Multiply two 2×2 matrices and you seem to need eight scalar multiplications. Strassen found a way with seven — and because you can apply it recursively to blocks, matrix multiplication drops from n3 to about n2.807. It was the shock that opened the whole field of fast matrix multiplication, still open today.
From the eight entries of A and B, form seven cleverly chosen products M1..M7 (each one multiplication of two sums/differences), then add and subtract them to get the four entries of C — using only 7 multiplications instead of 8, at the cost of more additions. Edit the matrices; the seven products and the result update live. live demo
“Strassen set out to speed matrix multiply and proved 7 is optimal” — he was trying to prove 8 is optimal, and it was Winograd who proved 7 is the floor. cited
The best possible exponent for matrix multiply is a famous open problem — still creeping toward 2. Strassen 1969, bound Winograd 1971
The seven products are pure scalar arithmetic; the matrices are 2-D arrays — and it runs bit-for-bit like the schoolbook product: