1887
banner image
No data available.
Please log in to see this content.
You have no subscription access to this content.
No metrics data to plot.
The attempt to load metrics for this article has failed.
The attempt to plot a graph for these metrics has failed.
Comparison of free energy methods for molecular systems
Rent:
Rent this article for
USD
10.1063/1.2378907
/content/aip/journal/jcp/125/18/10.1063/1.2378907
http://aip.metastore.ingenta.com/content/aip/journal/jcp/125/18/10.1063/1.2378907

Figures

Image of FIG. 1.
FIG. 1.

The slope of the free energy as a function of for changing the Lennard-Jones size of a neutral particle in a box of explicit water. Results for both TI and AIM methods are shown for dynamics steps. The data show the averages (data points) and standard deviations (error bars) from 16 independent simulations for each method. The figure demonstrates that AIM has the ability to sample the path more efficiently, thus producing a much smoother and more precise profile compared to TI. Thus, AIM is preferred over TI for computing the potential of mean force for this system. In addition, the smoothness of the profile suggests that the switching function of Eq. (25) used in this report is adequate.

Image of FIG. 2.
FIG. 2.

(a) “Fast-growth” unidirectional free energy difference estimates obtained for changing the Lennard-Jones size of a neutral particle in a box of explicit water. Results are shown for both SEPS and Jarz methods as a function of the number of dynamics steps used in the simulation. For both methods, fast-growth work values were generated by simulating roughly 2000 dynamics steps per path, which is ten times shorter than optimal. The solid horizontal line represents the best estimate of the free energy difference based on averaging all results shown in Table I at dynamics steps. The averages (data points) and standard deviations (errorbars) are from 16 independent simulations. (b) Histograms of the work values used to generate the free energy estimates for both the SEPS and Jarz methods. The plots demonstrate the potential usefulness of using path sampling over regular Jarzynski averaging. Specifically, if the work values are fast growth and unidirectional, then SEPS is able to bias the work values in such a way to improve the free energy estimate. Note that for all the SEPS data shown, the first 50 work values are thrown away for equilibration, as described in Sec. IV E.

Image of FIG. 3.
FIG. 3.

The slope of the free energy as a function of for a changing the charge of a Lennard-Jones particle in a box of explicit water from to . Results for both TI and AIM methods are shown for dynamics steps. The data show the averages (data points) and standard deviations (error bars) from 16 independent simulations for each method. The differences between TI and AIM are too small to resolve on the plot shown; however, it should be noted that the average uncertainty in the data for AIM is and for TI is , suggesting that AIM has the ability to produce a more precise profile compared to TI. Thus, AIM is preferred over TI for computing the potential of mean force for this system. The smoothness of the profile also suggests that the switching function of Eq. (24) used in this report is adequate.

Tables

Generic image for table
Table I.

Free energy difference estimates in units of kcal/mol obtained for changing the Lennard-Jones size of a neutral particle in a box of explicit water. Results are shown for various methods described in the text as a function of the number of dynamics steps used in the simulation. Table entries are the mean estimates from 16 independent simulations with the standard deviation shown in parentheses. For single-ensemble path sampling (SEPS and BSEPS) and Jarzynski methods (Jarz and BJarz), only the most efficient results are shown. The table shows that in the limit of long simulation times ( dynamics steps) all methods produce average estimates that roughly agree. The table also shows that AIM provides the most precise long-simulation estimate.

Generic image for table
Table II.

Number of dynamics steps necessary to be within a specified tolerance of the correct result , average estimate at dynamics steps for all methods, for growing a Lennard-Jones particle in explicit solvent. The first column is the method used to obtain the estimate. The second column is the number of dynamics steps needed to estimate within of with an uncertainty less than . The third column is the number of dynamics steps needed to obtain an estimate within with an uncertainty less than .

Generic image for table
Table III.

Free energy difference estimates in units of kcal/mol obtained for changing the charge of a Lennard-Jones particle from to in a box of explicit water. Results are the averages from 16 independent simulations for various methods described in the text as a function of the number of dynamics steps used in the simulation. The standard deviation is shown in parentheses. For single-ensemble path sampling (SEPS and BSEPS) and Jarzynski methods (Jarz and BJarz), only the most efficient results are shown. The table shows that in the limit of long simulation times ( dynamics steps) all methods produce average estimates that roughly agree. The table also shows that AIM and BJarz approaches provide the most precise long-simulation estimate.

Generic image for table
Table IV.

Number of dynamics steps necessary to be within a specified tolerance of the correct result , average estimate at dynamics steps for all methods, for charging a Lennard-Jones particle in explicit solvent. The first column is the method used to obtain the estimate. The second column is the number of dynamics steps needed to estimate within of with an uncertainty less than . The third column is the number of dynamics steps needed to obtain an estimate within with an uncertainty less than .

Generic image for table
Table V.

Number of dynamics steps necessary to be within of the analytical result for with a or less standard deviation for the two-dimensional model in Ref. 10. The first column is the barrier height of the potential energy surface in units. The second and third columns are the total numbers of dynamics steps using SEPS with, respectively, 200 work values and 20 000 work values. The fourth column is the total number of dynamics steps using TI with using 51 equally spaced values of . For both TI and SEPS, half of the generated data was thrown away for equilibration.

Loading

Article metrics loading...

/content/aip/journal/jcp/125/18/10.1063/1.2378907
2006-11-14
2014-04-18
Loading

Full text loading...

This is a required field
Please enter a valid email address
752b84549af89a08dbdd7fdb8b9568b5 journal.articlezxybnytfddd
Scitation: Comparison of free energy methods for molecular systems
http://aip.metastore.ingenta.com/content/aip/journal/jcp/125/18/10.1063/1.2378907
10.1063/1.2378907
SEARCH_EXPAND_ITEM