water research 44 (2010) 1797–1808

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Modeling the UV/hydrogen peroxide advanced oxidation

process using computational fluid dynamics

Scott M. Alpert a,b,*, Detlef R.U. Knappe b, Joel J. Ducoste b

a
Hazen and Sawyer, P.C., 4944 Parkway Plaza Blvd., Suite 375, Charlotte, NC 28217, USA
b
Department of Civil, Construction and Environmental Engineering, North Carolina State University, Mann Hall,
Box 7908, Raleigh, NC 27695, USA

article info abstract

Article history: The use of numerical models for the design and optimization of UV/H2O2 systems must
Received 31 May 2009 incorporate both reactor design (hydrodynamics, lamp orientation) and chemical kinetics
Received in revised form (reaction mechanisms, kinetic rate constants). This study was conducted to evaluate the
1 November 2009 performance of comprehensive CFD/UV/AOP models for the degradation of an indicator
Accepted 2 December 2009 organic contaminant. The combination of turbulence sub-models, fluence rate sub-models,
Available online 21 December 2009 and kinetic rate equations resulted in a comprehensive and flexible design tool for
predicting the effluent chemical composition from a UV-initiated AOP reactor. The CFD
Keywords: model tended to under predict the percent removal of methylene blue compared to pilot
Ultraviolet reactor trials under the same operating conditions. In addition, the percent difference
Advanced oxidation between the pilot and the CFD results increased with increasing flow rates. The MSSS
Methylene blue fluence rate sub-model predicted higher contaminant removal values than the RAD-LSI
Emerging contaminants sub-model while the different two-equation turbulence sub-models did not significantly
impact the predicted removal for methylene blue in the tested reactor configuration. The
overall degradation of methylene blue was a strong function of the second-order kinetic
rate constant describing the reaction between methylene blue and the hydroxyl radical. In
addition, the removal of methylene blue was sensitive to the concentration of dissolved
organic carbon in the water matrix since DOC acts as a scavenger of hydroxyl radicals.
ª 2009 Elsevier Ltd. All rights reserved.

1. Introduction pipe and fittings, lamp number, and lamp orientation) and
chemical kinetics (reaction mechanisms and kinetic rate
The use of ultraviolet radiation-initiated (UV-initiated) constants). While some numerical techniques have been
advanced oxidation processes (AOP) is rapidly becoming an developed for understanding UV AOP performance, these
attractive alternative for the degradation of emerging organic techniques are limited in their applicability for analyzing full-
contaminants that are not easily removed using conventional scale UV/AOP systems because of the need to combine
water treatment processes. Of the available UV-initiated hydrodynamic reactor models, fluence rate models, and
AOPs, UV/hydrogen peroxide (H2O2) is one of the more chemical kinetic models (Sharpless and Linden, 2003; Pareek
promising technologies. Design of UV/H2O2 systems requires et al., 2003; Crittenden et al., 1999). As a result, water profes-
knowledge about both system configuration (reactor design, sionals need more appropriate numerical tools to use as part

* Corresponding author at: Hazen and Sawyer, P.C., 4944 Parkway Plaza Blvd., Suite 375, Charlotte, NC 28217, USA. Tel.: þ1 704 357 3150;
fax: þ1 704 357 3152.
E-mail addresses: salpert@hazenandsawyer.com (S.M. Alpert), knappe@eos.ncsu.edu (D.R.U. Knappe), jducoste@eos.ncsu.edu
(J.J. Ducoste).
0043-1354/$ – see front matter ª 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.watres.2009.12.003
1798 water research 44 (2010) 1797–1808

of the design process and for optimizing UV/AOP systems. The recommendations by Sharpless and Linden (2003) was that
objective of this research was to evaluate the sensitivity of a detailed hydrodynamic model coupled with a UV irradiance
a UV/H2O2 CFD model performance to numerical parameters distribution profile be developed to examine the effect of
(fluence rate sub-model, turbulence sub-model, and kinetic optical path length and multiple-lamp reactors for UV/H2O2
rate constants) and to system characteristics (flow rate, systems.
hydrogen peroxide dose, and radical scavengers). Computational fluid dynamics (CFD) is a technique for
The reaction mechanisms for the degradation of organic numerically solving the fluid dynamics equations through
contaminants by UV-initiated AOPs typically consist of space and time, including the conservation of mass, conser-
a complex chain of fast chemical reactions. As such, the vation of momentum, and conservation of energy. Combined
resulting intermediates and products from these processes with appropriate boundary and initial conditions, these gov-
may be highly sensitive not only to the UV fluence rate within erning equations can describe both the physical and chemical
the reactor but also the level of turbulence and mixing (Bal- changes within a reactor. A CFD model for UV systems also
dyga and Bourne, 1999; Spalding, 1998; Marchisio and Barresi, includes the spatial variations of fluence rate within the UV
2003). Design factors, such as upstream hydraulic configura- reactor.
tions, internal reactor layout, and lamp arrangement, may Limited studies have been performed with CFD to simulate
influence process performance (Ducoste et al., 2005; Ducoste UV/AOPs. Pareek et al. (2003) used CFD combined with
and Linden, 2005; Liu et al., 2007). Water quality effects, a discrete-ordinate radiation transport equation for UV
including the impact of UV absorption and radical scavengers, intensity and modified k-3 turbulence equations to model
will determine the size of the treatment system and the a heterogeneous, multi-phase photocatalytic reactor system
appropriate placement in the treatment scheme (Linden et al., for the photodegradation of a spent Bayer liquor. However, the
2004). applied radiation model did not incorporate refraction and
Researchers have previously demonstrated the importance was used to describe a simple bench scale reactor. Mohseni
of combining UV reactor hydrodynamics with fluence rate and Taghipour (2004) used CFD to evaluate the heterogeneous
models to predict the effectiveness of the disinfection process. photocatalytic oxidation of gas-phase vinyl chloride by the UV
Studying the UV-initiated advanced oxidation of organic irradiation of a TiO2-coated surface; however, this research
contaminants, Linden et al. (2004) recognized the importance did not incorporate a fluence rate model and only described
of non-ideal reactor characteristics (hydrodynamics and flu- the flux of VC toward the TiO2-coated surface. Based on the
ence rate) on the overall AOP performance. Sharpless and authors’ literature review, no published research was found
Linden (2003) concluded that the development of a predictive that investigated the sensitivity of both the turbulence model
UV/AOP model that incorporates reactor hydrodynamics selection and the use of more rigorous fluence rate models on
would allow design simulations that optimize lamp place- UV/AOP simulations.
ment, minimize UV screening, and improve prediction of Typically, full-scale UV photoreactors will operate in
contaminant removal in different UV reactors. Thus, an a turbulent flow regime, and thus, simulations of a UV process
effective CFD simulation must include turbulence models, must incorporate the impact of turbulent mixing on any
fluence rate models, and accurate reaction mechanisms chemical reactions that occur within these UV photoreactors.
describing the oxidation of the contaminant. Although the standard k-3 turbulence model is often used for
Many studies have been performed to evaluate the trans- the examination of hydrodynamics within a flow-through
formation of organic contaminants in UV-initiated advanced reactor, the incorporation of other turbulence models may
oxidation processes (e.g., Sharpless and Linden, 2003; Bali provide better characterization of the turbulent behavior
et al., 2003; Devlin and Harris, 1984; Scheck and Frimmel, within an advanced oxidation reactor. One of the goals of this
1995). Detailed experiments have been performed in several study was to provide better guidance for the selection of
studies to ascertain the reaction pathways involved in the turbulence models for UV system analysis. Liu et al. (2007)
degradation of the parent compound to its intermediate and were able to evaluate the accuracy of the turbulence model
final products (Stefan et al., 1996, 2000; Alnaizy and Akger- with experimental fluid mechanics measurements and were
man, 2000). The determination of these reaction pathways able to assess the impact of the turbulence model selection on
provides a more detailed picture of the photo-oxidation UV disinfection performance. While their results showed
process and allows the development of numerical kinetic a slight sensitivity of the microbial inactivation to the turbu-
models that predict the rate of conversion of reactants to lence model selection, the analysis revealed that the sensi-
products. Detailed reaction pathways are specific to the tivity of microbial inactivation to the turbulence model
parent compound and chemical constituents in the water. selection was a function of the UV operating conditions and
However, once the reaction pathway is defined, researchers the UV response kinetics of the target microorganisms. A
can combine the reaction kinetics of these UV-initiated AOPs similar sensitivity study needs to be completed for UV-initi-
with simulation models that describe the irradiance distri- ated AOPs since the proper characterization of the turbulent
bution profiles and the hydrodynamic and turbulent charac- mixing intensity may be critical for fast competitive chemical
teristics of a UV reactor system. Sharpless and Linden (2003) reactions typically associated with UV/AOPs.
developed an NDMA degradation model that assumed ideal UV fluence rate can be defined as the total radiant power
mixing for a batch reactor. As a result, the model could not incident from all directions onto an infinitesimally small
account for variations in the UV fluence rate distribution or sphere of cross-sectional area dA, divided by dA (Bolton, 2001).
variations in fluence distributions due to different fluid The UV fluence rate varies with distance from the lamp and is
element paths within a continuous flow reactor. One of the a function of the absorptive characteristics of the media
 water research 44 (2010) 1797–1808 1799

through which the UV radiation passes. In advanced oxidation 1.563  106 M and the inlet hydrogen peroxide concentration
reactions, both the organic contaminant and H2O2 absorb UV was set equal to the inlet methylene blue concentration (M)
radiation, reducing the transmitted fluence rate, and conse- multiplied by a ratio ranging from 139 to 267. Since PHOENICS
quently, changing the effectiveness of hydroxyl radical uses a structured rectangular grid, the grid spacing in all three
production throughout the UV reactor. Other factors, such as dimensions was manually created.
reflection, refraction, shadowing, and lamp characteristics, To determine grid independence in this study, several grid
influence the spatial distribution of fluence rate within a UV densities were created that maintained fine grid spacing
reactor. Several models have been developed to characterize within the reactor in areas of expected high gradients and that
the spatial distribution of the UV fluence rate. These models were symmetrical about the UV lamp inside the reactor. Based
include the Multiple Point Source Summation (MPSS) model on preliminary results, two grid systems that differed in the
(Jacob and Dranoff, 1970; Bolton, 2000); the Line Source Inte- number of cells in each direction by approximately 10% (total
gration (LSI) model (Blatchley, 1997); the Multi-Segment cell count of 1,154,688 versus a total of 859,154 cells across the
Source Summation (MSSS) model (Liu et al., 2004); the modi- domain) were selected for evaluation of grid independence.
fied LSI model (RAD-LSI) (Liu et al., 2004); and the radiative The coarser grid was verified as sufficient to achieve grid
transfer equation using the Discrete Ordinate (DO) method independence, while minimizing run time and memory
(Fiveland, 1984; Stamnes et al., 1988; Liou and Wu, 1996). Liu requirements, since the change in average species concen-
et al. (2004) directly evaluated the performance of these trations, velocity, and turbulence quantities changed by no
models with experimental measurements of the fluence rate more than 1% as a function of the grid refinement.
at different points within a UV reactor, the results being In this study, the turbulent flow in the UV reactor was
a compilation of the strengths and weaknesses of each mod- simulated using three two-equation turbulence models
el’s ability to predict the fluence rate. Ducoste and coworkers (standard k-3, RNG k-3, and k-u) to assess the sensitivity of
further showed that the microbial log inactivation and the turbulence model selection on the indicator compound
shape of the fluence distribution were sensitive to the fluence degradation. While other more complicated turbulence
rate model selection (Ducoste et al., 2005; Liu et al., 2007). models exist, such as the Reynolds Stress Model or Large Eddy
Their results suggest the need to evaluate the influence of the Simulation model, the three two-equation models selected for
fluence rate model selection on any photoreactive process this research provide reasonable and stable results without
such as UV-initiated AOPs. being numerically intensive. The standard k-3 model used in
this study was that proposed by Launder and Sharma (1974)
(as cited in Wilcox, 2004) to solve the turbulence stress closure
2. Materials and methods problem. A second closure model, the Renormalized Group
(RNG) k-3 model, was developed by Yakhot and Orszag (1986)
2.1. Numerical model development using techniques from the renormalization group theory. The
third model, the k-u model, uses transport equations for k and
A cylindrical stainless steel UV reactor that was approxi- the Reynolds Mean Stress (RMS) fluctuating vorticity u. The
mately 15.2 cm (6 inches) in diameter and 182.9 cm (6 feet) in standard k-3 model is the most popular of the two-equation
length and contained one low-pressure UV lamp installed closure models and performs relatively well for most types of
axially was used in the pilot experiments. A three-dimen- flows. The RNG k-3 model tries to solve the k-3 singularity
sional solid model of the UV pilot reactor was created in problem at wall boundaries (where k approaches zero) and
AutoCAD 2004 (Autodesk, Inc., San Rafael, CA) and included reduces the higher level of dissipation that is predicted using
geometric elements for the reactor, inlet and outlet piping, the the standard k-3 model. However, the RNG k-3 model is not as
UV lamp/quartz sleeve assembly with wiper mechanism, five accurate in predicting free-shear (no wall) flows. The k-u
internal baffle plates, and the inlet and outlet planes. All model has been proven relatively accurate for boundary layer
dimensions were measured directly on the pilot reactor and (wall-bounded) flows, and, with the 1998 revisions, also works
translated as full-scale into the AutoCAD model. This model well for free-shear flows (Wilcox, 2004). The equations for
was then imported into the CFD software PHOENICS (CHAM, each of the turbulence sub-models (k-3, RNG k-3, and k-u) are
Ltd, Wimbledon, England) and appropriate boundary condi- included within the PHOENICS library and were selected
tions for inlet and outlet velocities, pressures, and turbulence through a user interface menu.
conditions were set. The inlet velocity normal to the inlet In this study, the RAD-LSI and MSSS fluence rate sub-
plane ranged from 0.064 to 0.384 m s1, which corresponded to models were incorporated into the CFD UV/AOP model to
a flow rate of 5 to 30 GPM (1.1–6.8 m3 h1). All inlet velocities simulate the spatial variation of UV fluence rate inside the
tangential to the inlet plane were set to zero. No external reactor (i.e., the fluence rate distribution). The RAD-LSI is
velocity restrictions were placed on the outlet and the gradi- a modified version of the line source integration model
ents of all variables, except for pressure, were set to zero in the (Blatchley, 1997) that was enhanced to account for the physics
flow direction at the outlet. The external pressure at the outlet of reflection at the lamp surface, refraction (bending only),
was zero gauge (100,000 Pa). The turbulent kinetic energy and and absorption, and provides better prediction of the fluence
energy dissipation rate inlet conditions are defined in rate near the lamp surface by using the radial model compo-
PHOENICS as kinlet ¼ ðI  UÞ2 and einlet ¼ ðkinlet Þ1:5 ð0:1  DÞ1 nent (Liu et al., 2004). Although the RAD-LSI is an improved
where U is the normal average inlet velocity and D is the pipe version of the LSI model, it is still based in part on the multiple
diameter. The inlet turbulence condition (I) was set equal to point source summation method (MPSS), which models
5%. The inlet methylene blue concentration equaled a lamp with a series of point sources. As discussed in Liu et al.
1800 water research 44 (2010) 1797–1808

Table 1 – Kinetic equations.

Reaction system Equations Constants (references)

1 H2 O2 þ hn/2,OH FH2 O2 ¼ 0:5

rUV ¼ 2FH2 O2 E0CFD 3H2 O2 ½H2 O2  3H2 O2 ¼ 19:6 M1 cm1
1
IUV ðW m2 Þ  100 ðcm m1 Þ U254 ¼ 471; 528 J Ein
E0CFD ¼ 1
U254 ðJ Ein Þ  1000 ðL m3 Þ

2 H2 O2 þ ,OH/H2 O þ HO2 , ð/Hþ þ ,O

2Þ k1 ¼ 2:7  107 M1 s1
(Buxton et al., 1988)
d½H2 O2 
¼ k1 ½H2 O2 ½,OH
dt
d½,OH
¼ k1 ½H2 O2 ½,OH
dt
d½,O2 
¼ k1 ½H2 O2 ½,OH
dt

3 H2 O2 þ ,O
2 /,OH þ O2 þ OH

k2 ¼ 0:13 M1 s1
(Weinstein and Bielski, 1979)
d½H2 O2  
¼ k2 ½H2 O2 ½,O 2
dt
 
d½,O2 
¼ k2 ½H2 O2 ½,O 2
dt
d½,OH 
¼ k2 ½H2 O2 ½,O 2
dt

4 ,OH þ ,OH/H2 O2 k3 ¼ 5:5  109 M1 s1

(Buxton et al., 1988)
d½,OH
¼ k3 ½,OH½,OH
dt
d½H2 O2 
¼ k3 ½,OH½,OH
dt

5 ,OH þ ,O2 /O2 þ OH



d½,OH 
¼ k4 ½,OH ½,O
2 k4 ¼ 7:0  109 M1 s1
dt
(Beck, 1969 as cited in
Crittenden et al., 1999)
d½,O2

¼ k4 ½,OH ½,O
2
dt

6 ,OH þ MB/Products kMB;$ OH ¼ 6:9  1010 M1 s1

d½,OH
¼ kMB;$ OH ½,OH½MB
dt
d½MB
¼ kMB;, OH ½,OH½MB
dt
fMB ¼ Methylene Blueg

7 DOC þ ,OH/Products kDOC;$OH ¼ 2:5  104 ðmg

L Þ
1
s1
(Larson and Zepp, 1988)
d½,OH
¼ kDOC;$OH ½,OHðDOCÞ
dt

8 HCO
3 þ ,OH/CO3 þ H2 O
$
kHCO3 ;$OH ¼ 8:5  106 M1 s1
(Buxton et al., 1988)
d½,OH 
¼ kHCO3 ;$OH ½,OH ½HCO
3
dt

9 NH2 Cl þ ,OH/,NHCl þ H2 O or NH2 Cl þ ,OH/,NH2 þ HOCl kNH2 Cl;$OH ¼ ð2:8  0:2Þ  109 M1 s1
(Johnson et al., 2002)
d½,OH
¼ kNH2 Cl;$OH ½,OH½NH2 Cl
dt
d½,OH 
Composite ¼ 2FH2 O2 3H2 O2 ½H2 O2 E0CFD  k1 ½H2 O2 ½,OH
dt     
þk2 ½H2 O2 ½,O2 k3 ½,OH ½,OH k4 ½,OH ½,O

2
 
kMB;$OH ½,OH ½MB kDOC;$OH ½,OHðDOCÞ
 
kHCO3 ;$OH ½,OH ½HCO kNH2 Cl;$OH ½,OH½NH2 Cl
 3
kOCl ;$OH ½,OH ½OCl kHOCl;$OH ½,OH½HOCl
d½,O2
    
¼ k1 ½H2 O2 ½,OH k2 ½H2 O2 ½,O 
2 k4 ½,OH ½,O2 
dt
d½H2 O2 
¼ FH2 O2 E0CFD 3H2 O2 ½H2 O2 k1 ½H2 O2 ½,OH
dt 
k2 ½H2 O2 ½,O 2 þk3 ½,OH½,OH

d½MB
¼ kMB;$OH ½MB½,OH
dt
 water research 44 (2010) 1797–1808 1801

(2004), the MPSS model with reflection, refraction, and system 1 in Table 1, which correlates the fluence rate values
absorption did not completely correct for the over-prediction calculated at each grid cell within the reactor (IUV) to the rate
of the fluence rate near the lamp surface and in regions near of direct photolysis of hydrogen peroxide.
the lamp ends. This over-prediction was corrected by Since the hydroxyl radical is very reactive, it will begin to
modeling the lamp as a series of differential cylindrical react immediately with other species in solution. The reaction
segments, where UV radiation is emitted normal to the between hydrogen peroxide and the hydroxyl radical
cylinder surface and decreases with the cosine of the angle produces the superoxide anion radical as described by reac-
between the unit normal and the direction vectors (Liu et al., tion system 2 in Table 1. Hydrogen peroxide reacts with the
2004). This fluence rate modeling approach is called the superoxide anion radical generating a hydroxyl radical
multiple segment source summation (MSSS), as described according to reaction system 3. Reaction system 4 describes
with more detail in Liu et al. (2004). Codes previously created the hydroxyl radical recombining with itself to form hydrogen
by this research group (Liu et al., 2004) for these two sub- peroxide. The two radical species, OH and O 2 , will also
models were modified as appropriate for the lamp character- combine with each other as shown in reaction system 5.
istics being evaluated in this research and called into the CFD Reaction system 6 in Table 1 describes methylene blue (MB)
software through user-created FORTRAN sub-routines in the oxidation by the hydroxyl radical. The reaction rate constant
PHOENICS environment. for methylene blue was determined by a competition kinetics
Methylene blue is a dye that was used as the indicator process. Several hydroxyl radical scavengers were present in
organic contaminant in this study. Many organic dyes are the background water matrix. These scavengers, including
decolorized (bleached) upon reaction with the hydroxyl radical. dissolved organic carbon (DOC), alkalinity (which at neutral
Dyes are good candidates for evaluation in an advanced pH is dominated by the bicarbonate ion), and chloramines (for
oxidation study since the decolorization can be analyzed online the water used in this research), will consume radicals
with a spectrophotometer. In addition, these dyes typically do according to reaction systems 7–9 in Table 1.
not undergo direct photolysis and are not reactive with The system of equations describing the generation of
hydrogen peroxide alone. The mechanism of decolorization hydroxyl radicals and the degradation of methylene blue (as
varies among the types of organic dyes, but, in general, these described in Table 1) was created in CFD using the PHOENICS
dyes contain aromatic ring structures that are subject to attack Input Language and included within the text file read by the
by the hydroxyl radical. The system of kinetic rate equations software upon the start of each run. The resulting equation
describing the production of the hydroxyl radical from for each of the species is shown at the end of Table 1. Two
hydrogen peroxide and the corresponding degradation of modifications to the reaction equations were made to
methylene blue is shown in Table 1. The photolysis of hydrogen improve convergence within the CFD code. First, the four
peroxide is shown as reaction system 1 in Table 1. species being tracked (methylene blue, hydrogen peroxide,
In Table 1, FH2 O2 is the quantum yield for the photolysis of the hydroxyl radical, and the superoxide anion radical) were
hydrogen peroxide (moles H2O2 per mole photons, or moles normalized to initial hydrogen peroxide or methylene blue
H2O2 per Einstein) and 3H2 O2 is the molar absorptivity of conditions. Second, pseudo-steady-state conditions for the
hydrogen peroxide (L mol1 cm1). The square brackets [ ] two radicals were assumed since the lifetimes of the two
represent the molar concentration of the species enclosed in radical species are very short. The CFD model results were
the brackets. The irradiance IUV at each individual grid cell considered converged to a stable solution when the net sum
centerpoint in the CFD domain was converted to a volumetric of the differences between the inlet and outlet volume frac-
photon flux term E0CFD as shown in Table 1. The initiation step tions (i.e., conservation of mass) was on the order of 103 or
for hydroxyl radical production is described by reaction below.

Spectrophotometer
Probe

Pressure Gauge (typ)

Peristaltic Pump (typ)

UV Reactor
Methylene Blue

H2O2 Sample Port (typ)

FI Propeller
Flow Meter

Centrifugal Submersible Pump

(in sub-floor storage tank)
GAC-treated tap water

Fig. 1 – Low-pressure UV pilot system schematic.

1802 water research 44 (2010) 1797–1808

Fig. 2 – Velocity distribution in reactor with five baffles and flow [ 20 GPM.

2.2. Experimental methods For the pilot-scale experiments, dry methylene blue (Alfa
Aesar, Acros Organics) and hydrogen peroxide solution (30% w/
The CFD results were validated with pilot-scale experiments. v, Fisher Scientific, Pittsburgh, PA) were used as purchased
A schematic of the low-pressure UV pilot-scale system is without further purification. The methylene blue and hydrogen
presented in Fig. 1. The UV pilot reactor was provided by peroxide were diluted separately with ultrapure laboratory
Degremont’s North American Research and Development water (Dracor High Purity Water Systems) (herein referred to as
Center (DENARD). A low-pressure UV lamp (Ondeo Degre- DI water). Reagents used for analyses, including potassium
mont, No. 61645-G01, 52 W nameplate output at 254 nm) was iodide, potassium iodate, sodium hydroxide, ammonium
installed axially along the length of the reactor. Five remov- molybdate tetrahydrate, and potassium hydrogen phthalate,
able perforated baffle plates were installed symmetrically were acquired as ACS-grade or similar solids from Fisher
about the center of the reactor. Sampling valves were installed Scientific (Pittsburgh, PA). Pilot influent water was pumped
within 0.9 m of the influent and effluent of the reactor. Two from a 10,000-gallon concrete reservoir filled with GAC-filtered
ball valves located upstream and downstream of the reactor (NORIT GAC400) municipal tap water from Raleigh, NC.
were used to throttle flow between 1.1 and 11.4 m3 h1 (5 and The methylene blue concentration in the pilot reactor
50 gallons per minute (GPM)). Two diffuser mechanisms were effluent was measured spectrophotometrically at a wave-
installed upstream of the reactor through which methylene length of 661 nm (Cary 50, Varian, Inc., Walnut Creek, CA)
blue and hydrogen peroxide solutions were injected into the using a fiber-optic spectrophotometric probe (C Technologies,
system. Flow rates of the contaminant and hydrogen peroxide Inc., Bridgewater, NJ) inserted into the reactor effluent pipe.
solutions were controlled by peristaltic pumps. The diffusers Total chlorine concentration of the pilot reactor influent was
were located immediately downstream of the flow regulating tested using a Hach Pocket Colorimeter. Total alkalinity was
valve, the turbulence from which provided additional mixing measured using Standard Method 2320 (APHA 1995).
of the chemicals prior to the reactor influent. Similar to the Hydrogen peroxide residual concentrations were measured
CFD models, the pilot system was used to measure methylene using the method described by Klassen et al. (1994). Total
blue degradation with either five internal baffles or with one organic carbon (TOC) was measured by the Town of Cary (NC)
internal baffle in the UV reactor. water quality laboratory using a Tekmar-Dohrmann Phoenix

Fig. 3 – Graphical results of RAD-LSI fluence rate sub-model in CFD.

 water research 44 (2010) 1797–1808 1803

Fig. 4 – Representative results at 20 GPM (4.5 m3 hL1) and 10.6 mg LL1 H2O2 of (a) pilot data and (b) CFD simulation (C6 equals
the normalized methylene blue concentration [MB]/[MB]0).

8000 analyzer. The pH of the water matrix for each pilot run concentrations of isopropyl alcohol (IPA) were added as
was measured using a Thermo Orion Model 410 pH meter a radical scavenger to crystallization dishes containing
calibrated weekly using a two-point standard. UV trans- aqueous solutions of methylene blue and hydrogen peroxide.
mittance at 254 nm (UVT254) was measured in a 1-cm path- The crystallization dishes were exposed to low-pressure UV
length quartz cuvette (Varian Cary 50 spectrophotometer radiation in a collimated beam apparatus. By varying the
zeroed with DI water). initial concentration of IPA and measuring the methylene blue
To determine the second-order rate constant for methy- concentration before and after exposure, a graph of the
lene blue, a competition kinetics method described by Rose- inverse of the pseudo-first-order rate constant ðk0MB Þ1 as
nfeldt and Linden (2004) was used. In this method, known a function of the initial IPA concentration was prepared. The

100

90
Percent Removal (%) of Methylene Blue

80

70

60

50

40

30

20
Flow = 5 Flow = 10 Flow = 20 Flow = 30 Flow = 20 Flow = 20 Flow = 20
{1.1} {2.3} {4.5} {6.8} {4.5} {4.5} {4.5}
MR 139.2 MR 150.3 MR 150 MR 144.2 MR 150 MR 199.7 MR 266.4
Pilot 92.2 76.5 57.7 38.7 57.7 64.1 72.0
CFD 91.8 78.1 47.1 29.9 47.1 58.6 62.4
Percent Difference 0.4% -2.1% 18.4% 22.7% 18.4% 8.6% 13.3%

Flow (GPM) {m3/hr} and [H2O2]:[MB] Molar Ratio (MR) Conditions

Fig. 5 – Comparison of pilot and CFD results for five-baffle trials.

1804 water research 44 (2010) 1797–1808

100

90

Percent Removal (%) of Methylene Blue

80

70

60

50

40

30

20
Flow = 10 {2.3} Flow = 20 {4.5} Flow = 20 {4.5} Flow = 20 {4.5}
MR 149.3 MR 149.3 MR 149.3 MR 200.3
Pilot 79.9 59.7 59.7 66.2
CFD 77.3 46.0 46.0 57.1
Percent Difference 3.3% 22.9% 22.9% 13.7%

Flow (GPM) {m3/hr} and [H2O2]:[MB] Molar Ratio (MR) Conditions

Fig. 6 – Comparison of pilot and CFD results for one-baffle trials.

slope and y-intercept of the line of best fit was used to deter- calculated values for the rate constant are high enough to be
mine the second-order rate constant for the reaction between in the range for which reactions may become diffusion
methylene blue and the hydroxyl radical (kMB,OH). limited. A diffusion limited reaction is one in which the
reaction rate is limited by the ability of diffusion to transfer
one reactant to the second reactant.
3. Results During this research, 8 combinations of flow rate and
hydrogen peroxide dose were evaluated using methylene blue
The results of the four replicate competition kinetics experi- in the pilot-scale reactor and the CFD model. Typical CFD results
ments using isopropyl alcohol (IPA) as the radical scavenger for velocity and fluence rate distribution through the reactor are
produced second-order rate constants for the reaction shown in Fig. 2 and Fig. 3, respectively. Representative data for
between methylene blue and the hydroxyl radical ranging the experiments conducted with a flow rate of 20 GPM (4.5 m3
from 4.40  1010 to 8.72  1010 M1 s1, with an average of h1) and an H2O2 concentration of 10.6 mg L1 are shown in
6.89  1010 M1 s1 and a 95% confidence interval of Fig. 4. The pilot and CFD results for all of the trials in which five
2.05  1010 M1 s1. Since the variation in rate constant values baffles were in the reactor are shown in Fig. 5. Similar data are
may be considered significant, as represented by the 95% compared for the trials with only one baffle in the reactor in
confidence interval, the sensitivity of the overall degradation Fig. 6. The CFD simulations for which the results are shown in
of methylene blue in the system being modeled was deter- Figs. 2–6 were conducted with the standard k-3 turbulence sub-
mined using CFD. In addition, it should be noted that the model and the RAD-LSI fluence rate sub-model.

Table 2 – Impact of turbulence sub-model on CFD results.

Turbulence model
a a
k-3 RNG k-3 k-ua k-3b RNG k-3b

Flow rate (GPM) [m3 h1] 20 [4.5] 20 [4.5] 20 [4.5] 30 [6.8] 30 [6.8]
Fraction of initial MB remaining 0.440 0.442 0.431 0.701 0.701
Methylene blue conversion (%) 56.0 55.8 56.9 29.9 29.9

MB ¼ Methylene Blue;
For all models: Initial MB Concentration ¼ 0.50 mg L1, Number of Baffles in Reactor ¼ 5, DOC ¼ 0.57 mg L1, Fluence Rate Sub-model ¼ RAD-LSI,
kMB,OH ¼ 6.9  1010 M1 s1.
a Molar Ratio of H2O2 to Initial MB ¼ 199.25, H2O2 Concentration ¼ 10.6 mg L1, Power Output at Lamp Surface ¼ 26.5 W, Fluence Rate at
Sensor ¼ 18.0 W m2, UVT254 ¼ 94.9%, pH ¼ 7.24, Alkalinity ¼ 22.2 mg L1 as CaCO3, Total Chlorine Residual ¼ 0.01 mg L1 as Cl2.
b Molar Ratio of H2O2 to Initial MB ¼ 144.2, H2O2 Concentration ¼ 7.66 mg L1, Power Output at Lamp Surface ¼ 30.2 W, Fluence Rate at
Sensor ¼ 15.8 W m2, UVT254 ¼ 91.9%, pH ¼ 7.49, Alkalinity ¼ 28.1 mg L1 as CaCO3, Total Chlorine Residual ¼ 0.14 mg L1 as Cl2.
 water research 44 (2010) 1797–1808 1805

Table 3 – Impact of fluence rate model on CFD results. Table 5 – Impact of DOC on CFD results.
Fluence rate sub-model DOC (mg L1)

RAD-LSI MSSS 0.0 0.5 1.0 2.0

2
Fluence rate at sensor (W m ) 18.0 24.0 Fraction of initial MB remaining 0.059 0.133 0.197 0.300
Fraction of initial MB remaining 0.440 0.386 Methylene blue conversion (%) 94.1 86.7 80.3 70.0
Methylene blue conversion (%) 56.0 61.4
MB ¼ Methylene Blue.
MB ¼ Methylene Blue. For all models: Flow Rate ¼ 10 gal min1 [2.3 m3 hr1], Initial MB
For all models: Flow Rate ¼ 20 gal min1 [4.5 m3 hr1], Initial MB Concentration ¼ 0.50 mg L1, Molar Ratio of H2O2 to Initial
Concentration ¼ 0.50 mg L1, Molar Ratio of H2O2 to Initial MB ¼ 150.0, H2O2 Concentration ¼ 7.98 mg L1, Power Output at
MB ¼ 199.25, H2O2 Concentration ¼ 10.6 mg L1, Power Output at Lamp Surface ¼ 31.6 W, Fluence Rate at Sensor ¼ 21.5 W m2,
Lamp Surface ¼ 26.5 W, UVT254 ¼ 94.9%, pH ¼ 7.24, Alkalin- UVT254 ¼ 94.9%, pH ¼ 7.25, Alkalinity ¼ 23.0 mg L1 as CaCO3, Total
ity ¼ 22.2 mg L1 as CaCO3, Total Chlorine Residual ¼ 0.01 mg L1 as Chlorine Residual ¼ 0.0 mg L1 as Cl2, Number of Baffles in Reac-
Cl2, Number of Baffles in Reactor ¼ 5, DOC ¼ 0.57 mg L1, Turbu- tor ¼ 5, Turbulence Sub-model ¼ Standard k-3, Fluence Rate Sub-
lence Sub-model ¼ Standard k-3, kMB,OH ¼ 6.9  1010 M1 s1. model ¼ RAD-LSI, kMB,OH ¼ 6.9  1010 M1 s1.

The three turbulence closure sub-models were evaluated

to determine the effect, if any, on the overall degradation
prediction of the advanced oxidation of the methylene blue. the reactor. While the low flow conditions produced good
Table 2 displays the methylene blue conversion percentages agreement between pilot and CFD results, the CFD-predicted
obtained when using the same experimental parameters removal was 8.6% to 22.7% less than that achieved in the pilot
with three turbulence sub-models at two different flow rates. for the higher flows. One potential explanation for this
Table 3 displays the effects of the fluence rate sub-model difference could be an over-prediction of radical scavenger
selection on the predicted methylene blue conversion effects in the model. More hydroxyl radicals would be avail-
percentage in the UV/H2O2 reactor. As shown in Table 3, the able for reaction if the DOC concentration, which will later be
MSSS sub-model predicted a higher fluence rate at the loca- shown to significantly impact the degradation of methylene
tion of the pilot reactor sensor than the RAD-LSI sub-model, blue, was lower in any individual pilot trial than that modeled.
which translated into a higher conversion percentage pre- The second trend in Fig. 5, as displayed for the data points
dicted for methylene blue. on the left side of the figure, is that the percent difference
To evaluate the model sensitivity to the second-order between the pilot and the CFD results increased with
kinetic rate constant for the reaction between the hydroxyl increasing flow rates. This trend suggests that there may be
radical and methylene blue, several values of (kMB,OH) were hydrodynamic effects that were not being considered. For
used in separate CFD models, including the values of the 95% example, small-scale eddy effects that may not be well-
confidence interval limits defined previously. The predicted described by the three two-equation turbulence models
methylene blue conversion percentages for the different rate considered as part of this study could influence the relation-
constants are shown in Table 4. The CFD model was also ship between turbulent mixing and chemical reaction, and,
evaluated for the sensitivity of the methylene blue conversion thus, result in differences between the CFD and pilot results.
to changes in dissolved organic carbon (DOC) by varying the Turbulence can be composed of many transient structural
inlet DOC concentration from 0 to 2 mg L1. A summary of the features that are difficult to capture with Reynolds Averaged
results is provided in Table 5 and graphically in Fig. 7. Navier–Stokes (RANS) equations. The RANS equations in CFD
have been shown to poorly capture the turbulent kinetic
energy and the extent of the recirculation zones in the wake
4. Discussion regions behind flow obstructions. One hypothesis for the
differences between the experimental and model results at
Examination of Fig. 5 reveals several trends. First, the CFD the higher flow rates in this study is the CFD RANS models did
model under predicts the methylene blue conversion within not adequately capture the flow characteristics downstream

Table 4 – Impact of second-order rate constant kMB,OH on CFD results.

kMB,OH (M1 s1)

1.0  109 6.9  1010 3.5  1010 8.9  1010 4.8  1010

Fraction of initial compound remaining 0.966 0.440 0.553 0.389 0.486

Compound conversion (%) 3.4 56.0 44.7 61.1 51.4

MB ¼ Methylene Blue.
For all models: Flow Rate ¼ 20 gal min1 [4.5 m3 hr1], Initial MB Concentration ¼ 0.50 mg L1, Molar Ratio of H2O2 to Initial MB ¼ 199.25, H2O2
Concentration ¼ 10.6 mg L1, Power Output at Lamp Surface ¼ 26.5 W, Fluence Rate at Sensor ¼ 18.0 W m2, UVT254 ¼ 94.9%, pH ¼ 7.24,
Alkalinity ¼ 22.2 mg L1 as CaCO3, Total Chlorine Residual ¼ 0.01 mg L1 as Cl2, Number of Baffles in Reactor ¼ 5, DOC ¼ 0.57 mg L1, Turbulence
Sub-model ¼ Standard k-3, Fluence Rate Sub-model ¼ RAD-LSI.
1806 water research 44 (2010) 1797–1808

100

+/- 0.25 mg/L

95

Percent Removal of Methylene Blue (%)

90

5.96%
85

80

75

70
y = -11.914x + 93.2
R2 = 0.9921
65 0.57 mg L-1
used in CFD

60
0.0 0.5 1.0 1.5 2.0
DOC Concentration (mg L-1)

Fig. 7 – Impact of DOC on advanced oxidation of methylene blue.

of the baffle plates in which enhanced mixing might have As seen in Table 4, the overall conversion of methylene
improved the reaction conditions between the higher blue is a strong function of the kinetic rate constant (kMB,OH) as
hydroxyl radical concentrations near the lamp and methylene expected. Within the range of the 95% confidence interval of
blue. The percent removal values in Table 2 indicate that the the rate constants determined experimentally, the percent
two-equation turbulence model selection (i.e., standard k-3, removal of methylene blue varied by 9.7% (i.e., 51.4% removal
RNG k-3, and k-u) for this reactor configuration did not at the lower limit and 61.1% removal at the upper limit). The
significantly impact the predicted removal for methylene overall removal of methylene blue was also shown to be
blue. sensitive to the concentration of DOC in the water matrix. Any
Another factor that may have contributed to the differ- variance in the influent DOC concentration between pilot runs
ences between experimental and numerical results is the could explain some of the differences between the pilot and
accuracy of the fluence rate model in simulating the actual CFD results (since the DOC of the pilot influent was only
fluence rate distribution in the pilot reactor. The UV fluence measured once in this research). In Fig. 7, the value of DOC
rate model in the CFD simulations was calibrated using an used in the CFD model (0.57 mg L1) is shown along with the
irradiance sensor located at the reactor wall at approximately resulting variability in methylene blue conversion if the DOC
the mid-point of the reactor length. The sensor output was changed by plus or minus 0.25 mg L1. Further, the rate
adjusted using a correction factor determined under the constant for the scavenging of radicals by DOC may also vary
collimated beam to account for quartz window reflection, based on the composition of the organic matter (Westerhoff
window transmittance, and sensor viewing angle. The et al., 2007).
resulting fluence rate reading for each pilot test was combined
with the experimentally-determined UVT254 for each test to
calibrate the lamp power in each model. Since the overall 5. Conclusions
advanced oxidation reaction (specifically, the hydroxyl radical
concentration) is sensitive to delivered UV dose, any discrep- For the reactor system evaluated within this research, the use
ancies between the delivered UV dose and that simulated in of CFD for the evaluation of UV-initiated advanced oxidation
the CFD model could contribute to differences between pre- processes for the degradation of an organic contaminant
dicted (model) and actual (pilot) methylene blue conversion. produced several conclusions:
In addition, the slope of the curve describing methylene blue
conversion as a function of UV dose becomes more negative,  Effective CFD models for AOP should incorporate rigorous
i.e., steeper, as UV dose decreases. Thus, the difference turbulence sub-models, fluence rate sub-models, kinetic
between predicted and actual methylene blue conversion rate equations, and proper characterization of the back-
would become more magnified at lower UV doses (higher flow ground water matrix.
rates). As a result, any uncertainty in the fluence rate model  In this research, the CFD model tended to under-predict the
could also explain the trend of an increasing discrepancy percent removal of methylene blue compared to corre-
between actual and predicted methylene blue removal as the sponding pilot reactor trials. The percent difference
flow rate increased. between the pilot and the CFD results increased with
 water research 44 (2010) 1797–1808 1807

increasing flow rates. It is likely that improper character- Association, Water Environment Federation, & American
ization of small-scale eddy turbulence effects at these Public Health Association, Washington, DC.
higher flows was a cause for the increased difference Baldyga, J., Bourne, J.R., 1999. Turbulent Mixing and Chemical
Reactions. John Wiley & Sons Ltd., Chichester, UK.
between the model and experimental results.
Bali, U., Catalkaya, E.C., Sengul, F., 2003. Photochemical
 The MSSS fluence rate sub-model predicted higher degradation and mineralization of phenol: a comparative
contaminant removal values than the RAD-LSI sub-model study. Journal of Environmental Science and Health,
under the same operating conditions. Part A-Toxic/Hazardous Substances & Environmental
 The selection of different two-equation turbulence sub- Engineering A38 (10), 2259–2275.
models had a negligible impact on the predicted conversion Blatchley III, E.R., 1997. Numerical modelling of UV intensity:
of methylene blue for the tested reactor configurations. application to collimated-beam reactors and continuous-flow
systems. Water Research 31 (9), 2205–2218.
 The overall transformation of methylene blue was a strong
Bolton, J.R., 2000. Calculation of ultraviolet fluence rate
function of the value of the second-order kinetic rate distributions in an annular reactor: significance of refraction
constant describing the reaction between methylene blue and reflection. Water Research 34 (13), 3315–3324.
and the hydroxyl radical. Bolton, J.R., 2001. Ultraviolet Applications Handbook, second ed.
 The overall removal of methylene blue was sensitive to the Bolton Photosciences, Inc, Ayr, ON Canada.
concentration of the radical scavenging dissolved organic Buxton, G.V., Greenstock, C.L., Helman, W.P., Ross, A.B., 1988.
Critical review of rate constants for reactions of hydrated
carbon in the water matrix.
electrons, hydrogen atoms and hydroxyl radicals (OH/O) in
aqueous solution. Journal of Physical and Chemical Reference
As oxidation pathways for emerging water contaminants Data 17 (2), 513–886.
are identified, a simulation model, such as the one described Crittenden, J.C., Hu, S., Hand, D.W., Green, S.A., 1999. A kinetic
herein, may become an important tool for the design and model for H2O2/uv process in a completely mixed batch
optimization of advanced oxidation systems. These numerical reactor. Water Research 33 (10), 2315–2328.
models will allow evaluation of multiple design scenarios, Devlin, H.R., Harris, I.J., 1984. Mechanism of the oxidation of
aqueous phenol with dissolved oxygen. Industrial and
including number of lamps, layout of reactors, and upstream
Engineering Chemical Fundamentals 23 (4), 387–392.
hydraulic conditions. However, there were significant differ- Ducoste, J., Liu, D., Linden, K.G., 2005. Alternative approaches to
ences between the CFD and pilot results at higher flow rates for modeling fluence distribution and microbial inactivation in
the reactor studied in this research. Thus, follow-up research ultraviolet reactors: Lagrangian versus Eulerian. Journal of
should include the evaluation of more advanced turbulence Environmental Engineering, 1393–1403.
models, including those incorporating the Reynolds Stress Ducoste, J., Linden, K.G., 2005. Hydrodynamic Characterization of
Model and/or the Large Eddy Simulation. The investigation of UV Reactors. Awwa Research Foundation, Denver, CO.
Fiveland, W.A., 1984. Discrete-ordinates solutions of the radiative
fluid dynamics on the advanced oxidation process should also
transport equation for rectangular enclosures. Journal of Heat
include the influence of the reaction time scale and turbulent Transfer 106, 699–705.
eddy time scales. In addition, a more rigorous method for Jacob, S.M., Dranoff, J.S., 1970. Light intensity profiles in
validation of the UV fluence (dose) distribution in UV/AOP a perfectly mixed photoreactor. AIChe Journal 16 (3),
reactors should be evaluated using multiple irradiance sensors 359–363.
and/or online chemical actinometers. As the reaction mecha- Johnson, H.D., Cooper, W.J., Mezyk, S.P., Bartels, D.M., 2002. Free
radical reactions of monochloramine and hydroxylamine in
nisms for additional emerging contaminants become avail-
aqueous solution. Radiation Physics and Chemistry 65,
able, especially quantification of specific byproduct formation,
317–326.
CFD should be considered as a tool for predicting the outcome Klassen, N.V., Marchington, D., McGowan, H.C.E., 1994. H2O2
of direct photolysis and advanced oxidation on these envi- determination by the I 3 method and by KMnO4 titration.
ronmentally-important compounds. Analytical Chemistry 66 (18), 2921–2925.
Larson, R.A., Zepp, R.G., 1988. Reactivity of the carbonate radical
with aniline derivatives. Environmental Toxicology and
Chemistry 7, 265–274.
Acknowledgements Linden, K.G., Sharpless, C.M., Andrews, S.A., Atasi, K.Z.,
Korategere, V., Stefan, M.I., Suffet, I., 2004. Innovative UV
The authors would like to thank the Water Research Foun- Technologies to Oxidize Organic and Organoleptic Chemicals.
dation (formerly the American Water Works Association Awwa Research Foundation. Publisher Place, Denver, CO.
Liou, B.T., Wu, C.Y., 1996. Radiative transfer in a multi-layer
Research Foundation) for funding this research; Infilco
medium with Fresnel interfaces. Heat and Mass Transfer 32
Degremont’s North American Research and Development (1-2), 103–107.
Center for providing pilot equipment; and North Carolina Liu, D., Ducoste, J., Jin, S., Linden, K.G., 2004. Evaluation of
State University for support of graduate fellowships. alternative fluence rate distribution models. Journal of
Water Supply: Research and Technology - AQUA 53 (6),
391–408.
references
Liu, D., Ducoste, J., Wu, C., Linden, K.G., 2007. Numerical
simulation of UV disinfection reactors: evaluation of
alternative turbulence models. Applied Mathematical
Alnaizy, R., Akgerman, A., 2000. Advanced oxidation of phenolic Modeling 31, 1753–1769.
compounds. Advances in Environmental Research 4 (3), Marchisio, D.L., Barresi, A.A., 2003. CFD simulation of mixing and
233–244. reaction: the relevance of the micro-mixing model. Chemical
APHA, 1995. Standard Methods for the Examination of Water and Engineering Science 58, 3579–3587.
Wastewater, nineteenth ed. American Water Works
1808 water research 44 (2010) 1797–1808

Mohseni, M., Taghipour, F., 2004. Experimental and CFD analysis Stamnes, K., Tsay, S.-C., Wiscombe, W., Jayaweera, K., 1988.
of photocatalytic gas phase vinyl chloride (VC) oxidation. Numerically stable algorithm for discrete-ordinate-method for
Chemical Engineering Science 59, 1601–1609. radiative transfer in multiple scattering and emitting layered
Pareek, V.K., Cox, S.J., Brungs, M.P., Young, B., Adesina, A.A., 2003. media. Applied Optics 27 (12), 2502–2509.
Computational fluid dynamic (CFD) simulation of a pilot-scale Stefan, M.I., Hoy, A.R., Bolton, J.R., 1996. Kinetics and mechanism of
annular bubble column photocatalytic reactor. Chemical the degradation and mineralization of acetone in dilute aqueous
Engineering Science 58, 859–865. solution sensitized by the UV photolysis of hydrogen peroxide.
Rosenfeldt, E.J., Linden, K.G., 2004. Degradation of endocrine Environmental Science and Technology 30, 2382–2390.
disrupting chemicals bisphenol A, ethinyl estradiol, and Stefan, M.I., Mack, J., Bolton, J.R., 2000. Degradation pathways
estradiol during IV photolysis and advanced oxidation during the treatment of methyl tert-butyl ether by the UV/
processes. Environmental Science and Technology 38 (20), H2O2 process. Environmental Science and Technology 34 (4),
5476–5483. 650–658.
Scheck, C.K., Frimmel, F.H., 1995. Degradation of phenol and Weinstein, J., Bielski, B.H.J., 1979. Kinetics of the interaction of
salicylic acid by ultraviolet radiation/hydrogen peroxide/ HO2 and O 2 radicals with hydrogen peroxide. The Haber–
oxygen. Water Research 29 (10), 2346–2352. Weiss reaction. Journal of the American Chemical Society 101
Sharpless, C.M., Linden, K.G., 2003. Experimental and model (1), 58–62.
comparisons of low- and medium-pressure Hg lamps for the Westerhoff, P., Mezyk, S.P., Cooper, W.J., Minakata, D., 2007.
direct and H2O2 assisted UV photodegradation of Electron pulse radiolysis determination of hydroxyl radical
N-nitrosodimethylamine in simulated drinking water. rate constants with Suwannee River fulvic acid and other
Environmental Science and Technology 37 (9), 1933–1940. dissolved organic matter isolates. Environmental Science and
Spalding, B., 1998. Turbulent Mixing and Chemical Reaction; The Technology 41 (13), 4640–4646.
Multi-Fluid Approach. A Lecture. Based on The International Wilcox, D.C., 2004. Turbulence Modeling for CFD, second ed. DCW
Symposium on the Physics of Heat Transfer in Boiling and Industries, Inc., La Canada, CA.
Condensation, from. http://www.simuserve.com/phoenics/d_ Yakhot, V., Orszag, S.A., 1986. Renormalization-group analysis of
polis/d_lecs/mfm/mfm0.htm. turbulence. Physical Review Letters 57 (14), 1722–1724.