Stringy Things

Notes on String Theory: Conformal Field Theory – Local Operators, the String Propagator, and Operator Product Expansions

1. Local Operators

In the last entry we introduced a theory of free massless scalars in flat 2-dimensions (i.e., a free X-CFT). From this we also introduced new terms and established notation relevant to our ongoing study of CFTs in string theory (Chapter 2 in Polchinski). What we now want to do is proceed with a review of a number of interrelated topics at the heart of stringy CFTs: namely, local operators, techniques with path integrals, string propagators, and finally operator product expansions. Each of these topics has a number of parts, and so we shall need to work piece by piece and then stitch everything together.

To begin, we note that in string perturbation theory, one of the main objects of interest is the expectation value of the path integral of a product of local operators (Polchinski, p.36). This interest is our entry point, and it represents a primary theme for much of the following discussion. So our first step should be to define what we mean by local operators. These objects may also be described as fields; however, in the context of CFTs, the notion of a field carries a different meaning than, for instance, the definition of a field in quantum field theory. In our case, a field may be viewed generally as a local expression, which may be the generic field {\phi} that enters the path integral in QFT, or as a composite operator {e^{i\phi}} or as a derivative {\partial^{n}\phi} (Tong, p.69). These are all different types of fields or local operators in the CFT dictionary.

With a definition of local operators in mind, we opened the discussion by mentioning the expectation value of the path integral as a primary object of interest. Let us now consider some general expectation value. Consider, for instance, {\mathcal{A}_{i}} that is some basis for a set of local operators. We may write the general expectation value as follows,

\displaystyle \langle \mathcal{A}_{i_{1}}(z_{1}, \bar{z}_{1}) \mathcal{A}_{i_{2}}(z_{2}, \bar{z}_{2}) ... \mathcal{A}_{i_{n}}(z_{n}, \bar{z}_{n}) \rangle \ \ (1)

If the basic idea, as mentioned, is to compute the expectation value of the path integral, a more technical or detailed description of our overarching interest is to understand the behaviour of this expectation value (1) in the limit of two operators taken to approach one another (Polchinski, p. 36). The tool that we use for such analysis is the operator product expansion (OPE). Understanding the definition of OPEs and how to compute them is one of the ultimate aims of studying stringy CFTs, important for more advanced topics that we will consider throughout the remainder of this paper. But before formally defining OPEs, it is useful to first build a deeper sense of intuition about their meaning. To do this, let as briefly review some more basics.

2. The Path Integral and Arbitrary Operator Insertions

What do we mean by path integral? And how do we understand this idea of local operator insertions? Additionally, how do we construct the important operator equations required to build a picture of OPEs? Polchinski offers several valuable contributions to a definition of the path integral, including a lengthy treatment in the Appendix of Volume 1. For our purposes, we might first emphasise the QFT view of the path integral as an integral over fields,

\displaystyle Z = \int [dX]e^{-S} \ \ (2)

We may describe this as a partition function. Now, if what we want to know is the expectation value given some operator, this implies that we want to employ the path integral representation to derive operator equations. For instance, as we read in Polchinski (p.34), given some operator we may compute,

\displaystyle \left\langle \mathcal{F}[X] \right\rangle = \int [dX]e^{-S}\mathcal{F}[X] \ \ (3)

Where {\mathcal{F}[X]} is some functional of X, typically a product of two operators, and where \langle \mathcal{F}[X] \rangle = \langle 0 \mid \mathcal{F} \mid 0 \rangle  . For multiple entries in the form,

\displaystyle  \mathcal{F}_{1}[X(z_{1}, \bar{z}_{1})] \mathcal{F}[X(z, \bar{z})] \mathcal{F}_{2}[X(z_{2}, \bar{z}_{2})]

We may write,

\displaystyle  \langle 0 \mid \mathcal{F}_{1} \mathcal{F} \mathcal{F}_{2} \mid 0 \rangle =\int [dX]e^{-S} \mathcal{F}_{1}[X(z_{1}, \bar{z}_{1})] \mathcal{F}[X(z, \bar{z})] \mathcal{F}_{2}[X(z_{2}, \bar{z}_{2})] \ \ (4)

There is a notion of time-ordering present in (4), which we will discuss later. For now, we should note that the path integral of a total derivative is always zero. This fact will prove useful in just a moment and on many other occasions in the future. As Polchinski reflects, `This is true for ordinary bosonic path integrals, which can be regarded as the limit of an infinite number of ordinary integrals, as well as for more formal path integrals as with Grassmann variables’ (Polchinski, pp. 34-35). Hence eq.(2.1.15) in Polchinski (p.35), where he considers the path integral with the inclusion of Grassmann variables,

\displaystyle  0 = \int [dX] \frac{\delta}{\delta X_{\mu}(z, \bar{z})} \exp (-S)

\displaystyle  = - \int [dX] \exp (-S) \frac{\delta S}{\delta X_{\mu} (z, \bar{z})}

\displaystyle = - \int \bigg \langle \frac{\delta S}{\delta X_{\mu} (z, \bar{z})} \bigg \rangle

\displaystyle = \frac{1}{\pi \alpha^{\prime}} \langle \partial \bar{\partial} X^{\mu}(z, \bar{z}) \rangle \ \ (5)

There is something interesting with this result. If we recall the action for the free X-CFT in the last post, remember that we found the classical EoM to be {\partial \bar{\partial} X^{\mu}(z, \bar{z}) = 0}. Notice, then, that the result (5) is the analogue statement in the quantum theory for the classical equations of motion. What is this telling us? Let us dig a bit deeper.

First, consider how the same calculation in (5) holds if we have arbitrary additional insertions `…’ in the path integral. We already considered what multiple entries in the path integral in (4). But there is a caveat: namely, these additional insertions cannot also be at {z} (something we will elaborate below). Second, in the case of multiple entries in the path integral, which implies that we may write something of the form {\int [dX] \frac{\delta}{\delta X^{\mu}(z, \bar{z})}[e^{-S}\mathcal{F}(z, \bar{z})]}, one can think of some of the insertions as preparing a state in the theory. In other words, we should note that these insertions prepare arbitrary initial and final states in the theory (Polchinski, p.35). These arbitrary initial and final states perform a similar role should we instead consider boundary conditions, except with the offered convenience that we may now write the following path integral statement,

\displaystyle \left\langle \partial\bar{\partial}X^{\mu}(z, \bar{z}) ... \right\rangle = 0 \ \ (6)

Now, if (5) is the analogous statement in the quantum theory for the classical equations of motion, look at (6). Notice, as an operator statement, it is the same as in the Hilbert space formalism,

\displaystyle \partial\bar{\partial}\hat{X}(z, \bar{z}) = 0 \ \ (7)

Polchinski describes (7) as holding for all matrix elements of the operator {\hat{X}(z, \bar{z})}, with all relations that hold (6) being operator equations (Polchinski, p.34). These two points are important. It should also be noted that (7) is Ehrenfest’s theorem, which makes a lot of sense because it is telling us something that we already know or suspect: namely, the expectation values of the operators obey the classical equations of motion. But, again, this proves true only when the additional insertions `…’ in the path integral are located away from {z}. So let us now look into this subtlety. If, for example, addition insertions cannot be coincident at {z}, then let us consider what happens when we do indeed have coincident points at {z}! It follows,

\displaystyle  0 = \int [dX] \frac{\delta}{\delta X_{\mu}(z, \bar{z})}[exp(-S)X^{\nu}(z^{\prime}, \bar{z^{\prime}})]

\displaystyle = \int d[X] exp(-S) [\eta^{\mu \nu}\delta^{2}(z - z^{\prime}, \bar{z} - \bar{z}^{\prime}) + \frac{1}{\pi\alpha^{\prime}}\partial_{z}\partial_{\bar{z}}X^{\mu}(z, \bar{z})X^{\nu}(z^{\prime}, \bar{z}^{\prime})]

\displaystyle = \eta^{\mu \nu} \langle \delta^{2}(z - z^{\prime}, \bar{z} - \bar{z}^{\prime}) \rangle + \frac{1}{\pi\alpha^{\prime}}\partial_{z}\partial_{\bar{z}}\langle X^{\mu}(z, \bar{z})X^{\nu}(z^{\prime}, \bar{z}^{\prime})\rangle \ \ (8)

Where the {\delta^{2} (z^{\prime} - z, \bar{z}^{\prime} - \bar{z})} term comes from differentiating {\frac{\delta X^{\mu}(z^{\prime}, \bar{z}^{\prime})}{\delta X^{\mu}(z, \bar{z})}} that appears in the computation. What we see in (8) is that at coincident points the classical equations of motion do not hold at the quantum level. This implies a few things. First, the good news is that we obtain our previous result that the EoM agrees as an operator statement of the ground state specifically under the conditions {z \neq z^{\prime}}. Second, the implication is clearly that with arbitrary additional insertions `…’ in the path integral, so long that these are far away from {z} and {z^{\prime}}, we may can rewrite (8) as,

\displaystyle \frac{1}{\pi \alpha^{\prime}} \partial_{z}\partial_{\bar{z}} \langle X^{\mu}(z, \bar{z}) X^{\nu}(z^{\prime}, \bar{z}^{\prime}) \ ... \ \rangle = -\eta^{\mu \nu} \langle \delta^{2}(z - z^{\prime}, \bar{z} - \bar{z}^{\prime}) \ ... \ \rangle \ \ (9)

Where the ellipses are, again, the additional fields. Importantly, we may note that the following holds as an operator equation,

\displaystyle \frac{1}{\pi \alpha^{\prime}} \partial_{z}\partial_{\bar{z}} X^{\mu}(z, \bar{z})X^{\nu}(z^{\prime},\bar{z}^{\prime}) = - \eta^{\mu \nu} \delta^{2}(z - z^{\prime}, \bar{z} - \bar{z}^{\prime}) \ \ (10)

We are going to want to solve this equation in the future, because the solution will prove useful when computing OPEs. In the meantime, what should be understood is that what we have accomplished here is that we’ve modified the EoM to take into account that there is a collision between points at {z} and {z^{\prime}}. And we have also found that this behaviour can be derived as an operator statement. The purpose and greater logic for such an exercise will become increasingly clear. Meanwhile, notice that we now have a product of operators. Although it will not be proven here, it follows that in the Hilbert space formalism this product in the path integral becomes time-ordered (Polchinski, p.36). We also see that the delta function appears when the derivatives act on the time-ordering.

To summarise, these last results signal what has already been alluded (however vaguely) about the definition of OPEs in the final paragraph of Section 1. If, moreover, the general theme is so far one of path integrals and local operator insertions, the picture we are ultimately constructing is one of such insertions inside time-ordered correlation functions. These correlation functions can then be held as operator statements.

3. Time-ordered Correlation Functions, Normal Ordering, and the String Propagator

Before formally introducing and defining OPEs, we should spend a few more moments developing the picture and building intuition. For example, when it comes to the idea of time-ordered correlation functions, we will learn that solving the operator equation (10) gives us,

\displaystyle \langle X^{\mu}(z, \bar{z})X^{\nu}(z^{\prime}, \bar{z}^{\prime}) \rangle = - \eta^{\mu \nu} \frac{\alpha^{\prime}}{2} \ln \mid z - z^{\prime} \mid^{2} \ \ (11)

The computation required to arrive at this result may not yet have much meaning and may be too forward thinking. We will come to understand it soon. What can be understood at this juncture are some of the pieces of this equation. The most important note is that (11) is the propagator of the theory of massless scalars that we have been working with in our study of CFTs (i.e., the free X-CFT). Notice, on the left-hand side of the equality, we a two-point correlation function. As it has been stated, correlation functions are time-ordered. Let us focus on this notion of time-ordering. For instance, consider a Wick expansion for {X^{\mu}(z, \bar{z})},

Where we have indicated the use of contraction notation that will be defined later. The first observation is that we have a two-point correlation function, and we have some term {T}. We also have colons on the right-hand side. For the {T} term, it indicates that the expression is time-ordered in the same way one will find in basic QFT (Polchinski, p.36). Writing {T} in full we find,

\displaystyle  T (X^{\mu} (z, \bar{z}), X^{\nu} (z^{\prime}, \bar{z}^{\prime}))

\displaystyle = X^{\mu} (z, \bar{z}) X^{\nu} (z^{\prime}, \bar{z}^{\prime}) \theta(z - z^{\prime}) + X^{\nu}(z^{\prime}, \bar{z}^{\prime})X^{\mu}(z, \bar{z})\theta(z^{\prime} - z) \ \ (13)

Now, looking again at (13), it is worth pointing out a few other things. Firstly, what we will learn in the future, particularly as we advance our discussion on CFTs, is that this time-ordering will prove very useful. Eventually we are going to want to make conformal transformations from an infinite cylinder to the complex plane, and we will learn that time-ordering on the cylinder corresponds to radially ordering on the complex plane. Reversely, we will see that radial ordering on the complex plane corresponds with time-ordering in the path integral. This is a featured point of study in Section 2.6 of Polchinski and it is something we will discuss later. Secondly, for the colons on the right-hand side, they indicate normal ordering. We saw normal ordering in the past discussion on the free string string spectrum using light-cone gauge quantisation. Notice, then, that on the far right-hand side we have a normal ordered product. The definition of normal ordered operators follows as (Polchinski, p.36),

\displaystyle :X^{\mu}(z, \bar{z}): = X^{\mu}(z, \bar{z}) \ \ (14)

And for the normal ordered product we have,

\displaystyle :X^{\mu}(z_{1}, \bar{z}_{1}), X^{\nu}(z_{2}, \bar{z}_{2}): = X^{\mu}(z_{1}, \bar{z}_{1}) X^{\nu}(z_{2}, \bar{z}_{2}) + \frac{\alpha^{\prime}}{2}\eta^{\mu \nu} \ln \mid z_{12} \mid^{2} \ \ (15)

Where {z_{ij} = z_{i} - z_{j}}. Furthermore, for arbitrary numbers of fields, the normal ordered product may be written as,

\displaystyle :X^{\mu_{1}} (z_{1}, \bar{z}_{1}) ... X^{\mu_{n}}(z_{n}, \bar{z}_{n}): = X^{\mu_{1}}(z_{1}, \bar{z}_{1}) ... X^{\mu_{n}}(z_{n}, \bar{z}_{n}) + \sum \text{subtractions} \ \ (16)

Where, for the subtractions, we sum the pairs of fields from the product and then replace each pair with its expectation value {\frac{\alpha^{\prime}}{2}\eta^{\mu \nu} \ln \mid z_{ij} \mid^{2}}. We will elaborate more on (16) later. Meanwhile, consider again the operator equation (10). If what we want to do is define a product of operators that would satisfy the classical EoM, then from (16) and using (10) we can compute,

\displaystyle \partial_{z} \partial_{\bar{z}} :X^{\mu}(z_{1}, \bar{z}_{1}) X^{\nu}(z_{2}, \bar{z}_{2}): = \partial_{z} \partial_{\bar{z}} X^{\mu}(z_{1}, \bar{z}_{1}) X^{\nu}(z_{2}, \bar{z}_{2}) + \frac{\alpha^{\prime}}{2}\eta^{\mu \nu} \partial_{z} \partial_{\bar{z}} \ln \mid z_{12} \mid^{2}

\displaystyle = - \pi \alpha^{\prime} \eta^{\mu \nu} \delta^{2}(z_{1} - z_{2}, \bar{z}_{1} - \bar{z}_{2}) + \frac{\alpha^{\prime}}{2} \eta^{\mu \nu} \partial_{z}\partial_{\bar{z}} \ln \mid z_{12} \mid^2

\displaystyle = - \pi \alpha^{\prime} \eta^{\mu \nu} \delta^{2}(z_{1} - z_{2}, \bar{z}_{1} - \bar{z}_{2}) + \pi \alpha^{\prime} \eta^{\mu \nu} \delta^{2} (z_{1} - z_{2}, \bar{z}_{1} - \bar{z}_{2}) = 0 \ \ (17)

Where, for the last line in the computation, we used the standard result,

\displaystyle  \partial \bar{\partial} \ln \mid z \mid^{2} = 2\pi \delta^{2}(z, \bar{z}) \ \ (18)

Which is derived from an application of Stokes’ theorem.

Importantly, (17) is precisely the property that Polchinski highlights in equation (2.1.23) on p.36 of his textbook. What (17) is telling us is that, on the last line, {- \pi \alpha^{\prime} \eta^{\mu \nu} \delta^{2}(z_{1} - z_{2}, \bar{z}_{1} - \bar{z}_{2})} are the quantum corrections to the classical EoM. It is also telling us that, as we want to define a product of operators that satisfy the classical EoM, we must necessarily induce normal ordering.

So what does this all mean? In order to further extend the picture being developed here, we are lead directly to a definition of OPEs.

4. Operator Product Expansions

We may now define operator product expansions. The definition follows (pp. 37-38) directly from the intuition and logic that we have so far established, notably that OPEs may be considered a direct statement about the behaviour of local operators as they approach one another. The formula for OPEs is as follows,

\displaystyle \langle \mathcal{O}_{i}(z, \bar{z})\mathcal{O}_{j}(z^{\prime}, \bar{z}^{\prime}) \rangle = \sum_{k} C_{ij}^{k}(z - z^{\prime}, \bar{z} - \bar{z}^{\prime})\langle \mathcal{O}_{k}(z^{\prime}, \bar{z}^{\prime}) \rangle \ \ (19)

Which is, again, an operator statement. This means that it also holds inside a general expectation value. Saving the general formula for OPEs until later, note that in (19) the {C_{ij}^{k}(z - z^{\prime}, \bar{z} - \bar{z}^{\prime})} should be considered as a set of functions that depend only on the separation between the two operators (i.e., there is translational invariance).

To summarise, if OPEs describe what happens when local operators approach one another other, we have already developed a sense of technical intuition for why the key idea is one of having two local operators inserted in such a way that they are situated close to one another but not at coincident points. As we have already discussed, upon insertion of local operators at {z_{1}} and {z_{2}} for example, we obtain some normal ordered product. Then, what we can do is compute their approximation by way of a string of operators at only one of the insertion points (Tong, p.69). There can be any number of operator insertions, which is of course why we have included `…’ in the general formula for the OPE (19); it denotes insertions that are not coincident at {z}. (From this point forward, the ellipse will be removed and the following statement will be implied). This leads us directly to an illustration of OPEs as provided in Polchinski’s textbook.

In figure 4.1, we see that we have a number of local operator insertions, {z_{1}} to {z_{4}}, hence what we would be computing is the expectation of 4 local operators. Given that the OPE describes the limiting behaviour of {z_{1} \rightarrow z_{2}} as a series, where the pair of operators are replaced by a single operator at {z_{2}}, one way to think about this is analogous to the Taylor series in calculus (i.e., the OPE plays a similar role in quantum field theory). In fact, the analogue of computing a Taylor series is apt, as we will see when we start computing OPEs.

Another thing to note is that the circle in the picture illustrates the radius of convergence, such that this radius is computed as the distance to the nearest other operator positioned on the circle. In CFTs, OPEs have a finite radius of convergence.

Now, from our previous discussions, and from the formal definition of OPEs, we can see quite clearly why they are always to be understood as statements which hold as operator insertions inside time-ordered correlation functions. Should one ask, ‘what are the observables in string theory?’, the answer is that we compute a set of correlation functions of local/composite operators at their insertion points. So, should we take for example the Polyakov action, {S_{P}}, and compute the correlation functions for the CFT, one motivation is to show the correlation function to be related to the scattering amplitude in 26-dimensional spacetime (in the case of the bosonic string). So, in perturbative string theory, we look at the critical theory – that is, the critical coefficients and components of the correlation function,

\displaystyle \langle A_{ij}(z_{i}\bar{z}_{j}) ... A_{ij}(z_{n}, \bar{z}_{n})) \rangle \ \ (20)

Where we are interested in the singular behaviour. Moreover, recall the definition of the normal ordered product (15). Notice that we have very interesting log behaviour. If what we want to know exactly is what will happen with the product of the two operators as {z_{1} \rightarrow z_{2}}, this implies that we have an operator singularity. As we start computing OPEs and moving forward in our study of string CFTs, it will become very clear why this singular behaviour is actually the only thing we care about.

In the next post, we will extend our discussion of OPEs. Following that, we will look to derive the Ward Identities and then turn our attention to the Virasoro algebra among other important topics in Chapter 2 of Polchinski’s textbook.

References

Joseph Polchinski. (2005). ‘String Theory: An Introduction to the Bosonic String’, Vol. 1.

David Tong. (2009). ‘String Theory’ [lecture notes].

Standard