= Solution
Use finite-dimensional $k$-modules or finite $R$-free lattices. We use the basic relative-projectivity facts that <Krull-Schmidt decomposition> holds with local endomorphism rings, <module vertices> are conjugate <p-subgroups>, and a <module source> $S$ at a <module vertex> $D$ satisfies $S\mid M\downarrow_D$ and $M\mid S\uparrow^G$. <Module induction> is transitive and <Mackey restriction formula> gives <module vertex> containment for restrictions and inductions. In particular, a summand induced from $D$ has <module vertex> conjugate into $D$, and <module restriction> to a subgroup containing a chosen <module vertex> retains a summand with that <module vertex>. The latter follows from the <module source> conditions and minimality: restricting further to $D$ must still contain its <module source>, whose <module vertex> is $D$.
The <module vertex>-$D$ form of <Green correspondence> states that for $N_G(D)\leq H\leq G$ there is a bijection between indecomposable $RG$-modules with <module vertex> $D$ and indecomposable $RH$-modules with <module vertex> $D$. Its maps select the unique <module vertex>-$D$ summands of <module restriction> and induction, using $H$-conjugacy for restricted <module vertices> and $G$-conjugacy for induced <module vertices>. They occur with multiplicity one and have the same <module sources>. Other induction summands have <module vertices> conjugate to proper subgroups of $D$. The <module restriction> remainder is relatively projective for
$$
\mathcal Y=\{H\cap aDa^{-1}:a\in G\setminus H\}.
$$
Relative projectivity for a family means being a summand of a finite sum of inductions from its members. We prove this form; no stronger description of the induction remainder is needed for the application.
Let $U$ have <module vertex> $D$ over $H$. The identity <double coset> in <Mackey restriction formula> gives
$$
(U\uparrow^G)\downarrow_H=U\oplus T.
$$
Since $U$ is relatively $D$-projective, a further Mackey decomposition puts every other <double coset> term in the relatively-$\mathcal Y$ family. No member of $\mathcal Y$ contains an $H$-conjugate of $D$. Otherwise, conjugation within $H$ would give $D\leq H\cap (ha)D(ha)^{-1}$ with $h\in H$, $a\notin H$. Equality of orders then makes $ha\in N_G(D)\leq H$, a contradiction. Therefore the <module restriction> above contains exactly one summand with <module vertex> $H$-conjugate to $D$, namely $U$ once.
Decompose $U\uparrow^G$ into <indecomposable modules>. All their <module vertices> are conjugate into $D$. Each one with a <module vertex> $G$-conjugate to $D$ can take $D$ itself as <module vertex> and contributes a <module vertex>-$D$ summand on <module restriction> to $H$. There can thus be at most one such summand and its multiplicity is one. There must be at least one, because smaller <module vertices> cannot restrict to produce $U$, with <module vertex> $D$, in the displayed decomposition. Define $g(U)$ to be that summand. Its <module restriction> contains $U$ once, with remainder a summand of $T$. All other induction summands have strictly smaller <module vertices>. This is <Green correspondence from Mackey multiplicity>.
Conversely take $M$ of <module vertex> $D$ over $G$ with <module source> $S$. Decompose $S\uparrow^H$ into <indecomposable modules>. Since $M\mid S\uparrow^G=(S\uparrow^H)\uparrow^G$, the local <endomorphism ring> and <Krull-Schmidt decomposition> give $M\mid U\uparrow^G$ for one indecomposable summand $U$ of $S\uparrow^H$. Its <module vertex> is $H$-conjugate to $D$: all <module vertices> are conjugate into $D$, and a smaller one could not induce $M$. The preceding uniqueness makes $g(U)=M$ and identifies $U$ as the unique <module vertex>-$D$ summand of $M\downarrow_H$. Define $f(M)=U$. This proves $f(g(U))=U$ and $g(f(M))=M$.
If $S$ is a <module source> of $U$, then $S\mid U\downarrow_D\mid M\downarrow_D$ and $M\mid U\uparrow^G\mid S\uparrow^G$. These two conditions and the common <module vertex> show that $S$ is also a <module source> of $M$. Thus the correspondence preserves <module sources>, including <trivial source modules>, and the proof is complete under the stated coefficient conventions.
For the required expression, a <p-local module> is a finite sum of inductions from $N_G(Q)$ for nontrivial <p-subgroups> $Q$. Induct on the order of a <module vertex> of $M$. Vertex 1 makes $M$ projective, so choose $L_1=L_2=P_1=0$, $P_2=M$.
For <module vertex> $D\ne1$, take $U=f(M)$ for $H=N_G(D)$. Its <module induction> is p-local and
$$
U\uparrow^G=M\oplus\bigoplus_iX_i
$$
with smaller <module vertices> for the $X_i$. By induction, write
$$
X_i\oplus A_i\oplus P_i\cong B_i\oplus Q_i,
$$
with $A_i,B_i$ p-local and $P_i,Q_i$ projective. Adding these identities and substituting the induction decomposition gives
$$
M\oplus\bigoplus_iB_i\oplus\bigoplus_iQ_i
\cong\left(U\uparrow^G\oplus\bigoplus_iA_i\right)\oplus\bigoplus_iP_i.
$$
Hence the explicit choices
$$
\boxed{L_1=\bigoplus_iB_i,\quad P_1=\bigoplus_iQ_i,\quad
L_2=U\uparrow^G\oplus\bigoplus_iA_i,\quad P_2=\bigoplus_iP_i}
$$
work. All sums are finite and <module vertex> orders strictly decrease, so the induction terminates. If $N_G(D)=G$, induction from that <normalizer> is still allowed by the definition of p-local.
Back to article page