THE JOURNAL OF CHEMICAL PHYSICS 134, 054106 (2011)

Sources of the deciencies in the popular SPC/E and TIP3P models of water
Pter T. Kiss and Andrs Baranyaia)
Institute of Chemistry, Etvs University, 1518 Budapest 112, P.O. Box 32, Hungary

(Received 18 November 2010; accepted 7 January 2011; published online 1 February 2011) Motivated by the results of Vega et al. [J. Phys. Condens. Matter 20, 153101 (2008)] about the phase diagram of water, and by the results of Kiss and Baranyai [J. Chem. Phys. 131, 204310 (2009)] about the properties of gas-phase clusters, we carried out a comparative study of the structure modeled by SPC/E and TIP3P interactions in ambient liquid water. The gas-phase clusters of SPC/E and TIP3P models show erroneous structures, while TIP4P-type models, either polarizable or not, provide qualitatively correct results. The trimers of SPC/E and TIP3P are planar in gas phase, contrary to experimental and TIP4P-type models. The aim of this study was to see whether traces of these false geometries characteristic to SPC/E and TIP3P in gas phase can also be found in the liquid phase. For this purpose we selected trimers formed by adjacent neighbors of water molecules in the liquid and calculated their geometrical features. We determined angles formed by the HO bonds of the molecules with OO vectors and with the normal vector of the OOO plane in the selected trimers. Our results showed that, despite high temperature, the SPC/E and TIP3P water contains larger number of planar arrangements than other TIP4P-type models. Although structural differences presented in this study are small, they are accurately detectable. These results weaken the reliability of studies obtained by the SPC/E or TIP3P models even in the liquid phase. 2011 American Institute of Physics. [doi:10.1063/1.3548869]
I. INTRODUCTION

The SPC/E interaction1 is the most popular rigid nonpolarizable model of water. This model was applied more than 8000 times to simulations.2 Immense popularity of the model can be attributed to its attractive features: partial charges are placed on the atoms; it has a tetrahedral HOH angle with a OH bond of 1 ; and polarization self-energy correction is also taken into account. In addition, SPC/E provides acceptable estimates for the structure, the internal energy, the density, and the diffusivity of ambient water.3 The model also predicts two challenging features reasonably well: the critical properties and the dielectric coefcient.3 Recent studies, however, established serious questions about the performance of the model. It seems that this remarkable behavior described above is rather the result of fortuitous cancellations, well known in integral equations of statistical physics, than inherent accuracy of the picture. Recently, Vega and co-workers carried out comprehensive studies of the phase diagram of water.4 They obtained 215 5 K for the melting point of hexagonal ice modeled by the SPC/E model at 1 bar pressure.4 The phase diagram of the SPC/E model was also erroneous. Not only it was severely distorted but also certain ice phases were entirely missing from the diagram.4 In the case of the TIP4P model5 there were shifts and inaccuracies in phase lines but the qualitative picture was correct even if the melting temperature was low for this model too (232 5 K).4 Recently, present authors carried out an extensive study of water clusters determined for different model interactions.6
a) Author to whom correspondence should be addressed. Electronic address:

bajtony@chem.elte.hu. 0021-9606/2011/134(5)/054106/5/$30.00

Clearly, for nonpolarizable rigid models such calculations provide inaccurate energy values; still the geometry is characteristic for the model. The principal angle of the dimer (the angle between the OO vector and the bisector of the acceptor molecule in the hydrogen bond) for the SPC/E model was 22.3 instead of the experimental 57 . This large discrepancy is typical for models with SPC-type structure contrary to nonpolarizable models with TIP4P-type structure where the angle was 4249 . A more remarkable nding was that the most stable trimer for the SPC/E model was planar contrary to experimental7 and all the other TIP4P-type models studied. This was the case for the tetramer too (see Fig. 1).6 Hexamer congurations have the following hierarchy in terms of intermolecular energy: prism cage < book < cyclic.8 The SPC/E has the strongest deviation from this hierarchy contrary to quantum chemical calculations8 and the best polarizable models.9, 10 So, SPC/E showed inaccuracies in predictions both for crystalline phases and gas clusters. Since this potential was devised for liquid water at ambient condition, it can still perform well under these circumstances. As for averages of important properties and the pair-correlation function calculated by SPC/E, the model does a good job,3 although the extremely low freezing temperature makes the temperature scale, and this way the results somewhat ambiguous (the temperature range between the freezing temperature and critical temperature for the TIP4P model is closer to the correct value, because both of these temperatures are shifted down by 40 K).4 If we want to study the difference in performance between the SPC/E and other models in liquid phase beyond the accuracy of calculated averages, we have to analyze the structure of liquid water. The natural way to go beyond the pair-correlation function would be to expand the correlation
2011 American Institute of Physics

134, 054106-1

Author complimentary copy. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

054106-2

P. T. Kiss and A. Baranyai

J. Chem. Phys. 134, 054106 (2011)

In Sec. II we describe the details of the calculations and present the results. Finally, in Sec. III we conclude the study.
II. RESULTS AND DISCUSSION

FIG. 1. Geometry of the most stable conformer of the trimer (upper row) and tetramer (lower row) for the TIP4P and the SPC/E model.

functions to higher order terms.11 Unfortunately, the large number of independent variables makes such an attempt extremely difcult.12 Even if we could overcome this technicality, we should dene what the correct structure of liquid water should be beyond the pair-correlation function, and which model gives a better description of it. We chose a simpler method instead of correlation functions. We have already identied the shortcomings of the SPC/E model for gas-phase clusters. These problems were not present for models with TIP4P-type structures, either polarizable or not. It is a natural question to ask whether traces of these problems manifested in the gas phase can also be identied in the liquid state. We searched for trimers formed by molecules reasonably close to one another. Three oxygen atoms form a plane. The gas-phase SPC/E trimer is completely at contrary to other models6 and experimental fact7 where two hydrogen atoms are positioned above and one hydrogen atom is positioned below the plane (see Fig. 1). Is this discrepancy among cluster structures noticeable in the highly disordered water phase at ambient conditions? Can we identify noticeable differences in trimer geometries between SPC/E and other models knowing that energy levels between geometrically similar cluster minima compared to kT are small? In this paper we carry out comparative calculations trying to obtain answers to the questions above. Primarily, we focus on the SPC/E model with respect to the TIP4P model, because these models have different geometries. When we use the word geometry, we mean not only the HOH angles and OH bond lengths but also most importantly the position of the negative charge. Models with TIP4P arrangements position the negative charge on the main axis of the molecule between the position of the oxygen and the hydrogen atoms. While SPC/E will be in the center of the paper, we will refer to TIP3P too.5 The latter model is frequently used in program packages simulating systems with biomolecules. The SPC/E and the TIP3P are similar because the negative charge is placed on the oxygen for both models. This causes the cluster geometry of TIP3P to show very similar cluster properties13 as the SPC/E despite the different HOH angles and OH bond lengths. The characteristic angle for the TIP3P dimer, 27 , is also small.5 We do not show TIP3P trimers and tetramers in Fig. 1 because they are also planar, and the slightly different bond angles and bond lengths do not make visually noticeable discrepancy between the arrangements of the two models.

We performed molecular dynamics simulations in the NPT (isotherm-isobar) ensemble with 432 molecules using the quaternion algorithm integrated by the Gear method.14 The long-range forces were treated by the reaction eld technique. The time step was 0.25 fs. The temperature was 298 K, and the pressure was 1 bar. The properties of the models we studied are presented in Table I. We studied the SPC/E, the TIP3P, and the TIP4P intermolecular potentials. We further created two models for this study. The model termed SPC/E_mod, as it can be seen in Table I, has SPC geometry with a negative charge at 0.15 from the oxygen atom, while TIP4P_mod has TIP4P geometry with a negative charge at 0.25 from the oxygen atom both on the main axis of the molecule. We chose the charges to obtain reasonable internal energy and density. In the case of SPC-type models the polarization correction was added to the calculated internal energy values. The criterion to choose a trimer was that the mutual OO distances should be smaller than d. We performed several test runs to choose the optimum distance of d. We opted for d = 3.2 which is close to the rst minimum of the OO pair-correlation function. This value was optimal because substantially longer distance would select fairly distorted triangles, while substantially shorter distance could exclude many atoms not having enough neighbors to the analysis. We found that small changes in d change the number of trimers identied in the system, but have only little impact on the result. The results presented below are statistics of more than 8 million trimers for each model. We determined the angle between the six OH vectors and the normal vector of the OOO plane. The results are shown for different models in Fig. 2. The distribution for each model has two peaks. The high peak at 90 represents the OH vectors in the OOO plane, while the broad maximum at smaller angles is
TABLE I. Properties of the models studied. SPC/E lOH / HOH /deg lOM / QH /esu /kJmol1 / U/kJmol1 /gcm3 /D Qxx/D Qyy/D Qzz/D 1.0000 109.47 0.00 0.4238 0.6502 3.1656 41.5 0.998 2.351 2.186 0.902 1.855 SPC/E_mod 1.0000 109.47 0.15 0.5150 0.6502 3.1656 42.6 0.997 2.114 2.665 0.383 2.282 TIP3P 0.9572 104.52 0.00 0.4170 0.6363 3.1507 40.0 0.984 2.347 1.761 0.078 1.682 TIP4P 0.9572 104.52 0.15 0.5200 0.6487 3.1536 41.4 0.991 2.177 2.203 0.112 2.091 TIP4P_mod 0.9572 104.52 0.25 0.6000 0.6487 3.1536 41.8 1.021 1.936 2.619 0.285 2.334

List of abbreviations: lOH : OH bond length; HOH : HOH bond angle; lOM : distance of the negative charge from the oxygen; QH : charge on the hydrogen; , : Lennard-Jones parameters; U: internal energy; : density; the dipole moment; Qxx , Qyy , and Qzz are the elements of the quadrupole moment.

Author complimentary copy. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

054106-3

Molecular models of water

J. Chem. Phys. 134, 054106 (2011)

0.011 0.010 0.009 SPC/E SPC/E_mod TIP3P TIP4P TIP4P_mod

probability

0.008 0.007 0.006 0.005 0.004 0.003 0 10 20 30 40

50

60

70

80

90

FIG. 3. A schematic picture of a selected trimer. For the sake of clarity oxygens are not shown. The normal vector of the OOO plane is marked by an arrow. Dark circles are the protons in hydrogen bonds. The wide arrow shows a proton positioning near the OOO plane.

angle
FIG. 2. Distribution of the angle between the normal vector of the OOO plane and the six OH bond vectors of the trimer. No selection was made.

created by the OH bond of protons not involved in hydrogen bonds of the actual trimer. The shape of the distributions is determined by two factors: the distance of the negative charge from the oxygen and the geometry of the molecule. The position of the second peak is determined by the HOH angle of the model. The smaller HOH angle of TIP4Ptype models, relative to the tetrahedral for SPC-type geometries, separates these models into two groups. This means that, for simple geometrical reasons, the nonbonded OH vectors of the SPC/E molecule are closer to the OOO plane than that of TIP4P-type models. It is known that in condensed phases the actual OH bond length and the correct HOH angle are somewhere between these two value pairs. In ambient liquid water, the average bond length and the angle should be around 0.9798 and 107 , respectively.15 The SPC/E model might be elegant but denitely incorrect with respect to its geometry (certainly, on the other side, this is also the case for the TIP4P geometry). The tetrahedral angle and the long OH bond enhance the hydrogen bond network in liquid water. Thus, one might argue that the need for longer OH bond and tetrahedral angle is to compensate for the lack of polarization. Therefore, the simplicity of the SPC/E geometry is not only attractive but also serves a useful purpose. To rene the rough picture given by Fig. 2, we carried out a selection among the trimers. We chose only trimers where all the three hydrogen bonds were present. In fact, we selected trimers topologically similar to the trimers of Fig. 1. We did this by calculating the scalar product of OH bonds with the corresponding OO vector. We identied three OH bonds with the largest scalar products as hydrogen-bond forming vectors. During this process we omitted structures when one molecule provided two protons into the hydrogen bond structure. We dropped situations when the direction of the OH vector was very close to the direction of the OO vector, but the former pointed away from the trimer. In Fig. 3 we show a schematic diagram of a selected trimer. For clarity we do not show oxygen atoms. The normal vector of the plane is shown by an arrow which is shifted to the position of an oxygen to show the calculated angles more clearly.

In Fig. 4 we show the distribution of the angle between the six OH vectors and the normal vector of the OOO plane for the selected trimers. The difference between Figs. 2 and 4 is that the latter is more structured, because we eliminated trimers topologically different from the gas-phase trimer. The following step was to separate the bonding and nonbonding OH vectors shown together in Fig. 4. For us the angle of the nonbonding OH vectors with the normal vector of the OOO plane was decisive (white circles in Fig. 3). In Fig. 5 one can see the high angle part of these distributions. If the angle between the nonbonding OH vector and the normal vector of the OOO plane is 90 , the actual nonbonding OH vector is in the OOO plane (this is almost the case for the hydrogen marked by the broad arrow in Fig. 3). It is clear from Fig. 5 that the SPC/E and the TIP3P are the interaction forms where the largest part of trimers is close to planar structures. Seeing that the less planar is the TIP4P_mod structure where the negative charge is 0.25 from the oxygen, we can conclude that the major factor determining the structure of the trimer is the position of the negative charge. If the negative charge is shifted from the oxygen toward the hydrogen, the trimer loses its tendency to form a planar structure.

FIG. 4. Distribution of the angle between the normal vector of the OOO plane and the nonbonding OH vectors from the selected samples. See text for details.

Author complimentary copy. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

054106-4

P. T. Kiss and A. Baranyai

J. Chem. Phys. 134, 054106 (2011)

Ih, it becomes obvious that the hydrogen bond network never connects trimers. While two hydrogen bonds can be properly positioned, the third points actually toward a fourth molecule. Nevertheless, for the majority of the selected trimers the gas-phaselike geometry held. Therefore, the geometry of the samples did not distort our results.

III. CONCLUSION

FIG. 5. Distribution of the angle between the normal vector of the OOO plane and the OH bond vectors. Selection was made for gas-phase clusterlike structures. See text for details.

This result is not surprising. In polarizable models, performing well from the gas phase to hexagonal ice,6, 10 it is characteristic that the negative charge is placed on the main axis of the molecule at certain distance from the oxygen atom. It was motivated primarily to approximate the quadruple moment of the gas-phase molecule (in our case this corresponds roughly to the TIP4P_mod potential). While there is certain amount of variation in the actual distance of the negative charge from the oxygen atom in different models,6, 10 this structure proved to work well all over the phase diagram.10 In Fig. 6 we show the distribution of the angles between OH bonds and the corresponding OO vectors for the SPC/E and the TIP4P models (in Fig. 3 the double-headed arrow indicates such an angle between the vector of the dark hydrogen and the corresponding OO vector). The distributions show that the hydrogen bonds lined up along the OO vectors with reasonable accuracy with the exception of a small peak at 55 . This small peak is an indication of the slightly tetrahedral character of the water. If one looks at the structure of ice

We do not have reliable experimental information about the many-body structure of liquid water. There is only an indirect way to qualify the models by calculating more and more measurable property as an average and evaluate the overall performance of the model. Acknowledging the problems in solid phases4 and in dilute gas,6 we showed that the manybody structure of the SPC/E model is different from that of TIP4P-type models even in liquid state. The difference is not big but accurately measurable. The discrepancy is caused primarily by the position of the negative charge and enhanced a bit by the geometry of the molecule. This is also true for the TIP3P model. It seems that water models work better if the negative charge is positioned at a certain distance from the oxygen, on the main axis of the molecule. Since the discrepancy in character is similar to the difference we found in gas-phase clusters, where the SPC/E and TIP3P clusters were proven to show erroneous geometry, we state with certainty that the actual many-body structure of real water is more similar to models created by TIP4P-type structures than to models created by SPC/E or TIP3P geometry. Based on pioneering attempts, there are new models exempt from these deciencies, to describe the force elds in water with reliable accuracy.16, 17 Many of the new models are even polarizable which is more time-consuming to calculate but with the increasing power of new computers this is no longer prohibitively expensive.9, 10, 18, 19 These models provide a chance to describe the properties of water all over the phase diagram with acceptable accuracy.10 Although the SPC/E was an excellent pioneering model, we believe that it is time to use other more trustable force elds to describe water in molecular details.

ACKNOWLEDGMENTS

A.B. gratefully acknowledges the support of OTKA (Grant No. K67783).

J. C. Berendsen, J. R. Grigera, and T. P. Straatsma, J. Chem. Phys. 91, 6269 (1987). 2 C. Vega, private communication (2010). 3 B. Guillot, J. Mol. Liq. 101, 219 (2002). 4 C. Vega, E. Sanz, J. L. F. Abascal, and E. G. Noya, J. Phys. Condens. Matter 20, 153101 (2008). 5 W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L. Klein, J. Chem. Phys. 79, 926 (1983). 6 P. T. Kiss and A. Baranyai, J. Chem. Phys. 131, 204310 (2009). 7 N. Pugliano and R. J. Saykally, Science 257, 1937 (1992). 8 S. S. Xantheas, C. J. Burnham, and R. J. Harrison, J. Chem. Phys. 116, 1493 (2002). 9 P. Paricaud, M. Psedota, A. A. Chialvo, and P. T. Cummings, J. Chem. Phys. 122, 244511 (2005). 10 A. Baranyai and P. T. Kiss, J. Chem. Phys. 133, 144109 (2010).
1 H.

FIG. 6. Distribution of angle between the OO vector and the corresponding OH bond. See text for details.

Author complimentary copy. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

054106-5
11 A.

Molecular models of water

16 H.

J. Chem. Phys. 134, 054106 (2011) W. Horn, W. C. Swope, J. W. Pitera, J. D. Madura, T. J. Dick, G. L. Hura, and T. Head-Gordon, J. Chem. Phys. 120, 9665 (2004). 17 J. L. F. Abascal and C. Vega, J. Chem. Phys. 123, 234505 (2005). 18 G. Lamoureux, E. Harder, I. V. Vorobyov, B. Roux, and A. D. MacKerell, Chem. Phys. Lett. 418, 245 (2006). 19 A. P. E. Kunz and W. F. van Gunsteren, J. Phys. Chem. A 113, 11570 (2009).

Baranyai and D. J. Evans, Phys. Rev. A 40, 3817 (1989); ibid. 42, 849 (1990); Z. Naturforsch. 40a, 27 (1991). 12 Cs. Peltz, A. Baranyai, A. A. Chialvo, and P. T. Cummings, Mol. Simul. 29, 13 (2003). 13 C. J. Tsai and K. D. Jordan, J. Phys. Chem. 97, 11227 (1993); Chem. Phys. Lett. 213, 181 (1993). 14 A. A. Chialvo, personal communication (October 10, 2005). 15 V. F. Petrenko and R. W. Withworth, Physics of Ice (Oxford University Press, New York, 1999).

Author complimentary copy. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp