How Do You Calculate The Half Life

Why the Standard Formula Fails You Before You Even Start

The half-life formula is t½ = ln(2) / , where is the decay constant. That's what every textbook says. It's also completely useless in practice unless you already know your decay constant, which almost never happens on a real workbench. What actually happens is you measure activity at two different times, plot it, and back-calculate. The formula is just the finish line, not the road. I spent three weeks trying to get half-life measurements on a tritium source that was supposed to be 12.3 years. The problem wasn't the math. It was that my GM tube had a dead time of about 180 microseconds, and at higher count rates the dead-time correction started eating into my data in a non-linear way. Every time I tried to use the standard linear regression on ln(activity) vs time, the residuals showed a clear curve. The source wasn't behaving oddly — the electronics were. Once I measured the dead time separately using the two-source method and applied a paralyzable model correction instead of the standard non-paralyzable one, the half-life landed within 2% of the accepted value. That difference between paralyzable and non-paralyzable models is the kind of thing nobody tells you until you've burned a week on it.

How Do You Calculate The Half Life From Raw Measurements

Here is what the actual process looks like. You take a radioactive sample and measure its activity at regular intervals. The sample needs to be stable enough that nothing else is changing — no chemical reactions, no geometric shifts in your detector setup, no temperature drift that moves the source closer or farther from the detector window. I once had a NaI detector where the PMT high voltage drifted with room temperature, and I spent two days thinking my isotope had a shorter half-life than it actually did before I realized the 15-degree swing in my basement lab was shifting the gain enough to change counted rates by about 4 percent. You need at least five or six data points, preferably more. The rule of thumb is that your total measurement window should cover at least one to two half-lives. If you're measuring something with a half-life of days or weeks, that's manageable. If it's minutes, you need fast electronics and a way to start the clock precisely at a known moment. If it's years, you're basically measuring a constant and hoping your calibration doesn't drift over months of data collection. Once you have your activity values A1 at time t1 and A2 at time t2, you use the relationship A = A0 * e^(-t). Rearranging gives you = ln(A1/A2) / (t2 - t1). Then t½ = ln(2) / . That's it on paper. In practice you plug all your points into a spreadsheet or Python script, do a least-squares fit on ln(A) versus t, and the slope of that line is -. The uncertainty on the slope gives you the uncertainty on your half-life directly.

The part people mess up is the uncertainty handling. You can't just average pairwise half-life calculations from adjacent points and call it a day. Each point has its own statistical error — Poisson counting statistics mean the standard deviation is sqrt(N) where N is your count number. If you're getting 10,000 counts per minute, your uncertainty is about 1 percent. If you're getting 100, it's 10 percent. Short measurements with low activity make the half-life result look precise when it actually isn't. Always weigh your regression by the inverse variance of each point, not just do an unweighted fit.

The Edge Cases Where This All Falls Apart

Parent-daughter decay chains are the biggest headache. If your sample produces a daughter isotope that is also radioactive and has a comparable half-life, your measured activity isn't from a single exponential anymore. You get a sum of exponentials, and fitting it with a single decay constant gives you garbage. I ran into this with a scandium-46 source that was picking up some manganese-56 contamination from the activation foil. The Mn-56 half-life is about 2.6 hours while the Sc-46 is about 83 days. In the first couple of days, the Mn was dominating the counts and made the early part of my decay curve look like it was dropping way faster than it should. I had to fit the early data separately to isolate the Mn component, subtract it, and then fit the residual with the Sc-46 decay. Takes about twenty minutes if you know what you're doing, or three days if you're figuring it out blind. Another thing that bites people is secular equilibrium. If you have a long-lived parent feeding a short-lived daughter, after about five daughter half-lives the daughter activity equals the parent activity. Your detector can't tell them apart if they emit similar radiation. You're not measuring the parent's decay anymore — you're measuring a system in equilibrium where the apparent half-life is the parent's half-life only because the daughter keeps getting replenished. This is common with radon daughters and in many generator systems. If you don't account for it, you'll calculate a half-life that looks right but means something entirely different. For very short half-lives measured with conventional lab equipment, the dead time issue I mentioned earlier gets worse fast. Some labs use a pulse generator to inject known dead-time losses and correct offline. Others switch to a different detector type entirely — a silicon surface barrier for alpha emitters, for example, which has microsecond dead times instead of the hundreds of microseconds you get from a GM tube. The choice of detector matters more than the math ever will.

Practical Steps That Actually Work

Set up your geometry and don't touch it. Mark the source position, the detector window position, everything. A shift of a few millimeters changes the solid angle enough to matter if you're aiming for sub-percent precision. Log the room temperature if you're using a scintillator. Calibrate your detector's efficiency at the energy you're measuring before you start taking data. An efficiency calibration done at one energy doesn't transfer well to another energy — the absolute efficiency can change by factors of two or three across a typical gamma spectrum. Record your live time, not your clock time. Most modern multi-channel analyzers do this automatically, but if you're using a simple counter and a stopwatch, the difference between real elapsed time and live counting time adds up, especially at high rates. My rule is that if your dead-time correction is above 1 or 2 percent, you need to be using live time gating or at least measuring your dead time carefully and applying a correction. For the regression itself, I use a simple Python script with scipy.optimize.curve_fit that models the decay as an exponential with a constant background term. Including the background as a free parameter in the fit is important. You can measure the background separately, but having it in the fit handles small drifts and keeps the uncertainty estimates honest. The script returns the half-life, its standard error, and the reduced chi-squared so you can check whether your error model is reasonable. If the reduced chi-squared is much larger than 1, your errors are underestimated — usually because you ignored dead time, background fluctuations, or geometry drift.

There's no shortcut around good experimental design. The formula is straightforward. The measurement is where everything goes wrong. I've seen people get half-life values that were off by 30 percent because they didn't account for background, and others who nailed it to within 0.5 percent by measuring for long enough and treating the statistics properly. The difference is almost never the calculation method.