There is another long-term precession model I found. Vondrák's New precession expressions, valid for long time intervals (2011) describes this. He includes code samples, depictions, and thoughts on the process that make this more readily available to me as a physics layman to comprehend. This method is not as long-term as Owen's, it only covers several hundreds of millennia (haha). One of the research goals is to
find relatively simple expressions of precession parameters, with accuracy comparable to the IAU 2006 model near the epoch J2000.0, and lower accuracy outside the interval ±1 millennium (up to several minutes of arc at the extreme epochs ±200 millennia).
See also:
I've translated the Fortran code by Vondrák into Python:
Here we present four computer subroutines that implement the long-term precession model. The chosen parameterization is Pₐ,Qₐ for the ecliptic pole and Xₐ,Yₐ for the equator pole. These results enable the precession matrix P to be computed. Two versions of the precession matrix are presented, the first for use with J2000.0 mean place and the second for use with Geocentric Celestial Reference System (GCRS) coordinates. Fortran code is used; implementations in other languages would be along very similar lines. The time argument for all four subroutines is Julian epoch (TT).
%pylab inline
# 2Pi
tau = 6.283185307179586476925287e0
# Arcseconds to radians
as2r = 4.848136811095359935899141e-6
# Obliquity at J2000.0 (radians)
eps0 = 84381.406 * as2r
Populating the interactive namespace from numpy and matplotlib
The Fortran subroutine ltp_PECL generates the unit vector for the pole of the ecliptic, using the series for PA,QA (Eq. (8) and Table 1)
def ltp_PECL(epj):
'''Long-term precession of the ecliptic'''
# There is a typographical error in the original coefficient
# C₇ for Qₐ should read 198.296701 instead of 198.296071
# Src: http://www.aanda.org/articles/aa/abs/2012/05/aa17274e-11/aa17274e-11.html
# Number of polynomial and periodic coefficients
npol = 4
nper = 8
# Polynomial coefficients
pqpol = array([
[+5851.607687,-1600.886300],
[-0.1189000,+1.1689818],
[-0.00028913,-0.00000020],
[+0.000000101,-0.000000437],
])
# Periodic coefficients
pqper = array([
[708.15, -5486.751211, -684.661560, 667.666730, -5523.863691],
[2309.00, -17.127623, 2446.283880, -2354.886252, -549.747450],
[1620.00, -617.517403, 399.671049, -428.152441, -310.998056],
[492.20, 413.442940, -356.652376, 376.202861, 421.535876],
[1183.00, 78.614193, -186.387003, 184.778874, -36.776172],
[622.00, -180.732815, -316.800070, 335.321713, -145.278396],
[882.00, -87.676083, 198.296701, -185.138669, -34.744450], # corrected coefficient per erratum
[547.00, 46.140315, 101.135679, -120.972830, 22.885731],
])
# Centuries since J2000
T = (epj-2000.0)/100.0
# Initialize Pₐ and Qₐ accumulators
P = 0.0
Q = 0.0
# Periodic Terms
for i in range(0,nper):
W = tau*T
A = W/pqper[i][0]
S = sin(A)
C = cos(A)
P = P + C*pqper[i][1] + S*pqper[i][3]
Q = Q + C*pqper[i][2] + S*pqper[i][4]
# Polynomial Terms
W = 1.0
for i in range(0,npol):
P = P + pqpol[i][0]*W
Q = Q + pqpol[i][1]*W
W = W*T
# Pₐ and Qₐ (radians)
P = P*as2r
Q = Q*as2r
# Form the ecliptic pole vector
Z = sqrt(max(1.0-P*P-Q*Q,0.0))
S = sin(eps0)
C = cos(eps0)
vec0 = P
vec1 = -Q*C - Z*S
vec2 = -Q*S + Z*C
vec = array([vec0,vec1,vec2])
return vec
The Fortran subroutine ltp_PEQU generates the unit vector for the pole of the equator, using the series for XA,YA (Eq. (9) and Table 2)
def ltp_PEQU(epj):
'''Long-term precession of the equator'''
# Number of polynomial and periodic coefficients
npol = 4
nper = 14
# Polynomial coefficients
xypol = array([
[+5453.282155,-73750.930350],
[+0.4252841,-0.7675452],
[-0.00037173,-0.00018725],
[-0.000000152,+0.000000231],
])
# Periodic coefficients
xyper = array([
[256.75, -819.940624, 75004.344875, 81491.287984, 1558.515853],
[708.15, -8444.676815, 624.033993, 787.163481, 7774.939698,],
[274.20, 2600.009459, 1251.136893, 1251.296102, -2219.534038],
[241.45, 2755.175630, -1102.212834, -1257.950837, -2523.969396],
[2309.00, -167.659835, -2660.664980, -2966.799730, 247.850422],
[492.20, 871.855056, 699.291817, 639.744522, -846.485643],
[396.10, 44.769698, 153.167220, 131.600209, -1393.124055],
[288.90, -512.313065, -950.865637, -445.040117, 368.526116],
[231.10, -819.415595, 499.754645, 584.522874, 749.045012],
[1610.00, -538.071099, -145.188210, -89.756563, 444.704518],
[620.00, -189.793622, 558.116553, 524.429630, 235.934465],
[157.87, -402.922932, -23.923029, -13.549067, 374.049623],
[220.30, 179.516345, -165.405086, -210.157124, -171.330180],
[1200.00, -9.814756, 9.344131, -44.919798, -22.899655],
])
# Centuries since J2000
T = (epj-2000.0)/100.0
# Initialize Pₐ and Qₐ accumulators
X = 0.0
Y = 0.0
# Periodic Terms
for i in range(0,nper):
W = tau*T
A = W/xyper[i][0]
S = sin(A)
C = cos(A)
X = X + C*xyper[i][1] + S*xyper[i][3]
Y = Y + C*xyper[i][2] + S*xyper[i][4]
# Polynomial Terms
W = 1.0
for i in range(0,npol):
X = X + xypol[i][0]*W
Y = Y + xypol[i][1]*W
W = W*T
# X and Y (direction cosines)
X = X*as2r
Y = Y*as2r
# Form the equator pole vector
veq0 = X
veq1 = Y
W = X*X + Y*Y
veq2 = 0.0
if(W < 1.0):
veq2 = sqrt(1.0-W)
veq = array([veq0, veq1, veq2])
return veq
The Fortran subroutine ltp_PMAT generates the 3 × 3 rotation matrix P, constructed using Fabri parameterization (i.e. directly from the unit vectors for the ecliptic and equator poles – see Sect. 5.4). As well as calling the two previous subroutines, ltp_PMAT calls subroutines from the IAU SOFA library. The resulting matrix transforms vectors with respect to the mean equator and equinox of epoch 2000.0 into mean place of date.
The original Fortran code uses the SOFA commands pxp()
and pn()
. They in return use a few more functions in the library. The modulus W returned from iau_PN is tossed out in the Fortran code. I just do similarly here so that everything is common between the functions. There are Python wrappers for both SOFA and ERFA, the open source version of SOFA, and there are elegant numpy linear algebra functions, but the following was simple enough and works too.
def pxp(a,b):
'''p-vector outer (=vector=cross) product.
Given: two p-vectors (a and b)
Return: a x b
'''
xa = a[0]
ya = a[1]
za = a[2]
xb = b[0]
yb = b[1]
zb = b[2]
axb = array([])
axb = append(axb,ya*zb - za*yb)
axb = append(axb,za*xb - xa*zb)
axb = append(axb,xa*yb - ya*xb)
return axb
def pn(p):
'''Convert a p-vector into modulus and unit vector.
Given: p-vector (p)
Return: modulus (r), and unit vector (u)
'''
# Modulus of p-vector
# http://www.iausofa.org/2013_1202_C/sofa/pm.c
w = sqrt(p[0]*p[0] + p[1]*p[1] + p[2]*p[2])
u = array([])
if(w == 0.0):
# zero a p-vector
# http://www.iausofa.org/2013_1202_C/sofa/zp.c
u = append(u,[0.0, 0.0, 0.0])
else:
# unit vector
# http://www.iausofa.org/2013_1202_C/sofa/sxp.c
s = 1.0/w
u = append(u, s*p[0])
u = append(u, s*p[1])
u = append(u, s*p[2])
r = w
return r, u
def ltp_PMAT(epj):
'''Long-term precession matrix
Given: EPJ d Julian epoch (TT)
Return: RP d precession matrix, J2000.0 to date
The matrix is in the sense P_date = RP x P_J2000,
where P_J2000 is a vector with respect to the J2000.0 mean
equator and equinox and P_date is the same vector with respect to
the equator and equinox of epoch EPJ.
'''
# Equator pole (bottom row of matrix)
peqr = ltp_PEQU(epj)
# Ecliptic pole
pecl = ltp_PECL(epj)
# Equinox (top row of matrix)
V = pxp(peqr, pecl) # P-vector outer product.
w, EQX = pn(V) # Convert a p-vector into modulus and unit vector
# Middle row of matrix
V = pxp(peqr, EQX)
# The matrix elements
rp = array([])
rp = append(rp, [EQX, V, peqr])
rp = rp.reshape(3,3)
return rp
The Fortran subroutine ltp_PBMAT generates the 3 × 3 rotation matrix P × B, where B is the “frame bias matrix” that expresses the relative orientation of the GCRS and mean J2000.0 reference systems. A simple first-order implementation of the frame bias is used, the departure from rigor being well under 1 μas.
def ltp_PBMAT(epj):
'''Long-term precession matrix, including GCRS frame bias.
Given: EPJ d Julian epoch (TT)
Return: RPB d precession-bias matrix, J2000.0 to date
The matrix is in the sense P_date = RPB x P_J2000,
where P_J2000 is a vector in the Geocentric Celestial Reference
System, and P_date is the vector with respect to the Celestial
Intermediate Reference System at that date but with nutation
neglected.
A first order bias formulation is used, of sub-microarcsecond
accuracy compared with a full 3D rotation.
'''
# Frame bias (IERS Conventions 2010, Eqs. 5.21 and 5.33)
DX = -0.016617 * as2r
DE = -0.0068192 * as2r
DR = -0.0146 * as2r
# Precession matrix.
rp = ltp_PMAT(epj)
# Apply the bias
rpb = array([])
rpb = append(rpb, rp[0][0] - rp[0][1]*DR + rp[0][2]*DX)
rpb = append(rpb, rp[0][0]*DR + rp[0][1] + rp[0][2]*DE)
rpb = append(rpb, -rp[0][0]*DX - rp[0][1]*DE + rp[0][2] )
rpb = append(rpb, rp[1][0] - rp[1][1]*DR + rp[1][2]*DX)
rpb = append(rpb, rp[1][0]*DR + rp[1][1] + rp[1][2]*DE)
rpb = append(rpb, -rp[1][0]*DX - rp[1][1]*DE + rp[1][2] )
rpb = append(rpb, rp[2][0] - rp[2][1]*DR + rp[2][2]*DX)
rpb = append(rpb, rp[2][0]*DR + rp[2][1] + rp[2][2]*DE)
rpb = append(rpb, -rp[2][0]*DX - rp[2][1]*DE + rp[2][2] )
rpb = rpb.reshape(3,3)
return rpb
These test cases as well are translated from the original Fortran code by Vondrák et al.
To assist in checking for correct use of the above subroutines, we present below the results of calling each of them for the following test date: -1374 (i.e. 1375 BCE) May 3 (Gregorian calendar) at 13:52:19.2 TT. The equivalent Julian date and Julian epoch are 1219339.078000 (TT) and -1373.5959534565 (TT) respectively.
The functions compute the precession matrix when given the date as a julian epoch. epj
is based on the SOFA calendar helper function and is supplied in case it is useful.
def epj(dj):
'''Julian Date to Julian Epoch'''
# based on http://www.iausofa.org/2013_1202_C/sofa/epj.c
DJ00 = 2451545.0 # Reference epoch (J2000.0), Julian Date
DJY = 365.25 # Days per Julian year
epj = 2000.0 + (dj - DJ00)/DJY;
return epj
print(epj(1219339.078000))
epj = -1373.5959534565 # julian epoch
-1373.59595346
pecl = ltp_PECL(epj)
pecl_test = array([+0.00041724785764001342,-0.40495491104576162693,+0.91433656053126552350])
print(pecl)
print(pecl_test)
[ 4.17247858e-04 -4.04954914e-01 9.14336559e-01] [ 4.17247858e-04 -4.04954911e-01 9.14336561e-01]
pequ = ltp_PEQU(epj)
pequ_test = array([-0.29437643797369031532,-0.11719098023370257855,+0.94847708824082091796])
print(pequ)
print(pequ_test)
[-0.29437644 -0.11719098 0.94847709] [-0.29437644 -0.11719098 0.94847709]
rp = ltp_PMAT(epj)
rp_test = array([
[+0.68473390570729557360,+0.66647794042757610444,+0.29486722236305384992],
[-0.66669482609418419936,+0.73625636097440967969,-0.11595076448202158534],
[-0.29437643797369031532,-0.11719098023370257855,+0.94847708824082091796],
])
print(rp)
print(rp_test)
[[ 0.68473391 0.66647794 0.29486715] [-0.66669482 0.73625636 -0.11595076] [-0.29437644 -0.11719098 0.94847709]] [[ 0.68473391 0.66647794 0.29486722] [-0.66669483 0.73625636 -0.11595076] [-0.29437644 -0.11719098 0.94847709]]
rpxb = ltp_PBMAT(epj)
rpxb_test = array([
[+0.68473392912753224372,+0.66647788221176470103,+0.29486722236305384992],
[-0.66669476463873305255,+0.73625641199831485100,-0.11595079385100924091],
[-0.29437652267952261218,-0.11719099075396051880,+0.94847706065103424635],
])
print(rpxb)
print(rpxb_test)
[[ 0.68473393 0.66647788 0.29486722] [-0.66669476 0.73625642 -0.11595079] [-0.29437652 -0.11719099 0.94847706]] [[ 0.68473393 0.66647788 0.29486722] [-0.66669476 0.73625641 -0.11595079] [-0.29437652 -0.11719099 0.94847706]]
The above computations were performed using quadruple precision (REAL*128) so that all the reported decimals (except perhaps for the least significant digit in rare critical cases) are correct. Note that on typical computers ordinary double precision has only 15–16 decimal places of precision, and this must be taken into account when comparing results.
If there is an error, it will definitely will be due to a precision error. It is on my to-do list to massage these errors out.
Here are the periodic results from -200000 to 200000 years showing each component of the vector separately.
ecl_p_matrix_list = []
equ_p_matrix_list = []
year = arange(-200001.0, 200000.0, 100)
for y in year:
p_ecl = ltp_PECL(y)
p_equ = ltp_PEQU(y)
ecl_p_matrix_list.append(p_ecl)
equ_p_matrix_list.append(p_equ)
axis([-200000,200000,-.5,1])
plot(year, ecl_p_matrix_list)
#savefig('ecl_p_matrix.png')
show()
axis([-200000,200000,-.8,1.1])
plot(year, equ_p_matrix_list)
#savefig('equ_p_matrix.png')
show()
So what did we get? What are the units? How do we use it?
The Report of the International Astronomical Union Division I Working Group on Precession and the Ecliptic (2006) discusses some of the different ways to compute the precession matrix P. P is the matrix we calculated above. This precession matrix is used to transform the current coordinates (of epoch J2000) to the coordinates of any date.
The new positions (future or past of J2000) are optained by multiplying the coordinates of the standard epoch by the precession matrix. p₁ = P·p₀, where P is the below precession matrix and p₀ is the original position. The precession matrix is unitless.
$$P = \begin{pmatrix} P_{11} & P_{12} & P_{13} \\ P_{21} & P_{22} & P_{23} \\ P_{31} & P_{32} & P_{33} \end{pmatrix}$$When these papers mention the coordinate system based upon the mean equator and equinox, they are talking about how the pole axis traces a cone (wobbles like a top) when they speak of equinox precession. The earth goes through one such rotation in approximately every 26,000 years. The ecliptic plane is the path the Sun appears to traverse across the celestial sphere. The ecliptic pole is perpindicular to this.
The pole of the equator is needed because the Earth's equatorial plane moves in respect to the celestial sphere and it is not perpendicular to the Ecliptic pole. The angle between the Earth's equatorial plane and the ecliptic pole, ∈, is called the obliquity of the ecliptic and is about 23.4°.
The term "mean" in mean equator, in respect to the Earth's equatorial plane, is used when nutation is averaged out or omitted. I am not accounting for nutation in these examples, so we should note that the 'true pole' does not point at the epoch's celestial pole because it nutates away from the mean one in reality.
The following gets the coordinates for Polaris (in respect to J2000) from PyEphem.
def compute_polaris(year):
polaris = ephem.star('Polaris')
polaris.compute(str(year),epoch='2000')
ra = polaris.a_ra
dec = polaris.a_dec
x = cos(dec) * cos(ra)
y = cos(dec) * sin(ra)
z = sin(dec)
p = array([[x], [y], [z]])
return p
def ra_dec(v):
x = v[0][0]
y = v[1][0]
z = v[2][0]
ra = math.atan2(y,x)
dec = math.atan2(z,sqrt(x*x + y*y))
return ra, dec
import ephem
# renaming these so that they don't overwrite the pylab degrees
from ephem import hours as hrs
from ephem import degrees as deg
p0 = compute_polaris(2000)
(ra, dec) = ra_dec(p0)
print('RA: {}'.format(hrs(ra)))
print('DEC: {}'.format(deg(dec)))
print('cartesian position of Polaris in the year=2000, epoch=2000:')
x = p0[0][0]
y = p0[1][0]
z = p0[2][0]
print('{}\nthis vector has length {}'.format(
(x,y,z),sqrt(x*x + y*y + z*z)))
RA: 2:31:47.10 DEC: 89:15:51.0 cartesian position of Polaris in the year=2000, epoch=2000: (0.010127331660770541, 0.007897050033378511, 0.99991753347673784) this vector has length 1.0
This is then translated from -200000 to 200000. Where P is recomputed for every epoch date.
$$p_{1} = P\cdot p_{0} = \begin{pmatrix} P_{11} & P_{12} & P_{13} \\ P_{21} & P_{22} & P_{23} \\ P_{31} & P_{32} & P_{33} \end{pmatrix} \begin{pmatrix} x\\ y\\ z \end{pmatrix} = \begin{pmatrix} P_{11}x + P_{12}y + P_{13}z \\ P_{21}x + P_{22}y + P_{23}z \\ P_{31}x + P_{32}y + P_{33}z \end{pmatrix}$$def pdp(a,b):
'''p-vector inner (=scalar=dot) product.'''
# this is akin to the SOFA iauPdp, but it handles 3x3 square
# matrices times (dot) column vectors 3x1
# http://www.iausofa.org/2013_1202_C/sofa/pdp.c
row = []
for r in range(3):
w = a[r][0]*b[0] + a[r][1]*b[1] + a[r][2]*b[2]
row.append(w)
adb = array(row)
return adb
p0 = compute_polaris(2000)
print('The position of Polaris at J2000 is')
print(p0)
epj = 100000
P = ltp_PBMAT(epj) # Precession matrix, GCRS
p1 = compute_polaris(epj)
p1 = pdp(P, p1)
print('The new position of Polaris in 100000 years is')
print(p1)
The position of Polaris at J2000 is [[ 0.01012733] [ 0.00789705] [ 0.99991753]] The new position of Polaris in 100000 years is [[ 0.31728427] [-0.15847663] [ 0.9349951 ]]
(ra, dec) = ra_dec(p0)
print('In hours of right ascension and degrees of declination')
print('The position of Polaris at J2000 is')
print(str(hrs(ra)),str(deg(dec)))
print('The new position of Polaris in 100000 years is')
(ra1, dec1) = ra_dec(p1)
print(str(hrs(ra1)),str(deg(dec1)))
In hours of right ascension and degrees of declination The position of Polaris at J2000 is ('2:31:47.10', '89:15:51.0') The new position of Polaris in 100000 years is ('-1:46:09.87', '69:13:38.5')
Awesome, we were able to compute the position of Polaris in approximately 100000 years. These coordinates look to be in the right place too, and if we checked the length of the new cartesian vector it would be 1.0.
from mpl_toolkits.basemap import Basemap
width = 9000000
bm = Basemap(width=width, height=width, projection='aeqd',
lat_0=70.0, lon_0=280.0)
bm.drawparallels(np.arange(-80,81,10))
bm.drawmeridians(np.arange(-180,180,10))
# Position of Polaris at J2000 is p0
years = arange(-13001, 13000, 100)
for year in years:
P = ltp_PBMAT(year) # Precession matrix, GCRS
p_1 = compute_polaris(year)
p_1 = pdp(P, p1)
(ra1, dec1) = ra_dec(p_1)
x, y = bm(degrees(ra1), degrees(dec1))
if(year == years[0]):
bm.plot(x,y, marker='s', color='r')
elif(year == years[-1]):
bm.plot(x,y, marker='o', color='b')
else:
bm.plot(x,y, marker='.', color='black', markersize=4)
#savefig('images/polaris_one_precession.png')
This is comparable to the results from the IAU precession model PyEphem inherits from libastro. See github-issue-61.ipynb (nbviewer) for further explanation. We also see that this precession model is periodic for 400000 years. The dates supplied are in Julian Epoch (TT). Because of this, the range of years ±13000 and ±200000, are centered on Julian Epoch 0.0, and not epoch J2000.
width = 9000000
bm = Basemap(width=width, height=width, projection='aeqd',
lat_0=70.0, lon_0=280.0)
bm.drawparallels(np.arange(-80,81,10))
bm.drawmeridians(np.arange(-180,180,10))
for year in range(-200001, 200000, 100):
P = ltp_PBMAT(year) # Precession matrix, GCRS
p_1 = compute_polaris(year)
p_1 = pdp(P, p_1)
(ra1, dec1) = ra_dec(p_1)
x, y = bm(degrees(ra1), degrees(dec1))
bm.plot(x,y, marker='o', color='blue', alpha=0.3, ms=3)
#savefig('images/polaris_long_term.png')
You can also convert the coordinates of another epoch to J2000 by inverting the precession matrix. We transpose this like so
$$P^{-1} = \begin{pmatrix} P_{11} & P_{12} & P_{13} \\ P_{21} & P_{22} & P_{23} \\ P_{31} & P_{32} & P_{33} \end{pmatrix}$$I'm really glad I made a precession matrix and implemented it. Going through all of the steps makes it easier to comprehend. What started all of this was that I wanted to plot the Big Dipper 100000 years in the past and future in proper-motion.ipynb (nbviewer). Now I definitely am up to trying that again.
If you would like a Python package of this precession model, see the source at github.com/digitalvapor/vondrak or the Vondrak Package on PyPi. You can install with pip install vondrak
.