Expressions

Expression Template

template <typename T, typename Derived, typename DerivativeType = T>
struct Expression

Represents a mathematical expression in a type. Active data types, such as AReal and FReal, as well as all mathematical expressions inherit from this class. All mathematical operations are defined on this type, rather than any specific derived class.

The derived classes are typically created transparently to the user. Expressions may hold their operands by reference: evaluate an expression within the full expression that creates it, and do not keep one past it (e.g. bound to auto); assign it to an active type instead.

Note that this class uses the CRTP pattern, where Derived is the derived class itself, so that static polymorphism can be used.

All global arithmetic operations defined in C++ are specialized for Expression, so that double or float can be replaced seamlessly with an active data type from XAD. This also includes comparisons.

Expression Traits

XAD also defines expression traits to find information about expressions in a templated context. This is typically only needed when custom functions dealing with the XAD expressions are added.

Direction enum

This enum indicates the direction of algorithmic differentiation associated with a type.

enum Direction {
    DIR_NONE,       // Not an algorithmic differentiation type
    DIR_FORWARD,    // Forward mode AD type
    DIR_REVERSE     // Reverse mode AD type
};

ExprTraits

This is the main traits class to get information on an AD type:

template <typename T>
struct ExprTraits {
    static const bool isExpr;      // true if an expression of XAD active type
    static const int numVariables; // Number of variables in an expression
    static const bool isForward;   // true if forward-mode AD
    static const bool isReverse;   // true if reverse-mode AD
    static const bool isLiteral;   // true if it's an elementary XAD active type
                                   // and not an expression
    static const Direction direction;  // direction of the expression or type
    static const std::size_t vector_size; // number of derivatives (N)
    typedef ... nested_type; // underlying type of the expression
                             // e.g. double for AReal<double>
    typedef ... value_type;  // the base active type of a more
                             // complex expression template
    typedef ... scalar_type; // Type when unwrapping a higher order
                             // expression, e.g. FReal<double> for
                             // an expression of AReal<FReal<double>>
};