WorksMathematics
A lower bound for the Bohr radius of the bidisc
Abstract
We prove that the Bohr radius of the bidisc, whose exact value is unknown, satisfies , improving Patel's bound . As in his argument, we pass to functions with positive real part and value at the origin; we bound the sum of the moduli of their coefficients of degree by instead of , where is the Sidon constant of , and we bound , and with computer-assisted certificates. In Lean 4 we also prove , assuming Patel's inequality coupling the coefficients of degrees two and three.
Introduction
For let be the unit polydisc and let the Schur class be the set of holomorphic maps . For and write , where . The Bohr radius of is
Bohr [9] proved that . The sharp value was found independently by Riesz, Schur and Wiener, and Bohr’s paper includes Wiener’s proof (see [8]). Boas and Khavinson [8] introduced for , proved and asked for the exact value of when . Defant, Frerick, Ortega-Cerdà, Ounaïes and Seip [11] showed that has order , and Bayart, Pellegrino and Seoane-Sepúlveda [6] proved as . No exact value of is known for any .
For the bidisc, functions of one variable show . Knese Knese [14, Corollary 1.2] proved : by his Theorem 1.1 and Andô’s inequality, is at least the root of . Baran, Pikul, Woerdeman and Wojtylak Baran, Pikul, Woerdeman and Wojtylak [5, Theorem 6.4] proved , and Griego Griego [12] lowered this to .
In an unrefereed solution to a problem on MathDB, Patel Patel [20] proved . He also reports the upper bound , from a rational inner function and an exact computation that his reproducibility package [21] carries out; we have not checked this bound, but if it holds, it and Theorem 1.1 confine to an interval of width about .
Sidon constants enter through the coefficients. For a finite set the Sidon constant is the least with
and we write
Bernal-González, Cabana, García, Maestre, Muñoz-Fernández and Seoane-Sepúlveda Bernal-González, Cabana, García, Maestre, Muñoz-Fernández and Seoane-Sepúlveda [7, Theorem 3.3] showed that is at least the root of , where is the Sidon constant of the -homogeneous polynomials in variables. For , (Remark 2.4). Newman’s bound , whose proof appears in Shapiro’s thesis [23 Chapter 3] (see also [19, (1)]), turns this equation into Knese’s, and Shapiro [23] showed that $S_k=\sqrt{k}$ if and only if $k\in\{1,2,4\}$. Few values of $S_k$ are known; {xref}Table 1` lists them with their sources.
We give Patel’s argument for the lower bound in full, confirm the exact rational computation behind it with our own programs, and improve the bound.
Theorem 1.1. The Bohr radius of the bidisc satisfies .
The gain over Patel’s bound, about , comes from the terms of degrees five, six and seven.
Patel reduces the problem to the coefficients of holomorphic functions on with and (§ 2). Carathéodory’s inequality on slices bounds sums of moduli of the coefficients of degree two and, with the help of a positive semidefinite Toeplitz form, of degree three, and a third inequality (Lemma 3.4) couples the two degrees; Patel combines the three into a bound for the coefficients of degrees two and three together (§ 3). The third inequality comes from the unitary realization of Schur functions on the bidisc [1, 4], which rests on Andô’s inequality [3]. In the degrees Patel uses the bound for the sum of the absolute values of the coefficients of degree of , which he derives from Knese’s estimate [14 Lemma 2.1].
Our improvement starts from a simple observation: Carathéodory’s inequality on slices also bounds this sum by (Lemma 2.3). By Newman’s bound and Shapiro’s theorem, for every , while . We prove , and with computer-assisted certificates built from positive definite Toeplitz matrices (Section 4); for our bound is slightly above the value that Ye et al. [24] determined. In the degrees we keep Newman’s bound , since the size of the certificates grows quickly with (Table 2). With rational upper bounds for and an explicit tail, an exact rational computation shows that Patel’s majorant, with these bounds, is less than at , and by continuity also at slightly larger , which proves Theorem 1.1 (§ 5). § 6 describes the computer checks and a formal proof in Lean, conditional on Lemma 3.4, of the non-strict bound , which needs only the computation at .
| bound | 2.187 | 2.363 | 2.552 |
| initial grid | |||
| boxes | 8,912 | 61,118 | 210,960 |
| distinct vertex labels | 269,769 | 4,333,585 | 46,344,804 |
| smallest margin |
Table 2. The certificates of Proposition 4.2.
The Carathéodory normalization and slices
Let the Carathéodory class be the set of holomorphic functions on with and . We write
and set and .
We use the following reduction.
Lemma 2.1. Let . If for every , then .
Proof. Constant functions satisfy . Let be nonconstant. By the maximum principle on . Multiplying by a unimodular constant does not change , so we may assume that lies in . Then lies in . Its coefficients of positive degree are those of multiplied by . Hence . □
Remark 2.2. The normalization loses nothing: if , then for every . Indeed, for the function maps into a compact subset of the right half-plane, hence into a disc with . Then lies in with , and becomes . Let . Thus .
Lemma 2.3. Let and . Then for , and .
Proof. Fix . The function is holomorphic on with . For ,
so . Letting , we obtain Carathéodory’s inequality . Now write with . As runs over , the quotient runs over and . Hence , and by the definition of ,
Remark 2.4. The same identification shows that : on a -homogeneous polynomial in two variables is, up to the factor , a polynomial in of degree at most . With this identification, the next corollary is the case of [7 Theorem 3.3].
Corollary 2.5. If and , then .
Proof. By Lemma 2.3, for every . Apply Lemma 2.1.
In [7] the corresponding estimate is proved for itself. There Wiener’s inequality takes the place of Carathéodory’s: the Taylor coefficients of a holomorphic satisfy for . Patel derives the bound for all from Knese’s estimate [14 Lemma 2.1], by a limiting argument with [20]. Lemma 2.3 with Newman’s bound gives it directly.
With the values of for in Table 1 and for , Corollary 2.5 gives . If the value of Ye et al. for and Newman’s bounds for and are replaced by the bounds of Proposition 4.2, it gives . Patel’s coupling of degrees two and three goes beyond this.
Table 1. Sidon constants .
The coefficients of degrees two and three
Throughout this section , and
so that and . Inner products are linear in the first argument, .
Lemma 3.1 (Patel [20]). .
Proof. Choose such that and have the same argument, and put and . Then , and . By Lemma 2.3, , and therefore .
Lemma 3.2. Let be a positive semidefinite Hermitian Toeplitz matrix of size , and let . Then , where .
Proof. Let be vectors in with , for example the columns of . Since is Toeplitz, the tuples and have the same Gram matrix. Hence extends to a well-defined linear isometry from the span of
onto the span of , and then to a unitary operator on the span of all the . Then , so . As is unitary, , and
$\xrefLemma 3.2 is a consequence of the Carathéodory–Toeplitz theorem. Ye et al. prove it in the same way [24 Lemma 3.1].
Proof. Replacing by for a suitable , we may assume that and have the same argument. Choose with , and let
The matrix is positive semidefinite. Since , it is also Toeplitz, with first row . By \xref`Lemma 3.2`, withc=(c_0,\ldots,c_3)^{\mathsf{T}}$,
Here and , which proves the first claim. For the second, apply it to from the proof of \xref`Lemma 2.3`. Then\lvert c_0\rvert+\lvert c_3\rvert=A_3/2$ and .
The proof above extends to a one-parameter family. For and , the matrix is positive semidefinite and Toeplitz, with first row , so the same argument shows
$\xrefLemma 3.3 is the case . The case is the inequality , from which Yu et al. obtain by the Cauchy–Schwarz inequality [25 (6)]. On its own, $\xrefLemma 3.3 gives only , more than . It is used together with the next inequality, which bounds in terms of .
Lemma 3.4 (Patel [20]). .
Proof. We use the transfer-function realization of Schur functions on the bidisc. The Schur–Agler class consists of the with for every pair of commuting strict contractions on a Hilbert space. By Andô’s inequality [3], every belongs to this class. By Agler’s theorem [1] (see also [4, 2]) there are a Hilbert space , complementary orthogonal projections and on , and an isometry of , the colligation, such that
For we have , so and the other operators , inverted below are invertible. We may take unitary, which matters if the realization is given with an isometric colligation, as in [14 Section 2]. Indeed, the operator on is unitary and agrees with on the first summand. We use it as the new colligation, with state space and projections and . Since the new is the old one, the new and agree with the old ones on , and the new projections map into itself, each term of the expansion of is unchanged. From now on , , , , , and denote the new data.
Apply this to , which lies in and vanishes at 0. Then , and since is unitary,
Put . These identities imply . Let . Then and . Since is a rank-one perturbation of and ,
Hence , and expanding the resolvent,
Relative to write
so that . Collecting the monomials , and in , we find
Put and . Since is an isometry, and , so . Since is an isometry, in the same way. Applying the Cauchy–Schwarz inequality to in the two groupings, we obtain
Finally , and likewise for , so .
Remark 3.5. Lemmas 3.1 and 3.4 are sharp. Equality holds in both for , where and , and for , where and .
Lemma 3.6 (Patel [20]). Let , and , and put
Then .
Proof. Expanding the squares verifies the identity
By Lemmas 3.1, 3.3 and 3.4, each term on the right is nonnegative. □
Thus is the Lagrangian bound, with multipliers , and , for the maximum of under the three inequalities of Lemmas 3.1, 3.3 and 3.4.
Sidon constants and certificates for small
The proof of Theorem 1.1 uses upper bounds for and for with . Degrees two and three are covered by Lemma 3.6.
The proof of Theorem 1.1 uses neither the value nor the value of below, so that it does not rely on these recent preprints. Ye et al. [24 Theorem 1.1] show that is the root in of . The proof of Proposition 4.2 rests on the following lemma.
Lemma 4.1. Let . For put
(a) .
(b) is convex.
(c) Let , and let be the Hermitian Toeplitz matrix with diagonal entries and entries for . If is positive definite, then .
(d) For , the arc lies in the triangle with vertices , and .
(e) , where .
Proof. (a) Write . If for all , then . Equality holds when attains this maximum and (any unimodular number if ). The substitution replaces by and by without changing , and a choice of and makes . (b) is a supremum of real-linear functions of . (c) By Lemma 3.2, for . Hence, by the Cauchy–Schwarz inequality,
(d) After a rotation, . The triangle is , and with satisfies and . (e) The map maps the set of admissible onto itself, since , and . Hence . □
Parts (a) and (c) are variants of the reduction to phase functionals and of the Toeplitz domination bound used by Ye et al. [24 (14) and Proposition 3.2], who normalize the phases differently [24 (17)]. Proposition 4.2 also differs from their proof in the covering of the phase space.
Proposition 4.2. , and .
Proof (computer-assisted). Let be , or . By Lemma 4.1(a) it suffices to show that for . By Lemma 4.1(e), which replaces by , we may also assume . For and the certificate covers by boxes, obtained from a uniform grid of boxes with half-widths by repeated bisection. For it covers in the same way, starting from a uniform grid of boxes with half-widths . By Lemma 4.1(d), the vectors with in the box lie in a product of triangles. By Lemma 4.1(b), the maximum of over this product is attained at one of its vertices. For each vertex that occurs, the certificate gives with positive definite and , so that by Lemma 4.1(c). A vertex with a coordinate occurs with two labels, phase and phase , and has a certificate for each. Table 2 gives the sizes of the three certificates, and § 6 describes the check. □
Numerical maximization, which is not part of the proof, gives and , so the bounds 2.363 and 2.552 are close to the true values. Further gains along these lines need bounds for with or sharper inequalities in the degrees two and three.
Proof of Theorem 1.1
Throughout the paper, lower bounds are rounded down and upper bounds up. Let . Let , , , , and for . By Table 1 and Proposition 4.2, for , and for all . Let , let , and fix and . Lemma 2.3 bounds by 2 for , by for and by for , and Lemma 3.6 bounds . Summing the series , we obtain the majorant
We take
All quantities in are then rational, and an exact computation shows . Since , and are fixed, is continuous on , so there is with . By Lemma 2.1, , which proves Theorem 1.1.
Remark 5.1 (Patel’s computation). With Patel’s parameters , and , with and with for , the function is the rational majorant that he evaluates in [20]; its value at his radius is less than 0.99999997857, below his bound .
Verification
Computer checks
The certificates of Proposition 4.2 are checked in the ball arithmetic of Arb [13], through python-flint [22]. The checker reads the numbers of the certificates as binary64 numbers, exactly, rebuilds the boxes from the recorded bisections and verifies at every vertex the condition stated in the proof of Proposition 4.2; for and , whose certificates are split into files, a further program checks, from the outputs of the file checks, that the files together cover the phase domain. The majorant of § 5 is evaluated in exact rational arithmetic, and so is Patel’s majorant in Remark 5.1. The programs and their output are archived with the certificates (see Data availability).
Formal proof
The formal proof is written in the Lean 4 proof assistant [15] with the Mathlib library [16]. The formal statements use Mathlib’s definitions. A family is admissible if the series converges absolutely at every point of and its value there has modulus less than 1. Bohr’s inequality holds at if for every admissible family and every finite set , and is the supremum of the radii at which it holds. It agrees with the definition in the introduction: a nonconstant maps into , and a constant of modulus 1 satisfies Bohr’s inequality with equality. The main theorem is the following.
Theorem 6.1 (formalized in Lean). Suppose that , the inequality of Lemma 3.4, holds for every family with whose series converges absolutely on to values with positive real part, that is, for the coefficients of every . Then .
The proof of the hypothesis rests on Andô’s inequality and Agler’s realization, which are not in Mathlib. The formal proof follows Sections 2 to 5 and checks the certificates of Proposition 4.2 in one of two ways.
(1) A checking program written in Lean validates the certificates, converted into Lean source files, in exact rational arithmetic and ball arithmetic, and a Lean proof shows that every part it accepts implies the bound of Lemma 4.1(c) on the boxes of that part, so that the converted files need not be trusted. These checks run as compiled code, through native_decide, so the trusted base includes Lean’s compiler and runtime; the rest of the proof uses only the standard axioms.
(2) A second formalization proves the three bounds of Proposition 4.2 with every check run by Lean’s kernel. With the step of the first proof that derives the main theorem from them, the main theorem follows from four theorems in which every check is run by the kernel; these are not combined into a single Lean theorem.
Data availability
The certificates of Proposition 4.2, the programs of § 6, the programs that produced the certificates, the controls and the output of all runs, and the Lean files of § 6.2, are archived at doi:10.5281/zenodo.23716440. The same files, except the 48 files of the certificate for , the 576 files of the certificate for and an archive of generated Lean files of the kernel-only proof of , are also in the repository https://github.com/tejasvi/bidisc-bohr-radius/tree/v6. The README of the archive lists the Lean theorems and projects, with the versions of Lean and Mathlib, and gives the commands that run the checks and build the proofs. The certificate for has the SHA-256 checksum ce3e9dd41a0dbd9c3a8ea49b04aa7aaf21f0fe557291a2084c7739461a460a14, and the archive lists the checksums of the other files.
Use of AI tools
Generative AI tools were used to find and prove the results, to write the programs, and to draft the text. The author reviewed the arguments and the programs and takes responsibility for the content.
References
ORCID: https://orcid.org/0000-0003-4668-3639
References
- [1]J. Agler, On the representation of certain holomorphic functions defined on a polydisc, in: Topics in Operator Theory: Ernst D. Hellinger Memorial Volume, Oper. Theory Adv. Appl. 48, Birkhäuser, Basel, 1990, 47–66.
- [2]J. Agler and J. E. McCarthy, Pick interpolation and Hilbert function spaces, Grad. Stud. Math. 44, Amer. Math. Soc., Providence, RI, 2002.DOI
- [3]T. Andō, On a pair of commutative contractions, Acta Sci. Math. (Szeged) 24 (1963), 88–90.
- [4]J. A. Ball and T. T. Trent, Unitary colligations, reproducing kernel Hilbert spaces, and Nevanlinna–Pick interpolation in several variables, J. Funct. Anal. 157 (1998), 1–61.DOI
- [6]F. Bayart, D. Pellegrino and J. B. Seoane-Sepúlveda, The Bohr radius of the n-dimensional polydisk is equivalent to √(log n)/n, Adv. Math. 264 (2014), 726–746.
- [7]L. Bernal-González, H. J. Cabana, D. García, M. Maestre, G. A. Muñoz-Fernández and J. B. Seoane-Sepúlveda, A new approach towards estimating the n-dimensional Bohr radius, Rev. R. Acad. Cienc. Exactas Fís. Nat. Ser. A Math. RACSAM 115 (2021), no. 2, Paper No. 44.
- [8]H. P. Boas and D. Khavinson, Bohr’s power series theorem in several variables, Proc. Amer. Math. Soc. 125 (1997), 2975–2979.arxiv.org/abs/math/9606203
- [9]H. Bohr, A theorem concerning power series, Proc. London Math. Soc. (2) 13 (1914), 1–5.DOI
- [11]A. Defant, L. Frerick, J. Ortega-Cerdà, M. Ounaïes and K. Seip, The Bohnenblust–Hille inequality for homogeneous polynomials is hypercontractive, Ann. of Math. (2) 174 (2011), 485–497.DOI
- [12]S. Griego, Improve C₅₉ upper bound, pull request no. 75 to the repository teorth/optimizationproblems, GitHub, May 2026, https://github.com/teorth/optimizationproblems/pull/75.
- [13]F. Johansson, Arb: efficient arbitrary-precision midpoint-radius interval arithmetic, IEEE Trans. Comput. 66 (2017), 1281–1292.arxiv.org/abs/1611.02831
- [14]G. Knese, Three radii associated to Schur functions on the polydisk, Proc. Amer. Math. Soc. Ser. B 12 (2025), 48–63; arXiv:2410.21693v3.DOI
- [15]L. de Moura and S. Ullrich, The Lean 4 theorem prover and programming language, in: Automated Deduction – CADE 28, Lecture Notes in Comput. Sci. 12699, Springer, 2021, 625–635.DOI
- [16]The mathlib Community, The Lean mathematical library, in: Proc. 9th ACM SIGPLAN Int. Conf. Certified Programs and Proofs (CPP 2020), 2020, 367–381.DOI
- [18]S. Neuwirth, The Sidon constant of sets with three elements, arXiv:math/0102145, 2001.DOI
- [19]S. Neuwirth, On the (Fourier analytic) Sidon constant of {0, 1, 2, 3}, arXiv:2603.28229, 2026 (note dated 2013).arxiv.org/abs/2603.28229
- [20]S. Patel, Improved lower and upper bounds for the bidisc Bohr radius, solution to the problem “Exact value of the Bohr radius of the bidisc”, MathDB, second solution on the page, posted 27 August 2026 (UTC) and edited the same day, https://mathdb.com/p/392062/exact-value-of-the-bohr-radius-of-the-bidisc#s2 (consulted 21 October 2026).
- [21]S. Patel, Complex phases and a certified upper bound for the bidisc Bohr radius: reproducibility package, version 1.0.0, Zenodo, 5 September 2026, doi:10.5281/zenodo.22341928.DOI
- [22]python-flint: Python bindings for FLINT, version 0.9.0 (FLINT 3.6.0), https://github.com/flintlib/python-flint.
- [23]H. S. Shapiro, Extremal problems for polynomials and power series, Master’s thesis, Massachusetts Institute of Technology, 1951.
- [24]Z. Ye et al., The exact Sidon constant of six consecutive frequencies, preprint, October 2026, https://github.com/ABCDEFGCoder/The-Sidon-Constant-of-n-Consecutive-Frequencies.
- [25]H. Yu et al., The exact Sidon constant of {0,1,2,3} and its formalization, preprint, ai.viXra.org:2606.0009, 2026 (dated 10 May 2026); Lean formalization at https://github.com/yhx-12243/Sidon3.