Example. The same four-cycle Schur matrix with fermions [fdip-009M]
AGENTDRAFTED
Take the alternating four-cycle with
\(t,m>0\), set \(a=m^2\), \(s=1+t+a\), and use
\(K=Q_{t,m}=L_t+aI\) now as a complex matrix. Order retained
vertices \(R=(0,2)\) before eliminated vertices \(U=(1,3)\).
Its blocks are
\[A=D=sI_2,\qquad B=C=-\begin {pmatrix}1&t\\t&1\end {pmatrix}.\]
The fermionic Schur formula gives
\[J_U(e^{-\bar \psi K\psi })
=s^2e^{-\bar \psi _R S\psi _R},\]
where
\[\begin {aligned}
S&=\begin {pmatrix}\alpha &-\beta \\-\beta &\alpha \end {pmatrix},\\
\alpha &=s-\frac {1+t^2}{s},\qquad \beta =\frac {2t}{s}.
\end {aligned}\]
This is exactly the real Gaussian effective precision. The
fermionic scalar is \(s^2\); the real Gaussian eliminated
factor in the cited example is \(2\pi /s\).
The two eigenvalues of \(S\) are
\[\frac {a(a+2+2t)}s,\qquad \frac {(a+2)(a+2t)}s.\]
Writing \(\mathcal D=a(a+2)(a+2t)(a+2+2t)\), it follows that
\[\det S=\frac {\mathcal D}{s^2},\qquad
Z_F(K)=s^2\det S=\mathcal D.\]
Positivity of \(K\) also allows the ordinary complex integral
with normalized volume \(\pi ^{-4}d^8z\). The
complex Gaussian theorem gives
\[Z_B(K)=\mathcal D^{-1},\qquad Z_F(K)Z_B(K)=1.\]
In contrast, the original real Gaussian, with exponent
\(-\phi ^TK\phi /2\) and volume \(d^4\phi \), has partition
function \((2\pi )^2/\sqrt {\mathcal D}\). These powers and
constants refer to different field spaces and measures.
At \(t=m=1\), \(s=3\), \(\mathcal D=45\), and
\[S=\begin {pmatrix}7/3&-2/3\\-2/3&7/3\end {pmatrix},
\qquad \det S=5.\]
The Grassmann eliminated factor \(9\) times the retained
factor \(5\) gives \(45\), while the normalized complex-boson
value is \(1/45\). At \(t=2,m=1\), \(s=4\),
\(\det S=105/16\); the product is \(16\cdot 105/16=105\).
These are the same two effective matrices compared in the
weighted-cycle example.
At \(m=0,t>0\), the eliminated block \((1+t)I_2\)
is still invertible and the partial Grassmann formula is
valid. The retained Schur matrix has a constant zero mode,
so \(\det K=0\): the full Grassmann integral vanishes.
The commuting full Gaussian instead diverges in its constant
mode. There is no finite inverse determinant, cancellation
identity or logarithmic contribution at this massless point.