Critical Speed & the Power-Duration Curve: The Runner's Complete Guide

Critical Speed is formally defined as the upper boundary of the heavy-exercise intensity domain — the fastest pace at which physiological systems can reach a metabolic steady state rather than drift inexorably toward their limits. Below CS, oxygen uptake, blood lactate, and intramuscular pH stabilize after an initial transient, meaning you could (in principle) hold the pace for a prolonged period. Above CS, the VO2 slow component develops progressively, blood lactate accumulates at an accelerating rate, phosphocreatine depletes, and pH falls — the textbook signature of the severe-intensity domain. The boundary between 'I can stabilize here' and 'I am on a countdown to failure' is Critical Speed, and it is one of the most important intensities in endurance physiology.

The concept originated with Monod & Scherrer (1965), who studied local muscle work on small synergist groups and observed a hyperbolic relationship between force output and endurance time. Moritani et al. (1981) extended the framework to whole-body exercise on the cycle ergometer, coining the term 'critical power.' Hill (1993) provided the canonical review of the model's mathematical structure, and Poole, Burnley, Vanhatalo & Jones (2016) synthesized three decades of running-specific research into a definitive physiological review. Jones et al. (2019) updated the training implications with particular attention to running. Across this literature, CS consistently falls within approximately 3–5% of the Maximal Lactate Steady State (MLSS) and within ±3% of LT2 determined by the 4 mmol·L⁻¹ fixed threshold or individualized methods.

The physiological significance of CS extends beyond mere pace prescription. It is the only commonly used threshold that is empirically derived from an individual's own power-duration relationship rather than from population-averaged assumptions or a single biomarker. A runner whose VO2 max sits at 60 mL·kg⁻¹·min⁻¹ might have a CS at 85% of velocity at VO2 max (vVO2max) if they have excellent mitochondrial density, or only 78% if they have lower aerobic economy — CS captures this individuality directly. In practical terms, Critical Speed is the pace at which you are maximally stressing aerobic metabolism without yet tipping into the runaway physiological regime that guarantees termination — which is exactly why it maps so well onto threshold training prescriptions.

The mathematical backbone of the critical power framework is the two-parameter hyperbolic model: t = D' / (P − CS), where t is time to exhaustion at a constant power (or speed) P above Critical Speed. Rearranged, the relationship states that the product (P − CS) × t equals a constant D', which has units of distance in meters for running. This equation has been validated in hundreds of studies across running, cycling, rowing, swimming, and kayaking (Jones et al. 2019). The hyperbolic structure means that effort above CS is bounded: at CS + 1 m/s you might last 150 seconds before failure, at CS + 2 m/s only 75 seconds, at CS + 4 m/s just 37 seconds. The curve has a vertical asymptote (infinite effort at CS + 0) and a horizontal asymptote at CS itself — you cannot sustainably run above CS, but you also cannot produce any finite effort above CS without drawing from D'.

CS is expressed in m/s (or equivalently, min/km or min/mile), while D' is expressed in meters — the finite distance 'bank' you can cover above CS before exhaustion. A useful mental model is that D' represents a reservoir of anaerobic capacity, phosphocreatine reserves, and tolerance for metabolite accumulation (H+, inorganic phosphate, K+) that depletes whenever you run faster than CS and partially refills when you run slower than CS. The running analog of cycling's CP (critical power, measured in watts) and W' (work above CP, measured in kilojoules) is CS (m/s) and D' (m). For Stryd users and other running power meter owners, the analogous terms CP and W' are often used directly, with CP expressed in watts and W' in kilojoules.

D' varies systematically with training history and event specialization. A well-trained middle-distance specialist (800 m to 1500 m) typically carries a D' of 300–400 m, reflecting extensive glycolytic capacity and high tolerance for lactate and H+ accumulation. A marathon specialist with the same CS might carry only 150–200 m of D', reflecting a physiology optimized for prolonged steady-state work rather than high-intensity bursts. This is not a defect — it is an adaptation to the demands of the event. Interval training (especially 30-second to 2-minute high-intensity repeats) primarily builds D', while threshold and tempo training primarily raises CS itself. Knowing your current D' tells you whether your training prescription should lean toward raising the ceiling (more tempo, more cruise intervals) or expanding the anaerobic reservoir (more VO2 max intervals, more lactate tolerance work).

Typical D' Ranges by Running Specialty

The gold-standard field test for CS and D' is the two time trials method: a 3-minute all-out effort and a 12-minute all-out effort, performed on a flat, windless track or certified course and separated by 48–72 hours of recovery. Compute mean speed for each (total distance divided by duration), then solve the linear version of the hyperbolic model: distance = CS × time + D'. With two (time, distance) points, two unknowns are uniquely determined — CS is the slope of the line through those points, and D' is the y-intercept. For example, if you cover 900 m in 180 s and 3,600 m in 720 s, CS = (3600 − 900) / (720 − 180) = 5.0 m/s and D' = 900 − 5.0 × 180 = 0 m (a pathological fit indicating a 3-minute effort that was not truly all-out). Realistic efforts give D' values of 150–300 m.

Burnley, Doust & Vanhatalo (2006) validated a single-session alternative: the 3-minute all-out exhaustion test. The protocol is brutally simple — run as hard as possible for 180 seconds with no pacing strategy, on a track or treadmill, with continuous pace recording. The terminal 30-second mean speed (150–180 s) equals CS, because by that point D' is fully depleted and the runner has collapsed onto the CS asymptote. The excess distance covered above CS over the full 180 seconds equals D'. Galbraith, Hopker, Lelliott, Tolfrey & Passfield (2014) further refined a 1500 m + 3000 m race-based protocol that produces CS estimates within ±2% of laboratory values for trained runners. The key to any field test is genuine maximal effort — submaximal pacing collapses the two-parameter fit and produces misleadingly low D' values.

For runners unable or unwilling to perform dedicated time trials, race-data modeling is a viable alternative. Collect three to four recent maximal race performances spanning a broad duration range — typically a 3K or 5K (8–20 min), a 10K (30–50 min), and a half marathon (75–120 min) — and fit the two-parameter model via least-squares regression of distance on time. Stryd and other running power meters automate this process using running power (in watts) rather than speed, producing a CP (critical power) and W' (kilojoules) output. The Stryd 'Critical Power Test' protocol uses a 9-minute and a 3-minute time trial as input. Regardless of method, retest every 8–12 weeks during focused training blocks, or after any performance breakthrough, to keep training zones calibrated to your current physiology.

Critical Speed overlaps with several other physiological thresholds, but each metric has a distinct derivation and best use case. Consider a representative 3:00 marathoner: VDOT would compute to approximately 52 (Daniels 4th ed.), lactate threshold 2 (LT2) pace would sit around 4:05/km, Critical Speed would fall near 4:00–4:05/km, and Stryd FTP (functional threshold power, treated as critical power) would be approximately 270 W. The four metrics are close but not identical, and their differences encode different physiological assumptions. VDOT is a race-predictor constructed from Daniels' population-averaged tables; it assumes 'typical' running economy (~200 mL·kg⁻¹·km⁻¹) and derives training paces from that single-number fit. LT2 is a lactate-specific threshold, most commonly defined as the running pace corresponding to 4 mmol·L⁻¹ blood lactate or individualized via baseline + 1.5 mmol·L⁻¹ methods.

Critical Speed, in contrast, is neither a population average nor a single biomarker — it is empirically derived from the individual's own power-duration curve and respects idiosyncratic running economy. A runner with unusually good economy (say, 180 mL·kg⁻¹·km⁻¹) will have a higher CS than VDOT predicts because they cover more distance per unit of VO2. A runner with poor economy (220 mL·kg⁻¹·km⁻¹) will have a lower CS than VDOT predicts. This individualization is why CS-based race predictions often outperform VDOT for runners at the tails of the economy distribution. Stryd's FTP is conceptually identical to CS, but implemented in watts via the Stryd power model; it is subject to assumptions about the power-speed relationship that may differ slightly from direct speed-based CS estimation.

The practical upshot: use VDOT as a quick, standardized estimate when you have a single recent race result and no equipment; use LT2 when you have lab access and want a lactate-anchored zone structure; use CS when you want the most physiologically grounded, individual-specific threshold for training prescription and race prediction; use Stryd FTP when you own a running power meter and want automated continuous tracking. The four converge for 'typical' runners but diverge meaningfully for runners with unusual economy, extreme D' values, or strong specialization in either short or long events.

Comparison of Key Running Threshold Metrics

The most direct application of CS is as a training zone anchor. Because CS is a mechanical, pace-based threshold rather than a lagged cardiovascular proxy, it keeps the intended metabolic stress constant across conditions that disrupt heart-rate zones — heat, cold, dehydration, fatigue, caffeine, altitude, and cardiovascular drift during long sessions. A common CS-based zone structure assigns easy running to below 80% CS (true aerobic base work), steady running to 80–90% CS (aerobic development with minimal lactate accumulation), threshold running to 90–100% CS (LT2-adjacent, sustainable for 20–60 minutes depending on individual D' and position within the zone), VO2 max intervals to 105–120% CS (heavily above CS, drawing significantly on D', sustainable for 3–8 minutes per rep), and neuromuscular work to above 120% CS (short reps at 30 seconds to 2 minutes, large D' drawdown per rep).

The value of CS-anchored prescription becomes most apparent in sessions sensitive to hitting exactly the right intensity. Consider a classic 6 × 5-minute cruise interval session targeted at LT2. If prescribed as '85% HRmax with 90-second jogs,' the first rep might hit the target correctly but by rep 4, cardiovascular drift has raised HR such that you are no longer running at LT2 — you are running 5–10 seconds per kilometer too easy relative to the intended metabolic stress. Prescribed as '98% CS with 90-second jogs,' the pace itself enforces the correct metabolic intensity regardless of HR drift. The same logic applies to VO2 max intervals (10 × 1 km at 105–110% CS is more reproducible than 10 × 1 km at '3K pace'), long tempo runs (40 min at 92% CS produces the intended sub-threshold stimulus), and progression long runs (final 20 min at 85% CS guarantees a productive but controlled finish).

CS-anchored training also clarifies the distinction between 'sub-threshold' and 'threshold' work — a distinction that modern Norwegian-influenced training models exploit heavily. Sub-threshold (85–95% CS) allows high cumulative volume at near-maximal aerobic stress without significant D' drawdown, meaning daily or near-daily exposure becomes feasible. True threshold work (95–100% CS) and supra-threshold work (100–105% CS) are more metabolically costly, recruit more D', and require longer recovery between sessions. The difference in recoverability between 95% CS and 102% CS is disproportionate to the small pace difference — which is why double-threshold days at 90–95% CS have become a dominant training modality among elite middle-distance runners over the past decade.

CS-Anchored Training Zones

Where CS anchors the ceiling of sustainable work, D' quantifies the finite anaerobic reservoir you can spend above it. The dynamics are asymmetric: D' depletes whenever you run above CS (at a rate proportional to how far above), and it recovers whenever you run below CS (at a rate that is emphatically non-linear, governed by intensity of recovery and individual kinetics). Skiba et al. (2012) developed the W'bal (W-prime balance) model, which tracks moment-to-moment D' expenditure and reconstitution during intervals. In running terms, the equivalent D'bal model lets you predict — and therefore design — interval sessions that hit specific D' depletion targets. A typical recreational runner's D' (200 m) corresponds to roughly 60 s at 105% CS, 40 s at 110% CS, or 25 s at 120% CS if spent continuously.

Concrete example: a runner with CS = 4.5 m/s (3:42/km) and D' = 200 m designs a 10 × 400 m session at 115% CS (5.175 m/s, or about 3:13/km) with 200 m jog recoveries at 60% CS (2.7 m/s). Each 400 m rep takes 77 seconds, during which the runner spends (5.175 − 4.5) × 77 = 52 m of D' — approximately 26% of their 200 m reservoir per rep. The 200 m jog recovery at 60% CS takes 74 seconds, during which D' reconstitutes at a rate governed by Skiba's exponential kinetics; typical reconstitution might replenish 70% of what was spent, meaning net depletion per rep-plus-recovery cycle is about 8% of D'. By rep 10, cumulative depletion reaches roughly 50–60% of D', producing a controlled high-stress stimulus without outright failure. By contrast, running the same intervals with full standing rest would allow near-complete reconstitution between reps and produce a less demanding session despite identical work totals.

The classic shapes of interval sessions — short/fast with short rest (fractional D' depletion, cumulative stress), long/moderate with longer rest (per-rep D' depletion near maximum), or mixed-ladder sessions (testing D' dynamics across a range of intensities) — are mathematical expressions of D' dynamics. Experienced coaches have intuited these patterns for a century (Billat, Pirnay, Petit et al. 2000 analyzed their physiological foundations), but the Skiba framework lets you quantify and tune them. Sessions that drain D' to 80% of capacity and then require sustained running near CS develop lactate shuttling and pH tolerance; sessions that rapidly cycle D' between 40% and 60% depleted train the reconstitution kinetics themselves. The emerging practical lesson from Vanhatalo, Jones & Burnley (2011) and subsequent work is that interval session design is better understood as D' management than as 'hard effort with rest.'

For events lasting longer than approximately 8 minutes — where CS is the dominant physiological determinant — race pace can be predicted from the two-parameter model as: average speed ≈ CS + (D' / race_time). Because race_time itself depends on average speed (and therefore on the answer), the equation must be solved iteratively or via direct substitution. Example: a runner with CS = 14.5 km/h (4.03 m/s) and D' = 220 m targets a 10 km race. First pass: assume 42 min, then v = 14.5 + (0.22 / 0.7) = 14.81 km/h → 10 km in 40:32. Second iteration: v = 14.5 + (0.22 / 0.675) = 14.826 km/h → 10 km in 40:28. The model converges rapidly to ~40:30 for this athlete.

Compared to alternative predictors, CS-based estimation has distinct strengths. Daniels' VDOT assumes a fixed relationship between race times across distances, rooted in population averages. The Riegel formula (t2 = t1 × (d2/d1)^1.06) assumes a universal fatigue exponent of 1.06, which works well for average athletes but systematically misclassifies the tails of the distribution. High-D' athletes — strong finishers with large anaerobic reserves — outperform Riegel predictions at shorter distances and underperform at longer ones; low-D' athletes (aerobic specialists who 'fade' in the final kilometers) show the opposite pattern. CS-based prediction respects individual profiles by using the athlete's own hyperbolic curve rather than imposing a universal exponent. Vanhatalo, Jones & Burnley (2011) reported that CS-based 5K-to-10K prediction errors were typically 1–2% for well-tested athletes, compared to 3–5% for Riegel with runners at the distribution tails.

Two caveats govern race-prediction accuracy. First, the model assumes the race duration sits within the validity window of the power-duration curve — approximately 2 to 60 minutes, with the sweet spot at 5 to 30 minutes. Shorter races (400 m, 800 m) require additional anaerobic modeling beyond the two-parameter form. Longer races (marathon, ultramarathon) introduce complications that the instantaneous CS+D' model cannot handle: glycogen depletion, thermoregulatory strain, biomechanical durability degradation, and the creeping rise of metabolic cost as muscle damage accumulates. For half marathons, CS-based predictions typically land within 1–3% of actual performance for well-trained runners; for marathons, they require durability adjustments (typically 3–8% slower than the naive CS prediction, depending on the runner's long-run economy retention) to match reality.

Critical Speed is powerful, but it is not a license to ignore the model's boundary conditions. The first and most important caveat: CS is not an infinite-duration sustainable pace. Empirical studies (Jones et al. 2019; Vanhatalo, Jones & Burnley 2011) consistently find that runners can hold exactly CS for only approximately 30–60 minutes before requiring slowdown, even though the mathematical model suggests indefinite sustainability. This discrepancy arises because the two-parameter hyperbolic fit is a simplification that ignores slow-drifting physiological variables (core temperature, glycogen, hydration, muscle damage) that matter over long durations. The practical implication: CS is a useful anchor for predictions and training zones in the 2–60 minute window, but it is not 'marathon pace' for most runners. Marathon pace typically sits at 87–94% of CS, with the exact percentage depending on training status, course profile, and heat.

Testing error is the second major caveat. Single-session CS determinations carry approximately 3–5% measurement error, primarily from pacing variability and effort calibration. Re-running the 3-minute test twice in one week will often yield results that differ by 2–4% despite identical physiology, simply because all-out pacing is difficult to replicate. D' is even more variable between sessions, with test-retest coefficients of variation of 10–15% reported in laboratory studies. Environmental factors amplify the problem: heat reduces apparent CS by approximately 3–5% per 10°C above 15°C wet-bulb (Périard & Racinais 2015), and altitude reduces CS by approximately 7–15% depending on elevation. Any single CS value should be treated as a best-estimate with meaningful uncertainty bands rather than a precise constant.

For marathon and longer distances, CS should be complemented by durability measurement. Durability — the retention of CS and running economy after 90+ minutes of prior running — is highly variable between athletes and highly trainable, but not captured by the standard CS test itself. Two runners with identical fresh CS may differ by 8–10% in their CS-at-90-minutes, producing very different marathon performances. For Stryd users and other running power meter owners, be aware that the device's estimated CP depends on the internal power model, which makes assumptions about the relationship between speed, gradient, wind, and metabolic cost that may not perfectly match your physiology — reported watt values can differ by 5–10% between runners at the same true metabolic intensity. Final recommendation: retest CS every 8–12 weeks or after significant fitness changes, maintain a healthy skepticism about small changes (±3% is noise), and layer durability work into marathon preparation to expose any gap between fresh and fatigued CS.