Geophysics has always sought a language to converse with the depths. Well drilling is a crude alphabet, laboratory core tests are slow decoding. A new approach turns humming porosity into a pure note. Each sandstone sample, captured by X-ray microtomography (with photometric accuracy akin to absorption spectroscopy), becomes a key. A light convolutional neural network 'presses' it, and at the output, Young's modulus and bulk modulus sound. There's no loose metaphor in this mechanism: the network architecture, reminiscent of a simplified visual analyzer, captures pore patterns as naturally as a musician recognizes an instrument by the wooden curves of its body. Hooke and Augustin-Louis Cauchy laid down the alphabet of elasticity three centuries ago; now the neural network reads it fluently, bypassing cumbersome finite-element symphonies. The Hashin-Shtrikman bounds – a kind of limits of elastic response, akin to the maximum mass of white dwarfs – set a strict range within which the prediction falls, ensuring physical meaningfulness.

The secret of the speed lies in hierarchical homogenization. A large chunk of rock is divided into thousands of sub-grids, and for each, the neural network predicts effective moduli. Trained to output normalized factors, it is lightning-fast: the output value is clamped between zero and one, just as a piano tuner knows a string won't sound above or below an octave. When predictions are assembled back into large blocks, the total error falls almost entirely on the mathematical 'upscaling', not on the network. The accuracy is phenomenal: the coefficient of determination R² exceeds 0.99, as if a chamber ensemble played to absolute unison. And here a curious parallel with astrophysics arises: just as gravitational lensing reconstructs the distribution of dark matter from many distorted images, geophysicists stack the 'lenses' of sub-grids to see the invisible structure of porous media.

Success on five different sandstones, from pure Fontainebleau to clayey Castlegate, promises the imminent advent of digital twins of entire formations. CO2 management, production optimization, fracture prediction – all of this will be able to be 'played out' in real time. The method subtly echoes ideas from effective field theories: without examining each pore, we accurately predict the macro-behavior of rock. Just as spacetime curvature emerges from matter distribution, effective elasticity is born from pore geometry. The next step is accounting for plasticity, fluid flows, and nonlinear effects; then we will literally be able to 'hear' how stone cracks under load, long before the actual experiment.