Circular Probable Error for Circular and Noncircular

Gaussian Impacts

by David W. Webb

ARL-MR-830 September 2012

Approved for public release; distribution is unlimited.

 NOTICES

Disclaimers

The findings in this report are not to be construed as an official Department of the Army position unless
so designated by other authorized documents.

Citation of manufacturer’s or trade names does not constitute an official endorsement or approval of the
use thereof.

Destroy this report when it is no longer needed. Do not return it to the originator.
Army Research Laboratory
Aberdeen Proving Ground, MD 21005-5066

ARL-MR-830 September 2012

Circular Probable Error for Circular and Noncircular

Gaussian Impacts

David W. Webb
Weapons and Materials Research Directorate, ARL

Approved for public release; distribution is unlimited.

 REPORT DOCUMENTATION PAGE Form Approved
OMB No. 0704-0188
Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering
and maintaining the data needed, and completing and reviewing the collection information. Send comments regarding this burden estimate or any other aspect of this collection of information,
including suggestions for reducing the burden, to Department of Defense, Washington Headquarters Services, Directorate for Information Operations and Reports (0704-0188), 1215 Jefferson
Davis Highway, Suite 1204, Arlington, VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to any penalty for failing to
comply with a collection of information if it does not display a currently valid OMB control number.
PLEASE DO NOT RETURN YOUR FORM TO THE ABOVE ADDRESS.
1. REPORT DATE (DD-MM-YYYY) 2. REPORT TYPE 3. DATES COVERED (From - To)
September 2012 Final January 2012
4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER
Circular Probable Error for Circular and Noncircular Gaussian Impacts
5b. GRANT NUMBER

5c. PROGRAM ELEMENT NUMBER

6. AUTHOR(S) 5d. PROJECT NUMBER

David W. Webb 622618AH80
5e. TASK NUMBER

5f. WORK UNIT NUMBER

7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION

REPORT NUMBER
U.S. Army Research Laboratory
ATTN: RDRL-WML-A ARL-MR-830
Aberdeen Proving Ground, MD 21005-5066
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR’S ACRONYM(S)

11. SPONSOR/MONITOR'S REPORT

NUMBER(S)

12. DISTRIBUTION/AVAILABILITY STATEMENT

Approved for public release; distribution is unlimited.

13. SUPPLEMENTARY NOTES

14. ABSTRACT
Circular probable error (CEP), defined as the radius of a circle centered at the aimpoint, which has a 50% probability of hit, is
one of many measures of precision used to characterize ballistic precision. When determining CEP for two-dimensional impact
data in which both dimensions are assumed to be independent, the case where both dispersion components are equal is a
straightforward application of the Rayleigh distribution. However, when the dispersion components differ, an approximate
CEP formula is necessary. This document evaluates the accuracy of the approximate CEP formula using Monte-Carlo
simulation and shows that its associated probability of hit is no more than 2.17% higher than the nominal value of 0.50.

15. SUBJECT TERMS

circular probable error, approximate formula, simulation, Rayleigh distribution, Wilson-Hilferty transformation
17. LIMITATION 18. NUMBER 19a. NAME OF RESPONSIBLE PERSON
16. SECURITY CLASSIFICATION OF: OF ABSTRACT OF PAGES
David W. Webb
a. REPORT b. ABSTRACT c. THIS PAGE 19b. TELEPHONE NUMBER (Include area code)
Unclassified Unclassified Unclassified UU 16 410-278-7014
Standard Form 298 (Rev. 8/98)
Prescribed by ANSI Std. Z39.18

ii
Contents

1. Introduction 1

2. CEP for Bivariate Independent Normal Impact Data 1

2.1 Case 1: Circular Data ( σ = σ X = σY ), Single Error Source ............................................1

2.2 Case 2: Noncircular Data ( σ X  σY ), Single Error Source .............................................2

2.3 Case 3: Circular Data, Multiple Error Sources...............................................................4
2.4 Case 4: Noncircular Data, Multiple Error Sources.........................................................4

3. Accuracy of the CEP Approximation for Noncircular Data 4

Appendix. MATLAB Code for Evaluating the Accuracy of Grubbs’ Approximate

Circular Probable Error (CEP) Formula 7

Distribution List 9

iii
INTENTIONALLY LEFT BLANK.

iv
1. Introduction

In ballistic performance testing, many statistics have been proposed as measures of precision.
Some of these statistics only consider the spread of the data in one direction, such as the
horizontal standard deviation and the extreme vertical dispersion. Other measures attempt to
capture a sense of the spread in both directions, e.g., extreme spread and circular probable error
(CEP). Grubbs1 describes each of these measures, along with several others in his seminal self-
published reference, informally known as the “Red Book.”

In this report, we examine the CEP, defined here as the radius of a circle centered at the aimpoint
which, in the long run, is impacted by 50% of the projectiles. We assume that impacts land on a
two-dimensional plane and that the horizontal (x) and vertical (y) locations are independent.*
Furthermore, we assume that the impacts follow a Gaussian (normal) distribution centered at the
aimpoint. We will first review the cases in which the only source of variation in each dimension
is the round-to-round variation, σx and σy. Then, we augment Grubbs’ approximate CEP
formula2 for those cases in which additional error sources contribute to the overall precision of
the weapon system. Finally, we evaluate the accuracy of the approximate CEP formula using
Monte-Carlo simulation.

2. CEP for Bivariate Independent Normal Impact Data

2.1 Case 1: Circular Data ( σ = σ X = σY ), Single Error Source

Because the impact locations are assumed to be centered at the aimpoint, we may set the
aimpoint to be (0,0) without loss of generality. Then, assume that and
are independent normal random variables; therefore, and , and
follows a Rayleigh distribution with parameter value 1. By definition, the CEP satisfies the
following:

, (1)

1Grubbs, F. E. Statistical Measures of Accuracy for Rifleman and Missile Engineers; Havre de Grace, MD, 1964.
*In referring to the two directions as “horizontal” and “vertical,” we imply that the target plane is upright, as is the case for
many weapon systems, including calibers as large as tank ammunition. However, for artillery systems, the ground is the most
likely target plane, and we instead refer to the x and y dimensions as “deflection” and “range,” respectively.
2Grubbs, F. E. Approximate Circular and Noncircular Offset Probabilities of Hitting. Operations Research 1964, 12, 51–62.

1
where is the Rayleigh cumulative distribution function. Since , we have
, which can be algebraically rearranged to obtain the following solution:

. (2)

2.2 Case 2: Noncircular Data ( σ X  σY ), Single Error Source

Assume that and are independent normal random variables; therefore,

and , and and are independent chi-square random
variables, each with one degree of freedom. Following a similar argument given by Grubbs,2 we
start by rewriting the probability statement as

. (3)

The left side of this last inequality is a weighted sum of independent chi-square random
variables, which, as Patnaik3 argues, has a distribution that can be approximated by a chi-squared
random variable when appropriately scaled by a positive constant a. This random variable is
denoted as , where n is the degrees of freedom. The approximation is found by matching the
first two moments of and . That is, we seek a and n such that
, (4)

and
. (5)

The left side of equation 4 is

, (6)

while the right side is

. (7)

The left side of equation 5 is

, (8)

and the right side is

. (9)

3Patnaik, P. B. The Use of the Range as an Estimator of Variance in Statistical Tests. Biometrika 1950, 37, 78–87.

2
Therefore, we have the following system of two equations in the two unknowns, a and n:

, (10)

and

. (11)

The solution to this system is and .

Therefore,

. (12)

By definition of the CEP, we seek a solution to

, (13)

which can be rewritten to obtain a probability statement based on the chi-square distribution in
equation 12:

. (14)

Next, we use the Wilson-Hilferty transformation, , which relates a chi-

square random variable to a standard normal.

. (15)

3
Since the median of the standard normal distribution is 0, the numerator of the right side of the
inequality needs to be set to 0 to find a solution to the CEP. That is,

, (16)

or

. (17)

Substituting for n, we arrive at the final solution of

. (18)

2.3 Case 3: Circular Data, Multiple Error Sources

When multiple sources contribute to overall weapon system accuracy (e.g., occasion-to-occasion,
location, crosswind), equation 18 still applies as long as these sources are uncorrelated and the
vertical and horizontal impacts are independent. However, in lieu of the single error source, ,
we use the total system error calculated from the usual root-sum-square formula,
.

2.4 Case 4: Noncircular Data, Multiple Error Sources

When the weapon system accuracy differs in the vertical and horizontal directions, the presence
of multiple error sources does not change equation 18 other than and are replaced by
and , respectively.

3. Accuracy of the CEP Approximation for Noncircular Data

The accuracy of the approximate formula for CEP (equation 18) can be determined using Monte-
Carlo simulation by randomly generating impact data under various assumed values of and
*
. The frequency with which the impacts fall within a circle of radius CEP should be close to
0.5. The MATLAB† code used to conduct this Monte-Carlo study is given in the appendix.
Figure 1 is a plot of values of vs. the frequency of hitting a circle of radius
CEP, based on 100,000,000 simulated impacts.

*If both and are increased by a common factor “k,” then it can be shown that the CEP will also increase by a factor of k.
Therefore, it is sufficient to just consider the ratio . In the simulation study, we fixed and varied .
†MATLAB is a trademark of The MathWorks, Inc.

4
Since the estimated hit probabilities are all slightly higher than 0.5, we see that Grubbs’ CEP
approximation is a slight overestimate. The approximation has an error between 0.47% and
2.17% when the horizontal and vertical errors are within two orders of magnitude of each other.
The relationship appears to have an asymptotic value of about 0.5073. A similar Monte-Carlo
study in which the order of magnitude of the error ratio was increased to as large as 6 upheld this
asymptotic behavior.

Figure 1. Relationship between relative size of horizontal and vertical errors and the hit probability of a circle
centered at the aimpoint and having radius given by Grubbs’ approximate formula.

5
INTENTIONALLY LEFT BLANK.

6
 Appendix. MATLAB Code for Evaluating the Accuracy of Grubbs’
Approximate Circular Probable Error (CEP) Formula

______________________
This appendix appears in its original form, without editorial change.

7
% OBTAIN P(HIT) ESTIMATES FOR HORIZONTAL ERRORS OF 1 TO 100
% KEEP VERTICAL ERROR EQUAL TO 1
i=[0:.01:2]; % exponent values
for j=1:length(i)
sigx(j)=10^i(j) % horizontal error
sig=sqrt(sigx(j)*sigx(j)+1); % root sum squares; NOTE sigy=1
v=2*(((sigx(j)^4)+1)/(sig^4)); % df from chi-square in Eq (6)
CEP=sig*(1-v/9)^1.5;
%
clear dat
for k=1:100 % generate 100 Ph estimates
imp=mvnrnd([0 0],[sigx(l)^2 0;0 1],1000000); % 1M simulated impacts
ph(k)=mean(imp(:,1).^2+imp(:,2).^2<=CEP^2); % hit frequency on CEP
end
phit(j)=mean(ph); % avg 100 hit frequencies to “incr n”
end

% GRAPHICS
plot(i, phit,'r-'); % error exponent versus Ph estimate
ylabel('Estimated Hit Probability within CEP')
xlabel('log_{10}(\sigma_{max} / \sigma_{min})')

8
NO. OF
COPIES ORGANIZATION

1 DEFENSE TECHNICAL
(PDF INFORMATION CTR
only) DTIC OCA
8725 JOHN J KINGMAN RD
STE 0944
FORT BELVOIR VA 22060-6218

1 DIRECTOR
US ARMY RESEARCH LAB
IMAL HRA
2800 POWDER MILL RD
ADELPHI MD 20783-1197

1 DIRECTOR
US ARMY RESEARCH LAB
RDRL CIO LL
2800 POWDER MILL RD
ADELPHI MD 20783-1197

9
NO. OF
COPIES ORGANIZATION

ABERDEEN PROVING GROUND

9 DIR USARL
(8 HC RDRL CII C
1 CD) B BODT
RDRL WML A
M ARTHUR
W OBERLE (1 CD)
C PATTERSON
R PEARSON
L STROHM
R YAGER
RDRL WML C
R THOMPSON

10