Article

pubs.acs.org/cm

Determination of Crystal Structure of Graphitic Carbon Nitride: Ab

Initio Evolutionary Search and Experimental Validation
Junjie Wang,*,†,‡ Dong Hao,† Jinhua Ye,† and Naoto Umezawa*,†,§
†
International Center for Materials Nanoarchitectonics (MANA), National Institute for Materials Science, 1-1 Namiki, Tsukuba,
Ibaraki 305-0044, Japan
‡
Materials Research Center for Element Strategy, Tokyo Institute of Technology, 4259 Nagatsuta-cho, Midori-ku, Yokohama,
Kanagawa 226-8503, Japan
§
Center for Materials research by Information Integration (CMI2), National Institute for Materials Science, 1-2-1 Sengen, Tsukuba,
Ibaraki 305-0047, Japan
*
S Supporting Information

ABSTRACT: Although graphitic carbon nitride (g-C3N4) is a promising

photofunctional material, its structure is poorly understood. Here, we
present a systematic study of stable crystal structures of g-C3N4 by ab initio
evolutionary searching. It was discovered that off-plane distortion of
heptazine units is a characteristic of the most stable structure, which
explains a known discrepancy between the lattice parameters determined
by X-ray diffraction (XRD) patterns and the planar structures modeled in
previous studies. A phase transition from a metastable phase to the global
minimum phase provides a reasonable explanation for the observed red
shift in photoabsorption edges upon high-temperature annealing. The
recently suggested salt-melt synthesis for g-C3N4 is subject to the
contamination of hydrogen, chlorine, and lithium according to our detailed
analysis of the crystal structures of C6N9H3-Li3Cl and C6N9H3-LiCl in
comparison with the measured XRD patterns of these samples. Finally, a
viable synthesis pathway for purifying high-crystallinity g-C3N4 is proposed.

1. INTRODUCTION assumed to construct an ideal model structure of g-C3N4.

Graphitic carbon nitride (g-C3N4), which is believed to possess The heptazine ring is energetically more favored according to a
a graphitelike layered structure, has attracted extensive interest computational stud,y2,19 and its dimension of about 7.13 Å is
because of its promise as a metal-free photocatalyst1−9 and its close to the value of the coplanar repetition distance of 6.788 Å
potential applications in optoelectronics.10−14 However, the assigned for the peak at 2θ = 13.04° in the XRD pattern.
basic characteristics of g-C3N4 (e.g., electronic and optical Hence, many researchers have suggested that heptazine rings
properties) are not well understood because the ground-state linked by trigonal nitrogen atoms constitute the most stable
crystal structure of g-C3N4 is still unclear. Clarification of the local connection pattern (see the structure illustrated in Figure
structure of g-C3N4 is, therefore, a challenge and crucial for 1a). Recent studies20,22 have also suggested that carbon nitride
understanding its properties, both experimentally and theoret- is composed of graphitelike A−B stacks. However, the exact
ically. stacking positions of the atoms with respect to the adjacent
Much effort15−26 has already been applied to identify the layers are still unclear because of the experimental limitations,
structures of g-C3N4. Normally, the g-C3N4 samples used in and further study is needed to clarify the stacking order as well
research are synthesized by thermal polycondensation in air as the local geometry.
(Figure 1a). When using this method, however, the greatest Recently, salt-melt synthesis (SMS) of g-C3N4, first proposed
obstacle to experimentally confirming the crystal structure of g- by Bojdys et al.,19 has emerged as an important alternative to
C3N4 is that a sample with high crystallinity cannot be obtained. conventional thermal polycondensation. By employing this
The information that can be obtained through crystallographic method, we obtained a uniform crystalline sample, as shown by
methods has been insufficient and limited to two slightly the fine peaks in the XRD pattern in Figure 1b. However, the
temperature-dependent X-ray diffraction (XRD) peaks at 2θ = crystal model proposed by Bojdys et al. does not match the
13.04° and 27.25°. Researchers believe that these two peaks diffraction patterns, possibly because it was fitted under the
correspond to an in-plane ordering with a repeated distance of
about 6.788 Å and a planar graphitic interlayer distance of Received: July 20, 2016
about 3.273 Å, respectively. Two kinds of elementary building Revised: March 12, 2017
blocks, triazine and heptazine (tris-s-triazine), have been Published: March 13, 2017

© 2017 American Chemical Society 2694 DOI: 10.1021/acs.chemmater.6b02969

Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

Figure 1. Synthesis of g-C3N4 by thermal polycondensation and SMS using the same precursor dicyandiamide (DCDA, C2H4N4). (a) XRD pattern
and suggested crystal structure of g-C3N4 synthesized by thermal polycondensation. In the thermal polycondensation process, DCDA was placed in
an alumina crucible with a lid and heated to 550 °C in an air atmosphere for 4 h. (b) XRD pattern of g-C3N4 synthesized by SMS. In the SMS
process, DCDA was thoroughly ground together with a mixture of LiCl and KCl and heated under an inert atmosphere at 400 °C for 6 h and then
sealed in a quartz glass ampule followed by heating at 580 °C for 6 h. In this and following figures, the blue and light gray balls represent N and C
atoms, respectively.

assumption that the sample was free from any potential searching.27−29 Starting from our experimental data of thermal
impurities such as hydrogen, chlorine, or lithium, which are the condensation, we revealed the most stable configuration of
constituent elements of the precursors for the SMS synthesis of heptazine-based g-C3N4 for the first time by ab initio
g-C3N4. Schnick and co-workers24 suggested a lamellar evolutionary searching. Our computational investigation
structure with intercalated LiCl based on thorough nuclear corresponding to SMS confirms that the highly crystalline
magnetron resonance spectroscopy and diffraction studies, SMS sample we obtained was not g-C3N4 but C6N9H3·LixCl (1
although their results did not reproduce the XRD result. Bojdys ≤ x ≤ 3). Finally, a possible reaction pathway from C6N9H3·
and co-workers25 reported that triazine-based g-C3N4 could be LixCl (1 ≤ x ≤ 3) to g-C3N4 is proposed based on our
detected at the gas−liquid and solid−liquid interfaces in the computational analysis.
reactor of SMS, which suggests that the SMS condition can
promote the production of g-C3N4 with triazine units. 2. COMPUTATIONAL AND EXPERIMENTAL DETAILS
However, the crystal structure of detected g-C3N4 could not
be determined because of a lack of observable XRD peaks for 2.1. Structure Prediction using ab initio Evolutionary
Algorithm. Our computational structure searches were performed
bulk g-C3N4. Hence, further studies on the effects of impurities using Universal Structure Predictor: Evolutionary Xtallography
on the SMS sample are required to precisely determine its (USPEX)27−29 and VASP codes.30,31 A detailed introduction to
crystal structure. structure searching and a typical USPEX input file for the structure
Figure 1 shows that the procedures for the syntheses of g- search of g-C3N4 can be found in the Supporting Information. We
C3N4 by thermal polycondensation and SMS involve air and performed first-principles calculations using the generalized gradient
inert atmospheres, respectively. Previous experimental inves- approximation in the Perdew−Burke−Ernzerhof (PBE)32 form as
tigations showed that the conditions of synthesis have a implemented in VASP. The projector-augmented wave method was
significant effect on the final structure of g-C3N4,1,2,19,25 employed to treat core electrons, while the N (2s22p3) and C (2s22p2)
although a clear explanation for the effect was not provided. electrons were treated as valence electrons. As we know, a nonlocal
In the present work, to answer this question, we first studied van der Waals (vdW) bonding exists between the carbon nitride layers
and cannot be properly described by standard DFT. In recent
the synthesis reaction pathway proceeding through intermedi- works,33−35 the influence of vdW interaction on the structural
ates to g-C3N4 from dicyandiamide (DCDA, C2H4N4) in air parameters and the electronic properties of layered structures has
and inert atmospheres by using density functional theory been highlighted. A strong vdW interaction between the g-C3N4 layers
(DFT) calculations. Then, we carried out a global search for is definitely expected. In the structure search, to correctly describe the
stable structures of g-C3N4 and intermediates corresponding to nonbonded dispersion forces between g-C3N4 layers, we included the
different atmospheres using ab initio evolutionary structure van der Waals correction by using the method of Tkatchenko and

2695 DOI: 10.1021/acs.chemmater.6b02969

Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

Scheffler with self-consistent screening (it is referred to as PBE+TS-

SCS in this paper).36−40 The plane-wave kinetic-energy cutoff was 550
eV, and the Γ-centered k-point mesh resolution was 2π × 0.06 Å−1.
2.2. Structure Optimization and Electronic Structure
Calculation. For the optimization of the most stable structures and
the subsequent band structure calculations, we improved the cutoff
energy and k-point sets to 900 eV and 2π × 0.04 Å−1, respectively. The
convergence criterion for the geometry optimization was set to require
the force on the atoms to be less than 0.001 eV/Å. Recent studies33,40
revealed that the vdW-corrected screened hybrid functional of Heyd,
Scuseria, and Ernzerhof (HSE)41,42 can be a good choice to extend
layered structures to obtain good descriptions of both nonbonded
dispersion and electronic structure. We had carried out a detailed test
by using different functionals (details of the functional test are
presented in Supporting Information and shown in Figure S1). After
considering all the optimized lattice parameters of g-C3N4, a, c, and c/
a, we concluded that HSE06-TS-SCS, which was developed by
Tkatchenko and Scheffler,36−40 gives the best overall estimation.
Therefore, the lattice constants and atomic coordinates obtained with
HSE06-TS-SCS were used for the calculations of electronic properties
Figure 2. Synthesis of g-C3N4 by thermal polycondensation and SMS
in the present work, which were also performed within the vdW
using the same precursor dicyandiamide (DCDA, C2H4N4) and its
correction framework of HSE06-TS-SCS.
possible reaction mechanism. (a) Pathways for the first three steps of
The chemical bonds in heptazine-based g-C3N4 were analyzed by
the condensation of DCDA. (b) Free energy profile in inert
conducting crystal orbital overlap populations (COOP) analysis, using
environment. (c) Free energy profile in an air environment. Both of
the LOBSTER code43−46 and density of states (DOS) calculations
the profiles were obtained using the Helmholtz free energy and were
based on VASP outputs. These calculations were performed using the
computed at 800 K, which is a typical temperature for the synthesis of
same settings as the band structure calculations with the exception of a
g-C3N4.
denser k-mesh (Monkhorst−Pack mesh solution of 2π × 0.02 Å−1) for
DOS and COOP analysis.
2.3. Phase Transition Calculation. To study the phase transition
of g-C3N4 in the present work, we performed the transition state through melem to form heptazine-based g-C3N4 or directly
searches using a synchronous transit method47 implemented in the transferring to triazine-based g-C3N4. It appears that different
CASTEP code.48 We performed complete linear synchronous transit structures of g-C3N4 can be obtained by carefully controlling
(LST)/ quadratic synchronous transit (QST) calculations by starting the condensation conditions. Therefore, to reveal the under-
with a calculation to optimize LST (see detailed introduction in the lying mechanism, we calculated the overall Helmholtz free
Supporting Information). energy profiles from DCDA to melem corresponding to
2.4. Synthesis of g-C3N4, C6N9H3-Li3Cl, and C6N9H3-LiCl. The
traditional g-C3N 4 was synthesized in static air by thermal different synthesis atmospheres. The details of the free energy
polycondensation of dicyandiamide (DCDA; >98.0%) according to computation are presented in the Supporting Information.
the procedure reported in a previous study.1 In a typical process, 2 g of Figure 2b shows the free-energy profile of DCDA condensation
DCDA was placed in a lidded alumina crucible, which was inserted in an inert atmosphere, which corresponds to the following
into a muffle furnace and heated to 550 °C at a ramp rate of 2.3 °C/ reactions:
min for 4 h. After cooling to 25 °C, the resultant yellow product was
ground into powder in an agate mortar. The eutectic SMS method 3C2N4H4 → 2C3N6H6
proposed in the literature19 was adopted to synthesize the high-
crystallinity sample of C6N9H3-Li3Cl and was modified for the → C6N11H 9 + NH3
synthesis of C6N9H3-LiCl. The details are described in Supporting
Information.
→ C6N10H6 + 2NH3 (1)
2.5. Materials Characterization. X-ray diffraction patterns were The figure indicates that the first step of melamine formation
recorded on an X-ray diffractometer with Cu Kα X-ray radiation (λ = by the molecular reaction of DCDA is a downhill exothermic
0.15418 nm) in a scanning angle (2θ) range of 5°−60° (X’pert PRO;
PANalytical Co., The Netherlands). The diffuse reflection spectra were
reaction with an energy release of 3.75 eV. However, the
measured with an integrating-sphere-equipped UV−visible recording successive deamination processes are endothermic. As illus-
spectrophotometer (UV-2600, Shimadzu Co., Japan) using BaSO4 as trated in Figure 1b, the first step of the SMS of DCDA is to
the reference, and the optical absorptions were converted from the heat the mixture of DCDA and LiCl/KCl in an open inert
respective reflectance spectra according to the Kubelka−Munk environment (Ar flux) at 400 °C. As demonstrated in the
equation. The simulated XRD patterns of the predicted structures in previous experiment,19 this temperature is too low to promote
the present work were obtained using analytical and crystallization the full condensation of DCDA, although the Ar flux will also
software (Materials Studio Reflex Powder Diffraction, BIOVIA). The help the deamination by transporting the produced ammonia
experimental XRD settings were adopted in the simulation to ensure away from the sample. The situation is not improved much by
the validity of the comparison.
the successive annealing at 600 °C under a vacuum in a sealed
quartz tube. Therefore, it is expected that the deamination
3. RESULTS AND DISCUSSION reaction shown in Figure 2a can only proceed to an incomplete
3.1. Pathway of the Condensation of DCDA. As degree. In fact, there should be a good chance of trapping the
proposed in the literature,19 Figure 2a illustrates the pathway final product during the formation of melam, which still has a
of thermal condensation of the DCDA starting material toward rather high concentration of ammonia.
g-C3N4. It shows that this condensation is driven by In the normal thermal polycondensation of DCDA (Figure
deamination and the formation of aromatic units. We can see 1a), the involvement of O2 from the surrounding air is
that melam can further branch into two channels: advancing unavoidable. Hence, we incorporated the following reaction to
2696 DOI: 10.1021/acs.chemmater.6b02969
Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

calculate the free energy profile of DCDA polycondensation in

air:
3C2N4H4 + 1.5O2 → 2C3N6H6 + 1.5O2
→ C6N11H 9 + 0.75O2 + 1.5H 2O + 0.5N2
→ C6N10H6 + 3H 2O + N2 (2)
After the introduction of O2, the deamination steps shown in
Figure 2c are thermodynamically favored, with a dramatic
energy decrease accompanying each step. Figure 2a illustrates
that the deamination process will lead to the production of
melem, a direct precursor of heptazine-based g-C 3N 4.
Consequently, our free energy calculations suggest that the
production of heptazine-based g-C3N4 is highly energetically
favored in an air atmosphere, which is consistent with the
results of previous experiments.
The free energy profile calculations offer important clues to
why the structures synthesized in thermal polycondensation
and SMS are so different. Therefore, we performed an
evolutionary structure search based on two assumptions: (1)
the air atmosphere in the thermal polycondensation process
favors the production of heptazine-based g-C3N4. It is worth
mentioning that defects have been reported in g-C3 N4
synthesized by thermal polycondensation.49 However, defects Figure 3. Prediction of heptazine-based g-C3N4. (a) Predicted most
were not considered in the present study because based on the stable configuration, phase 1; (b) second stable distorted config-
calculated free energy profile (Figure 2c), we believe C3N4 will uration, phase 2; (c) most stable planar configuration, phase 3. The
be the dominant phase in the product. The effect of defects on atoms in the middle layer of the lattices are labeled with yellow crosses.
the properties of g-C3N4 is certainly important and must be
studied in future investigations. (2) The high-crystallinity
sample obtained by SMS contains a rather high concentration
of extra ammonia. Lithium and chlorine can also be included
owing to the use of LiCl in the SMS synthesis.
3.2. Model Structures of g-C3N4 Optimized under
Assumption 1. 3.2.1. Ab Initio Evolutionary Structure
Search. The following clues to the g-C3N4 structure obtained
from thermal polycondensation1,2,18 were gathered from the
evolutionary structure search: an in-plane repetition distance of
6.788 Å, which is comparable with the size of a heptazine unit
(about 7.13 Å) and an interlayer distance of about 3.273 Å.
Therefore, it is reasonable to assume that there are two layers of
C3N4 in a primitive lattice and that each layer consists of one
heptazine unit, which is composed of six carbon and eight
nitrogen atoms. The search was performed with 28 atoms per
unit cell. More than 4000 structures were sampled in this search
to confirm the reliability of the final result. The most stable
structures, phases 1−3, are shown in Figure 3. The pathway in
Figure 2a shows that melem can be regarded as the direct
precursor for the heptazine-based g-C3N4. To confirm the
thermal stability of g-C3N4, we calculated the Helmholtz free Figure 4. (a) Detailed structure of a distorted heptazine unit and (b)
energy change from melem to the most stable g-C3N4 (phase possible rearrangement route from phase 2 to phase 1. The red arrows
indicate the directions of movement of the atoms.
1) in air using the following equation:
C6N10H6 + 1.5O2 → 2C3N4 + 3H 2O + N2 (3)
through the atoms Nlink, Ncen, and C1. The significant difference
The calculated energy change of −17.20 eV per formula unit between g-C3N4 and the typical honeycomb layered structures
shows that the condensation of heptazine-based g-C3N4 is (e.g., graphite and h-BN) is that each layer consists of heptazine
thermodynamically favored. All the predicted structures exhibit units and pores. Therefore, g-C3N4 is more susceptible to
graphitelike sheets that consist of polymerized heptazine units distortion. Our calculation shows that the distortion will
and pores with comparable sizes, and they are aligned with the considerably stabilize the structure. The computed phonon
A−B stacking configuration. Interestingly, the N and the C dispersion of phase 1 shows no imaginary frequency (Figure
atoms in phases 1 and 2 were found to be distorted from those S2a), thus confirming the dynamic stability of the thermody-
in the planar-layered structure represented by phase 3. Figure namically stable phase.
4a shows that the peripheral nitrogen atoms, Nper, and the The relative energy difference and lattice parameters of three
carbon atoms C2 and C3 are shifted around an axis passing predicted g-C3N4 structures and two planar structures reported
2697 DOI: 10.1021/acs.chemmater.6b02969
Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

Table 1. Comparison of Energetics, Lattice Parameters, and Band Gap of Predicted g-C3N4 Structures with Those Reported in
the Literature18 (T1 and T2)a
lattice constant (Å) Eg (eV)
structure ΔE(eV)/C3N4 symmetry a c c/a volume (Å3) direct indirect
phase 1 0 C2221 6.898(1.62) 6.679(2.03) 0.968 275.800 2.87 −
phase 2 0.056 Aba2 6.893(1.55) 6.754(3.18) 0.980 278.377 3.17 3.14
phase 3 0.313 Amm2 7.061(4.01) 6.414(−2.02) 0.908 276.873 2.66 2.27
T1 0.329 P̅6m2 7.060(4.00) 6.536(−0.15) 0.926 282.130 2.64 2.27
T2 0.532 P̅6m2 7.060(4.00) 7.141(9.09) 1.011 308.265 2.43 2.17
experiment − − 6.788 6.546 0.964 261.86 2.86 −
a
Detailed information on the crystal structure of the predicted structures can be found in the Crystallographic Information Files of Predicted
Structures.

by Tyborski et al.18 (hereafter referred to as T1 and T2) are energy (Figure S3(a)) shows that phase 1 possesses lower free
listed in Table 1; the experimental data are also listed for energy than phase 2. Therefore, there is a thermodynamic
comparison. The vertical lattice constant, c, which is twice the purpose for the transition from phase 2 to pPhase 1.
interlayer distance, shows a dependence on the stacking modes. Thermodynamically, the rate at which a system overcomes
For example, the values of c are 6.414, 6.536, and 7.141 Å for the energy barrier for the transition from a metastable phase to
the planar configurations of phase 3, T1, and T2, respectively. the most stable phase is increased as the temperature is raised.
The A−A stacking structure of T2 exhibits a much larger value Hence, the higher the reaction temperature, the greater the
of c than those of the A−B stacking structures of phase 3 and probability for the phase transition from phase 2 (with a wider
T1. On the other hand, the lateral lattice constant a is direct band gap, 3.17 eV) to phase 1 (with a smaller direct band
influenced by the distortion. In the earlier studies,1 researchers gap, 2.87 eV). Consequently, the reaction-temperature depend-
gave an ambiguous explanation that the possible tilt angularity ence of the band gap of g-C3N4 can be observed.
of g-C3N4 could be the reason that the lateral lattice constant Our structural studies of phases 1 and 2 showed that the
(∼ 6.81 Å) is smaller than the size of heptazine units (∼ 7.13 three Nlink−C bonds located at the corner of a heptazine unit
Å). Our predicted crystal structures explicitly reveal that this (Figure 4a) are about 11% longer than other N−C bonds
difference originates from the distortion of the basic g-C3N4 (Table S1). Among the three Nlink−C bonds, two (Nlink−C2)
unit. are identical to each other and are 1% longer than the third one
Here, we propose the concept of “micro” and “macro” A−B (Nlink−C4). On the basis of this observation, a possible
stacking sequences. We name the stacking sequence of −C− rearrangement route from phase 2 to phase 1 was proposed
N−C−N− along the c direction observed in phases 2 and 3 as (as illustrated in Figure 4b): first, the two longer Nlink−C bonds
“micro” A−B stacking, where there is only a shift of C and N break; next, a 60° rotation of a heptazine unit occurs; and
atoms with respect to those in the adjacent layers, without finally, the lattice and the atomic positions relax to the
rotation of the heptazine units. However, phase 1, the most configuration of phase 1.
stable configuration of g-C3N4, exhibits a unique structure in To support our hypothesis, a search for a transition state
which the heptazine units in one layer are rotated by 60° with between phase 2 and phase 1 was carried out based on the
respect to the basic units in the adjacent layers. If we regard the route illustrated in Figure 4b. A reaction pathway between the
heptazine units and the pore regions as “macro” A and B sites, initial and the final configurations was interpolated using the
respectively (Figure 3a), phase 1 exhibits a perfect “macro” A− synchronous transit method.47,48 Finally, we found a transition
B stacking arrangement in which the heptazine units sit exactly state with an energy barrier of 1.87 eV/C3N4 referenced to
on the pores of the adjacent layer. The lowest energy of phase 1 phase 2 (Figure S4a). After the introduction of entropy
reveals that the “macro” A−B stacking manner can stabilize the contributions of phase 1, phase 2, and the transition state
g-C3N4 structure to the greatest extent. Moreover, neither bare (Figure S4b), the energy barriers of the phase transition were
PBE nor HSE06 calculations yields significant differences in the updated, as shown in Figure S4(c). Following the classical
total energy of phase 1 and the metastable phases, and the absolute rate theory,50−52 the jump rate to overcome the energy
lattice parameter c of phase 1 is largely overestimated when the barrier in the experimental temperature range can be estimated
vdW interaction is omitted (Figure S1). These results prove by the equation
that vdW interaction plays an important role in stabilizing the
layered structures of g-C3N4. ⎛ E ⎞
Γ = Γ0 exp⎜ − b ⎟ ,
3.2.2. Phase Transition. It was observed in previous ⎝ kBT ⎠ (4)
experiments1−4 that the optical absorption edge of g-C3N4 is
shifted toward the longer-wavelength region with increasing where Γ is the jump rate of the phase transition from the initial
condensation temperature, indicating a slight dependence of to the final structure, Γ0 is the ratio of the vibrational
the band gap on the reaction temperature. To study the effect frequencies of the initial configuration to the frequencies of the
of annealing temperature on the variation in band gap of g- transition state structure, kB is the Boltzmann constant, T is the
C3N4, a possible phase transition between different phases of g- temperature, and Eb is the energy barrier shown in Figure
C3N4 was investigated, and the results are discussed in this S4(c). The most important feature of eq 4 is that an increase of
section. It is reasonable to assume that the g-C3N4 obtained in a few degrees in the temperature results in a sizable increase in
the experiments was a mixture of the different phases, such as the jump rate by several orders of magnitude. This indicates
phase 1 and phase 2, which possess different band gaps, as that the jump rate is dominated by the energy barrier Eb rather
discussed in the following section. The computed Gibbs free than Γ0, and thus it is common to use a typical value51−53 of
2698 DOI: 10.1021/acs.chemmater.6b02969
Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

Figure 5. Electronic structures of the predicted phases 1 and 3. Calculated band structures of (a) phase 1 and (b) phase 3; calculated partial charge
densities of π1, π1*, and LP states at the VBM and CBM of (c) phase 1 and (d) phase 3.

atomic vibration frequency between 1012 and 1013 s−1 for Γ0. A (Figure 5a), whereas phases 2 and 3 possess indirect band gaps
value of 5 × 1012 was adopted for Γ0 in our estimation. The of 3.14 eV (Figure S5) and 2.27 eV (Figure 5b), respectively.
jump rates of the phase transition between phase 1 and Pphase Meanwhile, the band gaps of T1 (2.27 eV) and T2 (2.17 eV) are
2 were estimated in the typical temperature range for the much narrower than the experimental value. The computed
synthesis of g-C3N4 (i.e., between 673 and 883 K),1−3 and the band gap of phase 1 is consistent with the experimental band
results are shown in Figure S4d. They show that a jump rate Γ gap of g-C3N4 synthesized by thermal condensation. Surpris-
≥ 1 s−1 (indicating that the transition takes place with a high ingly, the interlayer distance (listed in Table 1) does not
possibility) for the phase transition from phase 2 to phase 1 can influence the band gap, which is contrary to the trend exhibited
be obtained when the temperature is higher than 770 K. In the by other layered materials.33,54 Instead, the distortion of the
same temperature range, the jump rate for the reverse transition heptazine unit was found to determine the band gap, and the
from phase 1 to phase 2 is much lower than that for the distortion of the heptazine unit is strongest in phase 2 (Table
transition from phase 2 to phase 1. Therefore, the transition S2). Consequently, phase 2 has the widest band gap, as shown
from the meta-stable state (phase 2) to the ground state (phase in Table 1. In a planar structure like that in phase 3, the valence
1) is the dominant reaction process at the elevated temperature band maximum (VBM) consists of antibonding states of lone
if we assume that phase 1 and phase 2 coexist in the as-prepared pair (LP) electrons of peripheral nitrogen, while pz orbitals on
samples. As the temperature increases, so does the fraction of these nitrogen sites form π1 bands right below the LP band
the material that has undergone the phase transition. Therefore, [Figure 5 (panels b and d) and Figure S6 (panel c and d)].
phase 1 with a smaller band gap would predominate. This It is understood that the repulsion between the LP orbitals in
explains the main features of the reaction-temperature depend- phase 3 is strong because lobes of the LP orbitals at different
ence of the band gap of g-C3N4. Nper sites lie on the same heptazine plane, and electrons
3.2.3. Electronic Properties of Predicted g-C3N4. We occupying these orbitals are closely distributed (Figure 5d). As
performed electronic structure calculations for phases 1−3, a result, the LP band is higher than the π1 band in phase 3
and their band gaps are presented in Table 1. Our (Figure 5b). The electronic structure at the VBM is drastically
computational results for the previously suggested planar- changed by the distortion of the heptazine unit, as observed in
shaped T1 and T2 phases are also listed in Table 1. It is the band structures for phase 1 and phase 2 (Figure 5a). The
understood that phase 1 possesses a direct band gap of 2.87 eV LP bands are largely shifted downward and partially hybridized
2699 DOI: 10.1021/acs.chemmater.6b02969
Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

with the π1 bands. This originates from the release of the LP

orbitals toward the off-plain open space; the lobes protrude
from the heptazine plane (Figure 5c), alleviating the Coulombic
repulsive interactions among the LP orbitals. The distortion,
therefore, greatly stabilizes the system, giving a clear
explanation of why phase 1 and phase 2 are energetically
favored over phase 3. We also observed that the Nper−Nper
distances across the peripheral nitrogen atoms are slightly
increased in phase 1 and phase 2 (Table S2). The on-site
LP−π1 hybridization, which also contributes to the stabilization
of the system, is manifested as resonance of the two orbitals in
the band structure (Figure 5a) and character-decomposed DOS
(Figure S6(b)) for phase 1. It was confirmed that the computed
ionization potential is greater in the distorted phase 1 (6.64 eV)
and phase 2 (6.79 eV) than in the planar phase 3 (6.33 eV);
these results are commensurate with the magnitude of the band
gap (Table 1). This mechanism is also supported by our
analysis of the effects of distortion on a single-layered heptazine
g-C3N4 nanosheet, where the band gap opening associated with
on-site LP-π1 hybridization was observed (Figure S7). This
result supports the idea that the interlayer interaction is not the
dominant factor for the determination of the band gap. On the
other hand, the distortion does not significantly alter the nature
of the conduction band minimum (CBM), which is mainly
composed of π orbitals of carbon (π1*). The electronic
structure of phase 1 is consistent with the result of a previous
photoelectron spectroscopy study on tri-s-triazine by Shahbaz
et al.55
As described in the previous section, the stability of g-C3N4
can be influenced by the stacking sequence (“micro” and
“macro” stacking sequences) and the distortion of the heptazine
unit. On the other hand, the electronic structures of g-C3N4
strongly depend on the distortion of the heptazine unit and are
not significantly influenced by the stacking order.
Figure 6. Flowchart of the evolutionary structure search for highly
3.3. Evolutionary Structure Search of High-Crystal- crystalline phases synthesized through SMS. The lattice parameters
linity Phases under Assumption 2. The XRD data shown in (including space group of P63cm) were used only for setting up a
Figure 1b, which will be referred to as the SMS high- series of input crystal structures as initial guesses, and they were
crystallinity phase in the following section, has been used as an generated by random number operators. The initial lattice parameters
initial guess for our ab initio evolutionary structure search. and the space group were fully relaxed and modified during the
However, the composition of the SMS high-crystallinity phase evolutionary search by applying genetic operators such as heredity,
is still unclear as discussed earlier. Figure 6 shows the flowchart mutation, etc. Therefore, the obtained structure for each composition
of our refinement process for finding consistent crystal possesses a space group that gives the global minimum of the enthalpy
structures with an experimentally measured XRD pattern set of formation in a wide range of crystal symmetries.
as the criterion for the evolutionary search. The experimental
lattice parameters were used as input guesses, and the instead of heptazine rings, and its total energy is only 0.010 eV
evolutionary search based on the adjusted compositions was higher than that of phase 1 per formula unit. The computed
continued until the criterion was satisfied. Following previous Gibbs free energy of the triazine-based g-C3N4 is about 0.1 kJ/
experimental studies, we assumed that C and N are the mol higher than that of the heptazine-based phase 1 at ambient
dominant elements in this highly crystalline phase in the first pressure (Figure S3(a)). In addition, the characteristics of this
step, and we introduced potential impurities such as hydrogen, structure are similar to those of the heptazine-based phase 1;
chlorine, or lithium in the subsequent steps until the criterion the g-C3N4 planes are distorted and a “macro” A−B stacking
was satisfied. Through this strategy, we efficiently approached sequence between the neighboring layers was also observed.
the most reliable structure by gradually increasing the However, the calculated phonon dispersion of the triazine-
complexity of the system. based g-C3N4 shows slightly negative frequencies (Figure
For the structure search of g-C3N4, we examined the lattice S2(c)). The present calculation result reaffirms that the
volume of the experimentally determined primitive cell of the heptazine-based g-C3N4 is the most stable phase, as suggested
SMS sample and found it to be about 1.5 times the cell size of by the experiments.2,21 However, a recent study56 reported that
phase 1 (discussed in the last section). In the evolutionary prospective crystal structures featuring computed dynamically
structure search, therefore, we used a larger cell that includes 42 unstable phases should not be fully excluded from the
atoms for g-C3N4 (i.e., a cell with the formula C18N24). possibility of synthesis. Moreover, the very slight difference
Interestingly, after sampling more than 3000 structures, a between the triazine- and heptazine-based phases reveals that
triazine-based g-C3N4 (Figure 7a) was identified as a metastable the triazine-based phase could be realized under certain
phase. This new g-C3N4 structure is composed of triazine rings conditions.25
2700 DOI: 10.1021/acs.chemmater.6b02969
Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

Figure 7. Predicted structures corresponding to the SMS condition: (a) triazine-based g-C3N4; (b), C3N5H3; (c) C2N3H-P63m; (d) C3N6H3-HCl;
(e) C6N9H3-LiCl; (f) C6N9H3-Li3Cl. The small pink and green balls represent H and Li atoms, respectively. The large green balls in (d)−(f)
represent Cl atoms. The atoms in the lower layer of the lattices are labeled with yellow crosses.

The space group and the lattice parameters of the predicted crystallinity phase synthesized by SMS, the parameters a and b
triazine-based are listed in Table 2, which shows that even are quite different from the experimental values. The
though g-C3N4 possesses the same space group as the high- comparison of simulated XRD and experimental patterns
2701 DOI: 10.1021/acs.chemmater.6b02969
Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

Table 2. Space Groups, Lattice Parameters, And Band Gaps C3N4. Therefore, we calculated the change in Helmholtz free
of Predicted Stable Structures Corresponding to SMSa energy for the transition from melam to g-C3N4 in an inert
atmosphere using the following equation:
space band gap
composition group lattice parameters (eV)
C6N11H 9 → 2C3N4 + 3NH3 (5)
g-C3N4 P63cm a = b = 8.004 Å α = β = 90° 3.31
(triazine)
The calculated energy change of the above reaction is about
c = 6.920 Å γ = 120°
2.14 eV per formula unit. The very small energy difference
C3N5H3 P63cm a = b = 8.718 Å α = β = 90° 4.07
between the triazine- and heptazine-based g-C3N4 shows that
c = 6.786 Å γ = 120°
the SMS of g-C3N4 is thermodynamically unfavorable.
C2N3H P63cm a = b = 8.550 Å α = β = 90° 4.47
The next step is to include H atoms in our structure search.
c = 6.662 Å γ = 120°
It is not wise to introduce an arbitrary quantity of H atoms, as it
P63m a = b = 8.539 Å α = β = 90° 4.38
will lead to too many compositions and eventually an infinite
c = 6.355 Å γ = 120°
structure search. As previously mentioned in connection with
C6N9H3·HCl Pm a = 8.433 Å α = β = 90° 3.72
Figure 2a, the deamination from the precursor to form g-C3N4
b = 8.396 Å γ = 119.90°
is energetically unfavorable in an inert environment. Therefore,
c = 6.504 Å
it is highly possible that extra ammonia (NH3) exists in the
C6N9H3·LiCl Cmcm a = b = 8.545 Å α = β = 90° 3.38
c = 6.648 Å γ = 121.08°
high-crystallinity phase synthesized by the SMS method. The
a
condensation pathways in Figure 2b show that three NH3
Detailed crystal information on these structures can be found in groups need to be removed from one melam (C6N11H9)
Crystallographic Information Files of Predicted Structures.
molecule to form g-C3N4. Therefore, in our structure search, we
adopted the two compositions of C6N10H6 and C6N9H3, which
(Figure S8) explicitly proves that they are different structures correspond to the condition of removing one and two NH3
and refutes the idea that the high-crystallinity phase synthesized groups from each C6N11H9 molecule, respectively. Since the
by SMS is g-C3N4. Moreover, based on the condensation size of one C6N11H9 molecule is comparable with the area of
pathway shown in Figure 2a, we know that melam (C6N11H9) the (001) plane of the high-crystallinity phase formed by SMS,
can be regarded as the direct precursor for the triazine-based g- it is reasonable to assume that there are two C6N10H6 or

Figure 8. Comparison of the simulated XRD patterns of (a) C3N5H3, (b) C2N3H (P63cm), (c) C6N9H3·LiCl, and (d) C6N9H3·Li3Cl with the
experimental XRD patterns of the highly crystalline samples synthesized by SMS and modified SMS.

2702 DOI: 10.1021/acs.chemmater.6b02969

Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

C6N9H3 molecules per unit cell. Two ab initio evolutionary HCl looks similar to that of C2N3H-P63m. The C2N3H layers
structure searches were carried out with the compositions of stack along the same “macro” A−B stacking mode as that of
C12N20H12 (44 atoms) and C12N18H6 (36 atoms). Finally, one C2N3H-P63m. The side view of C6N9H3·HCl shows that Cl
most-stable structure with the composition of C3N5H3 (the atoms are placed on the same plane as the C2N3H layers. The
simplest formula) and two most-stable structures with the introduced H atoms are linked with the nitrogen atoms in the
composition of C2N3H (the simplest formula) were obtained, triazine rings. Our calculation shows that the band gap gets
and they are shown in Figure 7 (panels b and c) and Figure S9, narrower after the introduction of Cl. However, the value of
respectively. 3.72 eV (Table 2) is still quite far from the visible spectrum.
It is interesting to note that all these structures are layered. More importantly, there is still little similarity between the
Figure 7b shows that the most stable configuration of C3N5H3 simulated XRD pattern of C6N9H3·HCl and the experimental
is composed of melem (C6N10H6) molecules, and it possesses pattern (Figure S11). Therefore, a new ab initio evolutionary
the space group suggested by the experiment, P63cm, but with structure search was carried out for the composition of
larger lattice parameters (Table 2). Two configurations with C6N9H3·LiCl (40 atoms/cell), and finally, the most stable
almost identical energies but with different structural configuration with space group of Cmcm (shown in Figure 7e)
parameters were found for the formula of C2N3H. The one was obtained.
with the space group P63m (Figure 7c) shows a lower energy Table 2 shows that the lattice parameters of the relaxed
than the one with the space group P63cm (Figure S9). structure of C6N9H3·LiCl are consistent with the experimental
However, the energy difference is only 0.0002 eV/C2N3H, data. The simulated XRD pattern of C6N9H3·LiCl (Figure 8c)
which is much smaller than the chemical accuracy and can be also shows good consistency with the experimental XRD
ignored. Figure 7c shows that the stacking of the neighboring pattern of the SMS high-crystallinity sample, except for the
layers in C2N3H-P63m shows a “macro” A−B sequence. As peak at 18.02°. This inconsistency may be attributed to the
discussed in the previous section, this stacking sequence can different symmetries of the predicted structure (Cmcm) and the
fully stabilize the structure and decrease the interlayer distance, experimental one (P63cm). In a previous research, Schnick and
which is proved by the values of the parameter c of C2N3H in co-workers24 found by elemental analysis that the composition
Table 2. The space group and the lattice parameters of the of the SMS highly crystalline phase was C12N17.5H6.3Cl1.5Li3.2.
predicted C3N5H3 and C2N3H-P63cm appear to be very To compare with their result, our formula for the predicted
consistent with the experimental data. The simulated diffraction structure can be rewritten as C12N18H6Cl2Li2, which shows a
peaks are compared with the experimental XRD patterns in lower Li content than the experimentally determined formula.
Figure 8 (panels a and b). It is clear that the XRD peak This indicates that extra Li atoms could exist in the lattice cell
positions of C3N5H3 are shifted toward lower angles when and might be helpful in stabilizing the P63cm symmetry.
compared to the experimental peaks, whereas the XRD peak As the next step, a series of new structure searches was
positions of C2N3H-P63cm are very consistent with the carried out for the systems with the composition of C6N9H3·
experimental data. Furthermore, the relative intensities of the Li1+xCl (x = 1, 2, ...). Finally, we arrived at a new structure
peaks of both C3N5H3 and C2N3H-P63cm are quite different (Figure 7f) with the composition of C6N9H3·Li3Cl and space
from those obtained in the experimental XRD data. The band group of P63cm; its simulated XRD pattern shows perfect
structures of all the predicted structures (Figure S10) were agreement with the experimentally observed XRD pattern
calculated, and they showed that the band gaps of C3N5H3, (Figure 8d) of the highly crystalline SMS sample. For the first
C2N3H-P63m, and C2N3H-P63cm are in the range from 4.07 to time, we explicitly proved the composition of the high-
4.47 eV (Table 2), which are far beyond the visible spectrum crystallinity SMS sample and obtained its crystal structure.
(1.65−3.27 eV). Previous investigations1−4 have shown the We noticed that the concentration of Li in the predicted
visible-light-activated photocatalysis property of the SMS high- structure of C6N9H3·Li3Cl is twice that of the experimentally
crystallinity materials. Therefore, it appears that our structure determined composition of C12N17.5H6.3Cl1.5Li3.2. This incon-
search within the composition of C−N−H has yet to reach the sistency comes from the difficulty in precisely determining the
right structure. composition of a compound with light elements. Figure 7
Before introducing new elements to the C−N−H system, we (panels e and f) shows that the structures of C6N9H3·LiCl and
needed to determine the best composition for the starting point C6N9H3·Li3Cl are quite similar; the voids of the layers of
for further structure searches. Table 2 shows that the lattice C6N9H3 are stacked upon each other, forming channels along
parameters of C3N5H3 are larger than those extracted from the the c axis by “micro” A−B stacking, while the Cl atoms sit at the
experimental data, while those of C2N3H are smaller. We expect centers of the void channels and are located on planes that are
that the introduction of other species will further expand the different from those of the C6N9H3 layers. However, because of
lattice volume but also cause further deviation of the structure the removal of the extra Li atoms, the overlap of the triazine
from the experimental results if C3N5H3 is adopted as the initial rings in the C6N9H3·LiCl cell (Figure 7e) is not as perfect as
composition. Thus, we chose the composition of C2N3H as the that in the C6N9H3·Li3Cl cell (Figure 7f). This deviation can be
starting composition for the subsequent evolutionary structure attributed to the transition from the “micro” A−B stacking to
search. the “macro” A−B stacking, which can stabilize the structure by
First, we introduced Cl atoms into the C2N3H-P63m unit cell decreasing the interlayer repulsion and lead to the change in
by adding one Cl atom per C2N3H layer. Meanwhile, the same symmetry from P63cm to Cmcm. Since the concentration of Li
number of H atoms was also added for charge compensation; in in C6N9H3·Li3Cl is three times that of Cl, it is not very stable
other words, C6N9H3·HCl (40 atoms/cell) was adopted as a from the point of view of charge compensation. Therefore, we
search composition for the C−N−H−Cl system. From the ab suspected that C6N9H3·Li3Cl might be a metastable phase in
initio evolutionary structure search, a structure with space comparison with C6N9H3·LiCl. We designed a modified
group of Pm was found to be the most stable configuration of experimental procedure based on the method reported by
C6N9H3·HCl (Figure 7d). The predicted structure of C6N9H3· Bojdys et al.19 to confirm our speculation. In the modified
2703 DOI: 10.1021/acs.chemmater.6b02969
Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

Figure 9. Proposed reaction pathway to synthesize highly crystalline g-C3N4 starting from C6N9H3·LiCl.

procedure, we boiled our final sample for 5 min instead of just The space group of the predicted triazine-based g-C3N4 is
washing it with boiling water. The sample was then centrifuged P63cm, which is the same as the symmetry of C6N9H3·Li3Cl
it at 16000 rpm for 10 min. This procedure was repeated at and also quite similar to that of C6N9H3·LiCl. Therefore, the
least 15 times. We present the XRD pattern of this modified stable structure of C6N9H3·LiCl can be a good starting point to
SMS sample in Figure 8c together with those of the SMS high- synthesize highly crystalline g-C3N4 by the postsynthesis
crystallinity phase and the predicted C6N9H3·LiCl result. It modification (PSM) method, which has been highlighted in
clearly shows that a new peak appears at about 18°, which is recent studies.58,59 These two structures of C6N9H3·HCl and
consistent with the XRD pattern of the predicted C6N9H3·LiCl C6N9H3 (C2N3H) can be regarded as the intermediates for the
structure. The appearance of this new peak indicates the transition from C6N9H3·LiCl to g-C3N4. Consequently, a
transition of the symmetry from P63cm to Cmcm and validates possible reaction pathway to produce highly crystalline g-
our hypothesis. Therefore, we claim that a more stable new C3N4 from C6N9H3·LiCl has been proposed, as shown in
phase of C6N9H3·LiCl has been discovered theoretically and Figure 9. The possible experimental procedure is described in
validated experimentally for the first time. detail as follows.
The computed band structure of C6N9H3·LiCl presents a The first step is to remove Li from C6N9H3·LiCl or, more
direct band gap of 3.38 eV, which is quite close to the precisely, to replace Li with H. Our experiments on the removal
experimental value of 3.07 eV (Figure S12). Moreover, both the of Li from C6N9H3·Li3Cl have proved that the lithium ion can
be easily removed because of its small size (about 0.76 Å).
top of the valence band and the bottom of the conduction band
However, the structure of C6N9H3·LiCl is indeed very stable
are very flat (Figure S11(a)), indicating the good photo-
and the charge compensation has already been achieved. The Li
absorption ability of C6N9H3·LiCl as a photofunctional
ions cannot be removed by further boiling and centrifuging.
material.57 This clearly supports the good photocatalytic Hence, we suggest dispersing C6N9H3·LiCl in an aqueous HCl
performance of this material1−9 even in the presence of a solution to replace Li+ with H+. The high densities of Li+ and
fairly wide band gap. H+ in C6N9H3·LiCl and in the HCl solution, respectively, will
Through a series of high-throughput ab initio evolutionary drive the ion exchange between Li+ and H+. The removal of Cl
structure searches, we confirmed that the high-crystallinity SMS ions in the second step could be slightly more difficult because
phase is not g-C3N4; instead, the predicted structure should be of their larger size, about 1.81 Å. Two possible solutions are
C6N9H3·Li3Cl. More importantly, we proved that the C6N9H3· suggested. The first solution is to peel the bulk C6N9H3·HCl
Li3Cl is a metastable phase, and we discovered the more stable into nanosheets by soft chemistry synthesis; the Cl ions can
structure with a composition of C6N9H3·LiCl theoretically and then be easily removed by a chemical method to obtain C2N3H
then validated our prediction by a modified experimental nanosheets. Through the oxidation of the C2N3H nanosheets,
procedure. The study so far explicitly shows the power of the graphitic carbon nitride can be synthesized. Another option is
state-of-the-art high-throughput structure search method for to heat C6N9H3·HCl in a gas flow that contains H2 at a
the prediction of structures of novel materials. One question medium to high temperature. The small H2 molecules can
that has been raised is the following: Can we give further clues diffuse into the C6N9H3·HCl to drag the Cl out in the form of
on how to synthesize new materials like highly crystalline g- HCl. The continuous gas flow can remove the HCl from the
C3N4? Our answer is definitely “Yes”. sample surface to perpetuate the reaction. However, the
2704 DOI: 10.1021/acs.chemmater.6b02969
Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

concentration of H2 and the operating temperature should be heptazine-based Phase 1 and Phase 2; (Figure S5)
carefully controlled to protect the sample from possible electronic structure of Phase 2; (Figure S6) electronic
hydrogen etching. Finally, highly crystalline g-C3N4 can be structures of distorted Phase 1 and planar Phase 3;
obtained by heat-treating C2N3H in an oxidizing environment. (Figure S7) electronic structures of distorted and planer
Recently, Bojdys and co-workers25 detected triazine-based g- g-C3N4 single layers; (Figure S8) simulated XRD pattern
C3N4 during SMS of poly(triazine imide) with intercalated and calculated band structure of triazine-based g-C3N4;
bromide ions (PTI/Br) at the gas−liquid and solid−liquid (Figure S9) predicted structure of C2N3H-P63 cm;
interfaces in the reactor. Since the halogens can be easily (Figure S10) band structures for C2N3H-P63 cm,
removed at the interface, this result supports the feasibility of C2N3H-P63m, and C3N5H3; (Figure S11) comparison
the last step of the proposed procedure of g-C3N4 shown in of simulated XRD pattern of predicted C6N9H3·HCl with
Figure 9. Therefore, the removal of halogens would be the key that of the highly crystalline SMS sample; (Figure S12)
step in the synthesis of high-crystallinity g-C3N4 by SMS. band structure and UV/vis absorption spectrum of
C6N9H3·LiCl; (Table S1) structure details of heptazine-
4. CONCLUSION based g-C3N4 configurations; (Table S2) structure and
A series of high-throughput ab initio evolutionary structure ionization potentials of the three most stable heptazine-
searches were carried out to predict the structures of g-C3N4 based g-C3N4 configurations; typical USPEX input file
and related compounds corresponding to two different for the prediction of triazine-based g-C3N4; crystallo-
synthesis methods: thermal polycondensation in an oxidizing graphic Information Files of predicted structures (PDF)

■
environment and SMS in an inert environment.
For the synthesis via thermal polycondensation, the AUTHOR INFORMATION
structures for the most stable heptazine-based g-C3N4 and Corresponding Authors
the robust metastable phases, phases 1, 2, and 3, were
*E-mail: wang.junjie0810@gmail.com
predicted. Our search results revealed that unlike other planar
*E-mail: umezawa.naoto@gmail.com
layered structures, the stable configurations of g-C3N4 are
distorted. On the basis of the results of structure prediction and ORCID
transition state searches, we demonstrated that the phase Junjie Wang: 0000-0002-6428-2233
transition from phase 2 to phase 1 is a new mechanism to Jinhua Ye: 0000-0002-8105-8903
explain the temperature dependence of the band gap of g-C3N4. Naoto Umezawa: 0000-0001-9572-9790
A complete and renewed understanding of the relationship Notes
between the electronic structure and the crystal configurations The authors declare no competing financial interest.

■
of g-C3N4 has been presented based on the newly discovered
structures of g-C3N4. A model was developed to explain the
ACKNOWLEDGMENTS
influence of the distortion of the heptazine unit on the
electronic structure of g-C3N4. J.W. is an International Research Fellow of the Japan Society
For the SMS process, we predicted a series of structures with for the Promotion of Science (JSPS). We acknowledge the
different compositions of C3N4, C3N5H, C2N3H, C6N9H3·HCl, financial support from JSPS through project P14207. This work
C6N9H3·Li3Cl, and C6N9H3·LiCl through extensive high- is partly supported by the World Premier International
throughput ab initio evolutionary structure searches. We Research Center Initiative on Materials Nanoarchitectonics
clarified that the controversial structure of the previously (MANA), MEXT, and by the Core Research for Evolutional
reported high-crystallinity SMS phase was C6N9H3·Li3Cl Science and Technology (CREST) program and Materials
instead of g-C3N4. Furthermore, we revealed that C6N9H3· research by Information Integration Initiative (MI2I) project of
LiCl is a more stable high-crystallinity phase than C6N9H3· the Japan Science and Technology Agency (JST). J.W. thanks
Li3Cl. By combining the theoretical structure search with Dr. Nguyen Thanh Cuong of NIMS, Prof. Tomofumi Tada and
improved SMS experiments, the structure of C6N9H3·LiCl has Prof. Hirofumi Akamatsu of Titech and Prof. Daniel
been validated for the first time. On the basis of the predicted Fredrickson of UW-Madison for useful discussions and
stable structures, a very promising reaction pathway to suggestions. N.U. thanks Dr. Christian Joachim and Dr.
synthesize highly crystalline g-C3N4 from C6N9H3·LiCl has Masakazu Aono of NIMS for their useful advice.
been proposed.

■ ASSOCIATED CONTENT
■ REFERENCES
(1) Wang, X. C.; Maeda, K.; Thomas, A.; Takanabe, K.; Xin, G.;
*
S Supporting Information Carlsson, J. M.; Domen, K.; Antonietti, M. A Metal-Free Polymeric
Photocatalyst for Hydrogen Production from Water under Visible
The Supporting Information is available free of charge on the Light. Nat. Mater. 2009, 8, 76−80.
ACS Publications website at DOI: 10.1021/acs.chemma- (2) Wang, Y.; Wang, X.; Antonietti, M. Polymeric Graphitic Carbon
ter.6b02969. Nitride as a Heterogeneous Organocatalyst: From Photochemistry to
Details of the computational and experimental methods; Multipurpose Catalysis to Sustainable Chemistry. Angew. Chem., Int.
(Figure S1) test results of different exchange-correlation Ed. 2012, 51, 68−89.
functionals; (Figure S2) computed phonon dispersions (3) Ong, W.-J.; Tan, L.-L.; Ng, Y. H.; Yong, S.-T.; Chai, S.-P.
Graphitic Carbon Nitride (g-C3N4)-Based Photocatalysts for Artificial
of heptazine-based Phase 1 and Phase 2 and triazine- Photosynthesis and Environmental Remediation: Are We a Step
based g-C3N4; (Figure S3) computed Gibbs free energy Closer To Achieving Sustainability? Chem. Rev. 2016, 116 (12), 7159−
and equations of states of heptazine-based Phase 1 and 7329.
Phase 2 and triazine-based g-C3 N4; (Figure S4) (4) Liu, J.; Liu, Y.; Liu, N.; Han, Y.; Zhang, X.; Huang, H.; Lifshitz,
calculation results for the phase transition between Y.; Lee, S.-T.; Zhong, J.; Kang, Z. Metal-Free Efficient Photocatalyst

2705 DOI: 10.1021/acs.chemmater.6b02969

Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

for Stable Visible Water Splitting via a Two-electron Pathway. Science (23) Dong, H.; Oganov, A. R.; Zhu, Q.; Qian, G.-R. The Phase
2015, 347, 970−974. Diagram and Hardness of Carbon Nitrides. Sci. Rep. 2015, 5, 9870.
(5) Liu, G.; Wang, T.; Zhang, H.; Meng, X.; Hao, D.; Chang, K.; Li, (24) Wirnhier, E.; Döblinger, M.; Gunzelmann, D.; Senker, J.;
P.; Kako, T.; Ye, J. Nature-Inspired Environmental “Phosphorylation” Lotsch, B. V.; Schnick, W. Poly(triazine imide) with Intercalation of
Boosts Photocatalytic H2 Production over Carbon Nitride Nanosheets Lithium and Chloride Ions [(C3N3)2(NHxLi1‑x)3 ·LiCl]: A Crystalline
under Visible-Light Irradiation. Angew. Chem., Int. Ed. 2015, 54, 2D Carbon Nitride Network. Chem. - Eur. J. 2011, 17, 3213−3221.
13561−13565. (25) Algara-Siller, G.; Severin, N.; Chong, S. Y.; Björkman, T.;
(6) Bhunia, M. K.; Yamauchi, K.; Takanabe, K. Harvesting Solar Palgrave, R. G.; Laybourn, A.; Antonietti, M.; Khimyak, Y. Z.;
Light with Crystalline Carbon Nitrides for Efficient Photocatalytic Krasheninnikov, A. V.; Rabe, J. P.; Kaiser, U.; Cooper, A. I.; Thomas,
Hydrogen Evolution. Angew. Chem., Int. Ed. 2014, 53, 11001−11005. A.; Bojdys, M. J. Triazine-Based Graphitic Carbon Nitride: a Two-
(7) Zhang, Y.; Schnepp, Z.; Cao, J.; Ouyang, S.; Li, Y.; Ye, J.; Liu, S. Dimensional Semiconductor. Angew. Chem., Int. Ed. 2014, 53, 7450−
Biopolymer-Activated Graphitic Carbon Nitride towards a Sustainable 7455.
Photocathode Material. Sci. Rep. 2013, 3, 2163. (26) Wang, J.; Shen, Y.; Li, Y.; Liu, S.; Zhang, Y. Crystallinity
(8) Merschjann, C.; Tyborski, T.; Orthmann, S.; Yang, F.; Modulation of Layered Carbon Nitride for Enhanced Photocatalytic
Schwarzburg, K.; Lublow, M.; Lux-Steiner, M.-Ch.; Schedel-Niedrig, Activities. Chem. - Eur. J. 2016, 22, 12449−12454.
Th. Photophysics of Polymeric Carbon Nitride: An Optical (27) Oganov, A. R.; Glass, C. W. Crystal Structure Prediction Using
Quasimonomer. Phys. Rev. B: Condens. Matter Mater. Phys. 2013, 87, ab initio Evolutionary Techniques: Principles and Applications. J.
205204. Chem. Phys. 2006, 124, 244704.
(9) Zhang, Y.; Mori, T.; Niu, L.; Ye, J. Non-Covalent Doping of (28) Glass, C. W.; Oganov, A. R.; Hansen, N. USPEXEvolutionary
Graphitic Carbon Nitride Polymer with Graphene: Controlled Crystal Structure Prediction. Comput. Phys. Commun. 2006, 175, 713−
Electronic Structure and Enhanced Optoelectronic Conversion. Energy 720.
Environ. Sci. 2011, 4, 4517−4521. (29) Zhu, Q.; Li, L.; Oganov, A. R.; Allen, P. B. Evolutionary Method
(10) Du, A. J.; Sanvito, S.; Smith, S. C. First-Principles Prediction of for Predicting Surface Reconstructions with Variable Stoichiometry.
Metal-Free Magnetism and Intrinsic Half-mMetallicity in Graphitic Phys. Rev. B: Condens. Matter Mater. Phys. 2013, 87, 195317.
Carbon Nitride. Phys. Rev. Lett. 2012, 108, 197207. (30) Kresse, G.; Furthmüller, J. Efficient Iterative Schemes for ab
(11) Du, A. J.; Sanvito, S.; Li, Z.; Wang, D.; Jiao, Y.; Liao, T.; Sun, Q.; initio Total-Energy Calculations Using a Plane-Wave Basis Set. Phys.
Ng, Y. H.; Zhu, Z.; Amal, R.; Smith, S. C. Hybrid Graphene and Rev. B: Condens. Matter Mater. Phys. 1996, 54, 11169.
Graphitic Carbon Nitride Nanocomposite: Gap Opening, Electron− (31) Kresse, G.; Furthmüller, J. Efficiency of ab-initio Total Energy
Hole Puddle, Interfacial Charge Transfer, and Enhanced Visible Light Calculations for Metals and Semiconductors Using a Plane-Wave Basis
Response. J. Am. Chem. Soc. 2012, 134, 4393−4397. Set. Comput. Mater. Sci. 1996, 6, 15−50.
(32) Perdew, J. P.; Ernzerhof, M.; Burke, K. Rationale for Mixing
(12) Wei, W.; Jacob, T. Strong Excitonic Effects in the Optical
Exact Exchange with Density Functional Approximations. J. Chem.
Properties of Graphitic Carbon Nitride g-C3N4 from First Principles.
Phys. 1996, 105, 9982.
Phys. Rev. B: Condens. Matter Mater. Phys. 2013, 87, 085202.
(33) Wang, J.; Umezawa, N.; Hosono, H. Mixed Valence Tin Oxides
(13) Zhou, Z.; Shen, Y.; Li, Y.; Liu, A.; Liu, S.; Zhang, Y. Chemical
as Novel van der Waals Materials: Theoretical Predictions and
Cleavage of Layered Carbon Nitride with Enhanced Photoluminescent
Potential Applications. Adv. Energy Mater. 2016, 6, 1501190.
Performances and Photoconduction. ACS Nano 2015, 9, 12480−
(34) Allen, J. P.; Scanlon, D. O.; Piper, L.; Watson, G. W.
12487. Understanding the Defect Chemistry of Tin Monoxide. J. Mater. Chem.
(14) Zhou, Z.; Shang, Q.; Shen, Y.; Zhang, L.; Zhang, Y.; Lv, Y.; Li,
C 2013, 1, 8194−8208.
Y.; Liu, S.; Zhang, Y. Chemically Modulated Carbon Nitride (35) Govaerts, K.; Saniz, R.; Partoens, B.; Lamoen, D. van der Waals
Nanosheets for Highly Selective Electrochemiluminescent Detection Bonding and the Quasiparticle Band Structure of SnO From First
of Multiple Metal-ions. Anal. Chem. 2016, 88, 6004−6010. Principles. Phys. Rev. B: Condens. Matter Mater. Phys. 2013, 87, 235210.
(15) Liu, A. Y.; Wentzcovitch, R. M. Stability of Carbon Nitride (36) Tkatchenko, A.; Scheffler, M. Accurate Molecular Van Der
Solids. Phys. Rev. B: Condens. Matter Mater. Phys. 1994, 50, 10362. Waals Interactions from Ground-State Electron Density and Free-
(16) Ortega, J.; Sankey, O. F. Relative Stability of Hexagonal and Atom Reference Data. Phys. Rev. Lett. 2009, 102, 073005.
Planar Structures of Hypothetical C3N4 Solids. Phys. Rev. B: Condens. (37) Ruiz, V. G.; Liu, W.; Zojer, E.; Scheffler, M.; Tkatchenko, A.
Matter Mater. Phys. 1995, 51, 2624. Density-Functional Theory with Screened van der Waals Interactions
(17) Fanchini, G.; Tagliaferro, A.; Conway, N.; Godet, C. Role of for the Modeling of Hybrid Inorganic-Organic Systems. Phys. Rev. Lett.
Lone-Pair Interactions and Local Disorder in Determining the 2012, 108, 146103.
Interdependency of Optical Constants of a−CN:H Thin Films. Phys. (38) Tkatchenko, A.; Di Stasio, R. A.; Car, R.; Scheffler, M. Accurate
Rev. B: Condens. Matter Mater. Phys. 2002, 66, 195415. and Efficient Method for Many-Body van der Waals Interactions. Phys.
(18) Tyborski, T.; Merschjann, C.; Orthmann, S.; Yang, F.; Lux- Rev. Lett. 2012, 108, 236402.
Steiner, M-Ch.; Schedel-Niedrig, Th. Crystal Structure of Polymeric (39) Gobre, V. V.; Tkatchenko, A. Scaling Laws for van der Waals
Carbon Nitride and the Determination of its Process-Temperature- Interactions in Nanostructured Materials. Nat. Commun. 2013, 4,
Induced Modifications. J. Phys.: Condens. Matter 2013, 25, 395402. 2341.
(19) Bojdys, M. J.; Müller, J.-O.; Antonietti, M.; Thomas, A. (40) Gao, W.; Tkatchenko, A. Electronic Structure and van der Waals
Ionothermal Synthesis of Crystalline, Condensed, Graphitic Carbon Interactions in the Stability and Mobility of Point Defects in
Nitride. Chem. - Eur. J. 2008, 14, 8177−8182. Semiconductors. Phys. Rev. Lett. 2013, 111, 045501.
(20) Thomas, A.; Fischer, A.; Goettmann, F.; Antonietti, M.; Müller, (41) Heyd, J.; Scuseria, G. E.; Ernzerhof, M. Hybrid Functionals
J.-O.; Schlögl, R.; Carlsson, J. M. Graphitic Carbon Nitride Materials: based on a Screened Coulomb Potential. J. Chem. Phys. 2003, 118,
Variation of Structure and Morphology and Their Use as Metal-Free 8207.
Catalysts. J. Mater. Chem. 2008, 18, 4893−4908. (42) Krukau, A. V.; Vydrov, O. A.; Izmaylov, A. F.; Scuseria, G. E.
(21) Kroke, E.; Schwarz, M.; Horath-Bordon, E.; Kroll, P.; Noll, B.; Influence of the Exchange Screening Parameter on the Performance of
Norman, A. D. Tri-s-Triazine Derivatives. Part I. From Trichloro-Tri- Screened Hybrid Functionals. J. Chem. Phys. 2006, 125, 224106.
s-Triazine to Graphitic C3N4 Structures. New J. Chem. 2002, 26, 508− (43) Hoffmann, R. Solids and Surfaces; Wiley-VCH: New York, 1988.
512. (44) Dronskowski, R.; Bloechl, P. E. Crystal Orbital Hamilton
(22) Fina, F.; Callear, S. K.; Carins, G. M.; Irvine, J. Structural Populations (COHP): Energy-Resolved Visualization of Chemical
Investigation of Graphitic Carbon Nitride via XRD and Neutron Bonding in Solids Based on Density-Functional Calculations. J. Phys.
Diffraction. Chem. Mater. 2015, 27, 2612−2618. Chem. 1993, 97, 8617−8624.

2706 DOI: 10.1021/acs.chemmater.6b02969

Chem. Mater. 2017, 29, 2694−2707
Chemistry of Materials Article

(45) Deringer, V. L.; Tchougréeff, A. L.; Dronskowski, R. Crystal

Orbital Hamilton Population (COHP) Analysis As Projected from
Plane-Wave Basis Sets. J. Phys. Chem. A 2011, 115, 5461−5466.
(46) Maintz, S.; Deringer, V. L.; Tchougréeff, A. L.; Dronskowski, R.
Analytic Projection from Plane-Wave and PAW Wavefunctions and
Application to Chemical-Bonding Analysis in Solids. J. Comput. Chem.
2013, 34, 2557−2567.
(47) Govind, N.; Petersen, M.; Fitzgerald, G.; King-Smith, D.;
Andzelm, J. A Generalized Synchronous Transit Method for
Transition State Location. Comput. Mater. Sci. 2003, 28, 250−258.
(48) Clark, S. J.; Segall, M. D.; Pickard, C. J.; Hasnip, P. J.; Probert,
M. I. J.; Refson, K.; Payne, M. C. First Principles Methods using
CASTEP. Z. Kristallogr. - Cryst. Mater. 2005, 220, 567−570.
(49) Zhou, Z.; Wang, J.; Yu, J.; Shen, Y.; Li, Y.; Liu, A.; Liu, S.;
Zhang, Y. Dissolution and Liquid Crystals Phase of 2D Polymeric
Carbon Nitride. J. Am. Chem. Soc. 2015, 137, 2179−2182.
(50) Vineyard, G. H. Frequencey Factors and Isotope Effects in Solid
State Rate Processes. J. Phys. Chem. Solids 1957, 3, 121−127.
(51) Janotti, A.; Van de Walle, C. G. Native Point Defects in ZnO.
Phys. Rev. B: Condens. Matter Mater. Phys. 2007, 76, 165202.
(52) Sholl, D. S.; Steckel, J. A. Density Functional Theory: A Practical
Introduction; Wiley-Interscience, 2009.
(53) Levitin, V .V. Atom Vibrations in Solids: Amplitudes and
Frequencies; Cambridge Scientific Publishers, 2004.
(54) Han, S. W.; Kwon, H.; Kim, S. K.; Ryu, S.; Yun, W. S.; Kim, D.
H.; Hwang, J. H.; Kang, J.-S.; Baik, J.; Shin, H. J.; Hong, S. C. Band-
Gap Transition Induced by Interlayer van der Waals Interaction in
MoS2. Phys. Rev. B: Condens. Matter Mater. Phys. 2011, 84, 045409.
(55) Shahbaz, M.; Urano, S.; LeBreton, P. R.; Rossman, M. A.;
Hosmane, R. S.; Leonard, N. J. Tri-s-triazine: Synthesis, Chemical
Behavior, and Spectroscopic and Theoretical Probes of Valence
Orbital Structure. J. Am. Chem. Soc. 1984, 106, 2805−2811.
(56) Tadano, T.; Tsuneyuki, S. Self-consistent phonon calculations of
lattice dynamical properties in cubic SrTiO3 with first-principles
anharmonic force constants. Phys. Rev. B: Condens. Matter Mater. Phys.
2015, 92, 054301.
(57) Yu, L.; Kokenyesi, R. S.; Keszler, D. A.; Zunger, A. Inverse
Design of High Absorption Thin-Film Photovoltaic Materials. Adv.
Energy Mater. 2013, 3, 43−48.
(58) Jiang, J.; Zhao, Y.; Yaghi, O. M. Covalent Chemistry beyond
Molecules. J. Am. Chem. Soc. 2016, 138, 3255−3265.
(59) Liu, Y.; Ma, Y.; Zhao, Y.; Sun, X.; Gándara, F.; Furukawa, H.;
Liu, Z.; Zhu, H.; Zhu, C.; Suenaga, K.; Oleynikov, P.; Alshammari, A.
S.; Zhang, X.; Terasaki, O.; Yaghi, O. M. Weaving of Organic Threads
into a Crystalline Covalent Organic Framework. Science 2016, 351,
365−369.

2707 DOI: 10.1021/acs.chemmater.6b02969

Chem. Mater. 2017, 29, 2694−2707