Namespaces
Variants
Actions

Difference between revisions of "Talk:Tikhonov-Phillips regularization"

From Encyclopedia of Mathematics
Jump to: navigation, search
(copied to discussion for TeXing)
 
(post-tex notes)
 
(22 intermediate revisions by the same user not shown)
Line 1: Line 1:
''Phillips–Tikhonov regularization''
+
Post $\TeX$ notes.
 
+
* Modified the norm subscripts to mention the space, $\left\|\cdot\right\|_1$ becomes $\left\|\cdot\right\|_{H_1}$, so as not cause confusion with the $L_p$ norms.
Fredholm integral equations of the first kind (cf. also [[Fredholm equation|Fredholm equation]]),
+
* Removed repeated reference for Fredholm Equation
 
+
* Added reference to  Sobolev space
<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/t/t120/t120100/t1201001.png"  /></td> </tr></table>
+
--[[User:Jjg|Jjg]] 18:00, 25 April 2012 (CEST)
 
 
present  special challenges to working scientists (cf. also [[Fredholm equation,  numerical methods|Fredholm equation, numerical methods]]; [[Fredholm  equation|Fredholm equation]]). Unlike equations of the second kind, for  which the existence of a unique stable solution can be guaranteed under  mild conditions, integral equations of the first kind are typically  ill-posed (cf. also [[Ill-posed problems|Ill-posed problems]]). The  blame for this misbehaviour lies with the kernel <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t1201002.png" />. The data function  <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t1201003.png" /> inherits some of  the structure and smoothness of the kernel and hence, if the kernel is  highly structured or very smooth, then a solution exists only for  functions <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t1201004.png" /> belonging to a  quite exclusive class of data. Furthermore, null functions for the kernel may be plentiful, so when solutions exist, they are seldom  unique. But the most troubling aspect of the first-kind equation is its  instability: perturbations in <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t1201005.png" /> which are very  small (with respect to reasonable metrics on the data space) may arise  from very large changes in <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t1201006.png" />. This instability  is a consequence of the Riemann–Lebesgue lemma (cf. [[Fourier  series|Fourier series]]) or, more generally, from the decay to zero of  the singular values of the kernel.
 
 
 
The basic existence  and uniqueness theory for Fredholm equations of the first kind with  square-integrable kernels was worked out by E. Picard  [[#References|[a13]]] in the early 1900s. Toward the 1950s, iterative  approximation methods for Fredholm equations of the first kind were  developed (see, e.g., [[#References|[a3]]], [[#References|[a7]]]), but  soon thereafter it became apparent that when these methods are applied  to practical problems, the approximations are cursed by the instability  noted above (see, e.g., [[#References|[a1]]], [[#References|[a9]]]).  This led to the development, in the early 1960s, of a non-iterative  stabilized approximation technique, now called the method of  regularization, by D.L. Phillips [[#References|[a12]]] and A.N. Tikhonov  [[#References|[a14]]].
 
 
 
The analysis of the method of  regularization is best carried out in the context of a [[Hilbert  space|Hilbert space]], where the integral operator is modeled by a  [[Compact operator|compact operator]] <img align="absmiddle"  border="0" src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t1201007.png" /> from a  Hilbert space <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t1201008.png" /> into a Hilbert  space <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t1201009.png" />. The norm of  <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010010.png" /> should be  sufficiently weak to accommodate realistic data functions, while that of  <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010011.png" /> should be strong  enough to impose desirable structure on the solution. In the method of  regularization a  "damped least-squares"  approximate solution of  <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010012.png" /> is sought by  minimizing the functional
 
 
 
<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/t/t120/t120100/t12010013.png"  /></td> </tr></table>
 
 
 
where  <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010014.png" /> is a  regularization parameter. The minimizer <img align="absmiddle"  border="0" src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010015.png" /> of  this functional is the solution of the well-posed second-kind equation
 
 
 
<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/t/t120/t120100/t12010016.png"  /></td> </tr></table>
 
 
 
(cf. also  [[Integral equation|Integral equation]]) and hence is unique and stable  with respect to perturbations of the data function <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010017.png" />. The  regularization parameter may be viewed as an arbiter between fidelity  and stability in the approximate solution; a choice of <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010018.png" /> that is too small  causes the approximate solution to inherit some of the instability of  the original ill-posed problem, while too large a choice tends to  over-smooth the approximate solution with consequent loss of  information.
 
 
 
A key point in practical problems is that  the data function is the result of observations and hence is subject to  error. The data function in hand is then an observation <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010019.png" /> satisfying  <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010020.png" />, where <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010021.png" /> is a bound for the observational error. One then takes the minimizer <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010022.png" /> of the functional  <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010023.png" /> as an approximate  solution. If <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010024.png" /> is chosen so that  <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010025.png" />, then <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010026.png" /> as <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010027.png" />, where <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010028.png" /> is the  minimal-norm least-squares solution of <img align="absmiddle"  border="0" src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010029.png" /> (cf.  also [[Least squares, method of|Least squares, method of]]). If <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010030.png" /> is in the range  of <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010031.png" /> and <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010032.png" />, then <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010033.png" />, and this order  is best possible (see [[#References|[a4]]], p. 41).
 
 
 
V.A.  Morozov [[#References|[a10]]] introduced an a posteriori parameter  choice strategy (already used informally by D.L. Phillips  [[#References|[a12]]]) called the discrepancy principle. According to  this principle, there is a unique <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010034.png" /> satisfying  <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010035.png" /> and <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010036.png" /> as <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010037.png" />. Furthermore, if  <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010038.png" /> is in the range  of <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010039.png" />, then <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010040.png" />, but this order  is best possible [[#References|[a4]]], p. 50. Discrepancy principles  that attain the optimal order <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010041.png" /> have been devised  by T. Raus, H. Gfrerer and H. Engl (see [[#References|[a2]]]). The  <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010042.png" /> brick wall for  ordinary Tikhonov regularization can be scaled by using an iterated  version of the regularization method. Iterated Tikhonov regularization  consists of sequentially minimizing the functionals
 
 
 
<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/t/t120/t120100/t12010043.png"  /></td> </tr></table>
 
 
 
where  <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010044.png" /> is the minimizer  of <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010045.png" /> and <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010046.png" /> is a suitable  sequence of regularization parameters. Under appropriate conditions,  this method attains an order of approximation <img align="absmiddle"  border="0" src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010047.png" /> for  any <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010048.png" /> (see  [[#References|[a5]]]).
 
 
 
The computational implementation  of the method of regularization requires a discretization procedure to  produce a finite-dimensional problem. The interpretation of the method  as a variational technique suggests finite-element methods, while its  interpretation as an integral equation of the second kind suggests  quadrature or degenerate-kernel methods (cf. also [[Degenerate  kernel|Degenerate kernel]]; [[Quadrature|Quadrature]]). In any case, a  successful computational experience relies on a delicate balancing of  regularization norm, regularization parameter, discretization  parameters, and characteristics of the data error. For more on  discretized ill-posed problems, see [[#References|[a6]]].
 
 
 
In  Tikhonov's original work, the regularizing norm <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010049.png" /> was taken to be a  Sobolev norm and the data norm <img align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010050.png" /> was the <img  align="absmiddle" border="0"  src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010051.png" />-norm. Therefore,  the Euler equation characterizing the regularized approximation is an  [[Integro-differential equation|integro-differential equation]] and the  convergence of the approximations is uniform. Phillips, however, used a  regularizing [[Semi-norm|semi-norm]] (the <img align="absmiddle"  border="0" src="https://www.encyclopediaofmath.org/legacyimages/t/t120/t120100/t12010052.png" />-norm  of the second derivative) and hence the Phillips method requires a more  detailed analysis than Tikhonov's method (see [[#References|[a8]]],  [[#References|[a11]]]).
 
 
 
====References====
 
<table><TR><TD  valign="top">[a1]</TD> <TD valign="top">  J. Douglas  Jr.,  "Mathematical programming and integral equations" , ''Symp.  Numerical Treatment of Ordinary Differential Equations, Integral and  Integro-differential Equations (Rome, 1960)'' , Birkhäuser  (1960)  pp.  269–274</TD></TR><TR><TD  valign="top">[a2]</TD> <TD valign="top">  H.W. Engl,  M.  Hanke,  A. Neubauer,  "Regularization of inverse problems" , Kluwer  Acad. Publ.  (1996)</TD></TR><TR><TD  valign="top">[a3]</TD> <TD valign="top">  V.M. Fridman,    "Method of successive approximations for Fredholm integral equations of  the first kind"  ''Uspekhi Mat. Nauk'' , '''11'''  (1956)  pp. 233–234  (In Russian)</TD></TR><TR><TD  valign="top">[a4]</TD> <TD valign="top">  C.W. Groetsch,    "The theory of Tikhonov regularization for Fredholm equations of the  first kind" , Pitman  (1984)</TD></TR><TR><TD  valign="top">[a5]</TD> <TD valign="top">  M. Hanke,    C.W. Groetsch,  "Nonstationary iterated Tikhonov regularization"  ''J.  Optim. Th. Appl.'' , '''98'''  (1998)  pp.  37–53</TD></TR><TR><TD  valign="top">[a6]</TD> <TD valign="top">  P.C. Hansen,    "Rank-deficient and discrete ill-posed problems" , SIAM (Soc. Industrial  Applied Math.)  (1998)</TD></TR><TR><TD  valign="top">[a7]</TD> <TD valign="top">  L. Landweber,   "An iteration formula for Fredholm integral equations of the first  kind"  ''Amer. J. Math.'' , '''73'''  (1951) pp.  615–624</TD></TR><TR><TD  valign="top">[a8]</TD> <TD valign="top">  A.K. Louis,    "Inverse und schlecht gestellte Probleme" , Teubner  (1989)</TD></TR><TR><TD  valign="top">[a9]</TD> <TD valign="top">  R.W. McKelvey,    "An analysis of approximative methods for Fredholm's integral equation  of the first kind"  ''Techn. Rep. David Taylor Model Basin, U.S. Navy  Dep., Washington, D.C.'' , '''19'''  (1956)</TD></TR><TR><TD  valign="top">[a10]</TD> <TD valign="top">  V.A. Morozov,    "On the solution of functional equations by the method of  regularization"  '
 
'Soviet Math. Dokl.'' , '''7'''  (1966)  pp.  414–417</TD></TR><TR><TD  valign="top">[a11]</TD> <TD valign="top">  V.A. Morozov,    "Regularization methods for ill-posed problems" , CRC  (1993)</TD></TR><TR><TD  valign="top">[a12]</TD> <TD valign="top">  D.L. Phillips,    "A technique for the numerical solution of certain integral equations  of the first kind"  ''J. Assoc. Comput. Mach.'' , '''9'''  (1962)  pp.  84–97</TD></TR><TR><TD  valign="top">[a13]</TD> <TD valign="top">  E. Picard,    "Sur un théorème générale relatif aux équations intégrales de première  espèce et sur quelques problèmes de physique mathématique"  ''Rend.  Circ. Mat. Palermo'' , '''29'''  (1910)  pp.  79–97</TD></TR><TR><TD  valign="top">[a14]</TD> <TD valign="top">  A.N. Tikhonov,    "Solution of incorrectly formulated problems and the regularization  method"  ''Soviet Math. Dokl.'' , '''4'''  (1963)  pp.  1035–1038</TD></TR></table>
 

Latest revision as of 16:00, 25 April 2012

Post $\TeX$ notes.

  • Modified the norm subscripts to mention the space, $\left\|\cdot\right\|_1$ becomes $\left\|\cdot\right\|_{H_1}$, so as not cause confusion with the $L_p$ norms.
  • Removed repeated reference for Fredholm Equation
  • Added reference to Sobolev space

--Jjg 18:00, 25 April 2012 (CEST)

How to Cite This Entry:
Tikhonov-Phillips regularization. Encyclopedia of Mathematics. URL: http://encyclopediaofmath.org/index.php?title=Tikhonov-Phillips_regularization&oldid=25383