gpt 6 astra ultra attempt on Math-14

Schanuel's Conjecture

Definitions and mathematical statement

Numbers $z_1,\ldots,z_n$ are ${\mathbb{Q}}$-linearly independent when $\sum_jq_jz_j=0$ with $q_j\in{\mathbb{Q}}$ forces all $q_j=0$. The transcendence degree of a field extension is the maximum size of an algebraically independent subset. Schanuel's assertion is \[ \dim_{{\mathbb{Q}}}\operatorname{span}_{{\mathbb{Q}}}\{z_1,\ldots,z_n\}=n \quad\Longrightarrow\quad \operatorname{trdeg}_{{\mathbb{Q}}}{\mathbb{Q}}(z_1,\ldots,z_n,e^{z_1},\ldots,e^{z_n})\ge n. \] It is quantified over every $n\ge1$ and every such complex tuple; the $z_j$ themselves need not be algebraic.

Formulation and concept references: [S1].

Short English statement

Do linearly independent complex numbers, together with their exponentials, always contain at least as much algebraic independence as Schanuel predicts?

Research attempt

Status and scope.

This is an unresolved attempt, dated 27 September 2026. The aim is to identify exactly what a reduction to independent inputs accomplishes, prove the algebraic-input case using a named classical theorem, and test a proposed induction on exponential towers. The elementary reductions below are not claimed to be new. None gives the missing transcendence estimate for arbitrary complex inputs.

The external theorem used is Lindemann--Weierstrass: exponentials of distinct algebraic numbers are linearly independent over the algebraic numbers. A primary proof is [E1]. Ax's theorem [E2] concerns exponentiation in differential fields. Its functional scope is different from the numerical assertion here; no specialization principle is assumed. The power-tower reference [S1] is retained as context, not used to turn a conditional conclusion into a theorem.

1. Removing dependent coordinates without losing exponentials.

For a finite tuple $z=(z_1,\ldots,z_m)$, put \[ K(z)=\mathbb Q(z_1,\ldots,z_m,e^{z_1},\ldots,e^{z_m}),\qquad \delta(z)=\operatorname{trdeg}_{\mathbb Q}K(z) -\dim_{\mathbb Q}\langle z_1,\ldots,z_m\rangle. \] Choose a basis $b_1,\ldots,b_r$ of this rational span from among the $z_i$. For a common positive denominator $D$ there are integers $a_{ij}$ with $Dz_i=\sum_j a_{ij}b_j$. Hence \[ (e^{z_i})^D=\prod_{j=1}^r(e^{b_j})^{a_{ij}}. \tag{28.1} \] The right side belongs to $K(b)$, including when some exponents are negative, because exponentials never vanish. Each $z_i$ already belongs to $K(b)$ and each $e^{z_i}$ is algebraic over it. Conversely $K(b)\subseteq K(z)$ because the basis was selected from the tuple. Thus $K(z)/K(b)$ is algebraic and \[ \operatorname{trdeg}_{\mathbb Q}K(z) =\operatorname{trdeg}_{\mathbb Q}K(b),\qquad \delta(z)=\delta(b). \tag{28.2} \] There is no logarithm branch choice in this proof: $28.1$ follows from the single-valued exponential addition law. The assertion for independent tuples is therefore equivalent to $\delta(z)\geq0$ for every finite tuple, including dependent ones. Rank zero consists of zero inputs and gives $K(z)=\mathbb Q$.

This reformulation is useful bookkeeping, but it supplies no lower bound on the transcendence degree. In particular, selecting a rational basis does not make its entries algebraic. Any argument applying Lindemann--Weierstrass immediately after this selection would be using an unproved hypothesis.

2. Two unconditional subcases.

For one nonzero complex number $z$, the conjecture follows at once. If $z$ is transcendental, the field already contains one transcendental element. If $z$ is algebraic, Lindemann--Weierstrass applied to $0,z$ makes $e^z$ transcendental. This establishes the entire rank-one case, not just real or algebraic rank-one inputs.

Now suppose $b_1,\ldots,b_r$ are algebraic and rationally independent. Let $P\in\overline{\mathbb Q}[X_1,\ldots,X_r]$ be a nonzero polynomial, written with its distinct monomials as \[ P(X)=\sum_{\nu\in A}c_\nu X_1^{\nu_1}\cdots X_r^{\nu_r}, \qquad c_\nu\ne0. \] At the exponentials its value is \[ P(e^{b_1},\ldots,e^{b_r}) =\sum_{\nu\in A}c_\nu\exp(\nu_1b_1+\cdots+\nu_rb_r). \tag{28.3} \] The exponents are algebraic and distinct: equality for two indices would give the nontrivial rational relation $(\nu-\nu')\cdot b=0$. Lindemann--Weierstrass therefore makes $28.3$ nonzero. Consequently $e^{b_1},\ldots,e^{b_r}$ are algebraically independent, and $\operatorname{trdeg}_{\mathbb Q}K(b)=r$ since the inputs are algebraic. Together with $28.2$, this proves the rank formulation for every tuple of algebraic numbers. This familiar corollary is a genuine subcase, with its dependence on the classical theorem explicit.

The exact point of failure for transcendental inputs is visible in $28.3$: its exponents need no longer be algebraic. Their distinctness survives, but distinct complex exponents alone give no nonvanishing theorem for their exponential values. For example $0$ and $2\pi i$ are distinct and both exponentials equal $1$. That example is a stress test of the proposed use of the theorem, not a counterexample to Schanuel: the input $2\pi i$ itself contributes transcendence.

3. What a smallest counterexample would look like.

Suppose, conditionally, that the conjecture fails, and let $r$ be its smallest failing rank. The rank-one argument gives $r\geq2$. Choose an independent counterexample $b_1,\ldots,b_r$. Each $(r-1)$-coordinate subtuple satisfies the conjecture by minimality, so its field has transcendence degree at least $r-1$. Its field is a subfield of $K(b)$, whereas failure says the latter degree is less than $r$. Therefore necessarily \[ \operatorname{trdeg}_{\mathbb Q}K(b)=r-1,\qquad \delta(b)=-1. \tag{28.4} \] More generally every rational subspace of the span of dimension less than $r$ has nonnegative predimension, by choosing a basis and using minimality. Thus a least-rank obstruction would be a deficit of exactly one, invisible on all lower-dimensional rational subspaces. This narrows a hypothetical obstruction but does not rule it out. In particular, transcendence degree need not increase when a new pair $(b_r,e^{b_r})$ is adjoined; proving that it must increase is precisely where an induction would require new input.

One can also see why merely counting $2r$ generators is ineffective. A field generated by many named numbers can have much smaller transcendence degree because of polynomial relations. Rational independence excludes linear relations between the inputs, not arbitrary relations mixing inputs and outputs. Equation $28.4$ does not supply an algorithm for recognizing that mixing from decimal approximations.

4. A tower induction, with its hypothesis exposed.

Define real numbers $a_0=1$ and $a_{j+1}=e^{a_j}$. Assume Schanuel's conjecture. Then $a_1,\ldots,a_m$ are algebraically independent for every $m\geq1$. Here is the full induction. For $m=1$, $a_1=e$ is transcendental by the preceding theorem. Suppose $a_1,\ldots,a_{m-1}$ are algebraically independent. Any rational relation among $1,a_1,\ldots,a_{m-1}$ is a polynomial relation in the latter $m-1$ numbers, so all its coefficients vanish. Apply Schanuel to that $m$-tuple. The resulting field is exactly \[ \mathbb Q(1,a_1,\ldots,a_{m-1}, e,e^{a_1},\ldots,e^{a_{m-1}}) =\mathbb Q(a_1,\ldots,a_m). \] It has transcendence degree at least $m$, and it is generated by $m$ numbers, so those generators are algebraically independent.

This is a conditional consequence, not an induction proving Schanuel. Already the step from $e$ to $(e,e^e)$ invokes the conjecture on $(1,e)$, whose second input is transcendental. Positivity and rapid growth of the tower do not justify that invocation. The same distinction appears in the tuple $(1,i\pi)$: it is rationally independent by real and imaginary parts, and Schanuel would give transcendence degree at least two for $\mathbb Q(i\pi,e,-1)$. After adjoining the algebraic number $i$, this is equivalent to algebraic independence of $e$ and $\pi$. Their separate transcendence does not imply that conclusion.

5. A functional argument cannot be specialized for free.

It is possible to prove directly that the functions $t,e^t$ are algebraically independent over $\mathbb C$. Suppose $\sum_{j=0}^M P_j(t)e^{jt}=0$ with polynomial coefficients and $P_M\ne0$. Divide by $e^{Mt}$ and take real $t\to+\infty$. Every term with $j<M$ tends to zero, forcing the nonzero polynomial $P_M(t)$ to tend to zero, which is impossible. Thus the meromorphic function field $\mathbb C(t,e^t)$ has transcendence degree two over $\mathbb C$. In particular, $\mathbb Q(t,e^t)$ has transcendence degree two over $\mathbb Q$ as well, so the next comparison can keep the base field fixed.

At $t=1$, however, the corresponding numerical field $\mathbb Q(1,e)$ has transcendence degree only one over $\mathbb Q$. At $t=0$ both values are algebraic. These observations do not contradict the numerical conjecture; they show that independence of functions is not a license to retain the same degree at a prescribed point. Ax's differential theorem [E2] should therefore not be inserted at the missing numerical induction step. A valid specialization theorem would need additional arithmetic hypotheses and a proof that the chosen point satisfies them.

Verification and remaining gap.

The companion exact-arithmetic checks use the coefficient vectors of $\sqrt2,\sqrt3,(\sqrt2+\sqrt3)/2$. They confirm rational rank two and the cleared relation $2z_3-z_1-z_2=0$, which gives $(e^{z_3})^2=e^{z_1}e^{z_2}$. The independence of $\sqrt2,\sqrt3$ used to interpret these vectors is elementary: a nontrivial rational dependence would make $\sqrt{3/2}$ rational, contradicting the parity of prime valuations in a rational square. The script also checks distinct exponent vectors for all pairs of monomial indices between zero and eight in each coordinate. These finite checks test denominator handling and monomial bookkeeping only; the general injectivity proof is the argument in $28.3$. No finite search or numerical precision is reported as a transcendence test.

The outcome is an exact rank reduction, the full rank-one and algebraic-input subcases, a description of any least-rank failure, and explicit conditional consequences. The first unresolved step is a lower bound for $\operatorname{trdeg}\mathbb Q(b,e^b)$ when the independent inputs include transcendental numbers and their known transcendence does not already reach their rational rank. Neither the basis reduction, the tower construction, nor differential algebra fills that step. Schanuel's conjecture remains UNRESOLVED.

Sources

• S1 Schanuel's Conjecture and the Transcendence of Power Towers. https://www.mdpi.com/2227-7390/9/7/717. Formulation source.
• S2 Combining the conjectures of Schanuel and Zilber-Pink. https://content.ems.press/assets/public/full-texts/serials/rlm/36/3/14299588/online/10.4171-rlm-1086.pdf. Further reference; not a proof endorsement.
• S3 Schanuel Property for Elliptic and Quasi-Elliptic Functions. https://webusers.imj-prg.fr/~michel.waldschmidt/articles/pdf/SchanuelPropertyAlmostAll.pdf. Further reference; not a proof endorsement.

• E1 F. Beukers, J. P. Bézivin, and P. Robba, An Alternative Proof of the Lindemann--Weierstrass Theorem, American Mathematical Monthly 97 $1990$, no. 3, 193--197. https://doi.org/10.2307/2324683. Primary proof of the classical nonvanishing theorem used in the algebraic-input case. Full-text scan inspected at https://datuan5pdes.wordpress.com/wp-content/uploads/2016/12/10-23072324683.pdf. Consulted 27 September 2026.
• E2 James Ax, On Schanuel's conjectures, Annals of Mathematics 93 $1971$, no. 2, 252--268. https://annals.math.princeton.edu/1971/93-2/p03. Primary research reference for the differential-algebraic setting; the publisher's abstract was inspected to check scope. No numerical specialization theorem is inferred. Consulted 27 September 2026.