University of Southern California Dissertations and Theses

Applications of continuous flow reactors towards screening catalytically active nanoparticle syntheses

(USC Thesis Other)

Applications of continuous flow reactors towards screening catalytically active nanoparticle syntheses

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Applications of Continuous Flow Reactors Towards Screening Catalytically
Active Nanoparticle Syntheses
by
Majed Sameer Madani
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(CHEMICAL ENGINEERING)
December 2023
Copyright 2023 Majed Sameer Madani

This dissertation is dedicated to my parents for their endless love, support and
encouragement
ii

Table of Contents
Dedication.................................................................................... ii
List of Tables ................................................................................ vii
List of Figures ............................................................................... viii
Abstract ...................................................................................... xv
Chapter 1: Background ................................................................... 1
1.1 Fundamentals of Fluid Mechanics in Milli- and Microfluidic Systems ........... 2
1.1.1 Flow in Round Capillaries .................................................. 3
1.1.2 Reynolds Number............................................................ 4
1.1.3 Fluidic Resistance ........................................................... 4
1.2 Diffusion .............................................................................. 5
1.3 Droplet Flow ......................................................................... 6
1.4 Computational Fluid Dynamics .................................................... 7
1.4.1 Microfluidic Passive Mixer Simulation ..................................... 8
1.5 Fundamentals and Governing Equations ........................................... 8
1.6 Mixing Efficiency Analysis .......................................................... 10
1.7 Implications and Significance ....................................................... 10
1.8 Applications of Microfluidics in Biomolecular systems: A Case Study on
Cholesterol’s Impact on Carbon Dioxide Permeability Across Lipid Membranes 12
1.8.1 Methodology ................................................................. 12
iii

1.8.2 Results & Discussion ........................................................ 14
1.8.3 Conclusion ................................................................... 16
References .................................................................................. 19
Chapter 2: Continuous Flow Reactors for Synthesizing Colloidal Nanoparticles ............................................................................ 20
2.1 Introduction .......................................................................... 20
2.2 Reactor Types for Flow Syntheses.................................................. 21
2.3 Flow Regimes and Their Impact on Nanoparticle Synthesis...................... 23
2.4 Single-Phase Flow ................................................................... 24
2.5 Multi-Phase Flow .................................................................... 25
2.6 Control of Fluid Flow in Reactors .................................................. 26
2.7 Transport Phenomena in Batch and Flow Reactors............................... 28
2.8 Scaling Nanoparticle Syntheses ..................................................... 29
References .................................................................................. 32
Chapter 3: Throughput Optimization of Molybdenum Carbide Nanoparticle Catalysts in a Continuous Flow Reactor using Design of
Experiment ................................................................... 33
3.1 Abstract .............................................................................. 33
3.2 Introduction .......................................................................... 34
3.3 Results and Discussion .............................................................. 38
3.3.1 Continuous Flow Millifluidic Synthesis and Reactor Setup ............... 38
3.4 Design of Experiments .............................................................. 40
3.5 Optimization via RSM: Maximizing Throughput ................................. 48
3.6 Experimental Procedures............................................................ 56
3.6.1 Continuous Flow Synthesis of MoC1−x Nanoparticles..................... 56
3.6.2 Millifluidic Flow Synthesis Platform ....................................... 56
iv

3.6.3 Furnace Temperature Stability ............................................. 59
3.6.4 Precursor Thermal Time Constant Estimation ............................ 61
3.6.5 Residence Time Calculation ................................................ 62
3.6.6 Reactor Cleaning ............................................................ 63
3.6.7 Catalytic Testing ............................................................ 64
3.6.8 Characterization ............................................................. 64
3.7 Conclusion............................................................................ 65
References .................................................................................. 73
Chapter 4: Machine Learning-Assisted Visible Light Spectrophotometry in
Continuous Flow Reactors for Kinetic Analysis of Ionic LiquidBased Platinum Nanoparticle Synthesis................................ 74
4.1 Introduction .......................................................................... 74
4.2 Preliminary Screening ............................................................... 76
4.3 Training Matrix ...................................................................... 80
4.4 Machine Learning-assisted visible light spectrophotmetry ........................ 81
4.5 Continuous Flow Synthesis of Pt Nanoparticles ................................... 85
4.6 Formation of IL-based Pt Nanoparticles ........................................... 86
4.7 Nanoparticle Size Estimation ....................................................... 86
4.8 Methods .............................................................................. 90
4.8.1 Partial Least Square Regression ............................................ 90
4.8.2 Artificial Neural Network ................................................... 90
4.8.3 Machine Learning Model Training ......................................... 90
4.9 Supplementary Machine Learning Models ......................................... 92
4.9.1 Artificial Neural Network (ANN)........................................... 92
4.9.2 Partial Least Squares Regression (PLSR) and Cross-Validation ......... 92
4.10 Concluding Remarks................................................................. 93
References .................................................................................. 96
v

Chapter 5: In situ Characterization for Screening Colloidal Nanoparticle
syntheses ...................................................................... 97
5.1 Introduction .......................................................................... 97
5.1.1 Integration of X-ray Scattering in Flow Reactors ......................... 99
5.1.2 Scattering Theory ........................................................... 100
5.2 Particle Size and Shape ............................................................. 101
5.3 Radius of Gyration .................................................................. 102
5.4 Porod’s Law .......................................................................... 102
5.5 Dimensional Analysis ................................................................ 103
5.6 Preliminary Reactor Design for In situ X-ray Scattering ......................... 105
5.6.1 Future Outlook .............................................................. 108
References .................................................................................. 109
Bibliography ................................................................................. 109
vi

List of Tables
2.1 Application Range of Tubular Reactors Materials ................................ 23
3.1 Coded Low (-1), Center Point (0), and High (+1) Experimental Values for
the Input Variables Investigated in the Full Factorial Screening Design. ........ 40
3.2 Full factorial matrix via coded values. ............................................. 42
3.3 Full factorial matrix of the corresponding real values ............................. 43
3.4 Real values of the levels investigated for each significant variable in the Doehlert
optimization of maximizing throughput. As depicted from the number of levels investigated for each variable, the variables most significantly affecting
throughput in decreasing order are concentration, flow rate, and amount of
oleylamine. ........................................................................... 46
3.5 Coded Doehlert optimization matrix appended (runs 1-18) with points on
each corner of the design space (runs 19-28). ...................................... 46
3.6 Optimization matrix of the corresponding real values ............................ 47
3.7 Table caption here ................................................................... 59
4.1 Preliminary Screening Design of Experiment Matrix. ............................. 78
4.2 Levels of each component for the training Matrix. ................................ 81
4.3 Comparison of predicted and actual nanoparticle sizes and yields for different
solvents. .............................................................................. 88
5.1 Photo-absorption cross-section for BMIM-Tf2N constituent elements at 8000
eV. .................................................................................... 103
vii

List of Figures
1.1 The historical progress of microfluidics over the last decades in chronological
order. ................................................................................. 2
1.2 Laminar flow. ........................................................................ 5
1.3 Microfluidic analysis showing (a) surface concentration, (b) surface velocity,
(c) concentration profile, and (d) velocity magnitude. ............................ 11
1.4 Illustrative schematic of the experimental approach. A buffer with a high CO2
content and another containing vesicles with pH-responsive fluorescent probes
meet in a Y-shaped channel. CO2 transits the lipid bilayer of the vesicles,
causing an internal pH drop and diminishing fluorescence. Continuous imaging is facilitated by the serpentine microchannel, which allows for prolonged
observation of the vesicle without immobilization. The composite fluorescence image displays a strobe-like series of snapshots, taken at 20-millisecond
intervals, documenting the temporal evolution of the vesicle’s fluorescence. .... 13
1.5 Short caption for LOF............................................................... 14
1.6 Exemplary data from a vesicle with composition 40-60 POPC-chol. The fluorescence has been normalized to the first point. Data points are derived from
each frame of the acquisition. The breaks in the data arise from the vesicle
leaving the field of view as it traverses the channel. The red line is a fit to
Equation 1.17, with a 1/τ = 1.58 s−1 and I∞ = 0.63. ............................. 15
2.1 Sperpentine-shaped milli-fluidic plate reactor (ca. 1.8 mL volume) made of
borosilicate glass. .................................................................... 22
viii

2.2 Illustration of advanced milli- and microfluidic systems for scaling nanoparticle
synthesis: (Left) Schematic of the self-optimizing parallel millifluidic reactor
with the control and feedback system for nanoparticle synthesis;16 (Right) (a)
Schematic of the parallel network assembled by connecting a distribution manifold to four droplet generators, showing a gradient of resistances across the
branches. (b) Comparison of droplet diameters produced under different flow
rates, illustrating the transition between flow invariant and flow dependent
regimes.9
.............................................................................. 29
3.1 Schematic of the millifluidic reactor system for the continuous flow synthesis
of MoC1−x nanoparticles. A syringe pump with a heated syringe (80 C) is
used to drive a precursor mixture into a reactor coil housed in a furnace. The
colors are representative of those empirically observed in the reaction, with
yellow (at reactor inlet) indicating the unreacted Mo(CO)6 precursor solution
and black indicating the MoC1−x nanoparticle suspension isolated into plugs
due to in situ gas evolution. An in-line thermal dissipation-type flow sensor at
the reactor outlet reports the relative volume of liquid and gas—an idealized
flow sensor signal is shown with corresponding liquid (detected by the sensor)
and gas (not detected) plugs. Product is collected in vials that can be isolated
with valves to allow for removal of partial product volumes during runs. The
system pressure is maintained by a fluidic controller (Fluigent) which is locked
on a fixed pressure of 20 psi to minimize in situ gas evolution. .................. 39
3.2 (a) Powder XRD pattern and (b) TEM image of the MoC1−x nanoparticles
synthesized under the base reaction conditions. ................................... 40
ix

3.3 Standardized Pareto charts for (a) residence time, (b) isolated yield, and (c)
throughput. The variables on the y-axis of the Pareto charts are defined as A =
flow rate (mL h-1), B = temperature (°C), C = amount of oleylamine (%vol),
and D = precursor concentration (mM). The vertical line in each Pareto chart
corresponds to = 5%, and (+) and (-) correspond to an increase or decrease
in the response, respectively, for the high value of a given factor. Main effects
plots for (d) residence time, (e) isolated yield, and (f) throughput, where (+)
and (-) correspond to the high and low levels for each factor, respectively. .... 44
3.4 Calculated response surface function demonstrating the reaction conditions
(precursor concentration, flow rate, and amount of oleylamine) that correspond to a specific throughput, illustrated by the color legend. The conditions for maximum throughput are indicated by a star. The bounds of this
parameter space are specific to this flow reactor and synthetic system; any
points outside of the parameter space are not feasible for this system. .......... 50
3.5 Visual representation of the design points for the optimization, corresponding
to the runs displayed in Table 3.5. ................................................. 51
3.6 Prediction variance plot of the design space. Concentration is set at the base
value (coded 0) for visual simplicity. ............................................... 52
3.7 (a) Powder XRD pattern and (b) TEM image of the MoC1–x nanoparticles
produced under optimized conditions, and (c) powder XRD pattern and (d)
TEM image of the MoC1–x nanoparticles produced under unoptimized conditions. .............................................................................. 53
3.8 (a) HRTEM and (b) SAED pattern of MoC1−x nanoparticles synthesized under optimized conditions. (c) HRTEM and (d) SAED pattern of MoC1−x
nanoparticles synthesized under batch conditions. The lattice fringes correspond to the (111) and (200) planes of the α-phase. .............................. 54
x

3.9 FT-IR spectra of Mo(CO)6 precursor and the supernatant of the first wash
of the MoC1−x nanoparticles (in the workup procedure). The spectra show
the representative n(CO) stretching region, demonstrating that there is no
unreacted Mo(CO)6 precursor in the supernatant resulting from the product
stream................................................................................. 55
3.10 CO2 conversion as a function of time-on-stream and (b) product selectivity
taken as an average of data from 16-20 h. Reaction conditions were 300 ◦C, 2
MPa, WHSV based on Mo content of 40 h
−1 and H2:CO2 molar ratio in the
feed of 2.7. .......................................................................... 55
3.11 Photographic images of the experimental setup. The top image shows the
entire millifluidic flow synthesis platform, and the bottom image shows the
reactor placed inside the furnace.................................................... 58
3.12 Representative data showing time-dependent furnace temperature at approaching a setpoint of 320 ◦C: (a) Ramp-up stage and (b) time-dependent temperature after reaching steady state and prior to running the experiment. ......... 60
3.13 Modeled time-dependent temperature of the precursor fluid (from 80 to 340 ◦C)
approximated by the lumped capacitance model. ................................. 62
3.14 Exemplary data for the sensor readout during product collection with a pump
flow rate set at QL = 40 mL h−1
. The data shown here correspond to a 2-min
time frame for experiment #1 in the full factorial matrix (Table 3.3). Data
are sampled at 10 Hz. Based on the count of data points above and below the
threshold throughout the experiment (14 min), the gas to liquid volumetric
ratio was found to be 3.74, which translates to QG = 150 mL h−1
. Therefore,
the residence time is calculated as τ = V/(QL + QG) = 11.7 min. ............... 63
xi

4.1 Illustration of the experimental setup for the preliminary screening phase of
the Design of Experiments (DoE). The configuration includes five syringe
pumps, each responsible for accurately delivering reagents into the system
at predetermined flow rates in accordance with the full factorial matrix design. Pumps 1 and 2 manage the introduction of the ionic liquid (IL) and its
mixture with polyvinylpyrrolidone (PVP), respectively, while Pumps 3 and 4
handle the ethylene glycol (EG) and the Pt salt dissolved in EG. The fifth
pump dispenses the Pt nanoparticles suspension. This setup enables precise
manipulation of reaction parameters to discern the influence of various concentrations on the spectral profile of the resultant Pt nanoparticles ............. 77
4.2 ANOVA Significance Plots for Principal Components 1, 2, and 3. Subplots
(a), (b), and (c) represent the significance of variables on PC1, PC2, and PC3,
respectively. The x-axis shows the LogWorth of each variable, a transformation of the p-value, while the y-axis lists the components (variables). The red
dashed line indicates the p=0.05 significance threshold. Variables to the right
of this line are considered significant. Notably, PtNP is the most significant
variable affecting PC1, and Pt(II) for PC2 while no variables significantly
affect PC3 at the 5% significance level. ............................................ 79
4.3 Schematic representation of the revised experimental setup employed during
the training phase. The setup features three syringe pumps, used to administer the ionic liquid with PVP, the ethylene glycol with Pt salt, and the Pt
nanoparticle solution. This configuration ensures a more refined control of the
components at increased resolution levels, as determined by the preliminary
screening outcomes. The solutions converge at a single microfluidic junction,
where passive mixers facilitate optimal homogenization before the mixture
progresses to the spectroscopic analysis segment .................................. 80
xii

4.4 Scatter plots illustrating the relationship between actual and predicted concentrations of PtNPs across ten folds of cross-validation. Each plot denotes a
separate fold, with data points representing individual predictions for training
(green) and validation (blue) datasets. The accompanying R2 values highlight
the model’s predictive accuracy, with training R2 values consistently high and
validation R² values showcasing the model’s generalization capability. .......... 83
4.5 Training and validation loss curves over successive epochs for each of the
ten folds in the K-fold cross-validation. The plots exhibit a sharp decline
in training loss (green) indicating effective learning, while the validation loss
(blue) plateaus close to the training loss, demonstrating the model’s ability to
generalize. The congruence of these losses across all folds signifies a balanced
and well-generalized model. ......................................................... 84
4.6 Experimental Setup for Continuous Flow Synthesis of Pt Nanoparticles ........ 85
4.7 Comparative analysis of Pt NP concentration against residence time across
three solvent systems. (a) BMPy, (b) BMPyrr, and (c) pure EG, with data
points categorized as "High" and "Low" to denote the initial precursor concentration levels. These plots demonstrate the distinctive kinetic behaviors
of Pt NP synthesis in various solvents, highlighting the influence of solvent
properties and ionic liquid presence on the reaction’s progression ............... 87
4.8 TEM images illustrating the variance in nanoparticle average size and standard deviation (SD) across different solvents and reaction conditions. From
top left to bottom right: BMPyrr in flow, BMPyrr in batch, BMPy in flow,
and BMPY in batch reactions. ..................................................... 89
5.1 A schematic of a tube with an in-flow reaction where in situ characterization
methods are utilized to screen reactions or characterize nanoparticle synthesis
processes across a wide range of the electromagnetic spectrum .................. 98
xiii

5.2 Transmission of the beam as a function of photon energy for three different
ionic liquids with a path length of 900 µm......................................... 104
5.3 Transmission of the beam as a function of photon energy for Quartz and
Borosilicate glass consider 100 µm and 200 µm walls ............................ 105
5.4 Transmission of the beam as a function of photon energy for Kapton tubes
with different wall thicknesses ...................................................... 106
5.5 Kapton-based tubular reactor.(a) Side perspective displaying inlets for cartridge heaters and sensors. (b) Interior view illustrating the arrangement of
cartridge heaters. (c) Frontal view showcasing the alignment of the Kapton
tubes within the grooves and the central slits designated for X-ray examination. .................................................................................. 107
xiv

Abstract
The dissertation presented herein is structured into chapters that delve into various research
domains within milli- and microfluidic systems. Part of this dissertation includes collaborative authorship. Chapter 1 introduces the fundamentals of fluid mechanics. In this chapter,
some highlights of the important physical phenomena that are dominant in milli- and microscale flow systems are presented, focusing on flow dynamics, diffusion, and computational
fluid dynamics simulations. It emphasizes the importance of fluid behavior in microscale
systems and introduces a case study on microfluidics applications in biomolecular systems
in which a portion of a manuscript I participated in as a third author is presented. Chapter 2 covers applications of continuous flow synthesis of colloidal nanoparticles using milliand microfluidics systems, highlighting the advantages of miniaturized systems in reactionbased nanoparticle syntheses. Chapter 3 is adapted from a published manuscript in which
I am a joint primary author. Chapter 3 describes the use of continuous flow methods for
screening the reaction parameters of catalytically active molybdenum carbide nanoparticle
synthesis with an emphasis on throughput optimization using a Design of Experiment approach. Chapter 4 introduces machine learning-assisted spectrophotometry, showcasing the
integration of machine learning algorithms for the kinetic analysis of ionic liquid-based platinum nanoparticle synthesis. Chapter 5 introduces in-situ characterization for continuous
flow reactors with a particular objective of studying the nucleation and growth kinetics of
nanoparticle synthesis using X-ray scattering. This chapter provides a critical evaluation of
flow reactor designs for in situ X-ray scattering analysis, focusing on the synthesis of ionic
liquid-based Pt nanoparticles.
xv

Chapter 1: Background
The study of microfluidics is a multidisciplinary field that includes parts of Physics, Chemistry, Engineering and Biotechnology in which fluid flow in miniaturized systems is utilized
to perform chemical, biochemical, or biomolecular operations. Microfluidic processes often
manipulate small amounts of fluids, using channels with dimensions on the order of tens to
hundreds of micrometers. Such small dimensional characteristics offer useful capabilities such
as low volume consumption, low cost, and short analysis time.1 The history of microfluidic
systems spans decades starting with the invention of photolithography in the 1950s, which led
to the development of one of the first Lab-on-a-chip (LoC) devices.2
In the 1990s, microchip
technology was developed by Manz and co-workers3, 4 where they introduced the concept
of the miniaturized total chemical analysis system by performing fast and efficient chemical
analysis with lower reagent consumption and shorter analysis time. Since the 2000s, research
microfluidic devices were further developed to include on-chip tissue and organ models in
order to overcome the limitations of in vitro and in vivo biological processes.5 On another
front, millifluidics involves manipulation of fluids in channels that are one millimeter across.
While providing similar capabilities to microfluidics, the choice of either milli- or microfluidic
systems depends on the goal of the project under investigation. In some aspects, millifluidics
offer certain advantages over smaller channels in microfluidics where issues of blockage due
to the formation of precipitates are more common.6 Over the past two decades, there have
been substantial advances in the design and development of milli- and microfluidic platforms
for various applications in chemical and biomolecular systems, including flow synthesis of
chemical intermediates, medical diagnostics, protein analysis, and drug delivery. The miniaturized nature of milli- and microfluidic systems allows minimal usage of reagents to perform
1

experiments and obtain results in a quicker manner. Comprehensive reviews with greater
depth of analysis were reported for microfluidics.7, 8 In this dissertation, applications of continuous flow reactors in milli- and microscale systems are discussed with an emphasis on
reaction screening of catalytically active nanoparticle syntheses.
Figure 1.1: The historical progress of microfluidics over the last decades in chronological
order.
1.1 Fundamentals of Fluid Mechanics in Milli- and Microfluidic Systems
At the microscale, the laws of physics remain the same as in macroscale systems. However,
fluid dynamics behaves differently at reduced channel dimensions where the scaling down
of fluids into sub-millimeter scale tends to favor the domination of one force over another.
To understand the full benefits of miniaturized systems, it is important to investigate the
physics governing said systems. In this section, analyses are shown considering flow in microchannels. To a large extent, millifluidics follow similar behavior with a relatively comparable
high surface-area-to-volume ratio.
For a given domain, a physical description of the space-time evolution of the velocity
field can be obtained from conservation of momentum and the conservation of mass. Under
the assumption that the fluid is Newtonian and incompressible, pressure-driven flow can be
described by the Navier-Stokes equation and the continuity equation:
2

ρ

∂v
∂t + (v · ∇)v

= −∇P + η∇2v (1.1)
∇ · v = 0 (1.2)
where ρ is the fluid density, v is the fluid velocity, P is the pressure, and η is the fluid
viscosity.
1.1.1 Flow in Round Capillaries
Flow in round channels is a common type of flow in milli- and microscale systems. The
rigorous characterization of pipe flow dates back to the 19th century where both Hagen
(1839), and Poiseuille (1841) described the relation between velocity and pressure head
showing its inverse proportionality to the tube diameter. Pressure-driven steady flow in
circular channels is described by:
1
r
∂
∂r
r
∂vz
∂r
= −
1
µ
∂P
∂z (1.3)
where r is the radial coordinate and z is the central axis of the channel. The solution to
equation (1.3) results in a parabolic velocity profile across the channel diameter:
vz = vmax
1 −
r
2
a
2

(1.4)
The maximum velocity is given by:
vmax =
a
2∆P
4µL
(1.5)
For such flow, the volumetric flow rate is directly proportional to the pressure drop
applied. Equation (1.5) can be expressed in another common form relating the volumetric
flow rate:
3

∆P =
8µLQ
πR4
(1.6)
This is known as the Hagen-Poiseuille equation, which describes the pressure drop for
incompressible laminar flow through a long cylindrical pipe. The Hagen-Poiseuille equation
is a solution to the Navier-Stokes equation under the assumption that the fluid is Newtonian,
incompressible, and fully developed.
1.1.2 Reynolds Number
Dimensionless ratios between two values are useful tools to describe which forces play dominant roles in describing the flow characteristics. One such ratio is the Reynolds number
(Re), which is defined as the ratio of inertial to viscous forces. The Re is a dimensionless
quantity used to characterize the fluid behavior within the channels, which is mathematically
expressed by:
Re =
ρvDh
µ
(1.7)
where ρ is the fluid density, v is the characteristic velocity, µ is the fluid viscosity, and Dh
is the hydraulic diameter of the system in question. Re< 2300 generally indicates a laminar
flow. As Re approaches 2300, the fluid enters a transition stage showing signs of turbulence.
Above 2300, the fluid is described as turbulent. In laminar flow, the liquids are transported
in uniform layers of thickness between fixed boundaries, where the mixing of the streamed
liquids occurs by diffusion across the liquid-liquid interfaces. This is a typical phenomenon
in microfluidic systems.
1.1.3 Fluidic Resistance
The Hagen-Poiseuille equation (eq 1.6) describes the relationship between pressure drop, flow
rate, and fluidic resistance. Analogous to Ohm’s law for electrical circuits (V = IR), both
4

Figure 1.2: Laminar flow.
electrical and fluidic resistances are directly proportional to length. Eq 1.6 can be expressed
as an analog to Ohm’s law for fluidic systems:
∆P = RhQ (1.8)
For circular geometry, the hydrodynamic resistance (Rh) can be calculated by:
Rh =
8µL
πR4
(1.9)
where µ is the fluid dynamic viscosity, L is the channel length, and R is the channel
radius. One notable difference between the Hagen-Poiseuille equation and Ohm’s law is that
the fluidic resistance is not inversely proportional to the cross-sectional area πR2
, but to
πR4
, unlike the electrical resistance.
For rectangular microchannels with high aspect ratios (i.e., w ≪ h or h ≫ w):
Rh =
12µL
wh3

1 −
192h
π
5w
tanh πw
2h

−1
. (1.10)
where w is the channel width and h is the channel height.9
1.2 Diffusion
In a similar manner to the velocity profile, the space-time evolution of the concentration
field of a given species can be described by the convection-diffusion equation:
∂C
∂t + D∇2C − ∇ · (vC) = 0 (1.11)
5

where C is the concentration of the solute, and D is the diffusion coefficient. Eq 1.11
can be simplified by examining the Péclet Number (Pe), which is a dimensionless number
describing the ratio of convective to diffusive transport of molecules in a fluid:
Pe =
vL
D
(1.12)
where D is the diffusion coefficient and L is the characteristic length. If Pe ≫ 1, the
diffusive mass transport term D∇2C in eq 1.11 can be omitted. If Pe ≪ 1, the convective
mass transport term ∇ · (vC), can be omitted, resulting in a reduced form to Fick’s second
law, ∂C
∂t = D∇2C, which can be used to correlate the time-dependent diffusion to reach
steady state in a channel flow.
Laminar layers glide past each other, and the transport between them is purely diffusive,
with mixing time inversely proportional to the diffusivities of the species being mixed. This is
due to the smaller dimensions of the systems leading to a shorter diffusion time for any given
molecule. An approximation for diffusion time scale is t ≈
x
2
2D
. One drawback, however, is the
insufficient mixing behavior, which is often required for reagent homogenization or sample
dilution in chemical or biological reactions. For the case of chemical reactions, multi-phase,
or droplet flow, can be utilized to mitigate the issue of insufficient mixing.
1.3 Droplet Flow
Droplet flow has been employed as a promising tool in chemical and biological experimentation. The principle behind droplet multiphase flow is based on the use of two immiscible
fluids such as liquid-liquid or liquid-gas to form droplets or slugs of one phase in the other.
The physical aspects that are involved in droplet or slug formation depend on the design of
the channels along with its material properties. Microfluidic platforms allow for the production of monodisperse droplets at rates exceeding tens of kilohertz while maintaining control
of their size and trajectory.10
6

Multiphase systems possess several features such as large interfacial areas, fast mixing,
and reduced axial dispersion that give them advantages in continuous flow reactions that cannot be achieved in batch systems. In addition, droplet flow helps eliminate fouling or reduce
channel clogging that may arise from deposition of materials on the channel walls.11 While
most droplet flow systems rely on oil as a carrier fluid, oil-free droplet generation, in which
liquid-in-gas is used in lieu of conventional liquid-in-liquid systems has also been demonstrated.12 The droplets generated in this manner is more suitable for chemistries that are oil
sensitive. This is known to be a concern for some chemistries where cross-contamination of
the droplet contents is an issue when surfactants are used.13 Oil-free droplet generation is
also useful for systems that requires high temperatures since most used synthetic oils such
as mineral oil or silicon oil have a limited operating temperature.10
1.4 Computational Fluid Dynamics
Computational fluid dynamics (CFD) has become an indispensable tool in the field of microfluidics, offering a detailed and precise way to simulate and analyze the behavior of fluids
in small-scale channels and devices. Microfluidics, dealing with the control and manipulation
of fluids at the microscale, involves a range of complex phenomena that are often challenging to study experimentally. CFD simulations provide a versatile platform to predict fluid
behavior, understand transport mechanisms, and optimize device design under a variety of
conditions. CFD provides numerical simulations of fluid flows based on conservation principles: energy, Newton’s second law, and mass conservation. These simulations transform
the differential equations of fluid flow into numerical data, enabling visualization of fluid behavior and interaction within microfluidic systems. This capability is essential for designing
and assessing the performance of microfluidic devices.
7

1.4.1 Microfluidic Passive Mixer Simulation
In this section, we investigate the simulation of a microfluidic passive mixer designed with
a herringbone structure using COMSOL Multiphysics. This simulation is pivotal in evaluating the mixing efficiency of the mixer, which is crucial for a wide range of applications in
microfluidic engineering. The herringbone structure introduces secondary flows and chaotic
advection, enhancing the mixing of species within the microchannel.14 This analysis not
only contributes to the understanding of fluid behavior in microscale passive mixers but also
serves as a foundation for optimizing mixer designs for more efficient and effective microfluidic applications.
1.5 Fundamentals and Governing Equations
The fundamentals of the simulation rely on the principles of computational fluid dynamics
(CFD). These principles are encapsulated in a set of governing equations, each representing
a core aspect of fluid behavior and interaction within this complex microfluidic system. The
Convection-Diffusion Equation plays a vital role in understanding the transport of species
within the mixer. It combines the effects of convection and diffusion and is given by:
∂ϕ
∂t + u · ∇ϕ = D∇2ϕ (1.13)
Where ϕ is the concentration of the species and D is the diffusion coefficient.
In our investigation, we employ the Steady State and Creeping Flow interfaces to model
the behavior of diluted species within a microfluidic mixer, adhering to the convectiondiffusion equation. The viscosity is posited to vary quadratically with the species concentration, encapsulated by the relationship µ = µ0(1 + αc2
), where α is a dimensionality constant
with units of m6/mol2
, and µ0 represents the viscosity at zero concentration.
To simulate more complex, multicomponent systems, the Transport of Concentrated
Species interface is utilized. This approach forms the theoretical basis for our COMSOL
8

simulations, enabling us to accurately model and analyze the microscale mixing process.
Diffusivity estimation is calculated using the Stokes-Einstein equation, which relates the
diffusivity D to the temperature T, the fluid’s dynamic viscosity η, and the hydrodynamic
radius r of the solute particle as D =
kBT
6πηr
, where kB is Boltzmann’s constant. in this study,
the diffusivity of 1-Butyl-1-methylpyrrolidinium trifluoromethanesulfonate ([BMP][TFO]) in
ethylene glycol at room temperature was calculated Utilizing the Stokes-Einstein equation,
and was found to be 2.9 × 10−11 m2/s. This value aligns well with the expected range for
ionic liquids.15
Finally, the Coefficient of Variation (CV) is used in the context of evaluating mixing
efficiency in a microfluidic system. The Coefficient of Variation provides a comprehensive
and comparative measure of the relative variability of concentration, essential for assessing
mixing efficiency in microfluidic systems. A lower CV indicates higher uniformity in the
mixing process, signifying efficient mixing within the microfluidic mixer. CV can be derived
as follows:
The mean concentration (µ) is calculated as the average of the species concentration at
each discrete point:
µ =
1
N
X
N
i=1
Ci (1.14)
where Ci represents the concentration at each point, and N is the total number of points.
The variance (σ
2
) and standard deviation (σ) are measures of the spread of the concentration values around the mean:
σ
2 =
1
N
X
N
i=1
(Ci − µ)
2
, σ =
√
σ
2
(1.15)
Finally, the Coefficient of Variation (CV) is the ratio of the standard deviation to the
mean:
CV =
σ
µ
(1.16)
9

Expressed as a percentage, the CV provides a dimensionless indicator of the relative variability of the concentration distribution.
1.6 Mixing Efficiency Analysis
Based on the simulation results, and using the Coefficient of Variation (CV) as a measure
of mixing efficiency, the variance was determined to be approximately 1.49 × 10−7 mol2/m4
,
calculated from the squared differences between the local concentration and the mean concentration over the specified surface. The standard deviation was around 3.86×10−4
, leading
to a CV of approximately 0.05. This low value of CV indicates a high degree of uniformity
in the mixing process, signifying efficient and consistent mixing of the species within the
system.
1.7 Implications and Significance
The results of this simulation have significant implications. The accurate estimation of diffusivity for an ionic liquid in a solvent like ethylene glycol advances our understanding of fluid
behavior in microfluidic applications. The low coefficient of variation in the mixing efficiency
analysis showcase the capability of the simulation to predict and analyze the effectiveness of
mixing processes.
10

(a) Surface concentration (b) Surface velocity
(c) Concentration profile (d) Velocity magnitude
Figure 1.3: Microfluidic analysis showing (a) surface concentration, (b) surface velocity, (c)
concentration profile, and (d) velocity magnitude.
11

1.8 Applications of Microfluidics in Biomolecular systems: A Case Study on Cholesterol’s Impact on Carbon Dioxide Permeability Across Lipid Membranes
Microfluidic technology stands out for its ability to manipulate small volumes of fluids within
micro-scale channels. In the realm of biomolecular research, microfluidics has emerged as
a revolutionary technology, offering new avenues to explore complex biomolecular processes
with exceptional precision and control. This cutting-edge technique has proven especially
valuable in studying the dynamics of cellular environments that closely mimic natural biological systems.
Recently, a study focusing on the role of cholesterol in the permeability of carbon dioxide
(CO2) across lipid membranes was presented. This study leveraged continuous flow microfluidics and fluorescence microscopy to unravel the effects of cholesterol on CO2 transport
through lipid bilayers. Employing Giant Unilamellar Vesicles (GUVs) as a simplified model
for cell membranes, a microfluidic-based fluorescence microscopy platform was utilized to
precisely observe and quantify how changes in cholesterol levels alter CO2 permeability.
1.8.1 Methodology
In this work, a microfluidic-based microscopy assay was utilized to investigate the permeability of carbon dioxide (CO2) across lipid bilayers enriched with varying concentrations
of cholesterol.16 The core of the experimental setup involved the use of Giant Unilamellar
Vesicles (GUVs), which were formed using the electroformation method. These vesicles, composed of a lipid bilayer, were designed to mimic cellular membranes. The GUVs were placed
in a custom-designed microfluidic device, allowing for precise control and manipulation of
the fluidic environment. To measure the permeability of CO2 across these lipid membranes,
a fluorescence-based approach was implemented in which a CO2-sensitive fluorescent dye was
added into the vesicles, which altered its fluorescence intensity based on the concentration
of CO2 within the vesicle. By monitoring the fluorescence intensity over time using high12

resolution microscopy, the permeability of CO2 into the GUVs was determined. This setup
provided a quantifiable and visual method to assess how cholesterol levels in the lipid bilayers influenced the permeation of CO2. The permeability was calculated based on the rate of
change in fluorescence intensity, which directly correlated with the CO2 concentration inside
the vesicles. This combination of microfluidics, fluorescence microscopy, and GUVs allowed
for a highly controlled and accurate assessment of gas permeability in a biomimetic system,
paving the way for a deeper understanding of the role of cholesterol in cellular gas transport
processes. The permeability of CO2 across phospholipid membranes was investigated for a
range of cholesterol concentrations. The fluorescence decay, normalized to the initial point,
adhered to an exponential model as shown in Equation 1.17:
Figure 1.4: Illustrative schematic of the experimental approach. A buffer with a high CO2
content and another containing vesicles with pH-responsive fluorescent probes meet in a Yshaped channel. CO2 transits the lipid bilayer of the vesicles, causing an internal pH drop and
diminishing fluorescence. Continuous imaging is facilitated by the serpentine microchannel,
which allows for prolonged observation of the vesicle without immobilization. The composite
fluorescence image displays a strobe-like series of snapshots, taken at 20-millisecond intervals,
documenting the temporal evolution of the vesicle’s fluorescence.
13

Figure 1.5: Diagram illustrating the cellular mechanisms of CO2 transport and pH regulation in extracellular fluid. CO2 diffuses into the cell, reacting with water to form carbonic
acid, which dissociates into bicarbonate and protons, mediated by carbonic anhydrase. This
process decreases intracellular pH and is balanced by bicarbonate exchange with the extracellular environment.
1.8.2 Results & Discussion
The permeability of CO2 across phospholipid membranes was investigated for a range of
cholesterol concentrations. The fluorescence decay, normalized to the initial point, adhered
to an exponential model as shown in Equation 1.17:
I(t) = I0e
−t/τ + I∞ (1.17)
In the above equation, I(t) represents the fluorescence intensity at time t, I0 is the initial
fluorescence intensity, τ is the decay time constant, and I∞ is the fluorescence intensity at
equilibrium. For a vesicle with a 40-60% POPC-cholesterol composition, the parameters
14

were determined as 1/τ = 1.58 s
−1 and I∞ = 0.63. (Figure 1.6)
Figure 1.6: Exemplary data from a vesicle with composition 40-60 POPC-chol. The fluorescence has been normalized to the first point. Data points are derived from each frame
of the acquisition. The breaks in the data arise from the vesicle leaving the field of view as
it traverses the channel. The red line is a fit to Equation 1.17, with a 1/τ = 1.58 s−1 and
I∞ = 0.63.
A simplistic transport model based on the permeability constant P and Equation 1.18,
given by
Cin = Cout
1 − e
−S
Pmt
V

, (1.18)
where Cin/out is the CO2 concentration inside/outside the vesicle, S is the surface area, and
V is the volume. The internal concentration of CO2 increases at a slower rate than what
equation 1.18 suggests due to a significant portion of CO2 converting into carbonate upon
hydration. The formula for the internal concentration, Cin is adjusted to reflect this and
includes the pKa value of CO2, denoted as pK. The refined model, accounting for CO2
hydration, is given by:
Cin = Cout
1 − e
−SPmt
V (1+10pH−pK)

, (1.19)
15

This equation does not measure CO2 directly; instead, the observable data is linked
to the concentration of HPTS that has lost a proton since a substantial fraction of CO2
will be hydrated to form carbonate. Therefore, to account for the pH changes and the
buffering capacity, the model was refined and numerically solved with rate constants sourced
from literature. The fitting of the experimental data to the refined model allowed for the
estimation of permeability constants from the observed fluorescence decay.
A suite of simulated data was generated, varying permeability to correlate with the fluorescence proportional to the concentration of unprotonated HPTS. The parameters 1/τ and
I∞ were deduced by fitting the simulated data to the exponential decay function. A numerical relationship was established to interpolate permeabilities from a measured decay rate.
The permeability constant for each vesicle was individually determined, and the standard
error of the mean was calculated based on the variance observed under specific conditions.
An offset of 0.06 s was incorporated to model the mixing time, as determined by the transit
time of vesicles from the Y-junction to the serpentine path in the experimental setup.
An important observation from the study is the permeability of bilayers with varying
cholesterol levels. It was found that the permeability of bilayers with high cholesterol levels
is significantly lower compared to those without cholesterol. Specifically, bilayers containing
significant cholesterol levels exhibited permeability values around 9.9 ± 1.0 × 10−4
cm/s. In
contrast, bilayers without cholesterol demonstrated permeability values of 9.6 ± 1.4 × 10−3
cm/s. This indicates that the presence of cholesterol in bilayers can reduce their permeability
by an order of magnitude.
1.8.3 Conclusion
This case study not only sheds light on the important role of cholesterol in cellular gas
transport but also exemplifies the power of microfluidics in biological research. By enabling
detailed and controlled studies of membrane dynamics, microfluidic technology is enhancing
our understanding of fundamental biological processes. As research in this field continues to
16

evolve, microfluidics is poised to remain at the forefront, driving discoveries and innovations
that deepen our comprehension of life at the molecular level.
17

References
[1] G. Whitesides. “The origins and the future of microfluidics”. In: Nature 442 (2006),
pp. 368–373.
[2] S. C. Terry, J. H. Jerman, and J. B. Angell. “A gas chromatographic air analyzer fabricated on a silicon wafer”. In: IEEE Trans. Electron Devices 26.12 (1979), pp. 1880–
1886.
[3] A. Manz, N. Graber, and H. M. Widmer. “Miniaturized total chemical analysis systems: A novel concept for chemical sensing”. In: Sensors and Actuators B: Chemical
1.1–6 (1990), pp. 244–248. issn: 0925-4005.
[4] D. J. Harrison et al. “Micromachining a miniaturized capillary electrophoresis-based
chemical analysis system on a chip”. In: Science 261.5123 (1993), pp. 895–897. doi:
10.1126/science.261.5123.895.
[5] A. M. Ghaemmaghami et al. “Biomimetic tissues on a chip for drug discovery”. In:
Drug Discov. Today 17.3–4 (2012), pp. 173–181.
[6] P. J. Kitson et al. “Configurable 3D-Printed Millifluidic and Microfluidic ‘Lab on a
Chip’ Reactionware Devices”. In: Lab Chip 12.18 (2012), pp. 3267–3271.
[7] Neil Convery and Nikolaj Gadegaard. “30 years of microfluidics”. In: Micro and Nano
Engineering 2 (2019), pp. 76–91. issn: 2590-0072.
[8] B. K. Gale et al. “A Review of Current Methods in Microfluidic Device Fabrication
and Future Commercialization Prospects”. In: Inventions 3.3 (2018), p. 60.
[9] Andrea Zanella and Andrea Biral. “MICROFLUIDIC NETWORKING: MODELLING
AND ANALYSIS”. In: (2012). url: https://api.semanticscholar.org/CorpusID:
14474657.
[10] Microfluidics for Pharmaceutical Applications: From Nano/micro Systems Fabrication to Controlled Drug Delivery. Micro and Nano Technologies. 2019, pp. 79–100.
18

[11] A. Günther and K. F. Jensen. “Multiphase microfluidics: From flow characteristics
to chemical and materials synthesis”. In: Lab Chip 6.12 (2006), pp. 1487–1503. doi:
10.1039/b609851g.
[12] Emily J. Roberts et al. In: ACS Applied Materials & Interfaces 11.31 (2019), pp. 27479–
27505.
[13] Pooyan Tirandazi and Carlos H. Hidrovo. In: J. Micromech. Microeng. 27 (2017),
p. 075020.
[14] Abraham D Stroock et al. “Chaotic mixer for microchannels”. In: Science 295.5555
(2002), pp. 647–651. doi: 10.1126/science.1066238. url: https://pubmed.ncbi.
nlm.nih.gov/11809963/.
[15] Jinlei Cui et al. “Diffusivity and Structure of Room Temperature Ionic Liquid in
Various Organic Solvents”. In: The Journal of Physical Chemistry B 124.44 (2020),
pp. 9931–9937. doi: 10.1021/acs.jpcb.0c07582.
[16] Matthew C Blosser et al. “Effect of cholesterol on permeability of carbon dioxide
across lipid membranes”. In: bioRxiv (Nov. 2020). Preprint. doi: 10.1101/2020.11.
16.384958.
19

Chapter 2: Continuous Flow Reactors for Synthesizing Colloidal Nanoparticles
2.1 Introduction
Catalysts are central to modern manufacturing. Estimates are that 85% of all commercially
produced chemical products involve a catalyst at some stage in the process of their manufacture and that approximately 80% of these catalytic processes involve solid heterogeneous
catalysts.1 Therefore, the demand for developing catalysts with higher selectivity, activity, and stability is still ongoing. Nanostructuring of materials has been explored over the
years to leverage the high surface-area-to-volume ratio for catalysis. The growing interest in
nanotechnology over the last few decades has led to an increasing demand in inorganic nanomaterials. However, nanomaterial manufacturing relies to a large extent on batch processes.
Industrial syntheses of colloidal nanoparticles in batch processes are typically achieved by
precipitation or impregnation methods. While these widely adopted methods are scalable,
and cost effective, they rely on supports, and suffer from challenges in obtaining high loadings in a single step as the impregnation method is limited by the adsorption capacity of the
support. In addition, they are more susceptible to a wide distribution of particle size and
morphologies where the inhomogeneous nucleation and growth rates that result from local
thermal inhomogeneities introduce batch-to-batch variability.2 The structure of the synthesized nanoparticle is directly correlated to its physical and chemical properties, which is an
important aspect of the nanoparticle catalytic activity. Therefore, enhanced control of heat
and mass transfer is essential for optimized structure-function relationship.3 Continuous flow
methods are superior to conventional batch methods in that regard where the control of reaction parameters is more precise.2 The surface-to-volume ratio in confined channels is very
20

high, so any chemical reaction in a smaller channel is greatly accelerated. Reaction times in
milli- and microfluidics systems are also much quicker than conventional batch devices due
to the smaller dimensions of the systems leading to a shorter diffusion time for any given
molecule.
This chapter aims to provide a comprehensive understanding of continuous flow reactors
in the context of nanoparticle synthesis. It will explore the fundamental principles, types of
reactors, flow control mechanisms, and the impact of flow regimes on nanoparticle properties.
A comparative analysis between batch and continuous flow processes, alongside real-world
case studies, will be presented to illustrate the advantages and challenges in scaling up
nanoparticle synthesis using continuous flow reactors.
2.2 Reactor Types for Flow Syntheses
In flow chemistry, chemical reactions occur in a continuously flowing stream rather than in
batch production. Fluids are moved into a reactor by pumps, where they react upon contact.
This technique is applicable at large scales for manufacturing and has also been adapted
for small-scale laboratory use. The common flow reactors used in milli- and micro flow
syntheses are plate and tubular reactors. The type of reactor used depends on the respective
application. Important factors include material selection, reactor geometry, and scalability.
These considerations directly impact the efficiency, selectivity, and reaction speed of the
reactors. Material choice ensures chemical resistance and durability, while reactor geometry
affects mixing efficiency and heat transfer.2
Despite low throughput and their tendency to clog, plate reactors offer outstanding heat
transfer characteristics due to their high surface area-to-volume ratios.4 Millifluidic plate
reactors offer higher throughput than traditional microfluidic plate reactors at the expense
of lower surface area-to-volume ratios. Plate reactors are usually machined from silicon,
glass, or stainless steel. The choice of material depends on the respective application under
investigation. Plate reactors are particularly useful in reactions requiring rapid heating
21

or cooling and are often used in applications involving highly exothermic or endothermic
reactions. While plate reactors offer many advantages, their expensive fabrication and limited
throughput make them less suitable for large scale manufacturing.2
Figure 2.1: Sperpentine-shaped milli-fluidic plate reactor (ca. 1.8 mL volume) made of
borosilicate glass.
Due to the high cost of fabricating plate reactors and their inherent limitations, tubular
or coil reactors have emerged as widely used alternative for flow syntheses. Commercial
tubular reactors that are commonly used in flow chemistry applications are usually made
of inert fluoropolymers (PTFE, PFA, and FEP) or stainless steel (SST). Commonly, these
tubings have outer diameters of 1/16 inch or 1/8 inch with various inner diameters ranging
from 0.01 inch to 1/16 inch.4 The selection of the appropriate material depends on the
experimental requirement or limitations. For instance, high temperature syntheses over ca.
220 °C prevent the usage of fluoropolymers due to the thermal limitations. For example,
Polytetrafluoroethylene, (PTFE) begins to deteriorate at 260 °C and decomposes above 350
°C. In such scenarios, glass coiled reactors or SST reactors are usually used. Table 2.1 gives a
summary of the approximate operating temperature and pressure range for common tubular
materials used in flow chemistry.2
22

Table 2.1: Application Range of Tubular Reactors Materials
Application PTFE PFA FEP SST
Low T, P (< 50◦C, < 5 bar) ✓ ✓ ✓ ✓
High T, P (< 200◦C, < 20 bar) ◦ ◦ ◦ ✓
Very high T, P (> 220◦C, > 20 bar) × × × ✓
*✓= acceptable to use; ◦= some concerns (due to material degradation or melting point); ×=not feasible.
PTFE = Polytetrafluoroethylene; PFA = Perfluoroalkoxy alkane; FEP = Fluorinated ethylene propylene;
SST = stainless steel
One major advantage that makes tubular reactors more robust is the flexibility of switching and controlling the reactor volume, in contrast to plate reactors where the dimensions
have to be predetermined prior to fabrication. Connections often or are often made of
Polyether ether ketone (PEEK), which is a thermoplastic with excellent mechanical and
chemical resistance properties. PEEK has a melting temperature of around 343 °C and is
resistant to attacks by both organic and aqueous environments. This robustness along with
its low cost makes PEEK an ideal candidate for setting up various continuous flow systems.5
2.3 Flow Regimes and Their Impact on Nanoparticle Synthesis
In micro- and milli-scale continuous flow reactors, the process of scaling down from larger
systems leads to a significant shift in the flow dynamics. This change is primarily attributed
to an enhanced surface-area-to-volume ratio in the fluid channels. Despite the miniaturization, the basic fluid dynamics principles remain consistent with those observed in larger,
macroscale systems. However, the interaction between the fluid and the channel walls becomes more pronounced due to the smaller scale, influencing the type of flow achieved.
Two main flow regimes are observed: laminar and turbulent. Laminar flow, typically
dominant in these smaller systems, features fluids moving in parallel, streamlined paths.
Turbulent flow, common in larger systems, involves chaotic and irregular motions that facilitate mixing in all directions, a result of vortices and flow fluctuations. The governing forces
for these regimes are different: laminar flow is primarily driven by viscous forces, whereas
23

turbulent flow results from dominating inertial forces.
Mixing in laminar flow is dependent on diffusion, with the rate of mixing being inversely
proportional to the diffusivities of the species involved. Turbulent mixing, though inherently
more complex, accelerates the process by integrating molecular diffusion with flow properties
that minimize diffusion length and resemble convective mixing.
To improve mixing efficiency in microfluidic environments, several active and passive
mixing strategies have been developed. Additionally, employing multiphase flows in these
channels can substantially reduce the time required for effective mixing
2.4 Single-Phase Flow
Achieving laminar flow mixing in continuous flow reactors is straightforward, typically using
either a T- or Y-shaped junction. In this setup, two miscible fluids converge into a single
channel, flowing parallel to each other. The mixing of these fluids primarily occurs through
diffusion across their interface. A key aspect of laminar flow is its parabolic velocity profile,
where the fluid speed is zero at the channel walls due to the no-slip boundary condition and
reaches maximum in the channel’s center. This velocity distribution results from the friction
between the fluid and the channel walls.
However, this parabolic flow pattern can pose challenges in nanoparticle synthesis. The
variation in nucleation times due to different fluid velocities across the channel can lead
to a broader particle size distribution. Additionally, the diffusion-driven mixing can create
concentration gradients both across and along the reactor, potentially resulting in uneven
reaction conditions. Moreover, the slower fluid velocity near the channel walls increases the
risk of fouling during nanoparticle synthesis, which will be explored further in section 3.1.
The mixing time in laminar flow reactors is directly proportional to the square of the channel
diameter and inversely proportional to the diffusivity of the mixed species.
To improve mixing in laminar flow, various advancements have been made. One approach
is to laminate the fluid streams into thin, interdigitated layers, reducing the distance required
24

for stream interdiffusion and hence, shortening the necessary reactor length for effective
mixing.6 Another technique involves designing the reactor architecture to induce chaotic
advection (see CFD example in Chapter 1). This can be done by introducing obstacles
in the channel, modifying the channel wall with ridges or grooves, or altering the channel
geometry into a zigzag or serpentine pattern, which disrupts the flow and introduces a
transverse velocity component.7
2.5 Multi-Phase Flow
In the realm of flow chemistry, significant advancements have been made to enhance mixing
within flow reactors, notably through the utilization of multiphase or segmented flow. This
approach is particularly effective in systems involving liquid-liquid or liquid-gas combinations, where immiscible phases are used to create distinct droplets or plugs of one phase
within another. In these setups, the reagents form dispersed-phase droplets or plugs, separated by an immiscible carrier phase, either liquid or gas. These droplets serve as consistent,
miniaturized reaction vessels, with volumes typically in the nanoliter to microliter range, ensuring uniform reagent concentrations. The continuous phase around these droplets not only
isolates them from each other but also from the channel walls. As these droplets traverse
downstream, the interaction with the stationary channel walls generates countercirculating
flows within them. The efficiency of this process can be further enhanced in curved channels,
which induce asymmetric vortices, akin to Dean flow, leading to improved mixing.
Droplet formation is commonly achieved using flow-focusing devices or T-junctions. In
flow-focusing setups, droplets are formed as the dispersed phase passes through an orifice,
flanked by immiscible fluid streams, leading to encapsulation by the higher-viscosity continuous phase under certain flow rate conditions. In contrast, T-junctions create droplets
by intersecting the dispersed phase with the continuous phase at a right angle, where the
flow of the continuous phase applies a viscous force to break apart the dispersed phase into
droplets, contingent on specific flow rates and ratios.
25

For applications requiring isolation of reagent streams before droplet formation, such as
in certain nanoparticle syntheses, multiple-inlet T-junctions are used. An example of this
is the synthesis of gold and silver nanoparticles, where metal salts and reducing agents are
introduced through separate inlets, with an additional solvent stream creating a barrier to
delay mixing and nucleation until droplet formation occurs.8
The process of droplet generation is influenced by several factors, including the interfacial
tension between the phases, the wetting properties of the channel walls, flow rates, their
ratios, and the fluid viscosities. Stable droplet formation typically requires good wettability
of the continuous phase inside the channel, governed by the interfacial energies between the
solid and the liquid. Reactor design, therefore, must consider solvent choices in relation to
channel surface properties, which can be modified through surface coatings.
Recent innovations, such as 3D-printed droplet generators, have shown the ability to
produce droplets across a wide range of inlet flow rate ratios and channel surface energies,
without affecting droplet diameter. This design allows for droplets to be formed perpendicular to the outlet, with the droplet size being determined by the outlet tubing’s inner
diameter, a feature that can be easily modified. Such advancements in fluidic component
design are propelling continuous flow manufacturing to new heights.9
2.6 Control of Fluid Flow in Reactors
Managing the flow of dispersed and continuous phases in flow reactors is accomplished
through various techniques. Syringe pumps are predominantly used for controlling flow
in micro- and milliscale reactors, particularly in lab settings. These pumps work by mechanically pushing the plunger of a mounted syringe to generate a specific flow rate for a set
volume of solution, offering simplicity in both setup and operation. Ideally, the flow rate set
on the pump should match the actual flow within the reactor; however, discrepancies often
arise due to factors like thermal expansion and in situ gas evolution, which may necessitate
additional flow monitoring tools or methods.
26

Syringe pumps can also face challenges in scale-up scenarios. For example, excessive
back-pressure may cause stalling, especially in cases of channel clogging. Their application
in large-scale operations is limited by the maximum syringe volume that fits into commercially available pumps. To circumvent these limitations, microfabricated pumps have been
introduced. These are driven by forces like piezoelectric, electromagnetic, centrifugal, or
pneumatic, allowing for a compact design. While micropumps excel in precise control of
small fluid volumes and are crucial for effective micromixing in nanofabrication, their relatively low total flow rates limit their scalability.10
Peristaltic pumps offer an alternative for fluid delivery, using a flexible tube that is
periodically squeezed to create negative pressure and propel the fluid. They share the ease
of operation with syringe pumps but without the volume limitations, making them more
suitable for larger-scale applications. However, they can introduce flow oscillations or pulses,
potentially affecting nanoparticle synthesis quality. Additionally, the chemical compatibility
of materials used in peristaltic pumps is crucial, especially for organic solvents, with special
materials like fluoroelastomers and silicones developed for handling aggressive chemicals.
High-performance liquid chromatography (HPLC) pumps are also used in continuous
flow setups. These pumps operate on a reciprocating piston mechanism, similar to peristaltic pumps, and are known for their precise flow control and high operating pressures,
although they are more costly. PEEK tubing, commonly used in HPLC systems, has material compatibility considerations with certain solvents.
Pressurized gas-based systems have gained recognition for their ability to maintain stable
flows over extended durations, suitable for large-volume fluid handling. These systems don’t
produce flow pulses due to the lack of mechanical pumping, but they may face challenges like
reagent contamination or evaporation in open environments. Careful selection of materials
and safety valves is essential to mitigate risks from corrosive vapors or gases. Real-time flow
rates in such systems can be accurately controlled using flow sensors, despite the flow rate
not being directly proportional to delivery pressure.
27

2.7 Transport Phenomena in Batch and Flow Reactors
In considering quantitative comparisons of transport phenomena between batch and continuous flow reactors, it’s important to consider aspects such as mass and heat transfer, reaction
kinetics, and mixing efficiency. In the context of nanoparticle synthesis, these factors significantly influence the control over particle size distribution, morphology, and reaction yields.11
In continuous flow reactors, the superior heat and mass transfer is a result of the high
surface-area-to-volume ratios of the micro- and millifluidic channels. In batch reactors, however, heat and mass transfer are often less efficient due to larger reaction volumes and reduced
surface interaction, leading to gradients in temperature and concentration. Convective mass
and heat transfer dominate in batch reactors, requiring high Re and optimized impeller geometries for thorough mixing. In contrast, microfluidic reactors enable more efficient heat
exchange and can mitigate temperature fluctuations from endo- or exothermic reactions,
prevalent in nanoparticle syntheses. Efficient heat transfer and effective passive mixing in
flow synthesis often result in nanoparticles requiring less time for reactions and achieving
higher yields compared to similar batch processes.9, 12, 13, 14
The reaction kinetics in continuous flow systems are often superior due to the controlled
environment. The Damköhler number (Da), which describes the ratio of reaction rate to
mass-transfer rate, is a critical parameter in this context. Da < 1 signifies a system limited
by reaction kinetics, where mixing occurs rapidly enough, whereas Da > 1 suggests that the
system is limited by mixing, potentially leading to undesired products and reduced yield.
In batch reactions, as the reactor size or reagent concentration increases, the control over
mixing and heating profiles decreases, often leading to lower quality materials with wider size
distributions and poorly defined morphologies. In contrast, micro- and millifluidic reactors
maintain a consistent Damköhler number even when scaled up, either by increasing run time
or parallelization, allowing for more consistent reaction kinetics and product quality15
28

2.8 Scaling Nanoparticle Syntheses
Continuous flow methods are increasingly viable for commercializing colloidal nanoparticle
synthesis due to their compatibility with automation and parallel processing. These methods
offer enhanced throughput and consistent results, addressing the limitations often encountered in batch processing. Additionally, the reduction in the volumes of solvents and reagents
used, which are exposed to specific temperatures at any given time, is smaller compared to
large tank reactors, leading to improved safety.
Riche et al.9 demonstrated a scaling solution for nanoparticle syntheses by introducing a
3D droplet-generating device that maintains consistent droplet sizes despite flow rate variations, crucial for scaling up batch chemistries to continuous flow systems. This technology is
adaptable for producing a wide range of droplet volumes, addressing traditional challenges
in nanoparticle synthesis like uniform fluidic behavior and channel fouling. Its application
in synthesizing platinum nanoparticles using ionic liquid solvents demonstrates doubled reaction yields compared to batch processes and highlights its potential for high-throughput,
parallelized nanoparticle production (Figure 2.2).
Figure 2.2: Illustration of advanced milli- and microfluidic systems for scaling nanoparticle
synthesis: (Left) Schematic of the self-optimizing parallel millifluidic reactor with the control and feedback system for nanoparticle synthesis;16 (Right) (a) Schematic of the parallel
network assembled by connecting a distribution manifold to four droplet generators, showing
a gradient of resistances across the branches. (b) Comparison of droplet diameters produced
under different flow rates, illustrating the transition between flow invariant and flow dependent regimes.9
29

Wang et al 16 reported a 16-channel millifluidic reactor that employs a multiphase gas–liquid
flow to continuously produce colloidal cesium lead bromide (CsPbBr3) quantum dots with
a throughput of approximately 1 liter per hour. The novel aspect of this reactor is its ability to self-optimize the synthesis process in real-time by monitoring the photoluminescence
characteristics of the quantum dots in situ, ensuring optimized reaction conditions and consistent product quality. The system is designed to overcome the limitations of segmented-flow
micro- and millifluidic reactors (SMRs), such as low throughput and compromised reaction
efficiency when scaled up for industrial production. The parallelization strategy of the reactor maintains the benefits of SMRs, including superior heat and mass transfer and excellent
reproducibility, while also achieving a higher throughput through linear scaling of the number
of channels.
This reactor represents the first instance of a high-throughput nanomaterial synthesis in a
feedback-controlled continuous flow parallel reactor, showcasing a significant step towards the
scaled-up production of photoluminescent quantum dots. With its high chemical resistance,
highly parallel slug generation, robust optimization algorithms, and substantial throughput,
this reactor has broad applications for the scaled-up manufacturing of various nanomaterials.
30

References
[1] Z. Ma and F. Zaera. “Title of the Chapter”. In: Encyclopedia of Inorganic and Bioinorganic Chemistry. Ed. by R. A. Scott. John Wiley Sons, Ltd., 2014.
[2] Matthew B. Plutschack et al. “The Hitchhiker’s Guide to Flow Chemistry”. In: Chemical Reviews 117.18 (2017), pp. 11796–11893.
[3] Emily J. Roberts et al. “Title of the Article”. In: ACS Applied Materials Interfaces
11.31 (2019), pp. 27479–2750.
[4] V. Hessel, J. C. Schouten, and A. Renken. Micro Process Engineering: A Comprehensive Handbook. John Wiley Sons, 2009.
[5] P. A. Gunatillake and R. Adhikari. “Nondegradable synthetic polymers for medical
devices and implants”. In: Biosynthetic Polymers for Medical Applications. Woodhead
Publishing, 2016, pp. 33–62.
[6] V. Hessel et al. “Laminar mixing in different interdigital micromixers: I. Experimental
characterization”. In: AIChE Journal 49.3 (2003), pp. 566–577. doi: https://doi.
org/10.1002/aic.690490304.
[7] Abraham D Stroock et al. “Chaotic mixer for microchannels”. In: Science 295.5555
(2002), pp. 647–651. doi: 10.1126/science.1066238.
[8] Laura L. Lazarus et al. “Two-Phase Microfluidic Droplet Flows of Ionic Liquids for the
Synthesis of Gold and Silver Nanoparticles”. In: ACS Applied Materials & Interfaces
4.6 (2012), pp. 3077–3083. doi: 10.1021/am3004413.
[9] C. T. Riche et al. “Flow Invariant Droplet Formation for Stable Parallel Microreactors”. In: Nat. Commun. 7 (2016), p. 10780. doi: 10.1038/ncomms10780.
[10] Sung-Yi Yang et al. “Size-controlled synthesis of gold nanoparticles using a micromixing system”. In: Microfluidics and Nanofluidics 8.3 (2010), pp. 303–311. doi:
10.1007/s10404-009-0461-2.
31

[11] Ryan L Hartman, Jonathan P McMullen, and Klavs F Jensen. “Deciding whether to
go with the flow: evaluating the merits of flow reactors for synthesis”. In: Angewandte
Chemie International Edition 50.33 (2011). Epub 2011 Jun 27. PMID: 21710673,
pp. 7502–7519. doi: 10.1002/anie.201004637.
[12] E. J. Roberts et al. “High-Throughput Continuous Flow Synthesis of Nickel Nanoparticles for the Catalytic Hydrodeoxygenation of Guaiacol”. In: ACS Sustainable Chem.
Eng. 5 (2017), pp. 632–639. doi: 10.1021/acssuschemeng.6b01888.
[13] C. D. Ahrberg, J. W. Choi, and B. G. Chung. “Droplet-Based Synthesis of Homogeneous Magnetic Iron Oxide Nanoparticles”. In: Beilstein J. Nanotechnol. 9 (2018),
pp. 2413–2420. doi: 10.3762/bjnano.9.227.
[14] B. F. Cottam et al. “Accelerated Synthesis of Titanium Oxide Nanostructures Using
Microfluidic Chips”. In: Lab Chip 7 (2007), pp. 167–169. doi: 10.1039/B615688C.
[15] Robert W. Epps et al. “Automated microfluidic platform for systematic studies of
colloidal perovskite nanocrystals: towards continuous nano-manufacturing”. In: Lab
Chip 17 (23 2017), pp. 4040–4047. doi: 10.1039/C7LC00884H. url: http://dx.doi.
org/10.1039/C7LC00884H.
[16] Lu Wang et al. “Self-optimizing parallel millifluidic reactor for scaling nanoparticle synthesis”. In: Chem. Commun. 56 (26 2020), pp. 3745–3748. doi: 10 . 1039 /
D0CC00064G. url: http://dx.doi.org/10.1039/D0CC00064G.
32

Chapter 3: Throughput Optimization of Molybdenum Carbide Nanoparticle Catalysts in a Continuous Flow Reactor using Design of Experiment*
Karadaghi, L. R.†
, Madani, M. S.†
, Williamson, E. M.†
, To, A. T., Habas, S. E., Baddour,
F. G., Schaidle, J. A., Ruddy, D. A., Brutchey, R. L., Malmstadt, M
*This chapter was adapted from a published work: ACS Applied Nano Materials 2022 5 (2),
1966-1975.
†These authors contributed equally to this work.
3.1 Abstract
Transition metal carbides (TMCs) have attracted significant attention because of their applications toward a wide range of catalytic transformations. However, the practicality of their
synthesis is still limited because of the harsh conditions in which most TMCs are prepared.
Recently, a solution-phase synthesis of phase-pure -MoC1–x nanoparticles was presented.
While this synthetic route yielded nanoparticles with exceptional catalytic performance, the
reaction parameter space was not explored, and catalyst throughput was not optimized for
scale-up. Continuous flow platforms coupled with statistical design of experiments (DoE) can
provide a powerful method for understanding the reaction parameter space for optimizations.
Here, we demonstrate the use of statistical DoE in tandem with response surface methodology for a parametric screening analysis to optimize the throughput of a MoC1–x nanoparticle
synthesis utilizing a millifluidic flow reactor. A full factorial design was implemented to evaluate four input variables (reaction temperature, flow rate, solvent fraction of oleylamine,
and precursor concentration) that carry statistically significant effects on three responses
33

(throughput, residence time, and isolated yield). A Doehlert matrix was implemented to
investigate each significant variable at a higher number of levels to optimize throughput.
Our results give a nonintuitive set of experimental conditions that resulted in an optimized
throughput of 2.2 g h–1. This translates to a 50-fold increase in throughput compared to the
previously reported batch method. The catalytic performance of the MoC1–x nanoparticles
produced under optimized throughput was demonstrated in the CO2 hydrogenation reaction. This DoE screening analysis and throughput optimization of MoC1–x synthesis open
the door to an increased feasibility for scale-up.
3.2 Introduction
The increasing demand for sustainable routes to produce fuels and chemicals motivates
the development of low-cost, Earth-abundant catalysts that maintain stability under catalytic conditions.1 Transition metal carbides (TMCs) possess inherent multifunctionality
that enables exceptional catalytic performance for a wide range of transformations, including hydrogenation,2, 3, 4, 5, 6 isomerization,7, 8, 9, 10 deoxygenation,11 and hydrodeoxygenation
reactions.12 Levy and Boudart were among the first to report that TMCs possess similarities in catalytic behavior and electronic structure to noble metals, such as Pt.13 However,
these early TMCs were orders of magnitude less active than noble metals, driving efforts
to synthesize TMCs with higher active surface areas. High-temperature carburization has
been demonstrated as a route to synthesize higher surface area TMC powders; however,
this approach requires harsh conditions with temperatures typically exceeding 600 °C that
limit the practicality of high-throughput industrial-scale manufacturing.14, 15, 16Molybdenum
carbide (particularly the α-MoC1−x and β-Mo2C phases) is an attractive TMC because of
superior reactivity and the potential to obtain higher specific surface areas (177–210 m2
g
−1
).17, 18, 19 While synthetic routes to high-surface area α-MoC1−x have been demonstrated,
they still rely on a high-temperature carburization step.20 We recently addressed some of
these shortcomings by developing the first mild solution-phase synthetic route to produce
34

phase-pure α-MoC1−x nanoparticles.21 This synthetic approach was enabled by the thermolytic decomposition of Mo(CO)6 at comparatively low temperatures (i.e., 320 C) and was
further extended to other TMC nanoparticles, including β-WC, using an analogous W(CO)6
precursor. The resulting α-MoC1−x nanoparticle catalysts exhibited a twofold increase in
activity on a per-site basis and an increase in selectivity toward C+
2 hydrocarbon products,
as compared to a lower surface area bulk α-MoC1−x catalyst for the thermocatalytic hydrogenation of CO2.
21 Despite the mild synthetic conditions employed to make this catalyst,
the volumetric scaling of nanoparticle synthesis by batch processes is not ideal because of
batch-to-batch variability and nonuniform reactor conditions.22 More specifically to the batch
MoC1−x nanoparticle synthesis, Mo(CO)6 precursor sublimation contributes to relatively low
isolated yield, and significant gas evolution during the reaction introduces safety concerns,
ultimately making process intensification and scale-up untenable. Consequently, this labscale batch reaction resulted in a relatively low isolated yield of 40–50% after a 1 h reaction
time, for a throughput of only ca. 0.047 g h−1 per reaction, reflecting the by-hand nature of
the process.21
Continuous flow systems for nanoparticle syntheses are an attractive alternative approach that eliminates many of the aforementioned problems encountered in traditional
batch processes.23 The superior heat and mass transport characteristics provided by the
small, millimeter-scale channels allow for greater control over heating by virtue of reduced
thermal masses and higher surface area-to-volume ratios. In contrast to batch syntheses,
where the inhomogeneous nucleation and growth rates that result from local thermal inhomogeneities introduce batch-to-batch variability (which is only further exacerbated when
volumetric scale-up is attempted), continuous flow approaches enable high-temperature reactions to be performed with improved product consistency and reproducibility,22, 24, 25, 26, 27
while generating higher product yields with shorter reaction times.21, 28, 29 Our previous unoptimized attempt to synthesize MoC1−x nanoparticles in flow resulted in a throughput of
0.77 g h−1
.
21
35

While our previous study demonstrated the successful synthesis of MoC1−x nanoparticles via continuous flow methods, the synthetic parameter space was undefined, resulting
in an unoptimized system. The experimental variable space for nanoparticle syntheses is
high dimensional and complex by nature, and, when coupled with the additional operational
parameters of a flow reactor, optimizations become nearly impossible via traditional one
variable at a time (OVAT) methods, which are prohibitively slow and costly for all but the
simplest reactions. With that being said, continuous flow methodologies are amenable to
the rapid screening of reaction conditions for the multivariate optimization of synthetic conditions.30, 31, 32 High-throughput experimentation has been used to increase the efficiency of
such investigations, including self-optimizing flow reactors that combine in situ analysis with
feedback algorithms like SNOBFIT to rapidly map a response surface and identify an target
optimum33, 34, 35, 36 or Bayesian optimizations that combine a mixture of theoretical, literature, and experimental screening data to learn general predictive schemes for a system.37, 38
However, in our case, in situ analysis of the target response (throughput) is implausible
because the resulting MoC1−x nanoparticles lack a spectral signature, which rules out implementation of automated self-optimization. This is not a unique case, as materials with
meaningful spectral signals are the exception rather than the rule. More complex machine
learning models, such as Bayesian optimizations, utilize a balance of both exploiting existing
data and exploring a design space, so yield optimizations require large libraries of existing
theoretical or literature data to pull from to effectively train a model.33, 34, 35, 36, 37, 38 This is
impractical for our system as this is novel chemistry with no library of preexisting data to
pull from, and an optimization must therefore be purely exploratory.
Considering the limitations of the system under study, statistical design of experiments
(DoE) in conjunction with response surface methodology (RSM) is the most efficient option for response surface creation and optimization. As a well-known, powerful optimization
method that has been proven effective for synthetic optimizations of nanoparticles, the combination of the two techniques offers more precise analysis of a unique system that is required
36

here, as the simplicity of the regression techniques enables elucidation of the parameter space
based solely on in-house experimental data and ex situ nanoparticle characterization in a
minimal number of experiments.39, 40, 41 This removes the difficulties encountered when considering the implementation of machine learning approaches while still efficiently providing
a comprehensive evaluation of the multivariate design space (including interaction effects)
and an accurate model of the response surface for identification of the optimal reaction
conditions.40, 41, 42, 43, 44, 45, 46 In addition, coupling such techniques with the advantages of
flow systems further advances the speed and efficiency of reaction parameter space elucidation for nanoparticle optimization. Herein, we utilize statistical DoE for a parametric
screening analysis to explore the experimental variables governing the MoC1−x nanoparticle synthesis in a continuous flow millifluidic reactor in tandem with RSM to optimize
MoC1−x nanofabrication with the goal of identifying the reaction conditions that maximize
the product throughput. We successfully use this approach to maximize the throughput of
catalytically active MoC1−x nanoparticles, discovering nonintuitive reaction conditions that
lead to maximum throughput, thereby demonstrating a robust route to scale-up.
37

3.3 Results and Discussion
3.3.1 Continuous Flow Millifluidic Synthesis and Reactor Setup
The colloidal MoC1−x nanoparticles were synthesized through a solution-phase thermolytic
decomposition of Mo(CO)6 in oleylamine (acting as a surface ligand47) and octadecene (high
boiling solvent), as previously described.21 High-throughput continuous flow synthesis was
performed using the millifluidic reactor configuration illustrated in Figure 3.1. A syringe
pump was used to deliver a precursor Mo(CO)6 solution at a constant flow rate (determined
as described in the Supporting Information) into a custom-made borosilicate millifluidic
glass reactor placed inside a convection furnace. As the precursor enters the heated zone,
it spontaneously decomposes to release CO gas and separate the liquid into isolated plugs
(Figure 3.1). Based on a transient heat transfer approximation (see 3.6.4 ), and assuming a
constant surface heat flux, the reagent solution reaches a reaction temperature of 240 C at ca.
2.5 s after entering the furnace and asymptotically approaches the final temperature of 340
C after ca. 13 s. The time required for the fluid to approach the steady-state temperature
is negligible relative to the minimum residence time achieved in this work (ca. 6 min) and
is significantly faster than ramping times in the batch “heating up” process (ca. 10 min).
Segmented flow prevents axial dispersion while recirculation within the liquid plugs facilitates
rapid mixing.48, 49 As gas evolution resulting from precursor decomposition can drive the flow
in either direction, a one-way valve is placed upstream of the furnace to prevent the backflow
of the reaction solution. In addition, the reactor is maintained at 20 psig to minimize in situ
gas evolution. The pressurized system elevates the reaction mixture boiling point above the
highest temperature used in the matrix of experimental conditions. As the product emerges
from the furnace, the reaction is thermally quenched, and the nanoparticle suspension passes
through an in-line flow sensor to determine the gas-to-liquid plug volume ratio (Figure 3.1).
The residence time of the reaction is calculated based on the ratio of gas to liquid. Exemplary
data and calculations are given in Figure 3.14. Although the plug size varies over time during
38

an experiment, this variation does not result in run-to-run inconsistencies as demonstrated
by experimental replicates. Beyond the flow sensor, the product is directed through a set of
manually operated valves to collection reservoirs.
Figure 3.1: Schematic of the millifluidic reactor system for the continuous flow synthesis
of MoC1−x nanoparticles. A syringe pump with a heated syringe (80 C) is used to drive
a precursor mixture into a reactor coil housed in a furnace. The colors are representative
of those empirically observed in the reaction, with yellow (at reactor inlet) indicating the
unreacted Mo(CO)6 precursor solution and black indicating the MoC1−x nanoparticle suspension isolated into plugs due to in situ gas evolution. An in-line thermal dissipation-type
flow sensor at the reactor outlet reports the relative volume of liquid and gas—an idealized
flow sensor signal is shown with corresponding liquid (detected by the sensor) and gas (not
detected) plugs. Product is collected in vials that can be isolated with valves to allow for
removal of partial product volumes during runs. The system pressure is maintained by a
fluidic controller (Fluigent) which is locked on a fixed pressure of 20 psi to minimize in situ
gas evolution.
The base reaction, denoted with the coded value “0” in Table 3.1, gives phase-pure, facecentered cubic α-MoC1−x nanoparticles (ICDD PDF # 03-065-8092). This reaction was run
at a flow rate of 25 mL h−1
, a temperature of 315 C, a Mo(CO)6 precursor concentration of
352 mM, and a 53% volume fraction of oleylamine in octadecene. These reaction conditions
resulted in an isolated yield of 69%, a residence time of 14 min, and product throughput
39

of 0.62 g h−1
. The powder X-ray diffraction (XRD) pattern of the resulting product shows
significant peak broadening consistent with small nanoparticles. Scherrer analysis of the
XRD pattern indicated a crystallite size of 2 nm. This size is qualitatively similar to the size
of the multipodal nanoparticles observed by transmission electron microscopy (TEM). These
data, shown in Figure 3.2 , are fully consistent with previous reports of this material.21
Figure 3.2: (a) Powder XRD pattern and (b) TEM image of the MoC1−x nanoparticles
synthesized under the base reaction conditions.
Table 3.1: Coded Low (-1), Center Point (0), and High (+1) Experimental Values for the
Input Variables Investigated in the Full Factorial Screening Design.
Input Variables Flow Rate Concentration Temperature Amount of Oleylamine
(mL h−1) (mM) (°C) (vol%)
Coded Low (-1) 10 78 290 5
Coded Center Point (0) 25 352 315 53
Coded High (+1) 40 625 340 100
3.4 Design of Experiments
The first step in employing DoE and predictive RSM is to determine the experimental input
variables that may affect the outcome of a specific system (e.g., throughput, isolated yield,
or residence time). These variables are typically chosen via prior knowledge of the system
and assessment of the current literature.21, 50 The input variables chosen for this system are
reaction temperature (C), volume fraction of oleylamine in octadecene (vol%), Mo(CO)6
40

precursor concentration (mM), and syringe pump flow rate (mL h−1
). Preliminary reactions
are then performed to determine the high and low bounds of each variable, the highest
and lowest values where the reaction produces MoC1−x nanoparticles as assessed by powder
XRD. This creates the bounds/edges of the parameter space specific to this flow reactor
and synthetic system that can be evaluated as part of the DoE. Any points outside of the
bounded design space presented here are not feasible for this system. The bounds were
established by analogous batch reactions, except for flow rate, which was bounded by the
mechanical limitations of the syringe pump and millifluidic reactor apparatus. Each variable
was chosen because of its typical importance in conventional nanoparticle syntheses. The
input variables and their corresponding high and low levels (coded +1 or -1, respectively) are
given in Table 3.1. Similar MoC1−x crystallite sizes, as assessed by Scherrer analysis of the
powder XRD data, were observed for all screening reactions. The responses, or the output
data that is subsequently analyzed, were then chosen based on the nature of this system and
the most important experimental aspects when considering scale up optimization, which in
this case were isolated yield, residence time (which is related in a complex way to syringe
pump flow rate due to gas evolution), and product throughput.
Once the high and low bounds are set, the combinations of k factors (variables) can
be investigated at N number of levels to create a factorial matrix in a statistical screening
design. To rapidly screen the effects of each factor and the combinations of the factors on
the responses, an investigation at two levels (N = 2) is standard, as it is the most robust and
efficient. Considering that the goal of an initial screening is to assess significance and not
to resolve the fine details of the system, a design involving solely the absolute high and low
extremes (i.e., the boundary conditions) of each variable enables an adequate investigation
of the entire expanse of the reaction space in the smallest possible number of experiments. A
full factorial design will then consist of Nk
combinations.50 For this system, the full factorial
consisted of 2
4
(16) experiments, since four variables, or factors, were investigated at two
levels each. The coded and real values of the full factorial matrix are given in Tables 3.2
41

and 3.3, respectively. In addition to the experiments given in the matrix, two replicates of
the center point of the design space, also known as the base reaction (coded as 0), and two
replicates chosen at random from the matrix were performed to help quantify run-to-run
variance and increase statistical significance. As a result of the complexity of the system, we
chose to perform the full factorial to ensure maximum data resolution (V ), which ensures
minimal to no confounding of the variable interactions.50
Table 3.2: Full factorial matrix via coded values.
Input Variables Responses
Block Flow Rate Temperature Amount of Oleylamine Concentration Residence Time Isolated Yield Throughput
(mL h−1) (C) (% vol) (mM) (min) (%) (g h−1)
1 1 -1 1 -1 11.7 0.69 0.219
1 -1 1 1 -1 18.3 0.55 0.044
1 -1 -1 -1 -1 19.1 0.60 0.048
1 1 1 -1 -1 9.3 0.36 0.115
1 0 0 0 0 14.7 0.69 0.618
1 0 0 0 0 13.0 0.69 0.618
1 -1 -1 1 1 24.0 0.95 0.605
1 1 1 1 1 18.6 0.71 1.809
1 1 -1 -1 1 6.0 0.33 0.841
1 -1 1 -1 1 21.0 0.23 0.147
1 -1 -1 1 -1 24.8 0.72 0.057
1 1 1 1 -1 9.7 0.70 0.223
1 1 -1 -1 -1 12.5 0.80 0.254
1 -1 1 -1 -1 34.0 0.22 0.017
1 1 -1 1 1 9.5 0.93 2.370
1 -1 1 1 1 30.0 0.20 0.127
1 -1 -1 -1 1 25.0 0.18 0.115
1 1 1 -1 1 6.9 0.05 0.127
1 -1 -1 1 -1 29.7 0.79 0.063
1 1 1 1 -1 11.4 0.70 0.223
The results of the full factorial screening are illustrated through Pareto charts and main
effects plots in Figure 3.3. The Pareto chart is used to determine which factors are statistically relevant; that is, the length of each bar is proportional to the value of a t-statistic
calculated for the corresponding effect. Any bars beyond the vertical error line represent
statistically significant factors at a determined significance level (α = 5%). For example,
the sole factor that affected the residence time was flow rate, with higher flow rates yielding shorter residence times (Figure 3.3a). For isolated yield, higher fractions of oleylamine
and lower temperatures had significant effects on the resulting isolated yield, as depicted
in Figure 3.3b. Lastly, high precursor concentration, high flow rate, high fractions of oleylamine, and the binary interactions between those three variables significantly affected the
42

Table 3.3: Full factorial matrix of the corresponding real values
Block Flow Rate Temperature Amount of Concentration
(mL h−1
) (°C) Oleylamine (% vol) (mM)
1 40 290 100 78
1 10 340 5 625
1 40 290 5 625
1 10 340 100 78
1 25 315 52.5 351.5
1 25 315 52.5 351.5
1 40 340 100 625
1 10 290 10 78
1 40 340 10 78
1 10 290 100 625
1 10 290 100 78
1 40 340 100 78
1 40 290 5 78
1 10 340 5 78
1 40 290 100 625
1 10 340 100 625
1 10 290 5 625
1 40 340 5 625
throughput (Figure 3.3c). The main effects plots help visualize the change in each indicated
response from the low level to the high level for each of the factors (Figure 3.3d-f), with a
steeper slope corresponding to a more significant effect on the given response.
43

Figure 3.3: Standardized Pareto charts for (a) residence time, (b) isolated yield, and (c)
throughput. The variables on the y-axis of the Pareto charts are defined as A = flow rate
(mL h-1), B = temperature (°C), C = amount of oleylamine (%vol), and D = precursor
concentration (mM). The vertical line in each Pareto chart corresponds to = 5%, and (+)
and (-) correspond to an increase or decrease in the response, respectively, for the high
value of a given factor. Main effects plots for (d) residence time, (e) isolated yield, and
(f) throughput, where (+) and (-) correspond to the high and low levels for each factor,
respectively.
44

Based on the significant impact of multiple variables on throughput from the screening
data, an optimization was performed with the goal of maximizing throughput as a single
response with important relevance to scale-up. Since only three variables (i.e., fraction of
oleylamine, precursor concentration, and flow rate) significantly affected throughput, the reactor temperature was fixed as a constant to the low level (−1) which corresponds to a real
value of 290 ◦C. The low level was chosen due to its apparent positive, though not statistically
significant, effect on the throughput (see Figure 3.3f), in addition to the decreased energy
consumption required for lower temperature reactions. To perform an optimization of the
experimental conditions and create a response surface of the parameter space, a second-order
design is needed51, 52. We employed the Doehlert, or uniform shell, design for this optimization because it allows for the critical factors to be investigated in differing amounts of detail
based on their significance53. That is, significant factors that have a clear hierarchy in regard
to their impact on the response can be assigned varying levels of investigation, weighing their
significance accordingly. This enables a more thorough investigation of the variables without
increasing the number of experiments and will increase the accuracy of the fitted model. The
spherical nature of the uniform shell model also allows for smooth movement throughout the
parameter space and more efficient model predictions in fewer experiments53. As illustrated
in Figure 3.3c, the variables that significantly affected throughput had a clear ranking of
significance with the amount of oleylamine having the smallest impact, flow rate having a
moderate impact, and precursor concentration having the greatest impact. We, therefore,
decided to investigate the corresponding variables at 3, 5, and 7 levels, respectively. Table
3.4 provides the real values of each investigated level for each variable. The entire Doehlert
matrix (coded and real values) is given in (Table 3.5 and 3.6). The Doehlert matrix for these
three factors corresponded to 13 experiments, as well as two random replicates in the design
space and two replicates of the center point, or base reaction.
45

Table 3.4: Real values of the levels investigated for each significant variable in the Doehlert
optimization of maximizing throughput. As depicted from the number of levels investigated
for each variable, the variables most significantly affecting throughput in decreasing order
are concentration, flow rate, and amount of oleylamine.
Levels
Input Variables 1 2 3 4 5 6 7
Concentration (mM) 115 194 273 352 431 509 588
Flow Rate (mL h−1
) 10 18 25 33 40 – –
Amount of Oleylamine (vol%) 14 53 92 – – – –
Table 3.5: Coded Doehlert optimization matrix appended (runs 1-18) with points on each
corner of the design space (runs 19-28).
Run Flow Rate (mL h−1) Amount of Oleylamine (%vol) Concentration (mM) Throughput (g h−1)
1 0 0 0 0.608927
2 1 0 0 0.888317
3 -1 0 0 0.254317
4 -0.5 0 -0.866 0.198324
5 0.5 0 -0.866 0.353127
6 -0.5 -0.817 -0.289 0.233224
7 0.5 -0.817 -0.289 0.342895
8 0.5 0.817 0.289 0.926832
9 -0.5 0.817 0.289 0.760111
10 0.5 0 0.866 0.837875
11 -0.5 0 0.866 0.608544
12 0 -0.817 0.577 0.467106
13 0 0.817 -0.577 0.345411
14 0.5 0 -0.866 0.364519
15 0.5 0 -0.866 0.364519
16 0 0.817 -0.577 0.340477
17 0 0.817 -0.577 0.330608
18 1 -1 -1 0.25
19 -1 -1 1 0.11
20 -1 -1 -1 0.05
21 1 1 1 2.4
22 -1 1 1 0.61
23 1 1 -1 0.22
24 1 -1 1 0.84
25 -1 1 -1 0.06
26 -1 1 -1 0.06
27 0 0 0 0.77
28 0 0 0 0.76
46

Table 3.6: Optimization matrix of the corresponding real values
Run Flow Rate (mL h−1
) Amount of Oleylamine (%vol) Concentration (mM)
1 25 52.5 351.5
2 40 52.5 351.5
3 10 52.5 351.5
4 17.5 52.5 114.65
5 32.5 52.5 114.65
6 17.5 13.7 272.46
7 32.5 13.7 272.46
8 32.5 91.3 430.54
9 17.5 91.3 430.54
10 32.5 52.5 588.35
11 17.5 52.5 588.35
12 25 13.7 509.31
13 25 91.3 193.69
14 32.5 52.5 114.65
15 32.5 52.5 114.65
16 25 91.3 193.69
17 25 91.3 193.69
18 40 5 78
19 10 5 625
20 10 5 78
21 40 100 625
22 10 100 625
23 40 100 78
24 40 5 625
25 10 100 78
26 10 100 78
27 25 52.5 351.5
28 25 52.5 351.5
47

3.5 Optimization via RSM: Maximizing Throughput
After successful collection of the experimental data representing the Doehlert matrix, RSM
can be performed to optimize the throughput. A surface response function was fitted to the
aforementioned data to serve as a predictive polynomial model of the curvature of the surface
in three dimensions (Eq. 3.1). This is done by depicting the response (i.e., throughput) in
the 3D parameter space as a function of the three input variables (i.e., concentration, flow
rate, and amount of oleylamine). The function is produced, initially, by performing a linear
regression to fit the input experimental data to the polynomial via regression coefficients.
The function is continually enhanced by performing an exchange algorithm54. This algorithm
tests all pairs of experimental runs, consisting of one that has been selected in the design
space and one that has not, making any exchanges that would increase the efficiency of the
model. Exchanges continue until no further improvements can be made by switching one
run that has been selected with one run that has not. This adjusts the function to minimize
the residual error term, which is a numerical value that represents the observed-predicted
value of each design point as each data point is introduced. Once the resulting function is
an optimal representation of the design space, it can then be extrapolated beyond the input
data values to predict the likely throughput response for every data point within the 3D
parameter space, providing a coherent model of our synthetic system.
Throughput (g h−1
) =0.578 + 0.319A + 0.235B + 0.370C
+ 0.0491A
2 + 0.105AB + 0.249AC
− 0.0577B
2 + 0.253BC − 0.0179C
2
(3.1)
where,
• A = flow rate (mL h−1
),
48

• B = amount of oleylamine (vol%),
• C = Mo(CO)6
concentration (mM).
Because all of the variables are normalized through the coded values, the relative change
of a variable is directly related to the size of its regression coefficient (Eq. 3.1). Therefore,
the equation using the coded values can be used to compare the absolute effect of the
variables on the response of the throughput. This means that if the model coefficients have
a large absolute value, the corresponding variable has a significant effect on the response.40
As shown by the statistical analysis of Eq. 3.1, throughput had a statistically significant
quadratic dependence on all three factors to the 95% confidence interval. Using the quadratic
terms in Eq. 3.1, the model predicts a maximum throughput of 2.1 g h−1 when the three
parameters (i.e., precursor concentration, flow rate, and amount of oleylamine) are set to
the coded values (1, 1, 1), respectively, which correspond to real values of 625 mM, 40 mL
h
−1
, and 100%. These predicted conditions for a maximum throughput are nonintuitive for
this synthetic system (specifically, 100% oleylamine and low temperature) and could not be
ascertained through chemical intuition or previous literature. Therefore, the implementation
of DoE is imperative for this optimization. A visual representation of the parameter space
based on the fitted model predictions of throughput plotted in three dimensions is given in
Figure 3.4, with the optimum condition indicated with a star.
Replicates of the center point in triplicate create a basis for the constant value in the
polynomial function (see Eq. 3.1). Additionally, the center point replicates along with the
replicates of two random points in the design allow for a variance estimate of the polynomial
function’s vector value, which is a simplified representation of the 3D function that contains
the coefficients of each term in the model and their directionality (positive or negative effect).
This essentially allows for the assessment and correction of run-to-run variability or sampling
error in the fitting of the model via experimental data and helps to calculate model accuracy.
This is done by extracting the square root of each component of the vector value, giving the
standard errors for the coefficients (regression error and residual error). The sum of squares
49

Figure 3.4: Calculated response surface function demonstrating the reaction conditions (precursor concentration, flow rate, and amount of oleylamine) that correspond to a specific
throughput, illustrated by the color legend. The conditions for maximum throughput are
indicated by a star. The bounds of this parameter space are specific to this flow reactor and
synthetic system; any points outside of the parameter space are not feasible for this system.
data allows the residual error value to be further split into pure error and lack of fit. Based
on the evaluation of the fitted model, the fit was able to explain 92% of the variability
in throughput with no serial autocorrelation of the residuals at a 5% significance level, as
indicated by R2
. The prediction variance increases slightly around the edges of the parameter
space (Figure 3.6) because of the spherical nature of the optimization design, as depicted in
Figure 3.5. This indicates an accurate model of the design space.
Several subsequent reactions were performed to validate the predicted model. First, the
optimized reaction conditions that were predicted to maximize throughput were performed in
triplicate. The XRD pattern and TEM image of the resulting optimized MoC1−x nanoparticles are provided in Figure 3.7a and 3.7b, respectively, and are consistent with the α-phase.
Selected area electron diffraction of these MoC1−x nanoparticles corroborates the assignment of the α-phase. The lattice fringes of MoC1−x nanoparticles were observed through
high-resolution TEM and suggest single crystalline particles (Figure 3.8). The measured
50

Figure 3.5: Visual representation of the design points for the optimization, corresponding to
the runs displayed in Table 3.5.
d-spacing (0.21 nm) corresponds to the (200) plane, in agreement with previous reports.21
The throughput optimization did not affect the MoC1−x nanoparticle phase or crystallite
size. The average isolated yield of this reaction, including the replicates, was 87 ± 3%. The
average experimental throughput achieved was 2.2 ± 0.04 g h−1
, compared to the predicted
throughput under these reaction conditions, which was 2.1 ± 0.3 g h−1
. The optimized
throughput represents a 3.5× increase compared to the base reaction, which afforded an
average throughput of 0.62 g h−1
.
In addition, triplicate reactions were carried out with conditions from a random, unoptimized point on the surface. The conditions of the unoptimized reaction consisted of a
flow rate of 30 mL h−1
, a Mo(CO)6 precursor concentration of 100 mM, and an 86% volume
fraction of oleylamine in octadecene. The average isolated yield resulting from these reaction
conditions, including replicates, is 86 ± 8%. The average experimental throughput achieved
under these conditions was 0.26 ± 0.02 g h−1
, which matches the predicted throughput of
0.25±0.3 g h−1
from the response surface under these specific conditions. The powder XRD
pattern and TEM image of the resulting nanoparticle product from the unoptimized reaction
are shown in Figure 3.7c and 3.7d, respectively. The predicted and experimental through51

Figure 3.6: Prediction variance plot of the design space. Concentration is set at the base
value (coded 0) for visual simplicity.
put values from both the optimized synthetic conditions and a random set of conditions,
both performed in triplicate, fall within standard error of the predicted throughput, thus
successfully validating the model.
To determine if there is any unreacted Mo(CO)6 precursor in the product stream, Fourier
transform infrared (FT-IR) spectra were obtained on the Mo(CO)6 precursor and compared
to those of the supernatant from the first wash in the MoC1−x nanoparticle work-up procedure
(Figure 3.9). From this analysis, we conclude that there is no evidence of unreacted Mo(CO)6
precursor in the product stream when the isolated yield is not quantitative. This suggests
that in those cases the precursor converts to clusters or ultrasmall nanoparticles that are
lost in the nanoparticle work-up. Nonetheless, the surface response can model throughput
in our experimental parameter space, including regions where the isolated yield is lower, as
evidenced by the excellent agreement between the model and experimental data.
Finally, the MoC1−x nanoparticles were supported on a carbon support and evaluated
as catalysts in the thermocatalytic reduction of CO2 using procedures and reaction conditions similar to those of our previous report (300 ◦C, 2 MPa, weight hourly space velocity
(WHSV) based on Mo content of 40 h−1 and H2:CO2 molar ratio in the feed of 2.7)21. Two
52

Figure 3.7: (a) Powder XRD pattern and (b) TEM image of the MoC1–x nanoparticles
produced under optimized conditions, and (c) powder XRD pattern and (d) TEM image of
the MoC1–x nanoparticles produced under unoptimized conditions.
catalysts were tested, one synthesized under the optimized throughput conditions described
here (termed “Optimized Flow” MoC1−x/C), and one synthesized via the previously reported
batch method (“Batch” MoC1−x/C). As presented in Figure 3.10a for the conversion as a
function of time-on-stream (TOS), both catalysts demonstrated an induction period where
conversion increased from ca. 17–21% over a period of 15 h TOS and then maintained stable activity without signs of deactivation over the next 5 h. This activity trend was nearly
identical to the performance observed previously for our C-supported nanoparticle MoC1−x
catalyst21. Major products from the CO2 hydrogenation reaction under these conditions
were CO, methane, and a mixture of C2+ hydrocarbons (Figure 3.10b). The C2+ hydrocarbons were composed mostly of C2–5 alkanes, which are characteristic of the products from
53

Figure 3.8: (a) HRTEM and (b) SAED pattern of MoC1−x nanoparticles synthesized under
optimized conditions. (c) HRTEM and (d) SAED pattern of MoC1−x nanoparticles synthesized under batch conditions. The lattice fringes correspond to the (111) and (200) planes
of the α-phase.
CO2 conversion through the Fischer–Tropsch reaction on molybdenum carbide catalysts.
Both catalysts exhibited a similar product distribution to the previously reported study21
,
with comparable selectivity to C2+ hydrocarbon products observed over the optimized flow
MoC1−x/C catalyst (11.7%) and the batch MoC1−x/C catalyst (10.7%).
54

Figure 3.9: FT-IR spectra of Mo(CO)6 precursor and the supernatant of the first wash of
the MoC1−x nanoparticles (in the workup procedure). The spectra show the representative
n(CO) stretching region, demonstrating that there is no unreacted Mo(CO)6 precursor in
the supernatant resulting from the product stream.
Figure 3.10: CO2 conversion as a function of time-on-stream and (b) product selectivity
taken as an average of data from 16-20 h. Reaction conditions were 300 ◦C, 2 MPa, WHSV
based on Mo content of 40 h
−1 and H2:CO2 molar ratio in the feed of 2.7.
55

3.6 Experimental Procedures
3.6.1 Continuous Flow Synthesis of MoC1−x Nanoparticles
Oleylamine (70% technical grade) and 1-octadecene (90%) were purchased from SigmaAldrich and dried by heating to 120 ◦C under vacuum for ca. 5 h prior to use. Mo(CO)6
(98%) was purchased from Sigma-Aldrich and used as received. A precursor solution was
prepared by mixing an appropriate portion of Mo(CO)6 with oleylamine and/or octadecene,
based on the desired vol% of oleylamine relative to the total solvent volume, and heating
the mixture to 140 ◦C for 1 h under N2. The solution was cooled to 100 ◦C and carefully
transferred to a glass syringe fitted with heat tape maintained at 80 ◦C to maintain a homogeneous precursor solution. The resulting MoC1−x nanoparticle product suspension (10
mL) was combined with hexanes in a 1:5 (vol/vol) ratio of hexanes/reaction mixture, before
being transferred equally into two 50 mL centrifuge tubes. The reaction mixtures were vortex mixed and bath sonicated before being precipitated each with 35 mL acetone through
centrifugation (6, 000 rpm, 20 min). The colorless supernatant was decanted and discarded,
and the black nanoparticle pellet was redispersed by adding 0.5 mL CHCl3. After vortex
mixing and bath sonicating the product suspension, the nanoparticles were re-precipitated
using 39 mL of ethanol through centrifugation (6, 000 rpm, 10 min). This washing step
with CHCl3 and ethanol was performed once more. The resulting nanoparticle pellet was
redispersed in CHCl3 and dried overnight under flowing N2(g) for further characterization.
Nanoparticle work-up was performed identically for all samples.
3.6.2 Millifluidic Flow Synthesis Platform
The millifluidic synthesis platform is shown in the main text (Figure 1). A 1.5” hole was
drilled through the center of the furnace (Paragon, Inc) to allow room for the inlet and the
outlet of the reactor while also acting as ventilation. A 37 mL coiled borosilicate glass reactor
(ID = 1.8 mm, OD = 3 mm) was custom-made with extended inlet and outlet capillaries
56

and placed inside the furnace. A fiber glass thermal insulator covered with aluminum foil
was placed ca. 9 cm above the furnace entrance to isolate the connectors from the heat
emerging from the furnace. Through the thermal insulator, the glass reactor inlet and outlet
are connected to polytetrafluoroethylene (PTFE) tubing (1.59 mm ID, 3.18 mm OD) using
1/4-28 PEEK Nuts, poly(ethene-co-tetrafluorethene) (ETFE) Flangeless Ferrule, 1/8” OD,
and PEEK union assembly (IDEX Health & Science). Shut-off valves, PEEK, 1/8” OD
(IDEX Health & Science) are used for flow stream configuration. Back pressure regulators
(PEEK, 20 psi, 1/8” OD) are connected to the collection jars. Silicon rubber heating tapes
(Brisk Heat®) heated to ca. 80 C are attached to tubing with Kapton tape and covered
with aluminum foil throughout the system. A temperature differential flow sensor (Flow
EZ™
, Fluigent) was added downstream of the reactor outlet for in-line identification of gas
and liquid plugs. A fluidic control system (MFCS™–EZ, Fluigent) was used for regulating
the pressure in the system. All-in-one (Fluigent) software was utilized for regulating the
pressure and monitoring the flow sensor data at a sampling frequency of 10 Hz.
57

Figure 3.11: Photographic images of the experimental setup. The top image shows the entire
millifluidic flow synthesis platform, and the bottom image shows the reactor placed inside
the furnace.
58

3.6.3 Furnace Temperature Stability
Temperature stability of the furnace (Paragon, Inc) was verified by placing thermocouples
at the four corners of the glass reactor. The temperature varied by less than 2 °C. Table 3.1
summarizes the ramp power used to operate the furnace. Time-dependent temperature for
the ramp-up stage and the steady-state stage are shown in Figure 3.4.
Ramp Power (℃ h
-1) Temperature Range (℃)
280 Room temperature to 280
100 280 to 300
50 300 to 315
20 315 to 320
Table 3.7: Table caption here
59

Figure 3.12: Representative data showing time-dependent furnace temperature at approaching a setpoint of 320 ◦C: (a) Ramp-up stage and (b) time-dependent temperature after
reaching steady state and prior to running the experiment.
60

3.6.4 Precursor Thermal Time Constant Estimation
Heat transfer analysis was carried out to approximate the time required for the precursor
fluid to reach the temperature of the furnace. We apply the lumped capacitance model. The
lumped capacitance is a valid approximation for low Biot number, Bi (eq 3.1),
Bi =
hLc
k
(3.2)
where h is the heat transfer coefficient, Lc is the characteristic length (Lc =
V
A ≈
R
2
for
cylindrical tubes), and k is the thermal conductivity. The Nusselt number for constant
surface heat flux for circular tubes is estimated as Nu = 4.36. Therefore, the heat transfer
coefficient, h (eq 3.2),
h =
k
DH
Nu (3.3)
where DH = 1.8 mm is the hydraulic diameter. This gives a heat transfer coefficient of
h = 339 W/(m2K).
Starting from the overall energy balance, the time required for the fluid to reach the final
temperature is determined from eqs 3.3-3.5:
ρcpV
dT
dt = −hAs(T(t) − T∞) (3.4)
d(T − T∞)
(T − T∞)
= −
hAs
ρV cp
dt (3.5)
(T(t) − T∞)
(Ti − T∞)
= e
−bt (3.6)
Considering octadecene as the fluid with the properties, ρ = 789 kg/m3
, cp = 2500 J/(kg K),
and b =
hAs
ρV cp
= 0.385 s
−1
, the time period for the fluid to reach 99% of the of the final tem61

perature can be estimated as:
(T(t) − T∞)
(Ti − T∞)
= e
−bt → 0.01 = e
−(0.385 s−1
)t → t ≈ 12.9 s (3.6) (3.7)
Figure 3.13: Modeled time-dependent temperature of the precursor fluid (from 80 to 340 ◦C)
approximated by the lumped capacitance model.
The Mo(CO)6 precursor begins to decompose at ca. 240 ◦C, which is reached approximately 2.5 s after the precursor enters the reactor. The setpoint temperature is reached in
the fluid after approximately 12.9 s. This analysis neglected the temperature gradient along
the furnace entrance as well as radiative heat transfer and assumes a lumped capacitance
model to be valid.
3.6.5 Residence Time Calculation
Upon exiting the furnace, the nanoparticle product solution was directed through an in-line
temperature differential flow sensor set to identify liquid plugs for an evaluation of the gasto-liquid volumetric ratio. Exemplary data are shown in Figure 3.6. The peaks indicate
a liquid phase flowing through the sensor, while the baseline represents a gas phase. To
neglect noise peaks that may arise from the sensor readout, a threshold is set to distinguish
62

the two phases. The threshold is set to 10× the standard deviation of the sensor baseline
readout when no liquid is flowing in a pressurized system. By counting the number of data
points above and below the threshold, the gas to liquid volumetric ratio can be determined.
The liquid volumetric flow rate (QL) corresponds to the set syringe pump flow rate. The
gas volumetric flow rate (QG) is determined by the ratio obtained from the flow sensor data
multiplied by QL. The residence time was, therefore, determined by dividing the measured
volume of the reactor (37 mL) by the total volumetric flow rate, τ = V/(QL + QG).
Figure 3.14: Exemplary data for the sensor readout during product collection with a pump
flow rate set at QL = 40 mL h−1
. The data shown here correspond to a 2-min time frame for
experiment #1 in the full factorial matrix (Table 3.3). Data are sampled at 10 Hz. Based
on the count of data points above and below the threshold throughout the experiment (14
min), the gas to liquid volumetric ratio was found to be 3.74, which translates to QG = 150
mL h−1
. Therefore, the residence time is calculated as τ = V/(QL + QG) = 11.7 min.
3.6.6 Reactor Cleaning
The cleaning procedure is kept consistent throughout all experiments by performing a complete system flush after re-filling the syringe with a new batch of precursor (i.e., a new %
oleylamine concentration). Piranha solution composed of H2SO4 (80%) and H2O2 (20%)
(v/v) was used to flush the reactor. The solution is injected by pressure-driven flow where
the inlet of the reactor is connected to a separate N2-pressurized jar containing piranha so63

lution. The reactor is then rinsed with water and flushed with ethanol. The reactor is dried
under heat prior to running the following experiment.
3.6.7 Catalytic Testing
MoC1−x nanoparticles were supported on Vulcan XC 72 R carbon from a batch preparation
and the optimized throughput flow preparation, with Mo contents of 3.54 and 5.10 wt%,
respectively. Activity and selectivity of the MoC1−x nanoparticle catalysts were evaluated
for the CO2 hydrogenation reaction following similar conditions described in our previous
report. Specifically, 0.70–1.0 g of supported catalyst was loaded in a 1
4
” ID stainless steel
reactor and pretreated under 95% H2/5% Ar flow (50 sccm) at 450 ◦C for 2 h before cooling
to 300 ◦C for reaction. Flowrates for CO2 and 95% H2/5% Ar were adjusted to achieve
the same weight-hourly space velocity (WHSV) of ca. 40 gCO gMo−1 h
−1 based on metal
composition (i.e., weight loading of Mo) for each catalyst and the same feed gas composition
of 26 : 70 : 4 mol% for CO2:H2:Ar, respectively (corresponding to a molar H2:CO2 ratio in the
feed of 2.7). Product analysis was performed online by an Agilent Technologies 7890B gas
chromatograph equipped with flame ionization detectors (FIDs) and a thermal conductivity
detector (TCD). Conversion was calculated as Σ(molar flow rate of C in all products)/(molar
flow rate of inlet CO2). The C-selectivity of product i was calculated as (molar flow rate of
C in product i)/Σ(molar flow rate of C in all products).
3.6.8 Characterization
Powder X-ray diffraction (XRD)
XRD patterns were collected on a Rigaku Ultima IV diffractometer operating at 40 mA and
44 kV with a Cu Kα X-ray source (λ = 1.5406 Å). Samples used for XRD were prepared by
drop-casting CHCl3 suspensions of nanoparticles onto a glass sample holder.
64

Transmission electron microscopy (TEM)
TEM and high-resolution TEM (HRTEM) images were acquired with a JEOL JEM2100F
(JEOL Ltd.) microscope operating at 200 kV. Selected area electron diffraction (SAED)
images were acquired with an FEI Talos F200C G2 (Thermo Fisher Scientific) microscope
operating at 200 kV. Each sample was dropcast on 400 mesh Cu grids coated with a lacey
carbon film (Ted Pella, Inc.) and dried overnight under vacuum at room temperature.
Thermogravimetric analysis (TGA)
TGA of the MoC1−x nanoparticles was performed on a TGA Q50 instrument. The organiccorrected isolated yield of the nanoparticles from each reaction was gravimetrically calculated
via TGA. To determine the organic ligand content, ca. 5 mg of the resulting nanoparticles
was isolated after workup and drying and heated to 650 ◦C under flowing N2 at a heating
rate of 10 ◦C min−1
.
Fourier transform infrared spectroscopy (FT-IR)
FT-IR spectra were collected using a Bruker Vertex 80 spectrophotometer using 16 scans, 4
cm−1
resolution, 4000-400 cm−1
range, and absorbance units as the operational parameters.
3.7 Conclusion
We successfully utilized a millifluidic continuous flow system coupled with DoE (Design of Experiments) and predictive RSM (Response Surface Methodology) to optimize the throughput
of α-MoC1−x nanoparticle production. This is the first report, to the best of our knowledge,
of a colloidal nanoparticle synthetic throughput optimization using DoE in a continuous flow
system. Through the investigation of four experimental variables—temperature, volume fraction of oleylamine, precursor concentration, and syringe pump flow rate—we performed a
first-order full factorial design, which allowed for robust information to be accessed regarding
the statistical significance of each variable in affecting three different responses (i.e., isolated
yield, residence time, and throughput). This comprehensive statistical screening provided
insight into singular variable effects on each response, as well as binary interaction effects
65

between the variables, which are nearly impossible to identify using traditional OVAT (One
Variable At a Time) methods. For example, throughput (the response with the most immediate relevance for scale-up) was affected by three of the four experimental variables (the high
levels of the volume fraction of oleylamine, precursor concentration, and flow rate), in addition to the binary interactions between each of these variables. All three of these variables
and their interactions were significant to the 95% confidence interval. The deconvolution of
these variable interactions, including the secondary positive effects on throughput from the
interactions of the high levels of each variable and the insignificance of temperature, would
not be possible without the use of this kind of data-driven screening method.
Upon completion of the screening design, a Doehlert uniform-shell optimization design
to maximize throughput with the three statistically significant variables (vide supra) was
employed, keeping reaction temperature constant at the low level. Each of these factors
was investigated at different levels, based on their significance, to fit the experimental data
to a quadratic equation that modeled throughput as a function of the investigated factors
via linear regression (RSM). The resulting model equation was graphically represented as a
response surface function with the ability to identify the optimized reaction conditions to
maximize the throughput, as well as the synthetic conditions for any given throughput. In
this case, a DoE optimization is superior to other simplex algorithms and machine learning
methods because of the novelty of this material. Because there is only one previous report
of colloidal α-MoC1−x nanoparticles in the literature,21 optimization designs that require
large data sets would not be feasible55, 56. Using DoE in conjunction with RSM for this
system results in incredibly accurate predictive power of the desired responses with a minimal
number of reactions, given the limited prior knowledge of the design space as a whole.
We achieved an average experimental optimized throughput of 2.2 g h−1
. This throughput represents a threefold increase over the throughput achieved with the base reaction
conditions. Coupling DoE with continuous flow methods also drastically increases throughput over the lab-scale batch reaction, which had a throughput of only 0.047 g h−1
, for a
66

total improvement of almost 50×. This impressive increase in throughput demonstrates
that the optimization through DoE and predictive RSM unlocks the robust exploration of
a given synthetic parameter space in a way that is not possible using OVAT methods. The
achievable throughput can be further increased through parallelization of the millifluidic reactor.Wang2020, 29 Considering a 16-channel parallel reactor, the achieved throughput with
linear scaling could reach 52 g day−1
, or 0.4 kg week−1ref12. This provides a clear pathway
to scaling the production to industrially relevant quantities57, 58. Catalytic testing in the
CO2 hydrogenation reaction confirmed comparable activity and selectivity of the MoC1−x
nanoparticles produced using the conditions for optimized throughput compared to our initial report of low throughput batch synthesis.21. The implications of this study further
promote the potential of scaling up colloidal nanoparticle catalysts at the industrial level.
67

References
[1] A. Marinas and R. A. Sheldon. “Utilisation of Biomass for Fuels and Chemicals: The
Road to Sustainability”. In: Catal. Today 167 (2011), p. 1.
[2] Y. Aoki, H. Tominaga, and M. Nagai. “Hydrogenation of CO on Molybdenum and
Cobalt Molybdenum Carbide CatalystsMass and Infrared Spectroscopy Studies”. In:
Catal. Today 215 (2013), p. 169.
[3] G. S. Ranhotra et al. “Catalysis over Molybdenum Carbides and Nitrides: I Catalyst
Characterization”. In: J. Catal. 108 (1987), p. 24.
[4] G. S. Ranhotra, A. T. Bell, and J. A. Reimer. “Catalysis over Molybdenum Carbides
and Nitrides: II. Studies of CO Hydrogenation and C2H6 Hydrogenolysis”. In: J.
Catal. 108 (1987), p. 40.
[5] C. Liu et al. “Preparation of Nanostructured Molybdenum Carbides for CO Hydrogenation”. In: RSC Adv. 4 (2014), p. 20948.
[6] H. Shou and R. J. Davis. “Multi-Product Steady-State Isotopic Transient Kinetic
Analysis of CO Hydrogenation over Supported Molybdenum Carbide”. In: J. Catal.
306 (2013), p. 91.
[7] F. H. Ribeiro et al. “Catalytic Reactions of N-Alkanes on -W2C and WC: The Effect
of Surface Oxygen on Reaction Pathways”. In: J. Catal. 130 (1991), p. 498.
[8] C. Phamhuu, M. J. Ledoux, and J. Guille. “Reactions of 2- and 3-Methylpentane,
Methylcyclopentane, Cyclopentane, and Cyclohexane on Activated Mo2C”. In: J.
Catal. 143 (1993), p. 249.
[9] E. Iglesia et al. “Bifunctional Reactions of Alkanes on Tungsten Carbides Modified
by Chemisorbed Oxygen”. In: J. Catal. 131 (1991), p. 523.
68

[10] F. H. Ribeiro et al. “Reactions of Neopentane, Methylcyclohexane, and 3,3-Dimethylpentane
on Tungsten Carbides: The Effect of Surface Oxygen on Reaction Pathways”. In: J.
Catal. 130 (1991), p. 86.
[11] M. M. Sullivan, C.-J. Chen, and A. Bhan. “Catalytic Deoxygenation on Transition
Metal Carbide Catalysts”. In: Catal. Sci. Technol. 6 (2016), p. 602.
[12] Z. Lin et al. “Hydrodeoxygenation of Biomass-Derived Oxygenates over Metal Carbides: From Model Surfaces to Powder Catalysts”. In: Green Chem. 20 (2018), p. 2679.
[13] R. B. Levy and M. Boudart. “Platinum-Like Behavior of Tungsten Carbide in Surface
Catalysis”. In: Science 181 (1973), p. 547.
[14] L. Volpe and M. Boudart. “Compounds of Molybdenum and Tungsten with High
Specific Surface Area: I Nitrides”. In: J. Solid State Chem. 59 (1985), p. 332.
[15] J. S. Lee, S. T. Oyama, and M. Boudart. “Molybdenum Carbide Catalysts: I Synthesis
of Unsupported Powders”. In: J. Catal. 106 (1987), p. 125.
[16] E. Iglesia et al. “Synthesis, Characterization, and Catalytic Properties of Clean and
Oxygen-Modified Tungsten Carbides”. In: Catal. Today 15 (1992), p. 307.
[17] D.-V. N. Vo et al. “Non-Linear ASF Product Distribution over Alkaline-Earth Promoted Molybdenum Carbide Catalysts for Hydrocarbon Synthesis”. In: Catal. Today
214 (2013), p. 42.
[18] D.-V. N. Vo et al. “Fischer–Tropsch Synthesis: Effect of Promoter Type on AluminaSupported Mo Carbide Catalysts”. In: Catal. Today 175 (2011), p. 450.
[19] P. A. Alaba et al. “Molybdenum Carbide Nanoparticle: Understanding the Surface
Properties and Reaction Mechanism for Energy Production towards a Sustainable
Future”. In: Renew. Sustain. Energy Rev. 91 (2018), p. 287.
69

[20] T. Hyeon, M. Fang, and K. S. Suslick. “Nanostructured Molybdenum Carbide: Sonochemical Synthesis and Catalytic Properties”. In: J. Am. Chem. Soc. 118 (1996),
p. 5492.
[21] F. G. Baddour et al. “An Exceptionally Mild and Scalable Solution-Phase Synthesis
of Molybdenum Carbide Nanoparticles for Thermocatalytic CO2 Hydrogenation”. In:
J. Am. Chem. Soc. 142 (2020), p. 1010.
[22] L. Zhang and Y. Xia. “Scaling up the Production of Colloidal Nanocrystals: Should
We Increase or Decrease the Reaction Volume?” In: Adv. Mater. 26 (2014), p. 2600.
[23] E. J. Roberts et al. “Continuous Flow Methods of Fabricating Catalytically Active
Metal Nanoparticles”. In: ACS Appl. Mater. Interfaces 11 (2019), p. 27479.
[24] K. F. Jensen. “Microreaction Engineering Is Small Better?” In: Chem. Eng. Sci. 56
(2001), p. 293.
[25] Focus. “F”. In: Lab Chip 4 (2004), 11N.
[26] J. Sui et al. “Continuous Synthesis of Nanocrystals via Flow Chemistry Technology”.
In: Small 16 (2020), p. 1902828.
[27] A. J. deMello. “Control and Detection of Chemical Reactions in Microfluidic Systems”.
In: Nature 442 (2006), p. 394.
[28] E. J. Roberts et al. “High-Throughput Continuous Flow Synthesis of Nickel Nanoparticles for the Catalytic Hydrodeoxygenation of Guaiacol”. In: ACS Sustainable Chem.
Eng. 5 (2017), p. 632.
[29] C. T. Riche et al. “Flow Invariant Droplet Formation for Stable Parallel Microreactors”. In: Nat. Commun. 7 (2016), p. 10780.
[30] R. M. Maceiczyk and A. J. deMello. “Fast and Reliable Metamodeling of Complex
Reaction Spaces Using Universal Kriging”. In: J. Phys. Chem. C 118 (2014), p. 20026.
70

[31] S. Li et al. “Automated Microfluidic Screening of Ligand Interactions during the
Synthesis of Cesium Lead Bromide Nanocrystals”. In: Mol. Syst. Des. Eng. 5 (2020),
p. 1118.
[32] P. W. Miller et al. “A Microfluidic Approach to the Rapid Screening of PalladiumCatalysed Aminocarbonylation Reactions”. In: Adv. Synth. Catal. 351 (2009), p. 3260.
[33] N. Holmes et al. “Self-Optimisation of the Final Stage in the Synthesis of EGFR
Kinase Inhibitor AZD9291 Using an Automated Flow Reactor”. In: React. Chem.
Eng. 1 (2016), p. 366.
[34] K. Abdel-Latif et al. “Self-Driven Multistep Quantum Dot Synthesis Enabled by Autonomous Robotic Experimentation in Flow”. In: Adv. Intell. Syst. 3 (2021), p. 2000245.
[35] C. Mateos, M. J. Nieves-Remacha, and J. A. Rincón. “Automated Platforms for Reaction Self-Optimization in Flow”. In: React. Chem. Eng. 4 (2019), p. 1536.
[36] B. L. Hall et al. “Autonomous Optimisation of a Nanoparticle Catalysed Reduction
Reaction in Continuous Flow”. In: Chem. Commun. 57 (2021), p. 4926.
[37] B. J. Shields et al. “Bayesian Reaction Optimization as a Tool for Chemical Synthesis”.
In: Nature 590 (2021), p. 89.
[38] A. M. Schweidtmann et al. “Machine Learning Meets Continuous Flow Chemistry:
Automated Optimization towards the Pareto Front of Multiple Objectives”. In: Chem.
Eng. J. 352 (2018), p. 277.
[39] N. M. Fhionnlaoich et al. “DoE-It-Yourself: A Case Study for Implementing Design
of Experiments into Nanoparticle Synthesis”. In: (2020). doi: 10.26434/chemrxiv.
8198420.v1. url: https://doi.org/10.26434/chemrxiv.8198420.v1.
[40] E. M. Williamson et al. “Statistical Multiobjective Optimization of Thiospinel CoNi2S4
Nanocrystal Synthesis via Design of Experiments”. In: ACS Nano 15 (2021), p. 9422.
71

[41] L. Mora-Tamez et al. “Controlled Design of Phase- and Size-Tunable Monodisperse
Ni2P Nanoparticles in a Phosphonium-Based Ionic Liquid through Response Surface
Methodology”. In: Chem. Mater. 31 (2019), p. 1552.
[42] S. A. Weissman and N. G. Anderson. “Design of Experiments (DoE) and Process
Optimization. A Review of Recent Publications”. In: Org. Process Res. Dev. 19 (2015),
p. 1605.
[43] E. J. Braham et al. “Navigating the Design Space of Inorganic Materials Synthesis Using Statistical Methods and Machine Learning”. In: Dalton Trans. 49 (2020),
p. 11480.
[44] N. D. Burrows et al. “Understanding the Seed-Mediated Growth of Gold Nanorods
through a Fractional Factorial Design of Experiments”. In: Langmuir 33 (2017),
p. 1891.
[45] Z. Liu et al. “Application of the Factorial Design of Experiments to Hydrothermal
Synthesis of Lithium Iron Phosphate”. In: Int. J. Appl. Ceram. Technol. 17 (2020),
p. 1231.
[46] E. J. Braham et al. “Machine Learning-Directed Navigation of Synthetic Design
Space: A Statistical Learning Approach to Controlling the Synthesis of Perovskite
Halide Nanoplatelets in the Quantum-Confined Regime”. In: Chem. Mater. 31 (2019),
p. 3281.
[47] S. Mourdikoudis and L. M. Liz-Marzán. “Oleylamine in Nanoparticle Synthesis”. In:
Chem. Mater. 25 (2013), p. 1465.
[48] H. Song, J. D. Tice, and R. F. Ismagilov. “A Microfluidic System for Controlling
Reaction Networks in Time”. In: Angew. Chem., Int. Ed. 42 (2003), p. 768.
[49] B. K. H. Yen et al. “A Microfabricated Gas–Liquid Segmented Flow Reactor for HighTemperature Synthesis: The Case of CdSe Quantum Dots”. In: Angew. Chem., Int.
Ed. 44 (2005), p. 5447.
72

[50] G. E. P. Box, J. S. Hunter, and W. G. Hunter. Statistics for Experimenters: Design,
Innovation, and Discovery. 2nd ed. Wiley series in probability and statistics. Hoboken,
N.J: Wiley-Interscience, 2005.
[51] L. Vera Candioti et al. “Experimental Design and Multiple Response Optimization.
Using the Desirability Function in Analytical Methods Development”. In: Talanta 124
(2014), p. 123.
[52] M. A. Bezerra et al. “Response Surface Methodology (RSM) as a Tool for Optimization in Analytical Chemistry”. In: Talanta 76 (2008), p. 965.
[53] D. H. Doehlert. “Uniform Shell Designs”. In: Appl. Stat. 19 (1970), p. 231.
[54] R. K. Meyer and C. J. Nachtsheim. “The Coordinate-Exchange Algorithm for Constructing Exact Optimal Experimental Designs”. In: Technometrics 37 (1995), p. 60.
[55] V. Fath et al. “Self-Optimising Processes and Real-Time-Optimisation of Organic
Syntheses in a Microreactor System Using Nelder–Mead and Design of Experiments”.
In: React. Chem. Eng. 5 (2020), p. 1281.
[56] C. P. Breen et al. “Ready, Set, Flow! Automated Continuous Synthesis and Optimization”. In: Trends Chem. 3 (2021), p. 373.
[57] Z. Dong et al. “Scale-up of Micro- and Milli-Reactors: An Overview of Strategies,
Design Principles and Applications”. In: Chem. Eng. Sci. X 10 (2021), p. 100097.
[58] J. A. Schaidle et al. “Transitioning Rationally Designed Catalytic Materials to Real
“Working” Catalysts Produced at Commercial Scale: Nanoparticle Materials. In”. In:
Catalysis (2017), p. 213.
73

Chapter 4: Machine Learning-Assisted Visible Light Spectrophotometry in Continuous Flow Reactors for Kinetic Analysis of Ionic LiquidBased Platinum Nanoparticle Synthesis*
Madani, M. S.†
, Pan, B.†
, Brutchey, R. L., Malmstadt, M
*This chapter is adapted from a manuscript in preparation
†These authors contributed equally to this work.
4.1 Introduction
Platinum nanoparticles (Pt NPs) have garnered attention in various scientific fields due to
their versatile applications. Over the past decades, there have been significant demonstration of their potential in optoelectronics,1
catalysis,2
fuel cells,3 and hydrogenation reactions.4
Among many synthesis strategies, the polyol reduction method stands out as a key technique
for fabricating metal nanostructures. This method is unique in that it employs polyol not
only as a solvent but also as a reducing agent.5 Moreover, the introduction of ionic species
into the polyol solution offers a strategic approach to manipulating the shape and overall
morphology of these Pt nanoparticles. Ionic Liquid-based (IL-based) synthesis emerges as
a promising avenue to further refine and enhance those syntheses. This is attributed to the
several advantages that ionic liquids present as solvents. Ionic liquids excel as solvents in
nanoparticle synthesis due to their ability to colloidally stabilize nanoparticles and induce
high nucleation rates, given their low interfacial tensions.6 Unlike the conventional volatile
and flammable organic solvents, ionic liquids are thermally stable, non-volatile, and exhibit
low flammability and toxicity, marking them as environmentally friendly alternatives.7
. Col74

loidal synthesis based on IL has been reported previously8, 9 and the ability to recycle ILs
post colloidal synthesis has also been demonstrated in previous work.10 While IL-based synthesis is proving to be a viable approach for colloidal NPs synthesis, the synthetic process
under such medium is not understood from a mechanistic perspective. Gaining a mechanistic
insight requires analytical characterization techniques which explores their nucleation and
growth kinetics advancing their potentials further and allowing for tunning of NPs properties.
Small-angle X-ray scattering (SAXS) is a powerful method used to obtain such information
as it provides qualitative and quantitative real-time information.11 However, when investigating ionic liquids (ILs) with SAXS, challenges arise. This difficulty is primarily due to the
high attenuation coefficient of ILs, which often contain elements like fluorine and sulfur with
high photoabsorption cross-section. These elements significantly reduce X-ray transmission
at low energies, necessitating the use of synchrotron radiation for effective analysis.12
Ultraviolet-Visible (UV-Vis) spectroscopy is a common tool for studying the kinetics
of nanoparticle synthesis. By monitoring the absorption spectrum of the nanoparticle solution over time, it provides real-time insights into the nucleation and growth phases of
nanoparticle synthesis. The Beer-Lambert Law, which links absorbance to concentration,
has been the traditional method for obtaining quantitative understanding in UV-vis spectroscopic analysis.13, 14 This straightforward relation has enabled researchers to determine
the concentration of various substances by examining their light absorption behavior within
the ultraviolet to visible spectrum. Numerous studies have been performed on Au, Ag, and
Cu colloidal suspensions among other metallic nanoparticles.15, 16, 17, 18, 19, 20 However, unlike
Au and Ag, other catalytically relevent noble metals such as Pt and Pd do not possess a
spectral signature in the near UV or visible.18, 19, 20. For instance, the surface plasmon peak
for Pt particles is in the 216 nm to 261 nm range, in the far UV.20, 21, 22 The absence of
distinct absorption in the visible range for certain noble metals such as Pt and Pd often
necessitates the utilization of the UV range to analyze Surface Plasmon Resonance (SPR)
peaks for concentration measurements, which may not be accessible for all labs or purposes.
75

Analyzing the extinction spectrum, which captures both absorption and scattering phenomena, can be extremely valuable in cases where the primary absorption peak falls outside
the accessible measurement range. This is because the extinction spectrum contains subsidiary peaks, or subtle spectral shifts that are indicative of the sample’s properties under
investigation, including its concentration. This is often reflected by the complex interactions
between the material and light, such as multiple scattering events, or changes in the refractive index.23 While these features may not be as pronounced as the primary absorption peak,
they can still exhibit quantifiable features obtained through appropriate analytical methods.
Machine Learning (ML) algorithms have been used as a tool to analyze complex relationships and can be trained to identify subtle spectral features which traditional methods
might overlook. These algorithms can be trained on datasets to recognize and interpret even
minute spectral shifts or subsidiary peaks within the extinction spectrum. By analyzing the
extinction spectrum in the available wavelength range through ML-assisted methods, one
can acquire useful data for concentration estimations. One of the most compelling arguments in favor of this methodology lies in its cost-effectiveness. Compared to traditional
spectroscopic techniques such as Raman spectroscopy, the financial overhead for setting up
a visible extinction spectroscopy experiment is considerably lower. In this work, we employ
in-line analysis of visible spectroscopy coupled with machine learning algorithms to demonstrate the growth kinetics of ionic-liquid-based Pt NP synthesis by polyol reduction. This
work not only showcases the ML-assisted methodology for visible light spectrophotometry,
but also paves the way for a first investigation on the growth kinetics of an ionic-liquid-based
Pt NP synthesis.
4.2 Preliminary Screening
In the initial phase of this project, we conducted a screening Design of Experiments (DoE)
using a full factorial matrix. This involved five factors at two levels, aiming to identify the
components with the most significant impact on the spectral data. The experimental setup,
76

as depicted in Figure 4.1, consisted of five pumps delivering different volume ratios. This
setup allowed for the control of the concentration of Pt salt to EG, the amount of PVP
in ionic liquid (IL), and the concentration of Pt nanoparticles in the final mixture. The
mixtures from Pumps 1 (IL) and 2 (IL/PVP), and Pumps 3 (EG) and 4 (K2P tCl4/EG),
were combined separately before being passed through passive microfluidic mixers at each
junction. This ensured thorough mixing before merging with the solution from Pump 5
(Pt NPs/IL/EG), as shown in Figure 4.1. These mixers played a crucial role in enhancing
the homogeneity of the solutions. Further details on the mixing process and the efficacy of
these mixers are available in the Supplementary Information. The flow rates were controlled
within a range of 0 to 100 µL/min, constrained by the system’s flow resistance. The design
matrix led to 32 unique experimental conditions, covering various combinations of the five
factors at two levels. However, the condition with all factors at zero was excluded, resulting
in 31 distinct experimental runs. (Table 4.1)
Figure 4.1: Illustration of the experimental setup for the preliminary screening phase of
the Design of Experiments (DoE). The configuration includes five syringe pumps, each responsible for accurately delivering reagents into the system at predetermined flow rates in
accordance with the full factorial matrix design. Pumps 1 and 2 manage the introduction
of the ionic liquid (IL) and its mixture with polyvinylpyrrolidone (PVP), respectively, while
Pumps 3 and 4 handle the ethylene glycol (EG) and the Pt salt dissolved in EG. The fifth
pump dispenses the Pt nanoparticles suspension. This setup enables precise manipulation of
reaction parameters to discern the influence of various concentrations on the spectral profile
of the resultant Pt nanoparticles
77

Experiment Components
Pt NPs Pt(II) PVP EG IL
1 100 0 100 100 0
2 0 100 0 100 100
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
31 100 100 100 100 .
.
.
Table 4.1: Preliminary Screening Design of Experiment Matrix.
Each experimental run produced spectral data corresponding to the specific composition
of the mixture. Spectral data obtained from the experiments were subjected to Principal
Component Analysis (PCA) to reduce their dimensionality. The principal components (PCs)
that result from this PCA are then used as the response variables (dependent variables) in
the DoE. On the other side, the various concentrations of the components in the mixture
act as the factors (independent variables) in the DoE. This setup allows for an in-depth understanding of how changes in the concentrations of different components affect the spectral
data, as captured by the principal components. Based on the DoE results, the mixture’s
components were evaluated for their significance. The Pt nanoparticles (PtNPs) stand out
as the most significant contributor, as evidenced by the highest LogWorth value and an extremely low P-value (0.00001), signifying strong statistical significance. This suggests that
PtNPs concentration has the greatest impact on the response variable as shown in Figure
4.2. Both PC 1 and PC 2 together explain 90% of the variance. The second PC shows
that Pt(II) is the most significant More details about the PCA and DoE are given in the
supplementary.
78

(a) Significance of Variables on PC1 (b) Significance of Variables on PC2
(c) Significance of Variables on PC3
Figure 4.2: ANOVA Significance Plots for Principal Components 1, 2, and 3. Subplots (a),
(b), and (c) represent the significance of variables on PC1, PC2, and PC3, respectively. The
x-axis shows the LogWorth of each variable, a transformation of the p-value, while the yaxis lists the components (variables). The red dashed line indicates the p=0.05 significance
threshold. Variables to the right of this line are considered significant. Notably, PtNP is the
most significant variable affecting PC1, and Pt(II) for PC2 while no variables significantly
affect PC3 at the 5% significance level.
79

4.3 Training Matrix
In the subsequent phase of the experiment, a training matrix was constructed to reflect the
insights gained from the initial screening phase. In this training matrix, more levels were
given to components with higher significance, increasing their resolution. This approach
allows for a more detailed exploration of the variable space for those components that have
a greater impact on the outcome, which can lead to a more precise understanding of their
effects. Therefore, PtNPs mixture and Pt(II)/EG were set at 24 and 3 levels respectively,
with the other components set at 2 levels (Table 4.2), resulting in 144 experiments. The
training setup was configured with three pumps: one for IL/PVP, another for EG/Pt salt,
and the third for Pt NPs. These solutions were combined at a single junction, then channeled
through passive mixers to ensure thorough mixing before entering the spectroscopy analysis
zone (Figure 4.3). The spectra obtained from this training matrix were utilized for training
the ML models. To align our training data with the specific solvent environment employed,
we adopted a segmented approach to the training matrix. Separate sets of experiments were
conducted using two distinct ionic liquids—BMPyrr and BMPy——in addition to a control
set using pure ethylene glycol (EG) as the solvent, ensuring the models’ predictions are
solvent-specific and thereby more reliable.
Figure 4.3: Schematic representation of the revised experimental setup employed during
the training phase. The setup features three syringe pumps, used to administer the ionic
liquid with PVP, the ethylene glycol with Pt salt, and the Pt nanoparticle solution. This
configuration ensures a more refined control of the components at increased resolution levels,
as determined by the preliminary screening outcomes. The solutions converge at a single
microfluidic junction, where passive mixers facilitate optimal homogenization before the
mixture progresses to the spectroscopic analysis segment
80

Table 4.2: Levels of each component for the training Matrix.
Component Level 1 Level 2 Level 3 Level 4 Level 5 Level 6 . . . Level 24
Pt NPs mixture 0 4 9 13 17 22 . . . 100
Pt(II)/EG 0 50 100
IL/PVP 0 100
4.4 Machine Learning-assisted visible light spectrophotmetry
In this work, we considered two well-established machine learning (ML) models. The inputs
for these models were spectral data obtained from the refined matrix experiments, covering a wavelength range from 420 nm to 650 nm with approximately 990 data points per
spectrum. These spectral data was used to train the models to predict the corresponding
molar concentrations detailed in the matrix. The obtained spectral data were divided into
training and validation sets, with cross-validation employed to ensure robust model training.
Through this training, a predictive relationship between the spectral characteristics and the
concentrations of Pt nanoparticles was established.
The architecture of the model consisted of an input layer, various hidden layers with
ReLU (Rectified Linear Unit) activations, and an output layer. The dimension of the input
layer matched the number of features in the spectral dataset, while the output layer was
aligned with the number of target variables. The hidden layers were adjusted to balance the
complexity of the model against computational efficiency, with their sizes and the number
of layers being determined through iterative trials.
For model training, k-fold cross-validation was employed to enhance robustness against
overfitting and ensure model generalizability. This method divides the dataset into k equalsized subsets, or "folds," training the model on k-1 folds and validating it on the remaining
fold. This cycle is repeated k times, with each fold used once as the validation set, allowing
every data point to contribute to both training and validation. An early stopping mechanism
was implemented to halt training when the validation loss stopped decreasing, optimizing
training time and preventing overfitting. Parameters such as learning rate and batch size
81

were empirically determined.
The results of our k-fold cross-validation are illustrated in a series of scatter plots Figure
4.4, each representing one of the ten folds utilized in our model evaluation. The scatter plots
depict the actual vs. predicted concentrations of the PtNP component, with distinct markers
for training and validation data points. The R-squared (R2
) values, are annotated for both
training and validation sets within each fold. The consistency of the high R2 values across
the training folds, all above 0.98, indicates an excellent fit of the model to the training data,
capturing the underlying patterns with high fidelity. The validation R2 values, although
slightly lower, are commendably robust, predominantly above 0.75, affirming the model’s
generalizability and its capability to perform well on unseen data.
The performance of our model over multiple training epochs is graphically represented,
showing the interplay between training and validation loss. (Figure 4.5) Consistently across
the ten folds, we observe a rapid decline in training loss, indicating effective learning from
the training data. The validation loss, while displaying initial fluctuations, stabilizes closely
with the training loss, demonstrating the model’s ability to generalize beyond the training
set. Notably, the validation loss in each fold closely mirrors the training loss after the initial
epochs, suggesting that the model is not overfitting to the training data. This tight coupling
between training and validation loss is indicative of a well-tuned model exhibiting high
predictive reliability. In folds such as 1, 2, and 10, the validation loss descends and plateaus
at values suggesting minimal discrepancy between the model’s performance on both seen
and unseen data. This consistent pattern reaffirms the model’s robustness and underscores
its potential efficacy.
82

Figure 4.4: Scatter plots illustrating the relationship between actual and predicted concentrations of PtNPs across ten folds of cross-validation. Each plot denotes a separate
fold, with data points representing individual predictions for training (green) and validation
(blue) datasets. The accompanying R2 values highlight the model’s predictive accuracy,
with training R2 values consistently high and validation R² values showcasing the model’s
generalization capability.
83

Figure 4.5: Training and validation loss curves over successive epochs for each of the ten
folds in the K-fold cross-validation. The plots exhibit a sharp decline in training loss (green)
indicating effective learning, while the validation loss (blue) plateaus close to the training
loss, demonstrating the model’s ability to generalize. The congruence of these losses across
all folds signifies a balanced and well-generalized model.
84

4.5 Continuous Flow Synthesis of Pt Nanoparticles
The synthesis of Pt NPs is performed using a continuous flow process that employs a polyol
reduction method. Leveraging the predictive model fine-tuned through machine learning
analysis in earlier phases, reaction pathway for the formation of Pt NPs in ionic liquid can
be characterized. Schematic of the continuous flow setup utilized in the testing phase is
depicted in Figure 4.6. The setup features two syringe pumps that regulate the inflow of
reactants. These reactants converge and proceed through a reaction chamber maintained
at 150 °C, facilitating the reduction reaction necessary for Pt NP synthesis. Post-reaction,
a quenching mechanism rapidly cools the product stream to halt the reaction and stabilize
the nanoparticles before they are subjected to spectroscopic analysis. The setup facilitates
the investigation of the influence of two distinct ionic liquids, BMPyrr and BMPy, on the
formation kinetics of the nanoparticles. A parallel control experiment using pure EG as the
solvent is conducted to compare the effects of ionic liquids against a standard solvent system.
This setup allows for real-time monitoring of reaction coordinates providing valuable data
on the mechanistic impact of ionic liquids in nanoparticle formation.
Figure 4.6: Experimental setup for the continuous flow synthesis of Pt nanoparticles. The
schematic illustrates the configuration used in the testing phase. Two syringe pumps are
employed to deliver the reactants at controlled flow rates into a reaction chamber, where
they combine and react at a sustained temperature of 150C. Following the reaction, the
product stream is rapidly quenched to halt the reaction prior spectroscopic analysis.
85

4.6 Formation of IL-based Pt Nanoparticles
The spectra collected from the reaction are then fed into the machine learning model that was
previously trained. This model processes the spectral data to predict the concentration of
Pt NPs over time, providing us with a deeper understanding of the reaction kinetics. Figure
4.7 shows the concentration of a Pt NPs over time under different conditions. Each plot
represents data from a different solvent environment. Each data point is labeled "High" or
"Low," referring to the initial concentration of Pt precursor used in the reaction. The plots
collectively provide comparative insights into the impact of ionic liquids on Pt NP synthesis,
with clear differences observed in the kinetics of nanoparticle formation and growth.
The comparison aims to shed light on how the presence of ionic liquids alters the formation
kinetics of Pt NPs. The data depicted in Figure 4.7 underscore the kinetic differences
induced by the solvent choice, with IL-based environments showing distinctive behavior
compared to the conventional EG solvent. The findings from this research pave the way for
a deeper understanding of IL roles in nanoparticle synthesis and open avenues for customizing
nanoparticle properties through strategic solvent selection
4.7 Nanoparticle Size Estimation
In conjunction with the concentration versus time data, the size of the nanoparticles was
also estimated. This allowed for a quantitative assessment of the particle dimensions derived
from the model-predicted concentrations. By integrating the molar concentration from the
predictive models with the lattice constants and molar mass specific to platinum, we could
infer the average size of the nanoparticles synthesized under varying conditions. This size
estimation complements the temporal concentration data, providing a more comprehensive
understanding of the nanoparticle synthesis process in the continuous flow reactors. The
calculated nanoparticle sizes serve as a metric for comparing the efficiency and quality of
synthesis across different solvent systems and reaction parameters. The relationship between
86

Figure 4.7: Comparative analysis of Pt NP concentration against residence time across three
solvent systems. (a) BMPy, (b) BMPyrr, and (c) pure EG, with data points categorized as
"High" and "Low" to denote the initial precursor concentration levels. These plots demonstrate the distinctive kinetic behaviors of Pt NP synthesis in various solvents, highlighting
the influence of solvent properties and ionic liquid presence on the reaction’s progression
the nanoparticle concentration in solution and the size of the nanoparticles can be defined
by the following equation:
C[molNPs · L
−1
] = C[mg · mL−1
] · 3 · a
3
2π · D3
· M
(4.1)
Here, C[molNPs · L
−1
] is the molar concentration of nanoparticles, C[mg · mL−1
] is the
mass concentration, a is the face-centered cubic (fcc) lattice constant, D is the diameter of
the nanoparticles, and M is the molar mass of the metal. For platinum (Pt), with a lattice
constant a = 0.391 nm and molar mass M = 195.084 g·mol−1
, this equation allows us to
87

estimate the average size of the nanoparticles based on the predicted concentration.
The mass concentration (C[mg · mL−1
]) can be further related to the initial precursor
concentration, the molar mass of the platinum precursor (K2P tCl4), and the yield of the
reaction by the relation:
C[mg · mL−1
] = C0 × molar mass of K2P tCl4 × Yield × molar mass of Pt (4.2)
The predicted concentration represents the molar concentration of nanoparticles obtained
from the model’s predictions. By integrating these equations, we can calculate the estimated
nanoparticle size based on the model’s concentration predictions and compare it to the actual
sizes obtained from TEM analysis. This comparison allows for the validation of the model’s
accuracy and provides insights into the synthesis process of Pt nanoparticles. A summary
of the comparison of predicted and actual nanoparticle sizes is shown in Table 4.3.
Table 4.3: Comparison of predicted and actual nanoparticle sizes and yields for different
solvents.
Solvents C0 (mg/mL) Cp (M) Yield Calc. Size, nm Act. Size, nm
BMPYRR OTf 1.25 0.000 009 86 94 % 2.02 2.86
BMPYRR OTf 0.625 0.000 007 16 94 % 1.78 2.86
BMPY OTf 1.25 0.000 002 24 10 % 1.57 1.47
BMPY OTf 0.625 0.000 001 29 10 % 1.50 1.47
EG alone 1.25 0.000 005 62 98 % 2.47 1.91
EG alone 0.125 0.000 002 35 98 % 1.53 1.91
88

(a) BMPyrr, flow (b) BMPyrr, batch
(c) BMPy, flow (d) BMPY, batch
Figure 4.8: TEM images illustrating the variance in nanoparticle average size and standard
deviation (SD) across different solvents and reaction conditions. From top left to bottom
right: BMPyrr in flow, BMPyrr in batch, BMPy in flow, and BMPY in batch reactions.
89

4.8 Methods
4.8.1 Partial Least Square Regression
PLSR preprocessing involved mean-centering and variance scaling to improve model robustness. The optimal number of latent variables was determined through 10-fold cross-validation
to minimize prediction error. Model performance was evaluated using the mean-squared error
(MSE) and coefficient of determination (R2
)
4.8.2 Artificial Neural Network
The ANN architecture was designed with an input layer matching the dimensionality of the
spectral data, multiple hidden layers, and an output layer yielding concentration predictions.
We employed the rectified linear unit (ReLU) activation function and Adam optimizer. Grid
search was conducted to tune the number of neurons and layers in the ANN to maximize
the cross-validated R2
. The network was trained using backpropagation with a batch size
determined by empirical testing for optimal convergence. Performance was quantified using
cross-validated MSE and R2 metrics.
4.8.3 Machine Learning Model Training
To compare different machine learning algorithms—Partial Least Squares Regression (PLSR),
and Artificial Neural Networks (ANN)—we employed k-fold cross-validation. This methodology allows for the assessment of the predictive performance of each algorithm on our
dataset. Scikit-learn was used for the implementation of all models, ensuring a consistent
and reproducible framework for our machine learning pipeline. In the k-fold cross-validation
approach, the dataset was partitioned into k equal-sized folds. For each iteration, one fold
was retained as the validation data for model validation, and the remaining k-1 folds were
used as training data. The cross-validation process was then repeated k times, with each of
the k folds used exactly once as the validation data. The results from the k folds were then
90

averaged to produce a single estimation, providing insight into the model’s generalization
ability on an independent dataset. The predictive performance of each model was evaluated
using the R2 value and the MSE, which provides a measure of how well unseen data will be
predicted by the model. The R2 values obtained from the cross-validation process were then
compared to select the model that exhibits the best trade-off between prediction accuracy
and model complexity. For each algorithm, a thorough grid search was conducted to tune
hyperparameters, ensuring optimal performance. The specific hyperparameters included the
number of latent variables for PLSR, and the number of neurons and layers in the ANN.
These hyperparameters were fine-tuned to maximize the cross-validated R2 value. Upon
completion of the cross-validation, the best-performing model based on best R2 value was
selected to predict the concentration of PtNP during the synthesis process
91

4.9 Supplementary Machine Learning Models
4.9.1 Artificial Neural Network (ANN)
An Artificial Neural Network (ANN) is a computational model inspired by the human brain,
composed of interconnected nodes or neurons in layers. The output of a neuron is calculated
as:
y = f
Xn
i=1
wixi + b
!
(4.3)
where f is the activation function like ReLU. In multi-layer ANNs, the output of the l-th
layer is given by:
Y
(l) = f
(l)
(W(l)Y
(l−1) + B
(l)
) (4.4)
The network is trained using backpropagation with a commonly used loss function, the Mean
Squared Error (MSE).
4.9.2 Partial Least Squares Regression (PLSR) and Cross-Validation
Partial Least Squares Regression (PLSR) establishes the relationship between the predictor
matrix X and the response matrix Y :
Y = XB + E (4.5)
PLSR decomposes X and Y into latent structures and aims to maximize the covariance
between the scores T and U:
max
w,c
cov2
(Xw, Y c) (4.6)
92

4.10 Concluding Remarks
This study signifies a major advancement in the field of nanoparticle synthesis, particularly highlighting the significant influence of ionic liquids (ILs) on the fabrication of Pt
NPs using polyol reduction. Our approach has diverged from traditional methodologies by
integrating machine learning-assisted spectrophotometry to predict the concentration trajectories during the synthetic process. This innovative approach marks a departure from
conventional analytical methods, offering offering a novel perspective on the mechanistic
pathways of IL-based Pt NP formation. The study has successfully demonstrated that ILs
have a substantial impact on the synthesis of Pt NPs, suggesting that ILs may alter the
reaction kinetics. Future work should expand on these initial findings to unravel the specific
mechanisms involved. The research presented here serves as a foundation for future explorations into the kinetic analysis of nanoparticle synthesis, emphasizing the significant role
of ionic liquids. By utilizing machine learning-assisted spectrophotometry, we have gained
the capability to predict changes in nanoparticle concentration. The predictive capabilities
of machine learning-assisted spectrophotometry, as demonstrated in our research, provides
a foundational framework for understanding concentration dynamics. Yet, to fully harness
the complexities of nanoparticle formation and evolution, a complementary approach that
incorporates synchrotron-based methods is essential. Synchrotron X-ray scattering offers
unparalleled resolution and sensitivity, making it an ideal technique to complement the temporal concentration predictions of machine learning models. By combining the temporal data
from spectrophotometry with the structural insights provided by synchrotron techniques, we
can achieve a holistic view of the synthesis process.
93

References
[1] Limny Esther Pérez-Jiménez et al. “Enhancement of optoelectronic properties of TiO2
films containing Pt nanoparticles”. In: Results in Physics 12 (2019), pp. 1680–1685.
issn: 2211-3797. doi: https : / / doi . org / 10 . 1016 / j . rinp . 2019 . 01 . 046. url:
https://www.sciencedirect.com/science/article/pii/S2211379718331802.
[2] C. Dong, C. Lian, S. Hu, et al. “Size-dependent activity and selectivity of carbon
dioxide photocatalytic reduction over platinum nanoparticles”. In: Nature Communications 9 (2018), p. 1252. doi: 10.1038/s41467-018-03666-2.
[3] Tetsuya Kimata et al. “Platinum nanoparticles prepared by ion implantation exhibit high durability for fuel cell applications”. In: APL Materials 11.6 (June 2023),
p. 061115. issn: 2166-532X. doi: 10.1063/5.0148263.
[4] Licheng Bai et al. “Explaining the Size Dependence in Platinum-Nanoparticle-Catalyzed
Hydrogenation Reactions”. In: Angewandte Chemie International Edition 55.50 (2016),
pp. 15656–15661. doi: https://doi.org/10.1002/anie.201609663.
[5] Thurston Herricks, Jingyi Chen, and Younan Xia. “Polyol Synthesis of Platinum
Nanoparticles: Control of Morphology with Sodium Nitrate”. In: Nano Letters 4.12
(2004), pp. 2367–2371. doi: 10.1021/nl048570a.
[6] J. Dupont and J. D. Scholten. “On the structural and surface properties of transitionmetal nanoparticles in ionic liquids”. In: Chemical Society Reviews 39 (2010), pp. 1780–
1804. doi: 10.1039/B822551f.
[7] T. Welton. “Room-temperature ionic liquids. Solvents for synthesis and catalysis”. In:
Chemical Reviews 99 (Aug. 1999), pp. 2071–2083.
[8] C. T. Riche et al. “Flow invariant droplet formation for stable parallel microreactors”.
In: Nature Communications 7 (2016), p. 10780.
94

[9] L. L. Lazarus et al. “Two-Phase Microfluidic Droplet Flows of Ionic Liquids for the
Synthesis of Gold and Silver Nanoparticles”. In: ACS Applied Materials & Interfaces
4 (June 2012), pp. 3077–3083. doi: 10.1021/Am3004413.
[10] Lanja R. Karadaghi et al. “Techno-Economic Analysis of Recycled Ionic Liquid Solvent Used in a Model Colloidal Platinum Nanoparticle Synthesis”. In: ACS Sustainable
Chemistry & Engineering 9.1 (2021), pp. 246–253. doi: 10.1021/acssuschemeng.
0c06993.
[11] Tao Li, Andrew J. Senesi, and Byeongdu Lee. “Small Angle X-ray Scattering for
Nanoparticle Research”. In: Chemical Reviews 116.18 (2016), pp. 11128–11180. doi:
10.1021/acs.chemrev.5b00690.
[12] B.L. Henke, E.M. Gullikson, and J.C. Davis. “X-Ray Interactions: Photoabsorption,
Scattering, Transmission, and Reflection at E = 50-30,000 eV, Z = 1-92”. In: Atomic
Data and Nuclear Data Tables 54.2 (1993), pp. 181–342. issn: 0092-640X. doi: https:
//doi.org/10.1006/adnd.1993.1013. url: https://www.sciencedirect.com/
science/article/pii/S0092640X83710132.
[13] “J. Chem. Educ.” In: 39.7 (July 1962), p. 333.
[14] T. G. Mayerhöfer, S. Pahlow, and J. Popp. “ChemPhysChem”. In: ChemPhysChem
21 (2020), p. 2029.
[15] K. Okitsu. “UV-Vis Spectroscopy for Characterization of Metal Nanoparticles Formed
from Reduction of Metal Ions During Ultrasonic Irradiation”. In: UV-VIS and Photoluminescence Spectroscopy for Nanomaterials Characterization. Berlin, Heidelberg:
Springer, 2013. doi: 10.1007/978-3-642-27594-4_4.
[16] Robina Begum et al. “Applications of UV/Vis Spectroscopy in Characterization and
Catalytic Activity of Noble Metal Nanoparticles Fabricated in Responsive Polymer
Microgels: A Review”. In: Critical Reviews in Analytical Chemistry 48.6 (2018), pp. 503–
516. doi: 10.1080/10408347.2018.1451299.
95

[17] R.M. Tilaki, A. Iraji zad, and S.M. Mahdavi. “Size, composition and optical properties
of copper nanoparticles prepared by laser ablation in liquids”. In: Appl. Phys. A 88.2
(2007), pp. 415–419.
[18] D. Paramelle et al. “A rapid method to estimate the concentration of citrate capped
silver nanoparticles from UV-visible light spectra”. In: Analyst 139.19 (2014), pp. 4855–
4861.
[19] X. Liu et al. “Extinction coefficient of gold nanoparticles with different sizes and
different capping ligands”. In: Colloids Surf. B Biointerfaces 58.1 (2007), pp. 3–.
[20] M.A. Garcia. “Surface plasmons in metallic nanoparticles: fundamentals and applications”. In: J. Phys. D Appl. Phys. 44.28 (2011).
[21] A. Bragau et al. “Platinum nanoparticles for nanocomposite membranes preparation”.
In: Rom. J. Inf. Sci. Technol. 13 (2010), pp. 350–357.
[22] “Langmuir”. In: 20.7 (2004). Publication Date: February 20, 2004, pp. 2915–2920. doi:
10.1021/la0361060. url: https://doi.org/10.1021/la0361060.
[23] Alberto Parola, Roberto Piazza, and Vittorio Degiorgio. “Optical extinction, refractive index, and multiple scattering for suspensions of interacting colloidal particles”.
In: J. Chem. Phys. 141.12 (2014), p. 124902. doi: 10.1063/1.4895961. url: https:
//doi-org.libproxy2.usc.edu/10.1063/1.4895961.
96

Chapter 5: In situ Characterization for Screening Colloidal Nanoparticle syntheses
5.1 Introduction
Over the past few decades, there has been a great deal of effort in investigating and understanding colloidal nanoparticle nucleation and growth processes. One of the major benefits of
using continuous flow systems is the ability to probe in-line to characterize chemical reactions
and nanoparticle syntheses.1
In situ characterization of nanoparticles is pivotal for real-time analysis and understanding of their synthesis and behavior. Such techniques enable scientists to observe nanoparticle
formation as it occurs, providing critical insights into their growth mechanisms and dynamics.2 This knowledge is crucial for optimizing the synthesis process, allowing for precise
control over nanoparticle characteristics like size, shape, and distribution.
Recent advancements in in situ characterization techniques have greatly enhanced our
ability to study colloidal nanoparticles. For instance, in situ TEM has become an invaluable
tool, allowing for the direct observation of the nucleation and growth of nanoparticles under
various conditions.3 This technique has been particularly transformative in understanding
the dynamics of nanoparticle synthesis at the nanoscale. Moreover, in situ X-ray diffraction
provides insights into the crystallographic structure of nanoparticles as they form, offering valuable data on phase transformations and structural evolution.4 These advancements
have not only deepened our understanding of nanoparticle properties but also aided in the
development of novel materials with tailored functionalities.
While in situ TEM offers unparalleled resolution and detail in the observation of colloidal
nanoparticles, it is not without its limitations. One significant drawback of in situ TEM is its
97

Figure 5.1: A schematic of a tube with an in-flow reaction where in situ characterization
methods are utilized to screen reactions or characterize nanoparticle synthesis processes
across a wide range of the electromagnetic spectrum
restriction to a very small volume of liquid solutions, which can limit the representativeness
of the observed phenomena. The conditions within the TEM, including high vacuum and
electron beam irradiation, can also alter the behavior of nanoparticles, potentially leading
to artifacts that do not represent their behavior under normal synthesis conditions. Furthermore, the high costs and technical expertise required for TEM make it less accessible for
routine analysis
In contrast, Small-Angle X-ray Scattering (SAXS) offers several advantages that address
some of these limitations. Unlike TEM, SAXS can analyze a larger sample volume, providing
a more statistically representative view of the nanoparticles. This method is also non98

destructive and does not significantly alter the sample during analysis. SAXS is particularly
powerful in providing averaged information over a large number of particles, making it ideal
for studying the size distribution and shape of nanoparticles in solution. Moreover, SAXS
experiments can often be conducted under more realistic environmental conditions, allowing
for a more accurate representation of the synthesis and behavior of nanoparticles in situ.
5.1.1 Integration of X-ray Scattering in Flow Reactors
The integration of X-ray scattering techniques, such as Small-Angle X-ray Scattering (SAXS)
and Wide-Angle X-ray Scattering (WAXS), into continuous flow reactors has significantly
enhanced the capability to study colloidal nanoparticle synthesis in real time. These techniques provide critical insights into the size, shape, and internal structure of nanoparticles
as they form.
A number of in-line techniques have been applied, including optical and fluorescence
microscopy, UV-vis spectroscopy, and small-angle X-ray scattering (SAXS). A review of
spectroscopically probed microreactors has been reported.1
Figure 5.1 shows schematic of a tube with an in-flow reaction where in situ characterization methods are utilized to screen reactions or characterize nanoparticle syntheses processes.
One example of studying nanoparticle nucleation and growth processes is in situ transmission
electron microscopy (TEM) which is capable of studying the real-time formation of nanoparticles under electron-beam illumination. However, some issues with TEM include that the
reaction conditions are limited to the small volume of liquid solutions, and only a limited
number of NPs are focused in the field of view. In contrast, X-ray scattering provides more
statistically rich data to describe NPs formation kinetics. This also became more efficient
with the use of synchrotron radiation facilities.
99

5.1.2 Scattering Theory
A simple layout for a conventional small angle X-ray scattering experiment is shown in Figure . In a typical X-ray scattering experiment, a highly collimated beam of monochromatic
X-rays is transmitted through a sample of interest. The scattered X-rays are collected at
a continuous range of scattering angles by a two-dimensional detector azimuthally in 360°,
while the transmitted primary beam is absorbed by a beam stop placed in front of the detector. Subsequently, the collected two-dimensional scattering pattern is radially integrated
to provide a one-dimensional scattering function I(q), where q is the length of the scattering
vector. The relationship between the scattering angle, the scattering vector, and the particle
wavelength is expressed by
q =
4π sin(θ)
λ
(5.1)
The scattering angle which is defined by 2θ dictates the probe length scale which is expressed
as D =
2π
q
, which illustrates a real space window of observation that typically ranges from
nm to µm.
X-ray scattering, therefore, offers information over a wide range of scattering angles.
This makes X-ray scattering a powerful tool for investigating the structural properties of
nanoparticles while providing higher statistics to describe their formation kinetics.2
X-ray Transmission
X-ray path length through a sample is determined by the depth of the sample or channel
which is parallel to the beam direction. The path length is an important parameter when
designing X-ray compatible devices. This will ensure achieving proper signal that can be
further reduced to meaningful data. The intensity of the beam at a distance x can be
expressed using the following equation
Ix = I0e
−µd (5.2)
100

where Ix is the intensity at depth x, I0 is the initial intensity, and µ is the linear attenuation
coefficient, which is a constant that describes the fraction of attenuated incident photons in
a monoenergetic beam per unit thickness of a material. Eq 5.2 can be rearranged by taking
the log of both sides to give an equation for µ
µ =
1
d
ln
I0
Ix

(5.3)
The transmission T through a sample of path length d follows T ∼ exp(−µd). The
detected scattering is proportional to Ix
I0
∼ d exp(−µd). This results in a maximum intensity
for d =
1
µ
, which is the ideal path length for a given setup. A minimum X-ray transmission
of 5% through the microfluidic device, with an optimal value near 33% is recommended in
order to process data. To achieve such goals, we must consider both the X-ray path length
and the material of the microfluidic chip being used.
5.2 Particle Size and Shape
The primary information obtained from a SAXS pattern is the size and shape of the nanoparticles. The intensity of scattered X-rays at different angles is directly related to these physical
properties. For spherical particles, the form factor P(q) can be used to describe the scattering
intensity:
I(q) = I(0) · P(q) · S(q) (5.4)
Here, P(q) is the form factor that depends on the particle size and shape, S(q) is the
structure factor that accounts for interparticle interactions, and I(0) is the forward scattering
intensity.
101

5.3 Radius of Gyration
The Guinier plot, a log plot of intensity versus q
2
, can be used to estimate the radius of
gyration Rg of the particles, providing an average size estimation. The Guinier approximation
is valid for small q values (small angles), where Rg is related to the intensity as:
ln[I(q)] = ln[I(0)] −
R2
g
q
2
3
(5.5)
5.4 Porod’s Law
At higher angles (large q values), Porod’s law comes into effect, providing information about
the surface characteristics of the nanoparticles. The intensity in this regime is inversely
proportional to q
4
, indicative of the interface between particles and the surrounding medium.
X-ray Absorption and Attenuation in Ionic Liquids
For X-ray absorption in materials, the mass absorption coefficient µ is a crucial parameter.
It quantifies how much X-ray beams are attenuated as they pass through a material. For a
pure substance, µ is directly related to the atomic absorption cross section σA (cm2/atom)
through the relation:
µ =
NA
A
σA (5.6)
where NA is Avogadro’s number and A is the atomic mass. For compounds, µ is the sum
of the absorption cross sections of the constituent atoms, given by:
µ =
NA
MW
X
i
xiσAi (5.7)
with MW being the molecular weight of the compound, xi the fraction of the i-th element,
and σAi its atomic absorption cross-section.
102

In the case of the ionic liquid 1-Butyl-3-methylimidazolium bis(trifluoromethanesulfonyl)imide
(BMIM-Tf2N), the mass absorption coefficient has been calculated using tabulated values
for the photo-absorption cross section. The following table illustrates these values for the
constituent elements at an energy level of 8000 eV.
Table 5.1: Photo-absorption cross-section for BMIM-Tf2N constituent elements at 8000 eV.
Element Atomic Number, Z Atomic Weight Photo-absorption Cross-Section (cm2/g)
Hydrogen 1 1 5.88e-03
Carbon 6 12.01 4.26
Fluorine 9 18.998 15.41
Nitrogen 7 14 6.984
Oxygen 8 16 11.22
Sulfur 16 32.07 91.31
The calculated µ for BMIM-Tf2N was found to be 21.79 cm2/g. Using this value, the
transmission T of X-rays through the material can be approximated by Beer-Lambert law:
T ≈ exp(−µx) (5.8)
where x represents the mass thickness of the material, given by the product of density
ρ and thickness t. This model provides a fundamental understanding of the interaction
between X-rays and ionic liquids, facilitating the characterization of such materials in various
scientific and industrial applications. Figure 5.4 shows the Transmission for three different
ionic liquids given a path length of 900 µm.
5.5 Dimensional Analysis
The design of the microfluidic chip or the capillary tube, choice of material, and temperature
limitations are all important factors to consider in the fabrication of X-ray compatible flow
devices. An important consideration is the width of the microfluidic channel (the dimension
which is perpendicular to the X-ray beam). The width should be wide enough to accommodate the size of the beam to avoid scattering that may arise from the channel edges.
103

Figure 5.2: Transmission of the beam as a function of photon energy for three different ionic
liquids with a path length of 900 µm
One important thing to note is that the beam size under consideration is the total beam
width rather than the full width at half maximum (FWHM), as weak signals from the outer
beam will cause parasitic background scattering due to interaction of the X-ray beam tail
with channel walls. The limitations of the channel width are greatly alleviated with the
advancement of synchrotron beamlines as the beam sizes are typically smaller ( 1 – 5 µm).
Another crucial design consideration relating to SAXS-compatible devices is to optimize the
channel depth (which is the dimension parallel to the beam path length) such that the depth
is as close as possible to the optimal path length. A major bottleneck to the fabrication of
X-ray compatible flow devices is the low energies (8 keV) for lab-based SAXS instruments.
For experiments that require higher temperatures, the choice of material is limited. Quartz
and Borosilicate glass are common choices for high temperature experiments. Higher energy
sources can accommodate thicker walls up to a certain thickness depending on the photon
energy used (Figure 5.3). Therefore, plate microfluidic chips made of quartz or borosilicate
104

glass are not suitable for lab-based SAXS due to low transmission.
Figure 5.3: Transmission of the beam as a function of photon energy for Quartz and Borosilicate glass consider 100 µm and 200 µm walls
An alternative material that is increasingly utilized is Kapton, a polyimide film known
for its excellent chemical stability, high thermal resistance, and particularly its X-ray transparency. Kapton can maintain its physical integrity at temperatures up to 400°C and has a
low absorption coefficient for X-rays, making it highly suitable for high-temperature SAXS
experiments even at lower energies. Figure 5.4 shows Transmission as funciton of photon
energy for Kapton tubes with 100 µm and 200 µm wall thicknesses.
5.6 Preliminary Reactor Design for In situ X-ray Scattering
In contrast to most commonly used flow cells or microfluidic devices in X-ray experiments,
we have proposed a tubular reactor made of X-ray-compatible polyimide tubes (0.9 mm ID
and 1 mm OD) embedded on a heated copper plate that can be used to perform in situ
characterization of nanoparticle syntheses at high temperatures. Figure 5.5 shows a copper
plate with 1 mm grooves in a serpentine form to mimic a serpentine plate reactor. he grooves’
curvature is carefully calculated to prevent kinking of the tubes. The copper plate acts as
a heater and a holder for the polyimide tube. A slit in the middle of the plate (1.5 mm
wide) provides the screening window where the X-rays pass through.The reactor’s design
105

Figure 5.4: Transmission of the beam as a function of photon energy for Kapton tubes with
different wall thicknesses
incorporates ten distinct measurement points along its length, with each point offering a
specific temporal resolution of the reaction profile at steady state. The inlet and the outlet
of the reactor are connected using commercially available fittings. Heating is achieved by
inserting cartridge heaters from side (5.5a) of the plate without covering the middle slit.
Copper possesses high thermal conductivity (386 W/m K), which makes ramp up time
fast. This design provides a robust and cost-effective method for in situ X-ray scattering
measurements in continuous flow at high temperatures for in situ SAXS.
106

Figure 5.5: Kapton-based tubular reactor.(a) Side perspective displaying inlets for cartridge
heaters and sensors. (b) Interior view illustrating the arrangement of cartridge heaters. (c)
Frontal view showcasing the alignment of the Kapton tubes within the grooves and the
central slits designated for X-ray examination.
107

5.6.1 Future Outlook
Implementing real-time X-ray scattering analysis during the synthesis process could provide
invaluable insights into the nucleation and growth kinetics of Pt nanoparticles in ILs. However, due to the challenges in analyzing ILs with X-ray scattering, primarily because of their
high attenuation coefficient, future work could focus on refining these techniques. This could
involve the use of more advanced X-ray sources like synchrotron radiation to counteract the
high photoabsorption cross-section of elements commonly found in ILs.
Exploring a broader range of ILs and varying the composition of Pt nanoparticles could
reveal new insights into the synthesis process and the properties of the resulting nanoparticles. This could lead to the development of nanoparticles with tailored properties for specific
applications.
Additionally, Integrating machine learning algorithms with X-ray scattering data could
significantly enhance the interpretation and predictive power of the experiments. This approach could be particularly effective in identifying subtle changes and patterns in the scattering data, leading to a more profound understanding of the synthesis process.
108

References
[1] Benjamin A Rizkin, Filip G Popovic, and Ryan L Hartman. “Review Article: Spectroscopic microreactors for heterogeneous catalysis”. In: Journal of Vacuum Science
& Technology A 37.050801 (2019).
[2] Tao Li, Andrew J Senesi, and Byeongdu Lee. “Chemical Reviews”. In: Chemical Reviews 116.18 (2016), pp. 11128–11180.
[3] James J. De Yoreo. “In-situ liquid phase TEM observations of nucleation and growth
processes”. In: Progress in Crystal Growth and Characterization of Materials 62.2
(2016). Special Issue: Recent Progress on Fundamentals and Applications of Crystal
Growth; Proceedings of the 16th International Summer School on Crystal Growth
(ISSCG-16), pp. 69–88. issn: 0960-8974. doi: https : / / doi . org / 10 . 1016 / j .
pcrysgrow . 2016 . 04 . 003. url: https : / / www . sciencedirect . com / science /
article/pii/S0960897416300043.
[4] C. S. Santos et al. “In operando EPR study for real-time monitoring of redox processes
in ionic liquid media during electrochemical cycling”. In: Scientific Reports 8.1 (2018),
p. 12413. doi: 10.1038/s41598- 018- 30392- y. url: https://www.nature.com/
articles/s41598-018-30392-y.
109

Bibliography
[1] K. Abdel-Latif et al. “Self-Driven Multistep Quantum Dot Synthesis Enabled by Autonomous Robotic Experimentation in Flow”. In: Adv. Intell. Syst. 3 (2021), p. 2000245.
[2] C. D. Ahrberg, J. W. Choi, and B. G. Chung. “Droplet-Based Synthesis of Homogeneous Magnetic Iron Oxide Nanoparticles”. In: Beilstein J. Nanotechnol. 9 (2018),
pp. 2413–2420. doi: 10.3762/bjnano.9.227.
[3] P. A. Alaba et al. “Molybdenum Carbide Nanoparticle: Understanding the Surface
Properties and Reaction Mechanism for Energy Production towards a Sustainable
Future”. In: Renew. Sustain. Energy Rev. 91 (2018), p. 287.
[4] Y. Aoki, H. Tominaga, and M. Nagai. “Hydrogenation of CO on Molybdenum and
Cobalt Molybdenum Carbide CatalystsMass and Infrared Spectroscopy Studies”. In:
Catal. Today 215 (2013), p. 169.
[5] F. G. Baddour et al. “An Exceptionally Mild and Scalable Solution-Phase Synthesis
of Molybdenum Carbide Nanoparticles for Thermocatalytic CO2 Hydrogenation”. In:
J. Am. Chem. Soc. 142 (2020), p. 1010.
[6] Licheng Bai et al. “Explaining the Size Dependence in Platinum-Nanoparticle-Catalyzed
Hydrogenation Reactions”. In: Angewandte Chemie International Edition 55.50 (2016),
pp. 15656–15661. doi: https://doi.org/10.1002/anie.201609663.
[7] Robina Begum et al. “Applications of UV/Vis Spectroscopy in Characterization and
Catalytic Activity of Noble Metal Nanoparticles Fabricated in Responsive Polymer
Microgels: A Review”. In: Critical Reviews in Analytical Chemistry 48.6 (2018), pp. 503–
516. doi: 10.1080/10408347.2018.1451299.
110

[8] M. A. Bezerra et al. “Response Surface Methodology (RSM) as a Tool for Optimization in Analytical Chemistry”. In: Talanta 76 (2008), p. 965.
[9] Matthew C Blosser et al. “Effect of cholesterol on permeability of carbon dioxide
across lipid membranes”. In: bioRxiv (Nov. 2020). Preprint. doi: 10.1101/2020.11.
16.384958.
[10] G. E. P. Box, J. S. Hunter, and W. G. Hunter. Statistics for Experimenters: Design,
Innovation, and Discovery. 2nd ed. Wiley series in probability and statistics. Hoboken,
N.J: Wiley-Interscience, 2005.
[11] A. Bragau et al. “Platinum nanoparticles for nanocomposite membranes preparation”.
In: Rom. J. Inf. Sci. Technol. 13 (2010), pp. 350–357.
[12] E. J. Braham et al. “Machine Learning-Directed Navigation of Synthetic Design
Space: A Statistical Learning Approach to Controlling the Synthesis of Perovskite
Halide Nanoplatelets in the Quantum-Confined Regime”. In: Chem. Mater. 31 (2019),
p. 3281.
[13] E. J. Braham et al. “Navigating the Design Space of Inorganic Materials Synthesis Using Statistical Methods and Machine Learning”. In: Dalton Trans. 49 (2020),
p. 11480.
[14] C. P. Breen et al. “Ready, Set, Flow! Automated Continuous Synthesis and Optimization”. In: Trends Chem. 3 (2021), p. 373.
[15] N. D. Burrows et al. “Understanding the Seed-Mediated Growth of Gold Nanorods
through a Fractional Factorial Design of Experiments”. In: Langmuir 33 (2017),
p. 1891.
[16] Neil Convery and Nikolaj Gadegaard. “30 years of microfluidics”. In: Micro and Nano
Engineering 2 (2019), pp. 76–91. issn: 2590-0072.
[17] B. F. Cottam et al. “Accelerated Synthesis of Titanium Oxide Nanostructures Using
Microfluidic Chips”. In: Lab Chip 7 (2007), pp. 167–169. doi: 10.1039/B615688C.
111

[18] Jinlei Cui et al. “Diffusivity and Structure of Room Temperature Ionic Liquid in
Various Organic Solvents”. In: The Journal of Physical Chemistry B 124.44 (2020),
pp. 9931–9937. doi: 10.1021/acs.jpcb.0c07582.
[19] James J. De Yoreo. “In-situ liquid phase TEM observations of nucleation and growth
processes”. In: Progress in Crystal Growth and Characterization of Materials 62.2
(2016). Special Issue: Recent Progress on Fundamentals and Applications of Crystal
Growth; Proceedings of the 16th International Summer School on Crystal Growth
(ISSCG-16), pp. 69–88. issn: 0960-8974. doi: https : / / doi . org / 10 . 1016 / j .
pcrysgrow . 2016 . 04 . 003. url: https : / / www . sciencedirect . com / science /
article/pii/S0960897416300043.
[20] A. J. deMello. “Control and Detection of Chemical Reactions in Microfluidic Systems”.
In: Nature 442 (2006), p. 394.
[21] D. H. Doehlert. “Uniform Shell Designs”. In: Appl. Stat. 19 (1970), p. 231.
[22] C. Dong, C. Lian, S. Hu, et al. “Size-dependent activity and selectivity of carbon
dioxide photocatalytic reduction over platinum nanoparticles”. In: Nature Communications 9 (2018), p. 1252. doi: 10.1038/s41467-018-03666-2.
[23] Z. Dong et al. “Scale-up of Micro- and Milli-Reactors: An Overview of Strategies,
Design Principles and Applications”. In: Chem. Eng. Sci. X 10 (2021), p. 100097.
[24] J. Dupont and J. D. Scholten. “On the structural and surface properties of transitionmetal nanoparticles in ionic liquids”. In: Chemical Society Reviews 39 (2010), pp. 1780–
1804. doi: 10.1039/B822551f.
[25] Robert W. Epps et al. “Automated microfluidic platform for systematic studies of
colloidal perovskite nanocrystals: towards continuous nano-manufacturing”. In: Lab
Chip 17 (23 2017), pp. 4040–4047. doi: 10.1039/C7LC00884H. url: http://dx.doi.
org/10.1039/C7LC00884H.
112

[26] V. Fath et al. “Self-Optimising Processes and Real-Time-Optimisation of Organic
Syntheses in a Microreactor System Using Nelder–Mead and Design of Experiments”.
In: React. Chem. Eng. 5 (2020), p. 1281.
[27] Ethan Y. Feng et al. “Investigation of the optical properties of uniform platinum, palladium, and nickel nanocrystals enables direct measurements of their concentrations
in solution”. In: Colloids and Surfaces A: Physicochemical and Engineering Aspects
601 (2020), p. 125007. issn: 0927-7757.
[28] N. M. Fhionnlaoich et al. “DoE-It-Yourself: A Case Study for Implementing Design
of Experiments into Nanoparticle Synthesis”. In: (2020). doi: 10.26434/chemrxiv.
8198420.v1. url: https://doi.org/10.26434/chemrxiv.8198420.v1.
[29] “Flow chemistry remains an opportunity for chemists and chemical engineers”. In:
ScienceDirect (). url: https://www.sciencedirect.com/science/article/pii/
S2211339820300290.
[30] Focus. “F”. In: Lab Chip 4 (2004), 11N.
[31] B. K. Gale et al. “A Review of Current Methods in Microfluidic Device Fabrication
and Future Commercialization Prospects”. In: Inventions 3.3 (2018), p. 60.
[32] M.A. Garcia. “Surface plasmons in metallic nanoparticles: fundamentals and applications”. In: J. Phys. D Appl. Phys. 44.28 (2011).
[33] A. M. Ghaemmaghami et al. “Biomimetic tissues on a chip for drug discovery”. In:
Drug Discov. Today 17.3–4 (2012), pp. 173–181.
[34] P. A. Gunatillake and R. Adhikari. “Nondegradable synthetic polymers for medical
devices and implants”. In: Biosynthetic Polymers for Medical Applications. Woodhead
Publishing, 2016, pp. 33–62.
[35] A. Günther and K. F. Jensen. “Multiphase microfluidics: From flow characteristics
to chemical and materials synthesis”. In: Lab Chip 6.12 (2006), pp. 1487–1503. doi:
10.1039/b609851g.
113

[36] B. L. Hall et al. “Autonomous Optimisation of a Nanoparticle Catalysed Reduction
Reaction in Continuous Flow”. In: Chem. Commun. 57 (2021), p. 4926.
[37] D. J. Harrison et al. “Micromachining a miniaturized capillary electrophoresis-based
chemical analysis system on a chip”. In: Science 261.5123 (1993), pp. 895–897. doi:
10.1126/science.261.5123.895.
[38] Ryan L Hartman, Jonathan P McMullen, and Klavs F Jensen. “Deciding whether to
go with the flow: evaluating the merits of flow reactors for synthesis”. In: Angewandte
Chemie International Edition 50.33 (2011). Epub 2011 Jun 27. PMID: 21710673,
pp. 7502–7519. doi: 10.1002/anie.201004637.
[39] B.L. Henke, E.M. Gullikson, and J.C. Davis. “X-Ray Interactions: Photoabsorption,
Scattering, Transmission, and Reflection at E = 50-30,000 eV, Z = 1-92”. In: Atomic
Data and Nuclear Data Tables 54.2 (1993), pp. 181–342. issn: 0092-640X. doi: https:
//doi.org/10.1006/adnd.1993.1013. url: https://www.sciencedirect.com/
science/article/pii/S0092640X83710132.
[40] Thurston Herricks, Jingyi Chen, and Younan Xia. “Polyol Synthesis of Platinum
Nanoparticles: Control of Morphology with Sodium Nitrate”. In: Nano Letters 4.12
(2004), pp. 2367–2371. doi: 10.1021/nl048570a.
[41] V. Hessel, J. C. Schouten, and A. Renken. Micro Process Engineering: A Comprehensive Handbook. John Wiley Sons, 2009.
[42] V. Hessel et al. “Laminar mixing in different interdigital micromixers: I. Experimental
characterization”. In: AIChE Journal 49.3 (2003), pp. 566–577. doi: https://doi.
org/10.1002/aic.690490304.
[43] N. Holmes et al. “Self-Optimisation of the Final Stage in the Synthesis of EGFR
Kinase Inhibitor AZD9291 Using an Automated Flow Reactor”. In: React. Chem.
Eng. 1 (2016), p. 366.
114

[44] T. Hyeon, M. Fang, and K. S. Suslick. “Nanostructured Molybdenum Carbide: Sonochemical Synthesis and Catalytic Properties”. In: J. Am. Chem. Soc. 118 (1996),
p. 5492.
[45] E. Iglesia et al. “Bifunctional Reactions of Alkanes on Tungsten Carbides Modified
by Chemisorbed Oxygen”. In: J. Catal. 131 (1991), p. 523.
[46] E. Iglesia et al. “Synthesis, Characterization, and Catalytic Properties of Clean and
Oxygen-Modified Tungsten Carbides”. In: Catal. Today 15 (1992), p. 307.
[47] “J. Chem. Educ.” In: 39.7 (July 1962), p. 333.
[48] K. F. Jensen. “Microreaction Engineering Is Small Better?” In: Chem. Eng. Sci. 56
(2001), p. 293.
[49] Lanja R. Karadaghi et al. “Techno-Economic Analysis of Recycled Ionic Liquid Solvent Used in a Model Colloidal Platinum Nanoparticle Synthesis”. In: ACS Sustainable
Chemistry & Engineering 9.1 (2021), pp. 246–253. doi: 10.1021/acssuschemeng.
0c06993.
[50] Tetsuya Kimata et al. “Platinum nanoparticles prepared by ion implantation exhibit high durability for fuel cell applications”. In: APL Materials 11.6 (June 2023),
p. 061115. issn: 2166-532X. doi: 10.1063/5.0148263.
[51] P. J. Kitson et al. “Configurable 3D-Printed Millifluidic and Microfluidic ‘Lab on a
Chip’ Reactionware Devices”. In: Lab Chip 12.18 (2012), pp. 3267–3271.
[52] Tae Joon Kwak et al. “Convex Grooves in Staggered Herringbone Mixer Improve
Mixing Efficiency of Laminar Flow in Microchannel”. In: PLOS ONE 11 (Nov. 2016),
pp. 1–15. doi: 10.1371/journal.pone.0166068.
[53] “Langmuir”. In: 20.7 (2004). Publication Date: February 20, 2004, pp. 2915–2920. doi:
10.1021/la0361060. url: https://doi.org/10.1021/la0361060.
115

[54] L. L. Lazarus et al. “Two-Phase Microfluidic Droplet Flows of Ionic Liquids for the
Synthesis of Gold and Silver Nanoparticles”. In: ACS Applied Materials & Interfaces
4 (June 2012), pp. 3077–3083. doi: 10.1021/Am3004413.
[55] Laura L. Lazarus et al. “Two-Phase Microfluidic Droplet Flows of Ionic Liquids for the
Synthesis of Gold and Silver Nanoparticles”. In: ACS Applied Materials & Interfaces
4.6 (2012), pp. 3077–3083. doi: 10.1021/am3004413.
[56] J. S. Lee, S. T. Oyama, and M. Boudart. “Molybdenum Carbide Catalysts: I Synthesis
of Unsupported Powders”. In: J. Catal. 106 (1987), p. 125.
[57] R. B. Levy and M. Boudart. “Platinum-Like Behavior of Tungsten Carbide in Surface
Catalysis”. In: Science 181 (1973), p. 547.
[58] S. Li et al. “Automated Microfluidic Screening of Ligand Interactions during the
Synthesis of Cesium Lead Bromide Nanocrystals”. In: Mol. Syst. Des. Eng. 5 (2020),
p. 1118.
[59] Tao Li, Andrew J Senesi, and Byeongdu Lee. “Chemical Reviews”. In: Chemical Reviews 116.18 (2016), pp. 11128–11180.
[60] Tao Li, Andrew J. Senesi, and Byeongdu Lee. “Small Angle X-ray Scattering for
Nanoparticle Research”. In: Chemical Reviews 116.18 (2016), pp. 11128–11180. doi:
10.1021/acs.chemrev.5b00690.
[61] Z. Lin et al. “Hydrodeoxygenation of Biomass-Derived Oxygenates over Metal Carbides: From Model Surfaces to Powder Catalysts”. In: Green Chem. 20 (2018), p. 2679.
[62] C. Liu et al. “Preparation of Nanostructured Molybdenum Carbides for CO Hydrogenation”. In: RSC Adv. 4 (2014), p. 20948.
[63] X. Liu et al. “Extinction coefficient of gold nanoparticles with different sizes and
different capping ligands”. In: Colloids Surf. B Biointerfaces 58.1 (2007), pp. 3–.
116

[64] Z. Liu et al. “Application of the Factorial Design of Experiments to Hydrothermal
Synthesis of Lithium Iron Phosphate”. In: Int. J. Appl. Ceram. Technol. 17 (2020),
p. 1231.
[65] Z. Ma and F. Zaera. “Title of the Chapter”. In: Encyclopedia of Inorganic and Bioinorganic Chemistry. Ed. by R. A. Scott. John Wiley Sons, Ltd., 2014.
[66] R. M. Maceiczyk and A. J. deMello. “Fast and Reliable Metamodeling of Complex
Reaction Spaces Using Universal Kriging”. In: J. Phys. Chem. C 118 (2014), p. 20026.
[67] A. Manz, N. Graber, and H. M. Widmer. “Miniaturized total chemical analysis systems: A novel concept for chemical sensing”. In: Sensors and Actuators B: Chemical
1.1–6 (1990), pp. 244–248. issn: 0925-4005.
[68] A. Marinas and R. A. Sheldon. “Utilisation of Biomass for Fuels and Chemicals: The
Road to Sustainability”. In: Catal. Today 167 (2011), p. 1.
[69] C. Mateos, M. J. Nieves-Remacha, and J. A. Rincón. “Automated Platforms for Reaction Self-Optimization in Flow”. In: React. Chem. Eng. 4 (2019), p. 1536.
[70] T. G. Mayerhöfer, S. Pahlow, and J. Popp. “ChemPhysChem”. In: ChemPhysChem
21 (2020), p. 2029.
[71] R. K. Meyer and C. J. Nachtsheim. “The Coordinate-Exchange Algorithm for Constructing Exact Optimal Experimental Designs”. In: Technometrics 37 (1995), p. 60.
[72] Microfluidics for Pharmaceutical Applications: From Nano/micro Systems Fabrication to Controlled Drug Delivery. Micro and Nano Technologies. 2019, pp. 79–100.
[73] P. W. Miller et al. “A Microfluidic Approach to the Rapid Screening of PalladiumCatalysed Aminocarbonylation Reactions”. In: Adv. Synth. Catal. 351 (2009), p. 3260.
[74] L. Mora-Tamez et al. “Controlled Design of Phase- and Size-Tunable Monodisperse
Ni2P Nanoparticles in a Phosphonium-Based Ionic Liquid through Response Surface
Methodology”. In: Chem. Mater. 31 (2019), p. 1552.
117

[75] S. Mourdikoudis and L. M. Liz-Marzán. “Oleylamine in Nanoparticle Synthesis”. In:
Chem. Mater. 25 (2013), p. 1465.
[76] K. Okitsu. “UV-Vis Spectroscopy for Characterization of Metal Nanoparticles Formed
from Reduction of Metal Ions During Ultrasonic Irradiation”. In: UV-VIS and Photoluminescence Spectroscopy for Nanomaterials Characterization. Berlin, Heidelberg:
Springer, 2013. doi: 10.1007/978-3-642-27594-4_4.
[77] D. Paramelle et al. “A rapid method to estimate the concentration of citrate capped
silver nanoparticles from UV-visible light spectra”. In: Analyst 139.19 (2014), pp. 4855–
4861.
[78] Alberto Parola, Roberto Piazza, and Vittorio Degiorgio. “Optical extinction, refractive index, and multiple scattering for suspensions of interacting colloidal particles”.
In: J. Chem. Phys. 141.12 (2014), p. 124902. doi: 10.1063/1.4895961. url: https:
//doi-org.libproxy2.usc.edu/10.1063/1.4895961.
[79] Limny Esther Pérez-Jiménez et al. “Enhancement of optoelectronic properties of TiO2
films containing Pt nanoparticles”. In: Results in Physics 12 (2019), pp. 1680–1685.
issn: 2211-3797. doi: https : / / doi . org / 10 . 1016 / j . rinp . 2019 . 01 . 046. url:
https://www.sciencedirect.com/science/article/pii/S2211379718331802.
[80] C. Phamhuu, M. J. Ledoux, and J. Guille. “Reactions of 2- and 3-Methylpentane,
Methylcyclopentane, Cyclopentane, and Cyclohexane on Activated Mo2C”. In: J.
Catal. 143 (1993), p. 249.
[81] Matthew B. Plutschack et al. “The Hitchhiker’s Guide to Flow Chemistry”. In: Chemical Reviews 117.18 (2017), pp. 11796–11893.
[82] G. S. Ranhotra, A. T. Bell, and J. A. Reimer. “Catalysis over Molybdenum Carbides
and Nitrides: II. Studies of CO Hydrogenation and C2H6 Hydrogenolysis”. In: J.
Catal. 108 (1987), p. 40.
118

[83] G. S. Ranhotra et al. “Catalysis over Molybdenum Carbides and Nitrides: I Catalyst
Characterization”. In: J. Catal. 108 (1987), p. 24.
[84] F. H. Ribeiro et al. “Catalytic Reactions of N-Alkanes on -W2C and WC: The Effect
of Surface Oxygen on Reaction Pathways”. In: J. Catal. 130 (1991), p. 498.
[85] F. H. Ribeiro et al. “Reactions of Neopentane, Methylcyclohexane, and 3,3-Dimethylpentane
on Tungsten Carbides: The Effect of Surface Oxygen on Reaction Pathways”. In: J.
Catal. 130 (1991), p. 86.
[86] C. T. Riche et al. “Flow Invariant Droplet Formation for Stable Parallel Microreactors”. In: Nat. Commun. 7 (2016), p. 10780. doi: 10.1038/ncomms10780.
[87] C. T. Riche et al. “Flow Invariant Droplet Formation for Stable Parallel Microreactors”. In: Nat. Commun. 7 (2016), p. 10780.
[88] C. T. Riche et al. “Flow invariant droplet formation for stable parallel microreactors”.
In: Nature Communications 7 (2016), p. 10780.
[89] Benjamin A Rizkin, Filip G Popovic, and Ryan L Hartman. “Review Article: Spectroscopic microreactors for heterogeneous catalysis”. In: Journal of Vacuum Science
& Technology A 37.050801 (2019).
[90] E. J. Roberts et al. “High-Throughput Continuous Flow Synthesis of Nickel Nanoparticles for the Catalytic Hydrodeoxygenation of Guaiacol”. In: ACS Sustainable Chem.
Eng. 5 (2017), pp. 632–639. doi: 10.1021/acssuschemeng.6b01888.
[91] E. J. Roberts et al. “High-Throughput Continuous Flow Synthesis of Nickel Nanoparticles for the Catalytic Hydrodeoxygenation of Guaiacol”. In: ACS Sustainable Chem.
Eng. 5 (2017), p. 632.
[92] E. J. Roberts et al. “Continuous Flow Methods of Fabricating Catalytically Active
Metal Nanoparticles”. In: ACS Appl. Mater. Interfaces 11 (2019), p. 27479.
119

[93] Emily J. Roberts et al. In: ACS Applied Materials & Interfaces 11.31 (2019), pp. 27479–
27505.
[94] Emily J. Roberts et al. “Title of the Article”. In: ACS Applied Materials Interfaces
11.31 (2019), pp. 27479–2750.
[95] C. S. Santos et al. “In operando EPR study for real-time monitoring of redox processes
in ionic liquid media during electrochemical cycling”. In: Scientific Reports 8.1 (2018),
p. 12413. doi: 10.1038/s41598- 018- 30392- y. url: https://www.nature.com/
articles/s41598-018-30392-y.
[96] J. A. Schaidle et al. “Transitioning Rationally Designed Catalytic Materials to Real
“Working” Catalysts Produced at Commercial Scale: Nanoparticle Materials. In”. In:
Catalysis (2017), p. 213.
[97] A. M. Schweidtmann et al. “Machine Learning Meets Continuous Flow Chemistry:
Automated Optimization towards the Pareto Front of Multiple Objectives”. In: Chem.
Eng. J. 352 (2018), p. 277.
[98] B. J. Shields et al. “Bayesian Reaction Optimization as a Tool for Chemical Synthesis”.
In: Nature 590 (2021), p. 89.
[99] H. Shou and R. J. Davis. “Multi-Product Steady-State Isotopic Transient Kinetic
Analysis of CO Hydrogenation over Supported Molybdenum Carbide”. In: J. Catal.
306 (2013), p. 91.
[100] H. Song, J. D. Tice, and R. F. Ismagilov. “A Microfluidic System for Controlling
Reaction Networks in Time”. In: Angew. Chem., Int. Ed. 42 (2003), p. 768.
[101] Eric A. Stach. “Real-time observations with electron microscopy”. In: Materials Today
11 (2008), pp. 50–58. issn: 1369-7021. doi: https://doi.org/10.1016/S1369-
7021(09)70007- 0. url: https://www.sciencedirect.com/science/article/
pii/S1369702109700070.
120

[102] Abraham D Stroock et al. “Chaotic mixer for microchannels”. In: Science 295.5555
(2002), pp. 647–651. doi: 10.1126/science.1066238. url: https://pubmed.ncbi.
nlm.nih.gov/11809963/.
[103] Abraham D Stroock et al. “Chaotic mixer for microchannels”. In: Science 295.5555
(2002), pp. 647–651. doi: 10.1126/science.1066238.
[104] J. Sui et al. “Continuous Synthesis of Nanocrystals via Flow Chemistry Technology”.
In: Small 16 (2020), p. 1902828.
[105] M. M. Sullivan, C.-J. Chen, and A. Bhan. “Catalytic Deoxygenation on Transition
Metal Carbide Catalysts”. In: Catal. Sci. Technol. 6 (2016), p. 602.
[106] S. C. Terry, J. H. Jerman, and J. B. Angell. “A gas chromatographic air analyzer fabricated on a silicon wafer”. In: IEEE Trans. Electron Devices 26.12 (1979), pp. 1880–
1886.
[107] Nguyen T. K. Thanh, N. Maclean, and S. Mahiddine. “Mechanisms of Nucleation and
Growth of Nanoparticles in Solution”. In: Chemical Reviews 114.15 (2014), pp. 7610–
7630. doi: 10.1021/cr400544s.
[108] R.M. Tilaki, A. Iraji zad, and S.M. Mahdavi. “Size, composition and optical properties
of copper nanoparticles prepared by laser ablation in liquids”. In: Appl. Phys. A 88.2
(2007), pp. 415–419.
[109] Pooyan Tirandazi and Carlos H. Hidrovo. In: J. Micromech. Microeng. 27 (2017),
p. 075020.
[110] L. Vera Candioti et al. “Experimental Design and Multiple Response Optimization.
Using the Desirability Function in Analytical Methods Development”. In: Talanta 124
(2014), p. 123.
[111] D.-V. N. Vo et al. “Fischer–Tropsch Synthesis: Effect of Promoter Type on AluminaSupported Mo Carbide Catalysts”. In: Catal. Today 175 (2011), p. 450.
121

[112] D.-V. N. Vo et al. “Non-Linear ASF Product Distribution over Alkaline-Earth Promoted Molybdenum Carbide Catalysts for Hydrocarbon Synthesis”. In: Catal. Today
214 (2013), p. 42.
[113] L. Volpe and M. Boudart. “Compounds of Molybdenum and Tungsten with High
Specific Surface Area: I Nitrides”. In: J. Solid State Chem. 59 (1985), p. 332.
[114] Lu Wang et al. “Self-optimizing parallel millifluidic reactor for scaling nanoparticle synthesis”. In: Chem. Commun. 56 (26 2020), pp. 3745–3748. doi: 10 . 1039 /
D0CC00064G. url: http://dx.doi.org/10.1039/D0CC00064G.
[115] S. A. Weissman and N. G. Anderson. “Design of Experiments (DoE) and Process
Optimization. A Review of Recent Publications”. In: Org. Process Res. Dev. 19 (2015),
p. 1605.
[116] T. Welton. “Room-temperature ionic liquids. Solvents for synthesis and catalysis”. In:
Chemical Reviews 99 (Aug. 1999), pp. 2071–2083.
[117] G. Whitesides. “The origins and the future of microfluidics”. In: Nature 442 (2006),
pp. 368–373.
[118] E. M. Williamson et al. “Statistical Multiobjective Optimization of Thiospinel CoNi2S4
Nanocrystal Synthesis via Design of Experiments”. In: ACS Nano 15 (2021), p. 9422.
[119] Sung-Yi Yang et al. “Size-controlled synthesis of gold nanoparticles using a micromixing system”. In: Microfluidics and Nanofluidics 8.3 (2010), pp. 303–311. doi:
10.1007/s10404-009-0461-2.
[120] B. K. H. Yen et al. “A Microfabricated Gas–Liquid Segmented Flow Reactor for HighTemperature Synthesis: The Case of CdSe Quantum Dots”. In: Angew. Chem., Int.
Ed. 44 (2005), p. 5447.
[121] Andrea Zanella and Andrea Biral. “MICROFLUIDIC NETWORKING: MODELLING
AND ANALYSIS”. In: (2012). url: https://api.semanticscholar.org/CorpusID:
14474657.
122

[122] L. Zhang and Y. Xia. “Scaling up the Production of Colloidal Nanocrystals: Should
We Increase or Decrease the Reaction Volume?” In: Adv. Mater. 26 (2014), p. 2600.
[123] H. Zheng, Y. S. Meng, and Y. Zhu. “Frontiers of in situ electron microscopy”. In:
MRS Bulletin 40 (2015), pp. 12–18. doi: 10 . 1557 / mrs . 2014 . 305. url: https :
//doi.org/10.1557/mrs.2014.305.
123

Abstract (if available)

Abstract The dissertation presented herein is structured into chapters that delve into various research domains within milli- and microfluidic systems. Part of this dissertation includes collaborative authorship. Chapter 1 introduces the fundamentals of fluid mechanics. In this chapter, some highlights of the important physical phenomena that are dominant in milli- and microscale flow systems are presented, focusing on flow dynamics, diffusion, and computational fluid dynamics simulations. It emphasizes the importance of fluid behavior in microscale systems and introduces a case study on microfluidics applications in biomolecular systems in which a portion of a manuscript I participated in as a third author is presented. Chapter 2 covers applications of continuous flow synthesis of colloidal nanoparticles using milli- and microfluidics systems, highlighting the advantages of miniaturized systems in reaction based nanoparticle syntheses. Chapter 3 is adapted from a published manuscript in which I am a joint primary author. Chapter 3 describes the use of continuous flow methods for screening the reaction parameters of catalytically active molybdenum carbide nanoparticle synthesis with an emphasis on throughput optimization using a Design of Experiment approach. Chapter 4 introduces machine learning-assisted spectrophotometry, showcasing the integration of machine learning algorithms for the kinetic analysis of ionic liquid-based platinum nanoparticle synthesis. Chapter 5 introduces in-situ characterization for continuous flow reactors with a particular objective of studying the nucleation and growth kinetics of nanoparticle synthesis using X-ray scattering. This chapter provides a critical evaluation of flow reactor designs for in situ X-ray scattering analysis, focusing on the synthesis of ionic liquid-based Pt nanoparticles

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Synthesis of high-quality nanoparticles using microfluidic platforms

PDF

High-throughput nanoparticle fabrication and nano-biomembrane interactions

PDF

Sustainable continuous flow syntheses of colloidal inorganic nanoparticle catalysts

PDF

Continuous flow synthesis of catalysts with custom made reactor with flow and temperature studies

PDF

Sustainable methods for manufacturing catalytically active nanoparticles

PDF

Cooperative effect of ionic liquid solvents in microfluidic devices for fabricating monodisperse metal nanoparticles

PDF

Intricate microfluidic devices for biopharmaceutical processes: forging ahead with additive manufacturing

PDF

Expanded functionality and scalability of modular fluidic and instrumentation components

PDF

Catalytic methane ignition over freely-suspended palladium nanoparticles

PDF

Development of a novel heterogeneous flow reactor: soot formation and nanoparticle catalysis

PDF

A modular microscale laboratory

PDF

Methanol synthesis in the membrane reactor

PDF

Fabrication of polymer films on liquid substrates via initiated chemical vapor deposition: controlling morphology and composition

PDF

Deep learning architectures for characterization and forecasting of fluid flow in subsurface systems

PDF

Methanol synthesis in a membrane reactor

PDF

Methanol synthesis in a contactor-type membrane reactor

PDF

Deep learning for characterization and prediction of complex fluid flow systems

PDF

Rapid microfluidic prototyping for biological applications via stereolithographic 3-D printing: a journey

PDF

Towards learning generalization

PDF

The patterning of polymer thin films on porous substrates via initiated chemical vapor deposition

Asset Metadata

Creator Madani, Majed Sameer (author)

Core Title Applications of continuous flow reactors towards screening catalytically active nanoparticle syntheses

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Chemical Engineering

Degree Conferral Date 2023-12

Publication Date 12/15/2023

Defense Date 12/05/2023

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag continuous flow,machine learning,microfluidics,millifluidics,nanoparticle synthesis,OAI-PMH Harvest

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Malmstadt, Noah (committee chair), Brutchey, Richard (committee member), Lee, Jay (committee member)

Creator Email msmadani@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113797220

Unique identifier UC113797220

Identifier etd-MadaniMaje-12563.pdf (filename)

Legacy Identifier etd-MadaniMaje-12563

Document Type Dissertation

Format theses (aat)

Rights Madani, Majed Sameer

Internet Media Type application/pdf

Type texts

Source 20231218-usctheses-batch-1116 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu