The conclusion of Aspect's experiment is statistically flawed
Published in Statistics
History and connection to Statistics
Considering Einstein's doubts about quantum mechanics, Einstein was puzzled by the strange "causality" in certain characteristics of quantum mechanics. In particular, the long distance connection between two particles that had briefly interacted, was a mystery to him. For, after all, the possibility of immediate interaction over a long distance could violate relativity.
Hidden parameters
Einstein postulated the existence of extra hidden parameters that could explain the quantum entanglement of particles over long distances. His aim was eventually a return to causality and locality in the philosophical basis of quantum mechanics. The latter was, and still is, a debated controversy.
Nowadays, one can read statements like: Aspect demonstrated conclusively that Einstein was wrong about quantum mechanics. It means that there are no Einsteinian extra parameters that may explain entanglement. My paper asks if that really is the case. Let us first look at the set up of Aspect's experiment.
Aspect's experiment and data analysis
The key to Aspect's experiment is Bell's correlation formula between measurements. In Bell's function, the Einsteinian variables are introduced as ancillary parameters. The ancillary parameters co-determine the outcome of measurements and are ruled by classical probability laws.
Interestingly, the hypothetical existence of Einstein's hidden ancillary parameters in Bell's formula also give rise to certain inequalities. Aspect wanted to employ these inequalities in his data analysis.
Methodology of data gathering and analysis in the experiment
Aspect set up his experiment without using explicit models for measurement and for the way the hidden ancillary variables influence the measured result. This practical abstraction enabled Aspect to employ an approximate form of Bell's correlation function. And that, in turn, enabled him to employ the inequalities.
Therefore, Aspect's approach is a completely classical probability approach to possible Einstein positive data in empirical reality. Furthermore, the “yes or no” violation of the inequalities in empirical reality is a litmus test for the presence or absence of Einstein data in empirical reality.
Wait a minute
But please note. The quantum correlation can violate the Bell inequalities.
Now, on the one hand Einstein data, in the form of the ancillary extra hidden parameters can mathematically not violate Bell inequalities. On the other hand, if they can not violate the inequalities, then they don't exist in empirical reality when in empirical reality the inequalities are violated with the data gathered in experiment.
The key here is that Aspect in the hypothesis of existence of Einstein data, on the one hand corners Einstein data as classical probability, while on the other hand, the data must be of non classical probability character, in order to exist.
There could be a statistical wrinkle here. My paper
https://rdcu.be/ftJkR
demonstrates that, indeed, such is the case. Aspect's experiment is statistically flawed and does not conclusively demonstrate the absence of Einstein data i.e. demonstrate non-locality.
This isn't a probability function
To give the reader a bit of a taste of what the paper is about, the following point is added. In the data analysis we take the settings of the measurement instrument parameters, a, for Alice and, b, for Bob. One can define an angle x=ange(a,b) in the plane spanned by a and b. The angle x is i the interval [0,2π).
Aspect requires in the data analysis that:
sin²(x/2) = Prob[≠,x]
With Prob[≠,x] the probability that, given the angle between Alice's measurement instrument parameter, a, and Bob's measurement instrument parameter b, is x, gives unequal results for Alice's measurement and Bob's measurement. But note, sin²(x/2), isn't monotone non-descending on the complete interval [0,2π). It's not a probability function.
So, Aspect required from Einstein data to be classical probability in nature, but, not to follow a monotone non-descending function on [0,2π) as a probability function. That's not possible. The hypothesis: "in the experiment there are Einstein positive data possible ", is false by statistical design of the experiment. No data can, at the same time, be classical probability based and not follow a monotone non-descending probability function. The details are in the paper
Thank you.
My expertise is twofold. In this particular case, my knowledge of statistical methodology counts. But of course one must have knowledge of quantum mechanics as well.
I never fully agreed with the "Einstein is wrong approach " of the nineties and early 2000. Together with Dr Nagata and Dr Nakamura I published work that showed in a very round about way that something could be wrong with the "Einstein is wrong " approach.
Other papers are, a.o. https://doi.org/10.1016/j.nexres.2025.101268. This paper shows that Einstein's idea about field theory of hidden variables might very well be correct. Another paper is about the Krein-Milman theorem from measure theory that questions the completeness of Bell's formula; https://ijqf.org/archives/7960
I am a pensioner. I earned my PhD in statistical methodology in 1992. The association with GDS Applied Mathematics BV is more like a consultant.
Because being a pensioner I could indulge myself also with foundations of mathematics. I think this paper https://rdcu.be/ewfW0 is interesting as well.
Follow the Topic
-
Journal of Statistical Theory and Applications
This journal is dedicated to the publication of original research papers in the area of statistical sciences and related applications.
Related Collections
With Collections, you can get published faster and increase your visibility.
Machine Learning and Advanced Statistical Methods in Energy and Environmental Research
This topical collection seeks to advance methodologies that tackle key challenges in environmental and energy research. It will particularly focus on how machine learning techniques, statistical theories, and simulations can enhance research methods and practical applications in energy and environmental-related fields. As energy markets become increasingly complex, robust statistical methodologies, data-driven machine learning models, and simulation techniques are essential for analysing vast datasets, improving forecasting accuracy, optimizing resource allocation, and guiding policy decisions.
This issue highlights innovative applications of machine learning, statistical methods, and simulations in energy pricing, demand forecasting, market dynamics, renewable energy integration, and sustainability assessments. It will explore how machine learning algorithms, such as deep learning, reinforcement learning, and Bayesian networks, can complement traditional statistical approaches. Additionally, it will emphasize the role of simulations in modelling energy systems, testing policy interventions, and assessing risk under uncertainty.
By integrating machine learning, statistical analysis, and simulations, this special issue seeks to advance both fields and promote interdisciplinary research, bridging theory with real-world applications.
This collection covers, though is not strictly limited to, the following areas:
• High-dimensional data analysis to address challenges related to large-scale energy sector datasets
• Machine learning and AI methods for predictive analysis with their applications in the energy sector
• Time series analysis for improved energy demand forecasting
• Advanced forecasting models for renewable energy generation and integration
• Statistical methods for smart energy pricing and market efficiency
• Improving energy efficiency through advanced data analytics
• Risk management and resilience modeling in energy markets and infrastructure
These topics are a selection of key focus areas, but we also welcome contributions exploring other innovative directions within energy and environmental research, particularly those leveraging advanced data-driven methodologies.
Publishing Model: Open Access
Deadline: Aug 31, 2026
Innovative Statistical Methods for Complex Data in Ophthalmology and Vision Science
Ophthalmology and vision science studies increasingly generate multi-modal, high-dimensional, and longitudinal datasets that present substantial statistical challenges. A distinctive feature of ophthalmic data is the inherent dependence between paired eyes within individuals, often coupled with repeated measures over time. These complex structures, along with irregular visit schedules, missing outcomes, and diverse data types such as imaging, functional and structural measures, and epidemiologic data, require specialized methodological approaches.
This collection focuses on the development and application of advanced statistical methodologies tailored to these challenges, including joint modeling approaches, regression techniques for high-dimensional data, Bayesian inference, and machine learning and deep learning methods with statistical guarantees. By showcasing innovations in statistical theory and practice, the collection aims to promote rigorous and reproducible analysis in ophthalmic research and to provide methodological insights with broad applicability to biomedical data science.
This Collection supports and amplifies research related to SDG 3: Good Health & Well-Being
Publishing Model: Open Access
Deadline: Oct 22, 2026
Please sign in or register for FREE
If you are a registered user on Research Communities by Springer Nature, please sign in