Appendix To Lecture 4: Remarks On Renormalization and Asymptotic Freedom

Appendix to Lecture 4: Remarks on
renormalization and asymptotic freedom

Edward Witten1
December 1996
1. Ambiguity in operator products. In the last lecture we consid-
ered Lagrangians of the form
Z 2 X
L = d4x( 21 (r)2 + m2 2 + 4!g 4 + "i(x)Oi(x));
i
where Oi are local functionals. We saw that dening correlation functions
of such Lagrangians to the rst order in "i (x) (i.e. "i "j = 0) is equivalent to
dening composite operators corresponding to Oi . Now let us work to the
second order in "i ("i "j "k = 0). This corresponds to considering products of
composite operators. Before we considered products of composite operators
at non-coinciding points x; y , and saw that such a product Oi (x)Oj (y ) is de-
ned automatically once we have dened Oi and Oj . Now we will allow the
points x and y to coincide. Then, as we know, the product is not automat-
ically dened, and its denition requires additional renormalization. This
means, we have to introduce counterterms in the Lagrangian, i.e. consider
a new Lagrangian of the form
X
L0 = L + Lk "i (x)Rk"j (x)Wk (; g)Ok(x);
i;j;k
where Lk ; Rk are dierential operators (in x) with constant coecients, and
Wk are some functions which diverge as ! 1. The functions Wk are
usually not uniquely determined and cannot be chosen canonically.
Let us consider an example. In one of the homework problems we com-
puted the 1-loop correction to the 1-particle irreducible bosonic 2-point
function in QED. We discussed that this correction can be computed
(in momentum space) as
Z
eikx hJ (x)J (0)idx;
where J (x) = is the operator of current, and the correlator in (1) is
in the theory of a free fermion. Now, as follows from the above discussion,
1
Notes by Pavel Etingof and David Kazhdan
1
the product J (x)J (x) is dened up to operators of lower order, and the
expectation value hJ (x)J (0)i is non-uniquely dened, which causes an
ambiguity in the computation of (1). However, this non-uniqueness occurs
only at x = 0. In fact, its is easy to show that the function hJ (x)J (0)i is
well dened up to adding a multiple of (x). Therefore, the ambiguity in
(1) is a constant (i.e. is independent of k).
In QED, to preserve gauge invariance, it is necessary to choose this
constant in such a way that the condition k (k) = 0 is satised. This
gives a unique way to x the constant.
2. Symmetry breaking. A symmetry that exists in a classical eld
theory may be lost in a particular renormalization scheme for the corre-
sponding quantum theory. Of course it is possible that there exists a better
scheme which preserves this symmetry, but it is also possible that there
exists no such scheme, i.e. the symmetry is broken at the quantum level.
PFor( example, consider N free massless fermions 1; ::: N , with Lagrangian
;i D i), i 2 S+ , 2 S . In this theory we have a U (N ) symme-
try. Let us add interactions in such a way that part of this symmetry is
preserved. For example, add a gauge eld A (i.e. regard the fermions to-
gether to make a section of the vector bundle S+
E , where E is an
N -dimensional Hermitian vector bundle over the spacetime), which takes
values in the Lie algebra of a subgroup H U (N ), and consider the La-
grangian ( ; DA ), where DA is the Dirac operator along the connection
A. The classical symmetry of this theory is the centralizer Z (H ) of H . In
quantum theory, however, this symmetry may fail for topological reasons.
In other words, a topological anomaly may appear.
3. An oversimplied version of experimental conrmation of
asymptotic freedom.
Consider a eld theory with electromagnetic and strong interactions,
which contains electrons (which interact only electromagnetically), and quarks
(which interact both electromagnetically and strongly). The Lagrangian of
such a theory can be written as follows. The elds are:
(i) An SU (3)-connection Ag (the eld of strong interactions),
(ii) A U (1)-connection Ae (the electromagnetic eld),
(iii) Quarks qi and an electron " (which are fermions with values in C 3
and C respectively).
Z The X
Lagrangian is
d4x( q (iD + A + A m )q + "(iD + A m)" + 1 F 2 + 1 F 2 ):
j g e i j e e2 Ae g2 Ag
j
Here e is the charge of the electron and g is the coupling of the strong
2
interaction.
Now suppose that we scatter two electrons against each other with mo-
menta p1 ; p2, and measure the amplitude of the event that after scattering
they will have momenta q1 ; q2 . As we know, this amplitude is dened by the
4-point function 4 (p1; p2; q1; q2). Let us try to compute this function and
thus predict the result of measurement.
First of all, we can use the fact that e2 is small. This means, we can
trust the perturbative expansion in powers of e.
To order e2 , we can assume that the electrons, during scattering, ex-
change only one photon, which does not interact while it moves from one
electron to the other. This corresponds to the following Feynman diagram:
Thus we have
4 (p1 ; p2; q1 ; q2) = e2 (G2((p1 q1 )2) + G2 ((p1 q2 )2 ));
where G2(p2) is the free photon propagator. Here G2 is regarded as an
operator from S
S to S
S , where S is the space of spinors.
Remark. In principle, we should include the diagrams where one of
the electrons exchanges a photon with itself, but we will not consider them,
regarding them as absorbed in the electron propagator.
To order e4, we have two possiblities.
1) The electrons could exchange two non-interacting photons. The am-
plitude of the corresponding 1-loop diagram
3
can be computed within the framework of QED.
2) The electrons could exchange only one photon, but on its way it could
split in an electron and positron, or in a quark and an antiquark. The rst
splitting scenario
is harmless, since it gives only one 1-loop diagram with no strong in-
teractions, and we can compute the amplitude of this diagram as in QED.
However, the second scenario (with quarks) really gives us trouble. Indeed,
the coupling constant g of the strong interaction is not small, so we cannot
trust the perturbation expansion in g and thus have to take into account
innitely many Feynman diagrams with any number of loops:
(here dotted lines denote the eld of strong interaction).
4
However, we know that the theory of strongly interacting quarks is
asymptotically free (when the number of quarks is not too large). Thus,
we should expect that the perturbative expansion in the eective coupling
geff g (ln p2) 1=2 should be valid at high momenta. This would mean that
at high momenta we can restrict to the 1-loop diagrams, which involve no
strong interactions. Since we have one such diagram for each type of quark,
P
the total amplitude of these diagrams is e2i i(p), where ei are charges
of quarks, and i are amplitudes of the corresponding diagrams for a particle
with charge 1. Of course, the functions i (p) depend on the (unmeasurable!)
masses mi of quarks, but at high momenta masses are irrelevant, and all
functions
P i are approximately equal to each other. Thus, the amplitude is
(p)( e2i ), where (p) is a universal function computed from QED (as in
one of the homework exercises).
It follows from asymptotic freedom that the (relative) error of this com-
putation is of order 1= ln p2. If we compute the two-loop correction (still
working to order e4 ), we will get an additional term of the order 1= ln p2 ,
and the error will be of order 1= ln2 p2. More generally, if we take into ac-
count N -loop diagrams, the error will be of order 1= lnN p2 . Thus, we get
an asymptotic series, with very slowly decaying terms, but at very high p
one can hope that it gives a reasonably good approximation to the 4-point
scattering amplitude. This approximation (to the 0-th order) could in prin-
ciple be checked experimentally, and can be regarded as a conrmation of
asymptotic freedom.
Remark. In practice, asymptotic freedom was checked experimentally
in a dierent way, but the ideology is similar to the one described above.
Correction to the text of lecture 3 (by Pavel Etingof)
Unfortunately, in Section 3.3 of Lecture 3 there is a wrong statement (no-
ticed by D.Freed). Namely, the statement \h:::O(x)O(x0)i jx x0 j [O] [O ] "
0
is incorrect. For example, it fails if O0 = 1 and O is a nontrivial operator.

This statement is true, however, if O0 = O, which is enough to make the
point which was being made in the text.

Appendix To Lecture 4: Remarks On Renormalization and Asymptotic Freedom

Uploaded by

Copyright:

Available Formats

Appendix To Lecture 4: Remarks On Renormalization and Asymptotic Freedom

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Appendix To Lecture 4: Remarks On Renormalization and Asymptotic Freedom

Uploaded by

Copyright:

Available Formats

Appendix to Lecture 4: Remarks on

renormalization and asymptotic freedom

(here dotted lines denote the eld of strong interaction).

is incorrect. For example, it fails if O0 = 1 and O is a nontrivial operator.

You might also like