Precision

Definitions of data types used in the MD simulations. More...

Macros
#define	TINKER_DETERMINISTIC_FORCE   1
#define	TINKER_REAL_SIZE   4
#define	TINKER_MIXED_SIZE   8
#define	TINKER_DOUBLE_PRECISION   0
#define	TINKER_MIXED_PRECISION   1
#define	TINKER_SINGLE_PRECISION   0

Typedefs
typedef unsigned long long	tinker::fixed
typedef float	tinker::real
typedef double	tinker::mixed
typedef mixed	tinker::time_prec
	Floating-point type for time. More...
typedef mixed	tinker::T_prec
	Floating-point type for temperature. More...
typedef mixed	tinker::vel_prec
	Floating-point type for velocities. More...
typedef mixed	tinker::pos_prec
	Floating-point type for coordinates. More...
typedef real	tinker::e_prec
	Floating-point type for the pairwise energy components. More...
typedef real	tinker::v_prec
	Floating-point type for the pairwise virial components. More...
typedef real	tinker::g_prec
	Floating-point type for the pairwise gradient components. More...
typedef mixed	tinker::energy_prec
	Floating-point type for total energies. More...
typedef mixed	tinker::virial_prec
typedef fixed	tinker::grad_prec

Detailed Description

Definitions of data types used in the MD simulations.

Macro Definition Documentation

TINKER_DETERMINISTIC_FORCE

#define TINKER_DETERMINISTIC_FORCE   1

Logical macro for the underlying type of energy gradients.

• If true, always uses fixed-point arithmetic to accumulate energy gradients, regardless of the underlying type of each individual component.
• If false, use type real, which can either be float or double based on the precision macro.

See also

TINKER_DOUBLE_PRECISION

TINKER_MIXED_PRECISION

TINKER_SINGLE_PRECISION

In general, evaluating energy, forces, etc. twice, we don't expect to get two identical answers, but we may not care as much because the difference is usually negligible. [See Why is cos(x) != cos(y)?] Whereas in MD, two simulations with the same initial configurations can easily diverge due to the accumulated difference. If, for whatever reason, you are willing to elongate the process of the inevitable divergence at the cost of slightly slower simulation speed, a more "deterministic" force (using fixed-point arithmetic) can help.

To accumulate floating-point values via fixed-point arithmetic, we first scale the values by a large integer then sum only the integer part.

// fixed sum = 0;
// FLT is float or double
FLT val2 = val * 0x100000000ull;
sum += (fixed)((long long)val2);

To get the floating-point sum, we have to cast the integer sum back to floating-point number, then divide it by the same large integer.

FLT val2 = (FLT)((long long)sum);
FLT answer = val2 / 0x100000000ull;

TINKER_DOUBLE_PRECISION

#define TINKER_DOUBLE_PRECISION   0

Macro for the precision mode. Types real and mixed have different definitions in different modes.

Macros	real	mixed
DOUBLE	double	double
MIXED	float	double
SINGLE	float	float

Only one of the precision macros can be set to 1 and the others will be set to 0.

TINKER_MIXED_PRECISION

#define TINKER_MIXED_PRECISION   1

Macro for the precision mode. Types real and mixed have different definitions in different modes.

Macros	real	mixed
DOUBLE	double	double
MIXED	float	double
SINGLE	float	float

Only one of the precision macros can be set to 1 and the others will be set to 0.

TINKER_MIXED_SIZE

#define TINKER_MIXED_SIZE   8

Number of bytes of the type mixed.

TINKER_REAL_SIZE

#define TINKER_REAL_SIZE   4

Number of bytes of the type real.

TINKER_SINGLE_PRECISION

#define TINKER_SINGLE_PRECISION   0

Macro for the precision mode. Types real and mixed have different definitions in different modes.

Macros	real	mixed
DOUBLE	double	double
MIXED	float	double
SINGLE	float	float

Only one of the precision macros can be set to 1 and the others will be set to 0.

Typedef Documentation

e_prec

typedef real tinker::e_prec

Floating-point type for the pairwise energy components.

energy_prec

typedef mixed tinker::energy_prec

Floating-point type for total energies.

fixed

tinker::fixed

64-bit unsigned integer type for fixed-point arithmetic.

g_prec

typedef real tinker::g_prec

Floating-point type for the pairwise gradient components.

grad_prec

tinker::grad_prec

Floating-point or fixed-point type for the total gradients.

See also

TINKER_DETERMINISTIC_FORCE

mixed

tinker::mixed

Floating-point type with higher precision (not lower than real).

See also

TINKER_MIXED_PRECISION

pos_prec

typedef mixed tinker::pos_prec

Floating-point type for coordinates.

real

tinker::real

Floating-point type with lower precision (not higher than mixed).

See also

TINKER_MIXED_PRECISION

T_prec

typedef mixed tinker::T_prec

Floating-point type for temperature.

time_prec

typedef mixed tinker::time_prec

Floating-point type for time.

v_prec

typedef real tinker::v_prec

Floating-point type for the pairwise virial components.

vel_prec

typedef mixed tinker::vel_prec

Floating-point type for velocities.

virial_prec

typedef mixed tinker::virial_prec

Floating-point type for total virials.