Abstract
In this paper we work in the ‘split’ discrete Clifford analysis setting, i.e. the m-dimensional function theory concerning null-functions, defined on the grid \({\mathbb {Z}}^m\), of the discrete Dirac operator \({\partial }\), involving both forward and backward differences, which factorizes the (discrete) Star-Laplacian. We show how the space \({\mathcal {M}}_k\) of discrete spherical monogenics homogeneous of degree k, is decomposable into irreducible \(\mathfrak {so}(m)\)-representations.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
In particle physics, the (massless) Dirac operator is a well-studied operator in the setting of elementary particle physics [14]. This operator can be studied in a more mathematical setting, namely Euclidean Clifford analysis in general dimensions, where the Dirac operator factorizes the Laplace operator, making Clifford analysis a refinement of harmonic analysis. The Dirac operator is a Spin(m)-invariant operator, or equivalently, invariant under its Lie algebra \(\mathfrak {so}(m)\).
The current paper deals with a special type of Clifford analysis, namely discrete Clifford analysis, where functions are defined on a grid \((h {\mathbb {Z}})^m\) rather than the continuous m-dimensional space \({\mathbb {R}}^m\). In recent years, a function theory studying discrete functions defined on the standard grid with mesh size 1 \(({\mathbb {Z}}^m)\) has given rise to a discrete counterpart of Euclidean Clifford analysis [2, 5]. Different choices have been made for the discrete Dirac operator [15, 16, 19], containing forward, backward and central differences.
The ‘split’ discrete Clifford analysis setting, which we are considering here (e.g. [4, 7, 13]) introduces a discrete Dirac operator \(\partial \) defined using only forward and backward differences, which factorizes the discrete star Laplace operator \(\Delta ^*\) [20]. The function theory has already lead to a number of results regarding polynomial solutions of this differential operator, namely a Taylor series decomposition [11], a Cauchy–Kowalewskaya extension theorem [12], etc.
In this paper, we are however more interested in the representation theoretical aspect of this theory. It is well known that in classical harmonic analysis, the Laplace operator is a rotational invariant operator, or equivalently, invariant under the Lie algebra \(\mathfrak {so}(m)\).
In [10], it has been shown that infinitesimal rotation operators can be defined in the split discrete Clifford analysis setting under which the star Laplacian is invariant. However, while the space \({\mathcal {H}}_k\) of discrete k-homogeneous harmonic polynomials is a representation of \(\mathfrak {so}(m)\), contrary to the classical harmonic case, it is not irreducible. This has been shown in [9], where a full decomposition has been made into irreducible representations. The aim of this paper is to do the same for the space \({\mathcal {M}}_k\) of discrete monogenic polynomials of arbitrary degree of homogeneity k. In the process, we will be able to define spinor spaces in the discrete setting.
In classical harmonic analysis, the infinitesimal ‘rotations’, i.e. the elements of the orthogonal Lie algebra corresponding to the rotation group SO(m), are given by the angular momentum operators \(L_{a,b} = x_a \partial _{x_b} - x_b \partial _{x_a}\). These operators satisfy the commutation relations
which are exactly the defining relations of the special orthogonal Lie algebra \(\mathfrak {so}(m)\) and they form endomorphisms of the space \({\mathcal {H}}_k({\mathbb {R}}^m,{\mathbb {C}})\) of scalar-valued harmonic k-homogeneous polynomials, thus transforming the latter in an (irreducible) \(\mathfrak {so}(m,{\mathbb {C}})\)-representation. To establish \({\mathcal {M}}_k({\mathbb {R}}^m,{\mathbb {S}})\), i.e. the spinor-valued homogeneous monogenics of degree k, classically as \(\mathfrak {so}(m,{\mathbb {C}})\)-representation, the following operators are considered
These operators are endomorphisms of the space of spinor-valued k-homogeneous polynomials in m vector variables which also satisfy the defining relations of \(\mathfrak {so}(m,{\mathbb {C}})\):
In [10], we developed similar operators in the discrete Clifford analysis setting: the angular momentum operators are discrete operators \(L_{a,b} = \xi _a \, {\partial }_b + \xi _b \, {\partial }_a, a \ne b\). For \(a = b\), we define \(L_{aa} = 0\). Then the operators \(\Omega _{a,b}\), acting on discrete functions f as \(\Omega _{a,b} \, f = L_{a,b}\, f \, e_b \, e_a\), satisfy the defining relations of the orthogonal lie algebra \({\mathfrak {s}}{\mathfrak {o}}(m)\):
Furthermore, they are endomorphisms of the space \({\mathcal {H}}_k\) of Clifford-algebra valued homogeneous harmonics of degree k, since \(\Omega _{a,b}\) commutes with \(\mathfrak {sl}_2 = \left\{ \Delta , \xi ^2, {\mathbb {E}} + \frac{m}{2} \right\} \), for all (a, b). Here \(\xi ^2\) is the square of the discrete vector variable \(\xi \) and \({\mathbb {E}}\) is the discrete Euler operator (see Sect. 2). In [9], we showed that \({\mathcal {H}}_k\) is the sum of \(2^{2m}\) isomorphic copies of the irreducible representation of \(\mathfrak {so}(m,{\mathbb {C}})\) with highest weight \((k,0,\ldots ,0)\).
The discrete Dirac operator \({\partial }\) is however not invariant under the operators \(\Omega _{a,b}\), hence \({\mathcal {M}}_k\) cannot be expressed as \(\mathfrak {so}(m,{\mathbb {C}})\)-representation by means of these operators. Therefore, we considered in [10] the operators \(L_{a,b} - \dfrac{1}{2}\) and the four-vector \(V_{a,b} = e_a\, e_b \,e_a^\perp \, e_b^\perp = - e_a^\perp \,e_a\,e_b^\perp \, e_b\). Let the operator \(dR(e_{a,b}), a \ne b\), act on discrete functions f as
For \(a = b\), we defined \(dR(e_{a,a}) = 0\). Note that, for the sake of continuity, we use the same notation for the discrete and continuous rotation operators. From this point on, we always refer to the discrete versions unless stated otherwise. The operators \(dR(e_{a,b})\) satisfy the defining relations of the special lie algebra \({\mathfrak {s}}{\mathfrak {o}}(m)\):
and commute with \(\mathfrak {osp}(1 | 2) = \left\{ {\partial }, \xi , {\mathbb {E}} + \frac{m}{2} \right\} \) which makes them endomorphisms of \({\mathcal {M}}_k\). As such, the space \({\mathcal {M}}_k\) is a reducible \(\mathfrak {so}(m,{\mathbb {C}})\)-representation. In [10], it was already suggested that \({\mathcal {M}}_k\) can be decomposed into irreducible parts of highest weight \((k)'_+ = \left( k + \frac{1}{2}, \frac{1}{2},\ldots ,\frac{1}{2}\right) \) resp. \((k)'_- = \left( k + \frac{1}{2}, \frac{1}{2},\ldots ,-\frac{1}{2}\right) \), but this was left as an open conjecture. In the following sections, we will show how this decomposition is done exactly.
2 Preliminaries
Let \({\mathbb {R}}^m\) be the m-dimensional Euclidian space with orthonormal basis \(e_j, j=1,\dots ,m\) and consider the Clifford algebra \({\mathbb {R}}_{m,0}\) over \({\mathbb {R}}^m\). Passing to the so-called ‘split’ discrete setting [7, 13], we embed the Clifford algebra \({\mathbb {R}}_{m,0}\) into the bigger complex one \({\mathbb {C}}_{2m,0}\), the underlying vector space of which has twice the dimension, and introduce forward and backward basis elements \( \mathbf{e }_j^{\pm }\) satisfying the following anti-commutator rules:
The connection to the original basis \(e_j\) is given by \(\mathbf{e}^+_j + \mathbf{e}^-_j = e_j, j=1,\ldots ,m\), which indeed implies that \(e_j^2=1\). We will often write \(e_j^\perp = \mathbf{e}_j^+ - \mathbf{e}_j^-\) and \(\mathbf{e}_j^+ \wedge \mathbf{e}_j^- = \mathbf{e}_j^+ \mathbf{e}_j^- - \mathbf{e}_j^- \mathbf{e}_j^+ = e_j^\perp e_j\).
Now consider the standard m-dimensional equidistant lattice \({\mathbb {Z}}^m\); the coordinates of a Clifford vector \({\underline{x}}\) will thus only take integer values. We construct a discrete Dirac operator factorizing the discrete Laplacian, using both forward and backward differences \(\Delta _j^{\pm }, j=1,\dots , m\), acting on Clifford-valued functions f as follows:
With respect to the \({\mathbb {Z}}^m\)-grid, the usual definition of the discrete Laplacian in \({\underline{x}}\in {\mathbb {Z}}^m\) is
This operator is also known as “Star Laplacian”; we will from now on simply write \(\Delta \). An appropriate definition of a discrete Dirac operator \({\partial }\) factorizing \(\Delta \), i.e. satisfying \({\partial }^2 = \Delta \), is obtained by combining the forward and backward basis elements with the corresponding forward and backward differences, more precisely
Denote the co-ordinate difference operators \({\partial }_j = \mathbf{e}^+_j \Delta ^+_j + \mathbf{e}^-_j \Delta ^-_j\) and consider the discrete co-ordinate vector variables \(\xi _j = \mathbf{e}_j^+ \, X_j^- + \mathbf{e}_j^- \, X_j^+, j= 1,\ldots ,m\), with \(X_j^\pm \) scalar operators. In order to receive an analogue of the classical Weyl relations \({\partial }_{x_j} x_k - x_k {\partial }_{x_j} = \delta _{jk}\), the co-ordinate vector variable operators \(\xi _j\) are defined by their interaction with the corresponding co-ordinate operators \({\partial }_j\), according to the skew Weyl relations, cf. [7]
which imply that \({\partial }_j\, \xi _j^k[1] = k \,\xi _j^{k-1}[1]\). The operators \(\xi _j\) and \({\partial }_j\) furthermore satisfy the following anti-commutator relations:
implying that \({\partial }_\ell \,\xi _j^k[1] = 0, j \ne \ell \).
The natural powers \(\xi ^k_j[1]\) of the operator \(\xi _j\) acting on the constant 1 are the basic discrete k-homogeneous polynomials of degree k in the variable \(x_j\), i.e. \({\mathbb {E}}\, \xi _j^k[1] = k\,\xi _j^k[1]\), where \({\mathbb {E}} = \sum _{j=1}^m \xi _j\,{\partial }_j\) is the discrete Euler operator. They constitute a basis for all discrete polynomials. Explicit formulas for \(\xi _j^k[1]\) are given for example in [7, 12]; furthermore \(\xi _j^k[1](x_j)=0\) if \( k \geqslant 2\,|x_j|+1\).
A discrete function taking values in the Clifford algebra \({\mathbb {C}}_{2m}\) is discrete harmonic (resp. left discrete monogenic) in a domain \(\Omega \subset {\mathbb {Z}}^m\) if \(\Delta f({\underline{x}}) = 0\) (resp. \({\partial }f({\underline{x}}) = 0\)), for all \({\underline{x}}\in \Omega \). The space of discrete harmonic (resp. monogenic) homogeneous polynomials of degree k (i.e. \({\mathbb {E}} f = k f\)) is denoted \({\mathcal {H}}_k\) (resp. \({\mathcal {M}}_k\)), while the space of all discrete harmonic (resp. monogenic) homogeneous polynomials is denoted \({\mathcal {H}}\) (resp. \({\mathcal {M}}\)). It is clear that
The respective dimensions of these spaces over \({\mathbb {C}}\) are:
The calculations are similar to the classical case (see e.g. [2]).
3 Orthogonal Lie Algebras
As the remainder of this paper will deal with the explicit construction of representations for the orthogonal algebra \(\mathfrak {so}(m, {\mathbb {C}})\), we will start by briefly introducing this Lie algebra \(\mathfrak {so}(m,{\mathbb {C}})\); a detailed description can be found for example in [17]. In even dimension \(m=2n\), the Lie algebra \(\mathfrak {so}(m, {\mathbb {C}})\) is generated as a matrix algebra as follows. Define \(E_{i,j}\) as the \((m \times m)\)-matrix with entry 1 on position (i, j), while all other entries are 0. Then we can define the matrices
for all \(1 \leqslant i,j \leqslant n\). Note that due to the symmetry of \(Y_{i,j}\) and \(Z_{i,j}\) in the indices, it suffices in fact that for those sets of operators, we pick \(i<j\). The matrices \(H_i, X_{i,j}, Y_{i,j}\) and \(Z_{i,j}\) generate exactly the Lie algebra \(\mathfrak {so}(2n, {\mathbb {C}})\) using the commutator as Lie bracket. When considering the same set of generators as \((2n+1 \times 2n+1)\)-matrices, in addition with the matrices
for all \(1 \leqslant i \leqslant n\), we find a set of generators for \(\mathfrak {so}(2n+1,{\mathbb {C}})\), again using the commutator as Lie bracket. From now on we consider these generators as abstract elements, satisfying the same relations as their matrix equivalents. Hence
The Cartan subalgebra is chosen as
independently of the parity of the dimension, i.e. \(\mathfrak {so}(2n,{\mathbb {C}})\) and \(\mathfrak {so}(2n+1,{\mathbb {C}})\) are both Lie algebras of rank n. The roots of \(\mathfrak {so}(m,{\mathbb {C}})\) (see also [21]) are determined by considering the action of the Cartan algebra on the other generators of the adjoint representation of \(\mathfrak {so}(m, {\mathbb {C}})\). Hence, for all \(1 \leqslant a, b, c, d \leqslant n\):
Here \(\left\{ L_a, 1 \leqslant a \leqslant n \right\} \) is a basis of the dual vector space \(\mathfrak {h}^*\) of the Cartan subalgebra \(\mathfrak {h}\), i.e. \(L_a \left( H_b \right) = \delta _{a,b}\). Note in particular that the Cartan subalgebra elements \(H_a\) appears in the commutator of a certain positive root with a negative root of the same index:
We thus deduce the following roots and root vectors.
Root | Root vector |
---|---|
\(m = 2n\) | |
\(L_a - L_b\) | \(X_{a,b}\) |
\(L_a + L_b\) | \(Y_{a,b}\) |
\(-L_a - L_b\) | \(Z_{a,b}\) |
\(m = 2n+1\) | |
\(L_a - L_b\) | \(X_{a,b}\) |
\(L_a + L_b\) | \(Y_{a,b}\) |
\(-L_a - L_b\) | \(Z_{a,b}\) |
\(L_a\) | \(U_a\) |
\(-L_a\) | \(V_a\) |
By the usual convention (see e.g. [17]), we choose the positive roots in even dimension to be
and negative roots
In odd dimension, one chooses positive roots
and negative roots
In [10], we introduced the algebra \(\mathfrak {so}(m,{\mathbb {C}})\) (up to an isomorphism) in the discrete Clifford analysis context. The generators of \(\mathfrak {so}(m,{\mathbb {C}})\) were not given in terms of the root vectors and Cartan subalgebra, but rather by the generators \(\left\{ dR(e_{a,b}): 1 \leqslant a \ne b \leqslant m \right\} \), see (1) satisfying the defining relations of \(\mathfrak {so}(m,{\mathbb {C}})\):
see [10]. The next step is to identify both realisations of \(\mathfrak {so}(m, {\mathbb {C}})\) in the discrete Clifford analysis setting, by determining the explicit expressions of the root vectors and Cartan subalgebra.
4 Decomposition of \({\mathcal {M}}_k\) in Irreducible Representations
Since the definition of the generators of \(\mathfrak {so}(m, {\mathbb {C}})\) differs in even and odd dimensions, we have to make a distinction. We start with the even dimensional case.
4.1 Even Dimension \(m =2n\)
Definition 1
We define the operators \(H_a, X_{a,b}, Y_{a,b}\) and \(Z_{a,b}\) \(\in \mathfrak {so}(m,{\mathbb {C}})\):
Note that, because \(dR(e_{a,b}) = - dR(e_{b,a})\), we find that \(Y_{b,a} = -Y_{a,b}\) and \(Z_{b,a} = -Z_{a,b}\). For \(X_{a,b}\), we find that \(X_{b,a} \ne X_{a,b}\) and that \(X_{a,a} = H_a\), hence we will only consider couples (a, b) with \(a \ne b\).
The original operators \(dR(e_{a,b})\) can be reconstructed as linear combinations of the operators \(X_{a,b}, Y_{a,b}, Z_{a,b}\) and \(H_a\).
Straightforward calculations, which make use of (2) show that these operators satisfy the same commutator relations as their matrix equivalents. In particular, we have the following lemma.
Lemma 1
The operators \(H_c, X_{a,b}, Y_{a,b}\) and \(Z_{a,b}, 1 \leqslant a, b, c, d \leqslant n\), satisfy the commutation relations of \(\mathfrak {so}(m, {\mathbb {C}})\):
This means that the operators \(X_{a,b}\) with \(a < b\) and \(Y_{a,b}\) are indeed positive root vectors corresponding to the roots \(L_a - L_b\), resp. \(L_a + L_b\). Furthermore, \(X_{a,b}\) with \(a > b\) and \(Z_{a,b}\) are negative root vectors corresponding to the roots \(L_a - L_b\) resp. \(-L_a - L_b\).
Proof
Since the commutator relations between the operators \(dR(e_{a,b})\) are the same as those between the operators \(\Omega _{a,b}\) of the harmonics, the proof is completely similar as the proof given in [9].
We already established in [10] that since the operators \(dR(e_{a,b})\) are endomorphisms of \({\mathcal {M}}_k, {\mathcal {M}}_{k}\) is a representation of \(\mathfrak {so}(2n,{\mathbb {C}})\) although this representation is not irreducible. The next step in our reasoning is to decompose \({\mathcal {M}}_k\) into irreducible representations of \(\mathfrak {so}(m, {\mathbb {C}})\). This is done by splitting 1 into a sum of idempotents, see further. When constructing irreducible representations, the main target is to find so-called highest weight vectors. These are in our context functions belonging to \({\mathcal {M}}_k\) which:
-
vanish under the action of all positive root vectors,
-
are simultaneous eigenfunctions for the action of all \(H_a\).
The property of such highest weight vectors is that they generate an entire irreducible representation by all consecutive actions of negative root vectors. The aim is to construct idempotents L such that there exists a discrete monogenic function \(P_k\), such that \(P_k L\) satisfies the conditions for a highest weight vector. Here \(P_k\,L\) denotes the multiplication of \(P_k\) with the idempotent L. For a function \(P_k \, L\) to be an eigenfunction of the maximal abelian subgroup \(\mathfrak {h}\), it must hold that \(L \, e_{2a-1}^\perp \, e_{2a}^\perp \) is again equal to L up to a (complex) constant. Consider, for \(a = 1, \ldots , n\), the Clifford elements
For the rest of this article, we will need the following notations. For \(F_a \in \left\{ L_a^\pm , M_a^\pm \right\} , a = 1,\ldots ,m\), denote
Furthermore, denote by \({\widetilde{F}}_a\) the idempotent
Then \(|{\widetilde{F}}_a|= 1 - |F_a|\) and \(\Vert {\widetilde{F}}_a \Vert = \Vert F_a \Vert \). Before we introduce the highest weight vectors, we will study the effect of multiplication by basis elements on these idempotents.
Lemma 2
The multiplication by \(e_a^\perp \) from the right on the idempotent \(F_a \in \left\{ L_a^\pm , M_a^\pm \right\} \) is:
As a result, for \(1 \leqslant a \leqslant n\), we have that
Denote, for \(1 \leqslant s_1 < s_2 \leqslant m\):
then we find that for \(1 \leqslant a < b \leqslant n\) and a general idempotent \(F = \prod _{s= 1}^m F_s\), with \(F_s \in \left\{ L_s^\pm , M_s^\pm \right\} \):
Proof
Note that
We may indeed summarize this as
From this it follows that
Hence we find that
Also important to note is that \(e^\perp _a \, e_a \, L_a^\pm = L_a^\pm \) and \(e^\perp _a\, e_a \, M^\pm _a = -M_a^\pm \) so for the idempotent \(F = \prod _{s = 1}^m F_s \), we find that
We thus get that
Analogously, we find that
Also
Finally
Consider the basic discrete monogenic k-homogeneous functions
From now on we denote \((k)'_+ = \left( k+ \frac{1}{2},\frac{1}{2},\ldots ,\frac{1}{2}\right) \) and \((k)'_- = \left( k+ \frac{1}{2},\frac{1}{2},\ldots ,-\frac{1}{2}\right) \). We will show under which conditions on the idempotent \(F = \prod _{s=1}^m F_s\), the space \(\text {span}_{{\mathbb {C}}} \left\{ g_k \, F \right\} \) is a weight space of \(\mathfrak {h}\) with weight \((k)'_+\) resp. \((k)'_-\).
Lemma 3
The polynomial \(g_k \, F \in {\mathcal {M}}_k, F = \prod _{s=1}^m F_s\) with \(F_s \in \left\{ L_s^\pm , M_s^\pm \right\} \), is a weight vector of \(\mathfrak {so}(m,{\mathbb {C}})\) with
-
weight \((k)'_+\) if \(k+ |F_1| + |F_2| + \Vert F_1 \Vert + \Vert F_2 \Vert \) is even and \(\Vert F_{2a-1}\Vert + \Vert F_{2a}\Vert + |F_{2a-1}| + |F_{2a}|\) is even for \(2 \leqslant a \leqslant n\).
-
weight \((k)'_-\) if \(k+ |F_1| + |F_2| + \Vert F_1 \Vert + \Vert F_2 \Vert \) is even, \(\Vert F_{2a-1}\Vert + \Vert F_{2a}\Vert + |F_{2a-1}| + |F_{2a}|\) is even for \(2 \leqslant a \leqslant n-1\) and \(\Vert F_{2n-1}\Vert + \Vert F_{2n}\Vert + |F_{2n-1}| + |F_{2n}|\) is odd.
Proof
We consider the action of the Cartan subalgebra-elements \(H_s, 1 \leqslant s \leqslant n\), on \(g_k\,F\). Since \(g_k\) only contains the vector variables \(\xi _1\) and \(\xi _2\), we will first consider \(H_1\):
We will also denote
In [6] it was established that \({\partial }_j \,g_k = (-1)^j \, k \, f_{k-1}\) hence
We thus get that
Notice that \(V_{12} \, g_k = (-1)^k \, g_k \, V_{12}\) since
Applying this, we find that
For \(g_k \, F\) to be a weight vector with weight \((k)'_+\) it must be an eigenfunction of \(H_1\) with eigenvalue \(k+ \frac{1}{2}\), hence it must hold that \(k+ |F_1| + |F_2| + \Vert F_1 \Vert + \Vert F_2 \Vert \) is even. We find 8 possible combinations for \(F_1\,F_2\):
-
k even:
$$\begin{aligned} F_1 \, F_2\in & {} \left\{ L_1^+\, L_2^+,\ L_1^- \, L_2^-, \ L_1^+\,M_2^+, \ L_1^- \,M_2^-, \right. \\&\quad \left. M_1^+\,L_2^+, \ M_1^+ \, M_2^+, \ M_1^- \, L_2^-, \ M_1^- \, M_2^- \right\} . \end{aligned}$$ -
k odd:
$$\begin{aligned}&F_1 \, F_2 \in \left\{ L_1^+\, L_2^-,\ L_1^- \, L_2^+, \ L_1^+\,M_2^-, \ L_1^- \,M_2^+, \ M_1^+\,L_2^-, \right. \\&\quad \left. M_1^+ \, M_2^-, \ M_1^- \, L_2^+, \ M_1^- \, M_2^+ \right\} . \end{aligned}$$
Next, we consider the action of \(H_a, 2 \leqslant a \leqslant n\), on \(g_k F\) to conclude under which conditions \(g_k\,F\) is an eigenfunction of \(H_a\) with eigenvalue \(\pm \frac{1}{2}\). Since the generator \(g_k\) only contains \(\xi _1\) and \(\xi _2\), it vanishes under the action of \(L_{2a-1,2a} = \xi _{2a-1}\,{\partial }_{2a} + \xi _{2a} \, {\partial }_{2a-1}\). Note that \(V_{2a-1,2a} \, g_k = g_k \, V_{2a-1,2a}\) since \(g_k\) contains only \(\mathbf{e}_1^\pm \) and \(\mathbf{e}_2^\pm \). Thus
This equals \(+\frac{1}{2} \, g_k \, F\) when \(\Vert F_{2a-1}\Vert + \Vert F_{2a}\Vert + |F_{2a-1}| + |F_{2a}|\) is even and \(-\frac{1}{2}\, g_k\,F\) otherwise.
We find that \(\Vert F_{2a-1}\Vert + \Vert F_{2a}\Vert + |F_{2a-1}| + |F_{2a}|\) is even for \(F_{2a-1} \, F_{2a}\) in
and odd for \(F_{2a-1} \, F_{2a} \) in
This proves the lemma.
Remark 1
In particular, we find that \(g_{2k} \, \prod _{s=1}^m L_s^+\) respectively \(g_{2k+1} \, L_1^+ \, L_2^-\, \prod _{s=3}^m L_s^+\) are weight vectors of \(\mathfrak {h}\) in \({\mathcal {M}}_{2k}\) resp. \({\mathcal {M}}_{2k+1}\) of weight \((2k)'_+\) resp. \((2k+1)'_+\).
Corollary 1
There are \(2^{2m-n}\) weight vectors \(g_k \, F\), with F one of the idempotents mentioned above, of weight \((k)'_+\) and \(2^{2m-n}\) weight vectors \(g_k \, F\), with F one of the above mentioned idempotents, with weight \((k)'_-\).
Proof
To obtain weight \((k)'_+\), one has eight choices for each factor \(F_{2s-1} \, F_{2s}\) in \(F, 1 \leqslant s \leqslant n\). We thus get \(8^{n} = 2^{3n} = 2^{4n-n} = 2^{2m-n}\) choices for the idempotent F. The same reasoning can be made for the weight \((k)'_-\).
We will now show that the weight vectors, defined in Lemma 3 are actually highest weight vectors, i.e. that they vanish under the action of all positive roots.
Theorem 1
The polynomials \(g_k \, F\), with
-
\(k+ |F_1| + |F_2| + \Vert F_1 \Vert + \Vert F_2 \Vert \) even
-
\(\Vert F_{2a-1}\Vert + \Vert F_{2a}\Vert + |F_{2a-1}| + |F_{2a}|\) even, for all \(2 \leqslant a \leqslant n-1\), and
-
\(\Vert F_{2n-1}\Vert + \Vert F_{2n}\Vert + |F_{2n-1}| + |F_{2n}|\) even resp. odd
are highest weight spaces with highest weight \((k)'_+\) resp. \((k)'_-\), i.e.
and
Proof
Lemma 3 tells us that these \(g_k \, F\) are weight vectors with weight \((k)'_\pm \) for the conditions stated above. The only thing that we still have to prove is that \(g_k\,F\) vanishes under the action of \(X_{a,b}\) and \(Y_{a,b}, a< b\).
We first consider the action of \(X_{a,b}\) on \(g_k\, F\). We make a distinction between \(a = 1\) and \(a \ne 1\). Take \(1 = a < b\), then
Now we use
Furthermore, since for \(b \ne 1,2\),
we find, for \(j \in \left\{ 2b-1, 2b\right\} \):
We get that
Using Lemma 2, we get
We thus see that this vanishes when
is even.
For \(1< a < b \leqslant n\) we get that
This will be zero when \( |F_{2a-1}| + \Vert F_{2a-1} \Vert + |F_{2a}| + \Vert F_{2a} \Vert \) is even, and this for all \(2 \leqslant a \leqslant n-1\).
Note that:
If we apply the appropriate change of sign in the second and last term of previous calculations, we immediately get that \(Y_{a,b}(g_k \, F) = 0\) for \(a < b\). Since \(Y_{a,b} = - Y_{b,a}\), this will also be zero for \(a > b\).
Remark 2
In particular, the polynomials \(g_{2k} \, \prod _{s=1}^m L_s^+\) and \(g_{2k+1} \, L_1^+ \, L_2^- \, \prod _{s=3}^m L_s^+\) are highest weight vectors with weight \((2k)'_+\) resp. \((2k+1)'_+\).
Remark 3
The dimension of \(\left( k\right) '_\pm \) is (see e.g. [17])
As the dimension of \({\mathcal {M}}_k\) equals
and as we found \(2^{3n}\) disjoint isomorphic copies of \(\left( k\right) '_+\) combined with \(2^{3n}\) copies of \(\left( k\right) '_-\), the space \({\mathcal {M}}_k\) is fully decomposed in \(2^{3n}\) copies of \(\left( k\right) '_+\) and \(2^{3n}\) copies of \(\left( k \right) '_-\).
Definition 2
We define the space of positive resp. negative spinors \({\mathbb {S}}_{2n}^\pm \) as the image under \(\mathfrak {so}(m,{\mathbb {C}})\) of the idempotents \(\prod _{s=1}^m L_s^+\), resp. \(\left( \prod _{s=1}^{m-1} L_s^+ \right) L_m^-\):
and
The elements \(L_1^+ \,L_2^+ \ldots \, L_{2n-1}^+ \, L_{2n}^+\), resp. \(L_1^+ \,L_2^+ \ldots \, L_{2n-1}^+ \, L_{2n}^-\) are highest weight vectors with weight \((0)'_+ = \left( \frac{1}{2}, \ldots , \frac{1}{2}\right) \) resp. \((0)'_- = \left( \frac{1}{2}, \ldots , \frac{1}{2},-\frac{1}{2}\right) \). Let us give an example to clarify this construction.
Example 1
Let \(m = 4\) (i.e. \(n=2\)) and consider \(L = L_1^+ \,L_2^+ \, L_3^+ \, L_4^+\). The Lie algebra \(\mathfrak {so}(4,{\mathbb {C}})\) is given in this context by
The elements \(dR(e_{12})\) and \(dR(e_{34})\) return L up to complex constant. The other four rotations give us (up to a complex constant) the idempotent \(L_1^+\,L_2^- \, L_3^-\, L_4^+\). Hence
Starting from \(L_1^+ \,L_2^+ \, L_3^+ \, L_4^-\), the rotations \(dR(e_{13}), dR(e_{14}), dR(e_{23})\) and \(dR(e_{24})\) lead us to the idempotent \(L_1^+ \,L_2^- \, L_3^- \, L_4^-\) which shows that
The space of positive/negative spinors \({\mathbb {S}}^\pm _4\) is 2-dimensional.
In general, the elements \(dR(e_{2a-1,2a})\) acting on an idempotent return the same idempotent up to a multiplicative complex factor. Since, for \(1 \leqslant a < b \leqslant n\):
with \(L^{2a,2b-1} = L_1^+ \, L_2^+ \, \ldots L_{2a-1}^+ \, L_{2a}^- \ldots L_{2b-1}^- \, L_{2b}^+ \ldots L_{2n-1}^+ \, L_{2n}^+\), similar to the notation used before, we see that \(dR(e_{a,b})\) acting on
changes the sign of an even number of \(L_a\)’s. The operator always leaves \(L_1^+\) and \(L_{2n}^+\) unchanged. The resulting idempotent will always have an even number of minus-signs. Starting from the idempotent L with all plus-signs, we thus get all possible idempotents of the following form:
where each place \(\; \underbrace{ \, . \quad . \, } \; \) consists of either \(L_{2a}^+ \, L_{2a+1}^+\) or \(L_{2a}^- \, L_{2a+1}^-, 1 \leqslant a \leqslant n-1\). We get \(2^{n-1}\) spinors belonging to the positive spinorspace and we have the following weight space decomposition
where \(V_{\left( \pm \frac{1}{2}, \pm \frac{1}{2},\ldots , \pm \frac{1}{2}\right) }\) denotes the weight space with weight \(\left( \pm \frac{1}{2}, \pm \frac{1}{2},\ldots , \pm \frac{1}{2}\right) \) and where the sum goes over all weights with an even number of minus-signs. The highest weight remains \(\left( \frac{1}{2}, \ldots , \frac{1}{2}\right) \) and the highest weight vector is L.
Starting from \(L_1^+ \, L_2^+ \ldots L_{2n-1}^+ \, L_{2n}^-\), we will generate all possible idempotents of the following form:
where each place \(\; \underbrace{ \, . \quad . \, } \; \) consists of either \(L_{2a}^+ \, L_{2a+1}^+\) or \(L_{2a}^- \, L_{2a+1}^-, 1 \leqslant a \leqslant n-1\). We thus also get \(2^{n-1}\) spinors belonging to the negative spinorspace and the following weight space decomposition:
where the sum goes over all weights with an odd number of minus-signs. The highest weight is still \(\left( \frac{1}{2}, \ldots , \frac{1}{2},-\frac{1}{2}\right) \) and the highest weight vector is \(L_1^+ \, L_2^+ \ldots L_{2n-1}^+ \, L_{2n}^-\).
4.2 Odd Dimension \(m=2n+1\)
We now extend the set of generators \(H_a, X_{a,b}, Y_{a,b}\) and \(Z_{a,b}\) of \(\mathfrak {so}(m,{\mathbb {C}})\) with 2n operators
where \(1 \leqslant a \leqslant n\). With the addition of these 2n operators, we are again able to reconstruct all original \(dR(e_{a,b})\)’s since \(\sqrt{2} \; dR(e_{2a-1,m}) = U_a + V_a\) and \(-\sqrt{2} \,i \, dR(e_{2a,m}) = U_a - V_a\).
The classic commutator relations follow immediately.
Lemma 4
For \(1 \leqslant a, b \leqslant n\), it holds that
In particular, \(U_b\) is a root vector of \(\mathfrak {so}(m, {\mathbb {C}})\) corresponding to the positive root \(L_b\) and \(V_b\) is a root vector corresponding with the negative root \(-L_b\), for all \(1 \leqslant b \leqslant n\).
Lemma 5
The operators \(U_c\) and \(V_d\) satisfy the following additional commutator relations with \(X_{a,b}, Y_{a,b}\) and \(Z_{a,b}, 1 \leqslant a, b, c,d \leqslant n\):
Proof
The statements follow immediately from the definitions of \(U_c\) and \(V_d\) and from the defining relations (2) which the operators \(dR(e_{a,b})\) satisfy.
We introduce four extra idempotents
and denote
We will now show that the highest weight vectors of weight \((k)'_+\) from the even-dimensional setting are also highest weight vectors with weight \((k)'_+\) in the odd-dimensional case when we multiply them with one of the four possible extra factors \(L_m^\pm \) and \(M_m^\pm \).
Theorem 2
The weight vectors \(g_k \, F, F = \prod _{s=1}^m F_s\) with \(F_s \in \left\{ L_s^\pm , M_s^\pm \right\} \) with
-
\(k+ |F_1| + |F_2| + \Vert F_1 \Vert + \Vert F_2 \Vert \) even
-
\(\Vert F_{2a-1}\Vert + \Vert F_{2a}\Vert + |F_{2a-1}| + |F_{2a}|\) even, for all \(2 \leqslant a \leqslant n\),
vanish under the positive root vectors \(U_a, 1 \leqslant a \leqslant n\), i.e. \(U_a \left( g_k \,F \right) = 0\), for all \(1 \leqslant a \leqslant n\).
Proof
Consider
Since \(g_k\) contains only the vector variables \(\xi _1\) and \(\xi _2\), we will make a distinction between \(a = 1\) and \(a \ne 1\). We start with assuming that \(a=1\):
Now we again use that, for \(j \ne 1,2\):
Hence
We complete the proof by noting that
Thus \(\sqrt{2}\,U_1(g_k \, F)\) will be zero since \(k+ |F_1| + |F_2| + \Vert F_1 \Vert + \Vert F_2 \Vert \) is even.
When \(a \ne 1\), the action of \(L_{2a-1,m}\) on \(g_k\, F\) results in zero hence
This will be zero when \(\Vert F_{2a-1}\Vert + \Vert F_{2a}\Vert + |F_{2a-1}| + |F_{2a}|\) is even, for all \( 2 \leqslant a \leqslant n\).
Corollary 2
The polynomials \(g_k \, F \in {\mathcal {M}}_k\) with \(F = \prod _{s=1}^m F_s, F_s \in \left\{ L_s^\pm , M_s^\pm \right\} \) such that
-
\(k+ |F_1| + |F_2| + \Vert F_1 \Vert + \Vert F_2 \Vert \) is even
-
\(\Vert F_{2a-1}\Vert + \Vert F_{2a}\Vert + |F_{2a-1}| + |F_{2a}|\) is even, for all \(2 \leqslant a \leqslant n\),
are highest weight vectors of weight \((k)'_+\). In particular, \(g_{2k} \,L\) and \(g_{2k+1}\,L'\) are highest weight vectors of weight \((2k)'_+\) resp. \((2k+1)'_+\).
Note that the particular choice for the last factor \(F_m \in \left\{ L_m^\pm , M_m^\pm \right\} \) does not affect the results.
We again count how many distinct highest weight vectors \(g_k\,F, F = \prod _{s=1}^m F_s\), with weight \((k)'_+\) we find: for each \(F_{2a-1}\,F_{2a}, 1 \leqslant a \leqslant n\), we have 8 possible combinations, namely
and for \(F_m\) we have four possible choices \(L_m^\pm , M_m^\pm \). Combining this, we find \(8^n \, 2^2 = 2^{3n+2} = 2^{2m-n}\) isomorphic irreducible representations with highest weight \((k)'_+\), each of which has dimension
Hence the total dimension of all isomorphic irreducible representations is
i.e. the dimensional analysis shows that \({\mathcal {M}}_k\) may be decomposed as \(2^{3n+2}\) isomorphic irreducible representations with highest weight \((k)'_+\).
Definition 3
We define the spinor space \({\mathbb {S}}_{2n+1}\) as the image under \(\mathfrak {so}(m,{\mathbb {C}})\) of the idempotent \(L = \prod _{s=1}^m L_s^+\):
The element \(L_1^+ \,L_2^+ \ldots \, L_{2n}^+ \, L_{2n+1}^+\) is a highest weight vector with weight \((0)'_+ = \left( \frac{1}{2}, \ldots , \frac{1}{2}\right) \) and thus generates an irreducible representation with the same weight.
Example 2
Let \(m = 5\) (i.e. \(n=2\)) and consider \(L = L_1^+ \,L_2^+ \, L_3^+ \, L_4^+ \, L_5^+\). We denote \(dR(e_{a,b})\) in short as (a, b). The Lie algebra \(\mathfrak {so}(5,{\mathbb {C}})\) is given in this context by the span over \({\mathbb {C}}\) of the ten elements (1, 2), (1, 3), (1, 4), (1, 5), (2, 3), (2, 4), (2, 5), (3, 4), (3, 5) and (4, 5). The idempotents involved interact in the following way under the action of \(\mathfrak {so}(m,{\mathbb {C}})\):
Hence
The spinorspace \({\mathbb {S}}_{5}\) is \(2^{2}\)-dimensional as expected.
In general, the rotations \(dR(e_{2a-1,2a}), a= 1,\ldots ,n\), acting on an idempotent return the same idempotent up to a multiplicative complex number. Again, we find that \(dR(e_{a,b})\) with \(1 \leqslant a, b \leqslant n\), changes the sign of an even number of \(L_i\)’s. The additional rotations \(dR(e_{2a-1,m})\) and \(dR(e_{2a,m})\), with \(1 \leqslant a \leqslant n\), act as follows on L:
The rotation always leaves \(L_1^+\) invariant. The resulting idempotent always will have an even number of minus-signs. Starting from the idempotent L with all plus-signs, we thus get all possible idempotents of the following form:
where each place \(\; \underbrace{ \, . \quad . \, } \; \) consists of either \(L_{2a}^+ \, L_{2a+1}^+\) or \(L_{2a}^- \, L_{2a+1}^-, 1 \leqslant a \leqslant n\). We thus get \(2^{n}\) spinors and we have the following weight space decomposition
where the sum goes over all weights with an even number of minus-signs.
Example 3
Let \(m = 7\) (i.e. \(n=3\)) and consider \(L = L_1^+ \,L_2^+ \, L_3^+ \, L_4^+ \, L_5^+\, L_6^+\,L_7^+\). We will again denote \(dR(e_{a,b})\) in short as (a, b). The Lie algebra \(\mathfrak {so}(5,{\mathbb {C}})\) is given in this setting by 21 elements and the corresponding spinorspace will be 8-dimensional. The idempotents involved interact in the following way under the action of \(\mathfrak {so}(m,{\mathbb {C}})\):
Hence
We indeed find an 8-dimensional spinorspace \({\mathbb {S}}_{7}\).
5 Conclusion and Future Research
The space \({\mathcal {M}}_k\) of discrete k-homogeneous monogenic polynomials is a reducible representation of \(\mathfrak {so}(m,{\mathbb {C}})\) which can, in the odd-dimensional case \(m = 2n+1\), be decomposed into \(2^{2m-n}\) isomorphic copies of the irreducible \(\mathfrak {so}(m,{\mathbb {C}})\)-representation with highest weight \((k)'_+\) and in the even-dimensional setting \(m =2n\), we find \(2^{2m-n}\) isomorphic irreducible representations with highest weight \((k)'_+\) and \(2^{2m-n}\) irreps of highest weight \((k)'_-\). This is done by means of an appropriate amount of idempotents.
Let \(g_k = \left( \xi _2 - \xi _1 \right) \left( \xi _2 + \xi _1 \right) \left( \xi _2 - \xi _1 \right) \left( \xi _2 + \xi _1 \right) \ldots [1], (k \text { factors})\), be a discrete homogeneous monogenic function of degree k and let
Denote \(\Vert L^\pm _a\Vert = 0, \Vert M^\pm _a \Vert = 1, | L^+_a | = |M_a^- | = 0\) and \(|L^-_a| = |M_a^+| = 1\).
In even dimension \(m = 2n\), the polynomial \(g_k \, F \in {\mathcal {M}}_k, F = \prod _{s=1}^m F_s\) with \(F_s \in \left\{ L_s^\pm , M_s^\pm \right\} \), is a highest weight vector of \(\mathfrak {so}(m,{\mathbb {C}})\) with
-
weight \((k)'_+\) when \(k+ |F_1| + |F_2| + \Vert F_1 \Vert + \Vert F_2 \Vert \) is even and \(\Vert F_{2a-1}\Vert + \Vert F_{2a}\Vert + |F_{2a-1}| + |F_{2a}|\) is even for \(2 \leqslant a \leqslant n\).
-
weight \((k)'_-\) when \(k+ |F_1| + |F_2| + \Vert F_1 \Vert + \Vert F_2 \Vert \) is even, \(\Vert F_{2a-1}\Vert + \Vert F_{2a}\Vert + |F_{2a-1}| + |F_{2a}|\) is even for \(2 \leqslant a \leqslant n-1\) and \(\Vert F_{2n-1}\Vert + \Vert F_{2n}\Vert + |F_{2n-1}| + |F_{2n}|\) is odd.
We find \(2^{2m-n}\) highest weight vectors in \({\mathcal {M}}_{k}\) with weight \(\left( k \right) '_+\) and \(2^{2m-n}\) weight vectors, with weight \((k)'_-\), each generating an irreducible \(\mathfrak {so}(m,{\mathbb {C}})\)-representation.
In odd dimensions \(m=2n+1\), the polynomial \(g_k \, F \in {\mathcal {M}}_k, F = \prod _{s=1}^m F_s\) with \(F_s \in \left\{ L_s^\pm , M_s^\pm \right\} \), is a weight vector of \(\mathfrak {so}(m,{\mathbb {C}})\) with weight \((k)'_+\) when \(k+ |F_1| + |F_2| + \Vert F_1 \Vert + \Vert F_2 \Vert \) is even and \(\Vert F_{2a-1}\Vert + \Vert F_{2a}\Vert + |F_{2a-1}| + |F_{2a}|\) is even for \(2 \leqslant a \leqslant n\). We find \(2^{2m-n}\) highest weight vectors in \({\mathcal {M}}_{k}\) with weight \(\left( k \right) '_+\) and thus as much irreps.
We have shown throughout this article how the space \({\mathcal {M}}_k\) of monogenic discrete k-homogeneous polynomials can be decomposed into irreducible representations of \(\mathfrak {so}(m,{\mathbb {C}})\). However, because of the presence of the basiselements \(e_a\) and \(e_a^\perp \) in the definition of the generators of the rotations, the spinorspace is no maximal left ideal. In future research we will investigate other possibilities to define rotations and the spinorspace in the hopes of writing the spinorspace as maximal left ideal. An equivalent description of \({\mathcal {H}}_k\) and \({\mathcal {M}}_k\) as SO(m) / Spin(m)-representations is also still work in progress.
References
Baaske, F., Bernstein, S., De Ridder, H., Sommen, F.: On solutions of a discretized heat equation in discrete Clifford analysis. J. Diff. Eq. Appl. 20(2), 271–295 (2014)
Brackx, F., Delanghe, R., Sommen, F.: Clifford analysis. Research Notes in Mathematics 76, (1982), Pitman Books Limited
Constales, D., Sommen, F., Van Lancker, P.: Models for irreducible representations of spin\((m)\). Adv. Appl. Clifford Algebr. 11(S1), 271–289 (2001)
De Bie, H., De Ridder, H., Sommen, F.: Discrete Clifford analysis: the one dimensional setting. Complex Var. Elliptic Eq. Int. J. 57(7–8), 903–920 (2012)
Delanghe, R., Sommen, F., Soucek, V.: Clifford Algebra and Spinor Valued Functions: A Function Theory for the Dirac Operator. Mathematics and its Applications, vol. 53. Kluwer Academic, Dordrecht (1992)
De Ridder, H.: Discrete Clifford analysis: function theory and integral transforms. Ghent University. Faculty of Sciences, Ghent, Belgium (2013)
De Ridder, H., De Schepper, H., Kähler, U., Sommen, F.: Discrete function theory based on skew Weyl relations. Proc. Am. Math. Soc. 138, 3241–3256 (2010)
De Ridder, H., De Schepper, H., Sommen, F.: Fueter polynomials in discrete Clifford analysis. Math. Zeit. 272(1–2), 253–268 (2012)
De Ridder, H., Raeymaekers, T.: Models for some irreducible representations of \({\mathfrak{so}}(m,\mathbb{C})\) in discrete Clifford analysis. In: Proceedings of the Session “Quaternion and Clifford Analysis” of the 10th ISAAC Conference Macau, Modern Trends in Hypercomplex Analysis, Trends in Mathematics, Birkhauser, Basel (to appear)
De Ridder, H., Raeymaekers, T., Sommen, F.: Rotations in discrete Clifford analysis. Appl. Math. Comput. 285, 114–140 (2016)
De Ridder, H., De Schepper, H., Sommen, F.: Taylor series expansion in discrete Clifford analysis. Complex Anal. Oper. Theory 8(2), 485–511 (2013)
De Ridder, H., De Schepper, H., Sommen, F.: The Cauchy–Kovalevskaya extension theorem in discrete Clifford analysis. Commun. Pure Appl. Anal. 10(4), 1097–1109 (2011)
De Schepper, H., Sommen, F., Van de Voorde, L.: A basic framework for discrete Clifford analysis. Exp. Math. 18(4), 385–395 (2009)
Dirac, P.A.M.: The quantum theory of the electron. In: Proceedings of the Royal Society of London , vol. 117 (1928)
Faustino, N.: Discrete Clifford analysis, Ph.D. thesis, Universidade de Aveiro, Aveiro, (2009)
Faustino, N., Kähler, U., Sommen, F.: Discrete Dirac operators in Clifford analysis. Adv. Appl. Clifford Algebr. 17(3), 451–467 (2007)
Fulton, W., Harris, J.: Representation theory, Graduate Texts in Mathematics 129, A first course, Readings in Mathematics. Springer, New York (1991)
Gilbert, J., Murray, M.A.M.: Clifford Algebras and Dirac Operators in Harmonic Analysis. Cambridge University Press, Cambridge (1991)
Gürlebeck, K., Hommel, A.: On finite difference potentials and their applications in a discrete function theory. Math. Meth. Appl. Sci. 25, 1563–1576 (2002)
Kenyon, R.: The Laplacian and Dirac operators on critical planar graphs. Invent. Math. 150(2), 409–439 (2002)
Knapp, A.W.: Lie Groups Beyond an Introduction, Progress in Mathematics, vol. 140. Birkhäuser Boston Inc., Boston (1996)
Sommen, F.: Functions on the spingroup. Adv. Appl. Clifford Algebr. 6(1), 37–48 (1996)
Stein, E., Weiss, G.: Generalization of the Cauchy–Riemann equations and representations of the rotation group. Am. J. Math. 90(1), 163–196 (1968)
Weinberg, S.: The Quantum Theory of Fields, Repr. with corr. Cambridge university press, Cambridge (1998)
Acknowledgements
The first author acknowledges the support of the Research Foundation—Flanders (FWO), Grant No. FWO13\(\backslash \)PDO\(\backslash \)039.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Uwe Kähler.
Rights and permissions
About this article
Cite this article
De Ridder, H., Raeymaekers, T. Spinor Spaces in Discrete Clifford Analysis. Complex Anal. Oper. Theory 11, 1113–1137 (2017). https://doi.org/10.1007/s11785-017-0644-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11785-017-0644-x