Knight Probability - Implementation

Here's the solution with caching (to avoid recomputing the same dp values):

function knightProbability(n, k, row, col)
    dp := 3D array [n][n][k + 1], all -1
    return solve(row, col, k, n)
function solve(r, c, k, n)
    if r < 0 or r >= n or c < 0 or c >= n then
        return 0
    if k = 0 then
        return 1
    if dp[r][c][k] ≠ -1 then
        return dp[r][c][k]
    moves := [(-2,-1),(-2,1),(-1,-2),(-1,2),(1,-2),(1,2),(2,-1),(2,1)]
    prob := 0
    for (dr, dc) in moves
        prob := prob + solve(r + dr, c + dc, k - 1, n) / 8
    dp[r][c][k] := prob
    return prob

Time: $O(n^2 \cdot k)$. Space: $O(n^2 \cdot k)$ for the $3$D memoization table.