gpt 6 astra ultra attempt on Math-370

Beal's Conjecture

Definitions and mathematical statement

Beal's conjecture asserts that for positive integers $A,B,C$ and exponents $x,y,z>2$, \[ A^x+B^y=C^z\quad\Longrightarrow\quad\gcd(A,B,C)>1. \] Equivalently, some prime must divide all three bases. If a prime divides two bases in such an equation, it divides the third, so excluding a common prime also enforces pairwise coprimality. The exponents may be different and need not be prime. Positivity excludes zero and signed degeneracies. Unlike a finiteness statement for exceptional solutions, the assertion prohibits every coprime solution in this exponent range.

Formulation and concept references: [S1].

Short English statement

Whenever the sum of two positive perfect powers equals a third and all exponents exceed two, must their bases share a prime factor?

Research attempt

Status: UNRESOLVED. No primitive solution is constructed and no proof excluding every primitive solution is obtained. The attempt reduces the possible exponent signatures, gives a complete elementary descent for a substantial quartic subcase, and proves a conditional finiteness statement from the $abc$ conjecture while explaining why finiteness falls short of Beal's assertion.

Statement and source audit.

The equation in the catalogue is the positive-integer equation \[ A^x+B^y=C^z,\qquad x,y,z>2. \] The formulation on Mauldin's page [S1], checked on 27 September 2026, asks that the bases have a common prime divisor. It allows unequal exponents and bases equal to one. Throughout this attempt a primitive solution means $\gcd(A,B,C)=1$. Results imported from Fermat's Last Theorem and Mihăilescu's theorem are explicitly identified below; their deep proofs are not replaced by elementary manipulations. Any use of $abc$ is conditional.

1. Primitivity, signs, and the unit bases.

If a prime divides any two of $A,B,C$, the equation forces it to divide the third. For example, $q\mid A,B$ gives $q\mid C^z$ and hence $q\mid C$. The other two cases follow by rearranging. Therefore a primitive solution is automatically pairwise coprime. Conversely, pairwise coprimality implies primitivity. This equivalence depends on the equation; it is not true for arbitrary triples of integers.

The case $C=1$ is impossible because the left side is at least two. If $A=B=1$, the equality $C^z=2$ is also impossible. If exactly one of $A,B$ is one, the equation becomes a difference of two perfect powers equal to one. The classical theorem of Mihăilescu says that the only consecutive nontrivial perfect powers, with both exponents greater than one, are $8$ and $9$. Their upper exponent is two, so this cannot occur with both $y,z>2$ (or both $x,z>2$ after interchanging the summands). Thus, using that external theorem, every hypothetical primitive solution has $A,B,C\ge2$.

Positivity matters in the estimates later: both $A^x$ and $B^y$ are strictly smaller than $C^z$. An argument allowing arbitrary signed bases or moving a negative term without checking parity would be treating a different equation.

2. Reduce exponents individually, not to a common exponent.

Every integer $m>2$ either has an odd prime divisor or is divisible by four. Choose a divisor $r$ of $x$, a divisor $s$ of $y$, and a divisor $t$ of $z$, each belonging to \[ \{4\}\cup\{\text{odd primes}\}. \] Then \[ (A^{x/r})^r+(B^{y/s})^s=(C^{z/t})^t. \] Raising a positive integer to a positive power does not change its prime support, so pairwise coprimality and primitivity are preserved. Hence excluding primitive solutions for every signature $(r,s,t)$ of this restricted type would prove the conjecture.

This reduction still leaves infinitely many mixed signatures. It is invalid to replace the three chosen divisors by a single common odd prime when no such divisor exists. In particular, signatures such as $(3,4,5)$ survive untouched. Fermat's Last Theorem concerns equal exponents; it does not settle all equations whose three exponents happen to be at least three.

There is a correct common-divisor subcase. If $d=\gcd(x,y,z)>2$, then \[ (A^{x/d})^d+(B^{y/d})^d=(C^{z/d})^d \] contradicts Fermat's Last Theorem. Thus there are no positive solutions at all in that subcase. If $d=2$, the transformed equation is a Pythagorean equation, for which positive solutions exist, so the same inference is unavailable. If $d=1$, it gives no restriction.

A common factor of the bases also cannot simply be divided out while preserving the original exponents. Writing $A=aD,B=bD,C=cD$ introduces different powers $D^x,D^y,D^z$. Conversely, if any solution exists, then for $L$ a common multiple of $x,y,z$ and an integer $u>1$, \[ (u^{L/x}A)^x+(u^{L/y}B)^y=(u^{L/z}C)^z \] is another solution with a common base divisor. This explains why producing large nonprimitive solutions is easy and irrelevant to disproving the assertion. For example, $2^3+2^3=2^4$ and $3^3+6^3=3^5$ satisfy the equation and its predicted common-divisor conclusion.

3. Complete descent for $X^4+Y^4=Z^2$.

We prove that there are no positive integers satisfying this auxiliary equation. If a solution exists, divide $X,Y$ by their greatest common divisor $d$. Since $d^4\mid Z^2$, valuation comparison gives $d^2\mid Z$, so this produces a primitive solution. Choose a primitive solution with $Z$ minimal. The two bases cannot both be odd, because fourth powers of odd integers are one modulo 16 and a square cannot be two modulo 16. They cannot both be even by primitivity. Interchanging them, take $X$ odd and $Y$ even.

The primitive Pythagorean triple $(X^2,Y^2,Z)$ therefore has the form \[ X^2=m^2-n^2,\qquad Y^2=2mn,\qquad Z=m^2+n^2, \] with $m>n>0$, $\gcd(m,n)=1$, and opposite parity. As $X^2\equiv1\pmod4$, the first identity forces $m$ odd and $n$ even. Unique factorization in $2mn=Y^2$ now gives \[ m=a^2,\qquad n=2b^2,\qquad Y=2ab, \qquad \gcd(a,b)=1. \] Indeed every odd prime appears with even valuation in each of the coprime factors, while the 2-adic valuation of $n$ is odd. Substitution yields \[ X^2+(2b^2)^2=a^4. \] This is another primitive Pythagorean triple: $X$ is odd, and a prime dividing $X$ and $b$ would divide $a$ from the displayed identity. Parameterize it as \[ X=r^2-s^2,\qquad 2b^2=2rs,\qquad a^2=r^2+s^2, \] with $r>s>0$ and $\gcd(r,s)=1$. Since $b^2=rs$ and the factors are coprime, $r=c^2$ and $s=d^2$ for positive integers $c,d$. Thus \[ c^4+d^4=a^2 \] is a new solution of the original auxiliary equation, with smaller last variable: $a<Z=a^4+4b^4$. This contradicts minimality. The descent is complete.

It follows that the Beal equation has no positive solutions whenever \[ 4\mid x,\qquad 4\mid y,\qquad 2\mid z. \] Set $X=A^{x/4},Y=B^{y/4},Z=C^{z/2}$ and apply the proved result. This includes some mixed signatures which the common-exponent argument alone misses. It does not cover, for example, a cubic summand with an odd exponent on the other summand. The descent relies twice on one leg already being a square and on a particular even leg having the form $2b^2$; those identities cannot be assumed for an arbitrary mixed signature.

4. What $abc$ would prove uniformly.

Assume the following standard $abc$ statement: for every $\delta>0$ there is a constant $K_\delta$ such that pairwise coprime positive integers $a+b=c$ satisfy \[ c\le K_\delta\operatorname{rad}(abc)^{1+\delta}, \] where $\operatorname{rad}$ denotes the product of distinct prime divisors. Apply this to a primitive Beal solution, with $a=A^x,b=B^y,c=C^z=M$. Its radical is $\operatorname{rad}(ABC)$, bounded above by $ABC$. Since $A<M^{1/x}$, $B<M^{1/y}$, and $C=M^{1/z}$, \[ M\le K_\delta M^{(1+\delta)\sigma},\qquad \sigma=\frac1x+\frac1y+\frac1z. \] The signature $(3,3,3)$ has no solution by Fermat's Last Theorem. In every other signature with exponents at least three, at least one exponent is at least four, so \[ \sigma\le\frac13+\frac13+\frac14=\frac{11}{12}. \] Choose the fixed number $\delta=1/22$. Then $(1+\delta)\sigma\le23/24$, and consequently \[ M^{1/24}\le K_{1/22},\qquad M\le K_{1/22}^{24}. \] This bound is uniform over all remaining exponent signatures. After the unit bases have been excluded, $A,B,C\ge2$, so each exponent is at most $\log_2M$ and all bases are bounded by $M$. Hence $abc$, together with the named established theorems, implies that only finitely many primitive Beal solutions can exist.

The conclusion is finiteness, not emptiness. An unknown $abc$ constant does not give a completed finite search, and even an effective bound would require checking every case below it. It is therefore incorrect to announce Beal's conjecture proved merely from this exponent-sum argument. There is also no unconditional $abc$ input being asserted here: the whole paragraph is a conditional implication, with the exact assumption and its use displayed.

5. Congruences and finite searches.

Congruences can discard base residue patterns for a fixed signature, but a locally admissible residue pattern need not lift to a positive integer solution. The obvious residues $A\equiv0,B\equiv C\equiv1$ modulo an auxiliary modulus already show why some patterns survive many simple tests. They do not satisfy a common-prime conclusion as an integer identity, and local admissibility supplies no global counterexample.

For reproducibility, the offline diagnostic uses exact integer powers, not floating-point roots. It stores every value $a^e$ with $1\le a\le100$ and $3\le e\le8$, retaining all representations of values such as $64$. It adds each pair of stored values and looks up every representation of the sum in the same table. It checks the common-divisor conclusion for each resulting equation, includes the displayed nonprimitive examples, and records the exact finite box searched. A separate check enumerates primitive Pythagorean parameters and verifies the algebra and parity identities used in the descent. No result outside that box is inferred from the computation.

Perfect-power representations must not be collapsed to a single arbitrary exponent: $a^e=b^f$ can lead to different exponent signatures, and the problem quantifies over all of them. Likewise, the search cutoff on $C$ is a cutoff on a base, not on the entire integer $C^z$. These distinctions are included to keep the computational claim checkable.

6. The remaining gap.

The proven reductions leave primitive mixed signatures with no common exponent divisor greater than two and outside the quartic descent pattern. Addressing all of them would require a uniform obstruction, or a reduction to finitely many cases with a fully effective and completed verification. Neither is provided. The argument does establish exactly why the equal-exponent theorem, quartic descent, and conditional radical bound are relevant, and exactly why their union is still insufficient. The unrestricted assertion remains unresolved by this attempt.

Sources

• S1 R. D. Mauldin, The Beal Conjecture and Prize, University of North Texas; announced in Notices Amer. Math. Soc. 44 $1997$, 1436-1437. https://sites.math.unt.edu/~mauldin/beal.html. Formulation source.
• E1 Preda Mihăilescu, Primary cyclotomic units and a proof of Catalan's conjecture, Journal für die reine und angewandte Mathematik 572 $2004$, 167--195. https://doi.org/10.1515/crll.2004.048. Added primary source: The theorem on consecutive perfect powers, used only for unit bases. Publisher bibliographic record consulted 27 September 2026.
• E2 Andrew Wiles, Modular elliptic curves and Fermat's Last Theorem, Annals of Mathematics 141 $1995$, 443--551. https://annals.math.princeton.edu/1995/141-3/p01. Added primary source: Established equal-exponent theorem, used as external input rather than reproved. Consulted 27 September 2026.