/prophet-secretary-through-blind-strategies

Collection of scripts that solve computational problems stated in the paper "Prophet secretary through blind strategies".

Primary LanguageRustMIT LicenseMIT

prophet-secretary-through-blind-strategies

Collection of scripts that solve computational problems stated in the paper Prophet Secretary Through Blind Strategies.

There are three results that need the help of computations to be proven and are listed below.

Lower bound

Result

There exists a nonincreasing function $\alpha: [0, 1] \to [0, 1]$ such that

  1. $\alpha(1) = 0$, and
  2. For all $x \in [0, 1]$, we have that

$$ \int_{ 0 }^{ x } \frac{ 1 - \alpha(y) }{ 1 - \alpha(x) } dy + \int_{ x }^{ 1 } \exp \left( \int_{ 0 }^{ y } \ln( \alpha(w) ) dw \right) dy \quad \in \quad [0.6653, 0.6720] . $$

Improved lower bound

Definitions

Let us define for $m \geq 1$ and $p \in (0, 1)$ the function $g_{ m , p } : [m] \to \mathbb{R}_+$ by

$$ g_{m, p}(k) = \begin{cases} \frac{ 1 }{ 1 - \frac{k}{m} ( 1 - p ) } &; k \leq \frac{m}{2} \\ \frac{ 2 }{ 1 + p } &; k > \frac{m}{2} \end{cases} $$

Result

There exists $1 \ge \alpha_1 \ge \ldots \ge \alpha_m \ge 0$ such that

$$ \min_{j \in [m+1]} f_j( \alpha_1, \ldots, \alpha_m ) \ge 0.66975 ,, $$

where, for every $j \in [m+1]$, $$ f_j( \alpha_1, \ldots, \alpha_m ) := \begin{cases} \sum\limits_{ k = 1 }^{ m } \left( \prod\limits_{ l \in [k-1] } \alpha_l \right)^{ \frac{1}{m} } \left( \frac{ 1 - \alpha_k^{ \frac{1}{m} } }{ -\ln \alpha_k } \right) &; j = 1 \ \sum\limits_{ k \in [j-1] } \frac{ 1 - \alpha_k }{ m( 1 - \alpha_j ) } + \sum\limits_{ k = j }^{ m } \left( \prod\limits_{ l \in [k-1] } \alpha_l \right)^{ \frac{1}{m} } g_{ m, \alpha_1 }( k - 1 ) \left( \frac{ 1 - \alpha_k^{ \frac{1}{m} } }{ -\ln \alpha_k } \right) &; j \in { 2, \ldots, m } \ \sum\limits_{ k \in [ m ] } \frac{ 1 - \alpha_k }{ m } &; j = m+1 ,. \end{cases} $$

Upper bound

Definitions

For $K \in [0, 3]$ and $\overline{t} \in [0, 1/3]$, define

$$ \alpha_{ K, \overline{t} }( t ) = \begin{cases} 1 &; 0 \leq t < \overline{t} \\ \beta_{ K, \overline{t} }( t ) &; \overline{t} \leq t \leq 1, \end{cases} $$

where $\beta_{ K, \overline{t} }$ is the solution of the following integro-differential equation

$$ \begin{align} \begin{cases} \dot{\beta}( t ) = - K \exp\left[ \int\limits_{ \overline{t} }^{ t } \ln \beta( s ) ds \right] & t \in ( \overline{t}, 1) \\ \beta( 1 ) = 0 \end{cases} \end{align} $$

Result

$$ \sup_{ K \in [0, 3] \quad \overline{t} \in [0, 1/3] } \min \left( 1 - \int\limits_{ 0 }^{ 1 } \alpha_{ K, \overline{t} }( s ) ds , \int\limits_{ 0 }^{ 1 } e^{ \int\limits_{ 0 }^{ s } \ln \alpha_{ K, \overline{t} }( w ) dw } ds \right) \le 0.675 $$