Otega A.
Ejegbavwo‡
a,
Anna A.
Berseneva‡
a,
Corey R.
Martin
a,
Gabrielle A.
Leith
a,
Shubham
Pandey
d,
Amy J.
Brandt
a,
Kyoung Chul
Park
a,
Abhijai
Mathur
a,
Sharfa
Farzandh
a,
Vladislav V.
Klepov
a,
Brittany J.
Heiser
a,
Mvs
Chandrashekhar
b,
Stavros G.
Karakalos
c,
Mark D.
Smith
a,
Simon R.
Phillpot
d,
Sophya
Garashchuk
a,
Donna A.
Chen
*a and
Natalia B.
Shustova
*a
aDepartment of Chemistry and Biochemistry, University of South Carolina, Columbia, South Carolina 29208, USA. E-mail: shustova@sc.edu
bDepartment of Electrical Engineering, University of South Carolina, Columbia, South Carolina 29208, USA
cCollege of Engineering and Computing, University of South Carolina, Columbia, South Carolina 29208, USA
dDepartment of Materials Science and Engineering, University of Florida, Gainesville, Florida 32611, USA
First published on 27th June 2020
Metal node engineering in combination with modularity, topological diversity, and porosity of metal–organic frameworks (MOFs) could advance energy and optoelectronic sectors. In this study, we focus on MOFs with multinuclear heterometallic nodes for establishing metal−property trends, i.e., connecting atomic scale changes with macroscopic material properties by utilization of inductively coupled plasma mass spectrometry, conductivity measurements, X-ray photoelectron and diffuse reflectance spectroscopies, and density functional theory calculations. The results of Bader charge analysis and studies employing the Voronoi–Dirichlet partition of crystal structures are also presented. As an example of frameworks with different nodal arrangements, we have chosen MOFs with mononuclear, binuclear, and pentanuclear nodes, primarily consisting of first-row transition metals, that are incorporated in HHTP-, BTC-, and NIP-systems, respectively (HHTP3− = triphenylene-2,3,6,7,10,11-hexaone; BTC3− = 1,3,5-benzenetricarboxylate; and NIP2− = 5-nitroisophthalate). Through probing framework electronic profiles, we demonstrate structure–property relationships, and also highlight the necessity for both comprehensive analysis of trends in metal properties, and novel avenues for preparation of heterometallic multinuclear isoreticular structures, which are critical components for on-demand tailoring of properties in heterometallic systems.
Herein, we utilize the versatility of metal node nuclearity to establish possible metal–property trends for frameworks containing mononuclear, binuclear, and pentanuclear heterometallic nodes (Scheme 1). We demonstrate changes in the electronic profile as a function of integration of a second metal. Furthermore, we probe changes in the electronic structure as a function of metal ensemble size (i.e., number of metal ions in the metal node), metal nature, and metal ratio (in the example of three series). With support from theoretical modeling, we demonstrate that the experimentally studied changes in the density of states (DOS) near the Fermi edge, distribution of the charge on the metal, as well as band gap and conductivity values correlate with each other and are governed by the nature of the second integrated metal.
Scheme 1 Schematic representation of the studied heterometallic MOFs: (top) mononuclear M3−XM′X(HHTP)2, (middle) binuclear M3−XM′X(BTC)2, and (bottom) pentanuclear M5−XM′X(NIP)4 systems. |
Fig. 1 (Top) Crystal structures of: (left) mononuclear heterometallic Cu3−XCoX(HHTP)228 and (right) pentanuclear heterometallic Cu5−XMnX(NIP)4 MOFs.29 (bottom) Crystal structures of binuclear monometallic Cu3(BTC)2 and binuclear heterometallic Cu3−XMX-(BTC)2 (M = Mn, Fe, and Co) MOFs. Insets show photographs of the MOF single crystals. The light blue, dark blue, green, orange, grey, and red spheres represent Cu, Co, Mn, Fe, C, and O atoms, respectively. H atoms were omitted for clarity. |
Comprehensive MOF analysis was performed using single-crystal X-ray diffraction, powder X-ray diffraction (PXRD), X-ray photoelectron spectroscopy (XPS), inductively coupled plasma mass spectrometry (ICP-MS), conductivity measurements, diffuse reflectance (DR) spectroscopy, thermogravimetric analysis (TGA), and density functional theory (DFT) studies. All prepared heterometallic MOFs were analyzed by PXRD to ensure crystallinity before and after transmetallation. The metal ratio was verified by ICP-MS analysis. Notably, all ICP-MS studies were performed on samples that underwent an extensive washing (∼one week) procedure using a Soxhlet apparatus to remove any residual M′-salts utilized for the integration of a second metal (M′). The discussion in this paper will be organized in the following order: preparation and characterization of the monometallic and corresponding heterometallic frameworks, then comprehensive analysis based on XPS, DR spectroscopy, and conductivity measurements with the support of theoretical modeling. The main emphasis of the presented studies is to reveal possible relationships between the observed experimental and theoretical values as a function of the chosen metal M′, i.e., establishing M′–property trends.
Heterometallic MOF | Synthesis T (°C)/time (h) | Evacuation T (°C)/time (h) |
---|---|---|
Cu2.0Mn1.0-HHTP | 85/16 | 85/6 |
Cu2.5Co0.5-HHTP | 85/16 | 85/6 |
Cu1.5Ni1.5-HHTP | 85/16 | 85/6 |
Cu2.6Rh0.4-HHTP | 85/16 | 85/6 |
Cu2.8Mn0.2-BTC | 90/24 | 160/24 |
Cu2.6Mn0.4-BTC | 90/48 | 160/24 |
Cu2.4Mn0.6-BTC | 90/72 | 160/24 |
Cu2.7Fe0.3-BTC | 90/24 | 160/24 |
Cu2.6Fe0.4-BTC | 90/48 | 160/24 |
Cu2.2Fe0.8-BTC | 90/72 | 160/24 |
Cu2.9Co0.1-BTC | 90/12 | 160/24 |
Cu2.82Co0.18-BTC | 90/42 | 160/24 |
Cu2.79Co0.21-BTC | 90/72 | 160/24 |
Cu2.7Ni0.3-BTC | 90/74 | 160/24 |
Cu1.6Zn1.4-BTC | 25/24 | 160/24 |
Cu4.8Mn0.2-NIP | 25/3.5 | 85/12 |
Cu4.4Fe0.6-NIP | 25/1 | 85/12 |
Cu4.8Rh0.2-NIP | 60/5 | 85/12 |
Despite the fact that a typical MOF transmetallation procedure results in polycrystalline samples, we were able to preserve single crystals of BTC-based frameworks containing Cu/Fe, Cu/Mn, and Cu/Co pairs. The crystal structures and crystallographic data for the heterometallic Cu2.4Fe0.6-BTC, Cu1.8Fe1.2-BTC, Cu2.4Mn0.6-BTC, Cu2.3Mn0.7-BTC, Cu1.9Co1.1-BTC, and Cu1.1Co1.9-BTC MOFs are shown in Fig. 1 and Tables S1 and S2,† highlighting the isoreticular nature of the monometallic and heterometallic analogues. For the synthesis of the zinc-containing Cu3−XZnX(BTC)2 system, a different parent scaffold, Zn3(BTC)2, was chosen due to unsuccessful attempts to integrate zinc in the copper-containing monometallic framework, Cu3(BTC)2. Thus, to prepare CumZnn-BTC, (m = 1.6, n = 1.4), we soaked Zn3(BTC)2 in a 1.01 M ethanol solution of Cu(NO3)2 at room temperature for 24 hours (Table 1). All BTC-based samples were analyzed by PXRD to ensure crystallinity before and after transmetallation (Fig. S7–S12†). Thermostability of the Cu3−XM′X(BTC)2 samples was determined by TGA and the corresponding TGA plots are shown in Fig. S13–S15.†
Evaluation of M⋯M Interactions by the Voronoi–Dirichlet Approach. Valence Band Structure, Density of States, Conductivity Measurements, Metal Oxidation States, and Optical Data Analysis in Combination with Theoretical Modeling.
To probe metal⋯metal interactions in the discussed mononuclear, binuclear, and pentanuclear systems, we employed the Voronoi–Dirichlet tessellation approach.32,33 In general, a Voronoi–Dirichlet polyhedron (VDP) for a selected atom in the crystal structure is shaped by an intersection of the planes dissecting the center of the lines that connect the selected atom with all surrounding atoms in the structure and are perpendicular to these lines. As a result, every inner point of a VDP is closer to the selected atom than to any other atom in the structure. This approach allows for estimation of interaction strength between two atoms, for instance, metals in the nodes, by calculating the solid angle (Ω) of a shared face of their VDP, expressed as percent of total VDP surface area as shown in eqn (1).32
Ω = S/Stotal × 100% | (1) |
Scheme 2 (Left) VDP of the cobalt atom in the crystal structure of mononuclear Co9(HHTP)4,28 (middle) VDP of the copper atoms in the crystal structures of binuclear Cu3(BTC)2,31 and (right) pentanuclear Cu5(NIP)4.29 The light blue, dark blue, and red spheres represent Cu, Co, and O atoms, respectively. The contact atoms except oxygen for the pentanuclear SBU were omitted for clarity. Gray arrows indicate a shared face of VDP with the area S, while Stotal stands for the total area of the VDP. |
For performance of the VDP analysis, we need to have access to single-crystal X-ray data, and thus, we chose Co9(HHTP)4 as an example of the mononuclear system.28 In Co9(HHTP)4, the cobalt atoms do not share a common VDP face, indicating no interactions between the metal atoms that is also supported by the distance between the metal atoms of 4.96 Å (Scheme 2). In Cu3(BTC)2 with binuclear nodes, the shortest distance between the metal atoms is 2.63 Å, and a corresponding solid angle was found to be 9.46% (Scheme 2). For comparison, an atom in an idealized octahedral environment has six VDP faces with a solid angle of 16.7% for each bond. In a pentanuclear Cu5(NIP)4 node, the M⋯M distances vary in a range of 3.20–3.50 Å (Table S5†).29 There are two unique non-zero contacts in the copper-based node with Ω of 2.60% per contact, giving a total value of 5.20% for the central Cu atom. Despite higher metal node nuclearity observed in Cu5(NIP)4, estimated Ω for Cu5(NIP)4 is almost twice as small as that found in Cu3(BTC)2. Notably, the constructed polyhedra were built taking all atoms in the second and third coordination spheres into account, and only metal nodes are shown in Scheme 2 for clarity. According to the VDP analysis, M⋯M interactions are not simply a function of metal node nuclearity, and therefore, a more in-depth crystallographic analysis is required for each system. BTC-MOFs could be used as a platform for understanding M–property correlations due to the pronounced M⋯M interactions.
To evaluate a large number of monometallic and heterometallic systems, we employed XPS as a powerful and non-destructive tool for fast prescreening of the changes in the valence band (VB) region. We simultaneously monitored the DOS near the Fermi level (EF, binding energy = 0 eV) and changes in the oxidation states of metals integrated into the MOF lattice. Prior to experimental analysis by XPS, all MOF samples were evacuated using the procedures based on the TGA results for the corresponding frameworks (Table 1, Fig. S4, S13–S15, S19, and S20†). The results acquired from XPS studies were compared against those obtained from DR analysis, conductivity measurements, and theoretical modeling. We initially began with M3−XM′X(BTC)2 due to a wider compositional range and diversity of metals available for integration inside the lattice without degradation of the parent framework.
Fig. 3 (a) A binuclear paddle-wheel metal node and graphical illustration of the results of conductivity measurements obtained for Cu3−XM′X(BTC)2 (M′ = Mn, Fe, and Co) as a function of M′ percentage. (b) Changes in conductivity (|Δσ|, dark blue triangles), experimentally measured band gaps (ΔEg(exp), red circles), calculated band gaps (ΔEg(calc), orange pentagons), estimated valence band onset values from the XPS data (ΔE′, black squares), and calculated (zCu × XCu) values (green pentagons) as a function of M′ performed for M3−XM′X(BTC)2 (M′ = Co, Ni, Mn, Fe, and Zn). The ΔEg, ΔE′, and |Δσ| values have been normalized to the mole fraction of M′ (XM′). The corresponding graphs with error bars are shown in Fig. S22.† (c) Crystal structure of parent Cu3(BTC)2 possessing the tbo topology (shown in inset). The red, gray, and light blue spheres represent O, C, and Cu atoms, respectively. H atoms were omitted for clarity. |
The XPS studies not only allowed us to estimate (E′* − E′)/XM′ values, but also shed light on the oxidation states of the incorporated metals (M′). As previously shown for the monometallic Cu3(BTC)2 system, the Cu(2p3/2) region of the XPS spectrum consists of two peaks at 933.0 and 934.7 eV that can be assigned to Cu1+ and Cu2+, respectively.36 For the heterometallic Cu3−XM′X(BTC)2 MOFs, a similar trend was observed, and the presence of Cu1+ and Cu2+ peaks was also detected (Fig. S27†). Analysis of the corresponding XPS regions for incorporated M′ allowed us to conclude that M′ inside Cu3−XM′X(BTC)2 possesses the following oxidation states: +2 (Co); +2 (Ni); +2 (Mn); +2 and +3 (Fe); and +2 (Zn). Based on the XPS data, we attempted to estimate how (zCu × XCu) and (zM′ × XM′) changes as a function of M′ with the assumption that the total charge of cations remains constant (eqn (2)).
zM′ × XM′ + zCu × XCu = constant | (2) |
We estimated the average charge on the copper ions by peak fitting the Cu(2p3/2) XPS data with contributions from Cu1+ and Cu2+ (Fig. S27†). For instance, if the ratio of Cu1+ to Cu2+ is 0.5 to 0.5 then zCu = 0.5 × (1+) + 0.5 × (2+) = (1.5+), where 1+ and 2+ are the charges on copper. The mole fractions of XM′ and XCu were estimated from the ICP-MS data. To find the constant from eqn (2), we used XPS data for monometallic Cu3(BTC)2. In this case, (zM′ × XM′) equals zero because of the absence of a second metal, M′, in the Cu3(BTC)2 structure. Therefore, zCu × XCu + zM′ × XM′ = zCu × (1) + zM′ × (0) = zCu. The constant in eqn (2) was estimated to be 1.69. Finally, the zM′ value was also calculated based on eqn (2) since zCu and XCu (or XM′) was estimated from the XPS and ICP-MS data, respectively. The corresponding values of (zCu × XCu)/(zM′ × XM′) for heterometallic BTC-samples with an integrated metal (M′) were found to be 1.44/0.25 (Co), 1.35/0.34 (Ni), 1.17/0.52 (Mn), 1.10/0.59 (Fe), and 0.89/0.79 (Zn, Table 2 and Fig. S27†). The calculated zCu × XCu follows the trends established for the experimental optical band gap and conductivity values (Fig. 3b). Thus, increase in the copper charge and its mole fraction correlates with the corresponding optical band gap decrease.
M′= | Co | Ni | Mn | Fe | Zn |
---|---|---|---|---|---|
a Samples with the maximum M′/Cu ratio were chosen for analysis. | |||||
Cu 3−X M′X(BTC)2 | |||||
z Cu × XCu | 1.44 | 1.35 | 1.17 | 1.10 | 0.89 |
z M′ × XM′ | 0.25 | 0.34 | 0.52 | 0.59 | 0.79 |
E′, eV | 0.29 ± 0.02 | 1.72 ± 0.12 | 1.66 ± 0.17 | 1.76 ± 0.14 | 1.50 ± 0.09 |
σ (×1011), S × cm−1 | 396.00 ± 0.19 | 101.00 ± 0.05 | 62.40 ± 0.08 | 36.30 ± 0.03 | 31.50 ± 0.01 |
E g(exp), eV | 3.22 ± 0.17 | 3.32 ± 0.13 | 3.30 ± 0.14 | 3.24 ± 0.10 | 3.27 ± 0.09 |
E g(calc), eV | 3.32 | 3.50 | 3.65 | 3.70 | 3.90 |
Cu 3−X M′X(HHTP)2 | |||||
z Cu × XCu | 1.19 | 0.77 | 1.01 | — | — |
z M′ × XM′ | 0.33 | 0.75 | 0.51 | — | — |
E′, eV | 1.20 ± 0.07 | 1.38 ± 0.08 | 1.11 ± 0.06 | — | — |
σ (× 107), S × cm−1 | 205.00 ± 1.36 | 0.99 ± 0.01 | 87.70 ± 0.88 | — | — |
E g, eV | 1.06 ± 0.01 | 1.20 ± 0.01 | 1.17 ± 0.01 | — | — |
To rationalize the observed experimental trends, we analyzed the electronic structure computed using the Vienna ab initio simulation package (VASP)37,38 with the plane wave basis set. The total and partial DOS were obtained from the single point calculations at experimental geometries using the hybrid HSE06 method39 followed by geometry optimization (see the ESI†). The results revealed that substitution of one of the two metal centers in the metal node of the MOF truncated model, Cu2(OAc)4 (Fig. S34†), resulted in an increase of the band gap in the order Co < Ni < Mn < Fe < Zn (Table 2) that is in agreement with the Eg values estimated from the Tauc plot analysis (Fig. S26†). Calculated ΔEg(calc)/XM′ also follows the experimental trend shown in Fig. 3b. The partial-DOS analysis suggests that the decrease in the band gap is associated with changes in the electronic structure near the Fermi level. In the case of M′ = Co, Ni, Mn, and Fe, the highest occupied molecular orbital (HOMO) is dominated by M′-3d-orbitals after substitution, in contrast to parent monometallic Cu2(OAc)4 where the HOMO is occupied by the O-2p-orbitals (Fig. 4a–e and S35†). At the same time, the lowest unoccupied molecular orbital (LUMO) is dominated by the Cu-3d-orbitals in the case of monometallic and heterometallic clusters. Integration of zinc inside the copper paddle-wheel node, according to theoretical calculations, does not significantly alter the electronic structure, and the band gap edges remain the same (Fig. 4f). The Zn-3d-orbitals lie deep inside the occupied orbitals and the band edges are still dominated by O-2p- and Cu-3d-orbitals that represent the HOMO and LUMO, respectively. These results indicate that cobalt substitution decreases the band gap of Cu-MOF the most, followed by band gaps for Ni < Mn < Fe; while zinc integration has almost no effect on the band gap.
The results of the performed Bader charge, atomic-dipole-corrected-Hishfeld-atomic charge, and Mulliken-charge analysis based on the B3LYP-D3/m6-31G* and ωB97X-V/6-31G* methods using the optimized geometry for the CuM′(OBn)4 (OBn− = benzoate; M′ = Co, Ni, Mn, Fe, and Zn) cluster are given in Table S13 and described in the ESI.†
As a next step in our analysis, we compared the observed trends for heterometallic MOFs with those known for doped inorganic oxides, which exhibit the electronic property tunability that has been studied for several decades.40,41 The challenge in the literature search was mainly associated with the typically narrow range of metals traditionally used as dopants for one set of studies. However, we found that Deepak and co-workers reported tuning of electronic properties of ZnO (a wurtzite-type structure) by doping with 3d divalent metals such as M′ = Co, Ni, and Mn.40 It was found that an increase in dopant concentration caused a decrease in the ZnO band gap values (Eg(ZnO) = 3.30 eV).40 Indeed, the reported Eg values of zinc oxide doped with Co, Ni, and Mn were found to be 2.95, 3.24, and 3.28 eV, respectively, for a substitution percentage of M′ at 5% (Fig. 5). Analysis of electronic properties revealed that a decrease in the band gap in the case of the Co dopant is the highest among the three systems, followed by Ni and Mn incorporated samples (Fig. 5). Such a behavior was attributed to the sp–d exchange interactions between electrons in conduction and valence bands (that are mostly s and p electrons) and dopant localized d electrons.42 In line with this trend, Lin and co-workers reported a theoretical study of the doped anatase phase of TiO2 with the same transition metals, M′ = Mn, Co, and Ni.41Ab initio band calculations based on DFT with the plane wave basis set were performed on the supercell of the anatase structure with a substitution percentage of M′ at 12.5%. The trend for Co, Ni, and Mn metals obtained in this study is the following: Eg (1.78 eV for Co:TiO2) < Eg (2.23 eV for Ni:TiO2) < Eg (2.32 eV for Mn:TiO2, Fig. 5). It has been demonstrated that the dopant energy levels occur in the middle of the band gap (at an “intermediate level”), leading to band gap narrowing.41 While TiO2 valence and conduction bands are dominated by O-2p and Ti-3d states, respectively, valence and conduction bands are still formed by O-2p and Ti-3d states modified by the dopant metal. On the example of these transition-metal doped oxides, we demonstrate that the trend established for ΔEg(lit.)/XM′ is in line with the trends found in our studies for experimental and calculated ΔEg(Cu3−XM′X(BTC)2)/XM′ (M′ = Co, Ni, and Mn, Fig. 5). Access to crystallographic data of heterometallic MOFs such as Cu3−XM′X(BTC)2 (M′ = Co, Fe, and Mn, Fig. 3c) allowed us to evaluate the dependence of a unit cell parameter, a, (Cu3−XM′X(BTC)2 belongs to the Fmm space group) as a function of the integrated metal, and therefore survey possible structural distortion. Maximum deviation in the unit cell parameter, a, in comparison with that of pristine Cu3(BTC)2 was found to be 0.09% for Cu1.8Fe1.2-BTC while for the rest of the BTC-systems Δa/a* varied in a range of 0.007% to 0.06% (Tables S3 and S4†). Notably, the distance comparison was performed on crystal structures with several M/M′ pairs (M = Cu, M′ = Co; M = Cu, M′ = Fe; and M = Cu, M′ = Mn), collected at the same temperature, 100 K. The evaluation of possible changes in Cu⋯M′ metal distances demonstrated that the largest change (1.02%) was observed for Cu2.4Mn0.6-BTC. The largest change in distances between metal nodes (0.09%) was observed for Cu1.8Fe1.2-BTC (Tables S3 and S4†). We also evaluated structural changes by calculating Δa/XM′ values. Since we have two crystal structures per metal composition, we estimated Δa/XM′ = [(a* − a1)/XM′1+ (a* − a2)/XM′2] × 0.5 (a1 and a2 = unit cell parameters of two heterometallic structures; a* = the unit cell parameter of the Cu3(BTC)2 structure; XM′1 and XM′2 = mole fraction of incorporated M′ in the particular structure; for more details see ESI†). In summary, there are no significant structural deviations to establish a correlation between Cu⋯M′ metal distances, metal node distances, or unit cell parameters, and the estimated Δa/XM′ values do not follow the trend based on conductivity, VB edge, and optical data of Cu3−XM′X(BTC)2 MOFs as shown in Fig. 3.
Fig. 5 Band gaps: measured ΔEg(exp) (red circles) and calculated ΔEg(calc) (orange pentagons) for Cu3−XM′X(BTC)2 MOFs. Literature data: measured ΔEg(lit1) for M′5%:ZnO (dark blue triangles)40 and calculated ΔEg(lit2) for M′12.5%:TiO2 (black squares).41 ΔEg values have been normalized to the M′ metal mole fraction (M′ = Co, Ni, and Mn). |
As a part of our studies, we surveyed the electronic structure changes in heterometallic MM′-MOFs with the same M and M′ but with a different M to M′ ratio (Fig. 3a). The choice of metal ratios was a balance between incorporation of the highest percentage of the second metal, M′, and preservation of framework integrity. According to conductivity measurements, the largest difference in electronic properties within the same composition was observed for Cu3−XCoX(BTC)2 systems. Indeed, changes from 2% to 7% of incorporated cobalt according to ICP-MS analysis resulted in a five-fold conductivity enhancement (Fig. 3a and Table S10†). The statistical difference between measured conductivity values was probed by employment of a variance test (ANOVA) integrated in the MATLAB package.43–45 As a result, it was demonstrated that the conductivity values of the Cu3−XCoX(BTC)2 samples with different cobalt percentages are indeed statistically different. In the other heterometallic systems with the same M/M′ pairs, the experimentally observed changes as a function of M to M′ ratio were less pronounced in comparison with those observed for Cu3−XCoX(BTC)2. For instance, changes in iron percentage from 9 to 25% in Cu3−XFeX(BTC)2 do not lead to significant changes of electronic properties as shown by conductivity measurements (Fig. 3). Indeed, for Cu3−XFeX(BTC)2, the estimated p-value was greater than 0.05, and thus, the null hypothesis, that measured conductivity values are the same, could not be rejected. At the same time, the one-way analysis of variance performed for the Cu3−XMnX(BTC)2 system demonstrates that the measured conductivity values are statistically different.
However, in contrast to BTC-frameworks, conductivity of heterometallic HHTP-MOFs decreases upon incorporation of a second metal in comparison with that of their monometallic analogues. We estimated the conductivity values as (2.10 ± 0.01) × 10−5 S × cm−1 for Cu2.5Co0.5-HHTP, (8.80 ± 0.09) × 10−6 S cm−1 for Cu2.0Mn1.0-HHTP, (9.90 ± 0.06) × 10−8 S cm−1 for Cu1.5Ni1.5-HHTP, and the lowest value was found to be (8.60 ± 0.02) × 10−9 S cm−1 for the Cu2.6Rh0.4-HHTP framework (Tables 2 and S10†); while conductivity measured under the same conditions for the parent Cu3(HHTP)2 framework was found to be (4.90 ± 0.02) × 10−4 S × cm−1 (Fig. 6a). Previous literature reports for similar 2D frameworks are in line with our studies.28,47 Thus, it was shown through theoretical modeling that the nickel-to-copper transmetallation procedure in M-HITP systems (HITP3− = 2,3,6,7,10,11-hexaaminotriphenylenesemiquinonate) possessing the same AAAA packing motif can result in changes of the framework electronic behavior from semiconducting to metallic due to different coordination environments adopted by nickel versus copper that likely leads to packing distortion.48 As we previously mentioned, Cu3(HHTP)2 possesses AAAA packing while layers of Co9(HHTP)4 alternate in the ABAB sequence.28 While the A layer in both frameworks consists of the M3(HHTP)2 two-dimensional honeycomb structure, the B layer in the case of Co9(HHTP)4 is formed by discrete Co3(HHTP) units resulting in a Co6(HHTP)2 layer (Fig. 6c).28 Therefore, we speculate that changes in electronic behavior of heterometallic HHTP-systems may be indicative of a structural distortion of the 2D sheets due to distinct coordination environments adopted by M and M′. PXRD analysis demonstrated that all Cu3−XM′X(HHTP)2 possess AAAA stacking that allows for comparison of electronic properties of bimetallic MOFs. To probe possible M′–property correlations similar to the BTC-system, we have evaluated |Δσ|/XM′ as shown in Fig. 6b. Although the unnormalized conductivity values (σ) decrease for Co < Mn < Ni < Rh, the corrected values |Δσ|/XM′ (σ* and σ = conductivity values for Cu3(HHTP)2 and Cu3−XM′X(HHTP)2, respectively) do not demonstrate M′–conductivity dependence (Fig. 6b). The optical data (Eg, Table 2) corroborated the trend observed for conductivity values, σ. The smallest band gap among all heterometallic HHTP-systems was detected after integration of Co (1.06 eV) and an increase in Eg values was observed for M′ = Mn (1.17 eV) < Ni (1.20 eV) < Rh (1.23 eV, Table S8†). However, adjustments based on the mole fraction of integrated metal by estimation of ΔEg/XM′ (Eg and Eg* = band gap values for Cu3(HHTP)2 and Cu3−XM′X(HHTP)2, respectively) did not result in the M′–Eg correlation similar to that observed for Cu3−XM′X(BTC)2 systems (Fig. 5).
Fig. 6 (a) A mononuclear metal node and graphical illustration of the results of conductivity measurements obtained for Cu3−XM′X(HHTP)2 as a function of M′ = Co, Mn, Ni, and Rh. (b) Changes in conductivity (|Δσ|, dark blue triangles), experimentally measured band gaps (ΔEg, red circles), and estimated valence band onset values from the XPS data (ΔE′, black squares) as a function of M′ performed for Cu3−XM′X(HHTP)2 (M′ = Co, Mn, Ni, and Rh). The ΔEg, ΔE′, and |Δσ| values have been normalized to the mole fraction of M′ (XM′). The corresponding graphs with error bars are shown in Fig. S37.† (c) Crystal structure of the Co-containing HHTP system possessing the bnn topology (shown in inset).28 The red, gray, and dark blue spheres represent O, C, and Co atoms, respectively. H atoms were omitted for clarity. |
The oxidation states for incorporated M′ = Mn, Ni, and Co inside the HTTP-systems coincide with the values observed for the BTC-systems. Indeed, XPS analysis of the Mn(2p), Ni(2p), and Co(2p) regions of both systems reveals the following oxidation states +2 (Mn), +2 (Ni), and +2 (Co) (see the ESI for more details, Fig. S28–S30†). Furthermore, analysis of the Rh(3d) region indicates the presence of rhodium in the +3 oxidation state (310 eV, Fig. S33†) for the HHTP system. According to XPS studies, the highest DOS near EF was detected for Cu3−XCoX(HHTP)2 and Cu3−XMnX(HHTP)2 based on the E′ values (Tables 2 and S6†). For other HHTP-systems where M′ = Ni (E′ = 1.38 eV for Cu3−XNiX(HHTP)2, Table S6†) and Rh (E′ = 1.51 eV for Cu3−XRhX(HHTP)2, Table S6†), the DOS near the Fermi edge are less pronounced (Fig. S40†). Overall, for the Cu3−XM′X(HHTP)2 system after incorporation of the first–row transition metals, E′ values vary in the range of 1.10–1.38 (eV) while for BTC-frameworks E′ changes from 1.50 eV (Zn) to 1.76 eV (Fe) with the exception of the Co-incorporated sample (E′ = 0.29 eV, Tables 2 and S6†). The larger E′ values are consistent with conductivity values, σ (Tables 2 and S10†), which demonstrate that HHTP-frameworks are in general more conductive than the BTC-systems. Similar to BTC-frameworks, we estimated (zM′ × XM′ and zCu × XCu) changes as a function of M′ in HHTP-systems (where zM′ (zCu) = charge on the metal (copper); XM′ (XCu) = mole fraction of M′ (Cu)). Similar to calculations performed for the BTC-systems (vide supra), the constant from eqn (2) was estimated to be 1.53 from the XPS spectrum of the monometallic Cu3(HHTP)2 sample.
The corresponding values of (zCu × XCu/zM′ × XM′) for incorporated Co, Mn, and Ni were found to be 1.19/0.33, 1.01/0.51, and 0.77/0.75, respectively (Table 2 and Fig. S39†).
Fig. 7 (a) A pentanuclear metal node and graphical illustration of the results of conductivity measurements obtained for Cu5−XM′X(NIP)4 as a function of M′ = Rh, Fe, and Mn. (b) Crystal structure of parent Cu5(NIP)4, possessing the tfz-d MOF topology (shown in inset).29 The red, gray, blue, and dark blue spheres represent O, C, N, and Cu atoms, respectively. H atoms were omitted for clarity. |
Footnotes |
† Electronic supplementary information (ESI) available. CCDC 2001459–2001464. For ESI and crystallographic data in CIF or other electronic format see DOI: 10.1039/d0sc03053h |
‡ These authors contributed equally. |
This journal is © The Royal Society of Chemistry 2020 |