You are on page 1of 19

MODELING OF PRACTICAL PHOTOELECTROCHEMICAL REACTORS

1 INTRODUCTION
The desire to convert solar energy to usable energy for both commercial and consumer use is an important goal, especially as global energy forecasts worsen. Photo-electrochemical systems, which are capable of producing molecular hydrogen and oxygen gas directly from water, are an appealing solution that allow for both energy conversion and storage. Over the last 30 years, scientists and engineers have focused mainly on developing materials that are highly efficient at capturing solar radiation and relatively inexpensive to manufacture (for example, TiO2 has been well studied). While such studies are the key to ultimately developing models that can compete against other forms of energy production, commercial reactors at the consumer or industrial scale will require further design work with regards to kinetics and transport processes, at the bulk semiconductor, interfacial, and bulk fluid scales. These studies can take place in parallel to the development of new materials. This study aimed to develop a first-order model of a complete photo-electrochemical reactor. A variety of phenomena were incorporated, including flow and transport within the electrolyte fluid, kinetics at electrode interfaces, and electronic effects within the semiconductor electrodes. All of these processes have been studied in typical (non-photo) electrochemical cells, but have not yet been applied to practical photoelectrochemical reactor designs. To ensure that the design was practical and verifiable, all geometries and materials were chosen to mimic a recently developed flow-through photo-electrochemical reactor. Most importantly, this model will provide insight into the limiting physics of the reactor design, be it transport, kinetics, or materials design. Finally, it can provide insight into the design of the next generation of reactors.

REACTOR GEOMETRY AND CONFIGURATIONS

A recently developed photo-electrochemical reactor design was chosen as the basis for this study. The reactor was designed to test photo electrodes in a flow reactor, similar to what a commercial reactor might resemble. This design can thus act as a stepping stone in the scale up from the lab bench to a full design. The reactor in a typical configuration is demonstrated in Figure 1. The reactor was composed of a series of milled PVDF plates, providing a variable number of flow channels. Plates A and F provide the top and bottom of the reactor, with plate A having a central hole to allow light to enter the reactor. Electrodes B and E represent the photo-anode and cathode (not necessarily in that order). Plate C represents a flow channel, with electrolyte entering from the hole at the top of the plate and leaving through an identical hole at the bottom. Finally, D is a thin membrane acting to separate the anodic and cathodic products while allowing the transfer of the supporting electrolyte. The modular design of the reactor allows for a variety of reactor configurations. A typical configuration, analogous to the exploded view, is shown in Figure 2. Electrolyte flows through the two channels, and exits through the opposite side of the channels. In the presented case, light enters through the top of the reactor, passes through the solution, and strikes the photo-anode layer lying at the bottom. However, a variety of other configurations are possible. The simplest consists of just the top and bottom plate (A and F), the electrodes (B and E), and a single channel (C). As long as electrochemical current flow is low, the combustion

of hydrogen and oxygen gases should not be an issue. In addition, a single channel design greatly simplifies the design. Both single and double channel systems were considered within this study.

A B

C D

C E W

Figure 1: A schematic of the reactor configuration, both disassembled and assembled.

Figure 2: The active region within the reactor, constructed from the exploded view in Figure 1.

MATERIAL SELECTION FOR THE PHOTO-ANODE

Previous efforts have established a large number of candidate materials for the photo-anode in a water splitting device. The first relatively efficient material discovery was for TiO2 (Fujishima and Honda 1971). A huge number of studies, on both TiO 2 and on newer materials have since been conducted. Modern possibilities under investigation include new semiconductor materials, the addition of photosensitive dyes, and the use of complex surface morphologies. Polycrystalline silicon-doped hematite ( -Fe2O3) was chosen as the primary material to compose the photo anode for several reasons. Most importantly, hematite is an abundant and cheap material, making it extremely economical. The band gap of hematite in various forms has been measured to be 2.0-2.2 eV (Cesar, et al. 2009), corresponding to an upper wavelength threshold of 620 nm, encompassing a large portion of the solar spectrum. Unfortunately, hematite has a conduction band energy lower than required for the evolution of hydrogen, requiring an additional small bias voltage. Furthermore, the diffusion length of minority carriers (photo-generated electrons and holes) is extremely short, about 2-4 nm (Kennedy and Frese 1978), restricting the active regions of hematite films to just next to the film/solution interface.

Silicon doped hematite thin films without catalytic surface sites were used throughout this study, although many possibilities have been tried in literature. To compensate for the short carrier diffusion length, nanostructured films have been proposed and tested (Cesar, et al. 2009). Another approach is to layer small amounts of hematite onto a structured host film, such as tungsten oxide (Sivula, Formal and Gratzel 2009). Recent efforts have been made to incorporate nanotube synthesis strategies (increasingly common throughout materials science) (Rangaraju, et al. 2009). Other groups have proposed the use of several stacked thin hematite sheets (mentioned as a solution in most hematite studies, such as in (Cesar, et al. 2009)). Furthermore, a variety of dopants and dyes have been used to modify the electronic structure and absorb higher wavelength light. Representative studies include doping with Si (Cesar, et al. 2009), Mo and Cr (Kleiman-Shwarsctein, et al. 2008), Ge and Mg (Sanchez, Steinfink and White 1982), Nb (Sanchez, et al. 1986) and Ti (Sartoretti, et al. 2003) among others. Finally, various catalytic additives have been deposited on hematite to improve performance such as Cobalt-Phosphate (Gamelin, et al. 2009).

COMPUTATIONAL METHODS

Due to the complexity of the model and the non-ideal geometry of the reactor, analytical solutions to the various design equations were not possible. Instead, the flexible numerical PDE package COMSOL was used. Several of the desired models were already provided within the software, simplifying model development. The description of the model will be in the context of the COMSOL model, but should be applicable to similar packages. Unless specified otherwise, solution time for each aspect of the model was on the scale of minutes, and memory usage was on the order of gigabytes. Numerical meshes were refined until results did not change with increased meshing or until memory usage exceeded 4 Gb. Several models with varying geometries and numbers of dimensions (2D/3D) were developed to assess fully understand the modeled phenomena. Reactors with either one chamber or two chambers separated with a membrane were considered. Additionally, models were developed both for the full 3D reactor and for 2D cross sections of the reactor (parallel to the mean flow direction). Obtaining solutions to the 2D case were straightforward, but additional considerations were necessary for the 3D case. The flow of fluid through the reactor was highly dependent on the entrance and exit regions, as detailed later. However, the transport of dissolved gases and ions in solution required boundary layer computational meshes, which added too much complexity to the nonlinear flow calculations. A hybrid method was instead used, demonstrated in Figure 3, where a regular (relatively coarse) mesh was used to solve for the flow profile, which was then transferred to a more detailed mesh incorporating only the reactor region around the electrodes.

(a) Triangular mesh for fluid flow calculations

(b) Logarithm-spaced mesh for transport calculations

Figure 3: Demonstration of hybrid meshing method used to solve for both fluid flow and flow-dependent transport phenomena. The inner rectangular mesh of (b) corresponds to the region on which the electrode sits.

5 5.1

DOMAIN MODELING GENERAL OVERVIEW

The wide range of phenomena present within the photo-electrochemical reactor, spread over several orders of magnitude in both time and space, suggested that each region within the reactor should be treated separately and coupled together. A simplified view of the reactor setup can be seen in Figure 4. The mesh cathode was treated as a standard electrode, with a fixed voltage source applying a potential difference between it and the anode (and providing electrons). At the electrode surface, the electrons were used to + convert aqueous H into gaseous H2. The electrolyte flow in each channel was considered to be independent, with the flow generally being laminar (justified later). A membrane separated the two channels to prevent the diffusion of H2 and O2 into the opposite compartments. Finally, the electronic structure of the photo-anode was considered (with corresponding light absorption and electron conduction) and used to evaluate the kinetics of oxygen evolution at its surface.

Figure 4: A simplified view of the various physical phenomena considered and their interactions.

The expected potential distribution across the cell for the system in Figure 4 using a Fe2O3 anode is shown in Figure 5. The electronic band gap for iron oxide is shown in shaded yellow. While the bandgap is relatively narrow, the ideal material would have a conduction band just above the H2O/H2 potential and a valence band just below the H2O/O2 potential (with enough excess to achieve reasonably fast kinetics). Within the photo-anode, both the valence and conduction bands are bent due to the semiconductor/solution interface. There is a large potential change across the interface, followed by small resistive changes across the solution and membrane. At the cathode, there is a second sharp change in potential across the cathode / solution interface, but due to the highly conductive nature of the metal, there is a uniform potential in the cathode. Models for all of these phenomena are considered throughout this report.

Figure 5: Overview of the potential losses for the entire two-chamber reactor. As described later, most reactor limitations occur either within the semiconductor or at the semiconductor/solution interface. The figure is not drawn to scale.

5.2

TOTAL POTENTIAL DROP AND THERMODYNAMIC TREATMENT

The potential distribution across the reactor is critical to understanding all aspects of the photoelectrochemical reactor. The total potential drop is described here as it is best described in an overall manner, rather than in the context of each particular section and interface, and is shown in Figure 3. The potential drop across the entire reactor is:

The potential drop across each of the interfaces is:

Where is the thermodynamic potential, is potential loss due to band bending in a semiconductor, and is the overpotential. In many electrochemical experiments, a potentiostat is used to control the interfacial potential drop. Thus, the entire potential drop will be: ( ) ( )

The thermodynamic potentials can be obtained from the Nernst equation, using well known constants for the separate reactions at the anode and cathode:

Where is the equilibrium potential, and are additional constants, and is the partial pressure of the relevant gas (H2 or O2). If the reaction is contributing to bubble formation, will be atmospheric pressure plus a diameter-dependent contribution from the curved bubble/solution interface. For the purposes of this study, was taken to be atmospheric pressure in all cases. The thermodynamic terms for the anode are cathode were thus: [ ] [ ] When a potentiostat is used to control the potential drop across one of the interfaces, the balance for the interface becomes:

When the thermodynamic potential and band bending potential can be estimated from local conditions (pH and such) the overpotential can be calculated.

5.3

BULK ELECTROLYTE

The bulk electrolyte, critical to the understanding of any electrochemical reactor, was modeled first to obtain an understanding of both the fluid dynamics and the distribution of ions (both the supporting electrolyte and the pH gradients). A high concentration of supporting electrolyte was assumed, such as 1M sodium sulfate. In addition, it was assumed that the pH of the inlet electrolyte could be well controlled. In the actual implementation, the electrolyte would be recycled through a reservoir, where such control could take place.

5.3.1 ELECTROLYTE FLUID FLOW PROFILES


The first aspect to be considered was the flow profile of the fluid within the central region of each o reactor. A comparison of the density and viscosity at 25 C of sea water (ionic concentration of 0.66 molal) of to that of pure water is shown in Table 1. Since the values for each were similar, and the fluid flow profile does not affect the flow profile of a fully developed laminar flow, the temperature dependent values for water were used.
Table 1: Comparison of the density and viscosity of sea water to pure water (CRC Press, Inc 2008-2009)

Pure Water Salt Walter

Density at 25 C (kg/m ) 997 1023

Viscosity at 25 C (cP) 0.89 0.97

The flow regime of the electrolyte solution was determined through an analysis of the transition flow rate for the central region of the flow channel. To accommodate the rectangular cross section, the hydraulic diameter was used in the calculation: [ ] [ The corresponding volumetric flow rate for this velocity was: [ ] ] [ ] [ ]

This flow rate is quite large (and greater than the capability of the pump employed in the reactor system. Thus, the flow regime throughout most of the reactor was assumed to be laminar (the entrance region would become turbulent at lower flow rates). The laminar flow profile was calculated using the full Navier-Stokes equation for incompressible fluids. Diffusion of solutes within the solution was assumed to have no impact on the flow profile, although diffusion influenced convection could be possible. Both the 2-dimensional (trivial, parabolic) case and 3dimensional case (using the full reactor) was solved, with the appropriate momentum balance with no body forces: ( ) [ ( ) ]

The boundary conditions for the inlet and outlet of the three-dimensional model were a laminar inflow and outflow. Along each surface (and along the electrode and membrane), no-slip conditions were used. Since each flow channel was symmetric in both directions perpendicular to the length of the reactor, only one quarter of the channel was modeled (and the solution transferred to the rest of the channel). The solution of the fluid flow profile for the two-channel reactor scenario is shown in Figure 6. Of particular importance was the distribution of flow over the surface of the electrode (and the corresponding design of the inlet region).

Figure 6: The fluid velocity profile, | |, for the two-channel reactor configuration at low flow rates.

5.3.2 MASS TRANSPORT OF GASEOUS PRODUCTS

The transport of products formed at each electrode, molecular hydrogen and oxygen, were modeled separately from the electrolyte concentration. At moderate current densities, the concentration of these species at the surface of the electrodes will exceed the thermodynamic saturation concentration and form bubbles within the flow. While reactors can be designed to account for bubble formation, any bubbles produced in this design would either be swept along with the flow, or trapped in the upper section of each channel. In either case, the semiconductor/solution surface area becomes unsteady and complicated. Some studies in the past have considered the effect of bubble formation on surface activity, for example (Vogt 1980). To simplify the model and experimental methods, the reactor was always run so that no bubble formation occurred and simple diffusive/convective flow dominated. The system was modeled as follows:

Where is the concentration of either oxygen of hydrogen and is the velocity field from the bulk electrolyte flow. The saturation concentration and diffusion coefficients for hydrogen and oxygen in water are shown in Table 2. Saturation concentrations were related to temperature through Henrys law (in equilibrium with a partial pressure of 1 atm) and empirical relation for the constant. The boundary conditions for the each species were coupled to the kinetics of the surface reactions on the electrode, with a zero inlet concentration. Example calculations and determination of the maximum current density before bubble formation is shown later.
Table 2: Diffusion coefficients and saturation concentrations for molecular hydrogen and oxygen. (CRC Press, Inc 2008-2009)

Saturation Concentration [ Hydrogen (H2) Oxygen (O2) [ [ ( ( [ ] [ ]

] )] )]

Diffusion Coefficient at 25 C [

5.3.3 ION CONCENTRATION AND POTENTIAL DISTRIBUTION


Determining concentration gradients of important ions in an electrolyte solution (in this case, H and OH ) is greatly complicated by the electric potential applied across the reactor channel. The resulting solution of the complex interactions that arise, including ion/ion interactions, diffusion induced convection, and electromagnetic/ion interactions, is called the tertiary current distribution. The problem is still being developed in literature, making inclusion in this first-order model difficult. Previous successes in the field include (Wang 2004), where a full computational model including induced convection for a PEM fuel cell was developed and experimentally verified, (Basha 2006), where semi-analytical solutions were developed for a variety of electrochemical geometries based on uniform activation potentials (which would not be true for semiconductor electrodes), and (Chung 2000), where natural convection was included in the solution.
+

The most widely accepted method for calculating ion transport within the reactors is the NernstPlanck equations, which model transport due to diffusion, electric potential, reaction, and convection. For + + each ion in solution (for example, Na , Cl , H , OH ), the flux is given by:

Where is the species diffusion coefficient, the species concentration, the species mobility constant, the species charge, the Farraday constant, the species concentration, the electric potential, and the flow velocity as calculated above. The diffusion coefficients for the various ions considered are (in + + units ( ): Na ,1.33, Cl , 2.03, H , 9.0, OH , 5.0 (CRC Press, Inc 2008-2009). The species mobility was related to the species diffusion coefficient through the Nernst-Einstein equation:

The transient balance for an infinitesimal fluid element in the reactor is then: ( )

Where is the total production or consumption from chemical reactions for species . At steady state, this becomes: ( )

In the fluid bulk (that is, excluding molecular-width boundary layers), any fluid element can be assumed to be neutral. The total charge of all species was related through:

This conservation equation, typically called the electroneutrality condition, allowed the differential equation for one of the ions in solution to be replaced by an algebraic relation rather, greatly simplifying the solution. In order to include the proton/hydroxide/water equilibrium, a pseudo chemical reaction was included. For the hydroxide and proton species, the following term was used: [ ] ( [ ][ ])

Where was the equilibrium constant of water ( ). The rate constant ( [ ] ) was chosen so that it was neither too high (causing numerical instability) or too low (allowing out of equilibrium proton/hydroxide concentrations). This constant needed slight adjusting depending on the pH gradients involved in the solutions.

5.4

MEMBRANE

The membrane for the reactor was assumed to contribute to the reactor model only through a small ohmic potential loss, . For the examples provided, this potential was set to 0. The membrane would become especially important when trying to model the separation of gases produced from the anode and cathode (H2 and O2), a mixture of which would be highly explosive.

5.5

BULK CHARACTERISTICS OF THE SEMICONDUCTOR PHOTO-ANODE

5.5.1 ELECTRONIC STRUCTURE


The ability of a semiconductor to absorb light at various wavelengths is entirely determined by its electronic structure, namely the position of the valence and conduction bands ( and ). For photons to be absorbed, they must be energetic enough to promote electrons from the valence to the conduction band ( ). Ideally, the valence band holes will be below the potential of the oxygen evolution potential so as to drive the production of oxygen gas. At the same time, the potential of conduction band electrons should be high enough to drive the production of hydrogen gas at the (non-photo) cathode.

5.5.2 ELECTRON AND HOLE MOBILITY AND RECOMBINATION RATES

Performance of the photo-electrochemical cell can be limited by transport effects of the minority carriers if the current is close to the limiting value. While models of the semiconductor/solution interface can be restricted to this low current case, a full understanding requires knowledge of transport parameters within the bulk semiconductor. The most widely used parameters to characterize this type of transport is the diffusion length of holes and electrons, and the recombination rate. Diffusion lengths for electrons and holes tend to be equated and represented as minority carrier diffusion lengths (roughly justified as both are effectively electron movement). Literature parameters for -Fe2O3 vary widely, and are assumed to be independent of hole/electron energies and isotropic in nature. Electron diffusion lengths have been reported of 2-4nm (Kennedy and Frese 1978) or 20nm (Dare-Edwards, et al. 1983). Despite many studies over the past two decades no improvements to these measurements have been made, although recent studies tend to cite (Kennedy and Frese 1978) more frequently.

5.5.3 LIGHT ABSORPTION


When the transport of electrons and holes within the semiconductor photo-anode is potentially limiting, the spatial distribution of photo-generated electron/hole pairs becomes important. The light intensity will be largest near the electrolyte/solution surface of the photo-anode (in the configuration shown in Figure 4) and decrease as photons are absorbed. If the photo-anode is thick enough, the light intensity at absorbable wavelengths should reach zero. A complete model of the propagation of light through the semiconductor interface should include effects from both absorption and scattering processes, especially when the semiconductor/solution interface is coated with particulate or porous catalyst material. Some recent studies have included scattering effects (Reyes-Coronado, et al. 2008), and others have even encouraged the phenomena to improve performance (Sheng, et al. 2009). However, as a simplification and because of the availability of UV/VIS absorption data for a range of materials, only light absorption was considered. The intensity profile was thus taken to be of the form: ( ) ( )

Where was the surface reflectivity, was the surface light intensity, was the experimentally measured effective absorption coefficient, and is the distance from the electrode/solution interface into the semiconductor film. Before attempting to identify absorption coefficients for the materials used in this study, the reliability of literature UV/VIS data for TiO 2, a well-studied material, was investigated. Data from three studies, where UV/VIS was conducted on samples of known thickness and absorption coefficients calculated, is shown in Figure 4. Absorption coefficients from (Aarik, et al. 1997), (Sumita, et al. 2002), and (Koelsch, et al. 2002) are shown for rutase, anatase, and brookite TiO2, generated using four techniques (pulsed layer deposition, o o PLD, atomic layer deposition, ALD, and sol-gel), generated at two temperatures (100 C and 300 C) and at various concentrations. Although there is rough agreement in the distribution of the absorption coefficients (the shape of the curve), there are significant differences in the absorption coefficient at any specific wavelength. This suggests that attempting to find literature absorption coefficients that correspond to newly developed samples will be unsuccessful, and that absorption coefficient curves will need to be collected with UV/VIS instrumentation for each new photo-anode prepared.

Figure 7: Comparison of literature values for the absorption coefficient of TiO 2 films made through a variety of techniques. (Koelsch, et al. 2002) did not specify units, and are thus shown separately.

Give examples of absorption lengths, as intensity assumed to have linear effect on rate of generation of electrons and holes. Discuss effect of porosity of semiconductor

5.6

PHOTO-ANODE AND SOLUTION INTERFACE

The kinetic and thermodynamic treatment of the reactions and charge profiles of the photo-anode and solution interface are necessary for a complete understanding of the photo-electrochemical process in Figure 4. Near the limiting current for a photo-electrochemical cell, the electron/hole transport properties in the bulk semiconductor will most likely dominate (since the generation of photons is usually lower than the maximum current that can be sustained through mass transport in the solution) (Bockris and Reddy, Modern Electrochemistry 2B 1998). However, below the limiting current the interface may play a large role. This is further evidenced by the effectiveness of adding large amounts of catalytic surface sites (Thampi, Kiwi and Grtzel 1987).

5.6.1 CATALYTIC SURFACE SITES


The use of catalytic additions to photo materials to increase the rate of surface reactions is extremely common. For example, Ruthenium and Ruthenium oxides were used to enhance the rate of a TiO 2-based photomethanation of CO2 (Thampi, Kiwi and Grtzel 1987). Assuming that these catalytic materials do not significantly alter the electronic structure of a semiconductor photomaterial, their impact will be purely kinetic. However, quantitatively determining the impact of various catalytic materials is complex, even in the gas phase with a known chemical pathway (Prasad, et al. 2009). DFT-based first-principles studies of TiO2 surfaces have been conducted recently (Valds, et al. 2008), but make no attempt at predicting the effects of additives or different morphologies at the reactor scale. Thus, the role of catalytic sites for this model must be confined to the kinetic rate constant from the Butler-Volmer method and will be fitted to experimental data.

5.6.2 ELECTRIFIED INTERFACE (ELECTRONIC DOUBLE LAYER)


The potential distribution at a semiconductor/solution interface is complicated by a relatively slow transition from the bulk material potential to the bulk solution. In a metal, where the conductive nature ensures the potential is nearly uniform everywhere inside, there will only be a small potential change approaching the surface. However, in the transport controlled flows of the semiconductor, there will be a much larger potential change depending on the local electric fields, the consumption rate at the surface, and

the local diffusion and mobility constants of the relevant charge carriers. Extensive progress has been made in describing the coupling of these interactions mathematically, beginning with (Bockris and Uosaki 1977), developing with (Albery, Bartlett and Dare-Edwards 1981), and largely completed with (Khan and Bockris 1984). Most work was conductor for planar geometries, but some models also took into account spherical interfaces (Albery and Bartlett 1984). More recent work has largely focused on quantum mechanical effects for understanding of highly idealized surfaces inappropriate for the work here (Valds, et al. 2008). The importance of electronic effects within photo-electrochemical reactors depends strongly on the operating conditions. When the current is close to the limiting current in a standard electrochemical reactor, performance is limited by traditional mass transport effects at the electrode/solution interface. Photoelectrochemical reactors could be limited by these effects, but are usually limited instead by transport within semiconductor electrodes (Bockris and Reddy 1998). When the current is well below the limiting current, traditional charge transport kinetics take over. This phenomenon was first confirmed experimentally for semiconductors using a p-GaP electrode (Uosaki and Kita 1981), where normal Tafel relations were identified with medium currents below the limiting current and well above zero current. If bubble formation is avoided, the limitations due to mass transport of the product dissolved gases will limit the operating current well below the limiting current for the cell.

5.6.3 BUTLER-VOLMER KINETICS


The most popular method of estimating the kinetics of charge transfer at an electrode/solution interface is the Butler-Volmer equation. The equation relates the current density to an overpotential based on a potential barrier argument (similar to that of an Arrhenius-type rate equation). The full equation, which is derived with discussion in (Bockris and Reddy 1998), is as follows: (
( )

Where is the exchange current density, is an empirical constant representing the position of the potential barrier with respect to the width of the Helmholtz layer, and is the overpotential. The exchange current density is further dependent on the concentration of reactants depending on the order of the reaction. However, for similar environments, with only the potential applied across the electrode/solution interface changing, the Butler-Volmer equation is sufficient. When is sufficiently small, on the order of 10 mV for normal electrochemical systems, the current density simplifies to:

At the opposite extreme (high overpotentials), the current density simplifies to:
( )

This equation forms the physical grounding for Tafel plots, where the overpotential is plotted versus the ) log of the current density. Effectively, the constants and ( can be determined. Although the semiconductor/solution interface is far more complicated than the simple model on which the Butler-Volmer equation is based, Tafel lines are observed when photocurrents are sufficiently low (Uosaki and Kita 1981). To test this observation and identify the scaling of Si-doped Fe2O3 electrode performance with respect to potential, data electrode performance data from a variety of recent sources were collected and are presented in Figure 8. Data for 10 electrodes from four sources are shown. (Gamelin, et al. 2009) developed electrodes with and without an added layer of Co-phosphate catalyst. The deviant line shows that the type of catalyst can greatly affect electrode scaling, and that perhaps the Co-phosphate additions were doing more than providing surface sites for improved reaction rates. (Kay, Cesar and Gratzel 2006) was the first of a series of efforts by

the Gratzel lab to improve Si-doped iron electrodes with nano-patterning. A follow-up study with variations in preparation method, (Cesar, et al. 2009), and experiments using WO 3 as a substrate to deposit Fe2O3 onto, (Sivula, Formal and Gratzel 2009) show similar behavior. All 10 electrodes exhibited Tafel relationships below approximately . Thus, Tafel relations were sufficient to represent the kinetics until bubble formation begins (estimated to be on the order of 1 ). The average Tafel slope (excluding the outlier) was calculated to be . The fitted Tafel relation is then:

Calculating

from this relation (assuming all experiments conducted at approximately 298K):

Interestingly, this value is quite close to the nave assumption of used in many electrochemical studies and derivations (Bockris and Reddy 1998). These relations also show how much the kinetics can vary depending on the synthesis conditions, even when considering a single material composition. will need to be a fitted variable for every material.

Figure 8: Tafel plots for a variety of Si-doped -Fe2O3 electrodes from literature. Data from 10 samples among four publications are shown, resulting in an average (excluding the one clear outlier where large amounts of surface catalyst were added) linear slope of , corresponding to . Bubble formation is estimated to occur at approximately 1 A/m2, below the limit of applicability of the observed Tafel relations.

5.7

BULK CHARACTERISTICS OF THE CATHODE

The kinetics and modeling of the cathodic reaction, hydrogen gas evolution, was disregarded for the setup listed in Figure 4 for several reasons. The cathode in this reactor is not photo-active, with of all of the major physics (electronic bandgaps and electron/hole generation and mobility) directly based on the material of the photo-anode. Since the preparation technique and morphology of the photo-anode has been shown in numerous instances to have a large effect on the performance of photo-catalytic reactors (a somewhat obvious observation), the kinetics of the alternate electrode should not be the rate limiting process. However, it has been suggested that photo-anode/photo-cathode pairs be used, especially if the materials complement each other by having different bandgaps (and thus having different absorbance spectra), such as those

demonstrated in (Licht, et al. 1998). These new systems would require additional considerations similar to those presented here for the photo-anode.

5.8

CATHODE AND SOLUTION INTERFACE

While hydrogen evolution at the cathode was assumed not to be the rate limiting process, terms for the current density and potential drop at the cathode solution interface were required to solve for the ion distribution within the reactor. The current density at the cathode/solution interface was modeled with a simplified Buttler-Volmer relation (Tafel kinetics):
( )

Where was a fitted parameter and the remaining terms were the same as shown above for the anode/solution interface. Due to the low resistance metallic nature of the cathode, the potential was considered to be an arbitrary uniform value . However, the overpotential was variable and calculated through:

The cathode potential, was calculated so that the total current at the cathode was equal to the total current at the anode. Using the overpotential relation in the kinetic relationship: [( )( ) ] [( [( )( )( ) ) ] ]

The total current at the cathode was then: [( [( )( )( ) ) ] ]

Moving constant terms outside of the integral: [( )( ) ] [ ( )( ) ]

The cathode current was then related to the anode current: [ ( )( [( ) )( ] ) ]

The cathode potential could then be calculated directly as a function of the total anode current and the total effect from the thermodynamic potentials: ( ) * [ ( )( ) ] +

6 6.1

RESULTS IMPROVEMENTS OF THE ELECTROLYTE FLOW PROFILES

To ensure maximum activity across the entire surface of each electrode, a uniform flow pattern was desired. The electrolyte flow from the relatively small inlet regions (tubular inlet, with a 4 mm radius) needed

to be distributed across the 10cm X 10cm electrode. Without a suitable mechanism to spread the incoming flow, most of the flow would occur down the centerline of each channel, with the remaining regions becoming dead volume. The initial design called for a series of posts to be used in the entrance region to spread the flow. The resulting flow profiles for inlet rates of [ ] and [ ] are shown in Figure 9. At the upper flow rate (a reasonable rate for an electrochemical reactor of this sort), there is uneven flow within the reactor. Several methods were attempted to improve the flow situation, including the removal of the corners along the edges of the inlet region and the addition of extra posts. The final recommended design chosen for its improved profile and ease of modification (only a small amount of milling necessary) is shown in Figure 10. The flow past the top most post is directed towards the upper edge of the reactor, making the flow more even. However, there is still substantial spatial variation in the flow pattern. Also, there is a noticeable minimum in the profile along the centerline of the reactor (in both the original and improved designs). A small reduction in the diameter of the first center post would likely improve this situation. 1[ml/min] 300 [ml/min] ) | |
Figure 9: Flow distribution for the default reactor design, with a 1[ml/min] inlet flow (left) and 300[ml/min] flow (right). A series of posts at the inlet are used for flow distribution over the entire reactor. However, this design is insufficient at high flow rates and leads to non-uniform flows in the main channel region.

| |

Figure 10: Recommended modifications to improve flow profile at 300[ml/min]. Shown on the lower half of the channel is the profile of the original design, and along the upper half the flow with simple modifications (shaving part of the corners in the inlet region).

6.2

MAXIMUM SUSTAINABLE CURRENT BEFORE BUBBLE FORMATION

| |

| |

As described above, bubble formation will impede the operation of the reactor, and the model and experimental studies were run so that bubble formation was minimal. However, this places a limit on the average current density, which can be calculated through the numerical model. Calculated profiles are shown in Figure 11. Two methods were used to determine this limiting density. First, the mass flux of gaseous species at the outlet of the reactor was calculated assuming a saturated liquid at the electrode/solution interface along one side of a single channel. Secondly, the highest uniform flux was found so that no point along the reactor channel reached saturation concentrations. For uniform surface concentration, the maximum current density was calculated to be 0.78A. For uniform flux, the maximum current density was 0.64A.

(A)

(B)

(C)

Figure 11: Hydrogen concentration profiles for the determination of maximum current density before bubble formation. (A) shows the overall profile, with overlaid velocity profile. (B) shows the case of uniform saturated concentration. (C) shows that of uniform flux.

Since the diffusion coefficient and saturation concentration are both strongly dependent on temperature, the limiting current density should as well. Figure 12 shows the diffusion coefficient, saturation o concentration, and current density, all normalized to values at 25 C for a range of temperatures. The increase in the diffusion coefficient at higher temperature more than makes up for the corresponding drop in the saturation concentrations. Temperature changes will also have an effect on the surface kinetics and properties of the semiconductor, which could result in further improvements to the current flow.

Figure 12: Diffusion coefficient, saturation concentration, and maximum sustainable current density as a function of temperature. All quantities are scaled to the respective values at 25oC.

Transport calculations for the dissolved gas were extended to the full reactor geometry in the single chamber case (one flow section with both the anode and cathode). The results can be seen in Figure 13 The oxygen saturation near the surface (0.1 mm away) of the anode is shown for two different uniform current densities. At , fluid flows from the left with zero gas saturation, and gradually picks up oxygen until saturation conditions are barely reached in the upper region where flow is slower. Additionally, a

countercurrent is produced by the flow near the entrance, effectively creating a dead zone in the reactor. By , most of the electrode surface is saturated with oxygen, driving bubble formation everywhere. 1.0 Dissolved O2 Saturation(C/Csat) 0.0
Figure 13: Saturation of oxygen in the fluid flow 0.1mm from the surface of the photo-anode. Fluid flows from the left to the right along the streamlines with 100 ml/min average flow velocity. The inner region covers the surface of the electrode, and the region 1cm past the electrode in each direction is included. At a uniform current density of 0.9 A/m 2, saturation is reached (i.e. bubble formation) at the upper right corner and in the dead zone of the upper left corner. By 2.3 A/m 2, nearly the entire electrode surface is predicted to be saturated.

BIBLIOGRAPHY

Aarik, Jaan, Aleks Aidla, Alma-asta Kiisler, Teet Uustare, and Vainio Sammelselg. "Effect of crystal structure on optical properties of TiO2 films grown by atomic layer deposition." Thin Solid Films, 1997: 270-273. Albery, WJ, and P. Bartlett. "The Transport and Kinetics of Photogenerated Carriers in Colloidal Semiconductor Electrode Particles." Journal of the Electrochemical Society, 1984: 315-325. Albery, WJ, P., A. Hamnett Bartlett, and M. Dare-Edwards. "The Transport and Kinetics of Minority Carriers in Illuminated Semiconductor Electrodes." Journal of the Electrochemical Society, 1981: 1492-1501. Arthur Adamson, Alice Gast. Physical chemistry of surfaces. New York: Wiley, 1997. Basha, Vijayasekaran Boovaragavan and C. Ahmed. "A novel approach for computing tertiary current distributions based on simplifying assumptions." Journal of Applied Electrochemistry, 2006: 745-757. Bockris, J., and A. Reddy. Modern Electrochemistry 2B. New York: Kluwer Academic/Plenum Publishers, 1998. Bockris, J., and K. Uosaki. "Theoretical treatment of the photoelectrochemical production of hydrogen (semiconductor electrode for solar applications)." International Journal of Hydrogen Energy, 1977: 123-138. Cesar, I., K. Sivula, A. Kay, R. Zboril, and M. Gratzel. "Influence of Feature Size, Film Thickness, and Silicon Doping on the Performance of." Journal of Physical Chemistry C, 2009: 772-782. Chung, Meng-Hsuan. "A numerical method for analysis of tertiary current distribution in unsteady natural convection multi-ion deposition." Electochimica Acta, 2000: 3959-3972. CRC Press, Inc. CRC Handbook of Chemistry and Physics, 89th Edition. CRC Press, Inc., 2008-2009. Dare-Edwards, M., J. Goodenough, A. Hamnett, and P. Trevellick. "Electrochemistry and Photoelectrochemistry of Iron (III) Oxide." Journal of the Chemical Society, Faraday Transactions, 1983: 2027-2041. Fujishima, A., and K. Honda. "Electrochemical Photolysis of Water at a Semiconductor Electrode." Nature, 1971: 37-38. Gamelin, D., D. Zhong, J. Sun, and H. Inumaru. "Solar Water Oxidation by Composite Catalyst/a-Fe2O3 Photoanodes." Journal of the American Chemical Society, 2009: 6086-6087.

Kay, A., I. Cesar, and M. Gratzel. "New Benchmark for Water Photooxidation by Nanostructured a-Fe2O3 Films." Journal of the American Chemical Society, 2006: 15714-15721. Kennedy, J., and K. Frese. "Photooxidation of Water at a-Fe2O3 Electrodes." Journal of the Electrochemical Society, 1978: 709-714. Khan, S., and J. Bockris. "A Model for Electron Transfer at the Illuminated p-Type Semiconductor-Solution." Journal of Physical Chemistry, 1984: 2504-2515. Kleiman-Shwarsctein, A., Y. Hu, A. Forman, G. Stucky, and E. McFarland. "Electrodeposition of r-Fe2O3 Doped with Mo or Cr as Photoanodes for Photocatalytic Water Splitting." Journal of Physical Chemistry C, 2008: 15900-15907. Koelsch, M., S. Cassaignon, J.F. Guillemoles, and J.P. Jolivet. "Comparison of optical and electrochemical properties of anatase and brookite TiO2 synthesized by the sol-gel method." Thin Solid Films, 2002: 312-319. Licht, S., P. Ramakrishnan, O. Khaselev, T. Soga, and M. Umeno. "Multiple-Bandgap Photoelectrochemistry: Inverted Semiconductor Ohmic Regenerative Electrochemistry." The Journal of Physical Chemistry B, 1998: 2546-2554. Prasad, V., A. Karim, Z. Ulissi, M. Zagrobelny, and D. Vlachos. "High throughput multiscale modeling for design of experiments, catalysts, and reactors: Application to hydrogen production from ammonia." Chemical Engineering Science, 2009. Rangaraju, R., A. Panday, K. Raja, and M. Misra. "Nanostructured anodic iron oxide film as photoanode for water oxidation." Journal of Physics D: Applied Physics, 2009: 10 pp. Reyes-Coronado, D, G Rodriguez-Gattorno, M E Espinosa-Pesqueira, C Cab, R de Coss, and G Oskam. "Phasepure TiO2 nanoparticles: anatase, brookite, and rutile." Nanotechnology, 2008: 1-10. Sanchez, C., M. Hendewerk, K. Sieber, and G. Somorjai. "Synthesis, Bulk, and Surface Characterization of Niobium-Doped Fe2O3 Single Crystals." Journal of Solid State Chemistry, 1986: 47-55. Sanchez, H., H. Steinfink, and H. White. "Solid solubility of Ge, Si, and Mg in Fe2O3 and Photoelectric Behavior." Journal of Solid State Chemistry, 1982: 90-96. Sartoretti, C., M. Ulmann, B. Alexander, J. Augustynski, and A. Weidenkaff. "Photoelectrochemical oxidation of water at transparent ferric oxide film electrodes." Chemical Physics Letters, 2003: 194-200. Sheng, Xianliang, Jin Zhai, Lei Jiang, and Daoben Zhu. " Enhanced photoelectrochemical performance of ZnO photoanode with scattering hollow cavities ." Applied Physics A: Materials Science & Processing, 2009: 473479. Sivula, K., F. Formal, and M. Gratzel. "WO3Fe2O3 Photoanodes for Water Splitting: A Host Scaffold, Guest Absorber Approach." Chemistry of Materials, 2009: 28622867. Sumita, Taishi, Tetsuya Yamaki, Shunya Yamamato, and Atsumi Miyashita. "Photo-induced surface charge separation of highly oriented TiO2 anatase and rutile thin films." Applied Surface Science, 2002: 21-26. Thampi, K., J. Kiwi, and M. Grtzel. "Methanation and photo-methanation of carbon dioxide at room temperature and atmospheric pressure." Nature, 1987: 506-508. Uosaki, K., and H. Kita. "Mechanistic Study of Photoelectrochemical Reactions at a P-GaP Electrode." Journal of the Electrochemical Society, 1981: 2153-2158.

Valds, A., Z. Qu, G. Kroes, J. Rossmeisl, and J. Nrskov. "Oxidation and Photo-Oxidation of Water on TiO2 Surface." The journal of physical chemistry C, 2008: 9872-9879. Vogt. Comprehensive Treatise on Electrochemistry, Vol 6. Plenum, 1980. Wang, Hyunchul Ju and Chao-Yang. "Experimental Validation of a PEM Fuel Cell Model by Current Distribution Data." Journal of the Electrochemical Society, 2004: 1954-1960.

You might also like