The　Bandpass-2　problem　is　a　variant　of　the　maximum　traveling　salesman　problem　arising　from　optical　communication　networks　using　wavelength-division　multiplexing　technology,　in　which　the　edge　weights　are　dynamic　rather　than　fixed.　The　previously　best　approximation　algorithm　for　this　NP-hard　problem　has　a　worst-case　performance　ratio　of　227426.　Here　we　present　a　novel　scheme　to　partition　the　edge　set　of　a　4-matching　into　a　number　of　subsets,　such　that　the　union　of　each　of　them　and　a　given　matching　is　an　acyclic　2-matching.　Such　a　partition　result　takes　advantage　of　a　known　structural　property　of　the　optimal　solution,　leading　to　a　70-2128≈0.5358-approximation　algorithm　for　the　Bandpass-2　problem.　©　2013,　Springer　Science+Business　Media　New　York.