\subsection{Removing the cubic height without asymptotics}\label{sec:airyformal} The extra term $T^3$ can be dealt with algebraically. This avoids using an asymptotic formula for a cubic Gaussian integral. \begin{lemma}[An algebraic Riccati obstruction]\label{lem:airyriccati} The equation $f_x=x-f^2$ has no solution algebraic over $\C(x)$. \end{lemma} \begin{proof} At a finite point, a finite branch of $f$ satisfies a regular analytic first-order equation and cannot branch. At a pole, put $g=1/f$. Then $g_x=1-xg^2$, again a regular equation; a zero of $g$ is simple. Thus the normalized algebraic curve of $\C(x,f)$ is unramified over every finite $x$. If its degree is $d$ and there are $r$ points over infinity, Riemann--Hurwitz gives $2g-2=-d-r$. This forces $d=r=1$. So $f$ would be rational. An integer-order pole of $f$ at infinity makes $f^2$ dominate both $x$ and $f_x$; if $f$ is bounded there, the term $x$ cannot cancel. Both possibilities are impossible. \end{proof} \begin{lemma}[Cubic Gaussian obstruction]\label{lem:cubicgaussian} Let $A,B,C\in\C(z)$ be regular at zero, with $A(0)=C(0)=0$, $B(0)=-1/2$ and $A\ne0$. Define the formal unit \[ I=\E_T\exp\bigl(AT^3+(B+1/2)T^2+CT\bigr). \] Then $I'/I$ is not rational in $z$. \end{lemma} \begin{proof} Write $M_j=\E[T^j\exp(AT^3+(B+1/2)T^2+CT)]$ and $J=M_1$. Gaussian integration by parts gives, for $j\geq0$, \begin{equation}\label{eq:momentrec} 3A M_{j+2}+2B M_{j+1}+C M_j+jM_{j-1}=0, \end{equation} where the last term is zero for $j=0$. Also $I'=A'M_3+B'M_2+C'J$ and $J'=A'M_4+B'M_3+C'M_2$. For clarity, the first two eliminations are \[ M_2=-\frac{2BJ+CI}{3A},\qquad M_3=\frac{(4B^2-3AC)J+(2BC-3A)I}{9A^2}. \] The case $j=2$ of \eqref{eq:momentrec} supplies $M_4$. In an algebraic extension choose $a^3=3A$ and put \[ X=\frac{B^2-3AC}{a^4},\qquad F=\frac JI, \qquad f=-aF-\frac B{a^2}. \] Substitution in these three recurrences, using $a'=A'/a^2$, gives \begin{align} \frac{I'}I&=U-aX'F, &U&=\frac{A'(2BC-3A)}{9A^2}-\frac{B'C}{3A},\label{eq:airyfirst}\\ f'&=X'(X-f^2).\label{eq:airypullback} \end{align} These identities follow by differentiation and polynomial substitution; they do not involve an integration contour or a limit of a Gaussian integral. They can also be checked by substituting $A=a^3/3$ in \eqref{eq:momentrec} before differentiating. If $A$ has order $q>0$, then $X$ has a pole of order $4q/3$, because $B^2-3AC$ is a unit. Thus $X'\ne0$ and $aX'$ is a nonzero rational function. If $I'/I$ were rational, \eqref{eq:airyfirst} would make $F$ rational and $f$ algebraic over $\C(z)$. The function field $\C(z,a)$ is finite over $\C(X)$, so $f$ would be algebraic over $\C(X)$. Equation~\eqref{eq:airypullback} contradicts Lemma~\ref{lem:airyriccati}. \end{proof}