CosmoSlider: Interactive Cosmology

• CosmoSlider is an interactive suite of visualization tools that enables real-time exploration of theoretical predictions and observational data through adjustable cosmological parameters.
• It employs numerical integration, Gaussian Process emulators, and TensorFlow Lite neural networks to deliver high-fidelity, sub-second predictions for observables such as CMB power spectra and distance moduli.
• By integrating extensive observational datasets and simulation-based models, CosmoSlider supports both educational exercises and advanced research workflows in cosmological inference.

CosmoSlider is a suite of interactive visualization and emulation tools for cosmology, fundamentally designed to provide real-time exploration of theoretical predictions and observational data across a wide range of cosmological models. Its key components—parameter sliders, immediate plotting, built-in data sets, and fast emulation routines—support both research and education by enabling direct manipulation of cosmological parameters and instant evaluation of their effects on observables such as the cosmic microwave background (CMB) power spectra, supernova distance moduli, baryon acoustic oscillations (BAO), and more. The CosmoSlider architecture spans Java/JavaScript applications for general cosmological parameter inference (Moldenhauer et al., 2012), simulation-driven emulators for cosmic shear analyses (Harnois-Deraps et al., 2019), and, most recently, TensorFlow Lite neural network models specialized for rapid CMB visualization on mobile and web platforms (Nygaard et al., 23 Jan 2026). These tools employ rigorous theoretical modeling, high-performance numerical methods, and advanced machine learning to deliver sub-second, high-fidelity predictions and facilitate both pedagogical and research-driven workflows in cosmological data analysis.

1. User Interface and Visualization Features

CosmoSlider implementations are characterized by intuitive, real-time parameter control and dynamic plotting environments. The original Java-based version (Moldenhauer et al., 2012) offers continuous sliders for $H_0$, $\Omega_m$ (internally split into $\Omega_b$ and $\Omega_c$), $\Omega_\Lambda$, $\Omega_k$, and dark energy equation-of-state parameters ($w_0$, $w_a$). Each parameter adjustment triggers instant recalculation of model predictions (e.g., $\mu(z)$, $H(z)$, BAO ratios, CMB shift parameters, and scale factor evolution) via numerically integrated FLRW equations. Observational data from 18 curated surveys are built in, with residuals against theory rendered on-the-fly.

In CMB-focused variants (Nygaard et al., 23 Jan 2026), the interface includes six sliders for standard $\Lambda$CDM parameters ($\Omega_b h^2$, $\Omega_c h^2$, $H_0$, $A_s$, $n_s$, $\tau_{\rm reio}$), overlaid with multi-spectrum plotting (TT, TE, EE, and lensing $\phi\phi$) and Planck/SPT data reference curves. Plot panels support hybrid log-linear scaling and peak-position annotation. Evaluation latency is typically $\sim$5–10 ms, enabling fluid exploration within lectures, classroom contexts, or individual study.

2. Theoretical Framework and Mathematical Formalism

CosmoSlider tools are grounded in the Friedmann–Lemaître–Robertson–Walker (FLRW) cosmological model, with explicit parametric forms for expansion rates and observable predictions. The expansion rate $E(z) = H(z)/H_0$ is calculated from

$$H(z) = H_0 \sqrt{ \Omega_m (1 + z)^3 + \Omega_k (1 + z)^2 + \Omega_\Lambda (1 + z)^{3(1 + w_0 + w_a)} \exp \left[ -\frac{3w_a z}{1 + z} \right] }$$

Distances and time scales are computed via

$$d_L(z) = (1+z) \frac{c}{H_0} \int_0^z \frac{dz'}{E(z')}$$

$$\mu(z) = 5 \log_{10} \left[ \frac{d_L(z)}{\text{Mpc}} \right] + 25$$

$$D_V(z) = \left[ \left( \frac{c z}{E(z)} \right) \left( d_A(z) \right)^2 \right]^{1/3}, \quad d_A(z) = \frac{d_L}{(1+z)^2}$$

$$t(z) = \int_z^\infty \frac{dz'}{(1+z') H(z')}, \qquad a(t) = \frac{1}{1+z}$$

For cosmic shear analyses (Harnois-Deraps et al., 2019), the focus extends to two-point correlation functions $\xi_\pm(\theta;p)$ and covariance matrices $\Sigma(p)$, with emulator predictions embedded in likelihood evaluation and Fisher forecasting pipelines.

CMB power spectrum emulation (Nygaard et al., 23 Jan 2026) uses a neural network mapping $E: \mathbb{R}^6 \rightarrow \mathbb{R}^{100}$, with $E(\theta)_i \approx C^{X}_{\ell_i}(\theta),\ X\in${TT, TE, EE}, and final $C_\ell^X(\theta)$ reconstructed by cubic-spline interpolation across sampled $\ell_i$.

3. Emulator Architectures and Computational Methods

CosmoSlider's emulation technologies span both physical modeling and machine learning. The original suite (Moldenhauer et al., 2012) uses numerical integration (Romberg’s method) on each slider update, feeding observables directly into $\chi^2$ scoring against survey data.

The cosmic shear variant (Harnois-Deraps et al., 2019) embeds a principal component–compressed, Gaussian Process regression emulator for both signal and covariance structure. Training occurs on 25 nodes of a Latin hypercube in $[\Omega_m, \sigma_8, h, w_0]$, each anchored by 800 pseudo-independent simulations (via matched-pair $N$-body + ray-tracing) yielding high-fidelity $\xi_\pm$ and $\Sigma$. Emulator kernel hyperparameters are fit via marginal likelihood maximization with squared-exp Gaussian kernels. On-the-fly evaluation produces sub-5% relative errors for $\xi_\pm$ and 10% for covariance diagonals inside the sampled hypercube.

For CMB spectra (Nygaard et al., 23 Jan 2026), the neural-network emulator is built in the CONNECT framework, exported as a $\sim$1–2 MB TensorFlow Lite model. It consists of four hidden layers (128 tanh neurons each) mapping six $\Lambda$CDM inputs to 100 $C_\ell$ outputs, with sub-percent error in mid/high-$\ell$ and $\leq$2% in the Sachs–Wolfe regime. Emulation replaces the need for precomputed grids or online Einstein–Boltzmann solves (CLASS, CAMB), yielding a $\sim$100-fold speedup ($\sim$5–10 ms per spectrum).

4. Integration of Observational Data and Statistical Inference

The tools incorporate extensive observational datasets for model validation and parameter constraint. The Java/JavaScript CosmoSlider (Moldenhauer et al., 2012) includes Union2 SN Ia compilations (up to $z \sim 1.4$), gamma-ray bursts ($\mu(z) > 5$), $H(z)$ cosmic chronometer measurements, BAO ratios ($r_s/D_V$ at $z \sim 0.1$–0.6), and CMB shift parameters ($R$, $l_a$, $z_*$). Each dataset is loaded with its redshift bins, observed values, and $1\sigma$ uncertainties. The code computes $\chi^2$ goodness-of-fit with residual and summary panels per probe.

Cosmic shear implementations (Harnois-Deraps et al., 2019) process two-point statistics and covariance matrices from large suites of $N$-body light-cone simulations, closely matching brute-force ensemble estimators to within 6% area error on parameter-error ellipses. Fisher forecasting is performed under Gaussian-likelihood assumptions, with explicit sensitivity to covariance evaluation at different $p$ points: shifting the cosmology where $\Sigma$ is evaluated changes constraint areas by factors of $2$–$5$, much larger than covariance-method choice impact.

CMB emulators (Nygaard et al., 23 Jan 2026) are calibrated to CLASS-generated spectra and allow overplotting of Planck and SPT measurements, facilitating direct visual comparison during interactive parameter sweeps.

5. Practical Implementation, Performance, and Deployment

CosmoSlider’s distributed versions address distinct deployment scenarios:

• Java/JavaScript/EJS version (Moldenhauer et al., 2012): Standalone, open-source under GPL-style license. Runs on Java 8+ or in-browser; installation via ZIP extract and JAR launch, with no command-line options required. Supports loading custom ASCII datasets, and numerical integration tolerances are user-configurable.
• Cosmic shear emulator (Harnois-Deraps et al., 2019): Algorithmic recipe for “on-the-fly” prediction within MCMC or grid samplers. Training is based on simulation campaigns of matched N-body plus ray tracing, with PCA+Gaussian Process emulators for fast likelihood evaluation. Interpolation at each parameter point replaces slow, direct simulation runs.
• Neural-network CMB emulator (Nygaard et al., 23 Jan 2026): Available as an iOS app or web-based tool, utilizing TensorFlow Lite deployment. Offline use requires $\sim$2 MB model storage and $<$20 MB RAM. Web usage benefits from model caching for repeated sessions. Evaluation time per spectrum is $5$–$10$ ms on consumer hardware, facilitating synchronous, interactive sessions in classroom or research settings.

6. Educational Applications and Exemplary Workflows

CosmoSlider has proven utility in both pedagogical and research contexts. Its real-time response to parameter changes enables intuitive investigation of cosmological phenomena, parameter degeneracies, and physical mechanisms. Suggested educational exercises with the CMB emulator (Nygaard et al., 23 Jan 2026) include tracking acoustic peak amplitude shifts under $\Omega_b h^2$ variation and reionization bump structure under $\tau_{\rm reio}$ shifts. Reported user feedback from university trials indicates increased confidence (>80% of students) in explaining parameter effects after short engagement periods.

For parameter inference workflows, CosmoSlider supports systematic adjustment of $\Lambda$CDM (or wCDM) parameters to recover best-fit scenarios (e.g., showing observable curves approaching survey data for $H_0=70$, $\Omega_b=0.045$, $\Omega_c=0.225$, $\Omega_\Lambda=0.73$, $w_0=-1.0$, $w_a=0$, and $\Omega_k=0$ yields $\chi^2_{\rm total}/$dof $\approx 1.0$ and model-data overlap for all probes (Moldenhauer et al., 2012)). The CMB emulator delivers physical intuition for the interplay of cosmological parameters by continuously visualizing resulting spectra.

Planned extensions for CosmoSlider include support for additional observables ($P(k)$, $H(z)$, lensing $\phi\phi$), custom emulator integration, enhanced plotting options, figure export, and deployment of Jupyter-notebook interactive controls (Nygaard et al., 23 Jan 2026).

7. Context, Variants, and Impact

CosmoSlider and its associated tools encapsulate current best practices in interactive cosmological model exploration, incorporating numerical, simulation-based, and machine-learning paradigms for accurate, rapid, and visually intuitive computation. Its methodology influences both education and research by lowering barriers to engagement with large datasets and high-dimensional model spaces.

Related tools—cosmo-SLICS for large simulation suites (Harnois-Deraps et al., 2019), COSMOEJS for observation-driven parameter fitting (Moldenhauer et al., 2012), and CONNECT–TensorFlow Lite neural-network emulators (Nygaard et al., 23 Jan 2026)—share the philosophy of instant feedback, broad coverage of parameter regimes, and rigorous statistical comparison. This suggests a growing consensus that hybrid simulation-emulator frameworks are central to cosmological inference and education. A plausible implication is the broader adoption of neural-network emulators and GPU-accelerated numerical codes for real-time model-data comparison across the cosmology community.

CosmoSlider thus represents a convergence of pedagogical utility, technical rigor, and computational efficiency in contemporary cosmological analysis.