I describe the treatment of gravity as a quantum effective field theory. This allows a natural separation of the (known) low energy quantum effects from the (unknown) high energy contributions. Within this framework, gravity is a well behaved quantum field theory at ordinary energies. In studying the class of quantum corrections at low energy, the dominant effects at large distance can be isolated, as these are due to the propagation of the massless particles (including gravitons) of the theory and are manifested in the nonlocal/nonanalytic contributions to vertex functions and propagators. These leading quantum corrections are parameterfree and represent necessary consequences of quantum gravity. The methodology is illustrated by a calculation of the leading quantum corrections to the gravitational interaction of two heavy masses.
UMHEP408 grqc/9405057
1 Introduction
We are used to the situation where our theories are only assumed to be provisional. They have been tested and found to be valid over a limited range of energies and distances. However, we do not know that they hold in more extremes situations. There are many examples of theories which have been superseded by new theories at higher energies, and we expect this process to continue. It is interesting to look at the incompatibility of general relativity and quantum mechanics in this light. It would not be surprising if there are new ingredients at high energy in order to have a satisfactory theory of quantum gravity. However, are there any conflicts between gravity and quantum mechanics at the energy scales that are presently accessible? If there are, it would be a major concern because it would mean our present theories are wrong in ways which cannot be blamed on new physics at high energy.
There is an apparent technical obstacle to addressing the compatibility of quantum mechanics and gravity at present energies, i.e., the non renormalizability of quantum gravity. Quantum fluctuations involve all energy scales, not just the energy of the external particles. Perhaps our lack of knowledge of the true high energy theory will prevent us from calculating quantum effects at low energy. In the class of renormalizable field theories, low energy physics is shielded from this problem because the high energy effects occur only in the shifting of a small number of parameters. When these parameters are measured experimentally, and results expressed in terms the measured values, all evidence of high energy scales disappears or is highly suppressed.[1] However in some nonrenormalizable theories, the influence of high energy remains. For example in the old Fermi theory of weak interactions, the ratio of the neutron decay rate to that of the muon has a contribution which diverges logarithmically at one loop. It is not the divergence itself which is the problem, as the ratio becomes finite in the Standard Model (with a residual effect of order ). More bothersome is the sensitivity to the high energy theory–the low energy ratio depends on whether the scale of the new physics is or .
However quantum predictions can be made in nonrenormalizable theories. The techniques are those of effective field theory, which has been assuming increasing importance as a calculational methodology. The calculations are organized in a systematic expansion in the energy. Effects of the high energy theory again appear in the form of shifts in parameters which however are determined from experiment. To any given order in the energy expansion there are only a finite number of parameters, which can then be used in making predictions. Using the techniques of effective field theory, it is easy to separate the effects due to low energy physics from that of the (unknown) high energy theory. Indeed, even the phrasing of the question raised in the opening paragraph is a byproduct of the way of thinking about effective field theory.
General relativity fits naturally into the framework of effective field theory. The gravitational interactions are proportional to the energy, and are easily organized into an energy expansion. The theory has been quantized on smooth enough background metrics.[2,3,4] We will explore quantum gravity as an effective field theory and find no obstacle to its successful implementation.
In the course of our study we will find a class of quantum predictions which are parameter free (other than Newton’s constant G) and which dominates over other quantum predictions in the low energy limit. These ’leading quantum corrections’ are the first modifications due to quantum mechanics, in powers of the energy or inverse factors of the distance. Because they are independent of the eventual high energy theory of gravity, depending only on the massless degrees of freedom and their low energy couplings, these are true predictions of quantum general relativity.
The plan of the paper is as follows. In Section 2, we briefly review general relativity and its quantization. Section 3 is devoted to the treatment of general relativity as an effective field theory, while the leading quantum corrections are described in more detail in Section 4. We give more details of a previously published example,[5] that of the gravitational interaction around flat space, in Section 5. Some speculative comments on the extreme low energy limit, where the wavelength is of order the size/life time of the universe are given in Section 6 with concluding comments in Section 7. An appendix gives some of the non analytic terms needed for the leading quantum corrections arising in loop integrals.
2 General Relativity and its Quantization
In this paper the metric convention is such that flat space is represented by .[6,7] The Einstein action is
(1) 
where is the metric tensor and .
(2) 
Heavy spinless matter fields interact with the gravitational field as described by the action
(3) 
The quantum fluctuations in the gravitational field may be expanded about a smooth background metric , with one possible choice being
(4) 
Indices here and in subsequent formulas are raised and lowered with the background metric. In order to quantize the field , we need to fix the gauge. Following ’t Hooft and Veltman[3] this entails a gauge fixing term
(5) 
with , and with the semicolon denoting covariant differentiation on the background metric. The ghost Lagrangian is
(6) 
for the complex FaddeevPopov ghost field .
The expansion of the Einstein action takes the form [3,8]
(7) 
where
(8)  
A similar expansion of the matter action yields
(9) 
with
(10)  
If the background metric satisfies Einstein’s equation
(11) 
the linear terms in , will vanish.
In calculating quantum corrections at one loop, we need to consider the Lagrangian to quadratic order
(12) 
The integration over the gravitational degrees of freedom
(13) 
yields a functional which in general is nonlocal and also divergent. The identification of the quantum degrees of freedom and the definition of a quantum theory is no more difficult than the quantization of Yang Mills theory, at least for small quantum fluctuations. The difficulties arise in applying this result. Because of the dimensionful coupling and the nonlinear nature of the theory, divergencies appear in places which cannot be absorbed into the basic parameters introduced this far. Since the coupling grows with energy, the theory is strongly coupled at very high energy, , and badly behaved in perturbation theory. We also do not have known techniques for dealing with large fluctuations in the metric, which may in principle be topologychanging in nature. However, the low energy fluctuations are very weakly coupled, with a typical strength for . Since small quantum fluctuations at ordinary energies behave normally in perturbation theory, it is natural to try to separate these quantum corrections from the problematic (and most likely incorrect) high energy fluctuations. Effective field theory is the tool to accomplish this separation.
3 Gravity as an Effective Field Theory
Effective field theory techniques[9,10] have become common in particle physics. The method is not a change in the rules of quantum mechanics, but is rather a procedure which organizes the calculation and separates out the effects of high energy from the known quantum effects at low energy. General relativity is a field theory which naturally lends itself to such a treatment. Perhaps the known manifestation of a effective field theory which is closest in style to gravity is chiral perturbation theory,[10] representing the low energy limit of QCD. Both are nonlinear, nonrenormalizable theories with a dimensionful coupling constant. If the pion mass were taken to zero, as can be easily achieved theoretically, long distance effects similar to those from graviton loops would be found. In addition we have had the benefit of detailed calculations and experimental verification in the chiral case, so that the workings of effective field theory are transparent. In this section, I transcribe the known properties of effective field theory to the gravitational system.
The guiding principle underlying general relativity is that of gauge symmetry, i.e., the local invariance under coordinate transformations. This forces the introduction of geometry, and requires us to define the action of the theory using quantities invariant under the general coordinate transformations. However this is not sufficient to completely define the theory, as many quantities are invariant. For example, each term in the action
(14) 
(where are constants and the ellipses denote higher powers of and ), are separately invariant under general coordinate transformations. Other physics principles must enter in order to simplify the action. For example, the constant is proportional to the cosmological constant , which experiment tells us is very small.[11] We therefore set (the renormalized value of) for the rest of this paper. Experiment tells us very little about the dimensionless constants , bounding ,[12] and coefficients of yet higher powers of R have essentially no experimental constraints. Einstein’s theory can be obtained by setting as well as forbidding all higher powers. However it is very unlikely that in fact . For example, quantum corrections involving the known elementary particles (whether or not gravity itself is quantized) will generate corrections to etc. Unless an infinite number of ”accidents” occur at least some of the higher order coefficients will be nonzero.
In practice there is no known reason to require that vanish completely. We can instead view the gravitational action as being organized in an energy expansion, and then reasonable values of do not influence physics at low energies. In order to set up the energy expansion, we note that the connection is first order in derivatives and the curvature is second order. When matrix elements are taken, derivitives turn into factors of the energy or momentum , so that the curvature is said to be of order . Terms in the action involving two powers of the curvature are of order . The graviton energy can be arbitrarily small and at low enough energies terms of order are small compared to those of order . The higher order Lagrangians will have little effect at low energies compared to the Einstein term, . This is why the experimental bounds on are so poor; reasonable values of give modifications which are unmeasurably feeble. In a pure gravity theory the expansion scale might be expected to be the Planck mass . For example, if we just consider the and terms in the Lagrangian, Einstein’s equation is modified to
(15) 
Unless or the influence of the terms is negligible.
In the literature[13] there are discussions of problems with theories. These include negative metric states, unitarity violation, an inflationary solution, and an instability of flat space. However J. Simon[14] has shown that these problems do not appear when the theory is treated as an effective field theory. Essentially, the problems arise from treating the description (without any higher terms) as a fundamental theory at high energy when the curvature is of order the Planck mass squared. Then the contribution is comparable to that of . Of course then yet higher powers of would also be comparable to the and terms, so that in this region we would not be able to say anything about the full expansion. In the low energy region the effect of is just a small correction to the behavior of the pure Einstein theory and no bad behavior is introduced.
The most general gravitational action will have an infinite number of parameters such as . At the lowest energy, only is important. However we can imagine in principle that we could do experiments with such high precision that we could also see the first corrections and measure . If we knew the ultimate correct theory of gravity, we might be able to predict . With our incomplete knowledge at low energy, we must treat them as free parameters. Quantum effects from both high energy and low energy particles have the potential to produce shifts in and it is the final (renormalized) value which experiments would determine.
It is crucial to differentiate the quantum effects of heavy particles from those of particles which are massless. The issue is one of scale. Virtual heavy particles cannot propagate long distances at low energies; the uncertainty principle gives them a range . On distance scales much larger than this, their effects will look local, as if they were described by a local Lagrangian. Even the slight nonlocality can be accounted for by Taylor expanding the interactions about a point. In a simple example, a particle propagator can be Taylor expanded
(16) 
In the coordinate space propagator obtained by a Fourier transform, the constant generates a delta function, hence a local interaction and the factors of are replaced by derivatives in a local Lagrangian. The quantum effects of virtual heavy particles then appear as shifts in the coefficients of most general possible local Lagrangian.
On the other hand, the quantum effects of massless particles cannot all be accounted for in this way. Some portions of their quantum corrections, for example the results of high energy propagation in loop integrals, do shift the parameters in the Lagrangian and are local in that respect. However the low energy manifestations of massless particles are not all local, as the particles can propagate for long distances. A simple example is again the propagator, now , which cannot be Taylor expanded about . The low energy particles (massless ones or ones whose mass is comparable to or less than the external energy scale) cannot be integrated out of the theory but must be included explicitly in the quantum calculations.
Our direct experience in physics covers distances from to cosmic distance scales. Although gravity is not well tested over all of those scales, we would like to believe that both general relativity and quantum mechanics are valid in that range, with likely modifications coming in at . For reasons discussed more fully in Section 6, I would like to imagine quantizing the theory in a very large, but not infinite, box. Roughly speaking, this is to avoid asking questions about wavelengths of order the size of the Universe, i.e., reaching back to the Big Bang singularity. However, the volume is taken large enough that we may ignore edge effects. We assume that any particles which enter this quantization volume (e.g., the remnants of the Big Bang) are either known or irrelevant. The curvature is assumed to be small and smooth throughout this space time volume. This situation then represents the limits of our ”known” confidence in both general relativity and quantum mechanics, and we would like to construct a gravitational effective field theory (GEFT) in this region.
The dynamical information about a theory can be obtained from a path integral. The results of the true theory of gravity will be contained in a generating functional
(17) 
where (gravity) represents the fields of a full gravity theory, represents matter fields and can be a set of source fields added to the Lagrangian (i.e., ) in order to allow us to probe the generating functional. The gravitational effective field theory is defined to have the same result
(18) 
Here is constructed as the most general possible Lagrangian containing and consistent with general covariance. It contains an infinite number of free parameters, such as described above. The effects of the high energy part of the true theory are accounted for in these constants. However, as discussed above, the low energy degrees of freedom must be accounted for explicitly, hence their inclusion in the path integral. Since we are only interested in the small fluctuation and low energy configurations of , we do not need to address issues of the functional measure for large values of . Any measure and regularization scheme which does not violate general covariance may be used. Because the coupling of the low energy fluctuations is so very weak, the path integral has a well behaved perturbative expansion. We have implicitly assumed that gravitons are the only low energy particles which are remnants of the full gravity theory. If any other massless particles result, they would also need to be included.
The most general effective Lagrangian will contain both gravitational and matter terms and will written in a derivative expansion
(19) 
The general gravitational component has already been written down
(20) 
The first two terms in the general matter Lagrangian for a massive field are
(21) 
Note that derivatives acting on a massive matter field are not small quantities–the ordering in the derivative expansion only counts derivatives which act on the light fields, which in this case is only the gravitons. In contrast, if the matter field were also massless, the ordering in energy would be different
(22) 
where the overbar is meant to indicate the parameters in the massless theory. For , we have the ”improved” action of Callen, Coleman and Jackiw[15]; however any value of is consistent with general covariance and the energy expansion. Note that since we are only working to , we may use the lowest order equations of motion to simplify .
Let us now discuss the expected size of the parameters in the effective action. Those parameters which are accessible to realistic measurements have been labeled by their conventional names. From the standpoint of the energy expansion it is of course a great shock that the renormalized value of the cosmological constant is so small . The cosmological bound is ,[11] where as the standard expectation would be a value 40 to 60 orders of magnitude larger. Effective field theory has nothing special to say about this puzzle. However it does indicate that at ordinary scales, is unimportant and that the energy expansion for gravity starts at two derivatives with . The constants are dimensionless. They determine the scale of the energy expansion of pure gravity which in general is
(23) 
We of course have no direct experience with this scale, but the expectation that is of order of the Planck mass would correspond to . The phenomenological bound [12], , is of course nowhere close to being able to probe this possibility.
For the constants in the matter Lagrangian, , the expectations are a bit more complicated, as we must distinguish between point particles which have only gravitational interactions and those which have other interactions and/or a substructure. The constants have dimension , and we will see by explicit calculation in Section V that they contribute to the form factors in the energymomentum vertex for the particle, being equivalent to the charge radius in the well known electromagnetic form factor. Loop diagrams involving gravitons shift by terms of order . In the absence of interactions other than gravity, it is consistent to have . However, for particles that have other interactions, the energy and momentum will be spread out due to quantum fluctuations and a gravitational charge radius will result. The expectation in this case is
(24) 
For composite particles, such as the proton, this will be of size of the physical radius, . For interacting point particles, such as the electron, this would be of order the scale of quantum correction .
The gravitational effective field theory is a full quantum theory and loop diagrams are required in order to satisfy general principles, such as unitarity. The finite infrared part of loop diagrams will be discussed more fully in the next section. Also obtained in the usual calculation of many loop diagrams are ultraviolet divergences. These arise in a region where the low energy effective theory may not be reliable, and hence the divergences may not be of deep significance. However as a technical matter they must be dealt with without influencing low energy predictions. The method is known from explicit calculations of the divergences in gravity [3,4,16,17], and from general effective field theory practice. The divergences are consistent with the underlying general covariance, and must occur in forms similar to terms in the most general possible effective Lagrangian. They can then be absorbed into renormalized values for the unknown coefficients which appear in this general Lagrangian. Moreover, it can be shown that for loops involving low order terms in the energy expansion, that the renormalization involves the coefficients which appear at higher order.
In a background field method, one can compactly summarize the one loop quantum corrections. The classical background field is determined by the matter fields and sources by the equations of motion
(25) 
The effective action is thus rendered into quadratic form in ,
(26) 
where
(27)  
with being a covariant derivative and
(28) 
Formally integrating over one finds
(29) 
While some of the finite portions of Z are difficult to extract (see the next section), since is in general a nonlocal functional, the divergences are local and are readily calculated. One loop divergences are known for gravity coupled to matter fields and the two loop result has been found for pure gravity. At one loop, the divergences due to gravitons have the form [3]
(30) 
where within dimensional regularization. This produces the following minimal subtraction renormalization of the gravitational parameters
(31)  
At two loops, the divergence of pure gravity is[16]
(32) 
The key feature is that higher powers of are involved at higher loops. This is a consequence of the structure of the energy expansion in a massless theory. A simple example can illustrate the essentials of this fact. Consider a four graviton vertex, Fig. 1a. Since each graviton brings a factor of (see Eq. 4 and recall that ) the Einstein action gives this a behavior
(33) 
where is representative of the external momentum, whereas the Lagrangian at order have the behavior.
(34) 
If we use two of the Einstein vertices in a loop diagram, Fig. 1b, the momenta could be either external or internal, for example
(35) 
If we imagine that the divergent integrals are regularized by dimensional regularization (which preserves the general covariance and which only introduces new scale dependent factors in logarithms not in powers), the Feynman integral must end up being expressed in terms of the external momenta. Because no masses appear for gravitons, the momentum power of the final diagrams can be obtained easily by counting powers of . The result is a divergence at order and can be absorbed into . Loop diagrams involve more gravitons than tree diagrams, hence more factors of , which by dimensional analysis means that they are the same structure as higher order terms in the energy expansion.
If we were to attempt a full phenomenological implementation of gravitational effective field theory at one loop order, the procedure would be as follows:

Define the quantum degrees of freedom using the lowest order
effective Lagrangian, as done in Section 2.

Calculate the one loop corrections.

Combine the effects of the order and Lagrangians (given earlier in this section) at tree level with the one loop corrections. The divergences (and some accompanying finite parts) of the loop diagrams may be absorbed into renormalized coefficients of the Lagrangian , using some renormalization scheme which does not violate general covariance.

Measure the unknown coefficients by comparison with some experimental measurement.

Having determined the parameters of the theory, one can make predictions for other experimental observables, valid to in the energy expansion.
In practice the difficulty arises at step 4: there is no observable that I am aware of which is sensitive to reasonable values of any of the coefficients. However the low energy content of the gravitational effective field theory is not just contained in these parameters. There is a distinct class of quantum corrections, uncovered in the above procedure, which are independent of the unknown coefficients. Moreover this class, the ”leading quantum corrections”, are generally dominant at large distances over the other one loop gravitational corrections. These are discussed more fully in the next section.
4 Leading Quantum Corrections
Although the ultraviolet behavior of quantum gravity has been heavily studied to learn about the behavior of general relativity as a fundamental theory, from the standpoint of effective field theory it is rather the infrared behavior which is more interesting. In the last section, the renormalization of the parameters in the effective action was described. Although a technical necessity, this has no predictive content. However the low energy propagation of massless particles leads to long distance quantum corrections which are distinct from the effects of the local effective Lagrangian.
A crucial distinction in this regard is whether the effective action may be expanded in a Taylor expansion in the momentum (or equivalently in powers of derivatives). If the result is analytic, it may be represented by a series of local Lagrangians with increasing powers of derivatives. However nonanalytic effects cannot be equivalent to local contributions, and hence are unmistakable signatures of the low energy particles. Moreover, the nonanalytic effects can be dominant in magnitude over analytic corrections. The expansion of the gravitational action is in powers of so that the first two terms of a matrix element will be
(36) 
As we will see in the next section, a graviton loop will have a logarithmic nonanalytic modification around flat space.
(37) 
When massive matter fields are included in loops with gravitons we may also have nonanalytic terms of the form instead of the logarithm.[18] Both of these effects have the property that they pick up imaginary components for timelike values of (i.e., for in this metric), as they are then part of the loop diagrams which are required for the unitarity of the matrix. The imaginary pieces arise from the rescattering of onshell intermediate states, and can never contained in a local Lagrangian. In addition, since can become very small, the nonanalytic pieces will satisfy and at low energy, thereby dominating over the analytic effect. We can see that the nonanalytic terms have a distinct status as the leading quantum corrections due to long distance effects of massless particles.
The leading nonanalytic effects have the extra advantage that they involve only the massless degrees of freedom and the low energy couplings of the theory, both of which are known independent of the ultimate high energy theory. The massless particles are the gravitons, photons and maybe neutrinos. Only the lowest energy couplings are needed, since higher order effects at the vertices introduce more powers of . The low energy couplings are contained in the Einstein action and only depend on the gravitational constant . So in distinction to the analytic contributions, which depend on the unknown parameters , the leading quantum corrections are parameter free.
Although our prime interest above has been the quantum corrections within the gravitational part of the theory, we note that similar comments can be made if interactions other than just gravity are present. For example a theory with massless particles, such as photons in QED, can also generate nonanalytic behavior in loop amplitudes when the photons are coupled to gravity. Let us call these Class II nonanalytic corrections as compared to the Class I nonanalytic effects found due to the quantum behavior of gravitons. In addition there is a district type of quantum predictions (Class III) which may also be predictions of the low energy theory once we allow interactions in addition to gravity. This occurs for analytic terms in the energy expansion which are accompanied by a parameter with dimension . The parameters in Eq. 21 are examples. The low energy theory can generate contributions to the parameter with inverse powers of a light mass, while the AppelquistCarazonne theorem[1] tells us that the effects of a high energy theory would generally produce inverse powers of a heavy mass. Therefore the low energy contribution can be dominant, and the uncertainty caused by unknown high energy theory is minimal. In the case of the parameters, if the particles were strongly interacting QCD would generate a gravitational charge radius corresponding to , which would be unlikely to be changed by the underlying quantum theory of gravity. Other examples in the case of QED plus gravity have been worked out by Behrends and Gastmans.[18] Classes II and III corrections (if present) are most often larger than the gravitational leading quantum corrections (Class I) because their form need not be expansions in the small quantity . [An exception is the gravitational potential at large distance, where analytic corrections have no effect on the term.]
5 Example: The Gravitational Potential
In this section I describe in detail an example which demonstrates the extraction of the leading quantum corrections. The gravitational interaction of two heavy objects close to rest described in lowest order by the Newtonian potential energy
(38) 
This is modified in general relativity by higher order effects in and by nonlinear terms in the field equations of order (which are of the same order as ). While a simple potential is not an ideal relativistic concept, the general corrections would be of the form[7]
(39) 
The number ”a” would depend on the precise definition of the potential and would be calculable in the PostNewtonian expansion. At some level there will be quantum corrections also. By dimensional analysis, we can figure out the form that these should take. Since they arise from loop diagrams, they will involve an extra power of , and if they are quantum corrections they will be at least linear in . If the effects are due to long range propagation of massless particles, the other dimensionful parameter is the distance . The combination
(40) 
is dimensionless and provides an expansion parameter for the long distance quantum effects. We then expect a modification to the potential of the form
(41) 
and our goal is to calculate the number b for an appropriate definition of the potential.[5]
The Newtonian potential can be found as the nonrelativistic limit of graviton exchange, Fig. 2. In the harmonic gauge, the graviton propagator is
(42) 
with
(43) 
with being the flat space matrix . The matter stress tensor has the onshell matrix element
(44) 
in our normalization convention
(45) 
[Here the subscript 0 indicates that this form holds before the inclusion of radiative corrections.] Graviton exchange then yields
(46) 
The static limit corresponds to and
(47) 
where the accounts for the covariant normalization factor. The Newtonian potential is then found from the Fourier transform of
(48) 
which in coordinate space yields
(49) 
Of course, despite the description of quantizing, gauge fixing etc., this is purely a classical result.
In order to define a quantum potential one can consider the set of one particle reducible graphs of Fig. 3, where the heavy dots signify the full set of radiative corrections to the vertex function and the graviton propagator. These corrections are given explicitly in Fig. 4, 5. It is this set which we will examine. Fortunately we will be able to extract the information that we need for the vacuum polarization from the work of others. This leaves the vertex correction to be worked out here.
The vertices required for the calculation follow from the Lagrangians given previously. For the vertices pictured in Fig. 6, we find
(50) 
and
(51)  
where
(52) 
The graviton vertex is found most easily by using Eq. 8 plus Eq. 5 with the background metric being expanded as , where we pick out the vertex with one external field and two quantum fields. After some work, this can be put into the form
(53)  
with defined in Eq. 52 and defined in Eq. 42.
The diagrams involved for the vertex correction are given in Fig. 4. We will argue below that Fig. 4d,e,f will not contribute to the leading quantum corrections, so that we need to calculate only diagrams 4b and 4c. These have the form
(54) 
Before proceeding, it is worth examining the structure of the answer. The general vertex may be described by two form factors
(55)  
with normalization condition . The expansion in energy corresponds to an expansion of the form factors in powers of . The one loop diagrams of Fig. 6 have an extra power of compared to the tree level vertex, and has dimensions , so that and form dimensionless combinations. However loop diagrams will also produce non analytic terms with the form and , which also are dimensionless. Also contributing to the form factors are the terms in the higher order Lagrangian as these give extra factors of . By working out these contributions and taking the general form of the loop diagrams from dimensional considerations we obtain form factors
(56) 
where are numbers which come from the computation of the loop diagrams. There can be no corrections in of the form because of the normalization condition . The constant can be chosen arbitrarily, with a corresponding shift in the constants and . The ellipses denote higher powers of . The constants and will in general be divergent, while and must be finite. For timelike, and pick up imaginary parts which correspond to the physical (on shell) intermediate states as described by unitarity. Recall that represents the unknown effects of the true high energy theory, while and come largely from the high energy end of the loop integrals. For these high energies we have no way of knowing if the loop integrals are well represented by the low energy vertices and low energy degrees of freedom–almost certainly they are not. Therefore it is logical, as well as technically feasible, to combine