Descent maps and projective image splittings
Let \(R\) be a commutative ring and \(q:M\to P\) a surjective linear map.
If \(M\) is projective and an endomorphism \(e:M\to M\) satisfies \[ q(e(m))=q(m),\qquad q(m)=0\Longrightarrow e(m)=0, \] then \(P\) is projective.
For the same surjection, the following split-sequence facts also hold:
- if \(M,P\) are projective, then \(\ker q\) is projective;
- if \(P,\ker q\) are projective, then \(M\) is projective.
Proof
Step 1. Descend a map through a surjection
More generally, let \(g:M\to Q\) be linear and assume \(\ker q\subseteq\ker g\). For \(p\in P\), choose \(m\) with \(q(m)=p\), and define \(\bar g(p)=g(m)\).
If \(q(m')=p\) too, then \(m-m'\in\ker q\), so \(g(m-m')=0\) and \(g(m)=g(m')\). Thus the value is independent of the lift. To check addition, lift \(p,p'\) by \(m,m'\); then \(m+m'\) lifts \(p+p'\), giving \(\bar g(p+p')=\bar g(p)+\bar g(p')\). A scalar multiple \(rm\) similarly proves scalar linearity.
The descended map satisfies \(\bar gq=g\). It is unique because \(q\) is surjective. This is ProjectiveRange.descend.
Step 2. Descend the endomorphism to a section
Apply Step 1 to \(g=e\). Its second hypothesis gives the needed kernel containment. The resulting map \(\bar e:P\to M\) satisfies \[ q\bar e(q(m))=q(e(m))=q(m). \] Surjectivity of \(q\) therefore gives \(q\bar e=1_P\). Thus \(P\) is a direct summand of the projective module \(M\), and is projective.
Step 3. Construct the kernel retraction
Suppose \(P\) is projective. Lift \(1_P\) through \(q\) to obtain a section \(\sigma:P\to M\) with \(q\sigma=1_P\).
Define \[ r:M\to\ker q,\qquad r(m)=m-\sigma(q(m)). \] This vector lies in the kernel because \(q(m)-q\sigma(q(m))=0\). For \(k\in\ker q\), \(r(k)=k-\sigma(0)=k\). Hence \(r\) retracts the kernel inclusion.
If \(M\) is projective, its direct summand \(\ker q\) is projective. This proves the second claim.
Step 4. Exhibit the splitting isomorphism
The maps \[ m\longmapsto(r(m),q(m)),\qquad (k,p)\longmapsto k+\sigma(p) \] are inverse linear maps between \(M\) and \(\ker q\oplus P\). For \(m\), the returned vector is \(m-\sigma(q(m))+\sigma(q(m))=m\). For \((k,p)\), applying \(q\) gives \(p\), and subtracting \(\sigma(p)\) leaves \(k\).
If both \(\ker q\) and \(P\) are projective, their direct sum is projective, so this isomorphism proves that \(M\) is projective.
References
The local source declarations below give the construction and proof.
Lean formalization
OAI.KLInvariance.ProjectiveRange.descend: IntegralPairings.lean:490.
OAI.KLInvariance.ProjectiveRange.descend_apply: IntegralPairings.lean:494.
OAI.KLInvariance.ProjectiveRange.projective_of_end: IntegralPairings.lean:500.
OAI.KLInvariance.ProjectiveRange.kernelRetraction: IntegralPairings.lean:509.
OAI.KLInvariance.ProjectiveRange.kernel_projective: IntegralPairings.lean:515.
OAI.KLInvariance.ProjectiveRange.splitEquiv: IntegralPairings.lean:526.
OAI.KLInvariance.ProjectiveRange.projective_of_kernel: IntegralPairings.lean:541.