Hello, I have written a small set of templates to represent a 1-variable real mathematical function, with the help of "expression templates", mentioned in blitz++ for arrays, and a simple metafunction to calculate the symbolic derivative. I feel I have reinvented the wheel but I don't know which libs from boost to reuse (MPL, function, bind...), or POOMA (the expr templates part of it). I have written all the templates as purely compile-time objects with all the function data stored in types. The main drawback is the inability to use floating-point literals. The objective is to store in compile-time arbitrary 1-var real functions like: f(x) = x^2 + sqrt(x - 3) and then to have a metafunction to calculate f'(x), the derivative. (Partials derivatives for later)... Here is what I came up with. Not sure my nomenclature is correct, I came across the terms in misc lib docs: The parse tree is: A "Function" can be a "Terminal", an "Application Expression", a "Unary expression" or a "Binary Expression" (The interface is inadequate because the user instantiates a new type (a new symbolic function) very verbosely). A "Terminal" can be a "Literal", a "Variable" or a "Elementary function". A "Literal" is a class template whose template argument is an int (drawback: cannot store floating points this way, and I should have ). A "Variable" just represents "x" in the usual notation "f(x)" A "Elementary Function" is a wrapper around <cmath>'s standard and perhaps boost::math's special 1-variable functions, like sin, cos, exp, .... "Application Expression" is to represent sin ( x - 2 ), sort of like callable type. "Unary Expression" is to represent op R (I have 2 so far: +R or -R) "Binary Expression" is to represent L op R (L / R) Op is an "Operator" (plus, minus, multiplies, divides, power)... As I said, I feel this must have been done already? Is there a library I can reuse? Is there any way to represent floating points while staying totally compile-time? Below is the code with an example call: Best regards, ---------------------------------------------------------------------------- ------- #include <iostream> #include <typeinfo> #include <cmath> #include #include namespace MathFunction { template <typename T> struct FunctionTag { typedef T function; }; // Terminals template <typename T> struct TerminalTag : FunctionTag< T > { typedef T terminal; }; template <int N> struct Literal : TerminalTag< Literal<N> > { typedef Literal<0> derivative; static double eval(double variable) { return static_cast<double>(N); } }; struct Variable : TerminalTag<Variable> { typedef Literal<1> derivative; static double eval(double variable) { return variable; } }; template struct UnaryExpression; template struct BinaryExpression; struct plus; struct minus; struct multiplies; struct divides; struct power; // Elementary functions template <typename T> struct ElemFunctionTag : FunctionTag< T > { typedef T elementaryfunction; }; struct Cos; struct Sin : ElemFunctionTag<Sin> { typedef Cos derivative; static double eval(double variable) { return std::sin(variable); } }; struct Cos : ElemFunctionTag<Cos> { typedef UnaryExpression derivative; }; struct Tan : ElemFunctionTag<Tan> { }; struct ATan : ElemFunctionTag<ATan> {}; struct Exp : ElemFunctionTag<Exp> {}; // Operators template <typename T> struct UnaryOperatorTag { typedef T unaryoper; }; template <typename T> struct BinaryOperatorTag { typedef T binaryoper; }; struct plus : UnaryOperatorTag<plus>, BinaryOperatorTag<plus> { static double eval(double var) { return var; } static double eval(double Lvar, double Rvar) { return Lvar + Rvar; } template struct derivative { typedef BinaryExpression type; }; }; struct minus : UnaryOperatorTag<minus>, BinaryOperatorTag<minus> { static double eval(double var) { return -var; } static double eval(double Lvar, double Rvar) { return Lvar - Rvar; } template struct derivative { typedef BinaryExpression type; }; }; struct multiplies : BinaryOperatorTag<multiplies> { static double eval(double Lvar, double Rvar) { return Lvar * Rvar; } template struct derivative { typedef BinaryExpression left; typedef BinaryExpression right; typedef BinaryExpression type; }; }; struct divides : BinaryOperatorTag<divides> { static double eval(double Lvar, double Rvar) { return Lvar/Rvar; } template struct derivative { typedef BinaryExpression left; typedef BinaryExpression right; typedef BinaryExpression > bottom; typedef BinaryExpression top; typedef BinaryExpression type; }; }; struct power : BinaryOperatorTag<power> { static double eval(double Lvar, double Rvar) { return std::pow(Lvar,Rvar); } template struct derivative { typedef BinaryExpression > Rminus1; typedef BinaryExpression right; typedef BinaryExpression type; }; }; struct root : BinaryOperatorTag<root> { static double eval(double Lvar, double Rvar) { return std::pow(Rvar,1.0/Lvar); } }; // Expression template struct UnaryExpression : FunctionTag< UnaryExpression > { BOOST_STATIC_ASSERT(( boost::is_sameOpTag::unaryoper,OpTag::value )); BOOST_STATIC_ASSERT(( boost::is_same::value )); typedef UnaryExpression derivative; static double eval(double variable) { return OpTag::eval(R::eval(variable)); } }; template struct BinaryExpression : FunctionTag< BinaryExpression > { BOOST_STATIC_ASSERT(( boost::is_same::value )); BOOST_STATIC_ASSERT(( boost::is_sameOpTag::binaryoper,OpTag::value )); BOOST_STATIC_ASSERT(( boost::is_same::value )); typedef OpTag::derivative::type derivative; static double eval(double variable) { return OpTag::eval(L::eval(variable), R::eval(variable)); } }; template struct ApplicationExpression : FunctionTag< ApplicationExpression > { BOOST_STATIC_ASSERT(( boost::is_same::value )); BOOST_STATIC_ASSERT(( boost::is_same::value )); typedef BinaryExpression< ApplicationExpression, multiplies, R::derivative > derivative; static double eval(double variable) { return L::eval( R::eval(variable) ); } }; template struct Derivative { BOOST_STATIC_ASSERT(( n>0 )); typedef Derivative::type type; }; template <typename F> struct Derivative { typedef F::derivative type; }; // Main interface template <typename K> struct Function { BOOST_STATIC_ASSERT(( boost::is_same::value )); typedef Function derivative; static double eval(double variable) { return K::eval(variable); } }; template <int N> std::ostream& operator<<( std::ostream& os, const Literal<N>& ) { os << N; return os; } std::ostream& operator<<( std::ostream& os, const Variable& ) { os << 'x'; return os; } template <typename T> std::ostream& operator<<( std::ostream& os, const ElemFunctionTag<T>& ) { os << T(); return os; } std::ostream& operator<<( std::ostream& os, const Sin& ) { os << "sin x"; return os; } std::ostream& operator<<( std::ostream& os, const Cos& ) { os << "cos x"; return os; } template std::ostream& operator<<( std::ostream& os, const ApplicationExpression& ) { os << L() << '(' << R() << ')'; return os; } template std::ostream& operator<<( std::ostream& os, const UnaryExpression& ) { os << OpTag() << R(); return os; } template std::ostream& operator<<( std::ostream& os, const BinaryExpression& ) { os << '(' << L() << ' '<< OpTag() << ' ' << R() << ')'; return os; } std::ostream& operator<<( std::ostream& os, const plus& ) { os << '+'; return os; } std::ostream& operator<<( std::ostream& os, const minus& ) { os << '-'; return os; } std::ostream& operator<<( std::ostream& os, const multiplies& ) { os << '*'; return os; } std::ostream& operator<<( std::ostream& os, const divides& ) { os << '/'; return os; } std::ostream& operator<<( std::ostream& os, const power& ) { os << '^'; return os; } std::ostream& operator<<( std::ostream& os, const root& ) { os << 'V'; return os; } template <typename K> std::ostream& operator<<( std::ostream& os, const Function<K>& ) { os << "f(x) ="<< K(); return os; } } int main(int argc, char* argv[]) { using namespace MathFunction; typedef Function< Sin > f; typedef Derivative::type fprime; std::cout<< f() <

...

I have written a small set of templates to represent a 1-variable real mathematical function, with the help of "expression templates", mentioned in blitz++ for arrays, and a simple metafunction to calculate the symbolic derivative. The shortest path could be to use Boost::Proto to leverage your expression templates implementation. Once your abstract syntax tree is built through compile time evaluation, a proto transformation of the AST can take care of building the symbolic derivative. Boost::proto is available at the vault IIRC. Thank you. I will read the docs. It seems Boost::proto will do the derivation for me, while I still have to write the AST myself. This AST, isn't there already some boost implementation of it somewhere? After all, writing symbolic mathematical functions would seem to me something many users require.

...

Is there any way to represent floating points while staying totally compile-time? Is your derivative compution NEEDED to be performaed at compile-time ? AS I envision it, best way should be to have your ET code generate a new AST representing the derivative, build a function object out of it and makign it callable on runtime value by the user. Performance is first factor for the code I am writing, so I am trying to run as much as possible at compile-time, as a matter of principle. But I guess making a function object out of the AST will be cheap. I will now move the literals from the compile-time to the run-time world, and will get all numeric types then.

Regards,