Tyler O'Neal, Staff Editor CHEMISTRY

Swedish researchers show how the model used to evaluate lockdowns was flawed

In a recent study, researchers from Imperial College London developed a model to assess the effect of different measures used to curb the spread of the coronavirus. However, the model had fundamental shortcomings and cannot be used to draw the published conclusions, claim Swedish researchers from Lund University, and other institutions, in the journal Nature.

The results from Imperial indicated that it was almost exclusively the complete societal lockdown that suppressed the wave of infections in Europe during spring. 

{media load=media,id=249,width=350,align=left,display=inline}

The study estimated the effects of different measures such as social distancing, self-isolating, closing schools, banning public events, and the lockdown itself.

"As the measures were introduced at roughly the same time over a few weeks in March, the mortality data used simply does not contain enough information to differentiate their individual effects. We have demonstrated this by conducting a mathematical analysis. Using this as a basis, we then ran simulations using Imperial College's original code to illustrate how the model's sensitivity leads to unreliable results," explains Kristian Soltesz, associate professor in automatic control at Lund University and first author of the article. 

The group's interest in the Imperial College model was roused by the fact that it explained almost all of the reduction in transmission during the spring via lockdowns in ten of the eleven countries modeled. The exception was Sweden, which never introduced a lockdown.

"In Sweden, the model offered an entirely different measure as an explanation for the reduction - a measure that appeared almost ineffective in other countries. It seemed almost too good to be true that an effective lockdown was introduced in every country except one, while another measure appeared to be unusually effective in this country", notes Soltesz. mqdefault cc6a0 {module INSIDE STORY}

Soltesz is careful to point out that it is entirely plausible that individual measures are affected, but that the model could not be used to determine how effective they were.

"The various interventions do not appear to work in isolation from one another, but are often dependent upon each other. A change in behavior as a result of one intervention influences the effect of other interventions. How much and in what way is harder to know, and requires different skills and collaboration", says Anna Jöud, associate professor in epidemiology at Lund University and co-author of the study.

Analyses of models from Imperial College and others highlight the importance of epidemiological models being reviewed, according to the authors.

"There is a major focus in the debate on sources of data and their reliability, but an almost total lack of systematic review of the sensitivity of different models in terms of parameters and data. This is just as important, especially when governments across the globe are using dynamic models as a basis for decisions", Soltesz and Jöud point out.

The first step is to carry out a correct analysis of the model's sensitivities. If they pose too great a problem then more reliable data is needed, often combined with a less complex model structure.

"With a lot at stake, it is wise to be humble when faced with fundamental limitations. Dynamic models are usable as long as they take into account the uncertainty of the assumptions on which they are based and the data they are led by. If this is not the case, the results are on a par with assumptions or guesses", concludes Soltesz.

Swiss scientist shows how nearby galaxies form their stars

Tyler O'Neal, Staff Editor CHEMISTRY

How stars form in galaxies remains a major open question in astrophysics. A new UZH study sheds new light on this topic with the help of a data-driven re-analysis of observational measurements. The star-formation activity of typical, nearby galaxies is found to scale proportionally with the amount of gas present in these galaxies. This points to the net gas supply from cosmic distances as the main driver of galactic star formation.

Stars are born in dense clouds of molecular hydrogen gas that permeates the interstellar space of most galaxies. While the physics of star formation is complex, recent years have seen substantial progress towards understanding how stars form in a galactic environment. What ultimately determines the level of star formation in galaxies, however, remains an open question.

In principle, two main factors influence the star formation activity: The amount of molecular gas that is present in galaxies and the timescale over which the gas reservoir is depleted by converting it into stars. While the gas mass of galaxies is regulated by competition between gas inflows, outflows, and consumption, the physics of the gas-to-star conversion is currently not well understood. Given its potentially critical role, many efforts have been undertaken to determine the gas depletion timescale observationally. However, these efforts resulted in conflicting findings partly because of the challenge in measuring gas masses reliably given current detection limits. Stars (white) form throughout the gas disk. (Illustration: Robert Feldmann){module INSIDE STORY}

Typical star formation is linked to the overall gas reservoir

The present study from the Institute for Computational Science of the University of Zurich uses a new statistical method based on Bayesian modeling to properly account for galaxies with undetected amounts of molecular or atomic hydrogen to minimize observational bias. This new analysis reveals that, in typical star-forming galaxies, molecular and atomic hydrogen is converted into stars over approximately constant timescales of 1 and 10 billion years, respectively. However, extremely active galaxies (“starbursts”) are found to have much shorter gas depletion timescales.

“These findings suggest that star formation is indeed directly linked to the overall gas reservoir and thus set by the rate at which gas enters or leaves a galaxy,” says Robert Feldmann, professor at the Center for Theoretical Astrophysics and Cosmology. In contrast, the dramatically higher star-formation activity of starbursts likely has a different physical origin, such as galaxy interactions or instabilities in galactic disks.

Far galaxies across cosmic history

This analysis is based on observational data of nearby galaxies. Observations with the Atacama Large Millimeter/Submillimeter Array, the Square Kilometer Array, and other observatories promise to probe the gas content of large numbers of galaxies across cosmic history. It will be paramount to continue the development of statistical and data science methods to accurately extract the physical content from these new observations and to fully uncover the mysteries of star formation in galaxies.

UH machine learning model boosts search for 'superhard' materials

Tyler O'Neal, Staff Editor CHEMISTRY

The model predicts promising new materials

Superhard materials are in high demand in industry, from energy production to aerospace, but finding suitable new materials has largely been a matter of trial and error based on classical materials such as diamonds. Until now.

Researchers from the University of Houston and Manhattan College have reported a machine learning model that can accurately predict the hardness of new materials, allowing scientists to more readily find compounds suitable for use in a variety of applications. The work was reported in Advanced Materials.

Superhard materials - defined as those with a hardness value exceeding 40 gigapascals on the Vickers scale, meaning it would take more than 40 gigapascals of pressure to leave an indentation on the material's surface - are rare. Researchers have developed a machine learning model that can accurately predict the hardness of new materials, allowing scientists to more readily find compounds suitable for use in a variety of applications.{module INSIDE STORY}

"That makes identifying new materials challenging," said Jakoah Brgoch, associate professor of chemistry at UH and corresponding author for the paper. "That is why materials like synthetic diamond are still used even though they are challenging and expensive to make."

One of the complicating factors is that the hardness of a material may vary depending on the amount of pressure exerted, known as load dependence. That makes testing a material experimentally complex and using computational modeling today almost impossible.

The model reported by the researchers overcomes that by predicting the load-dependent Vickers hardness based solely on the chemical composition of the material. The researchers report finding more than 10 new and promising stable borocarbide phases; work is now underway to design and produce the materials so they can be tested in the lab.

Based on the model's reported accuracy, the odds are good. Researchers reported the accuracy at 97%.

First author Ziyan Zhang, a doctoral student at UH, said the database built to train the algorithm is based on data involving 560 different compounds, each yielding several data points. Finding the data required poring over hundreds of published academic papers to find data needed to build a representative dataset.

"All good machine learning projects start with a good dataset," said Brgoch, who is also a principal investigator with the Texas Center for Superconductivity at UH. "The true success is largely the development of this dataset."

In addition to Brgoch and Zhang, additional researchers on the project include Aria Mansouri Tehrani and Blake Day, both with UH, and Anton O. Oliynyk from Manhattan College.

Researchers traditionally have used machine learning to predict a single variable of hardness, Brgoch said, but that doesn't account for the complexities of the property like load dependence, which he said still aren't well understood. That makes machine learning a good tool, despite earlier limitations.

"A machine learning system doesn't need to understand the physics," he said. "It just analyzes the training data and makes new predictions based on statistics."

Machine learning does have limitations, though. "The idea of using machine learning isn't to say, 'Here is the next greatest material,' but to help guide our experimental search," Brgoch said. "It tells you where you should look."