Surface Science 605 (2011) 1754–1762

Contents lists available at ScienceDirect

Surface Science

j ourna l homepage: www.e lsev ie r.com/ locate /susc

The oxidation of Fe(111)

Robert Davies, Dyfan Edwards, Joachim Gräfe, Lee Gilbert, Philip Davies,Graham Hutchings, Michael Bowker ⁎Wolfson Nanoscience Laboratory and Cardiff Catalysis Institute, School of Chemistry, Cardiff University, Cardiff CF10 3AT, UK

⁎ Corresponding author.E-mail address: bowkerm@cf.ac.uk (M. Bowker).

0039-6028/$ – see front matter © 2011 Elsevier B.V. Adoi:10.1016/j.susc.2011.06.017

a b s t r a c t

a r t i c l e i n f o

Article history:Received 26 April 2011Accepted 14 June 2011Available online 21 June 2011

Keywords:OxidationFe(111)Methanol adsorptionSTMXPSSurface structure

The oxidation of Fe(111) was studied using Auger electron spectroscopy (AES), low energy electron diffraction(LEED), X-ray photoelectron spectroscopy (XPS), ion scattering spectroscopy (ISS) and scanning tunnellingmicroscopy (STM). Oxidation of the crystalwas found to be a very fast process, even at 200 K, and theAuger O signalsaturation level is reached within ~50×10−6 mbar s. Annealing the oxidised surface at 773 K causes a significantdecline in apparent surface oxygen concentration and produces a clear (6×6) LEEDpattern,whereas after oxidationat ambient temperaturenopatternwasobserved. STMresults indicate that theoxygen signalwas reduceddue to thenucleation of large, but sparsely distributed oxide islands, leaving mainly the smooth (6×6) structure between theislands. The reactivity of the (6×6) layer towards methanol was investigated using temperature programmeddesorption (TPD), which showed mainly decomposition to CO and CO2, due to the production of formateintermediates on the surface. Interestingly, this removes the (6×6) structure by reduction, but it can be reformedfrom the sink of oxygenpresent in the large oxide islands simply by annealing at 773 K for a fewminutes. The (6×6)appears to be a relatively stable, pseudo-oxide phase, that may be useful as a model oxide surface.

ll rights reserved.

© 2011 Elsevier B.V. All rights reserved.

1. Introduction

The interaction of oxygen with iron is of great interest in a widerange of technological areas including corrosion, metallurgy, datarecording media and construction and research in the area has a longhistory. As a result, the majority of studies in the literature concerns theoxidation of iron sheet in ambient conditions and there is relatively littlework using pure oxygen and well-defined surfaces, particularly the(111) plane,which, because of its open structure, is likely to be themostreactive of the low index surfaces. Studies of oxygen adsorption andoxide growth on Fe(110) and (100) surfaces have been reported usingSTM [1–3], LEED/AES [2,4–7], and XPS [8–10] and although it isgenerally agreed that oxide formation follows oxygen chemisorption,there remains somedebate over the kinetics,mechanism and structuresevolving at the earliest stages of oxidation.

A study by Qin et al. described AES, LEED and STM results on theoxidation of Fe(111) at 300 and 500 K. At 300 K there was a rapidinitial uptake of oxygen, with saturation of the Auger signal achievedafter around 200 L exposure, whilst at 500 K oxygen uptake is directlyproportional to exposure and no saturation is observed in the O/Fepeak ratio up to 4200 L [1]. In addition, they report the formation ofdifferent oxide phases at the two different temperatures, with bothFe3O4 and Fe2O3 forming at 300 K but predominantly Fe3O4 at 500 K.STM images showed that after oxidation at 300 K, the surface was

uniformly covered in small (5 to 15 nm wide) oxide islands, whereasoxidation at 500 K resulted in significantly larger islands (100 to300 nm wide) with patches of bare substrate in between.

Formation of oxide islands after room temperature oxidation ofFe(111) was previously reported in an STM and AES study which alsoreported dewetting of the islands from the surface after annealing,resulting in the appearance of larger oxide islands separated by regionsof Fe accompanied by a decrease in the Auger O/Fe peak ratio [2].

An XPS study of the oxidation of polycrystalline iron found that theoxide films produced contained amixture of Fe3O4 and γ-Fe2O3with nodistinct boundaries between them [10]. Other groups havealso reportedmixtures of oxide states at the surface after oxidation of iron [11–14].

The aforementionedwork [10] proposed an oxidationmechanism inwhich oxygen adsorption is followed by thin layer oxide formation viaplace exchangewhich terminates after a fewmonolayers. Subsequently,electron tunnelling from the metal towards the oxygen produces anelectric field which induces outward cation diffusion leading to Fe3O4

growth, whilst inward anion diffusion produces γ-Fe2O3 growth. Thereis some debate within the literature with regards to which mechanisticstep is rate-determining. Grosvenor et al. have measured activationenergies for Fe oxidation by both O2 and water vapour and found thevalues to be the samedespite the fact that oxidationbyO2 resulted in theformation of thicker oxide layers. From this they concluded that placeexchange is the rate determining step [15]. Other groups are of theopinion that the electric field driven ionic diffusion is rate determiningsince the reaction between the metal and oxygen takes place at theoxide/oxygen interface and therefore the metal cations must diffusethrough the oxide layer for the reaction to continue [16–18].

Fig. 1. a.) Auger uptake curves for adsorption at 298 K. b.) Auger uptake curves foradsorption at 400 K and 500 K.

Fig. 2. Auger uptake curve for adsorption at 773 K. The exposure is only shown up to80×10−6 mbar s to emphasise the shape of the curve at low exposures, but the Auger signalhas already saturated—higher exposures up to 500×10−6 mbar s give a ratio of 0.4(±0.05).

1

1.5

2

2.5

3

3.5

4

4.5

5

300 400 500 600 700 800

O/F

e A

ug

er R

atio

Temperature/K

Fig. 3. The effect of annealing temperature, after oxygen adsorption at 298 K for1000×10−6 mbar s, on the Auger signal ratio (measured after cooling).

0

0.05

0.1

0.15

0.2

0.25

0.3

0 0.5 1 1.5 2 2.5 3 3.5

Sti

ckin

g P

rob

abili

ty

Uptake/monolayers

Fig. 4. Sticking probability dependence on oxygen uptake at 290 K.

1755R. Davies et al. / Surface Science 605 (2011) 1754–1762

The aim of the current work was to use a wide range ofcomplementary surface sensitive techniques to explore the rate ofoxidation and structures formed during the oxidation of Fe(111) inmore detail.

2. Experimental

The experiments were carried out on three pieces of equipment: asmall ultra high vacuum (UHV) system for TPD and LEED/Augermeasurements (UHV I), a custom-designed Omicron Multiprobe UHVsystem (UHV II), and a system equipped with XPS and a thermalmolecular beam system (UHV III). UHV I comprises of a stainless steelchamber maintaining a base pressure of ~5×10−10 mbar equippedwitha PSP ISIS ion gun for sample cleaning, anOCI systemfor lowenergyelectron diffraction (LEED) and Auger electron spectroscopy (AES), anda Hiden quadrupole mass spectrometer for TPD measurements. TheFe(111) crystal was mounted on a holder which was attached directlyonto the system manipulator. W wires were spot welded onto theunderside of the sample holder to achieve direct heating. Thetemperature was recorded through a thermocouple inserted into asmall hole in the side of the crystal. For TPD experiments and annealing,methanol and oxygen were dosed via a leak valve through a stainless

steel dosing tube, directed at the sample. The methanol source (Fisher,99.8%) was cleaned by repeated freeze–pump–thaw cycles.

UHV II consists of fourUHVchambers pumped by three turbopumpsand four ion pumps, giving typical base pressures of ~1×10−9 mbar.

0

0.05

0.1

0.15

0.2

0.25

0.3

300 400 500 600 700 800

Init

ial S

tick

ing

Pro

bab

ility

Temperature/K

Fig. 5. The dependence of the initial sticking probability of oxygen on Fe(111) (solidpoints) and of the steady-state sticking probability (open points) upon sampletemperature during adsorption.

1756 R. Davies et al. / Surface Science 605 (2011) 1754–1762

Thepreparation chamber (P) comprises a fast entry lockandanOmicronISE ion gun for sputtering. The analysis (A) chamber is equipped withOmicron SpectaLEED apparatus, a twin anode X-ray source for X-rayphotoelectron spectroscopy (XPS, here we used Al Kα radiation), anOmicron ISE ion gun for ion scattering spectroscopy (ISS) and an energyanalyser used with both XPS and ISS. Adjacent to the preparationchamber, there is the SPM chamber which houses an Omicron variabletemperature surface probe microscope (VT-SPM) which is usedpredominantly for scanning tunnelling microscopy (STM) and whichgenerally has a pressure of b10−10 mbar. The Fe(111) crystal wasmounted on a standardOmicron sample holderwhich could be securelytransferred between the chambers using three horizontal manipulatorsand a linear horizontal probe. Sample heating can be achieved on any ofthe three manipulators using a resistive heating filament. Temperaturewas recorded through a thermocouple attached to the manipulatorclose to sample holder. Oxygen was dosed via a stainless steel dosingtube. The surfacewas cleanedbycycles ofAr+ ionbombardment (1 keV,20 min) followed by annealing in vacuum for ~20 min.

The molecular beam system (UHV III) andmethodology have beendescribed in detail elsewhere [19,20]. It comprises a stainless steelUHV chamber maintaining a base pressure of ~2×10−10 mbar (95%H2) which increased to ~6×10−10 mbar during the course of some ofthe adsorption experiments. It consists of the molecular beam itself,which delivers a circular beam at the sample of 2.9 mm. diameter witha flux of 1.5×1017 molecules m−2 s−1, or about 0.015 monolayers ofmolecules s−1 with respect to the number of surface atoms, togetherwith a PSP Ltd XPS system (dual anode source and 100 mmhemispherical analyser). The Fe(111) crystal was mounted on a

a

Fig. 6. a) The clean (1×1) at 109 eV beam energy. b) LEED pattern due to oxygen adsorption on

custom-built holder in the chamber centre, with the thermocoupledirectly attached to the crystal. Also in the main UHV chamber is aquadrupole mass spectrometer (QMS) (Hiden Analytical Limited)employed for residual gas analysis (RGA). The QMS provides anindication of the angle-integrated partial pressure within the vacuumchamber, since no direct line of sight exists between the sample anddetector. The use of a molecular beam reactor system allows stickingand kinetic information to be obtained, and is described in more detailin previous literature [19–22]. The system relies on principlesoriginally presented by King andWells [21], which compares pressurechanges when the molecular beam directly hits the sample with thatwhen it is blocked by an inert material to determine the stickingprobability. The method removes problems associated with massspectrometer calibration and enables sticking probabilities to bedetermined with an accuracy of ±0.02.

3. Results and discussion

Figs. 1–3 show the results of Auger measurements of the uptake ofoxygen on the Fe(111) surface under a variety of conditions. Here wemakeuse of themainO(KLL) peak at ~510 eV, and themain Fe LVVpeakat ~640 eV. What is very noticeable about these data is that the O:Feratio becomes very highwith the O:Fe ratio saturating at between 4 and5, the saturation level appearing to be nearly independent of dosingpressure or temperature up to ~600 K (Fig. 1). However, as shown inFig. 3, if we anneal a surface dosedwith oxygen at ambient temperatureto higher temperatures, then the signal ratio drops significantly after~620 K, reaching a ratio of ~1, and annealing for longer times reducesthis value further. Similarly, if we carry out adsorption at 773 K (Fig. 2),then the saturation signal is much lower than it is for the loweradsorption temperatures, reaching a saturation value of ~0.4, afterwhich a clear (6×6) LEED pattern is observed. It must be noted that theAuger ratio of ~4 for the low temperature adsorption is the same valuethat we have found for a bulk γ-Fe2O3(0001) crystal [23]. There appearsto be some pressure dependence of the oxidation process in Fig. 1a andb, but it is small over two orders ofmagnitude of pressure change, and ismost noticeable at ambient temperature as a slight negative orderdependence on oxygen pressure. It is actually close to zero order in gasphase oxygen pressure.

Sticking probability data for oxygen adsorption on Fe(111),determined with the molecular beam system described above, areshown in Figs. 4 and 5. The initial sticking probability is 0.25 at ambienttemperature, and decreases with increasing coverage, approaching alower, apparently steady-state value of 0.07(±0.02). Fig. 5 shows thecrystal temperature dependence of the initial sticking probability (thatat the start of adsorption, at very low coverage) and of the steady statesticking value observed at high exposures (the maximum exposure in

b

the Fe(111) surface, followed by annealing to 773 K— the (6×6) at 100 eV beam energy.

A

B

ab

dc

4.9Å

e

Fig. 7. STM images of oxidation of Fe(111). (a) surface exposed to 100×10−6 mbar s O2 at 300 K; (c) and (e) surface exposed to 100×10−6 mbar s O2 at 300 K, then annealed to773 K for 20 min; (b) and (d) line profiles corresponding to (a) and (c) respectively.

1757R. Davies et al. / Surface Science 605 (2011) 1754–1762

these molecular beam studies was ~20×10−6 mbar s). The initialsticking probability is fairly constant up to around 600 K, after which itsharply declines, and becomes the same as the steady state value byaround 750 K. Surprisingly, the latter appears to be independent oftemperature, at least within the experimental error.

Such adsorption measurements are rather more accurate than theAuger and the XPS described below, since they directly measure theoxygen uptake at the surface, and so we can estimate the oxidation rateof the sub-surface layers rather accurately, at least in the limit of a fewlayers of oxidation. Since the steady state sticking probability in thecurves of Fig. 5 is 0.07(±0.02), then the oxidation rate in the beam is theproduct of this and the beam flux given above giving 10−3 mono-layers s−1. If we assume a linear dependence upon oxygen pressure,then the oxidation rate at 1 mbar pressure would be 3×104 mono-layers s−1, or approximately 6 μm s−1. At atmospheric pressure ofoxygen thiswould be ~1 mm s−1. This is an unreasonably high rate, butthere is unlikely to be such a linear dependence at high pressure/deep

oxidation due to diffusion limitations as the layer gets thicker, and thepresence of other factors (humidity andCO2presence in air).However, itindicates that the initial oxidation of iron, assuming the surface could beobtained clean in the first place, is extremely fast under aerobic conditions.

After adsorption at room temperature no LEED pattern is visible, asalso found by Qin et al. [1], even though the clean surface exhibited agood (1×1) pattern (Fig. 6a). However, after the annealing to 773 Kshown in Fig. 3, a very clear (6×6) LEED patternwas seen (Fig. 6b), thispattern also being reported occasionally in the literature [24]. Thisappears to correspond to a surface of lower oxygen concentration thanwas present at ambient temperature, since, upon annealing the O:Feratio drops from ~4 to ~0.7. Sowemay ask ourselveswhat happened tothe oxygen apparently lost by heating? There are three possibilities: i) itis desorbed during heating; ii) it is lost to the bulk of the crystal; iii) itforms crystallites of iron oxide which occupy a small part of the surfacearea and therefore give a small O:Fe Auger signal ratio. Massspectrometry was used to show the absence of a molecular oxygen

0

5

10

15

20

25

30

35

40

300 400 500 600 700 800

MethaneCarbon DioxideMethanolFormaldehydeHydrogenWater

Mas

s sp

ectr

om

eter

sig

nal

/a.u

.

Temperature/K

Fig. 8. TPD after dosing ~12000 mbar s of methanol on to the (6×6) surface. The methane and carbon dioxide signals are offset for clarity.

1758 R. Davies et al. / Surface Science 605 (2011) 1754–1762

desorption eliminating possibility i). To resolve possibilities ii) andiii) we have used STM to image the surface before and after annealing(Fig. 7). At room temperature there is no LEED pattern, and, inagreement with Qin et al.[1], we see no clear structure in STM, just asomewhat roughened surface, as shown in Fig. 7a and in the line profilein Fig. 7b. However, after annealing, when LEED shows the (6×6)pattern, the surface is mostly smooth with a small number of relativelylarge islands present (Fig. 7c), thus supporting possibility iii) above. Theline profiles in Fig. 7d show that these features are between 100 and200 nm in diameter and ~6 nm high, and are very rough particles.Experiments show that the structure and number density of thesefeatures dependon theamount of oxygendosed. This kindof structure isalso similar to that seen by Qin et al. [1] after annealing.

Obviously, due to the large size of the oxide islands formed, iron hasbeen transported from the original surface into these islands. Originally,before annealing, they existed in the rough, grainy film formed byadsorption at ambient temperature. Some kind of phase separation/sintering thenoccurswhenannealing toproduce the large islandsof oxideand the thin film of (6×6) pseudo-oxide. Where exactly the Fe comesfromduring the initial oxidation is uncertain, but since the film covers thewhole surface it is likely to be a fairly homogeneous process. If it waspreferentially at stepswemight expect to see the stickingprobability dropas step sites are lost in the initial stages of adsorption, whereas we see afairly constant sticking probability in the oxidation regime of Fig. 5.

It proved very difficult to get consistent atomic resolution imagingof the inter-particle region. However, in Fig. 7e, taken in the inter-particle region, we can resolve some details of the (6×6) structurewhich confirms that it is due to a coincidence lattice in which 5spacings of the overlayer coincide with 6 of the substrate Fe(111)surface. Since the lattice spacing of Fe(111) is 0.41 nm, then theoverlayer spacing from such a structure should be 0.49 nm, which isvery close to that measured on the STM image. In turn this is close tothe packing expected for haemetite, α-Fe2O3, but quite different fromthe 0.59 nm expected for magnetite or maghemite, γ-Fe2O3.

Further support for the idea that the extra oxygen from the originallayer is present in these large islands comes from the effect of theadsorption of a reductant on the oxidised surface. As shown in Fig. 8, ifwe adsorb methanol on top of the (6×6) structure and carry out TPD,we remove oxygen from the surface as CO2 and H2O, and if weexamine the LEED just after this desorption process, we see adisordered surface with faint (1×1) spots. The main desorptionproduct is hydrogen which desorbs in two channels at 440 K and640 K. CO2 desorbs coincidently with hydrogen in the 440 K peak,whereas a mix of products is observed at 630 K, CO2, H2O, H2, CH4 andH2CO. The approximate carbon selectivities in this peak are 1/3 eachfor each of the three C-containing products. The products are mainlyoxidised and thus oxygen must have been removed from the surface.The mechanism would appear to be the following

CH3OHþ S þ Os→CH3Oa þ OHa

CH3OHþ S þ OHa→CH3Oa þ H2O þ Ov

CH3Oa→H2COþ Ha

2Ha→H2 þ 2S

CH3Oa þ Os þ 2S→CO2þ 3Haþ Ov

CH3Oa þ Ov→CH3a þ Os

CH3a þ Ha→CH4þ 2S

OHa þ Ha→H2O þ 2S

where S represents a surface site, Os is a surface oxygen anion, Ovis a surface oxygen vacancy and subscript a indicates an adsorbedspecies.

1759R. Davies et al. / Surface Science 605 (2011) 1754–1762

Thus we propose that the significant methane production iseffectively due to annealing of surface anion vacancies formed duringadsorption and by CO2 production during the TPD.

At present this mechanism is tentative, but the details of methanoladsorptionon a varietyof ironoxide surfaceswill be reported in thenearfuture. However, the important point in relation to the current paper isthat after annealingat773 K for 15 minafter theTPDagood (6×6)LEEDpattern develops once again, and the Auger signal ratio is 0.7,notwithstanding the fact that oxygen has been removed from thesurface as CO2 andH2O. This confirms that diffusion of oxygen can occurand implies that the islands act as reservoirs for oxygen to reform the(6×6) structure after it was reduced by themethanol. Indeed, the samewas found for several adsorption–desorption cycles with methanol,presumably because there is sufficient oxygen in the islands to reformthe (6×6)monolayer timeafter time. In turn this implies that the (6×6)is a thermodynamically more preferred situation for the oxygen,compared with being in the islands. The exact composition of theislands is unknown, since in suchexperiments they occupy aminority ofthe surface area. Thus, in Fig. 7c for instance, although the islands appearto dominate the image to the human eye, they actually occupy only~15% of the surface area on average, with large areas of flat (6×6)between them. Presumably, if treatment was carried out in theappropriate manner (perhaps prolonged dosing at elevated tempera-ture) we might be able to get considerable coverage of the surface bythis oxidic phase.

Fig. 9. (a) XP spectra of the Fe(2p) region recorded after exposure of the clean Fe(111) surfaspectra are exposure in units of 10−6 mbar s) at 290 K and (b) difference spectra generated

XP spectra of the Fe(2p3/2) region show the Fe peak at 707.5 eV forthe clean surface shifting to 711.0(±0.2)eV with increasing oxygentreatment, Figs. 9 and 10. This is consistent with the formation of Fe3+

[25–27]. The clean metal peak at 707.5 eV was subtracted from thespectra of the oxidised surfaces using the maximum intensity ofthe clean peak that is possible without generating a negative peak inthe residual spectrum. No binding energy shifts weremade during thesubtraction. Such difference spectra also suggest the presence of athird component at low oxygen exposures (b130 L at 290 K, b50 L at500 K) centred at approximately 708 eV, thuswe cannot eliminate thepossibility of a small fraction of the overlayer being in the 2+oxidation state in the early stages of the oxidation. However, itscontribution, if any, in the latter stages is very small. Quantification ofthe XP data, Fig. 11, shows reasonable agreement with the Auger datain Fig. 2; oxidation at 500 K occurs at the same rate as at 298 K in theearly stages but continues to higher oxygen concentrations withhigher oxygen doses. If we anneal an oxidised layer adsorbed at roomtemperature to 773 K, then the Fe(2p) spectrum reverts to a near-metallic one, with a main narrow peak centred at 707.5 eV (Fig. 12),consistent with most of the surface now being metallic, due tocoverage by only a thin layer of oxygen in the (6×6) structure.

The XP data lead us to the conclusion that the oxidation of the(111) surface is most likely to be in the form of Fe2O3, and theevidence we have is that the initial layers and the (6×6), whichdominates the surface area after annealing, are also predominantly α-

ce to increasing amounts of oxygen gas at a pressure of 10−6 mbar (the numbers on theby subtracting the clean spectrum from that after oxygen dosing.

Fig. 10. (a) A series of XPS spectra of the Fe(2p) region after various doses of oxygen onto the clean Fe(111) surface at a pressure of 10−6 mbar at 500 K (the numbers on the spectraare exposure in units of 10−6 mbar s) and (b) difference spectra after removal of the metallic Fe component.

1760 R. Davies et al. / Surface Science 605 (2011) 1754–1762

Fe2O3. Although this contrasts with some thin film work for whichmagnetite dominates for very thin films of iron oxide [27], it isinteresting to compare these results with the very recent findings of

Fig. 11. Uptake curves from the XPS data, showing the change in coverage in the surfaceregion probed by XPS.

Xue et al. [28]. They produced Fe films on Mo(110) and found anordered, thin layer of Fe2O3, after oxidation which transformed intoFe3O4 upon annealing in vacuum to 800 K. This was also associatedwith a change in LEED pattern. In our case we have a (6×6) structurewhich is stable over the range from room temperature (afterannealing) to ~800 K, and is most consistent with a thin layer ofhaematite from STM. Qin et al. consider the thin layer formed afterannealing to 800 K to be magnetite by the qualitative analysis of thelineshape of the broad Auger peaks at ~40–55 eV [1]. It may be thatthe thin film behaves a little differently, in terms of the transitiontemperature for these structures, from the bulk iron crystal.

Thereareanumberof interesting featuresof theoxidation results. Firstof all, it is very surprising thatbulkoxidationof the crystal is so fast. Evenat200 K, the oxidation within the region analysed by Auger appears to becomplete within a dose of 100 L (data not shown). From the XPSmeasurementswe estimate the inelasticmean free path of electrons to be~1.7 nmusing themethods of Powell et al. [29]. Fromthiswecanestimatethe thickness of the oxide layer at an exposure of 260×10−6 mbar s as~2.5 nm for 500 K adsorption and ~0.2 nm (or ~one monolayer) afterannealing the latter to 773 K. It is evident that the oxide layer is a littlethinner at ambient temperature, since Fe0 is still seen in the spectra inFig. 9. Clearly then, the diffusion of Fe cations is extremely fast under bothconditions and must proceed over a low barrier (Ea for Fe oxidation isreportedelsewhereas32+/−6 kJ/mol [15]). Theoxidation ratemeasuredby the molecular beam technique has a probability of ~0.07, with little

Fig. 12. (a) XPS spectra before oxygen dosing, after dosing 70×10−6 mbar s and after annealing to 773 K; (b) difference spectra between clean and dosed surface (lower curve) andbetween annealed and clean surface (upper curve).

Fig. 13. Schematicmodel of the oxidation process and oxide evolutionwith annealing andsputtering. In the top two panels we show nucleation of islands resulting in the formationof a rough oxide film at the surface at ambient temperature. After annealing at hightemperature, then dewetting of the surface occurs to leave a phase-separated systemwithlarge oxide islands and very thin film oxide (the (6×6)) between them. Short sputteringmay give the impression of a fairly clean surface, since it cleans away the majority of thesurface area covered by (6×6), but cannot clean away themuch thicker big islands,whichmay cover a small fraction of the surface area, but still contain a lot of oxygen. If thatsputtered surface is then annealed the (6×6) is recovered bydiffusion of oxygenout of thelarge islands to cover the surface with the thin layer oxide again.

1761R. Davies et al. / Surface Science 605 (2011) 1754–1762

apparent dependence upon temperature,whichmaybe a reflection of thelow barrier to cation diffusion.

However, the initial adsorption (occurring fast compared with thespectroscopy data — in ~10×10−6 mbar s) looks to be classicalsecond order in nature (Fig. 4), following a (1-θ)2 dependence, andapproaches the ‘steady state’ oxidation rate. Thus it appears thatinitial adsorption occurs into a single chemisorbed layer, followed bynucleation and growth of the islands seen in STM. Thus the highsticking value for the clean surface diminishes to that for adsorptionon the oxide islands, which eventually cover the surface in a grainyfilm of disordered oxide. We are engaged in computer modelling ofthese kinetics and will report the results in the future.

From all of these data we can construct a schematic model of theoxidation, as shown in Fig. 13. In the top right we begin with a cleansurface, which nucleates very small oxide islands, which eventuallycover thewhole surfacewith a rough, grainy film (top left). If we thenheat to above ~650 K (middle right), then the (6×6) forms due to de-wetting of the surface and nucleation of the much bigger oxideislands seen in STM. If we sputter the surface at ambient temperaturethen it's easy to remove the (6×6) quickly, butmuchmore difficult toremove the massive oxide islands (bottom left). With a brief annealwe can obtain an apparently ‘clean’ surface with a good (1×1) LEEDpattern, but prolonged annealing forms the (6×6) again (bottomright). This is due to out-diffusion of oxygen from the remainingislands to the clean surface to reform the relatively stable (6×6)structure. Similar effects are observed after reducing the surfacewithmethanol as described above. It is important, under such circum-stances, to sputter hot, in order to continuously deplete oxygen fromthe oxide islands and eventually remove them completely to producea truly clean surface.

4. Conclusions

Oxidation of the surface of Fe(111) proceeds extremely fast, evenat ambient temperatures, equating with initial rates of mm s−1 if they

1762 R. Davies et al. / Surface Science 605 (2011) 1754–1762

could be extrapolated to atmospheric pressure. Under the latterconditions, the surface becomes fully oxidised within the region ofsurface analysis by XPS in ~10−4 mbar s, and forms a granular,disordered surface, which seems to largely consist of Fe2O3. Afterannealing above 600 K, or by adsorbing at high temperatures, theoxide growth is rather different. A good, ordered LEED pattern isobtained because the surface consists of two phases — areas of largeparticles of oxide with dominant, flat, inter-particle areas of the(6×6) structure. It appears that there is communication of oxygenbetween these two areas, since, if the oxygen is removed from the thinlayer areas, it is replaced by oxygen from the oxide islands, and thishas implications for the cleaning of iron surfaces. It may be possible tomake nicely ordered, thin layer iron oxide model catalysts on ironsurfaces in a similar way to which thin alumina films are produced onNiAl crystal surfaces, for instance [30]. It would be interesting toexamine this possibility for significant systems — for example Aunanoparticles on such a model catalyst, since Au/Fe2O3 has beenshown to be an excellent low temperature CO oxidation catalyst [31].It would also be interesting to compare such materials with modelcatalysts prepared on thin film Fe3O4 formed on Pt(111) [27,30].

It would be of considerable interest to expand these findings toconditions more relevant to iron oxidation at ambient conditions, inparticular by examining the effect of water and carbon dioxide,especially the former, on the rates of the initial oxidation of iron singlecrystals, and to examine the structural dependence of these rates byusing other iron single crystal planes.

Acknowledgements

We are grateful to EPSRC for funding DE and to the Sir CharlesWright Endowment fund at Cardiff University for part support of thestudentship for RD.

References

[1] F. Qin, N.P. Magtoto, M. Garza, J.A. Kelber, Thin Solid Films 444 (2003) 179.[2] J.-S. Lin, B. Ekstrom, S.G. Addepalli, H. Cabibil, J.A. Kelber, Langmuir 14 (1998) 4843.[3] A. Wight, N.G. Condon, F.M. Leibsle, G. Worthy, A. Hodgson, Surf. Sci. 331–333

(1995) 133.[4] G. Gafner, R. Feder, Surf. Sci. 57 (1976) 37.[5] A.J. Pignocco, G.E. Pellisser, Surf. Sci. 7 (1967) 261.[6] C. Leygraf, S. Ekelund, Surf. Sci. 40 (1973) 609.[7] G.W. Simmons, D.J. Dwyer, Surf. Sci. 48 (1975) 373.[8] G. Pirug, G. Broden, H.P. Bronzel, Surf. Sci. 94 (1980) 323.[9] V. Stambouli, C. Palacio, H.J. Mathieu, D. Landolt, Appl. Surf. Sci. 70 (71) (1993) 240.

[10] A.P. Grosvenor, B.A. Kobe, N.S. McIntyre, Surf. Sci. 565 (2004) 151.[11] S.J. Roosendaal, B. van Asselen, J.W. Elsenaar, A.M. Vredenburg, F.H.P.M. Habraken,

Surf. Sci. 442 (1999) 329.[12] T.-C. Lin, G. Seshadri, J.A. Kelber, Appl. Surf. Sci. 119 (1997) 83.[13] B. Sinkovic, P.D. Johnson, N.B. Brookes, Phys. Rev. Lett. 65 (1990) 1647.[14] C.R. Brundle, T.J. Chuang, K. Wandelt, Surf. Sci. 68 (1977) 459.[15] A.P. Grosvenor, B.A. Kobe, N.S. McIntyre, Surf. Sci. 574 (2005) 317.[16] S.J. Roosendaal, J.P.R. Bakker, A.M. Vredenburg, F.H.P.M. Habraken, Surf. Sci. 494 (2001)

197.[17] A.T. Fromhold Jr., E.L. Cook, Phys. Rev. Lett. 17 (1966) 1212.[18] N. Cabrera, N.F. Mott, Rep. Prog. Phys. 12 (1948) 163.[19] M. Bowker, P.D.A. Pudney, C.J. Barnes, J. Vac. Sci. Technol., A 8 (2) (1990) 816.[20] M. Bowker, Appl. Catal. 160 (1997) 89.[21] D.A. King, M.G. Wells, Surf. Sci. 29 (1972) 454.[22] D.A. King, M.G. Wells, Proc. R. Soc. London, Ser. A 339 (1974) 245.[23] P.R. Davies, R. Davies, D. Edwards and M. Bowker, unpublished data.[24] U. Seip, M.C. Tsai, J. Kuppers, G. Ertl, Surf. Sci. 147 (1984) 65.[25] D. Briggs, M.P. Seah (Eds.), Practical Surface Analysis, 2nd edition, Wiley,

Chichester, 1990, vol 1.[26] K. Wandelt, Surf. Sci. Rep. 2 (1982) 1.[27] W. Weiss, W. Ranke, Prog. Surf. Sci. 70 (2002) 1.[28] M. Xue, S. Wang, K. Wu, J. Guo, Q. Guo, Langmuir 27 (2011) 11.[29] C.J. Powell, A. Jablonski, S. Tanuma, D.R. Penn, J. Electr. Spectr. Rel. Phenom. 68

(1994) 605.[30] H.-J. Freund, G. Pacchioni, Chem. Soc. Rev. 37 (2008) 2224 and references therein.[31] M. Haruta, T. Kobayashi, H. Sano, N. Yamada, Chem. Lett. 16 (1987) 405.