OpenMath

Proof source

Nonsmooth scalar curvature and the two level currents

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. Nonsmooth scalar curvature and the two level currents

Nonsmooth scalar curvature and the two level currents

Keep the original hypotheses and notation of Section sec:r11-level-disintegration. Thus QQ is any original symmetric middle-positive separately closed current of degree (p,p)(p,p), p<np<n, and ff is a locally bounded difference of classical psh functions. All identities are identities of original coefficient-Radon currents on ordinary open sets, tested against every original compact smooth complementary form. The positive level-current theorem of that section is the only new input to the present extension.

Definition (Lower and upper slope conventions). Write Lf−(t)=D(max⁡(f,t)Q)−1{f>t}D(fQ),Lf+(t)=D(max⁡(f,t)Q)−1{f≥t}D(fQ).(SC1)\begin{split} \mathsf L_f^{-}(t) &=D(\max(f,t)Q)-\mathbf1_{\{f>t\}}D(fQ),\\ \mathsf L_f^{+}(t) &=D(\max(f,t)Q)-\mathbf1_{\{f\geq t\}}D(fQ). \end{split} \tag{SC1} The superscripts indicate the lower and upper values assigned to the derivative of the scalar hinge at its corner; in particular Lf−(t)=LfQ(t)\mathsf L_f^{-}(t)=\mathsf L_f^Q(t) from (LC3). The background QQ is suppressed in this section’s notation only.

Proposition (Two positive level currents and their jump). Both currents in (SC1) are middle-positive and carried by {f=t}\{f=t\}. They satisfy Lf+(t)=L−f−(−t),Lf−(t)−Lf+(t)=1{f=t}D(fQ).(SC2)\mathsf L_f^{+}(t)=\mathsf L_{-f}^{-}(-t),\qquad \mathsf L_f^{-}(t)-\mathsf L_f^{+}(t) =\mathbf1_{\{f=t\}}D(fQ). \tag{SC2} For every original compact smooth test η\eta, the scalar function t↦⟨Lf−(t),η⟩t\mapsto\langle\mathsf L_f^{-}(t),\eta\rangle is right continuous and has bounded variation on each compact level interval. Its left limit at every tt is ⟨Lf+(t),η⟩\langle\mathsf L_f^{+}(t),\eta\rangle. On any compact real interval of levels, the local positive coefficient masses of both families are uniformly bounded. Both admit the energy coarea identity (LC7), with the same energy current ΓQ(f)\Gamma_Q(f).

Status: NEW_PROVED.

Proof. The elementary identity max⁡(−f,−t)=max⁡(f,t)−f−t\max(-f,-t)=\max(f,t)-f-t and D(tQ)=0D(tQ)=0 give the first formula in (SC2); Theorem thm:r11-positive-level-current applied to −f-f proves positivity and the carrier of Lf+(t)\mathsf L_f^{+}(t). Subtracting (SC1) proves the second formula.

For a fixed compact test, the function t↦⟨D(max⁡(f,t)Q),η⟩t\mapsto\langle D(\max(f,t)Q),\eta\rangle is Lipschitz: the potentials differ by at most ∣t−s∣|t-s|, and the fixed signed coefficient measure μQ,Dη\mu_{Q,D\eta} has finite variation on the test support. Continuity from below and above for the fixed signed coefficient measures of D(fQ)D(fQ) gives lim⁡s↓t1{f>s}D(fQ)=1{f>t}D(fQ),lim⁡s↑t1{f>s}D(fQ)=1{f≥t}D(fQ),\begin{split} \lim_{s\downarrow t}\mathbf1_{\{f>s\}}D(fQ) &=\mathbf1_{\{f>t\}}D(fQ),\\ \lim_{s\uparrow t}\mathbf1_{\{f>s\}}D(fQ) &=\mathbf1_{\{f\geq t\}}D(fQ), \end{split} in local coefficient total variation. This proves the stated one-sided continuity after combining the two terms of (SC1). The first term is Lipschitz in the level, and the second is a tail of a fixed finite signed measure after pushforward by ff; their sum therefore has bounded variation on compact intervals.

For the uniform mass assertion, fix a compact original moment stage and a positive trace control χβn−p−1\chi\beta^{n-p-1}, where β\beta is the sum of the elementary positive moment squares. The closed current Q′=βn−p−1∧QQ'=\beta^{n-p-1}\wedge Q has degree (n−1,n−1)(n-1,n-1) and Lf±,Q′(t)=βn−p−1∧Lf±,Q(t)\mathsf L_f^{\pm,Q'}(t)=\beta^{n-p-1}\wedge \mathsf L_f^{\pm,Q}(t) by the bounded smooth-slot identities. The bounded-variation density bound (LC12), applied to ff and −f-f, bounds the two positive trace masses uniformly in tt. Finite original coefficient domination proves the assertion for every compact stage, including any actual moment refinement.

Finally the difference in (SC2), integrated in tt against a bounded Borel scalar function, is zero: for each point xx the indicator of the singleton {f(x)}\{f(x)\} has Lebesgue integral zero, and Fubini applies to each fixed finite coefficient variation of D(fQ)D(fQ). Therefore (LC7) for the minus convention also proves it for the plus convention. ◻

Theorem (An exact scalar curvature formula with a real measure). Let J⊂RJ\subset\mathbf R be an open interval, let I⋐JI\Subset J contain the range of ff near the compact test support, and let FF be a finite convex real function on JJ. Let νF=F′′\nu_F=F'' be its positive distributional second-derivative measure, and write F−′,F+′F'_-,F'_+ for its finite one-sided derivatives on II. Then F(f)F(f) is locally a bounded difference of classical psh functions, and D(F(f)Q)=F−′(f)D(fQ)+∫JLf−(t) dνF(t)=F+′(f)D(fQ)+∫JLf+(t) dνF(t).(SC3)\boxed{\quad D(F(f)Q) =F'_-(f)D(fQ)+\int_J\mathsf L_f^{-}(t)\,d\nu_F(t) =F'_+(f)D(fQ)+\int_J\mathsf L_f^{+}(t)\,d\nu_F(t). \quad} \tag{SC3} The integrals are weak integrals of coefficient-Radon currents; they are well defined locally, middle-positive, and carried by the actual inverse image of a carrier of νF\nu_F. Only levels in a compact interval near the local range contribute.

The same formula holds for a real difference of two finite convex functions, with its signed second-derivative measure and the corresponding signed integral. All multipliers in (SC3) use the actual given Borel values. The construction commutes with ordinary restriction and every actual moment refinement.

Status: NEW_PROVED.

Proof. Write f=a−bf=a-b on the chosen witness stage. On a compact interval strictly larger than II and contained in JJ, the convex function FF is Lipschitz. Extend its restriction affinely past the two endpoints using its interior one-sided slopes; this gives a global convex Lipschitz extension F^\widehat F agreeing with FF on II. Choose LL greater than the absolute values of its slopes. On the compact range box of (a,b)(a,b) the function (s,t)⟼F^(s−t)+L(s+t)(SC4)(s,t)\longmapsto\widehat F(s-t)+L(s+t) \tag{SC4} is convex and nondecreasing in each variable. The frozen convex nondecreasing composition rule makes (SC4), composed with (a,b)(a,b), a bounded classical psh function. Subtracting L(a+b)L(a+b) gives a bounded-difference presentation of F(f)F(f).

Choose c<inf⁡I≤sup⁡I<dc<\inf I\leq\sup I<d with [c,d]⋐J[c,d]\Subset J. The elementary real convex representation, on II, is F(s)=A+Bs+∫(c,d)(s−t)+ dνF(t),F−′(s)=B+νF((c,s)).(SC5)F(s)=A+Bs+\int_{(c,d)}(s-t)_+\,d\nu_F(t),\qquad F'_-(s)=B+\nu_F((c,s)). \tag{SC5} The constants A,BA,B absorb the affine contribution of levels outside (c,d)(c,d). The measure in (SC5) is finite. The potentials (f−t)+(f-t)_+ are uniformly bounded on the compact stage, so Fubini against the fixed coefficient measures of QQ and distributional differentiation give D(F(f)Q)=BD(fQ)+∫(c,d)D((f−t)+Q) dνF(t).(SC6)D(F(f)Q)=B D(fQ) +\int_{(c,d)}D((f-t)_+Q)\,d\nu_F(t). \tag{SC6} The positive level masses are uniformly bounded by Proposition “Two positive level currents and their jump”; their compact test pairings are Borel. They consequently have a well-defined weak integral with local coefficient-Radon control. Insert D((f−t)+Q)=1{f>t}D(fQ)+Lf−(t)D((f-t)_+Q)=\mathbf1_{\{f>t\}}D(fQ)+\mathsf L_f^-(t) in (SC6), and apply Fubini to the fixed signed coefficient measures in its first term. Its multiplier is exactly B+νF((c,f))=F−′(f)B+\nu_F((c,f))=F'_-(f). This proves the first equality in (SC3).

Replacing the strict indicator by the weak indicator gives B+νF((c,f])=F+′(f)B+\nu_F((c,f])=F'_+(f) and the plus level current, proving the second equality. This also checks explicitly the jump contribution at a real atom of νF\nu_F: changing slope conventions changes the curvature multiplier by νF({f})D(fQ)\nu_F(\{f\})D(fQ) and changes the level integral by its negative, through (SC2).

For any Borel carrier AA of νF\nu_F, the positive scalar kernel construction in (LC9), applied to the finite measure νF\nu_F in place of Lebesgue measure for measurability only, makes the weak integral a positive measure kernel. Each level current is carried by {f=t}\{f=t\}, so its integral is carried by f−1(A)f^{-1}(A). No assertion that this inverse image is closed is required. Finite polarization gives the same carrier assertion for every original coefficient. Linearity treats a difference of convex scalar functions. Intrinsic definitions and the already established refinement invariance of each level current prove the final assertions. ◻

Corollary (Original Kato inequalities and the exact defect). For ff as above, put f+=max⁡(f,0)f_+=\max(f,0) and use sgn⁡(0)=0\operatorname{sgn}(0)=0. Then D(f+Q)−1{f>0}D(fQ)=Lf−(0)≥0,D(∣f∣Q)−sgn⁡(f)D(fQ)=Lf−(0)+Lf+(0)≥0.(SC7)\begin{split} D(f_+Q)-\mathbf1_{\{f>0\}}D(fQ)&=\mathsf L_f^-(0)\geq0,\\ D(|f|Q)-\operatorname{sgn}(f)D(fQ) &=\mathsf L_f^-(0)+\mathsf L_f^+(0)\geq0. \end{split} \tag{SC7} Both defects are positive coefficient-Radon currents carried by the actual zero set of ff. The same identities hold after every original positive coefficient trace and against every original complementary test. Below top degree the defects need not be closed.

Status: DIRECT_COROLLARY of the new two-level theorem.

Proof. The first identity is (SC1). Apply it to −f-f and add, using ∣f∣=f++(−f)+|f|=f_++(-f)_+ and L−f−(0)=Lf+(0)\mathsf L_{-f}^-(0)=\mathsf L_f^+(0) from (SC2). The two strict indicators give sgn⁡(f)\operatorname{sgn}(f) with the printed value at zero. Positivity and the carriers were proved before the identities were added. Nonclosedness is allowed by the explicit differential defects in (LC5), and Proposition prop:r11-nonclosed-level-example prevents a universal closedness inference. ◻

What this calculus resolves and what it leaves open.

Formula (SC3) integrates a real curvature measure against actual level currents of a bounded psh difference. It therefore gives a precise replacement for formal multiplication by a real Dirac mass in a nonsmooth scalar chain rule. Its hypotheses do not define an arbitrary pullback of a cycle through a moment map, and no relation quotient for analytic delta presentations follows. Moreover the level currents retain the defects (LC5). Treating them as separately closed backgrounds in a second BT iteration is a FAILED_ATTEMPT; (LC13)–(LC15) give its concrete failure already on the original two-torus.

References requiring source context

Some references in this note depend on its original surrounding manuscript. Their source labels are retained in the reading edition; its TeX is available above.

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 58181ad09cd3381bd3106709d70b12d62c5fc6a54cc3e235ebd36633fd51bef0

Frozen-original SHA-256 58181ad09cd3381bd3106709d70b12d62c5fc6a54cc3e235ebd36633fd51bef0

\section{Nonsmooth scalar curvature and the two level currents}
\label{sec:r11-nonsmooth-scalar-curvature}

Keep the original hypotheses and notation of
Section~\ref{sec:r11-level-disintegration}.  Thus $Q$ is any
original symmetric middle-positive separately closed current of
degree $(p,p)$, $p<n$, and $f$ is a locally bounded difference
of classical psh functions.  All identities are identities of
original coefficient-Radon currents on ordinary open sets, tested
against every original compact smooth complementary form.  The
positive level-current theorem of that section is the only new
input to the present extension.

\begin{definition}[Lower and upper slope conventions]
Write
\[
 \begin{split}
 \mathsf L_f^{-}(t)
    &=D(\max(f,t)Q)-\mathbf1_{\{f>t\}}D(fQ),\\
 \mathsf L_f^{+}(t)
    &=D(\max(f,t)Q)-\mathbf1_{\{f\geq t\}}D(fQ).
 \end{split}
 \tag{SC1}
\]
The superscripts indicate the lower and upper values assigned to
the derivative of the scalar hinge at its corner; in particular
$\mathsf L_f^{-}(t)=\mathsf L_f^Q(t)$ from (LC3).
The background $Q$ is suppressed in this section's notation only.
\end{definition}

\begin{proposition}[Two positive level currents and their jump]
\label{prop:r11-two-level-currents}
Both currents in (SC1) are middle-positive and carried by
$\{f=t\}$.  They satisfy
\[
 \mathsf L_f^{+}(t)=\mathsf L_{-f}^{-}(-t),\qquad
 \mathsf L_f^{-}(t)-\mathsf L_f^{+}(t)
                      =\mathbf1_{\{f=t\}}D(fQ).
 \tag{SC2}
\]
For every original compact smooth test $\eta$, the scalar function
$t\mapsto\langle\mathsf L_f^{-}(t),\eta\rangle$ is right
continuous and has bounded variation on each compact level interval.
Its left limit at every $t$ is
$\langle\mathsf L_f^{+}(t),\eta\rangle$.  On any compact real
interval of levels, the local positive coefficient masses of both
families are uniformly bounded.  Both admit the energy coarea
identity (LC7), with the same energy current $\Gamma_Q(f)$.
\end{proposition}

\noindent\textbf{Status: \texttt{NEW\_PROVED}}.

\begin{proof}
The elementary identity
$\max(-f,-t)=\max(f,t)-f-t$ and $D(tQ)=0$ give the first
formula in (SC2); Theorem~\ref{thm:r11-positive-level-current}
applied to $-f$ proves positivity and the carrier of
$\mathsf L_f^{+}(t)$.  Subtracting (SC1) proves the second formula.

For a fixed compact test, the function
$t\mapsto\langle D(\max(f,t)Q),\eta\rangle$ is Lipschitz:
the potentials differ by at most $|t-s|$, and the fixed signed
coefficient measure $\mu_{Q,D\eta}$ has finite variation on the
test support.  Continuity from below and above for the fixed
signed coefficient measures of $D(fQ)$ gives
\[
 \begin{split}
 \lim_{s\downarrow t}\mathbf1_{\{f>s\}}D(fQ)
       &=\mathbf1_{\{f>t\}}D(fQ),\\
 \lim_{s\uparrow t}\mathbf1_{\{f>s\}}D(fQ)
       &=\mathbf1_{\{f\geq t\}}D(fQ),
 \end{split}
\]
in local coefficient total variation.  This proves the stated
one-sided continuity after combining the two terms of (SC1).
The first term is Lipschitz in the level, and the second is a
tail of a fixed finite signed measure after pushforward by $f$;
their sum therefore has bounded variation on compact intervals.

For the uniform mass assertion, fix a compact original moment
stage and a positive trace control $\chi\beta^{n-p-1}$, where
$\beta$ is the sum of the elementary positive moment squares.
The closed current $Q'=\beta^{n-p-1}\wedge Q$ has degree
$(n-1,n-1)$ and
$\mathsf L_f^{\pm,Q'}(t)=\beta^{n-p-1}\wedge
\mathsf L_f^{\pm,Q}(t)$ by the bounded smooth-slot identities.
The bounded-variation density bound (LC12), applied to $f$ and
$-f$, bounds the two positive trace masses uniformly in $t$.
Finite original coefficient domination proves the assertion for
every compact stage, including any actual moment refinement.

Finally the difference in (SC2), integrated in $t$ against a
bounded Borel scalar function, is zero: for each point $x$ the
indicator of the singleton $\{f(x)\}$ has Lebesgue integral
zero, and Fubini applies to each fixed finite coefficient
variation of $D(fQ)$.  Therefore (LC7) for the minus convention
also proves it for the plus convention.
\end{proof}

\begin{theorem}[An exact scalar curvature formula with a real measure]
\label{thm:r11-stieltjes-scalar-curvature}
Let $J\subset\mathbf R$ be an open interval, let $I\Subset J$
contain the range of $f$ near the compact test support, and let
$F$ be a finite convex real function on $J$.  Let $\nu_F=F''$
be its positive distributional second-derivative measure, and
write $F'_-,F'_+$ for its finite one-sided derivatives on $I$.
Then $F(f)$ is locally a bounded difference of classical psh
functions, and
\[
 \boxed{\quad
 D(F(f)Q)
  =F'_-(f)D(fQ)+\int_J\mathsf L_f^{-}(t)\,d\nu_F(t)
  =F'_+(f)D(fQ)+\int_J\mathsf L_f^{+}(t)\,d\nu_F(t).
 \quad}
 \tag{SC3}
\]
The integrals are weak integrals of coefficient-Radon currents;
they are well defined locally, middle-positive, and carried by
the actual inverse image of a carrier of $\nu_F$.
Only levels in a compact interval near the local range contribute.

The same formula holds for a real difference of two finite convex
functions, with its signed second-derivative measure and the
corresponding signed integral.  All multipliers in (SC3) use the
actual given Borel values.  The construction commutes with
ordinary restriction and every actual moment refinement.
\end{theorem}

\noindent\textbf{Status: \texttt{NEW\_PROVED}}.

\begin{proof}
Write $f=a-b$ on the chosen witness stage.  On a compact interval
strictly larger than $I$ and contained in $J$, the convex function
$F$ is Lipschitz.  Extend its restriction affinely past the two
endpoints using its interior one-sided slopes; this gives a global
convex Lipschitz extension $\widehat F$ agreeing with $F$ on $I$.
Choose $L$ greater than the absolute values of its slopes.  On the
compact range box of $(a,b)$ the function
\[
                  (s,t)\longmapsto\widehat F(s-t)+L(s+t)
 \tag{SC4}
\]
is convex and nondecreasing in each variable.  The frozen
convex nondecreasing composition rule makes (SC4), composed
with $(a,b)$, a bounded classical psh function.  Subtracting
$L(a+b)$ gives a bounded-difference presentation of $F(f)$.

Choose $c<\inf I\leq\sup I<d$ with $[c,d]\Subset J$.
The elementary real convex representation, on $I$, is
\[
 F(s)=A+Bs+\int_{(c,d)}(s-t)_+\,d\nu_F(t),\qquad
 F'_-(s)=B+\nu_F((c,s)).
 \tag{SC5}
\]
The constants $A,B$ absorb the affine contribution of levels
outside $(c,d)$.  The measure in (SC5) is finite.  The
potentials $(f-t)_+$ are uniformly bounded on the compact
stage, so Fubini against the fixed coefficient measures of
$Q$ and distributional differentiation give
\[
 D(F(f)Q)=B D(fQ)
           +\int_{(c,d)}D((f-t)_+Q)\,d\nu_F(t).
 \tag{SC6}
\]
The positive level masses are uniformly bounded by
Proposition~\ref{prop:r11-two-level-currents}; their compact
test pairings are Borel.  They consequently have a well-defined
weak integral with local coefficient-Radon control.  Insert
$D((f-t)_+Q)=\mathbf1_{\{f>t\}}D(fQ)+\mathsf L_f^-(t)$
in (SC6), and apply Fubini to the fixed signed coefficient
measures in its first term.  Its multiplier is exactly
$B+\nu_F((c,f))=F'_-(f)$.  This proves the first equality
in (SC3).

Replacing the strict indicator by the weak indicator gives
$B+\nu_F((c,f])=F'_+(f)$ and the plus level current, proving
the second equality.  This also checks explicitly the jump
contribution at a real atom of $\nu_F$: changing slope
conventions changes the curvature multiplier by
$\nu_F(\{f\})D(fQ)$ and changes the level integral by its
negative, through (SC2).

For any Borel carrier $A$ of $\nu_F$, the positive scalar
kernel construction in (LC9), applied to the finite measure
$\nu_F$ in place of Lebesgue measure for measurability only,
makes the weak integral a positive measure kernel.  Each level
current is carried by $\{f=t\}$, so its integral is carried
by $f^{-1}(A)$.  No assertion that this inverse image is
closed is required.  Finite polarization gives the same
carrier assertion for every original coefficient.
Linearity treats a difference of convex scalar functions.
Intrinsic definitions and the already established refinement
invariance of each level current prove the final assertions.
\end{proof}

\begin{corollary}[Original Kato inequalities and the exact defect]
\label{cor:r11-original-kato}
For $f$ as above, put $f_+=\max(f,0)$ and use
$\operatorname{sgn}(0)=0$.  Then
\[
 \begin{split}
 D(f_+Q)-\mathbf1_{\{f>0\}}D(fQ)&=\mathsf L_f^-(0)\geq0,\\
 D(|f|Q)-\operatorname{sgn}(f)D(fQ)
                 &=\mathsf L_f^-(0)+\mathsf L_f^+(0)\geq0.
 \end{split}
 \tag{SC7}
\]
Both defects are positive coefficient-Radon currents carried by
the actual zero set of $f$.  The same identities hold after
every original positive coefficient trace and against every
original complementary test.  Below top degree the defects
need not be closed.
\end{corollary}

\noindent\textbf{Status: \texttt{DIRECT\_COROLLARY}} of the new
two-level theorem.

\begin{proof}
The first identity is (SC1).  Apply it to $-f$ and add, using
$|f|=f_++(-f)_+$ and
$\mathsf L_{-f}^-(0)=\mathsf L_f^+(0)$ from (SC2).
The two strict indicators give $\operatorname{sgn}(f)$ with
the printed value at zero.  Positivity and the carriers were
proved before the identities were added.  Nonclosedness is
allowed by the explicit differential defects in (LC5), and
Proposition~\ref{prop:r11-nonclosed-level-example} prevents a
universal closedness inference.
\end{proof}

\paragraph{What this calculus resolves and what it leaves open.}
Formula (SC3) integrates a real curvature measure against actual
level currents of a bounded psh difference.  It therefore gives
a precise replacement for formal multiplication by a real Dirac
mass in a nonsmooth scalar chain rule.  Its hypotheses do not
define an arbitrary pullback of a cycle through a moment map,
and no relation quotient for analytic delta presentations follows.
Moreover the level currents retain the defects (LC5).  Treating
them as separately closed backgrounds in a second BT iteration
is a \texttt{FAILED\_ATTEMPT}; (LC13)--(LC15) give its concrete
failure already on the original two-torus.