Difference between revisions of "Dictator"

From Boolean Zoo
Jump to: navigation, search
(Created page with "== Definition == A function <math>f:\{-1,1\}^n \to \{-1,1\}</math> is called a '''dictator function''' if there exists an index <math>i</math> such that <math>f(x) = 1</math>...")
 
(Added more properties, refined the definition a bit.)
Line 1: Line 1:
 
== Definition ==
 
== Definition ==
A function <math>f:\{-1,1\}^n \to \{-1,1\}</math> is called a '''dictator function''' if there exists an index <math>i</math> such that <math>f(x) = 1</math> if and only if <math>x_i = 1</math>. In other words, the function is a dictator if and only if its output is controlled by only single bit from the input.
+
A function <math>f:\{-1,1\}^n \to \{-1,1\}</math> is called a '''dictator function''' if there exists an index <math>i</math> such that either <math>f(x) = 1</math> if and only if <math>x_i = 1</math> for all <math>x</math>, or <math>f(x) = -1</math> if and only if <math>x_i = 1</math> for all <math>x</math>. In other words, the function is a dictator if its output is controlled by only single bit from the input.
 +
 
 +
Sometimes, the term "dictator" refers only to the function which returns 1 when <math>x_i = 1</math>, and the function which returns -1 is called an "anti-dictator".  
  
 
== Properties ==  
 
== Properties ==  
* Dictators are locally testable: Given query access to a function <math>f</math>, only a constant number of queries is needed in order to determine if it is a dictator with high probability. <ref>Ryan O'Donnell, Analysis of Boolean functions, [http://www.contrib.andrew.cmu.edu/~ryanod/?p=1153 Theorem 7 in section 7.1]</ref>
+
* The Fourier representation of a dictator function is either <math>f(x) = x_i</math> or <math>f(x) = -x_i</math>.
* Dictators maximize mutual information: If <math>X,Y</math> are two iid random vectors in <math>\{-1,1\}^n</math> and <math>f,g</math> are two Boolean functions, then the mutual information <math>I(f(X), g(Y))</math> is maximized when <math>f</math> is a dictator and <math>g = \pm f</math>. <ref>Georg Pichler, Pablo Piantanida, Gerald Matz, [https://arxiv.org/abs/1604.02109 Dictator Functions Maximize Mutual Information]</ref>
+
* In a dictatorship, the <math>i</math>-th variable has [[influence]] 1, while all others have influence 0. Conversely, if a Boolean function <math>f</math> has total influence 1, then it is a dictator.<ref>https://math.stackexchange.com/questions/64449/a-boolean-function-with-total-influence-1-must-be-a-dictatorship</ref>
 +
* Dictators are locally testable: Given query access to a function <math>f</math>, only a constant number of queries is needed in order to determine if it is a dictator with high probability. <ref>Ryan O'Donnell, Analysis of Boolean functions, [http://www.contrib.andrew.cmu.edu/~ryanod/?p=1153 Theorem 7 in section 7.1]/</ref>
 +
* Dictators maximize mutual information: If <math>X,Y</math> are two correlated random vectors in <math>\{-1,1\}^n</math> with iid entries, and <math>f,g</math> are two Boolean functions, then the mutual information <math>I(f(X), g(Y))</math> is maximized when <math>f</math> is a dictator and <math>g = \pm f</math>. <ref>Georg Pichler, Pablo Piantanida, Gerald Matz, [https://arxiv.org/abs/1604.02109 Dictator Functions Maximize Mutual Information].</ref>
 +
* If a Boolean function <math>f</math>'s Fourier representation is close to a degree-1 polynomial, then <math>f</math> is close to the dictator function. <ref>Friedgut, Ehud; Kalai, Gil; Naor, Assaf (2002). "Boolean functions whose Fourier transform is concentrated on the first two levels". Adv. Appl. Math. 29 (3): 427–437. [https://doi.org/10.1016/S0196-8858%2802%2900024-6 doi:10.1016/S0196-8858(02)00024-6.]</ref>
 +
* TODO: someone more knowledgeable should add something about the Unique Games Conjecture.
  
 
== References ==
 
== References ==
 
<references/>
 
<references/>

Revision as of 13:05, 29 August 2018

Definition

A function [math]f:\{-1,1\}^n \to \{-1,1\}[/math] is called a dictator function if there exists an index [math]i[/math] such that either [math]f(x) = 1[/math] if and only if [math]x_i = 1[/math] for all [math]x[/math], or [math]f(x) = -1[/math] if and only if [math]x_i = 1[/math] for all [math]x[/math]. In other words, the function is a dictator if its output is controlled by only single bit from the input.

Sometimes, the term "dictator" refers only to the function which returns 1 when [math]x_i = 1[/math], and the function which returns -1 is called an "anti-dictator".

Properties

  • The Fourier representation of a dictator function is either [math]f(x) = x_i[/math] or [math]f(x) = -x_i[/math].
  • In a dictatorship, the [math]i[/math]-th variable has influence 1, while all others have influence 0. Conversely, if a Boolean function [math]f[/math] has total influence 1, then it is a dictator.[1]
  • Dictators are locally testable: Given query access to a function [math]f[/math], only a constant number of queries is needed in order to determine if it is a dictator with high probability. [2]
  • Dictators maximize mutual information: If [math]X,Y[/math] are two correlated random vectors in [math]\{-1,1\}^n[/math] with iid entries, and [math]f,g[/math] are two Boolean functions, then the mutual information [math]I(f(X), g(Y))[/math] is maximized when [math]f[/math] is a dictator and [math]g = \pm f[/math]. [3]
  • If a Boolean function [math]f[/math]'s Fourier representation is close to a degree-1 polynomial, then [math]f[/math] is close to the dictator function. [4]
  • TODO: someone more knowledgeable should add something about the Unique Games Conjecture.

References

  1. https://math.stackexchange.com/questions/64449/a-boolean-function-with-total-influence-1-must-be-a-dictatorship
  2. Ryan O'Donnell, Analysis of Boolean functions, Theorem 7 in section 7.1/
  3. Georg Pichler, Pablo Piantanida, Gerald Matz, Dictator Functions Maximize Mutual Information.
  4. Friedgut, Ehud; Kalai, Gil; Naor, Assaf (2002). "Boolean functions whose Fourier transform is concentrated on the first two levels". Adv. Appl. Math. 29 (3): 427–437. doi:10.1016/S0196-8858(02)00024-6.