What hills actually cost you (and why the downhills don't pay it back)

The first hilly marathon I ran, I did the arithmetic on the elevation profile the night before and decided it was fine. Nine hundred feet of climbing, nine hundred feet of descending, net zero. What goes up comes down. I would give a little back on the way up and take it back on the way down, and my goal time was safe.

It was not safe. I finished six minutes off it, and none of the six came from the climbs I had worried about. They came from the last four miles, on flat ground, on quads that a series of "free" descents had quietly destroyed.

The maths of hills is not symmetrical, and that asymmetry is the whole story.

The exchange rate, up and down

The number most coaches use, and the one that has held up best against my own splits, is roughly 12 to 15 seconds per mile for every 1% of uphill grade. Going downhill you get back about 8 seconds per mile per 1% of grade, and only for the first few percent.

So a mile at a steady 2% climb costs you around 25 to 30 seconds against your flat pace. The matching 2% descent hands back about 16. You are down roughly ten seconds a mile on a stretch of road that looks, on paper, like it netted out to nothing.

Run that over the marathon I mentioned. Nine hundred feet of gain spread across 26.2 miles averages a little over 34 feet of climb per mile, which is about a 0.65% average grade on the up sections if half the course is climbing. Call it 8 seconds a mile lost on the ups and 5 given back on the downs. Three seconds a mile, times 26. Just over a minute. That is the direct cost, and it is smaller than most people fear.

The other five minutes came from somewhere else.

The real bill arrives late

Downhill running is eccentric loading. Your quadriceps are lengthening under load with every footstrike instead of shortening, which is the muscle action that produces actual microtrauma. You are not paying for the descent while you run it. You pay two hours later, when the fibres you shredded at mile 8 stop cooperating at mile 22.

This is why Boston is the classic trap. It is a net downhill course, about 450 feet lower at the finish than the start, and it has a reputation for being slow. The Newton hills get the blame. The Newton hills are not the problem. The problem is the first four miles, which drop roughly 300 feet, run at goal pace or faster with fresh legs and a tailwind of adrenaline. Everyone banks two minutes there and pays four back after mile 20.

A useful reframe: on a hilly course, treat downhills as an expense, not income. Run them relaxed, let gravity do it, keep your cadence quick and your steps short so you are not braking with each landing. Reaching for free speed on a steep descent is the single most expensive thing you can do in the first half of a race.

Where the physiology actually flips

There is a well-known set of treadmill data from Minetti and colleagues (2002) measuring the metabolic cost of running at grades from -45% to +45%. Two findings from it are worth carrying around.

First, cost per distance rises almost linearly on the way up, which is why the "seconds per 1% grade" rule works at all in the range real roads occupy.

Second, and less intuitively, the cheapest running is not on the flat. It is at about -10% grade — a fairly steep descent. Go steeper than that and the cost starts climbing again, because you stop running and start controlling a fall. At around -20% and beyond, descending is metabolically as expensive as flat running, and structurally far worse.

So the mountain trail race where you scream down a -25% fire road is not giving you anything back. You are spending energy and wrecking your legs. That is worth knowing before you write a split plan for one.

Naismith's rule is not for you

Someone in every trail running forum brings up Naismith's rule: one hour per 3 miles, plus an extra hour per 2,000 feet of ascent. It is a hiking rule from 1892 and it is fine for hiking.

For running, the ascent term is roughly usable and the base pace is nonsense. If you want a single number for vertical, the field convention among mountain runners is closer to one extra minute per 100 feet of gain at moderate effort, which works out slower than Naismith's 30 seconds per 100 feet because Naismith assumed you were walking the flats too. Use it as a sanity check on a trail race, never as a plan for a road race.

Adjusting a predicted time

Any race time prediction starts from a flat-course assumption, because the race you fed it was probably flat. Two adjustments get you close on a hilly one.

Vertical gain per mile. Total elevation gain divided by distance is the number that matters, not total gain. A 10K with 500 feet of gain (80 ft/mile) is a much harder race than a marathon with 900 (34 ft/mile), even though the marathon's number is bigger.

A rough surcharge. For road races on rolling terrain, adding 1.5 to 2 seconds per mile for every 10 feet of gain per mile lands close for most runners. At 34 ft/mile that is 5 to 7 seconds a mile — right in line with the split-by-split estimate above, and it has the advantage of including some of the late-race damage.

For genuinely mountainous courses, stop adjusting and start looking at actual finish times. Pull last year's results, find three or four runners whose flat 10K or half times you know, and see what they ran. Course-specific data beats any formula once the profile gets serious.

What I do now

I run the ups by effort, not by pace, and I accept whatever the watch says. I run the downs at the pace gravity offers and no faster, with short steps. And I set my goal time from the course profile rather than from my flat PR, which sounds obvious and took me two blown marathons to actually do.

Build the pace band from the adjusted target, not the flat one. A band that tells you to hold 7:30 up a 3% grade is not a plan, it is a way of finding out early that you cannot.