🔭 astrophysics

CosmoSlider: An educational tool for cosmology

CosmoSlider is a lightweight, cross-platform educational tool that utilizes a TensorFlow Lite neural-network emulator to enable real-time, interactive exploration of how multiple cosmological parameters influence cosmic microwave background power spectra, thereby facilitating intuitive understanding without the computational burden of traditional solvers.

Original authors: Andreas Nygaard, Steen Hannestad, Thomas Tram

Published 2026-01-26

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Andreas Nygaard, Steen Hannestad, Thomas Tram

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to teach someone how the universe works, specifically how the "afterglow" of the Big Bang (called the Cosmic Microwave Background, or CMB) looks. To do this, you usually have to run incredibly complex computer simulations that take a long time to finish. It's like trying to bake a cake to show someone how changing the amount of sugar affects the taste, but the baking process takes three hours every time you tweak the recipe.

CosmoSlider is a new tool created by researchers Andreas Nygaard, Steen Hannestad, and Thomas Tram that solves this problem. Think of it as a "Cosmic Mixing Board" that lets you change the ingredients of the universe in real-time and instantly see how the "taste" (the visual pattern of the universe) changes.

Here is how it works, using simple analogies:

The Problem: The "Pre-Cooked Meal" vs. The "Instant Chef"

Traditionally, to show students how changing a parameter (like the amount of dark matter) changes the universe, scientists had two bad options:

The Animation: Show a video where one number changes slowly. This is like showing a video of a cake rising, but you can't stop it or change the recipe halfway through.
The Pre-Computed Grid: Imagine trying to make a giant library of every possible cake you could bake (low sugar/high flour, high sugar/low flour, etc.). If you want to change six different ingredients at once with high precision, you would need to bake one million different cakes and store them all. This takes up so much computer memory that it's impossible to fit on a phone or a standard website.

The Solution: The "Magic Taste-Tester" (Neural Network)

Instead of baking a million cakes and storing them, the authors built a "Magic Taste-Tester" (a neural network emulator).

How it works: They taught this computer program the rules of how the universe works. Now, instead of looking up a pre-baked cake in a library, the program instantly calculates what the cake would look like if you changed the ingredients right now.
The Result: You can slide a knob to change the amount of dark matter, the expansion rate of the universe, and other factors all at the same time. The screen updates instantly, showing you exactly how the universe's "fingerprint" changes. It's like having a chef who can instantly conjure up a new cake for every single tiny adjustment you make.

What You Can Do With It

The tool is called CosmoSlider and comes in two forms:

An iPhone/iPad App: A full-featured app where you can even bring in your own custom "recipes" (different cosmological models) if you are a teacher or researcher.
A Website: A lighter version that works in any web browser, perfect for a teacher to project on a screen during a lecture.

Why It Matters for Learning

The authors explain that this tool is designed for education.

Building Intuition: Just as you learn to drive by feeling the car respond to the steering wheel, students can learn cosmology by "feeling" how the universe responds to changes. They can drag sliders and immediately see, "Oh, if I add more dark matter, these peaks in the graph get taller."
No Waiting: There is no "Compute" button to press and no waiting for the computer to finish. The feedback is immediate, making it easy to explore and make mistakes.
Visual Aid: It helps students connect the math and physics they hear in class to a visual picture they can manipulate.

What's Next?

The creators plan to add more features soon, such as:

Saving the pictures you make as high-quality images for reports.
Changing colors and backgrounds.
Adding more types of data points from real-world experiments (like other telescopes) to compare against your custom universe.

In short: CosmoSlider is a lightweight, instant "universe simulator" that lets anyone play with the knobs of the cosmos to see how the universe works, without needing a supercomputer or waiting hours for results. It turns complex physics into an interactive, hands-on experience.

Technical Summary of "CosmoSlider: An educational tool for cosmology"

Problem Statement
Teaching the dependence of Cosmic Microwave Background (CMB) power spectra on cosmological parameters is a central component of modern cosmology education. Traditional methods rely on either static theoretical depictions or animations where a single parameter varies at a time. While interactive tools allowing simultaneous variation of multiple parameters exist, they face significant computational and storage constraints. Generating a full $N$ -dimensional grid of precomputed spectra for $N$ parameters with resolution $M$ requires $M^N$ spectra. For a standard six-parameter $\Lambda$ CDM model with a modest resolution of $M=10$ , this necessitates $10^6$ spectra, requiring hundreds of gigabytes of RAM (as illustrated in Figure 2 of the paper). Consequently, existing interactive tools often compromise by using low resolutions, limiting the number of free parameters, or enforcing constraints between sliders. Conversely, tools that compute spectra on-demand using Einstein–Boltzmann solvers (e.g., class or CAMB) lack real-time interactivity, requiring users to wait seconds for computation or forcing a "compute" button press that halts interaction.

Methodology
To overcome the trade-off between interactivity and computational cost, the authors developed CosmoSlider, a lightweight visualization tool that utilizes a neural-network emulator rather than precomputed grids or on-the-fly solvers.

Emulation Framework: The tool employs the connect framework to train an accurate artificial neural network emulator for the six parameters of the $\Lambda$ CDM model.
Deployment Architecture: The emulator is converted into a TensorFlow Lite model, enabling efficient execution on mobile devices and web browsers.
- iOS Version: Built with SwiftUI, allowing for the import of custom models (converted from connect) and offering features like model switching and literature resources.
- Web Version: Built with JavaScript and the TensorFlow Lite plugin, designed as a lightweight, static webpage suitable for embedding in educational materials.
Visualization Strategy: The tool visualizes observables (TT, TE, EE, and $\phi\phi$ power spectra) using native plotting utilities. To ensure smooth real-time updates (milliseconds per evaluation) while maintaining high slider resolution ( $10^2$ to $10^3$ ), the emulator computes only 100 $C_\ell$ values across the multipole range $\ell \in [2, 2500]$ . The remaining spectrum is reconstructed via cubic spline interpolation. The interface supports a hybrid logarithmic-linear scaling for the x-axis to detail both low and high multipoles.

Key Contributions

Real-Time Multi-Parameter Exploration: CosmoSlider enables the simultaneous variation of multiple cosmological parameters with immediate visual feedback, a capability previously limited by the storage requirements of precomputed grids or the latency of solvers.
Cross-Platform Accessibility: The tool is available as both a native iOS application (supporting iPhones and iPads) and a web-based tool, removing the need for specialized hardware or software installation for the web version.
Extensibility: The iOS version allows users to import custom neural network emulators trained with the connect framework, enabling the tool to adapt to different cosmological models beyond the standard $\Lambda$ CDM.
Educational Integration: The tool is designed specifically for classroom and self-guided use, featuring "Help" and "Literature" pages within the iOS app to bridge the gap between visual intuition and theoretical understanding.

Results and Capabilities
The paper demonstrates that the neural network emulator can evaluate a CMB spectrum in milliseconds, allowing for smooth slider interactions without the "warm-up" periods or mandatory wait times associated with running Einstein–Boltzmann solvers. The tool successfully visualizes the CMB power spectra (temperature and polarization) alongside observed data points from Planck, SPT, and other experiments. The authors note that the resolution of the sliders can be arbitrarily high ( $M \approx 10^2$ ) without impacting user experience, a significant improvement over the $M=10$ limit imposed by precomputed grid approaches for six dimensions.

Significance and Claims
The authors position CosmoSlider not as a replacement for rigorous numerical calculations using Einstein–Boltzmann solvers, but as a complementary educational resource. Its primary significance lies in lowering the technical barrier for students and lecturers to explore parameter dependencies in the CMB. By providing immediate visual feedback, the tool aids in building physical intuition regarding how cosmological parameters influence anisotropies. The paper claims that this approach demonstrates how modern machine learning emulation techniques can make complex physical models accessible in educational contexts, suggesting that similar methods could be valuable in other areas of physics education. Future updates are planned to include data export options, further customization, and the inclusion of additional observables (such as the matter power spectrum) and datasets.