TITLE: On the difference between a projective chain complex and a level-wise projective chain complex
QUESTION [7 upvotes]: Let R be an associative ring with a unit, and consider the standard projective model structure of non-negatively graded (left) R-module, $Ch_R$. A map $f:M\to N$ in $Ch_R$ is a weak equivalence if it induces isomorphisms $H_kM\to H_kN$ on all homologies, and a cofibration if each $f_k:M_k\to N_k$ is a monomorphism, with a projective R-module as its cokernel.
Let A be a projective R-module. The chain complex $(*)$ $0\to A\to A\to0$ is a projective object in the world of chain complexes, $Ch_R$. So is any direct sum of such chain complexes. It is a known result (see [DS95, 7.10] for instance) that any acyclic object in $Ch_R$ which is level-wise projective is isomorphic to a direct sum of chains complexes as in $(*)$, with $A$'s projective.
We have here two notions, of (i) a chain complex which is level-wise project, and of (ii) a projective object in $Ch_R$. My gut's feeling is that (i) doesn't imply (ii). In the other way around I'm not sure. I would be glad to have an answer in both directions.   :-)
[DS95] = Dwyer & Spalinski, Homotopy theories and model categories.

REPLY [9 votes]: Edit: I put my answer into a broader perspective. 
In the following [We] refers to Weibel's "An Introduction to Homological Algebra". 
Recall that a chain complex $C$ is split exact if it is acyclic and $Z_n$ (the cycles) is a direct summand of $C_n$ for each $n$ [We, Def. 1.4.1, Ex. 1.4.2]. Moreover, let $Ch$ be the category of (unbounded) chain complexes over an abelian category with enough projectives. Then, by [We, Ex. 2.2.1] and my answer in  
$\qquad$https://mathoverflow.net/questions/103056/when-is-an-acylic-chain-complex-contractible 
we have: 

For a chain complex $P$ the following are equivalent: 

$P$ is a projective object in $Ch$ 
$P$ is a split exact complex of projectives 
$P$ is a contractible complex of projectives


Also from the link we obtain the following examples where $Ch_R$ denotes the category of unbounded chain complexes of modules over the ring $R$: 

If $R$ is hereditary (e.g. PID's, Dedekind domains), then the projective objects of $Ch_R$ are exactly the acyclic complexes of projective $R$-modules.
If $R$ is any ring with unit and the acyclic complex $P$ of projective $R$-modules is bounded below, then $P$ is a projective object in $Ch_R$. 

However, not all acyclic complexes of projective or free modules are projective objects in $Ch_R$. A counter-example (due to Dold) is given in [We, Example 1.4.2]: 

Over $R=\mathbb{Z}/4$ the following complex is exact 
$$\cdots \to \mathbb{Z}/4 \xrightarrow{2} \mathbb{Z}/4 \xrightarrow{2} \mathbb{Z}/4 \to \cdots $$
But it's no projective object in $Ch_R$ since $Z_n = \mathbb{Z}/2$ can't be a direct summand of $C_n = \mathbb{Z}/4$. 


Added: Let $Ch_b \subseteq Ch$ be the subcategory of chain complexes that are bounded below. In contrast to $Ch$ the following holds in $Ch_b$: 

The projective objects in $Ch_b$ are exactly the acyclic chain complexes (bounded below) of projectives. 

Proof: Let $P$ be an acyclic chain complex of projectives that is bounded below. We have already seen in example 2 above (compare also [We, Ex. 1.4.1 2.]) that $P$ is projective in $Ch$. Since $P \in Ch_b$ it's a projective object in $Ch_b$  
Now let $P \in Ch_b$ be a projective object. The same proof as in $Ch$ shows that each $P_i$ is projective (consider objects of the abelian category as chain complexes concentrated in a single degree). Also the same proof as in $Ch$ can be used to see that $P$ is acyclic: The mapping cone of $id: P[1] \to P[1]$ yields a short exact sequence 
$$0 \to P[1] \to \operatorname{cone}(id_{P[1]}) \to P \to 0.$$
But $P, P[1] \in Ch_b$ and by definition $\operatorname{cone}(id)_i = P_i\oplus P_{i+1}$ whence it is also bunded below.  So the short exact sequence is in $Ch_b$ and splits since $P$ is a projective object. Hence $id_{P[1]}$ is nullhomotopic. So in particular, $P[1]$  and thus $P$ is acyclic. q.e.d.

REPLY [5 votes]: Here is a general nonsense fact: if $F \dashv U : \mathcal{A} \to \mathcal{B}$ is an adjunction and $U$ preserves epimorphisms, then $F$ preserves projective objects. Epimorphisms in $\textrm{Ch}(R)$ are precisely the levelwise epimorphisms in $\textrm{Mod}(R)$, so if we take $\mathcal{A} = \textrm{Mod}(R)$, $\mathcal{B} = \textrm{Ch}(R)$, and set $U(M)$ to be $M$ considered as a chain complex concentrated in degree $n$, and $F(K_\bullet) = K_n$, we have an adjunction $F \dashv U$ satisfying the above hypotheses, and therefore $F$ preserves projective objects. Thus,
Proposition. A projective chain complex is necessarily levelwise projective. 　◼
The converse is false, as you anticipated. Consider the following chain complexes:
$$P_\bullet = L_\bullet = (R^3 \to R^3) \text{ with differential given by }
\begin{pmatrix}
0 & -1 & 1 \newline
1 & 0 & -1 \newline
-1 & 1 & 0
\end{pmatrix}$$
$$K_\bullet = (R^3 \to R^4) \text{ with differential given by } \begin{pmatrix}
0 & -1 & 0 \newline
1 & 0 & -1 \newline
-1 & 1 & 0 \newline
0 & 0 & 1
\end{pmatrix}$$
Let $f_\bullet : K_\bullet \to L_\bullet$ be the map given in degree $1$ by the identity and in degree $0$ by the matrix
$$f_0 = \begin{pmatrix}
1 & 0 & 0 & 1 \newline
0 & 1 & 0 & 0 \newline
0 & 0 & 1 & 0 \newline
\end{pmatrix}$$
It is clear that $f_\bullet$ is an epimorphism. Geometrically, $P_\bullet$ is the simplicial chain complex of a triangle $S^1$, $K_\bullet$ is the simplicial chain complex of the interval $[0, 1]$ subdivided into 3 segments, and $f$ is the quotient map that identifies the endpoints of $[0, 1]$. Intuitively, we expect that there is no morphism $g : S^1 \to [0, 1]$ such that $f \circ g = \textrm{id}$, and this turns out to be true in the chain complex world as well... provided $R$ is not the trivial ring. Indeed, if $f_0 \circ g_0 = \textrm{id}$ and $f_1 \circ g_1 = \textrm{id}$, then by elementary algebra we must have $g_1 = \textrm{id}$ and
$$g_0 = \begin{pmatrix}
a & b & c \newline
0 & 1 & 0 \newline
0 & 0 & 1 \newline
1 - a & -b & -c
\end{pmatrix}$$
for some $a, b, c$, but if $g_\bullet$ is to be a chain map, $a, b, c$ must satisfy
$$\left\lbrace\begin{aligned} b & = c \newline
a & = 1 + c \newline
a & = b \newline
b & = c 
\end{aligned}\right.$$
which implies $0 = 1$. So $P_\bullet$ cannot be projective in $\textrm{Ch}(R)$, even though it is degreewise projective.