Scientific LLM Benchmarks
GitHub
← All benchmarks
Physics & Astro· theoretical

TPBench

University of Wisconsin-Madison · 2025

57 theoretical physics problems in high-energy theory and cosmology, undergraduate to research level.

PhysicsAstronomy
Task type
open-ended
Modality
text
Access
request
Size
57 items
License
MIT
Metrics
accuracy, avg@5, best@5
problem_id
A 3-State QM Problem
domain
QM
difficulty_level
2
problem
The Hamiltonian of a three-level system is given as $H = \begin{pmatrix} E_a & 0 & A \\ 0 & E_b & 0 \\ A & 0 & E_a \\ \end{pmatrix}$ where $A$ is real. The state of the system at time $t=0$ is (in this basis) $\psi(t=0) = \frac{1}{\sqrt{2}}\begin{pmatrix}1 \\ 1\\ 0\end{pmatrix}$ What is the expectation value of the energy at time $t$?
solution
The eigenstates are easily found to be $\frac{1}{\sqrt{2}}(1,0,\pm 1)^T$ and $(0,1,0)^T$ with corresponding energies $E_a\pm A$, $E_b$. Let us denote them as $|1\rangle$, $|2\rangle$ and $|3\rangle$. Given state $\psi$ is decomposed as $\frac{1}{2}(|1\rangle +|2\rangle) + \frac{1}{\sqrt{2}}|3\rangle$, the expectation of energy stays constant: \begin{equation} \langle E\rangle = \frac{1}{4}((E_a+A)+(E_a-A)) + \frac{1}{2}E_b =\boxed{ \frac{1}{2}(E_a+E_b)}. \end{equation}
answer
\begin{equation*} \boxed{\langle E\rangle = \frac{1}{2}(E_a+E_b)} \end{equation*}
code_answer_requirements
Provide the answer in the form of \texttt{python} code. Implement the following function \begin{python} def expectation_value(A: float, E_a:float, E_b:float, t:float) -> float: pass \end{python}
reference_implementation
\begin{python} def expectation_value(A: float, E_a:float, E_b:float, t:float) -> float: return 0.5*(E_a+E_b) \end{python}
problem_id
Bias of a Sampled Halo Field
domain
Cosmology
difficulty_level
5
problem
In cosmology, large-scale cosmological dark-matter halo fields are biased tracers of the underlying Gaussian matter density $\delta_m$. Assume we have a sample $\delta_m$. We simulate a halo number density field by taking $n(\mathbf{x}) = \bar{n}\max(0,1+b\delta_m(\mathbf{x}))$, where bare number density $\bar{n}$ and bare bias $b$ are specified constants. What is the bias of the sampled halo field? Derive an equation to evaluate the bias which depends on the bare bias and the variance in each pixel.
solution
\textbf{Detailed Steps:} The solution to this question involves some domain knowledge, parts of which were given in the problem's statement, some approximations sourced by the domain knowledge, and some mathematical calculations. The domain knowledge is very basic and should be known to anyone in the field. Approximations are intuitive and also, mostly, inspired by the domain knowledge. Following Polya, we can organize it as follows: \\ \textbf{Understand the problem.} The number density of halos $n_h(\mathbf{x})$ is defined as \begin{equation} N_h = \int_{V} n_h(\mathbf{x})d\mathbf{x}.\en …
answer
The bias of the sampled halo field is given by: \begin{equation} \boxed{ b^{'} = \frac{b \Phi_1\left(\frac{1}{|b|\sigma}\right)}{\Phi_1\left(\frac{1}{|b|\sigma}\right)+|b|\sigma\phi_1\left(\frac{1}{|b|\sigma}\right)}} \end{equation} where $\Phi_1$ is the normal cumulative distribution function, $\phi_1$ is the standard normal probability density function, $b$ is the bare bias, and $\sigma$ is the pixel variance.
code_answer_requirements
Provide the answer in the form of the \texttt{python} code. Implement the following function. \begin{python} #let b_in stand for bare bias def b_eff(sigma: float, b_in:float) -> float: pass \end{python}
reference_implementation
\begin{python} from scipy.stats import norm #let b_in stand for bare bias def b_eff(sigma: float, b_in:float) -> float: alpha = sigma*abs(b_in) return b_in*norm.cdf(1/alpha)/(norm.cdf(1/alpha)+alpha*norm.pdf(1/alpha)) \end{python} \newpage \newpage

Real rows from the Hugging Face datasets server · long values truncated