Conjecture Atlas

Proof source

Sharp binary rooftop energy obstruction in original smooth backgrounds

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

    The result below concerns a proposed sufficient estimate, not the literal continuity question itself. It constructs one allowed original smooth strongly positive background and bounded actual potentials for which e(P(u,v))>e(u)+e(v)e(P(u,v))>e(u)+e(v). It also proves that the uniform binary constant 22 is sharp in degree one across the original background class. This does not exclude constants depending on a fixed background, direct finite-family bounds, or C1.

    Original geometry and frozen inputs.

    Keep the prescribed finite extension K/QpK/\mathbf Q_p, its unchanged absolute value, a uniformizer π\pi, and λ=−log⁡∣π∣>0\lambda=-\log|\pi|>0. Set X=(Gm,K2)an,t=(log⁡∣T1∣,log⁡∣T2∣),Q=(−λ,λ)2,V=t−1(Q),X=(\mathbf G_{m,K}^2)^{\rm an},\quad t=(\log|T_1|,\log|T_2|),\quad Q=(-\lambda,\lambda)^2, \quad V=t^{-1}(Q), βt=d′t1∧d′′t1+d′t2∧d′′t2,S0=βt∧[X].\beta_t=d't_1\wedge d''t_1+d't_2\wedge d''t_2, \qquad S_0=\beta_t\wedge[X]. The closed box lies in the ordinary interior of the larger original Laurent affinoid t−1([−2λ,2λ]2)t^{-1}([-2\lambda,2\lambda]^2). All backgrounds below extend to its ordinary neighborhood. We use the unchanged relative strong completion with its encoded approximation boundary, and retain the full original test sheaf. Here n=2n=2, p=q=1p=q=1 and d=2d=2. Frozen R17 FP1–FP6 supplies canonical finite convex Gauss traces, their completed linear moment identity, and compact toric test pairings for S0S_0. IF20–IF21 supplies full original toric calibration and quadratic energy equality for bounded Lipschitz convex zero-boundary potentials. The actual-core admission argument following IF20 supplies finite mass, compact negative sublevels and canonical no loss for these neighborhood-extendible functions. Frozen HC14–HC15 places every original compact smooth scalar in the actual difference space. Finite relative rooftops exist and have their frozen greatestness property.

    Lemma (Identification of a bounded polyhedral toric rooftop). Let f1,f2f_1,f_2 be bounded Lipschitz convex functions on Q0=(−1,1)2Q_0=(-1,1)^2, continuous and zero on its boundary, negative inside, with convex neighborhood extensions. Suppose their real convex envelope h=conv⁡(min⁡(f1,f2))h=\operatorname{conv}(\min(f_1,f_2)) has the same properties. Set Fi=fi(t/λ)F_i=f_i(t/\lambda) and H=h(t/λ)H=h(t/\lambda). Then the completed relative rooftop for S0S_0 equals HH in the original relative quotient.

    Proof. The IF20 actual-core argument admits all three functions. Their bounded convex neighborhood extensions have finite Laplacian mass, compact strict negative sublevels, finite energies and eventually unchanged deep truncations. Thus H⪯P:=PS0(F1,F2)⪯FiH\preceq P:=P_{S_0}(F_1,F_2)\preceq F_i. To justify trace order without evaluating a relative class at exceptional Gauss points, replace any Borel representative of PP by its pointwise clipping between HH and min⁡(F1,F2)\min(F_1,F_2). This agrees with PP outside the fixed union of allowed relative null sets and hence represents the same completed member. It satisfies the two order bounds everywhere. The canonical convex trace gPg_P agrees with this representative on the Gauss section almost everywhere. Therefore h(z)≤gP(λz)≤min⁡(f1(z),f2(z))h(z)\le g_P(\lambda z)\le\min(f_1(z),f_2(z)) almost everywhere, and hence everywhere by convex continuity. Greatestness of the real convex envelope forces gP(λz)=h(z)g_P(\lambda z)=h(z).

    This trace equality alone is not used as an injectivity assertion. For every compact smooth real φ\varphi on QQ, frozen FP6 gives BS0(P−H,φ∘t)=−∫Q(gP−h(t/λ))Δφ dt=0.(1)B_{S_0}(P-H,\varphi\circ t) =-\int_Q(g_P-h(t/\lambda))\Delta\varphi\,dt=0. \tag{1} Choose compact smooth φk\varphi_k converging to h(t/λ)h(t/\lambda) in H01(Q)H_0^1(Q). Such approximation applies because hh is Lipschitz and zero on the rectangle boundary. The compact pullbacks are actual energy differences. Expanding their difference energy by FP6 and IF21 gives BS0(φk∘t−H,φk∘t−H)=∫Q∣∇φk−∇(h(t/λ))∣2dt⟶0.(2)B_{S_0}(\varphi_k\circ t-H,\varphi_k\circ t-H) =\int_Q|\nabla\varphi_k-\nabla(h(t/\lambda))|^2dt \longrightarrow0. \tag{2} Quadratic energy Cauchy–Schwarz and (1) imply BS0(P−H,H)=0B_{S_0}(P-H,H)=0. Consequently eS0(P)=eS0(H)+IS0(P,H).e_{S_0}(P)=e_{S_0}(H)+I_{S_0}(P,H). But H⪯PH\preceq P gives eS0(P)≤eS0(H)e_{S_0}(P)\le e_{S_0}(H) by frozen energy monotonicity. Thus IS0(P,H)=0I_{S_0}(P,H)=0 and the completed potentials are the same relative member. No identification from trace equality alone, and no restriction of the original test family, has occurred. ◻

    Lemma (Uniformly comparable smooth backgrounds). Let S=A(t)∧[X]S=A(t)\wedge[X] be smooth, separately closed and positive on a neighborhood of the closed box, with constants 0<c≤C<∞0<c\le C<\infty such that cβt≤A≤Cβt.(3)c\beta_t\le A\le C\beta_t. \tag{3} Then its degree-one strong cone, relative null ideal and relative order identify canonically with those for S0S_0. In particular their relative rooftops agree.

    Proof. The degree-one source of each fixed normalized tester is linear in the background. Applying (3) to every tester, then taking suprema and ordinary-open outer regularizations, gives two-sided capacity comparisons and hence equality of the null ideals. The same source comparisons for bounded actual psh functions give equivalence of finite mass and power-energy bounds. Compact negative sublevels and the ordinary boundary conditions are unchanged. The canonical no-loss estimates for the truncations transfer by the same fixed constants, so the bounded actual defining cores coincide with their printed qualifications.

    For every difference a−ba-b of members of this common actual core, global Green exchange and positivity of the quadratic energy forms for S−cS0S-cS_0 and CS0−SCS_0-S give cIS0(a,b)≤IS(a,b)≤CIS0(a,b).(4)c I_{S_0}(a,b)\le I_S(a,b)\le C I_{S_0}(a,b). \tag{4} Both residual backgrounds remain original positive separately closed backgrounds. Thus the strong Cauchy defining rows and their equivalences are the same. The capacity comparisons identify their limiting relative Borel realizations. This proves equality of the completed cones with their encoded approximation boundary, rather than imposing an ordinary boundary value on an arbitrary completed member. Their relative orders and admissible lower families are consequently identical, so greatestness identifies their rooftops. ◻

    Theorem. In the original degree-one scope q=1q=1, the universal binary estimate e(P(U,V))≤2(e(U)+e(V))e(P(U,V))\le2\bigl(e(U)+e(V)\bigr) has optimal constant 22, even when the backgrounds are smooth strongly positive and the two obstacles belong to the bounded actual core. Precisely, for every 0<C<20<C<2 there are one allowed original smooth background and bounded actual obstacles with e(P(U,V))>C(e(U)+e(V)).e(P(U,V))>C\bigl(e(U)+e(V)\bigr). The background is allowed to depend on CC. No assertion of sharpness for a single fixed smooth background is made.

    Proof. The upper bound follows from the frozen rooftop root inequality r(P(U,V))≤r(U)+r(V)r(P(U,V))\le r(U)+r(V) and (r(U)+r(V))2≤2(r(U)2+r(V)2)(r(U)+r(V))^2\le2(r(U)^2+r(V)^2), since d=q+1=2d=q+1=2.

    For sharpness fix 0<a<10<a<1, and work in normalized coordinates (x,y)=t/λ∈Q0=(−1,1)2(x,y)=t/\lambda\in Q_0=(-1,1)^2. Put ua(x,y)=max⁡{x−11−a,−x+11+a,y−11+a,−y+11−a},va(x,y)=ua(y,x).u_a(x,y)=\max\left\{ \frac{x-1}{1-a},-\frac{x+1}{1+a}, \frac{y-1}{1+a},-\frac{y+1}{1-a}\right\}, \qquad v_a(x,y)=u_a(y,x). These are convex pyramids over the four zero-height corners, with apices (a,−a,−1)(a,-a,-1) and (−a,a,−1)(-a,a,-1), respectively. Their common real convex envelope is ha(x,y)=max⁡{∣x∣−11−a,∣y∣−11−a,∣x+y∣2−1}.(5)h_a(x,y)=\max\left\{ \frac{|x|-1}{1-a},\frac{|y|-1}{1-a},\frac{|x+y|}{2}-1 \right\}. \tag{5} To verify this, denote the two planar apices by A=(a,−a)A=(a,-a) and B=(−a,a)B=(-a,a) and the corners by C±±=(±1,±1)C_{\pm\pm}=(\pm1,\pm1). The six affine branches in (5) are lower supporting planes through the triples (C−−,C−+,B), (C−−,C+−,A), (C−+,C++,B), (C+−,C++,A),(C_{--},C_{-+},B),\ (C_{--},C_{+-},A),\ (C_{-+},C_{++},B),\ (C_{+-},C_{++},A), (C−−,A,B), (C++,A,B).(C_{--},A,B),\ (C_{++},A,B). They are below all six vertex heights and their projected triangles tile the square. Affine interpolation on each triangle proves both the common-minorant inequality and greatestness, by the same elementary supporting-plane argument. All three functions are globally polyhedral convex, negative inside the square, Lipschitz, continuous and zero on its boundary. Their original moment pullbacks are therefore qualified bounded actual core potentials.

    On the diagonal one has the exact profiles ua(s,s)=va(s,s)=∣s∣−11+a,ha(s,s)=∣s∣−1.u_a(s,s)=v_a(s,s)=\frac{|s|-1}{1+a},\qquad h_a(s,s)=|s|-1. Consequently, for Eℓ(f)=∫−11∣(d/ds)f(s,s)∣2dsE_\ell(f)=\int_{-1}^1|(d/ds)f(s,s)|^2ds, Eℓ(ua)=Eℓ(va)=2(1+a)2,Eℓ(ha)=2,E_\ell(u_a)=E_\ell(v_a)=\frac2{(1+a)^2},\qquad E_\ell(h_a)=2, and Eℓ(ha)Eℓ(ua)+Eℓ(va)=(1+a)22⟶2(a↑1).(6)\frac{E_\ell(h_a)}{E_\ell(u_a)+E_\ell(v_a)} =\frac{(1+a)^2}{2}\longrightarrow2 \quad(a\uparrow1). \tag{6}

    Use the same entirely smooth original backgrounds Sδ,ε=[εβt+λ2D ⁣(((t2−t1)/λ)2+δ2)]∧[X],δ,ε>0.S_{\delta,\varepsilon} =\left[\varepsilon\beta_t+ \lambda^2D\!\left(\sqrt{((t_2-t_1)/\lambda)^2+\delta^2}\right) \right]\wedge[X], \qquad\delta,\varepsilon>0. They satisfy εS0≤Sδ,ε≤(ε+2/δ)S0\varepsilon S_0\le S_{\delta,\varepsilon} \le(\varepsilon+2/\delta)S_0 on the original neighborhood. The preceding lemmas identify the completed rooftop of the two actual obstacles with ha(t/λ)h_a(t/\lambda) for every positive pair of parameters. For f∈{ua,va,ha}f\in\{u_a,v_a,h_a\}, full original torus calibration and global Green exchange give the following exact smooth energy formula. Indeed the background matrix in the tt coordinates is εI+χδ′′(y−x)(1−1−11)\varepsilon I+\chi_\delta''(y-x) \left(\begin{smallmatrix}1&-1\\-1&1\end{smallmatrix}\right); its cofactor is εI+χδ′′(y−x)(1111)\varepsilon I+\chi_\delta''(y-x) \left(\begin{smallmatrix}1&1\\1&1\end{smallmatrix}\right). The factor λ2\lambda^2 in the background definition makes the change t=λ(x,y)t=\lambda(x,y) cancel in the energy integral. Local convex mollification as in IF20 passes the calibration on all original tests to each polyhedral potential; the zero-boundary compact clipping argument of IF21 justifies global Green exchange. All coefficients are bounded and smooth for fixed positive parameters. Thus eδ,ε(f(t/λ))=ε∫Q0∣∇f∣2dxdy+∫Q0δ2(fx+fy)2((y−x)2+δ2)3/2 dxdy.(7)e_{\delta,\varepsilon}(f(t/\lambda)) =\varepsilon\int_{Q_0}|\nabla f|^2dxdy +\int_{Q_0}\frac{\delta^2(f_x+f_y)^2} {((y-x)^2+\delta^2)^{3/2}}\,dxdy. \tag{7} For fixed aa the gradients are bounded. With r=y−xr=y-x, the inner cross-sectional integral of (fx+fy)2(f_x+f_y)^2 is continuous at r=0r=0: away from finitely many diagonal breakpoints the active plane is locally fixed; if a whole interface lies on the diagonal, its adjacent affine functions have the same tangential derivative. Dominated convergence therefore applies to the cross sections. The nonnegative kernel in (7) has total real-line mass 22 and concentrates at zero, so lim⁡δ↓0eδ,ε(f(t/λ))=εE0(f)+2Eℓ(f),E0(f)=∫Q0∣∇f∣2dxdy.(8)\lim_{\delta\downarrow0}e_{\delta,\varepsilon}(f(t/\lambda)) =\varepsilon E_0(f)+2E_\ell(f),\qquad E_0(f)=\int_{Q_0}|\nabla f|^2dxdy. \tag{8} The limiting denominator below is strictly positive, and (6)–(8) give lim⁡ε↓0lim⁡δ↓0eδ,ε(ha(t/λ))eδ,ε(ua(t/λ))+eδ,ε(va(t/λ))=(1+a)22.(9)\lim_{\varepsilon\downarrow0}\lim_{\delta\downarrow0} \frac{e_{\delta,\varepsilon}(h_a(t/\lambda))} {e_{\delta,\varepsilon}(u_a(t/\lambda))+ e_{\delta,\varepsilon}(v_a(t/\lambda))} =\frac{(1+a)^2}{2}. \tag{9} Given C<2C<2, choose a<1a<1 with the right side greater than CC, then choose a sufficiently small positive ε\varepsilon, followed by a sufficiently small positive δ\delta. Fix these parameters. The resulting single original smooth background and two actual obstacles give the required strict inequality. No potential or completion is transferred to a singular limiting background. ◻

    An explicit fixed smooth counterexample.

    The family above contains a particularly simple fixed example. Take a=1/2a=1/2 and ε=δ=1/100\varepsilon=\delta=1/100. Then u=max⁡{2(x−1),−23(x+1),23(y−1),−2(y+1)},v(x,y)=u(y,x),u=\max\{2(x-1),-\tfrac23(x+1),\tfrac23(y-1),-2(y+1)\}, \qquad v(x,y)=u(y,x), h=max⁡{2(∣x∣−1),2(∣y∣−1),12∣x+y∣−1}.h=\max\{2(|x|-1),2(|y|-1),\tfrac12|x+y|-1\}. For the single original smooth background S=S1/100,1/100S=S_{1/100,1/100} defined above, the completed rooftop of U=u(t/λ)U=u(t/\lambda) and V1=v(t/λ)V_1=v(t/\lambda) is H=h(t/λ)H=h(t/\lambda). The background form is bounded between (1/100)βt(1/100)\beta_t and (200+1/100)βt(200+1/100)\beta_t. Its full original smooth energy formula is (7).

    Here is an exact evaluation rather than a numerical or limiting check. For a triangle TT of area ATA_T whose three vertex values of L=y−xL=y-x are aT<bT<cTa_T<b_T<c_T, affine cross-section integration gives ∫Tχδ′′(L)=2ATcT−aT(χδ(cT)−χδ(bT)cT−bT−χδ(bT)−χδ(aT)bT−aT),χδ(r)=r2+δ2.(10)\int_T\chi_\delta''(L) =\frac{2A_T}{c_T-a_T} \left(\frac{\chi_\delta(c_T)-\chi_\delta(b_T)}{c_T-b_T} -\frac{\chi_\delta(b_T)-\chi_\delta(a_T)}{b_T-a_T}\right), \qquad \chi_\delta(r)=\sqrt{r^2+\delta^2}. \tag{10} For a plane f=αx+γy+κf=\alpha x+\gamma y+\kappa on TT, its energy contribution is εAT(α2+γ2)\varepsilon A_T(\alpha^2+\gamma^2) plus (α+γ)2(\alpha+\gamma)^2 times (10). The four triangles joining each apex to its four boundary edges give the obstacle energies; the six triangles in the theorem’s proof give the rooftop energy. All relevant vertex values of LL are distinct. Exact substitution gives

    eS(U)=eS(V1)=−1610001225+225+24000145,eS(H)=−31000125+13100+24000125.\begin{aligned} e_S(U)=e_S(V_1) &=-\frac{16\sqrt{10001}}{225}+\frac2{25} +\frac{2\sqrt{40001}}{45},\\ e_S(H)&=-\frac{3\sqrt{10001}}{25}+\frac{13}{100} +\frac{2\sqrt{40001}}{25}. \end{aligned}

    Consequently eS(H)−eS(U)−eS(V1)=1000145−240001225−3100>10045−2⋅201225−3100=73180>25.(11)e_S(H)-e_S(U)-e_S(V_1) =\frac{\sqrt{10001}}{45} -\frac{2\sqrt{40001}}{225}-\frac3{100} >\frac{100}{45}-\frac{2\cdot201}{225}-\frac3{100} =\frac{73}{180}>\frac25. \tag{11} The strict inequalities use only 10001>100\sqrt{10001}>100 and 40001<201\sqrt{40001}<201. Thus coefficient-one energy subadditivity fails in this one explicit smooth original background, with an exact positive gap.

    Scope.

    The sharpness statement excludes a background-independent universal binary constant strictly below 22 across the original degree-one class. It does not exclude a subcritical constant C(S)<2C(S)<2 for each fixed smooth background, or a direct arbitrary-finite-family estimate e(P(Ui:i∈F))≤C(S,n)∑i∈Fe(Ui)e(P(U_i:i\in F))\le C(S,n)\sum_{i\in F}e(U_i) whose constant is independent of ∣F∣|F|. It does not disprove SR8 or produce a strongly convergent row with literal discontinuity. C1 remains open in the full arbitrary-background scope on the basis of this argument.

    Exact frozen source identity.

    The R17 whole source for FP1–FP6, IF20–IF21, and the actual-core and compact-test qualifications has SHA-256 8389d2ce81aa5fd1b563767a3c207a06d0b50e661fb06ee978da2d39634c49be. The finite relative rooftop definition and greatestness are recorded in frozen R12 LG2, SHA-256 5001a36a41048e2c6c94189dffcec2f0610537d90ddb64e8c0932770c861988d. The completion and positive quadratic energy form remain frozen inputs in their exact original scopes. The explicit smooth energies above were independently recomputed from the printed triangulations with exact arithmetic; the strict final gap uses the displayed rational bounds.

    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 be4285bb60488e262522b78b87392486db30a049e3a26b08d9a4fda1435f53bb

    Frozen-original SHA-256 be4285bb60488e262522b78b87392486db30a049e3a26b08d9a4fda1435f53bb

    \documentclass{article}
    \usepackage{amsmath,amssymb,amsthm}
    \newtheorem{lemma}{Lemma}
    \newtheorem{theorem}{Theorem}
    \begin{document}
    \title{Sharp binary rooftop energy obstruction in original smooth backgrounds}
    \date{}\maketitle
    
    The result below concerns a proposed sufficient estimate, not the literal
    continuity question itself. It constructs one allowed original smooth
    strongly positive background and bounded actual potentials for which
    $e(P(u,v))>e(u)+e(v)$. It also proves that the uniform binary constant $2$ is sharp in degree
    one across the original background class. This does not exclude constants
    depending on a fixed background, direct finite-family bounds, or C1.
    
    \paragraph{Original geometry and frozen inputs.}
    Keep the prescribed finite extension $K/\mathbf Q_p$, its unchanged
    absolute value, a uniformizer $\pi$, and $\lambda=-\log|\pi|>0$. Set
    \[
     X=(\mathbf G_{m,K}^2)^{\rm an},\quad
     t=(\log|T_1|,\log|T_2|),\quad Q=(-\lambda,\lambda)^2,
     \quad V=t^{-1}(Q),
    \]
    \[
     \beta_t=d't_1\wedge d''t_1+d't_2\wedge d''t_2,
     \qquad S_0=\beta_t\wedge[X].
    \]
    The closed box lies in the ordinary interior of the larger original
    Laurent affinoid $t^{-1}([-2\lambda,2\lambda]^2)$. All backgrounds below
    extend to its ordinary neighborhood. We use the unchanged relative
    strong completion with its encoded approximation boundary, and retain
    the full original test sheaf. Here $n=2$, $p=q=1$ and $d=2$.
    Frozen R17 FP1--FP6 supplies canonical finite convex Gauss traces,
    their completed linear moment identity, and compact toric test pairings
    for $S_0$. IF20--IF21 supplies full original toric calibration and
    quadratic energy equality for bounded Lipschitz convex zero-boundary
    potentials. The actual-core admission argument following IF20 supplies
    finite mass, compact negative sublevels and canonical no loss for these
    neighborhood-extendible functions. Frozen HC14--HC15 places every original
    compact smooth scalar in the actual difference space. Finite relative
    rooftops exist and have their frozen greatestness property.
    
    \begin{lemma}[Identification of a bounded polyhedral toric rooftop]
    Let $f_1,f_2$ be bounded Lipschitz convex functions on
    $Q_0=(-1,1)^2$, continuous and zero on its boundary, negative inside,
    with convex neighborhood extensions. Suppose their real convex envelope
    $h=\operatorname{conv}(\min(f_1,f_2))$ has the same properties. Set
    $F_i=f_i(t/\lambda)$ and $H=h(t/\lambda)$. Then the completed relative
    rooftop for $S_0$ equals $H$ in the original relative quotient.
    \end{lemma}
    \begin{proof}
    The IF20 actual-core argument admits all three functions. Their bounded
    convex neighborhood extensions have finite Laplacian mass, compact strict
    negative sublevels, finite energies and eventually unchanged deep
    truncations. Thus $H\preceq P:=P_{S_0}(F_1,F_2)\preceq F_i$.
    To justify trace order without evaluating a relative class at exceptional
    Gauss points, replace any Borel representative of $P$ by its pointwise
    clipping between $H$ and $\min(F_1,F_2)$. This agrees with $P$ outside
    the fixed union of allowed relative null sets and hence represents the
    same completed member. It satisfies the two order bounds everywhere.
    The canonical convex trace $g_P$ agrees with this representative on the
    Gauss section almost everywhere. Therefore
    \[
     h(z)\le g_P(\lambda z)\le\min(f_1(z),f_2(z))
    \]
    almost everywhere, and hence everywhere by convex continuity. Greatestness
    of the real convex envelope forces $g_P(\lambda z)=h(z)$.
    
    This trace equality alone is not used as an injectivity assertion.
    For every compact smooth real $\varphi$ on $Q$, frozen FP6 gives
    \[
     B_{S_0}(P-H,\varphi\circ t)
           =-\int_Q(g_P-h(t/\lambda))\Delta\varphi\,dt=0.       \tag{1}
    \]
    Choose compact smooth $\varphi_k$ converging to $h(t/\lambda)$ in
    $H_0^1(Q)$. Such approximation applies because $h$ is Lipschitz and
    zero on the rectangle boundary. The compact pullbacks are actual energy
    differences. Expanding their difference energy by FP6 and IF21 gives
    \[
     B_{S_0}(\varphi_k\circ t-H,\varphi_k\circ t-H)
           =\int_Q|\nabla\varphi_k-\nabla(h(t/\lambda))|^2dt
              \longrightarrow0.                              \tag{2}
    \]
    Quadratic energy Cauchy--Schwarz and (1) imply $B_{S_0}(P-H,H)=0$.
    Consequently
    \[
     e_{S_0}(P)=e_{S_0}(H)+I_{S_0}(P,H).
    \]
    But $H\preceq P$ gives $e_{S_0}(P)\le e_{S_0}(H)$ by frozen energy
    monotonicity. Thus $I_{S_0}(P,H)=0$ and the completed potentials are
    the same relative member. No identification from trace equality alone,
    and no restriction of the original test family, has occurred.
    \end{proof}
    
    \begin{lemma}[Uniformly comparable smooth backgrounds]
    Let $S=A(t)\wedge[X]$ be smooth, separately closed and positive on a
    neighborhood of the closed box, with constants $0<c\le C<\infty$ such that
    \[
                        c\beta_t\le A\le C\beta_t.             \tag{3}
    \]
    Then its degree-one strong cone, relative null ideal and relative order
    identify canonically with those for $S_0$. In particular their relative
    rooftops agree.
    \end{lemma}
    \begin{proof}
    The degree-one source of each fixed normalized tester is linear in the
    background. Applying (3) to every tester, then taking suprema and ordinary-open
    outer regularizations, gives two-sided capacity comparisons and hence
    equality of the null ideals. The same source
    comparisons for bounded actual psh functions give equivalence of finite
    mass and power-energy bounds. Compact negative sublevels and the ordinary
    boundary conditions are unchanged. The canonical no-loss estimates for
    the truncations transfer by the same fixed constants, so the bounded
    actual defining cores coincide with their printed qualifications.
    
    For every difference $a-b$ of members of this common actual core, global
    Green exchange and positivity of the quadratic energy forms for
    $S-cS_0$ and $CS_0-S$ give
    \[
     c I_{S_0}(a,b)\le I_S(a,b)\le C I_{S_0}(a,b).              \tag{4}
    \]
    Both residual backgrounds remain original positive separately closed
    backgrounds. Thus the strong Cauchy defining rows and their equivalences
    are the same. The capacity comparisons identify their limiting relative
    Borel realizations. This proves equality of the completed cones with
    their encoded approximation boundary, rather than imposing an ordinary
    boundary value on an arbitrary completed member. Their relative orders
    and admissible lower families are consequently identical, so greatestness
    identifies their rooftops.
    \end{proof}
    
    \begin{theorem}
    In the original degree-one scope $q=1$, the universal binary estimate
    \[
                           e(P(U,V))\le2\bigl(e(U)+e(V)\bigr)
    \]
    has optimal constant $2$, even when the backgrounds are smooth strongly
    positive and the two obstacles belong to the bounded actual core.
    Precisely, for every $0<C<2$ there are one allowed original smooth
    background and bounded actual obstacles with
    \[
                           e(P(U,V))>C\bigl(e(U)+e(V)\bigr).
    \]
    The background is allowed to depend on $C$. No assertion of sharpness
    for a single fixed smooth background is made.
    \end{theorem}
    
    \begin{proof}
    The upper bound follows from the frozen rooftop root inequality
    $r(P(U,V))\le r(U)+r(V)$ and
    $(r(U)+r(V))^2\le2(r(U)^2+r(V)^2)$, since $d=q+1=2$.
    
    For sharpness fix $0<a<1$, and work in normalized coordinates
    $(x,y)=t/\lambda\in Q_0=(-1,1)^2$. Put
    \[
     u_a(x,y)=\max\left\{
     \frac{x-1}{1-a},-\frac{x+1}{1+a},
     \frac{y-1}{1+a},-\frac{y+1}{1-a}\right\},
     \qquad v_a(x,y)=u_a(y,x).
    \]
    These are convex pyramids over the four zero-height corners, with apices
    $(a,-a,-1)$ and $(-a,a,-1)$, respectively. Their common real convex
    envelope is
    \[
     h_a(x,y)=\max\left\{
     \frac{|x|-1}{1-a},\frac{|y|-1}{1-a},\frac{|x+y|}{2}-1
                             \right\}.                         \tag{5}
    \]
    To verify this, denote the two planar apices by $A=(a,-a)$ and
    $B=(-a,a)$ and the corners by $C_{\pm\pm}=(\pm1,\pm1)$.
    The six affine branches in (5) are lower supporting planes through
    the triples
    \[
     (C_{--},C_{-+},B),\ (C_{--},C_{+-},A),\
     (C_{-+},C_{++},B),\ (C_{+-},C_{++},A),
    \]
    \[
                       (C_{--},A,B),\ (C_{++},A,B).
    \]
    They are below all six vertex heights and their projected triangles
    tile the square. Affine interpolation on each triangle proves both
    the common-minorant inequality and greatestness, by the same elementary supporting-plane argument. All three functions are
    globally polyhedral convex, negative inside the square, Lipschitz,
    continuous and zero on its boundary. Their original moment pullbacks
    are therefore qualified bounded actual core potentials.
    
    On the diagonal one has the exact profiles
    \[
     u_a(s,s)=v_a(s,s)=\frac{|s|-1}{1+a},\qquad
     h_a(s,s)=|s|-1.
    \]
    Consequently, for $E_\ell(f)=\int_{-1}^1|(d/ds)f(s,s)|^2ds$,
    \[
     E_\ell(u_a)=E_\ell(v_a)=\frac2{(1+a)^2},\qquad
     E_\ell(h_a)=2,
    \]
    and
    \[
     \frac{E_\ell(h_a)}{E_\ell(u_a)+E_\ell(v_a)}
                             =\frac{(1+a)^2}{2}\longrightarrow2
                             \quad(a\uparrow1).                \tag{6}
    \]
    
    Use the same entirely smooth original backgrounds
    \[
     S_{\delta,\varepsilon}
     =\left[\varepsilon\beta_t+
     \lambda^2D\!\left(\sqrt{((t_2-t_1)/\lambda)^2+\delta^2}\right)
                                                       \right]\wedge[X],
     \qquad\delta,\varepsilon>0.
    \]
    They satisfy
    $\varepsilon S_0\le S_{\delta,\varepsilon}
    \le(\varepsilon+2/\delta)S_0$ on the original neighborhood.
    The preceding lemmas identify the completed rooftop of the two actual
    obstacles with $h_a(t/\lambda)$ for every positive pair of parameters.
    For $f\in\{u_a,v_a,h_a\}$, full original torus calibration and
    global Green exchange give the following exact smooth energy formula.
    Indeed the background matrix in the $t$ coordinates is
    $\varepsilon I+\chi_\delta''(y-x)
    \left(\begin{smallmatrix}1&-1\\-1&1\end{smallmatrix}\right)$;
    its cofactor is
    $\varepsilon I+\chi_\delta''(y-x)
    \left(\begin{smallmatrix}1&1\\1&1\end{smallmatrix}\right)$.
    The factor $\lambda^2$ in the background definition makes the change
    $t=\lambda(x,y)$ cancel in the energy integral. Local convex mollification
    as in IF20 passes the calibration on all original tests to each
    polyhedral potential; the zero-boundary compact clipping argument of
    IF21 justifies global Green exchange. All coefficients are bounded and
    smooth for fixed positive parameters. Thus
    \[
     e_{\delta,\varepsilon}(f(t/\lambda))
     =\varepsilon\int_{Q_0}|\nabla f|^2dxdy
     +\int_{Q_0}\frac{\delta^2(f_x+f_y)^2}
                           {((y-x)^2+\delta^2)^{3/2}}\,dxdy.    \tag{7}
    \]
    For fixed $a$ the gradients are bounded. With $r=y-x$, the inner
    cross-sectional integral of $(f_x+f_y)^2$ is continuous at $r=0$:
    away from finitely many diagonal breakpoints the active plane is locally
    fixed; if a whole interface lies on the diagonal, its adjacent affine
    functions have the same tangential derivative. Dominated convergence
    therefore applies to the cross sections. The nonnegative kernel in (7)
    has total real-line mass $2$ and concentrates at zero, so
    \[
     \lim_{\delta\downarrow0}e_{\delta,\varepsilon}(f(t/\lambda))
                  =\varepsilon E_0(f)+2E_\ell(f),\qquad
     E_0(f)=\int_{Q_0}|\nabla f|^2dxdy.                         \tag{8}
    \]
    The limiting denominator below is strictly positive, and (6)--(8) give
    \[
     \lim_{\varepsilon\downarrow0}\lim_{\delta\downarrow0}
     \frac{e_{\delta,\varepsilon}(h_a(t/\lambda))}
     {e_{\delta,\varepsilon}(u_a(t/\lambda))+
                             e_{\delta,\varepsilon}(v_a(t/\lambda))}
                             =\frac{(1+a)^2}{2}.                \tag{9}
    \]
    Given $C<2$, choose $a<1$ with the right side greater than $C$,
    then choose a sufficiently small positive $\varepsilon$, followed by
    a sufficiently small positive $\delta$. Fix these parameters. The
    resulting single original smooth background and two actual obstacles
    give the required strict inequality. No potential or completion is
    transferred to a singular limiting background.
    \end{proof}
    
    
    \paragraph{An explicit fixed smooth counterexample.}
    The family above contains a particularly simple fixed example. Take
    $a=1/2$ and $\varepsilon=\delta=1/100$. Then
    \[
     u=\max\{2(x-1),-\tfrac23(x+1),\tfrac23(y-1),-2(y+1)\},
     \qquad v(x,y)=u(y,x),
    \]
    \[
     h=\max\{2(|x|-1),2(|y|-1),\tfrac12|x+y|-1\}.
    \]
    For the single original smooth background $S=S_{1/100,1/100}$ defined
    above, the completed rooftop
    of $U=u(t/\lambda)$ and $V_1=v(t/\lambda)$ is $H=h(t/\lambda)$.
    The background form is bounded between $(1/100)\beta_t$ and
    $(200+1/100)\beta_t$. Its full original smooth energy formula is (7).
    
    Here is an exact evaluation rather than a numerical or limiting check.
    For a triangle $T$ of area $A_T$ whose three vertex values of $L=y-x$
    are $a_T<b_T<c_T$, affine cross-section integration gives
    \[
     \int_T\chi_\delta''(L)
     =\frac{2A_T}{c_T-a_T}
      \left(\frac{\chi_\delta(c_T)-\chi_\delta(b_T)}{c_T-b_T}
           -\frac{\chi_\delta(b_T)-\chi_\delta(a_T)}{b_T-a_T}\right),
     \qquad \chi_\delta(r)=\sqrt{r^2+\delta^2}.                 \tag{10}
    \]
    For a plane $f=\alpha x+\gamma y+\kappa$ on $T$, its energy contribution
    is $\varepsilon A_T(\alpha^2+\gamma^2)$ plus $(\alpha+\gamma)^2$
    times (10). The four triangles joining each apex to its four boundary
    edges give the obstacle energies; the six triangles in the theorem's
    proof give the rooftop energy. All relevant vertex values of $L$ are
    distinct. Exact substitution gives
    \begin{align*}
     e_S(U)=e_S(V_1)
         &=-\frac{16\sqrt{10001}}{225}+\frac2{25}
                                      +\frac{2\sqrt{40001}}{45},\\
     e_S(H)&=-\frac{3\sqrt{10001}}{25}+\frac{13}{100}
                                      +\frac{2\sqrt{40001}}{25}.
    \end{align*}
    Consequently
    \[
     e_S(H)-e_S(U)-e_S(V_1)
         =\frac{\sqrt{10001}}{45}
             -\frac{2\sqrt{40001}}{225}-\frac3{100}
         >\frac{100}{45}-\frac{2\cdot201}{225}-\frac3{100}
          =\frac{73}{180}>\frac25.                            \tag{11}
    \]
    The strict inequalities use only $\sqrt{10001}>100$ and
    $\sqrt{40001}<201$. Thus coefficient-one energy subadditivity fails in
    this one explicit smooth original background, with an exact positive gap.
    
    \paragraph{Scope.}
    The sharpness statement excludes a background-independent universal binary
    constant strictly below $2$ across the original degree-one class. It does
    not exclude a subcritical constant $C(S)<2$ for each fixed smooth
    background, or a direct arbitrary-finite-family estimate
    $e(P(U_i:i\in F))\le C(S,n)\sum_{i\in F}e(U_i)$ whose constant is
    independent of $|F|$. It does not disprove SR8 or produce a strongly
    convergent row with literal discontinuity. C1 remains open in the full
    arbitrary-background scope on the basis of this argument.
    
    \paragraph{Exact frozen source identity.}
    The R17 whole source for FP1--FP6, IF20--IF21, and the actual-core and
    compact-test qualifications has SHA-256
    \texttt{8389d2ce81aa5fd1b563767a3c207a06d0b50e661fb06ee978da2d39634c49be}.
    The finite relative rooftop definition and greatestness are recorded
    in frozen R12 LG2, SHA-256
    \texttt{5001a36a41048e2c6c94189dffcec2f0610537d90ddb64e8c0932770c861988d}.
    The completion and positive quadratic energy form remain frozen inputs
    in their exact original scopes. The explicit smooth energies above were
    independently recomputed from the printed triangulations with exact
    arithmetic; the strict final gap uses the displayed rational bounds.
    \end{document}