Limits...
Accelerated search for materials with targeted properties by adaptive design.

Xue D, Balachandran PV, Hogden J, Theiler J, Xue D, Lookman T - Nat Commun (2016)

Bottom Line: Here we show how an adaptive design strategy, tightly coupled with experiments, can accelerate the discovery process by sequentially identifying the next experiments or calculations, to effectively navigate the complex search space.We synthesize and characterize 36 predicted compositions (9 feedback loops) from a potential space of ∼800,000 compositions.Of these, 14 had smaller ΔT than any of the 22 in the original data set.

View Article: PubMed Central - PubMed

Affiliation: Theoretical Division, Los Alamos National Laboratory, MS-B262, Los Alamos, New Mexico 87545, USA.

ABSTRACT
Finding new materials with targeted properties has traditionally been guided by intuition, and trial and error. With increasing chemical complexity, the combinatorial possibilities are too large for an Edisonian approach to be practical. Here we show how an adaptive design strategy, tightly coupled with experiments, can accelerate the discovery process by sequentially identifying the next experiments or calculations, to effectively navigate the complex search space. Our strategy uses inference and global optimization to balance the trade-off between exploitation and exploration of the search space. We demonstrate this by finding very low thermal hysteresis (ΔT) NiTi-based shape memory alloys, with Ti50.0Ni46.7Cu0.8Fe2.3Pd0.2 possessing the smallest ΔT (1.84 K). We synthesize and characterize 36 predicted compositions (9 feedback loops) from a potential space of ∼800,000 compositions. Of these, 14 had smaller ΔT than any of the 22 in the original data set.

No MeSH data available.


Related in: MedlinePlus

Inference and design combination.The relative performance of various regressor:selector combinations on the NiTi SMA training data set. On the abscissa, we plot the number of initial random picks, taken from the training set, for building the statistical inference model. On the ordinate, we plot the average number of picks required to find the alloy in the training set with the lowest thermal hysteresis (ΔT). The best regressor:selector finds the optimal alloy in as few picks as possible. We conclude that SVRrbf:KG (continuous red line) is the best regressor:selector combination for the NiTi SMA problem. Random picks are given as continuous blue line.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4835535&req=5

f2: Inference and design combination.The relative performance of various regressor:selector combinations on the NiTi SMA training data set. On the abscissa, we plot the number of initial random picks, taken from the training set, for building the statistical inference model. On the ordinate, we plot the average number of picks required to find the alloy in the training set with the lowest thermal hysteresis (ΔT). The best regressor:selector finds the optimal alloy in as few picks as possible. We conclude that SVRrbf:KG (continuous red line) is the best regressor:selector combination for the NiTi SMA problem. Random picks are given as continuous blue line.

Mentions: We emphasize that a priori it is not clear which regressor:selector combination to use. According to the ‘no-free-lunch theorem'27, there is no universal optimizer for all problems. Thus, we investigated the performances of several regressor:selector combinations as a function of the size of the data using cross-validation and found that SVRrbf:KG outperformed every other regressor:selector combination on the training set. Our approach (see Methods) is based on determining the average number of samples needed to find the best value when training on randomly chosen initial subsets of the training data. In particular, we randomly selected without replacement a given number of samples from the training data, trained using a given regressor:selector combination pair and counted the total number of tries needed to find the best sample in the training data. This was repeated 2,000 times with different sets of randomly selected samples and we included the initial random picks in the overall count. Figure 2 shows the relative performance of the various regressor:selector combinations on the NiTi training data. The plotting symbols are slightly larger than the s.d. of the measurements. Some of the algorithms perform worse than random, agreeing with the no-free-lunch theorem that guarantees such algorithms exist. The GPM:Min combination, which for design merely chooses the best estimate from the GPM regressor, showed the best performance for sample sizes 2 and 3; however, as more samples were included its performance began to deteriorate. For samples sizes from 4 to 8, SVRrbf:KG and SVRrbf:EGO have nearly identical performance and are better than the other regressor:selector pairs. As SVRrbf works well or better than the other techniques beyond three training samples—as we have more than three training samples and as we do not have any compelling argument for using less than our full training set on the problem—the results indicate that we should choose SVRrbf:KG. Thus, the 22 initial samples are adequate enough for our training set, because beyond ∼5 randomly chosen training samples, we are better off using samples chosen by the design loop than by random guessing.


Accelerated search for materials with targeted properties by adaptive design.

Xue D, Balachandran PV, Hogden J, Theiler J, Xue D, Lookman T - Nat Commun (2016)

Inference and design combination.The relative performance of various regressor:selector combinations on the NiTi SMA training data set. On the abscissa, we plot the number of initial random picks, taken from the training set, for building the statistical inference model. On the ordinate, we plot the average number of picks required to find the alloy in the training set with the lowest thermal hysteresis (ΔT). The best regressor:selector finds the optimal alloy in as few picks as possible. We conclude that SVRrbf:KG (continuous red line) is the best regressor:selector combination for the NiTi SMA problem. Random picks are given as continuous blue line.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4835535&req=5

f2: Inference and design combination.The relative performance of various regressor:selector combinations on the NiTi SMA training data set. On the abscissa, we plot the number of initial random picks, taken from the training set, for building the statistical inference model. On the ordinate, we plot the average number of picks required to find the alloy in the training set with the lowest thermal hysteresis (ΔT). The best regressor:selector finds the optimal alloy in as few picks as possible. We conclude that SVRrbf:KG (continuous red line) is the best regressor:selector combination for the NiTi SMA problem. Random picks are given as continuous blue line.
Mentions: We emphasize that a priori it is not clear which regressor:selector combination to use. According to the ‘no-free-lunch theorem'27, there is no universal optimizer for all problems. Thus, we investigated the performances of several regressor:selector combinations as a function of the size of the data using cross-validation and found that SVRrbf:KG outperformed every other regressor:selector combination on the training set. Our approach (see Methods) is based on determining the average number of samples needed to find the best value when training on randomly chosen initial subsets of the training data. In particular, we randomly selected without replacement a given number of samples from the training data, trained using a given regressor:selector combination pair and counted the total number of tries needed to find the best sample in the training data. This was repeated 2,000 times with different sets of randomly selected samples and we included the initial random picks in the overall count. Figure 2 shows the relative performance of the various regressor:selector combinations on the NiTi training data. The plotting symbols are slightly larger than the s.d. of the measurements. Some of the algorithms perform worse than random, agreeing with the no-free-lunch theorem that guarantees such algorithms exist. The GPM:Min combination, which for design merely chooses the best estimate from the GPM regressor, showed the best performance for sample sizes 2 and 3; however, as more samples were included its performance began to deteriorate. For samples sizes from 4 to 8, SVRrbf:KG and SVRrbf:EGO have nearly identical performance and are better than the other regressor:selector pairs. As SVRrbf works well or better than the other techniques beyond three training samples—as we have more than three training samples and as we do not have any compelling argument for using less than our full training set on the problem—the results indicate that we should choose SVRrbf:KG. Thus, the 22 initial samples are adequate enough for our training set, because beyond ∼5 randomly chosen training samples, we are better off using samples chosen by the design loop than by random guessing.

Bottom Line: Here we show how an adaptive design strategy, tightly coupled with experiments, can accelerate the discovery process by sequentially identifying the next experiments or calculations, to effectively navigate the complex search space.We synthesize and characterize 36 predicted compositions (9 feedback loops) from a potential space of ∼800,000 compositions.Of these, 14 had smaller ΔT than any of the 22 in the original data set.

View Article: PubMed Central - PubMed

Affiliation: Theoretical Division, Los Alamos National Laboratory, MS-B262, Los Alamos, New Mexico 87545, USA.

ABSTRACT
Finding new materials with targeted properties has traditionally been guided by intuition, and trial and error. With increasing chemical complexity, the combinatorial possibilities are too large for an Edisonian approach to be practical. Here we show how an adaptive design strategy, tightly coupled with experiments, can accelerate the discovery process by sequentially identifying the next experiments or calculations, to effectively navigate the complex search space. Our strategy uses inference and global optimization to balance the trade-off between exploitation and exploration of the search space. We demonstrate this by finding very low thermal hysteresis (ΔT) NiTi-based shape memory alloys, with Ti50.0Ni46.7Cu0.8Fe2.3Pd0.2 possessing the smallest ΔT (1.84 K). We synthesize and characterize 36 predicted compositions (9 feedback loops) from a potential space of ∼800,000 compositions. Of these, 14 had smaller ΔT than any of the 22 in the original data set.

No MeSH data available.


Related in: MedlinePlus