Constraining Black Hole Parameters in Non-Commutative Geometry using Machine Learning

This paper employs CUDA-accelerated machine learning to constrain non-commutative black hole parameters by analyzing shadow behaviors and energy emission rates, ultimately demonstrating that the proposed model is consistent with Event Horizon Telescope observations of the $SgrA^*$ black hole.

The Gist

Imagine the universe as a giant, complex video game. In this game, the most mysterious characters are Black Holes. For a long time, scientists have tried to figure out exactly what these "bosses" look like and how they behave using the rules of standard physics (General Relativity). But recently, scientists have started asking: "What if the game's code has a glitch? What if space and time aren't perfectly smooth, but are actually made of tiny, fuzzy pixels?"

This is the world of Non-Commutative (NC) Geometry. It's a theory suggesting that at the tiniest scales, the rules of the game change slightly.

This paper is like a detective story where the author, Maryem Jemri, tries to solve a mystery: Do these "fuzzy pixel" black holes actually exist in our real universe, or are they just math games?

Here is how she solves the case, broken down into simple steps:

1. The Setup: Building the "Fuzzy" Black Hole

First, the author builds a theoretical model of a black hole. But this isn't a normal one. She adds three special ingredients to her recipe:

• A "Cloud of Strings": Imagine the black hole is wrapped in a fuzzy blanket made of tiny vibrating strings.
• Dark Energy: The invisible force pushing the universe apart, acting like a background pressure.
• The "Fuzziness" (Non-Commutativity): This is the main character. It's a parameter (let's call it $b$) that controls how "pixelated" or fuzzy the space around the black hole is.

2. The Super-Computer: Using CUDA as a High-Speed Camera

To see what these fuzzy black holes would look like, she needs to run millions of calculations. Doing this on a normal computer would take years. So, she uses CUDA, which is like giving the computer a fleet of super-fast racing cars (GPUs) to do the work all at once.

She simulates how light travels around these black holes. Since black holes are so heavy, they bend light like a funhouse mirror. This creates a dark circle in the middle called a Shadow.

• The Analogy: Imagine shining a flashlight at a bowling ball in a foggy room. The ball blocks the light, creating a shadow. The shape and size of that shadow tell you about the ball.
• The Result: She finds that changing the "fuzziness" parameter ($b$) changes the size and shape of the shadow. A higher $b$ makes the shadow bigger and more distorted.

3. The Real-World Check: The Event Horizon Telescope (EHT)

Now, she has a bunch of theoretical shadows. But do they match reality?
She compares her computer-generated shadows to real photos taken by the Event Horizon Telescope (EHT). The EHT is a giant telescope network that actually took pictures of two famous black holes: M87* (a giant one in a distant galaxy) and Sgr A* (the one right in the center of our own Milky Way).

She asks: "If I tweak the fuzziness ($b$) and the other ingredients, does my computer shadow look like the real photo?"

• The Finding: She finds that for the black hole in our galaxy (Sgr A*), specifically the version observed by the Keck telescope, there is a specific range of "fuzziness" that makes the theoretical shadow match the real photo perfectly.

4. The AI Detective: Machine Learning

There are so many combinations of ingredients (fuzziness, string clouds, dark energy, spin, charge) that checking them one by one is still too slow. So, she brings in a Machine Learning assistant.

• The Analogy: Imagine you have a giant box of 20,000 different puzzle pieces. You want to find the ones that fit the picture of the real black hole. Instead of trying every piece, you train a smart robot (a Neural Network) to look at the pieces and say, "Yes, this fits" or "No, this doesn't."
• The Training: She feeds the robot thousands of examples of her computer shadows, telling it which ones match the EHT photos and which ones don't.
• The "Voting" System: To make sure the robot isn't just guessing, she uses a clever trick. She shows the robot the same puzzle piece 100 times with tiny, almost invisible changes. If the robot says "Yes" 99 times and "No" once, it takes a vote and goes with the majority. This makes the decision very reliable.

5. The Verdict

The AI detective did its job with incredible accuracy (over 97% correct!).

• The Conclusion: The study finds that the "fuzzy" black hole model does match the observations of Sgr A* (our galaxy's black hole) as seen by the Keck telescope.
• The Limit: However, the "fuzziness" parameter ($b$) can't be just any number. It has to be small (less than about 0.44) to fit the picture. If it were too big, the shadow would look wrong.

Summary

In short, the author used a super-fast computer to simulate "fuzzy" black holes, then used a smart AI to compare those simulations to real photos of our galaxy's black hole. The result? The "fuzzy" theory works! It fits the real data, suggesting that our universe might indeed have a slightly "pixelated" structure at the smallest scales, at least around black holes.

What the paper does NOT claim:

• It does not claim this proves string theory is definitely true (it just says the model is consistent with the data).
• It does not claim this technology can be used for anything other than studying black holes right now.
• It does not claim we can see these "pixels" with our eyes; it's a mathematical constraint based on the shape of the shadow.

Technical Summary

Technical Summary: Constraining Black Hole Parameters in Non-Commutative Geometry using Machine Learning

Problem Statement
The paper addresses the challenge of constraining the parameters of rotating and charged black holes within Non-Commutative (NC) geometry, specifically those embedded in a spacetime containing a cloud of strings and a quintessence (dark energy) sector. While theoretical models in NC geometry offer extensions to General Relativity motivated by string theory, validating these models requires confronting their predictions with observational data, particularly the black hole shadow images provided by the Event Horizon Telescope (EHT) collaboration. The authors aim to determine which regions of the NC parameter space are consistent with the observed shadows of M87* and Sgr A* (both VLTI and Keck datasets), utilizing high-performance computing and machine learning to navigate the complex, multi-dimensional parameter space.

Methodology
The study employs a three-pronged approach combining theoretical modeling, high-performance numerical simulation, and machine learning classification:

1. Theoretical Framework: The authors model the spacetime metric of a rotating, charged NC black hole incorporating a cloud of strings (parameter $\alpha$) and quintessence (parameters $N$ and $w$). The NC geometry is characterized by a deformation parameter $\Theta$, reduced to a single parameter $b = 8\sqrt{\Theta}/\sqrt{\pi}$. The metric function includes perturbative NC corrections and external field contributions.
2. CUDA-Accelerated Numerical Simulations:
   • Horizon Analysis: The existence of event horizons is determined numerically by solving the horizon equation ($\Delta(r)=0$) across a grid of parameters ($a, b, \alpha, N$) using NVIDIA GPU-accelerated CUDA code.
   • Shadow and Emission Calculations: Using the Hamilton-Jacobi formalism and the Carter method, the authors derive equations of motion for null geodesics. CUDA-based parallel computing is utilized to calculate the celestial coordinates ($X, Y$) defining the black hole shadow and the energy emission rate. The shadow radius ($R_s$) is compared against the Schwarzschild radius to compute the fractional deviation ($d$), which is then matched against EHT observational bounds (1$\sigma$ and 2$\sigma$ confidence intervals).
3. Machine Learning Classification:
   • Data Generation: A dataset of 20,000 samples is generated by varying the parameters $(b, N, \alpha)$. Each sample is labeled binary (1 or 0) based on whether the calculated shadow radius falls within the EHT observational constraints for M87*, Sgr A* (VLTI), or Sgr A* (Keck).
   • Model Architecture: A Fully Connected Neural Network (FCNN) is trained to classify parameter sets. The architecture consists of an input layer, hidden layers with ReLU activation, and a softmax output layer.
   • Voting Strategy: To enhance robustness against small numerical variations, a voting mechanism is employed where multiple perturbed versions of an input tuple are processed independently, and the final prediction is determined by majority vote.

Key Contributions

• High-Performance Numerical Framework: The paper establishes a CUDA-based framework that significantly accelerates the computation of black hole shadows and energy emission rates in complex NC geometries, allowing for dense parameter space scans that would be computationally prohibitive on standard CPUs.
• Parameter Constraints: The study provides specific numerical constraints on the NC parameter $b$ and its interplay with the string cloud parameter $\alpha$, the quintessence parameter $N$, the charge $Q$, and the spin $a$.
• Machine Learning Application: The work demonstrates the efficacy of FCNNs in classifying theoretical black hole configurations against observational data, achieving high accuracy in distinguishing consistent from inconsistent parameter sets.

Results

• Shadow Behavior: The NC parameter $b$ is found to control both the size and global shape of the shadow. Increasing $b$ leads to larger, more extended shadows, while the rotation parameter $a$ and charge $Q$ induce D-like deformations and size reductions, respectively. The string cloud parameter $\alpha$ and quintessence parameter $N$ primarily influence the distortion and overall size, with large $N$ and small $\alpha$ producing D-like shapes.
• Energy Emission: The energy emission rate exhibits non-monotonic behavior with respect to charge $Q$ and spin $a$ (increasing to a critical value then decreasing). Conversely, increasing $\alpha$, $N$, or $b$ generally suppresses the energy emission rate.
• Observational Constraints: The numerical analysis reveals that while the theoretical parameter space is vast, only limited regions are compatible with EHT data. Specifically, the NC parameter $b$ is consistently constrained to be positive and generally below approximately 0.44 to satisfy the 1$\sigma$ and 2$\sigma$ bounds for M87* and Sgr A*.
• Machine Learning Performance: The trained FCNN demonstrates high classification accuracy. For the 1$\sigma$ constraint, test accuracies range from 0.9779 to 0.981 across the different black hole datasets. The voting strategy further improves stability, with voting accuracies reaching up to 0.9815. The model performs slightly better under the stricter 1$\sigma$ constraints compared to the 2$\sigma$ constraints, particularly for the Sgr A* Keck dataset.

Significance and Claims
The paper claims that the proposed non-commutative model is consistent with the observational findings of the Sgr A* Keck black hole. The authors emphasize that their work bridges theoretical predictions in NC geometry with empirical data through a robust combination of GPU-accelerated simulations and machine learning. They posit that this approach provides a powerful framework for probing fundamental properties of black holes in non-trivial gravity theories. The authors maintain a modest stance, noting that while the classification results are strong, the approach should be interpreted cautiously. They suggest that future work should incorporate more advanced learning methods, standard evaluation tools like ROC curves, and a deeper analysis of observational uncertainties and modeling assumptions. The study concludes that CUDA-accelerated machine learning is a viable and efficient tool for constraining black hole parameters in complex geometries.