Solution (source code)

= Solution

Let $R_i(x)=\left.\partial x_{g(\theta)}/\partial\theta_i\right|_{\theta=0}$ be the infinitesimal orbit tangent vectors. Since $F_j(x_0)=0$, Taylor expansion gives
$$
F_j((x_0)_{g(\theta)})=\sum_iM_{ji}(x_0)\theta_i+O(|\theta|^2),\qquad M_{ji}(x_0)=\sum_a\frac{\partial F_j}{\partial x_a}(x_0)R_i^a(x_0).
$$
This explains the stated linearization. For a linear <group representation>, $R_i(x)=t_ix$ in the corresponding action convention.

Two assumptions are implicit in the advertised integration formula. \b[The slice must be regular, and the determinant sign must be fixed if the absolute value is omitted.] Uniqueness alone does not imply regularity: translations on the line with $F(x)=x^3$ have a unique gauge-fixed representative but $M(0)=0$. For ordinary real integration the <Dirac delta distribution> change-of-variables formula uses $|\det M|$. The choice $F(x)=-x$ instead has a unique regular zero and negative determinant, illustrating why its sign cannot simply be dropped.

Assume a <regular gauge slice> with one representative on each orbit, and choose the action convention $(x_{g_0})_{g(\theta)}=x_{g(\theta)g_0}$, matching the specified right-invariance of <Haar measure>. For the unique $x_0=x_{g_0}$ with $F(x_0)=0$, put $g=g(\theta)g_0$. The Haar density is one at the identity, so changing from $\theta$ to $F$ near the unique zero gives
$$
I(x):=\int_Gd\mu(g)\,\delta^{(p)}(F(x_g))=\frac1{|\det M(x_0)|}.
$$
The integral is orbit invariant by the right-invariance of <Haar measure>. Thus $\Delta(x)=I(x)^{-1}$ is also orbit invariant, and on the gauge slice it equals $|\det M(x)|$. Insert $1=\Delta(x)I(x)$ into the original integral and use its invariant volume measure and invariant $f$:
$$
\begin{aligned}
\int d^nx\,f(x)&=\int_Gd\mu(g)\int d^nx\,f(x)\Delta(x)\delta^{(p)}(F(x_g))\\
&=\int_Gd\mu(g)\int d^ny\,f(y)\Delta(y)\delta^{(p)}(F(y)).
\end{aligned}
$$
In the second line set $y=x_g$. This proves the <finite-dimensional Faddeev-Popov gauge reduction>
$$
\boxed{\int d^nx\,f(x)=C\int d^nx\,\delta^{(p)}(F(x))|\det M(x)|f(x),\qquad C=\int_Gd\mu(g).}
$$
Choose a positive orientation of the regular slice to obtain the question's $\det M$ form. If the group volume is infinite, its cancellation in normalized quantities is formal or regulated. Multiple intersections and residual zero modes invalidate the simple identity; in a gauge theory these are the <Gribov ambiguity> and residual <gauge transformations>.

The <Yang-Mills action> counts every <gauge orbit> redundantly in its <path integral>, and its quadratic gauge-field operator has gauge zero modes. A <gauge fixing> condition removes this redundancy, while the <Faddeev-Popov determinant> supplies the orbit-to-slice Jacobian. This is a local perturbative construction with the same regularity qualifications as the finite-dimensional derivation.

Take the linear condition $G[A]=F^\mu A_\mu$. Under an infinitesimal <gauge transformation> $\delta A_\mu=D_\mu\omega$, its change is $\delta G=F^\mu D_\mu\omega$. Hence the <Faddeev-Popov operator> is $\mathcal M[A]=F^\mu D_\mu$. For <Lorenz gauge>, $F^\mu=\partial^\mu$. The <gauge covariant derivative> acts in the <Adjoint representation>.

Average the sharp constraint $G=h$ over $h$ with Gaussian weight $\exp[-i\int h^2/(2\xi)]$. This replaces it by $\exp[-i\int G[A]^2/(2\xi)]$. The <Gaussian gauge fixing with an auxiliary field> identity rewrites that factor as
$$
\int Db\,\exp\left[i\int\left(b\cdot G[A]+\frac\xi2b\cdot b\right)\right]\propto\exp\left[-\frac i{2\xi}\int G[A]\cdot G[A]\right],
$$
by completing the square. The bosonic <Nakanishi-Lautrup field> $b$ is nondynamical: its algebraic field equation is $b=-G[A]/\xi$. At $\xi=0$ it is a <Lagrange multiplier> enforcing $G[A]=0$.

The <Grassmann Gaussian integral> exponentiates the determinant, up to a field-independent phase, as
$$
\det\mathcal M[A]\propto\int D\bar c\,Dc\,\exp\left[i\int\bar c\cdot\mathcal M[A]c\right].
$$
The independent anticommuting <Faddeev-Popov ghost field> $c$ and <Faddeev-Popov antighost field> $\bar c$ are spacetime scalars in the <Adjoint representation>. They represent the Jacobian and do not describe physical external fermions. Combining these factors gives
$$
\boxed{\mathcal L_q=-\frac14F^{\mu\nu}\cdot F_{\mu\nu}+b\cdot F^\mu A_\mu+\frac\xi2b\cdot b+\bar c\cdot F^\mu(D_\mu c).}
$$
The parameter $\xi$ selects the Gaussian averaging of the gauge condition. In an anomaly-free regulated treatment, physical <gauge-invariant> observables are independent of this choice. One way to express this is <BRST symmetry>: with $sA=Dc$, $s\bar c=-b$ and $sb=0$, the gauge and ghost terms are $s[-\bar c\cdot(G+\xi b/2)]$. The ghost transformation supplies the <off-shell nilpotence of the Yang-Mills BRST quartet>, and <gauge-fixing parameter independence from BRST symmetry> explains why these auxiliary fields do not change the physical content.