Fredholm alternative

The Fredholm alternative is the infinite-dimensional rescue of a fact you trust for square matrices: for the system A x = b with A square, either A is invertible (a unique solution for every b) or it is not (the homogeneous system has nonzero solutions and b must satisfy compatibility conditions). The alternative says this clean either/or survives in infinite dimensions for one crucial family of operators — identity minus a compact operator.

Precisely: let K be a compact operator on a Hilbert (or Banach) space and consider T = I - K. Then exactly one of two situations holds. Either T is invertible, so T x = y has a unique solution for every y; or the homogeneous equation T x = 0 has a nonzero solution, in which case its solution space is finite-dimensional, the adjoint equation T^* z = 0 has a solution space of the same finite dimension, and T x = y is solvable precisely when y is orthogonal to every solution of T^* z = 0.

Why it is the infinite-dimensional rank-nullity: for I - K, injectivity and surjectivity become equivalent again — uniqueness implies existence and vice versa, exactly as for square matrices. The shift operator showed this equivalence dies for general bounded operators; the Fredholm alternative shows it is reborn for compact perturbations of the identity. The dimensions of the kernel and cokernel match, so the index of I - K is zero.

Why it matters: this is the backbone of integral equations. An equation f(x) - integral K(x,y) f(y) dy = g(x) with a nice kernel is exactly (I - K) f = g with K compact, so the Fredholm alternative tells you precisely when it is uniquely solvable and, when it is not, the finitely many solvability conditions on g. It is the rigorous foundation for boundary-value problems solved via integral-equation methods.