Structured methodology for constructing and verifying mathematical proofs in statistical research
Constructs and verifies rigorous mathematical proofs for statistical theorems and derivations.
npx claudepluginhub data-wise/scholarThis skill inherits all available tools. When active, it can use any tool Claude has access to.
Structured methodology for constructing and verifying mathematical proofs in statistical research
Use this skill when working on: mathematical proofs, theorem development, derivations, consistency proofs, asymptotic arguments, identification proofs, or verifying proof correctness.
Every rigorous statistical proof should contain:
\begin{theorem}[Name]
\label{thm:name}
Under Assumptions \ref{A1}--\ref{An}, [precise claim].
\end{theorem}
\begin{proof}
The proof proceeds in [n] steps.
\textbf{Step 1: [Description]}
[Content with justification for each transition]
\textbf{Step 2: [Description]}
[Content]
\vdots
\textbf{Step n: Conclusion}
Combining Steps 1--[n-1], we obtain [result], completing the proof.
\end{proof}
Goal: Show that a causal/statistical quantity is uniquely determined from observed data distribution.
Standard Structure:
Template:
\begin{theorem}[Identification of $\psi$]
Under Assumptions \ref{A:consistency}--\ref{A:positivity}, the causal effect
$\psi = E[Y(a)]$ is identified by
\[
\psi = \int E[Y \mid A=a, X=x] \, dP(x).
\]
\end{theorem}
\begin{proof}
\begin{align}
E[Y(a)] &= E[E[Y(a) \mid X]] && \text{(law of iterated expectations)} \\
&= E[E[Y(a) \mid A=a, X]] && \text{(A\ref{A:exchangeability}: $Y(a) \indep A \mid X$)} \\
&= E[E[Y \mid A=a, X]] && \text{(A\ref{A:consistency}: $Y = Y(A)$)} \\
&= \int E[Y \mid A=a, X=x] \, dP(x) && \text{(definition)}
\end{align}
which depends only on the observed data distribution.
\end{proof}
Goal: Show that an estimator converges to the true parameter value.
Standard Structure:
Key Tools:
Template:
\begin{theorem}[Consistency]
Under Assumptions \ref{A1}--\ref{An}, $\hat{\theta}_n \xrightarrow{p} \theta_0$.
\end{theorem}
\begin{proof}
Define $M_n(\theta) = n^{-1} \sum_{i=1}^n m(O_i; \theta)$ and
$M(\theta) = E[m(O; \theta)]$.
\textbf{Step 1: Uniform convergence}
By [ULLN conditions], $\sup_{\theta \in \Theta} |M_n(\theta) - M(\theta)| \xrightarrow{p} 0$.
\textbf{Step 2: Unique maximum}
$M(\theta)$ is uniquely maximized at $\theta_0$ (by identifiability).
\textbf{Step 3: Conclusion}
By standard M-estimation theory, Steps 1--2 imply $\hat{\theta}_n \xrightarrow{p} \theta_0$.
\end{proof}
Goal: Establish $\sqrt{n}(\hat{\theta}_n - \theta_0) \xrightarrow{d} N(0, V)$.
Standard Structure:
Key Tools:
Template:
\begin{theorem}[Asymptotic Normality]
Under Assumptions \ref{A1}--\ref{An},
\[
\sqrt{n}(\hat{\theta}_n - \theta_0) \xrightarrow{d} N(0, V)
\]
where $V = E[\phi(O)\phi(O)^\top]$ and $\phi$ is the influence function.
\end{theorem}
\begin{proof}
\textbf{Step 1: Score equation}
$\hat{\theta}_n$ solves $\mathbb{P}_n[\psi(O; \theta)] = 0$ where $\psi = \partial_\theta m$.
\textbf{Step 2: Taylor expansion}
\[
0 = \mathbb{P}_n[\psi(O; \hat{\theta}_n)] = \mathbb{P}_n[\psi(O; \theta_0)]
+ \mathbb{P}_n[\dot{\psi}(O; \tilde{\theta})](\hat{\theta}_n - \theta_0)
\]
\textbf{Step 3: Rearrangement}
\[
\sqrt{n}(\hat{\theta}_n - \theta_0) = -\left(\mathbb{P}_n[\dot{\psi}]\right)^{-1}
\sqrt{n} \mathbb{P}_n[\psi(O; \theta_0)]
\]
\textbf{Step 4: Apply CLT}
$\sqrt{n} \mathbb{P}_n[\psi(O; \theta_0)] \xrightarrow{d} N(0, \text{Var}(\psi))$ by CLT.
\textbf{Step 5: Slutsky}
$\mathbb{P}_n[\dot{\psi}] \xrightarrow{p} E[\dot{\psi}]$ by WLLN. Apply Slutsky's theorem.
\end{proof}
Goal: Show estimator achieves semiparametric efficiency bound.
Standard Structure:
Template:
\begin{theorem}[Semiparametric Efficiency]
$\hat{\theta}_n$ is semiparametrically efficient with influence function
\[
\phi(O) = [optimal formula]
\]
achieving the efficiency bound $V_{\text{eff}} = E[\phi(O)^2]$.
\end{theorem}
Goal: Show estimator is consistent if either nuisance model is correctly specified.
Standard Structure:
Template:
\begin{theorem}[Double Robustness]
The estimator $\hat{\psi}_{DR}$ is consistent if either:
\begin{enumerate}
\item The outcome model $\mu(a,x) = E[Y \mid A=a, X=x]$ is correctly specified, or
\item The propensity score $\pi(x) = P(A=1 \mid X=x)$ is correctly specified.
\end{enumerate}
\end{theorem}
\begin{proof}
The estimating equation has the form:
\[
\psi - \hat{\psi}_{DR} = E\left[\frac{(A-\pi)(Y-\mu)}{\pi(1-\pi)}\right] + o_p(1)
\]
The bias term $(A-\pi)(Y-\mu)$ is zero in expectation if either:
\begin{itemize}
\item $E[A-\pi \mid X] = 0$ (propensity correctly specified), or
\item $E[Y-\mu \mid A, X] = 0$ (outcome correctly specified).
\end{itemize}
\end{proof}
For each step, verify:
| Error | Example | Fix |
|---|---|---|
| Interchanging limits | $\lim \sum \neq \sum \lim$ | Verify DCT/MCT conditions |
| Division by zero | $1/\pi(x)$ when $\pi(x)=0$ | State positivity assumption |
| Incorrect conditioning | $E[Y \mid A,X] \neq E[Y \mid X]$ | Check independence structure |
| Wrong norm | $|f|2$ vs $|f|\infty$ | Verify which space |
| Missing measurability | Random variable not measurable | State measurability |
| Error | Example | Fix |
|---|---|---|
| Circular reasoning | Using result to prove itself | Check logical dependency |
| Unstated assumption | "Clearly, X holds" | Make all assumptions explicit |
| Incorrect quantifier | $\exists$ vs $\forall$ | Be precise about scope |
| Missing case | Not handling $\theta = 0$ | Enumerate all cases |
| Error | Example | Fix |
|---|---|---|
| Confusing $\xrightarrow{p}$ and $\xrightarrow{d}$ | Different convergence modes | State which mode |
| Ignoring dependence | Applying iid CLT to dependent data | Check independence |
| Wrong variance | Using population variance for sample | Distinguish estimator/parameter |
| Symbol | Meaning |
|---|---|
| $Y(a)$ | Potential outcome under treatment $a$ |
| $Y(a,m)$ | Potential outcome under $A=a$, $M=m$ |
| $M(a)$ | Potential mediator under treatment $a$ |
| $NDE$ | Natural Direct Effect: $E[Y(1,M(0)) - Y(0,M(0))]$ |
| $NIE$ | Natural Indirect Effect: $E[Y(1,M(1)) - Y(1,M(0))]$ |
| $TE$ | Total Effect: $E[Y(1) - Y(0)] = NDE + NIE$ |
| $P_M$ | Proportion Mediated: $NIE/TE$ |
| Symbol | Meaning |
|---|---|
| $\theta_0$ | True parameter value |
| $\hat{\theta}_n$ | Estimator based on $n$ observations |
| $\phi(O)$ | Influence function |
| $\mathbb{P}_n$ | Empirical measure |
| $\mathbb{G}_n$ | Empirical process: $\sqrt{n}(\mathbb{P}_n - P)$ |
| Symbol | Meaning |
|---|---|
| $\xrightarrow{p}$ | Convergence in probability |
| $\xrightarrow{d}$ | Convergence in distribution |
| $\xrightarrow{a.s.}$ | Almost sure convergence |
| $O_p(1)$ | Bounded in probability |
| $o_p(1)$ | Converges to zero in probability |
This skill works with:
Version: 1.0 Created: 2025-12-08 Domain: Mathematical Statistics, Causal Inference
Creating algorithmic art using p5.js with seeded randomness and interactive parameter exploration. Use this when users request creating art using code, generative art, algorithmic art, flow fields, or particle systems. Create original algorithmic art rather than copying existing artists' work to avoid copyright violations.
Applies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.
Create beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.