Density functional theory for Baxter’s sticky hard spheres in confinement

Density functional theory for Baxter’s sticky hard spheres in confinement

Hendrik Hansen-Goos Yale University, New Haven, Connecticut 06520, USA    Mark A. Miller University Chemical Laboratory, Lensfield Road, Cambridge CB2 1EW, UK    J.S. Wettlaufer Yale University, New Haven, Connecticut 06520, USA NORDITA, Roslagstullsbacken 23, SE-10691 Stockholm, Sweden
July 2, 2019

It has recently been shown that a free energy for Baxter’s sticky hard sphere fluid is uniquely defined within the framework of fundamental measure theory (FMT) for the inhomogeneous hard sphere fluid, provided that it obeys scaled-particle theory and the Percus-Yevick (PY) result for the direct correlation function [Hansen-Goos and Wettlaufer, J. Chem. Phys. 134, 014506 (2011)]. Here, combining weighted densities from common versions of FMT with a new vectorial weighted density, we derive a regularization of the divergences of the associated strongly confined limit. Moreover, the simple free energy that emerges is exact in the zero-dimensional limit, leaves the underlying equation of state unaffected, and yields a direct correlation function distinct from the PY expression. Comparison with simulation data for both the bulk pair correlation function and the density profiles in confinement shows that the new theory is significantly more accurate than the PY-based results. Finally, the resulting free energy is applicable to a glass of adhesive hard spheres.

61.20.Gy, 82.70.Dd, 64.60.ah

Colloidal suspensions are known to exhibit a variety of collective phenomena such as crystallization, gel formation, glass transition and percolation BuRuPi07 (); Luetal08 (); EbWaCa11 (). Unlike in molecular systems these properties can be measured not only in terms of bulk sampling such as x-ray diffraction, heat capacity measurements, or mechanical properties, but also on a microscopic level thanks to optical single-particle observation techniques such as confocal microscopy. An ideal theoretical tool to study the structural properties down to the particle size is classical density functional theory (DFT) Ev79 (), an excellent example application being the purely entropic hard-sphere system. Indeed, within the framework of fundamental measure theory (FMT) Ro89 (); KiRo90 (); KiRo93 (), for arbitrary external potentials both the inhomogeneous liquid and the crystalline phases are described by a simple and accurate free energy. However, a more realistic model of colloidal suspensions must account for attractive interactions. Such a model is provided by Baxter’s sticky hard spheres (SHS) which interact via an infinitely narrow and deep attractive well, the limit being taken such that the second virial coefficient remains finite Ba68 (). The model is particularly appealing because it is soluble analytically within the Percus Yevick (PY) approximation. The PY results obtained by Baxter provide a framework to interpret experimental phase diagrams of colloidal suspensions, including the liquid-liquid transition, crystallization, percolation and, to some extent, the glass transition, revealing the existence of attractive and repulsive glasses Phetal02 (). Quantitative interpretation of single-particle observations in colloidal suspensions rely on a theoretical approach capable of resolving the microscopic structure of Baxter’s SHS as accurately as does FMT for hard spheres. Previous DFT approaches relied on Taylor expansions of the free energy around the homogeneous bulk fluid which use the PY result for the direct correlation function while crudely approximating higher order terms. Recently, using a set of weighted densities from FMT for the hard-sphere fluid HaWe11 (), a first non-perturbative DFT was derived. When applied to density profiles near a planar wall, or the SHS confined between two parallel planar hard walls, the theory was shown to be a significant improvement over previous approaches. However, based on experience with Rosenfeld’s original FMT for the hard-sphere fluid, it is expected that the FMT for the SHS would exhibit an unphysical divergence for strongly confined fluids, the most extreme test being the zero-dimensional (0D) limit.

In this Letter, we recast the original FMT for the SHS, which is based on Kierlik’s and Rosinberg’s scalar weighted densities KiRo90 (); KiRo93 (), by combining Rosenfeld’s vectorial weighted density Ro89 () with a new vectorial weighted density. In consequence, we identify the origin of the divergence and regularize it in the spirit of Schmidt et al.’s approach to Rosenfeld’s FMT for the hard-sphere system Roetal97 (). Rather than deriving a cumbersome expression by conserving the underlying PY correlation function, we seek a simple density functional. When the stickiness of the particles is strong, the modified theory (exact in the 0D limit) provides a better description of the pair-correlation function than does the PY-based FMT, while the accurate results at moderate stickiness are left virtually unchanged. For the SHS confined in a spherical cavity, the new theory is significantly better than its predecessor. Our free energy can be used to study a wide range of the structural properties of inhomogeneous colloidal suspensions with short-ranged attractions. In particular, this is the first free energy applicable to the glassy state where confinement due to caging is ubiquitous. In fact, it may overcome previous theoretical pathologies associated with Baxter’s adhesive hard-sphere treatment of the glass transition. These problems were based on using the PY direct correlation function alone Foetal00 (), rather than the complete free energy provided here.

Consider a fluid of monodisperse hard spheres with radius . We define the locally averaged packing fraction with weighted densities and . Rosenfeld’s excess free energy density (measured in units of ) is then with , , and Ro89 (). While has been extremely successful in describing an unconfined fluid, with an inhomogeneity caused, say, by a planar wall, it is problematic when the fluid is highly confined (see Ta00 () and refs. therein). For the most extreme confinement, the density distribution becomes simply a delta-peak, i.e., , where , and is the packing fraction (). The weighted densities for this distribution are , , and , where . The integration of can be readily performed yielding which is the exact result for a 0D cavity holding an average number of particles Roetal97 (). Examining further for such a cavity shows that vanishes (as it should) while exhibits a strong negative divergence. This is due to the factor in front of the vector term in which would have to be unity for its scalar and vectorial terms to cancel. However, the resulting expression for the excess free energy does not yield the PY direct correlation function, thereby destroying the excellent experimental agreement of Rosenfeld’s for the HS fluid. Several routes have been devised to regularize . A particularly appealing approach consists of introducing a new weighted density that reduces to in the bulk fluid (where the vectorial weighted densities vanish as is spatially uniform), and vanishes in the 0D limit (where ). Using , a regularized excess free energy density proposed by Schmidt et al. Roetal97 () is


From the properties of it is immediately clear that is exact for a 0D cavity while the free energy of the bulk fluid is unaffected. However, the third term differs by order from that of , which is very small for moderately inhomogeneous fluids. In fact, it affects only the direct correlation functions of order 4 and higher whereas the PY direct correlation function (which is second order) is preserved. Both unconfined and confined fluids are extremely well described by ; accurately predicting solid-liquid coexistence in hard sphere systems whereas dramatically overstabilizes the crystal due to the negative divergence for strongly peaked density distributions. However, Tarazona’s modification of , which uses a tensorial weighted density Ta00 (), performs better with regard to the width of the density peaks in the crystal.

How does the picture change as we introduce Baxter’s attractive surface interaction into the system? Rosenfeld’s set of weighted densities (i.e. , , and ) is known to be insufficient for constructing any reasonable free energy density for the SHS fluid because they cannot yield the -function within the direct correlation function HaWe11 (). However, when and are supplemented by the weighted density , the set is sufficient to construct a free energy density for Baxter’s SHS fluid where is uniquely defined by (a) requiring the density functional to yield the PY direct correlation function and (b) imposing consistency with scaled particle theory HaWe11 (). The dimensionless coefficients and are functions of , which in the bulk fluid is related to the packing fraction via . They are obtained by solving and , where the integration constants are chosen such that and vanish for . Here , where is the cavity function at contact, and is obtained as the smaller of the two solutions to .

In order to regularize the divergences in the 0D limit we rewrite using vectorial weighted densities so that the resulting functional vanishes. This guarantees that is exact for a cavity that can hold only one particle at a time, and thus the corresponding free energy is unaffected by the sticky interaction. We use the identity KiRo93 () in , and for reasons that will become clear below, we split the term in half and substitute for in only one of the terms. Moreover, because is always integrated on , we can apply Green’s first identity and neglect the boundary integral. The result is , where and . The coefficients , , , and are functions of , and the new vectorial weighted density is .

Figure 1: Coefficient functions in the attractive part of the excess free energy density for SHS with different Baxter parameters . In the coefficients are functions of , which in the homogeneous bulk fluid is related to the packing fraction via . The non-divergent expression , Eq. (2), relies on replacing by and by in .

Replacing only in half of the split term has two important consequences. Firstly, it insures that the function remains finite in the limit . Secondly, it leads to a symmetry between and that is reflected in their equality in the limits and , where and . However, comparing with for any reveals their differences (Fig. 1). When we replace with , is unaffected in the bulk fluid. Above we calculated the weighted densities , and in the 0D limit. These are now supplemented by , and , from which we have . Hence, this modification regularizes . Note, that because vanishes as for then even if and take the same values in this limit it is insufficient to insure that vanishes in the 0D limit, where large values of occur due to the delta-function in . Furthermore, larger powers of the vectorial weighted densities could be used to modify the terms and individually while maintaining for example the PY direct correlation function which underlies . However, the symmetry of and motivates the simple remedy introduced here. The value of giving up Baxter’s PY result for the direct correlation function is assessed in terms of the performance of the modified functional in comparison to simulations.

The regularization of requires two steps. We replace by and note that even for asymptotic values of the argument , while ; and while , these function differ (see Fig. 1). Therefore, our substitution is justified by the fact that it becomes exact in the limit of low density () and weak “stickiness” (). Like , in the bulk fluid remains unaffected by the substitution.

After replacing by , is still divergent. This is a consequence of the factor in front of the vectorial term, which comes from matching the and limits of and . In analogy to where the term is regularized by replacing it with we now regularize by substituting . Introducing the modified weighted density (which equals in the bulk fluid and vanishes in the 0D limit) we can write down the regularized version of as


where and are functions of .

Figure 2: Pair-correlation function of a SHS fluid with and from MC simulations (solid line) compared to results from the PY-based FMT (dotted line) and the regularized FMT (dashed line).

As a first test of this regularized free energy density we calculate the pair correlation function by minimizing the density functional , where is the functional for the ideal gas, is the chemical potential and is the external potential, chosen so that it represents a particle of the SHS fluid at position . The resulting density profile gives , where is the bulk fluid density. In Fig. 2 we show the case of strong adhesion, , and compare the FMT results with MC simulations obtained using the algorithm described in MiFr04 (). The regularized is superior to the PY-based , which is remarkable because has been optimized for the 0D limit and not at all for . Other, larger , MC simulations for (see MiFr04b ()) are equally well described by the two FMTs, which give very similar results for .

Figure 3: The percolation line as obtained from MC simulations Le01 () compared to the PY result (dashed line) and the results based on the non-diverging FMT constructed here (solid line).

Alternatively, an analytical expression for the direct correlation function can be obtained from the excess free energy through


where and run through and . The partial derivatives of are evaluated in the bulk fluid and derivatives with respect to vectorial quantities and are executed formally while the fact that and in bulk ensures that the resulting expressions are scalars. The symmetric coefficients are convolutions of the weight functions that are associated with the respective weighted densities. Therefore for . For example, is the overlap volume of two spheres with radius and center-to-center distance . Moreover, , , , , , and are required in Eq. (3).

In particular, the coefficient of the -function in is . This result can be used to calculate the percolation line which, under the approximations described in ChGl83 (), is . As long as is not too large, both the classic PY result and the modified FMT describe the percolation threshold as determined in simulations, the latter being somewhat more reliable in the range of intermediate (Fig. 3).

Figure 4: Density profiles for sticky hard spheres with Baxter parameter and in a spherical cavity of radius . The FMT results of the previous PY-based FMT and the present regularized FMT were obtained in the grand canonical ensemble and transformed to the canonical ensemble (with correction of order ) for a comparison with canonical MC simulations that were performed using the algorithm of MiFr04 ().

Finally, we consider a confinement scenario for which our theory is expected to be the most relevant. We compare the previous PY-based FMT and the present regularized FMT to canonical MC simulations for a SHS fluid with and in a spherical cavity of radius . This system was considered previously Goetal97 () for a hard-sphere fluid (). To compare the FMTs with the simulations we have to transform the former from the grand canonical ensemble to the canonical ensemble which can be achieved up to corrections of order by using the scheme of Goetal97 () that is based on results from Saetal96 (). Fig. 4 shows that the regularized FMT is clearly superior to the previous PY-based FMT, especially in the center of the spherical cavity. Interestingly, while the previous FMT underestimates the density in the center when is small, it overestimates the density in the center when applied to hard spheres; large (see Goetal97 ()). The crossover occurs around where the regularized and the PY-based FMT yield very similar results. In the limit of large cavity radius , planar wall density profiles are recovered; comparison (not shown) with existing MC simulations JaBr94 () shows that the PY-based and the regularized FMT describe the density profiles equally well. In particular, the theories yield identical contact values for the density at a hard wall. This results from being equal to in bulk and hence the underlying pressure being the PY compressibility result for both theories. Given that both FMTs obey the contact theorem the contact values of the density must be identical. This illustrates the importance of preserving the bulk fluid properties while removing divergences for highly peaked density distributions in order to obtain a density functional theory that is robust and accurate for a broad spectrum of settings.


  • (1) S. Buzzaccaro, R. Rusconi, and R. Piazza, Phys. Rev. Lett. 99, 098301 (2007).
  • (2) P.J. Lu, E. Zaccarelli, F. Ciulla, A.B. Schofield, F. Sciortino, and D.A. Weitz, Nature 453, 499 (2008).
  • (3) A.P.R. Eberle, N.J. Wagner, and R. Castañeda-Priego, Phys. Rev. Lett. 106, 105704 (2011).
  • (4) R. Evans, Adv. Phys. 28, 143 (1979).
  • (5) Y. Rosenfeld, Phys. Rev. Lett. 63, 980 (1989).
  • (6) E. Kierlik and M.L. Rosinberg, Phys. Rev. A 42, 3382 (1990).
  • (7) S. Phan, E. Kierlik, M.L. Rosinberg, B. Bildstein, and G. Kahl, Phys. Rev. E 48 618 (1993).
  • (8) R.J. Baxter, J. Chem. Phys. 49, 2770 (1968).
  • (9) K.N. Pham, A.M. Puertas, J. Bergenholtz, S.U. Egelhaaf, A. Moussaïd, P.N. Pusey, A.B. Schofield, M.E. Cates, M. Fuchs, and W.C.K. Poon, Science 296, 104 (2002).
  • (10) H. Hansen-Goos and J.S. Wettlaufer, J. Chem. Phys. 134, 014506 (2011).
  • (11) Y. Rosenfeld, M. Schmidt, H. Löwen, and P. Tarazona, Phys. Rev. E 55, 4245 (1997).
  • (12) G. Foffi, E. Zaccarelli, F. Sciortino, P. Tartaglia, and K.A. Dawson, J. Stat. Phys. 100, 363 (2000).
  • (13) P. Tarazona, Phys. Rev. Lett. 84, 694 (2000).
  • (14) M.A. Miller and D. Frenkel, J. Chem. Phys. 121, 535 (2004).
  • (15) M.A. Miller and D. Frenkel, J. Phys.: Condens. Matter 16, S4901 (2004).
  • (16) S.B. Lee, J. Chem. Phys. 114, 2304 (2001).
  • (17) Y.C. Chiew and E.D. Glandt, J. Phys. A: Math. Gen. 16, 2599 (1983).
  • (18) A. González, J.A. White, F.L. Román, S. Velasco, and R. Evans, Phy. Rev. Lett. 79, 2466 (1997).
  • (19) J.J. Salacuse, A.R. Denton, and P.A. Egelstaff, Phys. Rev. E 53, 2382 (1996).
  • (20) A. Jamnik and D. Bratko, Phys. Rev. E 50, 1151 (1994).
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
The feedback must be of minimum 40 characters and the title a minimum of 5 characters
Add comment
Loading ...
This is a comment super asjknd jkasnjk adsnkj
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test description