Customizing continuous chemistry and catalytic conversion for carbon–carbon cross-coupling with 3dP

1.1 The Suziki–Miyaura reaction and palladium nanoparticles

The Suzuki-Miyaura reaction, predominantly involving a boronic acid, an organohalide, and a Pd containing catalyst, is one of the most useful transformations in organic synthesis (Boström et al. 2018). For example, in the pharmaceutical industry, it has been estimated that palladium catalysts are used in one fifth of all carbon–carbon (C–C) bond forming reactions (Carey et al. 2006; Thomas 2012). Due to its extensive use, there has been an increased emphasis on the development of reusable Pd catalytic systems. One approach towards developing more sustainable Pd catalysts focuses on use of nanoparticles as a potential alternative to homogeneous systems (Balanta, Godard, and Claver 2011). There are two different hypotheses concerning the mechanism of Pd nanoparticle catalytic action in the Suzuki reaction: (1) the Pd(0) solid catalyst is immobilized in a solid phase where the reaction mixture contacts the surface (heterogeneous), and (2) Pd is dispersed in the reaction mixture either is atomic form Pd(0) or as a leached Pd(Ar)X species that initiates the homogeneous catalytic cycle. In either case, there has been increased interest in moving towards the use of Pd nanoparticles as a solid phase catalyst (Garrett and Prasad 2004). As such, the design of effective supports for Pd nanoparticles has become increasingly important in the efforts aimed towards intensification of Suzuki–Miyaura catalytic reaction processes (Alonso et al. 2018; Baran 2017; Díaz-Sánchez et al. 2019).

1.2 Process intensification

While process intensification (PI) efforts can take place at any level of granularity (e.g. over an entire process or a single operation), there has been an increased focus on targeting more efficient unit operation design (Freund and Sundmacher 2008). In general, a more efficient catalytic unit operation is accomplished by achieving reduced processing time, increased production volumes, consistent quality and uniformity, or less waste (Keil 2018; Gerven and Stankiewicz 2009; McMillin, Luxon, and Ferri 2020). Gerven and Stankiewicz distilled three targets for PI: driving forces and specific surface area, molecular processing experience, and process synergy (Gerven and Stankiewicz 2009). The multiphase nature of heterogeneous catalytic processes presents numerous challenges in addressing these targets (Arizmendi-Sánchez and Sharratt 2008). In the case of heterogeneous catalytic reaction engineering and reactor design, barriers to presenting a uniform processing experience arise from non-uniform distributions of catalyst on the solid support, non-uniform distribution of flow contacting the support, or a combination thereof. Here, we specifically address the challenges associated with delivering a uniform processing experience.

Although there are numerous strategies for intensifying heterogeneous catalytic systems, the most common approach is to translate batch unit operations to continuous processes. Nearly all approaches that leverage continuous processing encounter common key issues including: (1) the relationship among the fluid phase flow rate, pressure, and reactor geometry (hydrodynamics); (2) immobilization of the heterogeneous catalyst; and (3) the residence time distribution and conversion of reactants to products. Additionally, for supported catalysts to be effective, the metal must be highly active, and strongly bound to the support in order to minimize metal leaching and maximize recyclability (Pagliaro et al. 2012; Péres-Lorenzo 2012).

Debate continues as to how supported Pd species participate as catalysts when immobilized on an interface, e.g. through a truly heterogeneous process or via homogeneous processes (Niu et al. 2012; Richardson and Jones 2007). However, it is commonly agreed that Pd nanoparticles are effective catalysts for Suzuki reactions because of the large surface to volume ratio that is manifest in stable colloidal nanoparticle dispersions. The other issues noted above arise from the hydrodynamic relationships associated with intrinsic and extrinsic features and the operating conditions associated with the physical design of the unit operation. Therefore, it is advantageous to consider strategies that seek to optimize the nanoparticle dispersion and the device hydrodynamics independently. One attractive approach to device design is the use of additive manufacturing (AM); i.e. a bottom-up design method that affords nearly independent variation of the geometric parameters of a device (Capel et al. 2018; Kalsoom et al. 2018; Xing et al. 2018). The key challenge is the incorporation of the catalyst into the AM design.

1.3 Additive manufacturing in heterogeneous catalysis

Additive manufacturing provides the ability to precisely control the geometry of the catalyst support. In industrial applications, heterogeneous catalysts are loaded into randomly packed beds and the reaction fluid flows through the porous bed. The high pressure drop and non-uniform channeling that occurs during flow through a packed bed of catalyst particles can be reduced by structuring the catalyst. Structuring the catalyst as a monolith and decreasing the channel length can reduce pressure drop across the reactor (Parra-Cabrera et al. 2018). Practically speaking, this means higher throughput can be realized with structured catalysts given the same pressure drop. Stuecker et al. demonstrated the dependence of throughput and conversion on catalyst geometry (Stuecker et al. 2004). Three dimensional printing (3dP) has also been used to create catalytically active mixing elements for both flow and batch operations (Avril et al. 2017; Penny and Hilton 2020).

Two strategies are used for immobilizing catalysts on a 3dP structure: integration and functionalization. Integration involves incorporating the catalyst into the print material as an additive or printing the catalytic material directly (Parra-Cabrera et al. 2018). Polymers are the most common filament printing material, but pure polymers are not typically active catalysts. Therefore, polymeric print materials must compounded with catalytically-active species, such as metal oxides, nanoparticles, or metal organic frameworks (MOFs) (Zhou and Liu 2017). There are two consequences: (1) each print material must be custom-prepared, risking print-platform incompatibility and (2) a large fraction of the catalyst in the printed support will be unexposed to the support surface. The second strategy for immobilizing catalysts is through post-print processing and chemical modification (Ambrosi, Moo, and Pumera 2016; Hurt et al. 2017; Lefevere et al. 2013; Stuecker et al. 2004). This approach opens a wider diversity of the additive manufacturing materials library and fabrication techniques, enabling structures and features that present advantages that overcome the challenges associated with post-fabrication functionalization.

In this work, the design of a 3dP fixed bed reactor for continuous Suzuki cross-coupling reactions is discussed. Control of the support geometry for catalytic loading, i.e. surface to volume ratio of support, as well as customization of the flow path using complex internal geometries is demonstrated. The hybrid digital-chemical design approach we use for catalytic reaction engineering is important, because it is scaleable and extensible to a wide range of industrial manufacturing scenarios.

2.1 Design of the 3dP fixed bed reactor

The design of a fixed bed unit operation produced by three dimensional printing (3dP) is divided into three elements: computer aided design, machine code rendering, and printing.

2.1.1 Computer aided design of fixed bed reactor: external features

The overall dimensions of the fixed bed device, including the geometry, were designed in CAD using AutoDesk AutoCAD 2019. For a fixed bed device, these dimensions include port connections (PC), headspace dimensions (H), bed length (BL), and overall part dimensions (L, D); see Figure 1A.

Figure 1:

Key issues in continuous heterogeneous reactor design. (1) Volumetric flow rate, pressure, and reactor geometry; (2) immbolization of heterogeneous catalyst; (3) residence time distribution and conversion.

Port connections were designed to be compatible with common laboratory equipment, such as a high performance liquid chromatograph (HPLC). In this case, the port connections had female 10/32 coned fittings at the inlet and outlet; see Figure 1B. The headspace before the fixed bed affects the manner in which the fluid impinges on the fixed bed; a headspace of concentric cylinders is shown in Figure 1A.

2.1.2 Digital design of fixed bed reactor: internal features

The internal features of the fixed bed were designed outside of the CAD platform. Using the CAD drawing file, the design was exported as a stereolithography file. Cura Version 4.6 was used to render machine instructions. Cura slices the AutoCAD fixed bed device design into layers that are printed sequentially. Following the Cura rendering, the resulting gcode was manipulated via custom Python software to achieve the desired internal geometry. Each printed layer followed a specific geometric pattern. Geometric patterns were selected from a library of patterns and custom-designed. The complex internal geometry of the fixed bed was accomplished by the sequential stacking of different geometric patterns in a single layer or by rotating the same geometric pattern by ordered or random angles ranging from 0 to 360°. The choice of the geometric pattern of a single cross section (g _ij ), the void fraction of a single cross section (ϵ), the angle of rotation between each layer (θ) see Figure 2B, and total layer thickness (h) were all specified. The final rendered internal bed for g ₂₂ is shown in Figure 2C.

Figure 2:

Design of a fixed bed reactor with 3dP. (A) External fixed bed reactor features include port connections (PC), headspace dimensions (H), bed length (BL), and overall part dimensions (L, D); (B) internal features of a fixed bed reactor include geometry of a single layer cross section (g _ij ), the angle of rotation between each layer (θ), and wall thickness (h _w); (C) rendered internal features of fixed bed using Cura 4.6 for g ₂₂; (D) fixed bed coupon printed for [g ₁₁, ϵ = 0.2, θ = (30, 60, 90, 120, 150), h _w = 1.2 mm]; scanning electron micrograph of internal geometry of a single cross-section (1000X). Scale bar is 500 μm.

2.2 Synthesis of palladium nanoparticles

Palladium nanoparticles (Pd NPs) were typically synthesized by dissolving 8.30 mg Na₂PdCl₄ (1.0 equivalent), 6.46 mg l-ascorbic acid (1.3 equivalents), and 11.76 mg K₂CO₃ (3.0 equivalents) in 30 mL of water. Following complete dissolution, the mixture was treated in a microwave oven at full power (1000 W, 2.45 GHz) at 70 °C for 2 min to reduce Pd(2+) to Pd(0). After reduction, HCl (1.0 M) and potassium carbonate were added to obtain the desired pH and ionic strength of the solution. Sodium tetrachloropalladate (Na₂PdCl₄) was used as received from Pressure Chemical. ACS grade l-ascorbic acid (AA), HCl, and potassium carbonate (K₂CO₃) were used without further purification from VWR. In all experiments, Ultrapure water (18.2 M Ω at 25 °C) was used from a MilliQ dispenser.

2.3 Characterization of palladium nanoparticle size and aggregation kinetics

The average Pd NP size and aggregation rate was determined by dynamic light scattering (DLS) using an ALV/CGS-3 Compact Goniometer equipped with a 633 nm laser, ALV/LSE-5004 multiple τ correlator, and two APD based single photon detectors. All DLS experiments were performed using cross-correlation at a detection angle of 90° for 20 s. DLS samples were prepared by diluting 200 μL of the pH and ionic strength adjusted Pd NP solution in 2.8 mL of water. Aggregation studies were performed by determining the mean size at 30 s time intervals over 1 h. All Pd NP concentrations were determined by digestion of a Pd NP solution using concentrated HCl, followed by quantification using inductively coupled plasma – optical emission spectrometry (ICP-OES, Varian Vista-MPX).

2.4 Residence time distribution in 3dP fixed bed reactors

The residence time distribution for the fixed bed 3dP reactor was determined by connecting the column inline to a UPLC (Waters, Alliance e2695 XE with UV/Visible Detection). 1,3,5-trimethoxy benzene was dissolved in ethanol (0.1 v:v) and 1 μL was injected into a pure ethanol mobile phase. Detector signal was monitored as function of time. Absorbance was recorded as a function of mobile phase flowrate. Average residence time was calculated by Equation (4) using the absorbance distribution obtained from the UPLC for each flowrate. 1,3,5-Trimethoxy benzene and ethanol (EtOH) were obtained from Millipore Sigma.

2.5 Palladium nanoparticle adsorption and desorption to a 3dP fixed bed

Adsorption and desorption of Pd NPs to 3dP Nylon fixed beds was measured as a function of bed geometry and total surface area. Fixed beds were prepared using the procedure in Sections 2.1.1–3. However, rather than locate the fixed bed geometry within a reactor, coupons consisting of the exposed internal geometry (10 layers with outer wall) were prepared, as shown in Figure 2D. Pd NPs were loaded onto the fixed bed 3dP coupon by adsorption from solution. The coupon was stirred in a diluted Pd NP solution for 1–3 h at 300 rpm. Following loading, coupons were washed with water to remove lightly bound or unbound Pd NPs from the coupon. The concentration of Pd NPs in solution was determined before and after loading by ICP-OES. Final Pd(0) uptake was determined using a mass balance between the solution and the coupon. Leaching from the coupons was studied by measuring the Pd NP concentration at 1 h intervals after placing the loaded coupon in 20 mL of a 1:1 v:v water and ethanol mixture at 60 °C for up to 24 h. The full 3dP fixed beds were loaded by flowing a 100 ppm Pd NP solution through the fixed bed at 0.67 mL/min for 2.2 and 11 h. CT images of the 3dP fixed bed reactor before and after loading were obtained using a Bruker Skyscan 1173 Micro-CT.

2.6 Palladium nanoparticle catalytic activity immobilized on 3dP fixed bed coupons: (batch)

The catalytic activity of Pd NPs on a 3dP coupon was assessed by monitoring the conversion in the C-C coupling of 4-bromotoluene and phenylboronic acid. In a typical experiment 20 mL water, 20 mL ethanol, 1.32 g K₂CO₃, 0.470 g of phenylboronic acid and the loaded coupon were added to a sample vial. Following addition, the sample vial was heated to 60 °C and stirred at 500 rpm before charging 4-bromotoluene. One hundred (100) μL samples were taken every 10 min for 1 h following a 15 min equilibration period. The samples were diluted with 1 mL of dichloromethane, filtered, and then analyzed by GC-MS. 4-bromotoluene was used without further purification from VWR. Phenylboronic acid was purchased from Chem-Impex.

2.7 Palladium nanoparticle catalytic activity immobilized on 3dP fixed bed reactors: Pd@3dP nylon (continuous)

3dP fixed bed reactors were assessed by monitoring the conversion in effluent residence volumes after flowing a 60 °C preheated mixture of 1-Bromo-3-(trifluoromethyl)benzene (Oakwood Chemical), phenylboronic acid, ethanol, and water mixture at 0.1 mL/min through the fixed beds in a 60 °C water bath for 4 h. The sample workup was the same as in Section 2.6. Desorption of Pd NPs was assessed by collecting aliquots of the effluent at residence volume intervals and analyzing Pd(0) concentration by ICP-OES.

3.1 Hydrodynamics in 3dP fixed bed reactors

In a continuous flow reactor, there are trade offs between the pressure drop imposed across the reactor, the volumetric flow rate of fluid through the reactor, and the residence time distribution of fluid elements in the reactor. All of these relationships are determined, at least in part, by the external and internal geometry of the reactor. To quantify these relationships, 3dP reactors were designed and tested using PLA and Nylon to insure compatibility with the solution conditions used in the Suzuki test reaction, described in Sections 2.6 and 2.7.

3.1.1 Relationship among the fluid phase flow rate, pressure, and reactor geometry

Consider the Ergun equation:

(1) Δ P L = 150 μ u 0 ϕ s 2 D P 2 ( 1 − ϵ ) 2 ϵ 3 + 1.75 ρ u 0 2 ϕ s D P ( 1 − ϵ ) ϵ 3

which relates the bed packing characteristics (ϵ, ϕ, D _p, L) and the fluid properties (μ, ρ) to the velocity in the bed (u ₀) and the pressure drop across the bed (ΔP). The Ergun Equation (1) considers flow of an incompressible, Newtonian fluid through a bed packed with spherical particles as analogous to flow through a collection of cylindrical channels having the same wetted surface area. The total solid surface area in the bed is analytically related to the size of the sphere. More generally, the Ergun equation can be written as

(2) Δ P L = 2 μ u 0 s 2 λ 1 ϵ 3 + f ρ u 0 2 s λ 2 2 ϵ 3

where (s) is the total bed surface area per unit volume (S ₀ L) of the empty column. In (2), the tortuosity (λ = l _i /L) in laminar (λ ₁) or turbulent (λ ₂) flow is explicit. In the low Reynolds number limit, the relationship between pressure drop and volumetric flow rate Q = u ₀ πD ²/4, where (D) is the reactor diameter in Figure 1A, follows the first term of the right hand side of Equation (2).

Equation (2) shows the design parameters available in 3dP fixed beds: (s) and (ϵ). The surface area per unit volume (s) is controlled by the pattern utilized in the digital design. These patterns can be selected from a library or custom designed. Figure 2C shows a selection of design choices – hereafter (g _ij ), which directly affects (s). Other parameters also contribute. Among these, the void fraction of the bed, (ϵ), is also independently controllable in the digital design. More indirectly, the bed tortuosity (λ _i ) is an emergent feature of the fixed bed which arises from choices of (s) and (λ), as well as conditions under which the bed is printed. Other geometric features were varied to facilitate device integration, see Figure 2B, but it was found that these did not affect the flow rate or (ΔP) in a significant way.

Although the bed characteristics (s, ϵ, λ _i ) are dictated by the digital design, a number of approximations must be applied to make an a priori estimate of the fluid throughput as a function of the pressure drop imposed across the reactor. Therefore, direct observation of (ΔP) as a function of (Q) for a fluid of (μ, ρ) can be used to extract these features for a given reactor. Representative experiments are shown in Figure 3 for variations of (g _ij ) and (ϵ).

Figure 3:

Pressure drop ΔP and volumetric flow rate Q in a 3dP fixed bed reactor. Water at 22 °C was flowed at 0 < Q < 5 mL/min through a fixed bed reactor with external dimensions from Figure 1A: D = 10.0 mm, L = 6.54 cm, BL = 6.00 cm, and h _w = 1.2 mm. Internal bed features were: (A) g ₁₁ in Figure 2B for beds of porosity ϵ = 0.3, 0.4 and 0.5; (B) g ₂₂ in Figure 2B for beds of porosity ϵ = 0.3, 0.4 and 0.5.

Figure 3 shows (ΔP) as a function of (Q) for (g _ij ) with beds of porosity 0.3 < ϵ < 0.5. In all cases, (ΔPQ) is linear, suggesting low Reynolds numbers flow. Therefore, the constant group ( s λ l ) can be determined uniquely for each bed (Ergun 1952a)

First consider Figure 3A. For (g ₁₁), ( s λ l ) is a constant. This suggests that the surface area per unit volume and the path length of a fluid element in this device is not a function of the imposed pressure drop. This is consistent with the fundamental assumptions of the Ergun equation, i.e. the flow through the bed is equivalent to flow through a collection of uniform cylindrical channels with equivalent surface area to the bed (Ergun 1952b). Now consider the same device with an alternate cross-sectional geometry, (g _ij  = 22). See Figure 3B. In this case, (g ₂₂), ( s λ l ) is affected by both the bed porosity and the volumetric flow rate through the bed. However at intermediate and high flow rates, ( s λ l ) are independent of operating conditions, for example fluid properties or velocity. This suggests a distribution of channels in the bed of varying resistance; i.e. the flow rate affects the surface area exposed to the bed.

For both internal geometries (g _ij ) shown in Figure 3, 7 × 10 − 3 < s λ l <   3 × 10 − 3 1 / m . To leading order, the larger is ( s λ l ) , the larger is the internal solid surface area per unit volume of the 3dP fixed bed. However, the larger is ( s λ l ) , the smaller is the volumetric flow rate per pressure drop. While these relationships are expected based on the assumptions in Equation (2), the opportunity to digitally design (s) is clearly presented. However to de-confound the internal surface area per unit volume from the bed tortuosity, further experiments are required.

3.1.2 Residence time distribution

A key aspect to understanding the conversion of reactants to products of a continuous reactor is the residence time distribution of fluid elements in the fixed bed. The residence time distribution is typically determined by the use of an inert marker, see Figure 4A. For a single straight channel, a pulse input U(t) broadens to a wider distribution C(t) according to axial dispersion alone, see Figure 4B. For flow distributed through a collection of channels of different pathlengths, the residence time distribution of fluid elements is determined using

(3) E ( t ) = C ( t ) ∫ 0 ∞ C ( t ) d t

where C(t) is the concentration of the marker in the effluent from the bed. From E(t), the mean residence time for all fluid elements is determined using

(4) τ = ∫ 0 ∞ t E ( t ) d t

Figure 4:

Residence time distribution (RTD) as a function of volumetric flow rate Q. (A) RTD experimental set-up; (B) RTD for axial flow in a straight tube; (C) representative RTD for a 3dP fixed bed reactor with dimensions from Figure 1A: D = 10.0 mm, L = 6.54 cm, BL = 6.00 cm, and h _w = 1.2 mm. Internal bed features used g ₁₁ in Figure 2B for beds of porosity ϵ = 0.3 at flow rates Q = 0.1, 0.2, 0.4 mL/min; (D) comparison of RTDs for g ₁₁ and g ₂₂ at Q = 0.1 mL/min.

Figure 4C shows C(t) and tE(t) for (ϵ) = 0.7 and (g ₁₁) in Figure 2B. Figure 4D compares C(t) for (g ₂₂) at 0.2 mL/min.

For a large enough body of observations of (ΔPQ) and E(t) for (g _ij ), we anticipate that supervised or unsupervised machine learning can be used to guide the selection of geometry (g _ij ) for specific design targets such as (ΔPQ) and (τ).

3.2 Immobilizing heterogeneous catalysts to 3dP fixed beds

Now, consider the second key issue associated with continuous heterogeneous catalytic processes: immobilization of the heterogeneous catalyst. A variety of synthetic approaches have been utilized to prepare Pd(0) nanoparticles in solution for Suzuki cross-coupling reactions, see Figure 5A. The surface chemistry of a metal nanoparticle is determined in part by the conditions under which it is synthesized. In turn, the surface chemistry of the particle affects its colloidal stability, interactions with macroscopic interfaces, and catalytic activity; see Figure 5B.

Figure 5:

Pd(0) nanoparticle structure-property relationships. (A) Pd synthesis conditions (laser photoreduction or microwave reduction) dictate the surface chemistry (uncapped or capped) of the Pd nanoparticle; and (B) the surface chemistry of the Pd particle affects its colloidal stability, loadability to interfaces, and catalytic activity.

3.2.1 Nanoparticle colloidal stability

For this example, Pd(0) nanoparticles capped with ascorbic acid, Pd(0)@ascorbate, were prepared by microwave chemical reduction according to the synthesis conditions in Methods and Materials. For these conditions, the mean size of the nanoparticle was R _H = 21 ± 6.5 nm at the 95% statistical confidence.

Nanoparticle dispersions are typically classified as stable or unstable with respect to aggregation. A stable dispersion is one in which the nanoparticles are present as monomers, and the rate of aggregation is slow. Monomeric particles are advantageous for heterogeneous catalysis, because they present the highest surface area per unit mass of catalyst. The colloidal interaction potential between particles in solution determines the stability of the dispersion. The total colloidal interaction potential (ϕ _T) is the sum of the attractive (ϕ _A) and repulsive (ϕ _R) potentials. While (ϕ _A) is determined by the composition of the particle, i.e. the Hameker constant, (ϕ _R) is more directly affected by the surface chemistry. The total interaction potential (ϕ _T) is related to the aggregation rate constant (k ₁₁) via the instability ratio (W ⁻¹).

The rate of aggregation of monomeric nanoparticles into larger agglomerates is a kinetic process (Gambinossi, Mylon, and Ferri 2015; Liu et al. 2012; Singh et al. 2019). Assuming irreversible association, the increase in concentration of dimers is proportional to the product of the initial rate constant of aggregation, (k ₁₁), and the square of the initial concentration of nanoparticle monomer, (C ₀).

(5) d C 2 d t = − 1 2 d C 1 d t = 1 2 k 11 C 0 2

The rate of aggregation (k ₁₁) of a dispersion of nanoparticles can be monitored using experimental approaches, such as dynamic light scattering (DLS). To experimentally determine (k ₁₁), the rate of change of particle hydrodynamic radius is measured as a function of solution condition, such as pH in Figure 6B or ionic strength in Figure 6C and Figure 6D, and analyzed using

(6) k 11 = 1 O C O R H , 1 ( d R H d t ) t → 0

where (C ₀) is the number density of nanoparticles per unit volume, (R _H,1) is the initial particle radius and (O) is an optical factor depending on the scattering vector and particle geometry. By measuring (k ₁₁), the interparticle potential can be determined. The aggregation rate constant is related to the interparticle interaction potential by

(7) k 11 = 8 k B T 3 η 2 R H ∫ 0 ∞ f ( h ) ( h + 2 R H ) 2 e Φ T ( h ) k B T d h

Figure 6:

Pd(0) nanoparticle aggregation kinetics. (A) Hydrodynamic radius R _H(t) for Pd(0)@ascorbate as a function of ionic strength; (B) aggregate growth rate as a function of pH, I = 0.15 M; (C) aggregrate growth rate as a function of ionic strength, pH = 4; and (D) aggregrate growth rate as a function of ionic strength, pH = 7.

In Equation (7), the interaction potential, (ϕ _T(h)), and the hydrodynamic retardation term, f(h), account for the surface and viscous forces manifest between two spheres during mutual approach (Kovalchuk and Starov 2012). Both terms are functions of the intersphere separation distance, (h), and the particle size, (R _H). Further details of aggregation theory and experiment can be found elsewhere (Barton et al. 2014; Kim et al. 2008; Moskovits and Vlčková 2005; Shrestha, Wang, and Dutta 2020; Zhang 2014).

Therefore, for each unique synthesis condition, the colloidal interaction potential can be determined and used to design the conditions under which catalytic nanoparticles are synthesized and stored. In Figure 6A, the evolution of the hydrodynamic radius as a function of time for varying ionic strength of the solution; pH = 6. Figure 6 (B–D) shows the rate of change of the mean hydrodynamic radius in solution, which is a proxy for the aggregation rate as per Equation (7). From Figure 6, it can be seen that for Pd(0)@ascorbate nanoparticles, loading solutions should be limited to ionic strengths (I) > 50 mM and pH > pK_a,ascorbate = 4.0. More detailed measurements of (k ₁₁) enable calculation of the inter-nanoparticle potential via Equation (7), and therefore more precise tuning of the stability of the dispersion as a function of solution chemistry.

3.2.2 Nanoparticle Loading and Leaching in 3dP fixed beds

Strictly speaking, nanoparticles in a liquid are a dispersion, rather than a solution. As such, nanoparticles do not adsorb to interfaces. Therefore, the preferential partitioning of catalytic metal nanoparticles to a macroscopic solid interface is referred to here as loading. In spite of the fundamental differences between adsorption and loading (or conversely desorption and leaching), the Langmuir equation can be used to describe the partitioning of nanoparticles to an interface (Balarak et al. 2017; Chen, Yao, and Kimura 2001)

(8) Γ Γ ∞ = C 0 a + C 0

Equation (8) relates the surface excess concentration Γ of nanoparticles to the bulk concentration C ₀. The Langmuir relationship presumes the loading species preferentially partitions to the interface and are space filling; i.e. the mass uptake is limited to a monolayer.

Using the digital design (g ₂₂) from Figure 2B, three different coupons of total fixed bed surface area (s) = 0.5, 1, 2 were prepared. Each Nylon coupon was immersed in a Pd(0) nanoparticle dispersion of a fixed concentration C ₀ = 10, 20, 50, 100 ppm in a stirred 50:50 aqueous: ethanol solution for 24 h. See Figure 7A. For each concentration, loading was determined by ICP-OES by differencing the Pd(0) in solution before and after immersion of the coupon in the Pd(0) dispersion. Figure 7B shows the mass uptake of ascorbate-capped Pd(0) nanoparticles to 3dP coupons of Nylon, i.e. Pd(0)@ascorbate@3dP Nylon. In Figure 7B, at low nanoparticle concentration, the mass uptake increases as the Pd(0) nanoparticle concentration increases. At high nanoparticle concentration, the mass uptake asymptotically approaches a maximum. Both of these features suggest Langmuir-like behavior. It should be noted however that the Langmuir framework is limited to monolayer adsorption. In Figure 7B, the mass uptake rather than the surface excess concentration is shown.

Figure 7:

Pd(0) nanoparticle loading and leaching from 3dP coupons. (A) Scheme for Pd(0) nanoparticle loading and experiments; (B) loading as a function of Pd(0) concentration in solution fit to the Langmuir adsorption isotherm in Equation (8) for three different coupons of A/A ₀ = 0.5, 1, and 2; (C) leaching from Pd(0)@Nylon 3dP coupons; and (D) maximum surface loading of Pd(0), i.e. the Langmuir adsorption isotherm maximum surface coverage, Γ_∞ × A ₀, as a function of A/A _0.

The extent to which Pd(0) leached from the Nylon coupon after loading was also measured. Coupons loaded at the highest concentration of Pd(0) nanoparticles C ₀ = 100 ppm were immersed in 50:50 aqueous:ethanol solution and stirred at 60 °C. Figure 7C shows the Pd(0) concentration as a function of time in the nanoparticle solution as measured by ICP-OES. Leaching was limited to very low levels at all times.

The data from each loading experiment was fit to the Langmuir adsorption isotherm in Equation (8). The maximum monolayer coverage, (Γ_∞), for each experiment was plotted as a function of coupon area, as shown in Figure 7D. In these experiments, the surface area of the fixed bed increased and decreased two-fold. As the surface area of the solid interface changes, (Γ_∞) should remain constant. It does not. At the highest concentration loading condition (Γ_∞) is statistically larger than the maximum coverage at lower concentration Pd(0) nanoparticle loading conditions. Therefore, it can be concluded that nanoparticle loading exceeds monolayer coverage at the highest loading concentrations. This suggests that catalytic effectiveness will be less than the ideal limit.

3.3 Reaction kinetics and catalytic activity on a 3dP Coupon (batch)

The reaction kinetics and catalytic activity for Pd(0)@ascorbate@3dP Nylon were measured using the reaction of phenyl boronic acid and 4-bromotoluene under conditions in the Materials and Methods section.

3.3.1 Reaction kinetics and catalytic effectiveness

As test reaction, the reaction of phenyl boronic acid (A) and 4-bromotoluene (B) was used; the reaction and experimental scheme are shown in Figure 8A. For the catalytic reaction shown in Figure 8A, the reaction rate law governing the rate of disappearance of 4-bromo toluene or the rate of appearance of the cross-coupled product (C) is

(9) r C = d C C d t = − d C A d t = k p K A K AB C A C B C * ( 1 + K A C A + K A K AB C A C B )

where (C _A) is the concentration of 4-bromotoluene, (C _B) is the concentration of phenyl boronic acid, (C _*) is the concentration of catalyst, and (k _p), (K _A), (K _B) and (K _AB) are the reaction rate constants associated with the reaction in Figure 8A.

Figure 8:

Reaction kinetics and catalytic activity on a 3dP coupon. (A) Cross-coupling reaction of phenyl boronic acid and 4-bromotoluene and experimental set-up; (B) concentration of 4-bromotoluene as a function of time and catalytic loading on the 3dP coupon; (C) psuedo-first order kinetic rate of reaction best fit lines; and (D) psuedo-first order reaction rate constant (k′) as a function of Pd(0) NP loading solution for 3dP coupon.

The reaction was ran under a flooding of excess of phenyboronic acid, i.e. C _B >> 1; see Materials and Methods for details. In this case, the rate law simplifies to

(10) r C = d C C d t = − d C A d t = k P C *

For these conditions, it is anticipated that (C _A) decreases linearly in time for a fixed (C _*). As (C _*) increases, the rate of reaction should also increase.

Test reactions were run in 50:50 aqueous:ethanol solutions and stirred at 60 °C using Pd(0)@ascorbate@3dP Nylon coupons loaded from dispersions of Pd(0) at different concentrations C ₀ = 10, 20, 50, 100 ppm i.e. different levels of (C _*).

Figure 8B shows (C _A) versus time. At low values of catalyst loading ( C * 25 ppm ) , the rate of reaction was approximately constant, suggesting the rate law in Equation (10). However, as C _* increases (C _* = 50, 100 ppm), the rate of reaction decreases with increasing time. Therefore, a pseudo first order kinetic model was used to describe the reaction kinetics. Figure 8C shows the reaction kinetics as fit to the pseudo first order rate law. As anticipated, the pseudo first order reaction rate constant, (k′) is a function of (C _*); as (C _*) increases (k′) increases. Figure 8D shows the pseudo first order rate constant as a function of (C _*). As predicted from the loading experiments shown in Figure 7D, the rate constant (k′) does not increase linearly with (C _*). However, because the deviation from linearity is small, it is anticipated that the loading is only slightly more than monolayer coverage. This is consistent with Figure 7D.

3.4 Conversion and leaching in a 3dP fixed bed reactor (continuous)

A 3dP fixed bed reactor with specifications: (D) = 10.0 mm; (L) = 6.54 cm; (BL) = 4.00 cm; (h _w) = 1.2 mm; (ϵ) = 0.4; (g ₂₂) was prepared.

The 3dP fixed bed was loaded with Pd(0) nanoparticles in flow. To load Pd(0) nanoparticles onto the fixed bed, the minimum volume of a Pd(0) dispersion to use for loading was calculated by considering a mass balance between the surface of the fixed bed and the volume of dispersion to be used. The minimum volume ( V min ) = Γ S 0 L C 0 . 4 < (V) < 20(V _min) was used. The loading flow rate was estimated by considering a balance between the timescale for Pd(0) advection through the fixed bed ( t A ∼ L v max ) and the timescale for diffusion from the dispersion to the fixed bed interface ( t D ∼ R 2 D ) . Therefore the maximum velocity ( v max ) < D L R 2 . The corresponding volumetric flow rate for loading is the product of the maximum velocity and the effective bed cross sectional area (S ₀), π 4 D 0 2 ( 1 − ϵ ) . The volume of a ( C * 100 ppm ) Pd(0) dispersion required to load a D ₀ = 0.5 cm, L = 3 cm fixed bed of (g ₂₂) and (ϵ) = 0.6 is 22 mL. The maximum loading flow rate for these conditions is (Q) = 0.1 mL/min. Figure 9B shows the rendered internal geometry and the 3dP fixed beds before and after loading, and distribution of Pd(0) loaded to the fixed bed internal geometry, as measured by CT.

Figure 9:

Substrate conversion in 3dP fixed bed reactor. (A) Cross-coupling reaction of phenyl boronic acid with 1-bromo-3-(trifluoromethyl)benzene; (B) 3dP fixed bed reactor rendered internal geometry, before and after loading, and distribution of Pd(0) on internal bed geometry by CT (left to right); (C) conversion as a function of Pd(0) nanoparticle loading for two different catalytic loading conditions; (D) Pd(0) leaching as a function of time in 3dP fixed bed reactor effluent.

The performance of 3dP fixed bed reactors was assessed using the reaction of phenyl boronic acid with 1-Bromo-3-(trifluoromethyl)benzene. See Figure 9A. The test reaction was run at 0.1 mL/min of a 10:1 1-Bromo-3-(trifluoromethyl)benzene:4-bromotoluenesolution at 60 °C. The fixed bed reactor was thermostatically controlled to 60 °C and the solution was preheated by a length of tubing immersed in the thermostatic bath. The flow rate was small, so aliquots were collected of sufficient volume to analyze by GC. The conversion as a function of residence time is shown in Figure 9B for two different loading conditions of Pd(0) into the fixed bed reactors. The sample workup was the same as in Section 2.6. Leaching of Pd NPs was assessed by collecting aliquots of the effluent at residence volume intervals and analyzing Pd(0) concentration by ICP-OES.

The conversion of substrate in the reactor effluent loaded at 20 V _min shows a sharp dropoff. This is due to the excess loading of Pd(0). Recall that if loading is perfectly efficient, a monolayer only requires V = V _min. It is likely that excess Pd(0) nanoparticles; i.e. that beyond a monolayer, is sloughed off initially. Note that although there is an initial decrease in substrate during the start-up of the reactor loaded at 4 V _min conditions, this is attributed to the removal of non-adsorbed or weakly bound Pd(0) nanoparticles from the reactor. Substrate conversion is again stable after approximately twenty (20) residence volumes have been processed.