Alphabeta Math
Pipeline-generated
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

✓ 12 results · all verified · 7 also independently AI-judged
Every result on this page is machine-checked by a proof checker and read in full and owner-audited; the judge is an additional, independent cross-model AI review of the proofs. The 5 not AI-judged were verified by owner audit (typically over a confirmed judge false positive), not failures.

Hilbert and Riesz Transforms

1 · Prerequisites

2 · Summary

This page proves the strict-range mapping theory of the periodic conjugate function and the L2 theory of the Hilbert and Riesz transforms on Rn. Every argument that needs a choice principle assumes Countable Choice and names the step that spends it; the remaining proofs are choice-free.

On the circle the conjugate function is first defined as the coefficient multiplier Cf^(k)=−isgn⁡(k)f^(k) on trigonometric polynomials, so constants lie in its kernel and no extension is built into the definition. The conjugate Dirichlet kernel is computed exactly and identified with the cotangent principal value under local C1 regularity; the finite kernel convolution is kept distinct from the limiting truncation. The square identity (Cg)2=g2+2C(gCg) for real mean-zero polynomials, Riesz interpolation of the bounded L² action, and complex Lp duality give the Marcel Riesz conjugate-function theorem: C extends uniquely to a bounded operator on Lp(T) for every 1<p<∞, while the exact L1 operator norm of the partial sums grows logarithmically and rules out compatible strong L1 or L∞ extensions. The same Lebesgue-constant lower bound, together with the Fejér means which converge in Lp, yields uniform Lp bounds for the Fourier partial sums and their norm convergence in the strict range.

On the line the page defines the truncated Hilbert transform Hε and its principal value, proves the uniform sine-integral bounds and the value π/2 under Countable Choice, and shows that on Schwartz functions the principal value is the tempered convolution with pv 1/(πx), whose Fourier multiplier is −isgn⁡ξ. The multiplier produces the L2 extension with ∥Hf∥2=∥f∥2, H2=−I, and skew-adjointness, with no zero-frequency exception on the line. For Rn the Riesz transforms are defined by the multipliers −iξj/∣ξ∣; polar coordinates identify that multiplier with the principal-value kernel cnxj/∣x∣n+1, giving the L2 bound, the finite square-sum identity ∑jRj2=−I, and the kernel size, first-difference and spherical cancellation estimates needed by the later singular-integral theory.

The closing remark maps the endpoint statements this page does not claim — weak (1,1), real-line strict-range Lp, almost-everywhere truncation convergence, real Hardy space and BMO — to the later pages that own them, and the companion page carries the interval, Poisson-kernel and endpoint counterexamples that do not refute those weaker conclusions.

3 · Logical flowchart

4 · Definitions, theorems and proofs

DefinitionDefinition: Literature-sourcedProof: Not applicablejudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

Conjugate function on the circle

Definition

Work on the torus T=R/Z with the conventions of Period-one Fourier coefficients, partial sums, and convolution on the torus: the characters are ek(x)=e2πikx for k∈Z, the Fourier coefficients of a one-period integrable f are f^(k)=∫01f(t)e−2πikt dt, and a trigonometric polynomial is a finite complex linear combination of characters.

Let f=∑∣k∣≤Mckek be a trigonometric polynomial, so that f^(k)=ck for ∣k∣≤M and f^(k)=0 for ∣k∣>M. The conjugate function of f is the trigonometric polynomial

Cf:=∑0<∣k∣≤M(−isgn⁡(k))f^(k) ek,sgn⁡(k):={1,k>0,0,k=0,−1,k<0.

Equivalently, Cf is the unique trigonometric polynomial whose Fourier coefficients are

Cf^(k)=−isgn⁡(k) f^(k)(k∈Z).

Consequently Cf^(0)=0: every constant trigonometric polynomial lies in the kernel of C.

The assignment C is complex-linear on the space of trigonometric polynomials. It also preserves real-valuedness: if f is real, then f^(−k)=f^(k)‾ for every k, and −isgn⁡(−k)f^(k)‾=−isgn⁡(k)f^(k)‾, so Cf^ has the conjugate symmetry that characterizes a real trigonometric polynomial.

Two conventions are fixed by this definition. The factor −i and the sign refer to the characters ek(x)=e2πikx and to the coefficient convention above; conjugating real functions, as in f↦−Cf, is the opposite sign convention. And C is defined on trigonometric polynomials alone: no bound on any Lp(T) norm and no extension to arbitrary integrable functions is asserted here. The extension to Lp(T) for 1<p<∞ is proved on this page after the kernel formula below.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passaudited 2026-10-02Open item page →

The conjugate Dirichlet kernel, and the periodic principal-value formula

Statement

Assume Countable Choice, and work on T=R/Z with the conventions of Period-one Fourier coefficients, partial sums, and convolution on the torus. Let f∈L2(T;C) and let N≥1. Put

KN(t):=2∑k=1Nsin⁡(2πkt)=cos⁡(πt)−cos⁡((2N+1)πt)sin⁡(πt)(t∉Z),

the conjugate Dirichlet kernel, and write CSNf=∑0<∣k∣≤N(−isgn⁡k)f^(k)ek for the conjugate partial sum. Then:

  1. CSNf is the convolution of f with KN: for almost every x, CSNf(x)=∫01KN(t)f(x−t) dt.
  2. Extend C to L2(T;C) by the square-summable coefficient family (−isgn⁡kf^(k))k∈Z; the resulting class Cf is the L2 limit of the partial sums CSNf.
  3. If a representative of f is C1 on an open interval containing x, then

Cf(x)=lim⁡ε↓0∫ε<∣t∣<1/2f(x−t)cot⁡(πt) dt,

the limit existing for almost every such x, and being the symmetric principal value about the singularity.

The finite kernel KN is not itself a cotangent truncation: KN equals cot⁡(πt) minus the oscillatory remainder cos⁡((2N+1)πt)/sin⁡(πt), and only the limit N→∞ of the convolutions recovers the principal value.

Facts & Assumptions

Given: Countable Choice, f∈L2(T;C), N≥1, and the characters ek(x)=e2πikx.

[F1]

C is defined on trigonometric polynomials coefficientwise by Cg^(k)=−isgn⁡(k)g^(k), it is complex-linear, kills constants, and preserves real-valuedness. Conjugate function on the circle

[F2]

Fourier coefficients, partial sums SNf, characters, and the torus convolution (f∗g)(x)=∫01f(x−t)g(t)dt are as defined there, and the torus integral is invariant under the reflections used below. Period-one Fourier coefficients, partial sums, and convolution on the torus

[F3]

The Dirichlet kernel is DN(t)=∑∣k∣≤Nek(t). Dirichlet and Fejer kernels

[F4]

For one-period integrable f, SNf(x)=∫01f(x−t)DN(t) dt=(f∗DN)(x) for every x. Fourier partial sums are Dirichlet convolutions

[F5]

For N≥1 and x∉2πZ, ∑n=1Nsin⁡(nx)=cos⁡(x/2)−cos⁡((N+1/2)x)2sin⁡(x/2). Finite sums of the sine harmonics

[F6]

Parseval: ∥f∥22=∑k∈Z∣f^(k)∣2 in the finite-subset-supremum sense, so the tails over {∣k∣>N} tend to 0. The Parseval identity for Fourier series

[F7]

Every square-summable coefficient family in ℓ2(Z;C) is the Fourier coefficient family of a unique L2 class, realized as the L2 limit of its symmetric partial sums. Riesz–Fischer: the Fourier coefficient map is onto the space of square-summable families

[F8]

Riemann-Lebesgue: if g is integrable on one period then g^(k)→0 as ∣k∣→∞. Riemann-Lebesgue lemma for Fourier coefficients

[F9]

Norm-convergent sequences in L2 have subsequences converging almost everywhere to a representative of the limit. Complex Lp completeness and almost-everywhere subsequences

[F10]

The real mean value theorem bounds the increment of a real C1 function by the supremum of its derivative times the interval length. Applied separately to the real and imaginary parts, it gives ∣g(t)−g(0)∣≤C∣t∣ for a complex C1 function on a compact interval about 0, with C=∥Re⁡g′∥∞+∥Im⁡g′∥∞. The mean value theorem, as the case g(x)=x of Cauchy's: for f continuous on [a,b] with a<b and differentiable on (a,b) there is c∈(a,b) with f(b)−f(a)=f′(c)(b−a)

[F11]

Dominated convergence. Dominated convergence

[F12]

Cosine addition formula: cos⁡(A+B)=cos⁡Acos⁡B−sin⁡Asin⁡B. The addition formulas for sine and cosine

Proof

technique · direct
1.1F1F3F5

By [F1] and [F3], KN:=CDN is the trigonometric polynomial with coefficients −isgn⁡(k) on 0<∣k∣≤N and 0 elsewhere, so KN(t)=∑0<∣k∣≤N(−isgn⁡k)ek(t)=2∑k=1Nsin⁡(2πkt), using ek−e−k=2isin⁡(2πkt). Applying [F5] with x=2πt gives 2∑k=1Nsin⁡(2πkt)=cos⁡(πt)−cos⁡((2N+1)πt)sin⁡(πt) for t∉Z, while KN(0)=0; in particular KN is odd, one-periodic, and ∫−1/21/2KN(t) dt=0.

1.2F1F2F4

By [F1], the conjugate partial sum is the trigonometric polynomial CSNf=∑0<∣k∣≤N(−isgn⁡k)f^(k)ek. Expanding KN from 1.1 and substituting u=x−t in each finite sum as in [F2] and [F4], ∫01KN(t)f(x−t) dt=∑0<∣k∣≤N(−isgn⁡k)f^(k)ek(x)=CSNf(x) for every x. The bounded finite kernel makes the integral exist for each translate of the L1 representative, and the finite coefficient calculation is exact; this keeps the finite-N object a polynomial-level convolution and makes no claim on any cotangent kernel.

2.1step 1.1step 1.2F10

Fix a point x at which some representative of f is C1 on an open interval containing x, and put g(t):=f(x−t)−f(x) for ∣t∣<1/2, so that ∣g(t)∣≤C∣t∣ for a constant C and all small t by [F10]. Step 1.2 gives CSNf(x)=∫01KN(t)f(x−t) dt; since KN is one-periodic, odd, and has ∫−1/21/2KN=0 by 1.1, that integral equals ∫∣t∣<1/2KN(t)g(t) dt. The closed form of 1.1 splits the kernel as cot⁡(πt)−cos⁡((2N+1)πt)/sin⁡(πt), so CSNf(x)=A(x)−RN(x) with A(x):=∫∣t∣<1/2cot⁡(πt)g(t) dt and RN(x):=∫∣t∣<1/2cos⁡((2N+1)πt)sin⁡(πt)g(t) dt, the integrands being defined and measurable off the null point t=0.

2.2step 1.2F6F7F9

Put ak:=−isgn⁡(k)f^(k). Since ∣ak∣≤∣f^(k)∣ for every k (with a0=0), [F6] gives ∑k∣ak∣2≤∥f∥22<∞. Thus [F7] supplies a unique class Cf∈L2(T;C) whose symmetric partial sums are exactly the CSNf of 1.2 and which is their L2 limit; by [F9] there is an increasing sequence Nj→∞ with CSNjf(x)→Cf(x) for almost every x.

3.1step 2.1F8F10F12

For the point x of 2.1, [F12] writes cos⁡((2N+1)πt)=cos⁡(2πNt)cos⁡(πt)−sin⁡(2πNt)sin⁡(πt), so RN(x)=∫∣t∣<1/2cos⁡(2πNt)cos⁡(πt)g(t)sin⁡(πt) dt−∫∣t∣<1/2sin⁡(2πNt)g(t) dt. Both t↦cos⁡(πt)g(t)/sin⁡(πt) and t↦g(t) are integrable on (−1/2,1/2): the second because f is L1 on the finite torus and the quotient is bounded near 0 by C; away from 0, 1/sin⁡(πt) is bounded and g∈L1, so the quotient is integrable there as well. Extending them by zero to one period, [F8] gives h^(N)→0 for these integrable functions, hence RN(x)→0 as N→∞; the convergence is at every such x, and no uniformity in x is claimed.

3.2step 2.1F10F11

Also at the point x of 2.1, for 0<ε<1/2 the oddness of cot⁡ gives ∫ε<∣t∣<1/2cot⁡(πt)f(x−t) dt=∫ε<∣t∣<1/2cot⁡(πt)g(t) dt, and by [F10] the function cot⁡(πt)g(t) is integrable on (−1/2,1/2); [F11] therefore gives ∫ε<∣t∣<1/2cot⁡(πt)g(t) dt→A(x) as ε↓0. So the symmetric principal value exists at x and equals A(x).

4.1step 2.2step 3.1step 3.2∎

Combining 3.1 and 3.2, for every x at which f is C1 near x the finite convolutions satisfy CSNf(x)=A(x)−RN(x)→A(x), so the full sequence CSNf(x) converges to the principal value at every such x; by 2.2 it also converges to Cf(x) along a subsequence for almost every x. Therefore Cf(x)=lim⁡ε↓0∫ε<∣t∣<1/2f(x−t)cot⁡(πt) dt for almost every x in the open set where f is C1 near x, as asserted.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

The periodic conjugate square identity for real mean-zero polynomials

Statement

Let g be a real-valued trigonometric polynomial on T=R/Z with zero mean, and let C be the conjugate function of Conjugate function on the circle. Then

(Cg)2=g2+2 C(g Cg).

The identity is asserted for trigonometric polynomials only; no extension to arbitrary Lp(T) inputs is claimed here, and the mean-zero hypothesis is not removable: for the constant polynomial g=1 one has Cg=0 and C(gCg)=0, so the right-hand side equals 1 while the left-hand side vanishes.

Facts & Assumptions

Given: A real-valued trigonometric polynomial g on T with zero mean; the conjugate function C on trigonometric polynomials.

[F1]

On trigonometric polynomials C acts coefficientwise by Cf^(k)=−isgn⁡(k)f^(k); it is complex-linear, kills constants, and preserves real-valuedness, so Cg is real-valued for real g. Conjugate function on the circle

[F2]

Characters satisfy ekel=ek+l and a trigonometric polynomial is a finite complex linear combination of characters. Fourier coefficients and trigonometric polynomials on the torus

[F3]

If p=∑k∈Fckek is a trigonometric polynomial, then its Fourier coefficient at j∈F is p^(j)=cj and p^(j)=0 for j∉F. The trigonometric characters are orthonormal in L2 of the torus

Proof

technique · direct
1.1F1F2F3

Write the finite expansion g=∑k∈Fckek given by [F2]. Since g has mean zero and 0∈F, [F3] gives c0=g^(0)=0; since g is real-valued, [F1] makes Cg=∑k∈F(−isgn⁡(k))ckek real-valued, so Cg=∑k∈F,0≠k(−isgn⁡k)ckek has zero mean as well. Adding the two expansions and using complex-linearity of C from [F1] gives F:=g+iCg=∑k∈Fck(1+sgn⁡k)ek=∑k>02ckek: a trigonometric polynomial whose only frequencies are strictly positive.

2.1step 1.1F1F2

By [F2], ekel=ek+l, so expanding the finite square and collecting terms shows that F2=∑k,l>04ckclek+l is a trigonometric polynomial whose only frequencies are strictly positive. On a polynomial carried by the characters ek with k>0, the coefficient rule of [F1] gives C(ek)=−iek, and complex-linearity gives C(F2)=−iF2.

2.2step 1.1F1

Expanding, F2=(g+iCg)2=g2−(Cg)2+2i g Cg; by 1.1 both u:=g2−(Cg)2 and v:=2g Cg are real-valued trigonometric polynomials with real-valued conjugate transforms, and F2=u+iv. By complex-linearity of C, C(F2)=Cu+iCv.

3.1step 2.1step 2.2

By 2.1 and 2.2, Cu+iCv=−iF2=−i(u+iv)=v−iu. Taking real and imaginary parts of this identity of trigonometric polynomials, whose four real and imaginary parts are real-valued by 2.2, gives Cu=v and Cv=−u.

4.1step 3.1F1∎

Substituting u=g2−(Cg)2 and v=2g Cg into Cv=−u gives 2C(gCg)=(Cg)2−g2, hence (Cg)2=g2+2C(gCg), which is the asserted identity.

TheoremStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

The Marcel Riesz conjugate-function theorem on the circle

Statement

Assume Countable Choice and the conventions of Period-one Fourier coefficients, partial sums, and convolution on the torus on the torus T=R/Z with normalized Haar measure m, so that m(T)=1: characters ek(x)=e2πikx, Fourier coefficients f^(k)=∫01f(t)e−2πikt dt, trigonometric polynomials as finite complex linear combinations of characters, and C the conjugate function of Conjugate function on the circle, with Cf^(k)=−isgn⁡(k)f^(k) and sgn⁡(0)=0.

  1. For every 1<p<∞ the operator C extends uniquely from the trigonometric polynomials to a bounded complex-linear operator Cp:Lp(T;C)→Lp(T;C), and the extensions are mutually consistent: Cpf=Cqf almost everywhere for every f∈Lp(T;C)∩Lq(T;C).
  2. Constants lie in the kernel: Cp1=0 for every 1<p<∞.
  3. No compatible endpoint extension exists: there is no bounded operator U:L1(T;C)→L1(T;C) with Up=Cp for every trigonometric polynomial p, and no bounded operator V:L∞(T;C)→L∞(T;C) with Vp=Cp for every trigonometric polynomial p. The failure is of strong-type boundedness; assertions about weak type (1,1), maximal truncations or a bounded mean-oscillation range are not made here.

Facts & Assumptions

Given: Countable Choice; the torus T=R/Z with normalized Haar measure m; the conjugate function C on trigonometric polynomials, Cf^(k)=−isgn⁡(k)f^(k).

[F1]

On trigonometric polynomials C is complex-linear, kills constants and preserves real-valuedness; the characters satisfy eaeb=ea+b and a trigonometric polynomial has only finitely many nonzero Fourier coefficients. Conjugate function on the circle Period-one Fourier coefficients, partial sums, and convolution on the torus

[F2]

For every real mean-zero trigonometric polynomial g one has (Cg)2=g2+2C(gCg). The periodic conjugate square identity for real mean-zero polynomials

[F3]

Parseval: for f,g∈L2(T;C), ∥f∥22=∑k∣f^(k)∣2 and ⟨f,g⟩=∫Tfg‾ dm=∑kf^(k)g^(k)‾, the sums being finite-subset-net limits whose value is also the limit of the symmetric partial sums ∑∣k∣≤N. The Parseval identity for Fourier series

[F4]

Hölder and Minkowski for complex Lp: for conjugate exponents and complex measurable functions, the integral of a product is bounded by the product of the norms, and the norm of a sum by the sum of the norms. Complex Holder, Minkowski, and the quotient norm

[F5]

Cesàro means: σNf=f∗FN; for 1≤p<∞ and f∈Lp(T;C) one has ∥σNf−f∥p→0; for continuous one-periodic f one has sup⁡x∣σNf(x)−f(x)∣→0. Cesaro and Abel means of a Fourier series Fejer means converge in L^p for 1 <= p < infinity Fejer means converge uniformly for continuous periodic functions

[F6]

The Fejér kernel satisfies FN≥0 and ∫01FN=1. The Fejer kernel is a positive approximate identity

[F7]

For one-period integrable f, SNf(x)=∫01f(x−t)DN(t) dt at every x, where DN is real-valued, even and ∫01DN=1. Fourier partial sums are Dirichlet convolutions Dirichlet and Fejer kernels

[F8]

On the continuous periodic functions with the supremum norm, ∥SN:C(T)→C(T)∥=∫01∣DN(t)∣ dt, and for N≥1 this number is at least 13πlog⁡(N+1). Fourier partial-sum operator norm equals the Lebesgue constant

[F9]

For every measure space and 1≤p≤∞ the complex space Lp is complete. Complex Lp completeness and almost-everywhere subsequences

[F10]

Riesz–Thorin: a complex-linear map defined on the complex finite simple functions with finite-measure support which is bounded with constants M0,M1 between Lp0→Lq0 and Lp1→Lq1, 1≤pi<∞, 1<qi<∞, extends uniquely to a bounded operator Lpθ→Lqθ with norm at most M01−θM1θ. Riesz-Thorin interpolation theorem

[F11]

For a finite measure space and 1<p<∞, every bounded complex-linear functional on Lp is integration against a unique h∈Lq with the bilinear pairing, and the norms agree. Complex Lp duality from real Lp duality

[F12]

Norm recovery: for 1≤p<∞ (and p=1 only for sigma-finite measures, which includes T), ∥f∥p=sup⁡{∣∫fg dm∣:∥g∥q≤1}. The Lp norm is the supremum of pairings against unit Lq functions

[F13]

On a finite measure space, ∥f∥p≤m(X)1/p−1/r∥f∥r for 1≤p<r<∞ and ∥f∥p≤m(X)1/p∥f∥∞. Finite-measure Lr includes into Lp for p<r

[F14]

Tonelli/Fubini for L1 functions on sigma-finite product spaces. Fubini's theorem for L^1 functions on a sigma-finite product

Proof

technique · direct
1.1F1F3givenalgebra

Let g be a real mean-zero trigonometric polynomial. By [F3] and the coefficient rule of [F1], ∥Cg∥22=∑k∣Cg^(k)∣2=∑k≠0∣g^(k)∣2=∑k∣g^(k)∣2=∥g∥22, where g^(0)=0 is used in the middle equality; again by [F1], Cg is real-valued with Cg^(0)=0, so Cg is real with zero mean. Moreover [F3] with the coefficient rule gives ∫Tg Cg=∑kg^(k)Cg^(k)‾=i∑ksgn⁡(k)∣g^(k)∣2, which is i times the real number ∑ksgn⁡(k)∣g^(k)∣2; since g Cg is real-valued, its integral is real, so ∫Tg Cg=0 and g Cg is a real mean-zero trigonometric polynomial.

1.2F1givenalgebra

For a trigonometric polynomial h put P+h:=∑k≥0h^(k)ek, and for a∈Z put Mah:=eah. Then P+ is complex-linear, (Mah) ^(k)=h^(k−a) for every k, and for every trigonometric polynomial h and every N≥1 one has SNh=M−NP+MNh−MN+1P+M−(N+1)h. Indeed the definitions and eaeb=ea+b of [F1] give (M−NP+MNh) ^(k)=1{k≥−N}h^(k) and (MN+1P+M−(N+1)h) ^(k)=1{k≥N+1}h^(k) for every k, and subtracting gives 1{∣k∣≤N}h^(k)=SNh^(k) for every k, which identifies the two trigonometric polynomials.

1.3F7F14givenalgebra

For f∈L1(T) and g∈L∞(T), ∫T(SNf)g dm=∫Tf(SNg) dm. Indeed [F7] gives SNf(x)=∫01f(x−t)DN(t) dt and SNg(s)=∫01g(s−u)DN(u) du with DN real and even, so the double integral of ∣f(x−t)DN(t)g(x)∣ is at most ∥f∥1∥g∥∞∥DN∥∞<∞ and [F14] applies; the substitution s=x−t and evenness of DN turn ∫ ⁣ ⁣∫f(x−t)DN(t)g(x) dt dx into ∫f(s)(∫DN(s−x)g(x) dx)ds=∫f(s)SNg(s) ds.

1.4F5F6givenalgebra

For f∈L1(T), each σjf is a trigonometric polynomial with ∥σjf∥1≤∥f∥1 and ∥σjf−f∥1→0; for f∈L∞(T), each σjf is a trigonometric polynomial with ∥σjf∥∞≤∥f∥∞ and ∥σjf−f∥1→0. This is [F5] with p=1 together with [F6]: from Fj≥0 and ∫01Fj=1 one gets ∣σjf∣=∣f∗Fj∣≤∣f∣∗Fj, hence ∥σjf∥1≤∥f∥1 and, for f∈L∞, ∣σjf(x)∣≤∥f∥∞ for every x.

1.5F7givenalgebra

For g∈L1(T) and every real x one has ∣SNg(x)∣≤∥DN∥∞∥g∥1 by [F7], hence ∥SNg∥∞≤∥DN∥∞∥g∥1: on the unit ball of L1, every SN is bounded by ∥DN∥∞.

2.1step 1.1F2F4givenalgebra

Put Am:=sup⁡{∥Cg∥2m/∥g∥2m:g a real mean-zero trigonometric polynomial, g≠0}. Then A1=1 by 1.1, and Am+1≤Am+Am2+1≤2Am+1 for every m≥1, so every Am is finite. Indeed fix m, put p:=2m and let g≠0 be real with zero mean; the square identity [F2], the identification of g Cg as a real mean-zero trigonometric polynomial in 1.1, and the definition of Am give ∥Cg∥2p2=∥(Cg)2∥p≤∥g∥2p2+2∥C(gCg)∥p≤∥g∥2p2+2Am∥g Cg∥p, while [F4] applied with exponents 2,2 to the functions ∣g∣p and ∣Cg∣p gives ∥g Cg∥p≤∥g∥2p∥Cg∥2p. Dividing by ∥g∥2p2>0 and writing u:=∥Cg∥2p/∥g∥2p≥0 yields u2≤1+2Amu, hence u≤Am+Am2+1; taking the supremum over g gives the recursion.

2.2step 1.2step 1.4step 1.5givenalgebra

Suppose a bounded linear V:L∞(T)→L∞(T) satisfies Vp=Cp for every trigonometric polynomial p; put M:=∥V∥ and Af:=12((∫Tf dm)1+f+iVf) for f∈L∞. Then ∥A∥L∞→L∞≤1+12M and Ap=P+p for every trigonometric polynomial p, by 1.2: the multiplier of 12(f+iVf) is 1 on positive frequencies, 0 on negative frequencies and 1/2 at zero, and the half-mean term supplies the remaining 1/2 at zero. For h∈L∞ with ∥h∥∞≤1 let hj:=σjh: by 1.4 these are trigonometric polynomials with ∥hj∥∞≤1 and ∥hj−h∥1→0, and 1.2 gives SNhj=M−NAMNhj−MN+1AM−(N+1)hj because M±ahj is again a trigonometric polynomial on which A acts as P+. Hence ∥SNhj∥∞≤2∥A∥L∞→L∞∥hj∥∞≤2+M, while 1.5 with ∥hj−h∥1→0 gives ∥SNhj−SNh∥∞→0, so ∥SNh∥∞≤2+M. Therefore ∥SN∥L∞→L∞≤2+M for every N, and in particular the continuous functions give ∥SN∥C→C≤2+M.

2.3step 1.3step 1.4F5F6F8givenalgebra

By [F8], ∥SN:C(T)→C(T)∥=∫01∣DN∣≥13πlog⁡(N+1) for every N≥1. Write LN:=∫01∣DN∣. Given δ>0, choose a continuous one-periodic g with ∥g∥∞≤1 and ∥SNg∥∞≥LN−δ/2. Choose x0 with ∣SNg(x0)∣=∥SNg∥∞, let c have modulus one with cSNg(x0)=∣SNg(x0)∣ (take c=1 if this value is zero), and set ψj:=cFj(⋅−x0). Then ∥ψj∥1=1 by [F6], and ∫Tψj(SNg) dm=c σj(SNg)(x0)→∥SNg∥∞ because SNg is continuous and its Cesaro means converge uniformly by [F5]. By pairing symmetry from 1.3 and ∥g∥∞≤1, for all sufficiently large j we have ∥SNψj∥1≥∣∫T(SNψj)g dm∣=∣∫Tψj(SNg) dm∣≥∥SNg∥∞−δ/2≥LN−δ. Since ∥ψj∥1=1, this gives ∥SN∥L1→L1≥LN−δ for every δ>0, hence ∥SN∥L1→L1≥LN≥13πlog⁡(N+1).

3.1step 2.2step 1.2step 1.4step 1.5givenalgebra

Suppose a bounded linear U:L1(T)→L1(T) satisfies Up=Cp for every trigonometric polynomial p; put M:=∥U∥ and define A on L1 by the same formula Af:=12((∫Tf dm)1+f+iUf). Then ∥A∥L1→L1≤1+12M and Ap=P+p for every trigonometric polynomial p, by the frequency check in 2.2. For f∈L1 with ∥f∥1≤1 and its Cesàro means fj=σjf, which are trigonometric polynomials with ∥fj∥1≤1 by 1.4, identity 1.2 gives SNfj=M−NAMNfj−MN+1AM−(N+1)fj and hence ∥SNfj∥1≤2∥A∥L1→L1≤2+M; by 1.5, ∥SNfj−SNf∥1≤∥DN∥∞∥fj−f∥1→0, so ∥SNf∥1≤2+M. Therefore ∥SN∥L1→L1≤2+M for every N.

3.2step 2.2step 2.3given

No bounded V:L∞(T)→L∞(T) satisfies Vp=Cp on trigonometric polynomials: such a V would give ∥SN∥C→C≤2+∥V∥ for every N by 2.2, while ∥SN∥C→C≥13πlog⁡(N+1) grows without bound by 2.3, a contradiction for N large.

3.3step 2.1F4F5F9givenalgebra

For every m≥1 there is a finite constant Km with ∥Cf∥2m≤Km∥f∥2m for every complex trigonometric polynomial f: writing f=Re⁡f+iIm⁡f and applying 2.1 to the real mean-zero parts gives ∥C(Re⁡f)∥2m≤Am∥Re⁡f−Re⁡f‾∥2m≤2Am∥Re⁡f∥2m≤2Am∥f∥2m, and likewise for the imaginary part, so Km:=4Am works; here ∥Re⁡f−Re⁡f‾∥≤2∥Re⁡f∥ uses ∣Re⁡f‾∣≤∥Re⁡f∥1≤∥Re⁡f∥2m and [F4]. Since the trigonometric polynomials are dense in L2m(T;C) by [F5] and that space is complete by [F9], C therefore has a unique extension to a bounded complex-linear operator C(m) on L2m with ∥C(m)∥≤Km; uniqueness holds because two continuous extensions of one map agree on the dense polynomial core.

4.1step 3.1step 2.3given

No bounded U:L1(T)→L1(T) satisfies Up=Cp on trigonometric polynomials: such a U would give ∥SN∥L1→L1≤2+∥U∥ for every N by 3.1, while ∥SN∥L1→L1≥13πlog⁡(N+1) grows without bound by 2.3, a contradiction for N large.

4.2step 1.1step 3.3F1givenalgebra

For all f,g∈L2(T;C) one has ∫T(C(1)f)g dm=−∫Tf(C(1)g) dm for the bilinear pairing. Indeed, for trigonometric polynomials f,g the coefficient identity ∫T(Cf)g=∑kCf^(k)g^(−k) and the rule Cf^(k)=−isgn⁡(k)f^(k) of [F1] give ∫T(Cf)g=−i∑ksgn⁡(k)f^(k)g^(−k)=−∫Tf(Cg), since Cg^(−k)=−isgn⁡(−k)g^(−k)=isgn⁡(k)g^(−k); both pairings are bounded bilinear functionals on L2×L2 (bounded by ∥f∥2∥g∥2 times the norm of C(1)), and they agree on the dense polynomial core, so they agree everywhere.

4.3step 3.3F5F10F13givenalgebra

Let p∈[2,∞). If p=2k, set Cp:=C(k). Otherwise choose m≥1 with 2m<p<2m+1 and let T be C(1) restricted to the complex finite simple functions on T. For r=2m and r=2m+1, approximate a simple function h by trigonometric polynomials in Lr using [F5]; since r≥2 and m(T)=1, [F13] gives convergence in L2 as well. The extensions therefore satisfy C(1)h=C(m)h and C(1)h=C(m+1)h as L2 classes, so T has the endpoint bounds Km,Km+1. Applying [F10] gives an extension Cp with ∥Cp∥≤Km1−θKm+1θ, where 1p=1−θ2m+θ2m+1. It agrees with C on polynomials: uniformly approximate a polynomial P by finite simple functions hj; then Cphj=C(m+1)hj→C(m+1)P=CP in L2m+1, hence in Lp by [F13], while boundedness gives Cphj→CpP in Lp.

5.1step 4.2step 4.3F4F11F12givenalgebra

Let 1<p<2 and q:=pp−1>2, so that 4.3 gives a bounded operator Cq on Lq with norm Kq. For f∈Lp the formula Λf(ψ):=−∫Tf(Cqψ) dm defines a complex-linear functional on Lq, bounded by ∥Λf∥≤Kq∥f∥p because [F4] bounds ∣Λf(ψ)∣≤∥f∥p∥Cqψ∥q≤Kq∥f∥p∥ψ∥q; by [F11] there is a unique h∈Lp with Λf(ψ)=∫Tψh dm for all ψ∈Lq and ∥h∥p=∥Λf∥≤Kq∥f∥p. Put Cpf:=h: then Cp is complex-linear and bounded with ∥Cp∥≤Kq. It extends the polynomial core, because for a trigonometric polynomial f and any ψ∈Lq the defining identity, the consistency of 4.3 and the skew-adjointness of 4.2 give ∫T(Cpf)ψ dm=−∫Tf(Cqψ) dm=−∫Tf(C(1)ψ) dm=∫T(Cf)ψ dm, so Cpf=Cf by the norm recovery [F12] applied to the difference in Lp.

6.1step 3.2step 4.1step 4.3step 5.1F1F5F13given∎

Collecting: for every 1<p<∞ the operator Cp of 4.3 (for p≥2) and of 5.1 (for 1<p<2) is a bounded complex-linear extension of C to Lp(T;C), and it is the only such extension because trigonometric polynomials are dense in Lp for p<∞ and continuous extensions of one map agree on a dense set. If 1<p≤q<∞ and f∈Lp∩Lq, choose trigonometric polynomials ϕk→f in Lq using [F5]; [F13] gives convergence in Lp, and the bounded extensions agree on polynomials, so their images converge to the same Lp limit. Thus Cpf=Cqf. Also Cp1=0 because C kills constants by [F1]. By 3.2 there is no bounded compatible L∞ extension and by 4.1 no bounded compatible L1 extension. This proves all three assertions.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

Uniform Lp bounds for periodic Fourier partial sums

Statement

Assume Countable Choice and let 1<p<∞. With the torus conventions of Period-one Fourier coefficients, partial sums, and convolution on the torus, write ∥SN∥:=∥SN∥Lp(T;C)→Lp(T;C) for the operator norm of the N-th Fourier partial sum. Then

sup⁡N≥0∥SN∥<∞.

Moreover the bound is explicit: with M:=∥Cp∥ the norm of the Lp extension of the conjugate function supplied by The Marcel Riesz conjugate-function theorem on the circle, one has sup⁡N≥0∥SN∥≤2+M.

Facts & Assumptions

Given: Countable Choice, 1<p<∞, and the torus conventions of Period-one Fourier coefficients, partial sums, and convolution on the torus: characters ek(x)=e2πikx, coefficients f^(k)=∫01f(t)e−2πikt dt, partial sums SNf=∑∣k∣≤Nf^(k)ek, trigonometric polynomials, and convolution.

[F1]

The torus carries the normalized translation-invariant Haar integral m with m(T)=1, and ∣∫Tf dm∣≤∥f∥1 for f∈L1(T). The one-dimensional torus and its normalized Haar integral Period-one Fourier coefficients, partial sums, and convolution on the torus

[F2]

The conjugate function C is defined on trigonometric polynomials coefficientwise by Cg^(k)=−isgn⁡(k)g^(k); it is complex-linear, kills constants, and the characters satisfy eael=ea+l. Conjugate function on the circle

[F3]

For every 1<p<∞ the operator C extends uniquely to a bounded complex-linear Cp on Lp(T;C) with ∥Cp∥=M<∞, and Cp1=0. The Marcel Riesz conjugate-function theorem on the circle

[F4]

For 1≤p<∞ and f∈Lp(T;C), the Cesaro means σNf=1N+1∑j=0NSjf are trigonometric polynomials and ∥σNf−f∥p→0; hence the trigonometric polynomials are dense in Lp(T;C). Fejer means converge in L^p for 1 <= p < infinity Cesaro and Abel means of a Fourier series

[F5]

On the finite measure space T one has ∥f∥1≤∥f∥p for 1≤p<∞. Finite-measure Lr includes into Lp for p<r

[F6]

Complex Lp carries the norm structure of Complex Holder, Minkowski, and the quotient norm, so the triangle inequality applies to finite sums.

[F7]

A trigonometric polynomial whose Fourier coefficients all vanish is the zero polynomial; equivalently, a finite family of distinct characters is linearly independent. The trigonometric characters are orthonormal in L2 of the torus

Proof

technique · direct
1.1F1F3F5F6

Define P+g:=12(g+iCpg)+12(∫Tg dm)1 for g∈Lp(T;C), with 1 the constant function. Then P+ is complex-linear and bounded with ∥P+g∥p≤(1+M2)∥g∥p, where M=∥Cp∥: indeed the triangle inequality of [F6] gives ∥12(g+iCpg)∥p≤12(1+M)∥g∥p, and ∥12(∫g)1∥p=12∣∫g∣≤12∥g∥1≤12∥g∥p by [F1] and [F5].

1.2F1F2

For a trigonometric polynomial p one has P+p=∑k≥0p^(k)ek. Indeed [F2] gives (p+iCp)^(k)=(1+sgn⁡k)p^(k), so 12(p+iCp) has coefficients p^(k) for k>0, 12p^(0) at k=0, and 0 for k<0; the constant function 1 has coefficients 1 at k=0 and 0 elsewhere, and ∫Tp dm=p^(0), so adding 12p^(0)1 yields exactly the coefficients p^(k) for k≥0 and 0 for k<0.

1.3F2F6

For a∈Z define the modulation Mah:=eah, a complex-linear map on Lp(T;C). Since ∣ea∣=1, one has ∣Mah∣=∣h∣ pointwise and hence ∥Mah∥p=∥h∥p: each Ma is an isometry. For a trigonometric polynomial h and every k, the coefficients satisfy Mah^(k)=h^(k−a), because eael=ea+l in [F2] gives ∫eah e−k dm=∫h e−(k−a) dm.

1.4F1F5F6

SN is bounded on Lp: for f∈Lp each ∣f^(k)∣≤∥f∥1≤∥f∥p by [F1] and [F5], so the defining finite sum gives ∥SNf∥p≤∑∣k∣≤N∥f∥p=(2N+1)∥f∥p.

2.1step 1.2step 1.3F7

For every trigonometric polynomial p and every N≥1, SNp=M−NP+MNp−MN+1P+M−(N+1)p. Indeed, by 1.2 and 1.3 the left-hand side of the identity has coefficients (M−NP+MNp)^(k)=(P+MNp)^(k+N)=1{k+N≥0}p^(k), and (MN+1P+M−(N+1)p)^(k)=1{k−(N+1)≥0}p^(k); their difference has coefficients 1{∣k∣≤N}p^(k)=SNp^(k) for every k, and two trigonometric polynomials with equal coefficients are equal by [F7].

2.2step 1.1step 1.3F6

The right-hand side of 2.1 defines a bounded operator on Lp with norm at most 2(1+M2)=2+M: by 1.3 each modulation is an isometry and by 1.1 ∥P+∥≤1+M2, so the triangle inequality of [F6] bounds the difference of the two composites by 2(1+M2).

3.1step 1.4step 2.1step 2.2F4

The identity of 2.1 holds for every f∈Lp(T;C), not only for trigonometric polynomials: both sides are bounded operators on Lp by 1.4 and 2.2, they agree on the set of trigonometric polynomials, and that set is dense in Lp by [F4]; two bounded operators agreeing on a dense set agree everywhere.

4.1step 1.4step 2.2step 3.1∎

By 3.1 and 2.2, ∥SNf∥p≤(2+M)∥f∥p for every f∈Lp and every N≥1, while for N=0 step 1.4 gives ∥S0f∥p=∣f^(0)∣≤∥f∥p≤(2+M)∥f∥p. Hence ∥SN∥≤2+M for every N≥0 and sup⁡N≥0∥SN∥≤2+M<∞, which is the assertion.

TheoremStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

Periodic Fourier partial sums converge in the strict Lp range

Statement

Assume Countable Choice, and use the torus conventions of Period-one Fourier coefficients, partial sums, and convolution on the torus and The one-dimensional torus and its normalized Haar integral: the torus T=R/Z carries normalized Haar measure m with m(T)=1, and SNf=∑∣k∣≤Nf^(k)ek for f∈Lp(T;C).

  1. For every 1<p<∞ and every f∈Lp(T;C), ∥SNf−f∥p⟶0(N→∞).
  2. At the endpoints the operator norms grow at least as the Lebesgue constants. For every N≥1, ∥SN∥L1→L1 ≥ ∫T∣DN∣ dm ≥ 13πlog⁡(N+1),∥SN∥L∞→L∞ ≥ ∫T∣DN∣ dm ≥ 13πlog⁡(N+1). Hence both families (∥SN∥L1→L1)N≥0 and (∥SN∥L∞→L∞)N≥0 are unbounded, and there exist f∈L1(T;C) and g∈L∞(T;C) such that (SNf) fails to converge in L1(T;C) and (SNg) fails to converge in L∞(T;C).

No failure of weak-type (1,1) or of any endpoint mapping weaker than norm convergence is asserted.

Facts & Assumptions

[F1]

The torus integral is normalized, m(T)=1, and translation invariant: ∫Th(x−t) dm(x)=∫Th(u) dm(u) for integrable h. The character ek(x)=e2πikx, the coefficient f^(k)=∫01f(t)e−2πiktdt, the partial sum SNf=∑∣k∣≤Nf^(k)ek, trigonometric polynomials and torus convolution (f∗g)(x)=∫01f(x−t)g(t)dt are defined as in the cited definition, as is ∥f∥p for the normalized measure. Period-one Fourier coefficients, partial sums, and convolution on the torus The one-dimensional torus and its normalized Haar integral

[F2]

For every one-period integrable f, every N≥0 and every x, SNf(x)=(f∗DN)(x)=∫01f(x−t)DN(t)dt. Fourier partial sums are Dirichlet convolutions

[F3]

DN=1+2∑k=1Ncos⁡(2πkt) is real, even, continuous and bounded, ∫TDN dm=1, and ∥DN∥1<∞. The Fejer kernel FM=1M+1∑j=0MDj satisfies FM≥0 and ∫TFM dm=1, so ∥FM∥1=1. Dirichlet and Fejer kernels The Fejer kernel is a positive approximate identity

[F4]

For g∈Lp(T;C) and 1≤p<∞ the Cesaro means satisfy σMg=g∗FM, are trigonometric polynomials, and ∥σMg−g∥p→0; hence trigonometric polynomials are dense in Lp(T;C). For a trigonometric polynomial P one has SNP=P whenever N≥deg⁡P. Fejer means converge in L^p for 1 <= p < infinity Cesaro and Abel means of a Fourier series

[F5]

For every 1<p<∞ one has Cp:=sup⁡N≥0∥SN∥Lp→Lp<∞. Uniform Lp bounds for periodic Fourier partial sums

[F6]

For every N≥1, ∥SN:C(T)→C(T)∥=∫T∣DN∣ dm≥13πlog⁡(N+1). Fourier partial-sum operator norm equals the Lebesgue constant

[F7]

For integrable complex h one has ∣∫h dm∣≤∫∣h∣ dm and ∥f+g∥p≤∥f∥p+∥g∥p; Tonelli's theorem applies to nonnegative measurable functions on the finite product T×T. The modulus of an integral is bounded by the integral of the modulus Complex Holder, Minkowski, and the quotient norm Tonelli's theorem for nonnegative measurable functions on a sigma-finite product

[F8]

For every 1≤p≤∞ the space Lp(T;C) is complete. Complex Lp completeness and almost-everywhere subsequences

[F9]

(Sequential uniform boundedness.) If X is a Banach space, Y a normed space over the same field, and Tk:X→Y, k∈N, are bounded linear maps with ∥Tkx∥≤Mx for every k and every x∈X, then sup⁡k∥Tk∥<∞. Sequential uniform boundedness under countable choice

Proof

technique · direct
1.1F1F2F3F7givenalgebra

For f∈L∞(T;C), the convolution formula [F2] and the integral triangle inequality [F7] give ∣SNf(x)∣≤∫T∣f(x−t)∣ ∣DN(t)∣ dm(t)≤∥f∥∞∥DN∥1 for every x, so SN is bounded on L∞ with ∥SNf∥∞≤∥f∥∞∥DN∥1. For f∈L1(T;C), Tonelli and the translation invariance of [F1] give ∥SNf∥1≤∫T∫T∣f(x−t)∣ ∣DN(t)∣ dm(t)dm(x)=∥f∥1∥DN∥1, so SN is bounded on L1. Both bounds are finite by [F3].

1.2F1F2F3F4algebra

For all M,N≥0 one has SN(FM)=FM∗DN=DN∗FM=σM(DN): the first equality is [F2], the second is the substitution t↦x−t in the absolutely convergent torus convolution, and the third is [F4]. Since ∥FM∥1=1 by [F3], the identity and the Fejer convergence of [F4] give ∥SN(FM)−DN∥1=∥σM(DN)−DN∥1→0 as M→∞.

1.3F2F6algebra

For N≥1, the unit ball of C(T) is contained in the unit ball of L∞(T;C), and for continuous f the partial sum SNf is a trigonometric polynomial, whose essential supremum equals its supremum; hence ∥SN∥L∞→L∞≥∥SN:C(T)→C(T)∥=∫T∣DN∣ dm≥13πlog⁡(N+1) by [F6].

1.4F4F5F7algebra

Let 1<p<∞, f∈Lp(T;C) and ε>0. By [F4] choose a trigonometric polynomial P with ∥f−P∥p<ε/(Cp+2), where Cp is the finite bound of [F5]. For N≥deg⁡P one has SNP=P by [F4], so the triangle inequality [F7] and the bound [F5] give ∥SNf−f∥p≤∥SN(f−P)∥p+∥P−f∥p≤(Cp+1)∥f−P∥p<ε. Hence SNf→f in Lp for every 1<p<∞.

2.1step 1.2F3algebra

For fixed N≥0 and every M≥0, step 1.2 and [F3] give ∥SN(FM)∥1≤∥SN∥1→1∥FM∥1=∥SN∥1→1, while ∥SN(FM)∥1→∥DN∥1; therefore ∥SN∥L1→L1≥∥DN∥1=∫T∣DN∣ dm.

3.1step 1.3step 2.1algebra

By step 1.3 and step 2.1, for every N≥1 both endpoint norms satisfy max⁡(∥SN∥L1→L1,∥SN∥L∞→L∞)≥∫T∣DN∣ dm≥13πlog⁡(N+1), and 13πlog⁡(N+1)→∞. Hence sup⁡N∥SN∥L1→L1=sup⁡N∥SN∥L∞→L∞=∞, which is the norm-growth assertion of part 2.

4.1step 1.1step 3.1F8F9given

Suppose no f∈L1(T;C) failed to converge. Then each f would have sup⁡N∥SNf∥1<∞, and since L1(T;C) is Banach by [F8] and each SN is a bounded linear operator on it by step 1.1, the sequential uniform boundedness principle [F9] would give sup⁡N∥SN∥L1→L1<∞, contradicting step 3.1. Hence there is f∈L1(T;C) with sup⁡N∥SNf∥1=∞; if (SNf) converged to some g in L1, then ∥SNf∥1≤∥g∥1+1 for all large N, a contradiction. So (SNf) does not converge in L1.

4.2step 1.1step 3.1F8F9given

The same argument with L∞(T;C) in place of L1(T;C): each SN is bounded on L∞ by step 1.1, this space is Banach by [F8], and sup⁡N∥SN∥L∞→L∞=∞ by step 3.1, so [F9] supplies g∈L∞(T;C) with sup⁡N∥SNg∥∞=∞, and (SNg) does not converge in L∞.

5.1step 1.4step 4.1step 4.2step 1.3step 2.1∎

Step 1.4 proves part 1, and steps 4.1 and 4.2 together with the norm lower bounds of steps 1.3 and 2.1 prove part 2.

DefinitionDefinition: Literature-sourcedProof: Not applicablejudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

Truncated Hilbert transform and principal value

Definition

Fix 1≤p<∞ and a function f∈Lp(R), with the Lp conventions of Complex Lp classes and Euclidean test-function conventions. For ε>0 and x∈R define the truncated Hilbert transform

Hεf(x):=1π∫∣x−y∣>εf(y)x−y dy=1π∫∣t∣>εf(x−t)t dt.

Each truncation is an ordinary Lebesgue integral over the complement of an interval of length 2ε around x, and it is absolutely convergent. For p=1 this follows from the pointwise bound ∣1/(x−y)∣<ε−1 on the domain of integration; for 1<p<∞ it follows from Hölder's inequality applied to the two half-lines x−y>ε and x−y<−ε, where ∣x−y∣−1 has finite Lq norm, with q conjugate to p (Complex Holder, Minkowski, and the quotient norm). Changing f on a null set changes no integral, so Hεf(x) is a well-defined number attached to the class of f; and Hεf is itself a measurable function of x.

The Hilbert transform in the principal-value sense is defined only where the truncations converge:

Hpvf(x):=lim⁡ε↓0Hεf(x),

whenever this limit exists in C. No almost-everywhere existence of this limit, and no bound of Hpvf in any Lp norm, is asserted by this definition. The definition also does not extend Hε to L∞: for the tail ∣x−y∣>ε the bound ∥f∥∞/ε is finite but the integral over an unbounded domain is not controlled, so the truncation of a merely bounded f need not converge absolutely at any x.

Three distinctions are recorded here for later use. First, Hεf is an integral of a truncated singular kernel, while the pairing of a test function with the principal-value distribution of 1/(πx) is a separate object; the two agree only under the convergence just defined. Second, the limit is taken symmetrically in ε about the singularity y=x, and unsymmetric truncations are a different object. Third, Hpvf is a pointwise partial function, whereas the L2(R) extension constructed later on this page is a single bounded operator agreeing with Hpv where the latter exists on a dense class.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

The sine integral under Countable Choice: uniform bounds and the value pi/2

Statement

Assume Countable Choice. Put f(u):=sin⁡(u)/u for u>0 and f(0):=1, and write S(T):=∫0Tf(u) du for T≥0. Then:

  1. S(T)→π/2 as T→∞; that is, the improper integral ∫0∞sin⁡(u)/u du converges and equals π/2.
  2. The partial integrals are uniformly bounded: ∣S(T)∣≤3 for every T≥0, moreover ∣S(T)∣≤T for 0≤T≤1, and more precisely ∣∫ABsin⁡(u)/u du∣≤2/A for all 1≤A<B.
  3. For every real z and every T≥0, reading the integrand at t=0 as z,

∫−TTsin⁡(tz)t dt=2sgn⁡(z) S(T∣z∣),

so that lim⁡T→∞∫−TTsin⁡(tz)/t dt=πsgn⁡(z) and ∣∫−TTsin⁡(tz)/t dt∣≤6 for every T≥0 and every real z.

The argument uses Countable Choice only; it does not invoke the published full-AC sine-integral lemma of the same name on the Dirichlet-kernel page.

Facts & Assumptions

Given: Countable Choice (The Axiom of Countable Choice (ACω)) and the functions f and S of the statement.

[F3]

Under Countable Choice a bounded Riemann integrable function on a compact interval is Lebesgue measurable, and its Riemann and Lebesgue integrals agree. A bounded Riemann integrable function on a closed bounded interval is Lebesgue measurable and has the same integral

[F4]

Fubini's theorem for L^1 functions on a sigma-finite product. Fubini's theorem for L^1 functions on a sigma-finite product

[F5]

Tonelli's theorem for nonnegative product-measurable functions on a sigma-finite product. Tonelli's theorem for nonnegative measurable functions on a sigma-finite product

[F6]

Dominated convergence. Dominated convergence

[F7]

Sine and cosine have derivatives cosine and minus sine, and sin⁡0=0, cos⁡0=1. The derivatives of sine and cosine are cosine and minus sine

[F8]
[F10]

1+x≤ex for every real x, hence e−x≤1/(1+x) for x≥0 and e−x→0 as x→∞. 1+x≤exp⁡(x) for every real x, hence (1−p)m≤exp⁡(−mp)

[F11]

A continuous function on a compact interval is Riemann integrable. A continuous function on [a,b] is Riemann integrable, by Heine-Cantor and Riemann's criterion

[F12]

Sine and cosine are 1-Lipschitz: ∣sin⁡u−sin⁡v∣≤∣u−v∣ and ∣cos⁡u−cos⁡v∣≤∣u−v∣. Sine and cosine are 1-Lipschitz on R

[F13]

Parity and the Pythagorean identity: sin⁡(−x)=−sin⁡x, cos⁡(−x)=cos⁡x, sin⁡2x+cos⁡2x=1, hence ∣sin⁡x∣≤1 and ∣cos⁡x∣≤1. Parity and the Pythagorean identity for sine and cosine

[F14]

Continuous maps on Euclidean spaces are Borel measurable, so the product integrands below are measurable. Continuous functions on Euclidean spaces are Borel measurable

[F15]

Change of variable for improper integrals: for a monotone differentiable surjection satisfying the proper hypotheses on compact truncations, the two improper integrals converge simultaneously and are equal, with orientation retained for decreasing parametrizations. Change of variable in an improper integral

[F16]

Principal arctangent: arctan⁡′=1/(1+x2) and arctan⁡x=∫0xdt/(1+t2). Principal arctangent: derivative, integral, power series, and the Gregory–Leibniz series

[F17]

Principal arctangent is a continuous strictly increasing bijection from R onto (−π/2,π/2), and arctan⁡(tan⁡x)=x on the principal interval. The principal inverse tangent arctan⁡:R→(−π/2,π/2)

[F18]

A convergent nonnegative improper Riemann integral on a half-line agrees with the Lebesgue integral of its integrand, under Countable Choice. A nonnegative improper Riemann integral on a half-line agrees with the Lebesgue integral

Proof

technique · direct
1.1F7F12F13

By [F7], sin⁡0=0 and sin⁡′(0)=cos⁡0=1, so sin⁡(u)/u→1=f(0) as u→0; and [F12] with v=0 gives ∣sin⁡u∣≤u for u≥0, while [F13] gives ∣cos⁡u∣≤1. Thus f is bounded by 1 on [0,∞) and continuous at 0 from the right.

1.2F2F8F9F10F18

For u≥0 and ε>0, [F8] and [F9] give (e−εu)′=−εe−εu, and [F10] bounds e−εu≤1/(1+εu), so e−εu→0 as u→∞; [F2] applied to u↦−e−εu/ε therefore gives ∫0Re−εu du=(1−e−εR)/ε for every R>0, and this tends to 1/ε as R→∞ when ε>0. By [F18] the nonnegative continuous function u↦e−εu is Lebesgue integrable on [0,∞) with ∫[0,∞)e−εu dλ(u)=1/ε.

1.3F2F7F8F9F10F12F13

For t>0, [F2] applied to s↦sin⁡(st)/t, whose derivative is cos⁡(st) by [F7] and [F9], gives ∫01cos⁡(st) ds=sin⁡(t)/t, and at t=0 both sides equal 1. For ε>0 and s∈[0,1] put H(t):=e−εt(−εcos⁡(st)+ssin⁡(st))/(ε2+s2); [F8], [F9] and [F7] give H′(t)=e−εtcos⁡(st), while [F10], [F12] and [F13] give H(0)=−ε/(ε2+s2) and H(t)→0 as t→∞.

2.1step 1.1F11

By 1.1 the quotient u↦sin⁡(u)/u is continuous on (0,∞) and extends continuously to u=0 with value 1, and it is bounded by 1 there; by [F11] it is Riemann integrable on every compact interval [0,T], T>0.

2.2step 1.1F1F2F7F8F9

Let 1≤A<B and 0≤ε≤1, and put w(u):=e−εu/u on [A,B]. By [F8] and [F9], w is continuously differentiable with w′(u)=−e−εu(εu+1)/u2≤0, so w is nonincreasing, w′>0 holds nowhere, and [F2] gives ∫AB∣w′∣=w(A)−w(B). Since (cos⁡u)′=−sin⁡u by [F7], [F1] applies with factors w and −cos⁡ and, using ∣cos⁡∣≤1 from 1.1, yields ∣∫ABw(u)sin⁡u du∣≤w(A)+w(B)+∫AB∣w′∣=2w(A)≤2/A; at ε=0 this is ∣∫ABsin⁡(u)/u du∣≤2/A.

3.1givenstep 2.1step 2.2F3F11

Under the given Countable Choice, [F3] applies on every compact interval: for the continuous integrands f, e−εtf and e−εtcos⁡(st) of steps 2.1 and 2.2, the proper Riemann integral on [0,T] or [A,B] equals the corresponding Lebesgue integral.

3.2step 2.1step 2.2F19

By 2.1 and [F19], for T≥1 one has S(T)=∫01f+∫1Tsin⁡(u)/u du with ∣∫01f∣≤1 and ∣∫1Tsin⁡(u)/u du∣≤2 by 2.2; for 0≤T≤1 the bound ∣S(T)∣≤T follows from ∣f∣≤1 in 2.1, and T≤1 gives ∣S(T)∣≤1. Hence ∣S(T)∣≤3 for every T≥0 and ∣S(T)∣≤T on [0,1].

4.1step 1.2step 1.3step 3.1F6

Fix s∈[0,1] and ε>0. By [F6] applied on [0,∞) to the functions t↦e−εtcos⁡(st)χ[0,T](t) as T→∞, which converge pointwise to t↦e−εtcos⁡(st) and are dominated by the integrable function e−εt of 1.2, and by 3.1 and 1.3, ∫[0,∞)e−εtcos⁡(st) dλ(t)=lim⁡T→∞H(T)−H(0)=ε/(ε2+s2).

4.2step 1.2step 2.2step 3.1F6

Let A≥1 and 0<ε≤1. The functions t↦e−εtf(t)χ[A,B](t) converge pointwise as B→∞ to t↦e−εtf(t)χ[A,∞)(t) and are dominated by the integrable function e−εtχ[A,∞), so [F6] with 2.2 and 3.1 gives ∣∫[A,∞)e−εtf(t) dλ(t)∣=lim⁡B→∞∣∫ABe−εtf(t) dt∣≤2/A. For ε=0, the bound in 2.2 makes ∫ABf(t) dt Cauchy as B→∞ and bounds the resulting improper tail by 2/A.

4.3step 2.1step 3.1F6

For fixed A>0, e−εtf(t)→f(t) as ε↓0 for every t∈[0,A], with ∣e−εtf(t)∣≤1 and [0,A] of finite measure, so [F6] and 3.1 give ∫[0,A]e−εtf(t) dλ(t)→∫[0,A]f(t) dλ(t)=S(A) as ε↓0.

5.1step 1.2step 1.3step 3.1step 4.1F4F5F14F15F16

Fix ε>0 and put Φ(s,t):=e−εtcos⁡(st) on [0,1]×[0,∞). By [F14] the integrand Φ is product measurable, and [F5] with 1.2 gives ∫[0,1]×[0,∞)∣Φ∣ d(λ⊗λ)≤∫[0,1](∫[0,∞)e−εt dλ(t))ds=1/ε<∞, so Φ∈L1 of the product and [F4] may be applied. By 1.3, 3.1 and 4.1, the outer s-integration of [F4] turns the t-inner integral into ε/(ε2+s2), while the outer t-integration turns the s-inner integral into e−εtsin⁡(t)/t; hence Jε:=∫[0,∞)e−εtf(t) dλ(t) satisfies Jε=∫01ε/(ε2+s2) ds, and [F15] with the substitution s=εv followed by [F16] gives Jε=∫01/εdv/(1+v2)=arctan⁡(1/ε).

6.1step 4.2step 4.3step 5.1F17

Let A≥1 and 0<ε≤1. By 4.2 and 4.3, ∣S(A)−Jε∣≤∣S(A)−∫[0,A]e−εtf dλ∣+2/A, so letting ε↓0 and using 5.1 gives ∣S(A)−π/2∣≤2/A: indeed arctan⁡(1/ε)→π/2 since [F17] makes arctan⁡ strictly increasing onto (−π/2,π/2), whence for every v<π/2 one has arctan⁡y>v for all y>tan⁡v, while arctan⁡(1/ε)<π/2 always. Letting A→∞ yields S(T)→π/2, so the improper integral ∫0∞sin⁡(u)/u du converges to π/2.

7.1step 3.2step 6.1F7F13F15F19∎

If z=0, the integrand with its assigned value at t=0 is identically zero, and the identity, bound and limit follow directly, with sgn⁡(0)=0. If T=0, both finite integrals vanish. For z≠0 and T>0, put g(t):=sin⁡(tz)/t for t≠0, g(0):=z; by [F7] and [F13], g is continuous and even, so [F15] with the substitution t↦−t on [0,T] and [F19] give ∫−TTg=2∫0Tg. By [F15] with the substitution u=∣z∣t (orientation retained, and sin⁡(−u)/(−u)=sin⁡(u)/u by [F13]) and [F7], ∫0Tg=sgn⁡(z)∫0T∣z∣sin⁡(u)/u du=sgn⁡(z)S(T∣z∣), so ∫−TTg=2sgn⁡(z)S(T∣z∣); step 3.2 bounds this by 6, and step 6.1 gives the limit πsgn⁡(z).

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

The Hilbert transform is the tempered convolution with pv(1/(pi x)) and has signum Fourier multiplier

Statement

Assume Countable Choice and use the e−2πixξ Fourier convention of Fourier transform of a tempered distribution. Define the tempered distribution W=pv⁡1πx by its pairing with a Schwartz test function, ⟨W,φ⟩ equal to

1π∫∣x∣>1φ(x)x dx+1π∫∣x∣<1φ(x)−φ(0)x dx.

Then, for every Schwartz function f:

  1. the principal value lim⁡ε↓0Hεf(x) of Truncated Hilbert transform and principal value exists at every x∈R, and equals (W∗f)(x) for the tempered convolution of Convolution of a tempered distribution with a schwartz function;
  2. hence Hf:=W∗f is a tempered distribution, and F(Hf)=−isgn⁡(ξ)f^(ξ), where sgn⁡(0)=0 and f^=Ff is the Schwartz transform of f.

The principal value is taken symmetrically about the singularity, and the statement is made for Schwartz functions only; no Lp mapping property and no almost-everywhere statement for general f is asserted.

Facts & Assumptions

Given: Countable Choice, the Schwartz space S(R) and its seminorms pαβ(φ)=sup⁡x∣xα∂βφ(x)∣, and the Fourier convention φ^(ξ)=∫φ(x)e−2πixξdx.

[F1]

The truncated Hilbert transform is Hεf(x)=1π∫∣t∣>εf(x−t)t dt, and the principal-value transform is its symmetric ε↓0 limit wherever it exists; the definition asserts no almost-everywhere existence by itself. Truncated Hilbert transform and principal value

[F2]

The sine integral satisfies ∫0∞sin⁡uudu=π2, its partial integrals obey ∣S(T)∣≤3 for all T≥0 and ∣S(T)∣≤T for 0≤T≤1, and ∫ABsin⁡uudu≤2A in absolute value for 1≤A<B. The sine integral under Countable Choice: uniform bounds and the value pi/2

[F3]

The Fourier transform of a tempered distribution is defined by ⟨Fu,φ⟩=⟨u,Fφ⟩; the pairing is bilinear with no conjugation. Fourier transform of a tempered distribution

[F4]

For u∈S′(Rn) and Schwartz φ one has F(u∗φ)=(Fu)(Fφ), the product being the product of a tempered distribution with a smooth polynomially bounded function. Fourier transform converts allowed tempered convolutions to products

[F5]

The tempered convolution is defined by (u∗φ)(x)=⟨uy,φ(x−y)⟩, a scalar function of x. Convolution of a tempered distribution with a schwartz function

[F6]

A tempered distribution is a continuous complex-linear functional on Schwartz space. Tempered distribution

[F7]

Schwartz seminorms pαβ(φ)=sup⁡x∣xα∂βφ(x)∣ are finite for φ∈S(R). Schwartz space and its seminorms

[F8]

If f∈S(Rn) then f∈Lp for every 1≤p≤∞, with norm bounded by a finite sum of Schwartz seminorms. Schwartz derivatives are integrable

[F9]

The Fourier transform is a topological automorphism of Schwartz space, so f^∈S for f∈S. Fourier transform is a topological automorphism of Schwartz space

[F10]

Mean value theorem: for differentiable φ, ∣φ(x)−φ(0)∣≤∥φ′∥∞∣x∣ on [−1,1]. The mean value theorem, as the case g(x)=x of Cauchy's: for f continuous on [a,b] with a<b and differentiable on (a,b) there is c∈(a,b) with f(b)−f(a)=f′(c)(b−a)

[F11]

Fubini for L^1 functions on a sigma-finite product. Fubini's theorem for L^1 functions on a sigma-finite product

[F12]

Dominated convergence. Dominated convergence

[F13]

Substitution for improper integrals, with orientation retained for decreasing parametrizations. Change of variable in an improper integral

Proof

technique · direct
1.1F6F7F10

For φ∈S(R) both integrals in the definition of W converge absolutely: on ∣x∣>1 the bound ∣φ(x)/x∣≤p20(φ)∣x∣−3 is integrable, and on ∣x∣<1 the bound ∣φ(x)−φ(0)∣/∣x∣≤p01(φ) from [F10] is integrable on a set of length two. Hence ∣⟨W,φ⟩∣≤2π(p20(φ)+p01(φ)) and W is a tempered distribution by [F6]. Moreover ∫ε<∣x∣<1φ(0)/x dx=0 by oddness of 1/x, so for 0<ε<1 the truncated pairing (1/π)∫∣x∣>εφ(x)/x dx equals the defining two-piece pairing with the local piece integrated over ε<∣x∣<1; consequently ⟨W,φ⟩=lim⁡ε↓01π∫∣x∣>εφ(x)/x dx.

2.1step 1.1F2F8F11F13

Fix φ∈S(R) and 0<ε<R. By [F11] applied on the product of the finite-measure annulus {ε<∣x∣<R} with R, using the integrable majorant ∣φ(ξ)∣/∣x∣ from [F8], Iε,R:=1π∫ε<∣x∣<Rφ^(x)x dx=1π∫Rφ(ξ)Λε,R(ξ) dξ with Λε,R(ξ):=∫ε<∣x∣<Re−2πixξx−1dx. Writing the exponential in cosine and sine, the cosine term is odd and integrates to zero, while [F13] with u=2πxξ gives Λε,R(ξ)=−2i∫εRsin⁡(2πxξ)xdx=−2isgn⁡(ξ)(S(2πR∣ξ∣)−S(2πε∣ξ∣)) for the partial sine integral S of [F2]; in particular ∣Λε,R(ξ)∣≤12.

2.2step 1.1F1F5F8F10F12

Fix f∈S(R) and x∈R. For 0<ε<1, [F1] gives Hεf(x)=1π∫∣t∣>εf(x−t)t dt; since ∫ε<∣t∣<1f(x)/t dt=0, this equals 1π∫ε<∣t∣<1f(x−t)−f(x)t dt+1π∫∣t∣>1f(x−t)t dt. The tail is absolutely convergent by [F8], and the first integral converges as ε↓0 by [F12], the integrand tending pointwise to f(x−t)−f(x)t and being dominated on (−1,1) by p01(f)=∥f′∥∞ thanks to [F10]. The resulting limit is exactly ⟨Wy,f(x−y)⟩=(W∗f)(x) by the defining formula of W in 1.1 and the convolution definition [F5]. Hence lim⁡ε↓0Hεf(x) exists at every x and equals (W∗f)(x).

3.1step 1.1step 2.1F2F3F12

By 1.1 and [F3], ⟨FW,φ⟩=⟨W,φ^⟩=lim⁡ε↓01π∫∣x∣>εφ^(x)/x dx. Holding ε fixed, [F2] gives Λε,R(ξ)→ρε(ξ):=−2isgn⁡(ξ)(π2−S(2πε∣ξ∣)) as R→∞, with ∣ρε(ξ)∣≤π+6. Thus [F12] against ∣φ∣ yields Iε,R→1π∫φρε as R→∞. Then S(2πε∣ξ∣)≤2πε∣ξ∣ for 2πε∣ξ∣≤1 by [F2] makes ρε(ξ)→−iπsgn⁡(ξ) pointwise as ε↓0, and a second application of [F12] gives 1π∫φρε→∫(−isgn⁡ξ)φ(ξ) dξ. Combining the two limits with the pairing identity gives ⟨FW,φ⟩=∫(−isgn⁡ξ)φ(ξ) dξ for every φ∈S, that is, FW=−isgn⁡(ξ) as tempered distributions.

4.1step 2.2step 3.1F3F4F9∎

By 3.1 and [F4] applied to the tempered distribution W and the Schwartz function f, F(W∗f)=(FW)(Ff)=−isgn⁡(ξ)f^(ξ), the product being that of the distribution −isgn⁡ with the Schwartz function f^∈S supplied by [F9] and [F3]. By 2.2 the same Hf=W∗f is the pointwise principal-value transform of f; thus the principal value defines the tempered convolution with pv⁡1πx and has the signum Fourier multiplier, as claimed.

CorollaryStatement: Literature-sourcedProof: AI-adaptedprecheck passaudited 2026-10-02Open item page →

The Hilbert transform is an L2 isometry and squares to minus the identity

Statement

Assume Countable Choice, use the e−2πixξ convention, and let m(ξ)=−isgn⁡(ξ) with sgn⁡(0)=0. The Hilbert transform of The Hilbert transform is the tempered convolution with pv(1/(pi x)) and has signum Fourier multiplier defines, on Schwartz functions, the operator Hf=(W∗f) with F(Hf)=mf^. Then H extends uniquely to a bounded operator on L2(R;C), still denoted H, and for every f∈L2(R;C),

∥Hf∥2=∥f∥2,H2f=−f.

In particular the single point ξ=0, where m vanishes, is a Lebesgue-null set and creates no zero-mode exception. Nonzero constant functions are not in L2(R), so there is no constant mode in the domain to transform.

Facts & Assumptions

Given: Countable Choice and the multiplier m(ξ)=−isgn⁡(ξ) of the Schwartz Hilbert transform.

[F1]

For Schwartz f the principal-value Hilbert transform Hf=W∗f satisfies F(Hf)=−isgn⁡(ξ)f^(ξ) as tempered distributions. The Hilbert transform is the tempered convolution with pv(1/(pi x)) and has signum Fourier multiplier

[F2]

A measurable multiplier m with finite essential supremum defines the bounded operator Tm=F2−1MmF2 on L2; its Schwartz-core action extends uniquely to L2, it depends only on the almost-everywhere class of m, and ∥Tm∥=∥m∥∞. Exact L2 Fourier multiplier norm

[F3]

Plancherel: F2 is a surjective linear isometry of L2, so ∥F2g∥2=∥g∥2 and F2−1(−g)=−F2−1g and F2−1(−F2g)=−g. Plancherel theorem

[F4]

Fourier transformation is injective on tempered distributions. Fourier transform is a topological automorphism of tempered distributions

Proof

technique · direct
1.1givenalgebra

The symbol satisfies ∣m(ξ)∣=1 for every ξ≠0, m(ξ)2=−1 for every ξ≠0, and the exceptional set is the Lebesgue-null singleton {0}.

2.1step 1.1F1F2F4

For Schwartz f, [F2] identifies Tmf with an L2 class whose regular tempered distribution has Fourier transform mf^. By [F1], the tempered distribution Hf has the same transform. Injectivity [F4] gives equality of these distributions, so Hf is represented by the L2 class Tmf. Thus no L2 membership of the principal value is assumed in this identification.

3.1step 1.1step 2.1F2F3

By [F2] the Schwartz-core action of Tm extends uniquely to a bounded operator on L2; by 2.1 the Schwartz action of H is that core action, so H=Tm on L2, and for g∈L2, ∥Hg∥2=∥m F2g∥2=∥F2g∥2=∥g∥2 because ∣m∣=1 almost everywhere by 1.1.

4.1step 1.1step 2.1F2F3∎

Likewise, on the Schwartz core H2f=Tm2f=F2−1(m2F2f)=F2−1(−F2f)=−f by 1.1 and [F3]; both H2 and −I are bounded on L2 and agree on the dense Schwartz core, so H2=−I on all of L2.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passjudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

The Hilbert transform is skew-adjoint on L2

Statement

Assume Countable Choice and use the first-variable-linear pairing ⟨f,g⟩=∫Rfg‾ dx on L2(R;C). Then for all f,g∈L2(R;C),

⟨Hf,g⟩=−⟨f,Hg⟩.

Equivalently H∗=−H: the Hilbert transform is skew-adjoint, and the statement is a statement about the L2 extension of the Schwartz principal-value operator, not about pointwise values.

Facts & Assumptions

Given: Countable Choice, the first-variable-linear pairing, and the L2 Hilbert transform H with symbol m(ξ)=−isgn⁡(ξ).

[F1]

The Hilbert transform is the L2 operator with multiplier m(ξ)=−isgn⁡(ξ), extending the Schwartz principal-value operator; ∥Hf∥2=∥f∥2 and H2=−I. The Hilbert transform is an L2 isometry and squares to minus the identity

[F2]

For a measurable symbol with essential supremum at most one the operator F2−1MmF2 acts on L2, and the Schwartz-core action of H extends uniquely to it. Exact L2 Fourier multiplier norm

[F3]

Plancherel: F2 is a surjective linear isometry that preserves the first-variable-linear inner product, ⟨f,g⟩=⟨F2f,F2g⟩. Plancherel theorem

Proof

technique · direct
1.1givenalgebra

The symbol satisfies m(ξ)‾=−m(ξ) for every ξ≠0: indeed −isgn⁡ξ‾=isgn⁡ξ=−(−isgn⁡ξ), and both sides vanish at ξ=0.

2.1step 1.1F1F2F3

Since F2 preserves the inner product by [F3] and H is the multiplier operator of [F1] with F2(Hf)=mF2f, one has ⟨Hf,g⟩=⟨mF2f,F2g⟩=∫mf^g^‾ dλ for all f,g∈L2, the last expression being an absolutely convergent integral because ∣m∣≤1 and f^,g^∈L2.

3.1step 1.1step 2.1F3∎

Applying 2.1 with the roles of f and g interchanged and conjugating the symbol by 1.1, ⟨f,Hg⟩=∫f^mg^‾ dλ=∫f^ m‾ g^‾ dλ=−∫mf^g^‾ dλ=−⟨Hf,g⟩, which is the asserted skew-adjointness.

DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-10-02Open item page →

Riesz transforms on Euclidean space

Definition

Assume Countable Choice, let n≥1, and let 1≤j≤n. On L2(Rn;C) define the j-th Riesz transform Rj by its Fourier multiplier

mj(ξ):={−i ξj/∣ξ∣,ξ≠0,0,ξ=0,Rjf^=mj f^,

where the Fourier transform is the unitary Plancherel extension F2 of Plancherel theorem and the multiplier acts by Rj=F2−1MmjF2. The symbol mj is measurable and ∣mj(ξ)∣≤1 for every ξ on account of ∣ξj∣≤∣ξ∣, so the published L2 multiplier theorem Exact L2 Fourier multiplier norm applies with essential supremum at most one: Rj is a well-defined bounded complex-linear operator on L2(Rn), it depends only on the almost-everywhere class of mj, and in particular the assigned value mj(0)=0 has no effect on the operator. The theorem also identifies Rj on Schwartz functions with the regular distribution of F2−1(mjF2f), and gives ∥Rj∥≤1.

The Riesz kernel attached to this definition is the function on Rn∖{0}

Kj(x):=cn xj∣x∣n+1,cn:=Γ((n+1)/2)π(n+1)/2,

with Γ the Euler integral. Since (n+1)/2>0, the published convergence theorem Euler's Gamma integral converges exactly for positive real parameters gives 0<Γ((n+1)/2)<∞, so cn is a positive finite constant and Kj is a smooth function on Rn∖{0}, odd under x↦−x and homogeneous of degree −n: Kj(tx)=t−nKj(x) for t>0.

This definition asserts only the multiplier description. It does not assert that the principal value lim⁡ε↓0∫∣y∣>εKj(y)f(x−y) dy exists for any particular f or x; that statement is proved separately for Schwartz functions, as is the identification of the limit with the L2 class Rjf. In dimension n=1 the constant collapses to c1=Γ(1)/π=1/π, by the value Γ(1)=1 of The real Gamma functional equation Γ(s+1)=sΓ(s), and K1(x)=1/(πx) is the line Hilbert kernel; the comparison of R1 with the Hilbert transform of the line is worked out on the examples page. The Fourier convention is the e−2πix⋅ξ convention of F2. Replacing its phase by e−ix⋅ξ leaves both mj and cn unchanged: the frequency rescaling ξ↦ξ/(2π) preserves ξj/∣ξ∣.

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passaudited 2026-10-02Open item page →

The Riesz transform is the principal value of its kernel, with the matching constant

Statement

Assume Countable Choice, use the e−2πixξ convention, and let Kj(x)=cnxj/∣x∣n+1 with cn=Γ((n+1)/2)/π(n+1)/2 be the Riesz kernel of Riesz transforms on Euclidean space. Then for every Schwartz function f∈S(Rn):

  1. the truncated integrals ∫∣y∣>εKj(y)f(x−y) dy converge as ε↓0 for every x∈Rn, with a limit that is continuous in x; and
  2. that continuous function is a representative of the L2 class Rjf, whose Fourier multiplier is −iξj/∣ξ∣.

Existence of the principal value is asserted only for Schwartz f, pointwise in x; no almost-everywhere convergence for general L2 or Lp inputs is claimed.

Facts & Assumptions

Given: Countable Choice, n≥1, 1≤j≤n, the Riesz kernel Kj, the symbol mj(ξ)=−iξj/∣ξ∣ for ξ≠0 with mj(0)=0, and the operator Rj=F2−1MmjF2 on L2(Rn).

[F1]

The Riesz kernel is Kj(x)=cnxj/∣x∣n+1 with 0<cn<∞, smooth and odd on Rn∖{0}, and ∣Kj(x)∣≤cn∣x∣−n; the operator Rj is the bounded L2 operator with symbol mj. Riesz transforms on Euclidean space

[F2]

For S(T):=∫0Tsin⁡uudu, one has S(T)→π2, ∣S(T)∣≤3 for all T≥0, and ∣∫ABsin⁡uudu∣≤2/A when 1≤A<B. Thus ∣∫ABsin⁡uudu∣≤6 for all 0<A<B: use ∣S(B)∣+∣S(A)∣≤6 if A<1, and the tail bound if A≥1. The sine integral under Countable Choice: uniform bounds and the value pi/2

[F3]

Polar coordinates: ∫Rnh dλn=∫0∞∫Sn−1h(rω)rn−1dσ(ω) dr for nonnegative Borel h and, by splitting real and imaginary parts into their positive and negative parts, for integrable complex Borel h, and the finite Borel measure σ is uniquely determined by this property. Polar coordinates decompose Lebesgue measure into r^{n-1} dr d sigma

[F4]

Vm(1)=πm/2/Γ(m/2+1) for m≥1, and volumes scale as Vm(ρ)=Vm(1)ρm. The closed form for the volume of the unit n-ball

[F5]

Fubini for L^1 functions on a sigma-finite product. Fubini's theorem for L^1 functions on a sigma-finite product

[F6]

Dominated convergence. Dominated convergence

[F7]

(u∗φ)(x)=⟨uy,φ(x−y)⟩ defines the tempered convolution for u∈S′ and Schwartz φ. Convolution of a tempered distribution with a schwartz function

[F8]

F(u∗φ)=(Fu)(Fφ) for u∈S′(Rn) and Schwartz φ. Fourier transform converts allowed tempered convolutions to products

[F9]

Fourier transformation is a topological automorphism of S′(Rn), hence injective. Fourier transform is a topological automorphism of tempered distributions

[F10]

⟨Fu,φ⟩=⟨u,Fφ⟩ with no conjugate on the right-hand side. Fourier transform of a tempered distribution

[F11]

For a continuous curve h:[a,b]→R2 differentiable on (a,b), the bound ∣h′(t)∣≤D implies ∣h(b)−h(a)∣≤D(b−a). Identify C with R2 when applying this inequality. The mean value inequality: if f:[a,b]→Rm is continuous and differentiable on (a,b) with ∥f′∥2≤M, then ∥f(b)−f(a)∥2≤M(b−a)

[F12]

The real one-variable chain rule applies to compositions of real scalar functions; below it is applied separately to the real and imaginary parts of each coordinate section of f. The chain rule, in one line from Carathéodory: if g is differentiable at c and f is differentiable at g(c), then f∘g is differentiable at c with (f∘g)′(c)=f′(g(c)) g′(c)

[F13]

Schwartz seminorms: for every integer N≥0 there is a finite constant CN(φ) with ∣φ(x)∣≤CN(φ)(1+∣x∣)−N. Schwartz space and its seminorms

[F15]

The Gamma function satisfies Γ(1)=1. The real Gamma functional equation Γ(s+1)=sΓ(s)

Proof

technique · direct
1.1F1F6F11F12F13

Fix f∈S(Rn) and x∈Rn. Write gx(y):=f(x−y)−f(x), and put Dk:=sup⁡z∣∂kf(z)∣<∞ for 1≤k≤n by [F13], and set M:=∑k=1nDk. Join x to x−y by the n coordinate segments with successive endpoints z(k)=x−∑ℓ=1kyℓeℓ, where z(0)=x. On the k-th segment, the real one-variable chain rule [F12] on both components gives (d/dt)f(z(k−1)−tykek)=−yk∂kf(z(k−1)−tykek) for 0<t<1; this follows from the definition of the coordinate partial derivative and is valid also when yk=0, when the curve is constant. Applying [F11] to this complex curve viewed in R2 bounds its increment by Dk∣yk∣. Telescoping gives ∣gx(y)∣≤∑kDk∣yk∣≤M∣y∣ for every y. Hence on ∣y∣<1 the bound ∣Kj(y)gx(y)∣≤cnM∣y∣1−n is integrable in n dimensions, while 1+∣y∣≤(1+∣x∣)(1+∣x−y∣) and the Schwartz bound [F13] with N=n+2 give ∣f(x−y)∣≤Cn+2(f)(1+∣x∣)n+2(1+∣y∣)−n−2. Thus on ∣y∣>1, ∣Kj(y)f(x−y)∣≤cnCn+2(f)(1+∣x∣)n+2∣y∣−n−2 is integrable. Since ∫ε<∣y∣<1Kj(y)f(x) dy=0 by oddness of Kj and symmetry of the annulus, ∫∣y∣>εKj(y)f(x−y) dy=∫ε<∣y∣<1Kj(y)gx(y) dy+∫∣y∣>1Kj(y)f(x−y) dy, and ε↓0 in the first term yields the absolutely convergent limit J(x):=∫∣y∣<1Kj(y)gx(y) dy+∫∣y∣>1Kj(y)f(x−y) dy. For xk→x, the sequence (xk) is bounded, so the tail constants (1+∣xk∣)n+2 have a common finite bound. This and the common small-ball bound cnM∣y∣1−n supply integrable dominators for [F6]; continuity of f gives pointwise convergence in both integrals, hence J(xk)→J(x).

1.2F1F3

For 0<ε<R and ξ≠0 put Λε,R(ξ):=∫ε<∣y∣<RKj(y)e−2πiy⋅ξdy. The cosine part of the integrand is odd in y, so it integrates to zero on the symmetric annulus, and Kj(y)=cnyj/∣y∣n+1 gives Λε,R(ξ)=−icn∫ε<∣y∣<Ryj∣y∣n+1sin⁡(2πy⋅ξ) dy. Polar coordinates [F3] turn this into Λε,R(ξ)=−icn∫εRdrr∫Sn−1ωjsin⁡(2πr ξ⋅ω) dσ(ω).

1.3F3F14

For ξ≠0 one has ∫Sn−1sgn⁡(ξ⋅ω)ωj dσ(ω)=Anξj∣ξ∣ with An:=∫Sn−1∣ω1∣ dσ: by [F3] the measure σ is invariant under the orthogonal map ω↦Rω, so substituting ω=RTu for an orthogonal map with R(ξ/∣ξ∣)=e1 (take R=I if v:=ξ/∣ξ∣=e1, and otherwise take R=I−2wwT/∣w∣2 with w=v−e1) and reflecting uk↦−uk for k≠1 (which preserves sgn⁡(u1)u1 and kills the other components by oddness) leaves only An(RTe1)j=Anξj/∣ξ∣.

1.4F3F4F5F14F15

An=2π(n−1)/2/Γ((n+1)/2): compute C:=∫Bn∣x1∣ dx twice. Polar coordinates [F3] give C=∫01rndr⋅An=An/(n+1); for n≥2, slicing at x1=t gives, by [F5], [F14] and [F4], C=∫−11∣t∣Vn−1(1)(1−t2)(n−1)/2dt=2Vn−1(1)/(n+1)=2π(n−1)/2/((n+1)Γ((n+1)/2)), hence An=(n+1)C=2Vn−1(1)=2π(n−1)/2/Γ((n+1)/2). For n=1 the sphere is S0={−1,1}: the defining identity of [F3], applied to functions supported in the annulus 1<∣x∣<2, shows that the measure σ is the counting measure δ−1+δ1, so A1=∫S0∣ω1∣ dσ=1+1=2, while 2π0/Γ(1)=2 by Γ(1)=1 of [F15]. Hence An=2π(n−1)/2/Γ((n+1)/2) for every n≥1, and cnAnπ/2=1 by cancellation of Γ((n+1)/2) and π(n+1)/2.

2.1step 1.2step 1.3step 1.4F2F6

In 1.2 let R=1/ε and ε↓0. For each fixed ω∈Sn−1 with ξ⋅ω≠0, the substitution u=2πr(ξ⋅ω) (with orientation, [F2]) gives ∫ε1/εsin⁡(2πr ξ⋅ω)rdr→π2sgn⁡(ξ⋅ω); when ξ⋅ω=0 the integral is zero. In all cases [F2] bounds its absolute value by 6, uniformly in ε and ω. Since the sphere has finite measure, [F6] on Sn−1 gives lim⁡ε↓0Λε,1/ε(ξ)=−icn∫Sn−1ωjπ2sgn⁡(ξ⋅ω) dσ(ω)=−icnπ2Anξj∣ξ∣=−iξj∣ξ∣ by 1.3 and the constant identity of 1.4.

3.1step 1.1step 2.1F5F6F10

Define the tempered distribution Wj by the symmetric principal-value pairing ⟨Wj,φ⟩:=lim⁡ε↓0∫ε<∣y∣<1/εKj(y)φ(y) dy for φ∈S; the two-piece bound of 1.1 shows the limit exists, is finite, and is Schwartz-continuous. By [F10], ⟨FWj,φ⟩=⟨Wj,φ^⟩=lim⁡ε↓0∫ε<∣y∣<1/εKj(y)φ^(y) dy; the double integrand is absolutely integrable since ∫ε<∣y∣<1/ε∣Kj(y)∣ dy∫Rn∣φ(ξ)∣ dξ<∞, so [F5] applies, giving ∫ε<∣y∣<1/εKj(y)φ^(y) dy=∫Rnφ(ξ)Λε,1/ε(ξ) dξ. By 2.1 the bracket converges to −iξj/∣ξ∣ pointwise off the null set {ξ=0}, and by the uniform bound of 2.1 it is dominated by a constant times ∣φ(ξ)∣; [F6] therefore yields ⟨FWj,φ⟩=∫Rnφ(ξ)(−iξj/∣ξ∣) dξ, i.e. FWj=mj as tempered distributions.

4.1step 1.1step 3.1F1F7F8F9∎

By [F8] and 3.1, F(Wj∗f)=(FWj)(Ff)=mjf^; by [F1] the L2 class Rjf has Fourier transform mjf^ as well, so the two tempered distributions agree and [F9] gives Wj∗f=Rjf. By [F7] and the definition of Wj in 3.1, the value (Wj∗f)(x)=⟨Wj,f(x− ⋅ )⟩ is exactly the limit J(x) of 1.1; the continuity in 1.1 therefore makes J a continuous representative of the L2 class Rjf, and the truncated integrals ∫∣y∣>εKj(y)f(x−y)dy converge to it at every x.

CorollaryStatement: Literature-sourcedProof: AI-adaptedprecheck passaudited 2026-10-02Open item page →

Riesz transforms are L2 contractions and square to minus the identity in sum

Statement

Assume Countable Choice and let n≥1. For the Riesz transforms R1,…,Rn of the multiplier definition,

∥Rjf∥2≤∥f∥2(j=1,…,n, f∈L2(Rn;C)),

and

∑j=1nRj2f=−f(f∈L2(Rn;C)).

Both statements are L2 statements only; no Lp bound for p≠2 is asserted, and the operators are the L2 operators of the definition, so all identities hold as classes (no pointwise statement is made).

Facts & Assumptions

Given: Countable Choice, the dimension n≥1, and the Riesz transforms Rj=F2−1MmjF2 with symbols mj(ξ)=−iξj/∣ξ∣ for ξ≠0 and mj(0)=0.

[F1]

Each Rj is defined as the bounded L2 operator with multiplier mj; the symbol is measurable with ∣mj(ξ)∣≤1 everywhere, the value at the origin is immaterial, and the definition asserts no more than the multiplier description. Riesz transforms on Euclidean space

[F2]

A measurable multiplier m with finite essential supremum defines the bounded operator Tm=F2−1MmF2 with ∥Tm∥L2→L2=∥m∥∞, and Tm depends only on the almost-everywhere class of m. Exact L2 Fourier multiplier norm

[F3]

Plancherel: F2 is a surjective complex-linear isometry of L2(Rn;C), so ∥F2g∥2=∥g∥2. Its inverse is complex-linear, hence F2−1(−g)=−F2−1g and F2−1(−F2g)=−g for every class g. Plancherel theorem

Proof

technique · direct
1.1F1F2

For every ξ≠0 the symbol values satisfy ∑j=1nmj(ξ)2=∑j=1n(−ξj2/∣ξ∣2)=−1, while mj(0)=0; the single point {0} is Lebesgue null. Hence the function s(ξ):=∑j=1nmj(ξ)2 is measurable, bounded with ∣s∣≤1, and equals the constant −1 almost everywhere.

1.2F1F2F3

By [F2] applied to the bounded measurable symbol mj of [F1], ∥Rj∥=∥mj∥∞≤1; consequently, for f∈L2 and using the isometry of [F3], ∥Rjf∥2=∥mj F2f∥2≤∥F2f∥2=∥f∥2.

2.1step 1.1F2F3

Since Rj=F2−1MmjF2, composition gives Rj2=F2−1Mmj2F2=Tmj2 in the notation of [F2], and summing the finitely many bounded operators gives ∑j=1nRj2=Ts for the almost-everywhere-−1 symbol s of 1.1.

3.1step 1.1step 2.1F2F3∎

By [F2] the operator Ts depends only on the almost-everywhere class of s, which by 1.1 is the class of the constant −1; hence Ts=T−1, and T−1f=F2−1M−1F2f=F2−1(−F2f)=−f by the linearity and isometry of [F3]. Therefore ∑j=1nRj2f=−f for every f∈L2(Rn;C).

LemmaStatement: Literature-sourcedProof: AI-adaptedprecheck passaudited 2026-10-02Open item page →

Riesz kernel size, difference and spherical-cancellation bounds

Statement

Assume Countable Choice, let n≥1 and 1≤j≤n, and let Kj(x)=cnxj/∣x∣n+1 with cn=Γ((n+1)/2)/π(n+1)/2 be the Riesz kernel of Riesz transforms on Euclidean space. Then:

  1. ∣Kj(x)∣≤cn∣x∣−n for every x≠0;
  2. with Cn:=cn 2n+1(3n+4) one has ∣Kj(x−h)−Kj(x)∣≤Cn ∣h∣ ∣x∣−(n+1) whenever x≠0 and ∣h∣≤∣x∣/2; and
  3. ∫Sn−1Kj(rω) dσ(ω)=0 for every r>0, where σ is the polar surface measure of Polar coordinates decompose Lebesgue measure into r^{n-1} dr d sigma on the unit sphere Sn−1={ω∈Rn:∣ω∣=1}.

The constant Cn is explicit and depends only on n; at n=1 it reads C1=28c1=28/π. These are the raw size, first-difference and cancellation estimates that a later singular-integral treatment consumes; no Calderón–Zygmund kernel definition is invoked here.

Facts & Assumptions

Given: Countable Choice, n≥1, 1≤j≤n, and the Riesz kernel Kj(x)=cnxj/∣x∣n+1 with 0<cn<∞.

[F1]

The Riesz kernel Kj(x)=cnxj/∣x∣n+1 has 0<cn<∞, is smooth, odd and homogeneous of degree −n on Rn∖{0}, so Kj(rω)=cnr−nωj whenever r>0 and ∣ω∣=1. Riesz transforms on Euclidean space

[F2]
[F3]

For n≥1 and x∈Rn the Euclidean norm satisfies ∥x∥∞≤∥x∥2, hence ∣xj∣≤∥x∥2=∣x∣ for every coordinate j. The finite and reverse triangle inequalities for a norm; and for n≥1 every norm N on Rn satisfies N(x)≤C∥x∥1 and is Lipschitz, hence continuous, for d2

[F4]

Mean value theorem: a real function continuous on a closed interval and differentiable on its interior has a point whose derivative equals the average rate of change. The mean value theorem, as the case g(x)=x of Cauchy's: for f continuous on [a,b] with a<b and differentiable on (a,b) there is c∈(a,b) with f(b)−f(a)=f′(c)(b−a)

[F6]

Polar coordinates: for n≥1 and every Borel h:Rn→[0,∞], ∫Rnh dλn=∫0∞∫Sn−1h(rω)rn−1 dσ(ω) dr, and σ is a finite Borel measure on Sn−1. Polar coordinates decompose Lebesgue measure into r^{n-1} dr d sigma

[F7]

Linear change of variables for Lebesgue measure, in particular λn(T[E])=∣det⁡T∣ λn(E) for invertible linear T. A linear map T of Rn sends Lebesgue measurable sets to Lebesgue measurable sets, with λn(T[E])=∣det⁡T∣ λn(E) when T is invertible and T[E] Lebesgue null when it is not

Proof

technique · direct
1.1F1F3givenalgebra

Fix x≠0. By [F1] the kernel is Kj(x)=cnxj∣x∣−(n+1), so ∣Kj(x)∣=cn∣xj∣∣x∣−(n+1)≤cn∣x∣−n by the coordinate bound ∣xj∣≤∣x∣ of [F3] and the positivity ∣x∣>0.

1.2F2F3givenalgebra

Fix x≠0 and h with ∣h∣≤∣x∣/2 and put A:=∣x−h∣. By [F2] applied to the Euclidean norm, ∣A−∣x∣∣≤∣h∣, so ∣x∣/2≤A≤3∣x∣/2; by [F3], ∣hj∣≤∣h∣ and ∣(x−h)j∣≤A≤3∣x∣/2. In particular A>0 and the kernel is defined at both arguments.

1.3F1F6F7givenalgebra

Let An:=∫Sn−1ωj dσ(ω) and H(x):=1{1<∣x∣<2} xj. The function H is Borel and ∫Rn∣H∣ dλn<∞; the map x↦−x is a linear bijection with ∣det⁡∣=1, so [F7] gives ∫RnH(−x) dλn(x)=∫RnH(x) dλn(x), while H(−x)=−H(x) gives ∫H dλn=−∫H dλn, that is, ∫RnH dλn=0. On the other hand [F6] applied to the nonnegative and the negative part of H gives ∫RnH dλn=∫12rn−1(∫Sn−1rωj dσ(ω))dr=An∫12rn dr with ∫12rn dr>0, so An=0. Hence for every r>0 the homogeneity [F1] gives ∫Sn−1Kj(rω) dσ(ω)=cnr−n∫Sn−1ωj dσ(ω)=cnr−nAn=0.

2.1step 1.2F1F4F5givenalgebra

Keep x≠0 and ∣h∣≤∣x∣/2 as in 1.2, put B:=∣x∣ and ψ(s):=s−(n+1); by [F5] with m=n+1≥2≥1 one has ψ′(s)=−(n+1)s−(n+2) on (0,∞). The interval with endpoints A and B lies in [B/2,∞) by 1.2. If A=B, then ∣ψ(A)−ψ(B)∣=0 and the following bound is immediate. If A≠B, [F4] on the interval with ordered endpoints min⁡(A,B)<max⁡(A,B) gives a point s with ψ(A)−ψ(B)=ψ′(s)(A−B) and therefore ∣ψ(A)−ψ(B)∣≤(n+1)(B/2)−(n+2)∣A−B∣≤(n+1)2n+2∣h∣B−(n+2). Insert ±(x−h)jψ(B) into the difference and expand: Kj(x−h)−Kj(x)=cn[(x−h)j(ψ(A)−ψ(B))−hjψ(B)], so by 1.2 and the preceding bound, and by ψ(B)=B−(n+1), ∣Kj(x−h)−Kj(x)∣≤cn[32(n+1)2n+2+2n+1]∣h∣B−(n+1)=Cn∣h∣∣x∣−(n+1) with Cn=cn2n+1(3n+4), since 32(n+1)2n+2=3(n+1)2n+1 and 3(n+1)+1=3n+4.

3.1step 1.1step 1.3step 2.1given∎

The three assertions are proved: ∣Kj(x)∣≤cn∣x∣−n for x≠0 is 1.1; the difference bound with the stated constant is 2.1, whose hypothesis ∣h∣≤∣x∣/2 keeps both arguments nonzero as recorded in 1.2; and the vanishing of every spherical integral ∫Sn−1Kj(rω) dσ(ω), r>0, is 1.3.

RemarkRemark: Literature-sourcedProof: Not applicablejudge pass (gpt-6.1-sol)audited 2026-10-02Open item page →

Endpoint map for Hilbert and Riesz transforms

Statement

This page proves the strict-range facts for the periodic conjugate operator and the L2 facts for the line Hilbert transform and the Euclidean Riesz transforms: The Marcel Riesz conjugate-function theorem on the circle bounds the conjugate operator on Lp(T) for 1<p<∞ and records the failure of compatible strong-type L1 and L∞ extensions; The Hilbert transform is an L2 isometry and squares to minus the identity identifies the line Hilbert transform as the L2 multiplier by −isgn⁡ with H2=−I; and Riesz transforms are L2 contractions and square to minus the identity in sum gives the Euclidean L2 contractions with ∑jRj2=−I.

Three distinct endpoints are deliberately not settled here, and the reader should not read this page as a negative statement about them. First, weak (1,1) bounds, the real-line strict-range Lp theory for the line and Riesz transforms, and the almost-everywhere convergence of truncated integrals are deferred to the later Calderón–Zygmund decomposition and singular-integrals material, which supplies the covering and Calderón–Zygmund kernel estimates this page stops short of. Second, the real Hardy space endpoint belongs to the later real Hardy space and maximal-function material. Third, the bounded mean-oscillation endpoint belongs to the later BMO material. Each of those later pages is named here by title only; no result from them is used as a premise anywhere on this page.

Two further distinctions are recorded. The periodic conjugate operator is presented through the circle's zero mode: constants lie in its kernel and the multiplier vanishes at frequency 0, whereas on the line the corresponding signum multiplier vanishes on a Lebesgue-null singleton and the L2 square identity holds with no zero-mode exception. The line Hilbert and Riesz endpoint questions are distinct from the circle's partial-sum operator norms. For the periodic conjugate operator itself, however, the Lebesgue-constant lower bound above is used to rule out compatible strong L1 and L∞ extensions.

Finally, the companion examples page constructs the interval indicator whose Hilbert transform is π−1log⁡∣x/(x−1)∣ and uses it to refute a bounded strong-type L1→L1 action and a bounded L∞→L∞ action compatible with the L2 transform. Those computations refute strong-type mapping only: they are consistent with a weak (1,1) bound and with a bounded BMO-valued endpoint, and they say nothing against the deferred results named above.

5 · Examples, counterexamples and false statements

None yet.

Sources