66FE6F0ACABE
We close the chapter with two estimates on how a weak geodesic depends on its endpoints. The first, proposition 17.9.1, shows that the essential supremum of $G_ t(u,v)-u$ is a linear function of $t$, and similarly for the essential infimum; essential suprema and infima are taken with respect to the volume form $\omega ^ n$. The second, proposition 17.9.2, concerns long segments from $\phi $ to the truncations of a singular potential.
Proposition 17.9.1.0583958ACA6D Let $u,v\in \mathcal E(X,\theta ,\phi )$ with $v\preceq u$, and put $C=\operatorname *{ess\, sup}_ X(v-u)$. Then $C$ is finite and
\[ \operatorname *{ess\, sup}_ X(G_ t(u,v)-u)=tC,\qquad 0\leq t\leq 1. \]
If moreover $u\sim v$ and $c=\operatorname *{ess\, inf}_ X(v-u)$, then $\operatorname *{ess\, inf}_ X(G_ t(u,v)-u)=tc$. These essential extrema equal the corresponding extrema over the common finite locus, and they do not change if a pluripolar set is removed.
Proof.
The function $v-u$ is finite almost everywhere, and $v\preceq u$ gives $v-u\leq C'$ for some constant $C'$; hence $C$ is finite. The inequality $v\leq u+C$ holds almost everywhere, hence everywhere, as recalled in section 16.11. Thus the supremum of $v-u$ over the common finite locus is at most $C$; it is at least $C$ because the common finite locus has full measure. The same argument applies to an infimum, and to the complement of any pluripolar set, which still has full measure. The same remarks apply to $G_ t-u$.
By theorem 17.7.8, $G_ t\leq (1-t)u+tv\leq u+tC$. The affine curve $v-C(1-t)$ is a competitor, since $v-C\leq u$, so $G_ t\geq v-(1-t)C$. On the common finite locus $v-u=(v-G_ t)+(G_ t-u)$, and taking essential suprema gives $C\leq (1-t)C+\operatorname *{ess\, sup}_ X(G_ t-u)$. Together with the first bound, this proves the equality for the supremum.
Now let $u\sim v$. The affine curve $u+ct$ is a competitor, since $u+c\leq v$, so $G_ t-u\geq ct$. Applying the equality for the supremum to the reversed segment, $G_ t(u,v)=G_{1-t}(v,u)$, gives $\operatorname *{ess\, sup}_ X(G_ t-v)=(1-t)\operatorname *{ess\, sup}_ X(u-v)=-(1-t)c$, that is, $\operatorname *{ess\, inf}_ X(v-G_ t)=(1-t)c$. Taking essential infima in $v-u=(v-G_ t)+(G_ t-u)$ gives $c\geq (1-t)c+\operatorname *{ess\, inf}_ X(G_ t-u)$, which proves the opposite bound.
$\square$
We now turn to long segments. Given a potential $u\leq \phi $ and $T>0$, we join $\phi $ to the truncation $\max \{ u,\phi -T\} $ by a weak geodesic parametrized by $[0,T]$. Changing $u$ by at most a constant $C$ changes this segment at time $s$ by at most $Cs/T$, an error that tends to zero as $T\to \infty $. Hence the limits of these segments as $T\to \infty $ depend only on the singularity type of $u$.
Proposition 17.9.2.A6533ED6CC68 Let $u\in \mathrm{PSH}(X,\theta )$ with $u\leq \phi $. For $T>0$ and $0\leq s\leq T$ put
\[ u_ T=\max \{ u,\phi -T\} ,\qquad L_ s^{u,T}=G_{s/T}(\phi ,u_ T). \]
For fixed $s\geq 0$, the potentials $L_ s^{u,T}$ increase with $T\geq s$ and satisfy $\phi -s\leq L_ s^{u,T}\leq \phi $. Their regularized limit
\[ R_ s^ u=\left(\sup _{T>s}L_ s^{u,T}\right)^* \]
is $\theta $-psh. If $v\in \mathrm{PSH}(X,\theta )$ satisfies $v\leq \phi $, and $u\leq v+C$ for a constant $C\geq 0$, then
\[ L_ s^{u,T}\leq L_ s^{v,T}+\frac{Cs}{T},\qquad R_ s^ u\leq R_ s^ v. \]
In particular, if $u\sim v$, then $R_ s^ u=R_ s^ v$ for every $s\geq 0$. Consequently, for an arbitrary $u\in \mathrm{PSH}(X,\theta )$ with $u\preceq \phi $, the potentials $R_ s^{u-c}$ do not depend on the constant $c$ with $u-c\leq \phi $. Denoting them by $R_ s^ u$, we have $R_ s^ u\leq R_ s^ v$ whenever $u\preceq v\preceq \phi $.
Proof.
Since $\phi -T\leq u_ T\leq \phi $, the potential $u_ T$ has bounded difference from $\phi $; in particular $u_ T\in \mathcal E(X,\theta ,\phi )$. By theorem 17.7.8, the Perron segment from $\phi $ to $u_ T$ is a weak geodesic. We regard it as a curve on $[0,T]$, so that its slice at time $s$ is $L_ s^{u,T}$.
Bounds. For $0\leq s\leq T$ put $q_ s=\max \{ u,\phi -s\} $. After the affine change of parameter, the complexification of the curve $(q_ s)_{0\leq s\leq T}$ is $(x,\zeta )\mapsto \max \{ u(x),\phi (x)-T\operatorname{Re}\zeta \} $. This is the maximum of two $p^*\theta $-psh functions, because $\operatorname{Re}\zeta $ is pluriharmonic, and hence it is $p^*\theta $-psh (proposition 5.9.5, applied after adding smooth local potentials of $p^*\theta $, which exist by lemma 5.4.2). Moreover $q_0=\phi $, since $u\leq \phi $, and $q_ T=u_ T$. As $|q_ s-q_ t|\leq |s-t|$ on the finite locus of $\phi $, the curve tends to $\phi $ and $u_ T$ in $L^1$ at the two ends. Thus $(q_ s)$ is a subgeodesic with endpoints $\phi $ and $u_ T$, and by the definition of the Perron segment, $L_ s^{u,T}\geq q_ s\geq \phi -s$. On the other hand, theorem 17.7.8 gives $L_ s^{u,T}\leq (1-s/T)\phi +(s/T)u_ T\leq \phi $.
Monotonicity in $T$. Let $T'>T$. The same subgeodesic, now on $[0,T']$, shows that $L_ T^{u,T'}\geq q_ T=u_ T$. By the last assertion of theorem 17.7.8, the restriction of the segment of length $T'$ to $[0,T]$ is the weak geodesic from $\phi $ to $L_ T^{u,T'}$. Since weak geodesics increase with their endpoints (proposition 17.7.9(1)), this restriction dominates the weak geodesic from $\phi $ to $u_ T$. Hence $L_ s^{u,T'}\geq L_ s^{u,T}$ for $0\leq s\leq T$.
The limit. Fix $s$. The family $(L_ s^{u,T})_{T>s}$ is increasing, bounded above by $\phi $ and bounded below by $\phi -s$. By theorem 5.10.2, applied after adding smooth local potentials of $\theta $, its regularized supremum $R_ s^ u$ is $\theta $-psh, and $\phi -s\leq R_ s^ u\leq \phi $. Since the family is increasing, the supremum may be taken over $T\geq A$ for any $A>s$ without changing $R_ s^ u$. For $s=0$, all the potentials $L_0^{u,T}$ and $R_0^ u$ equal $\phi $.
Comparison. Suppose that $v\leq \phi $ and $u\leq v+C$ with $C\geq 0$. Then
\[ u_ T\leq \max \{ v+C,\phi -T\} \leq v_ T+C. \]
By the monotonicity of weak geodesics in their endpoints and the formula for constant shifts of the endpoints, proposition 17.7.9(1) and (2), we have $G_ t(\phi ,u_ T)\leq G_ t(\phi ,v_ T+C)=G_ t(\phi ,v_ T)+tC$ for $0\leq t\leq 1$. Taking $t=s/T$ gives the bound $L_ s^{u,T}\leq L_ s^{v,T}+Cs/T$. Now fix $s$ and $A>s$. For $T\geq A$ we get $L_ s^{u,T}\leq R_ s^ v+Cs/A$, and taking the regularized supremum over $T\geq A$ gives
\[ R_ s^ u\leq R_ s^ v+\frac{Cs}{A}. \]
Letting $A\to \infty $ yields $R_ s^ u\leq R_ s^ v$.
If $u\sim v$, then $u\leq v+C$ and $v\leq u+C$ for some $C\geq 0$, and the comparison applied in both directions gives $R_ s^ u=R_ s^ v$. If $u\preceq \phi $ is arbitrary, any two constants $c,c'$ with $u-c\leq \phi $ and $u-c'\leq \phi $ give potentials $u-c\sim u-c'$, so $R_ s^{u-c}=R_ s^{u-c'}$. Finally, if $u\preceq v\preceq \phi $, choose constants $c,c'$ with $u-c\leq \phi $ and $v-c'\leq \phi $. Then $u-c\leq v-c'+C$ for some $C\geq 0$, and the comparison gives $R_ s^ u=R_ s^{u-c}\leq R_ s^{v-c'}=R_ s^ v$.
$\square$
Comments
Corrections, questions, and mathematical discussion are welcome. Sign in with GitHub to comment. Comments are public.