OR WAIT null SECS
Understanding the impact of reactive mass transfer and local flow in multiphase systems is crucial for maximizing reaction selectivity and minimizing the formation of byproducts. The authors study the influence of mixing on fast liquid–liquid reactions. The iodination of L-tyrosine was used to demonstrate the relationship of droplet size and shape on reaction selectivity and to verify computational predictions. By understanding that droplet dynamics affect the yield and selectivity of fast reactions, the formation of byproducts can be minimized by optimizing operating parameters.
Chemical reactions in pharmaceutical and fine-chemical applications typically are carried out in systems that include at least one liquid phase. In many cases, another phase exists, and the reaction may be carried out in a gas–liquid, liquid–liquid, or solid–liquid system. In these systems, the mixing and the phase dynamics are highly complex and may have an impact on product distribution. Such reactions are termed mixing-sensitive. Problems associated with mixing-sensitive reactions often are seen during process scale-up, where impurities and byproducts that didn't exist at the laboratory scale suddenly appear at the production scale. Examples of mixing-sensitive GMP reactions include the hydrogenation of a dinitrile over a Raney-type nickel catalyst (1), the liquid-phase hydrogenation of benzene in metal-hydride slurry systems (2), and the diastereoselective iodohydroxylation of an alkyl enamide to form an intermediate for the HIV inhibitor "Crixivan" (Merck, Whitehouse Station, NJ) (3). In all these systems, the transport in the reactor can influence the selectivity of the chemical reactions, which causes the product distribution to deviate from the one prescribed by the kinetics. In addition, transport effects can influence the enantioselectivity of chemical reactions such as asymmetric catalytic hydrogenations. A specific example of such a reaction is the asymmetric hydrogenation of ethyl pyruvate (4).
For slow reactions, large-scale mixing patterns in the reactor are important for product distribution. For relatively fast reactions, however, such as reactions with shorter time scales than the mixing time scale, the problems associated with mixing-sensitive reactions are even more complicated because micromixing dictates the product distribution and the selectivity. Fast multiphase reactions occur close to the interface. Therefore, the local mixing environment has the strongest impact. Thus, for fast reactions taking place in liquid–liquid systems, the wake and the mixing behind the individual droplets determines the selectivity and byproduct formation. This article explores the impact of wake dynamics of individual droplets on the product distribution of a fast liquid–liquid reaction. To the best of the authors' knowledge, this is the first paper that addresses this effect.
Multiphase reactions are an important and interesting class of reactions, and significant efforts in the literature have been devoted to this topic (5–9). The importance of mixing in reactive systems is paramount because it brings various reactive species in molecular contact, which is a prerequisite for a chemical reaction to take place. Better mixing brings reactants together faster, thus enhancing the rates at which reactions occur. Poor mixing may impede certain reactions, thus altering a process's product distribution (10–14). If fast side reactions are suppressed by low mixing rates (because the reactants for the side reactions are not brought into contact), however, then poor mixing can improve product quality.
Although micromixing and its influence on reaction behavior have been studied (15–18), the industrially important case of reactive micromixing in bubble and droplet swarms has received comparatively little attention. During the past five years, our group has investigated the micromixing effects close to individual bubbles and in bubble–droplet swarms (19–22). The main results of our studies have shown that the selectivity and yield of fast chemical reactions are determined by local micromixing—that is, by the local flow pattern around individual bubbles and droplets in a swarm. If the reaction is heterogeneously catalyzed, then the local density of the catalyst close to the bubble is important as well. Thus, for fast liquid–liquid–solid reactions, the liquid and particle flow close to individual droplets determines reactor performance (23).
The most important feature in terms of local mixing is the dynamic behavior of the droplet (or bubble) wake, which has been extensively studied (24–25). Studies of wake phenomena and fluid flow past solid obstacles date as far back as Leonardo da Vinci (26). Subsequent studies focused on the development of highly accurate pendulum clocks for the determination of a ship's location at high sea. The construction of these clocks required a detailed understanding of the airflow and drag around a swinging pendulum (The Longitude Act was passed in England in 1714, in which Parliament promised a prize of 20,000 pounds for the solution of the "longitude problem." To win the prize, the inventor had to construct an accurate and reliable pendulum clock, requiring a precise understanding of the wake behind the pendulum) (27).
The wake behind droplets and bubbles consists of a primary wake moving in close association with the droplet and a secondary wake extending far downstream (28). Typically three different wake types are observed: a steady-wake without circulation (see Figure 1a), a steady wake with a well-developed circulation zone that can grow significantly (see Figure 1b), and an unsteady wake with vortical structures and vortex shedding (see Figure 1c) (22, 25, 28–33).
Clearly the mixing in these wakes is very different. For example, Figure 2 shows the contour plots of dissolving gas from a bubble rising with the three different wake types. It can be seen that in the case of the steady wake, the gas is concentrated in a small region behind the bubble (see Figures 2a and 2b). In the case of vortex shedding, the gas is rapidly dispersed into the liquid phase (see Figure 2c). It is obvious that in the case of fast, mixing-sensitive reactions, these different types of wakes will result in different product distributions. In an article published in 2000, the authors addressed this problem by analyzing computationally this phenomenon for a fast parallel-consecutive reaction network:
in which A is the liquid-phase reactant, G is the dissolving gas, P is the product, and BP the byproduct (19). This could be a generic network corresponding to a hydrogenation and overhydrogenation. A typical result of our study is shown in Figure 3, which plots the selectivity YP toward the reaction as a function of the bubble Reynolds number. The initial decrease of YP coincides with the onset of recirculation. Recirculation leads to a significantly increased residence time of the product P in the wake and reduces the concentration of the reactant A in the wake. Effectively, the recirculating pattern in the wake acts as a barrier for reactant A, which cannot enter the wake. Thus, because the concentration of A is low, a larger fraction of the gas G reacts with P to form the secondary product BP, leading to a decrease of the selectivity. At Reynolds numbers greater than 52, vortex shedding occurs, which causes a sudden increase of selectivity (see Figure 3). Once vortex shedding occurs, patches of the dissolved gas G (and product P) are quickly convected away from the bubble into regions rich in A, thus leading to significantly higher selectivities toward P. In effect, the transition from a closed wake to an open wake qualitatively changes the mixing behavior, which can lead to a large change in the reaction selectivity.
This study involves a similar reaction network to prove our computational results experimentally. For safety reasons, however, a liquid–liquid system was chosen in favor of a gas–liquid reaction.
Reaction and chemicals. To verify the computationally predicted effects previously published (19–22), an extensive series of experiments was performed. The iodination of l-tyrosine to form 3-iodo-l-tyrosine and 3,5-diiodo-l-tyrosine was studied in a liquid–liquid system. This is a competitive-consecutive second-order reaction, previously studied in a single liquid phase by Paul and Treybal (11), that occurs naturally within the human body. The compound l-tyrosine aids in the production of thyroid hormones by acting as a carrier and allowing iodine to enter the thyroid cells. Figure 4 shows the reaction system.
We purchased 98% pure l-tyrosine from Sigma Aldrich (St. Louis, MO). Samples of 3-iodo-l-tyrosine and 3,5-diiodo-l-tyrosine were also purchased from Sigma Aldrich for use as standards for high-performance liquid chromatography (HPLC). All other chemicals (sodium hydroxide, potassium iodide, iodine, sodium phosphate dibasic, sodium phosphate tribasic dodecahydrate, methylene chloride, and glycerin) were purchased from either Fisher Scientific or Sigma Aldrich and were a minimum of 98% pure.
Setup. The experimental setup is shown in Figure 5. An aqueous continuous phase containing known amounts of l-tyrosine, potassium iodide, phosphate buffers, sodium hydroxide, and glycerin was charged to the column. The dispersed organic phase consisting of a known concentration of iodine in methylene chloride was charged to a syringe. A PTFE needle was attached to the syringe and the dispersed phase was fed into the continuous phase subsurface at a constant rate using a syringe pump. The dispersed phase (i.e., the droplet) travelled down the continuous phase and was removed from the bottom of the column by a very small effluent rate. Hydrodynamics around the droplet mixed the reactants. The volume of the continuous phase in the column remained fairly constant during a given experiment. The column was 18 in. in effective length with a 3-in. internal diameter. The column was designed with five discharge ports to allow top and bottom removal of a volume of the liquid phase. Only the middle portion of the continuous phase was used for sampling.
We developed an HPLC method to quantify and analyze the results. The HPLC method uses a linear ramp of the mobile phase to ensure peak separation and to shorten the run time of the method. All samples were analyzed with an HPLC instrument ("series 1100," Hewlett Packard) and software (ChemStation Rev. A.10.01, Agilent Technologies). The column was a 250 × 4.6 mm i.d., 5 μm C8 column (ACE, Advanced Chromatography Technologies). All samples were analyzed at a wavelength of 210 nm. The mobile phase consisted of acetonitrile (solvent A) and water containing 0.05 v% of perchloric acid and 0.1 v% of phosphoric acid (solvent B). The wavelength and solvents were chosen to minimize the run time while providing good separation and peak formation of each reaction component. (For more details, see Raffensberger ).
Procedure. The aqueous l-tyrosine solution was prepared using the recipe listed in Table I. The water was saturated with methylene chloride to prevent solvent mass transfer from the dispersed phase. The resulting concentration of l-tyrosine solution used in the experiments was 0.084 ± 0.001 mg l-tyrosine/mL water.
Table I: l-Tyrosine solution components.
The organic iodine solution was prepared using the recipe listed in Table II. The resulting concentration of iodine in methylene chloride was 31.7 ± 0.1 mg iodine/mL methylene chloride. The solution was aged for several hours to ensure all of the iodine had dissolved.
Once both solutions were prepared, the continuous phase for a given experiment was made. The continuous phase was 1250 mL in volume and contained 75 mL of the l-tyrosine solution described previously. The remainder of the continuous phase consisted of water saturated with methylene chloride, glycerin, and 1.2 g each of sodium phosphate dibasic and sodium phosphate tribasic dodecahydrate. For example, a 30% glycerin solution would contain 75 mL l-tyrosine solution, 375 mL glycerin, 800 mL of deionized water saturated with methylene chloride, 1.2 g of sodium phosphate dibasic, and 1.2 g of sodium phosphate tribasic dodecahydrate. The solution was mixed for at least 5 min, and the pH and temperature of the continuous phase were checked. The pH of the continuous phase at the start was between 10 and 11. For all of the experiments reported, the starting temperature varied between 22 and 25 °C.
Table II: Iodine solution components.
The continuous phase was charged to the column and allowed to settle to remove all visible air droplets. The iodine solution was injected subsurface using a 20-gauge PTFE needle. Droplets were released at least 4 s apart, thus guaranteeing that a local steady-state had been reached. The iodine charge times for the reported experiments varied between 0.25 and 3.0 h. For analysis purposes, only a 300-mL sample was collected from the middle section of the column with a set of valves. All of the remaining continuous phase was drained to waste, and the sample was analyzed using HPLC.
The viscosity of the solution was controlled by adding glycerin, which lowers the droplet Reynolds number and changes the wake dynamics of a droplet. The glycerin, however, does not affect the reaction nor the HPLC analysis. The glycerin concentration was fixed at either 10% or 30% (and 50%) for the set of experiments. Table III contains information about the experimental properties of the two phases as well as the droplets' parameters.
Table III: Experimental properties.
Although the amount of iodine that phase transfers is not known ahead of time, the initial molar ratio of l-tyrosine to iodine can be determined a posteriori by using the measured concentration of reaction products. Although the absolute amounts cannot be known, the "molar charge ratio"—that is, the initial moles of l-tyrosine to the total moles of iodine that transferred phases—can be calculated as:
in which Ao are the moles of l-tyrosine available for reaction; Bo are the moles of iodine that transfer phases and are available for reaction; P is the concentration of 3-iodo-l-tyrosine in moles per liter, and BP stands for the byproduct concentration of 3,5-diiodo-l-tyrosine in moles per liter. A is the concentration of l-tyrosine at the end of the reaction in moles per liter. The selectivity YP of the reaction was determined by dividing the total amount of desired product formed by the total amount of products formed:
The experiments focused on identifying how the selectivity varies with the initial molar charge ratio for experiments run with 10% and 30% glycerin.
Images of the droplets falling through the stagnant liquid were captured using a high-speed CCD camera that records images at the rate of 500 frames/s. The camera was set to span the entire 3-in. width of the column, and the images were captured after the droplets had fallen approximately 25 cm. Figure 6 shows snapshots of droplets falling through 10% and 30% glycerin solution. From the detailed analysis of all the snapshots (34), one can see that the droplets change shape continuously as they fall. The droplets change from a horizontally elongated cap shape to an elliptical shape and then back to a cap shape. When a droplet fell through the 10% glycerin solution, it was observed to have both path and shape oscillations. The droplets were always released in the center of the column. In the case of the 10% solution, the droplet clearly has moved off center, moving from one side of the column to the other in a zig-zag fashion as it falls, as illustrated in the schematic drawing. Although not visible in the snapshots, this clearly indicates the existence of vortex shedding, as shown by many groups (24).
As the viscosity of the continuous phase increased, the droplet size increased. The droplets became less elongated, and the shape oscillations were less pronounced as the viscosity increased. In addition, the droplet remained at the center of the column, which indicated a closed (steady) wake.
Figure 7 presents a set of snapshots with a time difference of 10 ms for a single droplet falling in a continuous phase with increasing viscosities (10, 30, and 50% glycerin solutions). These snapshots show that as the viscosity increases, the droplet falls at a slower speed, has less pronounced shape oscillations, and becomes more spherical in shape. Zig-zagging was observed only in the lowest viscosity (10%) case.
At the lowest viscosity, the vortex shedding provides good mixing. Because mixing is strong, when products are formed in the wake they are quickly convected downstream and a fresh supply of reactants is brought into the wake region. Thus, as was found with our computational example, the selectivity should be maximized when operating in the vortex-shedding regime because the reaction products do not remain in the wake and therefore, the potential for overreaction to the undesired product is minimized.
Experimental results are shown in Figure 8. The results shown are for 17 different experiments with the charge time of the iodine solution varying from 0.25 h to 2.0 h. When the amount of iodine that transfers phases exceeds the amount of l-tyrosine available for reaction, the initial molar charge ratio is less than one, and the selectivity rapidly decreases as the charge ratio decreases. This result is expected because these conditions favor the second, undesired reaction in our reaction network. If the initial amount of l-tyrosine available for reaction exceeds the amount of iodine that transfers phases, then the initial molar-charge ratio is greater than one, and the selectivity approaches 100%. When the initial molar-charge ratio is equal to one, however, then the effects of mixing can be observed. For this set of conditions, the selectivity at an initial molar charge ratio of one is approximately equal to 85 ± 2%. As shown in Figure 8, the data are reproducible and minimal scatter is observed.
On the basis of theoretical studies, the authors expected that if the flow regime was changed from vortex shedding to a steady-flow pattern (either closed wake or closed wake with recirculation), then the selectivity should decrease as a result of mixing effects. In the steady-wake regime, the fluid flows around the droplet. As the Reynolds number increases, a recirculation region directly behind the droplet develops and increases in size. Therefore, if a reaction occurs in the recirculation region, then the product P will continue to circulate in the wake and can only leave by diffusion. Fresh reactant will not be transported into the recirculation region except by diffusion. For this reaction system, this means that once the desired product is formed, it will remain in the recirculation region, and no fresh reactant will be transported into the region (except by diffusion). Therefore, as the iodine transfers phases, it will preferentially react with 3-iodo-l-tyrosine.
Results for the selectivity as a function of the molar-charge ratio in the steady-flow regime are shown in Figure 9. The continuous phase contained 30% glycerin. Fifteen experiments were run to generate this trend. As in the previous results, one sees that if the iodine is fed in excess, then the selectivity tends to drop quickly, thereby indicating that the second reaction is favored. If the l-tyrosine is fed in excess, the selectivity approaches 100%. At an initial molar-charge ratio of one, however, the selectivity for this set of experiments is approximately 74 ± 2%. This value is substantially smaller when compared with the vortex-shedding regime (10% glycerin in the continuous phase) result of 85 ± 2%.
In this study, a fast liquid–liquid reaction system showed experimentally how droplet wake dynamics can influence the selectivity of the reaction network. The main conclusions observed in this study are:
To the best of the authors' knowledge, this is the first experimental study addressing the micromixing effects of bubbles and droplets. Understanding the effect of reactive mass transfer and local flow in such systems is crucial for maximizing reaction selectivity and minimizing the formation of byproducts. These byproducts can be detrimental to a process because additional manufacturing steps often are needed to separate, dispose, and rework the byproducts. Full-scale manufacturing can be severely constrained by parameters set early in process development. A poor understanding of multiphase reactive flows can impede efficient and effective scale-up. This article is intended to contribute to a better understanding and to the minimization of problems during scale-up of mixing sensitive reactions.
The authors acknowledge support of this work by Merck and Company. J.K. acknowledges funding by NSF through a CAREER Award (CTS-0093129) and NSF Grant CTS 02098764.
Jodi Raffensberger is a process engineer at Merck and Company. Benjamin Glasser is an associate professor, and Johannes G. Khinast* is an associate professor at the Department of Chemical and Biochemical Engineering, Rutgers University, 98 Brett Rd., Piscataway, NJ 08854-8058. Khinast also is Marie Curie Chair at the TU Graz, Austria, tel. +43 316 873 7978, email@example.com
*To whom all correspondence should be addressed.
Submitted: Dec. 12, 2005. Accepted: June 23, 2006. Keywords: Mixing sensitivity, multiphase flows, selectivity
1. E. Crezee et al., "Three-Phase Hydrogenation of D-Glucose over a Carbon-Supported Ruthenium Catalyst — Mass Transfer and Kinetics," Reactor and Catalysis Engineering 251 (1), 1–17 (2003).
2. B.W. Hoffer et al., "Mass Transfer and Kinetics of the Three-Phase Hydrogenation of a Dinitrile over a Raney-Type Nickel Catalyst," Reactor and Catalysis Engineering 59 (2), 259–269 (2004).
3. C.R. LeBlond et al., "Harvesting Short-Lived Hypoiodous Acid for Efficient Diastereoselective Iodohydroxylation in Crixivan Synthesis," Tetradhedron Letters 42, 8603–8606 (2001).
4. Y. Sun et al., "Asymmetric Hydrogenation of Ethyl Pyruvate: Diffusion Effects on Enantioselectivity" J. Catalysis 161 (2), 759–765 (1996).
5. P.A. Ramachandran and R.V. Chaudhari, Three-Phase Catalytic Reactors (Gordon & Breach, New York, NY, 1983).
6. P.L. Mills and R.V. Chaudhari, "Reaction Engineering of Emerging Oxidation Processes," Catal. Today 48 (1–4), 17–29 (1999).
7. W. Halwachs and K. Schugerl, "Investigation of Reactive Extraction on Single Droplets," Chem. Eng. Sci. 37, 1073–1084 (1983).
8. M.M. Sharma, "Multiphase Reactions in the Manufacture of Fine Chemicals," Chem. Eng. Sci. 48, 1749–1758 (1988)
9. B. Li and B.W. Brooks, "Modeling and Simulation of Semibatch Emulsion Polymerization," J. Appl. Poly. Sci. 48, 1811–1823 (1993).
10. P.V. Danckwerts, "The Effect of Incomplete Mixing on Homogeneous Reactions," Chem. Eng. Sci. 8, 93–102 (1958).
11. E.L. Paul and R.E. Treybal, "Mixing and Product Distribution for a Liquid-Phase, Second-Order, Competitive-Consecutive Reaction," AIChE J. 17, 718–724 (1971).
12. J.R. Bourne, E. Crivelli, and P. Rys, "Chemical Selectivities Disguised by Mass Diffusion V: Mixing-Disguised Azo Coupling Reactions, 6th Communication on the Selectivity of Chemical Processes," Helv. Chim. Acta. 60, 2944–2957 (1977).
13. A.W. Nienow et al., "A New Pair of Reactions to Characterize Imperfect Macromixing and Partial Segregation in a Stirred Semi-Batch Reactor," Chem. Eng. Sci. 47, 2825–2830 (1992).
14. J. Baldyga and J.R. Bourne, Turbulent Mixing and Chemical Reactions (Wiley & Sons, New York, NY, 1999).
15. R. David and J. Villermaux, "Interpretation of Micromixing Effects on Fast Consecutive Competing Reactions in Semi-Batch Stirred Tanks by a Simple Interaction Model," Chem. Eng. Commun. 54, 333 (1987).
16. J. Villermaux and L. Falk, "A Generalized Mixing Model for Intial Contacting of Reactive Fluids," Chem. Eng. Sci. 49, 5127 (1994).
17. R.O. Fox, "On the Relationship between Lagrangian Micromixing Models and Computational Fluid Dynamics," Chem. Eng. Process. 37, 521–535 (1998).
18. S. Sundaresan, "Modeling the Hydrodynamics of Multiphase Flow Reactors: Current Status and Challenges," AIChE J. 46, 1102–1105 (2000).
19. J.G. Khinast, "Impact of 2-D Bubble Dynamics on the Selectivity of Fast Gas–Liquid Reactions," AIChE J. 47, 2304–2319 (2001).
20. J.G. Khinast, A.A.Koynov, and T.M.Leib, "Reactive Mass Transfer at Gas–Liquid Interfaces: Impact of Microscale Fluid Dynamics on Yield and Selectivity of Liquid-Phase Cyclohexane Oxidation," Chem. Eng. Sci. 58, 3961–3971 (2003).
21. A. Koynov and J. Khinast, "Effects of Hydrodynamics and Lagrangian Transport on Chemically Reacting Bubble Flows," Chem. Eng. Sci. 59, 3907–3927 (2004).
22. A. Koynov, G. Tryggvason, and J.G. Khinast, "Mass Transfer and Chemical Reactions in Reactive Swarms with Dynamic Interfaces," AIChE J. 51 10, 2786–2800 (2005)
23. J. Raffensberger et al., "Influence of Particle Properties on the Yield and Selectivity of Fast Heterogeneously Catalyzed Gas–Liquid Reactions," Int. J. Chem. Reactor Eng. 1, A–15 (2003).
24. L.-S. Fan and K. Tsuchiya, Bubble Wake Dynamics in Liquids and Liquid–Solid Suspensions (Butterworth-Heinemann, Boston, MA, 1990).
25. D. Bhaga and M.E. Weber, "Bubbles in Viscous Liquids: Wakes, Shapes, and Velocities," J. Fluid Mech. 105, 61 (1981).
26. A.E. Popham, The Drawings of Leonardo da Vinci (Jonathan Cape, London, UK, 1946).
27. D. Sobel, Longitude (Penguin Books, New York, NY, 1995).
28. T.Y. Wu, "Inviscid Cavity and Wake Flows," in Basic Developments in Fluid Dynamics, M. Holt, Ed. (Academic Press, New York, NY, 1968).
29. J.T. Lindt, "Note on the Wake Behind a Two-Dimensional Bubble," Chem. Eng., Sci. 26, 1776 (1971).
30. J.T. Lindt, "On the Periodic Nature of the Drag on a Rising Bubble," Chem. Eng., Sci. 27, 1775 (1972).
31. J.H. Hills, "The Two-Dimensional Elliptical Cap Bubble," J. Fluid Mech. 68, 503 (1975).
32. P.G. Saffman, "Viscous Fingering in Hele-Shaw Cells," J. Fluid Mech. 73, 173 (1986).
33. T. Miyahara, K. Tsuchiya, and L.-S. Fan, "Wake Properties of a Single Gas Bubble in a Three-Dimensional Liquid-Solid Fluidized Bed," Int. J. Multiphase Flow 14, 749 (1988).
34. J.A. Raffensberger, Impact of Catalyst, Bubble, and Droplet Properties on the Selectivity of Fast Multiphase Reactions, PhD Thesis, Rutgers University (2004).