Adsorption Sites of Individual Metal Atoms on Ultrathin MgO(100) Films
Edgar Fernandes, Fabio Donati, François Patthey, Srdjan Stavri?, Željko ?ljvan?anin, Harald Brune
AAdsorption Sites of Individual Metal Atoms on Ultrathin MgO(100) Films
Edgar Fernandes, Fabio Donati, Fran¸cois Patthey, Srdjan Stavri´c, ˇZeljko ˇSljvanˇcanin,
2, 3 and Harald Brune Institute of Physics, Ecole Polytechnique F´ed´erale de Lausanne (EPFL), Station 3, CH-1015 Lausanne, Switzerland Vinˇca Institute of Nuclear Sciences (020), University of Belgrade P.O.Box 522, 11001 Belgrade, Serbia Texas A&M University at Qatar, Doha, Qatar (Dated: October 15, 2018)We use Ca doping during growth of one and two monolayer thick MgO films on Ag(100) toidentify the adsorption sites of individual adatoms with scanning tunneling microscopy. For thiswe combine atomic resolution images of the bare MgO layer with images of the adsorbates andthe substitutional Ca atoms taken at larger tip-sample distance. For Ho atoms, the adsorptionsites depend on MgO thickness. On the monolayer, they are distributed on the O and bridge sitesaccording to the abundance of those sites, 1 / / I. INTRODUCTION
Single atoms on ultrathin insulating layers grown onmetal surfaces have spectacular magnetic properties ,in particular very long spin-coherence and even longerspin-relaxation times lending single adatom qubitsand memories feasible. Moreover, they exhibit multiplestable charge states , when adsorbed in the vicinityof defects they may catalyse chemical reactions . Addi-tionally, on surfaces of bulk oxides they are astonishinglystable and currently considered as single atom cata-lysts . These remarkable properties emerge from theinteraction of the atom with the surface depending crit-ically on the adsorption site. Knowing this site is there-fore mandatory to understand the thermal stability, cat-alytic properties, charge state, and finally the symmetryof the crystal field that determines the lifetime of mag-netic quantum states .Field ion microscopy reveals the adsorption site, butis limited to strongly bound species on metal surfaces .Scanning tunneling microscopy (STM) and atomic forcemicroscopy (AFM) are more versatile and now widelyemployed to determine the adsorption sites of adatoms.However, all examples in the literature are limited tosingle element surfaces . On the surfaces of ioniccrystals or thin films, such as MgO or NaCl, one of-ten ignores which of the two sublattices gives rise tothe atomic STM and AFM contrast. Density functionaltheory (DFT) calculations report contradicting resultsabout the STM contrast on MgO/Ag(100) , whileAFM contrast of NaCl was interpreted based on molecu-lar markers for which adsorption geometry was obtainedfrom DFT . Specific to STM, the tunnel parame-ters required for atomic resolution on insulating layersimply very small tip-sample distances. Under these con-ditions, adsorbed atoms are frequently displaced or evendesorbed, which further complicates the determination oftheir adsorption site. For example, light adsorbates suchas H can get displaced by tip-sample interactions even un-der moderate tunnel conditions yielding fictitious adsorp- tion sites . As an alternative approach, electron param-agnetic resonance (EPR) was used to indirectly indentifythe adsorption site of Au on thick MgO layers . How-ever, for the same atoms on 3 ML of MgO on Ag(100),the adsorption site remains debated . These issuesare general for any single atom on the surfaces of ioniccrystals or thin films and call for a direct and reliableexperimental method.Here we introduce dilute Ca doping to mark the Mgsublattice in STM images of one and two monolayer thickMgO films grown on Ag(100). To determine the orienta-tion and size of the atomic MgO lattice, we record atomicresolution images on adsorbate-free areas. This lattice isoverlaid onto STM images of the Ca dopants and ad-sorbates taken at larger tip-sample distance. On thisgrid all Ca atoms are on identical sites demonstratingthe reliability of our method to mark the Mg positions.Comparison of the adsorbates’ positions with this MgOgrid unequivocally identifies the adsorption sites of Ho,Au, Co, and Fe adatoms as function of MgO thickness.These four examples are motivated by Ho being the firstsingle atom magnet , the adsorption sites of Au beingdebated , Co having the highest possible magneto-crystalline anisotropy for a 3 d element , and finally Feon MgO being the first system where electron spin reso-nance (ESR) with the STM was demonstrated and spin-coherence times measured on a single atom. DFT calcu-lations reveal the charge transfer and binding energy ofthe adsorbates on mono- and bilayer MgO/Ag(100), aswell as on the (100) surface of bulk MgO.We start by giving details on the experiment and onthe density functional theory (DFT) calculations. Theresults and discussion section is divided into five parts.Section A focuses on the characterization of the pristineand Ca-doped MgO thin films. Sections B and C de-scribe the experimental determination of the adsorptionsites of Ho, as well as STM atomic manipulation experi-ment on these atoms. Sections D and E present resultson the adsorption site of Co, Fe, and Au. Section IV concludes the manuscript. a r X i v : . [ c ond - m a t . m t r l - s c i ] M a y II. TECHNICAL DETAILSA. Experimental
The Ag(100) surface was prepared in ultra high vac-uum by repeated cycles of Ar + sputtering (800 eV,10 µ A / cm ) and subsequent annealing to 770 K. MgOthin films were grown by evaporating Mg from a Knudsencell under a partial pressure of oxygen of 1 × − mbarand with the sample kept at 770 K, as described inRef. 32. These conditions yields an MgO growth rate ofabout 0.1 monolayers per minute. We define one mono-layer (ML) as one MgO(100) unit cell per Ag(100) sub-strate atom. Calcium-doped MgO films were preparedby co-evaporating Ca and Mg under the conditions de-scribed above and with a significant lower Ca than Mgflux, adjusted to obtain the desired dopant concentration.Ho, Co, Fe, and Au atoms were evaporated from e -beamevaporators onto the sample in the STM at T dep ≈
10 Kand p < × − mbar. STM measurements were per-formed with a home-built STM at T STM = 4 . . Differential conductance (d I/ d V ) spectrawere acquired with a Lock-In amplifier using a bias mod-ulation at 1397 Hz and working at closed feedback loopto minimize the tip-surface interaction at large biases. B. Density Functional Theory calculations
The DFT calculations for Ho adatoms onMgO(100)/Ag(100) were carried out using the Wien2kcomputer code , with the same computational setup asthe one described in Ref. 5, i.e. , using the generalizedgradient approximation (GGA) and on-site Coulombinteractions. DFT calculations of the Co and Auadatoms on thin MgO(100) films on Ag(100) wereperformed with the GPAW code , based on the realspace grid implementation of the projector augmentedwave (PAW) method . Exchange-correlation effectswere described employing the Perdew-Burke-Ernzerhoffunctional (PBE) . For Co, the calculations wereperformed within the GGA+U approach , whichcombines the standard PBE exchange-correlation func-tional with on-site Coulomb interaction, using a U valueof 2 eV. The MgO(100)/Ag(100) surface was modeledwith a 3 × k -points for sampling of Brillouinzone . Open boundary conditions are applied perpen-dicular to the surface with 7 ˚A of vacuum separating theoxide/metal slabs from the cell boundaries. To increasenumerical stability of the calculations, the electronicstates were occupied according to the Fermi-Dirac dis-tribution with a broadening of 0.1 eV. Atomic positions were relaxed using the BFGS algorithm . III. RESULTS AND DISCUSSIONA. STM characterization of pristine andCa-doped MgO thin films
Figure 1(a) shows an STM image of the MgO/Ag(100)surface with the bare substrate coexisting with MgO lay-ers of two thicknesses. Since their apparent heights arestrongly bias-dependent, and sometimes inverted withrespect to the expectation from morphology , we usethe energies of field emission resonances to determine theMgO thickness . The d I /d V spectra in Fig. 1(c) exhibittwo resonances with distinct energy separations of 0.69 Vand 0.44 V; the first is characteristic for the MgO mono-,and the second for the bilayer . Note that a very re-cent paper proposes an MgO thickness calibration thatdiffers by one layer from the one used in the current liter-ature and also in the present study. However, our methodto determine the adsorption sites is independent of theMgO thickness.The atomically-resolved STM image of 2 ML MgO ofFig. 1(b) shows a square lattice of protrusions represent-ing one ionic sub-lattice . The period of 2 . ± .
03 ˚Aagrees very well with the Ag(100) nearest neighbor dis-
MgO1 MLMgO2 ML Ag(100) 1 nm10 nm(c) (b)(a) 0 d I/ d V ( A r b . un i t ) Figure 1. (a) STM image of Ag(100) partially covered byMgO ( V t = 1 V, I t = 100 pA). Dark spots are attributedto point defects in the oxide at the interface. (b) Atomicallyresolved image of 2 ML MgO ( V t = −
10 mV, I t = 10 nA).(c) Field emission resonance spectra recorded above 1 and2 ML MgO as well as clean Ag(100) ( V t = 1 V, I t = 100 pA,peak-to-peak modulation amplitude V mod = 10 mV). Ca 2 ML MgO 1 ML MgO pm Figure 2. STM image of one and two monolayers of MgOwith the substitutional Ca atoms appearing as protrusions( V t = −
515 mV, I t = 100 pA). Note that the apparent heightof the MgO layers is inverted to their thickness. tance of 2.89 ˚A. In addition, the STM image showsno superstructure, such as moir´e patterns or disloca-tions. Both observations provide direct evidence forthe MgO(100) film being uniformly and compressivelystrained by 3 % to form a pseudomorphic (1 ×
1) struc-ture on Ag(100). This confirms early diffraction stud-ies that reveal that this lateral compression leads to avertical expansion of the unit cell by 3.6 % . Whetherthe protrusions in this image represent the Mg or theO species has been a matter of debate in theory .Our Ca doping method introduced hereafter determinesit unequivocally for the respective STM tip and tunnelparameters.Figure 2 shows an overview image of 1 and 2 ML Ca-doped MgO. For the employed tunnel parameters, Caatoms are imaged as small protrusions with apparentheights of 73 ± ± i.e. , 88 and 50 objects re-spectively on 1 and 2 ML MgO). In addition, the verysimilar apparent heights observed on 1 and 2 ML MgOsuggests that all Ca atoms are localized in the topmostMgO layer. Ca atoms buried in the second layer wouldappear with different apparent height and be located onthe other lattice site. Therefore Ca protrusions alwaysmark the Mg lattice positions irrespectively of the localMgO thickness. Our conclusion of Ca surface segrega- Ho A Ho A Ho B pm (b)(a) 5 nm 5 nm Figure 3. STM images of (a) 1 ML and (b) 2 ML undopedMgO after the adsorption of 5 × − ML of Ho. Two species,Ho A and Ho B , are discerned by their apparent heights of 220 ± vs. ± T dep ≈
10 K, V t = −
20 mV, I t = 20 pA). tion is supported by the literature . Upon annealingof Ca-rich MgO crystals, Mg atoms at the surface arereplaced with Ca. For our employed Ca-doping of theorder of 0.5% pratically all the Ca atoms are sufficientlyfar from each other to appear as individual and identicalprotrusions, see Figures 2 and 4. B. Adsorption site determination of Ho adatomson thin MgO films
Figures 3(a) and (b) show 5 × − ML of Ho depositedat 10 K on one and two monolayer of MgO/Ag(100).While two species with characteristic apparent heightscoexist on 1 ML MgO, on 2 ML almost exclusively thespecies with smaller apparent height (Ho A ) occurs.Figure 4 shows a single MgO layer with substitutionalCa atoms, as well as both adsorbed Ho species. Calciumatoms appear as faint spots with an apparent height of67 ±
12 pm for this tunneling setpoint. Note that thehigher uncertainty stems from the standard deviation ob-tained from the 5 Ca protrusions. The tunneling condi-tions yielding atomic resolution on MgO move the Hoatoms and therefore, we imaged a bare MgO spot of thesame sample with atomic resolution and extracted theorientation and lattice constant of the MgO lattice fromit. This lattice was then overlaid onto Fig. 4 and one ofits Mg atoms aligned with one of the substitutional Caatoms. All other Ca species fall exactly onto Mg sites il-lustrating the precision of the alignment. This techniquehas been applied on images up to 20 ×
20 nm containingup to 12 Ca atoms (not shown here) with the same reli-ability. Comparing the Ho positions with the overlayedMgO lattice for the shown image and for many additionalones, we infer that Ho A adsorbs on O while Ho B is ona bridge site. Thus, the preferred adsorption site on 2MgO layers is on top of oxygen. In agreement, our DFTcalculations identify this site for 2 ML MgO/Ag(100) asthe most stable one, see Table I. They also show thatthis site is favored for MgO(100) bulk. Therefore, from CaHo A Ho B OMg pm
Figure 4. STM image of 1 ML MgO grown with 5 × − MLCa doping and the same amount of Ho atoms adsorbed ontoit. The orientation and spacing of the overlaid MgO latticewere determined from an atomically resolved image recordedon a bare MgO spot of the same sample. This lattice wasthen translated to bring one of its Mg atoms in coincidencewith one of the Ca species ( V t = −
20 mV, I t = 20 pA). .The very different abundance of both Ho species on 1and 2 ML MgO can be traced back to MgO thickness de-pendent dissipation of the adsorption energy. On 1 MLMgO, the abundance of both species (Ho A : 35 . ± . B : 64 . ± . A : 91 . ± . B :8 . ± . . The fact that this occurs more readily on2 than on 1 ML MgO is related to the dissipation of theadsorption energy via electron-hole pair excitation in thesubstrate that is more efficient for atoms adsorbingon thinner MgO layers. C. STM manipulation of Ho adsorption sites
On 1 ML MgO/Ag(100), the DFT calculationsidentify the Mg site as the most favorable one for Ho atoms. Although Ho atoms do not spontaneously reachthis site after deposition, we can populate it by atomicmanipulation. For the atomic manipulation we applyvoltage ramps with the STM tip placed above the Hoatoms. To prevent major modifications of the probedarea by high electric fields, the voltage is ramped whilekeeping the feedback loop closed, i.e. , the tunnelingcurrent stays constant while the tip retracts smoothly.Abrupt changes in the vertical position of the tipdetected during the ramp evinces a modification ora displacement of the probed atom . Figures 5(a-c)illustrate the result of this process on a few selected Hoatoms on 1 ML MgO. Ramping the bias up to − A atoms [Fig. 5(a)] switches them to Ho B . As aresult, in Fig. 5(b) all the atoms have the same apparentheight and are adsorbed on the bridge site. We note thatthis operation is reversible, i.e. , positive bias rampingup to +1 V on top of a Ho B transforms it back intoHo A . Conversely, further ramping with negative biasesup to − . B atoms irreversibly switchthem to a new Ho C species, see Fig. 5(c). With anapparent height of 141 ± A or Ho B . Similar atomic manipulationsyield to an equivalent sequence of adsorption sites alsoon two MgO layers. Using the unchanged atoms asreference, we identify the possible displacement of theswitched adatoms by subtracting subsequent images.Figure 5(d), obtained by subtracting (a) from (b), showsasymmetric spots at the positions of the three circledatoms. The white arrows point into the direction ofthe displacement and indicate that the switched atoms pm B Ho B Ho B Ho B Ho B Ho A Ho C Ho B Ho A Ho C (e)(d) (c)(b)(a) Figure 5. STM images of Ho atoms on 1 ML MgO. Atomsindicated with circles are successively transformed from Ho A (a) to Ho B (b), and finally to Ho C (c) by applying nega-tive voltage ramps, see text for details (a-c: V t = −
100 mV, I t = 20 pA). (d) Subtraction of images (a) from (b) using theunchanged atoms for precise alignment. (e) (b) − (c). Thewhite (brown) color indicates levels above (below) the zeroplane. Arrows indicate the directions of the atomic displace-ments. have moved along two perpendicular directions, whichcorrespond to the two possibilities of hopping from anO to bridge site. Interestingly, by subtracting Fig. 5(b)from (c), we observe that the switched atoms have beendisplaced perpendicularly to their previous direction, i.e. , from a bridge to an Mg site, see Fig. 5(e).Ho C adatoms are remarkably stable. Voltage rampsup to ±
10 V have no effect on them, while Ho A or Ho B are transformed or desorbed under these conditions. Fig-ure 6(a) shows the adsorption site of Ho C to be on topof Mg. The grid marks the O sites and was determinedwith the same method as the one used for Fig. 4.As shown by our DFT calculations for 1 ML MgO, thisextraordinary stability results from a strong relaxationof the surrounding O neighbors and of the underlyingMg atom making this binding site 4-fold O coordinated,see Fig. 7(a). In agreement with experiment, this sitehas the highest binding energy, followed by the bridgeand O sites, whose atomic geometries are shown in Figs.7(b) and (c), respectively. See Table I for the differencesin binding energy and charge state of the atoms in therespective sites.On two ML MgO, adsorption on top of Mg is calculatedto be less favorable due to the presence of the subsurfaceMgO layer preventing large relaxation of the surface lat-tice, see Figure 7(f). Nevertheless, experiment still findsthis site as the most stable one, although only reachableafter atomic manipulations. Note also that the order ofthe charge states is in agreement with the atomic manip-ulation from O via bridge to Mg sites requiring increas-ingly negative voltages. As also observed for Au and Ag (b)(a) 2 nm 2 nm Ho C Ho A Ho B Ho C Figure 6. (a) STM image of all three Ho species on 1 MLMgO ( V t = −
20 mV, I t = 20 pA). Ho C has an apparentheight of 141 ± C atoms on 2 ML MgO.The lattice marks the O atoms that appear as protrusions( V t = −
20 mV, I t = 5 nA). Note that (b) has been recordedusing the same tunneling parameters all along the image scan.The grid overlays the atomic protrusions, located at O latticepositions. Only half of the grid is shown to better distinguishthe atomic protrusions in the lower part. M L M g O M L M g O (f)(e)(d) (c)(b)(a) Figure 7. DFT adsorption geometries of a Ho adatom on one(a-c) and two (d-f) MgO monolayer in order of decreasingbinding energy from left to right. Color legend - red: O,green: Mg, gray: Ag, blue: Ho on NaCl(100) , this indicates that the atoms becomemore and more positively charged along the transforma-tion sequence.The stability of the Ho C species enables imaging itunder tunnel conditions that yield atomic resolution onMgO. These conditions displace or desorbs Ho atoms onthe other two sites. The lattice overlaid onto Fig. 6(b)marks the O atoms that appear as protrusions. Accord-ing to our experience, this contrast is by far the mostcommon one in low-bias images of the MgO surface. Onlyin very rare cases of tip chemistry and tunnel parametersare the Mg atoms imaged as protrusions. This clarifiesthe controversial DFT results on the STM contrast ofMgO/Ag(100) . Table I. DFT binding energy differences ∆ E (eV) and chargetransfers ∆ q ( e ) for individual Ho, Au and Co atoms on O,bridge (br), and Mg sites on 1 and 2 ML thick MgO/Ag(100)and on the (100) surface of MgO bulk. The site with thehighest binding energy is taken as reference; positive valuessignify a decrease in binding energy.atom site 1 ML MgO/Ag 2 ML MgO/Ag MgO(100)∆ E ∆ q ∆ E ∆ q ∆ E ∆ q O 0.90 0.26 0.00 0.12 0.00 − − − − O 0.04 − − − − − − − − − − − − − − − FeCo Ho A Ho B CoCo (b)(a) Figure 8. STM images of (a) co-adsorbed Co and Ho and (b) Co and Fe on 1 ML MgO(100)/Ag(100). The lattices areinferred from atomically resolved images of the bare MgO surface of the respective samples and mark the O positions. ((a) V t = −
20 mV, I t = 20 pA, (b) V t = −
50 mV, I t = 20 pA). D. Adsorption site of Co and Fe on MgO
Our method of Ca doping can be applied to determinethe adsorption site of any adatom on MgO. Once the siteof one species is determined, one can use that species asmarker to identify the sites of other atoms that are co-adsorbed. We use Ho atoms as marker for O and bridgesites to determine the adsorption site of co-adsorbed Coadatoms. Figure 8(a) shows a 1 ML thick MgO regionwith co-adsorbed Ho and Co atoms. The overlayed gridhas again been extracted from atomically resolved imagesof the substrate and then been brought to coincidencewith the Ho A atoms. It therefore marks the O atoms andone sees that all Co atoms on that image are adsorbedon-top of O, in agreement with Ref. 5. Notice that Co isalways on O, independent of the MgO layer thickness (upto 3 layers) in excellent agreement with DFT. Using Coas a marker for the O sites, we further determine thatFe also adsorbs on top of O, Fig. 8(b), confirming for-mer DFT calculations . Additionally, we notethe presence of an elongated object of apparent height of135 ± . We find thattheir axis is aligned along the O sublattice and each ofthe two constituent atoms is directly above or at leastvery close to an O site. This result indicates that ourmethod can be extended to few-atoms clusters or smallmolecules. E. Adsorption sites of Au on MgO
We now apply our method to Au atoms, for whichcontradicting results were reported in the literature . Figure 9(a) shows 1 ML MgO with co-adsorbed Au andHo, and a grid with lattice spacing and orientation beingagain extracted from an atomically resolved bare MgOspot of the same sample. Ho A and Ho B atoms are usedto align the grid such that it represents the O positions.All the Au atoms shown in Fig. 9(a) are on bridge sites.A statistical analysis over 50 Au atoms on 1 ML MgOindicates that they almost exclusively adsorb on bridgesites, with a small fraction (8 ± and DFTcalculations .The use of adsorbates or substitutional atoms asatomic markers is the key for identifying an adatom’sadsorption site on ionic or more general multi-elementsurfaces. STM images of Au atoms on 3 ML MgO wereinterpreted with an atomic lattice that was not calibratedwith substitutional doping or other means, and a close toequivalent occupation of O and Mg sites was inferred .A small translation of this lattice by half of its latticeparameter would identify the two species as the two dif-ferently oriented bridge sites, thus compatible with ourpresent finding and with former DFT calculations for Auatoms on 1-4 ML MgO/Mo(100) . On mono- and bilay-ers MgO films, the presence of the substrate allows foran effective charge transfer to the adsorbed atom. Con-versely, the charge transfer is reduced in absence of ametal support, as our DFT calculations and Refs. 30 and31 show. In addition to the reduced charge transfer to thesubstrate, DFT predicts a change of the adsorption sitefrom bridge to oxygen going from ultrathin MgO filmson Ag(100) to MgO bulk. This result is in agreementwith former EPR experiment, which reported adsorptionon top of O for 20 ML MgO/Ag(100) . We thereforeinfer that the transition between the bridge and the Osite occurs for MgO thickness above 3 ML . Co AuAu Ho B Ho A pm Figure 9. STM image of co-adsorption of Au and Ho (a) andAu and Co (b) on 1 ML MgO/Ag(100). The overlayed latticesmark oxygen positions in both cases. All Au atoms are foundon the bridge site (a-b: V t = −
100 mV, I t = 20 pA). IV. CONCLUSION
We presented a viable way to experimentally determinethe adsorption site of adatoms on surfaces made of twoor more elements and to interpret atomically resolvedSTM images thereof. For the specific case of MgO, wedetermine the adsorption sites of Ho, Co, Fe, and Au forthe MgO mono- and bilayers grown on Ag(100). Theseresults are of importance for the understanding of thefascinating electronic, catalytic, and magnetic propertiesof individual adatoms on thin films of ionic crystals.
ACKNOWLEDGMENTS
We acknowledge support from the Swiss National Sci-ence Foundation under Projects 140479 and 157081, aswell as the Serbian Ministry of Education and Scienceunder grants ON171033 and ON171017. The DFT cal-culations were performed at the PARADOX-IV super-computer at the Scientific Computing Laboratory (SCL)of the Institute of Physics Belgrade. A. J. Heinrich, J. A. Gupta, C. P. Lutz, and D. M. Eigler,Science , 466 (2004). A. F. Otte, M. Ternes, K. v. Bergmann, S. Loth, H. Brune,C. P. Lutz, C. F. Hirjibehedin, and A. J. Heinrich, Nat.Phys. , 847 (2008). I. G. Rau, S. Baumann, S. Rusponi, F. Donati,S. Stepanow, L. Gragnaniello, J. Dreiser, C. Piamonteze,F. Nolting, S. Gangopadhyay, O. R. Albertini, R. M. Mac-farlane, C. P. Lutz, B. A. Jones, P. Gambardella, A. J.Heinrich, and H. Brune, Science , 988 (2014). S. Baumann, W. Paul, T. Choi, C. P. Lutz, A. Ardavan,and A. J. Heinrich, Science , 417 (2015). F. Donati, S. Rusponi, S. Stepanow, C. W¨ackerlin,A. Singha, L. Persichetti, R. Baltic, K. Diller, F. Patthey,E. Fernandes, J. Dreiser, ˇZ. ˇSljivanˇcanin, K. Kummer,C. Nistor, P. Gambardella, and H. Brune, Science ,318 (2016). W. Paul, K. Yang, S. Baumann, N. Romming, T. Choi,C. P. Lutz, and A. J. Heinrich, Nat. Phys. , DOI:10.1038/NPHYS3965 (2017). F. D. Natterer, K. Yang, W. Paul, P. Willke, T. Choi,T. Greber, A. J. Heinrich, and C. P. Lutz, Nature ,226 (2017). J. Repp, G. Meyer, F. E. Olsson, and M. Persson, Science , 493 (2004). F. E. Olsson, S. Paavilainen, M. Persson, J. Repp, andG. Meyer, Phys. Rev. Lett. , 176803 (2007). W. Steurer, J. Repp, L. Gross, I. Scivetti, M. Persson, andG. Meyer, Phys. Rev. Lett. , 036801 (2015). A. S. W¨orz, K. Judai, S. Abbet, J. M. Antonietti, U. Heiz,A. D. Vitto, L. Giordano, and G. Pacchioni, Chem. Phys.Lett. , 266 (2004). Z. Novotny, G. Argentero, Z. Wang, M. Schmid,U. Diebold, and G. S. Parkinson, Phys. Rev. Lett. , 216103 (2012). X. F. Yang, A. Wang, B. Qiao, J. Li, J. Liu, and T. Zhang,Acc. Chem. Res. , 1740 (2013). G. S. Parkinson, Z. Novotny, G. Argentero, M. Schmid,J. Pavelec, R. Kosak, P. Blaha, and U. Diebold, Nat.Mater. , 724 (2013). R. Bliem, J. van der Hoeven, A. Zavodny, O. Gamba,J. Pavelec, P. E. de Jongh, M. Schmid, U. Diebold, andG. S. Parkinson, Angew. Chem. Int. Ed. , 13999 (2015). C. H¨ubner, B. Baxevanis, A. A. Khajetoorians, andD. Pfannkuche, Phys. Rev. B , 155134 (2014). S. C. Wang and G. Ehrlich, Phys. Rev. Lett. , 2297(1989). H. Brune, J. Wintterlin, G. Ertl, and R. J. Behm, Euro-phys. Lett. , 123 (1990). J. Repp, G. Meyer, K. H. Rieder, and P. Hyldgaard, Phys.Rev. Lett. , 206102 (2003). F. Donati, Q. Dubout, G. Aut`es, F. Patthey, F. Calleja,P. Gambardella, O. V. Yazyev, and H. Brune, Phys. Rev.Lett. , 236801 (2013). T. Eelbo, M. Wa´sniowska, M. Gyamfi, S. Forti, U. Starke,and R. Wiesendanger, Phys. Rev. B , 205443 (2013). N. Lopez and S. Valeri, Phys. Rev. B , 125428 (2004). A. Malashevich, E. I. Altman, and S. Ismail-Beigi, Phys.Rev. B , 165426 (2014). H. Y. T. Chen and G. Pacchioni, Phys. Chem. Chem. Phys. , 21838 (2014). K. L¨ammle, T. Trevethan, A. Schwarz, M. Watkins,A. Shluger, and R. Wiesendanger, Nano Lett. , 2965(2010). G. Teobaldi, K. L¨ammle, T. Trevethan, M. Watkins,A. Schwarz, R. Wiesendanger, and A. L. Shluger, Phys.Rev. Lett. , 216102 (2011). C. Klein, A. Eichler, E. L. D. Hebenstreit, G. Pauer,R. Koller, A. Winkler, M. Schmid, and P. Varga, Phys.Rev. Lett. , 176101 (2003). M. Yulikov, M. Sterrer, M. Heyde, H.-P. Rust, T. Risse,H.-J. Freund, G. Pacchioni, and A. Scagnelli, Phys. Rev.Lett. , 146804 (2006). M. Sterrer, T. Risse, M. Heyde, H.-P. Rust, and H.-J.Freund, Phys. Rev. Lett. , 206103 (2007). K. Honkala and H. H¨akkinen, J. Phys. Chem. C , 4319(2007). G. Pacchioni, L. Giordano, and M. Baistrocchi, Phys. Rev.Lett. , 226104 (2005). J. Pal, M. Smerieri, E. Celasco, L. Savio, L. Vattuone, andM. Rocca, Phys. Rev. Lett. , 126102 (2014). R. Gaisch, J. Gimzewski, B. Reihl, R. Schlittler,M. Tschudy, and W. Schneider, Ultramicroscopy , 1621(1992). P. Blaha, K. Schwarz, G. Madsen, D. Kvasnicka, J. Luitz,WIEN2k: An Augmented Plane Wave plus Local OrbitalsProgram for Calculating Crystal Properties (2001). J. Enkovaara, C. Rostgaard, J. J. Mortensen, J. Chen,M. Du(cid:32)lak, L. Ferrighi, J. Gavnholt, C. Glinsvad,V. Haikola, H. A. Hansen, H. H. Kristoffersen, M. Kuisma,A. H. Larsen, L. Lehtovaara, M. Ljungberg, O. Lopez-Acevedo, P. G. Moses, J. Ojanen, T. Olsen, V. Petzold,N. A. Romero, J. Stausholm-Møller, M. Strange, G. A.Tritsaris, M. Vanin, M. Walter, B. Hammer, H. H¨akki-nen, G. K. H. Madsen, R. M. Nieminen, J. K. Nørskov,M. Puska, T. T. Rantala, J. Schiøtz, K. S. Thygesen, andK. W. Jacobsen, J. Phys. Cond. Mat. , 253202 (2010). P. E. Bl¨ochl, Phys. Rev. B , 17953 (1994). J. J. Mortensen, L. B. Hansen, and K. W. Jacobsen, Phys.Rev. B , 035109 (2005). J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev.Lett. , 3865 (1996). V. I. Anisimov, J. Zaanen, and O. K. Andersen, Phys.Rev. B , 943 (1991). S. L. Dudarev, G. A. Botton, S. Y. Savrasov, C. J.Humphreys, and A. P. Sutton, Phys. Rev. B , 1505(1998). H. J. Monkhorst and J. D. Pack, Phys. Rev. B , 5188(1976). D. C. Liu and J. Nocedal, Math. Program. , 503 (1989). S. Baumann, I. G. Rau, S. Loth, C. P. Lutz, and A. J.Heinrich, ACS Nano , 1739 (2014). S. Schintke, S. Messerli, M. Pivetta, F. Patthey, L. Libi-oulle, M. Stengel, A. De Vita, and W.-D. Schneider, Phys.Rev. Lett. , 276801 (2001). J. Wollschl¨ager, J. Viernow, C. Tegenkamp, D. Erd¨os,K. M. Schr¨oder, and H. Pfn¨ur, Appl. Surf. Sci. , 129(1999). S. Valeri, S. Altieri, A. d. Bona, P. Luches, C. Giovanardi,and T. S. Moia, Surf. Sci. , 311 (2002). S. Stankic, M. Sterrer, P. Hofmann, J. Bernardi, O. Di-wald, and E. Kn¨ozinger, Nano Lett. , 1889 (2005). P. W. Tasker, E. Colbourn, and W. MacKrodt, J. Am.Ceram. Soc. , 74 (1985). R. C. McCune and P. Wynblatt, J. Am. Ceram. Soc. ,111 (1983). D. Z. Gao, M. B. Watkins, and A. L. Shluger, J. Phys.Chem. C , 14471 (2012). O. B¨unermann, H. Jiang, Y. Dorenkamp, A. Kandrat-senka, S. Janke, D. J. Auerbach, and A. M. Wodtke, Sci-ence , 1346 (2015). H. Brune, Science , 1321 (2015). A. Markovits, J. C. Paniagua, N. L´opez, C. Minot, andF. Illas, Phys. Rev. B , 115417 (2003). K. Neyman, C. Inntam, V. Nasluzov, R. Kosarev, andN. R¨osch, Appl. Phys. A , 823 (2004). S. Baumann, F. Donati, S. Stepanow, S. Rusponi, W. Paul,S. Gangopadhyay, I. G. Rau, G. E. Pacchioni, L. Grag-naniello, M. Pivetta, J. Dreiser, C. Piamonteze, C. P. Lutz,R. M. Macfarlane, B. A. Jones, P. Gambardella, A. J. Hein-rich, and H. Brune, Phys. Rev. Lett. , 237202 (2015). O. R. Albertini, A. Y. Liu, and B. A. Jones, Phys. Rev.B , 214423 (2015). S. Baumann,