Triple the fun: tris(ferrocenyl)arene-based gold(i) complexes for redox-switchable catalysis†

The modular syntheses of C3-symmetric tris(ferrocenyl)arene-based tris-phosphanes and their homotrinuclear gold(i) complexes are reported. Choosing the arene core allows fine-tuning of the exact oxidation potentials and thus tailoring of the electrochemical response. The tris[chloridogold(i)] complexes were investigated in the catalytic ring-closing isomerisation of N-(2-propyn-1-yl)benzamide, showing cooperative behaviour vs. a mononuclear chloridogold(i) complex. Adding one, two, or three equivalents of 1,1′-diacetylferrocenium[tetrakis(perfluoro-tert-butoxy)aluminate] as an oxidant during the catalytic reaction (in situ) resulted in a distinct, stepwise influence on the resulting catalytic rates. Isolation of the oxidised species is possible, and using them as (pre-)catalysts (ex situ oxidation) confirmed the activity trend. Proving the intactness of the P–Au–Cl motif during oxidation, the tri-oxidised benzene-based complex has been structurally characterised.


Materials and methods
All reactions and manipulations were carried out under an atmosphere of either nitrogen or argon using standard Schlenk line techniques unless stated otherwise. Thin-layer chromatography (TLC) with silica gel 60 F254 on glass or aluminium sheets available from Merck KGaA was used for monitoring the tris-phosphane synthesis reactions. Column chromatography was performed using a Biotage Isolera 1 automatic purification system with SNAP (silica, particle diameter: 0.040 to 0.065 mm) and SNAP Ultra (silica, spherical particle, diameter: 0.025 mm) cartridges using solvents purged with nitrogen prior to use. The fractions were detected by an integrated UV/Vis detector. Molecular sieves (4 Å) were activated at 300 °C in vacuo for a minimum of 3 h. Dry, oxygen-free solvents (THF, CH2Cl2, Et2O, hexanes, and toluene) were obtained from an MBraun Solvent Purification System MB SPS-800 and directly stored over 4 Å molecular sieves, except for THF, which was further distilled from potassium/benzophenone and stored over 4 Å molecular sieves.
For the use in NMR measurements, CD2Cl2 and CDCl3 were dried by stirring over P2O5 at room temperature for several days, followed by vacuum transfer into a storage flask and degassing by the freeze-pump-thaw method. The solvents were stored over 4 Å molecular sieves.
NMR spectra were recorded with a BRUKER Avance III HD 400 MHz NMR spectrometer at 25 °C (frequencies of 1 H: 400.13 MHz; 11 B 128.38 MHz; 13 19 F and 31 P) of ferrocenyl protons are abbreviated as pt or pq and their observable coupling constants J are given. Assignment of 1 H and 13 C signals to the respective chemical entities are based on 1 H, 1 H COSY, phase-sensitive 1 H, 13 C HSQC and 1 H, 13 C HMBC NMR experiments. TMS was used as the internal standard in the 1 H and 13 14 For the determination of the effective magnetic moments by Evans' NMR experiment, 15 a defined amount of sample was weighed in and dissolved in 0.6 mL of CD2Cl2/CH2Cl2 (50:1 v/v), a sealed capillary with the same solvent mixture but without sample also present in the NMR tube. The diamagnetic contribution of parent complex [1a(Au)3] was determined through the same experiment. Diamagnetic contributions for the anions were approximated as described by Bain and Berry. 16 Electrospray ionisation (ESI) mass spectrometry was performed with an ESI ESQUIRE 3000 PLUS spectrometer with an IonTrap analyser from Bruker Daltonics, or a MicroTOF spectrometer from Bruker Daltonics with a ToF analyser in positive mode. As solvents for the measurements, pure CH2Cl2 or mixtures of CH2Cl2 and CH3CN were used. Dry, oxygen-free solvents were used for air-sensitive species. Elemental analyses were performed with a VARIO EL elemental analyser from Heraeus. Melting points were determined with a Gallenkamp MPD350·BM2.5 melting point device and are reported uncorrected. FTIR spectra were obtained with a PerkinElmer FT-IR spectrometer Spectrum 2000 as KBr pellets and with a Thermo Scientific Nicolet iS5 with an ATR unit in the range from 4000 to 400 cm 1 . UV/Vis spectra were recorded on a PerkinElmer UV/VIS-NIR Lambda 900 spectrometer in quartz cuvettes (d = 10 mm) or double-chamber cuvettes (d = 2 x 4.37 mm). Sample concentration was in the range of 3·10 5 mol·L 1 .

Crystallography
The data were collected on a Gemini-CCD diffractometer (RIGAKU INC.) using Mo-K radiation ( = 0.71073 -scan rotation. Data reduction was performed with CrysAlis Pro 17 including the program SCALE3 ABSPACK 18 for empirical absorption correction. The structure solution and anisotropic full-matrix least-squares refinement on F 2 of all non-hydrogen atoms for 4a, 4b, 4d, 4e, SP1, SP2, SP3, 1a, 1d, [1a(BH3)3], and [1c(Au)3] was performed with SHELXS-97 (direct methods). 19 The structure solution and anisotropic full-matrix least-squares refinement on F 2 of all non-hydrogen atoms for [1a(Au)3](TEF)3 was performed with SHELXT-2018 (dual-space method). 20 Except for disordered solvent molecules, all non-hydrogen atoms were refined with anisotropic thermal parameters and the HFIX command was used to locate all hydrogen atoms for non-disordered regions of the structure. Structure figures were generated with Mercury (versions 3.8 and 3.10). 21

Electrochemistry
Cyclic voltammetry (CV) measurements on 1.0 mmol·L 1 analyte solutions in dry, oxygen-free dichloromethane containing 0.1 mol·L 1 (nBu4N)BF4 or (nBu4N)BAr F 4 as supporting electrolyte have been conducted in a three-electrode setup (GAMRY Instruments, SP-50 potentiostat by BioLogic Science Instruments) under a blanket of nitrogen at room temperature. The glassycarbon working electrode (ALS; surface area 0.07 cm 2 ) and the counter electrode (neoLab; platinum wire, 99.9%) were immersed in the analyte solution, while the reference electrode (ALS; Ag/AgNO3 (0.01 mol·L 1 ) in 0.1 mol·L 1 tetrabutylammonium hexafluorophosphate in dry, oxygen-free CH3CN) was connected to the cell via a bridge tube (filled with the supporting electrolyte) through Vycor tips. The reference electrode was calibrated against decamethylferrocene as an internal standard at the end of the CV experiment, 22 and the results were converted to the FcH/[FcH] + scale in accordance with the IUPAC requirements. 23

Computational methods
All calculations were carried out with the ORCA program package. 24,25 All geometry optimisations were performed at the BP86-D3BJ/def2-TZVP [26][27][28][29][30] level of theory in the gas phase. For gold, the respective effective core potentials, namely def2-ECP, 31 were used. Frequency calculations were carried out to confirm the nature of stationary points found by geometry optimisations.

Synthetic procedures
NMR spectra for all compounds described in the following, including spectra for Evans' NMR experiment, can be found at the end of this document.

1.2.5.
1,3,5-Tris(1-diphenylphosphanyl-1'-ferrocenylene)benzene (1a) -General procedure for the synthesis of 1,3,5-(1-diphenylphosphanyl-1'-ferrocenylene)arenes (1a-d) A stirred solution of 2.36 g (2.72 mmol, 1.00 eq.) 4a in 6 °C (ethyl acetate/N2(l)) and 5.9 mL (1.53 mol·L 1 , 9.0 mmol, 3.3 eq.) nBuLi in n-hexane were added dropwise over the course of 30 min, resulting in a strong intensification of the orange colour. The mixture was kept stirring for 1.5 °C, followed by the dropwise addition of 2.07 g (9.39 mmol, 3.45 eq.) chlorodiphenylphosphane dissolved in 40 mL THF at the same temperature. After slowly warming to room temperature overnight and stirring at 60 °C for 1 hour, TLC (hexanes/ethyl acetate, 10:1) indicated full conversion of 4a. For quenching, a degassed saturated aqueous solution of NH4Cl (40 mL) was added to the fervently stirred mixture via cannula. The phases were separated, and the aqueous phase was extracted with diethyl ether (2x 20 mL). The combined organic phases were dried over degassed MgSO4, filtered, and then the product mixture was adsorbed on Celite® and subjected to column chromatography (hexanes/ethyl acetate, gradient 99:1 to 6:1). Crystals suitable for XRD were obtained from a so-obtained fraction at 7 °C. The fractions were combined and the solvent removed under reduced pressure, 1a was filtered over degassed silica using CH2Cl2, and the solvent was removed in vacuo, yielding 1a (2.01 g, 62% yield) as an orange foam.  15 1 The results of the CHN analysis for 1a are just outside the exceptionally allowed range for analytical purity (±0.5%) and are only provided to illustrate the best value obtained to date.

P P P}-tris[chloridogold(I)] ([1c(Au)3])
Under stirring and protected from light, a solution of 59 mg (50 eq.) 1c in 2 mL CH2Cl2 was added to 50 mg (156 3.15 eq.) [AuCl(tht)] in 1 mL CH2Cl2 and kept stirring at room temperature overnight after which the protection from light was discontinued. The reaction mixture was filtered, its volume was reduced to approx. one third and complex [1c(Au)3] was precipitated from the dark red solution using 5 mL diethyl ether, yielding a red powder which was dried in vacuo (69 mg, 74%). Crystals suitable for XRD were obtained by layering a solution of [1c(Au)3] in CHCl3 with diethyl ether and keeping it at 7 °C for two days.

1,1'-Diacetylferrocinium teflonate (8)
A colourless solution of 1.07 g (1.00 mmol, 1.00 eq.) silver teflonate in 10 mL CH2Cl2 was transferred to an orange solution of 270 mg (1.00 mmol, 1.00 eq.) 1,1'-diacetylferrocene in 10 mL CH2Cl2 leading to an colour change to dark teal and precipitation of a grey metallic solid (elemental silver). The solution was stirred for 18 h, filtered over dry Celite®, and the filter residue was extracted with 10 mL CH2Cl2. The crude product was precipitated by adding 150 mL n-pentane. After decanting off the mother liqueur, the residue was re-dissolved in 10 mL CH2Cl2 and layered with 50 mL n-hexane. After a few days at 4 °C, crystalline 8 was obtained as dark blue needles. The crystals were washed twice with 10 mL n-pentane and dried in vacuo (950 mg, 77%).  13

Monooxidised complex [1a(Au)3](TEF)
To a yellow solution of 30.0 mg (16.0 , 1.00 eq.) [1a(Au)3] in 3 mL CH2Cl2 cooled to 0 °C (ice/water), a similarly cooled, dark-brown solution of 21.0 mg (16.8 eq.) 8 in 2 mL CH2Cl2 was added in small portions via cannula under stirring. The colour of the oxidant solution vanished immediately upon mixing with the solution of [1a(Au)3], which in turn slightly darkened in colour. Stirring at 0 °C was continued for 30 min after which the volatiles were removed in vacuo at 0 °C. The brown residue was taken up in 2 mL diethyl ether, to which 4 mL pentanes were added under vigorous stirring. The orange supernatant, containing 1,1'-diacetylferrocene as confirmed by 1 H NMR spectroscopy, was filtered off the resulting light-green precipitate, and this procedure was repeated twice, until the washing solution was colourless. The light-green solid was dried in vacuo at room temperature (45 mg, quant.). 3 Integrals cannot reliably be determined and are thus excluded for all paramagnetic species.    19 3 2 ). 4 No other resonances than the three phenyl-ppm; they might be too broad to be identified.
For activation of the pre-catalysts by halide abstraction, 67 L of a dried (stirred over 3 Å molecular sieves overnight, then filtered) stock solution of NaBAr F 4 in acetonitrile (5 mg·mL -1 ) were added to an inert-gas NMR tube. The solvent was removed in vacuo and the colourless residue was carefully dried using a heat gun at low power. Next, 0.1 mL of the respective pre-catalyst stock solution were added to the NMR tube, which was then put in an ultra-sonicating bath for 5 min. The catalytic reaction was started (t = 0) upon addition of 0.5 mL of the substrate stock solution.
All reactions were followed by time-resolved 1 H NMR spectroscopy at 25 °C, using 12 scans and d = 10 s. For evaluation of the conversion/yield, spectra were stacked and jointly subjected to auto phase correction and baseline correction (polynomial fit) by MestReNova (version 12.0.0-20080). Spectra, which had not been properly shimmed, were removed from further analyses. Using the "Concentration Graph" function of MestReNova's built-in data analysis suite, signals of interest were integrated against the signal from the internal standard's methoxy protons (9H, 3.71-3.79 ppm). As best suited for determining the yield, the ortho-phenyl protons (HO) of 6 (2H, 7.93-7.99) were chosen (vide infra, section 5). Cross-checking against other 1 H NMR signals of both 5 and 6 gave similar yields, conversions, and turn-over frequencies (TOF) than those reported.  (2) 130 (2) 130 (2) 130 (2) 130 (2) 130 (2) 130(2) Crystal system / Space group

Structural parameters
4e with respect to the other compounds is apparent, in line with slightly longer Carene CCp bonds. While no significant differences among the are found, the steric strain put on 4e co-planarity of the cyclopentadienyl substituents (180° and 0°, respectively). Two short Fe···H distances of 2.86(5) Å (Fe(3)···H9a) and 2.85(5) Å (Fe(1)···H9b) and one slightly longer one of 3.05(4) Å (Fe(2)···H8c) are all well below the sum of the van der Waals radii (3.61 Å), 38 yet very much longer than the experimentally determined bond lengths of a ferrocenyl hydride (1.34-1.62 Å). 48 If steric constraints or energy gain from attractive interactions are at the heart of these contacts cannot be decided here.

Fig. S3
Molecular structures of bromoferrocenylene-containing side products SP1, SP2, and SP3 including part of their atom numbering scheme. As apparent from the depiction, SP3 crystallises with two independent molecules in the asymmetric unit. Thermal ellipsoids are set at the 50% probability level, and co-crystallised solvent for SP2 as well as hydrogen atoms have been omitted for clarity.
Crystallographically characterised side products SP1-3 (Fig. S3) demonstrate the utility of the Negishi protocol for constructing complex molecular architectures from comparatively simple building blocks. Utilising the bromine substituents for further, stepwise functionalisation opens pathways for ferrocene-rich systems which might be of interest for molecular electronics and redoxswitchable catalysis. The structural parameters (Table S4) of SP1-3 do not differ significantly from their intended counterparts 4b,d regarding crucial bond lengths and angles. SP1 is furthermore characterised by two almost perfectly coplanar fulvalenide bridges, likewise found for other, still very rarely prepared tris-/terferrocenes, 49,50 and the synperiplanar arrangement of the two bromoferrocenylene substituents. Pentaferrocenylene derivative SP2 14.90°), illustrating the flexibility of ferrocene for accommodating sterically demanding substituents.

SP1
SP2 1.880 (7) 1.895 (8) 1.870 (7) 1.894 (7) 1.882 (7) 1.879 (7) C ( 1.343 (8) 1.347 (7) 1.342 (7) 1.342 (8)      The solid-state molecular structures of 1a and 1d (Table S5) show virtually identical parameters relevant for their use as ligands in coordination chemistry, indicating that the change in the arene backbone does not influence the diphenylphosphanyl moiety. The all-syn conformation and other structural parameters of 1a are similar to 1c. 13 Converting 1a to [1a(BH3)3] leads to a significant shortening of the C P ips P bond, while the other bond lengths and distances of the tris(ferrocenylene)benzene core remain very similar. The twist between the C6H3 core and the cyclopentadienyl rings of [1a(BH3)3] is notably smaller (8.81° vs. 14.64-27.66°); however, as all compounds display free rotation about the C C ips(fc) C Link bonds, these solid-state parameters are of limited practical value. It is interesting to note that, just as 1a and 1c are very similar in most structural aspects, so are their BH3 analogues, including the crystallographic C3 symmetry represented by space group R3 . 13   a Intramolecular metal-metal distances; the shortest intermolecular Au-Au distance 6.4204(9) Å; b Intramolecular separation between gold and the corresponding iron centre; c Ct C denotes the calculated centre of gravity of the carbon-substituted cyclopentadienyl ring; d Ct P denotes the calculated centre of gravity of the P-substituted cyclopentadienyl ring.
Due to poor quality of the only available single crystals of [1c(Au)3], bond lengths and angles are less accurate than for the trisphosphanes and bromoferrocenylene precursors. The large residual electron densities around the gold atoms are presumably caused by this low crystal quality, as employing other absorption correction methods did not yield better results, and no disorder can unambiguously be assigned to the structure model. Despite numerous attempts, we were not able to grow crystals of better quality. However, the data are still sufficient to unambiguously prove the chemical structure of the given compound.

Table S7
Comparison of solid-state molecular structures of trinuclear gold complexes containing three metal-bound phosphorus atoms (excl. phosphides) listed in the CSD. Entries are sorted alphabetically. Only non-polymeric entries have been considered and analysed in terms of tris-phosphane ligand, approximate C3 (or higher) symmetry (given in brackets if potentially so in solution) and for the presence of aurophilic interactions (given in brackets if not recognised as such in the corresponding CSD entry but matching the distance criterion of d(Au···Au) < 3.5 Å). 51

CSD Identifier Code
Oxidation State of Gold Tris-Phosphane Ligand C3 symmetry Aurophilic Interactions

Au(4)···Fe(4) / Au(5)···Fe(5) /
Due to quality of the only available single crystals of {[1a(Au)3]}2(TEF)6, bond lengths and angles are less accurate than for the tris-phosphanes and bromoferrocenylene precursors. Weak residual electron density is found close to the gold atoms; potentially, a part of the dimeric hexa-cation suffers from disorder, which however could not be properly modelled. Due to relatively weak scattering and strong disorder, only four of the six teflonate anions necessary to balance the hexa-cationic charge of the {[1a(Au)3]2} 6+ dimer could be crystallographically localised and refined accordingly. The presence of the other two teflonate anions was deduced from the electron count, as 1992 electrons per unit cell have been removed using the SQUEEZE routine implemented in PLATON. 19 Four teflonate anions (465 electrons per anion) require 1860 electrons; the missing 132 electrons are most likely contributed from the solvents of crystallisation, 1,2-dichloroethane (50 electrons per molecule) and pentane (40 electrons per molecule). In the same way, a volume of 955 Å 3 per teflonate anion (as judged from the unoccupied volume, assuming dense packing in the asymmetric unit) compares well to the literature value of 17 Å 3 / non-hydrogen atom (one teflonate anion consists of 57 atoms, thus a value of 16.8 Å 3 / atom results). 52 As shown in Fig. S7, void space is clearly visible in the final structure solution, leaving room for four additional teflonate anions per unit cell. In the space-filling model (Fig. S7, right), the separation of the hexa-cationic dimers by co-crystallised solvent and the teflonate anions is illustrated. The closest separation between two of such dimers has been determined to about 4.65 Å (hydrogen-to-hydrogen distance between H120 (phenyl ring at P5) and H98 (ferrocenyl moiety at Fe4 of a neighbouring dimer).   . S9); g Not fully reversible; additional, scan rate-dependent reduction peak at 534 mV (cf. Fig. S9).

Electrochemical data
All tris(1-bromo-1'-ferrocenylene)arenes show one broad, quasi-reversible redox event in the BF4 -based SE (Fig. S8, left). For the mesitylene (4e) and trifluorobenzene (4b) cores, a noticeable deviation from reversibility (ia/ic 1) is linked to a cathodically shifted reduction event which depends on prior oxidation (cf. Fig. S9). This behaviour points towards a fast chemical reaction following the initially Fe-centred oxidation (EC mechanism), 13 generating a species which is harder to reduce (and which can be re-oxidised, too) appearing after the first scan. As this behaviour is linked to 1,3,5-trisubstitued arene cores, this phenomenon is most likely linked to the substituents themselves. Fluorinated (hetero)aromatics are known to generate radical cations and anions on the electrochemical timescale, supporting this hypothesis. [53][54][55] In contrast, 4a-e yield well-defined cyclic voltammograms when the BAr F 4 -based SE is employed (Fig. S8, left), and without indications for follow-up chemistry. With the exception of 4d, the three consecutive oxidation steps are separated by more than  Table S9 0 1/2 in Table S9). Given the disconjugated nature of 4d due to its methylene spacers, the individual ferrocenylene moieties behave more independently and no mixed-valency behaviour is to be expected. I 0 1/2(4d) and E 0 2/3(4d) reflect purely electrostatic effects, 56 while the other compounds might experience a certain stabilisation in their monoand dicationic states.  The redox behaviour of the non-coordinated or unprotected tris-phosphanes 1a-d is dominated by the well-known irreversibility of the first oxidation, which has been attributed to either an involvement of the lone pair of electrons at the phosphorus atom in the HOMO (highest occupied molecular orbital) or by electron transfer from said lone pair of electrons onto the generated iron(III) ion immediately after oxidation. [57][58][59] While the first oxidation on its own is irreversible in the BF4 -based SE (Fig. S10, left), it appears more (quasi-)reversible in the BAr F 4 -based SE (Fig. S10, right). In the case of the benzylic arene core of 1d (blue), even the first two oxidations (dotted lines) are reversible, while for the other arene cores the second oxidation results in a loss of reversibility of the first. In both SE, further oxidation events, potentially P-centred, are recorded at higher potentials. The striazine core of 1c leads to the appearance of cathodically shifted reduction events upon former oxidation in both supporting electrolytes as we have previously noted. 13  In engaging the lone pairs of electrones of the phosphorus atoms in 1a-c in a bond to BH3, the electrochemistry of the tris(ferrocenyl)arene core becomes more recognisable again (Fig. S11). In the BF4 -based SE (Fig. S11, left), much like for the bromine-substituted precursors 4a-c, all ferrocenyl groups are oxidised (and reduced) at the same potential with no sign of the mono-or dicationic intermediates. Again, the trifluorobenzene core of [1b(BH3)3] (green) gives rise to a cathodically shifted reduction tied to the previous oxidation (dashed vs. solid line), speaking for an EC mechanism to occur. In the BAr F 4 -based SE (Fig. S11, left), three well-separated redox events without apparent chemical interference can be recorded. The peak-to-peak separations E o 1/2 and E o 2/3 (Table S11)    Comparing [1a(Au)3] to mono-gold complex 7, their redox potentials E 0 (Table S12) are very similar (E 0 (7) = 320 mV (vs. FcH/[FcH] + , recorded in (nBu4N)PF6/CH2Cl2)), 60 although care has to be taken given the broad waves of [1a(Au)3] and the slight difference in the supporting electrolyte. For the C3N3-based tris-chloridogold(I) complex [1c(Au)3] in the BAr F 4 -based SE (Fig. S12, right), broad, unstructured reduction events appear after the third oxidation (black solid line) has been cycled through. The first (grey dotted line) and second oxidation (dark grey dashed line) processes, when cycled through individually, do not lead to such behaviour as markedly, even though some weak cathodic currents can be detected starting at about 380 mV vs. the FcH/[FcH] + couple. It can thus be speculated that chemical reactions as part of an EC process transfer a part of the oxidised complex into species which are considerably more difficult to reduce. Such behaviour has already been documented for a mono-gold(I) complex of 1c and might thus relate to the s-triazine core itself, 13 as the C6H3-and C6F3-based complexes do not feature these reduction events under these conditions, while in the BF4 -based SE (Fig. S12, left), all three compounds show very similar behaviour.
Aiming to study this behaviour in more detail, the cyclic voltammograms for [1c(Au)3] have been recorded at three different scan speeds (20, 100, and 200 mV·s 1 ). As can be seen from the corresponding Randles-Sev ik plot (Fig. S13), 61 the third oxidation becomes more reversible at higher and less reversible at lower scan speeds, supporting the notion of an EC mechanism being the underlying reason for the non-reversibility of the third oxidation. Further proof can be seen in the intensity reduction of the cathodically shifted reduction events for both 20 and 200 mV·s 1 . In the former case, the chemically generated species will either diffuse away or might be too unstable to be available for reduction, while in the latter case no sufficient time for its formation is available.       DFT calculations have been carried out to provide a better understanding of the electrochemical behaviour of [1a(Au)3]. The highest occupied molecular orbitals (HOMO ) are mostly Fe-centred ( Fig. S14 and S15), while the lowest unoccupied molecular orbital (LUMO) encompasses contributions from iron and the P-bound phenyl rings (Fig. S16). This is in line with the (first) oxidation taking place at the ferrocenylene moieties as found by cyclic voltammetry (cf. section 4). Further supporting this notion, a difference-density plot between native [1a(Au)3] and a monooxidised species [1a(Au)3] + (removing one electron from [1a(Au)3]; no changes in geometry, solvent or anion influences have been considered) shows the iron atoms to be involved most strongly (Fig. S17).
As to the presumed increase in electrophilicity of the gold(I) centres upon oxidation of the ferrocenylene groups, unoccupied orbitals with significant gold contribution ( Fig. S18 and S19) were found to decrease in energy going from [1a(Au)3] to [1a(Au)3] + .

Catalysis
In order to test for catalytic activity of oxidant 8 and the reaction mixture of 8 and reductant 9 themselves, 3.0 mol% of 8 and both 3.0 mol% of 8 and 3.3 mol% of 9 (with respect to substrate 5, representing the maximum amounts used throughout this study) were separately added to an appropriately diluted portion (0.6 mL, diluted with pure CD2Cl2) of the substrate stock solution (0.5 mL) as 60 s under argon. Upon introduction of just 8, a discolouration from green to pink took place; correspondingly, peaks for reduced 8, i.e. 1,1'-diacetylferrocene, are present in the 1 H NMR spectra ( Figure S21, left), amounting to almost 100% of the maximum concentration of 1.8 mmol·L 1 . While it is unclear what caused the reduction of 8, neither 8 nor 1,1'-diacetylferrocene or any other reduction side-or byproducts catalyse the ring-closing isomerisation of 5 to 6 as it becomes  apparent from the constant concentration of 5 (Fig. S20, top). When both 8 and 9 are introduced to the substrate stock solution, no product formation or appreciable conversion of substrate 5 takes place, either. Again, the relative concentration of reduced 8 is close to 100% (Fig. S20, bottom). It is worth pointing out that no signal for the methyl protons of native 9 is detectable (expected spectral region represented by dashed blue box in Fig. S21) and that the position of the amide proton (light blue spectral region C3 in Fig. S21) is slightly different for both experiments. Both pure reductant 9 63 and NaBAr F 4 64 have been previously tested for their catalytic inactivity in this type of catalysis. Illustrating, exemplarily, how the catalytic runs were evaluated using 1 H NMR spectroscopy and the built-in "data analysis" module of MestReNova, Fig. S22 depicts both the stack of individual 1 H NMR spectra (bottom) and the resulting concentration-overtime profiles (top) for the chosen proton resonances. While only the spectral regions of interest are included in this stack, not all of them have been used to follow the reaction. The two multiplets in the aromatic region (7.4-7.6 ppm) correspond to the protons in meta and para position of the phenyl ring of both substrate 5 and product 6. The singlet at 6.07 ppm corresponds to the three aromatic protons of the internal standard of which the nine methoxy protons (3.75 ppm) have been chosen for a fixed concentration reference ([Internal standard] 6.0 mM). The olefinic protons of product 6 (4.38 and 4.81 ppm) have also not been integrated. In the absence of any paramagnetic species (phase IV), the ortho phenyl resonances of 5 (light blue circles) give a significantly lower apparent concentration of about 43 mM than deduced from both the amide (golden circles) and the methylene (navy circles) resonances amounting to about 54 mM, despite a quite long delay time d = 10 s being used in the pulse sequence (a standard 1 H NMR experiment employs d = 2 s). Upon addition of oxidant 8 (phase V), the apparent concentration jumps to about 60 mM, in line with the initial theoretical concentration of [5]0 = 60 mM. Furthermore, the ferrocenylene resonances of 1,1'diacetylferrocene (reduced 8) appear in the 1 H NMR spectra (highlighted in bright red). Even though the effective concentration of paramagnetic species is not changed upon addition of reductant 9 (phase VI), a sudden drop in apparent concentration occurs for some resonances of both 5 and 6. Notably, all signals of product 5 become significantly broadened and, presumably paramagnetically, shifted in phase VI, while all signals of substrate 5 remain unaltered in shape and position. Whether this is due to a selective interaction of the decamethylferrocenium cation ([9] + ) with 6 or an interaction between (diamagnetic) [1a(Au)3] and 6 is unclear. Gratifyingly, the resonances of [1a(Au)3] re-appear after reduction (highlighted in yellow). The signal broadening and shift of 6 is again reversed when the reaction mixture is re-oxidised with another equivalent of 8 (phase VII), again causing a discontinuity in the apparent concentrations back to more "expectable" values, thus suggesting the oxidised catalyst (in this case, [1a(Au)3] + ) to be a paramagnetic relaxation agent. Further oxidation with two additional equivalents of 8 (phase VIII) does not result in another concentration discontinuity but is indeed mirrored in an increase in signal intensity of 1,1'-diacetylferrocene. Taking everything into account, the ortho phenyl protons Ho of 6 (C2, dark green circles) provide the best indicators of the reaction progress when oxidised species are involved, given their separated position, especially from signals of [1a(Au)3] and reduced 8, and comparatively weak discontinuities upon addition of 8 and 9.
In order to investigate the unexpected catalytic inactivity of [1c(Au)3] towards the conversion of 5 to 6, the interaction of complex and substrate was studied using 1 °C (Fig. S23). For this experiment, spectra of both complexes and the substrate in CD2Cl2 were first recorded separately (Fig. S23, top) and superimposed. Following, a Young's NMR tube was charged with a defined amount of [1a(Au)3] or [1c(Au)3] by evaporating a stock solution in vacuo. The NMR tube was cooled to 100 °C (ethanol/N2(l)). A stock solution of 5 in CD2Cl2 was added to the pre-cooled tube, so that a stoichiometric ratio complexto-substrate of 1:3 was achieved. The NMR °C prior to the measurement. Each sample was put into the pre-°C) as quickly as possible and left to equilibrate at this temperature for several minutes before locking, shimming, and measuring. While both [1a(Au)3] and [1c(Au)3] lead to a significant shielding of the amide proton resonance of 5 this very resonance splits into a broad, doublet-like structure only for [1c(Au)3] (Fig. S23, bottom right). The other resonances of 5 are hardly affected by the presence of either [1a(Au)3] or [1c(Au)3]. It is, however, interesting to note that the presence of 5 leads to a splitting ([1a(Au)3]) or narrowing ([1c(Au)3]) of the ferrocenylene resonances associated with the P-substituted C5H4 ring in both cases. This observation might hint at desymmetrisation of the trinuclear complexes upon (reversible and fast) coordination of 5 to the gold atoms; however, no changes in the corresponding 31 P NMR signals have been observed. As a side observation, cooling to 60 °C does not slow down the rotation about the Carene C5H4 bonds to observe decoalescence.
As Fig. S24 shows, both [1a(Au)3] (purple squares) and 7 (pink dots) can be activated for the ring-closing isomerisation of propargyl amide 5 to oxazoline 6 by chloride abstraction, using a stoichiometric amount of NaBAr F 4 prior to addition of substrate 5. In contrast to directly employing native [1a(Au)3] as a catalyst -where this tri-gold complex shows superior performance when compared to mono-gold analogue 7 (cf. Fig. 4, main article) -an anti-cooperative effect can be deduced from the yield-overtime plot and the correspondingly derived linear regressions (Fig. S31 and S32). In fact, 7/NaBAr F 4 operates with a 6.4-fold higher TOF of 20.1±0.6 h 1 vs. 3

.15±0.01 h 1 determined for [1a(Au)3]/NaBAr F 4.
To better understand the inferior performance of [1a(Au)3]/NaBAr F 4 when compared to 7/NaBAr F 4, an NMR experiment was conducted where [1a(Au)3], in a higher concentration suitable for the acquisition of 1 H and 31 P{ 1 H} NMR spectra, was treated with a stoichiometric amount of NaBAr F 4 in the absence of both substrate 5 and the internal standard 1,3,5-trimethoxybenzene, otherwise following the same activation protocol. The results of this experiment are shown in Fig. S25. Right after the abstraction process (dark slate blue, middle), almost no signal attributable to native [1a(Au)3] (green, top) can be discerned, and severe line broadening is apparent from both the 1 H and the 31 P{ 1 H} NMR spectra. The latter indicates the presence of two major species, with the resonance at 42.9 ppm suggesting a cationic, P,P'-dicoordinate gold(I) species. 65,66 The sample was left overnight (prepared under inert conditions, sealed under an argon atmosphere and protected from light), and another set of spectra was recorded (bottom, dark purple), clearly showing a change with time, in line with our previous observation of a change in rate  Table S13. constant over time. Employing the inverse-gated pulse sequence zgig (spectrum not shown), the integrals for the three main phosphorus-containing species were determined to approx. 2:1:2. Apparently, some [1a(Au)3] had re-formed, its 1 H NMR signals significantly broadened and falling together into just three signals. A further upfield shift for the broad 31 P resonance is associated with a loss of intensity; the extremely broad signals previously observed in the 1 H NMR spectrum have almost completely vanished, leaving a much simpler pattern in the spectral region characteristic for ferrocene (4-5 ppm), the four major resonances suggesting a highly (C3-)symmetric species. At this point, we can but speculate about the nature of the catalytically active species in solution. Despite the cationic dicoordinate gold(I) species being an obvious candidate, similar cationic, P,P'-dicoordinate gold(I) complexes that we have tested in the same transformation did -by far -perform worse; furthermore, the formation of a very similar species upon reduction of [1a(Au)3] 3+ (vide infra) casts further doubt on this hypothesis, leaving the species linked to the broadened resonance at about 20 ppm as a likely suspect. Yet, the presence of substrate might yield a different picture altogether.
In addition, the oxidation-induced activation of 7 using one equivalent of oxidant 8 was investigated, and the obtained results are also depicted in Fig. S24 (green stars). In its native state at 1 mol% Au catalyst loading, 7 displays no appreciable catalytic activity, unsurprising given its lack of activity at 3 mol% Au loading. Upon addition of 8, notably without any discernible activation period in contrast to the oxidation-induced activation of [1a(Au)3], a sharp increase of catalytic activity can be witnessed. The TOF of [7] + was determined to 10.6±0.4 h 1 , that is, higher than in situ generated [1a(Au)3] 3+ , isolated [1a(Au)3](TEF)3 or [1a(Au)3]/NaBAr F 4, yet significantly smaller than for 7/NaBAr F 4. It can thus be concluded that the chloride abstraction provides a better means of activation for 7 than the oxidation, yet without the added benefit of comparatively easy reversibility. That is even more so to case for [1a(Au)3] where (threefold) oxidation and chloride abstraction yield similarly active species, with the former allowing for mutual transformation of activity states into one another. and 31 P{ 1 H} NMR (right) spectra acquired before (top, dark green), right after (middle, dark slate blue), and long after (bottom, dark purple) the halide abstraction using 1.0 eq. of NaBAr F 4. Spectra have been acquired in CD2Cl2, the corresponding 1 H NMR signals for CHDCl2 and CH2Cl2 are marked with asterisks (*). Only the relevant spectral region is shown. 31 P{ 1 H} NMR spectra after the halide abstraction have been acquired using the zgpg pulse sequence for fast-relaxing nuclei and can hence not be integrated properly for concentration determination. For the graphic depiction, an exponential line broadening of 5 Hz has been applied to the 31 P{ 1 H} NMR spectra.
In order to test whether the presence of substrate during the oxidation would (negatively) influence the performance of the catalytically active oxidised complex during the conversion of 5 to 6, 67 the oxidation of [1a(Au)3] with one, two, or three equivalents prior to substrate addition was carried out (Fig. S26, TOF cf. bottom of Fig. S26 and Table S13). Oxidant 8 (20, 40 or 60 of a stock solution) and pre-catalyst [1a(Au)3] (0.1 mL of a stock solution) were, under argon, mixed in a Young's NMR tube and left to stand for 10 minutes. Afterwards, the reaction was initiated by addition of the substrate stock solution (0.5 mL) and the reaction was followed by NMR spectroscopy as usual. In all three cases, the TOF obtained by linear fits of the yield-over-time plots are in good agreement with those obtained by oxidising [1a(Au)3] in the presence of substrate 5 with the exception of the mono-oxidised reaction (red trace in Fig. S26 which is slightly faster than its equivalent 1.8 h 1 vs. 1.2 h 1 ). Induction periods are observed, too, and are thus most likely not a direct result of oxidation in the presence of 5, but more likely resulting from the interaction of oxidised complexes and substrate 5 itself. Furthermore and in agreement with the oxidation in the presence of substrate, addition of reductant 9 leads, after some delay, to a halting of the conversion. A second addition of oxidant, as for the other reactions, did lead to renewed catalytic activity, although the observed activities are somewhat lower than for the initial species, particularly in the case of going from the twofold to the mono-oxidised species (blue trace in Fig. S26).  Table S13. Bottom: Turn-over-frequencies (TOF) obtained by linear regression for the three different reaction phases (colour code as above).
In an attempt to elucidate the behaviour of [1a(Au)3] upon stepwise oxidation/reduction using 8/9, a corresponding NMR experiment was conducted ( Fig. S45 and S46 Fig. S45) appear at their literature-described chemical shifts. 68 Upon further addition of up to three equivalents of 8, those signals shift to higher (*) and lower (**) field, indicative of either being strongly affected by the paramagnetic species present in solution or partaking in fast electron-transfer events between different oxidised species. The latter is more likely, given that the oxidation potential of the 1,1'-diacetylferrocenium cation in CH2Cl2 (E 0 = 490 mV vs. . Although the reaction seems to proceed quite selectively, a second species with a 31 P NMR chemical shift between 45 and 40 ppm is formed in a small quantity. Upon stepwise addition of up to three equivalents of reductant 9, the 1 H

Fig. S27
Stacked, individually scaled 1 H NMR spectra before (red, top) and after the successive addition of one (orange), two (green), and three (teal) equivalents of oxidant 8, followed by the successive addition of two (slate blue) and three (purple, bottom) equivalents of reductant 9. Only regions of interest are shown, and residual solvent signals have been excised from the spectra as indicated by the broken lines. All spectra belong to the same sample prepared and kept under an argon atmosphere, and have been acquired in CD2Cl2. The sample was protected from light and maintained at room temperature in between measurements. Asterisks denote the signal attributable to the ferrocenyl (*) and acetyl (**) resonances of reduced 8 (i.e., 1,1'-diacetylferrocene). The dashed arrow highlights the paramagnetic shift experienced by the phenyl protons of [1a(Au)3] upon oxidation and reduction. NMR spectral features characteristic of 1,1'-diacetylferrocene are recovered (Fig. S45), while those of the P-bound phenyl ring protons of [1a(Au)3] remain broadened even after a prolonged waiting time which had been chosen to understand why the addition of 9 in the RSC experiments resulted in a delayed attenuation of catalytic activity. The 1 H NMR signals of the ferrocenylene moiety might re-appear but are hard to discern buried beneath the signals of 1,1'-diacetylferrocene. Similarly, although the addition of 9 also leads to a reversal of the upfield shift for the 31 P NMR resonance of [1a(Au)3] n+ , no full but only partial recovery of [1a(Au)3] can be observed, accompanied by the presence of a probably P,P'ppm, vide supra) 65,66 indicative of chloride loss and a broadened signal at and electron-transfer equilibria.

Fig. S28
Stacked, individually scaled 31 P{ 1 H} NMR spectra before (red, top) and after the successive addition of one (orange), two (mustard), and three (green) equivalents of oxidant 8, followed by the successive addition of two (light blue) and three (navy) equivalents of reductant 9 as well as 14 h after the addition of 9 (purple, bottom). All spectra belong to the same sample prepared and kept under an argon atmosphere, and have been acquired in CD2Cl2 using the zgpg pulse sequence for fast-relaxing nuclei for all spectra except for native [1a(Au)3] (top). They can hence not be integrated properly for concentration determination. For the graphic depiction, an exponential line broadening of 10 Hz was applied. The sample was protected from light and maintained at room temperature in between measurements.    Aiming to study the catalytic stability of [1a(Au)3](TEF)3 (representative for all other oxidised species) and to, in the same way, shed light on the observed induction periods, a second batch of 5 was added to an almost-finished catalytic run (Fig. S37). The catalytic activity, determined over the same relative increase of product concentration, drops from 5.0±0.1 h 1 to 3.2±0.1 h 1 .
No visible catalyst degradation (e.g., precipitation of solids, formation of a gold mirror or colour changes) took place; potentially, at these relatively high concentrations of oxazoline, product inhibition of the catalyst might already operate. Most strikingly and similar to the in situ re-oxidation of [1a(Au)3], no second induction period is observed. Taken together with the independence of the order of addition of oxidant and substrate (vide supra), these findings indicate that the formation of the catalytically active species takes place only after oxidation but involves the substrate. Such amide-assisted activation has already been suggested by Heinze and co-workers, supported through DFT calculations, in their study of valence tautomerism from Fe III /Au I to Fe II /Au II . 70 Time-resolved UV/Vis spectroscopy of mixing 1 eq. of    , not yet mixed, in a two-compartment UV/Vis cuvette (path length 2x 0.476 cm). After this spectrum had been recorded, the cuvette was vigorously shaken (t = 0 min) and successively measured at the indicated times.  Table S13.  Table S13.

Author Contributions
AS has carried out the syntheses and characterisation of the compounds (except for diacetylferrocenium teflonate 8), including the electrochemical, the catalysis, and the NMR experiments, and has written the original draft. PC has acquired and solved the solid-state structures of all compounds but 4a and conducted and evaluated the computational calculations. LD has prepared and characterised diacetylferrocenium teflonate 8. EHH has supervised and administered the project, helped in acquiring funding for AS and PC and supported AS in the writing of the draft.