betaFeedback

WorksMathematics

An asymptotic ball algebra with countably determined spectra and a nonseparable character space

Abstract

Let X=∂B2⊂C2X=\partial\mathbb B_2\subset\mathbb C^2 and let R\mathcal R be the restrictions to XX of functions in the ball algebra A(B2‾)A(\overline{\mathbb B_2}). The unital uniform Banach algebra A={(fn)∈ℓ∞(N,C(X)):distC(X)(fn,R)⟶0} \mathcal A=\{(f_n)\in\ell^\infty(\mathbb N,C(X)):\mathrm{dist}_{C(X)}(f_n,\mathcal R)\longrightarrow0\} is inverse-closed in ℓ∞(N,C(X))\ell^\infty(\mathbb N,C(X)). Consequently one countable family of evaluation characters generates the spectrum of every element of A\mathcal A. Nevertheless Δ(A)Δ(\mathcal A) contains continuum many pairwise disjoint nonempty open subsets and is therefore nonseparable. This gives a negative answer to a problem posed by Ohrysko, Wojciechowski and Graham.

The statement and the base algebra

For a commutative unital complex Banach algebra AA, let Δ(A)\Delta(A) denote its character space with the Gelfand topology. A subset S⊂Δ(A)S \subset\Delta(A) generates all spectra if

σA(a)={φ(a):φ∈S}(a∈A).\sigma_A(a) = \{\varphi(a) : \varphi\in S\} \qquad(a \in A).

Problem 18 of [1] asks whether the existence of such a countable SS forces Δ(A)\Delta(A) to be separable. In Nikolski’s terminology [2 Definition 0.2.1], the spectrum of AA is 1-visible from S⊂Δ(A)S \subset\Delta(A) if σA(a)={φ(a):φ∈S}\sigma_A(a) = \{\varphi(a) : \varphi\in S\} for every a∈Aa \in A. The classical Wiener–Pitt phenomenon, observed by Wiener and Pitt [3] and proved rigorously by Šre˘ıder [4], illustrates the failure of this property for Fourier evaluations on the convolution algebra M(T)M(\mathbb{T}) of complex Borel measures on the circle. There exist measures μ\mu such that {μ^(n):n∈Z}⊊σM(T)(μ)\{\widehat{\mu}(n) : n \in\mathbb{Z}\} \subsetneq\sigma_{M(\mathbb{T})}(\mu). Our construction shows that 1-visibility from a countable family is nevertheless compatible with a nonseparable character space.

Theorem 1. There is a commutative unital semisimple uniform Banach algebra AA such that all spectra are generated by one countable family of characters, but Δ(A)\Delta(A) has continuum many pairwise disjoint nonempty open subsets.

Write

B2={z=(z1,z2)∈C2:∣z1∣2+∣z2∣2<1},X=∂B2.B_2 = \{z = (z_1,z_2) \in\mathbb{C}^2 : |z_1|^2 + |z_2|^2 < 1\}, \qquad X = \partial B_2.

Let A(B2)A(B_2) be the algebra of continuous functions on B2‾\overline{B_2} that are holomorphic in B2B_2, and put

R={F∣X:F∈A(B2)}⊂C(X).\mathcal{R} = \{F|_X : F \in A(B_2)\} \subset C(X).

The maximum-modulus principle gives ∥F∥B2‾=∥F∣X∥X\|F\|_{\overline{B_2}} = \|F|_X\|_X. Thus restriction is an isometric algebra isomorphism onto the closed unital subalgebra R\mathcal{R}. For h∈Rh \in\mathcal{R} denote its unique ball-algebra extension by h~\widetilde{h}, and define

χ0(h)=h~(0).\chi_0(h) = \widetilde{h}(0).

This is a character of norm one. If zi∈Rz_i \in\mathcal{R} denotes the iith coordinate function on XX, then

χ0(z1)=χ0(z2)=0,∣z1(x)∣2+∣z2(x)∣2=1(x∈X).(1)\chi_0(z_1) = \chi_0(z_2) = 0, \qquad|z_1(x)|^2 + |z_2(x)|^2 = 1 \qquad(x \in X). \tag*{(1)}

Lemma 2. The subalgebra R\mathcal{R} is inverse-closed in C(X)C(X).

Proof. Let h=F∣X∈Rh=F|_{X}\in\mathcal{R} and assume hh has no zeros on XX. By continuity and compactness there is 0<r<10<r<1 such that FF has no zeros on the shell r<∥z∥≤1r<\lVert z\rVert\leq1. The function 1/F1/F is holomorphic on B2∖rB2\mathbb{B}_{2}\setminus r\mathbb{B}_{2}. The classical Hartogs extension theorem extends it to a holomorphic function GG on B2\mathbb{B}_{2}; see, for example, [5 Theorem 2.3.2]. Since FG=1FG=1 on the shell, the identity theorem gives FG=1FG=1 throughout B2\mathbb{B}_{2}. The function GG agrees with 1/F1/F near the boundary and hence extends continuously to B‾2\overline{\mathbb{B}}_{2}. Therefore 1/h=G∣X∈R1/h=G|_{X}\in\mathcal{R}.

Notice that inverse-closedness here is a statement about single functions. The pair z1,z2z_{1},z_{2} has no common zero on XX, although χ0\chi_{0} annihilates both.

The asymptotic algebra and its spectra

Give E=ℓ∞(N,C(X))E=\ell^{\infty}(\mathbb{N},C(X)) its usual pointwise product and norm

∥f∥E=sup⁡n∥fn∥X.\lVert f\rVert_{E}=\sup_{n}\lVert f_{n}\rVert_{X}.

Define

A={f=(fn)∈E:dist⁡(fn,R)⟶0}.(2)\mathcal{A}=\{f=(f_{n})\in E:\operatorname{dist}(f_{n},\mathcal{R})\longrightarrow0\}. \tag*{(2)}

Equivalently,

A=ℓ∞(N,R)+c0(N,C(X)).(3)\mathcal{A}=\ell^{\infty}(\mathbb{N},\mathcal{R})+c_{0}(\mathbb{N},C(X)). \tag*{(3)}

Indeed, approximants hn∈Rh_{n}\in\mathcal{R} with ∥fn−hn∥X→0\lVert f_{n}-h_{n}\rVert_{X}\to0 can be chosen bounded whenever ff is bounded.

Lemma 3. The set A\mathcal{A} is a norm-closed unital subalgebra of EE and is inverse-closed in EE.

Proof. The set c0(N,C(X))c_{0}(\mathbb{N},C(X)) is an ideal of EE, and ℓ∞(N,R)\ell^{\infty}(\mathbb{N},\mathcal{R}) is a unital subalgebra. Hence (3) is a unital subalgebra. To check closedness, suppose f(k)→ff^{(k)}\to f uniformly in EE. The inequality

dist⁡(fn,R)≤∥fn−fn(k)∥X+dist⁡(fn(k),R)\operatorname{dist}(f_{n},\mathcal{R})\leq\lVert f_{n}-f_{n}^{(k)}\rVert_{X}+\operatorname{dist}(f_{n}^{(k)},\mathcal{R})

gives lim sup⁡ndist⁡(fn,R)≤∥f−f(k)∥E\limsup_{n}\operatorname{dist}(f_{n},\mathcal{R})\leq\lVert f-f^{(k)}\rVert_{E}. Letting k→∞k\to\infty proves f∈Af\in\mathcal{A}.

For inverse-closedness, let f∈Af\in\mathcal{A} be invertible in EE, so that ∣fn(x)∣≥δ>0|f_{n}(x)|\geq\delta>0 for all nn and xx. Choose bounded hn∈Rh_{n}\in\mathcal{R} such that ϵn=∥fn−hn∥X→0\epsilon_{n}=\lVert f_{n}-h_{n}\rVert_{X}\to0. For all sufficiently large nn, ∣hn∣≥δ/2|h_{n}|\geq\delta/2 on XX. Lemma 2 gives hn−1∈Rh_{n}^{-1}\in\mathcal{R}, and

∥fn−1−hn−1∥X≤2ϵnδ2⟶0.\lVert f_{n}^{-1}-h_{n}^{-1}\rVert_{X}\leq\frac{2\epsilon_{n}}{\delta^{2}}\longrightarrow0.

The finitely many initial coordinates do not affect condition (2). Thus f−1∈Af^{-1}\in\mathcal{A}.

For n∈Nn\in\mathbb{N} and x∈Xx\in X let ϵn,x(f)=fn(x)\epsilon_{n,x}(f)=f_{n}(x). These are characters and separate the elements of A\mathcal{A}, so A\mathcal{A} is semisimple. Its norm is uniform, since it is a closed subalgebra of the commutative C∗C^{*}-algebra EE.

Proposition 4. Let D⊂XD\subset X be any countable dense subset. Then

S={ϵn,d:n∈N, d∈D}S=\{\epsilon_{n,d}:n\in\mathbb{N},\ d\in D\}

is countable and generates all spectra in A\mathcal{A}. More explicitly,

σA(f)=⋃n∈Nfn(X)‾={fn(d):n∈N, d∈D}‾.(4)\sigma_{\mathcal{A}}(f)=\overline{\bigcup_{n\in\mathbb{N}}f_{n}(X)}=\overline{\{f_{n}(d):n\in\mathbb{N},\ d\in D\}}. \tag*{(4)}

Proof. Each value fn(x)f_{n}(x) belongs to the spectrum by evaluation at ϵn,x\epsilon_{n,x}, so the closure of their union is contained in σA(f)\sigma_{\mathcal{A}}(f). If λ\lambda is outside that closure, then ∣fn(x)−λ∣|f_{n}(x)-\lambda| has a uniform positive lower bound. Lemma 3 shows that (f−λ1)−1∈A(f-\lambda1)^{-1}\in\mathcal{A}. This proves the first equality. The second follows from continuity of every fnf_{n} and density of DD.

Restriction to the countable set N×D\mathbb{N}\times D embeds EE, and hence A\mathcal{A}, isometrically into ℓ∞(N×D,C)\ell^{\infty}(\mathbb{N}\times D,\mathbb{C}). The image of A\mathcal{A} is closed and inverse-closed there, by the same uniform lower-bound argument. Since N×D\mathbb{N}\times D is countable, this proves the last realization assertion of Theorem 1.

Finite-coordinate and ultrafilter characters

For L⊂NL \subset\mathbb{N} put

(eL)n=1L(n)1X.(e_L)_n = 1_L(n)1_X.

Every eLe_L belongs to ℓ∞(N,R)⊂A\ell^\infty(\mathbb{N},\mathbb{R}) \subset\mathcal{A} and is an idempotent. Write en=e{n}e_n=e_{\{n\}}, and let Zi∈AZ_i \in\mathcal{A} be the constant sequence whose every term is ziz_i, for i=1,2i=1,2.

Lemma 5. If φ∈Δ(A)\varphi\in\Delta(\mathcal{A}) and φ(en)=1\varphi(e_n)=1, then φ=εn,x\varphi=\varepsilon_{n,x} for some x∈Xx \in X. In particular,

∣φ(Z1)∣2+∣φ(Z2)∣2=1.|\varphi(Z_1)|^2 + |\varphi(Z_2)|^2 = 1.

Proof. The algebra enAe_n\mathcal{A} is exactly C(X)C(X) supported in coordinate nn, since every such sequence belongs to c0(N,C(X))c_0(\mathbb{N},C(X)). Restricting φ\varphi gives a unital character of C(X)C(X), hence evaluation at some xx. For every f∈Af \in\mathcal{A}, multiplicativity gives φ(f)=φ(enf)=fn(x)\varphi(f)=\varphi(e_nf)=f_n(x). Now apply (1). □

Lemma 6. For every free ultrafilter pp on N\mathbb{N} there is a character Φp∈Δ(A)\Phi_p \in\Delta(\mathcal{A}) satisfying

Φp(Z1)=Φp(Z2)=0,Φp(eL)={1,L∈p,0,L∉p.(5)\Phi_p(Z_1)=\Phi_p(Z_2)=0,\qquad \Phi_p(e_L)= \begin{cases} 1, & L \in p,\\ 0, & L \notin p. \end{cases} \tag*{(5)}

Proof. For f∈Af \in\mathcal{A}, choose bounded hn∈Rh_n \in\mathbb{R} with ∥fn−hn∥X→0\lVert f_n-h_n\rVert_X \to0 and set

Φp(f)=lim⁡n→pχ0(hn).(6)\Phi_p(f)=\lim_{n\to p}\chi_0(h_n). \tag*{(6)}

If knk_n is a second choice, then

∣χ0(hn)−χ0(kn)∣≤∥hn−kn∥X⟶0.|\chi_0(h_n)-\chi_0(k_n)|\leq\lVert h_n-k_n\rVert_X \longrightarrow0.

A free ultrafilter contains every cofinite set, so (6) is independent of the choice. Linearity follows by using hn+knh_n+k_n as approximants to a sum. For products, if hnh_n approximates fnf_n and knk_n approximates gng_n, then hnknh_nk_n approximates fngnf_ng_n, because the sequences are bounded. The multiplicativity of χ0\chi_0 and of scalar ultrafilter limits gives multiplicativity of Φp\Phi_p. Also Φp(1)=1\Phi_p(1)=1 and ∣Φp(f)∣≤∥f∥E|\Phi_p(f)|\leq\lVert f\rVert_E. Finally, choose the exact approximants ziz_i and 1L(n)1X1_L(n)1_X to obtain (5). □

Continuum many disjoint open subsets

Set

V={φ∈Δ(A):∣φ(Z1)∣2+∣φ(Z2)∣2<1/4}.V=\{\varphi\in\Delta(\mathcal{A}):|\varphi(Z_1)|^2+|\varphi(Z_2)|^2<1/4\}.

This is open and nonempty by Lemma 6. Lemma 5 shows that every φ∈V\varphi\in V annihilates every ene_n, and therefore every finite sum of the ene_n.

Choose a family F\mathcal{F} of continuum many infinite subsets of N\mathbb{N} with pairwise finite intersections. Such a family is obtained by enumerating the finite binary words and taking the sets of prefixes of distinct infinite binary sequences. For L∈FL\in\mathcal{F} define

UL=V∩{φ:∣φ(eL)−1∣<1/2}.(7)U_L=V\cap\{\varphi:|\varphi(e_L)-1|<1/2\}. \tag*{(7)}

These sets are open. Since eLe_L is idempotent, φ(eL)∈{0,1}\varphi(e_L)\in\{0,1\}; hence the second condition in (7) simply means φ(eL)=1\varphi(e_L)=1.

Each ULU_L is nonempty: choose a free ultrafilter pp containing LL and use Φp\Phi_p. If L≠ML\ne M and φ∈UL∩UM\varphi\in U_L\cap U_M, multiplicativity gives

1=φ(eL)φ(eM)=φ(eL∩M).1=\varphi(e_L)\varphi(e_M)=\varphi(e_{L\cap M}).

But L∩ML\cap M is finite, and every character in VV annihilates finite-coordinate idempotents. This is a contradiction. The sets ULU_L are therefore pairwise disjoint.

A countable dense subset would have to meet each of these continuum many pairwise disjoint nonempty open sets. Consequently Δ(A)\Delta(\mathcal{A}) is nonseparable, completing the proof of Theorem 1.

Remark 7 (Two-function form of the open sets). For L⊂NL \subset\mathbb{N} put

AL=1−eL+eLZ1,BL=eLZ2.A_L = 1-e_L+e_LZ_1,\qquad B_L=e_LZ_2.

At every coordinate evaluation one has

∣εn,x(AL)∣2+∣εn,x(BL)∣2=1.|\varepsilon_{n,x}(A_L)|^2+|\varepsilon_{n,x}(B_L)|^2=1.

Nevertheless, if pp is free and L∈pL \in p, then Φp(AL)=Φp(BL)=0\Phi_p(A_L)=\Phi_p(B_L)=0. Moreover,

UL={φ:∣φ(AL)∣2+∣φ(BL)∣2<1/4}.U_L=\{\varphi:|\varphi(A_L)|^2+|\varphi(B_L)|^2<1/4\}.

Thus the disjoint open sets require two simultaneous conditions. There is no conflict with (4), which concerns one element at a time.

Remark 8 (The Shilov boundary). The Shilov boundary of AA is exactly the closure of {εn,x:n∈N,x∈X}\{\varepsilon_{n,x}:n\in\mathbb{N},x\in X\}, and is separable. Indeed, this closure is a closed norming set. Conversely, finite-coordinate functions from C(X)C(X) provide functions supported in arbitrarily small neighborhoods of each εn,x\varepsilon_{n,x}, showing that all these evaluations are strong boundary points. The open set VV lies outside this closed boundary. In particular, nonseparability is not being inferred merely from a nonseparable closed subset of a possibly separable space: it follows from the explicit disjoint open sets above.

Remark 9 (A general construction principle). The same proof works for any compact Hausdorff separable XX and any closed unital inverse-closed subalgebra R⊂C(X)R\subset C(X) admitting a character χ\chi outside the evaluation image of XX. By compactness, there are finitely many ri∈Rr_i\in R with χ(ri)=0\chi(r_i)=0 and inf⁡x∈X∑i∣ri(x)∣2>0\inf_{x\in X}\sum_i|r_i(x)|^2>0. Replace z1,z2z_1,z_2 by these rir_i throughout, and use a countable dense subset of XX in Proposition 4. No metrizability is used. The ball algebra on ∂B2\partial B_2 supplies an explicit instance of all the hypotheses.

Remark 10 (The global implication for corona theorems). The preceding amplification principle supplies the additional argument needed to connect Problem 18 with corona theorems. Let Ω\Omega be a separable complex domain and let H=H∞(Ω)H=H^\infty(\Omega). Put F={δz:z∈Ω}⊂Δ(H)F=\{\delta_z:z\in\Omega\}\subset\Delta(H). The space FF is compact and separable: evaluations at a countable dense subset of Ω\Omega are dense in it. Restriction of the Gelfand transform identifies HH isometrically with a closed unital subalgebra R⊂C(F)R\subset C(F). This subalgebra is inverse-closed: a uniform lower bound on FF is a uniform lower bound on Ω\Omega, so the reciprocal is again bounded and holomorphic. If the corona is nonempty, then Δ(R)\Delta(R) has a character outside FF. The amplification principle therefore produces an algebra with countably generated spectra and nonseparable character space. Consequently, a positive answer to Problem 18 for all commutative unital Banach algebras would imply the corona theorem for H∞(Ω)H^\infty(\Omega), including balls and polydisks. This implication is stronger than merely applying a putative positive answer directly to H∞(Ω)H^\infty(\Omega), which would give only separability of its character space.

Use of artificial intelligence

The proof was obtained in collaboration with OpenAI’s GPT 6.1 Sol model. The author has read the entire manuscript and takes full responsibility for its contents.

References

  1. [1]P. Ohrysko, M. Wojciechowski, and C. C. Graham, Non-separability of the Gelfand space of measure algebras, Ark. Mat. 54 (2016), no. 2, 525–535. doi:10.1007/s11512-016-0240-8, Definition 16 and Problem 18.DOI
  2. [2]N. Nikolski, In search of the invisible spectrum, Ann. Inst. Fourier (Grenoble) 49 (1999), no. 6, 1925–1998. doi:10.5802/aif.1743.DOI
  3. [3]N. Wiener and H. R. Pitt, On absolutely convergent Fourier–Stieltjes transforms, Duke Math. J. 4 (1938), no. 2, 420–436. doi:10.1215/S0012-7094-38-00435-1.DOI
  4. [4]Yu. A. Šreĭder, The structure of maximal ideals in rings of measures with convolution, Mat. Sb. (N.S.) 27(69) (1950), no. 2, 297–318 (in Russian). English translation in Amer. Math. Soc. Transl. no. 81 (1953), 28 pp.
  5. [5]L. Hörmander, An Introduction to Complex Analysis in Several Variables, 3rd revised ed., North-Holland Mathematical Library, vol. 7, North-Holland, Amsterdam, 1990.

Paper details

vlnc:W_s28bvdxe
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY-SA 4.0
Usage

views · 0

downloads · 0

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

  • Mathematics

GPT-6.1 Sol (bare)

View statement
Looking for related papers…
Contents