\subsection{A uniform cut criterion in any degree} \label{sec:uniformcutdecay} Intersecting critical curves need not admit a simple formal normal form. The following criterion avoids choosing one. Its hypothesis concerns the full phase $tP-q$, uniformly in large complex $t$; bounded critical points alone would not suffice. \begin{theorem}\label{thm:uniformcuts} Let $P,q\in\C[x_1,\ldots,x_n]$, with $P$ nonconstant. Suppose there are fixed $c,R_0,t_0>0$ such that \begin{equation}\label{eq:uniformphasegradient} \|\nabla_x(tP-q)(x)\|\ge c\|x\| \quad(|t|\ge t_0,\ \|x\|\ge R_0). \end{equation} Along every fixed ray $t=Re^{i\theta}$, the full rapid-decay homology has a flat cut basis $\Gamma_i(t)$, attached to the critical values $c_i(t)$ of $tP-q$, such that for every fixed polynomial $a$, \begin{equation}\label{eq:normalizedcutdecay} e^{-c_i(t)}\int_{\Gamma_i(t)}a(x)e^{tP-q}\, dx_1\cdots dx_n\longrightarrow0. \end{equation} Each $c_i(t)=\alpha_i t+O(1)$, where $\alpha_i$ is a critical value of $P$. If all critical values of $P$ lie in a closed half-plane through zero, a supporting $t$ ray makes every flat period of every such fixed form tend to zero. \end{theorem} We give the completeness, tail and limiting-form arguments separately. None follows just from counting the phase-critical points. \begin{proof} For every sufficiently large $t$, \eqref{eq:uniformphasegradient} makes the phase-critical set finite and contained in one fixed ball. It also gives complex tameness at infinity. Pham's tame Milnor decomposition~\cite[Part~I and Appendix]{Pham} identifies the complete middle rapid-decay homology with the local vanishing homologies carried along cuts from its critical values. The rank is the sum of their Milnor numbers, including degenerate points. If this rank is zero the conclusion is immediate. Otherwise retain all these local generators; identical critical values are grouped, not discarded. On the chosen ray, the critical points and values are algebraic functions of $R$, after a finite cover. Enlarge $R_1$ so branches and multiplicities stabilize. Choose $|\eta|=1$ with $\Re\eta\ge1/2$ so the cuts \[ S_t(x)=c_i(t)-\eta s,\qquad s\ge0,\quad S_t=tP-q, \] are disjoint for $R>R_1$, except at identical grouped endpoints. There are only finitely many forbidden choices: every nonzero difference of two algebraic critical values has one limiting direction. Choose this auxiliary cut direction after fixing $\theta$; it imposes no genericity condition on $\theta$ itself. In particular a supporting $t$ ray need not be a generic ray. Along a cut, normalized complex gradient transport is \[ x_s=-\eta\,\overline{\nabla S_t}/\|\nabla S_t\|^2, \qquad S_t(x(s))_s=-\eta. \] Outside the fixed ball, $|d\|x\|^2/ds|\le2/c$. All local vanishing generators start in that ball. With fixed sufficiently large $A$ and $B>2/c$, they therefore have representatives inside the carrier \begin{equation}\label{eq:uniformcarrier} \|x\|^2\le A+Bs. \end{equation} At its outer boundary the transport points strictly inward relative to this inequality. No trajectory escapes at finite cut length, and the cuts avoid all other critical values. We must control a single family of representatives. The gradient flow itself need not be semialgebraic. Instead compactify $x$ by $x/\sqrt{1+\|x\|^2}$ and $s$ by $s/(1+s)$. Take the closure of each constrained cut carrier and mark its rapid boundary $G$, where $s=\infty$. Inequality~\eqref{eq:uniformcarrier} forbids spatial infinity at finite $s$. A compatible semialgebraic triangulation computes the relative homology of this compact pair by locally finite semialgebraic cut chains, with boundary allowed only in $G$. These representatives are admissible rapid-decay chains: on the carrier, \[ s\geq(\|x\|^2-A)/B,\qquad |e^{-\eta s}|\leq e^{-s/2} \leq \exp\!\left(-\frac{\|x\|^2-A}{2B}\right). \] Semialgebraic approach to a boundary has at most power-law rates, so this decay dominates every boundary pole of a polynomial form. The chain-current boundary identity applies on finite phase bands; the tail estimate below justifies passage to the improper integral. Thus the compact-pair calculation supplies admissible cycle representatives, without identifying arbitrary singular chains with integration chains. Their exponential is exactly $e^{c_i-\eta s}$. For the full unconstrained cut carrier, mark and remove any spatial boundary at finite $s$; it is not a rapid boundary. The constrained carrier, the full carrier, their rapid boundaries and this excluded boundary form a finite family of semialgebraic subsets in the same compactification. Apply the compatible finite-family form of Hardt triviality to projection onto $R$. It gives one trivialization on an unbounded interval, preserving these subsets and their inclusions. The precise form is stated in Drusvyatskiy--Lewis~\cite[Theorem~2.25]{DrusvyatskiyLewis}; it is Hardt's theorem. Semialgebraic triangulation and the chain-current boundary rule are as in Hardt--Lambrechts--Turchin--Voli\'c~\cite[Sections~2.1 and~3, especially Theorem~3.5]{HLTV}. For precision, the pairs in this application are as follows. In the compactified $(x,s)$ space write $\overline K_i(R)$ for the closure of the constrained cut carrier, $G_i(R)=\overline K_i(R)\cap\{s=\infty\}$, and write $\overline K_i^{\rm full}(R)$ for the unconstrained closure. Let $E_i(R)$ be its spatial boundary at finite $s$, and put $G_i^{\rm full}=\overline K_i^{\rm full}\cap\{s=\infty\}$. The relative pairs and their inclusion are \[ (\overline K_i,G_i)\longrightarrow (\overline K_i^{\rm full}\setminus E_i, G_i^{\rm full}\setminus E_i). \] The constrained pair has no $E_i$, by \eqref{eq:uniformcarrier}. At $s=0$ a local vanishing cycle closes by its collapsing thimble; that endpoint is included in $K_i$, rather than declared a rapid boundary. In degree $n$, projection to $x$ takes these relative classes to the rapid-decay homology of $e^{S_t}$. The tame cut comparison above says that the images, summed over the cuts, contain the entire vanishing-homology decomposition. We use this image; we need not identify every relative group of an individual carrier with the full rapid-decay group. The compatible Hardt trivialization preserves both pairs, their inclusion, and hence the induced maps on their relative homology. Gradient transport already showed that the constrained inclusion contains every local vanishing generator. At one parameter choose finitely many integral relative chains whose images are a basis over $\C$ of the full rapid-decay space. A nonprimitive integral image causes no problem over $\C$. Their number is the total tame Milnor rank, and their images are independent at this parameter by construction. Transport these chains by the compatible trivialization. They form a jointly semialgebraic family with fixed integer multiplicities and supports satisfying \eqref{eq:uniformcarrier}. Their images are a flat basis of the actual homology local system: on each compact parameter interval the transport preserves the relative pairs and is proper on every finite phase band. The induced comparison maps commute with this transport, so the image basis remains independent and spanning. This establishes flatness and completeness before any mass estimate is made; a uniform bound for arbitrary representatives would not establish either property. Here is the needed uniform mass bound. In a fixed semialgebraic family of $d$-chains with bounded multiplicities, the mass inside a ball of radius $L\ge1$ is at most $C L^d$, uniformly in the parameters. Partition tangent planes by their largest coordinate $d$-minor. The Cauchy--Binet identity gives a lower bound $\binom Nd^{-1/2}$ for some such minor in ambient dimension $N$. The area formula then bounds that piece by its coordinate projection volume times the number of preimages. Semialgebraic uniform finiteness bounds those finite fibers independently of the parameters; see, for example, \cite[Theorem~2.32]{DrusvyatskiyLewis}. Sum the finitely many pieces and multiplicities. Apply the same argument to boundary chains. This proves uniformity, rather than merely pointwise finite volume. Lift the cut chains to $(x,s)$. Their bands $j\le s\le j+1$ lie in balls of radius $O(1+j)$ by \eqref{eq:uniformcarrier}. Their fixed semialgebraic descriptions and the mass estimate give, with constants independent of $R$, \begin{equation}\label{eq:uniformcuttail} \left|e^{-c_i(t)}\int_{s\ge L}a e^{S_t}\,dx_1\cdots dx_n\right| \le C\sum_{j\ge\lfloor L\rfloor}(1+j)^M e^{-j/2}. \end{equation} The polynomial amplitude changes only $M$. This tail tends to zero uniformly in $R$. It remains to bound a fixed phase band. A bounded critical branch $x(r)$, $r=1/t$, satisfies \[ q(x(r))=-c_i(r)-r c_i'(r). \] At a stationary point the chain rule gives $c_i'(r)=-P(x(r))/r^2$; eliminating $P$ gives this identity. Since $q(x(r))$ is bounded, its Puiseux expansion forces the only possible polar term of $c_i$ to be $\alpha_i/r$. Its coefficient is $\alpha_i=P(x(0))$, and $\nabla P(x(0))=0$. Hence $c_i(t)=\alpha_i t+O(1)$ even at critical collisions. Choose a generic fixed truncation level $L$. The truncated lifted chains have compact supports and uniformly bounded mass and boundary mass. The latter follows from the same semialgebraic estimate applied to their slices; after a finite stratification there is an arbitrarily large sequence of such levels. By the Federer--Fleming compactness theorem, every sequence $R\to\infty$ has a subsequence converging to an integral $n$-current. We use the version with both mass and boundary-mass bounds, Simon~\cite[Chapter~6, Section~3, Theorem~3.15]{SimonGMT}. Its rectifiability conclusion is essential. On the lifted chain, \[ P(x)-\alpha_i= \frac{c_i(t)-\alpha_i t+q(x)-\eta s}{t}. \] Every limiting current is therefore supported in $\{P=\alpha_i\}\times[0,L]$. Stratify this complex algebraic level set into smooth complex strata. Almost everywhere for the rectifiable limiting current, its tangent plane projects into the tangent of one of these strata. Each has complex dimension less than $n$. The holomorphic $n$-form $dx_1\cdots dx_n$ thus vanishes on that tangent plane. Lower-dimensional strata carry no $n$-mass; the same argument covers singular strata of larger real dimension. Testing with the fixed smooth form $e^{-\eta s}a(x)dx_1\cdots dx_n$, with a compact cutoff, gives zero on every subsequential limit. Consequently the normalized integral on $0\le s\le L$ tends to zero. This does not assert that the real volume of the chains tends to zero. First let $R\to\infty$, then let $L\to\infty$ through the allowed levels in \eqref{eq:uniformcuttail}. This proves \eqref{eq:normalizedcutdecay}. Finally choose $\theta$ with $\Re(e^{i\theta}\alpha)\le0$ for all critical values $\alpha$ of $P$. The factors $e^{c_i(t)}$ are then bounded, so every basis period tends to zero. A flat cycle is a constant linear combination of this full basis, proving the last assertion. \end{proof} \begin{corollary}\label{cor:twocriticalvalues} Let $q$ be a full affine Gaussian quadratic and assume \eqref{eq:uniformphasegradient}. If $P$ has at most two critical values, every original marked Gaussian period is either zero or is not a finite exponential-polynomial. In particular $P$ is neither pure nor safe. \end{corollary} \begin{proof} Any two complex numbers lie in a closed half-plane through zero. For two nonzero values normalize them to $z_1,z_2$ on the unit circle. If $z_1+z_2\ne0$, use the direction $-\overline{z_1+z_2}/|z_1+z_2|$; if their sum is zero, use $i\overline{z_1}$. Zero or missing values add no condition. A nonzero constant original marked period would, by the rational seed restriction and actual-period extension in Lemma~\ref{lem:seedfunctional}, be represented by a flat period of that fixed polynomial form. Its generic exponential system has $f=t$, $g=tP-q$ on smooth affine space and regular input $\mathcal O$; $f,g$ are algebraically independent because $P$ is nonconstant. Remove the finite exceptional parameter set of the Hien--Roucairol comparison. Theorem~\ref{thm:uniformcuts} makes the resulting constant period tend to zero, a contradiction. No convergence of the formal moment series was assumed. A constant-coefficient operator reduces a nonzero finite exponential-polynomial to $c e^{\beta t}$ while replacing its original marker by another fixed polynomial in $P$ times that marker. Shift $P$ by $-\beta$ and normalize. Its gradient is unchanged and its critical values are still at most two. The same contradiction applies. Since $\Phi_1(0)=1$, it cannot be a polynomial, proving also the safety exclusion. \end{proof}