Tailoring the electron and hole Landé factors in lead halide perovskite nanocrystals by

quantum confinement and halide exchange

Mikhail O. Nestoklon1 ,∗ Erik Kirstein1 , Dmitri R. Yakovlev1,2 ,† Evgeny A. Zhukov1 , Mikhail M. Glazov2 ,
Marina A. Semina2 , Eugeniyus L. Ivchenko2 , Elena V. Kolobkova3,4 , Maria S. Kuznetsova5 , and Manfred Bayer1
1
Experimentelle Physik 2, Technische Universität Dortmund, 44227 Dortmund, Germany
2
Ioffe Institute, Russian Academy of Sciences, 194021 St. Petersburg, Russia
3
ITMO University, 199034 St. Petersburg, Russia
4
St. Petersburg State Institute of Technology, 190013 St. Petersburg, Russia
5
Spin Optics Laboratory, St. Petersburg State University, 198504 St. Petersburg, Russia
arXiv:2305.10586v1 [cond-mat.mes-hall] 17 May 2023

The tunability of the optical properties of lead halide perovskite nanocrystals makes them highly
appealing for applications. Both, halide anion exchange and quantum confinement pave the way
for tailoring their band gap energy. For spintronics applications, the Landé g-factors of electrons
and hole are of great importance. By means of the empirical tight-binding and k·p methods, we
calculate them for nanocrystals of the class of all-inorganic lead halide perovskites CsPbX3 (X =
I, Br, Cl). The hole g-factor as function of the band gap follows the universal dependence found for
bulk perovskites, while for the electrons a considerable modification is predicted. Based on the k·p
analysis we conclude that this difference arises from the interaction of the bottom conduction band
with the spin-orbit split electron states. The model predictions are confirmed by experimental data
for the electron and hole g-factors in CsPbI3 nanocrystals placed in a glass matrix, measured by
time-resolved Faraday ellipticity in a magnetic field at cryogenic temperatures.

I. INTRODUCTION tracted a lot of attention during the last decade due to

their excellent optical properties and technologically sim-
The development of quantum technologies and spin- ple synthesis [16–19]. The quantum confinement in NCs
tronics calls for a search of suitable material platforms. strongly affects the electronic and optical properties. It
The recently emerging lead halide perovskite semicon- is known, that in conventional semiconductor quantum
ductors [1, 2] offer a new testbed for spin-dependent phe- dots the confinement leads to a renormalization of the
nomena. Among them, the optical orientation of charge carriers’ g-factors [20]. This effect has been studied in
carrier spins, their coherent spin precession, spin mode- nanostructures of III–V and II–VI semiconductors and is
locking, spin-flip Raman scattering, and dynamic nu- well-understood from the k·p theory [21] and atomistic
clear polarization have been demonstrated for bulk lead calculations [22]. Universal dependences of the g-factors
halide perovskites [3–9] and their nanocrystals [10–14]. on the band gap have been demonstrated for quantum
Information on the electron and hole Landé factors (g- wells [23, 24] and III–V quantum dots [22], however, in
factors), which determine the Zeeman splitting of their a much narrower range of band gap variations compared
spin states in an external magnetic field, is of key im- to bulk perovskites [9]. All that calls for an investigation
portance for understanding the spin-dependent phenom- of the g-factors in perovskite NCs.
ena and for resulting spintronic applications. The mag- Modern calculation methods allow one to compute the
netic field allows one to unravel otherwise hidden infor- electronic band structure of bulk semiconductors with
mation on the band structure, e.g., by splitting degen- reasonable precision [25], but the use of DFT for nanos-
erate spin states and activating optically-forbidden dark tructures is challenging. For a qualitative analysis of
exciton states [15]. Electron and hole g-factors were mea- nanostructures with sizes of tens of nanometres, the ef-
sured for a set of bulk hybrid organic-inorganic and all- fective mass approximation [26] is the method of choice,
inorganic lead halide perovskites by pump-probe Kerr ro- while for nanostructures of few nm size empirical atom-
tation and spin-flip Raman scattering [9]. It was shown istic methods are preferable [27, 28]. The empirical tight-
that the carrier g-factors obey a universal dependence binding method is one of the simplest approximations
on the band gap energy across different perovskite ma- suitable for an accurate description of the band struc-
terial classes, which can be summarized in a universal ture of conventional semiconductors at the lowest pos-
semi-phenomenological expression. This empirical result sible computation cost [28]. The ETB method within
was corroborated by atomistic calculations based on the the sp3 d5 s∗ basis in the nearest neighbor approximation
combination of the density functional theory (DFT) and gives a precise description of the band structure of bulk
empirical tight-binding (ETB) methods. III–V [29] and group IV [30] semiconductors. It has been
Lead halide perovskite nanocrystals (NCs) have at- shown recently that this method can be used to model
the band structure of inorganic lead halide perovskites
with meV-range precision [31].
In this study we use the ETB method within the
∗ [email protected] sp3 d5 s∗ basis applying the nearest neighbor approxima-
† [email protected] tion to calculate the electron and hole g-factors in all-
 2

inorganic lead halide perovskite NCs based on CsPbI3 , edge length of the cubic NC, see Figure 1. This is in line
CsPbBr3 and CsPbCl3 . We show that the hole g-factors with the effective mass approximation and with the k·p
in NCs follow the empirical trend, which was first ob- calculations of CsPbI3 nanocrystals in Ref. 35. For all
served for bulk perovskites[9]. The electron g-factors three materials the confinement-induced band gap renor-
show a significant renormalization, which results in a de- malization is significant and reaches ∼ 1 eV for small
viation from the bulk empirical trend. Using the k·p nanocrystal sizes with a ∼ 3 nm. In the inset of Figure 1a
calculations we uncover the origin of this behavior and we compare the calculated band gap energy as function
show that it is due the quantum-confinement induced of the NC size with the corresponding dependence of the
admixing of the excited heavy- and light-electron states maximum energy of the Faraday ellipticity amplitude,
to the ground electron state. We measured the electron measured at low temperature (see below). The size of
and hole g-factors by means of time-resolved Faraday el- the NCs is evaluated via scanning transmission electron
lipticity on ensembles of CsPbI3 NCs of different sizes. microscopy (STEM) at room temperature, see Supple-
The measured g-factors are in good agreement with the mentary Information S4. The agreement between calcu-
theory predictions. lation and experiment is good, the difference (∼ 30 meV)
may be attributed to the neglected exciton binding en-
ergy, the effect of the dielectric contrast not included in
II. EMPIRICAL TIGHT-BINDING the modeling, or underestimated masses of the carriers in
CALCULATIONS our ETB parametrization. However, the determination
of the experimental band gap energy is also an involved
We use an ab initio inspired ETB method to compute problem as the photoluminescence line is much broader
the band gap and g-factors of charge carriers quantum- than the Faraday ellipticity spectrum for the measured
confined in cubic-shaped nanocrystals of CsPbX 3 (X = NCs.
I, Br, Cl) in the cubic phase. In the ETB method,[32] The electron and hole g-factors are calculated by in-
the i-th electron wave function Φi (r) is expanded in the troducing a weak magnetic field in the ETB Hamil-
basis of orthogonal atomic-like functions φα (r) : tonian (2) following a standard procedure by using
X the Peierls substitution[37, 38]: the vector-potential-
i
Φi (r) = Cnα φα (r − rn ) , (1) dependent phase, and the Zeeman term, µB (σ·B), where
nα µB is the Bohr magneton, σ is the vector of Pauli matri-
ces describing the spin and B is the magnetic field, are
where n enumerates the atoms with coordinates rn , the added to the off-diagonal and diagonal matrix elements
index α runs through different orbitals and spins, and of the tight-binding Hamiltonian (2), respectively.
i
Cnα are expansion coefficients. We use the sp3 d5 s∗ vari-
For bulk, the g-factors of different materials show
ant of the method [29] with twenty orbitals per atom. As
trends which may be understood in the framework of
demonstrated in Refs. 9 and 31, this method provides an
k·p theory.[9] The value of the bulk electron g-factors is
accurate description of the band structure of lead halide
mostly determined by a constant contribution (given by
perovskites. In Ref. 31 it is also demonstrated that the
the bare g-factor and the interaction with remote bands)
method gives a reasonable description of the Pb-plane
and the interaction with the valence band which changes
terminated surface without considering additional passi-
with the band gap. The value of the hole g-factors is
vation and reconstruction.
given by the constant bare g-factor and the interaction
We start from the calculation of the electron and hole
with the three low-lying conduction bands, the latter con-
states in NCs. In the basis (1), the Schrödinger equation
tribution is band-gap-dependent. The similar considera-
for the nanocrystal reduces to the eigenvalue problem for
tion may be applied for III-V semiconductors and used
a sparse matrix:
to account for the change of g-factors in nanostructures
X due to quantum confinement. The goal of our study is
Hnα,n0 ς Cni 0 ς = Ei Cnα
i
. (2)
to accurately check whether such generalization may be
n0 ,ς
done also for perovskite nanostructures.
Here Hnα,n0 ς are the matrix elements of the ETB Hamil- The resulting values of the g-factors of the electron and
tonian calculated following Ref. 33 and Ei is the energy of hole ground states as function of the effective band gap
the i-th electron state. The solutions of Eq. (2) near the in NCs are presented in Figure 2 by symbols. For a given
band-edge can be found using the Thick-Restart Lanczos material, with increasing Eg the electron g-factor de-
algorithm [34]. creases whereas the hole g factor increases. The g-factors
The effective band gap EgNC = Eg + Ee + Eh is the sum are isotropic, which is expected for cubic nanocrystals in
of the bulk material band gap Eg and the electron and the cubic phase. Note that we use the updated ETB
hole confinement energies, Ee and Eh , respectively. It is parametrization which better reproduces the g-factors of
extracted as the energy difference between the lowest con- bulk perovskites, see for details.
fined conduction band state and the uppermost confined In bulk lead halide perovskites, a universal dependence
valence band state, and depends on the NC size accord- of the charge carrier g-factors on the band gap has been
ing to an almost linear function of 1/a2 , where a is the found [9]. The hole g-factor in NCs only weakly devi-
 3

5 5
(a) (b)

EgNC (eV)
2
1.8 CsPbCl3
4 4
1.6 CsPbBr3
5 10 15
a (nm) CsPbI3
EgNC (eV)

EgNC (eV)
3 CsPbCl3 3
3×3×3
CsPbBr3
2 2 4×4×4
CsPbI3 5×5×5
1 1
0 2 4 6 8 10 12 14 0 0.1 0.2 0.3 0.4
a (nm) a (nm-2)
-2

FIG. 1. Calculated band gap of NCs as a function of NC edge length a (panel (a)) and of 1/a2 (panel (b)). In the ETB
calculations the integer number of monolayers is multiplied with the lattice constant a0 = 0.561 nm, 0.586 nm, 0.624 nm for
CsPbX3 with X = Cl, Br and I, respectively [36]. The crosses show the results of numerical calculations and the lines are
guides to the eye. In the inset, a comparison of the calculated band gap energies (red line) with experimental data for the
CsPbI3 NCs samples used in this study (blue dots) is shown. Details of the sample characterization are given in the Supporting
Information S4, the NC size was evaluated via STEM, and the energy is given by the maximum in Faraday ellipticity spectra
of Figure 3a. The error bars give the half width at half maximum of both the Faraday ellipticity spectra (Figure 3a) and the
STEM characterization (Supporting Information Figure S6).

ates from this universal curve shown by the dashed line band mixing. Both energy denominators and wh depend
in Figure 2b. On the other hand, the electron g-factor on the hole size-quantization energy.
shows a strong effect of the quantum confinement, drop- The significant renormalization of the conduction band
ping significantly with an increase of the effective band g-factors results from the complex conduction band
gap (a decrease of the NC size) and even changing sign structure: In lead halide perovskites the lowest conduc-
from positive to negative, see Figure 2a. tion band is formed by spin 1/2 states and the next
To provide a qualitative interpretation of the results bands represent heavy and light electrons with spin 3/2.
of the ETB calculations and reveal the origin of the de- The magneto-induced mixing with the spin-orbit split-
viation from the universal dependence, we extend the off heavy- and light-electron bands results in the elec-
k·p model[9] to account for the size quantization in per- tron g-factor renormalization. The calculation within the
ovskite NCs. For the hole g-factors, straightforward ap- k·p model, which explicitly accounts for the band mix-
plication of the Kane model for spherical NCs with infi- ing (Supporting Information S2), results in the following
nite barriers yields (see Supporting Information S2 and expression for the electron g-factor
Refs. 20 and 21 for details)
2 4 p2 we
ge (Ee ) = − + +∆gremote + δgeso , (4)
4 p2
 
1 1 3 3 m0 Eg + Ee
gh (Eh ) = 2 − wh − . (3)
3 m0 Eg + Eh Eg + Eh + ∆
| {z }
cb−vb
| {z } | {z }
vb−cb vb−(he/le) with we accounting for the confinement-induced band
Here p is the interband momentum matrix element, m0 mixing, ∆gremote being the remote bands’ contribu-
is the free-electron mass, Eg and ∆ are the band gap tion [9], and δgeso being the contribution which arises from
and spin-orbit splitting in the bulk material, respectively, the size-quantization induced mixing with the split-off
and EhR is the hole size quantization energy. The factor electron band
wh = d3 rfh2 (r) 6 1, where fh (r) is the valence band γ̄ 2 Ee
envelope function, accounts for the confinement-induced δgeso = −40 , (5)
γ1 ∆
 4

FIG. 2. The closed symbols show the g-factors of electrons (a) and holes (b), calculated for various pervoskite NCs by ETB.
The open symbols show the g-factors calculated in ETB for the bulk crystals. The orange and grey dashed lines are the results
of a k·p model fit to reproduce the experimental data for bulk crystals, see Ref. 9. (c) Sketch of the band structure of cubic
lead halide perovskites in vicinity of the direct band gap at the R point.

γ1 and γ̄ are the Luttinger parameters. While deriv- we selectively address NCs with specific mean sizes di-
ing Eq. (5) we assumed that Ee /∆  1 (see Support- rectly related to the exciton transition energy, Figure 1.
ing Information S2 for the complete analysis and com- In Figure 3a a transmission spectrum measured for
parison with ETB). Equations (4) and (5) explain the sample #1 at T = 5 K is shown. Its edge, centered at
strong renormalization of the electron g-factor provided 1.697 eV, originates from the exciton absorption. The
by the quantum confinement: the g-factor significantly spread of the absorption edge by about 26 meV from
decreases with increasing electron size-quantization en- 1.684 to 1.710 eV is due to the NC size dispersion.
ergy, in agreement with the ETB calculations. We apply the time-resolved Faraday ellipticity (TRFE)
technique to measure the electron and hole g-factors ().
This is an all-optical pump-probe technique using pulsed
III. g-FACTORS MEASURED IN CsPbI3 NCs lasers to address the coherent spin dynamics of carriers
in a magnetic field [41]. Recently, this technique was
Published information on the electron and hole g- successfully used for investigation of CsPb(Cl,Br)3 NCs
factors in lead halide perovskite NCs is limited. The in glass [14] and CsPbBr3 solution-grown NCs [12, 13].
reported experimental data are related to specific sizes Thereby, spin-oriented carriers are photogenerated by
of CsPbBr3 and CsPb(Br,Cl)3 NCs [12–14, 39, 40], but circularly-polarized pump pulses and the dynamics of
no systematic study on their size and band gap depen- their spin polarization are detected through variations of
dence is available. In order to check the role of quan- the ellipticity of the linearly polarized probe pulses [42].
tum confinement for the carrier g-factors that we predict Spectrally narrow (1.5 meV width) laser pulses with du-
by model calculations, we experimentally examine a set ration of 1.5 ps (repetition rate of 76 MHz) were used.
of CsPbI3 nanocrystals embedded in a fluorophosphate The laser photon energy was scanned across the exciton
glass matrix. Three samples with NC sizes varying in absorption band, exciting narrow subsets of NC sizes, so
the range of 8–16 nm are studied. Their NC size distri- that we could measure the spectral dependences of the
butions are centered at 13.8 nm (sample #1), 11.7 nm g-factors and compare with the modeling results for their
(sample #2), and 10.7 nm (sample #3), see Supporting dependence on EgNC , i.e. on the carrier confinement en-
Information S4. The NC size was changed by the synthe- ergies.
sis procedure (), providing a rather broad size dispersion A typical TRFE dynamics trace measured on sample
within each sample. By tuning the laser photon energy #1 at the laser photon energy of 1.691 eV is shown in
 5

(a)

Faraday Ellipticity (a. u.)

#1 #2 #3 electron 200 (d)

Zeeman splitting (µeV)

Transmission (a. u.)
1 1
T=5K
150
ge = +2.4
t = 0 ps
0.5 BV = 0.5 T 0.5 100
50 gh = −0.2
0 0 0
1.66 1.68 1.7 1.72 1.74 1.76 1.78 0 0.5 1 1.5
Energy (eV) Magnetic field BV (T)
(b) (c)
Excitation 1.691 eV
Faraday Ellipticity

Faraday Ellipticity
BV = 0.25 T BV = 0.1 T
hole 0.5 T
1.0 T
electron
1.5 T
0 0.5 1 1.5 0 0.5 1
Time (ns) Time (ns)
FIG. 3. (a) Transmission spectrum of CsPbI3 NCs (sample #1, red line, right axis) and spectral dependences of Faraday
ellipticity amplitudes of the electron signal measured for the three studied samples (for BV = 0.5 T at zero time delay, left
axis). T = 5 K. (b) Faraday ellipticity dynamics (blue dots) measured for sample #1 at 1.691 eV in BV = 0.25 T. The red
dashed line is a fit with two components using the equation for the signal AFE given in . The individual hole and electron
components are given below. (c) Faraday ellipticity dynamics measured for sample #1 at different magnetic fields. (d) Magnetic
field dependence of the electron and hole Zeeman splittings. Experimental data given by the symbols and lines are linear fits
to evaluate the corresponding g-factors. The sign of the hole g-factor is determined from another experiment.

Figure 3b. The measurement is performed in a magnetic dependences of the electron and hole Zeeman splittings
field of BV = 0.25 T applied perpendicular to the light evaluated by using EZ,e(h) = ~ωL,e(h) are shown in Fig-
wavevector (Voigt geometry). The carrier spins, which ure 3d. Both dependences can be fitted with a linear
are initially oriented along the wavevector, undergo Lar- function, which allows us to evaluate the g-factors accord-
mor precession about the magnetic field. In the TRFE ing to ge(h) = EZ,e(h) /(µB BV ): ge = +2.4 and gh = −0.2.
dynamics this leads to oscillating signals, which are de- We assign the fast Larmor precession frequency to the
caying with the spin dephasing time T2∗ . As is typical for electrons and the slow one to the holes, referring to the
bulk lead halide perovskites [8] and also their NCs [13], model predictions (Figure 2) and also comparing with
the spin dynamics are contributed by two oscillating sig- the experimental data for bulk materials having compa-
nals provided by the electrons and the holes. We fit rable band gaps [9]. Note that the TRFE is insensitive
the experimental dynamics (blue circles in Figure 3b) to the g-factor sign, but the latter can be obtained from
by a function accounting for these two decaying oscil- the spectral dependence of the g-factor value and the
latory components (). The fit is shown by the dashed model considerations. Namely, for gh < 0 its absolute
red line, and the individual electron and hole dynamics value decreases with growing EgNC , which is the case for
are given by the solid lines in Figure 3b. The following the studied CsPbI3 NCs, but for gh > 0 it increases, see
parameters are evaluated from the fit: for the electrons, Figure 2 and Ref. 9 for details.
the Larmor precession frequency ωL,e = 55.9 rad/ns and
∗ Very importantly, the dependences EZ,e(h) (BV ) shown
T2,e = 170 ps, and for the holes, ωL,h = 3.9 rad/ns
∗ in Figure 3d has no offset for extrapolation to zero mag-
and T2,h = 500 ps. Their amplitudes are about equal:
Se ≈ Sh . The electron g-factor is always positive for netic field. This allows us to assign the measured spin
these energies [9]. dynamics to resident electrons and holes confined in the
NCs and not to carriers bound in excitons. In the latter
With increasing magnetic field the Larmor preces- case the electron-hole exchange interaction, which in per-
sion frequencies increase and the spin dephasing times ovskite NCs amounts to 0.5 − 2.0 meV [15], would result
shorten, as one can see in Figure 3c. The magnetic field in spin beats with a single-exciton frequency. We sug-
 6

gest that the resident carriers in the NCs appear from

long-living photocharging, where either the electron or
the hole from a photogenerated electron-hole pair escapes 3 bulk
from the NC core. As a result, some NCs in the ensemble
become charged with electrons, some with holes, while
the rest stays neutral. This situation has been studied
2 electron
for CsPbBr3 NCs synthesized in solution and more argu-

g-factor
ments for its validity can be found in Ref. 13.
We measured three samples of CsPbI3 NCs by TRFE
and tuned the laser photon energy across the exciton ab- 1 CsPbI3 NCs
sorption band of the NC ensemble. The spectral depen-
dences of the electron spin amplitudes are shown in Fig-
hole
ure 3a. The amplitudes are normalized on their largest
value for each sample. The respective maxima are lo- 0
cated at 1.690 eV (sample #1), 1.705 eV (sample #2), bulk
and 1.735 eV (sample #3). The spin dynamics were de-
tected in the spectral range from 1.682 to 1.772 eV cor- −1
responding to NC sizes of 8 − 16 nm, see the insert of 1.65 1.7 1.75 1.8 1.85
Figure 1a. The measured electron and hole g-factors are Eg or EgNC (eV)
plotted in Figure 4. Note, that the hole signal was not
detectable in the sample #3 so that we show only data
FIG. 4. Energy dependence of the electron (circles) and hole
for its electron g-factor.
(triangles) g-factors measured at T = 5 K in CsPbI3 NCs:
In order to compare the experimental data with the data from samples #1 (black open symbols), #2 (blue sym-
theoretical predictions, we show in Figure 4 the model bols) and #3 (green symbols). Red lines are ETB model
calculations taken from Figure 2. One can see that in results for NCs. The orange and grey dashed lines are the
agreement with our ETB calculations and the qualita- universal dependences for bulk crystals taken from Ref. 9.
tive k·p analysis, the experimental data for CsPbI3 NCs
clearly show a strong deviation of the electron g-factor
from the universal bulk dependence (orange dashed line), the band structure of these materials. Based on the
while the hole g-factor closely follows this dependence ETB and k·p calculations we have demonstrated that
(grey dashed line). Note that the clear deviation from the significant deviation of the electron g-factor from the
the univeral behaviour is seen even for relatively large universal bulk dependence [9] arises from the quantum-
nanocrystals and can be naturally explained by admix- confinement induced admixture of the heavy- and light-
ture of the split-off bands to the ground electron band by electron bands to the electron ground state.
the size quantization (Supporting Information S2). The calculations were performed for cubic shaped NCs,
Care has to be exercised when comparing the experi- while for NCs in glass, due to the growth method,
mental data with the calculations. In Figure 4 we show droplets of pseudo spherical shape are expected as con-
the experimental data as a function of excitation energy, firmed by STEM, see Supplementary Information S4.
corresponding to the ground state energy of the excitons Still, the ETB calculations agree well with the experi-
in NCs, while the results of the calculations are shown as mental data and we expect that the g-factor does not sig-
function of the effective band gap of the NCs EgNC . We nificantly depend on NC shape.[43] The elongated shape
argue that the exciton binding energy does not change and low-symmetry crystal phase[9, 44] of NCs both lead
the results and their interpretation qualitatively as the to an anisotropy of the g-factor, these effects and their in-
binding energy is not too large (of the order of a few tens terdependences are yet to be investigated. However, the
of meV) and leads to an overall shift, which smoothly main trends demonstrated here are not affected by this
depends on NC size. anisotropy. Nevertheless, the random orientation of the
NC anisotropy axes in an ensemble should be accounted
in the analysis of the experimental data as it might lead
IV. DISCUSSION to small uncertainties of the measured g-factor value pre-
sented in Figure 4.
The obtained theoretical results match well the exper-
imental findings. Previously, it had been reported that
the experimentally observed[12, 13] electron g-factors in V. CONCLUSIONS
large CsPbBr3 NCs are significantly (∼ 10%) lower than
the bulk value[7] while the hole g-factor is marginally In conclusion, we have analyzed theoretically the role
larger. Our results provide the explanation for this ob- of confinement and halide exchange on the electron and
servation and demonstrate that this is a general feature hole g-factors in lead halide perovskite nanocrystals.
common for perovskite-based NCs, which originates from Both the empirical tight-binding and k·p approaches sug-
 7

gest a strong deviation of the electron g-factor from the factor values has been discussed in Ref. 49.
bulk values of the previously established universal de- To reach better agreement with the experimental data,
pendence of the g-factor on the band gap energy with we choose a new set of tight-binding parameters to re-
decreasing NC size. For the hole g-factor only small devi- produce more precisely the experimentally measured g-
ations from the universal dependence are expected. The factors in the bulk crystals CsPbX3 , X = I, Br, Cl, at the
theory results are confirmed by experiments on CsPbI3 cost of a less accurate description of the lower halide p-
NCs. The combination of halide exchange, also consid- bands. The comparison of the band structure calculated
ering mixed compounds like CsPb(I,Br)3 , CsPb(Br,Cl)3 using DFT[31, 45, 46] and ETB with the new parameters
or CsPb(I,Cl)3 , and NC size and shape variation offers a is presented in Supporting Information S1.
tool of great flexibility for tailoring the electron and hole
g-factors and for selecting the desired values for them. b. k · p calculations. For qualitative analysis of the
We are convinced that very similar trends will be found ETB calculations we use the k·p 8-band approach de-
for the g-factors in hybrid organic-inorganic NCs, as the veloped in Ref. 9. Note that this approach is valid only
band gap is mostly controlled by the halogen exchange for large NCs, where the wave vector associated with the
and is only weakly dependent on the cation type. This is quantum confinement which is inversely proportional to
of great importance for both basic research on the spin the NC size is small enough to ignore the non-parabolicity
physics and for spintronic applications of lead halide per- of the bands. Details of the k·p calculations are presented
ovskite NCs. in Supporting Information S2.
c. Synthesis of CsPbI3 NCs. The studied CsPbI3
nanocrystals embedded in fluorophosphate Ba(PO3 )2 -
VI. METHODS AlF3 glass were synthesized by rapid cooling of a glass
melt enriched with the components needed for the per-
a. Tight-binding calculations. To model the quan- ovskite crystallization. The details of the method are
tum confined NC states we use the sp3 d5 s∗ tight-binding given in Refs. 14 and 50. Samples of fluorophosphate
method. It has been demonstrated that this method per- (FP) glass with the composition BaI2 -doped 35P2 O5 –
mits the precise description of the band structure in in- 35BaO–5AlF3 –10Ga2 O3 –10PbF2 –5Cs2 O (mol. %) were
organic lead halide perovskites [31]. Moreover, it allows synthesized using the melt-quench technique. The glass
for the realistic description of the surface of these ma- synthesis was performed in a closed glassy carbon cru-
terials: ETB gives a qualitatively correct description of cible at the temperature of T = 1050◦ C. About 50 g of
the PbI [001] surface of CsPbI3 without passivation [31]. the batch was melted in the crucible for 30 min., then the
In Ref. 9 this approach was used to demonstrate that glass melt was cast on a glassy carbon plate and pressed
the g-factors of charge carriers follow simple trends and to form a plate with a thickness of about 2 mm. Samples
depend almost exclusively on the band gap energy. with a diameter of 5 cm were annealed at the temperature
For the empirical tight-binding calculations of the g- of 50◦ C below Tg = 400◦ C to remove residual stresses.
factors in bulk materials we follow the procedure ex- The CsPbI3 perovskite NCs were formed from the glass
plained in detail in the Supplementary Information of melt during the quenching. The glasses obtained in this
Ref. 9. However, as shown in Ref. 9, the values of the bulk way are doped with CsPbI3 NCs. The dimensions of
g-factors calculated within ETB strongly deviate from the NCs in the initial glass were regulated by the con-
the experimentally measured g-factors. We attribute the centration of iodide and the rate of cooling of the melt
origin of this deviation to peculiarities of the DFT calcu- without heat treatment above Tg . Three samples were in-
lations underlying the ETB parametrization and to the vestigated in this paper, which we label #1, #2 and #3.
ETB model itself. First, the modified Becke-Johnson Their technology codes are EK7, EK31 and EK8, respec-
(mBJ) exchange-correlation potential[45, 46] is good to tively. They differ in the NC sizes, which is reflected by
obtain a band gap energy close to experimental data, but relative spectral shifts of their optical spectra.
it underestimates the renormalization of the energies of The change of the NC size was achieved by chang-
the conduction band secondary minima and, thus, the ing the concentration of iodine in the melt. Due to the
carrier effective masses.[47] Second, for the ETB model high volatility of iodine compounds and the low viscos-
used, an accurate description of the halide p-bands is ity of the glass-forming fluorophosphate melt at elevated
hardly possible without inclusion of the interaction be- temperatures, an increase in the synthesis time leads
tween next-nearest neighbors [48]. The lack of such inter- to a gradual decrease in the concentration of iodine in
action in our model is compensated by the interaction of the equilibrium melt. Thus, it is possible to completely
the halide p-states with other bands [31], but, as a result, preserve the original composition and change only the
this interaction is overestimated. The effective masses of concentration of iodine due to a smooth change in the
the charge carriers are correct as long as the dispersion synthesis duration. Glasses with different NC sizes were
is reasonably well described in the full Brillouin zone, synthesized using different time intervals. Glasses with
but more subtle properties, including the g-factor values, the photoluminescence lines centered at 1.801, 1.808 and
are affected by the overestimated interband interactions. 1.809 eV (room temperature measurements) were syn-
The importance of the ETB parametrization for the g- thesized within 40, 35 and 30 min, respectively.
 8

d. Time-resolved Faraday ellipticity (TRFE): The probe time delay due to spin dephasing. When both elec-
coherent spin dynamics were measured by a pump-probe trons and holes contribute to the Faraday ellipticity sig-
time-resolved technique [41, 42]. We use a titanium- nal, which is the case for the studied perovskite NCs, the
sapphire (Ti:Sa) laser emitting 1.5 ps long pulses with signal can be described as a superposition of two decaying
∗
a spectral width of about 1 nm (1.5 meV) at a pulse rep- oscillatory functions: AFE = Se cos(ωL,e t) exp(−t/T2,e )+
∗
etition rate of 76 MHz (repetition period TR = 13.2 ns). Sh cos(ωL,h t) exp(−t/T2,h ). Here Se(h) are the signal am-
The laser photon energy was tunable in the spectral range plitudes that are proportional to the spin polarization of
∗
of 1.265 − 1.771 eV (700 − 980 nm). The laser beam was electrons (holes), and T2,e(h) are the carrier spin dephas-
split into two beams, pump and probe. The probe pulses ing times. The g-factors are evaluated from the Larmor
were delayed with respect to the pump pulses by a me- precession frequency ωL,e(h) by |ge(h) | = ~ωL,e(h) /(µB B).
chanical delay line. Both pump and probe beams were Note, that the electron and hole Zeeman splitting is
modulated using photo-elastic modulators. The pump EZ,e(h) = ~ωL,e(h) .
beam helicity was modulated between σ + and σ − circu-
lar polarization at the frequency of 50 kHz. The probe
beam was kept linearly polarized, but its amplitude was
modulated at the frequency of 84 kHz. The polarization
ACKNOWLEDGMENTS
of the transmitted probe beam was detected with a bal-
anced photodiode and analyzed, via a lock-in technique,
with respect to the variation of its ellipticity (Faraday M.O.N. acknowledges support by TU Dortmund Uni-
ellipticity). versity core funds. E.K. and D.R.Y. acknowledge fi-
The measurements were performed at the low tempera- nancial support by the Deutsche Forschungsgemeinschaft
ture of T = 5 K, with the sample placed in cooling helium via the SPP2196 Priority Program (Project YA 65/26-
gas. The cryostat was equipped with a vector magnet 1). E.A.Zh. and M.B. are thankful to the Interna-
with three pairs of orthogonally oriented coils, allowing tional Collaborative Research Center TRR160 (Project
us to apply magnetic fields up to 3 T in any direction. We A1). M.M.G., M.A.S. and E.L.I. acknowledge support of
used only the Voigt geometry where the magnetic field the Russian Foundation for Basic Research (Grant No.
is perpendicular to the light k-vector (BV ⊥ k). In the 19-52-12038). E.V.K and M.S.K. acknowledge the Saint-
transverse magnetic field, the Faraday ellipticity ampli- Petersburg State University (Grant No. 94030557). We
tude oscillates in time, reflecting the Larmor spin pre- thank Volker Brandt from TU Dortmund for the STEM
cession of the carriers. It decays with increasing pump- imaging.

[1] Z. V. Vardeny and M. C. Beard, Hybrid Organic Inor- N. E. Kopteva, D. N. Dirin, O. Nazarenko, M. V. Ko-
ganic Perovskites: Physical Properties and Applications valenko, A. Baumann, J. Höcker, V. Dyakonov, and
(in 4 Volumes) (Singapore: World Scientific Publishing M. Bayer, Nat. Comm. 13, 3062 (2022).
Co. Pte. Ltd., United States, 2022). [10] M. O. Nestoklon, S. V. Goupalov, R. I. Dzhioev, O. S.
[2] A. Vinattieri and G. Giorgi, Halide Perovskites for Pho- Ken, V. L. Korenev, Y. G. Kusrayev, V. F. Sapega,
tonics (AIP Publishing LLC, 2021). C. de Weerd, L. Gomez, T. Gregorkiewicz, J. Lin,
[3] D. Giovanni, H. Ma, J. Chua, M. Grätzel, R. Ramesh, K. Suenaga, Y. Fujiwara, L. B. Matyushkin, and I. N.
S. Mhaisalkar, N. Mathews, and T. C. Sum, Nano Letters Yassievich, Phys. Rev. B 97, 235304 (2018).
15, 1553 (2015), pMID: 25646561. [11] S. Strohmair, A. Dey, Y. Tong, L. Polavarapu, B. J.
[4] C. Zhang, D. Sun, C.-X. Sheng, Y. X. Zhai, K. Miel- Bohn, and J. Feldmann, Nano Letters 20, 4724 (2020),
czarek, A. Zakhidov, and Z. V. Vardeny, Nature Physics pMID: 32453960.
11, 427 (2015). [12] M. J. Crane, L. M. Jacoby, T. A. Cohen, Y. Huang, C. K.
[5] R. Wang, S. Hu, X. Yang, X. Yan, H. Li, and C. Sheng, Luscombe, and D. R. Gamelin, Nano Letters 20, 8626
J. Mater. Chem. C 6, 2989 (2018). (2020).
[6] P. Odenthal, W. Talmadge, N. Gundlach, R. Wang, [13] P. S. Grigoryev, V. V. Belykh, D. R. Yakovlev, E. Lhuil-
C. Zhang, D. Sun, Z.-G. Yu, Z. Valy Vardeny, and Y. S. lier, and M. Bayer, Nano Letters 21, 8481 (2021).
Li, Nature Physics 13, 894 (2017). [14] E. Kirstein, N. E. Kopteva, D. R. Yakovlev, E. A.
[7] V. V. Belykh, D. R. Yakovlev, M. M. Glazov, P. S. Grig- Zhukov, E. V. Kolobkova, M. S. Kuznetsova, V. V.
oryev, M. Hussain, J. Rautert, D. N. Dirin, M. V. Ko- Belykh, I. A. Yugova, M. M. Glazov, M. Bayer, and
valenko, and M. Bayer, Nat. Comm. 10, 673 (2019). A. Greilich, Nature Communications 14, 699 (2023).
[8] E. Kirstein, D. R. Yakovlev, M. M. Glazov, E. Evers, [15] P. Tamarat, L. Hou, J.-B. Trebbia, A. Swarnkar, L. Bi-
E. A. Zhukov, V. V. Belykh, N. E. Kopteva, D. Kud- adala, Y. Louyer, M. I. Bodnarchuk, M. V. Kovalenko,
lacik, O. Nazarenko, D. N. Dirin, M. V. Kovalenko, and J. Even, and B. Lounis, Nature Communications 11, 6001
M. Bayer, Advanced Materials 34, 2105263 (2022). (2020).
[9] E. Kirstein, D. R. Yakovlev, M. M. Glazov, E. A. Zhukov, [16] L. Protesescu, S. Yakunin, M. I. Bodnarchuk, F. Krieg,
D. Kudlacik, I. V. Kalitukha, V. F. Sapega, G. S. Dim- R. Caputo, C. H. Hendon, R. X. Yang, A. Walsh, and
itriev, M. A. Semina, M. O. Nestoklon, E. L. Ivchenko, M. V. Kovalenko, Nano Lett. 15, 3692 (2015).
 9

[17] S. Yakunin, L. Protesescu, F. Krieg, M. I. Bodnarchuk, [42] I. A. Yugova, M. M. Glazov, E. L. Ivchenko, and A. L.
G. Nedelcu, M. Humer, G. D. Luca, M. Fiebig, W. Heiss, Efros, Phys. Rev. B 80, 104436 (2009).
and M. V. Kovalenko, Nat. Comm. 6, 8056 (2015). [43] I. D. Avdeev, S. V. Goupalov, and M. O. Nestoklon,
[18] A. Dey, J. Ye, A. De, E. Debroye, S. K. Ha, E. Bladt, Phys. Rev. B 107, 035414 (2023).
et al., ACS Nano 15, 10775 (2021). [44] R. Ben Aich, S. Ben Radhia, K. Boujdaria, M. Chamarro,
[19] A. Younis, C.-H. Lin, X. Guan, S. Shahrokhi, C.-Y. and C. Testelin, The Journal of Physical Chemistry Let-
Huang, Y. Wang, T. He, S. Singh, L. Hu, J. R. D. Re- ters 11, 808 (2020), pMID: 31931571.
tamal, J.-H. He, and T. Wu, Advanced Materials 33, [45] F. Tran and P. Blaha, Phys. Rev. Lett. 102, 226401
2005000 (2021). (2009).
[20] E. L. Ivchenko, Optical spectroscopy of semiconductor [46] R. A. Jishi, O. B. Ta, and A. A. Sharif, J. Phys. Chem.
nanostructures (Alpha Science, Harrow UK, 2005). C 118, 28344 (2014).
[21] A. A. Kiselev, E. L. Ivchenko, and U. Rössler, Phys. Rev. [47] D. Waroquiers, A. Lherbier, A. Miglio, M. Stankovski,
B 58, 16353 (1998). S. Poncé, M. J. T. Oliveira, M. Giantomassi, G.-M. Rig-
[22] A. Tadjine, Y.-M. Niquet, and C. Delerue, Phys. Rev. B nanese, and X. Gonze, Phys. Rev. B 87, 075121 (2013).
95, 235437 (2017). [48] R. Kashikar, B. Khamari, and B. R. K. Nanda, Phys.
[23] I. A. Yugova, A. Greilich, D. R. Yakovlev, A. A. Kiselev, Rev. Materials 2, 124204 (2018).
M. Bayer, V. V. Petrov, Y. K. Dolgikh, D. Reuter, and [49] J.-M. Jancu, R. Scholz, E. A. de Andrada e Silva, and
A. D. Wieck, Phys. Rev. B 75, 245302 (2007). G. C. La Rocca, Phys. Rev. B 72, 193201 (2005).
[24] A. A. Sirenko, T. Ruf, M. Cardona, D. R. Yakovlev, [50] E. Kolobkova, M. Kuznetsova, and N. Nikonorov, Journal
W. Ossau, A. Waag, and G. Landwehr, Phys. Rev. B of Non-Crystalline Solids 563, 120811 (2021).
56, 2114 (1997).
[25] J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem.
Phys. 124, 219906 (2006).
[26] J. M. Luttinger and W. Kohn, Phys. Rev. 97, 869 (1955).
[27] A. Zunger, in Quantum theory of real materials, Vol. 348,
edited by J. R. Chelikowsky and S. G. Louie (Kluwer
academic publisher, 1996) Chap. 13, p. 173.
[28] R. Benchamekh, M. Nestoklon, J.-M. Jancu, and
P. Voisin, in Semiconductor Modeling Techniques,
Springer Series in Materials Science, Vol. 159, edited by
X. Marie and N. Balkan (Springer, 2012) Chap. 2, p. 19.
[29] J.-M. Jancu, R. Scholz, F. Beltram, and F. Bassani, Phys.
Rev. B 57, 6493 (1998).
[30] Y. M. Niquet, D. Rideau, C. Tavernier, H. Jaouen, and
X. Blase, Phys. Rev. B 79, 245201 (2009).
[31] M. Nestoklon, Computational Materials Science 196,
110535 (2021).
[32] P.-O. Löwdin, J. Chem. Phys. 18, 365 (1950).
[33] J. C. Slater and G. F. Koster, Phys. Rev. 94, 1498 (1954).
[34] K. Wu and H. Simon, SIAM J. Matrix Anal. Appl. 22,
602 (2000).
[35] Q. Zhao, A. Hazarika, L. T. Schelhas, J. Liu, E. A. Gauld-
ing, G. Li, M. Zhang, M. F. Toney, P. C. Sercel, and J. M.
Luther, ACS Energy Letters 5, 238 (2020).
[36] M. A. Becker, R. Vaxenburg, G. Nedelcu, P. C. Ser-
cel, A. Shabaev, M. J. Mehl, J. G. Michopoulos, S. G.
Lambrakos, N. Bernstein, J. L. Lyons, T. Stöferle, R. F.
Mahrt, M. V. Kovalenko, D. J. Norris, G. Rainò, and
A. L. Efros, Nature 553, 189 (2018).
[37] M. Graf and P. Vogl, Phys. Rev. B 51, 4940 (1995).
[38] Y. Kim, Z. Hu, I. D. Avdeev, A. Singh, A. Singh,
V. Chandrasekaran, M. O. Nestoklon, S. V. Goupalov,
J. A. Hollingsworth, and H. Htoon, Small 17, 2006977
(2021).
[39] M. Fu, P. Tamarat, H. Huang, J. Even, A. L. Rogach,
and B. Lounis, Nano Letters 17, 2895 (2017), pMID:
28240910.
[40] M. Isarov, L. Z. Tan, M. I. Bodnarchuk, M. V. Kovalenko,
A. M. Rappe, and E. Lifshitz, Nano letters 17 8, 5020
(2017).
[41] D. R. Yakovlev and M. Bayer, Coherent spin dynamics
of carriers, in Spin Physics in Semiconductors, edited by
M. I. Dyakonov (Springer Berlin Heidelberg, Berlin, Hei-
delberg, 2008) pp. 135–177.
 S1

Supplementary Information

S1. TIGHT-BINDING PARAMETERS

The ETB parametrization in Ref. S1 gives an almost perfect description of the band structure of bulk cubic CsPbX3 ,
calculated within the DFT approach. However, the resulting bulk g-factors of electrons and holes significantly deviate
from the experimental data obtained in Ref. S1. We attribute this to an incorrect parametrization of the interaction
with the halide p-band. To improve the modeling of the g-factors, we changed the parameters of the interaction
between the Pb and the halide atoms, the modified parameters are given in Table S1. This change results in a slightly
larger difference between the ETB and DFT band structures for CsPbX3 in the cubic phase, but the g-factors are much
closer to the experimental data. We also changed the diagonal energy of the Pb p-orbital for CsPbI3 and CsPbBr3
to match the low-temperature experimental band gap of MAPbI3 and CsPbBr3 from Ref. S1. The band structure,
compared with DFT calculations, is shown in Fig. S1. The values of the carrier effective masses and g-factors are
given in Table S2.

TABLE S1. Modified ETB parameters used in the calculations. We give only the parameters which differ from the parameters
presented in Ref. S1. All values are given in eV.
CsPbI3 CsPbBr3 CsPbCl3
Epc 4.81 5.39 6.46
ppσ −2.47 −2.61 −2.79
ppπ 0.00 0.05 0.20
pa d c σ 2.62 2.90 3.51

TABLE S2. Effective masses and g-factors of electrons and holes calculated for the bulk materials using the parameters from
Ref. S1, corrected in accordance with Table S1.
CsPbI3 CsPbBr3 CsPbCl3
Eg (eV) 1.652 2.352 3.090
∆ (eV) 1.258 1.436 1.526
me /m0 0.168 0.219 0.315
mh /m0 0.145 0.191 0.250
ge +3.23 +1.77 +0.95
gh −0.33 +0.66 +1.14

The band gaps agree well with the combined experimental and theoretical (DFT) studies on bulk crystals giving
1.72 eV, 2.31 eV, and 2.99 eV for CsPbX3 in Ref. S2.

S2. K·P MODEL

To describe the electron and hole g-factor renormalization within the k·p approach, we consider the 8 band model
that includes the two-fold degenerate valence band states and the six states in the conduction band: the bottom
two-fold degenerate (spin-orbit split) conduction band and the higher band that is four-fold degenerate at the R point
of the Brillouin zone and consisting of the heavy-electron and light-electron branches.
The Bloch amplitudes at the R point are taken in the form
valence band:
(
uvb,1/2 (r) = iS(r)| ↑i,
(S1a)
uvb,−1/2 (r) = iS(r)| ↓i,
 S2

FIG. S1. Band structure calculated for bulk CsPbI3 , CsPbBr3 , and CsPbCl3 using ETB with the parameters from Ref. S1,
corrected in accordance with Table S1 (black lines) compared with DFT calculations (green dashed lines). For details of the
DFT calculations see the Supporting Information of Ref. S1.

bottom conduction band:


X (r) + i Y(r)
ucb,1/2 (r) = − sin ϑ Z (r)| ↑i − cos ϑ √ | ↓i,


2 (S1b)
X (r) − i Y(r)
ucb,−1/2 (r) = + sin ϑ Z (r)| ↓i − cos ϑ √ | ↑i,


2
split-off c.b. (light electron):

X (r) + i Y(r)
ule,1/2 (r) = cos ϑ Z (r)| ↑i − sin ϑ √ | ↓i,


2 (S1c)
X (r) − i Y(r)
ule,−1/2 (r) = cos ϑ Z (r)| ↓i + sin ϑ √ | ↑i,


2
split-off c.b. (heavy electron):

+ i Y(r)
uhe,3/2 (r) = − X (r)√ | ↑i,


2 (S1d)
X (r) − i Y(r)
uhe,−3/2 (r) = √ | ↓i.


2
 S3

Here ↑ and ↓ denote the basic spin-1/2 spinors, the function S(r) is invariant in the group Pm3̄m, and the functions
X (r), Y(r), Z(r) transform like the corresponding coordinates, the subscripts vb, cb, le, and he refer, respectively, to
the valence band, (bottom) conduction band, spin-orbit split-off light electron, and spin-orbit split-off heavy electron.
For the two latter bands the notations h and l are used in the same way as for
p the holes in III-V
p or II-VI semiconductors.
In what follows we focus on the perovskite cubic phase where cos ϑ = 2/3, sin ϑ = 1/3, and the space group
of symmetry is P m3̄m (or O1h ). At the R point, the light and heavy split-off electrons are degenerate forming a
quadruplet with total angular momentum 3/2. The representations at the R point are R6− and R8− (corresponding to
the representations Γ− −
6 , Γ8 of the point group Oh in the Koster notation[S3]) for the conduction bands and R6 (or
+
+
Γ6 ) for the valence band. We define the band gap Eg as the energy difference between the bottom conduction band
and the top valence band, and let ∆ > 0 be the spin-orbit splitting between the two-fold and the four-fold conduction
bands. The interband momentum matrix element is defined by

p = ihZ |p̂z | S i = ihX |p̂x | S i = ihY |p̂y | S i . (S2)

Here p̂x,y,z are the components of the momentum operator, and we take the phases of the Bloch functions in such a
way that p is real.
We start the analysis of the g-factors with the case of the holes. Following Ref. S1 we obtain for a bulk crystal of
cubic symmetry

4 p2
 
1 1
gh = 2 − − , (S3)
3 m0 Eg Eg + ∆

where m0 is the free electron mass. The second contribution is related to the magnetic-field-induced k·p-mixing with
the bottom conduction band. The third term is related to the mixing with the higher 4-fold degenerate conduction
band. The remote band contributions in the case of holes are rather small.[S1] Within the k·p-approach for simplicity
of the analysis we consider the size-quantization effect in a NC of spherical symmetry and expect a relatively weak
effect of the NC shape on the g-factors, cf. Ref. S4. Kiselev et al. S5 derived in the k·p method the following equation
for the electron g factor in a spherical quantum dot, composed of III-V based semiconductor material A embedded in
the wider band gap matrix of material B:
g = 2 + [gA (Ee ) − 2] wA + [gB (Ee ) − 2] wB (S4)
+ [gB (Ee ) − gA (Ee )] VQD (R) f 2 (R) .
Here R and VQD (R) = 4πR3 /3 are the radius and volume of the quantum dot, f (r) is the conduction-electron
scalar envelope, gA (E) and gB (E) are defined by Eq. (3) ofRthe main text, where the energy Eh is replaced by the
electron confinement energy Ee , wA and wB are the integrals drf 2 (r), taken over the A and B volumes, respectively.
Importantly, the sum wA + wB differs from unity because of the confinement-induced conduction band-valence band
mixing [S5, S6]. If the potential barriers are high, the values of f 2 (R) and wB become negligible, and the g factor
tends toward 2 + wA [gA (Ee ) − 2]. Since the role of the conduction band in III-V based semiconductors is in effect
taken by the valence band in perovskites, the hole g-factor in the perovskite NC is given by

4 p2
 
1 1
gh (Eh ) = 2 − wh − . (S5)
3 m0 Eh + Eg Eh + Eg + ∆

Here Eh is the hole quantization energy, and the factor wh is given by

1
wh (E) = 2m(E)c2 (E)
, (S6)
1+ ~2 E

where
2 p2
 
1 2 1
= +
m(E) 3 m20 Eg + E Eg + ∆ + E

and
~2 p 2
 
2 2 1
c (E) = + .
3m20 (Eg + E)2 (Eg + ∆ + E)2

This is in agreement with Eq. (3) of the main text.

 S4

bulk, hole bulk, electron NCs

(a) (b) (c) le, ± 12

he, ± 32

δgso
e
cb, ± 12
2
p
Eg +Δ
2
p2 p2
− Ep g Eg Eg +Ee

Δgremote Δgremote

vb, ± 12
vb2

FIG. S2. Illustration of the bands important for the calculation of: (a) hole g-factor in bulk perovskites,[S1] Eq. (S3) (b)
electron g-factor in bulk perovskites,[S1] Eq. (S7), and (c) electron g-factor in perovskite NCs, Eq. (S8).

Now we turn to the case of the conduction band electrons. The analysis, following the same lines as above, but
starting from the bulk electron g-factor[S1],

2 4 p2 1
ge = − + + ∆gremote , (S7)
3 3 m0 Eg
yields the expression for the NCs:
2 4 p2 we
ge (Ee ) = − + + ∆gremote + δgeso . (S8)
3 3 m0 Eg + Ee
Here the term −2/3 comes from the spin structure of the Bloch amplitudes, the second term results from the
conduction-valence band k·p-mixing with Ee being the electron quantization energy, we is an energy-dependent
coefficient, and δgeso is contributed by the higher conduction band R8− . The band contributions to the resulting g-
factors in bulk materials and NCs are illustrated in Fig. S2. In order to get a reasonable estimate, we first consider
the two-band approximation excluding the band R8− as if ∆  Eg . In this model the energy spectrum in absence of
the magnetic field is symmetric, we = wh , and the electron and hole quantization energies coincide, Ee = Eh . Leaving
only the linear in Ee/h terms in (S6) in the limit ∆ → ∞, we obtain

4 p2 1
gh (Eh ) ≈ 2 − . (S9a)
3 m0 2Eh + Eg

2 4 p2 1
ge (Ee ) = − + + ∆gremote + δgeso . (S9b)
3 3 m0 2Ee + Eg
Finally, we include the split-off conduction band in the considerations. Importantly, its presence results in a strong
renormalizationS6 of the electron g-factor described by the last term in Eq. (S7). We derive an analytical formula for
δgeso for a spherically-symmetric NC assuming that the electron quantization energy Ee  ∆. For simplicity, we also
assume the inequality ∆  Eg . Table S3 presents the 2×4 matrix H6c,8c (k) of the off-diagonal matrix elements that
mix the conduction bands R6− and R8− in the extended k·p approach. The matrix is similar to that describing the
coupling between the valence bands Γ7 and Γ8 in III-V semiconductors [S7–S9]. Standard notations are used, namely,
√ 2
3~ γ3
H= kz (kx − iky ) , (S10)
m
√ 20
3~ 
γ2 kx2 − ky2 − 2iγ3 kx ky ,


I=
2m0
~2 γ2 2
kx + ky2 − 2kz2 ,


G−F =
m0
 S5

TABLE S3. The off-diagonal components of the conduction-electron Hamiltonian H6c,8c .

he, +3/2 le, +1/2 le, −1/2 he, −3/2
∗
√ √ p √
cb, +1/2 H / 2 (G − F )/ 2 − 3/2H − 2I
√ ∗ √ √
− 3/2H ∗ −(G − F )/ 2 H/ 2
p
cb, −1/2 2I

and γ2 , γ3 are the dimensionless Luttinger band parameters, applied here for the complicated perovskite conduction
band [S10]. In the following we ignore the difference between γ2 and γ3 , and use the notation γ̄ for them. In this
spherical approximation
√ 2
3~ γ̄
I= (kx − iky )2 .
2m0

In the presence of a magnetic field B = ∇ × A, the electron wave vector in Eqs. (S10) should be replaced by
e e
K = k − A = −i∇ − A (S11)
~c ~c
and (kx − iky ) by (Kx − iKy ). We evaluate the spin splitting of the conduction band electron in a NC for B k z,
making use of second-order perturbation theory as follows
1
δgeso µB Bz = − hcb, 1s, +1/2|H6c,8c H8c,6c |cb, 1s, +1/2i
∆  (S12)
− hcb, 1s, −1/2|H6c,8c H8c,6c |cb, 1s, −1/2i .
†
Here, |cb, 1s, ±1/2i is the Kramers conjugate pair of the zero-dimensional electron ground state 1s, H8c,6c = H6c,8c ,
and we assume the spin-orbit splitting ∆ to exceed by far the size-quantization energies of the states contributing to
δgeso .
Extracting the Bz -linear contributions by virtue of the relations
ie
kx ky − ky kx = Bz ,
~c
eBz
(kx + iky )(kx − iky ) → , (S13)
~c
eBz 2
(kx + iky )2 (kx − iky )2 → 4 (k + ky2 ),
~c x
we arrive at
~2 hkz2 i 2
δgeso = −60 γ̄ . (S14)
m0 ∆
Here hkz2 i = hkx2 i = hky2 i is the average value of kz2 over the 1s electron envelope function. Note that for a spherically-
symmetric or cubic NC ~2 hkz2 i/2me = Ek /3, where Ek is the contribution of the kinetic energy to the electron
confinement energy Ee . For a NC with infinite barriers Ek = Ee . Equation (S14) can be recast as
γ̄ 2 Ek
δgeso = −40 , (S15)
γ1 ∆
where we took into account that Ek = 3~2 γ1 hkz2 i/(2m0 ).
To demonstrate the significance of the interaction with the split-off band we present rough estimates within the
Kane two-band model, where the Luttinger parameter γ̄ = p2 /(3m0 Eg ) and m0 /me = 2γ̄(= γ1 ). From Eq. (S15) we
obtain
Ee
δgeso ≈ −20γ̄ . (S16)
∆
Even for Ee ∼ 0.1∆ the split-off band admixture contribution δgeso is on the order of unity due to the large prefactor.
Figure S3 shows the comparison of the electron g-factors calculated after Eqs. (S8), (S16) shown by the dashed lines
with the ETB results. We used the same value of p (see caption of Figure S3), which reproduces the g-factors in the
bulk material, and we took into account that the electron quantization energy is the same as the hole quantization
energy due to comparable effective masses. One can see that the analytical expressions derived in the k·p-model
describe the initial rapid drop of the electron g-factors with increasing confinement. Further, the saturation-like
behavior is beyond the contribution of first-order in Ee , Eqs. (S14) – (S16).
 S6

4
(a) halide exchange I Br Cl
3

electron g-factor
2

0
large small NC size
−1
2 (b) CsPbCl3
hole g-factor

CsPbBr3
1 CsPbI
3

−1
1.5 2 2.5 3 3.5 4 4.5 5
Eg or ENC
g
(eV)

FIG. S3. Closed symbols show the g-factors of electrons (upper panel) and holes (lower panel) calculated for CsPbX3 (X = I,
Br, Cl) NCs in ETB. Open symbols show the g-factors (from an atomistic approach) in the bulk crystals. Dashed lines show
the result of the k·p-approach. For the holes we disregard the effect of size quantization.

S3. TEMPERATURE DEPENDENCE OF THE BAND GAP

We measured transmission spectra with a halogen lamp and a 0.5 m monochromator with an attached charge-
coupled devices camera. The spectra measured for the sample #1 in the temperature range from 8 up to 285 K
are shown in Fig. S4(a). The transmission has a step like decrease around 1.7 eV, which reflects the strong light
absorption above the band gap.
With increasing temperature the CsPbI3 NCs’ central absorption energy E0 shows a monotonic high energy shift by
60 meV. The shift is linear for temperatures below 100 K and gradually saturates with further temperature increase.
The slope of the edge also increases with rising temperature, it is almost linear for temperatures above 50 K but
shows saturation dependence for low temperatures. In perovskites, the lattice shrinkage dependent gap change is far
stronger than electron-phonon interaction (Varshni-term), which acts in the opposite energy direction. We assume
a linear shrinkage of the lattice constant with temperature, resulting in a linear dependence of the band gap on
temperature [S11, S12]. Together, it reads

αT 2
E0 (T ) = E0 (T = 0) − + ζT (S17)
T +β |{z}
| {z } lattice shrinkage
Varshni

for describing the band gap shift. We fit the experimental dependence with the parameters ζ = 0.447 meV/K as the
linear slope, Eg (T = 0) = 1.692 eV as the zero temperature band gap, and the Varshni parameters α = 1.2 meV/K
and β = 1300 K.

S4. SCANNING TRANSMISSION ELECTRON MICROSCOPY – HIGH-ANGLE ANNULAR

DARK-FIELD

The evaluation of the NC sizes was done by means of scanning transmission electron microscopy using the high-
angle annular dark-field method (STEM-HAADF). For these measurements the samples were grinded with an agate
mortar. The powder was placed on a carbon coated copper grid. The TEM images were recorded using a Talos
 S7

(a) (b)
1.76

1.74

Transmission (arb. u.)

E0 (meV)
T= 1.72
285 K
198 K
160 K 1.7
E0 95 K
50 K 1.68
8K
1.6 1.7 1.8 1.9 0 100 200 300
Energy (eV) Temperature (K)
FIG. S4. (a) Transmission spectra of CsPbI3 NCs (sample #1) measured from cryogenic to room temperature. The spectra
are shifted vertically for clarity. The arrow indicates the center of the curve E0 . (b) Temperature dependence of E0 . The line
shows the fit with Eq. S17.

F200X machine of Thermo Fisher with the acceleration voltage of 200 kV, current of 130 pA and high-angle annular
dark field detector (0.16 nm resolution) [1024×1024 pixel Thermo-Fisher SuperX]. Due to the dielectric glass matrix
in which the CsPbI3 NCs are embedded, charges can easily accumulate on the samples and, therefore, the scanning
fails due to distraction of the electron beam. The best results can be achieved on the edges of the powder grains.
Overall, on several grains dozens of sharp or blurry bright spots can be identified as the CsPbI3 NCs, e.g. shown in
Fig. S5(a). Typically the NCs remain blurry, thus the real morphology is hidden but from the sharp NC images a
non-cuboid almost spherical shape can be found.

An elemental analysis was performed to confirm the presence of CsPbI3 NCs, Fig. S5(b-d). Interestingly, the
analysis shows a rather homogeneous distribution of all elements over the full grain. Noteworthy, within the presented
cut of the image the full area shows the presence of the elements (black/white for the absence). Within the glass
melt still a high amount of the perovskite-forming elements are solved. However, for iodine and lead at some spots a
higher concentration of the elements can be seen. These spots of high concentration coincide with the positions where
before the NCs were identified, thus confirming the presence of NCs. Note that in order to have a better contrast,
the images were post-processed with the graphic software inkscape. The brightness to contrast ratio and color were
tuned, further a slight blurring was applied and image stacking was used. The rough estimation of the NC density
within the 2D images is 2 × 1010 cm−2 . A precise recalculation to the 3D density is not possible due to the unknown
NC thickness and hidden NCs at higher depth. However, we obtain an estimate of 3 × 106 measured NCs within the
200 µm laser spot.

Finally, for a larger set of measured grains the NC size was evaluated using the calibrated ruler. The size distribution
is shown in the histograms in Fig. S6(b-d). Another example of a STEM image for sample #3 with smaller NCs is
shown in Fig. S6(a). For evaluation of the NC size the software ImageJ, and partially contrast post-processing were
used. For the image-post processing the vanilla STEM images are sharpened by a Microsoft PowerPoint algorithm
and further the brightness and contrast settings adjusted by ImageJ. Afterwards the NC size was evaluated by an
approximation via two lines placed manually inside the NC to take into account the slight elliptical shape of the
NCs. In total about 1000 NCs were evaluated. Overall, NCs with sizes varying from 8 up to 16 nm were found.
The histograms were fitted assuming Gaussian distributions, neglecting the asymmetry of having more larger than
smaller NCs, from which average sizes of 13.8 nm (sample #1), 11.7 nm (sample #2), and 10.7 nm (sample #3) were
evaluated. A certain trend of having a broader size distribution for a larger average size is present, originating from
difficulties in the growth control.
 S8

(a) (b)

(c) (d)

FIG. S5. (a) STEM-HAADF images of glass grains with CsPbI3 NCs (sample #1) measured at room temperature. The
bright spots are NCs. The plateau-like brightness steps result from the sharpening process to better identify the NCs. (b)
STEM-HAADF image of another grain which was analyzed for elemental composition in (c-d). (c,d) Detected iodine (c) and
lead (d) signal within the sample region presented in (b).

S5. LATTICE CONSTANT OF CUBIC MATERIALS

In ETB we use the lattice constant of cubic materials. We comment on these values as the corresponding data
presented in different publications are not fully consistent. This is mainly explained by the fact that bulk cubic
perovskites are not stable at low temperatures. In Ref. S13, for cubic CsPbI3 , the high-temperature experimental
value 0.6289 nm from Ref. S14 is used and for cubic CsPbCl3 they use the value of 0.5605 nm extrapolated from
other experimental data in Ref. S15. In Ref. S15 one may also find the lattice constant 0.5874 nm for cubic CsPbBr3 .
Note that the high-temperature lattice constant of cubic CsPbI3 extrapolated to room-temperature[S14] would give
0.6249 nm.
In Ref. S16, the values 0.6238, 0.5865, and 0.5610 nm are used for I-, Br- and Cl-based cubic perovskites. Calcu-
lations using the WIEN2k package (version 23.2) with the PBEsol exchange-correlation potential and default high-
precision settings (-prec 3n) give respectively 0.6243, 0.5856, and 0.5619 nm. These values are in reasonable agree-
ment with experimental data on the lattice constants of low-symmetry CsPb(I,Br)2 in Ref. S17.

[S1] E. Kirstein, D. R. Yakovlev, M. M. Glazov, E. A. Zhukov, D. Kudlacik, I. V. Kalitukha, V. F. Sapega, G. S. Dimitriev,

M. A. Semina, M. O. Nestoklon, E. L. Ivchenko, N. E. Kopteva, D. N. Dirin, O. Nazarenko, M. V. Kovalenko, A. Baumann,
J. Höcker, V. Dyakonov, and M. Bayer, Nat. Comm. 13, 3062 (2022).
[S2] S. Tao, I. Schmidt, G. Brocks, J. Jiang, I. Tranca, K. Meerholz, and S. Olthof, Nature Communications 10, 2560 (2019).
 S9

(a) (b) 40 300

13.8nm ± 3.8nm

Σ Counts (#) Σ Counts (#)

Count (#)
#3 #1 200
20
100

0 0
(c)
11.7nm ± 3.4nm

Count (#)
#2
100
500

0 0
(d) 60 300
10.7±1.9 nm

Σ Counts (#)
Count (#) #3
40 200

20 100

0 0
0 10 20
size (nm)
FIG. S6. (a) STEM-HAADF image of a glass grain with CsPbI3 NCs (sample #3). (b-d) Histograms of the NC size distribution
(diameter) for samples #1, #2 and #3, respectively (left axis), binned by 1 nm; the right axis shows the sum of NC counts for
specific sizes.

[S3] G. F. Koster, J. O. Dimmock, R. G. Wheeler, and H. Statz, The Properties of the Thirty-Two Point Groups (M.I.T.
Press, Cambridge, 1963).
[S4] I. D. Avdeev, S. V. Goupalov, and M. O. Nestoklon, Phys. Rev. B 107, 035414 (2023).
[S5] A. A. Kiselev, E. L. Ivchenko, and U. Rössler, Phys. Rev. B 58, 16353 (1998).
[S6] E. L. Ivchenko, Optical spectroscopy of semiconductor nanostructures (Alpha Science, Harrow UK, 2005).
[S7] G. Bir and G. Pikus, Symmetry and Strain-Induced Effects in Semiconductors (Wiley, New York, 1974).
[S8] G. E. Pikus and A. N. Titkov, in Optical Orientation, Modern Problems in Condensed Matter Sciences, Vol. 8, edited by
F. Meier and B. Zakharchenya (Elsevier, 1984) pp. 73–131.
[S9] R. Winkler, Spin-orbit Coupling Effects in Two-Dimensional Electron and Hole Systems (Springer, Berlin, 2003).
[S10] J. M. Luttinger and W. Kohn, Phys. Rev. 97, 869 (1955).
[S11] X. Zhang, G. Pang, G. Xing, and R. Chen, Materials Today Physics 15, 100259 (2020).
[S12] S. Yu, J. Xu, X. Shang, E. Ma, F. Lin, W. Zheng, D. Tu, R. Li, and X. Chen, Advanced Science 8, 2100084 (2021).
[S13] R. A. Jishi, O. B. Ta, and A. A. Sharif, J. Phys. Chem. C 118, 28344 (2014).
[S14] D. Trots and S. Myagkota, Journal of Physics and Chemistry of Solids 69, 2520 (2008).
[S15] R. L. Moreira and A. Dias, Journal of Physics and Chemistry of Solids 68, 1617 (2007).
[S16] M. A. Becker, R. Vaxenburg, G. Nedelcu, P. C. Sercel, A. Shabaev, M. J. Mehl, J. G. Michopoulos, S. G. Lambrakos,
N. Bernstein, J. L. Lyons, T. Stöferle, R. F. Mahrt, M. V. Kovalenko, D. J. Norris, G. Rainò, and A. L. Efros, Nature
553, 189 (2018).
[S17] M. C. Brennan, S. Draguta, P. V. Kamat, and M. Kuno, ACS Energy Letters 3, 204 (2018).