56 Chapter 3 The Dirac Field
the matrix element can be written
(0| ap ag} |0) = (2n)°6® (p — q)s"* - A(p),
where A(p) is so far undetermined. Note, however, that if the norm of a state
is always positive (as it should be in any self-respecting Hilbert space), A(p)
must be greater than zero. We can now go back to (3.93), and write
dp 1
(Q)3 25,
(0| v(x)h(y) |0) =
Lv ww@oatpye ?-
-la: Ga Bis +m) ACDye HP.
This expression is properly invariant under boosts only if A(p) is a Lorentz
scalar, i.e., A(p)=A(p?). Since p? = m?, A must be a constant. So finally we
obtain
Bp 1
eP(Z—y) . A, .
(2n)3 2B, (3.94)
(0| a(x) ¥e(y) 0) = (1, +m),
Similarly, in the amplitude (0|%(y)W(2) |0), we want the only contri-
butions to be from the positive-frequency terms of #(y) and the negative-
frequency terms of (x). So ay, still annihilates the vacuum, but 6p, does not.
Then by arguments identical to those given above, we have
(OFA) ()= Ge +m), | SEig rem? B, (8.95)
where B is another positive constant. The minus sign is important; it comes
from the completeness relation (3.67) for }7 vd and the sign of x in the ex-
ponential factor. It implies that we cannot have (0| [y(x), ¥(y)] |0) = 0 out-
side the light-cone: The two terms (3.94) and (3.95) would indeed cancel if
A= ~B, but this is impossible since A and B must both be positive.
The solution, however, is now at hand. By setting A = B = 1, it is easy
to obtain (outside the light-cone)
(0| va(e)¥o(y) 10) = — (0| Bo(y)Yalz) [0).
That is, the spinor fields anticommute at spacelike separation. This is enough
to preserve causality, since all reasonable observables (such as energy, charge,
and particle number) are built out of an even number of spinor fields; for any
such observables O; and Oo, we still have [O; (x), O2(y)] = 0 for (x—y)? < 0.
And remarkably, postulating anticommutation relations for the Dirac field
solves the negative energy problem. The equal-time anticommutation relations
will be
(val), Bj (9)} = 6 — y)ba05
3.96
{valx),voly)} = {Ube why)} = ~