Ann Glob Anal Geom DOI 10.1007/s10455-011-9266-0
ORIGINAL PAPER
The Weierstrass factorization theorem for slice regular functions over the quaternions
Graziano Gentili · Irene Vignozzi
Received: 8 November 2010 / Accepted: 5 April 2011 © Springer Science+Business Media B.V. 2011
Abstract The class of slice regular functions of a quaternionic variable has been recently introduced and is intensively studied, as a quaternionic analogue of the class of holomorphic functions. Unlike other classes of quaternionic functions, this one contains natural quatern- ionic polynomials and power series. Its study has already produced a rather rich theory having steady foundations and interesting applications. The main purpose of this article is to prove a Weierstrass factorization theorem for slice regular functions. This result holds in a formula- tion that reflects the peculiarities of the quaternionic setting and the structure of the zero set of such functions. Some preliminary material that we need to prove has its own independent interest, like the study of a quaternionic logarithm and the convergence of infinite products of quaternionic functions.
Keywords Functions of a quaternionic variable · Weierstrass factorization theorem · Zeros of hyperholomorphic functions
Mathematics Subject Classification (2000) 30G35 · 30C15 · 30B10
1 Introduction
The well-known theory of Fueter regular functions [9,10,25] is the first and most celebrated candidate among a few others, in the search for a quaternionic analogue of the theory of holo- morphic functions of one complex variable. This well-developed theory [3,19Ð21] inspired also the setting of Clifford algebras (see, e.g. [2]). Recently, Gentili and Struppa proposed a different approach, which led to a new notion of holomorphicity (called slice regular- ity) for quaternion-valued functions of a quaternionic variable [15,16]. Unlike Fueter’s, this
G. Gentili (B) · I. Vignozzi Dipartimento di Matematica “U. Dini”, Università di Firenze, Viale Morgagni 67/A, 50134 Firenze, Italy e-mail: [email protected]fi.it I. Vignozzi e-mail: [email protected]fi.it 123 Ann Glob Anal Geom theory includes the polynomials and the power series of the quaternionic variable q of the n ∈ H type n≥0 q an, with coefficients an . Furthermore, the analogs (sometime peculiarly different) of many of the fundamental properties of holomorphic functions of one complex variable can be proven in this new setting, like the Cauchy and Pompeiu representation formu- las and Cauchy inequalities, the maximum (and minimum) modulus principle, the identity principle, the open mapping theorem, the Morera theorem, the power and Laurent series expansion, the Runge approximation theorem, to cite only some of the most significant (see [4Ð7,12Ð16,24]). In fact, the theory of slice regular functions is already rather rich and well established on steady foundations, and appears to be of fundamental importance to construct a functional calculus in non commutative settings [8]. Let H denote the skew field of quaternions. Its elements are of the form q = x0 + 2 2 2 ix1 + jx2 + kx3 where the xl are real, and i, j, k are such that i = j = k =−1, ij =− ji = k, jk =−kj = i, ki =−ik = j. We set Re(q) = x0,Im(q) = ix1 + jx2 +kx3, | |= 2 + 2 + 2 + 2 ( ) ( ) | | q x0 x1 x2 x3 and call Re q ,Imq and q the real part,theimaginary part and the module of q, respectively. The conjugate√of the√ quaternion q,definedasq¯ = Re(q)−Im(q) = x0 −ix1 − jx2 −kx3, satisfies |q|= qq¯ = qq¯ and allows the definition ¯ inverse q = q−1 = q . S of the of any element 0as |q|2 If is the unit sphere of purely imaginary S ={ ∈ H : 2 =− }={ = + + : 2 + 2 + 2 = } quaternions, i.e., q q 1 q ix1 jx2 kx3 x1 x2 x3 1 , then every quaternion q which is not real (i.e. with Im(q) = 0) can be written as q = x + Iy = ( ), =| ( )| = Im(q) ∈ S ∈ S for x Re q y Im q and I |Im(q)| . Moreover, using the same I , we can write q =|q|eϑ I ,forsomeϑ ∈ R. We can now state the following definition, see [4]. Definition 1.1 Let be a domain in H. A function f : → H is said to be slice regular if, for every I ∈ S, its restriction fI to the complex line L I = R + RI passing through the origin and containing 1 and I has continuous partial derivatives and satisfies 1 ∂ ∂ ∂ I f (x + yI) := + I fI (x + yI) = 0, 2 ∂x ∂y in I = ∩ L I . In what follows, we will simply call regular a slice regular function. If is a generic domain of H, then there are examples of regular functions defined on which are not even continuous, see e.g. [4]. In [5], and in this same article [4], a class of pathology-preventing domains of definition for regular functions were introduced, as follows:
Definition 1.2 Let be a domain in H, intersecting the real axis. If I = ∩ L I is a domain in L I C for all I ∈ S then we say that is a slice domain. As it is well known, the domains of holomorphy are the most natural domains of definition for holomorphic functions of a complex variable. The same can be said, in the setting of qua- ternionic regular functions, for slice domains which have the following additional property:
Definition 1.3 A subset C of H is axially symmetric if, for all x + yI ∈ C with x, y ∈ R, I ∈ S, the whole set x + yS ={x + yJ : J ∈ S} is contained in C. For the sake of simplicity, we will call such a C a symmetric set. As it is proved in [4], any regular function defined on a slice domain can be (uniquely) extended to the smallest sym- metric slice domain which contains , and for this reason we will always consider regular functions defined on symmetric slice domains. 123 Ann Glob Anal Geom
One of the interesting, basic features of regular functions on slice domains is the structure of their zero set: it consists of isolated points (zeros) and isolated spheres of type x + yS, with x, y ∈ R (spherical zeros). The fact that pointwise product does not preserve regularity led to the definition of a regular product (the ∗-product) which maintains regularity and allows the study of a factorization for quaternionic regular polynomials, as well as the construction of different notions of multiplicity for their roots (and for the zeros of regular functions), [11,17,18,23]. This article is devoted to the study of a quaternionic version of the well-known Weierst- rass factorization theorem, which will be valid for entire regular functions, i.e. for regular functions defined on the whole of H. The main result that we obtain is in fact the following.
Theorem 1.4 (Weierstrass factorization theorem for regular functions) Let f be an entire regular function. Let: m ∈ N be the multiplicity of 0 as zero of f , {bn}n∈N ⊆ R \{0} be the sequence of the (non zero) real zeros of f , {Sn = xn + ynS}n∈N be the sequence of the spherical zeros of f ,and{an}n∈N ⊆ H \ R be the sequence of the non real zeros of f with isolated multiplicity greater then zero. If all the zeros listed above are repeated according to their multiplicities, then there exists a never vanishing, entire regular function h and, for all ∈ N ∈ δ ∈ = ( ) +| ( )|S , , ∈ N n ,thereexistcn Sn, n San Re an Im an , rn n mn such that f (q) = qm R(q) S(q) A(q) ∗ h(q) where, with obvious notations for the infinite ∗-product, ∞ − − qb−1+ 1 q2b−2+···+ 1 qrn b rn R( ) = ( − 1) n 2 n rn n , q 1 qbn e n=0 ∞ ( 2) ( n ) 2 ( ) 2Re(cn ) 1 2 2Re cn 1 n 2Re cn q 2qRe cn q + q +···+ q S(q) = − + 1 e |cn |2 2 |cn |4 n |cn |2 n , |c |2 |c |2 n=0 n n ∞ A( ) = ∗ ( − δ−1) ∗ ( ) q 1 q n gn q n=0 and where, for all n ∈ N, the function gn is the never vanishing entire regular function whose restriction to the plane Ln = R + R[Im(δn)] is given by − zδ−1+ 1 z2δ−2···+ 1 zmn δ mn | ( ) = n 2 n mn n . gn Ln z e
In particular, we can choose rn = n = mn = n for all n ∈ N.
The above result is achieved in several steps. Some of them are inspired by, and similar to, the steps that lead to the complex version of the Weierstrass factorization theorem, while other are new and quite diverse, mainly due to the peculiarities of the quaternionic setting and the fact that the structure of the zero-set of the entire regular functions includes spherical zeros, which are not present in the case of the entire holomorphic functions. This article is organized as follows. After having presented some necessary preliminary results, in Sect.3, we study criteria to establish the convergence of infinite products of quater- nions, and this latter necessity leads us to the (peculiar) definition of a quaternionic Logarithm. Then, in Sect.4, we investigate the uniform convergence of infinite products of regular func- tions. The results so found give us a tool to find, in Sect.5, conditions that can guarantee the uniform convergence of infinite products of regular functions to a regular function. In Sect. 6 we perform the study of different type of convergence-producing regular factors, some of 123 Ann Glob Anal Geom which arise only in the quaternionic setting. Section 7 is dedicated to give the definitions of spherical and isolated multiplicities for the zeros of a regular function; to do this, we follow the path which brought to the same definitions in the case of quaternionic regular polynomi- als. Eventually, in Sect.8, we prove the Weierstrass factorization theorem for entire regular functions. In all this procedure, the technical difficulties encountered are quite relevant, but we believe that the effort made to overcome them brought to a neat final result.
2 Preliminaries
This section is devoted to present some basic results that point out the main features of slice regular functions and of their domains of convergence. We begin with some technical prop- erties of the quaternions, and recall that (see e.g. [16]) for all I ∈ S and for all J ∈ S such that I ⊥ J,wehavethatIJ ∈ S and 1, I, J, K = IJform a basis for H with the same algebraic properties of the standard basis 1, i, j, k = ij. Moreover, the following result holds Proposition 2.1 Let I and J be two elements in S,let I, J∈R be the Euclidean scalar product of I and J in R4, and let I × J ∈ S be their natural vector product in R3. Then the quaternionic product I J can be decomposed through the following formula: IJ =− I, J+I × J. One of the interesting features of the regular functions is a splitting property, which turns out to be a key tool in the theory: Lemma 2.2 (Splitting Lemma) If f is a regular function on a slice domain then, for every I ∈ S and every J ⊥ IinS, there exist two holomorphic functions F, G : I → L I such that
fI (z) = F(z) + G(z)J for all z ∈ I . The Splitting Lemma naturally leads to a version of the identity principle for regular functions (see [4]). Theorem 2.3 (Identity Principle) Let f : → H be a regular function on a slice domain and let Z f ={q ∈ : f (q) = 0} be the zero-set of f . If there exists I ∈ S such that L I ∩ Z f has an accumulation point, then f ≡ 0 on . We will now recall some results (proved in [4,11,16]) on the algebraic and topological structure of the zero-set of a regular function. Theorem 2.4 Let f be a regular function on a symmetric slice domain .Ifthereexist x, y ∈ R and distinct imaginary units I, J ∈ S such that f (x + yI) = f (x + yJ) = 0,then f (x + yL) = 0 for all L ∈ S. If f is not identically zero, its zero set Z f consists of isolated points or isolated 2-spheres of the form S = x + yS, with x, y ∈ R,y= 0. It becomes now natural to give the following (see e.g. [11]) Definition 2.5 Let f be a regular function on a symmetric slice domain . An isolated 2-sphere S = x + yS of zeros of f is called a spherical zero of f . Any point q0 belonging to a spherical zero is called a generator of the spherical zero. Any zero of f not belonging to a spherical zero is said isolated zero or non spherical zero. 123 Ann Glob Anal Geom
For a regular function defined on a symmetric slice domain, both the set of spherical zeros and the set of isolated zeros have cardinality at most countable. Moreover
Corollary 2.6 Azeroq0 ∈/ R of a regular function f defined on a symmetric slice domain is a generator of a spherical zero if and only if its conjugate q0 is a zero of f as well. For the sake of completeness, we state here a result proven in [13], that we will use in the sequel.
Proposition 2.7 Let f be a regular function on a symmetric slice domain .Iff(L I ) ⊆ L I for some I ∈ S then the zero-set Z f consists of isolated points belonging to L I or isolated 2-spheres of type x + yS. As a consequence, if f (q) = 0 for every q ∈ L I then f (q) = 0 for every q ∈ H. Finally, if f (L I ) ⊆ L I for all I ∈ S then the zero-set Z f consists of real isolated points or isolated 2-spheres of type x + yS.
Since pointwise product of regular functions is not in general regular, inspired by the classic case of polynomials with coefficients in a non commutative algebra (see e.g. [22]), the ∗-multiplication (or regular multiplication) between regular functions f and g defined on the open unit ball B of H was introduced in [11] by means of the power series expansion, to guarantee the regularity of f ∗ g. The definition of ∗-multiplication was extended in [4] to the case of regular functions on a slice domain and it is based on the following result. Lemma 2.8 (Extension Lemma) Let be a symmetric slice domain and choose I ∈ S.If fI : I → H is holomorphic, then setting 1 I f (x + yJ) = [ fI (x + yI) + fI (x − yI)] + J [ fI (x − yI) − fI (x + yI)] 2 2 extends fI to a regular function f : → H. The function f is the unique such extension and it is denoted by ext( fI ). In order to define the regular product of two regular functions f, g on a symmetric slice domain ,letI, J ∈ S, with I ⊥ J, and choose holomorphic functions F, G, H, K : I → L I such that for all z ∈ I
fI (z) = F(z) + G(z)J, gI (z) = H(z) + K (z)J. (1)
Let fI ∗ gI : I → H be the holomorphic function defined by
fI ∗ gI (z) =[F(z)H(z) − G(z)K (z¯)]+[F(z)K (z) + G(z)H(z¯)]J. (2) Using the Extension Lemma 2.8, the following definition is given in [4]:
Definition 2.9 Let ⊆ H be a symmetric slice domain and let f, g : → H be regular. The function
f ∗ g(q) = ext( fI ∗ gI )(q) defined as the extension of (2) is called the regular product of f and g.
We now recall a few properties of the regular multiplication, see [11]and[4].
Remark 2.10 Let ⊆ H be a symmetric slice domain and let f, g : → H be regular. If f (L I ) ⊆ L I for all I ∈ S,then f ∗ g(q) = f (q) g(q). If f (L I ) ⊆ L I or g(L I ) ⊆ L I for all I ∈ S,then f ∗ g(q) = g ∗ f (q). 123 Ann Glob Anal Geom
An alternative expression of the regular product of two regular functions was introduced in [11]. In particular we have the next proposition proven in [4].
Theorem 2.11 Let f , g be regular functions on a symmetric slice domain . For all q ∈ , if f (q) = 0 then f ∗ g(q) = 0,else
f ∗ g(q) = f (q)g( f (q)−1qf(q))
Corollary 2.12 Let f , g be regular functions on a symmetric slice domain .Then f ∗ g(p) = 0 if and only if f (p) = 0 or f (p) = 0 and g( f (p)−1 pf(p)) = 0.
Definition 2.13 Let f be a regular function on a symmetric slice domain and suppose f splits on I as in formula (1), fI (z) = F(z)+G(z)J. We consider the holomorphic function c( ) = (¯) − ( ) fI z F z G z J (3) and define, according to the Extension Lemma 2.8,theregular conjugate of f by the formula c( ) = ( c)( ) = ( (¯) − ( ) ). f q ext fI q ext F z G z J
Furthermore, the following definition is given under the same assumptions.
Definition 2.14 The symmetrization of f is defined as s = ∗ c = c ∗ = ( c ∗ )( )) = ( ( ) (¯) + ( ) (¯)). f f f f f ext fI fI q ext F z F z G z G z (4)
s s Remark 2.15 It turns out that f (L I ) ⊆ L I for all I ∈ S, and hence the zero set of f has the property described in Proposition 2.7. Moreover, if f, g are regular functions on a symmetric slice domain, then is easy to verify that ( f ∗ g)c = gc ∗ f c and ( f ∗ g)s = f s gs = gs f s.
The zero-sets of f c and f s are characterized in [11,13] as follows.
Theorem 2.16 Let f be a regular function on a symmetric slice domain . For all x, y ∈ R with x + yS ⊆ , the zeros of the regular conjugate f c on x + yS are in one-to-one corre- spondence with those of f . Moreover, the symmetrization f s vanishes exactly on 2-spheres (or singleton) x + yS on which f has a zero. Note that x + yS is a 2-sphere if y = 0 and a real singleton {x} if y = 0.
3 Quaternionic logarithm and infinite products of quaternions
We consider an infinite product of quaternions ∞ q0q1 ...qi ...= qi i=0 and, for n ∈ N, we denote by Qn = q0q1 ...qn the partial products. In analogy with the complex case (see [1]), we give the following definition. ∞ Definition 3.1 The infinite product i=0 qi is said to converge if and only if at most a finite number of the factors are zero, and if the partial products formed by the non vanishing factors tend to a finite limit which is different from zero. 123 Ann Glob Anal Geom
In what follows, we will always refer to an infinite product assuming that the vanishing factors are a finite number, and we will check its convergence just looking at the product of the non-vanishing terms. We point out that in a convergent product we have that limi→∞ qi = 1. = −1 In fact, this is clear by writing qi Qi−1 Qi . It is therefore preferable to write all infinite products in the form ∞ (1 + ai ) i=0 so that limi→∞ ai = 0 is a necessary condition for their convergence.
Remark 3.2 In Definition 3.1, the requirement that the partial products of non vanishing ∞ ( + ) factors of i=0 1 ai tend to a finite limit different from zero finds its motivation in the ∞ ( + ) complex case. In fact, assuming this requirement, an infinite product i=0 1 ci with ∈ C ∞ ( + ) ci converges simultaneously with the series i=0 Log 1 ci , whose terms represent the values of the principal branch of the logarithm (see [1]). The convergence is not necessar- ily simultaneous if the limit of the infinite product is zero, as the following example shows. Let ∞ (1 − 1/i) (5) i=0 and consider the corresponding series ∞ Log(1 − 1/i). (6) i=0 S Denoting the partial sums of (6)bySn,wehavethatQn = e n . Since limn→∞ Sn =−∞, then ∞ Sn (1 − 1/i) = lim Qn = lim e = 0. n→∞ n→∞ i=0 Therefore, the infinite product (5) converges (to zero) while the series (6)diverges(to−∞).
To follow a similar approach to study the convergence of infinite products in the quatern- ionic case, we need to introduce a logarithm on H.Theexponential function on H is naturally defined as ∞ qn exp(q) = eq = n! n=0 and it coincides with the complex exponential function on any complex plane L I .
Definition 3.3 Let ⊆ H be a connected open set. We define a branch of the quaternionic logarithm (or simply a logarithm)on a function f : → H such that for every q ∈ e f (q) = q.
First of all, since exp(q) never vanishes, we must suppose that 0 ∈/ .Ifweset Im(q)/|Im(q)| if q ∈ H \ R I = q any element of S otherwise 123 Ann Glob Anal Geom
θ we have that for every q ∈ H \{0} there exists a unique θ ∈[0,π] such that q =|q|e Iq . Moreover θ = arccos(Re(q)/|q|). The function arccos(Re(q)/|q|) will be called the prin- cipal quaternionic argument of q and it will be denoted by ArgH(q) for every q ∈ H \{0}. Hence, we are ready to define the principal quaternionic logarithm. Definition 3.4 Let ln be the natural real logarithm. For every q ∈ H \ (−∞, 0], we define the principal quaternionic logarithm (or simply principal logarithm)ofq as Re(q) Log(q) = ln |q|+arccos I . |q| q This same definition of principal logarithm was already given in [20], in the setting of Clif- ford algebras. In fact, Definition 3.4 can be obtained by specialising to the case of quaternions the paravector logarithm introduced in [20, Definition 11.24, p. 231]. It is also important to state that: Proposition 3.5 The principal logarithm is a continuous function on H \ (−∞, 0].
Proof The function q → ln |q| is clearly continuous on H \{0}. The function q → Re(q) Im(q) ∈ H \ R arccos |q| |Im(q)| is defined and continuous for every q . Moreover, since for every strictly positive real r, and for every q ∈/ R,wehave Re(q) lim arccos Iq = 0 q→r |q| → Re(q) H \ (−∞, ] then q arccos |q| Iq is continuous on 0 . We observe that, if s, instead, is a strictly negative real number, then Re(q) lim arccos Iq q→s |q| does not exist. Remark 3.6 Our approach to the definiton of a quaternionic logarithm is inspired by the search for an inverse of the exponential map. While the principal logarithm obvioulsy satis- fies the relation eLog(q) = q in H \ (−∞, 0], the equality Log(eq ) = q is valid only in the domain {q ∈ H :|Im(q)| <π}. As one may expect, we can prove the following nice result Proposition 3.7 The principal quaternionic logarithm coincides with the principal complex logarithm on any complex plane L I , with I ∈ S.
Proof Let I ∈ S. First of all notice that, since L I = L−I , every quaternion q ∈ L I \ R, can be written both as x + yI and as x − y(−I ), for suitable x, y ∈ R.If(x, y) ∈ R2 \ (−∞, 0], then by Definition 3.4,wehavethat x y Log(x + yI) = ln x2 + y2 + arccos I x2 + y2 |y| 123 Ann Glob Anal Geom and x (−y) Log(x − y(−I )) = ln x2 + y2 + arccos (−I ) x2 + y2 |y| coincide. Now, we can consider the restriction of the principal quaternionic logarithm to L I : 2 2 y Log(x + yI) = ln( x + y ) + ArgH(x + yI) I. |y| φ( + ) = ( + ) y ( + ) [ ,π) We set x yI ArgH x yI |y| . The function ArgH x yI , with values in 0 , is the not oriented angle from the positive real half line to the vector x + yI. When y > 0, then x + yI belongs to the upper half plane of L I ,andφ(x + yI) = ArgH(x + yI).On the other hand if y < 0, we obtain that x + yI belongs to the lower half plane of L I ,and φ(x + yI) =−ArgH(x + yI). Therefore, φ(x + yI) coincide with the imaginary part of the principal complex logarithm of L I , and the proof is complete.
In the complex case, it is well known that the argument of a product is equal to the sum of the arguments of the factors (up to an integer multiple of 2π). In the quaternionic case, we have the following lemma. ∈ N θ ,...,θ ∈[ ,π) n θ <π Lemma 3.8 Let n and let 1 n 0 be such that i=1 i . Then for every set {I1,...,In}⊆S, we have that n θ1 I1 θn In ArgH(e ...e ) ≤ θi . i=1 = θ ,θ ,...,θ ∈[ ,π) Proof By induction on n.Forn 1, the thesis is straightforward. Let 1 2 n 0 n θ <π φ ∈[,π) be such that i=1 i .Let 0 be the principal quaternionic argument of the θ θ θ θ φ product e 1 I1 ...e n−1 In−1 and let J ∈ S be such that e 1 I1 ...e n−1 In−1 = e J . Consider the φ θ product e J e n In :
φ J θn In e e = cos φ cos θn + cos φ sin θn In + sin φcos θn J + sin φ sin θn JIn.
From the formula (2.1), we have that JIn =− J, In+J × In, and hence we obtain that φ J θn In φ J θn In cos (ArgH(e e )) = Re e e = cos φ cos θn − sin φ sin θn J, In.
Since J, In≤|J||In|=1andsinφ sinθn ≥ 0 we obtain the inequality
φ J θn In cos (ArgH(e e )) ≥ cos φ cos θn − sin φ sin θn = cos (φ + θn). The function cos (x) decreases in [0,π], and hence we have
φ J θn In ArgH(e e ) ≤ φ + θn. Thanks to the induction hypothesis, we have the thesis.