Limits...
Relative Entropy and Optimization-Driven Coarse-Graining Methods in VOTCA.

Mashayak SY, Jochum MN, Koschke K, Aluru NR, RΓΌhle V, Junghans C - PLoS ONE (2015)

Bottom Line: We illustrate the new methods by coarse-graining SPC/E bulk water and more complex water-methanol mixture systems.The CG potentials obtained from both methods are then evaluated by comparing the pair distributions from the coarse-grained to the reference atomistic simulations.In addition to the newly implemented methods, we have also added a parallel analysis framework to improve the computational efficiency of the coarse-graining process.

View Article: PubMed Central - PubMed

Affiliation: Department of Mechanical Science and Engineering, Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, Illinois, 61801, United States of America.

ABSTRACT
We discuss recent advances of the VOTCA package for systematic coarse-graining. Two methods have been implemented, namely the downhill simplex optimization and the relative entropy minimization. We illustrate the new methods by coarse-graining SPC/E bulk water and more complex water-methanol mixture systems. The CG potentials obtained from both methods are then evaluated by comparing the pair distributions from the coarse-grained to the reference atomistic simulations. In addition to the newly implemented methods, we have also added a parallel analysis framework to improve the computational efficiency of the coarse-graining process.

No MeSH data available.


Computation times for the construction of neighbor lists: simple vs. grid search.Shown are the results for the Lennard-Jones fluid as a function of particles in the system. In addition to the data points for the simple and grid search algorithms, lines indicate the scaling law with 2 and 1 as the exponent, respectively. These exponents result from the cost of the simple and grid search algorithm: π“ž(N2) and π“ž(N), respectively. The cut-off for the neighbor search was set to 1.6Οƒ, which roughly corresponds to the first minimum in the radial distribution function.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4507862&req=5

pone.0131754.g003: Computation times for the construction of neighbor lists: simple vs. grid search.Shown are the results for the Lennard-Jones fluid as a function of particles in the system. In addition to the data points for the simple and grid search algorithms, lines indicate the scaling law with 2 and 1 as the exponent, respectively. These exponents result from the cost of the simple and grid search algorithm: π“ž(N2) and π“ž(N), respectively. The cut-off for the neighbor search was set to 1.6Οƒ, which roughly corresponds to the first minimum in the radial distribution function.

Mentions: First, we compare the performance of the simple search and grid search algorithms. For comparison, we measure the computational time for constructing neighbor-list for 1 frame on 1 core only. In Fig 3, the timing results for the neighbor-list creation as a function of the number of particles are shown. The computational times of the simple and grid search algorithms follow the expected scaling of π“ž(N2) and π“ž(N), respectively. The grid search algorithm shows a very good agreement with our expectation. However, the simple search algorithm performs worse than expected for the systems with 262636 and 492038 particles. The performance loss of the simple search algorithm is likely due to the large amount of data the algorithm has to process in the inner loops. For large data, cache memory limits may exceed causing bad scaling. An additional benchmark, not reported here, confirmed the dependence of the grid search algorithm on the neighbor cut-off distance, whereas the simple search is unaffected. However, even after increasing the cut-off from 1.6 to 2.6Οƒ, it is found that the grid search performs an order of magnitude faster than the simple search. The cross-over, where the simple search becomes faster than the grid search depends on the cut-off distance, but is typically too large for productive uses, and hence, the grid search should be preferred over the simple search in general.


Relative Entropy and Optimization-Driven Coarse-Graining Methods in VOTCA.

Mashayak SY, Jochum MN, Koschke K, Aluru NR, RΓΌhle V, Junghans C - PLoS ONE (2015)

Computation times for the construction of neighbor lists: simple vs. grid search.Shown are the results for the Lennard-Jones fluid as a function of particles in the system. In addition to the data points for the simple and grid search algorithms, lines indicate the scaling law with 2 and 1 as the exponent, respectively. These exponents result from the cost of the simple and grid search algorithm: π“ž(N2) and π“ž(N), respectively. The cut-off for the neighbor search was set to 1.6Οƒ, which roughly corresponds to the first minimum in the radial distribution function.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4507862&req=5

pone.0131754.g003: Computation times for the construction of neighbor lists: simple vs. grid search.Shown are the results for the Lennard-Jones fluid as a function of particles in the system. In addition to the data points for the simple and grid search algorithms, lines indicate the scaling law with 2 and 1 as the exponent, respectively. These exponents result from the cost of the simple and grid search algorithm: π“ž(N2) and π“ž(N), respectively. The cut-off for the neighbor search was set to 1.6Οƒ, which roughly corresponds to the first minimum in the radial distribution function.
Mentions: First, we compare the performance of the simple search and grid search algorithms. For comparison, we measure the computational time for constructing neighbor-list for 1 frame on 1 core only. In Fig 3, the timing results for the neighbor-list creation as a function of the number of particles are shown. The computational times of the simple and grid search algorithms follow the expected scaling of π“ž(N2) and π“ž(N), respectively. The grid search algorithm shows a very good agreement with our expectation. However, the simple search algorithm performs worse than expected for the systems with 262636 and 492038 particles. The performance loss of the simple search algorithm is likely due to the large amount of data the algorithm has to process in the inner loops. For large data, cache memory limits may exceed causing bad scaling. An additional benchmark, not reported here, confirmed the dependence of the grid search algorithm on the neighbor cut-off distance, whereas the simple search is unaffected. However, even after increasing the cut-off from 1.6 to 2.6Οƒ, it is found that the grid search performs an order of magnitude faster than the simple search. The cross-over, where the simple search becomes faster than the grid search depends on the cut-off distance, but is typically too large for productive uses, and hence, the grid search should be preferred over the simple search in general.

Bottom Line: We illustrate the new methods by coarse-graining SPC/E bulk water and more complex water-methanol mixture systems.The CG potentials obtained from both methods are then evaluated by comparing the pair distributions from the coarse-grained to the reference atomistic simulations.In addition to the newly implemented methods, we have also added a parallel analysis framework to improve the computational efficiency of the coarse-graining process.

View Article: PubMed Central - PubMed

Affiliation: Department of Mechanical Science and Engineering, Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, Illinois, 61801, United States of America.

ABSTRACT
We discuss recent advances of the VOTCA package for systematic coarse-graining. Two methods have been implemented, namely the downhill simplex optimization and the relative entropy minimization. We illustrate the new methods by coarse-graining SPC/E bulk water and more complex water-methanol mixture systems. The CG potentials obtained from both methods are then evaluated by comparing the pair distributions from the coarse-grained to the reference atomistic simulations. In addition to the newly implemented methods, we have also added a parallel analysis framework to improve the computational efficiency of the coarse-graining process.

No MeSH data available.