What You Actually Need When You Open a Blank Document

Most people think a math reference sheet is a list of formulas to memorize. It's not. It's a tool you build for the work you actually do. I used to write these from scratch at the start of every project. Then I stopped doing that around 2018 and just kept updating one master sheet over the years. It lives on my local drive now and has saved me probably hundreds of hours across whatever projects landed on my desk. The first version I made was for a structural analysis job where we were running hand calculations to sanity-check FEA output. I pulled together everything from beam deflection formulas to matrix inverse shortcuts. Took me three days to assemble because I didn't have a standard format. The second version took an afternoon because I just copied the template and dropped in the new material. That's the main point of this entire

Core Math Reference Sheet

discussion right there, which is that the value isn't in having the formulas, it's in having a consistent layout you can trust when you're tired and the clock is ticking.

How to structure one without overthinking it

Start with the section you use most often. For me that was linear algebra. Write out matrix multiplication rules, determinant properties, eigenvalue definitions, and the Jacobi iteration method in that order. Don't add commentary next to each formula. Just the formula, the variable definitions, and the boundary conditions where it applies. If a formula breaks under certain conditions, note the condition inline. Example: Gauss-Seidel converges for strictly diagonally dominant or symmetric positive-definite matrices. Put that next to the update equation. The second section should be your numerical methods. Finite difference approximations for first and second derivatives, the trapezoidal rule, Simpson's rule, Newton-Raphson with its convergence criteria. I learned the hard way that most people write down Newton-Raphson without noting what happens when the derivative is near zero. In practice, if |f'(x)| drops below 1e-8 during iteration, switch to bisection for that step. I keep a small side note about this fallback on my sheet because I've lost half a day to a stuck Newton loop before. Calculus sections should include integration by parts patterns, substitution rules, and the fundamental theorem stated both ways. Not for learning calculus, but for reminding yourself which convention your code uses versus which convention your paper uses. I once spent forty-five minutes debugging a sign error that traced back to a textbook using a different orientation for the surface integral than my simulation code. A quick reference line showing both conventions side by side would have prevented that. For differential equations, list the standard forms: separable, first-order linear with integrating factor, exact equations, and the Laplace transform table for common functions. Add a note about when Laplace fails, which is usually when the forcing function grows faster than exponential order. I've seen engineers blindly apply Laplace transforms to systems with polynomial-exponential growth and then wonder why the inverse transform didn't exist. Linear systems deserve their own subsection even inside linear algebra. LU decomposition steps, backward substitution, condition number estimation through norm ratios. The condition number is the part most people skip. A matrix with a condition number above 1e12 will give you garbage results in double precision regardless of how clean your algorithm is. Write that threshold somewhere visible. I keep it at the top of my linear systems section now because I learned that lesson during a project where the stiffness matrix was ill-conditioned and nobody noticed until the displacement vector looked wrong by. The specific problem that changed how I build these sheets Around 2020 I was working on a heat transfer simulation where the boundary condition involved a piecewise temperature profile. The analytical solution required evaluating an integral with a discontinuous integrand. My reference sheet had the standard integration formulas but none of them covered the case where the integrand changed definition mid-domain. I ended up deriving the split-integral approach on the fly, which cost me about six hours of actual work time. After that, I added a new subsection for handling piecewise and discontinuous functions, including the rule for splitting integrals at discontinuity points and the Fourier series approach for periodic piecewise inputs. That six-hour gap is exactly why the reference sheet exists as a living document rather than a one-time project. Every time you hit a wall, the solution goes into the sheet. Not as a paragraph of explanation. Just the formula, the condition, and a one-line note about when it applies. The sheet grows slowly and becomes more useful over time. Where this approach falls apart A single static reference sheet cannot replace working knowledge. If you're staring at a formula you've never seen before, having it on your sheet doesn't help you apply it correctly. The sheet works best for formulas you use repeatedly but want to avoid re-deriving. For novel problems, you still need to understand the derivation path. The sheet also becomes a liability if it gets too long. I had one version that ran past forty pages because I kept adding edge cases. By the time I needed something urgent, I couldn't find it fast enough. Keep it under twenty pages. If you need more, split it into two documents by domain. A core sheet and an advanced supplement. Finally, don't treat it as a substitute for verification. I once used a misremembered formula for Gaussian quadrature weights that looked correct on my sheet. It was wrong by a factor of two. The reference sheet reflected my error, not the truth. Always cross-check derived formulas against a trusted source at least once, preferably before you build anything critical on top of them. What to include versus what to leave out Include formulas you use more than twice in a single project. Include variable definitions that aren't obvious from context. Include the limits of validity for each formula. Leave out derivations, proofs, and introductory explanations. You aren't building a textbook. You're building a lookup tool. If you find yourself writing more than two sentences next to a formula, you're doing it wrong. Move that content to a separate notes file. Start with what you need today. Add to it when the need arises. The best reference sheets are the ones that grow organically instead of being engineered upfront.