Balinski proves the result based on the correctness of the simplex method for finding the minimum or maximum of a linear function on a convex polytope (the linear programming problem). The simplex method starts at an arbitrary vertex of the polytope and repeatedly moves towards an adjacent vertex that improves the function value; when no improvement can be made, the optimal function value has been reached.
If S is a set of fewer than d vertices to be removed from the graph of the polytope, Balinski adds one more vertex v0 to S and finds a linear function ƒ that has the value zero on the augmented set but is not identically zero on the whole space. Then, any remaining vertex at which ƒ is non-negative (including v0) can be connected by simplex steps to the vertex with the maximum value of ƒ, while any remaining vertex at which ƒ is non-positive (again including v0) can be similarly connected to the vertex with the minimum value of ƒ. Therefore, the entire remaining graph is connected.
Ziegler, Günter M. (2007) [1995], "§3.5: Balinski's Theorem: The Graph is d-Connected", Lectures on Polytopes, Graduate Texts in Mathematics, vol. 152, Springer-Verlag, p. 95–, ISBN 978-0-387-94365-7. 978-0-387-94365-7 ↩
Balinski, M. L. (1961), "On the graph structure of convex polyhedra in n-space", Pacific Journal of Mathematics, 11 (2): 431–434, doi:10.2140/pjm.1961.11.431, MR 0126765. /wiki/Michel_Balinski ↩
Steinitz, E. (1922), "Polyeder und Raumeinteilungen", Encyclopädie der mathematischen Wissenschaften, Band 3 (Geometries), pp. 1–139. /wiki/Ernst_Steinitz ↩