Conjecture Atlas

Proof source

An independent derivation of the three-arm obstruction

Complete proof text in the unsigned reading edition.

Download sourceBrowse research records

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

Contents of this source
  1. An independent derivation of the three-arm obstruction

An independent derivation of the three-arm obstruction

Status. PROVED-INTERNAL: failure of strong subadditivity of the original outer capacity in every ambient dimension n≥2n\ge2, with remaining degree q=1q=1. This does not refute Borel compact-inner regularity, because strong subadditivity is not a necessary axiom of a Choquet capacity.

Use the original torus, box VV, normalized moment map τ:V→Q=(−1,1)n\tau:V\to Q=(-1,1)^n, and smooth positive background S=βn−1[X]S=\beta^{n-1}[X] from the audited linear trace lemma. Put γ=(n−1)!λn−2.\gamma=(n-1)!\lambda^{n-2}. That lemma, for every bounded original normalized classical psh tester aa, gives a continuous convex g:Q→[0,1]g:Q\to[0,1] with τ∗(Da∧S)=γΔg.(SA1)\tau_*(Da\wedge S)=\gamma\Delta g. \tag{SA1} It uses the convex almost-everywhere representative of a∘σa\circ\sigma, not an unproved pointwise identification.

Fix 0<r<1/30<r<1/3. In (−1,1)2(-1,1)^2 let C={(0,−s):0≤s≤r},D={(s,0):0≤s≤r},E={(−s,0):0≤s≤r},A=C∪D,B=C∪E.C=\{(0,-s):0\le s\le r\},\quad D=\{(s,0):0\le s\le r\},\quad E=\{(-s,0):0\le s\le r\},\quad A=C\cup D,\quad B=C\cup E. Write J=[−r,r]n−2J=[-r,r]^{n-2}, with the usual singleton and volume-one convention if n=2n=2, and put F^=τ−1(F×J)(F=A,B,C,D,E),κ=γ(2r)n−2.\widehat F=\tau^{-1}(F\times J)\quad(F=A,B,C,D,E),\qquad \kappa=\gamma(2r)^{n-2}. These are compact subsets in the original ordinary topology; no assumption that their inverse images consist only of skeleton points is made.

Lemma (Paired jumps on adjacent perpendicular arms). For every convex g:(−1,1)2→[0,1]g:(-1,1)^2\to[0,1], Δg(C)≤2r,Δg(A)≤2r1−r,Δg(B)≤2r1−r.(SA2)\Delta g(C)\le2r,\qquad \Delta g(A)\le {2r\over1-r},\qquad \Delta g(B)\le {2r\over1-r}. \tag{SA2}

Proof. The slice formula gives ∂xxg(C)=∫0r(∂x,+g(0,−s)−∂x,−g(0,−s)) ds.\partial_{xx}g(C)=\int_0^r \bigl(\partial_{x,+}g(0,-s)-\partial_{x,-}g(0,-s)\bigr)\,ds. The transverse Lebesgue factor in the other slice formula gives ∂yyg(C)=0\partial_{yy}g(C)=0. Every slice at x=0x=0 has derivative jump at most 22, by secants toward x=±1x=\pm1 and 0≤g≤10\le g\le1. This proves the first bound. In particular, trace Hessian measures have no point atoms, so the common endpoint of the arms causes no additional term.

Set L=(1−r)−1L=(1-r)^{-1}. Every subgradient pp of gg at a point of [−r,r]2[-r,r]^2 satisfies ∣p1∣+∣p2∣≤L|p_1|+|p_2|\le L: apply the supporting inequality at displacement (1−r−ε)(sgn⁡p1,sgn⁡p2)(1-r-\varepsilon)(\operatorname{sgn}p_1,\operatorname{sgn}p_2) and then let ε↓0\varepsilon\downarrow0.

For almost every s∈(0,r)s\in(0,r), the restrictions of gg to both axes are differentiable in their tangential directions at (0,−s)(0,-s) and (s,0)(s,0). Thus all subgradients at the first point have the same second component bb, and all those at the second have the same first component cc. The projections of these compact subdifferentials onto the normal coordinate are intervals with endpoints the corresponding one-sided directional derivatives. Write their extremal elements as pC+=(a,b),pC−=(a−h,b),pD−=(c,d),pD+=(c,d+k),p_C^+=(a,b),\quad p_C^-=(a-h,b),\qquad p_D^-=(c,d),\quad p_D^+=(c,d+k), where h,k≥0h,k\ge0 are exactly the normal derivative jumps. Monotonicity of the subgradient map applied to pC+p_C^+ and pD−p_D^- gives (pD−−pC+)⋅(s,s)≥0,c+d≥a+b.(p_D^--p_C^+)\cdot(s,s)\ge0, \qquad c+d\ge a+b. The ℓ1\ell^1 bounds on pC−p_C^- and pD+p_D^+ give h≤a+L−∣b∣,k≤L−∣c∣−d.h\le a+L-|b|,\qquad k\le L-|c|-d. Therefore h+k≤2L+a−d−∣b∣−∣c∣≤2L+c−b−∣b∣−∣c∣≤2L.h+k\le2L+a-d-|b|-|c| \le2L+c-b-|b|-|c|\le2L. Only the normal components charge the respective arms: ∂yyg(C)=0\partial_{yy}g(C)=0 and ∂xxg(D)=0\partial_{xx}g(D)=0 by their transverse Lebesgue factors. Integrating h+kh+k in ss proves the bound for AA. Reflection of the first coordinate proves the bound for BB. ◻

Lemma (The needed outer upper bound is non-circular). For any compact F⋐QF\Subset Q, the family Cn\mathcal C_n of convex functions g:Q→[0,1]g:Q\to[0,1] satisfies inf⁡F⊂U⋐Q, U open sup⁡g∈CnΔg(U)=sup⁡g∈CnΔg(F).(SA3)\inf_{F\subset U\Subset Q,\ U\text{ open}}\ \sup_{g\in\mathcal C_n}\Delta g(U) =\sup_{g\in\mathcal C_n}\Delta g(F). \tag{SA3} Consequently the right side, multiplied by γ\gamma, bounds the original CS∗(τ−1F;V)C_S^*(\tau^{-1}F;V) from above.

Proof. Interior slope bounds and the Arzela–Ascoli diagonal argument make Cn\mathcal C_n compact for locally uniform convergence. Under such convergence the positive measures Δg\Delta g converge weakly on each protected compact region: distributional convergence follows immediately, and uniform local mass bounds upgrade it to convergence against continuous compact tests.

Choose closed compact neighborhoods Fj↓FF_j\downarrow F inside QQ, and choose gjg_j whose mass on FjF_j is within 1/j1/j of the supremum. Extract a locally uniformly convergent subsequence with limit gg. For each fixed mm, eventual containment Fj⊂FmF_j\subset F_m and Portmanteau give lim⁡jsup⁡f∈CnΔf(Fj)≤Δg(Fm).\lim_j\sup_{f\in\mathcal C_n}\Delta f(F_j)\le\Delta g(F_m). Let m→∞m\to\infty and use continuity from above of this single locally finite measure. The result is at most sup⁡fΔf(F)\sup_f\Delta f(F). Monotonicity supplies the reverse bound. Taking the interiors of the FjF_j proves (SA3).

For every original tester the pushforward is given by (SA1), so its mass on the ordinary open set τ−1U\tau^{-1}U is at most γsup⁡gΔg(U)\gamma\sup_g\Delta g(U). Taking these particular original open neighborhoods gives the claimed outer upper bound. This argument uses compactness of the explicit convex family, not a general Borel outer-regularity theorem for the original tester family. ◻

Proposition (Strict failure of original outer strong subadditivity). One has CS∗(C^;V)=2rκ,CS∗(A^∪B^;V)≥4rκ,C_S^*(\widehat C;V)=2r\kappa,\qquad C_S^*(\widehat A\cup\widehat B;V)\ge4r\kappa, whereas CS∗(A^;V), CS∗(B^;V)≤2r1−rκ.C_S^*(\widehat A;V),\ C_S^*(\widehat B;V) \le {2r\over1-r}\kappa. Since A^∩B^=C^\widehat A\cap\widehat B=\widehat C, these imply CS∗(A^∪B^;V)+CS∗(A^∩B^;V)−CS∗(A^;V)−CS∗(B^;V)≥2r(1−3r)1−rκ>0.(SA4)C_S^*(\widehat A\cup\widehat B;V)+C_S^*(\widehat A\cap\widehat B;V) -C_S^*(\widehat A;V)-C_S^*(\widehat B;V) \ge {2r(1-3r)\over1-r}\kappa>0. \tag{SA4} At r=1/6r=1/6, the left two capacities sum to at least κ\kappa, while the right two sum to at most 4κ/54\kappa/5.

Proof. For n=2n=2, combine (SA2) and (SA3). The explicit normalized toric tester ∣x∣|x| has trace source 2δ0(dx) dy2\delta_0(dx)\,dy, so it gives CC mass 2r2r. The tester ∣y∣|y| gives D∪ED\cup E mass 4r4r. These provide the lower bounds using actual continuous classical psh testers and hence also lower bounds for outer capacity.

For general nn, fix the last n−2n-2 coordinates and apply the two-dimensional calculation to the resulting convex slice, then integrate over JJ. The xx and yy Hessian components acquire the factor (2r)n−2(2r)^{n-2}. Every Hessian component in one of the other coordinate directions gives (A∪B)×J(A\cup B)\times J zero mass: its transverse projection includes the planar three-arm set, which has two-dimensional Lebesgue measure zero; the slice masses are locally uniformly bounded. The same applies to A×JA\times J, B×JB\times J, and C×JC\times J. The explicit testers ∣x∣|x| and ∣y∣|y| are independent of the other coordinates and give exactly the claimed lower bounds. Finally multiply by γ\gamma from (SA1) and apply (SA3) to the compact products. ◻

Limits of the conclusion.

The capacity, normalization, background, field, absolute value, and full tester class are unchanged. The counterexample concerns strong subadditivity, not ordinary subadditivity. It is an obstruction to a proposed route to B1, not a counterexample to B1 itself. The compact-set outer argument above is confined to this explicit moment-trace setting and cannot be exported to arbitrary compact sets or arbitrary backgrounds. No claim of worldwide novelty is made without a separate literature comparison.

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 cbdbb8de8e632b7fc36d8c893adf0fc963076a42c261ee8ce6fb2e0433ace1fe

Frozen-original SHA-256 cbdbb8de8e632b7fc36d8c893adf0fc963076a42c261ee8ce6fb2e0433ace1fe

\section{An independent derivation of the three-arm obstruction}
\noindent\textbf{Status.} \texttt{PROVED-INTERNAL}: failure of strong subadditivity of the original outer capacity in every ambient dimension $n\ge2$, with remaining degree $q=1$. This does not refute Borel compact-inner regularity, because strong subadditivity is not a necessary axiom of a Choquet capacity.

Use the original torus, box $V$, normalized moment map $\tau:V\to Q=(-1,1)^n$, and smooth positive background $S=\beta^{n-1}[X]$ from the audited linear trace lemma. Put
\[
 \gamma=(n-1)!\lambda^{n-2}.
\]
That lemma, for every bounded original normalized classical psh tester $a$, gives a continuous convex $g:Q\to[0,1]$ with
\[
 \tau_*(Da\wedge S)=\gamma\Delta g.
 \tag{SA1}
\]
It uses the convex almost-everywhere representative of $a\circ\sigma$, not an unproved pointwise identification.

Fix $0<r<1/3$. In $(-1,1)^2$ let
\[
 C=\{(0,-s):0\le s\le r\},\quad
 D=\{(s,0):0\le s\le r\},\quad
 E=\{(-s,0):0\le s\le r\},\quad
 A=C\cup D,\quad B=C\cup E.
\]
Write $J=[-r,r]^{n-2}$, with the usual singleton and volume-one convention if $n=2$, and put
\[
 \widehat F=\tau^{-1}(F\times J)\quad(F=A,B,C,D,E),\qquad
 \kappa=\gamma(2r)^{n-2}.
\]
These are compact subsets in the original ordinary topology; no assumption that their inverse images consist only of skeleton points is made.

\begin{lemma}[Paired jumps on adjacent perpendicular arms]
For every convex $g:(-1,1)^2\to[0,1]$,
\[
 \Delta g(C)\le2r,\qquad
 \Delta g(A)\le {2r\over1-r},\qquad
 \Delta g(B)\le {2r\over1-r}.
 \tag{SA2}
\]
\end{lemma}
\begin{proof}
The slice formula gives
\[
 \partial_{xx}g(C)=\int_0^r
 \bigl(\partial_{x,+}g(0,-s)-\partial_{x,-}g(0,-s)\bigr)\,ds.
\]
The transverse Lebesgue factor in the other slice formula gives
$\partial_{yy}g(C)=0$. Every slice at $x=0$ has derivative jump at most $2$, by secants toward $x=\pm1$ and $0\le g\le1$. This proves the first bound. In particular, trace Hessian measures have no point atoms, so the common endpoint of the arms causes no additional term.

Set $L=(1-r)^{-1}$. Every subgradient $p$ of $g$ at a point of $[-r,r]^2$ satisfies $|p_1|+|p_2|\le L$: apply the supporting inequality at displacement $(1-r-\varepsilon)(\operatorname{sgn}p_1,\operatorname{sgn}p_2)$ and then let $\varepsilon\downarrow0$.

For almost every $s\in(0,r)$, the restrictions of $g$ to both axes are differentiable in their tangential directions at $(0,-s)$ and $(s,0)$. Thus all subgradients at the first point have the same second component $b$, and all those at the second have the same first component $c$. The projections of these compact subdifferentials onto the normal coordinate are intervals with endpoints the corresponding one-sided directional derivatives. Write their extremal elements as
\[
 p_C^+=(a,b),\quad p_C^-=(a-h,b),\qquad
 p_D^-=(c,d),\quad p_D^+=(c,d+k),
\]
where $h,k\ge0$ are exactly the normal derivative jumps.
Monotonicity of the subgradient map applied to $p_C^+$ and $p_D^-$ gives
\[
 (p_D^--p_C^+)\cdot(s,s)\ge0,
 \qquad c+d\ge a+b.
\]
The $\ell^1$ bounds on $p_C^-$ and $p_D^+$ give
\[
 h\le a+L-|b|,\qquad k\le L-|c|-d.
\]
Therefore
\[
 h+k\le2L+a-d-|b|-|c|
       \le2L+c-b-|b|-|c|\le2L.
\]
Only the normal components charge the respective arms: $\partial_{yy}g(C)=0$ and $\partial_{xx}g(D)=0$ by their transverse Lebesgue factors. Integrating $h+k$ in $s$ proves the bound for $A$. Reflection of the first coordinate proves the bound for $B$.
\end{proof}

\begin{lemma}[The needed outer upper bound is non-circular]
For any compact $F\Subset Q$, the family $\mathcal C_n$ of convex functions $g:Q\to[0,1]$ satisfies
\[
 \inf_{F\subset U\Subset Q,\ U\text{ open}}\ \sup_{g\in\mathcal C_n}\Delta g(U)
       =\sup_{g\in\mathcal C_n}\Delta g(F).
 \tag{SA3}
\]
Consequently the right side, multiplied by $\gamma$, bounds the original $C_S^*(\tau^{-1}F;V)$ from above.
\end{lemma}
\begin{proof}
Interior slope bounds and the Arzela--Ascoli diagonal argument make $\mathcal C_n$ compact for locally uniform convergence. Under such convergence the positive measures $\Delta g$ converge weakly on each protected compact region: distributional convergence follows immediately, and uniform local mass bounds upgrade it to convergence against continuous compact tests.

Choose closed compact neighborhoods $F_j\downarrow F$ inside $Q$, and choose $g_j$ whose mass on $F_j$ is within $1/j$ of the supremum. Extract a locally uniformly convergent subsequence with limit $g$. For each fixed $m$, eventual containment $F_j\subset F_m$ and Portmanteau give
\[
 \lim_j\sup_{f\in\mathcal C_n}\Delta f(F_j)\le\Delta g(F_m).
\]
Let $m\to\infty$ and use continuity from above of this single locally finite measure. The result is at most $\sup_f\Delta f(F)$. Monotonicity supplies the reverse bound. Taking the interiors of the $F_j$ proves (SA3).

For every original tester the pushforward is given by (SA1), so its mass on the ordinary open set $\tau^{-1}U$ is at most $\gamma\sup_g\Delta g(U)$. Taking these particular original open neighborhoods gives the claimed outer upper bound. This argument uses compactness of the explicit convex family, not a general Borel outer-regularity theorem for the original tester family.
\end{proof}

\begin{proposition}[Strict failure of original outer strong subadditivity]
One has
\[
 C_S^*(\widehat C;V)=2r\kappa,\qquad
 C_S^*(\widehat A\cup\widehat B;V)\ge4r\kappa,
\]
whereas
\[
 C_S^*(\widehat A;V),\ C_S^*(\widehat B;V)
                         \le {2r\over1-r}\kappa.
\]
Since $\widehat A\cap\widehat B=\widehat C$, these imply
\[
 C_S^*(\widehat A\cup\widehat B;V)+C_S^*(\widehat A\cap\widehat B;V)
 -C_S^*(\widehat A;V)-C_S^*(\widehat B;V)
 \ge {2r(1-3r)\over1-r}\kappa>0.
 \tag{SA4}
\]
At $r=1/6$, the left two capacities sum to at least $\kappa$, while the right two sum to at most $4\kappa/5$.
\end{proposition}
\begin{proof}
For $n=2$, combine (SA2) and (SA3). The explicit normalized toric tester $|x|$ has trace source $2\delta_0(dx)\,dy$, so it gives $C$ mass $2r$. The tester $|y|$ gives $D\cup E$ mass $4r$. These provide the lower bounds using actual continuous classical psh testers and hence also lower bounds for outer capacity.

For general $n$, fix the last $n-2$ coordinates and apply the two-dimensional calculation to the resulting convex slice, then integrate over $J$. The $x$ and $y$ Hessian components acquire the factor $(2r)^{n-2}$. Every Hessian component in one of the other coordinate directions gives $(A\cup B)\times J$ zero mass: its transverse projection includes the planar three-arm set, which has two-dimensional Lebesgue measure zero; the slice masses are locally uniformly bounded. The same applies to $A\times J$, $B\times J$, and $C\times J$. The explicit testers $|x|$ and $|y|$ are independent of the other coordinates and give exactly the claimed lower bounds. Finally multiply by $\gamma$ from (SA1) and apply (SA3) to the compact products.
\end{proof}

\paragraph{Limits of the conclusion.}
The capacity, normalization, background, field, absolute value, and full tester class are unchanged. The counterexample concerns strong subadditivity, not ordinary subadditivity. It is an obstruction to a proposed route to B1, not a counterexample to B1 itself. The compact-set outer argument above is confined to this explicit moment-trace setting and cannot be exported to arbitrary compact sets or arbitrary backgrounds. No claim of worldwide novelty is made without a separate literature comparison.