#include <algorithm>
#include <bitset>
#include <cassert>
#include <chrono>
#include <cmath>
#include <complex>
#include <cstdint>
#include <fstream>
#include <functional>
#include <iomanip>
#include <iostream>
#include <limits>
#include <list>
#include <map>
#include <queue>
#include <random>
#include <regex>
#include <set>
#include <sstream>
#include <stack>
#include <string>
#include <sys/timeb.h>
#include <vector>

using namespace std;

#define repr(i, a, b) for (int i = (int)(a); i < (int)(b); i++)
#define rep(i, n) repr(i, 0, n)
#define reprrev(i, a, b) for (int i = (int)(b)-1; i >= (int)(a); i--)
#define reprev(i, n) reprrev(i, 0, n)
#define repi(itr, ds) for (auto itr = ds.begin(); itr != ds.end(); itr++)
#define chmin(mi, val) mi = min(mi, val)
#define chmax(ma, val) ma = max(ma, val)
#define all(a) (a).begin(), (a).end()
#define rall(a) (a).rbegin(), (a).rend()
#define mp make_pair
#define mt make_tuple
#define INF 1050000000
#define INFR INT_MAX
#define INFL (long long)(4e18)
#define INFLR LLONG_MAX
#define EPS (1e-10)
#define MOD 1000000007
//#define MOD 998244353
#define PI 3.141592653589793238
#define RMAX 4294967295

using vi = vector<int>;
using vvi = vector<vector<int>>;
using vvvi = vector<vector<vector<int>>>;
using vvvvi = vector<vector<vector<vector<int>>>>;
using ll = long long;
using vll = vector<ll>;
using vvll = vector<vector<ll>>;
using vvvll = vector<vector<vector<ll>>>;
using vd = vector<double>;
using vvd = vector<vector<double>>;
using vb = vector<bool>;
using vvb = vector<vector<bool>>;
using vc = vector<char>;
using vvc = vector<vector<char>>;
using vs = vector<string>;
using vvs = vector<vector<string>>;
using Pi = pair<int, int>;
using vPi = vector<Pi>;
using vvPi = vector<vector<Pi>>;
using vvvPi = vector<vector<vector<Pi>>>;
using vvvvPi = vector<vector<vector<vector<Pi>>>>;
using Pll = pair<ll, ll>;
using vPll = vector<Pll>;
using Pd = pair<double, double>;
using vPd = vector<Pd>;
template <class T>
using vec = vector<T>;
template <class T>
using pql = priority_queue<T, vector<T>, greater<T>>;
using Comp = complex<double>;

// vvvvvvvvvvvvvvvvvvvvvvv debug output vvvvvvvvvvvvvvvvvvvvvvv
// vector input
template <typename T>
istream &operator>>(istream &is, vector<T> &vec) {
    for (T &x : vec) is >> x;
    return is;
}
// pair
template <typename T, typename U>
ostream &operator<<(ostream &os, const pair<T, U> &pair_var) {
    os << "(" << pair_var.first << ", " << pair_var.second << ")";
    return os;
}
// vector
template <typename T>
ostream &operator<<(ostream &os, const vector<T> &vec) {
    os << "{";
    for (int i = 0; i < vec.size(); i++) {
        os << vec[i] << (i + 1 == vec.size() ? "" : ", ");
    }
    os << "}";
    return os;
}
// deque
template <typename T>
ostream &operator<<(ostream &os, const deque<T> &vec) {
    os << "{";
    for (int i = 0; i < vec.size(); i++) {
        os << vec[i] << (i + 1 == vec.size() ? "" : ", ");
    }
    os << "}";
    return os;
}
// map
template <typename T, typename U>
ostream &operator<<(ostream &os, const map<T, U> &map_var) {
    os << "{";
    repi(itr, map_var) {
        os << *itr;
        itr++;
        if (itr != map_var.end()) os << ", ";
        itr--;
    }
    os << "}";
    return os;
}
// set
template <typename T>
ostream &operator<<(ostream &os, const set<T> &set_var) {
    os << "{";
    repi(itr, set_var) {
        os << *itr;
        itr++;
        if (itr != set_var.end()) os << ", ";
        itr--;
    }
    os << "}";
    return os;
}
// multiset
template <typename T>
ostream &operator<<(ostream &os, const multiset<T> &set_var) {
    os << "{";
    repi(itr, set_var) {
        os << *itr;
        itr++;
        if (itr != set_var.end()) os << ", ";
        itr--;
    }
    os << "}";
    return os;
}

#define DUMPOUT cerr

void dump_func() {
    DUMPOUT << endl;
}
template <class Head, class... Tail>
void dump_func(Head &&head, Tail &&... tail) {
    DUMPOUT << head;
    if (sizeof...(Tail) > 0) {
        DUMPOUT << ", ";
    }
    dump_func(std::move(tail)...);
}
#ifdef DEBUG_
#define DEB
#define dump(...)                                                              \
    DUMPOUT << "  " << string(#__VA_ARGS__) << ": "                            \
            << "[" << to_string(__LINE__) << ":" << __FUNCTION__ << "]"        \
            << endl                                                            \
            << "    ",                                                         \
        dump_func(__VA_ARGS__)
#else
#define DEB if (false)
#define dump(...)
#endif

// ^^^^^^^^^^^^^^^^^^^^^^^ debug output ^^^^^^^^^^^^^^^^^^^^^^^

string YN(bool y, int id = 0) {
    if (id) cout << id;
    return (y ? "YES" : "NO");
}
string yn(bool y, int id = 0) {
    if (id) cout << id;
    return (y ? "Yes" : "No");
}
string ON(bool y, int id = 0) {
    if (id) cout << id;
    return (y ? "OK" : "NG");
}

int dir4[4][2] = {{0, -1}, {-1, 0}, {1, 0}, {0, 1}};
int dir8[8][2] = {{-1, -1}, {0, -1}, {1, -1}, {-1, 0},
                  {1, 0},   {-1, 1}, {0, 1},  {1, 1}};
char dirchar[4] = {'<', '^', '>', 'v'};

// [a,b)
int irand(int a, int b) {
    static mt19937 Rand(static_cast<unsigned int>(time(nullptr)));
    uniform_int_distribution<int> dist(a, b - 1);
    return dist(Rand);
}

// [a,b)
double drand(int a, int b) {
    static mt19937 Rand(static_cast<unsigned int>(time(nullptr)));
    uniform_real_distribution<double> dist(a, b);
    return dist(Rand);
}

// https://qiita.com/IgnorantCoder/items/3101d6276e9bdddf872c
template <typename A, typename F>
inline auto transform(const A &v, F &&f) {
    using result_type =
        decltype(std::declval<F>()(std::declval<typename A::value_type>()));
    vector<result_type> y(v.size());
    std::transform(std::cbegin(v), std::cend(v), std::begin(y), f);
    return y;
}

// 多次元vector生成
template <class T>
vector<T> make_v(size_t size, const T &init) {
    return vector<T>(size, init);
}
template <class... Ts>
auto make_v(size_t size, Ts... rest) {
    return vector<decltype(make_v(rest...))>(size, make_v(rest...));
}

template <typename T>
T Max(vector<T> a) {
    return *max_element(all(a));
}
template <typename T>
T Min(vector<T> a) {
    return *min_element(all(a));
}
template <typename T>
T Sum(vector<T> a) {
    return accumulate(all(a), (T)0);
}

// mapでカウントとかする
template <typename T>
void Add(map<T, int> &m, T item) {
    if (m.find(item) == m.end()) {
        m[item] = 1;
    } else {
        m[item]++;
    }
}

// mapでカウントとかする
template <typename T>
void Erase(map<T, int> &m, T item) {
    if (m.find(item) == m.end()) {
    } else {
        if (m[item] == 1) {
            m.erase(item);
        } else {
            m[item]--;
        }
    }
}

// デフォルト値つきのmapのget
template <typename T, typename U>
U Get(map<T, U> m, T key, U def) {
    if (m.find(key) == m.end()) {
        return def;
    } else {
        return m[key];
    }
}

template <typename T>
inline bool Contains(const set<T> &t, const T &key) {
    return t.find(key) != t.end();
}

template <typename T, typename U>
inline bool Contains(const map<T, U> &t, const T &key) {
    return t.find(key) != t.end();
}

template <class T>
struct Edge {
    int from, to;
    T cost;
    bool operator<(Edge e) {
        return cost < e.cost;
    }
};

template <class T>
ostream &operator<<(ostream &os, Edge<T> &edge) {
    os << "(" << edge.from << "->" << edge.to << ":" << edge.cost << ")";
    return os;
}
template <class T = int>
class Graph {
    int n;
    bool directed;
    vector<vector<Edge<T>>> edges;

  public:
    Graph(int n, bool directed)
        : n(n), directed(directed), edges(vector<vector<Edge<T>>>(n)) {}

    void add_edge(int s, int t, T cost) {
        edges[s].push_back(Edge<T>{s, t, cost});
        if (!directed) {
            edges[t].push_back(Edge<T>{t, s, cost});
        }
    }
    Graph() {}

    vector<Edge<T>> operator[](size_t i) const {
        return edges[i];
    }

    int size() const {
        return n;
    }
};

//======================================================

class BitVector {
    vector<int> cumulative_sum;
    int pos = 1;
    vector<int> pos_select;

  public:
    vector<vector<int>> select; // select[b][n] = n 番目（0-indexed）の b の位置

    BitVector(int n)
        : cumulative_sum(n + 1), pos_select(2, 0), select(2, vector<int>(n)) {
        cumulative_sum[0] = 0;
    }

    // 数 b を追加。n 回呼ばれる
    void push(int b) {
        cumulative_sum[pos] = cumulative_sum[pos - 1] + b;
        select[b][pos_select[b]] = pos - 1;
        pos++;
        pos_select[b]++;
    }

    int operator[](int idx) const {
        return cumulative_sum[idx + 1] - cumulative_sum[idx];
    }

    // [0, n) に x がいくつあるか
    int rank(int n, int b) const {
        if (n < 0) return 0;
        n = min(n, size());
        if (b == 1) {
            return cumulative_sum[n];
        } else {
            return n - cumulative_sum[n];
        }
    }

    int size() const {
        return cumulative_sum.size() - 1;
    }
};
ostream &operator<<(ostream &os, const BitVector &bv) {
    os << "{";
    for (int i = 0; i < bv.size(); i++) {
        os << bv[i] << (i + 1 == bv.size() ? "" : ", ");
    }
    os << "}";
    return os;
}

class WaveletMatrix {
    int len;       // 数列の長さ
    int digit = 0; // 値の桁数
    vector<BitVector> bit_vectors;
    vector<int> num_zero;
    map<int, int> start; // sorted における各値の開始位置
    vector<int> sorted;
    vector<vector<long long>> cumulative_sum; // 各段での累積和

  public:
    WaveletMatrix(vector<int> v) {
        len = v.size();
        int max_value = *max_element(all(v));
        while (max_value > 0) {
            max_value >>= 1;
            digit++;
        }
        if (digit == 0) digit++;

        bit_vectors.reserve(digit);
        cumulative_sum.reserve(digit);
        num_zero = vector<int>(digit, 0);
        rep(d, digit) {
            bit_vectors.emplace_back(len);
            cumulative_sum.emplace_back(len + 1, 0);
            // 累積和を記録
            rep(i, len) {
                cumulative_sum[d][i + 1] = cumulative_sum[d][i] + v[i];
            }
            // ビットベクトルの構築
            rep(i, len) {
                bit_vectors[d].push(v[i] >> (digit - 1 - d) & 1);
            }
            // 安定ソート
            vector<int> temp;
            temp.reserve(len);
            rep(i, len) {
                if ((v[i] >> (digit - 1 - d) & 1) == 0) {
                    temp.push_back(v[i]);
                    num_zero[d]++;
                }
            }
            rep(i, len) {
                if ((v[i] >> (digit - 1 - d) & 1) == 1) {
                    temp.push_back(v[i]);
                }
            }
            v = temp;
        }
        reprev(i, len) {
            start[v[i]] = i;
        }
        sorted = v;
    }

    // n 番目（0-indexed）の値を得る
    int operator[](int n) const {
        int ans = 0;
        rep(d, digit) {
            ans <<= 1;
            ans += bit_vectors[d][n];
            if (bit_vectors[d][n] == 1) {
                n = num_zero[d] + bit_vectors[d].rank(n, 1);
            } else {
                n = bit_vectors[d].rank(n, 0);
            }
        }
        return ans;
    }

    // [0, n) に x がいくつ含まれるか
    int rank(int n, int x) {
        if (start.find(x) == start.end()) return 0;
        rep(d, digit) {
            if ((x >> (digit - 1 - d) & 1) == 1) {
                n = num_zero[d] + bit_vectors[d].rank(n, 1);
            } else {
                n = bit_vectors[d].rank(n, 0);
            }
        }
        return n - start[x];
    }

    // n 番目（0-indexed）の x の位置（無ければ-1）
    int select(int n, int x) {
        if (start.find(x) == start.end() || sorted[start[x] + n] != x)
            return -1;
        n = start[x] + n;
        reprev(d, digit) {
            if ((x >> (digit - 1 - d) & 1) == 1) {
                n = bit_vectors[d].select[1][n - num_zero[d]];
            } else {
                n = bit_vectors[d].select[0][n];
            }
        }
        return n;
    }

    // [l, r) で n 番目（0-indexed）に小さい値
    int quantile(int l, int r, int n) {
        rep(d, digit) {
            if (bit_vectors[d].rank(r, 0) - bit_vectors[d].rank(l, 0) > n) {
                // 0
                l = bit_vectors[d].rank(l, 0);
                r = bit_vectors[d].rank(r, 0);
            } else {
                // 1
                n -= bit_vectors[d].rank(r, 0) - bit_vectors[d].rank(l, 0);
                l = num_zero[d] + bit_vectors[d].rank(l, 1);
                r = num_zero[d] + bit_vectors[d].rank(r, 1);
            }
        }
        return sorted[l];
    }

    // [l, r) で出現回数が多い順に k 個の値と頻度を返す
    vector<pair<int, int>> topk(int l, int r, int k) {
        vector<pair<int, int>> res;
        res.reserve(k);
        priority_queue<tuple<int, int, int, int>> pq;
        pq.push(mt(r - l, 0, l, r));
        while (!pq.empty() && res.size() < k) {
            int dummy, d;
            tie(dummy, d, l, r) = pq.top();
            pq.pop();
            if (d == digit) {
                // 終わり
                res.push_back(mp(sorted[l], r - l));
                continue;
            }
            int l0 = bit_vectors[d].rank(l, 0);
            int r0 = bit_vectors[d].rank(r, 0);
            int l1 = num_zero[d] + bit_vectors[d].rank(l, 1);
            int r1 = num_zero[d] + bit_vectors[d].rank(r, 1);
            pq.push(mt(r0 - l0, d + 1, l0, r0));
            pq.push(mt(r1 - l1, d + 1, l1, r1));
        }
        return res;
    }

    // [l, r) に含まれる x 未満の値の個数
    int less_freq(int l, int r, int x) {
        if (x <= 0) return 0;
        if (l >= r) return 0;
        int ans = 0;
        rep(d, digit) {
            if ((x >> (digit - 1 - d) & 1) == 1) {
                ans += bit_vectors[d].rank(r, 0) - bit_vectors[d].rank(l, 0);
                l = num_zero[d] + bit_vectors[d].rank(l, 1);
                r = num_zero[d] + bit_vectors[d].rank(r, 1);
            } else {
                l = bit_vectors[d].rank(l, 0);
                r = bit_vectors[d].rank(r, 0);
            }
        }
        return ans;
    }

    // [l, r) に含まれる [a, b) の範囲の値の和
    long long range_sum(int l, int r, int a, int b) {
        if (l >= r) return 0;
        long long ans = 0;
        stack<tuple<int, int, int, int>> st;
        st.push(mt(0, 0, l, r)); // d, 範囲内理論最小値, l, r
        while (!st.empty()) {
            int d, min_value;
            tie(d, min_value, l, r) = st.top();
            st.pop();
            if (d == digit) {
                // 終わり
                if (a <= sorted[l] && sorted[l] < b) {
                    ans += (long long)sorted[l] * (r - l);
                }
                continue;
            }
            // この範囲内に [a, b) の値が無いなら飛ばす
            if (min_value >= b || min_value + (1 << (digit - d)) <= a) continue;
            // この範囲内が全て [a, b) に入るなら累積和
            if (a <= min_value && min_value + (1 << (digit - d)) <= b) {
                ans += cumulative_sum[d][r] - cumulative_sum[d][l];
                continue;
            }

            int l0 = bit_vectors[d].rank(l, 0);
            int r0 = bit_vectors[d].rank(r, 0);
            int l1 = num_zero[d] + bit_vectors[d].rank(l, 1);
            int r1 = num_zero[d] + bit_vectors[d].rank(r, 1);
            if (l0 < r0) st.push(mt(d + 1, min_value, l0, r0));
            if (l1 < r1)
                st.push(mt(d + 1, min_value + (1 << (digit - d - 1)), l1, r1));
        }
        return ans;
    }
};

int main() {
    int N, K;
    cin >> N >> K;
    vi A(N);
    cin >> A;

    WaveletMatrix wm(A);

    ll ans = INFL;
    rep(i, N - K + 1) {
        // [i, i+K) で揃えるコスト
        int median = wm.quantile(i, i + K, K / 2);
        int less_num = wm.less_freq(i, i + K, median);
        int more_num = K - less_num;
        ll less_sum = wm.range_sum(i, i + K, 0, median);
        ll more_sum = wm.range_sum(i, i + K, median, INF);
        ll cost = 0;
        cost += (ll)median * less_num - less_sum;
        cost += more_sum - (ll)median * more_num;
        chmin(ans, cost);
    }
    cout << ans << endl;

    return 0;
}