We generalize the notion of the best separable approximation (BSA) and best W-class approximation (BWA) to arbitrary pure state entanglement measures, defining the best zero-E approximation (BEA). We show that for any polynomial entanglement measure E, any mixed state ρ admits at least one "S-decomposition," i.e., a decomposition in terms of a mixed state on which E is equal to zero, and a single additional pure state with (possibly) non-zero E. We show that the BEA is not in general the optimal S-decomposition from the point of view of bounding the entanglement of ρ, and describe an algorithm to construct the entanglement-minimizing S-decomposition for ρ and place an upper bound on E(ρ). When applied to the three-tangle, the cost of the algorithm is linear in the rank d of the density matrix and has accuracy comparable to a steepest descent algorithm whose cost scales as d 8 log d. We compare the upper bound to a lower bound algorithm given by Eltschka and Siewert for the three-tangle, and find that on random rank-two three-qubit density matrices, the difference between the upper and lower bounds is 0.14 on average. We also find that the three-tangle of random full-rank three qubit density matrices is less than 0.023 on average.Non-classical correlations in quantum states such as entanglement distinguish quantum from classical information theory. The ability to calculate entanglement of mixed quantum states is relevant for the analysis of tomography data for systems of multiple qubits in several implementations [1][2][3]. Multipartite systems can contain multiple inequivalent types of entanglement that cannot be converted into one another by local operations and classical communication [4].One approach to characterizing pure-state entanglement in a system of qubits associates a polynomial function that is invariant under determinant 1 local operations with each type of entanglement [5][6][7]. Examples of such polynomial invariants include the concurrence for two qubits [8] and the three-tangle, which quantifies the amount of entanglement in a three-qubit system that cannot be accounted for by entanglement between pairs of the qubits [9].A polynomial invariant E is extended to mixed states by way of the convex roof, given for a rank-d density matrix ρ by:where E = {p i , |ψ i } is a pure-state ensemble for ρ and Υ ρ is the set of all such ensembles. Caratheodory's theorem allows us to restrict the optimization to ensembles containing no more than d 2 elements [10]. An ensemble that minimizes eq. (1) is said to be minimal. We consider the rank d of the density matrix d, rather than the dimension of the Hilbert space on which it acts, because d is the parameter that determines the computational difficulty of the convex roof minimization. A number of special cases of computation of the convex roof have been solved for cases of restricted rank [11][12][13].Minimal ensembles have been found analytically for the concurrence of arbitrary two-qubit mixed states [8], and for the three-tangle of rank-two mixtures of gener...