◄ WORLD V · SONNY 5DART 209 · a helldive at the net

THE ADJOINT move the matrix to the other side of the inner product

For every linear map A there is an adjoint A* that satisfies ⟨Ax, y⟩ = ⟨x, A*y for all vectors — you may apply A on the left of the inner product or its adjoint on the right, and get the same number. In finite dimensions A* is just the transpose Aᵀ. This is the duality under every gradient: reverse-mode automatic differentiation (backprop) is the adjoint of the forward pass — the transpose walked backwards.

THE TECHNIQUE ⟨Ax, y⟩ = ⟨x, Aᵀy⟩ for all x, y

Pick a matrix and two vectors; compute ⟨Ax, y⟩ and ⟨x, Aᵀy⟩ independently and watch them land on the same scalar — the matrix slid across the inner product: live demo


HISTORY & CREDIT Hilbert / Riesz, early 1900s

“The transpose is just bookkeeping — rows become columns.” — the transpose is the adjoint, the dual map that lets a matrix cross the inner product. That is not cosmetic: it is precisely why backpropagation works — the backward pass applies Aᵀ where the forward pass applied A. cited

1900s · Hilbert, Riesz, Fredholm — the adjoint operator on inner-product spaces; ⟨Ax,y⟩=⟨x,A*y⟩ defines A*.
1960s · adjoint / costate method — control theory (Pontryagin) solves gradient problems by running the adjoint system backwards.
1970–86 · Linnainmaa; Rumelhart–Hinton–Williams — reverse-mode AD / backpropagation: the adjoint of the forward pass, the transpose walked in reverse.

Forward applies A; the gradient applies Aᵀ in the opposite order. Every backward pass in every neural net is this one duality, run at scale. the adjoint

RECOMMEND FOR I-13 ⟨Ax,y⟩ = ⟨x,Aᵀy⟩ to the bit, computed

On the canonical compiler, with A=[[1,2],[3,4]], x=[5,6], y=[7,8], both sides of the adjoint identity equal 431:

$ i13 run adjoint.i13 # A=[[1,2],[3,4]], x=[5,6], y=[7,8] = 431 = 431 difference = 0 -- the matrix slides across the inner product
Recommend: the adjoint is LIT for I-13 — verified ⟨Ax,y⟩ = ⟨x,Aᵀy⟩ = 431, difference 0, from a flat matrix and two vectors. This is the exact duality that backpropagation runs — forward applies A, the gradient applies Aᵀ — so the dart is also the seed of reverse-mode AD (dart 206's mirror).