@lakutata/math 0.0.0-stage → 3.1.3

package/vendor/symengine/symengine/functions.cpp

@@ -0,0 +1,3913 @@
+#include <symengine/visitor.h>
+#include <symengine/symengine_exception.h>
+#include <symengine/eval_double.h>
+#include <symengine/limits.h>
+#include <symengine/real_double.h>
+#include <symengine/complex_double.h>
+
+namespace SymEngine
+{
+
+extern RCP<const Basic> &i2;
+extern RCP<const Basic> &i3;
+extern RCP<const Basic> &i5;
+extern RCP<const Basic> &im2;
+extern RCP<const Basic> &im3;
+extern RCP<const Basic> &im5;
+
+RCP<const Basic> sqrt(RCP<const Basic> &arg)
+{
+    return pow(arg, div(one, i2));
+}
+RCP<const Basic> cbrt(RCP<const Basic> &arg)
+{
+    return pow(arg, div(one, i3));
+}
+
+extern RCP<const Basic> &sq3;
+extern RCP<const Basic> &sq2;
+extern RCP<const Basic> &sq5;
+
+extern RCP<const Basic> &C0;
+extern RCP<const Basic> &C1;
+extern RCP<const Basic> &C2;
+extern RCP<const Basic> &C3;
+extern RCP<const Basic> &C4;
+extern RCP<const Basic> &C5;
+extern RCP<const Basic> &C6;
+
+extern RCP<const Basic> &mC0;
+extern RCP<const Basic> &mC1;
+extern RCP<const Basic> &mC2;
+extern RCP<const Basic> &mC3;
+extern RCP<const Basic> &mC4;
+extern RCP<const Basic> &mC5;
+extern RCP<const Basic> &mC6;
+
+// sin_table()[n] represents the value of sin(pi*n/12) for n = 0..23
+static const RCP<const Basic> *sin_table()
+{
+    // lakutata: the tables are never destroyed, the threads of the addon may use them when the process exits
+    static const RCP<const Basic> *table = new RCP<const Basic>[24]{
+        zero, C0,  C1,  C2,  C3,  C4,  one,       C4,  C3,  C2,  C1,  C0,
+        zero, mC0, mC1, mC2, mC3, mC4, minus_one, mC4, mC3, mC2, mC1, mC0};
+    return table;
+}
+
+static const umap_basic_basic &inverse_cst()
+{
+    static const umap_basic_basic &inverse_cst_ = *new umap_basic_basic{
+        {C3, i3},
+        {mC3, im3},
+        {C2, mul(i2, i2)},
+        {mC2, mul(im2, i2)},
+        {C4, integer(12)},
+        {mC4, integer(-12)},
+        {C5, i5},
+        {mC5, im5},
+        {C6, integer(10)},
+        {mC6, integer(-10)},
+        {div(one, i2), integer(6)},
+        {div(minus_one, i2), integer(-6)},
+    };
+    return inverse_cst_;
+}
+
+static const umap_basic_basic &inverse_tct()
+{
+    static const umap_basic_basic &inverse_tct_ = *new umap_basic_basic{
+        {div(one, sq3), mul(i2, i3)},
+        {div(minus_one, sq3), mul(im2, i3)},
+        {sq3, i3},
+        {mul(minus_one, sq3), im3},
+        {add(one, sq2), div(pow(i2, i3), i3)},
+        {mul(minus_one, add(one, sq2)), div(pow(i2, i3), im3)},
+        {sub(sq2, one), pow(i2, i3)},
+        {sub(one, sq2), pow(im2, i3)},
+        {sub(i2, sq3), mul(mul(i2, i2), i3)},
+        {sub(sq3, i2), mul(mul(im2, i2), i3)},
+        {sqrt(add(i5, mul(i2, sqrt(i5)))), div(i5, i2)},
+        {mul(minus_one, sqrt(add(i5, mul(i2, sqrt(i5))))), div(im5, i2)},
+        {one, pow(i2, i2)},
+        {minus_one, mul(minus_one, pow(i2, i2))},
+    };
+    return inverse_tct_;
+}
+
+Conjugate::Conjugate(const RCP<const Basic> &arg) : OneArgFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Conjugate::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a_Number(*arg)) {
+        if (eq(*arg, *ComplexInf)) {
+            return true;
+        }
+        return false;
+    }
+    if (is_a<Constant>(*arg)) {
+        return false;
+    }
+    if (is_a<Mul>(*arg)) {
+        return false;
+    }
+    if (is_a<Pow>(*arg)) {
+        if (is_a<Integer>(*down_cast<const Pow &>(*arg).get_exp())) {
+            return false;
+        }
+    }
+    // OneArgFunction classes
+    if (is_a<Sign>(*arg) or is_a<Conjugate>(*arg) or is_a<Erf>(*arg)
+        or is_a<Erfc>(*arg) or is_a<Gamma>(*arg) or is_a<LogGamma>(*arg)
+        or is_a<Abs>(*arg)) {
+        return false;
+    }
+    if (is_a<Sin>(*arg) or is_a<Cos>(*arg) or is_a<Tan>(*arg) or is_a<Cot>(*arg)
+        or is_a<Sec>(*arg) or is_a<Csc>(*arg)) {
+        return false;
+    }
+    if (is_a<Sinh>(*arg) or is_a<Cosh>(*arg) or is_a<Tanh>(*arg)
+        or is_a<Coth>(*arg) or is_a<Sech>(*arg) or is_a<Csch>(*arg)) {
+        return false;
+    }
+    // TwoArgFunction classes
+    if (is_a<KroneckerDelta>(*arg) or is_a<ATan2>(*arg)
+        or is_a<LowerGamma>(*arg) or is_a<UpperGamma>(*arg)
+        or is_a<Beta>(*arg)) {
+        return false;
+    }
+    // MultiArgFunction class
+    if (is_a<LeviCivita>(*arg)) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> Conjugate::create(const RCP<const Basic> &arg) const
+{
+    return conjugate(arg);
+}
+
+RCP<const Basic> conjugate(const RCP<const Basic> &arg)
+{
+    if (is_a_Number(*arg)) {
+        return down_cast<const Number &>(*arg).conjugate();
+    }
+    if (is_a<Constant>(*arg) or is_a<Abs>(*arg) or is_a<KroneckerDelta>(*arg)
+        or is_a<LeviCivita>(*arg)) {
+        return arg;
+    }
+    if (is_a<Mul>(*arg)) {
+        const map_basic_basic &dict = down_cast<const Mul &>(*arg).get_dict();
+        map_basic_basic new_dict;
+        RCP<const Number> coef = rcp_static_cast<const Number>(
+            conjugate(down_cast<const Mul &>(*arg).get_coef()));
+        for (const auto &p : dict) {
+            if (is_a<Integer>(*p.second)) {
+                Mul::dict_add_term_new(outArg(coef), new_dict, p.second,
+                                       conjugate(p.first));
+            } else {
+                Mul::dict_add_term_new(
+                    outArg(coef), new_dict, one,
+                    conjugate(Mul::from_dict(one, {{p.first, p.second}})));
+            }
+        }
+        return Mul::from_dict(coef, std::move(new_dict));
+    }
+    if (is_a<Pow>(*arg)) {
+        RCP<const Basic> base = down_cast<const Pow &>(*arg).get_base();
+        RCP<const Basic> exp = down_cast<const Pow &>(*arg).get_exp();
+        if (is_a<Integer>(*exp)) {
+            return pow(conjugate(base), exp);
+        }
+    }
+    if (is_a<Conjugate>(*arg)) {
+        return down_cast<const Conjugate &>(*arg).get_arg();
+    }
+    if (is_a<Sign>(*arg) or is_a<Erf>(*arg) or is_a<Erfc>(*arg)
+        or is_a<Gamma>(*arg) or is_a<LogGamma>(*arg) or is_a<Sin>(*arg)
+        or is_a<Cos>(*arg) or is_a<Tan>(*arg) or is_a<Cot>(*arg)
+        or is_a<Sec>(*arg) or is_a<Csc>(*arg) or is_a<Sinh>(*arg)
+        or is_a<Cosh>(*arg) or is_a<Tanh>(*arg) or is_a<Coth>(*arg)
+        or is_a<Sech>(*arg) or is_a<Csch>(*arg)) {
+        const OneArgFunction &func = down_cast<const OneArgFunction &>(*arg);
+        return func.create(conjugate(func.get_arg()));
+    }
+    if (is_a<ATan2>(*arg) or is_a<LowerGamma>(*arg) or is_a<UpperGamma>(*arg)
+        or is_a<Beta>(*arg)) {
+        const TwoArgFunction &func = down_cast<const TwoArgFunction &>(*arg);
+        return func.create(conjugate(func.get_arg1()),
+                           conjugate(func.get_arg2()));
+    }
+    return make_rcp<const Conjugate>(arg);
+}
+
+bool get_pi_shift(const RCP<const Basic> &arg, const Ptr<RCP<const Number>> &n,
+                  const Ptr<RCP<const Basic>> &x)
+{
+    if (is_a<Add>(*arg)) {
+        const Add &s = down_cast<const Add &>(*arg);
+        RCP<const Basic> coef = s.get_coef();
+        auto size = s.get_dict().size();
+        if (size > 1) {
+            // arg should be of form `x + n*pi`
+            // `n` is an integer
+            // `x` is an `Expression`
+            bool check_pi = false;
+            RCP<const Basic> temp;
+            *x = coef;
+            for (const auto &p : s.get_dict()) {
+                if (eq(*p.first, *pi)
+                    and (is_a<Integer>(*p.second)
+                         or is_a<Rational>(*p.second))) {
+                    check_pi = true;
+                    *n = p.second;
+                } else {
+                    *x = add(mul(p.first, p.second), *x);
+                }
+            }
+            if (check_pi)
+                return true;
+            else // No term with `pi` found
+                return false;
+        } else if (size == 1) {
+            // arg should be of form `a + n*pi`
+            // where `a` is a `Number`.
+            auto p = s.get_dict().begin();
+            if (eq(*p->first, *pi)
+                and (is_a<Integer>(*p->second) or is_a<Rational>(*p->second))) {
+                *n = p->second;
+                *x = coef;
+                return true;
+            } else {
+                return false;
+            }
+        } else { // Should never reach here though!
+            // Dict of size < 1
+            return false;
+        }
+    } else if (is_a<Mul>(*arg)) {
+        // `arg` is of the form `k*pi/12`
+        const Mul &s = down_cast<const Mul &>(*arg);
+        auto p = s.get_dict().begin();
+        // dict should contain symbol `pi` only
+        if (s.get_dict().size() == 1 and eq(*p->first, *pi)
+            and eq(*p->second, *one)
+            and (is_a<Integer>(*s.get_coef())
+                 or is_a<Rational>(*s.get_coef()))) {
+            *n = s.get_coef();
+            *x = zero;
+            return true;
+        } else {
+            return false;
+        }
+    } else if (eq(*arg, *pi)) {
+        *n = one;
+        *x = zero;
+        return true;
+    } else if (eq(*arg, *zero)) {
+        *n = zero;
+        *x = zero;
+        return true;
+    } else {
+        return false;
+    }
+}
+
+// Return true if arg is of form a+b*pi, with b integer or rational
+// with denominator 2. The a may be zero or any expression.
+bool trig_has_basic_shift(const RCP<const Basic> &arg)
+{
+    if (is_a<Add>(*arg)) {
+        const Add &s = down_cast<const Add &>(*arg);
+        for (const auto &p : s.get_dict()) {
+            const auto &temp = mul(p.second, integer(2));
+            if (eq(*p.first, *pi)) {
+                if (is_a<Integer>(*temp)) {
+                    return true;
+                }
+                if (is_a<Rational>(*temp)) {
+                    auto m = down_cast<const Rational &>(*temp)
+                                 .as_rational_class();
+                    return (m < 0) or (m > 1);
+                }
+                return false;
+            }
+        }
+        return false;
+    } else if (is_a<Mul>(*arg)) {
+        // is `arg` of the form `k*pi/2`?
+        // dict should contain symbol `pi` only
+        // and `k` should be a rational s.t. 0 < k < 1
+        const Mul &s = down_cast<const Mul &>(*arg);
+        RCP<const Basic> coef = mul(s.get_coef(), integer(2));
+        auto p = s.get_dict().begin();
+        if (s.get_dict().size() == 1 and eq(*p->first, *pi)
+            and eq(*p->second, *one)) {
+            if (is_a<Integer>(*coef)) {
+                return true;
+            }
+            if (is_a<Rational>(*coef)) {
+                auto m = down_cast<const Rational &>(*coef).as_rational_class();
+                return (m < 0) or (m > 1);
+            }
+            return false;
+        } else {
+            return false;
+        }
+    } else if (eq(*arg, *pi)) {
+        return true;
+    } else if (eq(*arg, *zero)) {
+        return true;
+    } else {
+        return false;
+    }
+}
+
+bool could_extract_minus(const Basic &arg)
+{
+    if (is_a_Number(arg)) {
+        if (down_cast<const Number &>(arg).is_negative()) {
+            return true;
+        } else if (is_a_Complex(arg)) {
+            const ComplexBase &c = down_cast<const ComplexBase &>(arg);
+            RCP<const Number> real_part = c.real_part();
+            return (real_part->is_negative())
+                   or (eq(*real_part, *zero)
+                       and c.imaginary_part()->is_negative());
+        } else {
+            return false;
+        }
+    } else if (is_a<Mul>(arg)) {
+        const Mul &s = down_cast<const Mul &>(arg);
+        return could_extract_minus(*s.get_coef());
+    } else if (is_a<Add>(arg)) {
+        const Add &s = down_cast<const Add &>(arg);
+        if (s.get_coef()->is_zero()) {
+            map_basic_num d(s.get_dict().begin(), s.get_dict().end());
+            return could_extract_minus(*d.begin()->second);
+        } else {
+            return could_extract_minus(*s.get_coef());
+        }
+    } else {
+        return false;
+    }
+}
+
+bool handle_minus(const RCP<const Basic> &arg,
+                  const Ptr<RCP<const Basic>> &rarg)
+{
+    if (is_a<Mul>(*arg)) {
+        const Mul &s = down_cast<const Mul &>(*arg);
+        // Check for -Add instances to transform -(-x + 2*y) to (x - 2*y)
+        if (s.get_coef()->is_minus_one() && s.get_dict().size() == 1
+            && eq(*s.get_dict().begin()->second, *one)) {
+            return not handle_minus(mul(minus_one, arg), rarg);
+        } else if (could_extract_minus(*s.get_coef())) {
+            *rarg = mul(minus_one, arg);
+            return true;
+        }
+    } else if (is_a<Add>(*arg)) {
+        if (could_extract_minus(*arg)) {
+            const Add &s = down_cast<const Add &>(*arg);
+            // lakutata: added term by term, so that a sum among the terms is
+            // flattened (its coefficient -1 would become 1 in the dictionary)
+            RCP<const Basic> negated = s.get_coef()->mul(*minus_one);
+            for (const auto &p : s.get_dict()) {
+                negated = add(negated, mul(p.second->mul(*minus_one), p.first));
+            }
+            *rarg = negated;
+            return true;
+        }
+    } else if (could_extract_minus(*arg)) {
+        *rarg = mul(minus_one, arg);
+        return true;
+    }
+    *rarg = arg;
+    return false;
+}
+
+// \return true if conjugate has to be returned finally else false
+bool trig_simplify(const RCP<const Basic> &arg, unsigned period, bool odd,
+                   bool conj_odd, // input
+                   const Ptr<RCP<const Basic>> &rarg, int &index,
+                   int &sign) // output
+{
+    bool check;
+    RCP<const Number> n;
+    RCP<const Basic> r;
+    RCP<const Basic> ret_arg;
+    check = get_pi_shift(arg, outArg(n), outArg(r));
+    if (check) {
+        RCP<const Number> t = mulnum(n, integer(12));
+        sign = 1;
+        if (is_a<Integer>(*t)) {
+            int m = numeric_cast<int>(
+                mod_f(down_cast<const Integer &>(*t), *integer(12 * period))
+                    ->as_int());
+            if (eq(*r, *zero)) {
+                index = m;
+                *rarg = zero;
+                return false;
+            } else if (m == 0) {
+                index = 0;
+                bool b = handle_minus(r, outArg(ret_arg));
+                *rarg = ret_arg;
+                if (odd and b)
+                    sign = -1;
+                return false;
+            }
+        }
+
+        rational_class m;
+        if (is_a<Integer>(*n)) {
+            // 2*pi periodic => f(r + pi * n) = f(r - pi * n)
+            m = mp_abs(down_cast<const Integer &>(*n).as_integer_class());
+            m /= period;
+        } else {
+            SYMENGINE_ASSERT(is_a<Rational>(*n));
+            m = down_cast<const Rational &>(*n).as_rational_class() / period;
+            integer_class t;
+#if SYMENGINE_INTEGER_CLASS != SYMENGINE_BOOSTMP
+            mp_fdiv_r(t, get_num(m), get_den(m));
+            get_num(m) = t;
+#else
+            integer_class quo;
+            mp_fdiv_qr(quo, t, get_num(m), get_den(m));
+            m -= rational_class(quo);
+#endif
+            // m = a / b => m = (a % b / b)
+        }
+        // Now, arg = r + 2 * pi * m  where 0 <= m < 1
+        m *= 2 * period;
+        // Now, arg = r + pi * m / 2  where 0 <= m < 4
+        if (m >= 2 and m < 3) {
+            sign = -1;
+            r = add(r, mul(pi, Rational::from_mpq((m - 2) / 2)));
+            bool b = handle_minus(r, outArg(ret_arg));
+            *rarg = ret_arg;
+            if (odd and b)
+                sign = -1 * sign;
+            return false;
+        } else if (m >= 1) {
+            if (m < 2) {
+                // 1 <= m < 2
+                sign = 1;
+                r = add(r, mul(pi, Rational::from_mpq((m - 1) / 2)));
+            } else {
+                // 3 <= m < 4
+                sign = -1;
+                r = add(r, mul(pi, Rational::from_mpq((m - 3) / 2)));
+            }
+            bool b = handle_minus(r, outArg(ret_arg));
+            *rarg = ret_arg;
+            if (not b and conj_odd)
+                sign = -sign;
+            return true;
+        } else {
+            *rarg = add(r, mul(pi, Rational::from_mpq(m / 2)));
+            index = -1;
+            return false;
+        }
+    } else {
+        bool b = handle_minus(arg, outArg(ret_arg));
+        *rarg = ret_arg;
+        index = -1;
+        if (odd and b)
+            sign = -1;
+        else
+            sign = 1;
+        return false;
+    }
+}
+
+bool inverse_lookup(const umap_basic_basic &d, const RCP<const Basic> &t,
+                    const Ptr<RCP<const Basic>> &index)
+{
+    auto it = d.find(t);
+    if (it == d.end()) {
+        // Not found in lookup
+        return false;
+    } else {
+        *index = (it->second);
+        return true;
+    }
+}
+
+Sign::Sign(const RCP<const Basic> &arg) : OneArgFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Sign::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a_Number(*arg)) {
+        if (eq(*arg, *ComplexInf)) {
+            return true;
+        }
+        return false;
+    }
+    if (is_a<Constant>(*arg)) {
+        return false;
+    }
+    if (is_a<Sign>(*arg)) {
+        return false;
+    }
+    if (is_a<Mul>(*arg)) {
+        if (neq(*down_cast<const Mul &>(*arg).get_coef(), *one)
+            and neq(*down_cast<const Mul &>(*arg).get_coef(), *minus_one)) {
+            return false;
+        }
+    }
+    return true;
+}
+
+RCP<const Basic> Sign::create(const RCP<const Basic> &arg) const
+{
+    return sign(arg);
+}
+
+RCP<const Basic> sign(const RCP<const Basic> &arg)
+{
+    if (is_a_Number(*arg)) {
+        if (is_a<NaN>(*arg)) {
+            return Nan;
+        }
+        if (down_cast<const Number &>(*arg).is_zero()) {
+            return zero;
+        }
+        if (down_cast<const Number &>(*arg).is_positive()) {
+            return one;
+        }
+        if (down_cast<const Number &>(*arg).is_negative()) {
+            return minus_one;
+        }
+        if (is_a_Complex(*arg)
+            and down_cast<const ComplexBase &>(*arg).is_re_zero()) {
+            RCP<const Number> r
+                = down_cast<const ComplexBase &>(*arg).imaginary_part();
+            if (down_cast<const Number &>(*r).is_positive()) {
+                return I;
+            }
+            if (down_cast<const Number &>(*r).is_negative()) {
+                return mul(minus_one, I);
+            }
+        }
+    }
+    if (is_a<Constant>(*arg)) {
+        if (eq(*arg, *pi) or eq(*arg, *E) or eq(*arg, *EulerGamma)
+            or eq(*arg, *Catalan) or eq(*arg, *GoldenRatio))
+            return one;
+    }
+    if (is_a<Sign>(*arg)) {
+        return arg;
+    }
+    if (is_a<Mul>(*arg)) {
+        // lakutata: the sign of each factor, where it is known (sign(2*sqrt(3))
+        // is 1, sign(-e) is -1); SymEngine kept the factors after the coefficient
+        RCP<const Basic> s = sign(down_cast<const Mul &>(*arg).get_coef());
+        map_basic_basic rest;
+        for (const auto &p : down_cast<const Mul &>(*arg).get_dict()) {
+            map_basic_basic single;
+            insert(single, p.first, p.second);
+            const RCP<const Basic> power = Mul::from_dict(one, std::move(single));
+            const RCP<const Basic> factor = sign(power);
+            // Unknown: sign(factor) itself; a factor that is a sign is its own
+            if (is_a<Sign>(*factor)
+                and eq(*down_cast<const Sign &>(*factor).get_arg(), *power)) {
+                insert(rest, p.first, p.second);
+            } else {
+                s = mul(s, factor);
+            }
+        }
+        if (rest.empty())
+            return s;
+        return mul(s,
+                   make_rcp<const Sign>(Mul::from_dict(one, std::move(rest))));
+    }
+    if (is_a<Pow>(*arg)) {
+        RCP<const Pow> pow_arg = rcp_static_cast<const Pow>(arg);
+        RCP<const Basic> s = sign(pow_arg->get_base());
+        // lakutata: a number only (the sign of a base -x is -sign(x), whose
+        // power came back here without end)
+        if (is_a_Number(*s) and not eq(*s, *pow_arg->get_base())) {
+            return sign(pow(s, pow_arg->get_exp()));
+        }
+    }
+    // lakutata: a number written as an expression (pi - 3), its sign from its
+    // value when it is clearly away from zero
+    if (free_symbols(*arg).empty()) {
+        try {
+            const std::complex<double> value = eval_complex_double(*arg);
+            if (value.imag() == 0 and std::abs(value.real()) > 1e-10)
+                return value.real() > 0 ? one : minus_one;
+        } catch (SymEngineException &) {
+        }
+    }
+    return make_rcp<const Sign>(arg);
+}
+
+Floor::Floor(const RCP<const Basic> &arg) : OneArgFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Floor::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a_Number(*arg)) {
+        return false;
+    }
+    if (is_a<Constant>(*arg)) {
+        return false;
+    }
+    if (is_a<Floor>(*arg)) {
+        return false;
+    }
+    if (is_a<Ceiling>(*arg)) {
+        return false;
+    }
+    if (is_a<Truncate>(*arg)) {
+        return false;
+    }
+    if (is_a<BooleanAtom>(*arg) or is_a_Relational(*arg)) {
+        return false;
+    }
+    if (is_a<Add>(*arg)) {
+        RCP<const Number> s = down_cast<const Add &>(*arg).get_coef();
+        if (neq(*zero, *s) and is_a<Integer>(*s)) {
+            return false;
+        }
+    }
+    return true;
+}
+
+RCP<const Basic> Floor::create(const RCP<const Basic> &arg) const
+{
+    return floor(arg);
+}
+
+// lakutata: the integer part of a real number written as an expression
+// (floor(sqrt(2)) is 1), from its value when it is clearly away from an
+// integer and small enough for a double to tell
+static bool integer_part_of(const RCP<const Basic> &arg, double (*part)(double),
+                            const Ptr<RCP<const Basic>> &value)
+{
+    if (is_a_Number(*arg) or not free_symbols(*arg).empty())
+        return false;
+    try {
+        const double x = eval_double(*arg);
+        if (not std::isfinite(x) or std::abs(x) > 1e15
+            or std::abs(x - std::round(x)) <= 1e-10 * std::max(1.0, std::abs(x)))
+            return false;
+        *value = integer(integer_class(static_cast<long long>(part(x))));
+        return true;
+    } catch (SymEngineException &) {
+        return false;
+    }
+}
+
+RCP<const Basic> floor(const RCP<const Basic> &arg)
+{
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_exact()) {
+            if (is_a<Rational>(*arg)) {
+                const Rational &s = down_cast<const Rational &>(*arg);
+                integer_class quotient;
+                mp_fdiv_q(quotient, SymEngine::get_num(s.as_rational_class()),
+                          SymEngine::get_den(s.as_rational_class()));
+                return integer(std::move(quotient));
+            }
+            return arg;
+        }
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        return _arg->get_eval().floor(*_arg);
+    }
+    if (is_a<Constant>(*arg)) {
+        if (eq(*arg, *pi)) {
+            return integer(3);
+        }
+        if (eq(*arg, *E)) {
+            return integer(2);
+        }
+        if (eq(*arg, *GoldenRatio)) {
+            return integer(1);
+        }
+        if (eq(*arg, *Catalan) or eq(*arg, *EulerGamma)) {
+            return integer(0);
+        }
+    }
+    if (is_a<Floor>(*arg)) {
+        return arg;
+    }
+    if (is_a<Ceiling>(*arg)) {
+        return arg;
+    }
+    if (is_a<Truncate>(*arg)) {
+        return arg;
+    }
+    if (is_a<BooleanAtom>(*arg) or is_a_Relational(*arg)) {
+        throw SymEngineException(
+            "Boolean objects not allowed in this context.");
+    }
+    if (is_a<Add>(*arg)) {
+        RCP<const Number> s = down_cast<const Add &>(*arg).get_coef();
+        umap_basic_num d = down_cast<const Add &>(*arg).get_dict();
+        if (is_a<Integer>(*s)
+            and not down_cast<const Integer &>(*s).is_zero()) {
+            return add(s, floor(Add::from_dict(zero, std::move(d))));
+        }
+    }
+    RCP<const Basic> value;
+    if (integer_part_of(arg, std::floor, outArg(value)))
+        return value;
+    return make_rcp<const Floor>(arg);
+}
+
+Ceiling::Ceiling(const RCP<const Basic> &arg) : OneArgFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Ceiling::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a_Number(*arg)) {
+        return false;
+    }
+    if (is_a<Constant>(*arg)) {
+        return false;
+    }
+    if (is_a<Floor>(*arg)) {
+        return false;
+    }
+    if (is_a<Ceiling>(*arg)) {
+        return false;
+    }
+    if (is_a<Truncate>(*arg)) {
+        return false;
+    }
+    if (is_a<BooleanAtom>(*arg) or is_a_Relational(*arg)) {
+        return false;
+    }
+    if (is_a<Add>(*arg)) {
+        RCP<const Number> s = down_cast<const Add &>(*arg).get_coef();
+        if (neq(*zero, *s) and is_a<Integer>(*s)) {
+            return false;
+        }
+    }
+    return true;
+}
+
+RCP<const Basic> Ceiling::create(const RCP<const Basic> &arg) const
+{
+    return ceiling(arg);
+}
+
+RCP<const Basic> ceiling(const RCP<const Basic> &arg)
+{
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_exact()) {
+            if (is_a<Rational>(*arg)) {
+                const Rational &s = down_cast<const Rational &>(*arg);
+                integer_class quotient;
+                mp_cdiv_q(quotient, SymEngine::get_num(s.as_rational_class()),
+                          SymEngine::get_den(s.as_rational_class()));
+                return integer(std::move(quotient));
+            }
+            return arg;
+        }
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        return _arg->get_eval().ceiling(*_arg);
+    }
+    if (is_a<Constant>(*arg)) {
+        if (eq(*arg, *pi)) {
+            return integer(4);
+        }
+        if (eq(*arg, *E)) {
+            return integer(3);
+        }
+        if (eq(*arg, *GoldenRatio)) {
+            return integer(2);
+        }
+        if (eq(*arg, *Catalan) or eq(*arg, *EulerGamma)) {
+            return integer(1);
+        }
+    }
+    if (is_a<Floor>(*arg)) {
+        return arg;
+    }
+    if (is_a<Ceiling>(*arg)) {
+        return arg;
+    }
+    if (is_a<Truncate>(*arg)) {
+        return arg;
+    }
+    if (is_a<BooleanAtom>(*arg) or is_a_Relational(*arg)) {
+        throw SymEngineException(
+            "Boolean objects not allowed in this context.");
+    }
+    if (is_a<Add>(*arg)) {
+        RCP<const Number> s = down_cast<const Add &>(*arg).get_coef();
+        umap_basic_num d = down_cast<const Add &>(*arg).get_dict();
+        // lakutata: the rest evaluated too (ceiling(5 + pi) is 5 + ceiling(pi), a number)
+        if (is_a<Integer>(*s) and not s->is_zero()) {
+            return add(s, ceiling(Add::from_dict(zero, std::move(d))));
+        }
+    }
+    RCP<const Basic> value;
+    if (integer_part_of(arg, std::ceil, outArg(value)))
+        return value;
+    return make_rcp<const Ceiling>(arg);
+}
+
+Truncate::Truncate(const RCP<const Basic> &arg) : OneArgFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Truncate::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a_Number(*arg)) {
+        return false;
+    }
+    if (is_a<Constant>(*arg)) {
+        return false;
+    }
+    if (is_a<Floor>(*arg)) {
+        return false;
+    }
+    if (is_a<Ceiling>(*arg)) {
+        return false;
+    }
+    if (is_a<Truncate>(*arg)) {
+        return false;
+    }
+    if (is_a<BooleanAtom>(*arg) or is_a_Relational(*arg)) {
+        return false;
+    }
+    if (is_a<Add>(*arg)) {
+        RCP<const Number> s = down_cast<const Add &>(*arg).get_coef();
+        if (neq(*zero, *s) and is_a<Integer>(*s)) {
+            return false;
+        }
+    }
+    return true;
+}
+
+RCP<const Basic> Truncate::create(const RCP<const Basic> &arg) const
+{
+    return truncate(arg);
+}
+
+RCP<const Basic> truncate(const RCP<const Basic> &arg)
+{
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_exact()) {
+            if (is_a<Rational>(*arg)) {
+                const Rational &s = down_cast<const Rational &>(*arg);
+                integer_class quotient;
+                mp_tdiv_q(quotient, SymEngine::get_num(s.as_rational_class()),
+                          SymEngine::get_den(s.as_rational_class()));
+                return integer(std::move(quotient));
+            }
+            return arg;
+        }
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        return _arg->get_eval().truncate(*_arg);
+    }
+    if (is_a<Constant>(*arg)) {
+        if (eq(*arg, *pi)) {
+            return integer(3);
+        }
+        if (eq(*arg, *E)) {
+            return integer(2);
+        }
+        if (eq(*arg, *GoldenRatio)) {
+            return integer(1);
+        }
+        if (eq(*arg, *Catalan) or eq(*arg, *EulerGamma)) {
+            return integer(0);
+        }
+    }
+    if (is_a<Floor>(*arg)) {
+        return arg;
+    }
+    if (is_a<Ceiling>(*arg)) {
+        return arg;
+    }
+    if (is_a<Truncate>(*arg)) {
+        return arg;
+    }
+    if (is_a<BooleanAtom>(*arg) or is_a_Relational(*arg)) {
+        throw SymEngineException(
+            "Boolean objects not allowed in this context.");
+    }
+    if (is_a<Add>(*arg)) {
+        RCP<const Number> s = down_cast<const Add &>(*arg).get_coef();
+        umap_basic_num d = down_cast<const Add &>(*arg).get_dict();
+        // lakutata: no integer taken out, trunc(5 - 1/2) is 4, not
+        // 5 + trunc(-1/2); SymEngine took it out
+        (void)s;
+        (void)d;
+    }
+    RCP<const Basic> value;
+    if (integer_part_of(arg, std::trunc, outArg(value)))
+        return value;
+    return make_rcp<const Truncate>(arg);
+}
+
+Sin::Sin(const RCP<const Basic> &arg) : TrigFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Sin::is_canonical(const RCP<const Basic> &arg) const
+{
+    // e.g. sin(0)
+    if (is_a<Integer>(*arg) and down_cast<const Integer &>(*arg).is_zero())
+        return false;
+    // e.g sin(7*pi/2+y)
+    if (trig_has_basic_shift(arg)) {
+        return false;
+    }
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> sin(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return zero;
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().sin(*arg);
+    }
+
+    if (is_a<ASin>(*arg)) {
+        return down_cast<const ASin &>(*arg).get_arg();
+    } else if (is_a<ACsc>(*arg)) {
+        return div(one, down_cast<const ACsc &>(*arg).get_arg());
+    }
+
+    RCP<const Basic> ret_arg;
+    int index, sign;
+    bool conjugate = trig_simplify(arg, 2, true, false,           // input
+                                   outArg(ret_arg), index, sign); // output
+
+    if (conjugate) {
+        // cos has to be returned
+        if (sign == 1) {
+            return cos(ret_arg);
+        } else {
+            return mul(minus_one, cos(ret_arg));
+        }
+    } else {
+        if (eq(*ret_arg, *zero)) {
+            return mul(integer(sign), sin_table()[index]);
+        } else {
+            // If ret_arg is the same as arg, a `Sin` instance is returned
+            // Or else `sin` is called again.
+            if (sign == 1) {
+                if (neq(*ret_arg, *arg)) {
+                    return sin(ret_arg);
+                } else {
+                    return make_rcp<const Sin>(arg);
+                }
+            } else {
+                return mul(minus_one, sin(ret_arg));
+            }
+        }
+    }
+}
+
+/* ---------------------------- */
+
+Cos::Cos(const RCP<const Basic> &arg) : TrigFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Cos::is_canonical(const RCP<const Basic> &arg) const
+{
+    // e.g. cos(0)
+    if (is_a<Integer>(*arg) and down_cast<const Integer &>(*arg).is_zero())
+        return false;
+    // e.g cos(k*pi/2)
+    if (trig_has_basic_shift(arg)) {
+        return false;
+    }
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> cos(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return one;
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().cos(*arg);
+    }
+
+    if (is_a<ACos>(*arg)) {
+        return down_cast<const ACos &>(*arg).get_arg();
+    } else if (is_a<ASec>(*arg)) {
+        return div(one, down_cast<const ASec &>(*arg).get_arg());
+    }
+
+    RCP<const Basic> ret_arg;
+    int index, sign;
+    bool conjugate = trig_simplify(arg, 2, false, true,           // input
+                                   outArg(ret_arg), index, sign); // output
+
+    if (conjugate) {
+        // sin has to be returned
+        if (sign == 1) {
+            return sin(ret_arg);
+        } else {
+            return mul(minus_one, sin(ret_arg));
+        }
+    } else {
+        if (eq(*ret_arg, *zero)) {
+            return mul(integer(sign), sin_table()[(index + 6) % 24]);
+        } else {
+            if (sign == 1) {
+                if (neq(*ret_arg, *arg)) {
+                    return cos(ret_arg);
+                } else {
+                    return make_rcp<const Cos>(ret_arg);
+                }
+            } else {
+                return mul(minus_one, cos(ret_arg));
+            }
+        }
+    }
+}
+
+/* ---------------------------- */
+
+Tan::Tan(const RCP<const Basic> &arg) : TrigFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Tan::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a<Integer>(*arg) and down_cast<const Integer &>(*arg).is_zero())
+        return false;
+    // e.g tan(k*pi/2)
+    if (trig_has_basic_shift(arg)) {
+        return false;
+    }
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> tan(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return zero;
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().tan(*arg);
+    }
+
+    if (is_a<ATan>(*arg)) {
+        return down_cast<const ATan &>(*arg).get_arg();
+    } else if (is_a<ACot>(*arg)) {
+        return div(one, down_cast<const ACot &>(*arg).get_arg());
+    }
+
+    RCP<const Basic> ret_arg;
+    int index, sign;
+    bool conjugate = trig_simplify(arg, 1, true, true,            // input
+                                   outArg(ret_arg), index, sign); // output
+
+    if (conjugate) {
+        // cot has to be returned
+        if (sign == 1) {
+            return cot(ret_arg);
+        } else {
+            return mul(minus_one, cot(ret_arg));
+        }
+    } else {
+        if (eq(*ret_arg, *zero)) {
+            return mul(integer(sign),
+                       div(sin_table()[index], sin_table()[(index + 6) % 24]));
+        } else {
+            if (sign == 1) {
+                if (neq(*ret_arg, *arg)) {
+                    return tan(ret_arg);
+                } else {
+                    return make_rcp<const Tan>(ret_arg);
+                }
+            } else {
+                return mul(minus_one, tan(ret_arg));
+            }
+        }
+    }
+}
+
+/* ---------------------------- */
+
+Cot::Cot(const RCP<const Basic> &arg) : TrigFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Cot::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a<Integer>(*arg) and down_cast<const Integer &>(*arg).is_zero())
+        return false;
+    // e.g cot(k*pi/2)
+    if (trig_has_basic_shift(arg)) {
+        return false;
+    }
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> cot(const RCP<const Basic> &arg)
+{
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().cot(*arg);
+    }
+
+    if (is_a<ACot>(*arg)) {
+        return down_cast<const ACot &>(*arg).get_arg();
+    } else if (is_a<ATan>(*arg)) {
+        return div(one, down_cast<const ATan &>(*arg).get_arg());
+    }
+
+    RCP<const Basic> ret_arg;
+    int index, sign;
+    bool conjugate = trig_simplify(arg, 1, true, true,            // input
+                                   outArg(ret_arg), index, sign); // output
+
+    if (conjugate) {
+        // tan has to be returned
+        if (sign == 1) {
+            return tan(ret_arg);
+        } else {
+            return mul(minus_one, tan(ret_arg));
+        }
+    } else {
+        if (eq(*ret_arg, *zero)) {
+            return mul(integer(sign),
+                       div(sin_table()[(index + 6) % 24], sin_table()[index]));
+        } else {
+            if (sign == 1) {
+                if (neq(*ret_arg, *arg)) {
+                    return cot(ret_arg);
+                } else {
+                    return make_rcp<const Cot>(ret_arg);
+                }
+            } else {
+                return mul(minus_one, cot(ret_arg));
+            }
+        }
+    }
+}
+
+/* ---------------------------- */
+
+Csc::Csc(const RCP<const Basic> &arg) : TrigFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Csc::is_canonical(const RCP<const Basic> &arg) const
+{
+    // e.g. Csc(0)
+    if (is_a<Integer>(*arg) and down_cast<const Integer &>(*arg).is_zero())
+        return false;
+    // e.g csc(k*pi/2)
+    if (trig_has_basic_shift(arg)) {
+        return false;
+    }
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> csc(const RCP<const Basic> &arg)
+{
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().csc(*arg);
+    }
+
+    if (is_a<ACsc>(*arg)) {
+        return down_cast<const ACsc &>(*arg).get_arg();
+    } else if (is_a<ASin>(*arg)) {
+        return div(one, down_cast<const ASin &>(*arg).get_arg());
+    }
+
+    RCP<const Basic> ret_arg;
+    int index, sign;
+    bool conjugate = trig_simplify(arg, 2, true, false,           // input
+                                   outArg(ret_arg), index, sign); // output
+
+    if (conjugate) {
+        // cos has to be returned
+        if (sign == 1) {
+            return sec(ret_arg);
+        } else {
+            return mul(minus_one, sec(ret_arg));
+        }
+    } else {
+        if (eq(*ret_arg, *zero)) {
+            return mul(integer(sign), div(one, sin_table()[index]));
+        } else {
+            if (sign == 1) {
+                if (neq(*ret_arg, *arg)) {
+                    return csc(ret_arg);
+                } else {
+                    return make_rcp<const Csc>(ret_arg);
+                }
+            } else {
+                return mul(minus_one, csc(ret_arg));
+            }
+        }
+    }
+}
+
+/* ---------------------------- */
+
+Sec::Sec(const RCP<const Basic> &arg) : TrigFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Sec::is_canonical(const RCP<const Basic> &arg) const
+{
+    // e.g. Sec(0)
+    if (is_a<Integer>(*arg) and down_cast<const Integer &>(*arg).is_zero())
+        return false;
+    // e.g sec(k*pi/2)
+    if (trig_has_basic_shift(arg)) {
+        return false;
+    }
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> sec(const RCP<const Basic> &arg)
+{
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().sec(*arg);
+    }
+
+    if (is_a<ASec>(*arg)) {
+        return down_cast<const ASec &>(*arg).get_arg();
+    } else if (is_a<ACos>(*arg)) {
+        return div(one, down_cast<const ACos &>(*arg).get_arg());
+    }
+
+    RCP<const Basic> ret_arg;
+    int index, sign;
+    bool conjugate = trig_simplify(arg, 2, false, true,           // input
+                                   outArg(ret_arg), index, sign); // output
+
+    if (conjugate) {
+        // csc has to be returned
+        if (sign == 1) {
+            return csc(ret_arg);
+        } else {
+            return mul(minus_one, csc(ret_arg));
+        }
+    } else {
+        if (eq(*ret_arg, *zero)) {
+            return mul(integer(sign), div(one, sin_table()[(index + 6) % 24]));
+        } else {
+            if (sign == 1) {
+                if (neq(*ret_arg, *arg)) {
+                    return sec(ret_arg);
+                } else {
+                    return make_rcp<const Sec>(ret_arg);
+                }
+            } else {
+                return mul(minus_one, sec(ret_arg));
+            }
+        }
+    }
+}
+/* ---------------------------- */
+
+// simplifies trigonometric functions wherever possible
+// currently deals with simplifications of type sin(acos())
+RCP<const Basic> trig_to_sqrt(const RCP<const Basic> &arg)
+{
+    RCP<const Basic> i_arg;
+
+    if (is_a<Sin>(*arg)) {
+        if (is_a<ACos>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ACos &>(*(arg->get_args()[0])).get_arg();
+            return sqrt(sub(one, pow(i_arg, i2)));
+        } else if (is_a<ATan>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ATan &>(*(arg->get_args()[0])).get_arg();
+            return div(i_arg, sqrt(add(one, pow(i_arg, i2))));
+        } else if (is_a<ASec>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ASec &>(*(arg->get_args()[0])).get_arg();
+            return sqrt(sub(one, pow(i_arg, im2)));
+        } else if (is_a<ACot>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ACot &>(*(arg->get_args()[0])).get_arg();
+            return div(one, mul(i_arg, sqrt(add(one, pow(i_arg, im2)))));
+        }
+    } else if (is_a<Cos>(*arg)) {
+        if (is_a<ASin>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ASin &>(*(arg->get_args()[0])).get_arg();
+            return sqrt(sub(one, pow(i_arg, i2)));
+        } else if (is_a<ATan>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ATan &>(*(arg->get_args()[0])).get_arg();
+            return div(one, sqrt(add(one, pow(i_arg, i2))));
+        } else if (is_a<ACsc>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ACsc &>(*(arg->get_args()[0])).get_arg();
+            return sqrt(sub(one, pow(i_arg, im2)));
+        } else if (is_a<ACot>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ACot &>(*(arg->get_args()[0])).get_arg();
+            return div(one, sqrt(add(one, pow(i_arg, im2))));
+        }
+    } else if (is_a<Tan>(*arg)) {
+        if (is_a<ASin>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ASin &>(*(arg->get_args()[0])).get_arg();
+            return div(i_arg, sqrt(sub(one, pow(i_arg, i2))));
+        } else if (is_a<ACos>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ACos &>(*(arg->get_args()[0])).get_arg();
+            return div(sqrt(sub(one, pow(i_arg, i2))), i_arg);
+        } else if (is_a<ACsc>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ACsc &>(*(arg->get_args()[0])).get_arg();
+            return div(one, mul(i_arg, sqrt(sub(one, pow(i_arg, im2)))));
+        } else if (is_a<ASec>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ASec &>(*(arg->get_args()[0])).get_arg();
+            return mul(i_arg, sqrt(sub(one, pow(i_arg, im2))));
+        }
+    } else if (is_a<Csc>(*arg)) {
+        if (is_a<ACos>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ACos &>(*(arg->get_args()[0])).get_arg();
+            return div(one, sqrt(sub(one, pow(i_arg, i2))));
+        } else if (is_a<ATan>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ATan &>(*(arg->get_args()[0])).get_arg();
+            return div(sqrt(add(one, pow(i_arg, i2))), i_arg);
+        } else if (is_a<ASec>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ASec &>(*(arg->get_args()[0])).get_arg();
+            return div(one, sqrt(sub(one, pow(i_arg, im2))));
+        } else if (is_a<ACot>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ACot &>(*(arg->get_args()[0])).get_arg();
+            return mul(i_arg, sqrt(add(one, pow(i_arg, im2))));
+        }
+    } else if (is_a<Sec>(*arg)) {
+        if (is_a<ASin>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ASin &>(*(arg->get_args()[0])).get_arg();
+            return div(one, sqrt(sub(one, pow(i_arg, i2))));
+        } else if (is_a<ATan>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ATan &>(*(arg->get_args()[0])).get_arg();
+            return sqrt(add(one, pow(i_arg, i2)));
+        } else if (is_a<ACsc>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ACsc &>(*(arg->get_args()[0])).get_arg();
+            return div(one, sqrt(sub(one, pow(i_arg, im2))));
+        } else if (is_a<ACot>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ACot &>(*(arg->get_args()[0])).get_arg();
+            return sqrt(add(one, pow(i_arg, im2)));
+        }
+    } else if (is_a<Cot>(*arg)) {
+        if (is_a<ASin>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ASin &>(*(arg->get_args()[0])).get_arg();
+            return div(sqrt(sub(one, pow(i_arg, i2))), i_arg);
+        } else if (is_a<ACos>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ACos &>(*(arg->get_args()[0])).get_arg();
+            return div(i_arg, sqrt(sub(one, pow(i_arg, i2))));
+        } else if (is_a<ACsc>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ACsc &>(*(arg->get_args()[0])).get_arg();
+            return mul(i_arg, sqrt(sub(one, pow(i_arg, im2))));
+        } else if (is_a<ASec>(*arg->get_args()[0])) {
+            i_arg = down_cast<const ASec &>(*(arg->get_args()[0])).get_arg();
+            return div(one, mul(i_arg, sqrt(sub(one, pow(i_arg, im2)))));
+        }
+    }
+
+    return arg;
+}
+
+/* ---------------------------- */
+ASin::ASin(const RCP<const Basic> &arg) : InverseTrigFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool ASin::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero) or eq(*arg, *one) or eq(*arg, *minus_one))
+        return false;
+    RCP<const Basic> index;
+    if (inverse_lookup(inverse_cst(), get_arg(), outArg(index))) {
+        return false;
+    }
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> asin(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return zero;
+    else if (eq(*arg, *one))
+        return div(pi, i2);
+    else if (eq(*arg, *minus_one))
+        return mul(minus_one, div(pi, i2));
+    else if (is_a_Number(*arg)
+             and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().asin(*arg);
+    }
+
+    RCP<const Basic> index;
+    bool b = inverse_lookup(inverse_cst(), arg, outArg(index));
+    if (b) {
+        return div(pi, index);
+    } else {
+        return make_rcp<const ASin>(arg);
+    }
+}
+
+ACos::ACos(const RCP<const Basic> &arg) : InverseTrigFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool ACos::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero) or eq(*arg, *one) or eq(*arg, *minus_one))
+        return false;
+    RCP<const Basic> index;
+    if (inverse_lookup(inverse_cst(), get_arg(), outArg(index))) {
+        return false;
+    }
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> acos(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return div(pi, i2);
+    else if (eq(*arg, *one))
+        return zero;
+    else if (eq(*arg, *minus_one))
+        return pi;
+    else if (is_a_Number(*arg)
+             and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().acos(*arg);
+    }
+
+    RCP<const Basic> index;
+    bool b = inverse_lookup(inverse_cst(), arg, outArg(index));
+    if (b) {
+        return sub(div(pi, i2), div(pi, index));
+    } else {
+        return make_rcp<const ACos>(arg);
+    }
+}
+
+ASec::ASec(const RCP<const Basic> &arg) : InverseTrigFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool ASec::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *one) or eq(*arg, *minus_one))
+        return false;
+    RCP<const Basic> index;
+    if (inverse_lookup(inverse_cst(), div(one, get_arg()), outArg(index))) {
+        return false;
+    } else if (is_a_Number(*arg)
+               and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> asec(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *one))
+        return zero;
+    else if (eq(*arg, *minus_one))
+        return pi;
+    else if (is_a_Number(*arg)
+             and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().asec(*arg);
+    }
+
+    RCP<const Basic> index;
+    bool b = inverse_lookup(inverse_cst(), div(one, arg), outArg(index));
+    if (b) {
+        return sub(div(pi, i2), div(pi, index));
+    } else {
+        return make_rcp<const ASec>(arg);
+    }
+}
+
+ACsc::ACsc(const RCP<const Basic> &arg) : InverseTrigFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool ACsc::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *one) or eq(*arg, *minus_one))
+        return false;
+    RCP<const Basic> index;
+    if (inverse_lookup(inverse_cst(), div(one, arg), outArg(index))) {
+        return false;
+    }
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> acsc(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *one))
+        return div(pi, i2);
+    else if (eq(*arg, *minus_one))
+        return div(pi, im2);
+    else if (is_a_Number(*arg)
+             and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().acsc(*arg);
+    }
+
+    RCP<const Basic> index;
+    bool b = inverse_lookup(inverse_cst(), div(one, arg), outArg(index));
+    if (b) {
+        return div(pi, index);
+    } else {
+        return make_rcp<const ACsc>(arg);
+    }
+}
+
+ATan::ATan(const RCP<const Basic> &arg) : InverseTrigFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool ATan::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero) or eq(*arg, *one) or eq(*arg, *minus_one))
+        return false;
+    RCP<const Basic> index;
+    if (inverse_lookup(inverse_tct(), get_arg(), outArg(index))) {
+        return false;
+    }
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> atan(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return zero;
+    else if (eq(*arg, *one))
+        return div(pi, mul(i2, i2));
+    else if (eq(*arg, *minus_one))
+        return mul(minus_one, div(pi, mul(i2, i2)));
+    else if (is_a_Number(*arg)
+             and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().atan(*arg);
+    }
+
+    RCP<const Basic> index;
+    bool b = inverse_lookup(inverse_tct(), arg, outArg(index));
+    if (b) {
+        return div(pi, index);
+    } else {
+        return make_rcp<const ATan>(arg);
+    }
+}
+
+ACot::ACot(const RCP<const Basic> &arg) : InverseTrigFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool ACot::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero) or eq(*arg, *one) or eq(*arg, *minus_one))
+        return false;
+    RCP<const Basic> index;
+    if (inverse_lookup(inverse_tct(), arg, outArg(index))) {
+        return false;
+    }
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> acot(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return div(pi, i2);
+    else if (eq(*arg, *one))
+        return div(pi, mul(i2, i2));
+    else if (eq(*arg, *minus_one))
+        return mul(i3, div(pi, mul(i2, i2)));
+    else if (is_a_Number(*arg)
+             and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().acot(*arg);
+    }
+
+    RCP<const Basic> index;
+    bool b = inverse_lookup(inverse_tct(), arg, outArg(index));
+    if (b) {
+        return sub(div(pi, i2), div(pi, index));
+    } else {
+        return make_rcp<const ACot>(arg);
+    }
+}
+
+ATan2::ATan2(const RCP<const Basic> &num, const RCP<const Basic> &den)
+    : TwoArgFunction(num, den)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(num, den))
+}
+
+bool ATan2::is_canonical(const RCP<const Basic> &num,
+                         const RCP<const Basic> &den) const
+{
+    if (eq(*num, *den) or eq(*num, *mul(minus_one, den)))
+        return false;
+    if (eq(*num, *zero)) {
+        return not(is_a_Number(*den) or eq(*den, *zero));
+    }
+    if (eq(*den, *zero)) {
+        return not is_a_Number(*num);
+    }
+    RCP<const Basic> index;
+    bool b = inverse_lookup(inverse_tct(), div(num, den), outArg(index));
+    if (b)
+        return false;
+    else
+        return true;
+}
+
+RCP<const Basic> ATan2::create(const RCP<const Basic> &a,
+                               const RCP<const Basic> &b) const
+{
+    return atan2(a, b);
+}
+
+RCP<const Basic> atan2(const RCP<const Basic> &num, const RCP<const Basic> &den)
+{
+    if (eq(*num, *zero)) {
+        if (eq(*den, *zero)) {
+            return Nan;
+        }
+        if (is_a_Number(*den)) {
+            RCP<const Number> den_new = rcp_static_cast<const Number>(den);
+            if (den_new->is_negative())
+                return pi;
+            else if (den_new->is_positive())
+                return zero;
+            else {
+                return Nan;
+            }
+        }
+        return make_rcp<const ATan2>(num, den);
+    } else if (eq(*den, *zero)) {
+        if (is_a_Number(*num)) {
+            RCP<const Number> num_new = rcp_static_cast<const Number>(num);
+            if (num_new->is_negative())
+                return div(pi, im2);
+            else
+                return div(pi, i2);
+        }
+        return make_rcp<const ATan2>(num, den);
+    }
+    RCP<const Basic> index;
+    bool b = inverse_lookup(inverse_tct(), div(num, den), outArg(index));
+    if (b) {
+        SYMENGINE_ASSERT(is_a_Number(*index));
+        RCP<const Number> index_num = rcp_static_cast<const Number>(index);
+        if (index_num->is_positive()) {
+            return add(div(pi, index), mul(div(pi, i2), sub(sign(den), one)));
+        } else {
+            return sub(div(pi, index), mul(div(pi, i2), sub(sign(den), one)));
+        }
+    } else {
+        return make_rcp<const ATan2>(num, den);
+    }
+}
+
+/* ---------------------------- */
+
+RCP<const Basic> Sin::create(const RCP<const Basic> &arg) const
+{
+    return sin(arg);
+}
+
+RCP<const Basic> Cos::create(const RCP<const Basic> &arg) const
+{
+    return cos(arg);
+}
+
+RCP<const Basic> Tan::create(const RCP<const Basic> &arg) const
+{
+    return tan(arg);
+}
+
+RCP<const Basic> Cot::create(const RCP<const Basic> &arg) const
+{
+    return cot(arg);
+}
+
+RCP<const Basic> Sec::create(const RCP<const Basic> &arg) const
+{
+    return sec(arg);
+}
+
+RCP<const Basic> Csc::create(const RCP<const Basic> &arg) const
+{
+    return csc(arg);
+}
+
+RCP<const Basic> ASin::create(const RCP<const Basic> &arg) const
+{
+    return asin(arg);
+}
+
+RCP<const Basic> ACos::create(const RCP<const Basic> &arg) const
+{
+    return acos(arg);
+}
+
+RCP<const Basic> ATan::create(const RCP<const Basic> &arg) const
+{
+    return atan(arg);
+}
+
+RCP<const Basic> ACot::create(const RCP<const Basic> &arg) const
+{
+    return acot(arg);
+}
+
+RCP<const Basic> ASec::create(const RCP<const Basic> &arg) const
+{
+    return asec(arg);
+}
+
+RCP<const Basic> ACsc::create(const RCP<const Basic> &arg) const
+{
+    return acsc(arg);
+}
+
+/* ---------------------------- */
+
+Log::Log(const RCP<const Basic> &arg) : OneArgFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Log::is_canonical(const RCP<const Basic> &arg) const
+{
+    //  log(0)
+    if (is_a<Integer>(*arg) and down_cast<const Integer &>(*arg).is_zero())
+        return false;
+    //  log(1)
+    if (is_a<Integer>(*arg) and down_cast<const Integer &>(*arg).is_one())
+        return false;
+    // log(E)
+    if (eq(*arg, *E))
+        return false;
+
+    if (is_a_Number(*arg) and down_cast<const Number &>(*arg).is_negative())
+        return false;
+
+    // log(Inf) is also handled here.
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact())
+        return false;
+
+    // log(3I) should be expanded to log(3) + I*pi/2
+    if (is_a<Complex>(*arg) and down_cast<const Complex &>(*arg).is_re_zero())
+        return false;
+    // log(num/den) = log(num) - log(den)
+    if (is_a<Rational>(*arg))
+        return false;
+    return true;
+}
+
+RCP<const Basic> Log::create(const RCP<const Basic> &a) const
+{
+    return log(a);
+}
+
+RCP<const Basic> log(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return ComplexInf;
+    if (eq(*arg, *one))
+        return zero;
+    if (eq(*arg, *E))
+        return one;
+
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().log(*_arg);
+        } else if (_arg->is_negative()) {
+            return add(log(mul(minus_one, _arg)), mul(pi, I));
+        }
+    }
+
+    if (is_a<Rational>(*arg)) {
+        RCP<const Integer> num, den;
+        get_num_den(down_cast<const Rational &>(*arg), outArg(num),
+                    outArg(den));
+        return sub(log(num), log(den));
+    }
+
+    if (is_a<Complex>(*arg)) {
+        RCP<const Complex> _arg = rcp_static_cast<const Complex>(arg);
+        if (_arg->is_re_zero()) {
+            RCP<const Number> arg_img = _arg->imaginary_part();
+            if (arg_img->is_negative()) {
+                return sub(log(mul(minus_one, arg_img)),
+                           mul(I, div(pi, integer(2))));
+            } else if (arg_img->is_zero()) {
+                return ComplexInf;
+            } else if (arg_img->is_positive()) {
+                return add(log(arg_img), mul(I, div(pi, integer(2))));
+            }
+        }
+    }
+
+    // lakutata: log(e**r) is r for an exact real number r, as SymPy
+    if (is_a<Pow>(*arg) and eq(*down_cast<const Pow &>(*arg).get_base(), *E)) {
+        const RCP<const Basic> &exponent = down_cast<const Pow &>(*arg).get_exp();
+        if (is_a<Integer>(*exponent) or is_a<Rational>(*exponent))
+            return exponent;
+    }
+
+    return make_rcp<const Log>(arg);
+}
+
+RCP<const Basic> log(const RCP<const Basic> &arg, const RCP<const Basic> &base)
+{
+    return div(log(arg), log(base));
+}
+
+LambertW::LambertW(const RCP<const Basic> &arg) : OneArgFunction{arg}
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool LambertW::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero))
+        return false;
+    if (eq(*arg, *E))
+        return false;
+    if (eq(*arg, *div(neg(one), E)))
+        return false;
+    if (eq(*arg, *div(log(i2), im2)))
+        return false;
+    return true;
+}
+
+RCP<const Basic> LambertW::create(const RCP<const Basic> &arg) const
+{
+    return lambertw(arg);
+}
+
+RCP<const Basic> lambertw(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return zero;
+    if (eq(*arg, *E))
+        return one;
+    if (eq(*arg, *div(neg(one), E)))
+        return minus_one;
+    if (eq(*arg, *div(log(i2), im2)))
+        return mul(minus_one, log(i2));
+    return make_rcp<const LambertW>(arg);
+}
+
+FunctionSymbol::FunctionSymbol(std::string name, const RCP<const Basic> &arg)
+    : MultiArgFunction({arg}), name_{name} {SYMENGINE_ASSIGN_TYPEID()
+                                                SYMENGINE_ASSERT(
+                                                    is_canonical(get_vec()))}
+
+      FunctionSymbol::FunctionSymbol(std::string name, const vec_basic &arg)
+    : MultiArgFunction(arg), name_{name}
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(get_vec()))
+}
+
+bool FunctionSymbol::is_canonical(const vec_basic &arg) const
+{
+    return true;
+}
+
+hash_t FunctionSymbol::__hash__() const
+{
+    hash_t seed = SYMENGINE_FUNCTIONSYMBOL;
+    for (const auto &a : get_vec())
+        hash_combine<Basic>(seed, *a);
+    hash_combine<std::string>(seed, name_);
+    return seed;
+}
+
+bool FunctionSymbol::__eq__(const Basic &o) const
+{
+    // lakutata: the same type code, so that the function wrappers compare too
+    // (is_a<FunctionSymbol> is false for them: a wrapper was not equal to
+    // itself, and cos(wrapper) recursed without end)
+    if (o.get_type_code() == get_type_code()
+        and name_ == down_cast<const FunctionSymbol &>(o).name_
+        and unified_eq(get_vec(),
+                       down_cast<const FunctionSymbol &>(o).get_vec()))
+        return true;
+    return false;
+}
+
+int FunctionSymbol::compare(const Basic &o) const
+{
+    SYMENGINE_ASSERT(is_a<FunctionSymbol>(o))
+    const FunctionSymbol &s = down_cast<const FunctionSymbol &>(o);
+    if (name_ == s.name_)
+        return unified_compare(get_vec(), s.get_vec());
+    else
+        return name_ < s.name_ ? -1 : 1;
+}
+
+RCP<const Basic> FunctionSymbol::create(const vec_basic &x) const
+{
+    return make_rcp<const FunctionSymbol>(name_, x);
+}
+
+RCP<const Basic> function_symbol(std::string name, const vec_basic &arg)
+{
+    return make_rcp<const FunctionSymbol>(name, arg);
+}
+
+RCP<const Basic> function_symbol(std::string name, const RCP<const Basic> &arg)
+{
+    return make_rcp<const FunctionSymbol>(name, arg);
+}
+
+// lakutata: the values of a wrapper for numbers, through create() and eval()
+double FunctionWrapper::evaluate_real(const double *arguments) const
+{
+    vec_basic values;
+    for (size_t i = 0; i < get_args().size(); i++)
+        values.push_back(real_double(arguments[i]));
+    return eval_double(*create(values));
+}
+
+std::complex<double>
+FunctionWrapper::evaluate_complex(const std::complex<double> *arguments) const
+{
+    vec_basic values;
+    for (size_t i = 0; i < get_args().size(); i++)
+        values.push_back(complex_double(arguments[i]));
+    return eval_complex_double(*create(values));
+}
+
+FunctionWrapper::FunctionWrapper(std::string name, const RCP<const Basic> &arg)
+    : FunctionSymbol(name, arg){SYMENGINE_ASSIGN_TYPEID()}
+
+      FunctionWrapper::FunctionWrapper(std::string name, const vec_basic &vec)
+    : FunctionSymbol(name, vec){SYMENGINE_ASSIGN_TYPEID()}
+
+      /* ---------------------------- */
+
+      Derivative::Derivative(const RCP<const Basic> &arg,
+                             const multiset_basic &x)
+    : arg_{arg}, x_{x}
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg, x))
+}
+
+bool Derivative::is_canonical(const RCP<const Basic> &arg,
+                              const multiset_basic &x) const
+{
+    // Check that 'x' are Symbols:
+    for (const auto &a : x)
+        if (not is_a<Symbol>(*a))
+            return false;
+    if (is_a<FunctionSymbol>(*arg) or is_a<LeviCivita>(*arg)) {
+        for (auto &p : x) {
+            RCP<const Symbol> s = rcp_static_cast<const Symbol>(p);
+            RCP<const MultiArgFunction> f
+                = rcp_static_cast<const MultiArgFunction>(arg);
+            bool found_s = false;
+            // 's' should be one of the args of the function
+            // and should not appear anywhere else.
+            for (const auto &a : f->get_args()) {
+                if (eq(*a, *s)) {
+                    if (found_s) {
+                        return false;
+                    } else {
+                        found_s = true;
+                    }
+                } else if (neq(*a->diff(s), *zero)) {
+                    return false;
+                }
+            }
+            if (!found_s) {
+                return false;
+            }
+        }
+        return true;
+    } else if (is_a<Abs>(*arg)) {
+        return true;
+    } else if (is_a<FunctionWrapper>(*arg)) {
+        return true;
+    } else if (is_a<PolyGamma>(*arg) or is_a<Zeta>(*arg)
+               or is_a<UpperGamma>(*arg) or is_a<LowerGamma>(*arg)
+               or is_a<Dirichlet_eta>(*arg)) {
+        bool found = false;
+        auto v = arg->get_args();
+        for (auto &p : x) {
+            if (has_symbol(*v[0], *rcp_static_cast<const Symbol>(p))) {
+                found = true;
+                break;
+            }
+        }
+        return found;
+    } else if (is_a<KroneckerDelta>(*arg)) {
+        bool found = false;
+        auto v = arg->get_args();
+        for (auto &p : x) {
+            if (has_symbol(*v[0], *rcp_static_cast<const Symbol>(p))
+                or has_symbol(*v[1], *rcp_static_cast<const Symbol>(p))) {
+                found = true;
+                break;
+            }
+        }
+        return found;
+    }
+    return false;
+}
+
+hash_t Derivative::__hash__() const
+{
+    hash_t seed = SYMENGINE_DERIVATIVE;
+    hash_combine<Basic>(seed, *arg_);
+    for (auto &p : x_) {
+        hash_combine<Basic>(seed, *p);
+    }
+    return seed;
+}
+
+bool Derivative::__eq__(const Basic &o) const
+{
+    if (is_a<Derivative>(o)
+        and eq(*arg_, *(down_cast<const Derivative &>(o).arg_))
+        and unified_eq(x_, down_cast<const Derivative &>(o).x_))
+        return true;
+    return false;
+}
+
+int Derivative::compare(const Basic &o) const
+{
+    SYMENGINE_ASSERT(is_a<Derivative>(o))
+    const Derivative &s = down_cast<const Derivative &>(o);
+    int cmp = arg_->__cmp__(*(s.arg_));
+    if (cmp != 0)
+        return cmp;
+    cmp = unified_compare(x_, s.x_);
+    return cmp;
+}
+
+// Subs class
+Subs::Subs(const RCP<const Basic> &arg, const map_basic_basic &dict)
+    : arg_{arg}, dict_{dict}
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg, dict))
+}
+
+bool Subs::is_canonical(const RCP<const Basic> &arg,
+                        const map_basic_basic &dict) const
+{
+    if (is_a<Derivative>(*arg)) {
+        return true;
+    }
+    return false;
+}
+
+hash_t Subs::__hash__() const
+{
+    hash_t seed = SYMENGINE_SUBS;
+    hash_combine<Basic>(seed, *arg_);
+    for (const auto &p : dict_) {
+        hash_combine<Basic>(seed, *p.first);
+        hash_combine<Basic>(seed, *p.second);
+    }
+    return seed;
+}
+
+bool Subs::__eq__(const Basic &o) const
+{
+    if (is_a<Subs>(o) and eq(*arg_, *(down_cast<const Subs &>(o).arg_))
+        and unified_eq(dict_, down_cast<const Subs &>(o).dict_))
+        return true;
+    return false;
+}
+
+int Subs::compare(const Basic &o) const
+{
+    SYMENGINE_ASSERT(is_a<Subs>(o))
+    const Subs &s = down_cast<const Subs &>(o);
+    int cmp = arg_->__cmp__(*(s.arg_));
+    if (cmp != 0)
+        return cmp;
+    cmp = unified_compare(dict_, s.dict_);
+    return cmp;
+}
+
+vec_basic Subs::get_variables() const
+{
+    vec_basic v;
+    for (const auto &p : dict_) {
+        v.push_back(p.first);
+    }
+    return v;
+}
+
+vec_basic Subs::get_point() const
+{
+    vec_basic v;
+    for (const auto &p : dict_) {
+        v.push_back(p.second);
+    }
+    return v;
+}
+
+vec_basic Subs::get_args() const
+{
+    vec_basic v = {arg_};
+    for (const auto &p : dict_) {
+        v.push_back(p.first);
+    }
+    for (const auto &p : dict_) {
+        v.push_back(p.second);
+    }
+    return v;
+}
+
+Sinh::Sinh(const RCP<const Basic> &arg) : HyperbolicFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Sinh::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero))
+        return false;
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_negative()) {
+            return false;
+        } else if (not down_cast<const Number &>(*arg).is_exact()) {
+            return false;
+        }
+    }
+    if (could_extract_minus(*arg))
+        return false;
+    return true;
+}
+
+RCP<const Basic> sinh(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return zero;
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().sinh(*_arg);
+        } else if (_arg->is_negative()) {
+            return neg(sinh(zero->sub(*_arg)));
+        }
+    }
+    RCP<const Basic> d;
+    bool b = handle_minus(arg, outArg(d));
+    if (b) {
+        return neg(sinh(d));
+    }
+    return make_rcp<const Sinh>(d);
+}
+
+Csch::Csch(const RCP<const Basic> &arg) : HyperbolicFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Csch::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero))
+        return false;
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_negative()) {
+            return false;
+        } else if (not down_cast<const Number &>(*arg).is_exact()) {
+            return false;
+        }
+    }
+    if (could_extract_minus(*arg))
+        return false;
+    return true;
+}
+
+RCP<const Basic> csch(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero)) {
+        return ComplexInf;
+    }
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().csch(*_arg);
+        } else if (_arg->is_negative()) {
+            return neg(csch(zero->sub(*_arg)));
+        }
+    }
+    RCP<const Basic> d;
+    bool b = handle_minus(arg, outArg(d));
+    if (b) {
+        return neg(csch(d));
+    }
+    return make_rcp<const Csch>(d);
+}
+
+Cosh::Cosh(const RCP<const Basic> &arg) : HyperbolicFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Cosh::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero))
+        return false;
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_negative()) {
+            return false;
+        } else if (not down_cast<const Number &>(*arg).is_exact()) {
+            return false;
+        }
+    }
+    if (could_extract_minus(*arg))
+        return false;
+    return true;
+}
+
+RCP<const Basic> cosh(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return one;
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().cosh(*_arg);
+        } else if (_arg->is_negative()) {
+            return cosh(zero->sub(*_arg));
+        }
+    }
+    RCP<const Basic> d;
+    handle_minus(arg, outArg(d));
+    return make_rcp<const Cosh>(d);
+}
+
+Sech::Sech(const RCP<const Basic> &arg) : HyperbolicFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Sech::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero))
+        return false;
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_negative()) {
+            return false;
+        } else if (not down_cast<const Number &>(*arg).is_exact()) {
+            return false;
+        }
+    }
+    if (could_extract_minus(*arg))
+        return false;
+    return true;
+}
+
+RCP<const Basic> sech(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return one;
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().sech(*_arg);
+        } else if (_arg->is_negative()) {
+            return sech(zero->sub(*_arg));
+        }
+    }
+    RCP<const Basic> d;
+    handle_minus(arg, outArg(d));
+    return make_rcp<const Sech>(d);
+}
+
+Tanh::Tanh(const RCP<const Basic> &arg) : HyperbolicFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Tanh::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero))
+        return false;
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_negative()) {
+            return false;
+        } else if (not down_cast<const Number &>(*arg).is_exact()) {
+            return false;
+        }
+    }
+    if (could_extract_minus(*arg))
+        return false;
+    return true;
+}
+
+RCP<const Basic> tanh(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return zero;
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().tanh(*_arg);
+        } else if (_arg->is_negative()) {
+            return neg(tanh(zero->sub(*_arg)));
+        }
+    }
+
+    RCP<const Basic> d;
+    bool b = handle_minus(arg, outArg(d));
+    if (b) {
+        return neg(tanh(d));
+    }
+    return make_rcp<const Tanh>(d);
+}
+
+Coth::Coth(const RCP<const Basic> &arg) : HyperbolicFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Coth::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero))
+        return false;
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_negative()) {
+            return false;
+        } else if (not down_cast<const Number &>(*arg).is_exact()) {
+            return false;
+        }
+    }
+    if (could_extract_minus(*arg))
+        return false;
+    return true;
+}
+
+RCP<const Basic> coth(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero)) {
+        return ComplexInf;
+    }
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().coth(*_arg);
+        } else if (_arg->is_negative()) {
+            return neg(coth(zero->sub(*_arg)));
+        }
+    }
+    RCP<const Basic> d;
+    bool b = handle_minus(arg, outArg(d));
+    if (b) {
+        return neg(coth(d));
+    }
+    return make_rcp<const Coth>(d);
+}
+
+ASinh::ASinh(const RCP<const Basic> &arg) : InverseHyperbolicFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool ASinh::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero) or eq(*arg, *one) or eq(*arg, *minus_one))
+        return false;
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_negative()) {
+            return false;
+        } else if (not down_cast<const Number &>(*arg).is_exact()) {
+            return false;
+        }
+    }
+    if (could_extract_minus(*arg))
+        return false;
+    return true;
+}
+
+RCP<const Basic> asinh(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return zero;
+    if (eq(*arg, *one))
+        return log(add(one, sq2));
+    if (eq(*arg, *minus_one))
+        return log(sub(sq2, one));
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().asinh(*_arg);
+        } else if (_arg->is_negative()) {
+            return neg(asinh(zero->sub(*_arg)));
+        }
+    }
+    RCP<const Basic> d;
+    bool b = handle_minus(arg, outArg(d));
+    if (b) {
+        return neg(asinh(d));
+    }
+    return make_rcp<const ASinh>(d);
+}
+
+ACsch::ACsch(const RCP<const Basic> &arg) : InverseHyperbolicFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool ACsch::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *one) or eq(*arg, *minus_one))
+        return false;
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_negative()) {
+            return false;
+        } else if (not down_cast<const Number &>(*arg).is_exact()) {
+            return false;
+        }
+    }
+    if (could_extract_minus(*arg))
+        return false;
+    return true;
+}
+
+RCP<const Basic> acsch(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *one))
+        return log(add(one, sq2));
+    if (eq(*arg, *minus_one))
+        return log(sub(sq2, one));
+
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().acsch(*_arg);
+        }
+    }
+
+    RCP<const Basic> d;
+    bool b = handle_minus(arg, outArg(d));
+    if (b) {
+        return neg(acsch(d));
+    }
+    return make_rcp<const ACsch>(d);
+}
+
+ACosh::ACosh(const RCP<const Basic> &arg) : InverseHyperbolicFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool ACosh::is_canonical(const RCP<const Basic> &arg) const
+{
+    // TODO: Lookup into a cst table once complex is implemented
+    if (eq(*arg, *one))
+        return false;
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> acosh(const RCP<const Basic> &arg)
+{
+    // TODO: Lookup into a cst table once complex is implemented
+    if (eq(*arg, *one))
+        return zero;
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().acosh(*arg);
+    }
+    return make_rcp<const ACosh>(arg);
+}
+
+ATanh::ATanh(const RCP<const Basic> &arg) : InverseHyperbolicFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool ATanh::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (eq(*arg, *zero))
+        return false;
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_negative()) {
+            return false;
+        } else if (not down_cast<const Number &>(*arg).is_exact()) {
+            return false;
+        }
+    }
+    if (could_extract_minus(*arg))
+        return false;
+    return true;
+}
+
+RCP<const Basic> atanh(const RCP<const Basic> &arg)
+{
+    if (eq(*arg, *zero))
+        return zero;
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().atanh(*_arg);
+        } else if (_arg->is_negative()) {
+            return neg(atanh(zero->sub(*_arg)));
+        }
+    }
+    RCP<const Basic> d;
+    bool b = handle_minus(arg, outArg(d));
+    if (b) {
+        return neg(atanh(d));
+    }
+    return make_rcp<const ATanh>(d);
+}
+
+ACoth::ACoth(const RCP<const Basic> &arg) : InverseHyperbolicFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool ACoth::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a_Number(*arg)) {
+        if (down_cast<const Number &>(*arg).is_negative()) {
+            return false;
+        } else if (not down_cast<const Number &>(*arg).is_exact()) {
+            return false;
+        }
+    }
+    if (could_extract_minus(*arg))
+        return false;
+    return true;
+}
+
+RCP<const Basic> acoth(const RCP<const Basic> &arg)
+{
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().acoth(*_arg);
+        } else if (_arg->is_negative()) {
+            return neg(acoth(zero->sub(*_arg)));
+        }
+    }
+    RCP<const Basic> d;
+    bool b = handle_minus(arg, outArg(d));
+    if (b) {
+        return neg(acoth(d));
+    }
+    return make_rcp<const ACoth>(d);
+}
+
+ASech::ASech(const RCP<const Basic> &arg) : InverseHyperbolicFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool ASech::is_canonical(const RCP<const Basic> &arg) const
+{
+    // TODO: Lookup into a cst table once complex is implemented
+    if (eq(*arg, *one))
+        return false;
+    if (eq(*arg, *zero))
+        return false;
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> asech(const RCP<const Basic> &arg)
+{
+    // TODO: Lookup into a cst table once complex is implemented
+    if (eq(*arg, *one))
+        return zero;
+    if (eq(*arg, *zero))
+        return Inf;
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().asech(*_arg);
+        }
+    }
+    return make_rcp<const ASech>(arg);
+}
+
+RCP<const Basic> Sinh::create(const RCP<const Basic> &arg) const
+{
+    return sinh(arg);
+}
+
+RCP<const Basic> Csch::create(const RCP<const Basic> &arg) const
+{
+    return csch(arg);
+}
+
+RCP<const Basic> Cosh::create(const RCP<const Basic> &arg) const
+{
+    return cosh(arg);
+}
+
+RCP<const Basic> Sech::create(const RCP<const Basic> &arg) const
+{
+    return sech(arg);
+}
+
+RCP<const Basic> Tanh::create(const RCP<const Basic> &arg) const
+{
+    return tanh(arg);
+}
+
+RCP<const Basic> Coth::create(const RCP<const Basic> &arg) const
+{
+    return coth(arg);
+}
+
+RCP<const Basic> ASinh::create(const RCP<const Basic> &arg) const
+{
+    return asinh(arg);
+}
+
+RCP<const Basic> ACsch::create(const RCP<const Basic> &arg) const
+{
+    return acsch(arg);
+}
+
+RCP<const Basic> ACosh::create(const RCP<const Basic> &arg) const
+{
+    return acosh(arg);
+}
+
+RCP<const Basic> ATanh::create(const RCP<const Basic> &arg) const
+{
+    return atanh(arg);
+}
+
+RCP<const Basic> ACoth::create(const RCP<const Basic> &arg) const
+{
+    return acoth(arg);
+}
+
+RCP<const Basic> ASech::create(const RCP<const Basic> &arg) const
+{
+    return asech(arg);
+}
+
+KroneckerDelta::KroneckerDelta(const RCP<const Basic> &i,
+                               const RCP<const Basic> &j)
+    : TwoArgFunction(i, j)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(i, j))
+}
+
+bool KroneckerDelta::is_canonical(const RCP<const Basic> &i,
+                                  const RCP<const Basic> &j) const
+{
+    RCP<const Basic> diff = expand(sub(i, j));
+    if (eq(*diff, *zero)) {
+        return false;
+    } else if (is_a_Number(*diff)) {
+        return false;
+    } else {
+        // TODO: SymPy uses default key sorting to return in order
+        return true;
+    }
+}
+
+RCP<const Basic> KroneckerDelta::create(const RCP<const Basic> &a,
+                                        const RCP<const Basic> &b) const
+{
+    return kronecker_delta(a, b);
+}
+
+RCP<const Basic> kronecker_delta(const RCP<const Basic> &i,
+                                 const RCP<const Basic> &j)
+{
+    // Expand is needed to simplify things like `i-(i+1)` to `-1`
+    RCP<const Basic> diff = expand(sub(i, j));
+    if (eq(*diff, *zero)) {
+        return one;
+    } else if (is_a_Number(*diff)) {
+        return zero;
+    } else {
+        // SymPy uses default key sorting to return in order
+        return make_rcp<const KroneckerDelta>(i, j);
+    }
+}
+
+bool has_dup(const vec_basic &arg)
+{
+    map_basic_basic d;
+    auto it = d.end();
+    for (const auto &p : arg) {
+        it = d.find(p);
+        if (it == d.end()) {
+            insert(d, p, one);
+        } else {
+            return true;
+        }
+    }
+    return false;
+}
+
+LeviCivita::LeviCivita(const vec_basic &&arg) : MultiArgFunction(std::move(arg))
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(get_vec()))
+}
+
+bool LeviCivita::is_canonical(const vec_basic &arg) const
+{
+    bool are_int = true;
+    for (const auto &p : arg) {
+        if (not(is_a_Number(*p))) {
+            are_int = false;
+            break;
+        }
+    }
+    if (are_int) {
+        return false;
+    } else if (has_dup(arg)) {
+        return false;
+    } else {
+        return true;
+    }
+}
+
+RCP<const Basic> LeviCivita::create(const vec_basic &a) const
+{
+    return levi_civita(a);
+}
+
+RCP<const Basic> eval_levicivita(const vec_basic &arg, int len)
+{
+    int i, j;
+    RCP<const Basic> res = one;
+    for (i = 0; i < len; i++) {
+        for (j = i + 1; j < len; j++) {
+            res = mul(sub(arg[j], arg[i]), res);
+        }
+        res = div(res, factorial(i));
+    }
+    return res;
+}
+
+RCP<const Basic> levi_civita(const vec_basic &arg)
+{
+    bool are_int = true;
+    int len = 0;
+    for (const auto &p : arg) {
+        if (not(is_a_Number(*p))) {
+            are_int = false;
+            break;
+        } else {
+            len++;
+        }
+    }
+    if (are_int) {
+        return eval_levicivita(arg, len);
+    } else if (has_dup(arg)) {
+        return zero;
+    } else {
+        return make_rcp<const LeviCivita>(std::move(arg));
+    }
+}
+
+Zeta::Zeta(const RCP<const Basic> &s, const RCP<const Basic> &a)
+    : TwoArgFunction(s, a){SYMENGINE_ASSIGN_TYPEID()
+                               SYMENGINE_ASSERT(is_canonical(s, a))}
+
+      Zeta::Zeta(const RCP<const Basic> &s)
+    : TwoArgFunction(s, one)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(s, one))
+}
+
+bool Zeta::is_canonical(const RCP<const Basic> &s,
+                        const RCP<const Basic> &a) const
+{
+    if (eq(*s, *zero))
+        return false;
+    if (eq(*s, *one))
+        return false;
+    if (is_a<Integer>(*s) and is_a<Integer>(*a)) {
+        auto s_ = down_cast<const Integer &>(*s).as_int();
+        if (s_ < 0 || s_ % 2 == 0)
+            return false;
+    }
+    return true;
+}
+
+RCP<const Basic> Zeta::create(const RCP<const Basic> &a,
+                              const RCP<const Basic> &b) const
+{
+    return zeta(a, b);
+}
+
+RCP<const Basic> zeta(const RCP<const Basic> &s, const RCP<const Basic> &a)
+{
+    if (is_a_Number(*s)) {
+        if (down_cast<const Number &>(*s).is_zero()) {
+            return sub(div(one, i2), a);
+        } else if (down_cast<const Number &>(*s).is_one()) {
+            return infty(0);
+        } else if (is_a<Integer>(*s) and is_a<Integer>(*a)) {
+            auto s_ = down_cast<const Integer &>(*s).as_int();
+            auto a_ = down_cast<const Integer &>(*a).as_int();
+            RCP<const Basic> zeta;
+            if (s_ < 0) {
+                RCP<const Number> res = (s_ % 2 == 0) ? one : minus_one;
+                zeta
+                    = mulnum(res, divnum(bernoulli(-s_ + 1), integer(-s_ + 1)));
+            } else if (s_ % 2 == 0) {
+                RCP<const Number> b = bernoulli(s_);
+                RCP<const Number> f = factorial(s_);
+                zeta = divnum(pownum(integer(2), integer(s_ - 1)), f);
+                zeta = mul(zeta, mul(pow(pi, s), abs(b)));
+            } else {
+                return make_rcp<const Zeta>(s, a);
+            }
+            if (a_ < 0)
+                return add(zeta, harmonic(-a_, s_));
+            return sub(zeta, harmonic(a_ - 1, s_));
+        }
+    }
+    return make_rcp<const Zeta>(s, a);
+}
+
+RCP<const Basic> zeta(const RCP<const Basic> &s)
+{
+    return zeta(s, one);
+}
+
+Dirichlet_eta::Dirichlet_eta(const RCP<const Basic> &s) : OneArgFunction(s)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(s))
+}
+
+bool Dirichlet_eta::is_canonical(const RCP<const Basic> &s) const
+{
+    if (eq(*s, *one))
+        return false;
+    if (not(is_a<Zeta>(*zeta(s))))
+        return false;
+    return true;
+}
+
+RCP<const Basic> Dirichlet_eta::rewrite_as_zeta() const
+{
+    return mul(sub(one, pow(i2, sub(one, get_arg()))), zeta(get_arg()));
+}
+
+RCP<const Basic> Dirichlet_eta::create(const RCP<const Basic> &arg) const
+{
+    return dirichlet_eta(arg);
+}
+
+RCP<const Basic> dirichlet_eta(const RCP<const Basic> &s)
+{
+    if (is_a_Number(*s) and down_cast<const Number &>(*s).is_one()) {
+        return log(i2);
+    }
+    RCP<const Basic> z = zeta(s);
+    if (is_a<Zeta>(*z)) {
+        return make_rcp<const Dirichlet_eta>(s);
+    } else {
+        return mul(sub(one, pow(i2, sub(one, s))), z);
+    }
+}
+
+bool Erf::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a<Integer>(*arg) and down_cast<const Integer &>(*arg).is_zero())
+        return false;
+    if (could_extract_minus(*arg))
+        return false;
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> Erf::create(const RCP<const Basic> &arg) const
+{
+    return erf(arg);
+}
+
+RCP<const Basic> erf(const RCP<const Basic> &arg)
+{
+    if (is_a<Integer>(*arg) and down_cast<const Integer &>(*arg).is_zero()) {
+        return zero;
+    }
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().erf(*_arg);
+        }
+    }
+    RCP<const Basic> d;
+    bool b = handle_minus(arg, outArg(d));
+    if (b) {
+        return neg(erf(d));
+    }
+    return make_rcp<const Erf>(d);
+}
+
+bool Erfc::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a<Integer>(*arg) and down_cast<const Integer &>(*arg).is_zero())
+        return false;
+    if (could_extract_minus(*arg))
+        return false;
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> Erfc::create(const RCP<const Basic> &arg) const
+{
+    return erfc(arg);
+}
+
+RCP<const Basic> erfc(const RCP<const Basic> &arg)
+{
+    if (is_a<Integer>(*arg) and down_cast<const Integer &>(*arg).is_zero()) {
+        return one;
+    }
+    if (is_a_Number(*arg)) {
+        RCP<const Number> _arg = rcp_static_cast<const Number>(arg);
+        if (not _arg->is_exact()) {
+            return _arg->get_eval().erfc(*_arg);
+        }
+    }
+
+    RCP<const Basic> d;
+    bool b = handle_minus(arg, outArg(d));
+    if (b) {
+        return add(integer(2), neg(erfc(d)));
+    }
+    return make_rcp<const Erfc>(d);
+}
+
+Gamma::Gamma(const RCP<const Basic> &arg) : OneArgFunction{arg}
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Gamma::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a<Integer>(*arg))
+        return false;
+    if (is_a<Rational>(*arg)
+        and (get_den(down_cast<const Rational &>(*arg).as_rational_class()))
+                == 2) {
+        return false;
+    }
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    return true;
+}
+
+RCP<const Basic> Gamma::create(const RCP<const Basic> &arg) const
+{
+    return gamma(arg);
+}
+
+RCP<const Basic> gamma_positive_int(const RCP<const Basic> &arg)
+{
+    SYMENGINE_ASSERT(is_a<Integer>(*arg))
+    RCP<const Integer> arg_ = rcp_static_cast<const Integer>(arg);
+    SYMENGINE_ASSERT(arg_->is_positive())
+    return factorial((arg_->subint(*one))->as_int());
+}
+
+RCP<const Basic> gamma_multiple_2(const RCP<const Basic> &arg)
+{
+    SYMENGINE_ASSERT(is_a<Rational>(*arg))
+    RCP<const Rational> arg_ = rcp_static_cast<const Rational>(arg);
+    SYMENGINE_ASSERT(get_den(arg_->as_rational_class()) == 2)
+    RCP<const Integer> n, k;
+    RCP<const Number> coeff;
+    n = quotient_f(*(integer(mp_abs(get_num(arg_->as_rational_class())))),
+                   *(integer(get_den(arg_->as_rational_class()))));
+    if (arg_->is_positive()) {
+        k = n;
+        coeff = one;
+    } else {
+        n = n->addint(*one);
+        k = n;
+        if ((n->as_int() & 1) == 0) {
+            coeff = one;
+        } else {
+            coeff = minus_one;
+        }
+    }
+    // lakutata: the double factorial 3*5*...*(2k - 1) in an integer of any size
+    // (an int overflowed from k = 10: gamma(-25/2) was wrong)
+    integer_class j = 1;
+    for (long i = 3; i < 2 * k->as_int(); i = i + 2) {
+        j *= i;
+    }
+    coeff = mulnum(coeff, integer(std::move(j)));
+    if (arg_->is_positive()) {
+        return div(mul(coeff, sqrt(pi)), pow(i2, n));
+    } else {
+        return div(mul(pow(i2, n), sqrt(pi)), coeff);
+    }
+}
+
+RCP<const Basic> gamma(const RCP<const Basic> &arg)
+{
+    if (is_a<Integer>(*arg)) {
+        RCP<const Integer> arg_ = rcp_static_cast<const Integer>(arg);
+        if (arg_->is_positive()) {
+            // lakutata: the factorials above the limit stay as written
+            if (arg_->as_integer_class() > GAMMA_LIMIT)
+                return make_rcp<const Gamma>(arg);
+            return gamma_positive_int(arg);
+        } else {
+            return ComplexInf;
+        }
+    } else if (is_a<Rational>(*arg)) {
+        RCP<const Rational> arg_ = rcp_static_cast<const Rational>(arg);
+        if ((get_den(arg_->as_rational_class())) == 2) {
+            if (mp_abs(get_num(arg_->as_rational_class())) > 2 * GAMMA_LIMIT)
+                return make_rcp<const Gamma>(arg);
+            return gamma_multiple_2(arg);
+        } else {
+            return make_rcp<const Gamma>(arg);
+        }
+    } else if (is_a_Number(*arg)
+               and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().gamma(*arg);
+    }
+    return make_rcp<const Gamma>(arg);
+}
+
+LowerGamma::LowerGamma(const RCP<const Basic> &s, const RCP<const Basic> &x)
+    : TwoArgFunction(s, x)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(s, x))
+}
+
+bool LowerGamma::is_canonical(const RCP<const Basic> &s,
+                              const RCP<const Basic> &x) const
+{
+    // Only special values are evaluated
+    if (eq(*s, *one))
+        return false;
+    if (is_a<Integer>(*s)
+        and down_cast<const Integer &>(*s).as_integer_class() > 1)
+        return false;
+    if (is_a<Integer>(*mul(i2, s)))
+        return false;
+#ifdef HAVE_SYMENGINE_MPFR
+#if MPFR_VERSION_MAJOR > 3
+    if (is_a<RealMPFR>(*s) && is_a<RealMPFR>(*x))
+        return false;
+#endif
+#endif
+    return true;
+}
+
+RCP<const Basic> LowerGamma::create(const RCP<const Basic> &a,
+                                    const RCP<const Basic> &b) const
+{
+    return lowergamma(a, b);
+}
+
+RCP<const Basic> lowergamma(const RCP<const Basic> &s,
+                            const RCP<const Basic> &x)
+{
+    // Only special values are being evaluated
+    if (is_a<Integer>(*s)) {
+        RCP<const Integer> s_int = rcp_static_cast<const Integer>(s);
+        if (s_int->is_one()) {
+            return sub(one, exp(mul(minus_one, x)));
+        } else if (s_int->as_integer_class() > 1) {
+            s_int = s_int->subint(*one);
+            return sub(mul(s_int, lowergamma(s_int, x)),
+                       mul(pow(x, s_int), exp(mul(minus_one, x))));
+        } else {
+            return make_rcp<const LowerGamma>(s, x);
+        }
+    } else if (is_a<Integer>(*(mul(i2, s)))) {
+        RCP<const Number> s_num = rcp_static_cast<const Number>(s);
+        s_num = subnum(s_num, one);
+        if (eq(*s, *div(one, integer(2)))) {
+            return mul(sqrt(pi),
+                       erf(sqrt(x))); // base case for s of the form n/2
+        } else if (s_num->is_positive()) {
+            return sub(mul(s_num, lowergamma(s_num, x)),
+                       mul(pow(x, s_num), exp(mul(minus_one, x))));
+        } else {
+            return div(add(lowergamma(add(s, one), x),
+                           mul(pow(x, s), exp(mul(minus_one, x)))),
+                       s);
+        }
+#ifdef HAVE_SYMENGINE_MPFR
+#if MPFR_VERSION_MAJOR > 3
+    } else if (is_a<RealMPFR>(*s) && is_a<RealMPFR>(*x)) {
+        const auto &s_ = down_cast<const RealMPFR &>(*s).i.get_mpfr_t();
+        const auto &x_ = down_cast<const RealMPFR &>(*x).i.get_mpfr_t();
+        if (mpfr_cmp_si(x_, 0) >= 0) {
+            mpfr_class t(std::max(mpfr_get_prec(s_), mpfr_get_prec(x_)));
+            mpfr_class u(std::max(mpfr_get_prec(s_), mpfr_get_prec(x_)));
+            mpfr_gamma_inc(t.get_mpfr_t(), s_, x_, MPFR_RNDN);
+            mpfr_gamma(u.get_mpfr_t(), s_, MPFR_RNDN);
+            mpfr_sub(t.get_mpfr_t(), u.get_mpfr_t(), t.get_mpfr_t(), MPFR_RNDN);
+            return real_mpfr(std::move(t));
+        } else {
+            throw NotImplementedError("Not implemented.");
+        }
+#endif
+#endif
+    }
+    return make_rcp<const LowerGamma>(s, x);
+}
+
+UpperGamma::UpperGamma(const RCP<const Basic> &s, const RCP<const Basic> &x)
+    : TwoArgFunction(s, x)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(s, x))
+}
+
+bool UpperGamma::is_canonical(const RCP<const Basic> &s,
+                              const RCP<const Basic> &x) const
+{
+    // Only special values are evaluated
+    if (eq(*s, *one))
+        return false;
+    if (is_a<Integer>(*s)
+        and down_cast<const Integer &>(*s).as_integer_class() > 1)
+        return false;
+    if (is_a<Integer>(*mul(i2, s)))
+        return false;
+#ifdef HAVE_SYMENGINE_MPFR
+#if MPFR_VERSION_MAJOR > 3
+    if (is_a<RealMPFR>(*s) && is_a<RealMPFR>(*x))
+        return false;
+#endif
+#endif
+    return true;
+}
+
+RCP<const Basic> UpperGamma::create(const RCP<const Basic> &a,
+                                    const RCP<const Basic> &b) const
+{
+    return uppergamma(a, b);
+}
+
+RCP<const Basic> uppergamma(const RCP<const Basic> &s,
+                            const RCP<const Basic> &x)
+{
+    // Only special values are being evaluated
+    if (is_a<Integer>(*s)) {
+        RCP<const Integer> s_int = rcp_static_cast<const Integer>(s);
+        if (s_int->is_one()) {
+            return exp(mul(minus_one, x));
+        } else if (s_int->as_integer_class() > 1) {
+            s_int = s_int->subint(*one);
+            return add(mul(s_int, uppergamma(s_int, x)),
+                       mul(pow(x, s_int), exp(mul(minus_one, x))));
+        } else {
+            // TODO: implement unpolarfy to handle this case
+            return make_rcp<const UpperGamma>(s, x);
+        }
+    } else if (is_a<Integer>(*(mul(i2, s)))) {
+        RCP<const Number> s_num = rcp_static_cast<const Number>(s);
+        s_num = subnum(s_num, one);
+        if (eq(*s, *div(one, integer(2)))) {
+            return mul(sqrt(pi),
+                       erfc(sqrt(x))); // base case for s of the form n/2
+        } else if (s_num->is_positive()) {
+            return add(mul(s_num, uppergamma(s_num, x)),
+                       mul(pow(x, s_num), exp(mul(minus_one, x))));
+        } else {
+            return div(sub(uppergamma(add(s, one), x),
+                           mul(pow(x, s), exp(mul(minus_one, x)))),
+                       s);
+        }
+#ifdef HAVE_SYMENGINE_MPFR
+#if MPFR_VERSION_MAJOR > 3
+    } else if (is_a<RealMPFR>(*s) && is_a<RealMPFR>(*x)) {
+        const auto &s_ = down_cast<const RealMPFR &>(*s).i.get_mpfr_t();
+        const auto &x_ = down_cast<const RealMPFR &>(*x).i.get_mpfr_t();
+        if (mpfr_cmp_si(x_, 0) >= 0) {
+            mpfr_class t(std::max(mpfr_get_prec(s_), mpfr_get_prec(x_)));
+            mpfr_gamma_inc(t.get_mpfr_t(), s_, x_, MPFR_RNDN);
+            return real_mpfr(std::move(t));
+        } else {
+            throw NotImplementedError("Not implemented.");
+        }
+#endif
+#endif
+    }
+    return make_rcp<const UpperGamma>(s, x);
+}
+
+bool LogGamma::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a<Integer>(*arg)) {
+        RCP<const Integer> arg_int = rcp_static_cast<const Integer>(arg);
+        if (not arg_int->is_positive()) {
+            return false;
+        }
+        if (eq(*integer(1), *arg_int) or eq(*integer(2), *arg_int)
+            or eq(*integer(3), *arg_int)) {
+            return false;
+        }
+    }
+    return true;
+}
+
+RCP<const Basic> LogGamma::rewrite_as_gamma() const
+{
+    return log(gamma(get_arg()));
+}
+
+RCP<const Basic> LogGamma::create(const RCP<const Basic> &arg) const
+{
+    return loggamma(arg);
+}
+
+RCP<const Basic> loggamma(const RCP<const Basic> &arg)
+{
+    if (is_a<Integer>(*arg)) {
+        RCP<const Integer> arg_int = rcp_static_cast<const Integer>(arg);
+        if (not arg_int->is_positive()) {
+            return Inf;
+        }
+        if (eq(*integer(1), *arg_int) or eq(*integer(2), *arg_int)) {
+            return zero;
+        } else if (eq(*integer(3), *arg_int)) {
+            return log(integer(2));
+        }
+    }
+    return make_rcp<const LogGamma>(arg);
+}
+
+RCP<const Beta> Beta::from_two_basic(const RCP<const Basic> &x,
+                                     const RCP<const Basic> &y)
+{
+    if (x->__cmp__(*y) == -1) {
+        return make_rcp<const Beta>(y, x);
+    }
+    return make_rcp<const Beta>(x, y);
+}
+
+bool Beta::is_canonical(const RCP<const Basic> &x, const RCP<const Basic> &y)
+{
+    if (x->__cmp__(*y) == -1) {
+        return false;
+    }
+    if (is_a<Integer>(*x)
+        or (is_a<Rational>(*x)
+            and (get_den(down_cast<const Rational &>(*x).as_rational_class()))
+                    == 2)) {
+        if (is_a<Integer>(*y)
+            or (is_a<Rational>(*y)
+                and (get_den(
+                        down_cast<const Rational &>(*y).as_rational_class()))
+                        == 2)) {
+            return false;
+        }
+    }
+    return true;
+}
+
+RCP<const Basic> Beta::rewrite_as_gamma() const
+{
+    return div(mul(gamma(get_arg1()), gamma(get_arg2())),
+               gamma(add(get_arg1(), get_arg2())));
+}
+
+RCP<const Basic> Beta::create(const RCP<const Basic> &a,
+                              const RCP<const Basic> &b) const
+{
+    return beta(a, b);
+}
+
+RCP<const Basic> beta(const RCP<const Basic> &x, const RCP<const Basic> &y)
+{
+    // Only special values are being evaluated
+    // lakutata: beta(x, 1 - x) is pi/sin(pi*x); SymEngine gave complexInfinity
+    // (the poles are where x or y is a nonpositive integer, below)
+    if (eq(*add(x, y), *one) and is_a<Rational>(*x)) {
+        return div(pi, sin(mul(pi, x)));
+    }
+
+    if (is_a<Integer>(*x)) {
+        RCP<const Integer> x_int = rcp_static_cast<const Integer>(x);
+        if (x_int->is_positive()) {
+            if (is_a<Integer>(*y)) {
+                RCP<const Integer> y_int = rcp_static_cast<const Integer>(y);
+                if (y_int->is_positive()) {
+                    return div(
+                        mul(gamma_positive_int(x), gamma_positive_int(y)),
+                        gamma_positive_int(add(x, y)));
+                } else {
+                    return ComplexInf;
+                }
+            } else if (is_a<Rational>(*y)) {
+                RCP<const Rational> y_ = rcp_static_cast<const Rational>(y);
+                if (get_den(y_->as_rational_class()) == 2) {
+                    return div(mul(gamma_positive_int(x), gamma_multiple_2(y)),
+                               gamma_multiple_2(add(x, y)));
+                } else {
+                    return Beta::from_two_basic(x, y);
+                }
+            }
+        } else {
+            return ComplexInf;
+        }
+    }
+
+    if (is_a<Integer>(*y)) {
+        RCP<const Integer> y_int = rcp_static_cast<const Integer>(y);
+        if (y_int->is_positive()) {
+            if (is_a<Rational>(*x)) {
+                RCP<const Rational> x_ = rcp_static_cast<const Rational>(x);
+                if (get_den(x_->as_rational_class()) == 2) {
+                    return div(mul(gamma_positive_int(y), gamma_multiple_2(x)),
+                               gamma_multiple_2(add(x, y)));
+                } else {
+                    return Beta::from_two_basic(x, y);
+                }
+            }
+        } else {
+            return ComplexInf;
+        }
+    }
+
+    if (is_a<const Rational>(*x)
+        and get_den(down_cast<const Rational &>(*x).as_rational_class()) == 2) {
+        if (is_a<Integer>(*y)) {
+            RCP<const Integer> y_int = rcp_static_cast<const Integer>(y);
+            if (y_int->is_positive()) {
+                return div(mul(gamma_multiple_2(x), gamma_positive_int(y)),
+                           gamma_multiple_2(add(x, y)));
+            } else {
+                return ComplexInf;
+            }
+        }
+        if (is_a<const Rational>(*y)
+            and get_den((down_cast<const Rational &>(*y)).as_rational_class())
+                    == 2) {
+            return div(mul(gamma_multiple_2(x), gamma_multiple_2(y)),
+                       gamma_positive_int(add(x, y)));
+        }
+    }
+    return Beta::from_two_basic(x, y);
+}
+
+// lakutata: whether polygamma(n, x) is moved right by its recurrence: a
+// negative rational x not too far left, an order n nonnegative integer
+static bool polygamma_shifted(const RCP<const Basic> &n,
+                              const RCP<const Basic> &x)
+{
+    if (not is_a<Integer>(*n) or down_cast<const Integer &>(*n).is_negative()
+        or not mp_fits_ulong_p(down_cast<const Integer &>(*n).as_integer_class())
+        or not is_a<Rational>(*x)
+        or not down_cast<const Rational &>(*x).is_negative()) {
+        return false;
+    }
+    return mp_abs(get_num(down_cast<const Rational &>(*x).as_rational_class()))
+           <= 1000 * get_den(down_cast<const Rational &>(*x).as_rational_class());
+}
+
+bool PolyGamma::is_canonical(const RCP<const Basic> &n,
+                             const RCP<const Basic> &x)
+{
+    // lakutata: the nonpositive integers and the negative rationals moved right
+    if ((is_a<Integer>(*x) and not down_cast<const Integer &>(*x).is_positive())
+        or polygamma_shifted(n, x)) {
+        return false;
+    }
+    if (is_a<Integer>(*x)
+        and down_cast<const Integer &>(*x).as_integer_class() > GAMMA_LIMIT) {
+        return true;
+    }
+    if (eq(*n, *zero)) {
+        if (eq(*x, *one)) {
+            return false;
+        }
+        if (is_a<Rational>(*x)) {
+            auto x_ = rcp_static_cast<const Rational>(x);
+            auto den = get_den(x_->as_rational_class());
+            if (den == 2 or den == 3 or den == 4) {
+                return false;
+            }
+        }
+    }
+    return true;
+}
+
+RCP<const Basic> PolyGamma::rewrite_as_zeta() const
+{
+    if (not is_a<Integer>(*get_arg1())) {
+        return rcp_from_this();
+    }
+    RCP<const Integer> n = rcp_static_cast<const Integer>(get_arg1());
+    if (not(n->is_positive())) {
+        return rcp_from_this();
+    }
+    if ((n->as_int() & 1) == 0) {
+        return neg(mul(factorial(n->as_int()), zeta(add(n, one), get_arg2())));
+    } else {
+        return mul(factorial(n->as_int()), zeta(add(n, one), get_arg2()));
+    }
+}
+
+RCP<const Basic> PolyGamma::create(const RCP<const Basic> &a,
+                                   const RCP<const Basic> &b) const
+{
+    return polygamma(a, b);
+}
+
+RCP<const Basic> polygamma(const RCP<const Basic> &n_,
+                           const RCP<const Basic> &x_)
+{
+    // Only special values are being evaluated
+    // lakutata: the poles are the nonpositive integers only (SymEngine gave
+    // complexInfinity for every number not positive); a negative rational is
+    // moved right by psi^(n)(x) = psi^(n)(x + 1) - (-1)^n n!/x^(n+1), the
+    // other numbers (complex ones) are kept for their numeric value
+    if (is_a<Integer>(*x_)
+        and not down_cast<const Integer &>(*x_).is_positive()) {
+        return ComplexInf;
+    }
+    if (polygamma_shifted(n_, x_)) {
+        const unsigned long order
+            = mp_get_ui(down_cast<const Integer &>(*n_).as_integer_class());
+        const RCP<const Basic> coefficient
+            = mul(integer(order % 2 == 0 ? -1 : 1), factorial(order));
+        RCP<const Basic> x = x_, correction = zero;
+        while (down_cast<const Rational &>(*x).is_negative()) {
+            correction = add(correction,
+                             mul(coefficient, pow(x, integer(-long(order) - 1))));
+            x = add(x, one);
+        }
+        return add(polygamma(n_, x), correction);
+    }
+    if (is_a_Number(*x_) and not is_a<Integer>(*x_) and not is_a<Rational>(*x_)) {
+        return make_rcp<const PolyGamma>(n_, x_);
+    }
+    // lakutata: not at the integers above the limit of the exact numbers (the
+    // value of digamma there is a harmonic number of as many terms)
+    if (is_a<Integer>(*x_)
+        and down_cast<const Integer &>(*x_).as_integer_class() > GAMMA_LIMIT) {
+        return make_rcp<const PolyGamma>(n_, x_);
+    }
+    if (is_a<Integer>(*n_) and is_a<Integer>(*x_)) {
+        auto n = down_cast<const Integer &>(*n_).as_int();
+        auto x = down_cast<const Integer &>(*x_).as_int();
+        if (n == 0) {
+            return sub(harmonic(x - 1, 1), EulerGamma);
+        } else if (n % 2 == 1) {
+            return mul(factorial(n), zeta(add(n_, one), x_));
+        }
+    }
+    if (eq(*n_, *zero)) {
+        if (eq(*x_, *one)) {
+            return neg(EulerGamma);
+        }
+        if (is_a<Rational>(*x_)) {
+            RCP<const Rational> x = rcp_static_cast<const Rational>(x_);
+            const auto den = get_den(x->as_rational_class());
+            const auto num = get_num(x->as_rational_class());
+            const integer_class r = num % den;
+            RCP<const Basic> res;
+            if (den == 2) {
+                res = sub(mul(im2, log(i2)), EulerGamma);
+            } else if (den == 3) {
+                // lakutata: by the remainder (psi(4/3) from psi(1/3)), not the
+                // numerator
+                if (r == 1) {
+                    res = add(neg(div(div(pi, i2), sqrt(i3))),
+                              sub(div(mul(im3, log(i3)), i2), EulerGamma));
+                } else {
+                    res = add(div(div(pi, i2), sqrt(i3)),
+                              sub(div(mul(im3, log(i3)), i2), EulerGamma));
+                }
+            } else if (den == 4) {
+                if (r == 1) {
+                    res = add(div(pi, im2), sub(mul(im3, log(i2)), EulerGamma));
+                } else {
+                    res = add(div(pi, i2), sub(mul(im3, log(i2)), EulerGamma));
+                }
+            } else {
+                return make_rcp<const PolyGamma>(n_, x_);
+            }
+            rational_class a(0), f(r, den);
+            for (unsigned long i = 0; i < (num - r) / den; ++i) {
+                a += 1 / (f + i);
+            }
+            return add(Rational::from_mpq(a), res);
+        }
+    }
+    return make_rcp<const PolyGamma>(n_, x_);
+}
+
+RCP<const Basic> digamma(const RCP<const Basic> &x)
+{
+    return polygamma(zero, x);
+}
+
+RCP<const Basic> trigamma(const RCP<const Basic> &x)
+{
+    return polygamma(one, x);
+}
+
+Abs::Abs(const RCP<const Basic> &arg) : OneArgFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool Abs::is_canonical(const RCP<const Basic> &arg) const
+{
+    if (is_a<Integer>(*arg) or is_a<Rational>(*arg) or is_a<Complex>(*arg))
+        return false;
+    if (is_a_Number(*arg) and not down_cast<const Number &>(*arg).is_exact()) {
+        return false;
+    }
+    if (is_a<Abs>(*arg)) {
+        return false;
+    }
+
+    if (could_extract_minus(*arg)) {
+        return false;
+    }
+
+    return true;
+}
+
+RCP<const Basic> Abs::create(const RCP<const Basic> &arg) const
+{
+    return abs(arg);
+}
+
+RCP<const Basic> abs(const RCP<const Basic> &arg)
+{
+    if (is_a<Integer>(*arg)) {
+        RCP<const Integer> arg_ = rcp_static_cast<const Integer>(arg);
+        if (arg_->is_negative()) {
+            return arg_->neg();
+        } else {
+            return arg_;
+        }
+    } else if (is_a<Rational>(*arg)) {
+        RCP<const Rational> arg_ = rcp_static_cast<const Rational>(arg);
+        if (arg_->is_negative()) {
+            return arg_->neg();
+        } else {
+            return arg_;
+        }
+    } else if (is_a<Complex>(*arg)) {
+        RCP<const Complex> arg_ = rcp_static_cast<const Complex>(arg);
+        return sqrt(Rational::from_mpq(arg_->real_ * arg_->real_
+                                       + arg_->imaginary_ * arg_->imaginary_));
+    } else if (is_a_Number(*arg)
+               and not down_cast<const Number &>(*arg).is_exact()) {
+        return down_cast<const Number &>(*arg).get_eval().abs(*arg);
+    }
+    if (is_a<Abs>(*arg)) {
+        return arg;
+    }
+
+    // lakutata: the numeric factor taken out, as SymPy (abs(-2*x) is 2*abs(x))
+    if (is_a<Mul>(*arg)
+        and not down_cast<const Mul &>(*arg).get_coef()->is_one()) {
+        map_basic_basic factors = down_cast<const Mul &>(*arg).get_dict();
+        return mul(abs(down_cast<const Mul &>(*arg).get_coef()),
+                   abs(Mul::from_dict(one, std::move(factors))));
+    }
+    RCP<const Basic> d;
+    // lakutata: abs(-u) is abs(u), evaluated again (abs(-abs(x)) is abs(x))
+    if (handle_minus(arg, outArg(d)) and not eq(*d, *arg))
+        return abs(d);
+    return make_rcp<const Abs>(d);
+}
+
+Max::Max(const vec_basic &&arg) : MultiArgFunction(std::move(arg))
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(get_vec()))
+}
+
+bool Max::is_canonical(const vec_basic &arg) const
+{
+    if (arg.size() < 2)
+        return false;
+
+    bool non_number_exists = false;
+
+    for (const auto &p : arg) {
+        if (is_a<Complex>(*p) or is_a<Max>(*p))
+            return false;
+        if (not is_a_Number(*p))
+            non_number_exists = true;
+    }
+    if (not std::is_sorted(arg.begin(), arg.end(), RCPBasicKeyLess()))
+        return false;
+
+    return non_number_exists; // all arguments cant be numbers
+}
+
+RCP<const Basic> Max::create(const vec_basic &a) const
+{
+    return max(a);
+}
+
+RCP<const Basic> max(const vec_basic &arg)
+{
+    bool number_set = false;
+    RCP<const Number> max_number, difference;
+    set_basic new_args;
+
+    for (const auto &p : arg) {
+        if (is_a<Complex>(*p))
+            throw SymEngineException("Complex can't be passed to max!");
+
+        if (is_a_Number(*p)) {
+            if (not number_set) {
+                max_number = rcp_static_cast<const Number>(p);
+
+            } else {
+                if (eq(*p, *Inf)) {
+                    return Inf;
+                } else if (eq(*p, *NegInf)) {
+                    continue;
+                }
+                difference = down_cast<const Number &>(*p).sub(*max_number);
+
+                if (difference->is_zero() and not difference->is_exact()) {
+                    if (max_number->is_exact())
+                        max_number = rcp_static_cast<const Number>(p);
+                } else if (difference->is_positive()) {
+                    max_number = rcp_static_cast<const Number>(p);
+                }
+            }
+            number_set = true;
+
+        } else if (is_a<Max>(*p)) {
+            for (const auto &l : down_cast<const Max &>(*p).get_args()) {
+                if (is_a_Number(*l)) {
+                    if (not number_set) {
+                        max_number = rcp_static_cast<const Number>(l);
+
+                    } else {
+                        difference = rcp_static_cast<const Number>(l)->sub(
+                            *max_number);
+
+                        if (difference->is_zero()
+                            and not difference->is_exact()) {
+                            if (max_number->is_exact())
+                                max_number = rcp_static_cast<const Number>(l);
+                        } else if (difference->is_positive()) {
+                            max_number = rcp_static_cast<const Number>(l);
+                        }
+                    }
+                    number_set = true;
+                } else {
+                    new_args.insert(l);
+                }
+            }
+        } else {
+            new_args.insert(p);
+        }
+    }
+
+    if (number_set)
+        new_args.insert(max_number);
+
+    vec_basic final_args(new_args.size());
+    std::copy(new_args.begin(), new_args.end(), final_args.begin());
+
+    if (final_args.size() > 1) {
+        return make_rcp<const Max>(std::move(final_args));
+    } else if (final_args.size() == 1) {
+        return final_args[0];
+    } else {
+        throw SymEngineException("Empty vec_basic passed to max!");
+    }
+}
+
+Min::Min(const vec_basic &&arg) : MultiArgFunction(std::move(arg))
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(get_vec()))
+}
+
+bool Min::is_canonical(const vec_basic &arg) const
+{
+    if (arg.size() < 2)
+        return false;
+
+    bool non_number_exists = false;
+
+    for (const auto &p : arg) {
+        if (is_a<Complex>(*p) or is_a<Min>(*p))
+            return false;
+        if (not is_a_Number(*p))
+            non_number_exists = true;
+    }
+    if (not std::is_sorted(arg.begin(), arg.end(), RCPBasicKeyLess()))
+        return false;
+
+    return non_number_exists; // all arguments cant be numbers
+}
+
+RCP<const Basic> Min::create(const vec_basic &a) const
+{
+    return min(a);
+}
+
+RCP<const Basic> min(const vec_basic &arg)
+{
+    bool number_set = false;
+    RCP<const Number> min_number, difference;
+    set_basic new_args;
+
+    for (const auto &p : arg) {
+        if (is_a<Complex>(*p))
+            throw SymEngineException("Complex can't be passed to min!");
+
+        if (is_a_Number(*p)) {
+            if (not number_set) {
+                min_number = rcp_static_cast<const Number>(p);
+
+            } else {
+                if (eq(*p, *Inf)) {
+                    continue;
+                } else if (eq(*p, *NegInf)) {
+                    return NegInf;
+                }
+                difference = min_number->sub(*rcp_static_cast<const Number>(p));
+
+                if (difference->is_zero() and not difference->is_exact()) {
+                    if (min_number->is_exact())
+                        min_number = rcp_static_cast<const Number>(p);
+                } else if (difference->is_positive()) {
+                    min_number = rcp_static_cast<const Number>(p);
+                }
+            }
+            number_set = true;
+
+        } else if (is_a<Min>(*p)) {
+            for (const auto &l : down_cast<const Min &>(*p).get_args()) {
+                if (is_a_Number(*l)) {
+                    if (not number_set) {
+                        min_number = rcp_static_cast<const Number>(l);
+
+                    } else {
+                        difference = min_number->sub(
+                            *rcp_static_cast<const Number>(l));
+
+                        if (difference->is_zero()
+                            and not difference->is_exact()) {
+                            if (min_number->is_exact())
+                                min_number = rcp_static_cast<const Number>(l);
+                        } else if (difference->is_positive()) {
+                            min_number = rcp_static_cast<const Number>(l);
+                        }
+                    }
+                    number_set = true;
+                } else {
+                    new_args.insert(l);
+                }
+            }
+        } else {
+            new_args.insert(p);
+        }
+    }
+
+    if (number_set)
+        new_args.insert(min_number);
+
+    vec_basic final_args(new_args.size());
+    std::copy(new_args.begin(), new_args.end(), final_args.begin());
+
+    if (final_args.size() > 1) {
+        return make_rcp<const Min>(std::move(final_args));
+    } else if (final_args.size() == 1) {
+        return final_args[0];
+    } else {
+        throw SymEngineException("Empty vec_basic passed to min!");
+    }
+}
+
+UnevaluatedExpr::UnevaluatedExpr(const RCP<const Basic> &arg)
+    : OneArgFunction(arg)
+{
+    SYMENGINE_ASSIGN_TYPEID()
+    SYMENGINE_ASSERT(is_canonical(arg))
+}
+
+bool UnevaluatedExpr::is_canonical(const RCP<const Basic> &arg) const
+{
+    return true;
+}
+
+RCP<const Basic> UnevaluatedExpr::create(const RCP<const Basic> &arg) const
+{
+    return make_rcp<const UnevaluatedExpr>(arg);
+}
+
+RCP<const Basic> unevaluated_expr(const RCP<const Basic> &arg)
+{
+    return make_rcp<const UnevaluatedExpr>(arg);
+}
+
+} // namespace SymEngine