A chemical reaction controlled by light-activated molecular switches based on hetero-cyclopentanediyls

Biradicals were applied as molecular switches to control chemical reactions that involve the activation of small molecules. The mechanism was studied by experimental and computational methods.


Experimental
General Information. If not stated otherwise, all manipulations were carried out under oxygen-and moisture-free conditions under an inert atmosphere of argon using standard Schlenk or Drybox techniques. Table S1: Origin and purification of solvents and reactants.
NMR signals were assigned using experimental data (e.g. chemical shifts, coupling constants, integrals where applicable) in conjunction with computed NMR data (GIAO method, cf. Computational details, p. 40). The signs of n J( 31 P, 31 P) coupling constants were derived from the calculated spectra.
For details regarding NMR spectroscopy under irradiation, see chapter 5.2 (p. 20).
IR spectra of crystalline samples were recorded on a Nicolet 380 FT-IR spectrometer equipped with a Smart Orbit ATR unit at ambient temperature.
Raman spectra of crystalline samples were recorded using a LabRAM HR 800 Horiba Jobin YVON Raman spectrometer equipped with an Olympus BX41 microscope with variable lenses. The samples were excited by a red laser (633 nm, 17 mW, air-cooled HeNe laser). Alternatively, a Buker VERTEX 70 FT-IR spectrometer equipped with a RAM II FT Raman module and an Nd:YAG solid state laser (1604 nm) was used. All measurements were carried out at ambient temperature unless stated otherwise.
Elemental analyses were obtained using an Elementar vario Micro cube CHNS analyser or a LECO TruSpec Micro CHNS analyser.

Melting points (uncorrected) were determined using a Stanford Research Systems EZ
Melt at a heating rate of 20 °C/min.
Mass spectra were recorded on a Thermo Electron MAT 95-XP sector field mass spectrometer using crystalline samples.
UV-Vis spectra were acquired on a Perkin-Elmer Lambda 19 UV-Vis spectrometer.

S5
2 Structure elucidation X-ray Structure Determination: X-ray quality crystals were selected in Fomblin YR-1800 perfluoroether (Alfa Aesar) at ambient temperature. The samples were cooled to 123(2) K during measurement. The data were collected on a Bruker D8 Quest diffractometer using Mo Kα radiation (λ = 0.71073 Å). The structures were solved by iterative methods (SHELXT) [4] and refined by full matrix least squares procedures (SHELXL). [5] Semi-empirical absorption corrections were applied (SADABS). [6] All nonhydrogen atoms were refined anisotropically, hydrogen atoms were included in the refinement at calculated positions using a riding model.
Obtaining a good data set of 2Dmp at low temperatures proved to be somewhat of a challenge since the light of the mounting lamp was sufficient to quantitatively isomerize 2Dmp to 3Dmp (which is irreversible at low temperatures), resulting in a microcrystalline material unsuitable for single-crystal X-ray structure analysis.
Fortunately, the thermal reverse reaction 3Dmp → 2Dmp is fast enough at ambient temperature that single crystals of 2Dmp stay intact for some time under light. We therefore mounted a crystal of 2Dmp at room temperature (298 K) and collected a full data set in the dark to verify the purity of the crystal. Afterwards, the crystal was slowly cooled to 123 K under strict exclusion of light, and a low temperature data set was recorded. This procedure allowed us to obtain a phase-pure diffraction pattern of 2Dmp · 0.5 PhF.
The co-crystallized solvent molecule in 2Dmp_298K was found to be strongly disordered and was therefore treated as a diffuse contribution to the overall scattering using Platon/Squeeze.
The co-crystallized solvent molecule in 2Dmp_123K was also found to be disordered, but the disorder was explicitly modelled during the structure refinement by splitting the molecule in two parts. (Due to its location on a crystallographic inversion centre, the overall number of possible orientations of the PhF molecule is doubled, i.e. the S6 disorder is modelled by four different orientations of the PhF molecule.) The position of the major component was fitted using DSR, [7] and the corresponding restraints were kept throughout the refinement. A second layer was added using the same restraints, and both parts were restrained to adopt similar structures using the SAME command. Additionally, the fluorobenzene layers were restrained to adopt a nearly mirror symmetric structure. The ADPs of all atoms within the PhF residue were restrained using the ISOR command. 3 Syntheses of starting materials 3

.1 Synthesis of DmpNC
DmpNC is synthesized according to a modified literature procedure. [8] The synthesis is carried out under non-inert conditions.
In a three-necked flask equipped with a reflux condenser and a dropping funnel,

Generation of the housane {[PN(Ter)]2(μ-CNDmp)} (3Dmp)
The housane 3Dmp can be generated by irradiation of a solution of 2Dmp with red (or white) light (λmax = 643 nm). A solution of the housane is stable for several weeks at

Generation of the housane {[PN(Ter)]2(μ-CN t Bu)} (3 t Bu)
The housane 3 t Bu can be generated by irradiating a solution of 4 t Bu2 in benzene or THF. A solution of the housane is stable for several weeks at −80 °C.
When t BuNC is added in large excess (0.25 mL, 184 mg, 2.21 mmol), the ratio between product and starting material increases to 9:1.
Isolation of the pure product failed, partly due to scrambling of the different isonitrile moieties after a few days. Moreover, the solution could not be concentrated in vacuo, since this led to removal of t BuNC and therefore re-formation of the starting material.

NMR spectroscopy under irradiation
To facilitate NMR spectroscopy under irradiation, we adopted a setup previously published by the Gschwind group, [12] who used a fibre-coupled light emitting diode (LED) to direct light into the NMR spectrometer.
In the original publication, the optical fibre was coupled with an LED by placing the polished end of the fibre directly on top of the diode. Unlike commercial solutions, this simple setup does not depend on sophisticated lens systems to focus light into the optical fibre; nonetheless, it offers efficient coupling since the area of the diode is almost completely covered by the cross section of the optical fibre.
Instead of an LED, we decided to use a laser diode (Oclaro HL63193MG, 638 nm, 700 mW), primarily due to the high optical output power at a low price (≈ 80 €) and secondly due to the narrow emission spectrum. Since the opening of the laser diode was only somewhat larger than the cross section of the optical fibre (10 m multimode fibre, 0.39 NA, high OH, 1000 µm core diameter, ThorLabs FT1000UMT), we decided to use the same optical coupling between the diode and the optical fibre as described above ( Figure S4). The diode was mounted in a metal casing to dissipate heat. The casing accepts a screwon brass cylinder that can be used to fix the optical fibre directly on top of the diode S21 and that also serves as an additional heat sink. Moreover, the cylinder may be inserted into an aluminium block (70×70×30 mm 3 ) to further dissipate heat ( Figure S5).
output) for more than one hour. At currents above 700 mA, heat quickly becomes a problem and the diode can only be switched on for short intervals of time.
As described in the original publication of the Gschwind group, [12] the outer cladding of the optical fibre was removed at the tip and the glass surface was roughened to allow uniform irradiation of the sample within the spectrometer. The fibre was inserted into an NMR tube with a coaxial insert ( Figure S7). To ensure inert conditions, all samples were prepared in a glovebox and the tubes were sealed with custom-made PTFE caps as well as 2-3 layers of PTFE tape.  To ensure that the optical fibre cannot exert lateral forces and thereby tilt the NMR tube in the spectrometer, a cap with a central bore was placed on top of the S23 spectrometer's sample inlet. The optical fibre is passed through the bore to keep it centred above the NMR tube ( Figure S8).

Photoisomerization
To study the photoisomerization process, crystals of 2Dmp (21 mg) were dissolved in THF-d8 (0.45 mL). PPh3 (11 mg, 0.06 mmol) was added as internal standard. The sample solution was filled in an NMR tube with a coaxial insert (vide supra). At first, NMR spectra were recorded in the dark, showing that only the biradical species 2Dmp was present ( Figure S9, top). The sample was then irradiated in the spectrometer (638 nm, 500 mA) for approx. two minutes. Subsequently, 1 H and 31 P NMR spectra were measured under irradiation, evidencing that the biradical had quantitatively isomerized to the housane-type species 3Dmp ( Figure S9, bottom).
In a next set of experiments, the behaviour of the t Bu derivative 2 t Bu under irradiation was investigated. Since it is formed in equilibrium from the adduct 4 t Bu2, crystals of the latter (14.9 mg, 0.02 mmol) were dissolved in THF-d8 (0.45 mL). PPh3 (11.5 mg, 0.06 mmol) was added as internal standard. The blue solution was then irradiated, resulting in quantitative formation of the housane species 3 t Bu ( Figure S10), i.e. both 2 t Bu and 4 t Bu2 were fully consumed.

Thermal equilibration
To investigate the thermal reverse reaction 3Dmp → 2Dmp, a solution of the biradical 2Dmp was irradiated in the NMR spectrometer as described above. After full conversion of the biradical to the housane 3Dmp, the laser diode was turned off and the thermal equilibration (Scheme S1) was traced by in-situ 31 P NMR spectroscopy.
Initially, a spectrum was collected every three minutes (75 scans).

S26
The time-dependent concentration data could be fitted as a first-order reaction according to the integrated time laws, where where κ is assumed to be 1 in the second equality. Hence, the Gibbs free energy of activation ΔG ‡ at 25 °C amounts to 88(4) kJ/mol.  As described before, dissolving crystals of 4 t Bu2 yielded an equilibrium mixture of 1, 2 t Bu and 4 t Bu2. Irradiation could shift the complete equilibrium towards the photoactivation product 3 t Bu, whose thermal reverse reaction (Scheme S2) was monitored by 31 P NMR spectroscopy. As in case of 3Dmp, the reverse reaction was found to be a first-order reaction. The species 1, 2 t Bu and 4 t Bu2 were recovered in their equilibrium ratios, i.e. the thermal reverse reaction of 3 t Bu is the rate determining step ( Figure S19).

S31
Scheme S2: 2 t Bu is continuously removed from the equilibrium by irradiation, leading to quantitative formation of the housane 3 t Bu. When the light source is switched off, the equilibrium mixture between 1, 2 t Bu and 4 t Bu2 is fully recovered. When a solution of 2Dmp in THF-d8 was treated with t BuNC, an equilibrium mixture of 2Dmp and 4Dmp t Bu was obtained (cf. chapter 4.5). Intriguingly, it was not possible to generate 3Dmp quantitatively. Even after prolonged irradiation (ca. 20 minutes), some amounts of 2Dmp and 4Dmp t Bu remained in solution ( Figure S20). Furthermore, as soon as the light was switched off, the housane species 3Dmp completely disappeared, and the equilibrium distribution of 2Dmp and 4Dmp t Bu was recovered. We attributed this behaviour to a catalytic influence of t BuNC on the thermal reverse reaction of the housane 3Dmp (Scheme S3).

Scheme S3:
The thermal equilibration of 3Dmp is most likely catalysed by t BuNC.
Figure S20: 31  The sample was cooled to −20 °C and irradiated. At this lower temperature, full conversion of 2Dmp to 3Dmp was observed. The light source was switched off and the reverse reaction was monitored by 31 P NMR spectroscopy as outlined before ( Figure   S21). The reverse reaction proceeded much faster than in the absence of t BuNC. Our results   An overview of the collected data sets is given in Table S3; Figure S23 and Figure S24 show the plots of 1/ = 1/ ( ) versus and the fitted curves according to eq. (3) for each data set.

Raman microscopy
Recording Raman spectra of 2Dmp and 3Dmp was anything but trivial. Due to lightinduced isomerization, samples of 2Dmp could not be excited using visible light (otherwise, a spectrum of a mixture of both isomers would be obtained). Therefore, we decided to use a 1064 nm laser, which should not be able to trigger isomerization. To acquire a spectrum of 3Dmp, a single crystal of 2Dmp was irradiated with a red laser at −80 °C, resulting in isomerization of the biradical to the housane. Raman spectra were then recorded using a red (633 nm) or an infrared (785 nm) laser. Still, it cannot be ruled out that small amounts of 2Dmp were detected in the spectrum of 3Dmp ( Figure S25).

Fluorescence spectroscopy
We attempted to record fluorescence spectra of a solution of 2Dmp in benzene.
However, no fluorescence was observed, in accordance with our calculations that predict a conical intersection between the S0 and S1 surfaces (cf. chapter 6.2) and thus radiationless deactivation of the first excited singlet state.

Nonetheless, previous investigations have shown that DFT methods can give
reasonable results if the multi-configurational character is not too large, [29] i.e. the contribution of the leading reference structure is at least 75 %. Thus, we were interested in the performance of different DFT methods, * especially with respect to the prediction of experimental data such as NMR spectra or UV-Vis absorption maxima. Furthermore, we used a model system to compare the results of different DFT functionals with accurate MRCI calculations (vide infra).

Prediction of experimental properties using DFT
In general, contrary to what is observed for "single-reference" molecules, pure density functionals outperformed hybrid DFT methods, as HF-type exchange tends to decrease

S41
the robustness towards non-dynamic correlation effects. This agrees with previously reported findings. [30] Of all functionals tested, the PBE functional was exceptionally well suited to reproduce experimental 31 P chemical shifts and coupling constants (GIAO method). [31][32][33][34][35] All functionals performed reasonably well in predicting molecular structures (for exemplary results see Table S4) as well as first excitation energies.
However, the calculated reaction (free) energies were notoriously bad, mostly due to underestimated stability of the biradical species (Table S5). Inclusion of dispersion (DFT-D3) [36,37] is paramount to produce at least qualitatively correct results. As becomes obvious from the calculated data, neglect of dispersion can to some degree cancel the errors made in the biradicals' energies; however, this is just serendipitous and does not mean that the "uncorrected" DFT functionals should be preferred to the D3 versions. a from singe crystal X-ray diffraction data; b literature data [9]

Model system
To gain further insight in the utility of DFT methods as well as to study the mechanism of the photochemical switching process, we performed a series of DFT, CAS and MRCI calculations on the model system 2H, i.e. all organic substituents were replaced by hydrogen atoms.
Firstly, the S0 and S1 potential energy surfaces were studied by CAS(8,6) calculations applying the def2-TZVP [38] basis set. The active space was chosen such that all π-type orbitals (A'' symmetry, point group CS, Figure S26) of 2H were included. The biradical 2H represents the global minimum on the S0 surface, while the housane-type species 3H corresponds to a local minimum. Both minima are connected by a single transition state (TS), i.e. the P-P bond formation/breaking as well as the folding of the ring system occur simultaneously. The electronic excitation of the biradical 2H to the S1 surface is S43 in the visible range of the electromagnetic spectrum (463 nm). Distortion of the ring system leads to a conical intersection (CI) between S1 and S0 surface which allows radiationless deactivation of the excited state ( Figure S27).  (Table S6). To obtain more reliable estimates of the relative energies discussed above, multireference CISD single point calculations (including the Davidson1 correction for the disconnected quadruples, MRCI+Q) were carried out, using the PBE/def2-TZVP geometries. The CAS(8,6) wave function was chosen as reference ( Figure S26). The agreement with the CAS results is reasonable; however, the correlated treatment lowers the first excitation energy by some 30 kJ/mol. Taking the MRCI+Q energies as benchmark values, it becomes clear that most density functionals underestimate the stability of the biradical 2H (Table S6). Nonetheless, the pure meta-GGA functionals τHCTH, [39] VSXC, [40] and, curiously, the hybrid functional B3LYP [41][42][43][44][45][46] (as implemented in Gaussian) perform remarkably well.  Note that the energy of the S1 state differs somewhat from the (more accurate) results described above; this is due to the CAS(6,6) reference that was chosen as a compromise between accuracy and computational effort. The description of the S1 state is somewhat S46 less complete in this set of computations as the reference space includes only four of the six orbitals of A'' symmetry. However, this is more of a technicality and does not influence the overall results. More excitation energies can be found in Table S7.

Biradical character
The wave function of the biradical 2 must be described by at least two determinants, Hence, the dominant configuration has a relative weight of 1 = 1 2 = 83 %. According to the scale of Miliordos et al, [47] this corresponds to a biradical character of

Electronic structure of S0 and S1 state
As discussed before, the dominant configuration of the ground state CAS/MRCI expansion contains two electrons in orbital π4 ( Figure S28 left), which is transannularly antibonding between the two P atoms. Despite contributions from the second CSF, which places two electrons into orbital π5 (net orbital occupations: π4 1.73 vs. π5 0.28), the interaction between those atoms can be classified as mostly antibonding in the ground state. The first excited singlet state can be described in several ways, depending on the method and reference wave function used: In a state-specific CAS calculation, the orbitals are variationally optimized for the first excited singlet state (i.e. the 2 nd root of the singlet CI matrix), leading to a unique set of orbitals which optimally describes the electronic structure of that state ( Figure S28 right) S48 necessarily orthogonal as they should be. This may be remedied by a state-averaged CAS calculation, yielding orbitals that are equally adequate to describe all states in question; however, these orbitals may be difficult to analyse in a chemically intuitive way.
A different approach is to use the occupied orbitals and improved virtual orbitals (IVOs) of a closed-shell HF calculation in an MRCI calculation. That is, the CAS reference is generated without re-optimization of the MOs/IVOs, and dynamic correlation is included by a CISD approach. The same orbitals are used for every state, thus making sure that all states are orthogonal to each other. Moreover, the orbitals are chemically interpretable. The drawback of this method is the high computational effort, making it only feasible for small (model) systems.
In case of 2H, the MRCI results for the ground state draw a very similar picture to the state-specific CAS calculation. The main contribution to the wave function comes from the CSF that places two electrons into the anti-bonding orbital π4. The S1 state is mainly described by an open-shell singlet CSF, with one electron residing in orbital π4 and the other one in orbital π5. This nicely agrees with the notion that one electron is promoted from the "HOMO" to the "LUMO" upon excitation ( Figure S29). However, it is important to note that other reference functions are needed to fully describe both species!

S49
All approaches described in this section give comparable results concerning the energy difference between the S0 and S1 states (cf. chapter 6.2).