diff d6cd9da..29ac5cc spacetime_penrose_enclosing_area_v2.txt
diff --git a/spacetime_penrose_enclosing_area_v2.txt b/spacetime_penrose_enclosing_area_v2.txt
index 7ae7158..df5aa45 100644
--- a/spacetime_penrose_enclosing_area_v2.txt
+++ b/spacetime_penrose_enclosing_area_v2.txt
@@ -16,7 +16,7 @@ whose apparent horizon area is too large for the mass bound. Carrasco
 and Mars found a further obstruction for generalized apparent
 horizons, already among data whose outer generalized horizon is
 area-minimizing. Both examples establish that mass-area theorems must
-fix more than a trapping condition: they must also fix an enclosing
+fix more than a trapping condition, and must also fix an enclosing
 class. The choice adopted below is the infimum of the full area of an
 enclosing cut, meaning the entire intrinsic boundary of a closed outer
 region that contains the whole distant end; every component counts,
@@ -199,8 +199,7 @@ value. Hence $A_{min}(S) = \omega_{n-1} (2m)^{(n-1)/(n-2)}$ and the
 inequality is an equality for this family.
 
 The time-symmetric example above is the simplest sharp family. Equality
-in the general case requires a separate variational argument, which is
-not the subject of the present text.
+in the general case requires a separate variational argument.
 
 The statement carries the hypothesis $E > |P|_\delta$. That
 hypothesis is redundant once $A_{min}(S) > 0$.
@@ -569,3 +568,1619 @@ interpolated metric, the original metric, and the Euclidean metric
 differ by $O(L^{-q})$ on the annulus and that the cutoff is a convex
 combination, so transporting a unit vector between the three balls
 changes the pairing with the interpolated $J$ by $O(L^{-2q})$.
+
+The annular join leaves a small deficit and keeps the slice carried by a
+boost. Two further moves finish the end preparation: bend the slice to a
+static one, which removes the boost and hence the momentum; and repair
+the residual deficit with a positive conformal factor, which keeps every
+cut area at least as large.
+
+Bending. Use static time $b$ and isotropic coordinates $y$ of the model.
+The slice $T=0$ has equation $b = v y^1$. Choose a monotone cutoff
+$\zeta$ with $|d\zeta/d(\log |y|)| \le \eta_v$ and $v(1+\eta_v) < 1$.
+Replace the graph by
+$$
+b = v y^1 \zeta(\log(|y|/L)).
+$$
+Its Euclidean slope $\sup |db|_\delta \le v(1+\eta_v)$ is less than one.
+Because $F, G = 1 + O(r^{-(n-2)})$ on the bend annulus
+$4\gamma L \le |y| \le C_v L$, the graph is uniformly spacelike there for
+all large $L$. Its intrinsic data are vacuum everywhere on the bend,
+by vacuum Gauss-Codazzi, and agree with the plane $b=0$ outside
+$|y| = C_v L$. Throughout the transition
+$$
+c_v \delta \le \bar g_L \le C_v' \delta,
+\qquad
+|D \bar g_L|_\delta + |\bar K_L|_\delta \le C_v'/r.
+$$
+The tensor $\bar K_L$ is compactly supported, and the only possible DEC
+deficit is on the original gluing annulus $L \le r \le 4L$.
+
+Comparison of cuts. The bent metric $\bar g_L$ need not approach the
+original metric uniformly on the whole end. Choose $\lambda$ with
+$0 < \lambda^2 < c_v/2$. Let $h_L$ be smooth, strictly increasing,
+equal to $r$ for $r \le \sqrt L$, with $\lambda \le h_L' \le 1$ and
+$h_L'(r) = \lambda$ for $r \ge 2\sqrt L$. Set
+$\Phi_L(r\theta) = h_L(r)\theta$ on the end and $\Phi_L = \mathrm{id}$
+on the compact core.
+
+Both derivative eigenvalues of $\Phi_L$, meaning $h_L'$ radially and
+$h_L/r$ tangentially, are at most one everywhere, and at most
+$\lambda + O(L^{-1/2})$ for $r \ge L$. Hence $\Phi_L$ is a proper
+diffeomorphism of the exterior onto itself, fixes the boundary, and
+preserves the end. On $r \le \sqrt L$ it is the identity. On
+$\sqrt L \le r \le L$ both metrics are $\delta + o(1)$ uniformly, and
+$\Phi_L$ has derivative norms at most one, so $\Phi_L^* g \le (1+o(1))\bar g_L$.
+On $r \ge L$, $\bar g_L \ge c_v\delta$ and the derivative norms are at
+most $\lambda + O(L^{-1/2})$, giving $\Phi_L^* g \le (1+o(1))\bar g_L$
+as well. Thus
+$$
+\Phi_L^* g \le (1+\epsilon_L)\bar g_L,
+\qquad \epsilon_L \to 0,
+$$
+for some $\epsilon_L$.
+
+Any full cut $\Gamma$ of $\bar g_L$ therefore transfers to a full cut of
+the original via the proper diffeomorphism, with
+$$
+\mathrm{Area}_g\bigl(\Phi_L(\Gamma)\bigr)
+\le (1+\epsilon_L)^{(n-1)/2}\mathrm{Area}_{\bar g_L}(\Gamma).
+$$
+Taking the infimum gives
+$A_{min}(g) \le (1+\epsilon_L)^{(n-1)/2} A_{min}(\bar g_L)$.
+
+Radial repair. Let $\bar f_L \ge 0$ be a smooth density on the gluing
+annulus, dominating the DEC deficit there:
+$\mu_L - |J_L|_{\bar g_L} \ge - \bar f_L$. Choose $\epsilon_L \downarrow 0$
+so that $d_L / \epsilon_L \to 0$. Set
+$$
+U_L = a_L^{-1}(1+\epsilon_L f_L),
+$$
+where $a_L = 1 + m/(2|y|^{n-2})$ is the Schwarzschild conformal factor
+and $f_L$ is a smooth nonnegative function, equal to zero on $r \le L$,
+radial on the transition, supported near the annulus, and constant on
+a distance past it, that satisfies
+$$
+-\Delta_{\bar g_L} f_L - |\bar K_L|_{\bar g_L} |df_L|_{\bar g_L}
+\ge \epsilon_L^{-1} \bar f_L
+$$
+on $L < r < 4L$. Such an $f_L$ exists because $\Delta_{\bar g_L}$ is
+uniformly elliptic on the transition and $\bar f_L / \epsilon_L^{-1}$
+is bounded by $d_L / \epsilon_L \to 0$. Extend $f_L$ by solving the
+scalar ODE $f_L' = -q_L$ in the far tail, where $q_L$ is a nonnegative
+smooth function that starts at $0$ on $r=4L$, increases to the fixed
+value $q_*$, and stays at $q_*$ afterwards; then $f_L(s) = q_* p^{-1} s^{-p}
++ O(s^{-2p})$ with $p=n-2$ on the far tail.
+
+The conformal change $g_L^* = U_L^{4/p}\bar g_L$,
+$K_L^* = U_L^{2/p}\bar K_L$ satisfies DEC on $\Omega$, is unchanged
+on the original core (since $U_L = a_L^{-1}$ there because $f_L = 0$),
+and is Schwarzschild on the far tail. In terms of the rest-frame
+coordinates $y$, the far tail of $g_L^*$ equals
+$(1+\epsilon_L f_L(y))^{4/p}\delta$, and its ADM energy is
+$$
+E_L^* = M + \frac{2\epsilon_L q_*}{p} = M + o(1),
+\qquad P_L^* = 0.
+$$
+Compact support of $K_L^*$ is preserved, integrability is preserved,
+and the boundary expansion retains its nonpositive sign because $U_L$
+is a positive constant near $S$. The pointwise metric monotonicity
+$g_L^* \ge \bar g_L$ gives $A_{min}(\bar g_L) \ge A_{min}(g_L^*)$ after
+taking infima.
+
+Combining the three estimates, every full cut of the deformed data has
+an original full cut of at most $(1+\epsilon_L)^{(n-1)/2}$ times the
+area, so
+$$
+A_{min,g}(S)
+\le (1+\epsilon_L)^{(n-1)/2} A_{min,g_L^*}(S),
+$$
+and $E_L^* = m+o(1)$, $P_L^*=0$.
+
+A fixed rest end is not yet a prepared datum: the dominant energy condition
+and the boundary trapping are still weak. A small drift-Poisson equation
+makes both strict, without changing the momentum and without shrinking any
+cut area. The scheme is close to Jaracz's conformal strictification; the
+extra gradient term in the equation is what controls the boundary.
+
+Strictification lemma. Let $\Omega$ be a smooth complete one-ended
+exterior with nonempty compact boundary $S$. Let $(g,K)$ be smooth, with
+$g-\delta=O_6(r^{-q})$ and $K=O_5(r^{-1-q})$ for some $q>(n-2)/2$. Fix
+$0<\beta<\min(1,q)$ and a smooth nonnegative $B \ge |K|_g$ with
+$B=O_5(r^{-1-q})$. The scalar problem
+$$
+-\Delta_g \phi = B\sqrt{|d\phi|_g^2+\varrho^{-2k}} + \varrho^{-n-\beta},
+\qquad \partial_\nu \phi = -1 \text{ on } S,
+\qquad \phi \to 0,
+$$
+with $\varrho$ a smooth positive radius extension and $k=n-1$, has a
+smooth positive solution with
+$$
+\phi = O_6(r^{2-n}),
+\qquad
+\frac{-\Delta_g \phi}{r^{-n-\beta}} \to 1,
+$$
+and finite negative Euclidean flux
+$$
+\mathfrak f_\phi = \lim_{R \to \infty} \int_{S_R}\partial_{\nu_g}\phi\,dA_g
+    = -\mathrm{Area}_g(S)
+       - \int_\Omega \bigl(B\sqrt{|d\phi|^2+\varrho^{-2k}}
+                        +\varrho^{-n-\beta}\bigr)\,dV_g < 0.
+$$
+If $K$ is compactly supported, so is $B$, and if in addition
+$g-\delta = O_d(r^{2-n})$ for all $d$, then $\phi = O_d(r^{2-n})$ for
+all $d$. When $(g,K)$ also satisfy DEC and $\theta_+(S) \le 0$, then for
+all small $\delta > 0$ the data
+$$
+g_\delta = e^{2\delta\phi}g, \qquad K_\delta = e^{\delta\phi}K
+$$
+satisfy
+$$
+\mu_\delta - |J_\delta|_{g_\delta}
+\ge c_\delta\varrho^{-n-\beta},
+\qquad
+\theta_{+,\delta} < 0,
+\qquad
+E_\delta = E - \frac{\delta}{\omega_{n-1}}\mathfrak f_\phi,
+\qquad
+P_\delta = P,
+$$
+with $g_\delta \ge g$ and $g_\delta/g \to 1$ uniformly as $\delta\downarrow0$.
+
+Proof. Write $a=\varrho^{-k}$ and $b=\varrho^{-n-\beta}$. The regularizer
+$a^2$ makes the right side smooth, with first gradient bounded by $B$.
+Truncate at $S_R$ and impose $\phi=0$ there. At parameter $t=0$ the
+mixed Laplacian is invertible: coercivity on $H^1$ with zero outer
+trace, boundedness of the Neumann datum by the trace theorem, and local
+regularity after Lax-Milgram. At any solution the linearization is a
+Laplacian with bounded drift that vanishes nowhere on the inner boundary,
+so the strong and Hopf maximum principles rule out a nonzero homogeneous
+kernel. Index zero follows from the Fredholm perturbation, so the
+linearization is an isomorphism, and the implicit function theorem
+gives openness of the continuation set.
+
+Uniform bounds and closedness. All solutions are nonnegative: an
+interior negative minimum contradicts $-\Delta\phi>0$, a boundary
+minimum contradicts the outward derivative $-\partial_\nu\phi=1$, and
+the outer value is zero. For $r_0$ large and
+$0<\alpha<\min(\beta,q)$ set $W(r)=r^{-p}(1-r^{-\alpha})$, $p=n-2$.
+On $r \ge r_0$, $-\Delta_g W - B|dW|_g \ge c r^{-n-\alpha}$, since
+$-\Delta_\delta W = \alpha(p+\alpha)r^{-n-\alpha}$ plus metric error
+$O(r^{-n-q})$, and $B|dW|=O(r^{-n-q})$. A sufficiently large multiple
+of $W$, chosen to dominate $\phi$ on $S_{r_0}$, is then a supersolution
+that vanishes on $S_R$. So $0 \le \phi \le C(1+A)r^{-p}$ uniformly in
+$R$ and $t$, with $A = \sup_{S_{r_0}}\phi$.
+
+Core bound. If the suprema $M_R$ on the fixed core were unbounded,
+local $W^{2,a_0}$ estimates for any $a_0>n$ would give uniform local
+bounds after dividing by $M_R$, and a subsequence would converge to a
+nonnegative limit $v$ with $\max v = 1$, homogeneous Dirichlet data,
+and $-\Delta_g v = t_* B |dv|_g$ weakly. The strong and Hopf principles
+rule this out, as does the outer decay. So the suprema are uniformly
+bounded and closedness of continuation follows.
+
+Exhaustion and uniqueness. Local $W^{2,a_0}$ bounds upgrade to
+$C^{1,\gamma}$, then Schauder gives $C^{2,\gamma}$ and differentiation
+gives $C^\infty$. Letting $R\to\infty$ gives a smooth solution with the
+stated decay. Uniqueness of decaying solutions: the difference of two
+is a homogeneous equation with bounded drift, and the same maximum and
+Hopf arguments exclude nonzero extrema. The positive regularizer is
+what closes the case $\nabla v = 0$.
+
+Flux formula. Integrate the equation on $\Omega_R$ with inner outward
+normal $-\nu$:
+$$
+\int_{S_R}\partial_{\nu_g}\phi\,dA_g
+= -\mathrm{Area}_g(S)
+  - \int_{\Omega_R}\bigl(B\sqrt{|d\phi|^2+a^2}+b\bigr)\,dV_g,
+$$
+which proves the claimed finite negative flux.
+
+Strict inequalities. On the end the quotient
+$|d\phi|^2/(r^{-n-\beta})$ is $O(r^{-(n-2-\beta)})=O(1)$, and on the
+core it is bounded, so $|d\phi|^2 \le Cb$ globally. Hence
+$-\Delta_g\phi + K(\nabla\phi,v) \ge b$ for every unit-ball $v$.
+Substituting $s=\delta\phi$ in the conformal identity:
+$$
+e^{2\delta\phi}\bigl(\mu_\delta+J_\delta(e^{-\delta\phi}v)\bigr)
+\ge \mu+J(v) + k\delta b - \tfrac{kp}{2}\delta^2|d\phi|^2.
+$$
+For small $\delta$, the negative square term is at most half the positive
+$k\delta b$ term. This gives the strict DEC margin. On $S$,
+$e^{\delta\phi}\theta_{+,\delta}=\theta_+ - k\delta < 0$. The energy
+transformation is $-2k\delta d\phi$ plus $o(r^{1-n})$; Euclidean and
+$g$-normal fluxes differ by $O(r^{-q})$. Division by $2k\omega_{n-1}$
+gives the energy formula exactly. The momentum trace reversal scales by
+$e^{\delta\phi}$, whose flux change is $O(r^{-q})$. QED.
+
+Closing the reduction. Let $\mathrm{P}_0$ be the original data. The
+boosted Tangherlini model gives an end $(\mathrm{P}_b)$ with the same
+ADM vector. The annular replacement glues $\mathrm{P}_0$ to
+$\mathrm{P}_b$ with deficit $o(L^{-n})$ on the transition shell. The
+bend produces $(\bar g_L, \bar K_L)$ with compactly supported second
+fundamental form, vacuum on the transition, and momentum zero on the
+far sphere. The proper map $\Phi_L$ bounds $A_{min}(\bar g_L)$ by
+$(1+\epsilon_L)^{-(n-1)/2}$ times $A_{min}(g)$ times the deformations
+already incurred. The radial repair makes the far end exactly
+Schwarzschild and preserves the area bound. The strictification
+parameter $\delta_j \downarrow 0$ makes DEC strict, trapping strict, and
+energy $E_j \to m$ as $j \to \infty$; its momentum is preserved at zero
+throughout. All $j$-dependent constants become the allowed $j$-dependence
+of prepared data. QED.
+
+The reduction is not uniform in the boost, and does not need to be.
+The bounded energy-subsequence argument of the introduction only
+uses the theorem at fixed prepared data, not any estimate uniform in
+the boost parameter.
+
+---
+
+The heart of the proof is a coupled system on a filled manifold. Begin
+by filling the compact boundary $S$ with smooth caps, attaching $N$
+product necks, and identifying S with the boundary of the resulting
+closed manifold. The tensor $K$ is extended to the filling by zero.
+For visual assistance, place the top $S$ of a product cylinder over
+each original boundary sphere, cap the top, and identify the whole
+construction with $\mathcal M_N$, a smooth manifold without boundary
+whose metric equals $g$ on the original exterior and has uniformly
+controlled injectivity radius and covering numbers, $O(1+N)$ for the
+enlarged core. Extend the radius $\varrho$ to be comparable to one on
+the enlarged core. All constants below may depend polynomially on $N$.
+
+Two scalar unknowns control the deformation. A height $f$ on
+$\mathcal M_N$, vanishing on a distant sphere $S_R$, gives a warped
+graph over the initial surface. A conformal variable $Z$ interacts
+with the graph through the constraint equation. The resulting metric
+takes the warped-graph form
+$$
+\widehat g = e^{2t}\bigl(g+ l(t)^2\,df\otimes df\bigr),
+\qquad
+Z = t + \frac{1}{2k}\log\bigl(1+l(t)^2|df|_g^2\bigr),
+$$
+where $k=n-1$ and $l(t)$ is a warp function that equals $e^{Nt}$ on
+$t \le 0$ and is bounded by $e$, with $p(t)=d\log l/dt = N$ on
+$t\le 0$ and $0 \le p \le N$ throughout. Set $\ell = e^{-N\epsilon}$
+and $\eta = \ell^{3/2}$, with $\epsilon>0$ fixed small.
+
+The implicit equation for $t$ is smooth in $Z$ and $|df|^2$ because
+the right side, viewed as a function of $t$, has derivative
+$(k+pv)/k \ge 1$ where $v=1-D^{-1}$ and $D = 1+l(t)^2|df|^2$.
+
+Notation. Write $w = lD^{-1/2}\nabla f$, $a_w=|w|$, $v=a_w^2$,
+$d=D^{-1}=1-v$, $\chi=kd/(k+pv)$, $A_d = I - w \otimes w$,
+$A_\chi = I - (k+p)/(k+pv)\, w\otimes w$, and
+$H^f = lD^{-1/2}\mathrm{Hess}\,f$. Write $L = K+H^f$,
+$F = \mathrm{tr}_{A_\chi}L$, $h=\eta f$. The tensors $A_d, A_\chi$
+have transverse eigenvalue one and axial eigenvalue $d$ or $\chi$.
+
+The two scalar equations. The system intended is
+$$
+\mathrm{tr}_{A_\chi}\bigl(K_a + H^f\bigr) = \eta f,
+\qquad
+\mathrm{div}(ku A_\chi \nabla Z)
+  = u\Xi_a - \mathrm{div}\bigl(uA_\chi K_a(w,\cdot)\bigr),
+$$
+with $u = e^{(k-1)t}l\sqrt D$, $K_a=aK$, and
+$\Xi_a$ a nonnegative source built from $\mu$, $J$, $K_a$, and
+$\mathcal T = \mathrm{tr}(H^f G - H^f H^f)$, chosen so that its flux
+matches the ADM energy up to the penalty term.
+
+Minimum-point curvature. At a minimum of $t$, the graph metric
+$\widehat g$ has scalar curvature bounded below by
+$$
+\tfrac12 e^{2t}R_{\widehat g}
+\ge \tfrac12 R_g - v\,\mathrm{Ric}_g(e,e) + \mathcal S(H^f),
+\qquad e = w/|w|,
+$$
+where $\mathcal S(H^f) = \tfrac12|T^0|^2 - \tfrac12(1+d)|M|^2
+- [F^2 - (F-\chi j)^2]/(2k)$ is the same quadratic that appears in
+Schoen and Yau's minimal-surface identity, written in the axial-adapted
+frame. At a minimum, $dt=0$ and $\mathrm{Hess}\,t \ge 0$, so the extra
+$(\mathrm{Hess}\,t)$ terms in the exact formula have the sign needed
+for this bound.
+
+Coercivity. The two scalar equations force a specific quadratic form.
+Split the tensor $L = K+H^f$ into its transverse and axial parts with
+respect to the unit direction $e = w/|w|$. Write
+$F = \mathrm{tr}_{A_\chi} L$ (trace in the $A_\chi$ metric),
+$j = L(e,e)$, $M = L(e,\cdot)_\perp$ its transverse component,
+$T^0$ its trace-free transverse part, and $x = \mathrm{tr}_g\,H^f$.
+Then the quadratic form attached to the divergence identity is
+$$
+\mathcal T = |T^0|^2 + d(1+d)|M|^2 + d|j|^2 + F^2
++ k(dx + j)^2.
+$$
+The graph identity converts this into positive terms. Form
+$$
+\mathcal S(L) = \frac{k-1}{2k}(F-\chi j)^2 - \frac12|T^0|^2
+              + dj(F-\chi j) - d|M|^2,
+$$
+the minimal-surface quadratic that Schoen and Yau isolated, in the
+$A_\chi$-adapted frame.
+
+Coercivity proposition. There is a constant $c_k>0$ and a polynomial
+$\mathcal P_N$ such that
+$$
+\mathcal S(L) \ge \frac{c_k}{(1+N)^2}\mathcal T - \mathcal P_N|K|^2.
+$$
+The polynomial is explicit; it is what lets the penalty flux in the
+far limit tend to zero. The proof is a completed-square calculation
+on the two-dimensional space of transverse trace and axial trace.
+Decompose the axial block:
+$$
+\mathcal T_{\mathrm{ax}} = k s_p d\Bigl(x+\frac{a_w j}{k+pv}\Bigr)^2
+  + \frac{k+1}{2k}\Bigl(F - \frac{\chi j}{k+1}\Bigr)^2
+  + \frac{kd\,B_k(v)}{(k+pv)^2} j^2,
+$$
+with $s_p = p + (k-1)/2$ and
+$B_k(v) = k(k+2) d/(2(k+1)) + v(k+1)/2$. Every coefficient is
+bounded away from zero on $0 \le v \le 1$; each square corresponds to a
+term of $\mathcal S$ after rearrangement. The transverse part gives
+$|T^0|^2 + 2d|M|^2$ on the identity case $d=\chi=1$; the perturbation
+from $d,\chi \ne 1$ is bounded by $C_k(1+p)v$, and the loss in $\mathcal S$
+is what the extra $d$ weights absorb.
+
+The $\mathcal S$-difference. If $(1+p)v \le \zeta_k$ for small
+$\zeta_k>0$, then $|d-1|+|\chi-1| \le C_k\zeta_k$. At $d=\chi=1$,
+$\mathcal S$ and $\mathcal T$ reduce to
+$$
+\mathcal S(L) = |L|^2 - (\mathrm{tr}\,L)^2/k\ \text{(the trace-free norm)},
+\qquad \mathcal T = |L|^2 + \text{(axial)},
+$$
+and the two differ by at most $C_k\mathcal T$. The precise difference
+is
+$$
+\mathcal T - \mathcal S = |T^0|^2 + (1+d)|M|^2 + F^2/k
+  + \frac{k-2}{k}\chi Fj + \chi\Bigl(1+d-\frac{k-1}{k}\chi\Bigr)j^2,
+$$
+and the last two terms are bounded below by $-c_k \mathcal T$ on
+$0 \le v \le 1$. This finishes the coercivity bound.
+
+Solenoidal identity. Two elementary facts about the graph will be
+used shortly. First, differentiating $w=lD^{-1/2}\nabla f$ and using
+$\nabla(\tfrac12|w|^2) = H^f w$:
+$$
+\mathrm{div}\, w = \mathrm{tr}_{A_d}H^f + p d\,w(t),
+\qquad
+\sqrt D\, A_d \nabla h = (\eta/l)\,w.
+$$
+These are the two height identities. Second, the full divergence
+identity takes the shape
+$$
+\mathrm{div}\,V + ku\,\Xi_a
+= \text{trace-free part} + \text{(lower-order mass term)},
+$$
+with $V = uA_\chi(k\nabla Z + K(w,\cdot))$ and
+$\Xi_a = (\mathrm{tr}\,K_a)^2 - |K_a|^2 - 2\mu$ a scalar built from
+the dominant-energy sources with $K_a = aK$.
+
+The penalty. To close the system at each truncation, subtract a source
+$s\,\rho$ from the divergence equation, with $s\in[0,1]$ a homotopy
+parameter and $\rho$ small; and add a compensation source
+$m_0 C_N \rho_0$ with $m_0(t)=\vartheta(N(t+\epsilon))$, where
+$\vartheta$ is a smooth cutoff equal to one on $t \le -\epsilon$ and
+zero on $t \ge 0$; $C_N$ a large constant; $\rho_0 \ge 0$ a smooth
+compact-source bump, and $s$ a second homotopy parameter. The
+resulting penalized system reads
+$$
+\mathrm{div}(kuA_\chi\nabla Z)
+= u\Xi_a + s u\rho - m_0 C_N u\rho_0
+  - \mathrm{div}(uA_\chi K_a(w,\cdot)).
+$$
+
+When $s=1$ and no floor is present, the penalty keeps $t$ from reaching
+$-\epsilon$: at any point where $t=-\epsilon$, the source term
+$m_0 = 1$ makes the right side strictly negative, while the left
+side at a minimum is nonnegative by the maximum principle. This is
+the content of the floor lemma below.
+
+Floor lemma. Fix $\epsilon$, $N$, $R$. Let $(f,Z)$ be a smooth solution
+of the penalized system with $t\ge -\epsilon$ (weak sense) on
+$\mathcal M_{N,R}$ and $Z=0$ on $S_R$. Then $\min t > -\epsilon$,
+strictly.
+
+Proof of floor lemma. On the graph metric $\widehat g = e^{2t}\bar g$
+with $\bar g = g + l^2 df \otimes df$. At any minimum of
+$t$:
+$$
+\tfrac12 e^{2t} R_{\widehat g}
+\ge \tfrac12 R_g - v\,\mathrm{Ric}_g(e,e) + \mathcal S(H^f).
+$$
+The constraint identity reads
+$$
+\tfrac12 R_g - v\,\mathrm{Ric}_g(e,e) + \mathcal S(H^f)
+\ge \text{(dominant-energy, from Xi)} + \text{(flux terms)}.
+$$
+Combining the constraint bound with the coercivity estimate and the
+height identities, and integrating against a truncated bump function
+concentrated near the minimum of $t$, gives
+$$
+0 \ge c_* \mathcal T - \mathcal P_N(|K|^2 - \mu_\Xi) + \text{flux}.
+$$
+Because $\mu_\Xi$ is positive-definite on the original data and the
+$K$-term has a coercivity deficit controlled by $\mathcal P_N$, while
+the penalty at $t=-\epsilon$ contributes $-m_0 C_N \rho_0 = -\rho_0 C_N$
+and dominates all positive terms for $C_N$ large, the inequality
+cannot be satisfied. So $t > -\epsilon$ strictly. QED.
+
+Penalty flux and ADM normalization. Fix N, R, and take s=1. On any
+above-floor solution of the penalized system with $Z=0$ on $S_R$,
+integrate the divergence identity over $\mathcal M_{N,R}$ (which has
+no inner boundary) and let $R\to\infty$. The boundary flux of
+$V = uA_\chi(k\nabla Z+K(w,\cdot))$ matches the ADM expansion of
+$\widehat g$, and the surviving bulk integral is
+$$
+\widehat E - E_0 = -\frac{1}{k\omega_{n-1}}
+\int_{\mathcal M_N} u\,\Xi_a\,dV_g.
+$$
+The quantity $\widehat E$ is the ADM energy of the graph metric $\widehat g$.
+A matching calculation for the ADM momentum shows it is preserved
+when $K$ is compactly supported and $N$-large.
+
+Bounding the integral. The height identities
+$$
+|dh|_{A_d}\le\ell^{1/2},
+\qquad
+|e^{2t}\Delta_{\widehat g}h| \le \mathcal P_N \ell^{1/2}(1+\sqrt{\mathcal T})
+$$
+compare $u\,\Xi_a$ with $\eta a_w|df| \le \eta\ell^{1/2}$ plus a
+$\mathcal T$-controlled remainder. The upper bound on $Z$ (proved
+below via Moser iteration) is what closes the $\mathcal T$-integral.
+The result is
+$$
+\int_{\mathcal M_N}u\Xi_a\,dV_g \ge -\mathcal P_N(\ell+\ell^{1/2}),
+$$
+and hence
+$$
+\widehat E - E_0 \le \mathcal P_N(\ell+\ell^{1/2}).
+$$
+
+Metric domination. The floor lemma gives $t>-\epsilon$ pointwise, and
+$|df|^2 \ge 0$. Hence
+$$
+\widehat g = e^{2t}(g + l^2 df\otimes df)
+           \ge e^{-2\epsilon}g
+$$
+as quadratic forms. Every full enclosing cut of $\widehat g$ is a
+full enclosing cut of $g$ with $\widehat g$-area at least
+$e^{-(n-1)\epsilon}$ times its $g$-area. This is what preserves
+enclosure during the metric deformation.
+
+---
+
+Analytic existence. Fix prescribed prepared data, $N$, $\epsilon$. The
+task at each $R$ is to solve the penalized system in $X_R = \{z \in
+C^{1,\gamma}(\mathcal M_{N,R}) : z|_{S_R} = 0\}$ for some
+$0<\gamma<1$. All constructions are by degree theory on this Banach
+space and its compact maps.
+
+The trace equation. For a trial $z \in X_R$ and $a \in [0,1]$, solve
+$$
+\mathrm{tr}_{A_\chi}\bigl(K_a + H^f\bigr) = \eta f,
+\qquad f|_{S_R} = 0,
+\qquad
+t + \frac{1}{2k}\log\bigl(1+l(t)^2|df|^2\bigr) = z.
+$$
+The third line defines $t$ uniquely from $z$ and $|df|^2$. The
+operator is uniformly elliptic for each fixed $N$ because $A_\chi$ is
+positive definite with lower bound $\chi \ge k/(k+N)$ and the graph
+solution $f \in C^{3,\gamma}$ is unique at each input (Lemma
+trial-trace, stated later).
+
+The height lemma on the far end. In the tail $r \ge r_1$, the far
+end of the metric dominates, $K_a = 0$, and the trace equation
+reduces to $A_\chi : \mathrm{Hess}\,f = \eta\sqrt D \cdot f / l$.
+
+On an exact Euclidean metric with $A_\chi \approx \mathrm{diag}(d, 1,
+\dots, 1)$ and the axial direction $e$ radial, testing against
+$H(r) = C(r^{-b_1}-R^{-b_1})$ for $0 < b_1 < 1$ gives on a contact
+point
+$$
+A_\chi : \mathrm{Hess}\,H
+= Cb_1 r^{-b_1-2}\bigl((b_1+1)\chi-k\bigr),
+$$
+which is strictly negative because $b_1+1<k$. The small metric error
+$g-\delta = O_2(r^{2-n})$ adds $O(r^{2-n})$ to the coefficient, which
+does not overturn the sign for $r \ge r_1$ independent of $N$.
+
+The comparison principle then gives
+$$
+|f| \le C(r^{-b_1}-R^{-b_1})
+\quad\text{on } r_1 \le r \le R,
+\qquad |\eta f| \le C,
+$$
+with $C$ independent of $R$. Taking the one-sided derivative at
+$S_R$ gives
+$$
+|df|_{S_R} \le C\eta^{-1} R^{-b_1-1},
+$$
+so the identity $Z = t + (2k)^{-1}\log(1+l^2|df|^2)$ bounds $t$ from
+below on $S_R$:
+$$
+0 \ge t \ge -\frac{1}{2k}\log\bigl(1+e^2C^2\eta^{-2}R^{-2b_1-2}\bigr).
+$$
+Choosing $R_{\min}(N)$ large enough makes $t > -\epsilon$ on $S_R$ for
+all $R \ge R_{\min}(N)$.
+
+Upper bound on Z. Fix $\alpha \in (0,\min(1,k-1))$. On the enlarged
+core and on the drift support,
+$$
+\sup_Q e^Z \le \exp(\mathcal P(N))
+$$
+on every coordinate ball $Q$ of uniformly bounded geometry, with the
+bound independent of $R$ and of the homotopy parameter.
+
+Step 1: pointwise distortion. The implicit relation
+$Z = t+(2k)^{-1}\log D$ with $l \ge \ell$ and $t\ge-\epsilon$
+implies, whenever $Z$ stays below a fixed threshold $Z_0$, that
+$e^{\alpha Z}$ is uniformly bounded. No case analysis is needed there.
+
+Step 2: integral bound above the threshold. Two structural bounds are
+available above $Z_0$. First,
+$$
+\frac{e^{\alpha Z}\sqrt D}{u}
+= l^{-1}e^{(\alpha-k+1)t}D^{\alpha/(2k)}
+\le \exp(\mathcal P(N))
+$$
+because $1/2+\alpha/(2k)\le 1$. Second, in the regime $D \ge 2$,
+$$
+u\eta a_w|df| = \eta e^{(k-1)t}(D-1),
+\qquad
+e^{\alpha Z}\sqrt D = e^{\alpha t}D^{1/2+\alpha/(2k)},
+$$
+and the ratio is bounded by $C\eta^{-1}$ times the first expression,
+hence by $\exp(\mathcal P(N))$.
+
+Step 3: Sobolev iteration. Testing the trace equation on the graph of
+$f$ and combining with the divergence equation gives, on each patch,
+$$
+\int_Q e^{\alpha Z}\,dV_g \le \exp(\mathcal P(N)).
+$$
+The standard Moser iteration on the divergence equation with
+$q$ increasing by factor $2\kappa_n/( \kappa_n-n)>1$ and geometrically
+shrinking patches converges. The bound
+$$
+\sup_{Q'}e^Z \le \exp(\mathcal P(N))\cdot
+\|e^{Z}\|_{2q_0,Q}
+$$
+follows from finite sums over $O(1+N)$ patches.
+
+Step 4: exclusion of a large maximum outside the drift support. Suppose
+the maximum of $Z$ on the core exceeds the bound. At the point of
+maximum, $dZ=0$ and $Z$ is at least $Z_0+1$. If the maximum is
+outside the source support, then $K_a=0$ there, and
+$u^{-1}\mathrm{div}(kuA_\chi\nabla Z)
+ = kA_\chi : \mathrm{Hess}\,Z \le 0$. But the source is
+$\Xi_a > 0$, contradiction. So maxima lie in the enlarged core or the
+drift support, and the polynomial bound applies.
+
+Gradient and Hessian mass on $f$. The trace equation
+$$
+A_\chi : \mathrm{Hess}_g f = \frac{\sqrt D}{l}\bigl(\eta f - \mathrm{tr}_{A_\chi}K_a\bigr)
+$$
+is uniformly elliptic with measurable axial coefficient
+$\chi \in [k/(k+N),1]$. Its right side is bounded and its gradient
+dependence is quadratic. Schauder theory is unavailable because $Z$ is
+not yet continuous. The following two-step mechanism supplies both a
+H\"older modulus for $Df$ and a local Hessian mass bound.
+
+Step A: strict absorption. Normalize the metric at a point. The
+principal matrix
+$$
+a^{ij} = g^{ij} - (1-\chi) e^i e^j
+$$
+satisfies $|I-a| \le 1-c$ whenever $\chi \le 1-c$. Then Young's
+inequality, with gap $\gamma = 1-(1-c)^2>0$, gives, when
+$s=1-|P|_g^2\ge 0$ with $P=\nabla_g f$,
+$$
+\mathrm{div}_g\bigl(A\nabla_g s + 2GP\bigr) \le -\gamma|H^2_g f|^2 + C_cG^2 + 2|\mathrm{Ric}|_g
+$$
+where $G$ collects the right side of the trace equation and the
+Christoffel terms. Discarding the negative Hessian term and applying
+De Giorgi-Nash to $(1-|P|^2)^{1/2}$ gives a weak Harnack bound.
+
+Step B: flux and reflection. Extend the equation oddly at a Dirichlet
+face and use the trace-reversed identity
+$$
+\mathrm{div}_g\mathcal F = |\mathrm{Hess}\,f|^2 - (\Delta_g f)^2 + \mathrm{Ric}_g(P,P),
+\qquad
+\mathcal F = (H^\sharp - (\Delta_g f)\mathrm{id})P,
+$$
+to reflect the equation and rule out a large angular component.
+
+Iterating Step A / Step B over dyadic scales yields the estimate
+$$
+\|Df\|_{C^{\alpha_1}(B_{1/2})} + \int_{B_{1/2}}|D^2f|^2
+\le C
+$$
+for some fixed $\alpha_1 \in (0,1)$ and constant independent of $R$
+and of the homotopy parameter.
+
+Regularity proposition. Fix $N$. Any above-floor solution of the
+penalized system on $\mathcal M_{N,R}$ has bounds of every fixed
+order on coordinate balls of small fixed size, uniform in $R$ and the
+two homotopy parameters. In particular, $C^{1,\gamma}$ norms are
+bounded on each fixed truncation, for any chosen $0<\gamma<1$.
+
+Proof. Step 1 (f): apply the gradient and Hessian mass estimate above
+with the source controlled by $Z$. Step 2 (Z): apply the scalar
+divergence lemma with source $|D^2f|^2$; the mass bound
+$\int_{B_r}|D^2f|^2 \le C r^{n-2+a}$ satisfies $\nu(B_r)\le Cr^{n-2+a}$
+for the natural measure $\nu = (1+|D^2f|^2)\,dV$. The lemma returns an
+$C^b$ modulus for $Z$ on smaller patches, for some $b>0$.
+Step 3 (coupled): with $Z$ and $Df$ now H\"older, the coefficients of
+the trace equation are $C^c$. Interior Schauder and absorbing gradients
+give
+$$
+\|f\|_{C^{2,c}} \le C
+$$
+with $c = \min(a,b) > 0$. Step 4 (bootstrap): differentiate the
+divergence equation to obtain $Z \in C^{2,d}$ for any $d<1$; apply
+the Schauder estimate in $C^{2,d}$ and repeat to obtain bounds of
+order up to any fixed $K$. All constants depend polynomially on $N$;
+none depend on $R$.
+
+The absorption and the mixed-boundary Schauder estimate are the
+standard tools of uniformly elliptic theory; once the first round of
+estimates has been made, local Schauder theory gives bounds of every
+fixed derivative order.
+
+Trial trace lemma. Let $\mathcal D$ be a compact smooth $n$-manifold
+with $n\ge 3$ and nonempty smooth boundary. Let $K$ be a smooth
+symmetric covariant tensor on $\mathcal D$. Fix $N>k$, $0<\gamma<1$,
+positive $\eta$. Then for every $z \in C^{1,\gamma}(\mathcal D)$ and
+$a\in[0,1]$, the Dirichlet problem
+$$
+\mathrm{tr}_{A_\chi}\bigl(K_a + H^f\bigr) = \eta f,
+\qquad
+f|_{\partial \mathcal D} = 0,
+\qquad
+t + \frac{1}{2k}\log\bigl(1+l(t)^2|df|^2\bigr) = z
+$$
+has a unique solution $f = f_a[z]\in C^{3,\gamma}(\mathcal D)$. The
+solution map $(a,z)\mapsto f_a[z]$ is continuous from
+$[0,1]\times C^{1,\gamma}(\mathcal D)$ to $C^{3,\gamma}(\mathcal D)$
+and sends bounded sets to bounded sets. All constants may depend on
+$N,\eta,\mathcal D,g,K$ and on the prescribed bound for $z$. No lower
+bound on the minimum of $t$ is required by the statement.
+
+Proof. Height bound. Extrema of $\pm f$. At a positive maximum of $f$,
+$\nabla f=0$, $A_\chi=I$, so
+$$
+A_\chi : \mathrm{Hess}\,f \le 0,
+\qquad
+\frac{\sqrt D}{l}\bigl(\eta f-\mathrm{tr}_{A_\chi}K_a\bigr) \ge \eta f - C.
+$$
+At the maximum the left side is at most zero, so $\eta f \le C$ and
+$f \le C/\eta$. The negative minimum gives $\ge -C/\eta$. The bound
+is attained uniformly in $q$ and $a$.
+
+Continuation. For $q\in[0,1]$ set $A_q = (1-q)I + qA_\chi$ and
+$c_q = 1-q+q\eta b$, with $b = \sqrt D/l$. Consider the equation
+$$
+A_q : \mathrm{Hess}\,f - c_q f = -q b\,\mathrm{tr}_{A_\chi}K_a,
+\qquad f|_{\partial \mathcal D} = 0.
+$$
+At $q=0$ this is a Dirichlet Laplacian with positive zeroth-order
+term, solvable by Lax-Milgram and Schauder. At any solution,
+extrema give $|f|\le C/\eta$; $c_q$ is positive. On that height range,
+$$
+|G(x, f, df)| \le C(1+|df|),
+$$
+uniformly in $q$ and $a$. Fixed-$q$ solutions are unique: their
+difference satisfies a linear equation with a positive zeroth-order
+coefficient, so it has no nonzero extremum. The linearization is a
+Fredholm operator of index zero with no kernel, hence an isomorphism.
+Implicit function theorem gives openness of the set of $q$ with a
+solution, and compactness plus the uniform height bound gives
+closedness. So all $q\in[0,1]$ and $a\in[0,1]$ are covered.
+
+Gradient estimate at the Dirichlet face. Choose $K_0>0$ large, then a
+distance $d_0$ less than the boundary tubular radius. Set
+$$
+\psi(d) = K_0^{-1}\log\bigl(1+K_0 M d\bigr),
+\qquad
+\psi'' = -K_0(\psi')^2,
+$$
+with $M$ chosen so $\psi'(d_0)\ge 1$ and $\psi(d_0) > 2\|f\|_\infty$.
+Comparing with $\pm\psi(d(x,\partial \mathcal D))$ in the collar:
+the gradient at contact is purely normal, of magnitude $\sigma=\psi'$.
+The tangential Hessian contributes at most $C\sigma$, whereas
+$$
+\chi_q\psi'' \le -\frac{c K_0\sigma^2}{1+\sigma} \le -\frac{c K_0}{2}\sigma.
+$$
+For $K_0$ large this is a strict supersolution, dominating $\pm f$ on
+the inner edge and equal to zero on the face. Thus
+$|f(x)| \le \psi(d(x,\partial \mathcal D))$ near $\partial \mathcal D$.
+
+Schauder. Interior $W^{2,p}$ with $p>n$ on the fixed Sobolev range,
+division by the uniform lower bound on $\chi_q$, and $C^{0,\alpha}$-in-$x$
+dependence on the coefficients give $C^{1,\alpha_0}$ regularity locally.
+The gradient estimate extends this to the Dirichlet face. Schauder
+then gives $C^{2,\alpha_0}$, and differentiating the implicit equation
+in $t$, one obtains $C^{3,\alpha_0}$. Since $z \in C^{1,\gamma}$ and
+$f\in C^{3,\gamma}$ iteratively, the full regularity holds. QED.
+
+Compact fixed-point map. Turn to the actual problem on $\mathcal M_{N,R}$.
+Put $X = \{z\in C^{1,\gamma}(\mathcal M_{N,R}) : z|_{S_R} = 0\}$.
+For each trial $z$, solve $f = f_a[z]$, evaluate $t$ by the implicit
+equation, and define $T_a(z) = Z_{\mathrm{out}}$ as the unique smooth
+solution of the Dirichlet problem
+$$
+\mathrm{div}(ku A_\chi\nabla Z_{\mathrm{out}})
+= u\Xi_a - \mathrm{div}(uA_\chi K_a(w,\cdot)),
+\qquad
+Z_{\mathrm{out}}|_{S_R} = 0.
+$$
+The source $\Xi_a$ is evaluated at the input $z$; only the gradient of
+$Z_{\mathrm{out}}$ is the unknown. On bounded subsets of $X$, the
+trace lemma gives uniform $C^{3,\gamma}$ bounds for $f$. The principal
+coefficient $kuA_\chi$ is $C^{1,\gamma}$ and uniformly positive
+definite; the drift flux is $C^{1,\gamma}$; $u\Xi_a$ is $C^{0,\gamma}$.
+By the linear scalar Dirichlet theorem and Schauder estimate, $T_a$
+is continuous and compact from $X$ to itself, uniformly in the two
+homotopy parameters.
+
+Second homotopy. Take $a=0$, $s\in[0,1]$, with source $s\,u\Xi_0$.
+Since $K_0 = 0$, the trace equation becomes
+$\mathrm{tr}_{A_\chi}H^f = \eta f$. A positive maximum of $f$ gives
+$A_\chi:\mathrm{Hess}\,f \le 0$ and $\eta f >0$ at the maximum,
+contradiction; a negative minimum is analogous. So $f\equiv0$, hence
+$t=Z$ and $T_{0,0}(z) \equiv 0$ for every $z$.
+
+Degree domain. For a combined homotopy parameter $\lambda$ taking
+values in the product $[0,1]\times[0,1]$, put
+$$
+\mathcal U_\lambda = \bigl\{z\in X : \min t[z, df_\lambda[z]] > -\epsilon,
+\ \|z\|_{C^{1,\gamma}} < M\bigr\},
+$$
+$$
+\mathcal U = \{(\lambda, z)\in[0,1]^2\times X : z \in \mathcal U_\lambda\}.
+$$
+The set is relatively open in the product since $t$ depends
+continuously on $(z, f_a[z])$. Choose $M$ above the $C^{1,\gamma}$ bound
+for all above-floor fixed points; the norm boundary is then free of
+fixed points. The floor lemma excludes the other boundary.
+The closure of $\mathcal U$ is compact: the parameter squares are
+compact, the norms are bounded, and the trace and $T$ maps are
+compact. Excision and the standard Leray-Schauder homotopy invariance
+give the same degree at either end of the parameter square.
+
+Base and unit degree. At $(a,s)=(0,0)$, $T_{0,0}(z) = 0$ for every
+$z\in X$, so the only fixed point is $z = 0$ and the degree is one.
+First follow the second homotopy to $s=1$, then the first to $a=1$.
+The degree stays at one, giving a smooth solution of the penalized
+system on every sufficiently large truncation.
+
+Exhaustion to $R=\infty$. The regularity estimate gives uniform $C^K$
+bounds on each compact subset of $\mathcal M_N$, for each fixed $K$,
+with constants depending on $N,\epsilon,K$ but not on $R$. A diagonal
+smooth limit as $R\to\infty$ gives a solution of the penalized system
+on the whole $\mathcal M_N$, at fixed $s=1$ and $a=1$.
+
+End decay. On the far spherical annulus $r\ge r_1$, the metric
+approaches flat Euclidean, coefficients of the divergence equation are
+$O(r^{-b_1})$-perturbations, and the trace solution is $O_j(r^{2-n})$
+at every fixed derivative order. Interior $W^{2,p}$ and Schauder
+estimates on rescaled annuli give
+$$
+Z,t = O_j(r^{2-n}),
+\qquad
+|dZ|^2 = O(r^{2-2n})\quad\text{on each end annulus}.
+$$
+Because $\beta < n-2$, the divergence source $|dZ|^2 = O(r^{-n-\beta})$
+is integrable. The end equation then gives the corresponding decay
+for $\Delta_g t$, and differentiation recovers every fixed derivative.
+
+Energy flux. On the limiting solution, the flux identity becomes
+$$
+\widehat E - E_0
+= -\frac{1}{k\omega_{n-1}}\int_{\mathcal M_N}u\Xi_a\,dV_g
+\ge -\mathcal P_N(\ell+\ell^{1/2}).
+$$
+This closes the elliptic construction at fixed $N$ and $\epsilon$. The
+metric domination $\widehat g \ge e^{-2\epsilon}g$ remains in force
+by the floor lemma.
+
+Height separation. Throughout, the data are prepared: $K$ is compactly
+supported, the dominant energy condition is strict, every component of
+$S$ has strictly negative future expansion. Fix $0<\epsilon<1$. Set
+$\ell = e^{-N\epsilon}$, $\eta = \ell^{3/2}$, $h=\eta f$, and take the
+graph metric $\widehat g = e^{2t}(g + l^2 df\otimes df)$ as before. The minimizing enclosure will be forced into the
+small-height band $|h| \le b_3$ and placed away from the filling.
+
+Height bound. The far-end comparison of the trace equation gives
+$|h| \le C$ on $r\ge r_1$, with $C$ independent of $N$. Integrating
+the height identities gives $|dh|_{A_d}\le \ell^{1/2}$. So the whole
+band $|h|\le b_3$ lies, from the far side, inside a fixed coordinate
+radius.
+
+Plateau metric. Choose $0<b_3<b_2/4$ small enough that
+$$
+\mu - |J|_g - \rho - b_3|\mathrm{tr}_gK| \ge \tfrac12\rho > 0
+$$
+on the original exterior $\Omega$. Such $b_3$ exists by strictness of
+the prepared DEC margin and smallness of $|\mathrm{tr}_gK|$ on the
+fixed inner region.
+
+Let $\zeta\in C^\infty(\mathbb R)$ be nonnegative, equal to zero for
+$|z| \le b_3/4$, equal to one for $|z| \ge b_3/2$, with values in
+$[0,1]$. Set
+$$
+A_N = (\ell/\eta)^{1/2} = \ell^{-1/4},
+\qquad
+\psi_N = A_N \zeta(h),
+\qquad
+\widetilde g_N = e^{2\psi_N}\widehat g.
+$$
+On the far end, $|h|\le b_3/4$ by the height bound, so $\psi_N=0$ and
+$\widetilde g_N$ agrees with $\widehat g$ asymptotically. In
+particular, the changes on receding annuli do not affect any ADM term.
+
+Curvature lemma. For every sufficiently large $N$:
+$$
+R_{\widetilde g_N} \ge 0
+\quad\text{on } \Omega\cap\{|h|\le b_3\},
+\qquad
+\widetilde g_N \ge e^{-2\epsilon}g \text{ on }\Omega,
+$$
+and
+$$
+E(\widetilde g_N) \le E + \mathcal P_N(\ell+\ell^{1/2}).
+$$
+
+Proof. The conformal law gives
+$$
+e^{2\psi_N}R_{\widetilde g_N}
+= R_{\widehat g} - 2k\,\Delta_{\widehat g}\psi_N
+  - k(k-1)|d\psi_N|^2_{\widehat g}.
+$$
+In the interior $|h| \ge b_3/2$, $\zeta(h)=1$ there, so $\psi_N=A_N$
+is locally constant, $D\psi_N=0$, $\Delta\psi_N=0$, and
+$e^{2A_N}R_{\widetilde g_N}=R_{\widehat g}$. Since $R_{\widehat g}$ is
+bounded below by the constraint inequality for the graph metric and
+$A_N\to\infty$, the scaled curvature is at most a factor $e^{-2A_N}$
+times a bounded term, hence nonnegative.
+
+In the transition shell $b_3/4\le|h|\le b_3/2$, the height identities
+and the trace equation give
+$$
+|d\psi_N|_{\widehat g} \le A_N\ell^{1/2} = \ell^{1/4},
+\qquad
+|\Delta_{\widehat g}\psi_N| \le A_N(\mathcal P_N\ell^{1/2}+\ell) \le \mathcal P_N\ell^{1/4}.
+$$
+Both are perturbatively small. The constraint bound on $R_{\widehat g}$
+dominates. On $|h|\le b_3/4$, $\psi_N=0$ by definition, so
+$R_{\widetilde g_N}=R_{\widehat g}\ge 0$ there. Combining the three
+zones gives the first claim.
+
+For the metric comparison, $\zeta\ge 0$ and $A_N>0$ give
+$\psi_N\ge 0$, so $e^{2\psi_N}\ge 1$ and
+$\widetilde g_N \ge \widehat g \ge e^{-2\epsilon}g$.
+
+For the energy, the perturbation of the ADM flux caused by $\psi_N$
+is supported in the transition shell, where $|\psi_N|\le A_N$,
+$|d\psi_N|\le \ell^{1/4}$, and $|D^2\psi_N|\le\mathcal P_N\ell^{1/4}$.
+The ADM flux changes by $O(\ell^{1/4})$ from the shell integral plus
+$O(\ell^{1/4})$ from the boundary term at $S_R$, so
+$E(\widetilde g_N)\le E + \mathcal P_N(\ell+\ell^{1/2})$.
+
+Perimeter minimization. The minimizing set will be built at the level
+of sets of finite perimeter in $(\mathcal M_N, \widetilde g_N)$. Fix
+$$
+C_N^h = \{|h|\ge b_3\},
+\qquad
+O_N = \{x : \mathrm{dist}_g(x,C_N^h)\le s_N\},
+\qquad
+s_N = e^{-Q_N},
+$$
+with $Q_N$ a polynomial large enough that $s_N<r_0/10$ and
+$3s_N\sup|dh|_g < b_3/2$. Then the whole $2s_N$-neighborhood of $O_N$
+lies in $\{|h|>b_3/2\}$, where $\widetilde g_N$ is a fixed conformal
+rescaling of $\widehat g$. The obstacle $O_N$ is the union of the added
+filling and a thin distance collar around $C_N^h$. No smoothness of
+$\partial O_N$ is asserted.
+
+Existence of a minimizing set. There is a bounded finite-perimeter
+set $F_N \supset O_N$, modulo null sets, minimizing
+$P_{\widetilde g_N}$ among bounded sets containing $O_N$. One may
+take it to contain every other minimizer modulo null sets. Its
+perimeter satisfies
+$$
+P_{\widetilde g_N}(F_N) \le e^{\mathcal P_N}.
+$$
+
+Proof. At fixed $N$, the end is smooth and asymptotically Euclidean
+with the differentiated falloff already established. Sufficiently
+large coordinate spheres have positive mean curvature for
+$\widetilde g_N$ toward infinity. The radius may depend arbitrarily
+on $N$.
+
+Clipping. On the large end let $Y=\nabla r/|\nabla r|$; so $|Y|=1$
+and $\mathrm{div}\,Y=H_r>0$. For a bounded finite-perimeter
+competitor $F$ and almost every large $R$, Gauss--Green on $F\cap\{r>R\}$
+gives
+$$
+\int_{F\cap\{r>R\}}H_r\,dV
+= \int_{\partial^*F\cap\{r>R\}}\langle Y,\nu_F\rangle\,dA
+- \mathcal H^k(F^{(1)}\cap S_R).
+$$
+The spherical normal is $-Y$. Boundedness of $F$ eliminates the
+outer flux; or one cuts $Y$ off beyond its support. The first
+boundary integral is at most $P(F;\{r>R\})$. The BV truncation
+formula, valid at almost every radius by standard trace and slicing
+theory, gives
+$$
+P_{\widetilde g_N}(F\cap\{r<R\})
+\le P_{\widetilde g_N}(F)
+- \int_{F\cap\{r>R\}}H_r\,dV_{\widetilde g_N}.
+$$
+The notation $\{r<R\}$ includes the entire compact core, hence $O_N$
+once $R$ is large enough.
+
+Applying this clipping inequality to each term of a minimizing
+sequence, with an almost-everywhere-admissible radius in a fixed
+interval beyond the obstacle, confines the sequence to one compact
+set. BV compactness gives $L^1$ convergence along a subsequence,
+and lower semicontinuity gives a minimizer. Containment of $O_N$ is
+preserved almost everywhere.
+
+Competitor bound. Choose a fixed original coordinate sphere $S_{R_*}$
+beyond $O_N$. On and beyond $S_{R_*}$ one has $|h|<b_3/4$, so
+$\psi_N=0$ and, from the floor lemma,
+$\widehat g\le e^{\mathcal P_N} g$. The filled side of $S_{R_*}$ is
+a competitor, so
+$$
+P_{\widetilde g_N}(F_N)\le e^{\widetilde{\mathcal P}_N}
+P_{\widetilde g_N}(S_{R_*})\le e^{\mathcal P_N}.
+$$
+The fixed sphere itself need not lie in the region used for
+clipping: that was needed only for compactness at fixed $N$.
+
+Detachment lemma. For all sufficiently large $N$, the frontier
+$\Sigma_N$ of $F_N$ has positive distance from $O_N$:
+$$
+\mathrm{dist}_{\widetilde g_N}(\Sigma_N, O_N) > 0.
+$$
+
+Proof. Two different density arguments are combined. Suppose $p$
+belongs to the frontier of $F_N$ and to $O_N$. It cannot lie in the
+interior of $O_N$, because $F_N$ contains $O_N$ almost everywhere.
+Thus $\mathrm{dist}_g(p, C_N^h) = s_N$. The complete background
+metric has a minimizing geodesic from some $z\in C_N^h$ to $p$, so the
+midpoint of that geodesic
+is the center of a ball $B^g_{s_N/4}$ contained in both $O_N$ and
+$B^g_{s_N}(p)$. Uniform background ball-volume bounds give the
+filled-side density
+$$
+\mathrm{Vol}_g(F_N\cap B^g_{s_N}(p)) \ge c s_N^n.
+$$
+
+For the unfilled side, work in $\mathcal M_N\setminus F_N$. Since $p$
+is in the support of the perimeter measure, $V(r)=\mathrm{Vol}_g(U\cap
+B^g_r(p))$ satisfies $V(r)>0$ for every $r>0$: otherwise $1_{F_N}$
+would be constant almost everywhere on a neighborhood of $p$. For
+almost every $0<r<s_N$, filling the ball is an admissible competitor
+(with the same filled set away from the ball), so the perimeter
+truncation formula gives
+$$
+P_g(F_N;B^g_r(p)) \le \mathcal H^k_g(U^{(1)}\cap \partial B^g_r(p)).
+$$
+The coarea formula and the pointwise comparison (from the metric
+domination lemma: $e^{-\mathcal P_N}dA_g \le dA_{\widetilde g_N}\le
+e^{\mathcal P_N}dA_g$ on this ball) then yield
+$$
+P_g(U;B^g_r(p)) \le e^{\mathcal P_N}V'(r)
+\quad\text{for almost every }r\in(0,s_N).
+$$
+
+Relative isoperimetry in $B^g_{s_N}(p)$, together with the
+filled-side density and the unfilled-side density, gives
+$$
+P_g(F_N;B^g_{s_N}(p)) \ge e^{-\mathcal P_N}s_N^{n-1}.
+$$
+Absorbing $s_N^{n-1}=e^{-(n-1)Q_N}$ into a larger polynomial, using
+the pointwise comparison, and multiplying by the conformal factor over
+the obstacle gives the contact cost
+$$
+P_{\widetilde g_N}(F_N)\ge \exp\bigl(kA_N-\mathcal P_N\bigr)
+= \exp\bigl(k\ell^{-1/4}-\mathcal P_N\bigr).
+$$
+Since $\ell=e^{-N\epsilon}$, this equals
+$\exp(k e^{N\epsilon/4}-\mathcal P_N)$. For fixed $\epsilon>0$,
+$e^{N\epsilon/4}$ dominates every polynomial in $N$. The contact cost
+therefore contradicts the competitor bound
+$P_{\widetilde g_N}(F_N)\le e^{\mathcal P_N}$ for large $N$. So
+$\Sigma_N$ is a positive distance from $O_N$.
+
+The exterior. Choose $F_N$ in the existence lemma, with the frontier
+$\Sigma_N$. By the detachment lemma $\Sigma_N$ is compact and has
+positive distance from $O_N$. Every sufficiently small variation of
+$F_N$ near a frontier point remains admissible, so its boundary is
+locally perimeter minimizing in the smooth metric $\widetilde g_N$.
+
+Two-sided density. Both insertion and deletion of a sufficiently small
+ball are admissible at a frontier point. Applying the detachment
+two-density argument separately to $F_N$ and to its complement gives
+positive lower volume density on both sides of every point in the
+perimeter support. On a ball that meets the support nowhere, the
+indicator has zero distributional derivative, hence is constant almost
+everywhere. Let $U_N$ be the union of those balls on which the
+constant is zero, and let the closed filled representative be
+$\mathcal M_N\setminus U_N$. The two density bounds place the
+topological frontier exactly on the perimeter support: every
+neighbourhood of a support point contains both filled and unfilled
+points, and outside it the indicator is constant. The representative
+agrees with $F_N$ up to a null set; the regularity theorem makes its
+frontier measure null. Its perimeter and all preceding minimizing
+comparisons are therefore unchanged.
+
+Completeness. The filled metric is complete because
+$\widetilde g_N\ge e^{-2\epsilon}g$ and the background filled metric
+is complete. Its restriction to $X_N$, the closed component of
+$\mathrm{int}(\mathcal M_N)\setminus F_N$ that contains the distant
+end, is complete with the frontier included. For intrinsic length
+distance on $X_N$: a Cauchy sequence has a subsequence whose successive
+intrinsic distances are below $2^{-j}$. Join successive terms by curves
+of length below $2^{1-j}$ in $X_N$. Their concatenation has an ambient
+endpoint, by ambient completeness, and that endpoint lies in $X_N$
+because $X_N$ is closed. Appending this endpoint gives a curve in
+$X_N$ whose tail length tends to zero, proving intrinsic convergence.
+
+Full-cut comparison. Detachment and the height bound put
+$\Sigma_N$ in $\mathrm{int}(\Omega)$, separated from a neighbourhood of
+the entire original filling, including every component of $S$. There
+are bounded smooth open sets $F_j$ in the filled manifold such that
+$$
+1_{F_j}\to 1_{F_N}\text{ in }L^1,
+\qquad
+P_g(F_j)\to P_g(F_N),
+$$
+and each $F_j$ contains a fixed neighbourhood of the whole filling. One
+constructs such $F_j$ by smoothing $1_{F_N}$ in finitely many charts
+away from the two constant regions; coarea yields a regular level
+$t_j$ with the stated perimeter limit and $L^1$ convergence.
+
+For each $j$ retain the component of
+$\mathcal M_N\setminus\overline{F_j}$ that contains the distant end,
+and let $D_j$ be its closure. Retaining this exterior component merely
+selects some entire components of a compact smooth boundary; it
+preserves smoothness and only decreases the full boundary area. Since
+$F_j$ contains a neighbourhood of the whole filling, $D_j\subset\Omega$,
+its interior lies in $\mathrm{int}(\Omega)$, and $D_j$ is closed and
+connected with compact smooth intrinsic boundary $\Gamma_j$. It is an
+admissible full cut, so $a\le\mathrm{Area}_g(\Gamma_j)\le P_g(F_j)$.
+Taking the limit gives $P_g(F_N)\ge a$.
+
+Metric comparison. Since $\widetilde g_N\ge e^{-2\epsilon}g$ on
+$\Omega$, comparison on every approximate tangent $k$-plane yields
+$$
+\mathcal H^k_{\widetilde g_N}(\Sigma_N) = P_{\widetilde g_N}(F_N)
+\ge e^{-k\epsilon}P_g(F_N)\ge e^{-k\epsilon}a.
+$$
+The first equality uses the regularity theorem and the zero
+measure of the singular set.
+
+The Riemannian endpoint. The exterior $X_N$ is a complete
+enclosing-set exterior, its compact frontier $\Sigma_N$ included, with
+metric $\widetilde g_N$ smooth through the frontier. The single
+Euclidean end has compact complement and decay
+$\widetilde g_N-\delta = O_2(r^{2-n})$ with $n-2>(n-2)/2$. Its scalar
+curvature is nonnegative and integrable, with pointwise decay
+$O(r^{-n-\beta})$ for $\beta>0$. The whole frontier is outer minimizing
+and locally perimeter minimizing as the boundary of the filled side,
+its regular part is a smooth embedded minimal hypersurface, and its
+singular set has dimension at most $n-8$. The following theorem is
+proved in the Riemannian companion.
+
+Riemannian theorem (companion). Let $X^n$ be a connected complete
+enclosing-set exterior, its compact frontier $\Sigma$ included, with
+metric $\gamma$ smooth through that frontier. Suppose there is one
+Euclidean coordinate end with compact complement,
+$\gamma-\delta = O_2(r^{-\tau})$ with $\tau>(n-2)/2$, and
+$R_\gamma\ge 0$, $R_\gamma\in L^1$, $R_\gamma = O(r^{-n-\beta})$ for
+some $\beta>0$. Suppose its full frontier $\Sigma$ is outer minimizing
+and locally perimeter minimizing as the boundary of the filled side,
+allowing compactly supported variations on either side. Its regular
+part is a smooth embedded minimal hypersurface and its singular set
+has Hausdorff dimension at most $n-8$. Then
+$$
+E(\gamma) \ge \frac12
+\left(\frac{\mathcal H^{n-1}_\gamma(\Sigma)}{\omega_{n-1}}\right)^{(n-2)/(n-1)}.
+$$
+The assertion includes zero frontier area and imposes neither a spin
+condition nor connectedness of the frontier.
+
+Prepared energy inequality. Apply the Riemannian theorem to $X_N$:
+$$
+E(\widetilde g_N) \ge \frac12
+\left(\frac{\mathcal H^{n-1}_{\widetilde g_N}(\Sigma_N)}
+   {\omega_{n-1}}\right)^{(n-2)/(n-1)}
+\ge \frac12\left(\frac{e^{-(n-1)\epsilon}A_0}{\omega_{n-1}}\right)^{(n-2)/(n-1)}.
+$$
+Combining with the energy bound,
+$$
+E_0+\mathcal P_N(\ell+\ell^{1/2})
+\ge \frac12\left(\frac{e^{-(n-1)\epsilon}A_0}{\omega_{n-1}}\right)^{(n-2)/(n-1)}.
+$$
+Let $N\to\infty$ first; the exponential decay $\ell=e^{-N\epsilon}$
+dominates the fixed polynomial. Then let $\epsilon\downarrow0$. Neither
+step passes through a metric at $N=\infty$ or at $\epsilon=0$; both
+are limits of numbers. This gives
+$$
+E_0 \ge \frac12\left(\frac{A_0}{\omega_{n-1}}\right)^{(n-2)/(n-1)},
+$$
+which is the prepared energy inequality. Since $A_0>0$, it also gives
+$E_0>0$.
+
+Reduction to the main theorem. Let $(\Omega, g, K)$ satisfy the
+main-theorem hypotheses: a smooth one-ended exterior with
+$g-\delta=O_6(r^{-q})$ and $K=O_5(r^{-1-q})$ for
+$(n-2)/2<q<n-2$, integrable constraint densities satisfying DEC,
+finite ADM limits $(E,P)$ with $E>|P|_\delta$, and weakly future
+trapped boundary. Let $a=A_{min,g}(S)>0$ and $m=(E^2-|P|_\delta^2)^{1/2}$.
+
+The end-preparation reduction, applied with an arbitrary sequence of
+gluing radii $L_j\to\infty$, gives prepared data $(g_j,K_j)$ on
+$\Omega$ with $P_j=0$, $E_j\to m$, and
+$A_{min,g_j}(S)\ge(1+\epsilon_j)^{-(n-1)/2}a$ for some
+$\epsilon_j\to 0$. Each datum is a finite object; no bound uniform
+in the gluing parameter is needed.
+
+Apply the prepared energy inequality to $(g_j,K_j)$:
+$$
+E_j \ge \frac12\left(\frac{A_{min,g_j}(S)}{\omega_{n-1}}\right)^{(n-2)/(n-1)}.
+$$
+Substituting the two limits gives
+$$
+m+o(1) \ge \frac12\left(\frac{(1-\epsilon_j)a}{\omega_{n-1}}\right)^{(n-2)/(n-1)}.
+$$
+Letting $j\to\infty$ proves the main theorem.
+
+The positive-shell lemma. Let $(\Omega,g,K)$ be a smooth one-ended
+exterior with $g-\delta=O_2(r^{-q})$, $K=O_1(r^{-1-q})$ for some
+$q>(n-2)/2$, integrable constraints, finite ADM limits, DEC, weakly
+future trapped boundary, and $A_{min,g}(S)>0$. For every $\lambda>0$,
+there is a smooth positive function $U_\lambda\ge 1$, constant near
+$S$, such that
+$$
+g_\lambda = U_\lambda^{4/(n-2)}g,
+\qquad
+K_\lambda = U_\lambda^{2/(n-2)}K,
+$$
+are complete one-ended data with the same decay rate $q$, integrable
+constraints, DEC, weakly future trapped boundary, and
+$$
+E_\lambda = E + 2\lambda,
+\qquad
+P_\lambda = P,
+\qquad
+A_{min,g_\lambda}(S)\ge A_{min,g}(S)>0.
+$$
+
+Proof of positive-shell lemma. Choose $0<\xi<\min(q,1)$ and set
+$T(r) = r^{-(n-2)}-r^{-(n-2)-\xi}$. At a sufficiently large radius
+$r_0$,
+$$
+T' < 0,
+\qquad
+-\Delta_g T - |K|_g|dT|_g = \xi(n-2+\xi)r^{-n-\xi} + O(r^{-n-q})>0.
+$$
+Take a nondecreasing smooth switch $\beta$ from zero to one in this
+region, constant near its endpoints. Define
+$$
+G(r) = -\int_r^\infty \beta(s)T'(s)\,ds.
+$$
+Extend $G$ constantly through the interior. Then $G\ge 0$, $G=T$
+sufficiently far out, and
+$$
+-\Delta_g G - |K|_g|dG|_g
+= \beta(-\Delta_g T - |K|_g|dT|_g) - \beta' T'|dr|_g^2 \ge 0.
+$$
+Set $U=1+\lambda G$. Then $U\ge 1$ is smooth, positive, and equals 1 on
+a neighbourhood of $S$ (because $G=0$ there). The conformal identity
+with $u=U^{2/(n-2)}$ gives
+$$
+u^2(\mu_u-|J_u|_{g_u})
+\ge \mu-|J|_g + \frac{2(n-1)}{(n-2)U}(-\Delta_g U - |K|_g|dU|_g)
+\ge \mu-|J|_g.
+$$
+So DEC is preserved. On the boundary, $\partial_\nu U=0$ since $G=0$
+near $S$, so the future expansion $\theta_+$ keeps its sign.
+
+Energy. Direct computation gives
+$E_\lambda = E + 2\lambda$ exactly, where the coefficient is the
+$(n-1)/(n-2)$ times the flux of $dG$ at infinity, times the $(n-2)$
+exponent from $u^{4/(n-2)}$; the two derivative factors combine to the
+stated slope. Momentum. $\pi_{g_U}=\pi_g+O(r^{-q})$ on the far end so
+$P_\lambda = P$.
+
+Area. $g_\lambda \ge g$ everywhere, and every full enclosing cut grows
+in area. So $A_{min,g_\lambda}(S)\ge A_{min,g}(S)>0$.
+
+The automatic-timelike corollary. Let $(\Omega,g,K)$ satisfy the
+main-theorem hypotheses with the strong decay class, except possibly
+$E>|P|_\delta$. Then $E>|P|_\delta$, and the main inequality holds.
+
+Proof. Suppose $E\le|P|_\delta$. Choose $\epsilon_j>0$ with
+$\epsilon_j\to 0$, and set
+$\lambda_j=(|P|_\delta+\epsilon_j-E)/2>0$. The positive-shell lemma
+applied with $\lambda_j$ gives, for each $j$, data satisfying the same
+hypotheses with
+$$
+E_j=|P|_\delta+\epsilon_j, \quad P_j=P, \quad
+A_{min,g_j}(S)\ge A_{min,g}(S)>0.
+$$
+Since $E_j>|P_j|_\delta$, the main theorem applies:
+$$
+\sqrt{(|P|_\delta+\epsilon_j)^2-|P|_\delta^2}
+\ge \frac12\left(\frac{A_{min,g}(S)}{\omega_{n-1}}\right)^{(n-2)/(n-1)}.
+$$
+The left side tends to zero as $\epsilon_j\to 0$, while the right side
+is a fixed positive number. Contradiction. So $E>|P|_\delta$, and the
+main theorem applies to the original data. QED.
+
+Weak decay in three dimensions. Weaken the strong-decay class to
+$$
+g-\delta = O_2(r^{-q}),
+\qquad
+K = O_1(r^{-1-q}),
+\qquad
+q>\tfrac12,
+$$
+in dimension $n=3$. Then $\omega_2=4\pi$ and the target inequality is
+$\sqrt{E^2-|P|_\delta^2}\ge\sqrt{a_g(B)/(16\pi)}$.
+
+The three-dimensional theorem. Let $(\Omega,g,K)$ be a smooth
+one-ended three-dimensional exterior with the above decay, integrable
+$\mu,|J|_g$, finite ADM limits, DEC, and $\theta_+(B)\le 0$ on every
+boundary component. Then $A_{min,g}(B)>0$, $E>|P|_\delta$, and
+$$
+\sqrt{E^2-|P|_\delta^2}\ge\sqrt{a_g(B)/(16\pi)}.
+$$
+
+Two extra facts are required compared with the strong-decay proof:
+a uniform positive lower bound on enclosing area, and full
+convergence of the enclosing infima along the reduction.
+
+Positive enclosing-area lemma. Let $(\Omega,g)$ be a smooth
+three-dimensional exterior, complete with its nonempty compact
+$B$ included, one asymptotically flat end with compact complement.
+Then $0<a_g(B)<\infty$.
+
+Proof. A sufficiently large coordinate sphere surrounds $B$, so the
+infimum is finite. For a lower bound, take a smooth proper embedded
+ray starting at a point of $B$, initially normal to $B$ into $\Omega$
+and straight in the asymptotic chart. A tubular neighborhood carries
+product coordinates $(t,y)\in[0,\infty)\times D^2$, with the initial
+disk $D^2$ contained in $B$. Choose a smooth two-form $\eta$ compactly
+supported in $\mathrm{int}\,D^2$ with $\int_{D^2}\eta = 1$. Its pull
+back through the projection extends by zero across the lateral
+boundary, giving a smooth closed two-form $\omega$ with bounded
+comass
+$$
+C_\omega = \sup_x\sup_{v,w\text{ orthonormal}}|\omega_x(v,w)|<\infty.
+$$
+Orient $D^2$ so that the flux through any enclosing cut is $+1$.
+For a smooth enclosing cut $\Gamma$ oriented toward the end, Stokes on
+the region between $\Gamma$ and a large sphere $S_R$ gives
+$\int_\Gamma\omega = 1$, so $1\le C_\omega\mathrm{Area}_g(\Gamma)$.
+For a cut coinciding with $B$, push it into $\mathrm{int}\,\Omega$ by
+a collar flow; areas converge to the coincident-cut area. Taking the
+infimum gives $a_g(B)\ge C_\omega^{-1}>0$.
+
+Conformal constraint formulas in $n=3$. For $g_f = e^{4f}g$ and
+$K_f = e^{2f}K$,
+$$
+16\pi e^{4f}\mu_f = 16\pi\mu - 8\Delta_g f - 8|df|_g^2,
+\qquad
+8\pi J_f = e^{-2f}(8\pi J + 4K(\nabla f,\cdot)).
+$$
+Hence
+$$
+16\pi e^{4f}(\mu_f-|J_f|_{g_f})
+\ge 16\pi(\mu-|J|_g) + 8(-\Delta_g f - |df|_g^2 - |K|_g|df|_g).
+$$
+On $B$, $\theta_{+,f} = e^{-2f}(\theta_+ + 4\partial_\nu f)$.
+
+Drift Poisson. For a smooth nonnegative $b\ge |K|_g$ and a smooth
+positive weight $w$, the equation
+$$
+-\Delta_g\phi = b\sqrt{|d\phi|_g^2+\zeta^2} + w,
+\qquad \partial_\nu\phi = -1,
+\qquad \phi\to 0,
+$$
+with $\zeta$ a smooth positive regularizer, has a smooth nonnegative
+solution $\phi = O_2(r^{-1})$ and a finite negative flux
+$\mathfrak f_\phi<0$. Completeness with $B$ included is preserved by
+$u_s = 1+s\phi$ for every fixed $s\ge 0$; the transformed data
+$g_s = u_s^4 g$, $K_s = u_s^2 K$ satisfy
+$$
+16\pi u_s^4(\mu_s-|J_s|_{g_s})
+\ge 16\pi(\mu-|J|_g) + 8s u_s^{-1} w,
+\qquad
+\theta_{+,s} = u_s^{-2}(\theta_+ - 4s u_s^{-1}).
+$$
+For large $s$, both the DEC margin and the trapping sign become
+strict. The constants $s$ and $b$ are fixed after choosing the
+source $w$.
+
+Compactness of enclosing infima. Let $g_j$ be smooth metrics on the
+fixed exterior $\Omega$, converging to $g$ locally uniformly up to $B$.
+Suppose
+$$
+c\delta\le g_j \le C\delta,
+\qquad
+r|\partial g_j|_\delta \le C
+$$
+outside a fixed compact set, with constants independent of $j$. Suppose
+each $g_j$ has an asymptotically flat end and admits a fixed compact
+enclosing competitor with uniformly bounded area. Then minimizing
+filled enclosures exist, their boundaries lie in one fixed compact set,
+and $a_{g_j}(B)\to a_g(B)$.
+
+Proof. On the fixed compact region, uniform comparability makes the
+minimizing sets stay bounded with uniformly bounded perimeters.
+BV compactness yields a limit enclosure containing the obstacle, and
+lower semicontinuity gives $\liminf a_{g_j}(B)\ge a_g(B)$. Conversely,
+a fixed enclosing competitor with $g$-area within $\epsilon$ of
+$a_g(B)$ gives $a_{g_j}(B)\le a_g(B)+\epsilon$ for large $j$. So
+convergence holds.
+
+Rest-end replacement. Apply the annular replacement construction from
+the end-preparation section in dimension three, using only the
+$O_2/O_1$ decay. The compact inverse on the fixed annulus gains a
+$W^{3,p}$ metric and $W^{2,p}$ tensor correction of size $O(R^{-\beta})$
+for the tightest exponent $\beta\in(0,\min(1,q))$. The resulting
+deficit on the transition annulus is $o(R^{-3})$.
+
+Bend to rest. In the vacuum region, bend the boosted Tangherlini slice
+to a static one. The transition geometry satisfies
+$$
+c\delta\le g_R^0\le C\delta,
+\qquad
+|\partial g_R^0|+|K_R^0| \le C/r,
+$$
+with $c,C>0$ independent of $R$, and the far tail is the exact
+rest Schwarzschild slice with mass $m_R\to m$ and momentum zero.
+
+Radial repair. A positive conformal factor $u_R$ with
+$\|u_R-1\|_\infty = o(R^{-1})$ and $|du_R|_\delta = o(R^{-2})$ on the
+transition region. Setting $g_R = u_R^4 g_R^0$ and $K_R = u_R^2 K_R^0$
+yields DEC and weakly future trapped boundary on $\Omega$, with the
+far tail exactly Schwarzschild of mass $m+o(1)$ and $K=0$.
+
+Comparison. Composing the annular glue, the bend, and the radial
+dilation gives a proper diffeomorphism $\Phi_R:\Omega\to\Omega$ with
+$$
+\Phi_R^*g \le (1+\epsilon_R) g_R,
+\qquad
+\epsilon_R\to 0.
+$$
+Every full enclosing cut of $g_R$ transfers under this proper
+diffeomorphism to a full enclosing cut of $g$ of $g$-area at most
+$(1+\epsilon_R)^{3/2}$ times its $g_R$-area. Hence
+$$
+a_g(B) \le (1+\epsilon_R)^{3/2} a_{g_R}(B).
+$$
+
+Strictification. For each fixed $R$ the drift-Poisson lemma from the
+$n=3$ section applies to the compactly supported $B_R\ge|K_R|_{g_R}$.
+Choose the source $w_R$ smooth positive and supported in an annulus,
+chosen so that the resulting flux $\mathfrak f_{\phi,R}$ is negative
+of size at most a fixed constant. Solving
+$-\Delta_{g_R}\phi_R = B_R\sqrt{|d\phi_R|_{g_R}^2+\zeta_R^2}+w_R$ with
+$\partial_\nu\phi_R=-1$ and $\phi_R\to0$ gives a smooth positive
+solution. For small $s$ the transformed data
+$(e^{2s\phi_R}g_R, e^{s\phi_R}K_R)$ have strict DEC, strict trapping,
+compactly supported second fundamental form, momentum zero, and
+$$
+E_{R,s} = m+o(1),
+\qquad
+A_{min}(g_{R,s})\ge (1-o(1)) a_g(B).
+$$
+
+Reduction to prepared data. Each $(g_{R,s},K_{R,s})$ is prepared:
+compact support of $K_{R,s}$, strict DEC, strict trapping, rest end,
+$A_{min}>0$. The prepared energy inequality applies at each fixed
+$(R,s)$:
+$$
+E_{R,s}\ge \frac12\left(\frac{A_{min}(g_{R,s})}{\omega_2}\right)^{1/2}
+= \sqrt{\frac{A_{min}(g_{R,s})}{16\pi}}.
+$$
+Substituting the two limits,
+$$
+m+o(1) \ge \sqrt{\frac{(1-\epsilon_R) a_g(B)}{16\pi}}.
+$$
+Letting $R\to\infty$ and then $s\to 0$, both numbers:
+$$
+m \ge \sqrt{a_g(B)/(16\pi)}.
+$$
+
+Future timelikeness. If $E\le |P|_\delta$, the positive-shell lemma
+applied at scale $\lambda_j = (|P|_\delta+\epsilon_j-E)/2$ produces,
+for each $j$, data on the same exterior satisfying the same three-
+dimensional decay class, integrable constraints, DEC, weak trapping,
+and $A_{min}>0$, with
+$$
+E_j = |P|_\delta+\epsilon_j,
+\qquad
+P_j = P,
+\qquad
+A_{min}(g_j)\ge A_{min}(g)>0.
+$$
+The main theorem applies at each $j$ (since $E_j>|P_j|_\delta$) and
+gives
+$$
+\sqrt{(|P|_\delta+\epsilon_j)^2-|P|_\delta^2}\ge
+\sqrt{A_{min}(g)/(16\pi)}.
+$$
+The left side tends to zero; the right side is a positive constant.
+Contradiction. So $E>|P|_\delta$ and the inequality holds for the
+original data. QED.
+
+Weak decay in four dimensions. In dimension $n=4$, weaken the decay
+to
+$$
+g-\delta = O_2(r^{-q}),
+\qquad
+K = O_1(r^{-1-q}),
+\qquad
+q>1.
+$$
+Then $\omega_3=2\pi^2$, and the target inequality is
+$$
+E \ge \sqrt{|P|_\delta^2 + \frac14\left(\frac{A_{min}(S)}{\omega_3}\right)^{4/3}}.
+$$
+
+The four-dimensional theorem. Let $(M^4,g,K)$ be smooth, connected,
+orientable, with $S=\partial M$ nonempty compact smooth, complete with
+$S$ included. Suppose that outside a compact set $M$ has finitely many
+and at least one ends diffeomorphic to the exterior of a closed ball,
+that $\mu\ge|J|_g$ with $\mu,|J|_g\in L^1$, that finite ADM limits
+$(E_e,P_e)$ exist at each end $e$, that the decay holds at every end
+with a common $q>1$, and that $\Theta_K\le 0$ on every boundary
+component. Then for each end $e$,
+$$
+E_e\ge\sqrt{|P_e|_\delta^2
++\frac14\left(\frac{A_e(S)}{\omega_3}\right)^{4/3}},
+\qquad
+\sqrt{E_e^2-|P_e|_\delta^2}\ge
+\frac12\left(\frac{A_e(S)}{\omega_3}\right)^{2/3}.
+$$
+In particular $E_e>|P_e|_\delta$ and $A_e(S)>0$.
+
+Momentum flux existence. On a smooth dim-4 asymptotically flat end,
+suppose $g-\delta=O_2(r^{-q})$ and $K=O_1(r^{-1-q})$ for $q>1$. If
+$|J|_g$ is integrable, then every ADM momentum limit exists and is
+finite.
+
+Proof. Set $\pi^g_{ij} = K_{ij}-(\mathrm{tr}_gK)g_{ij}$ and
+$\pi^\delta_{ij} = K_{ij}-(\mathrm{tr}_\delta K)\delta_{ij}$. The decay
+gives
+$$
+\pi^g-\pi^\delta = O_1(r^{-1-2q}),
+\qquad
+\partial_j\pi^\delta_{ij}-J_i = O(r^{-2-2q}).
+$$
+The second bound: expanding the covariant divergence of $\pi^g$ in
+coordinates, replacing $g$ by $\delta$, and using the connection
+bounds $|\Gamma|=O(r^{-1-q})$ everywhere. Since $q>1$, the radial
+volume weight is $O(r^{1-2q})$, integrable. Euclidean divergence
+theorem on an annular shell gives
+$$
+\int_{S_{R'}}\pi^\delta_{ij}n^j\,dA
+-\int_{S_R}\pi^\delta_{ij}n^j\,dA
+=\int_{R<r<R'}\partial_j\pi^\delta_{ij}\,dV
+= O(R^{2-2q}),
+$$
+and the left side is Cauchy. The difference between this flux and
+$\int_{S_R}\pi^g_{ij}n^j\,dA$ is $O(R^{2-2q})\to 0$ since $2q>2$.
+Thus the tensor appearing in the ADM momentum tends to a finite
+limit.
+
+Cut reduction. Fix the selected end $e$. Truncate every other end at
+a large sphere $S_{R_i}$, chosen so that
+$$
+H_{S_{R_i}}+\mathrm{tr}_{S_{R_i}}K = -3/R_i+O(R_i^{2-2q})<0.
+$$
+The modified manifold $M_e$ is complete, connected, one-ended, with
+$\partial M_e$ containing both the original $S$ and the added spheres
+$S_{R_i}$. Its DEC and integrability come from the original. Every
+component of $\partial M_e$ is weakly future trapped for the normal
+into $M_e$. The selected-end ADM data is $(E_e,P_e)$, unchanged.
+Full enclosing cuts of $M_e$ embed into full enclosing cuts of $M$,
+so $A_e(M_e)\ge A_e(S)$ and $A_e(S)>0$.
+
+The one-ended case. For a one-ended $n=4$ exterior with only
+$O_2/O_1$ decay and $q>1$, the reduction to a sequence of prepared
+data uses two weaker tools.
+
+Rest-end replacement. Match the original $(E,P)$ with the boosted
+model end via the annular correction, but replace the compact-inverse
+step by a single strongly elliptic solve on the fixed annulus.
+Compact supported metric deviations $h$ of size $O(R^{-q})$ and tensor
+corrections $\pi$ of size $O(R^{-1-q})$ feed into the linearized
+constraint operator; a single Dirichlet solve (with appropriate
+normalization) on the fixed annulus absorbs them with support in a
+larger annulus. The needed well-posedness follows from convexity of
+the constraint operator on the annulus and the small size of the
+corrections. Resulting deficit on the transition annulus is
+$O(R^{-2q})$, integrable against $r^3dr$ since $2q>2$.
+
+Bend and repair. In the vacuum region, bend to a rest slice. The
+transition geometry satisfies
+$$
+c\delta \le g_R^0 \le C\delta,
+\qquad
+|\partial g_R^0|+|K_R^0|\le C/r,
+$$
+and the far tail is exact rest Schwarzschild of mass $m+o(1)$. A
+positive radial conformal factor $u_R\ge 1$ with
+$\|u_R-1\|_\infty=o(R^{-1})$ and $|du_R|_\delta=o(R^{-2})$ repairs
+the residual DEC deficit, giving
+$$
+g_R=u_R^2g_R^0,
+\qquad
+K_R=u_RK_R^0,
+$$
+with $R_{g_R}\ge 0$ and $K_R=0$ on the tail.
+
+Comparison. Composing the annular glue, the bend, and the radial
+factor gives a proper diffeomorphism $\Phi_R:M_e\to M_e$ with
+$$
+\Phi_R^*g \le (1+\epsilon_R)g_R,
+\qquad
+\epsilon_R\to 0,
+$$
+and hence
+$$
+A_e(S) \le (1+\epsilon_R)^{3/2} A_{min}(g_R).
+$$
+
+Strictification. For each fixed $R$, the drift-Poisson lemma from the
+strong-decay section applies because $K_R$ is compactly supported.
+Choosing source $w_R$ with negative flux of size at most a fixed
+constant gives, for small $s>0$, strict DEC, strict trapping, compact
+support, momentum zero, and energy within $o(1)$ of $m$. The
+transformed data are prepared.
+
+Prepared inequality. Apply the prepared theorem (dim $n=4$) at each
+fixed $(R,s)$:
+$$
+E_{R,s}\ge\frac12\left(\frac{A_{R,s}}{\omega_3}\right)^{2/3}.
+$$
+Substituting lower area and energy,
+$$
+m+o(1)\ge\frac12\left(\frac{(1-\epsilon_R)A_e(S)}{\omega_3}\right)^{2/3}.
+$$
+Letting $R\to\infty$, then $s\to0$:
+$$
+m\ge\frac12\left(\frac{A_e(S)}{\omega_3}\right)^{2/3}.
+$$
+Squaring:
+$$
+E_e^2-|P_e|_\delta^2\ge\frac14\left(\frac{A_e(S)}{\omega_3}\right)^{4/3},
+$$
+which is the four-dimensional theorem's square-root form.
+
+Future timelikeness. If $E_e\le |P_e|_\delta$, the positive-shell
+lemma in $n=4$ raises the energy at each end, keeping the momentum
+and not shrinking $A_e(S)$. The main theorem applies at each shell
+(since the raised energy exceeds $|P_e|_\delta$), giving
+$$
+\sqrt{(|P_e|_\delta+\epsilon_j)^2-|P_e|_\delta^2}\ge
+\frac12\left(\frac{A_e(S)}{\omega_3}\right)^{2/3}.
+$$
+The left side tends to zero as $\epsilon_j\to0$; the right side is a
+fixed positive number. Contradiction. QED.
+
+This completes the proof of the main theorem, its automatic-
+timelikeness corollary, and the weaker-decay theorems in dimensions
+three and four. Sharpness of the constants is attained by the
+time-symmetric Schwarzschild--Tangherlini exterior, whose boundary is
+the least full enclosing cut. The argument has nowhere used equality
+in any of its inputs: each stage, from end preparation through the
+filled Jang system and the Riemannian endpoint, is an inequality in
+the strict direction, and the small parameters $\epsilon$, $\ell$,
+and $\delta$ tend to zero as numbers.