4D nanoimaging of early age cement hydration

Abstract

Despite a century of research, our understanding of cement dissolution and precipitation processes at early ages is very limited. This is due to the lack of methods that can image these processes with enough spatial resolution, contrast and field of view. Here, we adapt near-field ptychographic nanotomography to in situ visualise the hydration of commercial Portland cement in a record-thick capillary. At 19h, porous C-S-H gel shell, thickness of 500 nm, covers every alite grain enclosing a water gap. The spatial dissolution rate of small alite grains in the acceleration period, ∼100 nm/h, is approximately four times faster than that of large alite grains in the deceleration stage, ∼25 nm/h. Etch-pit development has also been mapped out. This work is complemented by laboratory and synchrotron microtomographies, allowing to measure the particle size distributions with time. 4D nanoimaging will allow mechanistically study dissolution-precipitation processes including the roles of accelerators and superplasticizers.

Full text

1
Article submitted to Nature Communications
4D nanoimaging of early age cement hydration
Shiva Shirani1, Ana Cuesta1, Alejandro Morales-Cantero1, Isabel Santacruz1, Ana Diaz2, Pavel
Trtik3, Mirko Holler2, Alexander Rack4, Bratislav Lukic4, Emmanuel Brun5, Inés R. Salcedo6 and
Miguel A. G. Aranda1*
Despite a century of research, our understanding of cement dissolution and precipitation
processes at early ages is very limited. This is due to the lack of methods that can image these
processes with enough spatial resolution, contrast and field of view. Here, we adapt near-field
ptychographic nanotomography to in situ visualise the hydration of commercial Portland
cement in a record-thick capillary. At 19h, porous C-S-H gel shell, thickness of 500 nm, covers
every alite grain enclosing a water gap. The spatial dissolution rate of small alite grains in the
acceleration period, ∼100 nm/h, is approximately four times faster than that of large alite grains
in the deceleration stage, ∼25 nm/h. Etch-pit development has also been mapped out. This work
is complemented by laboratory and synchrotron microtomographies, allowing to measure the
particle size distributions with time. 4D nanoimaging will allow mechanistically study
dissolution-precipitation processes including the roles of accelerators and superplasticizers.
1Departamento de Química Inorgánica, Cristalografía y Mineralogía, Universidad de Málaga, 29071-Málaga,
Spain. 2Laboratory for Macromolecules and Bioimaging, Paul Scherrer Institut, 5232 Villigen PSI,
Switzerland. 3Laboratory for Neutron Scattering and Imaging, Paul Scherrer Institut, 5232 Villigen PSI,
Switzerland. 4ESRF The European Synchrotron, 71 Rue des Martyrs, 38000 Grenoble, France. 5Université
Grenoble Alpes, Inserm UA7 STROBE, 38000 Grenoble, France. 6Servicios Centrales de Apoyo a la
Investigación, Universidad de Málaga, 29071-Málaga, Spain. email: g_aranda@uma.es

2
Portland concrete is the world’s largest fabricated commodity, ∼20 billion tonnes/yr. The enormous
production of Portland cement (PC), ∼4 billion tonnes/yr, results in ∼2.7 billion tonnes/yr of CO2
emissions1. Therefore, there are many attempts to decrease the cement CO2 footprint1,2. In order to
rationally design approaches to decrease the embodied carbon content of binders, maintaining the
performances, cement hydration understanding is key. Unfortunately, there are many unanswered
questions3 regarding the complex dissolution and precipitation processes that lead the setting and
early hardening of cements4.
The hydration of PC can be divided into five periods3, see Fig. 1b. Stage-I is the initial dissolution
(first minutes); stage-II is the low activity, induction, period (some hours); stage-III is the acceleration
(several hours until the maximum of the heat flow trace); stage-IV is the deceleration (tens of hours);
and stage-V is the diffusion-controlled hydration (months to years). There was a strong debate about
the mechanism responsible of the induction period. However, it is now accepted that it is the
dissolution controlled by undersaturation5 and not the protective membrane theory. Conversely, there
is no agreement in the mechanism(s) to explain the transition from acceleration to deceleration, when
there is a degree of hydration of just 10-20% and plenty of space for the hydrates to grow. The most
advanced theories, recently discussed6, are based on heterogeneous nucleation and growth within
confined regions taking into account the initial particle size distributions7, for instance, see the
reaction zone hypothesis8. There are alternatives9 like the needle model6 where alite, the most
abundant component of PC, hydrates10,11 to yield nonstoichiometric calcium-silicate-hydrate (C-S-H)
gel,12 which nucleates and grows as needles. Neither the dissolution of small grains, nor water
diffusion, nor etch‐pits coalescence, nor C-S-H gel impingement −alone− can currently explain the
transition from the acceleration to the deceleration periods3. The factors affecting the C-S-H gel
growth in these two periods, III and IV, are not known. Moreover, the role of etch pits13 needs to be
better understood as well as the consequences of the spatial gap which opens between the dissolving
(inward) alite grains and the growing (mainly outward) C-S-H gel14,15. Finally, the density evolution
of the C-S-H gel shells is also unknown.
On the one hand, in situ laboratory16 and synchrotron17 powder diffraction allow following the
phase development with time. These studies yield volume-averaged information which misses any
spatial feature like particle size dependence. On the other hand, electron microscopy (EM) techniques
yield very valuable information, with high spatial resolution, but they only give snapshots, as the
experimental conditions are not compatible with the hydration in relevant conditions. In situ
tomography18,19 can contribute to filling this gap. In cements, modelling20 and microstructural
characterization methods21 acknowledge the growing importance of X-ray microtomography (µCT)
in their different modalities22. Moreover, µCT is being widely used23 and in particular to follow in
situ 4D (3D + time) some specific features of cement hydration24–33,34,35. However, none of these
works combine the four required features for relevant contributions to the understanding of the
mechanism(s) of Portland cement hydration at early ages: (i) water to cement mass ratio (w/c) close
to 0.50, (ii) submicrometer spatial resolution, (iii) good contrast to be able to identify the different
evolving components (more than eight), and (iv) relatively large scanned volume to allow hydration
to progress with appropriate particle sampling, the particle sizes of commercial PCs have Dv,50∈10-
20 µm.
So far, ptychographic X-ray computed tomography (PXCT)36, which merges scanning X-ray
microscopy and coherent diffraction imaging37–39, met the first three requirements. Hence, it was
applied to several binders within capillaries of ∼40 µm of diameter40,41 using a photon energy of 6.2
keV. The second study41 provided valuable information about C-S-H gel hydrated for 5 months: an
average stoichiometry of (CaO)1.80(SiO2)(H2O)3.96 with a mass density of 2.11 gcm-3 and an electron
density of 0.64 e-Å-3. Moreover, it allowed quantifying a 6.4 vol% of a second amorphous component,
iron-siliceous hydrogarnet, with 2.52 gcm-3 and 0.76 e-Å-3. Here, we have used PXCT in near-field
configuration39,42 to acquire data in a record-thick capillary of ∼160 µm, employing a higher photon
energy, 8.93 keV. This configuration, and the iterative algorithms that allowed the reconstructions,39
now meet simultaneously the four stringent requirements with current data collections of 3-4 hours
allowing 4D measurements of spatially resolved data during the first four days of cement hydration.

3
On the one hand, the relatively slow overall acquisition time is the main limitation of this work, but
this is expected to improve, see last section. On the other hand, the excellent spatial resolution and
contrast of ptychographic nanotomography gave quantitative values of relevant parameters in the
dissolution-precipitation processes like alite spatial dissolution rates and etch-pit growth rate. These
values can help to test the above described models.
Result and discussion
Initial cement analysis and cement hydration study. Two commercial PCs have been used and
their laboratory X-ray powder diffraction (LXRPD) patterns were analysed by the Rietveld method,
see Figure S1. The elemental and mineralogical analyses are given in Tables S1 and S2, respectively.
The cements have very similar compositions but differ in their textural properties, see Fig. 1a and
Table S3. The specific surface areas for PC-52.5 and PC-42.5 were 2.27 and 1.25 m2g-1, respectively.
PC-52.5, with finer particles Dv,50∼12 µm, was used for the PXCT study in order to have more
hydrating particles in the analysed volume. It is noted that, in a very recent nanotomographic study,7
researchers milled alite very extensively, all particles<10 µm, in order to fit them within a small field
of view (FOV) of ∼50 µm.
The calorimetric study, see Fig. 1b, gave the hydration kinetic features at the relevant w/c ratios of
0.50 and 0.40. Table S4 reports the cumulative heat at the hydration times where the imaging data
were acquired. Moreover, it also lists the degree of hydration (DoH) at those times to be used as a
reference in the imaging studies; for a detailed explanation of this type of calculation, see reference43.
As expected, PC-52.5 releases more heat than PC-42.5, mainly because of its finesse. The DoH at the
maxima of the heat flow peaks were 19% and 10% for PC-52.5 and PC-42.5, respectively. These
values are well-known3,44,45 but they cannot be explained by current models.
Fig. 1. Initial characterization of the employed Portland cements. a, Textural analysis of the two
cements: PC-52.5 and PC-42.5, with Blaine values of 409 and 368 m2kg-1, respectively. (top) Particle size distributions,
(bottom) cumulative volume variation as measured by laser scattering. b, Isothermal calorimetric study, T=25 ºC, for the
cement pastes prepared with w/c values of 0.50 and 0.40, referenced to 1 g of anhydrous cement. (Top) Heat flow curves
(where the typical hydration stages are sketched), (bottom) cumulative heat traces. c, Laboratory Rietveld plots (MoKα1
radiation, λ=0.7093 Å) for PC-52.5 paste, w/c=0.40, within a capillary of 1 mm of diameter. The main diffraction peaks
are labelled with the contributing crystalline component by using the cement notation: tetracalcium aluminoferrite (C4AF),
belite (C2S), alite (C3S), tricalcium aluminate (C3A), calcite (Cc), portlandite (CH) and ettringite (AFt). (Top) Pattern
collected at 22 h, (bottom) pattern collected at 96 h of hydration.

4
Fig. 1c displays the LXRPD Rietveld fits for hydrating PC-52.5 taken in the same capillary where
the Lab-µCT imaging study was carried out. The in situ LXRPD data were analysed by the Rietveld
method resulting in the quantitative phase analyses reported in Table S5. From these data, the DoHs
of the different clinker phases are derived at 22, 50 and 96 h of hydration, also used as reference.
In situ multicontrast X-ray tomographic studies of cement hydration. Two additional in situ X-
ray imaging studies were carried out to place the results of the PXCT nanoimaging in context.
Emphasis is placed on the accuracy of the results that can only be estimated by comparison. The
FOVs were cylinders. Fig. 2a displays one orthoslice for each work: (i) Lab-µCT for PC-52.5-
w/c=0.40, FOV=1000×940 µm (φ×L); (ii) Syn-µCT for PC-42.5-w/c=0.50, FOV=700×1190 µm; and
(iii) PXCT for PC-52.5-w/c∼0.40, FOV=170×30 µm. Syn-µCT data has a larger w/c value, showing
higher porosity (the darkish micro-regions) see the central panel in Fig. 2a.
Fig. 2b shows enlarged views to illustrate the evolution of the different components and the spatial
resolutions. The PXCT study has a much higher resolution and contrast at the expense of a smaller
FOV. A minor artefact can be seen in the Syn-µCT data as some borders have grey-values too white.
This is a common feature for inline propagation-based data that cannot be fully corrected with the
employed Paganin algorithm46. Fig. 2c shows the histogram evolutions with time. At 19 h (blue
traces), there is plenty of free water, which displaces the main peak toward smaller grey-
values/electron densities. As hydration progresses, and as expected, the main peak in the histograms
densifies and the amount of clinker components decreases. The histogram evolutions for the same
cement paste from Lab-µCT and PXCT, see corresponding panels in Fig. 2c, were very similar giving
confidence to the relevancy of the nanoimaging results in spite of the limited amount of volume
scanned to have submicrometer resolution.
Fig. 2. In situ multicontrast X-ray tomographic studies of cement hydration. a, Selected orthoslices
for the three imaging approaches. (Top) Attenuation-contrast laboratory data (Lab-µCT) for a PC-52.5 paste with
w/c=0.40. (Middle) Inline propagation-based phase-contrast synchrotron data (Syn-µCT) for a PC-42.5 paste with
w/c=0.50 [phases retrieved by using the Paganin algorithm46], (bottom) Quantitative phase-contrast, phases retrieved by
near-field PXCT for a PC-52.5 paste scanning a region with w/c∼0.40. The thicknesses of the capillaries are given. b,
Enlarged views for the three approaches, at the three hydration ages, to qualitatively illustrate the quality of the data
(contrast and spatial resolution). Clinker minerals are seen as whitish particles, porosity as darkish regions, and hydrates
have intermediate grey tones. c, Histograms for the different tomographic studies and hydration times. The histograms
were obtained by computing the largest possible volumes without including the glass capillary walls.
The spatial resolutions were characterised by two approaches. Figures S2-S4 display line profiles
of sharp interfaces between high and low density components. Additionally, Fourier-shell-correlation
(FSC)47 plots are displayed in Figures S5-S7. Taken all together, the spatial resolution of the

5
tomograms is close to 2 voxels for the three approaches. More details are given in the Supplementary
Information (SI). Moreover, Fig. 3 compares the level of details that it can be obtained with Syn-µCT
and PXCT at 19 h of hydration. The latter shows the C-S-H gel shell and a porosity gap between the
shell and the dissolving alite particle that it is not observed in the Syn-µCT data because the lack of
spatial resolution and contrast.
Fig. 3. Comparison of phase-contrast synchrotron tomography and ptychographic X-ray
computed tomography. Left, Enlarged view of Syn-µCT data [voxel size: 650 nm] of PC hydration showing the
hydrating alite particle (whitish) surrounded by C-S-H reaction product (greyish). These reaction products are highlighted
with red arrows. Right, Enlarged view of the PXCT data [voxel size: 186 nm] of PC hydration. The C-S-H gel shells
surrounding the alite particles are clearly visible (pink arrows). There is a water gap between the shell and the alite grain
due to the inward dissolution of alite. Moreover, etch-pits on the surfaces of the alite particles are also visible. Selected
etch-pits are highlighted with blue circles. The highest spatial resolution and better contrast of PXCT data allow to
visualize submicrometre features of the dissolution-precipitation processes which are not visible in propagation-based
Syn-µCT.
4D nanoimaging of cement hydration. In situ near-field PXCT data were taken as detailed in
methods. To ensure the relevance of the results, the scanned volumes were assessed. Firstly, the w/c
ratio was 0.41(2), as determined from the absorption data41, see Table S6. This w/c value is fully
consistent with the obtained degrees of hydration. Secondly, possible signatures of radiation damage
were explored. The mean electron density values of the whole sample were 0.600, 0.599 and 0.591 e-
Å3, for the 19, 47 and 93 h datasets, respectively. The spatial resolution from FSC was 470 and 500
nm, for the 47 and 93 h data, respectively. Hence, radiation damage cannot be discarded but it is
small, if any. Thirdly, seven sets of components within the tomograms were identified: air, water,
AFt/C-S-H gel/others, CH, Cc, C3A/C3S/C2S and C4AF, using the electron density and the absorption
data in the bivariate plots, see Figures S8 and S9. The calculated electron density and attenuation
length values are given in Table S7, and Figure S10 displays the histogram evolution in logarithm
scale.
The crux of our results is the 4D submicrometer features of cement hydration, see Figs. 4 to 6. It
is underlined that PXCT readily distinguishes air and water porosities. The paste evolution is
displayed in Fig. 4a, showing a partly reacted binder plenty of capillary water at 19 h. The main
change from 19 to 43 h is the large consumption of capillary water (dark-grey regions in Fig. 4a-19h)
and the densification of C-S-H gel. The main evolution from 47 to 93 h is the appearance of chemical
shrinkage, evidenced by the development of many air-containing (black) regions. Importantly, Fig.
4c and Figs. S11-S12 show the evolution of etch pits with time, including etch pit coalescence, which
contributes, in addition to the consumption of small alite particles, to the decrease of specific surfaces
in the deceleration period. The etch pit growth rate, from 19 to 93 h, was estimated as ∼1 µm/d. Etch
pits at very early ages, i.e. 2-4 h, were imaged by EM with sizes of ∼15 nm48 which is out of in situ
PXCT capabilities.

6
Chiefly, the spatial dissolution rate of alite was determined, see Fig. 5a and Figs. S13-S18. From
22 measurements along different particles, the dissolution rate between 19 and 47 h was 25(14) nm/h.
This value compares well with 36 nm/h from the reaction zone model8 but poorly with 84 nm/h
obtained from the same dataset by using the boundary nucleation and growth model49. It is noted that
from 47 to 93 h, the dissolution rate of relatively flat faces of alite drops sharply but the etch-pits
growth (dissolution) rate does not significantly decrease. Moreover, the analysis of the 12 largest
Hadley grains (hollow-shell microstructure, see Fig. 6a-b and larger view at S19)14,15,50,51 gave 2.6(3)
µm of size. Here, their time-evolution can be followed and it is noted that hollow regions are filled
with water at 19 and 47 h but dried at 93 h, see Fig. 6a, directly evidencing the water diffusion through
the C-S-H shells. Additionally, the analysis of the 15 smallest alite particles leaving an unhydrated,
very small, core gave an average particle size of 3.4(5) µm. Therefore, it is concluded that alite
particles smaller than ∼3.0 µm are fully hydrated in the 4-19 h range, leading to a spatial dissolution
rate of ∼100 nm/h. The largest size of the Hadley grains found here agrees well with previous
works50,51 (at 24 h) reporting a maximum size of 5.0 µm with shells of 500 nm. Hence, it seems to be
a 3-4 fold difference between the spatial dissolution rate of small alite grains in the acceleration period
and that of large alite grains in the deceleration stage.
Fig. 4. 4D nanoimaging study highlighting the etch-pit evolution. a, PXCT orthoslices at the studied
hydration ages showing the evolution of the PC-52.5-w/c∼0.40 paste which includes: i) dissolution of cement particles,
ii) portlandite growth, iii) C-S-H gel densification at 2 and 4 days, and iv) chemical shrinkage. Examples of these features
are labelled with red arrows. b, 3D rendering of a volume including a fraction of large alite particle highlighted in a, to
show the evolution of five selected etch-pits, which are labelled. These images also show the full reaction of small alite
particles, featured with blue circles. c, 3D representation of the segmented particle, shown in the previous panels, to
highlight the evolution of the etch pits. It is noted that three branches within etch-pit #1 at 47 h coalesce at 93 hours of
hydration which may mean a reduction in surface with hydration time.

7
Fig. 5. C-S-H gel shells and their evolution. a, Selected 2D views of the PXCT data. The fourth panel display
the electron density profiles corresponding to the straight lines in the previous plots. At 19 h, the C-S-H shell covers every
alite particle. The line profile at 19 h, blue trace in the fourth panel, shows water porosity (gap) between the alite particle
and its shells. For these two particles, the shells have a thickness of ∼550 nm and an electron density of ∼0.47 e-Å-3. This
C-S-H densifies at 2 days to ∼0.53 e-Å-3. The alite particles partly dissolve and air porosity starts to develop (black
regions). b, 3D rendered views of the ML segmented C-S-H gel shells with the colour signalling its thickness. (left) View
superimposing the C-S-H shells to half of the studied capillary. (Right) 3D view of the segmented C-S-H shells.
Fig. 5a and Figs. S13-S16 show views detailing the hydration of alite particles and showing the C-
S-H gel shells that surround all alite grains enclosing a gap. Similar plots containing belite particles,
Figs. S17-S18, did not show gaps, as expected. C-S-H shells on alite have been extensively analysed
by EM6,14,15,50,52, but here their electron density and spatial evolutions can be followed. The C-S-H
shells for these two alite particles, Fig. 5a, have ∼0.47 e-Å-3 which increases to ∼0.53 e-Å-3 at 47 h.
This means a very low mass density shell. For instance, ettringite, a phase with 32 crystallization
water molecules has ∼0.56 e-Å-3, and mature C-S-H gel, with (CaO)1.8(SiO2)(H2O)4.0 composition,
i.e. including gel pore water, had 0.64 e-Å-3.41 As this result is critical for discarding diffusion as the
mechanism for the deceleration period, a larger study was carried out. 20 shells were analysed giving
0.51(4) e-Å-3 and 500(120) nm, for the average electron density and thickness, respectively.

8
Moreover, this study also yields the average width of the gap, 490(140) nm. At 24 h, the thickness of
the shells and gaps were reported as ∼500 and ∼300 nm from EM51. Moreover, a gap of 490 nm
developed in 15 h, time between the end of the induction period and the first nanoimaging
measurement, which means an alite spatial dissolution rate of ∼33 nm/h. Therefore, the dissolution
rates of large and small alite grains differ at least three-fold, which should be considered for
modelling. To further study the C-S-H shells, the components were segmented by Machine Learning
(ML) as described in methods following the procedure summarized in Fig. S20. Subsequently, the C-
S-H shells were segmented as detailed in Fig. S21. The results of the C-S-H shell segmentation are
presented in Fig. 5b with a relatively constant shell size of ∼450 nm. Fig. S22 shows a 2D comparison
of the raw data and the resulting segmented shells yielding a reasonable good agreement.
Finally, Fig. 6 shows interesting (directly-observed) nano-features. Figs. 6a-b exhibit the water
diffusion through the C-S-H shells for some Hadley grains between 19 and 93 h of hydration. Larger
regions are shown in Figs. S19 and S23, respectively. Fig 6c displays the water porosity evolution
within a calcite particle evidencing that some small pore regions, 2-3 µm in size, were water filled at
19 and 47 h and dried at 93 h, see also Fig. S24. This observation directly explains the indirect result
of water transport within limestone grains obtained by X-ray dark-field tomography30. Fig. 6d, larger
view in Fig. S25, shows how hydration progresses along the surface of an alite particle but it stops at
C4AF intergrown regions, evidencing the importance of alkanolamines at early ages as accelerators53.
Moreover, Fig 6d also illustrates that C-S-H gel can dissolve to leave a dry capillary pore. Fig. S26
presents at 47 h the full dissolution of a grain of a size of 4 µm at 19 h. This implies a very large
dissolution rate, >75 nm/h, which is likely C3A.

9
Fig. 6. Hydration time evolution of selected nano-features directly visualised by near field
PXCT. a, hollow-shell microstructure, also known as Hadley grain. The Hadley grains are fully hydrated small alite
particles that contain a void within the original boundary of the anhydrous grain. The hollow regions are filled with water
at 19 and 47 h (blue arrows) but dried at 93 h (red arrow) directly evidencing the porosity of the C-S-H shells. b, evolution
of water porosity (dark-grey) to air porosity (black). c, evolution of water porosity inside a calcite grain, if connected to
the surface. d, evolution of alite dissolution (hydration) which stops at the C4AF intergrown regions, highlighted by brown
arrows. This panel also illustrates that (recently precipitated) hydrates can dissolve, blue rectangles.
Segmentations of the X-ray tomographies. Fig 7a displays orthoslices of the trained models
overlaid on the three raw datasets and Fig 7b shows the time evolution of the segmented components.
For the PXCT study, the average electron densities and the segmented volumes are reported in Table
S8. Moreover, Fig. S27 displays the evolution of the segmented water capillary porosity. The amount
of anhydrous components, together with the initial values, allow determining the DoH from the three
imaging studies which are reported in Table S9 and summarised in Fig 7b. Table S9 also gives the
DoH from calorimetry and LXRPD as references. The agreement for Syn-µCT is noteworthy likely
due to the large volume probed and the relatively good spatial resolution. The DoHs from Lab-µCT
are overestimated probably due to the relatively poor spatial resolution and contrast, which do not
allow us to quantify the anhydrous particles smaller than ∼3 µm. Finally, PXCT yields underestimated
values for the DoH likely due to the limited scanned volume.

10
Fig. 7. ML training and segmentation results for the three datasets with different contrast
mechanisms. a, ML trained models overlaid on the three raw datasets. b, 3D rendering of the segmented volumes at
the three studied hydration ages. The DoH values determined from microtomography (bold) are compared to the ones
from calorimetry (italics). The number of quantified components in the Lab-µCT and Syn-µCT datasets are four: i)
porosity (air and water), ii) LDH (low-density hydrates: mainly C-S-H gel and ettringite), iii) HDH (high-density hydrates:
mainly portlandite and calcite), and iv) UCP (anhydrous cement particles: all unreacted clinker phases). The number of
quantified components in the PXCT datasets is seven: i) air porosity, ii) water porosity, iii) LDH (low-density hydrates:
mainly C-S-H gel and ettringite), iv) portlandite, v) calcite, vi) C3A/C3S/C2S, and vii) C4AF.
Importantly, the high resolution and excellent contrast of the PXCT study allowed us to track down
the individual hydration evolution of 1407 particles with connected anhydrous volume, at 19 h, of 1
µm3 or larger. For this particle tracking statistical study, they were arbitrarily classified into four
groups having connected volumes (µm3) of 1.0≥vol1>27.0≥vol2>216.0≥vol3>1000.0≥vol4. A rough
estimation of their sizes in µm could be ∛(vol) or 1.0≥size1>3.0≥ size2>6.0≥size3>10.0≥ size4. The
groups contained 1117, 204, 61 and 20 particles, respectively, and the corresponding percentages
with respect to the UCP volume were 5.4, 12.9, 21.5 and 60.2%. It is not possible to calculate the
DoH of these groups at 19 h because there are no reference values at t0. The degree of reaction,
between 47 and 19 h, was 69.4%, 59.3%, 37.2% and 15.9% for the groups classified as vol1, vol2,
vol3 and vol4, respectively, and they account for 12.9, 26.4, 27.7 and 33.0% of the total dissolved
volume. In other words, the twenty particles of vol4 group has a degree of reaction of 15.9 but it
accounts for 33% of the dissolved volume in that period. A similar study can be carried out between
93 and 47 h. The degree of reaction during this period decreased to 14.7%, 11.6%, 6.7% and 3.6%
for the corresponding groups. In terms of dissolved volume percentage in this period, the values are
6.8, 16.9, 25.3 and 51.0%, respectively. This simple analysis shows that 51% of the observed reaction
during this diffusion-limited period is due to the 20 largest particles as they account for a very large

11
fraction of the volume. The strong decrease of hydration rate in this time range is in line with a
diffusion controlled hydration stage.
Finally, Fig. 8a displays the hydration evolution of the segmented anhydrous cement particles. It
is readily visible that smaller particles dissolve faster than larger ones. Moreover, the segmentation
output allows us to classify the particles and to follow their volumes and cumulative volumes, see
Fig. 8b. The cumulative volume results for Lab-µCT and PXCT are very similar and in agreement
with the expected DoH from calorimetry showing a large variation between 19 and 47 h and a small
variation at 93 h. Moreover, the hydration of PC42.5 is slower because of their larger average particle
sizes, with similar DoH variation between the two studied time intervals, see Fig 8b, in full agreement
with their calorimetric traces, see Fig. 1b. This reflects a good accuracy of the obtained results.
Fig. 8. Hydration study as a function of time and cement particle sizes. a, 2D views of the segmented
anhydrous cement particles as a function of time. b, Volumes and cumulative volumes for the anhydrous cement particles
as a function of the particle size, represented in logarithmic scale for easy comparison with Fig. 1. The particle sizes are
computed as the mean Feret diameter frequency which measures object size along directions. The particles are represented
grouped in sets of 2.5 microns, i.e. 0-2.5, 2.5-5.0 µm, etc. The maximum and Dv,50 values for the Lab-µCT and Syn-µCT
are given in the panels. These data are scattered for PXCT because the limited height of the studied cylinder yields a poor
representative elementary volume for this feature.
Implications and outlook
The hydration of PC, in relevant conditions, has been measured with unprecedented spatial resolution
and contrast. As expected but not directly measured so far, the hydration of Portland cement at 1 day
or earlier is dominated by the small particles, smaller than 3 µm, and the hydration after 3 days is
very much dependent of the large alite particles, larger than about 10 µm. The nanoimaging work
shows the C-S-H gel shells surrounding every alite grain which does not preclude diffusion. The
measured alite spatial dissolution rates, ∼100 nm/h for small grains in the acceleration period and ∼25
nm/h for large particles in the deceleration stage impose constraints on the cement hydration models.
Etch-pit growth rate, ∼40 nm/h, and coalescence have also been measured but better spatial resolution

12
is required for etch-pit quantification. The configuration employed here already allows studying the
roles of admixtures (accelerators, retarders, superplasticizers, etc.) by measuring C-S-H shell density,
C-S-H gel spatial distribution and alite spatial dissolution and etch-pit growth rates. For instance, it
would be possible to measure if the acceleration produced by CaCl2 is due to a lower density of the
C-S-H shells (higher water diffusion) or if it is mainly due to higher calcium supersaturation.
The current spatial resolution of in situ near-field PXCT, ∼370 nm, can be improved by increasing
the number of projections, without larger acquisition times. Moreover, so far the time resolution is
modest, i.e. 3-4 h for a complete tomogram, but this is expected to improve in fourth-generation
synchrotron sources with tailored beamlines for ptychography. Time resolutions of ∼1 h will open the
way to accurately study the processes in the acceleration period. However, in these cases, radiation
damage could be an issue if the total dose is not kept low. Higher spatial resolution and shorter
acquisition times will allow to thoroughly study the acceleration stage which will be beneficial to
rationally design new accelerator admixtures with the final aim of developing low carbon cements
with competitive mechanical strengths and durability performances. Moreover, dissolution-
precipitation processes with moderate reaction rates take place also in several other fields like
geochemistry or biomaterials, which could benefit from the reported investigation.
Methods
Material provenance and initial characterization. Two different types of commercial cement were used: a
CEM I 52.5 R (PC-52.5) and a CEM I 42.5 R (PC-42.5) which conform to EN 197–1. The full
characterization of these anhydrous materials is performed with X-ray Fluorescence (XRF) and Laboratory
X-ray Powder Diffraction (LXRPD) using the Rietveld Method, see Fig. S1 and Tables S1, S2. Moreover, a
polycarboxylate ether (PCE) superplasticiser, containing 35 wt% of the active matter, was employed to
efficiently fill the very narrow capillaries required for the PXCT nanoimaging study. The characterisation of
this PCE has been recently reported54.
Textural analysis. The textural characterization was carried out in dry conditions. Particle Size Distribution
(PSD) data were measured by laser diffraction in a MasterSizer 3000 equipment (Malvern). Specific surface
areas were determined by N2 adsorption isotherms in ASAP 2420 equipment (Micromeritics, USA). The air
permeabilities were measured with the Blaine fineness apparatus (Controls) according to EN 196–6. The
density of the samples was measured with a helium Pycnometer (Accupyc II 1320 Pycnometer,
Micromeritics). The resulting data are reported in Table S3.
Isothermal calorimetry. The pastes were prepared using the PC-52.5 and PC-42.5 and with w/c mass ratios
of 0.40 and 0.50, respectively. For PC-52.5, 0.43 wt% (by weight of cement) of PCE was used. Water and
superplasticizer were mixed by magnetic stirring for 1 minute. Then, the cement was mixed with the
water/suspension and shaken, for 1 minute manually and 1 minute with a laboratory vortex mixer. Finally,
the pastes were introduced into the glass ampoules. The calorimetric data were taken in an eight-channel
Thermal Activity Monitor (TAM) instrument. Data were collected for up to 7 days at 25 °C and at 20 °C and
for PC-52.5 and PC-42.5, respectively. The first 45 minutes were required for the thermal stabilization of the
system. The employed conditions mimic the ones used for the imaging studies.
Laboratory X-ray Powder Diffraction (LXRPD) and Data Analysis. The same paste prepared for the
calorimetry measurement (PC-52.5-w/c=0.40) was used to fill the capillaries of ~1 mm in diameter. LXRPD
measurements were collected on a D8 ADVANCE diffractometer (Bruker AXS) using strictly
monochromatic Mo-Kα1 radiation (λ=0.7093 Å). This diffractometer is located at SCAI, University of
Malaga. The incident beam was formed by a primary monochromator with a focusing mirror and a 2 mm
anti-scatter slit. Moreover, 2.5° Soller slits were used for the incident and transmitted beams. An EIGER
detector (from DECTRIS, Baden, Switzerland) was used which is optimised for Mo anodes. This was used
with an aperture of 4 × 21 degrees, working in VDO mode. Data collection was performed from 3 to 35° (2θ)
for 2 h and 10 min. Rietveld quantitative phase analysis was performed with GSAS software.
Laboratory X-ray computed microtomography experiment (Lab-µCT). Lab-µCT experiments were
carried out at ~25 °C for the same capillary used in the LXRPD data collection and scanning the same region
with time. Lab-µCT experiment was performed on a SKYSCAN 2214 (Bruker) scanner at SCAI, University
of Malaga. Images were obtained using an X-ray tube with a LaB6 source filament and employing a 0.25 mm

13
Al foil to minimise the beam hardening effect. This source was operated at 55 kV and 130 µA. The CCD3
detector with a physical pixel size of 17.427 µm was set in a middle position with a source-to-detector
distance of 315.449 mm and a source-to-sample distance of 9.051 mm which yielded a voxel size of 1.00 µm
(binning 2×2). Finally, the projections were acquired every 0.22⁰ over 360⁰ with a total of 1637 projections
per tomogram and using an exposure time of 1.9 s. This results in an overall recording time of 3.5 h per
dataset. Image reconstruction of the CTs was carried out using Bruker NRecon software (version 2.1.0.1) and
by applying Gaussian smoothing and beam hardening correction.
Because Lab-µCT and LXRPD data were acquired in the same capillary, but not in the same equipment,
the description of the timing is important. Lab-µCT data were collected from 17.5 h until 21.0 h, using
cement-water mixing time as the reference. This dataset is labelled as 19 h. Then, the capillary was transfer
to the diffractometer, in the same room, and the LXRPD data were taken from 21h 15 min to 23h 25min.
This dataset is labelled as 22 h. The data collections at 47 h and 93 h followed the same protocol but it is less
important at these hydration ages, as the kinetics are slower.
Synchrotron X-ray computed microtomography experiment (Syn-µCT). PC-42.5 with a w/c ratio of
0.50 was used to prepare the paste for this experiment and PCE was not required. The paste was manually
mixed for 3.5 minutes and then introduced in a capillary of ~0.7 mm of diameter. Inline propagation-based
phase contrast microtomographic data were acquired at ID19 beamline of the European Synchrotron ESRF
in Grenoble, France. The measurements were performed at 21.5 °C, temperature of the experimental hutch,
using a photon energy of 19 keV. The distance between the sample and the detector was 15 mm. The total
time to record a full tomogram was 6 minutes with 0.05 s exposure time. 6000 projection angles were
acquired over a 360 degree tomographic scan. For reconstructions, Paganin phase retrieval of the projections
was performed.46 The resulting voxel size was 0.65 µm. Further experimental details are given in the SI.
Near-field ptychographic X-ray computed tomography (PXCT). The paste employed for the in situ
nanoimaging study was PC-52.5 with a w/c ratio of 0.50 and 0.43 wt% of PCE. The suspension was mixed
with a mechanical stirrer at 800 rpm for 3.5 minutes and then introduced in a glass capillary of ~0.2 mm in
diameter. Near-field PXCT data were taken at the cSAXS beamline of the Swiss Light Source (SLS) at the
Paul Scherrer Institute, Villigen, Switzerland. Near-field ptychography42 is a variant of X-ray
ptychography37,39 in which the sample is scanned across a coherent divergent illumination and magnified
images of the sample are recorded on the detector at each scanning position. Ptychographic phase retrieval
algorithms are employed to reconstruct the complex transmission function of the specimen, with both
absorption and phase, at each angular projection. We then repeat measurements of many projections at
different incident angles of the X-rays onto the sample and combine them using standard tomographic
reconstruction methods to obtain 3D maps of the electron density and the absorption coefficient of the
specimen. In near-field ptychography, if a sufficient number of projections are recorded, the spatial
resolution is limited by the magnified pixel size, which is determined by the pixel size of the detector, the
distance between the sample and the detector, the divergence of the illumination and the position of the
specimen from the point source of the beam, i.e. the focus. We performed our measurements using a high-
stability instrument designed for high-resolution PXCT working in air55,56 and at the hutch temperature of 25
ºC, using a photon energy of 8.93 keV. The total time to record a full tomogram was between 3 and 4 hours,
including dead time during the motion of stages in between acquisitions. The voxel size was 186.6 nm.
Further experimental details are given in the SI.
Tomographic data analysis. A supervised Machine-Learning (ML) image analysis approach was used to
segment the different components of imaged samples, using the IPSDK Explorer software (version 3.2.0.0
for Windows™, Reactiv’IP, Grenoble, France). The plot profiles along the scans electron density/grey value
slices and the 3D rendering visualization were done by using Dragonfly software (version 2022.1 for
Windows™, Object Research Systems (ORS) Inc., Montreal, Canada). More information about data analysis
can be found in the SI.
Data availability
The twelve reconstructed tomograms ‘raw’ data in tiff format, and the laboratory characterisation data, can
be freely accessed on Zenodo at https://doi.org/10.5281/zenodo.7030107, and used under the Creative
Commons Attribution license.
Online content

14
Any methods, additional references, Nature Research reporting summaries, source data, extended
data, supplementary information, acknowledgements, peer review information; details of author
contributions and competing interests; and statements of data and code availability are available at
https://doi.org/10.1038/xxxxxxxxxxxxx.
Received: XXXXXXXXXXX
Published online: XXXXXXXXXXXX
References
1. Habert, G. et al. Environmental impacts and decarbonization strategies in the cement and
concrete industries. Nat. Rev. Earth Environ. 1, 559–573 (2020).
2. Monteiro, P. J. M., Miller, S. A. & Horvath, A. Towards sustainable concrete. Nat. Mater.
16, 698–699 (2017).
3. Scrivener, K. L., Ouzia, A., Juilland, P. & Kunhi Mohamed, A. Advances in understanding
cement hydration mechanisms. Cem. Concr. Res. 124, 105823 (2019).
4. Ioannidou, K. et al. The crucial effect of early-stage gelation on the mechanical properties of
cement hydrates. Nat. Commun. 7, 12106 (2016).
5. Juilland, P., Gallucci, E., Flatt, R. & Scrivener, K. L. Dissolution theory applied to the
induction period in alite hydration. Cem. Concr. Res. 40, 831–844 (2010).
6. Ouzia, A. & Scrivener, K. L. The needle model: A new model for the main hydration peak of
alite. Cem. Concr. Res. 115, 339–360 (2019).
7. Neubauer, J. et al. Evolution of the particle size distribution of tricalcium silicate during
hydration by synchrotron X-ray nano-tomography. Cem. Concr. Res. 156, 106769 (2022).
8. Masoero, E., Thomas, J. J. & Jennings, H. M. A reaction zone hypothesis for the effects of
particle size and water-to-cement ratio on the early hydration kinetics of C3S. J. Am. Ceram.
Soc. 97, 967–975 (2014).
9. Shen, D., Wang, X. & Wu, S. Determining hydration mechanisms for initial fall and main
hydration peak in tricalcium silicate hydration using a two-scale hydration simulation model.
Cem. Concr. Res. 156, 106763 (2022).
10. Pustovgar, E. et al. Understanding silicate hydration from quantitative analyses of hydrating
tricalcium silicates. Nat. Commun. 7, 10952 (2016).
11. Cuesta, A. et al. Multiscale understanding of tricalcium silicate hydration reactions. Sci. Rep.
8, 8544 (2018).
12. Allen, A. J., Thomas, J. J. & Jennings, H. M. Composition and density of nanoscale calcium–
silicate–hydrate in cement. Nat. Mater. 6, 311–316 (2007).
13. Juilland, P., Nicoleau, L., Arvidson, R. S. & Gallucci, E. Advances in dissolution
understanding and their implications for cement hydration. RILEM Tech. Lett. 2, 90–98
(2017).
14. Kjellsen, K. O. & Lagerblad, B. Microstructure of tricalcium silicate and Portland cement
systems at middle periods of hydration-development of Hadley grains. Cem. Concr. Res. 37,
13–20 (2007).
15. Gallucci, E., Mathur, P. & Scrivener, K. Microstructural development of early age hydration
shells around cement grains. Cem. Concr. Res. 40, 4–13 (2010).
16. de Matos, P. R. et al. In-situ laboratory X-ray diffraction applied to assess cement hydration.
Cem. Concr. Res. 106988 (2022) doi:10.1016/j.cemconres.2022.106988.
17. Aranda, M. A. G. Recent studies of cements and concretes by synchrotron radiation
crystallographic and cognate methods. Crystallogr. Rev. 22, 150–196 (2016).
18. Lewis, J. A. et al. Linking void and interphase evolution to electrochemistry in solid-state
batteries using operando X-ray tomography. Nat. Mater. 20, 503–510 (2021).

15
19. Zhang, X. et al. In situ imaging of amorphous intermediates during brucite carbonation in
supercritical CO2. Nat. Mater. 21, 345–351 (2022).
20. Bullard, J. W. et al. Measurement and modeling needs for microstructure and reactivity of
next-generation concrete binders. Cem. Concr. Compos. 101, 24–31 (2019).
21. Monteiro, P. J. M. et al. Advances in characterizing and understanding the microstructure of
cementitious materials. Cem. Concr. Res. 124, 105806 (2019).
22. Withers, P. J. et al. X-ray computed tomography. Nat. Rev. Methods Prim. 2021 11 1, 1–21
(2021).
23. Brisard, S., Serdar, M. & Monteiro, P. J. M. Multiscale X-ray tomography of cementitious
materials: A review. Cem. Concr. Res. 128, 105824 (2020).
24. Gallucci, E., Scrivener, K. L., Groso, A., Stampanoni, M. & Margaritondo, G. 3D
experimental investigation of the microstructure of cement pastes using synchrotron X-ray
microtomography (μCT). Cem. Concr. Res. 37, 360–368 (2007).
25. Gastaldi, D. et al. In situ tomographic investigation on the early hydration behaviors of
cementing systems. Constr. Build. Mater. 29, 284–290 (2012).
26. Parisatto, M. et al. Examining microstructural evolution of Portland cements by in-situ
synchrotron micro-tomography. J. Mater. Sci. 50, 1805–1817 (2015).
27. Hu, Q. et al. Direct measurements of 3d structure, chemistry and mass density during the
induction period of C3s hydration. Cem. Concr. Res. 89, 14–26 (2016).
28. Hu, Q. et al. Direct three-dimensional observation of the microstructure and chemistry of
C3S hydration. Cem. Concr. Res. 88, 157–169 (2016).
29. Bae, S. et al. In situ soft X-ray spectromicroscopy of early tricalcium silicate hydration.
Materials (Basel). 9, 1–16 (2016).
30. Prade, F. et al. Time resolved X-ray Dark-Field Tomography Revealing Water Transport in a
Fresh Cement Sample. Sci. Rep. 6, 29108 (2016).
31. Moradian, M. et al. Direct observation of void evolution during cement hydration. Mater.
Des. 136, 137–149 (2017).
32. Geng, G. et al. Synchrotron X-ray nanotomographic and spectromicroscopic study of the
tricalcium aluminate hydration in the presence of gypsum. Cem. Concr. Res. 111, 130–137
(2018).
33. Moradian, M. et al. Multi-scale observations of structure and chemical composition changes
of portland cement systems during hydration. Constr. Build. Mater. 212, 486–499 (2019).
34. Vigor, J. E., Bernal, S. A., Xiao, X. & Provis, J. L. Time-resolved 3D characterisation of
early-age microstructural development of Portland cement. J. Mater. Sci. 57, 4952–4969
(2022).
35. Poirier, M. et al. Synchrotron X-ray micro-tomography investigation of the early hydration
of blended cements: A case study on CaCl2-accelerated slag-based blended cements. Constr.
Build. Mater. 321, 126412 (2022).
36. Dierolf, M. et al. Ptychographic X-ray computed tomography at the nanoscale. Nature 467,
436–439 (2010).
37. Pfeiffer, F. X-ray ptychography. Nat. Photonics 12, 9–17 (2018).
38. Miao, J., Ishikawa, T., Robinson, I. K. & Murnane, M. M. Beyond crystallography:
diffractive imaging using coherent x-ray light sources. Science 348, 530–5 (2015).
39. Guizar-Sicairos, M. & Thibault, P. Ptychography: A solution to the phase problem. Phys.
Today 74, 42 (2021).
40. Cuesta, A. et al. Chemistry and Mass Density of Aluminum Hydroxide Gel in Eco-Cements
by Ptychographic X-ray Computed Tomography. J. Phys. Chem. C 121, 3044–3054 (2017).
41. Cuesta, A. et al. Quantitative disentanglement of nanocrystalline phases in cement pastes by
synchrotron ptychographic X-ray tomography. IUCrJ 6, 473–491 (2019).

16
42. Stockmar, M. et al. Near-field ptychography: phase retrieval for inline holography using a
structured illumination. Sci. Rep. 3, 1927 (2013).
43. Linderoth, O., Wadsö, L. & Jansen, D. Long-term cement hydration studies with isothermal
calorimetry. Cem. Concr. Res. 141, 106344 (2021).
44. Taylor, H. F. W. Cement chemistry. (Thomas Telford Pub, 1997).
45. Scrivener, K. L., Juilland, P. & Monteiro, P. J. M. Advances in understanding hydration of
Portland cement. Cem. Concr. Res. 78, 38–56 (2015).
46. Paganin, D., Mayo, S. C., Gureyev, T. E., Miller, P. R. & Wilkins, S. W. Simultaneous phase
and amplitude extraction from a single defocused image of a homogeneous object. J.
Microsc. 206, 33–40 (2002).
47. van Heel, M. & Schatz, M. Fourier shell correlation threshold criteria. J. Struct. Biol. 151,
250–262 (2005).
48. Bellmann, F., Sowoidnich, T., Ludwig, H. & Damidot, D. Cement and Concrete Research
Dissolution rates during the early hydration of tricalcium silicate. Cem. Concr. Res. 72, 108–
116 (2015).
49. Scherer, G. W., Zhang, J. & Thomas, J. J. Nucleation and growth models for hydration of
cement. Cem. Concr. Res. 42, 982–993 (2012).
50. Diamond, S. The microstructure of cement paste and concrete––a visual primer. Cem. Concr.
Compos. 26, 919–933 (2004).
51. Kjellsen, K. O. & Justnes, H. Revisiting the microstructure of hydrated tricalcium silicate - A
comparison to Portland cement. Cem. Concr. Compos. 26, 947–956 (2004).
52. Li, X. & Scrivener, K. L. Cement and Concrete Research Impact of ZnO on C 3 S hydration
and C-S-H morphology at early ages. Cem. Concr. Res. 154, 106734 (2022).
53. Xu, Z. et al. Research on cement hydration and hardening with different alkanolamines.
Constr. Build. Mater. 141, 296–306 (2017).
54. Morales-Cantero, A. et al. C-S-H seeding activation of Portland and Belite Cements: an
enlightening in situ synchrotron powder diffraction study. Cem. Concr. Res. 161, 106946
(2022).
55. Holler, M. et al. An instrument for 3D x-ray nano-imaging. Rev. Sci. Instrum. 83, 073703
(2012).
56. Holler, M. et al. X-ray ptychographic computed tomography at 16 nm isotropic 3D
resolution. Sci. Rep. 4, 3857 (2014).
Acknowledgements
Financial support from PID2019-104378RJ-I00 research grant, which is co-funded by FEDER, is gratefully
acknowledged. SLS is thanked for granting beamtime at cSAXS. ESRF is thanked for beamtime at ID19.
ToScA (United Kingdom) is gratefully acknowledged for awarding Jim Elliott Award to Shiva Shirani, which
supported her stay at ESRF. Dr. Manuel Guizar-Sicairos is thanked for his assistance with the ptychography
data processing. IRS is thankful for funding from PTA2019-017513–I.
Author contributions
M.A.G.A. conceived, designed and supervised this study. S.S. and I.S. did initial rheological and laboratory-
tomographic studies to fill the 200-micron capillaries. S.S. and A.M.-C. carried out the laboratory
characterization. S.S., A.D., P.T. and M.H. carried out the synchrotron ptychographic experiment. S.S., B.L.
and A.R. conducted the synchrotron microtomographic experiment. I.R.S. did the laboratory diffraction and
microtomographic experiments. S.S. did all the X-ray imaging data analysis with assistance of M.A.G.A., A.D.
and A.C. The machine learning segmentation was carried out by S.S. under E.B.'s supervision. M.A.G.A. wrote
the first draft. S.S. prepared all the figures, with help of A.C. for bivariate and Rietveld plots. All authors
discussed the results and commented on the manuscript.

17
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/xxx
Correspondence and requests for materials should be addressed to M.A.G.A.
Peer review information Nature Materials XXXXXXXXXX.
Reprints and permissions information is available at www.nature.com/reprints.

1
Supplementary information
4D nanoimaging of early age cement hydration
In the format provided by the authors and unedited

2
Supplementary Information
4D nanoimaging of early age cement hydration
Shiva Shirani1, Ana Cuesta1, Alejandro Morales-Cantero1, Isabel Santacruz1, Ana Diaz2, Pavel Trtik3, Mirko
Holler2, Alexander Rack4, Bratislav Lukic4, Emmanuel Brun5, Inés R. Salcedo6 and Miguel A. G. Aranda1*
1Departamento de Química Inorgánica, Cristalografía y Mineralogía, Universidad de Málaga, 29071-Málaga,
Spain.
2Laboratory for Macromolecules and Bioimaging, Paul Scherrer Institut, 5232 Villigen PSI, Switzerland.
3Laboratory for Neutron Scattering and Imaging, Paul Scherrer Institut, 5232 Villigen PSI, Switzerland.
4ESRF The European Synchrotron, 71 Rue des Martyrs, 38000 Grenoble, France.
5Université Grenoble Alpes, Inserm UA7 STROBE, 38000 Grenoble, France.
6Servicios Centrales de Apoyo a la Investigación, Universidad de Málaga, 29071-Málaga, Spain.
*Corresponding author: g_aranda@uma.es

3
• Supplementary Methods
Synchrotron X-ray computed microtomography experiment (Syn-µCT).
Microtomographic scans were acquired at the 150 m-long beamline ID19 of the European Synchrotron ESRF
in Grenoble, France. A so-called single-harmonic undulator (type: u17.6, gap 16.5 mm) was chosen as a
source due to its excellent photon flux density at a narrow bandwidth around approximately 19 keV photon
energy. The u17.6 allows beamline ID19 to be operated only with the two mandatory windows (0.8 mm
diamond in the front-end and 0.5 mm Beryllium in the experimental hutch) and an 0.7 mm-thick Aluminium
attenuator and hence, guaranties a homogeneous wave front: which is suited for high-sensitivity
measurements by means of inline propagation-based phase contrast. The propagation distance between
sample and detector was set to 15 mm. The indirect high-resolution detector consisted of a so-called
revolver-microscope by the French company OptiquePeter (Lentilly, France)1, the system lens-couples an
8.7 µm-thin LSO:Tb (Tb-doped Lu2SiO5) single-crystal scintillator with a 10× Olympus microscope (0.3NA) to
a sCMOS-based camera (type: pco.edge, PCO AG, Germany)2. The effective pixel size of the detector
assembly is approximately 0.6 µm. 6000 projection angles were acquired over a 360 degree tomographic
scan with an exposure time of 0.05 s, i.e. 5 minutes scan, at ~21.5 °C. During this experiment, the ESRF
operated in so-called 4bunch timing mode with a reduced ring current of maximum 20 mA. The estimated
flux density at the sample position was 4.2×1011 photons.s-1.mm-2. Phase retrieval of the projections was
performed using the Paganin algorithm3, considering the ratio of the refractive and absorption index /
equal to 70. In order to retrieve the microstructural content introduced by the inherent smoothing
characteristics of the phase retrieval method, a Gaussian unsharp mask was applied. The voxel size with the
employed configuration, to fully image a capillary of 0.7 mm of diameter, was 0.65 µm.
The tomographic reconstructions were performed using the open-source tomography software available
at the ESRF, relying on the sub-packages NXtomoMill and NABU4. Given its straightforward Graphics
Processing Unit (GPU)-based implementation, the full volume reconstructions were performed on the
Power9 cluster using the gold-standard filtered back projection (FBP) algorithm. The projections are first
corrected for beam profile illumination (flat field), dark current noise of the detector (dark field) and filtered
for any potential pixel outliers arising from stray photons. The reconstructed volume, consisting of
2490×2490×1950 pixels, is cast to 16 bit format considering the 10-90% of the volume histogram and
cropped to the region of interest. In the case of the reconstructions used for the Fourier Shell Correlation
(FSC) analysis,5 the original reconstruction is split into two sub-sampled reconstructions considering the
number of the reconstructed either even or odd projections.
Near-field ptychographic X-ray computed tomography (PXCT).
The measurements were carried out with a high-stability instrument designed for high-resolution PXCT
working in air and at room temperature6,7, using a photon energy of 8.94 keV. The coherent illumination
was defined with a Fresnel zone plate (FZP) of 120 µm diameter and 60 nm outer-most-zone width, which
at this energy had a focal distance of 51.9 mm. The FZP had locally displaced zones, specifically designed to
produce am optimal illumination for ptychograph8. The flux of the X-ray beam was 1.7×108 photons.s-1 at
the sample position. The sample was placed at 13 mm downstream the focus, where the illumination had
a size of about 30 µm. Ptychographic scans were recorded following the positions of a Fermat spiral9 with
an average step size of 6 or 7 µm and a field of view of 186 µm × 30 µm (horizontal × vertical). The field of
view must be larger than the size of the capillary to include an air region at both sides of the sample, which
is needed for successful tomographic reconstructions and for quantitative contrast. At each scanning
position, magnified images of the sample were recorded with an in-vacuum Eiger 1.5M detector10 with a
pixel size of 75 µm placed at 5.237 m downstream the sample, with an acquisition time of 0.1 s. A scan
speed of ∼5 Hz was achieved thanks to a combined motion of the FZP and the sample, while achieving an
effective static illumination on the sample during acquisition11. Near-field ptychographic scans were
repeated at 420 rotation angles of the sample in equal intervals from 0 to 180 deg. We recorded a total of
3 tomograms at different times from the start of the cement hydration at ∼25 °C, the temperature of the
experimental hutch. The first tomogram was recorded with an average scanning step size of 6 µm, it started

4
at 17 h and finished at 20h and 55 minutes, after water mixing, i.e. 3h 55 min of total acquisition time. This
scan is hereafter labelled 19 h dataset. The other two tomograms were recorded with a step size of 7 µm
lasting 3h 6 min. The scans labelled 47 and 93 h started at 46 and 92 h (after water mixing), respectively.
The scan times include the dead time during motion of stages in between acquisitions. The dose absorbed
by the specimen during data acquisition was estimated to be 0.7 and 0.5 MGy for the tomograms with 6
and 7 µm of step size, respectively.
Near-field ptychographic reconstructions were performed for each projection using the Ptycho Shelves
package12 developed by the Coherent X-ray Scattering group at PSI, using 5000 iterations of a difference
map algorithm13 adapted for near-field geometry. The pixel size of the images, determined by geometric
magnification, is 186.64 nm and we estimate by Fourier ring correlation5 that the 2D resolution of each
reconstructed image is about 200 nm. For each tomographic dataset, projections were aligned with sub-
pixel accuracy and processed for phase tomographic reconstruction from phase projections as previously
reported14,15. The 3D spatial resolution was estimated by FSC5. The resolution obtained, see subsection
dedicated to the spatial resolution, was limited by the number of projections, which was chosen to have
reasonable scan times.
Tomographic data analysis.
The segmentation was done on a Volume of Interest (VOI) corresponding to the inner part of the capillaries
for each imaged sample. The total volume of these VOIs varies depending on the sample sizes, amounting
to ∼1×105 μm3 for each PXCT dataset and ∼1×108 μm3 for Syn-µCT and Lab-µCT samples. A supervised
Machine-Learning (ML) image analysis approach was used to segment the different components of the
scanned samples, using the IPSDK Explorer software (version 3.2.0.0 for Windows™, Reactiv’IP, Grenoble,
France). This software allows us to manually label voxels on a selected training dataset (approximately 31
voxels for each component on average for PXCT, 20 voxels on average for Syn-µCT and Lab-µCT) and to
rapidly obtain test results to determine if the labelling is sufficient or if it requires more information/re-
training. These test results are obtained after a first learning step using a random forest method. It is also
possible for the user to keep or remove features used in the random forest decision trees based on their
relevance. This method permitted to segment the components with comparable grey values and/or
electron densities, overlaid ML models on raw datasets are shown in Fig. 7.
On the one hand, the good contrast and the high spatial resolution in PXCT allowed to classify the
components into seven categories. They are given next from higher to lower electron densities: i) C4AF
(yellow) with highest values; ii) C3S/C2S/C3A (dark brown) which are the clinker particles; iii) calcite (pink);
iv) portlandite (green); v) the rest of the hydrated phases with lower electron densities were labelled as
‘Low-Density Hydrates’ (light brown), i.e. C-S-H gel, iron-silicon-hydrogarnet, hemicarbonate and ettringite;
vi) water porosity (blue); and vii) air porosity (black). On the other hand, due to the contrast and spatial
resolution limitations in the two other modalities, Syn-µCT and Lab-µCT, the components were classified
into four categories. The classification from higher to lower grey-values was: i) clinker particles (dark
brown), i.e. C4AF/C3S/C2S/C3A; ii) a component labelled ‘High-Density Hydrates’ (green), being mainly
portlandite and calcite; iii) another component labelled ‘Low-Density Hydrates’ (light brown), being mainly
C-S-H gel, iron-silicon-hydrogarnet, hemicarbonate and ettringite; and iv) porosity (black) which contain
both water and air. It is noted that Syn-µCT and Lab-µCT microtomographies do not allow to distinguish
water from air porosities due to the similarities in their X-ray attenuation values.
This ML approach also permits to mitigate the influence of partial volume effects in-between labelled
components for accurate quantitative analysis of PXCT, i.e. mean electron density. Selected results after
the PXCT segmentation procedure are summarised S.I. video.1. In addition, after grains were segmented
using the ML approach described above, the C-S-H gel shell thickness was computed on PXCT imaged
sample at 19 h, see Fig. 6 and S.I. video.2. The wall thickness script computes the object thickness. For a
given pixel, the thickness is the radius of the largest circle centred on this pixel entirely included in the
object. The steps of the data analysis process are shown in flowcharts, see Fig. S20 and S21. A further post-
segmentation data analysis calculation was carried out in order to show the particle size distribution

5
evolution with hydration time. The anhydrous cement particles, at the three hydration times, were
classified by computing their mean Feret diameters. Fig. 8b displays the volume percentage of the
segmented grains (and their cumulative volumes) as function of the particle sizes that can be compared
with the initial characterization by laser diffraction, see Fig. 1a. 3D rendering visualization was done using
Dragonfly software (version 2022.1 for Windows™, Object Research Systems (ORS) Inc., Montreal, Canada).
Spatial resolution analysis.
The spatial resolutions were characterised by two approaches. On the one hand, Figures S2-S4 display line
profiles of sharp interfaces between high-density and low-density components. The steeper the slope, the
higher the spatial resolution. From this approach, the following estimations were obtained ∼2.1 µm, ∼1.1
µm and ∼240 nm for Lab-µCT, Syn-µCT and PXCT, respectively. On the other hand, FSC plots5 are displayed
in Figures S5-S7 giving spatial resolution values of ∼1.9 µm, ∼0.65 µm and ∼450 nm for the three type of
tomographies employed in this study. Moreover, the FSC trace for PXCT data at 19 h shows a smooth
decrease in the 0.0-0.2 spatial frequency range which is likely due to the hydration of cement during the 4-
hour measurement. As expected, this behaviour is not shown at later ages. Taken both results together,
the 3D spatial resolution in the tomograms is close to two voxels or ∼2.0 µm, ∼1.2 µm and ∼370 nm for Lab-
µCT, Syn-µCT and PXCT, respectively.
Water/cement ratio estimation of the scanned sample by PXCT
The w/c ratios of the scanned capillaries in a selected region can be calculated at the different ages
according to the procedure previously reported16. The final β-mean values obtained by PXCT were used
after converting to µ values, see Table S6. Then, using the mineralogical compositions of the anhydrous
cement (given in the Supporting information, Table S2) the µ value is estimated, taking into account the µ
value of free water, 22.2 cm-1. For instance, for the 19 h sample, it can be estimated that the paste was
composed of 69.9 wt% PC and 30.1 wt% water to account for the overall µ of the paste. This calculation
yielded a w/c ratio of 0.39, see Table S6.
Chemical reactions
I. The chemical reactions used for the FW calculation, see Table S5, are: (1) the consumption of water by
the hydration of C4AF, with C3S which is the source of silicates, to give amorphous iron siliceous hydrogarnet
(Fe-Si-Hg) and crystalline portlandite; (2) the hydration of C3S to yield amorphous C-S-H gel and crystalline
portlandite; (3) the hydration of C3A, consuming a calcium sulfate source, to yield ettringite if there are
enough sulfates available (which is the case here); and (4) the possible carbonation of portlandite gives
crystalline (and amorphous) calcium carbonate(s) and it releases free (capillary) water.
II. In the absence of belite hydration, the chemical reaction contributing to C-S-H gel formation, see Table
S5, is just (2). It is underlined that a small fraction of the consumed alite did not result in C-S-H gel but in
the formation of iron siliceous hydrogarnet from the ferrite hydration (reaction 1).
It is noted here that for the calculations presented in Table S5, the amount of C3S which is needed for the
silicate groups in iron-silicon-hydrogarnet, is calculated first from the degree of hydration of C4AF (applying
reaction #1). Then, the portlandite and C-S-H gel contents are determined from the reaction of C3S after
subtracting the number obtained in the process described just above.
Ca4Al2Fe2O10 + 1.68Ca3SiO5 + 11.68H2O → 2Ca3AlFe(SiO4)0.84(OH)8.64 (amorph.) + 3.04Ca(OH)2 (1)
Ca3SiO5 + 5.2H2O → 1.2Ca(OH)2 + (CaO)1.8SiO2(H2O)4.0 (2)
Ca3Al2O6 + 3CaSO4·2H2O + 26H2O → Ca6Al2(SO4)3(OH)12·26H2O (3)
Ca(OH)2 + CO2 → CaCO3 + H2O (4)

6
• Open access raw data availability and description
The following raw data has been openly deposited on Zenodo and can be acceded at:
https://doi.org/10.5281/zenodo.7030107
1. Tomographic reconstructed raw data of all the X-ray imaging modalities (twelve tomograms) in 16 bit
and .tif format. The size of the files is also given in the following table.
Folder Name
Sub-folders Label
File Size*
Additional Information
Phase-PXCT
1_PXCT-19h
736.9MB
4D Synchrotron Ptychographic X-
ray Computed Tomography
Cement Hydration (Delta Dataset)
2_PXCT-47h
3_PXCT-93h
Syn-microCT
4_Syn-microCT-19h
52.8GB
4D Synchrotron Phase-contrast
Microtomography Cement
Hydration
5_Syn-microCT-47h
6_Syn-microCT-93h
Lab-microCT
7_Lab-microCT-19h
7.5GB
4D Laboratory Attenuation-
contrast Microtomography
Cement Hydration
8_Lab-microCT-47h
9_Lab-microCT-93h
Absorption-PXCT
10_abs-PXCT-19h
722.9MB
Absorption Dataset for 4D
Synchrotron Ptychographic X-ray
Computed Tomography Cement
Hydration (Beta Dataset)
11_abs-PXCT-47h
12_abs-PXCT-93h
Net total size of all datasets: 82.8GB
*These are zipped files. The original file size for every single 16 bit Syn-microCT dataset is 22.8GB, for each
Lab-microCT dataset is 3.0GB and for each PXCT dataset is 500MB.
2. Laboratory raw data
2.1. Particle size distribution (PSD)
“PSD” labelled folder contains two files in .mmes format.
2.2. Isothermal calorimetry
“Calorimetry” labelled folder contains six files in .xlsx excel format.
2.3) Laboratory X-ray powder diffraction (LXRPD)
“LXRPD” labelled folder contains five files in .ASC text format.

7
• Supplementary Tables
Table S1. Chemical (elemental) analysis (by X-ray fluorescence) of the two employed Portland cements in
this investigation. All data expressed in weight percentages of the corresponding oxides.
CaO
SiO2
SO3
Al2O3
Fe2O3
MgO
K2O
Na2O
Others
LoI
PC-52.5
62.2
20.4
3.6
4.9
3.2
1.5
1.1
0.3
0.5
2.3
PC-42.5
62.9
19.7
3.4
5.0
3.4
1.5
1.1
0.3
0.3
2.7
Table S2. Rietveld quantitative phase analysis of the employed anhydrous Portland cements.
C3S
β-C2S
C4AF
o-C3A
C
H2
C
H0.5
Cc
Q
K

CaO
PC-52.5
61.0
11.8
11.5
8.2
1.9
2.0
2.8
0.8
-
-
PC-42.5
61.6
11.1
10.8
7.8
0.9
2.1
3.7
0.6
1.1
0.4
Table S3. Textural details of the two cements.
Density (gcm
-3
)
BET (m
2
g
-1
)
Blaine (m
2
Kg
-1
)
Dv,10 (μm)
Dv,50 (μm)
Dv,90 (μm)
PC-52.5
3.108(1)
2.27(1)
409(8)
1.8
11.5
32.7
PC-42.5
3.126(1)
1.25(1)
368(1)
2.1
18.0
50.0
Table S4. Selected cumulative heat release data (from the isothermal calorimetry study) for the two
employed cements. All values in J per gram of Portland cement.
19 h
47 h
93 h
7 d
Maximum peak
Heat/J
DoH
$
/%
Heat/J
DoH
$
/%
Heat/J
DoH
$
/%
Heat/J
DoH
$
/%
Time/h
Heat/J
DoH
$
/%
PC-52.5, w/c=0.40
180.8
33.7
279.8
52.4
300.7
56.3
313.5
58.7
14.2
102.9
19.3
PC-52.5, w/c=0.50
157.2
29.4
306.6
57.4
346.9
64.9
365.2
68.4
15.6
102.7
19.2
PC-42.5, w/c=0.40
116.5
22.1
220.5
41.9
270.8
51.5
291.5
55.4
10.4
50.4
9.6
PC-42.5, w/c=0.50
114.4
21.7
223.3
42.4
279.8
53.1
310.0
58.9
11.0
53.4
10.2
$Total heat of hydration of PC52.5 as calculated in ref.17 option-1: 534 J.
$Total heat of hydration of PC42.5 as calculated in ref.17 option-1: 526 J.

8
Table S5. Rietveld quantitative phase analysis (of MoKα1 radiation powder diffraction data) results (wt%)
for the studied PC-52.5 pastes, w/c=0.40. Laboratory X-ray diffraction data were taken on the same capillary
used for the laboratory, attenuation-contrast, microtomographic study. The data are referenced to 100
grams of paste.
phases
t0*
t=22h
t=50h
t=96h
C3S
43.57
18.9
17.0
16.3
β-C2S
8.43
8.3
8.7
9.1
C3A
5.86
2.9
1.9
1.5
C4AF
8.21
7.7
5.8
5.4
Cc
2.00
3.1
3.2
3.6
CH
-
10.6
13.5
13.8
AFt
-
12.2
12.4
12.4
AFm-Hc
-
0.4
0.4
C-S-H
&
- 20.3 20.6 20.9
Fe-Si-Hg&
-
0.5
2.8
3.1
FW&
28.57
14.4
12.9
12.7
DoH C3S (%)
- 56 61 63
DoH C2S (%)
-
0
0
0
DoH C3A (%)
-
50
67
74
DoH C4AF (%)
-
5
29
34
* This cement also has at t0: 1.4 wt% of gypsum, 1.4 wt% of bassanite and 0.6 wt% of quartz.
& C-S-H, Fe-Si-Hg and FW (free water) contents calculated from the assumed chemical reactions as described
in supplementary methods.
Table S6. Experimental mean β values converted to µ values obtained from a VoI in the PXCT study. The
calculation of the w/c ratio of this region at the different hydration ages is also included. More details about
this calculation is given in the Supplementary Methods.
Scan Experimental
β
*
Experimental
µ/cm-1
Weight /g
Weight /wt%
w/c ratio
water
cement
water
cement
PXCT-19h
1.316×10
-7
119.1
55.0
139.4
28.3
71.7
0.39
PXCT-47h
1.265×10
-7
114.5
57.2
132.8
30.1
69.9
0.43
PXCT-93h
1.274×10
-7
115.3
56.8
133.9
29.8
70.2
0.42
*The β values have been calculated using the same volume used for the calculation of delta but excluding the regions
where the electron density is smaller than 0.24 e-Å3, as they have been considered air porosity. It is noted that air is
not included in this calculation.

9
Table S7. Expected (from the crystallographic data,18,19) mass and electron densities. The attenuation length values are calculated$ for the employed wavelength,
E=8.93 keV. The measured electron densities for selected components from the PXCT data, where relatively large volumes could be chosen, are also given.
Cement Phases Abbrev. Formula
Mass
density
/gcm
-3
Electron
density
/e
-
Å
-3
Measured electron density /e-Å-3
(within selected volumes)
19 h 47 h 93 h
Attenuation
length /µm µ /cm-1

#0, air
-
-
∼0.00
∼0.00
#1, water
-
H
2
O
1.00
0.33
1388.6
7.2
7.96 10-09
#2, ettringite
AFt
Ca
6
Al
2
(SO
4
)
3
(OH)
12
.26H
2
O
1.78
0.56
146.7
68.2
7.53 10-08
#3, calcium silicate
hydrate C-S-H (CaO)1.8(SiO2)(H2O)4 2.11% 0.64% 128.6 77.8 8.59 10-08
#4, portlandite
CH
Ca(OH)
2
2.23
0.69
0.62(2)
0.649(6)
0.651(5)
61.5
162.6
1.80 10-07
#5, calcium carbonate
Cc
CaCO
3
2.71
0.82
0.782(3)
0.776(3)
0.776(3)
66.3
150.8
1.67 10-07
#6, tricalcium aluminate
C
3
A
Ca
3
Al
2
O
6
3.05
0.91
49.0
204.0
2.25 10-07
#7, alite
C
3
S
Ca
3
SiO
5
3.15
0.95
0.936(2)
0.931(2)
0.932(1)
41.8
239.1
2.64 10-07
#8, belite
C
2
S
Ca
2
SiO
4
3.30
0.99
0.98(2)
0.98(1)
0.98(1)
43.3
231.1
2.55 10-07
#9, ferrite
C
4
AF
Ca
2
AlFeO
5
&
3.73
1.10
26.4
379.4
4.19 10-07
% There are not expected values for an amorphous material. The quoted values (italics) were determined for five months cured C-S-H gel by PXCT16.
& The reported values are for stoichiometric Ca2AlFeO5, i.e. an Al/Fe molar ratio of 1.0, which is an approximation as this ratio could be different from 1.0.
$ The attenuation length values have been calculated from20, https://henke.lbl.gov/optical_constants/atten2.html
The µ values were calculated from:  [] = 
  []×
Finally,
β
was calculated as =


10
Table S8. Component segmentation (vol%) and average electron densities obtained by PXCT at the different hydration ages; expected electron densities (from
crystallographic data when it is possible) are also given for reference.
Component
Expected
electron
density
/e
-
Å
-3
19 h
47 h
93 h
Vol /% Electron
density* /e-Å-3
Electron
density& /e-Å-3 Vol /% Electron
density* /e-Å-3
Electron density&
/e-Å-3 Vol /% Electron
density* /e-Å-3
Electron
density& /e-Å-3
Capillary
-
-
-
0.63(1)
-
-
0.63(1)
-
-
0.63(1)
Air porosity
0.00
0.2
0.01(5)
-
4.1
0.10(1)
-
6.7
0.10(1)
-
Water porosity
0.33
15.1
0.33(6)
-
2.2
0.33(5)
-
2.8
0.32(6)
-
LD-Hydrates
0.38-0.53
45.5
0.50(4)
-
56.5
0.52(4)
-
51.8
0.53(4)
-
Portlandite
0.69
4.6
0.63(1)
0.62(2)
13.1
0.62(4)
0.65(1)
15.7
0.62(4)
0.65(1)
Calcite
0.82
2.5
0.74(2)
0.78(1)
2.2
0.74(4)
0.78(1)
2.3
0.74(5)
0.78(1)
Belite/Alite/C3A
0.99/0.95/0.91
30.4
0.90(1)
0.94(1)/0.98(2)
20.3
0.90(4)
0.93(1)/0.98(1)
19.3
0.90(5)
0.93(1)/0.98(1)
Ferrite
1.10
2.1
1.02(2)
-
1.5
1.01(4)
-
1.4
1.02(4)
-
*Electron densities, from full volume, were obtained by segmentation excluding the external voxels to avoid partial volume effect
&Electron densities, from particle picking, were obtained by the average of 10 cubes for the capillary; 5, 4, 5 and 6 grains for portlandite, calcium carbonate, alite and belite,
respectively.

11
Table S9. Volume percentages for the cement pastes at the different ages of hydration determined by the techniques used in this paper. Degrees of hydration are
also included.
Technique Hydration
age /h
Porosity
/vol%:
air & water
LD-hydrates
/vol%
Portlandite &
calcite /vol %
Anhydrous
components /vol %:
C3S/C2S/C3A/C4AF
DoH /%
Calorimetry
PC-52.5, w/c=0.40
19
-
-
-
-
34
22
-
-
-
-
39
47-50
-
-
-
-
53
93-96
-
-
-
-
56
Calorimetry
PC-42.5, w/c=0.50
19
-
-
-
-
22
47
-
-
-
-
42
93
-
-
-
-
53
0
56.1
$
-
-
39.9
$
-
LXRPD
PC-52.5, w/c=0.40
22
26.4
39.6
11.6
22.0
45
50
22.3
44.8
13.7
20.6
48
96
21.9
45.5
14.2
19.9
50
0
56.1
$
39.9
$
-
Lab-µCT
PC-52.5, w/c=0.40
19
5.6
35.7
35.7
23.5
41
47
2.1
21.7
60.8
15.4
61
93
2.1
20.2
62.8
15.0
62
0
61.5
#
-
-
35.0
#
-
Syn-µCT
PC-42.5, w/c=0.50
19
1.9
43.8
27.3
27.0
23
47
14.5
32.9
31.7
20.8
41
93
13.7
34.3
35.2
16.9
52
0
56.1
$
39.9
$
-
PXCT
PC-52.5, w/c=0.40
19
15.3
45.5
7.1
32.1
20
47
6.3
56.5
15.3
21.8
45
93
9.5
51.8
18.0
20.7
48
$ The amount of water and clinker phases is 96.0 vol%. The remaining 4.0 vol% is due to the minor components: gypsum, bassanite, calcite and quartz
# The amount of water and clinker phases is 96.4 vol%. The remaining 3.6 vol% is due to the minor components: gypsum, bassanite, calcite and quartz

12
• Supplementary Figures
Figure S1. Laboratory Rietveld plots for the anhydrous cements. (a) PC-52.5 (CuKα1 radiation, λ=1.5416
Å). (b) PC-42.5 (MoKα1 radiation, λ=0.7093 Å).

13
Figure S2. (Left) Selected orthoslice and (Right) grey-value profile of the yellow line (shown in the right
panel) including a sharp interface for the laboratory, attenuation-contrast, microtomographic study (PC-52.5
paste with w/c=0.40, dataset at 4 days of hydration). The estimated spatial resolution, from this approach, is
close to 2.1 µm.
Figure S3. (Left) Selected orthoslice and (Right) grey-value line profile of a sharp interface for the
synchrotron, phase propagation based-contrast, microtomographic study (PC-42.5 paste with w/c=0.50,
dataset at 4 days of hydration). The estimated spatial resolution, from this approach, is close to 1.1 µm. The
star symbol highlights the small artefact (edge enhancement not fully corrected by the Paganin algorithm)
which is commonly observed in in-line propagation-based phase-contrast synchrotron tomography.

14
Figure S4. (Left) Selected orthoslice and (Right) line profile of a sharp interface for the near-field
ptychographic X-ray computed tomographic study (PC-52.5 paste with w/c=0.40, dataset at 4 days of
hydration). The estimated spatial resolution, from this approach, is close to 240 nm.

15
Figure S5. Fourier Shell Correlation plots for the laboratory, attenuation-contrast, microtomographic study,
PC-52.5 paste (w/c=0.40) at (a) 19 h, (b) 47 h and (c) 93 h of hydration. The cuts between the FSC traces
and the threshold lines give an indication of the spatial resolution of each tomogram.

16
Figure S6. Fourier Shell Correlation plots for the propagation-based synchrotron phase-contrast X-ray
computed microtomographic study, PC-42.5 paste (w/c=0.50) at (a) 19 h, (b) 47 h and (c) 93 h of hydration.
The FSC traces do not cut the threshold indicating that, from this approach, the overall spatial resolution is
limited by the sampling (pixel size).

17
Figure S7. Fourier Shell Correlation plots from the near-field ptychographic X-ray computed tomographic
study for the PC-52.5 paste (w/c=0.40) at (a) 19 h, (b) 47 h and (c) 93 h of hydration. The cuts between the
FSC traces and the threshold lines give an indication of the spatial resolution of each tomogram.

18
Figure S8. Selected PXCT orthoslices at 19 and 93 h of hydration. (Top) Electron density datasets. (Bottom)
Absorption datasets.

19
Figure S9. Bivariate histograms of electron densities and absorption indexes (β) for the PXCT study of the
PC-52.5 paste (w/c=0.40) at the three studies hydration ages. The positions of the different components, as
expected from the crystallographic data, are given. C-S-H gel position (component #3) is not given, as it is a
non-stoichiometric solid, but it should be close to ettringite (#2). The electron density and absorption of air
is zero but the partial volume effect slightly displaces the values, see pink arrows. For the
numbers/components, the reader is referred to Table S7.
Figure S10. Volume-of-interest histogram of the electron densities for the PXCT study of the PC-52.5 paste
(w/c=0.40) at (blue) 19 h, (green) 47 h and (red) 93 h of hydration. (a) Logarithmic scale, (b) Linear scale.
The expected (from the crystal structures) electron densities for the different components are labelled in (a).
C-S-H gel is an amorphous solid with variable water content and Ca/Si molar ratio and therefore, the expected
electron densities are a range.

20
Figure S11. Second etch-pit evolution picture. (Top row) PXCT orthoslices at the three studied ages. (Intermediate) 3D rendering of a volume including a fraction of
the alite particle highlighted in the top panels. (Bottom) 3D representation of the segmented particle to highlight the evolution of the etch pits.

21
Figure S12. Selected PXCT vertical views at the studied ages showing the evolution of the PC-52.5 paste.
This series is intended to show the evolution of water porosity (dark-grey) towards air porosity (black) with
time. Moreover, the enlarged views (right images) show the evolution of a small alite particle, initial size
about 3 µm, which develops etch-pits of sizes of ∼700 nm, highlighted with red arrows.
Figure S13. (Left panel) Selected 2D view of the PXCT data at 19 h. (Right) Electron density profile
corresponding to the yellow straight line. The line profile signals the water porosity region (blue) surrounding
the two alite particles, with the sizes of the C-S-H gel shells and their electron densities given in green.

22
Figure S14. Study of the alite dissolution and C-S-H gel (shell) densification with hydration time. Same
region than that shown in Fig. S13.
Figure S15. (Left panel) Selected 2D view of the PXCT data at 19 h. (Right) Electron density profile
corresponding to the yellow straight line. The line profile signals the water porosity region (blue) surrounding
the two alite particles, with the sizes of the C-S-H gel shells and their electron densities given in green.

23
Figure S16. Study of the alite dissolution and C-S-H gel (shell) densification with hydration time. Same
region than that shown in Fig. S15. Capillary pore water is still visible at 93 h in the top left region of the
image.
Figure S17. (Left panel) Selected 2D view of the PXCT data at 19 h. (Right) Electron density profile
corresponding to the yellow straight line in the right image. The C-S-H porous shell covers every alite particle
but it does not surround belite neither calcite. The line profile signals the water porosity region (blue)
surrounding alite, with the size of the C-S-H gel shell and its electron density given in green.

24
Figure S18. (Left panel) Selected 2D view of the PXCT data at 19 h. (Right) Electron density profile
corresponding to the yellow straight line. The C-S-H porous shell covers every alite particle but it does not
surround belite grains. The line profile signals the water porosity region (blue) surrounding alite, with the
size of the C-S-H gel shell and its electron density given in green.
Figure S19. Selected PXCT vertical views at the studied ages showing the evolution of the PC-52.5 paste.
This series is intended to show the evolution of water porosity (dark grey) towards air porosity (black) with
time. The enlarged views (right images) show the change of a hollow-shell volume, also known as Hadley
grain, with hydration time. The hollow-shells, Hadley grains, are fully hydrated small alite particles that
contain a void within the original boundary of the anhydrous grain. The hollow regions of the Hadley grains
are filled with water at 19 and 47 h but dried at 93 h, see enlarged pictures to the right. This illustrates that
most of the capillary pores with sizes larger than ∼1 µm are already water emptied at 93 h of hydration, see
bottom right. Moreover, it also illustrated that the C-S-H shells are porous as they allow the water diffusion
from the inner regions towards the exterior.

25
Fig S20. Flow chart describing the data treatment in this work and detailing the machine learning training
steps (boxes in grey).

26
Figure S21. Flow chart detailing the steps for the C-S-H shell segmentation in the 19 h PXCT dataset.
Figure S22. Views, at the same scale, comparing the C-S-H shell as observed in the PXCT raw dataset at 19
h of hydration (left) and the C-S-H shell segmentation output applying the procedure detailed in Fig. S21.

27
Figure S23. Selected PXCT orthoslices at the studied ages showing the evolution of the PC-52.5 paste. The
enlarged views (bottom) show the evolution of porosity within the paste, where several pores of sizes smaller
than ∼2 µm are dried (red arrows) at 93 h but other larger, pores keep filled with water (blue arrows).
Figure S24. Selected PXCT orthoslices at the three studied ages showing the evolution of the PC-52.5 paste.
The enlarged views (bottom) show the evolution of a large calcite particle with internal pores. At 19 and 47
h of hydration some pores are filled with water (highlighted with blue arrows) which are connected to the
surface. At 93 h, these pores are empty (red arrows) releasing water for further hydration.

28
Figure S25. Selected PXCT orthoslices at the studied ages. The chemical shrinkage is evident at 93 h because
the appearance of empty (black) regions. The enlarged views (bottom) show: (1) the hydration of a large alite
particle with aluminoferrite, C4AF, intergrown, i.e. the whitest regions, see brown arrows. In addition to the
etch-pit evolution, it can be seen that hydration stops at the regions where C4AF is exposed to the hydration
medium; (2) the blue rectangles highlight the dissolution of C-S-H gel particles to give dry (air-filled) pores.
Figure S26. Selected PXCT vertical pictures with enlarged views (bottom) displaying the hydration pathway
of a very fast dissolving particle, i.e. a 4 µm particle fully dissolved between 19 and 47 h. The electron density
of this small volume, 0.91 e-Å-3, is compatible with C3S or C3A. It already shows a gap at 19 h indicating a
highly soluble component. At 47 h, two hydrate rods of diameter smaller than 1 µm, morphologically
suggesting ettringite, grow in the pristine region which is filled with capillary water. At 93 h, the volume is
fully occupied by hydrate(s). The dissolution rate between 19 and 43 h is faster than 75 nm/h suggesting C3A.
However, the chemical nature of this tine particle could not be firmly established.

29
Figure S27. Capillary water porosity evolution, with the obtained spatial resolution (i.e. approximately two
voxels). Field of view ∼160 µm. (Left) ML segmentation output. (Right) 2D orthoslices of the raw PXCT
datasets.

30
• Supplementary Movies
• Video-1: "Summary of 4D nanoimaging of cement hydration" 43 seconds.
A summarized display of the cement paste hydration evolution as seen by this nanoimaging study. The
progress of the different components is displayed after segmentation by Machine-Learning. Moreover,
key changes like water porosity evolution or shrinkage development are highlighted on the video by
embedded written text.
• Video-2: " C-S-H shell characterization at 19 hours " 17 seconds.
A movie revealing the arrangement of the 3D segmented C-S-H shells through the 19 h nanoimaging
dataset.
The size of each short video is standard 640×480 pixels in mp4 format as suggested by Nature journal.
Therefore, users can download them quickly.
Article cover image:
Title: "X-ray nanoimaging of a hydrating cement paste at early ages"
Description: The precipitating calcium silicate hydrate shells (blue) surround the dissolving alite particles
(yellow) with regions of calcium aluminoferrite highlighted in orange. For better visualisation: only the C-
S-H shells in the left part, the three components in the middle region, just the anhydrous cement particles in
the right part. The gaps, approximately 500 nm, between the C-S-H shells and the dissolving alite particles
are readily visible in the central part of the image.
The cover image is a high resolution 5000×4005 pixels size in .tif format.
Cover image credit: Shiva Shirani, Cover design: Maziar Moussavi

31
• Supplementary References
1. Douissard, P. A. et al. A versatile indirect detector design for hard X-ray microimaging. J. Instrum. 7,
P09016 (2012).
2. Martin, T. et al. LSO-based single crystal film scintillator for synchrotron-based hard X-ray micro-
imaging. IEEE Trans. Nucl. Sci. 56, 1412–1418 (2009).
3. Paganin, D., Mayo, S. C., Gureyev, T. E., Miller, P. R. & Wilkins, S. W. Simultaneous phase and
amplitude extraction from a single defocused image of a homogeneous object. J. Microsc. 206, 33–
40 (2002).
4. Vigano, N., Cloetens, P., di Michiel, M., Rack, A. & Tafforeau, P. Redefining the ESRF tomography
software. in Digital Holography and Three-Dimensional Imaging 2021 paper DF2G.4 (OSA Technical
Digest, Optica Publishing Group, 2021). doi:10.1364/DH.2021.DF2G.4.
5. van Heel, M. & Schatz, M. Fourier shell correlation threshold criteria. J. Struct. Biol. 151, 250–262
(2005).
6. Holler, M. et al. An instrument for 3D x-ray nano-imaging. 83, 073703 (2012).
7. Holler, M. et al. X-ray ptychographic computed tomography at 16 nm isotropic 3D resolution. Sci.
Rep. 4, 3857 (2014).
8. Odstrčil, M., Lebugle, M., Guizar-Sicairos, M., David, C. & Holler, M. Towards optimized
illumination for high-resolution ptychography. Opt. Express 27, 14981 (2019).
9. Huang, X. et al. Optimization of overlap uniformness for ptychography. Opt. Express 22, 12634
(2014).
10. Dinapoli, R. et al. EIGER: Next generation single photon counting detector for X-ray applications.
Nucl. Instruments Methods Phys. Res. Sect. A Accel. Spectrometers, Detect. Assoc. Equip. 650, 79–
83 (2011).
11. Odstrcil, M., Lebugle, M., Lachat, T., Raabe, J. & Holler, M. Fast positioning for X-ray scanning
microscopy by a combined motion of sample and beam-defining optics. J. Synchrotron Radiat. 26,
504–509 (2019).
12. Wakonig, K. et al. PtychoShelves, a versatile high-level framework for high-performance analysis of
ptychographic data. J. Appl. Cryst 53, 574–586 (2020).
13. Thibault, P., Dierolf, M., Bunk, O., Menzel, A. & Pfeiffer, F. Probe retrieval in ptychographic
coherent diffractive imaging. Ultramicroscopy 109, 338–343 (2009).
14. Odstrčil, M., Holler, M., Raabe, J. & Guizar-Sicairos, M. Alignment methods for nanotomography
with deep subpixel accuracy. Opt. Express 27, 36637–36652 (2019).
15. Guizar-Sicairos, M. et al. Phase tomography from x-ray coherent diffractive imaging projections.
Opt. Express 19, 21345 (2011).
16. Cuesta, A. et al. Quantitative disentanglement of nanocrystalline phases in cement pastes by
synchrotron ptychographic X-ray tomography. IUCrJ 6, 473–491 (2019).
17. Linderoth, O., Wadsö, L. & Jansen, D. Long-term cement hydration studies with isothermal
calorimetry. Cem. Concr. Res. 141, 106344 (2021).
18. Aranda, M. A. G., Cuesta, A., De la Torre, A. G., Santacruz, I. & León-Reina, L. Diffraction and
crystallography applied to hydrating cements. in Cementitious Materials: Composition, Properties,
Application (ed. Pöllmann, H.) 31–60 (De Gruyter, 2017). doi:10.1515/9783110473728-003.
19. De la Torre, A. G., Santacruz, I., Cuesta, A., León-Reina, L. & Aranda, M. A. G. Diffraction and
crystallography applied to anhydrous cements. in Cementitious Materials (ed. Pöllmann, H.) 3–29
(De Gruyter, 2017).
20. Henke, B. L., Gullikson, E. M. & Davis, J. C. X-Ray Interactions: Photoabsorption, Scattering,
Transmission, and Reflection at E = 50− 30,000 eV, Z = 1−92. At. Data Nucl. Data Tables 54, 181–
342 (1993).