What does this equation mean?

Because a genuine tensor decomposition composes with itself, the rank number converts directly into an asymptotic cost exponent, and DeepMind’s own paper does not spell out what that conversion yields — this is the article’s own arithmetic, shown rather than asserted. Applying Strassen’s 2×2 trick twice to reach 4×4 gives an algorithm whose cost scales as $n^{\log_2 7}$ ; applying it k times to reach a 2^k $\times$ 2^k matrix costs 7^k multiplications for n = 2^k , the same exponent however you slice the recursion. A genuine rank-48 4×4 decomposition, applied the same way, would scale as $n^{\log_4 48}$ instead: $\log_4(49) = \log_2(7) \approx 2.8074 \qquad \log_4(48) \approx 2.7925$. The gap is small — about 0.015 in the exponent, on the order of half a percent — and it does not come close to the roughly 2.37 exponent that a completely different lineage of methods (the laser method and its descendants) has established as the theoretical frontier for matrix multiplication in general. Small-matrix rank hunting, whether by AlphaEvolve, by flip graphs, or by hand, has never been competing on that frontier; it is a different, more concrete kind of result, closer to a real building block than to a record book entry. But within its own lane — better recursive building blocks at small, practically implementable sizes — a rank reduction from 49 to 48 in a genuine, composable decomposition is not a rounding error. It is the exact kind of small, provable gain the field spent fifty-six years failing to find.