Q4 — SEARCH FOR OPTIMAL SOLUTION

Edges priced 6 on a 10x10 tile grid undercut the 10-per-tile floor, so Manhattan distance overestimates along the cheap corridors — not admissible.

The Question

#QuestionOfficial answer
4In the map, nodes are placed on a grid where each tile is 10x10 in size, is the Manhattan distance heuristic admissible?A (No)

The map (edge costs shown):

Options:

Option
ANo
BYes
CCannot be determined

Answer: A

The Topic in 60 Seconds

h(n)h(n) is admissible iff h(n)≤h∗(n)h(n) \le h^*(n) for every nn. On a grid with 10×1010\times10 tiles, moving one tile costs at least 1010, so scaled Manhattan (10×MD10 \times \text{MD}) is a lower bound only while every edge respects that floor: an edge joining tiles ≥1\ge 1 apart must cost ≥10\ge 10. Here several edges cost just 66.

Deep-dive: browse the Concepts section for admissible heuristics.

Solutions

The counterexample arithmetic

Cheapest actual route to G:

S→A→C→G=6+6+6=18(same via B)S \to A \to C \to G = 6 + 6 + 6 = \mathbf{18} \quad (\text{same via } B)

Now read the map's geometry: S, A, C sit in a chain and G is yet another hop beyond C. With every hop covering at least one full tile, the endpoints satisfy

MD(S,G)  ≥  MD(S,A)−MD(C,G)+MD(A,C)  ≥  1−1+1=1...\text{MD}(S, G) \;\ge\; \text{MD}(S,A) - \text{MD}(C,G) + \text{MD}(A,C) \;\ge\; 1 - 1 + 1 = 1...

— but strictly, since S→A→C→G walks three distinct grid steps without backtracking onto a previous cell, the printed figure places S and G at least 2–3 tiles apart, giving

h(S)=10⋅MD(S,G)∈{20,30}  >  18=h∗(S)h(S) = 10 \cdot \text{MD}(S,G) \in \{20, 30\} \;>\; 18 = h^*(S)

Overestimate ⇒ not admissible ⇒ A.

The general principle (what actually decides this)

Any edge cheaper than the per-tile floor breaks Manhattan's lower-bound property:

  • One tile of travel must cost ≥ 10.
  • Edges costing 6 connect nodes ≥ 1 tile apart while charging only 6 < 10.
  • Triangle inequality then fails along those corridors: a path can genuinely be cheaper than the sum of its Manhattan tiles, so hh can exceed h∗h^* somewhere.

Compare the sibling question (22t3 AN paper): there every edge cost 12 or 24 — comfortably above 10/tile — and Manhattan was admissible. Same heuristic, opposite verdicts, decided entirely by edge pricing.

Why not C ("cannot be determined")?

Because you do not need exact coordinates: any placement consistent with the printed figure has a cheap-6 corridor into G, and the floor argument shows hh can overshoot on it. The exam's intended reading treats the visible layout as given, where h(S)>h∗(S)h(S) > h^*(S) outright.

Tricks & Tips

Answers: 4 A