Accepted Manuscript

Title: Modeling of luminescence-based oxygen sensing by

redox-switched energy transfer in nanocrystalline TiO2 :Sm3+

Authors: M. Eltermann, V. Kiisk, A. Berholts, L. Dolgov, S.

Lange, K. Utt, R. Jaaniso

PII: S0925-4005(18)30524-0
DOI: https://doi.org/10.1016/j.snb.2018.03.034
Reference: SNB 24319

To appear in: Sensors and Actuators B

Received date: 24-10-2017

Revised date: 8-3-2018
Accepted date: 9-3-2018

Please cite this article as: M.Eltermann, V.Kiisk, A.Berholts, L.Dolgov, S.Lange,
K.Utt, R.Jaaniso, Modeling of luminescence-based oxygen sensing by redox-
switched energy transfer in nanocrystalline TiO2:Sm3+, Sensors and Actuators B:
Chemical https://doi.org/10.1016/j.snb.2018.03.034

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
Modeling of luminescence-based oxygen sensing by redox-switched
energy transfer in nanocrystalline TiO2:Sm3+

M. Eltermann, V. Kiisk, A. Berholts, L. Dolgov†, S. Lange, K. Utt, R. Jaaniso*

Institute of Physics, University of Tartu, W. Ostwald St 1, Tartu EE50411, Estonia

T
Highlights

R IP
 Oxygen sensitivity of Sm-doped TiO2 was studied from 100 ppm to 100% of O2

SC
 Oxygen responses of stationary and time-resolved luminescence were measured

 An original model is developed for rare-earth-doped semiconducting materials


U
Model describes oxygen dependence of complex time-resolved luminescence decays
N
 Luminescence sensitivity is linked to defect densities and surface-to-volume ratio
A
M

Abstract
ED

It is shown that the luminescence of Sm3+ ions, which were doped into anatase nanopowder, is highly
sensitive to the presence of oxygen gas. The luminescence was effectively excited through TiO2 band-
PT

to-band absorption and exhibited oxygen sensitivity in a wide concentration range from pure O2 gas
down to 100 ppm of O2 in a nitrogen atmosphere. An increase in oxygen concentration led to a stronger
intensity and longer lifetime of Sm3+ luminescence, an exactly opposite behavior to luminescence-based
E

sensors described by Stern-Volmer law. An original physical model is developed for describing such
CC

luminescent enhancement mechanism: it is proposed, that the adsorbed oxygen suppresses the inherent
luminescence quenching of Sm ions taking place via a resonant energy transfer to acceptor defects in
A

the material. Electron transfer between the adsorbed oxygen and these defects changes the structure of
electronic energy levels and hence the energy accepting ability of the latter. The model allows describing
the oxygen-dependent non-exponential luminescence decays in a quantitative manner and relates the
luminescence characteristics to the material parameters such as the surface-to-volume ratio of
nanocrystallites and the density of acceptor defects and gas adsorption sites.

1
Keywords: titania, nanopowder, samarium, oxygen sensor, luminescence, modeling

†
Present address: Key Laboratory of Bioinorganic and Synthetic Chemistry of Ministry of Education,
School of Chemistry, Sun Yat-Sen University, Guangzhou 510275, P. R. China

* Corresponding author, e-mail: raivo.jaaniso@ut.ee

T
R IP
SC
U
N
A
M
ED
E PT
CC
A

2
Introduction

Dioxygen is an important gas due to its role in metabolism, combustion, and in various industrial
applications. Among different type of oxygen sensors, optical sensors [1] have been more widely spread
during the last decade. In addition to general advantages such as high tolerance to electrical noise and
remote access to the sensing volume, the optical sensors can be operated in the imaging mode and have
demonstrated more stable operation in the applications which have traditionally used electrochemical
sensors [2].

T
Optical oxygen sensing is usually done by photoluminescent (PL) probes embedded into the oxygen
permeable polymer or porous sol-gel matrices. Mostly organic probe molecules with energetic levels

IP
resonant with triplet oxygen levels and with long excited state lifetime are used (see [1,3] and references

R
therein). The sensing effect of these materials is based on luminescence quenching via so-called
collisional energy transfer process between the probe and oxygen molecules. Similar optical materials

SC
have also been developed for pressure sensitive paints which luminescence is actually responding to the
partial pressure of oxygen in the air [4]. Organic fluorophores, however, are susceptible to photo-

U
bleaching and cannot withstand elevated temperatures, a prerequisite in many industrial applications.
Therefore, there is an interest to develop luminescence based inorganic oxygen-sensitive materials
N
which could be applicable in a broader range of environments.
A
A different emerging class of optical gas sensors is based on the influence of surface redox reactions on
M

PL centers of metal oxide semiconductor materials. These materials, widely used for electrical gas
sensing [5], are naturally high temperature tolerant and chemically inert. The sensing mechanism of
ED

oxygen in them relies on the fact that absorption and desorption of oxygen on their surface releases or
traps charge carriers. Many of these oxides have visible PL due to different lattice defects. Several works
have attempted to harness this PL as a gas sensor signal with various semiconductor oxide nanomaterials
PT

such as SnO2 [6–8], ZnO [8–15], In2O3 [16], TiO2 [8,17,18] and WO3 [8]. Additionally, influence of
oxidizing gases on PL has been reported for several materials with a wider bandgap, more concretely,
E

for nanostructures of ZrO2 [19], BaTiO3, SrTiO3, HfO2 [20] and MgO [12,21]. In particular, oxygen
sensitivity has been addressed or reported in several of these nanostructures [8,12–14,18–21].
CC

Even a more elegant way to engineer the metal oxides for PL gas sensing is by intentional doping. A
notable class of stable dopants is rare earth ions, which are efficient emitters even at high temperatures
A

and allow spectral and time-domain filtering due to arrow spectral lines and long PL lifetime,
respectively. Several early examples [22–25] demonstrated the potential of such approach. Very recently
oxygen sensing has been systematically studied in different rare earth doped metal oxide or composite
nanomaterials [26–31]. With the exception of a rather complex composite system of
Sr4Al14O25:Eu2+/Dy3++AgNP in polymers [28], all the investigated materials (Sm:TiO2 [26], Eu:TiO2
[27], Pr:K0.5Na0.5NbO3 [29], Sm:TiO2/MOF [30], Eu:ZrO2 [31]) showed not quenching but enhancement

3
of luminescence with increasing oxygen pressure. In our previous work [26], we proposed a qualitative
model for such behavior, caused by energy transfer from PL centers to plurivalent oxide defects, the
latter switched between different charge states by redox processes on the surface via electron transfer.

In the present work, a quantitative model is elaborated for describing the stationary and time-dependent
PL as a function the oxygen content. In particular, for the describing the decay curves, a physical model
based equations are derived and used instead of earlier phenomenological approach with Kohlrausch
functions. The decay curves will be studied and fitted with the model equations in a wide range of
oxygen concentrations from 100 ppm to 100% at normal pressure. The PL characteristics (quantum

T
yield, lifetime) are precisely linked to the material parameters such as the concentration of PL-quenching

IP
defects, density of gas adsorption and electron trapping sites as well as surface-to-volume ratio.

R
SC
Experimental

TiO2 doped with Sm3+ was prepared via the sol-gel technique. Ti(OC4H9)4 solution in butanol and

U
SmCl3•6H2O solution in methanol were used as precursors. A mixture of the two solutions was added
dropwise to distilled water while stirring. A white precipitate formed. The precipitate was dried using
N
rotary evaporator and then calcined in the air at 800 °C for 2 hours. The resulting white powder was
A
dispersed in distilled water using an ultrasound probe to reduce agglomeration of the crystallites. Finally,
the suspension was transferred to fused silica substrates by drop coating in the air. The substrates were
M

pretreated in air plasma for cleaning the surface and increasing its hydrophilicity. Two different powders
with 0.5 and 3 at% Sm3+ dopant concentration were made. The elemental analysis made by X-ray
ED

fluorescence showed that the residual concentration of other impurities was <0.1 %. Cl is the major
candidate for an additional impurity because of SmCl3 dopant source but the analysis showed that only
0.02-0.03 wt% of Cl remained in the samples after annealing.
PT

Scanning electron micrographs were acquired with FEI NanoSem 450. Raman scattering measurements
were performed using Renishaw inVia micro-Raman setup with excitation at 514 nm.
E
CC

The gas sensitivity experiments were carried out while the samples were under test gas flow inside a
sample chamber equipped with a temperature controlled sample stage (Linkam THMS350V). The gas
was fed into the chamber using mass flow controllers (Brooks, model SLA5850), the composition of the
A

test gas was varied by setting the flow rates of different source gases individually while keeping the total
flow rate constant (200 ml/min). The source gases were nitrogen, oxygen and nitrogen mixture with 10
% of oxygen (all 99.999 % pure). Photoluminescence (PL) was excited with a 355 nm nanosecond
pulsed Nd:YAG laser (Elforlight SPOT) with a repetition rate of 5 kHz for spectral or 200 Hz for PL
decay measurements, dispersed using a monochromator (LOMO MDR-23) and detected by a CCD
camera (Andor DU240-BU) or by a photomultiplier tube (Hamamatsu R2949) with a photon counting

4
multiscaler (Fast ComTec P7888). Before gas sensing measurements, the samples were always
preheated at 150°C for > 10 min in the flow of dry gas to ensure stable initial conditions and to minimize
water vapor side effects. All measurements were performed at atmospheric pressure and at 25 °C.

Results and discussion

The deposited material showed a hierarchical structure of agglomerated nanocrystallites as seen in SEM
micrographs (Figure 1). The grain size was estimated to be 40 nm, based on a set of measurements at

T
randomly chosen locations on the SEM micrographs. The crystalline phase of the annealed samples was

IP
determined from Raman scattering measurements. Raman spectra showed only strong peaks
corresponding to anatase phase [32,33] and were indistinguishable for the samples with different (0.5-3

R
%) Sm3+ concentrations (Figure 2). The Raman peaks are known to shift and broaden in the case of grain

SC
sizes below 20 nm [34]. These shifts were not observed, indicating the grain size to be indeed more than
20 nm.

U
N
A
The PL spectrum of TiO2:Sm3+ under UV excitation is shown in Figure 3. It consists of several rather
narrow lines in the red part of the spectrum that originate from radiative electron transitions 4G5/2→6HJ
M

inside the 4f-shell of the Sm3+ ion [35]. Three spectra, shown in Fig. 3 were measured at different oxygen
concentrations in the mixture of N2 and O2 gases. One can see that the intensity of Sm3+ lines strongly
ED

depends on the concentration of oxygen in the ambient. The fine structure of the Sm3+ lines and their
intensity relative to each other is unchanged when the content of oxygen is varied between 0 % and 100
%. It indicates that the crystallographic surrounding of the emitting ions remains unchanged during
PT

changes in gas composition and only one crystalline site of Sm3+ is present.

In order to investigate the time-dependent responses of Sm3+ luminescence to changes in oxygen

E

concentration, the integral intensity of the strongest transition 4G5/2-6H7/2 around 615 nm was monitored.
CC

A time series of the spectra was recorded by CCD, whereas the intensity of each spectrum was integrated
as shown by the shaded area in the inset of Fig. 3. Figure 4 illustrates the responses of integral PL to
step-changes of oxygen content at the input of measurement chamber. In each measurement cycle, the
A

sample was exposed to oxygen-containing gas for 10 min and then to pure nitrogen for 10 min before
the next exposure to oxygen. As compared with our initial study [26], where only the oxygen contents
near the atmospheric composition were probed, we changed the oxygen content over 4 orders of
magnitude, starting from 100 ppm trace level up to 100% at normal pressure. A closer look at the
characteristic response and recovery times is given in the inset of Figure 4. At room temperature, the
measured response time was under 1 min and recovery time about 5 min. We note that the response time

5
of the sensor material should be even smaller, as the measured response includes the instrumental time
(≈20 s) of changing the gas composition.

A way to clarify the sensing mechanism is to investigate the PL decay curves (Figure 5), which may
contain rich information about underlying processes. The PL decay kinetics showed a systematic trend
with changing oxygen concentration – the decays became faster as ambient O2 concentration decreased.
At the same time, as one can see on the semi-log graphs, the decay profiles appear to be strongly non-

T
exponential. The qualitative description of the luminescence decay was proposed in our earlier work
[26]: it was proposed that the excitation energy transfer from Sm3+ to TiO2 defects causes the accelerated

IP
initial phase of the decay and that the delayed excitation of Sm3+ via electron traps causes the slow tail

R
of the decay. In the current work, we outline a quantitative model and validate it on the experimental
data obtained with different samples and at different conditions.

SC
We assume that the excited state of Sm3+ (4G5/2) is depopulated by two processes: by natural decay with
a rate constant k0 and by excitation energy transfer to lattice defects with matching excitation energies.

U
If these defects are randomly distributed in the material, then the PL decay follows the well know law
N
[36]:
A
𝛽
𝑢(𝑡) = 𝑒 −𝑘0 𝑡−(𝑐/𝑐0 )(𝑘0 𝑡) (1)
M

The power index  depends on two factors: i) the type of interaction between the Sm3+ donors and energy
acceptors and ii) on the Euclidean dimensionality of the acceptor space (see Appendix I). The most
common case is dipolar interaction in 3D space – the Förster’s energy transfer with =1/2. Parameter c
ED

is the acceptor concentration and the constant c0 characterizes the energy transfer strength and is
inversely proportional to the 3rd power of the characteristic (Förster’s) energy transfer radius. Previous
PT

studies have estimated the natural decay time 𝜏0 = 1/𝑘0 of Sm3+ luminescence in TiO2. Values 258 –
350 μs have been obtained [37–39], so it is reasonable to assume 𝜏0 ≈ 300 μs. In Figure 5, the dashed
black line denotes a purely exponential decay with 𝜏0 = 300 μs. The measured PL decays faster at the
E

initial stage after the exciting impulse, which clearly indicates the presence of quenching.
CC

However, the situation is further complicated as the tail part of the decay is slower than the natural
decay. Note that the energy of the exciting quantum is bigger than the bandgap energy of anatase [40].
A

In such case, this UV quantum is absorbed by TiO2 host and then the energy is transferred to Sm ions
[41]. So we conclude that this slow decay must be caused by a delayed excitation of the Sm3+ ions,
where the initial excitation is somehow trapped for an extended period in the TiO2 matrix before the
energy is transferred to a nearby unexcited Sm3+ ion. The binding energy of these trap states must be
relatively small (comparable to the thermal vibration energy 𝑘B 𝑇), as the resulting escape times are still
as small as several milliseconds at room temperature. Indeed, several studies suggest the existence of

6
such shallow traps in TiO2 [42–44]. Taking into the account the presence of delayed excitation processes,
the decay law can be generalized as follows (see Appendix I):

𝐼(𝑡) = 𝐼0 𝑢(𝑡) + 𝐼1 𝑘𝑒 −𝑘𝑡 ⊗ 𝑢(𝑡), (2)

where the first term describes the instantly excited PL centers and the second term those PL centers
excited with a delay. It is assumed here that the delayed population occurs with a constant rate k, which
corresponds to the presence of a single type of trap level. The ⊗ symbol marks convolution. The
convolution reflects the fact that different PL centers start to decay at different time delays from the
laser pulse and the convolution sums over all possible delay times from 0 to 𝑡. Parameters I0 and I1

T
characterize the relative weights of directly excited and delayed luminescence, respectively. When we

IP
fitted the experimental decay curves with the equation (2), the tails of the curves were still not described

R
in a satisfactory manner (see Fig. 6). Therefore, we further assume the presence of traps with different
depths, which leads to a distribution of delayed population rates. The equation (2) is now further

SC
generalized to equation

∞
(3)
𝐼(𝑡) = 𝐼0 𝑢(𝑡) + 𝐼1 ∫ 𝑑𝑘 𝜌(𝑘) 𝑘𝑒
0 U
−𝑘𝑡
⊗ 𝑢(𝑡)
N
where (k) is the distribution of delayed population rates. If to adopt the commonly used exponential
A
distribution of trap depths [45] and assume thermally activated release of carriers from traps, then the
resulting 𝑘-distribution is given be an equation
M

𝑘 𝑅 (4)
𝜌(𝑘)𝑘 = 𝑅 ( )
𝑘𝑀
ED

where 0 ≤ 𝑘 ≤ 𝑘𝑀 , and 𝑅 = 𝑘B 𝑇⁄Δ𝐸 (see Appendix II). The two parameters kM and R determine the
scale and shape of the distribution 𝜌(𝑘). All the decay curves could be well fitted with equations (3) and
PT

(4), which elaborated form is given in the Appendix I as (A8) and (A17). The fitting procedure was
carried out in Mathcad 15, where the integral was evaluated by discrete convolution (based on an FFT
E

algorithm) using the convol function from the signal processing extension pack. The least-squares
CC

optimization was performed with Minerr (using Levenberg–Marquardt algorithm) and the shot noise
statistics of the data was taken into account. The corresponding fitting curves are superimposed over all
experimental curves in Figure 5. Figure 6 illustrates the role and importance of different model
A

components.

At fitting the resulting function 𝐼(𝑡) to the experimental decay curves it was considered that some of the
parameters should remain constant. The natural lifetime 𝜏0 was fixed at 300 µs (see the discussion

7
above). The value of 1⁄2 for 𝛽 corresponds to the scenario of infinite 3D crystal lattice where the Sm3+
ions are sufficiently far from each other so that they do not interact. This choice was suitable for the
samples doped with 0.5 at% of Sm, but in the case of 3 at% doping the smaller values of 𝛽,
approximately equal to 3/8, led to an improvement of the fit. Such value of  is characteristic for dipole-
quadrupole interaction, which has been observed, e.g., for Sm3+ self-quenching in KYP4O12 [46]. This
may indicate the interaction between the Sm3+ ions in our 3% sample too, so in the case of the 3 %
sample 𝛽 was fixed at 3/8. The other possible reason for a lower value of  is that Sm ions are somewhat
preferentially located near the surface in case of higher concentration as a result of segregation [47].

T
This would lead to effectively lower value of D in the equation (A1) and hence to the value <0.5 even

IP
in the case of dipolar energy transfer mechanism. At the fixed values of 𝜏0 and 𝛽 the adjustment of the
remaining free parameters 𝐼0 , 𝐼1 , 𝑘𝑚𝑎𝑥 , 𝑅, and 𝑐/𝑐0 provided good fits of all experimental curves. The

R
average values of parameters (2-3 samples were studied at each Sm concentration) together with error
margins are given in Table I. The values of 2, characterizing the quality of decay curve fitting, are also

SC
given in the table.

U
Across the oxygen concentration series the parameters 𝐼0 , 𝐼1 , 𝑅, and 𝑘𝑚𝑎𝑥 generally varied less than
30%. However, the parameter 𝑐/𝑐0 , which is proportional to the concentration c of acceptor centers,
N
changed substantially and decreased systematically with increasing content of O2 gas (Figure 7). So we
A
identify the change in the concentration of acceptor centers as the main factor connecting the changes
in O2 concentration and PL intensity of Sm3+.
M
ED

Interaction of gaseous oxygen with the surfaces of metal oxides is generally accompanied by a charge
transfer. Various negatively charged oxygen species may be formed at the surface of TiO2 but at room
temperature the negatively charged molecular ion is the dominant adsorbate [48]. In order to explain the
PT

decreasing number of energy acceptors with increasing oxygen pressure, it is quite natural to assume,
that the charge transfer occurs from acceptor defects to the increasing number of 𝑂2− . As the result of
E

charge donation, the energy level structure of the acceptor completely changes. Consequently, its energy
CC

levels are generally turned off resonance with luminescent Sm ions and it ceases to act as an energy
transfer acceptor. If one assumes that the charge transfer occurs between the adsorbed oxygen and
acceptor defects only, then one can write
A

∆𝑛 = −∆𝑛𝑎𝑑𝑠 (5)

where n is the change in the number of acceptor centers and nads is the change in the number of
adsorbed 𝑂2− . If to consider a nanocrystal with volume V and surface area S, then one can write n=cV
and nads=cadsS, where cads is the surface density of 𝑂2− . A linear relation between the acceptor
concentration and the concentration of surface adsorbents can be derived now from equation (5):

8
 𝑐 𝑐𝐷 𝑐𝑎𝑑𝑠 𝑆 (6)
= − ( )
𝑐0 𝑐0 𝑐0 𝑉

where cD is the concentration of acceptors in the absence of charged surface species.

This allows us to fit an adsorption isotherm (a function that connects the adsorbate concentration and
gas pressure) directly to the ambient oxygen content vs 𝑐/𝑐0 data points of Figure 7. The dependence
depicted in Figure 7 has a vague resemblance to the Langmuir isotherm but spans more orders of O2
concentration. Tóth isotherm [49],

T
1⁄𝑚
(𝐾𝑥)𝑚
𝑐ads = 𝑐𝑠𝑎𝑡 [ ] (7)
1 + (𝐾𝑥)𝑚

IP
gives a better description, as shown by fitting results in Figure 7. This isotherm is a generalization of

R
the Langmuir and Freundlich isotherms and is believed to be more suitable for a broad pressure range

SC
and heterogeneous substrates. Instead of oxygen partial pressure we used the oxygen content (vol%) x
at normal pressure in equation (7). A fitting with global parameters m and K was performed on the
datasets of samples with similar Sm concentration. The resulting parameter values are given in the last

U
columns of Table I. Note that these two parameters do not have a distinct physical meaning, however,
N
the fact that all isotherms for a given Sm concentration are described by the same parameters, shows
that the functional form of the isotherm is similar. Different functional form in the case of 3% Sm doping
A
may result from the formation of different surface sites because of additionally induced defects or partial
M

Sm segregation.

The other two parameters, cD/c0 and c/c0=csat(S/V)/c0, which are describing the acceptor concentration
ED

and concentration of adsorption centers, respectively, varied somewhat in different samples with same
Sm concentration. An example of this variation and its influence on the isotherm can be seen in Figure
7 for three samples with 0.5 at% Sm.
PT

For characterizing the sensitivity, a direct output quantity, such as the PL intensity or lifetime has to be
plotted against the gas concentration. Such information is gathered in Figure 8, where different relative
E

responses are compared for one sample. The term ‘relative’ means here that the intensity or lifetime is
CC

divided by the same quantity obtained in the absence of oxygen. Firstly, with square dots the PL intensity
is shown, as obtained from the type of data as depicted in Figure 4. In fact, the same information is
A

contained in the decay data as the stationary luminescence signal is generally given by an integral

𝑆 = ∫ 𝐼(𝑡) 𝑑𝑡 (8)
0

By calculating the areas under the decay curves, such as shown in Figure 5, a different set of data points
(represented by circular dots in Fig. 8) are obtained for the same sample. One can see that these data

9
points have higher values than those directly derived from stationary spectra, which is an unexpected
result as both datasets should represent the same physical quantity. This discrepancy is most probably
caused by saturation of the Sm excited state in case of stationary measurements. The average excitation
intensity was 20 times higher in these measurements as compared to decay measurements. In case of
such inequality of experimental conditions the relative signals in Fig. 8 should be smaller for more
saturating conditions i.e. for higher excitation intensity and longer excited state lifetime. As the lifetime
is increasing with oxygen concentration, the slope of the PL relative intensity of the stationary signal is
smaller in Fig. 8.

T
So far the comparison of the data obtained directly from the stationary measurements and from decay

IP
areas was independent of a concrete model. Within our model, a following equation can be derived in
case of Förster’s transfer (Appendix I):

R
𝐼0 + 𝐼1
𝑆= [1 − √𝜋𝑞 exp(𝑞 2 ) erfc(𝑞)]

SC
(9)
𝑘0

where we used a (conventional) notation 2q=c/c0. The expression in the brackets is a well-known result

U
for luminescence yield in case of resonant energy transfer [50]. As one can see, the equation (9) does
not include any detailed data on electron traps causing the delayed PL. Consequently, the delayed
N
excitation complicates the decay function and increases the average lifetime but does not alter the
A
stationary luminescence signal and its dependence on the oxygen pressure.
M

Using the equation (9) together with equations (6) and (7) a theoretical curve can be drawn for
comparison in Fig. 8 (dashed line). One can see that the model curve is in reasonable agreement with
ED

the (circular) data points obtained directly by calculating the areas below the decay curves according to
equation (8). This confirms our previous (qualitative) argument [26] that the luminescence quenching
by energy transfer is the only process involved in Sm:TiO2 and the excitation processes leading to the
PT

population of the Sm3+ 4G5/2 level are not altered by oxygen adsorption.

This, indeed, allows to realize a lifetime-based sensing with this material, a detection mode, which is
E

generally considered as a more stable one as compared to intensity measurements. A single quantity
CC

characterizing the PL decay is the average decay time, defined by equation:

∞
∫0 𝑡 × 𝐼(𝑡) 𝑑𝑡
〈𝜏〉 = ∞ (10)
∫0 𝐼(𝑡)𝑑𝑡
A

In Figure 8, the values of average lifetimes relative to <>0 are given by triangular dots. The absolute
values of average lifetime varied for a given 0.5 at% doped sample between <>0=283 s in pure
nitrogen atmosphere to 598 s in pure oxygen atmosphere.

10
The data in Figure 8 are presented in log-log axes. One can see that both quantities, I/I0 and <>/<>0,
depend quite slowly from oxygen concentration, whereas the sensitivity is bigger in the trace oxygen
range (<1 vol%). In the limited ranges, above or below 1% oxygen content, the dependencies of PL
intensities can be approximated to power laws with exponent between 0.15 and 0.27, respectively.

The overall qualitative scheme of the excitation-emission process in TiO2:Sm3+ is illustrated in Figure
9. Absorption of UV photons excites electrons from valence to conduction band. Some of the electrons
become trapped in the above-mentioned shallow traps whereas the rest of the electron-hole pairs
instantly form trapped exciton states which excite Sm3+ ions after a non-radiative energy transfer. An

T
excited Sm3+ ion either emits a photon or is quenched by a quenching defect (processes marked with (a)

IP
and (b) in Figure 9, respectively). At the same time, electrons released from the shallow traps (after
some delay) continue to excite the PL centers with a decreasing rate. The mentioned quenching defect

R
is probably a specific lattice defect (such as an F-center or charged oxygen vacancy [48]), which can be

SC
switched between two charge states. However, it causes significant quenching of PL only in a certain
charge state which energy level structure allows transitions in resonance with Sm3+. According to the
literature survey in Ref. [51] both oxygen vacancies (VO/V+O) or Ti intersititials (Tiint/Ti2+int/Ti3+int) can

U
have such energy levels in different charge states. The oxygen vacancies are more probable in our Sm-
N
doped samples as compensating more easily the charge mismatch between Sm3+ and Ti4+. The charge
state of this defect is controlled by the ambient oxygen concentration because adsorption (desorption)
A
of an oxygen molecule at the surface captures (releases) an electron which is exchanged with a
M

quenching defect.

As Fig. 9 demonstrates, there is a complex sequence of processes determining the PL signal. The
ED

excitation through band-to-band absorption with a following energy transfer to Sm is especially

beneficial for rare-earth based nanostructural or thin film optical sensors because it provides a large
optical absorption coefficient. The excitation process, however, is not influenced by oxygen in the
PT

studied material. The luminescence signal is related to the oxygen pressure via the energy transfer
parameter 2q=c/c0. This parameter varies between 2qmin=cD/c0-csat(S/V)/c0 (at high oxygen pressures)
E

and 2qmax= cD/c0 (in the absence of oxygen). Its span (and hence the overall sensitivity) can be controlled
CC

by adjusting the surface-to-volume ratio (S/V) or surface density of adsorption centers csat. The
concentration of the acceptor defects cD also influences the range of parameter q and hence the signal
conditions for oxygen detection as 2qmin>0 if csat(S/V)<cD.
A

Finally, it is important to consider possible future generalizations and elaborations of the theory for the
current class of luminescent gas sensor materials. Firstly, we note that the present model can also
describe the situation when luminescence efficiency is decreased with oxygen concentration (e.g., Ref.
[28]). This will occur when the internal defects are forming the energy acceptors in their oxidized state.
Formally, this situation is handled by reversing the signs in the right sides of equations (5) and (6).

11
Secondly, the model can be further developed for other oxidizing (or also reducing) gases. The type of
the gas will not alter the energy transfer process which directly influences the luminescence, but the
isotherms may be different and have to be established in each particular case. The effect of humidity is
also important for real-life sensors and have to be included in the theoretical description in the future.
The preliminary example of the effect of humidity on the luminescence response of the studied material
is given in the Supplementary Material.

T
Conclusions

IP
Current study thoroughly modeled the photoluminescence (PL) response of Sm3+ in TiO2 nanocrystals
to changes in ambient oxygen. It is shown that both the PL intensity and decay time of Sm ions can be

R
used for gaseous O2 detection in a wide concentration range starting from pure O2 down to trace amounts

SC
of less than 100 ppm.

Analysis of PL decay kinetics revealed that after a band-to-band optical excitation of TiO2 the excited

U
states of Sm3+ ions are populated either instantaneously or in a delayed manner through shallow electron
traps and subsequently quenched by internal defects via resonant energy transfer. The amount of active
N
quenching defects is controlled by oxygen adsorption to the nanocrystals, which is the core reason of
A
oxygen sensitivity.
M

The elaborated model can serve as an important analytical tool for the description of a full class of
luminescent materials as well as for further material development as it clearly relates general structural
and functional sensor characteristics. In other words, it can provide a framework for detailed studies and
ED

optimization of luminescence-based solid state gas sensors through adjustments of material inherent
properties (given by the model parameters kM, R, cD, csat, S/V) by doping conditions or modified
PT

preparation routes.
E
CC

Acknowledgement

This work was supported by institutional research funding IUT34-27 of the Estonian Ministry of
A

Education and Research. The authors are grateful to Peeter Ritslaid for measuring the content of Cl in
the samples.

12
Appendix I. Luminescence in the presence of energy transfer and delayed
excitation

We consider an ensemble of identical PL centers surrounded by a random distribution of resonant energy

acceptors. After instantaneous excitation of the PL centers the observed PL decay follows the law [36]:

𝐷
𝐼instant (𝑡) = 𝐼0 exp[−𝑘0 𝑡 − 2𝑞(𝑘0 𝑡)𝛽 ] , 𝛽= (𝐴1)
𝑛
where k0 is the natural decay rate, 𝐷 is the Euclidean dimensionality of the acceptor space, 𝑛 depends

T
on the type of interaction responsible for the energy transfer (e.g, n=6 for dipolar and n=8 for dipole-
quadrupole interaction) and 𝐼0 is proportional to the number of initially excited PL centers. 2𝑞 = 𝑐⁄𝑐0

IP
is a relative concentration of acceptors, where 1⁄𝑐0 = 𝛼𝑅30 . Here 𝑅0 is the characteristic radius of

R
energy transfer and is a constant (close to unity) which includes a contribution from the mutual

SC
orientations of donor and acceptors. For 𝐼0 = 1 we denote the resulting normalized decay profile as

𝛽
𝑢(𝑡) = 𝑒 −𝑘0 𝑡−(𝑐/𝑐0 )(𝑘0 𝑡) (A2)

U
The observed slow tail of the decay curve implies that a part of PL centers are excited with a delay. This
N
occurs as the initially trapped photocreated charge carriers of the host are gradually released by thermal
agitation. If there were just single type of such traps and the retrapping probability were negligible, then
A
the number of trapped carriers would decay exponentially as 𝐼1 𝑒 −𝑘𝑡 , where 𝐼1 is proportional to the
M

initial number of trapped carriers and 𝑘 is a rate constant of the release process. Correspondingly,
additional excited PL centers would be created at a time-dependent rate of
ED

𝑑 (A3)
𝑟(𝑡) = − (𝐼 𝑒 −𝑘𝑡 ) = 𝐼1 𝑘𝑒 −𝑘𝑡 .
𝑑𝑡 1
PT

During a short time interval 𝑡 … 𝑡 + 𝑑𝑡 an amount 𝑟(𝑡)𝑑𝑡 of new PL centers are excited. These PL
centers again start to decay according to the law 𝑢(𝑡) (provided that the acceptor distribution doesn’t
E

change and there is no spatial correlation between the traps and acceptors). Hence, the resulting decay
CC

would be convolution of the two processes:

𝑡 𝑡
′ ′ )−(𝑐/𝑐 )[𝑘 (𝑡−𝑡 ′ )]𝛽
𝐼delayed (𝑡) = 𝐼1 ∫ 𝑟(𝑡 ′ )𝑢(𝑡 − 𝑡 ′ ) 𝑑𝑡 ′ = 𝐼1 ∫ 𝑘𝑒 −𝑘𝑡 𝑒 −𝑘0 (𝑡−𝑡 0 0 𝑑𝑡 ′ . (𝐴4)
0 0
A

Taking into the account the presence of both, instantaneous and delayed excitation processes, the decay
law can be written as follows

𝐼(𝑡) = 𝐼𝑖𝑛𝑠𝑡𝑎𝑛𝑡 + 𝐼𝑑𝑒𝑙𝑎𝑦𝑒𝑑 = 𝐼0 𝑢(𝑡) + 𝐼1 𝑘𝑒 −𝑘𝑡 ⊗ 𝑢(𝑡) (A5)

where the ⊗ symbol marks convolution.

13
We further assume the presence of traps with different depths, which leads to the distribution of delayed
population rates. We consider the rate constant 𝑘 as a continuous variable and correspondingly introduce
a probability density function 𝜌(𝑘), so that the product 𝜌(𝑘)𝑑𝑘 gives the fraction of traps which populate
PL centers at the rate of 𝑘 … 𝑘 + 𝑑𝑘. Again, assuming negligible retrapping, each trap independently
contributes to the PL in the form of Eqn. (A4). Therefore, the resulting delayed PL decay would be an
average of Eqn. (A4) over the distribution 𝜌(𝑘). An average of a 𝑘-dependent quantity, for example
∞
𝐼(𝑘), is ∫0 𝐼(𝑘)𝜌(𝑘)𝑑𝑘, hence from Eqn. (A4) we obtain

∞ 𝑡

T
′ ′ )−(𝑐/𝑐 )[𝑘 (𝑡−𝑡 ′ )]𝛽
𝐼delayed (𝑡) = 𝐼1 ∫ 𝑑𝑘 𝜌(𝑘) ∫ 𝑑𝑡 ′ 𝑘𝑒 −𝑘𝑡 𝑒 −𝑘0 (𝑡−𝑡 0 0 (𝐴6)
0 0

IP
The equation (A5) is now generalized to equation

R
∞
(A7)
𝐼(𝑡) = 𝐼0 𝑢(𝑡) + 𝐼1 ∫ 𝑑𝑘 𝜌(𝑘) 𝑘𝑒 −𝑘𝑡 ⊗ 𝑢(𝑡)

SC
0

The double integral in the second term of equation (A7) makes it inefficient for fitting the experimental

U
decay curves. Note that after swapping the order of integration in (A6), the equation (A7) can be
rewritten as
N
𝐼(𝑡) = 𝐼0 𝑢(𝑡) + 𝐼1 𝐹(𝑡) ⊗ 𝑢(𝑡) (A8)
A

where
M

∞
(A9)
𝐹(𝑡) = ∫ 𝑑𝑘 𝜌(𝑘) 𝑘𝑒 −𝑘𝑡
ED

If the function F(t) can be presented in an analytical form, then only a single integral remains to be
evaluated in Eq. (A8) at fitting the experimental data. It turns out that for the particular distribution
PT

derived in Appendix II this condition is met (see equation (A17)).

For data analysis, it is also useful to derive the connection between the decay signal at short-pulse
E

excitation and stationary luminescence at cw excitation. Assuming that no saturation is present, the
CC

stationary signal is given by

𝑆 = ∫ 𝐼(𝑡) 𝑑𝑡
A

(A10)
0

By substituting (A8) into (A10) one can write S=S0+S1, where in case of Förster’s energy transfer
(=0.5)

14
 ∞
𝐼0
𝑆0 = 𝐼0 ∫ 𝑢(𝑡) 𝑑𝑡 = [1 − √𝜋𝑞 exp(𝑞 2 ) erfc(𝑞)] (A11)
𝑘0
0

∞ 𝑡 ∞ ∞
′ ′ ′ ′ ′
𝑆1 = 𝐼1 ∫ 𝑑𝑡 ∫ 𝑑𝑡 𝐹(𝑡 )𝑢(𝑡 − 𝑡 ) = 𝐼1 ∫ 𝑑𝑡 𝐹(𝑡 ) ∫ 𝑑𝑡 𝑢(𝑡) (A12)
0 0 0 0

∞ ∞
The first integral in the last expression ∫0 𝑑𝑡 ′ 𝐹(𝑡 ′ ) = ∫0 𝑑𝑘𝜌(𝑘) and equals to unity due to the

normalization of the distribution (k). Hence the final result for the stationary PL is

T
𝐼0 + 𝐼1
𝑆= [1 − √𝜋𝑞 exp(𝑞 2 ) erfc(𝑞)] (A13)
𝑘0

R IP
Appendix II. Distribution of delayed population rates

SC
We consider that the binding energies of the trapped charge carriers (the energies needed to untrap the
carrier) follow an exponential distribution [45]:

𝜌𝐸 (𝐸) =
1
Δ𝐸
𝐸
exp (− )
Δ𝐸
U (A14)
N
where Δ𝐸 characterizes the spread of the distribution. A basic physical consideration implies that the
A
probability of de-trapping follows the Arhhenius law, so that the rate constant
M

𝐸 (A15)
𝑘 = 𝑘𝑀 exp (− )
𝑘B 𝑇
ED

where 𝑘B is the Boltzmann constant, 𝑇 is temperature and kM is an attempt-to-escape frequency.

For simplicity, we do not constrain the binding energy, hence 𝐸 = 0 … ∞ and the rate constant is
PT

respectively k=0…kM. Now we have to make a transition from the random variable 𝐸 with a distribution
𝜌(𝐸) to the random variable 𝑘 with a distribution 𝜌(𝑘). If 𝐸 … 𝐸 + 𝑑𝐸 and 𝑘 … 𝑘 + 𝑑𝑘 represent
corresponding infinitesimal intervals, the probabilities over the intervals must be equal: 𝜌(𝐸)𝑑𝐸 =
E

𝜌(𝑘)𝑑𝑘. Assuming 𝑘𝑀 ≈ const, one can easily evaluate the derivative 𝑑𝐸⁄𝑑𝑘 and thereby find the 𝑘-
CC

distribution:

𝑅 𝑘 𝑅 (A16)
A

𝜌(𝑘) = ( )
𝑘 𝑘𝑀

where 𝑅 = 𝑘B 𝑇⁄Δ𝐸 . Hence, the two parameters 𝑠 and 𝑅 (or Δ𝐸) determine the scale and shape of the
distribution 𝜌(𝑘).

In the case of distribution (A16) one can analytically evaluate the integral

15
 ∞
𝑅 Γ(𝑅 + 1) − Γ(𝑅 + 1, 𝑘𝑀 𝑡) (A17)
𝐹(𝑡) = ∫ 𝑑𝑘 𝜌(𝑘) 𝑘𝑒 −𝑘𝑡 = ,
0 (𝑘𝑀 )𝑅 𝑡𝑅+1

where Γ denotes the (incomplete) gamma function.

References

[1] X. Wang, O.S. Wolfbeis, Optical methods for sensing and imaging oxygen: materials,
spectroscopies and applications, Chem. Soc. Rev. 43 (2014) 3666–3761.
doi:10.1039/C4CS00039K.
[2] O.S. Wolfbeis, Luminescent sensing and imaging of oxygen: Fierce competition to the Clark

T
electrode, BioEssays 37 (2015) 921–928. doi:10.1002/bies.201500002.

IP
[3] M. Quaranta, S.M. Borisov, I. Klimant, Indicators for optical oxygen sensors, Bioanal Rev. 4
(2012) 115–157. doi:10.1007/s12566-012-0032-y.
[4] J.W. Gregory, H. Sakaue, T. Liu, J.P. Sullivan, Fast Pressure-Sensitive Paint for Flow and

R
Acoustic Diagnostics, Annu. Rev. Fluid Mech. 46 (2014) 303–330. doi:10.1146/annurev-fluid-
010313-141304.

SC
[5] R. Jaaniso, O.K. Tan (Eds), Semiconductor gas sensors, Woodhead, Cambridge, 2013.
[6] G. Faglia, C. Baratto, G. Sberveglieri, M. Zha, A. Zappettini, Adsorption effects of NO2 at ppm

U
level on visible photoluminescence response of SnO2 nanobelts, Appl. Phys. Lett. 86 (2005)
011923. doi:10.1063/1.1849832.
N
[7] A. Setaro, A. Bismuto, S. Lettieri, P. Maddalena, E. Comini, S. Bianchi, C. Baratto, G.
Sberveglieri, Optical sensing of NO2 in tin oxide nanowires at sub-ppm level, Sens. Actuators B:
A
Chem. 130 (2008) 391–395. doi:10.1016/j.snb.2007.09.015.
[8] V.M. Zhyrovetsky, D.I. Popovych, S.S. Savka, A.S. Serednytski, Nanopowder Metal Oxide for
M

Photoluminescent Gas Sensing, Nanoscale Res. Lett. 12 (2017) 132. doi:10.1186/s11671-017-

1891-5.
[9] E. Comini, C. Baratto, G. Faglia, M. Ferroni, G. Sberveglieri, Single crystal ZnO nanowires as
ED

optical and conductometric chemical sensor, J. Phys. D: Appl. Phys. 40 (2007) 7255.
doi:10.1088/0022-3727/40/23/S08.
[10] D. Valerini, A. Cretì, A.P. Caricato, M. Lomascolo, R. Rella, M. Martino, Optical gas sensing
PT

through nanostructured ZnO films with different morphologies, Sens. Actuators B: Chem. 145
(2010) 167–173. doi:10.1016/j.snb.2009.11.064.
[11] R. Aad, V. Simic, L.L. Cunff, L. Rocha, V. Sallet, C. Sartel, A. Lusson, C. Couteau, G.
E

Lerondel, ZnO nanowires as effective luminescent sensing materials for nitroaromatic

derivatives, Nanoscale 5 (2013) 9176–9180. doi:10.1039/C3NR02416D.
CC

[12] J.R. Sanchez-Valencia, M. Alcaire, P. Romero-Gómez, M. Macias-Montero, F.J. Aparicio, A.

Borras, A.R. Gonzalez-Elipe, A. Barranco, Oxygen Optical Sensing in Gas and Liquids with
Nanostructured ZnO Thin Films Based on Exciton Emission Detection, J. Phys. Chem. C. 118
(2014) 9852–9859. doi:10.1021/jp5026027.
A

[13] A. R. Gheisi, C. Neygandhi, A. K. Sternig, E. Carrasco, H. Marbach, D. Thomele, O. Diwald, O2

adsorption dependent photoluminescence emission from metal oxide nanoparticles, Phys. Chem.
Chem. Phys. 16 (2014) 23922–23929. doi:10.1039/C4CP03080J.
[14] X. Liu, Y. Sun, M. Yu, Y. Yin, B. Yang, W. Cao, M.N.R. Ashfold, Incident fluence dependent
morphologies, photoluminescence and optical oxygen sensing properties of ZnO nanorods
grown by pulsed laser deposition, J. Mater. Chem. C. 3 (2015) 2557–2562.
doi:10.1039/C4TC02924K.

16
[15] M. Madel, J. Jakob, F. Huber, B. Neuschl, S. Bauer, Y. Xie, I. Tischer, K. Thonke, Optical gas
sensing by micro-photoluminescence on multiple and single ZnO nanowires, Phys. Status Solidi
A 212 (2015) 1810–1816. doi:10.1002/pssa.201431688.
[16] W. Shirbeeny, W.E. Mahmoud, Synthesis and characterization of transparent optical gas sensor
device made of indium oxide pyramid like nanoarchitectures, Sens. Actuators B: Chem. 191
(2014) 102–107. doi:10.1016/j.snb.2013.09.058.
[17] P. Wang, T. Xie, L. Peng, H. Li, T. Wu, S. Pang, D. Wang, Water-Assisted Synthesis of Anatase
TiO2 Nanocrystals:  Mechanism and Sensing Properties to Oxygen at Room Temperature, J.
Phys. Chem. C. 112 (2008) 6648–6652. doi:10.1021/jp800409f.
[18] D. Pallotti, E. Orabona, S. Amoruso, P. Maddalena, S. Lettieri, Modulation of mixed-phase
titania photoluminescence by oxygen adsorption, Appl. Phys. Lett. 105 (2014) 031903.

T
doi:10.1063/1.4891038.

IP
[19] F. Fujishiro, S. Mochizuki, Reversible photo-induced spectral change and defect creation in
ZrO2, Phys. Status Solidi (c) 6 (2009) 354–357. doi:10.1002/pssc.200879818.

R
[20] S. Mochizuki, T. Saito, K. Yoshida, UV-laser-light-controlled photoluminescence of metal oxide
nanoparticles in different gas atmospheres: BaTiO3, SrTiO3 and HfO2, Physica B: Condens.

SC
Matter. 407 (2012) 2889–2894. doi:10.1016/j.physb.2011.08.057.
[21] N. Siedl, D. Koller, A.K. Sternig, D. Thomele, O. Diwald, Photoluminescence quenching in
compressed MgO nanoparticle systems, Phys. Chem. Chem. Phys. 16 (2014) 8339–8345.
doi:10.1039/C3CP54582B.

U
[22] S. Mochizuki, H. Araki, Reversible photoinduced spectral transition in Eu2O3–γAl2O3
N
composites at room temperature, Physica B: Condens. Matter. 340 (2003) 913–917.
doi:10.1016/j.physb.2003.09.204.
A
[23] V. Reedo, S. Lange, V. Kiisk, A. Lukner, T. Tätte, I. Sildos, Influence of ambient gas on the
photoluminescence of sol-gel derived TiO2:Sm3+ films, in: International Society for Optics and
M

Photonics, 2006: p. 59460F. doi:10.1117/12.639162.

[24] T. Tachikawa, T. Ishigaki, J.-G. Li, M. Fujitsuka, T. Majima, Defect-Mediated
Photoluminescence Dynamics of Eu3+-Doped TiO2 Nanocrystals Revealed at the Single-Particle
ED

or Single-Aggregate Level, Angew. Chem. Int. Ed. 47 (2008) 5348–5352.

doi:10.1002/anie.200800528.
[25] W. Di, X. Wang, X. Ren, Nanocrystalline CePO4:Tb as a novel oxygen sensing material on the
basis of its redox responsive reversible luminescence, Nanotechnology 21 (2010) 075709.
PT

doi:10.1088/0957-4484/21/7/075709.
[26] M. Eltermann, K. Utt, S. Lange, R. Jaaniso, Sm3+ doped TiO2 as optical oxygen sensor material,
Optical Materials 51 (2016) 24–30. doi:10.1016/j.optmat.2015.11.020.
E

[27] N.A.F. Almeida, J. Rodrigues, P. Silva, N. Emami, M.J. Soares, T. Monteiro, J.A. Lopes-da-
CC

Silva, P.A.A.P. Marques, Pressure dependent luminescence in titanium dioxide particles

modified with europium ions, Sens Actuators B: Chem. 234 (2016) 137–144.
doi:10.1016/j.snb.2016.04.157.
[28] I. Aydin, K. Ertekin, S. Demirci, S. Gultekin, E. Celik, Sol-gel synthesized Sr4Al14O25:Eu2+/Dy3+
A

blue–green phosphorous as oxygen sensing materials, Optical Materials 62 (2016) 285–296.

doi:10.1016/j.optmat.2016.10.019.
[29] W. Tang, Y. Sun, S. Wang, B. Du, Y. Yin, X. Liu, B. Yang, W. Cao, M. Yu, Pr3+-doped (K0.5
Na0.5 )NbO3 as a high response optical oxygen sensing agent, J. Mater. Chem. C. 4 (2016)
11508–11513. doi:10.1039/C6TC04216C.
[30] H. Weng, X.-Y. Xu, B. Yan, Novel multi-component photofunctional nanohybrids for ratio-
dependent oxygen sensing, J. Colloid Interface Sci. 502 (2017) 8–15.
doi:10.1016/j.jcis.2017.04.081.

17
[31] L. Puust, V. Kiisk, M. Eltermann, H. Mändar, R. Saar, S. Lange, I. Sildos, Leonid Dolgov, L.
Matisen, R. Jaaniso, Effect of ambient oxygen on the photoluminescence of sol–gel-derived
nanocrystalline ZrO2:Eu,Nb, J. Phys. D: Appl. Phys. 50 (2017) 215303. doi:10.1088/1361-
6463/aa6c48.
[32] I.R. Beattie, T.R. Gilson, Single Crystal Laser Raman Spectroscopy, Proceedings of the Royal
Society of London A: Mathematical, Physical and Engineering Sciences. 307 (1968) 407–429.
doi:10.1098/rspa.1968.0199.
[33] O. Frank, M. Zukalova, B. Laskova, J. Kürti, J. Koltai, L. Kavan, Raman spectra of titanium
dioxide (anatase, rutile) with identified oxygen isotopes (16, 17, 18), Phys. Chem. Chem. Phys.
14 (2012) 14567–14572. doi:10.1039/C2CP42763J.
[34] S.K. Gupta, R. Desai, P.K. Jha, S. Sahoo, D. Kirin, Titanium dioxide synthesized using titanium

T
chloride: size effect study using Raman spectroscopy and photoluminescence, J. Raman
Spectrosc. 41 (2010) 350–355. doi:10.1002/jrs.2427.

IP
[35] G.H. Dieke, Spectra and Energy Levels of Rare Earth Ions in Crystals, John Wiley & Sons,
1968.

R
[36] A. Blumen, J. Manz, On the concentration and time dependence of the energy transfer to

SC
randomly distributed acceptors, J. Chem. Phys. 71 (1979) 4694–4702. doi:10.1063/1.438253.
[37] V. Kiisk, M. Šavel, V. Reedo, A. Lukner, I. Sildos, Anatase-to-rutile phase transition of
samarium-doped TiO2 powder detected via the luminescence of Sm3+, Phys. Procedia 2 (2009)
527–538. doi:10.1016/j.phpro.2009.07.038.

U
[38] Ž. Antić, R.M. Krsmanović, M.G. Nikolić, M. Marinović-Cincović, M. Mitrić, S. Polizzi, M.D.
N
Dramićanin, Multisite luminescence of rare earth doped TiO2 anatase nanoparticles, Mater.
Chem. Phys. 135 (2012) 1064–1069. doi:10.1016/j.matchemphys.2012.06.016.
A
[39] W. Luo, R. Li, X. Chen, Host-Sensitized Luminescence of Nd3+ and Sm3+ Ions Incorporated in
Anatase Titania Nanocrystals, J. Phys. Chem. C. 113 (2009) 8772–8777. doi:10.1021/jp901862k.
M

[40] H. Tang, H. Berger, P.E. Schmid, F. Lévy, G. Burri, Photoluminescence in TiO2 anatase single
crystals, Solid State Commun. 87 (1993) 847–850. doi:10.1016/0038-1098(93)90427-O.
[41] K.L. Frindell, M.H. Bartl, M.R. Robinson, G.C. Bazan, A. Popitsch, G.D. Stucky, Visible and
ED

near-IR luminescence via energy transfer in rare earth doped mesoporous titania thin films with
nanocrystalline walls, J. Solid State Chem. 172 (2003) 81–88. doi:10.1016/S0022-
4596(02)00126-3.
PT

[42] H. Tang, F. Lévy, H. Berger, P.E. Schmid, Urbach tail of anatase TiO2, Phys. Rev. B. 52 (1995)
7771–7774. doi:10.1103/PhysRevB.52.7771.
[43] A. Yamakata, T. Ishibashi, H. Onishi, Time-resolved infrared absorption spectroscopy of
E

photogenerated electrons in platinized TiO2 particles, Chem. Phys. Lett. 333 (2001) 271–277.
doi:10.1016/S0009-2614(00)01374-9.
CC

[44] L.J. Antila, F.G. Santomauro, L. Hammarström, D.L.A. Fernandes, J. Sá, Hunting for the elusive
shallow traps in TiO2 anatase, Chem. Commun. 51 (2015) 10914–10916.
doi:10.1039/C5CC02876K.
A

[45] A. Rose, Space-Charge-Limited Currents in Solids, Phys. Rev. 97 (1955) 1538–1544.

doi:10.1103/PhysRev.97.1538.
[46] M. Malinowski, B. Jacquier, G. Boulon, W. Wolinski, Fluorescence quenching in Sm3+ doped
KYP4O12 crystals, J. Luminescence 39 (1988) 301-311. doi:10.1016/0022-2313(88)90011-7
[47] E. Setiawati, K. Kawano, T. Tsuboi, H. J. Seo, Studies on thermal migration of Eu ion doped
into TiO2 nanoparticles, Jpn. J. Appl. Phys. 47 (2008) 4651-4657. doi:10.1143/JJAP.47.4651
[48] P. Deák, B. Aradi, T. Frauenheim, Quantitative theory of the oxygen vacancy and carrier self-
trapping in bulk TiO2, Phys. Rev. B. 86 (2012) 195206. doi:10.1103/PhysRevB.86.195206.

18
[49] J. Tóth, Uniform interpretation of gas/solid adsorption, Adv. Colloid Interface Sci. 55 (1995) 1–
239. doi:10.1016/0001-8686(94)00226-3.
[50] T. Förster, Experimentelle und theoretische Untersuchung des zwischenmolekularen Übergangs
von Elektronenanregungsenergie, Z. Naturforsch. A. 4 (2014) 321–327. doi:10.1515/zna-1949-
0501.
[51] B. Weiler, A. Gagliardi, P. Lugli, Kinetic Monte Carlo Simulations of Defects in Anatase
Titanium Dioxide, J. Phys. Chem. C 2016, 120 (2016) 10062−10077. doi:
10.1021/acs.jpcc.6b01687

T
R IP
SC
U
N
A
M
ED
E PT
CC
A

19
Biographies

Marko Eltermann received his Master’s degree in physics from the University of Tartu in 2014. He is
currently pursuing a PhD degree in the University of Tartu, Institute of Physics. His research is focused
on the optical gas sensing properties of doped TiO2 material.

Valter Kiisk received his PhD in solid-state physics from the University of Tartu in 2006. His research
activities mainly involve spectroscopic studies and applications of intrinsic and rare earth luminescence
in various metal-oxide nanomaterials. Currently he is a senior research fellow at the University of Tartu,

T
exploring rare earth activated oxide materials for luminescent gas sensing.

IP
Raivo Jaaniso received his PhD in solid-state physics from the Institute of Physics at the Estonian

R
Academy of Sciences in 1988. His research interests have covered a wide area from laser spectroscopy

SC
and spectral hole burning to pulsed laser deposition of thin films and, most recently, development of
luminescent and semiconductor gas sensor materials. He is the head of the laboratory of sensor
technologies at the University of Tartu.

U
N
A
M
ED
E PT
CC
A

20
 T
Figure 1. SEM image of the TiO2:Sm3+ Figure 2. Raman spectrum of the TiO2:Sm3+

IP
nanopowder. nanopowder.

R
SC
U
N
A
M
ED

Figure 3. PL emission spectra of TiO2:Sm3+ (3 at%) nanopowder (excited by 355 nm laser) at three
different oxygen concentrations (0%, 1% and 100%) in the O2/N2 mixture at normal pressure. The
PT

inset shows the area that was integrated in order to obtain a single numeric parameter to describe
the intensity of the Sm3+ impurity emission.
E
CC
A

21
 6

Relative PL intensity I/I0

90 %
~1min ~5min
5

90 %
4

0.1 % O2
3

0.01 % O2
100 % O2

10 % O2
2

T
1 % O2

IP
1
0 20 40 60 80 100

R
Time (min)

SC
Figure 4. The temporal response curves for a set of gas exchange cycles in a wide oxygen
concentration range of 100–0.01 vol% measured at 25 °C. Inset shows characteristic response times
during one cycle.

U
N
A
M
ED
E PT
CC
A

Figure 5. PL decay kinetics of the 3 % (a) and 0.5 % (b) TiO2:Sm3+ sample in a wide oxygen
concentration range. The colored dots represent experimental data, solid black lines theoretical fits (Eq.
3) and the dashed straight line an exponential decay with characteristic time 300 µs.

22
 T
R IP
SC
Figure 6. Experimental PL decay of TiO2:Sm3+ (0.5 at%) nanopowder at 1% oxygen content (dotted
U
curve) compared to theoretical curves corresponding to simple exponent, equations (1), (2), and (3).
N
The parameters obtained at the final fitting with equation (3) were used for all theoretical curves
A
shown in the figure.
M

cD/c0= Toth isoterms:

4
ED

m = 0.28
5.0 K = 1700 (1/vol%)

3
c/c0=
PT

3.8
c/c0

2 3.4 3.8
E

2.8
CC

1
3.4

0
A

0.01 0.1 1 10 100

Oxygen content (vol%)

Figure 7. Acceptor concentration c/c0 as a function of oxygen content derived from the fitting of PL
decays. The data for 3 different samples all doped with 0.5 at% of Sm are shown. The smooth curves
describe Tóth isotherms found by global fitting with the common values of parameters m and K.

23
 5

I/I0, <>/<>0 3

T
IP
1
0.01 0.1 1 10 100

R
Oxygen content (%)

SC
Figure 8. Various types of TiO2:Sm3+ sensor relative responses as the functions of oxygen
concentration. The PL intensity data were obtained directly from stationary measurements (square
dots) or from the areas below the decay signals (round dots). The model curve given by equation (9)

U
is shown by the dotted line. Triangular dots indicate the sensor signal obtained by calculating the
N
average lifetimes from the decay curves according to eq. (10).
A
M
ED
E PT
CC
A

Figure 9. A simplified energy diagram of the TiO2:Sm3+ sensor material. The Sm3+ ion is exited either
instantly (blue dashed line) or in a delayed manner (red dashed line). Once the excitation has reached
the Sm3+ ion it can (a) emit a photon or (b) be quenched by a defect. This defect can be ‘switched off’
by electron transfer to surface oxygen species.

24
Table I. Model parameters obtained by fitting the experimental decay curves and isotherms of defect
concentrations.

[Sm]  0  R kM I1/I0 2 K m
at% ms 1/ms 1/vol%
0.5 0.300 1/2 0.8±0.3 4.5±2.0 1.4±0.6 1.1-1.6 1700 0.28
3 0.300 3/8 0.7±0.1 3.5±1.3 2.0±0.3 1.5-2.0 100 0.45

T
R IP
SC
U
N
A
M
ED
E PT
CC
A

25