There is no royal road to science, and only those who do not dread the
fatiguing climb of its steep paths have a chance of gaining its luminous
summits. (Karl MARX;Marx,1872)
This appendix contains various unrelated formulae that I find useful in my daily practice, but that Ido not necessarily remember. The problem is that gathering them here suppresses any incentive tomemorize them, therefore making it essential to have them written here.
C.1 3D Quantities and Volume Integrals
C.1.1 Differential Operators
The three most common coordinate systems are represented in Fig. C.1. In what follows, we expressthe differential operators in these systems.
Figure C.1:Most common coordinate systems.Licensed under CC BY-SA 4.0.
C.1.1.1 Gradient
Cartesian:
(C.1)
Cylindrical:
(C.2)
Spherical:
(C.3)
C.1.1.2 Laplacian
Cartesian:
(C.4)
Cylindrical:
(C.5)
Spherical:
(C.6)
C.1.1.3 Divergence
Cartesian:
(C.7)
Cylindrical:
(C.8)
Spherical:
(C.9)
C.1.1.4 Curl
Cartesian:
(C.10)
Cylindrical:
(C.11)
Spherical:
(C.12)
C.1.2 Vectorial Analysis
C.1.3 Integral Theorems
C.1.4 Dust Heating and Cooling: Two Ways of Slicing the Pis
One of the most elementary equations for a grain is the relation betweenits absorption efficiency and the power it absorbs or emits, given inEq. () andEq. (). Yet, it oftencauses problems to newcomers, who, by trying to visualize the rays, get the numerical factor in the integrand() wrong. I have seenseveral improper values: ,,etc. The solution is of course to explicitly write the integral, which I do below. I also provide twoalternative visual solutions to count the factor the right way.
The power absorbed by a spherical grain of radius, exposed to an isotropicISRF with mean intensity ,, canexpressed as:
(C.31)
where andare the grain surface andsolid angle elements. indicates the direction of the incident rays. The possible orders of the integrals over these twoelements is illustrated in Fig. C.2.
C.1.4.1 Solution 1
The case represented in Fig. C.2.acorresponds to the case where we first integrate over. Thus, parallel rays with agiven direction interceptthe grain on a surface :
(C.32)
Then, we need to integrate over all the possible ray directions:
(C.33)
C.1.4.2 Solution 2
The case represented in Fig. C.2.bcorresponds to the case where we first integrate the flux over theray directions, on an incident surface:
(C.34)
There is only a factor here, as we integrate the flux on a surface element, somewhere on the grain. The rest of thegrain shields this surface from the radiation coming from the other hemisphere. In addition, thisintegration is weighted by the inclination of the rays on the surface (this is the classicalflux formula). We thus have the flux received by a surface element of the grain, from allpossible ray directions. We now simply need to integrate over the whole grain surface::
(C.35)
C.2 Statistics
C.2.1 General Formulae
C.2.1.1 Moments of a PDF
Definitions.If we have a joint PDF, ,of two variables, and ,the first moments are the following.
Normalization:
(C.36)
Mean:
(C.37)
Variance:
(C.38)
Skewness:
(C.39)
Covariance:
(C.40)
Standard-deviation:
(C.41)
Correlation coefficient:
(C.42)
Estimators.The most commonly used non-robust estimators are the following.
Mean:
(C.43)
Standard-deviation:
(C.44)
Skewness:
(C.45)
Correlation coefficient:
(C.46)
C.2.1.2 Marginalization
Marginalizing over a parameter, :
(C.47)
Comparing two data sets, and :
(C.48)
C.2.1.3 Variable Change
If we have two sets of random variables, and , suchthat ,then the relation between their PDF is:
(C.49)
where the Jacobian of the transformation is:
(C.50)
C.2.1.4 Combining Uncertainties
If we have a set of random variables, ,the uncertainty of an arbitrary function of these parameters,, isgiven by:
(C.51)
where is the covariance matrix of the variable set. In the 2D case, posing, wehave:
(C.52)
and Eq. ()gives the usual expression:
(C.53)
Similarly, the covariance of two functions of the parameter set,and, is:
(C.54)
Rant about systematics.There is a long-lasting laboratorian legend that “systematics must be non-quadratically summed”.This is true in some cases and false in others. Everything depends on what we are talkingabout.
1.
If we are measuring a flux, ,with noise, ,and calibration uncertainty (systematics), ,the total uncertainty on the flux will be, according to Eq. ():.This is because .The fluctuations of the detector’s signal at the time of the observation do not have anythingto do with the error the instrument’s team made by deriving the calibration factor.
2.
Now, if we are summing the flux in two pixels with same flux and noise levels, we willget, using Eq. ():,because the calibration factors of the pixels were correlated. The total calibration uncertaintyis this time linearly summed: ,because (the error due to the calibration uncertainty is the same for both pixels).
3.
If we are now summing two systematics, such as the calibration and the backgroundsubtraction uncertainties, and ,we will sum them quadratically: .This is because the error the instrument’s team made deriving the calibration factor isindependent of the error we have made by selecting a region in one of the corners of ourmap, assuming it was free of galaxy emission.
C.2.2 Useful Probability Distributions
C.2.2.1 Binomial Distribution
Discrete probability distribution to get successes out of tries, eachone having a probability :
(C.55)
with and .
C.2.2.2 Poisson Distribution
Discrete probability distribution to get events per unit time knowing the mean expected number,, ofsuch events per unit time:
(C.56)
with and. The superposition of twoPoissonian events is alsoPoissonian with mean .It is the limit of the binomial distribution to large numbers:
(C.57)
C.2.2.3 Gaussian Distribution
(C.58)
with ,and all superior moments equal to 0. It is the limit of a Poisson distribution when:.
Multivariate form.A multivariate normal law of mean and covariance matrix is defined as:
(C.59)
Error function.Noting the CDF of a reduced normal law, the error functionis defined such that:
(C.60)
It is thus:
(C.61)
C.2.2.4 Student’s tDistribution
It is defined as:
(C.62)
with being the degree of freedom. Its mean is 0 and its standard-deviation, for, is.
C.2.2.5 Split-Normal Distribution
It is, to my mind, the most convenient asymmetric distribution:
(C.63)
Posing:
(C.64)
the first moments are:
(C.65)
C.2.2.6 Lorentzian Distribution
(C.66)
Its mean is , itsFWHM is ,but its standard-deviation is not defined.
C.2.3 Drawing random variables from an arbitrary distribution
C.2.3.1 The Rejection Method
The rejection methodis a widely used technique to draw a random variable,, from an arbitrary PDF,. It requires the ability to easilydraw a random variable, ,from a proposal distribution, ,such that . In caseis finiteover ,we can take:
(C.67)
The algorithm is the following.
1.
Draw a random variable from .
2.
Draw a uniform random variable between 0 and 1, .
3.
If ,then is accepted. Otherwise, if ,this draw is rejected, and we need to go back to the first step.
The closer isfrom ,the lower the rejection rate will be, and the faster the method will be. It is illustrated in Fig. C.3.a.
Figure C.3:Methods for drawing random numbers from arbitrary distributions.Panel(a)represents the rejection method applied to the distribution in blue, with the proposal in red.We have represented a first rejected draw and a second accepted one. Panel(b)represents theCDF method applied to the distribution in panel(a). Licensed under CC BY-SA 4.0.
C.2.3.2 Inverting the CDF
A general Monte-Carlo technique to draw a random variable,, from anarbitrary PDF, ,consists in drawing a uniform random variable between 0 and 1,,and inverting the Cumulative Distribution Function(CDF) of: