Imagine you have a class Square in C++, which represents a square on the board. Two options are readily apparent for storage, assuming the grid itself is dense:

class Square
{
public:
    Square()
    {
        // initialize `nearestNeighbors_`
    }

    const std::array<Square*, 4>& GetNearestNeighbors() const
    {
        return nearestNeighbors_;
    }

private:
    std::array<Square*, 4> nearestNeighbors_;
};

OR

class Square
{
public:
    Square()
    {
        // initialize `nearestNeighbors_`
    }

    std::array<Square*, 4> GetNearestNeighbors() const
    {
        std::array<Square*, 4> nearestNeighbors = /* ... init .../;

        return nearestNeighbors;
    }
};

Either solution would work. The first consumes more memory; the second requires more calculations. Which would make more sense for a real game map?