OpenMath

Proof source

Least-energy source majorants and the Perron–variational gap

Complete proof text in the unsigned reading edition.

Download sourceBrowse research records

Reading edition of the complete source note. Original labels and explicit equation tags are retained; references use source titles where compiled numbering is unavailable. Current result status is recorded separately in the project register.

Contents of this source
  1. Least-energy source majorants and the Perron–variational gap

Least-energy source majorants and the Perron–variational gap

Retain the full hypotheses (ED1), the original arbitrary background SS, and the unchanged strong cone C\mathscr C. Put d=q+1d=q+1. For dual functionals write ℓ⪯pshλ⟺ℓ(v)≤λ(v)for every v∈C.(PV1)\ell\preceq_{\rm psh}\lambda \quad\Longleftrightarrow\quad \ell(v)\leq\lambda(v)\quad\text{for every }v\in\mathscr C. \tag{PV1} This is the order tested by negative energy potentials. It is neither ordinary measure order nor middle positivity of currents.

Lemma (A one-sided Bregman bound). For a,b∈Ca,b\in\mathscr C, e(a)−e(b)−dJ(b)(a−b)=∑j=0q−1(q−j)∫VΓN^(a[j],b[q−1−j])(a−b),I(a,b)≤e(a)−e(b)−dJ(b)(a−b)≤qI(a,b).(PV2)\begin{split} e(a)-e(b)-dJ(b)(a-b) &=\sum_{j=0}^{q-1}(q-j) \int_V\Gamma_{\widehat N(a^{[j]},b^{[q-1-j]})}(a-b),\\ I(a,b)&\leq e(a)-e(b)-dJ(b)(a-b)\leq qI(a,b). \end{split} \tag{PV2} Every displayed energy is finite and nonnegative.

Status: DIRECT_COROLLARY of the frozen mixed variation and the coefficient computation (JD9). The bounded Bregman calculation was already present in R3.

Proof. Apply polynomial Taylor expansion to e((1−t)b+ta)e((1-t)b+ta) and retain its first derivative rather than canceling it by averaging. The same beta integral as in (JD9) gives the coefficients q−jq-j. Their unweighted sum is I(a,b)I(a,b) by the frozen discrepancy identity, proving both inequalities. The completed mixed energy calculus justifies all global integrals before this algebraic computation is made. ◻

Theorem (The unique least-energy curvature majorant). Let ℓ∈DS,s∗\ell\in\mathbb D_{S,\mathrm s}^* be nonnegative on C\mathscr C, and put u=Π(ℓ)u=\Pi(\ell). Then ℓ⪯pshJ(u),∥J(u)∥∗=∥ℓ∥∗.(PV3)\ell\preceq_{\rm psh}J(u),\qquad \|J(u)\|_*=\|\ell\|_*. \tag{PV3} Among all v∈Cv\in\mathscr C satisfying ℓ⪯pshJ(v)\ell\preceq_{\rm psh}J(v), the potential uu is the unique one of least energy. Every such vv satisfies 0≤I(u,v)≤q(e(v)−e(u)).(PV4)0\leq I(u,v)\leq q\bigl(e(v)-e(u)\bigr). \tag{PV4} Consequently any row of these majorants with e(vj)→e(u)e(v_j)\to e(u) converges strongly to uu on the full original current calculus.

The map ℓ↦J(Π(ℓ))\ell\mapsto J(\Pi(\ell)) is an idempotent, positively homogeneous, dual-norm-preserving retraction on this positive dual cone. Its norm continuity is (NI11). The phrase least energy does not assert a least element for the order (PV1), and norm preservation does not mean preservation of the total mass of a measure.

Status: NEW_PROVED for the all-degree majorant characterization and its quantitative energy gap. The normal decomposition is the preceding proved R12 input.

Proof. The normal residual in (NI7) is nonpositive on C\mathscr C, which proves the order assertion in (PV3). Positivity of ℓ\ell on this cone gives p(ℓ)=∥ℓ∥∗p(\ell)=\|\ell\|_*, so (NI9) and (ED10) prove its norm equality.

For a competitor vv, the majorant condition tested at uu gives B(u,v[q])=J(v)(u)≥ℓ(u)=e(u)B(u,v^{[q]})=J(v)(u)\geq\ell(u)=e(u). Use (PV2) with (a,b)=(u,v)(a,b)=(u,v) to obtain I(u,v)≤e(u)−e(v)−d(B(u,v[q])−e(v))=e(u)+qe(v)−dB(u,v[q])≤q(e(v)−e(u)).\begin{split} I(u,v)&\leq e(u)-e(v)-d\bigl(B(u,v^{[q]})-e(v)\bigr)\\ &=e(u)+q e(v)-d B(u,v^{[q]}) \leq q(e(v)-e(u)). \end{split} This proves (PV4), minimality, and uniqueness by relative discrepancy separation. It also proves the approximation assertion. The retraction, homogeneity and continuity follow from the corresponding exact properties of (NI7)–(NI11). No positivity of the signed residual measure is inferred from its potential-test order. ◻

For a literal positive Radon datum μ\mu, (PV3) says ∫V(−v) dμ≤∫V(−v) dμ^u(v∈C),∫V(−u) dμ=e(u).(PV5)\int_V(-v)\,d\mu\leq\int_V(-v)\,d\widehat\mu_u \quad(v\in\mathscr C),\qquad \int_V(-u)\,d\mu=e(u). \tag{PV5} One may call this a nonlinear energy sweeping of the datum, provided (PV1), the specified source class, and the absence of a total-mass claim remain part of the definition.

Lemma (The variational source cannot have a positive excess). Let μ\mu be a positive original Radon measure whose literal functional belongs to DS,s∗\mathbb D_{S,\mathrm s}^*, and put u=Π(ℓμ)u=\Pi(\ell_\mu). If μ^u≥μ\widehat\mu_u\geq\mu as original measures, then μ^u=μ\widehat\mu_u=\mu.

Status: DIRECT_COROLLARY of radial variational orthogonality and the frozen completed weak maximum inequality.

Proof. For every a∈Ca\in\mathscr C, apply the completed weak maximum inequality to aa and the zero potential, assigning the equality set to aa. Since max⁡(a,0)=0\max(a,0)=0, it gives 1{a=0}μ^a=0.(PV6)\mathbf1_{\{a=0\}}\widehat\mu_a=0. \tag{PV6} All involved measures also ignore the relative pole set. Now let τ=μ^u−μ≥0\tau=\widehat\mu_u-\mu\geq0. It is dominated by μ^u\widehat\mu_u, so it ignores both sets just mentioned. Radial stationarity gives the finite equality ∫V(−u) dτ=e(u)−ℓμ(u)=0.\int_V(-u)\,d\tau =e(u)-\ell_\mu(u)=0. The integrand is strictly positive and finite on the remaining full carrier {−∞<u<0}\{-\infty<u<0\}. Its increasing level exhaustion therefore makes τ\tau vanish there too. Thus τ=0\tau=0 as a Radon measure. No subtraction of infinite unweighted total masses has occurred. ◻

Theorem (An exact scalar gap between Perron and variational potentials). Let μ\mu be a positive original Radon measure with a literal strong-continuous energy functional. Suppose the completed subsolution family Sμ={v∈C:μ^v≥μ}(PV7)\mathscr S_\mu =\{v\in\mathscr C:\widehat\mu_v\geq\mu\} \tag{PV7} is nonempty. Let HμH_\mu be its greatest relative member, constructed by the frozen R11 full Perron theorem (SE1)–(SE8), and let uμ=Π(ℓμ)u_\mu=\Pi(\ell_\mu). Define ΔS(μ)=e(Hμ)−e(uμ).(PV8)\Delta_S(\mu)=e(H_\mu)-e(u_\mu). \tag{PV8} Then ΔS(μ)≥0,I(Hμ,uμ)≤qΔS(μ).(PV9)\Delta_S(\mu)\geq0,\qquad I(H_\mu,u_\mu)\leq q\Delta_S(\mu). \tag{PV9} Moreover the following conditions are equivalent:

  1. ΔS(μ)=0\Delta_S(\mu)=0;

  2. Hμ=uμH_\mu=u_\mu as relative completed classes;

  3. μ^uμ=μ\widehat\mu_{u_\mu}=\mu on every original test;

  4. there exists v∈Cv\in\mathscr C with μ^v=μ\widehat\mu_v=\mu;

  5. there exists a row vj∈Sμv_j\in\mathscr S_\mu with e(vj)→e(uμ)e(v_j)\to e(u_\mu).

If they hold, this source potential is unique. The family is nonempty, in particular, whenever μ≤Aμψ\mu\leq A\mu_\psi for an actual first-energy ψ\psi and finite A≥0A\geq0.

Thus the general subsolution Perron construction and the general variational construction are linked by a single nonnegative scalar gap. Its universal vanishing is not claimed: the compact-source examples (ML22)–(ML28) have ΔS(μ)>0\Delta_S(\mu)>0 in every 2≤q<r2\leq q<r.

Status: NEW_PROVED for this all-degree criterion and quantitative comparison. The two constructions themselves are explicitly frozen or previously proved inputs. Unqualified vanishing of the gap is FALSE, not a theorem hidden in the word Perron.

Proof. The R11 theorem applies to the nonempty family in (PV7), keeps its measure lower bound, and makes its relative envelope an actual member of C\mathscr C. Thus HμH_\mu is the greatest member of that family. Every measure majorant is also a majorant for the potential-test order (PV1). Applying (PV4) proves (PV9).

If the gap is zero, (PV9) gives equality of the two relative potentials. The common potential is a subsolution, so Lemma “The variational source cannot have a positive excess” makes its source exactly μ\mu. Conversely, a source potential vv has zero normal residual and hence is the unique variational potential uμu_\mu. It belongs to (PV7), so Hμ⪰uμH_\mu\succeq u_\mu. Energy monotonicity gives e(Hμ)≤e(uμ)e(H_\mu)\leq e(u_\mu), and (PV9) gives equality. This proves the first four equivalences and uniqueness.

Every v∈Sμv\in\mathscr S_\mu is below HμH_\mu, so e(v)≥e(Hμ)e(v)\geq e(H_\mu). A row in condition (5) therefore forces the gap to be zero. If the gap is zero, the constant row vj=uμv_j=u_\mu supplies condition (5). Alternatively (PV4) directly shows that any such row converges strongly to uμu_\mu and its source inequalities pass to all original compact tests.

For the printed dominated datum, A1/qψA^{1/q}\psi is a subsolution (the zero case is immediate), and its literal datum functional satisfies the strong-dual hypothesis by mixed Holder and (EV1). The nonlinear compact-source examples satisfy these same assumptions but have no source in C\mathscr C. The equivalence therefore makes their gap strictly positive. None of these arguments needs a common actual lower barrier for the whole subsolution family. ◻

Proposition (A quantitative positive-test lower bound for the gap). Under the preceding theorem put u=uμu=u_\mu, H=HμH=H_\mu, R=r(H)R=r(H), and define ε+(μ,u)=sup⁡0≤ϕ∈Ac0,0(V)∥ϕ∥ed≤1(∫Vϕ d(μ−μ^u))+.(PV10)\varepsilon_+(\mu,u)= \sup_{\substack{0\leq\phi\in A_c^{0,0}(V)\\ \|\phi\|_{\rm ed}\leq1}} \left(\int_V\phi\,d(\mu-\widehat\mu_u)\right)_+. \tag{PV10} This is finite. If R>0R>0, then ΔS(μ)≥ε+(μ,u)2q2Rq−1.(PV11)\Delta_S(\mu)\geq \frac{\varepsilon_+(\mu,u)^2}{q^2R^{q-1}}. \tag{PV11} If R=0R=0, the datum and all quantities are zero. Also ε+(μ,u)=0\varepsilon_+(\mu,u)=0 if and only if the source equation holds. Every positive test in (PV10) is from the full original sheaf, not a selected moment family.

Status: DIRECT_COROLLARY of the new gap theorem and the proved source-dual estimate.

Proof. All original compact smooth scalars lie in DS\mathbb D_S by (HC14)–(HC15). Finiteness follows by the dual norm of ℓμ−J(u)\ell_\mu-J(u). Since μ^H≥μ\widehat\mu_H\geq\mu, each nonnegative test in (PV10) gives ∫ϕ d(μ−μ^u)≤∫ϕ d(μ^H−μ^u)=(J(u)−J(H))(ϕ).\int\phi\,d(\mu-\widehat\mu_u) \leq\int\phi\,d(\widehat\mu_H-\widehat\mu_u) =(J(u)-J(H))(\phi). Consequently ε+≤∥J(H)−J(u)∥∗\varepsilon_+\leq\|J(H)-J(u)\|_*. The energy minimum gives r(u)≤Rr(u)\leq R. Equations (NI2) and (PV9) now give ε+≤qR(q−1)/2ΔS(μ)1/2\varepsilon_+\leq qR^{(q-1)/2}\Delta_S(\mu)^{1/2}, which is (PV11). When R=0R=0, the subsolution H=0H=0 already forces μ=0\mu=0.

Vanishing in (PV10), together with original positive scalar test detection, says μ^u≥μ\widehat\mu_u\geq\mu. Lemma “The variational source cannot have a positive excess” upgrades this to equality. The reverse implication is immediate. This is a full original-measure statement, not an ordinary-to-delta-test promotion. ◻

Source text and edition identity

This reading copy omits personal attribution and places existing mathematical question marks over their relation symbols. The mathematical content and source status are retained.

Reading-copy SHA-256 76769b5072af4d886ea97abd1b56e63084d3cdc858dd0635ed3cafd5bd75a702

Frozen-original SHA-256 76769b5072af4d886ea97abd1b56e63084d3cdc858dd0635ed3cafd5bd75a702

\section{Least-energy source majorants and the Perron--variational gap}
\label{sec:r12-perron-variational-gap}

Retain the full hypotheses (ED1), the original arbitrary background
$S$, and the unchanged strong cone $\mathscr C$. Put $d=q+1$.
For dual functionals write
\[
       \ell\preceq_{\rm psh}\lambda
       \quad\Longleftrightarrow\quad
       \ell(v)\leq\lambda(v)\quad\text{for every }v\in\mathscr C.
 \tag{PV1}
\]
This is the order tested by negative energy potentials. It is
neither ordinary measure order nor middle positivity of currents.

\begin{lemma}[A one-sided Bregman bound]
\label{lem:r12-one-sided-Bregman}
For $a,b\in\mathscr C$,
\[
 \begin{split}
 e(a)-e(b)-dJ(b)(a-b)
  &=\sum_{j=0}^{q-1}(q-j)
       \int_V\Gamma_{\widehat N(a^{[j]},b^{[q-1-j]})}(a-b),\\
 I(a,b)&\leq e(a)-e(b)-dJ(b)(a-b)\leq qI(a,b).
 \end{split}
 \tag{PV2}
\]
Every displayed energy is finite and nonnegative.
\end{lemma}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of
the frozen mixed variation and the coefficient computation (JD9).
The bounded Bregman calculation was already present in R3.

\begin{proof}
Apply polynomial Taylor expansion to $e((1-t)b+ta)$ and
retain its first derivative rather than canceling it by
averaging. The same beta integral as in (JD9) gives the
coefficients $q-j$. Their unweighted sum is $I(a,b)$ by
the frozen discrepancy identity, proving both inequalities.
The completed mixed energy calculus justifies all global
integrals before this algebraic computation is made.
\end{proof}

\begin{theorem}[The unique least-energy curvature majorant]
\label{thm:r12-least-energy-curvature-majorant}
Let $\ell\in\mathbb D_{S,\mathrm s}^*$ be nonnegative on
$\mathscr C$, and put $u=\Pi(\ell)$. Then
\[
       \ell\preceq_{\rm psh}J(u),\qquad
       \|J(u)\|_*=\|\ell\|_*.
 \tag{PV3}
\]
Among all $v\in\mathscr C$ satisfying
$\ell\preceq_{\rm psh}J(v)$, the potential $u$ is the
unique one of least energy. Every such $v$ satisfies
\[
               0\leq I(u,v)\leq q\bigl(e(v)-e(u)\bigr).
 \tag{PV4}
\]
Consequently any row of these majorants with
$e(v_j)\to e(u)$ converges strongly to $u$ on the full
original current calculus.

The map $\ell\mapsto J(\Pi(\ell))$ is an idempotent,
positively homogeneous, dual-norm-preserving retraction on
this positive dual cone. Its norm continuity is (NI11).
The phrase \emph{least energy} does not assert a least element
for the order (PV1), and norm preservation does not mean
preservation of the total mass of a measure.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} for the
all-degree majorant characterization and its quantitative
energy gap. The normal decomposition is the preceding
proved R12 input.

\begin{proof}
The normal residual in (NI7) is nonpositive on $\mathscr C$,
which proves the order assertion in (PV3). Positivity of
$\ell$ on this cone gives $p(\ell)=\|\ell\|_*$, so (NI9)
and (ED10) prove its norm equality.

For a competitor $v$, the majorant condition tested at $u$
gives $B(u,v^{[q]})=J(v)(u)\geq\ell(u)=e(u)$.
Use (PV2) with $(a,b)=(u,v)$ to obtain
\[
 \begin{split}
 I(u,v)&\leq e(u)-e(v)-d\bigl(B(u,v^{[q]})-e(v)\bigr)\\
       &=e(u)+q e(v)-d B(u,v^{[q]})
                         \leq q(e(v)-e(u)).
 \end{split}
\]
This proves (PV4), minimality, and uniqueness by relative
discrepancy separation. It also proves the approximation
assertion. The retraction, homogeneity and continuity follow
from the corresponding exact properties of (NI7)--(NI11).
No positivity of the signed residual measure is inferred
from its potential-test order.
\end{proof}

For a literal positive Radon datum $\mu$, (PV3) says
\[
      \int_V(-v)\,d\mu\leq\int_V(-v)\,d\widehat\mu_u
               \quad(v\in\mathscr C),\qquad
      \int_V(-u)\,d\mu=e(u).
 \tag{PV5}
\]
One may call this a nonlinear energy sweeping of the datum,
provided (PV1), the specified source class, and the absence
of a total-mass claim remain part of the definition.

\begin{lemma}[The variational source cannot have a positive excess]
\label{lem:r12-variational-positive-excess}
Let $\mu$ be a positive original Radon measure whose literal
functional belongs to $\mathbb D_{S,\mathrm s}^*$, and
put $u=\Pi(\ell_\mu)$. If $\widehat\mu_u\geq\mu$
as original measures, then $\widehat\mu_u=\mu$.
\end{lemma}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of
radial variational orthogonality and the frozen completed
weak maximum inequality.

\begin{proof}
For every $a\in\mathscr C$, apply the completed weak
maximum inequality to $a$ and the zero potential, assigning
the equality set to $a$. Since $\max(a,0)=0$, it gives
\[
                   \mathbf1_{\{a=0\}}\widehat\mu_a=0.
 \tag{PV6}
\]
All involved measures also ignore the relative pole set.
Now let $\tau=\widehat\mu_u-\mu\geq0$. It is dominated
by $\widehat\mu_u$, so it ignores both sets just mentioned.
Radial stationarity gives the finite equality
\[
                       \int_V(-u)\,d\tau
                                  =e(u)-\ell_\mu(u)=0.
\]
The integrand is strictly positive and finite on the
remaining full carrier $\{-\infty<u<0\}$. Its increasing
level exhaustion therefore makes $\tau$ vanish there too.
Thus $\tau=0$ as a Radon measure. No subtraction of
infinite unweighted total masses has occurred.
\end{proof}

\begin{theorem}[An exact scalar gap between Perron and variational potentials]
\label{thm:r12-perron-variational-gap}
Let $\mu$ be a positive original Radon measure with a
literal strong-continuous energy functional. Suppose the
completed subsolution family
\[
             \mathscr S_\mu
               =\{v\in\mathscr C:\widehat\mu_v\geq\mu\}
 \tag{PV7}
\]
is nonempty. Let $H_\mu$ be its greatest relative member,
constructed by the frozen R11 full Perron theorem (SE1)--(SE8),
and let $u_\mu=\Pi(\ell_\mu)$. Define
\[
                  \Delta_S(\mu)=e(H_\mu)-e(u_\mu).
 \tag{PV8}
\]
Then
\[
              \Delta_S(\mu)\geq0,\qquad
              I(H_\mu,u_\mu)\leq q\Delta_S(\mu).
 \tag{PV9}
\]
Moreover the following conditions are equivalent:
\begin{enumerate}
\item $\Delta_S(\mu)=0$;
\item $H_\mu=u_\mu$ as relative completed classes;
\item $\widehat\mu_{u_\mu}=\mu$ on every original test;
\item there exists $v\in\mathscr C$ with $\widehat\mu_v=\mu$;
\item there exists a row $v_j\in\mathscr S_\mu$ with
      $e(v_j)\to e(u_\mu)$.
\end{enumerate}
If they hold, this source potential is unique. The family
is nonempty, in particular, whenever $\mu\leq A\mu_\psi$
for an actual first-energy $\psi$ and finite $A\geq0$.

Thus the general subsolution Perron construction and the
general variational construction are linked by a single
nonnegative scalar gap. Its universal vanishing is not
claimed: the compact-source examples (ML22)--(ML28) have
$\Delta_S(\mu)>0$ in every $2\leq q<r$.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}} for this
all-degree criterion and quantitative comparison. The two
constructions themselves are explicitly frozen or previously
proved inputs. Unqualified vanishing of the gap is
\texttt{FALSE}, not a theorem hidden in the word Perron.

\begin{proof}
The R11 theorem applies to the nonempty family in (PV7),
keeps its measure lower bound, and makes its relative
envelope an actual member of $\mathscr C$. Thus $H_\mu$
is the greatest member of that family. Every measure
majorant is also a majorant for the potential-test order
(PV1). Applying (PV4) proves (PV9).

If the gap is zero, (PV9) gives equality of the two
relative potentials. The common potential is a subsolution,
so Lemma~\ref{lem:r12-variational-positive-excess} makes
its source exactly $\mu$. Conversely, a source potential
$v$ has zero normal residual and hence is the unique
variational potential $u_\mu$. It belongs to (PV7), so
$H_\mu\succeq u_\mu$. Energy monotonicity gives
$e(H_\mu)\leq e(u_\mu)$, and (PV9) gives equality.
This proves the first four equivalences and uniqueness.

Every $v\in\mathscr S_\mu$ is below $H_\mu$, so
$e(v)\geq e(H_\mu)$. A row in condition (5) therefore
forces the gap to be zero. If the gap is zero, the
constant row $v_j=u_\mu$ supplies condition (5).
Alternatively (PV4) directly shows that any such row
converges strongly to $u_\mu$ and its source inequalities
pass to all original compact tests.

For the printed dominated datum, $A^{1/q}\psi$ is a
subsolution (the zero case is immediate), and its literal
datum functional satisfies the strong-dual hypothesis by
mixed Holder and (EV1). The nonlinear compact-source
examples satisfy these same assumptions but have no source
in $\mathscr C$. The equivalence therefore makes their
gap strictly positive. None of these arguments needs a
common actual lower barrier for the whole subsolution family.
\end{proof}

\begin{proposition}[A quantitative positive-test lower bound for the gap]
\label{prop:r12-positive-test-gap-bound}
Under the preceding theorem put $u=u_\mu$, $H=H_\mu$,
$R=r(H)$, and define
\[
 \varepsilon_+(\mu,u)=
   \sup_{\substack{0\leq\phi\in A_c^{0,0}(V)\\
                   \|\phi\|_{\rm ed}\leq1}}
                   \left(\int_V\phi\,d(\mu-\widehat\mu_u)\right)_+.
 \tag{PV10}
\]
This is finite. If $R>0$, then
\[
       \Delta_S(\mu)\geq
          \frac{\varepsilon_+(\mu,u)^2}{q^2R^{q-1}}.
 \tag{PV11}
\]
If $R=0$, the datum and all quantities are zero. Also
$\varepsilon_+(\mu,u)=0$ if and only if the source equation
holds. Every positive test in (PV10) is from the full
original sheaf, not a selected moment family.
\end{proposition}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of
the new gap theorem and the proved source-dual estimate.

\begin{proof}
All original compact smooth scalars lie in $\mathbb D_S$
by (HC14)--(HC15). Finiteness follows by the dual norm of
$\ell_\mu-J(u)$. Since $\widehat\mu_H\geq\mu$, each
nonnegative test in (PV10) gives
\[
 \int\phi\,d(\mu-\widehat\mu_u)
       \leq\int\phi\,d(\widehat\mu_H-\widehat\mu_u)
       =(J(u)-J(H))(\phi).
\]
Consequently $\varepsilon_+\leq\|J(H)-J(u)\|_*$.
The energy minimum gives $r(u)\leq R$. Equations (NI2)
and (PV9) now give
$\varepsilon_+\leq qR^{(q-1)/2}\Delta_S(\mu)^{1/2}$,
which is (PV11). When $R=0$, the subsolution $H=0$
already forces $\mu=0$.

Vanishing in (PV10), together with original positive
scalar test detection, says $\widehat\mu_u\geq\mu$.
Lemma~\ref{lem:r12-variational-positive-excess} upgrades
this to equality. The reverse implication is immediate.
This is a full original-measure statement, not an
ordinary-to-delta-test promotion.
\end{proof}