We study the truncated Dirichlet functional Φn(f)=∑k=1nk∣ak∣2,f(z)=∑k=0∞akzk, when f belongs to the Bloch space. If n≥27, we find a normalized Bloch function fn such that Φn(fn)>nBn2, where Bn=max{∣an∣:∥f∥B=1}, thus proving that an inequality by Kayumov and Wirths cannot be extended to large values of n. Nevertheless, we show that the inequality Φn(f)≤nBn2 holds for all normalized functions when n≤9.
Introduction
Let D be the open unit disc of the complex plane C. The Bloch spaceB is defined as the space of all analytic functions in D such that
ρB(f)=∣z∣<1sup{∣f′(z)∣(1−∣z∣2)}<∞.(1)
Note that ρB is a seminorm, and it is well-known that B becomes a Banach space when endowed with the norm ∥f∥B=∣f(0)∣+ρB(f). An important property of B is that it is Möbius invariant, i.e., ρB(f)=ρB(f∘φ) for every disk automorphism φ. Actually, as shown by Rubel and Timoney [11], it is the largest Banach space of holomorphic functions (admitting a nontrivial continuous linear functional) with this property.
Another Möbius invariant space is the Dirichlet spaceD. We say that an analytic function f belongs to D if
ρD(f)2=∫D∣f′(z)∣2dA(z)<∞,
where dA(z) represents the normalized area measure of D. Actually, D is the unique Hilbert space with this property [2] and it is easy to see that the inclusion D⊂B is proper. In the case that f is univalent, the change of variables ζ=f(z) shows that ρD2(f) is precisely the normalized area of f(D). If f(z)=∑k≥0akzk, Parseval’s identity yields the alternative representation
ρD2(f)=k=1∑∞k∣ak∣2.
We refer the reader to the seminal paper [1] and the monograph [4] for further information about B and D, respectively.
In 1978, Wirths [12] showed that if n≥2 and f(z)=∑k≥0akzk is a normalized Bloch function (i.e., a function with ∥f∥B=1), then
∣an∣≤2nn+1(n−1n+1)2n−1=:Bn.(2)
Furthermore, equality is possible if and only if f is a rotation of z↦Bnzn (see [7 Corollary 2]). If n=1, it trivially follows from (1) that ∣a1∣≤1 so we define B1=1 accordingly.
Related to this bound, consider the quadratic functional
Φn(f)=k=1∑nk∣ak∣2,f∈B.
In other words, Φn(f) is the truncation of the (perhaps divergent) series defining ρD2(f). In 2017, Kayumov [5] proved that
Φn(f)≤nBn2(3)
for all f∈B with ∥f∥B≤1 when n≤5. Actually, his proof yields that
k=[n/2]∑nk∣ak∣2≤nBn2,
where [x] denotes the integer part of x, for all n[5 p. 467]. These results were later complemented by Kayumov and Wirths [6], who showed that (3) holds for n=6 as well. Other related extremal problems for coefficients of Bloch functions have been studied since then (e.g., [8] and [10]).
In [3], it was proved that (3) holds for all n when the coefficients of f are non-negative. Furthermore,
nBn2≤max{Φn(f):∥f∥B=1}≤(n−1)n−1nn<e4nBn2,
so the bound nBn2 is of the correct order of magnitude. Even though it was not made explicit in the original paper [5], it was mentioned in [9 p. 6] that (3) could be valid for every n.
Despite this evidence, the first main result of this work is to show that (3) does not hold when n is large enough.
Theorem 1. Let n≥27. Then, there exists f∈B with ∥f∥B=1 such that
Φn(f)>nBn2.
Nevertheless, we are able to show that (3) holds for all n≤9. Namely, we use a refinement of the methods of [5] to prove the following:
Theorem 2. Suppose that 6≤n≤9. Then we have
Φn(f)≤nBn2
for all normalized Bloch functions. Furthermore, equality holds if and only if f(z)=λBnzn for some unimodular constant λ.
Observe that, as stated before, the case n=6 was already known [6]. However, the present argument will represent a new proof of this fact.
To the best of our knowledge, the cases 10≤n≤26 are still open.
This work is organized as follows: the construction of the counterexamples to (3) which lead to the proof of Theorem 1 can be found in Section 2. Indeed, we will present one function for the case n=27 and another that covers all other cases n≥28. Section 3 is devoted to the positive solution for n∈{6,7,8,9}. In Section 4, we extend Theorem 1 to the family of generalized weighted Dirichlet functionals Φnt. Finally, an appendix with a series of elementary but technical results is included for the correct understanding of the main sections.
AI statement
AI statement. This work has been carried out with assistance from Claude Opus 5.5 (Anthropic). It began with the aim of continuing the work [3] and attempting to prove (3) for every n. When asked about the case n=7, it produced a first draft of solution for 6≤n≤9. In particular, it proposed the use of Lemma 5 and the restricted parameter κ that can be found in Section 3. While exploring a suitable replacement of Lemma 5 for larger n, it found a counterexample for (3) when n≥28. Further discussions led to the failure of the case n=27 as well. All these proofs have been distilled, simplified and written in their final form by the authors.
Acknowledgements
Acknowledgements. The three authors were partially supported by the grant PID2025-168349NB-I00 from Ministerio de Ciencia, Innovación y Universidades and by the project CIGE/2026/195 from Conselleria d’Educació, Cultura i Universitats, Generalitat Valenciana. The work of the second author was also supported by the grant PID2024-160326NA-I00 from the same ministry.
Failure of (3) for large n
In this section we will construct the counterexample needed to prove Theorem 1. In light of [3 Theorem 2], the coefficients of this function must not have constant argument. Actually, the same proof yields that it is not possible that
argak=α+kβ,1≤k,
for certain α,β∈R. This represents the main difficulty of the following lemma.
Lemma 3. Let n≥4. Consider the function
fε(z)=εz+Bnzn−2n−1ε(n+12n)2Bn2z2n−1,ε>0.(4)
Then, there exists εn>0 such that ρB2(fε)=1+(n+1)216ε2 for all 0<ε<εn.
(recall the definition of Bn (2)). If r>0 and ε are fixed, observe from (5), (6) and (7) that ∣fε′(reit)∣2 is a concave quadratic function of cos((n−1)t). That quadratic function attains its maximum at its vertex
and equality in (8) is possible when e(n−1)ti is a purely imaginary number. This in particular shows that ρ∞2(fε)≥1+(n+1)216ε2 for every ε>0.
We will now show that the converse inequality holds if ε is small enough. If r=0, it is clear that ∣fε′(0)∣2=ε2<1+(n+1)216ε2 if ε<1. From now on assume that 0<r<1 and thus vε(r) is well defined. In order to simplify the later notation, write R=(n−1n+1r2)n−1 so that
vε(r)=8εn+1(R1−R).
Observe that vε(r) is a continuous and strictly decreasing function of R (and therefore of r) and monotonic with respect to ε.
Even if there is no t such that cos((n−1)t)=vε(r), the following inequality holds:
(2) When 0<r<r0, in particular vε(r)>vε(r0)>vε(rn)=0. Pick εn<1 such that vεn(r0)≥1. For every 0<ε<εn the quantity fε′(reit)2 is then maximized when e(n−1)ti=1. Thus
since R<1 because r<r0<rn. The monotonicity of the function rn−1(1−r2) (see Lemma 10 in the Appendix) yields that we may also choose εn such that nBnr0n−1(1−r02)+εn<1. Under this new assumption, we have that fε′(reit)(1−r2)<1 if 0<r<r0. This concludes the proof.
Proof of Theorem 1 for n≥28. Let fε be the function defined in (4) and assume that 0<ε<εn. Then we have that
The proof concludes by observing that the sequence in the left-hand side of (9) is increasing and that 28(2927)27>4. □
Observe that n=28 is the smallest integer such that (9) holds. Hence, a counterexample for n=27 must be necessarily different. To this end, we will add an extra high-order term which will not contribute to Φn.
Proof of Theorem 1 for n=27. Given ε>0, consider the polynomial
This elementary but ingenious inequality allows us to deduce (3) for 1≤n≤5 when substituting by the critical radius rn=n+1n−1 (see [5]). However, it is not sufficient by itself to cover the cases 6≤n≤9.
We will divide the analysis into two complementary propositions based on the relative magnitude of the n-th Taylor coefficient ∣an∣ with respect to the first one ∣a1∣.
We first consider the case where the n-th coefficient is relatively small. In this case, the defect generated by the intermediate coefficients combined with Parseval’s identity is sufficient to establish the desired inequality under some additional technical hypotheses which will be checked later in Proposition 8.
Proposition 4. Let n≥3. Consider the sequence
(wk)wk=k(n+1n−1)k−1,1≤k,
and define
(wmin)wmin=2≤k≤n−1min{wk}.
Assume that there exists a number κ such that the conditions
Applying hypothesis (H1) guarantees that ∑k=1nk∣ak∣2<nBn2. □
We now turn to the complementary statement of Proposition 4 where the n-th coefficient is dominant. To this end, we first need the following refinement of (11).
Lemma 5. Let r∈(0,1) and m≥1 be an integer. If f(z)=∑k≥0akzk and ∥f∥B≤1, then
Proof. Consider 0<r<1 and the real-valued function G(t)=∣f′(reit)∣2, which is continuous on [0,2π] and clearly satisfies
0≤G(t)≤(1−r2)21,∀t∈[0,2π].
Writing G(t)=f′(reit)f′(reit) and taking into account that the power series of f′(z) converges absolutely on {∣z∣=r}, we obtain that the Fourier coefficients (cl)l∈Z of G are given by
Proof. Applying Lemma 5 for m=n−1 at rn=n+1n−1, we have
k=1∑∞kwk∣ak∣2+∣R∣≤(1−rn2)21=nwnBn2.
Subtracting wnΦn(f) from both sides yields the desired inequality since w1=1. □
Similarly to Proposition 4, we will need to control the interaction between a1, the dominant an and a2n−1. Observe that this last coefficient does not play a role for Φn(f) but it is present in the remainder R introduced in Lemma 6. This task will require delicate estimates based on further technical hypotheses which we can safely assume when 6≤n≤9.
Proposition 7. Let n≥3 and consider the sequence introduced in (wk) and the quantity wmin defined in (wmin). Define
wmax=2≤k≤n−1max{wk}.
Assume that there exists a number κ satisfying (H0) and the additional conditions
2κwn+4(wmin−wn)κ2wn2wmax<1.
and
2κwn+wn−1<1.
Suppose that f(z)=∑k≥0akzk with ∥f∥B≤1 satisfies
n∣an∣2≥κ2wn34(1−wn)2∣a1∣2.(16)
then
k=1∑nk∣ak∣2≤nBn2.
Furthermore, equality holds if and only if f(z)=λBnzn for a certain unimodular constant λ.
It remains to characterize the equality case Φn(f)=nBn2. Suppose that Φn(f)=nBn2, so that wn(nBn2−Φn(f))=0. Since y2n−1<y1 gives a strict inequality above, we only need to analyze the case y2n−1≥y1. We split up this analysis into two parts
If y1>0, since y2n−1≥y1>0 and the hypothesis (H3) is strict, we get that the inequality in (20) is actually strict. Therefore, wn(nBn2−Φn(f))>0.
If y1=0, by the previous case we can suppose that y2n−1=0. Now, since hypotheses (H0) and (H2) are strict, we obtain the inequalities in (21) and (19) force yk=0 for k≥n+1. Then, analysing the squares in (18) we get yk=0 for 2≤k≤n−1. Therefore, Φn(f)=nBn2 holds if and only if f(z)=λBnzn for a certain unimodular constant λ.
This finishes the proof. □
Proposition 8. For n∈{6,7,8,9} there exists κ=κ(n) such that the hypotheses (H0), (H1), (H2) and (H3) are satisfied.
Proof. To complete the proof, we summarize the required numerical values in the Table 1 and Table 2.
n
w1
w2
w3
w4
w5
w6
w7
w8
w9
wmax
wmin−wn
6
1
710
4975
343500
24013125
1680718750
4975
168073125
7
1
23
1627
1627
256405
512729
40965103
1627
4096729
8
1
914
2749
7291372
656112005
1968333614
531441823543
47829696588344
7291372
4782969823543
9
1
58
2548
125256
125256
31256144
1562528672
78125131072
390625589824
125256
39062535176
Table 1. Values of (wj)1≤j≤n for 6≤n≤9.
n
κ
κwn2(wn−1)
(H1)
(H2)
(H3)
6
21
93753886
1.77577<10.9806
32941723028125<1
3361413261<1
7
41
51038056
4.38177<12.8426
41943041617651<1
3276813159<1
8
41
8235431805375
6.6811<14.7010
34867844011730263843<1
531441292102<1
9
41
73728199199
11.4974<16.5568
214697265625214468927488<1
390625272927<1
Table 2. Verification of (H1), (H2) and (H3) for 6≤n≤9.
Proof of Theorem 2. The result follows directly by combining Propositions 4, 7 and 8.
Remark 1.The proof technique presented here reaches its natural limit at n=9 and cannot be extended to n≥10. Indeed, hypothesis (H0) strictly requires that wmin>wn and, for example, for n=10 we have
w10−w2=10(119)9−1118≈0.0067.
Generalized weighted Dirichlet functionals
In [3], the authors also investigated a generalized weighted truncated Dirichlet functional defined for any parameter t≥0 by
Φnt(f)=k=1∑nkt∣ak∣2,
and analyzed the validity of the corresponding bound
Φnt(f)≤ntBn2(22)
for all functions f(z)=∑k=0∞akzk in B satisfying ∥f∥B≤1. As noted in [3], if t1≤t2 then the validity of (22) for t1 implies its validity for t2. Hence, the inequality (22) holds unconditionally for all t≥2 by applying (11), while for functions with non-negative Taylor coefficients it remains true for t≥1 (see [3]). However, when t<1, the estimate fails for all sufficiently large n, establishing t=1 as the critical threshold for this inequality among functions with non-negative coefficients.
The function considered in Theorem 1 to disprove inequality (3) for large n is also suitable as a counterexample for the generalized inequality (22) when t<2.
Corollary 9. Let 1≤t<2. Then, there exists nt≥2 such that for every n≥nt we have that
Φnt(f)>ntBn2.
for a function f∈B with ∥f∥B=1.
Proof. Let nt be such that
n2−t(n+1n−1)n−1>4,n≥nt.
The index nt certainly exists because the sequence is unbounded. Now, consider n≥nt and fε the function defined in (4) and assume that 0<ε<εn, where εn is the constant defined in Lemma 3. Then, proceeding as in the proof of Theorem 1 we have that
Let x1=n(n−1)(n+1)(n−2) be the unique zero of g′′. Observe that x1<1 because of the fact that (n+1)(n−2)=n2−n−2<n(n−1).
Since g′ is strictly decreasing in (x1,n−1n+1] and g′(1)=0, we have that g′(x1)>0. But g′(0)<0, so Bolzano’s theorem yields that there exists x2∈(0,x1) such that g′(x2)=0. Given the monotonicity of g′ on that interval, x2 must be the unique root there.
Summing up, we deduced that g is decreasing on [0,x2] and on [1,n−1n+1]; and increasing on [x2,1]. Hence
g(x2)<g(1)=n+14≤1=g(0).
Again, Bolzano’s intermediate values theorem yields that there exists a (unique) x0∈(0,x2) such that g(x0)=n+14. Given the monotonicity of g, we will have that
x0≤x≤n−1n+1max{g(x)}=max{g(x0),g(1)}=n+14.
Lemma 12.Consider the polynomial
gε(ζ)=14ε+14ζ−14εζ2+εζ3.
Then there exists 0<R1<1<R2<(1314)13 and ε0=ε0(R1,R2) such that
Proof. If x≤0 the right hand side is 0. If 0<x<C, the inequality is reduced to βC≥βx. Finally, if x≥C, then x−C+βC≥β(x−C)+βC=βx. □
References
Departamento de Matemáticas, Universidad Autónoma de Madrid, 28049 Madrid, Spain Email address:[email protected]
Departamento de Matemáticas, Universidad Autónoma de Madrid, 28049 Madrid, Spain Email address:[email protected]
Departamento de Matemáticas, Universidad de Alicante, 03690 Sant Vicent del Raspeig, Spain Email address:[email protected]
References
[1]J. M. Anderson, J. Clunie, and Ch. Pommerenke, On Bloch functions and normal functions, J. Reine Angew. Math. 270 (1974), 12–37. MR 361090DOI
[2]J. Arazy and S. D. Fisher, The uniqueness of the Dirichlet space among Möbius-invariant Hilbert spaces, Illinois J. Math. 29 (1985), no. 3, 449–462. MR 786732DOI
[3]I. Efraimidis, A. Mas, and D. Vukotić, Estimates for truncated area functionals on the Bloch space, Proc. Amer. Math. Soc. 151 (2023), no. 9, 3845–3854. MR 4607629arxiv.org/abs/2208.10626
[4]O. El-Fallah, K. Kellay, J. Mashreghi, and T. Ransford, A primer on the Dirichlet space, Cambridge Tracts in Mathematics, vol. 203, Cambridge University Press, Cambridge, 2014. MR 3185375
[5]I. R. Kayumov, A note on an area-type functional of Bloch functions, Lobachevskii J. Math. 38 (2017), no. 3, 466–468. MR 3659227
[6]I. R. Kayumov and K.-J. Wirths, Coefficient inequalities for Bloch functions, Lobachevskii J. Math. 40 (2019), no. 9, 1319–1323. MR 4021521DOI
[8]————, On the sum of squares of the coefficients of Bloch functions, Monatsh. Math. 190 (2019), no. 1, 123–135. MR 3998335arxiv.org/abs/1904.10846
[9]______, Inequalities of Carlson type for α-Bloch functions, Mediterr. J. Math. 17 (2020), no. 3, Paper No. 83, 9. MR 4099644
[10]R. Khasianov, On the weighted sum of squares of the coefficients of Bloch functions, Anal. Math. Phys. 16 (2026), no. 1, Paper No. 13, 17. MR 5013332
[11]L. A. Rubel and R. M. Timoney, An extremal property of the Bloch space, Proc. Amer. Math. Soc. 75 (1979), no. 1, 45–49. MR 0529210DOI
[12]K.-J. Wirths, Über holomorphe Funktionen, die einer Wachstumsbeschränkung unterliegen, Arch. Math. (Basel) 30 (1978), no. 6, 606–612. MR 492223