Time & Space Complexity

🎯 Learning Objectives

• Define time and space complexity classes (P, NP, PSPACE, EXP)
• Explain the time hierarchy theorem
• Analyze relationships between complexity classes
• Distinguish between deterministic and nondeterministic complexity
• Understand the P vs. NP problem

1. Introduction to Complexity

1.1 Intuition

1.2 Asymptotic Notation

2. Time Complexity Classes

2.1 Definition

2.2 The Time Hierarchy Theorem

2.3 P vs. NP

3. Space Complexity Classes

3.1 Definition

3.2 Space vs. Time

• $P \subseteq NP \subseteq PSPACE = NPSPACE$ (Savitch's theorem)
• $P \subsetneq EXP$ (time hierarchy theorem)
• $PSPACE \subseteq EXP$ (space ≤ time)
• $P \neq EXPSPACE$ (if $P = PSPACE$, then $P = EXP$ too, which contradicts the hierarchy theorem)

3.3 Savitch's Theorem

4. The Landscape

5. Common Pitfalls

Pitfall 1: Confusing NP with "Not Polynomial"

Pitfall 2: Assuming P ≠ NP is the Default

Pitfall 3: Equating "Hard" with "Intractable"

6. Key Concepts Reference

7. 📝 Practice Questions

Q1: Why do we use polynomial time as the definition of "efficient" rather than linear or quadratic?

Answer: Polynomial time $(n^k)$ is robust: (1) It's closed under composition (polynomial of polynomial is polynomial). (2) It's independent of machine model (polynomial time on any reasonable model is polynomial on a TM). (3) Most practical algorithms run in polynomial time. (4) Problems with super-polynomial lower bounds are truly intractable. Linear time is too restrictive (many algorithms are O(n log n) or O(n²)), and exponential time includes truly infeasible problems. Q2: Show that any problem in NP can be solved in exponential time.

Answer: If $L \in NP$, there's an NTM N that decides $L$ in $O(n^k)$ time. We can simulate N using a DTM via breadth-first search of the computation tree. The branching factor of N is at most $b$ (finite), and the tree depth is $O(n^k)$. The total number of configurations is at most $b^{O(n^k)}$ which is $O(2^{n^k})$ for some $k$. So $L$ can be decided in $O(2^{n^k})$ time → $NP \subseteq EXP$. This is proven, not conjectured. Q3: State and explain Savitch's theorem.

Answer: Savitch's theorem: $NSPACE(f(n)) \subseteq SPACE(f(n)^2)$ for $f(n) \geq \log n$. Intuition: Nondeterministic space can be simulated deterministically with only quadratic space overhead. The proof uses divide-and-conquer: to check if a configuration can reach another in $t$ steps, check if there's a midpoint that can be reached in $t/2$ steps from the start and can reach the end in $t/2$ steps. This recursive approach reuses space (depth $O(f(n))$, each level $O(f(n))$ space → total $O(f(n)^2)$).

Key implication: $PSPACE = NPSPACE$ (polynomial space is unaffected by nondeterminism), unlike time where $P \neq NP$ is believed. Q4: What is the difference between a reduction $A \leq_p B$ and $A \leq_m B$?

Answer: $A \leq_m B$ (mapping reduction) requires only that the reduction function $f$ is computable (in finite time). $A \leq_p B$ (polynomial-time reduction) requires that $f$ is computable in polynomial time. $\leq_p$ is more restrictive — it preserves polynomial-time computability. $\leq_m$ is used in undecidability proofs; $\leq_p$ is used in NP-completeness proofs. All $\leq_p$ reductions are also $\leq_m$ reductions, but not vice versa. Q5: Why don't we know if P = PSPACE?

Answer: We know $P \subseteq NP \subseteq PSPACE$, but we can't prove proper inclusion for either containment. Proving $P \subsetneq PSPACE$ would imply $P \neq NP$ (since $NP \subseteq PSPACE$). Proving $P = PSPACE$ would imply $NP = PSPACE$ (since $NP \subseteq PSPACE$). Either result would be revolutionary, and both seem far beyond current techniques. The time hierarchy theorem gives $P \neq EXP$ but nothing separates $P$ from $PSPACE$ or $NP$ from $PSPACE$ with current methods. Q6: Place these problems in the most likely complexity class: (a) finding a path in a graph, (b) 3SAT, (c) QBF, (d) generalized chess.

Answer: (a) Path in graph → P (BFS/DFS runs in O(n+m) time) (b) 3SAT → NP (verifiable in polynomial time, believed not in P) (c) QBF → PSPACE-complete (quantified boolean formulas are the canonical PSPACE-complete problem) (d) Generalized chess → EXP (on an n×n board, generalized chess is EXP-complete) Q7: Explain the Time Hierarchy Theorem's significance.

Answer: The Time Hierarchy Theorem proves that $TIME(f(n)) \subsetneq TIME(f(n)^2)$ for time-constructible $f$. This means: (1) More time allows solving strictly more problems — there's always a decidable problem that needs more than $O(f(n))$ time. (2) $P \subsetneq EXP$ (since $P = \bigcup_k TIME(n^k)$ and $EXP = \bigcup_k TIME(2^{n^k})$, and the hierarchy theorem shows $TIME(n^k) \subsetneq TIME(2^n)$ for any $k$). (3) The proof uses diagonalization — constructing a TM on the fly that disagrees with every $O(f(n))$ time TM. Q8: If someone proves P = PSPACE, what other problems would be efficiently solvable?

Answer: If P = PSPACE, then every problem solvable in polynomial space is also solvable in polynomial time. This would include: (1) All problems in NP (SAT, TSP, graph coloring) — P = NP follows from P = PSPACE since NP ⊆ PSPACE. (2) All PSPACE-complete problems (QBF, generalized geography, many games). (3) Problems in PH (polynomial hierarchy) — the entire hierarchy collapses to P. This would be revolutionary — the biggest discovery in computer science — but is widely believed to be false.

8. 🔗 Cross-References

• Week 11 - NP-Hardness: Polynomial reductions
• Week 12 - NP-Completeness: Complete problems
• BSCS4021 (Advanced Algorithms): Algorithm analysis Join Discord PreviousReductionsNextNP-Hardness