In a 1995 paper, Choudhry gave a table of solutions to the quartic Diophantine equation,

$a^4+nb^4 = c^4+nd^4\tag{1}$

for $n\leq101$. Seiji Tomita recently extended this to $n<1000$ and solved all $n$ except $n=967$ (which was later found by Andrew Bremner).

It can be shown there is an infinite number of $n$ such that $(1)$ is solvable, such as $n=v^2-3$. In general, given the Chebyshev polynomials of the first kind $T_m(x)$

$$\begin{aligned} T_1(x) &= x\\ T_2(x) &= 2x^2-1\\ T_3(x) &= 4x^3-3x\\ \end{aligned}$$

etc, and the second kind $U_m(x)$

$$\begin{aligned} U_1(x) &= 2x\\ U_2(x) &= 4x^2-1\\ U_3(x) &= 8x^3-4x\\ \end{aligned}$$

etc, define, $$n = \frac{T_m(x)}{x}$$ $$y = \frac{U_{m-2}(x)}{x}$$

then one can observe that,

$(n + n x + y)^4 + n(1 - n x - x y)^4 = (n - n x + y)^4 + n(1 + n x + x y)^4\tag{2}$


  1. Is $(1)$ solvable in the integers for all positive integer $n$ where $a^4\neq c^4$? (This has also been asked in this MSE post.)
  2. How do we show that $(2)$ is indeed true for all positive integer $m$?

Question 2: The reason must be that Chebyshev polynomials solve Fermat-Pell equations. The difference between the two sides of equation (2), $$ (n+nx+y)^4 + n(1−nx−xy)^4 = (n−nx+y)^4 + n(1+nx+xy)^4, $$ factors as $8nx(n+y) \phantom. ((x^2-1)y^2 + 2n(x^2-1)y + 1 - n^2)$. The last factor is a quadratic in $n$ with leading coefficient $1$, so has solutions iff the discriminant is a square. This yields the equation $(x^2-1) u^2 + 1 = t^2$ where $u=xy$, which is a Fermat-Pell equation in $u$ and $t$ with fundamental solution $(u,t)=(1,x)$, etc.


I am pretty sure (2) is easy to prove by identities for Chebyshev polynomials and I think sage proved it.

Let $U_{n-1},U_{n-2},T_{n-1},T_{n-2}$ be variables, use the recurrence for $T_n$ to compute the identity for $m$ and reduce it modulo the identities $ T_n(x)^2 - (x^2-1) U_{n-1}(x)^2 = 1 $ and $ U_n(x) = xU_{n-1}(x) + T_n(x) $

Here is sage code that appears to prove it (I suppose I don't even need induction).

def titoide2():
    def titochebide1(n,y,x):
        return (n+n*x+y)**4+n*(1-n*x-x*y)**4-( (n-n*x+y)**4+n*(1+n*x+x*y)**4 )
    def smalchste(a,b,x):
        recurrence for U_n and T_n
        return 2*x*a-b

    # u1=U_{n-1},u2=U_{n-2}, t1=T_{n-1},t2=T_{n-2}

    t3=smalchste(t1,t2,x) #T_n
    t4=smalchste(t3,t1,x) #T_{n+1}
    u3=smalchste(u1,u2,x) #U_n

    r1=titochebide1(t3/x,u2/x,x) #hypothesis for m, currently not used

    ide=[t1^2-(x^2-1)*u2^2-1,t3^2-(x^2-1)*u1^2-1,u1-(u2*x+t1),u3-(u1*x+t3) ,t4^2-(x^2-1)*u3^2-1] 
    ##ide=[numerator(r1),t1^2-(x^2-1)*u2^2-1,t3^2-(x^2-1)*u1^2-1,u1-(u2*x+t1),u3-(u1*x+t3) ,t4^2-(x^2-1)*u3^2-1] #hypothesis doesn't appear to be needed

    # T_n(x)^2 - (x^2-1) U_{n-1}(x)^2 = 1 and U_n(x) = xU_{n-1}(x) + T_n(x)  from https://en.wikipedia.org/wiki/Chebyshev_polynomials

    ide=[numerator(_) for _ in ide]
    r2=titochebide1(t4/x,u1/x,x) # for m+1
    r2q=prq(numerator(r2)) # reduce it modulo identities
    print 'r2q=',r2q
    if r2q==0:
        print ' identity holds'
        print ' couldn\'t prove it'

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.