"On the History of Unified Field Theories. Part II. (ca. 1930 – ca. 1965)"
Hubert F. M. Goenner 
1 Introduction
2 Mathematical Preliminaries
2.1 Metrical structure
2.2 Symmetries
2.3 Affine geometry
2.4 Differential forms
2.5 Classification of geometries
2.6 Number fields
3 Interlude: Meanderings – UFT in the late 1930s and the 1940s
3.1 Projective and conformal relativity theory
3.2 Continued studies of Kaluza–Klein theory in Princeton, and elsewhere
3.3 Non-local fields
4 Unified Field Theory and Quantum Mechanics
4.1 The impact of Schrödinger’s and Dirac’s equations
4.2 Other approaches
4.3 Wave geometry
5 Born–Infeld Theory
6 Affine Geometry: Schrödinger as an Ardent Player
6.1 A unitary theory of physical fields
6.2 Semi-symmetric connection
7 Mixed Geometry: Einstein’s New Attempt
7.1 Formal and physical motivation
7.2 Einstein 1945
7.3 Einstein–Straus 1946 and the weak field equations
8 Schrödinger II: Arbitrary Affine Connection
8.1 Schrödinger’s debacle
8.2 Recovery
8.3 First exact solutions
9 Einstein II: From 1948 on
9.1 A period of undecidedness (1949/50)
9.2 Einstein 1950
9.3 Einstein 1953
9.4 Einstein 1954/55
9.5 Reactions to Einstein–Kaufman
9.6 More exact solutions
9.7 Interpretative problems
9.8 The role of additional symmetries
10 Einstein–Schrödinger Theory in Paris
10.1 Marie-Antoinette Tonnelat and Einstein’s Unified Field Theory
10.2 Tonnelat’s research on UFT in 1946 – 1952
10.3 Some further developments
10.4 Further work on unified field theory around M.-A. Tonnelat
10.5 Research by and around André Lichnerowicz
11 Higher-Dimensional Theories Generalizing Kaluza’s
11.1 5-dimensional theories: Jordan–Thiry theory
11.2 6- and 8-dimensional theories
12 Further Contributions from the United States
12.1 Eisenhart in Princeton
12.2 Hlavatý at Indiana University
12.3 Other contributions
13 Research in other English Speaking Countries
13.1 England and elsewhere
13.2 Australia
13.3 India
14 Additional Contributions from Japan
15 Research in Italy
15.1 Introduction
15.2 Approximative study of field equations
15.3 Equations of motion for point particles
16 The Move Away from Einstein–Schrödinger Theory and UFT
16.1 Theories of gravitation and electricity in Minkowski space
16.2 Linear theory and quantization
16.3 Linear theory and spin-1/2-particles
16.4 Quantization of Einstein–Schrödinger theory?
17 Alternative Geometries
17.1 Lyra geometry
17.2 Finsler geometry and unified field theory
18 Mutual Influence and Interaction of Research Groups
18.1 Sociology of science
18.2 After 1945: an international research effort
19 On the Conceptual and Methodic Structure of Unified Field Theory
19.1 General issues
19.2 Observations on psychological and philosophical positions
20 Concluding Comment

2 Mathematical Preliminaries

For the convenience of the reader, some of the mathematical formalism given in the first part of this review is repeated in a slightly extended form: It is complemented by further special material needed for an understanding of papers to be described.

2.1 Metrical structure

First, a definition of the distance ds between two infinitesimally close points on a D-dimensional differential manifold M D is to be given, eventually corresponding to temporal and spatial distances in the external world. For ds, positivity, symmetry in the two points, and the validity of the triangle equation are needed. We assume ds to be homogeneous of degree one in the coordinate differentials dxi connecting neighboring points. This condition is not very restrictive; it includes Finsler geometry [510, 199, 394, 4*] to be briefly discussed in Section 17.2.

In the following, ds is linked to a non-degenerate bilinear form g (X, Y ), called the first fundamental form; the corresponding quadratic form defines a tensor field, the metrical tensor, with 2 D components gij such that

∘ -----i--j ds = gijdx dx , (1 )
where the neighboring points are labeled by xi and xi + dxi, respectively11. Besides the norm of a vector --------- |X | := ∘ gijXiXj, the “angle” between directions X, Y can be defined by help of the metric:
i j cos(∠ (X, Y )) := gijX--Y--. |X ||Y |

From this we note that an antisymmetric part of the metrical’s tensor does not influence distances and norms but angles.

We are used to g being a symmetric tensor field, i.e., with gik = g(ik) with only D (D + 1)∕2 components; in this case the metric is called Riemannian if its eigenvalues are positive (negative) definite and Lorentzian if its signature is ± (D − 2)12. In this case, the norm is ∘ ---------- |X | := |gijXiXj |. In space-time, i.e., for D = 4, the Lorentzian signature is needed for the definition of the light cone: gijdxidxj = 0. The paths of light signals through the cone’s vertex are assumed to lie in this subspace. In unified field theory, the line element (“metric”) gik is an asymmetric tensor, in general. When of full rank, its inverse ik g is defined through13

gmigmj = δji, gimgjm = δji. (2 )
In the following, the decomposition into symmetric and antisymmetric parts is denoted by14:
gik = hik + kik, (3 ) gik = lik + mik. (4 )
hik and lik have the same rank; also, hik and lik have the same signature [27]. Equation (2*) looks quite innocuous. When working with the decompositions (3*), (4*) however, eight tensors are floating around: hik and its inverse hik (indices not raised!); kik and its inverse kik; lik ⁄= hik and its inverse lik ⁄= hik, and finally ik ik m ⁄= k and its inverse mik.

With the decomposition of the inverse jm g (4*) and the definitions for the respective inverses

h hik = δk; k kik = δk; l lik = δk; m mik = δk, (5 ) ij j ij j ij j ij j
the following relations can be obtained:15
lik = l(ik) = h-hik + kkimkknh (6 ) g g mn
ik [ik] k ik h im kn m = m = -k + -h h kmn (7 ) g g
where g =: det(gik) ⁄= 0, k =: det(kik) ⁄= 0, h =: det(hik) ⁄= 0. We also note:
g = h + k + hhklhmnk k , (8 ) 2 km ln
ij ij ij ir js ir js gg = hh + kk + hh h krs + kk k hrs. (9 )
Another useful relation is
h g2 = --, (10 ) l
with l = det(lij). From (9*) we see that unlike in general relativity even invariants of order zero (in the derivatives) do exist: kh, and hklhmnkkmkln; for the 24 invariants of the metric of order 1 in space-time cf. [512, 513, 514].

Another consequence of the asymmetry of gik is that the raising and lowering of indices with gik now becomes more complicated. For vector components we must distinguish:

→y. := gkjyj, ←y .:= yjgjk, (11 ) k k
where the dot as an upper index means that an originally upper index has been lowered. Similarly, for components of forms we have
→k kj ← k jk w. := g wj, w . := wjg . (12 )
The dot as a lower index points to an originally lower index having been raised. In general, → . ← .→ k. ← k. y k ⁄= yk,w ⁄= w. Fortunately, the raising of indices with the asymmetric metric does not play a role in the following.

An easier possibility is to raise and lower indices by the symmetric part of gjk, i.e., by hjk and its inverse hij.16 In fact, this is often seen in the literature; cf. [269*, 297*, 298*]. Thus, three new tensors (one symmetric, two skew) show up:

ˇkij := hishjtk ⁄= kij, ˇl := h h lst ⁄= l , st st ij is jtij ijij mˇij := hishjtm ⁄= mij, ˇh := h .
Hence, Ikeda instead of (9*) wrote:
ij ij 1 ij ij ir j. ρ ijrs gg = h[h (1 + 2kijˇk ) + ˇk − ˇk ˇks + 2h-𝜖 krs],
with 1 ijlm ρ := 8𝜖 ˇkijˇklm. For a physical theory, the “metric” governing distances and angles must be a symmetric tensor. There are two obvious simple choices for such a metric in UFT, i.e., hik and lik. For them, in order to be Lorentz metrics, h < 0 (l := det(lij) < 0) must hold. The light cones determined by h ik and by lik are different, in general. For further choices for the metric cf. Section 9.7.

The tensor density formed from the metric is denoted here by

ˆgij = √ −-g gij, ˆg = (√ −-g)−1 g . (13 ) ij ij

The components of the flat metric (Minkowski-metric) in Cartesian coordinates is denoted by ηik:

ηik = δ0iδ0k − δ1iδ1k − δ2iδ2k − δ3iδ3k.

2.1.1 Affine structure

The second structure to be introduced is a linear connection (affine connection, affinity) L with 3 D components k L ij; it is a geometrical object but not a tensor field and its components change inhomogeneously under local coordinate transformations.17 The connection is a device introduced for establishing a comparison of vectors in different points of the manifold. By its help, a tensorial derivative ∇, called covariant derivative is constructed. For each vector field and each tangent vector it provides another unique vector field. On the components of vector fields X and linear forms ω it is defined by

+ i i+ ∂Xi i j + ∂ ωi j ∇kX = X ∥k := ---k + L kj X , ∇k ωi := ωi+∥k = ---k − Lki ωj. (14 ) ∂x ∂x
The expressions + ∇kXi and ∂∂Xxik are abbreviated by i X +∥k and Xi,k = ∂kXi. For a scalar f, covariant and partial derivative coincide: ∂f- ∇if = ∂xi ≡ ∂if ≡ f,i. The antisymmetric part of the connection, i.e.,
Sij k= L[ijk] (15 )
is called torsion; it is a tensor field. The trace of the torsion tensor l Si =: Sil is called torsion vector or vector torsion; it connects to the two traces of the linear connection Li =: Lill ;L&tidle;j =: L ljl as Si = 1∕2 (Li − L&tidle;i ). Torsion is not just one of the many tensor fields to be constructed: it has a very clear meaning as a deformation of geometry. Two vectors transported parallelly along each other do not close up to form a parallelogram (cf. Eq. (22*) below). The deficit is measured by torsion. The rotation +∇ ω − ∇+ ω k i i k of a 1-form now depends on torsion S r ki:
+ + ∂ ω ∂ω ∇k ωi − ∇i ωk = ---i − ---k− 2Skirωr. ∂xk ∂xi

We have adopted the notational convention used by Schouten [537*, 540*, 683]. Eisenhart and others [182*, 438*] change the order of indices of the components of the connection:

− i i − ∇kXi = X − := ∂X--+ Ljki Xj, ∇k ωi := ω i∥k = ∂-ωi − Likjωj, (16 ) ∥k ∂xk − ∂xk
whence follows
− − ∂ ωi ∂ωk r ∇k ωi − ∇i ωk = ---k − ---i + 2Ski ωr. ∂x ∂x

As long as the connection is symmetric this does not make any difference because of

+ i − i i j ∇kX − ∇kX = 2S[kj] X = 0. (17 )
For both kinds of derivatives we have:
+ l − l ∇k (vlwl) = ∂(v-wl); ∇k(vlwl) = ∂-(vwl-). (18 ) ∂xk ∂xk

Both derivatives are used in versions of unified field theory by Einstein and others.18

A manifold provided with only a linear (affine) connection L is called affine space. From the point of view of group theory, the affine group (linear inhomogeneous coordinate transformations) plays a special rôle: with regard to it the connection transforms as a tensor; cf. Section 2.1.5 of Part I.

The covariant derivative with regard to the symmetrical part of the connection L l = Γ l (kj) jk is denoted by 0 ∇k such that19

0 ∂Xi 0 ∂ω ∇kXi = Xi∥k = ---- + Γ kjiXj, ∇k ωi = ωi∥k = ---i− Γkji ωj. (19 ) 0 ∂xk 0 ∂xk
In fact, no other derivative is necessary if torsion is explicitly introduced, because of20
+ 0 + 0 ∇kXi = ∇kXi + Skm iXm, ∇k ωi = ∇k ωi − Skim ωm. (20 )
In the following, k Γij always will denote a symmetric connection if not explicitly defined otherwise. To be noted is that: + s 0 λ[i,k] = ∇ [kλi] + λsSki = ∇ [kλi].

For a vector density of coordinate weight z ˆXi, the covariant derivative contains one more term (cf. Section 2.1.5 of Part I):

+ ˆi − i ∇k Xˆi = ∂X-- + L iˆXj − z L rXˆi, ∇kXˆi = ∂X-- + L iXˆj − z L rXˆi. (21 ) ∂xk kj kr ∂xk jk rk
The metric density of Eq. (13*) has coordinate weight z = 1.21 For the concept of gauge weight cf. (491*) of Section 13.2.

A smooth vector field Y is said to be parallelly transported along a parametrized curve λ (u) with tangent vector X iff for its components Y i Xk (u) = 0 ∥k holds along the curve. A curve is called an autoparallel if its tangent vector is parallelly transported along it in each point:22

Xi ∥kXk (u) = σ(u)Xi. (22 )
By a particular choice of the curve’s parameter, σ = 0 may be reached. Some authors use a parameter-invariant condition for auto-parallels: XlXi ∥kXk (u) − XiXl ∥kXk (u) = 0; cf. [284*].

A transformation mapping auto-parallels to auto-parallels is given by:

j j j Lik → L ik + δ(iωk). (23 )
The equivalence class of auto-parallels defined by (23*) defines a projective structure on MD [691], [690]. The particular set of connections
k k 2 k (p)Lij =: L ij − ------δ (iLj ) (24 ) D + 1
with Lj =: L jmm is mapped into itself by the transformation (23*), cf. [608].

In Section 2.2.3, we shall find the set of transformations L j → L j+ δj-∂ωk ik ik i∂x playing a role in versions of Einstein’s unified field theory.

From the connection L k ij further connections may be constructed by adding an arbitrary tensor field T to it or to its symmetrized part:

L¯ikj = L ijk + Tij k, (25 ) k k k k L¯¯ij = L (ij) + Tij = Γij + T&tidle;ij. (26 )
By special choice of T or &tidle; T we can regain all connections used in work on unified field theories. One case is given by Schrödinger’s “star”-connection:
∗ k k 2-k L ij = L ij + 3δi Sj, (27 )
for which ∗ k ∗ k Lik = Lki or ∗ Si = 0. The star connection thus shares the vanishing of the torsion vector with a symmetric connection. Further examples will be encountered in later sections; cf. (382*) of Section 10.3.3.

2.1.2 Metric compatibility, non-metricity

We now assume that in affine space also a metric tensor exists. In the case of a symmetric connection the condition for metric compatibility reads:

Γ r r ∇kgij = gij,k − grjΓki − girΓkj = 0. (28 )
In Riemannian geometry this condition guaranties that lengths and angles are preserved under parallel transport. The corresponding torsionless connection23 is given by:
Γ k = {k } = 1gks(∂jgsi + ∂igsj − ∂sgij). (29 ) ij ij 2
In place of (28*), for a non-symmetric connection the following equation was introduced by Einstein (and J. M. Thomas) (note the position of the indices!)24
0 = g ik∥l =: gik,l − grkLirl − girL rlk. (30 ) +−
As we have seen in Section 2.1.1, this amounts to the simultaneous use of two connections: + k k k k Lij =: L(ij) + Sij = Lij and − k k k k L ij =: L(ij) − S ij = L ji.25 We will name (30*) “compatibility equation” although it has lost its geometrical meaning within Riemannian geometry.26 In terms of the covariant derivative with regard to the symmetric part of the connection, (30*) reduces to
⋅ r 0 = g+ik−∥l =: gik∥0l − 2S (i|l|k) + 2kr[iSk]l . (31 )
In the 2nd term on the r.h.s., the upper index has been lowered with the symmetric part of the metric, i.e., with hij. After splitting the metric into its irreducible parts, we obtain27
⋅ r 0 = g i+k−∥l =: hik∥0l + kik∥0l − 2S (i|l|k) + 2kr[iSk]l ,
or (cf. [632*], p. 39, Eqs. (S1), (A1)):
r r hik∥0l + 2hr (iSk)l = 0, kik∥0l + 2kr[iSk]l = 0. (32 )
Eq. (32*) plays an important role for the solution of the task to express the connection L by the metric and its first partial derivatives. (cf. Section 10.2.3.)

In place of (30*), equivalently, the ±-derivative of the tensor density gˆik can be made to vanish:

ik ˆg+− = ˆgik + ˆgskL i+ ˆgisL k− ˆgikL s = 0. (33 ) ||l ,l sl ls (ls)

From (30*) or (33*), the connection L may in principle be determined as a functional of the metric tensor, its first derivatives, and of torsion.28 After multiplication with νs, (33*) can be rewritten as − ∇i→ν k = gks&tidle;∇iνs −, where ∇&tidle; is formed with the Hermitian conjugate connection (cf. Section 2.2.2) [396].29


Although used often in research on UFT, the ±-notation is clumsy and ambiguous. We apply the ±-differentiation to (2*), and obtain: mj mj m+−j j (gmi g )∥±l = gm+−i∥l g + gmi g ∥l = (δi)∥±l . While the l.h.s. of the last equation is well defined and must vanish by definition, the r.h.s. is ambiguous and does not vanish: in both cases j δ+i = − Sijl ⁄= 0 −, j δ−i∥l = Sijl ⁄= 0 +. Einstein had noted this when pointing out that only j+ −j δ i∥l = 0 = δi∥l + − but j− j+ δi ∥l ⁄= 0, δ i ∥l ⁄= 0 + − ([147*], p. 580). Already in 1926, J. M. Thomas had seen the ambiguity of (AiBj )∥±l and defined a procedure for keeping valid the product rule for derivatives [607]. Obviously, 0 j ∇k δi = 0.

A clearer presentation of (30*) is given in Koszul-notation:

± + − ∇Z g (X, Y ) := Zg (X, Y ) − g(∇Z X, Y ) − g (X, ∇Z Y). (34 )
The l.h.s. of (34*) is the non-metricity tensor, a straightforward generalization from Riemannian geometry:
± ± ± Q (Z, X, Y ) := ∇ g (X,Y ) = ZlXiY kg = − ZlXiY k Q . (35 ) Z i+k−∥l lik
(34*) shows explicitly the occurrence of two connections; it also makes clear the multitude of choices for the non-metricity tensor and metric-compatibility. In principle, Einstein could have also used:
++ + + ∇ Zg(X, Y ) := Zg (X, Y ) − g(∇Z X, Y) − g(X, ∇Z Y ), (36 ) − − − − ∇ Zg(X, Y ) := Zg (X, Y ) − g(∇Z X, Y) − g(X, ∇Z Y ), (37 ) 00 0 0 ∇Z g(X, Y ) := Zg (X, Y ) − g(∇Z X, Y) − g(X, ∇Z Y ). (38 )
and further combinations of the 0- and ±-derivatives. His adoption of (30*) follows from a symmetry demanded (Hermitian or transposition symmetry); cf. Section 2.2.2.

An attempt for keeping a property of the covariant derivative in Riemannian geometry, i.e., preservation of the inner product under parallel transport, has been made by J. Hély [249]. He joined the equations 0 = g−ik−∥l; 0 = gi+k+∥l to Eq. (30*). In the presence of a symmetric metric hij, in place of Eqs. (25*), (26*) a decomposition

L ijk = {kij)h + u ijk (39 )
with arbitrary u ijk can be made.30 Hély’s additional condition leads to a totally antisymmetric u k ij.

We will encounter another object and its derivatives, the totally antisymmetric tensor:

√ --- ijkl √--- ijkl 𝜖ijkl := − g ηijkl, 𝜖 := (1∕ − g) η , (41 )
where ηijkl is the totally antisymmetric tensor density containing the entries 0, ±1 according to whether two indices are equal, or all indices forming an even or odd permutation. For certain derivatives and connections, the object can be covariantly constant [473*, 484*]:
ijk l ijkl 𝜖 = 0, 𝜖++++ = 𝜖−−−− = 𝜖ijklL s. (42 ) i0j0k00l||r ||r ||r [rs]

2.2 Symmetries

2.2.1 Transformation with regard to a Lie group

In Riemannian geometry, a “symmetry” of the metric with regard to a C ∞-generator X = ξa -∂a ∂x of a Lie algebra (corresponding, locally, to a Lie-group)

l [X (i),X (j)] = cij X (l),
is defined by
ℒξgab = 0 = gab,c ξc + gcb ξc,a + gac ξc,b. (43 )
The vector field ξ is named a Killing vector; its components generate the infinitesimal symmetry transformation: xi → xi′ = xi + ξi. Equation (43*) may be expressed in a different form:
g ℒ ξgab = 2∇ (aξb) = 0. (44 )

In (44*), g ∇ is the covariant derivative with respect to the metric gab [Levi-Civita connection; cf. (29*)]. A conformal Killing vector η satisfies the equation:

l ℒηgab = f(x )gab. (45 )

2.2.2 Hermitian symmetry

This is a generalization (a weakening) of the symmetrization of a real symmetric metric and connection:31 Hermitian “conjugate” metric and connection are introduced for a complex metric and connection by

&tidle; k k &tidle;gik := gki; L ij := Lji . (46 )
In terms of the real tensors h ,k ,L k,S k ik ik ij ij, i.e., of g = h + i k ik ik ik, L k = Γ k + i S k ij ij ij obviously k k &tidle;gik = ¯gik, &tidle;L ij = ¯L ij holds, if the symmetry of hik and the skew-symmetry of kik are taken into account. For a real linear form + &tidle; − ωi : (∇k ωi) = ∇i ωk. Hermitian symmetry then means that for both, metric and connection, k k &tidle;gik = gik,L&tidle;ij := L ij is valid. For the determinant g of a metric with Hermitian symmetry, the relation g = ¯g holds.

The property “Hermitian” (or “self-conjugate”) can be generalized for any pair of adjacent indices of any tensor (cf. [149*], p. 122):

&tidle;A...ik...(grs) := (47 )
&tidle; Aij is called the (Hermitian) conjugate tensor. A tensor possesses Hermitian symmetry if A&tidle;...ik...(grs) = A...ik...(grs). Einstein calls a tensor anti-Hermitian if
A&tidle;...ik...(grs) := = − A...ik...(grs). (48 )
As an example for an anti-Hermitian vector we may take vector torsion l Li = L [il] with &tidle; Li = − Li. The compatibility equation (30*) is Hermitian symmetric; this is the reason why Einstein chose it.

For real fields, transposition symmetry replaces Hermitian symmetry.

! &tidle; k k ! k &tidle;gij := gji = gij, L ij := Lji = Lij , (49 )
with A&tidle;ij = Aji.

In place of (47*), M.-A. Tonnelat used

&tidle;A...ik...(Lrst ) :=;Lrst ) (50 )
as the definition of a Hermitian quantity [627*]. As an application we find &tidle;g = g i+k−∥l i−k+∥l and &tidle;+ik− i−k+ ˆg ||l = ˆg ||l.

2.2.3 λ-transformation

In (23*) of Section 2.1.1, we noted that transformations of a symmetric connection Γ k ij which preserve auto-parallels are given by:

′Γ ijk = Γijk+ λiδ kj + λjδki , (51 )
where λi is a real 1-form field. They were named projective by Schouten ([537*], p. 287). In later versions of his UFT, Einstein introduced a “symmetry”-transformation called λ-transformation [156*]:
′Γ k = Γ k+ λ δk. (52 ) ij ij j i
Einstein named the combination of the “group” of general coordinate transformations and λ-transformations the “extended” group U. For an application cf. Section 9.3.1. After gauge- (Yang–Mills-) theory had become fashionable, λ-transformations with λi = ∂iλ were also interpreted as gauge-transformations [702*, 23*]. According to him the parts of the connection irreducible with regard to diffeomorphisms are “mixed” by (52*), apparently because both will then contain the 1-form λi. Under (52*) the torsion vector transforms like ′Sk = Sk − 3λi 2, i.e., it can be made to vanish by a proper choice of λ.

The compatibility equation (30*) is not conserved under λ-transformations because of g ik∥l → g ik∥l − 2gi(kλl) +− +−. The same holds for the projective transformations (51*), cf. ([430*], p. 84). No generally accepted physical interpretation of the λ-transformations is known.

2.3 Affine geometry

We will speak of affine geometry in particular if only an affine connection exists on the 4-manifold, not a metric. Thus the concept of curvature is defined.

2.3.1 Curvature

In contrast to Section 2.1.3 of Part I, the two curvature tensors appearing there in Eqs. (I,22) and (I,23) will now be denoted by the ±-sign written beneath a letter:

K i = ∂ L i− ∂ L i+ L iL m − L iL m , (53 ) + jkl k lj l kj km lj lm kj K i = ∂ L i− ∂ L i+ L iL m − L iL m . (54 ) − jkl k jl l jk mk jl ml jk
Otherwise, this “minus”-sign and the sign for complex conjugation could be mixed up.

Trivially, for the index pair j,k, i i K− jkl ⁄= &tidle;K+ jkl. The curvature tensors (53*), (54*) are skew-symmetric only in the second pair of indices. A tensor corresponding to the Ricci-tensor in Riemannian geometry is given by

K jk := K ljkl = ∂kLljl− ∂lL klj + LkmlLljm − L lm lLkjm. (55 ) + +
On the other hand,
l l l l m l m K− jk := K− jkl = ∂kLjl − ∂lL jk + Lmk Ljl − L ml Ljk . (56 )
Note that the Ricci tensors as defined by (55*) or (56*) need not be symmetric even if the connection is symmetric, and also that K jk ⁄= K&tidle;jk − + when &tidle;K denotes the Hermitian (transposition) conjugate. Thus, in general
− K [jk] := ∂[kSj] + ∇lSlkj. (57 ) −

If the curvature tensor for the symmetric part of the connection is introduced by:

K ijkl = ∂kΓ lji− ∂lΓ kij + ΓkmiΓljm − Γ lm iΓkmj , (58 ) 0
K i (L ) = K i (Γ ) + S i − S i + S iS m − S iS m . (59 ) − jkl 0 jkl jl∥0k jk∥0l mk jl ml jk
The corresponding expression for the Ricci-tensor is:
l l l l m l m K0 jk := K0 jkl = ∂kΓ lj − ∂lΓ kj + Γ km Γlj − Γlm Γ kj , (60 )
whence follows:
K0 [jk] := ∂[kΓ j] (61 )
with l Γ k = Γkl. Also, the relations hold (for (63*) cf. [549*], Eq. (2,12), p. 278)):
l m m l K+ jk = K0 jk + Sjk∥l − Sj∥k − Sjk Sm − Sjl Skm, (62 ) 0 0
l m m l K− jk = K0 jk − S jk∥l + Sj∥0k − Sjk Sm − S jl Skm. (63 ) 0
A consequence of (62*), (63*) is:
K − K = − 2S + 2S l , K + K = 2K + 2S lS m − 2S mS . (64 ) + jk − jk j∥0k jk ∥0l + jk − jk 0 jk km lj jk m
Another trace of the curvature tensor exists, the so-called homothetic curvature32:
V = K j = ∂ L j− ∂ L j. (65 ) +kl + jkl k lj l kj
V = K j = ∂ L j− ∂ L j, (66 ) − kl − jkl k jl l jk
such that V− kl − V+ kl = 2∂lSk − 2∂kSl. For the curvature tensor, the identities hold:
K i − 2 ∇ S i+ 4S rS i= 0, (67 ) − {jkl} − {j kl} {jk l}r ∇ {kK i + 2K i S r = 0. (68 ) − − |j|lm} −jr{k lm}
where the bracket {... } denotes cyclic permutation while the index |j| does not take part.

Equation (68*) generalizes Bianchi’s identity. Contraction on i,j leads to:

V + 2K = 2∇ S l+ 4S rS + 4∇ S , (69 ) jk [jk] l jk jk r [j k]
or for a symmetric connection (cf. Section of Part I, Eq. (38)):
V + 2K = 0. jk [jk]
These identities are used either to build field equations without use of a variational principle, or for the identification of physical observables; cf. Section 9.7.

Finally, two curvature scalars can be formed:

ij ij K+ = g K+ ij , K− = g K− ij. (70 )

For a symmetric connection, an additional identity named after O. Veblen holds:

K i + K i + K i + K i = 0. (71 ) 0 jkl,m 0 ljm,k 0 mlk,j 0 kmj,l

The integrability condition for (30*) is ([399*], p. 225), [51]:

griRk + gkrRi = 0 (72 ) rlm rlm

For a complete decomposition of the curvature tensor (53*) into irreducible parts with regard to the permutation group further objects are needed, as e.g., ajkl b ajkl b 𝜖 K+ jkl = 2𝜖 ∂[kSl]j; cf. [348].

2.3.2 A list of “Ricci”-tensors

In many approaches to the field equations of UFT, a generalization of the Ricci scalar serves as a Lagrangian. Thus, the choice of the appropriate “Ricci” tensor plays a distinct role. As exemplified by Eq. (64*), besides K+jk and K− jk there exist many possibilities for building 2-rank tensors which could form a substitute for the unique Ricci-tensor of Riemannian geometry. In ([150*], p. 142), Einstein gives a list of 4 tensors following from a “single contraction of the curvature tensor”. Santalò derived an 8-parameter set of “Ricci”-type tensors constructed by help of l m l m K− ik, Sik∥l, V− ik(Γ ), Si∥k, Sk∥i, S ik Sm, SiSk, Sim Skl ∗ ∗ ∗  ([524*], p. 345). He discusses seven of them used by Einstein, Tonnelat, and Winogradzki.33 The following collection contains a few examples of the objects used as a Ricci-tensor in variational principles/field equations of UFT besides K + ik and K − ik of the previous section.34 They all differ in terms built from torsion. Among them are:

Her 1 K− ik = − 2-(K− ik + &tidle;K− ik) = Pik = L ik l,l − LimlL lk m − 1-(Lill ,k + L lk l,i) + 1L ikm (L mll+ Llml)([147 ], p. 581 ) (73 ) 2 2 1- &tidle; l l m = 2(K0 ik + K0 ik) + Sik∥0l + Sim Skl ; (74 ) 1 1 P ∗ik = L ik l,l − Lim lL lk m − -(L(il)l ,k + L(lkl) ,i) +-Likm(L mll+ L lm l)([150], p. 142) (75 ) 2 2 Her = K− ik + S[i,k]([371], p. 247–248) (76 ) 2 (1)Rik = − K ik + --(∂iSk − ∂kSi )([632], p. 129); (77 ) − 3 (2)R = ∂ L l− ∂ L l+ L lL m − L lL m + 1(∂ S − ∂ S ) − 1S S ([632], p. 129) ik l ik k (il) ik (lm) im lk 3 i k k i 3 i k 2 l m 1 m 1 = − K0 ik − 3-S[i∥k] + Sim S kl + 3S ik Sm − 3SiSk; (78 ) 0 (3)Rik = (2)Rik − 1V ik,([632], p. 129); (79 ) 2+ Her 1 Uik = K− ik − --[Si,k − Sk,i + SiSk] (80 ) 3 = K ik − Sikl∥l + Sim lS kml − 2S [i∥k] − 2S ikm Sm − 1SiSk,([151], p. 137; (81 ) 0 0 3 0 3 3 ∗ + l l m l 1 m l l R ik = − K ik + ∇kSi = L ik ,l − L im L lk − L (il) ,k +--Lik (L ml + L lm )([156], p. 144) (82 ) − ∗∗ 2∗ √ --- R ik = R ik − [(log( − g)),i]||k([156], p. 144) (83 ) −
Further examples for Ricci-tensors are given in (475*), (476*) of Section 13.1.

One of the puzzles remaining in Einstein’s research on UFT is his optimism in the search for a preferred Ricci-tensor although he had known, already in 1931, that presence of torsion makes the problem ambiguous, at best. At that time, he had found a totality of four possible field equations within his teleparallelism theory [176]. As the preceding list shows, now a 6-parameter object could be formed. The additional symmetries without physical support suggested by Einstein did not help. Possibly, he was too much influenced by the quasi-uniqueness of his field equations for the gravitational field.

2.3.3 Curvature and scalar densities

From the expressions (73*) to (81*) we can form scalar densities of the type: ikHer ˆg K− ik to ik ˆg Uik etc. As the preceding formulas show, it would be sufficient to just pick gˆikK 0 ik and add scalar densities built from homothetic curvature, torsion and its first derivatives in order to form a most general Lagrangian. As will be discussed in Section 19.1.1, this would draw criticism to the extent that such a theory does not qualify as a unified field theory in a stronger sense.

2.3.4 Curvature and λ-transformation

The effect of a λ-transformation (52*) on the curvature tensor K ijkl − is:

K i → K i + 2 ∂ λ δ i. (84 ) − jkl − jkl [k l] j
In case the curvature tensor K i + jkl is used, instead of (52*) we must take the form for the λ-transformation:35
′ k k k Γij = Γij + λiδj . (85 )


K ijkl → K ijkl + 2 ∂[kλl]δji. (86 ) + +
also holds. Application of (52*) to K i + jkl, or (85*) to K i − jkl results in many more terms in λk on the r.h.s. For the contracted curvatures a λ-transformation leads to (cf. also [430]):
K jk → K jk − 2∂[kλj], V jk → V jk + 2∂[jλk], V jk → V jk + 8∂ [jλk]. (87 ) − − + + − −
If λi = ∂iλ, the curvature tensors and their traces are invariant with regard to the λ-transformations of Eq. (52*). Occasionally, ′ k k k Γij = Γij + (∂iλ )δj is interpreted as a gravitational gauge transformation.

2.4 Differential forms

In this section, we repeat and slightly extend the material of Section 2.1.4, Part I, concerning Cartan’s one-form formalism in order to make understandable part of the literature. Cartan introduced one-forms 𝜃ˆa (ˆa = 1,...,4) by 𝜃ˆa := h ˆaldxl. The reciprocal basis in tangent space is given by eˆj = hlˆj∂∂xl. Thus, 𝜃ˆa(eˆj) = δˆa ˆj. An antisymmetric, distributive and associative product, the external or “wedge”(∧)-product is defined for differential forms. Likewise, an external derivative d can be introduced.36 The metric (e.g., of space-time) is given by ηˆıkˆ𝜃ˆı ⊗ 𝜃ˆk, or glm = ηˆıˆkhˆilhˆkm. The covariant derivative of a tangent vector with bein-components X ˆk is defined via Cartan’s first structure equations,

ˆ Θi := D 𝜃ˆı = d𝜃ˆı + ωˆıˆl ∧ 𝜃 l, (88 )
where ˆı ω ˆk is the connection-1-form, and ˆı Θ is the torsion-2-form, ˆı ˆı ˆl ˆm Θ = − Sˆlmˆ 𝜃 ∧ 𝜃. We have ω ˆıˆk = − ωˆkˆı. The link to the components L[ijk] of the affine connection is given by ω ˆı = h ˆıhm L l𝜃 ˆr ˆk l ˆk ˆrm37. The covariant derivative of a tangent vector with bein-components ˆ X k then is
DX ˆk := dX ˆk + ωˆk X ˆl. (89 ) ˆl

By further external derivation on Θ we arrive at the second structure relation of Cartan,

D Θ ˆk = Ω ˆk∧ 𝜃 ˆl. (90 ) ˆl
In Eq. (90*) the curvature-2-form ˆ ˆ Ω kˆl = 12 Rkˆlmˆˆn𝜃ˆm ∧ 𝜃ˆn appears, which is given by
Ω ˆk = d ωˆk + ωkˆ∧ ω ˆk. (91 ) ˆl ˆl ˆl ˆl
Ω ˆkˆk is the homothetic curvature.

A p-form in n-dimensional space is defined by

ˆi1 ˆi2 ˆip ω = ωˆi1ˆi2...ˆipdx ∧ dx ∧ ⋅⋅⋅ ∧ dx
and, by help of the so-called Hodge *-operator, is related to an (n-p)-form)38
∗ 1 ˆi1ˆi2...ˆip ˆk ˆk ˆk ω:= --------𝜖ˆk1ˆk2...ˆkn−p ωˆi1ˆi2...ˆipdx 1 ∧ dx 2 ∧ ⋅⋅⋅ ∧ dx n−p. (n − p)!

2.5 Classification of geometries

A differentiable manifold with an affine structure is called affine geometry. If both, a (possibly non-symmetric) “metric” and an affine structure, are present we name the geometry “mixed”. A subcase, i.e., metric-affine geometry demands for a symmetric metric. When interpreted just as a gravitational theory, it sometimes is called MAG. A further subdivision derives from the non-metricity tensor being zero or ⁄= 0. Riemann–Cartan geometry is the special case of metric-affine geometry with vanishing non-metricity tensor and non-vanishing torsion. Weyl’s geometry had non-vanishing non-metricity tensor but vanishing torsion. In Sections 2.1.3 and 4.1.1 of Part I, these geometries were described in greater detail.

2.5.1 Generalized Riemann-Cartan geometry

For the geometrization of the long-range fields, various geometric frameworks have been chosen. Spaces with a connection depending solely on a metric as in Riemannian geometry rarely have been considered in UFT. One example is given by Hattori’s connection, in which both the symmetric and the skew part of the asymmetric metric enter the connection39 [240*]:

H k kl L ij = 1 ∕2 h (gli,j + gjl,i − gji,l) (92 ) = {k }h + 1∕2 hkl(kli,j + kjl,i + kij,l), (93 ) ij
where hkl is the inverse of hkl = g(kl). As described in Section 6.2 of Part I, its physical content is dubious. As the torsion tensor does not vanish, in general, i.e.,
H k kl Sij = h (kl[i,j] + 1∕2kij,l) (94 )
this geometry could be classified as generalized Riemann–Cartan geometry.

2.5.2 Mixed geometry

Now, further scalars and scalar densities may be constructed, among them curvature scalars (Ricci-scalars):

jk jk jk K+ := g K+jk = l K+ (jk) + m K+ [jk], (95 ) jk jk jk K− := g K− jk := l K− (jk) + m K− [jk]. (96 )
Here, ljk and mjk come from the decomposition into irreducible parts of the inverse of the non-symmetric metric g jk. Both parts on the r.h.s. could be taken as a Lagrangian, separately. The inverse jk h of the symmetric part, i.e., of g(jk) = hjk could be used as well to build a scalar: jk h K+(jk). Mixed geometry is the one richest in geometrical objects to be constructed from the asymmetric metric and the asymmetric connection. What at first may have appeared as an advantage, turned out to become an ‘embarras de richesses’: defining relations among geometric objects and physical observables abound; cf. Section 9.7.

Whenever a symmetric tensor appears which is independent of the connection and of full rank, it can play the role of a metric. The geometry then may be considered to be a Riemannian geometry with additional geometric objects: torsion tensor, non-metricity tensor, skew-symmetric part of the “metric” etc. These might be related to physical observables. Therefore, it is moot to believe that two theories are different solely on the basis of the criterion that they can be interpreted either in a background of Riemannian or mixed geometry. However, by a reduction of the more general geometries to a mere Riemannian one plus some additional geometric objects the very spirit of UFT as understood by Einstein would become deformed; UFT explicitly looks for fundamental geometric objects representing the various physical fields to be described.

2.5.3 Conformal geometry

This is an “angle preserving” geometry: in place of a metric m gij(x ) a whole equivalence class γij(xl) = ρ2(xn )gij(xm ) with a function ρ(xn ) obtains. Geometrical objects of interest are those invariant with regard to the transformation: gij(xm ) → ¯gij = ρ2(xn)gij(xm). One such object is Weyl’s conformal curvature tensor:

i i 1 i i i i 1 i i C jkl := K+ jkl − n −-2(δkK+ jl − δlK+ jk + gjlK+ k − gjkK+ l) + (n-−-1)(n-−-2) K+ (δlgjk − δkgjl), (97 )
where n is the dimension of the manifold (n = 4: space-time). Cijkl is trace-free. For n > 3,Cijkl = 0 is a necessary and sufficient condition that the space is conformally flat, i.e., γij(xl) = ρ2(xn)ηij(xm ) ([191], p. 92).

If ξk is a Killing vector field for gij, then ξk is a conformal Killing vector field for ¯gij; cf. Eq. (45*) in Section 2.1.2.

A particular sub-case of conformal geometry is “similarity geometry”, for which the restricted group of transformations acts gij(xm ) → γij(xl) = k2gij(xm ), with a constant k, cf. Section 3.1.

2.6 Number fields

In Section 2.3 of Part I, the possibility of choosing number fields different from the real numbers for the field variables was stated. Such field variables then would act in a manifold with real coordinates. A more deeply going change is the move to an underlying manifold with coordinates taken from another number field, e.g., complex spaces. The complex number field was most often used in connection with unified field theory in both roles. cf. A. Einstein, (complex space, Section 7.2), J. Moffat, (complex field on real space, Section 13) and A. Crumeyrolle, (hypercomplex manifold, Section 11.2.2).

As hypercomplex numbers are less well known, we briefly introduce them here. Let z = x + 𝜖y, x,y real and consider the algebra with two elements I,𝜖, where I is the unit element and 𝜖2 = I. z is called a hypercomplex number. A function f(z) = P (x,y) + 𝜖Q(x,y ) will be differentiable in z if

∂P ∂Q ∂P ∂Q ----= ---, ----= ----. (98 ) ∂x ∂y ∂y ∂x
The product of two identical real manifolds of dimension n can be made into a manifold with hypercomplex structure.
  Go to previous page Scroll to top Go to next page