Top 100 Unsolved Mysteries of the Universe, Episode 98: The Joint Inversion Problem of the CMB and Large-Scale Structure. Imagine holding two cosmic archives. One is the universe's baby photograph: the cosmic microwave background, with temperature spots, polarization traces, and lensing distortions, like faint fingerprints left when the universe first became visible. The other is the mature city map built over billions of years: galaxy surveys, weak lensing, redshift-space distortions, cluster counts, and future 21-cm tomography, showing roads, junctions, traffic jams, reconstruction scars, and empty lots. The hard question is not whether both archives can be fed into a supercomputer to produce a tidy parameter set. It is whether the old photograph and the mature map are really describing the same construction process. The CMB is like an aerial photo of the foundation just after the concrete was poured: mostly linear patterns, shallow wrinkles, and no complex neighborhoods yet. Large-scale structure is like today's city of high-rises, bridges, subways, and traffic, where it is hard to tell which roads came from the first blueprint and which were later squeezed into place by mergers, feedback, commerce, and local accidents. Mainstream cosmology is powerful here. It can combine the CMB, galaxy surveys, lensing, BAO, RSD, cluster counts, simulations, covariance matrices, foreground cleaning, reconstruction pipelines, and posterior analysis in one joint machine. But the danger sits inside that power. The CMB mostly gives us an early, cleaner, more linear window. Large-scale structure arrives covered with late-time nonlinear growth, galaxy bias, baryonic feedback, sample selection, foreground leftovers, and survey-pipeline repainting. The same parameter can explain an early transfer function and also absorb a late-time systematic. A posterior curve may look beautiful without proving that one physical map has been recovered. It may simply mean that two ledgers from different eras, written with different readout standards and noise sources, have been glued together by a flexible parameter bucket. Like forcing a childhood photo and an adult face onto the same template, overlap is not the same as understanding growth. Joint cosmic inversion has the same weakness: we do not merely need image alignment. We need the construction diary between the early blueprint and the late road network. EFT's rewrite is not that the CMB is wrong, not that large-scale-structure surveys are useless, and not that joint inference should be abandoned. EFT rejects only the automatic upgrade from "the joint fit works" to "the physical ground map is unified." In EFT, the CMB is treated as the earliest, coarsest, and hardest-to-erase memory of the cosmic base state. Large-scale structure is that same memory later thickened by supply routes, mergers, feedback, environmental layering, and node hierarchy. The early universe did not merely throw down random seeds, and the late universe did not grow freely from nothing. There should be an inheritance chain from base texture to road network. EFT therefore asks a sharper question: within the same sky regions, can CMB cold and hot spots, polarization residues, and lensing distortions be translated into later galaxy filaments, voids, cluster nodes, weak-lensing shear, and velocity flows using one environmental grammar? If a directional residue is weak in the CMB, should it be amplified later in the galaxy web? If an early texture corresponds to a future matter corridor, do its ranking, orientation, node level, and environmental layering preserve a cross-window order? If yes, the baby photo and the city map are signing the same skeleton. If not, even a beautiful joint posterior may be only statistical glue. The EFT demand can be summarized in four words: cross-window migration. This is not adding one more free parameter. It is giving every observational window the same identity card. The early window should leave the base texture. The middle window should show how routes open. The late window should test whether nodes, flows, shear, and clustering grew along the same skeleton. Otherwise, the larger the joint fit becomes, the easier it is to smooth away the physical differences that matter most. CMB is the first old photograph. Large-scale structure is the later aerial map. Twenty-one-centimeter tomography is like scanning the construction process layer by layer. Clusters and weak lensing are like weighing how solid today's roads and junctions have become. A genuinely good joint inversion should not simply make all pictures look comfortable. It should predict the later road network from the early texture, then retrodict the early texture from the later road network, with direction, strength, hierarchy, and environment still matching at each step. The guardrail matters. EFT is not throwing away modern statistics. It is adding a harder physical acceptance test. The question is not only which parameters feel most comfortable. The question is whether those parameters hide a mechanism loop that can travel from CMB to large-scale structure and back again. Then joint inversion stops being a giant cosmological jigsaw puzzle and becomes a truth test for the universe's original plate: did the faint early wrinkles really grow into the same skeleton we now see in the galaxy web, velocity field, lensing field, and cluster nodes? EFT's core rewrite is this: joint inversion is not about piling up more data until the fit looks smoother. It is about making cosmic readouts from different ages recognize their shared ancestry. It is not a search for the parameter bucket best able to absorb tension, but for a real construction chain that migrates across windows and closes across cosmic time. Tap the playlist for more. Next episode: Unknown Systematics and New-Physics Degeneracy in Cosmological Data. Follow and share - our new-physics explainer series will help you see the whole universe more clearly.
EFT
Language