Skip to content
GitLab
Explore
Sign in
Primary navigation
Search or go to…
Project
SWIFTsim
Manage
Activity
Members
Labels
Plan
Issues
Issue boards
Milestones
Wiki
Code
Merge requests
Repository
Branches
Commits
Tags
Repository graph
Compare revisions
Snippets
Deploy
Releases
Model registry
Monitor
Incidents
Analyze
Value stream analytics
Contributor analytics
Repository analytics
Model experiments
Help
Help
Support
GitLab documentation
Compare GitLab plans
Community forum
Contribute to GitLab
Provide feedback
Keyboard shortcuts
?
Snippets
Groups
Projects
Show more breadcrumbs
SWIFT
SWIFTsim
Commits
a651fa11
Commit
a651fa11
authored
5 years ago
by
Matthieu Schaller
Browse files
Options
Downloads
Patches
Plain Diff
Implemented the new updated MAC
parent
69f724ab
No related branches found
No related tags found
1 merge request
!1077
Improved multipole acceptance criterion (MAC)
Changes
3
Hide whitespace changes
Inline
Side-by-side
Showing
3 changed files
src/gravity_properties.c
+1
-2
1 addition, 2 deletions
src/gravity_properties.c
src/gravity_properties.h
+6
-10
6 additions, 10 deletions
src/gravity_properties.h
src/multipole_accept.h
+136
-43
136 additions, 43 deletions
src/multipole_accept.h
with
143 additions
and
55 deletions
src/gravity_properties.c
+
1
−
2
View file @
a651fa11
...
@@ -32,6 +32,7 @@
...
@@ -32,6 +32,7 @@
#include
"gravity.h"
#include
"gravity.h"
#include
"kernel_gravity.h"
#include
"kernel_gravity.h"
#include
"kernel_long_gravity.h"
#include
"kernel_long_gravity.h"
#include
"restart.h"
#define gravity_props_default_a_smooth 1.25f
#define gravity_props_default_a_smooth 1.25f
#define gravity_props_default_r_cut_max 4.5f
#define gravity_props_default_r_cut_max 4.5f
...
@@ -86,8 +87,6 @@ void gravity_props_init(struct gravity_props *p, struct swift_params *params,
...
@@ -86,8 +87,6 @@ void gravity_props_init(struct gravity_props *p, struct swift_params *params,
/* Opening angle */
/* Opening angle */
p
->
theta_crit
=
parser_get_param_double
(
params
,
"Gravity:theta"
);
p
->
theta_crit
=
parser_get_param_double
(
params
,
"Gravity:theta"
);
if
(
p
->
theta_crit
>=
1
.)
error
(
"Theta too large. FMM won't converge."
);
if
(
p
->
theta_crit
>=
1
.)
error
(
"Theta too large. FMM won't converge."
);
p
->
theta_crit2
=
p
->
theta_crit
*
p
->
theta_crit
;
p
->
theta_crit_inv
=
1
.
/
p
->
theta_crit
;
/* Mesh dithering */
/* Mesh dithering */
if
(
periodic
&&
!
with_external_potential
)
{
if
(
periodic
&&
!
with_external_potential
)
{
...
...
This diff is collapsed.
Click to expand it.
src/gravity_properties.h
+
6
−
10
View file @
a651fa11
...
@@ -26,10 +26,6 @@
...
@@ -26,10 +26,6 @@
#include
<hdf5.h>
#include
<hdf5.h>
#endif
#endif
/* Local includes. */
#include
"kernel_gravity.h"
#include
"restart.h"
/* Forward declarations */
/* Forward declarations */
struct
cosmology
;
struct
cosmology
;
struct
phys_const
;
struct
phys_const
;
...
@@ -58,14 +54,14 @@ struct gravity_props {
...
@@ -58,14 +54,14 @@ struct gravity_props {
/* -------------- Properties of the FFM gravity ---------------------- */
/* -------------- Properties of the FFM gravity ---------------------- */
/*!
T
re
e opening angle (Multipole acceptance criterion)
*/
/*!
A
re
we using the adaptive opening angle?
*/
double
theta_crit
;
int
use_adaptive_tolerance
;
/*!
Square of opening angle
*/
/*!
Accuracy parameter of the advanced MAC
*/
double
theta_crit2
;
float
adaptive_tolerance
;
/*!
Inverse of
opening angle */
/*!
Tree
opening angle
(Multipole acceptance criterion)
*/
double
theta_crit
_inv
;
double
theta_crit
;
/* ------------- Properties of the softened gravity ------------------ */
/* ------------- Properties of the softened gravity ------------------ */
...
...
This diff is collapsed.
Click to expand it.
src/multipole_accept.h
+
136
−
43
View file @
a651fa11
...
@@ -23,67 +23,160 @@
...
@@ -23,67 +23,160 @@
#include
"../config.h"
#include
"../config.h"
/* Local includes */
/* Local includes */
#include
"binomial.h"
#include
"integer_power.h"
#include
"minmax.h"
#include
"multipole_struct.h"
#include
"multipole_struct.h"
/**
/**
* @brief Checks whether
a cell-cell interaction can be appromixated by a M-M
* @brief Checks whether
The multipole in B can be used to update the field
*
interaction using the distance and cell radius
.
*
tensor in A
.
*
*
* We use the multipole acceptance criterion of Dehnen, 2002, JCoPh, Volume 179,
* We use the MAC of Dehnen 2014 eq. 16.
* Issue 1, pp.27-42, equation 10.
*
*
* We also additionally check that the distance between the multipoles
* Note: this is *not* symmetric in A<->B unless the purely geometric criterion
* is larger than the softening lengths (here the distance at which
* is used.
* the gravity becomes Newtonian again, not the Plummer-equivalent quantity).
*
*
* @param r_crit_a The size of the multipole A.
* @param props The properties of the gravity scheme.
* @param r_crit_b The size of the multipole B.
* @param A The gravity tensors that we want to update (sink).
* @param theta_crit2 The square of the critical opening angle.
* @param B The gravity tensors that act as a source.
* @param r2 Square of the distance (periodically wrapped) between the
* @param r2 The square of the distance between the centres of mass of A and B.
* multipoles.
* @param epsilon_a The maximal softening length of any particle in A.
* @param epsilon_b The maximal softening length of any particle in B.
*/
*/
__attribute__
((
always_inline
,
const
))
INLINE
static
int
gravity_M2L_accept
(
__attribute__
((
nonnull
,
pure
))
INLINE
static
int
gravity_M2L_accept
(
const
double
r_crit_a
,
const
double
r_crit_b
,
const
double
theta_crit2
,
const
struct
gravity_props
*
props
,
const
struct
gravity_tensors
*
restrict
A
,
const
double
r2
,
const
double
epsilon_a
,
const
double
epsilon_b
)
{
const
struct
gravity_tensors
*
restrict
B
,
const
float
r2
)
{
const
double
size
=
r_crit_a
+
r_crit_b
;
/* Order of the expansion */
const
double
size2
=
size
*
size
;
const
int
p
=
SELF_GRAVITY_MULTIPOLE_ORDER
;
const
double
epsilon_a2
=
epsilon_a
*
epsilon_a
;
const
double
epsilon_b2
=
epsilon_b
*
epsilon_b
;
// MATTHIEU: Make this mass-dependent ?
/* Compute the error estimator (without the 1/M_B term that cancels out) */
float
E_BA_term
=
0
.
f
;
for
(
int
n
=
0
;
n
<=
p
;
++
n
)
{
E_BA_term
+=
binomial
(
p
,
n
)
*
B
->
m_pole
.
power
[
n
]
*
integer_powf
(
A
->
r_max
,
p
-
n
);
}
E_BA_term
*=
8
.
f
;
E_BA_term
*=
max
(
A
->
r_max
,
B
->
r_max
);
E_BA_term
/=
(
A
->
r_max
+
B
->
r_max
);
/* Multipole acceptance criterion (Dehnen 2002, eq.10) */
/* Compute r^(p+2) */
return
(
r2
*
theta_crit2
>
size2
)
&&
(
r2
>
epsilon_a2
)
&&
(
r2
>
epsilon_b2
);
#if SELF_GRAVITY_MULTIPOLE_ORDER % 2 == 1
const
float
r_to_p_plus2
=
integer_powf
(
sqrtf
(
r2
),
(
p
+
2
));
#else
const
float
r_to_p_plus2
=
integer_powf
(
r2
,
((
p
/
2
)
+
1
));
#endif
/* Get the mimimal acceleration in A */
const
float
min_a_grav
=
A
->
m_pole
.
min_old_a_grav_norm
;
/* Get the maximal softening length in B */
const
float
max_softening
=
B
->
m_pole
.
max_softening
;
/* Get the relative tolerance */
const
float
eps
=
props
->
adaptive_tolerance
;
/* Get the basic geometric critical angle */
const
float
theta_crit
=
props
->
theta_crit
;
const
float
theta_crit2
=
theta_crit
*
theta_crit
;
/* Get the sum of the multipole sizes */
const
float
rho_sum
=
A
->
r_max
+
B
->
r_max
;
if
(
props
->
use_adaptive_tolerance
)
{
/* Test the different conditions */
/* Condition 1: We are in the converging part of the Taylor expansion */
const
int
cond_1
=
rho_sum
*
rho_sum
<
r2
;
/* Condition 2: We are not below softening */
const
int
cond_2
=
max_softening
*
max_softening
<
r2
;
/* Condition 3: The contribution is accurate enough
* (E_BA / r^(p+2) < eps a_min) */
const
int
cond_3
=
E_BA_term
<
eps
*
min_a_grav
*
r_to_p_plus2
;
return
cond_1
&&
cond_2
&&
cond_3
;
}
else
{
/* Condition 1: We are in the converging part of the Taylor expansion */
const
int
cond_1
=
rho_sum
*
rho_sum
<
theta_crit2
*
r2
;
/* Condition 2: We are not below softening */
const
int
cond_2
=
max_softening
*
max_softening
<
r2
;
return
cond_1
&&
cond_2
;
}
}
}
/**
/**
* @brief Checks whether a particle-cell interaction can be appromixated by a
* @brief Checks whether The multipole in B can be used to update the particle
* M2P interaction using the distance and cell radius.
* pa
*
* We use the multipole acceptance criterion of Dehnen, 2002, JCoPh, Volume 179,
* Issue 1, pp.27-42, equation 10.
*
*
* We also additionally check that the distance between the particle and the
* We use the MAC of Dehnen 2014 eq. 16.
* multipole is larger than the softening length (here the distance at which
* the gravity becomes Newtonian again, not the Plummer-equivalent quantity).
*
*
* @param r_max2 The square of the size of the multipole.
* @param props The properties of the gravity scheme.
* @param theta_crit2 The square of the critical opening angle.
* @param pa The particle we want to compute forces for (sink)
* @param r2 Square of the distance (periodically wrapped) between the
* @param B The gravity tensors that act as a source.
* particle and the multipole.
* @param r2 The square of the distance between pa and the centres of mass of B.
* @param epsilon The softening length of the particle.
*/
*/
__attribute__
((
always_inline
,
const
))
INLINE
static
int
gravity_M2P_accept
(
__attribute__
((
nonnull
,
pure
))
INLINE
static
int
gravity_M2P_accept
(
const
float
r_max2
,
const
float
theta_crit2
,
const
float
r2
,
const
struct
gravity_props
*
props
,
const
struct
gpart
*
pa
,
const
float
epsilon
)
{
const
struct
gravity_tensors
*
B
,
const
float
r2
)
{
/* Order of the expansion */
const
int
p
=
SELF_GRAVITY_MULTIPOLE_ORDER
;
/* Compute the error estimator (without the 1/M_B term that cancels out) */
float
E_BA_term
=
8
.
f
*
B
->
m_pole
.
power
[
p
];
/* Compute r^(p+2) */
#if SELF_GRAVITY_MULTIPOLE_ORDER % 2 == 1
const
float
r_to_p_plus2
=
integer_powf
(
sqrtf
(
r2
),
(
p
+
2
));
#else
const
float
r_to_p_plus2
=
integer_powf
(
r2
,
((
p
/
2
)
+
1
));
#endif
/* Get the estimate of the acceleration */
const
float
old_a_grav
=
pa
->
old_a_grav_norm
;
/* Get the maximal softening length in B */
const
float
max_softening
=
B
->
m_pole
.
max_softening
;
/* Get the relative tolerance */
const
float
eps
=
props
->
adaptive_tolerance
;
/* Get the basic geometric critical angle */
const
float
theta_crit
=
props
->
theta_crit
;
const
float
theta_crit2
=
theta_crit
*
theta_crit
;
if
(
props
->
use_adaptive_tolerance
)
{
/* Test the different conditions */
/* Condition 1: We are in the converging part of the Taylor expansion */
const
int
cond_1
=
(
B
->
r_max
)
*
(
B
->
r_max
)
<
r2
;
/* Condition 2: We are not below softening */
const
int
cond_2
=
max_softening
*
max_softening
<
r2
;
/* Condition 3: The contribution is accurate enough
* (E_BA / r^(p+2) < eps * a) */
const
int
cond_3
=
E_BA_term
<
eps
*
old_a_grav
*
r_to_p_plus2
;
return
cond_1
&&
cond_2
&&
cond_3
;
}
else
{
/* Condition 1: We are in the converging part of the Taylor expansion */
const
int
cond_1
=
(
B
->
r_max
)
*
(
B
->
r_max
)
<
theta_crit2
*
r2
;
// MATTHIEU: Make this mass-dependent ?
/* Condition 2: We are not below softening */
const
int
cond_2
=
max_softening
*
max_softening
<
r2
;
/* Multipole acceptance criterion (Dehnen 2002, eq.10) */
return
cond_1
&&
cond_2
;
return
(
r2
*
theta_crit2
>
r_max2
)
&&
(
r2
>
epsilon
*
epsilon
);
}
}
}
#endif
/* SWIFT_MULTIPOLE_ACCEPT_H */
#endif
/* SWIFT_MULTIPOLE_ACCEPT_H */
This diff is collapsed.
Click to expand it.
Preview
0%
Loading
Try again
or
attach a new file
.
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Save comment
Cancel
Please
register
or
sign in
to comment