Iodine is a trace constituent of the atmosphere that is particularly efficient at forming new particles. While sulfuric acid (H2SO4)1,2,3, methanesulfonic acid1,4 and nitric acid5 all require an additional vapour (ammonia, NH3 or dimethylamine (DMA)) to form particles, highly oxygenated organic molecules (HOMs)6 and iodine7,8,9 can do so alone. Iodine nucleation rates exceed those of H2SO4 (in excess NH3) at comparable concentrations of iodic acid (HIO3)10. Furthermore, HIO3 growth rates of nanoparticles are both charge- and dipole-enhanced, exceeding the neutral collision rate10,11.

Currently, iodine particle formation is rarely represented in atmospheric models—such models form most particles from the nucleation of H2SO4 and include iodine primarily because of its ozone-destroying potential12. While sulfur emissions are projected to decrease due to pollution control measures (probably to a few tens of teragrams of SO2 per year by 2100 (ref. 13), iodine emissions have been increasing due to human activity. Iodine is primarily emitted from oceans by the reaction of O3 with iodide (I) dissolved in surface waters, which liberates volatile iodine species (hypoiodous acid (HOI) and iodine (I2)) to the atmosphere14,15. This marine source is enhanced as a result of O3 pollution on local and hemispheric scales16,17 as well as the thinning of sea ice18, and now accounts for iodine emissions of ~3 Tg yr−1 (refs. 19,20). Over the past 70 years, iodine concentrations have tripled in ice-core records in Greenland18, Alpine glaciers17 and tree-ring records in Tibet21.

Iodine is highly reactive and participates in catalytic reaction cycles that enhance its atmospheric impact. A catalytic role is well known for O3 loss, but has, as of yet, not been suggested for particle formation. Iodine in the lower stratosphere has a 6–15 and 400–1,000 times higher O3 destruction potential per atom than bromine and chlorine22. Extremely low mixing ratios of iodine oxide (IO) radicals (for example, ~0.1 parts per trillion by volume (pptv); IO = 10−13 volume mixing ratio) can therefore affect the lifetime of climate-active gases (for example, O3 and CH4)19,23,24. This chemical reactivity extends to heterogeneous reactions involving aerosol iodide (I)14,15 and iodate (IO3) (refs. 25,26 and references therein), which is the thermodynamically most stable form of iodine. The efficient multiphase chemistry of IO3 is markedly different from that of inert aerosol sulfate (SO42−), which accumulates without further chemical conversion until it is scavenged from the atmosphere by wet or dry deposition.

Iodine is ubiquitous in the atmosphere22,23,27,28, and HIO3 has been detected in coastal marine air9,10,29, the Arctic and Antarctic boundary layer9,10,30,31,32, various continental sites10 and in the lower free troposphere10,33. Several precursors for HIO3 have been suggested: hydrated iodine atoms10,34, hydrated IO radicals34, iodine dioxide (OIO) radicals35 and larger iodine oxides (I2O3, I2O4 and I2O5; refs. 34,36,37,38). However, these mechanisms remain speculative and have not been demonstrated experimentally, leaving atmospheric HIO3 observations unexplained. Recent field observations of iodine-induced nucleation over remote oceans31 and of IO3 in stratospheric aerosols22 suggest a widespread role of iodine particle formation, but the conundrum of the missing HIO3 source mechanism blocks our ability to connect iodine sources to particle formation in atmospheric models.

Results and discussion

CLOUD measurements

In this Article we report iodine chemistry and particle formation experiments under marine boundary layer conditions at the CERN CLOUD chamber (Methods). Because of the large chamber volume (26.1 m3) and associated long wall-loss lifetime (~8 min; comparable to typical condensation rates in the atmosphere), precursor gas-phase concentrations do not need to be increased above atmospheric levels (Supplementary Table 1). Experiments were conducted at 283 K and 263 K, with I2 at a typical volume mixing ratio of 8 pptv (range of <0.5–330 pptv), 40% relative humidity (RH, <3–90%) and 40 ppbv O3 (<1–80 parts per billion by volume (ppbv)). The chemistry is driven by photolysis of I2, which is measured by cavity-enhanced differential optical absorption spectroscopy (CE-DOAS; Methods) and bromide chemical ionization mass spectrometry (Br-MION-CIMS). HIO3 is measured quantitatively by NO3-CIMS, and HOI by Br-MION-CIMS. Both instruments also allow insights into the evolution of other iodine species (IO, OIO, I2O2, I2O4 and so on; Methods).

The measurements are accompanied by chemical box modelling, building on state-of-the-art iodine chemistry (Methods). The model is constrained by measurements of I2 concentrations, actinic fluxes, temperature, humidity and losses of molecules to the chamber walls (stainless steel, characterized via H2SO4) and chamber dilution (~2 h). Established iodine chemistry only contains a single reaction predicted from theory35 that could form HIO3 from OIO + OH. This reaction does not form HIO3 in the HOx-free conditions when I2 is photolysed by green light10. Even if OH radicals were present, they would be predominately scavenged by other species. The model base case does not form any HIO3 or HOI under the experimental conditions probed (Fig. 1). Based on the comprehensive experimental evidence of this work, and supported by theoretical calculations, the base case model is extended to include the following two reactions:

$${{{\rm{IOIO}}}}+{{{{\rm{O}}}}}_{3}\to {{{{\rm{IOIO}}}}}_{4}$$
$${{{{\rm{IOIO}}}}}_{4}+{{{{\rm{H}}}}}_{2}{{{\rm{O}}}}\to {{{{\rm{HIO}}}}}_{3}+{{{\rm{HOI}}}}{+}^{(1)}{{{{\rm{O}}}}}_{2}$$

and considers an update to the thermal lifetime of IOIO (extended model, Supplementary Section 3).

Fig. 1: Coincident formation of HIO3 and HOI in the early stages of iodine oxidation.
figure 1

ag, Time-resolved measurements of key iodine species (a,b,d show precursors to HIO3 (f) and HOI (g), and c and e show higher-oxide routes) are compared with model predictions after the start of I2 photolysis at green wavelengths within the CERN CLOUD chamber. Measured concentrations (grey lines) of HIO3 and HOI exceed 107 molecules per cm3 (molec cm−3) within minutes. Established gas-phase iodine chemistry (model base case, dashed blue lines) forms neither HIO3 nor HOI, contrary to the observations, and overestimates the concentrations of IOIO and I2O4. The extended model (solid red line), including reactions (R1) and (R2) and considering a longer thermal lifetime of IOIO, achieves good mass and temporal closure for HIO3, HOI, IOIO and I2O4.

Source data

Figure 1 shows that HIO3 and HOI concentrations rapidly increase to exceed 1 × 107 molecules per cm3 (molec cm−3) within a few minutes of the onset of I2 photolysis by green light (grey lines). While zero HIO3 and zero HOI are predicted by the base case model (current state of the art), the extended model achieves excellent agreement with regard to the measured concentrations and the timing of HIO3 and HOI formation. The extended model also improves the closure of timing and concentrations measured for OIO, IOIO and I2O4. Measured HIO3 concentrations reach a steady state after ~8 min, consistent with the wall-loss lifetime of other sticky molecules3 measured at CLOUD (Extended Data Fig. 1). HOI continues to accumulate due to a lower effective wall uptake. Notably, the IO radical concentrations closely resemble those in the remote marine boundary layer (compare Supplementary Table 1) and do not exceed 1 pptv (1 pptv = 2.68 × 107 molec cm−3 at 273 K and 1 atm pressure). The timing of IO radicals is predicted very well from both the base case and the extended model, reflecting the high level of trust in the gas-phase chemical kinetics during the early stages of the iodine photolysis experiments. Interestingly, iodine oxide clusters IxOy (x ≥ 2, y ≥ 3) larger than IOIO are formed too late to explain the rapid formation of HIO3 as an early generation product (Extended Data Fig. 2 and Supplementary Table 2).

Figure 2 shows that the extended model accurately predicts the measured HIO3 production rates, pHIO3, over a wide range of I radical production rates, pI (104–106 molec cm−3 s−1). Here, pHIO3 is calculated from HIO3 concentration measurements and the well-known loss rates to the chamber walls, and pI is calculated from the photolysis of I2. The HIO3 yield, defined as the ratio of pHIO3 and pI, is a function of the experimental conditions and varies between 10 and 20%. This variability is most pronounced for low pI (<105 molec cm−3 s−1) and is quantitatively explained by the wall loss of HIO3 precursors becoming progressively more relevant at lower gas concentrations. We corroborated that HIO3 formation from I atoms is a multistep process by carrying out an experiment with enhanced stirring (by two fans at the top and bottom of the chamber), thereby decreasing the wall accommodation lifetime of HIO3 from the standard ~8 min to ~2 min, while holding all other parameters constant. The HIO3 concentration decreased by more than one order of magnitude, indicating that the HIO3 suppression exceeds that expected from a change in lifetime alone (Extended Data Fig. 1). The extended model reproduces this superlinear response under the reasonable assumption of efficient reactive uptake of IO radicals on the chamber walls (red dashed line, Fig. 2). Indeed, if the extended model is run while disregarding IO wall loss (blue dashed line, Fig. 2), a constant and high yield of ~20% applies over the full pI range probed.

Fig. 2: HIO3 yield η and rate order.
figure 2

The HIO3 production rate pHIO3 scales in first order with the I atom production rate pI (median (solid line) and 25–75% and 5–95% inter-percentile ranges (dark and light grey shading)). The yield η is substantial (~20%) and near constant for pI larger than 105 molec cm−3 s−1. At smaller pI, losses of intermediates to chamber walls reduce η. This effect is captured by the model (red line (median)) and is explained by IO radical wall losses (compare blue dashed and red dotted lines (medians)). If larger IxOy clusters were the HIO3 precursor, a higher-order yield would be expected—this is not consistent with the observations.

Source data

That HIO3 formation is first order in pI (Fig. 2) explains the presence of HIO3 over remote oceans, where pI is low (Supplementary Table 1)10,31. This finding also carries key mechanistic information, in that it is incompatible with the hypothesis that larger IxOy (x ≥ 3) species are HIO3 precursors34 at CLOUD. If such IxOy were the precursor, the HIO3 yield would not be constant, but would increase progressively with pI, and pHIO3 would follow a higher-order rate law (Supplementary Fig. 2). This is not observed. We regularly detect I2O2 and I2O4, in agreement with predictions by the extended model (Fig. 1 and Extended Data Fig. 3), indicating that there is no fundamental limitation to our analytical capabilities to detect IxOy species. Interestingly, I2O3 is generally not detected, except in experiments that employ extremely high I2 concentrations (ppbv levels), which can bias reaction pathways to favour the formation of larger IxOy species (Supplementary Table 1 and Supplementary Section 5). Quantum chemical calculations support that the \({{{{\rm{I}}}}}_{2}{{{{\rm{O}}}}}_{3}\cdot {{{{{\rm{NO}}}}}_{3}}^{-}\) cluster is thermally stable (Supplementary Fig. 3) and should be observable. Including the formation of HIO3 from IOIO in the extended model reduces the predicted I2O3 by approximately a factor of two (Extended Data Fig. 3), and improves predictions about IOIO, in close agreement with observations (Fig. 1). The remaining discrepancy for I2O3 reflects the uncertainty in larger IxOy chemistry39. We conclude that IxOy species larger than IOIO are not needed as precursors for HIO3 under typical conditions at CLOUD.

HIO3 formation from IOIO is robust against variations in O3, H2O and temperature (Extended Data Fig. 4 and Supplementary Fig. 1). This suggests that neither O3 nor H2O are rate-limiting to HIO3 formation under the conditions probed. The rate-limiting step is the formation of IOIO, which is fully converted into HIO3 (Extended Data Fig. 4). We observe excellent closure between pHIO3 and pIOIO during the O3 ramps, where pIOIO is based on the well-known IO + IO rate coefficients40.

At O3 concentrations below a few ppbv, the chemistry slows down sufficiently that other sinks become relevant for IOIO (for example, wall loss and thermal decomposition), resulting in a slight dependence of the pHIO3-to-pIOIO ratio on O3. That slight dependence is captured by the extended model (assuming an IOIO wall uptake coefficient γwall(IOIO) = 1). In contrast, a pronounced O3 sensitivity would be expected if IOH2O or OIO were HIO3 precursors (Supplementary Fig. 1). The absence of an O3 and H2O sensitivity is difficult to reconcile with any mechanism that does not quantitatively convert a single precursor. The comprehensive evidence (Supplementary Table 2 and Supplementary Section 2) strongly supports a rapid and quantitative conversion of IOIO into HIO3 and HOI.

Quantum chemical calculations

We employed quantum chemical calculations (density functional theory (DFT) methods M062X/aug-cc-pVTZ-PP, followed by coupled-cluster single-point energy corrections; Methods) to explore the reactivity of IOIO with O3, H2O and other available reactants to form HIO3 and HOI. IOIO reacts reasonably quickly with O3 to form HIO3, HOI and singlet oxygen via reaction sequences (R1) and (R2).

Figure 3 shows the reaction coordinate. The reactions (R1) and (R2) are exothermic and free of prohibitively large barriers. Accurately predicting energies and rate coefficients for iodine is challenging because of the inherent complexity of iodine atoms (atom size, number of electrons and relativistic effects). The strong sensitivity towards varying levels of theory is illustrated by comparing bond dissociation energies (BDEs) and proton affinities for simple iodine oxides where measurements are available (Table 1). The method used in this study has improved skill in the coupled-cluster part of the calculations, primarily due to a more balanced description of the basis set on iodine and the other atoms (Methods), and is found to reproduce experimental values within ~3 kcal mol−1 (with the exception of the OIO BDE), which translates into approximately one order of magnitude uncertainty in rate constants.

Fig. 3: Quantum chemical calculations support HIO3 and HOI as co-products of hypoiodide IOIO oxidation.
figure 3

Reaction coordinate for the gas-phase reactions (R1) and (R2) as free energy ΔG(T = 298 K). The energies are calculated using theory at the CCSD(T)/CBS(T,Q)//M062X/aug-cc-pVTZ-PP level of theory. ΔG(TS3) (not rate-limiting) is calculated at the CCSD(T)/aug-cc-pVTZ-PP//M062X/aug-cc-pVTZ-PP level, due to memory limitations. The reaction coordinate supports that atmospheric concentrations of O3 and H2O lead to a quantitative conversion of IOIO into HIO3, HOI and singlet O2.

Table 1 Comparison of different levels of theory with experimental values

The transition states in Fig. 3 translate into the rate coefficients for reactions (R1) and (R2) at 298 K as shown in Table 1 (for temperature dependencies see Supplementary Fig. 6). Notably, the experimentally derived k1 ≥ 1.5 × 10−13 molec−1 cm3 s−1 is supported within the error margins of theory and maintains the quantitative conversion of IOIO into HIO3 even at low O3 concentrations (Supplementary Fig. 1). Our results led to a reassessment of the thermal lifetime of IOIO, which is predicted to be substantially longer than previously thought (Table 1), consistent with observations of IOIO (Extended Data Fig. 3), and its persistently quantitative conversion into HIO3 even at extremely low O3 concentrations at 263 K (Supplementary Fig. 5 and Supplementary Section 3). Reaction (R2) is predicted to proceed with k2 = 5.7 × 10−16 molec−1 cm3 s−1 at 298 K (Table 1 and Supplementary Fig. 6), corresponding to a typical conversion of IOIO4 into HIO3 within fractions of a second. Competing pathways of IOIO4 into other products than HIO3 were investigated (Supplementary Fig. 4, Supplementary Table 4 and Supplementary Section 3), but found to be unlikely. The marginal detection of IOIO4 (Extended Data Fig. 3 and Supplementary Section 2.4) is consistent with a value of k2 ≈ 2.0 × 10−16 molec−1 cm3 s−1 at 263 K. The detection of IOIO4 at the observed levels suggests that reaction (R2) is enhanced by water reacting with hot IOIO4 (Supplementary Section 3.3); assuming a lower k2 from thermalized IOIO4 leads to IOIO4 accumulation in the extended model that is not observed. We recommend temperature-dependent rate coefficients for k1 and k2 for the development of atmospheric models (Supplementary Section 3). Overall, the theory-predicted rates support the experimentally derived rates within the uncertainty of the calculations.

Atmospheric observations

The laboratory-derived mechanism can explain field measurements of HIO3 concentrations in the remote free troposphere. We use concurrent measurements of HIO3 (in situ, NO3-CIMS), IO radicals (near-observatory, MAX-DOAS) and particle surface area measurements at the Maïdo observatory41 to assess the relevance of CLOUD findings in the real world. The observatory is located in the southern Indian Ocean on Réunion Island at an elevation of 2,200 m, and is frequently exposed to lower free tropospheric air (mornings) and anabatic orographic flows from the ocean (afternoons). The laboratory conditions at CLOUD closely match the conditions at the Maïdo observatory (Supplementary Table 1 and Methods) regarding IO concentrations (single pptv), condensational sink (~10−3 s−1) and temperature (~283 K).

Figure 4 shows pHIO3 in the field and laboratory on a common IO radical concentration axis. pHIO3 is calculated from HIO3 concentrations and the condensation sink surface area, assuming a steady state. IO radical concentrations are measured directly at the Maïdo observatory, and taken from the extended model at CLOUD. The solid line shown in Fig. 4 is not a fit to the data; it corresponds to pIOIO at 283 K and serves as a transfer standard to propagate the mechanistic finding of quantitative IOIO conversion into HIO3 from CLOUD (Extended Data Fig. 4) to the field observations. The excellent consistency between the laboratory experiments and field observations demonstrates the atmospheric relevance of the proposed HIO3 mechanism.

Fig. 4: Comparison with field measurements.
figure 4

Good consistency is observed between HIO3 production rates measured in the CLOUD laboratory (red) and at the Maïdo field site (blue). IO radical concentrations at CLOUD overlap with those found in the remote lower free troposphere. The solid black line is the IOIO formation rate from IO radicals (at 283 K), and corresponds to the rate-limiting step of HIO3 formation under both field and laboratory conditions.

Source data

The ability of our HIO3-formation mechanism to predict simultaneous field measurements of HIO3 and IO radicals in the remote free troposphere is anything but trivial (Supplementary Fig. 7), and demonstrates the ability to approximate atmospherically relevant experimental conditions at CLOUD. Interestingly, HIO3 concentrations at Maïdo increase rapidly already under twilight conditions during sunrise (Supplementary Fig. 7 and Supplementary Section 4). He and colleagues10 had predicted the efficient formation of iodine oxoacids under cloudy daylight conditions, and Supplementary Fig. 7 provides field evidence in support of the rapid activation of iodine reservoir species into iodine oxoacids in the absence of ultraviolet irradiation.

Atmospheric implications

The mechanism provides a source of HIO3 that is effective even at low iodine concentrations, and will allow atmospheric models to test HIO3 field observations. Such model development will also help guide future laboratory experiments and field observations. The near-linear rate law of pHIO3 in pI also enables HIO3 formation and subsequent particle formation beyond hotspots at lower iodine concentrations in the background atmosphere31,42,43.

The gas-phase formation mechanism of HIO3 we present here facilitates a missing connection between iodine sources and particle formation in atmospheric models, as illustrated in Fig. 5. The activation of iodine reservoir species (Fig. 5, step 1) liberates iodine radicals, which rapidly form IO radicals and HIO3 (step 2) via reactions (R1) and (R2). Iodine oxoacid particle formation and growth (step 3) is driven by HIO3 in most atmospheric environments. Indeed, recent field observations of particle formation events over the remote Arctic Ocean indicate that all of the observed events were driven by HIO3 (ref. 31). IxOy species may also contribute locally in coastal hotspots with extremely high iodine concentrations. Freshly nucleated iodine particles are composed almost entirely of HIO3 (ref. 10); HIO3 is a strong acid (pKa = 0.8; ref. 44) that dissociates to form IO3. IO3 is known to undergo reduction reactions that ultimately form more volatile iodine species (for example, HOI, I2 and IO), which are re-emitted to the gas phase (step 4). Field observations and laboratory experiments show that IO3 is reduced via iron redox chemistry, H2O2, nitrite, photosensitized reactions, photolysis and numerous other species (refs. 25,26 and references therein), with the overall effect of recycling iodine to the gas phase. The HIO3 formation mechanism thus completes a catalytic iodine reaction cycle, by which a single iodine atom can repeatedly form HIO3, driving particle formation. For each HIO3 molecule produced from I, three O3 molecules are consumed. The re-emission of reduced iodine species thus constitutes a multiphase reaction cycle that destroys O3.

Fig. 5: Simplified gas-phase iodine chemistry in the remote atmosphere.
figure 5

After activation of iodine reservoirs (step 1), HIO3 is efficiently formed (step 2) and subsequently nucleates and grows particles extremely efficiently (step 3). Iodate (IO3) can be reduced and re-emitted to the gas phase (step 4), closing an iodine-catalysed reaction cycle forming particles and destroying O3. HIO3 formation from IOIO links iodine sources and new particle formation even at lower IO concentrations. This mechanism is currently missing from atmospheric models.

That iodine recycling controls the iodine partitioning between the gas and particle phases is corroborated by field measurements of the size-resolved iodine activity in radioactive fallout45. Among the primary radioactive elements, 132Te, 137Cs and 103Ru abundances were found to correlate with the aerosol volume distribution, whereas 131I correlated with the aerosol surface area distribution instead. These empirical observations hint at efficient recycling occurring on timescales of hours to days, consistent with rapid HIO3 formation. Notably, although the reactive uptake of HOI on aerosols is known to be fast46, this reaction de facto removes halides from aerosols to the gas phase. A gas-phase source of HIO3 adds iodine to particles and, in conjunction with iodine recycling, provides a plausible explanation for the correlation of particulate 131I with the aerosol surface area distribution at the molecular level. Particulate IO3 is the primary reservoir of total (gas and particle) iodine in the stratosphere22. Whether HIO3 forms in the stratosphere—and controls iodine partitioning between the gas and particle phases—deserves further study. The HIO3-formation mechanism fills a gap in the representation of the geochemical iodine cycle in current atmospheric models.

Iodine particle formation has heretofore been considered to have only limited global importance19. This deserves re-evaluation in light of efficient HIO3 formation even at low concentrations, the catalytic role of iodine in particle formation, and the increased global iodine source in recent decades. Iodine particle formation is probably already relevant on global scales today, and will become even more important in view of decreasing global sulfur emissions and increasing iodine emissions in a future climate.


CLOUD experiments

Laboratory experiments were carried out at the CERN CLOUD chamber3,50 in Geneva, Switzerland as part of the CLOUD12 and CLOUD13 campaigns during 2017 and 2018. The CLOUD chamber is a temperature-controlled, electropolished stainless-steel reaction vessel with a volume of 26.1 m3. Experiments were carried out at temperatures of 283 and 263 K. The chamber was operated as a continuous-flow reactor, and ultra-pure N2 and O2 at 250–300 l min−1 were continuously replenished at a pressure of 1 atm, resulting in an air exchange time of ~80 min. Two fans at the top and bottom of the chamber established near-homogeneous mixing (mixing time ~2 min). Trace gases were added at the bottom of the chamber. I2 was produced from sublimating iodine crystals (Sigma-Aldrich, 99.999% purity), and concentrations inside the chamber were varied in the range 0.5 pptv < [I2] < 330 pptv (typically ~8 pptv). O3 was generated from UV irradiation of dry synthetic air, and the chamber was humidified using ultrapurified water, resulting typically in [O3] = 40 ppbv (range < 1–80 ppbv) and RH = 40% (<3–90%).

A typical experiment explored the formation of HIO3 following the selective photolysis of I2 using green light (light-emitting diodes (LEDs) centred at 523 nm, I2 photolysis frequencies \({j}_{{{{{\rm{I}}}}}_{2}}\le {6.5}\times {10}^{-3}\,{{{{\rm{s}}}}}^{-1}\)) in the presence of O3 and humidity (Fig. 1). Actinic frequencies were spectrally determined using a spectrometer and dedicated iodine actinometry experiments (Supplementary Section 6.3). Actinic fluxes of light sources at variable intensity were monitored during actual experiments by photodiodes. Sensitivity studies during individual experiments followed the response in the HIO3 concentration to variations in O3 (for example, Supplementary Fig. 1), chamber wall loss during variations of the fan mixing speed (for example, Extended Data Fig. 1) and by varying selected environmental parameters. The typical duration of individual experiments varied from a few tens of minutes to a few hours, depending on the experimental conditions.

I2 was measured by closed-path CE-DOAS51 using the unique ro-vibronic absorption bands between 508 and 554 nm. CE-DOAS is inherently calibrated from knowledge of the absorption cross-section. The I2 limit of detection is 8 pptv for an integration time of 10 min. Median I2 concentrations were below 8 pptv during most experiments, but elevated to up to 1.7 ppbv to calibrate the Br-MION-CIMS, which also provides precise I2 measurements at low concentrations. The Br-MION-CIMS is composed of an atmospheric-pressure interface–time of flight mass spectrometer (APi-TOF) coupled to a chemical ionization unit, using dibromomethane (CH2Br2) as the reagent gas. The CH2Br2 is fed into the sheath flow of the inlet and illuminated by a soft X-ray source. The produced bromide anions are directed into the sample flow by a negative electric field, and cluster with neutral molecules (I2) in the sample air. The overall uncertainty of the resulting I2 time series is estimated to be better than 30%52. The I2 constraint imposed to the model assimilates the lower bound of the measured I2 time series (within the 30% uncertainty), which results in the best closure between measured and predicted HIO3. Iodine radical production rates, pI, are calculated from the photolysis rate of I2 concentrations.

HIO3 was measured by a NO3-CIMS system comprising an APi-TOF coupled to a chemical ionization unit that uses nitric acid as the reagent gas. It is used extensively for detecting H2SO4, highly oxygenated organic molecules and HIO3. Details of the instrument used in the present study are provided in ref. 53. The NO3-CIMS has an ion filter integrated into its sampling line to avoid confusion with ions and charged clusters from the CLOUD chamber. It thus measures only neutral molecules and clusters in CLOUD. The uncertainty of the HIO3 measurement is estimated to be 50%.

The characteristic time for the deposition of sticky molecules to the chamber walls is 440 s with standard mixing by the fans (Extended Data Fig. 1), as characterized via H2SO4 loss rates. The loss to walls is the well-defined dominant sink of HIO3. Experiments that formed a large particle surface area (measured by nSEMS, nano-SMPS or long-SMPS) competitive to chamber wall loss were discarded in this study to avoid introducing uncertainty due to the other less-well-defined sinks for HIO3 and other iodine species. The HIO3 production rates were calculated from measured concentrations under the assumption of a steady state. Periods with rapid changes of HIO3 concentration are not considered in, for example, Fig. 2.

Box modelling

The photochemical box model builds on the framework described in refs. 22,23,24 and represents state-of-the-art iodine chemistry and HOx chemistry25,48. Here, the model is extended by the chamber auxiliary mechanism, which includes losses of gases to the chamber walls and to particles, losses by dilution and the actinic fluxes of the chamber lights. IO, OIO, IOIO, I2O3, I2O4, HI and HIO3 are assumed to be lost to the walls with the same rate constant as H2SO4, the prototypical sticky molecule. Accommodation of molecules to the CLOUD chamber walls is limited by transport, not by diffusion. Thus, the effective wall accommodation coefficient of molecules (most iodine species are reasonably sticky54,55,56, with accommodation coefficients of multiple tens of percent or even unity) used in the model is enhanced over the accommodation coefficient for individual collisions57. Extended Data Fig. 1 provides evidence for the efficient loss of iodine species to the chamber walls. The model is constrained by measurements of I2, O3 and H2O, photolysis frequencies (I2, IO, OIO, HOI, I2O2, I2O3 and I2O4), temperature and the aforementioned loss mechanisms. HOI is both lost to the walls and produced on the chamber walls through heterogeneous chemistry14, which also proceeds in dark conditions. This study did not make an attempt to describe the uptake and release of HOI at the molecular level. An empirical uptake efficiency of 25%, relative to H2SO4, establishes closure in regard to the temporal evolution and concentrations of HOI (Extended Data Fig. 3). See Supplementary Section 6 for more details.

Quantum chemical calculations

For the reactants, intermediates, transition states and products in Fig. 3 with multiple possible conformers, a systematic conformer sampling was carried out using the MMFF method in the Spartan ’18 program. The conformer sampling algorithm in Spartan allows for pre-optimization and the elimination of duplicate structures, which is computationally more efficient than other conformer sampling approaches like MS-TOR. Geometry optimization and frequencies were calculated using DFT methods (M062X/aug-cc-pVTZ-PP) with the ultrafine grid, followed by coupled-cluster single-point energy corrections at the CCSD(T)//CBS/aug-cc-pV(T,Q)Z-PP level of theory. Iodine pseudopotentials were taken from the Environmental Molecular Sciences Laboratory (EMSL) basis set library58,59. The accuracy of the final energetics is critical to reliably estimate the rate of conversion of IOIO4 to HIO3, which was simulated using the master equation solver for multi-energy well reactions (MESMER) program.

Final product fractions were calculated using the MESMER program60. In the simulation, IOIO + O3 was modelled to directly lead to IOIO4 using the MesmerILT method with a pre-exponential value of 2.7 × 10−14 molec−1 cm3 s−1, which corresponds to the transition-state-theory-derived bimolecular rate. The unimolecular isomerization reactions of intermediate complexes were treated using the SimpleRRKM method with Eckart tunnelling. The MesmerILT method with a pre-exponential value of 2.0 × 10−10 molec−1 cm3 s−1 was used for the bimolecular reaction of IOIO4 with H2O, with the latter set as the excess reactant with a defined initial concentration. All intermediate complexes were assigned as ‘modelled’ with Lennard–Jones potentials of σ = 6.5 Å and ϵ = 300 K. These are identical to those used by Galvez and colleagues for their iodine systems61. MESMER uses the exponential down (ΔEdown) model for simulating the collisional energy transfer; a value of 225 cm−1 was used in the simulations, which is within the 175–275 cm−1 range recommended by MESMER for nitrogen bath gas.

The energetics of ozonolysis reactions are difficult to calculate accurately using single-reference methods. The inherent uncertainties are probably even more pronounced for complex iodine-containing systems. Although no experimental values are available for the gas-phase ozonolysis reaction of iodine systems, proton affinities (PAs) and BDEs of simple molecules such as HI, HOI, IO and OIO are available. Table 1 shows that the differences between the literature values and the theoretical values calculated in this work are less than 3 kcal mol−1 (with the exception of the BDE of OIO). Previous quantum chemical calculations on iodine oxide reactions34,47 are included in Table 1 for comparison, highlighting the improved skill of the method used in this study in the coupled-cluster part of the calculation, as benchmarked through comparisons with experimental PAs and BDEs. Previous studies used a double-zeta basis set (LanL2DZ) for I atoms, but a larger triple-zeta basis set (aug-cc-pVTZ) for O and H atoms, leading to substantial overestimation of the exothermicity of bond-forming reactions involving iodine. Our approach uses a large basis set for all atoms, substantially reducing this overestimation.

Field measurements

The field data were collected during an intensive operating period in April 2018 at the Maïdo observatory41, Réunion island, southern Indian Ocean (21° S, 55° E). The observatory is located at 2,200 m above sea level and is frequently exposed to lower free tropospheric air (mornings) and flows from the ocean (afternoons). Near-instrument altitude volume mixing ratios of IO radicals were retrieved from CU MAX-DOAS scattered sunlight observations. The retrieval62,63 leverages the high sensitivity of the limb viewing geometry to the atmospheric layers nearest to the instrument altitude, allowing for the parameterization of aerosol effects on the observed light path. Gas-phase HIO3 was measured directly by a NO3-CIMS system using a methodology similar to that used in the laboratory experiments. The instrument was calibrated in the field in its actual field campaign sampling configuration by in situ-produced H2SO4, which resulted in a calibration factor of c = 1.7 × 1010 molec cm−3. This same calibration factor was used for all quantifications, so the determined concentrations here represent lower limits. The uncertainty of the determined [HIO3] was estimated similarly as [H2SO4], at −50% and +100% following the work in ref. 64. Particles were size-selected by a differential mobility particle sizer and counted with a condensation particle counter to determine the available particle surface area. The box modelling constraints are described in Supplementary Section 4.1. TUV calculated spectral fluxes65 were used to determine the photolysis frequencies of individual iodine species.