Acta Chim. Slov. 2004, 51, 589-609. 589 Scientific paper TOPOLOGICAL INDICES DERIVED FROM THE G(a,b,c) MATRK, USEFUL AS PHYSICOCHEMICAL PROPERTY INDICES Anton Perdih,* Branislav Perdih Mala vas 12, SI-1000 Ljubljana, Slovenia Received 05-07-2004 Abstract From the G(a,b,c) matrix at least 20 groups of topological indices can be derived. Each of these groups contains an infinite number of possible indices. Some of these indices correlate by abs(r)max > 0.99, and a number of them by abs(r)max > 0.9 with particular tested 29 resp. 31 physicochemical properties of alkanes. Therefore, their use in QSPR/QSAR models is recommended. The indices of these groups are not useful to predict the physicochemical properties: de, Zc, A, Cohesive Energy Density, and Solubility Parameter. Key words: alkane, index, matrix, QSAR, QSPR, topological Introduction Mathematical topological methods occupy an eminent plače in the field of predietion of properties and activities of chemical compounds, and even materials. These methods, known under the acronym QSPR/QSAR (quantitative-structure-property or structure-activity relationship) are normally, but not always, based on graph-theoretical deseriptors, where molecules are seen as chemical graphs, i.e. as a set of vertices attached to each other by a set of non-metrical conneetions.1 These deseriptors are known also as topological indices. They are the simplest means of deseribing the strueture of a molecule, characterizing it by a simple number.2 A huge number of topological indices are known3'4 but in spite of that, interest in topological indices has grown remarkably during recent years. A substantial part of topological indices is derived from one or another matrix associated with molecular strueture. Estrada5 developed a matrix that enables the derivation of an infinite number of indices. We6 presented a type of matrices, i.e. the generalized vertex-degree vertex-distance matrices, that enable the derivation of an infinite number of indices, too, and we have shown that these matrices represent a step in unification of several matrices which have been used to derive topological indices, i.e. of the adjacency matrix, the distance matrix, the reciprocal distance matrix, etc. The A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... 590 Acta Chim. Slov. 2004, 51, 589-609. characteristics of some groups of indices derived by means of the generalized vertex-degree vertex-distance matrices have been studied, namely: The summation derived W(m,n) indices,6 which besides the Wiener index7 W also include some Ivanciuc indices.8 The summation derived "mean degree of vertices" indices,9 which include the Randič index %l° The summation derived Vij(m,n) indices,11 which besides the Wiener index7 and several Ivanciuc indices8'12"16 also include the Randič index %W The susceptibilities for branching of the W(m,n) indices,17 and the difference derived indices.18 The largest eigenvalues of the matrices from which the Vij(m,n) indices11 have been derived. One index of this group has a long tradition, namely h.19'20 It is also a member of the indices21"23 which are the largest eigenvalues of the matrices from which the W(m,n) indices6 or the "mean degree of vertices" indices9 are derived. The multiplication derived indices, which include the Gutman index k24 They are derived by multiplication of the non-diagonal elements of the matrix.25 The aim of the present study is to give a survey of the large number of indices groups that can be derived from the above mentioned generalized vertex-degree vertex-distance matrices and especially to indicate where in these groups of indices, consisting of infinite numbers of them, are to be expected the best indices, useful for the prediction of properties or activities of chemical compounds or materials. These indices could then be used in methods known under the acronym QSPR/QSAR (quantitative-structure-property or structure-activity relationship). The aim of the present study is not to look for the best QSPR/QSAR models but to indicate, using a set of physicochemical properties of alkanes, how and where the best indices useful for that can be looked for. Data and notations Physicochemical properties (PP) The data for the boiling point (BP), density (d), the critical data Te, Pc, Ve, Zc, oce, and de, the standard enthalpy of formation for the ideal gas (AHf°g), the enthalpy of vaporisation (AHv), the Antoine constants A, B, and C, as well as the Pitzer's acentric factor (co) and the refraetive index (nD) were taken from the CRC Handbook26 or from Lange's Handbook27. The data for the liquid molar volume (Vm), the intrinsic molar volume (Vi), the intermolecular volume (V"), the ratios Tc2/Pc and Tc/Pc used instead of the van der Waals parameters a0 and b0, the ratio BP/Tc (reduced BP), the molar A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... Acta Chim. Slov. 2004, 51, 589-609. 591 refraction (MR), cohesive energy density (CED) and its square root, the solubility parameter (Sol. par.) were calculated from data presented in the handbooks. The data for Octane Numbers (BON, MON, RON) were taken from: Pogliani,28 Balaban and Motoc,29 Gutman et al.,24 and Morley;30 those for vapour pressure (logVP) from Goli and Jurs,31 and those for the entropy (S) and quadratic mean radius (R2) from Ren.32 Surface tension (ST) data were taken from Li.33 Notations The structures of alkanes are presented in shorthand, e.g. Hp is n-heptane, Oct is n-octane, 223M5 is 2,2,3-trimethylpentane, 3E2M5 is 3-ethyl-2-methylpentane, etc, cf. also Fig. 1. Matrix and indices Let there be the matrix G(a,b,c), composed of elements g = v!axv7bxd/, where v, and v,- is the degree of vertex i and j, respectively (in alkanes it is the number of C-C bonds the carbon in question is involved in) and d!y is the shortest distance from vertex i to vertex j (in alkanes it is the smallest number of bonds between the carbons in question). 1 1 2,3-dimethyl pentane: 1—2—3^1—5 =23M5 v, distance matrix G(a ,b,c) matrix 1 0 1 2 3 4 2 3 0 3b 3b 2c 2b 3c 4C T 3C 3 1 0 1 2 3 1 2 3a 0 3a3b 3a2b T 3a 3c 3a 3a 2c 3 2 1 0 1 2 2 1 3a 2c 3a3b 0 3a2b 3a 2c 3a 2c 3a 2 3 2 1 0 1 3 2 2a 3c 2a3b 2c 2a3b 0 2a 2a 3c 2a 2c 1 4 3 2 1 0 4 3 4C 3b 3c 3b 2c 2b 0 4C 3C 1 2 1 2 3 4 0 3 T 3b gb 2c 2b 3c 4C 0 3C 1 3 2 1 2 3 3 0 3C 3b 2c 3b 2b 2c 3C 3C 0 vj 1 3 3 2 1 1 1 Figure 1. 2,3-dimethyl pentane: its formula, its label, its distance matrix composed of distances d9, its vectors of degrees of vertices, v, and v,, as well as the G(a,b,c) matrix derived from them. v, - vector of degrees of starting vertex, v, v, - vector of degrees of target vertex, v, normal script: matrix elements, common to both halves of the matrix bold: matrix elements, not common to both halves of the matrix A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... 592 Acta Chim. Slov. 2004, 51, 589-609. Vertex i is the starting vertex whereas vertexy is the target vertex; a, b, and c are the exponents. For the purpose of this study, the value of diagonal elements in the matrix G(a,b,c) is set to g„ = 0, since d« = 0 and the contribution of hydrogen atoms is set to zero, whereas the contribution of ali other atoms is set to one, i.e. only the contribution of the structure of the molecule in question to the value of the matrix and its elements is considered. As an example, the G(a,b,c) matrix of the 2,3-dimethyl pentane structure ispresented in Figure 1. To derive the indices from the G(a,b,c) matrix, one or both halves of the matrix or only some parts of them can be used. Indices derived by summation or multiplication of non-diagonal elements of the G(a,b,c) matrix are considered here. The definitions of indices studied here are presented in Table 1. Table 1. Definitions of indices studied in the present paper. VL(a,b,c) = Y(V x vb x d„c)left V^b,c) = I(v;a x vb x d/)nght V(a,b,c) = VL(a,b,c) + VD(a,b,c) AV(a,b,c) = VL(a,b,c) - VD(a,b,c) PV(a,b,c) = VL(a,b,c) x VD(a,b,c) QV(a,b,c) = VL(a,b,c) / VD(a,b,c) VC(a,b,c) = Z(V!a x Vyb x d c elements, common to both halves of the matrix) VNL(a,b,c) = Z(V!a x v.b x d/ elements, found only on the left side of the main diagonal of the matrix) VND(a,b,c) = Z(V!a x v.b x d/ elements, found only on the right side of the main diagonal of the matrix) VN(a,b,c) = VNL(a,b,c) + VND(a,b,c) PVN(a,b,c) = VNL(a,b,c) x VND(a,b,c) QVN(a,b,c) = VNL(a,b,c) / VND(a,b,c) PL(a,b,C) = n(v,a >< V,b X dAeft PD(a,b,c) = n(v;a x vb x d;c)nght M(a,b,c) = PL(a,b,c) x pD(a,b,c) QP(a,b,c) = PL(a,b,c) / PD(a,b,c) VP(a,b,c) = PL(a,b,c) + PD(a,b,c) AP(a,b,c) = PL(a,b,c) - PD(a,b,c) PC(a,b,c) = rKv/a x v,b x dc elements, common to both halves of the matrix) PNL(a,b,c) = n(v,a x v,b x dc elements, found only on the left side of the main diagonal of the matrix) PND(a,b,c) = n(v,a x v,b x dc elements, found only on the right side of the main diagonal of the matrix) PN(a,b,c) = PNL(a,b,c) x PND(a,b,c) VPN(a,b,c) = PNL(a,b,c) + PND(a,b,c) APN(a,b,c) = PNL(a,b,c) - PND(a,b,c)___________________________________ A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... Acta Chim. Slov. 2004, 51, 589-609. 593 Expression of indices The values of indices derived from the G(a,b,c) matrix can be expressed analytically, Table 2. The expressions for multiplication-derived indices are in general simple, whereas for the summation-derived ones they are more complicated. Table 2. Analytical expressions of some indices derived from the G(a,b,c) matrix for the structure of 2,3-dimethyl pentane (23M5)._________________________________________________________________ Summation-derived indices VL(a,b,c) = 2a3c+2b(l+2c+3c)+3a(l+2c)+3b(2+3x2c+3c)+2a3b(l+2c)+3a3b+2c+3x3c+2x4c VD(a,b,c) = 2a(l+2c+3c)+2b3c+3a(2+3x2c+3c)+3b(l+2c)+3a2b(l+2c)+3a3b+2c+3x3c+2x4c V(a,b,c) = (2a+2b)(l+2c+2x3c)+(3a+3b)(3+4x2c+3c)+(2a3b+3a2b)(l+2c)+2(3a3b+2c+3x3c+2x4c) AV(a,b,c) = (2b-2a)(l+2c)+(3b-3a)(l+2x2c+3c)+(2a3b-3a2b)(l+2c) VC(a,b,c) = 3c(2a+2b)+(3a+3b)(l+2c)+3a+b+2c+3x3c+2x4c VNL(a,b,c) = 2b(l+2c)+3b(l+2x2c+3c)+2a3b(l+2c) VND(a,b,c) = 2a(l+2c)+3a(l+2x2c+3c)+3a2b(l+2c) VN(a,b,c) = (2a+2b)(l+2c) +(2a3b+3a2b)(l+2c)+(3a+3b)(l+2x2c+3c) Multiplication-derived indices p t u \ _ 23a+3b+llco3(a+3b+2c) VC(a h f 1 = 9a+b+7c33a+3b+5c P (a'hV> = 93a+3b+llc33(3a+b+2c) PN /a h r"> = 22a+2b+4c36b+c yrs u ' \ _ 22(3a+3b+llc)ol2(a+b+c) p-vr (^h r^ = 92a+2b+4c36a+c \rp(a i. c\ = 23a+3b+llco3(a+b+2c)/'o6b_|_o6a\ nw„ u '\ _ 24(a+b+2c)-26a+6b+2c AP^a h c"» = 23a+3b+llc33(a+b+2c)n6b 36a) VpWa'h c"> = 92a+2b+4c3cn6b+36aN> QP(a,b,c) = 36(b"a) APN(a,b,c) = 22a+2b+4c3c(36b-36a) Characteristics of indices The characteristics of several groups of these indices, as well as the presentation which known indices belong to which group of the indices presented here, have been a\ en eke uere 6,9,11,17,18,21-23,25 Secondaij indices From the indices considered above, several groups of secondary indices can be derived. We used the following ones, exemplified by those derived from the whole-matrix index V(a,b,c): \(&,b,c)/(N2-N), which is the average value of the non-diagonal element of the matrix, where N is the number of vertices in the graph, i.e. the carbon number of the alkane in question; V(a,b,c)/JV, which is the mean row-sum or the mean column sum of A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... 594 Acta Chim. Slov. 2004, 51, 589-609. the non-diagonal matrix elements; original indices raised to several exponents, e.g. V2(a,b,c), V1/2(a,b,c), etc, furthermore VMIv(a,b,c), where v is the degree of vertex in the graph, Viw(a,b,c), or V1/(m"A°(a,b,c), as well as lnV(a,b,c). Note: (N2-N) in the exponent means (N2-N). Among the multiplication derived indices, exemplified here by the whole-matrix index M(a,b,c), we tested their logarithm, lnM(a,b,c), the exponent \IN, which gives the geometric mean of the row products of the matrix, the exponent \I(N2-N), which gives the geometric mean of the non-diagonal elements of the matrix, and when feasible, several other exponents, e.g. 2, 72, V3, Vio, etc, or even Vio10. To derive secondary indices from the half-matrix indices, we used as a rule the exponents in the form 2/k or k/2, as applicable, instead of l/k or k, where k means any number, including N or N2-N. Among the indices derived from other parts of the matrix, besides the above-mentioned ways to derive the secondary indices, the following approaches were used: VC(a,b,c)/M7 is the average value of the non-diagonal elements common to both halves of the matrix, where NC is the number of elements common to both halves of the matrix. NC = VC(0,0,0) PClwc(a,b,c), on the other hand, is the geometric mean of the non-diagonal elements common to both halves of the matrix. Similarly, VNL(a,b,c)MW is the average value of the non-diagonal elements not common to both halves of the matrix, where NN is the number of elements not common to both halves of the matrix. NN = VNL(0,0,0) = VND(0,0,0) PNL1/AW(a,b,c) is the geometric mean of the non-diagonal elements not common to both halves of the matrix. Usefu/ness of indices The aim of present paper is to screen the 20 groups of indices, each containing an infmite number of possible indices, and the secondary indices derived from them, for most promising indices and not to find the best QSPR models. Therefore, the usefulness of the indices derived from the G(a,b,c) matrix was tested using a simple linear model of the form PP = mxl(a,b,c) + n, where PP means the physicochemical property in question, and I(a,b,c) means the index in question. To keep the results simple and A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... Acta Chim. Slov. 2004, 51, 589-609. 595 comparable, only one criterion of the goodness of relation is used, i.e. the correlation coefficient r. In most our cases maxima of the correlation coefficients coincide with minima of the standard error, therefore it is not given. The values of abs(r) > 0.99 are considered to be potentially useful.2 For screening purposes, also abs(r) > 0.9 is indicated in some cases. Results Labelling the indices Labelling the indices derived from the G(a,b,c) matrix is not a simple task. Some well-known indices, e.g. the Wiener index,7 W, or the Randič index,10 %, can be unambiguously labelled by a single letter since their value is determined by the structure taken into account and by the way of their derivation, which is well defined. The indices derived from the G(a,b,c) matrix, on the other hand, form several groups, each one consisting of an infinite number of indices. The indices of each group are derived in a different way and/or from different parts of the source matrix and on their derivation several data are used, i.e. the degrees of those two vertices which are in question at a particular tirne, the distance between them, as well as the three exponents to which the degrees of vertices and the distance between them are raised. Altogether, eight data are used besides the structure and to identify such an index unambiguously, at least a four-item label is needed, which presents that index as a function of exponents a, b, and c. In fact, four- to six-item labels have to be used to identify these indices unambiguously. Survey of indices that might be useful to predict the values of particular tested physicochemical properties One-digit values of exponents a, b, and c For screening the most promising indices, we used one-digit values of exponents a, b, and c, i.e. whole numbers from -20 (which gives results close to those using -») to 5, as well as + or -: 0.1 to 0.9. In correlations we used ali available data. Lacking data, i.e. for ali alkanes except octanes among R2 and S, for propane and butanes at ST, for propane at BON and MON, for 2,3-dimethyl butane (23M4) at RON, as well as for 2,2,3,3-tetramethyl butane (2233M4) at ST, BON, MON, RON, logVP, and VP/Pc, were not considered in correlations. A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... 596 Acta Chim. Slov. 2004, 51, 589-609. Table 3 indicates that when the size a«Jbranching of alkanes are considered and the best combination of tested one-digit values of exponents a, b, and c is used, then the best results, abs(rmax) > 0.999, are observed among the van der Waals parameters a0 and b0, represented here by Tc2/Pc and Tc/Pc. They are followed by MR, AHf°g, co, BP, AHv, VP/Pc, BP/Tc, Ve, Te, Pc, and logVP, where abs(rmax) > 0.99 is observed. Table 3. The highest (PP) of alkanes from obtain the values of indices (right) values of correlation coefficients observed using data of physicochemical properties propane to octanes inclusive, as well as one-digit values of exponents a, b, and c, to the original indices derived from the G(a,b,c) matrix (left) or also of secondaiy PP Index a b Tc2/Pc Tc/Pc MR AHf°g co BP AHv VP/Pc BP/Tc Ve Pc Te logVP Vm B ST C ac BON RON d MON V" de CED Sol.par. A Zc VL VL V V VL V V PL VL VL VL V V VL V V VL VL VC VC VL VC VL VL V PL PL PL PL 0 -0.6 0.9 -0.4 -0.5 0 -0.6 -0.4 0.3 -1 -0.5 -0.4 -0.7 -0.6 -0.6 -1 0.5 -1 1 -0.4 1 -1 1 -1 -2 -2 0.2 2 -3 3 0.9 -0.8 -0.5 -0.5 -0.5 -0.5 -0.4 -0.9 0.4 -0.9 0.4 -0.9 0.2 -2 0.9 0.1 2 -1 -4 -1 4 -5 4 -5 -2 4 -3 -5 -5 -5 0.5 -5 -4 -6 -4 -6 -6 -5 -5 -4 -4 -4 -6 -2 -6 -0.2 -5 0.5 0.5 -3 -0.3 -0.2 0.2 I"max 0.998 < 0.999 < 0.999 < 0.996 < 0.993 < 0.995 < 0.996 < 0.988 < 0.993 < 0.992 < -0.986 < 0.987 < -0.990 = 0.987 0.982 0.956 -0.971 0.952 -0.924 0.936 0.940 -0.918 0.925 0.905 < < < < < < < < Sec. index a b < and = refer to non-rounded data In exponent, (N2-N) stands for (N2-N) 0.9997 0.999 0.999 0.998 0.998 0.998 0.997 0.997 0.995 0.994 -0.994 0.992 -0.990 0.988 0.984 -0.977 -0.971 0.954 -0.950 0.945 0.942 0.940 0.925 0.910 v4/7 yjV/Lv v3/7 v2/7 p 2I{N2-N) V -JV/Ev -JV/Ev V PL2/(N2-N) 2/(N2-N) 1/7 1/3 PV VL v2/7 V vL1/2 v3/7 v2/N VL pVl/7 p 2/(N2-N) FLV(N2-N) VL 0.5 -0.5 -0.5 3 1 -0.6 -0.5 -0.5 0 -1 -1 -1 -0.5 1 -1 -? 0.4 2 4 11 2 -3 3 1 0.5 0.1 0 0.1 2 -0.6 -0.5 -5 0.4 3 -0.8 -0.6 -0.5 -0.4 -0.5 5 -0.9 0.1 2 4 1 -0.9 2 -0.9 c 0.4 1 0.5 0.9 1 -4 -5 -4 1 2 -? -4 -6 -? -5 3 -4 2 0.6 0.9 -0.9 -5 -0.2 0.849 < 0.852 Vw/Sv -4 -1 -5 0.778 < 0.839 PY/(N2-N) -2 -2 -? 0.766 < 0.833 P\/(N2-N) -2 -2 -? -0.756 < -0.789 2Vl7(7V2-A0 3 -0.5 -4 0.753 < -0.769 p^*2-*) 2 4 -0.7 c A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... Acta Chim. Slov. 2004, 51, 589-609. 597 In most cases the summation-derived original indices derived from the G(a,b,c) matrix give the best result (22 out of 29 cases). The rest represent the multiplication-derived indices. When also tested secondary indices are considered, then in 18 cases the summation-derived indices give the best result, in 5 cases the multiplication-derived ones, and in 6 cases the mixed, summation-and-multiplication-derived indices give the best result. Among them prevail by far the whole-matrix and half-matrix indices. The values of exponents a or b, giving rise to abs(r)max > 0.99 are between -1 and +1. The values of exponent c are mostly between -? and +1. Table 4 illustrates the best observed correlation coefficients of original indices derived from the G(a,b,c) matrix using one-digit values of exponents a, b, and c and the physicochemical properties Tc/Pc, BP, and the Solubility Parameter, which is formally equal to (AHv/Vm)1/2. In former cases, the whole-matrix and half-matrix summation-derived indices (VL and V), as well as multiplication-derived indices (PL and M) give rise to best results, whereas the mixed type indices (PV and VP) and indices derived from other parts of the matrix (PC and VC) are not so good. In the čase of Solubility Parameter only low values of correlation coefficients are observed. Anyway, these are provisional conclusions that have to be rechecked by fine-tuning of exponents a, b, and c. For this reason, the values of exponents a, b, and c are not given in Table 4. The observed "best ten" types of secondary indices are: Tc/Pc: Vw/Sv, V3/7, VL1/3, V1/3, V2/7, PV1/7, V1/2, VL1/2, VLW/Sv, V/N BP: Vw/Sv, V3/7, V1/2, VL1/2, V1/3, VL1/3, V4/7, V2/7, PV1/7, V1/7 Table 4. The best correlation coefficients observed using one-digit values of exponents a, b, and c, as well as data for alkanes from propane to octanes inclusive and the original indices derived from the G(a,b,c) matrix Tc/Pc BP Sol.Par. 0.766 0.999 vL 0.995 vL PL 0.994 V 0.995 V -0.753 vL -0.991 PL -0.988 Pl 0.702 PC -0.991 M -0.983 PC -0.679 VC 0.987 PV -0.981 M 0.658 M -0.977 PC 0.981 PV -0.647 PV 0.972 VC -0.957 VP 0.643 VP -0.968 VP 0.952 VC -0.641 V A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... 598 Acta Chim. Slov. 2004, 51, 589-609. Table 5. The highest values of correlation coefficients observed using data of physicochemical properties of octanes as well as one-digit values of exponents a, b, and c to obtain the values of the original indices derived from the G(a,b,c) matrix (left) or also secondaiy indices (right) P.P. Index a b c I"max = rmax^ ec. index V2, VL2 a 1 b 1 c Tc/Pc PV 1 1 -0.8 -0.999 -0.999 -0.8 BP/Tc PV 1 0.9 -2 -0.998 = -0.998 PV 1 0.9 -2 co VL -0.1 -0.2 -1 -0.997 = -0.997 VL -0.1 -0.2 -1 RON Vl -2 -0.6 -4 0.993 < 0.995 VL1/4 -2 -0.5 -4 MON Vl 1 0 2 -0.993 < -0.993 VP12 0.1 0.1 0.4 Tc2/Pc Pl 0.1 0.2 0.09 0.987 < 0.992 pLl/2 0.2 0.3 0.1 BON Vl 0.6 0.1 1 -0.988 < 0.989 InV -4 -0.6 -4 C PV -0.6 2 0 0.984 < 0.984 PV2 2 0 -0.1 Pc VL 1 0.4 -2 0.970 < 0.979 M1/10 2 0.5 -3 Ud Pl 0.5 0.03 -2 0.976 < 0.972 pLl/3 0.1 0.03 -3 d Pl 0.5 0.2 -2 0.963 < 0.967 pLl/4 1 -1 -5 AHf°g V -2 4 2 0.961 < 0.967 yl/7 -2 4 2 S Vl -0.4 -1 -1 -0.966 < -0.971 VCINC -0.4 4 -0.4 R2 V -0.5 0 -5 0.960 < 0.960 InV -0.5 0 -5 AHv PV -0.4 0.4 -8 -0.961 < -0.961 lnPV -0.5 0.6 -? V" PL 0.6 0.3 -2 -0.954 < -0.957 pLl/3 0.9 -0.4 -4 Vm Pl 0.6 0.3 -2 -0.955 < -0.957 pLl/4 1 -0.7 -5 ST V 5 3 2 -0.940 < -0.940 y6/7 5 3 2 BP Vl -? 4 2 -0.932 ^ -0.932 VL -? 4 2 MR V 3 1 2 0.894 < 0.921 InV 5 3 5 Te Vl 4 5 3 -0.881 ^ -0.881 VL 4 5 3 A VP 4 2 -0.9 0.823 < -0.852 pLi/4 3 5 -1 B AP 0.5 -2 2 0.842 ^ -0.842 AP 0.5 -2 2 Ve VP -0.2 -0.1 0.3 0.851 < 0.864 lnVL 5 2 4 de VP -0.2 -0.1 0.3 -0.837 < -0.859 lnVL 5 2 4 ac PL 0.1 0.3 -0.2 -0.849 < -0.851 VP1/8 3 1 -2 Zc VC -? 4 0.9 0.831 < 0.844 VC2 -? 4 1 logVP VN -? 2 1 0.775 < 0.779 VC/NC -4 -4 2 CED PL 0 -4 -0.5 0.731 < 0.736 VPN1/2 -2 0.5 -1 Sol.par. Pl 0 -4 -0.5 0.721 < 0.727 PNL2 -0.3 0.1 -0.3 VP/Pc VN -? 3 2 0.645 < 0.668 lnPV 7 5 2 < and = refer to non-rounded data Table 5 indicates that when only branehing of alkanes is considered and the best combination of tested one-digit values of exponents a, b, and c is used, then there is not observed any result having abs(rmax) > 0.999. The best result, abs(rmax) > 0.99, is observed testing the van der Waals parameter b0, represented here by Tc/Pc. It is A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... Acta Chim. Slov. 2004, 51, 589-609. 599 followed by the result of BP/Tc, co, RON, MON, and the van der Waals parameter a0, represented here by Tc2/Pc. In other cases, abs(rmax) < 0.99. In most cases the summation-derived original indices derived from the G(a,b,c) matrix give the best result (12 out of 31 cases). The rest represent the multiplication-derived indices (8 cases), the mixed type indices (9 cases) and the indices derived from other parts of the matrix (2 cases). When also tested secondary indices are considered, then in 12 cases the summation-derived indices give the best result, in 7 cases the multiplication-derived ones, in 7 cases the mixed, summation-(or subtraction)-and-multiplication-derived or multiplication-(or division)-and-summation-derived indices give the best result, whereas in 5 cases the indices derived from other parts of the matrix do. The values of exponents a or b, giving rise to abs(r)max > 0.99 are again in most cases between -1 and +1. The values of exponent c are mostly between -5 and +2. The best correlation coefficients observed using data of octanes and the primary indices derived from the G(a,b,c) matrix using one-digit values of exponents a, b, and c with Tc/Pc are: PV (-0.998) > VL (-0.998) > V (-0.998) > M (-0.994) > PL (-0.992) > VP (-0.992) > VC (0.966) > AP (0.928) > QV (-0.918) > AV (0.912) > VNL (-0.895) > VN (-0.892) > PNL (0.889) > APN (0.873) > VPN (-0.800) > PN (0.774) > QP (-0.752) > QVN (-0.750) > PC (0.743) > PVN (-0.709). The observed "best ten" types of secondary indices are PV4/7, PV, VL2, V2, PV1/2, VL, V, V6/7, V5/7, and PV1/3. The best observed indices giving rise to abs(r)max > 0.99, derived using two-digit values of exponents a, b, and c Some indices which give rise to the highest abs(r)max > 0.99 using two-digit values of exponents a, b, and c, are presented in Table 6 for alkanes from propane to octanes inclusive as well as for octanes only. We can see that, as a rule, the secondary indices derived from indices originating from the G(a,b,c) matrix give rise to the best correlation coefficients. Some of them are only slightly better than those obtained using one-digit values of exponents a, b, and c. In some cases, several local maxima of similar value are observed. The first digit in the value of exponents a, b, and c, used to derive an index from the G(a,b,c) matrix, defines in most tested cases the first three decimals of the A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... 600 Acta Chim. Slov. 2004, 51, 589-609. correlation coefficient. The second digit in the exponents a, b, and c improves in most tested cases the value of the third to fifth decimal, depending on how far from the best value of the exponent is its one-digit appimimation. The third digit in the exponents a, b, and c improves the value of the fifth or higher decimal of the correlation coefficient. For our purpose, five decimals in the value of the correlation coefficient are considered sufficient, therefore only two digits in the value of exponents a, b, and c are needed. Table 6. The highest values of correlation coefficients observed using data of physicochemical properties BP/Tc and Tc/Pc of alkanes from propane to octanes inclusive resp. of octanes only, as well as one-digit and two-digit values of exponents a, b, and c to obtain the values of the best indices derived from the G(a,b,c) matrix C3-C8 No. of Index c8 No. of Index digits a b c r digits PV a b c r BP/Tc 1 1 0.9 -2 -0.99755 BP/Tc 1 p 2KN2-N) 0 0.4 1 0.99522 2 TTT>\1(N2-N) 2.1 2.1 -12.6 -0.99879 2 -D 2KN2-N) 0.010 0.37 0.92 0.99953 2 p 2KN2-N) 0.97 1.03 -6.1 -0.99849 2 M -0.164 0.20 -0.111 -0.99848 2 M1/(7V2-A0 1.5 0.0 -5.6 -0.99834 2 PL1/3 0.056 0.030 -0.51 -0.99805 Tc/Pc 1 v Nl^y -0.2 0 1 0.99933 2 PV 1.26 0.82 -1.63 -0.99764 2 YWJy -0.27 -0.0091 1.06 0.99936 2 VlMZv -0.33 -0.20 1.14 0.99936 Tc/Pc 1 2 v2, VL2 vL2 1 1.01 1 0.99 -0.8 -0.99851 -0.88 -0.99859 1 VN/Ev -0.5 0.1 1 0.99915 2 v2 1.04 1.00 -0.80 -0.99855 2 vMIv -0.32 -0.32 1.15 0.99931 2 pV4/7 1.03 1.03 -0.86 -0.99807 2 v7*^ -0.51 -0.023 1.08 0.99930 2 V 1.03 1.04 -0.87 -0.99793 2 v7*^ -0.29 -0.30 1.10 0.99930 2 v4/7 1.05 1.05 -0.88 -0.99737 2 v7*^ 1.34 -0.26 0.88 0.99920 2 lnVL 1.04 0.76 -2.1 -0.99736 Discussion The G(a,b,c) matrix The G(a,b,c) matrix is in general non-symmetric, i.e. it is composed of two non-equal halves (cf. Fig. 1). It is symmetric only in special cases, i.e. when a = b, as well as when «-alkanes are considered. There are also other features in the G(a,b,c) matrix. On the one hand, if we use a positive value of exponent c, we have a usual form of the vertex-degree vertex-distance A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... Acta Chim. Slov. 2004, 51, 589-609. 601 matrix, which includes also the simple distance matrix. If, on the other hand, we use negative values of exponent c, we have to do with a reciprocal type of the vertex-degree vertex-distance matrix, which includes besides several other reciprocal types of matrix also the adjacency matrix (a = b = 0, c = -?).6 The primary indices derived from the G(a,b,c) matrix The asymmetry of the G(a,b,c) matrix enables deriving a great number of index groups. They can be derived by summation of its non-diagonal elements: V(a,b,c), VD(a,b,c), VL(a,b,c), VC(a,b,c), VN(a,b,c), VND(a,b,c), VNL(a,b,c); by multiplication of them: M(a,b,c), PD(a,b,c), PL(a,b,c), PC(a,b,c), PN(a,b,c), PND(a,b,c), PNL(a,b,c); by multiplication of summation derived indices: PV(a,b,c), PVN(a,b,c); by summation of multiplication derived indices: VP(a,b,c), VPN(a,b,c), etc. They can be the highest eigenvalues of the matrix, L(a,b,c) (not considered here), or they can be derived by other ways one can imagine, e.g. the difference between the values of the indices derived from two halves of the matrix: AV(a,b,c), AP(a,b,c), APN(a,b,c), the quotients of them QV(a,b,c), QVN(a,b,c), QP(a,b,c), etc. The indices VL(a,b,c), VD(a,b,c), V(a,b,c), PV(a,b,c), VC(a,b,c), PL(a,b,c), PD(a,b,c), M(a,b,c), VP(a,b,c), and PC(a,b,c) represent the size as well as branching of the molecule. They thus belong to the BIM- type34 of indices. They were used to test data of ali alkanes from propane to octanes inclusive, as well as of octanes only. The indices AV(a,b,c), QV(a,b,c), VNL(a,b,c), VN(a,b,c), PVN(a,b,c), QVN(a,b,c), QP(a,b,c), AP(a,b,c), PNL(a,b,c), PN(a,b,c), VPN(a,b,c), and APN(a,b,c) do not reflect the size but only branching of the molecule. Therefore, they belong to the BIA-type34 of indices. For this reason they were used to index the physicochemical properties of octanes only. These facts are reflected in Tables 3, 5, 7, and 8. It has been reasonably expected that the characteristics of these indices would vary widely, which proved to be true. There is also an additional source of variation of characteristics of these indices. It is the exponents a, b, and c. Here we have an infinite number of possibilities to choose the value of each of these exponents, as well as of the combinations of them. Using these indices, we have at our disposal at least 20 groups of them, each group containing an infinite number of possible indices. A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... 602 Acta Chim. Slov. 2004, 51, 589-609. The secondary indices derived from the G(a,b,c) matrix From these groups of, let us say, primary indices derived from the G(a,b,c) matrix, we can derive an infinite number of groups of, let us say, secondary indices. We can derive them by division of values of primary indices using a variable factor like N (the number of vertices in the graph, which is equal to the number of rows or columns in the matrix), or N2-N (the number of non-diagonal elements of the matrix), or some other variable that one finds plausible, e.g. NC (the number of elements common to both halves of the matrix), NN (the number of elements not common to both halves of the matrix), etc, etc. We can derive them by raising the values of primary indices to an appropriate exponent, which can be any real number or a variable like N, N2-N, NIJy (v is the degree of vertex in the graph), etc, etc. We can also draw the logarithms of the values of primary indices. Here we have an interesting situation. Namely, the logarithms of the multiplication-derived indices are in fact summation-derived indices since logx(TI(v;a x v,b x d/)) = Z(axlogxv! + bxlogxv7 + cxlogxd/y). Because logxI = logyI / logyx, the base of the logarithm does not influence the correlation coefficient in this čase. Due to this fact, the natural logarithm (In) was chosen here. Usefulness of indices derived from the G(a,b,c) matrix Then another question arises: Are ali of these indices useful? And if not, then which of them are useful? To obtain some indication to the answer, correlation of the values of a number of these indices with the values of 29 physicochemical properties of alkanes from propane to octanes inclusive, as well as of 31 physicochemical properties of octanes, was performed. A wide spread of values of correlation coefficients was observed. Since only the best correlations are interesting, the indices giving rise to them are presented in Table 3 for alkanes from propane to octanes inclusive, and in Table 5 for octanes only. As a rule, the maxima of correlation coefficients coincide with minima of standard error data and for this reason the data of standard error are not given in these tables. From Table 3 it can be seen that when the influence of the size a«Jbranching of alkanes is to be considered, then a number of physicochemical properties can be well2 A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... Acta Chim. Slov. 2004, 51, 589-609. 603 correlated with some of the tested indices. Less well expressed is the correlation when only the influence of branching of alkanes is considered; see Table 5. From the data, which were the basis for construction of Table 3 and 5 we can deduce a lot of information: 1. Separate data for «-alkanes are not given in these tables, but as a rule of thumb the best correlations between the tested physicochemical properties and indices of alkanes are observed when considering «-alkanes; lower correlations are observed when ali alkanes are considered and the lowest ones when only octanes are considered. Thus, this rule of thumb is: abs(r)max: «-alkanes » ali alkanes > octanes. 2. Which group of indices is the most promising to index well a physicochemical property? According to the criterion abs(r)max > 0.99 for the correlation with any one of the tested physicochemical properties,2 these are the following indices: a. Testing data of ali alkanes, from propane to octanes inclusive: Original indices and/or some tested secondary indices derived from them: V(a,b,c), VL(a,b,c), PV(a,b,c), PL(a,b,c), M(a,b,c), and VP (a,b,c); b. Testing data of octanes: Original indices and/or some tested secondary indices derived from them: V(a,b,c), VL(a,b,c), PV(a,b,c), M(a,b,c), PL(a,b,c), and VP(a,b,c); c. Other tested indices [VC(a,b,c), PC(a,b,c), AV(a,b,c), QV(a,b,c), VNL(a,b,c), VN(a,b,c), PVN(a,b,c), QVN(a,b,c), QP(a,b,c), AP(a,b,c), PNL(a,b,c), PN(a,b,c), VPN(a,b,c), APN(a,b,c) and the tested secondary indices derived from them] do not give rise to abs(r)max > 0.99. 3. From the G(a,b,c) matrix can be derived good (original and/or secondary) indices for the following physicochemical properties, since abs(r)max > 0.99: a. Testing data of ali alkanes, from propane to octanes inclusive: Tc/Pc as well as: VL(a,b,c): MR, Tc2/Pc, BP, AHf°g, AHv, Ve, Te, BP/Tc, logVP, Pc V(a,b,c): MR, Tc2/Pc, BP, AHf°g, AHv, Ve, Te, BP/Tc, logVP, co, PV(a,b,c): MR, Tc2/Pc, BP, AHf°g, AHv, Ve, Te, logVP PL(a,b,c); MR, Tc2/Pc, BP, AHf°g, AHv, Ve, BP/Tc, co, VP/Pc M(a,b,c): MR, Tc2/Pc, BP, AHv VP(a,b,c): Pc A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... 604 Acta Chim. Slov. 2004, 51, 589-609. b. Testing data of octanes: Tc/Pc, BP/Tc, co, as well as: VL(a,b,c): RON, MON V(a,b,c): RON PV(a,b,c): MON PL(a,b,c): MON, Tc2/Pc M(a,b,c): MON VP(a,b,c): MON, Tc2/Pc 4. The indices derived from the G(a,b,c) matrix which might be interesting as indices for the physicochemical properties, since 0.9 < abs(r)max < 0.99: a. Testing data of alkanes from propane to octanes inclusive: Ali groups of tested indices. b. Testing data of octanes: Ali groups of them, even PVN(a,b,c) and PC(a,b,c), which are the worst cases. 5. The original and the tested secondary indices derived from the G(a,b,c) matrix are not good indices for the following physicochemical properties, since abs(r)max < 0.9: a. Testing data of ali alkanes, from propane to octanes inclusive: de, Zc, A, Cohesive energy density, Solubility parameter, as well as: V(a,b,c): MON, V" PV(a,b,c): BON, V" VC(a,b,c): VP/Pc, V" PL(a,b,c): nD, d M(a,b,c): BON, MON, RON, V", nD, d VP(a,b,c): V", nD, d PC(a,b,c): ST, BON, MON, RON, V", nD, d b. Testing data of octanes: logVP, VP/Pc, Te, Ve, de, ac, Zc, A, B, CED, Sol.par., as well as PV(a,b,c): BP VC(a,b,c): AHf°g, MR, BP, ST AV(a,b,c): MR QV(a,b,c): MR, nD, d, Vm, V", AHf°g, BP, ST VNL(a,b,c): MR, Pc, Tc/Pc VN(a,b,c): MR, ST, Vm PVN(a,b,c): Ali of them except BON, MON, RON, Tc2/Pc QVN(a,b,c): Ali of them except nD, d, Vm, V" M(a,b,c): BP, MR VP(a,b,c): BP, MR, ST PC(a,b,c): Ali of them except BON, MON, RON, Tc2/Pc QP(a,b,c): Ali of them except RON, nD, d, Vm, V" AP(a,b,c): R2, MR, ST PNL(a,b,c): BP, MR, ST, Tc/Pc, AHv, AHf°g, Pc, C PN(a,b,c): Ali of them except nD, d, Vm, V" VPN(a,b,c): Ali of them except BON, RON, S, BP/Tc, co, nD, d, Vm, V" APN(a,b,c): Ali of them except BON, MON, RON, S, BP/Tc, Tc2/Pc, co, C, nD, d, Vm, V" A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... Acta Chim. Slov. 2004, 51, 589-609. 605 Evidently, the indices derived from the G(a,b,c) matrix are not good indices of some physicochemical properties of alkanes, namely of de, Zc, A, Cohesive energy density, and Solubility parameter. Well predieted are, referring to alkanes of different size and branehing, Tc2/Pc, Tc/Pc, MR, AHf°g, co, BP, AHv, VP/Pc, BP/Tc, Ve, Te, Pc, and logVP, whereas referring to alkanes of the same size are well predieted Tc/Pc, BP/Tc, co, RON, MON, and Tc2/Pc. Other physicochemical properties are predieted worse than one would desire. 6. The whole-matrix as well as the half-matrix indices are in general good indices, whereas the indices derived from elements common to both halves of the matrix as well as the indices derived from elements not common to both halves of the matrix are in general not. They are to be tested for their usefulness as the second or the third index in multiparameter relationships. Using tested indices derived from the G(a,b,c) matrix and optimizing the values of exponents a, b, and c one can arrive, considering particular physicochemical properties of alkanes, to quite good correlations. This can be achieved by proper seleetion of the type of the index, of values of exponents a, b, and c, as well as by proper seleetion of modification of the index (secondary indices). Using these indices, the maximum of correlation coefficient in the space of exponents a, b, and c is in most cases flat and in several cases there are observed several local maxima of similar value. Looking for the best QSPR/QSAR model and using one or more of these indices, one should test the values of indices at these local maxima as well as in vicinity of them. We compared the best observed indices derived from the G(a,b,c) matrix using two-digit values of exponents a, b, and c with a seleetion of known indices: %,w D (the largest eigenvalue of the distance matrix),34 EA,35 ID,36 J,37 lh19'20 Mi,38 MTI39 (the values were taken from ref40), Sch-S,41 Sch-TF,41 Xu,42 W,7 and Z.43 This comparison is presented in Table 7 for alkanes from propane to oetanes inclusive and in Table 8 for oetanes only. When alkanes from propane to oetanes inclusive are taken into consideration, Table 7, Tc2/Pc, Tc/Pc, MR, BP, AHv, and Ve are well predieted (r > 0.99) by Mi, Xu, ID, Xu, x, and Mi, respectively, but the best indices derived from the G(a,b,c) matrix are better in this respect. There is no čase that an index from the above mentioned seleetion would be better than the best observed index derived from the G(a,b,c) matrix. A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... 606 Acta Chim. Slov. 2004, 51, 589-609. When only octanes are considered, Table 8, no index of the above mentioned selection gives rise to abs(r)max > 0.99, whereas a number of indices derived from the G(a,b,c) matrix do. In most cases, the indices derived from the G(a,b,c) matrix are appreciably better than those from the above mentioned selection. Table 7. The highest values of correlation coefficients and standard errors observed using data of physicochemical properties (PP) of alkanes from propane to octanes inclusive, as well as wifh the best observed indices derived from the G(a,b,c) matrix using two-digit values of exponents a, b, and c (left) and with a selection of known indices (right) PP Index(a,b,c) sxl0x I"max 0.9997 Index sxl0x 20.80 I"max Tc2/Pc V4/7(0.49, 0.50, 0,41) 6.67 0.997 Tc/Pc V^-0.56, 0.039, 1.08) 1.34 0.999 Xu 3.10 0.996 MR V3/7(-0.52, 0.00, 0.50) 3.09 0.999 ID 4.61 0.997 AHf°g V2/7(3.0, 0.090, 0.91) 4.68 0.998 Sch-S 14.86 0.979 co PL2/^(1.12, 2.1, 0.99) 3.84 0.998 D 18.38 0.949 BP V^-0.62, -0.60, -3.9) 2.76 0.998 Xu 5.36 0.992 AHv V^-0.47, -0.50, -5.3) 7.07 0.997 X 10.92 0.993 VP/Pc PL2/^(-0.83, -5.5, -3.1) 3.00 0.997 Xu 26.00 -0.762 BP/Tc PL2/^(0.029, 0.40, 0.89) 1.70 0.995 D 4.16 0.972 Ve PV1/7(0.087, 3.0, 1.64) 7.45 0.995 Ui 10.41 0.990 Pc VL1/3(0.98, -0.86, -oo) 4.32 -0.994 Xu 9.75 -0.968 Te V2/7(-1.37, -0.75, -3.8) 5.77 0.994 ID 9.44 0.983 logVP V(-0.50, -0.50, -5.7) 9.46 -0.990 Xu 10.89 -0.987 Vm VL1/2(1.05, -0.46, -oo) 3.33 0.988 ID 4.31 0.980 ST V2W(-oo, 4.8, 3.0) 3.10 -0.986 X 7.16 0.921 B V3/7(-1.42, -0.52, -5.1) 2.42 0.985 Xu 3.13 0.975 C VL(0.56, -0.67, -oo) 2.14 -0.972 D 2.54 -0.960 ac VL(0.094, -1.48, -3.1) 8.49 0.956 D 9.47 0.945 BON PL2/(m-^(4.3, 2.1, 0.61) 1.04 -0.951 D 2.66 -0.603 RON VC2(-2.4, -0.74, -5.0) 1.05 0.948 D 2.82 -0.513 d VL (2.3, 1.36, -0.54)/N 1.53 0.947 Sch-S 2.05 0.904 MON VL (-2.2, -0.79, -4J)/N 0.97 0.945 EA 2.43 0.577 IlD VL(2.8, 1.61, -0.53) 8.58 0.929 Sch-S 11.37 0.871 V" VLwffiv(0.99, -0.87, -oo) 3.34 0.917 Sch-TF 3.70 0.897 In exponent, (N2-N) stands for (N2-N) x: Integer to adjust the value of s to the same order of magnitude If we compare the correlation coefficients observed here with those reported by some other authors using more than one variable in their equations, we can see that in A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... Acta Chim. Slov. 2004, 51, 589-609. 607 several cases, by using one index derived from the G(a,b,c) matrix, we can obtain results which are comparable to or not much worse than those obtained by using models with several (2 to 5) other indices at once, Table 9. Table 8. The highest values of correlation coefficients observed using data of physicochemical properties of octanes, as well as the best observed indices derived from the G(a,b,c) matrix using two-digit values of exponents a, b, and c (left), and with a selection of known indices (right) P.P. Index(a,b,c) I"max Index I"max Tc/Pc VL2(1.03, 1.01, -0.84) -0.999 Xu 0.917 BP/Tc PV(1.07, 1.00, -0.77) -0.999 Ui 0.927 co VL(-1.08, -0.22, -1.20) -0.997 Ui 0.956 RON VL1/4(-2.0, -0.60, -4.0) 0.995 ID -0.964 BON lnV(-oo -1.37, -3.6) 0.994 Xu -0.961 MON VP1/2(0.128, 0.128, 0.33) -0.993 D -0.941 Tc2/Pc PL1/2(0.20, 0.30, 0.111) 0.992 D 0.962 C PV2(1.75, -0.35, -0.071) 0.986 ID -0.920 Pc M1/10(0.20, o.45, -3.0) 0.979 J 0.793 nD PL1/3(1.01, -0.58, -4.9) 0.977 Sch-TF -0.794 S VC (-0.27, 4.6, -0.29)/NC -0.975 Ui 0.942 AHf°g V1/7(-2.1, 4.1, 2.0) 0.968 ID -0.709 d PL1/4(1.00, -1.08, -5.0) 0.967 Sch-TF -0.745 R2 lnV(0.47, 0.00, -5.0) 0.963 ID 0.902 AHv lnPV(-0.40, 0.48, -«>) -0.962 ID 0.886 V" PL1/3(0.90, -0.45, -3.9) -0.958 Sch-TF 0.717 Vm PL1/4(1.00, -0.93, -4.7) -0.958 D -0.941 ST V6/7(5.1, 3.1, 2.2) -0.944 Z 0.453 BP VL(—, 3.9, 2.1) -0.937 Z 0.742 MR lnV(4.9, 2.9, 5.1) 0.921 J -0.629 Table 9. Comparison of correlation coefficients and standard errors observed by some other authors and our data. No. of Li33 | Ivanciuc44 | Ivanciuc 45 Table 7, here variables 5 5 3 2 1 Indices X** W** various n, ib I(a,b,c) r s r s r s r s r s BP 0.997 3.31 0.998 3.07 0.994 2.79 0.984 4.94 0.998 2.76 Te 0.996 4.76 0.998 4.54 0.992 5.77 Pc 0.992 0.50 0.990 0.55 -0.994 0.43 ST 0.998 0.22 0.989 0.29 -0.977 0.31 AHv 0.990 0.63 0.975 0.81 0.997 0.30 nD 0.984 0.0025 0.960 0.0038 0.929 0.0086 d 0.990 3.73 0.961 7.30 indices 0.947 15.3 **: Molecular connectivity indices, resp. Extended Wienei A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... 608 Acta Chim. Slov. 2004, 51, 589-609. Thus, in several instances a properly selected index of the type described here gives as good a correlation as if several classic indices are used at once. Naturally, there are also exceptions. On the other hand, when using the Randič10 index x one can expect several good results since % = VlC-1^,-1^,-?) = 1/2V(-1/2,-1/2,-?), which is close to many good indices derived from the G(a,b,c) matrix. The exponents which are used to derive the Wiener7 index, W = VL(0,0,1) = V2V(0,0,1) are not so close to the exponents of the best observed indices as there are exponents to be used to derive the Randič index x. This is the reason why the Randič index x is often a better index than the Wiener index. Extending the idea of the G(a,b,c) matrix and the diversity of indices derived from it to other sources of indices would open additional possibilities to find good indices for particular needs. Conclusion The indices derived from the G(a,b,c) matrix in the ways described here are promising indices of several physicochemical properties of alkanes. Therefore, their use in QSPR/QSAR models is recommended. References 1. L. Pogliani, Indian J. Chem. 2003, 42A, 1347–1353. 2. Z. Mihalić, N. Trinajstić, J. Chem. Educ. 1992, 69, 701–712. 3. M. Randić, C. Basak, M. Pompe, M. Novič, Acta Chim. Slov. 2001, 48, 169–180. 4. R. Todeschini, V. Consonni, Handbook of Molecular Descriptors, Wiley-CH, New York 2000. 5. E. Estrada, Chem. Phys. Lett. 2001, 336, 248–252. 6. A. Perdih, B. Perdih, Acta Chim. Slov. 2002, 49, 67–110. 7. H. Wiener, J. Am. Chem. Soc. 1947, 69, 17–20. 8. O. Ivanciuc, Rev. Roum. Chim. 1989, 34, 1361–1368. 9. A. Perdih, B. Perdih, Acta Chim. Slov. 2002, 49, 291–308. 10. M. Randić, J. Am. Chem. Soc. 1975, 97, 6609–6615. 11. A. Perdih, B. Perdih, Acta Chim. Slov. 2003, 50, 95–114. 12. O. Ivanciuc, Rev. Roum. Chim. 1999, 44, 519–528. 13. O. Ivanciuc, Rev. Roum. Chim. 2000, 45, 587–596. 14. A. T. Balaban, D. Mills, O. Ivanciuc, S. C. Basak, Croat. Chem. Acta. 2000, 73, 923–941. 15. O. Ivanciuc, T. Ivanciuc, A.T. Balaban ACH-Models Chem. 2000, 137, 57–82. 16. O. Ivanciuc, J. Chem. Inf. Comput. Sci. 2000, 40, 1412–1422. 17. A. Perdih, B. Perdih, Acta Chim. Slov. 2002, 49, 497–514. 18. A. Perdih, B. Perdih, Acta Chim. Slov. 2003, 50, 83–94. 19. L. Lovasz, J. Pelikan, Period. Math. Hung. 1973, 3, 175–182. 20. D. M. Cvetković, I. Gutman, Croat. Chem. Acta 1977, 49, 115–121. 21. A. Perdih, B. Perdih, Acta Chim. Slov. 2002, 49, 309–330. A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ... Acta Chim. Slov. 2004, 51, 589-609. 609 22. A. Perdih, B. Perdih, Acta Chim. Slov. 2002, 49, 467-482. 23. A. Perdih, B. Perdih, Acta Chim. Slov. 2003, 50, 161-184. 24. I. Gurman, W. Linert, I. Lukovits, Ž. Tomovič, J. Chem. Inf. Comput. Sci. 2000, 40, 113-116. 25. A. Perdih, B. Perdih, Acta Chim. Slov. 2003, 50, 513-538. 26. D. R. Lide, CRC Handbook of Chemistry and Physics, 76th Ed., CRC Press, Boca Raton 1995-96. 27. J. A. Dean, Lange's Handbook of Chemistry, McGraw-Hill, New York 1985. 28. L. Pogliani, J. Phys. Chem. 1995, 99, 925-937. 29. A. T. Balaban, I. Motoc, MATCH (Commun. Math. Chem.) 1979, 5, 197-218. 30. C. Morley, Combust. Sci., Technol. 1987, 55, 115-123. 31. E. S. Goli, P.C. Jurs, J. Chem. Inf. Comput. Sci. 1999, 39, 1081-1089. 32. B. Ren, J. Chem. Inf. Comput. Sci. 1999, 39, 139-143. 33. X. H. Li, Chem. Phys. Lett. 2002, 365, 135-139. 34. A. Perdih, M. Perdih, Acta Chim. Slov. 2000, 47, 231-259. 35. Y.-Q. Yang, L. Xu, C.-Y. Hu, J. Chem. Inf. Comput. Sci. 1994, 34, 1140-1145. 36. M. Randič, J. Chem. Inf. Comput. Sci. 1984, 24, 164-175. 37. A.T. Balaban, Chem. Phys. Lett. 1982, 89, 399-404. 38. M. Randič, Acta Chim. Slov. 1997, 44, 57-77. 39. H.P. Schultz, J. Chem. Inf. Comput. Sci. 1989, 29, 227-228. 40. Z. Mihalič, S. Nikolič, N. Trinajstič, J. Chem. Inf. Comput. Sci. 1992, 32, 28-37. 41. H.P. Schultz, T.P. Schultz, J. Chem. Inf. Comput. Sci. 2000, 40, 107-112. 42. B. Ren, J. Chem. Inf. Comput. Sci. 1999, 39, 139-143. 43. H. Hosoya, Buli. Chem. Soc. Japan 1971, 44, 2332-2339. 44. O. Ivanciuc, T. Ivanciuc, A.T. Balaban, Internet Electron. J. Mol. Des. 2002, 1, 467-487. 45. O. Ivanciuc, T. Ivanciuc, D. Cabrol-Bass, A.T. Balaban, Internet Electron. J. Mol. Des. 2002, I, 319-331. Povzetek Iz matrike tipa G(a,b,c) lahko izvedemo vsaj 20 skupin topoloških indeksov. Vsaka od teh skupin ima neskončno mnogo teh indeksov. Številni od njih korelirajo z nekaterimi od 29 odnosno 31 upoštevanih fizikokemijskih lastnosti alkanov z abs(r)max > 0.99, in še več od njih z abs(r)max > 0.9. Zato priporočamo njihovo uporabo v modelih za QSPR/QSAR. Neprimerni pa so za opis fizikokemijskih lastnosti dc, Zc, A, gostote kohezivne energije in topnostnega parametra. A. Perdih, B. Perdih: Topological Indices Derived From the G(a,b,c) Matrix, Useful as Physicoche. ...