Homogeneous Catalyst Recycling and Separation of a Multicomponent Mixture Using Organic Solvent Nanofiltration

2.1 Materials

Ethanol (EtOH), ethyl acetate (EtOAc), and cyclohexane (Cy) are the components of the ternary mixture used in this work. The most important properties of the above-mentioned solvents are summarized in Tab. 1.

For the experiments, the commercially available DuraMem® 300 membrane from Evonik Industries was selected. The membrane consists of modified polyimide (PI), creating a solvent-resistant polymeric membrane, designed for usage in OSN applications. It performs best in polar, more specific polar aprotic solvents. The DuraMem® 300 has a molecular weight cut-off (MWCO) of 300 Da which allows 90 % of the solute with a molecular weight of 300 g mol⁻¹ or the size of a styrene oligomer with the weight of 300 Da to be retained by the membrane. This value is determined by Evonik Industries according to the rejection of styrene oligomers dissolved in toluene.

To determine the catalyst retention rate, the Lewis acid magnesium trifluoromethanesulfonate (magnesium triflate (Mg(OTf)₂)) was employed. The molecular weight of 322.44 g mol⁻¹ is slightly higher than the MWCO stated by Evonik Industries. It is also one of the less expensive triflate catalysts which still has a higher molecular weight than 300 g mol⁻¹.

The membrane was tested with the Evonik MetCell® test bench. The piping and instrumentation (P&I) scheme is given in Fig. 1. It consists of two 4'' membrane test cells which are connected in series, a gas pressure control unit to pressurize the system with nitrogen gas, a refilling pump to recycle the permeate for keeping the concentration of the test mixture constant, a gear pump to circulate the feed, as well as a feed tank.

2.2 Experimental Procedure

The Evonik MetCell® cross-flow setup consists of two test cells, which are connected in series, having an active membrane area of 51 cm² each. The feed is pressurized with gaseous nitrogen. A piston pump is used to recirculate the permeate back into the feed tank to avoid changes in the composition of the mixture during the experiments. The whole mixture is circulated through the test cells with a gear pump. The feed flow rate is about 60 L h⁻¹; according to the Evonik Industries this is sufficient to avoid concentration polarization.

Before starting the experiments, the used membranes had to be flushed with the used solvent mixture containing ethanol, cyclohexane, and ethyl acetate, to wash off production residues. This cleaning step had to be done to ensure the full contact between membrane and solvent. A further washing of the membrane with the respective solvent mixture without the dissolved catalyst had to be performed before each measurement to adjust the membranes to the new solvent and to wash off catalyst residues. The system was emptied and the solvent mixture containing the catalyst was transferred into the system, which was then filled up with the current solvent mixture.

After flushing the membrane in the test cells, the membrane was pressurized for 4 h to reach the steady state. To determine the retention, one permeate sample per test cell and one retentate sample was taken simultaneously every hour for 4 h. The values for the retention presented in this work are averages of the measured retention. To avoid compaction influences on the measurements, the experiments were carried out in order with increasing pressure, first 30 bar, then 50 bar TMP.

2.3 Analytics

To calculate the retention rate, concentrations of solutes in the permeate and the feed have to be determined. So, the retention rate R_i can be calculated with Eq. 1:

(1)

According to this formula, a higher concentration of the solute in the permeate than in the retentate results in a negative retention rate. Effects like this have already been observed ¹⁰ and lead to an enrichment of some of the components.

For the quantification of the catalyst, ¹⁹F nuclear magnetic resonance (NMR) spectroscopy was applied. To establish a measurement procedure using only NMR to quantify both the composition of the ternary mixture as well as the catalyst concentration, gas chromatography (GC) helped to verify the results of the ¹H NMR measurements.

2.3.1 Determination of the Mixture Composition Using GC

GC was applied to cross-validate the parameters chosen for the NMR analytics. The GC system was a Focus GC system (Thermo Fischer Scientific, MA, USA) with liquid autosampler, split/splitless injection, and flame ionization detector. An Agilent CP-Wax 52 CB column coated with bonded polyethylene glycol (PEG), having 50 m length, 0.20 mm inner diameter, and 0.20 μm thickness, was employed. The injector temperature was 523 K and helium (1.5 bar constant pressure) served as carrier gas; split injection of 1 µL each sample (split ratio 50:1) was done. Temperature program: initial temperature was 323 K, hold for 10 min, followed by a temperature ramp of 12 K min⁻¹ until 473 K; hold time 15 min; total run time 37.5 min. Substances were identified by retention times of the particular compounds. However, the concentration of the catalyst could not be quantified by GC.

For all samples, dodecane in toluene was selected as an internal standard. To ensure an equal concentration of the internal standard for all samples, a stock solution of 2809.9 mg dodecane in 150 mL (129.4 g) toluene was used. By means of the peak area in the gas chromatogram, along with the respective correction factor k_f,i, the mass fraction ω_i of the solvent components can be calculated, using Eq. 2 with the dimensionless peak area A_GC,i:

(2)

Each sample was prepared by adding 300 µL of the sample to 1 mL of prepared dodecane in toluene stock solution. The samples were weighed after each step.

2.3.2 Determination of the Mixture Composition Using¹H NMR Spectroscopy

To analyze the samples via NMR with the automatization procedure, a preparation of the sample is necessary. The NMR sample was prepared by transferring 400 µL of the sample into an NMR tube. As a lock reference, 100 µL of a prepared mixture, consisting of 24.15 g of deuterated DMSO from Euriso-top with 519.6 mg of 2,3,4,5,6-pentafluorophenol from abcr, was added to the sample in the tube and thoroughly mixed. The ¹H NMR analysis of all experiments was done on a Bruker AC 300 from Bruker BioSpin GmbH. The ¹H NMR sequence consists of the parameters summarized in Tab. 2.

Table 2. Parameters used for NMR analysis.

Parameter	1H NMR	DOSY NMR	19F NMR
Solvent	DMSO	DMSO	DMSO
Temperature [K]	298	298	298
Experiment	1D	1D	1D
Nucleus	1H	1H	19F
Probe	5 mm PABBO BB-1H/D	5 mm QNP 1H/13C/31P/19F	5 mm PABBO BB-1H/D
	Z-GRD Z104275/0117	Z-GRD Z8660/0001	Z-GRD Z104275/0117
Number of scans [–]	32	8	64
Receiver gain [–]	5.6	9.0	2050.0
Relaxation delay [s]	4	5	8
Pulse width [s]	9.6	10.5	8.5
Acquisition time [s]	7.9692	0.9765	0.7341
Spectrometer frequency [MHz]	299.61	600.07	281.88
Acquired size [–]	65 536	4096	65 536
Spectral size [–]	131 072	65 536	131 072

The feed and the permeate composition can be calculated using the spectra generated from ¹H NMR measurements (see an exemplary spectrum (Fig. S1) in the Supporting Information). Note that the molar percentage of the catalyst is very low compared to the molar percentage of the solvent components (EtOH, Cy, EtOAc). Therefore, it is not taken into account for the number of the solvent components, meaning the solvent is considered a ternary mixture. Regarding literature in ¹H NMR spectroscopy, the integral area of a peak is directly proportional to the number of nuclei contributing to the signal ²⁶.

By dividing the integral area of a certain peak by n_i, the number of contributing nuclei, the molar fraction x_i of the solvent components can be calculated regarding Eq. 3:

(3)

A_NMR,i is defined as the absolute value of the peak area and n_i is the number of nuclei contributing to the integral area of component i.

2.3.4 Determination of the Catalyst Concentration Using¹⁹F NMR Spectroscopy

To analyze the samples via NMR with the automatization procedure, also a sample preparation is necessary. The NMR sample was prepared in the same way as described in Sect. 2.3.3 for ¹H NMR measurements. The ¹⁹F NMR analysis of all experiments was done with the Bruker AC 300 from Bruker BioSpin GmbH. The ¹⁹F NMR sequence consists of the parameters presented in Tab. 2.

Considering the singlet at –79.11 ppm for magnesium triflate and the three signals of the standard (q, tq, tt), ranging from –173.99 to –163.34 ppm, the molar percentage of the standard and the catalyst can be calculated by Eq. 3 (see an exemplary spectrum in Fig. S2). Therefore, the absolute value of the catalysts' integral area must be divided by 6, whereas the absolute value of the standards integral area has to be divided by 5, taking into account the contributing nuclei for triflate and standard, respectively. The concentration of the catalyst c_cat can be calculated according to Eq. 4:

(4)

where c_is is defined as the concentration of the internal standard in mol L⁻¹, and V_is and V_s are the volumes of the internal standard and the sample in m³. Considering the volume contraction by solving pentafluorophenol in DMSO-d6 as negligible, V_is can be determined according to Eq. 5:

(5)

Where m_DMSO-d6 is defined as the mass of DMSO-d6 in kg and ρ_DMSO-d6 denotes the density of DMSO-d6 in kg m⁻³. Considering this, Eq. 3 is transformed into Eq. 6:

(6)

Using the concentration, the retention rate of component i, R_i in percent, can be calculated with Eq. 1.

2.4 Determination of the Hydrodynamic RadiusR_H

Applying the Einstein-Stokes equation, the hydrodynamic radius R_H of the components in the ternary mixture can be calculated with the Boltzmann constant k_B and the temperature T by Eq. 7:

(7)

By means of DOSY the diffusion coefficient D_i can be determined. With the viscosity blending number (VBN; Eq. 8, ASTM D7152), the kinematic viscosity ν_i of component i in m²s⁻¹, the kinematic viscosity of the mixture ν_mixture can be calculated with the VBN of the mixture (Eq. 9) according to Eqs. 10 and 11:

(8)

(9)

(10)

(11)

The density of the mixture ρ_mixture was on the one hand determined experimentally by weighing 1 mL of the mixture multiple times and on the other hand by calculating the density with Eq. 12. The results are in good accordance with each other.

(12)

The hydrodynamic radius of the mixture was found by multiplying the hydrodynamic radius of the components with their molar fraction in the mixture according to Eq. 13:

(13)

2.5 Determination of the Permeate Flux

The flux J through the membranes was calculated by measuring the weight of the permeate samples over a certain time span. Note that only the ternary solvent system was considered, calculating the flux. The overall flux can be found according to Eq. 14:

(14)

Where m_s is defined as the mass of the sample in g, A as the total membrane area in m², and t as the time in h. The flux of each solvent component can be calculated considering the mass fraction ω_i of the component. The mass fraction can be determined by considering the molar fraction using Eq. 15:

(15)

where x_i denotes the molar fraction of component i in percent and M_i the molecular weight of component i in g mol⁻¹.

3.1 Permeate Fluxes

The permeability changes significantly depending on the composition of the mixture used for the experiments which ranged from 2.75 × 10⁻³ to 9.435 × 10⁻² L bar⁻¹m⁻²h⁻¹. Especially the mixtures containing high amounts of Cy and EtOAc allow only little permeation. Increasing the pressure from 30 to 50 bar, TMP does increase the overall flux across the membrane. Fig. 2 displays the flux for the different mixtures as a function of the pressure and the membrane area on the righthand side (yellow triangle). Only the permeability of the mixture with a very low Cy fraction rises from 4.429 × 10⁻² to 9.435 × 10⁻² L bar⁻¹m⁻²h⁻¹ with increasing TMP.

Since the DuraMem® membrane performs best in polar, more specific polar aprotic solvents, the decrease in permeability is most likely caused by the increasing fraction of the nonpolar component, in this case Cy ²⁷. Nevertheless, the membrane performance also decreases with increasing fraction of EtOAc, which is classified as polar aprotic solvent.

In an approach to correlate the permeability to the size of the molecules in the ternary mixtures, the hydrodynamic radius of the different components was calculated using the diffusion coefficient (Fig. 3). The results for the diffusion coefficient for the different components can be found in Tab. S1. According to Eq. 7, the hydrodynamic radius was determined after averaging the diffusion coefficient for each component. Considering the separation mechanism of nanofiltration membranes is also dependent on size exclusion as well as pore diffusion, smaller molecules could pass the membrane more easily.

It was possible to get an insight into the change of the hydrodynamic radius of a certain substance depending on its molar fraction in the ternary mixture. A tendency can be seen for EtOH and Cy. In both cases, the molecules' hydrodynamic radius decreases with increasing molar fraction of the component in the mixture, respectively (Fig. 3). For EtOAc, a contrary trend can be seen due to the intermolecular influences of the molecules. The fact that EtOH follows the trend of Cy can be caused by the fact that EtOH has the tendency to form dimers in the presence of nonpolar solvents, especially if water is absent. However, this could not be verified via NMR spectroscopy so far and has to be further investigated to prove this hypothesis ^28-30. The change in the composition of the solvent mixture alters the diffusion coefficient of the different components leading to varying hydrodynamic radii of the molecules in the mixture. On the one hand, the buildup of EtOH dimers, which are observed to be formed in unipolar solvents in the absence of water, is an influencing factor and on the other hand, the change of conformation of Cy, which is caused by the energy that is induced in the mixture by mechanical stress and pressure, also enhances this effect ³¹.

Fig. 4 supports the trend seen in Fig. 3, while showing that a higher EtOAc concentration increases the overall hydrodynamic radius of all the components. It is believed that the polarity of the mixture influences the hydrodynamic radii of the components in the mixture. Consequently, changing the composition of the mixture leads to different polarities of the mixtures which influence the hydrodynamic radii. Unfortunately, the final quantification of this effect was not possible. Therefore, the hydrodynamic radii were employed, which can be easily measured and, thus, quantified, for the decision on how to compose the used mixture.

3.2 Catalyst Retention Measurement in the Ternary Mixture

The DuraMem® 300 membrane has shown promising results for the recycling of the Lewis acid magnesium trifluoromethanesulfonate from the ternary mixture. With a molecular weight of 322.44 g mol⁻¹, the catalyst is beyond the MWCO of 300 g mol⁻¹. This suggests catalyst retention rates of 90 % and higher.

Fig. 2 in fact demonstrates that the predicted minimum retention rate of 90 % is neither reached for all mixtures nor for the two TMPs. It can be observed that with increasing TMP the retention rates rise for all mixtures. This is due to compaction of the membrane which results in smaller pores, leading to a denser membrane and, thus, to higher retention.

The retention rate is significantly lower for the mixtures with high Cy fraction as demonstrated in Fig. 2. During all of the experiments, it was observed that the components (fluorinated ethylene propylene (FEP) capillaries, polytetrafluoroethylene (PTFE) connectors, silicone capillaries for the permeate collection) swell when in contact with Cy or mixtures containing high fractions of Cy. As a result, it was not possible to reuse the components which had to be connected to metal or glass counterparts because of increased outer diameters or wall thicknesses. This supposedly also happens to the PI membrane which results in larger pores, thereby reducing the retention of the catalyst molecules. Cy seems to block the membrane, so other molecules cannot permeate through the membrane which leads to a reduced overall permeability (Fig. 2).

3.3 Retention Measurement in the Ternary Mixture

The aim of the conducted experiments was to test if OSN was applicable to catalyst recovery from the presented ternary mixture. However, during the experiments, fractioning of the ternary mixture was observed which results in the retention of Cy. Therefore, the permeate consists of considerably lower amounts of Cy and higher amounts of EtOH and EtOAc.

Fig. 5 illustrates the retention rate of the different components at 30 and 50 bar TMP. The retention rate of Cy varies between 44.3 and 77.6 % which resulted in negative retention of some of the other components. Considering that Cy is the solvent in this system which should be recycled anyhow, this effect shifts the application of OSN in the context of recycling into a new perspective. The results so far suggest that both a catalyst recycling as well as recycling of the organic solvent is possible in one separation step, even though consecutive steps involving OSN have to be carried out to reach the retention rates which are achieved for the catalyst for Cy.

The approach to correlate this effect to the size of the Cy molecule was not successful, even though keeping in mind that the membrane also rejects molecules which are smaller than 300 Da but with a lower retention rate. According to Fig. 5, this becomes more difficult since EtOAc has almost the same molecular weight as Cy; however, it does not get rejected by the membrane and instead has a negative retention rate which results in the enrichment of EtOAc in the permeate. Therefore, the influence of the polarity of the mixture might play a considerable role as described in Sect. 3.1. Nevertheless, the hydrodynamic radii offer a good guidance which takes into account the molecule size as well as the polarity of the mixture.

Depending on the consistency of the mixture, EtOH is enriched in the permeate or retained in the retentate. When the initial EtOH concentration is low, like in the mixture 0.1/0.1/0.8, the negative retention rate is especially high. In this case, the retention rate varies between 4.0 and 22.4 %.