Vertexing and Kinematic
Fitting, Part I:
Basic Theory
Lectures Given at SLAC
Aug. 5 – 7, 1998
Paul Avery
University of Florida
avery@phys.ufl.edu
http://www.phys.ufl.edu/~avery
Paul Avery
1
Basic theory
Overview of plan
 1st of several lectures on kinematic fitting
 Focus in this lecture on theory
 Plan of lectures




Lecture 1: Basic theory
Lecture 2: Introduction to the KWFIT fitting package
Lecture 3: Vertex fitting
Lecture 4: Building virtual particles
 References
 KWFIT
http://www.phys.ufl.edu/~avery/kwfit/ or
http://w4.lns.cornell.edu/~avery/kwfit/
 Several write-ups on fitting theory and constraints
http://www.phys.ufl.edu/~avery/fitting.html
Paul Avery
2
Basic theory
What is Kinematic Fitting?
Kinematic fitting is a mathematical procedure in which one uses the
physical laws governing a particle interaction or decay to improve
the measurements of the process.
For example, the fact that the tracks coming from D0  K    
decay must come from a common space point can be used to improve
the 4-momentum and positions of the daughter particles, thus
improving the mass resolution of the D0 .
Physical information is supplied via constraints. Each constraint is
expressed in the form of an equation expressing some physical
condition that the process must satisfy. In the example above, each
track contributes 2 constraints (r– and z) to the vertex requirement,
giving 8 constraints in all.
Vertexing is only one example. We can require instead that the
invariant mass of the particles be equal to 1.8654. This is known as a
mass constraint. We will discuss mass constraints later.
Paul Avery
3
Basic theory
Implementation of
Constraints
Constraints are generally implemented through a least-squares
procedure. Each constraint equation is linearized and added, via the
Lagrange multiplier technique, to the  2 equation of the tracks using
the covariance matrices of the tracks. Each track contributes a 7
parameter “measurement”, and the 7  7 track covariance matrix is
the generalization of the  2 for a single measurement.
One then minimizes the  2 simultaneously with the constraint conditions. The constraints “pull” the tracks away from their unconstrained
values, and the resulting  2 one obtains with n constraints is
distributed like a standard  2 with n degrees of freedom, if gaussian
errors apply. A histogram of fits to, say, 10,000 decays would clearly
show this distribution. Of course, since track errors are only
approximately gaussian, the actual distribution will have more events
in the tail than predicted by theory. Still, knowledge of the
distribution allows one to define reasonable  2 cuts.
For example, in the vertexing example D0  K     , there are a
total of 8 constraints, but 3 unknown parameters must be determined
(the D0 vertex). The total number of degrees of freedom is thus 2*4
– 3 = 5.
Paul Avery
4
Basic theory
Trivial Example
Let’s work out all the least squares machinery for a simple example.
Suppose we have two measurements, x1 and x2 with (independent)
errors 0.1. Now we impose the condition that we want the two variables to sum to 6. Why 6? I don’t know; just humor me for now.
6
x2
6
x1
Without the constraint condition, the total  2 of the measurements
could be written
dx  x i  dx  x i
2
2 
1
10
 12
2
2
20
 22
where x10 and x20 are the initial measurements of x1 and x2 , and
 1   2  01
. . Since there is no reason yet for the measurements to
stray from their initial values,  2  0 initially.
Paul Avery
5
Basic theory
The constraint is imposed using the Lagrange multiplier method, e.g.
2 
d
x1  x10
 12
i d
2
x2  x20
i  2ax  x  6f
2
1
 22
2
where  is a lagrange multiplier which must be determined (the factor of 2 is inserted to simplify the algebra).
We minimize the  2 by setting the partial derivatives wrt x1 , x2 and
 to 0. This yields, using  1   2  
1  2
 x1  x10 /  2    0
2 x1
d
i
1  2
 x2  x2 0 /  2    0
2 x 2
d
i
1  2
 x1  x2  6  0
2 
Paul Avery
6
Basic theory
We solve for the unknowns x1 , x2 and :
1
  2 12 x10  12 x2 0  6
d
x  3 d
x
x  3 d
x

Paul Avery
1
1
2
10
 x2 0
2
1
2
10
 x2 0
7
i
i
i
Basic theory
Error Analysis and
Covariance Matrix
The solution is only half the story, because we also care about the
errors and correlations of the updated parameters. From the above
discussion, we expect that the constraint will reduce the errors of the
original measurement.
We calculate the errors for x1 and x2 directly from the definition of
standard deviation, by averaging over all possible measurements.
 2x 
z


f  x  x  x 2 dx
First write the deviations from the mean
x1  x1  x1  0.5 x10  x2 0
x2  x2  x2
Paul Avery
d
 0.5d
x
10
8
i
i
 x2 0
Basic theory
The error information for more than one variable is elegantly
expressed in terms of the “covariance matrix”. For example, let
x1
x
.
x2
F
IJ
G
HK
The covariance matrix of the two variables wrt one another is
Vx ij  xix j , or in matrix form
Vx  xx


T
1
2
1
1
1
2
2
x x
F
G
Hx x
1
x I
F
G
Jax x f
x K
H
x x I
x x J
K
1
2
2
2
It is clear from the definition that Vx is symmetric (Vx ij  Vx ji ) and
the diagonal elements are just the squares of the standard deviations
(Vx ii   i2 ).
Paul Avery
9
Basic theory
For our toy problem, the initial and final covariance matrices are
Vx 0 
F
G
H0
2
0

2
2
I
J
K
Thus the errors are
 x1 
 x2 
F
G
 G2

G

H2
Vx

2

2
2
I
2J
 J
J
2 K

2
2
 0.071
 0.071
which are substantially smaller than before.
The correlation coefficient rx1 x2 is commonly used to express the
variation of one parameter with another. It is defined as
Vx1 x2
rx1 x2 
 x1 x2
Our simple constraint leads to rx1 x2  1, i.e., every fluctuation of x1
upward is matched by an equal fluctuation of x2 downward. This, of
course, was expected. Other kinds of constraints lead to different
correlations.
Paul Avery
10
Basic theory
General Constrained Fits
with Tracks
Kinematic fitting involving tracks is straightforward, although a bit
more complicated:
1. The initial tracks are defined by 7 parameters apiece
2. Each track has a 7  7 non-diagonal covariance matrix
3. There are typically > 1 constraints
4. The constraints are generally non-linear
Non-linearity is not a difficult problem since we expand about a
point close to the final answer anyway. Suppose that there are m
variables  and r constraints H   0:
H1    0
H2    0

Hr     0
Paul Avery
11
Basic theory
The constraints can be expanded to first order about the point  A:
H 
0   i A  j   jA  Hi  A
a fc
j
h af
 j
c
h
  Dij  j   jA  di
j
which can be written more naturally as a matrix equation
0  D  d
where      A and
F
G
DG
G

G
G
H
H1
 1
H2
 1
Hr
 1
H1
 2
H2
 2

Hr
 2




H1
 m
H2
 m

Hr
 m
I FD
J
G
D
J
G
J

G
J
G
J
K HD
11
D12  D1m
21
D22  D2 m
r1



Dr 2  Drm
I
J
J
J
J
K
FH a fI
G
H a
 fJ
J
dG
 J
G
G
HH a fJ
K
Paul Avery
1
A
2
A
r
A
12
Basic theory
For example, for our simple example of two variables satisfying the
constraint x1  x2  6  0 expanded about the point x A  3, 3 , we
get
D 1 1
af
a f
d0
The constraint equation becomes
x1
D
0
x2
F
IJ
G
HK
where x1  x1  3 and x2  x2  3.
Paul Avery
13
Basic theory
Matrix formulation of 2 problem
Kinematic fits differ only in how the matrices D and d are specified.
The complete  2 equation can be written compactly in matrix form:
a f dV i a   f  2 aD   d f
a
  f V a
   f 2 D  d
2   0
1
i
T
0
 0 ij
0 j
1
k
kl
l
k
T
0
0
where  0 are the unconstrained parameters,      A , and
1
1
2
2




m
r
FI
G
J
G
J
G
J
G
HJ
K
Paul Avery
14
FI
G
J
G
J
G
J
G
HJ
K
Basic theory
The first term in the  2 expression is the general form for a set of m
correlated variables. When the variables are uncorrelated, it collapses
to the familiar expression
d

 i   i0
i
i
2
 2i
The second term is the sum of the products of each of r Lagrange
multipliers  i by its corresponding constraint.
This works for the simple example:
d

2
1

d
x1  x10
Paul Avery
F 0 IFx  x I  2 1ax  3f 1ax  3f 0
iG
G
J
J
x

x
0
H
K
H K
1
  x1  x10 , x2  x2 0
2
2
i d
2
x2  x2 0
2
i
2
2
1
10
2
20
a
 2  x1  x2  6
15
1
2
f
Basic theory
General solution of 2 problem
We set to zero the partial derivatives of the  2 wrt each of the m
variables  and r Lagrange multipliers , giving a total of m + r
equations, enough to solve for all the unknowns.
The solution is demonstrated in my first fitting note, CBX 91–72.
Without going into details, the solution is
   0  V 0 DT 
a
  VD D 0  d
f
with covariance matrix
V  V 0  V 0 DT VDDV 0
The auxiliary matrix VD and  2 are
d
VD  DV 0 DT
1
i
 2   T VD1
a
  T D  0  d
f
Physically, VD is the covariance matrix of the lagrange multipliers .
Paul Avery
16
Basic theory
The following points should be noted:
1. Kinematic fitting problems are fully specified by the D and d
matrices. Once they are set, the solution is determined by the
above equations.
2. The solution requires the inverse of only a single matrix, the r  r
matrix DV 0 DT , which is used to obtain VD .
3. It can be shown that the new covariance matrix V has diagonal
elements smaller than the initial covariance matrix V 0 . Thus the
constraints are doing their job.
4. The  2 does not require the evaluation of V01, although the
formal definition uses that matrix. This is a great simplification
and permits the use track representations with non-invertible
covariance matrices (such as that used in KWFIT).
5. The  2 can be written as a sum of r terms, one per constraint. It’s
then possible to look at each of these terms separately in order to
get more discriminating power than from the overall  2 .
Paul Avery
17
Basic theory