2018MNRAS.478..791N

Recent work has suggested that extreme [O III] emitting star-forming galaxies are important to reionization. Relatedly, [O III]/[O II] has been put forward as an indirect estimator of the Lyman Continuum (LyC) escape fraction (f_esc) at z >= 4.5 when the opaque IGM renders LyC photons unobservable. Using deep archival U-band (VLT/VIMOS) imaging of a recently confirmed overdensity at z∼ 3.5, we calculate tight constraints on f_esc for a sample (N = 73) dominated by extreme [O III] emitters. We find no LyC signal (f_ esc_^ rel^ < 6.3^+0.7_–0.7 per cent at 1σ) in a deep U-band stack of our sample (31.98 mag at 1σ). This constraint is in agreement with recent studies of star-forming galaxies spanning z ∼ 1-4 that have found very low average f_esc. Despite the galaxies in our study having an estimated average rest-frame EW([O III]λ5007) ∼ 400 Å and [O III]/[O II] ∼ 4 from composite SED fitting, we find no LyC detection, which brings into question the potential of [O III]/[O II] as an effective probe of the LyC - a majority of LyC emitters have [O III]/[O II] > 3, but we establish here that [O III]/[O II] > 3 does not guarantee significant LyC leakage for a population. Since even extreme star-forming galaxies are unable to produce the f_ esc_∼10-15per cent required by most theoretical calculations for star-forming galaxies to drive reionization, there must either be a rapid evolution of f_esc between z∼ 3.5 and the epoch of reionization, or hitherto observationally unstudied sources [e.g. ultrafaint low-mass galaxies with log (M/M_☉) ∼ 7-8.5] must make an outsized contribution to reionization.