Are all localizing subcategories of stable homotopy categories coreflective? (Q2445288)

Let \(T\) be a triangulated category admitting a combinatorial model. Assuming Vopěnka's principle (a large cardinal axiom), the authors prove that every localizing subcategory of \(\mathcal T\) is coreflective (its inclusion admits a right adjoint) and that every colocalizing subcategory is reflective (its inclusion admits a left adjoint). Thus, in particular, Hom-orthogonality establishes a bijective correspondence between the localizing and colocalizing subcategories of~\(\mathcal T\). This answers, at least partially, questions asked by Neeman, Hovey, Palmieri and Strickland. The proof is not symmetric: the statement on colocalizing subcategories only requires -- besides Vopěnka's principle -- that the underlying category of the model be locally presentable, while the statement on localizing subcategories needs cofibrant generation as well (i.e., that the model be combinatorial), together with several technical results about homotopy colimits in combinatorial simplicial model categories. Moreover, Vopěnka's principle is used differently in the two cases. The results are not quite symmetric either: localizing subcategories are actually proved to be coreflective because they are generated by a single object. Under the same hypotheses, the authors do not know whether this is also true for colocalizing subcategories, although the existence (in ZFC) of non-singly generated torsion-free classes of abelian groups points towards a negative answer. There also remains the question whether these results can be proved without appealing to Vopěnka's principle, although the authors establish a partial result towards showing independence from ZFC. The article is well written. The discussion of the literature, related results and open questions is thorough; all necessary facts about homotopy colimits are conveniently recollected; and the setup is optimized (the results apply equally to ``semi-(co)-localizing'' subcategories), so that for instance the consequences for t-structures and nullity classes can be easily drawn.