Metal-ligand cooperative activation of HX (X=H, Br, OR) bond on Mn based pincer complexes

Reversible dissociation of H (cid:0) X bond (M (cid:0) L + H (cid:0) X ! M(X)-L(H); D G HX ) is an important step during pre-activation, catalysis and possible deactivation of acid-base cooperative pincer based transition metal catalysts (M (cid:0) L). Herein we carried out a high-throughput computational investigation of the thermodynamic stability of different adducts in various functionalized Mn(I) based pincer complexes. We used a combination of density functional theory (DFT) and density functional tight binding (DFTB) calculations to analyze D G HX of > 700 (M(X)-L(H)) intermediates based on functionalized variants of four pincer type ligand scaffolds derived from PCP, CNC, PNP and SNS ligands. We discovered linear scaling relations between D G HX of various species. Strongest correlations were found between species of similar size and chemical nature e.g. D G tBuOH correlated best with D G iPrOH and worst with D G HBr . Such scaling relations can be useful for property based screening of catalysts and selection of (co)solvent/substrate/base for optimized reaction conditions. We also investigated the influence of the ligand backbone and the functionalization of donor and backbone sites in the ligand. Our analysis reveals the crucial role of the second coordination sphere functionalization for the reactivity of the complexes with impact in some cases exceeding that of the variation of the functional groups directly attached to the donor atoms.


Metal-ligand cooperative activation of HX (X = H, Br, OR) bond on Mn based pincer complexes
Annika M. Krieger, [a] Vivek Sinha,* [a] Adarsh V. Kalikadien, [a] and Evgeny A. Pidko* [a] Reversible dissociation of HÀ X bond (MÀ L + HÀ X!M(X)-L(H); DG HX ) is an important step during pre-activation, catalysis and possible deactivation of acid-base cooperative pincer based transition metal catalysts (MÀ L). Herein we carried out a highthroughput computational investigation of the thermodynamic stability of different adducts in various functionalized Mn(I) based pincer complexes. We used a combination of density functional theory (DFT) and density functional tight binding (DFTB) calculations to analyze DG HX of > 700 (M(X)-L(H)) intermediates based on functionalized variants of four pincer type ligand scaffolds derived from PCP, CNC, PNP and SNS ligands. We discovered linear scaling relations between DG HX of various species. Strongest correlations were found between species of similar size and chemical nature e. g. DG tBuOH correlated best with DG iPrOH and worst with DG HBr . Such scaling relations can be useful for property based screening of catalysts and selection of (co)solvent/substrate/base for optimized reaction conditions. We also investigated the influence of the ligand backbone and the functionalization of donor and backbone sites in the ligand. Our analysis reveals the crucial role of the second coordination sphere functionalization for the reactivity of the complexes with impact in some cases exceeding that of the variation of the functional groups directly attached to the donor atoms.

Introduction
Pincer complexes are important catalysts in organometallic chemistry for multiple applications such as transformation and synthesis of imines, amines, peptides, pyridines, pyrroles, acetals, and carboxylic acid derivatives, such as esters, ketones and amides. [1][2][3] Owing to their success with (de)hydrogenation of a wide scope of substrates, pincer complexes have been adopted favorably by the pharmaceutical, fine chemicals and the energy industry [4] (representative examples shown in Figure 1a,b). The well-defined geometry and tridentate coordination mode of this class of complexes offers a stable catalytic structure. Most highly active pincers such as Ru-MACHO complex [5,6] and Nozaki's Ir-PNP complex [7] are based on expensive Ru and Ir metals. Catalytic systems based on such metals are not desirable for large scale ubiquitous applications due to high cost and limited availability. Several successful examples of pincer catalysts based on earth abundant 3d transition metals (TMs) such as Fe and Mn have been realized in the last decade. [8,9] However, the activity and stability of such catalysts based on first row TMs remains a challenge. Therefore, the development and optimization of catalysts based on 3d transition metals is an active and highly sought after area of research. [10][11][12] Manganese is particularly attractive as the active metal in such catalysts in view of its high biocompatibility, which is of interest for industries in the food or pharmaceutical sector.
Functionalization of the ligand scaffold can be used to explore the chemical space of TM pincers in the pursuit of highly active and stable catalysts based on first row TMs. In such an approach one can start with a "skeleton" complex bearing a TM center coordinated to a pincer scaffold. Selected sites on the scaffold or metal center can be functionalized generating an ensemble of new TM complexes using various combinations of functional groups. [13][14][15][16] Experimentally only a handful of functionalized variants of pincer ligand scaffolds have been reported. Moreover, synthesis and subsequent testing of the catalytic activity of functionalized TM complexes quickly becomes intractable. Theoretical consideration of the functionalized variants that are synthetically not accessible can provide an insight into rational design principles. In this regard, computational methods are relevant and can be applied to screen through a large ensembles of functionalized TM complexes. [17,18] Recently such approaches have been applied to screen TM complex including pincer complexes for activity, regioselectivity and ligand effects. [19][20][21] Herein we screened the effect of functionalization of the ligand backbone on the stability of potential catalytic intermediates on Mn(I)-pincer complexes. The focus was on the determination of the effect of the type of functionalization and the functionalization site (backbone/donor site). By choosing functionalization sites directly coordinated to the Mn(I)-center, the effect of the first coordination sphere can be analyzed. The effect of the second coordination sphere is rationalized by investigating functionalization on a site not directly coordinated to the metal. We chose five representative pincer ligand scaffolds, namely, PNP-(bis(3-phosphaneylpropyl)amine)-, SNS-(azanediylbis(ethane-1-thiol))-, CNC-(bis(2-(1H-3λ 4 -imidazol-3-yl) ethyl)amine)-, PNN-(N 1 -(2-phosphaneylethyl)ethane-1,2-diamine)-, and PCP-(N 1 ,N 3 -bis(phosphaneyl)benzene-1,3-diamine)backbones coordinated to a Mn(I) center stabilized by CO ligands as illustrated in Figure 1c. Our analysis included the pristine complexes as well as their catalytically relevant intermediates resulting in over~1200 structures based on the five selected pincer ligand scaffolds ( Figure 1c).
Pincer complexes based on these ligands have been reported for various transition metals, including manganese ( Figure 1b). [8,13,30,31,[22][23][24][25][26][27][28][29] SNS-, [32][33][34][35][36][37] CNC-, [38][39][40][41][42][43] and PCP-ligands [44][45][46][47][48] are primarily known for their use in 4d and 5d transition metal catalysis (Figure 1a). Especially in 4d and 5d-transition metal catalysis high turnover frequencies and turnover numbers are reported for the complexes. [31,40,42,49] Literature on 3d-transition metals generally reports lower catalytic efficiencies, [30,48] which indicates that there is opportunity to maximize their potential towards sustainable catalyst systems. Catalysts based on 3d metals are known to be more prone to deactivation and formation of resting states, limiting their reactivity. [4,50] Herein, we investigated Mn(I) -pincers as potential (de) hydrogenation catalysts. Possible applications are the storage of H 2 in unsaturated moieties such as CO 2 , the reduction of organic substrates with H 2 gas or with hydrogen donors such as i-PrOH. [51][52][53] Figure 2 illustrates representative catalytic cycles of dehydrogenation of methanol to formaldehyde and hydrogenation of acetone to isopropanol. The Br adduct (M(Br)À L(H)) is a common precursor to the active form of Mn-pincers. [11] The activation of the catalyst is commonly carried out by the  reaction with a strong base (e. g. KOH or KOt-Bu). This yields a 5-coordinated complex (MÀ L), regarded as a pristine activated catalyst and which features a Lewis acid site on the metal and the ligand can act as a Brønsted base i.e. the metal can coordinate with an electron donating species while the ligand can accept a H + . The activated catalyst is susceptible for potential deactivation/inhibition through the metal-ligand cooperative addition of alcohol/water/base resulting in the formation of À OR adducts. [54] Alkoxide adduct of hydrogen donating alcohols such i-PrOH, MeOH and EtOH are often formed as intermediates in the course of catalytic hydrogenation reactions, and can even act as the resting states limiting the catalytic performance depending on their stability. [50,[55][56][57][58][59][60] Competitive bonding of other species such as the solvent or the nucleophile base to the metal can slow down or even deactivate the catalyst.
For example, water can compete with methanol for the catalytically active site via the formation of a stable hydroxide adduct upon reaction with the catalyst or via ligand exchange with the methoxide adduct ( Figure 2). To continue the catalytic cycle, the alkoxide adduct must convert to the hydride adduct (M(H)-L(H)), which in turn regenerates the catalyst by hydrogen evolution/transfer. Catalytic turnover is inhibited if the alkoxide adduct is very stable compared to the hydride adduct. Similarly, an excessively stable hydride adduct would render it inactive towards hydrogen liberation resulting in an adverse effect on the catalysis.
When employed as a hydrogenation catalyst, the hydride adduct is formed in the first step via heterolytic H 2 dissociation. A less stable hydride would be prone to H 2 recombination instead of catalytic turnover. On the other hand, excessive hydride stability would make it less reactive towards the hydride transfer steps of the catalytic cycle. The alkoxide adduct formed upon the hydride transfer to C = O must dissociate to regenerate the catalyst.
Transfer hydrogenation reactions with e. g. isopropanol often proceeds via the formation of an intermediate alkoxide species. Excessive stability of this alkoxide adduct would have a deactivating effect and adversely affect the catalytic turnover. From a mechanistic perspective the relative stability of hydride, hydroxide and alkoxide intermediates are important for the catalytic turnover. Their relative stability, in accordance with the Sabatier's principle should be balanced and any excessive stabilization/destabilization would have an inhibiting effect on the catalyst.
In the present work we investigate functional strategies to tune the stability of aforementioned intermediates. Our results allow comparison of different ligand scaffolds and functionalization strategies in a common framework. We have identified linear free energy scaling relations (LFESRs) between various intermediates. By analyzing relative stabilities of various adducts we analyze their competitive binding at the metal center. We describe the impact of functionalization of the complexes near the metal center and on the ligand backbone. Finally, we draw conclusions about catalyst behavior and formulate perspectives on catalyst activation strategies, choice of solvent environment and possible deactivation species. The paper is organized as follows: in the computational methods section we describe the functionalization approach, computational model to investigate the thermodynamic parameters, and details of the quantum chemical calculations applied. Next we describe and discuss the results from our calculations. Finally we summarize the results and present our conclusions.

Computational methods
Electronic structure calculations on the transition metal complexes were carried out either by using the extended density functional tight-binding (DFTB) or the density functional theory (DFT) methods. For a large number of calculations, a full DFT based approach is computationally expensive. Extended DFTB calculations via the xTB code from Grimme's group has recently emerged as a rapid tool with reasonable accuracy to predict geometry and thermochemistry of various chemical systems including TM complexes. [61] We therefore performed xTB calculations on all the complexes in our paper. On a selected number of TM complexes (432) based on CNC, PCP and PNN ligands, we also performed DFT calculations. xTB calculations on the same 432 complexes were compared with DFT based predictions to determine the accuracy of xTB.

Extended density functional tight-binding calculations
Extended DFTB calculations were performed using the xTB software suite (version 6.3.3). [62,63] The GFN2-xTB method was applied for geometry optimization, using the verytight criteria. Hessian matrix calculations were performed for all optimized geometries to verify the absence of imaginary frequencies and that each geometry corresponds to a local minimum on its respective potential energy surface (PES). The GBSA solvation model parametrized for THF as implemented in xTB was used to account for solvent effects.

Density functional theory calculations
Density functional theory (DFT) calculations were performed using the Gaussian 16 C.01 suite of software. [64] All geometries were optimized using the BP86 functional with a def2-SVP basis set in the gas phase. [65] This combination of functional and basis set gives reliable geometry predictions accompanied with low computational costs. [66,67] Hessian calculations were performed to ensure that all optimized geometries were a minima on the PES (no imaginary mode). Zero-point energy and Gibbs free energy corrections to electronic energy were obtained from hessian calculations within the harmonic approximation under standard conditions (298.15 K, 1 bar). Single point (SP) energy calculations with the SMD [68] solvation (THF) model were carried out using the PBE0 hybrid functionals [69] with a triple zeta basis set (def2-TZVP) to further refine the electronic energies. SPs were also carried out using the BP86 functional. [70] We denote such composite methods BP86/def2-SVP//XC/def2-TZVP (THF), as bp86(thf) or pbe0(thf) depending on the exchange-correlation (XC) functional used for the SP single-point calculations. This allowed to investigate the impact of solvation (bp86(gas) vs bp86(THF)) and the functional (bp86(THF) vs pbe0(THF)) on the computed free energies. All DFT calculations were performed with dispersion correction (D3). [71] DFT calculations were performed for a selected number of complexes bearing CNC, PNN and PCP ligands. We focused our investigation on the addition of HÀ X species (X = H, OH, MeO, EtO, i-PrO, t-BuO, Br) across the catalyst which is represented as MÀ L, where M represents the metal center and L represents the ligand. Addition of HÀ X across MÀ L leads to formation of M(X)À L(H) species, where metal forms an adduct with X and the ligand gets protonated. We estimated the thermodynamic stability of M(X)À L(H) by computing the Gibbs free energy change under standard conditions upon addition of HÀ X moiety across MÀ L bond (eq. (1)).

Functionalization strategy
All functionalized geometries were obtained via an in-house developed automated python based workflow. [72] We chose two different functionalization sites: four R 1 sites which are located near the metal center, and two R 2 sites (only one in case of PCP) which functionalize the ligand backbone as shown in Figure 1. We chose to perform symmetric functionalizations meaning all four R 1 sites were kept the same, and both R 2 sites were also functionalized with the same ligand. R 1 and R 2 were however not constrained to be the same. In addition to functionalization sites on the ligand, seven Mn-adducts were also considered which included vacant site (pristine complex), H, Br, MeO, i-PrO, EtO and t-BuO adducts. This functionalization scheme gener-ated~1200 geometries of metal complexes. Out of these 1225 geometries, we filtered geometries where the pristine complex had a hemi-labile ligand resulting in a total of 732 geometries which are discussed in this work. We found that most PNN based complexes resulted in hemi-labile ligand. Hemi-lability can arise as an artifact of xTB based geometry optimization or it can be genuinely present in the system. Since this would require further investigation we excluded all xTB based results for the PNN catalyst.

Comparison of xTB with DFT
Low computational cost and wide applicability of xTB calculations make them suitable for high throughput screening of TM complexes. The accuracy of xTB calculations has not been tested for Mn complexes. To investigate the accuracy of xTB calculations with respect to DFT based results, we computed CO stretching frequencies and DG HX for selected complexes. To further compare the performance of xTB and DFT, we analyzed the computed CO stretching frequencies (ν(CO)) for the carbonyl moieties in the Mn pincers. A comparison of the results obtained with the DFT and xTB methods for a representative case of Mn-PCP complex is presented in Figure 4. Both methodologies reveal a similar trend in computed ν(CO). For a given R 1 , electron donating R 2 groups give rise to lower ν(CO). The plots also show that the electron donating effect of R 1 functionalziation in this case is more important that that of the R 2 , because of the major role of the electronic effects at the metal center on the coordinated CO ligands. Furthermore, we observed a good agreement between xTB and DFT results when all systems with CF 3 functionalization are excluded from the dataset (R 2 = 0.85). xTB calculations seem to overestimate the CO stretching for CF 3 functionalized ligands for all pincer complexes considered in this study (see SI). Nevertheless, the comparison of DG HX and ν(CO) parameters point to a qualitative agreemenet between the results obtained with the xTB and DFT methods.

Scaling Relations and competitive adduct formation
The activation of HÀ X bonds is assumed to proceed via a metalligand cooperative heterolytic cleavage over the MÀ L site in all of the pincer complexes discussed here. We therefore, expect similar trends in Gibbs free energy for addition of HÀ X species, with differences arising from the nature of MÀ X bonding. Such similarities practically manifest themselves in scaling or linear free energy relationships between different substrates. Such relationships imply that having computed the DG HX 1 for a substrate HX 1 , one can estimate the DG HX i for all other species that follow a linear scaling relation with HX 1 . Figure 5 shows a significant correlation between the DFT-computed DG HBr and DG H2 using DFT (R 2 = 0.91), especially for PCP and CNC complexes. Therefore, the Gibbs free energy of the bromide adduct formation can be used to estimate the relative energy of formation of the hydride species.
In our experience metal hydride complexes are not described well using xTB. [73] Often the MÀ H bond was found to be very elongated (> 2.8 Å) (see SI) in the structures obtained by the xTB-based geometry optimizations. However, Br complexes are described well and reasonable geometries were obtained with both xTB and DFT optimizations. Therefore, the scaling relationship that we observed between the bromide and hydride adduct becomes particularly practical because the DG HBr computed using xTB can be directly used to estimate the stability of the active hydride intermediates in screening studies. Figure 5 also shows that while both PCP and PNN complexes react with H 2 in an exergonic manner, the reaction is endergonic with MnÀ CNC complexes. DFT calculations also revealed linear scaling relations among hydride, iso-propoxide, bromide and hydroxide adducts. The correlation of DG H 2 with OH and i-PrO adduct was found to be rather weak (R 2 = 0.53 and 0.55) (see SI). We attribute this weaker correlation to differences in M-X bonds formed upon addition of HÀ X (MÀ OR vs. MÀ H) (vide infra). DG H 2 O and DG iPrOH were found to have a moderate correlation with R 2 = 0.71. We also investigated the correlation coefficient between the xTB-computed DG HX values. The resulting correlation matrix is shown in Figure 6. We observe that Gibbs free energies for addition of chemically similar species have higher correlation coefficient. For example, all alkoxides correlate well among each other but have relatively poor correlation with hydrides and bromides. Therefore, isopropoxide and ethoxide have the strongest correlation followed by the correlation between methoxide and ethoxide. The weakest correlation is between the Mn-alkoxides and the catalyst precursor (MnÀ Br).

Adduct stability modulations via ligand modifications
Next we investigated the impact of the functionalization of donor site vs backbone on DG HX . For this purpose we examined DG HBr in two scenarios: 1) R 2 = H with R 1 varied to investigate the impact of functionalization at the donor site 2) R 1 = H with R 2 varied to investigate the impact of functionalization on the ligand backbone. Functionalizations on the ligand either on donor or acceptor site can have a stabilizing or a destabilizing   effect in two ways: 1) via (de)stabilization of the pristine catalyst 2) Electronic and/or steric (de)stabilization of the adduct moieties. The pristine complex can be (de)stabilized via geometric distortions and the resulting strain in the ligand scaffold introduced by functionalization. The adduct moiety such as an alkoxide can be destabilized by increased electronic density at the metal center. On the other hand, the LÀ H bond in the adducts can be favorably stabilized by an increased electronic density at the ligand. Larger alkyl groups such as t-Bu/cy can also destabilize bulky alkoxides by steric repulsion. An accurate consideration of electronic effects is necessary to examine the impact of functionalization. Therefore we examined the impact of functionalization of DG HX using DFT calculations.  Table S5. For both PCP and PNN complexes, electron donating groups at R 1 destabilize the adduct leading to higher (more positive) DG HX . In contrast to PCP and PNN complexes, the CNC complexes exhibit a different trend. The electron withdrawing CF 3 groups also destabilize Br and OH adducts. A detailed analysis of this divergent trend is beyond the scope of this paper. We speculate that these differences are related to the nature of MnÀ C coordination in the CNC complexes. Furthermore, the R 1 functionalizations in this case are not performed on the C moiety coordinating the Mn center unlike for the other complexes where the functionalized P/N/S are directly bound with the metal site.

Functionalization on ligand backbone
Primarily two factors control the impact of functionalization on the ligand backbone: 1) electronic effect which influences the basicity of the acceptor site and 2) strain in the geometry upon functionalization. Electron releasing groups should increase the basicity of the acceptor site and, therefore, are expected to have a stabilizing effect. At the same, since electron releasing groups such as a t-Bu are bulkier they are also expected to introduce a high amount of strain in the coordinated ligand geometry. R 1 = i-Pr resulted in 6 CPC and PNN, and 5 PCP complexes with Br coordinated to the Mn. The resulting DG HBr are presented in Table S7.
Variation of R 2 seems to have minimal impact on the PCP scaffold where DG HBr shows little variation with the R 2 = H being the most stable complex, and R 2 = CF 3 being the least stable. This behavior is expected since the R 2 site is located further away from the proton acceptor site on the ligand. For both Mn-PNN and MnÀ CNC complexes, R 2 = cy results in most favorable adduct formation. R 2 = CF 3 leads to most destabilized adduct for the MnÀ CNC, whereas R 2 = t-Bu is most destabilizing functionalization for Mn-PNN. The additional stabilization of cy substitution is in contrast with the destabilization introduced by other electron donating groups namely Ph, i-Pr and t-Bu. The Hamett constants of cy and i-Pr are À 0.05 and À 0.04, respectively, indicating similar electron releasing behavior via inductive effect. [74] We attribute the observed destabilizing effect of Ph, i-Pr and t-Bu groups, despite their electron donating nature, to the geometric strain. Analysis of geometric strain showed that R 2 = cy scaffold had an effective stabilization of 4.15 kcal mol À 1 relative to R 2 = H. In contrast while R 2 = i-Pr suffered a destabilization of 1.37 kcal mol À 1 (Table S5-S6). Therefore both steric and inductive effects play an important role in stabilization of the Br adducts.
Heatmap plots for the functionalization on donor and ligand backbone sites respectively represent the average impact of the choice of the functionalization groups ( Figure S9). Both donor and backbone site functionalization leads to a great spread in average Gibbs free energy of formation, which can vary by up to 20 kcal mol À 1 between different functionalizations. This reflects the degree of tunability of the catalytic properties that these ligand scaffolds can offer. Especially interesting are the qualitative changes in the average stability, where for example the substitution of CF 3 on the backbone site can turn the adduct formation from exergonic to endergonic. The relative range and standard deviation does not change significantly within the data sets.

Catalytic adduct species
With regards to the catalytic cycle, different adduct stabilities that play a significant role in activity and efficiency are examined. The heatmaps of the average formation Gibbs free energies show the difference between the investigated adducts ( Figure S9). The results indicate that the bromide adduct sustains substantial stability upon functionalization, especially for CNC complexes. The stabilities of the other adducts with similar structure such as MeO and EtO, or t-BuO and i-PrO are quite similar.
Boxplots comparing the difference formation Gibbs free energies for the catalysts with different ligands are shown in Figure 7 and 8 allow to analyze the trends in more detail. One interesting aspect in (de)hydrogenation chemistry is the choice of the base activator/promotor for the catalytic reaction. An enhanced stability of a complex formed with the base can result in inhibition and the catalyst can remain in a resting state. Here, we considered a model (de)hydrogenation of i-PrOH reaction as an example.
In Figure 7 and 8, the stability of isopropxide to the adduct of two potential bases (KOH and KOtBu, respectively) is compared. The results show that a competition between the base (KOtBu) and i-PrOH for complexation with the Mn center is highly likely. However, the i-PrO adduct was generally found to be more stable than the OH adduct for all Mn-pincers studied here. This concludes that KOH may represent a better choice for the systems where i-PrOH dehydrogenation is important. Consistent with our observation, Beller and co-workers observed that switching from KOH to KOtBu lead to catalyst deactivation in dehydrogenation of methanol catalyzed by a Mn-PNP pincer complex. [54]

Conclusions
In this investigation, we carried out a computational analysis of the effect of functionalization on different Mn-adducts of five types of pincer catalysts. Here, a data augmented approach was employed using fast xTB optimizations to analyze stabilities of metal adducts that can play a significant role in the catalytic cycle. The xTB results were compared to DFT calculations, which showed qualitative agreement and helped identifying the accuracy boundary of the accelerated xTB methodologies for studying Mn(I)-pincer complexes. We identified linear scaling relation between Gibbs free energy for formation of different adducts, which can be used for rapid screening purposes.
Functionalization of the donor site directly affects the metal center activation, as illustrated by the changes in the computed CO stretching frequencies. Increased electronic density at the metal center and geometric strain both have destabilizing effects on the formation of alkoxide, hydroxide, bromide and hydride adducts. Comparison of relative stabilities of the isopropoxide adduct with hydroxide and tert-butoxide adducts showed that KOtBu can have a poisoning effect during i-PrOH dehydrogenation, and that KOH would be a more suitable base.
In the outlook, this work is a first step in mechanism based high throughput screening of pincer ligand based catalysts. Development of data-augmented approaches to screen and design highly active homogeneous catalysts is an ongoing effort in our laboratory.

Author contributions
A. K. carried out calculations and analysed the results. V. S. conceived and supervised the project. A. V. K. generated functionalized geometries of pincer complexes, coordinated in carrying out xTB and DFT calculations and compiling datasets from calculations. E. A. P. played a supervisory role and directed the project. All the authors discussed the results and wrote the manuscript.  Keywords: data-driven catalysis · linear scaling relations · catalyst design · (de)hydrogenation catalysis · catalyst deactivation