Lemma 9.11.19.5D5A847F99AC Let $C$ be a closed full-dimensional convex cone in a Euclidean space, let $g$ be finite and convex, and let $\ell $ be a linear form. Then
\[ \int _ C\exp \bigl (2\langle \ell ,x\rangle -2g(x)\bigr )\, \mathrm dx<\infty \quad \Longleftrightarrow \quad \langle \ell ,v\rangle <(\operatorname {rec}g)(v) \quad (v\in C\setminus \{ 0\} ). \]
This condition is also equivalent to $\ell $ being an interior point of $\{ a:\sup _ C(\langle a,x\rangle -g(x))<\infty \} $. The integral has the same finiteness on every translate of $C$.
Proof.
Set $F=g-\ell $. Suppose its recession is positive on $C\setminus \{ 0\} $. Then $F(x)\geq \gamma |x|-A$ on $C$ for some $\gamma >0$ and $A$. Otherwise there are $x_ j\in C$ with $|x_ j|\to \infty $ and $F(x_ j)/|x_ j|\leq 1/j$. After taking a subsequence, $x_ j/|x_ j|\to v\in C$ with $|v|=1$. For fixed $t>0$, convexity gives
\[ F\left(t\frac{x_ j}{|x_ j|}\right) \leq \left(1-\frac t{|x_ j|}\right)F(0) +\frac t{|x_ j|}F(x_ j). \]
Continuity gives $F(tv)\leq F(0)$ for every $t$, contradicting positive recession. A bound on the remaining compact set gives the asserted $A$. Exponential decay proves integrability. The same bound persists when $\ell $ is changed by a sufficiently small linear form, so $\ell $ lies in the stated interior.
Conversely, if $(\operatorname {rec}F)(v)\leq 0$ for some nonzero $v\in C$, then $F(tv)\leq F(0)$ for $t\geq 0$. Choose a compact ball $K$ in the interior of $C$. On the half-tube $\{ tv+x/2:t\geq 1,\ x\in K\} $, convexity gives
\[ F(tv+x/2)\leq \tfrac 12F(2tv)+\tfrac 12F(x). \]
Thus $\mathrm e^{-2F}$ has a positive lower bound on a set of infinite volume. If instead $\ell $ is in the stated interior, for each nonzero $v\in C$ one may perturb $\ell $ within that interior by a linear form positive on $v$. Dividing its defining upper bound along $tv$ by $t$ gives the strict recession inequality for $\ell $. Finally, translating $C$ replaces $g(x)$ by $g(x+b)$ and multiplies the exponential by a constant; recession is unchanged.
$\square$
Comments
Corrections, questions, and mathematical discussion are welcome. Sign in with GitHub to comment. Comments are public.