An N-term approximation in an orthonormal basis uses at most basis functions. For a chosen index set , its optimal coefficients are the inner products with those functions, and its squared error is the sum of the omitted squared coefficients by the Parseval identity.
New to topics? Read the docs here!