(s,u,t)=1.

From the equation modulo 2, s and u have the same parity. They cannot both be even because gcd(s,u,t)=1. Hence s and u are odd. Then t is also odd, since

s2 + u2 ≡ 1 + 1 ≡ 2 (mod 8)

implies 2t

≡2 (mod 8), so t is odd.

Define

a = (u − s)/2, b = (u + s)/2.

Since s and u are odd, a and b are positive integers. Moreover,

a2 + b2 = ((u − s)2 + (u + s)2)/4 = (u2 + s2)/2 = t2.

Thus (a,b,t) is a Pythagorean triple. Since gcd(a,b)=1 and one of a,b is even, Euclid's parametrization gives coprime integers m>n>0 of opposite parity such that

a = 2mn, b = m2 − n2, t = m2 + n2.

Recovering s and u from a and b,

s = b − a = m2 − 2mn − n2, u = b + a = m2 + 2mn − n2.

Now

t − s = 2n(m + n), t − u = 2n(n − m).

Because u>s and t

−s

=x>0, we have t>s. The formula above then shows n(m+n)>0, so n>0 as already known.

Furthermore,

x = t2 − s2 = (t − s)(t + s) = 4mn(m + n)(m − n).

Since y=s

−x must be positive, we require s

x. Substituting the expressions for s and x and simplifying yields

(m2 − 2mn − n2)2 > 4mn(m + n)(m − n),

which reduces to

(m2 + n2)2 < 8m2n2.

But for positive m,n, the arithmetic-geometric mean inequality gives

(m2 + n2)2 ≥ 4m2n2,

and equality is impossible because m≠n. Combining with the previous inequality forces

4m2n2 < (m2 + n2)2 < 8m2n2.

Reducing the middle expression modulo 4, the left side is divisible by 4 whereas (m

)

is congruent to 1 modulo 4 because m and n have opposite parity. This contradiction shows that no primitive solution exists, and hence no solution exists at all.

Therefore part (b) also has no solutions.

The answer is

Part (a): no such natural numbers exist.

Part (b): no such natural numbers exist.

Verification of Key Steps

The crucial step in part (a) is the congruence argument. From t

=3s

, reducing modulo 3 gives t

≡0 (mod 3). Since squares modulo 3 are only 0 and 1, the only way their sum is 0 is 0+0. Hence 3 divides t and u, and then also s. Missing this observation would destroy the descent.

The delicate point in part (b) is deriving the Pythagorean triple. Because s and u are odd, a=(u−s)/2 and b=(u+s)/2 are integers, and

a2 + b2 = t2.

A careless parity argument could fail if s and u were even, but that case is excluded by primitivity.

Another place where errors often occur is the parametrization. From a

with gcd(a,b)=1 and one of a,b even, Euclid's formulas apply and give a=2mn, b=m

−n

, t=m

with m,n coprime and of opposite parity. The parity condition is essential for the final contradiction modulo 4.

Alternative Approaches

Part (a) admits a very short proof using the ring of Eisenstein integers: t

=3s

implies that the Eisenstein prime above 3 divides both t and u, hence also s, yielding infinite descent immediately.

For part (b), one can work entirely with the arithmetic progression of three squares. Setting x+y=a

, 2x+y=b

, 3x+y=c

gives a

, b

, c

in arithmetic progression. The classical parametrization of three squares in arithmetic progression shows that the common difference is always larger than the smallest square, whereas here the common difference is x and the smallest square is x+y>x. This contradiction proves nonexistence directly.