Namespaces
Variants
Actions

Difference between revisions of "Statistical experiments, method of"

From Encyclopedia of Mathematics
Jump to: navigation, search
(Importing text file)
 
m (fixing superscripts)
 
(One intermediate revision by one other user not shown)
Line 1: Line 1:
A method of numerical calculation that interprets required unknown values as characteristics of a convenient (related) random phenomenon <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s0873801.png" />; this phenomenon is simulated numerically, whereafter the required values are estimated using the simulation of observations of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s0873802.png" />. As a rule, an unknown <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s0873803.png" /> is sought in the form of the [[Mathematical expectation|mathematical expectation]] <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s0873804.png" /> of some random variable <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s0873805.png" /> on a probability space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s0873806.png" /> that describes <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s0873807.png" />, and the independent observations <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s0873808.png" /> are simulated (see [[Independence|Independence]]). Then, by the [[Law of large numbers|law of large numbers]]
+
<!--
 +
s0873801.png
 +
$#A+1 = 112 n = 0
 +
$#C+1 = 112 : ~/encyclopedia/old_files/data/S087/S.0807380 Statistical experiments, method of
 +
Automatically converted into TeX, above some diagnostics.
 +
Please remove this comment and the {{TEX|auto}} line below,
 +
if TeX found to be correct.
 +
-->
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s0873809.png" /></td> </tr></table>
+
{{TEX|auto}}
 +
{{TEX|done}}
  
When <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738010.png" />, the random error of this formula can be roughly estimated in probability using the [[Chebyshev inequality|Chebyshev inequality]] or, asymptotically, by the [[Central limit theorem|central limit theorem]]
+
A method of numerical calculation that interprets required unknown values as characteristics of a convenient (related) random phenomenon  $  \phi $;
 +
this phenomenon is simulated numerically, whereafter the required values are estimated using the simulation of observations of  $  \phi $.  
 +
As a rule, an unknown  $  z $
 +
is sought in the form of the [[Mathematical expectation|mathematical expectation]]  $  z = {\mathsf E} Z( \omega ) $
 +
of some random variable  $  Z( \omega ) $
 +
on a probability space  $  ( \Omega , {\mathcal A} , {\mathsf P}) $
 +
that describes  $  \phi $,
 +
and the independent observations  $  \phi $
 +
are simulated (see [[Independence|Independence]]). Then, by the [[Law of large numbers|law of large numbers]]
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738011.png" /></td> <td valign="top" style="width:5%;text-align:right;">(1)</td></tr></table>
+
$$
 +
z  \approx  \zeta _ {N}  = \
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738012.png" /></td> </tr></table>
+
\frac{Z( \omega  ^ {( 1)} ) + \dots + Z( \omega  ^ {( N)} ) }{N}
 +
.
 +
$$
  
The mathematical expectation <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738013.png" /> can also be estimated "by simulation" , which enables one to make an a posteriori [[Confidence estimation|confidence estimation]] of the accuracy of the calculation. The random phenomenon is usually simulated by means of a sequence of independent random numbers, uniformly distributed on the interval <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738014.png" /> (cf. [[Uniform distribution|Uniform distribution]]). For this purpose a measurable mapping <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738015.png" /> from the unit hypercube of countable dimension
+
When  $  {\mathsf E} | Z( \omega ) |  ^ {2} < \infty $,
 +
the random error of this formula can be roughly estimated in probability using the [[Chebyshev inequality|Chebyshev inequality]] or, asymptotically, by the [[Central limit theorem|central limit theorem]]
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738016.png" /></td> </tr></table>
+
$$ \tag{1 }
 +
{\mathsf P} \left \{ | z - \zeta _ {N} | <  
 +
\frac{a \sigma ( Z) }{N  ^ {1/2} }
  
onto <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738017.png" /> is used; <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738018.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738019.png" />, where the function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738020.png" /> depends essentially only on coordinates with small indices. The problem thus formally reduces to the calculation of the integral
+
\right \}  \approx \
 +
\mathop{\rm erf} ( a) ,\ \
 +
$$
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738021.png" /></td> </tr></table>
+
$$
 +
\sigma  ^ {2}  = {\mathsf E} | Z( \omega ) |  ^ {2} - [ {\mathsf E} Z( \omega )]  ^ {2} .
 +
$$
  
using the simplest quadrature formula with equal weights and random abscissae <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738022.png" />. It follows from (1) that the amount of calculation needed to achieve the desired accuracy <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738023.png" /> of the calculation of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738024.png" /> is determined, given a fixed confidence level <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738025.png" />, by the product <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738026.png" />, where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738027.png" /> is the mathematical expectation of the amount of calculation needed to construct a single realization of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738028.png" />; it increases rapidly as <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738029.png" /> diminishes. A successful choice of a model with sufficiently small <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738030.png" /> is therefore of great value. In particular, it may prove more useful in the original integral representation to make a priori an analytic integration over some of the variables <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738031.png" />, change some other variables, break the integration cube down into parts, separate the main part of the integral, use groups of dependent points <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738032.png" /> which give the exact quadrature formula for any desired class of functions, etc. The most advantageous "model" can be chosen by estimating roughly the values of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738033.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738034.png" /> in small preliminary numerical experiments. In making a series of calculations, a noticeable higher degree of accuracy can be obtained by appropriate statistical processing of "observations" and by choosing a corresponding program of "experiments" .
+
The mathematical expectation  $  {\mathsf E} | Z( \omega ) |  ^ {2} $
 +
can also be estimated  "by simulation" , which enables one to make an a posteriori [[Confidence estimation|confidence estimation]] of the accuracy of the calculation. The random phenomenon is usually simulated by means of a sequence of independent random numbers, uniformly distributed on the interval $ \{ {x } : {0 \leq  x \leq  1 } \} $ (cf. [[Uniform distribution|Uniform distribution]]). For this purpose a measurable mapping $ f $
 +
from the unit hypercube of countable dimension
  
A large class of models used in the method of statistical experiments is related to the scheme of random walks. In the simplest case, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738035.png" /> is a square matrix of order <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738036.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738037.png" />, where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738038.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738039.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738040.png" />; <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738041.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738042.png" />. Consider a Markov [[Random walk|random walk]] <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738043.png" /> through <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738044.png" /> states <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738045.png" />, with transition probabilities <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738046.png" /> from <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738047.png" /> to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738048.png" />, up to the transition at a random <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738049.png" />-st step to an extra absorbing state <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738050.png" />, with absorption probability <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738051.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738052.png" />. Under the assumption that the moving particle changes its weight according to the rule <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738053.png" />, if the <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738054.png" />-th random transition was from <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738055.png" /> to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738056.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738057.png" />, the solution of the equation <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738058.png" /> using a Neumann series can be interpreted coordinatewise as
+
$$
 +
= \mathop{\times} _ { k= 1} ^  \infty  \{ {x _ {k} } : {0 \leq  x _ {k} \leq  1 } \}
 +
\ \
 +
\left ( \textrm{ with Lebesgue  volume  } d V = \prod _ { k } dx _ {k} \right )
 +
$$
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738059.png" /></td> <td valign="top" style="width:5%;text-align:right;">(2)</td></tr></table>
+
onto  $  ( \Omega , {\mathcal A} , {\mathsf P}) $
 +
is used; $  \Omega = f( H) $,
 +
$  P = Vf ^ { - 1 } $,
 +
where the function  $  f $
 +
depends essentially only on coordinates with small indices. The problem thus formally reduces to the calculation of the integral
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738060.png" /></td> </tr></table>
+
$$
 +
\int\limits _ { H } Z( f( x))  dV
 +
$$
  
where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738061.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738062.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738063.png" />. Every "trajectory" <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738064.png" /> is simulated by its sequence of random numbers <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738065.png" />; the transition to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738066.png" /> is completed at the <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738067.png" />-th step from <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738068.png" /> when <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738069.png" />. The amount of work involved in constructing the trajectory and calculating the functional from it is proportional to its "length" <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738070.png" />; in this scheme <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738071.png" />.
+
using the simplest quadrature formula with equal weights and random abscissae  $  \mathbf x  ^ {( n)} $.  
 +
It follows from (1) that the amount of calculation needed to achieve the desired accuracy  $  \epsilon $
 +
of the calculation of  $  z $
 +
is determined, given a fixed confidence level  $  \mathop{\rm erf} ( a) $,  
 +
by the product  $  N \tau = \epsilon  ^ {- 2} a  ^ {2} \sigma  ^ {2} ( Z) \tau ( Z) $,  
 +
where  $  \tau $
 +
is the mathematical expectation of the amount of calculation needed to construct a single realization of  $  Z( \omega ) $;
 +
it increases rapidly as  $  \epsilon $
 +
diminishes. A successful choice of a model with sufficiently small  $ \sigma ^ {2} \tau $
 +
is therefore of great value. In particular, it may prove more useful in the original integral representation to make a priori an analytic integration over some of the variables  $  x _ {i} $,
 +
change some other variables, break the integration cube down into parts, separate the main part of the integral, use groups of dependent points  $  \mathbf x  ^ {n} $
 +
which give the exact quadrature formula for any desired class of functions, etc. The most advantageous  "model" can be chosen by estimating roughly the values of $  \sigma  ^ {2} $
 +
and  $ \tau $
 +
in small preliminary numerical experiments. In making a series of calculations, a noticeable higher degree of accuracy can be obtained by appropriate statistical processing of  "observations" and by choosing a corresponding program of  "experiments" .
  
When simulating random walks in continuous time, the motion must be made discrete. Suppose it is necessary to calculate the fraction <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738072.png" /> of radiation emanating from a sphere of radius <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738073.png" />, at the centre of which a source is situated. The motion of the radiated particles is rectilinear; on a path <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738074.png" /> with probability <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738075.png" /> a particle interacts with the medium, so that it is absorbed with probability <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738076.png" />, and is spherically-symmetrically dispersed with probability <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738077.png" />. The problem is solved by simulating the  "particle"  trajectories corresponding to the given stochastic differential description of the motion. Instead of breaking down the approximate path of the particle into steps <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738078.png" /> and testing at each step whether interaction has taken place, it is possible, by means of the exponential distribution with density <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738079.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738080.png" />, to generate the length of an <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738081.png" />-th random run <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738082.png" /> by means of a single random number, and find the next point of interaction <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738083.png" />. Moreover, it is possible not to perform a type of interaction with the medium, but to allow for absorption by a weight factor according to the rule <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738084.png" />. The polar and azimuthal angles <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738085.png" /> of the new direction of the motion are then looked for; <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738086.png" /> is distributed with uniform probability on the interval <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738087.png" />, and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738088.png" /> is distributed with uniform probability on the semi-interval <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738089.png" />. They define the unit vector <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738090.png" /> of the new direction of the motion. The simulation continues until the  "particle"  leaves the sphere, i.e. until the first event <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738091.png" />, where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738092.png" /> is the length of the path up to the boundary of the sphere, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738093.png" />. The average weight of the  "particles"  that have left the sphere provides an estimate of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738094.png" />. The integral expression obtained for the required quantity (which also follows from the integral transport equation) can be transformed into an integral along those trajectories <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738095.png" /> that do not leave the sphere. The run <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738096.png" /> must then be performed according to the conditional distribution with density <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738097.png" />; the new weight is defined by the rule <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738098.png" />, and on every trajectory the functional <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s08738099.png" /> is calculated. Then <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380100.png" />, where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380101.png" /> is a continuous function within <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380102.png" />. In this model, the trajectories are infinite, but the contribution of the later segments (those with a high number, if the segments are numbered beginning with the first one starting at the origin) is small; their simulation can therefore be stopped as soon as <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380103.png" /> by introducing a small systematic error into <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380104.png" />. The described scheme gives quite good results when <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380105.png" />. However, for large <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380106.png" /> its use may lead to false conclusions. When <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380107.png" />, departure from the sphere is rare, and is generally only achieved by trajectories all segments of which are long "on the average" . If <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380108.png" /> is not sufficiently large, then it is highly probable that these a-typical trajectories with a relatively large value of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380109.png" /> will not occur, and this may lead to underrated (though not to zero) estimates both of the required average <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380110.png" /> and the variance <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380111.png" />, i.e. an a posteriori measure of the error. Accuracy can be increased here, if an exponential transformation is used, by simulating trajectories by means of the exponential distribution with increased mean run and by compensating this by an extra exponential factor in the weight.
+
A large class of models used in the method of statistical experiments is related to the scheme of random walks. In the simplest case,  $  B $
 +
is a square matrix of order  $  m $,
 +
$  b _ {ij} = r _ {ij} \cdot p _ {ij} $,
 +
where  $  | r _ {ij} | < 1 $,
 +
$  p _ {ij} \geq  0 $,
 +
$  1 \leq  i, j \leq  m $;
 +
$  p _ {ij} + \dots + p _ {im} < 1 $,
 +
$  i = 1 \dots m $.
 +
Consider a Markov [[Random walk|random walk]]  $  w = \{ \theta ( 0), \dots \theta ( \nu ) \} $
 +
through  $  m $
 +
states  $  \theta _ {1} \dots \theta _ {m} $,
 +
with transition probabilities  $  p _ {ij} $
 +
from  $  \theta _ {i} $
 +
to  $  \theta _ {j} $,
 +
up to the transition at a random  $  ( \nu + 1 ) $-st step to an extra absorbing state  $  \theta _ {0} $,
 +
with absorption probability  $  p _ {i0} = 1 - ( p _ {i1} + \dots + p _ {im} ) $,
 +
$  p _ {00} = 1 $.
 +
Under the assumption that the moving particle changes its weight according to the rule  $  \rho _ {k} = \rho _ {k-} 1 r _ {ij} $,
 +
if the  $  k $-th random transition was from  $  \theta _ {i} $
 +
to  $  \theta _ {j} $,
 +
$  \rho _ {0} = 1 $,
 +
the solution of the equation  $  ( \mathbf 1 - B) y = g $
 +
using a Neumann series can be interpreted coordinatewise as
 +
 
 +
$$ \tag{2 }
 +
y _ {e}  =  g _ {e} + ( Bg) _ {e} + ( B  ^ {2} g) _ {e} + \dots =
 +
$$
 +
 
 +
$$
 +
= \
 +
{\mathsf E} [ g( \theta ( 0)) + \rho _ {1} g ( \theta
 +
( 1)) + \dots + \rho _  \nu  g( \theta ( \nu ))],
 +
$$
 +
 
 +
where  $  \theta ( 0) = \theta _ {e} $,
 +
$  g( \theta _ {s} ) = g _ {s} $,
 +
$  s = 1 \dots m $.
 +
Every  "trajectory"  $  \omega  ^ {( k)} $
 +
is simulated by its sequence of random numbers  $  x _ {n}  ^ {( k)} $;
 +
the transition to  $  \theta _ {j} $
 +
is completed at the  $  n $-th step from  $  \theta ( n- 1) = \theta _ {i} $
 +
when  $  p _ {i0} + \dots + p _ {i,j-} 1 \leq  x _ {n}  ^ {( j)} \leq  p _ {i0} + \dots + p _ {ij} $.
 +
The amount of work involved in constructing the trajectory and calculating the functional from it is proportional to its  "length"  $  \nu $;
 +
in this scheme  $  {\mathsf E} \nu < \infty $.
 +
 
 +
When simulating random walks in continuous time, the motion must be made discrete. Suppose it is necessary to calculate the fraction $  b $
 +
of radiation emanating from a sphere of radius $  R $,  
 +
at the centre of which a source is situated. The motion of the radiated particles is rectilinear; on a path $  ds $
 +
with probability $  a  ds $
 +
a particle interacts with the medium, so that it is absorbed with probability $  1- q $,  
 +
and is spherically-symmetrically dispersed with probability $  q $.  
 +
The problem is solved by simulating the  "particle"  trajectories corresponding to the given stochastic differential description of the motion. Instead of breaking down the approximate path of the particle into steps $  \Delta s $
 +
and testing at each step whether interaction has taken place, it is possible, by means of the exponential distribution with density $  p( s) = a  \mathop{\rm exp} (- as) $,
 +
s\geq  0 $,  
 +
to generate the length of an $  n $-th random run s _ {n} $
 +
by means of a single random number, and find the next point of interaction $  \mathbf r _ {n} $.  
 +
Moreover, it is possible not to perform a type of interaction with the medium, but to allow for absorption by a weight factor according to the rule $  \rho _ {n} = q \rho _ {n-} 1 $.  
 +
The polar and azimuthal angles $  ( \theta , \phi ) $
 +
of the new direction of the motion are then looked for; $  \cos  \theta $
 +
is distributed with uniform probability on the interval $  [- 1, 1] $,  
 +
and $  \phi $
 +
is distributed with uniform probability on the semi-interval $  [ 0, 2 \pi ) $.  
 +
They define the unit vector $  \mathbf e _ {n} $
 +
of the new direction of the motion. The simulation continues until the  "particle"  leaves the sphere, i.e. until the first event s _ {n} \geq  l _ {n} $,  
 +
where $  l _ {n} $
 +
is the length of the path up to the boundary of the sphere, $  | \mathbf r _ {n} + l _ {n} \mathbf e _ {n} | = R $.  
 +
The average weight of the  "particles"  that have left the sphere provides an estimate of $  b $.  
 +
The integral expression obtained for the required quantity (which also follows from the integral transport equation) can be transformed into an integral along those trajectories $  \omega $
 +
that do not leave the sphere. The run s _ {n} $
 +
must then be performed according to the conditional distribution with density $  a[ 1-  \mathop{\rm exp} (- al _ {n} )]  ^ {- 1}  \mathop{\rm exp} (- as) $;  
 +
the new weight is defined by the rule $  \rho _ {n+} 1 = q \rho _ {n} [ 1-  \mathop{\rm exp} (- al _ {n} )] $,  
 +
and on every trajectory the functional $  \beta = \sum \rho _ {n}  \mathop{\rm exp} (- al _ {n} ) $
 +
is calculated. Then $  b = {\mathsf E} \beta $,  
 +
where $  \beta ( \omega ( \mathbf x )) $
 +
is a continuous function within $  H $.  
 +
In this model, the trajectories are infinite, but the contribution of the later segments (those with a high number, if the segments are numbered beginning with the first one starting at the origin) is small; their simulation can therefore be stopped as soon as $  \rho _ {n} \leq  \delta $
 +
by introducing a small systematic error into $  b $.  
 +
The described scheme gives quite good results when $  aR \sim 1 $.  
 +
However, for large $  R $
 +
its use may lead to false conclusions. When $  aR \gg 1 $,  
 +
departure from the sphere is rare, and is generally only achieved by trajectories all segments of which are long "on the average". If $  N $
 +
is not sufficiently large, then it is highly probable that these a-typical trajectories with a relatively large value of $  \beta $
 +
will not occur, and this may lead to underrated (though not to zero) estimates both of the required average $  b $
 +
and the variance $  {\mathsf D} \beta $,  
 +
i.e. an a posteriori measure of the error. Accuracy can be increased here, if an exponential transformation is used, by simulating trajectories by means of the exponential distribution with increased mean run and by compensating this by an extra exponential factor in the weight.
  
 
It follows from formula (2) that by solving a system of linear equations via the method of statistical experiments, it is possible to find approximately only one unknown variable without calculating the others. This important property justifies the use of the method of statistical experiments, in spite of its slow convergence, for example, in solving boundary value problems for elliptic differential equations of the second order, when a solution has to be found at only one given point. In particular, for the Laplace equation the solution is written in the form of an integral over Wiener trajectories, i.e. the trajectories of a Brownian motion. The solution of certain boundary value problems for the meta-harmonic (including biharmonic) equations can be written in the form of integrals over the space of random trajectories of a Brownian particle with matrix weight. The simulation of the Brownian trajectories themselves, which undergo an infinitely large number of collisions for any interval of time, can be constructed in large sections by explicit specific methods.
 
It follows from formula (2) that by solving a system of linear equations via the method of statistical experiments, it is possible to find approximately only one unknown variable without calculating the others. This important property justifies the use of the method of statistical experiments, in spite of its slow convergence, for example, in solving boundary value problems for elliptic differential equations of the second order, when a solution has to be found at only one given point. In particular, for the Laplace equation the solution is written in the form of an integral over Wiener trajectories, i.e. the trajectories of a Brownian motion. The solution of certain boundary value problems for the meta-harmonic (including biharmonic) equations can be written in the form of integrals over the space of random trajectories of a Brownian particle with matrix weight. The simulation of the Brownian trajectories themselves, which undergo an infinitely large number of collisions for any interval of time, can be constructed in large sections by explicit specific methods.
Line 33: Line 168:
 
In solving non-linear equations by the method of statistical experiments, more complex models are used of flows of many particles that interact stochastically with the medium and with each other, including cascades of multiplying particles.
 
In solving non-linear equations by the method of statistical experiments, more complex models are used of flows of many particles that interact stochastically with the medium and with each other, including cascades of multiplying particles.
  
Apart from its slow convergence, this method has other shortcomings, including the inadequate reliability of the a posteriori estimation (1) of the random error. It can become less reliable as a result of both  "poor quality"  (e.g., correlation) of the random numbers used and  "non-typicality"  (e.g., low probability) of the results of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087380/s087380112.png" /> making a main contribution to the integral.
+
Apart from its slow convergence, this method has other shortcomings, including the inadequate reliability of the a posteriori estimation (1) of the random error. It can become less reliable as a result of both  "poor quality"  (e.g., correlation) of the random numbers used and  "non-typicality"  (e.g., low probability) of the results of $  \omega $
 +
making a main contribution to the integral.
  
 
Another name for the method of statistical experiments — the [[Monte-Carlo method|Monte-Carlo method]] — relates largely to the theory of modifying the method of statistical experiments.
 
Another name for the method of statistical experiments — the [[Monte-Carlo method|Monte-Carlo method]] — relates largely to the theory of modifying the method of statistical experiments.
  
 
For references, see [[Monte-Carlo method|Monte-Carlo method]].
 
For references, see [[Monte-Carlo method|Monte-Carlo method]].
 
 
  
 
====Comments====
 
====Comments====

Latest revision as of 11:25, 31 January 2022


A method of numerical calculation that interprets required unknown values as characteristics of a convenient (related) random phenomenon $ \phi $; this phenomenon is simulated numerically, whereafter the required values are estimated using the simulation of observations of $ \phi $. As a rule, an unknown $ z $ is sought in the form of the mathematical expectation $ z = {\mathsf E} Z( \omega ) $ of some random variable $ Z( \omega ) $ on a probability space $ ( \Omega , {\mathcal A} , {\mathsf P}) $ that describes $ \phi $, and the independent observations $ \phi $ are simulated (see Independence). Then, by the law of large numbers

$$ z \approx \zeta _ {N} = \ \frac{Z( \omega ^ {( 1)} ) + \dots + Z( \omega ^ {( N)} ) }{N} . $$

When $ {\mathsf E} | Z( \omega ) | ^ {2} < \infty $, the random error of this formula can be roughly estimated in probability using the Chebyshev inequality or, asymptotically, by the central limit theorem

$$ \tag{1 } {\mathsf P} \left \{ | z - \zeta _ {N} | < \frac{a \sigma ( Z) }{N ^ {1/2} } \right \} \approx \ \mathop{\rm erf} ( a) ,\ \ $$

$$ \sigma ^ {2} = {\mathsf E} | Z( \omega ) | ^ {2} - [ {\mathsf E} Z( \omega )] ^ {2} . $$

The mathematical expectation $ {\mathsf E} | Z( \omega ) | ^ {2} $ can also be estimated "by simulation" , which enables one to make an a posteriori confidence estimation of the accuracy of the calculation. The random phenomenon is usually simulated by means of a sequence of independent random numbers, uniformly distributed on the interval $ \{ {x } : {0 \leq x \leq 1 } \} $ (cf. Uniform distribution). For this purpose a measurable mapping $ f $ from the unit hypercube of countable dimension

$$ H = \mathop{\times} _ { k= 1} ^ \infty \{ {x _ {k} } : {0 \leq x _ {k} \leq 1 } \} \ \ \left ( \textrm{ with Lebesgue volume } d V = \prod _ { k } dx _ {k} \right ) $$

onto $ ( \Omega , {\mathcal A} , {\mathsf P}) $ is used; $ \Omega = f( H) $, $ P = Vf ^ { - 1 } $, where the function $ f $ depends essentially only on coordinates with small indices. The problem thus formally reduces to the calculation of the integral

$$ \int\limits _ { H } Z( f( x)) dV $$

using the simplest quadrature formula with equal weights and random abscissae $ \mathbf x ^ {( n)} $. It follows from (1) that the amount of calculation needed to achieve the desired accuracy $ \epsilon $ of the calculation of $ z $ is determined, given a fixed confidence level $ \mathop{\rm erf} ( a) $, by the product $ N \tau = \epsilon ^ {- 2} a ^ {2} \sigma ^ {2} ( Z) \tau ( Z) $, where $ \tau $ is the mathematical expectation of the amount of calculation needed to construct a single realization of $ Z( \omega ) $; it increases rapidly as $ \epsilon $ diminishes. A successful choice of a model with sufficiently small $ \sigma ^ {2} \tau $ is therefore of great value. In particular, it may prove more useful in the original integral representation to make a priori an analytic integration over some of the variables $ x _ {i} $, change some other variables, break the integration cube down into parts, separate the main part of the integral, use groups of dependent points $ \mathbf x ^ {n} $ which give the exact quadrature formula for any desired class of functions, etc. The most advantageous "model" can be chosen by estimating roughly the values of $ \sigma ^ {2} $ and $ \tau $ in small preliminary numerical experiments. In making a series of calculations, a noticeable higher degree of accuracy can be obtained by appropriate statistical processing of "observations" and by choosing a corresponding program of "experiments" .

A large class of models used in the method of statistical experiments is related to the scheme of random walks. In the simplest case, $ B $ is a square matrix of order $ m $, $ b _ {ij} = r _ {ij} \cdot p _ {ij} $, where $ | r _ {ij} | < 1 $, $ p _ {ij} \geq 0 $, $ 1 \leq i, j \leq m $; $ p _ {ij} + \dots + p _ {im} < 1 $, $ i = 1 \dots m $. Consider a Markov random walk $ w = \{ \theta ( 0), \dots \theta ( \nu ) \} $ through $ m $ states $ \theta _ {1} \dots \theta _ {m} $, with transition probabilities $ p _ {ij} $ from $ \theta _ {i} $ to $ \theta _ {j} $, up to the transition at a random $ ( \nu + 1 ) $-st step to an extra absorbing state $ \theta _ {0} $, with absorption probability $ p _ {i0} = 1 - ( p _ {i1} + \dots + p _ {im} ) $, $ p _ {00} = 1 $. Under the assumption that the moving particle changes its weight according to the rule $ \rho _ {k} = \rho _ {k-} 1 r _ {ij} $, if the $ k $-th random transition was from $ \theta _ {i} $ to $ \theta _ {j} $, $ \rho _ {0} = 1 $, the solution of the equation $ ( \mathbf 1 - B) y = g $ using a Neumann series can be interpreted coordinatewise as

$$ \tag{2 } y _ {e} = g _ {e} + ( Bg) _ {e} + ( B ^ {2} g) _ {e} + \dots = $$

$$ = \ {\mathsf E} [ g( \theta ( 0)) + \rho _ {1} g ( \theta ( 1)) + \dots + \rho _ \nu g( \theta ( \nu ))], $$

where $ \theta ( 0) = \theta _ {e} $, $ g( \theta _ {s} ) = g _ {s} $, $ s = 1 \dots m $. Every "trajectory" $ \omega ^ {( k)} $ is simulated by its sequence of random numbers $ x _ {n} ^ {( k)} $; the transition to $ \theta _ {j} $ is completed at the $ n $-th step from $ \theta ( n- 1) = \theta _ {i} $ when $ p _ {i0} + \dots + p _ {i,j-} 1 \leq x _ {n} ^ {( j)} \leq p _ {i0} + \dots + p _ {ij} $. The amount of work involved in constructing the trajectory and calculating the functional from it is proportional to its "length" $ \nu $; in this scheme $ {\mathsf E} \nu < \infty $.

When simulating random walks in continuous time, the motion must be made discrete. Suppose it is necessary to calculate the fraction $ b $ of radiation emanating from a sphere of radius $ R $, at the centre of which a source is situated. The motion of the radiated particles is rectilinear; on a path $ ds $ with probability $ a ds $ a particle interacts with the medium, so that it is absorbed with probability $ 1- q $, and is spherically-symmetrically dispersed with probability $ q $. The problem is solved by simulating the "particle" trajectories corresponding to the given stochastic differential description of the motion. Instead of breaking down the approximate path of the particle into steps $ \Delta s $ and testing at each step whether interaction has taken place, it is possible, by means of the exponential distribution with density $ p( s) = a \mathop{\rm exp} (- as) $, $ s\geq 0 $, to generate the length of an $ n $-th random run $ s _ {n} $ by means of a single random number, and find the next point of interaction $ \mathbf r _ {n} $. Moreover, it is possible not to perform a type of interaction with the medium, but to allow for absorption by a weight factor according to the rule $ \rho _ {n} = q \rho _ {n-} 1 $. The polar and azimuthal angles $ ( \theta , \phi ) $ of the new direction of the motion are then looked for; $ \cos \theta $ is distributed with uniform probability on the interval $ [- 1, 1] $, and $ \phi $ is distributed with uniform probability on the semi-interval $ [ 0, 2 \pi ) $. They define the unit vector $ \mathbf e _ {n} $ of the new direction of the motion. The simulation continues until the "particle" leaves the sphere, i.e. until the first event $ s _ {n} \geq l _ {n} $, where $ l _ {n} $ is the length of the path up to the boundary of the sphere, $ | \mathbf r _ {n} + l _ {n} \mathbf e _ {n} | = R $. The average weight of the "particles" that have left the sphere provides an estimate of $ b $. The integral expression obtained for the required quantity (which also follows from the integral transport equation) can be transformed into an integral along those trajectories $ \omega $ that do not leave the sphere. The run $ s _ {n} $ must then be performed according to the conditional distribution with density $ a[ 1- \mathop{\rm exp} (- al _ {n} )] ^ {- 1} \mathop{\rm exp} (- as) $; the new weight is defined by the rule $ \rho _ {n+} 1 = q \rho _ {n} [ 1- \mathop{\rm exp} (- al _ {n} )] $, and on every trajectory the functional $ \beta = \sum \rho _ {n} \mathop{\rm exp} (- al _ {n} ) $ is calculated. Then $ b = {\mathsf E} \beta $, where $ \beta ( \omega ( \mathbf x )) $ is a continuous function within $ H $. In this model, the trajectories are infinite, but the contribution of the later segments (those with a high number, if the segments are numbered beginning with the first one starting at the origin) is small; their simulation can therefore be stopped as soon as $ \rho _ {n} \leq \delta $ by introducing a small systematic error into $ b $. The described scheme gives quite good results when $ aR \sim 1 $. However, for large $ R $ its use may lead to false conclusions. When $ aR \gg 1 $, departure from the sphere is rare, and is generally only achieved by trajectories all segments of which are long "on the average". If $ N $ is not sufficiently large, then it is highly probable that these a-typical trajectories with a relatively large value of $ \beta $ will not occur, and this may lead to underrated (though not to zero) estimates both of the required average $ b $ and the variance $ {\mathsf D} \beta $, i.e. an a posteriori measure of the error. Accuracy can be increased here, if an exponential transformation is used, by simulating trajectories by means of the exponential distribution with increased mean run and by compensating this by an extra exponential factor in the weight.

It follows from formula (2) that by solving a system of linear equations via the method of statistical experiments, it is possible to find approximately only one unknown variable without calculating the others. This important property justifies the use of the method of statistical experiments, in spite of its slow convergence, for example, in solving boundary value problems for elliptic differential equations of the second order, when a solution has to be found at only one given point. In particular, for the Laplace equation the solution is written in the form of an integral over Wiener trajectories, i.e. the trajectories of a Brownian motion. The solution of certain boundary value problems for the meta-harmonic (including biharmonic) equations can be written in the form of integrals over the space of random trajectories of a Brownian particle with matrix weight. The simulation of the Brownian trajectories themselves, which undergo an infinitely large number of collisions for any interval of time, can be constructed in large sections by explicit specific methods.

In solving non-linear equations by the method of statistical experiments, more complex models are used of flows of many particles that interact stochastically with the medium and with each other, including cascades of multiplying particles.

Apart from its slow convergence, this method has other shortcomings, including the inadequate reliability of the a posteriori estimation (1) of the random error. It can become less reliable as a result of both "poor quality" (e.g., correlation) of the random numbers used and "non-typicality" (e.g., low probability) of the results of $ \omega $ making a main contribution to the integral.

Another name for the method of statistical experiments — the Monte-Carlo method — relates largely to the theory of modifying the method of statistical experiments.

For references, see Monte-Carlo method.

Comments

In the Western literature this method is almost universally known as the Monte-Carlo method.

References

[a1] B. Ripley, "Stochastic simulation" , Wiley (1987)
[a2] G. Marsaglia, A. Zaman, W.W. Tsang, "Toward a universal random number generator" Statistics and Prob. Letters , 9 : 1 (1990) pp. 35–39
[a3] S.M. Ermakov, V.V. Nekrutkin, A.S. Sipin, "Random processes for the classical equations for mathematical physics" , Kluwer (1989) (Translated from Russian)
How to Cite This Entry:
Statistical experiments, method of. Encyclopedia of Mathematics. URL: http://encyclopediaofmath.org/index.php?title=Statistical_experiments,_method_of&oldid=13776
This article was adapted from an original article by N.N. Chentsov (originator), which appeared in Encyclopedia of Mathematics - ISBN 1402006098. See original article