Contents of this source
A differential module with actual piecewise smooth coefficients
All currents and tests in this section are original CLD objects over the fixed with its original absolute value. Piecewise smooth (PS) means the actual codimension-zero sheaf, with tangential matching and the full actual moment equivalence, as in the frozen R6 PS multiplication theorem. It does not mean an arbitrary Borel coefficient array. Backgrounds remain arbitrary original symmetric middle-positive separately closed currents.
The inputs used below are precisely: the frozen R4 uniform normal-wall variation estimate for positive separately closed currents; the frozen weighted joint bounded BT theorem; the R6 canonical PS multiplication and tangency theorem; and the codimension-zero case of the R6 anchored Courant presentation. No analytic positive-codimension relation quotient is assumed.
The exact frozen pins are thm:r6-delta-module-generation in delta_module_generation.tex, lem:r6-level-tangency and thm:r6-pws-multiplication in pws_tangent_extension.tex, lem:r4-wall-normal-factor and thm:r4-piecewise-test-continuity in potential_quotient_next.tex, and thm:relative-monotone in the frozen foundational relative_monotone.tex. Their full archived paths, SHA256 values, and exact uses are recorded in PWS_DEPENDENCY_CHECK.json.
A simultaneous smoothing estimate with moving closed tails
On a compact stage in one original moment presentation, the frozen anchored construction writes every PS form as a finite sum of words where the are continuous real PL functions in those moments and is smooth on a neighborhood of the relevant finite coordinate image. Finite sums and original smooth coefficient forms are allowed. The latter can also be included by adding the original affine moment coordinates to the list in (P1).
Here is the exact use of the frozen generation theorem. In its codimension-zero case, every lifted constant full-dimensional balanced cycle on a connected real neighborhood is a constant multiple of the ambient integration current. Its printed anchored coefficient is , which is (P1). Its Courant functions are continuous PL but need not be convex. Refining by their finitely many walls and adding a sufficiently large positive sum of absolute affine wall functions writes each one locally as with both convex PL. Indeed their gradient jumps are bounded normal multiples along finitely many walls; a large sum of absolute wall functions makes every jump nonnegative. Convexity on all segments then proves (P2). This is the PL case of the frozen R6 continuous-PS difference lemma.
Use one nonnegative compact smooth convolution kernel of total mass one and zero first moment, with radius , for all real representatives on a slightly larger stage. Write and . Each positive row is smooth convex and decreases as , by the convexity/Jensen argument already printed for PL slots. The signed row is the convolution of itself. It converges locally uniformly, and all its first coefficient derivatives have one compact-stage bound. Let be the smooth form obtained from (P1) by these replacements.
Lemma (Uniform normal-factor error for the differentiated rows). Fix a compact original test stage, a PS form as above, and an original compact smooth complementary form . There are finitely many fixed original positive compact smooth controls and a constant such that, for every allowed positive separately closed diagonal current of the degree needed to pair with , The same estimate holds after multiplication by a Borel scalar of absolute value at most one. It applies to each word arising by differentiating the coefficient part of (P1) once or twice. All actual refinements are allowed by making a common original moment presentation with the chosen test.
Status: DIRECT_COROLLARY of the frozen R4 normal-factor variation estimate, with the required special smoothing rows verified here.
Proof. Refine the finitely many PL subdivisions by one finite arrangement of affine hyperplanes . Choose on each face an ambient gradient from any incident chamber. These are legitimate PS representatives by the frozen tangency theorem.
Across a facet, continuity of makes the difference of its two constant gradients a scalar multiple of . Compare two points at distance at most by the segment between them. The successive gradient changes are bounded normal multiples at the finitely many crossed walls. Every crossed wall satisfies at the initial point . At a point on several walls the same statement holds after choosing an incident chamber; its ambiguity is a sum of the same conormals. Averaging the gradient differences against the kernel therefore gives where the Borel coefficients have a common bound and are supported in the indicated shrinking wall tubes. All identities are understood modulo the face conormal ambiguity, which every allowed annihilates before Borel restriction.
The scalar functions are uniformly, so smoothness of on a compact coordinate neighborhood gives the same bound for their scalar coefficient error. Expand the difference of the finite wedge products in (P1) by replacing one factor at a time. It is a sum of an bounded coefficient form and bounded forms with one factor or , supported in the corresponding wall tube. All other first-derivative coefficients are uniformly bounded. Multiplication by the fixed smooth preserves this description.
Original coefficient domination bounds the first part by . The R4 normal-factor variation estimate bounds every remaining part by . That frozen estimate is a total-variation estimate and already permits arbitrary bounded Borel coefficients in its other factors. It therefore proves (P3) and its scalar-weight version. If the complementary degree of is zero, only scalar words can occur, and uniform scalar convergence proves the estimate without a wall argument.
Replacing by one of its finitely many derivatives and deleting or adding finitely many first-derivative letters gives exactly the same coefficient description. This proves the assertion needed for once- and twice-differentiated words. ◻
Lemma (Weighted BT convergence against these moving PS rows). Let be any positive mixed bounded BT tail over a fixed allowed background, whose finitely many classical slots decrease to locally bounded limits, and let be the limiting tail. Let be bounded classical weights with locally bounded limit; a fixed bounded classical is allowed. Let be one of the rows just constructed, where . Then on every original compact smooth complementary test. The assertion also holds with the fixed PS coefficient in place of its moving row.
Status: NEW_PROVED as the weighted PS convergence used in the intrinsic extension.
Proof. Unweighted frozen BT convergence bounds the finitely many positive controls of in (P3), uniformly in . The weights have a common finite compact-stage bound. Thus (P3) shows that replacing by in (P5) has error tending to zero, even after multiplication by .
To handle the fixed , choose another fixed smoothing parameter . The test is original smooth, so the frozen weighted joint theorem gives The errors in replacing by on either side are uniformly , by (P3) and the common weight and control bounds. First let and then . This proves both assertions. In particular the passage with a possibly discontinuous classical weight uses the frozen weighted BT theorem; it is not inferred from weak convergence of Radon measures against Borel tests. ◻
Intrinsic derivatives and the corrected operator
Definition (The full PS coefficient module over allowed backgrounds). For , let consist of original currents locally admitting a finite expression where each is an allowed arbitrary original positive separately closed current and is an actual PS form of the needed bidegree. Products in (P6) are the frozen canonical PS products. Equality means equality of original currents on all original tests, not equality of witnesses. Neither symmetry nor separate closedness of is assumed.
Every such belongs to the frozen tangential Radon module, even when it is not closed. Indeed, write a PS multiplier by its finite smooth-face/Borel-stratum formula. For every original smooth level , both conormals annihilate the corresponding restriction of each closed before any face restriction. Bounded coefficient multiplication and further Borel restriction preserve these zero measures, so the same conormal identities hold for . The R6 PS multiplication theorem consequently applies to itself. Its finite face formula is intrinsic: a zero coefficient-Radon current has zero coefficient measures, which remain zero after each bounded coefficient multiplication. Thus further PS multiplication and associativity do not depend on a witness (P6), and do not require to be separately closed.
Theorem (Derivative-Radon closure with every interface retained). The module is stable under actual PS exterior coefficients, , and . In particular Their derivatives are their intrinsic original distribution derivatives. They can be computed in any local anchored presentation by the usual graded product rule, using the following replacements for every continuous PL generator over a closed tail : In the bottom two rows the entry means insertion into that word and absorption of the curvature into the closed tail; . Several curvature factors mean the frozen mixed BT products, expanded linearly in their positive and negative slots. All identities hold on every original test and every actual moment refinement. Degrees outside are zero.
Status: NEW_PROVED for the intrinsic differential module and its interface formulas. The existence of the undifferentiated module is INTERNAL_INPUT from R6.
Proof. PS multiplication and its associativity are frozen inputs. Consider one word in (P6), using (P1)–(P2), and its smooth row . Lemma “Uniform normal-factor error for the differentiated rows” gives in local coefficient variation on every fixed test.
Differentiate the smooth word once or twice. A coefficient derivative of creates another first-derivative letter. Differentiating by gives , and differentiating by gives . The other same-prime derivatives and all derivatives of curvature factors are zero. After expanding (P2), every term in , , or is another row of the type in Lemma “Uniform normal-factor error for the differentiated rows” multiplied by a positive closed BT tail over , with at most two inserted convex PL smoothing slots. The signs are exactly the usual graded signs; no derivative of a Borel face indicator is discarded.
Lemma “Weighted BT convergence against these moving PS rows”, with weight one, gives a coefficient-Radon limit for each finite term. It is its stated PS coefficient times the bounded BT tail of the limiting PL potentials. Distributional differentiation is continuous on the original test space, so these limits are precisely , , and . Their finite expressions belong to , because every positive mixed tail is again an allowed closed background. This also proves (P8).
The intrinsic distribution derivative is independent of every choice of representative, subdivision, DC splitting, and smoothing row. The finite calculation therefore proves the asserted independence rather than assuming it. The frozen anchored presentation and PS multiplication apply after every common actual refinement. Localizing on compact supports and gluing gives the full original test statement. ◻
Theorem (Intrinsic corrected bounded insertion on the PS module). For and an original locally bounded classical psh function , with , the intrinsic distribution is a coefficient-Radon current belonging to . For every witness (P6), In particular (P10) is independent of that witness. It is linear over actual PS exterior coefficients, local in the ordinary topology, and commutes with and in every permitted bidegree. Iteration gives These products are mixed symmetric and have joint decreasing continuity on every original compact smooth test. No middle-order preservation for an arbitrary positive nonclosed member of is asserted.
Status: NEW_PROVED for the PS coefficient extension, including bounded classical weights which are not assumed continuous.
Proof. All four scalar multiplications in (P9) are intrinsic coefficient-Radon multiplications by (P7). Thus its left side is an intrinsic distribution before proving the right side. For one witness term, use the smooth rows of the previous proof. The already proved smooth corrected Leibniz identity gives On the right, the fixed positive closed current and (P3) give convergence in local coefficient variation to .
On the left, each of is the finite sum described in the derivative proof. Lemma “Weighted BT convergence against these moving PS rows”, now with the fixed classical weight , proves convergence after multiplying each of these currents by . Passing their printed distribution derivatives to the limit proves (P10). This passage is the reason the normal-wall estimate and weighted theorem were established before defining the extension. It does not differentiate a PS coefficient through without accounting for its corners.
Every right side in (P10) belongs to the module. Since the left side depends only on , representation independence follows. Canonical PS associativity proves PS coefficient linearity. Ordinary locality and compatibility with all refinements follow from the intrinsic formula and the already original PS products.
For completeness the differential commutation is an intrinsic operator calculation, not an unproved PS Leibniz rule. The anticommutation of the two differentials gives, with all scalar factors defined by (P7), Module closure permits both sides. Formula (P11) follows by induction from (P10); every tail is positive and closed. Frozen mixed symmetry proves its symmetry. For a decreasing tuple the frozen products converge with uniform compact positive controls. The fixed-PS version of (P5), with weight one, gives the stated continuity. Finite localization handles all original test supports. ◻
An explicit nonclosed first-derivative example
Corollary (The full nonsymmetric PL gradient product). Let be any allowed -background, , and let be locally bounded convex PL functions of original moments. Define by canonical PS multiplication Then all the following are intrinsic coefficient-Radon identities on the complete original test sheaf: for every locally bounded classical . No symmetry of is required. For , is the positive gradient-energy current and The first equality in (P16) is the frozen R6 scalar PS product rule; the new assertion is stability under the intrinsic operator with every printed derivative retained.
Status: DIRECT_COROLLARY of the two new module theorems and the indicated frozen energy identity.
Proof. Apply (P8) to the two first-derivative letters, remembering their total degree one. This gives the first three lines of (P15), including the negative sign in . Formula (P10) gives the last line. The frozen scalar PS Leibniz formula for gives (P16). All background products in the formulas are the existing bounded BT products over the same arbitrary . ◻
The exact limits of the extension.
The argument proves a differential module and corrected insertion, not OPEN-GQ. A finite PS/PL response belongs to this module once its pure-cycle slots have their prescribed bounded presentations. It does not follow that a zero actual positive-codimension delta-form has zero response. In particular no equality on an auxiliary-field cap background is inferred from an unproved analytic quotient. The smooth freezing argument alone does not transfer ordinary order on an arbitrary nonclosed PS response: piecewise constant coefficient backgrounds are generally not separately closed. Theorem thm:r10-ps-module-positivity resolves this finite actual PS positivity question by a genuinely closed conormal penalty, followed by exact tangency on each carrier. No claim of signed actual-relation descent follows from that order theorem.
The scalar class cannot be enlarged silently to arbitrary Borel coefficient functions. On an original annulus let be its unit Gauss point, , and . Original smooth top-form integration has no atom at , so , whereas . A coefficient-linear extension over all Borel witnesses would therefore send the same zero current to both zero and a nonzero measure. This is a concrete FALSE extension, not a GQ counterexample. In (P5), the Borel point values of bounded classical weights are admitted only through the frozen weighted BT theorem.