0000-thesis.Rnw

%%%%%%%%
% To compile the thesis:
% $ Rscript -e "library(knitr); library(methods); knit('thesis.Rnw')"; latexmk -pdf -bibtex thesis
% Note that latexmk settings are set in the ./.latexmkrc file
%%%%%%%%

\RequirePackage{luatex85}
\PassOptionsToPackage{final}{graphicx}
\PassOptionsToPackage{final}{hyperref}

% For instructions on how to switch thesis to final mode, see \input{README.md}.
% Please avoid using absolute paths in either LaTeX or R code
% (in LaTeX code, use paths relative to the thesis root, and in R code please use the here::here() function).

\documentclass[webedition,openright,titles,swedish,english]{LuaUUThesis}
% printedition: configure thesis for physical printing (webedition by default)
% openright: force chapters to start on right-hand page (important for print edition)
% titles: show current chapter and section in top outer corner of each page (like a proper book)
% note that cleveref language switching does not work when languages are specified
% as babel package options, they need to be here. The main language should be loaded last.
% https://tex.stackexchange.com/a/20991/10824
% https://tex.stackexchange.com/questions/248341/how-to-make-a-custom-crefformat-dependent-on-language-chosen-with-babel/252847

%\input{assets/preamble/preamble.Rnw}
<<child=here::here('assets/preamble/preamble.Rnw'), label='preamble-LaTeX', cache=FALSE>>=
@

% For "LaTeX Workshop" (VSCodium extension) to do its autocomplete magic,
% our bib-files must be included here, in the main Rnw
% That also means \usepackage{biblatex} must be included in tikzLualatexPackages
\addbibresource{assets/references/library.bib}

\begin{document}
% \VerbatimFootnotes activates fancyvrb
\VerbatimFootnotes

% ----------------------------------------------------------------------
% R SETUP
% ----------------------------------------------------------------------
%\input{assets/preamble/R.Rnw}
<<child=here::here('assets/preamble/R.Rnw'), label='preamble-R', cache=FALSE>>=
@
% ----------------------------------------------------------------------


%%% FRONTMATTER
% includes abstract, dedication, list of papers, list of contents, other TOCs, glossaries
%\input{assets/components/frontmatter.Rnw}
<<child=here::here('assets/components/frontmatter.Rnw'), label='latex-frontmatter'>>=
@



%%% MAIN MATTER
\mainmatter


%%% CHAPTER:
%%% INTRODUCTION
\chapter{Introduction}
\label{ch:introduction}

The Sun is the primary source of energy for all living organisms,
and together with the cycle of day and night it is
the motive force powering the planet \cite{Schneider2005}.
Humans have always organized their lives and their settlements around the sun,
to draw warmth from it or to seek shade from it, since the earliest civilizations
\cite{Perlin2013}.
The modern world has, over the last couple of centuries, slowly turned to fossil fuels
for both domestic and industrial uses, and as we discovered more and more coal, gas,
and oil we have allowed our industry, agriculture, transportation, and
domiciles to become almost entirely dependent on them.
The resultant soot and pollution were obvious to anyone, but the effect of
\carbondiox\ on the climate perhaps less so. Still, the first scientific
reports warning about its heating effect on the atmosphere were
published so long ago \cite{Arrhenius1896,Tyndall1861,Jackson2019}
that they predate the discovery of the electron \cite{Thomson1897}.
By the end of the 1950s, anyone who read a mainstream newspaper
would have been aware of the basic idea of climate change due to anthropogenic
combustion of fossil fuels \cite{Hudson2023}.

Fossil fuels are the remains of ancient flora and fauna, compressed deep inside the Earth
for tens of millions of years. Importantly, these hydrocarbons
have been removed from the biosphere for just as long.
And we have been reintroducing them at a mind-boggling rate:
in a single year, we burn \num{55} trillion tonnes (that's \qty{55e15}{\kg}) of
this ancient carbon, which is by mass a hundred times more than everything
alive today (if we only count the carbon atoms).
Perhaps even more worryingly, over half of all the anthropogenic \gls{GHG} deposited
into the atmosphere has occurred since 1990, and all around the world we are still
prospecting for new oil and gas fields.

Due to the urgency of transitioning the energy system to a renewable basis,
scientific and technological advancements cannot solve it alone.
To successfully transition the energy system and industry to the power of sunlight,
economic incentives and policy must also align with this aim \cite{Nordhaus1972,Nordhaus1994,IRENA2019}.
And our societies must realize the need for an energy diet that respects
the physical limits of our biosphere \cite{Persson2022,Meadows1972}.




%%%% SECTION
\section{A plentiful source of exergy}
\label{intro:solar-spectrum}


The Sun is and always has been the main source of exergy in the solar system.
From it light radiates in all directions (of the Sun's total radiance about
two parts out of \num{100000} impinges Earth).
Light is nature's way of transferring energy through space.%
%\input{assets/aux/chunks/luminosity.Rnw}
<<child=here::here('assets/aux/chunks/luminosity.Rnw')>>=
@
The day-side of Earth continuously receives
\qty{\Sexpr{formatC(1e-12 * luminosity.hemispherical, format="fg", digits=3)}}{\tera\watt}
of sunlight (just outside the atmosphere), which after correcting
for Earth's albedo and the fraction irradiance absorbed by the atmosphere
(\Sexpr{earth.albedo} and \Sexpr{atmospheric.absorption}, respectively)
yields a solar power of
\qty{\Sexpr{formatC(1e-12 * luminosity.sealevel, format="fg", digits=3)}}{\tera\watt}.%
\seemore{\cref{photoec}}%
This is instantaneous \emph{power}, and summing over the number of hours in a year
yields an annual solar energy input of
\qty{\Sexpr{formatC(luminosity.sealevel * 8766 * 3600 * 1e-18, format="fg", digits=7)}}{\exa\joule},
which is around \num{10000} times greater than the world's annual energy consumption,%
\footnote{%
   Which was \qty{178899}{\TWh} or \qty{644}{\exa\joule} in 2022
   (equivalent to 280 million barrels of oil per day) \cite{Wolfson2018}.
}
or in other words: in one hour the planet receives more solar energy than humanity
expends in a whole year.
Next to solar power all other energy forms pale in comparison.

As far as we are concerned, sunlight is infinite, and only our imagination and
technology limits what we can use it for.
This is where modern chemistry and engineering comes into the fore.

%\input{assets/aux/fig/fig_0100-solar-spectrum-decorated.Rnw}
<<child=here::here('assets/aux/fig/fig_0100-solar-spectrum-decorated.Rnw'), label='0100-solar-spectrum-decorated'>>=
# \label{fig:solarspectrum-photoec}
# \label{tab:solarspectrum-amounts}
@

In this thesis, I will make the case for a more widespread use of photocatalysis
for production of solar fuels and for the proliferation of photocatalysts for
the degradation of environmental pollutants by exploiting low-dimensional
\ZnO\ alone and in combination with \ch{CdS} and \ch{Fe2O3}.




%%%% SECTION
\section{The energy transition and resource scarcity}
\label{intro:energy-transition}

The idea of an energy system based on renewable power is not new at all.
Exactly one hundred years ago, the evolutionary biologist \mbox{J.\ B.\ S.\ Haldane}
laid out a practical vision for the future energy system of England, saying
\cite{Haldane1923,Rajeshwar2008}
\enquote{the country will be covered with rows of metallic windmills}
feeding electricity through the grid to power stations where
\enquote{the surplus power will be used for the electrolytic decomposition of water
into oxygen and hydrogen}, and that the initial costs would be very considerable,
but \enquote{energy will be as cheap in one part of the country as another, so that
industry will be greatly decentralized; and that no smoke or ash will be produced}.
Of course, by that time the United Kingdom had been burning fossil fuels
for almost \qty{150}{\years} already, and people were getting really
worried about the visible smoke and smog engulfing their cities.
But we should note that Haldane envisioned this pollution-free energy system to
happen \emph{four hundred years} into the future, which is quite pessimistic, although
to be fair we have not managed to beat his prediction yet.

A known problem with the energy transition is that all development of renewable
energy systems is currently, and for the foreseeable future, built on logistics
fed with \emph{non}-renewable power \cite{Roos2021}.
Thus, producing the materials needed to provide renewable power
(\gls{PV} panels, wind turbines, electrolysers, \etc)
necessarily equates to emission of \gls{GHG}.
If we expect to operate all other industry as usual, then the energy transition
itself will necessarily produce even more emissions.
The only reasonable way out of this conundrum is to realize that the energy
transition, if managed responsibly, must lead to a sort of self-imposed energy scarcity
(perhaps that is a \enquote{good thing}, considering how wasteful the modern
mass-consumption economy has been).

Equally important is the question of how we will source the chemicals
necessary to build the required renewable power generation infrastructure.
\Cref{fig:0100-elements-of-photocatalysis} considers the availability of the chemical
elements commonly used in \glsdisp{PC}{photocatalytic} compounds.


Water is a critical resource for the energy transition.
Worldwide, we use more than \qty{4e12}{\cubic\metre} of freshwater each year.%
\footnote{Agriculture accounts for \qty{70}{\percent}, industry \qty{19}{\percent} and households \qty{10}{\percent}.}
Yet, three billion people are affected by water scarcity, and two billion of them
lack access to safe drinking water \cite{WWAP2020}.
By 2050, the demand for freshwater is projected to increase by more than \qty{40}{\percent}.\\
A perhaps overlooked aspect of our fossil-based
energy conversion system is that electrical power generation consumes a lot of freshwater.
In fact, of the various electrical power generation methods at our
disposal (coal/gas, nuclear, hydro, solar, wind, \etc) the renewable
generators (except for hydropower) consume hundreds or even thousands of times
\emph{less} freshwater to generate the same amount of electricity \cite{Wilson2012}.%
\footnote{%
   With that in mind, we can see that operating a desalination plant on
   non-renewable generated electricity would actually waste water, or in the best case
   simply represent a transfer of freshwater resources from the desalination plant
   to the power plant powering it.
   Desalination plants should be solar powered. Recent research have shown
   that it is even possible to simultaneously desalinate and produce electricity
   using a composite \glstext{PV}-membrane device \cite{Wang2019}.
}
In many countries, the electrical generation sector consumes \emph{more} freshwater
than their agricultural sector.\\
The electrolysis of water into \hydrogen\ and \oxygen\ (\cref{rxn:water-splitting}),
also called \emph{water splitting}, requires around \qty{9}{\kg}
of \water\ to produce \qty{1}{\kg} of \hydrogen
and is an \enquote{uphill} reaction with a large positive Gibbs free energy,
$\state{G}=\qty{238}{\kJ\per\mol}$ or $\state{G}=\qty{1.23}{\eV}$,
comprising two half reactions, \namely, the \glsfirst{HER} and the \glsfirst{OER}
(here shown under acidic conditions at \gls{STP}):
% otherwise pagebreak inside the reactions block (too bad we couldn't make subreactions immune to pagebreaks)
\clearpage
% Be warned - vertical and horizontal spaces are held together by duct-tape here
\begin{subreactions}\begin{reactions}%
4 \proton{} + 4 \electron{} &-> 2 \hydrogen{} && $\qquad\qquad\qquad\qquad$ \AddRxnDesc{Hydrogen~evolution} "\label{rxn:hydrogen-evolution}" \\%
2 \water{} + 4 \hole{} &-> \oxygen{} + 4 \proton{} && $\qquad\qquad\qquad\qquad$ \AddRxnDesc{Oxygen~evolution} "\label{rxn:oxygen-evolution}"%
\end{reactions}\end{subreactions}%
\addtocounter{reaction}{-1}%
\vspace{-\baselineskip}%
% note that the following lengths must be adjusted if the horizontal
% extent of any reactions are changed!
\hspace{13mm}%
\begin{minipage}{59mm}%
   \vspace{-\baselineskip}%
   %\rule disappears inside minipage, but \hrulefill is visible, weird
   \hrulefill%
\end{minipage}%
\begin{reaction}%
\qquad{}\qquad{}\quad{} 2 \water\lqd{} -> 2 \hydrogen\gas{} + \oxygen\gas{} $\quad\enspace\enspace E^0=\qty{1.23}{\voltNHE}$ \AddRxnDesc{Overall~water~splitting} "\label{rxn:water-splitting}"%
\end{reaction}%
But in practice, an electrolyser splitting water may require up to ten times more water
than stoichiometry demands.%
\footnote{%
   Taking into account water treatment to meet the electrolyser's purity requirements,
   water for cooling the electrolyser, and water for cooling the produced \hydrogen.}
Even when accounting for this, the water resources required for a renewably based
energy system are significantly lower than for a fossil-based system.


Even if we accept the notion that \emph{total} primary energy production must
decrease to achieve the energy transition, the inherent distributed nature
of renewable power generation \cite{Bollen2011} might actually mean that a larger
number of people get access to more electrical power than today.
There is no inherent benefit nor any technical reason to centralize solar power generation;
its inherently distributed nature makes it a more resilient power generation system.


Photoelectrochemistry lies at the cross-roads between energy and water, two
resources that will always be vital to society, but perhaps now more than ever.
It is our responsibility as a field to advice on how to produce sustainably sourced
photocatalysts for solar fuels production or for the photodegradation of pollutants.
I believe our aim must be to convert this carbon society into a photon society.




%%%% SECTION
\section{Photocatalysis with inorganic semiconductors}
\label{intro:photocatalysis}

Over \qty{90}{\percent} of all chemicals and fuels are estimated to be produced
with the aid of some catalyst \cite{NilssonPingel2018}, and catalysts contribute directly
to around a third of the world's economy \cite{Ma2005}.
Most commercial catalytic processes are heterogeneous, and commonly employ disperse
particles on high-surface area supports.

\glsdisp{PC}{Photocatalysis (PC)} is the acceleration of a photoreaction
by the presence of a catalyst \cite{Mills1997,Suppan1994}.%
\footnote{%
   Note that the light is \emph{not} the catalyst, but rather a reactant.
   The source of the light can be anything, really, but most commonly
   natural sunlight. There is, as we have seen, a lot of it, and the
   price cannot be beat.
   It could also be simulated sunlight, or a lamp, as long as it emits
   photons with an energy above the \glsdisp{band_gap}{band gap} of
   the \glsdisp{PC}{photocatalyst}.}
A benefit of \gls{PC} is that it may (and indeed often does) rely
on atmospheric oxygen gas as the oxidant and that it can be carried out at
\gls{STP} \cite{Akyol2004}.
Photocatalysis can be employed in quite varied settings, in this thesis
we will primarily concern ourselves with the production of
energy-rich chemicals (\eg, solar fuels such as \hydrogen) and the degradation of
harmful organic compounds in water (water purification).

Photocatalysts are an important complement to \gls{PV} cells in the renewable
energy transition. Whereas \gls{PV} generates electrons in an external circuit,
\gls{PC} utilizes locally photogenerated charge carriers that can induce a chemical
redox reaction.
Many cheap, non-toxic and commonly available semiconductor materials are also
quite effective photocatalysts (\cf\ \cref{fig:0100-elements-of-photocatalysis}).

Heterogeneous catalysis is when the catalyst and catalysed species
are in different phases. In this context, heterogeneous photocatalysis is the process
of inducing chemical reactions using an irradiated photocatalyst's surface, \eg,
inorganic semiconductor surfaces \cite{Rajeshwar2015}, usually while immersed in
aqueous solution.
At the fundamental level an irradiated semiconductor nanoparticle can be considered
an electrochemical cell performing work, whose anodic and cathodic photoreactions
are in steady-state conditions. The photocurrent and the internal photovoltage
give the electric power of the cell, and the particle/cell achieves an
electrochemical potential that depends on the relative rate of its two half-cell
reactions \cite{Rajeshwar2015}.

%\input{assets/aux/fig/fig_0100-elements-of-photocatalysis.Rnw}
<<child=here::here('assets/aux/fig/fig_0100-elements-of-photocatalysis.Rnw')>>=
# \label{fig:0100-elements-of-photocatalysis}
@

But despite \ch{TiO2} having the honour of practically defining the field of
heterogeneous inorganic photocatalysis, the study of \ZnO\ as a \gls{PC} actually
predates \ch{TiO2}.
The earliest description of photocatalysis using \zincox\ dates back to 1911
in experiments by Alexander Eibner bleaching Prussian blue using
illuminated \ZnO\ \cite{vonEibner1911}, and its earliest use for
the photolysis of water is from 1953 \cite{Markham1953,Markham1955}.
In contrast to anatase \ch{TiO2} with its indirect band gap,
wurtzite \zincox\ has a direct band gap and thus higher absorption coefficient,
and similar positions of the conduction and valence band edges
(\qty{-0.5}{\voltNHE} and \qty{+2.8}{\voltNHE} at \pH{\,\num{5}}),
and have the advantage of a significantly higher charge carrier mobility
\cite{Boschloo2006,Jacobsson2012a}.

%\input{assets/aux/fig/fig_0100-bandgaps-all.Rnw}
<<child=here::here('assets/aux/fig/fig_0100-bandgaps-all.Rnw')>>=
# \label{fig:0100-bandgaps-all}
@



In metal oxides, the \gls{VBE} is determined primarily by the orbitals on the
\oxide{} anion, and the \gls{CBE} by metal cations and their bonding geometry.

The band gap of a semiconductor can be tuned using different methods:
\via\ the quantum size effect, or phase transformations, or solid solutions
with other elements.
In this thesis we have explored the quantum size effect on the band gap of \ZnO\
(\cf\ \cref{ch:quantum-confinement}).

Perhaps it would not be a mischaracterization to say that
the ultimate goal of photocatalysis is to design a \gls{PC} made out of
cheap and non-toxic materials with excellent stability in water under irradiation;
with a band gap $\geq\qty{1.6}{\eV}$ whose band edges straddle the \water\ redox
potentials;
and that performs the photocatalysed reaction at a high rate and for a long time.
Although the search for that holy graal will surely continue,
in the present the pressing need for quickly implementable solar-based
energy generation suggests that we should make do with what we have.
And what we have is very cheap \gls{PV} panels and an increasing
number of electrolysers.






%%%% SECTION
\section{The race to the bottom: benefits and risks}
\label{intro:nanoparticles-history-risks}

Many physical and chemical properties of a material change when
the constituent particles become sufficiently small.
A particle with a diameter between \qtyrange{10}{100}{\nm} is said to be
\enquote{nano}-sized, while larger particles than that are commonly referred
to as sub-micron or micron-sized.
But how small is a nanoparticle?
Getting a grip on that is mind-bendingly hard.
In \cref{fig:0100-nanoparticles-are-small} we try to put it into some perspective.

Of course, there is a limit to how small stuff we can build, and that is set by the size
of the atoms themselves. An \oxide{} ion has a diameter of \qty{0.28}{\nm},
and a \ch{Zn^{2+}} ion has a diameter of \qty{0.15}{\nm} (based on ionic crystal radii).

%\input{assets/aux/fig/fig_0100-nanoparticles-are-small.Rnw}
<<child=here::here('assets/aux/fig/fig_0100-nanoparticles-are-small.Rnw')>>=
# \label{fig:0100-nanoparticles-are-small}
@

If particles are made into nanometre dimensions, the most noticeable change is
the large increase in specific surface area (area per weight of the material),
and thus the potential to substantially increase chemical activity at the exposed surfaces.
The size of the material also changes how light interacts with it.
If a nanoparticle is made small enough (depending on its composition)
something rather special happens: its electronic states are literally
\emph{squeezed} by the constricted dimensions of the nanoparticle, which
results in drastic changes in the material's electronic and optical
properties.
This phenomenon is termed \emph{quantum confinement} and is elaborated
in \cref{ch:quantum-confinement}.

Although humans have benefitted from the use of nanoparticles or nanostructured materials
for millennia \cite{Montanarella2022} (often unwittingly), we are by no means
alone in doing so.
Many birds have iridescent plumage and many insect exoskeletons have intricate
nanostructured surfaces that create pigment-less (structural) colouration by reflecting
incident light in very specific ways \cite{Wolpert2009,Dunning2023}.
In terms of sophistication, engineered nanostructures cannot yet compete
with their counterparts in nature.
The study of such nanostructures is a major interest for applications towards
improving the efficiency of \gls{PV} panels or in technological applications such
as waveguides towards light-based computing.


Modern scientific investigations into nanoparticles were first undertaken
in the late \century{19} century by Scottish scientist John Aitken when he
took an interest in something as mundane as dust.
At the time, it was obvious that the transparency
of the air correlated to the amount of dust (and to the relative humidity).
Aitken was the first that set out to measure these effects, and created a method
for measuring these suspended dust particles in air \cite{Aitken1888},
then miniaturized the device and promptly set about travelling
across Europe to collect measurements (\enquote{the portable apparatus is reduced
to such dimensions that it adds but little to one's personal luggage}) \cite{Aitken1891}.
Aitken's device was the first example of a condensation particle counter,
which works by introducing airborne particles into a supersaturated vapour,
where the particles act as nucleation points for the vapour to form droplets
that are then counted using optical techniques.
Using this technique one can detect the presence of particles down to
a few nanometres in size.
This was the first demonstration of a characterization technique
to quantify nanoparticles (particles below \qty{50}{\nm} are generally
undetectable with light-scattering optical particle counters), but less than
\qty{50}{\years} later it was supplanted by the electron microscope.

Nanotechnology is the latest in a long line of technologies that promises
better living through chemistry \cite{Schummer2008}.
But it would behoove us to root our expectations in history, which can
be an unforgiving teacher if forgotten. Only during the \century{20}, chemical
industry managed to spread DDT over every continent, freon into the stratosphere,
and lead in every city by the combustion of leaded petrol.
To be fair, the chemical industry has also been the source of some
of our modern world's greatest successes, such as the catalytic Haber--Bosch
process that has fed the world, vaccines that have wiped out polio and many other
debilitating diseases, and the production of an amazing range of materials that
have made our lives healthier and safer.
Perhaps it is no wonder that we believe the nano-fication of chemistry
will lead to similar rapid advances.

I guess what I am trying to convey is that we must make ourselves aware of the
risks, and to do that some creativity and humility is called for.
The risks of nanosized materials may not always be the same as those experienced
by earlier generations of chemists, and widely spreading a technology that
\enquote{works in the lab} comes with its own set of new risks.
Chemists as well as chemical industry regulators (especially in the
EU and USA) have a good grasp of the risks with most industrially manufactured
chemicals.
The main risks we need to consider are those towards human health and towards the environment.
This is something that is still being actively researched
\cite{Mitchell2019,Kim2016a,Reed2012,Maynard2011}, and for that reason those working
in the field would do well to be cautious and avoid hyperbole when expressing the
benefits of nanotechnology.
When it comes to risk management, the hardest challenge might be to figure out
what exactly we should be measuring in the first place; what material
and particle attributes are most closely associated with health risks, and at what
exposure levels? \cite{Maynard2015}





%%%% SECTION
\section{Will chemistry come to the rescue?}
\label{intro:chemistry-rescue}

There is no denying that many of the technological and environmental challenges
ahead of us are well-suited for chemical solutions.
Chemistry has developed technologies and materials that can reduce pollution,
improve waste management, and clean up contaminated environments.
For example, catalytic converters in cars use chemical reactions to convert harmful
exhaust gases into less harmful compounds before they are released into the air.
Chemistry has also helped us develop materials such as activated carbon and zeolites that
can absorb and remove pollutants from air and water.
Additionally, chemistry has played a crucial role in developing principles of
\enquote{green chemistry}, that aims to design chemical products and processes
that are safe, sustainable, and minimize waste.

A not yet realized application for photocatalytic degradation lies
in the complete destruction of synthetic chemicals from various waste streams.
Synthetic chemicals of all kinds are produced in increasing quantities all over
the world, and pollution by their release into the environment is growing \cite{Bernhardt2017}.%
In particular, pharmaceuticals are dispersed in the environment \via\
water treatment plants that are presently ill-equipped to remove such compounds
from the effluent, but worse than that, the pharmaceutical production plants
themselves are a major source of local pollution and are
frequently located in countries with lax environmental regulations.
The Swedish government agency Naturvårdsverket published a report in 2017 \cite{Naturvardsverket2017}
on advanced cleaning methods of effluent from pharmaceutical residues.
The report mentions \gls{AOP} as a technology not yet in production, but showing
promise for certain use cases, such as effluent with higher concentrations of pollutants.
Ozonation has been in use in one treatment plant in Linköping since 2015.

Semiconductors are an excellent choice for photoactive materials, because their
band gaps are well-suited to produce charge separation under solar irradiation.
For a \emph{single} semiconductor material to split water (\cf\ \cref{rxn:water-splitting}),
its band gap must straddle the hydrogen and oxygen evolution reduction potentials
of \water\ (\cf\ \cref{fig:0100-bandgaps-all}) and it must allow efficient transport
of both electrons and holes.

Unfortunately, such a semiconductor does not exist (or has not been found yet, despite
well-organized research into photoelectrochemistry since at least the 1970s \cite{Maeda2011}).
Wide gap semiconductors such as \ch{TiO2} or \ZnO\ have band edges suitable
for total water splitting, but on the other hand their band gaps are so large
that they will only absorb \gls{UV} photons which make up a very small part
of the solar spectrum (\cf\ \cref{tab:solarspectrum-amounts}).



%%%% SECTION
\section{Scope of this thesis}
\label{intro:scope-thesis}

Throughout this thesis we have studied the
application of sunlight towards the synthesis of solar fuels or
the degradation of chemicals
with photocatalysts synthesized using cheap and easily scaled methods.

The work involved laboratory-scale wet-chemical synthesis
of composite nanostructured films and colloidal low-dimensional nanoparticles;
their experimental characterization using a variety of spectroscopic, microscopic,
and diffractive methods;
and proof-of-concept tests of their photocatalytic performance.
Concurrently I got the opportunity to set up a laboratory from scratch (that was really fun!)
and to create a personalized data analysis pipeline with an eye towards
reproducibility and the sharing of code and data.

In the first paper
\seepaper{P1}
we set out to synthesize a large surface area nanodigitated \glsdisp{PC}{photocatalyst}
\enquote{sandwich} of a wide gap semiconductor (\ZnO) coated with \ch{CdS}
for visible-light sensitization, which allowed us to optimize electrodeposition
parameters and study the effect of \ch{CdS} coating thickness \vs\ \gls{PC} performance.
Here the goal was to create a highly photoactive \gls{PC} material with visible-light sensitivity
\via\ an experimentally simple and economically scalable method, which we
achieved by using electrodeposition and chemical-bath deposition (both very
established methods, and relatively simple in operation), although the visible-light
sensitivity came at the cost of introducing the well-known toxicity of cadmium.

In our second paper we
\seepaper{P2}
looked for ways to avoid cadmium as well as making the outer coating
more \pH\ and photoresistant (\ch{CdS} degrades under irradiation unless sacrificial
agents are introduced in the electrolyte solution), and what could be more
durable than literal rust?
But for that to work the well-known extremely poor hole conductivity of \hematite\
required the application of an extremely thin film on the \ZnO\ \glsplural{NR},
which essentially demanded the use of \gls{ALD}.
This allowed us to study how varying the \hematite\ layer thickness affected
the properties of the photoelectrodes.

In the third paper
\seepaper{P3}
we took a step back to study the question of how \ZnO\ nanoparticle
diameter in the quantum size regime influenced the redox potential
of the photogenerated excitons.
We did this by synthesizing colloidal \zincox\ suspensions at varying
average \gls{QD} diameter and measure their \gls{PC} activity under
simulated sunlight and simulators tracking the photodegradation of \gls{MB} dye.
This kind of colloidally suspended \gls{PC} could be used to
combat distributed pollutants using nothing more than natural sunlight and
then naturally precipitate into relatively harmless \ZnO\ sediment, a process
aided by the ultrasmall size and large band gap of the \gls{PC} particles.

In our fourth paper
\seepaper{P4}
we take a deep dive into the vibronic properties of low-dimensional \ZnO\ nanocrystals,
\via\ a series of Raman spectroscopy measurements where we quantify
their phonon--phonon and electron--phonon coupling as a function of particle diameter.


This thesis also includes a chapter discussing the relative ease with which
analytical reproducibility can be achieved with the use of freely available software,
and also briefly presents some scientific software tools
this author has publicly shared during this work.
% data analysis chapter with an eye towards improved reproducibility in science via code and data sharing


\Cref{fig:0100-wordcloud-thesis-corpus} shows a word cloud of keywords;
not compiled ourselves, but rather \emph{generated}
from the corpus of all \emph{titles} of all bibliographic records
\emph{cited in this thesis}.
In a way, it is a \emph{meta}-visualization: the thesis querying itself.

%\input{assets/aux/fig/fig_0100-wordcloud-thesis-corpus.Rnw}
<<child=here::here('assets/aux/fig/fig_0100-wordcloud-thesis-corpus.Rnw'), label='0100-wordcloud-thesis-corpus'>>=
# \label{fig:0100-wordcloud-thesis-corpus}
@






%%% CHAPTER:
%%% SYNTHESIS AND PREPARATION
\chapter{Synthesis and preparation}
\label{ch:synthesis}


In my synthetical work I strove for the simple, as far as possible.
That mindset was based on the idea that if chemistry is to contribute to the
conversion of our energy system at any scale, its proposed solutions must
first and foremost be scaleable.
That same idea was also behind our choice of elements---zinc and iron being
plentiful and very cheap and practically non-toxic.
Granted, we also used cadmium which is far from non-toxic, but on account of
its sulfide being a commonly employed visible-light absorber in solar cells and because
its synthesis method is very simple, it was too tempting a target to overlook.

At an early stage of this work we attempted to synthesize anatase \ch{TiO2}
\glsplural{NR} on \gls{TCO}, but after attempting several approaches
including
electrochemical deposition with templated membranes
\cite{Limmer2001,Limmer2002,Chu2003,Chu2003b,Limmer2003a,
Limmer2004,Chu2004,Limmer2005,Yan2006},
cathodic electrodeposition in acidic baths
\cite{Natarajan1996,Karuppuchamy2001,Chenthamarakshan2002,Karuppuchamy2002,
deTacconi2003,deTacconi2004,deTacconi2005,Karuppuchamy2005,Karuppuchamy2006,
Karuppuchamy2006a,Somasundaram2006,Karuppuchamy2007,Chettah2008,Chigane2009,
Zhou2009,Chen2010h,Elhajj2010},
cathodic electrodeposition in alkaline baths
\cite{Ohya2002,Sawatani2005,Wessels2008,Wessels2008a}, and
anodic electrodeposition
\cite{Flood1995,Lokhande2005,Wessels2006,Tsai2009,Wu2009b,Wessels2010,Wu2011}
without achieving the desired synthetical outcome apart from
impressive clouds of hydrochloric acid (inside the fume hood, mind you) we looked
for an alternative photocatalytic metal oxide that could be electrodeposited
at mild processing conditions.

Zinc oxide is a semiconductor with known experimental methods of cathodic electrodeposition
of thin films or nanostructures in aqueous electrolytes \cite{Peulon1998,Pauporte2009},
which we could easily adapt.
\ZnO\ wurtzite has attractive optical, electrical, and thermal properties, such as
a wide and direct band gap (\qty{3.3}{\eV} in the bulk at \gls{RT}),
a high absorption coefficient (due to its direct band gap),
a large exciton binding energy%
\footnote{%
   The exciton binding energy is the energy needed to separate the electron and hole
   of an exciton into free charge carriers, \ie, the energy difference between the
   fundamental gap and the optical gap.
   It is significantly larger than $kT$ (\qty{26}{\milli\eV} at \gls{RT}), giving
   \ZnO\ exciton emission even above \gls{RT}.
   For comparison, \ch{GaN} has an exciton binding energy of \qtyrange{18}{28}{\meV} \cite{Karpina2004}.
   }
\qty{60}{\milli\eV},
a high carrier mobility $>\qty{100}{\cm\squared\per\volt\per\second}$ \cite{Janotti2009},
and a high thermal conductivity,
which makes \ZnO%
\footnote{%
   On a personal note, zinc was the first element that made a mark
   during my undergraduate education: at one of my cohort's very
   first lectures at the Arrhenius laboratory, our lecturer
   Lars-Johan Norrby proclaimed zinc to be his favourite element on the periodic table!}
considered for applications in \glsdisp{PC}{photocatalysis},
\glsdisp{PV}{solar cells}, \glsplural{LED}, transparent conducting films, and more
and a viable alternative to \ch{TiO2} anatase.

\ZnO\ is amphoteric, reacting with both acids and bases, but it is nearly insoluble
around neutral \pH.%
\footnote{%
   \ZnO\ solubility at \qty{298}{\kelvin} (\qty{25}{\celsius}) is
   \qty{1.23e-5}{\mol} per \qty{100}{\gram} \water\ \cite{Ellis1984},
   or ${\approx}\qty{1.0}{\mg}$ per \qty{100}{\gram}.}
In acids it easily forms salts,
and with increasingly alkaline conditions (starting from around \pH{\,\num{9}})
it is known to form monomeric hydroxide complexes from \ch{ZnOH+} to \ch{[Zn(OH)4]^{2-}}
\cite{Greenwood1997,Ekberg2016}.




\section[Electrodeposition of ZnO NR arrays]{\texorpdfstring{Electrochemical deposition of \ZnO\ NR arrays}{Electrochemical deposition of ZnO NR arrays}}
\label{synthesis:electrodeposition}

Electrochemical deposition is cheap, scalable, and affords
relatively good control of the conditions during synthesis \cite{Schwarzacher2006}.
In addition, semiconductor thin-film electrodeposition is a mature technology,
widely used in many industrial settings, and its basis in solution chemistry
allows a wide range of synthetical parameters.
Perhaps ironically then, the main disadvantages of electrodeposition
are also related to solution chemistry: (relatively) complicated sample handling and residual
electrolyte in the product.

Despite being cheap, electrodeposition allows
shape-controlled synthesis of \ZnO\ \glsplural{NR} \emph{without}
any capping agents or surfactant molecules that preferentially inhibit the
growth of particular crystallographic facets, which was attractive to my
experimental sensibilities.

We created \ZnO\ \glsplural{NR} by cathodic deposition at \qty{-0.7}{\voltSCE}
for \qty{90}{\minute} with an initial pulse at \qty{-1.3}{\voltSCE} with a duration
of \qtylist[list-final-separator={ or }]{0.1;0.5;1.0}{\second}.
The aqueous electrolyte solution was
\reactant[concentration=0.1,concentration-unit=\milli\Molar]{zincchloride}
with \reactant[concentration=0.1,solvent=\water]{potassiumchloride}.
The electrodeposition was done at a temperature of \qty{80}{\celsius} maintained
using a thermostat-controlled water bath.

Our primary electrodeposition setup (see \cref{{fig:0201-electrodeposition-schematic}})
used a CH~Instruments 760C%
\footnote{%
   CH~Instruments 760C was a compact bench-top bipotentiostat/galvanostat
   with a current range from \qty{100}{\micro\ampere} to \qty{1}{\pico\ampere} and a
   voltage range $\pm\qty{10}{\volt}$.
}
bipotentiostat connected to a three-electrode cell with
a CHI111 \ch{Ag}/\ch{AgCl} reference electrode
and a CHI115 \ch{Pt} wire counter electrode.
Our secondary electrodeposition setup (see \cref{fig:0301-ED-cell-autolab})
used either a {\textmu}AUTOLABIII%
\footnote{%
   The {\textmu}AUTOLABIII was a compact bench-top potentiostat with
   a current range from \qty{10}{\milli\ampere} to \qty{10}{\nano\ampere} and a
   voltage range $\pm\qty{5}{\volt}$ with a resolution of \qty{3}{\micro\volt}.
   Controlled via a laptop computer running GPES~v4.9 software.
}
potentiostat or an AUTOLAB PGSTAT302N%
\footnote{%
   The PGSTAT302N was a high-current potentiostat/galvanostat with
   a current range from \qty{1}{\ampere} to \qty{10}{\nA} and a
   voltage range $\pm\qty{10}{\volt}$ with a resolution of \qty{3}{\micro\volt}.
}
connected to a three-electrode cell with an AUTOLAB \gls{RE}
and a coiled \ch{Pt}-wire \gls{CE}.

The \gls{WE} was \gls{FTO} glass, \ie, sodalime%
\footnote{%
   Sodalime glass is a very common type of glass commonly used in containers, plate glass,
   microscope slides, pasteur pipettes and more.
   It does complicate analysis slightly due to containing several different oxides (in order of abundance):
   \ch{SiO2}, \ch{Na2O}, \ch{CaO}, \ch{MgO}, and \ch{Al2O3}.
   Before deposition each cut glass substrate were prepared by ultrasonication
   in deionized \water, followed by an \EtOH\ and then an acetone rinse,
   and finally dried with \ch{N2}.
}
glass coated with a compact
layer of fluorine-doped \tinox (TEC-7).
The TEC-7 substrate has a sheet resistance of \qty{7}{\ohm\per\sq},
is insoluble in dilute acids or alkalis,
and microcrystalline with a highly oriented structure.
Its impedance is low enough to allow direct-current cathodic deposition.
The \gls{WE} was kept at a potential negative to the reduction potential
of \ch{O2\gas}, which causes oxygen gas near the electrode to be reduced to
hydroxide ions (\cref{rxn:oxygen-reduction}). The hydroxide ions react with the
zinc ions and form \ch{Zn(OH)2}, \ch{ZnOOH} or \ZnO\ (\cref{rxn:Zn-OH-O})
depending on the temperature.
% https://tex.stackexchange.com/questions/681457/half-cell-reactions-and-sum-reaction-below-horizontal-rule-with-chemmacros
% \begin{minipage}{\textwidth}%
% surrounding the whole shebang with another minipage caused horizontal width issues
\begin{subreactions}\begin{reactions}
1/2 O2 + \water{} + 2 \electron{} &<=> 2 \hydroxide{} \AddRxnDesc{Oxygen~reduction~in~water} "\label{rxn:oxygen-reduction}" \\
Zn^{2+} + 2 \hydroxide{} &-> \ZnO{} + \water{} \AddRxnDesc{Zinc~hydroxide~to~oxide} "\label{rxn:Zn-OH-O}"
\end{reactions}\end{subreactions}
% reverting the counter so the sum reaction gets the same major number as the half-cell reactions
\addtocounter{reaction}{-1}
\vspace{-\baselineskip}%
% https://tex.stackexchange.com/a/269888/10824
% this must be messiest LaTeX code in the entire thesis, just to achieve this
% ridiculously simple thing (rule extending the width of the reactions)
% I have given up on customising the thickness of the hrule
% https://tex.stackexchange.com/questions/371286/draw-a-horizontal-line-in-latex
% note that these values are hard-coded and must be adjusted if the horizontal
% extent of any reactions are changed
\hspace{14mm}%
\begin{minipage}{75mm}
   \vspace{-\baselineskip}
   %\rule disappears inside minipage, but \hrulefill is visible, weird
   \hrulefill
\end{minipage}
\begin{reaction}
Zn^{2+}\aq{} + 2 \electron{} + 1/2 O2\aq{} -> \ZnO\sld{} \phantom{ZnOZnOZnO}
\AddRxnDesc{Zinc~oxide~electrodeposition} "\label{rxn:ZnO-dep}"
\end{reaction}
%\end{minipage}
% https://stackoverflow.com/questions/2170819/margin-figures-in-latex
\marginpar{\scriptsize\itshape%
   \includegraphics[width=\marginparwidth]{synthesis/electrodeposition/model.pdf}
}
The oxygen was actively supplied to the electrolyte solution by either saturating
the headspace with \ch{O2\gas} flow or by bubbling it into the electrolyte solution.%
\footnote{%
   I found that bubbling \ch{O2\gas} directly into the electrolyte solution
   was better to reach and maintain saturation \ch{[O2\aq]}, but only if
   it could be done without the bubbles disturbing the electrolyte solution
   near either the \gls{WE} or \gls{RE}.
   The alternative was to maintain \oxygen\ overpressure in the headspace,
   which required the gas flow to be higher than the leak-rate out
   of the headspace, and that the electrolyte solution was given enough time
   prior to deposition to become saturated with \oxygen.
}
In any case, the solution was saturated in oxygen \emph{before} the deposition
started, such that the oxygen concentration was kept as-close to constant as possible
during the electrodeposition.
The role of the supplied oxygen gas was to supply a source for the hydroxide ions needed at
the working electrode. But note that this role could have been fulfilled by other means,
for example using hydrogen peroxide, nitrate or other ions such as phosphate,
carbonate, or sulfate.

\seemore{\cref{refelectrodes}}
\seemore{\hlink{https://solarchemist.se/2016/11/24/reference-electrodes}{Blog}}




%\input{assets/aux/fig/fig_0201-electrodeposition-cell-schematic.Rnw}
<<child=here::here('assets/aux/fig/fig_0201-electrodeposition-cell-schematic.Rnw')>>=
# \label{fig:0201-electrodeposition-cell-schematic}  (a)
# \label{fig:0201-electrodeposition-cap-schematic}   (b)
# \label{fig:0201-electrodeposition-WE}              (c)
# \label{fig:0201-ED-CHI}                            (d)
# \label{fig:0201-electrodeposition-schematic}
@

Maintaining a steady temperature of \qty{80}{\celsius} was just as important
as maintaining a steady \ch{O2\aq} concentration.
The target temperature was low enough to allow the use of a water bath,
and for practical reasons we used a hot plate with a built-in magnetic stirrer
which meant that evaporation of the water bath itself needed to
be counter-acted since depositions were \qty{1.5}{\hour} long.
For that purpose I developed a custom-made cover made with small peripheral holes to allow
the temperature probe access to the bath and a large centre-hole for placing the
deposition cell securely (to avoid it bobbing in the water).
The cover was successful in limiting evaporation and kept the
water bath's level constant for over \qty{1.5}{\hour} at \qty{80}{\celsius}.
We also improved on the gas inlet by adapting a long glass pipette for the
purpose (instead of a typical gas needle which is metallic and interfered with
the current inside the cell).

To improve reliability and repeatability of the electrochemical depositions
I created a simple paper-based form to jot down the deposition parameters for each run
(since the potentiostats' software unfortunately offered no way to save such metadata).
These \hlink{https://github.com/solarchemist/electrodeposition-templates}{simple templates}
also allowed us to record non-potentiostat parameters
such as gas flow rate, stirring or not, temperature, \etc.


%\input{assets/aux/fig/fig_0201-ED-cell-autolab.Rnw}
<<child=here::here('assets/aux/fig/fig_0201-ED-cell-autolab.Rnw')>>=
# \label{fig:0301-ED-Autolab02-a}          (a)
# \label{fig:0301-ED-electrodes-distance}  (b)
# \label{fig:0301-ED-electrode-insulated}  (c)
# \label{fig:0301-ED-cell-autolab}
@

The secondary electrodeposition setup (see \cref{fig:0301-ED-cell-autolab}) used
a larger-volume cell, and was used for some projects after the completion of \cref{P1}.
Due to the shallower angle of the \gls{WE} in this cell (compared to the perfectly
vertical configuration in the primary setup) it was necessary to immerse the
entire \gls{WE} in the electrolyte solution and thus to fully insulated its electrical
contacting.
The longer distances from the cell ports to the electrolyte solution also
necessitated the use of a larger \gls{RE} and \gls{CE} (coiled \ch{Pt} wire).
The larger cell also meant a longer \glsname{WE}--\glsname{CE} distance
$\approx\qty{25}{\mm}$ (about twice as long as in the primary setup).

The fully insulated contact was sturdy and created in the following way:
the electrical wire was passed through
a disposable plastic pipette (with its bulbous part cut off) and contacted to
the \gls{TCO} surface with electrically conductive silver lacquer,
then covered with a strip of aluminium tape.
The tape and the end of the plastic pipette were then fixed into place by
colourless epoxy glue to add sturdiness and chemical stability to the contact,
and the whole contact was then covered with a thin coat of nail varnish (\cref{fig:0301-ED-electrode-insulated}).
The nail varnish was applied primarily to define the active surface area, and
for purposes of easier identification since we had several
available colours.





%%%% SECTION
\section[Chemical bath deposition of CdS films]{\texorpdfstring{Chemical bath deposition of \ch{CdS} films}{Chemical bath deposition of CdS films}}
\label{synthesis:chemical-bath-deposition}

\ch{CdS} nanoparticles were chemical bath-deposited onto
\qty{1.5}{\hour} electrodeposited \zincox\ \glsplural{NR} arrays.
An aqueous bath solution of \reactant[concentration=2,concentration-unit=\milli\Molar]{cadmiumsulfate},
\reactant[concentration=10,concentration-unit=\milli\Molar]{thiourea}, and
\reactant[concentration=1]{ammonia} was
thermostatically maintained at \qty{60}{\celsius} before immersing the
substrate for a set amount of time (up to \qty{2}{\hour}).
The substrate was always suspended vertically, and the solution was continuously
stirred during the deposition.
To prepare the deposition bath (per above), I started by dissolving
\reactant[mass=1.2365]{cadmiumsulfatehydrate} in \reactant[concentration=12.8]{ammonia}
to $[\ch{Cd^{2+}}]=\qty{24}{\milli\Molar}$
then volumetrically diluted it by a factor of eight to
\reactant[%
   concentration=3,%
   concentration-unit=\milli\Molar,%
   solvent={\qty{12.8}{\Molar} \ch{NH3\aq}}]{cadmiumsulfate}.
The cadmium sulfate salt was sourced from Merck (\ch{3 CdSO4 * 8 H2O}, \emph{pro analysi}).
Next I dissolved \reactant[mass=0.2284]{thiourea} in \qty{100}{\milli\L}~\water\
for a concentration of \qty{30}{\milli\Molar}.
The thiourea powder was sourced from Merck, \emph{for synthesis}.
% print only the number: \reactant*[switch=true]{CdSO4}

%\input{assets/aux/fig/fig_0203-CBD-synthesis.Rnw}
<<child=here::here('assets/aux/fig/fig_0203-CBD-synthesis.Rnw')>>=
# \label{fig:0203-CBD-station}
@


For a \gls{CBD}, first \qty{5}{\milli\L} of \reactant*[switch=true]{thiourea}
was added to each deposition bath (\cf\ \cref{fig:0203-CBD-station})
and allowed to reach thermal equilibrium at the set temperature
(at least \qty{15}{\minute} \qty{60}{\celsius}).
Simultaneously, a container with \reactant*[switch=true]{cadmiumsulfate} was also heated.
The substrates were lowered into each of the four deposition baths, and
\qty{10}{\mL} of \reactant*[switch=true]{cadmiumsulfate} was then added to each to start
the \gls{CBD} reaction.
The point of adding the \ch{Cd}-precursor at the last moment was to avoid unnecessarily
subjecting the \ch{ZnO}/\ch{Fe2O3} substrate to the strongly alkaline solution.
After the deposition was completed, the back of the glass substrate was
cleaned with a cotton swab soaked in $\approx\qty{0.75}{\milli\Molar}$ \ch{HCl},
then both sides were rinsed with deionized water and blow-dried with \ch{N2}.

The earliest report \cite{Nagao1968} of chemical bath deposition of \ch{CdS} from an alkaline
solution with an ammonium salt at room temperature produced thick films in the cubic structure.
Later work \cite{Nair1988,Mondal1983} found that the thickness of the deposit
\emph{decreased} with bath temperature.


The \ch{CdS} deposition proceeds along the following steps
(most mechanistic studies of \gls{CBD} of \ch{CdS} have proposed the so called
\emph{ion-by-ion} mechanism, which is what we present here) \cite{Hodes2003,Tak2009}:
\begin{subreactions}\begin{reactions}
NH3 + H2O             <=>& NH4\pch[] + OH\mch[]   \AddRxnDesc{Ammonium~formation} "\label{rxn:CBD-ammonia}" \\
Cd\pch[2] + 4 NH3     <=>& [Cd(NH3)4]\pch[2]      \AddRxnDesc{Formation~of~cadmium~complex} "\label{rxn:CBD-Cd-complex}" \\
(NH2)2CS + OH\mch[]    ->& CH2N2 + H2O + HS\mch[] \AddRxnDesc{Formation~of~hydrogen~sulfide} "\label{rxn:CBD-thiourea}" \\
HS\mch[] + OH\mch[]    ->& S\mch[2] + H2O         \AddRxnDesc{Release~of~sulfide~ion} "\label{rxn:CBD-sulfide}" \\
Cd\pch[2] + S\mch[2]  <=>& CdS\sld{}              \AddRxnDesc{Cadmium~sulfide~ion~by~ion} "\label{rxn:CBD-CdS-ionbyion}" \\
Cd\pch[2] + 2 OH\mch[] ->& Cd(OH)2\sld{}          \AddRxnDesc{Cadmium~hydroxide~precipitation} "\label{rxn:CBD-OH-cluster}" \\
Cd(OH)2 + S\mch[2]     ->& CdS\sld{} + 2 OH\mch[] \AddRxnDesc{Cadmium~hydroxide~dissolution} "\label{rxn:CBD-CdS-OH-cluster}"
\end{reactions}\end{subreactions}
usually described with the following overall reaction
\addtocounter{reaction}{-1}
\begin{reaction}
[Cd(NH3)4]\pch[2] + (NH2)2CS + 2 OH\mch[] ->
   CdS + CH2N2 + 4 NH3 + 2 H2O \AddRxnDesc{Deposition~of~cadmium~sulfide} "\label{rxn:CBD-dep}"
\end{reaction}

Ammonia dissolved in water produces a hydroxide ion (\cref{rxn:CBD-ammonia}) and
forms a cadmium tetramine complex (\cref{rxn:CBD-Cd-complex}) which is responsible
for slowing the release of solvated \ch{Cd\pch[2]} ions.
Sulfide ions are released (\cref{rxn:CBD-sulfide}) by the hydrolysis of thiourea
in alkaline solution (\cref{rxn:CBD-thiourea}) \cite{Ortega-Borges1993,Hodes2003}.
When the product of \ch{[Cd\pch[2]]} and \ch{[S\mch[2]]} exceed the solubility
product%
\footnote{%
   The \glsname{solubility_product} of \ch{CdS} has variously been reported in literature as
   \num[retain-unity-mantissa=false]{1e-25} \cite{Ortega-Borges1993},
   \num[retain-unity-mantissa=false]{1e-28} \cite{Hodes2003},
   \num{7.1e-28} \cite{Oladeji1997}, and
   \num{1.4e-29} \cite{Nair1988}.
}
of \ch{CdS} it precipitates (\cref{rxn:CBD-CdS-ionbyion}) either homogeneously
in the solution (forming colloids) \cite{Hodes1987} or
heterogeneously at the substrate surface \cite{Kaur1980}, leading to a film.
Note that formation of the hydroxide is not desired, but may still occur (which
is why it is included in the schema), but if it does occur the likelihood of sulfide
substituting for the hydroxide and forming the desired sulfide is very high due
to the much higher \glsname{solubility_product} (\num{2e-14}) of \ch{Cd(OH)2}.




%%%% SECTION
\section{Atomic layer deposition of iron oxide}
\label{synthesis:atomic-layer-deposition}

After having successfully coated electrodeposited \zincox\ \glsplural{NR} with
\ch{CdS} and achieved the intended effect of sensitizing \ZnO\ to visible
light, we considered other conformal coatings
that could alleviate two of the main drawbacks with \ch{CdS}, \namely,
the alkaline conditions during its deposition, and the toxicity of the \ch{Cd\pch[2]} ion.

Hewing to the principle of simplicity we established earlier, and considering
our alternatives, we landed on the abundant
\hematite\ (\glstext{hematite}), \ie, the principal component of rust.
The iron oxides and iron oxide hydroxides (see \cref{tab:0300-pxrd-pdfs}) have
a particularly well-studied
chemistry, which is no wonder considering their immense economic importance.
As the most thermodynamically stable iron oxide phase at \gls{STP}, \glstext{hematite} is
the end product of all steel corrosion, and has as such been studied in detail.
\hematite\ has a band gap of \Eg{2.2} (\qty{564}{\nm}), \ie, well into the visible
range, and is also non-toxic, resistant to chemical corrosion in both dark and
illuminated conditions, and its \gls{VBE} is suitable for oxygen evolution
from \water\ (\cf\ \cref{fig:0100-bandgaps-all}).
The main drawbacks with \glstext{hematite} is its indirect band gap and its short
intrinsic hole diffusion length (typically \qtyrange{2}{4}{\nm} \cite{Kennedy1978},
but up to \qty{20}{\nm} when considering the effect of the electrolyte and band bending
\cite{Edvinsson2018}), which was one of our main motivations behind minimizing the thickness
of the \gls{hematite} layer and the reason for using \gls{ALD}.

%\input{assets/aux/fig/fig_0202-ALD-ferrocene-process.Rnw}
<<child=here::here('assets/aux/fig/fig_0202-ALD-ferrocene-process.Rnw')>>=
# \label{fig:ALD-ferrocene-process}
@

By utilizing self-limiting surface reactions, \gls{ALD} can evenly deposit
thin films on a substrate of arbitrary shape and form.
The primary restriction is that the substrate needs to be thermostable, which was
not a problem for our samples, being essentially oxide on glass.
\Gls{ALD} is a slow deposition method, but no other technique at
our disposal could deposit a perfectly even layer
of material on a relatively rough substrate (such as our \ch{ZnO} nanorod samples),
in particular not without etching the amphoteric \zincox\ layer during
its deposition.

Binary compounds (such as \hematite) are usually deposited in
four steps for each atomic layer (called a \emph{cycle}, \cf\ \cref{fig:ALD-ferrocene-process}).
The first step contains the metal precursor, in this case ferrocene, which is
transported in the gas phase by a carrier gas to the deposition chamber
which was kept at \qty{450}{\celsius}.
The precursor is chemisorbed to the surface of the substrate forming one monolayer.
In the second step the carrier gas is flushed to rinse off any excess precursor
from the first step.
The third step introduces the second precursor, which react with the already chemisorbed
molecules on the surface, ideally forming one atomic layer of the desired phase.
The fourth step is another flushing operation.
Our \gls{ALD} process was based on previous work by Fondell \textit{et al.} \cite{Fondell2014a,Fondell2014}
and varied the number of cycles in a series: \numlist{35;50;62;75;87;100},
ideally corresponding to a hematite layer thickness of \qtyrange{2}{20}{\nm}.
Reaction conditions were optimized for hematite formation.

The \gls{ALD} instrument was a Picosun 8482 R series reactor from the Finnish Sun Oy company.




%%%% SECTION
\section{\texorpdfstring{\ch{ZnO} quantum dot synthesis}{ZnO quantum dot synthesis}}
\label{synthesis:quantum-dot}

Low-dimensional \zincox\ (wurtzite) \glsplural{NP} were synthesized as a colloidal%
\footnote{%
   A solution that scatters light is a \emph{suspension}.
   A \emph{colloidal} solution does not.
   Our prepared \zincox\ solutions were colloidal, but gradually turned into suspensions
   after weeks or months of storage.
}
solution in \solvent{ethanol}, by adapting a
synthetical route first described by \textcite{Spanhel1991} and also
by \cite{Hoyer1993,Hoyer1994}, and later improved by \textcite{Meulenkamp1998}.
The synthesis was a batch method involving two precursor solutions
prepared at room temperature
(for amounts, masses and concentrations see \cref{tab:0204-precursors-QD}).

In the first precursor solution zinc acetate dihydrate, \reactant[purity=99.5]{zincacetatedihydrate},
Fluka, was dissolved in \solvent[purity=99.5]{ethanol}.
In the second precursor solution lithium hydroxide monohydrate,
\reactant[purity=99]{lithiumhydroxidehydrate}, Sigma-Aldrich,
was dissolved in \solvent[purity=99.5]{ethanol}.
Note that \reactant*[switch=true]{lithiumhydroxidehydrate} may
have contained up to \qty{1}{\percent} of \ch{Li2CO3} (as specified on the bottle).
This can cause some challenges during \gls{QD} synthesis, which we will expand on later.

%\input{assets/aux/tab/tab_0204-QD-precursors.Rnw}
<<child=here::here('assets/aux/tab/tab_0204-QD-precursors.Rnw')>>=
# \label{tab:0204-precursors-QD}
@

%\input{assets/aux/fig/fig_0204-QD-synthesis-setup.Rnw}
<<child=here::here('assets/aux/fig/fig_0204-QD-synthesis-setup.Rnw')>>=
# \label{fig:0204-synthesis-QD}
@

Preparation of precursor solutions was as follows:
\begin{itemize}[after=\vspace{\baselineskip}]
\item add approximately \solvent[volume=45]{ethanol} to the Zn-reactor (\qty{100}{\mL} Erlenmeyer flask)
      and a magnetic stirrer bar; set temperature of water bath to boiling point of \solvent{ethanol},
\item turn on \ch{N2\gas} flow into the empty Li-reactor (round bottom flask) for a few minutes
      to purge \ch{CO2\gas} from it; add a magnetic stirrer bar;
      set temperature of water bath to \qty{45}{\celsius},
\item quantitatively transfer the weighed mass of \reactant{zincacetatedihydrate} to the boiling \solvent{ethanol},
\item add approximately \solvent[volume=50]{ethanol} and quantitatively transfer the weighed mass
      of \reactant{lithiumhydroxidehydrate} to the round-bottom flask,
\item once the mixtures have dissolved and the solutions
      are transparent and colourless move both reactors to an ice bath,
\item quantitatively mix the two precursor solutions to \qty{100}{\mL}
      (fill to the mark with \solvent{ethanol} as needed). Stir.
      Note the time ($t=\qty{0}{\minute}$ for \gls{QD} growth).
\end{itemize}

Both salts were easy to handle, weigh up and transfer.
The zinc acetate \reactant*[switch=true]{zincacetatedihydrate} dissolved easily after
a few minutes stirring in boiling \reactant{ethanol}.
But this solution was not stable for long: after a few hours it formed a milk-white suspension.

The lithium hydroxide \reactant*[switch=true]{lithiumhydroxidehydrate} has a more
challenging chemistry, primarily due to its carbonate salt.
Whereas both the dry and the slaked hydroxide have moderate solubility in \water\ and \EtOH,
with increasing solubility with increasing temperature, the
lithium carbonate, on the other hand, has poor solubility in \water\ and \EtOH,
and its solubility \emph{decreases} with increasing temperature.%
\footnote{%
   The solubility of \ch{Li2CO3} in water is \qty{1.54}{\gram} at \qty{0}{\celsius},
   and \qty{1.33}{\gram} at \qty{25}{\celsius} \cite{CRC104}.
   This points to an entropy effect, \ie, the solvation of \ch{Li+} and \ch{CO3^{2-}}
   in water decreases the overall entropy of the solution.}
In practice, this meant that during dissolution of
\reactant[solvent=\EtOH,concentration=0.14]{lithiumhydroxidehydrate} we needed
to take care to limit the competing dissolution reaction of \ch{Li2CO3}, which was
driven by the presence of dissolved \ch{CO2}.
Since the hydroxide is only moderately soluble to start with, we needed to use
as large a volume of solvent as possible. To avoid encouraging the formation
of the carbonate, \ch{CO2\gas} should be kept out of the headspace, and in addition
one should avoid heating the solution (definitely not boil it) so as not decrease
the solubility of any lithium carbonate that may form.
In short, the higher the temperature and the higher the flow rate of \ch{N2\gas},
the more lithium ions were lost as insoluble carbonate, and
the more solvent was lost to evaporation, which were both detrimental to our process.

For the same reason, the lithium hydroxide solution \reactant*[switch=true]{lithiumhydroxidehydrate} must be stirred only gently.
Even moderate stirring will draw in more air (which even under \ch{N2\gas} flow may contain
\ch{CO2\gas}) into the solution, making it easier to form \ch{Li2CO3}.
This reaction also neutralizes at least one hydroxide ion, thus decreasing the
precursor solution's apparent \pH.
Of course, strictly speaking \pH\ is not defined in a non-aqueous solvent,
but we may still use it as a proxy to discuss the changing chemistry of \EtOH\ as
air (containing \carbondiox) equilibrates with it.
To investigate this, I measured the proton activity (apparent \pH) of \EtOH\ of
two different purities using a regular \pH-meter,
and found \pH{\,\num{8.35}} (\qty{99.5}{\percent}) and \pH{\,\num{8.19}} (\qty{99.96}{\percent})
immediately after opening the factory-sealed bottle;
and \pH{\,\num{7.83}} and \pH{\,\num{7.40}} after aerating the samples overnight.
These observations suggest that equilibration with air, just like for \water, significantly
alter the carbonate chemistry of the solvent, which may influence the \gls{QD} synthesis,
and that varying trace amounts of \water\ in \EtOH\ were likely not decisive.
All \ZnO\ \gls{QD} in this work were synthesized using \solvent[purity=99.5]{ethanol}
for reasons of cost, and used more or less immediately after opening the bottle.%
\footnote{%
   With the benefit of hindsight, I should perhaps have let the \solvent{ethanol}
   solvent aerate fully (until the \pH{} settled) before preparing the precursors.
   This did not effect the quality of individual batches, but may have improved
   repeatability between batches.}

To prepare precursor solutions at precisely the intended concentration it is important to
maintain the desired solvent volume, which was challenging since \solvent{ethanol} evaporated
so readily, especially when boiling.
Instead of sealing the glassware (which changes the boiling point of the solvent)
we fitted the \ch{Li}-reactor with a water-cooled reflux tube which successfully captured
and returned effectively all the evaporated solvent.





%%% CHAPTER:
%%% CHARACTERISATION METHODS
\chapter{Characterization methods}%
\label{ch:characterization-methods}


%%% SECTION
\section{UV-Vis spectroscopy}
\label{methods:uvvis-spectroscopy}

An object under irradiation can interact with the light in various ways.
From an optical perspective, the incident light has to penetrate the surface of
the object to interact with the material in the bulk of the object. And then the
light has to penetrate another surface to exit the object \cite[p.\,71]{Stenzel2005}.

For light incident on an object, the light may
\begin{enumerate*}[label=(\alph*)]
\item be transmitted through the object (in a well-defined direction),
\item be diffusely scattered at the object interfaces or in its interior,
\item be specularly reflected, or
\item be absorbed at the object interfaces or in its interior.
\end{enumerate*}
If we posit that these processes are the only interactions that can occur
between the incident light and the object, then conservation of energy requires:
\begin{equation}\label{eq:TRSA=1}
\glsname{transmittance} + \glsname{reflectance} + \glsname{scattering} + \glsname{absorbance} = 1
\end{equation}
where \glsname{transmittance} is the \glstext{transmittance},
\glsname{reflectance} is the \glsdisp{reflectance}{specular reflectance},
\glsname{scattering} is the \glsdisp{scattering}{diffuse scattering}, and
\glsname{absorbance} is the \glstext{absorbance}.

The \glstext{transmittance} is also related to the \glstext{absorption_coefficient}
and the path length, $L$ in the familiar Beer-Lambert law
(also illustrated in the margin):
\begin{equation}\label{eq:transmittance}
\glsname{transmittance} \equiv \frac{I}{I_0} = \exp(-\glsname{absorption_coefficient}L)
\end{equation}

In most of my work, we were mainly interested in the optical absorption.
\marginpar{%
%\input{assets/aux/fig/fig_0301-beer_lambert_law.Rnw}
<<child=here::here('assets/aux/fig/fig_0301-beer_lambert_law.Rnw')>>=
@
}
If scattering can be neglected, \cref{eq:TRSA=1} simplifies to
\begin{equation}
\glsname{absorbance} = 1  - \glsname{transmittance} - \glsname{reflectance}
\end{equation}
or in terms of incident and transmitted light
\begin{equation}
\glsname{absorbance} =
\log\left(\frac{I_0}{I}\right) =
-\log(\glsname{transmittance})
\end{equation}
In practice, \gls{absorbance} is the base-10 logarithm of the spectral radiant power
transmitted through a reference sample divided by that
transmitted through the investigated sample, both observed in identical cells.

Note that \emph{absorptance}, defined as $-\log(1-\frac{I_\text{abs}}{I_0})$ \cite{Cohen2008}
is sometimes used in lieu of \glsname{absorbance} for samples where scattering
is not negligible, is algebraically equivalent to \glsname{absorbance}, since
$I_\mathrm{abs} = I_0 - I$, and thus the expression above can be rearranged
to $\log\left(\frac{I_0}{I}\right)$.

The \glstext{absorbance} can thus be determined by measuring both the \glstext{transmittance}
and the \glstext{reflectance} of a sample. Or, if reflectance can be neglected, then simply
by measuring its transmittance.

For a thin film, the amplitude and intensity of reflected or transmitted beams
of light can in principle be determined by setting up Maxwell's equations
and applying the appropriate boundary conditions \cite{Heavens1955}.
In practice that is very cumbersome and not something often done.
More to the point, for a thin film the \glstext{absorbance} equals the
\glstext{absorption_coefficient} times the film thickness, $d$: $A=\alpha d$.
The \glstext{absorption_coefficient} of a thin film can be expressed in terms
of the reflectance and transmittance \cite{Green2012}:
\begin{equation}\label{eq:alpha-thickness}
\glsname{absorption_coefficient} = \frac{1}{d}\ln\left(\frac{1-R}{T}\right).
\end{equation}
And if the thickness, $d$, is not known, simply rearrange the equation so the
left-hand side gives you $\glsname{absorption_coefficient}d = \glsname{absorbance}$.


We have employed \gls{UV-Vis-NIR} spectroscopy extensively, and with a few
different instruments.
For \insitu\ experiments on films or colloidal suspensions I used an
OceanOptics HR2000+ high-resolution fibre-optic spectrometer,
with a Mikropack DH-2000-BAL \gls{UV-Vis-NIR} light source using
both a deuterium lamp and a halogen lamp.
This setup was controlled \via\ a computer running SpectraSuite on Microsoft Windows~XP.
\Exsitu\ experiments on films were measured on the venerable PerkinElmer Lambda~900
\gls{UV-Vis-NIR} spectrophotometer equipped with an integrating sphere
(\eg, \cref{P1}, \cref{fig:P01-uvvisir-absorbance}).

\Gls{UV-Vis-NIR} spectroscopy gave us optical transmittance, reflectance, and/or absorbance
for both thin film and colloidal samples, which allowed me to track semiconductor band edges
and dye absorption peaks, among other features.



%%%% SECTION
\section{Photoluminescence spectroscopy}
\label{methods:photolum}

In \glsfirst{PL} spectroscopy the sample is excited with photons with higher
energy than its band gap and the sample's emission spectrum is recorded \cite{Pazoki2020},
\ie, photons in and photons out.
I have used \gls{PL} to study surface or interface phenomena involving
the absorption of photons that generate electron-hole pairs, which is of central
importance in \glstext{PC} \cite{Kundu2011a}.

In \gls{UV-Vis} fluorospectrometry (also known as \glsfirst{PL} spectrometry)
the sample under study is placed in front of a light source with a suitable bandpass filter
between them so that only a particular wavelength in the \gls{UV-Vis} region illuminates the sample.
Since the sample's emission can be assumed to be isotropic, the direction of excitation was set at
an angle to the detector to avoid detecting the incident light.
Special care was taken to keep stray light out of the spectrometer
(the room was darkened and the instrument was covered).

I recorded fluorescence spectra of electrodeposited \ch{ZnO} \glsplural{NR} on
\gls{TCO}-glass with and without \gls{CBD} \ch{CdS}.
Room-temperature steady-state emission spectra were recorded on a Fluorolog~3%
\footnote{%
   The Fluorolog 3's control software was based on the Origin~Pro software suite,
   with all recorded data saved as Origin plots by default. Not very convenient
   if you are not an Origin user.
   Using Origin's built-in command prompt, the data could be exported to ASCII
   in a two-liner: \texttt{fdlog.openpath(); }%
   % verb in footnote required fancyvrb package
   % https://tex.stackexchange.com/questions/203/how-to-obtain-verbatim-text-in-a-footnote
   \verb+{doc -e W {save -w %H %B%H.txt}}+%
   .%
}
(Horiba Jobin Yvon)
equipped with double-grating excitation and emission monochromators and a
\qty{450}{\watt} \ch{Xe}-lamp.
The main sample series (\ch{CdS}-coated \ch{ZnO} \glsplural{NR} on \gls{TCO}-glass)
was measured using right-angle (RA) detection with the sample substrate at \ang{30}
and always with the EE (electrode--electrolyte) surface facing the incident beam,
in order to minimize inner-filter effects.
The excitation and emission slits were set to obtain \qtylist{2;3;6}{\nm} and
\qtylist{4;6;8}{\nm} band passes, respectively (depending on excitation wavelength).
Integration time was set to \qty{0.1}{\second} per point.
A \qty{400}{\nm} long-pass filter was placed \emph{after} the sample to reduce
second-order diffraction by the excitation monochromator
(for $\glsname{wavelength}_\text{\glsname{excitation}}=\qty{350}{\nm}$ and \qty{375}{\nm}).
The emission spectra were corrected for the spectral sensitivity of the
detection system by using a calibration file of the detector response.





%%%% SECTION
\section{Raman spectroscopy}
\label{methods:raman-spectroscopy}

As we previously discussed (\cref{methods:uvvis-spectroscopy}), light interacts
with matter in a variety of ways.
When incident light is scattered by an object, it is overwhelmingly likely to
be \emph{elastically} scattered, meaning the light is propagated without changing
its wavelength (or equivalently, its energy or frequency).
With Raman spectroscopy we measure the \emph{inelastic} spectrum scattered from the
sample under illumination by laser light (high-power density, coherent, monochromatic, collimated light).
In this process the vast majority of the incident laser light is scattered
\emph{elastically} (Rayleigh scattering), and only about one inelastic scattering event
occurs for every \num[retain-unity-mantissa=false]{1e7} elastic scattering events (\qty{0.1}{\ppm}).
The inelastically scattered light has a very slightly different wavelength than the incident light,
because it has exchanged vibrational (or rotational) quanta with the specimen, which
can only occur if the polarizability of the specimen's electron cloud is changing
during the vibration \cite{Rahman2020a}, and thus carries with it information on the vibrational state
of the specimen which is correlated to its chemical bonding, chemical composition,
and phases in crystalline materials.

For some crystal structures, Raman and \gls{IR} are complementary, in the sense
that the crystal's vibrational modes are either Raman \emph{or} \gls{IR} active.
A lattice vibration can be simultaneously Raman and \gls{IR}-active only in
non-centrosymmetric crystal structures (\eg, polar crystals such as \ZnO,
\cf\ \cref{tab:0300-pxrd-pdfs}).
All molecules and solids have vibrational spectra, and nearly all vibrations
are either Raman or \gls{IR} active \cite{Dyer2005}.

Inelastic scattering of light by matter was first theorized by \textcite{Smekal1923} and
shortly after by \textcite{Kramers1924,Kramers1924a,Kramers1925}. This was around
the same time as Venkata Raman was pondering the molecular diffraction of deep bodies
of seawater during a voyage across the Indian Ocean \cite{Raman1922},
which led him and Krishnan to experimentally describe the Raman effect a few years later
\cite{Raman1928,Raman1928a}. The Raman effect was also independently described
experimentally in the USSR by Landsberg \& Mandelstam \cite{Landsberg1928,Landsberg1928a}.
The earliest investigations of the Raman effect in semiconductors occurred after
the advent of the laser in the 1960s \cite{Cardona1974}.

The terminology we use to assign vibrational spectral modes was determined
by an ICSU Joint Commission for Spectroscopy in 1953 \cite{Cohen2008} but is
commonly referred to as \textcite{Mulliken1930,Mulliken1955,Mulliken1956} notation.

In this thesis we concern ourselves with the Raman effect in crystalline semiconductors.
Semiconductors always have more than one atom in their unit cell, which leads
the dispersion relations (the relationship between frequency and wave vector) to
always exhibit two kinds of phonons with which photons can interact, \namely,
the \emph{optical} and \emph{acoustic} modes of lattice vibration.
The interaction with optical phonons%
\footnote{
   They are called \emph{optical} phonons because they can be
   excited by electromagnetic radiation in the optical wavelength region.
}
is what typically generates Raman scattering, and
the interaction with acoustic phonons results in so called Brillouin scattering
(first predicted by \textcite{Brillouin1922} in 1922) \cite{Pankove1975}.
Both optical and acoustic phonons have longitudinal and transverse modes,
depending on whether the vibration is along the direction of propagation
or orthogonal to it.

An incident photon of a certain wavelength can give or gain some of its energy
to the lattice in the form of a phonon.
The precise amount of energy gained or lost by the lattice is compensated by
an increase (Stokes) or decrease (anti-Stokes) in the wavelength (energy) of
the scattered light \cite{Loudon1964,Pankove1975}.
A phonon is a many-body-type quasiparticle;%
\footnote{%
   Some other commonly encountered quasiparticles are
   \textbf{polaritons} (coupled photon--optical phonon modes),
   \textbf{excitons} (an electron--hole pair),
   \textbf{polarons} (coupled electron--phonon modes),
   \textbf{plasmons} (a collective excitation of all the electrons in the crystal, \ie, a quantum of plasma oscillation),
   \textbf{magnons} (a collective spin excitation of the lattice),
   and
   \textbf{trions} (a localized excitation of two holes and one electron or two electrons
   and one hole that may occur in optically excited semiconductor nanostructures).}
a quantized lattice vibration and describes a collective motion of atoms in the lattice.
Phonons play an important role in determining a material's thermal, optoelectronic,
and electrical transport characteristics \cite{Senga2019}.
At $\glsname{absolute_temperature}=\qty{0}{\kelvin}$, a crystal lattice would contain no phonons.
At any temperature above absolute zero all matter vibrates to some extent,
due to lattice vibrations being quantized.
A material's phonon structure is fundamentally related to its thermodynamic
properties.

%\input{assets/aux/fig/fig_3300-vibrational-energy-levels.Rnw}
<<child=here::here('assets/aux/fig/fig_3300-vibrational-energy-levels.Rnw')>>=
# \label{fig:0300-vibrational-energy-levels}
@

Another way to look at this interaction is that the incident photon excites a phonon,
causing the semiconductor to re-emit the remaining (or gained) energy as a photon.
This scattering process is generally isotropic, which is why Raman scattering is
commonly measured in the antiparallel direction to the incident beam. Also, both
the incident and emitted light typically have limited penetration depth in the sample,
which also restricts detection to the reflection direction
(just like for \gls{SEM} and many other techniques with opaque samples).

The created phonon can in turn interact with the lattice to produce a second phonon
(of lower energy than the first), and this cascade process can be repeated,
generating what is called second-, third-, fourth-order (and so on) Raman scattering,
or more generally termed \emph{multiphonon} scattering.
Note that if the energy of the phonon coincides with the energy of an exciton (\emph{resonance}),
the intensity of the higher-order phonon can be strongly enhanced \cite{Pankove1975}.

In polar semiconductors, \ie, materials with a permanent electric polarization
due to the relative displacement of positive and negative ions in the crystal
structure (in other words, ionic crystals), with few defects
interaction between this polar electric field and
the optical phonons is expected to be the dominant scattering mechanism at \gls{RT}.
This so called polar--optical-phonon scattering interacts with the charge carriers
in what is known as the Fröhlich interaction \cite[p.\,40]{Stroscio2005},
which strongly influences the semiconductor's carrier mobility.

Raman spectroscopy is non-destructive, simple, and often (but not always) a quick method
to determine which crystal phases are present in a solid or solute sample.
It yields information about bonds and crystallinity, and we can use it to tell apart
different polymorphs of the same composition (\eg, \glsxtrlong{hematite} from maghemite)
which is not always straight-forward with \gls{XRD}.
One of the major weaknesses of Raman spectroscopy is fluorescence from a sample,
which can completely swamp the inelastically scattered signal and be very hard
to overcome.



\subsection{Resonant Raman scattering}
\label{methods:resonant-Raman}

As we have seen, regular phonon scattering (\ie, non-resonant) occurs through
excitation to virtual intermediate electronic transitions.
In this section we will briefly describe \emph{resonant} Raman scattering by optical phonons
as utilized in \cref{P4}.

When the wavelength (\ie, energy or frequency) of the incident light approaches one
of the electronic transitions of the crystal (such as the band gap in a semiconductor),
the Raman scattering can be enhanced by many orders of magnitude (this effect
was first reported in crystals by \textcite{Ovander1962}).
Achieving resonance used to be hard because the wavelength of the incident
laser light must be close to the band gap under study, and it is only with the
development of high-power \gls{UV} laser sources
that wide gap semiconductors such as \ZnO\ came within reach of resonance studies.
But even with the intensity gained from resonance, resonant Raman experiments
may be saddled with an even stronger phonon-aided luminescence that can drown out
all or some of the Raman modes, which is a drawback particularly affecting samples
with any organic residues (even very small amounts).

Under electronically exciting conditions (\ie, resonance) Raman can provide
information on vibrations in the excited state and can also give clues towards
the charge separation/transfer mechanisms in the material \cite{Rahman2020a}.
Polar crystals often complicate resonance by
leading to the introduction of multiphonon processes caused by
intraband Fröhlich interaction \cite{Cardona1983}.

%\input{assets/aux/fig/fig_0303-raman-setup.Rnw}
<<child=here::here('assets/aux/fig/fig_0303-raman-setup.Rnw')>>=
# \label{fig:0403-raman-setup}
@

A challenge when using resonant Raman for wide gap materials
is the need to utilize \gls{UV} light, which puts added constraints on the
optical components in the Raman spectrometer, which is why
we used a near-UV objective and \gls{UV} lenses in the spectrometer's
optical path. All measurements were recorded at ambient conditions.
Resonant Raman was measured with the \qty{325}{\nm} excitation line of a \ch{He}-\ch{Cd} laser,
using a 15x/0.30 NUV objective lens (Thorlabs).
The detector was calibrated against the \qty{1332}{\per\cm} line of diamond.
All measurements in the resonant experiment series used the exact same acquisition
parameters.
Cosmic rays were removed either by the detection and removal algorithm
built-in to the Renishaw software suite, or manually at a later stage by
linear interpolation over the affected data points.
Peak and baseline fitting was performed with the algorithm by \textcite{Davies2008}
(also see \cref{method-development:spectral-peak-fitting}).




%%%% SECTION
\section{Photocatalysis}
\label{methods:photocatalysis}

In order to evaluate the \glsdisp{PC}{photocatalytic} or
\glsdisp{PEC}{photoelectrochemical} properties of the
catalysts we synthesized, we employed both solar simulators
(emulating the standard \gls{AM15G} spectrum to varying degrees of fidelity)
and custom or off-the-shelf lamps.
I designed and built many of the experimental setups from scratch.


\subsection{The solar spectrum}
\label{pc:solar-spectrum}

To get a firm grasp of the solar spectrum it helps to first
understand the black body spectrum and the concept of \gls{AM}.

The \glstext{spectral_radiance} of an ideal black body%
\footnote{%
   Every object at above \qty{0}{\kelvin} emits and absorbs electromagnetic
   radiation to some degree. An object that absorbs all radiation is a black body
   and by definition has an emissivity of unity. For a black body the
   spectral distribution of this radiation depends \emph{only} on the object's
   \glstext{absolute_temperature}.}
is given by Planck's law%
\footnote{%
   By the year 1900, physicists had been trying to understand the dependence of the black body spectrum
   as a function of temperature and wavelength for almost half a century.
   Several proposed equations existed, but as they were within the confines of
   classical physics they all failed to match observations at higher frequencies.
   As the story is commonly told, in an act of desperation (Planck feared being scooped)
   he submitted an equation that seemingly worked but made use of Boltzmann's entropy law which
   rested on the assumption of atoms (quanta) and was statistical,
   both phenomena disliked by Planck who was a conservative physicist.
}
\cite[sec.\,2.3]{Suppan1994}
\begin{equation}\label{eq:planck}
\glsname{spectral_radiance} =
   \frac{2{\pi}\glsname{Planck_constant}\glsname{speed_of_light}^2}{\glsname{wavelength}^5}%
   \frac{1}{\exp\left(\displaystyle\frac{%
      \glsname{Planck_constant}\glsname{speed_of_light}}{%
      \glsname{wavelength}\glsname{Boltzmann_constant}\glsname{absolute_temperature}}%
   \right) - 1}
\end{equation}
This represents the emitted power per unit area of
emitting surface and per unit wavelength. \glsname{spectral_radiance} is called
the \glstext{spectral_radiance}, and when referring to a black body it is considered
independent of viewing angle (isotropic), independent of time, and independent of
position on the emitting surface (spatially homogeneous).

%\input{assets/aux/fig/fig_0101-blackbody-spectralradiance.Rnw}
<<child=here::here('assets/aux/fig/fig_0101-blackbody-spectralradiance.Rnw')>>=
# \label{fig:blackbody-spectralradiance}
@

%\input{assets/aux/tab/tab_0100-solarconstants.Rnw}
<<child=here::here('assets/aux/tab/tab_0100-solarconstants.Rnw')>>=
# \label{tab:0100-solarconstants}
@

By plugging the surface temperature of the Sun (see \cref{tab:0100-solarconstants})
into \cref{eq:planck} we can accurately model its radiance and the irradiance on Earth%
\footnote{%
   For an \emph{emitting} surface, the term is \textbf{radiance},
   and for an \emph{absorbing} surface it is \textbf{irradiance}.}
(stars are to a very good approximation black bodies).
Once the sunlight enters our atmosphere, though, things become much more messy,
and we are better off leaving analytical equations behind and embracing
international standards.
To understand the \glsdisp{AM15G}{standard solar spectrum}
we need to briefly familiarize ourselves with the \gls{AM} concept.

%\input{assets/aux/fig/fig_airmass.Rnw}
<<child=here::here('assets/aux/fig/fig_airmass.Rnw')>>=
# \label{fig:airmass}
@

The \glsxtrlong{AM} concept (see \cref{fig:airmass}) offers a compact way to express
the influence of atmospheric absorption and scattering (both of which depend on
path length in the atmosphere, \ie, \glsxtrlong{AM}) on the terrestrial irradiance
for all latitudes except for the highest.
Just outside the atmosphere, the \glsxtrlong{AM} is zero at all latitudes.
On the equator with the sun at zenith, \glsxtrlong{AM} is unity.
Assuming zero altitude (sea level) and a homogeneous plane-parallel atmosphere
(\ie, one in which density is constant and Earth's curvature is ignored),
\seemore{\cref{photoec}}
\gls{AM} is the secant of the zenith angle in degrees, $z$:
\begin{equation}\label{eq:airmass}
\mathrm{\glsname{AM}} = \sec(z) = 1/\cos(z)
\end{equation}
This equation breaks down above \ang{75} latitude, which is a problem
for astronomers but perfectly fine for our purposes.%
\footnote{%
   Better models of the atmosphere \cite{Woolard1966} demonstrate that
   \glsxtrlong{AM} reaches at most around \num{38} for astronomical objects
   towards the horizon ($z=\ang{90}$).
}




\subsection{Photocatalysis rigs}
\label{pc:rigs}

I designed several rigs for photocatalytic testing over the course of my work.
Some were tailored towards thin film evaluation and others towards the
evaluation of colloidal photocatalysts, all at room temperature
(no active cooling of the \gls{PC} reactor).
Our light sources are described in \cref{box:light-sources}.

%\input{assets/aux/box/box_light-sources.Rnw}
<<child=here::here('assets/aux/box/box_light-sources.Rnw'), cache=TRUE>>=
# \label{box:light-sources}
@

After a few not-so-successful lighting setups
involving white-light \glsplural{LED} and \enquote{black light}
\glsplural{CFL} (both turned out to be too low-intensity for any experiments),
we achieved our first working \gls{PC} reactor (\cref{fig:0400-pc-blue-beaker})
using a high-intensity blue-light \gls{LED} array.
%
The reactor for this lamp was simply a beaker wherein the sample was placed
face-up on a custom-machined acrylic platform so that a magnetic bar could
freely spin underneath it and stir the solution.
Progress monitoring was done by periodically transferring a sample
of the solution to a cuvette for \exsitu{} \gls{UV-Vis} spectrophotometry.
This setup was used in \cref{P1}.

We quickly evolved the setup to use a quartz cuvette to allow \insitu{} \gls{UV-Vis} tracking,
but with the down-side of no longer being able to stir the solution
(see \cref{fig:0400-pc-blue-vertical-with-condensers}).
We managed to improve the intensity and uniformity of the light from
the \gls{LED} array with the addition of an optical tube (containing a convex lens
with adjustable position along its length) and two aspheric condenser lenses
(Thorlabs ACL-5040, $d=\qty{50}{\mm}$, $f=\qty{40}{\mm}$) to the optical path,
producing a dense and even patch of light approximately \qty{2}{\cm} in diameter.
We chose to continue mounting the light source vertically because the quartz cuvette
only had two transparent sides.
This setup was in \cref{P2}, where we also experimented with a mercury arc lamp.

I also performed a number of \gls{PC} experiments with a mercury arc lamp
(as a \enquote{poor-man's} solar simulator), which resulted in quite
poor photocatalytic degradation, but some interesting morphological
changes of our core--shell nanorods reported in \cref{P2}.

With our third and fourth generation of the \gls{PC} reactor
(\cref{fig:0400-pc-solarsimulator-thinfilm,fig:0400-pc-solarsimulator-suspension})
we graduated to using a solar simulator with a xenon arc lamp.
The proximate cause for this setup was to have the light impinge directly onto a
thin-film photocatalyst that was lying on the bottom of the cuvette (facing up).
The ultimate cause was that the quartz cuvette had only two transparent sides,
which meant \insitu\ spectrophotometry would not have been possible with horizontal illumination.
The light from the solar lamp hit a mirror placed at a \ang{45} angle above the cuvette.
The dye degradation was tracked \insitu\ using our fibre-optic spectrophotometer.

%\input{assets/aux/fig/fig_0300-pc-rigs.Rnw}
<<child=here::here('assets/aux/fig/fig_0300-pc-rigs.Rnw')>>=
# \label{fig:0400-pc-blue-beaker}                    (a)
# \label{fig:0400-pc-blue-vertical-with-condensers}  (b)
# \label{fig:0400-pc-solarsimulator-thinfilm}        (c)
# \label{fig:0400-pc-solarsimulator-suspension}      (d)
# \label{fig:0400-pc-rigs}
@

Constructing these \gls{PC} rigs gave me ample reason to learn more
about basic optics like how to collimate visible light with lenses
or redirecting beams with mirrors, and the use of optical tables.
For the earliest setups an optical tube fitted with a convex lens
movable along the tube's length
(labelled as \enquote{\ch{Al} tube \& lens} in \cref{fig:0400-pc-rigs})
was the only beam-forming component. As such, I devoted some thought to
whether the inside of the tube would reflect the (small) UV emission of the
blue \gls{LED} (which is what we desired).
A metallic aluminium surface would be highly UV-reflective, but the question
in my mind was how reflective a naturally oxidized aluminium surface would be.
Aluminium oxide itself is a quite good \gls{UV} absorber. But the natural oxide layer
should be quite thin.
I proceeded under the assumption that UV reflectivity in the tube was sufficient,
with consideration given to the fact that the blue \gls{LED} was expected to
produce relatively little sub-\qty{400}{\nm} light.
In hindsight, a \gls{UV} reflectance experiment would have settled the matter,
but at the time I was young and reckless.



\subsection{Dyes}
\label{pc:dyes}

When evaluating a photocatalyst with regard to its effectiveness to photodegrade
a pollutant, we most often want to substitute the actual pollutant for a model
compound that
\begin{enumerate*}[label=(\alph*)]
\item is reasonable safe to work with,
\item has chemical properties that as far as possible match the pollutant,
\item and has a distinct absorption band in the \gls{UV-Vis-NIR} region for easy tracking.
\end{enumerate*}
As many pollutants of interest are polycyclic aromatic compounds,
and aromaticity often produces colour, \emph{dyes}%
\footnote{%
   Soluble colourants are called \emph{dyes},
   and insoluble colourants are called \emph{pigments}.
}
were perhaps an obvious model for \glsdisp{PC}{photocatalytic} degradation.

%\input{assets/aux/sch/sch_0302-tested-dyes.Rnw}
<<child=here::here('assets/aux/sch/sch_0302-tested-dyes.Rnw')>>=
# \label{sch:0302-tested-dyes}
@

The source of the colour of all dyes is the electronic transitions of their conjugated ring system.
And since they are more-or-less water-soluble, they are also salts (the aromatic part may
be cationic or anionic, and the formal charge may be localized or not).

Dyes containing the \ch{\bond{sb}N\bond{db}N\bond{sb}} functional group are called
\emph{azo} dyes, which \cref{sch:RB5-structure,sch:EBT-structure,sch:MO-structure} are
examples of.
Out of the dyes presented in this thesis (see \cref{sch:0302-tested-dyes}),
\gls{MB} exhibited the least changes to its absorption spectrum on the
addition of \EtOH\ (discussed further in \cref{results:P12-methylene-blue}).
Both \Gls{EBT} and \gls{MO} exhibited pronounced changes in their spectrum
as well as poor linearity with concentration on addition of \EtOH, and
\gls{RB5} decolourized completely in less than an hour under illumination
in the same conditions.

\gls{MB} is a cationic dye in the thiazin family of structures, and one of many
derivates of phenothiazine. It has brilliantly strong colour (high \glstext{absorption_coefficient})
making it very useful in the dyeing industry \cite{Tayade2009}, and was historically
used as an anti-malarial drug (the first such drug actually).
Around \qty{40}{\years} ago \gls{MB} was a popular dye sensitizer in thiazine
photogalvanic cells \cite{Mills1999}
(this was before the discovery of \ch{Ru}-based \glsplural{DSSC}).

Both dye \emph{degradation} and \emph{decolourization} will lead to a decrease in the optical
absorption spectrum, but whereas decolourization (also called \emph{bleaching}) is reversible
and leaves the dye molecule conformationally intact, degradation is irreversible and
eventually leads to the mineralization of the organic phase
(producing water, carbonates, nitrates, sulfates, \etc).

Photocatalytic degradation is the complete breakdown of the bonds in the target molecule
by the action of reactive species in solution formed by the photoexcited charges
on the photocatalyst's surface.
This degradation is commonly thought to proceed first \via\ a two-step attack of photogenerated
\ch{OH^.} radicals in solution on the sulfur of the thiazine ring to form
sulfoxide then sulfone:
% it is possible to put \chemfig{} inside reaction env
% https://tex.stackexchange.com/questions/511592/triple-bond-inside-escaped-chemfig-formula-within-chemformulas-reaction
% also, for these chemfigs I want roman font in order to match \ch{} output, so resetting \printatom before and after subreactions env
\renewcommand*\printatom[1]{\ensuremath{\mathrm{#1}}}
\begin{subreactions}\begin{reactions}%
% note, \fplus, \fscrp, and \fsscrp are defined by the chemmacros package
% anchor=180+\chargeangle moves the charge symbol a little further away by using TikZ anchor point,
% see chemfig manual for details
"\chemfig{R=\charge{90[anchor=180+\chargeangle]=\fsscrp}{S}-R\chemprime}" + OH^. &-> "\chemfig{R-\charge{90[anchor=180+\chargeangle]=\fsscrp}{S}(=[::-30]O)(-[::30]R\chemprime)}" + \proton{} \AddRxnDesc{Photodegradation~MB~forming~sulfoxide} "\label{rxn:PC-MB-sulfoxide}" \\%
"\chemfig{R-\charge{90[anchor=180+\chargeangle]=\fsscrp}{S}(=[::-30]O)(-[::30]R\chemprime)}" + OH^. ->&  "\chemfig{\charge{90[anchor=180+\chargeangle]=\fsscrp}{S}(-[::-150]R)(=[::-30]O)(=[::30]O)(-[::150]R\chemprime)}" + \proton{} \AddRxnDesc{Photodegradation~MB~forming~sulfone} "\label{rxn:PC-MB-sulfone}"%
\end{reactions}\end{subreactions}%
\renewcommand*\printatom[1]{\ensuremath{\mathsf{#1}}}
The formed sulfone group on the central ring makes the thiazine ring unstable causing it to
cleave leaving two benzenic rings and leading to the total loss of colour.
Hydroxyl radicals are expected to be the most potent of the photogenerated \gls{ROS},
but in principle the step-by-step photodegradation of the dye may proceed \via\ \gls{ROS}
in solution, or \via\ oxidation or reduction of adsorbed dye molecules or fragments.
It is perhaps useful to remember that at this point in the reaction, decolourization is total
despite the reaction not yet reaching complete mineralization. This is a result of our
choice of tracking variable: decolourization will occur as soon as the first bond is broken
(roughly requiring 2 holes), but complete mineralization of \gls{MB} requires
\num{102} holes \cite{Mills1999}.
Note that once decolourization has occurred, the degradation process can no longer
be spectrometrically tracked, and without using other analytical methods (such as
mass spectrometry) we cannot know how much time elapses until complete mineralization,
only that once the first irreversible degradation step has occurred,
mineralization will occur eventually.

The so called \emph{leuco}%
\footnote{% λευκός
   Greek $\uplambda\upepsilon\upupsilon\upkappa\upomicron\upzeta$ (leukos) \enquote{white},
   \cf\ \emph{leucocyte} (white blood cell).}
form of the dye is formed in the reversible reaction
\begin{reaction}
MB+ + \proton{} + 2 \electron{} <=> LMB
\AddRxnDesc{Formation~of~leuco~methylene~blue} "\label{rxn:leuco-methylene-blue}"
\end{reaction}
($E^0=\qty{0.011}{\voltNHE}$ at \pH{\,\num{7}})
that breaks the conjugated bonding and shifts the absorption maximum far into
\gls{NIR}, causing \emph{decolourization} without degradation.
\Gls{MB} may also form the semi-reduced form \ch{MB^.-} with an absorption maximum
at \qty{420}{\nm}, which readily disproportionates into MB and LMB. In addition,
\gls{MB} has a colourless oxidized form, \ch{MB^.+} which is unstable in alkaline
conditions \cite{Mills1999}.





%%%% SECTION
\section{Electron microscopy}%
\label{methods:electron-microscopy}

Electron microscopy allows us to resolve features
impossible to resolve with \gls{VLM}.
Furthermore, due to the rich interactions between high-energy electrons
and matter (see \cref{fig:0305-electronbeam-interactions}), a host of related
techniques can be used to study elemental composition and more.

%\input{assets/aux/fig/fig_0305-sem_scattering.Rnw}
<<child=here::here('assets/aux/fig/fig_0305-sem_scattering.Rnw')>>=
# \label{fig:0305-electronbeam-interactions}
@



\subsection{Scanning electron microscopy}%
\label{methods:scanning-electron-microscopy}

I made extensive use of \gls{SEM} to inspect the electrodeposited \zincox\ nanoarrays
and their different coatings.

Like many other scientific advances, the electron microscope was the result
of scientific and technical advances progressing in lockstep.
Diffraction with X-rays having been established some two decades earlier,
diffraction with electrons soon followed.
Louis de Broglie theorized (in 1925) that electrons (and all other matter,
for that matter) have wavelike properties, which would give them a very short
wavelength, which would mean much higher resolving power than \glsplural{VLM}.
Funnily enough, electron optics and the electron microscope were invented
in 1932 by \textcite{Knoll1932,Knoll1932a,Knoll1932b}%
\footnote{%
   Max Knoll, with a doctorate from the Institute for High Voltage Technology
   at the University of Berlin, soon after took a job at Telefunken to work in
   the budding field of television design.}
who had not heard about de Broglie's
ideas \cite{Williams1996} and embarked on the project because they thought the
wavelength limit did not apply to electrons (it does).
Within a year, the resolution limit of \gls{VLM} was surpassed.
Commercial production of \glsplural{TEM} began in Germany in 1939,
and in many other places after 1945.
Despite being commercially available for so long, the electron microscope is still a very expensive
piece of equipment, but it has also seen impressive improvements in resolution
due to higher beam brightness and ingenious corrections of
the spherical and chromatic aberrations that are inherent to electron optics.
The first \glsname{SEM} was patented by \textcite{vonArdenne1939} who managed
to \emph{scan} a very small raster with a finely focused electron beam.
Compared to \glsplural{TEM}, commercial interest was slower to develop for
\glsplural{SEM} and did not take off until the 1960s.

The interactions between the direct beam
(also called the incident beam, or \emph{primary electrons})
and the sample (\cf\ \cref{fig:0305-electronbeam-interactions})
can be divided into elastic and inelastic scattering.

Elastic scattering may entail a change in direction of the primary electrons,
but no detectable change in energy. They are referred to as backscattered electrons,
because it is easier to detect the elastically scattered electrons that happened
to deflect \emph{away} from the direct beam than the ones more or less in its path.
Back-scattering increases with increasing atomic number $Z$, making heavier atoms
appear brighter in back-scatter mode. Back-scattered electrons have a large information
depth in the sample  (they come from deep inside the sample).

Almost all the kinetic energy carried by the direct beam will eventually be
dissipated by inelastic scattering events, ending up as heat in the sample.
The small proportion of the direct beam ending up as back-scatter, secondary
electrons, or X-rays form the whole basis for imaging or analysis.
%
Inelastic scattering can be further subdivided into phonons, plasmons,
and inner shell excitations, and so called secondary electrons.
Secondary electrons is a somewhat loosely defined term, used to describe any electrons with
less than around \qty{50}{\eV} in kinetic energy escaping from
the surface of the sample \cite{Goodhew2001}.
They are most likely the result of inelastic scattering
very close to the sample surface, and
their yield (\ie, the number of secondary electrons per primary electron)
is often above unity which makes them very useful for \gls{SEM} imaging.
Because of their low kinetic energy, the secondary electron detector is
positively charged to attract them.
To form the two-dimensional micrograph from the secondary electron signal, it is simply
synchronized in time with the scanning of the direct beam over the sample.

I used \gls{SEM} to record the micro-morphology of our thin film samples
deposited \via\ \gls{ECD}, \gls{CBD}, or \gls{ALD} on various types of
\glsxtrlong{TCO} substrates.
As \gls{TCO} on glass makes for a fairly poor conductor (certainly in terms of
dissipating the electron beam without charging up the sample), I always
contacted an exposed part of the \gls{TCO} film with the metallic \gls{SEM}
sample holder.
For the same reason I generally avoided electron beam voltages higher than \qty{3}{\kV}.



\subsection{Transmission electron microscopy}
\label{methods:transmission-electron-microscopy}

In a \glsdisp{TEM}{transmission electron microscope} a highly focused beam of electrons is generated either
thermionically or by field emission (the latter can produce beams only \qty{1}{\nm}
in diameter at the sample).
The emitted beam is passed through a series of electromagnetic condenser lenses
and objective lenses before hitting the sample, and then a series of projector lenses
to form the resulting image.

A \gls{TEM} specimen is typically no thicker than the mean free path of
electrons at the accelerating voltage of the beam, or more precisely
we could say that the likelihood of a single scattering event to occur while
passing through the specimen is \qty{50}{\percent}.

The core--shell nanorods studied in \cref{P1} were sliced with the
focused ion beam (FIB) to prepare thin specimens of individual nanorods for imaging in the \gls{TEM}.
To improve conductivity, the specimens were coated with \ch{Pd}/\ch{Au}, and \ch{Pt}.




%%%% SECTION
\section{X-ray methods}%
\label{methods:xray}

\subsection{Thin-film X-ray diffraction}%
\label{methods:XRD}

A special case of scattering is when the material is periodic and the
incident light is monochromatic and parallel, resulting in \emph{diffraction},
which is
the constructive interference of elastically scattered electromagnetic waves
by electrons in a crystalline material.
As interatomic distances in crystals typically are a few ångströms,
X-rays are the most suitable light source.

X-rays scattered by different lattice planes will interfere constructively only
when their difference in path length is equal to an integer $n$ times their
wavelength $lambda$, and thus give rise to a reflection in the diffractogram.
This is Bragg's law:
\begin{equation}\label{eq:xrd-bragg}
2d\sin\theta = n\glsname{wavelength}
\end{equation}
where $d$ is the adjacent crystal plane spacing and $\theta$ is the angle
between the incident X-ray beam and the normal of the reflecting lattice plane,
and $n$ is the diffraction order.
The distance $d$ between two parallel and adjacent atomic planes in
a cubic crystal system is $d = a/\sqrt{h^2+k^2+l^2}$, and in the
hexagonal crystal system:
\begin{equation}\label{eq:hexagonal-dhkl}
\frac{1}{d^2} = \frac{4}{3}\frac{h^2 + hk + k^2}{a^2} + \frac{l^2}{c^2}
\end{equation}

\Gls{XRD} was not very useful for our thin film samples, which were thin
and non-monolithic and thus realized very little interaction volume.
Instead, we used \gls{GI-XRD} at very incident low angles and even then had some
trouble quantifying the sample's reflections.
We mainly used diffraction as a phase identification technique, but \gls{XRD}
is of course capable of yielding many sorts of structural information,
such as crystal structure, average grain size, preferred crystal orientation,
epitaxy, periodic defects,
and the presence and magnitude of strain in the material.

The relation between the grazing incidence angle, $\alpha$, and the depth
traversed by the X-rays into the material, $\tau$
\begin{equation}\label{eq:xrd-gi}
\tau_{1/e} = \frac{\sin\alpha}{\mu}
\end{equation}
which is the depth at which the intensity of the X-ray beam has decreased
to $1/e$. $\mu$ denotes the linear attenuation coefficient, which is a
material-specific value.

\Gls{XRD} measurements were also employed to determine the average crystallite sizes
in solid materials.
Smaller particle sizes cause a broadening of the reflections, which can be
calculated by the Scherrer equation \cite{Holzwarth2011}
\begin{equation}\label{eq:xrd-scherrer}
g = \frac{K\glsname{wavelength}}{\Beta\cos\theta}
\end{equation}
where $g$ is the crystallite size, $K$ is  the particle shape factor which is
typically \num{0.9} for spherical particles, $B$ is the \gls{FWHM} of the reflection
(minus the instrumental broadening), and $\theta$ is the diffraction angle.

%\input{assets/aux/tab/tab_0300-pxrd-pdfs.Rnw}
<<child=here::here('assets/aux/tab/tab_0300-pxrd-pdfs.Rnw')>>=
# \label{tab:0300-pxrd-pdfs}
@

To identify material phases we compared the measured $d$-spacings and their
fitted intensities with reference values from the JCPDS \glsplural{PDF}
\cite{McMurdie1986} (see \cref{tab:0300-pxrd-pdfs}).




\subsection{X-ray spectroscopy}
\label{methods:XRF}
\label{methods:XPS}
\label{methods:xray-spectroscopy}

In addition to \gls{XRD}, some projects in this work also
involved \glsxtrlong{XRF} or \glsxtrlong{XPS}.
In both of these techniques X-rays are absorbed by the sample, and they only differ
(on a conceptual level) by what happens next:
with \gls{XRF} we detect the emitted characteristic X-rays, and
with \gls{XPS} we detect the emitted photoelectrons.
A major practical difference is that \gls{XPS} requires a vacuum between the
sample and the detector, but with the upside of a much higher energy resolution
and a very surface-sensitive signal.

Whereas observed peaks in \gls{XPS} are due to \emph{elastically}
scattered photoelectrons (\emph{inelastically} scattered photoelectrons in fact
make up the background), observed peaks in \gls{XRF} are due entirely to
\emph{inelastically} scattered X-ray radiation (\ie, fluorescence).

For a maximum fluorescence yield, the incident X-ray energy should
lie just above the \glstext{binding_energy} of the electronic transition
of interest (for incident X-ray energy much higher the yield decreases, analogous
to how photons much higher than the \gls{band_gap} are a waste of energy).
Therefore, we always selected a secondary target with atomic number just larger
than the heaviest element in our sample, usually \ch{Ge} ($Z=\num{32}$).

Our \gls{XRF} experiments were performed on a
PANalytical Epsilon~5 with a \ch{W} (tungsten) anode
operated at \qty{75}{\kV} and \qty{8}{\mA}.
Each recorded spectrum was usually set to a live time of \qty{240}{\second}
with dead time varying between \qty{10}{\percent} and \qty{15}{\percent}.

\Gls{XPS} is based on the \emph{photoelectric effect} (of Einstein and Nobel Prize fame),
stating that a surface exposed to electromagnetic radiation of energy large enough
to overcome its first ionization energy emits electrons \cite{Einstein1905}.
The kinetic energy of the emitted photoelectron is related to its \glstext{binding_energy},
$\glsname{binding_energy} = \glsname{Planck_constant}\glsname{frequency} - \glsname{kinetic_energy}$
\cite{Einstein1905}.
The \glsfirst{binding_energy}, is the difference between the total energy of the system
in its initial state (before X-ray excitation) and final state (after emission of a core-level electron).
From a conservation of energy standpoint, then, the following holds:
% https://tex.stackexchange.com/questions/17528/show-equation-number-only-once-in-align-environment
\begin{align}\begin{split}\label{eq:xps}
\glsname{Planck_constant}\glsname{frequency} + E_\mathrm{initial}(\mathrm{N}) &= E_\mathrm{final}(\mathrm{N}{-}1) + \glsname{kinetic_energy}\\
\glsname{binding_energy} &= E_\mathrm{final}(\mathrm{N}{-}1) - E_\mathrm{initial}(\mathrm{N})\\
\glsname{Planck_constant}\glsname{frequency} &= \glsname{binding_energy} + \glsname{kinetic_energy}
\end{split}\end{align}
where $E_\mathrm{initial}(\mathrm{N})$ is the total energy of the neutral system with
$N$ electrons, and $E_\mathrm{final}(\mathrm{N}{-}1)$ is the total energy of the
now cationic system \emph{after} an electron with \glsfirst{kinetic_energy} has been ejected
due to its absorption of the X-ray photon \glsname{Planck_constant}\glsname{frequency}.

Assuming the Fermi levels of the sample and the \gls{XPS} spectrometer
are aligned (true if the sample is conductive and in good electrical contact with the spectrometer)
the \glstext{binding_energy} can be determined from the experimentally measured
\gls{kinetic_energy} \cite{Greczynski2023}:
\begin{equation}
\glsname{binding_energy} =
\glsname{Planck_constant}\glsname{frequency} - \glsname{work_function} - \glsname{kinetic_energy}
\end{equation}

Since the emitted core electron has to pass through the electron density
at the atom's surface, the \glsname{binding_energy} measured by \gls{XPS}
experiences a \emph{chemical shift} that depends on the valence electron charge density
due to chemical bond formation \cite{Greczynski2023}.
The realization that the chemical shift observed in core-level \gls{XPS}
provides information about the chemical environment in the sample
was made by \textcite{Siegbahn1968} at Uppsala University.
The simplest way I have found to conceptualize the chemical shift is probably in terms of
electronegativity: if the valence electrons of the atom in question are being
pulled away by a more electronegative neighbour, the core electrons experience
less \emph{screening} of the nuclear charge, and should thus be bound harder
than otherwise, causing a chemical shift towards higher \glstext{binding_energy}.

Samples need to maintain electrical neutrality during the measurement.
Insulating or semiconducting samples will otherwise experience charge-up, leading to
mismatch in \glstext{Fermi_level} between the sample and the spectrometer.
This can be corrected by use of a flood gun (low energy electron gun)
inside the sample chamber to neutralize the sample.
But even so, it is difficult to perfectly compensate for this charge-up, which
will lead to an observed shift in the \glstext{binding_energy} making it
necessary to calibrate the energy scale using a known elementary peak.
We calibrated against the hydrocarbon surface contamination peak
\eltr{C}{1s} which has a known value of \qty{285.0}{\eV}.

If the atom is in an excited state (such as at a higher oxidation state)
the chemical shift is perturbed further, causing a so called \enquote{shake up}
transition at the higher energy side of the core level transition.
The $\Delta E$ between the core level and the shake up transition is highly
characteristic of the chemical oxidation state of the atom.

The \gls{XPS} at the department of chemistry (in the \glsname{ESCA}-lab) has a
tungsten anode which we operated with an aluminium cathode that emits radiation
at \qty{1487}{\eV}.
The measurements were performed using a PHI~5500 spectrometer, using
monochromatic \ch{Al} $K\alpha$ radiation (\qty{1487}{\eV}), with an electron
emission angle of \ang{45} (tungsten anode, aluminium cathode).
The dimensions of the probed region were approximately \qty{2}{\mm} by \qty{4}{\mm}.
We used a flood gun since our samples (thin semiconducting films on glass) were
not sufficiently electrically conductive.

Peak fitting was done with Fityk~v1.3.1 \cite{Wojdyr2010}.
I stripped the baseline of each \eltr{C}{1s} spectrum using a spline function
and the transitions were fitted using three pseudo-Voigt kernels, and
the offset of the fitted \eltr{C}{1s} transition from the reference value
\qty{285.0}{\eV} was calculated.
Finally, the energy scale of each survey spectrum and all its transitions
were offset accordingly.

Apart from using X-ray spectroscopy one can also probe the characteristic X-ray energies
from core levels using \gls{EDS}, which was used to record the equivalent of \gls{XRF} spectra
but on the micro-scale
as well as to create 2D maps of elemental composition over areas of interest.
The elemental \emph{fluorescence yield} (\cf\ \cref{fig:xray-fluorescence-yield})
applies also for \gls{EDS}, which had the practical effect of making lighter elements
effectively invisible.





%%% CHAPTER:
%%% ALGORITHMS & DATASETS
\chapter[Algorithms \& datasets]{Algorithms \& datasets}
\label{ch:algorithms-datasets}


It seems that we live in a time of transitions.
The energy transition is underway, even if at times the speed appears glacial.
Simultaneously, the scholarly publishing system is indeed morphing into
one built on open science.
But whereas the energy transition is driven by an urgent need to halt
\gls{GHG} emissions, the transition to open science has been inevitable \cite{Dyson2000}
since the Web was gifted to the world by Berners-Lee and CERN \cite{Hoffmann2023}
and dropped the marginal cost of distribution of text (and later images and videos)
to effectively zero.

\emph{Open science} or \emph{open scientific knowledge} commonly refers to the
\enquote{open  access  to  scientific  publications, research  data,  metadata,
[\ldots], software, and source code} under a \emph{libre} licence and
free of charge \cite{KungligaBiblioteket2023,UNESCO2021}.
This chapter mainly concerns itself with open access to \emph{research data},
as defined by OECD \cite{OECD2007} almost \qty{20}{\years} ago:

\blockquote{%
   [\ldots] factual records [\ldots] used as primary sources for scientific research,
   and that are commonly accepted in the scientific community as necessary to validate
   research findings. A research data set constitutes a systematic, partial representation
   of the subject being investigated.}

The Swedish Research Council stated in 2020 that
the transition to open access to research data shall be fully implemented no later
than 2026.%
\footnote{\hlink*{https://www.vr.se/english/mandates/open-science/open-access-to-research-data/the-way-towards-open-access-to-research-data.html}{https://www.vr.se/english/mandates/open-science/open-access-to-research-data/the-way-towards-open-access-to-research-data.html}}
% https://links.solarchemist.se/shaare/s5JKTQ
Which may sound ambitious, but a lot has actually been happening both across Europe
and the world in recent years, for example:
guidelines have been published for the scientific management of FAIR data
(also specifically for chemistry data)
\cite{McEwen2023,Scheffler2022,Lamprecht2020,KungligaBiblioteket2019,Wilkinson2016},
as well as practical guides on how to manage data to ensure reusability \cite{Borgman2022}
and a radical idea on how to finance it \cite{Mons2020}.
FAIR principles have also been introduced for \emph{research software} \cite{Barker2022},
and a light shone on the perceived barriers to sharing \cite{Gomes2022}.

There is clearly a lot of inertia in science, but I would like to believe that
once it picks up speed the transition will be unstoppable.
That also means that the best time to influence this transition is right now.
% So, experiment! Share your data, or code, or both!
There are different levels of software reproducibility, and you should strive for
the highest one you can manage:%
\footnote{These criteria were formulated by Steve Crawford \orcid{0000-0002-8969-5229} in a NASA talk.}
\begin{enumerate*}[label=(\roman*),itemjoin={{; }},itemjoin*={{; and }}]
   \item software is not mentioned
   \item software is mentioned, but not described
   \item software is described, but not made available (or by request only)
   \item software is made openly available
   \item \ldots with a permissible licence
   \item \ldots with documentation and testing
   \item software is shared as a generalized package.
\end{enumerate*}

The benefits of sharing data are, I would like to think, self-evident.
By also sharing the code that was used to turn that data into the reported results,
the combined artefact gains more value than the sum of its parts.
It is natural to worry that one's code is not \enquote{good enough} to be publicly shared,
but in the words of the \emph{Code is science Manifesto} \cite{Yehudi2019}:
% https://codeisscience.github.io/manifesto/manifesto.html
\blockquote{%
   You do not have to be a computer scientist or professional software developer
   to write code, and your code does not have to be perfect in order to be published.
   If code produces paper-ready results, the code too is paper-ready.}
By sharing your data \emph{and} your code, you give others the ability
to stand on your shoulders, instead of just admiringly looking at you flexing them.
Shared data and code is a prerequisite to \emph{validate} and \emph{reproduce}
research findings.

An important part of a reproducible scientific workflow is to
express visualizations and tables as code rather than producing them by copy-pasting
data into a point-and-click tool.
% inspired by Wickhams's remark in his interview by Will Chase 2020-04-06
% https://links.solarchemist.se/shaare/AxpsQA
Likewise, and this might not be a popular opinion, the command line is powerful
and can save a serious amount of time doing otherwise mundane computation tasks \cite{Perkel2021}.

The scientific quality of your shared data can be significantly improved by
including units and/or uncertainties in the data itself \cite{Hanisch2022}.
This can be achieved with both \R\ and Python, but is still somewhat of a challenge.
In \cref{P3} we managed to include uncertainties (but not units) in the data
itself, such that the error propagation then became automatic.
In fact, in 2022 the Physical Chemistry division of IUPAC launched the project
\enquote{Digital representation of units of measurement} in collaboration
with CODATA and CIPM \cite{Frey2020,CODATA2020}.

To make a paper completely reproducible you can use the following approach:%
\footnote{These criteria were inspired by Jeff Leek \hlink*{https://simplystatistics.org/posts/2015-12-11-instead-of-research-on-reproducibility-just-do-reproducible-research}{https://simplystatistics.org/posts/2015-12-11-instead-of-research-on-reproducibility-just-do-reproducible-research}.}
\begin{enumerate*}[label=(\roman*),itemjoin={{, }},itemjoin*={{, and }}]
   \item use Git for version control
   \item use Quarto for your analysis code and \emph{generate} the desired output format using \texttt{pandoc}
   \item when your paper is done post it to a preprint server
   \item post your data to an appropriate repository like \hlink{https://snd.gu.se}{SND}
      or a general-purpose repository like Zenodo
   \item publish any software you develop to a controlled repository like CRAN
   \item participate in the post-publication discussion on the Fediverse and with a blog
\end{enumerate*}
The point is to inextricably tie the data to the manuscript by expressing
visualizations as code, in addition to putting all of it under version control.
Such a manuscript can be called a \emph{reproducible manuscript}.


The rest of this chapter briefly presents some of the algorithms and datasets that
I have shared publicly%
\footnote{
   Almost all \via\ \hlink{https://github.com/solarchemist}{GitHub},
   but also \via\ Git hosting services
   \hlink{https://codeberg.org/solarchemist}{Codeberg} and
   \hlink{https://git.solarchemist.se}{Gitea}.
}
and that this thesis has reused in some form.
I have always used Git to have version control (this allows a specific version
of the code/data to be referenced, which is invaluable in case the implementation
changes or in case of bugs).
And to make it easier for others to get started quickly, I have almost always
managed to share these tools as \R\ packages.%
\footnote{%
   An \R\ package is a way of collecting code, data, and documentation into
   a structured and standardized collection of files that can also explicitly
   state its dependencies.
}



\section{Notes on spectral peak fitting}
\label{method-development:spectral-peak-fitting}

Peak fitting is a very complex problem, but luckily it is one that has already
been solved many times over. Unless you have very
specific needs, there is no need to implement your own peak fitting algorithm.
In the \R\ ecosystem, there is of course the well-regarded \Rpackage{hyperSpec} package,
but I never felt comfortable with its attempt to do \emph{everything}.
My preference is software that hews closer to the Unix philosophy of
\enquote{doing one thing and doing it well}. I believe this software philosophy
also happens to agree well with the scientific principle
of parsimony \cite{Gauch2003}.

I was lucky to have decided from the get-go to only consider peak fitting
software that enabled a reproducible analysis, meaning not only the output
but also the input parameters must be possible to put under version control.
So I was delighted when the \texttt{diffractometry} CRAN package crossed my path.
Here was an extremely well-documented, fully automatic, \R-based peak fitting
algorithm \cite{Davies2008}
specifically geared for X-ray diffractometry but whose documentation \cite{Davies2010}
noted it could be useful for the analysis of Raman, FTIR, NMR or mass spectrometry data.

I tested its baseline identification and peak decomposition functions successfully
on my own Raman, \gls{XRF}, Mössbauer and \gls{XRD} data, and created simple
wrapper scripts to work around some of the idiosyncracies of the \texttt{diffractometry}
package (primarily by reining in its tendency to happily repeat an already accomplished fitting).
The algorithm provided all the expected peak parameters such as centroid position,
area, height, \gls{FWHM}, goodness of fit values, \etc.
The resonant Raman analysis in \cref{P4} was completed using my \texttt{diffractometry} wrapper.
I found that this algorithm struggled when presented with spectra that required
fitting multiple kernels to each physical peak, as it lack the ability to \enquote{lock}
certain parameters and re-run the fitting.

Sticking to my guns, I started looking for alternatives.
Enter Fityk by \textcite{Wojdyr2010}, a GUI peak fitting program which thanks to
an ingenious built-in command-line terminal coupled with a built-in logging
mechanism affords the same degree of reproducibility by
automatically writing the corresponding text command (including arguments) for
every user interaction in the GUI to a text file that is then \emph{executable}
by the Fityk built-in terminal.

In light of this parity with regards to reproducibility and marked improvements in many
other respects, Fityk has become this author's tool of choice for spectral peak fitting.



\section{Converting between reference electrode scales}
\label{algorithm:electrochemical-scales}

The algorithm at the heart of \cref{refelectrodes} was not complicated,
\seemore{\hlink{https://solarchemist.se/2016/11/24/reference-electrodes}{\cref*{refelectrodes}}}
as the electrochemical reference scales are essentially related by
constant offsets, but to make the algorithm more useful it also takes
into consideration each reference scale's temperature and concentration dependence.
This involved creating a matrix for each reference scale, and then
finding the desired point in $(\glsname{absolute_temperature},C)$-space (so to speak) by linear
interpolation.

This package supports conversion to/from any of the following reference electrodes:
standard hydrogen, silver--silver chloride, mercury calomel, lithium, sodium,
magnesium, and the absolute vacuum scale. I designed the internals of the package
to make it relatively easy to extend it with more reference electrode scales or
with additional temperature/concentration values.



\section{Tauc fit algorithm}
\label{algorithm:tauc}

In order to facilitate the calculation of optical band gap values from hundreds of
\gls{UV-Vis} spectra, I designed a semi-automatic algorithm based in part on work
shared with me by a former colleague.
Many papers in the literature that rely on band gap values determined \via\ Tauc fitting
unfortunately lack the necessary information for someone else to reproduce the
analysis.
In this algorithm such reproducibility is near guaranteed since all the input
parameters (that control the Tauc fit) must be provided explicitly as function
arguments. If the user additionally uses version control (as they should),
\emph{and} create their Tauc plots with code, reproducibility should be ensured.
In fact, that is precisely how the Tauc fits in \cref{P1} were managed.

Tauc fitting builds on the known fact \cite{Tauc1966,Viezbicke2015} that
the product of a semiconductor's \gls{absorption_coefficient} and
the incident photon energy \ch{\photon} is proportional to the square root of
the energy difference between its \gls{band_gap} and the incident photon energy $E$:
\begin{equation}\label{eq:ahv-Eg}
\glsname{absorption_coefficient}\glsname{Planck_constant}\glsname{frequency}
\propto
\left(E-\glsname{band_gap}\right)^{1/r}
\end{equation}
where the exponent $r$ indicates the nature of the electronic transition:
\begin{enumerate*}[label={},itemjoin*={{ and }}]
   \item $r=1/2$ for direct allowed transitions,
   \item $r=3/2$ for direct forbidden transitions,
   \item $r=2$ for indirect allowed transitions,
   \item $r=3$ for indirect forbidden transitions.
\end{enumerate*}

A detailed description of the implementation of my Tauc algorithm can be found
% the starred variant of \cref*{} prevents hyperlinking that cross-ref
\seemore{\hlink{https://github.com/solarchemist/uvvistauc}{\cref*{uvvistauc}}}
in the vignette to the \texttt{uvvistauc} package (link in the margin).

The \texttt{uvvistauc} package is a \emph{dependency} of \cref{P1}.
Of course, the paper has many other dependencies,%
\footnote{For the sake of brevity, let us limit ourselves to \emph{software} dependencies.}
such as the SpectraSuite software and driver on the computer,
its operating system, and any software running on the hardware of the spectrometer itself.
Ideally, the source code for all these dependencies would be freely available
to everyone, so that the behaviour of the system could be validated and data
reproducibility ensured.




\section{Dataset of band edges and gaps}
\label{dataset:bandgaps}

This dataset lists the energy position (potential) of the valence and conduction
band edges for some semiconductors in contact with an aqueous electrolyte.
It is certainly not comprehensive, but manages to include most binary and tertiary
semiconductors found in the literature.
\Cref{tab:0400-bandgaps} shows a subset of the dataset, in this case with
the band edge potentials at the \pH\ of \gls{PZZP} for each material (all in the dark).
You may note that \cref{fig:0100-bandgaps-all} is also built on this dataset.

This sort of \enquote{band edge level} plot showing relative or absolute position
of the \gls{VBE} and \gls{CBE} at the surface of various semiconductor electrodes
abounds in the photoelectrochemical literature.%
\footnote{%
   For some examples in the primary literature, see
   \textcite{Hernandez-Alonso2009,Burnside1999,Matsumoto1996,Serpone1995,Kung1977,Kohl1977,Bolts1976,Laflere1974};
   in review articles see
   \textcite{Kudo2009,NavarroYerga2009,Gratzel2001,Mills1997,Linsebigler1995a,Nozik1978},
   \textcite[p.\ 173]{Gleria1975};
   and in the secondary literature see
   \textcite[p.\ 89]{Miller2008}, \textcite[p.\ 289]{Morrison1980}, and \textcite[p.\ 313]{Sato1998}.
}

The package was designed with functions to add new data (new materials),
to create \LaTeX-formatted tables of the data, or to plot the dataset.
\seemore{\hlink{https://github.com/solarchemist/bandgaps}{\cref*{bandgaps}}}
By recording the \pH\ at \gls{PZZP}, and explicitly noting whether the
material follows Nernstian \pH\ dependence,
the dataset can provide relative band edge levels at any given \pH\ between
multiple semiconductors provided they are all Nernstian.
For oxide semiconductors in aqueous solutions, \ch{\proton} and \ch{\hydroxide} are the dominant
adsorbed species, and therefore the potential drop across the Helmholtz layer, $V_\mathrm{H}$,
and the \gls{flatband_potential} changes systematically with \pH.
The \pH\ at which the potential drop across the Helmoltz layer is zero is called
the \glsxtrlong{PZZP} \cite{Nozik1978}.
The change of $V_\mathrm{H}$ with \pH\ then follows a straight-forward Nernstian relationship:
\begin{equation}\label{eq:nernstian-pH}
V_\mathrm{H} = 0.059\left(\pH_\mathrm{\glsname{PZZP}} - \pH\right).
\end{equation}

%\input{assets/aux/tab/tab_0403-bandgaps.Rnw}
<<child=here::here('assets/aux/tab/tab_0403-bandgaps.Rnw')>>=
# \label{tab:0400-bandgaps}
@



\section{Dataset of X-ray transition lines}
\label{dataset:xraylines}

For analysing \gls{XRF} spectra (as well as \gls{EDS}), it is most useful
to have at hand a list of all X-ray transitions of the elements.
\seemore{\hlink{https://github.com/solarchemist/xraylines}{\cref*{xraylines}}}
Freely available data could be found on the web in the form of HTML-formatted
\enquote{Tables of Physical \& Chemical Constants} from Kaye \& Laby
hosted by the National Physical Laboratory,%
\footnote{%
   \hlink*{https://web.archive.org/web/20190511222636/http://www.kayelaby.npl.co.uk/atomic\_and\_nuclear\_physics/4\_2/4\_2\_1.html}{https://web.archive.org/web/20190511222636/http://www.kayelaby.npl.co.uk/atomic\_and\_nuclear\_physics/4\_2/4\_2\_1.html}.%
}
or in our PANalytical Epsilon~5 \gls{XRF} instrument's software,%
\footnote{%
   Version 2.0J/ICSW 2.8 of Dec 7, 2010 with kernel PW5050.
   Contained data on every transition's energy and relative intensity,
   as well as edge energies for each element. Which were stored in a user-accessible
   format but had to be manually transcribed.}
or in the primary literature  as poorly formatted tables in PDF files.
None of these tables were very convenient to work with, to say the least.

I combined the Epsilon~5 dataset with that of Kaye \& Laby, and where the they diverged
(usually at the third decimal) I gave priority to the former on account of it being more
recent, and extended this combined dataset by adding natural line widths and
fluorescence yields for each transition from \textcite{Krause1979,Krause1979a}.
\seemore{\hlink{https://solarchemist.se/2014/09/13/xray-line-energies}{Blog}}
To avoid ambiguity I relabelled all transitions that used Siegbahn notation into
the systematic IUPAC notation; I was helped by table~10.11 of \textcite{Lengyel1998}.

%\input{assets/aux/fig/fig_xray-overview.Rnw}
<<child=here::here('assets/aux/fig/fig_xray-overview.Rnw')>>=
# \label{fig:xray-fluorescence-yield}  (a)
# \label{fig:xray-moseley}             (b)
# \label{fig:xray-overview}
@

As a quality check, the dataset can be used to recreate the classic Moseley plot%
\footnote{
   First constructed by \textcite{Moseley1913}, and at the time a major breakthrough
   (this was before the discovery of the proton). It identified the existence of
   up-to-then unknown elements corresponding to
   $Z$=\numlist{43;61;75} (\ch{Tc}, \ch{Pr}, and \ch{Re}) \cite{Egdell2020}.
   Perhaps more importantly, it demonstrated that the elements in a solid sample could
   be identified by measuring the spectrum of the X-rays it emits.
   % http://www.physics.ox.ac.uk/history.asp?page=moseley
}
(see \cref{fig:xray-moseley}).




\section{Dataset of some properties of the elements}
\label{dataset:periodicdata}

Frustrated by the lack of freely available tools to plot one's own periodic table,
and realizing that \texttt{ggplot2} \cite{Wickham2016} could be quite easily
adapted for the task, I put together a small dataset and designed a proof-of-concept
table of elements plot that worked for my own purposes.
The IUPAC campaign for the year of chemistry in 2019 gave me the impetus to
clean up the generated periodic table and share the code publicly.
\seemore{\hlink{https://github.com/solarchemist/periodicdata}{\cref*{periodicdata}}}
The eagle-eyed reader might notice that \cref{fig:0100-elements-of-photocatalysis}
was in fact created using this dataset.





%%% CHAPTER:
%%% Quantum confinement in low-dimensional ZnO
%\input{assets/components/optical-quantum-confinement.Rnw}
<<child=here::here('assets/components/optical-quantum-confinement.Rnw'), label='ch-quantum-confinement', warning=FALSE>>=
# \chapter{Optical quantum confinement in low-dimensional ZnO}
@





%%% CHAPTER:
%%% RESULTS
\chapter[Results]{Results}
\label{ch:results}

In this chapter I summarize results from our published papers (\cref{P1,P3})
and results from our not-yet published manuscripts (\cref{P2,P4}), as well
as some work on the growth of \ZnO\ \glsplural{QD} that are not presently part
of any manuscript (\cref{results:P21-tracking-growth-ZnO-QDs}).



%%% PAPER I % \cref{P1} %%% (P01)
%\input{assets/code/P01/300-text.Rnw}
<<child=here::here('assets/components/0501P-ZnO-CdS-nanoarrays.Rnw'), label='0500-ZnO-CdS-nanoarrays', warning=FALSE>>=
# \section{PC properties of ZnO/CdS nanoarrays}
# %\input{assets/code/P01/300-text.Rnw}
@




\clearpage
%%% PAPER II % \cref{P2} %%% (P03)
%\input{assets/components/0503P-Ultrathin-ironox-ZnO-NRs.Rnw}
<<child=here::here('assets/components/0503P-Ultrathin-ironox-ZnO-NRs.Rnw'), label='0500-Ultrathin-ironox-ZnO-NRs', warning=FALSE>>=
# \section{Zinc oxide/iron oxide core--shell nanorods}
# %\input{assets/code/P03/300-text.Rnw}
@



\clearpage
%%% PAPER III % \cref{P3} %%% (P12)
%\input{assets/components/0512P-ZnO-QDs-sizes-photocatalysis.Rnw}
<<child=here::here('assets/components/0512P-ZnO-QDs-sizes-photocatalysis.Rnw'), label='0500-ZnO-QDs-sizes-PC', warning=FALSE>>=
# \section{Non-degradative processes in methylene blue dye}
# %\input{assets/code/P12/301-methylene-blue.Rnw}
# \section{Photocatalysis with quantum-confined ZnO}
# %\input{assets/code/P12/302-PC-quantum-confined-ZnO.Rnw}
@


% \clearpage
%%% Characteristics of growth of ZnO QDs (unpublished)
%\input{assets/components/0521P-ZnO-QDs-growth-suspension.Rnw}
<<child=here::here('assets/components/0521P-ZnO-QDs-growth-suspension.Rnw'), label='0500-ZnO-QDs-growth-suspension', warning=FALSE>>=
# \section{Long-term tracking of growth of ZnO QDs}
# %\input{assets/code/P21/300-text.Rnw}
@


\clearpage
%%% PAPER IV % \cref{P4} %%% (P25)
%\input{assets/components/0525P-Phonon-confinement-ZnO.Rnw}
<<child=here::here('assets/components/0525P-Phonon-confinement-ZnO.Rnw'), label='0500-Phonon-confinement-ZnO', warning=FALSE>>=
# \section{Non-resonant Raman in low-dimensional ZnO}
# %\input{assets/code/P25/301-nonresonant.Rnw}
# \section{Resonant Raman in low-dimensional ZnO}
# %\input{assets/code/P25/302-resonant.Rnw}
@




%%% CHAPTER:
%%% SUMMARY AND OUTLOOK
%\input{assets/components/summary-and-outlook.Rnw}
<<child=here::here('assets/components/summary-and-outlook.Rnw'), label='ch-summary', warning=FALSE>>=
# \chapter{Summary and outlook}
@






%%% CHAPTER:
%%% SAMMANFATTNING
\begin{otherlanguage}{swedish}
\chapter{Populärvetenskaplig sammanfattning}
\label{ch:sammanfattning}

%\input{assets/components/0900-sammanfattning.Rnw}
<<child=here::here('assets/components/0900-sammanfattning.Rnw'), label='sammanfattning'>>=
@
\end{otherlanguage}




%%% BACKMATTER
% Acknowledgements, References, Colophon, Lists of ..., gitmeta page
%\input{assets/components/backmatter.Rnw}
<<child=here::here('assets/components/backmatter.Rnw'), label='latex-backmatter'>>=
@

\end{document}