Cement and Concrete Research 65 (2014) 96–104

Contents lists available at ScienceDirect

Cement and Concrete Research

journal homepage: http://ees.elsevier.com/CEMCON/default.asp

Magnesium potassium phosphate cement paste: Degree of reaction,

porosity and pore structure
Hongyan Ma, Biwan Xu, Zongjin Li ⁎
Department of Civil and Environmental Engineering, The Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong

a r t i c l e i n f o a b s t r a c t

Article history: Although magnesia–phosphate cements have been studied and applied in several fields for many years, the
Received 13 January 2014 theoretical background on this kind of chemically bonded ceramics has not been sufficiently well established
Accepted 18 July 2014 for the quantitative prediction of material properties. In this study, the stoichiometric factors of the chemical re-
Available online 12 August 2014
action in magnesium potassium phosphate cement (MKPC) paste are analyzed, and the degree of reaction of this
cement is defined. Based on the stoichiometric factors and the degree of reaction, the porosity of MKPC paste,
Keywords:
Chemically bonded ceramics (D)
which is essential for predictions of both the mechanical and transport properties, is calculated. In addition,
Reaction (A) the pore structure is simulated by a newly developed computer model. The calculated porosities and the simu-
Thermal analysis (B) lated pore structures are both found to be consistent with the results measured by mercury intrusion
Microstructure (B) porosimetry (MIP).
Pore size distribution (B) © 2014 Elsevier Ltd. All rights reserved.

1. Introduction
Magnesia–phosphate cements (MPCs), or magnesium phosphate
cements as sometimes referred to, are termed chemically bonded ce-
ramics [1], as they are low-temperature high-strength materials formed
by through-solution acid-based reaction between dead burnt magnesia
and phosphate [2]. During decades of development, MPCs have been
widely applied in many fields, such as the repair and rehabilitation of
Abbreviations: ADP, ammonium dihydrogen phosphate; COV, coefficient of variation;
DTG, differential thermogravimetric analysis; H, water; KDP, potassium dihydrogen phos-
phate; KMP, potassium metaphosphate; M, magnesia; MAP, magnesium ammonium
phosphate hexahydrate, MgNH4PO4·6H2O; MAPC, magnesium ammonium phosphate ce-
ment; MIP, mercury intrusion porosimetry; MKP, magnesium potassium phosphate hexa-
hydrate, MgKPO4·6H2O; MKPC, magnesium potassium phosphate cement; MPC,
magnesia–phosphate cement; REV, representative elementary volume; TGA, thermogra-
vimetric analysis; XRD, X-ray diffraction; Lb 200, weight loss below 200 °C; LN 200, weight
loss above 200 °C; M/P, magnesia-to-phosphate molar ratio; Mi, molar mass of matter i,
where i can be H, KDP, KMP, M, or MKP; mKDP, initial weight of KDP before reaction;
mun
KDP, weight of unreacted KDP; m˜M , a mass parameter which is defined as
m˜M ¼ MM⋅M=PηM; R600, remaining weight of MKPC paste when heated to 600 °C;
RKMP, weight component of R600 due to KMP; RM, weight component of R600 due to magne-
sia; vH, volume of remaining bulk water in the REV; Vi, molar volume of matter i, where i
can be H, KDP, M, or MKP; vIP, volume of inner reaction products; vij, initial volume of mat-
ter j in the REV, where j can be H, KDP, or M; vjr, reacted volume of matter j in the REV,
where j can be KDP or M; vOP, volume of outer reaction products; W/C, water-to-cement
mass ratio; αKDP, degree of reaction of KDP; αM, degree of reaction of magnesia; γ, volume
of reaction products formed when 1 unit volume of magnesia reacted; γH, volume of water
consumed when 1 unit volume of magnesia reacted; ηM, purity of the dead burnt magne-
sia; ρi, density of matter i, where i can be H, KDP, M, or MKP; ϕ, porosity.
⁎ Corresponding author. Tel.: +852 23588751.
E-mail addresses: mhy1103@gmail.com (H. Ma), xubiwan@ust.hk (B. Xu),
zongjin@ust.hk (Z. Li).
civil structures [3], stabilization of toxic matter and nuclear waste
[4–7], treatment of waste water [8,9], and dental and bone restorations
[10,11]. In civil engineering, MPCs are frequently used as repair mate-
rials for pavements and structures, as they set rapidly and produce
high strength patches with low permeability, little drying shrinkage
and good durability [3,12–16]. In early applications of MPCs in civil re-
pair engineering, ammonium dihydrogen phosphate (ADP) was usually
used as the phosphate component, with boric acid or borax compounds
as the reaction retarder [3,13]. More strictly, the ADP-based MPC can be
termed as magnesium ammonium phosphate cement (MAPC), as the
main reaction product in the cement paste is magnesium ammonium
phosphate (MAP) hexahydrate, with the formula MgNH4PO4·6H2O,
which is a naturally existing crystal more commonly known as struvite
[17]. However, as the byproduct of the reaction forming struvite, ammo-
nium gas would generate an unpleasant odor, leading to attempts to re-
place ADP by potassium dihydrogen phosphate (KDP) [12,18–21]. As
compared with ADP, KDP has smaller dissociation constant and lower
solubility, which would lower the reaction rate to some extent, so that
a retarder may not even be necessary, in some applications. The KDP-
based MPC can be termed as magnesium potassium phosphate cement
(MKPC), as the principal reaction product in the magnesia–KDP–water
ternary system has been identified as magnesium potassium phosphate
(MKP) hexahydrate, with the formula MgKPO4·6H2O, which is more
popularly known as struvite-(K). Struvite-(K) is isostructural with
struvite [22], and has been found recently to be also a naturally existing
mineral [23], which reflects its stability somewhat.
Most of the studies on MPCs focused directly on the engineering
properties of the resulting paste or mortar, such as setting time, strength
and durability, and attempted to obtain an optimized formula based on
the raw materials available [3,4,12,14,15,18]. In recent years, some

http://dx.doi.org/10.1016/j.cemconres.2014.07.012
0008-8846/© 2014 Elsevier Ltd. All rights reserved.
 H. Ma et al. / Cement and Concrete Research 65 (2014) 96–104
researchers have begun to investigate the reaction mechanisms of MPCs
[24–28]. However, theories that can be used to predict properties of
MPC-based materials have not been established. Since MPCs have
been considered as promising candidates for the partial replacement
of Portland cement in construction and repair projects [21,29], it is im-
perative to develop such theories so that MPC-based materials can be
designed in much the same way as Portland cement-based materials
are designed, namely according to the properties of the raw materials
and the desired performance. The present study aims to develop theo-
ries and methods as the basis of property prediction, and the scope is
limited to MKPC since this seems to be more favored in recent research
[4–7,12,30]. The stoichiometric factors of the chemical reaction in MKPC
paste are first analyzed, and the degree of reaction of this cement is then
defined through thermal analysis. Based on the stoichiometric factors
and the degree of reaction, the porosity of MKPC paste, which is essen-
tial for predictions of both the mechanical and transport properties, is
calculated. The pore structure is also simulated by a newly developed
computer model. At last, the theory for porosity calculation and the
model for pore structure simulation are validated by experimental re-
sults from mercury intrusion porosimetry (MIP) measurements.
2. Materials and experiments
Dead burnt magnesia powder (calcined at 1500 °C for 5 h) with a
purity of 95.1%, powder KDP (chemical reagent) and deionized water
were used as the raw materials for preparation of the MKPC paste. The
chemical composition of the dead burnt magnesia is shown in Table 1.
The milled magnesia and KDP powders were analyzed with a laser par-
ticle size analyzer, and the particle size distributions are shown in Fig. 1.
In this study, pastes with the magnesia-to-phosphate molar ratios (M/P)
of 4, 6, 8 and 12, and the water-to-cement mass ratio (W/C, where the
cement contains magnesia and KDP) of 0.20 were prepared by mixing
the raw materials according to the mix proportions. The mixing proce-
dure involved firstly dry-mixing the powders for 1 min and then further
mixing them with water for 3 min, in a vertical-axis planetary mixer.
The mixed fresh pastes were cast into plastic tubes of diameter 2 cm
and then sealed by plastic sheets. After 12 h, the specimens were de-
molded and sealed in airtight plastic bags.
At the ages of 3, 7 and 28 days, the specimens were taken out of the
bags. Cylinders 2 cm long were cut from the specimens using a diamond
saw, with the saw blade cooled by ethanol. Near-surface portions of
each cylinder were removed, and the left central part was sawn into
small cuboids with the smallest dimensions of 3–5 mm. The obtained
cuboids were then vacuum-dried to constant weight. As recommended
by Ma [31], such cuboids were used as samples for porosity measure-
ments by MIP. In addition to the porosity, pore size distribution curves
could also be obtained by MIP, based on the well-known Washburn
equation in which theoretical parameters were chosen according to
Ma [31]. A Micromeritics AutoPore IV 9500 was used for MIP measure-
ments, and the maximum pressure that could be applied was 30,500 psi
(210 MPa), which corresponds approximately to a minimum detectable
pore diameter of 6 nm. Diamond [32] pointed out that MIP cannot pro-
vide valid estimates for pore size distributions of materials with com-
plex pore structures, mainly due to the ink–bottle effect, and MIP
results are valuable only to provide porosities and threshold pore
diameters. However, as an easy and widely used method for pore struc-
ture characterization, MIP is still very useful in the comparative assess-
ment of the pore structure and in providing physically or statistically
Table 1
Chemical composition of the dead burnt magnesia (mass %).
MgO SiO2 CaO Fe2O3
95.10 3.71 0.78 0.27
97
Fig. 1. Particle size distribution curves of magnesia and KDP powders.
meaningful parameters for developed models [31]. Recently, innovative
schemes for MIP measurement have been developed to enable more ac-
curate estimation of the pore size distribution [33,34], and intrusion
curves from multi-cycle MIP measurements have been proposed as
criteria for judging the reasonability of a pore size distribution deduced
via other techniques [35,36]. In the present study, MIP measurements
were conducted in two continuous intrusion–extrusion cycles. Pore
size distribution curves deduced from the 1st intrusion were used for
porosity determination and qualitative comparison of the pore struc-
tures, and the curves deduced from the 1st and 2nd intrusions were
used for validating the computer model, by comparing them with the
pore size distribution generated by the computer model.
The degree of reaction of MKPC was assessed by thermogravimetric
analysis (TGA). To do so, dried cuboid samples as described above were
crushed, pestled and sieved, and small particles ranging from 200 μm to
500 μm were collected as samples. In a thermal analysis, about 30 mg of
a sample was put in an alumina top-opened crucible and heated from
room temperature to 600 °C at a rate of 10 °C/min. The weight loss
data was determined and recorded for further analysis. Nitrogen gas
was chosen as the dynamic atmosphere, and corundum as the reference
material.
3. From degree of reaction to porosity
The principal reaction in MAPC paste is the through-solution
reaction between the dead burnt magnesia and the phosphate that pro-
duces struvite [26]. However, besides struvite, other minerals, such as
Mg3(PO4)2·4H2O and Mg3(NH4)2(HPO4)4·8H2O, can also be formed
[13], which makes the quantitative analysis of this material more com-
plicated. Contrarily, in the XRD (X-ray diffraction) patterns of MKPC
paste reacted to some degree, all peaks other than the reactants corre-
spond to MKP [6,12,25,29], and evidence of amorphous products is
very weak. Therefore, assuming MKP to be the sole reaction product in
MKPC will not introduce considerable error in the quantitative descrip-
tion of the microstructure. Another assumption in this study is that com-
ponents other than magnesium oxide in the dead burnt magnesia are
not only inert in the MKPC reaction but also stable in thermal analysis.
3.1. Stoichiometric factors
As frequently cited [6,18,29], the reaction in a MKPC paste can be
written as
MgO þ KH2 PO4 þ 5H2 O→MgKPO4  6H2 O: ð1Þ
MnO
According to Eq. (1), 1 mol of magnesium oxide reacts with 1 mol of
0.14
KDP and 5 mol of water, and produces 1 mol of MKP. The molar mass,
98 H. Ma et al. / Cement and Concrete Research 65 (2014) 96–104

Table 2 and KMP (potassium metaphosphate, KPO3) of which the molar ratio
Properties of the 4 substances involved in the MKPC reaction. equals the initial M/P of the MKPC. Let R600 denote the remaining weight
Items M KDP H MKP at 600 °C, so the weights of magnesia and KMP in the residua can thus
Molar mass Mi (g/mol) 40.30 136.09 18.02 266.47
be calculated respectively as
Molar volume Vi (cm3/mol) 11.25 58.21 18.02 143.03
Density ρi (g/cm3) 3.58 2.34 1 1.86 eM
R600  m
RM ¼ ð4Þ
e M þ M KMP
m

molar volume and density of the four substances involved in Eq. (1) are
listed in Table 2, in which M, KDP, H and MKP denote MgO, KH2PO4, H2O R600  M KMP
and MgKPO4·6H2O, respectively. RKMP ¼ ð5Þ
e M þ M KMP
m

3.2. Degree of reaction

where MKMP is the molar mass of KMP, and

MKPC is a type of binary cement consisting of magnesia and KDP,

M M  M=P
thus its degree of reaction can be defined by the degrees of reaction of eM ¼
m : ð6Þ
ηM
the two components. When either component is exhausted, or water
is not available any more, the reaction ends. The degree of reaction of
KDP can be obtained through thermogravimetic analysis (TGA), and In Eq. (6), MM is the molar mass of magnesia, and ηM is the purity
the degree of reaction of magnesia can then be determined through (magnesium oxide mass percentage) of the dead burnt magnesia. Cor-
the stoichiometry and the material proportion. The TGA curves of a respondingly in the raw materials, the weight of the initial magnesia
typical MKPC paste and unreacted KDP, representing the weight loss equals RM, and that of the initial KDP is
following an increase of temperature, and the corresponding differen-
tial curves (DTG) are shown in Fig. 2. After being dried, a MKPC paste, MKDP
mKDP ¼ RKMP  ð7Þ
reacted to some degree, is composed of MKP, unreacted magnesia and MKMP
unreacted KDP. In thermal analysis, the decomposition of MKP starts
at around 60 °C, and almost completes at around 200 °C [37]. Unreacted
dead burnt magnesia does not change within this temperature range,
while as shown in Fig. 2, the decomposition of unreacted KDP occurs
above 200 °C. It can therefore be assumed that the decomposition tem-
perature range of MKP does not overlap that of KDP. Thus, the weight
loss of a MKPC paste above 200 °C (LN200) can be used to determine
the amount of unreacted KDP. The decomposition of MKP and that of
KDP can be written as

60–200BC
MgKPO4  6H2 O → MgO þ KPO3 þ 6H2 O↑ ð2Þ

N200BC
KH2 PO4 → KPO3 þ H2 O↑ ð3Þ

Note that the equations are written in this way for the convenience
of calculation, rather than to give accurate chemical descriptions of the
residua after heating.
According to the above discussion, the residua of a MKPC paste at
600 °C can be seen as consisting of two components, viz. magnesia

Fig. 2. TGA and DTG curves of KDP and a typical MKPC paste. Fig. 3. Flowchart of the microstructure simulation of MKPC paste.
 H. Ma et al. / Cement and Concrete Research 65 (2014) 96–104 99

where MKDP is the molar mass of KDP. Employing Eq. (3) and LN200, the
weight of unreacted KDP can be calculated as

un M KDP
mKDP ¼ LN200  ð8Þ
MH

where MH is the molar mass of water. The degree of reaction of KDP can
then be obtained as

un
mKDP
α KDP ¼ 1− : ð9Þ
mKDP

In light of Eq. (1), the degree of reaction of magnesia is written as

α KDP  ηM
αM ¼ : ð10Þ
M=P
Fig. 5. TGA results expressed in remaining weight–temperature curves.

3.3. Porosity Eq. (11) is derived on the basis of 1 g of MKPC solid, and it can be re-
written as
In a MKPC paste, when M/P and W/C are given and αKDP and αM have
e M þ MKDP Þ−ηM  α KDP  ðV MKP −V KDP Þ þ α M  V M  M=P
W=C  ηM  ðm
been determined, the porosity, or the volume fraction of the pore ϕ¼ :
e M þ MKDP Þ þ ηM  V KDP þ V M  M=P
W=C  ηM  ðm
phases, including water and air voids, can then be calculated as
ð12Þ

α KDP e M =ρM
m M =ρ
W=C− V þ  α þ KDP KDP  α KDP
e M þ MKDP MKP m
m e M þ M KDP M m e M þ M KDP
ϕ¼ : 4. Computer model for pore structure simulation
e M =ρM
m M =ρ
W=C þ þ KDP KDP
e M þ MKDP m
m e M þ MKDP In the reaction between MPC and water, phosphate is dissolved con-
ð11Þ tinuously in water, and the reaction occurs on the wetted surface of the

(a) (b) (c)

(d) (e) (f)

Fig. 4. A n-step closing operation based image processing for pore structure extraction and pore representation: (a) initial; (b) step 1; (c) step 2; (d) a middle step; (e) step n − 1; (f) final
representation of the pore structure, where darker indicates larger pores.
100 H. Ma et al. / Cement and Concrete Research 65 (2014) 96–104

magnesia particles. The reaction products grow around the magnesia

particles and connect with each other, developing the cementitious ma-
trix [26]. This reaction mechanism is employed to direct the numerical
simulation of the microstructure, or pore structure, of MKPC paste.
Given the W/C and M/P of a MKPC paste, the initial volumes of magne-
sia, KDP and water in a specific volume of paste, e.g. in a REV (representa-
tive elementary volume), are readily obtained as viM, viKDP and viH,
respectively. In a digitized REV, as will be shown later, the initial volumes
can be obtained simply by counting the corresponding voxels. To achieve
a microstructure determined by αM and αKDP, the volume of KDP that
should be reacted, or dissolved into the water for reaction, is

r i
vKDP ¼ vKDP  α KDP ð13Þ

and the volume of magnesia reacted is

r i
vM ¼ vM  α M : ð14Þ

As the reaction products (i.e. MKP) grow around the magnesia par-
ticles, they will take up the space originally occupied by the reacted
magnesia. This part of reaction products are defined as the inner
products, since they form inside the original boundary of the magnesia
particles, and the volume is denoted by vIP. Clearly,
r
vIP ¼ vM : ð15Þ

According to Eq. (1) and Table 2, when magnesium oxide is trans-

formed into MKP, the volume is enlarged by a factor of γ = VMKP/VM =
12.71. The reaction products other than the inner products are defined
as outer products, and the volume is denoted by vOP. vOP can be calculated
as
r
vOP ¼ ðγ−1Þ  vM : ð16Þ

Chemically speaking, the inner and the outer products have no dif-
ferences. They are only distinguished from each other by locations for
the sake of numerical simulation. In addition, when 1 unit volume of
magnesium oxide is reacted, the volume of water consumed in the reac-
tion is γH = 5VH/VM = 8.01. Thus, the volume of bulk water remaining
in the system can be calculated as

i r
vH ¼ vH −γ H  vM : ð17Þ

The unreacted MKPC paste is simulated by randomly distributed

spheres representing the magnesia and KDP particles in a cubic REV
(100 × 100 × 100 μm3) with periodic boundary conditions, according
to the particle size distributions, M/P and W/C. The particle size distribu-
tions in the simulation system copy the real ones obtained through
measurements, except that particles smaller than 1 μm and those larger
than 50 μm are cut off, in order to save the computational cost. Then,
Fig. 6. MIP results expressed by cumulative porosity curves: (a) repeatability of MIP
results; (b) pore structure evolution of paste with the M/P of 4; (c) pore structure compar-
Table 3
ison of pastes with different M/P at the age of 3 days.
Results of TGA experiments and calculations.

M/P Age Lb200 LN200 R600 αKDP αM Calculated

(days) ϕ (%) the REV, including the magnesia and KDP particles, is discretized
into 500 × 500 × 500 cubic voxels. Thus, the voxel size (side length)
4 3 17.49 2.15 80.36 0.572 0.136 11.42
7 18.03 2.01 79.96 0.598 0.142 10.27
is 0.2 μm. In the reaction simulation, the consumption of the reactants
28 18.66 1.82 79.52 0.634 0.151 8.69
6 3 18.81 0.77 80.42 0.802 0.127 9.20 Table 4
7 19.29 0.67 80.04 0.827 0.131 8.31 Porosities of MKPC pastes as measured by MIP.
28 19.71 0.59 79.70 0.847 0.134 7.59
8 3 17.97 0.23 81.80 0.929 0.110 10.74 Age M/P = 4 M/P = 6 M/P = 8 M/P = 12
7 18.19 0.17 81.64 0.947 0.113 10.24 (days)
ϕ (%) COV (%) ϕ (%) COV (%) ϕ (%) COV (%) ϕ (%) COV (%)
28 18.75 0.12 81.13 0.962 0.114 9.77
12 3 14.23 0.09 85.68 0.963 0.076 17.63 3 11.61 0.91 8.68 1.54 9.83 1.37 15.62 1.36
7 14.64 0.01 85.35 0.996 0.079 16.91 7 10.13 1.14 7.40 1.66 9.04 1.56 14.29 1.39
28 14.71 0 85.29 1 0.080 16.81 28 8.18 1.53 6.31 1.92 7.92 1.55 13.98 1.34
 H. Ma et al. / Cement and Concrete Research 65 (2014) 96–104
and the formation of the reaction products are both processed through a
voxel operation, so that their real-time volumes can be easily computed
by counting the voxels accordingly. By inputting a target status, which is
defined by αM and αKDP, the microstructure of the MKPC paste can be sim-
ulated in several steps, as shown in Fig. 3. First, KDP particles, which are
spherical assemblies of KDP voxels in the simulation system, are eroded
inwards from their surfaces to simulate the dissolution. The erosion is
conducted on all KDP particles layer by layer (each of 1 voxel thick),
until the volume of the eroded KDP achieves the value calculated from
Eq. (13). Simultaneously, all magnesia particles are processed in a similar
manner, but the eroded magnesia voxels are immediately transformed
into inner products. This operation stops once the volume of inner prod-
ucts equals the value calculated from Eqs. (14) and (15). In the following
step, from the original boundaries of the magnesia particles, outer product
voxels are formed outwards, layer by layer, until vOP calculated from
Eq. (16) is achieved. When the growing outer product layers make con-
tact with each other, only the free surfaces that are still exposed to the
open space are allowed to grow further in their outer-pointing normal
directions. In this step, the solid skeleton of the paste has been formed.
The open space left in the REV represents the pore phases, including
the remaining bulk water, and the voids formed due to the fact that the
volume of the reaction product is smaller than that of the reactants. The
pore structure is analyzed by a closing operation-based algorithm. This
algorithm considers the REV as 500 digital images, 500 × 500 pixels
each, in any of the three (x-, y-, and z-) directions. Each image is proc-
essed by the closing operation, as illustrated in Fig. 4. Fig. 4(a) shows
the initial state of a digital image in which white represents all solid
phases while dark represents the pore phase. The closing operation is
a direct handling of the white (solid) pixel assemblage in n steps using
disk structure elements with gradually increasing diameter from d1 to
dn. In the ith step, the closing is simply defined as a dilation followed
by an erosion using the same disk structure element with diameter of
di. In the dilation, due to the expansion of the solid phase, part of the
pore pixels along the interface are converted to solid pixels. In the
following erosion operation, some of the just converted pixels are con-
verted back to pore pixels. After this cycle, the pixels that cannot be con-
verted back are marked as pores with diameters from di − 1 to di. Step by
step, as illustrated in Fig. 4(b) through (e), the closing operation pro-
gresses until all the pore pixels are converted to solid pixels. In Fig. 4(f),
pores with different diameters determined in this way are displayed by
different gray levels, and the darker the color is, the larger the pore it rep-
resents. In the output file, each pore voxel is assigned to a cylindrical pore
with a specific diameter and 1 voxel high. After the algorithm is applied in
all the three directions, each pore voxel will have three pore diameters,
and the largest one is taken as the diameter of the pore to which the
voxel belongs. By counting the pore voxels with different diameters, a
pore size distribution curve can be obtained. As the cylindrical pore
shape assumption is consistent with the Washburn equation, which is
normally used to analyze MIP results, the simulated pore size distribution
curves can be compared with MIP results for model verification.
Finally, water adsorption is conducted by replacing pore voxels with
water voxels, from the smallest pores to the larger ones gradually, until
the volume of the remaining bulk water calculated from Eq. (17) is
achieved. Note that for the purpose of microstructure simulation, the
volume of the REV is kept constant in the computer model, and no
shrinkage is considered. The moisture condition of the MKPC paste, ob-
tained through the water adsorption algorithm, is not used in this study,
but it can be employed in future work for analyses of the shrinkage and
transport properties of the paste.
5. Results and validations
5.1. TGA results and degrees of reaction
Typical TGA curves representing the changes of the remaining
weight percentages following the increase of temperature are shown
101
in Fig. 5. Based on the TGA curves, the weight loss below 200 °C
(Lb200), the weight loss above 200 °C (LN200) and the remaining weight
at 600 °C (R600) for all dried MKPC pastes tested were determined and
are listed in Table 3. Each value is determined by averaging three results
from separate measurements. The COVs (coefficients of variation) are as
small as 0.05%–0.4%, as the samples for TGA are relatively homogeneous
due to the sampling method, proving the high reproducibility and
reliability of the measurements. In the light of Eqs. (4) through (10),
the average values of LN200 and R600 are used to calculate the degree of
reaction of KDP (αKDP) and of magnesia (αM) at different ages, in
MKPC pastes with different M/P value. The calculated results are listed
in Table 3. The results clearly show that with the same W/C of 0.2, higher
values of M/P result in higher degrees of reaction of KDP, but leads to
lower degrees of reaction of the magnesia at each age.
5.2. MIP results
MIP results, expressed by the cumulative porosity curves, are plotted
in Fig. 6. Fig. 6(a) shows three repeated results of a sample obtained
through separate tests, and the average curve. The good reproducibility
of this method is clearly shown. Curves shown in Fig. 6(b) and (c) are all
average curves obtained in this way. Fig. 6(b) illustrates the pore struc-
ture evolution of the MKPC paste with the M/P of 4. It can be seen that,
as the age or degree of reaction increases, the porosity decreases, while
the pore size distribution mode does not change much. The pore struc-
tures of MKPC pastes with different M/P values, at the age of 3 days, are
compared in Fig. 6(c). According to this comparison, the M/P of 6 gives
the lowest porosity, which is followed by the M/P of 8, 4 and 12 in
order of increasing porosity. In addition, the lower the porosity is, the
finer the pore structure (or the smaller the critical pore diameter)
seems to be. The porosities of all test MKPC samples, as averaged from
three separate MIP measurements, are listed in Table 4, together with
the corresponding COVs. As shown by the data in Table 4, the aforemen-
tioned porosity order holds not only at the age of 3 days, but also at
other ages.
5.3. Porosity calculation and validation
Based on the TGA results, the porosities of MKPC pastes were calcu-
lated using Eq. (12), and the results are listed in Table 3. From the results,
it can be seen that the MKPC paste with the M/P of 6 gives the lowest po-
rosity, followed by pastes with the M/P values of 8 and 4, and the M/P of
12 gives the highest porosity. This order is consistent with the one re-
vealed by MIP as shown in the previous section. It is worth noting that
the compressive strength and flexural strength of MPC are also not
monotonic functions of M/P, but optimized M/P values to achieve high
strength always exist and depend on the chemical composition of the
Fig. 7. Comparison of the calculated and measured porosities.
102 H. Ma et al. / Cement and Concrete Research 65 (2014) 96–104

magnesia [12,13,38]. To understand why such optimized M/P values al-
ways exist, a comprehensive study on the reaction kinetics of the mag-
nesia–KDP–water ternary system is needed in the future.
Porosities calculated from Eq. (12) and porosities measured by MIP
are directly compared in Fig. 7. Almost all data points in Fig. 7 lie
above the equality line, which means that the calculated porosities are
more or less higher than the measured porosities. It has been shown
that MIP normally underestimates the porosity, because the smallest
pores, and some large pores separated from the interconnected pore
network by the smallest pores, are not accessible to mercury [36]. In a
mature MKPC paste with a specific W/C, the higher the M/P, the more
unreacted magnesia particles will remain. With more magnesia parti-
cles, the pore structure becomes more tortuous, which may bring a
larger error to the MIP measurement. Considering this point and the
comparison shown in Fig. 7, the accuracy of the model developed for po-
rosity calculation of MKPC paste is acceptable.

initial state initial state

low degree of reaction low degree of reaction

high degree of reaction high degree of reaction

(a) (b)
Fig. 8. Simulated microstructures of MKPC pastes (W/C = 0.2): (a) M/P = 4; (b) M/P = 12, where red, white, green, blue and black represent unreacted magnesia, unreacted KDP, reaction
product, remaining water and void, respectively.
 H. Ma et al. / Cement and Concrete Research 65 (2014) 96–104 103

5.4. Simulated pore structure and validation

Two examples of the simulated microstructural evolution of MKPC

paste are shown in Fig. 8. The M/P values of the two cases are 4 and
12, respectively, and the W/C values are both 0.2. The 2D digital images
shown are cross-sectional views of the simulated 3D microstructures. It
is worth noting that the simulated pore structure is influenced by the
voxel size adopted in the simulation, and a smaller voxel means a higher
resolution. Pore size distribution curves of an arbitrary system, obtained
from simulations adopting different voxel sizes, are shown in Fig. 9. It
can be seen that when the voxel size changes from 0.5 μm to 0.2 μm,
the pore size distribution curves shifts toward smaller pore size indicat-
ing finer pore structure, and the porosity becomes slightly higher. As the
simulation uses voxels to approximate smooth boundaries, a smaller
voxel size certainly results in more accurate simulation results. A further
reduction in voxel size from 0.2 μm to 0.1 μm, however, does not result
in obvious modification any more, yet the computation time is in-
creased by one order of magnitude. Hence the voxel size of 0.2 μm
seems to strike a good balance between simulation accuracy and
computation time. This is the reason why, in the present study, the
100 × 100 × 100 μm3 REV is digitized into 500 × 500 × 500 voxels.
The pore size distribution curves from the simulation and from the
MIP of the same systems (M/P = 4 and 12, and W/C = 0.2), with com-
parable porosities, are compared in Fig. 10. The cumulative porosity
curves shown in Fig. 6 are all deduced from the 1st intrusion of mercury
in a MIP procedure, and are named the MIP 1st intrusion curves. As con-
cluded by researchers [35,36], the MIP 1st intrusion curve overestimates
the volumes of small pores due to the ink–bottle effect, while the 2nd
intrusion curve underestimates the volumes of pores in its accessible
range, thus the realistic pore size distribution curve should lie some-
where between the MIP 1st and 2nd intrusion curves. Therefore, in
the comparisons shown in Fig. 10, MIP 2nd intrusion curves are also
added for the sake of judging whether the simulation is possible to real-
istically represent the pore structure. From Fig. 10, it's evident that the
simulated curves, in the small pore-size range (smaller than several mi- Fig. 10. Comparison of the simulated pore size distribution curve with MIP results:
crons), lie right between the MIP 1st and 2nd intrusion curves, thus suc- (a) M/P = 4; (b) M/P = 12.
cessfully represent the pore structures of MKPC pastes. The failure in the
large pore-size range can be attributed mainly to the cut-off of the large
particles in the simulations, and can be overcome by improving the reaction are analyzed. Based on thermogravimetric analysis, the degree
computational efficiency and using larger REV in future versions of the of reaction of magnesia and that of KDP are determined and used to
computer model. define the degree of reaction of the MKPC paste. A model is developed
to calculate the porosity of the MKPC paste, from the stoichiometric
6. Conclusions factors, the degrees of reaction, the W/C and the M/P. Given the same
W/C of 0.2, a higher M/P results in a higher degree of reaction of KDP,
In the present study, MKP, or struvite-(K), is assumed to be the sole but leads to a lower degree of reaction of magnesia at each age. Howev-
reaction product in MKPC paste, and the stoichiometric factors of the er, as revealed by both the theoretical model and the results of MIP, po-
rosity is not a monotonic function of M/P, but the M/P of 6 gives the
lowest porosity at different ages. According to the MIP results, the
pore size distribution mode does not change much with increasing
ages or decreasing porosities in a specific MKPC paste. However, in
pastes of a specific age and with different M/P values, a lower porosity
seems to correspond to a finer pore structure. As proven by a direct
comparison of the porosities calculated from the model and those mea-
sured by MIP, the accuracy of the model is acceptable. In addition, a
computer model is developed in this study to simulate the microstruc-
ture of MKPC paste, and it has been validated by comparing the simulat-
ed pore size distribution curves with the MIP results.
Porosity and pore structure are essential for an engineering material,
like MKPC paste, as they significantly affect both the mechanical and
transport properties. The relation between the microstructure and de-
gree of reaction of MKPC paste has been well defined in the present
study. However, to predict the properties of MKPC-based materials for
design purpose, the reaction kinetics must also be properly understood.
The reaction kinetics of MKPC, to the authors' knowledge, has not
yet been studied, and would be an interesting topic to pursue in the
Fig. 9. Influence of voxel size on the simulated pore size distribution curves. near future.
104 H. Ma et al. / Cement and Concrete Research 65 (2014) 96–104
Acknowledgements
Financial support from the China Ministry of Science and Technology
under 2009CB623200 and from the Hong Kong Research Grants Council
under 615810 is gratefully acknowledged.
References
[1] D.M. Roy, New strong cement materials: chemically bonded ceramics, Science 235
(1987) 651–658.
[2] B.E. Abdelrazig, J.H. Sharp, B. El-Jazairi, The chemical composition of mortars made
from magnesia–phosphate cement, Cem. Concr. Res. 18 (1988) 415–425.
[3] S.S. Seehra, S. Gupta, S. Kumar, Rapid setting magnesium phosphate cement for
quick repair of concrete pavements — characterisation and durability aspects,
Cem. Concr. Res. 23 (1993) 254–266.
[4] I. Buj, J. Torras, D. Casellas, M. Rovira, J. de Pablo, Effect of heavy metals and water
content on the strength of magnesium phosphate cements, J. Hazard. Mater. 170
(2009) 345–350.
[5] I. Buj, J. Torras, M. Rovira, J. de Pablo, Leaching behaviour of magnesium phosphate
cements containing high quantities of heavy metals, J. Hazard. Mater. 175 (2010)
789–794.
[6] S.R. Iyengar, A. Al-Tabbaa, Developmental study of a low-pH magnesium phosphate
cement for environmental applications, Environ. Technol. 28 (2007) 1387–1401.
[7] J. Yang, J. Shin, C. Lee, C. Heo, M. Jeon, K. Kang, Stabilization of Cs–Re trapping filters
using magnesium phosphate ceramics, J. Radioanal. Nucl. Chem. 295 (2013) 211–219.
[8] Y. Liu, S. Kumar, J. Kwag, C. Ra, Magnesium ammonium phosphate formation, recov-
ery and its application as valuable resources: a review, J. Chem. Technol. Biotechnol.
88 (2013) 181–189.
[9] T. Masuda, I. Ogino, S.R. Mukai, Optimizing the dimensions of magnesium ammoni-
um phosphate to maximize its ammonia uptake ability, Adv. Powder Technol. 24
(2013) 520–524.
[10] G. Mestres, M. Ginebra, Novel magnesium phosphate cements with high early
strength and antibacterial properties, Acta Biomater. 7 (2011) 1853–1861.
[11] C. Moseke, V. Saratsis, U. Gbureck, Injectability and mechanical properties of magne-
sium phosphate cements, J. Mater. Sci. Mater. Med. 22 (2011) 2591–2598.
[12] F. Qiao, C. Chau, Z. Li, Property evaluation of magnesium phosphate cement mortar
as patch repair material, Constr. Build. Mater. 24 (2010) 695–700.
[13] Q. Yang, X. Wu, Factors influencing properties of phosphate cement-based binder
for rapid repair of concrete, Cem. Concr. Res. 29 (1999) 389–396.
[14] Q. Yang, B. Zhu, X. Wu, Characteristics and durability test of magnesium phosphate
cement-based material for rapid repair of concrete, Mater. Struct. 33 (2000) 229–234.
[15] Q. Yang, B. Zhu, S. Zhang, X. Wu, Properties and applications of magnesia–phosphate
cement mortar for rapid repair of concrete, Cem. Concr. Res. 30 (2000) 1807–1813.
[16] Q. Yang, S. Zhang, X. Wu, Deicer-scaling resistance of phosphate cement-based
binder for rapid repair of concrete, Cem. Concr. Res. 32 (2002) 165–168.
[17] A. Whitaker, J.Y. Jeffery, The crystal structure of struvite, MgNH4PO4·6H2O, Acta
Crystallogr. B 26 (1970) 1429–1440.
[18] Z. Ding, Z. Li, High-early-strength magnesium phosphate cement with fly ash, ACI
Mater. J. 102 (2005) 375–381.
[19] D. Singh, A.S. Wagh, US patent no. 5846894, 1998.
[20] A.S. Wagh, S. Jeong, D. Singh, US patent no. 5830815, 1998.
[21] A.S. Wagh, S. Jeong, D. Singh, High strength phosphate cement using industrial
byproduct ashes, in: A. Azizinamini, D. Darwin, C. French (Eds.), High Strength Con-
crete (Proceedings of the First International Conference on High Strength Concrete,
Held in Kona, Hawaii, July 13–18, 1997), ASCE, Reston, VA, 1999, pp. 542–553.
[22] M. Mathew, L.W. Schroeder, Crystal structure of a struvite analogue, MgKPO4·6H2O,
Acta Crystallogr. B 35 (1979) 11–13.
[23] S. Graeser, W. Postl, H. Bojar, P. Berlepsch, T. Armbruster, T. Raber, et al., Struvite-
(K), KMgPO4·6H2O, the potassium equivalent of struvite—a new mineral, Eur. J.
Mineral. 20 (2008) 629–633.
[24] C. Chau, F. Qiao, Z. Li, Potentiometric study of the formation of magnesium potassium
phosphate hexahydrate, J. Mater. Civ. Eng. 24 (2012) 586–591.
[25] F. Qiao, C. Chau, Z. Li, Calorimetric study of magnesium potassium phosphate ce-
ment, Mater. Struct. 45 (2012) 447–456.
[26] E. Soudée, J. Péra, Mechanism of setting reaction in magnesia–phosphate cements,
Cem. Concr. Res. 30 (2000) 315–321.
[27] E. Soudée, J. Péra, Influence of magnesia surface on the setting time of magnesia–
phosphate cement, Cem. Concr. Res. 32 (2002) 153–157.
[28] Z. Ding, B. Dong, F. Xing, N. Han, Z. Li, Cementing mechanism of potassium phos-
phate based magnesium phosphate cement, Ceram. Int. 38 (2012) 6281–6288.
[29] C. Chau, F. Qiao, Z. Li, Microstructure of magnesium potassium phosphate cement,
Constr. Build. Mater. 25 (2011) 2911–2917.
[30] C.X. Qian, J.M. Yang, Effect of disodium hydrogen phosphate on hydration and hard-
ening of magnesium potassium phosphate cement, J. Mater. Civ. Eng. 23 (2011)
1405–1411.
[31] H. Ma, Mercury intrusion porosimetry in concrete technology: tips of measurement,
pore structure parameter acquisition and application, J. Porous. Mater. 21 (2014)
207–215.
[32] S. Diamond, Mercury porosimetry — an inappropriate method for the measurement
of pore size distributions in cement-based materials, Cem. Concr. Res. 30 (2000)
1517–1525.
[33] J. Zhou, G. Ye, K. van Breugel, Characterization of pore structure in cement-based
materials using pressurization–depressurization cycling mercury intrusion
porosimetry (PDC–MIP), Cem. Concr. Res. 40 (2010) 1120–1128.
[34] J. Kaufmann, Characterization of pore space of cement-based materials by combined
mercury and Wood's metal intrusion, J. Am. Ceram. Soc. 92 (2009) 209–216.
[35] J. Kaufmann, R. Loser, A. Leemann, Analysis of cement-bonded materials by multi-
cycle mercury intrusion and nitrogen sorption, J. Colloid Interface Sci. 336 (2009)
730–737.
[36] H. Ma, Z. Li, Realistic pore structure of Portland cement paste: experimental study
and numerical simulation, Comput. Concr. 11 (2013) 317–336.
[37] S. Zhang, H. Shi, S. Huang, P. Zhang, Dehydration characteristics of struvite-K
pertaining to magnesium potassium phosphate cement system in non-isothermal
condition, J. Therm. Anal. Calorim. 111 (2013) 35–40.
[38] A. Wang, Z. Yuan, J. Zhang, L. Liu, J. Li, Z. Liu, Effect of raw material ratios on the com-
pressive strength of magnesium potassium phosphate chemically bonded ceramics,
Mater. Sci. Eng. C 33 (2013) 5058–5063.