A Caged Neutral 17-Valence-Electron Iron(I) Radical [Fe(CO) 2 (Cl)(P((CH 2 ) 10 ) 3 P)] • : Synthetic, Structural, Spectroscopic, Redox, and Computational Studies

and (3) computational analysis of the neutral pentacoordinate iron dicarbonyl chloride radical trans-[Fe(CO) 2 (Cl)(P-((CH 2 ) 10 ) 3 P)] • (1a • ). The stability is attributed to a new motif of steric protection afforded by a cage-like, triply transspanning diphosphine ligand. Furthermore, the mechanistic sequences involved are believed to involve topologically unusual steps tantamount to turning a molecule inside out. The many unique physical properties of 1a • are thoroughly interpreted, often with the aid of the DFT calculations.

■ RESULTS Synthesis of the Title Complex. The iron tricarbonyl complexes 2 in Scheme 1 have been under study in the group of one author for some time. 11 They feature trans-spanning diphosphines with three P(CH 2 ) n P tethers, the lengths of which can be varied. It was recently found that irradiation with a Hanovia mercury lamp (450 W) in the presence of excess PMe 3 in hexanes or CH 2 Cl 2 afforded the corresponding free dibridgehead diphosphines 3. 12,13 These could be isolated as white solids in 46-77% yields. All are capable of homeomorphic isomerization, a topological process that turns the molecules inside out, equilibrating in,in and out,out isomers without the need for pyramidal inversion of the phosphorus bridgeheads. This is illustrated in Scheme 2, and a video is available in the Supporting Information of a recent publication. 14 As further elaborated on in the Discussion section, the Fe(CO) 3 moiety is believed to escape by an initial ironphosphorus bond cleavage, followed by homeomorphic isomerization of the diphosphine to give an adduct of out,out-3 and Fe(CO) 3 . Subsequent attack of PMe 3 would displace the η 1diphosphine from iron. Although a small amount of trans-Fe(CO) 3 (PMe 3 ) 2 can be detected during photolyses, the conversion never exceeds 5%, presumably because this complex is itself photoactive under these conditions.

When 2b-e were photolyzed in hexane or CH 2 Cl 2 , the yellow solutions became cloudy orange. 15 However, CH 2 Cl 2 solutions of 2a turned green. 16 Workup afforded a paramagnetic green material in 31% yield based upon the structure established below (3a was also produced). The IR spectrum exhibited two strong ν CO bands (1944 and 1861 cm -1 ) at higher frequencies than those of the precursor (1853 and 1841 cm -1 ), and microanalysis supported the formulation trans-[Fe(CO) 2 (Cl)(P-((CH 2 ) 10 ) 3 P)] • (1a • ). A mass spectrum showed a strong ion corresponding to [M -2(CO)] + . As illustrated in Figure s12, solid 1a • decomposed over the course of 20 h in air. Solutions decomposed over the course of 1.5 h.

Crystallography. Single crystals of a hexane hemisolvate of 1a • could be grown, and the X-ray structure was solved as described in Table 1 and the Experimental Section. The (CH 2 ) 10 segments exhibited two conformations, which could be modeled by a 59:41 occupancy ratio. Thermal ellipsoid plots of the dominant isomer are provided in Figure 2, and bond lengths and angles about iron are summarized in Table 2. The trigonalbipyramidal geometry was evidenced by a P-Fe-P angle of 178. 53(5)°, P-Fe-CO and P-Fe-Cl angles ranging from 89.65 (7)°to 91.64(7)°, and OC-Fe-CO and Cl-Fe-CO angles ranging from 107. 5(1)°to 126.57(8)°.

Complex 2a, which differs from 1a • only by a CO/Cl replacement (identical tether lengths), was viewed as an especially valuable reference compound. Accordingly, the crystal structure of a hexane solvate of 2a was solved. Again, the (CH 2 ) 10 segments exhibited two conformations, which in this case was best modeled by a 54:46 occupancy ratio. The thermal ellipsoid plots are presented side-by-side with those of 1a • in Figure 2. As can be seen in Table 1, the space group, crystal system, and Z value of 2a and 1a • are identical (P2 1 /c, monoclinic, 4), and the unit cell dimensions vary by less than 0.4%, showing the lattices to be essentially isostructural. 17 As would be intuitively expected from the reduced number of valence electrons and presumably attenuated back-bonding in 1a

• , the Fe-C bonds [1.793(5) and 1.802(5) Å] are longer than those of 2a [1.764(2), 1.763(2), and 1.768(2) Å], 2c [1.761(3) and 2 × 1.764(2) Å], and other closely related 18-valenceelectron Fe(CO) 3 adducts. 11a The CO bond lengths [1.074(5) and 1.095(5) Å] are in turn shorter [2a, 1.157(2), 1.160(2), and 1.162(2) Å; 2c, 1.162(3) and 2 × 1.164(3) Å]. It merits note in passing that complexes of the formula Fe(CO) 3 (PPh 2 R) 2 can additionally adopt square pyramidal geometries, 18 and that isomerization can take place upon oxidation. 19 Additional Physical Characterization. UV-visible spectra of 1a • and 2a were recorded in CH 2 Cl 2 . As depicted in Figure 3, 1a • exhibits a moderately intense band at 382 nm (ε 1230 M -1 cm -1 ) superimposed on an absorption tail, and a broader and weaker band at 692 nm (ε 490 M -1 cm -1 ). The latter is of course primarily responsible for the green color. The precursor 2a displays a shoulder of modest intensity at 359 nm (ε 510 M -1 cm -1 ), and did not absorb above 450 nm. The underlying Next, the magnetic susceptibility (χ) of 1a • was determined by Evans' method. The value, μ eff 1.67 μ B , is typical of complexes with a spin (S) of 1 / 2 . Two VT SQUID measurements were made with powdered samples. Both gave identical results, as depicted in Figure 4 (top), indicating a magnetic moment (μ B ) ranging from 1.76 (300 K) to 1.63 (2 K).

As shown in Figure 4 (bottom), CW X-band EPR spectra of 1a • were recorded under several conditions. An isotropic spectrum in liquid CH 2 Cl 2 at 293 K (black trace) exhibited a triplet due to 31 P coupling with a g iso value of 2.05 and a superhyperfine coupling constant (A iso ) of 2.48 mT (23.7 × 10 -4 cm -1 ). A spectrum of the corresponding frozen glass (95 K) also gave a signal with a g iso value of 2.05, but without coupling (red trace).

Next, zero-field 57 Fe Mossbauer spectra of solid samples of 1a • and 2a were recorded at 77 K. As shown in Figure 5 (top), 1a • exhibits a positive isomer shift (IS or δ) of 0.14 mm s -1 , a conspicuously small quadrupole splitting (QS) of 0.29 mm s -1 , and a line width (Γ) of 0.34 mm s -1 . In contrast, 2a (bottom) displays a negative isomer shift of -0.12 mm s -1 , a larger quadrupole splitting of 2.38 mm s -1 , and a line width of 0.27 mm s -1 . These properties are summarized in Tables 3 and 4, together with the experimental data from the literature [20][21][22] and DFT computational results 23 described and interpreted below.

Cyclic voltammograms of 1a • and 2a are depicted in Figure 6. These were recorded in CH 2 Cl 2 under the standard conditions summarized in the caption. Each gave a partially reversible oxidation with i c /i a values of 0.42 and 0.75, respectively. The E 1/2 values (vs Fc 0/+ ) show that the oxidation of 1a • , which would give the 16-valence-electron species 1a + , is only slightly less thermodynamically favorable (with respect to any arbitrary oxidizing agent) than that of 2a to give 2a •+ (0.02 vs -0.01 V; ΔE 1/2 = 0.03 V). A replicate determination on another apparatus but at 200 instead of 100 mV s -1 gave i c /i a values of 0.70 and 0.88 and a ΔE 1/2 of 0.04 V (Figure s10). Surprisingly, neither complex underwent reduction when scans were extended to -2.0 V. Computational insight is provided below.

Computational Characterization. To help interpret the preceding data, DFT calculations were carried out using standard functionals and protocols extensively benchmarked for the electrochemical and Mossbauer calculations, as detailed in the Experimental Section. For computational structures and data, molecules are designated with primes (1a′ • and 2a′). To provide additional perspective, three acyclic model complexes with PMe 3 ligands were also studied: the 17-valence-electron radical trans-[Fe(CO) 2 (Cl)(PMe 3 ) 2 ] • (7′ • ), the corresponding 18-valence-electron anion trans-[Fe(CO) 2 (Cl)(PMe 3 ) 2 ] - (7′ -), and the 18-valence-electron adduct trans-Fe-(CO) 3 (PMe 3 ) 2 (8′). 24 The first two represent experimentally unknown compounds, although 7′ • has been proposed as a transient, 7 as described in the Discussion section.

As illustrated in Figure 7, gas phase energy minimization afforded structures of 1a′ • and 2a′ that were very close to those obtained crystallographically. Bond lengths and angles associated with the central iron atoms and carbonyl ligands are incorporated into Table 2, and the excellent agreement is apparent. The conformations along the P-Fe-P axes (Figure 2, Table 1. Summary of Crystallographic Data

1a • •0.50C 6 H 14 2a•0.45C 6 H 14 empirical formula C 35 H 67 ClFeO 2 P 2 C 35.7 H 66.3 FeO 3 P 2 formula weight 673.12 661.37 temperature [K] 110.0 100.00 diffractometer Bruker Venture XtaLAB Synergy, Dualflex, HyPix wavelength [Å] 0.71073 1.54184 crystal system monoclinic monoclinic space group P2 1 /c P2 1 /c unit cell dimensions a [Å] 11.4017(6) 11.45057(5) b [Å] 11.7236(6) 11.76785(6) c [Å] 27.9438(14) 28.05972(13) α [deg] 90 90 β [deg] 96.4986(19) 96.6447(4) γ [deg] 90 90 V [Å 3 ] 3711.2(3) 3755.61(3) Z 4 4 ρ calc [Mg m -3 ] 1.205 1.170 μ [mm -1 ] 0.593 4.252 F(000) 1464 1422 crystal size [mm 3 ] 0.249 × 0.082 × 0.048 0.44 × 0.28 × 0.12 θ limit [deg] 1.467-25.729 3.171-70.073 index range (h, k, l) -13/13, -14/13, -34/34 -14/14, -13/13, -35/35 reflections collected 70838 69531 independent reflections 7060 7106 R(int) 0.0493 0.0350 completeness to θ [%] 100.0 99.8 max and min transmission 0.4684 and 0.2960 1.000 and 0.225 data/restraints/ parameters 7060/1284/534 7106/159/633 goodness-of-fit on F 2 0.866 1.052 R indices (final) [I > 2σ(I)] R 1 0.0779 0.0377 wR 2 0.1813 0.0938 R indices (all data) R 1 0.0794 0.0389 wR 2 0.1824 0.0947 largest diff peak and hole [e Å -3 ] 0.864 and -0.534 0.579 and -0.287 Figure 2. Thermal ellipsoid plots of the molecular structures of 1a • • 0.5C 6 H 14 (top) and 2a•0.45C 6 H 14 (bottom) (50% probability levels, dominant conformations, hydrogen and solvate atoms omitted).

right) were also nicely reproduced. Figure 8 shows that the corresponding PMe 3 -substituted model compounds 7′ • and 8′ have very similar structures, but now with perfectly staggered P-Fe-P conformations. Interestingly, the anionic 18-electron complex [Fe(CO) 2 (Cl)(P(CH 3 ) 3 ) 2 ] -(7′ -) has a much longer Fe-Cl bond than 7′ • (2.60 vs 2.36 Å), while the other bonds to iron are somewhat shorter (0.06-0.07 Å).

To help understand the Fe-Cl bond length trends, the spin densities of 7′ • and 1a′ • were calculated. As can be derived from the former (Figure 9A), the electron lost from 7′ -comes from the iron 3d xy orbital, which is antibonding with respect to the chlorine 3p x orbital (x-axis perpendicular to the FeP 2 Cl plane). The spin density motif calculated for the full molecule 1a′ • (Figure 9B) is, as expected, very close to that of 7′ • . Parts C and D of Figure 9 show the corresponding plot for the HOMO. As expected, the spin density is dominated by the unpaired electron in this orbital, but the spin density shows spin polarization with the opposite spin perpendicular to the majority spin.

A reviewer noted that the experimental Fe-P bonds in 1a • were longer than those in 2a, although one might have expected a shortening due to the oxidized iron having a shorter bond radius in 1a • , and asked if this lengthening was due to decreased π-bonding. While parallel trends have been observed with other complexes of iron and phosphorus donor ligands where the corresponding radical cations have been crystallographically characterized 25 a more thorough examination of this issue was undertaken as this same trend is reproduced by the calculations for both the actual complexes, 1a′ • and 2a′, and the simplified model complexes 7′ • and 8′. The Quantum Theory of Atoms in Molecules (QTAIM) 26 provides a quantum mechanically accurate method to calculate atomic charges. The QTAIM results for 8′ and 7′ • (Table s1) show that, in spite of 7′ • having a higher formal oxidation state than 8′, the atomic charges differ Table 2. Selected Crystallographic Distances (Å) and Angles (deg) for 1a • and 2a and the Corresponding Computational Data for 1a′ • , 2a′, and PMe 3 -Substituted 7′ • , 7′ -, and 8′

1a • •0.5C 6 H 14 1a′ • 2a•0.45C 6 H 14 2a′ 7′ • 7′ - 8′ P•••P 4.517(1) 4.61 4.4323(7) 4.52 4.52 4.39 4.45 Fe-P(1) 2.2603(10) 2.31 2.2199(5) 2.26 2.27 2.20 2.23 Fe-P(2) 2.2577(11) 2.31 2.2128(5) 2.27 2.27 2.20 2.23 Fe-C(31) 1.793(5) 1.78 1.764(2) 1.77 1.79 1.73 1.77 Fe-C(32) 1.802(5) 1.79 1.763(2) 1.77 1.79 1.73 1.77 Fe-C(33) 1.768(2) 1.77 1.77 Fe-Cl 2.3128(12) 2.36 2.36 2.60 Fe-O(1) 2.866(4) 2.94 2.922(1) 2.93 2.94 2.89 2.93 Fe-O(2) 2.897(4) 2.94 2.925(2) 2.93 2.94 2.89 2.93 Fe-O(3) 2.928(2) 2.93 2.93 C(31)-O(1) 1.074(5) 1.15 1.157(2) 1.16 1.15 1.18 1.16 C(32)-O(2) 1.095(5) 1.16 1.162(2) 1.16 1.15 1.18 1.16 C(33)-O(3) 1.160(2) 1.16 1.16 C(1)-P(1) a 1.814(5) 1.86 1.903(4) 1.86 1.84 1.85 1.84 C(2)-P(1) a 1.843(5) 1.86 1.837(4) 1.86 1.84 1.85 1.84 C(3)-P(1) a 1.850(6) 1.85 1.804(4) 1.86 1.84 1.85 1.84 C(4)-P(2) a 1.835(6) 1.86 1.878(5) 1.86 1.84 1.85 1.84 C(5)-P(2) a 1.839(6) 1.86 1.821(5) 1.86 1.84 1.85 1.84 C(6)-P(2) a 1.821(7) 1.85 1.768(4) 1.86 1.84 1.85 1.84 P(1)-Fe-P(2) 178.53(5) 176.3 178.18(2) 176.9 173.0 170.8 179.9 P(1)-Fe-C(31) 89.7(1) 89.52(6) P(1)-Fe-C(32) 91.0(1) 91.31(6) P(1)-Fe-C(33) 89.98(6) P(1)-Fe-Cl 89.89(4) 88.76 86.48 85.38 P(2)-Fe-Cl 89.83(4) 88.47 86.48 85.38 P(2)-Fe-C(31) 89.3(2) 88.94(6) P(2)-Fe-C(32) 90.3(1) 90.31 by less than 0.10 electrons. This small change occurs because the increased donation from Cl, a strong σ and π donor, offsets the expected electron loss from the increase in formal oxidation state. Thus, in spite of the increase in formal oxidation state, the bond radius of the iron should be similar. One can gain some insight into the evolution of the π-bonding by examining the bond lengths. Replacing a CO in 8′ with Cl -to produce 7′ - causes a substantial increase in the back-bonding of the other ligands and both the Fe-C and Fe-P bond lengths decrease substantially and the C-O and P-C bond lengths increase substantially, a characteristic signature of increased backbonding. When 7′ -is oxidized to produce 7′ • the bond length changes described above are reversed, an indication that the back-bonding in 7′ -is reduced in 7′ • . Relative to 8′, 7′ • appears to have somewhat less backbonding. Table 5 compiles the upper valence molecular orbitals (MOs) for the three model complexes, and Figure 10 illustrates how the energies of the MOs evolve from the 18-electron tricarbonyl 8′ to 7′ -and then 7′ • . It is helpful to include 7′ -in this sequence as 7′ • is an unrestricted calculation with different MOs for α and β spins, and 7′ -shows how the chloride ligand splits the iron 3d orbitals before they are split further by the unrestricted calculation in 7′ • .

As depicted in Figure 10, 8′ displays the typical symmetry and orbital distribution for a 3d 8 trigonal-bipyramidal complex. 27 The two most stable iron 3d MOs, HOMO-2 and HOMO-3, are only π bonding with the carbonyl ligands. The two higher energy iron MOs, HOMO and HOMO-1, are π bonding with some of the carbonyls and σ antibonding with others. When one carbonyl ligand is replaced with a chloride ligand, the degeneracy of both of these pairs is broken. Comparison of HOMOs of 8′ and 7′ -demonstrates that the strong destabilization of two MOs in 7′ -is due to the π antibonding character with the chloride ligand. When a β electron is removed from the HOMO of 7′ -, the MO splits and the energy of the β MO increases, becoming the LUMO of 7′ • β. All of the MOs in 7′ • β are higher in energy due to the loss of an exchange integral, but they follow the same order as 7′ • α.

Attention was next turned to the cyclic voltammetry data in Figures 5 and s10. First, consider the oxidation of 2a by 1a + as expressed in eq 1:

The sum of the E 1/2 values for the half-reactions (+0.03 to +0.04 V) predict that it should be exergonic by -0.69 to -0.92 kcal mol -1 , as calculated from the Nernst equation. Thus, the energies of each of the four species (i.e., 2a′/1a′ + /2a′ •+ /1a′ • ) 3 and 4. were calculated as described in the Experimental Section. As summarized in Table 6, when the geometries were optimized with solvent corrections, the reaction was computed to be exergonic in CH 2 Cl 2 by -0.68 kcal mol -1 in accord with the experimental data. However, the process was predicted to be slightly endergonic in the gas phase and to have mild solvent dependency. In spite of 1a′ + having only 16 valence electrons, it has a singlet ground state because the strong π donation from the Cl destabilizes the HOMO of 1a′ • such that removal of its remaining electron offsets the alternative of removing the electron from d x 2 -y 2 orbital that is strongly stabilized by π-backbonding of the two CO ligands. Next, attention was focused on modeling the UV-visible spectra in Figure 3 with TD-DFT. The spectra computed for 1a′ • and 2a′ are depicted in Figure 11. No intensity is observed for 2a′ in the visible region, consistent with the white to pale yellow color of 2a and its experimental UV-visible spectrum (Figure 3). However, 1a′ • exhibits four moderately strong absorptions in the blue and orange/red regions, consistent with the green color of 1a • and its UV-visible spectrum. As expected, the PMe 3 -substituted model complexes 7′ • and 8′ show very similar spectra, as illustrated in Figure s11.

■ DISCUSSION Mechanism of Formation of 1a • . Most of the pentacoordinate 17-valence-electron iron(I) carbonyl radicals in Figure 1 were accessed by simple oxidative or reductive pathways. In contrast, the title complex 1a • is prepared by the photolysis of an iron carbonyl phosphine complex (2a) in the absence of conventional oxidizing or reducing agents (Scheme 1). Such photolyses can labilize either the carbonyl or phosphine ligands, 28 and the dichloromethane solvent is the only possible source of the chlorine atom in 1a • . Over the last 40 years, the coordination chemistry of dichloromethane has been extensively developed, evolving from conjecture 29 to spectroscopically characterized intermediates 30 to isolable adducts. 31 Thus, the postulation of an initial adduct 4a (Schemes 1 and 3) seems plausible.

This would be followed by inner sphere chlorine atom transfer by one of several mechanistic variants, 32 generating the ironcentered radical 1a • and the carbon-centered radical • CH 2 Cl. Indeed, many dichloromethane complexes undergo carbonchlorine bond cleavage, but with the next observable being an addition product of the type L y M(Cl)(CH 2 Cl). 30b However, this might be suppressed by the cage-like diphosphine ligand. An experimental and computational study involving the transient 18-valence-electron complex (η 5 -C 5 H 5 )Re(CO) 2 (ClCH 2 Cl) implicated subsequent thermal conversion to the 17-valenceelectron species [(η 5 -C 5 H 5 )Re(CO) 2 (Cl)] • . 33 None of the other dibridgehead diphosphine complexes 2b-e (Scheme 1) give an analogous iron(I) chloride complex upon photolysis. As shown in Scheme 3, the mechanism of demetalation of 2a-e to the free dibridgehead diphosphines 3a-e is believed to involve initial iron-phosphorus bond cleavage and homeomorphic isomerization to yield the η 1diphosphine adducts 5a-e. Since 2a has the smallest diphosphine cage, its conversion to 5a would be expected to be slower. Indeed, the barriers to isomerization from out,out-3a,b,c to in,in-3a,b,c decrease from 27.6 to 14.9 to 13.7 kcal mol -1 (ΔG ⧧ 353 K ). 12 Thus, it would be no surprise if CO/CH 2 Cl 2 photosubstitution of 2a to give the dichloromethane complex 4a and then 1a • were to become competitive (Scheme 3).

A less direct pathway to 1a • would involve a CO/CH 2 Cl 2 substitution of 5a, followed by intermediates of the types 6a and 7a (Scheme 3). If this sequence were operative, it would seem likely that the 5b-e generated from 2b-e would undergo analogous substitutions, and ultimately form some 1b-e • as byproducts. However, there has been no evidence for the generation of significant quantities of these species.

Redox Properties of 1a • . As illustrated by the cyclic voltammograms in Figures 6 and s10, 1a • can undergo a partially reversible oxidation to the 16-valence-electron cation 1a + in CH 2 Cl 2 . There is seemingly the possibility for dichloromethane coordination to 1a + as well. The octahedral coordination geometry that would result has abundant precedent. 11a,34,35 Interestingly, no evidence was seen for reduction of 1a • , including cathodic scans out to -2.0 V, close to the solventimposed limit. Reduction would presumably yield 1a -, an 18valence-electron species analogous to computationally characterized 7′ -.

The E 1/2 value for the couple 1a • /1a + (0.020 V) is also a puzzle. This is only slightly greater than that for the tricarbonyl complexes 2a/2a + (-0.010 V), each of which have one additional valence electron. Nonetheless, the oxidation of 1a • is only slightly less thermodynamically favorable than 2a, consistent with the DFT data in Table 6. As shown in Figure 10, when the strong π acceptor, CO, is replaced by the strong π donor, Cl -, 36 two of the iron d orbitals are markedly destabilized. The HOMO is so strongly destabilized that the 17-electron neutral species becomes more stable than the 18electron anion. Loss of an electron from the HOMO of the 18electron anion stabilizes the remaining electron in this MO; now its orbital energy is nearly identical to that of the HOMO of the 18-electron tricarbonyl system. Accordingly, the E 1/2 value for the 1a • /1a + couple is only slightly different from that of the 2a/ 2a + couple. However, the E 1/2 values for couples involving higher homologues of the tricarbonyl complexes, 2b/2b + and 2c/2c + , indicate much more thermodynamically favorable oxidations (-0.136 and -0.146 V). 11a,37 Thus, there is a significant cage size effect.

Other Pentacoordinate Iron(I) Radicals. Of the isolable iron carbonyl radicals in Figure 1, the four reported by Berke, 6 trans-[Fe(CO) 2 (Br)(P(OiPr) 3 ) 2 ] • , trans-[Fe(CO) 2 (I)(P-(OiPr) 3 ) 2 ] • , trans-[Fe(CO) 2 (Br)(PEt 3 ) 2 ] • , and trans-[Fe-(CO) 2 (I)(PEt 3 ) 2 ] • , are the most relevant. All are especially air-sensitive as solids, whereas 1a • survives more than 6 h (Figure s12). The PEt 3 adducts give IR ν CO bands (X = Br, 1955/1884 cm -1 ; I, 1956/1886 cm -1 ) comparable to those of 1a • (1944/1861 cm -1 ). Using similar synthetic approaches as Berke, others later attempted the isolation of our computational model complex trans-[Fe(CO) 2 (Cl)(PMe 3 ) 2 ] • (7′ • ). 7 The green material, characterized by IR (1984/1935 cm -1 ), was reported to be thermally sensitive and rapidly react with air, and could never be obtained in pure form. These observations vividly illustrate the stabilization imparted by the dibridgehead diphosphine cage of 1a • . Coincidentally, Berke has also studied 7′ • by DFT, but communicated only a few structural parameters. 38 Berke also reported the EPR spectra of his four radicals, as well as P(OMe) 3 analogs that could not be isolated in pure form. The two PEt 3 adducts give g and A 0 values (2.057/2.55 and 2.078/ 2.40) very close to those of 1a • (2.05/2.48). The radical trans-[Fe(CO) 2 (I)(PEt 3 ) 2 ] • could furthermore be crystallographically characterized, and extended Huckel calculations were reported for the PH 3 analog. 6 In several cases, Berke could reduce his radicals with Na/Hg under nitrogen and obtain diamagnetic adducts with Fe-N 2 -Fe linkages. Under similar conditions using sodium, no reaction was observed with 1a • , in parallel with the cyclic voltammetry data.

The Mossbauer data for 1a • and 2a in Figure 5 can be compared to those of the 18-valence-electron complexes trans-Fe(CO) 3 (L) 2 (L = CO, PPh 3 , PCy 3 ), as summarized in Tables 3 and 4. However, among the iron-based radicals in Figure 1, only [Fe(CO) 5 ] •+ Al(OC(CF 3 ) 3 ) 4 -has been characterized by Mossbauer spectroscopy. In all cases, the DFT calculations give quadrupole splittings (QS, Table 3) and isomer shifts (IS, Table 4) that are in excellent agreement. Also, the computational data for the PMe 3 model compounds are very close to those of their experimental counterparts.

The most interesting Mossbauer comparisons involve the data for 1a • and 2a versus those of [Fe(CO) 5 ] •+ Al(OC(CF 3 ) 3 ) 4 -Table 5. MOs of Model trans PMe 3 Complexes 8′, 7′ -, and 7′ • with Energies in Atomic Units Inorganic Chemistry and Fe(CO) 5 . In both pairs of complexes, the paramagnetic species exhibits a dramatically lower quadrupole splitting (0.29 vs 2.38 mm s -1 and 0.53 vs 2.55 mm s -1 ), with that of 1a • being conspicuously small. The larger splitting of the diamagnetic iron(0) complexes arises from the d 8 configurations, which, as compared to a spherical d 10 atom, have less electron density along the P-Fe-P axis. When another electron is lost from the d orbital in the (CO) 2 FeCl plane to give iron(I), the electron density becomes more spherical, reducing the quadrupole splitting.

Also, the isomer shifts of the paramagnetic iron(I) complexes are greater than those of their diamagnetic iron(0) counterparts (0.14 vs -0.12 mm s -1 and 0.17 vs -0.08 mm s -1 , respectively). This is opposite to some literature generalizations regarding oxidation state trends [e.g., Fe(0) > Fe(I)]. 22b,23 However, fuller treatments highlight the independent roles of ligand fields and covalency, coordination numbers, and bond lengths. 22b,23a Furthermore, Peters has reported an iron carbonyl triphosphine system that can be isolated in three oxidation states, and his isomer shift trend parallels ours [Fe(II) > Fe(I) > Fe(0)]. 9c The doublet character and (in the case of 1a • ) the strong donation from the chloride ligand would be expected to enhance isomer shifts. So to sum, pairs of closely related pentacoordinate paramagnetic iron(I) and diamagnetic iron(0) complexes appear to give diagnostically different isomer shifts and quadrupole splittings.

Overview and Conclusion. The steric stabilization afforded by the dibridgehead diphosphine cage of the radical 1a • can also be visualized with space-filling representations, as shown in Figure 12. The equatorial CO and Cl ligands are visible only in a "peekaboo" mode, and the tightly fitting methylene chains strongly shield the metal. Despite these favorable factors, at least two potential Achille's heels remain. One is direct air oxidation (Figure s12), presumably by some outer sphere Table s3 for details.

process. Another would be homeomorphic isomerization, a pathway we consider operative for isosteric 2a as outlined in Scheme 3. This would expose the iron atom, facilitating a variety of possible degradation reactions. For this reason, related complexes with less conformationally flexible phosphorusphosphorus linkages have been a long-standing synthetic goal of one author. 39 In conclusion, this work has established a new strategy for the stabilization of organometallic radicals based upon the steric shielding provided by cage-like trans-spanning dibridgehead diphosphine ligands. Importantly, these ligands can also accommodate square planar 35,40 and octahedral 11a,34,35 coordination geometries, so this approach could have considerable generality. However, such efforts will be facilitated by the development of other synthetic routes, as that used for 1a • (Schemes 1 and 3) is only feasible for one cage size. Nonetheless, the different reaction modes of 2a-e as a function of cage size provide valuable insight regarding homeomorphic isomerization in coordination chemistry. It is also easy to envision substituted dibridgehead diphosphines that provide even more steric shielding for the L y M core, and the analogous dibridgehead diarsines and distibines are also available. 13 ■ SAFETY STATEMENT Caution! The 450 W photochemical lamp emits considerable heat during use, requiring an external cooling well. 12,13 Interruption of the water f low in the quartz cooling well can lead to glass failure, solvent ignition, and other potential hazards. Attached hosing should be checked thoroughly before operation of the lamp. Ultraviolet light produced by the lamp is damaging to biological tissues. When in use, proper PPE should be worn (e.g., UV protective goggles/glasses, lab coat, gloves) to ensure that exposure is minimized. Care should also be taken when using liquid nitrogen for Schlenk line traps to avoid condensation of liquid oxygen f rom air.

■ EXPERIMENTAL SECTION General Procedures. Reactions and workups were conducted under inert atmospheres. Chemicals were treated as follows: hexanes, CH 2 Cl 2 , and toluene, dried and degassed using a Glass Contour solvent purification system; PMe 3 (Strem, 98%) used as received; silica gel (40-63 μm mesh, Silicycle) flame-dried and left under vacuum for 1 day before use.

trans-Fe(CO) 2 (Cl)(P((CH 2 ) 10 ) 3 P) (1a • ). A flame-dried Schlenk flask was charged with trans-Fe(CO) 3 (P((CH 2 ) 10 ) 3 P) (2a; 0.278 g, 0.446 mmol), PMe 3 (0.69 mL, 6.69 mmol), and CH 2 Cl 2 (10 mL), and placed in front of a water-cooled quartz immersion well of a Hanovia 450 W lamp. As illustrated with photographs elsewhere, 12,13 the sample was irradiated overnight with stirring. The solvent was removed by an oil pump vacuum. The residue was dissolved in hexanes and applied to a small pipet column of silica gel. The column was rinsed with hexanes (eluting 3a) and then CH 2 Cl 2 . The dark green fractions were collected, and the solvent was removed by an oil pump vacuum. The residue was washed several times with hexanes and dried under vacuum to give 1a

• (0.088 g, 0.140 mmol, 31%) as a green solid, dec pt. 166 °C (open Scheme 3. Mechanism Proposed for Conversions of the Iron Tricarbonyl Complexes 2a-e to the Free Diphosphines 3a-e and the Title Complex 1a • Figure 12. Space-filling representations of 1a • . Inorganic Chemistry capillary). Anal. Calcd for C 32 H 60 ClFeO 2 P 2 (630.07): C, 61.00; H, 9.60. Found: C, 61.24; H, 9.76. IR (powder film, cm -1 ): 2922 (m), 2852 (m), 1944 (s, ν C�O ), 1861 (s, ν C�O ), 1456 (w), 1259 (w), 1074 (w), 1017 (w), 948 (w), 804 (w), 713 (m). HRMS (ESI, m/z): calcd for C 30 H 60 P 2 FeCl [M -2CO] + : 573.3203, found: 573.3198; calcd for C 32 H 61 P 2 FeO 2 [M -Cl + H] + : 595.3491, found: 595.3480. Computations. Input models were built with Avogadro software. Except as otherwise noted the density functional theory (DFT) calculations were performed with the Gaussian 16 program (revision C.01). 41 Geometry optimizations and frequency calculations in the gas phase used the B3LYP 42 functional and the 6-311G(d) 43 basis set for all atoms. Tight convergence criteria were used for the optimizations. Wave function stability calculations were performed to confirm the absence of lower-energy numerical solutions for all computed structures. Ultrafine grids (99,590 points per atom) were used as implemented in the Gaussian software. Calculations involving redox phenomena were performed with TPSS 44 functional with the 6-311G(d) basis set for all atoms. The TPSS functional has performed well for predicting structures and electrochemical properties of other iron and nickel complexes. 45 Initially, dichloromethane and acetonitrile solvent corrections with SMD 46 were applied to the gas-phaseoptimized geometries. The solvent dependency suggested that optimizing the structure in CH 2 Cl 2 would bring its predicted value closer to the experimental one. Mossbauer calculations were done in the gas phase with B3LYP/def2-TZVP 47 method/basis set combination in the Orca 5.0.3 software package, 48