betaFeedback

WorksMathematics

A BMO Extension of Oskolkov's Criterion for Rectangular Trigonometric Fourier Sums

Abstract

We give a sufficient condition in terms of sectional BMO for almost everywhere Pringsheim convergence of rectangular trigonometric Fourier sums. For continuous functions, the condition includes Oskolkov's criterion, and an example shows that the inclusion is strict.

Introduction

Let T=R/Z\mathbb{T} = \mathbb{R}/\mathbb{Z}, with normalized Lebesgue measure mm, and write m2=m⊗mm_2 = m \otimes m for the product measure on T2\mathbb{T}^2. For f∈L1(T2)f \in L^1(\mathbb{T}^2), put

f^(k,ℓ)=∫T2f(x,y)e−2πi(kx+ℓy) dx dy,(1)\widehat{f}(k,\ell) = \int_{\mathbb{T}^2} f(x,y)e^{-2\pi i(kx+\ell y)}\,dx\,dy, \tag*{(1)}
SM,Nf(x,y)=∑∣k∣≤M∑∣ℓ∣≤Nf^(k,ℓ)e2πi(kx+ℓy).(2)S_{M,N}f(x,y) = \sum_{|k|\le M}\sum_{|\ell|\le N}\widehat{f}(k,\ell)e^{2\pi i(kx+\ell y)}. \tag*{(2)}

The Pringsheim limit means that min⁡(M,N)→∞\min(M,N)\to\infty, with the two indices otherwise independent. For f∈C(T2)f \in C(\mathbb{T}^2), define

ω(f,δ)C=sup⁡s2+t2≤δ2(x,y)∈T2∣f(x+s,y+t)−f(x,y)∣.\omega(f,\delta)_C = \sup_{\substack{s^2+t^2\le\delta^2\\(x,y)\in\mathbb{T}^2}} |f(x+s,y+t)-f(x,y)|.

Fefferman [3] constructed a continuous function whose rectangular Fourier sums diverge in the Pringsheim sense. Bakhbukh and Nikishin [1] obtained a divergence example with

ω(f,δ)C=O(1log⁡(1/δ))(δ↓0).(3)\omega(f,\delta)_C = O\left(\frac{1}{\log(1/\delta)}\right)\quad(\delta\downarrow0). \tag*{(3)}

Oskolkov [9 Theorem 5] proved almost-everywhere Pringsheim convergence under the condition

ω(f,δ)C=o(1log⁡(1/δ)log⁡log⁡log⁡(1/δ)).(4)\omega(f,\delta)_C = o\left(\frac{1}{\log(1/\delta)\log\log\log(1/\delta)}\right). \tag*{(4)}

Masyutin [8] later obtained criteria for LpL^p functions, 1<p<∞1<p<\infty, in terms of their LpL^p modulus of continuity.

For u∈L1(T)u \in L^1(\mathbb{T}), write

[u]BMO(T)=sup⁡I⊂T1∣I∣∫I∣u−uI∣ dx,uI=1∣I∣∫Iu dx,(5)[u]_{\mathrm{BMO}(\mathbb{T})}=\sup_{I\subset\mathbb{T}}\frac{1}{|I|}\int_I|u-u_I|\,dx,\qquad u_I=\frac{1}{|I|}\int_Iu\,dx, \tag*{(5)}

where II ranges over arcs. Define the sectional seminorm

[f]sBMO=max⁡{ess sup⁡y∈T[f(⋅,y)]BMO(T),ess sup⁡x∈T[f(x,⋅)]BMO(T)}.(6)[f]_{\mathrm{sBMO}}=\max\left\{\operatorname*{ess\,sup}_{y\in\mathbb{T}}[f(\cdot,y)]_{\mathrm{BMO}(\mathbb{T})},\operatorname*{ess\,sup}_{x\in\mathbb{T}}[f(x,\cdot)]_{\mathrm{BMO}(\mathbb{T})}\right\}. \tag*{(6)}

2020 Mathematics Subject Classification. Primary 42B05; Secondary 42A20, 42B35.

Key words and phrases. Rectangular Fourier sums, almost-everywhere convergence, sectional BMO, modulus of continuity.

We write f∈sBMO⁡(T2)f \in\operatorname{sBMO}(\mathbb{T}^2) when f∈L1(T2)f \in L^1(\mathbb{T}^2) and (1.6) is finite. For 0<δ≤1/20 < \delta\le1/2, put

Ω1(f,δ)=sup⁡∣h∣≤δess sup⁡y∈T[f(⋅+h,y)−f(⋅,y)]BMO⁡(T)(7)\Omega_1(f,\delta)=\sup_{|h|\le\delta}\operatorname*{ess\,sup}_{y\in\mathbb{T}}[f(\cdot+h,y)-f(\cdot,y)]_{\operatorname{BMO}(\mathbb{T})} \tag*{(7)}
Ω2(f,δ)=sup⁡∣h∣≤δess sup⁡x∈T[f(x,⋅+h)−f(x,⋅)]BMO⁡(T)(8)\Omega_2(f,\delta)=\sup_{|h|\le\delta}\operatorname*{ess\,sup}_{x\in\mathbb{T}}[f(x,\cdot+h)-f(x,\cdot)]_{\operatorname{BMO}(\mathbb{T})} \tag*{(8)}
Ω(f,δ)=max⁡{Ω1(f,δ),Ω2(f,δ)}.(9)\Omega(f,\delta)=\max\{\Omega_1(f,\delta),\Omega_2(f,\delta)\}. \tag*{(9)}

All asymptotic conditions below concern δ↓0\delta\downarrow0.

Theorem 1. Let f∈sBMO⁡(T2)f \in\operatorname{sBMO}(\mathbb{T}^2), nq=22qn_q=2^{2^q}, and

aq=2qΩ(f,1nq+1).a_q=2^q\Omega\left(f,\frac{1}{n_q+1}\right).

If, for every η>0\eta>0,

∑q=1∞exp⁡(−η/aq)<∞,(10)\sum_{q=1}^{\infty}\exp(-\eta/a_q)<\infty, \tag*{(10)}

where a summand with aq=0a_q=0 is interpreted as zero, then SM,Nf(x,y)→f(x,y)S_{M,N}f(x,y)\to f(x,y) for almost every (x,y)(x,y) as min⁡(M,N)→∞\min(M,N)\to\infty.

Corollary 1. Let f∈sBMO⁡(T2)f \in\operatorname{sBMO}(\mathbb{T}^2). If

Ω(f,δ)=o(1log⁡(1/δ)log⁡log⁡log⁡(1/δ)),\Omega(f,\delta)=o\left(\frac{1}{\log(1/\delta)\log\log\log(1/\delta)}\right),

then SM,Nf→fS_{M,N}f\to f almost everywhere in the Pringsheim sense.

Proof. For δq=(22q+1)−1\delta_q=(2^{2^q}+1)^{-1}, we have log⁡(1/δq)≍2q\log(1/\delta_q)\asymp2^q and log⁡log⁡log⁡(1/δq)∼log⁡q\log\log\log(1/\delta_q)\sim\log q. Consequently aq=o(1/log⁡q)a_q=o(1/\log q). Given η>0\eta>0, eventually η/aq≥2log⁡q\eta/a_q\ge2\log q whenever aq>0a_q>0; thus the summands in (1.10) are at most q−2q^{-2}. Apply Theorem 1. □\square

For f∈C(T2)f\in C(\mathbb{T}^2), the bound

Ω(f,δ)≤2ω(f,δ)C(11)\Omega(f,\delta)\le2\omega(f,\delta)_C \tag*{(11)}

shows that Corollary 1 includes Oskolkov’s criterion. The next example shows that (1.10) can hold for a continuous function that does not satisfy Oskolkov’s hypothesis.

Theorem 2. There is F∈C(T2)F\in C(\mathbb{T}^2) such that, for every η>0\eta>0,

∑q=1∞exp⁡(−η2qω(F,(nq+1)−1)C)<∞,(12)\sum_{q=1}^{\infty}\exp\left(-\frac{\eta}{2^q\omega(F,(n_q+1)^{-1})_C}\right)<\infty, \tag*{(12)}

with the usual zero-summand convention, but

rlog⁡log⁡r ω(F,2−r)C↛0(r→∞).(13)r\log\log r\,\omega(F,2^{-r})_C\not\to0\qquad(r\to\infty). \tag*{(13)}

Thus Oskolkov’s hypothesis need not hold when Theorem 1 applies.

One-dimensional estimates

Write Sn(1)S_n^{(1)} and Sn(2)S_n^{(2)} for Fourier partial sums in xx and yy, respectively. Explicitly,

Sn(1)f(x,y)=∑∣k∣≤n(∫Tf(s,y)e−2πiks ds)e2πikx,S_n^{(1)}f(x,y)=\sum_{|k|\le n}\left(\int_{\mathbb{T}}f(s,y)e^{-2\pi i k s}\,ds\right)e^{2\pi i kx},
Sn(2)f(x,y)=∑∣ℓ∣≤n(∫Tf(x,t)e−2πiℓt dt)e2πiℓy.S_n^{(2)}f(x,y)=\sum_{|\ell|\le n}\left(\int_{\mathbb{T}}f(x,t)e^{-2\pi i\ell t}\,dt\right)e^{2\pi i\ell y}.

By Fubini,

SM,N=SM(1)SN(2)=SN(2)SM(1).(14)S_{M,N}=S_M^{(1)}S_N^{(2)}=S_N^{(2)}S_M^{(1)}. \tag*{(14)}

A polynomial in one coordinate may have measurable coefficients depending on the other coordinate.

Lemma 1. There is an absolute CJ>0C_J>0 such that, for f∈sBMO⁡(T2)f\in\operatorname{sBMO}(\mathbb{T}^2), n≥1n\geq1, and ν=1,2\nu=1,2, there exists a polynomial Pn(ν)P_n^{(\nu)} of degree at most nn in coordinate ν\nu. The residual gn(ν)=f−Pn(ν)g_n^{(\nu)}=f-P_n^{(\nu)} has mean zero in the active coordinate, and

ess sup⁡y[gn(1)(⋅,y)]BMO⁡(T)≤CJΩ1(f,1n+1),(15)\operatorname*{ess\,sup}_{y}[g_n^{(1)}(\cdot,y)]_{\operatorname{BMO}(\mathbb{T})}\leq C_J\Omega_1\left(f,\frac{1}{n+1}\right), \tag*{(15)}
ess sup⁡x[gn(2)(x,⋅)]BMO⁡(T)≤CJΩ2(f,1n+1).(16)\operatorname*{ess\,sup}_{x}[g_n^{(2)}(x,\cdot)]_{\operatorname{BMO}(\mathbb{T})}\leq C_J\Omega_2\left(f,\frac{1}{n+1}\right). \tag*{(16)}

Proof. Set m=⌊n/2⌋+1m=\lfloor n/2\rfloor+1 and use

Fm(t)=1m(sin⁡(πmt)sin⁡(πt))2,Jn(t)=Fm(t)2∫TFm(s)2 ds.F_m(t)=\frac{1}{m}\left(\frac{\sin(\pi mt)}{\sin(\pi t)}\right)^2,\qquad J_n(t)=\frac{F_m(t)^2}{\int_{\mathbb{T}}F_m(s)^2\,\mathrm{d}s}.

Thus FmF_m has degree m−1m-1, and the nonnegative kernel JnJ_n has degree 2(m−1)≤n2(m-1)\leq n and integral one. The bounds Fm(t)≤Cmin⁡{m,(m∣t∣2)−1}F_m(t)\leq C\min\{m,(m|t|^2)^{-1}\} and ∫TFm2≍m\int_{\mathbb{T}}F_m^2\asymp m give

∫T∣t∣Jn(t) dt≤Cn+1.(17)\int_{\mathbb{T}}|t|J_n(t)\,\mathrm{d}t\leq\frac{C}{n+1}. \tag*{(17)}

where t∈[−1/2,1/2]t\in[-1/2,1/2] is the representative of a point of T\mathbb{T}. For ν=1\nu=1, take

Pn(1)(x,y)=∫Tf(x−t,y)Jn(t) dt.P_n^{(1)}(x,y)=\int_{\mathbb{T}}f(x-t,y)J_n(t)\,\mathrm{d}t.

This is a polynomial in xx with measurable coefficients in yy. With δ=(n+1)−1\delta=(n+1)^{-1}, Minkowski’s inequality for the BMO seminorm gives

[gn(1)(⋅,y)]BMO⁡≤∫TJn(t)[f(⋅,y)−f(⋅−t,y)]BMO⁡ dt.[g_n^{(1)}(\cdot,y)]_{\operatorname{BMO}}\leq\int_{\mathbb{T}}J_n(t)[f(\cdot,y)-f(\cdot-t,y)]_{\operatorname{BMO}}\,\mathrm{d}t.

For a fixed t∈[−1/2,1/2]t\in[-1/2,1/2], set

N(t)=max⁡{1,⌈∣t∣δ⌉},h=tN(t).N(t)=\max\left\{1,\left\lceil\frac{|t|}{\delta}\right\rceil\right\},\qquad h=\frac{t}{N(t)}.

Then ∣h∣≤δ|h|\leq\delta and N(t)≤1+∣t∣/δN(t)\leq1+|t|/\delta. Write uy=f(⋅,y)u_y=f(\cdot,y) and (τsu)(x)=u(x+s)(\tau_su)(x)=u(x+s). The identity

uy−τ−tuy=∑j=0N(t)−1τ−jh(uy−τ−huy)u_y-\tau_{-t}u_y=\sum_{j=0}^{N(t)-1}\tau_{-jh}(u_y-\tau_{-h}u_y)

is a telescoping sum. For each fixed tt, the definition of Ω1\Omega_1, together with ∣h∣≤δ|h|\leq\delta, gives [uy−τ−huy]BMO⁡≤Ω1(f,δ)[u_y-\tau_{-h}u_y]_{\operatorname{BMO}}\leq\Omega_1(f,\delta) for almost every yy. The triangle inequality and translation invariance of the BMO seminorm therefore imply

[f(⋅,y)−f(⋅−t,y)]BMO⁡≤N(t)Ω1(f,δ)≤(1+∣t∣δ)Ω1(f,δ).[f(\cdot,y)-f(\cdot-t,y)]_{\operatorname{BMO}}\leq N(t)\Omega_1(f,\delta)\leq\left(1+\frac{|t|}{\delta}\right)\Omega_1(f,\delta).

The exceptional set of yy’s may depend on tt; Fubini’s theorem shows that this estimate holds for almost every tt for almost every yy. We may therefore integrate it against Jn(t)J_n(t). Since Jn≥0J_n\geq0, ∫TJn=1\int_{\mathbb{T}}J_n=1, and (2.4) gives δ−1∫T∣t∣Jn(t) dt≤C\delta^{-1}\int_{\mathbb{T}}|t|J_n(t)\,\mathrm{d}t\leq C, we obtain

[gn(1)(⋅,y)]BMO⁡≤Ω1(f,δ)(1+1δ∫T∣t∣Jn(t) dt)≤(1+C)Ω1(f,δ)[g_n^{(1)}(\cdot,y)]_{\operatorname{BMO}}\leq\Omega_1(f,\delta)\left(1+\frac{1}{\delta}\int_{\mathbb{T}}|t|J_n(t)\,\mathrm{d}t\right)\leq(1+C)\Omega_1(f,\delta)

for almost every yy. Taking the essential supremum proves (2.2); interchanging xx and yy proves (2.3). □

For u∈L2(T)u \in L^2(\mathbb{T}), let

Hu=∑k∈Z(−isgn⁡k)u^(k)e2πikx,S~nu=Sn(Hu).\mathcal{H}u=\sum_{k\in\mathbb{Z}}(-i\operatorname{sgn} k)\widehat{u}(k)e^{2\pi ikx},\qquad\widetilde{S}_n u=S_n(\mathcal{H}u).

We use two standard one-dimensional facts. First, the Fefferman–Stein decomposition [4] says that every mean-zero g∈BMO⁡(T)g \in\operatorname{BMO}(\mathbb{T}) has a representation

g=a+Hb,∥a∥∞+∥b∥∞≤C[g]BMO⁡(T).(18)g=a+\mathcal{H}b,\qquad\|a\|_{\infty}+\|b\|_{\infty}\le C[g]_{\operatorname{BMO}(\mathbb{T})}. \tag*{(18)}

Second, Hunt’s endpoint estimate for ordinary and conjugate Fourier sums [5, 9] gives absolute A,c>0A,c>0 such that every u∈L∞(T)u\in L^\infty(\mathbb{T}) satisfies

m{x:sup⁡n≥0max⁡{∣Snu(x)∣,∣S~nu(x)∣}>t}≤Aexp⁡(−ct/∥u∥∞),t>0.(19)m\left\{x:\sup_{n\ge0}\max\{|S_nu(x)|,|\widetilde{S}_nu(x)|\}>t\right\}\le A\exp(-ct/\|u\|_\infty),\qquad t>0. \tag*{(19)}

For u=0u=0, the left-hand side is zero. The estimate for Cg=sup⁡n∣Sng∣\mathcal{C}g=\sup_n|S_ng| in [7 Remark 6.1] is an L∞→BMO⁡L^\infty\to\operatorname{BMO} statement.

Lemma 2. There are absolute A,c>0A,c>0 such that, for every mean-zero g∈BMO⁡(T)g\in\operatorname{BMO}(\mathbb{T}) and u>0u>0,

m{x:sup⁡n≥0∣g(x)−Sng(x)∣>u[g]BMO⁡(T)}≤Ae−cu.(20)m\left\{x:\sup_{n\ge0}|g(x)-S_ng(x)|>u[g]_{\operatorname{BMO}(\mathbb{T})}\right\}\le Ae^{-cu}. \tag*{(20)}

If [g]BMO⁡=0[g]_{\operatorname{BMO}}=0, the maximal remainder vanishes almost everywhere.

Proof. Let σ=[g]BMO⁡(T)\sigma=[g]_{\operatorname{BMO}(\mathbb{T})}. If σ=0\sigma=0, the mean-zero function gg vanishes almost everywhere. Otherwise choose a,ba,b in (2.5). Since SnS_n commutes with H\mathcal{H},

sup⁡n∣Sng∣≤sup⁡n∣Sna∣+sup⁡n∣S~nb∣.\sup_n|S_ng|\le\sup_n|S_na|+\sup_n|\widetilde{S}_nb|.

Apply (2.6) to aa and bb, and use ∥a∥∞+∥b∥∞≤Cσ\|a\|_\infty+\|b\|_\infty\le C\sigma. This yields

m{x:sup⁡n∣Sng(x)∣>uσ}≤A1e−c1u.m\{x:\sup_n|S_ng(x)|>u\sigma\}\le A_1e^{-c_1u}.

John–Nirenberg [6], applied to the whole circle and using gT=0g_{\mathbb{T}}=0, likewise yields m{∣g∣>uσ}≤A2e−c2um\{|g|>u\sigma\}\le A_2e^{-c_2u}. Finally sup⁡n∣g−Sng∣≤∣g∣+sup⁡n∣Sng∣\sup_n|g-S_ng|\le|g|+\sup_n|S_ng|; a union bound proves (2.7). □

Let M(1)\mathcal{M}^{(1)} and M(2)\mathcal{M}^{(2)} be the Hardy–Littlewood maximal operators along the indicated coordinates, applied to the absolute value of their arguments.

Lemma 3. For h∈L1(T2)h\in L^1(\mathbb{T}^2), m≥0m\ge0, and ν=1,2\nu=1,2,

∣Sm(ν)h∣≤Clog⁡(m+2)M(ν)halmost everywhere.(21)|S_m^{(\nu)}h|\le C\log(m+2)\mathcal{M}^{(\nu)}h\qquad\text{almost everywhere}. \tag*{(21)}

Proof. In the active coordinate the Dirichlet kernel obeys ∣Dm(t)∣≤Cmin⁡{m+1,∣t∣−1}|D_m(t)|\le C\min\{m+1,|t|^{-1}\}. Split the convolution into ∣t∣≤(m+1)−1|t|\le(m+1)^{-1} and dyadic annuli. Each piece is bounded by a fixed multiple of the one-dimensional maximal function, and there are O(log⁡(m+2))O(\log(m+2)) pieces. □

We also use, for 1<s<∞1<s<\infty,

∥M(ν)H∥Ls(T2)≤Cs∥H∥Ls(T2).(22)\|\mathcal{M}^{(\nu)}H\|_{L^s(\mathbb{T}^2)}\le C_s\|H\|_{L^s(\mathbb{T}^2)}. \tag*{(22)}

Proof of Theorem 1

Put δq=(nq+1)−1\delta_q=(n_q+1)^{-1}, and for ν=1,2\nu=1,2 set εq(ν)=CJΩν(f,δq)\varepsilon_q^{(\nu)}=C_J\Omega_\nu(f,\delta_q). Choose Pq(ν)=Pnq(ν)P_q^{(\nu)}=P_{n_q}^{(\nu)} from Lemma 1 and write gq(ν)=f−Pq(ν)g_q^{(\nu)}=f-P_q^{(\nu)}. Each active-coordinate section of gq(ν)g_q^{(\nu)} has zero mean and BMO seminorm at most εq(ν)\varepsilon_q^{(\nu)}, outside a null set of passive coordinates. Define

Bq(ν)=sup⁡nq≤n<nq+1∣f−Sn(ν)f∣.\mathcal{B}_q^{(\nu)}=\sup_{n_q\le n<n_{q+1}}\left|f-S_n^{(\nu)}f\right|.

Since Sn(ν)Pq(ν)=Pq(ν)S_n^{(\nu)}P_q^{(\nu)}=P_q^{(\nu)} for n≥nqn\ge n_q, we have

Bq(ν)=sup⁡nq≤n<nq+1∣gq(ν)−Sn(ν)gq(ν)∣.\mathcal{B}_q^{(\nu)}=\sup_{n_q\le n<n_{q+1}}\left|g_q^{(\nu)}-S_n^{(\nu)}g_q^{(\nu)}\right|.

Write ε=εq(1)\varepsilon=\varepsilon_q^{(1)} and, for a passive coordinate yy, put gy=gq(1)(⋅,y)g_y=g_q^{(1)}(\cdot,y) and σy=[gy]BMO(T)\sigma_y=[g_y]_{\mathrm{BMO}(\mathbb{T})}. For almost every yy, gyg_y has mean zero and σy≤ε\sigma_y\le\varepsilon. Also,

Bq(1)(x,y)≤sup⁡n≥0∣gy(x)−Sngy(x)∣.\mathcal{B}_q^{(1)}(x,y)\le\sup_{n\ge0}\left|g_y(x)-S_ng_y(x)\right|.

Suppose ε>0\varepsilon>0. If σy>0\sigma_y>0, then uε≥uσyu\varepsilon\ge u\sigma_y, so Lemma 2 yields

m{x:Bq(1)(x,y)>uε}≤m{x:sup⁡n≥0∣gy(x)−Sngy(x)∣>uσy}≤Ae−cu,u>0.m\{x:\mathcal{B}_q^{(1)}(x,y)>u\varepsilon\}\le m\left\{x:\sup_{n\ge0}|g_y(x)-S_ng_y(x)|>u\sigma_y\right\} \le Ae^{-cu},\qquad u>0.

If σy=0\sigma_y=0, the mean-zero section gyg_y vanishes almost everywhere and the set on the left is empty. Thus the same bound holds for almost every yy. Since m(T)=1m(\mathbb{T})=1, Fubini gives

m2{Bq(1)>uε}=∫Tm{x:Bq(1)(x,y)>uε} dy≤Ae−cu,u>0.m_2\{\mathcal{B}_q^{(1)}>u\varepsilon\}=\int_{\mathbb{T}}m\{x:\mathcal{B}_q^{(1)}(x,y)>u\varepsilon\}\,dy \le Ae^{-cu},\qquad u>0.

For ν=2\nu=2, fix xx instead and integrate over xx. We have therefore proved, for both ν=1,2\nu=1,2 whenever εq(ν)>0\varepsilon_q^{(\nu)}>0, that

m2{Bq(ν)>uεq(ν)}≤Ae−cu,u>0.(23)m_2\{\mathcal{B}_q^{(\nu)}>u\varepsilon_q^{(\nu)}\}\le Ae^{-cu},\qquad u>0. \tag*{(23)}

Finally, if εq(ν)=0\varepsilon_q^{(\nu)}=0, almost every active-coordinate section of gq(ν)g_q^{(\nu)} has BMO seminorm zero and mean zero. Such a section is zero almost everywhere, so Bq(ν)=0\mathcal{B}_q^{(\nu)}=0 almost everywhere by Fubini. We construct λq≥1\lambda_q\ge1 with

∑q≥1e−λq<∞,aqλq⟶0.(24)\sum_{q\ge1}e^{-\lambda_q}<\infty,\qquad a_q\lambda_q\longrightarrow0. \tag*{(24)}

Indeed, (1.10) implies aq→0a_q\to0. For k≥1k\ge1, choose 1≤Q1<Q2<⋯1\le Q_1<Q_2<\cdots so that, for q≥Qkq\ge Q_k, aq≤1/ka_q\le1/k and

∑q≥Qkexp⁡[−1/(kaq)]≤2−k.\sum_{q\ge Q_k}\exp[-1/(ka_q)]\le2^{-k}.

For Qk≤q<Qk+1Q_k\le q<Q_{k+1}, set λq=(kaq)−1\lambda_q=(ka_q)^{-1} when aq>0a_q>0, and λq=q\lambda_q=q otherwise. For the finitely many q<Q1q<Q_1, put λq=q\lambda_q=q. The sums over successive blocks are at most 2−k2^{-k}, while the zero cases contribute at most ∑qe−q\sum_q e^{-q}. This proves the first part of (3.2); on a nonzero block aqλq=1/ka_q\lambda_q=1/k, and in zero cases the product is zero. Thus the second part holds as well. In particular,

2εq(ν)λq≤CJaqλq⟶0.(25)2^{\varepsilon_q^{(\nu)}}\lambda_q\le C_Ja_q\lambda_q\longrightarrow0. \tag*{(25)}

Choose KK with cK≥2cK \ge2, where cc is from (23), and set

Cν=max⁡{1,sup⁡q≥1Bq(ν)Kεq(ν)λq}.C_\nu=\max\left\{1,\sup_{q\ge1}\frac{\mathcal{B}^{(\nu)}_q}{K\varepsilon^{(\nu)}_q\lambda_q}\right\}.

Declare a quotient with εq(ν)=0\varepsilon^{(\nu)}_q=0 to be zero, on a common full-measure set where the corresponding block vanishes. For t>1t>1, (23) gives

m2{Cν>t}≤A∑q≥1e−cKtλq≤Ae−t∑q≥1e−λq,m_2\{C_\nu>t\}\le A\sum_{q\ge1}e^{-cKt\lambda_q} \le Ae^{-t}\sum_{q\ge1}e^{-\lambda_q},

because λq≥1\lambda_q\ge1 and cK≥2cK\ge2. Consequently Cν∈Ls(T2)C_\nu\in L^s(\mathbb{T}^2) for every finite s≥1s\ge1, and

Bq(ν)≤Kεq(ν)λqCν(26)\mathcal{B}^{(\nu)}_q\le K\varepsilon^{(\nu)}_q\lambda_qC_\nu \tag*{(26)}

almost everywhere for every qq.

By (22), M(1)C2\mathcal{M}^{(1)}C_2 and M(2)C1\mathcal{M}^{(2)}C_1 are finite almost everywhere.

Suppose M≤NM\le N, and choose qq with nq≤N<nq+1n_q\le N<n_{q+1}. Then

f−SM,Nf=(f−SM(1)f)+SM(1)(f−SN(2)f).f-S_{M,N}f=(f-S^{(1)}_M f)+S^{(1)}_M(f-S^{(2)}_Nf).

For almost every yy, f(⋅,y)∈BMO⁡(T)f(\cdot,y)\in\operatorname{BMO}(\mathbb{T}) and has a finite mean. John–Nirenberg gives f(⋅,y)∈L2(T)f(\cdot,y)\in L^2(\mathbb{T}). Carleson’s theorem [2] and Fubini therefore imply

SM(1)f(x,y)⟶f(x,y)(27)S^{(1)}_M f(x,y)\longrightarrow f(x,y) \tag*{(27)}

for almost every (x,y)(x,y).

Using Lemma 3, (26), and M≤N<nq+1M\le N<n_{q+1}, we obtain, on the same full-measure set,

∣SM(1)(f−SN(2)f)∣≤Clog⁡(M+2)M(1)(f−SN(2)f)≤C′2qεq(2)λqM(1)C2⟶0.\left|S^{(1)}_M(f-S^{(2)}_Nf)\right|\le C\log(M+2)\mathcal{M}^{(1)}(f-S^{(2)}_Nf) \le C'2^q\varepsilon^{(2)}_q\lambda_q\mathcal{M}^{(1)}C_2\longrightarrow0.

by (25), as N→∞N\to\infty. When N<MN<M, reverse the roles of the two coordinates. Together with (27), this proves the Pringsheim limit.

Proof of Theorem 2

For k≥1k\ge1, set

Rk=22k+2,Nk=22Rk,Ak=12Rklog⁡Rk.R_k=2^{2^{k+2}},\qquad N_k=2^{2^{R_k}},\qquad A_k=\frac{1}{2^{R_k}\log R_k}.

Here Rk+1=Rk2R_{k+1}=R_k^2, Nk=nRkN_k=n_{R_k}, and

AkNkAk+1Nk+1<12,Ak+1Ak<12,Ak−1<Nk−1Ak(k≥2).(28)\frac{A_kN_k}{A_{k+1}N_{k+1}}<\frac{1}{2},\qquad \frac{A_{k+1}}{A_k}<\frac{1}{2},\qquad A_{k-1}<N_{k-1}A_k\quad(k\ge2). \tag*{(28)}

Define

g(x)=∑k=1∞Aksin⁡(2πNkx),F(x,y)=g(x)+g(y).g(x)=\sum_{k=1}^{\infty}A_k\sin(2\pi N_kx),\qquad F(x,y)=g(x)+g(y).

The series converges uniformly because ∑kAk<∞\sum_k A_k<\infty, so F∈C(T2)F\in C(\mathbb{T}^2).

Put dq=2−2q=1/nqd_q=2^{-2^q}=1/n_q. If k≥2k\ge2 and Rk−1<q≤RkR_{k-1}<q\le R_k, then dq≤Nk−1−2d_q\le N_{k-1}^{-2}. For every ∣h∣≤dq|h|\le d_q, (28) gives

∣g(x+h)−g(x)∣≤2πdq∑ℓ<kAℓNℓ+2∑ℓ≥kAℓ≤4πAk−1Nk−1+4Ak≤CAk.|g(x+h)-g(x)|\le2\pi d_q\sum_{\ell<k}A_\ell N_\ell+2\sum_{\ell\ge k}A_\ell \le4\pi\frac{A_{k-1}}{N_{k-1}}+4A_k\le CA_k.

Hence

ω(F,dq)C≤C′Ak(Rk−1<q≤Rk).(29)\omega(F,d_q)_C \le C' A_k \qquad(R_{k-1}<q\le R_k). \tag*{(29)}

Since ω(F,(nq+1)−1)C≤ω(F,dq)C\omega(F,(n_q+1)^{-1})_C\le\omega(F,d_q)_C, the denominator bq=2qω(F,(nq+1)−1)Cb_q=2^q\omega(F,(n_q+1)^{-1})_C in (12) satisfies

bq≤C′2qAk=C′2q−Rklog⁡Rk.b_q\le C'2^q A_k=\frac{C'2^{q-R_k}}{\log R_k}.

Thus, for every η>0\eta>0, the sum over one block is

∑Rk−1<q≤Rke−η/bq≤∑j=0∞exp⁡[−ηC′2jlog⁡Rk]≤CηRk−η/C′.\sum_{R_{k-1}<q\le R_k}e^{-\eta/b_q} \le\sum_{j=0}^{\infty}\exp\left[-\frac{\eta}{C'}2^j\log R_k\right] \le C_\eta R_k^{-\eta/C'}.

The last expression is summable in kk, since RkR_k grows doubly exponentially. The finitely many initial terms cause no difficulty. This proves (12).

Now set hk=(4Nk)−1<dRkh_k=(4N_k)^{-1}<d_{R_k}. For ℓ>k\ell>k, Nℓ/NkN_\ell/N_k is divisible by 4, so sin⁡(2πNℓhk)=0\sin(2\pi N_\ell h_k)=0. For ℓ<k\ell<k, 0<2πNℓhk<π/20<2\pi N_\ell h_k<\pi/2, so those sine terms are nonnegative. It follows that

ω(F,dRk)C≥∣F(hk,0)−F(0,0)∣≥Ak.\omega(F,d_{R_k})_C\ge\lvert F(h_k,0)-F(0,0)\rvert\ge A_k.

With rk=2Rkr_k=2^{R_k}, we have dRk=2−rkd_{R_k}=2^{-r_k} and

rklog⁡log⁡rk ω(F,2−rk)C≥log⁡(Rklog⁡2)log⁡Rk⟶1.r_k\log\log r_k\,\omega(F,2^{-r_k})_C\ge\frac{\log(R_k\log2)}{\log R_k}\longrightarrow1.

This proves (13). In fact, the specific FF has uniformly convergent rectangular Fourier sums, since its Fourier coefficients are absolutely summable.

Department of Mathematical Sciences, United Arab Emirates University, P.O. Box 15551, Al Ain, Abu Dhabi, United Arab Emirates

Email address: [email protected]

Email address: [email protected]

References

  1. [1]M. Bakhbukh and E. M. Nikishin, The convergence of the double Fourier series of continuous functions, Siberian Math. J. 14 (1973), 832–839. doi:10.1007/BF00975888.DOI
  2. [2]L. Carleson, On convergence and growth of partial sums of Fourier series, Acta Math. 116 (1966), 135–157.DOI
  3. [3]C. Fefferman, On the divergence of multiple Fourier series, Bull. Amer. Math. Soc. 77 (1971), 191–195.DOI
  4. [4]C. Fefferman and E. M. Stein, Hᵖ spaces of several variables, Acta Math. 129 (1972), 137–193.DOI
  5. [5]R. A. Hunt, On the convergence of Fourier series, in Orthogonal Expansions and Their Continuous Analogues, Southern Illinois University Press, 1968, pp. 235–255.
  6. [6]F. John and L. Nirenberg, On functions of bounded mean oscillation, Comm. Pure Appl. Math. 14 (1961), 415–426.
  7. [7]G. A. Karagulyan, Bounded oscillation operators on BMO spaces, Math. Z. 312 (2026), article 12. doi:10.1007/s00209-025-03901-9.DOI
  8. [8]D. I. Masyutin, Estimate of the rate of convergence of Fourier sums for functions from Lebesgue classes on a set of full measure, arXiv:2606.28786 (2026).arxiv.org/abs/2606.28786
  9. [9]K. I. Oskolkov, An estimate of the rate of approximation of a continuous function and its conjugate by Fourier sums on a set of total measure, Math. USSR-Izv. 8 (1974), no. 6, 1372–1386.DOI

Paper details

vlnc:W_yr5x5ayh
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents