next up previous contents
Next: Pathways for Lennard-Jones Clusters Up: Characterizing the Topography of Previous: Traversing the Potential Energy Home: Return to my homepage

The PES of the 13-atom Morse Cluster

 As we have noted in our consideration of Morse clusters in earlier chapters, decreasing the range of the potential increases the ruggedness of the PES. This effect is particularly noticeable in the number of stationary points on the PES (Table 2.1). Here we hope to quantify these ideas further by analysing the topologies of the M13 PES in detail. Given the ubiquity of rugged energy landscapes in a diverse range of fields--spin glasses, superconductors, protein folding and optimization problems such as the infamous travelling salesman problem--the results may be of wide interest[285].

The samples of M13 stationary points were produced by the method described in the previous section. For $\rho_0$=3, 4, 6, 10 and 14 we used as our values of nev 15, 15, 6, 3 and 2, respectively. The decreasing number of eigenvectors searched at shorter range was necessary because of the escalating requirements for computer time and disk space to generate and store the data for such large samples of stationary points. Consequently, the incompleteness of the samples is likely to increase with $\rho_0$and be greater for the transition states than minima. The size of the samples is given in Table 2.1 and some of the average properties are catalogued in Table 6.1. Some properties have a fairly simple dependence on $\rho_0$.For example, the increase in the mean vibrational frequency as the range decreases correlates with the increased curvature of the pair potential at its minimum (Figure 2.2).


 
Table 6.1: Some average properties of the M13 stationary point samples at $\rho_0$=3, 4, 6, 10 and 14. $\overline\nu$ is the geometric mean frequency, nts is the number of geometrically distinct transition states connected to a minimum, ngm is the number of steps in minima space along the shortest path to the global minimum, Sgm is the total length in configuration space of the shortest pathway to the global minimum, $\omega$ is the magnitude of the imaginary frequency of a transition state, S is the length of a single rearrangement pathway, D is the displacement in configuration space between two connected minima, $\gamma$ is the moment ratio of displacement, bd is the smaller (downhill) barrier height, bu is the larger (uphill) barrier height, $\Delta E$ is the difference in energy between two connected minima ($b_u=b_d+\Delta E$), and nbasin is the number of basins on the PES. $\overline\nu$, nts, ngm and Sgm are averages over all minima; $\omega$, S, D and $\gamma$, are averages over all transition states; and bd, bu and $\Delta E$ are averages over all non-degenerate transition states.
  $\rho_0=3$ $\rho_0=4$ $\rho_0=6$ $\rho_0=10$ $\rho_0=14$
$h \overline\nu/\epsilon$ 0.216 0.340 0.682 1.843 3.680
nts 4.89 8.26 11.54 8.04 8.52
ngm 1.00 1.38 2.40 3.65 3.70
Sgm / r0 2.597 3.079 3.884 3.952 3.770
$h \omega/\epsilon$ 0.184 0.262 0.469 1.051 1.450
S/r0 3.012 2.905 1.874 1.119 1.083
D/r0 1.607 1.477 1.169 0.842 0.819
$b_{d}/\epsilon$ 0.114 0.483 0.545 0.585 0.786
$b_{u}/\epsilon$ 4.512 3.693 2.087 1.473 1.540
$\Delta E/\epsilon$ 4.398 3.210 1.542 0.888 0.786
$13/\gamma$ 7.161 5.269 4.748 5.051 4.889
nbasin 1 1 3 54 262

More interesting are those properties that give insight into the global topography of the PES. As can be seen from Figure 4.4 the energy gap between the global minimum, the 13-atom icosahedron, and the other low-lying minima decreases as $\rho_0$ is increased. This flattening of the surface is also apparent in the average values of $\Delta E$, the difference in energy between two connected minima; $\Delta E$ is effectively a local measure of the slope of the PES.[*] The averages of the barrier heights show that the barrier to go downhill (bd) increases with $\rho_0$.At $\rho_0$=4, the probability distribution of bd is dominated by the low barrier peak (Figure 6.1), but as $\rho_0$ increases the magnitude of this peak diminishes. The greater similarity between the distributions for bd and bu at larger $\rho_0$ is a reflection of the smaller value of $\Delta E$. We expect, based on the results of Chapter 5, that the combined effect of the larger downhill barriers and the flatter PES will be to make relaxation to the global minimum more difficult at shorter range.


  
Figure 6.1: Probability distributions of the barrier heights of uphill (dashed line) and downhill (solid line) rearrangements for M13 at $\rho_0$=4, 6, 10 and 14, as marked. Transition states mediating degenerate rearrangements have been excluded from the samples.
\begin{figure}
\epsfxsize=15cm
\centerline{\epsffile{mm.barrier.eps}}\end{figure}

In a comparison of the topographies of Lennard-Jones and potassium chloride clusters, it was noted that the rearrangements for potassium chloride clusters were much more cooperative[252]. It was suggested that this might have been one of the reasons for the greater ease with which the potassium chloride clusters could reach the global minimum. The moment ratio of displacement, $\gamma$, can be used to measure the cooperativity of a rearrangement. It is defined by[286]
\begin{displaymath}
\gamma = { N \sum_i \left[Q_i(s)-Q_i(t)\right]^4 \over
 \left( \sum_i \left[Q_i(s)-Q_i(t)\right]^2 \right)^2 }, \end{displaymath} (6.1)
where Qi(s) is the value of the nuclear Cartesian coordinate Qi for minimum s, etc. $\gamma$ was evaluated using the relative positions of the two minima at the end-points. If only a single atom moves then $\gamma=N$ and the rearrangement is strongly localized, while if all atoms move through the same distance then $\gamma=1$ and the process is completely cooperative. $N/\gamma$ therefore can be interpreted as the number of atoms involved in the rearrangement. Somewhat surprisingly, apart from the greater cooperativity at $\rho_0$=3, $\gamma$ does not depend significantly on the range of the potential (Table 6.1). This is in contrast to a study which compared rearrangements of LJ55 and (C60)55 clusters, which found that cooperative rearrangements were much less likely for (C60)55 clusters[90]. This difference may reflect the fact that the structure of small clusters is less dependent on the range than for larger clusters.


  
Figure 6.2: Histograms of ngm, the number of steps a minimum is away from the global minimum, for M13 at $\rho_0$=4, 6, 10 and 14, as marked.
\begin{figure}
\epsfxsize=15cm
\centerline{\epsffile{mm.ngm.eps}}\end{figure}

The shortest reaction pathway to the global minimum was calculated for each minimum on the PES. If we look at ngm, the number of steps between minima along one of these pathways, it is clear that ngm increases as the range decreases (Figure 6.2). At $\rho_0$=3 all the minima are directly connected to the global minimum, yet at $\rho_0$=14, a minimum can be as far as eight steps away. The ngm histograms also show that the connectivity patterns for the M13 PES are quite different from the standard model PES that we considered in Chapter 5 for which the number of minima increased exponentially with the number of steps away from the global minimum. It is also interesting to note the large number of minima that are directly connected to the global minimum. For example, at $\rho_0$=6 we found 568 geometrically distinct transition states which connect the icosahedron to 213 geometrically distinct minima. If one takes into account the reaction path degeneracy, the number of transition states surrounding a single permutational isomer of the icosahedron is at least $67\,836$.


  
Figure 6.3: Shortest pathways to the global minimum for each minimum of M13 at $\rho_0$=4, 6, 10 and 14 as marked. The reaction coordinate is the number of steps away from the global minimum. Half integer values correspond to transition states.
\begin{figure}
\epsfxsize=15cm
\centerline{\epsffile{mm.gmpaths.eps}}\end{figure}

In Figure 6.3, we show the shortest reaction path to the global minimum for each minimum. Again the larger barrier heights and the increased flatness of the PES are apparent at short range. At $\rho_0$=14 there are clearly some pathways for which the minima do not monotonically decrease in energy as the global minimum is approached. This will be considered further when we examine the basin structure of the PES.


  
Figure 6.4: Probability distributions of Sgm, the length of the shortest pathway from a minimum to the global minimum, for M13 at $\rho_0$=4, 6, 10 and 14 as marked.
\begin{figure}
\epsfxsize=15cm
\centerline{\epsffile{mm.Sgm.eps}}\end{figure}

Although ngm increases with $\rho_0$ the average total length of the reaction paths in configuration space (Sgm) is relatively constant for $\rho_0\ge 6$, and for smaller $\rho_0$ shows a much weaker dependence on the range than ngm (Table 6.1 and Figure 6.4). This is not too surprising if we consider the dependence of the path length for a single rearrangement on $\rho_0$: this length decreases significantly as the range is reduced (Figure 6.5). The pathways simply become more rugged rather than longer: more transition states must be crossed.


  
Figure 6.5: Probability distributions of S, the length of a single rearrangement pathway, for the transition states of M13 at $\rho_0$=4, 6, 10 and 14 as marked.
\begin{figure}
\epsfxsize=10cm
\centerline{\epsffile{mm.Sts.eps}}\end{figure}

The topology of the surface was further analysed in terms of the structure of the basins on the PES. For $\rho_0$=3 and 4, every minimum lies on a sequence of minima which are monotonically decreasing in energy and lead down to the global minimum; the PES consists of a single basin. For $\rho_0$=6 all but two of the minima lie on such monotonic sequences. By Kunz and Berry's definition these two minima also represent basin bottoms, albeit very shallow; only one step needs to be taken to reach a minimum in the basin surrounding the icosahedron. For $\rho_0$=10 and 14 many more shallow basins appear on the PES (Table 6.1). Despite this increase in the number of basins, 96.8% of the minima located are in the icosahedral basin for $\rho_0$=14 (Figure 6.6e).

One consequence of the definition of a basin is that it is possible for a minimum to lie on monotonic sequences which lead down to the bottom of different basins. Indeed high energy minima may lie in numerous basins (Figure 6.6f). In this regard the division of minima space into basins differs from the division of configuration space into wells surrounding minima, since there is no equivalent of the steepest descent path which uniquely maps each point in configuration space to a minimum. Instead the division of minima space is more like a Venn diagram where overlap can occur between regions.


  
Figure 6.6: `Tree' diagrams showing the basin structure for M13 at (a) $\rho_0$=6, (b) $\rho_0$=10, and (c) $\rho_0$=14. The endpoints are at the energies of the minima at the bottoms of the basin. Nodes occur when two basins or metabasins become connected; The y coordinate of the node corresponds to the energy of the lowest energy minimum which lies in both (meta)basins, and the x coordinate to the average energy of the all the basin bottoms in the resulting metabasin. Scatter plots of (d) the energy (relative to the bottom of a basin) of the lowest energy minimum which is in both the basin concerned and a lower energy basin against the basin energy, (e) the number of minima in a basin against the basin energy and (f) the number of basins a minimum belongs to against the energy of the minimum for M13 at $\rho_0$=14
\begin{figure}
\epsfxsize=14cm
\centerline{\epsffile{mm.basin.eps}}\end{figure}

One method to analyse this basin structure is with `tree' diagrams (Figure 6.6); these are in some ways similar to the equilibration trees considered in Chapter 5. Two basins are considered to become `connected' and to form a new metabasin when the energy is that of the lowest energy minimum which lies in both basins. Each point in the tree diagram represents a basin or the creation of a new metabasin, and the nodes occurs at the energies at which the basins or metabasins become connected. The tree diagrams, therefore, follow the pattern of connections that occurs between basins and metabasins as the energy is increased.

It is noticeable that in the tree diagrams virtually all the basins are directly connected to the line corresponding to the metabasin of the global minimum, i.e. the basin which requires least energy to enter is the metabasin of the global minimum. The basin surrounding the global minimum is clearly the dominant feature even on the more rugged PES's for $\rho_0$=10 and 14. Furthermore, many of the lines joining a basin to the icosahedral basin are virtually horizontal implying that the depth of most basins (measured in terms of the energy, relative to the basin bottom, of the lowest energy minimum that is also in another basin) is small (Figure 6.6d). The average depths for $\rho_0$=10 and 14 are 0.237 and $0.067\,\epsilon$, respectively. Most of the basins are not sufficiently deep to act as effective traps. The lower average depth for $\rho_0$=14 results from the large number of very shallow basins at high energy; 163 of the 262 basins lie above $-33.5\,\epsilon$.

Although the PES's at $\rho_0$=10 and 14 involve a large number of basins, these PES's do not equate with the multiple funnel scenarios outlined in §5.4 since the basins are too shallow for them to be categorized as separate funnels. However, their roughness is still likely to hinder relaxation to the global minimum, even if not to as large an extent as multiple funnels.


next up previous contents
Next: Pathways for Lennard-Jones Clusters Up: Characterizing the Topography of Previous: Traversing the Potential Energy Home: Return to my homepage
Jon Doye
8/27/1997