#include <bits/stdc++.h>
// #pragma GCC optimize("Ofast,unroll-loops")
#define F first
#define S second
#define SZ(x) ((int)(x).size())
#define ALL(x) x.begin(), x.end()
#define MP make_pair
using namespace std;
using ll = long long;
using vi = vector<int>;
using vl = vector<ll>;
using vvi = vector<vi>;
using vvl = vector<vl>;
using pii = pair<int, int>;
using pll = pair<ll, ll>;
using vii = vector<pii>;
using vll = vector<pll>;
const int maxn = 20000 + 5;
const ll INF = 0x3f3f3f3f3f3f3f3f;
ll dp[2][2 * maxn][2];
ll dp_pre[2 * maxn][2];
ll dp_suf[2 * maxn][2];
void sol() {
int n;
cin >> n;
vi a(n + 1);
vi pos;
for (int i = 1; i <= n; i++) {
cin >> a[i];
pos.push_back(a[i] - 1);
pos.push_back(a[i]);
}
sort(ALL(pos));
pos.erase(unique(ALL(pos)), pos.end());
int m = SZ(pos);
for (int i = 0; i < m; i++) {
dp[0][i][0] = dp[0][i][1] = 0;
}
for (int i = 1; i <= n; i++) {
int now = i & 1;
int pre = now ^ 1;
for (int j = 0; j < m; j++) dp[now][j][0] = dp[now][j][1] = INF;
for (int j = 0; j < m; j++) {
dp[now][j][0] = min(dp[now][j][0], dp[pre][j][0] + abs(pos[j] - a[i]));
if (j < m - 1) dp[now][j][0] = min(dp[now][j][0], dp_suf[j + 1][1] + abs(pos[j] - a[i]));
dp[now][j][1] = min(dp[now][j][1], dp[pre][j][1] + abs(pos[j] - a[i]));
if (j > 0) dp[now][j][1] = min(dp[now][j][1], dp_pre[j - 1][0] + abs(pos[j] - a[i]));
}
for (int c = 0; c < 2; c++) {
dp_pre[0][c] = dp[now][0][c];
for (int j = 1; j < m; j++) dp_pre[j][c] = min(dp_pre[j - 1][c], dp[now][j][c]);
dp_suf[m - 1][c] = dp[now][m - 1][c];
for (int j = m - 2; j >= 0; j--) dp_suf[j][c] = min(dp_suf[j + 1][c], dp[now][j][c]);
}
}
ll ans = INF;
for (int i = 0; i < m; i++) {
for (int c = 0;c < 2; c++) {
ans = min(ans, dp[n&1][i][c]);
}
}
/*
for (int i = 0; i < m; i++) {
for (int c = 0;c < 2; c++) {
if (ans == dp[n&1][i][c]) cerr << pos[i] << '\n';
}
}
*/
cout << ans << '\n';
}
signed main() {
cin.tie(0)->sync_with_stdio(0);
sol();
}
