6

Consider the sums of cubes of three distinct positive integers in arithmetic progression, i.e. $a^3 + (a+b)^3 + (a+2b)^3$ where $a$ and $b$ are positive integers.
$$5643 = 1^3 + 9^3 + 17^3 = 6^3 + 11^3 + 16^3$$ is the smallest positive integer that has two such representations. $$255816 = 8^3 + 34^3 + 60^3 = 18^3 + 38^3 + 58^3 = 43^3 + 44^3 + 45^3$$ is the smallest that has three such representations. Is there a positive integer with four or more such representations, and if so what is the smallest?

The sequence of positive integers with one or more representations is OEIS sequence A306213. Those with two or more is A359055.

If representations where the three cubes do not need to be positive are allowed, then I know of examples with up to $11$ representations. Thus $$ \eqalign{42878854656 &= (-17232)^3 + 24^3 + 17280^3\cr &= (-6900)^3 + 144^3 + 7188^3\cr &= (-6704)^3 + 152^3 + 7008^3\cr &= (-3132)^3 + 528^3 + 4188^3\cr &= (-1812)^3 + 912^3 + 3636^3\cr &= (-1032)^3 + 1224^3 + 3480^3\cr &= (-292)^3 + 1552^3 + 3396^3\cr &= (-24)^3 + 1672^3 + 3368^3\cr &= 1038^3 + 2112^3 + 3186^3\cr &= 1635^3 + 2304^3 + 2973^3\cr &= 1728^3 + 2328^3 + 2928^3\cr }$$

Robert Israel
  • 448,999
  • 1
    Remark: Since $(c-d)^3+c^3+(c+d)^3=3c(c^2+2d^2)$, this is the same as asking about the number of representations of $n/3$ by the reducible binary cubic form $c(c^2+d^2)$. Maybe this was addressed in work of Michael Bennett or Shabnam Akhtari? – Greg Martin Dec 15 '22 at 18:35
  • 4
    Many of the solutions share similar prime factors, if you check numbers of form $N=2^u\cdot3^v\cdot 19\cdot 67$ then you find examples with $4$ representations, e.g. $346380489216$ has representations $1188^3+3888^3+6588^3$, $1728^3+4104^3+6480^3$, $4248^3+4824^3+5400^3$ and $4665^3+4864^3+5063^3$. – Sil Dec 15 '22 at 19:06
  • 3
    Here is an example of five representations: $2^9\cdot 3^{12}\cdot 5^3 \cdot11^3\cdot19\cdot 67=x^3+y^3+z^3$ with $(x,y,z)$=$(65340,213840,362340)$, $(95040, 225720, 356400)$, $(142992, 243000, 343008)$, $(233640, 265320, 297000)$, $(256575, 267520, 278465)$. Found simply by starting with $19\cdot 67$ and multiplying by various cube powers (since that preserve original solution, but I don't have rigorous process). – Sil Dec 15 '22 at 22:02
  • @Sil, is there an explanation for the factor 19 more common appearance in the factorisation of the number $N$ being a sum of cubes? – user25406 Dec 16 '22 at 12:01
  • 1
    @user25406 I am not sure, maybe it has to do with fact that it is a prime of form $a^2+2b^2$, but there are many of these... – Sil Dec 17 '22 at 01:34
  • 1
    $1571596223448176114112000=x^3+y^3+z^3$ has six representations $(x,y,z)=(43040592, 73143000, 103245408), (19667340, 64365840, 109064340), (28607040, 67941720, 107276400), (57772440, 77421960, 97071480), (77229075, 80523520, 83817965), (70325640, 79861320, 89397000)$. – Tomita Dec 17 '22 at 12:57

2 Answers2

3

I think there is no point in trying to guess any number. First you need to write a parametrization of the solutions of the equation.

$$(A-B)^3+A^3+(A+B)^3=(C-Q)^3+C^3+(C+Q)^3$$

$$A=4(24a^2-8ab+b^2)$$

$$B=177a^2-59ab+6b^2$$

$$C=4(33a^2-10ab+b^2)$$

$$Q=132a^2-49ab+6b^2$$

Then you can substitute it into the formula and then play with the coefficients. If this fails, then you will have to solve a system of nonlinear equations. There will be very cumbersome calculations.

There was a question. When there are solutions $ A−B>0 $ and $ C−Q>0 $

Each attempt at a solution gives a new formula for some reason. It didn't fit.

$$A=2(36a^2-4ab+b^2)$$

$$B=64a^2-ab+3b^2$$

$$C=2(32a^2+b^2)$$

$$Q=74a^2-11ab+3b^2$$

The following formula gives the necessary solutions.

$$A=528a^2-40ab+b^2$$

$$B=64a^2-25ab+3b^2$$

$$C=128a^2+b^2$$

$$Q=764a^2-95ab+3b^2$$

The general formula should contain 3 more parameters.

At the moment, we have managed to obtain such a formula.

$$A=64k^{4}t^{3}x^{2}+2ty^2$$

$$B=2k(27k^6-18k^{4}t^2+20k^2t^4+8t^6)x^2+(9k^2+2t^2)(3k^2-2t^2)xy+3ky^2$$

$$C=8tk^2(9k^4-4k^2t^2+4t^4)x^2+8kt(3k^2-2t^2)yx+2ty^2$$

$$Q=64k^3t^4x^2+(3k^2-2t^2)^2xy+3ky^2$$

individ
  • 4,301
  • This is a necessary condition (even if not explicit) but not at all sufficient. – Piquito Dec 17 '22 at 18:03
  • This particular parametrization seems problematic, no choice of $a,b$ is possible to have $A-B>0$ and $C-Q>0$. – Sil Dec 18 '22 at 10:58
  • I have given as an example the simplest possible formula - which describes some special cases. Full parameterization can look extremely cumbersome. For example so. https://math.stackexchange.com/questions/4591666/solving-cubic-systems-of-diophantine-equations/4595633#4595633 When I write a cumbersome formula, they constantly draw disadvantages for me. – individ Dec 18 '22 at 11:55
2

The smallest integer with four representations is \begin{align} 57426732000&=331^3+2000^3+3669^3\\ &=1530^3+2460^3+3390^3\\ &=2150^3+2620^3+3090^3\\ &=2265^3+2640^3+3015^3. \end{align}

The smallest integer with five representations is \begin{align} 57629053893312000&=65340^3+213840^3+362340^3\\ &=95040^3+225720^3+356400^3\\ &=142992^3+243000^3+343008^3\\ &=233640^3+265320^3+297000^3\\ &=256575^3+267520^3+278465^3 \end{align}

To prove the first one, we can just search the range $[1,57426732000]$ very quickly with a sieve. By using the fact that any solution $N$ must be of form $N=3c(c^2+2d^2)$ for integers $c>d>0$ (see Greg Martin's comment), the search range $A\leq N \leq B$ gives bounds $$ A\leq 3c(c^2+2d^2)\leq 3c(c^2+2(c-1)^2)=3c(3c^2-4c+2)\\ 3c(c^2+2)\leq 3c(c^2+2d^2)\leq B $$ In our case this implies $185696\leq c \leq 267818$ and $1\leq d\leq c-1$ (we can also stop iterating over $d$ when $3c(c^2+2d^2)> B$). For each of $c,d$ in this range we increment count of representations of $3c(c^2+2d^2)$.

Following C++ implementation searches the given range in few seconds:

#include <iostream>
#include <map>

using namespace std;

const long long A = 1; const long long B = 57426732000; const int repre = 4;

map<long long, char> counts;

int main() {

long long c = 1;
// 3c^3-4c^2+2c&gt;=A/3
while (3 * (3 * c * c * c - 4 * c * c + 2 * c) &lt; A) {
    c += 1;
}
long long min_c = c;

c = min_c;
// c^3+2*c&lt;=B/3
while (3 * (c * c * c + 2 * c) &lt;= B) {
    c += 1;
}
long long max_c = c - 1;

// search the range
c = min_c;
while (c &lt;= max_c) {
    long long d = 1;
    while ((d &lt;= c - 1) &amp;&amp; (3*c*c*c+6*c*d*d&lt;=B)) { 
        //c ^ 3 + 2 * c * d^2 &lt;= B / 3
        long long N = 3 * c * (c * c + 2 * d * d);
        counts[N] += 1;
        d += 1;
    }
    c += 1;
}

int idx = 1;
for (auto&amp; N : counts) {
    if (N.second &gt;= repre) {
        std::cout &lt;&lt; idx &lt;&lt; &quot; &quot; &lt;&lt; N.first &lt;&lt; std::endl;
        idx += 1;
    }
}

return 0;

}

Note: The above code can give list of the A359055 sequence by printing terms with at least two representations (I've tested this and for repre=2 and B=173704947768 we get first 10000 terms).

Increasing the bounds we find that first couple of smallest $N$ with four representations are therefore $$ 57426732000,346380489216,459413856000,1235190627000,1550521764000,\dots. $$

As for the five representations, the sieve above would require too much memory, but we can modify it into a segmented sieve where we search intervals of fixed length. The following code snippet verified the smallest integer mentioned above (it gets faster over time, but still took couple hours to finish on my PC):

#include <iostream>
#include <map>

using namespace std;

map<long long, char> counts;

long long isqrt(long long n) { if (n == 0) return 0;

long long y = n;
long long x = n + 1;

while (y &lt; x) {
    x = y;
    y = (y + n / y) &gt;&gt; 1;
}

return x;

}

int main() {

int search_repre = 1;
int search_up_to = 5;

const long long segment_size = 1000000000000;
const long long start = 1;

long long A = start;
long long B = start + segment_size - 1;
while (true) {

    cout &lt;&lt; &quot;Sieving [&quot; &lt;&lt; A &lt;&lt; &quot;,&quot; &lt;&lt; B &lt;&lt; &quot;]&quot; &lt;&lt; endl;
    counts.clear();

    long long c = 1;
    // 3c^3-4c^2+2c&gt;=A/3
    while (3 * (3 * c * c * c - 4 * c * c + 2 * c) &lt; A) {
        c += 1;
    }
    long long min_c = c;

    c = min_c;
    // c^3+2*c&lt;=B/3
    while (3 * (c * c * c + 2 * c) &lt;= B) {
        c += 1;
    }
    long long max_c = c - 1;

    // search the range
    for(long long c = min_c; c &lt;= max_c; ++c) {

        // c ^ 3 + 2 * c * d ^ 2 &gt;= A/3
        long long min_d = (isqrt(A/(6*c) - (c * c)/2));
        if (min_d &lt; 1)
            min_d = 1;

        // c ^ 3 + 2 * c * d ^ 2 &lt;= B/3
        // integer division floors down, so we use ceiling and 
        // ceil(n/m)=floor((n-1)/m)+1
        // to make sure we do not miss some solutions when upper bound is round down
        long long max_d = isqrt((B-3*c*c*c-1)/(6*c)+1);
        if (max_d &gt; c - 1)
            max_d = c - 1;

        for(long long d = min_d; d &lt;= max_d; ++d) {
            //c ^ 3 + 2 * c * d^2 &lt;= B / 3
            long long N = 3 * c * (c * c + 2*d*d);
            counts[N] += 1;
        }
    }

    int idx = 1;
    for (auto&amp; N : counts) {
        if (N.second &gt;= search_repre) {
            std::cout &lt;&lt; idx &lt;&lt; &quot; &quot; &lt;&lt; N.first &lt;&lt; &quot; &quot; &lt;&lt; int(N.second) &lt;&lt; std::endl;
            idx += 1;
            search_repre += 1;

            if (search_repre &gt; search_up_to) {
                return 0;
            }
        }
    }

    A = B + 1;
    B += segment_size;
}

return 0;

}

Sil
  • 16,612
  • 1
    This answer deserves to be the best! – Piquito Dec 17 '22 at 09:18
  • 1
    Your number $57426732000$ is also a sum of four cubes because it is not congruent to $\pm4$ modulo $9$. An old conjecture still unproved is an analogue for cubes of Lagrange's theorem on sum of four squares. It is almost assumed that it is true that all integer is a sum of four cubes, however it remains to prove it for numbers of the form $9n\pm4$. – Piquito Dec 17 '22 at 09:52