題解/算法 {1145. 北极通讯网络}

link

Solution

---

Firstly, let’s set a fallible device;
Let the ordered edges of MST be e1, e2, ..., em, for the largest edge em: (a-b), we make a, b to be Super-Point;
This greedy device is erroneous, e.g., the MST is a---(10)---b---(1)---c---(7)---d, $k=3$, firstly, in a-b, we set a, b Super-Point; when we arrive at c-d, $k=1$ can not eliminate this edge 7, so the answer is 7. The correct device is make a, b, d (b can be replaced to c) to Super-Point; then for c, it can c -> b to arrive at b, once a point arrived a Super-Point, then it can go to everywhere costless; so answer is 1;

---

It is very important to comprehend this problem, especially understand the nature of Super-Point;

Given you a totally connected undirected graph $G$ (contains $n\ast (n-1)/2$ undirected edges)
The goal is to make every pair of points to be accessible; suppose the answer is $D$, i.e., all edges can be used are $\le D$, or you can imagine that a graph $G1$ consists of $n$ points and with all edges are $\le D$.
There are two cases for the accessibility of two points:
1 a -> ... -> b where every edge belongs to $G1$
2 a -> ... s -> b where $s$ is a Super-Point, then every edge in $...$ is $\le D$; it is vital to realize that s -> b is costless i.e., once you arrived at a Super-Point, then you can go to any point with no cost;
Therefore, due to the existence of Super-Point, $G1$ maybe not connected graph;

Lets discuss by cases:
+ if $k=0$ that is no Super-Point, then $G1$ must be a connected-graph, $D$ be the maximal edge in $G1$; to make $D$ as small as possible, (according to the property of MST, i.e., a MST is $a1\le a2\le ...$, for any ST is $b1\le b\mathrm{2...}$, then we have $bi\ge ai$, as a lemma, the maximal edge in MST must be the least value among that in all ST), we just to make $G1$ be the MST of $G$;

+ If $k\ge n$, all points are Super-Point, there could be no edges in $G1$, so the answer is $0$.

+ Now, $k\in [1,n)$, let all Super-Points be $s1,...,sk$; Let $s1,...,sk$ also denotes a set, initially the set of $si$ has only one point $si$.
. Let us only focus on all Normal-Points (non-Super-Point) $a1,a2,...$, if the distance of $ai,aj$ is $\le D$, then we connect a edge between them; after this process, all isolated points $a1,a2,...$ originally will become a graph consists of several connected-sub-graphs and all edges are $\le D$.
For any sub-graph $SG$ (it’s connected, and consists of $ai$), we reduce it to be a ST (Spanning-Tree) (that will not affect the connection). Due to the goal of this problem, there must exists a edge $(a,b)w$ where $a\in SG$, $b$ is a Super-Point and $w\le D$ (only this is satisfied, $SG$ can only reach to any other point); such a edge maybe not unique, we can just choose one. Then, we connected a edge between $(a,b)$ for all $SG$; as a result, every $SG$ belongs to a unique root $si$. Finally, it becomes a graph in which there exists a Tree rooted by $si$; in other words, the graph consists of $k$ connected-sub-graphs where every sub-graph is also a Tree (i.e., we can add $k-1$ additional edges to make it to be a ST of $G$)
Now, you can imagine that there are $k$ points $s1,...,sk$ in $G$, every point is also a independent Spanning-Tree $STi$ with all edges $\le D$.
The cost is the maximal edge in the graph ($max(max(ST1),max(ST2),...))$)
Because it corresponds to a partition of the ST of $G$ (or you can say that it is a sub-graph of ST of G), that is, if we remove $k-1$ edges of the ST of $G$, then it becomes such a graph; (the cost equals to the maximal edge in the graph)
Let the ordered sequence of a ST is $e0,e1,...e_{n-2}$, we need to remove $k-1$ edges, of course, we should remove the maximal $k-1$ edges, the new maximal edge is $e_{n-1-k}$; now, it becomes $k$ sub-STs, the cost $max(max(STi))$ equals $e_{n-1-k}$;
Due to the property of ST: if MST is $a1\le a2,...$, then any ST will be $b1\le b\mathrm{2...}$ where $bi\ge ai$; Therefore, $e_{n-1-k}$ in MST must $\le$ that in any other ST;

Code

mst.Work();
sort( Paths.begin(), Paths.end());
double ans;
if( k == 0){
    ans = *Paths.rbegin();
}
else if( k >= n){
    ans = 0;
}
else{
    ans = Paths.at( Paths.size() - 1 - (k - 1));
}