Experimental study of the mutual interactions between waves and tailored turbulence

When surface waves interact with ambient turbulence, the two affect each other mutually. Turbulent eddies get redirected, intensified and periodically stretched and compressed, while the waves suffer directional scattering. We study these mutual interactions experimentally in the water channel laboratory at the Norwegian University of Science and Technology (NTNU) Trondheim. Long groups of waves were propagated upstream on currents with identical mean flow but different turbulence properties, created by an active grid at the current inlet. The subsurface flow in the spanwise–vertical plane was measured with stereo particle-image velocimetry. Comparing the subsurface velocity fields before and after the passage of a wave group, a strong enhancement of streamwise vorticity is observed which increases rapidly towards the surface for k0z ≳ -0.3 (z: vertical distance from still surface; k0: carrier wavenumber) in qualitative agreement with theory. Next, we measure the broadening of the directional wave spectrum at increasing propagation distance. The rate of directional diffusion is greatest for the turbulent case with the highest energy at the longest length scales whereas the highest total turbulent kinetic energy overall did not produce the most scattering. The variance of directional spectra is found to increase linearly as a function of propagation time.


Introduction
Synthetic photochemistry has grown tremendously in the past decade 1 and radical species are no longer considered to be difficult-to-handle intermediates 2 . Selective visible-light synthetic methods have been rapidly developed alongside photoredox catalysis 3 and milder visible-light sources have been used for popular light-driven reactions that were historically performed under UV-light irradiation [4][5][6] . The long-studied Paternò-Büchi (PB) reaction is a light-driven [2+2] heterocycloaddition reaction between an olefin and a carbonyl compound 7,8 . As first described by Paternò in 1909, this formidable light-driven reaction leads, in a single step, to a strained oxetane, which is often present in natural products and drugs 9 . This reaction and other light-driven processes have been the subject of intense mechanistic investigations (Fig. 1a), which identified diverse and divergent reaction manifolds [10][11][12][13] . In 1954, Büchi reported that, in PB reactions, the ground-state olefin reacts with the triplet excited state (T 1 ) of the carbonyl counterpart, generating a 1,4 biradical intermediate (Fig. 1a, left) 8,14,15 . In 1972, Turro reported the possibility of generating an exciplex between the reagents, possibly leading to different product distributions. 10,12 A few years later, Mattay reported that, when a charge-transfer (CT) complex between the reagents forms, the reaction proceeds through a more selective concerted mechanism 11 . The formation of a CT state proceeds through either an exciplex between an electronrich olefin and the excited ketone, or the formation of an electron-donor-acceptor (EDA) complex, which is a ground-state association characterized by a red-shifted absorption with respect to the single reactants 16 . Importantly, the spatial arrangement of the reagents determines the geometries of the CT states and the final oxetane stereochemistry 17 . As hypothesized by Kochi 18 , different stereoisomers can emerge when the cycloaddition process passes through different CT states, which lead to different spatial arrangements (for example, exciplex versus EDA complex) (Fig. 1b). In these cases, the wavelength selection is relevant for the stereochemical outcome. This hypothesis is confirmed by light-driven processes which use new types of substrates that can generate EDA complexes (such as indoles) 19,20 . These scaffolds allow broad control over the electronic and steric parameters. By using different indole-derived heterocycles, it is thus possible to channel the reactivity towards the intended stereochemical variant 21 .
Based on detailed mechanistic investigations, we herein report how to control the stereochemical outcome of light-driven [2+2] heterocycloadditions. Our study includes biorelevant indole heterocycles, aromatic ketones, and α-ketoesters. We rationalized our experimental findings by using various spectroscopic techniques, ground state and transition absorption spectroscopy (TAS), emission spectroscopy, and DFT calculations (Fig. 1c). By manipulating specific physicochemical parameters (light source or steric factors), the reaction's diastereoselectivity can be deliberately altered towards the activity of two stereodivergent reaction manifolds (exciplex versus EDA complex), leading to opposite diastereomers. These findings were extended to different indole derivatives and implemented under continuous flow irradiation, allowing access to elusive stereochemical variants on a preparative scale.

Spectroscopic identification of three different reaction manifolds
We initiated our study by selecting as the mechanistic probe the [2+2] heterocycloaddition reaction between indoles 1a-b and aryl ketones 2a-b (Fig. 2), where no diastereoselection is envisaged due to the symmetric nature of the ketone (see Supplementary Section E). Interestingly, we discovered that according to our experimental findings, and in agreement with literature, the reaction can proceed through three alternative reaction manifolds (pathways i-iii in Fig. 2), leading to a single regio-and diastereoisomer.

Conventional triplet mechanism -pathway i
When indole 1a and benzophenone 2a are involved (Fig. 2, top), a conventional radical trapping mechanism is observed 18,19 . This is a PB process commonly mediated by a UV-light source (250-360 nm) that promotes, after intersystem crossing (ISC) from the singlet state, the generation of the long-lived triplet state (T 1 ). The T 1 is then trapped by the electron-rich 1a (triplet-state I in Fig. 2, top), generating the corresponding 1,4-biradical 4. This mechanism was confirmed by TAS-experiments at a ns scale (Fig. 2, top-right 1,4-biradical box). In fact, by TAS, we observed the decay of the T 1 (max abs. at 525 nm) along with the concomitant appearance of the 1,4-biradical 4 (max abs. at 390 nm) 22 . This transient intermediate leads to the oxetane product 3a within a hundred ns (see Supplementary Section C). The experimental data clearly suggest a pure-triplet mechanism as the only operative pathway, as no traces of a radical ion-pair (IP) were detected. Additionally, a PET (photoinduced electron transfer) pathway can be excluded owing to its high endergonicity (ΔG PET = 10.3 kcal·mol -1 ). Importantly, it is well documented by several literature reports that when the ΔG PET > 0 kcal·mol -1 , PET processes are prevented and the reaction proceeds through a radical-trapping mechanism 23,24 .

Exciplex-based pathway -pathway ii
In the presence of a more electron-deficient ketone such as 2b (m-CF 3 benzophenone) and under diluted conditions ([2b] = 0.01M in acetone) (Fig. 2, middle) we observed an alternative mechanism. Analogously to pathway i, no ground state interactions between 1a and 2b were detected (see Supplementary Section C). Nevertheless, TAS experiments revealed the appearance of the transient radical IP (5 + 6) upon T 1 decay, as supported by the spectral changes at λ=780 nm characteristic of the benzophenone radical anion (Fig. 2, radical-ion-pair box and Supplementary Sections C and D). Additionally, no traces of the biradical 4 were observed at λ < 400 nm, indicating that the radical IP evolves through a concerted or quasi-concerted mechanism to the oxetane product 3 25 . This data implies that after the ketone excitation, the T 1 state of 2b interacts with the indole generating the exciplex II, that possesses a charge transfer character consistent with the exergonicity of the photoinduced electron transfer (PET) process (ΔG PET = -9.7 kcal·mol -1 ). The direct spectroscopic signature of the exciplex was investigated for benzil 7 and indole 1b (vide infra). This is because the low visible-light absorption of benzophenone 2b at 405 nm led to the selective excitation of the EDA complex with a negligible contribution of the ketone direct excitation.

EDA-complex-based pathway -pathway iii
This pathway foresees the interaction of the two reactants at the ground state level into an EDA complex under more concentrated conditions ([2b] = 0.1 M in Acetone) (Fig. 2, bottom). UV-Vis titration experiments with the electron-rich indole 1b and the electron-poor benzophenone 2b, revealed the appearance of a red-shifted CT band (Fig. 2, EDA complex, bottom left). A modest binding constant (K EDA = 0. 47 M -1 , in acetone at 25 °C) was extrapolated for the equilibrium of formation of the complex (see Supplementary Section C). Within an EDA-complex-based manifold, a stereochemically alternative radical IP (5 + 6) is directly produced upon PET (λ max 723 nm). As monitored by TAS, a favourable PET process (ΔG PET = -9.7 kcal·mol -1 ) occurs delivering the radical IP, whose spectral assignment is confirmed by spectroelectrochemistry (Fig. 2, alternative radical-ion-pair box and Supplementary Sections C and D). This species decays with a time-constant of 180 ns into the oxetane product.
We envisaged that the three diverse reaction manifolds observed by TAS are governed by alternative orbital interactions. For instance, the interactions within pathway iii differ from pathway i and pathway ii, which lack the π orbital overlap of the indole with the half-filled n P orbital of the excited ketone. Indeed, it is well established that for EDA-complexes alternative π-π interactions are relevant 16 . Importantly, diverse orbital interactions may lead to diverse spatial orientations of the reagents and finally alter the stereoselectivity of the process. For this reason, we decided to gain insights into the diverse molecular orientations by using DFT methods.

Computational insights into the exciplex versus the EDA-complex pathway
Aiming at developing a novel stereodivergent process, we focused our attention on the exciplex-based pathway ii and the EDA-complex-based pathway iii, with 1b and 2b as reactants.
Under diluted conditions ([2b] = 0.01 M in acetone), no EDA complex is detected, meaning that under these conditions the only active pathway is the exciplex-based pathway ii. Conversely, when more concentrated conditions ([2b] = 0.1 M in acetone) are used the EDA-complex pathway iii starts to be competitive.
To gain insight into the two diverse arrangements involved in the PET-based pathways (exciplex versus EDA complex), we employed DFT calculations at the (U)M06-2X/6-311+ +G(d,p)/_IEFPCM(acetone) level of theory ( Fig. 3) 17 . This functional is a well-established method to describe charge transfer complexes both in singlet and triplet states 26,27 .

Excited-state (T 1 ) exciplex arrangement, pathway ii
Upon direct excitation of the ketone 2b, the corresponding T 1 state is generated (Fig. 3a). At this juncture, the T 1 interacts with the ground state 1b 10 . In this case, the oxygen atom of the T 1 2b and the C2 position of 1b are in close proximity (Fig. 3a, n p -π interaction, 2.56 Å). Interestingly, the spin density map of the computed exciplex is almost equally distributed between the indole and the ketone (1.03 and 0.97 respectively (Fig. 3)). However, a large dipole moment is observed upon excitation (μ=15.6 D). This altered charge distribution indicates a remarkable CT character as reflected by natural population analysis. In fact, a total of 0.90 charge unit is transferred from indole to the ketone, corroborating the PET-based pathway ii, where the radical ion-pair has been detected experimentally by laser flash photolysis (LFP) measurements.

Ground-state (S0) EDA-complex arrangement, pathway iii
As experimentally observed, when concentrated conditions are used ([2b]=0.1 M in acetone) an EDA-complex is generated between the reagents (Fig. 3b). This complex is held together through a π-π stacking interaction between the electron-rich aromatic ring of indole 1b and the electron-poor aromatic ring of the ketone 2b (interaction ~3.6 Å (  27,28 . Additionally, the analysis of the frontier molecular orbitals of the EDA-complex reveals that the HOMO is located on the indole moiety while the LUMO is distributed along the ketone resulting in a dipole moment (μ) of 4.8 D (Fig. 3). To better elucidate the photoinduced electron transfer observed after irradiation of the EDA complex, we performed vertical excitation calculations employing a time-dependent density functional theory (TD-DFT) method. This technique allowed us to analyze the corresponding excited state involved in the PET process. For the S 1 excited-state, a HO-MO-LUMO contribution of 83.5% has been computed (3.66 eV), showing a remarkable charge transfer character (Fig. 3, μ=16.2 D). Indeed, natural population analysis of S 1 indicates that upon vertical excitation of the EDA complex a substantial charge (0.79) is transferred from the indole moiety 1b to the ketone 2b, which resembles the experimentally observed radical ion-pair system.

The impact of the light source on the diastereodifferenciation process
Based on the different geometries and orbital interactions obtained by DFT calculations (Fig. 4), it is possible to represent the two alternative reaction manifolds by using a simplified potential energy surface diagram (Fig. 4b) 28 . The spatial arrangements of the molecules are retained and directly transferred to the final oxetane product due to the concerted-character of PET-based PB reactions 11,12 . For instance, when the reaction mixture is irradiated at 405 nm, the CT-band of the EDA complex is directly excited (Fig. 4b left, red lines). This irradiation leads to the formation of a radical IP with a preferred π-π interaction. However, 2b can be directly excited at 370 nm. In this case the n p -π interaction will govern the spatial arrangement of the IP (Fig. 4b right, green lines). Thus, a light stereodifferentiation process can occur when switching the irradiation source from UV-to visible-light.
As per the DFT calculations, there are two alternative sets of interactions, which do not result in any product diversification when using symmetric benzophenones. These are crucial when moving to prochiral carbonyls. We therefore evaluated benzil 7, which is a class of synthetically relevant prochiral carbonyls used in several polar and radical reactions ( Fig. 4a) 29,30 . We hypothesized that the endo:exo selectivity 24 of the reaction would depend on the competition between the two alternative reaction manifolds (exciplex versus EDA complex). The conventional pure-triplet mechanism (pathway i) was ruled out by the LFP experiments of 7 with 1b (see Supplementary Section F), which instead confirmed the photogeneration of the CT state, in agreement with the favorable ΔG PET (-5.8 kcal·mol -1 ). Thus, by altering the irradiation wavelength and the parameters that influence the EDA complex formation, the endo:exo selectivity can be deliberately altered in a diastereodivergent light-driven process (see Supplementary Sections E and F). Specifically, 8-endo will be the preferred product in an exciplex-based mechanism, while 8-exo will be favoured following an EDA complex pathway (Fig. 4a, left). Remarkably, we observed the formation of 8-endo as a single diastereoisomer when running the reaction at 370 nm (>99:<1 d.r., 370 nm Fig. 4c). The endo selectivity results from the direct excitation of the free ketone 7, participating in an exciplex with 1b. 19 As hypothesized, when moving to less energetic 405 and 456 nm irradiation wavelengths, the exo selectivity increases, with the d.r. moving to 77:23 (456 nm Fig. 4c, left column). This is in line with the activity of the excited EDA complex between benzil 7 and the indole 1b, whose presence was confirmed by the UV-Vis absorption spectra (see Supplementary Sections E). Increasing the concentration from 0.01 to 0.1 M favors the EDA complex pathway, resulting in a d.r. of 93:7 and 68:32 upon irradiation at 370 and 405 nm, respectively. No significant variation was observed at 456 nm, indicating that 64:36 d.r. is the limit value under these reaction conditions (Fig. 4c). Following previous experimental observations, we reasoned that performing the reaction under reduced temperature should inhibit the interconversion between the two reaction manifolds, thus magnifying the observed light stereodifferentiation process. 31 Indeed, when running the reaction at -78 °C (in acetoned 6 at 0.1 M), the d.r. moved from >99:<1 at 370 nm to 52:48 at 456 nm. This d.r. variation is the highest ever registered for PB light-diastereodifferentiation processes, and it confirmed our initial working hyphothesis [32][33][34] . Furthermore, these experimental data reveal that the choice of the light is crucial. Reproducibility issues in diastereoselective PB processes involving heterocyclic systems can be now rationalized and possibly solved by selecting the appropriate irradiation wavelength 34 . As a control experiment, we performed a reaction in an aromatic solvent, which interferes with the π-π interactions in the EDA complex. Accordingly, when using C 6 D 6 as solvent, we observed reduced activity of the EDA-complex pathway, resulting in a d.r. no lower than 80:20 at 465 nm (Fig. 4c, right  column). It is worth noting that across all the performed reactions we have always observed complete regiocontrol. The change of the solvent (for example, acetone versus PhMe Reichardt parameter: acetone (electron-transition energy, ET(30) = 42.2 kcal·mol -1 ), toluene (ET(30) = 33.9 kcal·mol -1 ), that has been reported to have high impact on the regiocontrol of other PB processes 35,36 , did not impact the high regiocontrol of the reaction between indole 1b and ketone 7. This behaviour is ascribable to the high steric hindrance of the indole precursors (bearing bulky tButoxy carbonyl, Boc and a tBu Silyl, OTBS groups at position 2).
In addition to the experimental data, we used DFT and TD-DFT calculations to compute the two different geometries for the EDA complex between ketone 7 and indole 1b (see Supplementary Section K). Notably, the excited state (S 2 , HOMO-LUMO contribution=98.4%) of this complex in the exo configuration has a larger dipole moment (μ=18.6 D) than in the endo geometry (μ=15.4 D). This indicates a substantial chargetransfer character and is in accordance with the experimental trend.

Spectroscopical characterization of the exciplex intermediate, pathway ii
In order to confirm the key reactivity of an exciplex between substrates 1b and 7, we performed transient absorption spectrometry (TAS) measurements at a fs timescale (Fig. 4dg). These experiments allowed us to investigate the dynamics of the systems within the first nanosecond after photoexcitation with a time resolution of about 150 fs (see Supplementary Section F). We compared the photophysics of the single benzil 7 in the absence and in the presence of indole 1b (Fig. 4d and  For 7, the excitation at 400 nm results in the formation of positive excited state absorption features (ESA). At early time periods, we recorded a broad absorption band centered at about 560 nm, which gradually evolved into a sharper feature with a maximum at about 525 nm. While the literature contains few reports of the ultrafast dynamics of 7, which are typically performed at longer time resolution, it is reasonable to assign the early time signal at 560 nm to the ESA from the instantaneously photoexcited S 1 state (S 1 →S n transition) 37 , 38 . This signal quickly decays (0.3 ps) while the band at 525 nm rises with the same time constant. Once formed, this band decays over a longer timeframe than the explored 600 ps. This behaviour is in agreement with a fast S 1 →T 1 ISC that moves the population from S 1 to T 1 on a sub-ps timescale, followed by the long-lived T 1 →T n absorption at 525 nm 39 .
The TAS spectra of the mixture 7-1b (Fig. 4e) are qualitatively similar to the TAS spectra of 7. This is not surprising, given that the excitation at 400 nm preferentially excites 7 and therefore its spectral features are expected to prevail in the mixture response. However, the dynamic behaviour is significantly different, with the ESA band at 525 nm decaying much quicker in the mixture than in 7 (Fig. 4f-g). The time trace of 7 alone could be fitted with a bi-exponential model characterized by a fast rise (t rise =0.3 ps) assigned to the S 1 →T 1 ISC and a slow t decay (>500ps) assigned to all the relaxation processes from the T 1 state to the S 0 ground state. Interestingly, an additional relaxation pathway is observed for the triplet state in the presence of 1b, with a time constant of about 50 ps, which is associated with the formation of triplet exciplex species as depicted in Fig. 4h. Indeed, the formation of the triplet state is known to lead to a planarization of the benzil molecule, 36 which favours the exciplex formation, as confirmed by our TD-DFT calculations (see Supplementary Section K). We can exclude the 50 ps time constant corresponding to the direct formation of a 7-1b CT state. This is because TAS experiments performed on the 7-1b mixture confirmed that this species gives rise to a characteristic ESA signal at significantly longer wavelengths (around 600 nm) and has much longer dynamics.
Adding to previous experimental evidence and supported by DFT calculations, the findings support the existence of an exciplex between substrates 1b and 7 under these experimental conditions, while offering a rare report of the activity of this elusive and fleeting intermediate.

The impact of the electronic and steric factors on the diastereodifferentiation process
Having verified the key activity of the exciplex under the light-driven stereodifferentiation processes, we next investigated how steric and electronic parameters can be used to rationally control the stereodifferentiation events. Specifically, we selected α-ketoesters 9 as a molecular probe to monitor the impact of steric and electronic parameters on the reaction outcome, while keeping a constant visible-light source set at 405 nm (Fig. 5). Interestingly, the steric repulsion between the R group and the substituent at the C-2 of the indole 1b (OTBS) will favor a carbonyl perpendicular orientation in the exciplex, leading to 10-exo (Fig. 5a, exciplex box). However, the 10-endo selectivity will be preferred when favoring a π-π interaction between the aryl group of 9 and the aromatic indole core (Fig. 5a, EDA-complex box).

Evaluation of the electronic factors
We began by evaluating the impact of the electronic factors on the reaction outcome (Fig.  5b). First, we selected various α-ketoesters 9 recording the corresponding UV-Vis spectra in the presence of 1b, thus assessing their ability to generate an EDA complex. In the case of α-ketoester with Ar = 3,5-CF 3 Ph, we observed a red-shifted CT band (see Supplementary Section G). Additionally, the highly negative ΔG PET = -17.1 kcal·mol -1 measured was consistent with the formation and activity of an EDA complex. No spectral variations were registered for the other α-ketoesters. We next examined the experimental outcome for all the substates under a 405 nm irradiation (Fig. 5b).
Using a visible-light source was crucial because it enabled the activity of the exciplex and the EDA complex pathways (see Supplementary Section G). This meant that the electronic and steric parameters were the key features in governing the selectivity of these photoreactions. In fact, both the EDA complex and free α-ketoesters absorb light under these conditions. Only the PB reaction with the electron-deficient 3,5-CF 3 Ph α-ketoester was highly endo-selective (84:16 d.r., Fig. 5b). The other α-ketoesters showed no groundstate association with 1b, while maintaining a high level of exergonicity (ΔG PET in the range of -6.0 to -9.5 kcal·mol -1 , Fig. 5b) and resulting in similar d.r. of about 70: 30. 11, 24 We depicted the differences in reactivity by plotting the obtained experimental d.r. against the E red of 9 and highlighting the different behavior of the 3,5-CF 3 Ph α-ketoester (Fig. 5b,  right).

Evaluation of the steric factors
We then investigated the role of the steric factors, tuning the R group within 9 (Fig. 5c). The Charton analysis allowed us to compare the sensitivities (ψ) of the two different manifolds (EDA complex versus exciplex) [41][42][43] . When increasing the R size (from Me to i Pr), the endo:exo selectivity was reversed, passing from 74:26 to 43:57 d.r. for Ar = Ph (exciplex manifold), and from 84:16 to 45:55 d.r. for Ar = 3,5-CF 3 Ph (EDA complex manifold). We rationalize this trend as reflecting an increased steric repulsion between the R group and the OTBS, favoring the perpendicular approach. Nevertheless, the EDA complex manifold experiences a 1.5 increased sensitivity (ψ = -1.51 versus ψ = -2.23 for the exciplex and EDA manifolds, respectively), in agreement with a greater dependence of the parallel approach (π-π interactions) towards steric variations. Finally, we sought to use these findings to deliberately alter the diastereoselective outcome of the PB process by manipulating the steric factors. It is worth mentioning that a light-diastereodifferentiation process was also observed for the reaction between the α-ketoester 9 (Ar = 3,5-CF 3 Ph) and the indole 1b, with the d.r. endo:exo passing from 84:16 under 370 nm irradiation to 57:43 under a 427 nm irradiation at -78°C. The lower d.r. variation registered with respect to the reaction performed with benzil 8 is likely because of the very similar absorption profile between the EDA complex and the single substrate (See Supplementary Section G). We selected the 3,5-CF 3 Ph α-ketoester because it is the only substrate that formed an EDA complex with 1b (ψ = -2.23). We observed an initial d.r. of 84:16 in favor of the endo product. The endo selectivity increased slightly at 40 °C (89:11). Extensive studies performed by Inoue, Mori and Abe have demonstrated that the use of high temperature (>40 °C) generally favours more strongly interacting such as π-π stacked EDA complexes over more flexible structures such as an exciplex 31 . When passing from R = Me to R = i Pr, the endo:exo selectivity was reversed with a d.r. of 45:55 and further magnified to 16:84 when lowering the temperature (Fig. 6). We thus isolated and characterized the elusive exo-isomer of oxetane 10 in synthetically useful isolated yield (58%). These results demonstrate that simple structural variations (Me versus i Pr) in the starting material can dramatically alter the stereochemical outcome of light-driven processes. The importance of the proper light-source selection is also key. No variations were observed when running the reaction at 370 nm. Moreover, these results indicate that overlooked minor modifications of a substrate can have a tremendous impact on the reaction outcome of [2+2] photocycloadditions.
Following the rational approach used for the previous reaction partners, computational analysis was performed (see Supplementary Section K). We computed the EDA complexes and the corresponding vertical excitations between the α-ketoester (9, R = Me) and indole 1b. Remarkably, the corresponding excited state (S 1 , HOMO-LUMO contribution=94.4%) of the EDA complex in the endo configuration (major diastereoisomer) has a larger chargetransfer character (μ=15.8 D) than the exo geometry (μ=14.3 D). This trend is in agreement with previous data on the benzil-indole couple. Moreover, similarly to the 1b-2b exciplex, the complex in the triplet state between 1b and 9 (R = Me) possesses a large dipole moment (m=15.9 D), as reflected by the charge transferred (0.95) from indole to the ketone, which supports the PET-based pathway.

Generality and synthetic potential of the developed diastereodifferentiation processes
Having established two alternative yet complementary methods to govern the stereochemical outcome of the PB reaction, we sought to extend our findings to different carbonyl precursors and to other types of indole derivatives (Fig. 7a). We selected different indoles characterized by various steric and electronic modifications at C-5 and C-6. These new substrates were then subject to the light-driven and the steric factor diastereodifferentiation process under the developed conditions. When changing the light source from UV (370 nm) to visible light (456 nm), a significant variation in the d.r. was observed in all cases. In particular, the presence of an electron donating group, EDG (OMe) stabilized the EDA complex, resulting in higher exo-selectivity (53:47). However, the presence of an electron withdrawing group EWG (CF3) did not particularly alter the reaction outcome, demonstrating the robustness of our findings.
In terms of the steric parameters, switching from Me-to i Pr-α-ketoester dramatically changed the stereochemical outcomes, with the d.r. inverted in all cases. Specifically, the electron-rich OMe group induces an EDA-complex manifold prevalence, while the CF3 group favors the exciplex manifold by disfavoring the π-π interactions with the electrondeficient α-ketoester 9.
After evaluating alternative indole derivatives, we explored the preparative potential of the developed light diastereodifferentiation process. The reaction between benzil 7 and indole 1b was readily implemented into a flow reactor ( Fig. 7b and Supplementary Section I) 44 . The reaction in flow furnished identical results with a shorter residence time (40 min) 45 . Furthermore, under the flow conditions, we ran the reaction at 1 mmol scale, isolating both the 8-endo and the 8-exo product with an overall yield as high as 88% and 0.57 g. This unprecedented light stereodifferentiation flow process lays the foundation for the use of our protocol in semipreparative medicinal chemistry settings 46,47 .

Conclusion
In conclusion, we have been able to gain control over the stereochemical outcome of light-driven [2+2] heterocycloaddition reactions involving diverse types of carbonyls and indole derivatives. Our results are based on experimental results and corroborated by in-depth investigations of the reaction mechanisms performed by photodynamic analyses, TAS-experiments and supported by spectroelectrochemistry and DFT calculations. With a rational approach, we have revealed the major role of a visible-light-driven EDA complex manifold that can channel the reactivity towards unconventional diastereoselectivity. The often-overlooked effect of the light-source variation was exploited for the development of new light-stereodifferentiation reactions (d.r. endo:exo passing from >99:<1 up to 43:57).
Additionally, the punctual modulation of the steric parameters of the starting ketone allows access to unprecedented diastereoinversion events under visible-light irradiation (d.r. endo:exo passing from 89:11 to 16:84). The generality and synthetic utility of our findings was further proved for a series of structurally diverse indoles (Fig. 7) and by implementing the light stereodifferentiation process in flow (Fig. 7c).
Our study demonstrates how the mechanistic understanding of light-driven processes can increase their synthetic potential to previously inaccessible structural targets. Further, we envisage a widespread utilization of our approach to the development of novel and selective photoreactions as well as to the mechanistic elucidation of unconventional light-driven reactivity.

General setup of the light-driven [2+2] cycloaddition reaction
The indole (1 equiv.) and the ketone (1 equiv.) substrates were added into a glass 4 mL glass vial. The solvent was added, and the solution was rapidly degassed with Ar for 1 min. The vial was placed at 5 cm distance from the light source. The reaction was stirred vigorously. To maintain a stable reaction temperature one fan was placed at the top of the photoreactor (20±2 °C) and the temperature was controlled by a thermometer (see Supplementary Fig.  S.A.4). The crude reaction mixture was analysed by 1 H-NMR spectroscopy to determine the product distribution. The isolated yield was afforded after column chromatography on silica gel.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material.

Europe PMC Funders Author Manuscripts
Europe PMC Funders Author Manuscripts