'Finite-Jet Sign Rigidity on Hyperplane Arrangements and Denjoy–Carleman Realization'

Finite-Jet Sign Rigidity on Hyperplane Arrangements
and Denjoy–Carleman Realization

Brian Theory

Reviewed revision — 27 September 2026
Quantitative additions — 13 October 2026
Scope clarifications and revision record — 14 October 2026

Abstract

Let A be a finite real affine hyperplane arrangement. We study which {±1}-valued chamber labelings can occur as the sign of a function with prescribed regularity. A first nonzero homogeneous Taylor term inherits the chamber signs weakly, and factor multiplicities force a constant crossing parity. This gives a short arrangement-level proof of the known smooth signature-flatness obstruction and, in finite regularity, explicit Cd orthant and CN arrangement rigidity statements.

For normalized compactwise Roumieu Denjoy–Carleman classes E{M}, the Denjoy–Carleman criterion yields an exact realization dichotomy. In the quasianalytic case the realizable chamber laws are precisely the signed parity laws; in the non-quasianalytic case every chamber labeling is realized by explicit selector formulas. Thus an arrangement of m distinct hyperplanes with r chambers admits exactly 2m+1 and 2r realizable sign laws in the two regimes. We also record corner-margin consequences and a fixed-dimensional polynomial comparison on a cube with coordinate-wall strips of width δ removed: signed parity has constant exact degree, non-affine laws have degree Θ(log ⁡ (1∕δ)), and any nonconstant law requires degree Θ(1∕δ) under a global amplitude bound and fixed margin 0 < γ < 1. The smooth obstruction, positive-flat existence, and approximation ingredients are classical; the contribution is their explicit formulation and synthesis, with finite-order statements and complete reductions.

2020 Mathematics Subject Classification. 26E10, 52C35.

Keywords. finite jets; hyperplane arrangements; Denjoy–Carleman classes; quasianalyticity; sign distributions; polynomial sign representation.

1 Introduction

A Boolean truth table is finite. Requiring the same truth value throughout an entire orthant of ℝn is not. This distinction becomes sharp when the Boolean value is extracted from the sign of a smooth real-valued function.

For example, the polynomial

F(x,y) = xy

is positive in the first and third quadrants and negative in the other two. With positive interpreted as 1 and negative as 0, it realizes equivalence (XNOR), an affine Boolean function over 𝔽2. By contrast, the AND sign pattern is positive only in the positive-positive quadrant and negative in the other three. The nonsmooth function min ⁡ (x,y) has exactly that sign pattern, but the results below show that any smooth realization of AND on all four quadrants must be flat at the origin. In a quasianalytic class this is impossible for a nonzero function; in a non-quasianalytic class it is possible by using flat one-sided selectors.

The first purpose of this paper is to isolate the local mechanism behind that phenomenon. It is not coordinate division, resolution of singularities, or a special feature of one ultradifferentiable class. It is the first nonzero homogeneous Taylor term. If a smooth function has a nonzero finite jet at the common corner, its leading polynomial already inherits the orthant signs in a weak sense. Polynomial factorization across coordinate hyperplanes then forces an affine parity law. The last factor-multiplicity step is elementary and has clear antecedents; the point of the finite-jet formulation is that it transports that obstruction from polynomials to arbitrary smooth germs and makes the subsequent regularity classification immediate.

The second purpose is to carry that mechanism to a finite affine hyperplane arrangement. If a smooth function is nonzero off the arrangement and has no flat point on it, the local leading polynomial at each point fixes a crossing parity for every incident hyperplane. Those parities propagate along each connected affine hyperplane and hence determine the sign on every chamber.

The third purpose is to combine these smooth statements with the classical Denjoy–Carleman theorem. We use throughout one compactwise Roumieu convention,

|Dαf(x)|≤ C KρK|α||α|!M |α|,x ∈ K ⋐ U,

and no derivative-stability assumption. In this normalization the quasianalytic criterion is

∑ k=0∞ Mk (k + 1)Mk+1 = ∞.

On the divergent side, quasianalyticity rules out the flatness required by a non-affine orthant law. On the convergent side, a classical compactly supported cutoff yields a one-sided selector, from which arbitrary Boolean laws are assembled without cancellation.

Prior work and contribution boundary. The C∞ arrangement obstruction belongs to the signature-flatness theory of Broglia and Tognoli [12]; Remark 4.6 gives the precise reduction from the hypotheses used here, including the homological point in their theorem. Their factorization analysis also contains the parity-of-multiplicity mechanism underlying Lemma 3.1, and their examples already use a first nonzero Taylor coefficient to detect incompatible crossing parities. Thus neither the smooth obstruction nor the leading-term idea is claimed as new.

The broader historical setting includes distributions of signs and signatures [5, 1, 2, 4, 3], the Varchenko–Gelfand algebra of chamber functions and its later arrangement-theoretic extensions [7, 8, 6], and the classical Denjoy–Carleman theory of quasianalyticity, cutoffs, and positive-flat functions [14, 13, 15]. Stronger division and monomialization frameworks are also available [11, 9, 10], but they are not needed for the elementary proofs below.

Exact Boolean sign operations also have a classical constructive theory in the R-function literature. Shapiro’s survey [16, Sections 3.3 and 3.6] describes root-based conjunction and disjunction and their finite-differentiability upgrades. These supply direct examples at the boundary of the finite-jet statements; they are credited explicitly in Section 8.3.

The note combines explicit finite-order rigidity, a self-contained proof for affine arrangements, and exact realization classifications with constructive formulas and chamber counts. Section 7 adds quantitative consequences: Taylor estimates turn vanishing jets into corner-margin bounds, while classical rational approximation, Sherstov’s majority-conjunction lower bound [17], and Bernstein–Jackson inequalities [18] give a fixed-dimensional degree classification after excluding the walls. These are derived consequences, with no claim of historical firstness or new approximation machinery. They distinguish exact realization near the common corner from sign correctness at a positive distance from the walls.

Organization. Section 2 fixes the sign and Denjoy–Carleman conventions. Section 3 proves the weak-sign polynomial lemma. Section 4 establishes finite-jet rigidity and the smooth arrangement theorem. Section 5 constructs one-sided selectors and proves the whole-orthant dichotomy. Section 6 gives the arrangement classification and exact counts. Section 7 derives corner-margin and polynomial-degree bounds. Section 8 records counterexamples and comparisons with R-functions and normal-crossing results.

2 Definitions, conventions, and classical input

2.1 Orthants and Boolean sign laws

For α = (α1,…,αn) ∈{0,1}n, put

𝜀i(α) = 2αi − 1 ∈{−1,+1}

and define the open orthant

Oα = {x ∈ ℝn : 𝜀 i(α)xi > 0 for every i}.

A function F defined on an open set U meeting all these orthants realizes a Boolean law Θ : {0,1}n →{0,1} on the orthants in U if F is nonzero on their intersections with U and

sgn ⁡ F(x) = 2Θ(α) − 1,x ∈Oα ∩ U.

Local realization at the common corner means realization on a centered ball contained in U; global realization means U = ℝn. We call Θ affine if

Θ(α) = b ⊕ e1α1 ⊕⋯ ⊕ enαn,b,ei ∈{0,1},

where ⊕ denotes addition in 𝔽2. There are 2n+1 affine Boolean functions. When n = 1, these are all four unary Boolean functions.

A smooth function is flat at p if DαF(p) = 0 for every multi-index α.

Definition 2.1 (Taylor-quasianalytic class). For the two generic corollaries below, a smooth function class is called Taylor-quasianalytic if, at every point of its domain, a function whose full Taylor jet vanishes has zero germ there. For a vector space of germs, this zero-jet implication is equivalent to injectivity of the Taylor map. Only the zero-jet implication is used in the two generic corollaries below. No division, composition, or sheaf axiom is intended by this shorthand. The divergent Denjoy–Carleman classes used later have this property by the Denjoy–Carleman theorem.

2.2 Affine arrangements

Let

A = {H1,…,Hm},Hj = {x ∈ ℝn : ℓ j(x) = 0},

be a finite family of distinct proper affine hyperplanes, where each nonconstant affine defining form ℓj is fixed once and for all. A chamber is a connected component of

ℝn ∖⋃ j=1mH j.

We write Ch ⁡ (A) for the chamber set. On a chamber C, every sgn ⁡ ℓj is constant. A chamber labeling h : Ch ⁡ (A) →{±1} is a signed parity law if

h(C) = c∏ j=1m sgn ⁡ (ℓ j|C)ej ,c ∈{±1},ej ∈{0,1}.
(1)

The word “signed” is deliberate: when c = −1, this is not literally a group character, and the realized chamber sign vectors need not form a group.

2.3 One Denjoy–Carleman convention

Throughout Sections 5–6, a normalized weight means a sequence M = (Mk)k≥0 such that

M0 = 1,Mk > 0,

with M nondecreasing and log-convex. For an open set U ⊆ ℝn, define the compactwise Roumieu class E{M}(U) by requiring that for every compact K ⋐ U there exist CK,ρK > 0 such that

|Dαf(x)|≤ C KρK|α||α|!M |α|(x ∈ K).
(2)

This convention keeps the factorial outside M. To compare with the common total-weight convention, set

Wk = k!Mk.

Then W is a weight sequence in the sense used in [14]: it is positive and log-convex, W0 = 1 ≤ W1, and Wk1∕k →∞. Its quotient sequence satisfies

Wk+1 Wk = (k + 1)Mk+1 Mk ,

so the Denjoy–Carleman series is precisely

∑ k=0∞ Mk (k + 1)Mk+1.
(3)

The Denjoy–Carleman theorem says that E{M} is quasianalytic exactly when (3) diverges; see [14, Theorem 3.6]. In the convergent case, nontrivial compactly supported functions exist; the cutoff form we use is supplied, for example, by [14, Corollary 3.12].

We will need only elementary closure properties in addition to this classical input.

Lemma 2.2 (Elementary closure). For a normalized weight M, the class E{M} is closed under finite sums, finite products, translations, and precomposition with affine linear functionals. For functions on ℝ it is also closed under the primitive G(t) = ∫ 0tg(s)ds.

Proof. Only products and primitives need comment. Log-convexity with M0 = 1 implies MjMk−j ≤ Mk. Combining this with Leibniz’ rule and the bounds (2) gives the same form of estimate with a larger radius constant.

For a one-variable g ∈E{M}(ℝ), define G(t) = ∫ 0tg(s)ds. On a compact interval, G itself is bounded. For k ≥ 1, G(k) = g(k−1), and monotonicity of M gives

|G(k)|≤ Cρk−1(k − 1)!M k−1 ≤ C′(ρ′)kk!M k

for suitable enlarged constants. For an affine functional ℓ(x) = a ⋅ x + b and a one-variable function g, the chain rule gives

Dα(g ∘ ℓ)(x) = aαg(|α|)(ℓ(x)).

The image of a compact set is compact, and the factors from a are absorbed into the radius constant. Translations are the one-variable special case of slope one. □

3 A weak-sign polynomial lemma

The finite-jet argument rests on an elementary fact: even a weak chamber sign prescription for a nonzero polynomial has parity form.

Lemma 3.1 (Polynomial weak-sign lemma). Let A = {H1,…,Hm} be a finite family of distinct proper affine hyperplanes in ℝn, with fixed defining forms ℓj. Let P be a nonzero real polynomial. Suppose numbers hC ∈{±1} are assigned to the chambers C ∈ Ch ⁡ (A) and satisfy

hCP(x) ≥ 0(x ∈ C).
(4)

Then there are unique c ∈{±1} and e1,…,em ∈{0,1} such that

hC = c∏ j=1m sgn ⁡ (ℓ j|C)ej (C ∈ Ch ⁡ (A)).
(5)

The polynomial is allowed to vanish at points inside a chamber.

Proof. For each j, let aj be the largest exponent for which ℓjaj divides P, and write

P = ℓjaj Qj,ℓj ∤ Qj.

The restriction of Qj to Hj is therefore a nonzero polynomial on Hj and cannot vanish on a relatively open subset of Hj.

Consider two chambers sharing a facet contained in Hj. In the relative interior of that facet choose a point p outside every other hyperplane and with Qj(p)≠0. In a sufficiently small ball about p, the sign of Qj is fixed, while ℓj changes sign when Hj is crossed. Hence the sign of P changes by the factor (−1)aj. Wherever P≠0, the weak inequality (4) forces its sign to equal the prescribed chamber sign. Thus every adjacency across Hj has the same ratio

hC′ hC = (−1)aj .

Put ej ≡ aj(mod2).

Now divide the chamber label by ∏ ⁡ j sgn ⁡ (ℓj)ej. The quotient is unchanged across every chamber adjacency. The chamber adjacency graph is connected: choose interior points of two chambers and perturb a segment, or use a two-segment polygonal path, so that it meets the hyperplanes transversely and never meets two at the same time. The quotient is therefore a single constant c ∈{±1}, proving (5).

For uniqueness, fix j and choose a point of Hj outside all other hyperplanes. The two local chambers on its two sides differ only in the sign of ℓj, so their label ratio determines ej. One chamber then determines c. □

Remark 3.2. The content of Lemma 3.1 is elementary polynomial factorization plus chamber adjacency. We use it as a transparent engine for the smooth results; we do not attach an independent novelty claim to it.

4 Finite-jet rigidity

4.1 Orthants

Theorem 4.1 (Finite-jet rigidity). Let F be C∞ on a neighborhood of the origin in ℝn. Assume that in some ball B centered at the origin,

F(x)≠0whenever every coordinate of x is nonzero.

Because each B ∩Oα is connected, continuity makes the sign of F constant there; write the resulting local orthant sign law as h : {±1}n →{±1}. If at least one derivative of F at the origin is nonzero, then there are unique c ∈{±1} and e ∈{0,1}n such that

h(s) = c∏ i=1ns iei .
(6)

Equivalently, every smooth realization of a non-affine Boolean orthant law is flat at the origin.

Proof. Let d be the least order of a nonzero derivative of F at the origin; d = 0 is allowed. The first nonzero homogeneous Taylor term is

Pd(x) = ∑ |α|=dDαF(0) α! xα,P d≠0.

For any v ∈ ℝn with no zero coordinate, Taylor’s formula along the ray tv, t ↓ 0, gives

F(tv) = tdP d(v) + o(td).

By the nonvanishing hypothesis and connectedness just noted, F(tv) has sign h(sgn ⁡ v) for every sufficiently small t > 0,

h(sgn ⁡ v)Pd(v) ≥ 0.

Thus the nonzero polynomial Pd satisfies a weak sign prescription on the coordinate-hyperplane arrangement. Lemma 3.1 gives (6) and its uniqueness.

Under the identification si = 2αi − 1, a sign law of the form (6) is equivalent to an affine Boolean function b ⊕ e1α1 ⊕⋯ ⊕ enαn. The contrapositive gives the flatness statement. □

Remark 4.2 (Finite regularity). The proof uses only the first nonzero Taylor order. If the first nonzero jet has order d, then a Cd Taylor expansion is enough. The theorem does not assert that nonflatness is necessary for an affine law: flat functions can also realize affine laws.

Theorem 4.3 (Finite-order orthant rigidity). Let d ≥ 0 and let F be Cd near the origin, nonzero whenever all coordinates are nonzero in some centered ball. If every derivative of order < d vanishes at the origin and at least one derivative of order d is nonzero, then the local orthant sign law is affine parity.

Proof. The order-d Taylor expansion supplies the same nonzero leading homogeneous polynomial used in Theorem 4.1; Lemma 3.1 then applies verbatim. □

Corollary 4.4 (Quasianalytic orthant rigidity). Let F belong to a Taylor-quasianalytic class on a neighborhood of the origin and suppose it realizes a whole local orthant law there. Then that Boolean law is affine.

Proof. If every derivative of F vanished at the origin, the defining zero-jet property would make the germ of F at the origin zero, contradicting nonvanishing on every local open orthant. Apply Theorem 4.1. □

4.2 Affine hyperplane arrangements

Theorem 4.5 (No-flat-point arrangement rigidity). Let A = {H1,…,Hm} be a finite family of distinct proper affine hyperplanes in ℝn. Let F ∈ C∞(ℝn) be nonzero on ℝn ∖⋃ ⁡ jHj. Assume that F is not flat at any point of ⋃ ⁡ jHj. Then there are unique c ∈{±1} and e1,…,em ∈{0,1} such that

sgn ⁡ F(x) = c∏ j=1m sgn ⁡ ℓ j(x)ej ,x∉⋃ jHj.
(7)

Proof. Fix p ∈⋃ ⁡ jHj. Choose a ball Bp small enough that every hyperplane meeting Bp contains p. After translating p to the origin, the incident hyperplanes form a central arrangement. Let Pp be the first nonzero homogeneous Taylor polynomial of F at p.

For every direction v outside the incident hyperplanes,

F(p + tv) = td(p)P p(v) + o(td(p)),t ↓ 0.

The function F has a fixed sign on each local chamber because it is continuous and nonzero there. Hence Pp inherits those signs weakly. Lemma 3.1 gives, for every hyperplane Hj through p, a local crossing parity ej(p) ∈{0,1}.

We claim that ej(p) is locally constant as p varies in Hj. Fix p ∈ Hj and its ball Bp. For p′ ∈ Hj sufficiently close to p, choose the local ball Bp′ contained in Bp and avoiding all hyperplanes not containing p′. The relatively open set Hj ∩ Bp′ contains a point q outside every other hyperplane: the excluded intersections form a finite union of proper affine subspaces of Hj. In dimension one, Hj is a point and distinct hyperplanes do not meet it. A sufficiently small transverse pair near q, on opposite sides of Hj, lies inside both balls and meets no other hyperplane. The ratio of the two actual signs of F is both (−1)ej(p) and (−1)ej(p′). Thus the bits agree. Since Hj is connected, ej(p) is a single global bit ej.

Now consider the global chamber adjacency graph. Across a generic facet contained in Hj, the sign of F changes by (−1)ej. The same is true of the product ∏ ⁡ j sgn ⁡ (ℓj)ej. Their quotient is therefore unchanged across every chamber adjacency. As in Lemma 3.1, the chamber adjacency graph is connected, so the quotient is a global constant c ∈{±1}. This proves (7). Uniqueness follows by crossing a generic point of each Hj. □

Remark 4.6 (Earlier signature-flatness theorem). Theorem 4.5 is an elementary affine-arrangement case of the signature-flatness theory of Broglia and Tognoli [12, Theorem 4.2]. To compare hypotheses, set

G = F∏ j=1mℓ j2.

Then G has zero set exactly ⋃ ⁡ jHj and the same chamber signs as F. If F has a first nonzero Taylor term at a point of the arrangement, the first nonzero term of G is its product with a nonzero polynomial, so nonflatness is preserved. For n ≥ 2 their homological condition is automatic on ℝn because Hn−1(ℝn; ℤ2) = 0; the hyperplane components admit global affine equations and satisfy the required connectivity. The case n = 1 is immediate from the one-dimensional crossing picture. Their admissible signatures reduce, for hyperplanes, to the signed parity laws in (1). This remark records the historical implication explicitly; the proof above is retained because it isolates the finite-jet mechanism needed for Theorem 4.7.

Theorem 4.7 (Uniform finite-order arrangement rigidity). Let N ≥ 0, let A = {H1,…,Hm} be a finite family of distinct proper affine hyperplanes in ℝn, and let F ∈ CN(ℝn) be nonzero on ℝn ∖⋃ ⁡ jHj. Assume that for every p ∈⋃ ⁡ jHj the N-jet of F at p is nonzero; equivalently, for each such p there is a multi-index α with |α|≤ N and DαF(p)≠0. Then the chamber sign law of F is a unique signed parity law of the form (7).

Proof. At each p on the arrangement choose the least d(p) ≤ N for which some derivative of order d(p) is nonzero. The Cd(p) Taylor expansion gives a nonzero leading homogeneous polynomial and therefore the same local crossing parities as in the proof of Theorem 4.5. The propagation of those parities along a hyperplane, and then across the connected chamber adjacency graph, uses only continuity and nonvanishing off the arrangement. Thus the remainder of that proof applies unchanged. □

Remark 4.8 (A finite-order obstruction). Equivalently, if a CN function is nonzero off the arrangement and realizes a chamber labeling that is not signed parity, then at some point of the arrangement its entire N-jet must vanish. The smooth theorem is a stronger all-orders analogue: it yields a single flat point. The finite-order statements alone do not imply that conclusion by letting N tend to infinity, because the point at which the N-jet vanishes may depend on N.

Corollary 4.9 (Quasianalytic arrangement rigidity). Let F be a nonzero globally defined member of a Taylor-quasianalytic class on ℝn, and suppose F is nonzero off a finite affine arrangement. Then its chamber sign law has the form (7).

Proof. At a flat point, the defining zero-jet property would make the germ of F there zero. Every neighborhood of a point on a finite hyperplane arrangement contains points off the arrangement, contradicting the assumed nonvanishing there. Thus F has no flat point on the arrangement, and Theorem 4.5 applies. □

5 Denjoy–Carleman realization on orthants

The finite-jet theorem supplies the rigidity half of the classification. The converse direction in a non-quasianalytic class is constructive.

Lemma 5.1 (One-sided selector). Let M be a normalized weight. If

∑ k=0∞ Mk (k + 1)Mk+1 < ∞,
(8)

then there exists ψ ∈E{M}(ℝ) such that

ψ(t) = 0(t ≤ 0),ψ(t) > 0(t > 0).

Proof. With Wk = k!Mk, condition (8) is the non-quasianalytic Denjoy–Carleman condition ∑ ⁡ Wk∕Wk+1 < ∞. By the classical cutoff theorem, choose a nonzero real-valued compactly supported function f in the corresponding Roumieu class; see [14, Theorem 3.6 and Corollary 3.12]. Translate f so that the left endpoint of its support is 0. Then

f(t) = 0(t ≤ 0),

and every interval (0,t), t > 0, contains a point at which f is nonzero: otherwise f would vanish on a right-neighborhood of 0, contradicting that 0 is the left endpoint of supp ⁡ f.

By Lemma 2.2, f2 and its primitive belong to the same class. Define

ψ(t) = ∫ 0tf(s)2ds.

For t ≤ 0 the integral is zero. For t > 0, the interval (0,t) contains a point where f≠0; continuity then gives an interval on which f2 > 0, and therefore ψ(t) > 0. □

Classical constructive antecedent. A stronger existence statement is classical. Rainer’s Remark 3.13, citing work of Langenbruch, records positive-flat ultradifferentiable functions for prescribed closed sets in the non-quasianalytic setting [14, Remark 3.13]; see also [13]. Taking the closed set to be the arrangement and changing the sign of such a function independently on each chamber already gives arbitrary chamber labelings, since flatness makes the chamberwise pieces glue with zero jet on every wall. The selector construction above is retained because it gives an explicit formula in the defining affine forms and makes the later counting argument transparent.

Definition 5.2. For a normalized weight M, let RM(n) be the set of Boolean functions Θ : {0,1}n →{0,1} realized by some globally defined F ∈E{M}(ℝn) that is nonzero whenever all coordinates are nonzero.

Theorem 5.3 (Whole-orthant classification). Let M be a normalized weight. Then

RM(n) = { {affine Boolean functions}, ∑ k=0∞ Mk (k + 1)Mk+1 = ∞, {all Boolean functions on {0,1}n},∑ k=0∞ Mk (k + 1)Mk+1 < ∞.

Proof. Assume first that the series diverges. By the Denjoy–Carleman theorem, E{M} is quasianalytic. If F realizes a whole-orthant law, then F is not the zero function in any neighborhood of the origin, so its Taylor jet at the origin cannot vanish identically. Theorem 4.1 implies that the law is affine. Conversely, every affine law is realized by the polynomial

F(x) = c∏ i=1nx iei

for suitable c ∈{±1} and ei ∈{0,1}, and every polynomial lies in the class.

Now assume the series converges and choose the selector ψ from Lemma 5.1. For α ∈{0,1}n, define

Ψα(x) = ∏ i=1nψ(𝜀 i(α)xi).

For an arbitrary Boolean function Θ, set

FΘ(x) = ∑ α∈{0,1}n(2Θ(α) − 1)Ψα(x).
(9)

By Lemma 2.2, FΘ ∈E{M}(ℝn). If x ∈Oα, then Ψα(x) > 0, while for every β≠α at least one factor of Ψβ(x) is zero. Hence

sgn ⁡ FΘ(x) = 2Θ(α) − 1.

Thus every Boolean function is realized. □

Corollary 5.4 (Gevrey dichotomy). For the Gevrey class Gs, s ≥ 1, with derivative growth Cρk(k!)s, the whole-orthant laws are exactly the affine Boolean functions when s = 1, and all Boolean functions when s > 1.

Proof. In the normalization (2), take Mk = (k!)s−1. Then

Mk (k + 1)Mk+1 = (k + 1)−s.

The corresponding series diverges exactly at s = 1 for s ≥ 1. Apply Theorem 5.3. □

Remark 5.5. For n = 1, the two sides of Theorem 5.3 contain the same four Boolean functions because every unary Boolean function is affine. The strict expressive difference begins at arity n ≥ 2.

6 Affine arrangements: classification and exact counts

The orthant construction extends directly once orthants are replaced by the realized sign cells of a finite affine arrangement.

Theorem 6.1 (Arrangement classification). Let M be a normalized weight. Let A be an arrangement of m distinct proper affine hyperplanes in ℝn, and write r = |Ch ⁡ (A)|.

If the series (3) diverges, the chamber sign laws realized by globally defined F ∈E{M}(ℝn) that are nonzero off the arrangement are exactly the signed parity laws (1); there are exactly 2m+1 such laws.

If the series (3) converges, every one of the 2r chamber labelings is realizable by such a function.

Proof. On the quasianalytic side, Corollary 4.9 gives the signed-parity form. Every choice of (c,e1,…,em) is realized by the polynomial

c∏ j=1mℓ j(x)ej .

These 2m+1 restrictions are distinct. Indeed, for each Hj choose a generic point of Hj outside all other hyperplanes. The adjacent chamber pair at that point differs only in the sign of ℓj, so its label ratio recovers ej; one chamber then recovers c.

On the non-quasianalytic side, let ψ be the selector from Lemma 5.1. For each chamber C, let

sj(C) = sgn ⁡ (ℓj|C)

and define

ΨC(x) = ∏ j=1mψ(s j(C)ℓj(x)).

Distinct chambers have distinct sign vectors: a nonempty fixed sign cell is an intersection of open half-spaces and is therefore convex and connected. Hence if x lies in a chamber D, then ΨD(x) > 0 and ΨC(x) = 0 for every C≠D.

Given any labeling h : Ch ⁡ (A) →{±1}, set

Fh(x) = ∑ C∈Ch ⁡ (A)h(C)ΨC(x).

Elementary closure gives Fh ∈E{M}(ℝn), and on chamber D the sum reduces to h(D)ΨD(x). Thus Fh has the prescribed sign and is nonzero off the arrangement. □

Example 6.2 (Three concurrent lines). Take three distinct lines through the origin in ℝ2, for example

x = 0,y = 0,x − y = 0.

There are r = 6 chambers. A full quasianalytic normalized Roumieu class E{M} therefore realizes 23+1 = 16 signed-parity chamber laws, whereas a full non-quasianalytic such class realizes all 26 = 64 labelings.

Remark 6.3 (When the counts coincide). Each added distinct hyperplane splits at least one existing chamber, so r ≥ m + 1. The numerical gap is strict exactly when r > m + 1. For m parallel hyperplanes one has r = m + 1, so the two counts coincide. Thus the quasianalytic/non-quasianalytic distinction is structural, but its cardinality shadow depends on the arrangement.

Table 1: The classification in the common notation of this paper. All exact realizability counts refer to the full normalized Roumieu classes E{M}.

Setting

Mechanism

Sign laws

Smooth germ with a nonzero jet at the common orthant corner

Leading homogeneous Taylor polynomial and Lemma 3.1

Affine parity only

CN function on an arrangement with nonzero N-jet at every wall point

Local leading terms plus parity propagation

Signed parity only

Quasianalytic E{M} on orthants

Taylor injectivity excludes the flatness required by Theorem 4.1

Exactly 2n+1 affine Boolean laws

Non-quasianalytic E{M} on orthants

One-sided flat selector

All 22n Boolean laws

Quasianalytic E{M} on m distinct hyperplanes

No flat points plus Theorem 4.5

Exactly 2m+1 signed-parity laws

Non-quasianalytic E{M} on an arrangement with r chambers

Chamber selectors

All 2r chamber labelings

7 Margins and exclusion of the coordinate walls

The qualitative classification imposes correctness arbitrarily close to every wall. Two quantitative consequences clarify this requirement. Throughout this section h : {−1,+1}n →{−1,+1}, with positive meaning true, and n is fixed. A non-affine law means one not of the form c∏ ⁡ i∈Ssi.

7.1 Corner margins under derivative bounds

Proposition 7.1 (A normalized corner estimate). Let k ≥ 0, 0 < α ≤ 1, and let F ∈ Ck,α(BR(0)) realize a non-affine orthant law. Suppose that DkF has Hölder seminorm at most H, using the Euclidean multilinear operator norm (absolute value for k = 0). Then its entire k-jet at zero vanishes and

|F(x)|≤ H Γ(α + 1) Γ(k + α + 1)∥x∥k+α(x ∈ B R(0)).
(10)

In particular, if nδ < R and h(s)F(δs) ≥ γ for all sign vectors s, then

γ ≤ H Γ(α + 1) Γ(k + α + 1)(nδ)k+α.

Proof. Any least nonzero derivative of order at most k would contradict Theorem 4.3. Thus all these derivatives vanish. For k = 0, (10) is the Hölder estimate from F(0) = 0. For k ≥ 1, write x = rv, ∥v∥ = 1, and use integral Taylor remainder:

|F(rv)|≤ H (k − 1)!∫ 0r(r − t)k−1tαdt = H Γ(α + 1) Γ(k + α + 1)rk+α.

Substitute r = nδ for the margin conclusion. □

Remark 7.2 (Smooth derivative budgets). For a smooth exact non-affine realization, Theorem 4.1 gives flatness. If, on BR(0), fixed constants A,B > 0 satisfy

∥DjF(x)∥ op ≤ ABjj!M j(j ≥ 0),

Taylor remainder at every order gives

|F(x)|≤ AhM(B∥x∥),hM(t) = inf ⁡ j≥0Mjtj.
(11)

For j = 0 use the amplitude bound; for j ≥ 1 the factorial cancels in the remainder. This is the usual associated-function estimate in the factorial-out convention; compare [14, Section 7]. Coordinate partial-derivative bounds of the form (2), with fixed constants on the same ball (or on a fixed compact subball), imply the corresponding operator bounds after replacing B by Bn. In the Gevrey case Mj = (j!)s−1, s > 1, optimization in j gives hM(t) ≤ Cexp ⁡ (−ct−1∕(s−1)) for small t > 0. A matching optimal-cutoff assertion for general weights requires additional assumptions and is not used here. These estimates concern spatial derivatives and do not assert failure of an optimization or learning procedure.

7.2 Polynomial signs at positive distance from the walls

Put

Qn = [−1,1]n,D δ,n = {x ∈ Qn : |xi|≥ δ for all i},0 < δ < 1.

Let dh(δ) be the least total degree of a real polynomial P satisfying

h(sgn ⁡ x)P(x) > 0(x ∈ Dδ,n).

There is no amplitude or coefficient bound in this definition. For fixed 0 < γ < 1, let dh,γbd(δ) be the least total degree subject instead to

∥P∥Qn,∞ ≤ 1,h(sgn ⁡ x)P(x) ≥ γ(x ∈ Dδ,n).
(12)

The amplitude bound covers the entire cube, including its excluded strips.

Theorem 7.3 (Fixed-dimensional degree comparison). For fixed n ≥ 1 the following statements hold.

(i)
If h(s) = c∏ ⁡ i∈Ssi, then dh(δ) = |S| for every 0 < δ < 1.
(ii)
If h is non-affine, then dh(δ) = Θh(log ⁡ (1∕δ)) as δ ↓ 0.
(iii)
If h is nonconstant and 0 < γ < 1 is fixed, then
dh,γbd(δ) = Θ h,γ(1∕δ)(δ ↓ 0).

In fact, dh,γbd(δ) ≥ γ1 − δ2∕δ.

Constants may depend on the fixed dimension, law, and margin. No uniform growing-arity, coefficient-size, or evaluation-cost assertion is intended.

Proof. Parity. The polynomial c∏ ⁡ i∈Sxi proves the upper bound. On Boolean vertices, fix coordinates outside S. Every polynomial of degree less than |S| has zero correlation with parity, by expanding into monomials and averaging over the remaining independent signs. Strict sign agreement would give positive correlation. This proves (i), including constants with S = ∅.

A logarithmic upper bound for every law. Let m = ⌈log ⁡ 2(1∕δ)⌉ and choose an integer L ≥ 1. Define

p(t) = ∏ j=0m(t + 2−j)2L,a(t) = p(t) − p(−t),q(t) = p(t) + p(−t),r(t) = a(t) q(t).

The two nonnegative polynomials p(t) and p(−t) have disjoint real zero sets, so q > 0 and |r|≤ 1 on ℝ. For 2−m ≤ t ≤ 1, some dyadic node u = 2−j satisfies u ≤ t ≤ 2u. Therefore

0 ≤ p(−t) p(t) = ∏ j=0m |2−j − t 2−j + t |2L ≤ 3−2L,|r(t) − 1|≤ 23−2L.

Oddness gives the negative-interval estimate. This is a dyadic rational construction of Newman type; see also [17, Theorem 2.4].

For u ∈ [−1,1]n, define the multilinear expectation extension

Hh(u) = ∑ s∈{−1,+1}nh(s)∏ i=1n1 + siui 2 .

It is bounded by one. If |ui − si|≤ 𝜀 for every coordinate, regard Hh(u) as the expectation of h under independent signs with means ui. Each sign differs from si with probability at most 𝜀∕2, so the union bound gives

h(s)Hh(u) ≥ 1 − n𝜀.
(13)

Choose L with 2n3−2L < 1. Substitution of r(xi) and multiplication by ∏ ⁡ iq(xi) > 0 give a polynomial of total degree at most 2Ln(m + 1), strictly correct on Dδ,n. Hence dh(δ) = On(log ⁡ (1∕δ)).

The non-affine lower bound. Every non-affine Boolean law has a non-affine two-dimensional face: otherwise every second Boolean difference of its 𝔽2-valued truth function would vanish identically, forcing its multilinear algebraic normal form to have degree at most one. Fix the other coordinates at the corresponding ± 1 values. Each of the eight non-affine binary laws is AND after input sign flips and possibly output negation, all of which preserve degree and the guarded square.

Choose an odd integer N ≤ 1∕δ with N ≍ 1∕δ. In a polynomial for binary AND substitute

x = 1 N∑ i=1Nz i,y = 1 N∑ i=1Nw i,zi,wi ∈{−1,+1}.

Each magnitude is at least 1∕N ≥ δ. Thus the substituted polynomial sign-represents the conjunction of two independent N-variable majorities on a finite Boolean cube, without increasing degree. Sherstov’s lower bound [17, Theorem 1.10] is Ω(log ⁡ N). His convention uses − 1 for true; input and output sign changes align the conventions. This proves (ii).

Bounded margin: lower bound. A nonconstant law changes along some Boolean edge. Fix the other coordinates at that edge’s ± 1 values. The resulting univariate slice p is bounded by one on [−1,1] and takes opposite values of magnitude at least γ at ± δ. The mean value theorem gives ξ ∈ (−δ,δ) with |p′(ξ)|≥ γ∕δ. Bernstein’s inequality [18, Theorem 24.1] gives, with d = deg ⁡ P,

γ δ ≤|p′(ξ)|≤ d 1 − ξ2 ≤ d 1 − δ2.

Bounded margin: upper bound. The clipped ramp vδ(t) = max ⁡ (−1,min ⁡ (1,t∕δ)) is 1∕δ-Lipschitz. Jackson’s inequality yields a polynomial u of degree at most k with

∥u − vδ∥[−1,1],∞ ≤ C kδ.

For completeness, the trigonometric form [18, Theorem 21.1] applies to the even periodic function vδ(cos ⁡ 𝜃). Symmetrizing its approximant preserves the error and leaves only cosine terms; cos ⁡ (j𝜃) = Tj(cos ⁡ 𝜃) converts it to an algebraic polynomial of the same degree. Choose a fixed error η > 0 and k ≥ C∕(ηδ). The polynomial r = u∕(1 + η) is bounded by one on the whole interval and differs from sgn ⁡ t by at most 𝜀 = 2η∕(1 + η) when |t|≥ δ. Choose η small enough that n𝜀 ≤ 1 − γ. Then Hh(r(x1),…,r(xn)) satisfies (12), by (13), and has total degree at most nk = On,γ(1∕δ). This proves (iii). □

Remark 7.4 (Interpretation and endpoints). Positive denominator clearing preserves signs but need not preserve a fixed margin after normalizing the polynomial on Qn. This explains why (ii) and (iii) concern different resources. Constants have degree zero in both models; when γ = 1, no nonconstant law satisfies (12), since the polynomial would equal a constant on an open box and hence everywhere. The theorem is a synthesis of established approximation inputs. It does not follow from qualitative finite-jet rigidity alone, nor does it apply that rigidity to a polynomial required to be correct only on Dδ,n.

8 Scope, counterexamples, and relation to nearby theory

8.1 Real zero support is weaker than a normal-crossing divisor

If one already has a factorization

F = u∏ ixiai

with a nonvanishing unit u, then the sign law is immediately the sign of u times a parity product of coordinate signs. Normal-crossing and monomialization theories therefore provide important neighboring context; see [11, 10]. They should not, however, be substituted for the weaker hypothesis used in Theorems 4.5 and 6.1.

Indeed,

F(x,y) = xy(x2 + y2)
(14)

is real analytic and has real zero set exactly equal to the union of the two coordinate axes. Its sign off those axes is sgn ⁡ (xy). But after extracting the generic first-order factors x and y, the residual factor x2 + y2 still vanishes at the origin and is not a unit. Thus set-theoretic real zero support on an arrangement does not imply a unit-times-monomial factorization in the original coordinates.

This is why the arrangement theorem above is stated and proved directly at the sign level. Resolution and monomialization results involve stronger divisor hypotheses and/or transformations of source and target; they remain relevant background, not a replacement for the present argument.

8.2 A common corner is essential for local orthant rigidity

The connectedness of a domain alone does not replace the common-corner hypothesis in Theorem 4.1. On the punctured plane U = ℝ2 ∖{0}, define

F(x,y) = x + y + x2 + y2.
(15)

This function is real analytic on U. It is positive in the positive-positive quadrant and in both mixed-sign quadrants, while it is negative in the negative-negative quadrant. In a mixed-sign quadrant,

x2 + y2 > |x + y|

because xy < 0, whereas in the negative-negative quadrant

x2 + y2 < |x + y| = −x − y.

Thus (15) realizes OR on the four quadrants of the punctured plane. There is no contradiction with Theorem 4.1: the origin, at which the finite-jet obstruction would live, has been removed.

8.3 Classical R-functions and finite differentiability

The preceding root formula is the classical R-disjunction. Changing the plus sign to a minus sign gives an R-conjunction; see [16, Section 3.3, equations (3.4)–(3.5)]. In particular, for any integer k ≥ 0, put r = (x2 + y2)1∕2 and

Fk(x,y) = rk(x + y − r),F k(0,0) = 0.

This realizes AND on the open quadrants. Its smooth angular factor and positive homogeneity of degree k + 1 give DjFk = O(rk+1−j) for j ≤ k; these derivatives extend by zero at the origin. The kth derivative is Lipschitz: for pairs at distance comparable to their radius use its O(r) size bound, and for closer pairs integrate its bounded derivative along a segment away from zero. Thus Fk ∈ Ck,1 and its k-jet vanishes. It is not Ck+1: that regularity and homogeneity would make it its homogeneous Taylor polynomial of degree k + 1, contrary to the non-affine sign law. On (t,t), t > 0, its value is 2k∕2(2 −2)tk+1. Hence the exponent k + 1 in Proposition 7.1 for α = 1 cannot be increased in general. The construction is classical; the estimate states its relation to the finite-order obstruction.

8.4 Analytic models and restricted domains

A finite composition of real-analytic maps is real analytic on its domain. If a scalar output realizes an orthant law on a full neighborhood of the common corner, Corollary 4.4 therefore forces affine parity. The hypothesis concerns independently variable coordinates throughout that neighborhood; it does not cover merely fitting finitely many Boolean representatives or a restricted set of reachable inputs.

Even an analytic formula can realize AND on every Dδ,2 when it depends on δ. For example,

Fδ(x,y) = x + y −x2 + y2 + δ2

is globally real analytic. In the positive-positive guarded quadrant it is increasing in each coordinate, so its minimum is (2 −3)δ > 0. If x ≤−δ, then Fδ(x,y) < x + y −|y|≤−δ, and the other negative-input case is symmetric. At the common corner Fδ(0,0) = −δ, so it fails the positive-positive law sufficiently near zero. This example imposes no global amplitude normalization and does not contradict Theorem 7.3. A nonsmooth alternative is min ⁡ (x,y) = x − max ⁡ (0,x − y), which realizes AND exactly on every open quadrant.

8.5 Smoothness alone does not imply rigidity

The selector construction in (9) gives C∞ realizations of every Boolean law in the non-quasianalytic case. For a non-affine law, Theorem 4.1 says that such a realization must be flat at the origin; the construction is consistent with exactly that requirement. The mathematical boundary is therefore not “smooth versus nonsmooth,” but rather the presence or absence of a nonzero finite jet at the relevant common boundary.

8.6 Signatures and chamber functions

The factor-parity mechanism in Lemma 3.1 is classical antecedent material. Broglia–Tognoli’s Proposition 2.4 records how parity of maximal factor multiplicities controls the induced signature [12, Proposition 2.4], while the subsequent signature and sign-distribution literature studies broader realization and approximation questions [5, 1, 2, 4, 3].

A distinct combinatorial tradition studies all functions that are constant on arrangement chambers. Varchenko and Gelfand introduced the algebra generated by hyperplane Heaviside functions [7]; see also [8, 6]. The present analytic question is narrower: regularity constraints select a small family of {±1}-valued chamber functions in the quasianalytic regime, while the non-quasianalytic regime permits them all.

8.7 Established machinery versus the sign classification

Stronger division, quasianalytic local-ring, resolution, and monomialization frameworks are available [15, 11, 9, 10]. The proofs here deliberately avoid that machinery: the rigidity input is only the first nonzero Taylor polynomial, and the realization input is only the classical Denjoy–Carleman criterion together with cutoffs or positive-flat functions. This is the sense in which the note is self-contained without claiming foundational priority.

9 Conclusion

A first nonzero Taylor term provides a compact bridge from local regularity to global chamber signs. In finite order it yields explicit Cd orthant and CN arrangement obstructions; in the smooth setting the same mechanism gives an elementary proof of the established signature-flatness consequence for affine arrangements. Combined with the Denjoy–Carleman criterion, the argument gives an exact realization classification: for m distinct hyperplanes with r chambers, the full normalized Roumieu class realizes exactly 2m+1 signed parity laws in the quasianalytic case and all 2r chamber labelings in the non-quasianalytic case.

With fixed derivative bounds, the same vanishing jets control how rapidly an exact non-affine score approaches zero at the corner. Excluding wall strips changes the polynomial problem: at fixed dimension non-affine sign laws require logarithmic degree in the inverse width, whereas a fixed margin under a whole-cube amplitude bound requires reciprocal-width degree for every nonconstant law. The complete reductions make these distinctions explicit. The finite-order formulation, arrangement proof, and synthesis remain the contribution; the smooth obstruction, R-function constructions, positive-flat theory, and approximation inputs retain their classical attribution.

References

[1]    F. Acquistapace and F. Broglia, Signatures and flatness, Journal für die reine und angewandte Mathematik 425 (1992), 61–70.

[2]    F. Acquistapace and F. Broglia, Correction to the paper “Signatures and flatness” (Vol. 425, pp. 61–70, 1992), Journal für die reine und angewandte Mathematik 444 (1993), 155–156.

[3]    F. Acquistapace and F. Broglia, More about signatures and approximation, Geometriae Dedicata 50 (1994), no. 2, 107–116. DOI: 10.1007/BF01265306.

[4]    F. Acquistapace, F. Broglia, and E. Fortuna, When is a distribution of signs locally completable?, Canadian Journal of Mathematics 46 (1994), no. 3, 449–473. DOI: 10.4153/CJM-1994-024-2.

[5]    E. Fortuna, Distributions de signes, Mathematika 38 (1991), no. 1, 50–62. DOI: 10.1112/S0025579300006434.

[6]    A. Björner, M. Las Vergnas, B. Sturmfels, N. White, and G. M. Ziegler, Oriented Matroids, 2nd ed., Encyclopedia of Mathematics and its Applications, vol. 46, Cambridge University Press, Cambridge, 1999.

[7]    A. N. Varchenko and I. M. Gel’fand, Heaviside functions of a configuration of hyperplanes, Functional Analysis and Its Applications 21 (1987), no. 4, 255–270. DOI: 10.1007/BF01077800.

[8]    D. Moseley, Equivariant cohomology and the Varchenko–Gelfand filtration, Journal of Algebra 472 (2017), 95–114. DOI: 10.1016/j.jalgebra.2016.10.010; arXiv:1110.5369.

[9]    A. Belotto da Silva, E. Bierstone, and M. Chow, Composite quasianalytic functions, Compositio Mathematica 154 (2018), no. 9, 1960–1973. DOI: 10.1112/S0010437X18007339; arXiv:1709.06629.

[10]    A. Belotto da Silva and E. Bierstone, Monomialization of a quasianalytic morphism, Annales Scientifiques de l’Ecole Normale Superieure 56 (2023), no. 5, 1583–1651. DOI: 10.24033/asens.2562; arXiv:1907.09502.

[11]    E. Bierstone and P. D. Milman, Resolution of singularities in Denjoy–Carleman classes, Selecta Mathematica 10 (2004), no. 1, 1–28. DOI: 10.1007/s00029-004-0327-0; arXiv:math/0108204.

[12]    F. Broglia and A. Tognoli, Approximation of C∞-functions without changing their zero-set, Annales de l’Institut Fourier 39 (1989), no. 3, 611–632. DOI: 10.5802/aif.1178.

[13]    M. Langenbruch, A general approximation theorem of Whitney type, Rev. R. Acad. Cien. Serie A. Mat. (RACSAM) 97 (2003), no. 2, 287–303. https://rac.es/ficheros/doc/00128.pdf.

[14]    A. Rainer, Ultradifferentiable extension theorems: a survey, Expositiones Mathematicae 40 (2022), no. 3, 679–757. DOI: 10.1016/j.exmath.2021.12.001; arXiv:2107.01061.

[15]    V. Thilliez, On quasianalytic local rings, Expositiones Mathematicae 26 (2008), no. 1, 1–23. DOI: 10.1016/j.exmath.2007.04.001; arXiv:math/0509273.

[16]    V. Shapiro, Semi-analytic geometry with R-functions, Acta Numerica 16 (2007), 239–303. DOI: 10.1017/S096249290631001X.

[17]    A. A. Sherstov, The intersection of two halfspaces has high threshold degree, SIAM Journal on Computing 42 (2013), no. 6, 2329–2374. DOI: 10.1137/100785260.

[18]    K. A. Kopotun, A. V. Prymak, and I. O. Shevchuk, Introduction to Approximation Theory, study guide, 24 January 2018, Theorems 21.1 and 24.1. https://mathanalysis.knu.ua/wp-content/uploads/2020/07/at-study-guide-january-22-2018-1-1.pdf.