Dimensional Regularization for One-Loop UV Divergences

Hello! Welcome back to our study of Quantum Field Theory.

In our last lesson, we took a crucial step beyond tree-level calculations and identified the fundamental one-loop diagrams in QED that produce infinite results. We saw that corrections to the electron propagator (self-energy), the photon propagator (vacuum polarization), and the QED vertex all involve integrals over an unconstrained loop momentum that diverge at high momenta (UV divergences). We concluded that to make sense of this, we need a rigorous mathematical procedure to "regularize" these integrals—a way to tame the infinity so we can systematically track and cancel it.

Today, we will learn the most powerful and widely used regularization technique in modern physics. Our learning outcome is to apply dimensional regularization to isolate UV divergences in one-loop integrals. We will see how a clever analytic continuation in the number of spacetime dimensions transforms a daunting divergent integral into a well-behaved function with a simple pole, cleanly separating the infinite part from the finite, physical part.

1. Why Not Just Cut Off the Integral?

The most intuitive way to handle an integral that diverges at high momentum, like $\int_{0}^{\infty} d k$ , is to simply impose a cutoff, $Λ$ , and integrate only up to that value: $\int_{0}^{Λ} d k$ . This is known as a momentum cutoff.

While intuitive, this method has a major flaw: it breaks fundamental symmetries. In particular, it often violates gauge invariance, the cornerstone symmetry of QED. Given your background in advanced physics, you'll appreciate that breaking a core symmetry of the theory is a significant problem, leading to incorrect physical predictions (like photons acquiring a mass).

The genius of dimensional regularization, developed by Gerard 't Hooft and Martinus Veltman (work for which they won the 1999 Nobel Prize), is that it regulates the integrals while preserving gauge invariance.

Introduction to Loop Calculations

To understand the core idea, let's start with a brief overview from the text 'Introduction to Loop Calculations' by Gudrun Heinrich.

Please read the short section 1.1, 'Dimensional regularisation'. Focus on the central concept: instead of working in 4 spacetime dimensions, we work in a general D-dimensional spacetime. The divergences that appear at D=4 will then manifest as simple poles in 1/ε, where we define D = 4 - 2ε. Note that this resource uses D = 4 - 2ε, while others use d = 4 - ε. Both conventions are common; they just rescale the ε parameter. We will mostly follow the d = 4 - ε convention in our derivations for simplicity, but it's good to be aware of both.

The key idea is analytic continuation. An integral that diverges for dimension $d = 4$ may be perfectly convergent for $d = 2$ or $d = 3$ . We treat the dimension $d$ as a continuous complex variable, evaluate the integral in a region of the complex $d$ -plane where it converges, and then analytically continue the resulting expression back to the point of physical interest, $d = 4$ .

2. The Anatomy of a One-Loop Integral

Let's see how this works in practice by calculating a generic one-loop integral. The most common one you'll encounter is the "bubble" diagram, which involves two propagators in the loop. This appears in the electron self-energy in QED or the self-energy of a scalar field in $ϕ^{4}$ theory.

The integral has the form:

I (q^{2}) = \int \frac{d ^{4} k}{( 2 π ) ^{4}} \frac{1}{[ k ^{2} - m ^{2} + i δ ] [( k - q ) ^{2} - m ^{2} + i δ ]}

Our goal is to evaluate this integral in $d$ dimensions. The process involves a few standard steps.

Step 1: Feynman Parametrization

First, we combine the two denominators into one using a clever identity known as Feynman's trick:

\frac{1}{A B} = \int_{0}^{1} d x \frac{1}{[ x A + ( 1 - x ) B ] ^{2}}

Applying this to our integral, we set $A = k^{2} - m^{2}$ and $B = (k - q)^{2} - m^{2}$ . The denominator becomes:

[x (k^{2} - m^{2}) + (1 - x) ((k - q)^{2} - m^{2})]^{2}

Expanding and rearranging this expression allows us to complete the square in the loop momentum $k$ . This is a bit of algebra, but the result is that the denominator can be written as:

[(k - q (1 - x))^{2} - Δ]^{2}

where $Δ = - q^{2} x (1 - x) + m^{2}$ .

Step 2: Momentum Shift and Wick Rotation

Now we are integrating over an expression that depends on $(k - q (1 - x))^{2}$ . Since the integration is over all $k$ , we can shift the integration variable to $l = k - q (1 - x)$ . The integral measure $d^{d} k$ becomes $d^{d} l$ , and our integral simplifies to:

I (q^{2}) = \int_{0}^{1} d x \int \frac{d ^{d} l}{( 2 π ) ^{d}} \frac{1}{[ l ^{2} - Δ + i δ ] ^{2}}

The next step is to move from Minkowski space to Euclidean space, where the momentum squared is $l_{E}^{2} = l_{1}^{2} + l_{2}^{2} + \dots + l_{d}^{2}$ . This is done via a Wick rotation, where we set the time-component of the momentum $l_{0} = i l_{d}$ . This changes $l^{2} = l_{0}^{2} - l^{2}$ to $- (l_{d}^{2} + l^{2}) = - l_{E}^{2}$ . The integral becomes:

I (q^{2}) = i \int_{0}^{1} d x \int \frac{d ^{d} l _{E}}{( 2 π ) ^{d}} \frac{1}{[ l _{E}^{2} + Δ - i δ ] ^{2}}

The factor of $i$ comes from the transformation of the integration measure.

Step 3: Integration in $d$ Dimensions

The integrand is now spherically symmetric in $d$ -dimensional Euclidean space. We can switch to spherical coordinates:

d^{d} l_{E} = l_{E}^{d - 1} d l_{E} d Ω_{d}

where $l_{E}$ is the radial coordinate and $d Ω_{d}$ is the solid angle element in $d$ dimensions. The integral over the solid angle gives the surface area of a unit sphere in $d$ dimensions, which has the general form:

\int d Ω_{d} = Ω_{d} = \frac{2 π ^{d /2}}{Γ ( d /2 )}

Here, $Γ (z)$ is the Euler Gamma function, a generalization of the factorial. You can think of this as a known result.

The radial integral is now:

\int_{0}^{\infty} d l_{E} \frac{l _{E}^{d - 1}}{( l _{E}^{2} + Δ ) ^{2}}

With a substitution $u = l_{E}^{2}$ , this integral can be expressed in terms of Gamma functions. The final result for the full momentum integral is a master formula. The following resource provides a table of these master formulas and walks through the derivation in a very pedagogical way.

7.3 Dimensional regularization

The document '7.3 Dimensional regularization' from a JLab lecture series provides an extremely clear, step-by-step derivation of a general dimensionally regularized integral. It also contains a handy table of the most common integral formulas.

Please study the following sections: Section from 'First, let us Wick rotate...' to '...makes sense for non-integer d as well as integer d.' (pages 202-204): This shows the full derivation of the volume of a d-dimensional sphere, a beautiful result worth understanding. Section from 'Returning to our integral Id(q,n)...' to the formula (7.2) (pages 204-205): This completes the evaluation of the integral in terms of Gamma functions. The table of formulas (7.3) - (7.6) on page 206: These are the key results we will use in practice. Pay close attention to formula (7.3), which corresponds to our integral with k=0 and an arbitrary power 'n' in the denominator.

From formula (7.3) in the resource with $n = 2$ , our integral becomes:

\int \frac{d ^{d} l}{( 2 π ) ^{d}} \frac{1}{( l ^{2} - Δ ) ^{2}} Wick i \int \frac{d ^{d} l _{E}}{( 2 π ) ^{d}} \frac{1}{( l _{E}^{2} + Δ ) ^{2}} = \frac{i}{( 4 π ) ^{d /2}} \frac{Γ ( 2 - d /2 )}{Γ ( 2 )} (Δ)^{d /2 - 2}

3. Isolating the Divergence

Now we have our result, expressed as a function of the dimension $d$ . Let's see what happens when we approach the physical dimension, $d = 4$ . We set $d = 4 - ϵ$ , where $ϵ$ is a small parameter that we will eventually take to zero.

The key term is the Gamma function: $Γ (2 - d /2)$ .

Γ (2 - d /2) = Γ (2 - (4 - ϵ) /2) = Γ (2 - 2 + ϵ /2) = Γ (ϵ /2)

The Gamma function $Γ (z)$ has simple poles at $z = 0, - 1, - 2, \dots$ . Near $z = 0$ , it has the expansion:

Γ (z) \approx \frac{1}{z} - γ_{E} + O (z)

where $γ_{E} \approx 0.577$ is the Euler-Mascheroni constant.

Therefore, for small $ϵ$ , our crucial term becomes:

Γ (ϵ /2) \approx \frac{2}{ϵ} - γ_{E} + O (ϵ)

The original ultraviolet divergence of the integral has been converted into a simple pole in $1/ ϵ$ !

Let's assemble the full expression for $I (q^{2})$ for $d = 4 - ϵ$ :

I (q^{2}) = \frac{i}{( 4 π ) ^{2 - ϵ /2}} \int_{0}^{1} d x Γ (ϵ /2) (Δ)^{- ϵ /2}

Expanding for small $ϵ$ :

$(4 π)^{- (2 - ϵ /2)} = \frac{1}{16 π ^{2}} (4 π)^{ϵ /2} = \frac{1}{16 π ^{2}} (1 + \frac{ϵ}{2} ln (4 π) + \dots)$
$Δ^{- ϵ /2} = e^{- \frac{ϵ}{2} l n Δ} = 1 - \frac{ϵ}{2} ln Δ + \dots$
$Γ (ϵ /2) = \frac{2}{ϵ} - γ_{E} + \dots$

Multiplying these out and keeping terms up to $O (ϵ^{0})$ , we get:

I (q^{2}) \approx \frac{i}{16 π ^{2}} \int_{0}^{1} d x (\frac{2}{ϵ} - γ_{E} + ln (4 π) - ln (Δ))

This expression is the main result. It has two parts:

A divergent part: $\frac{i}{16 π ^{2}} \frac{2}{ϵ}$
A finite part: $\frac{i}{16 π ^{2}} \int_{0}^{1} d x (- γ_{E} + ln (4 π) - ln (\frac{m ^{2} - q ^{2} x ( 1 - x )}{μ ^{2}}))$

Wait, where did the $μ^{2}$ come from? This is a crucial subtlety. The original coupling constant (e.g., $λ$ in $ϕ^{4}$ theory) is dimensionless only in $d = 4$ . In $d \neq = 4$ dimensions, it acquires a mass dimension. To keep the coupling dimensionless for any $d$ , we introduce an arbitrary mass scale $μ$ and write the coupling as $g \to g μ^{(4 - d) /2}$ . This $μ$ naturally appears in the calculation to keep the arguments of logarithms dimensionless. Our final expression for the finite part is then:

I_{finite} (q^{2}) = \frac{i}{16 π ^{2}} \int_{0}^{1} d x (- γ_{E} + ln (4 π) - ln (\frac{m ^{2} - q ^{2} x ( 1 - x )}{μ ^{2}}))

We have successfully isolated the UV divergence as a clean $1/ ϵ$ pole.

Test your understanding!

Using the master formula from the resource LINK (specifically, eq. 7.3), evaluate the integral
$I^{'} = \int \frac{d ^{d} p}{( 2 π ) ^{d}} \frac{1}{( p ^{2} - Δ ) ^{3}}$
in $d = 4 - ϵ$ dimensions and identify the leading term as $ϵ \to 0$ . Is this integral divergent or finite?

Show answer

The master formula is:
$\int \frac{d ^{d} p}{( 2 π ) ^{d}} \frac{1}{( p ^{2} - Δ ) ^{n}} = \frac{i ( - 1 ) ^{n}}{( 4 π ) ^{d /2}} \frac{Γ ( n - d /2 )}{Γ ( n )} (Δ)^{d /2 - n}$
Here, we have $n = 3$ . Plugging this in:
$I^{'} = \frac{i ( - 1 ) ^{3}}{( 4 π ) ^{d /2}} \frac{Γ ( 3 - d /2 )}{Γ ( 3 )} (Δ)^{d /2 - 3} = \frac{- i}{( 4 π ) ^{d /2}} \frac{Γ ( 3 - d /2 )}{2} (Δ)^{d /2 - 3}$
Now, let $d = 4 - ϵ$ . The Gamma function argument is:
$3 - d /2 = 3 - (4 - ϵ) /2 = 3 - 2 + ϵ /2 = 1 + ϵ /2$
As $ϵ \to 0$ , this argument goes to 1. Since $Γ (1) = 1$ , $Γ (1 + ϵ /2)$ is finite and approaches 1.
$lim_{ϵ \to 0} I^{'} = \frac{- i}{( 4 π ) ^{2}} \frac{Γ ( 1 )}{2} (Δ)^{- 1} = \frac{- i}{32 π ^{2} Δ}$
The result is finite. This makes sense from power counting: the integral goes like $\int k^{3} d k / k^{6} \sim \int d k / k^{3}$ , which converges in 4D. Dimensional regularization correctly confirms this.

Conclusion

In this lesson, we have developed the primary tool for handling divergent integrals in modern quantum field theory. We've moved from a state of paralysis in the face of infinities to having a concrete, systematic algorithm for taming them.

Key Takeaways:

Dimensional Regularization (DR) is a technique that regulates UV divergent integrals by analytically continuing the dimension of spacetime from $d = 4$ to $d = 4 - ϵ$ .
Its primary advantage over simpler methods like a momentum cutoff is that it preserves crucial symmetries, most notably gauge invariance.
The standard procedure involves Feynman parametrization, momentum shifting, Wick rotation, and evaluating the resulting integral in $d$ -dimensional spherical coordinates.
This procedure expresses the integral in terms of Euler Gamma functions, $Γ (z)$ .
The UV divergences of the original integral manifest as simple poles of the form $1/ ϵ$ as we take the limit $d \to 4$ .
The process cleanly separates the integral into a divergent piece (proportional to $1/ ϵ$ ) and a finite piece that depends on the physical kinematics and an arbitrary mass scale $μ$ .

Preview of the Next Lesson:
We are now fully equipped to perform a real one-loop calculation in QED. In the next lesson, we will apply the machinery of dimensional regularization to the vacuum polarization diagram. We will calculate its divergent and finite parts, and in doing so, we will see how the need to absorb the $1/ ϵ$ pole leads directly to the concept of a running coupling constant—the idea that the strength of the electromagnetic force depends on the energy at which it is measured. This is one of the most profound and experimentally verified predictions of QFT.

Recommended resources