The Milos VF is part of the South Aegean Volcanic Arc (SAVA), an arc which was formed in the eastern Mediterranean by subduction of the African plate beneath the Aegean microplate (Fig. 1; Nicholls, 1971; Spakman et al., 1988; Duermeijer et al., 2000; Pe-Piper and Piper, 2007; Rontogianni et al., 2011). The present-day Benioff zone is located approximately 90 km underneath Milos (Hayes et al., 2018). The upper plate is influenced by extensional tectonics (e.g. McKenzie, 1978; Pe-Piper and Piper, 2013), which is evident on the island of Milos as horst and graben structures (Fig. 2).

The Milos VF is exposed on the islands of the Milos archipelago: Milos, Antimilos, Kimolos and Polyegos. The focus of this study is Milos, which has a surface area of 151 km2. The geology and volcanology of Milos have been extensively studied in the last 100 years. The first geological map was produced by Sonder (1924). This work was extended by Fytikas et al. (1976) and Angelier et al. (1977) and the subsequent publications of Fytikas et al. (1986) and Fytikas (1989). Interpretations based on volcanic facies of the complete stratigraphy were made by Stewart and McPhie (2003, 2006). More detailed studies of single volcanic centres (e.g. the Bombarda volcano and Fyriplaka complex) were published by Campos Venuti and Rossi (1996) and Rinaldi and Venuti (2003). Milos has also been extensively studied for its epithermal gold mineralization, summarized by Alfieris et al. (2013). Milos was known during the Neolithic period for its export of high-quality obsidian. Today the main export product is kaolinite mined from hydrothermally altered felsic volcanic units in the centre of the island (e.g. Alfieris et al., 2013).

The geology of Milos can be divided into four main units: (1) metamorphic basement, (2) Neogene sedimentary rocks, (3) volcanic sequences and (4) the alluvial cover. The metamorphic basement crops out at the south-west, south and south-east of Milos (Fig. 3) and is also found as clasts in many volcanic units. The metamorphic rocks include lawsonite-free jadeite eclogite, lawsonite eclogite, glaucophane schist, quartz–muscovite–chlorite and chlorite–amphibole schist (Fytikas et al., 1976, 1986; Grasemann et al., 2018; Kornprobst et al., 1979). The exposed units belong to the Cycladic Blueschist Unit (Lower Cycladic nappe), whereas eclogite pebbles in the phreatic eruption products called “green lahar” by Fytikas (1977) are derived from the Upper Cycladic nappe (Grasemann et al., 2018).

On top of this metamorphic basement, Neogene fossiliferous marine sedimentary rocks were deposited (e.g. Van Hinsbergen et al. 2004). This sedimentary sequence can be divided into a lower unit A and upper unit B that is unconformably overlain by volcaniclastic sediments (Van Hinsbergen et al., 2004). Unit A is 80 m thick and consists of fluviatile–lacustrine, brackish and shallow marine conglomerate, sandstone, dolomite and limestone. Unit B is 25–60 m thick and consists of sandstone overlain by a succession of alternating marls and sapropels, suggesting a deeper marine setting (Van Hinsbergen et al., 2004). Five volcanic ash layers that contain biotite are found in this Neogene sedimentary sequence, either suggesting that volcanic eruptions in small volume already occurred in the Milos area or that these ash layers are derived from larger eruptions of volcanic centres further away from Milos (van Hinsbergen et al., 2004). Age determinations by bio-magneto- and cyclo-stratigraphy suggested that deposition of Unit A started at approximately 5 Ma, and that Milos subsided 900 m in 0.6 Myr (Van Hinsbergen et al. 2004) due to extension. This subsidence happened ca. 1.0–1.5 Myr before the onset of the main phase of Pliocene–recent volcanism on Milos.

The Pliocene–recent volcanic sequence of Milos has been subdivided into different units by Angelier et al. (1977) and Fytikas et al. (1986). In addition, Stewart and McPhie (2006) provided a detailed facies analysis of the different volcanic units. The subdivision by Angelier et al. (1977) is not constrained well due to their limited amount of age data. The subdivision of volcanic units by Fytikas et al. (1986) and facies descriptions of Stewart and McPhie (2006) are summarized below. It is important to note that according to Stewart and McPhie (2006), the five volcanic cycles described by Fytikas et al. (1986) are difficult to match with existing age data and the continuous progression in volcanic construction (Fig. 4). For example, the first phase of Fytikas et al. (1986), the Basal Pyroclastic Series, contains the large pumice-cone/crypto-dome volcanoes according to Stewart and McPhie (2006). Two of these pumice-cone/crypto-dome volcanoes are much younger and intercalated between the Complex of Domes and Lava Flows (CDLF) of Fytikas et al. (1986).

The first volcanic unit deposited in the Milos area is the Basal Pyroclastic Series (BPS) (Fytikas et al., 1986) or submarine felsic pumice-cone/crypto-dome volcanoes (Stewart and McPhie, 2006, Figs. 2–4). This unit consists of thickly bedded pumice breccia with a rhyolitic–dacitic composition. These rhyolites–dacites are aphyric or contain quartz–feldspar ± biotite phenocrysts. Graded sandstone and bioturbated and fossil-rich (in situ bivalve shells) mudstone are intercalated, indicating a marine environment and a water depth of several hundreds of metres (e.g. Stewart, 2003; Stewart and McPhie, 2006), whereas later degassed magmas with a similar composition intruded as sills and crypto-domes. The BPS has been strongly affected by hydrothermal fluids, especially the proximal deposits (e.g. Kilias et al., 2001).

The second volcanic unit was named the Complex of Domes and Lava Flows (Fytikas et al., 1986), and the volcanic facies of this unit are described as submarine dacitic and andesitic domes by Stewart and McPhie (2006). This phase of effusive submarine volcanism was predominantly andesitic and dacitic in composition and produced microcrystalline rocks with phenocrysts of pyroxene, amphibole, biotite and plagioclase. The eruption centres were mainly located along NNE faults and formed up to 300 m thick deposits extending over areas of 2.5 to 10 km2 around the eruption centres. In the north-eastern part of Milos, an andesitic scoria cone provided scoria lapilli and bombs to deeper water settings. Sandstone intercalated in the CDLF contains both igneous and metamorphic minerals suggesting input from the basement. Rounded pebbles of rhyolite and dacite indicate that some of the volcanic deposits were above sea level or in very shallow, near-shore environments (e.g. Stewart and McPhie, 2006).

The third volcanic unit is called the Pyroclastic Series and Lava Domes (PSLD) by Fytikas et al. (1986) and belongs to the submarine-to-subaerial dacitic and andesitic lava domes of Stewart and McPhie (2006). This highly variable group is dominated by rhyolitic, dacitic and andesitic lavas, domes, pyroclastic deposits and felsic pumiceous sediments (Stewart and McPhie, 2006). Thickness varies between 50–200 m, and the deposits are located in the eastern and northern parts of Milos (Figs. 2 and 3). The initial pyroclastic layers were subaqueously deposited and the extrusion of a dome resulted in the deposition of talus around the margins by mass flow. On top of the dome sand- and siltstone with fossils (Ostrea fossil assemblage) and traction–current structures suggest that the top of the dome was above wave base. The youngest deposits of this unit are dacitic and andesitic lavas and domes. These domes generated subaerial block-and-ash flow and surge deposits. Paleosols within these deposits are a clear indicator that some areas were above sea level. The last unit of the PSLD is represented by large subaerial rhyolitic lava that contains quartz and biotite phenocrysts and is found near Halepa in the southern central part of Milos.

The fourth unit consists of the subaerially constructed rhyolitic complexes of Trachilas and Fyriplaka (CTF) (Fytikas et al., 1986), which Stewart and McPhie (2006) interpreted as subaerial rhyolitic lava–pumice cones. These two volcanic complexes are built from rhyolitic pumice deposits and lavas that contain quartz and biotite phenocrysts (10 modal%–20 modal%). The deposits have a maximum thickness of 120 m and decrease to several metres' thickness in the distal parts. Basement-derived schist is found as lithic clasts (Fytikas et al., 1986). In addition, the Kalamos rhyolitic lava dome, which outcrops on the southern coast of Milos, produced lava that spread westwards to the Fyriplaka beach (Fig. 2). This lava belongs to this fourth phase and is probably derived from an older volcano and not the Fyriplaka complex (Campos Venuti and Rossi, 1996).

The fifth volcanic unit comprises deposits from phreatic activity, especially in the northern part of the Zefiria Graben and near Agia Kiriaki (Fig. 2 of Stewart and McPhie, 2006). Many overlapping craters are surrounded by lithic breccias that are composed of variably altered metamorphic basement clasts and volcanic clasts. This phreatic activity has continued into historic times (Traineau and Dalabakis, 1989). Fytikas et al. (1986) referred to this unit as “green lahar”, although it is indicated that this deposit is not a lahar but the product of phreatic eruptions in the last 0.2 Myr.

Previous geochronological work is summarized in Table 1. Angelier et al. (1977) reported six K–Ar ages (0.95–2.50 Ma). These ages were used in combination with field observations to divide the Milos volcanic succession into four units. However, the samples from Fyriplaka, the fourth unit, were too young to be dated by Angelier et al. (1977). Fytikas et al. (1976, 1986) published 16 K–Ar ages for Milos (0.09–3.50 Ma) including an age of 0.09–0.14 Ma for the Fyriplaka complex. Fytikas et al. (1986) also obtained three K–Ar ages for Antimilos (0.32 ± 0.05 Ma), Kimolos (3.34 ± 0.06 Ma) and Polyegos (2.34 ± 0.17 Ma). Traineau and Dalabakis (1989) dated the very young phreatic deposits by 14C dating and found ages between 200 BCE and 200 CE. Matsuda et al. (1999) published two K–Ar ages of 0.8 ± 0.1 (MI-1) and 1.2 ± 0.1 Ma (MI-4) for the Plakes dome that was also studied by Fytikas et al. (1986). Bigazzi and Radi (1981) published two fission track ages of 1.54 ± 0.18 and 1.57 ± 0.15 Ma for obsidians of Bombarda–Adamas and Dhemeneghaki, respectively. Later fission track studies by Arias et al. (2006) (1.57 ± 0.12 and 1.60 ± 0.06 Ma) confirmed these ages. The fission track ages are younger than the K–Ar ages given by Angelier et al. (1977; 1.84 ± 0.08 Ma for Dhemeneghaki) and Fytikas et al. (1986; 1.71 ± 0.05 Ma for Bombarda). In the most recent geochronological study of the Milos VF, Stewart and McPhie (2006) published four SHRIMP U–Pb zircon ages: Triades dacite facies (1.44 ± 0.08 and 2.18 ± 0.09 Ma), Kalogeros crypto-dome (2.70 ± 0.04 Ma) and the Filakopi Pumice Breccia (2.66 ± 0.07 Ma). All uncertainties reported here are 1 standard deviation uncertainties as reported in the original publications, except for the 14C ages for which uncertainties were not specified.

The previous geochronological work for the Milos VF is mainly based on K–Ar ages. However, K–Ar ages may show undesirable and unresolvable scatter due to various problems: (1) inaccurate determination of radiogenic argon due to either incorporation of excess argon or incomplete degassing of argon during the experiments; (2) inclusion of cumulate or wall rock phenocrysts in bulk analyses; (3) disturbance of a variety of geological processes such as slow cooling or thermal reheating; (4) unrecognized heterogeneities due to separate measurements of potassium and argon content by different methods; (5) requirement of relatively large quantities (milligrams) of pure sample (e.g. Lee, 2015). In addition to these methodological issues, in the case of Milos we observe that hydrothermal alteration caused substantial kaolinitization, in particular of the felsic volcanic samples, that most likely has affected the K–Ar systematics. Some of these issues are also valid for the 40Ar/39Ar method. However, the K–Ar method does not allow testing whether ages are compromised.

40Ar/39Ar ages only need isotopes of argon to be measured from a single aliquot of sample with the same equipment that can eliminate some of the problems with sample inhomogeneity. Furthermore, step heating and multiple single-fusion experiments can shed light on sample inhomogeneity due to partial alteration effects. The high sensitivity of modern noble gas mass spectrometers for 40Ar/39Ar measurements results in very small sample amounts needed for analysis, which can yield more information on the thermal or alteration histories than larger samples. Moreover, other argon isotopes (36Ar, 37Ar and 38Ar) can be used to infer some information about the chemical compositions (i.e. Ca and Cl) of samples. A high-resolution laser incremental heating method of 40Ar/39Ar dating allows us to resolve the admixture of phenocryst-hosted inherited 40Ar in the final temperature steps of the incremental-step heating experiments.

Samples were collected from all major volcanic units on Milos island based on the studies of Fytikas et al. (1986), Stewart and McPhie (2006), and our own observations in the field. Photos of the sample locations and thin sections can be found in Supplement file I. Approximately 2 kg of fresh pumice clasts or lava was sampled from each unit. Samples were cut into ∼ 5 cm3 cubes using a diamond saw to remove potentially altered surfaces and obtain the fresh interior parts. These cubes were ultra-sonicated for 30 min in demi-water to remove dust and seawater and dried in an oven overnight at 50 ∘C. Dry sample cubes were crushed in a steel jaw crusher, and this fraction was split into two portions of roughly equal size. One of them was powdered in an agate shatter box and agate ball mill to a grain size of less than 2 µm for the major-element analysis. The second fraction was sieved to obtain a grain size of 250–500 µm for 40Ar/39Ar dating.

Heavy-liquid density separation techniques (IJlst, 1973) were used to purify mineral separates (groundmass, biotite, amphibole) required for the 40Ar/39Ar dating. Different densities of heavy liquids were used to obtain groundmass (2700 ≤ρ≤ 3000 kgm-3), biotite (2900 ≤ρ≤ 3100 kgm-3) and amphibole (∼ 3100 ≤ρ≤ 3200 kgm-3). A Frantz isodynamic magnetic separator was used to remove the magnetic minerals from the non-magnetic minerals and groundmass. The samples for 40Ar/39Ar analysis were purified by handpicking under a binocular optical microscope to select mineral grains without visible alteration and inclusions.

The mineral and groundmass samples were wrapped in either 6 or 9 mm aluminium foil and packed in 20 mm aluminium cups, which were vertically stacked. Based on stratigraphy and previous geochronological constraints > 1 Ma samples and the < 1 Ma samples were irradiated for 7 and 1 h, respectively, in irradiation batches VU108 and VU110 in the Cadmium-Lined In-Core Irradiation Tube (CLICIT) facility of the Oregon State University Training Research, Isotopes, General Atomics (TRIGA) reactor. The neutron flux for all irradiations was monitored by standard bracketing using the Drachenfels sanidine (DRA; 25.52 ± 0.08 Ma, modified from Wijbrans et al., 1995, and calibrated relative to Kuiper et al., 2008) and Fish Canyon Tuff sanidine (FCs; 28.201 ± 0.023 Ma, Kuiper et al., 2008) with Min et al. (2000) decay constants.

In total, 24 samples (8 groundmasses, 15 biotites and 2 amphiboles; for sample G15M0026 both biotite and amphibole were analysed) were measured by either 40Ar/39Ar fusion and/or incremental heating techniques. For incremental heating experiments, 80–100 grains per sample were loaded into a 25-hole (surface per hole ∼ 36 mm2) copper tray together with single-grain standards in ∼ 12 mm2 holes. The tray was prebaked in a vacuum (10-5–10-6 mbar) at 250 ∘C overnight to remove atmospheric argon and subsequently baked overnight at 120 ∘C in the ultra-high vacuum sample chamber (< 5 × 10-9 mbar) and purification system connected to a Thermo Scientific Helix multi-collector (MC) mass spectrometer.

Samples and standards were heated with a focused laser beam at 8 % power using a 50 W continuous-wave (CW) CO2 laser. The released gas was cleaned by exposure to a cold trap cooled by a Lauda cooler at -70 ∘C, a SAES NP10 at 400 ∘C, a Ti sponge at 500 ∘C and cold SAES ST172 Fe–V–Zr sintered metal. The five isotopes of argon were measured simultaneously on five different collectors: 40Ar on the H2-Faraday, 39Ar on the H1-Faraday or the H1-CDD, 38Ar on the AX-CDD, 37Ar on the L1-CDD, and 36Ar on the L2-CDD for 15 cycles with 33 s integration time (CDD: compact discrete dynodes; AX: axial; H: high-mass side; L: low-mass side). The Faraday cups on H2 and H1 were equipped with amplifiers with 1013 Ω resistors in their feedback loop. Procedural blanks were measured every two or three analyses in different sequences, and air shots were measured every 8–12 h to correct the instrumental mass discrimination. The gain between different collectors was monitored by measuring CO2 on mass 44 in dynamic mode on all collectors. Gain was generally stable over periods of weeks. Note that because samples, standards and air calibration runs are measured during the same period, gain correction does not substantially change the final age results. The raw mass spectrometer data output was converted by an Excel macro script designed in-house to be compatible with the ArArCalc 2.5 data reduction software (Koppers, 2002). The 40Ar/36Ar atmospheric air value of 298.56 from Lee et al. (2006) is used in the calculations. The correction factors for neutron interference reactions are (2.64 ± 0.02) × 10-4 for (36Ar/37Ar)Ca, (6.73 ± 0.04) × 10-4 for (39Ar/37Ar)Ca, (1.21 ± 0.003) × 10-2 for (38Ar/39Ar)K and (8.6 ± 0.7) × x10-4 for (40Ar/39Ar)K. All uncertainties are quoted at the 1σ level and include all analytical errors (i.e. blank, mass discrimination with neutron interference correction and analytical error in J-factor, the parameter associated with the irradiation process).

A reliable plateau age is defined as experiments with at least three consecutive steps overlapping at 2σ, containing > 50 % of the 39ArK, with a mean square weighted deviate (MSWD) value < 2.5 and with an 40Ar/36Ar inverse isochron intercept that does not deviate from atmospheric argon at 2σ. All the inverse isochron ages used the same steps as used in the weighted mean ages, and all relevant analytical data for the age calculations following standard practices (Schaen et al., 2020) can be found in Supplement file II.

Major-element concentrations were measured by X-ray fluorescence spectroscopy (XRF) on a Panalytical AxiosMax. A Panalytical Eagon2 was used to create 40 mm fused glass beads of Li2B4O7/LiBO2 (65.5:33.5 %, Johnson & Johnson Spectroflux 110) with a 1:6 sample–flux ratio that were melted at 1150 ∘C. Sample powders were ignited at 1000 ∘C for 2 h to determine loss on ignition (LOI) before being mixed with the Li2B4O7/LiBO2 flux. Interference-corrected spectra intensities were converted to oxide-concentrations against a calibration curve consisting of 30 international standards. The precision, expressed as the coefficient of variation (CV), is better than 0.5 %. The accuracy, as measured on the international standards AGV-2, BHVO-2, BCR-2 and GSP-2, was better than 0.7 % (1 relative standard deviation, RSD) (Supplement file III).

The minimum and/or maximum eruption volume of each volcano during each eruption period is derived from the ranges of thickness and surface areas that are reported in Campos Venuti and Rossi (1996) and Stewart and McPhie (2006). We converted these volumes to dense rock equivalent (DRE) based on the magma type of different deposits. This analysis only includes the onshore deposits and results in a smaller estimate for larger pyroclastic volumes. The DRE volume is calculated using the equation of Crosweller et al. (2012): 1DRE(km3)=tephra vol(km3)tephra density(kgm-3)magma density(kgm-3).

Tephra density is assumed to be 1000 kgm-3 (Crosweller et al., 2012). Magma density varies depending on the magma type. Here we used 2300 kgm-3 for rocks with a SiO2 range of 65 wt%–77 wt% and 2500 kgm-3 for all samples with SiO2 < 65 wt%. DRE corresponds to the unvesiculated erupted magma volume. Therefore, we did not convert the volume of some crypto-dome and lavas from Profitis Illias (G15M0017), Triades (G15M0021-24), Dhemeneghaki (G15M0032B) and Halepa (G15M0013) to the DRE since they contain less than 5 % vesicles.

In this section, we present our groundmass, biotite and amphibole 40Ar/39Ar results for 11 volcanic units of Milos. The 40Ar/39Ar ages range from 0.06 to 4.10 Ma and cover most of the major volcanic units of Milos. Tables 2 and 3 show the 40Ar/39Ar results of incremental heating steps and single-grain fusion analyses, respectively. Note that the Irr-ID column in these two tables represents the irradiation ID of the analytical experiment (e.g. VU108-, VU110-) and the top right superscripts (G, B, A, O) in the sample IDs (e.g. G15M0029G, G15M0021B) refer to groundmass, biotite, amphibole and obsidian.

Major-element results are given in Table 4. The SiO2 compositions range from 54 to 78 wt% (basaltic andesite to rhyolite, see Fig. 10a). The most felsic samples (SiO2 > 75 wt%) belong to the Fyriplaka and Trachilas complexes. Our data overlap with those of previous studies and display a similar range in SiO2-K2O (Francalanci and Zellmer, 2019, and references therein). The samples of Polyegos are similar to the Fyriplaka and Trachilas complexes, whereas the older Milos samples overlap with Kimolos and Antimilos (Fytikas et al., 1986; Francalanci et al., 2007).

Although some samples of Antimilos are tholeiitic, all of the Milos volcanic units belong to the calc-alkaline and medium- to high-K series (Fig. 10b). A mafic inclusion, sample G15M0022, has high K2O (6 wt%), similar to sample G15M0021 (7.2 wt%). Both of them were collected from the Cape Vani area (Fig. 2). The SiO2 wt% vs. our 40Ar/39Ar ages diagram (Fig. 11b) shows that there is a tendency for the volcanic units to become more felsic over time. In the diagram with K2O/SiO2 vs. age there is no significant change (Fig. 11c).

Figure 11a shows the cumulative volcanic output volume of the Milos VF over time. This diagram shows that the Milos VF can be separated into three periods: periods I (∼ 3.3–2.13 Ma) and III (1.48–0.00 Ma) are characterized by low volcanic output volumes, whereas Period II (2.13–1.48 Ma) shows a rapid increase in volcanic output volume. Periods I and II are built up in a submarine setting, whereas Period III is in a subaerial setting. The Milos VF was largely (∼ 85 % by volume) constructed in a submarine setting before ∼ 1.48 Ma (periods I and II) (Fig. 11a). During Period III (1.48 Ma–present), only a small volume (∼ 15 %) of rhyolitic magma was added from different eruption vents. See the details of periods I–III in Sect. 4.3.2.

More than half of our 40Ar/39Ar ages derived for this study are based on high-resolution laser incremental heating method. All incremental-step heating experiments are reproducible, except for the sample G15M0017 which gave the oldest age. The total fusion experiments of this study gave an at least 5 times smaller analytical uncertainty (1 SE on average ≤ 0.01 Ma) than the previous studies using conventional K–Ar (Angelier et al., 1977; Fytikas et al., 1976, 1986; Matsuda et al., 1999) and SHRIMP U–Pb zircon methods (Stewart and McPhie, 2006). Fission track dating on obsidians of the Milos VF produced two ages (Bigazzi and Radi, 1981; Arias et al., 2006), which seem to overlap with the K–Ar and 40Ar/39Ar ages but with larger uncertainty. U–Pb zircon ages could indicate the timing of zircon formation at high temperature (> 1000 ∘C) in magma chambers significantly prior to volcanic eruption (e.g. Flowers et al., 2005). On the other hand, the lower closure temperature of K-rich minerals (< 700 ∘C) makes the K–Ar and 40Ar/39Ar ages better suited to determine the timing of the extrusion of volcanic products (e.g. Grove and Harrison, 1996; Cassata and Renne, 2013).

The MSWD value, as a measure of the scatter of the individual step ages, is based on the error enveloping around the data point. The decrease in error will automatically cause an increase in MSWD (e.g. York, 1968; Wendt and Carl, 1991). The MSWD values reported in this study are relatively high. In part this is caused by the fact that modern multi-collector mass spectrometers used for 40Ar/39Ar dating can measure the isotope ratios very precisely, which in turn would increase the MSWD. It will be more valuable and challenging to find a plateau or isochron age which meets the MSWD criteria (< 2.5) by modern multi-collector 40Ar/39Ar dating than by K–Ar or 40Ar/39Ar dating using a single detector instrument (e.g. Mark et al., 2009).

Potential drawbacks of the 40Ar/39Ar method are its dependence on neutron irradiation causing the production of interfering argon isotopes that need to be corrected for. The uncertainty in the ages of standards that are required to quantify the neutron flux also needs to be incorporated in the final ages as there are uncertainties related to decay constants (Supplement file II). Finally, recoil can occur during irradiation. Minerals such as biotite can be prone to recoil, yielding slightly older ages (e.g. Hora et al., 2010).

In this section, our 40Ar/39Ar results are compared with previously published geochronological data and subsequently used to refine the stratigraphy of the Milos VF. In the last part, we will discuss the temporal variations in major elements and the volumetric volcanic output rate of the Milos VF.

Figure 12 compares previous published K–Ar, U–Pb zircon and fission track ages from the same volcanic units with the new 40Ar/39Ar data of this study. In general, there is a good agreement; however, 6 ages out of 23 differ significantly from previous studies and will be discussed below.

The obsidian fission track ages (Bigazzi and Radi, 1981; Arias et al., 2006) for the Dhemeneghaki volcano are 0.25 Myr younger than the K–Ar ages (1.84 Ma, Angelier et al., 1977) and the 40Ar/39Ar age of this study (1.825 Ma, G15M0032B). The good agreement between the K–Ar and 40Ar/39Ar ages suggests that the fission track ages record a different, lower-temperature event than the K–Ar and 40Ar/39Ar ages. In addition, the larger uncertainty of fission track ages (> 0.05 Ma) also overlaps with the 40Ar/39Ar age at 2σ. We assume that the 40Ar/39Ar age is the correct extrusion age for the obsidian of the Dhemeneghaki volcano.

Angelier et al. (1977) reported one dacite sample in the north-west of Milos with an age of 1.71 Ma (Angelier_3, location 3 in Fig. 3 of Angelier et al., 1977). Argon loss could result in these ages (Angelier_3–5 in Fig. 12) being younger than our 40Ar/39Ar groundmass ages of 1.97 ± 0.01 Ma (dacite sample G15M0021 and -22).

The amphibole of sample G15M0004 of the Adamas dacitic lava dome, located ∼ 1 km north of the rhyolitic Bombarda volcano, gave an inverse isochron age of 1.95 Ma ± 0.45 Ma. This age overlaps with the K–Ar age for the Adamas lava dome of 2.03 ± 0.06 Ma (dacite M 66) of Fytikas et al. (1986). The large analytical uncertainty of our sample G15M0004 is caused by a combination of low 40Ar∗ yields and clustering of data points that define the inverse isochron showing the excess argon was identified by the 40Ar/39Ar method (40Ar/36Ar 319.51 ± 14.70), whereas the presence of excess argon cannot be tested by the K–Ar technique, implying that the Fytikas et al. (1986) might be slightly old.

The Korakia andesite has an age of 1.59 ± 0.25 Ma (M 103, Fytikas et al., 1986) and was deposited in a submarine–subaerial environment on top of the Sarakiniko Formation, which was dated based on paleomagnetic polarity in combination with a K–Ar age (1.80–1.85 Ma, Stewart and McPhie, 2003, and references therein). The much older 40Ar/39Ar groundmass age (2.68 ± 0.01 Ma) of Korakia andesite sample G15M0029 is unreliable, and it could indicate the emplacement age of the Kalogeros crypto-dome (2.70 ± 0.04 Ma, Stewart and McPhie, 2006) or represent a geologically meaningless age with only 23 %–27 % of the total 39Ar released in the plateau. In this case, the K–Ar age of 1.59 ± 0.25 Ma is regarded as the likely eruption age for the Korakia andesite, although its argon loss or excess Ar component is unknown.

We obtained 40Ar/39Ar ages of 3.41–4.10 and 3.06 ± 0.02 Ma, respectively, from the groundmasses of dacite samples G15M0017 and G15M0015 in the south-west of Milos (Figs. 2 and 13b). Both of these samples are derived from the coherent dacite facies of the rhyolitic Profitis Illias volcano based on Fig. 11 of Stewart and McPhie (2006). Sample G15M0015 yielded much higher radiogenic 40Ar (41.77 %) than that of sample G15M0017 (< 10 % of 40Ar∗), and the rhyolite sample M 164 from Fytikas et al. (1986) (23.5 % of 40Ar∗) gave an estimate the eruptive age of 3.08 ± 0.08 Ma to the Profitis Illias volcano which is much younger than that given by our sample G15M0017 (Fig. 12). Therefore, we consider our 40Ar/39Ar ages of 3.06 ± 0.02 Ma as the best estimate of the emplacement age of the coherent dacite facies of the Profitis Illias volcano.

A basaltic andesite dyke near Kleftiko on the south-western coast of Milos has a K–Ar age of 3.50 ± 0.14 Ma, which only gave 13.9 % of 40Ar∗ (Fytikas et al. 1986). This age is significantly older than the eruptive ages of the Profitis Illias volcano, which the dyke intruded (Stewart, 2003). Although containing relatively low 40Ar∗ (16.87 %), our 40Ar/39Ar age of 2.66 ± 0.01 Ma with 67.27 % of 40Ar∗ from the groundmass of basaltic andesitic sample G15M0016 of the dyke near Kleftiko is probably an accurate intrusion age.

Unfortunately, we were not able to date all key volcanic units of the Milos VF. This was due to three factors: (1) we did not collect samples from all units; (2) some of the collected samples were not fresh enough after inspection of thin sections; and (3) some of the 40Ar/39Ar data indicate that the K–Ar decay system was disturbed. Therefore, we include published age information to establish a complete high-resolution geochronology for the Milos VF.

The published volcanic units that we include are the Profitis Illias volcano (3.08 ± 0.08 Ma with 23.5 40Ar∗ (%), Fytikas et al., 1986), the Mavro Vouni lava dome (2.50 ± 0.09 Ma with 55.2 40Ar∗ (%), Angelier et al., 1977) in the south-western part of Milos, the Bombarda volcano (1.71 ± 0.05 Ma with 24.3 40Ar∗ (%), Fytikas et al., 1986) and the Plakes volcano (0.97 ± 0.06 Ma with 10.2 40Ar∗ (%), Fytikas et al., 1986, and 0.8–1.2 Ma with 5.4–11.9 40Ar∗ (%), Matsuda et al. 1999). Scoria deposits that Stewart and McPhie (2006) attributed to an andesitic scoria cone between Milos and Kimolos were produced in a submarine setting and maybe occasionally above sea level. No age data for this deposit have been published so far. However, the stratigraphic position of this scoria deposit is between MIL 365 (2.66 Ma, Stewart and McPhie, 2006) and M103 (1.59 Ma, Fytikas et al., 1986), which is shown in Fig. 10 of Stewart and McPhie (2006). Therefore, this scoria cone was likely active in the north-eastern part of the Milos VF between 2.6 and 1.6 Ma.

Fytikas et al. (1986) also analysed a pumice coming from the Sarakiniko deposits east of Adamas (1.85 ± 0.10 Ma with 13.6 40Ar∗ (%), Fytikas et al., 1986) (Fig. 2). This unit is reworked pyroclastic sediment of the Adamas lava dome (Rinaldi and Venuti, 2003). Therefore, the K–Ar age from the Sarakiniko unit is not regarded as an eruption age in this study. We did not sample the neighbouring islands of the Milos VF and also did not attempt to date the products of the recent phase of phreatic activity from which Traineau and Dalabakis (1989) obtained 14C ages of 200 BCE and 200 CE.