. 2023 Mar 15;120(12):e2216805120. doi: 10.1073/pnas.2216805120

Table 1.

IRAS captures physical relations

	Combination	#observables	Corr
I.9.18	$F - \frac{G m_{1} m_{2}}{{(i_{2} - i_{1})}^{2} + {(j_{2} - j_{1})}^{2} + {(k_{2} - k_{1})}^{2}}$	10	0.91
II.36.38	$f - \frac{μ_{m} B}{K_{b} T} - \frac{μ_{m} α M}{ϵ c^{2} K_{b} T}$	9	0.87
I.11.19	A − x₁y₁ − x₂y₂ − x₃y₃	7	0.92
I.29.16	$x - \sqrt{x_{1}^{2} + x_{2}^{2} - x_{1} x_{2} cos (θ_{1} - θ_{2})}$	5	0.92

Observations generated in FSReD (8) for four physical equations describing the law of force in planetary motion, spontaneous magnetization, scalar product of vectors, and the interference effect (28) serve as inputs to IRAS. The observables are corrupted by noise as described in the main text. The final column lists the Pearson correlation between the known physical equation and the combination learned by IRAS. The high correlation values testify to a correct identification of the underlying physical equation.