How can we enhance our understanding of the uncertainties that arise when experiment and theory come together? A new focus issue in JPhysG explores ways to assess uncertainty and improve both the precision and accuracy of nuclear models.

A schematic illustration of systematic (accuracy) and statistical (precision) uncertainties of nuclear models.

For a long time researchers have built frameworks, theories and models to understand and predict physics, and then put those ideas to the test in an experiment. This cycle of “observation-theory-prediction-experiment-“, known simply as the scientific method, is the foundation we use to build our knowledge and understanding of all systems and phenomena.

The scientific method is good, and it works, but we often ignore a vital component that could refine it still further. That component is a full understanding of the uncertainties and errors within a given theory, model or experiment, and how they interact with one another.

David Ireland of the University of Glasgow in the UK and Witek Nazarewicz of Michigan State University in the US have tackled this problem within the field of nuclear physics, although the approaches presented apply equally well to any discipline that uses the scientific method. In an idea that has developed over several years they have edited a new focus issue in Journal of Physics G: Nuclear and Particle Physics that aims to answer the question:  “What is the best way to use experimental data in the formulation of theoretical models that attempt to explain the results and make predictions for new observables?”

Ireland and Nazarewicz sum up the importance and difficulty of this in the simple figure (above) from the issue’s editorial:

“The figure illustrates in a schematic way the challenge of quantifying model uncertainties, by categorising these uncertainties in two ways: systematic (accuracy; how near the model is to reality), and statistical (precision; how well known are model parameters). There are many techniques to assess statistical uncertainties, which result from our imperfect knowledge of model parameters. Systematic uncertainties, on the other hand, are very tough to assess, as exact models are seldom available in nuclear physics. Good agreement between different models (precision) does not guarantee that they are near the truth; all models can be systematically off due to missing knowledge.”

While the papers in this issue are focussed on nuclear physics, many of the techniques are rooted in statistics and numerical methods that can be applied across a variety of scientific disciplines. Many of the techniques are not entirely new, but it is important for the nuclear physics community to embrace them and use them correctly to improve our models and experiments. As an example, Richard Casten (Yale University) guides us through a range of tests for testing theories that will be familiar to scientists from other fields, whilst David Higdon (Los Alamos National Laboratory) and his team  assess and apply Bayesian approaches to nuclear density functional theory.

Interested? The full focus issue, entitled ‘Enhancing the interaction between nuclear experiment and theory through information and statistics‘  is now available to read in Journal of Physics G: Nuclear and Particle Physics.

Despite the far-reaching topics covered in this collection of articles, Ireland and Nazarewicz stress the importance of continuing to develop new theories and ways to cope with errors and uncertainties – while also ensuring that researchers make use of techniques that are already available.

The two editors ponder this problem in the final words from their editorial:

“It is our assertion that a paradigm shift is needed…to enhance the coupling between theory and experiment, and we hope that this collection of articles is a good start.”

The good news is that the community is already responding, with the third ISNET (Information and statistics in nuclear experiment and theory) workshop due to take place this coming November. Further ahead, a workshop dedicated to Bayesian methods in nuclear physics is planned for summer 2016.

Oh, and the answer to the question posed by Ireland and Nazarewicz in their focus issue? The contributions show that it’s complicated, and not without its own uncertainty.

This work is licensed under a Creative Commons Attribution 3.0 Unported License

Figure & excerpt: from D G Ireland and W Nazarewicz 2015 J. Phys. G: Nucl. Part. Phys. 42 030301. Copyright IOP Publishing 2015.

Front image: angular and energy dependence of insolvabilities, adapted from Jannes Nys et al 2015 J. Phys. G: Nucl. Part. Phys. 42 034016. Copyright IOP Publishing 2015.

Categories: Journal of Physics G: Nuclear and Particle Physics

Tags: Computational physics, Featured, Nuclear Physics, Nuclear reactions, Nuclear structure, Statistical physics