//#pragma GCC optimize("Ofast")
//#pragma GCC optimize("unroll-loops")
//#pragma GCC target("sse,sse2,sse3,ssse3,sse4,popcnt,abm,mmx,avx,tune=native")
#include <bits/stdc++.h>
using namespace std;
using ll = long long;
using ull = unsigned long long;
using db = double;
using ld = long double;
template<typename T> using V = vector<T>;
template<typename T> using VV = vector<vector<T>>;
#define fs first
#define sc second
#define pb push_back
#define mp make_pair
#define mt make_tuple
#define eb emplace_back
#define lb lower_bound
#define ub upper_bound
#define all(v) (v).begin(),(v).end()
#define siz(v) (ll)(v).size()
#define rep(i,a,n) for(ll i=a;i<(ll)(n);++i)
#define repr(i,a,n) for(ll i=n-1;(ll)a<=i;--i)
#define ENDL '\n'
typedef pair<int,int> Pi;
typedef pair<ll,ll> PL;
constexpr ll mod = 998244353;
constexpr ll INF = 1000000099;
constexpr ll LINF = (ll)(1e18 +99);
const ld PI = acos((ld)-1);
const vector<ll> dx={-1,0,1,0},dy={0,1,0,-1};
template<typename T,typename U> inline bool chmin(T& t, const U& u){if(t>u){t=u;return 1;}return 0;}
template<typename T,typename U> inline bool chmax(T& t, const U& u){if(t<u){t=u;return 1;}return 0;}
template<typename T> inline T gcd(T a,T b){return b?gcd(b,a%b):a;}
inline void yes() { cout << "Yes" << ENDL; }
inline void no() { cout << "No" << ENDL; }

template<typename T,typename Y> inline T mpow(T a, Y n) {
  T res = 1;
  for(;n;n>>=1) {
    if (n & 1) res = res * a;
    a = a * a;
  }
  return res;
}

template <typename T> V<T> prefix_sum(const V<T>& v) {
  int n = v.size();
  V<T> ret(n + 1);
  rep(i, 0, n) ret[i + 1] = ret[i] + v[i];
  return ret;
}

template<typename T>
istream& operator >> (istream& is, vector<T>& vec){
  for(auto&& x: vec) is >> x;
  return is;
}

template<typename T,typename Y>
ostream& operator<<(ostream& os,const pair<T,Y>& p){
  return os<<"{"<<p.fs<<","<<p.sc<<"}";
}

template<typename T> ostream& operator<<(ostream& os,const V<T>& v){
  os<<"{";
  for(auto e:v)os<<e<<",";
  return os<<"}";
}

template<typename ...Args>
void debug(Args&... args){
  for(auto const& x:{args...}){
    cerr<<x<<' ';
  }
  cerr<<ENDL;
}
template <long long MOD = mod>  // MODを適宜設定すること
struct Mint {
  using M = Mint;
  ll a;
  Mint(ll x = 0) {
    a = x % MOD;
    if(a < 0) a += MOD;
  }  // Mintで

  M pow(ll n) {
    M b = *this, r = 1 % MOD;
    while(n) {
      if(n & 1) r *= b;
      b *= b;
      n >>= 1;
    }
    return r;
  }

  M inv() { return pow(MOD - 2); }

  M& operator+=(M r) {
    a += r.a;
    if(a >= MOD) a -= MOD;
    return *this;
  }
  M& operator-=(M r) {
    a += MOD - r.a;
    if(a >= MOD) a -= MOD;
    return *this;
  }
  M& operator*=(M r) {
    a = 1LL * a * r.a % MOD;
    return *this;
  }
  M& operator/=(M r) { return (*this) *= r.inv(); }
  M operator+(M r) const { return M(a) += r; };
  M operator-(M r) const { return M(a) -= r; };
  M operator*(M r) const { return M(a) *= r; };
  M operator/(M r) const { return M(a) /= r; };
  M operator-() const { return M() - *this; }

  friend ostream& operator<<(ostream& os, const M& r) { return os << r.a; }
};
using M = Mint<mod>;

/*
using M = Mint<mod>;
M xとint iの演算はx.a+=i またはx+=M{i}
M(x).pow(exp)
*/

template <typename T> struct BIT {
  int n;
  vector<T> bit;  // 1-indexed

  BIT() : n(-1) {}

  BIT(int n_, T d) : n(n_), bit(n_ + 1, d) {}
  // initialization2 n_要素数 d初期値

  T sum(int i) {
    T s = bit[0];
    for(int x = i + 1; x > 0; x -= (x & -x)) s += bit[x];
    return s;
  }
  // 1からiまでの和(1-indexed)

  void add(int i, T a) {
    for(int x = i + 1; x <= n; x += (x & -x)) bit[x] += a;
  }
  // iにa加える

  T lower_bound(T w) {
    if(w <= 0) return 0;
    T x = 0, r = 1;  // xは横の位置を管理するイメージ
    while(r < n) r <<= 1;
    for(T k = r; k > 0; k >>= 1) {  //上の層から見る
      if(x + k <= n && bit[x + k] < w) {
        w -= bit[x + k];
        x += k;  //右の要素に移る
      }
    }
    return x + 1;
  }
  // indまでの区間和がw以上になるような最小のindを求める
  T query(int l, int r) { return sum(r - 1) - sum(l - 1); }  // [l,r)の和(0-indexed)
};

template<typename T> 
struct compression{
  V<T> tmp;
  map<T,T> ma;

  void push(T x){
    tmp.emplace_back(x);
  }

  void build(){
    sort(all(tmp));
    tmp.erase(unique(all(tmp)),tmp.end());
    rep(i,0,siz(tmp))ma[tmp[i]]=i;//
  }

  void change(T& x){
    assert(ma.count(x));
    x=ma[x];
  }

  void change(V<T>& v){
    rep(i,0,siz(v)){
      assert(ma.count(v[i]));
      v[i]=ma[v[i]];
    }
  }
};

signed main(){
  cin.tie(0);cerr.tie(0);ios::sync_with_stdio(false);
  cout<<fixed<<setprecision(20);
  
  ll n;cin>>n;
  V<ll> v(n);cin>>v;
  V<ll> id=v;
  compression<ll> com;
  rep(i,0,n)com.push(v[i]);
  com.build();
  com.change(id);


  BIT<ll> bl(n+5,0),br(n+5,0),numl(n+5,0),numr(n+5,0);
  rep(i,0,n){
    br.add(id[i],v[i]);
    numr.add(id[i],1);
  }


  M ans=0;
  rep(j,0,n){
    br.add(id[j],-v[j]);
    numr.add(id[j],-1);

    M l=numl.query(id[j]+1,n),r=numr.query(0,id[j]);
    ans+=r*bl.query(id[j]+1,n)+l*br.query(0,id[j])+r*l*v[j];

    bl.add(id[j],v[j]);
    numl.add(id[j],1);
  }
  cout<<ans<<ENDL;
}
//! ( . _ . ) ! 
//CHECK overflow,vector_size,what to output?
//any other simpler approach?
//list all conditions, try mathematical and graphic observation