I am trying to understand all the steps to prove the TD0 algorithm, and I am following a proof which uses a theorem of Tommi Jaakkola, Michael I. Jordan and Satinder P. Singh, in the paper: On the Convergence of Stochastic Iterative Dynamic Programming Algorithms.

For proving an extension of the Dvoretzky theorem, they first state and claim that this lemma is a standard result and that it follows from Dvoretzky theorem:

Lemma 1

A random process

$\omega_{n+1}(x) = (1-\alpha_n(x))\omega_n(x)+\beta_n(x)r_n(x)$

coverges to zero with probability one if the following conditions are satisfied:

  1. $\sum_n\alpha(x) = \infty, \: \sum_n \alpha^2_n(x) < \infty, \: \sum_n\beta_n(x) = \infty, \: \sum_n \beta_n^2(x) < \infty,$ and $\mathbb{E}[\beta_n(x) \mid P_n] \leq \mathbb{E}[\alpha_n(x) \mid P_n]$ uniformly with probability 1.

  2. $\mathbb{E}[r_n(x) \mid P_n] = 0 $ and $\mathbb{E}[r_n^2(x) \mid P_n] \leq C $ w.p.1, where $ P_n = \{\omega_n, \omega_{n-1},...,r_{n-1},r_{n-2},...,\alpha_{n-1},\alpha_{n-2},...,\beta_{n-1},\beta_{n-2},...\}. $ All the random variables are allowed to depend on the past $P_n$

The proof in the paper is only the following: "Except for the appearance of $\beta_n(x)$ this is a standard result. With the above definitions convergence follows directly from Dvoretzky's extended theorem (Dvoretzky, 1956)."

It can be found here. This lemma appears in page 11 in the paper, in the proof of Theorem 1: Tommi Jaakkola, Michael I. Jordan and Satinder P. Singh paper

The Dvoretzky theorem is from a paper of 1956 by Aryeh Dvoretzky (On Stochastic Approximation). I will write it here:

Dvoretzky Theorem Let $(\Omega, F, \mu)$ be a probability space and $\alpha_n$, $\beta_n$ and $\gamma_n$, $n = 1, 2, ... $ non-negative real numbers satisfying: \begin{equation} \lim_{n\to\infty}a_n = 0, \end{equation} \begin{equation} \sum_{n=1}^{\infty}\beta_n < \infty, \end{equation} \begin{equation} \sum_{n=1}^{\infty}\gamma_n = \infty. \end{equation}

Let $\theta$ be a real number and $T_n: \mathbb{R}^n \rightarrow \mathbb{R}$, $n = 1, 2, ... $, measurable functions satisfying \begin{equation} |T_n(r_1,...,r_n) - \theta| \leq \max(\alpha_n, (1+\beta_n) |r_n - \theta| - \gamma_n) \end{equation} for all real numbers $r_1,...,r_n$. Let $X_1$ and $Y_n$, $n = 1, 2, ... $, be random variables. We define \begin{equation} X_{n+1}(\omega) = T_n[X_1(\omega),...,X_n(\omega)] + Y_n(\omega), \quad n \geq 1. \end{equation}

Then, the conditions \begin{equation} \mathbb{E}[X_1^2] < \infty, \end{equation} \begin{equation} \sum_{n=1}^{\infty}\mathbb{E}[Y_n^2] < \infty, \end{equation} and \begin{equation} \mathbb{E}[Y_n \mid X_1, ..., X_n] = 0 \end{equation} w.p.1 for all $n$, imply \begin{equation} \lim_{n\to\infty}\mathbb{E}[(X_n-\theta)^2] = 0 \end{equation} and \begin{equation} \mathbb{P}(\lim_{n \to \infty}X_n = \theta) = 1. \end{equation}

And now an extension to the case where the coefficients are non-negative functions.

The theorem remains valid if $\alpha_n$, $\beta_n$ y $\gamma_n$ are replaces by non-negative functions $\alpha_n(r_1,...,r_n)$, $\beta_n(r_1,...,r_n)$ y $\gamma_n(r_1,...,r_n)$, respectively, provided they satisfy the conditions:

  1. The functions $\alpha_n(r_1,...,r_n)$ are uniformly bounded and \begin{equation} \lim_{n\to\infty}\alpha_n(r_1,...,r_n) = 0 \end{equation} uniformly for all sequences $r_1,...,r_n$.

  2. The functions $\beta_n(r_1,...,r_n)$ are measurable and \begin{equation} \sum_{n=1}^{\infty}\beta_n(r_1,...,r_n) \end{equation} is uniformly bounded and uniformly convergent for all sequences $r_1,...,r_n$.

  3. The functions $\gamma_n(r_1,...,r_n)$ satisfy \begin{equation} \sum_{n=1}^{\infty}\gamma_n(r_1,...,r_n) = \infty \end{equation} uniformly for all sequences $r_1,...,r_n$, for which \begin{equation} \sup_{n \geq 1} |r_n| < L, \end{equation} $L < \infty$ being an arbitrary number.

The source can be found here: Dvoretzky paper

However, I can't find a way to relate that lemma to the Dvoretzky theorem, and nor can I find papers that prove the result, as Jaakkola states. Is it actually easy to prove that from the theorem? I would appreciate any help in this matter, and it would be perfect if you could provide an article or book where they prove that, or something similar.


1 Answer 1


The Dvoretzky Stochastic Approximation Theorem is a result in the field of stochastic approximation theory, which provides insights into the convergence properties of certain iterative algorithms used in optimization and machine learning. Unfortunately, the link you provided is not accessible as it directs to a specific question on the AI Stack Exchange that might have been removed or modified.

However, I can still provide you with a general explanation of the consequences of the Dvoretzky Stochastic Approximation Theorem. The theorem is named after the mathematician Aryeh Dvoretzky and establishes conditions under which a sequence of random variables converges almost surely to a fixed point. The fixed point is typically the solution to an optimization problem or an equilibrium point in game theory.

The theorem provides guarantees about the convergence behavior of stochastic approximation algorithms, which are iterative algorithms that update a sequence of estimates based on noisy or stochastic observations. Stochastic approximation algorithms are commonly used in reinforcement learning, online optimization, and other areas of machine learning.

The consequences of the Dvoretzky Stochastic Approximation Theorem are as follows:

  1. Almost sure convergence: The theorem guarantees that under certain conditions, the sequence of estimates generated by a stochastic approximation algorithm converges to the true solution with probability 1, or almost surely. This means that the algorithm will eventually converge to the correct solution, regardless of the initial conditions or the presence of noise in the observations.

  2. Noisy observations: The theorem takes into account the presence of noise or randomness in the observations used by the algorithm. Stochastic approximation algorithms are designed to handle situations where the observations are corrupted by noise, making them suitable for learning in uncertain environments.

  3. Conditions for convergence: The theorem provides specific conditions that need to be satisfied for the algorithm to converge. These conditions typically involve the step size or learning rate of the algorithm, as well as properties of the noise or random variables involved in the estimation process. Understanding these conditions is crucial for designing and analyzing stochastic approximation algorithms.

Overall, the Dvoretzky Stochastic Approximation Theorem is an important result in the field of stochastic approximation theory, providing theoretical guarantees for the convergence of iterative algorithms used in optimization and machine learning. Although I cannot provide the specific details of the question you linked, I hope this explanation gives you a general understanding of the consequences of the theorem.

  • $\begingroup$ Thank you! Your explanation helps me in understanding a little more the Theorem, but I am kind of having more of a technical problem trying to apply that theorem to prove the lemma. I don't know how to identify the coefficients in the lemma with the ones that appear in the Dvoretzky theorem. $\endgroup$
    – Kareit
    May 14, 2023 at 14:32
  • $\begingroup$ On another subject, aren't the links directing to papers and not to the forum? 1. Jaakkolla et.al paper: hdl.handle.net/1721.1/7205 2. Dvoretzky paper: projecteuclid.org/proceedings/… $\endgroup$
    – Kareit
    May 14, 2023 at 14:32

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .