Skip to content

Appendix D Solutions

← Back to problems　|　Back to chapter

Table of Contents

Basic

Medium

Advanced

Basic

B-1. Determining Mass Dimension (Yukawa Interaction)

→ Back to problem

Solution Strategy

From the mass dimension of the Lagrangian density \([\mathcal{L}] = 4\), we determine \([g]\) by summing the dimensions of each factor in the interaction term.

Calculation

The Yukawa interaction term is:

\[ \mathcal{L}_{\text{int}} = -g\,\bar{\psi}\psi\,\phi \]

The mass dimensions of each field, from equations (D.3) and (D.4) in the text, are:

  • \([\psi] = [\bar{\psi}] = 3/2\)
  • \([\phi] = 1\)

From \([\mathcal{L}_{\text{int}}] = 4\):

\[ [g] + [\bar{\psi}] + [\psi] + [\phi] = [g] + \frac{3}{2} + \frac{3}{2} + 1 = [g] + 4 = 4 \]

Final Answer

\[ \boxed{[g] = 0} \]

The Yukawa coupling constant is dimensionless.

Consistency Check

The QED coupling constant \(e\) also has \([e] = 0\) (see the text). The Yukawa interaction \(\bar{\psi}\psi\phi\) has a total field dimension of \(3/2 + 3/2 + 1 = 4\), which has the same structure as QED's \(\bar{\psi}\gamma^\mu\psi A_\mu\) with \(3/2 + 3/2 + 1 = 4\). Both are renormalizable interactions, and the fact that their coupling constants are dimensionless is consistent with this.

B-2. Determining Mass Dimension (Scalar Field in 6-Dimensional Spacetime)

→ Back to problem

Solution Strategy

In \(d\)-dimensional spacetime, \([d^d x] = -d\) and \([S] = 0\) imply \([\mathcal{L}] = d\). We determine \([\phi]\) from the kinetic term and substitute into the \(\phi^3\) interaction.

Calculation

Determining \([\phi]\):

In \(d\) dimensions, \([\mathcal{L}] = d\). Focusing on the kinetic term \(\frac{1}{2}(\partial_\mu\phi)(\partial^\mu\phi)\):

\[ [\partial_\mu]^2 + [\phi]^2 = 2 + 2[\phi] = d \]
\[ [\phi] = \frac{d-2}{2} \]

For \(d = 6\):

\[ [\phi] = \frac{6-2}{2} = 2 \]

Determining \([g]\):

From \(\mathcal{L}_{\text{int}} = -\frac{g}{3!}\phi^3\):

\[ [g] + 3[\phi] = [g] + 6 = d = 6 \]
\[ [g] = 0 \]

Final Answer

\[ \boxed{[\phi] = 2, \qquad [g] = 0} \]

In six-dimensional spacetime, the coupling constant of \(\phi^3\) theory is dimensionless, making it a renormalizable theory.

Verification

In four dimensions, \([\phi] = 1\) and the \(\phi^3\) coupling constant has \([g] = 4 - 3 \times 1 = 1\) (mass dimension 1), making it super-renormalizable. In six dimensions, \(\phi^3\) becomes exactly renormalizable (\([g]=0\)), which is consistent with the known result. Furthermore, the coupling constant of the renormalizable \(\phi^4\) theory in four dimensions has \([\lambda] = 4 - 4 \times 1 = 0\), but in six dimensions \([\lambda] = 6 - 4 \times 2 = -2\), making it non-renormalizable. This correspondence is also consistent.

B-3. Direct Calculation of Feynman Parameters

→ Back to problem

Solution Strategy

Apply equation (D.24) to combine the denominators, then complete the square to arrange into the form \(\ell^2 - \Delta\).

Calculation

Introduction of Feynman parameters:

Setting \(A = k^2 - m^2\) and \(B = (k+q)^2 - m^2\), we apply equation (D.24):

\[ \frac{1}{AB} = \int_0^1 dx\;\frac{1}{[xA + (1-x)B]^2} \]

Expanding the contents of the denominator:

\[ xA + (1-x)B = x(k^2 - m^2) + (1-x)((k+q)^2 - m^2) \]
\[ = xk^2 - xm^2 + (1-x)(k^2 + 2k\cdot q + q^2 - m^2) \]
\[ = k^2 + 2(1-x)k\cdot q + (1-x)q^2 - m^2 \]

Completing the square:

We perform the change of variables \(\ell = k + (1-x)q\). From \(k = \ell - (1-x)q\):

\[ k^2 = \ell^2 - 2(1-x)\ell\cdot q + (1-x)^2 q^2 \]
\[ 2(1-x)k\cdot q = 2(1-x)\ell\cdot q - 2(1-x)^2 q^2 \]

Adding these together:

\[ k^2 + 2(1-x)k\cdot q = \ell^2 - (1-x)^2 q^2 \]

Therefore the denominator becomes:

\[ \ell^2 - (1-x)^2 q^2 + (1-x)q^2 - m^2 = \ell^2 + (1-x)[1-(1-x)]q^2 - m^2 \]
\[ = \ell^2 + x(1-x)q^2 - m^2 \]

Defining \(\Delta \equiv m^2 - x(1-x)q^2\):

\[ xA + (1-x)B = \ell^2 - \Delta \]

Final Answer

\[ \frac{1}{(k^2 - m^2)((k+q)^2 - m^2)} = \int_0^1 dx\;\frac{1}{(\ell^2 - \Delta)^2} \]

where \(\ell = k + (1-x)q\), and

\[ \boxed{\Delta = m^2 - x(1-x)q^2} \]

Verification

Special case \(q = 0\): We get \(\Delta = m^2\), and the original expression becomes \(1/(k^2 - m^2)^2\), while the Feynman parameter representation gives \(\int_0^1 dx\;1/(\ell^2 - m^2)^2\) (with \(\ell = k\)). The \(x\) integration trivially yields 1, which is consistent.

Vanishing of linear terms in \(\ell\): We have confirmed that the denominator after the change of variables contains no terms linear in \(\ell\). This ensures that integrands with odd powers of \(\ell\) vanish by symmetry.

B-4. Signs in Wick Rotation

→ Back to problem

(a) Transformation of \(\ell^2\)

Substituting \(\ell_0 = i\ell_0^E\):

\[ \ell^2 = \ell_0^2 - \vec{\ell}^{\,2} = (i\ell_0^E)^2 - \vec{\ell}^{\,2} = -(\ell_0^E)^2 - \vec{\ell}^{\,2} \]
\[ \boxed{\ell^2 = -\ell_E^2} \]

where \(\ell_E^2 \equiv (\ell_0^E)^2 + \vec{\ell}^{\,2}\) is the Euclidean norm squared (positive definite).

(b) Transformation of \(d^4\ell\)

From \(\ell_0 = i\ell_0^E\), we have \(d\ell_0 = i\,d\ell_0^E\). Since the spatial components are unchanged:

\[ d^4\ell = d\ell_0\,d^3\vec{\ell} = i\,d\ell_0^E\,d^3\vec{\ell} \]
\[ \boxed{d^4\ell = i\,d^4\ell_E} \]

(c) Transformation of \(1/(\ell^2 - \Delta + i\varepsilon)^3\)

Substituting the result from (a), \(\ell^2 = -\ell_E^2\):

\[ \ell^2 - \Delta + i\varepsilon = -\ell_E^2 - \Delta + i\varepsilon \]

When \(\ell_E^2 \geq 0\) and \(\Delta > 0\), we have \(-\ell_E^2 - \Delta < 0\), and the \(i\varepsilon\) becomes unnecessary (there is no need to avoid poles). Therefore:

\[ \frac{1}{(\ell^2 - \Delta + i\varepsilon)^3} = \frac{1}{(-\ell_E^2 - \Delta)^3} = \frac{1}{(-1)^3(\ell_E^2 + \Delta)^3} = \frac{-1}{(\ell_E^2 + \Delta)^3} \]
\[ \boxed{\frac{1}{(\ell^2 - \Delta + i\varepsilon)^3} = \frac{-1}{(\ell_E^2 + \Delta)^3}} \]

Combining everything: The transformation of a typical loop integral is

\[ \int \frac{d^4\ell}{(\ell^2 - \Delta + i\varepsilon)^3} = \int \frac{i\,d^4\ell_E}{(-1)^3(\ell_E^2 + \Delta)^3} = \frac{i}{(-1)^3}\int \frac{d^4\ell_E}{(\ell_E^2 + \Delta)^3} = -i\int \frac{d^4\ell_E}{(\ell_E^2 + \Delta)^3} \]

For general power \(n\):

\[ \int \frac{d^4\ell}{(\ell^2 - \Delta + i\varepsilon)^n} = \frac{i}{(-1)^n}\int \frac{d^4\ell_E}{(\ell_E^2 + \Delta)^n} = i(-1)^{-n}\int \frac{d^4\ell_E}{(\ell_E^2 + \Delta)^n} \]

Verification

Dimensional check: Both \(d^4\ell\) and \(d^4\ell_E\) have mass dimension 4. The factor \(i\) is dimensionless. The powers in the denominator are the same. The dimensions are consistent.

Counting factors of \(i\): One factor of \(i\) comes from \(d^4\ell\), and \((-1)^n\) comes from the denominator. For \(n=2\), the overall factor is \(i \cdot (-1)^{-2} = i\); for \(n=3\), it is \(i \cdot (-1)^{-3} = -i\). This agrees with the standard textbook results.

B-5. Solid Angle of a 4-Dimensional Sphere

→ Back to problem

Calculation

We substitute each value into the formula \(\Omega_d = \frac{2\pi^{d/2}}{\Gamma(d/2)}\).

\(d = 2\):

\[ \Omega_2 = \frac{2\pi^{2/2}}{\Gamma(2/2)} = \frac{2\pi^1}{\Gamma(1)} = \frac{2\pi}{1} = 2\pi \quad \checkmark \]

\(d = 3\):

\[ \Omega_3 = \frac{2\pi^{3/2}}{\Gamma(3/2)} = \frac{2\pi^{3/2}}{\frac{\sqrt{\pi}}{2}} = \frac{2\pi^{3/2} \cdot 2}{\sqrt{\pi}} = 4\pi^{3/2 - 1/2} = 4\pi \quad \checkmark \]

\(d = 4\):

\[ \Omega_4 = \frac{2\pi^{4/2}}{\Gamma(4/2)} = \frac{2\pi^2}{\Gamma(2)} = \frac{2\pi^2}{1} = 2\pi^2 \quad \checkmark \]

Final Answer

\[ \boxed{\Omega_2 = 2\pi, \qquad \Omega_3 = 4\pi, \qquad \Omega_4 = 2\pi^2} \]

All results agree with known values.

Verification

\(d = 2\): The circumference of a circle \(2\pi r\) evaluated at \(r = 1\) gives \(2\pi\). \(d = 3\): The surface area of a sphere \(4\pi r^2\) evaluated at \(r = 1\) gives \(4\pi\). Both are geometrically correct.

B-6. Verification of the \(2\pi\) Factors

→ Back to problem

Solution Strategy

Substitute the definition of \(\tilde{f}(p)\) into the expression for \(f(x)\), and show that \(f(x)\) is reproduced using the integral representation of the delta function.

Calculation

Substitute the definition of \(\tilde{f}(p)\) into the expression for \(f(x)\):

\[ f(x) = \int \frac{d^4p}{(2\pi)^4}\;\tilde{f}(p)\,e^{ipx} = \int \frac{d^4p}{(2\pi)^4}\left[\int d^4x'\;f(x')\,e^{-ipx'}\right]e^{ipx} \]

Exchange the order of integration:

\[ = \int d^4x'\;f(x')\int \frac{d^4p}{(2\pi)^4}\;e^{ip(x - x')} \]

Here, from the integral representation of the delta function (D.20):

\[ \int \frac{d^4p}{(2\pi)^4}\;e^{ip(x - x')} = \delta^{(4)}(x - x') \]

(This follows from equation (D.20), \(\int d^4p\;e^{ip\cdot y} = (2\pi)^4\delta^{(4)}(y)\), by setting \(y = x - x'\) and dividing both sides by \((2\pi)^4\).)

Therefore:

\[ f(x) = \int d^4x'\;f(x')\;\delta^{(4)}(x - x') = f(x) \]

Final Answer

It has been shown that \(f(x)\) is identically reproduced. The Fourier transform conventions (D.21) and (D.22) are consistent. \(\square\)

Verification

We also confirm the reverse direction. Substitute the expression for \(f(x)\) into the definition of \(\tilde{f}(p)\):

\[ \tilde{f}(p) = \int d^4x\;f(x)\,e^{-ipx} = \int d^4x\left[\int \frac{d^4p'}{(2\pi)^4}\;\tilde{f}(p')\,e^{ip'x}\right]e^{-ipx} \]
\[ = \int \frac{d^4p'}{(2\pi)^4}\;\tilde{f}(p')\int d^4x\;e^{i(p'-p)x} = \int \frac{d^4p'}{(2\pi)^4}\;\tilde{f}(p')\;(2\pi)^4\delta^{(4)}(p'-p) = \tilde{f}(p) \]

The reverse direction is also consistent.

B-7. General Feynman Parameter Formula (\(n = 3\))

→ Back to problem

Calculation

Writing out the formula:

Setting \(n = 3\) in equation (D.25):

\[ \frac{1}{A_1 A_2 A_3} = 2!\int_0^1 dx_1\,dx_2\,dx_3\;\frac{\delta(1 - x_1 - x_2 - x_3)}{[x_1 A_1 + x_2 A_2 + x_3 A_3]^3} \]
\[ = 2\int_0^1 dx_1\,dx_2\,dx_3\;\frac{\delta(1 - x_1 - x_2 - x_3)}{[x_1 A_1 + x_2 A_2 + x_3 A_3]^3} \]

Eliminating \(x_3\):

Using the delta function \(\delta(1 - x_1 - x_2 - x_3)\), we set \(x_3 = 1 - x_1 - x_2\). The condition \(x_3 \geq 0\) requires \(x_1 + x_2 \leq 1\). Therefore:

\[ \frac{1}{A_1 A_2 A_3} = 2\int_0^1 dx_1\int_0^{1-x_1} dx_2\;\frac{1}{[x_1 A_1 + x_2 A_2 + (1-x_1-x_2) A_3]^3} \]

Description of the Integration Region

The integration region in the \((x_1, x_2)\) plane satisfies: - \(x_1 \geq 0\) - \(x_2 \geq 0\) - \(x_1 + x_2 \leq 1\)

This is the region forming a right isosceles triangle (simplex) with vertices at \((0,0)\), \((1,0)\), and \((0,1)\).

Final Answer

\[ \boxed{\frac{1}{A_1 A_2 A_3} = 2\int_0^1 dx_1\int_0^{1-x_1} dx_2\;\frac{1}{[x_1 A_1 + x_2 A_2 + (1-x_1-x_2) A_3]^3}} \]

Verification

Case \(A_1 = A_2 = A_3 = A\): The denominator becomes \(A^3\), and the integral gives the area of the triangle \(\frac{1}{2}\). Therefore the right-hand side \(= 2 \times \frac{1}{2} \times \frac{1}{A^3} = \frac{1}{A^3}\). The left-hand side is also \(\frac{1}{A^3}\). They agree.

B-8. Estimating Divergences from Mass Dimension

→ Back to problem

Solution Strategy

We examine the behavior of the integrand as \(k \to \infty\). In four dimensions, \(d^4k \sim k^3\,dk\), so when the integrand behaves as \(k^n\), the radial integral takes the form \(\int dk\;k^{n+3}\). This diverges when \(n + 3 \geq -1\) (i.e., \(n \geq -4\)). The superficial degree of divergence is \(D = n + 4\) (divergent for \(D \geq 0\), logarithmically divergent for \(D = 0\)).

Calculation

(a) \(\displaystyle\int \frac{d^4k}{(2\pi)^4}\;\frac{1}{k^2 - m^2}\)

As \(k \to \infty\), the integrand \(\sim 1/k^2\). Combined with the measure \(d^4k \sim k^3\,dk\):

\[ \int dk\;\frac{k^3}{k^2} = \int dk\;k \to \infty \]

Superficial degree of divergence: \(D = 4 - 2 = 2\) (quadratic divergence).

(b) \(\displaystyle\int \frac{d^4k}{(2\pi)^4}\;\frac{1}{(k^2 - m^2)^2}\)

As \(k \to \infty\), the integrand \(\sim 1/k^4\).

\[ \int dk\;\frac{k^3}{k^4} = \int dk\;\frac{1}{k} = \int \frac{dk}{k} \to \infty \]

Superficial degree of divergence: \(D = 4 - 4 = 0\) (logarithmic divergence).

(c) \(\displaystyle\int \frac{d^4k}{(2\pi)^4}\;\frac{k^2}{(k^2 - m^2)^3}\)

As \(k \to \infty\), the integrand \(\sim k^2/k^6 = 1/k^4\).

\[ \int dk\;\frac{k^3}{k^4} = \int \frac{dk}{k} \to \infty \]

Superficial degree of divergence: \(D = 4 + 2 - 6 = 0\) (logarithmic divergence).

Final Answer

Integral Behavior as \(k \to \infty\) Degree of divergence \(D\) Type of divergence
(a) \(k^3/k^2 = k\) \(2\) Quadratic (power) divergence
(b) \(k^3/k^4 = 1/k\) \(0\) Logarithmic divergence
(c) \(k^3 \cdot k^2/k^6 = 1/k\) \(0\) Logarithmic divergence

Consistency Check

Verification by dimensional analysis: The integral in (a) should have mass dimension \(4 - 2 = 2\), which is consistent with \(D = 2\). For (b), the mass dimension is \(4 - 4 = 0\), i.e., dimensionless, consistent with \(D = 0\) (logarithmic divergence takes the form \(\ln\Lambda\), which is dimensionless). For (c), the mass dimension is \(4 + 2 - 6 = 0\), which is the same.

Medium

M-1. Complete Reduction of a One-Loop Integral Using Feynman Parameters

→ Back to problem

(a) Introduction of Feynman Parameters

Setting \(A = k^2 - m^2 + i\varepsilon\) and \(B = (k-p)^2 - m^2 + i\varepsilon\), we apply equation (D.24):

\[ \Sigma(p^2) = \frac{g^2}{2}\int_0^1 dx\int \frac{d^4k}{(2\pi)^4}\;\frac{1}{[x(k^2 - m^2) + (1-x)((k-p)^2 - m^2) + i\varepsilon]^2} \]

Expanding the contents of the denominator:

\[ x(k^2 - m^2) + (1-x)((k-p)^2 - m^2) \]
\[ = xk^2 + (1-x)(k^2 - 2k\cdot p + p^2) - m^2 \]
\[ = k^2 - 2(1-x)k\cdot p + (1-x)p^2 - m^2 \]

(b) Change of Variables and Determination of \(\Delta\)

We perform the change of variables \(\ell = k - (1-x)p\). By a calculation similar to D3:

\[ k^2 - 2(1-x)k\cdot p = \ell^2 - (1-x)^2 p^2 \]

Therefore the denominator becomes:

\[ \ell^2 - (1-x)^2 p^2 + (1-x)p^2 - m^2 = \ell^2 + x(1-x)p^2 - m^2 \]
\[ = \ell^2 - \Delta \]

where:

\[ \boxed{\Delta = m^2 - x(1-x)p^2} \]

Therefore:

\[ \Sigma(p^2) = \frac{g^2}{2}\int_0^1 dx\int \frac{d^4\ell}{(2\pi)^4}\;\frac{1}{(\ell^2 - \Delta + i\varepsilon)^2} \]

(c) Wick Rotation

Setting \(\ell_0 = i\ell_0^E\), and using the results of D4:

  • \(d^4\ell = i\,d^4\ell_E\)
  • \(\ell^2 = -\ell_E^2\)
  • \((\ell^2 - \Delta + i\varepsilon)^2 = (-\ell_E^2 - \Delta)^2 = (\ell_E^2 + \Delta)^2\)

(Since \(\ell_E^2 + \Delta > 0\), the \(i\varepsilon\) is no longer needed.)

Therefore:

\[ \int \frac{d^4\ell}{(\ell^2 - \Delta + i\varepsilon)^2} = \frac{i\,d^4\ell_E}{(\ell_E^2 + \Delta)^2} \]
\[ \Sigma(p^2) = \frac{g^2}{2}\int_0^1 dx\;i\int \frac{d^4\ell_E}{(2\pi)^4}\;\frac{1}{(\ell_E^2 + \Delta)^2} \]

(d) Angular Integration and Logarithmic Divergence

Using spherical coordinates in 4-dimensional Euclidean space, \(d^4\ell_E = \ell_E^3\,d\ell_E\,d\Omega_4\), with \(\Omega_4 = 2\pi^2\) (result from D5):

\[ \int \frac{d^4\ell_E}{(2\pi)^4}\;\frac{1}{(\ell_E^2 + \Delta)^2} = \frac{\Omega_4}{(2\pi)^4}\int_0^\infty d\ell_E\;\frac{\ell_E^3}{(\ell_E^2 + \Delta)^2} \]
\[ = \frac{2\pi^2}{16\pi^4}\int_0^\infty d\ell_E\;\frac{\ell_E^3}{(\ell_E^2 + \Delta)^2} = \frac{1}{8\pi^2}\int_0^\infty d\ell_E\;\frac{\ell_E^3}{(\ell_E^2 + \Delta)^2} \]

Evaluation of the radial integral:

Substituting \(u = \ell_E^2\) gives \(du = 2\ell_E\,d\ell_E\), so \(\ell_E\,d\ell_E = du/2\):

\[ \int_0^\infty d\ell_E\;\frac{\ell_E^3}{(\ell_E^2 + \Delta)^2} = \frac{1}{2}\int_0^\infty du\;\frac{u}{(u + \Delta)^2} \]

Partial fraction decomposition:

\[ \frac{u}{(u+\Delta)^2} = \frac{(u+\Delta) - \Delta}{(u+\Delta)^2} = \frac{1}{u+\Delta} - \frac{\Delta}{(u+\Delta)^2} \]

Therefore:

\[ \frac{1}{2}\int_0^\Lambda du\left[\frac{1}{u+\Delta} - \frac{\Delta}{(u+\Delta)^2}\right] \]

(An ultraviolet cutoff \(\Lambda^2\) is introduced as the upper limit in \(u\), where \(\Lambda\) corresponds to the cutoff in the original \(\ell_E\).)

\[ = \frac{1}{2}\left[\ln(u+\Delta) + \frac{\Delta}{u+\Delta}\right]_0^{\Lambda^2} \]
\[ = \frac{1}{2}\left[\ln(\Lambda^2 + \Delta) + \frac{\Delta}{\Lambda^2 + \Delta} - \ln\Delta - 1\right] \]

In the limit \(\Lambda^2 \gg \Delta\):

\[ \approx \frac{1}{2}\left[\ln\frac{\Lambda^2}{\Delta} - 1 + O(\Delta/\Lambda^2)\right] \]

Confirmation of logarithmic divergence: As \(\Lambda \to \infty\), the \(\ln\Lambda^2\) term survives, and the integral is logarithmically divergent.

Final Result

\[ \Sigma(p^2) = \frac{ig^2}{16\pi^2}\int_0^1 dx\left[\ln\frac{\Lambda^2}{\Delta} - 1\right] + O(1/\Lambda^2) \]

where \(\Delta = m^2 - x(1-x)p^2\). The integral exhibits a logarithmic divergence \(\sim \ln\Lambda^2\).

Consistency Check

Dimensional analysis: Considering \(\phi^3\) theory (renormalizable in 6 dimensions) in 4 dimensions, \([g] = 1\). \(\Sigma\) is the scalar field self-energy with \([\Sigma] = 2\). The right-hand side gives \(g^2 \times (\text{dimensionless}) = 1^2 \times 2 = 2\)... more precisely, in 4-dimensional \(\phi^3\) theory \([g] = 1\) so \([g^2] = 2\), and \(\ln(\Lambda^2/\Delta)\) is dimensionless. Therefore \([\Sigma] = 2\). The self-energy should have mass dimension 2, which is consistent.

Check of the degree of divergence: As seen in D8(b), the integral of \(1/(\ell^2)^2\) is logarithmically divergent (\(D = 0\)). This is consistent with the result obtained here.

M-2. Verification of the Validity of Wick Rotation

→ Back to problem

(a) Position of the Poles

The denominator of the Feynman propagator is \(\ell^2 - \Delta + i\varepsilon = \ell_0^2 - \vec{\ell}^{\,2} - \Delta + i\varepsilon\). Defining \(\omega^2 \equiv \vec{\ell}^{\,2} + \Delta > 0\), the poles are at:

\[ \ell_0^2 = \omega^2 - i\varepsilon \]
\[ \ell_0 = \pm\sqrt{\omega^2 - i\varepsilon} = \pm(\omega - i\varepsilon') \]

where \(\varepsilon' = \varepsilon/(2\omega) > 0\) (in the limit \(\varepsilon \to 0^+\)).

Therefore: - Positive pole \(\ell_0 = +\omega - i\varepsilon'\): slightly below the real axis (on the fourth quadrant side) - Negative pole \(\ell_0 = -\omega + i\varepsilon'\): slightly above the real axis (on the second quadrant side)

\[ \boxed{\text{The poles are located in the second and fourth quadrants}} \]

(b) Contribution from the Arc

We rotate the integration contour of \(\ell_0\) from the real axis to the imaginary axis by 90° counterclockwise. The closed contour consists of:

  1. Along the real axis from \(-R\) to \(+R\)
  2. A quarter-circle arc in the first quadrant (radius \(R\))
  3. Along the imaginary axis from \(+iR\) to \(-iR\) (reverse direction)... rather, more precisely: real axis → arc in the first quadrant → positive imaginary axis, and negative real axis → arc in the second quadrant → positive imaginary axis.

More precisely stated: we rotate the integration path along the real axis through the first and second quadrants to the imaginary axis.

On the quarter-circle arc, setting \(\ell_0 = Re^{i\theta}\) (\(0 \leq \theta \leq \pi/2\)), we have \(|\ell_0| = R\). A typical factor in the denominator is:

\[ |\ell_0^2 - \omega^2 + i\varepsilon| \sim |\ell_0^2| = R^2 \quad (R \to \infty) \]

For a product of \(n\) propagators, the denominator grows as \(\sim R^{2n}\). Meanwhile, the integration measure on the arc is \(|d\ell_0| = R\,d\theta\).

The entire integrand (when there are no powers of \(\ell_0\) in the numerator) behaves on the arc as:

\[ \sim \frac{R\,d\theta}{R^{2n}} = \frac{d\theta}{R^{2n-1}} \]

Since \(2n - 1 \geq 1\) when \(n \geq 1\), the contribution from the arc vanishes as \(R \to \infty\).

(c) Justification via Cauchy's Integral Theorem

From (a), the poles are in the second and fourth quadrants. The closed contour rotated 90° counterclockwise (real axis → arc in the first quadrant → imaginary axis) encloses the first quadrant. Since no poles exist in the first quadrant, by Cauchy's integral theorem:

\[ \oint_C f(\ell_0)\,d\ell_0 = 0 \]

From (b), the contribution from the arc is zero, so:

\[ \int_{-\infty}^{+\infty} f(\ell_0)\,d\ell_0 + \int_{\text{imaginary axis (top to bottom)}} f(\ell_0)\,d\ell_0 = 0 \]

Substituting \(\ell_0 = i\ell_0^E\) (\(\ell_0^E\): from \(-\infty\) to \(+\infty\)) on the imaginary axis gives \(d\ell_0 = i\,d\ell_0^E\). Taking the orientation into account:

\[ \int_{-\infty}^{+\infty} d\ell_0\;f(\ell_0) = i\int_{-\infty}^{+\infty} d\ell_0^E\;f(i\ell_0^E) \]

This is the Wick rotation identity, justified as a direct consequence of Cauchy's integral theorem. \(\square\)

Consistency Check

Consistency of pole positions: If we reverse the sign of \(i\varepsilon\) (to \(-i\varepsilon\)), the poles move to the first and third quadrants, and the Wick rotation can no longer be justified. This corresponds to the anti-Feynman prescription and contradicts the physical causality condition. The structure in which Feynman's \(+i\varepsilon\) prescription makes the Wick rotation possible is consistent.

M-3. Derivation of the Basic Formula for Dimensional Regularization

→ Back to problem

Goal

Derive the following:

\[ \int \frac{d^d\ell_E}{(2\pi)^d}\;\frac{1}{(\ell_E^2 + \Delta)^n} = \frac{1}{(4\pi)^{d/2}}\;\frac{\Gamma(n - d/2)}{\Gamma(n)}\;\frac{1}{\Delta^{n-d/2}} \]

(a) Angular Integration

In \(d\)-dimensional spherical coordinates, \(d^d\ell_E = \ell_E^{d-1}\,d\ell_E\,d\Omega_d\). Since the integrand depends only on \(\ell_E = |\ell_E|\), the angular integration yields the solid angle \(\Omega_d\):

\[ \int \frac{d^d\ell_E}{(2\pi)^d}\;\frac{1}{(\ell_E^2 + \Delta)^n} = \frac{\Omega_d}{(2\pi)^d}\int_0^\infty d\ell_E\;\frac{\ell_E^{d-1}}{(\ell_E^2 + \Delta)^n} \]

Substituting \(\Omega_d = \frac{2\pi^{d/2}}{\Gamma(d/2)}\):

\[ = \frac{2\pi^{d/2}}{(2\pi)^d\,\Gamma(d/2)}\int_0^\infty d\ell_E\;\frac{\ell_E^{d-1}}{(\ell_E^2 + \Delta)^n} \]

Simplifying the prefactor:

\[ \frac{2\pi^{d/2}}{(2\pi)^d\,\Gamma(d/2)} = \frac{2\pi^{d/2}}{2^d\pi^d\,\Gamma(d/2)} = \frac{1}{2^{d-1}\pi^{d/2}\,\Gamma(d/2)} \]

(b) Reducing the Radial Integral to a Beta Function

Substitute \(t = \ell_E^2/\Delta\). Then \(\ell_E = \sqrt{\Delta t}\), \(d\ell_E = \frac{\sqrt{\Delta}}{2\sqrt{t}}\,dt\).

\[ \ell_E^{d-1} = (\Delta t)^{(d-1)/2} = \Delta^{(d-1)/2}\,t^{(d-1)/2} \]
\[ (\ell_E^2 + \Delta)^n = (\Delta t + \Delta)^n = \Delta^n(1+t)^n \]

Therefore the radial integral becomes:

\[ \int_0^\infty d\ell_E\;\frac{\ell_E^{d-1}}{(\ell_E^2 + \Delta)^n} = \int_0^\infty \frac{\sqrt{\Delta}}{2\sqrt{t}}\,dt\;\frac{\Delta^{(d-1)/2}\,t^{(d-1)/2}}{\Delta^n(1+t)^n} \]
\[ = \frac{\Delta^{1/2 + (d-1)/2 - n}}{2}\int_0^\infty dt\;\frac{t^{(d-1)/2 - 1/2}}{(1+t)^n} \]
\[ = \frac{\Delta^{d/2 - n}}{2}\int_0^\infty dt\;\frac{t^{d/2 - 1}}{(1+t)^n} \]

Comparing this integral with the integral representation of the Beta function:

\[ B(a, b) = \int_0^\infty dt\;\frac{t^{a-1}}{(1+t)^{a+b}} = \frac{\Gamma(a)\Gamma(b)}{\Gamma(a+b)} \]

we identify \(a = d/2\), \(a + b = n\), i.e., \(b = n - d/2\):

\[ \int_0^\infty dt\;\frac{t^{d/2 - 1}}{(1+t)^n} = B\!\left(\frac{d}{2},\;n - \frac{d}{2}\right) = \frac{\Gamma(d/2)\,\Gamma(n - d/2)}{\Gamma(n)} \]

(This converges when \(n - d/2 > 0\). In dimensional regularization, \(d\) is treated as a continuous variable and the result is analytically continued.)

Assembling the Full Result

\[ \int \frac{d^d\ell_E}{(2\pi)^d}\;\frac{1}{(\ell_E^2 + \Delta)^n} = \frac{1}{2^{d-1}\pi^{d/2}\,\Gamma(d/2)} \cdot \frac{\Delta^{d/2-n}}{2} \cdot \frac{\Gamma(d/2)\,\Gamma(n-d/2)}{\Gamma(n)} \]

The \(\Gamma(d/2)\) factors cancel:

\[ = \frac{1}{2^d\,\pi^{d/2}} \cdot \frac{\Gamma(n-d/2)}{\Gamma(n)} \cdot \frac{1}{\Delta^{n-d/2}} \]

Now verify directly:

\[ (4\pi)^{d/2} = 4^{d/2}\,\pi^{d/2} = 2^d\,\pi^{d/2} \]

Therefore:

\[ \frac{1}{2^d\,\pi^{d/2}} = \frac{1}{(4\pi)^{d/2}} \]

Final Answer

\[ \boxed{\int \frac{d^d\ell_E}{(2\pi)^d}\;\frac{1}{(\ell_E^2 + \Delta)^n} = \frac{1}{(4\pi)^{d/2}}\;\frac{\Gamma(n - d/2)}{\Gamma(n)}\;\frac{1}{\Delta^{n-d/2}}} \]

Consistency Checks

Case \(d = 4\), \(n = 2\):

\[ \frac{1}{(4\pi)^2}\;\frac{\Gamma(2-2)}{\Gamma(2)}\;\frac{1}{\Delta^0} = \frac{1}{16\pi^2}\;\frac{\Gamma(0)}{1} \]

\(\Gamma(0)\) has a pole. This corresponds to the logarithmic divergence seen in S1. In dimensional regularization, setting \(d = 4 - 2\epsilon\), one expands \(\Gamma(\epsilon) = 1/\epsilon - \gamma_E + O(\epsilon)\), and the \(1/\epsilon\) pole represents the logarithmic divergence. This is consistent.

Dimensional analysis: The mass dimension of the left-hand side is \(d - 2n\) (\(d^d\ell_E\) has dimension \(d\), the denominator has dimension \(2n\)). On the right-hand side, \(\Delta^{-(n-d/2)}\) has dimension \(-2(n-d/2) = d - 2n\). They match.

M-4. Restoring Units and Estimating Cross Sections

→ Back to problem

(a) Confirming the mass dimension of \(\sigma\)

In natural units, \([\alpha] = 0\) (the fine-structure constant is dimensionless), \([s] = 2\) (dimension of energy squared).

\[ [\sigma] = \frac{[\alpha]^2}{[s]} = \frac{0}{2} = -2 \]
\[ \boxed{[\sigma] = -2} \]

The cross section should have dimensions of "area," and \([\text{length}^2] = (-1)^2 = -2\) (in mass dimension). This is consistent.

(b) Numerical calculation of \(\sigma\)

\(\sqrt{s} = 10\) GeV, i.e., \(s = 100\) GeV\(^2\).

\[ \sigma = \frac{4\pi\alpha^2}{3s} = \frac{4\pi}{3 \times 137^2 \times 100}\ \text{GeV}^{-2} \]

Computing the numerical value:

\[ 137^2 = 18769 \]
\[ 3 \times 18769 \times 100 = 5{,}630{,}700 \]
\[ \sigma = \frac{4\pi}{5{,}630{,}700}\ \text{GeV}^{-2} = \frac{12.566}{5{,}630{,}700}\ \text{GeV}^{-2} = 2.232 \times 10^{-6}\ \text{GeV}^{-2} \]

Using the conversion factor (D.6): \(1\ \text{GeV}^{-2} = 0.3894\ \text{mb} = 3.894 \times 10^{8}\ \text{pb}\)

\[ \sigma = 2.232 \times 10^{-6} \times 3.894 \times 10^{8}\ \text{pb} \]
\[ \boxed{\sigma \approx 869\ \text{pb} \approx 0.87\ \text{nb}} \]

(c) Expected number of events

The number of events is given by \(N = \sigma \cdot \mathcal{L} \cdot T\).

\(\mathcal{L} = 10^{33}\ \text{cm}^{-2}\text{s}^{-1}\), \(T = 1\ \text{day} = 86{,}400\ \text{s}\).

First, convert \(\sigma\) to cm\(^2\):

\[ \sigma \approx 869\ \text{pb} = 869 \times 10^{-40}\ \text{m}^2 = 869 \times 10^{-36}\ \text{cm}^2 = 8.69 \times 10^{-34}\ \text{cm}^2 \]
\[ N = 8.69 \times 10^{-34} \times 10^{33} \times 86{,}400 \]
\[ = 8.69 \times 10^{-1} \times 86{,}400 \]
\[ = 0.869 \times 86{,}400 \approx 75{,}000 \]
\[ \boxed{N \approx 7.5 \times 10^4\ \text{events/day}} \]

Consistency check

Order-of-magnitude verification: It is experimentally known that the cross section for \(e^+e^- \to \mu^+\mu^-\) is of order \(\sim 1\) nb at \(\sqrt{s} = 10\) GeV, so \(0.87\) nb is reasonable. A luminosity of \(10^{33}\) cm\(^{-2}\)s\(^{-1}\) corresponds to a B-factory class accelerator, and the estimate of tens of thousands of events per day is realistic.

Advanced

A-1. The \(\gamma_5\) Problem in Dimensional Regularization and the ABJ Anomaly

→ Back to problem

(a) Inconsistency of \(\{\gamma_5, \gamma^\mu\} = 0\) in \(d\) Dimensions

Method 1: Use \(\gamma^\alpha\gamma_\alpha = d\) (trace in \(d\) dimensions) first.

Consider the following quantity:

\[ T \equiv \text{Tr}[\gamma_5\,\gamma^\alpha\gamma_\alpha\,\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] \]

Using \(\gamma^\alpha\gamma_\alpha = d\):

\[ T = d\;\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] \tag{1} \]

Method 2: Move \(\gamma^\alpha\) to the right, using the anticommutation relation with \(\gamma_5\).

From \(\{\gamma_5, \gamma^\alpha\} = 0\), we have \(\gamma_5\gamma^\alpha = -\gamma^\alpha\gamma_5\). Therefore:

\[ T = \text{Tr}[\gamma_5\,\gamma^\alpha\,\gamma_\alpha\,\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] \]

Move \(\gamma_5\) from the far left to the right of \(\gamma^\alpha\):

\[ \gamma_5\,\gamma^\alpha = -\gamma^\alpha\,\gamma_5 \]
\[ T = -\text{Tr}[\gamma^\alpha\,\gamma_5\,\gamma_\alpha\,\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] \]

Next, move \(\gamma_5\) to the right of \(\gamma_\alpha\):

\[ \gamma_5\,\gamma_\alpha = -\gamma_\alpha\,\gamma_5 \]
\[ T = (-1)^2\text{Tr}[\gamma^\alpha\gamma_\alpha\,\gamma_5\,\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] \]

Similarly, move \(\gamma_5\) successively to the right of \(\gamma^\mu, \gamma^\nu, \gamma^\rho, \gamma^\sigma\). Each move produces a factor of \((-1)\). After 4 anticommutations in total:

\[ T = (-1)^{2+4}\text{Tr}[\gamma^\alpha\gamma_\alpha\,\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma\,\gamma_5] \]

Let us redo this more carefully.

Move \(\gamma_5\) to the right of the first \(\gamma^\alpha\) (one factor of \(-1\)), then to the right of \(\gamma_\alpha\) (one factor of \(-1\)), to the right of \(\gamma^\mu\) (\(-1\)), to the right of \(\gamma^\nu\) (\(-1\)), to the right of \(\gamma^\rho\) (\(-1\)), to the right of \(\gamma^\sigma\) (\(-1\)). A total of 6 anticommutations:

\[ T = (-1)^6\,\text{Tr}[\gamma^\alpha\gamma_\alpha\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma\gamma_5] \]

Using the cyclicity of the trace \(\text{Tr}[AB\cdots Z] = \text{Tr}[ZAB\cdots]\):

\[ \text{Tr}[\gamma^\alpha\gamma_\alpha\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma\gamma_5] = \text{Tr}[\gamma_5\gamma^\alpha\gamma_\alpha\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] = T \]

Therefore, from Method 2:

\[ T = (+1)\cdot T = T \]

This does not produce a contradiction. A more refined approach is needed.

Correct approach: Consider a form where \(\gamma^\alpha\) is not sandwiched between \(\gamma_5\) and \(\gamma_\alpha\). Consider the quantity:

\[ T' \equiv \text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] \]

\(0 = \text{Tr}[\gamma_5\{\gamma^\alpha, \gamma_\alpha\}\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma]\)... this is also trivial.

Let us take a different approach. Consider the following identity:

\[ \gamma^\alpha\gamma^\mu\gamma_\alpha = (2-d)\gamma^\mu \quad \text{(identity in $d$ dimensions)} \]

Using this, compute:

\[ \text{Tr}[\gamma_5\,\gamma^\alpha\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma\gamma_\alpha] \]

in two different ways.

Method A: Move \(\gamma_\alpha\) to the left and use \(\gamma^\alpha\gamma_\alpha = d\).

Bring \(\gamma_\alpha\) to the far left using the cyclicity of the trace:

\[ \text{Tr}[\gamma_5\,\gamma^\alpha\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma\gamma_\alpha] = \text{Tr}[\gamma_\alpha\gamma_5\,\gamma^\alpha\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] \]

Assuming \(\{\gamma_5, \gamma_\alpha\} = 0\), we have \(\gamma_\alpha\gamma_5 = -\gamma_5\gamma_\alpha\):

\[ = -\text{Tr}[\gamma_5\gamma_\alpha\gamma^\alpha\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] = -d\;\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] \]

Method B: Repeatedly use the contraction formulas for \(\gamma^\alpha\cdots\gamma_\alpha\).

\[ \gamma^\alpha\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma\gamma_\alpha \]

Repeatedly applying the contraction formulas in \(d\) dimensions:

\[ \gamma^\alpha\gamma^\mu\gamma_\alpha = (2-d)\gamma^\mu \]
\[ \gamma^\alpha\gamma^\mu\gamma^\nu\gamma_\alpha = 4\eta^{\mu\nu} - (4-d)\gamma^\mu\gamma^\nu \]
\[ \gamma^\alpha\gamma^\mu\gamma^\nu\gamma^\rho\gamma_\alpha = -2\gamma^\rho\gamma^\nu\gamma^\mu + (4-d)\gamma^\mu\gamma^\nu\gamma^\rho \]
\[ \gamma^\alpha\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma\gamma_\alpha = 2(\gamma^\sigma\gamma^\rho\gamma^\nu\gamma^\mu + \gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma) - (4-d)\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma \]

At \(d = 4\), the last term vanishes, giving \(\gamma^\alpha\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma\gamma_\alpha = 2(\gamma^\sigma\gamma^\rho\gamma^\nu\gamma^\mu + \gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma)\).

Inserting the result of Method B into the trace:

\[ \text{Tr}[\gamma_5\,\gamma^\alpha\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma\gamma_\alpha] = 2\,\text{Tr}[\gamma_5(\gamma^\sigma\gamma^\rho\gamma^\nu\gamma^\mu + \gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma)] - (4-d)\,\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] \]

In four dimensions, \(\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] = -4i\epsilon^{\mu\nu\rho\sigma}\), and \(\text{Tr}[\gamma_5\gamma^\sigma\gamma^\rho\gamma^\nu\gamma^\mu] = -4i\epsilon^{\sigma\rho\nu\mu} = -4i\epsilon^{\mu\nu\rho\sigma}\) (by the total antisymmetry of the Levi-Civita symbol, an even number of transpositions leaves the sign unchanged... indeed \(\epsilon^{\sigma\rho\nu\mu}\) is obtained from \(\epsilon^{\mu\nu\rho\sigma}\) by 4 adjacent transpositions, giving \(+\epsilon^{\mu\nu\rho\sigma}\)).

Therefore, the result of Method B at \(d = 4\) is:

\[ = 2 \times 2 \times \text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] = 4\,\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] \]

(The \((4-d)\) term vanishes at \(d = 4\).)

Comparison of Method A and Method B:

Method A: \(-d\;\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma]\)

Method B: \(4\;\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] - (4-d)\;\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma]\)

\(= (4 - 4 + d)\;\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] = d\;\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma]\)

Setting the two equal:

\[ -d\;\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] = d\;\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] \]
\[ \Rightarrow\quad 2d\;\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] = 0 \]

Since \(d \neq 0\):

\[ \text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] = 0 \]

However, in four dimensions \(\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] = -4i\epsilon^{\mu\nu\rho\sigma} \neq 0\).

\[ \boxed{\text{Contradiction}} \]

That is, if one assumes \(\{\gamma_5, \gamma^\mu\} = 0\) for all \(\mu\) in \(d\) dimensions, it forces traces involving \(\gamma_5\) to vanish, contradicting the known four-dimensional result.

(b) The 't Hooft–Veltman Prescription

In the 't Hooft–Veltman (HV) prescription, the \(d\)-dimensional \(\gamma^\mu\) is decomposed into two parts:

\[ \gamma^\mu = \hat{\gamma}^\mu + \tilde{\gamma}^\mu \]
  • \(\hat{\gamma}^\mu\): components in the 4-dimensional subspace (\(\mu = 0, 1, 2, 3\))
  • \(\tilde{\gamma}^\mu\): components in the remaining \((d-4)\)-dimensional subspace

\(\gamma_5\) is defined solely in terms of the 4-dimensional \(\gamma\) matrices:

\[ \gamma_5 = i\gamma^0\gamma^1\gamma^2\gamma^3 \]

The (anti)commutation relations are:

\[ \{\gamma_5, \hat{\gamma}^\mu\} = 0, \qquad [\gamma_5, \tilde{\gamma}^\mu] = 0 \]

That is, \(\gamma_5\) anticommutes with the 4-dimensional components \(\hat{\gamma}^\mu\), but commutes with the \((d-4)\)-dimensional components \(\tilde{\gamma}^\mu\).

Implications for the triangle diagram (AVV vertex):

In the triangle diagram, one computes a one-loop diagram with one axial-vector (A) vertex and two vector (V) vertices. The axial vertex contains \(\gamma^\mu\gamma_5\).

In the HV prescription, the \((d-4)\)-dimensional component \(\tilde{k}\) of the loop momentum \(k\) commutes with \(\gamma_5\), leading to additional terms that differ from the naive 4-dimensional calculation. Specifically:

  • Contributions from only the 4-dimensional components can be adjusted to satisfy the Ward identity for the vector current
  • However, contributions from the \((d-4)\)-dimensional components leave a finite remainder in the limit \(\epsilon = (4-d)/2 \to 0\)

This finite remainder demonstrates that the regularization procedure breaks the axial symmetry, and constitutes the origin of the anomaly.

(c) Physical Consequences of the ABJ Anomaly

Classically, for massless fermions the axial current \(j_5^\mu = \bar{\psi}\gamma^\mu\gamma_5\psi\) is conserved:

\[ \partial_\mu j_5^\mu = 0 \quad \text{(at the classical level)} \]

However, when computing quantum corrections (triangle diagrams), the \(\gamma_5\) issue discussed above implies that regardless of the regularization scheme employed, it is impossible to simultaneously maintain both vector current conservation and axial current conservation.

Prioritizing the gauge invariance of QED (vector current conservation) on physical grounds, the axial current conservation law is violated at the quantum level:

\[ \partial_\mu j_5^\mu = \frac{e^2}{16\pi^2}\,F_{\mu\nu}\tilde{F}^{\mu\nu} \]

where \(\tilde{F}^{\mu\nu} = \frac{1}{2}\epsilon^{\mu\nu\rho\sigma}F_{\rho\sigma}\) is the dual field strength tensor.

Physical consequences:

  1. \(\pi^0 \to \gamma\gamma\) decay: Without the axial current anomaly, the two-photon decay rate of the neutral pion would be zero, but the anomaly predicts a finite decay rate that agrees precisely with experiment.

  2. Anomaly cancellation conditions: For gauge anomalies to cancel in the Standard Model, specific combinations of fermion charges must vanish. This provides strong constraints on the generation structure of quarks and leptons.

  3. Non-renormalization of the anomaly (Adler–Bardeen theorem): The ABJ anomaly is determined exactly at one loop and receives no higher-loop corrections. This reflects the topological nature of the anomaly.

Consistency Check

Alternative perspective on the contradiction in (a): The requirement that \(\text{Tr}[\gamma_5\gamma^\mu\gamma^\nu\gamma^\rho\gamma^\sigma] = 0\) in \(d\) dimensions means that the \(\epsilon^{\mu\nu\rho\sigma}\) tensor cannot be naturally extended to \(d\) dimensions. The \(\epsilon^{\mu\nu\rho\sigma}\) is an inherently 4-dimensional object, which is consistent with the fact that \(\gamma_5\) is a concept intrinsic to four dimensions.

A-2. Relationship between Feynman Parameters and the Mellin–Barnes Representation

→ Back to problem

(a) Proof of the Mellin–Barnes Representation

Identity to be proved:

\[ \frac{1}{(A + B)^n} = \frac{1}{\Gamma(n)}\;\frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty} ds\;\Gamma(-s)\Gamma(n+s)\;\frac{A^s}{B^{n+s}} \]

Proof: We evaluate the \(s\) integral on the right-hand side as a sum of residues at the poles \(s = k\) (\(k = 0, 1, 2, \ldots\)) of \(\Gamma(-s)\). \(\Gamma(-s)\) has a first-order pole at \(s = k\), with residue:

\[ \text{Res}_{s=k}\,\Gamma(-s) = \frac{(-1)^k}{k!} \]

(This follows from the fact that \(\Gamma(z)\) has a pole at \(z = -k\) with \(\text{Res} = (-1)^k/k!\). The pole of \(\Gamma(-s)\) at \(s = k\) corresponds to the pole of \(\Gamma(z)\) at \(z = -k\).)

Closing the integration contour to the right (assuming \(|A/B| < 1\), i.e., \(|A| < |B|\)), we pick up the poles at \(s = k\) (\(k = 0, 1, 2, \ldots\)):

\[ \frac{1}{2\pi i}\int ds\;\Gamma(-s)\Gamma(n+s)\frac{A^s}{B^{n+s}} = \sum_{k=0}^{\infty}\frac{(-1)^k}{k!}\;\Gamma(n+k)\;\frac{A^k}{B^{n+k}} \]

Using \(\Gamma(n+k) = (n+k-1)!\) (when \(n\) is a positive integer) and dividing by \(\Gamma(n) = (n-1)!\):

\[ \frac{1}{\Gamma(n)}\sum_{k=0}^{\infty}\frac{(-1)^k}{k!}\;\frac{\Gamma(n+k)}{\Gamma(n)}\;\frac{A^k}{B^{n+k}} \cdot \Gamma(n) \]

Simplifying:

\[ = \frac{1}{\Gamma(n)}\sum_{k=0}^{\infty}\frac{(-1)^k}{k!}\;\Gamma(n+k)\;\frac{A^k}{B^{n+k}} \]

Using the fact that \(\frac{\Gamma(n+k)}{\Gamma(n)\,k!} = \binom{n+k-1}{k}\):

\[ = \frac{1}{B^n}\sum_{k=0}^{\infty}\binom{n+k-1}{k}\left(-\frac{A}{B}\right)^k \]

By the generalized binomial theorem \((1+z)^{-n} = \sum_{k=0}^{\infty}\binom{n+k-1}{k}(-z)^k\) (\(|z| < 1\)):

\[ = \frac{1}{B^n}\left(1 + \frac{A}{B}\right)^{-n} = \frac{1}{(A+B)^n} \]

\(\square\)

By analytic continuation, the restriction \(|A/B| < 1\) can be removed, and the identity holds for general \(A, B\).

(b) Mellin–Barnes Representation of the One-Loop Integral

Starting point: The integral after introducing Feynman parameters (as in S1):

\[ I(p^2) = \int_0^1 dx\int \frac{d^d\ell}{(2\pi)^d}\;\frac{1}{(\ell^2 - \Delta + i\varepsilon)^2} \]

where \(\ell = k - (1-x)p\), \(\Delta = xm_1^2 + (1-x)m_2^2 - x(1-x)p^2\).

After Wick rotation, applying the formula from S3:

\[ I(p^2) = \frac{i}{(4\pi)^{d/2}}\;\frac{\Gamma(2-d/2)}{\Gamma(2)}\int_0^1 dx\;\frac{1}{\Delta^{2-d/2}} \]
\[ = \frac{i}{(4\pi)^{d/2}}\;\Gamma(2-d/2)\int_0^1 dx\;\frac{1}{\Delta^{2-d/2}} \]

where \(\Delta = xm_1^2 + (1-x)m_2^2 - x(1-x)p^2\).

Applying the Mellin–Barnes representation:

We split \(\Delta\) into two terms. For example:

\[ \Delta = [xm_1^2 + (1-x)m_2^2] - x(1-x)p^2 \equiv C - D \]

where \(C = xm_1^2 + (1-x)m_2^2\), \(D = x(1-x)p^2\).

Alternatively, a more useful decomposition separates \(m_1^2\) and \(m_2^2\). We decompose the power of \(\Delta\) inside the Feynman parameter integral using Mellin–Barnes.

Applying the formula from (a) to \(\Delta^{-(2-d/2)}\), with \(\Delta = C + D'\) (\(C, D'\) being an appropriate decomposition of \(\Delta\)):

\[ \frac{1}{\Delta^{2-d/2}} = \frac{1}{\Gamma(2-d/2)}\;\frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty} ds\;\Gamma(-s)\Gamma(2-d/2+s)\;\frac{C^s}{D'^{2-d/2+s}} \]

Substituting this into the \(x\) integral, the \(x\) integration reduces to a Beta function, and ultimately only the Mellin–Barnes integral in \(s\) remains.

Equivalence with the Feynman parameter representation: If the Mellin–Barnes integral is evaluated as a sum of residues, each residue corresponds to a term in the \(x\)-expansion of the Feynman parameter integral. Conversely, starting from the Feynman parameter representation and separating terms in \(\Delta\) using Mellin–Barnes yields the same result. Both are different representations of the same integral and are equivalent.

(c) Asymptotic Expansion (\(m_1 = 0\), \(m_2 = m\), \(p^2 = -Q^2\), \(Q^2 \gg m^2\))

Setup: \(m_1 = 0\), \(m_2 = m\), \(p^2 = -Q^2\).

\[ \Delta = (1-x)m^2 + x(1-x)Q^2 \]

The result after performing the \(\ell_E\) integral following the Feynman parameter integration:

\[ I = \frac{i}{(4\pi)^{d/2}}\;\Gamma(2-d/2)\int_0^1 dx\;\frac{1}{[(1-x)m^2 + x(1-x)Q^2]^{2-d/2}} \]

This can be factored as \(\Delta = (1-x)[m^2 + xQ^2]\).

Separation using Mellin–Barnes:

We separate \(m^2\) and \(xQ^2\) inside \(\Delta^{-(2-d/2)}\) using Mellin–Barnes. Setting \(n = 2 - d/2\), \(A = m^2\), \(B = xQ^2\):

\[ \frac{1}{(m^2 + xQ^2)^n} = \frac{1}{\Gamma(n)}\;\frac{1}{2\pi i}\int ds\;\Gamma(-s)\Gamma(n+s)\;\frac{(m^2)^s}{(xQ^2)^{n+s}} \]

Substituting this into the \(x\) integral and including the \((1-x)^{-n}\) factor, we perform the \(x\) integration. The \(x\) integral becomes a Beta function:

\[ \int_0^1 dx\;(1-x)^{-n}\;x^{-(n+s)} = B(1-n, 1-n-s) = \frac{\Gamma(1-n)\Gamma(1-n-s)}{\Gamma(2-2n-s)} \]

(Convergence conditions are handled by analytic continuation.)

Asymptotic expansion for \(Q^2 \gg m^2\):

Closing the Mellin–Barnes contour to the right, we pick up residues at the poles \(s = 0, 1, 2, \ldots\) of \(\Gamma(-s)\). Due to the factor \((m^2)^s/(Q^2)^{n+s}\), in the limit \(m^2/Q^2 \to 0\) the residues at low \(s\) are dominant.

Residue at \(s = 0\):

\[ \text{Res}_{s=0}\,\Gamma(-s) = -1 \]

(The residue of \(\Gamma(-s)\) at \(s = 0\) is \((-1)^0/0! = 1\), but care must be taken with signs. The Laurent expansion of \(\Gamma(z)\) at \(z = 0\) is \(\Gamma(z) = 1/z - \gamma_E + \cdots\), so \(\text{Res}_{z=0}\Gamma(z) = 1\). The pole of \(\Gamma(-s)\) at \(s = 0\) corresponds to the pole of \(\Gamma(z)\) at \(z = 0\), and \(\text{Res}_{s=0}\Gamma(-s) = -\text{Res}_{z=0}\Gamma(z) = -1\).)

More precisely, let us recompute the exact residues. \(\Gamma(z)\) has a first-order pole at \(z = -n\) (\(n = 0, 1, 2, \ldots\)):

\[ \text{Res}_{z=-n}\,\Gamma(z) = \frac{(-1)^n}{n!} \]

The pole of \(\Gamma(-s)\) at \(s = k\) corresponds to \(-s = -k\), i.e., \(z = -k\). The residue in \(s\) is:

\[ \lim_{s \to k}(s-k)\Gamma(-s) = \lim_{z \to -k}(-z-k)\Gamma(z) \cdot \frac{ds}{ds} \]

Since \(z = -s\), we have \(dz = -ds\). \((s-k) = -(z-(-k)) = -(z+k)\).

\[ \lim_{s \to k}(s-k)\Gamma(-s) = \lim_{z \to -k}(-(z+k))\Gamma(z) \cdot 1 = -\text{Res}_{z=-k}\Gamma(z) = -\frac{(-1)^k}{k!} = \frac{(-1)^{k+1}}{k!} \]

Therefore at \(s = 0\): \(\text{Res} = \frac{(-1)^1}{0!} = -1\)

At \(s = 1\): \(\text{Res} = \frac{(-1)^2}{1!} = 1\)

Setting \(d = 4 - 2\epsilon\), we have \(n = 2 - d/2 = \epsilon\).

Contribution from \(s = 0\):

\[ \frac{(-1)}{\Gamma(\epsilon)}\;\Gamma(\epsilon)\;\frac{(m^2)^0}{(Q^2)^{\epsilon}} \cdot \frac{\Gamma(1-\epsilon)\Gamma(1-\epsilon)}{\Gamma(2-2\epsilon)} \cdot \frac{1}{(1-x)^{...}} \]

Since the calculation becomes complicated, it is more transparent to directly obtain the first two terms for the \(d = 4\) (\(\epsilon = 0\)) case.

Direct asymptotic expansion:

Focusing on the finite part after regularization in the \(\epsilon \to 0\) limit at \(d = 4\):

\[ I \sim \frac{i}{16\pi^2}\int_0^1 dx\;\ln\frac{\mu^2}{(1-x)[m^2 + xQ^2]} \]

When \(Q^2 \gg m^2\), for most of the range \(x \in [0,1]\) we have \(xQ^2 \gg m^2\), so:

\[ \ln\frac{\mu^2}{(1-x)xQ^2}\left[1 + O\!\left(\frac{m^2}{xQ^2}\right)\right] \]

In the Mellin–Barnes residue calculation, the residue at \(s = 0\) gives the leading term, and the residue at \(s = 1\) gives the next-to-leading term.

Result for the first two terms:

\[ I(Q^2) = \frac{i}{(4\pi)^{d/2}}\;\frac{\Gamma(\epsilon)}{(Q^2)^\epsilon}\left[\frac{\Gamma^2(1-\epsilon)}{\Gamma(2-2\epsilon)} - \epsilon\;\frac{m^2}{Q^2}\;\frac{\Gamma^2(1-\epsilon)\Gamma(1+\epsilon)}{\Gamma(2-\epsilon)} + O\!\left(\frac{m^4}{Q^4}\right)\right] \]

In the \(\epsilon \to 0\) limit:

\[ \frac{\Gamma^2(1-\epsilon)}{\Gamma(2-2\epsilon)} \to \frac{1}{1} = 1 \]

The second term leaves a finite contribution since \(\epsilon \cdot \Gamma(\epsilon) = 1 + O(\epsilon)\):

\[ -\frac{m^2}{Q^2}\;\frac{\Gamma^2(1)\Gamma(1)}{\Gamma(2)} = -\frac{m^2}{Q^2} \]

Final Answer

The first two terms of the asymptotic expansion for \(Q^2 \gg m^2\):

\[ \boxed{I(Q^2) = \frac{i}{(4\pi)^{d/2}}\;\frac{\Gamma(\epsilon)}{(Q^2)^\epsilon}\left[1 - \frac{m^2}{Q^2} + O\!\left(\frac{m^4}{Q^4}\right)\right]} \]

For \(d = 4\) (\(\epsilon \to 0\)), \(\Gamma(\epsilon) = 1/\epsilon - \gamma_E + O(\epsilon)\), and the \(1/\epsilon\) pole represents the ultraviolet divergence. The finite part contains logarithms \(\ln(Q^2/\mu^2)\) and power corrections of \(m^2/Q^2\).

Consistency Check

\(m = 0\) limit: Setting \(m = 0\), we have \(\Delta = x(1-x)Q^2\), and the \(x\) integral gives:

\[ \int_0^1 dx\;[x(1-x)]^{-\epsilon} = \frac{\Gamma^2(1-\epsilon)}{\Gamma(2-2\epsilon)} \]

This agrees with the first term in the result above.

Dimensional analysis: \([I] = d - 4 = -2\epsilon\). The right-hand side has dimension \((Q^2)^{-\epsilon}\), which gives \(-2\epsilon\), consistent.

Comparison with the Feynman parameter representation: For \(m = 0\), the Feynman parameter integral can be evaluated analytically, yielding \(\Gamma^2(1-\epsilon)/\Gamma(2-2\epsilon)\). This confirms agreement with the \(s = 0\) residue of the Mellin–Barnes representation.