Geometric algebra and the Clifford star product

Date: 2015-10-07; view: 486.

Starting point for geometric algebra [1] and [3] is an n-dimensional vector space over the real numbers with vectors a,b,c, … A multiplication, called geometric product, of vectors can then be denoted by juxtaposition of an indeterminate number of vectors so that one gets monomials A, B, C, … These monomials can be added in a commutative and associative manner: A + B = B + A and (A + B) + C = A + (B + C), so that they form polynomials also denoted by capital letters. The so obtained polynomials can be multiplied associatively, i.e., A (BC) = (AB) C and they fulfill the distributive laws (A + B) C = AC + BC and C (A + B) = CA + CB. Furthermore, there exists a null vector a0 = 0 and the multiplication with a scalar λa = aλ, with . The connection between scalars and vectors can be given if one assumes that the product ab is a scalar iff a and b are collinear, so that is the length of the vector a. These axioms define now the Clifford algebra Cℓ (V) and the elements A, B, C, … of Cℓ (V) are called Clifford or c-numbers.

Since the geometric product of two collinear vectors is a scalar, the symmetric part of the geometric product is a scalar denoted . The product a · b is the inner or scalar product. One can then decompose the geometric product into its symmetric and antisymmetric part:

(3.1)

where the antisymmetric part is formed with the outer product. For the outer product one has obviously a b = −b a and a a = 0, so that a b can be interpreted geometrically as an oriented area. The geometric product is constructed in such a way that it gives information over the relative directions of a and b, i.e., ab = ba = a · b a b = 0 means that a and b are collinear whereas ab = −ba = a b a · b = 0 means that a and b are perpendicular.

With the outer product one defines simple r-vectors or r-blades

Ar=a1 a2 ar,

(3.2)

which can be interpreted as r-dimensional volume forms. The geometric product can then be generalized to the case of a vector and an r-blade:

aAr=a·Ar+a Ar,

(3.3)

which is the sum of an (r−1)-blade and a (r + 1)-blade . Applying this recursively one sees that each c-number can be written as a polynomial of r-blades, and using a set of basis vectors e₁, e₂, … , e_r a c-number reads:

(3.4)

A is called a multivector or r-vector if the highest appearing grade is r. It decomposes into several blades:

(3.5)

where _n projects onto the term of grade n. A multivector A_r is called homogeneous if all appearing blades have the same grade, i.e., A_r = A_{r r}. The geometric product of two homogeneous multivectors A_r and B_s can be written as

ArBs= ArBs r+s+ ArBs r+s-2+ + ArBs |r-s|.

(3.6)

The inner and the outer product stand now for the terms with the lowest and the highest grade:

(3.7)

One should note that the inner and outer product here in the general case do not correspond anymore to the symmetric and the antisymmetric part of the geometric product. For example, in the case of two bivectors one has A₂ B₂ = B₂ A₂, so that the outer product is symmetric. Actually one finds for the symmetric and the antisymmetric parts of A₂B₂:

(3.8)

In general the commutativity of the outer and the inner product is given by:

(3.9)

and both products are always distributive:

(3.10)

Only the outer product of r-vectors is in general associative, i.e., A (B C) = (A B) C, for the inner product one gets:

(3.11)

If one has to calculate several products of different type, the inner and the outer product always have to be calculated first, i.e.,

(3.12)

The formalism of geometric algebra briefly sketched so far can now be described with Grassmann variables and the Clifford star product, that turns the Grassmann algebra into a Clifford algebra. To make the equivalence even more obvious we go over to the dimensionless Grassmann variables

(3.13)

These variables play here the role of dimensionless basis vectors and will therefore be written in bold face, whereas the θ_i played in the discussion of the first section the role of dynamical variables with dimension . In the σ_n-variables the Clifford star product (2.12) has the form

(3.14)

As a star product the Clifford star product is associative and distributive.

To show how the geometric algebra described with Grassmann variables and the Clifford star product looks like, we first consider the two-dimensional euclidian case. One has then two Grassmann basis elements σ₁ and σ₂, so that a general element of the Clifford algebra is a supernumber A = a₀ + a₁σ₁ + a₂σ₂ + a₁₂σ₁ σ₂ = A ₀ + A ₁ + A ₂ and a vector corresponds to a supernumber with Grassmann grade one: a = a₁σ₁ + a₂σ₂. The Clifford star product of two of these supernumbers is

(3.15)

where the symmetric and the antisymmetric part of the Clifford star product is given by:

(3.16)

and

(3.17)

which are terms with Grassmann grade 0 and 2, respectively. Note that now a juxtaposition like ab is just as in the notation of superanalysis the product of supernumbers and not the Clifford product, which we want to describe explicitly with the star product (3.14). The σ_i form an orthogonal basis under the scalar product: .

The unit 2-blade i = σ₁σ₂ can be interpreted as the generator of -rotations because by multiplying from the right one gets

(3.18)

so that a vector x = x₁σ₁ + x₂σ₂ is transformed into . The relation describes then a reflection and furthermore one has with (2.16): , so that i corresponds to the imaginary unit. The connection between the two-dimensional vector space with vectors x and the Gauss plane with complex numbers z is established by star multiplying x with σ₁:

(3.19)

Such a bivector that results from star multiplying two vectors is also called spinor. While the bivector i generates a rotation of when acting from the right, the spinor z generates a general combination of a rotation and dilation when acting from the right. One can see this by writing with . Acting from the right with z causes then a dilation by |z| and a rotation by , one has for example: σ_{1 C}z=x, which is the inversion of (3.19). Here one can see that the formalism of geometric algebra reproduces complex analysis and gives it a geometric meaning.

After having described the geometric algebra of the euclidian 2-space we now turn to the euclidian 3-space with basis vectors σ₁, σ₂, and σ₃ and with the Clifford star product (3.14) for d = 3. The basis vectors are orthogonal: σ_i · σ_j = δ_ij and a general c-number written as a supernumber has the form

This multivector has now four different simple multivector parts. Besides the scalar part a₀ there is the pseudoscalar part corresponding to I₃ = σ₁σ₂σ₃, which can be interpreted as a right handed volume form, because a parity operation gives (−σ₁) (−σ₂) (−σ₃) = −I₃. Moreover I₃ has also the properties of an imaginary unit: and I_{3 C}I₃ = I₃ · I₃ = −1. While the pseudoscalar I₃ is an oriented volume element the bivector part with the basic 2-blades

(3.21)

describes oriented area elements. Each of the i_r plays in the plane it defines the same role as the i of the two-dimensional euclidian plane defined above. Star-multiplying with the pseudoscalar I₃ is equivalent to taking the Hodge dual, for example to each bivector B = b₁i₁ + b₂i₂ + b₃i₃ corresponds a vector b = b₁σ₁ + b₂σ₂ + b₃σ₃, which can be expressed by the equation B = I_{3 C}b. This duality can for example be used to write the geometric product of two vectors a = a₁σ₁ + a₂σ₂ + a₃σ₃ and b = b₁σ₁ + b₂σ₂ + b₃σ₃ as:

a Cb=a·b+I3 C(a×b),

(3.22)

where and a×b=ε^klma_kb_lσ_m. Furthermore one finds:

σ1×σ2=-I3 Cσ1 Cσ2=-I3 Cσ1σ2=σ3

(3.23)

and cyclic permutations. Note also that one gets with the nabla operator _x=σ₁∂_x1+σ₂∂_x2+σ₃∂_x3 for the gradient of a vector field f = f₁ (x₁, x₂, x₃) σ₁ + f₂ (x₁, x₂, x₃) σ₂ + f₃ (x₁, x₂, x₃) σ₃:

x Cf= x·f+ x f=divf+I3 Crotf.

(3.24)

The multivector part of (3.20) with even Grassmann grade have the basis 1, i₁, i₂, i₃ and form a closed subalgebra under the Clifford star product, namely the quaternion algebra. The multivector part of (3.20) with odd grade does not close under the Clifford star product, but nevertheless one can reinvestigate the definition of the Pauli functions in (2.14). Replacing in (2.14) the scalar i by the pseudoscalar I₃ one sees that the basis vectors σ_i fulfill

(3.25)

which justifies denoting them σ_i. With the pseudoscalar I₃ the trace (2.18) can be written as Tr (F) = 2∫dσ₃ dσ₂dσ₁ F = 2∫dσ₃ dσ₂ dσ₁I_{3 C}F. So one has here achieved with the Clifford star product a cliffordization of the three-dimensional Grassmann algebra of the σ_i.

Just as in the two-dimensional case one can also consider in three dimensions the role of spinors and rotations. To this purpose one first considers a vector transformation of the form

x→x′=-u Cx Cu,

(3.26)

where u is a three-dimensional unit vector: u = u₁σ₁ + u₂σ₂ + u₃σ₃ with . This transformation can be identified as a reflection in a plane orthogonal to u if one decomposes x into a part collinear to u and a part orthogonal to u:

x=x +x =(x·u+xu) Cu,

(3.27)

with x = (x · u) u and x = (xu) _Cu = (xu) · u. One can easily check that

(3.28)

This decomposition of x can most easily be obtained if one just star-divides x _Cu = x · u + x u by u, which gives with u^-1 _C=u:

x=(x·u) Cu-1 C+(xu) Cu-1 C=(x·u)u+(xu) Cu=x +x .

(3.29)

Using (3.28) one sees that the transformation (3.26) turns x into x^′=-u _Cx _Cu=-x +x , so that only the component collinear to u is inverted, which amounts to a reflection at the plane where u is the normal vector. Two successive transformations (3.26) lead to:

(3.30)

where U can be written as:

(3.31)

where the angle between the unit vectors u and v is described by an bivector A = v u = vu = |vu|A₀. Hereby the unit bivector A₀ = vu/|vu| defines the plane in which the angle lies, while the magnitude |vu| gives the angle in radians, furthermore it fulfills A_{0 C}A₀ = −1. If one chooses for example the basis vectors σ_k for u and v, A₀ would be given by one of the bivectors in (3.21). The additional factor 1/2 in (3.31) will become clear if one investigates the action of the transformation (3.30). To this purpose one proceeds analogously to the discussion of the reflection (3.26). One first decomposes the vector x into a part x in the plane defined by A and a part x perpendicular to that plane. This is done analogously to (3.29) by star-dividing x _CA = x · A + x A by A which leads to

x=(x·A) CA-1 C+(xA) CA-1 C=x +x ,

(3.32)

with x _C A = −A _Cx and x _CA = A _Cx . We then have for the transformation (3.30):

(3.33)

So the component perpendicular to the plane defined by A is not changed while the component in this plane is rotated with the help of the spinor by an angle of magnitude |A|, just as described in the two-dimensional case above. One sees here why the rotation in the two-dimensional case could be written just by acting with a spinor from the right. This is due to the fact that when the vector lies in the plane of rotation one has

(3.34)

A rotation can be described with the bivector A, but also with the dual vector a defined by A = I_{3 C}a, where the direction of a defines the axis of rotation, while the magnitude gives the radian |a| = |A|. So U can also be written as:

(3.35)

which corresponds to the star exponential (2.24).

The formalism described so far can easily be generalized to the case of d euclidian dimensions. Just as there is a duality in the space spanned by the σ_i there is also the duality between the spaces spanned by the σ_i and the σⁱ. This duality is expressed by the relation . The σⁱ-vectors can be constructed with the help of the pseudoscalar, which is for the d-dimensional euclidian case I_d = σ₁σ₂ σ_d. The hyperplane for which the basis vector σ_j is normal is given for an d-dimensional euclidian space by the (d−1)-blade , where means that this basis vector is missing. The corresponding dual vector is then given by

(3.36)

where is the inverse d-dimensional pseudoscalar.

Note also that the multiple Clifford star product leads to an expansion of Wick type. For example, the Clifford product of four basis vectors is given by

σi1 Cσi2 Cσi3 Cσi4=σi1σi2σi3σi4+σi1σi2δi3i4-σi1σi3δi2i4+σi1σi4δi2i3+σi2σi3δi1i4-σi2σi4δi1i3+σi3σi4δi1i2+δi1i2δi3i4-δi1i3δi2i4+δi1i4δi2i3,

(3.37)

where the contraction of σ_i and σ_j is given by δ_ij. This suggests to use the star product formalism also in the realm of quantum field theory [10].