The Complexity Class Hierarchy

Core Idea

The major complexity classes form a hierarchy: L ⊆ NL ⊆ P ⊆ NP ⊆ PSPACE ⊆ EXPTIME. The time hierarchy theorem, proved by diagonalization, guarantees that strictly more time yields strictly more computational power: DTIME(n) ⊊ DTIME(n²). Similarly, the space hierarchy theorem shows DSPACE(log n) ⊊ DSPACE(n). It is proven that P ⊊ EXPTIME, so the hierarchy is strict overall, but the intermediate separations — P vs. NP, NP vs. PSPACE — remain open. The polynomial hierarchy extends NP with alternating quantifiers, analogous to the arithmetical hierarchy.

How It's Best Learned

Study the hierarchy theorem proofs as applications of diagonalization — the same technique used for the halting problem. Then map out which containments are proven strict and which remain open, building an accurate picture of current knowledge.

Common Misconceptions

• The complexity class hierarchy is not fully proven strict at every level; P ≠ NP and NP ≠ PSPACE are both unresolved.
• If P = NP were proven, the entire polynomial hierarchy would collapse to P, dramatically simplifying our picture of computational complexity.

Explainer

You already understand P and NP as the heart of the complexity landscape — P is what deterministic polynomial time can decide, NP is what nondeterminism adds. But these two classes sit inside a much richer landscape. Think of computational resources — time and space — as two different currencies. Using more of one can sometimes substitute for the other, and the complexity hierarchy maps out exactly which tradeoffs are possible and which boundaries are real.

PSPACE is the class of problems solvable with polynomial *space*, regardless of how long computation takes. Because space can be reused across time steps, PSPACE is potentially much larger than NP: a problem might require exponential time but only polynomial space to solve, since you can reuse the same memory cells for many different subcomputations. The canonical PSPACE-complete problem is TQBF (totally quantified Boolean formulas), which asks whether a formula with alternating universal and existential quantifiers is true — the same structure as adversarial games. This connects PSPACE to problems like determining the winner of a board game under optimal play.

Above PSPACE sits EXPTIME — problems solvable in exponential time — and we *do* know that P ⊊ EXPTIME is a strict containment. The proof uses diagonalization, the same technique you saw with Cantor and the halting problem. The time hierarchy theorem shows that given strictly more time, you can solve strictly more problems: DTIME(n^k) ⊊ DTIME(n^k+1) for all k. The argument constructs a language by diagonalizing against all machines running in time T(n), building a problem that deliberately answers differently from every such machine. Similarly, the space hierarchy theorem shows DSPACE(log n) ⊊ DSPACE(n).

The polynomial hierarchy (PH) extends NP with alternating quantifiers: Σ₁ = NP, Π₁ = co-NP, Σ₂ = NP^NP (NP with an NP oracle), and so on. This mirrors the arithmetical hierarchy you may know from logic. A key structural fact: if P = NP, then PH collapses to P — every level becomes equivalent. This is a reason (not a proof!) that P ≠ NP is expected: the polynomial hierarchy seems to be genuinely infinite. The current picture is: we know P ⊊ EXPTIME strictly, we know P ⊆ NP ⊆ PSPACE ⊆ EXPTIME, but the intermediate separations — P vs. NP, NP vs. PSPACE — remain among the deepest open problems in mathematics.