Negative Poisson’s ratio in 1T-type crystalline two-dimensional transition metal dichalcogenides

Abstract

Materials with a negative Poisson’s ratio, also known as auxetic materials, exhibit unusual and counterintuitive mechanical behaviour—becoming fatter in cross-section when stretched. Such behaviour is mostly attributed to some special re-entrant or hinged geometric structures regardless of the chemical composition and electronic structure of a material. Here, using first-principles calculations, we report a class of auxetic single-layer two-dimensional materials, namely, the 1T-type monolayer crystals of groups 6–7 transition-metal dichalcogenides, MX₂ (M=Mo, W, Tc, Re; X=S, Se, Te). These materials have a crystal structure distinct from all other known auxetic materials. They exhibit an intrinsic in-plane negative Poisson’s ratio, which is dominated by electronic effects. We attribute the occurrence of such auxetic behaviour to the strong coupling between the chalcogen p orbitals and the intermetal t_2g-bonding orbitals within the basic triangular pyramid structure unit. The unusual auxetic behaviour in combination with other remarkable properties of monolayer two-dimensional materials could lead to novel multi-functionalities.

Introduction

The Poisson’s ratio of a material characterizes its response to uniaxial load and is given by ν_ab=−ɛ_b/ɛ_a, where ɛ_a is an applied strain in the a axis direction and ɛ_b is the resulting strain in a transverse b axis direction. Counter-intuitively, negative Poisson’s ratio (auxetic) materials¹ expand laterally when stretched and contract laterally when compressed. They can lead to enhanced mechanical properties, such as shear modulus², indentation resistance³ and fracture toughness⁴. The unusual auxetic effect itself and concomitant enhancements in other material properties offers enormous potential in many technologically important applications^5,6,7, such as biomedicine⁸, sensors⁹, fasteners¹⁰ and protective equipments¹¹.

Auxetic effect has been reported in a number of natural and man-made materials and structures in bulk form^5,6,12,13, for example, cubic metals^14,15, α-cristobalite (SiO₂)¹⁶, α-TeO₂ (ref. 17), the zeolite mineral natrolite¹⁸, honeycombs¹⁹, foams⁷, microporous polymers^20,21, composites^22,23, ceramics²⁴, molecular auxtics²⁵, metal-organic frameworks²⁶, bucklicrystals²⁷ and origami structures^28,29,30. Geometric considerations dominate the literature in understanding such auxetic effects and designing new auxetic materials. For most of these auxetic materials, the auxetic effect is explained by some special re-entrant structure or the crystal structure that can be viewed as being made up of rigid building blocks linked by flexible hinges^{1,19,31,32,33}, independent of their chemical composition and electronic structure.

Auxetic effect has also been recently reported in several monolayer two-dimensional (2D) materials. For example, the out-of-plane negative Poisson’s ratio was discovered in phosphorene^34,35, GeS³⁶ and monolayer arsenic³⁷. The in-plane negative Poisson’s ratio was also predicted in borophene³⁸ and three theoretically proposed but not-yet-synthesized materials (that is, the penta-graphene^39,40, hα-silica⁴¹ and Be₅C₂ (ref. 42)). Similar to that in the bulk auxetic materials, the auxetic behaviour in these 2D materials is also considered to originate mainly from the puckered or buckled crystal structure.

In this study, using quantum mechanical first-principles calculations (see Methods section), we report a class of auxetic single-layer 2D materials with an intrinsic in-plane negative Poisson’s ratio. They differ from other known auxetic materials not only in their crystal structure but also in the microscopic origin of auxetic behaviour. These materials are the 1T-type crystalline monolayers of groups 6–7 transition metal dichalcogenides, 1T-MX₂ (M=Mo, W, Tc, Re; X=S, Se, Te). In contrast to those known bulk or 2D auxetic materials, the in-plane auxetic behaviour discovered in groups 6–7 1T-MX₂ cannot be explained merely from their geometric structure because the non-auxetic behaviour is also found in other groups of MX₂ compounds with the same 1T-type structure. This dichotomy between auxetic and non-auxetic behaviour in the 1T-MX₂ compounds is explained by their distinct electron structures. The in-plane stiffness of those 1T-MX₂ materials is predicted to be order of 10² GPa, at least three orders of magnitude higher than man-made auxetic materials. The high in-plane stiffness and the auxetic behaviour in combination with other remarkable electronic and optoelectronic properties of the single-layer 2D materials⁴³ could lead to novel multi-functionalities, such as nanoscale auxetic electrodes and sensors.

Results

Crystal structure

The single layers of 2D transition metal dichalcogenides are formed by a hexagonally packed layer of metal (M) atoms sandwiched between two layers of chalcogen (X) atoms (Fig. 1). Each chalcogen atom forms the apex of a triangular pyramid that has three metal atoms at its base. The symmetry of the chalcogen array about each metal atom is either octahedral or trigonal prismatic. The former is often referred to as the 1T phase, whereas the latter as the 1H phase. Depending on the combination of the metal and chalcogen elements, one of the two phases is thermodynamically preferred. Most group-6 MX₂ compounds thermodynamically prefer the 1H phase⁴⁴, but the metastable 1T phase is also observed^45,46,47. For other groups of layered MX₂ compounds, most crystallize in the high-symmetry 1T or low-symmetry distorted 1T phase^44,48. The 1H-MX₂ compounds are known to be non-auxetic in the plane due to their hexagonal in-plane crystalline structure. We hence focus on 42 monolayer MX₂ compounds in the high-symmetry 1T phase (Table 1).

Figure 1: Structure of monolayer 1T-MX₂.

(a) Crystal structure. The basic X-M₁-M₂-M₃ triangular pyramid unit is marked. The rectangular outline displays the unit cell adopted in our calculation. It contains two MX₂ formula units. (b) Local structure of M-centred octahedron. The metal atoms form three one-dimensional chains in the directions of y=x, z=y and z=−x in the local reference frame. The M–M interaction is through the t_2g-orbital coupling. (c) Schematic configuration of DOS showing the gradual filling of d orbitals from group 4 (d⁰) to group 10 (d⁶) 1T-MX₂. The horizontal bars denote the corresponding Femi level of the system. t_2g and t_2g* correspond to the intermetal t_2g-bonding and t_2g-antibonding states, respectively. (d) Predicted M–X–M bond angles in the relaxed structure of strain-free 1T MX₂. Note, in the triangular pyramid as shown in Fig. 1a, ∠M₁XM₂=∠M₂XM₃=∠M₃XM₁=∠MXM.

Table 1 Predicted in-plane stiffness for 42 monolayer 1T-MX₂ compounds.

Poisson’s ratio results

Figure 2a shows our calculated Poisson’s ratio results (ν_ab) for 42 1T-MX₂ compounds in the b axis direction subjected to a 5% tensile strain applied along the a axis direction. Remarkably, we find that the sign of Poisson’s ratio strongly depends on the d-electron count. All 12 1T-MX₂ compounds from group 6 (d²) and group 7 (d³) exhibit negative Poisson’s ratios, ranging from −0.03 to −0.37. Seven of them (that is, TcTe₂, ReTe₂, WTe₂, WSe₂, MoSe₂, ReS₂ and TcS₂) have a Poisson’s ratio <−0.1, higher in magnitude than that of borophene (−0.04 along a and −0.02 along b)³⁸, rendering them more promising candidates for specific applications in mechanical nanodevices. For other groups of 1T-MX₂ compounds, we find positive Poisson’s ratios ranging from 0.09 to 0.53.

Figure 2: Poisson’s ratios.

(a) Poisson’s ratio, ν_ab=−ɛ_b/ɛ_a, calculated for a 5% strain applied along the a axis (that is, ɛ_a=5%). (b,c) Poisson’s ratios for ZrS₂ and MoS₂ as a function of strain applied along the a axis (b) and b axis (c).

Figure 2b,c shows our calculated Poisson’s ratios (ν_ab and ν_ba) as a function of applied strain in two example compounds, non-auxetic ZrS₂ and auxetic MoS₂. For both compounds, Poisson’s ratio varies slowly as applied strain goes from −5% to 5%, suggesting a dominant linear elastic behaviour within the strain range considered. (Note the Poisson’s ratio at a large strain (that is, >5% or <−5%) may strongly depend on the strain. This behaviour is not pursued in this work since such large strains are often experimentally inaccessible.) The small differences between ν_ab and ν_ba reflect a nearly isotropic auxetic or non-auxetic behaviour inside the 1T-structure plane. Therefore, the d-electron count dependence of the sign of Poisson’s ratio as shown in Fig. 2a does not change with respect to the amount of the applied strain within the linear elastic range and the loading direction inside the plane.

Stiffness

To compare the stiffness (Young’s modulus) of a single-layer material with bulk materials, we calculate its 3D in-plane stiffness (Y_3D) from 2D in-plane stiffness (Y_2D) and effective layer thickness (t) via Y_3D=Y_2D/t. The Y_2D is directly derived from first-principles total energies as a function of uniaxial strain. The effective layer thickness t can also be uniquely determined from first-principles calculated bending energy⁴⁹. Here for simplicity, we approximate t as t=t₀+0.8 Å, where t₀ is the distance between the top and bottom chalcogen atom layers and the 0.8 Å is the total effective decay length (0.4 Å in each layer side) of electron density into the vacuum. The 0.8 Å is derived from the first-principles calculated layer thickness for 1H-MoS₂ (ref. 49). MSe₂ and MTe₂ may have different decay lengths than MS₂. However, such difference should be less than one time of magnitude. Hence, using a different decay length does not induce one time of magnitude difference in the calculated 3D in-plane stiffness.

Table 1 shows that the 3D in-plane stiffness of almost all 1T-MX₂ compounds lies in between 100 and 300 GPa. Among the auxetic d²–d³ 1T-MX₂ compounds, WS₂ and ReSe₂ are the stiffest, having a stiffness of ∼290 GPa; TcTe₂ is the softest, having a stiffness of ∼80 GPa. Man-made auxetic materials typically have a stiffness in the range from ∼10⁻⁵ to ∼1 GPa, and naturally occurring auxetic bulk solids exhibit a stiffness of ∼10¹–10² Pa (ref. 50). Therefore, even considering the uncertainty of our calculated 3D stiffness (less than one order of magnitude) that may be caused by using different approximations for effective layer thickness, the 3D stiffness values predicted for 1T-MX₂ compounds are among the highest in the naturally occurring crystalline solids and are at least three orders of magnitude higher than man-made auxetic materials.

The fact that both auxetic and non-auxetic materials are found in the same 1T-structure type implies that the auxetic effect is not a purely geometric property. The d-electron count dependence of electronic structure must be involved. In the 1T structure, the d orbitals of the octahedrally coordinated transition metal split into two groups, d_xy,yz,zx (t_2g) and (e_g). In what follows, we shall show that (i) transition metals interact with each other through t_2g-orbital coupling, and (ii) the coupled t_2g orbitals are further coupled with the lone-pair electrons of chalcogen atoms. It is the gradual filling of such t_2g-p-hybridized bands that leads to the different behaviour of Poisson’s ratio.

Intermetal t_2g-orbital coupling

In the ideal 1T phase, the M-centred octahedra share edges, forming three one-dimensional M-chains along the directions of lines y=x, y=z, and z=−x, respectively, within the local reference frame of the octahedra (Fig. 1b). The metal atoms can interact with each other through the coupling between their t_2g orbitals. This coupling gives rise to t_2g-bonding states and t_2g-antibonding states, with no energy gap in between due to the weak coupling nature. The t_2g states are mostly located within the gap between the bonding and antibonding bands of the M–X bonds (Fig. 1c).

The progressive filling of these t_2g bands from group 4 (d⁰) to group 10 (d⁶) species leads to different M–M bonding or antibonding character at the Fermi level. In d¹-d³ 1T-MX₂, the Fermi level crosses the t_2g-bonding states; the highest occupied bands close to the Fermi-level thus exhibit a stronger bonding character as we go from d¹ to d³. This bonding character attracts the metal atoms towards each other, leading to an intermetal distance shorter than that in the ideal 1T structure. In d⁵–d⁶ 1T-MX₂, since the t_2g bonding states can accommodate up to six electrons (three from each metal), all t_2g-bonding states are filled and the Fermi level crosses the t_2g-antibonding states. Hence the highest occupied bands in the vicinity of the Fermi level exhibit antibonding character, repelling metal atoms from each other.

The existence of the intermetal t_2g-orbital interactions is reflected by the d-electron count dependence of the M–X–M bond angles (∠MXM) as illustrated in Fig. 1d. The ideal 1T phase has regular octahedra with ∠MXM=90°. In the d⁰ 1T-MX₂ compounds, the ∠MXM deviates least from 90°. This is expected since all t_2g states are almost completely unoccupied and the intermetal d–d interaction is marginal. For the d¹–d³ 1T-MX₂, all have acute ∠MXM, decreasing with the increasing d-electron count. This trend arises from the increasing intermetal t_2g-bonding character in going from d¹ to d³, which shortens the intermetal distance. In the d⁵–d⁶ 1T-MX₂, the ∠MXM jumps up to over 90°, consistent with the intermetal t_2g-antibonding character.

Figure 1d also shows that the chalcogen atoms have minor effect on ∠MXM compared with the transition metals with different d-electron counts, but a trend can still be observed: the ∠MXM decreases with increasing atomic number of the chalcogen. For example, the ∠MXM of TiS₂, TiSe₂ and TiTe₂ decreases from 89.6° to 88.2° to 86.0°. This trend is not associated with the intermetal t_2g-orbital interaction; instead it is intrinsic to the spatial distribution of the lone-pair charge density relative to that of the M–X bonds around the chalcogen.

t_2g–p-orbital coupling

The intermetal t_2g orbitals are further coupled with chalcogen p orbitals in 1T-MX₂. It can be seen from their projected density of states (DOS) as shown in Fig. 3. In the 4d transition metal disulfides with the ideal 1T structure, we find that the DOS of sulfur 3p and metal t_2g states overlap, as manifested by their similar DOS peak shapes and positions in energy. The t_2g–p-orbital overlap is marginal in d⁰ ZrS₂, but it increases quickly in going from d¹ NbS₂ to d⁶ PdS₂. This trend is clear not only in the energy range from −12 to −7 eV, where the major peaks of 3p DOS are located, but also near the Fermi level.

Figure 3: DOS of 4d MS₂ in the ideal 1T structure.

The t_2g-p orbital coupling manifests itself in the overlap of their DOS. The local reference frame in the octahedral is used for projecting DOS. The DOS shown in the figure are t_2g=d_xy+d_yz+d_zx, e_g=+ and p=p_x+p_y+p_z. The vertical dashed lines show the position of Fermi level. The energy is aligned to the vacuum level.

The t_2g–p-orbital interaction is attractive because the X ligand has one lone electron pair and acts as a sigma donor. In d¹–d³ MX₂, the t_2g–p coupling force draws atom X towards the intermetal bond centres, because the t_2g states are the intermetal bonding states spreading over the M–M bond centers. In d⁵–d⁶ MX₂, the t_2g–p coupling force attracts atoms M and X towards each other, because the t_2g states are antibonding and localized near the metal atoms. The d-electron count dependence of t_2g–p interaction direction plays a key role in determining the structure deformation presented below.

Deformation mechanism

To understand the microscopic origin of Poisson’s ratios, let us now look into the resulting structural relaxation subjected to a tensile strain applied along the a axis. Due to the centrosymmetric nature of the 1T phase, the whole relaxation process manifests itself in the triangular pyramid unit as illustrated in Fig. 4. For the stretch along the M₁–M₃ axis (that is, axis a), the resulting relaxation involves only atoms M₂ and X moving inside the Q–X–M₂ plane. Hence, two relations always hold during relaxation: $d_{{\text{M}}_1{\text{M}}_2}=d_{{\text{M}}_2{\text{M}}_3}$ and ∠M₁XM₂=∠M₃XM₂.

Figure 4: Deformation mechanism.

The solid and dashed M–X bonds indicate, respectively, the initial and final configurations at each relaxation step. The force is applied along the lattice-a direction. The red dashed arrows indicate the direction of the t_2g-p orbital interaction. The hollow blue arrows show the resulting movement of the X and M₂ atoms within the Q–X–M₂ plane.

We analyse the relaxation process by decomposing it into three consecutive steps: (i) atom X relaxes along the line Q–X, (ii) atom X rotates around the M₁–M₃ axis, and (iii) atom M₂ relaxes along the line Q–M₂. In the first two steps, the lattice constant b is fixed to the value found in the relaxed strain-free 1T structure. In the third step, b varies as atom M₂ moves along the Q–M₂ line, leading to different Poisson’s ratio behaviour.

Figure 4 shows the detailed structural relaxation in the three consecutive steps described above for 1T-MX₂ with ∠QXM₂<90° (Fig. 4a) and with ∠QXM₂>90° (Fig. 4b) separately. Each step can be understood in the way that atom X (or atoms X and M₂) relaxes to conserve the M–X bond length (d_MX) since d_MX is energetically dominant. After the first two steps of the relaxation, it can be seen that (i) both $d_{{\text{M}}_1{\text{M}}_2}$ (also $d_{{\text{M}}_2{\text{M}}_3}$) and ∠M₁XM₂ and ∠M₃XM₂ (Supplementary Fig. 1) increase in all 1T-MX₂ compounds no matter whether ∠QXM₂ is larger or smaller than 90°, and (ii) ∠XQM₂ increases in the 1T-MX₂ with ∠QXM₂<90° but decreases in the 1T-MX₂ with ∠QXM₂ >90° (Supplementary Fig. 1). The changes in $d_{{\text{M}}_1{\text{M}}_2}$ and $d_{{\text{M}}_2{\text{M}}_3}$, ∠M₁XM₂ and ∠M₂XM₃ and ∠XQM₂ thus store the strain energy, which will be partially released in the subsequent third step relaxation.

The third-step relaxation determines the sign of Poisson’s ratio. The negative Poisson’s ratio of d²–d³ MX₂ can be attributed to the strong t_2g–p-orbital coupling. Such strong coupling implies a large amount of strain energy stored in the decreased ∠XQM₂ after the second step. This part of strain energy will be released in this third step through atom M₂ relaxing along the increased b-lattice direction, leading to a negative Poisson’s ratio. The strength of t_2g–p-orbital coupling depends not only on the d-electron count of the transition metal but also on the chalcogen atom. This dependence explains why the Poisson’s ratio of the compounds from same d² or d³ group also differs from one another as shown in Fig. 2a.

For d⁰–d¹ MX₂, the positive Poisson’s ratio results from the marginal or weak intermetal t_2g coupling and t_2g–p coupling. Such weak couplings imply that the strain energy stored in $d_{{\text{M}}_1{\text{M}}_2}$ and $d_{{\text{M}}_2{\text{M}}_3}$ and ∠XQM₂ is also marginal or small. The major strain energy that can be released in the third step is thus stored in the increased ∠M₁XM₂ and ∠M₂XM₃. Therefore, it is energetically favourable that atom M₂ relaxes in the b-decreasing direction, reducing the increase in ∠M₁XM₂ and ∠M₂XM₃, and resulting in a positive Poisson’s ratio. For d⁵–d⁶ MX₂, the positive Poisson’s ratio originates from the fact that the t_2g–p coupling aligns along the M–X bond and does not energetically affect the change in ∠XQM₂. In other words, the strain energy stored in the decreased ∠XQM₂ is small. Since the t_2g antibonding is also generally weak, the relaxation of atom M₂ is energetically favourable in the b-decreasing direction, giving rise to a positive Poisson’s ratio. This deformation mechanism is similar to that in d⁰–d¹ compounds.

Simply saying, the negative Poisson’s ratio in d²–d³ MX₂ results from the strong attractive coupling between the intermetal t_2g-bonding states and the X p states, which prevents atoms X and M₂ relaxing toward the ∠XQM₂-increasing direction. The positive Poisson’s ratio arises from lack of such strong t_2g–p coupling in other groups of 1T-MX₂.

Discussion

The monolayer MX₂ materials involve transition metals where strong correlation effects may be not well captured by the new strongly constrained and appropriately normed (SCAN) meta-GGA functional. To check the robustness of our results, we also calculated the Poisson’s ratio for 12 d²–d³ MX₂ by using the HSE06 hybrid functional⁵¹. The results are summarized in Supplementary Table 1. It shows that the Poisson’s ratio of eight 1T-MX₂ compounds (that is, MoSe₂, MoTe₂, WSe₂, WTe₂, TcTe₂, ReS₂, ReSe₂, ReTe₂) remains negative, whereas for other four compounds (that is, MoS₂, WS₂, TcS₂, TcSe₂) their Poisson’s ratio changes the sign from negative to slightly positive, which is still very interesting and useful for applications. Although it is found that the SCAN lattice constants agree better with experiment than the HSE06 ones for most of the compounds listed in this table, it is uncertain whether SCAN predicts a more accurate Poisson’s ratio than HSE06 since the semilocal SCAN functional could make larger density-driven error in the energy than HSE06 does for the system under stretching⁵². This uncertainty calls for experimental validation and further theoretical study. Nevertheless, the auxetic behaviour we find is robust in most of the d²–d³ MX₂ compounds. The less negative Poisson’s ratio predicted by HSE06 (Supplementary Table 1) further indicates that the auxetic behaviour originates from the strong p–d coupling. In general, compared with the semilocal SCAN functional, HSE06 yields more localized metal d and chalcogen p orbitals and hence the weaker hybridization between them, which leads to less negative Poisson’s ratios in HSE06.

Our predicted in-plane auxetic behaviour is intrinsic in the 1T structure without any external engineering and occurs in the elastic region. This is different from the extrinsic auxetic behaviour reported in the epitaxial oxide thin-film^53,54 and the engineered 2D materials, such as the wrinkled graphene⁵⁵, graphane⁵⁶ and borophane⁵⁷. Recently, the negative Poisson’s ratio was also reported in metal nanoplates⁵⁸, pristine graphene⁵⁹ and semi-fluorinated graphene⁶⁰. The negative Poisson’s ratio claimed there corresponds to the ratio calculated from ν_ab=−∂ɛ_a/∂ɛ_b under large stains, differing from that calculated from ν_ab=−ɛ_a/ɛ_b (the original definition of Poisson’s ratio) as we followed here. The Poisson’s ratios calculated from ν_ab=−ɛ_a/ɛ_b for pristine graphene⁵⁹ and semi-fluorinated graphene⁶⁰ are actually both positive under a strain <∼15%, and for metal nanoplates, it is also positive under a strain <∼4%.

Finally, it is noteworthy that the auxetic behaviour of d²–d³ MX₂ compounds is predicted in the high-symmetry 1T phase. This phase is known to be metastable or dynamically unstable in both d² and d³ MX₂ compounds^44,61,62,63. However, experimentally, relevant phase diagrams of monolayer materials differ from those of bulk materials. The kinetic barriers between the different phases of monolayers may arise and be affected by many external factors, such as interfaces, underlying substrate, temperature, strain and impurities. Therefore, it is not uncommon to observe the undistorted 1T phase synthesized experimentally. For instance, although no kinetic barrier is found from first-principles calculations between the unstable 1T phase and dynamically stable distorted 1T phase, the undistorted 1T monolayer structures of MoS₂, MoSe₂, WS₂ and WSe₂ are observed from the exfoliation using Li-intercalation method^45,64. For MoS₂, the coexistence of 1T and 1H domains is also observed in the same monolayer^46,47. Such heterogeneous monolayers with auxetic and non-auxetic domains are particularly intriguing since they could lead to novel functionality.

Methods

All calculations were performed using density functional theory and the plane-wave projector augmented-wave⁶⁵ method as implemented in the VASP code⁶⁶. The new SCAN meta-generalized gradient approximation was used^67,68. SCAN is almost as computationally efficient as PBE-GGA functional, yet it often matches or exceeds the accuracy of the more computationally expensive hybrid functionals in predicting the geometries and energies of diversely bonded systems⁶⁸. Supplementary Table 2 shows our calculated lattice constants for 1T-MX₂ compounds. They agree very well with available experimental data⁴⁴, especially for groups 4–7 1T-MX₂ whose errors are within 1%. An energy cutoff of 500 eV was used. The monolayer structure is modelled in an orthorhombic supercell that contains two formula units (Fig. 1a) and a 20 Å vacuum space inserted in the out-of-plane direction. A 24 × 14 × 1 k-point grid was used to sample the Brillouin zone during structure relaxation. All atoms were fully relaxed until their atomic forces were <0.005 eV Å⁻¹. The effects of spin-orbit coupling on the structural deformation are considered to be minor and hence not included in our study.

The Poisson’s ratio is calculated from the engineering strain (ɛ), which is defined as the change in length ΔL per unit of the original length L, that is, ɛ=ΔL/L. The applied uniaxial strain is realized in our calculations by fixing the lattice parameter to a value different from its equilibrium value during structural relaxation. The resulting strain in the transverse direction is extracted from the fully relaxed structure subjected to an applied strain.

Data availability

The authors declare that the data supporting the findings of this study are available within the paper and its Supplementary Information files.