Remarks by Arkadiusz Jadczyk on Evans' "Covariant Derivatives"

Remark 1 (received on January 26, 2007)

Indeed, Evans' exterior "covariant derivative" DÙ defined in [2,(17.16)] by

(D')                 D Ù Xab = d Ù Xab + ωac Ù Xcb .

is not covariant. The simplest example is the group SU(2) with its bi-invariant metric and basis of left-invariant vector fields.

The curvature tensor can be easily computed:

Rijkl = ½ (δilδjk − δik δjl)

The connection coefficients are also easily calculated:

ωikj = ½ εikj

(cf. for instance Coquereaux and Jadczyk, "Riemannian geometry, fiber bundles, Kaluza-Klein theories and all that", World Scientific, 1988, pp. 10-13)

Therefore ωcb Ù Rac ≠ 0 and it is evidently not a horizontal form - in other words: it does not define a tensorial object or, as physicists use to say: it is not gauge-invariant. Therefore (D') does not define a "covariant" operation.

Remark 2 (received on January 28, 2007)

Evans asserts in (4.8) of his GCUFT book (vol.1)

Dμ Γρμρ = ∂μ Γρμρ + Γμρλ Γλμρ .                                 [1,(8)] and [2,(4.8)]

To see that the "covariant derivative of the index contracted Christoffel symbol" is not covariant, it is enough to consider the simplest 1-dimensional case. Let the manifold M be the half-line M := (0,∞). We endow it with natural coordinate x in (0,∞) and the metric tensor g11(x) := 1 for all x in (0,∞). This is a flat metric, and the only Christoffel symbol in one dimension, Γ111, vanishes identically:

Γ111 = 0, for all x in (0,∞).

In particular Evans' D1 Γ111 = 0 identically.

Since D1 Γ111 is supposedly a scalar, it should vanish identically in any other coordinate system.

Let now x' be another coordinate, taking values in R given by x'(x) := log x , the inverse transformation is x(x') = ex'. We can easily compute the metric tensor component in the primed system:

g1'1'(x') = [∂ x/∂x']2 = e2x' .

This gives us the Christoffel symbol in the primed coordinate system:

Γ1'1'1' = ½ g1'1' ∂g1'1'/∂x' = 1 , for all x' in R.

Therefore Evans' covariant derivative is now:

D1' Γ1'1'1' = ∂Γ1'1'1'/∂x' + Γ1'1'1' Γ1'1'1' = g1'1' Γ1'1'1' Γ1'1'1' = e−2x' · 1 = e−2x' ≠ 0

for all x'. In particular at x'=0 (x=1) we get Evan's "covariant derivative" equal 1 while, if believed to be covariant, it should have value 0.

It is indeed hard to invent something more non-covariant!

Remarkable, that Evans' assertion could pass the refereeing of a prestigious journal.


[1] M.W. Evans, A Generally Covariant Wave Equation for Grand Unified Field Theory,
        Foundations of Physics Letters, Vol. 16, No. 6, Dec. 2003, pp. 513-547(35), Springer